Statical Fluid Mechanics MONIN

STATISTICAL
FLUID MECHANICS:
Mechanics of Turbulence
Volum e 1
A. S. Monin and A. M. Yaglom

English edition updated, augmented and revised
by the authors
Edited by John L. Lumley
TH E M IT PRESS
Cambridge, Massachusetts, and London, England
Originally published in 1965 by N auka Press, Moscow, under the
title Statisicheskaya gidromekhanika — Mekhanika Turbulentnosti^
by Andrey S. Monin and Akiva M. Yaglom.
Translated from the Russian by Scripta Technica, Inc.
Second printing, D ecember 1973
Third printing, March 1977
Fourth printing. May 1979
English translation copyright © 1971 by
The Massachusetts Institute of Technology
All rights reserved. N o part of this book may be reproduced in
any form or by any means, electronic or mechanical, including
photocopying, recording, or by any information storage and
retrieval system, without permission in writing from the publisher.
ISBN 0 262 13062 9 (hardcover)
Library of Congress catalog card number: 70-110232
Printed in the United States of America
AUTHORS' PREFACE
TO THE ENGLISH EDITION
The flows o f fluids with which one has to deal in engineering, and
which one meets in nature, are turbulent in the overwhelming
majority o f cases, and their description demands a statistical
approach. Laminar flows, which are quite accessible to individual
descriptions, occur with exotic infrequency. We are persuaded that
fluid mechanics cannot be hmited to the study o f these seldom
encountered special cases, and that the classical description o f
individual laminar flows, for all its unquestioned importance and
value, must be considered only as an introductory chapter to the
theory o f real turbulent flows, in which the objects of investigation
are the properties o f ensembles of flows, arising from macroscop-
ically identical external conditions.
We would like such a view to spread through the entire fluid
mechanics community, and this is one o f the goals of our book. For
this reason we are very happy with the appearance of the English
edition, which will help spread our views on fluid mechanics in
English-speaking countries.
We have tried to summarize in this book the majority of the
fundamental works and ideas o f modern turbulence theory, de
scribed and discussed in studies from various countries. This is why
the book has turned out to be so thick. But, of course, our own
scientific interests must be reflected in the selection o f material, and
considerable attention has been given to problems on which we
ourselves have done substantial w ork-particularly the theory of the
local structure of developed turbulence due to A. N. Kolmogorov,
whose students we were, and the theory of turbulence in stratified
media, which finds wide applications, primarily in geophysics.
The mechanics o f turbulence is a lively, actively developing
science, and after the Russian edition of our book saw the light
(Volume 1 in 1965 and Volume 2 in 1967), a large number of new
iv AUTHORS'PREFACE
and interesting worlds appeared in this area. The present edition has
been replenished with information on many new works (the
bibliograpiiy included in the book has, as a result, grown by several
hundred entries) and by the inclusion o f a series o f improvements
and additions (all o f this work was carried out by the second author).
Particularly great changes were undergone by Section 2, dealing with
hydrodynamic stability and transition to turbulence, and Section 8 ,
in which are set forth experimental results on turbulence in a
thermally stratified boundary layer.
We consider it our pleasure to express our deep gratitude to our
editor. Prof. J. L. Lumley, who did a great deal to improve the
English edition o f our book.
A. S. Monin
A. M. Yaglom
EDITOR'S PREFACE
TO THE ENGLISH EDITION
This is a translation of the first Russian edition of CxaTHCTHHecKaa

rH.^poMcxaHHka. In editing the manuscript of the translation, exten
sive use was made of the translation of Volume 1 prepared by the Joint
Publications Research Service at the behest of K.L. Calder of the U.S.
Environmental Science Services Administration, and of the transla
tion of Chapter 4, Volume 1, prepared at the instigation of F. PasquOl
of the British Meteorology Research Division. The cooperation of
both these individuals is gratefully acknowledged.
The edited English manuscript was sent to Yaglom, who made
very extensive additions, corrections and revisions to the technical
content, after which it was again edited. Russian technical style tends
to be turgid with internal cross-references which sound redundant to
an English ear; in addition, the inflected character of the language
makes possible a complexity of sentence structure that is dizzying. A
conscientious translator hesitates to paraphrase too freely, with the
consequence that the translation retains an unmistakable flavor of
the original. Yaglom has an excellent ear for English style, and as he
revised he also made innumerable suggestions for freer paraphrases,
which sound more natural in English. As editor, I wish to express my
gratitude for his cooperation, giving him full credit for what is good
in this translation, while taking the blame for remaining inade
quacies.
J. L. Lumley
FOREWORD
The theory o f turbulence discussed in this book is based on the usual

macroscopic description of flows o f liquids and gases, considered as
continua, and on the classical equations o f fluid mechanics. However,
unlike ordinary fluid mechanics, the theory o f turbulence does not
study individual fluid flows but the statistical properties o f an
ensemble of flows having macroscopically identical external condi
tions. Hence, the title o f this book, Statistical Fluid Mechanics.
However, we must stress that questions dealing with the deduction o f
the macroscopic equations of fluid mechanics from the statistical
laws o f the kinetic theory o f gases, which sometimes are also referred
to by this title, are not discussed in the present work.
The basic concepts and ideas relating to turbulence and our
approach to the presentation of the mechanics of turbulence are
outlined in the Introduction, together with a brief sketch o f the
historical development of the theory of turbulence, and the plan o f
both volumes o f the present work. The Introduction is intended to
describe in the most general terms the main problems and methods
o f the recent theory o f turbulence and to explain some o f its
practical applications. Naturally, we have had to make use here of
many concepts which will be expanded in detail only in subsequent
parts of the book.
The book is divided into chapters, sections and subsections. The
numbering of the chapters and sections runs continuously, while the
subsections are numbered within each section. The number o f the
section and of the subsection is indicated in the subsection number
separately (e.g., the fourth subsection of the second section is
denoted as 2.4). The equations are also numbered in a similar
manner: e.g., Eq. (2.15) is the fifteenth equation in the second
section. When referring to the works listed in the bibliographies at
the end o f each volume, we cite the name o f the author and, in
brackets, the date o f publication o f the work. In cases when several
works o f a given author are cited which all appeared in the same
year, these are denoted further by letters of the alphabet. Except in
viii FOREWORD
the Introduction, the initials of the authors are given only in cases
where one must distinguish between two authors with the same
surname.
We have tried, as far as possible, to use generally accepted notation
in this book. In the numerous cases where different authors use
different symbols for the same quantities, we usually decided to
select one of these, and not to introduce new notation. In several
cases this has led to the same symbol being used for different
quantities in different parts of the book. Sometimes, for various
reasons, it also proved convenient to denote the same quantity by a
different symbol in different sections of the book. In all these cases
the notation used is specified in the text.
This book represents the combined work o f both authors. We feel
that we must point out the great influence on the writing of it, and
on our own work in the field of turbulence, of our frequent
discussions with Andrei Nikolaevich Kolmogorov, our Professor in
our student days. We have tried to reflect many o f Kolmogorov’s
ideas in this book.
We are also greatly indebted to A. M. Obukhov who was one of
those who initiated the writing of this book, and with whom we
discussed the selection o f the material and the details o f the
exposition o f many questions. Some parts of the manuscript were
read by L. A. Dikiy, Ye. A. Novikov and V. I. Tatarskiy, who made a
number of profitable comments. In discussions on the analysis o f the
experimental data on atmospheric turbulence, A. S. Gurvich and L.
V. Tsvang were active. G. S. Golitsyn helped us in the preparation
and editing o f the manuscript. To all these colleagues we wish to
express our sincere gratitude.
A. S. Monin
A. M. Yaglonrt
CONTENTS
Authors’ Preface to the English Edition iii

Editor’s Preface to the English Edition v
Foreword vii
Introduction 1
Chapter 1 Laminar and Turbulent Flows 27

1. Equations o f Fluid Dynamics and Their Consequences 27
1.1 System o f Dynamical Equations for an
Incompressible Fluid 27
1.2 Simple Flows o f Incompressible Fluid 31
1.3 The Reynolds Number and Similarity Criteria 38
1.4 Flows with Large Reynolds Numbers; The
Boundary Layer 41
1.5 General Equation o f the Heat Budget and the
Thermal C onduction E quation; Forced and
Free Convection 52
1.6 Similarity Criteria for a Thermally hihomogeneous
Fluid; The Thermal Boundary Layer 59
1.7 Small Oscillations o f a Compressible Fluid 64
2. Hydrodynamic histability and Transition to Turbulence 71
2.1 Concept o f Turbulence; Empirical Data on Transition
to Turbulence in Tubes and Boundary Layers 71
2.2 Turbulent Flow Past Solid Bodies; Boundary-layer
Separation, the Drag Crisis and the Mechanism of
Boundary-layer Transition 79
2.3 H ydrodynam ic histability 85
2.4 Simple Examples o f Absolutely Unstable Fluid
Flows 89
2.5 Mathematical Form ulation o f the StabiUty Problem
for Infinitesimal Disturbances 92
2.6 Stability o f Flow between Two Rotating Cylinders 96
2.7 Stability o f a Layer o f Fluid Heated from Below 107
2.8 Stability o f Parallel Flows 1 11
2.9 Stability to Finite Disturbances; G row th o f
Disturbances and Transition to Turbulence 147
X CONTENTS
Cluipter 2 Miithematical Description of Turbulence.

Mean Values and Correlation Functions 205
3. Methods o f Taking Averages. Fields o f Fluid Dynamic
Variables as Random Fields 205
3.1 Practical Methods o f Averaging and Reynolds
Conditions 205
3.2 Random Fields o f Fluid Dynamic Variables and
Probability Averaging 209
3.3 Concept o f Ergodicity. Statistical Form ulation o f
the F undam ental Turbulence Problem 214
3.4 Characteristic Functions and the Characteristic
Functional 218
4. Moments o f Fluid Dynamic Fields 222
4.1 Moments and Cum ulants o f R andom Variables 222
4.2 Moments and Cum ulants o f R andom Fields 226
4.3 Random Fields with a Normal Probability Distribution
(Gaussian Fields) 231
4.4 D etermination o f the M oments and Cumulants o f a
R andom Field According to Its Characteristic Functional 237
4.5 Stationary R andom Functions 243
4.6 H omogeneous R andom Fields 246
4.7 The Ergodic Theorem 249
Chapter 3 Reynolds Equations and the Semiempirical
Theories of Turbulence 257
5. Turbulent Shear Flows in Tubes, Boundary Layers, etc. 257
5.1 Reynolds E quations 257
5.2 General Form o f the Mean Velocity Profile Close
to a Rigid Wall 264
5.3 Flow Close to a Sm ooth Wall; Viscous Sublayer and
Logarithmic Boundary Layer 270
5.4 Flow Along a Rough Wall; Roughness Parameter and
Displacement Height 284
5.5 Turbulent Flows in Tubes and Channels; Skin
Friction Laws 296
5.6 Turbulent Boundary Layer on a Flat Plate 311
5.7 Profile o f Concentration o f a Passive A dmixture
Close to a Wall; Mass- and Heat-Transfer in a
Turbulent Boundary Layer 327
5.8 Free Turbulence 347
5.9 Semiempirical Theories of Turbulence 364
6. The Energy Balance Equation and Its Consequences 373
6.1 Equation for the Reynolds Stress Tensor 373
6.2 Equation o f Turbulent Energy Balance 381
6.3 General Concept o f the Viscosity and Thermal
Diffusivity 388
CONTENTS Xi
6.4 Energy Balance in a Compressible Fluid 393

6.5 The Richardson Num ber and the Eddy Viscosity in a
Thermally Stratified Medium 398
6.6 Turbulence in the Planetary Boundary Layer o f the
Atmosphere 404
6.7 D istribution o f Suspended Particles in a Turbulent
Flow 412
Chapter 4 Turbulence in a Thermally Stratified Medium 417

7. Generalization o f Logarithmic Layer Theory to Thermally
Stratified Flows 417
7.1 Thermally Stratified T urbulent Boundary Layer as a
Model o f the A tmospheric Surface Layer 417
7.2 Application o f Dimensional Reasoning to Turbulence
in a Stratified Medium 425
7.3 The F orm o f the Universal Functions Describing
Turbulence in a Stratified Medium 430
7.4 Further Discussion o f the Universal Functions;
Interpolation Formulas and Semiempirical Formulas 443
7.5 General Similarity Hypothesis for a T urbulent Regime
in a Stratified Boundary Layer and Its Application
to the Characteristics o f Turbulent Fluctuations 456
8. Comparison o f the Theoretical Deductions with the Data 468
8.1 Wind-Velocity Profiles in the Surface Layer o f the
A tmosphere 468
8.2 Data on Tem perature and H umidity Profiles 486
8.3 Methods o f Measuring Turbulent Fluxes o f
M om entum , Heat, and Water V apor 504
8.4 D etermination o f T urbulent Fluxes from Data on the
Profiles o f Meteorological Variables 509
8.5 Characteristics o f Wind Velocity and Tem perature
F luctuations in the A tmospheric Surface Layer 517
Chapter 5 Particle Dispersion in a Turbulent Flow 527

9. The Lagrangian Description o f Turbulence 527
9.1 The Lagrangian Dynamic Equations o f an
Incompressible Viscous Fluid 527
9.2 Lagrangian Turbulence Characteristics 535
9.3 Displacement Characteristics o f a Single Fluid
Particle; the Case o f Homogeneous Turbulence 540
9.4 Fluid Particle Displacements in Grid Turbulence and
in Turbulent Shear Flows 547
9.5 The Lagrangian Velocity Correlation Function and Its
Relationship to the Eulerian Statistical
Characteristics 569
XII CONTENTS
10. Turbulent Diffusion 579

10.1 Problem o f the Description o f Turbulent Diffusion 579
10.2 Interaction Between Molecular and Turbulent
Diffusion 591
10.3 Semiempirical E quation o f T urbulent Diffusion 606
10.4 Diffusion in a Field o f Homogeneous Turbulence
and in Simple Shear Flows 614
10.5 Diffusion in the Atmospheric Surface Layer 640
10.6 Diffusion with a Finite Velocity 676
Bibliography 695
Author Index 747
Subject Index 757
INTRODUCTION
This book is devoted to the mechanics of turbulent flows o f liquids

and gases. It begins with a resum 6 o f the basic equations of fluid
mechanics and, generally speaking, demands of the reader no prior
knowledge in the field of turbulence. For this reason it seemed to ais
that it would be worthwhile to explain here, if only briefly, what will
be dealt with in the book.
It is known that all flows of liquids and gases may be divided into
two sharply different types; the quiet smooth flows known as
“laminar” flows, and their opposite, “turbulent” flows in which the
velocity, pressure, temperature and other fluid mechanical quantities
fluctuate in a disordered manner with extremely sharp and irregular
space- and time-variations. As a typical example. Fig. 1 shows a trace
of the time fluctuations o f the wind velocity, the vertical component
of velocity and the temperature in the surface layer of the
atmosphere. The data of Fig. 1 were obtained by measuring the
velocity and temperature with special precise, low-inertia devices.
The complicated nature of these curves indicates at once that the
corresponding motion of the air was turbulent. The variety o f
fluctuations o f different periods and amplitudes observed in the
traces shown in Fig. 1 illustrates the complicated structure o f
turbulence in contrast to simple laminar flows. This complicated
structure affects many properties of flows which differ greatly in the
laminar and turbulent cases. Thus, turbulent flows possess far greater
ability to transfer momentum (figuratively speaking, a turbulent
medium has an enormous effective viscosity) and hence in many
cases they exert far greater forces on rigid bodies in contact with the
fluid. Similarly, turbulent flows possess an increased ability to
transfer heat, soluble admixtures, and suspended particles and to
propagate chemical reactions (in particular, combustion). Due to the
presence o f internal inhomogeneities, turbulent flows are able to
scatter sound waves and electromagnetic waves and to induce
fluctuations o f their amplitudes, phases, etc.
2 STATISTICAL FLUID MECHANICS
The indicated properties of turbulent flows may evidently be very

significant for many problems in natural science and technology.
Thus the question o f how frequently turbulent flows are encoun
tered is o f undoubted interest. It is found that an overwhelming
majority o f the flows actually encountered in nature and technology
are, in fact, turbulent flows, while the laminar flows, which are stud
ied in detail in fluid mechanics, occur only as fairly rare exceptions.
FIG. 1. Typical example of a recording of fluctuations of temperature, wind

velocity and the vertical component of wind velocity.
In particular, the various motions o f the air in the earth’s

atmosphere, from a slight breeze in the surface layer (to which the
measurements o f Fig. 1 correspond) up to general atmospheric
circulation o f planetary scale, are turbulent. Atmospheric turbulence
plays a fundamental role in the transfer of heat and moisture by air
masses, in evaporation from the surface of land or water, and in
thermal and dynamic interaction between the atmosphere and the
underlying surface which has a considerable effect on changes in the
weather. Atmospheric turbulence determines also the spreading of
admixtures in the air, the production of wind waves on large lakes,
seas, and the ocean, the production of wind currents in the ocean,
the buffeting of airplanes and other aircraft and the vibrations of
many structures on the earth’s surface. Finally, turbulent fluctua
tions o f the refractive index affect significantly the propagation of
light and radio waves from terrestrial and cosmic sources.
INTRODUCTION 3
Turbulent, too, are the flows o f water in rivers, lakes, seas and
oceans, and also the motions o f gases in interstellar nebulae having an
enormous scale many orders greater than the earth. Finally,
practically all flows in pipes encountered in technology and
engineering are turbulent, e.g., in water-pipes, gas mains, petroleum
pipelines, the nozzles of jet engines, etc; and also the motions in
boundary layers over the surface o f moving aircraft, in liquid or gas
high-speed jets issuing from a nozzle, in the wakes behind rapidly
moving rigid bodies—propeller blades, turbine blades, bullets, pro
jectiles and rockets. Thus turbulence is literally all around us both in
nature and in engineering devices using flows o f liquids and gases;
therefore its study is extremely im portant from the practical
viewpoint.
Turbulent flows are also o f great interest from a purely theoretical
point o f view as examples o f nonlinear mechanical systems with a
very great number o f degrees of freedom. Indeed, the motion o f any
continuous medium, strictly speaking, is described by an infinite
number o f generalized coordinates (e.g., by the coefficients o f the
expansion o f the velocity field with respect to some complete system
of functions of the spatial coordinates). For laminar motion
these coordinates can usually be chosen in such a way that only a
few o f the corresponding degrees o f freedom will be excited, i.e., will
actually take part in the motion. However, for turbulent motion, an
enormous number o f degrees o f freedom are always excited, and
hence the variation vvdth time of any physical value will be described
here by functions containing a vast number of Fourier components,
i.e., by functions o f an extremely complicated nature (cf. again Fig.
1). Therefore, in this case it is practically hopeless to attem pt to
describe the individual time variations o f all the generalized
coordinates corresponding to the excited degrees o f freedom (i.e., to
find a mathematical expression for the time-dependence of the fields
of velocity, pressure, etc., o f a single individual flow). The only
possibility in the theory o f turbulence is a statistical description,
based on the study o f specific statistical laws, inherent in phenomena
en masse, i.e., in large ensembles of similar objects. Thus only
statistical fluid mechanics, which studies the statistical properties o f
the ensembles o f fluid flows under macroscopically identical external
conditions, can provide a turbulence theory.
The theory o f turbulence by its very nature cannot be other than
statistical, i.e., an individual description of the fields o f velocity,
pressure, temperature and other characteristics of turbulent flow is in
principle impossible. Moreover, such description would not be useful

even if possible, since the extremely complicated and irregular nature
o f all the fields eliminates the possibility of using exact values o f
them in any practical problems. As a result, the contrasting o f the
“semiempirical” and statistical theories of turbulence often found in
the literature is meaningless. The semiempirical theory, of course, is
also statistical and differs from other theories o f turbulence not by
rejecting the use of statistical characteristics, but only in the means
used to determine them.
In discussing the statistical nature of the theory of turbulence,
comparison is often made with the kinetic theory of gaseS, which
investigates many-particle systems o f interacting molecules. This
comparison is reasonable in the sense that in both these theories an
exact description of the evolution of the individual mechanical
system is theoretically impossible and practically useless. However, it
must be remembered that between the statistical mechanics of the
ensembles of molecules investigated by Gibbs, Boltzmann and others,
and the statistical fluid mechanics of a viscous fluid, there is a
fundamental difference. This is connected primarily with the fact
that the total kinetic energy of a molecular ensemble is time-
invariant (in any case, under the simplified assumptions on molecular
interactions adopted in the kinetic theory o f gases) while the kinetic
energy of real fluid flow is always dissipated to heat due to the
action o f viscosity. Of less importance, but still significant, is the fact
that molecular ensembles are discrete by nature and their time
evolution is described by systems of ordinary differential equations,
while in fluid mechanics we are dealing with the motion of a
continuous medium, which is described by partial differential
equations. Thus the analogy with the kinetic theory of gases is of
relatively little help in the formulation of the theory of turbulence,
and is useful only for a preUminary understanding of the concept of
a statistical approach to physical theory.
Far more fruitful, perhaps, is the analogy between the theory of
turbulence and quantum field theory, which is connected with the
fact that a system of interacting fields is also a nonlinear system with
a theoretically infinite number o f degrees o f freedom. From this
follows the similarity o f the mathematical techniques used in both
theories. This allows us to hope that the considerable advances in the
one wall also have a decisive effect on the development o f the other.
However, at present, quantum field theory is encountering great
difficulties connected with its first principles, while in the theory of
INTRODUCTION 5
turbulence there are only partial successes; consequently, the major

influence of one o f these theories upon the other still lies in the
future.
In spite o f the fact that the fluid flows encountered in nature and
in technical devices are, as a rule, turbulent, in all existing courses on
fluid mechanics, at best only a few sections are devoted to the theory
o f turbulence. These sections contain usually only some disjointed
remarks on the methods o f statistical description o f disordered fluid
flows and on some statistical characteristics o f such flows. The
monograph literature devoted to turbulence is also very poor and
amounts in all to only a few titles (almost all of these may be found
in the bibliography at the end o f this book); moreover, a great many
o f these refer to books of relatively narrow subject matter. It is not
difficult to understand why the situation is so unsatisfactory.
Turbulent flows are considerably more complicated than laminar
flows and require essentially new methods for their study. These
methods differ from the classical methods o f mathematical physics,
which for almost two centuries have been considered as the only
ones suitable for the quantitative study o f the laws of nature. The
mathematical techniques needed for the logically accurate formula
tion o f the statistical mechanics of continuous media, i.e., the theory
of random fields, was devised only in the last 25-30 years and is still
almost unknown outside a small group o f experts in probability
theory. It was during these years also that the modern theory of
turbulence was formulated (which even now is still far from being
complete). However, we feel that the existing achievements in this
field certainly deserve a considerably high place in the necessary bulk
of knowledge o f every scientist studying fluid mechanics and
theoretical physics, and if this has not yet occurred it is only because
the theory o f turbulence is relatively new. We are convinced that in
the future the place o f this theory in textbooks and fluid mechanics
curricula for colleges and universities, in the education plans of all
specialists in fluid mechanics and theoretical physics, and in various
research projects, will rapidly increase. If our book can contribute in
some measure to these developments we shall indeed be happy.
* * *
Let us now turn to a brief history o f turbulence. We shall

enumerate the methods and results, the detailed description o f which
constitutes the fundamental part o f this book. Simultaneously, we
shall try to explain why it seems worthwhile to us to write such a

book now at the present stage o f development of statistical fluid
mechanics.
The existence of two sharply different types o f flow, now called
laminar and turbulent, had already been pointed out in the first half
o f the nineteenth century. However, a theory of turbulence came
only with the outstanding pioneering works of Osborne Reynolds
(1883; 1894). In these works Reynolds turned his attention first to
the conditions under which the laminar flow of fluid in pipes is
transformed into turbulent flow. The study of these conditions led
him to a general criterion for dynamic similarity o f -flows of a
viscous, incompressible fluid. In the absence of external forces, and
with geometric similarity, this criterion is the coincidence o f the
values o f the so-called “Reynolds number” Re = ULIv, where U and
L are characteristic scales of velocity and length in the flow and v is
the kinematic viscosity o f the fluid. From the dynamical viewpoint,
the Reynolds number may be interpreted as the ratio o f typical
values o f the inertial and viscous forces acting within the fluid. The
inertial forces which produce mixing o f the different volumes of
fluid moving inertially with different velocities, also produce a
transfer of energy from large- to small-scale components o f motion
and hence assist the formation in the flow o f sharp, small-scale
inhomogeneities that characterize a turbulent flow. The viscous
forces, on the contrary, assist in the smoothing out of small-scale
inhomogeneities. Thus we may expect that flows with sufficiently
small values o f Re will be laminar, while those with sufficiently large
Re will be turbulent. This is a fundamental result which Reynolds
formulated.
Reynolds also contributed another im portant step in the theory of
turbulence. He proposed representing the values of all the hydro-
dynamical quantities in turbulent flow as sums o f mean (regular) and
fluctuating (irregular) components, giving up the practically hopeless
attempts to describe the details of individual hydrodynamic fields,
and studying only the mean values which vary comparatively
smoothly with space and time. For the determination of the mean
values, Reynolds proposed to use time-averaging or space-averaging,
but in fact he used only certain algebraic properties of the operation
of averaging. These properties permit considerable simplification in
application to the equations o f fluid mechanics. Consequently,
although nowadays when investigating turbulence we understand
averaging in a different sense from that o f Reynolds, all his
INTRODUCTION 7
deductions still remain valid, since the properties o f averaging which

he used proved to agree completely with the present understanding
o f the operation.
Here we will explain how averaging is now understood in the
theory of turbulence. In present-day statistical fluid mechanics, it is
always implied that the fluid mechanical fields of a turbulent flow
are random fields in the sense used in probability theory. In other
words, every actual example of such a field is considered as a
“sample” taken from the “statistical ensemble of all possible fields.”
Such a statistical ensemble is described by a given probability
measure on the set of functions o f space-time points, which satisfy
the necessary kinematic and dynamic conditions (arising from the
laws of fluid mechanics). The averaging o f any fluid mechanical
quantities may be understood then as probability averaging with
respect to the corresponding statistical ensemble. Moreover, all the
properties o f the operation of averaging which Reynolds required
foUow from the well-known properties o f the mean value of
probability theory (i.e., the mathematical expectation) which are
described in probability textbooks. This immediately eliminates
many difficulties inherent in the application of time- or space-
averaging. O f course, the interpretation in the real world of the
results o f the formal theory is (in such a formulation) found to
require the use o f certain assumptions of ergodicity, but this
situation is common to all applications of statistical physics.
In the preceding paragraph we departed from the chronological
principle of describing the fundamental stages in the formation of
the theory o f turbulence. However, before returning to the historical
survey, let us give the present-day formulation of the general
problem o f the statistical description of turbulent flows (or, in short
the “problem of turbulence” ). For simphcity, we shall confine
ourselves to the case o f an incompressible fluid. In this case the flow
is determined completely by a solenoidal velocity field (i.e., a
velocity field without divergence) u{x, t) = [ui{x, t ) , ii2 {x, t),
«3 (x, ^)} where «i, « 2, are components of velocity, at is a point o f
space, t is the time (the pressure field may be expressed in terms o f
the velocity field with the aid of the equations of motion). The
problem of turbulence is reduced here to finding the probability
distribution P(rfco) in the phase space of turbulent fiow Q = {co}, the
points (0 of which are all possible solenoidal vector fields «(Jf, t)
which satisfy the equations o f fluid mechanics and the boundary
conditions imposed at the boundaries of the flow. In many cases a
narrower (i.e., a less complete) formulation of the problem o f

turbulence also proves useful; here, only synchronous values o f the
fluid mechanical quantities (i.e., those occurring at the same instant
o f time) are considered. With this approach, the problem of
turbulence consists o f finding a one-parameter family o f probability
distributions Pt (dco) in the phase spaces Q< = {co}, the points w o f
which are all possible solenoidal vector fields u(x) which satisfy the
corresponding boundary conditions (these boundary conditions may
be time-dependent, and as a result, we introduce subscript ^ into the
notation Q< for the phase space). The time-dependence o f the
distributions P((do)) reflects the evolution of the velocity field in
accordance with the equations o f fluid mechanics. In fact, if we write
the solution o f these equations symbolically in the form u(x, /) =
T/U(x, 0 ) where Tg is the corresponding nonlinear operator, then
obviously for any measurable set A o f the space the equality
Pi (A) = Po(T_tA) holds where /I belonging to Qois the set o f all
initial velocity fields u(x, 0 ) which under the action of the operator
Ti are transformed into fields u<x, t) belonging to the set A.
Consequently, we see that the family o f probability distributions
Pt (da) is defined in principle uniquely by the given initial distribu
tion Po{d(iy). As a result, in the formulation under discussion, the
problem o f turbulence is a problem of the evolution o f the
probability measure in a functional space with given initial condi
tions.
A complete determination o f the probability distribution P(dco)or
at least P, (rfco) is extremely difficult, and at present is far from being
solved. However, in many practical cases it is by no means necessary
that the problem o f turbulence be solved completely. In most
practical cases it is sufficient to determine only some o f the simplest
numerical characteristics o f the probability distributions for the fluid
mechanical fields o f a turbulent flow, for example, the mean values
of the velocity and pressure at different space-time points, the
second moments of the fluctuations o f the fields at a given point
(e.g., the variances o f the velocity fluctuations which describe the
intensity o f turbulence or the second moments of velocity com
ponents and temperature which express the components o f the
turbulent fluxes o f mom entum and heat), and the correlation
coefficients between the fluctuations o f the fluid mechanical fields at
two points o f space-time (describing the space-time statistical
coherence or the structure o f the hydrodynamical fields of turbulent
flow). Thus it is natural for the majority of investigations in the
INTRODUCTION 9
theory o f turbulence to be devoted not to the problem of turbulence

as a whole, but to attempts to determine (although approximately)
some simplified characteristics o f the probability distribution for
fluid mechanical fields of turbulent flow. These investigations led to
a number o f valuable results on the physical nature and properties o f
turbulence and give satisfactory answers to many im portant practical
questions.
Returning now to the short outline o f the history o f the theory of
turbulence, we must mention first the interesting work o f Geoffrey
Taylor (1921) on turbulent transport in which the im portant role of
the correlation functions (i.e., the mixed second moments) o f the
velocity field was first demonstrated (not for a simple Eulerian
velocity at a given point, but for the more complicated Lagrangian
velocity o f a given fluid particle). However, in general, the idea that
the correlation functions and other statistical moments o f fluid
mechanical fields must be recognized as the fundamental character
istics o f turbulence was first stated by L. V. Keller and A. A.
Friedmann (1924), who proposed a general method of obtaining
(using the equation o f motion of a real fluid) the differential
equations for the moments o f arbitrary order of fluid mechanical
fields o f turbulent flow. The determination o f all such moments
(given some general assumptions) is equivalent to the determination
of the corresponding probability distribution in the functional space
P(dio) or Pt{dZ), i.e., the solution o f the problem of turbulence.
Therefore, the total infinite Friedmann-Keller system o f equations
for all possible moments gives an analytical formulation of the
problem o f turbulence. But this system of equations is very
complicated; any finite subsystem of this system is always nonclosed,
i.e., it contains more unknowns than there are equations in the given
subsystem (the impossibility o f obtaining a closed system of
equations for a finite number of moments is a direct consequence o f
the nonlinearity of the equations o f fluid mechanics). Thus when we
study the appUcation of the Friedmann-Keller method to a finite
number o f moments, there arises the closure problem of the
equations for the moments. This problem is in many ways analogous
to the closure problem o f a chain of equations for multiple
distribution functions in the kinetic theory o f gases.
It may be said that most o f the theoretical work on the dynamics
of turbulence has been devoted (and still is devoted) to ways o f
overcoming the difficulties associated with the closure problem. So
far, this has not been completely possible. Nevertheless, many
im portant results of great practical value have been obtained in the

theory o f turbulence, in two roundabout directions: the first,
devoted to the description of large-scale components o f turbulence
(the scale is comparable to the scales o f the overall flow); and the
second, to the description o f small-scale components. The basic
difference in behavior between these two types o f components lies in
the fact that the large-scale characteristics of turbulence depend
considerably on the geometry o f the boundaries o f flow and the
nature o f the external forces and hence will be very different for
different types o f flow. The small-scale characteristics, to a large
extent, possess a universal character.
The large-scale components make the fundamental contribution to
the transfer o f momentum and heat through a turbulent medium,
and thus their description is necessary for im portant engineering
problems such as the calculation o f drag and heat transfer in the fluid
flows past rigid bodies. Therefore, it is natural th at in the
development o f the theory o f turbulence attention was first given to
the study o f the large-scale components. The urgent needs of
technology demanded the carrying out o f a large am ount o f
experimental research into the properties of the large-scale com
ponents o f turbulence in the case of flows in pipes, channels,
boundary layers and in free turbulent flows (jets, wakes, etc.). On
the basis o f these experiments “semiempirical theories o f turbu
lence” were formulated. These theories allow the experimental data
to be systematized and used to predict the results o f subsequent
similar experiments.
The formulation of the semiempirical theories of turbulence was
an im portant advance in the development o f statistical fluid
mechanics. This step started as early as the second decade of this
century, but it came to full fruition in the following two decades
(1920’s and 1930’s). However, the possibilities o f the semiempirical
method have not yet been fully exhausted, and valuable work is still
going on in this field. Decisive advances in the development o f the
semiempirical approach to the theory o f turbulence were made by
Geoffrey Taylor (1915; 1932), Ludwig Prandtl (1925) and Theodor
von Karman (1930).
The semiempirical theories o f turbulence are based on the analogy
between turbulence and molecular chaos. Its fundamental concepts
include the mixing length (analogous to the mean free path o f
molecules), the intensity o f turbulence (analogous to the root-mean-
square velocity of molecules) and the coefficients o f turbulent
INTRODUCTION 11
viscosity, thermal conductivity and diffusion. Using the stated

analogy it is postulated that a linear dependence exists between the
turbulent stress tensor and the tensor of the mean rate of strain and
also between the turbulent flux of heat (or passive admixture) and
the mean temperature gradient (or gradient o f the admixture
concentration). These assumed relationships are supplemented by
certain hypothetical laws which are established in general form with
the aid of qualitative physical considerations, or simply by “guess
ing” from considerations of simplicity. These assumptions (or some
simple corollaries drawn from them ) are verified against empirical
material, and at the same time, values o f the undefined constants
occurring in the semiempirical relationships are determined. If the
results obtained prove satisfactory, the deductions drawn from them
are extended to the whole class o f turbulent flows related to the flow
to which the empirical data chosen for the verification actually refer.
The semiempirical theories of the 1920’s and 1930’s usually con
sidered only the simplest statistical characteristics of turbulence. As a
rule, the hypotheses adopted in these theories permit the closure of the
first equations of the infinite Friedmann-Keller system—the so-called
Reynolds equations, containing only single-point first and second mo
ments of the fluid mechanical field. An im portant role in the semi
empirical theories is also played by symmetry considerations and the
use of certain simplified hypotheses of similarity (which are, in particu
lar, an essential part of all semiempirical theories of turbulent jets and
wakes). However, the similarity hypotheses based on real physical ideas
of the mechanism of turbulence do not form a unique basis for these
theories and are always supplemented (sometimes even without any
real necessity) by assumptions of a more specialized and artificial
nature. Thus, for example, one of the most im portant deductions of
the semiempirical theory is that of the universal logarithmic law for
the profile of mean velocity in pipes, channels and boundary layers
on a flat plate. (Universal here means that the law holds for all
sufficiently large Reynolds numbers.) At present, it is known that
this law may be derived from a single similarity hypothesis about the
probability distribution of the various quantities in turbulent flows
in a half-space or from dimensional considerations based on natural
assumptions concerning the physical quantities which determine the
turbulent regime in this case. Nevertheless, in semiempirical theories
this result was always justified by some special hypotheses, and,
unfortunately, such artificial deductions even now are predominant
in textbooks on fluid mechanics.
The semiempirical theories of turbulence are valuable for solving a

number o f im portant practical problems. However, the hypotheses
adopted in these theories frequently have no reliable physical
foundation and contribute little to the understanding of the physical
nature o f turbulence. The theory of the universal steady statistical
regime o f the small-scale components o f turbulence for very high
Reynolds numbers, however, is a completely different matter. This
theory is an immediate consequence of the new similarity hypotheses
for small-scale components proposed by A. N. Kolmogorov (1941a,
b) [the same conclusions were reached by A. M. Obukhov (1941)
based on a special model of the physical processes affecting the
evolution o f these com ponents]. The formulation of this theory was
another great step in the development of statistical fluid mechanics.
Before proceeding to a description of the contributions to the
theory of turbulence made by Kolmogorov and Obukhov, it is
necessary to mention in this historical survey two of their pred
ecessors. One of these was the English scientist, Lewis Richardson
(1922; 1926), and the other was Geoffrey Taylor (1935a) whom we
have mentioned twice before. In his outstanding paper, published in
1922, and in some other papers, Richardson put forward some
penetrating ideas on the physical mechanism of turbulent mixing in
the case o f large Reynolds numbers (of which Httle notice was taken
at the time). According to his assumptions, developed turbulence
consists o f a hierarchy o f “ eddies” (i.e., disturbances or nonhomo
geneities) o f various orders. Here, the “eddies” of a given order arise
as a result o f the loss o f stability of larger “eddies” of the preceding
order, borrowing their energy, and, in their own turn, losing their
stability and generating smaller “eddies” of the following order to
which they transmit their energy. Thus there arises a peculiar
“cascade process,” of breaking-down of eddies in which the energy
of the overall flow is transmitted to motions of smaller and smaller
scale, down to motion of the smallest possible scale, which is stable.
To be stable, these extremely small-scale motions must be character
ized by a sufficiently small Reynolds number. Thus it follows that
viscosity will play an im portant role and, consequently, there will be
considerable dissipation o f the kinetic energy into heat. The
corresponding physical picture of developed turbulence is expressed
neatly in the following rhyme, which first appeared on page 6 6 of
Richardson’s book (1922) and has since been often quoted (usually
without the exact reference and the last line):
INTRODUCTION 13
Big whorls have little whorls,

Which feed on their velocity;
And little whorls have lesser whorls,
And so on to viscosity
(in the molecular sense).
Richardson put forward these general ideas only in qualitative

form and did not make any deductions that could be formulated in
the precise language of mathematics. However, his intuition was so
powerful that in his paper o f 1926 he was, nevertheless, able to
establish by purely empirical means one of the general quantitative
laws o f small-scale turbulent motion, which follows in fact from the
mathematical theory of the cascade process of breaking-down of
eddies. This law consists o f the fact that the effective diffusion
coefficient for a cloud of admixtures in a developed turbulence is
proportional to the four-thirds power o f the characteristic scale-
length o f the cloud. In 1941, when Kolmogorov and Obukhov
formulated the general quantitative theory of small-scale components
o f turbulence, Richardson’s “four-thirds law” was actually the only
empirical result which had indicated the existence o f some simple
general rules controlling the small-scale structure o f turbulence.
Taylor’s work also played a large part in the theory of small-scale
turbulent motion. G. I. Taylor (1935a) introduced the concept of
“homogeneous and isotropic turbulence,” determined by the condi
tion that all the finite-dimensional probability distributions o f the
fluid mechanical quantities at a finite number of space-time points
are invariant under any orthogonal transformations (shifts, rotations
and reflections) o f a system o f three-dimensional coordinates.
Homogeneous and isotropic turbulence is a special case o f turbulent
flow in which the structure o f the statistical moments o f the fluid
mechanical fields and the form of the corresponding Friedmann-Keller
equations is extremely simple. In this simplified case, all the
theoretical difficulties connected with the problem o f closure o f the
Friedmann-Keller equations still apply. However, the corresponding
equations are far more amenable to mathematical analysis than the
general equations corresponding to arbitrary turbulence, and by using
them it is possible to obtain many specific results which explain
individual aspects o f turbulent flow.
In itself, the model of homogeneous and isotropic turbulence is
not suitable for describing any real turbulent flows, since the
assumptions of homogeneity and isotropy are not fulfilled for real
flows (three-dimensional homogeneity assumes, in particular, that

there are no boundaries in the flow and that the mean velocity is
strictly constant). However, the mathematical technique of homo
geneous and isotropic turbulence, after certain generalizations,
proves very valuable for describing the properties of small-scale
components o f real turbulent flows. This is because the statistical
regimes o f these components as, following Kolmogorov, we will
explain later, may be taken quite naturally to be homogeneous and
isotropic. In other words, any developed turbulence with sufficiently
high Reynolds number may be considered to be locally homogeneous
and locally isotropic, which immediately simplifies the mathematical
investigation.
We shall now consider the fundamental ideas of Kolmogorov’s
“theory of locally isotropic turbulence.” We shall use this compara
tively short expression rather than the longer one given above. First,
Kolmogorov made an im portant addition to the assumptions on the
cascade process o f transfer of energy from large-scale components
(which obtain their energy immediately from the mean flow) to
components of smaller and smaller scale, by noting that due to the
chaotic nature o f this transfer of energy, the orienting effect of the
mean flow must be weakened with each breaking-down. Conse
quently, for sufficiently small-scale components of turbulence (i.e.,
for sufficiently large “order numbers” ) this orienting effect will
produce no results at all. In other words, in spite o f the fact that the
mean flow and the largest-scale nonhomogeneities of any real
turbulent motion are, in general, nonhomogeneous and anisotropic,
the statistical regime o f sufficiently small-scale fluctuations of any
turbulence with very high Reynolds number may be taken to be
homogeneous and isotropic. Moreover, it is natural to expect that the
characteristic periods of different orders will be smaller, the greater
their order. Therefore, for fluctuations o f sufficiently high order
(i.e., sufficiently small spatial scales), these periods will be much
smaller than the time of perceptible variation of the mean flow.
Therefore, the regime o f such fluctuations will be quasi-steady-
practically steady over an interval o f time containing a large number
o f characteristic periods.
Thus, we see that for sufficiently small-scale fluctuations, a
homogeneous, isotropic and practically steady statistical regime will
prevail, characterized by the presence o f a mean flux of energy e to
the largest o f the fluctuations under consideration and the dissipa
tion o f energy (equal to this flux) as heat under the action of the
INTRODUCTION 15
viscosity, which is concentrated principally in the range of minimum

scale fluctuations. Arguing from this, Kolmogorov formulated the
hypothesis that the statistical regime o f sufficiently small-scale
components of velocity of any turbulence with sufficiently large
Reynolds number will be universal and determined by only two
dimensional parame_ters—the mean rate o f dissipation o f energy (per
unit mass o f fluid) e and the coefficient of viscosity v. Hence, with
the aid of simple dimensional considerations, we may deduce that
the scale of the greatest fluctuations on which the viscosity still
exerts a considerable effect must be o f the order 77 =
Consequently, it is natural to assume that when the range of
fluctuations subject to the universal statistical regime under discus
sion extends to scales much greater than t) , then there must exist an
“inertial subrange” of scales (many times smaller than the typical
length-scale o f the whole flow L, but many times greater than t)), in
which the viscosity will no longer play any part, i.e., the statistical
regime will be determined by a single parameter e. This assumption is
Kolmogorov’s second main hypothesis.
These hypotheses o f Kolmogorov allow a number of concrete
deductions to be formulated on the statistical characteristics of the
small-scale components of turbulence. The most im portant of them is
the “two-thirds law” (deduced by Kolmogorov) according to which
the mean square o f the difference between the velocities at two
points o f a turbulent flow at a distan^ce r apart, when r lies in the
“inertial subrange,” is equal to C{zryi\ where C is a universal
numerical constant. Another form o f this assertion [first put forward
by Obukhov (1941)] is the “five-thirds law,” according to which the
spectral density o f the kinetic energy o f turbulence over the
spectrum o f wave numbers k in the inertial subrange has the form
where Ci is a new numerical constant (which is simply
connected with C ). There are also many other consequences of these
hypotheses, but we shall not dwell on them here.
The work of Kolmogorov has served as a basis for all subsequent
developments o f the theory o f the local structure of turbulence and
its applications in the 1940’s and 1950’s. During this period
investigations were made concerning the local structure o f not only
the velocity field, but also the fields o f concentration of passive (i.e.,
dynamically neutral) admixtures, temperature (including the case of a
thermally stratified heavy fluid in which, due to the effect of
buoyancy, the temperature can no longer be considered as a
“passive” admixture), pressure, and turbulent acceleration. The
deductions obtained concerning the statistical properties of small-

scale components of turbulence have found an application in other
problems, for example, on the relative dispersion of particles and the
breakdown of drops in a turbulent medium, on the generation of
wind waves on the surface of the sea, on the generation of a magnetic
field in a turbulent flow of conducting fluid and the structure of
inhomogeneities in the electron density in the ionosphere, on
fluctuations of the refractive index in the atmosphere and the
scattering and fluctuations o f electromagnetic waves in the turbulent
atmosphere, and to many other interesting questions.
The consequences of Kolmogorov’s theory, first and foremost the
“ two-thirds law” and “ five-thirds law” given above, were tested
many times during the 1940’s and 1950’s on results of measurements
in real turbulent flows. However, it was found that in laboratory
experiments (normally carried out in wind-tunnels) the Reynolds
number is not large enough for a perceptible “inertial subrange” to
exist in the turbulence spectrum, and, consequently, the results of
wind-tunnel measurements carried out over 2 0 years cannot be used
to verify these laws. Measurements in nature, on the other hand,
where the Reynolds number has, as a rule, far higher values than
those observed in laboratory conditions, have until recently given
results with considerable statistical scatter. Consequently, although
the general ensemble o f experimental data unquestionably supported
the theory, it had still not been verified in a completely direct
manner, and reliable estimates could not be made for the numerical
parameters occurring in the theory. Only in the 1960’s has there
been any real change in the situation, and now several researchers
have made extremely precise measurements of the characteristics of
turbulence in various natural and artificial turbulent flows with very
large Reynolds numbers. The results of these measurements are in
excellent agreement with each other, and furnish a definite confirma
tion of the correctness o f the theory, finally allowing the constants C
and Cl to be determined with sufficient accuracy.
At present, strong arguments exist for considering the theoretical
treatm ent and experimental verification of the idea of a universal
statistical regime of small-scale components o f velocity as being
basically completed. However, for the further development o f the
theory of turbulence, we require essentially new ideas, the search
for which we shall describe a little later. Thus it is appropriate now
to sum up the results of this im portant period in the development of
the theory o f turbulence connected with the elaboration of this idea.
INTRODUCTION 17
The attem pt to carry out this task to some extent is one of the
purposes o f our book.
There are certain works which deal with the search for new
avenues o f approach in the development of statistical fluid mechan
ics. In the papers of A. N. Kolmogorov (1962) and A. M. Obukhov
(1962) delivered a week apart at two international conferences on
the theory of turbulence held in Marseilles in the early autumn 1961,
a method is proposed for further refinement of the basic description
of the local structure of turbulence with large Reynolds numbers. The
principle o f the method is as follows: as we have seen in the
theoretical discussion above, the statistical characteristics of small-
scale turbulence are assumed to depend only on the mean value s , of
the rate o f dissipation of energy. In fact, however, the field of
dissipation o f energy also undergoes disordered fluctuations, and
there is reason to think (both theoretically and purely empirically)
that its range of variation may be extremely large. The statistical
characteristics of this field evidently may depend on the peculiarities
of the large-scale flow: in particular, we must expect the variance of
the field of £ to increase with the increase of Re. Therefore, the
statistical characteristics o f small-scale turbulent motion, defined by
the single quantity z must be interpreted only as conditional mean
values, obtained on the condition that e is strictly fixed (and equal to
e). However, the unconditional mean values which are obtained by
averaging the results of calculations with fixed e over the fluctuations
in value o f this parameter will, generally speaking, differ from the
conditional mean values (and may themselves be shown to be not
universal, i.e., they are different for different types of large-scale
flows, and depend, in particular, on Re). The works of Kolmogorov
and Obukhov were devoted to methods of estimating the effect of
this fact. To obtain specific quantitative results in this way, we first
need to discover the statistical properties of the field of the
dissipation of energy s(at, 0 , i e., in other words, we require a more
detailed study o f the mechanism of dissipation of energy in turbulent
flow. Preliminary estimates show that the corrections to the
“two-thirds” and “ five-thirds” laws obtained by taking into account
the fluctuations of the rate of energy dissipation will be close to the
order of accuracy o f the best experimental data presently available.
We must also note some special variants of closure hypotheses for
the equations of turbulence, which were developed in "a series of
papers by R. Kraychnan (1959; 1962b and others) based on the
assumption that direct interactions between triads of space Fourier
components o f the velocity field play a considerably greater part

than do the indirect interactions (by way of all the other Fourier
components). We should also mention the method of describing
large-scale components o f turbulence proposed by W. Malkus
(1954b) [see also Townsend (1962b) and Spiegel (1962)] based on
the use o f the hypothetical variational principal of “maximum
dissipation” and representing hydrodynamical fields in the form o f a
superposition o f a finite number of corresponding characteristic
functions. In recent years, both these approaches have evoked many
controversies—they lead to some interesting and likely results, but
they are based on unverified hypotheses and involve a number of
difficulties. Thus, for example, the first approximation o f Kraich-
nana’s theory leads to equations for the turbulence spectrum in the
small-scale range which do not agree with the predictions of
Kolmogorov’s theory (which have been substantiated excellently)
and indicate an unspecified dependence between the statistical
characteristics of the small-scale components and the mean square
velocity which characterizes the large-scale, energy-containing eddies.
Malkus’ theory for free convection leads to results which disagree
with deductions drawn from the natural similarity theory for the
turbulence in a stratified medium (and indicate an unexpected
dependence of the large-scale components of convective flow on the
molecular thermal conductivity o f the fluid). The same theory,
applied to flow between parallel plates, leads to a very disputed
conclusion concerning the dependence of the large-scale character
istics on the molecular viscosity, however large the Reynolds
number, and to a departure from the generally accepted (and well
supported by experiment) universal logarithmic law for the mean
velocity profile. Therefore, at present, it is still impossible to say
whether or not these approaches (or some simple modifications o f
them) will be fruitful for the development of the theory of
turbulence.
Finally, mention must be made o f the prospects of solving the
general problem o f turbulence which are connected with the use of
the techniques o f “ characteristic functionals” o f fluid mechanical
fields. These characteristic functionals define uniquely the proba
bility distributions P(dm) or Pi(dco) in the phase space of turbulent
flow, and hence finding them would give a complete solution o f the
problem o f turbulence. In the work of Eberhard Hopf (1952) an
equation is deduced for the characteristic functional of a turbulent
velocity field in an incompressible field. This is an equation in
INTRODUCTION 19
functional derivatives, and its most remarkable feature is linearity.

Thus, although the dynamics o f a fluid is nonlinear, the fundamental
problem o f statistical fluid mechanics, formulated in terms of
characteristic functionals, proves to be a linear problem. We should
also note that H o p f s equation is formally similar to the Schwinger
equations of quantum field theory, which are equations in functional
derivatives for the Greens function o f interacting quantum fields (we
have already mentioned the strict analogy between the theory of
turbulence and quantum field theory).
The solution o f H o pfs equation has encountered considerable
difficulties, first, because it is still not clear exactly how specific
problems for this equation must be formulated, and, second, because
of the absence at present o f any general methods o f solving equations
in functional derivatives (or even of general results on the existence
and uniqueness o f such solutions). In recent years the new
mathematical techniques of function space integrals (i.e., of integrals
of functionals over some function space) has attracted a great deal of
attention in connection with the theory o f equations in functional
derivatives. Today, it is formally possible to write down a solution of
H op fs equation in the form o f a specific function space integral with
respect to some “generalized measure” in a function space (which
does not possess some properties of the usual measures of measure
theory, and hence calls to mind the notorious “ Feynman measure”
arising in quantum mechanics and quantum field theory). However,
writing down this solution remains a purely formal process, and is of
little assistance in the effective formulation and study o f the required
specific solutions.
***
We conclude this introduction by discussing briefly the contents

o f this book. It is, of course, completely impossible in one or two
volumes to exhaust the whole range o f questions connected with the
problem o f turbulence. (We may note that the book by Chandrase
khar (1961), which in fact deals almost entirely with the material o f
our Sects. 2.6 and 2.7 and its generalizations only, contains
approximately 700 pages!) We shall not even attem pt to be
exhaustive but will select the material which, in our opinion, assists
our understanding of the physical nature of turbulence. Conse
quently, we shall not dwell on specific engineering applications or
mathematical details connected with the calculation of the
20 STATISTICAL FLU ID MECHANICS
complicated statistical characteristics. We shall confine ourselves

throughout to only the simplest flows and simplest problems. Thus
we shall deal with flows in straight circular tubes only; also, we con
sider boundary layers only along a flat plate and in the absence o f a
pressure gradient in the ambient flow; diffusing particles are taken
only to be points and to have the same specific gravity as the
surrounding medium, etc. Further, we shall not discuss problems o f
turbulent flows o f a conducting fluid in a magnetic field. The
discussion of compressibility occupies only one section in the
chapter on isotropic turbulence, so as to enable us to demonstrate
some o f the new physical effects connected with it. However, we
include a chapter on turbulence in a medium with vertical density
stratification (for the simplest case o f a two-dimensional flow in a
half-space z > 0 ) since the effect o f stratification on turbulence has a
clear physical character and may be investigated with the aid of
general methods that are used widely in other parts of the book.
The inclination to make the book as “physical” as possible also
dictated the choice o f methods of investigation. The equations of
turbulent motion are always unclosed (containing more unknowns
than equations) and hence problems of the turbulence theory cannot
be reduced directly to finding a unique solution of some differential
equation (or equations) determined by known initial and/or boun
dary values. Under these conditions, it is inevitable that, in addition
to the equations of motion, some additional considerations be
introduced. In our opinion, among all these additional considera
tions, it is the considerations o f similarity (based on the invariance of
the problem under some group o f transformations) which make the
most obvious physical sense as well as dimensional considerations
(based on the isolation of physical quantities which effect the
turbulent flow under investigation). We have tried to consider most
explicitly the deductions from the ideas of dimensionality and
similarity which may be used in the theory of turbulence far more
widely than is normally proposed. The semiempirical theory of
turbulence which employs more specialized hypotheses occupies
relatively little space in this book; our description of the “ classical”
applications o f the semiempirical theory to flow in tubes, channels
and boundary layers is especially b rief (This subject is discussed in
detail in the monographs of S. Goldstein (1938), L. G. Loitsyanskiy
(1941) and H. Schlichting (1960), together with the semiempirical
theories o f “ free turbulence” which are omitted from our book
entirely.) However, we have included several comparatively new and
INTRODUCTION 21
less-known applications o f the semiempirical theories and considered

a number of applications of the semiempirical theory of turbulent
transport (which normally is the only theory considered in discus
sions of the problem of the spreading of admixtures in turbulent
flows). Moreover, we have made a detailed analysis o f various
hypotheses on the transfer o f energy through the spectrum of
isotropic turbulence, similar in character to the semiemperical
hypotheses, but which throw some light on the physical mechanism
of turbulent mixing and which previously have not been discussed in
full.
The great attention paid in this book to similarity and dimension
ality is also conditioned by the fact that Kolmogorov’s theory of
locally isotropic turbulence (which is based entirely on these
methods) is given a great deal of space here. We have already noted
that one of the principal incentives for writing this book was a desire
to summarize the development of the idea of a universal local
structure in any turbulent flow for sufficiently large Reynolds
number. We have also discussed the improved version of this idea put
forward by Kolmogorov and Obukhov in 1961, together with
additional data on this question obtained later. We also consider
briefly the ideas of Kraichnan on small-scale features of turbulence.
But the controversial theory of Malkus which contradicts the results
of some recent calculations [see Reynolds and Tiederman (1966)],
except for the historical outline in the Introduction, will only be
mentioned briefly.
While wishing to present a systematic description of the theo
retical principles o f statistical fluid mechanics, we did not want to
give our book a formal mathematical character, and have tried
throughout to reinforce the theoretical deductions by the analysis of
experimental data. The combination o f theoretical and experimental
approaches, which is extremely fruitful in all investigations of natural
phenomena, is especially necessary in statistical fluid mechanics
where the theory is often still of a preUminary nature and is almost
always based on a number of hypotheses which require experimental
verification. However, we have avoided introducing experimental
results which have no theoretical explanation and which do not serve
as a basis for some definite theoretical deductions, even if these data
are in themselves very interesting or practically important. As a
source of experimental data on turbulence, in this book we have very
frequently used atmospheric turbulence. To a certain extent, this is
connected with the fact that both o f us have worked in the Institute
of Atmospheric Physics for many years and are far better acquainted
with atmospheric turbulence than with other types of turbulent
flow. However, in addition, there are weighty reasons which have
turned our attention to this type of turbulence. The fact is, that the
atmosphere, which von Karman himself (1934) called “ . . .a giant
laboratory for ‘turbulence research,’ ” possesses very valuable proper
ties which make it especially suitable for the verification of the
deductions o f modem statistical theory. We have already observed
that atmospheric motion is usually characterized by far larger
Reynolds numbers than flows created in the laboratory, and
therefore is far more convenient for investigating specific laws
relating to the case o f very large Re. Moreover, the geometrical
conditions o f atmospheric turbulence (namely, the conditions of a
two-dimensional flow in a half-space bounded by a rigid wall,
referring to the wind in the surface level of the atmosphere, where in
many cases the “wall” may be considered as plane and homogeneous;
or the conditions of flow in an infinite space, which give a good
description of the air motion in the free atmosphere) are simpler
than in most laboratory experiments. The only additional comphca-
tion, which arises on transition from laboratory to atmosphere, is the
necessity of taking into account the thermal stratification, but, as we
have already observed, this complication leads to additional interest
ing theoretical discussions and extends the number of laws observed
experimentally and permits physically based explanations.
The widespread nature of turbulent flows, their great significance
for a number o f diverse engineering problems, and their interest to
theoreticians, has led to the fact that the literature on the subject is
enormous—it numbers many tens o f thousands of papers, scattered
throughout a vast number of physical, mathematical, mechanical,
chemical, meteorological, oceanographical, and engineering journals.
Unfortunately, however, the theory of turbulence is extremely
difficult, and until now it has not been advanced very far; therefore,
many of the works referred to are purely empirical, or contain only
the very beginnings of the theory, or are even erroneous. These facts,
of course, greatly complicate any survey of the literature of
statistical fluid mechanics. In this book we originally proposed to
confine ourselves to the minimum number of necessary references;
however, having spent a great deal of time on the study of the
literature (without which it would have been impossible to decide
what was necessary), we decided that it would be more expedient
not to limit the references, especially to recent works. We have done
INTRODUCTION 23
this to give the reader a sufficiently complete picture of the present

state of investigations in the theory o f turbulence, to aid him in
finding the necessary information, and, as it were, give him his
bearings on the infinite sea of books and papers. In a number of
places throughout the book we have even included brief surveys o f
the literature, in which we have tried, as far as possible to give a short
description of the content o f a considerable number o f typical works
referring to the particular section o f the theory under discussion. We
realize, of course, that a large number of references encumbers to
some extent, the text of a book and also (since it is completely
impossible to cite even the majority o f the existing works)
considerably increases the number of authors who could feel they
had been unjustly passed over in so extensive a survey. We should
like, first, to offer our apologies to all such authors and to warn the
reader that we in no way claim that the works included in our
bibliography are necessarily the best or the most im portant, nor do
we pretend that it is a complete survey. We realize that as far as
priority is concerned (a subject which we did not especially study)
there may well be a number of slips.
Of course, in selecting material for this book our personal
scientific tastes played a definite part. In particular, we have included
a number of results from our own research. A large proportion of
these results is given in a revised and augmented form which now
seems better to us; often, also, we present the results o f other
authors in a revised form, taking into account later data and also the
general approach, terminology and notation adopted in this book. It
is natural, therefore, that in many cases the works cited contain only
an equivalent (or similar) form of the equation or deduction in
connection with which they are quoted.
Often, we also depart from the analysis given in the original work
as a result of our desire to explain every assumption used in the
course o f the various deductions as clearly as possible. In particular,
we always try to clarify which results follow immediately from the
equations of motion (i.e., from the general laws o f physics), which
follow from dimensional considerations (i.e., from the specific
hypothesis on the complete list of physical parameters affecting the
given effect), which require the introduction of special empirical
hypotheses (and what these are), and which are simply empirical
facts.
Statistical fluid mechanics makes wide use of the results and
methods of classical fluid mechanics and probability theory.
Therefore, a knowledge of these two subjects will greatly assist the

understanding o f this book. Nevertheless, we hope that our book will
be accessible also to those who have only a general mathematical and
physical training. With these readers in mind, we have included in the
first two chapters the basic necessary concepts from classical fluid
mechanics (beginning with the equations of continuity and motion)
and o f the theory o f probablility (commencing with the concept of
probability itselO- In these as in all subsequent chapters we have
tried to place fundamental stress on the principles involved without
lingering over technical details. Similarly, we never discuss methods
o f solving differential equations or other standard mathematical
techniques, but quote the answer (which is sometimes not easy to
find) immediately. At the same time we discuss in comparative detail
certain insufficiently widely known but im portant mathematical
questions traditionally excluded from books and papers intended for
those specializing in mechanics and physics (e.g., questions o f the
ergodic theorems or o f spectral representations of random fields);
this explains the fact that only two chapters of the book are devoted
to the mathematical theory o f random fields.
For technical reasons it has proved more convenient to issue the
book in two volumes. The first volume includes questions which can
be dealt with without recourse to spectral representations. (Probably
many o f our colleagues will be surprised that such problems have
been collected into a thick book.) Here we discuss general equations
o f fluid mechanics and their simplest corollaries (concluding with the
somewhat more specialized theory o f the small oscillations of a
compressible gas); the question o f hydrodynamic instability and
transition to turbulence is considered (including elements of the
nonlinear theory of instability); some elementary notions are given
from probability theory and the theory of random fields (including
conditions o f ergodicity); the application of dimensional and
similarity considerations to turbulent flows in tubes, channels,
boundary layers and in free turbulent flows (jets, wakes, etc.) are
discussed in detail and there is a brief survey of the main ideas and
results o f the semiempirical theory of turbulence (with specific
examples); the theory of similarity for turbulence in a medium with
vertical density stratification is discussed in detail, and the results
obtained are compared with the extensive empirical data on wind
and temperature fluctuations in the surface layer o f the atmosphere;
the Lagrangian characteristics of turbulence and the theory of
turbulent diffusion are described.
INTRODUCTION 25
The second volume of the book will begin with a mathematical

chapter devoted to the spectral theory of random fields (including
fields that are inhomogeneous but locally homogeneous); there is a
detailed description of the theory o f isotropic turbulence (here we
are concerned principally with the various methods of closure o f the
moment equations o f incompressible isotropic turbulence, but some
results relating to compressible fluids are also given); the general
theory o f the universal local structure of turbulence with sufficiently
large Reynolds numbers is considered (including the theory of
relative diffusion, i.e., the increase in size o f a cloud of an admixture
carried by a turbulent flow); the basic concepts relating to the
propagation of electromagnetic and sound waves in a turbulent
medium are given and, finally, we consider the general formulation of
the problem of turbulence, based on the study of the characteristic
functionals of hydrodynamic fields.
/ LAMINAR AND
TURBULENT FLOWS
1. EQUATIONS OF FLUID DYNAMICS AND

THEIR CONSEQUENCES
1.1 System of Dynamical Equations for an
Incompressible Fluid
Turbulence is a complex physical phenomenon, and its theoretical
investigation must be based on the fundamental laws of physics,
expressed in the equations of hydro- and thermodynamics. Hence,
we shall begin our discussion with a brief resume of these equations
and some of their more im portant consequences, restricting our
selves, of course, to those equations and facts which we shall need in
the subsequent discussion [for a more detailed treatm ent of fluid
mechanics see, e.g.. Lamb (1932), Goldstein (1938), Landau and
Lifshitz (1963), Kochin, Kibel’, and Roze (1964), Longwell (1966),
and Batchelor (1967)]. As usual, we shall describe the flow of fluid*
' Here and henceforth, we shall use the word “fluid” to denote any liquid or gaseous
medium.
27
by the velocity field u{x, t)={u\{xx, x^, ats, /), x^. I),
X2, X3, /)} and by the fields o f any two thermodynamic
quantities, e.g., the pressure p ( x , t) and the density p(jf, t) or the
temperature T {x, /)—in all, by five functions of four variables. In
addition, we shall also need the values of the molecular transport
coefficients which determine the physical properties of the flu id -th e
coefficient of viscosity ju, or kinematic viscosity ~ the sec
ond coefficient of viscosity C. Later, we shall need the coefficient
of thermal conductivity x (and thermal diffusivity or thermometric
conductivity X= where Cp is the specific heat at constant
pressure o f the fluid).
The simplest equation of fluid mechanics describing the physical
law o f conservation o f mass is the equation of continuity
# + ■^ = 0 “ ■'>
or
Here and henceforth, we shall always adopt Einstein’s summation

convention, according to which whenever an index occurs twice in a
single-term expression, the summation is carried out over the three
possible values of the index, so that — , for example, has the
same meaning as ^ The fundamental dynamic equations

a= 1
expressing Newton’s second law apphed to a small volume of fluid,

i.e., the equation o f conservation of momentum, have the form
d (pui) ! ^ (P«/«a)___V \
dt ^----- dx^
+ — v'î
/ dus du^ 2 duo d f duo \
+ 3 ^ = 1^2,3 (1.3)
În the deduction of the general equations, we shall always denote the coordinate axes
by Oxi, i = 1, 2, 3 in order to use widely this convenient summation convention. In specific
examples, however, when the directions of the three axes are defined in some special
manner, we shall often also use the letters x, y , z i o denote the coordinates.
LAMINAR AND TURBULENT FLOWS 29
where pA', = pX, (jc, t) is the tth component of the external force-
density at the point x at the instant t. From Eq. (1.1) it follows that
the density p may be taken outside the derivative in the first term
of the left side of Eq. (1.3), taking pu„ outside the s i g n i n the
second term at the same time. For the coefficients of viscosity n and
the variation of these values in space (due to their dependence on
the temperature which varies from point to point) is almost always
negligible. Thus terms containing derivatives of these coefficients are
usually ignored completely. However, this means that the values of
and ^ in Eq. (1.3) may also be taken outside the derivative, so that
these equations become
Since the four equations (1.2) and (1.4) contain five unknown
functions, they do not yet form a closed system. In practice,
however, the variation of the density p of a moving fluid particle
proves to be so small in many cases that this too may be ignored.
This is particularly true in the case of ordinary liquids, i.e., liquids in
the normal sense of the word. In the case of gaseous media in steady
motion [with velocity u{x) independent o f time] the variations of
density due to the variations of pressure may be ignored completely,
provided the absolute velocity u = 1 a I at every point of the flow is
small in comparison with the velocity of sound a. In general, for
unsteady motion, variations of the density p will be negligibly small
if, in addition to the condition u a, the condition T ^ L j a is also
fulfilled, where T and L are the characteristic time and length scales
in which the velocity of the fluid u {x, t) undergoes a perceptible
change; see Landau and Lifshitz (1963), Sect. 10. In all the above
cases, the density p may generally be taken as constant throughout
the whole space filled by the fluid. Then p will represent a
preassigned constant value (characteristic of the properties of the
medium) and the four equations (1.2) and (1.4) will now suffice for
the four unknown functions t) i = 1, 2, 3, and p ( x , t) to be
determined (for given initial and boundary conditions).
In the present case o f an incompressible fluid, these equations take
a simpler form, since a number o f terms containing derivatives of p
(or expressed in terms of such derivatives) may be omitted. Thus, in
this case we may omit ail tlie terms on the ieft side of the equation
of continuity ( 1 . 2 ), so that
dx, = 0 . (1.5)
In the equations o f motion (1.4) the last term on the right side thus
becomes identically equal to zero, and these equations are then
transformed into the well-known Navier-Stokes equations
which are the fundamental dynamical equations o f a viscous

incompressible fluid. We see that they contain only a single
coefficient o f viscosity, since the second viscosity is im portant only
when compressibility is taken into account.
It is easy to see that the pressure p may be eliminated from Eqs.
(1.6) from the very beginning. In fact, for this purpose we need only
apply to both sides of Eqs. (1.6) the operation of taking the curl,
which in tensor notation is written as the operation , where
is a completely skew-symmetric tensor of the third rank (i.e.,
= 0 if the indices k , ^ , i are not all different, and Shpj = -f- 1 o r — 1
if (k, p, 0 is obtained from (1, 2, 3) by an even or odd permutation,
respectively). Assuming, for simplicity, that there are no external
forces, we arrive at a system of three equations
= ^= 1.2,3. (1.7)
where
are the components of the vorticity vector. In principle we may

determine the three functions uîx, t) from Eqs. (1.7) and (1.8);
then, in order to find the pressure field it is only necessary to solve
Poisson’s equation
P dx^d^' (1.9)
which is obtained by applying the operation to Eqs. (1.6) [with

X, = 0 ], i.e., by taking the divergence of the Navier-Stokes vector
equation.
From Eq. (1.9) it follows that, accurate to within a harmonic
function of x ,
4it ./ dx^dx^ \x — x'\
where the integration is taken over the whole volume occupied by

the fluid. In particular, if the fluid flow then takes place in the whole
unbounded space, then the corresponding harmonic function may be
only a constant. Since only derivatives of the pressure appear in the
equation o f motion, the constant term in the expression for the
pressure will generally play no role at all; thus it is permissible here
to assume that Eq. (1.9') is accurate without any additions. For
flows in finite regions, the harmonic addition to the right side of Eq.
(1.9') must be determined from the boundary conditions for the
pressure; here, in a number of cases this may also prove to be a
constant and hence may be ignored.
1.2 Simple Flows of Incompressible Fluid
To obtain a unique solution o f the system o f equations
(1.5)—(1.6) it is necessary to fix the initial values o f the fields
Ui {X, t) and to take into account that the velocity must vanish on
the surfaces o f all rigid bodies immersed in the flow. However, since
this system is nonUnear, finding its exact solutions in explicit
analytical form is, in general, very difficult. Only for a few special
flows are the nonlinear terms of Eqs. ( 1 .6 ) identically equal to zero;
in this case the system of equations of fluid mechanics is greatly
simplified, and in such cases its exact solution is usually found
without much trouble. Below, we give some exact solutions of this
type; examples of more compUcated exact solutions of Eqs.
(1.5)—(1.6) may be found, for instance, in Landau and Lifshitz
(1963), Kochin, Kibel’, and Roze (1964), and Batchelor (1967). In
all the examples considered below we shall assume that there are no
external body forces X,-^; moreover, the flow will always be assumed
steady, i.e., time-independent.
^We observe that the body force o f gravity X = (0, 0, —p|r) which is always present
may easily be eliminated when there is no free surface of the fluid, by subtraction from the
true pressure p of the hydrostatic pressure Po = — + const.
We note first, that in the case of a steady flow the- total

momentum of the fluid in a fixed part of space does not vary with
time; i.e., the momentum loss connected with the friction of rigid
bodies submerged in the fluid is balanced by the gain in momentum
due to the action of forces that produce the fluid motion (in the case
of Xi= 0, due to the action of the pressure drop). In other words, in
a steady regime the force of the pressure drop acting over the whole
fluid to produce motion is balanced by the retarding force acting on
the fluid (which differs only in sign from the total force exerted by
the flow on the body submerged in it). The equation expressing this
equality which allows us to establish a relationship between the
characteristic velocity of the flow and the pressure drop is usually
called the '’“'drag law. ”
The force W exerted by the flow on the body is equal to the
integral over the surface of the body 2 of the flux of momentum
along the normal to this surface. The flux of the ith com ponent of
momentum per unit area perpendicular to the axis Ox^ has the form
p«j«h+p 6 ,ft— Oift, where 6 ik is the unit tensor (6<a = 1 fort = ^, 6 ,-ft = 0
for i ^ k) and ah is the viscous stress tensor, which in an
incompressible fluid is equal to + (1-61)1-
Since the velocity of the fluid on the surface of a rigid body equals
zero, we obtain for the ith component of the force W
where n, is the component of the unit vector of the outward normal

to the surface, and f,- is the integral o f aihfih over the surface of the
body. The first term on the right side of Eq. (1.10) describes the
transfer of momentum to the body by the pressure forces and is
independent of the viscosity of the fluid. In several cases (e.g., in the
case of flow along a flat plate or flows in straight tubes) this
component is equal to zero. The second component represents the
skin friction drag, which in a viscous fluid is always nonzero. In the
case of geometrically similar bodies, the expression pU^S may be
taken as characteristic of the force W, where U is the characteristic
velocity and 5 some characteristic area, e.g., the area of the surface
of the body or the area of its midsection. The dimensionless quantity
is called the drag coefficient o f the body. If the drag is due only to
the friction (IF, = /=■,), the drag coefficient is denoted b y Q and called
the coefficient o f friction (or the friction factor, or skin friction
coefficient).
Examples o f exact solutions o f the fluid-mechanical equations
1. We begin with the simplified case o f a two-dimensional flow of

viscous fluid between two parallel planes, one of which is stationary
and the other moving with constant velocity U. We denote the
distance between the planes by H and choose a coordinate system in
which the equations of these planes will have the form 2 = 0
(stationary plane) and z = H (moving plane), while the Ox axis is
directed along the vector U. In this case all the hydrodynamic
quantities will be dependent only on z, while the velocity of the
fluid will everywhere lie along theO;e axis. As a result, Eq. (1.5) and
the second equation of Eqs. (1. 6 ) will be satisfied identically, and the
first and third equations o f Eqs. (1. 6 ) will take the forms
—0- —0 (112)
dz^ dz
where u (2 ) = «i (2 ) is the only nonzero component of the velocity of

the flow. Consequently, p — const and u = az + b\ taking the
boundary conditions « = 0 for 2 = 0 , u = U for z = H into
consideration, we obtain
u{z) = ^ z . (1.13)
Thus the flow is described by a linear velocity profile, while the

mean velocity Um ~~y - frictional force per unit area of the walls
2 = 0 and z = H equals
On the plane 2 = 0 this force is directed along the Ox axis, while on

z = H it has the opposite direction. If we put
(1.15)
where Cf is the skin friction coefficient, then
The steady plane flow of fluid between infinite parallel planes

described by Eqs. (1.13) and (1.16) is clearly a mathematical
idealization, which, however, may prove useful in certain cases. In
handbooks on fluid mechanics, this idealization is sometimes called a
plane Couette flow.
2. Let us now consider a steady two-dimensional flow between

two fixed parallel planes z = 0 and z = H under the action of an
external force applied at infinity which produces a pressure gradient
along the Ox axis. It is easy to see that in this case the velocity o f the
flow at all points will also be directed along the Ox axis and will
depend only on z, so that u(z) = u i ( z ) will again be the only
nonzero com ponent of the field ». Here Eq. (1.5) and the second
equation o f Eqs. (1.6) will also be satisfied identically, while the first
and third equations o f Eqs. (1.6) will take the form
= ^ = 0-
It is clear therefore that in this case the pressure will be independent

of z; i.e., it will be constant over the whole thickness of the layer of
fluid, and consequently the first equation of Eq. (1.17) will be
satisfied only for
- ^ = const , u{z) = — - ^ - ^ z ‘^ -\-a z-\-b, (1.18)
where Aip is the drop in pressure between the planes x = xo and x =

Xo + I. Taking into account the boundary conditions (« = 0 for 2 = 0
and for z = H ), we obtain
//\2 //2
(1.19)
so that in this case the velocity profile will have a parabolic form.
The sheer stress t on both walls will equal
du
( 1.20 )
dz ~ 2 I
Hence, the dimensionless skin friction coefficient C/ of Eq. (1.15)

will equal
( 1.21)
H U ^ ~~ Re
(because in the present case
The idealized flow described by the exact solution (1.19) is

sometimes called plane Poiseuille flow.
3. We now turn to the case of a steady flow of fluid in a straight,

infinitely long, circular cylindrical tube o f given diameter D. The axis
Ox is taken along the axis of the tube; then the only nonzero
com ponent of the velocity field will be z) = u(y, z). Thus in
this case also, Eq. (1.5) will be satisfied identically; further, it follows
from the second and third equations of Eqs. (1.6) t h a t - ^ = - ^ =
0; i.e., that the pressure is constant across the section of the tube and
depends only on x . We now introduce into the plane Oyz the
cylindrical coordinates (r, cp). Then by symmetry of the tube,
u{y, z) = u{r), and hence in these coordinates the first equation of
Eqs. (1. 6 ) will take the form
( 1.22)
dx r dr \ dr )
Since the left side depends only on x and the right side on r, we must
have ^ = const — — wher e A/p is the pressure drop along a

section of tube length /. Taking into account the boundary
condition u(D/2) = 0 and the fact that u{r) is bounded for all r, we
obtain
(1.23)
Thus also in this case, the velocity profile (along any diameter) will be
parabolic. The mean velocity o f flow defined by the equation
2it /?
^ m = - ^ J f u{r)rdrd(f, (1-24)
0 0
is equal here to ^ 2 ” »hence, the discharge of fluid (i.e., the

volume of fluid flowing through a cross section of the pipe in unit
time) equals
Q= (1-25)
du îp
The shear stress on the wall is equal to t = the
same result may also be obtained by considering that the total shear
stress nDl • t on an element of the tube of length / must be balanced
by the force o f the pressure drop acting on the cross section of the
tube Consequently, in this case the skin friction coefficient
equals
= where Re = - ^ . (1.26)
The solution (1.23) of the equations of fluid mechanics also refers

to an idealized flow, since the tube is taken to be infinitely long, and
the flow, to be strictly constant. However, it is found that in cases of
real finite tubes, the flow o f fluid along them under constant
pressure head A;p quite often becomes practically steady extremely
quickly, while at distances from the intake of the tube of the order
RÛ
o f — ■— [according to the theoretical calculations o f Schiller (1934),
cited a SO in Goldstein (1938), Vol. 1, Sect. 139, even fo rx > 0.115

the effect of the finite length of the tube also ceases to be
perceptible, and Eq. (1.23) is a good description of the flow. Thus
the laws (1.23), (1.25), and (1.26) may be checked experimentally;
historically speaking, they were first demonstrated experimentally
(by G. Hagen in 1839 and J. Poiseuille in 1840-1841) and only later
justified theoretically (by J. Stokes in 1845). Hagen also showed that
these laws are correct only in the case of sufficiently high viscosity v
and sufficiently low velocity fact which we shall discuss in
detail in Sect. 2. A flow of incompressible fluid described by Eqs.
(1.23) and (1.25) is often called a circular Poiseuille flow (or a
Hagen-Poiseuille flow) or (in cases when there is no danger of
confusion with the similar plane flow) simply a Poiseuille flow.
4. As our last example we shall consider the steady motion of a

fluid in an annulus between two coaxial infinite cylinders of radii/?.i
and /?2 > Ri, rotating about their axes with angular velocities Qi and
Q2; the pressure drop along the axes of the cylinders is taken equal to
zero. In this case all the hydrodynamic quantities will be independ
ent of the X coordinate, reckoned along the axis of the cylinders.
Further, if in the plane O 1/2 perpendicular to the axis o f the cylinders
we introduce cylindrical coordinates (r, cp), then from considerations
of symmetry it is clear that only the com ponent of velocity u = u ,
will be nonzero and that the velocity u and the pressure p will
depend only on r. It follows that in this case the equation of
continuity (1.5) and the first equation o f Eqs. (1.6) will be satisfied
identically; for the second and third equations of Eqs. (1.6), after
transformation to cyhndrical coordinates, these will take the form
1 dp _ w’ dû I 1 u _^ 9 7 -^
J~dr~~' 1F ^ T ~ d r~ l^ ~ ^ '
The general solution of the second of Eqs. (1.27) has the form
u{r) == + y ; since at the same time a (r) must equal Qi/?i fo rr = /?i
and Q2R 2 for r = R 2 , the unknown function u (r) will be defined by
the equation
/\2 —Aj /\2 — '

Using the first equation of Eqs. (1.27) it is now very easy to find p(r),
but we shall not do so here. The flow given by Eq. (1.28), in many
respects, is similar to that o f Example 1, but it may be reproduced
far more simply and with comparative accuracy in the laboratory.
This flow is sometimes called circular Couette flow, or flow between'
cylinders.
1.3 The Reynolds Number and Similarity Criteria
All exact solutions o f the equations of fluid mechanics given above
refer to extremely simple idealized flows. In more complex cases it is
generally impossible to find an exact solution, and we have to resort
to a numerical solution of Eqs. (1 .5 )-(1 .6 ) using some approxima
tion method. In such cases it is extremely im portant to be able to
estimate the orders of magnitude of the terms in our equations to
determine which of the terms may be ignored in which cases. As a
result, we may confine our attention to simplified equations that are
much easier to integrate.
We shall assume that we are dealing with a steady flow of viscous
incompressible fluid in the absence o f body forces. Using Eq. (1.9)
makes it easy to show that the terms of Eqs. (1.6) containing the
pressure are, in general, of the same order as the nonlinear terms
necessary to compare the orders of the terms
which describe the effect of the forces of inertia and

the forces of viscous friction. We take U to be some typical velocity
(e.g., the maximum velocity o f flow, or some mean velocity) and L
to be some typical length (e.g., the mean dimension of a body
immersed in the flow, or the typical distance between rigid walls, or,
in general, a distance along which the velocity of the flow undergoes
a perceptible change A(7 of the order o f U). In this case, the order of
the first derivative of the velocity field will be given by the ratio UjL
and the second derivative by the ratio t//L^, so that terms describing
the inertia forces will have order W^jL and terms describing the
friction forces will have order vi7/Z,2. The ratio of these quantities.
is called the Reynolds number of the flow; it is a very important

characteristic o f flow, defining the relative role of the inertia forces
and the friction forces in the dynamics [we have already introduced
this number in Eqs. (1.16), (1.21), and (1.26)]. In the case of small
Reynolds numbers, the viscosity has a considerable effect on the
entire flow by smoothing out all existing small-scale inhomogeneities;
hence space variations of the fluid mechanical quantities must
take place very smoothly in the case o f small Reynolds numbers. In
the case of large Reynolds numbers the dominant role in the flow is
played by the inertia forces, the action o f which leads to the transfer
of energy from the large-scale components of motion to the
small-scale components and consequently to the formation of sharp
local irregularities; we shall be concerned with this type of flow in all
subsequent sections o f this book.
The concept of the Reynolds num ber greatly simplifies the
investigation o f geometrically similar flows of fluid, such as, flows of
fluid in tubes with cross sections of given form, or flow of an infinite
stream o f fluid past a rigid body of given form. Since in these cases
we are concerned with the whole set o f flows with geometrically
similar boundaries, the properties of the boundaries will be deter
mined uniquely by the same length L (in our examples, L may be
taken as the mean diameter o f the cross section of the tube or the
mean diameter of the body washed by the flow). Furtherm ore, the
flow itself will be described by some typical velocity U (for example,
the maximum velocity in a fixed cross section of the tube, or the
velocity o f the flow incident on a body). Finally, the only
dimensional parameter occurring in Eqs. (1.5) and (1.6) is the
viscosity coefficient v, which describes the physical properties of a
fluid. Thus, for a steady flow in the absence of body forces,
geometrically similar flows will depend only on the length L and the
dimensional parameters V and v, which have the dimensions
\U\ = LT-^; [v1 = l27’- \ (1.30)
where L and T denote, as usual, the dimensions of length and time,

respectively. It is clear that from the parameters L, L), and v, a unique
dimensionless com bination—Reynolds number Re = may be
formulated. Hence, if we measure length in units o f L, velocity in

units of U, and pressure in units of pf/*, then Eq. (1.5) and the
boundary conditions for all geometrically similar flows will be of the
same form, while Eq. (1.6) will be characterized almost entirely by
the value of the dimensionless parameter Re, which will appear in
this equation after transformation to dimensionless form. Conse

quently, in the case o f geometrically similar flows, the velocity,
measured in terms of U will depend only on the coordinates
measured in terms o f L , and on the number Re. In other words,
Ui{x) = U ^ i { ^ ^ , Rej, (1.31)
where 9 , ( 1 , Re) is some universal function describing the entire set of

flows under consideration (we must point out that in Examples 1 to
4 of the previous subsection, the functions 9 ,, due to various special
reasons, were completely independent o f Re). In the same way, we
have p{x) = ?U^<fo{-^ ^ where 90 is a new universal function.
Further, if W is the force exerted by the flow upon some rigid body
w
washed by the flow, then the drag coefficient -------- will
depend only on Re, while in the definition o f Cw may also be

replaced by the area S o f the whole surface (washed by the flow) of
the body (so that we can assume thatC^ = where t is the mean
friction force per unit surface). In other words, when the correspond
ing Reynolds numbers are equal (but, generally speaking, only when
they are equal), geometrically similar flows will also exhibit
mechanical similarity; i.e., they will possess geometrically similar
configurations of the stream lines and will be described by the same
functions of the dimensionless coordinates (this is called Reynolds’
similarity law). This law is of great importance in the theoretical
study o f flows occurring under similar conditions, for coordinating
the presentation of observational data on such flows, and for the
modeling o f flows encountered in practical problems. Furtherm ore,
it indicates that all possible geometrically similar flows will be
described by a one-parameter family of solutions of Eqs. (1.5)—(1.6);
this of course greatly simplifies the problem of integrating these
equations numerically.
We must stress once again that Reynolds’ similarity law holds only
for steady flows o f incompressible fluid, on which external forces
exert no appreciable effect. However, for motions that depend
largely on body forces (e.g., on the force of gravity) and also for
nonsteady motions described by some typical period T which differs
from LfU, the similarity law is more complicated; in such a case, for
mechanical similarity, it is necessary not only that the Reynolds
numbers be equal, but also that certain additional dimensionless
“ similarity criteria” be equal. In the case of flows of a compressible
fluid, the number of similarity criteria will be even greater; we shall
discuss this further in Sect. 1.6 .
We must also point out that in certain cases a very slight departure
from geometrical similarity may lead to a total breakdown of
mechanical similarity. Thus, for example, a slight change in the
conditions in the intake section o f a circular tube, introducing slight
disturbances into the flow, may completely change the character of
the flow in the tube (see Sect. 2).
1.4 Flows with Large Reynolds Numbers;
The Boundary Layer
We shall now assume that the Reynolds number of the flow is very
large. In this case, the nonlinear inertia terms of Eqs. (1.6) will be far
greater in magnitude than the terms containing the viscosity. Thus, at
first glance, the effect o f the viscosity may appear to be simply
ignored. However, this is not so; to ignore the terms containing v in
Eqs. (1.6) would reduce the order of these differential equations.
Moreover, the solutions o f the simplified equations of an ideal fluid
thus obtained would no longer satisfy the “ no slip” boundary
conditions which require that the velocity vanish upon all rigid
surfaces bounded by the fluid. At the same time, we know that for a
viscous fluid (however small the viscosity) “ adhesion” is bound to
take place. Therefore, in the motion of a viscous fluid with a large
Reynolds number, only at a distance from the rigid walls does the
flow become close to that which would occur in the case of an ideal
fluid (with zero viscosity); close to the walls a thin layer exists in
which the velocity o f the flow changes very rapidly from zero on the
wall to the value on the outer boundary of the layer. This value is
very close to that which would be obtained in the case of an ideal
fluid. The rapid variation of velocity within this layer (known as the
boundary layer) means that within it, the effect of the friction forces
is by no means small, but of the same order as the effect of the
forces o f inertia.
Thus, for large Reynolds numbers, the viscosity of the fluid plays
an effective role only in the region close to the boundary walls. In
this region the terms containing v cannot be ignored; however, here
the equations o f fluid mechanics can be simplified considerably due
to specific features of the motion of a thin film of fluid in contact

with a rigid body. Therefore, in free space, the flow will be
determined by the equations of an ideal fluid, while for the
boundary conditions we need to use the conditions on the upper
surface of the corresponding boundary layer (or layers).
Let us now consider the simple case o f a boundary layer on a flat
plate of length L and infinite width lying in the plane Oxy (so that it
occupies the region 0 < jc < Z, of this plane) and washed by a flow of
fluid, moving with given velocity U in theO;c direction. First, before
considering the mathematical solution of this problem, let us make a
few qualitative preliminary remarks. Since in this case the character
istic length will be L and the velocity U, the nonlinear terms in Eqs.
( 1 .6 ) [or, more precisely, in the first of these equations] will be of
the order U^/L in the region of the boundary layer. On the other
hand, the longitudinal velocity u = u\ will vary in this region from
« = 0 on the surface of the plate to a value of the order of U on the
outer edge o f the boundary layer.
Let us take the thickness o f the boundary layer to be S.'*
We now obtain that is o f the order o f U/6 \ and the main term
v -^ describing the friction force will be of the order o fv U / 6 ^.
However, within the boundary layer the friction force will have the
same order as the inertia force. Hence WIL ~ vi7/5^, i.e..
Re = - ^ . (1.32)
U
Thus the thickness o f the boundary layer depends on the Reynolds

number o f the flow. The greater this number (i.e., the smaller the
viscosity of the fluid and the greater the free stream velocity), the
thinner will be the boundary layer.
To determine the thickness of the boundary layer above a given
point of the plate at a distance x from its leading edge, we must
replace the total length of the plate L by the length x . Thus
^This is, of course, not an absolute concept since the boundary layer does not have a
clearly defined edge, and the velocity u only tends asymptotically to U with increasing
distance fi:om the plate in the Oz direction. Hence 6 is understood to denote the distance
from the plate (in the Oz direction) at which the velocity u attains a given, sufficiently
great fraction of U, e.g., 0.99U (this definition of 6 is often used in fluid mechanics). There
exist other somewhat less formal definitions of the thickness of the boundary layer which,
however, in practice give equivalent results-see the closing remarks of this subsection.
8 —^ at the point (1.33)
i.e., as the distance x from the leading edge increases, the thickness
of the boundary layer increases in proportion to\/~x. As far as the
shear stress x on the wall is concerned (t is equal to the friction force
per unit area of the plate), we have t = p. (where must be
taken for z = 0). S i n c e w e have or, in view of Eq.

(1.32)
(1.34)
Thus the shear stress here is proportional to 3/2 power of the

velocity; the corresponding skin friction coefficient
naturally depends only on the Reynolds number and decreases as this

number increases.
We now turn to the deduction of the equations defining the
velocity field within the boundary layer. Since in the case under
discussion the motion will be strictly two-dimensional (in the plane
Oxz) and independent o f y, the general equations (1.5) and (1.6) take
the form
dw ,„ I dw I dp , { d^w , d^w \ n
du , dw „
where u = Ui, ay = « 3 . We shall now attem pt to determine the order

of magnitude of the individual terms of these equations. We recall
that within the boundary layer, u is equal to some finite fraction of
the velocity U, i.e., it is o f the same order as U, while the derivatives
dx
and 4dz- are respectively of the order UIL,an 6.Ulb. Integrating the
third equation of Eqs. (1.36) with respect to z from 0 to somez = c6 ,

0 < c < l , constituting a finite part of the total thickness 6 of the
boundary layer, we obtain
Therefore, it is clear that w is approximately times less

than u. Taking into account Eq. (1.32), we may assert that in the
first equation of Eqs. (1.36) the terms u w and are all
of the order o f UÎL, while the te r m v -^ ^ is of much smaller order

i/2 1
hence this latter term may be ignored. Further, it is clear
from the second equation o f Eqs. (1.36) that the term y will be
of an order not greater than —jjr = 'X ' • x (since no term of this
second equation is o f a higher order); thus with accuracy to terms of
relative order we may ignore transverse variations of pressure and
assume that -§7 “ i.e., p = p(A:). However, in this case the pressure
p (x) within the boundary layer may be replaced by the value on the
upper edge. Hence the term — t he first equation of Eqs.
(1.36) may be determined independently from the equations of
an ideal fluid which describe the flow outside the boundary
layer; putting v = 0 in the first of Eqs. (1.6) we find t hat — =
dU ^ dU ^ ^ \ dp , , dU 1 at/2
W + ^ I F or (in the steady c a s e ) - = = ^ T h u s ,
to find u and w we obtain a system of two equations:
du. . du . du I dp . dû ( i ^ o\
This system, together with the boundary conditions
u — 'W— 0 ,{ o T Z = 0 , 0 ^ x ^ L ] u - > - ( J { x , t ) asz-ôo (1.40)
permits the functions u (a;, z, t) and w {x, 2 , t) to be determined

uniquely from the initial values of one of these functions.
Equations (1.38) and (1.39) make up the well-known boundary
layer equations which were first obtained by Prandtl in 1904. Later,
both Prandtl and other authors proposed several different derivations
of equations. In particular, it was established that Prandtl’s equations
are also correct for the two-dimensional flow past a curved surface (if
the curvature is not too great) and that they may be obtained
formally from the general equations of fluid mechanics as a first
approximation by the expansion of all the terms in series in powers
of [see Kochin, Kibel’, Roze (1964), part 2, Chapt. II, Sect. 29;
Goldstein (1938), Vol. 1, Chapt. IV; see also the special monographs
by Loitsyanskiy (1941; 1962b) and Schlichting ( I960)]. In general,z
must be taken as the coordinate reckoned along the normal to the
surface and x as the longitudinal coordinate in the tangent plane.
Let us now consider in greater detail the simplest case of a flat
plate washed by a steady flow with constant velocity U in the Ox
direction. In this case, the terms and in Eq. (1.38)
disappear, so that this equation takes the particularly simple form
du I du dû /"I 41\
We assume that the length L is infinite. Since it is natural to think

that in the flow considered the velocities in every section a:= const
of the boundary layer will not depend greatly on the velocities in the
next sections downstream, the assumption of infinite length will have
no great effect on the solution anywhere except in the narrow
neighborhood of the trailing edge and in the region lying
behind the plate, which we shall not deal with here. With L = 00 the
boundary conditions (1.40) become
u — w — Q for z = 0, a: > 0; « -»• t / as 2 -> cx>. (1.42)
Thus we only have to find a solution of Eqs. (1.39) and (1.41)

which satisfies the conditions (1.42).
To eliminate v and U from the equations, we turn to the

dimensionless values
^1
^
— -f- > Z\ — 7
nr
— 2 1/ ^ f >
^ ^ (1.43)'
u -/ L
«. = 7 r - '^^ = — u — = ' ^ V w '
defined in accordance with Eqs. (1.32) and (1.37); the length L in

Eq. (1.43) can now be chosen as desired (since the conditions o f the
problem do not contain any characteristic lengths). Substituting Eqs.
(1.43) into Eqs. (1.41), (1.39), and (1.42), we obtain
+ ^ + ^ = 0 , (1.44)
dx^ dz^ dij dx^ dzj
Ui = Wi = 0 for 2 i = 0, JCi 0; Ml - ► 1 as Zi <x>. (1.45)
However, since the length L may be chosen arbitrarily, only a

solution independent of L has any meaning. Thus it is clear that
«i = - ^ can depend only on the combination fj = -^ 1— = z |/ ^ ^
of the variables Z\ and X\ which does not contain L . The function
= w |/ which is also expressed in terms of le), x, v and LJ
only, must depend on the same combination. Thus we see that
where / and U are two universal functions of a single variable

(independent o f t / a n d v). From Eq. (1.46), in particular, it follows
that the velocity profiles for u and w above all points of any plate
washed by a flow at constant velocity must be similar to each other.
If we put
/( ^ ) = 'P'(^), (1-47)
Eq. (1.39) is transformed into the form / i ("17) = y ('i^'p'— <p); then
Eqs. (1.41) and (1.42) are easily reduced to the form:
cpcp" + 2«p'" = 0; cp(0) = tp'(0) = 0 , cp'(oo) = l. (1.48)

Thus our system of equations reduces to a single nonlinear,

third-order, ordinary differential equation with three boundary
conditions. This equation may be integrated numerically, for
example, by resolving the function 9 (11) as a power series close to
V = 0, and as an asymptotic series as t) -> 00 (this method of
solution was used in 1908 by Blasius, who was the first to investigate
the problem); later, other numerical methods were applied to Eq.
(1.48) by Topfer, Barstow, Goldstein, Howarth and others [see the
references in Goldstein (1938), Schlichting (1960), Longwell (1966)
and Loytsyanskiy (1941, 1962b)]. The profiles obtained in these
works for the longitudinal and vertical velocities in a boundary layer
above a plate are given in Figs. 2 and 3. The thickness 5 of the
boundary layer (defined as the value o f z for which u(z) =0.991/) is
given by
(1.49)
[cf. Eq. (1.33)]. The shear stress at a point of the plate distant x
from the leading edge is given by the equation
10
ae
k
0.6 f
/
u + 108-W‘
OA • lez-w ^
f 0 3.s^ ro^
2
• S.46-/0^
// • 7.28-JO^
0.2
/
10 2.0 3.0 AO s.o 6.0 7.0
FIG. 2. Longitudinal velocity profile in the boundary layei on a flat plate.

Experimental data according to Nikuradse’s measurements [see Schlichting
(I9 60 )].
FIG. 3. Vertical velocity profile in the boundary layer on a flat plate.
since the numerical integration shows that <p"(0)= 0.332, then

consequently
0.332 r, Ux
• = — (1.50)
pt/2
For a plate of length L and wetted on both sides, the total friction
force (per unit breadth of the plate) will equal
F = 2 j t(A:)dx = 1.328ptAZ73Z, (1.51)
i.e., it will be proportional to the 3/2 power of the velocity and the
square root of the length o f the plate (the comparatively slow
increase of F with increase of L is explained quite naturally by the
fact that the stress t(jc) decreases with increase of the distance from
the leading edge due to the increase of thickness of the boundary
layer). Instead o f F, we normally use the dimensionless skin friction
coefficient Cf, which for a flat plate of length L and breadth B is
determined from the equation
FB 1.328 UL
Re = (1.52)
The experimental verification of the theory of flow past a flat

plate was carried out, in particular, by I. M. Burgers and B. G. Van
der Hegge Zijnen in 1924, M. Hansen in 1928 and J. Nikuradse in

1942 [for references to the original works see, e.g., Schlichting
(I9 6 0 )]. It was established that if the Reynolds number Re = is
sufficiently large, the observed values of the skin friction coefficient
are described with sufficient accuracy by Eq. (1.52). In exactly the
same way, the measured values o f the thickness of the boundary
layer 8 and the form o f the longitudinal velocity profile u {z) within
Ux
this layer for values of Re^ = —i;;- that are neither too large nor too
small (i.e., everywhere except in a narrow strip near the leading edge
dû dû
of the plate where the condition dx^ c dz^ is not fulfilled, and
UL
also, perhaps the traihng edge of the plate in the case when is
very great) prove to be in extremely close agreement with the
theoretical predictions. The slight divergence of theory and experi
ment observed in early measurements may be explained entirely by
the effect of the finite thickness of the plate and the form of the
sharpening on the leading edge, and also by the presence of fairly
76
76
/
74 y
/+ +
72 y
h
70 >
6
S
4
2
4 6 6 W 72
Ûx
70-^
FIG. 4. Dependence of the boundary-layer
thickness 5 on a flat plate on the distance jc
from the leading edge of the plate [accord
ing to the data of Hansen (1928)].
small pressure gradients in the flow; these phenomena were specially

considered in Nikuradse’s experiments, and he was able to obtain an
almost ideal agreement with the calculated profile of u(z) [see Fig. 2
in which Nikuradse’s data are p lo tte d ]. However, for large values of
the velocity U and length L, beginning from some value z for which
RCx is of the order 3 X 10^-3 X 10®, all the laws deduced above are
greatly distorted [see, for example, Fig. 4, where results of Hansen’s
measurements o f the thickness of a boundary layer are given. These
results show that beginning at about Re* = 3.2 X 10’ , this thickness
begins to increase far more rapidly than would be expected from Eq.
(1.49)]. The causes producing these distortions will be discussed in
detail in Sect. 2.
So far, we have defined the thickness of the boundary layer on a purely conditional basis
as the value of z for which the longitudinal velocity u (z) attains a value equal to some
predetermined fraction of U sufficiently close to unity (for example, equal to 99% of U).
We were forced to adopt this somewhat vague definition because u(z) approaches U only
asymptotically as 2 ->oo, and nowhere attains a terminal value; strictly speaking, we should
regard the boundary layer as not having a sharp boundary. Nevertheless, the transverse
dimension of the boundary layer may be described by quantities which have definite
physical meaning. One such quantity is the displacement thickness 6*, defined by the
equation ^
a {z)
dz. (1.53)
- / [ ■ U
The meaning of this quantity is that 6* is equal to the distance through which the flow is
displaced outwards due to the frictional reduction of the longitudinal velocity in the
boundary layer. Let us in fact consider a streamline at a distance Zq from the plate at its
leading edge (see Fig. 5). Since the flux of fluid through every section of the stream tube
contained between this streamline and the surface of the plate must be the same, while the
longitudinal velocity u decreases downstream due to friction on the plate, the section of the
stream tube must increase with increase o f ;c. At a distance jc from the leading edge of the
plate, the streamline is a distance Zi from the plate, i.e., it has been displaced through a
distance Zi — , which may be determined from the condition of constant flux of the fluid
through the initial section of the stream tube and through the section with longitudinal
FIG. 5. Schematic form of a streamline in the

boundary layer on a flat plate.
coordinate x :
where
<1
— = -----(1.54)
0
As 2 i ->oo the difference Zi — Zq tends precisely to the displacement thickness 6*. For the
case of steady flow past a flat plate
oo
6 * = ] / ^ / ( 1 . 5 5 )
0
where we may obtain the equation
6* « 1.73 (1-56)
We note in this case that our previous calculation of the thickness of the boundary layer 6
(defined according to the relationship U (h )= 0.99 U) gives a value almost three times the
displacement thickness 6*. Due to the “displacement” of the streamlines, the vertical
component of velocity w does not tend to zero as -r -> oo
(see Fig. 3) but approaches some
asymptotic value, which proves to be equal to
w (00) = 0 .8 9 5 (/ (1 .5 7 )
Thus the displacement shows that the flow is distorted even outside the boundary layer
(however the thickness of the layer is defined), i.e., the boundary layer has an effect on the
outer flow which sometimes must be taken into account in calculations.
Another quantity that characterizes the transverse dimension of a boundary layer is the
momentum thickness, defined as
The term momentum thickness derives from the fact that the momentum transmitted by
the friction forces per unit time to an element of the plate of unit breadth and extending
from the leading edge x = 0 to a section of the plate at a given x (i.e., the momentum lost
per unit time by the column o f fluid lying above this part of the plate) may be shown to be
equal to . That is, it is characterized by the length 6**. In the case of a steady flow
of constant velocity U past a plate, this length will be [from Eqs. (1.46), (1.47), and
(1.48)1:
= ]/^ ^ / V' (1 - f ' ) '^’1 = 2?" (0) ] / ^ = 0-664 : (1.59)

0
therefore, 6** here is almost three times smaller than 6*.

We may note further that boundary layer theory also permits us to explain why
Poiseuille flow which represents an exact solution of the equations of fluid mechanics
RÛ
occurs only at distances of the order of — froiri the intake of a tube (see subsection 1.3,
Example 4). Close to the mouth of the tube the retarding action of the walls is exerted only
within the thin boundary layer, while in the central part of the tube, the fluid will move
with constant velocity, experiencing no viscous effects. Proceeding further into the tube,
the thickness of the boundary layer on the walls of the tube will increase in proportion to
- j j , where x is the distance from the mouth. Finally, at a distance ^ from the
mouth, the boundary layer fills the whole cross section of the tube, and viscous effects
become evident through the whole flow. Only then, of course, can the theoretical parabolic
velocity profile given by Eq. (1.23) be established [cf. Schiller (1932; 1934) or Goldstein
(1938), Vol. 1, Chap. 7, Sect. 139].
1.5 General Equation of the Heat Budget and the

Thermal Conduction Equation; Forced and
Free Convection
In what follows, we shall be concerned mainly with flows of an
incompressible fluid which are described by Navier-Stokes equations
(1.6) and the continuity equation (1.5). However, since Chapt. 4 will
be devoted to the more general case of a thermally inhomogeneous
fluid, and the second part of the book briefly considers turbulence in
a compressible medium, it will be beneficial at this point to discuss
the general system of equations of mechanics of a compressible fluid.
The results o f this discussion allow us to write down the equation for
the temperature of an incompressible thermally inhomogeneous
fluid, which will be important in the subsequent analysis.
In a compressible medium the continuity equation (i.e., the
balance o f mass) and the dynamical equations (i.e., the balance of
the three components of momentum) take the forms (1.2) and (1.4),
respectively. Since these four equations contain five unknowns, to
obtain a closed system we must add to them a fifth eq uation -th e
equation of the budget of heat, which expresses the physical law of
conservation of energy. In the most general form, this equation is
where e is the internal energy of unit mass o f fluid [ so that the sum
p(uV2 + e) gives the total energy per unit volume of moving fluid],
>w= e - \ - - ^ is known as the enthalpy,® x is the coefficient of
thermal conductivity and T is the temperature. Further, in Eq.
(1.60),
r
[du, duj 2 du \ du
is the viscous stress tensor which appears under the derivative on

the right side of the momentum equation (1.3) [see, for example,
Landau and Lifshitz (1963), Sect. 49]. From the energy equation
(1.60) we may also deduce the equation for the entropy budget; in
fact, using the thermodynamic equations expressing de and dw in
terms of dp, dp, and ds, where s is the entropy of unit mass of fluid,
and also taking into account Eqs. (1.2) and (1.4), we may transform
Eq. (1.60) into the form
[see Landau and Lifshitz (1963)]. Thus, when Eqs. (L 2) and (1.4)
hold, Eqs. (1.60) and (1.62) are equivalent to each other, and either
may be used equally well for deriving a closed system of equations
for flows o f a compressible fluid.
To obtain a closed system, it is necessary only to express the
thermodynamic quantities e and or s in terms of the pressure p
and the density p (or temperature T) with the aid of the general
equations o f thermodynamics and the equations of state of the
medium under consideration (connecting p, p, and T). We shall
confine ourselves here to the simplified case when the equation of
^We should note at this point that the use here and henceforth of thermodynamic
quantities and relationships requires special reservations, since a moving fluid with nonzero
velocity and temperature gradients does not constitute a system in thermodynamic
equilibrium. It may be shown, however, that for the moderate gradients encountered in real
fluid flows, the fundamental thermodynamic quantities may be defined in such a way that
they satisfy the ordinary equations of thermodynamics of media in equilibrium. [See, e.g.,
Landau and Lifshitz (1963), Sect. 49, and also Tolman and Fine (1948), and special manuals
on the kinetic theory of gases-(Chapman and Cowling (1952), Hirschfelder, Curtiss and Bird
(1954)1.
state is that for an ideal gas, i.e., o f the form
P = R 9T (1.63)
where T is the temperature in degrees Kelvin, and the constant R .is

equal to the difference between the specific heats of the medium at
constant pressure Cp and constant volume c„;
R = c ,-c ,. (1.64)
Moreover, in accordance with the deductions of the kinetic theory of

gases and with the experimental data for real fluids, we may assume
that the specific heats Cp and are individually constant (i.e.,
independent of temperature). In this case it is not difficult to show
that e = c^T + eo, w = e +^ = CpT + eo and s = — R In p +
In 7 + const = —R In p + Cp In T + const [see, for example,
Landau and Lifshitz (1958), Sect. 43 ]. Substituting these expres
sions for e and w into Eq. (1.60) and applying Eqs. (1.2) and (1.4),
w ithout difficulty we may transform the general equation of the
heat budget into the form:
+ = + + (1.65)
or, which is equivalent, into the form
where
da^
du^ du^
du^ I/ du^ I ddu^
u^ 22 du^ ^ \ ^ dUy du^
dx^ dx.^
V’ / <^“ 3 2 du., / da., V
• ( 1 -6 6 )
The quantity e is pT multiplied by the increase in the entropy s per

unit time, caused by the transition of a part of the kinetic energy
into heat as a result of internal friction; in other words, e is the
quantity o f heat liberated per unit time per unit mass o f fluid. With
the additional presence o f influx o f heat due to radiation, chemical
reactions, phase transitions or any other causes, a further term pQ

must be added to the right sides of Eqs. (1.60), (1.62), (1.65), and
(1.65'), where Q is the additional influx of heat transmitted to unit
mass per unit time. Equations (1.2), (1.4), and (1.65) or (1.65'),
together with Eqs. (1.63) and (1.66) give a general system of five
equations for determining the five unknown functions «j(jc, t), / = 1 ,
2, 3; p (x. t) and p (j»r. t) or T (x. i).
An equation o f the type (1.65') may also be obtained in the case
of thermally inhomogeneous liquids (in the usual sense o f the word).
Here the role o f the state equation will be played by the law o f
thermal expansion of the fluid
P - P o= - P P o( 7 ' - 7 ' o) (1.67)
where p and po are the densities at temperatures T and To,

respectively, and p is the coefficient of thermal expansion. It may be
shown that under these conditions, Eq. (1.62) takes the form
while in the expression for e in the case of a liquid we may assume

dUy
that = 0 , so that
[see, for example, Howarth (1953), Vol. II, Chapt. 14]. The
coefficient p for ordinary liquids is very small (for example, for
water p 1.5 X lO"'* per degree at 15°C); thus in Eq. (1.69) we may
ignore the term containing this coefficient. We then obtain
dr , dr
dt
where X — is the coefficient of thermal diffusivity (or thermo

metric conductivity) of the medium, which here and henceforth will
be assumed to have a constant value [cf. the transition from Eq.
(1.3) to Eq. (1.4) in Sect. 1.1]. The term 7 ^ in Eq. (1.70) describes
the general heating of the medium caused by the internal friction of
the fluid; this heating under real-life conditions generally plays a
completely insignificant role and may be entirely ignored. Then Eq.,
(1.70) simplifies still further and transforms into the ordinary
equation o f heat conduction in a moving medium
We shall make wide use of this equation in subsequent work.

It is most im portant that Eq. (1.71) be applicable not only to
liquids, but also to gases, provided that the velocity of these gases is
much less than the corresponding sound velocity a = ] / f y ,
Y= — . It may be shown that at such velocities, variations in pressure

Cy
will play a considerably smaller role in Eq. (1.65') than variations in
temperature. Therefore terms containing p on the right side may be
ignored; then Eq. (1.70) may be obtained quickly or, ignoring also
the heating of the medium due to internal friction, Eq. (1.71). We
should observe that comparison o f Eqs. (1.70) and (1.65) shows that
in Eq. (1.65), unlike Eq. (1.65') it is impossible to ignore the term
— l ocal compressions and expansion caused by variations
of density during heating and cooling, must be taken into account in
the heat equation, even in the case of small velocities. However, in
the dynamic equations, for motion at velocities much less than the
sound velocity, the velocity field may be assumed to be completely
“ incompressible,” provided the variations of density produced by the
temperature inhomogeneities are small in absolute value, i.e.,
provided the absolute differences of temperature in the flow are
small in comparison with the mean absolute temperature To- Under
these conditions, it may also be assumed in the expression for e that
dujd x^= 0, i.e., e may once again be taken in the form of Eq. (1.69).
In the case o f an incompressible medium Tds = de\ thus, in an
incompressible fluid e will be exactly equal to the increase of internal
energy per unit mass per unit time, i.e., the am ount of kinetic energy
dissipated (transformed into heat) per unit time per unit mass of
fluid [cf. Landau and Lifshitz (1963), Sect. 16]. Although in the
calculation of the temperature field this quantity may generally be

ignored, it is nevertheless a very important physical characteristic of
the motion; for brevity, we shall refer to it simply as the “rate of
energy dissipation.”
We must stress once again that Eq. (1.72) is identical in form to
the diffusion equation
f + (1.72)
which describes the variation in concentration t) o f some

passive admixture in the medium (i.e., an admixture which has no
effect on the dynamics of the flow). The only difference is that in
the latter case the coefficient x must be interpreted not as the
coefficient of thermal diffusivity, but as the molecular coefficient of
diffusion. Since Eqs. (1.71) and (1.72) are identical in form,
henceforth, when considering Eq. (1.72), we shall always bear in
mind that the quantity d may denote here either the concentration
of some passive admixture or the temperature. Taking into account
the importance of the temperature variations for many applications,
for brevity, we shall always refer to 0 simply as the temperature;
however, it must be remembered that the symbol ■& may in fact
always denote the concentration of some passive admixture, and in
those cases when we do not consider the temperature as a passive
admixture, we shall denote it by T instead of 0.
The passive admixture assumption implies that the velocity field
tti(x, t) may be determined independently of the field •& from the
ordinary system of equations of fluid mechanics of an incompressible
fluid and then substituted into Eq. (1.72). If we take O to be the
temperature, then the temperature differences between different
points of the fluid must be sufficiently small for the changes which
they produce in the physical properties o f the fluid to have no effect
On the motion. However, at the same time, these temperature
differences must be large enough for the heating resulting from
internal friction to be negligible in comparison with them. In such
cases the temperature inhomogeneities will simply move with the
fluid, at the same time being smoothed out under the influence of
molecular thermal conductivity; the motion of masses o f non-
uniformly heated fluid which arises under such conditions is usually
called forced convection.
An important case o f flows, in which the temperature cannot be

considered as a passive admixture, is that of a nonuniformly heated
fluid in a gravitational field. Flows of this class arise due to
Archimedean forces, which cause an upward buoyancy of the
warmer volumes of fluid and a downward sinking of the cooler,
volumes. Such a motion of a temperature-inhomogeneous fluid is
called free convection. Let us determine the equations of motion in
this case. We assume that the velocity of our motion is sufficiently
small for the variations in the density produced by variations of
pressure (but not of temperature) to be ignored. Therefore, it follows
that we may use the ordinary equation of incompressibility (1.5) and
the Navier-Stokes equations ( 1 .6 ), taking into account in Eqs. ( 1 .6 ),
however, the body force A’ = — ge^ (where is the unit vector of the
axis Ox^ = Oz) and remembering that the density p depends on the
temperature. We now assume that the (absolute) temperature
T{xi X2, Xi, t) = T{x, y, z, t) may be put in the form T =Tq + Ti,
where To is some constant mean value, and Ti is a small deviation
from To. It is clear in this case that p = po + Pi, where po is the
constant density corresponding to To, and pi = p — pois defined as
P, = - P P o7', (1 .7 3 )
(the coefficient of thermal expansion p = — ^ ( 4 ^ ) ^

which satisfies Eq. (1.63) will clearly be equal to l/To). We note that
for T = To = const and p = po = const, the pressure po will not,
however, be constant but will decrease with height as the weight of
the column o f liquid above the point under discussion decreases:
Po = — Po^-^3 + c o n s t . (1 .7 4 )
Putting p = Po + pi, with accuracy to small first-order quantities, we

have
P ^^3 Po ".*^3 Po P(“) <^-'^3 Po <^•'^3

Thus it is clear that the third Navier-Stokes equation will be written
in this case in the form
(the presence in this equation o f a term containing T\ shows

immediately that the temperature cannot be considered a passive
admixture in this case). The first and second Navier-Stokes equa
tions, however, will have the usual form ( 1 .6 ) in this case; the term p
may be replaced from the very beginning by a constant po, while p is
taken to mean the deviation p\ o f the pressure from the mean value
o f the pressure po and is dependent on z=xz. Finally, the equation
for the temperature, as always when the medium may be assumed
incompressible, will take the form of the ordinary equations of heat
conduction o f a moving fluid
(as usual, we ignore the term with e in the equation for the
temperature). We thus obtain an approximate system o f five
equations (1.5), (1.6) w itht = 1, 2; (1.75) and (1.76) with respect to
five unknown functions {x, i), i = 1, 2, 3, pi{x, t) a n d ri(jc , t).
These equations describe the free convection of the fluid; they were
first considered by J. Boussinesq as far back as the nineteenth
century. Therefore they are called the equations of free convection
or Boussinesq equations (or Boussinesq approximation for fluid
mechanics equations). A detailed discussion of the approximations
which lead to these equations can be found, e.g., in Chandrasekhar
(196 D o r in Mihaljan (1962).
1.6 Similarity Criteria for a Thermally Inhomogeneous

Fluid; The Thermal Boundary Layer
We have already seen that two geometrically similar flows o f
incompressible fluid will also be mechanically similar provided the
Reynolds numbers o f the two flows are identical. In the case o f a
thermally inhomogeneous or a compressible fluid, however, this
assertion is not'true. For mechanical and thermal similarity to exist
between two geometrically similar flows we require that several
dimensionless characteristics, “ similarity criteria” —which we shall
now consider-should be equal simultaneously.
Let us take first the simple case o f flows of a nonuniformly heated
fluid in which the temperature may be considered as a passive
admixture. In this case the flow will be described by the usual
equations (1.5)—(1.6) of incompressible fluid mechanics (with
constant p) to which, however, must be added the equation o f heat
conduction (1.72) for the temperature. For simplicity, we shall

consider only cases o f steady motion, i.e., we assume thatUi, p an d d
are all time-independent. There are two constant coefficients v and x
in our equations (both having the dimension where L and T
symbohze the dimensions of length and time). Furthermore, the
boundary conditions of the problem in the case of geometrical
similarity will be described by some length scale L, velocity scale U,
and also a temperature difference scale •&i — do (for example, a typical
temperature difference between the rigid walls and the flow).
However, since the temperature is passive in this case, and has no
effect on the dynamics, the unit of temperature may be chosen
arbitrarily. Thus we must assume that
where 0 is a new dimension symbol independent of L and T (for

example, a “ centigrade degree”). However, in this case, it is possible
to formulate only two independent dimensionless combinations of v,
%,L, U, anddi — do; we may take for example, the Reynolds number
Re = and the Prandtl number
Pr = ^ . (1.77)
When d is the concentration of a passive admixture and not the

temperature, so that x is the diffusion coefficient, the ratio v/x is
sometimes called the Schmidt number and is denoted by Sc.
Sometimes instead of Pr it is convenient to use the Pecl 6 t number
Pe = - ^ = R e - P r (1.78)
which determines the order of the ratio o f the advective terms

and the term x^îJin Eq. (1.73), and which plays a role in the
investigation of the temperature field analogous to that o f the
Reynolds number in dynamic problems. However, the Prandtl
number has an advantage over the Pecl^t number in that it depends
only on the nature of the moving fluid (i.e., it is a characteristic o f
the medium) and not on the special features of the motion. For air
(and other diatomic gases), Pr « 0.7 and is almost independent o f the
temperature; in general, for most gases, Pr is of the order of unity.

For liquids (except liquid metals), the Prandtl number is usually
greater than unity and decreases sharply with rise in temperature; for
example, for water (at atmospheric pressure), Pr 7 at 20°C, and Pr
~ 2 at 80°C. For very viscous technical oils the Pr number is much
greater than for water, and may reach values of several hundreds or
even thousands. On the other hand, for liquid metals, which are
distinguished by their very great thermal diffusivity, the Prandtl
number takes values very much less than unity. For example, for
mercury, Pr « 0.023 at 20°C; for other molten metals this number
often proves to be even smaller (of the order of lO’^). When x is the
coefficient o f diffusion of a passive admixture, the Prandtl number,
or (what is exactly the same thing) the Schmidt number, also
possesses a very different order for gases and for liquids. In gaseous
media the coefficients o f diffusion and viscosity are of the same
order, so that v /x ^ 1. For liquids of the type of water, v/x ~ 10®,
and for very viscous liquids this ratio becomes still larger and may
reach values of the order o f 10 ®and more.
To obtain similarity of the temperature fields in two geometrically
similar flows, both the Reynolds numbers Re and the Prandtl
numbers Pr (or Re and Pe) must be equal. In general, for flows that
are only geometrically similar, the temperature fields will be deter
mined by
Re, Pr) (1.79)
which differs from the similar equation (1.31) for the velocity field
in that here the function cp now depends on two dimensionless
parameters.
In exactly the same way, if we let q denote the density of the heat
flux through the surface o f a body immersed in the fluid (i.e., we put
q — — x - ^ , where n is the outward normal to the surface of the
body), then
= • R e, P r ) .
In other words, at every point on the boundary of the body

the dimensionless ratio Nu = will depend only on the
Reynolds and Prandtl numbers. However, if we replace the heat flux
at the point by the mean heat flux <7 mper unit area of the surface of
the body, which characterizes the total heat-exchange between the
body and the fluid, then in the equation
N- = i ; ( ^ = 'K f e .P r ) (1.80)
the function (|» will now be a universal function o f the two variables,
which determine the dependence o f the heat-exchange on the
physical properties of the fluid and the scales o f length, velocity, and
temperature difference for the whole set of geometrically similar
flows. In heat engineering, the quantity Nu is called the Nusselt
number. Instead of the Nusselt number we may also use as the
characteristic o f heat-exchange the heat-transfer coefficient (other
wise called the Stanton number)
Ch= Nu Nu (1.80')
Cp?U (9, — 9o) Re Pr “ Pe *
Let us now turn to the case of free convection. Here the

differential equations of the problem contain three dimensional
coefficients v, %and (where for an ideal gas p = ^ ); the boundary
conditions will now be described by the typical length scale L and
the typical temperature difference scale Tm— To(note that there will
be no typical velocity in this case). It is clear from these quantities
that it is possible to form two dimensionless combinations. For
example, we may take the Prandtl number and the Grashof number
Qr _ (1.81)
Instead of the Grashof number, we may also use the Rayleigh

number derived from it;
Ra ^ ^ Gr • P r . (1.81')
V.
Thus, in the case of free convection in geometrically similar flows,

the velocity and temperature fields will be similar only when the
similarity criteria Pr and Gr (or Pr and Ra) are equal; generally
speaking, however, in the case of geometric similarity we shall obtain

only relationships o f the form
• P’- 7'. = ( 7 ' . - r . ) P r , O r),

Or),
where i = 1, 2, 3, <p and (p are new universal functions. In general,

for flows of an incompressible fluid, described by the complete
system of equations (1.2), (1.4), (1.63), (1.65), and (1.66), the
number o f similarity criteria is even greater; however, we shall not be
concerned with this.
For flows with very large Peclet numbers Pe = - ^ ^ (e.g., when the Reynolds number is
great and the Prandtl number is less than or of the order of unity), the term xV^ d in the
equation of heat conduction will be far smaller than the nonlinear terms Wj, ■; thus, to a
first approximation, the effect of the molecular thermal conductivity may be ignored.
However, just as in the case of the viscous effect in dynamical problems, this effect may be
ignored only at a distance from the rigid bodies immersed in the flow. A thin thermal
boundary layer may be formed close to the surfaces of such bodies. In this layer the
temperature varies rapidly from the temperature of the body , to a temperature close to
the temperature do which would be observed in the free flow if no rigid body were present.
Moreover, the effect of the thermal conductivity in this layer will be of the same order as
the effect of the temperature advection, described by the terms . Therefore, we
may deduce [in a manner similar to the deduction of Eq. (1.32)] that the thickness of
the thermal boundary layer will be of the order ^ - . In other words, it

VTT ] / Re • Pr
will also be inversely proportional to K Re and for Prandtl numbers of the order o f unity it
^ ^
will be of the same order as 6. Taking into account that a =s — x - ^ X — - , we
^ dn 5,
easily find that for large Reynolds numbers
Within the thermal boundary layer the heat-conduction equation may be simplified in a
manner analogous to the simplification of the equation of motion within the ordinary
boundary layer. In particular, in the case of a steady flow of fluid at a given temperature Oq
past a flat plate maintained at some other temperature Oi, the heat-conduction equation
within the thermal boundary layer will be written in the form
which is analogous to Eq. (1.41). For the boundary conditions, it is clear that in this case
they wiU take the form » = S, for 2 = 0 and as c o . Putting 9 (jf, i ) ^
» i-(9 ,-9 o )0 I k and substituting this equation together with Eqs.

V fj)
(1.46) and (1.47) into Eq. (1.83) we may easUy obtain
e " 4 - ^ < y 0 ' = O ; 0(O ) = O. 6 ( c o ) = l (1 .8 4 )
where (p is a solution o f Eq. ( 1 . 4 8 ) . Thus, a simple linear equation is obtained for 0 (ri). The
explicit solution of this was first given by Pohlhausen in 1 9 2 1 [see Goldstein ( 1 9 3 8 ) , Vol.
2 , Sect. 2 6 8 ; Howarth ( 1 9 5 3 ) , Vol. 2 , Chapt. 1 4 , Sect. 1 3 ; Schlichting ( 1 9 6 0 ) , Chapt.l4,
Sect. 7 , Longwell ( 1 9 6 6 ) , Chapt. 2 , Sect. 7 - 1 5 ] . . In particular, when Pr = 1 comparison of

Eqs. ( 1 . 8 4 ) and ( 1 . 4 8 ) shows that the solution o f 0 ( t ] ) has the form 6 ( t i ) = (p ' ( t i ) . Thus in
this case the temperature profile proves to be exactly similar to the velocity profile.
However, for other values of Pr, the function 0 ( t i ) may easily be found numerically from
the function q)'(T]) known from Fig. 2 (see Fig. 6).
FIG. 6. Temperature profiles in the boundary layer on a flat plate for

different Prandtl numbers.
1.7 Small Oscillations of a Compressible Fluid

Flows o f compressible fluid described by the general system of equations (1.2), (1.4),
(1.63), (1.65), and (1.66) generally are very complex, and their theoretical study involves
great difficulties. We shall restrict ourselves at this stage to the simple case of small
oscillations about a position of rest (or motion with constant velocity) for which the
method of linearization of the equations may be used. Carrier and Carlson (1946), Yaglom
(1949) and Kovasznay (1953) have shown that in this case all possible motions may be
divided into oscillations of three types, with considerable differences in character. This
division will play an important role in the discussion of isotropic turbulence in a
compressible gas and the processes of wave propagation in a turbulent medium in Volume 2
of this book.
L et us c o n sid e r a sta tio n a ry m ass o f id eal gas o f c o n s ta n t d e n s ity po and c o n s ta n t

pressure Po (fo r a gas m o v in g a t c o n s ta n t v e l o c it y , ii m a y , o f c o u r se , b e r e d u c e d t o th e case
o f a sta tio n a ry gas b y tr a n sfo r m in g to a n e w in ertial s y ste m o f c o o r d in a te s ). We n o w a ssu m e
th at at th e in itia l in sta n t sm all d istu rb a n c es arise in th e gas, ch a r a cter iz e d b y sm all varia tio n s
of th e h y d r o d y n a m ic q u a n titie s W/ ( x , t), p' (JC, 0 = P (•*. 0 — Po (-*» 0 =
p { x , t) — pg. Where 1’ <C * 1 (here <jo=

Po Po ^0 y Po
Cp
is th e s o u n d v e lo c ity in th e u n d istu r b e d m e d iu m , ^ = — ). In th is c ase, th e tim e-v a ria tio n
Cl,
o f th e field f U/ (x, t), p' ( x , t ) a n d /?' ( x , t ) in th e first a p p r o x im a tio n m a y b e d e te r m in e d
fro m th e s y ste m o f e q u a t io n s ( 1 .2 ) , ( 1 . 4 ) , ( 1 . 6 3 ) , ( 1 . 6 5 ) , a n d ( 1 . 6 6 ) , lin ea r iz e d w ith r e s p e c t
to th e v e lo c ity c o m p o n e n ts , a n d w ith r e s p e c t to th e d e v ia tio n s o f pressure, d e n s ity and
te m p er a tu re fro m th e c o r r e sp o n d in g u n d istu r b e d v a lu es /?o, po a n d T q = T h e n th e

Kpo
e q u a tio n of sta te (1 .6 3 ) p erm its us to e lim in a te im m e d ia te ly te m p er a tu re fr o m th e
u n k n o w n s a n d to o b ta in a c lo se d s y ste m o f fiv e e q u a tio n s w ith fiv e u n k n o w n s . A s
u n k n o w n s , w e m a y ta k e, fo r e x a m p le , th e th re e c o m p o n e n ts o f v e lo c ity and s o m e tw o
sin g le-v alu ed fu n c tio n s o f pressure a n d d e n s ity . A c c o r d in g to K o v d szn a y ( 1 9 5 3 ) a n d Chu
and K o v d szn a y ( 1 9 5 8 ) , w e m a y tak e as th e s e t w o fu n c tio n s th e d im e n sio n le ss pressure
P = an d th e e n tr o p y d iv id e d b y : S = = lo g ------ f- c o n s t. T h e n to w ith in
sm all q u a n titie s o f h ig h er order:
= (P - Po) - (S - So). = (7 - 1) ( P - P o ) + ( S - So).

Po ^0
T h e lin ea r iz e d s y s te m o f h y d r o d y n a m ic e q u a tio n s in th e variables Ui, P a n d S w ill ta k e th e

form :
dP dS , du^ „ (1 .8 5 )
dt dt dx^
^ + + < -1.2.3. (..86,
|p -X V *5 -(T (-l)x v V = a (1.87)
T ra n sfo rm in g E qs. ( 1 . 8 5 ) a n d ( 1 . 8 6 ) b y in tr o d u c in g as u n k n o w n s in ste a d o f th e c o m p o n e n t s

du^
o f th e v e lo c it y fie ld th e c o m p o n e n ts o f th e v o r tic ity a n d o f th e
d iv e r g e n c e D = , fr o m E q . ( 1 . 8 6 ) w e h a v e
ox^
— •vV*o)ft = 0. *= 1 ,2 .3, (1.88)

dt
dt
—V, D+ P = 0, = (1-89)
w h ile E q . ( 1 . 8 5 ) , ta k in g E q . ( 1 . 8 7 ) in t o a c c o u n t, b e c o m e s
S + D = 0. (1 .9 0 )
F irst, w e o b se rv e th a t th e v o r tic ity c o m p o n e n ts (Os o c c u r o n ly in E qs. ( 1 . 8 8 ) w h ic h

agree w ith th e lin ea r iz e d e q u a t io n s fo r a v o r tic ity fie ld in a n in c o m p r e s sib le flu id . In th is
c o n n e c t io n w e recall th a t in an in c o m p r e s sib le flu id w e m a y fo r m u la te th e v e lo c it y fie ld Uf
u n iq u e ly in te r m s o f th e v o r tic ity fie ld d)k a n d th e c o r r e sp o n d in g b o u n d a r y c o n d it io n s ; in a
c o m p r e ss ib le m e d iu m , o n th e o th e r h a n d , th e v e lo c it y fie ld m a y b e p r e se n te d in th e fo r m
o f a su m o f in c o m p r e s sib le (s o le n o id a l) a n d ir r o ta tio n a l ( p o te n tia l) c o m p o n e n t s , th e latter
o f w h ic h is in d e p e n d e n t o f th e v o r tic ity fie ld . T h u s fo r m o t io n w h ic h rep re sen ts o n ly a
w e a k p e r tu r b a tio n o f th e s ta te o f r est, th e s y s te m o f h y d r o d y n a m ic e q u a tio n s in th e first
a p p r o x im a tio n d iv id e s in t o a c lo se d s y s t e m o f e q u a t io n s in th e c o m p o n e n t s o f th e v o r tic ity
(O/i, d e scr ib in g th e in c o m p r e s sib le f lo w , and a s y ste m o f e q u a tio n s in D, P a n d S ,
d e scr ib in g th e ir r o ta tio n a l p o t e n tia l f lo w . In th is a p p r o x im a tio n , th e flu c t u a tio n s o f
pressure an d e n tr o p y are c o n n e c t e d o n ly w ith th e c o m p r e ss ib le irro ta tio n a l f lo w , i.e ., th e y
are a b se n t in th e in c o m p r e s sib le (v o r tic it y ) c o m p o n e n t o f f lo w . In h ig h er a p p r o x im a tio n s in
th e th e o r y o f sm a ll d istu rb a n c es, th e s e t w o c o m p o n e n ts w ill in te r a c t w ith e a c h o th e r , th u s
c a u sin g a d d itio n a l ch a n g e s in pressure a n d e n tr o p y (w e shall d iscu ss th is b r ie fly at th e e n d
o f th is s u b s e c tio n ).
In ord er to in v e stig a te in d e ta il th e ty p e s o f m o t io n d e scr ib e d b y th e lin ea r iz e d s y ste m
( 1 . 8 7 ) - ( 1 . 9 0 ) and , in p articular, to s h o w th a t th e e q u a tio n s in D, P a nd S a c tu a lly d e scr ib e
tw o d iffe r e n t ty p e s o f o s c illa tio n s , w e u se th e ord in a ry F o u rier m e th o d s . F o r th is p u rp o se
w e a ssu m e th a t all th e fie ld s iHj (X, ^), D ( x , t ) , P ( x , /) , and S { X t t ) are p e r io d ic fu n c tio n s
o f th e sp a ce c o o r d in a te s w ith a g iv en w a v e -n u m b e r v e c t o r ^ . M oreover, w e r ep la ce th e
o rdinary tim e t b y th e d im e n sio n le ss tim e t = aqki^ w h e r e ^ = and r ep la ce th e
D
d iv erg en ce b y th e d im e n sio n le ss d iv e r g e n c e :
0 )y (X. f) = loy (X) = D (T) e ' * * ,
P (X. t ) = P (x ) S (X. t) = S (x )
T h e n E qs. ( 1 . 8 8 ) ta k e th e fo r m
fifo)y(x) v/s
coy, 7 = 1 , 2 , 3. (1 .9 1 )
dz ao
a n d th e r e m a in in g e q u a tio n s are tr a n sfo r m e d in t o th e f o llo w in g s y ste m o f ord in a ry lin ear
e q u a tio n s :
di ciq ciq (1 .9 2 )
di ^ ao
T h e re fo r e it f o llo w s th a t th e a m p litu d e s o)j ( / ) o f th e c o m p o n e n ts o f th e v o r tic ity fie ld w ill

b e d a m p e d w ith tim e , a c co r d in g to th e law
a>y(0 = c o y ( 0 ) e - ’*’' . (1 .9 3 )
w h ile th e a m p litu d e s D ( / ) , P ( / ) , an d S ( 0 o f th e v e lo c ity d iv e r g e n c e, pressure an d

e n tr o p y fie ld s w ill b e o f th e fo r m
w here Aî, X2 a n d >.3 are th ree r o o ts o f th e c h a ra cteristic e q u a tio n o f th e s y s te m ( 1 .9 2 ) :
x ,+ iiL ± n ) lx = + /i+ J :^ \x + 3 * = o . a . 94)

^0 \ % ) %
A fu rth er sim p lific a tio n o f th e r esu lts m a y b e o b ta in e d if w e o b se rv e th a t E qs.
( 1 . 9 1 ) - ( 1 . 9 4 ) c o n ta in d im e n sio n le ss para m eters o f very d iffe r e n t ord ers o f m a g n itu d e . In
fa c t, all th e c o e f fic ie n ts o f th e s y ste m ( 1 . 9 2 ) and E q . ( 1 . 9 4 ) m a y b e e x p r e s s e d in te r m s o f
th e f o llo w in g th ree d im e n sio n le ss param eters:
Vi* V, Cp
h o w e v e r, in Eqs. ( 1 .9 1 ) , th e d im e n sio n le ss p a ram eter I = —

V app ears, w h ic h d iffe r s fr o m 6 1
^0
4 4 C
o n ly in th e cha n g e o f th e v is c o s ity c o e f f ic ie n t = — i _ - . | -------w h ic h
o po po
d e fin es th e in tern a l fr ic tio n in a p o te n tia l flo w , fr o m th e ordin a ry k in e m a tic v is c o s ity v . B u t
Jt| is analogous to the ordinary Prandtl number P r — y is of the same order of

m ag n itu d e; th u s in a g a seo u s m e d iu m Jti ^ 1; ju st as y 1 (fo r air y 1 -4 ). A t th e sa m e
tim e 5 = — is o f th e sam e o rder as th e ratio JL o f th e le n g th o f th e m e a n fr e e p a th o f th e

do A
2tz
gas / to th e w a v e len g th o f th e d istu rb a n c e A = — (sin c e v ^ v l w h e r e v is th e r o o t m e a n
k
square v e lo c ity o f th e th erm a l m o t io n o f th e m o le c u le s , a ' ^ v a n d 1 /A ). T h u s, in all
cases w h e n th e m o t io n o f th e gas m ay b e d e scr ib e d b y ordinary flu id m e c h a n ic a l e q u a tio n s
6 1 (in particular, for air under normal conditions — « 0.5 X lO '^ cm, i.e., even if
ao
A « 1 mm, 6 < 10“^). M oreover, = — B is o f th e sam e order o f m a g n itu d e , and h e n c e
6 ,C l.
N o w , ig n o r in g in E qs. ( 1 . 9 1 ) - ( 1 . 9 2 ) th e term s o f order 6 (o r 6 i), w e o b ta in th e
s im p lifie d s y ste m
£fa) ; ( 0 d S {t) d D (t) d P {t)

- r - " - - r - o - o,(..9«
In a c c o r d a n c e w ith th e first tw o e q u a tio n s o f E q . ( 1 . 9 6 ) , th e v o r tic ity fie ld (Oj (JC, 0 a n d th e

e n tr o p y fie ld S (X,/ ) in th e a p p r o x im a tio n u n d er d is c u s sio n w ill b e fix e d in sp a c e. T h e last
tw o e q u a tio n s o f E q. ( 1 . 9 6 ) s h o w th at
o r, p r o c ee d in g at o n c e fr o m th e single F ou rier c o m p o n e n t t o th e to t a l fie ld s D {x , i) and

P (X , t),
d ^ D 2 i-f2 j-s 2 tt2 rx
T h u s fo r th e pressure fie ld P(X, i) a n d th e fie ld o f th e v e lo c ity diverg en ce D ( x , t) w e have

o b ta in e d id e n tic a l w a v e e q u a tio n s w h ic h d escrib e w a v es p ro p a g a ted at th e s o u n d v e lo c ity Oo-
We see th a t in th e z er o th -o r d e r a p p r o x im a tio n (w ith resp e c t to a sm all pa ra m eter 6 ) the
d istu rb a n c es o f th e h y d r o d y n a m ic q u a n tities break d o w n in to th ree n o n in te r a ctin g
c o m p o n e n ts . T h ese are c a lle d th e incompressible vorticity mode w h ic h is d e scr ib e d b y the
v o r tic ity fie ld a>y (jc) an d is tim e-in v a ria n t (o r d is p la c e d w it h o u t cha ng e at th e un pertu rb ed
v e lo c ity ), th e entropy mode w h ic h is d escrib ed b y th e also sta tio n a ry (or d is p la c e d w ith
v e lo c ity U) e n tr o p y fie ld S ( x ) p r o d u c e d b y in itia l te m p er a tu re in h o m o g e n e itie s , and th e
poten tia l or acoustic mode w h ic h is c o n n e c te d w ith flu c tu a tio n s o f th e p o te n tia l part o f th e
v e lo c ity fie ld an d w ith flu c t u a tio n s o f pressure, and w h ic h is r ep resen ted by a set o f w aves
p r o p a g a ted at th e u n p e rtu r b e d so u n d v e lo c ity Aq-
I f w e a lso ta k e in t o a c c o u n t e f f e c t s o f order 6 , th en w e fin d at o n c e th a t th e v e lo c ity
fie ld o f th e v o r tic ity m o d e w ill b e s lo w ly d a m p e d un der th e in flu e n c e o f v is c o s ity in
a c co r d a n ce w ith Eq. ( 1 . 9 3 ) . I f w e w ish to w rite d o w n th e e n tr o p y and a c o u s tic m o d e s also
w ith a c cu ra c y o f 6 , it is c o n v e n ie n t first, to rew rite Eq. ( 1 . 9 4 ) in th e form
+ + + + = (1 - 9 4 ')
\ Tlj / \
w h ic h c o n ta in s o n ly th e para m eters ( 1 .9 5 ) . It is th e n e a sy to d e te r m in e th e th ree r o o ts o f

E q. ( 1 .9 4 ) t o term s o f ord er b y th e u sual p o w e r series m e th o d . A s a r esu lt, w e o b ta in
1h _- l ; . ’'i+ T ----
— 18.„
1
X 3=_— 8,.
T h e se e q u a t io n s s h o w th a t i f w e restrict o u rselv es to term s o f ord er n o t e x c e e d in g 6 i, and

e lim in a te th e v o r tic ity m o d e fr o m th e d is c u s sio n , th e n in a c o m p r e ssib le m e d iu m th ere w ill
e x is t th re e d iffe r e n t w a v e s w ith th e sam e w a v e-n u m b er v e c to r k. T w o o f th e s e w a v e s w ill b e
d a m p e d w ith th e sam e d e c r e m e n t -------^ --------- Ojao« = ------------ 2 ------------ and w ill b e
p r o p a g a ted w ith th e u n d istu r b e d s o u n d v e lo c ity in th e d ir e c tio n o f th e v e cto rs k and

— k \ th e th ird w a v e w ill b e sta tio n a ry w ith r e s p e c t to th e u n d istu r b e d f lo w , and w ill have
decrem ent (It is n o t d if fic u lt t o s h o w th a t e v en i f te r m s o f o rder 6 i are

ao
ta k e n in t o a c c o u n t, th is c h a n g e s o n ly th e im agin a ry parts o f th e r o o ts Xi a n d ^ 2 ,i.e ., it leads
to th e a p p ea r a n c e o f a w e a k d e p e n d e n c e o f th e v e lo c ity o f p r o p a g a tio n o f th e a c o u s tic
w a v e s o n th e c o e f fic ie n ts o f v is c o s ity and th er m a l d iffu s iv ity [c f. Y a g lo m ( 1 9 4 8 ) ] . ) B y
a ssu m in g fu rth er th a t th e d e p e n d e n c e o f D ( r ) , P ( t ) and S ( t ) o n t w ill b e d e te r m in e d by
th e fa c to r e I = 1, 2 , 3 , a n d s u b s titu tin g th e c o rr e sp o n d in g particular s o lu tio n s in th e
s y ste m ( 1 . 9 2 ) , it is e a sy to s h o w th a t th e g en era l so lu tio n o f th is s y s te m , t o term s o f order
6 i in c lu siv e , m a y b e p r e se n ted in th e fo r m o f a su m o f th e fo llo w in g tw o s o lu tio n s:
S (x ) = 5oeH P ( t) = 0, D ( x) = — ( 1- 99)
a nd
D (x ) = + D < 2 V ’^],
(1.100)
P(T) = D » ) [ , - + - M + 1 6, 1/-^ + D(2) [ - / 4 - 5,
T hus it is clear th a t in th e a p p r o x im a tio n u n d e r c o n s id e r a tio n , th e e n tr o p y m o d e o f ou r
d istu rb a n c e w ill c o r r e sp o n d to an arbitrary in itia l e n tr o p y (o r te m p er a tu re ) d is tr ib u tio n
w h ic h d o e s n o t m o v e in sp a ce b u t w h ic h is grad u a lly s m o o t h e d o u t un d er th e a c tio n o f
th erm a l c o n d u c t iv ity . C o n n e c te d also w ith th is m o d e is th e very w e a k ( o f ord er 6 i) v o lu m e
d istr ib u tio n o f th e p o t e n tia l m o d e o f v e lo c ity and th e v e lo c ity d iv e r g e n c e D = - ^ ^ ^ .T h is

OXg^
d is tr ib u tio n is set up b y th erm a l e x p a n s io n a n d c o m p r e ss io n o f e le m e n ta r y v o lu m e s o f th e
th er m a lly in h o m o g e n e o u s m e d iu m . T h e a c o u s tic m o d e , h o w e v e r , w ill b e c o m p o s e d o f
p ressu re w aves and v e lo c ity d iv e r g e n c e w aves w h ic h are of th e fo rm
v ,+ (y -l)x
i {kx ± aokt)----------L----- kH
Doe ^ , p r o p a g a ted w ith v e lo c ity ao, and sligh tly d a m p e d b y
m o le c u la r v is c o s ity an d th erm al c o n d u c t iv ity ; th e s e w a v e s g e n e ra te also an e x tr e m e ly w e a k
transfer o f e n tr o p y ( o f th e ord er o f 61 ).
T h is r e s o lu tio n o f arbitrary d istu rb a n c es o f h y d r o d y n a m ic fie ld s in t o v o r tic ity , e n tr o p y
a n d a c o u s tic m o d e s m a y e a sily b e e x te n d e d t o all ord ers o f m a g n itu d e w ith r e s p e c t t o 6 a nd
6 1 ; h o w e v e r , in p r a c tic e , in all ca ses it is s u ffic ie n t to r estrict o u r selv e s t o th e te r m s g iv en
a b o v e . O f s o m e w h a t greater sig n ifica n ce is th e q u e s tio n o f th e c h a n g e s in d u c e d in in d iv id u a l
m o d e s o f th e d istu rb a n c es b y th e n o n lin e a r term s in th e h y d r o d y n a m ic e q u a tio n s . In order
to ta k e th is e f f e c t in t o a c c o u n t, it is n e c essa ry t o p r o c e e d to th e n e x t a p p r o x im a tio n in
s m a ll-d istu rb a n ce th e o r y w h ic h c o n s ists o f preserving in E qs. ( 1 . 8 7 ) - ( 1 . 9 0 ) te r m s th a t are
b ilin ea r w ith r e s p e c t t o th e h y d r o d y n a m ic q u a n titie s . In th e s e term s th e b a sic q u a n titie s
m a y b e d e fin e d a c c o r d in g t o th e e q u a t io n s o f th e fir st a p p r o x im a tio n , i.e ., th e y are a ssu m ed
t o agree w ith th e s o lu tio n s o f lin ea rized e q u a tio n s g iv e n a b o v e . T h u s E qs. ( 1 . 8 7 ) - ( 1 ; 9 0 ) in
th e su b s e q u e n t a p p r o x im a tio n w ill c o n ta in s o m e fairly sm all a d d itio n a l c o m p o n e n t s o f
k n o w n fu n c tio n a l fo r m , w h ic h are n a tu rally tran sferred t o th e righ t side o f th e e q u a tio n and
c o n s id e r e d as th r e e -d im e n sio n a l “ s o u r c e s ” o f th e c o rr e sp o n d in g h y d r o d y n a m ic fie ld s. T h e
s o lu tio n o f th e s e e q u a tio n s w ill d iffe r fr o m th a t giv e n a b o v e fo r th e h o m o g e n e o u s e q u a tio n s
( 1 . 8 7 ) - ( 1 . 9 0 ) b y th e a d d itio n a l term s p r o d u c e d b y th e c o rr e sp o n d in g so u r ce s, i.e ., b y
b ilinear c o m b in a tio n s or “ in te r a c tio n s ’* o f in d iv id u a l c o m p o n e n ts o f th e lin ear th e o r y w ith
ea ch o th e r . A gain w r itin g th e s o lu tio n o f th e lin ea r iz e d e q u a tio n s in th e fo rm o f th e sum o f
v o r tic ity , e n tr o p y an d a c o u s tic m o d e s , w e o b ta in six d if fe r e n t paired b ilin ea r (an d
q u a d r a tic ) c o m b in a tio n s o f th e s e th re e m o d e s , i.e ., six d iffe r e n t “ in t e r a c tio n s ,” e a c h o f
w h ic h , in its tu rn , c a n m a k e d e fin ite a d d itio n s t o s o lu tio n s d e scr ib in g a n y o f th e th re e
m o d e s , th a t is, th e y “ g e n e r a te ” th is m o d e .
H o w e v e r , it is natu ra l th a t m a n y o f th e 18 s ec o n d -o r d e r e f f e c t s th u s p r o d u c e d w ill b e
e x p r e s s e d b y b ilin ear c o m b in a tio n s th a t c o n ta in th e fa c to r s 6 or 6 1 , a n d w ill th e r e fo r e p la y
a v ery sm a ll r o le . T h u s, fo r e x a m p le , fr o m th e fa c t th a t (t o term s o f o rd er 6 1 ) t h e e n tr o p y
m ode c o n ta in s o n ly flu c t u a tio n s of e n tr o p y , we ca n dedu ce th a t in th e z e r o th
a p p r o x im a tio n w ith r e s p e c t t o 6 , th is m o d e w ill n o t in ter a c t w ith it s e lf at all, w h ile its
in te r a c tio n w ith th e v o r tic ity m o d e w ill in tr o d u c e a d e fin ite c o n tr ib u tio n [ c o n n e c te d ,
e v id e n tly , w ith th e a d v e c tiv e te r m s «« o f E q. ( 1 . 6 2 ) ] o n ly in th e e n tr o p y m o d e o f
th e d istu rb a n c e fie ld . A c o m p le te a n a ly sis fr o m th is v ie w p o in t o f all th e p o s sib le

“ in t e r a c tio n s ” m a y b e fo u n d in th e p a p er b y C hu a nd K o v a szn a y ( 1 9 5 8 ) . T h e r esu lts o f th is
a n a ly sis are illu stra ted b y th e ta b le b e lo w , w h ic h in d ic a te s th e o rd er w ith r e s p e c t t o 61 o f all
18 s ec o n d -o r d e r e f f e c t s , w ith a b r ie f e x p la n a t io n o f th e p h y sic a l m e a n in g o f th e z ero -o rd er
e ffe c ts . In a d d itio n to th is, fo r all z ero -o rd er e f f e c t s w ith r e s p e c t t o 6 |, C h u a nd K o v a szn a y
G e n e r a tio n
I n te r a c tio n A c o u s t ic V o r tic ity E n tr o p y
M ode M ode M ode
0 ( 1)
a c o u s tic m o d e 0 ( 6 ,) 0 ( 6 ,)
“ sound» 5 ca tte r in g b y
w ith a c o u s tic
s o u n d ” an d “ n o n lin
m ode
ear d is to r tio n o f
so u n d w a v e s
0 ( 1) 0 ( 1)
v o r tic ity m o d e “ g e n e r a tio n o f “ str etch in g o f v o r te x 0 (6 ,)
w ith v o r tic ity sou n d b y lin e s ” or “ in ertial
m ode v o r t ic it y ” transfer o f th e vor
tic ity b y a r o ta t
io n a l v e lo c ity fie ld
e n tr o p y m o d e w ith 0 (5 ,) 0 (5 2 ) 0 (8 ,)
e n tr o p y m o d e
0 ( 1) 0 ( 1)
a c o u s tic m o d e “ sca tte rin g o f 0 (5 ,)
s o u n d b y r o ta “ tra n sfer o f vo r
w ith v o r tic ity
tio n a l in h o m o t ic it y b y s o u n d
m ode
g e n e itie s o f th e w aves”
v e lo c it y f ie ld ”
0 ( 1) 0 ^1 ) , 0 ( 1)
a c o u s tic m o d e w ith “g e n e r a tio n o f v or
e n tr o p y m o d e “ s ca tte rin g o f
tic it y b y th e in ter
s o u n d b y te m p er a
a c tio n o f e n tr o p y “ tra n sfer o f h e a t
tu re in h o m o
a n d p r e ssu r e ” b y s o u n d -w a v e s ”
g e n e itie s ”
(B jerk n es e ff e c t ) .
0 ( 1)
v o r tic ity m o d e 0 (5 ,) 0 ( 8 .)
“ tra n sfer o f h e a t
w ith e n tr o p y
b y v o r t ic it y ”
m ode
give e x p lic it e x p r e s s io n s fo r th e c o rr e sp o n d in g th r e e -d im e n sio n a l so u r ce s w h ic h m u st b e

a d d e d t o th e right sid e o f th e lin ea r iz e d e q u a tio n s . H o w e v er , w e shall n o t s tu d y in d e ta il all
th e zer o -o r d e r e f f e c t s , a c o n s ise r a b le p r o p o r tio n o f w h ic h , as m a y b e s ee n fr o m th e ta b le,
refer t o a sim p le s h ift o f h y d r o d y n a m ic a l in h o m o g e n e itie s b y th e v e lo c ity fie ld fr o m o n e
flu id e le m e n t t o a n o th e r . We shall c o n s id e r a m o r e d e ta ile d d is c u s sio n o f o n ly th e m o s t
im p o r ta n t in te r a c tio n o f th e v o r tic ity m o d e w ith it s e lf (reca llin g th a t in an o r d in a r y , w e a k ly
c o m p r e ss ib le m e d iu m , th e v o r tic ity m o d e as a ru le, c o n s id e r a b ly e x c e e d s all th e o th e r s in
m a g n itu d e ).
S in ce th e v o r tic ity m o d e is c o n n e c te d o n ly w ith th e d istu rb a n c es o f th e v e lo c ity fie ld , its
in te r a c tio n w ith it s e lf w ill arise o n ly fr o m b ilin ea r term s w ith r e s p e c t to v e lo c ity in th e
h y d r o d y n a m ic e q u a tio n s . In th e e n tr o p y e q u a tio n (i.e ., h e a t b u d g e t) th er e are su c h bilin ea r
term s (in th e term p e ) b u t w ith th e c o e f f ic ie n t o f v is c o s ity , fr o m w h ic h it is clear th a t th e
e n tr o p y m o d e g e n e r a tio n b y th is in t e r a c tio n w ill b e o f ord er n o t e x c e e d in g 6 . A m u c h m o r e
ds
im p o r ta n t r o le w ill b e p la y e d b y th e te r m s o f th e e q u a tio n s o f m o t io n w h ic h are
bilinear w ith r esp e c t to th e v e lo c ity ; th e s e w ill d e te r m in e aU th e “ s ig n ific a n t” sec o n d -o r d e r

e ffe c ts , c o n n e c te d w ith th e in te r a c tio n o f th e v o r tic ity m o d e w ith its e lf.
K e e p in g th e te r m s in th e e q u a t io n s o f m o t io n , a n d th e n p r o c e e d in g t o th e
OXfi
v o r tic ity e q u a t io n , w e o b t a in E q . ( 1 . 7 )
w h ic h d iffe rs fr o m E q. ( 1 . 8 8 ) in th e rig h t s id e w h ic h d e scr ib e s th e p r o c e s s o f g e n e r a tio n o f

v o r tic ity d u e t o th e str etch in g o f th e v o r te x lin es. A ssu m in g in th is right sid e th a t th e fie ld
Ua is in c o m p r e s sib le an d sa tisfie s E q . ( 1 . 8 8 ) , w e o b ta in th e first sig n ifica n t “ sec o n d -o r d e r
e f f e c t ” - t h e g e n e r a tio n o f th e v o r tic ity m o d e o f th e v e lo c ity fie ld b y th e in te r a c tio n o f th is
m o d e w ith its e lf. T h e e f f e c t o f th is in te r a c tio n o n th e a c o u s tic m o d e m a y b e d e te r m in e d in
th e z er o a p p r o x im a tio n w ith r e s p e c t t o 6 , o m it tin g in th e h y d r o d y n a m ic e q u a t io n s all term s
c o n ta in in g th e c o e f fic ie n ts o f v is c o s ity a n d th er m a l d iffu s iv ity , a n d all n o n lin e a r te r m s
dui
e x c e p t th e term s in th e e q u a t io n s o f m o t io n . B u t in th is c a se it is e a sy t o see th a t
in ste a d o f th e last e q u a t io n o f E q. ( 1 . 9 7 ) w e o b ta in
4 “-S-
H ere th e right sid e e v id e n tly d e scr ib e s th e a d d itio n a l flu c t u a tio n s o f pressure w h ic h arise
fr o m th e flu c t u a tio n s o f t h e v e lo c ity fie ld in a c c o r d a n c e w ith E q. ( 1 . 9 ) w h ic h h a v e th e fo r m
S u b stitu tin g in t o th e rig h t sid e o f E q . ( 1 . 1 0 1 ) fo r m/(JC, / ) an in c o m p r e s sib le v e lo c ity fie ld

sa tisfy in g th e “ v o r tic ity e q u a t io n ” ( 1 . 8 8 ) o r ( i f w e d o n o t a ssu m e th a t th e v e lo c ity
flu c t u a tio n s are sm a ll) E q . ( 1 .7 ) , w e m a y fin d th e s o u n d fie ld g e n e ra ted b y th e in te r a c tio n
o f an in c o m p r e s sib le v e lo c it y fie ld w ith its e lf; th e right sid e o f E q . ( 1 . 1 0 1 ) , e v id e n tly in th is
dd^ duo ^2 (f/ 1/
)
case m a y also b e w r itte n as — ;;------- ^ or a s _________ ^____ t - . T h e g e n e r a tio n o f so u n d b y a
^-<3 dx^dx^
r o ta tio n a l f lo w d e s c r ib e d b y E q . ( 1 . 1 0 1 ) w a s s tu d ie d b y L ig h th ill ( 1 9 5 2 ; 1 9 5 4 ); th is is th e
s e c o n d sig n ifica n t sec o n d -o r d e r e f f e c t c o n n e c te d w ith th e in te r a c tio n o f th e v o r te x m o d e o f
th e v e lo c ity fie ld w ith its e lf.
2. HYDRODYNAMIC INSTABILITY AND TRANSITION

TO TURBULENCE
2.1 Concept of Turbulence; Empirical Data on Transition to

Turbulence in Tubes and Boundary Layers
In the preceding section, equations were given describing fluid
motion, and some of the simpler solutions were indicated. We
pointed out also that these solutions do not by any means always
correspond to any flow that may actually be observed. Thus, in Sect.
1.2, Example 3, we noted that the flow in a tube is described by Eqs.
(1.2 3 )-(1 .2 6) only in the case of sufficiently high viscosity and
sufficiently low mean velocity, while in Sect. 1.4 we stressed that the
Blasius solution o f the boundary-layer equation on a flat plate gives
good agreement with the data only in the case of fairly small values
of Uxh. The same is true for all other examples of fluid flows. As a
rule, the theoretical solutions of the equations of fluid mechanics,
whether exact or approximate, gives a satisfactory description of real
flows only under certain special conditions. If these conditions are
not met, the nature o f the flow undergoes a sharp change, and instead
of a regular smooth variation of the fluid mechanical quantities in
space and time, we observe irregular fluctuations of all these
quantities which have the complex nature indicated in Fig. 1.
Thus, flows of fluid may be divided into two very different classes:
smooth, quiet flows which vary in time only when the forces
acting or the external conditions are changing, known as laminar
flows; and flows accompanied by irregular fluctuations of all the
fluid mechanical quantities in time and space, which are called
turbulent flows.
The difference between laminar and turbulent regimes of flow is
revealed in a number o f phenomena which are of great significance
for many engineering problems. For example, the action of a flow on
rigid walls (i.e., friction on walls) in the case of a turbulent regime is
considerably greater than in the case of a laminar regime (since the
transfer of momentum in a turbulent medium is much more
intensive). The presence of irregular fluctuations of the velocity leads
also to a sharp increase in mixing of the fluid; extremely intensive
mixing is often considered as the most characteristic feature of
turbulent motion. With increase of the mixing, there is also a sharp
increase in the thermal conductivity of the fluid, etc. For all these
reasons, the determination of the conditions of transition from a
laminar to a turbulent regime is a very pressing problem. Moreover,
finding the mechanism o f the initiation of turbulence must aid our
understanding of its nature and facilitate the study of the laws of
turbulent flow, which are of great importance in practical work.
A detailed survey o f the experimental data relating to the
transition from laminar to turbulent flow is to be found in the
articles of Schlichting (1959), Dryden (1959), and Tani (1967),
containing many additional references. Many hundreds of scientific
papers and several dozens of survey articles are devoted to the theory
of this question [see, e.g., the surveys by Stuart (1963; 1965), Shen
(1964), Reid (1965), Segel (1966), Drazin and Howard (1966), and
Gortler and Velte (1967)]; see also the special monographs of Lin
(1955), Chandrasekhar (1961), and Betchov and Criminale (1967).
In this book we shall confine ourselves to a consideration of the most
important (and simplest) flows only, and shall devote most of our
attention to the principles involved. For the details of the mathe
matical calculations and a detailed description of the experiments,
and also for matters relating to more complex flows, the reader is
referred to the aforementioned sources and special articles.
The first results on the conditions for transition to turbulence
were obtained by Hagen (1839). Hagen studied flows of water in
straight circular tubes of fairly small radius, and established that with
gradual decrease in the viscosity of water (brought about by
increasing its temperature), the velocity of flow for the same
pressure-head first increases to some limit and then begins to
decrease again. The jet of water issuing from the tube before this
Hmit is reached has a smooth form, but after passing through this
limit it suffers sharp fluctuations. Hagen explained these phenomena
by saying that for sufficiently low viscosity values, internal motions
and vortices are produced, leading to increased resistance, and,
consequently, to a decrease in the velocity of flow. Hagen showed
also that variation in the nature of the flow may be effected by
changing the pressure-head (i.e., the mean velocity) or the radius of
the tube; he could not, however, obtain any general criterion for the
transition from laminar to turbulent flow.
A general criterion for transition to turbulence was established by
O. Reynolds (1883) using the concept of mechanical similarity of flows
of viscous fluid. This criterion consists of the fact that the flow will
be laminar so long as the Reynolds number Re = f7Z,/v does not
exceed some critical value Rec-, while for Re > Recr it will be
turbulent. As explained in Sect. 1.3, the Reynolds number has the
significance of a ratio of characteristic values of the forces of inertia
and viscosity. The inertia forces lead to the approach of initially
remote fluid volumes and thus contribute to the formation of sharp
inhomogeneities in the flow. The viscous forces, on the other hand,
lead to the equalizing of velocities at neighboring points, i.e., to the
smoothing of small-scale inhomogeneities. Thus for small Re, when
the viscous forces predominate over the inertia forces, there can be
no sharp inhomogeneities in the flow, i.e., the fluid mechanical
quantities will vary smoothly and the flow will be laminar. For large
Re, on the other hand, the smoothing action of the viscous forces
will be weak, and irregular fluctuations—sharp, small-scale inhomoge
neities—will arise in the flow, i.e., the flow will become turbulent.
The Reynolds criterion is explained by these considerations.
To verify this criterion experimentally and to measure Recr,
Reynolds carried out a series of experiments with water flows in
circular glass tubes connected with a reservoir. In these experiments a
source of colored fluid was placed on the axis of the tube at the
intake. For small Re, the colored water took the form of a thin,
clearly defined jet, which indicated a laminar regime of flow. As Re
increased, at the instant of passing through the critical value, the
form of the colored jet sharply changed; at quite a small distance
from the intake into the tube, the je t spread out and waves appeared
in it; farther on, separate eddies were formed and towards the end of
the tube the whole of the liquid was colored. If in such an
experiment the flow is illuminated by an electric spark, it may be
seen that the colored mass consists o f more or less distinct swirls
which indicate the presence of vorticity. Transient phenomena are
observed for subcritical Re numbers close to Recr in a laminar flow.
These phenomena consist of the appearance of short-term flashes of
high-frequency fluctuations, which were first observed by Reynolds,
in the form of peculiar “ turbulent slugs” which fill the entire cross
section of the tube but only for fairly short sections of its length.
(According to Schiller (1934), the appearance of such “bursts” is
connected with the alternate formation and breakdown of large-scale
vortices on the inner wall of the tube, at its upstream end). In the
initial part o f the tube, with Re > Recr, the flow has a similar
character. But as Re increases, the length of this initial part where
the flow is not entirely turbulent, decreases rapidly. According to
R otta’s measurements (1956), in an intermediate regime of this kind,
the mean fraction of the time in which a turbulent regime is observed
at a given point (the “ coefficient of interm ittency” ) increases
monotonically with increase of the distance x from the intake into
the tube. The existing experimental data (the most detailed being
Lindgren’s (1957; 1959; 1961) allow this increase to be explained by
the fact that the local velocity of the leading edge of a “ turbulent
slug” exceeds the local velocity of its trailing edge and hence the
“slug” increases in length as it moves along the tube (occasionally
some slugs will overtake others, coalescing to form a single, larger
slug).
Reynolds’ experiments were carried out in tubes of various

diameters with a smooth intake connected to the reservoir. Variation
of the numbers Re = Dl\ (where U,„ is the mean velocity o f flow)
was accomplished both by changing to a new tube (with another
value of the diameter D), and by varying the rate of flow and the
viscosity o f the water (by altering the temperature). The value of
RCpr in these experiments was on the average close to 12,830.
However, such results were obtained only by exercising the greatest
care for the absence o f disturbances in the water before entering the
tube. Reynolds’ investigations showed that the value of Recr,
corresponding to the transition from laminar to turbulent flow,
depends considerably on the degree of disturbance in the water
entering the tube (or, as we say, the “ initial turbulence,” which is
determined principally by the conditions at the intake of the tube)
and therefore in different experiments may prove to differ very
much. It is impossible to produce an absolutely steady flow in the
tube. Therefore, disturbances will always be in the flow, charac
terized by relatively rare fluctuations o f velocity, the origin o f which
is partly connected with the vortices separating from the leading edge
of the tube. The intensity of these disturbances (which may be
characterized by the parameter U'/Um where U' is the typical
magnitude of the fluctuations of the x-component of velocity) may
be fairly large, but due to their infrequent recurrence these
disturbances by themselves do not alter the laminar nature of the
flow and have no considerable effect on the velocity profile, the
pressure drop or other mean characteristics o f the flow. For
sufficiently small Reynolds numbers, the disturbances which arise are
attenuated when displaced downstream. However, as the Reynolds
number increases, at the instant that it attains its critical value
(which depends on the intensity of the disturbances and possibly on
their scale and frequency), the disturbances suddenly generate
turbulence. It is found that the value of R ec r corresponding to the
transition to turbulence is the smaller, the greater the intensity o f the
disturbances. Thus, for example, in the case of a tube with a sharp
entrance, pushed through the plane wall of the reservoir, the end o f
the tube will create considerable disturbances, and R ecr will equal
approximately 2800. Conversely, if the degree of disturbance at the
intake into the tube is decreased strongly by some means or other,
we can delay the transition from laminar to turbulent flow until very
high Reynolds numbers (this is called persistence o f the laminar
regime). Thus Barnes and Coker (1905) were able to delay the
transition to turbulence in tubes to values Recr == 2 0 ,0 0 0 , and

Ekman (1911) up to Recr « 50,000. Later, Comolet (1950) obtained
laminar flow in a tube at Reynolds numbers in the range
70,000—75,000, and Pfenninger (1961) [who arranged twelve special
screens for damping the flow disturbances between the inlet o f the
tube and the atmosphere] found that a tube flow can be fully
laminar even at Re = 100,000.
These results show that the Reynolds number in itself is not a
unique criterion for transition to turbulence; for a flow in a tube it is
apparently impossible to find a universal critical value Recr such that
for Re > Recr, the flow regime is bound to be turbulent. To establish
the upper value of Re for laminar flows in tubes it is necessary to
have some knowledge o f the level of inlet turbulence of the laminar
flows considered. Thus the upper bound of the Re numbers of
laminar flows will generally be a function of the parameter U'lUm
(which apparently increases monotonically as the parameter de
creases).
Without knowing the degree of disturbance of the laminar flow,
we can find only a rather weak criterion which indicates the
conditions under which only a laminar regime of flow is possible. For
this we must determine the critical Reynolds number Recrmin
corresponding to the transition from laminar to turbulent flow for
the largest possible level of disturbance of the laminar flow at the
intake into the tube. With Re < Recrmin the flow will always remain
laminar, i.e., any disturbance, regardless of its intensity, will be
damped.
Experiments to measure RCcrminWere performed by Reynolds
himself. Since in these experiments it is necessary to introduce into
the tube fluid that is as disturbed as possible, the method of dye
injection is clearly unsuitable here. As a result, we must determine
the transition to a turbulent regime in some other way (for example,
by the variation o f the skin friction law which determines the
dependence o f the mean velocity on the pressure drop). In Reynolds’
experiments the minimum critical value o f Re = C'mD/v was found to
be Recrmin ~ 2030. Values close to this (lying between 1800 and
2320) were also obtained in all subsequent investigations [including
the recent works by Lindgren (1959; 1961) and Sibulkin (1962)].
Similar results were also valid for flows in channels o f rectangular
cross section [see, e.g., Narayanan and Narayana (1967)].
Results on the criteria for transition to turbulence in many ways
similar to the above were obtained in the study of boundary-layer
flows around a solid body. Let us consider, for example, the

boundary layer formed on a flat plate by a flow of constant velocity
U, flowing parallel to the plate. The Reynolds number of the
boundary layer may be defined, e.g., by Res = Ui/v where 6 is the
thickness of the boundary layer. Alternatively we may use the more
easily measured number Re* = Ux/v where x is the distance from the
leading edge of the plate measured along the flow. The numbers Rej
and Re* are connected by a functional dependence; for example, for
a laminar flow, by the results of Sect. 1.4, Rej 5 ]/ Re^ [cf. Eq.
(1.49)]. Proceeding downstream, both Rej and Re* increase, and at
some point Xct, they attain the “ critical value” when the flow
sharply changes its nature and becomes turbulent. Thus, for x < Xcr
(more precisely, for Res < Reger and Re* < Re.t cr) the flow in the
boundary layer will be laminar while for x > Xcr (i.e., Rej > Re scr
and Re* > R e * c r ) it will be turbulent. In the immediate neighbor
hood of X = A'cr a “ mixed regime” is formed in which only discrete
“bursts” of turbulence are observed; these arise at a point in the
form of “ turbulent spots” which grow in size and coalesce with each
other as they move downstream [see, e.g., the survey papers by
Schlichting (1959), Sect. 14, Coles (1962), Stuart (1965) and
also Elder (I9 6 0 )]. The appearance of “ turbulent spots” leads to the
presence, at points near Xcr of alternating laminar and turbulent
regimes; at the beginning of the transition region both the frequency
of occurrence and the duration of the turbulent regime is negligible,
while at the end o f it, the laminar regime occurs only very rarely and
briefly.
The first measurements of the critical Reynolds number of a
boundary layer were carried out in 1924 by Burgers and Van der
Hegge Zijnen who studied a flow of air past a flat glass plate in a
wind-tunnel. Shortly thereafter, similar measurements were also
made by Hansen (1928) [see, in particular. Fig. 4 ]. According to the
data of these authors
Re^„ = ( i ^ ) - 3 . 1 0 5 ^ 5 - 10 ®.
V ^ /cr
which corresponds to
Re5cr = (— ) ~ 2 7 5 0 - h 3500,
\ ^ /cr
which is of the same order o f magnitude as the value of Recr for flow
in a tube. Later, it was shown that as in the case of flows in tubes,
the critical Reynolds number of a boundary layer depends consider
ably on the disturbance level of the ambient flow, depending on
which Re* may vary from 1 X 10® to almost 3 X 10® (see Fig.'7,
which illustrates the dependence of Rexcr on U'lU, where 3 U'îs
taken equal to the mean square velocity fluctuation of the ambient
flow). However, we observe in Fig. 7 that the value of Re* c r w i l l not
increase without bound, but will tend to a definite limit (of the order
3X 10® ) above which the boundary layer will be turbulent no matter
how small the disturbance level of the initial flow. On the other
hand, in the case o f flow in a circular tube, experiment does not
indicate a sharp reduction in the rate of increase of Recr as U'lU
tends to zero; thus it seems likely that here Recr-^<» as U'IU - ^ 0
(i.e., for any value of Re, the flow in a tube will remain laminar
provided that a sufficiently low disturbance level of the intake flow
can be guaranteed). We shall discuss this important difference
between flows in boundary layers and in tubes in greater detail in
Sect. 2.8.
FIG. 7. D e p e n d e n c e o f th e critica l R e y n o ld s n u m b e r fo r th e b o u n d a r y layer

o n a flat p la te o n th e free-strea m d istu rb a n c e lev el [a fter D r y d e n ( 1 9 4 9 ) ) .
(T h e d iffe r e n t s y m b o ls o n th e figu re d e n o t e th e d ata o f d iffe r e n t in vestig a
to r s.)
The fact that the critical Reynolds number is considerably smaller

for flow along a rough plate with natural or artificial irregularities is
connected with the turbulizing effect of disturbances in the ambient
flow. According to the data of numerous measurements, even one

isolated irregularity can bring about the transition of a laminar
boundary layer into a turbulent one, provided that the height Hq of
this irregularity is o f the order of the “ displacement thickness” §* »
0.38 o f the boundary layer in its neighborhood. Even more
important is the effect of a number of irregularities scattered over
the whole plate [see, e.g., Schlichting (1959), Chapt. X or Dryden
(1959), Sect. 5]. The value of Re* cr is also changed considerably by
even quite small longitudinal pressure gradients in the incident flow.
However, we shall not dwell on this point, but confine ourselves to
considering only one special effect connected with the presence of a
negative pressure gradient.
2.2 Turbulent Flow Past Solid Bodies; Boundary-layer
Separation, the Drag Crisis and the Mechanism
of Boundary-layer Transition
Transition to turbulence in a flow o f viscous fluid past a solid
body may occur not only in the boundary layer, but also by the
formation of a turbulent wake behind the body due to the separation
of macroscopic vortices from its surface.
FIG . 8 . S c h e m a tic fo r m o f t h e s tr ea m lin e s (d o t t e d )

a n d th e v e lo c ity p r o file s (s o lid ) a t d if fe r e n t p o in ts
in th e f lo w p a st a right c y lin d er .
The formation o f a turbulent wake is generally connected with the

retarding action of a negative longitudinal pressure gradient in the
flow. Let us consider, for example, a right circular cylinder washed
by an irrotational flow perpendicular to its axis (see Fig. 8 which
shows the flow past the upper part of the cylinder). Outside the
boundary layer the fluid may be assumed ideal and its m otion—
irrotational. The streamlines of this potential motion are closest
together on the upper part o f the cylinder (point C) where,
consequently, the tangential velocity u will attain a maximum. By
the Bernoulli equation
«2 ,
-g- + p/p = const
du 1 dp
[which follows from the equation given in Sect.
1.4, immediately before Eq. (1.38)] the pressure at C in the outer
flow will attain a minimum, so that it will decrease along AC and
increase along C E . Such changes in the pressure along the surface of
the body will also take place in the boundary layer (since across the
boundary layer the pressure hardly varies). Consequently, on CE the
fluid in the boundary layer must move in the direction of increasing
pressure, which leads to its retardation. This retardation will have the
strongest effect, of course, on the fluid particles moving on the very
surface of the cylinder, i.e., possessing the least velocity. At some
point D downstream from C, these particles will come to a standstill;
beyond D they will be moving backwards with respect to the fluid
particles further from the cylinder surface which have not yet been
retarded. The formation o f counterflows on the surface of the body
beyond D forces the outer flow away from the surface of the
cylinder; and there occurs what we call ‘ separation of the boundary
layer” from the surface, with formatior of a dividing stream-surface
DF in the fluid. It is clear that if the velocity U decreases sufficiently
rapidly beyond C, separation of the boundary layer is bound to
occur. Even if the boundary layer is laminar before separation, after
separation it will behave as a free submerged jet and will quickly
become turbulent (for considerably smaller Re than for an unsepa
rated boundary layer since the presence of the wall has a stabiUzing
effect on the flow). The dividing stream-surface DF, which is a
surface of tangential discontinuity of the velocity, is very unstable
(see below) and is quickly transformed into one or more vortices. In
the region FDE beyond the dividing stream-surface a large-scale
vortex is formed close to the cylinder, with a second such vortex
being formed on the lower part of the cylinder. These vortices
separate in turn from the cylinder, are carried downstream and
gradually are dispersed; in their place new vortices are formed (see,
e.g., the figures in Hinze (1959), p. 8 ).
As a result, a turbulent wake is formed behind the body in which
the motion is rotational, while outside the layer the motion is
irrotational (i.e., potential). In fact, the fluid outside the boundary
layer may be assumed to be ideal; it follows, therefore, that during
its motion, the circulation of the velocity along any closed contour is
conserved, and hence in the case of steady motion the curl of the
velocity is constant along streamlines. Therefore, it is evident that a
region of turbulent rotational flow at a distance from a body can
only arise when streamlines leave the boundary layer (in which the
motion will be rotational due to the viscosity), i.e., when there is a
direct mixing o f the fluid from the boundary layer with that o f the
space outside it.
It is also clear that the streamlines cannot leave the region of flow
in which the curl o f the velocity is nonzero, i.e., the region o f the
turbulent wake (although they can enter the wake from the region of
potential flow). In other words, fluid can flow into the turbulent
wake from the potential region but cannot flow out o f the turbulent
wake. At the same time, turbulent fluctuations of velocity can
penetrate from the wake into the region of potential flow, although
with considerable attenuation. In fact, for potential motion o f an
incompressible fluid, the equations o f motion (1.7) will be satisfied
identically. Therefore, in this case the flow will be described by the
single condition o f incompressibility (1.5). This condition is equiva
lent to Laplace’s equation 9 = 0 for the velocity potential <p, which
defines the velocity: m, = d<fldxi. Let z be the coordinate across the
wake. Then the field y, 2 ) which describes the velocity
fluctuations may conveniently be decomposed into periodic compo
nents of the form 9 = 9 0 (2 ) Fr omv^ <p = 0 , it follows that
= where k = V k \- { - k l is the wave number, inversely
proportional to the horizontal (Ojcj/-plane) scale o f the periodic
fluctuations under consideration. Discarding the physically meaning
less solution for cpo, which increases with increasing 2 , we find that
the attenuation o f the fluctuation amplitude in the region 2 > 0 is
given by the factor Thus the fluctuations are attenuated the
quicker, the smaller the scale. Consequently, at a sufficient distance
inside the potential motion, only comparatively smooth large-scale
fluctuations arise. For such fluctuations the energy dissipation does
not play a large role; thus almost all the dissipation in the flow will
take place in the rotational turbulent wake.''
The considerable energy dissipation in the whole region of the
turbulent wake, together with the formation of a dividing stream-
surface leads to a considerable increase in drag of bodies having
’ T h e a n a ly sis d e scr ib e d b e lo n g s t o L a n d a u a n d U f s h it z ( 1 9 6 3 ) ; it a lso p la y s th e cen tra l
ro le in d e ta ile d in v e s tig a tio n s o f th e v e lo c ity fie ld o u t s id e a tu r b u le n t r eg io n o f flo w m a d e
b y PhilUps ( 1 9 5 5 ) a n d S tew a r t ( 1 9 5 6 ) .
separated boundary layers. This drag, as a rule, will be the smaller,

the narrower the turbulent wake, i.e., the further along the surface of
the body is the point of separation. For sufficiently large Reynolds
numbers for which, however, the boundary layer remains laminar up
w
to the separation point, the drag coefficient Cm — ------- (where W
is the total drag and 5 is the area of the body or its cross section) is
independent of Re because the position of the separation point is
independent of Re (it is found from the equation = 0 in
=0
which it may be shown that the Reynolds number does not occur).
However, as we approach Reynolds numbers for which the boundary
layer becomes turbulent before the separation point for a laminar
boundary layer, the separation point will move downstream, and in
this case the turbulent wake becomes considerably narrower and the
drag of the body decreases sharply (by several times). This
phenomenon is called the drag crisis. It is explained by the fact that
the momentum transfer within a boundary layer increases sharply
when it becomes turbulent. Therefore, the entrainment of fluid from
the outer flow by the boundary layer is considerably increased so
that the fluid particles in the boundary layer move in the direction of
increasing pressure much longer than in the case of a laminar
boundary layer.
The drag crisis in the case of flow past a sphere was first observed
by Eiffel (1912). The transition from large to small drag occurs in
this case for Reynolds number Re = UD/v (where D is the diameter
of the sphere) close to 5.0 X 10®; the drag coefficient decreases,
approximately from 0.5 at Re = 10® to 0.15 at Re = 10®. Later, it
was also found that the minimum value of Cw in very careful
experiments is less than 0.1. The coefficient Cw for a circular
cylinder behaves in the same way. The dependence of the drag
coefficient for a sphere and a circular cylinder on Re is shown in Fig.
9. It is clear from the above discussion that the drag crisis will arise
the earlier, the greater the disturbance level of the ambient flow, i.e.,
the smaller the critical Reynolds number for transition to a turbulent
regime in the boundary layer. This is confirmed clearly by the
experiments of Prandtl (1914) who tried to achieve passage through
the drag crisis in the flow past a sphere by fitting a wire ring around
the sphere, i.e., by introducing additional disturbances into the flow
to cause transitions in the boundary layer.
The necessary condition for separation of the boundary layer is an

increase of pressure in the direction of flow along some part of the
surface of the body. This condition is satisfied not only for flow past
convex surfaces, but also in other cases, e.g., flow in an expanding
conical tube (diffuser) or in a sharply bent tube. In these cases, also,
separation of the boundary layer may occur.
FIG. 9. D e p e n d e n c e o f th e drag c o e f fic ie n ts o f

a sphere and a circular c y lin d e r o n th e R e y n o ld s
n u m b er.
T h e sep a ra tio n o f th e b o u n d a r y la y er u n d er th e a c tio n o f a n e g a tiv e lo n g itu d in a l pressure

g ra d ien t m a y also e x p la in to a certain e x t e n t th e e f f e c t o f d istu rb a n c es in th e a m b ie n t flo w
o n th e v a lu e o f R c x c r . We m a y assu m e th a t th is e f f e c t is c o n n e c te d w ith th e g e n e r a tio n o f
flu c t u a tio n s o f th e lo n g itu d in a l pressure gra d ien t b y th e s e d istu rb a n c es, le a d in g to th e
ra n d o m fo r m a tio n o f in d iv id u a l s p o ts o f u n sta b le S -shap ed v e lo c ity p r o file s (as at p o in t E ,
Fig. 8 ) a nd h e n c e t o sep a ra tio n and tr a n sitio n o f th e b o u n d a r y layer. O n th e b a sis o f th is
h y p o th e s is , G. I. T a y lo r ( 1 9 3 6 a ) tried to e stim a te th e o r e tic a lly th e d e p e n d e n c e o f th e critical
R e y n o ld s n u m b e r R e x cr o n th e in itia l tu rb u le n c e lev el o f th e a m b ien t flo w (later, W ieghardt
( 1 9 4 0 ) gave a sh o rter d e d u c t io n o f T a y lo r ’s results).
In his w o r k , T a y lo r p r o c e e d e d fr o m th e a p p r o x im a te v o n K a r m a n - P o h lh a u s e n th e o r y o f
a lam inar b o u n d a r y la y er in th e p resen ce o f a lo n g itu d in a l pressure g ra d ien t d pjdx.
A c c o r d in g to th e th e o r y th e fo rm o f th e v e lo c ity p r o file in d iffe r e n t s e c tio n s o f th e
b o u n d a r y la yer d e p e n d s o n ly o n a single d im e n sio n le ss p aram eter A = ------- . In a
b o u n d a r y la yer o n a fla t p la te d p ld x = 0, b u t th ere m a y e x is t flu c tu a tio n s o f pressure. T h u s

T a y lo r p r o p o s e d th a t th e character o f th e m o t io n in a fix e d s e c tio n is d e te r m in e d h ere b y
th e p aram eter A = (w h ere p ' is th e pressure flu c tu a tio n and 6 / 6 x sig n ifies a
ty p ic a l va lu e o f th e deriv a tiv e d/d x). In o th e r w o r d s, a c co r d in g to T a y lo r , th e p o in t o f

tr a n sitio n fro m lam inar to tu rb u le n t flo w is d e te r m in e d b y th e param eter A a tta in in g s o m e
1 hp'
critica l value. B u t th e e q u a tio n s o f m o t io n s h o w t h a t ----- -- m u st b e o f th e sa m e order
... 1 Zu'
o f m a g n itu d e as a ' ^ — w h e r e u' is th e f lu c t u a tio n o f th e lo n g itu d in a l v e lo c ity
5 a*
/2
in th e a m b ie n t flo w . F u rth e r, w e m a y p u t , w h e r e [ / ' i s a t y p ic a l v a lu e o f th e
v e lo c ity flu c t u a tio n s , a nd \ is th e so -c a lle d T a y lo r m ic r o sc a le o f tu r b u le n c e w h ic h is
d e te r m in e d fr o m th e c o n d it io n
/ hu' \21
I = (th is sca le w ill b e u se d several tim e s in
later s e c tio n s o f th e b o o k ) . T h e scale X m a y b e e x p r essed in term s o f th e e x te r n a l (in teg ra l)

scale o f tu r b u le n c e L (w h ic h d e fin e s th e order o f m a g n itu d e o f th e g r ea test d is ta n c e at
w h ic h s o m e r e la tio n b e t w e e n th e in s ta n ta n e o u s va lu es o f th e v e lo c ity flu c t u a tio n s is
/ \2
apparent) as follows: the mean rate of energy dissipation e ^ v I I is, on the one hand,
v i/'*
p r o p o r tio n a l t o — p — a n d , o n th e o th e r , fo r large v a lu e s o f ( 7 'L /v , it is a lso p r o p o r tio n a l to
U ' ^ j L (w e shall d isc u ss th is fa c t in g reater d e ta il in V o lu m e 2 o f th is b o o k ) . C o n s e q u e n tly ,
tp' u ,2
Ix
T a k in g in t o a c c o u n t th a t fo r a lam inar b o u n d a r y la y er o n a fla t p la te 6 ^ we

o b ta in
B2 hp' ( u 'Y h ( X yi2 ( Ux yi»

I t ) I— J •
T h u s, ta k in g R e x c r = cp(A cr), w e o b ta in
Ux\ U'
(2.1)
R e,
/cr u
W e m a y ta k e th e le n g th L in th is e q u a t io n t o b e s o m e ch a r a cter istic d im e n s io n o f t h e d e v ic e
g e n e ra tin g th e tu r b u le n c e (f o r e x a m p le , i f th e tu r b u le n c e is s e t u p b y a grid in a
w in d -tu n n e l, th e n L w ill b e a p p r o x im a te ly eq u a l t o th e d is ta n c e b e t w e e n th e r o d s o f th e
g rid).
QtZ
//I //
C/ 1 iu/
0M8 ■ -o
<
k ^r=?2.7cm
^D=2l7cm
----
ao4
0.8 f.2 16 2.0 ZA 2.3
Ijf / DVh
F IG . 1 0 . D e p e n d e n c e o f o n j j - (- 2 " j for flo w
past a sp here.
T h e resu lt ( 2 .1 ) , a c co r d in g t o w h ic h Re^r d e p e n d s o n ly o n - q - * has b e e n d e d u c e d
here fo r th e case o f f lo w pa st a fla t p la te. H o w e v e r , it m a y b e g e n e ra lize d t o f lo w s p ast o th e r

b o d ie s b y rep la cin g th e c o o r d in a te x w ith .9 , th e d is ta n c e fr o m th e p o in t w h e r e th e flo w
im p in g e s o n th e b o d y t o th e p o in t o f tr a n sitio n o f th e b o u n d a r y la y er fr o m a la m inar t o a
tu r b u le n t reg im e , r e c k o n e d a lo n g th e c o n to u r o f th e b o d y , or b y s o m e c h a ra cteristic
d im e n sio n o f th e b o d y D , an d b y u sin g , in ste a d o f R e , « Ux}y/, R e , = 6 / s / v or R e = U D I\ •
In sp ite o f t h e la c k o f rigor in its d e d u c t io n , th is r esu lt is in v ery g o o d a g r e e m e n t w ith
e x p e r im e n ta l d a ta (s e e , e .g .. F ig. 1 0 w h ic h giv es d a ta o f D r y d e n , S c h u b a u e r , M o ck and
S k ram stad ( 1 9 3 7 ) , w h o m e a su r ed th e v a lu e o f R e cr fo r b o u n d a r y lay ers o n sp h eres o f
d iffe r e n t radii w ith d iffe r e n t v a lu e s o f th e in t e n s it y an d sca le o f th e tu r b u le n c e in th e free
strea m ).
2.3 Hydrodynamic Instability

In the theoretical analysis of the transition problem, we must
start from the fact that the velocity and pressure fields in any fluid
flow, whether laminar or turbulent, are solutions of the equations of
fluid mechanics for given initial and boundary conditions. Steady
laminar flow, in particular, is described by steady solutions o f these
equations; however, in the case of turbulent flow, each individual
example o f the flow corresponds to a very complex nonsteady
solution o f the dynamical equations. The impossibility o f the
existence o f laminar flow for sufficiently high Reynolds numbers (in
spite of the fact that the equations of fluid mechanics have a steady
solution for any Re) clearly shows that not every solution cor
responds to a real fluid motion. It is natural to associate this fact
with the familiar postulate that real motion must not only satisfy the
dynamical equations but must also be stable to small disturbances. In
other words, small disturbances of the motion which are always
present, must be damped in time so as not to change the general
nature o f the flow. On the other hand, if these disturbances increase
with time, then considerable distortion of the initial motion occurs
and, consequently, this motion cannot continue to exist for any
considerable time.
Therefore, we may expect that the value of Rccr corresponds to
the point at which stability is lost; for Re < Recr the laminar flow is
stable, and with Re > Recr it is unstable and becomes turbulent
under the influence of the existing small disturbances. But if this is
the case, then by the mathematical study of laminar solutions o f the
equations of motion, we may (at least in principle) theoretically
determine the corresponding critical Reynolds number.
Turbulence is characterized by very complicated irregular fluctua
tions of the velocity and other characteristics of the flow. The
mechanics of the initiation of these fluctuations may be explained

from the viewpoint o f general mechanics. It is only necessary to
consider the flow o f fluid as a dynamical system with a very large
number of degrees o f freedom in which self-excited oscillations arise
as a result of the influx of energy from outside.
The appHcation of the concept of degrees of freedom necessitates
immediate introduction of some generalized coordinates which
describe the configuration of the flow uniquely. To define these
generalized coordinates we may begin from a decomposition o f the
motion into elementary components. These components must
be such that the sum of all their energies is equal to the total
energy o f the flow while the state of each of them is characterized by
a fairly small number of parameters. The parameters o f all the
elementary components o f motion will be generalized coordinates of
the flow, and the number of these coordinates which can vary under
given external conditions will be the total number of degrees of
freedom of the flow. From a mathematical viewpoint, decomposition
of the motion into elementary components is equivalent to the
expansion of the velocity field in terms of an orthogonal system of
functions; each o f these functions will describe the velocity field of
the corresponding elementary component of the motion, while the
coefficients o f the expansion will be generalized coordinates of the
flow. The choice o f an arbitrary orthogonal system of functions is
dictated by the form o f the boundaries of the flow. For flows in a
finite volume, the orthogonal system o f coordinates will always be
countable; thus, these flows will have no more than a countable set
o f generalized coordinates.®
For steady laminar flow the values of the generalized coordinates
will be defined uniquely by given external and boundary conditions
so that the number of degrees of freedom of a laminar flow is zero.
The number o f degrees of freedom o f a turbulent flow in a finite
volume will be very great, but it is also finite. In fact when the
velocity field is expanded in a series in terms of orthogonal
functions, the various components will describe elementary motions
* T h e c o n c e p t o f a f lo w in an in fln ite sp a ce w ill a lw a y s b e an id e a liz a tio n . U sin g th is
id e a liz a tio n , in s te a d o f e x p a n d in g th e v e lo c it y fie ld as a series in term s o f an o r th o g o n a l
s y s te m o f fu n c tio n s , w e m u st u se an e x p a n s io n sim ilar t o th a t o f a F o u rier -S tie ltje s integral
(w h ic h w ill b e u se d e x te n s iv e ly in V o lu m e 2 ) and t o a llo w th e e x is t e n c e o f a c o n tin u o u s
s p e c tr u m . T h e r e fo r e , a n u n b o u n d e d f lo w o f flu id m a y p o sse ss e v e n a c o n tin u u m o f
g e n e ra lize d c o o r d in a te s . H o w e v e r , in b o u n d e d sp a c e-tim e r eg io n s, flu c t u a tio n s w ith a
c o n t in u o u s s e t o f w a v e n u m b e r s, o f c o u r se , m a y a lso b e a p p r o x im a te d b y a c o u n ta b le
n u m b e r o f h a r m o n ic o s c illa tio n s .
of different scales. As the order o f the component increases

indefinitely, the corresponding scale tends to zero. However, due to
the viscosity, fluctuations of too small a scale cannot exist. Thus
with steady external conditions the coefficients of the expansion of
the velocity field in terms of orthogonal functions of sufficiently
high order will be independent of time. This means that the number
o f degrees of freedom of the flow will be finite. Also, the number of
degrees of freedom must increase with decrease of the coefficient o f
viscosity, in other words, with increase of Re. According to the
estimate of Landau and Lifshitz (1963), the number of degrees o f
freedom of a turbulent flow in a finite volume will be proportional
to Re'/‘ for large enough Re, where Re is the Reynolds number o f the
overall flow (this result will be discussed in Volume 2). Consequently,
the number o f degrees of freedom will increase rapidly with increase
of Re, and for developed turbulence with large Reynolds numbers it
will reach enormous values.
If we are to describe not only the configuration o f a system but
also its change in time, we require in addition to the values o f its
generalized coordinates the values of the corresponding generalized
velocities. The choice of values of all the generalized coordinates and
generalized velocities defines some point in the phase space of a
system which characterizes completely its instantaneous state. The
change of the state o f the system is described in the phase space by a
definite lin e-th e phase trajectory of the system. The plotting o f the
phase trajectory is a convenient way o f describing the evolution o f a
system.
Let us consider the evolution of a fluid flow under fixed steady
external conditions (in particular, with constant influx of energy)
but with various initial conditions. Each set of initial conditions has a
corresponding phase trajectory, coming from the corresponding
phase point. The question o f the behavior of these trajectories over
long intervals of time is an interesting one. From statistical
mechanics it is known that dynamical systems with a large number of
degrees of freedom and steady external conditions are inclined to
converge to some limiting equilibrium regime in which the mean
influx of energy equals the mean dissipation of energy of the system,
and the total energy has a fixed value and fixed distribution among
the different degrees of freedom. We may put forward the hypothesis
that for a broad class o f flows two possible limiting regimes
exist—laminar and turbulent. Thus every phase trajectory of a flow
will in the course of time either tend asymptotically to the point
corresponding to laminar flow or else approach some “limit cycle”

corresponding to a steady turbulent regime. The criterion for
transition to turbulence must permit us to predict from the starting
point o f the phase trajectory which of the two limiting regimes will
occur.
The possibility of decomposing the motion of the fluid into
elementary components means that a flow may be considered as a set
o f interacting elementary nonlinear oscillators. In each o f these
oscillators, due to the influx of energy, self-excited oscillations may
arise. The possibility of such oscillations arising is determined by the
relationship between the energy E* acquired by the oscillator and the
energy E~ lost by it for various ampHtudes of the oscillations a (see
Fig. 11). If E~ > E* for all ampHtudes (Fig. 11a), then the
oscillations will evidently be damped for any initial amplitude, and
the system will be stable to any disturbances. If E~ < £+ for
ai < a < Co, but E~ > E* for a < Oi or a > flo (Fig. 1 lb), then the
oscillations with initial amphtude a<ai will be damped, but those
with initial amplitude a>Ui will increase, until their amplitude
attains the equilibrium value ao. In this case, the system will be stable
to small disturbances but unstable to disturbances of sufficiently
large amplitude (such a system is called a system with hard
self-excitation). Finally, if £+ > £ - for any amplitude, however small
(Fig. 1 Ic), the system will be unstable to infinitely small dis
turbances (i.e., absolutely unstable) and will practically always be in
a regime of self-excited oscillation with amplitude Oo (system with
soft self-excitation).
a) b) c)
FIG . 1 1 . D iffe r e n t variants o f th e d e p e n d e n c e o f th e en erg y g a in ed and lo st b y an
o s c illa to r o n th e a m p litu d e o f th e o s c illa tio n s .
From the discussion below, it is very natural to think that in fluid

flows all three situations shown in Fig. 11 may arise; however, the
exact conditions which would allow us to determine in every case
which situation actually exists in a given flow are still unknown.
2.4 Simple Examples of Absolutely Unstable

Fluid Flows
We cite above the experimental data relative to the dependence of
Recr for flows in tubes and boundary layers on the intensity o f the
initial disturbances and the persistence o f the laminar regime. These
data show that for R e slightly greater than Recr mm the flows will be
self-excited systems with hard excitation (one possible mechanism of
the excitation o f oscillations in such systems is given by Taylor’s
theory outlined in Sect. 2.2).
We shall now show some simple examples of fluid motions that are
unstable even with respect to infinitely small disturbances; i.e., from
the viewpoint o f the theory of oscillations they are systems with soft
excitation.
One o f the simplest examples of an absolutely unstable flow is the
flow near a surface o f tangential velocity discontinuity which we
have mentioned above. In this case,
the absolute instabihty may be e x - _ ________ _
plained qualitatively with the aid of
simple physical considerations. Let us
consider an ideal fluid of zero viscos
ity, two layers of which slide over
each other with equal and opposite
1 j o . r FIG . 1 2 . S c h e m a tic fo r m o f th e stream -
velocities U and - U, forming a surface d is tr ib u tio n d o s e
of discontinuity of velocity. Let us t o a d istu rb ed surfa ce o f ta n g e n tia l
assume that as a result of some dis- v e lo c it y d is c o n tin u ity ,
turbance on the surface of discon
tinuity, a small-amplitude wave is formed (see Fig. 12). For
simplicity, we assume that this wave is nonprogressive. Under these
circumstances, the streamlines above the wavecrest will draw closer
together, i.e., the velocity will increase, while in the troughs the
streamlines will become further apart and the velocity will decrease.
According to Bernoulli’s equation, u^l2 + p/p = const, the pressure
will fall above the crest and rise in the troughs (in Fig. 12 this is
denoted by the plus and minus signs). Thus a transverse pressure
gradient arises in the fluid, tending to increase the amplitude of the
wave. Later, this increase in amplitude leads to the wave disintegrat
ing into individual vortices, forming the beginning o f the turbulent
zone.
In a real fluid, o f course, the waves which arise are progressive, but
the processes o f their evolution are similar. These processes may be
observed for example, in experiments with a jet issuing from an

orifice and then expanding in a space filled with the same (but
motionless) fluid (the boundary of such a je t may be considered as a
surface o f tangential velocity discontinuity). An accurate quantita
tive analysis of the instability of a surface o f tangential velocity
discontinuity was obtained first by Helmholtz (1868) [cf. Lamb
(1932), Sect. 232, or Landau and Lifshitz (1963), Sect. 30]. For
viscous fluid, the sliding o f the two layers over each other is, of
course, impossible, and instead o f the surface of discontinuity, there
exists between the two flows a narrow transition layer in which the
velocity profile will be S-shaped. The investigation of the stability of
such a layer will be more complicated; however, here we may show
also (both theoretically and experimentally) that it is very unstable
(see below. Sect. 2.8). We note further that the absolute instability
of a surface o f tangential velocity discontinuity is the simplest case
of so-called Helmholtz instability— absolute instability of a
special type o f surface of discontinuity, separating two regions of
flow, fllled with the same or different fluids, moving with different
velocities. A survey of the main results in this field and references to
later literature may be found in Birkhoff (1962).
Another simple example of absolute instability is the equilibrium
in a gravitational field of stationary stratified fluid with variable
density p = p ( 2 ) increasing with height. Consequently, with any
function p {z) the equations o f motion of incompressible fluid will
allow a solution » (x, y, z, t) = 0 , corresponding to a state of rest;
the gravitational field will only produce a vertical pressure variation
according to the law
uu
1 dp
i.e., p{z) = g J ^ {z') dz'-\-cons\. ( 2 .2 )
Now suppose that as a result o f some disturbance, some element of

the fluid is displaced from level 2 to a new level z' = z + h. U the
density p decreases with height, then for h > 0 the element will tend
to move downwards under the force of gravity, and for /t < 0 it will
tend to rise under the action o f the buoyancy, so that the
equilibrium will be stable. However, if the density increases with
height, then for any value of h the displaced element will tend to
move even further from its original position, and the state of
equilibrium will be absolutely unstable. Moreover, for an ideal fluid
(without friction), the equations o f motion will also have a steady

solution for any p = p(z) and arbitrary vertical profile o f the
A-component of velocity u = u{z) [and zero components of velocity
along the other axes]. Using the same argument, this flow will be
absolutely unstable for dp/dz > 0. For rfp/c/2 < 0, the question of
stability of the flow is considerably more complex; at this early stage
in the discussion we can only say, by similarity, the criterion o f
stability here must be expressed in terms of the so-called Richardson
number, i.e., the dimensionless parameter
= (2.3)
T h e ca se o f a flu id th a t is stra tified w ith r esp e c t to th e z a x is is o f g reat in ter e st fo r

m e te o r o lo g ic a l p r o b le m s, w h e r e such str a tific a tio n arises fro m th e te m p er a tu re p r o file
T = 7 (2 ) .
H o w e v er , in th is case, w e c a n n o t a ssu m e th a t th e flu id is in c o m p r e s sib le , b u t
m u st u se th e e q u a tio n o f state and th e e le m e n ta r y th e r m o d y n a m ic id e n titie s [s e e , e .g .,
L andau and L ifsh itz ( 1 9 6 3 ) , S e c t. 4 ] . T h e n w e fin d th a t a flu id e le m e n t d is p la c e d fr o m lev el
z t o lev el z + h , w ill b e lig h te r fo r /t > 0 th a n th e surro u n d in g air, b u t h ea v ier fo r fi < 0
w h e n a n d o n ly w h e n
( ( 2 .4 )
\ dT
w h e r e T is n o w th e a b s o lu te te m p er a tu re and V is th e sp e c ific v o lu m e . C o n d itio n ( 2 . 4 ) w ill

b e a c o n d it io n o f a b s o lu te in sta b ility o f th e sta te o f rest in th e p r e se n c e o f a te m p er a tu re
p r o file T = T( z ) . W hen th e m e d iu m m a y b e c o n s id e r e d as an id e a l gas, =
R V ^
— = -7^ , so th a t th e c rite rio n o f in sta b ility w ill ta k e th e fo rm
P T
dz Cp ■{ R
(th e c rite rio n w ill b e fo u n d in th is fo rm in all t e x t b o o k s o f d y n a m ic a l m e te o r o lo g y ). In
m e te o r o lo g y ,
T— 1 g is ca lled th e adiabatic temperature gradient (fo r air, th is
T ^
g ra d ien t is a p p r o x im a te ly 1 ° C /1 0 0 m ). T h e th er m a l s tr a tific a tio n o f th e air for w h ic h
— ( d T ld z ) ’\s greater, eq u a l o r less th a n Ga, is ca lle d , r e s p e c tiv e ly , sta b le, n e u tra l, or u n sta b le
s tr a tific a tio n .
A n o th e r r ep r e s e n ta tio n o f th e c o n d it io n s o f in s ta b ility ( 2 . 4 ) o r ( 2 . 5 ) is o f t e n u se d in
m e te o r o lo g y ; th is is c o n n e c te d w ith th e in t r o d u c tio n o f th e so-ca lled potential temperature,
d e fin e d b y
T-J
T (2.6)
\ P
w h ere po is so m e standard pressure (u su a lly ta k en as th e n o rm a l sea-level pressu re), in stea d

o f th e ordinary te m p er a tu re T. B y th e e n tr o p y e q u a tio n fo r an id e a l gas (S e c t. 1 .7 ),
Cp In 0 = 5 + c o n s t. T h e re fo r e , th e p o t e n tia l te m p er a tu re d o e s n o t vary in adia b a tic
pro cesses, so th a t d is eq u a l to th e te m p er a tu re w h ic h th e air w o u ld atta in i f b r o u g h t
adia b a tica lly to standard pressure Po. It is e a sy to see th a t ------( /^ .T h u s u sin g th e
c o n c e p t o f th e p o te n tia l te m p er a tu re , th e c rite rio n o f in s ta b ility ( 2 .4 ) m a y b e fo r m u la te d as

fo llo w s : th e sta te o f rest w ill b e u n sta b le i f d Q l d z < 0 (i.e ., i f th e p o t e n tia l te m p er a tu re
d ecrea ses w ith h e ig h t) and sta b le i f dQ/dz > 0 (i.e ., i f th e p o te n tia l te m p er a tu re in creases
w ith h e ig h t).
I f an arbitrary w in d v e lo c ity p r o file e x ists, th e m o t io n in th e ca se o f u n sta b le
s tr a tific a tio n w ill h k e w is e b e u n sta b le ; fo r sta ble str a tific a tio n , h o w e v e r , th e s ta b ility or
in sta b ility o f th e m o t io n m u st b e d e te r m in e d in s o m e w a y b y th e v alue o f th e R ic h a r d so n
n u m b er
g (dT (j\ £_
TI \s dz 0- dz ,
R i- , <2-3)
2.5 Mathematical Formulation of the Stability Problem for

Infinitesimal Disturbances
The conditions for transition to turbulence are by no means as
easy to find in every case as in the above examples. In general, the
most effective means o f investigating the stability is the general
method o f small disturbances. We shall now discuss the basic idea of
this method with reference to the flow o f an incompressible fluid of
constant density p.
The method of small disturbances is based on writing the velocity
field Ui(x,t) and the pressure p ( x , t ) which satisfy the dynamic
equations in the form u^ = Ui -t-aj, p = P + where Ui{x,t)SLnd
P {x, t) are particular solutions o f the equations under investigation,
and u[, p'aTQ small disturbances. Taking into account that Ui and P
themselves satisfy the equations of motion, and ignoring quadratic
terms in the disturbances, we obtain linear equations for u'l and p' in
the form
du\ du, , dU, 1 dp

— I" + “ a •;?;r = — T "st : + ^ ^
(2.7)
dXr.
Differentiating Eq. (2.7) containing da\jdt with respect to i,

summing with respect toATj.and using the last of Eq. (2.7), we may
express p' in terms of with the aid o f an equation analogous to Eq.
(1.9'). Then the general solution of Eq. (2.7) will be determined by
fixing only the initial values u ' (jc, 0) of the function a' {x, t) . We
may thus (at least in principle) attem pt to establish conditions under
which not all solutions of the corresponding initial value problem
will be damped in time; these conditions will also be conditions of
absolute instability® o f the solution Ui, P- Of course, if the
corresponding conditions are not satisfied, so that the solution P
will be absolutely stable (with respect to infinitesimal disturbances),
there still remains the possibiUty that with respect to finite
disturbances «•, p' (described by essentially nonlinear equations), this
solution may yet be unstable, i.e., that the flow described by this
solution will be a system with hard self-excitation. However, to
verify this possibility we shall require quite different methods of
investigation (see below. Sect. 2.9).
When the solution Ui = Ui(x), P ^ P ( x ) describes a steady
laminar flow of fluid, the coefficients of the system of equations
(2.7) will evidently be time-independent. In this case, the system will
have particular solutions of the form
the time-dependence of which is given by the exponential factor

with, generally speaking, a complex “ frequency” (o. The permissible
values of the characteristic frequency o) and the corresponding
amplitudes / „ (jc), (a:) will then be determined from the eigenvalue
problem for a linear system o f partial differential equations. When
the coefficients o f this system are independent of some space
coordinates, the number o f independent variables in the system may
be reduced by assuming that the dependence of the amplitudes
and on the corresponding coordinates will also be exponential,
with a given “ wave number” (i.e., the spatial scale of the dis
turbances is prescribed in the directions of the coordinate axes along
’ W e m u st p o in t o u t th a t th is te r m in o lo g y has n o t b e e n c o m p le te ly s e ttle d . O fte n th e

c o n d it io n o f a b s o lu te in s ta b ility is ta k en to m ea n th e c o n d itio n th a t E q. ( 2 . 7 ) w ill have an
in crea sin g s o lu tio n , i.e ., cases w h e r e o n ly d a m p e d and n eu tra lly stable s o lu tio n s are p resen t
are in c lu d e d in th e sta b le class. H ere, h o w e v e r , it w ill b e m o r e c o n v e n ie n t to c o n sid e r as
u n sta b le s o lu tio n s th o s e fo r w h ic h th ere e x is t s a t lea st o n e u n d a m p e d d istu rb a n ce.
which the undisturbed flow is homogeneous. Thus, for example, if

the undisturbed flow depends only on the coordinate a*, then we may
put
fjx ) = ^ ^ (2.9)
where the characteristic frequency <o=^m^k, and the amplitudes

/a;; s, g,„. ft, will be determined from the eigenvalue problem
for a system o f ordinary differential equations containing the
parameters k\ and k->. Similar equations, with exp j / + /t.x,)]
replaced by exp (/AjX,), and and i>„. , (x,), will be
obtained for flows which depend essentially on xz and only.
The system o f eigenfunctions (je) will frequently be complete in
the space o f all vector functions / = (/i, fz, fs) satisfying the
continuity equation dfjdxi = 0 and the necessary boundary condi
tions (if f j x ) = e'*"''/;,,.(X ,, x,) or f j x ) = e‘ , (.v,),
then it is sufficient that the functions ^ (a-,„ ^ 3) or ,, (A'^j for
any fixed k, or fixed k i , ^2 will form a complete system in the corre
sponding space of vector functions o f one or two variables). ‘ °
When the system of eigenfunctions is complete, any initial values
« (AT, 0 ) can be expanded in a series (or integral) in terms of these
eigenfunctions. Thus the general solution of the initial value problem
for Eqs. (2.7) may be expressed as a superposition of elementary
exponentially time-dependent solutions. Therefore the general sta
bility problem may be reduced here to the corresponding eigenvalue
problem. For the stability of such a laminar flow with respect to
infinitesimal disturbances it is necessary and sufficient that all
characteristic frequencies co have a negative imaginary part Imm <
0. When there is a space homogeneity with respect to one or more
coordinates, the various characteristic frequencies co will, generally
speaking, depend on the space “ scales” o f the disturbances (i.e., on
the wave numbers k\ or and ^ 2) and on Re. As Re -> 0, the
imaginary parts of all frequencies w will tend to negative values
However, the problem o f completeness in the theory o f hydrodynamic stability is not
a s im p le o n e . F r o m a m a th e m a tic a l v ie w p o in t , th e eig e n v a lu e p r o b le m s w h ic h arise h ere are
th o se o f a lin ear n o n se lf-a d jo in t o p e r a to r in a fu n c tio n sp a c e. T o esta b lish th e c o m p le te n e s s
o f th e s y ste m o f e ig e n fu n c t io n s (o r at le a st o f th e e ig e n fu n c t io n s and r ela ted a s so c ia te d
fu n c tio n s ) o f th is o p e r a to r , w e m a y o f t e n use th e th e o r e m o f K e ld y sh ( 1 9 5 1 ) [se e a lso
K e ld y sh a n d L id sk iy ( 1 9 6 3 ) ) . H o w e v er , in m u c h o f th e litera tu re, th e a s su m p tio n o f
c o m p le te n e s s is u se d w it h o u t a n y ju s tific a tio n , and in so m e a p p lic a tio n s th e a s su m p tio n is
clearly in c o r r ec t (se e th e e n d o f th is s e c tio n ).
(because for p = const, the state of rest is always stable). However, as

Re increases, the imaginary parts of certain frequencies may increase,
and, finally, become positive. Let us assume that the eigenvalue
problem corresponding to a fixed disturbance scale (i.e., fixed fei or
and k 2) has a discrete spectrum o f characteristic frequencies
(this assumption is fulfilled for many important flows). Here, the
critical value of Re corresponding to the transition to instability o f a
laminar flow with respect to infinitesimal disturbances o f a given
scale will be determined from the equation max Imco, = 0. The
smallest o f these critical values of Re for disturbances o f various
scales will be the critical number Re^r of the flow, i.e., for Re >
Recr, the laminar flow will be absolutely unstable and for Re < Recr
it will be stable. Of course this Rec- characterizing the instability
with respect to infinitesimal disturbances, must not be less than the
critical Reynolds number characterizing the stability o f the flow with
respect to finite disturbances. Thus, with the notation of Sect. 2.1, it
might be denoted by Rccr max- On the other hand, the instability o f a
flow for Re > Recr max means only that for such Reynolds numbers
the corresponding laminar flow cannot exist; however, it does
not signify that the flow is bound to become turbulent. In fact,
it is possible that after loss of stability, a given laminar flow may be
transformed into a new laminar flow which is now stable; also,
transition to a turbulent regime occurs only after loss of stability o f
this new laminar flow, at Reynolds numbers considerably greater
than Recr max (see the examples in subsections 2.6- 2.7).
Before ending our discussion, we must stress once again that the
possibility of expanding an arbitrary solution of the system (2.7) in a
series in terms of particular solutions of the form of Eq. (2.8) will
occur often, but not always—this fact is frequently forgotten in
presentations of the theory of hydrodynamic stability. In particular,
the situation is more complicated if the system (2.7) is singular (i.e.,
for example, if a coefficient of a leading derivative in this system
becomes zero at some point). In this case, the completeness of the
system of eigenfunctions cannot be proved simply, and even the very
concept of eigenfunctions must be defined with care. Often, even
with a fixed disturbance scale a continuous part of the spectrum of
eigenvalues arises with corresponding eigenfunctions satisfying un
usual boundary conditions or possessing a more complicated struc
ture (e.g., not vanishing at infinity or having discontinuities o f the
derivatives at a singular point). In applications, these “ improper”
eigenfunctions sometimes simply go unnoticed and, hence, the

system o f “ elementary solutions” of form ( 2 .8 ) is obviously
incomplete [see Case (1962); Lin (1961); Lin and Benney (1962);
Drazin and Howard (1966)]. If for some reason, the system of
eigenfunctions is incomplete, the investigation of the corresponding
eigenvalue problems is clearly insufficient for the solution o f the
stability problem. For a complete study o f such cases, we must
investigate the behavior of the general solution of the corresponding
initial value problem. This investigation is very complicated; how
ever, in the special case o f an ideal fluid with v = 0 , it nevertheless
has been possible to obtain a number of conclusive results (see the
works o f Dikiy (1960a,b) and Case (1960a,b) which we shall discuss
in detail later).
2.6 Stability of Flow between Two Rotating Cylinders
One important example o f absolute instability, which is amenable
to complete mathematical analysis, is the instability of steady
circular Couette flow between rotating cylinders. Let Ri and Qi be
the radius and angular velocity of the inner cylinder, and R 2 and
those o f the outer cylinder. In cylindrical coordinates r, (f, z with
the Oz axis along the axis of the cyUnders, the velocity field o f such
a Couette flow will be defined by the familiar equations
( 2 . 10)
(see Eq. (1.28) subsection 1.2).

First, let us ignore the effect o f viscosity. Then we may define the
criterion o f instability from the following elementary physical
considerations. In a steady laminar flow, the centrifugal force acting
on an element o f the fluid will be balanced by the radial pressure
gradient. Now let an element of mass m move under the action o f the
disturbance from a position with coordinate ro to a position with
coordinate r > ro- Then by the law of conservation of angular
momentum m rU J r ), in the new position its velocity will equal
'"o
roU^(ro)lr and therefore a centrifugal force m - ^ will a c to n
it. The equilibrium will be unstable if this force is greater than the
radial pressure gradient at a distance r from the axis, which is equal

in magnitude to the undisturbed value of the centrifugal force at the
distance r. Hence, the condition of instability [which, for such an
inviscid Couette flow was established by Rayleigh (1916a)], will
have the form
('■o)P — \rU^ (r )]2 > 0 for r > Tq,
or, in other words,
^ (ry /< o . (2 . 1 1 )
Following Coles (1965), and taking into account that U^\ris the
angular velocity of the flow, and that d{rU^)lrdr is the axial vor-
ticity, we may easily rephrase the criterion ( 2 . 1 1 ) as follows: a
flow is unstable i f the vorticity (local rotation) is opposite in sense to
the angular velocity (overall rotation). In such a form, this criterion
is apparently valid for many inviscid circulatory flows.
Using Eq. (2.10), Rayleigh’s criterion of instability may be
reduced to the form — If the cylinders are
rotating in opposite directions, then l)^ will change sign somewhere
between the cylinders, and in this case the flow will certainly be
unstable. When both rotate in the same direction, we may put Q, >
0, Q2 > 0, and then U^> 0 everywhere; in this case the Rayleigh
criterion of instability takes the form
A rigorous mathematical derivation of this result, obtained by

applying disturbance theory to the inviscid fluid was given by Synge
(1933) for the case of axisymmetric (i.e., independent o f cp)
disturbances o f the velocity [see also Shen (1964)], and by
Chandrasekhar (1960) for the case of arbitrary disturbances o f the
flow.
A more complete analysis (taking into account also the effect of
the viscosity) may be carried out only by the method of small
disturbances, first applied to this problem in a classical paper by
G. I. Taylor (1923). Since the undisturbed velocity field (2.10) here
depends only on the /-coordinate, then, by Eqs. (2.8) and (2.9), the
disturbances of the velocity and pressure may be sought in the form
u ,(x, p 'i x , 0
Here 2nlk is the wavelength of the disturbance in the Oz direction, n

is a nonnegative integer, determining the dependence of the
disturbance on the angle <p, and / ( r ) = / ^ . ^
(r) ] and g (r) — „ (r) are the r-dependent “ amplitudes” of the
disturbance with given axial wavenumber k , azimuthal wavenumber
n and characteristic frequency (o. Substituting from Eqs. (2.10) and
(2.13) into the general system of equations (2.7) and taking into
account the boundary conditions « (r, q?, 2 , /) = 0 for r = /?i and
r = /?2,we arrive at the eigenvalue problem determining the spectrum
o f permissible frequencies for given k and n. It may be shown [see,
for example, Di Prima (1961)], that this problem, after certain
transformations (including, in particular, the elimination of /f') and
g), may be reduced to the following system of two differential
equations in two unknown functions /Wand f(<p):
(2-14)
where A and B are determined from Eq. (2.10) and N is the

differential operator
l d-‘ \ d \ ./ nu^{r)\
N = - A 7 r > + 7 1 ? - 7 ^ - 'n - ‘ [ '" - ^ 7 - ] -
The boundary conditions for which the sixth-order system (2.14)

must be solved, take the form
= ioxr==R„ r = R ,.{ 2 .\ 6 )
The eigenvalue problem (2.14)—(2.16) has no singularities. There

fore it is apparently not difficult to show that this problem will
always have a discrete set of eigenvalues (Oj = (Oj(^, n, Qi, Q2, Ri, Ri),
while the corresponding eigenfunctions/L 7 * . f o r m a

complete system in the space o f function pairs satisfying
the boundary conditions (2.16). Consequently, it follows that the
initial velocities « ' (at, 0), defined with the aid o f Eq. (2.13) and
corresponding to the eigenfunctions with all possible k and integral
nonnegative /i, will form a complete system in the space o f vector
functions of = (r, q>, z) satisfying the solenoidal condition and be
coming zero for r = R\ and r = R 2. Thus, investigation o f the eigen
value problem (2 .14 )-(2 .1 6 ) completely exhausts the stability
investigation of Couette flow between rotating cylinders. However,
this eigenvalue problem is very complicated, and only recently have
there appeared several works devoted to it. We shall discuss these
works at the end of this section; first, however, let us consider the
simplified approach proposed by G. I. Taylor (1923) and used by al
most all subsequent investigators.
Instead o f solving the complete eigenvalue problem for the general
system o f equations (2.14), almost all investigators studying the
stability problem for circular Couette flow have assumed that n = 0 ,
i.e., they limited their attention to axisymmetric disturbances o f the
velocity o f the basic flow. Under this assumption, the system is
considerably simplified, and may be rewritten as
where
____ L — (2 17')
r dr r^— d r \ d r ^ r ) - )
The boundary-value problem (2.17)—(2.16) for fixed k has a

denumerable number o f eigenvalues oyj = o)j(li, Qi, Q2, Ri, R 2) and
eigenfunctions/^(r) = /;»>. (r)- However, the corresponding func
tions j (r) now will not form a complete system in the space
o f permissible initial fields »'(jc, 0 ) = u' {r, z) [if only because all
these functions are independent of rp]. Nevertheless, the region of
the(Qi, Q2)-plane to which there correspond unstable disturbances
o f the form exp[i{kz — (o /)]/(r) [i.e., disturbances of this form with
lm u )> 0 ] is usually identified simply with the whole region of
instability [see, e.g., Lin (1955); Chandrasekhar (1961); Stuart
(1963)]. In other words, it generally is assumed, even if this is not
specifically stated, that for all values Ri, Rz, Qi and J.22, for which
there is at least one eigenvalue coj(^, Qi, ^ 2, Ri, ^ 2)with n = ^ 0 and
lm u)y> 0 , there will also be an eigenvalue Qi, Q2, Ri, R 2)
[corresponding to n = 0] with lmo)y> 0 (i.e., as the Reynolds
number increases, a nonnegative imaginary part will first appear in
the case n = 0). Until recently there was no doubt that this
assumption was correct and in fact most of the existing experimental
data (although not all) give excellent confirmation of the identity of
the region o f stability determined by considering only axisymmetric
disturbances with the entire region of stability of a flow between
rotating cyUnders. However, in the last few years, some new results
(both experimental and theoretical) have been obtained. These
results show that the study of the eigenvalue problem (2.17)—(2.16)
[which corresponds to n = 0 ] is not sufficient for a complete
solution o f the Couette flow stability problem. We shall discuss these
new results later; first, we will consider the older results, based on
the assumption that it is permissable to take n = 0 .
Let us now assume that the radii /?, and R 2 are fixed. It is easy to
see that for sufficiently small Qi and Q2, all the eigenvalues
(Oj(k, Qi, Q2) and n, Qi, Qz)will have negative imaginary part
(because the state o f rest is always stable). If we now increase the
angular velocities Qi and Qz without changing the ratio Q2/S2 i(that is,
we increase the Reynolds number without loss of geometric
similarity), then for certain QjQi there will be no zero or negative
imaginary part for any Qi, Q2) at any Re (i.e., the motion will
always remain stable). For other values of Q2/Q 1 at some Recr =
^ 2 R lh , there will first appear a value of k^r such that some
®i(^cr. Q2)will have a zero imaginary part (i.e., the motion will
become unstable).
It is interesting that in this problem for all the transition
from stability to instability will occur by the appearance o f an
eigenvalue COj such that not only does l m a ) j = 0 , but in factcoj = 0 (a
rigorous mathematical proof o f this fact, in general, is difficult, and
apparently, has been given only for the special case when Rz — R\<.
{Ri + Rz)l2,and under some additional restirictions, by Pellew and
Southwell (1940) and Meksyn (1946). However, both the experimen
tal data and the results of direct numerical procedures show that it
has a more general character). This means that the loss of stability in
circular Couette flow leads to the development of a new (secondary)
steady motion with velocity field « = where » ' = exp
(ikz) f ( r ) . This rnotion will have the form of so-called Taylor
cellular toroidal vortices spaced regularly along the axis of

the cyUnders (the schematic
form of the streamline pattern
of these vortices, computed
from the corresponding eigen
functions fir), is shown in the
figure on p. 1 0 1 ).
-sc
The fact that the transition
from stability to instability
proceeds through a steady D
state, corresponding to the o
purely zero eigenvalue co = 0 ,
is sometimes called the “prin
ciple of exchange of sta
bilities” in the theory of
hydrodynamic stability. This
principle was often assumed
and given great significance in FIG. 12a. S tre a m lin e p a tte rn o f v o r te x dis-
tu rb a n c e fo r flo w b e tw e e n ro ta tin g c y lin d er s.
the past, but it is now known ( A fte r Shen, 1964).
that it holds only in certain
special cases, and breaks down
for many simple flows (e.g., for plane Poiseuille flow).
With further increase of the Reynolds number in Couette flow,
there will be a whole range of wavenumbers k for which there are
eigenvalues coj with lmcoy> 0. However, experiment shows that in
this case the disturbed motion preserves the form of a regular sequence
of toroidal vortices for a fairly long time, but with an axial wavenumber
which cannot be deduced from the linear theory (see below. Sect. 2.9).
With further significant increase of Re, this periodic motion becomes
unstable and is transformed into another regular motion (this time
nonaxisymmetric); with still further increase of the velocity other
(more comphcated) forms of flow appear and disappear in an orderly
sequence, until finally the flow takes the form of disordered turbulent
motion [see Coles (1965) and also subsection 2.9].
The boundary-value problem (2.17)—(2.16) is considerably sim
pler than (2.14)—(2.16).‘ * Nevertheless, solving Eqs. (2.17)—(2.16)
^ ^We m a y n o t e , in particular, th a t sin ce th e p rin cip le o f e x c h a n g e o f sta b ilitie s a p p lies

for E qs. ( 2 . 1 7 ) - ( 2 . 1 6 ) , w h e n d e te r m in in g th e b o u n d a r ie s o f th e reg io n o f sta b ility w it h o u t
p r e te n tio n to rigor, w e m a y a ssu m e in th is ca se th a t co = 0 . F or p r o b le m ( 2 . 1 4 ) - ( 2 . 1 6 ) w ith
n ^ 0, h o w e v e r , th is p rin cip le d o e s n o t a p p ly , a n d th u s th e a d d itio n a l p a ram eter R e co
o c cu rs in th e c a lc u la tio n s.
Still causes considerable difficulty. Consequently, the numerical

calculation of eigenvalues was first performed by Taylor (1923) only
for the case of a very small gap between the cylinders (i.e., for
/?,< C (î + ^ 2) / 2 ). In this case both the equations and the
numerical work may be simplified considerably. Taylor’s method-is
based on the expansion of the solution of Eqs. (2.17) in terms of a
special set of orthonormal functions; the differential equations then
reduce to an infinite system of linear algebraic equations. Later, this
method was improved somewhat by Synge (1938); other numerical
methods (in particular, a method based on the reduction of the
boundary-value problem (2-17)-(2.16) to a variational problem,
Galerkin’s method and some direct numerical methods) also applica
ble only in the small-gap approximation were worked out by Pellew
and Southwell (1940), Meksyn (1946), Chandrasekhar (1954; 1961),
Di Prima (1955), Steinman (1956), Duty and Reid (1964), Walowit,
Tsao and Di Prima (1964), Reed and Harris (1964) and others. Only
rather recently has a more general approach been proposed [Witting,
(1958)] which uses an expansion o f the solution in powers o f the
parameter d//?o (where d = R 2 — Ri,Ro= (Ri + Rz)l2). In the works
of Chandrasekhar (1958), Chandrasekhar and Elbert (1962),
Kirchgiissner (1961), Sparrow, Munro and Jonsson (1964), Yu
Yun-Sheng and Sun Dah-Dhen (1964), Roberts (1965), and Capriz,
Ghelardoni and Lombardi (1966), still more general methods have
been developed for the numerical calculation of the eigenvalues and
eigenfunctions of (2.17)—(2.16) for arbitrary Ri and i?2 ; see also
surveys by Chandrasekhar (1961) and Di Prima (1963). We should
also point out that the existence of the instability of the flow o f a
viscous fluid between rotating cylinders for certain values of Q iand
fi 2was rigorously proved by Krylov (1963) and by Yudovich (1966a)
without using any numerical methods, but this proof provides no
means of determining the boundary of the region of instability.
However, the numerical methods allow us to find this boundary with
high accuracy and to show that it has roughly the same form in all cases.
For an example. Fig. 13a shows the region of the(S2i, Q2) plane in
which unstable disturbances independent of (p are possible for
Ri!R\ = 113 (according to Taylor’s calculations and in excellent
agreement with all later calculations). Figure 13b gives the corre
sponding region for R 2lR\ = 2 (according to Chandrasekhar’s
calculations). In both cases (and also for all other R 2IR 1) the
numerical procedures lead to the finding of unstable disturbances
only if Q>/Q\ < {R\IR2) ^‘ . Therefore it is natural to think that in the
region > {R\IR‘i)'^ instability will not occur for any Re. Thus
here the viscosity has only a stabilizing effect [though in fact for
viscous flow with Q./Qi rigorous proof of stability has
been given only for axisymmetric disturbances and for some very
special nonaxisymmetric disturbances: see, e.g., Chandrasekhar
(1961), Sect. 70, and Schultz-Grunow (1967)]. When Qa/fii = const
< (/?I//?2) ^ however, according to the data of Fig. 13, instability of
infinitesimal disturbances must arise for sufficiently great Re.
The flows corresponding to the unshaded region o f Fig. 13 are
certainly stable with respect to infinitesimal disturbances, but as
experiment shows, for sufficiently large Re they are unstable with
respect to finite disturbances, in this case loss of stability leading to
the spontaneous formation of regions of developed turbulence (this
is an example of “ catastrophic transition” in the terminology of Coles
(1965), who performed a brilliant experimental investigation of
different types of transitions in circular Couette flow). The boundary
o f the region of stabihty to infinitesimal disturbances, calculated by
Taylor, is in excellent agreement with the experimental data found
by Taylor himself and by other investigators, using various visualiza
tion techniques. This was the first great (even epoch-making) success
o f the hydrodynamic disturbance theory (see Fig. 13a, where the
dots denote the measured values o f a number of pairs (Qi, Qj) for
which loss of stability first occurred). Calculations for the case
RÎRx = 2 are also in excellent agreement with the experimental data
FIG . 1 3 . P o s itio n o f th e reg io n o f in s ta b ility in th e p lan e (Q ,, Qj'j fo r C o u e tte

f lo w b e tw e e n c y lin d e r s , fo r v a lu es o f th e r a t i o T h e reg io n o f in sta b ility is
sh a d ed in th e figu re. T h e d o t t e d lin es in d ic a te th e b o u n d a r y o f th e r eg ion o f in sta
b ility fo r th e c o r r e sp o n d in g f lo w o f an in v iscid flu id (a fte r R a y le ig h ).
o f Donnelly and Fultz (1960) [dots in Fig. 13b, see also Donnelly
(1962)]. Snyder (1968b) made a detailed comparison of all the
existing stability calculations (for axisymmetric disturbances of
circular Couette flow) with the experimental results of various
investigators (including his own). With a few exceptions, excellent
agreement was found. The exceptional cases can usually be explained
by end effects; another explanation related to the results considered
below is discussed by the author.
The experimental findings of G. I. Taylor (1923) and Donnelly and
Fultz (1960) [shown in Fig. 13], appear to confirm the agreement
between the regions o f instability associated with axisymmetric
disturbances and the regions of general instability; in other words,
they are in excellent agreement with the assumption that for
monotonically increasing Reynolds number, circular Couette flow
always becomes unstable first to disturbances with « = 0. At the
same time, however, as early as the work o f Lewis (1928) it had been
shown that for n = negative and rather large (i.e., for
comparatively high rotational speeds o f the cyhnders in opposite
directions), upon loss o f stabihty motions appear which are
sometimes said to be of “pulsating form,” i.e., which do not appear
to be axisymmetric. For many years Lewis’ observation did not
attract the attention of anyone. However, recent experiments by
Nissan, Nardacci and Ho (1963) and Snyder [see Snyder and
Karlsson (1965); Snyder (1968a); Krueger, Gross and Di Prima
(1966)] completely confirmed its correctness. Furthermore, from a
purely mathematical viewpoint, the investigation of the stability of
circular Couette flow cannot be considered complete until the
behavior of nonaxisymmetric disturbances has been analyzed. Fi
nally, extensive experiments (which will be discussed further in Sect.
2.9) show that the loss o f stability of axisymmetric Taylor vortices in
the gap between the cylinders, arising upon further increase of the
Reynolds number, very often leads to the formation o f a new
nonaxisymmetric flow. Consequently, the examination of the behav
ior of nonaxisymmetric disturbances of the flow between cylinders
must be considered an im portant and interesting problem.
Apparently, Di Prima (1961) was the first whose work was
devoted to the mathematical analysis of the behavior of nonaxisym
metric disturbances o f circular Couette flow. He examined the
general eigenvalue problem (2.14)—(2.16), and under a series of
simplifying assumptions (including, in particular, the assumption of
smallness of the gap d = R2 - R ib y comparison with Rq =(Ri + R 2 )/2 ,
and that of m= > 0 ) a method was displayed whereby suc

cessive (rather crude) numerical estimates are obtained for the
smallest value o f Re = ^ id ^ lv at which, for n = 0 , 1, 2 and 3, one
o f the eigenvalues o f the system (2.14)—(2.16) appeared for the
first time with a nonnegative imaginary part (specific calculations
were carried out for the cases d/Ro = 1 / 8 , /i= 0 and d/Ro = 1 / 8 ,ju =
0.5). The results obtained show (as anticipated) that Recr corresponds
to axisymmetric disturbances. However, a somewhat surprising result
was that the critical Reynolds numbers for nonaxisymmetric
disturbances with small values of n turned out to be only slightly
larger (by a few percent in all cases) than Recr for axisymmetric
disturbances with n = 0 .
Later, Roberts (1965) calculated more precise values of Recr for
several fixed values o f 17 =/?i //?2 (to be exact, for 17 = 0.95, 0.90, 0.85
and 0.75), several values of n (in all cases not exceeding 4) and the
value n = 0 (i.e., ^ 2= 0). He also found that Recr for /li = 0 always
increases (although not strongly) with increasing n. The first work in
which an asymptotic analysis o f the eigenvalue problem
(2 .14 )-(2.16 ) was carried out also for ju < 0 (again in the limit of
small gap), was the dissertation o f Krueger (1962). According to his
estimates for the case of small gap between the cylinders and n ^
—0 .8 , nonaxisymmetric disturbances with n = 1 become unstable for
lower angular velocities o f the cylinders than those at which
axisymmetric disturbances first become unstable. Krueger, Gross and
Di Prima (1966) made a more complete investigation of this
question. They used a straightforward numerical procedure, and
calculated (with the help of a digital computer) critical values o f the
so-called Taylor number Ta (proportional to Q.]R\d^ jv^ , so that Ta <»
(Re)^ = (Q,/?i/i^)^ for fixed 5 = d I R , ) for a series o f values of n and ju
= Q j/i 2 i in the interval 0 > ju > —1.25 (in the hmit of small gap
between the cyUnders). Thus it turned out that if ju > -0 .7 8 , then
Tacr increases with increasing n (i.e., axisymmetric disturbances
become unstable first), but for ix < - 0 .1 8 the minimum value o f Tacr
is attained for n ^ 0 , and with decreasing ju the critical value of n
(giving the absolute minimum o f Tacr or Recr) increases (of course,
not continuously; n cannot take on nonintegral values). For example,
for 5 = dIR I = 1/20 and m = -0 .7 0 , -0 .8 0 , -0 .9 0 , -1 .0 0 , and
—1.25, with increasing rotational speed instability first appears for
disturbances with « = 0, 1, 3, 4, and 5, respectively. On the other
hand, for fixed {jl, the values «cr do not increase with increasing
dimensionless gap width 6 = d / R i , i.e., with decreasing 17 =
(calculations with the approximations o f small gap not made, show

that for M = - 1 and 17 = 0.95, 0.90, 0.80, 0.70, and 0.60, the greatest
instability is attained for disturbances with n = 4, 3, 2, 2, and 2,
respectively). However, it is important to note that the difference
between the minimum value Tacr or Recr and the value Tacr or Rccr
determined from the outcome o f an examination of axisymmetric
disturbances only, is usually found to be rather small; for example,
with 5 = 1/20 and /n = - 1 , the critical value Re^^ (corresponding to « =
4) is only roughly 4% less than that Re for which axisymmetric
disturbances first become unstable. This last circumstance induced
Krueger, Gross and Di Prima to carry out additional cumbersome
calculations using the complete eigenvalue problem (2.14)—(2.16)
[without the simplification resulting from the assumption of small
gap 6 ], to verify that the conclusions reached previously would not
change. It was found that for values o f m o f order —1, even for 6 =
1 / 2 0 , using the approximation of small gap noticeably changes the
numerical values Tacr; however, the change is almost identical for all
n and k, so that all qualitative conclusions bearing on the comparison
o f the stability o f disturbances for various values of n, are exactly the
same in the small-gap approximation and in the exact formulation o f
the problem.
The results of Krueger, Gross and Di Prima are in good agreement
with the experimental results o f Nissan, Nardacci and Ho (1963) and
Snyder and Karlsson (1965), according to which, in particular, for ju
less than roughly —0.75, the secondary motion which appears on the
loss o f stability of circular Couette flow always turns out to be
nonstationary and nonaxisymmetric (having a wavy or spiral struc
ture). The conclusions of Coles (1965; and other works) also agree
with these results. Coles carried out a detailed investigation of
transition in Couette flow between two relatively short cylinders
(appreciably shorter than those of most of the other investigators).
Finally, the work of Snyder (1968a) was particularly concerned with
verification of the conclusions of Krueger, Gross and Di Prima. Snyder
investigated experimentally the stability of circular Couette flow (for
four different values o f tj 0.96, 0.80, 0.50, and 0.20) over a
wide range of ju = including large negative values. Here, for t? =
0.96 (corresponding obviously to small gap) excellent agreement
with the above-mentioned theory is found throughout; in particular,
the values of ju corresponding to the first appearance of the most
unstable disturbances with « = 1, 2, 3, and 4 were almost exactly as
if they had been based on the calculations of Krueger, Gross and Di
Prima. Snyder obtained analogous results also for rj = 0.80, 0.50, and
0 .2 0 (where the assumption o f small gap cannot be considered well
satisfied). According to all these results, the form of the stability
boundary shown in Fig. 13a must be changed slightly in the upper
left corner of the figure to take into account the nonaxisymmetric
disturbances (in Fig. 13b the range of values t?, in which it is nec
essary to take into account disturbances w ith/t 0 ,is not shown).
In closing, we should note that the problem of the stability of flow
between cylinders is one which has been subjected to the most detailed
study in the theory of hydrodynamic stability. In addition to the
results mentioned above, a great many problems also exist which
have been subjected to general analysis, arising from the addition to
the original problem of axial or circumferential pressure gradients,
and/or the addition of axial, circumferential or radial magnetic fields,
the replacement of the usual fluid by a non-Newtonian fluid, etc.
[see, for example, Chandrasekhar (1961), Di Prima (1963), Krueger
and Di Prima (1964), and a series of other w orks].
2.7 Stability of a Layer of Fluid Heated from Below

When a vertical, downwards temperature gradient exists in a fluid,
there is additional destabilization of the flow due to buoyancy,
which is similar to the effect o f the centrifugal forces in curved flow
when the rotational velocity o f the fluid decreases with distance
from the center of curvature. On the other hand, an upwards
temperature gradient has a stabilizing effect on the flow, e.g., the
case of curved flow with velocity increasing with distance from the
center o f curvature. Thus it is not surprising that the stability
problem o f a thin layer of fluid between two infinite planes at
different temperatures is very similar mathematically to the stability
problem of an incompressible fluid between two rotating
cyUnders. ‘ ^
To solve this new problem we must take as our starting point the
Boussinesq equations of free convection, discussed at the end of
Sect. 1.5. For a layer of fluid bounded by the rigid planes ats = 0 and
X3 = H, which are maintained at constant temperatures To and T uthe
12 In fa c t J e ffr e y s ( 1 9 2 8 ) , s h o w e d th a t th e first o f th ese p r o b le m s is str ictly e q u iv a le n t to

a particular case o f th e s e c o n d ; c f. Lin ( 1 9 5 5 ) S e c t. 7 .3 . T h e a n a lo g y b e t w e e n th e
c en tr ifu g a l and b u o y a n c y e f f e c t s a lso p la y s a c en tra l ro le in G o r tle r ’s ( 1 9 5 9 ) sta b ility
in v e stig a tio n ; the sa m e id ea w as used b y D eb ler ( 1 9 6 6 ) to o b ta in n u m erica l r esu lts o n th e
s ta b ility o f a h o r iz o n ta l layer o f flu id w ith h e a t so u r ce s d ir e c tly fr o m d a ta o n s ta b ility o f
circular C o u e tte flo w .
boundary conditions will have the form: Ui = 0 , T = To for =0;

Ui = 0 , T = Ti for xb = H. In the case of a free constant-temperature
surface of the layer, the boundary condition o f the velocity is
changed so as to ensure constant pressure on this surface. There are
also other types o f mathematical boundary conditions, correspond
ing to different physical conditions on the boundaries of the layer
[see, e.g.. Sparrow, Goldstein and Jonsson (1964), or Hurle, Jakeman
and Pike (1967)]. The steady state (the stability of which we are
discussing), independent of the boundary conditions, will be the
state o f rest, for which* ^
Mj — 0, T — T (a:3) — ^0 H 77 ~ -'•s*
(2.18)
P ^ P ( X 3 ) = P o — gpX3.
Putting T = T{x 3) + T', P = Pixs) + p' and linearizing the Bous-

sinesq equations with respect to the disturbances uu T' and P', we
obtain a system o f five equations with five unknowns, from which it
is easy to eliminate all the variables except T'. Now, if we transform
to dimensionless variables, and then, following Eqs. (2.8) and (2.9)
we seek T'(x, t) in the form o f a product
r(A :, 0 = e x p + 0(C), C= (2.19)
we obtain the following eigenvalue problem :
( I p - ‘ "I ( ^ - + » )(| : - + /» ■ p .)» + F Ra . 0 = 0 ,^^
Pr = v/z, R a = g^
with boundary conditions on rigid surfaces with fixed temperature
®= I ? = ^ P--) = 0 (2 .2 0 ')
0= = = O ( 2 .2 0 ")
1 3 0 f c o u r se , in th is an alysis, w e hav e assu m ed fo llo w in g th e usual a s su m p tio n s o f th e

B o u ssin esq a p p r o x im a tio n , th a t th e variatio n o f d e n s ity P w ith h e ig h t is n eglig ib ly sm all, as
far as th e d istu rb a n c e e q u a tio n s are c o n c e r n e d . T h e r e fo r e , o u r r esu lts are ap p lic a b le o n ly to
th e case o f r elativ ely th in layers, w ith sm all values o f H o w e v e r , in th e case o f very th in
layers o f flu id th ese results o n c e again b e c o m e in a p p lic a b le d u e to th e b r e a k d o w n o f th e
c o n d it io n s o f lin earity o f th e te m p er a tu re p r o file and certain o th e r c o n d it io n s a ssu m ed in
th e th e o r y un der d is c u s sio n [see S u tto n ( 1 9 5 0 ) , Segel and Stuart ( 1 9 6 2 ) ] .
[see, for example, Chandrasekhar (1961), Stuart (1963), and Lin

(1955)]. Similar boundary conditions are also obtained for rigid
boundaries o f finite conductivity or for boundaries with variable
temperature, but fixed heat flux or fixed Unear relation between the
heat flux and the temperature. In every case, we end up with an
eigenvalue problem which, for given Prandtl number Pr, contains
only two parameters: the wave number k and the Rayleigh number
Ra. Consequently, to any given values o f k and Ra will correspond an
associated set o f eigenvalues (Oi(*. Ra). It is found that for negative
or small positive values of Ra (i.e., when the lower boundary is at
the lower temperature, or when the lower boundary is only slightly
warmer than the upper), all the eigenvalues Ra)for all values of
k will have a negative imaginary part. However, after some “critical
value” Racr (i.e., after some critical temperature difference /"o — 7’t,
which also depends on the distance H between the planes), there will
appear a value k = k^r for which one o f the eigenvalues coj(^cr. Racr)
has a zero imaginary part. In this case, just as in the problem o f the
stability o f flow between rotating cyHnders, the principle of
exchange o f stabilities will apply, i.e., the first e ig e n v a lu e o), (^, Ra)
with Imw, = 0 proves to be zero [a rigorous mathematical proof of
this fact was given by Pellew and Southwell (1940)]. Thus the loss o f
stability of the state of rest on attaining the critical temperature
difference To— 7'i leads to the initiation of a steady convection
periodic with respect to Xx and Xi.
From a qualitative viewpoint, all the basic features of the
transition from stability to instability in a layer o f fluid heated from
below were described very early by Rayleigh (1916b), who analyzed
the mathematically much simpler (but physically unreal) problem of
convection in a layer o f fluid between two free boundaries at
constant temperature. Mathematically, this problem reduces to an
eigenvalue problem for the differential equation ( 2 .2 0 ) with the
boundary conditions (2.20") for C = 0 and ^ = 1. To find the values
of Racr and k^r it is sufficient to consider Eq. (2.20) withco = 0 , i.e.,
+ = ( 2 .21 )
Solutions of this equation, which satisfy Eq. (2.20") for ^ = 0 and

^ = 1, are all of the form 0 = sin nn^, n = I, 2, . . . . Consequently,
for sufficiently large values of Ra, we shall have a sequence of
different neutral disturbances (with co = 0 ). the wave numbers of
which will satisfy
+ ( 2 .2 2 )
where n is a positive integer. The minimum Rayleigh number for

every given k will correspond to a disturbance with n = 1 ; Racr will
(jc’ -f- k'^Y
accordingly be defined from the condition R a c r= m in ---- pj---- .
Thus we obtain
Racr = 27it4/4 « 657.5, h , = it / 2 / 2 « 2.2.
For the real problem of convection in a layer between rigid

fixed-temperature boundaries, the analogous calculation demands the
use o f numerical methods because of the more complicated
boundary conditions. Such calculations were carried out by Jeffreys
(1962), Low (1929), and Pellew and Southwell (1940) [see also Lin
(1955), Reid and Harris (1958), and Chandrasekhar (1961)]; it is
found that in this case Racr ~ 1708 and 3.12. The values of k
and Ra for the first 10 eigenfunctions of this eigenvalue problem can
be found in Catton (1966). Further, we note that the value of k
determines only the periodicity o f the flow arising in the (xi, JCz) plane
but not its form; in fact, it is not difficult to see that the function
exp [i{kiXilH + k 2X2lH)] in Eq. (2.19), without changing the conclu
sions which follow, may be replaced by an arbitrary function X2)
satisfying the following equation:
The explicit form of the function cp will determine the form o f the
cells into which the convection breaks down, but it cannot be
determined uniquely on the basis o f linear disturbance theory.
However, the data o f numerous experiments (described, for example,
in Chandrasekhar (1961) [see also Stuart (1963) and the references
cited there) show that the flow which arises most often breaks down
into a set of cells (known as Benard cells) in the form of hexagonal
prisms, where in the middle the fluid moves upward and on the edges
it moves downward, or vice versa. To such cells there corresponds
is w o r th n o t in g th a t h e x a g o n a l c ells are n o t th e o n ly fo r m o f cellular c o n v e c t io n
o b s e rv e d in th e e x p e r im e n ts . M ore d e ta ile d d is c u s sio n o f th is q u e s tio n w ill b e g iven in S e c t.
2 .9 in c o n n e c t io n w ith th e n o n lin e a r th e o r y o f th er m a l c o n v e c t io n .
a completely defined form of the function , which was found by

Christopherson (1940). Comparison of the velocity field correspond
ing to such an eigenfunction o f our problem with the data of
observations of Benard cells may be found in Chandrasekhar (1961)
and in the article by Stuart (1964) [who also used the results of
nonlinear convection theory, which we shall discuss in Sect. 2.9]; it
gives good agreement with experiment.
The situation is analogous when the upper boundary o f the fluid is
a free boundary with fixed temperature; in this case we need only
replace the boundary condition (2.20') on S = 1 by (2.20"). As a
result, we arrive at a new eigenvalue problem, for which numerical
calculations show that the loss o f stabiUty occurs at Racr ~ 1 1 0 0
[Pellew and Southwell (1940); Chandrasekhar (1961)]. This value of
Racr, similar to the value o f Racr for convection between rigid
boundaries is confirmed excellently by experiment [ c f , for example,
Chandrasekhar (1961), Sutton (1950), Thompson and Sogin
(1966)]. Values o f Racr corresponding to some other boundary
conditions for temperature, may be found, for example, in the
articles of Sparrow, Goldstein and Jonsson (1964), Hurle, Jakeman
and Pike (1967) and Nield (1967). There are also many papers on
other effects in the convection problem (first of all, on surface
tension effects which apparently play an im portant role in Benard’s
original experiments); the reference to these papers may be found in
Stuart (1965) and Nield (1967; 1968).
With further increase o f To — Ti (i.e., with increase of Ra) above
the value of Racr the steady “cellular” convection first preserves its
character, but then becomes unstable, and, for Ra ~ 500,000,
disordered turbulent motion arises. It is important to note that the
transition to turbulent flow takes place gradually in this case—in the
form of a sequence o f discrete “jum ps,” characterizing transitions to
a sequence o f different regimes of convection each o f which is more
complicated and less ordered than the previous one [see Malkus
(1954a), Willis and Deardorff (1967)].
2.8 Stability of Parallel Flows
Above, we considered two examples of the application of the
method o f small disturbances to the investigation of hydrodynamic
stability. However, from the viewpoint of the experimenter or
engineer, both these examples are rather specialized. Far more
suitable for experimental verification and for practical application
are the cases of flows in a circular tube and along a flat plate (which
m STATISTICAL FLUID MECHANICS
is why they were given particular attention at the beginning of this

section). Nevertheless, to illustrate the small disturbance method,
first, we considered the flow between rotating cyUnders and free
convection in a layer between two planes of constant temperature.
This was done because in these two examples (due to the presence of
additional effects connected with a centrifugal force in the first case
and with buoyancy in the second), the method o f small disturbances
leads to comparatively simple eigenvalue problems without any
singularities even in the limiting case = 0 ); therefore quite definite
results can be obtained in these examples. For flows in tubes and
boundary layers, however, use o f the small disturbance method
encounters quite considerable mathematical difficulties, which even
now cannot be considered to have been completely solved.
Because of the complexity of the mathematical analysis of
stability o f flows in tubes and boundary layers, practically all of the
relevant investigations have considered only the simplest plane-
parallel, two-dimensional flows, which to some degree may be
regarded as models o f the real flows in question. Therefore, we shall
also begin our discussion with the stabiUty problem for plane-parallel
flows and only then shall we point out what is known concerning the
stability of real flows in tubes and boundary layers. Further, we note
that the equations of fluid dynamics ( 1 .6 ) show that a steady
plane-parallel flow of viscous fluid with only one nonzero com
ponent of velocity U = «i, dependent only on the coordinate Xs =
is possible only on the condition that the dependence of the profile
o f U(z) on 2 is quadratic. In other words, such a flow is always a
combination of a Couette flow with linear velocity profile and a
Poiseuille flow with a parabolic profile (see Examples 1 and 2, at the
end o f Sect. 1.2). However, keeping in mind that we are interested in
plane-parallel flows as possible models of more complicated real
flows, we shall also consider more complicated profiles of ll{z)\r\ the
hope that the results obtained may be appHcable to flows that are
not strictly plane-parallel.
We shall assume the axis Oxi = Ox to be directed along the flow,
and that the velocity of the undisturbed flow is independent of
X2 = y and is given by an arbitrary function Uixs) = U(z) . It is
important to note that in finding the criterion o f instability o f such a
plane-parallel flow, we may limit our consideration to only two-
dimensional disturbances o f the form u ' = { u' {x, z, t), w'{x, z,
since the more general three-dimensional disturbances lose their
stability later (for higher Reynolds numbers) than two-dimensional
disturbances. This assertion, proved by Squire (1933) [see also Lin

(1955), Sect. 3.1, Betchov and Criminale (1967), Sect. 18] may be
explained by writing the disturbance as the superposition of
elementary plane-wave components. Two-dimensional disturbances
will correspond to waves o f the form exp ii k x - o j t) propagated along
the undisturbed flow, and general three-dimensional disturbances to
waves of the form exp i(kiX + k 2 )’ — ojr) propagated in a direction
which does not coincide with the direction of the undisturbed flow.
If the coordinate frame is rotated in the (.v, y) plane so that the
new x-axis is in the direction of the wave, then the basic flow has
two components U(z), V(z), and the wave propagates in the
x-direction, and is independent of y. Further, the equations for u\
a?', p' are independent of V and v' in such a case. These
considerations show that a wave component of a disturbance with a
given wave-number vector is in fact affected only by the component
of velocity of the undisturbed flow in the direction of this
wave-number vector, which equals Ukijk, where h is the x-
component of k , and k is its modulus. Mathematically, this is seen by
the fact that for plane-parallel flow with velocity U = V{z),
substitution of Eqs. (2.8) and (2.9) withx:i = x,xz = y, jc, = 2 in the
general system (2.7) after certain transformations [and elimination
of all unknown functions except the function [( 2 ) = which
gives the component u'^{x, t)==w' (x, t), of the disturbance u' {x, /))
leads to the equation
(k,U - /V - k ^ J /, = k\,
(2.24)
This equation together with the corresponding boundary conditions

determines the eigenfrequencies of the problem, w = Wj, / = 1 , 2 ,. . ..
Now, if we consider a two-dimensional disturbance with the same
modulus o f the wave-number vector k (i.e., a disturbance of the form
(2.8)—(2.9) with /ji=feand /i2 = 0), then the new equation (2.24) will
differ from the old one only in the replacement o ih U h y k U , i.e., it will
be identical to the above equation, but with the velocity U (z) increased
by a factor ofkjki. Thus, for wave disturbances propagated at some
angle to the undisturbed flow, the effective Reynolds number is less
than for disturbances with the same wavelength propagated in the
direction o f the undisturbed flow, and therefore the critical Reynolds
number for two-dimensional disturbances will be subcritical for
three-dimensional disturbances. Consequently, in finding the criterion
11 4 STATISTICAL FLUID MECHANICS
o f Stability for plane-parallel flows, it is always permissible to assume

that ki = 0 , i.e., that the disturbance is independent of y. We can also
assume then that u!^(x. I) = v '(x, t) = 0 , because the fluctuations v'
have no effect on the system of equations in u' and w' (and will be
damped in time).* ®
For two-dimensional disturbances of a plane-parallel flow, the
general linearized equations (2.7) take the form
du' du' dU _ 1 dp'
w'
dt dx dz ~ P dx
dw' dw' _ _ _ 1 dp'
dt dx ~ dz +
V (2.25)
P
du' , d w ' __ ^
'5:^" + ~
where
is the two-dimensional Laplace operator. The boundary conditions

for these equations are that both components o f velocity u' and w'
become zero on the soHd boundaries of the flow (or, in the case of
an unbounded flow, at infinity, i.e., as 2 - > oo or as — oo or as
z ^ ± o o , depending on the nature o f the unbounded flow). Using
the equation o f continuity, in the first two equations o f Eq. (2.25),

we transform from the components of velocity «' and w' to the
stream function xp, putting
dif ,
After eliminating the pressure, we obtain the following partial

* *T h is d o e s n o t m e a n , o f c o u r se , also th at in th e c ase R e > R c c r o n ly tw o -d im e n s io n a l
d istu rb a n c es in d e p e n d e n t o f y are im p o r ta n t. S in ce th e e x istin g d a ta d e fin ite ly s h o w th a t in
tr a n sitio n to tu r b u le n c e of a tw o -d im e n s io n a l boundary layer, th re e -d im en sio n a l
d istu rb a n c es p lay a m ajor ro le [se e, fo r e x a m p le , K le b a n o ff, T id str o m and Sargent ( 1 9 6 2 ) ] ,
th e in v e s tig a tio n o f th e b e h a v io r o f su ch d istu rb a n c es at su p ercritical R e y n o ld s n u m b e r s is
also o f in ter e st. S u ch an in v e s tig a tio n w a s carried o u t , in particular, b y W atson ( 1 9 6 0 b ) and
M ichael ( 1 9 6 1 ) . T h e y s h o w e d th a t w ith in th e fr a m e w o r k o f th e linear d istu rb a n c e th e o r y ,
th ere a lw a y s e x ists a range o f values o f R e , R e c r < R e < R e i for a n y plan e-parallel flo w
w ith in w h ic h , o f all th e u n sta b le w a ve d istu rb a n c es, th e m o s t rapidly increasin g (i.e ., th a t
p osse ssin g th e gr ea test value o f I m tj) w ill b e s o m e tw o -d im e n s io n a l d istu rb a n c e (a lth o u g h ,
in th e case o f d istu rb a n c es w ith fix e d w ave n u m b e r k, for certain v alu es o f k th e m o s t
u n stab le w ill b e th r e e -d im e n sio n a l d istu rb a n c es). A p p a r e n tly , h o w e v e r , th r e e -d im e n sio n a l
d istu rb a n c es w ill o f t e n b e g in to play a fu n d a m e n ta l r ole for values o f R e o n ly s ligh tly
greater th a n R ecr, as a resu lt o f e ss e n tia lly n o n lin ear e f f e c t s (see b e lo w . S e c t. 2 .9 ).
differential equation for the stream function:
To find the instability criterion, it is necessary to find the solution

t|3 (x, z, t) o f this equation for the given boundary conditions and
given initial valueil)(-»i:, 2 , 0 ) = i|)o(a:, z ) . Here, tfoC-*:, z) is an arbitrary
function, nonzero within a finite region of space (and satisfying the
boundary conditions and small disturbance condition | Vi|)ol^ f/).
The nature o f the time-variation of max |Vi})(a:, 2 , 01 will depend on
(x, i)
the Reynolds number o f the undisturbed flow; for sufficiently small
Re, this value will always be damped, and for sufficiently large Re, it
may sometimes prove to increase with time. The boundary between
these regions o f values of Re (if they exist) will define the critical
Reynolds number.
In general, the solution of this initial value problem is very
complicated. However, since the coefficients of Eq. (2.26) are
independent o f t and x, this equation may have particular solutions
of the form
{x, z, t) = e‘ (*^-“Ocp{z) = (e), (2.27)
where c = (njk. If there are a sufficient number of such “wave-like

solutions” that it is always possible to resolve the arbitrary solution
z , /) o f the initial value problem into solutions o f the form
(2.27) with all possible (real) wave numbers k , then the general
problem o f finding the value o f Recr will be reduced to determining
only the permissible values of c. As a result, we must substitute Eq.
(2.27) into (2.26); we then obtain the following ordinary differential
fourth-order equation for (j)(z):
{U — c) (cp" — - U \ = — ^ (cp>V_ 2 k y + k \ ) (2.28)
(where the primes denote differentiation with respect to z). Equation

(2.28) is called the Orr-Sommerfeld equation. The boundary condi
tions for this equation are that q) and cp' become zero on the
boundaries o f the flow (which may be at infinity). The homogeneous
equation (2.28) with the given homogeneous boundary conditions
will have a nontrivial solution only for certain special values of the
parameter c. These eigenvalues o f the problem will be, generally

speaking, complex numbers c = Ci + icz, dependent on the wave
number k and on the viscosity v (i.e., on the Reynolds number Re).
The real part C| will have the meaning of the phase velocity of
propagation o f the wave, and the imaginary part C2 will define the
time variation o f the amplitude of the wave, which is described by
the factor Therefore, for £■2 < 0 the amplitude o f the wave will
be damped, for Cz= 0 it will remain constant (in this case the wave is
called neutral), while for Ca > 0 it will increase with time. Thus Recr
can be defined as the value of Re for which Cz(k, Re) is first equal to
zero, at least for one value o f k.
To find the instability criterion, it is only necessary to define cz as
a function o f k and R e. Further, we note that at the value R e = Recr
where C2 (^, Re) first becomes equal to zero, the real part ci of the
eigenvalue c{k,'Re) with Inn c = C2 = 0 will not, generally speak
ing, be equal to zero (breakdown of the “ principle of exchange
o f stabilities” )- This means that the eigenvalue c usually corresponds
to propagation o f the wave (2.27) along the Ox axis. For values o f Re
somewhat larger than Recr, a fairly small range of values of k will be
found, for which Cz will be positive (and Ci nonzero). Waves with
values o f k from this range will form a wave packet, which will grow
with time, moving simultaneously downstream. In fact, the value of
kcz will be greatest somewhere near the center o f this range o f values
of k ; thus, close to this center {kc,^ — 0 , and the group velocity
^ (^c) (A:c,) of our wave packet will be real, i.e., it will

represent the true velocity o f displacement of the packet [cf. Landau
and Lifshitz (1963), Sect. 29]. In this respect, the instability under
discussion will differ from that of the flows considered in Sects. 2.6
and 2.7, where the unstable disturbances did not move, and, at a
given point, increased to a finite value.
Stability o f Plane-Parallel Flows o f an Ideal Fluid
Substituting some actual velocity profile L^(z)into Eq. (2.28) we
arrive at a very complicated eigenvalue problem, the solution of
which requires the use o f sophisticated and cumbersome mathe
matical procedures. To simphfy these procedures, we begin with an
attem pt to use experimental data, according to which the critical
Reynolds number for many plane-parallel flows is extremely large.
Consequently, we may expect that for Reynolds numbers close to the
LAMINAR AND TUf^BULENT FLOWS 117
critical value, the terms on the right side of the Orr-Sommerfeid

equation (2.28), which describe the action of the viscous forces on a
small disturbance, will be small compared with the terms on the left
side. Therefore, first, we may try considering the fluid as ideal, i.e.,
we ignore the right side of Eq. (2.28) and consider the abridged
equation:
{U - c) (cp" - k^<f) - U"<f = 0, (2.29)
which was studied in detail by Rayleigh (1880; 1887; 1895; 1913).

Since this equation (called the Rayleigh equation) is not of fourth
order but only of the second, we can no longer demand that four
boundary conditions be satisfied. However, we must confine our
selves to the ordinary requirements for a flow of ideal fluid, e.g., that
on the boundary walls only the normal component of velocity is
equal to zero. Bearing in mind the definition of the stream function
\jj and Eq. (2.27), we conclude that the condition (^ = 0 must be
satisfied on the walls.
However, by completely ignoring the viscosity and using the
abridged equation (2.29), we encounter a number o f difficulties.
These are connected, first, with the fact that for real eigenvalues c
(corresponding to neutral waves, which are of considerable interest
for finding the instability criterion) there may be a value of z (say,
z = Zo) for which the velocity of the undisturbed flow U(zo) = t/o
will equal c, so that zo is a singular point of Eq. (2.29). Moreover,
Rayleigh [and later, by another method, Tollmien and Pretsch
(1946)] showed that in the case of plane-parallel flow of an ideal
fluid, the phase velocity c o f the neutral waves will always be
between the minimum and the maximum velocity o f the undisturbed
flow (i.e., there must be a singular point 2 o
within the flow.' * \i U' (zo) + 0, then, in the neighborhood of this
l^ T h e sim p le st p r o o f is the fo llo w in g [c f. D razin and H ow ard 1 9 6 6 ) ) . L et us rew rite th e

Rayleigh equation (2 .29) as
((U - cS^F'Y - k'âi - ( ) 2 f 0 , F ^ A il - c).
Multiplying this equation by the com plex conjugate function F * and integrating from to
^ 2 (where and Zg are boundaries o f the flow , so that F = 0 a t ^ = z^ and^ = z^) ob
tains for nonsingular F
I.
T h e last e q u a tio n im p lie s th at F c a n n o t b e n on sin g u la r w h e n c is real and th e r e fo r e th a t c
( c o n t ’d p . 1 1 8 )
point we may put U{z) — c [2 — ], and, as 2 Zo, the

function cp" will tend to infinity asThus the
^ —ô)
x-component of velocity of a neutral disturbance in the neighbor
hood of a singular point will take the form
“ -------- U' (z'o) " i^-ô h

Thus, for a neutral wave, one of the two linearly independent
solutions o f the abridged equation (2.29) will be discontinuous and
multivalued, which gives rise to the question: which branch o f the
multivalued function are we to take?
There is the further difficulty that when c in Eq. (2.29)
[corresponding to the eigenfunction (f( 2 )] is a complex number, its
complex conjugate c* will also be an eigenvalue of this equation
[corresponding to the eigenfunction (p* (2 ) ]. Consequently, in addi
tion to the damped wave, the equation for the stream function will
always have a solution in the form of a growing wave. Therefore, for
an inviscid fluid it is meaningless to define the stable case as the case
when only damped oscillations occur. Thus the very definition of
stability based on the consideration of elementary wave-like solu
tions must be changed. Stability now must be defined as the absence
of growing wave-like disturbances (so that the criterion for stability
is that the Rayleigh equation has only real eigenvahies c for any k).
The usual method of overcoming all these difficulties consists of
going back to the complete Orr-Sommerfeld equation (2.28). Here, we
select the proper branch of the solution of the abridged equation
(2.29), which describes neutral oscillations, as well as the wave solu
tions o f the equation with lmc=jfe 0 which have a physical meaning, by
carefully investigating the asymptotic behavior of solutions of the
fourth-order equation (2.28) as v -> 0 (i.e., as Re 0 0 ).’ ''
cannot lie beyond the range o f V.

N ote further that for com plex c = f k '2 with Cg 0 the imaginary part o f the
integral relation above implies t h a t c j = R e c cannot lie beyond the range o f U (this result
was also first demonstrated by Rayleigh). Finally, simple manipulations with both the real
and imaginary parts o f the same integral give the semicircle theorem by Howard (1 9 6 1 ).
This theorem states that all eigenvalues c with Im c 0 (corresponding to unstable
waves) lie in a semicircle o f the com plex c-plane with center in the real point
radius eigenvalue c, real or complex,
must lie in or on the circle with center 2 (^max " and radius
^ ^We n ote that in the case o f neutral w a v e s -w it h real c - w e need to take into account
the viscosity only in the neighborhood o f the singular point z = Zq, where an additional
“internal boundary layer” is introduced.
In particular, by using this approach it was found that the

ampHfied and damped wave solutions of Eq. (2.29) which have
physical meaning, are not conjugate to each other, and that in certain
cases, such solutions must be constructed by “sticking together”
different expressions in different domains of z [see, for example, Lin
(1955), Chapt. 8 ; Stuart (1963); Shen (1 9 6 4 );and Reid (1965)]. Of
course, this approach to inviscid stability theory is fairly cumber
some, and the advantages given by the relative simplicity of Eq.
(2.29) are in fact lost. However, there is another approach to the
same problem which allows the use of the relative simplicity o f the
inviscid equations; this will be discussed later.
Let us now give the main results o f the stability theory of
plane-parallel inviscid flows. The im portant works o f Rayleigh
(mentioned above) served as a starting point of the theory. In
particular, as early as 1880 he showed that i f U"(z) is nonzero
everywhere within the flow, then the abridged equation (2.29)
cannot possess complex eigenvalues with Im 0. The proof o f this
theorem is very simple. Rewriting Eq. (2.29) in the form
multiplying throughout by the complex-conjugate function tp*and

integrating from z = Zi to 2 = Z2 (where and Z2 are the boundaries
o f the flow on which cp(2 )is equal to zero) we obtain
f i W ? + k'^W\'^)dz+ = (2.30)
With c-i 4= 0, the imaginary part of this equation can become zero
only if C/"(z) changes sign somewhere between 2 , and 22. Thus we
obtain the result cited. It was widely accepted for many years that
this Rayleigh theorem gave complete proof of the stability (in the
sense of the absence of growing disturbances) of plane-parallel flows
o f ideal fluid, the velocity profile of which does not possess an
inflection point. (We note that for an ideal fluid with v = 0, any
velocity profile U ( 2 ) will be permissible from the viewpoint o f the
equations o f motion, unlike the case of a viscous fluid.) However,
this, in fact, is misleading. The singularity of Eq. (2.29) connected
with the coefficient o f a higher derivative becoming zero, leads to the
appearance o f a continuous spectrum o f eigenvalues c (in addition to

the usual discrete spectrum) while the eigenfunctions for eigenvalues
o f the continuous spectrum are “generahzed functions,” i.e., they do
not satisfy the usual conditions imposed on eigenfunctions (for
example, they may contain Dirac’s -functions) and very often
simply go unnoticed (in particular, the proof of Rayleigh’s theorem
refers only to simple discrete eigenvalues). It is interesting that the
existence o f the continuous spectrum was known to Rayleigh
[(1894), Vol. 2, pp. 3 9 1 -4 0 0 ], but its importance was completely
disregarded until recently. If we confine ourselves to a discrete
spectrum, the boundary-value problem for the Rayleigh equation
(2.29) will usually have no more than a finite number of eigenfunc
tions for every fixed k. Consequently, the set of these functions will
evidently be incomplete (e.g., for a Couette flow, where U" 0,
there will be no eigenfunctions which vanish on the two boundaries;
later, we shall see that this situation occurs rather often). However, if
we also consider the continuous part o f the c-spectrum, we must first
define precisely what we mean by corresponding eigenfunctions and
investigate their behavior as / oo; then we must prove that the
system of functions obtained is complete, which is no easy task.
Thus, in fact, the proof of stability o f an inviscid flow by considering
only elementary wave-like disturbances does not follow from
Rayleigh’s theorem alone, but requires additional sophisticated
procedures.
In this connection Case (1960a,b) and Dikiy (1960a,b) in
dependently of each other, stated that for inviscid stability analy
ses it is advantageous to discard the consideration of elementary
wave-like solutions o f the form (2.27). Instead, one should solve the
general initial value problem for the partial differential equation
(2.26) with zero right side, i.e., with v = 0 , and initial conditions
o|5(a:, z , 0) = \1)o(a:, z ) [this is the second approach to the inviscid
stability problem which we mentioned eariier]. Both authors solved
the initial value problem by using the Laplace transform with respect
to time, and obtained the solution in the form of a Laplace integral,
the asymptotic behavior of which as t - ^ o o may be studied by
ordinary methods o f the theory of functions of a complex variable.
The integrand in the corresponding Laplace integral may also be used
for accurate determination o f the set o f eigenfunctions (for both the
continuous and the discrete spectrum) o f Eq. (2.29); the complete
ness o f this set o f functions will then be obtained automatically
(since the solution o f an arbitrary initial value problem is expanded
in it). However, as already mentioned, the consideration o f elemen

tary wave-like solutions satisfying Eq. (2.29) is quite unnecessary for
the study of stability, since the behavior as / c x d of the general
solution of the initial value problem will immediately determine

whether or not the flow will be stable. Case (1960a) and Dikiy
(1960b) independently of each other, outlined the proof o f the fact
that in all cases, increasing solutions of the initial value problem for
Eq. (2.26) can exist only when the corresponding equation (2.29)
has discrete complex eigenvalues c, for which Imc > 0 (or multiple
discrete real eigenvalues c). More precisely, they also outlined the
proof that the discrete spectrum alone is associated with instability
(while the part of the Laplace integral respresentation of z, t)
over the continuous spectrum always decays like 1 // as ^ o o ) .
According to Rayleigh’s theorem, complex eigenvalues of Eq. (2.29)
cannot exist if the profile D(z') has no inflection points. A more
detailed investigation of Eq. (2.29), described in Lin (1955), Sect.
8 .2 , shows that a change of sign o f ( z ) within the flow is also
necessary for the existence of purely real eigenvalues c; in other
words, in the absence of an inflection point of the velocity profile
Eq. (2.29) has no discrete eigenvalues at all. Therefore, the results o f
Case and Dikiy give the first rigorous justification for the deduction
which is generally made from Rayleigh’s theorem, concerning the
stability o f inviscid plane-parallel flows without inflection points of
the velocity profile. It is interesting that for the special case of a
plane Couette flow in an ideal fluid, the approach described here was
used as early as Orr (1906—1907), who solved the general initial
value problem for a small disturbance in such a flow and showed that
the solution always decays [cf. Drazin and Howard (1966), p. 2 8 ];
later, however, this result of Orr’s was set aside. Dikiy and Case
showed also [c f especially Case (1961)] that the solution o f the
problem with initial condition ajj(x, z, 0 ) = z)for the complete
equation (2.26) always tends to the solution o f the same problem for
the abridged equation as v->-0. Thus, in this respect also the initial
value problem has a considerable advantage over the boundary-value
problem for Eq. (2.28), in which there is no simple connection
between the eigenvalues and eigenfunctions of the complete and
abridged equations [cf. Lin (1961), and Lin and Benney (1962)]. An
extensive review of the initial-value-problem approach to the study
o f inviscid hydrodynamic stability can be found in Drazin and
Howard (1966); rigorous proofs of some results in this field, only
outlined by Case and Dikiy, were given (under additional restric

tions) by Rosencrans and Sattinger (1966).
Let us now assume that the velocity profile has an inflection
point, i.e., that l)"(z) becomes zero at least at one point. According
to the theorem of Rayleigh, this condition is necessary but not
sufficient for instability of inviscid plane-parallel flow. One very
simple and stronger condition necessary for instabihty was noted by
Fj^rtoft (1950) and Holland (1953). To obtain this condition we
must only add the real part o f Eq. (2.30) to the imaginary part o f the
same equation multiplied by the factor (c^ - Ug)/c2 where Vg = U(zq)
is the value of the velocity at the inflection point. Then we get
U"iU -
dz = - + k^\‘p\^) dz < 0 .
(U - + cl
Therefore, not only must U"(z) vanish in some point z = z^, but also
U"(U - Ug) must be negative somewhere in the field o f flow. In
particular, if (J(z) is a monotonic function and U'(z) vanishes in one
point Zq only, then for instability the inequality U"(U - U^) < 0 must
be fulfilled at every z Zq ( s o that, for example, the flow with the
velocity profile shown in the figure below a may be unstable, but the
a
U 7 b
u
Two types o f velocity profiles with one inflection point.
flow with the velocity profile of the figure b is always stable). This new
necessary condition of instability is also insufficient, in general. For
example, Tollmien (1935) showed that a flow with the velocity
profile UU) = sinz in the layer < 2 < Zg. < 0 < Zg. is stable if
although the Fj^rtoft—Holland condition is vahd here
[cf. Drazin and Howard (1966), p. 35]. However, in the same paper,
Tollmien proved that the presence of an inflection point o f the
velocity profile is sufficient for instability in very im portant cases of
channel flows with symmetric velocity profiles and flows o f the

boundary-layer type with velocity profiles of the form shown in Fig.
a [see also, Lin (1955), Sect. 8.2; Shen (1964), Sect. G, 7; and
Drazin and Howard (1966)]. For all such flows the vanishing of
U"{z)at some point implies that the Fj^toft-H ^land condition is also
fulfilled and that the Rayleigh equation (2.29) will have eigenvalues c
such that lmc>0. One other necessary condition of instability
which is also sufficient in the case of a monotonic profile U (z) with
only one inflection point was given by Rosenbluth and Simon
(1964).
The results o f Case, Dikiy, and Rosencrans and Sattinger indicate
that the continuous spectrum of the Rayleigh equation is generally
(or even always) stable. Thus the investigation of the instability
characteristics of various plane-parallel flows of nonviscous fluid is
equivalent to the finding of all discrete eigenvalues and eigenfunc
tions o f the corresponding Rayleigh equation. This eigenvalue
problem is difficult to solve explicitly in the case of smoothly
varying functions U{z). However, when U{z) is a piecewise-linear
polygonal contour, one can easily find exponential solutions o f the
stability equation for different z-regions and then obtain the explicit
eigenvalue relation by using the natural “joining conditions” at the
corners of U (z). This method was first used by Helmholtz (for the
schematic case of a tangential velocity discontinuity), Kelvin and
Rayleigh [see especially the classical book by Rayleigh (1894), Vol.
2, Chapt. XXI]; it was also recently used by Drazin and Howard
(1962; 1966) and Michalke and Schade (1963).
The introduction of electronic computers changed the situation;
now it is easy to solve the eigenvalue problem for the Rayleigh
equation numerically even in the case of smooth profiles t/ ( 2 ). Many
examples o f such solutions can be found in surveys by Drazin and
Howard (1966), Sect II.4, and Betchov and Criminale (1967),
containing references to original papers.
The results for the cases of a Bickley jet with U(z)= sech^z,
-00 < z < 00, and a hyperbolic-tangent free boundary layer with t/(z)=
0.5(1 + tanhe), -00 < 2 < c», were compared with experimental data
obtained by Sato (1960), Freymuth (1966) and others who measured
the growth o f artificial disturbances introduced in jets and jet
boundary layers by sound from a loudspeaker [see the review by
Michalke and Freymuth (1966)].
The first results of the comparison were satisfactory but more
detailed investigations show that the experimental data deviate in
some respects from the calculated ones. One possible explanation is

that the theory considers strictly wave-like disturbances growing in
time, while in the experiments we have time-periodic disturbances
growing spatially in the direction of the mean flow. To use the
theory for the explanation of the experimental data, one transforrtis
the temporal growth rate linearly into a spatial growth rate by means
of the disturbance phase velocity (as was done first by Schubauer
and Skramstad for the case of a viscous boundary layer; we shall
discuss this work later). However, the transformation indicated is not
a rigorous one, as was explained by Gaster (1962; 1965). Therefore
the theory of time-periodic spatially growing disturbances based on
the investigation o f the eigenvalue problem for the Rayleigh equation
(2.29) with fixed (real) frequency o> = kc and unknown eigenvalues
(generally complex) k is more suitable for the description of this
experimental data. Such a theory was developed by Michalke (1965b)
for the hyperbolic-tangent shear layer and by Betchov and Criminale
(1966; 1967) for several inviscid jet and wake profiles. The results of
Michalke’s calculations were in excellent agreement with the experi
mental findings o f Freymuth (1966) for all moderate Strouhal
numbers St = cj8/U q (where 8 and Vq are characteristic scales of
boundary-layer thickness and jet velocity). For greater values o f the
Strouhal number it is necessary to take into account the deviation of
the real velocity profile from the hyperbolic tangent (see Michalke
(1968a) on spatially growing disturbances in free shear layers of
different velocity profiles).
In the case of a fluid having density varying with height (i.e., with
the z coordinate), we will retain even with nonzero viscosity the
same difficulties which arose in the instability investigation o f
inviscid plane-parallel flows; i.e., even when v # 0 , the corresponding
Orr-Sommerfeld equation will have a singularity at the point where
f/( 2 ) = c [see, for example, Dikiy (1960a)]
Consequently, we are not justified in transferring to this case the
constant density rules for selecting the proper branch o f the
multivalued solutions, as was done by Schlichting (1935b). The
instability o f a horizontally stationary layer o f a heavy fluid of
variable density was studied in the so-called Rayleigh-Taylor insta
bility theory [see, e.g., Chandrasekhar (1961), Chapt. X; Selig
18 In fa c t, i f in a d d itio n to n o n z e r o v is c o s ity , w e a lso ta k e in t o a c c o u n t n o n z e r o th erm al

c o n d u c t iv ity , w e shall o n c e again o b ta in an ord in ary eig e n v a lu e p r o b le m (th is tim e o f the
six th o r d e r) w it h o u t a n y singula rity in th e c o e f f ic ie n t o f th e le a d in g term [se e K o p p e l
(1964)].
(1964)]. At present, the results for viscous heterogeneous fluid in

horizontal motion are very incomplete [cf. Drazin (1962)]. We must
note that for the case o f variable density, the reduction of the
determination of the instability criterion to an eigenvalue problem,
without considering a continuous spectrum, is unjustified even if v
0 , due to the incompleteness o f the corresponding system o f
eigenfunctions. Therefore, for flow o f fluid with density varying with
height, both for v = 0 and v =/= 0 , a strict stability analysis demands
the study o f the asymptotic behavior as ^ oo of the solution o f the
corresponding initial value problem. This analytical problem is very
difficult and what progress has been made in this field was achieved
fairly recently and only on the assumption that v = 0 (i.e., for an ideal
fluid).
Thus, Eliassen, Hyland and Riis (1953) and, later, more accur
ately, Dikiy (1960a) and Case (1960b), investigated the asymptotic
behavior o f the solution of the initial value problem for the stream
function o f the disturbance for a two-dimensional flow o f a
nonhomogeneous heavy fluid with exponentially decreasing density,
filling an unbounded half-space and possessing a Unear velocity
profile. They showed that just as in the case o f plane-parallel flows of
an ideal fluid, instability (in the sense o f the presence o f increasing
disturbances) can arise in this case only when there exist wave-like
disturbances o f the form (2.27) with increasing amplitudes (i.e., with
Im c > 0 ).
By virtue of the results of the investigation o f the eigenvalue
problem carried out by G. I. Taylor in 1915 [but published much
later; see Taylor (1931)], supplemented recently by a more strict
mathematical analysis of the behavior of the zeros o f the confluent
hypergeometric functions, carried out independently by Dikiy
(1960c) and Dyson (1960), it follows that such a flow of
inhomogeneous fluid will be stable (in the sense indicated) for all
positive values of Richardson number Ri (and not only for Ri > 1/4
as was asserted in the well-known books and survey paper o f Prandtl
(1949) and Schlichting (1960; 1959). In fact, in this case the
corresponding eigenvalue problem has no discrete eigensolutions
when 0 < Ri < 1/4 and only stable ones when Ri > 1 /4 . Later, a
similar means of investigation was used by Miles (1961) for the
general plane-parallel flow o f inviscid heterogeneous heavy fluid with
a practically arbitrary density profile p (z.) [such that p' (z) < 0 for all
z, and that the density decreases monotonically with h eigh t], and any
monotonic velocity profile i7(z). Miles showed that in this case also.
it follows from the existence of unstable disturbances that unstable

waves exist with stream function o f the form (2.27). In addition, he
gave a detailed investigation of possible wave-like disturbances in this
flow, and as a result, was able to show that the flow under discussion
will certainly be stable if at all o f its points Ri > 1/4 (special cases of
the last result are contained in the works of Goldstein (1931) and
Drazin (1958) on wave-like disturbances in some special flows of
ideal fluid o f variable density). Miles’ general result was also proved
much more simply (with the aid of some integral inequalities similar
to those used by Rayleigh) and for more general conditions (without
restrictions on the velocity and density profiles) by Howard (1961).
In the same work the semicircle theorem was also proved for general
variable-density flows (in fact the theorem was first demonstrated for
this general case). In addition, Case (1960c) also analyzed the
behavior o f the solution o f the initial value problem for disturbances
in a motionless ideal fluid in the half-plane z > 0 with exponentially
increasing density; he naturally found a solution which increases in
time without limit (see above, Sect. 2.4), and estimated its maximum
rate of increase. Some other examples of the analysis o f initial value
problems for disturbances in fluids with density increasing with z
(but with nonzero velocity gradients) may be found in the work of
Eliassen, Holland and Riis (1953). A detailed investigation of the
possible wave-like disturbances in some special plane-parallel flows
with density stratification was made by Miles (1963) and Howard
(1963). They showed that the principle of “exchange o f stabilities”
(see Sect. 2.6) holds only under certain rather special conditions,
while the region o f instability in the (k, Ri)-plane may consist here of
several disconnected parts.
A detailed discussion of the inviscid stability theory of plane-
parallel flows (both for homogeneous and heterogeneous fluids and
also under the action o f various force fields) can be found in the
excellent survey by Drazin and Howard (1966).
th e General Theory o f Instability fo r Plane-Parallel Flows o f Viscous

Fluid
Let us now return to the more complex case of flows of viscous
fluid; the density P , as usual will now be assumed constant. In this
case, the Orr-Sommerfeld equation (2.28) plays a central role. It is a
nonsingular equation which has a purely discrete spectrum; therefore
there is no reason to doubt that any solution of Eq. (2.26) may be
represented by superposition of “ plane waves” of tlie form (2.27).*’

Here we may limit our attention entirely to the study o f the usual
eigenvalue problem for a discrete spectrum of the Orr-Sommerfeld
equation (2.28). The first attem pts at such a study were made (using
mathematical methods that were not very strict) circa 1910 by a
number of authors (W. Orr, A. Sommerfeld, R. von Mises, L. Hopf,
et al.) for the special case o f plane Couette flow with a linear velocity
profile. These attempts led to the result that such a flow is stable for
all Reynolds numbers. This result seemed, on the one hand, fairly
natural (since Orr (1 90 6 -1 9 07 ) had shown rigorously that in the
absence of viscosity, a Couette flow is stable, while the action of
viscosity is naturally assumed to be stabilizing) but, on the other
hand, it clearly contradicts the empirical facts on the transition to
turbulence o f all known flows for sufficiently great Reynolds
numbers. At the beginning o f the 1920’s, Prandtl (1921) and Tietjens
(1925) considered the question of the stability of flows with a
velocity profile consisting of segments of straight lines. They came to
the most unexpected conclusion that in the presence of viscosity,
such flows are unstable for any Reynolds numbers (however small).
During the same few years, Heisenberg’s important work (1924) on
the stability of a plane Poiseuille flow, using the method o f small
disturbances, appeared. In this work, a result was obtained which, at
the time, seemed paradoxical (but which was, nevertheless, correct).
By investigating in detail the asymptotic behavior of the solution of
the corresponding Orr-Sommerfeld equation with large Re (i.e., small
v), a Poiseuille flow which in the absence of viscosity will be stable
to small disturbances, will become unstable in the case of a viscous
fluid, for sufficiently large Reynolds numbers. Heisenberg’s result
was obtained by sophisticated mathematical procedures, and on a
physical (rather than mathematical) level of rigor; therefore it is not
surprising that it aroused serious doubts for a long time, and proofs
of the stability of Poiseuille flows continued to be pubUshed for
many more years (apparently the last such “proof” was that o f C. L.
Pekeris in 1948). The physical mechanism of the very interesting
phenomenon o f instability caused by viscosity was outhned by
Prandtl in the 1920’s, but did not become clearly understandable
19 Nevertheless, we must note that the question o f rigorous proof of the possibility of
such a r e p re sen ta tio n w as a p p are n tly stu d ied o n ly in a very o ld pap er b y H au p t ( 1 9 1 2 ) and
pro bab ly also in th e u n p u b lish ed d isser ta tio n o f S c h e n s te d ( 1 9 6 0 ) . H e n c e th ere is n o reason
at presen t t o c on sid e r th e q u e s tio n as b e in g c o m p le t e ly so lved .
until much later [Lin (1955), Chapt. 4; see also, Betchov and
Criminale (1967)].
In the work o f Tollmien (1929), the method of small disturbances
was used to investigate the stability o f boundary-layer flow, this
being considered as plane-parallel, with a velocity profile composed
o f straight-line and parabolic segments. Tollmien was the first to
obtain the form o f the curve o f neutral stability in the {k. Re) plane,
dividing the regions o f stable and unstable disturbances. Later,
Tollmien (1930; 1947) and Schlichting (1933a, b; 1935a) extended
these results to arbitrary velocity profiles. In 1944—1945, the whole
theory o f stabiHty o f plane-parallel flows was reconsidered critically
by Lin (1945) who reworked the basic examples and improved the
numerical results o f Tollmien and Schlichting. Nevertheless, the
complexity o f the methods used for the asymptotic analysis of the
solution o f the Orr-Sommerfeld equation (2.28) leads to the fact that
even now the results obtained cannot be taken as definitive in certain
respects. The difficulty is that the asymptotic series used normally
have a singularity at the point z where U(z) — c = 0 , while the initial
equation is regular at that point. Therefore, there is great interest in
the problem of finding uniformly convergent asymptotic expansions,
but the formulation of such expansions encounters great difficulties
[see, for example, Lin and Benney (1962)].
Many recent results in the extensive field o f refined mathematical
analysis o f various asymptotic approximations for solutions o f the
Orr-Sommerfeld equation can be found in Stuart (1963), Shen
(1964) and Reid (1965; 1966); see also, Goering (1959); Chen,
Joseph and Sparrow (1966); Graebel (1966); Betchov and Criminale
(1967), Appendix II; Tam (1968); and Huppert (1968).
We have already pointed out that, strictly speaking, a plane-
parallel flow o f viscous fluid can only be a combination of Couette
and Poiseuille flows. Therefore it is natural to begin the study of
special plane-parallel flows of viscous fluid with consideration of
these two examples. Let us first consider a plane Couette flow. At
present, there is little doubt that this flow is stable with respect to
any infinitely small disturbance; nevertheless, so far, a strict proof of
this fact has not been given. Most investigations of this problem use
certain asymptotic expansions for the study o f limiting cases (usually
cases of very large Re) and direct numerical procedures for the
calculation o f eigenvalues and eigenfunctions at moderate parameter
values. This approach was developed first by Southwell and Chitty
(1930) [however, they confined their attention to purely imaginary
eigenvalues, taking on faith the “principle of exchange of stabil

ities” )- Recently, this method has been developed widely in
connection with considerable improvements to the technique of
asymptotic approximation and the use of computers for numerical
calculations. One of the most important contributions was made by
Grohne (1954), who discovered the existence o f a sequence o f modes
of wave-like disturbances in a plane Couette flow at given k and Re
and computed the complex eigenvalues c = c(k, Re) for the twelve
lowest modes. All these eigensolutions -correspond to damped
oscillations with ca = Imc < 0 for all k and Re. The form of the
twelve functions C2 = cz(Re) for one fixed k given by Grohne can be
found also in Shen (1964) and Betchov and Criminale (1967)
[although it must be borne in mind that numerically Grohne’s results
do not agree very well with those of subsequent computations by
Gallagher and Mercer (1964)]. In subsequent numerical works using
computers, Riis (1962) [who also investigated wave-like disturbances
in a flow o f heterogeneous viscous fluid with a linear velocity
profile], Gallagher and Mercer (1962; 1964), Deardorff (1963),
Birikh (1965) [he gave special consideration to the case o f small
values o f R e ], Hains (1967), Ponomarenko (1968c), and others, only
damped oscillations were found. Therefore, it is very likely that no
unstable disturbances exist in plane Couette flow, although the
problem cannot be considered definitely settled at present since the
entire domain of the parameters, which cannot be handled by
asymptotic estimates, is not exhausted. Apparently, the only exact
mathematical result relating to the eigenvalues of Eq. (2.28) with
U{z) = Az, for arbitrary Reynolds number, belongs to Dikiy (1964),
who proved (without use of numerical methods) that all the purely
imaginary eigenvalues c of this equation will satisfy the inequality Im
c < - k /R e (where Re is the Reynolds number, defined for the value
H, = HI2 of the half-width o f the channel, and the maximum
velocity o f the flow U = 2AHi). This result, in particular, shows that
in this case the transition to instability cannot occur while satisfying
the “ principle o f exchange o f stabilities,” but it does not guarantee
that instability, in general, is impossible.
The situation for plane Poiseuille flow is far more satisfactory.
Here, Lin’s careful calculations based on asymptotic expansions o f a
special type, and applicable for large Re, confirmed the fundamental
conclusion of Heisenberg on the instability of this flow for large
enough (but finite) Re and permitted the form of the curve of
neutral stability lmc(fc, Re) = 0 to be found. Afterwards the
problem was reexamined by Thomas (1953) with the aid of the

numerical solution o f the corresponding eigenvalue problem, and
then also by Shen (1954), Lock (1955), Nachtsheim (1964), Grosch
and Salwen (1968), and others, who used improved forms o f the
asymptotic expansion and/or better computers. The results of all the
calculations are in satisfactory agreement with each other (see, for
example, Fig. 14). The instability of the plane Poiseuille flow at
sufficiently large Re was also rigorously proved by Krilov (1964)
without the use of numerical methods. Shen (1954) used Lin’s methods
to calculate the form o f a family of stability curves lmc(&, Re) =
const, which define the set o f disturbances o f plane Poiseuille flow
with a given growth rate [cf. Lin (1955); Shen (1964); Betchov and
Criminale (1967)]. The same calculations were later repeated by
Grosch and Salwen (1968), whose results agree well with those of
Shen.
FIG . 1 4 . T h e fo r m o f th e n eu tra l curve in th e (A:,Re)

p lan e fo r a p lan e P o ise u ille f lo w a c co r d in g t o th e cal
c u la tio n s o f L in ( 1 9 4 5 ) and L o c k ( 1 9 5 5 ) .
The form o f the “neutral curve” lmc(/:, Re), - 0, where

Re = UHt/v (//, is the half-width of the channel and U is the
maximum velocity of the undisturbed flow), found by Lin and Lock
is shown in Fig. 14. The critical Reynolds number Recr (correspond
ing to the point o f this curve farthest to the left) is equal to about
6000 according to Lock, and approximately 5300 according to Lin;
in both calculations, it corresponds to the value « I/H i (close
values o f Recr ~ 5780 and k » \.02/Hi were obtained by Thomas
whereas according to Nachtsheim Re^;- 5767, kcr ^ 1.02/Hi and ac
cording to Grosch and Salwen Re^r 5750, kcr ^ 1.025/H. As Re->-
°°, both branches of the neutral curve (both upper and lower)
tend to zero (the upper branch is asymptotically proportional to
Re '/", and the lower is proportional to Re-'A). Thus, as Re increases,

the disturbances with fixed (but not too great) k lie first in the
region o f stability (i.e., they are damped), then fail in the region of
instability, and, finally, once again lie in the region o f stability. From
this we see why in the limit as Re (i.e., as v->-0) the flow
becomes stable with respect to any disturbance.
Grohne (1954) proved that the Orr-Sommerfeld equation has
higher modes for the eigenvalues also in the case of the basic
Poiseuille velocity profile. He calculated the twelve lowest complex
eigenvalues c = c(k,Re) and found that all the modes, except for the
first (which is the only mode studied by all the authors mentioned
above), are strongly damped. Some of Grohne’s results for plane
Poiseuille flow are reprinted in Shen (1964) and Betchov and
Criminale (1967). Later, the eigenvalues and eigenfunctions of the
higher modes were recalculated by Grosch and Salwen (1968). They
obtained results in full qualitative (but not too close quantitative)
agreement with Grohne’s results.
The most general, stricfly steady, and plane-parallel flow of
viscous fluid is the combined plane Couette-Poiseuille flow arising in
a plane channel with a moving upper wall in the presence of a
constant longitudinal pressure gradient. The velocity profile of such a
flow may be written in nondimensional form as i/i(zi) =
( 4 - / l ) z , - ( 4 - Z 4 ) z ? , 0 < z , < l , where z^ = U{z^M)IU(O.SH),
and /4 = C/i(l) is the nondimensional parameter proportional to the
velocity of the upper wall and changing from A = 0 for pure
Poiseuille flow to A = 2 for pure Couette flow. The stability of such
Couette-Poiseuille flow was studied by Potter (1966) and Hains
(1967) whose results were in agreement with each other. According
to their results the superposing of a Couette flow on a Poiseuille flow
always has a stabilizing effect and increases Re^;. considerably; for ex
ample, the change of the upper wall velocity from zero (for Poiseuille
flow) to 10% of the maximum Poiseuille velocity leads to an increase
of Recrof 236%. The plane Couette-Poiseuille flow apparently becomes
stable to all infinitesimal disturbances (i.e., Recr tends to infinity)
when A reaches a value of about 0.55—long before the linear velocity
profile is reached when A = 2.
Similar results were obtained by Potter (1967) also for the case of
model flow with a symmetric parabolic nondimensional velocity
profile of the form t / , ( z , ) = ‘ ( / l - l ) z ] + / l z i ^-1 , 0 > z i > - l ; f / , ( z i ) =
0 < z i < 1. Here, also, yl = 0 corresponds to Poiseuille flow
and increasing A corresponds to increasing Recr; complete stability to
all infinitesimal disturbances at all finite Reynolds numbers is

reached apparently when A ^ 0.437 (i.e., long before the symmetrical
linear velocity profile is reached when A = J).
Further Results on the Stability o f Plane-parallel Flows o f Viscous
Fluid. The Case o f Boundary-layer Flows
Let us now consider the most important case of flow in a
boundary layer over a flat plate. According to the results of Sect.
1.4, since the thickness of the boundary layer increases compara
tively slowly and the vertical velocity is considerably less than the
horizontal velocity, it is reasonable to assume that in this case also
we may use, with fairly good precision, the theory of stability for
plane-parallel flows.^® Hence in their calculations Tollmien and
Schlichting paid considerable attention to the stability of plane-
parallel flows with velocity profiles identical or close to the Blasius
profile (Fig. 2, Sect. 1.4). As a result, they were able to plot the
neutral curve Im c(A’,Re) = 0 , i.e., to determine the boundary of the
region o f instability in the (k,Re)-plane, which, according to their
data, has approximately the same form as the curve Im c{k, Re) = 0
in Fig. 14. The critical Reynolds number Rea* cr = ^ ^ .where 6 * is
the displacement thickness of the boundary layer defined by Eq.
(1.53), which, according to Tollmien’s calculations equals 420, by
Schlichting’s equals 575.
According to Eq. ( 1.56), this means that Re* cr = ( Ux/v) cr = 0.6 X
10®, according to Tollmien and Re* cr = 11 X 10®, according to
Schlichting. These values are somewhat lower than those given in
Sect. 2.1 because the loss of stability does not have to be
accompanied by an instantaneous transition to the turbulent regime.
At the point of loss of stability only some growing oscillations occur
which are amplified as they move downstream. For some larger value
of X transition to developed turbulence occurs.
On the other hand, however, the values obtained for Recr cannot
in themselves be taken as a confirmation of Tollmien-Schlichting
2 0 S o m e p relim in ary e stim a te s o f th e e ff e c t o f a d d itio n a l term s in th e O r r -S o m m erfeld
e q u a tio n s for a lm o s t (b u t n o t e x a c t ly ) plan e-parallel flo w s , w h ic h p e rm it th is a s su m p tio n to
b e partly ju s tifie d , w ere o b ta in e d as early as b y P retsch (1 9 4 1 a ) . L ater, th is p r o b le m w as
also s tu d ied b y o th e r a u th o r s, and w a s d isc u sse d b y R e id ( 1 9 6 5 ) and B e t c h o v and C rim inale
( 1 9 6 7 ) , S e c t. 5 4 , w h e r e refer e n c e s to o th e r w o r k s can also b e fo u n d . N e v e r th e le s s, th e
q u e s tio n o f th e e x a c t n e s s o f th e plane-parallel a p p r o x im a tio n in th e stu d y o f s ta b ility o f
quasi-parallel flo w s is at p r e sen t far fr o m b e in g d e fin ite ly s e ttle d a n d th e m ain argu m e n t in
favor o f th e p e r m issib ility o f such a p p r o x im a tio n is th e rather g o o d a g r ee m e n t b e tw e e n th e
th e o r e tic a l r esu lts an d th e e x p e r im e n ta l data.
theory, and thus it is not surprising that as early as G.I. Taylor (1938),
doubt was expressed as to the legitimacy o f applying this theory to
real flows in a boundary layer that is not, in fact, plane-parallel.
However, the direct experimental verification o f the Tollmien-
Schlichting theory, carried out by Schubauer and Skramstad (1947),
using the example of a flow in a boundary layer, confirmed in basic
outline all the deductions of the theory, showing that doubts as to its
correctness were unfounded. After this, Lin (1945) and Shen (1954)
repeated the calculations of Tollmien and Schlichting and obtained
once again the value Re 8»cr = 420. However, the form they obtained for
the neutral curve differed slightly from that found previously (their
result is shown in Fig. 15). More recently, the same problem was
reexamined several times by different investigators with the aid of
machine computations [see the comprehensive review by Betchov
and Criminale (1967); cf. also Reid (1965)] without finding any
significant disagreement with the results of Lin and Shen.
0/f /
V
02 \
%»
FIG. 1 5 . T h e fo r m o f t h e neu tral curve in th e
(A:,Re5 *) p lan e fo r a b o u n d a r y layer o n a flat
plate.
The experiments of Schubauer and Skramstad were carried out

at the National Bureau of Standards in Washington, D.C., in a
wind-tunnel having extremely low initial turbulence. The parameter
U'lU in this tunnel, when certain precautionary measures are taken,
may attain values o f the order 0.0003—0.0002. This proves to be
very important, since there are some data showing that when the
values o f exceed 0 . 0 0 2 (i.e., in particular, for the values
obtained in all the older tests), the transition to turbulence in the
boundary layer is apparently affected strongly by the influence of
the free-stream disturbances [in accordance with Taylor’s scheme
described in Sect. 2.2, or as a result of another mechanism o f
interaction o f the free-stream with the boundary layer flow studied
by Criminale (1967)]. However, when U'lU < 0.002, the main role in
transition is played by random two-dimensional disturbances of

sinusoidal form, the amplitude of which under certain conditions
increases downstream in accordance with the deductions of the linear
disturbance theory. The presence of such regular oscillations was
demonstrated as early as 1940 by Schubauer and Skramstad using
careful hot-wire anemometer observations. Later, for more accurate
verification o f the deductions o f the theory, they used a thin metal
ribbon placed in the boundary layer, set in oscillation by an
electromagnet and producing artificial disturbances of fixed fre
quency (0 . They succeeded in demonstrating neutral (neither growing
nor damped) almost purely sinusoidal fluctuations of velocity,
corresponding to points o f the neutral curve in the stability diagram.
This classical experiment of Schubauer and Skramstad was repeated
in subsequent years by other members of the National Bureau of
Standards group, confirming the previous findings (cf. Betchov and
Criminale (1967), Chapt. VI). Similar experiments have also been
performed by Wortmann (1955) using water as the fluid and special
flow visualization technique, and by Burns, Childs, Nicol and Ross
(1959), who used a silvered lamina placed in the flow instead of a
hot-wire anemometer to observe the oscillations of the reflection of a
light beam directed on to the lamina caused by the z-component of
the velocity disturbances. The observations by Wortmann and by
Bums et al., lead to results which agree closely with those o f
Schubauer and Skramstad.
To compare the experimental data with the deductions of the
theory, it is convenient to consider, instead o f the more usual
space-periodic disturbances of the form (2.27) with fixed (real) wave
number fe , time-periodic disturbances of the form
with fixed (real) frequency co and variable (generally speaking,
complex) wave number k. In other words, it is expedient to
investigate the space variation along the plate (i.e., with increase of
x) o f the disturbances o f given frequency co, set up at a fixed point
of the flow. At the time of the Schubauer-Skramstad experiment, the
theory was developed for space-periodic disturbances only; therefore,
these authors simply transformed the temporal growth rate linearly
into a spatial growth rate by means of the disturbance phase velocity.
However, this procedure is not a rigorous one and if there is any
dispersion it seems more reasonable to transform x from temporal to
spatial growth using the group rather than the phase velocity. This
question was studied later by Gaster (1962; 1965), who showed that
for disturbances in the boundary layer with comparatively small
amplification rates, for which the group and phase velocities are
roughly equal, the Schubauer-Skramstad transformation has rather
high precision, i.e., it can be justified (see also Betchov and Criminale
(1967), Sect. 53). On the other hand, when modem computers are
utilized, the transition from the eigenvalue problem for unknown
complex c (with fixed real fe) to the eigenvalue problem for unknown
complex k (with fixed real co) does not lead to any serious additional
difficulties. Therefore, it is not surprising that numerical stability
calculations for spatially growing disturbances in the Blasius bound
ary layer have been carried out independently during 1964—1966 by
several investigators (all the references and some of the results
obtained can be found in Sect. 53 in Betchov and Criminale).
Figure 16 gives the comparison of
the experimental data by Schubauer — n f- - - - - - - - - - - - -
and Skramstad (1947) [black dots] and IP

by Bums et al. (1959) [crosses] on the
frequency o f the neutral disturbances 300
with the theoretical neutral curve Im
*((0 , Re) = 0 on the (co, Re)-plane, -
obtained with the aid o f the phase-

velocity transformation from the data
of Shen (1954). The agreement be
tween theory and experiment proves to 700
be completely satisfactory. Let us note
in this respect that the neutral curve for
\V
the space-periodic and for the time- — L- 1 1 1 ■ 1 1 1 . — 1— 1_ _ _ 1__
periodic disturbances is always exactly WOO 2000 Ref

the same, because if Imc = 0 , we have FIG. 1 6 . C a lc u la te d fo r m o f th e n e u
Im CO = Imfcc = 0. Figure 17 gives the tral curve in th e (w .R e ^ * ) p lan e fo r
data of Schubauer and Skramstad re- th e b o u n d a r y lay er o n a fla t p la te,
w - X- X ^1. 1. r and e x p e r im e n ta l d ata o n th e fre-
latmg this time to the wave numbers o f q „ e n c y o f th e n e u tra l o s c illa tio n .
the neutral disturbances; these data also
lie fairly well on the theoretical curve. The agreement proves to be
even slightly better when the theoretical curve obtained by digital
computation in 1959—1964 is used instead o f the older curve shown
in Fig. 15 (this new neutral curve can be found in Betchov and
Criminale’s book). The rate of spatial growth Im* = Re)
measured by Schubauer and Skramstad also is in quite good
agreement with the results of the theoretical computations. Further
material on the comparison o f the data with the deductions o f the
theory may be found in surveys of Lin (1955), Schlichting (1959;
1960), Dryden (1959), Stuart (1963), Shen (1964), and Betchov and
Criminale (1967).
F IG . 1 7 . V a lu e s o f th e w a v e n u m b e r s o f neutral d istu rb a n c es in th e
b o u n d a r y layer o n a fla t p la te a c co r d in g t o th e d a ta o f S c h u b a u e r an d
Sk ram stad .
Analogous calculations of the neutral curve and the stability

diagram were also carried out for many other boundary-layer
profiles, corresponding, for example, to flows past curved surfaces or
in the presence o f a favorable or adverse pressure gradient, to flow
over a porous plate when there is suction of the fluid through the
plate, to wall-jet flow, and so on. The results o f these calculations
may be found in Lin (1 9 5 5 ), Shen (1 9 6 4 ), Reid (1 9 6 5 ), Hughes and
Reid (1 9 6 5 ), Betchov and Criminale (1 9 6 7 ), and many other sources.
Results o f the same type for plane-parallel channel flow between
solid walls with nonquadratic velocity profiles may be found, for
example, in Birikh (1 9 6 6 ) and Chen and Sparrow (196 7).
It is interesting that the form o f the neutral curve obtained
depends considerably upon whether or not the velocity profile has a
point o f inflection (i.e., whether becomes zero for some z). In
dz^
the latter case, the neutral curve on the stability diagram will have
the same character as in cases of plane Poiseuille flow or the Blasius
boundary layer (see Figs. 14 and 15), i.e., as R e oo both its
branches are drawn towards the abscissa. Thus, when the velocity
profile has no inflection point, any disturbances of fixed wavelength
(or fixed frequency) with increasing Reynolds number will finally
become stable. However, when the velocity profile does possess an
inflection point, the upper branch of the neutral curve will have a
finite asymptote as Re oo (see Fig. 18) and the ordinate o f this
asymptote depends upon the distance of the inflection point from

the wall. In other words, there exists here a range of wavelengths (or
frequencies) such that the correspond
ing disturbances are unstable however
large the value of the Reynolds num
ber. Obviously, this is connected di
rectly with the definitive role played by
the presence o f an inflection point in
the velocity profile, for the inviscid
instability problem.
So far, we have spoken only of
FIG . 1 8 . S c h e m a tic fo r m o f th e neutral and unstable disturbances in
neutral curve in th e ( * , R e) p lan e boundary-layer flows. It is natural to
for a flo w w ith v e lo c ity p r ofile i ^ n x i. x- j
thmk that all the results mentioned
p o ssessin g an in f le c t io n p o in t ( e .g „
for a b o u n d a r y la yer in th e pres- above relate to the lowest eigenmodc
e n c e o f an a dverse pressure gradi- that there is also an infinite se-
■ quence o f strongly damped higher
modes similar to that found by Grohne
(1954) for the cases of plane Couette and Poiseuille flow. However,
nothing is presently known about these higher eigenmodes.
Information about all the modes is o f some importance in investigat
ing the fate o f a spatially localized disturbance introduced into the
flow at an initial time. In fact, here we must consider the initial value
problem with rather general initial conditions decomposed into an
integral over all the space-periodic eigenmodes o f the form
^ ,fcj,z) where A:= (A:, + Ar|)‘/^ and the index n
refers to the number of the mode. Because all the evidence at hand
indicates that only the first mode can be unstable and that all the
others are damped, it is reasonable to hope that in the asymptotic esti
mates for sufficiently large values of r, only the terms of the expansion
with n = 1 will play an important role. However, the corresponding
approximation is difficult to handle because of our rather poor
knowledge of the eigenfunctions o f the Orr-Sommerfeld equation,
even for the case o f the first mode. Therefore in the concrete
example o f the initial value problem for a three-dimensional localized
disturbance in the Blasius boundary layer considered by Criminale
and Kovasznay (1962) [see also, Betchov and Criminale (1967)]
some rather crude simplifying assumptions were made, and the
results obtained are not too close to the experimental data of
Vasudeva (1967), who tried to check the theory. However, the
qualitative features o f the solution o f Criminale and Kovasznay agree
well with the experimental results. The same is true for the
calculations o f Tam (1967) who used the method of Case (1960a)
and Dikiy (1960b) and showed that strongly localized “pointlike”
disturbances in a slightly supercritical plane flow develop into a
three-dimensional turbulent spot when moving downstream; such
spots were observed by many experimentalists. On the other hand,
lordanskiy and Kulikovskiy (1965) have sketched a very general
asymptotic analysis o f the fate o f a two-dimensional localized [in the
(x,z>plane] disturbance in a two-dimensional flow at high enough
Reynolds numbers. They assumed that the wave numbers tend to
zero on the branches o f the neutral curve lmc(/e,Re ) = 0 and
concluded that in this case the localized disturbance apparently must
decay as < co in any finite region o f the flow (see also related paper
by Kulikovskiy (1966), analyzing the stability of plane Poiseuille
flow in a channel o f finite length).
Stability o f Certain Plane-Parallel Flows in an Unbounded Space
As a concrete example of a flow with a profile having an inflection
point, let us consider a plane-parallel flow in the infinite space
—oo < 2 < oo, which has a velocity profile of the form
t/(z) = ^ o s e c h 2 ^ (2.31)
(see Fig. 19a). According to the calculations of Schlichting and

Bickley (see, for example, Schlichting (1960), or Goldstein (1938),
7 ^,
Un Un
FIG . 1 9 . S c h e m a tic fo r m o f th e v e lo c ity p r o files fo r s o m e plan e-parallel flo w s in an u n

b o u n d e d sp ace: a - p l a n e je t; b - f l o w w ith a ta n g en tia l v e lo c ity d is c o n tin u ity ; c - and d -
m ix in g laye r s o f t w o p lan e f lo w s o f d iffe r e n t v e lo c ity .
Vol. 1, Sect. 57) an equation of this type represents the similarity

solution for the longitudinal velocity profile «i = in a laminar
plane jet (issuing from an infinitely thin linear aperture along the line
a: = 0, z = 0, into a space filled with the same fluid). Here the
parameters (Jo ^nd H will depend on x only comparatively weakly,
while the transverse velocity «3 = w, will be small in comparison with
the longitudinal component. Thus a plane-parallel flow with profile
(2.31) may be considered as a model o f a plane jet at a great distance
from the aperture; in this connection its stability has been studied by
a number o f authors [in particular, by Curie (1956); Tatsumi and
Kakutani (1958); Howard (1959); Klenshow and Elliott (1960); and
Soprunenko (1965)].
As always, determination of the neutral curve in the plane (k, Re)
for a flow with profile (2.31) leads to the investigation o f an
eigenvalue problem for the corresponding Orr-Sommerfeld equation
(2.28). The unboundedness of this flow permits some simphfication
of the calculation in comparison with the case of channel flow
between rigid walls (since in this case it is only necessary to consider
those solutions o f Eq. (2.28) which are damped at infinity).
Moreover, because o f the symmetry o f the profile (2.31), all the
eigenfunctions cp(z) in this case may be divided into even and odd
with respect to z corresponding to antisymmetric and to symmetric
disturbances of velocity. Since both experiment and crude theoreti
cal estimates [cf., for example, Petrov (1937)] show that distur
bances with even <p(z) are always unstable, in the detailed calcula
tions we may confine ourselves only to the regionO < z < oo and the
boundary conditions cp(oo) = (t'(oo) = 0 and (p'(0 ) = (p"'(0 ) = 0 .
The results o f these calculations, carried out by the two different
methods o f Curie (1956) and of Tatsumi and Kakutani (1958) are
shown in Fig. 20. We observe that they are generally in fairly good
agreement (except in the 'egion of small values of k, where Curie’s
method is not sufficiently accurate). These results were also
confirmed by the machine computations of Kaplan, who found
simultaneously the stability curves Imc = C2 (k. Re) = const [see
Betchov and Criminale (1967)]. The numerical results o f Soprun
enko showed some small deviations from the data o f Fig. 20 (rather
minor ones), but they contain much additional information (for
example, Soprunenko also investigated symmetric velocity distur
bances and found that Recr « 90, i.e., it is much greater than Recr
for antisymmetric disturbances, as was expected). According to Fig.
20, as Re ^ o o , all disturbances withife < 2 ///are unstable; moreover.
as Re increases, the region of unstable wave numbers expands

monotonically. Thus, increasing the viscosity can only have a
stabilizing effect on the disturbances (unlike the situation for flows
with a velocity profile without an inflection point). The Recr here
prove to be very small (which is also typical for flows with velocity
profile having an inflection point); according to the careful numerical
calculations o f Klenshow and Elliot, Re,.,. — 3.7 and
kcT ^ 0.25/// (we note that for real jets the value Re ~ 4 is generally
attained in a region o f the flow in which the jet cannot be considered
plane-parallel). The theoretical results for the flow considered were
compared with the experimental data o f Sato and Sakao (1964) on
the development o f artificial disturbances induced in a plane jet by a
loudspeaker both by the experimenters themselves and by Soprun-
enko. The experimental data o f Sato and Sakao are rather scattered,
but in the mean they do not contradict the deductions from
theoretical stability calculations.
FIG . 2 0 . F o r m o f th e n e u tra l curve o n th e (/r, R e) p lan e for

a plan e j e t , a c co r d in g to th e c a lc u la tio n s o f Curie ( 1 9 5 6 ) and
T a ts u m i and K a k u ta n i ( 1 9 5 8 ) .
The stability of the idealized model o f a plane jet with polygonal

velocity profile of the form U (z) = t/o when \z\ < H and U(z) = 0
when \z\ > H, was studied by Yamada (1 9 6 4 ), who used the
corresponding Orr-Sommerfeld equation with natural “matching
conditions” at the comers o f the velocity profile. He found that in
this case, Re^r 4 and kcr ^ 0 .2 5 /// for antisymmetric velocity
disturbances, whereas Rccr *=» 2 0 and kcr ÎH for symmetric

disturbances in reasonable agreement with the results for the jet with
a smooth velocity profile. More interesting is that Yamada disclosed
the existence o f a sequence of eigenmodes of both antisymmetric
and symmetric types; to every mode there corresponds a special
neutral curve, only the first mode leading to the values of Recr and
k cr mentioned above.
Another im portant type o f flow in an unbounded space with a
velocity profile U ( 2 ) having an inflection point is that for which
U{z) Uo as z-*- + 0 0 and 1/ ( 2 )-»■ — Uo as 2 —0 0 . The simplest
flow of this type is the idealized flow with a broken velocity profile
shown in Fig. 19b; this flow describes a plane surface of tangential
velocity discontinuity. More real profiles of the same type are shown
in Figs. 19c and 19d; these correspond quantitatively to the laminar
mixing zone o f two plane-parallel flows, flowing one above the other
with different velocities. As already mentioned when considering the
stability of ideal fluid flows, neglecting viscosity the instabihty o f
flows with the velocity profiles shown in Figs. 19b and 19c was
strictly proved as early as Helmholz, Kelvin, and Rayleigh; the
instability o f the flow o f a nonviscous fluid with a profile o f the
form shown in Fig. 19d has now also been investigated in every detail
[see, for example, Michalke (1964)]. Taking the viscosity into
account, the stability o f a flow with a profile of the type in Fig. 19d
was studied by Lessen (1950); however, Lessen’s results refer only to
the case of sufficiently large Reynolds numbers and actually show
only that here Recr < 20 [in Lin (1955), it was assumed erroneously
from this that in this case Recr is close to 20]. Later, Esch (1957)
investigated the eigenvalue problem for the Orr-Sommerfeld equation
(2.28) numerically, with the f u n c t i o n ( 2 ) shown in Fig. 19c, and for
all values of Re. He was the first to find that the flow with such a
velocity profile is, in fact, unstable for all Re. Still later, Tatsumi and
Gotoh (1960) showed that this result is not connected with the
special form o f the profile studied by Esch, but is related to a wide
range o f profiles, corresponding to a plane mixing zone of two
parallel flows [although Lessen and Ko (1966) studied the stabiUty
of a half-jet flow similar to the plane mixing zone, they obtained
numerically a neutral curve of the same form as shown in Fig. 20 and
a nonzero value of Re^;- of the order of a few u n its]. The general form
of the neutral curve in the (k, Re)-plane for a plane mixing zone,
which is in accordance with the deductions of Tatsumi and Gotoh, is
shown in Fig. 21. This figure presents the results obtained by
Betchov and Szewczyk (1963) by numerical integration of the

Orr-Sommerfeld equation for the hyperbolic-tangent velocity profile
U(z) = t/otanh (z/H) (of the type shown in Fig. 19d). The details of
this integration procedure and many additional results of it may be
found in Betchov and Criminale (1967), Sect. 13.
FIG. 2 1 . Position o f the neutral curve and

the region o f instabihty (shaded) in the
(/j,Re) plane for the mixing layer between
two plane-parallel flows according to the
data o f Betchov and Szewczyk (1 9 6 3 ).
The Stability o f Circular Poiseuille Flow and other Axisymmetric

Flows
The problem of the instability and transition to turbulence of
Poiseuille flow in a circular tube is probably the most intriguing and
interesting stability problem closely related to the classical experi
ments of Osborne Reynolds described at the beginning of this
section. However, this problem is also very difficult and our
knowledge in this field until now has been rather poor. Therefore, we
shall begin our discussion with the general analysis of stability for
axisymmetric flows (with some applications to axisymmetric jets and
wakes) and only at the end shall we pass to the special case of
circular Poiseuille flow.
First, let us consider a general steady axisymmetric flow, i.e.,
parallel flow with the velocity U{y,z) = U{r), r = (y^ + z^)*/^,which is
everywhere parallel to the x-axis and depends only on the distance r
from the line y = z = 0. Applying the small disturbance theory, we
have to use cylindrical coordinates (x,r,tp). Then the coefficients of
the system o f linearized equations will not depend on x and hence
disturbances of the velocity and pressure may be sought here in the
form u' {x, t) = f(r) exp [/(A:x + mp —coOl, p(x, t) = g(r) exp [i{kx +
-co/^)] (cf. Eq. (2.13), Sect. 2.6). Upon substitution of these
expressions into the equations we obtain, as usual, a complicated
eigenvalue problem, related (but of course not identical) to the
problem (2.14 ) - ( 2 . 16).
When considering the stability of plane-parallel flows we assume

first that the Reynolds number is so high that it is possible to
disregard the viscosity, i.e., to consider the fluid as ideal. Let us do
the same for the axisymmetric flow. Then our eigenvalue problem is
simplified considerably and after eliminating the functions f ^ ,
and g (related to the axial and circumferential components of
velocity and pressure) from the system of linearized equations for
the disturbance amplitudes, we obtain the following single equation
for one unknown function /('')(r) =F(r):
d(rF)
W - c) — - (U - c)F - rF ] = 0
dr dr \r^2 +
where as usual c = oj/k is the eigenvalue of the problem. This

equation was given by Rayleigh (1892), who founded the theory of
hydrodynamic stability o f ideal fiuid flows in the axisymmetric case
too [see also the important papers by Batchelor and Gill (1962), and
Schade (1962), on the general analysis of the stability o f axisym
metric flows]. The corresponding boundary conditions state that
F(r) = 0 on the solid (cylindrical) boundaries of the flow and F(r) -*■
0 as r oo for unbounded flow, while F ( 0 ) = 0 if « 1 a n d F ( 0 ) is
bounded if « = 1 in the case o f a flow enveloping the axis /• = 0 .
The Rayleigh equation for F(r), after dividing hy U - c,
multiplying by the complex conjugate of rF, and integrating the
imaginary part o f the new equation, gives
I q(r) \‘^Q'{r)dr = 0 ,
1
where Ca = Im c, q{r) = rF(r)l(U — c), Q(r) = rU'Kn^ + k^r^), and r,

and K2 are the boundaries o f the flow (the cases ri = 0 and/or /-j = <»
are not excluded). Therefore, the necessary condition for the
existence of complex eigenvalues c (i.e., of growing wave-like
disturbances) is that Q'(r) should change sign at some point of the
flow (this condition also was demonstrated first by Rayleigh). It
follows, for example, that in circular Poiseuille flow (with U(r)
— /-^) nonaxisymmetric growing inviscid disturbances are not
possible. The case o f axisymmetric disturbances (with « = 0) in
Poiseuille flow is exceptional in that Q' is then identically zero
(similar to the case of plane-parallel Couette flow where U" is
identically zero). The Rayleigh equation for F(r) in this case has no
eigensolutions at all.
This necessary condition for instability is evidently similar to the
necessity o f the inflection point in the velocity profile for instability
o f plane-parallel inviscid flow. The stronger necessary condition of
Fj^rtoft-Hîland for the instability of plane-parallel flow also has a
simple counterpart in the stability theory o f axisymmetric flows: if
Us is the velocity at the point where Q' changes sign, then a complex
eigenvalue c may exist only if the inequality Q'(JJ — Us) < 0 holds
somewhere in the field of the flow [the corresponding proof is
similar to that for the Fj^rtoft-Hîland condition; see Batchelor and
Gill (1962)]. Rayleigh showed also that Ci = Re c cannot lie beyond
the range of U(r); moreover, Howard’s semicircle theorem which
restricts the range of the possible values c in the complex c-plane for
plane-parallel primary flow, is also vahd (and has exactly the same
formulation) for axisymmetric primary flow [Batchelor and Gill
(1962)]. Finally, the results by Tollmien (1935) showing that the
simplest necessary conditions for inviscid instabiUty of plane-parallel
flows are in many important cases also sufficient, can be transferred
to the axisymmetric flows [see Schade (1962)]. There is Httle doubt
that the results o f Case and Dikiy on the initial value problem for
disturbances in plane-parallel inviscid flow can also be proved for
axisymmetric flow, although this problem has apparently not been
studied by anyone.
If we disregard the possible continuous spectrum of the c-values,
which, in general, is apparently stable (on grounds of similarity with
the plane-parallel case), the investigation of the instability character
istics o f specific axisymmetric inviscid flow is equivalent to finding
the discrete eigenvalues and eigenfunctions of the Rayleigh equation
for F(r). This problem was studied by Batchelor and Gill (1962) for
the idealized jet with top-hat profile (i.e., U(r) = Uq if r < R and U(r) =
0 if r > 7?) and for a more realistic jet profile of the form U(r) ~
(^2 + ^ 2 j - i . latter case, only disturbances with n = 1 appear to
be unstable. Similar results were obtained by Sato and Okada (1966)
for the axisymmetric wake with profile o f the form U(r) = Uq —
exp(-Z>r^); here also unstable disturbances may exist only for « = 1 .
Sato and Okada computed the growth rate o f the unstable
disturbances and compared the results with experimental data on the
evolution o f artificial disturbances induced in the wake o f a slender
and sharp-tailed axisymmetric body; the agreement between theory
and experiment was rather good. Stability calculations for
axisyminetric jet flows were also performed by Michalke and Schade

(1963).
Now let us return to the considerably more difficult problem of
the stability o f axisymmetric flows o f viscous fluid. The effect o f the
viscosity on the stability o f axisymmetric disturbances in the
axisymmetric jet was considered briefly by Gill (1962), but the
overwhelming majority o f work in this field is connected with the
investigation of circular Poiseuille flow. The corresponding eigen
value problem is extremely complicated; therefore all the investi
gators have considered only special cases. Particular attention was
paid to the case o f (/îndependent axisymmetric disturbances (n = 0 ).
This special problem was studied by Sexl (1927a, b), Pretsch
(1941b), Pekeris (1948), Belyakova (1950), Corcos and Sellars
(1959), Sexl and Spielberg (1959), Gill (1965a) and some others. In
all cases only damped eigenmodes were found. A comprehensive
review of all previous results was given by Gill (1965a) who also
computed the spatial damping rate o f some disturbances with the aid
of a numerical solution of the eigenvalue problem with fixed (real)
frequency w and unknown (complex) k, and compared them with
the experimental data of Leite (1959) on the damping o f artificial
periodic nearly axisymmetric disturbances in pipe flow. Since no
growing axisymmetric solution o f the disturbance equations has yet
been found, the majority o f the workers in the field are sure that a
Poiseuille flow in a tube (like a plane Couette flow) is stable for all
Reynolds numbers with respect to infinitesimal axisymmetric distur
bances. However, we must note that the situation with regard to a
rigorous proof o f this fact is even slightly less satisfactory than for a
plane Couette flow.
Until now only a first step has been made in the direction of
solving the general problem o f the behavior of nonaxisymmetric
disturbances (n ^ 0) in circular Poiseuille flow. Until 1968 only very
special nonaxisymmetric disturbances were studied at all [for
example, the x-independent ones with /: = 0 , or such that m^(x, t) =
0; see Kropik (1964), where additional references may be fo u n d ]. All
these disturbances appear to be stable also. Therefore, it had been
customary to think that Poiseuille flow in a circular tube is stable to
any infinitesimal disturbance and that instability here can take place
only when disturbances o f finite amplitude are introduced (in other
words, that circular Poiseuille flow is a self-excited system with hard
excitation). This idea agrees well with the experimental data on the
persistence o f the laminar regime in tubes and we have already
mentioned it in this connection in Sects. 2.1 and 2.4. However, the

experimental findings of Lessen, Fox, Bhat and Liu (1964) and
especially o f Fox, Lessen and Baht (1968), raised some doubts as to
the correctness o f this view o f the problem. These investigators
produced artificially (by means of a small vibrating vane, whose
plane is parallel to the tube axis, with the motion in a direction
perpendicular to that o f the main flow) in laminar circular Poiseuille
flow, the first mode of the azimuthally periodic disturbance (with n =
1). Observing the development of the disturbance as it propagates
downstream, they found both damped and growing disturbances
(and also almost neutral ones). By varying both the Reynolds
number o f the flow and the frequency of vane vibration, they
succeeded in obtaining the experimental neutral curve on the (co,
Re)-plane (which corresponds to the eigenvalue problem with fixed
real co and unknown complex k for a disturbance proportional to
exp [/ (±1/’+ kx —o )/)]). The neutral curve is similar in form to the
curve shown in Fig. 16 (but it is smoother); the corresponding value
o f RecK is about 2130. Therefore, Fox et al. assumed that circular
Poiseuille flow is unstable to infinitesimal nonaxisymmetric distur
bances with n = 1 when Re is greater than about 2130. Taken at face
value, this assumption did not seem to be completely improbable.
This was because nothing was known at that time about the solutions
of the complex eigenvalue problem for the system of coupled
ordinary differential equations which describes the behavior o f the
disturbances with « 0 in circular Poiseuille flow. However, the
careful numerical investigation o f the problem undertaken by Lessen,
Sadler and Liu (1968) for the case n = 1 showed that several
eigenmodes exist for the problem, but all o f them are damped in the
range of wave and Reynolds numbers which Fox, Lessen and Bhat
(1968) found to be unstable. Hence, no linear instability exists in
this range for disturbances with « = 1. In addition. Lessen et al.
investigated limited portions of the (k, Re)-plane corresponding to
much higher Re values (up to Re = 30,000) and also observed no
instabilities, nor any indication o f the likelihood o f them for higher
Reynolds number. Disturbances with n > 1 were not considered by
Lessen et al.; however, they present some qualitative arguments
indicating that such disturbances would be more stable than those
with n = I. Their expectations are in good agreement with the
finding o f Salwen and Grosch (1968), who gave results of a
numerical investigation o f the behavior o f disturbances with n < 5;
they found no instability for Re < 10,000. Therefore, it is now clear
that the instabilities observed by Fox, Lessen and Bhat were not
caused by the linear instability o f infinitesimal disturbances, and the
hypothesis that circular Poiseuille flow is stable to any infinitesimal
disturbance^ now seems again to be highly probable.
The stability problem for rotating axisymmetric flows (with a
circumferential component of velocity V(r) different from zero) is
similar in many respects to the problem for parallel axisymmetric
flows. One such problem was considered in detail in Sect. 2.6; two
other special cases o f this problem were studied by Pedley (1967)
and Michalke and Timme (1967). The stability theory of general
parallel flows [with velocity U(y, z) = (r, <^)] parallel to the x-axis
but dependent on two variables was begun by Hocking (1968).
2.9 Stability to Finite Disturbances; Growth of Disturbances and
Transition to Turbulence
The small disturbance theory discussed above in many cases
permits the conditions of loss of stability of laminar flows to be
determined theoretically. These results allow some im portant facts
connected with the transition to turbulence to be explained.
However, it is quite clear that the loss of stability does not in itself
constitute such a transition, and that the linear theory discussed in
Sects. 2.5 through 2.8 can, at best, describe only the very beginning
o f the process o f initiation o f turbulence, but cannot give a complete
picture o f this process. Moreover, for certain important flows, for
example, Couette flow between two planes or Poiseuille flow in a
tube, the small disturbance theory cannot, in principle, aid our
understanding of the transition to turbulence observed experi
mentally, since these flows are apparently stable with respect to all
infinitely small disturbances. On the other hand, for example, in the
case o f a Poiseuille flow between two plates (close to a real flow in a
plane channel) stability theory, although it implies a possible loss o f
stability, indicates a considerably higher value of R e „ than that at
which transition actually occurs. The data of Davies and White
(1928) show that transition of a plane Poiseuille flow actually takes
place for Re = UH\N ^ 1000, while, according to the linear theory,
Recr ~ 6000. Finally, the cardinal difference between the motion
which arises after loss o f stability in the flow between rotating
cylinders, or in the state o f rest o f a fluid heated from below, on the
one hand, and plane Poiseuille flow of boundary-layer flow, on the
other hand, cannot be explained from the viewpoint of linear
disturbance theory. All these facts demand that great stress be placed
on working out a more complete nonlinear theory o f the initiation of

turbulence; at present, this is only in the initial stage of development.
Energy Balance o f a Finite Disturbance
The simplest approach to the investigation of the stability o f a
flow with respect to disturbances o f finite amplitude is connected
with the use of the “energy m ethod,” originated by O. Reynolds
(1894) [hence it is older than the small disturbance theory]. The
essence o f this method consists in determining the energy balance of
a disturbance in a given flow, to ascertain the conditions under which
the energy o f such a disturbance will increase (or decrease) with
time.
Let Ui { x , t) and P { x , t) be the velocity and pressure of the
undisturbed flow, and u[ {x, /)and p ' {x, t) the velocity and pressure
of the disturbance. In this case, both fields Uf = Ui~{- u\ and must
satisfy the Navier-Stokes equations (1.6) and the continuity equation
(1.5) [as usual, the fluid is assumed incompressible]. Subtracting the
equations for «/ and [/,• from each other, we obtain the following
equations for the disturbance velocity u' = (« ', w'):
du, du, dU, du, 1 dp'
(2.32)
dx„
(these differ from Eqs. (2.7) only by the presence in the first of them
o f a nonlinear term in u', which is omitted in the linear disturbance
theory). Then, multiplying the first of Eqs. (2.32) by mJ/2, summing
over i and integrating the result over the whole region o f flow, we
obtain the energy balance equation, which determines the variation
in time o f the total energy o f the disturbance. In particular, if the
flow takes place in a finite volume, bounded by rigid walls (moving
or nonmoving), on which «i' = 0 , the disturbance energy balance
equation may be written in the form
(2.33)
where dV is an element o f volume [it was actually used in this form

by O. Reynolds (1894) and Orr (1 9 0 6 -1 9 0 7 )]. The same result will

also be true for many unbounded flows; for example, it will be
correct if the disturbance may be considered periodic for all
coordinates with respect to which the flow is unbounded [in this
case, the integration in Eq. (2.33) along such axes must be carried
out over one period o f the disturbance o n ly ].
The first term on the right side o f Eq. (2.33) describes the
exchange of energy between the undisturbed flow and the distur
bance and, as a rule, is positive (the transfer of energy usually takes
place from the undisturbed flow to the disturbance). However, the
second term on the right describes the dissipation of energy due to
the viscosity, and is always negative. The relative value of the two
terms on the right side of Eq. (2.33) will determine whether the
energy o f the disturbance decreases or increases.
Now if we transform Eq. (2.33) to dimensionless quantities,
measuring distance in units of the characteristic length L, velocity in
units o f the characteristic velocity 17, and time in units of LjU, the
dimensional coefficient v in the second term on the right side will be
transformed into the dimensionless coefflcient 1/Re. Hence, it is
clear that if the Reynolds number is sufficiently small, the negative
second term on the right side will dominate the positive first term,
and the energy o f any disturbance will be damped, i.e., the flow will
be stable to disturbances of any magnitude. Proceeding from Eq.
(2.33), we may obtain certain estimates from below o f Rec- min >which
bounds the range of “sufficiently small” Reynolds numbers, within
which the energy o f any disturbance can only decrease. Thus, for
example, Serrin (1959) showed that for any pair o f solenoidal vector
fields u' (jc) and U {x) in an arbitrary bounded region with diameter
D, such that » ' becomes zero on the boundary o f the region, the
inequality
- / " X ^ ( | | ) ’<' 1' +%/ »' ’
3it2
a . /• ,
j u 'U V ,
will hold, where t/max is the maximum o f the modulus o f the field
U(x). Substituting these equations into Eq. (2.33), we find that in
the case o f a flow in a region of diameter D, the energy £ ( / ) o f the
disturbance obviously satisfies the inequality
£ ( / ) < £ ( 0 )exp
such that
Serrin obtained analogous inequalities for the fields u' (x) zndU{x),
corresponding to flows in a straight tube o f arbitrary cross section
with maximum diameter D or in a channel o f maximum width D. All
that is necessary is to replace the constant in the second equation
of Eq. (2.34) in the first case by 2n^ and in the second case by it*;
Serrin likewise showed that Re'cr mm ~ 4.43 for flows in a
straight tube (of arbitrary section) and Recrmin>- n « 3.14 for flows
in a plane channel. Later, Velte (1962) improved the value o f the
constant coefficients in equations of the type of the second equation
(2.34) and showed that Rcor min > » 7 . 7 for flows in a bounded
region o f diameter D, Recr mm > V 4.7 %« 6 . 8 for a flow in a straight
tube of diameter D and Recr mm > « ^ 6 . 0 for a flow in a chan
nel of width D. At the same time, Velte showed that his numerical
values are very close to the largest values which can be obtained with
the aid of general inequalities of the type (2.34) for fairly general
domains o f the type considered. Still later, Sorger (1966) considered
more special flow regions (such as the interior o f a sphere or a circle,
or the domain between two concentric spheres or circles), and for
these regions found the exact values of the coefficients in the
inequalities, thus further improving the stability bounds obtained
previously for wider classes o f stability problems. Sorger’s results
have some points in common with the previous finding by Serrin
(1959) that inequalities o f the same type can be found also for fields
u' (a:) and £/(jf) in the annulus between concentric cylinders of radii
and Rz. Using them, Serrin obtained the following sufficient
condition for stability to arbitrary disturbances of Couette flow
between cylinders:
\ ' < ( R 2 — R\) I n ( 2. 35)
Figure 22 shows the region (2.35) o f the (Qi, i22)-plane for the case
Studied experimentally by Taylor, with Ri = 3.55, Rz= 4.03 together

with the region o f instability (transferred from Fig. 13a) o f the
corresponding flow with respect to
infinitesimal disturbances.
The above-mentioned results of
Serrin, Velte and others, are de
duced in a fairly simple manner,
and are extremely general, but they
give very rough estimates o f Recrmin
lower than the experimental data
by several orders of magnitude (we
recall, for example, that the experi
- // <7 //
mental value o f Recrmin for a
straight tube is close to 2 0 0 0 ). FIG . 2 2 . P o s itio n o f th e in s ta b ility reg io n
Sufficient conditions for the stabil t o in fin ite s im a l d is tu rb a n c es an d t h e sta
b ility r eg io n t o a n y fin ite d is tu r b a n c e fo r
ity o f circular Couette flow, found C o u e tte f lo w b e t w e e n c y lin d er s s tu d ied
by Serrin, later were somewhat b y T a y lo r . T h e u p p er sh a d e d r e g io n c or
strengthened by Sorger (1967) uti r e s p o n d s t o in s ta b ility t o in fin ite s im a l
d istu rb a n c es, w h ile f lo w s co rr e sp o n d in g
lizing more exact inequalities. to p o in ts o f t h e sh a d e d strip are d e fin ite ly
Somewhat better results are ob sta b le t o a n y fin ite d istu rb a n c e. T h e c o n
tained if, instead of using general tin u o u s lin e in th e figu re is th e b o u n d a r y
o f th e r e g io n o f in s ta b ility fo r th e case o f
(but rough) inequalities o f type an in v isc id flu id (c f. F ig. 1 3 a ).
(2.34), we find the disturbances
u '{x) for which the ratio o f the
influx o f energy from the primary flow to the dissipation of energy
due to the action of viscosity, will be the maximum possible. Let us
consider, for example, a plane-parallel flow between the planes 2 = 0
and z = H, with velocity V = U (z), parallel to the ac-axis and for
simplicity, take the disturbance »'also to be two-dimensional: u' =
[(«'(ai:, z, t ) , w'(x, z, ^)] . Measuring distance in units of H , velocity in
units oft/max = maxU ( 2 )and time in units o f ///(/max, we may rewrite
the general energy balance equation (2.33) in the form
dE
dt
^fdxdz, ( 2 .33 ')

Re = i/raax///^'-
By the equation o f continuity [the second equation of Eq. (2.32)]

we may now introduce the stream function ip, putting «' = —dilj/(32,
w' = d'îdx. Then the condition o f instability dEldt ^ 0 will be

written in the form
1
Re < ■' VI rI --------
(^<py d x d z
The maximum of the functional F[^] over the set o f all possible
functions \|j satisfying the given boundary conditions, in this case will
also have the meaning o f 1/Rcrmin (since for Reynolds numbers less
than the reciprocal o f this maximum, the energy o f the disturbance
will certainly decrease; o f course, for two-dimensional disturbances
only). It may be expected that the values of R e c r m m obtained in this
way will be larger than those which follow from the general
inequalities (2.34), which do not take the specific form of the
primary flow into account. It is clear, however, that they may
certainly be considerably lower than the true values of Re«r min, since
using this method we consider the velocity field of the disturbance
(and the derivative dEjdt) only at one moment of time, and do not
take into account that the time evolution of the sum of this field and
the velocity field o f the undisturbed flow must satisfy the system of
fluid dynamic equations. In fact, attem pts to determine the
instability critierion by the energy method, working out values o f
F [if] for certain special functions (leading to a relatively high value
of this functional), or by solving the corresponding variational
problem (usually only for some special set o f disturbances), which
were undertaken by Lorentz (1907), Orr (1906—1907), Hamel
(1911), von Kdrmdn (1924) and others, all led to low values of
R e c r m in (approximately an order o f magnitude less than the values of
Recrmin actually observed). Analyzing this fact, Petrov (1938) came

to the conclusion that the values o f at which takes a
maximum, taking into account the time variations o f all the
functions, will apparently in no case generate a dynamically possible
motion. Thus the energy method can never give an exact value of
RCcr mto, but is only suitable for making preliminary, very rough
estimates o f this value.
It is interesting to note that a similar approach can also be de
veloped for thermal convection problems and here it proves to be
considerably more successful. This fact was demonstrated by
Sorokin (1953; 1954) and Joseph (1965; 1966).
In convection problems the Navier-Stokes equations must evi

dently be replaced by the Boussinesq equations (see Sect. 1.5). The
additional term -g/3T' on the right side of the Boussinesq equation
for the component U3 of the velocity, produces an extra term on the
right side of the equation of the energy balance (2.33) which now
takes the form
The extra term in this equation can easily be estimated by the

inequality
1/2
g/3y*’/^T'dV < g/3 J u f d V . T'^dV < 2g/3(EH)^2
where H = - I T'^dV is the intensity of the temperature distur-

2J
bances. The heat-conduction equation for T and the condition T =
const on the boundaries of the flow (together with the periodicity
conditions for unbounded flows) imply the equation of the balance
of the intensity H:
dV. (2.33"')
Clearly, - [ T u \ — dy < 2y(EH)^^ where y = (VT)^^^ is the

J ^*1
maximal value of the gradient of undisturbed temperature. Moreover,
if T" = 0 on the boundaries of the flow, then the following analog of
the second inequality (2.34) can be proved:
Here a = 3 for a bounded flow region of diameter D and a = 1 for

flow in a channel of width D (in the latter case the disturbances must
be considered as spatially periodic in planes parallel to the channel
walls and all the integrals must be taken over one period only). Using
Eq. (2.34) and all the other above-mentioned inequalities, it is not

difficult to derive the following general deduction from the two
balance equations (2.33") and (2.33"'):
Ve U) + < E(0) + exp[-f(Aa7T^ - VRa)«]

f 7x V rx
where Re = Ra = g/SyD'^/^Y’ - Re^)/2a7T^ is as

sumed to be positive, 8/D^ is the best value of the coefficient before
the integral in the right side of the second inequality (2.34), and
^ = ki//D^ when \^ u /x < 1 whereas ^ when > 1 [cf.
Joseph (1965)]. It follows from this result that the convective
motion will be universally stable to any disturbance of the velocity
and/or temperature if
0 < Ra <
For the special case of a motionless horizontal fluid layer (for

which a = 1, 5 = 3.7 n^, and Re = 0) we obtain the result: Rac r min >
1.85n^^ = 180. The last result can easily be improved; in fact the first
inequality (2.34) is clearly unsatisfactory in the case of motionless
fluid when its left side is equal to zero. If we simply omit the first
term in the right side of Eq. (2.33') and then repeat all the
arguments, we obtain twice as good an estimate: Ra^^ > 360 (this
compares with the value Ra^^, = 1708 obtained from linear theory). A
similar improvement can also be made in the estimate of the
boundary of the universal stability region on the (Ra,Re)-plane if one
uses a different estimate of the first term in the right side of Eq.
(2.33') [giving zero for the fluid at rest] and another definition of
Re-number [see Joseph (1965; 1966)].
The preceding results on the limits of stability of convective flows
are similar to the Serrin-Velte-Sorger results for nonconvective flows:
they do not depend on any specific details of the flow geometry or
the distributions of the primary velocity and temperature field.
Considerably stronger results can be achieved by Using all the
available information on the basic flow. With that end in view,
Joseph proposed first to fix the positive parameter^ = Re/(Ra )^^2
and another positive parameter X and then to seek the greatest value
of Re compatible with the condition —[ y E ( 0 + \g^H(t)] < 0 as a
dt
solution of a specific variational problem for a given primary flow. If
Ra = Ra{\, fi) is the solution of the problem, we can easily find the
largest of the values of Ra(A,/z) over A for a fixed point n [i.e., the
value Ra^ = m a x R a ( A ,f t ) ] ; then the point (R a^,/iV R a^) on the
(Ra,Re)-plane will belong to the optimum boundary of the stability
region.
For the special case o f pure convection in a motionless fluid,
Sorokin (1953;1954) developed the same approach considerably
earlier and produced surprisingly strong results. For simplicity, let us
consider only the case of a horizontal fluid layer of width D bounded
by rigid planes. It is easy to see that the derivative—[yE(t) + Xg^H(t)]
dt
takes the smallest value when A = 1; hence we must consider this value
of A, only. After introducing the new variables u . = =
u'i[(Ti - TQ)u/Dg^x^^'^^ of the dimension of temperature and the
nondimensional coordinates y- = x^/D, the condition
- [ y E ( t ) + g^H(t)] < 0
dt
takes the form
dV > 0 .
Therefore the optimal value of Ra compatible with this inequality

(i.e., the value of Ra^,c r m in
can be found as the solution of the
following variational problem:
- / UgTW = 1/2
= M ax
where the maximum is considered over the class of functions v(y, t),
T'(y,t) satisfying the conditions v = T' = 0 at yg = 0 or 1, div v = 0,
/[?©■ •
Using ordinary techniques o f variational calculus we can reduce this

problem to determining a minimum eigenvalue of a specific eigenvalue

problem for the system of partial differential equations. It is not diffi
cult to verify that this eigenvalue problem agrees with the eigenvalue
problem of linear stability theory corresponding to the case of
neutrally stable disturbances (i.e., to disturbances meeting the
principle o f exchange of stabilities) in a motionless layer of fluid
heated from below. This fact was demonstrated by Sorokin (1953),
who gave simultaneously a very simple proof of the validity of the
principle of exchange of stabilities in thermal convection problems
(and also of the sharper result that all the characteristic frequencies
(i>,- corresponding to the eigenvalue problem of small disturbance
theory in a motionless fluid heated from below are purely imagin
ary); later, the same fact was discovered by Joseph (1965). The
unknown minimum eigenvalue Ra^,^ replaces the parameter Ra of
linear disturbance theory in the boundary-value problem obtained.
Hence the best estimate of which can be achieved with the
aid of the energy method agrees with the minimal value of Ra
compatible with the existence o f infinitesimal neutrally stable
disturbances, i.e., Ra^^ = Ra^^ = 1708 (because Ra^.^. cannot
be greater than Ra^^). We see that the energy method in the case of
pure convection, gives the exact value of Ra^^. simultaneously it
shows that in the framework of Boussinesq’s approximate theory of
convection, a layer of fluid heated from below is stable to arbitrary
periodic disturbances (and not only to infinitesimal ones) at all
Rayleigh numbers below that given by the Unear disturbance theory.
Sorokin and Joseph give more general results for pure convection
problems relating to more general boundary conditions and other
flow regions; in all cases the turn out to coincide with a
value given by linear disturbance theory. In particular, steady
disturbances of any magnitude do not exist at Ra below Ra^^. and,
therefore, the state of rest is the only steady solution of the Boussinesq
equations at subcritical values of Rayleigh number; this special result
was proved by various methods and by Ukhovskiy and Yudovich
(1963), Howard (1963), Sani (1964), and Platzman (1965). Joseph
(1966) also studied in detail the case of a plane Couette flow heated
from below. He found that in this case the variational technique of the
energy method gives the stabiHty region in the form:(Re)2 + Ra < 1708.
It is worth noting that for pure convection (when Re = 0), the
corresponding estimate (Ra^.^, = 1708) is exact, whereas for the
pure dynamical problem (w henRa = 0) this estimate (Re^.^. > 41)
is poor. Other examples of the application of the energy method to
thermal convection problems (corresponding to flows in the presence

of internal heat sources) can be found in Joseph and Shir (1966),
Joseph and Carmi (1966), and Joseph, Goldstein and Graham
(1968).
The Adm issibility o f Linearization in the Investigation o f Hydrodynam ic Stability
In d y n a m ic a l p r o b le m s w e hav e seen th at th e en e rg y m e th o d le ad s to v ery rough

e stim a te s o f R e ^ ^ . T h is m a y b e e x p la in e d b y th e fa c t th a t it m a k es o n ly very in c o m p le te
use of th e d y n a m ic a l e q u a tio n s , sin c e it is b a se d e x c lu siv e ly u p o n e q u a litie s (and
in e q u a h tie s ) relatin g to a single in sta n t. S u b sta n tia lly m o r e p ro m isin g (b u t also m u c h m o r e
d if fic u lt ) is th e ap p ro a c h b ased u p o n tracing th e te m p o r a l d e v e lo p m e n t o f a fin ite
d istu rb a n c e a fter a lo n g tim e w ith th e aid o f th e fu ll s y ste m o f e q u a t io n s o f flu id d y n a m ic s.
T h e first q u e s tio n w h ic h arises in c o n n e c t io n w ith th is ap p ro a c h c o n c e r n s th e
in ter rela tion b e t w e e n th e linear th e o r y , d e scr ib e d in th e p re v io u s s e c tio n s, and th e strict
n on lin e a r th e o r y o f fin ite d istu rb a n c es. R e sta tin g it d if fe r e n t ly , th e q u e s tio n c o n c e r n s th e
e x t e n t to w h ic h th e stab ility (o r in s ta b ility ) o f s o lu tio n s o f th e lin earized e q u a t io n s o f flu id
m e c h a n ic s en ta ils a lso th e sta b ility (or in s ta b ility ) o f th e c o r r e sp o n d in g s o lu tio n s o f the
c o m p le te n o n lin e a r s y ste m o f e q u a tio n s .
B e fo r e a n sw erin g th is q u e s tio n it is n ecessa ry t o d e fin e e x a c t ly w h e n th e s o lu tio n o f th e
n o n lin ea r s y ste m is c a lled stable. S u ch a d e fin itio n h ad already b e e n g iven at th e e n d o f th e
last c e n tu r y b y L y a p u n o v . T o sta te it, first, it is n e c essa ry to c h o o s e so m e so rt o f “ n o r m ” in
fu n c tio n sp a c e, w h ic h a llo w s o n e t o m easu re th e m a g n itu d e o f a fu n c tio n (su c h a n o r m , for
e x a m p le , m a y b e th e m a x im u m o f th e fu n c t io n , o r th e integral o f its square, or, in th e case
o f v e c to r f u n c tio n s u ( x ) , th e spatial average o f th e valu e o f th e integral o f u } { x ) + ,
e tc .) . H aving d o n e th is, a s o lu tio n o f th e n o n lin e a r e q u a t io n s is called stable according to
Lyapunov, if, fr o m th e sm alln e ss o f a d istu rb a n c e o f th is s o lu tio n at th e initial in s ta n t i = 0 ,
it f o llo w s th a t th is d istu rb a n c e w ill b e sm all for all ^ > 0 . S ta te d e x a c t ly , th e s o lu tio n is
c a lled stable if, fo r arbitrarily sm all e > 0 , o n e can fin d a 8 = 8 ( e ) , such th a t for an “ in itial
d is tu r b a n c e ” (i.e ., o f a d if fe r e n c e in in itial c o n d it io n s at t = 0 fr o m th e in tial c o n d it io n s o f
th e s o lu tio n u n d er in v e s tig a tio n ) w h ic h is less th a n 8 a c co r d in g to th e n o r m , th e n o r m o f th e
d iffe r e n c e b e tw e e n th e tw o c o rr e sp o n d in g s o lu tio n s ( “ d is tu r b a n c e ” ) fo r all t > 0 w ill n o t
e x c e e d e.
T h is d e fin itio n d iffe rs fro m th a t w h ic h w e u se d in th e linear th e o r y , w h e r e b a sic a lly w e
had a sp e c tr u m o f eig e n v a lu e s a>j, and for s ta b ility w e req uired th a t <0 fo r all
H o w e v er , usua lly in c a ses w h e r e th e linear th e o r y le ad s to in s ta b ility , L y a p u n o v ’s s ta b ility
c o n d it io n is also n o t satisfie d . A c tu a lly , i f th e in itia l d istu rb a n c e is c h o s e n v ery sm all th e n it
w ill a p p a re n tly b e w e ll d e scr ib e d b y th e lin earized e q u a tio n s . F o r th is reason i f > 0
for at least o n e j , th e in itia l d istu rb a n c e can b e c h o s e n su ch th a t fo r sm all t it g r o w s
p r o p o r tio n a l to e x p [o ;^ i], w here = Im > 0 . Then, as th e d istu rb a n c e becom es
r elatively large, th e linear a p p r o x im a tio n cea se s to a p p ly ; th e in f lu e n c e o f th e n o n lin e a r
term s u su a lly le ad s first, to th e d istu rb a n c e b e g in n in g t o g r o w slo w er th a n w o u ld fo ll o w
from th e lin ear th e o r y , an d th e n e ven cea sin g to g r o w , i.e ., it rem a in s b o u n d e d (w e w ill
e n c o u n t e r e x a m p le s o f su ch b e h a v io r b e lo w ) . H o w e v e r , it is im p o r ta n t to n o t e th a t i f w e
d ec re a se fu rth er th e m a g n itu d e o f th e in itial d istu rb a n c e (k e e p in g th e sam e fo r m ) , w e o b ta in
o n ly a lon ger tim e interval du rin g w h ic h th e linear th e o r y is a su ita b le d e s c r ip tio n o f th e
flo w , th e s u b s e q u e n t fa te o f th e d istu rb a n c e b e in g th e sam e; th u s th e m a x im u m value
ac h ie v e d is n o t less th an in th e first case. T h is m e a n s th a t th e s o lu tio n o f th e n o n lin e a r
e q u a t io n (o r s y s te m ) is n o t stab le ac co r d in g to L y a p u n o v .
A p p a r e n tly , un der very b roa d c o n d it io n s th e reverse im p lic a tio n h o l d s - f r o m th e
s ta b ility o f th e lin ea r iz e d e q u a t io n s it fo llo w s th a t th e s o lu tio n o f th e c o m p le t e n o n lin e a r
s y ste m o f e q u a t io n s is sta b le in th e sen se o f L y a p u n o v . T h e a s su m p tio n th a t lin ea r iz a tio n o f
th e e q u a t io n s o f m o t io n is p o ssib le for sta b ility in v e s tig a tio n s has ju st th is sen se (n a m e ly ,
th at a o n e -t o -o n e rela tio n sh ip e x ists b e tw e e n th e ab se n c e o f e ig e n valu es o f th e lin earized

e q u a tio n s havin g p o sitiv e im aginary parts, and th e stab ility in th e sen se o f L y a p u n o v o f th e
s o lu tio n o f th e c o m p le te n o n lin e a r s y ste m o f e q u a t io n s ). In stab ility th eo r y th is a s su m p tio n
is a lw a y s ta k e n o n faith [s e e , for e x a m p le , L in ( 1 9 5 5 ) , S e c t. 1 . 1 ] , b u t, in general, it is n o t at
all e a sy to p rov e rig o r o u sly . F u rther, le t us n o t e th a t, i f th is a s su m p tio n is true, th en the
linear a p p r o x im a tio n gives a c o m p le te answ er to th e q u e s tio n o f w h e th e r or n o t th e given
flu id f lo w a llo w s s e lf-e x c itin g o s c illa tio n s w ith s o ft e x c ita tio n ; o b v io u sly th e s tu d y o f
o s c illa tio n s w ith hard e x c it a t io n a lw a y s req uires th e use o f th e c o m p le te s y ste m o f n o n lin e a r
e q u a tio n s .
H o w e v er , in several s p e c ific cases, th e p rese n c e o f sta b ility (in th e sen se o f L y a p u n o v ) o f
th e s o lu tio n s o f th e n o n lin e a r e q u a tio n s o f flu id m e c h a n ic s , fo r w h ic h th e c o r r e sp o n d in g
lin earized s y ste m s are s ta b le , m a y b e prov en c o m p a r a tiv e ly sim p ly and e le g a n tly . T h is w a s
su b sta n tia lly clarified b y A r n o l’d (1 9 6 5 a ; 1 9 6 6 a ,b ). L et us e x a m in e th a t s im p le st o f th e
e x a m p le s h e s t u d i e d - p l a n e parallel flo w o f an inviscid and in c o m p re ssib le flu id in a
tw o -d im e n s io n a l c h a n n el b e tw e e n rigid w a l l s - a n d le t us, fo llo w in g A r n o l’d, lim it our
c o n s id e r a tio n to tw o -d im e n s io n a l d istu rb a n c es o f th e flo w . F o r sim p lic ity , w e w ill m easu re
all le n g th s w ith a ch ar a cter istic le n g th L ^ , and all v e lo c itie s w ith a c h a r a cteristic v e lo c ity
U q , so th a t h e n c e fo r th all q u a n titie s m ay b e c o n s id e r e d n o n d im e n s io n a l (w h ic h m e a n s th at
w e m ay ta k e arbitrary n u m erical f u n c tio n s o f th e m , and add to g e th e r an y tw o q u a n titie s ). L et
U(z) = - b e th e n o n d im e n s io n a l p ro file o f v e lo c ity o f th e prim ary f lo w , and

d z
th e n o n d im e n s io n a l d istu rb a n c e v e lo c ity (s o th a t 'I' is th e stream fu n c tio n o f th e b a sic flo w ,
and is th e stream f u n c tio n o f th e d istu rb a n c e; A 'p = and A - -^ th e

dz 2 dx dz
v o r tic ity o f th e prim ary f lo w an d th e v o r tic ity o f th e d istu rb a n c e). T h e e q u a tio n s o f m o t io n
here r e d u c e t o a single e q u a t io n fo r th e c o n s e r v a tio n o f v o r tic ity + \p'):
^ = 0. ( 2 .3 7 )
dt dz dx dx dz
We c o n s id e r d istu rb a n c es w h ic h are p e r io d ic in th e c o o r d in a te x and in d o in g so , w e d o n o t

lo s e g e n e ra lity b e c a u s e th e resu lt w ill n o t d e p e n d o n th e le n g th o f th e p e r io d , w h ic h m a y b e
tak en to b e arbitrarily large. F r o m E q. ( 2 . 3 7 ) it f o llo w s im m e d ia te ly th at th e to ta l en ergy o f
th e flo w E = ^ (w h e r e th e integral in dx m a y c o n v e n ie n tly b e ta k en over a
single p e r io d ) is c o n ser v e d in tim e (le t us recall th a t w e are con sid e rin g th e case o f z ero
v is c o s ity ) . It is k n o w n also th a t in th e f lo w o f an id eal flu id th e v o r tic ity A v//is c o n ser v e d ,
and h e n c e an arbitrary fu n c tio n o f th e v o r tic ity c o n s e q u e n tly , th e q u a n tity
rr dj^
J^= J j ^ { A \ p ) d x d z also d o e s n o t c h an ge in tim e [th e e q u a lity = 0 m a y easily b e
o b ta in e d fr o m Eq. ( 2 . 3 7 ) ] .
d<if
^
N o w le t us assu m e th a t th e v e lo c it y p r o file u i z ) = - has n o in f le c t io n p o in ts . T h e n
j3^ cfi’
U iz ) = - 9^ 0 for all 2 , so th a t A4' = is a m o n o t o n ic fu n c tio n o f z and A'I'm a y
b e ta k en in ste a d o f 2 as a n e w transverse c o o r d in a te . In particular, it f o llo w s fr o m th is th at
th e stream fu n c tio n = ( 2 ) m a y a lso b e c o n s id e r e d a fu n c tio n o f A 'l';
'I' = A(A>I'). (2 .3 8 )
In discussing now the invariant integral G —E + we shall show that the function of a
single variable (s)m ay be chosen in such a form that the functionals of the stream
function z) [satisfying the boundary condition <// = const on the rigid walls of the
channel] takes on its minimum value for z) [i.e., for the plane-parallel steady
flow under investigation). This minimum value is understood here to be of a local
character-minimum relative to nearby flows differing from the basic flow by small
disturbances. However, from this it follows immediately that if the initial disturbance
2 , 0 ) is small, then in the function space o f the fields z) the “level lines” of
G[v/;] = G [ ^ { z) + \p'(x, 2 , 0 )] topologically will present the appearance of small ellip
soids surrounding the extremal point \ p = ^ l z ) . Since G[i//] is an invariant functional,
yp{x,z,t) = ^ { z ) + \ p ' i x , z , t ) for all t will belong to the given “level line,” i.e., the
disturbance 2 , t) will remain small for all time. To make the foregoing assertion
rigorous, we must still define how the magnitude o f the disturbance is to be measured, i.e.,
we must introduce a norm into the function space of fields 2 ). For such a norm it is
convenient to take the value of the functional
If' [ u ^ + ( V X u')^]dxdz
(which is well defined, since we have nondimensionalized both and Ax'). In this case
it is easy to show, with the help of equations for the first and second variations of G [i//]
which will be introduced below, that there exist a function ^ a system of coordinates x, 2 ,
and constants K ^ > 0 , K 2 >K jand > 0 such that
< g [ 4 ' + 1/ / I - g C * ] <
if H[ 4 i'] < for chosen <t> and coordinate system. Moreover, C[i//'] does not change with
time, and //[i/z ] changes continuously; therefore, it follows immediately that if
H[\p'] < Kê/K^ for t = 0 and e < then also H[4’'] < e for all ( > 0, i.e., that the
solution xpix, 2 , t) of the nonlinear equation (2.37) is stable in the sense of Lyapunov.
Let us proceed now to the construction of the functional
= / / [ ^ {A\ p) \ dx dz
= //»
which possesses the properties referred to above. First, let us find the variation of this
functional at the point \ p - ^ { z ) , i.e., the leading term of the difference
G [^ + - G [ ^ ] . With the help of the usual rules of variational calculus, we obtain
+ '(A^)A^']dxdz '(A ^) ~ ^ ]A i//'d x d z
where is the derivative of the function <p (while i/;' designates the disturbance stream
function). By virtue of Eq. (2.38) 8G = 0, if <d'(s) = A ( s ) , i.e., i// = ^ ( 2) will then be the
extremal point of the functional G[i//]. Analogously the second variation of the functional
G[i//] with d>=A may be calculated:
jy [
[A ^ '^ A \p 'd x d z =
d xd z
w h e r e A ' is ) = d A / d s an d E q. ( 2 . 3 8 ) is u sed fo r th e c a lc u la tio n o f a ' ( A ^ ) . H o w e v e r , th e

a s su m p tio n o f th e a b s e n c e o f in f le c t io n p o in ts o f th e v e lo c ity p r o file im p lie s th a t U'î z) has
th e sam e sign for all z. S in ce th e e q u a t io n s o f flu id m e c h a n ic s are G a lilea n invarian t, w e
m ay a lw a y s tr an sform t o a n e w s y ste m o f c o o r d in a te s , m o v in g w ith a c o n s ta n t v e lo c ity
relative to th e o ld , so th a t U ( 2 ) h a s o n ly o n e sign fo r all z , n a m e ly , th e sa m e as U " { z ). In
su ch a case, U / U " > 0 , i.e ., B^G[\p] > 0 fo r »// = ^ { z ) . . T h is m e a n s th a t th e p o in t \p = ^
really c o r r e sp o n d s to a lo c a l m in im u m o f th e fu n c tio n a l G [ \ p ] , w h ic h w a s to b e p r o v e n . T h e
in e q u a litie s r elatin g G [ ^ + i//'] - G [ ^ ] and fo ll o w at o n c e fr o m th e fa c t th a t
V / U " is b o u n d e d fr o m a b o v e a n d fr o m b e lo w .
We a lready k n o w th a t R a y le ig h h a d s h o w n th a t plan e-parallel flo w o f an in viscid flu id in
th e a b s e n c e o f in f le c t io n p o in ts in th e v e lo c ity p r o file , is stable in th e sen se th a t n o
eig e n v a lu e s o) o f th e lin earized s y ste m e x is t ha vin g p o s itiv e im a ginary parts. N o w w e see th a t
in th is case it is p o s sib le a lso to p r o v e rig o r o u sly th a t i f w e c h o o s e th e in itia l d istu rb a n c e
s u ffic ie n t ly sm all, th e n w ith in th e fu ll n o n lin e a r th e o r y it w ill b e sm all for all i > 0, i.e ., th a t
h ere s o ft e x c i t a t io n o f fin ite o s c illa tio n s is n o t p o s sib le (a t a n y e v e n t, w ith in o u r lim ita tio n
to tw o -d im e n s io n a l d is tu rb a n c es ). In A r n o ld ’s w o r k s it w a s a lso sh o w n th a t th e a s su m p tio n
o f th e a b s e n c e o f in f le c t io n p o in ts o f U i z ) m a y b e r ep laced b y several o th e r w ea k e r
c o n d it io n s ; w ith th e s e , a n a lo g o u s a r gu m en ts p e r m it th e p r o o f o f s ta b ility in th e sen se o f
L y a p u n o v o f m a n y plan e-parallel flo w s w ith p r o file s U ( 2 ), h avin g in f le c t io n p o in ts , fo r
w h ic h s ta b ility o f th e lin earized s y ste m o f e q u a tio n s w a s s h o w n b y T o llm ie n ( 1 9 3 5 ) [ o f
c o u r se , also o n ly w ith in th e lim ita tio n to tw o -d im e n s io n a l d is t u r b a n c e s ] . F u rth e r, th e sam e
m e th o d s are s h o w n to b e ap p lic a b le for th e p r o o f o f s ta b ility ( w it h o u t th e u se o f th e
lin ear iz e d e q u a t io n s ) o f m a n y tw o -d im e n s io n a l flo w s o f in viscid in c o m p r e s sib le flu id s,
hav in g curvilinear stream -lines. A n a n a lo g o u s p r o o f o f s ta b ility in th e sen se o f L y a p u n o v is
s h o w n t o b e a p p lic a b le to z o n a l flo w s o f an id e a l (or e v en v is c o u s) flu id o n a sphere [c f.
D ik iy (1 9 6 5 b )]. H o w e v e r , fo r th r e e -d im e n sio n a l d istu rb a n c es o f th e f lo w (an d t h r e e -
d im e n sio n a l f lo w s ) , th e r ea so n in g p r e se n te d here is s h o w n to b e in s u f fic ie n t. It p e r m its o n ly
a f e w partial resu lts to b e o b t a in e d an d d o e s n o t reso lv e fin a lly th e q u e s tio n o f th e s ta b ility
in th e s en se o f L y a p u n o v o f th e s o lu tio n s o f th e n o n lin e a r e q u a t io n s o f flu id m e c h a n ic s [c f.
A r n o ld ( 1 9 6 5 b ) ; D ik iy ( 1 9 6 5 a ) ] .
Y u d o v ic h ( 1 9 6 5 a ) has o u t lin e d q u ite a d if fe r e n t ap p ro a c h to th e p r o o f o f th e general
th e o r e m o f a d m iss ib ility o f lin ea r iz a tio n for th e in v e s tig a tio n o f s ta b ility o f sta tio n a ry
s o lu tio n s o f th e e q u a t io n s o f m o t io n o f a v is c o u s in c o m p r e s sib le flu id (i.e ., t o th e p r o o f o f
th e a s se r tio n o f a o n e -t o -o n e r e la tio n b e t w e e n th e e x is t e n c e o f e ig e n v a lu e s o f th e
O rr-S o m m er fe ld e q u a tio n ly in g in th e u p p e r h a lf-p la n e o f th e c o m p le x o;-plane, and th e
stab ility in th e sen se o f L y a p u n o v o f th e s o lu tio n s o f th e n o n lin e a r s y s te m o f N a v i e r - S t o k e s
e q u a t io n s ). H o w e v e r , in th is w o r k , c o n s id e r a b ly m o r e c o m p lic a t e d n o r m s in v e lo c ity -fie ld
sp ace are in t r o d u c e d a nd s o p h is tic a te d m a th e m a tic a l te c h n iq u e s are u se d ; c o n s e q u e n t ly , b o t h
th e s ta te m e n t o f th e r esu lts a n d th eir p r o o f see m t o b e rather aw k w a r d an d c o m p a r a tiv e ly
u n e n lig h te n in g .
Landau s Theory
The investigation of the admissibility of linearization in the studies
of hydrodynamic stability is significant for rigorous justification of
the highly developed linear stability theory; however, it is not the
primary function o f the nonlinear stability theory. Considerably

more attractive o f course is the hope of obtaining new fundamental
physical results on the behavior of finite disturbances in various fluid
flows, with the aid o f the complete system of nonlinear dynamical
equations. It is o f interest both in the investigation of disturbances in
a flow with Reynolds numbers^ * less than the value R e c r = Retr m a x
defined from the linear stability theory (to determine the critical
Reynolds number for finite disturbances of a given amplitude) and in
the investigation o f disturbances with Re > Recr (to study the
further evolution o f weak disturbances, which increase exponentially
according to the linear theory). At present, however, there exist only
a few isolated results for both these cases and almost all o f them
relate only to Reynolds numbers only slightly different from
Recr [ for a more exact estimate o f the range of permissible values of
Re, see, e.g., Stuart (I9 6 0 )].
The most general results on the behavior of finite disturbances
with R e close to Recr which are independent even of the actual form
o f the equations of fluid dynamics, were demonstrated by Landau
(1944) [see also, Landau and Lifshitz (1963), Sect. 27]. Let us
assume that R e > Recr, but that R e — Recr is small. Since with R e =
Recr there first appears a disturbance with “ frequency” w which has
a zero imaginary part, with small positive R e — Rccr there will exist
an infinitesimal disturbance with velocity field of the form
«(jf, t) = A ( t ) f{ x ) , (2.39)
w here/1(0 = ^ _ im a > 0 and as Re ->Recr (so

that Y ^ with sufficiently small Re — R ecr) and fix ) is the
eigenfunction o f the corresponding eigenvalue problem. Therefore, it
is clear that A {t) satisfies the equation
= (2.40)
However, Eq. (2.40) is correct only within the framework of the

linear disturbance theory. A s^ (r) increases, there will inevitably come
an instant when the theory is no longer valid and must be replaced
by a more complete one which takes into account terms in the
dynamical equations that are nonlinear in the disturbances. Then the
21 F o r s im p lic ity , w e shall sp eak a b o u t R e y n o ld s n u m b e r o n ly , a lth o u g h in s o m e cases th e

in itia tio n o f in s ta b ility w ill b e d e te r m in e d b y tr a n sitio n th rou gh a c ritical value o f s o m e
o th e r n o n d im e n s io n a l param eter o f th e sam e ty p e .
right side of Eq. (2.40) may be considered as the first term of the
expansion of in series in powers of A and A * (where the
asterisk denotes the complex conjugate). In the subsequent approxi
mation (which applies for larger t) it is also necessary to take into
account terms o f the next order of the series—the third-order terms;
however, it must be considered that the motion (2.39) is accompan
ied by rapid (in comparison with the characteristic time 1/y of
increase o f the amplitude) periodic oscillations, described by the
factor in the expression for A (t). These periodic oscillations do
not interest us; hence to exclude them, it is convenient to average the
expression d\A\‘^ldt over a period of time that is large in comparison
with 2k/(oi (but small in comparison with 1/y). Since third-order terms
in A and A * will evidently all contain a periodic factor, they will
disappear during the averaging. For the fourth-order terms, after
averaging there will remain only a term proportional to Thus,
retaining terms of not higher than the fourth-order, we will have an
equation o f the form
d\A\
= 2-rM P — 8 |yi|'' (2.41)
dt
(since the period of averaging is much less than l/-f the terms |/1P and
|i4|^ practically will not change with averaging so that Eq. (2 .4 1 ) may
be considered an exact equation for the amplitude of the averaged
disturbance). The sign of the coefficient 5 cannot be ignored;
generally speaking, it must be expected that it can be either positive
or negative (and can also be zero, but only in exceptional cases).
The general solution o f Eq. (2 .4 1 ) may be written in the form
| ^ ( / ) | 2 _ ------(2 .4 2 )
where C is an undetermined constant of integration. From Eq. (2 .4 2 )

it follows that if 6 > 0 and | A (0 ) p = cs/ 2Y sufficiently small,
the amplitude A (t) will first increase exponentially (in accordance
with the linear theory), but then the rate of the increase slows, and
as t^ o o the amplitude will tend to a finite value/I (oo) =
independent of A (0). We now note that 7 is a function of the
Reynolds number which becomes zero at Re = Recr and may be
expanded as a series in powers of Re — Ren- (the latter fact may be

deduced from the small disturbance theory) while 3 /= 0 for Re =
R e or. T h u s f ~ ( R e — R ecr) an d, c o n s e q u e n t l y , / 4 ( o o ) = |- 4 |m a x ~ ( R e
—Recr)'^’ for small Re — RCcr-
However, if 6 < 0, the solution (2.42) o f Eq. (2.41) will formally
become infinite for t = [In (2y/C161)]/2y. Consequently, consider
ably earlier it will have attained such large values that henceforth one
cannot use Eq. (2.41), which is obtained by keeping only the first
two terms in the expansion o{d\A\'^ldtm powers of \A\\ The ultimate
amplitude /l(oo) will be determined in this case by the coefficients
of higher powers of 1^4P which are different from zero at Re =R ecr-
Therefore /4 (oo) will be finite and roughly constant however small
Re — Recr. In this case Eq. (2.41) is evidently inapplicable for
analysis of the processes at Re > Re,;,., but it may be applied to the
investigation of the behavior of finite disturbances for Re < Re^.^. In
fact, for Re < Recr, the coefficient will be negative, i.e., small
disturbances of the form (2.39) will be damped. Since the second
term on the right side of Eq. (2.41) [equal to —81^41''] will be
positive for 6 < 0, then, for sufficiently large \A\\ the derivative
d\A\^ldi (averaged over the specifically chosen period of time) may
become positive, i.e., the motion will become unstable even for R e <
Recr with respect to finite disturbances. For amplitudes that are not
too large, for which Eq. (2.41) may be used, the amplitude |/11 will
increase if \A\ > (2 |'j|/|6 |)'/2. Considering that |-f| ~ (Recr — Re), we
find with respect to disturbances of given amplitude \A\ that the flow
will be unstable for Re > ReA cr = Recr — a|^ P , where a > 0.
Therefore, in this case ReorminWill be less than the value of Recr
given by the linear theory. An exact evaluation of Rccr mm, however,
cannot be carried out on the basis of the approximate equation
(2.41), which is applicable only for small |/4|, since such an evaluation
requires R ca c r t o be defined for arbitrarily large values of I/ll.
Now let us return to the case of 6 > 0, Re > Recr- Here the
increase o f the disturbances (2.39) for Re slightly greater than Recr
may be described as a soft self-excitation of an elementary oscillator,
leading, finally, to the establishment of steady periodic oscillations
with a small (but finite) amplitude proportional to (Re-Recr)’'^^ An
essential feature here is that Eq. (2.41) defines only the amplitude of
these oscillations; the phase, however, is not defined uniquely by the
external conditions, but depends on the random initial phase of the
disturbances, i.e., it may in fact be arbitrary. Thus the ultimate
regime o f steady oscillations of such an oscillator is characterized by
the presence o f one degree of freedom (unlike the case of steady

laminar flow which is defined uniquely by the boundary conditions,
and hence does not possess any degrees o f freedom at all).
With further increase of Re, this ultimate periodic motion may
itself become unstable to small disturbances (x, t). The instability
of a flow with velocity field £ / ( j f ) + a, ( at, where »,is the ultimate
value of the disturbance (2.39) which depends also on Re — R ec rl
may in principle be investigated with the aid o f the ordinary method
of disturbances. For this purpose, it is only necessary to investigate
the particular solutions o f a linear equation in the disturbance (x,i)
o f the form » 2= ex p ( —/co/)/(jc, <)[ w h e re /is a periodic function of
t with period 2 u/(oi| and to determine the frequency (o= co2, with
which, as Re increases, there will first appear (for Re = Rejcr) a
positive imaginary part. Then we may expect that with Re somewhat
greater than Rez or, oscillations with this frequency will increase to
some finite limit. Therefore, as /->- oo quasi-periodic oscillations occur
with two periods 2 it/coi and 2 it/(02 now possessing two degrees of
freedom (the phases of the oscillations). With further increase of Re,
a series of new oscillators will subsequently be excited (i.e., lead to
an oscillatory regime). It is natural to think that the intervals
between corresponding “ critical” Reynolds numbers will decrease
continuously, and that oscillations which arise will be o f higher and
higher frequency and smaller scale. Consequently, for sufficiently
large Re, the motion will have very many degrees of freedom and be
very complex and disordered. Such motion corresponds to the “limit
cycle” o f phase trajectories in which certain generalized coordinates
of the flow assume fixed values, and only the coordinates corre
sponding to the phases o f the corresponding oscillators will vary with
time (according to equations of the form <p(t) = (o(+ a).
A trajectory describing a “limit cycle” occupies a region in the
phase space which corresponds to all possible sets o f initial phases of
the oscillations of the oscillators, and in the course of time, passes
through practically all points of this region. In fact, at instants of
time tn = 2 nnl(0 i,n = 0 , 1 , 2 , . . . , at which the phase<pi (/) =(Oi/+ai
takes the value ai, the phase <P2 (0 o f any other oscillation will take
values 2 urt(02/(0 i + a 2, n = 0, 1 , 2 , . . . . Since individual frequencies coi
and ®2 will, generally speaking, be incommensurable (excluding very
special cases which are most unhkely), the latter set will contain
values, which after reduction to the interval [ 0 , 27t ], will be as close as
desired to any preassigned number in this interval. Thus it follows
that the developed turbulent motion which arises in this case will
possess a definite “ ergodicity,” revealed in the fact that in time, the
fluid will pass through states as close as desired to any possible state
o f motion.
These general considerations are the essence of Landau’s theory of
the initiation of turbulence. They are very obvious and physically
convincing, but not rigorous and cannot be considered as complete.
In fact, Eq. (2.41) is based on the assumption that only one unstable
disturbance will be excited at small positive values o f Re — Recr
although many such disturbances will often exist for Re > Recr and
the interactions between them may be of considerable importance
(see, e.g., Eckhaus (1965), where a rather full discussion o f Landau’s
approach to nonlinear instability can be found). There are also
data showing that the exceptional cases cannot be removed from the
analysis of hydrodynamic instability when at some critical value o f
the parameters o f the flow, not only = 0 but also 6 = 0 (thus the
term of the equation for |/iP o f the order |/4|® plays an important part;
see, e.g., Ponomarenko (1965). However, the most important defect
of Landau’s theory is that, so far, it has not been verified by direct
calculations in a single concrete case, and that the mechanism of
transition to turbulence which it describes is certainly not universal.
Thus, for example, in the cases of plane Couette flow or circular
Poiseuille flow in the absence of nonaxisymmetric disturbances the
Reynolds number Recr = R^cr max defined by linear stability theory
will apparently equal infinity (i.e., Eq. (2.40) will be meaningless).
Consequently, turbulent motion must occur as a result of instability
with respect to finite disturbances, while probably from the very
beginning, it possesses a very large number of degrees of freedom.^ ^
In the case of boundary-layer transition to turbulence, it is
possible that an important part is played by the fact that the
^ ^ H o w e v e r , w e n o t e th a t T a ts u m i ( 1 9 5 2 ) m a d e an a tt e m p t t o e x p la in th e in itia tio n o f
tu r b u le n t f lo w in a circular tu b e o n th e b a sis o f th e lin ear th e o r y o f a x is y m m e tr ic
d istu rb a n c es, a p p lie d to th e in ta k e r eg io n o f f lo w w h e r e a p a r ab olic H a ge n -P o iseu ille p r o file
has n o t y e t b e e n fo r m e d su c c e s sfu lly . O n th e o th e r h a n d , GiU ( 1 9 6 5 b ) fo u n d th a t o n ly a
sm all c h an ge in th e linear v e lo c ity p r o file o f a p la n e C o u e tte f lo w or in th e p ara b o lic p r o file
o f circular P o ise u ille f lo w is req uired t o c h a n g e th e m fr o m stab le to u n sta b le w ith r e s p e c t to
in f in ite s im a l d istu rb a n c es. T h e r e fo r e h e p r o p o s e d th e “ p artially n o n lin e a r ” m e c h a n is m fo r
in s ta b ility o f th e s e flo w s , b a s e d o n th e a s s u m p tio n th a t sm all (b u t n o t in fin ite s im a l)
d istu rb a n c es (a x is y m m e t r ic in th e case o f circular P o ise u ille f lo w ) p r o d u c e a sm all d is to r tio n
o f th e prim ary v e lo c ity p r o file w h ic h m a k es p o s sib le th e s u b s e q u e n t d e v e lo p m e n t o f
g r o w in g d istu rb a n c es in th e fr a m e w o r k o f th e o rd in ary hn ear s ta b ility th e o r y .
unstable disturbances which arise at some value of the Reynolds

number will be carried downstream into a region with higher
Reynolds number. In any case, the existing data on transition in a
boundary layer (which we shall discuss below also will not fit into
the framework of the above considerations. The described mechan
ism of initiation of turbulence recalls far more the generation of a
turbulent wake in the flow past finite rigid bodies (see above. Sect.
2.2). Apparently, the data referring to this process may be explained
on the basis of Landau’s theory if we assume that for flow past a
body, 6 < 0 and the critical Reynolds numbers for instabilities of
different orders are close to each other. Unfortunately, the calcula
tion o f the critical Reynolds number for flow past a finite body, even
within the framework o f the linear theory, is an extremely complex
problem which does not yield an exact mathematical solution.
Therefore, quantitative comparison o f Landau’s theory with the
empirical data on transition in a wake is still not possible.
The cases o f circular Couette flow, free convection between
parallel plates, and plane Poiseuille flow have been studied far more
completely. In these cases, the linear stability theory can yield quite
definitive results (see above, Sects. 2.6—2.8). Below, we shall discuss
in greater detail the question of the application of Landau’s theory
(or, more precisely, Landau’s equation (2.41) and some of its
generahzations) to these three types of flow.
Nonlinear Instability Effects in a Circular Couette Flow
and a Layer o f Fluid Heated from Below
In either of the above-mentioned two cases the motion arising for
Re = Recr (or Ra = Racr) is nonperiodic but steady (i.e., cjj = 0 ) .
Nevertheless, the considerations developed above are also completely
applicable to these cases since the unstable motion proves here to be
periodic with respect to certain space coordinates (the 2 -coordinate
in the first case, and the x- and y-coordinates in the second);
therefore, instead of time-averaging, we may average over these
coordinates. The second, more im portant complication arises because
the unstable disturbances refer here to a continuous spectrum (they
depend on the continuously varying wave number k)\ strictly
speaking, for Re > Re cr (or Ra > Racr) a continuous set of distinct
unstable disturbances will always exist. Nevertheless, this also finally
proves not to be a m atter of principle, since the data clearly show
that for small Re — Recr (or Ra — R a c r ) there always “survives” (and
attains a finite value) only a single disturbance with sharply defined
wave number. We shall discuss the possible cause of this later; mean
while we shall take it as an axiom that in both cases it is permissible to
confine ourselves to investigation of individual disturbances of the
form (2.41) with fixed wave number k and fixed value (o = t'T.
defined by the linear theory.
All the data on flows between concentric cylinders, and on layers
of fluid heated from below, show that at subcritical values of a
characteristic nondimensional parameter (i.e., for Re < R ec r or,
respectively, R a < Racr) there exists no steady motion different from
the laminar Couette flow or, respectively, the state of rest. In
discussing the applications o f the energy method to thermal
convection problems, it was mentioned that for a layer o f fluid
heated from below, this result was strictly proved by several
investigators [namely, Sorokin (1954); Ukhovskiy and Yudovich
(1963); Howard (1963); Sani (1964); Platzman (1965); and Joseph
(1965)] with the aid of the existence conditions for the solutions of
the complete system of nonlinear dynamical equations. The corre
sponding result for fluid in the annulus between rotating cylinders
apparently has not been proved rigorously until now, even for the
simplest axisymmetric flows (although some related theoretical
deductions can be found in the works cited below); nevertheless,
there is no doubt that this result is also valid for the flow in the
annulus. At the same time, for Re = Recr (or Ra = Racr) in both
cases “branching” o f the steady solutions of the corresponding
nonlinear equations will apparently occur. At this point, additional
steady solutions will arise, differing from the ordinary solution by
the presence of periodic terms in z (or in the x- and ^-coordinates),
the ampUtude o f which for small values o f Re — Recr (or Ra — Racr)
is proportional to (Re —Recr)'^’ [or (Ra — Racr)'''*]. Therefore, for
slightly supercritical Reynolds numbers (or Rayleigh numbers) two
different branches o f steady solutions exist which coincide for the
critical Reynolds (Rayleigh) number. By slowly increasing the
nondimensional parameter above the critical value, we can reach a
second critical value, at which a new branching of the steady
solutions will occur. The laminar-turbulent transition in both cases
may be understood as a process of repeated branching of steady
solutions of nonlinear equations in full accordance with the general
scheme proposed by Landau.
These statements are very plausible from the viewpoint of existing
experimental evidence concerning a circular Couette flow and a layer
of fluid heated from below (see, e.g., the data of Malkus (1954a),
and of Willis and Deardorff (1967), on the discrete transitions in a

layer of fluid, and the data of Coles (1965), and of Schwarz,
Springett and Donnelly (1964), on discrete transitions in circular
Couette flow), although in Couette flow transitions to nonsteady
periodic regimes also appear to be important. Their strict proof must
be the first step in the rigorous mathematical theory of transition
phenomena; later, it would be necessary to explain also why one
particular solution is realized physically and another possible
solution is not, and why only one particular value o f wave number k
is observed in the experiments for slightly supercritical non-
dimensional parameter values although a continuum of /^-values is
mathematically possible. However, even this first step is not an easy
one and requires the very heavy machinery of the advanced
topological and functional-analytical methods.
Sorokin (1954) was the first to formulate the general result
regarding the branching o f steady solutions of the nonlinear
Boussinesq equations at Ra = Racr; he demonstrated its plausibility
by the use of a formal expansion technique. Later, the same formal
technique was used by Sorokin (1961) to analyze the branching
phenomena in a bounded flow region in which the fluid is driven by
moving walls [the mathematically rigorous analysis of these phenom
ena was given considerally later; see Yudovich (1967)]. Related
phenomena were also demonstrated by Brushlinskaya (1965) for
some specific simplified “ model” fluid dynamical problems allowing
formulation in terms o f a finite system of ordinary differential
equations; in her proof she used substantially finite-dimensional
methods. The general topological theory of the bifurcations of the
solutions o f operator equations in Banach spaces was developed by
Krasnosel’skiy (see, for example, Krasnosel’skiy, 1964) and was first
applied to fluid dynamical problems by Velte (1964) and Yudovich
(1965b). Later, the same techniques was used by Velte (1966) to
show that in circular Couette flow with S2^='0 (i.e., in the case of a
stationary outer cylinder) an additional branch of steady solutions
forming Taylor vortices occurs at the critical Reynolds number
Re = Re(A:) for any choice of the wave number k; simultaneously
even more general results (relating to an arbitrary value of ii = ^ 2 !
1 > 0 and also to higher-order critical Reynolds numbers) were
demonstrated similarly by Yudovich (1966a) and Ivanilov and
Yalovlev (1966). A rigorous proof that for circular Couette flow the
branching solution has an expansion in powers o f (Re - Rec/-)*''^ was
first given by Kirchg^sner [see Gorfler and Velte (1967)] who used
the general analytical method of solving nonlinear integral equations
developed by Lyapunov and Schmidt at the beginning of this

century.
Convection in a fluid layer heated from below (and o f some other
convection problems) was investigated in detail by Yudovich (1966b;
1967a,b) with the aid o f a combination of the topological method of
Krasnosel’skiy and the analytical method o f Lyapunov-Schmidt. He
showed that when Ra slowly increases and goes through the critical
value Racr two new steady solutions of a given periodicity in the
(x.y)-plane occur, both having an expansion in powers of (Ra —
Racr)'^'- Moreover, the equilibrium solution turns out to be unstable
for supercritical Ra numbers, and the other two solutions are stable
with respect to small disturbances of the same periodicity (so that
consideration of disturbances of different periodicities is necessary to
explain the special role of hexagonal cells in convection phenomena
demonstrated by experiment).
Later, Ovchinnikova and Yudovich (1968) investigated analyti
cally the case of circular Couette flow between two cylinders
rotating in the same direction; in particular, a rigorous proof of many
o f Kirchgassner’s results can be found in this paper. Specifically,
using the small gap approximation, Ovchinnikova and Yudovich
proved that at a given wave number k Couette flow becomes unstable
when the Re number reaches (from below) the critical value Recr(A:);
with Re > Recr(A:) a second steady flow of the same periodicity
arises which is stable to any axisymmetric disturbance, and has an
expansion in powers of (Re — Recr)'^’ These results were not proved
rigorously by the authors without the use of a small gap approxima
tion, since they depend on the positivity of two constants formed
from complicated integrals; however, at any fixed Re >Recr(/f) and
k, these constants may be evaluated numerically, and Ovchinnikova
and Yudovich considered some examples of this type (in all o f them
the constants turn out to be positive). The numerical solutions for
the new steady flow with k = k„ were used by Ovchinnikova and
Yudovich also for the evaluation o f the torque for Re > Recr; their
results agree very well with existing experimental data, and with the
theoretical results of Davey (1962) discussed later. Meyer (1967)
evaluated the torque in a similar way by direct numerical solution of
the time-dependent equations for a two-dimensional disturbance in a
circular Couette flow (with the outer cylinder fixed, i.e., for Q2 = 0)
at Re > Recr. The numerical solutions o f Meyer all approached a
steady state as / which corresponds to a torque close to that
calculated by Davey. Convection in a fluid layer heated from below
and o f a flow between rotating cylinders (with > 0) at R a >

Racr or R e > Recr, respectively, were considered briefly by Ivanilov
(1966; 1968) who outlined a proof of the existence o f a great
number of steady flow regimes (apparently unstable) in these cases
with relatively close values of velocity distributions and energy.
Let us now discuss a more elementary approach which does not
use advanced mathematical techniques (and does not pretend to be
mathematically rigorous) but applies the dynamical equations for
nonrigorous quantitative explanation of the existing experimental
data. The main method in such an approach is to obtain equations o f
the type o f Landau’s equation (2.34) which describe the evolution of
disturbances that are unstable according to the linear theory directly
from the dynamic equations of the problem (combined with specific
plausible approximations). Stuart (1958) and Davey (1962) consider
this problem in connection with Couette flows between cylinders. In
the first of these works it is assumed that d = Rz — Ri
(Ri + R 2 ) / 2 = Ro and that only the inner cylinder rotates (with
angular velocity Qi). Here, instead of the Reynolds number it is
convenient to use the Taylor number Ta = (R e)^; the
Couette flow becomes unstable for Ta > Tacr ~ 1708. Stuart further
assumed that the difference Ta —Tacr is small (but positive) and that
at the instant t = 0 there arises a disturbance periodic in z that is
unstable according to the linear theory. He assumed also that the
form o f this disturbance (which can be found from linear stability
theory) varies only slightly with time (so that approximate equation
(2.39) for the amplitude A {t) is applicable). Consequently, the
dependence o f the amplitude ^4 (<) on time can easily be determined
from the equation of balance of disturbance energy (i.e., Eq. (2.33),
which is a corollary of the dynamical equations). Hence, one obtains for
wws an equation of the form (2.41) with coefficients y and 5 expressed
explicity in terms of Taylor’s number Ta, the Reynolds number Re =
^ i R i d l v , the wave num b er/: = kcr and the eigenfunction correspond
ing to the eigenvalue problem (2.1 6)-(2.17 ), evaluated by Chand
rasekhar (1953). The most important result o f the calculations is in fact
5 < 0. This result agrees with the nonexistence of unstable disturbances
at T a< T ac/- and means that there exists a finite value lyAUas =
2 (7 / 5 )*'^\ while |/l|max ~ (1 — Tacr/Ta)'/*~(Ta — Tacr)'^' for small
Ta — Tacr > 0. Knowing li4|max and the spatial form of the unstable
disturbance (which is given by the linear disturbance theory), Stuart
was also able to calculate the torque, i.e., the moment of the
frictional forces acting on the surface of the cylinders. The values
which he obtained for this torque proved to be in very close

agreement with G. I. Taylor’s data obtained by direct measurement
(1936b) up to values of Ta approximately ten times greater than
Tacr-
A more precise equation for the amplitude o f a disturbance in
circular Couette flow was obtained by Davey (1962) who used a
method similar to that o f Stuart (1960) and Watson (1960) [their
works will be discussed later]. Davey took into account that an
initial axisymmetric disturbance of the form a' {x) = A due
to the nonlinearity o f the dynamical equations, will also generate
higher harmonics (proportional to n = 2, 3, . . .) and that the
dependence on r o f the form o f this disturbance will also vary
slightly with time. Hence the velocity field o f the disturbance will be
written here in the form
u '{x, /) = »o(r, t)-\-tt^{r, . . . (2.43)
(where the term «o(^, 0 describes the distortion of the form of the
laminar Couette flow produced by the disturbance). Further, it is
assumed that as ^->-0, on the right side of Eq. (2.43) only the term
K, (r, t) is conserved, while, for very small / > 0 , this term becomes
the solution u ' = defined from the linear disturbance
theory. Then, for slightly greater positive t, this term will become the
leading term, while Ui{r, t) for such values of t may be written as
Uy (r, t) = A {t)f{r) + higher-order terms. (2.44)
Substituting Eqs. (2.43) and (2.44) into the nonlinear equations

(2.32) [instead of using a sole equation (2.33) as was done by Stuart
(1958)], we may once again obtain for A {t) an equation of the form
(2.41), where is determined from the linear disturbance theory and
8 = 81 + 82 + 83. (2.45)
Here 61 > 0 defines the influx o f energy from the basic disturbance
«i (r, t) to the laminar Couette flow which produces the distortion
t) [only this term, in fact, is taken into account in Stuart’s work
where the changes in disturbance form were neglected], 62 describes
the generation o f higher harmonics by the basic disturbance, and 63 is
the distortion of its radial form. For all three terms on the right side
of Eq. (2.45), Davey obtained cumbersome equations (containing
solutions of the corresponding eigenvalue problem of the linear

theory). Then for the cases a) Qz = 0 ; b)d«Cô, ^ 1;
and c) Rz = 2 Ri, Q2 = 0 , he calculated the values of these terms
numerically. In all cases the coefficient 6 turns out to be positive;
in case a) its values agree approximately with the results of Stuart’s
less precise calculations, and in cases b) and c) they also lead to
values o f the torque under supercritical conditions which agree very
well with existing experimental data.
The agreement between the calculated and observed values of the
frictional torque is convincing evidence that Landau’s equation
(2.41) obtained by Stuart and Davey with 6 > 0 gives a rather
accurate description of the real growth of an axisymmetric distur
bance that is unstable according to the Unear theory. However, this is
still indirect evidence since it is not the value of the amplitude itself
which is compared with experiment, but the integral characteristic of
the flow computed according to that characteristic—the total torque
acting on the cylinder. A more direct experimental verification of the
applicability of Landau’s theory (and of Davey’s calculations) to
flow between rotating cylinders was carried out by Donnelly
and Schwarz (1965) [see also Donnelly (1963), and Donnelly and
Schwarz (1963)] and by Snyder and Lambert (1966). Donnelly
and Schwarz used a special ion technique for measuring the flow
disturbances. They filled the gap between the cylinders with an
electrolyte C C I 4 (where the nonrotating outer cylinder was o f radius
R 2 = 2 cm and seven interchangeable inner cylinders were used of
various radii R i , most of which corresponded to small gap d = R^ —
R 2 < R 2 ), and measured the current passing through it to a
collector-a small plate on the outer cylinder, which could move with
constant velocity along the O2 axis. F o rT a > T a c r, in the electrolyte
between the cylinders a regular set o f steady toroidal vortices occurs,
the velocity field of which takes the form u' (Jf) = A f {r)e^^\ where
the coefficient y4 is /4 ( 0 0 ) = Amax of Landau’s theory. These vortices
disrupt electrically charged fluid layers around the electrodes, and
hence affect the current passing through the electrolyte. In cal
culating this phenomenon, it is seen that the appearance of vortices
must correspond to the appearance in the expression for the current j
of an additional term of the form Aj = CA cos kz, where C is a
specific constant. The measurement results confirm that for
Qi >Qcr = {v^TacrlRid^)''‘ such a component periodic in 2 will in
fact exist. The square of its amplitude (CA)^ turns out to be
proportional to 2 ^ — — Ta — T a^ (which is in complete
accordance with Landau’s theoretical prediction) right up to the

considerably greater value £2i = Qzcr. after which the sharp decrease
in ion signal begins (see Fig. 23 which corresponds to measurements
at Ri = 1.9 cm; other examples of such curves can be found in
Donnelly and Schwarz (1965). Based on the general propositions of
Landau’s theory, we may assume that the sharp break in the law
(Ta—Tacr)'^* when Qi = Q2cr, is connected with Taylor’s
number attaining its “second critical value” Ta2 cr, where the toroidal
vortices become unstable and decompose into more complex
disturbances.
FIG. 23. D e p e n d e n c e o f th e square o f th e a m p litu d e o f th e a d d i

tio n a l current (A /)^ in an e le c t r o ly t e b e tw e e n r o ta tin g c y lin d er s
o n th e angular v e lo c ity o f th e in tern al c y lin d er [A c c o r d in g to
D o n n e lly ( 1 9 6 3 ) ] .
The data in Fig. 23 concern only the ultimate amplitude

A(oo) = of Landau’s theory. In addition, Donnelly and Schwarz
(1963; 1965) produced another verification of Eq. (2.41), based on
the use of Eq. (2.42) for finite values of t. With this goal they rewrote
Eq. (2.42) in the form
A^(() „2y/
(2.42')
where the undetermined constant C was successfully eliminated

through a selection of the origin o f time from the condition
A(0 ) = Mmax- Donnelly and Schwarz then produced a series of

experiments, in each o f which the speed 0 ^ = 3 rad/sec (which
appears to be subcritical from the conditions o f the experiment) was
suddenly increased to the value shown on the right of each of the
curves in Fig. 23a. As a result, fluctuations in the ionic current
immediately arose with a variable amplitude proportional to A(t).
The change in this current with time, measured from the moment
corresponding to the amplitude A = M ^ a x ’ plotted in Fig. 23a
together with the points calculated from Eq. (2.42') with values of y
also indicated to the right of each of the curves. The agreement
between theory and experiment is excellent. Values of y correspond
ing to each o f the curves in Fig. 23a may also be calculated from the
solution o f the linearized equations [i.e., of the eigenvalue problem
(2.16)—(2.17)] with given values of Rj^, Rg, Jli and fig = ^ê
associated wave number k in Eq. (2.17) m aybe chosen equal toAr^^,
since experiments show that for moderate supercritical values of Re
time (sec)
FIG. 23a. I o n cu rren t a m p litu d e as a f u n c t io n o f tim e u p o n su d d e n ly in creasin g th e sp e ed

n 1 fr o m b e lo w critical t o v a riou s su p ercritica l values. T h e d o t s rep re sen t c o r r e sp o n d in g
th e o r e tic a l curves for v alu e s y s h o w n a t th e right o f e a ch trace (A fte r D o n n e lly and S chw arz
19 65).
or Ta the axial wavelength of the Taylor vortices remains approxi

mately constant, i.e., almost indistinguishable from 2 n/k^^; instead of
this one may choose the value of k from the maximum instability
criterion (i.e., from the condition Imo) = y = max), which, for
moderate Ta-Ta^.^., leads to practically the same result. Such
calculations o f the value of y were carried out by Roberts (1965);
they led to results which agreed well with the values o f y determined
from the curves in Fig. 23a (which again confirms th aty ~ (Ta-Ta^,^),
as also follows from the data in Fig. 23).
The results o f the brilliant experiments o f Donnelly and Schwarz
substantially confirm the theoretical concepts developed by Landau;
however, quantitatively, they do not appear to be sufficiently
complete. Equation (2.42') contains a single parameter y; for this
reason the data of Fig. 23a permit verification only of the results of
the calculation o f this parameter, following from the linear theory
(together with the qualitative conclusion regarding the validity of
equations like Eq. (2.41) with a positive coefficient S). No
conclusions regarding the values of S may be elicited from the data
of Fig. 23. From this viewpoint, the results o f Snyder and Lambert
(1966) appear to be more informative. They used a special hot
thermistor anemometer, measuring the total velocity gradient at the
gap wall. Since Davey in his calculations considered only axisym-
metric disturbances, while circular Couette flow for small gap
becomes unstable with respect to nonaxisymmetric disturbances for
Ta only slightly in excess of the critical value Ta^^., Snyder and
Lambert chose for verification the third case studied by Davey
= 2, fig = where instability with respect to nonaxisym
metric disturbances ensues for substantial values o f Ta-Ta^j,. In their
apparatus a small axial flow (the corresponding axial Reynolds
number was about 2), was created by which they were able to fix the
dependence o f the velocity gradient on the z-coordinate, and
thereupon single out (with the help o f a spectral analyzer) the basic
disturbance, proportional to cos kz, and its first, second and third
harmonics. They found that the values obtained for the amplitudes
o f the basic disturbance and its first harmonic (proportional to cos
2 kz) are in excellent agreement with the results of the calculations
based on Davey’s values of y, 8 and 83 , over a very wide range of
Taylor numbers Ta (ranging right up to 4Taj,j.).
The theoretical conclusions o f Stuart (1958) and Davey (1962) are
in good agreement with the experimental results of Donnelly and
Schwarz (1963; 1965) and Snyder and Lambert (1966), and taken
together, they describe satisfactorily the nonlinear development of
axisymmetric Taylor vortices. However, this does not mean that all
the principal nonlinear effects in circular Couette flow may be
regarded as understood. In fact, the experiments of many investi
gators [and first o f all the work o f Coles (1965)] show, that upon an
increase o f the Reynolds number (or Taylor number) above the
critical value, transitions to new nonaxisymmetric flow regimes
appear rather quickly. Coles identified a whole series o f successive
discrete transitions from one regime to another; if one characterizes
the state o f flow with m vortices between the two ends of the
cylinders, each of which has an azimuthal wave number (number of
azimuthal waves) n, by the symbol m/n, then for ftg = ®
increasing speed of rotation the following states appeared
successively at quite definite (and repeatable) speeds: 28/0 (Taylor
vortices); 28/4; 24/5; 22/5; 22/6; etc. (in all 25 different s ta te s .^
Among these, in all cases with n ^ 0 the boundaries between
successive vortices had wave-like form. The results of Coles differ
somewhat from the results of other investigators; thus, for example,
Schwarz, Springett and Donnelly (1964), having undertaken an
investigation of Couette flow in an apparatus of much greater length
(containing up to 260 Taylor vortices) and with a much smaller gap
between the cylinders, found the first transition to a nonaxisym
metric state with n = \ (for Ta only 3 —8% in excess of Ta^|.).
However, it is im portant to note that these experiments also confirm
the appearance of nonaxisymmetric states of flow.
The experiments described above create the impression that with
increasing Ta (or Re) for comparatively moderate supercritical values
axisymmetric Taylor vortices become unstable to nonaxisymmetric
disturbances, having roughly the same axial wave number k, but a
differing axial phase (otherwise the boundary between successive
vortices would remain plane, and not become wavy). The same
conclusion follows also from the theoretical calculations of Davey,
Di Prima and Stuart (1968) in which a nonstandard method was used
to investigate the stability of Taylor vortices (in the case of small
gap). These investigators assumed that at the moment t = 0 in
already n o t e d in S e c t. 2 .6 in th e c ase o f circular C o u e tte f lo w , w h ic h rem ains

Stable to in fin ite s im a l d istu rb a n c es fo r all R e y n o ld s n u m b e r s, an in crease in th e R e y n o ld s
nu m b er le ad s to a q u ite d if fe r e n t t y p e o f tr a n sitio n to tu r b u le n c e , c h a r a cterized b y th e
a p p earan ce o f sep arate tu r b u le n t reg io n s in th e f lo w [c f. C oles ( 1 9 6 5 ) ; V a n A tta ( 1 9 6 6 ) ;
C o les a n d V a n A tta ( 1 9 6 7 ) ] .
circular Couette flow there arise four independent disturbances of

the following form:
AcfQir) cos fez; A^fiir) sin fez; B ^ f 2 (r) cosfeze*’"'’; Bgf^(r) sinfeze*"*^.
(2.46)
With the evolution of time, all four amplitudes = Aît),
Ag = A^{t), = B^(t) and B^ = B ,(/) will vary, interacting among
one another; in addition, they will create a distortion of the velocity
profile o f the basic Couette flow, which will change slightly their
own form, and will produce higher axial and circumferential
harmonics. All these effects may be taken into account if one
postulates a velocity field u ' = u'(r, tp,z,t) in the form of a Fourier
series in z and substitutes this series into the Navier-Stokes
equations (taking advantage in the boundary conditions as well, of
the simplification arising from the assumption of small gap), and sets
equal to each other the coefficients of similar terms of the Fourier
series in the right and left sides of the resulting equality. Carrying out
the necessary calculations and retaining in the resulting equations
only terms of order no higher than fourth in the amplitudes A and B
[which corresponds to retaining only terms of order \A\^ and on
the right side of Eq. (2.41)] Davey et al. obtained a system of four
nonlinear differential equations for the amplitudes A^, A^, B^, B^.The
system obtained is similar in form to the systems (2.49) and (2.52),
written below (but much more complex—eight lines are required to
write it); its right side contains nine numerical coefficients [of the
type of coefficients y, a, S i , and 62 of the system (2.52)], defined
through cumbersome manipulation of the solution of the corre
sponding eigenvalue problems (2.14)-(2.16), describing the separate
disturbances (2.46). Davey et al. investigated all the stationary
solutions of this system for supercritical values of Ta; one of
these solutions turned out to be, naturally, the Taylor-vortex
flow, for which = (2y/S)*, A^ = B^ = B^ = 0 [here y and S
are the same as in Eqs. (2.41) and (2.45)]. Considering further
the case o f small disturbances to Taylor vortices, i.e., postulating
Aît) = (2y/S)^ + a^(t), Ag = B^ = Bît), B^ = Bît) where
Op, Ag, Bp, Bg are of small magnitude, after linearization o f the
general ampUtude equations, one obtains new equations of the form
da
1 : =-V .
dA^ dB^ dB^
dt
These new equations determine (within the framework of linearized

stabihty theory) the temporal evolution of disturbances of the form
(2.46) to a flow o f Taylor-vortex form. Here yg is a coefficient
describing within the hnearized theory the temporal evolution of a
nonaxisymmetric disturbance, proportional to e \ p [ i ( k z + and
Sjg and § 2 6 two more o f the nine coefficients mentioned above,
occurring in the amplitude equations.
Since y > 0 forTa > 0 ^ ( 0 is damped. Consequently, stability
or instability of Taylor vortices is determined by the sign of the real
part o f the combinations yg + yS^g/S and yg + ySgg/S.
Exact calculations of the coefficients yg, S^g and Sgg for various
K, «, 0 ,2 ’ ^ 1 ’ ^2 appear to be an extremely complicated and
cumbersome problem. For this reason Davey et al. considered only
the case 0,2 = 0,fe = and introduced several simplifying approxi
mations. Here for the simplification adopted, Re[yg + yS^g/S] < 0
for all Ta-Ta^j. > 0 (i.e., the Taylor vortices are stable to relatively
in-phase nonaxisymmetric disturbances). At the same time
Re[yg + yS2 B/^l becomes positive for Ta/Ta^j. = 1.08; conse
quently, the Taylor vortices become unstable to out-of-phase
nonaxisymmetric disturbances for Ta only 8% in excess of Ta^,j. (as a
result o f the simplifications made in the numerical computations this
result can be considered only as a rough estimate of the magnitude
Tagcr)- Moreover, the dependence o f Re[yg + yS2 B/S] on the
azimuthal wave number n turns out to be rather weak, but, in
general, one may say that Re[yg + yS2 B/S] decreases with increasing
n, so that the initial instability must occur for n = 1. All these
conclusions agree rather well with the experimental findings of
Schwarz, Springett and Donnelly, obtained under conditions more
similar to those assumed in the theory of Davey et al. (where the
length o f the cylinders was considered infinite, and the gap small),
than those characterizing the experiments of Coles.
The theory o f Davey et al., of course, is not quite ex act-th ey
considered only the interaction of four disturbances o f the special
form (2.46), while for every Ta > Ta^.^. in circular Couette flow there
will exist infinitely many unstable disturbances (corresponding to the
interval of unstable wave numbers k, containing expanding
with increasing Ta-Ta^^,). For this reason, even the question of why,
for moderate positive Ta - Ta^j. (or Re-Re^.^.), strongly z-periodic
states of flow between cylinders arise, characterized by a single value
of the wave number k, cannot be regarded at present as completely
resolved. Some general considerations concerning this question (and
not even using in a specific form the equations of fluid mechanics,

i.e., having to do with a wide class of physical processes, described by
nonlinear partial differential equations) may be found in Pono
marenko (1968a).
Let us turn now to the mathematically much simpler problem of
convection in a layer of fluid heated from below. We have already
pointed out that for Ra > Ra^,^, stationary solutions describing states
o f rest turn out to be nonunique; in addition to these additional
stationary cellular solutions arise, the amplitude of which is
proportional to (Ra-Ra^p^ [for small(Ra-Ra^j,)/Ra]. This fact agrees
well with the idea that for Ra > Ra^^. in a layer of fluid heated from
below, there occur softly excited oscillations (spatially, rather than
temporally) corresponding to Landau’s schema with § > 0. Addi
tional cellular solutions o f the nonlinear Boussinesq equations were
studied by Sorokin (1954), Gor’kov (1957), Malkus and Veronis
(1958), Kuo (1961) [in this paper a numerical method o f analysis
was used for the first tim e]; Bisshopp (1962), and others. It is
important that stationary solutions of the nonhnear convection
equations be strongly nonunique.
Let us begin with the simplest approach of Stuart (1958), based on
an approximate assumption (2.39) [the so-called “ shape assump
tion” ] that the form of the disturbances does not change with time,
and coincides with the form of the unstable disturbances which arise
for Ra = Ra^^. Then, for every choice of solution v5(x^, X2 >of Eq.
(2.23) with k = k„^,c r describing dependence of the unstable distur-
bance on the horizontal coordinate, one may easily obtain an
equation o f the form (2.41) for the amplitude A = Ait) through the
balance equations (2.33") and (2.33"'), where y (Ra-Ra^^,) is
obtained from linear disturbance theory, while 8 is simply related to
the eigenfunctions (of the eigenvalue problem) which describes the
neutrally stable infinitesimal disturbance [cf., for example, Roberts
(1966)]. The assumption of preservation (“shape assumption” ) of
the Landau-Stuart equation (2.39) does not appear to be exact, and is
suitable only as a first approximation for Ra slightly in excess of
Ra^,j,; however, the more exact methods of Gor’kov, Malkus and
Veronis, Kuo, Bisshopp and others, also permit a large number of
distinct stationary solutions to be obtained in the form of two-
dimensional waves, proportional to cosik^x^^ + ^2 ^2 ^’ fef +
(so-called rolls), square cells, hexagonal cells, etc. An extremely
general method for the construction of stationary solutions to the
nonlinear convection equations for moderate values of Ra > Ra^^
T80 STATISTICAL FLUID MECHANICS
[suitable, however, up to Ra several times cf. Busse (1967a)]

was developed by Schliiter, Lortz and Busse (1965). This method
utilized an expansion in powers o f a small parameter « (related to the
expansions used by G or’kov and by Malkus and Veronis for the
construction of some specific solutions) of the form
u (x ) = eu<l>(x) + f2„< 2)(x) + . . . ,T ( x ) = ,
Ra = Ra^^ + «Ra<l> + f2 R a < 2 ) + ... .
If we set equal the coefficients of equal powers of e on both sides of

the system o f Boussinesq equations, and each time use the boundary
conditions of the problem, we obtain for«*^^(x), the usual
linearized convection equations; for terms of higher order in the
series, a sequence of systems of inhomogeneous partial differential
equations is obtained. It is well known that such an inhomogeneous
system will have solutions only if appropriate existence conditions
are satisfied (namely, if a scalar product is defined in the space of the
pairs u(x), T'(x),when the right side of the system must be orthogonal
to all solutions of the adjoint homogeneous system). These existence
conditions allow one to successively define all the Ra^'"\ Subse
quently, u^'"Hx) and m = 2 , 3 , . . . in many cases may be found
uniquely with the help o f solutions of the corresponding inhomogen
eous system (if for we choose some suitably defined
solution o f the linearized problem), and finally we may define even
the param eter« from the equality Ra - Ra^^. = fRa^^^ + Ra^^) ^ .
[c f specifically, an example of the application of this approach to
the solution o f a simple model nonlinear partial differential equation
analyzed in the survey paper by Segel (1966)].
Schliiter et al. studied the stationary solutions obtained if the
zero-order approximation is chosen proportional to
the function
ly
^ ’ (2.47)
n = -N
/ = 1,2: C_„ = c*„-, =

= const. In this case the existence conditions for the solu
tion of the second-order system (for , t ^2)) reduced to the
condition Ra^^^ = 0 (thus if we limit ourselves to terms of order
no higher than P‘, then e ~ (Ra - Ra^,j.)^). If we take = 0 , then

the system o f equations T*2 >has unique solutions for arbitrary
choice o f the functions (2.47) in the zero-order approximation.
However, the existence conditions for the solution o f the third-order
system not only define Ra^^) uniquely, but also impose a restriction
on the function if we are given a set of 2 -vectors
jj, expression (2.47), then a stationary
solution o f the nonlinear third-order system will exist only for a
rather narrow choice of the coefficients |C^|, . . . , [C^]. Thus, not
every solution of the linearized problem can be a zero-order
approximation to the stationary solution of the nonlinear system of
convection equations. In addition, Schliiter et al. showed that the
number of functions from which one may construct
stationary solutions uix),T'{x) up to terms of all orders in e,
nevertheless, appears to be infinite. Thus, for example, all existence
conditions can easily be satisfied in the “regular case” (in which all
angles between neighboring k-vectors are equal, and |Cj|^ =
. . . = |C^ | 2 = 1 / 2 N; the regular case includes rolls, square cells and
hexagons). This is also true for even more general “ semiregular” cases
[cf. Segel (1965);B usse( 1967c)].
For Ra much greater than Racr the expansion procedures of
Gor’kov, Malkus and Veronis, and Schluter et al. are difficult to
apply. In these cases the direct numerical procedures seem to be
preferable. Such numerical procedures (based on various simplifying
approximations) were used for finding specific steady solutions of
the Boussinesq equations (mostly the simplest two-dimensional rolls)
for different boundary conditions and different Pr values in the
works of Kuo (1961), Herring (1963; 1964), Deardorff (1964),
Fromm (1965), Veronis (1966), Busse (1966b), Roberts (1966),
Schneck and Veronis (1967), Plows (1968) and some other authors.
The results obtained in these works agree well in many respects with
the existing data on convective heat transfer at large enough Rayleigh
numbers, and on the mean characterisitcs o f temperature and
velocity fields under such conditions. In particular, it is worth noUng
that the computations of all these investigators show that the vertical
profile o f the mean (i.e., averaged over horizontal coordinates)
temperature at large Ra differs strongly from the linear profile which
is observed at Ra < Ra^^. Namely, as Ra is increased, a thick region
in the center of the fluid layer achieves a nearly isothermal state in
the mean (the thickness o f this region increases with increasing Ra),
and almost all change in the mean temperature is concentrated in
two thin thermal boundary layers near the boundaries of the flow.
The temperature field for large values of Ra is characterized by a
large mass of nearly isothermal fluid in the center and a specific
mushroom-shaped form o f the isotherms. It is an interesting detail
that in the almost isothermal central region a small positive vertical
temperature gradient (i.e., a reversal of temperature gradient) occurs
when R a/R a„ is greater than several units. These results are most
distinct on the figures presented by Veronis (1966); however, they
can be observed on the data of all other investigators beginning with
Kuo (1961). The reversal o f the temperature gradient in the thick
central region of the layer of fluid (at R a/R acr = 16) was observed
also experimentally by Gille (1967) with the aid o f precise
interferometric measurements.
Let us now discuss a very important question concerning the
existence of a preferred disturbance mode which is the only one
occurring in real physical layers o f fluid. This question is closely
connected with the stability theory for cellular convective motions
which is at present far from being complete. As we know from Sect.
2.7, the linear stability theory leads to the conclusion that for R a >
R a„ there must exist an infinite set of unstable infinitesimal
disturbances (with exponential growth rates) corresponding to some
range of values of the wave number k surrounding the value k = kci
at which instability first appears. The most unstable (i.e., the
most rapidly increasing) disturbances will correspond to one definite
ft-value, but there will be an infinite set of such disturbances also
[since for given k the horizontal form o f a disturbance may be
described by an arbitrary function a;2) satisfying Eq. (2.23)].
Experiments show, however, that under each specific set of
conditions there will always arise only a disturbance having a strictly
defined form (corresponding largely to a division of the horizontal
plane into a set o f regular hexagonal cells, although there are other
possibilities), and strictly defined finite amplitude. Landau’s theory
permits only this steady amplitude to be found [with the aid of Eq.
(2.41) obtained from the nonlinear Boussinesq equations on the
assumption that the function ( x i . x j ) is known; cf. Gor’kov
(1957)]. However, this theory says nothing of why disturbances with
several different values of k never arise in the fluid, and why among
all the disturbances with given k, only those with one definite form
^{X\,X 2) are actually observed. The fact that in a number of cases,
nonlinear interactions of disturbances differing in wave number may
lead to the vigorous growth o f distrubances of one given wave
number, due to the suppression of all the rest, is explained partially
by the calculations of Segel (1962). In this work a simple “ pair

interaction” of two rolls independent of the^rj coordinate in a layer
bounded both above and below by plane-free boundary conditions)
was considered. In other words, Segel studied the evolution of a
disturbance for which the velocity component
where ^ = Xi/H, -q = x^/H, f = X 3 /H, and H is the depth of the
layer, takes the form:
u ^ { x , t) — (t) cos (C) + A 2 (t) cos /S/2 (C) 4 - small com plem ent.
(2.48)
Then, applying the methods of Stuart (1960) and Watson (1960a),
which we shall discuss later, Segel obtained in the first nonlinear
approximation the “ amplitude equations” for the functions/ 4 ,andy 42
o f the following form:
dA
which for /!•>, = 0 or /li = 0 , clearly yield an equation which is

equivalent to Landau’s equation (2.41) for the amplitude o f a single
disturbance. The system (2.49), evidently, has the following steady
solutions:
(I) Ai = A 2 — 0,
(II) A, = 0, ^ 2 = (T2/ 8/ ‘.
(III) = >^2 = 0,
(IV) yi, = (T.82 - T2P.)'^’ (S182 -
The stability o f these solutions may be verified with the ordinary

methods of stability theory of differential equations (or the
nonlinear theory of oscillations). It is found that in the most
im portant case, when 71 > 0 ,7 2 > 0 , 6 2 > 0 , 6 2 > 0 , it follows from the
stability of even one o f the solutions (II) and (III) that the solution
(IV) cannot be stable. Thus it is clear that a wide class of situations
exists in which the ultimate state will always contain only one roll
(with a definite wave number) but not a mixture of both. A more
detailed analysis o f the equations when Ra is just above Racr shows
that if the linear theory growth rate ifi of the first roll is more than
twice the growth rate 72 of the second roll, then in the final
equilibrium state only the first roll appears (i.e., (Ill) is the only
stable steady solution). Similarly if 12 > only solution (II) is
stable. That is, if one o f the two competing primary disturbances has
a sufficiently great advantage according to the linear theory, only the
advantaged disturbance will appear in the final state while the
disadvantaged one ultimately decays. On the other hand, if 72 < <
2 ’(2, then both solutions (II) and (III) are locally stable and the firial
state is either (II) or (III), depending on the initial conditions.
However, in this case the initial conditions for very small distur
bances are much more likely to be such that the ultimate state
corresponds to solution (III) [cf. Segel (1966)]; in this sense, the
disturbance with the greater linear growth rate will always be more
advantaged in the nonlinear theory also. In principle, it is also
possible [for certain values o f the coefficients in Eq. (1.49)] that the
interaction o f one stable and one unstable disturbance (i.e., the case
with > 0 and ^2 < 0) may lead to increase of the unstable
disturbance and the final establishment of a “ mixed state” corre
sponding to a solution of the type (IV). It is probable that such a
situation may also sometimes be encountered in mechanically driven
flow (for example, in the case of plane-parallel flow). However, for
thermal convection problems with Ra number only slightly greater
than Racr and instability of both primary rolls (i.e., with f j > 0, Ti ^
0) one of the two rolls will necessarily decay. This explains to some
extent how the interaction between disturbances lead to the fact that
for small Ra—Ra<;r, out of the whole range of unstable disturbances,
only that with one definite value of wave number k is actually
observed. The same results were obtained by Segel for the more
general case o f N parallel “nonreplicated” rolls (such that if k and /
are the wave numbers o f two o f the rolls, then (k + l ) / 2 is not among
the wave numbers considered; here also a single disturbance
ultimately appears and if one of the rolls has much greater linear
growth rate than all the others, it necessarily survives [cf. Segel
(1966)]. A more general approach to the problem of the mechanism
of selection o f a single preferred wave number from the whole band
of unstable wave numbers for given Ra > Racr is outlined by
Ponomarenko (1 968a).
Now let us consider disturbances with a given wave number k
only. Here, also, a question arises on the selection o f the preferred
mode because there are an infinite set o f different disturbance forms
with the horizontal wave number k. However, only one o f these
forms is observed in real experiments performed under fixed external
conditions. Therefore the investigation of the stability o f the
different steady solutions of the nonlinear Boussinesq equations for
Ra > Racr is o f great importance. One o f the most complete
investigations of this type was carried out by Schliiter, Lortz and

Busse (1965). These authors considered all the steady solutions
evolved from infinitesimal disturbances o f the horizontal form (2.47)
at a slightly supercritical value o f the Rayleigh number and studied
the stability of the finite amplitude cellular motions so obtained with
respect to infinitesimal disturbances o f similar form. As a result, they
came to the rather unexpected conclusion that all the cellular
solutions with the exception o f the simplest two-dimensional rolls
[which correspond to = 1 in Eq. (2.47)] are certainly unstable.
For the exceptional case of the rolls with given horizontal wave
number k, Schliiter et al. showed that they are stable to all
infinitesimal disturbances with the same wave num ber k if only this
wave number belongs to the band of unstable (according to the linear
stability theory) wave numbers. Finally, Schluter et al. investigated
the stability o f two-dimensional rolls of finite amplitude to infinitesi
mal disturbances of horizontal wave number ki ^ k. They found
that when Ra — Racr is small enough the rolls with wave number k <
k „ (where k „ is the wave number of the infinitesimal disturbance
FIG. 23b. S t a b ility range o f r o lls at R a y le ig h n u m b e r s c lo s e t o

critical [A fte r S c h lu te r, L o rtz, an d B u sse ( 1 9 6 5 ) ] .
which is neutrally stable at Rajr) cannot be stable to disturbances

with arbitrary wave numbers. However, if k is greater than k cr and k
— ^cr is small enough (of the order of Ra —Rac,) the rolls with wave
number k are stable with respect to all possible infinitesimal
disturbances. The full range o f the stable two-dimensional rolls for
small enough values o f the difference Ra — Racr > 0 found by
Schliiter et al. is shown in Fig. 23b.
The results o f Schliiter, Lortz and Busse are obtained by an
expansion procedure in powers o f a small parameter e and are valid
for Rayleigh numbers close to the critical value only. The general
stability analysis of the solutions of the Boussinesq equations at
higher Rayleigh numbers is difficult to perform. However, in the
particular limiting case of infinite Prandtl number, the Boussinesq
equations undergo considerable simplification and the stability
problem becomes accessible to analysis. Using numerical methods,
Busse (1967b) computed steady solutions of the Boussinesq equa
tions with Pr = oo in the form o f rolls for a wide range of Ra numbers
and investigated the stability of the solutions obtained with the aid
of the usual linear stability theory. He found that stable rolls
correspond to a narrow elongated region on the (Ra, A:)-plane [see
Fig. 2 3 c ]. The range o f stable wave numbers for all Rayleigh
FIG . 2 3 c . S t a b ility r e g io n o f r o lls fo r a w id e range o f R a y le ig h n u m b e r s

in th e case Pr = «> [ A ft e r Bu s s e { \ 9 6 1 ) ] .
numbers below 22,600 is restricted to a small band (almost

independent o f the value of Ra) surrounding ^cr- At Ra = 22,600 all
two-dimensional solutions of the Boussinesq equations (at least when
Pr = °°) become unstable. It is important to note that the value 22,600
is of the same order of magnitude as the value of Ra at which the
second discrete transition (in cellular convection) was observed
experimentally.
The results shown in Figs. 23b and 23c are in qualitative
agreement with the fact that the characteristic scale o f the horizontal
cells does not change considerably when the Rayleigh number
increases. However, there is another contradiction between these
theoretical results and the experimental findings. The theory asserts
that the only stable form of fmite-amplitude cellular convection is
two-dimensional rolls. Such two-dimensional convective cells were in
fact observed by several investigators including Silveston (1958;
1963), Koschmieder (1966), Chen and Whitehead (1968), Rossby
(unpublished Ph.D. thesis; MIT, 1966), and some others. However,
even more usual are observations in which cellular convection takes
the form of beautiful regular hexagonal Benard cells; this noteworthy
fact was twice stated by us in a previous discussion of the convection
problem. Now we must explain this fact from the viewpoint of
stability theory.
The explanation is based on the finding that additional effects
neglected in the Boussinesq approximation often play a dominating
part in the real laboratory experiments. At the end of Sect. 2.7 we
mentioned that surface tension effects played an important role in
the original Benard experiments; the same is apparently tnie for
many other experiments with liquid layers bounded above by an air
surface [cf. Koschmieder (1967); see also the corresponding theory
o f Scanlon and Segel]. The experimental data by Koschmieder
(1966; 1967) and Chen and Whitehead (1968) show also that the size
and the form o f the tank in which the convection takes place is of
great importance for the form o f the convection patterns [cf. also
the theory by Davis (1969)]. However, Palm (1960) was probably the
first to point out that the usual Boussinesq equations of free
convection cannot in principle provide a satisfactory explanation of
the basic peculiarities of hexagonal cellular convections. He justified
his categorical statement with reference to the experiments of
Tippelskirch (1956), which definitely show that the character of the
circulation in the cells is determined by the form of dependence of
the coefficient of viscosity v on the temperature T (with d^jdT < 0 ,
the fluid rises in the center of the cells and sinks at the edges, and
with dvIdT > 0 it rises on the edges and sinks at the center). Hence
Palm took as his basis more complicated equations which also take
into account the possible dependence of v ^ n T, (and estimated the
effect of this dependence on the value of Rac/-. He assumed further
that at the initial instant o f time there arises in the fluid some “basic
disturbance” in the form o f a roll (for example, proportional to cos
(kxzlH) = coskr\, where t) = xz/H, H is the depth of the fluid layer,
and independent of g = x JH ) on which is then imposed a weak
“background” o f various other disturbances of small amplitude, with
the same (most unstable) value o f the wave-number vector k. In this
case it is natural to assume that the fundamental role will be played
by the “pair interactions” o f the basic disturbance with the others.
In accordance with this, Palm confined himself to the study of the
evolution o f disturbances with vertical velocity «3 (jc, 0 = “ (i. i1. C. 0
where I = xJH of the form
«3(?. ■»l> 0 = IA (0 c o s ^ i7 + > l2 ( 0 c o s ^ i5 c o s A 2 ^ J /( C ) ,
kl-^kl= ^k\ (2.50)
The disturbances with k\ + k l = k'^ will apparently be especially

closely connected with the basic disturbance because quadratic
combinations o f these disturbances (which enter into the equations
of fluid mechanics) may once again generate terms of the same form
as the basic disturbance. One may further assume that the interaction
o f such disturbances with the basic one may, under certain
conditions, lead to the mutual amphfication of both, as a result o f
which it is only these disturbances that will finally play an important
role. On the basis o f such heuristic considerations. Palm proposed,
first, to consider the special case when
«3 (5. t ) = [ A ( 0 c o s kr\-\-A 2 {t) COS ( Y s k\/2) c o s ( ^ V 2 )]/(C )

(2.51)
(since here the two terms are strongly connected with each other).
Restricting the boundary conditions, for simplicity, to the physically
unreal “ free-free” case of convection between two free surfaces of
constant temperature [this simplification was later adopted also by
Segel and Stuart (1962), Palm and 0iann (1964), and Segel (1965)]
and taking for definiteness, f{ Z ) = sinX^, Palm deduced a system of
differential equations for the amphtudes Ai{t) and After
dropping terms of order higher than the third in the amplitudes, this
system has the form
T = - T ''^ 2 - - (282 - 8i/2) AiAl (2.52)

dAi ,.4 A s a3 / as s \ a2
= lf-^2 — °AiA2-- §2-^2--- (482 --- 8i) A 1A 2,
where a, fii, 62 are constant coefficients, Ra — Racr and

a Idv/dTl [cf. Segel and Stuart (1962)]. We see that the variation
o f viscosity with temperature generates second-order terms on the
right side o f the amplitude equations, whereas only first- and
third-order terms are present in these equations for a fluid with
temperature-independent properties [cf. Eqs. (2.49)]. It is important
that the system (2.52) has simple steady solutions of the form
A 2 = ± 2 A j, (2.53)
corresponding exactly to hexagonal prismatic cells. Moreover, Palm

showed that when a ¥= 0 (i.e., when dyldJ 0 , but not when d \jdt =
0 ) only solutions o f this type will be stable with respect to small
disturbances o f ampHtudes Ai andyl 2. so that precisely these will be
realized in the hmit as ^ oo. Since, as one can show, it is to these
stable solutions that the simple link between direction o f circulation
in the cells and the sign of dvIdT, observed experimentally corre
sponds, Palm concluded that these results fully explain the funda
mental experimental facts.
Later, Palm’s theory was critically reconsidered and generaUzed by
Segel and Stuart (1962), Palm and 0iann (1964), Segel (1965), Busse
(1967), Palm, Ellingsen and Gjevik (1967), and Davis and Segel
(1968). All these authors also took as starting point the third
approximation of disturbance theory. It was shown that certain
conclusions of Palm (1960) were not completely correct. First, it was
found that the solutions (2.53) o f Eqs. (2.52) are in fact stable only
for values of Ra — Racr that are not too large (i.e., Ra must be
smaller than some value Rai > Racr which depends on a typical
“scale” a of the variation o f viscosity and tends to Ra^^ as a ^ 0 ).
Moreover, when a 0, then even for a small range o f subcritical
values o f Ra (namely, for Rao < Ra < Racr where Racr — Rap is o f
order ), hexagonal steady motions of quite definite finite ampli
tude will exist, which are stable to infinitesimal disturbances.
Therefore the hexagonal convection cells were stable with respect to
all infinitesimal disturbances with the same horizontal wave numbers

(and apparently also to all other infinitesimal disturbances with no
exception; cf. Busse (1967c), and Ponomarenko (1968b) in the
whole range Rao < Ra < Ra, of Ra-values. 'For Ra > R a i, the only
stable solution of the amplitude equations is that which corresponds
to a convection pattern in the form o f two-dimensional rolls.
Moreover, the rolls are stable to all infinitesimal disturbances not
only for Ra > R a j, but for a wider range Ra > Raa where Racr <
Ra 2 > Rai- All other forms o f convection pattern are certainly
unstable; hence for the range Rao < Ra < R aj, only hexagons are
stable for Ra? < Ra < Rai both hexagons and rolls are stable, and
for Ra > Raj only rolls are stable. When the Rayleigh number Ra is
slowly increased the convection pattern starts growing at Racr and
takes the form o f the stable steady hexagonal cells. At Ra = Rai the
hexagonal convection pattern becomes unstable and transforms into
rolls, which are the only stable form o f convection at such high Ra.
With decreasing Rayleigh number the transition from rolls to
hexagons occurs at Ra = Raa (when rolls become unstable) and the
convection decays after Ra = Rao has been passed. We see that when
the Rayleigh number at first increases slowly and then slowly
decreases, a hysteresis effect must be observed. As a ->■ 0 (i.e., the
viscosity variations disappear), all the values Rao, R a i, and Ra 2 tend
to Racr and the results become identical with those of Schluter,
Lortz, and Busse.
The results presented above on the influence of viscosity variation
were obtained by Segel (1965) for a model case of “free-free”
boundary conditions and one special form o f dependence of viscosity
or temperature. Later, Palm, Ellingsen, and Gjevik (1967) considered
all possible types of boundary conditions with combinations of rigid
and free planes, and calculated all the “critical values” Rao, R a i, and
Ra 2 for these cases. However, the first and most general results were
obtained by Busse in his dissertation in 1962 [which was published
considerably later; see Busse (1967c)]. Busse used the parametric
expansion approach (described above in connection with the work
by Schluter et al.) and took into account the slight variation with
temperature not only o f the viscosity but of the thermal conductiv
ity, specific heat at constant pressure, and thermal expansion
coefficient (for all the combinations o f “ free” and “rigid” boundary
conditions). He found that all the effects considered imply the same
stability situation, which was described above for the case when only
viscosity depends on temperature. Later, Davis and Segel (1968)
showed that even in a fluid with constant properties, the hexagonal

cells will appear for Ra sufficiently near to Racr if the boundary
condition at the free top surface allows for its deformation. A very
general approach to the problem o f establishing hexagonal convec
tion cells was outUned by Ponomarenko (1968b), who did not use a
specific form o f dynamic equation but emphasized the leading role
o f second-order terms on the right side o f the amplitude equations
[one can note that these terms disappear for constant fluid
properties and that they make the equation for the amplitude of one
isolated disturbance different from the usual Landau equation
(2.41)].
It is worth noting that the transition from hexagonal convection
cells to rolls when the Rayleigh number is slowly increasing above
the critical value was really observed in the experiments o f Silveston
(1958). However, much more experimental work is clearly needed
for complete verification o f all the theoretical predictions presently
known.
Nonlinear Instability Effects in Plane-parallel Flows and Boundary
Layers. Transition to Turbulence
Let us now discuss the nonlinear development of disturbances in
plane-parallel flows. We shall first consider plane Poiseuille flow
which has the advantage that the linear stability theory was
completely successful in this case. However, the nonlinear analysis
proves to be much more complicated. The first attem pt in this
direction was made by Meksyn and Stuart (1951). They found, with
a number of simplifying assumptions, th at subcritical finite-
amplitude instabilities exist in a plane Poiseuille flow, i.e., that the
critical Reynolds number for two-dimensional disturbances o f finite
amplitude \A\ is less than Recr o f the linear stability theory and
decreases as \A I increases. This result compels us to assume that here
6 < 0. On the other hand, in the later work o f Stuart (1958), the
integral equation of energy balance (2.33') was used to obtain a
Landau equation (2.41) for the amplitude o f finite unstable
disturbance; here, with other simpUfying assumptions (the most
important being that the spatial form o f the disturbance does not
change with time and agrees strictly with the form of the
eigenfunction o f the linear stability equations), it was found that 6 >
0. A desire to resolve this controversy stimulated Stuart (1960) and
Watson (1960a) to carry out a more complete analysis o f the
behavior of two-dimensional, wave-like finite disturbances in a plane
Poiseuille flow with Re close to Recr- The Stuart-Watson analysis is

based on the use of the complete system o f dynamic equations and
expansions o f the type (2.43) and (2.44) [this analysis was later
taken as a model in Davey’s work (1962) which we have discussed
earlier]. Here also some approximations are made, but these seem to
be more natural and reasonable than those of Meksyn and Stuart
(1951) and Stuart (1958). The results of Stuart and Watson were
somewhat extended and obtained additional support in the work of
Eckhaus (1965) based on expansion of all the functions in terms of
the eigenfunctions o f the linear Orr-Sommerfeld equation. Another
elegant formulation of an expansion method closely related to the
initial Stuart-Watson approach (but adapted to the treatment of both
two- and three-dimensional, wave-like disturbances) was proposed by
Reynolds and Potter (1967).
All the above-mentioned works lead to the justification o f an
equation o f the form (2.41) for the amplitude of the disturbance,
and show that the coefficient 5 is made up of three terms 5 ,, S j , and
8 3 [which have the same meaning as the terms in Eq. (2.45)]. Not all
these terms were correctly taken into account by Meksyn and Stuart
(1951) and by Stuart (1958). In the latter work only the term 5i
was considered, whereas in the earlier paper the terms 5i and 63
were estimated approximately and Sj was fully disregarded. It was
also found that for all three terms, explicit expressions may be given
which contain the eigenvalues and eigenfunctions of the correspond
ing linear Orr-Sommerfeld equation (and also of the adjoint
equation) in a complex manner. The numerical calculation o f these
terms (and of their sum 6 ) is a very complex problem in numerical
analysis which, however, is accessible for modern high-speed com
puters. The corresponding computations were performed independ
ently by Reynolds and Potter (1967) [who used their own
modification o f the Stuart-Watson approach] and by Pekeris and
Shkoller (1967) [based on the Eckhaus eigenfunction expansion
m ethod]. The results of these two papers do not coincide numeri
cally (one reason being that they use different normalizations and
somewhat different definitions of the amplitude U l), but both
results have the same general behavior and imply close values of the
ratios o f 6 = 5(A:,Re) at different points o f the (A:,Re)-plane.
Reynolds and Potter computed the values 5 j , 62 > and 5 = 5i +
62 + S 3 at the critical point (kcj, Recr), at four other points of the
neutral stability curve on the (fc, Re)-plane, and at two points in the
neighborhood o f the neutral curve. They found that 5, is positive at
all points [in complete agreement with the previous results of Stuart
(1958)] and that at the critical point and the points on the upper
branch of the neutral curve 63 is negative and much greater in
absolute value than Sj [so that 5, + 8 3 is negative; this result agrees
with the approximate deductions of Meksyn and Stuart (1951)]. The
value 62 is also positive and o f the same order as 8 1 ; therefore the
sum 5 = 5, + 52 + 5j is negative at the critical point and points of
the upper branch o f the neutral curve (and is determined here
primarily by the process o f distortion o f the vertical profile o f the
basic disturbance). Of course, the most important is the result
obtained at the critical point: it indicates that finite disturbances in
plane Poiseuille flow lose stability earlier than infinitesimal ones (i.e.,
subcritical finite-amplitude instabilities exist) and that supercritical
finite-amplitude equilibrium states of the type known for a circular
Couette flow and a layer of fluid heated from below are rather
unlikely to be observed in plane Poiseuille flow (in full agreement
with the experimental evidence). However, the Reynolds and Potter
computations at the points of the lower branch of the neutral
stability curve reveal a surprising result (confirmed also by the data
of Pekeris and Shkoller) that 5 is positive (and comparatively small in
absolute value) at these points. Hence, periodic motions of finite
amplitude could theoretically exist in the case of plane Poiseuille
flow if the disturbance could be kept very “ pure,” i.e., higher
wave-number contributions could be suppressed (apparently this
would be very difficult to achieve experimentally).
Pekeris and Shkoller evaluated the coefficient 6 = 5(A:,Re) for an
extensive region o f the (A:,Re)-plane (using equations which are
reasonable in the vicinity of the neutral curve) and obtained results
that, in general, agreed with Reynolds and Potter’s conclusions; their
main result is presented in Fig. 23d.
We have already pointed out that Landau’s equation (2.41), for
6 > 0 , leads to physically interesting results only for disturbances in
flows with Re > RCcr, and in the case 5 > 0 only for finite
disturbances in flows with Re < Recr- For 6 > 0 and R e < Re^r this
equation is less interesting and for a < 0, and Re > Re^;. it quickly
becomes inapplicable; therefore in both cases it is reasonable to
supplement this equation by subsequent terms of the expansion in
powers of \A\'‘‘. The general equation for d\A\ÎdU taking into account
all terms of such an expansion, will evidently be of the form
FIG . 2 3 d . T h e r eg io n s o f p o s itiv e 7 a n d o f p o s itiv e 6 o n t h e (k, R e )-p la n e fo r th e case o f

p la n e P o ise u ille f l o w [A fte r P ekeris an d ShkoU er ( 1 9 6 7 ) ) .
(from which Landau’s equation is obtained, if we take into account

only the first two terms o f the right side). An equation of the form
(2.54) for the amplitude o f a two-dimensional disturbance in a
plane-parallel flow was obtained by Watson (1960a) with the aid of a
special expansion technique, proceeding from the equations of
motion. Here, however, the expressions for the coefficients Um with
m > 1 are far more complex than for the coefficient Oi = —6, so that
their evaluation still seems to be almost impossible. In the subse
quent work o f Watson (1962) a similar analysis was carried out for
plane Poiseuille flow, based on the representation of the two-
dimensional disturbances u' ( jut, t) in the form A ( x ) f { z , t) where the
amplitude A (x) in the linear approximation is equal to Here,
equations are obtained for A(x) which differ from Eqs. (2.41) and
(2.54) only in the replacement of t by x. The evaluation of the
coefficients of these equations naturally rests on the same difficult
ies as the evaluation of the coefficients in the equations for the
time-dependent amplitude A (<).
Similar results may be obtained also for many other types of

plane-parallel flows. As an example, Reynolds and Potter (1967)
investigated the case o f the combined plane Couette-Poiseuille flow
[i.e., the flow studied within the framework of linear stability theory
by Potter (1966), and Hains (1967)] for different relative intensities
o f the Couette component (in the range o f intensities for which
unstable infinitesimal disturbances exist). For all such flows, the
critical point on the (fc,Re)-plane was found through application of
linear stability theory and nonlinear analysis was carried out at the
critical point only. Calculations o f Reynolds and Potter show that 5
is negative at the critical point for all flows considered and is
determined primarily by the summand 6 3 . The dependence of 151on
intensity o f the Couette component of the flow is rather complex
(with two local minima); however, 5 ^ as Recr °° and this
result makes it probable that finite disturbances are unstable even if
all infinitesimal disturbances are stable (i.e., if Rccr = °°).
Unfortunately, the Stuart-Watson approach based on a specific
expansion in the vicinity o f the neutral stability curve is inapplicable
to fluid flows with Recr = (for example, to plane Couette flow
which is apparently unstable to disturbances of finite amplitude
according to experimental evidence). Therefore in the paper o f
Kuwabara (1967) on nonlinear instability o f plane Couette flow, the
approximate method of Meksyn and Stuart (1951) was used in
combination with the mathematical method of Galerkin. As a result
the “ neutral stability surface” in three-dimensional (fc,Re,/l)-space
was obtained, where A is the amplitude o f the disturbance (measured
in terms o f its mean energy). According to Kuwabara’s result plane
Couette flow is stable for any (infinitesimal or finite) disturbance if
Re < Recr min 45,000. Above this critical value at every fixed Re,
instability appears only in a rather small region of (k, A )-plane. The
crude method used is evidently insufficient for obtaining quantita
tively exact results; nevertheless, one can hope that Kuwabara’s
results are qualitatively correct.
For free plane-parallel flows w ithout rigid boundaries and with a
velocity profile having an inflection point (some examples of such
profiles are shown in Fig. 19) linear instability occurs even with v =
0. Therefore it seemed natural to expect that the nonlinear
development of disturbances in such flows could be explained within
the framework of the mechanics of an ideal (nonviscous) fluid. The
first attem pt in this direction was made by Schade (1964), who
considered the problem o f the determination of the value of
Landau’s coefficient 5 for a flow o f ideal fluid in the whole space

with a velocity profile of the form U(z) = Uq tanh (zlH). Schade
introduced some simplifying assumptions and using them determined
analytically the value of 6 for this flow [see also the discussion of
Schade’s work by Michalke (1965a), and Stuart (1967)]. Since the
value o f 5 so obtained turns out to be positive, the unstable
disturbances in a free shear layer with a hyperbolic-tangent velocity
profile must tend to a finite periodic equilibrium state as t
according to Schade’s theory. This conclusion and some related
results on the qualitative features o f the equilibrium state obtained
similarly by Michalke (1965a) and Stuart (1967) agree satisfactorily
with the findings o f Sato (1956; 1960), Freymuth (1966), Browand
(1966) and other experimenters for a region relatively close to the
origin o f the shear layer (which corresponds to not too great values
of t). The analytical result o f Schade is nevertheless quite doubtful
since some o f his assumptions are apparently incorrect [see the
discussion of this question in the survey lecture o f Michalke (1968)].
It is also worth noting that Schade’s assumption that the basic flow
will not be modified by the disturbance contradicts the exact
solution o f the nonlinear equations o f the neutral stability problem
for a free shear layer with hyperbolic-tangent velocity profile found
by Stuart (1967). Schade’s result disagrees also with that o f Gotoh
(1968) who considered the same problem for a viscous fluid flow and
found an expression for the constant 5 as a function o f the Reynolds
number which does not tend to a finite limit as Re ^ °o (i.e., in the
ideal fluid approximation). However, Michalke has noted (in a letter
to one o f the authors o f this book) that some of the objections to
Schade’s work are valid relative to G otoh’s work too; therefore the
question o f the exact value o f the Landau constant for a shear layer
with a hyperboUc-tangent velocity profile (either in the inviscid case,
or in the viscid one) cannot be considered as solved.
Even more important, however, is the fact that all the data on the
evolution of disturbances in a free shear layer at greater downstream
distances show peculiar behavior which it is impossible to explain
with the aid of Landau’s equation for the amplitude o f a disturbance
for any value o f the coefficient 5. In fact, although near the origin of
the layer only oscillations with the most unstable frequency are
significant, further downstream some other harmonic components
also appear, and still further, the frequency spectrum becomes
continuous and the free shear layer becomes turbulent. Moreover,
when additional harmonic components appear in the spectrum a
remarkable effect is observed. At first, only the higher harmonics of
the most unstable oscillation (henceforth called the fundamental

oscillation), with two and three times the frequency of the
fundamental, are observed. However, somewhat further downstream,
a subharmonic component with half the frequency o f the funda
mental also appears, and shortly after its appearance, it becomes the
most significant harmonic component o f the flow (see, for example,
the excellent data of Sato (1959), and Browand (1966), obtained in
the presence of artificially produced fundamental oscillation). The
simplest nonlinear mechanism of self-interaction of the fundamental
with itself (through quadratic terms o f the dynamic equations)
would explain the existence o f higher harmonics (taken into account
in the Stuart-Watson expansion procedure); however, the origin of
subharmonic oscillation could not be explained in that way.
Theoretically it is possible that the subharmonic is related to the
slow growth of shear-layer thickness with distance downstream;
however, such an explanation does not seem probable. It is more
natural to think that the great difference between the behavior of the
real oscillations in a free shear layer and the predictions of Landau’s
theory could be explained without drawing on the small deviations
of the shear-layer flow from strict plane-parallelism. The transition to
turbulence of the shear layer at great downstream distances leads one
to suppose that the finite-amplitude equilibrium state is unstable
here to small disturbances. Michalke and Timme (1967) attem pted to
verify this supposition using a special vortex model o f the disturbed
shear-layer flow. They studied the inviscid instability of an isolated
cylindrical (i.e., two-dimensional) vortex and found that a single
vortex o f the type which occurs in a shear layer can be unstable with
respect to cylindrical disturbances. This fact explains the breakdown
of vortices forming the equilibrium state of a disturbed shear layer
and the transition to turbulence. However, the Michalke-Timme
model is a qualitative one and their approach is rather crude. A more
realistic theory must deal with exact solutions of the nonlinear
equations of fluid dynamics representing disturbed shear-layer flow,
and study the stability o f these solutions. An equivalent formulation
consists o f a consideration o f the evolution of several different
wave-like disturbances in a free shear layer, taking into account their
mutual interactions, i.e., the effect of each of them on the
development of the rest (the existence o f such interactions is obvious
from the nonlinearity o f the dynamic equations).
As already mentioned in connection with other types of fluid
flows, the question of the interaction of finite disturbances is one of
the least explored in fluid mechanics. Kelly (1967) obtained the first
results which touch upon this interaction for a free shear layer. He
investigated the stability o f inviscid plane-parallel shear flow,
consisting of a nonzero mean component, together with a compo
nent periodic in the direction of flow and with time (such a flow
describes the equilibrium state of a single disturbance in a shear
layer). It was found that the interaction of the periodic component
o f the flow with a disturbance with twice the wavelength and
frequency o f the basic periodic component can produce a wave
which is o f the same wave number and frequency as the disturbance
and which can therefore reinforce it. As a result, the subharmonic
component with half the frequency o f the basic periodic component
begins to grow in full agreement (which is even quantitative in some
details) with the experimental data of Sato (1959) and Browand
(1966). Later, Kelly (1968) also considered some model examples of
interaction o f neutrally stable disturbances in two specific shear
flows. He showed that resonant interaction is possible and increases
the rate at which energy can be transferred from the mean flow to
each disturbance, leading to simultaneous amplification of both
disturbances.
The question of interaction of disturbances seems to be important
also for the analysis of nonlinear processes in a boundary layer on a
flat plate. The hnear analysis for this case was considered in detail in
Sect. 2.8. In particular, we have already pointed out that the
experiments o f Schubauer and Skramstad show the initial develop
ment o f unstable disturbances in such a flow to be in complete
agreement with the deductions of the linear Tollmien-Schiichting-Lin
theory. However, subsequent very interesting experimental observa
tions of several research groups [see, for example, Schubauer and
Klebanoff (1956); Hama, Long, and Hegarty (1957); Klebanoff and
Tidstrom (1959); Klebanoff, Tidstrom, and Sargent (1962); Kovasz-
nay, Komoda, and Vasudeva (1962); Tani and Komoda (1962);
Hama and Nutant (1963); and the review articles by Kovasznay
(1965); Stuart (1965); and Tani (1967)] showed conclusively that
this is the case only in the first stage of development of the
disturbances. After this first stage, the situation changes considerably
and a sequence of events occurs in a definite order finishing with
transition to developed turbulence. The most important in this
sequence o f events are the following: 1) the appearance of distinctly
three-dimensional disturbances; 2) the nonlinear amplification of
three-dimensional waves and the appearance of a streamwise vortex
system; 3) the development of a high-shear layer and the generation
o f high-frequency fluctuations; and 4) the development of random

“ turbulent spots,” the growth and repeated coalescing o f which lead
to transition to turbulence o f the whole boundary layer.
The generation o f weak three-dimensional disturbances which vary
with lateral (“spanwise” ) position can be explained by the influence
o f small spanwise irregularities in the free-stream or in the upstream
boundary layer. For supercritical Reynolds numbers, such three-
dimensional waves may be unstable, and for sufficiently large
R e-R ecr, three-dimensional waves may even become the most
rapidly increasing (i.e., the most unstable). The growth o f unstable
three-dimensional waves will clearly lead an initially two-dimensional
disturbance into a three-dimensional form. The development of
three-dimensionality in the boundary-layer flow is observed best in
“controlled” experiments with a special vibrating-ribbon technique
producing waves that are made spanwise-periodic artificially at the
ribbon position [see, e.g., Klebanoff and Tidstrom (1959); Kleban-
off, Tidstrom and Sargent (1962); Kovasznay, Komoda and Vasu-
deva (1962); Tani and Komoda (1962); Komoda (1967)]. These
experiments show that a three-dimensional wave leads to develop
ment of a whole system of longitudinal vortices with axes directed
along the mean stream (these streamwise vortices are o f course very
different from the two-dimensional spanwise vortices observed when
a free shear layer begins to roll up w ithout losing its two-dimensional
structure). The development o f a streamwise-vortex system leads to a
flow with a very distinct three-dimensional structure and a sharp
redistribution o f the intensity of fluctuation in the spanwise
direction 0 x 2 ~ Oy [see Fig. 24 taken from the paper by Klebanoff,
Tidstrom, and Sargent (1962)]. The time evolution of disturbances
at this stage is nonlinear and its theoretical analysis is impossible
without taking account of the interactions between the different
waves present.
For this purpose Stuart (1962) considered the behavior in a
plane-parallel flow o f a disturbance which is composed o f a
two-dimensional and a three-dimensional wave of the same stream-
wise wave number:
u (x,y,z,t) = A^{t)coskxf^{z) + ^ 2 ( 0 cosfex cos Zy/2 ( 2 ) +
+ small complement.
Using the expansion technique given in Stuart (1960) and Watson

(1960a), Stuart obtained for the amplitudes Ai and >42 a system of
l/'/U
FIG. 24. Dependence o f the relative magnitude
of a typical fluctuation o f longitudinal velocity
i f on the “ lateral” coordinate >»in a boundary
layer on a flat plate at different distances x from
the oscillating metal strip which creates the dis
turbance: O -fo r X = 7.6 cm; A -fo r j: = 15.2 cm;
X -fo r jc = 19 cm.
equations of the form (2.49). Thus in this case there will also exist
steady solutions of the four types (I)—(IV) and the most im portant
question will be the stability o f these steady solutions to some
disturbance. Unfortunately, the quantitative investigation o f stabil
ity, even with respect to only a specific disturbance o f the form
(2.55), requires the determination o f the values o f the coefficients of
the corresponding system of equations (2.49), which is extremely
difficult (we have pointed out earlier in this section, that the
evaluation of the coefficient 6 ,, which is equal to the coefficient 26 in
the corresponding Landau equation (2.41) for a single am plitude/I i ,
was only recently performed for one example of plane-parallel flow,
namely, for a plane Poiseuille flow). However, the data of Fig. 24
create the impression that for certain k and I a mixed steady solution
of type (IV) may prove to be stable, the stability being not only with
respect to disturbances o f the form (2.55), but also with respect to
many other ordinary types o f disturbances.
A similar conclusion may also be reached on the basis o f the

results of Benney (1961; 1964) and Lin and Benney (1962) who
approached the same problem along different lines [this approach is
also presented in Betchov and Criminale (1967) Chapt V III]. These
authors proposed a solution o f the equations of fluid mechanics in
the form o f a series o f disturbances o f increasing order
u ( x , t)==u(^){x)-\-auW{x, + 0+ •••, (2-56)
where a 1 is a dimensionless coefficient, defining the ratio of the

amplitude of the disturbance to the amplitude of the undisturbed
flow «(°)(jp). In actual calculations, in Benney (1961) and Lin and
Benney (1962), the example chosen asuS^'îx) is that o f plane-parallel
flow in an unbounded space, with velocity profile U(,{z) = i/otanh z,
shown in Fig. 19d, while in Benney (1964) it is a flow in the
half-space z > 0 with profile t/o (z) linearly increasing to some value
z = H, and then assuming a constant value Uo(H). The primary
disturbance (jt, t) in both cases is taken to be of the form
«(J)(j*r, = (2.57)
Here Ui(z) = [U i{z), Vi{z)], k and c are found from the ordinary
(“two-dimensional”) linear stability theory (and correspond to a
slightly unstable disturbance at Re somewhat greater than Recr),
Ui{z) = [«i( 2 ), Vi{z), W\(z)] is determined from the linear theory
for small three-dimensional disturbances, and the ratio |j,A describes
the relative role o f two- and three-dimensional disturbances. The
calculations are carried out accurately to second-order disturbances
(of the order ofa^); they show that even with (iA!^>l the interaction
o f two- and three-dimensional disturbances leads to the formation of
secondary longitudinal (streamwise) vortices and to considerable
redistribution o f the energy of the disturbances in the Oy direction.
As a result, the overall motion [described by the three terms on the
right side of Eq. (2.56)] is very close to that which is actually observed
in a boundary layer [which has a completely different velocity profile
U, (z)\.
The elucidation o f the im portant role o f three-dimensional
disturbances in the process o f transition to turbulence in a boundary
layer, prompted Meksyn (1964) to return once again to the problem
o f the stability o f a plane Poiseuille flow with respect to finite
disturbances. Adopting assumptions close to those used earlier in the

work of Meksyn and Stuart (1951), Meksyn carried out a similar
analysis with reference now to finite three-dimensional disturbances
(with velocity field o f the form u { x , 0 = S
where tii and na are arbitrary integers). After *some very cumbersome
approximate calculations, he found that for certain finite three-
dimensional disturbances, the critical Reynolds number Re^p may
certainly be fairly close to 1000, the value given experimentally for
Recrminfor this flow.
The theoretical investigation of the next two stages of transition
to turbulence in a boundary layer, leading to the appearance of
“random ” turbulent spots and subsequent transition to developed
turbulence, is very difficult and has not, at present, proceeded very
far. It is possible that the appearance of such a “spot” is connected
with many elementary disturbances becoming unstable at once, and
as a result, a complex “mixed” regime is set up, possessing a large
number of degrees o f freedom. It is clear, however, that in this case
the development o f an instantaneous velocity profile with a region of
large shear (and associated inflection point) plays an important part.
Such high-shear layers with an inflection point were predicted by
Betchov (1960) as a result o f a superposition of a secondary
stream wise vortex flow on the primary two-dimensional flow; they
were observed by Kovasznay, Komoda, and Vasudeva (1962) and by
many subsequent investigators [see, e.g., Tani (1967), and Komoda
(1967)]. A high-shear layer in a flow field will usually be unsteady;
its thickness and the corresponding drop in the mean velocity may
vary rapidly with time, leading’ to the development, immediately in
front o f the spots, of a region of flow characterized by sharply
increased instability. The fairly simple calculations carried out by
Greenspan and Benney (1963) [within the framework o f linear
disturbance theory] for a model time-dependent shear layer with
velocity profile of the type given in Fig. 19c, but with changing
velocity-difference 2U q = 2Uo(t) and thickness 2H = showed
that the development o f such a layer may lead to violent generation
of small-scale (i.e., high-frequency) fluctuations, the energy o f which
attains high values in a very short time. These results are in good
qualitative agreement with the existing data relating to breakdown
(i.e., the instantaneous generation of intense high-frequency fluctua
tions) and transition to turbulence of a boundary layer flow [see, in
particular. Miller and Fejer (1964)].
The vortices disintegrated from the high-shear layer [the “hairpin

eddies” in the terminology of Klebanoff et al. (1962)] are the
embryo turbulent spots. When traveling downstream (at a speed
greater than the wave speed of the primary unstable disturbance)
they break down into smaller vortices, which again break down into
even smaller vortices, etc. (the observation of this cascade process of
wave breakdown was made by Hama, Long and Hegarty (1957), with
the aid of water-flow visualization by hydrogen bubbles). During this
stage the distribution loses its regular “wavelike” form and transfroms
into a complex spot-like structure. The final stage is the growth of
localized spots and their joining up in a developed turbulent boundary
layer, characterized by the wholly irregular (“random ” ) nature of the
fluctuations of all fluid mechanical quantities in all points and all
moments of time. The investigation of such developed turbulence will
be the subject of all remaining chapters of this book. However, we
shall require a number of results from the mathematical theory of
probability and the theory o f random functions, and we now turn to
a short exposition of these ideas.
MATHEMATICAL DESCRIPTION OF
TURBULENCE. MEAN VALUES AND
CORRELATION FUNCTIONS
3. METHODS OF TAKING AVERAGES. FIELDS OF FLUID

DYNAMIC VARIABLES AS RANDOM FIELDS
3.1 Practical Methods of Averaging and Reynolds Conditions
We have already mentioned that the characteristic feature of
turbulent motion of a liquid or gas is the presence o f disordered
fluctuations o f the fluid dynamic variables of the flow. As a result,
both the spatial and temporal dependence of the instantaneous val
ues of the fluid dynamic fields have a very complex and confused
nature. Moreover, if turbulent flow is set up repeatedly under the
same conditions, the exact values of these fields will be different
each time. Let us return once again to Fig. 1 which shows the depen
dence of certain fluid dynamic quantities in a turbulent flow on
time. We see that all these curves consist of a set of fluctuations of
diverse periods and amplitudes, superimposed upon each other with
out any obvious regularity. The distributions of instantaneous values
of the fluid dynamic variables in space have a similar nature; they
205
constitute a disordered set of three-dimensional fluctuations of di

verse amplitude, wavelength and orientation. Due to this extreme
disorder and the sharp variation in time and space of the fields o f all
the fluid dynamic quantities, in the study of turbulence it is neces
sary to use some method o f averaging which will enable us to pass
from the initial fluid dynamic fields to smoother, more regular mean
values o f the flow variables. These variables may then be investigated
by means of the usual methods of mathematical analysis.
The question o f the definition o f mean values is a delicate one in
the theory o f turbulence, and has a long history. In practice, to
determine the mean values we most generally use time- and
space-averaging over some interval o f time or region o f space. We
may also consider a more general space-time averaging of the
functions/(jCi, X3, t) = f ( x , /)> given by the equation
/(Xi, X2, X3. t) =

00
— J* J " / (-^1 -^2 -^3 ^3> ^ '^) (^ li ^2' ^3>

— 00
(3.1)
Here the overbar indicates averaging and (o (§ , t ) is some weighting

function (usually nonnegative) which satisfies the normalization
condition
00
/ / / / “ (^1’ ^3. ^)d\,d%^d\ 3 dz = \. (3.2)
If the function co is equal to zero outside some four-dimensional

region and takes a constant value within it, then Eq. (3.1) is simple
averaging over a given region o f space-time. Putting <o(|, t ) =
(0 ( I ) 6 ( t ) or (0 (|, t ) = CO( t ) 6 (|), where 6 is Dirac’s delta-function, and
(1) ( I ) and CO( t ) are functions which possess a constant value on some
parallelepiped or segment and are equal to zero outside it, we obtain
space- or time-averaging, respectively. However, it is clear that the
mean value defined by Eq. (3.1), generally speaking, will depend on
the form o f the weighting function co (in particular, when averaging
over some time interval or region of space, it will depend on the
length o f the interval or the form and volume of the region). Thus
Eq. (3.1) gives rise to many different “mean values,” and it is
necessary to discover which of these is the “best.”
MATHEMATICAL DESCRIPTION OF TURBULENCE 20 7
In choosing some particular “averaging rule,” one must first

formulate the general requirements which such a rule should have.
From the viewpoint of the theory o f turbulence, the most im portant
of these general requirements is, o f course, that the application of
this rule to the differential equations o f fluid dynamics will allow us
to obtain sufficiently simple equations for the mean values of the
fluid dynamic variables. Reynolds, who founded the theory o f
turbulence, was well aware o f this fact; he used only the simplest
form of averaging over some time interval, but at the same time he
indicated the natural general conditions which any averaging applica
ble in fluid dynamics must satisfy. However, although not all the
necessary general conditions were formulated correctly by Reynolds
(1894), by slightly refining his postulates, it is easy to arrive at the
conclusion that the following five relationships must be satisfied:
f+ g = -f+ g > (3.3)

a f= a f, if a = const, (3.4)
a = a, if a = const, (3.5)
-§7 = - ^ > where s is x^, x^ or t, ( 3 .6 )
T g = fl (3.7)
At present, conditions (3.3)—(3.7) are generally known as the

Reynolds conditions.
Condition (3.6) may also be replaced by a more general condition
o f commutivity o f the operations o f averaging and going to the limit:
lim /„ = lim /„. (3.6')

n->co n->oo
Substituting into Eq. (3.7) successively g = I, g = Ji, a n d ^ = h' =

h — Ji (we shall always use the prime to indicate the fluctuation of
the corresponding quantity, i.e., deviation from its average value) and
using also Eqs. (3.5) and (3.3) we obtain the following im portant
consequences from the Reynolds conditions:
/= /. / ' = / - / = 0, fh = /k . f h ' = f h ' = 0. (3.7')
It is clear that conditions (3.3), (3.4), (3.5), and (3.6) [or (3.6')]
will be fulfilled for any averaging (3.1) with arbitrary weighting
function (o, satisfying Eq. (3.2). The situation is different with the
more complex condition (3.7). Thus, for example, if we use time- or
space-averaging over some interval, then, strictly speaking, we can
show that this condition will not be satisfied exactly for any choice
o f interval. However, it is not difficult to argue in favor o f the fact
that the averaging interval may be chosen in such a way that this
condition will be satisfied approximately with comparatively high
accuracy. For this it is necessary only that the averaging interval be
large in comparison with the characteristic periods of the fluctuating
quantity /' = / — /, but small in comparison with the periods of the
averaged quantity / [see, for example, Kochin, Kibel’, Roze (1964),
Vol. 2, Chapt. Ill, Part C ] . Reynolds confined himself to this type of
argument; however, at present these qualitative considerations can
hardly be called convincing.
T h e p o s sib ility o f c h o o s in g th e averaging interval to b e in te r m e d ia te b e t w e e n th e p e r io d s

o f th e flu c tu a tin g a n d m ean fie ld s a ssu m es th a t tu r b u le n t m o t io n m a y b e r esolv ed in t o a
c o m p a r a tiv e ly s m o o t h and s lo w ly v arying “ m e a n m o t i o n ” w ith a very irregular “ flu c tu a tin g
m o t i o n ” su p e r im p o s e d o n it, w h ile b e t w e e n th e fr e q u e n c y range c h a r a cteristic o f th e o n e
m o t io n and th a t o f th e o th e r , th ere is a c o n sid e ra b le gap. In o th e r w o r d s, it is assu m ed h ere
th at th e F ou rier tra n sform (w ith r e sp e c t to tim e or th e c o o r d in a te s ) o f th e fu n c tio n f is
d iffe r e n t fr o m zer o in s o m e reg io n c lo s e to z ero and in s o m e h ig h -fr e q u e n c y region (or
region o f h igh w ave n u m b e r s), an d is e q u a l to zer o in th e in ter v e n in g gap b e t w e e n th ese
region s. T h is p ictu re c o r r e sp o n d s m o r e or less to rea lity in th e case o f a n u m b e r o f artificial
tu rb u le n t flo w s set up in th e la b o r a to r y . H o w e v er , in th e c ase, for e x a m p le , o f natural
tu rb u le n t m o t io n in th e e a r th ’s a tm o sp h e r e and in th e sea, it is b y n o m e a n s alw a y s
a p p h c a b le , sin c e a tm o sp h e r ic an d m a rin e tu rb u le n c e o f t e n p o sse ss a w id e c o n tin u o u s
s p e ctru m . _
T h e m o s t lo gical d e d u c t io n fr o m th e id e a th a t th e m e a n valu e f and th e flu c t u a tio n / ' o f
th e fu n c tio n f d iffe r p r i ^ i p a l l y in th eir ch aracteristic p e r io d s (or w a v e le n g th s ) c o n s ists o f
d e fin in g th e m ean va lu e f as th a t part o f th e r e p r e s e n ta tio n o f th e f u n c tio n f (as a Fou rier
integral) w h ic h c o r r e sp o n d s to in te g r a tio n over th e range o f v alu es o f th e co rr e sp o n d in g
variable (f r e q u e n c y or w ave n u m b e r ) w h ic h are less in a b s o lu te value th an s o m e fix e d
n u m b e r po- It is e a sy to see th a t here th e c o n d it io n s ( 3 . 3 ) , ( 3 . 4 ) , ( 3 . 5 ) , an d ( 3 . 6 ) w ill b e
s a tisfie d , s in c e th is averaging is a sp ecial case o f th e averaging d e fin e d b y E q . ( 3 .1 ) .
S im ilarly , th e first tw o c o n d it io n s o f E q. ( 3 .7 ') w ill b e satisfie d . H o w e v e r , th e c o n d it io n
( 3 . 7 ) w ill n o t , g e n erally sp ea k in g , h o ld ; for this to b e sa tisfie d , it is n e c essa ry to im p o s e o n
th e f u n c tio n s f and g s o m e v ery sp ecial c o n d it io n s th a t are in c o m p a tib le w ith th e
a s su m p tio n th a t th eir F o u rier tran sform is e v er y w h e r e d iffe r e n t fr o m zer o [o n th is p o in t,
see th e d e ta ile d in v e s tig a tio n o f I za k so n ( 1 9 2 9 ) , and th e n o t e o f K am p^ d e F ^riet ( 1 9 5 1 ) ] .
We shall m e n t io n fu rth er th a t B ir k h o ff, Kamp(i d e F ^ r ie t, R o tta and several o th e r s have
p u b lish e d a n u m b e r o f p ap ers [r e fe r e n c e s to w h ic h m a y b e fo u n d , for e x a m p le , in K am p ^ d e
F ^ riet’s survey a r ticle ( 1 9 5 6 ) and in R o t t a ’s article ( I 9 6 0 ) ] , w h ic h are d e v o te d t o th e
in v e s tig a tio n o f gen era l “ averaging o p e r a tio n s .” It w as a ssu m ed in th ese w o r k s th a t th e
averaging o p e r a tio n un d er d is c u s sio n sa tisfies th e R e y n o ld s c o n d it io n s ( 3 . 3 ) - ( 3 . 7 ) [or so m e
related c o n d it io n s o f th e sam e ty p e ] e x a c t ly , an d is d e fin e d over a s p e c ific s u b set o f
fu n c tio n a l sp a ce (i.e ., over sets o f fu n c tio n s w h ic h sa tisfy s o m e special c o n d it io n s ). In
certain cases th e results o b ta in e d m a y give a c o m p le t e d e s c r ip tio n (in ab stract algebraical
te r m s) o f all such o p e r a tio n s. H o w e v er , all th e s e in v e s tig a tio n s are o f a fo r m a l m a th e m a tic a l

n ature an d th eir resu lts d o n o t fin d d irect a p p lic a tio n in th e th e o r y o f tu rb u le n c e.
F u rth e rm o re , th e y are n o t e v e n n e c essa ry , sin c e in p r e sen t-d ay tu r b u le n c e th e o r y th e
q u e s tio n o f th e m e a n in g o f averaging is r esolv ed in a c o m p le te ly d if fe r e n t m a n n er, and ,
m o r e o v e r , in such a w a y th at all th e R e y n o ld s c o n d it io n s are e v id e n tly satisfie d (th is,
h o w e v e r , raises th e n e w q u e s tio n o f s a tisfy in g th e e r g o d ic ity c o n d it io n s w h ic h w ill b e
d isc u sse d b e lo w ) .
3.2 Random Fields of Fluid Dynamic Variables

and Probability Averaging
The use of time-, space- or space-time averaging, defined by some

equations o f the form (3.1), is very convenient from the practical
viewpoint, but leads to a great many unavoidable analytical
difficulties in theoretical calculations. Moreover, this type of
averaging has the great disadvantage that the question of the form of
the function w ( | , t ) most suitable for the given problem must be
resolved each time before use. For all these reasons it is desirable in
the theory o f turbulence to avoid the use of this type of averaging
altogether, and to adopt instead some other method of defining the
mean value, a method that has simpler properties and is more
universal. A convenient definition of this type, which we shall use
throughout this book, is found in the probability-theory treatment
of the fields o f fluid dynamic variables in a turbulent flow as random
fields.
The basic feature o f the probability-theory approach (or, more
commonly, the statistical approach) to the theory o f turbulence is
the transition from the consideration o f a single turbulent flow to
the consideration of the statistical ensemble o f all similar flows,
created by some set o f fixed external conditions. To understand the
implication of this, let us consider as an example a particular class of
flows past a right circular cylinder arising in a wind-tunnel. The
fundamental difference between laminar and turbulent flow past
such a cylinder is as follows: for laminar flow, if we place two
identical cylinders in two identical wind-tunnels in similar positions
(or, what is essentially the same thing, repeat the experiment twice
with the same cylinder and the same wind-tunnel) then, at a given
time t after switching on the motor, and at a given point x of the
working section of the tunnel we shall obtain the same value o f
t) [the component o f velocity along the axis Oxi] and likewise
identical values o f the other fluid dynamic variables of the flow
(which, in principle, may in every case be found with the aid o f the
solution of some problem with boundary and initial conditions for
the Navier-Stokes equations). However, for turbulent flow, the effect

o f small uncontrollable disturbances in the flow and in the initial
conditions leads to a situation in which, when an experiment is
performed a second time under practically the same conditions we
shall obtain two different values o f «i {x, t) and the other fluid
dynamic variables. In this case one may therefore introduce the
concept of the “ ensemble of all values o f u i ( x , t ) obtained in all
possible experiments on turbulent flow past a cylinder under given
external conditions” and the value of Ui (jt, i) obtained in any actual
experiment is then considered as one “realization” chosen at random
from this ensemble. ‘
If we now fix the external conditions and repeat the experiment
many times under these conditions, recording each time the value
obtained f o r « i( ^ , then the arithmetic mean of all these values
will, in practice, be fairly stable. In other words, if initially we have a
sufficiently large number of experiments, then when this number is
increased still further, the mean value will usually vary very little,
oscillating about some constant value (this stability o f the mean
value indicates that our collection of similar experiments does, in
fact, constitute a statistical ensemble). In this case the value about
which the arithmetical mean o f Ui (x, t) oscillates is called the
probability mean o f the velocity Ui {x, t) and is denoted by the
symbol t). (Henceforth, we shall always use an overbar to
denote only the probability mean.)
Similarly, the mean values o f all the other fluid dynamic variables
taken over the whole ensemble o f similar experiments, are found to
be stable and for a sufficiently large number of experiments,
generally deviate only slightly from some constant value. O f especial
interest for our purposes is the indicator function x«, (j:, r “ ")>
u" > which is equal to zero if the value o f «i (je, t) is greater than u"
or less than u', and equal to unity otherwise. The number p { u \ u” )
about which the arithmetic mean o f this indicator function oscillates
1 W e m u s t n o t e here th a t in th e c a se o f n atu ral tu r b u le n c e (e .g ., a tm o sp h e r ic tu r b u le n c e ),

th e c h o ic e o f th e set o f sim ilar e x p e r im e n ts w ill p r e se n t greater d if f ic u lt y , b e c a u s e th e
“ e x te r n a l c o n d it io n s ” (in th is case, p rim arily, th e m e te o r o lo g ic a l c o n d it io n s ) c a n n o t b e
r e p e a te d w h e n e v e r w e w ish . H o w e v e r , in th is case a lso, w e are u s u ally a b le t o s e le c t a
n u m b e r o f situ a tio n s in w h ic h all th e c o n d it io n s w h ic h are e sse n tia l for d e te r m in in g a g iv e n
m easu r ed valu e (e .g ., fo r th e m e a s u r e m e n t o f th e w in d v e lo c ity at a h e ig h t o f 2 m , th ere
e x is t th e m e a n w in d v e lo c it y , th e w in d an d th e te m p er a tu re g r a d ien ts and w in d d ir e c tio n )
w ill b e p ractic a lly id e n tic a l. M oreover, th e e n s e m b le o f th e s e s itu a tio n s w ill a lso fo r m a
sta tistic a l e n s e m b le o f “ sim ilar e x p e r im e n ts ” w h ic h w ill b e sim ilar to th e e n s e m b le o f all
p o s sib le flo w s p ast e q u iv a le n t c y lin d e r s in id e n tic a l w in d -tu n n e ls.
MATHEMATICAL DESCRIPTION OF TURBULENCE 211
is evidently equal to the frequency of occurrence o f experiments in

wh i c h t h e val ue o f ui{x, t) satisfies the inequality
u ' ^ u i ( x , t) < m ". This num berp(«', u") is called the probability
that «i (jf, t) will take a value in the range between u' and u". Usually
this number p{u', u") may be represented as an integral from u' tow"
of some nonnegative function p{u) called the probability density
function (or, briefly, the probability density) of«i {x, t). Therefore,
the set o f all u for which p(«) 0, will give the “ set o f possible
values o f <)” o f which we have already spoken; we shall call the
actual value t) observed in one of the experiments a sample
value (or a realization) of the Xi component o f velocity. The fact o f
the existence o f the density p{u) is sometimes expressed in the
following form;
t)< u -\-d u \= p (u )d u ,
where the symbol P{...} denotes the probability of the condition

specified in the braces being satisfied. The probability mean Ui{x, t)
o f u i { x , t ) , therefore, may obviously be expressed in terms of p(«)
with the aid o f the equation
oo
tti ( at, t) = j up{u)du. (3.8)
At the same time, knowledge o f the probability density p{u) enables

us to determine also the probability mean of arbitrary functions of
Ui(x, t):
OO
F \u^(x, 0 1 = / F {u)p(u)du. (3.8')
In probability theory, variables u having a definite probability

density are called random variables; the set of all possible probabili
ties p(m', « " ) = P{u' < u < u " ) corresponding to u is called its
probability distribution.
Thus we may conclude that from the viewpoint o f probability
theory, the value o f the velocity at a point of a turbulent flow is a
random variable described by a definite probability distribution.
So far, we have discussed only the values «i {x, t) of the velocity
component at a fixed point x and at a fixed instant t. However, we
may apply a similar approach to the whole field of values o fu i{ x , t),

i.e., to the function Ui{x,t) = Ui(xu xz, xs, t) of the four variables.
Repeating the same experiment (to establish some turbulent flow)
several times and under the same external conditions, we shall obtain
a new field U i(x,t) every time. Here, also, we may speak o f the
“ensemble o f possible fields u i { x ,t ) . ” Moreover, each individual
field observed in some actual turbulent flow is considered as a
“representative” chosen at random from this ensemble [in other
words, as a sample value or a realization of the random field
u i{x, /)]. It now remains only to consider here how to rephrase the
proposition of the existence of the probability density which we
used for a single value «i {x, i).
For the field Mi {x , t) to be random, it is necessary, first, that the
value Ui (M) = «i (at, t) o f this field at a fixed space-time point M =
( x , t) be a random variable. Hence to every combination o f values of
X and t there must correspond its own probability density Pm{u)
dependent on M = [ x ,t ) . However, this is not all: if we choose two
values t■^) and «i(Al2) = «i(j(f2. ^2) of our velocity
component, the arithmetic mean of any function of these two values
should also be statistically stable. This means that for values
and Ui{Mi) there must exist a two-dimensional probability density
pM^M^ (Ml, M2)defined by the relationship
P {«1 < «1 (M,) < «1 + du^, «2 < «1 (M 2) < «2 + du<i] =
In other words, for the variation o f the values of «i(Mi) and«i(M 2) in

a large number of turbulent flows with the same external conditions,
the proportion of cases in which the value o f «i(Mi) will lie in the
range between and u'[, while at the same time the value of «i(Af2)
lies in the range between «' and must oscillate about some fixed
value [equal to the double integral from u[ to u'[ and from «' to u'!^ of
some nonnegative function “ 2)! • Moreover, if Mi = (;r,, /i),
M z = {X2 < k ) , . . . , Mjv = ( X[^, trf) are N arbitrary space-time points,
then there must exist a corresponding function of the N variables,
pM^M^...Mjv(Ui, U/v). (3 .9 )
defined by the relationship
P {«1 < «1 ( M l X Ml H - « 2 < « l ( ' ^ 2 ) < « 2 + ^^“ 2. •••.

Un Mjv + dUf^] =
= Pm^M^... Myv(«i, «2. • . U^)dUjdU 2 . . . dUff.
This function is the N-dimensional probability density of the values

of the random variables«i(Mi), «i(M2), . . . , «i(Mjv).The existence
o f all possible probability densities justifies considering the field
Ui{x, t) as random; to determine it completely (i.e., to determine the
probability distribution in the functional space of all possible field
values) it is necessary to determine the whole family o f functions
(3.9) corresponding to all possible positive integers N and all possible
choices o f N points of space-time. Thus we may consider the
turbulent flows to be identical if they have identical (one
dimensional and multidimensional) probability densities. Con
sequently, if some set of densities is close to that which describes a
given turbulent flow, then this set will define some approximate
statistical model of the flow.
The functions (3.9) clearly must all be nonnegative and such that
the integral of each over all variables is equal to unity. Moreover,
they must also satisfy some conditions of symmetry and consistency.
Thus by the very definition, the density (3.9) for any Mi, Mz, . . . ,
M n must satisfy the equation
Pm^M^ ... Al^ {Uj, •••) tllî •••>

(3.10)
where iu k, • • •. is any permutation of the integers I, 2, N.

Further, if n < N, then for any N points Mi, M 2, . . . , M„,
Mn+i, . . . , Mn the equation
... «„) =
00 00
~ f ' ' ' f P-'^\ ••• ” •* ^H + l> •••> ^ n ) ^^n+1 •••

—CO —00
(3.11)
must be satisfied. Any family o f nonnegative functions (3.9)

possessing the properties (3.10)—(3.11) and the property that
00
J pj^f{u)du— 1 for all M = {x, t) defines some probability distribu

tion in the space o f functions Ui{M) = u \ ( x , t) o f fou r variables [i.e.,
it defines a random fieldui(M) = Ui(x, 01. The probability mean F
of an arbitrary function F(ui, uz, . . . , un ) of the values
= U2 = Ui{Mi), . . . , Un = Ui(Ms)
will then be defined as the integral

oo oo oo
/ / ••• / “ 2.........
—oo —cxD —oo
... «2- •••. u^)du^du^...du.[^, (3.12)
w h e r e ... Afyv («i, Hj , is the corresponding probability

density.
It is natural to assume that in a turbulent flow the field Mj (at, t),
the fields of the remaining velocity components, and also the fields
of pressure p{x, t), density p(ji:, 0 [in the case of a compressible
fluid], temperature T (jc, t) [when the temperature of the fluid is not
homogeneous] and the other fluid dynamic variables, will also be
random fields. In this case each of these fields will have a
corresponding system o f multidimensional probability densities
(3.9). Moreover, the different fluid dynamic fields in a turbulent
flow are statistically interconnected, and account must be taken that
for these fields there also exist joint probability densities of the
values of one of the fields at some given Nt points of space-time,
values of a second field at given N 2 points, values of a third field at
given N 3 points, etc. Thus it follows that if we have any function of
the fluid dynamic variables of a turbulent flow, we may determine its
mean value as the integral of the product of this function with the
joint probability densities of all its arguments, extended over the
whole range of variation of these arguments [cf. Eq. (3.12)]. Then
the conditions (3.3)—(3.7) will be transformed into well-known
properties o f probability means, the proof of which is given in
textbooks on probabihty theory; therefore, they are satisfied exactly
and require no special justification.
3.3 Concept of Ergodicity. Statistical Formulation of the

Fundamental Turbulence Problem
The approach discussed in Sect. 3.2, which treats the fields of
hydrodynamic variables of a turbulent flow as random fields, was
initiated by the works of Kolmogorov and his school [see, e.g.,
Millionshchikov (1939)] and the work of Kampe de Feriet (1939).
At present, this approach is generally accepted in all investigations on
the theory of turbulence [see, e.g., the special survey articles of
Kampe de Feriet (1953), and Obukhov (1954) and the monographs
by Hinze (1959) and Lumley and Panofsky (1964)]. Adopting the
assumption of the existence of probability distributions for all fluid

dynamic fields, we may further make wide use of the mathematical
techniques of modern probability theory; the operation of averaging
is then defined uniquely and has all the properties naturally required
of it. However, it is essential to note that with this approach, an
important additional question arises concerning the comparison of
theoretical deductions with the data of direct measurements.
According to our new definition, the mean value is understood as
the mean taken over all possible values of the quantity under
discussion. Thus to determine empirically mean values with compara
tively high accuracy we should need results of a large number of
measurements carried out in a long series of repeated similar
experiments. In practice, however, we generally do not have such a
series of experiments, and thus are obliged to determine the mean
values from data taken in the course of a single experiment.^ In all
such cases based on a single experiment, we normally use simplified
averaging of the data over some time or space interval. Thus we see
that the assumption of the existence of probability distributions does
not by itself eliminate the problem of the validity of using ordinary
time or space mean values in the theory of turbulence, but only
alters the formulation of the problem. Instead of investigating the
special properties of particular methods of averaging, we must now
discover how close the empirical mean values obtained by these
methods lie to the probability mean value (the theory is concerned
only with this). The position is completely analogous to that in
ordinary statistical mechanics for systems with a finite number of
degrees of freedom, where the theoretical “ mean over all possible
states of the system” (more often called the “ensemble mean” ) may
also be replaced by the directly observed time-mean. In statistical
mechanics it is well known that such a change is generally made on
the basis of the assumption that as the averaging interval becomes
infinitely great, the time-means converge to the corresponding en
semble means. In certain special cases, the vahdity of this assumption
may be proved strictly (e.g., with the aid of G. D. Birkhoff’s ergodic
theorem) and in all other cases it is adopted as an additional, highly
likely, hypothesis (the “ ergodic hypothesis” ). In the theory o f turbu-
2 In th is r esp e c t, th e o n ly e x c e p t io n is th a t o f tu r b u le n t d if fu s io n e x p e r im e n ts , in w h ic h
a w h o le c lo u d o f id e n tic a l p a rticles is g e n erally released (e .g ., a p u f f o f s m o k e ), an d th e n
“ averaging over th e c lo u d ” is carried o u t . T h is averaging ov er th e c lo u d is in a d e fin ite sen se ,
e q u iv a le n t to averaging o ver a s e t o f sim ilar e x p e r im e n ts .
lence when the averaging interval is made infinitely large, the concept
of the convergence of time or space means to the corresponding
probability mean sometimes is introducedâlso as a special “ ergodic
hypothesis.” With regard to the time mean, the correctness o f this
hypothesis in a number o f cases is supported, in particular, by Lan
dau’s general ideas on the nature of developed turbulence, described
in Sect. 2.9. We note, however, that in several cases, the legitimacy of
replacing the probabihty means of the fluid dynamic fields by space
or time means may also be proved strictly with the aid of the “er
godic theorems” of the modern theory of random processes and
fields. Due to the great importance o f this question, we shall deal
with it in greater detail in Sect. 4.7; at this point, however, we shall
attem pt to formulate the general turbulence problem as a problem of
the probability distributions for the corresponding fluid dynamic
fields.
We have already seen that for laminar motion the fluid dynamic
equations permit a single-valued determination of all the fluid
dynamic variables at any future instant according to the initial values
of the fluid dynamic fields (and the corresponding boundary
conditions). For an incompressible fluid, it is sufficient to know only
the initial values of the velocity field (or the field of the vorticity); in
the case of a compressible fluid, however, the initial values of five
independent fluid dynamic fields must be given (e.g., three com
ponents of velocity, pressure and temperature). In turbulent flows,
the initial values of the corresponding fluid dynamic fields will also
determine, with the aid of the fluid dynamic equations all of their
future values.^ Here, however, these future values will depend
considerably on extremely small, uncontrollable disturbances of the
initial and boundary conditions. Moreover, they will be so complex
and confused in form that their exact determination is useless and
the integration o f the corresponding differential equations practically
impossible. Here, only the probability distributions for the corre
sponding fluid dynamic fields are o f interest and not the exact
3\Ve m u s t n o t e h e r e, h o w e v e r , th a t in th e literatu re o n th e th e o r y o f tu r b u le n c e , it is
s o m e tim e s sta te d th at in a tu r b u le n t flo w th e flu id d y n a m ic e q u a tio n s , in general, are
in a p p lic a b le . I f o n e ig n ores c o m p le t e ly u n ju s tifie d asser tio n s, th e n th e o n ly im p o r ta n t
q u e s tio n h e r e is w h e th e r th e m o le c u la r flu c t u a tio n s c a n c a u se r a n d o m “ sp la s h e s ” c a p a b le o f
tr a n s m ittin g e n e rg y to sm aller-scale flu id d y n a m ic d istu rb a n c es, an d , th u s, for e x a m p le ,
s tim u la tin g tr a n sitio n to tu r b u le n c e . A t p r e se n t, it is a lm o s t w id e ly agreed th a t e v en i f s u ch
p r o c esses are p o s sib le , th eir r ole is, in e v ery ca se , e x tr e m e ly sm all, so t h a t it m a y b e
c o m p le t e ly ig n o r e d t o first a p p r o x im a tio n (s e e b e lo w , th e b e g in n in g o f S e c t. 5 .1 ) .
values. Consequently, for turbulent flows, the fluid dynamic

equations will be used only for investigating the corresponding
probability distributions or of values defined by these distributions.
Further, we note that for it to be possible to apply the equations
of fluid dynamics to random fields defined by their probability
distributions, these distributions must satisfy some regularity condi
tions, which ensure that the realization of the corresponding fields
may be assumed continuous and sufficiently smooth—having all the
space and time derivatives which enter into the dynamic equations.
Let us now assume that the probability distributions referring to
values o f the fields at a fixed initial instant of time t = U satisfy these
regularity conditions. In this case, every actual realization of the
fluid dynamic fields will vary regularly in time in accordance with
the time-variation of the solution corresponding to the given initial
(and boundary) conditions. Consequently, the whole set of possible
initial fluid dynamic fields will have changed after time t > 0 into a
strictly defined set of functions of the space coordinates corre
sponding to the instant t = U + x. Therefore, it follows that the
probability density for any fluid dynamic field at the instant t > U
may be determined (in principle, in every case) from the initial
probability densities. To do this, we need only evaluate with the aid
of the fluid dynamic equations what set of initial conditions will
correspond to one or another range of values of the field at the
instant t, and then find the probability of this set of initial
conditions. Thus, in a turbulent flow, the equations of fluid
dynamics will determine uniquely the evolution in time of the
probability distribution of all the fluid dynamic fields. This means
that a more or less arbitrary choice (taking into account only certain
“regularity conditions” ) may be made of the probability distribution
at only one fixed instant of time; then, all remaining probability
distributions, corresponding to the values of the fluid dynamic fields
at all possible points of space time, will be determined uniquely by
the equations of motion. Consequently, the fundamental problem of
the theory of turbulence (e.g., for the case of an incompressible
fluid) may be formulated as follows. Given the probability distribu
tion of the values o f the three velocity components at different
points o f space at the instant t = to, concentrated on a set of doubly
differentiable solenoidal vector fields, it is required to determine the
probability distribution of the values of the velocity and pressure
fields at all subsequent times (including distributions for values at
several different times). For a compressible fluid, instead of the
probability distributions of the three components of velocity, it is

necessary only to proceed from the probability distributions of the
values of five independent fluid dynamic quantities. Unfortunately,
this general problem is too difficult, and at present no approach to
its complete solution is yet envisaged. Therefore, we shall postpone
further consideration of this problem until the concluding chapter of
the second volume of our book; in the remaining chapters we shall
deal only with more particular problems where instead of probability
distributions, less complete statistical characteristics of random fields
are considered.
3.4 Characteristic Functions and the Characteristic Functional
In m a n y cases in ste a d o f th e p r o b a b ility d e n sitie s ( 3 . 9 ) it is c o n v e n ie n t to c o n sid e r th eir

Fou rier transform s
®2.......®Ar) =
N
= //••• J ^ (“l’ “2.......“at) '^“2 • • •

-oo ( 3 .1 3 )
T hese Fou rier tran sform s are c a lled th e characteristic fu nctions o f th e c o r r e sp o n d in g

p r o b a b ility d istr ib u tio n s; a c co r d in g to E q . ( 3 . 1 2 ) , t h e y m a y a lso b e w r itte n in th e fo r m
N \
(3 .1 4 )
... 02....... 0;v)= ( '■ 2
k —\ J
It is clear th a t th e c h a r a cteristic fu n c tio n d e fin e s u n iq u e ly its co rr e sp o n d in g p r o b a b ility

d istr ib u tio n ; in fa c t
“ 2’ • • • ’ “ at) =
-iLou
— 00
(3 .1 5 )
b y th e w e ll-k n o w n in v e r sio n fo r m u la o f F ou rier integrals. T h u s , giv in g th e c h a ra cteristic

fu n c tio n is e q u iv a le n t to givin g th e c o r r e sp o n d in g p r o b a b ility d e n s ity .
B y d e fin itio n , th e ch a r a cter istic fu n c tio n s are c o m p le x -v a lu e d c o n tin u o u s fu n c tio n s o f
th e a r g u m e n ts 0 | , 0 2 , . . .,0jv, p o s se s sin g th e f o llo w in g p r op erties:
and
(w h e r e th e asterisk d e n o t e s c o m p le x c o n ju g a te ) fo r any integer /i, a n y real 6^/\

O^ând a n y c o m p le x Cu . . In fa c t, th e le ft side o f E q.
( 3 . 1 7 ) is eq u a l to th e m e a n valu e o f th e n o n n e g a tiv e q u a n tity
A=1
w h e n c e f o ll o w s th e giv e n in e q u a lity . It m a y a lso b e s h o w n th a t a n y c o n tin u o u s fu n c tio n o f
N variables p o s se s sin g th e p r o p e r tie s ( 3 . 1 6 ) and ( 3 . 1 7 ) w ill b e th e c h aracteristic fu n c tio n o f
s o m e A^-dimensional p r o b a b ility d istr ib u tio n [w h ic h , o f c o u r se , m a y have n o p rob a b ility
d e n s ity an d c o r r e sp o n d , fo r e x a m p le , t o a d isc re te ty p e ; see B o c h n e r (1933-, 1 9 5 9 ) ] . We
shall r efer to this fa c t in a n o th e r c o n n e c tio n and in th e s e c o n d v o lu m e o f th e b o o k .
It is n o t d if fic u lt to see th a t th e c o n d it io n s o f s y m m e tr y and c o m p a t ib ilit y ( 3 . 1 0 ) and
( 3 . 1 1 ) , a p p lie d t o th e ch aracter istic fu n c tio n s , w ill b e c o m e
........ = 0/,........ V (3-18)
.......= ..............................”)•
T h u s th e r a n d o m fie ld o f a n y flu id d y n a m ic q u a n tit y m a y also b e d e fin e d b y a fa m ily o f
ch ara cter istic fu n c tio n s ( 3 . 1 4 ) sa tis fy in g E qs. ( 3 . 1 8 ) an d ( 3 . 1 9 ) .
F r o m E q. ( 3 . 1 9 ) it is see n th a t th e ch ar acter istic fu n c t io n o f th e p r o b a b ility d istr ib u tio n
o f th e values o f a fie ld o n a giv en set o f N p o in ts d e te r m in e s , in an e x tr e m e ly sim ple
m a n n e r, th e ch ara cter istic f u n c tio n s o f th e valu es o f th e fie ld o n an y su b s e t o f th ese p o in ts .
T h u s it is natural to m ak e an im m e d ia te a t t e m p t t o d e te r m in e all th e p r o b a b ility
d is tr ib u tio n s ch aracterizin g th e fie ld w ith th e aid o f a single q u a n t i t y - “ th e ch aracteristic
f u n c tio n o f th e p r o b a b ility d is tr ib u tio n fo r values o f th e fie ld at all p o ssib le p o in t s .” It is
f o u n d th a t su ch a d e fin itio n o f a r a n d o m fie ld w ith th e aid o f a single q u a n t i t y - “ the
c h a r a cteristic f u n c t io n a l” - i s a c tu a lly p o ssib le (th is is o n e o f th e m o s t im p o r ta n t advan tag es
o f u sin g th e ap p ro a c h b a se d o n c h a racteristic fu n c tio n s in s te a d o f o n th e p r o b a b ility
d e n s ity ) . T h e p o s sib ility o f su ch a d e fin itio n o f ran d om fu n c tio n s w as first p o in te d o u t b y
K o lm o g o r o v ( 1 9 3 5 ) ; sin c e th e n , a n u m b e r o f w o r k s h ave b e e n d e v o te d to th is q u e s tio n ,
b o t h fr o m th e p u rely m a th e m a tic a l as w ell as th e ap p lie d v ie w p o in t s (a m o n g th e la tter,
sp e cia l m e n t io n m u st b e m a d e o f th e im p o r ta n t paper o f H o p f ( 1 9 5 2 ) , w h ic h w e shall
discuss in greater d e ta il in th e s e c o n d v o lu m e o f th e b o o k ) . Here w e shall give o n ly a b r ie f
d e s c r ip tio n o f th e e ss e n c e o f th e m a tte r, w it h o u t d iscu ssin g th e m a th e m a tic a l d eta ils.
F o r s im p lic it y , w e shall c o n sid e r first, in ste a d o f a ra n d om fie ld o f fo u r variables, a
ra n d o m f u n c tio n u { x ) o f a single variable d e fin e d in th e fin ite s e g m e n t x ^ b o f th e a:
axis. T h e f u n c tio n u ( x ) is d e fin e d b y all p r o b a b ility d is tr ib u tio n s fo r th e values u ( x i ) ,
.. . ,/ ^ ( j ^ .\ ) o f th is fu n c tio n over an arbitrary s y s te m o f N p o in ts X i , X 2 , . . .,A /v ,su ch
u { X 2 ) ,
th a t a < AT/^ < 6. We n o w le t the n u m b e r N b e c o m e in f in it e ly gr ea t, c h o o s in g th e p o in ts

in su c h a w a y th a t all th e d is ta n ce s b e tw e e n tw o n e ig h b o rin g p o in ts te n d to z er o , and
c h o o s in g fo r th e p aram eters Ofethe p r o d u c ts o f Xu + \ — Xk an d th e v alu es at th e p o in ts x ^ o f
s o m e f u n c tio n 0(Jt:) d e fin e d o n [a, b]. I f th e fu n c tio n 0 ( j f ) is su ch th at th e in tegra l
h
U[0 (A)] = f e(x)u (X) dx (3.20)
N
e x is ts fo r a lm o s t all r ea liza tio n s o f th e fu n c t io n u ( x ) th e n ^ th e
k=\
^ T h a t is, all r e a liza tio n s e x c lu d in g , p erh ap s, s o m e set o f e x c e p t io n a l r ea liza tio n s, w ith

to ta l p r o b a b ility e q u a l t o zer o .
integral ( 3 . 2 0 ) as N - > o o , T a k in g th e lim it as ooin E q. ( 3 . 1 4 ) w e o b ta in
^ [0 (Jc)] = e x p {iu [0 (a :)]} = exp I / J 0 '( ^ ) u ( x ) d x i . (3 .2 1 )
(D [0 (jc)] is th e va lue o f th e c h a r a cteristic f u n c tio n o f th e r a n d o m variable w [0(;c)] w h e n

th e a r g u m e n t o f th is f u n c t io n is e q u a l to u n ity . T h u s , fo r a g iv e n 0 (-^ ), th is w ill b e s o m e
c o m p le x n u m b e r . E q u a tio n ( 3 . 2 1 ) assigns t o e a ch fu n c tio n 0 (jc ) s o m e c o m p le x n u m b e r ,
i.e ., (D [0(;c)] is a fu n c t io n o f a fu n c tio n o r , as it is u s u ally c a lle d , a W e shall call
th is fu n c tio n a l th e characteristic functio n al o f th e r a n d o m fu n c t io n u ( x ) .
I f w e k n o w th e ch ar a cter istic fu n c tio n a l fo r so m e r a n d o m fu n c tio n « ( j c ) , th e n w e can
d e te r m in e all fin ite -d im e n sio n a l p r o b a b ility d e n s itie s Px , x , ,. . , x y f ( “ l» ^2» • • • ’ ^ n )'
th is, it is s u f f l d e n t t o s u b s titu te as th e fu n c tio n a l argument^© (V) o f t h e fu n c tio n a l d )[0 (jc }]
th e s p e cia l fu n c t io n
6(-«) = 9i8(-^ — + —-«2)+ ••• + ~
w h e r e 0 i, . . . , 0^^ are arbitrary n u m b e r s, an d 6 (jc ) is D ir a c ’s d e lta - fu n c t io n , so th a t E q.

( 3 . 2 2 ) is e q u a l t o z er o e v e r y w h e r e e x c e p t a t th e p o in ts x u X2 , .... Xn
S u b s titu tin g E q . ( 3 . 2 2 ) in t o ( 3 . 2 1 ) , w e o b ta in
$ [9 (X)] = exp j / 2 ^ ) = ’f x........... (01- • • - ^N> (3-23)
5 We n o t e at th is p o in t th a t in th e d is c u s sio n o f th e ch ar a cter istic fu n c t io n a l ( 3 . 2 1 ) it is

n o r m a lly a ssu m ed th a t 0(->c) is a s u ffic ie n t ly s m o o th f u n c tio n (e .g ., a c o n t in u o u s or
c o n tin u o u s a n d w -tim es d iffe r e n tia b le f u n c tio n , n b e in g a g iv e n n u m b e r ). C o n s e q u e n tly ,
str ictly sp e ak in g, in ste a d o f t h e “ im p ro p er f u n c t i o n ’* ( 3 . 2 2 ) w e m u st c o n sid e r a s e q u e n c e o f
s m o o t h f u n c tio n s 0^^^ (jf), « = 1, 2, . . . su ch th a t fo r a n y interval [ a , p] c o n ta in in g n o n e
o f th e p o in ts / = 1, 2 , . . .
lim
n->oo J
f O'"* (X) d x = 0
w h ile fo r a n y s u ffic ie n t ly sm all e > 0 a n d / = 1, 2 , . . . ,
xi^-
Hm J e*"* {X) d x = 0^.
In th is case
lim <S>(ef"' (a:)] = $ [6 (^)],

n->oo
where 0(jc) is th e f u n c tio n ( 3 . 2 2 ) . T h e r e fo r e , th e value o f O[0(Af)] fo r su c h Q{x) w ill

alw a y s b e d e fin e d b y th e c h aracteristic fu n c tio n a l o f s m o o t h fu n c tio n s
In this case the characteristic functional will be transformed into the characteristic function
of a multidimensional probability distribution for u(x\ ), u ( x 2 ), . . u ( x n ) , and the
corresponding probability density may be found with the aid of the inversion formula for
Fourier integrals.
The characteristic functional 0 [0 (x )] possesses the following properties, which are
analogous to the properties (3.16) and (3.17) of the characteristic functions
*[6 W] l eu) =o=l . (3-14)
2 2 W -0/W lc*c;>O (3.25)

*=1 /=1
for any function 0i (Jc),----- - On (x) and complex numbers Ci,.. Cn (the latter property is
called the property of positive definiteness o f the functional (D[0(jc)]). However, the
converse assertion, i.e., a continuous (in some natural sense) positive-definite functional
(D[0 (jc)] which possesses property (3.24) will always be a characteristic functional of some
random function, will be correct only if a considerably more general definition of a random
function is adopted than that used here [see Gel’fand and Vilenkin (1964), Prokhorov
(1961)].
When we consider, instead of a random function of one variable w(x), a random field
Ux(x, t) dependent on four variables, the procedure is completely analogous. Here the
characteristic functional is given by
* [0 (•*. 01 = exp / J J J J 0(JC, t ) u, ( x , t) d X i d X i d x i d t (3.26)
which contains as argument the function 0 (jc, t) of four variables.^

In this case the characteristic functional will define uniquely all the probability
distributions for the field Wi(x, ^)[and will possess properties analogous to Eqs. (3.24) and
(3.25i).
When considering several statistically interrelated random functions of random fields, we
must consider a characteristic functional which is dependent on several functions. Thus, for
example, the velocity field of a turbulent flow U ( x , t) = { u \ ( x j ) , U2 (x, t), Uz{x, t ) ) is
defined uniquely by the characteristic functional
^ [e (X, t ) ] = ^ [0 , (X. t ), 02 (X , t). 03 (X. 0 1 =
(3.27)
: exp I / / / / / s 8ft (JC. t)d x d t
*=i
which is dependent on three functions of four variables. In general, for an A^-dimensional
random field u (Jc) = (jc), . . . , ( x ) j in the space of points JC, we will have
4- [0 (JC)] = « [6j (JC)........... 6 ^ (JC)] = exp . /

I S I ^ ^ ^ ( x ) U j ^ ( x ) d x Y1 (3 - 2 7 ')
/=i J
^To avoid difficulties of convergence of the integrals at infinity, from the outset we may
restrict ourselves to only those functions 0 ( x, i ) which are all identically equal to zero
outside some bounded region of four-variable space.
222 STATISTICAL FLUID MECMANICS
T h e fu n c tio n a l <D[0 (JC)] w ill also p o sse ss p r o p e r tie s ( 3 . 2 4 ) and ( 3 . 3 5 ) [r e p lac in g th e scalar
a r g u m en t 0 ( a') b y th e v e c to r a r g u m e n t 0(jc)]. T h e p r o b a b ility d is tr ib u tio n for th e v e lo c ity
field at a given tim e t w ill b e d e fin e d b y th e fu n c tio n a l
^ [0 ( x l t ] = exp I i / / / s 9* ( X) (Jc, t ) d x (3 .2 8 )
w h ic h is d e p e n d e n t o n th e triad o f fu n c tio n s 0 ( x ) = { 6 j (jc), 02 (JC), 63 (Jc)} o f th ree

variables a n d o n e scalar p aram eter t. In a c c o r d a n c e w ith th e rem arks m a d e at th e b e g in n in g
o f th is s u b s e c tio n o n single-v alu ed d e p e n d e n c e o f th e p r o b a b ility d istr ib u tio n fo r th e
v e lo c ity fie ld o f an in c o m p r e s sib le flu id at every in sta n t o f tim e o n th e p r o b a b ility
d is tr ib u tio n at th e in itial in s ta n t / = 0 , th e ch a r a cter istic fu n c tio n a l (D[0 (a:), m u st, in th e
case o f an in c o m p r e s sib le flu id , b e d e fin e d u n iq u e ly b y its initial value (D[0 (jc), 0 ] and,
m o r e o v er , th e fu n c tio n a l d) [ 0 ( ^ , 0 ] o f Eqs. ( 3 . 2 7 ) m u st also b e d e fin e d u n iq u e ly w ith
resp e c t to cJ)[0(.Jc), 0 ] ) . F or a c o m p r e ss ib le flo w , th e p o s itio n is m o r e c o m p lic a t e d . Here w e
m u st c o n sid e r a fu n c tio n a l o f t h e ty p e
[0 (X), 04 (xi 05 (xh n =

00 p 3
= exp dx
—00
(3 .2 9 )
(p (JC, t ) is th e d e n s ity fie ld , T (JC, t) is th e te m p er a tu re fie ld ), w h ic h is d e p e n d e n t o n five

f u n c tio n s o f th ree variables and o n e scalar argu m e n t. F o r th is fu n c tio n a l th er e m u st b e , in
th e case o f a c o m p r e ss ib le flu id , a sing le-valu ed d e p e n d e n c e at an y in sta n t / > 0 o n th e
c o rr e sp o n d in g initial value (D [ 0 (Jc), 6 4 (JC), 6 5 ( X ) , 0 ].
4. MOMENTS OF FLUID DYNAMIC FIELDS

4.1 Moments and Cumulants of Random Variables
In the previous section we saw that for the complete statistical
specification of the fields o f the fluid dynamic variables of a
turbulent flow we must define all multidimensional probability
distributions for the values of these variables over all possible sets of
space-time points. However, the determination of these multidimen
sional distributions is very complex, and can rarely be carried out
with sufficient accuracy; moreover, the distributions themselves are
often not at all suitable for practical application because of their
awkwardness. Therefore, in practice, one almost always restricts
oneself to a consideration of only the simpler statistical parameters,
which describe some particular statistical property of the flow.
The most important of these parameters are the moments of the
probability distributions. If we have a system of N random variables
ui, «2. • • •, «.v with an A/-dimensional probability density

p {uu «2 , •. •, un ) , then the moments of these variables are defined by
the expressions
CX)
^ / / ••• / '“ 2 ' • • • ( « 1- " 2- • • • - " n) du, d u ^ . . .

(4.1)
where ku k-i,..., kN are nonnegative integers, the sum of which gives

the order of the moment. In particular, the moments o f first order
are the mean values o f the quantities Uu «2 , . . . , «jv.
In addition to the ordinary moments ... syy, it is sometimes
convenient to consider special combinations of them. For example,
we often use the central moment, i.e., the moment of the deviations
of «i, «2, . . . , Un from their corresponding mean values:
... — (W] tti) ' («2 ^ 2)*^ • • • (Wyv • (4.2)
Opening up the brackets in the right side of Eq. (4.2), it is easy to

express the central moment bk^h^...kN terms o f ... kj^, and
ordinary lower-order moments. In particular, for = 1, we have
6i = 0, b‘z — B 2 — B i , 1)3 = Bz — 3 B i B 2 ~ \~ 2 B i , (a^)

b .^ B i- 4 B 1B 3 + 6 BjB 2 - 3Bl ...
(the moment b^ = a'^^ is called the variance of u, and 0^ = 7 ^ is

called the standard deviation of u. Similarly, the general second
central moment b n = (mi — Uj) ( 1/2 ~ « 2 ) is called the covariance
of the variables Ui and ti2 . If« t= 0, < = 1, . . . , N, then the central
moments will coincide with the ordinary moments, so that bk^k^ ...
is a special case of the moment Bk^k^... “ i has some definite
dimensionality, then the corresponding moments or central moments
will also be dimensional; however, the ratios
for example, are always dimensionless. The quantity s is called the

skewness or the skewness factor of the random variable u (or of the
corresponding probability distribution) and 5 is its flatness factor
while the difference (6 —3) is called the excess.
Other combinations of the moments ... of special interest
are the cumulants (or semiinvariants 5*^6^ ...kp,- The general defini
tion o f these quantities will be given a little later (see Sect. 4.2); for
now we shall merely note that the cumulant (like the
central moment ... *yy) is obtained by subtracting from the
moment 5*^*^... a special polynomial in the lower-order moments.
In particular, for A/ = 1, the cumulants of the first five orders are
given by the following equations;
= $2 = 32 — B\ = b2, S3 = Ba — 3 B iB 2 -{-2 B i = bs,
54 = 54 — iBiBz — 3 5 2 + 1 2B\B2 — 65i = b i — 3bl

<Sg— b^ — XQb^b^- (4.5)
In the multidimensional case, the cumulants of the second and third

orders also coincide with the corresponding central moments, while
the general fourth-order cumulant is given by
•^nn = ^1111 îiooôon ^loioôioi ^looiôuo- (4.6)
It will be clear from what follows that in certain cases the cumulants
will be especially suitable to characterize the probability distribu
tions; however, at present, we shall restrict our discussion as already
indicated.
The various moments of «i, «2 , • • •, mjv cannot assume arbitrary
values, but must satisfy certain conditions which take the form of
inequalities. Thus, for example, if all the exponents kx, *2 , • • •, kn are
even, then the moment Bk k ... obviously cannot be negative.
Further, if = 1, then
2 (4.7)
ft = 0 i = 0
where Bo = 1, and co, C|, . . . , c„ are arbitrary real numbers, since the
left side of Eq. (4.7) is equal to the mean value of the nonnegative
n 2
quantity S CtU" . Taking n = 2 and Cq = 0, we obtain, in particular

L*=o
1^31<(52^4)^ and | 5 | < 8 ^ (4.8)
(since bk is a special case of the moment fi/,). Similar inequalities may

also be deduced for higher-order moments and moments of multi
dimensional distributions. Nevertheless, even within these limits, the
arbitrariness in the choice of possible values of the various moments
is still very great; hence the determination of all the moments
without exception provides considerable information about the
corresponding probability distributions.
In many cases, defining all the moments is simply equivalent to
defining the distribution itself (see below, the end of this subsec
tion). Thus, using all the corresponding moments instead of the
probability density does not lead to any loss in completeness of the
statistical description. In practice, however, all of the moments are
never known, and therefore one usually considers only certain
lower-order moments. These, of course, do not provide a single
valued definition of the distribution, but describe only some
particular properties of it. Nevertheless, the approach to the study of
random variables, based on the consideration of only a few of their
lower-order moments, often proves to be very valuable; later, we
shall see that in the theory of turbulence, this approach allows us to
obtain a whole series of results of considerable interest.
It is ea sy t o see th a t th e m o m e n ts o f th e ra n d o m variables U\, . . . . Un m a y b e

e x p r e s s e d sim p ly in te r m s o f th e c o r r e sp o n d in g c h aracteristic fu n c tio n q ) (0 i, . . . , 0 jv ) • In
fac t, it fo llo w s from th e c o m p a r is o n o f Eqs. ( 4 . 1 ) and ( 3 . 1 3 ) th at
, .......M
« . (4.9)
=0;V=O
^ = ^ + *2+ • • • + V
In particular, it fo ll o w s th a t i f th e c h aracteristic f u n c tio n can b e r e p re sen te d b y its T a y lo r
series, th e n
»(•.■«,.......• » ) - s
^1, k^, .... kjsf ...
w-w
T h u s, in th is c a se, i f w e k n o w all th e m o m e n t s o f th e d is tr ib u tio n , w e c a n d e te r m in e th e

c h ara cteristic fu n c tio n (an d h e n c e th e p r o b a b ility d e n s ity ) u n iq u e ly . It is n o t d if fic u lt to
s h o w th a t th e u n iq u e n e s s o f th e d e te r m in a tio n o f th e d e n s ity fr o m th e set o f m o m e n ts w ill
also h o ld w hen th e series (4 .1 0 ) is convergent o n ly in som e r eg io n o f valu es o f
Oi, 0 2 , . . . , 0N [ for th e o n e -d im e n s io n a l case, th e general c o n d it io n fo r th is u n iq u e n e s s
is given , fo r e x a m p le , in th e b o o k o f A k h iez e r ( 1 9 6 5 ) ) .
A lso u sin g th e ch a ra cter istic fu n c tio n , it is ea sy t o fo r m u la te th e g eneral d e fin itio n o f

c u m u la n ts o f ra n d o m variables. F o r th is p u rp o se w e m u st c o n s id e r th e lo ga rith m o f th e
c h a ra cteristic fu n c tio n il?(0 i, 0 2 , Bat) = In (p (0 i, 02, 0^); th e c u m u
lan ts d e fin e d as fo llo w s :
s .. . (4 .1 1 )
■ = ^ dft'i
1 2 ^ =®yv=o
... + V
T a k in g in t o a c c o u n t th a t (p(0, 0, . . 0) = 1, it is n o w e asy to o b ta in E qs. ( 4 . 5 ) - ( 4 . 6 )
and, in general, to e x p r ess a n y c u m u la n t in te rm s o f m o m e n ts or c en tral m o m e n ts .
4.2 Moments and Cumulants of Random Fields

In the theory of turbulence we are concerned with random
fields—random functions u(M) of a point M of four-dimensional
space-time. The Kth-order moments of such a field are the mean
values o f products of K values o f the field
^ u u . . . u { ^ v ^ 2 ....... M/f) = a(M i)a(Ai2). .. (4 .1 2 )
These moments depend on the coordinates of the points at which the

values are taken. Thus a Kth-order moment, generally speaking, is a
function of 4K variables. However, we must keep in mind that some
o f the points Mi, M 2, . . . , Mk may coincide with each other; the
number of different points among them defines the “ type” o f the
moment. In this connection, we shall distinguish moments of the
one-point, two-point, three-point, etc., types (more briefly, one-
point, two-point, three-point, etc., moments). If the type of the
moment is less than its order, then the corresponding moment
[«(M,)1*. l« (A l2 )^ ...|« (iW ^ )l^ ^
will be denoted by the s y m b o l u...u, .... u...« (-î. ^ 2> ■

where the subscript groups corresponding to different points of
space time are separated by commas.
The mean values of the products of values of several different,
statistically related random fields are called foint moments of these
fields. Thus, for example, the field o f the velocity vector u {M) =
{«,(yW), « 2 (^)> will have 3^ different (ordinary and joint)
moments of order K, which together constitute a single three-
dimensional m om ent tensor o f rank K. In particular, we have the
most important two-point, second- and third-order moment tensors

of the velocity field:
Bij (yWi, M 2) = Ui (Ml) UjiM^), ^

Bt,. u (^J.> ^ 2) = ( ^ 2)
(where the tensor ^ is clearly symmetric with respect to the

indices i and /). Similar notation will be used for the joint moments
of other fluid dynamic fields; for example, the two-point joint
moments of the pressure and velocity or the pressure and tempera
ture will be denoted by the symbols M^), / = 1, 2, 3 or
^ 2), respectively. For moments of order greater than their
type, the groups of indices relating to different points will be
separated by commas; for e x a m p l e , p , M 2, Af3)will denote
a three-point, sixth-order moment of velocity, pressure and tempera
ture, containing four components of velocity and forming a
fourth-rank tensor that is symmetric in the pairs of indices i, j and k,
I. The central moments (i.e., the moments of the fluctuations of the
fluid dynamic fields—their deviations from their mean values) for
fields with nonzero mean values, will be denoted in the same way as
the ordinary moments, but with B replaced by b or else with the
addition of “ primes” to the corresponding indices. One-point
moments may also be denoted conveniently by placing a bar above
the relevant symbol (for example, u'>or uv)', also, for the variances
b u u { ^ ) = [u(M) — u(M )Y of the field of «( M) we shall also use
sometimes one of the special symbols or U'^ (we have already used
the latter in Sect. 2; see, for example. Figs. 7 and 24).
When the arguments Mj, M 2, . . . ,Mk are arbitrary points of
four-dimensional space time, we shall call the corresponding mo
ments space-time moments. Very frequently, however, in the theory
of turbulence one considers only moments in which the values of all
the fields refer to the same instant; these are normally called space
moments. Sometimes, also, we deal with time moments—mtan values
of products of values of the fluid dynamic fields at the same point
(but at different instants). Henceforth, when we speak simply of
“ moments” we shall always mean space moments; on the other hand,
if we are discussing time or space-time moments, we shall always
make special mention of the fact.
In this book, we shall often be dealing with correlation functions,

i.e., two-point, second-order moments.’ The correlation function
Buu{Mi, Al2) = u(M i)u(M 2) of the field u(M) is symmetrically
dependent on the arguments Mi and M 2
B„JM „ = M,). (4.14)
Moreover, it possesses the property that
2 (4-15)
i= i y=i
for any nonnegative integer n and any choice of n points Mi, , M„

and n real numbers Ci, . . . , c„, since the left side of Eq. (4.15) is iden-
2
tical to the mean value o f the nonnegative quantity
In particular, with /t = 2, we have the inequality
M,)\ < , (4.16)
which follows from Eq. (4.15). Later, we shall see that any function
Buu{Mi, M 2) which satisfies Eqs. (4.14) and (4.15) may be the
correlation function of some random field (see Sect. 4.3). The joint
two-point moment M2) = u (M i)v(M 2) is often called the
cross-correlation function of the fields u and v. This fimction
satisfies an inequality, analogous to Eq. (4.16)
(4.17)
moreover, it is obvious that
B ,, (A/,, M,) = B ,, (M „ M , ) . (4.18)
If we define (Ml, M2) =Wi(Afi)«/(M 2), where «i(M), Un (M)

are N arbitrary fields, then
O th er te r m in o lo g y is o f t e n e n c o u n te r e d in th e literatu re. F or e x a m p le , in m a th e m a tic a l

w o r k s th e fu n c tio n ®^se th e centered fu n c tio n b n u ( M i , M 2 )y is o f t e n called
th ecovariance (or covariance function) an d th e term c o rr e la tio n fu n c t io n {oi autocorrelation
function) is o f t e n reserved for a c o n e l a t i o n c o e f f ic ie n t bim{MiyM 2 )la ^{M i)o u i^ 2 )'
i i (M,, Mj) CiCj > 0 (4.19)
for any choice of points M^, . . . , M„, real numbers c,, . . . , c„ and
integers , k„ (which take values from 1 to N). Two-point
moments o f order greater than 2 will represent correlation functions
of some new fields which are products of the original fields; such
two-point moments are sometimes called higher-order correlation
functions.
The central two-point, second-order moments
buu (^1- ^^2) = [ « ( ^ 1) — « (^ i)l [« ( ^ 2) — « iî)] =

= {M„ M,) - u (M .)« (M,) (4.20)
and
(yH„ M 2) = [u (Ml) — u (Ai,)] [v {M^) — V (VWj)] =

= M 2) ~ u ( M ~ ) ^ i ^ ) (4.20')
give the correlation functions of the fluctuations of the correspond

ing fields. Sometimes, when there is no risk of confusion, we shall
call them simply correlation functions (in accordance with some
authors). Of course, the correlation functions of fluctuations have all
the properties of ordinary correlation functions. Another extremely
important result is that when we divide the function b u u i^ u ^^2)by
a u (M i)a u (^ 2) or divide 6 ut>(Aî. M2) by au(Aî)ai,(iVf2) we obtain the
correlation coefficient between «(Afi) and «(M 2) or between «(Mi)
and V (M 2); hence the correlation functions of the fluctuations will
become zero whenever the corresponding correlation coefficient is
zero. It is natural to assume that for any fluid dynamic variable u or
pair o f such variables u and v the statistical connection between the
values M(Ml) and m(M2) or « (Mi) and v (M2) [ which is characterized by
the value of the corresponding correlation coefficient] will become
infinitely attenuated as the points Mi and M2 become infinitely far
apart (in space and/or in time). Consequently, the correlation func
tions of the fluctuations of the fluid dynamic fields will always tend
to zero as Mi and M2 become infinitely far apart. This fact defines an
important property of the correlation functions o f fluctuations.
which ordinary correlation functions, generally speaking, do not

possess.
It is well known that for independent random variables the mean
value of a product is equal to the product of the mean values of
individual factors. Since the values of the fluid dynamic variables at
extremely remote points are almost independent, it follows that any
central moments of the fluid dynamic fields (taken, for example, at
points Ml, . . . , jWjv) with order equal to their type, approach zero
when one of the points is infinitely far from the others. However,
when the order of the central moments is greater than their type, this
assertion does not hold. In exactly the same way, central moments of
order / O 4 will not, generally speaking, tend to zero if it is not one
point, but a group of points that is infinitely distant from the others.
At the same time, when, for example, the general central moment is
of the fourth order, it is not difficult to verify that the difference
= M,) (4.21)
will tend to zero whenever the position of the points
Ml, M 2, Ms, Mi changes in such a way that the distance between at
least two of them becomes infinitely great. It may be shown that for
any other moment (central or ordinary, it makes no difference) we
may always choose a combination o f lower-order moments such that
the difference between the original moment and this combination
will tend to zero, when any two of the points on which the moment
depends become infinitely far apart (see below; the text printed in
small type). The corresponding differences between a moment of
order K and specially chosen combinations of lower-order moments
will coincide exactly with the cumulants of the random variables
discussed in Sect. 4.1. Therefore they are called the cumulants (or
semiinvariants) of order K of these fields.
U sin g th e d e fin itio n o f a c u m u la n t given at th e e n d o f S e c t. 4 . 1 , it is n o t d if fic u lt to
s h o w th a t th e c u m u la n ts o f th e flu id d y n a m ic variables o f a tu rb u le n t flo w a c tu a lly d o
p o sse ss th is p r o p e r ty . In fac t, le t us c o n sid e r an arbitrary c u m u la n t
.. =
d^ln<p(0,, 0,.......G,,-)
(4 . 22 )
e, = 9 , = . . . = 0 ; v = o
/< ■= * ,+ ... + k ^ ,
where (p(0i, O2 , 0 n ) is the characteristic function of the values a, (M i ), W2(A^2),

. . . , Wiv ( M n ) o f th e variables Uu ^ 2 , . . . , Un (s o m e or all o f th ese m a y b e id e n tic a l) at
th e p o in ts M i, M 2 , . . . , M k (s o m e o f w h ic h m a y also b e id e n tic a l). N o w le t th e sy ste m o f
p o in ts M l, M 2 , . . . , M iv vary in su ch a w a y th a t th e d if fe r e n c e b e t w e e n at le a st tw o o f th e m
(fo r e x a m p le , b e tw e e n M < a n d M j ) b e c o m e s in f in it e ly great. In th is ca se , th e s y ste m is b o u n d
to d iv id e in t o at le ast t w o s u b s y ste m s, su ch th a t every p o in t o f th e first is in f in it e ly far from
every p o in t o f th e s e c o n d s u b s y ste m (it is s u ffic ie n t , for e x a m p le , to ta k e as th e first
su b s y ste m th e set o f all p o in ts in f in it e ly far fr o m th e p o i n t M j ) . B u t sin c e th e statistic a l
c o n n e c t io n b e t w e e n a n y t w o flu id d y n a m ic variables b e c o m e s in f in it e ly a tt e n u a te d as th e
d ista n ce in creases b e t w e e n th e p o in ts at w h ic h th e variables are ta k en , it f o ll o w s th a t th e N
ra n d o m variables (M i), . . . , (M^^) d iv id e in t o a t least tw o g r o u p s su c h th at th e
variables o f th e first gro u p are fin a lly p ra c tic a lly in d e p e n d e n t o f th e variables o f th e s e c o n d
group. We n o w u se th e fa c t th a t th e ch ara cter istic fu n c tio n o f a s e t o f t w o g ro u p s o f
sta tistic a lly in d e p e n d e n t ra n d o m variables is eq u a l t o th e p r o d u c t o f th e c h a racteristic
fu n c tio n s o f e a ch gro u p sep a ra tely . [T h is fo ll o w s fr o m th e fa c t th a t th e m u ltid im e n sio n a l
p r o b a b ility d e n s ity o f t w o g rou p s o f in d e p e n d e n t r a n d o m variables is e q u a l t o th e p r o d u c t
o f th e p r o b a b ility d e n s itie s for th e s e tw o g r o u p s b e c a u s e o f E q. ( 3 . 1 3 ) . ) T h e r e fo r e , w h e n
th e set o f p o in ts M^ , . . vari es as d e scr ib e d , th e fu n c t io n ? (Oj, ^2 ’ * * ®/y)
te n d s t o d e c o m p o s e in t o th e p r o d u c t cp(0j......... 0^) • 9 ( 6 ^^. , . . . , 0^^, w h e r e n < N .
S u b stitu tin g th is e x p r e s s io n fo r (p^0^, 0^, . . . , 0^j m t o E q. (i.22) and ta k m g in t o a c c o u n t
th a t /Ji 1 fo r all i w e d e m o n s tr a te th at th e c u m u la n t Sj^ ... ^ 2’ ‘
w h e n th e s y ste m o f p o in ts , M^, . . . , varies in th is m a n n er, w ill te n d to zero.
4.3 Random Fields with a Normal Probability Distribution

(Gaussian Fields)
Similar to random variables, the complete determination of the
probability distribution for random fields assumes, generally speak
ing, the determination o f all moments of all possible orders. The only
exceptions in this respect will be cases where some additional
conditions exist on the probability distributions, which enable us to
use certain given moments to determine the rest. We shall consider in
this subsection, a particular but very important case of this kind,
where we may confine ourselves to finding only the first- and
second-order moments. Specifically, we consider the case of a
Gaussian field, i.e., when all the probability distributions of its values
are multidimensional normal (or Gaussian) distributions.
We recall that an A/-dimensional probability distribution is said to
be normal (or Gaussian) if the corresponding probability density
takes the form
N
P(«i. « 2 .........“ yv) = Cexp j — y y] S ' , — flj) . (4.23)
I ;.*=! I
Here Uj, / = 1, . . . , N are arbitrary real c o n s t a n t s ; /, k = \ , . . . ,

N are real constants such that || gju II is a positive-definite matrix (i.e.,
S Sikî^k > 0 for any real Ci, . . . , Cjvnot all equal to zero), and C is
a constant, determined from the condition
1 ^
c f ... fe A d Ui . . . d u i^ = \ (4.24)
(it is not difficult to verify that C — 0 ^ / ( 2 u) \ where C? = | ^ | = det

ll^j/,ll).«
The constants uj and gj*in Eq. (4.23) are simply connected with
the first and second moments of the distribution. In fact, substitut
ing Eq. (4.23) into Eqs. (4.1) and (4.2) it is easy to show that
Uj = a j, = — «/)(«* — = (4.25)
where, as before, G = ||gyft|| and — the co-factor of the

element gjk in the determinant G (so that the matrices || gjk II and
II bjk II are mutually inverse). From Eq. (4.25) we also obtain an
expression for the ordinary (noncentral) second moments of the
distribution (4.23):
G
B j, = u j u , = ^ + a j a , . (4.26)
We see that for a normal probability distribution, the first and

second moments completely define the probability density. Conse
quently, they define, in general, all the statistical characteristics of
the corresponding random variables, and, in particular, all the
higher-order moments. Since the ordinary (noncentral) moments of
any order are expressed simply in terms of the central moments and
the mean values, it is sufficient to consider here only the problem of
the evaluation o f the central higher-order moments. It is not difficult
S G e n e r a lly sp e ak in g, w e m a y a lso c o n s id e r cases w h e n is o n ly a n o n n e g a tiv e
m a tr ix , i.e ., w h e n ^ b e c o m e z er o fo r c ertain n o n n e g a tiv e v alu es o f c i , C / v .
In th is ca se , it is o n ly n e c essa ry t o c o n sid e r th a t th e ra n d o m variables / / i , W a t are lin early

d e p e n d e n t, an d th a t th e p r o b a b ility d is tr ib u tio n w ith d e n s ity ( 4 . 2 3 ) is c o n c e n tr a te d e n tir e ly
in s o m e linear s u b sp a c e o f th e A^-dimensional sp a c e w ith d im e n s io n a lity less th an N . S u ch
p r o b a b ility d is tr ib u tio n s are calle d d e g e n e ra te (or im p ro p er ) n orm a l d is tr ib u tio n s ; h o w e v e r ,
t h e y are n o t o f in ter e st fo r o u r p u rp o se s.
to see that all the central moments of odd order of a normal

distribution will equal zero; for the central moments of even order,
these may be evaluated with the aid of a general rule deduced by
Isserlis (1918). According to this rule, if wu wz, Wzk are 2K
arbitrary random variables (some o f which may be identical), which
have a jointly normal probability distribution with zero mean, then
W1W2 ■ ■ ■ = 2 ■ wi^wi^ . . . (4-27)
where the sum on the right side is extended over all possible divisions
of the 2 K indices 1, 2, . . 2 K into K pairs (ii, t'z), U),
• • • - ihK -v Hk) [it is not difficult to calculate that the number of
terms on the right side of Eq. (4.27) is equal to = 1 •3 •
5 . . .{2K— 1)]. Thus, it follows that with 4-^2+ .. . +kN = 2 K
^*1*2 — (^^1 — — ^ 2) ^ ■ i^N ^ —

—i • • • ^‘2K- 1‘2K’ (4-28)
where the factors j have the same meaning as in Eq. (4.25) and the
sum on the right side is taken over all (2/C)\j2^K\ partitions of the 2K
indices 1, 1, . . . , 1; 2, 2 , , 2 \ . . .; N, N , . . . , N (where 1 is
repeated k^ tim e s,. . ., is repeated /jn times) into K pairs (t'l, *2).
(«3, k),- ■ ■ , ( h K - u t2K). In particular.
îni — (^*1 ^ 1) Uh ^ 2) (^3 ^^3) (^4 ^ 4) —

— ^ 12^34+ ^13^24 ^14^23’ (4.29)
while the central moments of the sixth order will be composed of 15

terms, etc. The method of providing this general rule is shown below.
From a comparison of Eqs. (4.6) and (4.29) it follows that for a
normal distribution S iin = 0. It may be shown that this result has a
very general character; all cumulatants of order /C > 3 of any
multidimensional Gaussian distribution are identically equal to zero
(see below).
We note that the normality of the probability distribution of the
random vector a = ( « i, Uz, . . . , Un ) is an independent property of
the special choice of the coordinate system; this follows directly

from the fact that any linear combination o f normally distributed
random variables will also have a normal probability distribution.
Let us now turn to the Gaussian random fields u{M) or u{M) =
{ui(M), where the probability distributions of any
finite number of its values are normal. As we have shown above, the
complete statistical determination of these fields reduces to the
determination of their mean values and correlation functions. All
other moments may then be found by Eq. (4.28) and the fact that
the central moments of odd order must be identically equal to zero.
It is essential to note that for any (one-dimensional or multidimen
sional) random field with finite moments of the first two orders, it is
always possible to choose a Gaussian field which will have the same
mean and the same correlation function (or functions). For example,
for a one-dimensional field u{M), it follows from Eq. (4.15) applied
to the correlation function of fluctuations of the field 6 „„ (Mi, M 2),
that for any choice of points Mi, Mj, . . . , Mjv an A^-dimensional
normal distribution may be found, with density
(ui, « 2, . . . , Un ), having mean values u{Mi), i = 1. . . . , N, and
second moments fi„„(Mi, Mj) = &utt(Mj, Mj) -F «(M ,)u(M j). Corre
sponding probability distributions for all possible choices of points
will, of course, satisfy the conditions of symmetry (3.1) and
compatibility (3.11), i.e., they will define some random field which
has the same moments of the first two orders as the initial random
field u (M ). This will also apply in the case of a multidimensional
random field u (M) ={«i(M ), . . . , ujv(M)} with the only difference
being that here, instead of Eq. (4.15), Eq. (4.19) must be used.®
Thus using only data on the moments of the first and second orders
in the approximate study o f random fields, we may always assume
that these fields have normal probability distributions. Later, we
shall see that the random fields of the fiuid dynamic variables of a
turbulent flow often in fact do prove to be close to Gaussian in many
respects; we shall make frequent use of this fact.
In th e s tu d y o f norm al v e lo c ity d is tr ib u tio n s it is very c o n v e n ie n t to u se c h aracteristic
fu n c tio n s . It is n o t d if fic u lt t o s h o w th a t in th e c ase o f th e p r o b a b ility d e n s ity ( 4 . 2 3 )
^ S in c e in th is case w e use o n ly th e p r o p e r tie s ( 4 . 1 5 ) and ( 4 . 1 9 ) o f th e c o rr e la tio n

fu n c tio n s , it f o llo w s , in particular, th a t an y fu n c tio n Bnu (M i, M 2 ), or s y ste m o f fu n c tio n s
M 2), p o sse ssin g th e s e p r o p e r tie s w ill b e a c o rr e la tio n f u n c tio n (or fu n c tio n s ) o f
so m e (o n e -d im e n s io n a l or m u ltid im e n s io n a l) r a n d o m fie ld .
9(0,. *2........ V ) = f ■■■ f ........... «/v) ^ “ , • • • =

—oo
N N \
{ k=\ A /^=i
w h e r e th e c o n s ta n ts Oj a n d bjk are th e sam e as in Eq. ( 4 . 2 5 ) [in th e d e d u c t io n o f this

e q u a t io n it is c o n v e n ie n t to u se th e r e d u c tio n to princip a l a x e s o f t h e q u a d r a tic form in th e
e x p o n e n t o f E q. ( 4 . 2 3 ) ] . T h u s th e c h a r a cteristic fu n c tio n o f a n o r m a l p r o b a b ility
d is tr ib u tio n has th e fo r m o f an e x p o n e n t ia l fu n c tio n o f a s ec o n d -o r d e r p o ly n o m ia l in th e
variables 0^, ^2 ’ * • c o n s ta n t term eq u a l to z er o . T h e c o n v e r s e a s sertion is
also true, i.e., to such a ch ar a cter istic f u n c tio n th ere w ill a lw a y s c o r r e sp o n d a p r o b a b ility
d e n s ity o f th e form o f E q. ( 4 . 2 3 ) . ' ®
F r o m Eq. ( 4 . 3 0 ) it f o llo w s th a t th e first- an d sec o n d -o r d e r c u m u la n ts o f th e G aussian
d is tr ib u tio n are e q u a l to c o n s ta n ts a j, / = 1...........N and bjh, /, k = \, . . /V,
r esp e c tiv e ly , w h ile all h igher-order cu m u la n ts are id e n tic a lly equal t o z ero . In a d d itio n , it is
easy to o b ta in from th is e q u a tio n th e general rule fo r ca lc u la tin g th e cen tral m o m e n ts o f
e v en order; for th is p u rp o se, w e m u st u se th e general e q u a tio n ( 4 . 9 ) in w h ic h , in stead o f
cp (Oj, . . . , 0 ^ ) w e m u st s u b s titu te th e fu n c tio n
exp
J.
havin g first e x p a n d e d it as a p o w e r series in 0^, . . . , 0^.

T h e m e t h o d o f cha ra cteristic fu n c tio n s en a b les us to give a s im p le p r o o f o f th e fa ct th a t
a n y linear c o m b in a tio n o f ra ndo m variables havin g a n orm al p r o b a b ility d istr ib u tio n w ill
N
also have a no rm a l d istr ib u tio n . In fa c t, i f Vj = ^ 7 = . • •» th e ch a racteristic
k=i
fu n c tio n o f t h e variables i;., . . v .. w ill o b v io u sly equal
1 M
6^) = e x p | ; 2 ( 2
N / M \ M M
exp 'S 1 (4 .3 1 )
k=\
w h ere <p(Oj, . . . , 0^) is th e ch aracteristic fu n c tio n of Wj, u^. T h e re fo r e , if
lo in fa ct, Eq. (4 .3 0 ) also in c lu d e s characteristic fu n c tio n s o f d e g e n e ra te n o r m a l

d istrib u tio n s; in th is case \\bjh II w ill n o t b e p o s itiv e -d e fin ite b u t o n ly a n o n n e g a tiv e -d e fin ite
m atrix . It is ea sy to see th a t a fu n c tio n o f th e fo rm o f E q. ( 4 . 3 0 ) m a y b e th e characteristic
fu n c tio n o f so m e p r o b a b ility d istr ib u tio n o n ly i f \\b jk II is a n o n n e g a tiv e -d e fin ite m a trix , an d
th u s it f o llo w s th a t th e chara cteristic f u n c tio n s w ill tak e th e fo r m (4 .3 0 ) o n ly
for (n o n d e g e n e ra te or d e g e n e ra te) n orm al d istr ib u tio n s.
cp (0 j . . . 0 ^ ) is given b y Eq. ( 4 . 3 0 ) , th e n ( 0 j ........... 0 ^ )^ w ill also b e an e x p o n e n t ia l

fu n c tio n o f a s e c o n d -d e g r ee p o ly n o m ia l in th e variables 0^, . . 0 ; ^ - In particular, it f o llo w s
that fo r c o m p le t e s p e c if ic a tio n o f th e p r o b a b ility d is t n b u t i o n o f V\, . . Um, it is o n ly
n e cessary to d e te r m in e th eir first- an d s ec o n d -o r d e r m o m e n t s , w h ic h m a y e asily b e fo u n d
from th e first- a n d s e c o n d -o r d e r m o m e n ts o f Wi, . . Un !
T h is a llo w s us t o w r ite in e x p lic it form th e ch ara cter istic fu n c tio n a l o f an arbitrary
G aussian r a n d o m fu n c tio n . First, le t u s c o n sid e r th e r a n d o m fu n c tio n u(x) o f a single
variable jc, d e fin e d in th e interval a ^ x ^ b . A c c o r d in g to E q. ( 3 . 2 1 ) , th e chara cteristic
fu n c tio n a l o f th is fu n c tio n is id e n tic a l to th e v alu e o f th e ch a ra cteristic fu n c tio n o f th e
ran d o m variable
u[fi(x)] = f u{x)Hx)dx
w ith th e a r g u m e n t o f this f u n c tio n eq u al to u n ity . B u t i f t h e p r o b a b ility d is tr ib u tio n s o f a n y

c o lle c t io n o f v alu e s o f th e f u n c tio n u(x) are no r m a l d istr ib u tio n s, th e n all a p p r o x im a tin g
su m s o f th e integral w[0(-^)], w h ic h are lin ear c o m b in a tio n s o f th e r a n d o m variables w (jc^) ,
a < X i < X 2 < . . . < X n < b, w ill ha v e n o r m a l p r o b a b ility d is tr ib u tio n s . T h u s th e integral
w [0 ( ^ ) ] w ill a lso b e a r a n d o m variable w ith a no r m a l d is tr ib u tio n law . T h e r e fo r e , t o fin d an
e x p lic it e q u a t io n fo r d)[0(.Jc)] it is o n ly n e cessary t o fin d th e first t w o m o m e n t s o f th e
ra n d o m variable a [ 0 ( ; c ) ] . T h e se t w o m o m e n t s are g iv e n b y th e e q u a tio n s
u [0(a:)] = J b{x)u{x)dx = J 0 (X) u (X) d x . (4.32)
b b
{u [6(a:)] }“= J f 9(-î) 0(-*^2 ) «(-î) «(-«2 ) d x , dX2
b b
= J J 0(a:,) 0(ATj) B (a:,, x , ) d x , dx^. (4 .3 3 )
T h u s it f o llo w s th at
b b
[u [0(at)]}2- [u [0(>:)]}2= f ®(^2 ) * ^2 ) ‘^■^2 . (4.34)
a a
w h e r e b ( x u X2) is th e c o rr e la tio n fu n c tio n o f t h e flu c t u a tio n s o f th e ra n d o m f u n c tio n w(-jc).

S u b stitu tin g t h e m o m e n ts ( 4 . 3 2 ) and ( 4 . 3 4 ) o f t h e r a n d o m variable u[Q(x)] in t o E q. ( 4 . 3 0 )
w ith N = 1, w e o b ta in
b b
<D[0(^)] = exp i f 0 (jc)« (a :) d x - - ^ I f 0 (Xi) 0 {X 2 ) b {Xi, X 2 ) d x i dXz
(4 .35 )
T his is th e general e x p r e s s io n for th e c h a r a cteristic fu n c tio n a l o f a G aussian r a n d o m

f u n c tio n given in K o lm o g o r o v ’s first n o t e ( 1 9 3 5 ) d e v o te d to th e c h a r a cteristic fu n c tio n a ls .
S im ilarly, th e c h a r a cteristic fu n c tio n a l (D [0(iV f)] (o r ( D [ 0 i ( M ) , 0 , v ( M ) ] ) o f th e
G aussian r a n d o m fie ld w ( ) ^ [or w ( M ) = { m, ( M ) , ( A i ) } ] is e q u a l to
$ [0 (M)] = exp i I ' 0 (M) u {M) d M —
— ^ f f D(Mi)6(Mi)b{M;,M2)dMtdM2 (4 .3 6 )
/• JL
* lO i W ......... 0^(iM )] = exp i I '^ d j{M )IIJ (M )d M ~
;=i
AT j
/ / S ^ 2) d M , d M , \ ; (4 .3 7 )
n aturally, in all cases it w ill b e d e fin e d u n iq u e ly b y th e c o r r e sp o n d in g m e a n va lu e s and

c o rr e la tio n fu n c tio n s .
4.4 Determination of the Moments and Cumuiants of a Random

Field According to Its Characteristic Functional
S in ce th e c h aracteristic fu n c tio n a l c o n ta in s a fu ll sta tistica l s p e c if ic a tio n o f a r a n d o m
fie ld , it w ill also c learly d e fin e all m o m e n ts an d c u m u ia n ts o f this field . We shall n o w d e d u c e
e x p lic it e q u a tio n s c o n n e c tin g th e m o m e n ts an d c u m u ia n ts w ith th e fu n c tio n a l tD[0]; w e shall
th us o b ta in a natural e x te n s io n o f Eqs. (4 .9 ) an d ( 4 . 1 1 ) w h ic h relate to th e fin ite
d im e n sio n a l case.
O n c e again, w e b e gin b y c o n s id e r in g th e r a n d o m fu n c tio n u ( x ) , a < x < b o f a single
variable. S in ce , as w e k n o w , m o m e n ts and c u m u ia n ts o f th e r a n d o m v e c to r u = [ uu .. Uy )
are e x p r e s s e d in te r m s o f th e partial derivativ es o f th e c o rr e sp o n d in g ch ar acter istic fu n c tio n
q )(0 i, . . . , O.v), first, w e ha ve to g en e ra lize th e c o n c e p t o f a derivative to th e case o f
fu n c tio n s o f an in f in it e n u m b e r o f variables, i.e ., o f t h e fu n c tio n a l <l>[0(jc)] w ith r esp e c t to
th e fu n c tio n 0 ( x ) . We recall th at in th e fin ite c a se th e fu n c tio n cp (0^, . . 0 ^ ) = cp (Q) is
said to be d iffe r e n tia b le at th e p o in t e “ = { e ? ............ 6 ^ ) if its in c r e m e n t rfcp =
? (0° + — Y (0 °) fo r a sm all variatio n rf0 = ^0^} of va lu es o f th e
a r g u m en t m a y b e w r itte n as
+ ... + k M )
k=l
(i.e ., w ith a c cu ra c y t o h igh er-order c o r r e c tio n s it is lin early d e p e n d e n t o n rfO). T h e partial

( 00 )
derivative — ----- , m o r e o v er , m a y b e d e fin e d as th e c o e f f ic ie n t Ah o f ^0^ in th e lin ear part
o f th e in c r e m e n t dcp, so th at
Sim ilarly, w e say th a t th e fu n c tio n a lO [0 (A :)] is d iffe r e n tia b le for 0 = Oo(''^), if, w lic n a small
in c r e m e n t 6 0 ( x ) is a d d e d to Oo(jf) th e principal part o f th e in c r e m e n t6 d > [0 o ( a:)] o f this
fu n c tio n a l is lin early d e p e n d e n t o n 6 0 (;c)
[Oo (X)] = <!>[Oo (X) + B0 (X)] - O [0o (x)] =

b r b
= J A ( x ) 86 ( x ) d x + o J I 50 (X) I d x ( 4 .3 8 )
a
(in o th e r w o r d s, i f th ere e x ists a derivative 5^ ^ [Oq (,v)] = [Oq ( ^ ) + Ôi ( ^ ) ] | / 7 = o ’

b
w h ic h m a y b e p u t in th e form BqO [0q ( x ) ] ^ ^ A ( x ) 0i ( x ) d x ) . T h e value o f th e fu n c tio n
A { x ) a t th e p o in t x = X] w ill, in th is case, b e ca lle d f unct i onal derivative o f (D[Oo(a:)| w ith

resp ect to 0(a :) at x =--k ^. T a k in g in to a c c o u n t th a t A ( x ) is th e c o e f f ic ie n t o f M ) ( x ) d x in
th e linear part o f th e in c r e m e n t 6 0 [ 0 o ( a : ) ] , it is c o n v e n ie n t to a d o p t th e n o t a tio n
hHx)d~x ~ ^
for th e fu n c tio n a l d erivative. T h is n o t a tio n stresses th a t th e fu n c tio n a l derivative is a

d o u b le lim it
i^)] _ d>[0o (x)+hO(x)\ - n%±xj] ^

60 ( x ) dx I j 80
w h ere 6 0 (jf) n o w in d ic a te s a fu n c tio n th a t is n o n z e r o o n ly in a small interval o f le n g th A x

su r ro u n d in g th e p o in t x . T h e r eq u ir e m e n t th a t th e fu n c tio n a l (D [0(jc)] b e d iffe r e n tia b le
im p o se s c ertain lim ita tio n s u p o n this fu n c tio n a l w h ic h , fo r e x a m p le , are n o t satisfie d b y so
sim p le a fu n c tio n a l as th e valu e 0 ( a:o) o f th e fu n c tio n 0 (;c ) at th e fix e d p o in t x = Xq. T h e
sim p le st e x a m p le o f a d iffe r e n tia b le fu n c tio n a l is th e fu n c tio n a l
u
4> [0 (a :)] = J A ( x ) f> ( x ) dx.
for w h ic h
b
J A (x )i)(x )d x
a
= A ix). ( 4 .4 0 )
I (jc) d x
In E q. ( 4 . 4 0 ) th e fu n c tio n a l d e riva tive is in d e p e n d e n t o f th e value o f th e
fu n c tio n 0 ( x ) at w h ic h th is derivativ e is ta k en . H o w e v er , in general, th is w ill c er ta in ly n o t be

so, and h e n c e it is m o r e c o rr e c t to w r ite Eqs. ( 4 . 3 8 ) and ( 4 . 3 9 ) in th e form
b
8 $ [00 ( ^ ) ] A [Oo (X); X , ] 86 ( x , ) d x „ I Oo (a-); a:, |.
a
T h u s th e fu n c tio n a l derivative o f th e fu n c tio n a l O [ 0 ( j f ) ] is o n c e again a fu n c tio n a l ofO (Jc),

w h ic h is still d e p e n d e n t o n th e p o in t Xi as a p aram eter. C o n s e q u e n tly , this fu n c tio n a l
d erivative w ill have derivatives o f d o u b le ty p e ; it m a y b e d iffe r e n tia te d w ith r e sp e c t to X\ in
th e usual m a n n e r and m a y a lso h av e its o w n fu n c tio n a l d erivative w ith r esp e c t to 0 (.v ) at the
p o in t X = x -2 w h ic h is th e s e c o n d fu n c tio n a l derivative o f th e original fu n c tio n a l d ) [ 0 ( j : ) ] :
824> (0 ( x ) \
(4.41)
66 (atj) dXi L dx, &6(A:|)rfAr, 80 {x^) dx^
T h e s e c o n d fu n c tio n a l derivative w ill b e a fu n c tio n a l X\, X2], d e p e n d e n t o n th e

pair o f p o in ts X\ and X2 ; h o w e v e r , th is fu n c tio n a l w ill n o t b e arbitrary, b u t w ill s a tisfy th e
c o n d it io n o f s y m m e tr y AYd(x) \ x^,x^= ^ [ 0 ( - ^ ) ;-^ 2 , wh i c h f o llo w s fr o m th e easily
p r ova b le id e n t ity
5^^ [6 (X)] (0 W ]
(4.42)
50 (jc,) d x i 60 (JC2 ) d x 2 ){X2) d X , h H x , ) d x ,
O n e o f th e s im p le st e x a m p le s o f a tw ic e d if fe r e n t ia b le fu n c tio n a l is
b b
a a
in th is ca se , it m a y e a sily b e ve rifie d th at
b b
/ / A {X, x ' ) 0 (AT) 0 (a:') d x dx'
a a
= ^ (ATi. X 2) + A {X 2 . ATi). (4.43)
50 (a:,) d x i 50 ( X 2 ) dX2
F u n c tio n a l d erivatives o f high er orders are d e fin e d in a sim ilar m a n n er; th e n t h
fu n c tio n a l derivative
5^0 [0 (X)]
)(x^)dxi ... U{Xn)dXn
( i f it e x is t s ) , w ill b e a fu n c tio n a l w ith r e sp e c t to 0 (jc ), d e p e n d e n t o n n p o in ts x\, . . Xn-
N o w le t O [0 (J c )) b e th e c h a r a cteristic fu n c tio n a l o f t h e r a n d o m fu n c tio n w(jc), given b y
E q. ( 3 . 2 1 ) . In th is case
J )h =0
o u
i J* u ( x ) 0 ^dx exp i J u{x)${x)dx
(sin c c averaging and d if fe r e n tia tio n are c o m m u ta tiv e o p e r a tio n s ). T liu s it f o llo w s th at
u
0^1^[Q(X)] i u (x ) e x p (4.44)
u(.x) 0 {x)dx
oU ( x ) d x ' /
A n a lo g o u s ly , w e m a y o b ta in t h e m o r e g en era l e q u a t io n
[6 (a:)]
I (a :,) d x ^ . . . Be ( j : „ ) d x „
w h ic h in d ic a te s th a t
if- {x)dix)dx , (4 .4 5 )
Bu...u (^1....... = «(^l) ••• =

[a (a:)]
= ( - 0" h H x,)d x, ... t% {x„)dx„
. (4 .4 6 )
9 U )= 0
E q u a tio n ( 4 . 4 6 ) is a g e n e r a liz a tio n o f E q. ( 4 . 9 ) fo r th e in f in it e d im e n sio n a l c ase.

T h e fu n c tio n a l derivativ es o f th e fu n c tio n a ls 0 [ 0 (x) ] an d d )[6 ( j : ) ] = (D [0i (a :), . . . ,
0iv ( JC)] d e p e n d e n t o n (scalar or v e c t o r ) fu n c tio n s o f a p o in t X o f m u ltid im e n s io n a l sp ace are
d e fin e d in a c o m p le t e ly sim ilar m an n er. T h u s, fo r e x a m p le , th e fu n c tio n a l(D [6 (jc)] is said to
be d iffe r e n tia b le w ith respect to 6/(JC) for a g iven valu e o f th e fu n c t io n 6 (a:) =
{01 ( j c ) , . .., 0 n (-^ )) i f t h e e q u a t io n
A =0
■J (JC): a:i] 6? (JCi) d X i (4 .4 7 )
is s a tisfie d , i.e., i f th e p rin cip al part o f th e in c r e m e n t o f (D[0 (jc)] fo r a fairly sm all ch a n g e

5 6/ ( x ) in th e f u n c tio n 0/ (jc) is lin early d e p e n d e n t o n 8 6 / (jc). In th is case, th e fu n c tio n a l
(4 .4 8 )
w h ic h is d e p e n d e n t o n th e p o in t a :i as a p aram eter is c a lled th e fu n c tio n a l d eriva tive (or,

m o r e p r e cise ly , th e partial fu n c tio n a l d e r ivativ e) o f ^ [0 (jc)] w ith r e sp e c t to 6 / ( j c ) at th e
p o in t x = X i, In o th e r w o r d s, th is deriv ative m a y b e d e fin e d as th e lim it
lim
I6.0(x)l->0
f M (X) dK
Axt
w h e r e 8/6 (JC)is a v e c to r fu n c t io n in w h ic h o n ly th e rth c o m p o n e n t is n o n z e r o , and th a t o n ly
in a sm all n e ig h b o r h o o d AjCj o f th e p o in t X \. T h e fu n c tio n a l d erivative ( 4 . 4 8 ) m a y th e n b e
d iffe r e n tia te d w ith r e s p e c t t o 6y (jc); w e th e n o b ta in th e s e c o n d fu n c tio n a l derivative:
524 . [ 8 (J C )]
(4 .4 9 )
Wi{x,)dx, lbj(x,)dx2 -
w h ic h is d e p e n d e n t o n th e pair o f p o in ts JCi andJC2 , an d w h ic h sa tisfie s th e c o n d it io n
[6 (X)1 [Q(Jt)l (4 . 50 )
B6,- (JC,) d jc , B6y (Jcj) d X 2 B0y (jCj) d x ^ Bfl< (JC,) d x , ■
T h e highcr-order fu n c tio n a l derivatives
[eW ] _______
are d e fin e d sim ilarly. I f O [0 (jc)] is ta k en to b e th e ch ara cter istic fu n c tio n a l ( 3 . 2 7 ' ) o f th e
yV -dimensional r a n d o m fie ld u (x) = [ux ( x ) , . . U n ( x ) ] , th e n b y a p r o c ess c o m p le t e ly
a n a lo g o u s t o th e d e d u c t io n o f Eq. ( 4 . 4 6 ) , w e o b ta in th e e q u a tio n
B (•'I...... (JC,) (X„) :

(4 .5 1 )
[9
••• (•«/.) 0(j»r)=O
(w e n o t e th a t here th e p r o p e r ty ( 4 . 5 0 ) o f f u n c tio n a l deriv atives b e c o m e s th e w e ll-k n o w n
p r o p e r ty ( 4 . 1 8 ) o f c o r r e la tio n fu n c tio n s ).
T h e c u m u la n ts o f th e r a n d o m fie ld u { x ) = [u\{x), ..., « iv (^ )} are n o w d e fin e d
w ith th e aid o f th e f o llo w in g e q u a tio n :
(4 .5 2 )
••• .........^ . . . 60, { x , ) d x „
6(jf)=0
w h e r e ® [8 ( x ) ] is th e c h a r a cter istic fu n c tio n a l o f th e fie ld . It is e a sy t o see th a t
8ln|exp /■ J (jc) , (jc) djc |

00
0; (JC,) dJC,
III (JCi) exp i J Stt, (jc) 6, (jc) d x
exp / J (jc) 0, (jc) d x

0(*)=0
5 Ui (JC,)exp i f Lu,(x)Q,(x)dx
S ,j (x„ JCj) = — ij
^h^j{x,)dx.
exp i J Stt. (jc) 0, (Jc)d x
6 (x)= 0
= «/ (*/) Uj (JC2) — ui(Xi) u j (Xi) = bij (4C„ X i)
and, in general, th a t S /j * * •» id e n tic a l t o th e c u m u la n t o f th e s y ste m o f

variables W/, (X j) , . . de f i ne d b y th e c o r r e sp o n d in g ch ar a cter istic f u n c tio n w ith
th e aid o f Eq. ( 4 . 1 1 ) . F rom E q. ( 4 . 5 2 ) it is ea sy t o d e d u c e all th e p r o p e r tie s o f c u m u la n ts
[p r o p e r ty ( 4 . 5 0 ) o f th e fu n c tio n a l derivative, fo r e x a m p le , m e a n s th a t th e c o rr e la tio n
f u n c tio n o f th e flu c t u a tio n s m u st hav e th e sa m e p r o p e r ty ( 4 . 1 8 ) as th e ordinary c o r r e la tio n
fu n c tio n s ).
F o r a fu n c tio n a l O [6 (jc)] p osse ssin g fu n c tio n a l derivatives o f all orders, un d er c ertain
c o n d it io n s w e m a y o b t a in a T aylo r-series e x p a n s io n
* l« W 1 - * | 0 | + S / «-<*.)■'*.+
i =\
+‘ /=iSy=iS/ /
2
In th e c ase o f th e c h ara cteristic fu n c tio n a l o f th e r a n d o m field U (x) th is e x p a n s io n

b e c o m e s an e x p a n s io n in c o rr e la tio n fu n c tio n s o f all p o ssib le orders
y=i
N N
^jk (^ 1. ^ 2 ) Oy ( X i ) h ( X2 ) dXx d x 2 +
2
y=i /f=i
N N
+ ••• + n l S ••• S /
.-,=1 /„=1
••• ••• (4 .5 4 )
It m a y b e s h o w n th a t for c o n v e r g e n c e o f th e series o n th e right side o f E q. ( 4 . 5 4 ) , it is

n ecessa ry and s u ffic ie n t th a t its /it h term te n d to z ero as /z - > c x d ; i f w e break o f f th is series
after a fin ite n u m b e r o f term s, for th e rem ainder w e m a y o b ta in an e stim a te sim ilar to th e
e stim a te for th e rem ainder o f an ord in ary T a y lo r scries [se e N o v ik o v ( 1 9 6 4 ) ] . We n o te ,
h o w e v e r, th a t i f w e sim p ly break o f f th e e x p a n s io n ( 4 . 5 4 ) o f th e fu n c tio n a l C )[0(;c)] after
any fin ite n u m b e r o f term s, w it h o u t tak in g th e rem ainder in to a c c o u n t, th e n w e arrive at a
fu n c tio n a l w h ic h p o sse sse s p r o p e r ty ( 3 . 2 4 ) o f th e cha ra cteristic fu n c tio n a l, b u t w h ic h
c er ta in ly d o e s n o t p o sse ss th e n ecessary p r o p e r ty ( 3 . 2 5 ) . In fact th is f o llo w s e v en from th e
fact th at a fu n c tio n a l eq u a l to th e sum o f a fin ite n u m b e r o f term s o f th e right side o f Eq.
( 4 . 5 4 ) d o e s n o t satisfy th e s im p le in e q u a lity | ^ [0 (jc)] | <^ 1, w h ic h fo llo w s fro m th e
in e q u a lity ( 3 . 2 5 ) w ith n = 2 (an d also fr o m th e very d e fin itio n ( 3 . 2 7 ' ) o f th e ch aracteristic
fu n c tio n a l 4> [6 (jc)]. T h e re fo r e , a ssu m in g th a t all th e m o m e n ts o f s o m e r a n d om fie ld o f
order h igh er th an s o m e given K are id e n tic a lly eq u a l to z er o , w e fin ally reach 2Lreductio ad
absurdum w h ic h c o n tr a d ic ts , for e x a m p le , th e c o n d it io n th at th e p r o b a b ility is a lw a y s less
than or eq u al to u n ity .
S o m e w h a t m o r e s a tisfa c to r y from this v ie w p o in t is th e a p p lic a tio n o f T a y lo r ’s e x p a n s io n
to th e lo garith m o f th e c h a r a cteristic fu n c tio n a l [0(jc)], i.e., th e r ep re sen ta tio n o f
^ [0 (;c)] in th e form
' N
* [eW] = exp i S/ Uj ( x ) Oy{ X ) d x —
1 7= 1
N N
bjk (JCi. -*2 ) Oy(JCi) e* (X2) dx, dXi +
/=! ft = l
N N
+ --- + fe ^ ••• ^
i,= l /„=1 (4 .5 5 )
... Q,{x„)dx^...dx^ + ...
I f w e c u t Eq. ( 4 . 5 5 ) short after a fin ite n u m b e r o f term s, w e o b ta in a fu n c tio n a l w h ic h

satisfies n o t o n ly Eq. ( 3 . 2 4 ) , b u t also th e n ecessary c o n d it io n for a c h a r a cteristic fu n c tio n a l
I * [0 I< 1. M oreover, i f w e restrict ou r selv e s o n ly to th e linear and qu adratic term s
in 6y(JC), in th e right side o f Eq. ( 4 . 5 5 ) , w e o b ta in a fu n c tio n a l w h ic h c er ta in ly is a
ch a ra cteristic fu n c tio n a l o f s o m e r a n d o m field , e .g ., o f th e G aussian r a n d o m field w ith
m o m e n ts o f th e first an d s e c o n d orders eq u al to th o s e o f th e in itial fie ld tt (Jc) [c f. Eq.

(4 .3 7 )1 . h o w e v e r , w e also tak e in to a c c o u n t th e th ird -order term s in b j { x ) in th e
e x p a n s io n ( 4 . 5 5 ) , i.e., i f w e p u t
N
®[6( jc)] = exp i ^ w y ( j c ) dj ( x ) d x —
I y=i
N N
bjk (^1. X 2) 6y {Xi) h (^ 2 ) dx ^ dX2 —
^ y = i k=i
N N N j
- i S S S / / / bjki {Xi, X 2 , x ^) Oj (X,) (X 2 ) 0/ (X,) dx, d x 2 dx, I. (4 .5 6 )
y = i *=1 / = ! J
or i f w e c u t sh ort th e series ( 4 . 5 5 ) after a n y o th e r fin ite n u m b e r o f term s o f order higher

th an th e s e c o n d inOy (j c ) , th e n w e o b ta in a fu n c tio n a l w h ic h w ill sa tisfy E q. ( 3 . 2 4 ) and th e
c o n d it io n | [0 (jc) ]1 1, b u t w h ic h , gen e ra lly sp ea k in g , d o e s n o t p o s se s s th e p r o p e r ty
(3 .2 5 ) w h ic h is n ecessary for c h a r a cteristic fu n c tio n a ls . T h u s, i f w e a ssu m e th a t all
ou m u la n ts o f th e fie ld U ( x ) o f order greater th an s o m e given / ( > 4 are e q u al to zero, w e
also fin a lly o b ta in a c o n tr a d ic tio n to th e o b v io u s p ro p e r tie s o f p r o b a b ility d istr ib u tio n s
[e .g ., t o th e fa c t th a t th e p r o b a b ility is n o n n e g a tiv e , from w h ic h c o n d it io n ( 3 .2 5 ) f o l l o w s ] .
In V o lu m e 2 o f th is b o o k w e shall have o c c a s io n to return to th is fa c t.
4.5 Stationary Random Functions

Let us now return to the important question (cf. Sect. 3.3) of the
conditions under which the time and space mean values of a random
field u {x, i) will converge to the probability mean value as the
averaging interval becomes infinitely great. We shall obtain some
special classes o f such fields, which are o f particular interest for the
theory of turbulence.
For definiteness, first, we shall discuss time-averaging only; then
the dependence of the field u{x, t) on x will have no significance,
and we can consider only the function u{t) of the single variable t. We
shall be interested in the question of the conditions under which the
random quantity
r /2
« r(0 = T / + (4.57)
-r/2
will converge to u{t) as 7 -^(X 5. “ One very important necessary

condition for convergence is deduced extremely simply. For any
1 ^We are c o n c e r n e d h e r e w ith th e c o n v e r g e n c e o f r a n d om variables w h ic h , in general,

»-“q u ires a s p ecia l d e fin itio n (see b e lo w , th e b e g in n in g o f S e c t. 4 . 7 ) .
bounded function the difference
r/2
tlj (^) Uj' (î) — rf «(^ + t)rfx — I '

- t /2 -r/2
+
’T + '' T+'*
y* « (s) rfs — J u (s) ds , (4.58)
where / and are fixed numbers (and, for example, ti > t) will
become infinitely small as T->- oo. Therefore, the limits as T - ► c» of
Mr(/) and Uriti) [provided that such exist] must become equal, i.e.,
the time mean value of«(/), defined as lim must be independent
of t. At the same time, the probability mean u(t) is, generally
speaking, a function o f t. Consequently, for the two means to be
equal, it is necessary that the following condition hold:
u{t) = U = const. (4.59)

The case is similar when time-averaging is used to determine higher-
order moments and other functions of values o f « ( / ) a t several points.
Thus, for example, if we use the time-mean (with respect to 0 o f the
product u(t)u{ti) = u {t)u { t + s) [where s = ti — t is assumed to be
fixed] to determine the correlation function /j), then we
obtain
T/2
T f s-\-x)d'^, (4.60)
- r /2
which can depend only on s = ti — t, but not on t and ti

individually. Therefore, th e time correlation function (ti t.^ can
equal the probability mean u{ti)u{t 2) = Buu{ti, tz) only if
Buuiiv = (4.61)
In e x a c t l y t he same way, for the A/th-order moment

h, tN) to be definable with the aid of
averaging, it is necessary for this moment to depend only on the N —
1 differences /z — h, . . . , tN — ti-
hi •••, tN) = Buu... u{h — ^1. •••> tN — î)- (4.62)
Finally, if we require that the mean values o f all possible functions of

«(^ 2), ••• , «(^jv)be obtained by means of time-averaging,
then we must confine ourselves to only those random functions
u{t) for which the A/-dimensional probability density ^ ^
(Mj, Mg' • • • > W/v) ^ h, ■ ■ ■ , ( n depends not on
the N parameters ti, tz, . . . . but only on the — 1 differences
ti — 11, . . . , /jv — h, i*e., they satisfy the condition
“ 2 ’ •••> ^ n ) — Pt^+h, t^+h,.... t^+h{.^v “ 2 ’ •••> “at) —
= Pt^- /,...... /, (« p «2- • • • - «;v)> (4-63)
where h is any real number. Condition (4.63) is extremely general, in

particular, conditions (4.59), (4.61) and (4.62) follow from it. We
note, further, that in the special case of a Gaussian random function
u{t), the general equations (4.62) and (4.63) follow from Eqs. (4.59)
and (4.61).
Thus we have obtained a special class o f random functions of time
for which all multidimensional probability densities satisfy Eq.
(4.63), i.e., they do not change when there is a shift of a
corresponding group of points /, through any time-interval h. Such
random functions are often encountered in the most diverse applied
problems; they are called stationary random functions or stationary
random processes (since random functions of time in scientific
literature are also often called random processes). A number of
special monographs or chapters in more general monographs have
been devoted to the mathematical theory of stationary random
functions [see, for example Doob (1953), Loeve (1955), Yaglom
(1962), Rozanov (1967)]. However, we shall restrict ourselves at this
point to a few remarks which have a direct bearing on the subject of
this book (see also Chapt. 6 in Volume 2).
The physical meaning o f the condition o f stationarity is perfectly
clear. It means that the physical process which has the function u(/)
as its numerical characteristic will be steady, i.e., that all the
conditions governing the process will be time-independent. With
reference to turbulence characteristics, the stationarity condition
means that the turbulent flow under discussion will be steady in the
ordinary fluid dynamic sense, i.e., all averaged characteristics o f the
flow (e.g., the mean velocity distribution, the mean temperature) and
also all external conditions (e.g., external forces, position of the

surfaces bounding the flow) will remain unchanged with the passage
of time. Flows which satisfy this conditipn with sufficient accuracy
may be obtained comparatively simply in the laboratory; however, in
the case o f natural turbulent flows, it is usually difficult to ensure
the invariance o f all averaged characteristics o f the flow (this is true
particularly in the case o f atmospheric turbulence, where the mean
values of all variables are usually very unstable and have a clearly
expressed diurnal and annual cycle). However, here too, for the
values of fluid dynamic variables considered over comparatively short
intervals o f time (for example, of several minutes or tens of minutes),
the corresponding random functions may often be considered as
stationary. In all such cases the probability means of the variables of
the flow may often be found by time-averaging; for this to be
possible, it is necessary that the time means asT <x>converge to the
probability mean and that the means taken over time T, in the
course o f which the process may be assumed stationary, will already
be fairly close to the limits corresponding to 7 ^ oo.
The values of any fluid dynamic variable at several points of a
stationary turbulent flow or the values of several such variables at
one or several points provide us with examples of multidimensional
stationary random processes, e.g., of vector functions u{t) ={ui{i),
. . . , Un{t)} such that the probability density for any choice of values
«/, ((2), • • •, (tN) does not vary when all the instants of time
ti, h, iff ars simultaneously shifted through the same arbitrary
time interval h. In this case, all the joint moments o f the functions
Uj(t) will depend only on the differences of the corresponding
instants o f time [for example, all cross-conelation functions
Bjh(U, tz)= Uj (t\) Uh (ts) will depend only on the argument t =
{t2 — /i) ] . It is clear that if the statistical characteristics dependent on
the values of several random functions o f time may be obtained by
means of time-averaging, the set of these random functions must
constitute a multidimensional stationary process; in the opposite case,
characteristics obtained by means o f time-averaging will depend on a
smaller number of variables than do the corresponding probability
means.
4.6 Homogeneous Random Fields
Before proceeding to an investigation of the conditions under
which the time means of functions o f stationary random processes
converge to the corresponding probability means, we shall consider
briefly the question of the changes which must be made in our

discussion if we work on the basis of space-averaging instead of
time-averaging. Here, of course, we must consider random functions
«(Jf) of the point x = {xu xz, X3), i.e., random fields in three-
dimensional space. The space means of such fields will be defined as
Af t B/2 C/2
Â, ~ABC f f f “ (-^1 + ^1’ -^2+ ^2> -^3+ y X

-Aft -B I2 -C /2
(4.64)
We must determine the conditions under which U a , b . c as /1-voo,

B -^ 0 0 , C-ôo(or at least one of these limiting processes occurs) will
tend in some definite sense to the probability mean u ( x ) . For this to
occur, we first require that a condition analogous to Eq. (4.59) will
be satisfied:
« (jc) = i / = const, (4.65)
i.e., the probability mean must be identical at all points of space. In

exactly the same way, if it is possible to determine the correlation
function = x^) with the aid o f space-averaging,
then it is necessary in every case that this function depend only on
the vector difference of its arguments:
•«2) = 'fiaa(-*2 — Jfl)- (4.66)
Finally, if the mean values of all possible functions o f the values of

the field at several points o f space may be obtained by space
averaging, it is necessary that the A'-dimensional probability density
of a(jcj), . . . , u { x ^ ) for any N and x^, . . . , should
depend only on the differences x^ — jc,, . . . , x ^ — jCj, i.e., it should
not change with any parallel displacement of the system of points
Jfi, ....... by addition of the same vector 3) to each:
........ “ 2- • • • ’ H ) = P x , + y , ..............................«2- • • • > « y v ) =
P x ^ - x ^ ........ x ^ - x ^ { . ^ V ^ 2' ' ' '' ^n)' (4-67)
For a Gaussian field u { x ) , Eq. (4.67) follows from Eqs. (4.65) and
(4.66). A field u ( x ) which satisfies the condition (4.67) is called a
statistically homogeneous field (or, briefly, a homogeneous field).

Thus, for space-averaging of any functions of the values of a random
field to lead to the same results as probability-averaging, the field in
question must be homogeneous. When we consider more general
functions of values of several fields a ,(j;),a 2 (•*).• • •. the
requirement is now that the multidimensional (vector) field u (a:)=:
{«, (jc), . . . , (Jc)) should be homogeneous, i.e., that all probability
densities of arbitrary collections o f the values of the components
M, (jc),. •., (jc) of the field u (jc) at some set of points of space will
not vary when all these points are displaced by the same v e c t o r .
With reference to fields o f fluid dynamic variables in a turbulent
flow, the assumption of homogeneity is always a mathematical
idealization, and can never be satisfied. In fact, for us to be able to
speak of homogeneity, the flow should fill an entire unbounded
space, and this postulate, applied to real flows, is always an
idealization. Further, we should require that all the mean values of
the flow (mean velocity, pressure and temperature) should be
constant throughout all space and that the statistical regime of the
fluctuations should not change from one point of space to another.
Of course, all these requirements can be satisfied with sufficient
accuracy only within some finite region o f space which is small in
comparison with the scales of the macroscopic inhomogeneities and
is sufficiently far from all rigid boundaries of the flow (or free
surfaces). Thus, in practice, we may only speak of the homogeneity
of fluid dynamic fields within some definite region, and not
throughout all space.* ^ Nevertheless, when considering turbulent
flow which is homogeneous within such a region, frequently it is
convenient to regard it as part of a fully homogeneous turbulence
filling all space; the value of such a postulate is connected with the
considerable mathematical simplicity of the homogeneous random
field model, which considerably simpUfies the theoretical analysis.
Thus the ergodic theorem [i.e., the theorem on the convergence of
the space means a, ^ (■*) to the probability mean u (jc)] may be
applied to a random field that is homogeneous only within a
bounded region provided that the dimensions of this region are
sufficiently large (cf. the remarks on stationary random functions at
the end of Sect. 4.5).
One very im portant method of artifically creating a turbulent
1 2 a r a n d o m fie ld is said to b e h o m o g e n e o u s in a r eg io n o f sp ace G i f E q . ( 4 . 6 7 ) h o ld s

fo r a n y a n d j; su c h th a t all th e p o in ts x^; ............
lie w ith in G.
flow that is very close to homogeneous will be discussed in Chapt. 7,

Vol. 2, in connection with the study o f homogeneous and isotropic
turbulence. Here we shall only observe that in addition to fields
which are homogeneous throughout all three-dimensional space, we
may also consider fields u ( x ) = u(x\, Xz) that are homogeneous
only in some plane (or along some axis), i.e., satisfying Eq. (4.67) for
all vectors y belonging to a given plane (or axis), but not, generally
speaking, satisfying Eq. (4.67) for other y. The values o f such a field
u (x) in any plane or line parallel to the direction of homogeneity will
clearly form a homogeneous field in this plane or Une. It is natural to
expect too that space-averaging may often be used when the fluid
dynamic fields are not homogeneous in all space, but are homogene
ous only in some plane (or along some line); however, in these cases,
it is necessary to consider averaging, not in three-dimensional space,
but only with respect to the corresponding plane or straight line (i.e.,
the triple integral in Eq. (4.64) must be replaced by a double or
single integral).
In many problems o f the theory of turbulence we may assume
that the corresponding fluid dynamic fields are homogeneous, at
least in one direction, or that they are stationary (or that they satisfy
both these conditions simultaneously). Thus if we could prove that
under conditions of homogeneity or stationarity the probability
means may be replaced by time or space means, this would have very
great practical significance. In fact, however, neither stationarity nor
homogeneity by itself is sufficient to ensure the convergence of the
time or space means to the probability mean values.* ^
However, as we shall now show, the necessary conditions for
convergence have a very general character. Consequently, in applied
problems, they may almost always be assumed to be satisfied.
4.7 The Ergodic Theorem

First, for definiteness, we shall discuss only stationary random
processes u{t) and time-averaging (exactly the same discussion, of
1 3 T o v erify th is a sse r tio n , it is s u ffic ie n t to c o n s id e r th e e x a m p le o f a s ta tio n a r y p r o c e s s
u ( t ) fo r w h ic h , w ith p r o b a b ility e q u a l to u n ity , u { t ) = u { 0 ) fo r a n y t (s o th a t a lm o s t all
r ea liza tio n s o f th e p r o c es s w ill b e rep re sen te d b y straight lin es parallel to th e r-axis). H ere,
c le arly , U t ( t) — ^ ( 0 ) f o r all T, an d i f w (0 ) has s o m e p r o b a b ility d is tr ib u tio n , th e n for th e
m e a n value u { t ) = « ( 0 ) to b e d e fin e d , it is n ecessa ry to ha v e a large n u m b e r o f d if fe r e n t
r ea liza tio n s o f th e fu n c tio n u ( t ) [e a c h o f w h ic h r ed u c es to a u n iq u e v alu e o f i^ (0 )) . In
a p p lie d q u e s tio n s , h o w e v e r , th is w ill a lw a y s b e c o n s id e r e d as an e x a m p le o f a r a n d o m
variable a = « (0 ), but not a ra n d o m fu n c tio n (in c id e n ta lly , th e m u ltid im e n s io n a l
p r o b a b ility d e n s itie s Pt ^t 2 • ord in ary fu n c t io n s in th is
case, b u t w ill b e e x p r e s s e d in term s o f th e D irac 6 - f u n c t io n ) .
course, with the time coordinate t replaced by the space coordinate

X, will apply to a homogeneous random field «(j:) on a straight line).
We begin by defining quite clearly what we mean by the convergence
of the random variables mt(0> defined by Eq. (4.57), as <» to a
constant u(t) = U. It will be convenient to consider random
variables u t to converge, as 7 ^ oo, to a limit U (generally speaking, a
random variable), if
lim Ittr — = (4.68)

r->oo
By Chebyshev’s inequality
it also follows from Eq. (4.68) that
lim P{|Mr — ^1 > e } = 0 , (4.70)

7' - > o o
i.e., that the probability o f the deviation of u t from U exceeding any

given e will tend to zero as T -ô o (and hence can be as small as
desired, provided that the value of T is chosen sufficiently large).
Equation (4.70) gives us sufficient justification for using in
practice the value of Ut instead o f U, where T is comparatively large;
thus it remains only to find under what conditions Eq. (4.68) will
hold.
It is easy to see that the left side of Eq. (4.68) where Mt is the time
mean (4.57) of the values o f a stationary random process u{t),
while i7 = « ( /) - its probability mean value—may be expressed in
terms of the correlation function of fluctuations «(^)
Ka (^) = \ u { t - \ - x ) - U ] [ u (t) - U ] = (x) - t /2 (4.71)
[see Eq. (4.83) below]. Therefore, the correctness of Eq. (4.68)

must be determined by some properties o f the function From
the viewpoint of applications to turbulence theory the following fact
is of fundamental significance: if buu{x) ^ 0 as t-> o o , then the
probab U ity th e o r y , E q. ( 4 . 6 8 ) is g e n erally c a lled th e c o n d it io n o f c o n v e r g e n c e in

q u a d ratic m e a n o f Ut to C/ as 7 -> o o , and E q . ( 4 . 7 0 ) is c a lled th e c o n d it io n o f
c o n v e r g e n c e in p r o b a b ility . T h u s u sin g C h e b y s h e v ’s in e q u a lity , w e h ave sh o w n th a t fr o m th e
c o n v e r g e n c e in q u ad ra tic m e a n o f U t to i / , th ere fo llo w s also its c o n v e r g e n c e in p r o b a b ility
to th e sa m e lim it.
convergence of « r ( 0 to the probability mean value is bound to

occur. This is a corollary of the following general theorem, first
proved by Slutskiy (1938): in order that Eq. (4.68) be satisfied for a
stationary random process u (t), it is necessary and sufficient that the
correlation function b u u ( t ) satisfy the condition
l \ m ^ fb ,,i^ )d x = 0 . (4.72)
r->co J
The proof of this theorem, normally called the law of large numbers
or the ergodic theorem* ^ for stationary random processes, is given at
the end of this subsection. It is clear that if buu{i) ->-0 as oo,
then the condition (4.72) will certainly be satisfied; thus it follows
that in this case, Eq. (4.68) will always be correct. Since the
correlation function buu (t) o f the fluctuations of the fluid dynamic
variable u{t) o f the turbulent flow can always be assumed to tend to
zero as T oo (see above. Sect. 4.2), in the theory of turbulence we
may always take as our starting point the fact that for steady flows
the probabiUty mean values of any fluid dynamic fields may be
determined by averaging over a sufficiently large interval of time.
If the rate of decrease o f buu (x) with increase o f t is such that
then there also exists a simple estimate o f the length T o f those

intervals, averaging over which is sufficient for the mean value U to
be obtained with sufficient accuracy. It may be shown that here,
with sufficiently large T, the asymptotic equation
\Ur-U\^ « 2 ^ (0) (4.74)
holds, where
oo
^ buu (0)
0
f (4-75)
T o a v o id m is u n d e rsta n d in g , w e s h o u ld p o in t o u t th a t in p r o b a b ility th e o r y th e n a m e
“ e rg o d ic th e o r e m for s ta tio n a ry r a n d o m p r o c e s s e s ” is s o m e tim e s g iven t o o th e r p r o p o s i
tio n s, r ela tin g t o th e c o n n e c t io n b e t w e e n th e p r o b a b ility m e a n s and th e tim e m ean s.
is a constant with dimensions of time, which may be called the

“ correlation time” (or the “ integral time scale” ) o f a stationary
function [Eq. (4.74)], and was first obtained by Taylor (1921).
Thus, for a reliable determination of U, iX is only necessary to use
time-averaging over a period T much greater than the corresponding
“correlation time” Ti. After selecting the desired accuracy (i.e., the
greatest permissible mean square error, when u is replaced by ht),
we may use Eq. (4.74) to determine the requisite averaging time T.' *
We will now discuss briefly the homogeneous random fields u (x)
and space-averaging. The case of homogeneous fields on a straight
line does not differ, in general, from the case o f stationary processes.
For homogeneous random fields in a plane or in a space, the values
of such a field on any straight line will form a homogeneous random
field on that line. Thus, provided the correlation function (r) of
the fluctuations o f this field is such that for at least one direction
(unit vector) to the function buu('cro) o f r satisfies Eq. (4.72) [e.g., if
buu{r) 0 as |.r | ^ oo in at least one direction], then the space mean
(or plane mean) will certainly converge to a constant u{x) = (J (this
convergence will hold even for averaging along a single parallel to To).
The general necessary and sufficient condition for the convergence in
quadratic mean of «a,b,c (at) for all Jf to as /I - ► cxs, B -v o o , C ^ cxd
for homogeneous fields in space, takes the form
lim f f f b,Ar)dr,dr^dr:, = 0 (4.76)

» -> c o , C -» o o ^
(the changes necessary for homogeneous fields on a plane are

obvious); the proof o f this condition differs little from the analogous
proof for the one-dimensional case. Instead of Eq. (4.74) we shall
now have
[ U y -U Y ^ 2 ^ b M (4.74')
where uv is the mean over the volume ABC = V, and Vi is the

“ correlation volume” (or “ integral volume scale” )
* *W e m a y n o t e h e r e th a t th e right side o f E q. ( 4 . 7 4 ) is c o m p le t e ly a n a lo g o u s t o th e
e x p r e s s io n for th e m e a n square error o f th e a r ith m e tic m e a n o f N in d e p e n d e n t
m e a s u r e m e n ts, i f w e p u t N = T I 2 T \ . T h u s , averaging a sta tio n a ry f u n c tio n w it h c o r r e la tio n
tim e T i ov er a p e r io d 2 NT\ is e q u iv a le n t t o averaging o ver N in d e p e n d e n t m e a s u r e m e n ts o f
th is fu n c t io n , i.e ., ov er sam p les.
CO (JO oo
VÎ = ‘'Bu(0) / /
0 —0 0 —00
f Ku (r) dr, dr, dr„ (4.75')
which is assumed finite in Eq. (4.74').

The results above may also be applied to the calculation of
higher-order moments or other probability rhean values of any
functions of values of « ( 0 or «(;«:) using time- or space-averaging.
Here we only have to replace the process «( / ) or the field u { x )h y a
new process or field which is a nonlinear function o f the original one.
For example, in the case of a stationary process u(t), for it to be
possible to obtain the A^th-order moment (4.62) by averaging with
respect to time t , the variables u(ti -I- x) u{t 2 -F t) ... m(/jv -I- t), we
require only that the 2 A^th-order moment
B u .,. uu ... u { ^ 2 •••> ^ 1' ^2 ^1 “ 1“ •••>
= a(/i) . . . + t) - [u(t^). . . u(t^)]^ (4.77)
for fixed ti, h, . . . , /jvbe a function of t, satisfying Eq. (4.72). Since

it is normally asserted on physical grounds that the correlation
coefficient between u ( ^ ) . .. m(/n) and u(ti + ' : ) . . . m(/n + "t) tends
to zero asT-)- o o , then, in practice, the necessary condition generally
may be assumed to be satisfied, and the use of time-averaging will be
justified. To estimate the necessary averaging time, it is necessary to
estimate the corresponding “ correlation time,” i.e., the integral of
the correlation coefficient. For Gaussian processes u( / ) t he moment
(4.77) may be expressed in terms of the correlation function and the
mean value of the process «(/), using the general rule (4.28) for
evaluating higher m oments. In particular, in the determination of the
mean square u^(t) = 6 u u ( 0 ) by time-averaging for a Gaussian
random process u( Owi t h zero mean, the role of the correlation
function &„„(/) will be played by
[„2 (/ + x) _ (^)j [„2 _ „2 2 b l (x). (4.78)
Thus Eqs. (4 .7 4)-(4 .7 5) here take the form
uu 0
We note that in this case, from the condition
T- ôo
[obtained from Eq. (4.72) by substituting for 6 „„(t)fro m (4.78)],

the initial condition (4.72) also follows. Moreover, for a Gaussian
stationary random process, it will follow from Eq. (4.80) that the
probability mean value of any function o f the process values (having
finite probability mean) will equal the limit as T -^ o o of the
corresponding time mean value for time T [see, for example,
Grenander (1950)]. A generalization of this last result to Gaussian
homogeneous random fields (and to some even more general random
functions) was formulated in Tempel’man’s note (1962); some
weaker results on such random fields were probed in the paper by
Birkhoff and Kampe de Feriet (1962).
T o prove th at i f th e c o n d it io n ( 4 . 7 2 ) is satisfie d E q. ( 4 . 6 8 ) m u st h o ld , w e e x p r e s s the

m e a n square o f th e d if fe r e n c e u t (0 — ^ term s o f th e c o rr e la tio n fu n c tio n 6 u u ( t ) :
( TI2
\ a j . - U \ ^ = ~ \ f [u{t)-U]dt
l - r /2
r /2 TI 2
= ^ f f [ u ( . t ) - U ] [u ( s ) - U ] d t d s
- r /2 - r /2
r /2 r /2 r t,
= ^ / / b ^ ^ {t-s )d td s = ~ J J b^^{x)dzdx^ (4.;8 1 )
-f/2 -f/2 0 0
[h aving m a d e th e s u b s titu tio n s t ~ s = x, / + s = Ti an d ta k in g in t o a c c o u n t th at
b u u { — x) = b u u ( r ) ] . I f b u u ( r ) sa tisfie s E q. ( 4 . 7 2 ) th e n for a n y 6 > 0 th er e w ill e x ist
To = T o ( 6 ) su c h th a t
f baa (■') for t, > T^. (4 .8 2 )

0
O n th e o th e r h a n d , s in c e l^ u « ( t ) | u ( 0 ) , fo r a n y t i
*1
(x ) dz < buu ( 0 ) Tj. (4 .8 3 )
0
Integ ra tin g E q. ( 4 . 8 3 ) w ith r e s p e c t t o Ti fr o m zer o to T q an d E q. ( 4 . 8 2 ) fr o m To t o s o m e

T > To, fo r a n y T > To w e fin d th a t
Jf Jr . (. t ). . .r f x , < - i>uumTt
- 2- - - ,\- - - - ^ - - - . (4 .8 4 )
0 0
C o n s e q u e n tly ,
T T.
4 ’
0 0
i.e ., for
r>]/ 7.^.
V
T h u s , fo r a n y 6 > 0^ fo r T > To an d T > j^
^ ''l
\Uj. — UY = - ^ f J bau (^ ) d i dZi < 5 . (4 .8 5 )
0 0
i.e., E q. ( 4 . 6 8 ) w ill h o ld .
I f th e lim it s h o w n o n th e le ft side o f E q. ( 4 . 7 2 ) e x is ts b u t is eq u a l to c = ^ 0 , th e n E q.
( 4 . 7 2 ) w ill b e satisfie d b y th e fu n c tio n b i ( x ) = 6 u u ( t ) — c . T h e r e fo r e , in th is case it is
clear th a t
im [Wy, — ^/]2 = ^ l i m f f baui'^) dz dz^^

->oo T-^ co ^ ^
T 1, r T,
f f Ibi (x) + c] dz dz^=^Vm^-^ f f + c=

0 0 0 0
i.e., U t d o e s n o t c o n v e rg e to C7 as 7 - > c x d . It m a y b e s h o w n , h o w e v e r , th a t
lim ~ I baa('^)dt
r->oo
0
for th e c o rr e la tio n c o e f f ic ie n t b u v ( r ) w ill a lw a y s e x is t (s e e , for e x a m p le , C h ap t. 6 , V o l. 2

o f t h i s j j o o k ) . C o n s e q u e n tly , it fo ll o w s th a t w h e n c o n d it io n ( 4 . 7 2 ) b reak s d o w n , th e tim e
m ean s U t w ill n o t con v e rg e to t h e co rr e sp o n d in g p r o b a b ility m e a n u { t ) = U asT - > o o .
I f th e c o r r e la tio n fu n c tio n ^ u u ( t ) satisfie s n o t o n ly E q. ( 4 . 7 2 ) b u t a lso th e stronger

c o n d it io n ( 4 . 7 3 ) , th e n E q . ( 4 . 7 4 ) f o llo w s e asily fr o m E q. ( 4 . 8 1 ) .
We n o t e th a t fr o m E q. ( 4 . 6 8 ) , gen e ra lly sp e ak in g, it still d o e s n o t fo ll o w th a t fo r a n y
r e a liza tio n o f th e r a n d o m p r o c e ss th e tim e m e a n s U j w ill a c tu a lly te n d , as ^ ^ o o , t o a
lim it w h ic h is id e n tic a l t o u { t ) = U. E q u a tio n ( 4 . 6 8 ) w o u ld n o t b e c o n tr a d ic te d , i f for an y
r ea lizatio n w e s o m e tim e s e n c o u n t e r e d values o f T as large as d esired fo r w h ic h U t w as
d ista n t fr o m U (s o th a t for in d ivid u a l r ea liza tio n s t h e lim it
lim
T- >oo
sim p ly d id n o t e x is t ). H o w e v e r , it m a y b e s h o w n th a t fo r a n y sta tio n a ry p r o c ess w(^) w ith

\u(t) \ < CO fo r a lm o s t all r ea liza tio n s (i.e ., fo r a ll r ea liza tio n s e x c e p t perhaps s o m e
“ singular” r e a liza tio n s h a vin g t o t a l p r o b a b ility e q u a l t o z e r o ) th e v alu es o f t h e tim e m e a n U t
w ill te n d to a d e fin ite lim it as T - > c o . (T h is is calle d th e B ir k h o ff-K h in c h in erg o d ic
th e o r e m ; see , fo r e x a m p le , D o o b ( 1 9 5 3 ) , C h a p t. X I, S e c t. 2 , R o z a n o v ( 1 9 6 7 ) , C h ap t. IV ,
S e c t. 5 .) B u t it fo ll o w s fr o m th is th a t in all cases, w h e n E q. ( 4 . 6 8 ) is sa tisfie d [i.e ., w h e n th e
c o r r e la tio n fu n c tio n ^ u u ( t ) sa tisfie s E q. ( 4 . 7 2 ) ] t h e lim it o f th e tim e m e a n s fo r m u la te d f or
in d ivid u al r e a liza tio n s o f th e p r o c e s s « ( / ) , w ith p r o b a b ility o n e e x is t an d are eq u a l t o u ( t ) .
3 REYNOLDS EQUATIONS AND THE
SEMIEMPIRICAL THEORIES OF
TURBULENCE
5. TURBULENT SHEAR FLOWS IN TUBES.

BOUNDARY LAYERS, ETC.
5.1 Reynolds Equations
The study of turbulent fluid flows naturally begins with the flows
in a circular tube and in the boundary layer on a flat plate. This is
because these are the easiest cases to reproduce in the laboratory and
have great importance for many engineering problems. The abundant
experimental material which has been collected on such flows
permits us to consider them as a standard against which we may
verify various theories and hypotheses on the nature of turbulence.
The discussion of the basic data on the most im portant integral
characteristics of flows in tubes and boundary layers, i.e., th e
longitudinal velocity profile, the flow rate and the skin friction law,
will also occupy a central place in this section. We shall also discuss
briefly the case of “ free turbulence” in which no perceptible effect is
produced by solid walls and in conclusion we shall consider certain
257
hypotheses on turbulent flows which are widely used in practical

calculations. First, however, we must present O. Reynolds’ general
considerations (1894) relating to arbitrary turbulent flows, which
constitute the basis o f the whole the 6 ry of turbulence.
We have already stated that the fluid dynamic fields of velocity,
pressure, temperature, etc., in a turbulent flow are so complex in
structure that it is practically impossible to describe them individu
ally. Therefore, initially, we must consider the whole ensemble of
such flows and study only its averaged statistical characteristics,
assuming that all the fluid dynamic fields concerned are random in
the sense explained in Sect. 3.2. Henceforth, we shall always assume
that such an approach is possible; i.e., we shall define as turbulent
only those flows for which there exists a statistical ensemble of
similar flows, characterized by some probability distribution (with
continuous density) for the values of all possible fluid dynamic
variables. In this connection, we must stress that the ordinary
definition of turbulent flows simply as flows accompanied by
disordered fluctuations of all the fluid dynamic variables, is quite
insufficient for the formulation of a mathematical theory of
turbulence. On the other hand, if a statistical ensemble of flows
exists, then the corresponding statistical description of all the fluid
dynamic variables will not be “incomplete” even from a purely
practical viewpoint. This is because knowledge of every detail of the
very confused individual flow is never required for any application,
and it is only the mean characteristics which are of interest. Of
course, in practice, it is not ensemble averaging which is usually
employed, but time- or space-averaging. Thus from a practical
viewpoint, it is also necessary that the random fields of the fluid
dynamic variables possess some ergodic properties. Henceforth, we
shall always assume this last condition to be satisfied without making
special mention of it.
The most important, and at the same time, the simplest statistical
characteristics of random fluid dynamic fields are their mean values.
The differences u' = u — u between individual values of the field u
and its mean value u are naturally called fluctuations of the field u.
The decomposition of all the fluid dynamic variables into mean
values and fluctuations played a fundamental role in Reynolds’
original considerations, and also in almost all subsequent investiga
tions of turbulence.
The mean values o f the fluid dynamic variables generally prove to
be very smooth and slowly varying; on the other hand, the
REYNOLDS EQUATIONS 259
lluctuations are characterized by great variation in time and space.

Generally speaking, it might be assumed that turbulent inhomogene
ities have arbitrarily small scales and periods, down to scales
comparable with the length o f the mean free path o f a molecule and
periods comparable with the mean interval between two successive
molecular collisions. If this is so, then the use in the theory of
turbulence o f the ordinary concepts and methods of continuum
mechanics (in particular, the differential equations of fluid dy
namics) will, of course, be invalid. However, by experiment,
turbulent inhomogeneities never have such small space-time dimen
sions as these. This is explained by the fact that for inhomogeneities
as small as these, the corresponding velocity gradients would be
extremely large; hence for very small-scale motions the energy loss in
overcoming the viscous friction would be so great as to make the
existence of such motion impossible. Consequently, the minimum
scales and periods o f turbulent inhomogeneities will always exceed
the scales and periods of molecular motion by several orders of
magnitude. More precisely, the length-scales o f the smallest inhomo
geneities observed in real turbulent flows o f air and water are of the
order of several millimeters, or, in extreme cases, of tenths of a
millimeter (see, e.g.. Volume 2, Chapt. 8 ) while under normal
conditions, the length of the mean free path of air molecules is o f the
order of 1 0 '^ cm, and for molecules o f water, it is of even smaller
order. As far as orders of magnitude are concerned, since the velocity
of fluid flows does not exceed the mean velocity of thermal motion
of molecules (close to 5 X 10“* cm/sec, or of the order o f sound
velocity), the characteristic periods o f turbulent fluctuations will
always exceed the mean time between two successive molecular
collisions by several orders o f magnitude. At distances comparable to
the dimensions of minimum inhomogeneities and for time intervals
comparable to the minimum periods o f the fluctuations, all the fluid
dynamic variables will vary smoothly, and may be described by
differentiable functions. Thus it follows that the description of
turbulent flows by means of the usual differential equations of fluid
dynamics is completely justified.
However, it is impossible to apply these equations directly, if only
because the fluid dynamic fields in turbulent flows are always
unsteady and depend strongly on the finest details o f the initial
conditions, details which are never known with sufficient precision.
Moreover, even if the initial conditions are known exactly, the
solution of the problem with them is excessively cumbersome and of
no practical use because of instability with respect to small

disturbances of the initial data. However, this does not mean that the
equations of fluid dynamics can never be used in the study of
turbulence. Due to the fact that the individual realizations o f the
fluid dynamic fields o f a turbulent flow will satisfy given differential
equations, the statistical characteristics of these fields will be
connected with a number o f important relationships, which are of
paramount importance in the theory o f turbulence.
The simplest connection o f this kind was established by Reynolds
by means of direct averaging of the equations o f fluid dynamics of an
incompressible fluid. Let us take as our starting point the equation of
balance of momentum (i.e., the Navier-Stokes equations (1.6)
multiplied by 3) in which it will now be convenient to transform the
term with the aid o f the continuity equation into the form
^ We apply the averaging operation to all terms of this

equation and use the fact that this operation commutes with the
operations of space and time differentiation [Eq. (3.6)] and also the
equation
= (“ i + + m; ) =
which follows directly from Eqs. (3.3) and (3.7'). In this case we
obtain the equations
+ + = + <= 1 . 2 , 3 ,
which are usually called the Reynolds equations. These equations

contain only smoothly varying averaged quantities; thus their use
does not give rise to any difficulties connected with the complexity
and irregularity o f the fluid dynamic variables of turbulent flows.
However, another difficulty does arise. It is connected with the
presence in the Reynolds equations of the new unknowns
= — pu[u'j [the reason for including the minus sign will be
explained below ], which characterize the fluctuating component of
the velocity field. The appearance of these new unknowns is
evidently due to the nonlinearity o f the equations of fluid dynamics.
Of course, when linear equations are averaged, no new terms arise.
Thus, for example, the averaged continuity equation will take the
simple form
I I - = 0. (5.2)
To understand the physical meaning o f the additional term

— p«'«' in Eqs. (5.1), let us consider the mean value of the
momentum flux density
Ij — = PUiti +
y
dui duj\
( is the viscous stress tensor in an incom
pressible fluid. This expression shows that in relation to the averaged
motion the role o f the viscous stress tensor is played by the tensor
= <s^j — pu\u’j = ; to obtain the sum here, the minus sign
was introduced into the expression for Thus, in turbulent flow,
in addition to the exchange of momentum between fluid particles
due to the forces o f molecular viscosity (described by the viscous
stress tensor) the transport o f momentum from one volume of fluid
to another occurs, induced by the mixing caused by the velocity
fluctuations. In other words, the effect of turbulent mixing on the
averaged motion is similar to that of an increase of viscosity; to
emphasize this, the Reynolds equations are sometimes written as:
1 — y rs 1 '^
The terms — pu\u'j in Eqs. (5.1) for the averaged motion have the
physical meaning of components of the tensor of the additional
stresses which are caused by the turbulent fluctuations, just as in
ordinary fluid mechanics, microscopic molecular motions lead to the
appearance o f viscous stresses. These additional stresses are called
Reynolds stresses in turbulence theory.
The component describing the transfer of the momentum o f the
fluid to a solid body washed by the flow is of particular interest in
studying the Reynolds stress tensor. Let 2 be some small area on the
surface of such a body, which may be considered to be approxi
mately plane and to coincide with part o f the plane Xs = 0. Further,
we assume that the direction o f the mean flow in the neighborhood
o f this plane is parallel to the Oxi axis. In this case, the frictional
forces acting on the area 2 will also be directed along the Oxi axis.
The magnitude of the frictional force acting on the unit area of the
surface of the body will be equal to the density of flux of the
ATi-component of momentum in the direction Oxz taken at a point of
the surface; i.e., it will be defined by the equation
"0 = (» 1 3 -P ^ ),,= 0 *
The variable to is called the shear stress at the wall. Since in a rigid
wall and u\ and their derivatives with respect to are evidently
equal to zero, while
- I dUi . dus \
Tfl may also be written in the following simpler form:
dUi (5.3)
Jf3=0
and may be called the viscous shear stress at the wall. However, if we
choose an area, parallel to 2 , above the surface o f the body, then the
total stress on it will be defined by
^ = ^13 — P ^ - ^5-4)
In the important special case when the mean velocity is everywhere

directed along the Ox\ axis, x will be equal to the shear stress
' = (5.4')
In this case, according to the ideas of Boussinesq (1877; 1897), we

can write, formally.
- p « X = p /C ^ . (5.5)
where /C is a new physical variable with dimensionality , called

the coefficient of eddy viscosity (or o f turbulent viscosity, or of
momentum transport). The sum v + K may therefore be called the

total (or effective viscosity) o f the turbulent flow. Unlike the
ordinary (molecular) viscosity v , the eddy viscosity /( does not
describe any physical properties of the fluid, but characterizes the
statistical properties of the fluctuating motion. Consequently, it does
not have to be a constant, but may vary in space and time, and may,
in principle, even take negative values (see below, Sect. 6.3).‘ It is
important initially to stress that the eddy viscosity K normally is
considerably greater than the molecular viscosity v; hence, at a
distance from the rigid boundaries, the total shear stress (defined by
the flux of the xrcom ponent of momentum in the direction o f the
0 ^ 3 axis) may be evaluated from the equation
T= = — p«J«3. (5.6)
The Reynolds equations (5.1) are equations o f the balance of

momentum o f the mean motion; the Reynolds stresses which occur
in them describe the turbulent transport of this momentum. Similar
balance equations may also be obtained for an arbitrary scalar
conservative quantity transferred by the fluid (e.g., for heat or a
passive substance like water-vapor, carbon dioxide, smoke or dust in
the atmosphere). Applying the averaging procedure to Eq. (1.72),
which describes the transport of the scalar quantity d in an
incompressible fluid, we obtain
+ d + = (5.7)
or, otherwise,
) (5-7)
[cf. Eqs. (5_.l) and (5.1')]. If 0 denotes the temperature, then the
vector describes the advective heat transport by the mean
*H ere w e sh o u ld n o t e th a t th e e d d y v is c o s ity , b y d e fin itio n , d e p e n d s d s o o n th e

averaging p r o c ed u r e u s e d , i.e ., o n th e c h o ic e o f th e sta tistic a l e n s e m b le o f sim ilar f lo w s , th e
a s sig n m e n t o f w h ic h m u st b e in c lu d e d in th e d e fin itio n o f a giv e n tu r b u le n t flo w . T h is is
e sp e c ia lly im p o r ta n t fo r “ n atu r a l” tu r b u le n c e in th e a tm o sp h e r e or o c e a n , w h e r e u su ally
a u n iq u e ly d e fin e d n atural “ e n s e m b le o f sim ilar f l o w s ” d o e s n o t e x is t , an d w h e r e averaging
over d if fe r e n t s ta tistic a l e n s e m b le s m a y le a d t o values o f K th a t d iffe r b y several ord ers o f
m a g n itu d e.
motion x , where x = Cppx is the heat transport due to the

molecular thermal conductivity, and
(5.8)
will be the density o f the turbulent heat flux in the direction o f the
Oxi axis. The turbulent fluctuations lead to the appearance o f an
additional heat flux, somewhat similar to that produced by the
molecular thermal conductivity. As in Eq. (5.5) we may write,
formally,
= (5.9)
where K# plays the part o f a new coefficient of thermal conductivity,

called the coefficient o f eddy (or turbulent) thermal conductivity or
the coefficient o f heat transport ', generally speaking, this coefficient
may differ for different coordinate axes. On the other hand, if in
Eqs. (5.7) and (5.7'), ^ represents the concentration of a passive
substance, then
y , = pF5; (5.8’)
will denote the density o f the turbulent flux of the substance in the
direction of the Oxi axis; in this case, the coefficient K& in the
equation
p 5 ^ = -p ;f,^ (5.9')
is called the coefficient o f eddy (or turbulent) substance diffusivity

or the coefficient o f substance transport [this last coefficient was
introduced by Schmidt (1917; 1925)].
All the preceding remarks on eddy viscosity may, o f course, be
repeated for the eddy thermal conductivity and eddy diffusion
coefficients. Section 6.3 discusses all these coefficients.
5.2 General Form of the Mean Velocity Profile

Close to a Rigid Wall
Let us now investigate the general properties o f turbulent flows
close to a rigid wall, parallel to the direction of the mean flow. The
results o f this study will be applicable to flows in a circular tube or in

a plane channel, and also to flows in a boundary layer on a flat plate
(in particular, to the surface layer of the neutrally stratified
atmosphere). We shall begin, first, by considering the simplified
idealized case o f a steady plane-parallel flow of fluid moving in the
direction of the Ox axis in the half-space z > 0, bounded by a rigid
wall, in the absence o f a mean pressure gradient.
For laminar steady plane-parallel flow with zero pressure gradient,
it follows from the first equation of fluid dynamics (see the first
example o f Sect. 1.2) that the velocity profile u(z) must be linear.
The first Reynolds equation for the analogous turbulent flow will
take the form
= (5.10)
which states that here, the flux of the x-component of momentum

along the Oz axis (directed from the fluid to the wall) will be the
same at any distance from the wall:
x(z) = p v -^ — pH'w' = xo = const, (5.11)
where to is the viscous shear stress on the wall z = 0. However, this

does not determine uniquely the profile o f the mean velocity u (z)
since, in addition to the function u (z ),it contains another unknown
u'w'. Nevertheless, certain deductions on the possible form of the
function M(z)may be obtained here by dimensional analysis. In fact,
the averaged characteristics of the flow under consideration at a
distance z from the wall can depend only on the shear stress Tq, the
z-coordinate, and the parameters o f the fluid v and p. Moreover, Tq
and p can enter into the kinematic characteristics o f the flow only in
the combination . which is independent o f the dimension o f mass.
Instead of the ratio it is convenient to use the quantity
which has the dimensions o f velocity and is therefore a natural scale

o f velocity for the flow close to the wall; we shall refer to this as the
friction velocity. Since only one dimensionless combination can

be formulated from u„ v, and z, the general form of the dependence
of the mean velocity profile u (z) on z, xo, p, and v may be written in
the form
(5.13)
or, equivalently, in the form
u + = /(z + ), where 2+= ^ - (5.13')
Here u+ and z+ are the dimensionless velocity and distance from the
wall, and f ( 2 +) is some universal function of a single variable. The
important result (5.13) in turbulence theory is called the universal
/aw o f the wall; it was first formulated by Prandtl (1925) [see also,
Prandtl (1932b)].
Equations (5.13) and (5.14) will of course be vahd only when the
wall may be assumed smooth, i.e., described by the simplified
equation 2 = 0 . Since the only length scale in the wall turbulence will
be the friction length 2 , = ^ , we can now give a quantitative
explanation of the requirement o f smoothness: the wall will be
dynamically smooth if the mean height ho o f the protrusions on it
satisfies the condition
= (5.14)
(for more precise data, see Sect. 5.4). Only in this case will the
velocity profile close to the wall be defined by Eq. (5.13). However,
for a rough wall which has protrusions with heights that do not
satisfy Eq. (5.14), these protrusions will also influence the distribu
tion of the mean velocity close to the wall. Consequently, in this case
Eq. (5.13) must be replaced by the more general equation
u{z)=uj[^, a, p, . . . ) . ( 5 . 15)
REYNOLDS EQUATIONS 26 7
where a, p, . . . are dimensionless parameters characteristic of the

form of the protrusions and their distribution over the surface of the
wall. Again, for extremely rough walls, the use of Eq. (5.15) causes a
further difficulty connected with the fact that here it is unclear from
what level the height z is to be measured; we shall return to this
question later in a more detailed discussion of flows near rough walls.
Before considering the form of the function f, let us discuss briefly
the applicability of Eqs. (5.13) and (5.15) to various turbulent flows
encountered in practice. We begin with a turbulent flow with zero
mean pressure gradient between two parallel walls moving with
respect to each other. Equations (5.10) and (5.11) in this case will be
exact; thus, for example, for smooth walls, for Eq. (5.13) to be valid,
it is necessary only that the distance z from one o f the walls be much
less than the distance H between the walls.^ However, a flow
between plane moving walls is very difficult to reproduce precisely in
the laboratory; consequently to verify Eqs. (5.13) and (5.15) we
must examine other more easily reproducible flows, having velocity
profiles which are described by these equations.
As already observed, one example o f such a flow is the air motion
close to the earth’s surface (over land or water) in the case of neutral
thermal stratification. O f course, in the earth’s atmosphere the
horizontal pressure gradient is always slightly different from zero
and, moreover, the wind velocity here is also affected by the coriolis
forces produced by the rotation o f the earth (see below. Sect. 6 .6 ).
However, in the lowest layer of air, these factors play only a small
part, and cause no perceptible variation o f the shear stress x with
height; simple estimates, which we shall give at the end of Sect. 6 .6 ,
show that in the atmosphere, right up to a height of about 50 m, t
can usually be taken to be practically constant. Consequently, within
this layer, for neutral stratification, we have sufficient grounds for
using the simphfied equations (5.11) and (5.15) [we note that in the
real atmosphere the underlying land surface is almost always rough,
since is generally of the order of 10 cm/sec and, consequently,
does not exceed some tenths of a m illim eter].
^ s tr ic tly speak in g, fo r tu rb u le n t flo w b e t w e e n m o v in g w a lls, an e q u a t io n o f th e form
M (z) = a ,/ j m u st a p p ly , w h e r e — H — z is th e d is ta n c e to th e s c c o n d
b o u n d in g w a ll. F o r Zi z , it is p erm issib le to ta k e — = c o , an d th is e q u a t io n th e n
re d u c e s t o E q. ( 5 . 1 3 ) .
Let us now discuss laboratory turbulent flows in tubes, channels

and boundary layers over a plate. In the simplest case o f a wide plane
channel, the flow may be considered approximately as a steady
plane-parallel flow between two infinite planes 2 = 0 and z = H =
2Hi, homogeneous in the planes z = const. In this case, the first and
third o f the Reynolds equations will take the form
dz dp dw'^ dp fF t
where, as usual, -: = pv — pu'w'. From the second equation o f Eq.

(5.16) it follows that p - f - d e p e n d s only on x (pêvidently
coincides with the mean pressure on the wall). But w''‘ cannot
depend on x; hence, using the first equation o f Eq. (5.16)
t= ^ 2 -}-const.
We denote the shear stress at the lower wall 2 = 0 by to; since by

symmetry the shear stress at the center of the channel (for z = H^)
must equal zero, the last equation may also be written as
Thus, f oTZ- ^Hi , we have •z{z) = xo = const; consequently, in this

case also, a layer exists, contiguous to the wall o f the channel, in
which the shear stress is practically constant. Within this layer, the
influence o f the pressure gradient may be ignored, and therefore Eqs.
(5.13) and (5.15) may be used (see also the additional remarks at the
end o f this subsection).
The case o f a steady pressure flow in a circular tube of diameter
D = 2R is completely analogous. Here, transforming the Reynolds
equations to cylindrical coordinates (x,r,<f>), we can rewrite the first
and second equations as:
llrz = ^ , (5.16')
r dr dx \ r dr r I dr
where
du^ ---
From these equations it follows that = 0 , i.e., that the

dxdr
longitudinal pressure gradient does not depend on z. Therefore
— = — where Pn is the static pressure on the wall. Now from the
dx dx
first equation (5.16') we obtain
dx 2 dx
where z = R - r is the distance from the wall. Consequently, with

z <K R , the shear stress t may once again be assumed constant, and
the mean velocity profile will be given by Eq. (5.13) [or Eq. (5.15)],
as in the case o f plane-parallel flow with zero pressure gradient in a
half-space.
Finally, the Reynolds equations are obtained in a boundary layer
on a flat plate by adding terms corresponding to the Reynolds
stresses, to the usual boundary-layer equations (1 .3 8 )-(1 .3 9 ) for
averaged fluid dynamic fields. By considering the fact that the
fluctuation velocity in the boundary layer is much less than the mean
longitudinal velocity u, and that the x derivative is much less than
the z derivative, retaining only the leading terms we find that the
third Reynolds equation for zero longitudinal pressure gradient will
take the form
- du , — du d —i— , , dû /e 1o\
From this it is possible to estimate the variation o f the shear stress t

along the vertical and to show that in the boundary layer there exists
a considerable layer within which the shear stress is practically
constant. Within this layer the flow may be assumed plane-parallel
and described by Eq. (5.10), while the mean velocity profile is given
by Eq. (5.13) or (5.15).
It is in te r e stin g t o n o t e h e r e th a t, a c c o r d in g to L u d w ie g a n d T illm a n n ( 1 9 4 9 ) , E q. ( 5 . 1 3 )
w ith th e o r d in ar y universal f u n c t io n / p roves t o b e co rr e c t a lso fo r a rather t h ic k p o r tio n o f
b o u n d a r y layers w ith c o n s id e r a b le (t h o u g h n o t t o o str o n g ) lo n g itu d in a l pressure gra d ie n ts,
w h ic h , it w o u ld s e e m , m u st ca u se grea t variatio n o f th e shear stress w ith in t h e p o r tio n o f th e
flo w u n d er c o n s id e r a tio n . T h is o b s e r v a tio n w as a lso c o n fir m e d later b y several o th e r
inv e stig a to r s. C o le s ( 1 9 5 5 ) e x p la in e d th is b y th e fa c t th a t in a b o u n d a r y la yer w ith
lo n g itu d in a l pressure g r a d ie n t, th e v aria tion o f a lo n g th e w a ll le ad s to a v aria tion o f r
a lo n g t h e vertical, w h ic h to a great e x t e n t , c o m p e n s a te s for th e ch a n g e in r in th e z -d ir e c tio n
p r o d u c e d b y th e m e a n pressure gra d ie n t [c f. R o t t a ( 1 9 6 2 a ; b ) ] . H o w e v e r , th e q u e s tio n is
n o t c o m p le t e ly clear; e .g ., M ellor ( 1 9 6 6 ) fo u n d w ith t h e aid o f sp e cific sem iem p ir ic a l
ar gu m en ts th at t h e e f f e c t o f th e pressure g r ad ien t o n th e v e lo c ity p r o file in a b o u n d a r y layer
near a w all is q u ite c o n sid e ra b le .
A n o th e r e x p la n a t io n o f th e c o m p a r a tiv e ly sm all in f lu e n c e o f t h e pressure grad ie n t o n th e
v e lo c ity p r o file m a y b e r ela te d t o th e fa c t (c o n fir m e d b o t h b y e x p e r im e n ta l d a ta and
a p p r o x im a te th e o r e tic a l c a lc u la tio n s) th a t th e v e rtical p r o files o f th e m e a n d y n a m ic a l
variables are o n ly s lig h tly s e n sitiv e , e v e n t o fairly large c h a n g es in t. T h u s , fo r e x a m p le , fo r a
layer in w h ic h r varies b y n o t m o r e th a n 15-20% , w e m a y a lm o s t a lw a y s use th e s im p lifie d
th e o r y , c o n sid e rin g th e la y er t o b e c o n sta n t-stre ss (i.e ., t = c o n s t). L ater, w e shall see th a t in
several ca ses, th e results for a c o n sta n t-stre ss layer g iv e a fair a p p r o x im a tio n e v en ov er th e
to t a l s e c tio n o f th e c h a n n e l (or t u b e ). In th e n e x t a p p r o x im a tio n t h e in f lu e n c e o f th e stress
variation m a y b e ta k e n in t o a c c o u n t b y a sim p le p e r tu r b a tio n p r o c ed u r e w h ic h w ill b e
d e s cr ib e d at t h e very e n d o f S e c t. 5 .5 .
5.3 Flow Close to a Smooth Wall; Viscous Sublayer and

Logarithmic Boundary Layer
Let us still assume that the wall 2 = 0 is dynamically smooth; thus
Eq. (5.13) holds. Then the form of the function / ( 2 +) occurring in
this equation may be determined explicitly for two limiting
cases, for large and for small values of the argument z+.
In the region very close to the wall, the fact that p u W = 0 and
To = pv ^ on the wall itself [see Eq. (5.3)] plays a fundamental
role. Thus for sufficiently small values of z, the viscous stresses will
be considerably greater in magnitude than the Reynolds stress
— pu^w^.The layer of fluid within whichv ^ usually
called the viscous sublayer} Within this sublayer we may assume
^ A t o n e tim e th e te r m “v is c o u s s u b la y e r ” and th e term “lam inar s u b la y e r ” w e r e b o t h
u se d , sin c e it w as a ssu m ed th a t in th is la yer th e f lo w w a s lam inar. H o w e v e r , th e d ata o f
d irect u ltr a m ic r o s c o p ic o b s e r v a tio n s o f th e m o t io n o f sm all p a r ticles or b u b b le s in flu id
c lo se t o a w a ll [P a g e an d T o w n e n d ( 1 9 3 2 ) ; F id m a n ( 1 9 5 9 ) ; O rlov ( 1 9 6 6 ) ; P o p o v ic h and
H u m m e l ( 1 9 6 7 ) ; K lin e, R e y n o ld s e t al. ( 1 9 6 7 ) ; and o th e r s ] and o f n u m e r o u s refin e d
h o t-w ir e a n e m o m e t e r m e a s u r e m e n ts in th e v ic in ity o f a w all [L a u fer ( 1 9 5 1 ; 1 9 5 4 );
K le b a n o ff ( 1 9 5 5 ) ; C o m te -B e U o t ( 1 9 6 3 ; 1 9 6 5 ) ; C o a n tic ( 1 9 6 6 ; 1 9 6 7 a ); K lin e, R e y n o ld s e t
al. ( 1 9 6 7 ) ; an d m a n y o th e r s ] s h o w q u ite d e fin ite ly th a t a lth o u g h th e m e a n v e lo c ity w ith in
th e su b la y e r is id e n tic a l to th e Hnear v e lo c ity p r o file o f a plan e-parallel lam inar f lo w w ith
zer o pressure g r a d ien t, th e f lo w w ith in it n e v e rth e le ss is n o t lam inar, b u t a c c o m p a n ie d b y
c o n s id e r a b le irregular flu c t u a tio n s . C o n s e q u e n tly , at p r e se n t, th e term “lam inar s u b la y e r ”
m u st b e c o n s id e r e d to b e m isle a d in g .
that
p v = To = const (5.19)
and therefore
u{z) = ^ z , /(z + ) = «+. (5.20)
Strictly speaking, Eq. (5.20) is only the first term of the expansion
o f the function f(z+) as a Taylor series in powers of z+ in a
neighborhood of the point z+= 0. The subsequent terms of this
expansion may be obtained by differentiating Eq. (5.11) with respect
to z at the point 2 = 0 . Here, naturally, u '= u '= t « '= 0 for 2 = 0; as a
result, when 2 = 0 all the derivatives of the velocity fluctuations with
respect to x and y are equal to zero and, furthermore, -0 ~ ^
from the continuity equation.
Thus
d û \ 1 / d
V
dû \ 1 (
,dz^ ” V '[d z^
d*u\ _ 1 _ 3 / du' d^w' \
[dz^
uw dz^ j^=o'
. d z * 7^=0 ~
Therefore, it follows that the Taylor expansion o f the mean velocity

profile in powers o f 2 has the form
_ u® u®
u {z)= --^z — c ^ ^ z ^ + c ^ - f z ^ +
(5.20')
f { z ; ) = ^ z ^ - c , z \ + c , 2^ ^ + ...
Here
' [ d ^ u ' w ' \ ____/ du' d V

^4—— 24u^ *
~ 8u®I dz
The minus sign is inserted because it seems intuitively clear that close
to the wall u'w '< 0 and hence while Cg =
ion 6 ( r expansion(5.20') was demonstrated (apparently

Iz O f/* ( 72’ = o
for the first time) in a somewhat different form by Murphree (1932)

[see also Townsend (1956)]. Later, certain authors [in particular,
Landau and Lifshitz (1963); Deissler (1955); Elrod (1957); Levich
(1962); Lyatkher (1968) et al.; cf. the closing remarks of this
subsection] formulated some arguments in favor o f the assumption
that c/. = 0, but all these are not rigorous.'* The existing experimental
data are also insufficient for reliable estimation of the coefficient c...
and therefore give no reason to conclude that C4 = 0. Nevertheless,
the fact that the second and third derivatives of / ( 2 +)are equal to
zero when z+= 6 is sufficient for the variation o f the velocity over a
considerable region to be very close to linear, i.e., for the concept of
a viscous sublayer to be justified.
The upper boundary of the viscous sublayer may be defined
conditionally, e.g., as the value of 2 for which u 'w '— 0.1 v— ;
dz
certain other definitions related to this one may also be used. In
every case the thickness o f the viscous sublayer can depend only
on the parameters and v . Consequently, it must be given by the
equation Z — a^ —//jj.
~ayZ^, where a„ is a universal constant of the
order of unity (the precise value of which will of course depend on
the chosen definition), which must be found from experimental data.
The fact that this constant depends on the choice of definition, i.e.,
has no unique value, is quite natural since the viscous sublayer does
not have a sharp upper boundary, but passes smoothly into the next
region of the fiow, in which the viscous stress and the Reynolds
stress have the same order of magnitude. Nevertheless, just as in the
case o f the numerical coefficient in the equation for the boundary
layer thickness 6 , the range o f permissible values of Uv in this case
proves not to be very broad. Most often the number 5 is chosen as a„,
i.e., it is assumed that 8.^ = 5 the basis for this choice is the
experimental data, given in Fig. 25.
Let us turn now to the second limiting case, to values of 2
considerably greater than In a developed turbulent flow at a
distance from rigid walls, the turbulent stresses are many times
^ T h e d e m o n s t i a t io n o f th e e q u a t io n = 0 p r o p o s e d b y Ohji ( 1 9 6 7 ) is in c o r r ec t; in fa c t
it is b a se d o n a p articular n o n r ig o r o u s a ssu m p tio n ( n o t fo r m u la te d e x p lic itly in th e paper).
L et u s n o t e a lso th a t c o n s id e r in g th a t in real c h a n n el- an d t u b e - f l o w x ( z ) c o n s t, b u t axldz =
c o n s t [c f. E qs. ( 5 . 1 7 ) an d (5 .1 7 * )] w e shall o b ta in a n o n z e r o value o f th e d im e n sio n le ss
c o e f f ic ie n t o f th e z ^ -te rm o f th e T a y lo r series fo r w (z); h o w e v e r , ^^3 = 0 e ven in th is case.
50 ■
m
m o
28 ♦ -0 /
0
A
y
K
26 A •<
a -2
24 b kt\
■ 9 '
* ^
20
••
.fj i f
16
16 .vf
7
t4
f2
to
m
(if
1 Set wbauer &
• /C/e*banoff
A Bre}bner & Bagley
t N i k wradse
/ fchardt & Mor t zf el d
5 *Rei ichardt & Schuh
oLat j f e r
6 ^ Dei iss/er
® Vain Driest
^Scf l ul t z- Gr unow
4
j OKIei banof f
*Fai je
2 ~r
11 ■
0
to 10^
FIG . 2 5 . U niversal d im e n sio n le ss m e a n v e lo c it y p r o file o f tu r b u le n t flo w c lo se to

a s m o o t h w all a c co r d in g t o th e d ata o f tu b e -, c h a n n el- and b o u n d a r y -la y e r m e a
s u r em en ts [a c c o rd in g t o K e stin and R ic h a r d so n ( 1 9 6 3 ) ] .
greater in magnitude than the viscous stresses. Consequently, for

sufficiently large z (let us say, for z>(>i) we may ignore the term
du
pv in Eq. (5.11) and assume that x o = — p u ' w ' . Hence for 2 > 6i
dz
the law of variation o f the mean velocity must not depend on the
viscosity v but must be determined only by the values o f the density
p and o f the momentum flux Tj passing through the fluid (and equal
to the total interaction force of the fluid with the wall per unit area
of the wall). Essential here is that we are speaking not o f the velocity
itself, but the law of variation of the mean velocity. In fact, by
considering the flow within a layer of negligible viscous stresses, we

can say nothing about the absolute values o f the velocity ^T(^).
However, we can study only the differences a(zi) — u{z 2) of the
values o f the velocity at two heights and Z2 within this layer, and,
in particular, the increase u{z) — a ( 6 j )of the velocity in the layer
(2 , 6 ;),where z > bi. This follows from the Galilean invariance o f the
equations of mechanics, according to which the addition of any
constant to all velocities m(2 ) for 2 > 6 ; cannot change the momentum
flux transmitted through the fluid. Consequently, the absolute value
of the velocity u{z), 2 > 6 / is not defined by the values to and 2 , but
depends also on the value of m(61) , i.e., on the law of variation of
velocity in the layer 2 < which itself is influenced by the viscosity
V. However, the value o f the gradient of the mean velocity at
height 2 for z > flj must be independent of v; i.e., it must be a
function of the parameters to, p, and z only. As easily seen from these
parameters, it is possible to formulate a unique combination =
u
~Y with dimension o f a velocity gradient. Thus for z > bi the
relationship
= (5.21)
dz z
must be satisfied, where .4 is a universal dimensionless constant. For

the velocity profile at z > bi from Eq. (5.21) we obtain the
logarithmic equation
u{z) = A u ^ \n z -\-A i, (5.22)
where /li is a new constant which, by the preceding argument, may

also depend on the viscosity v. The layer of fluid in which Eq. (5.22)
is satisfied is called the logarithmic layer, or logarithmic boundary
layer', the existence o f this layer is extremely important for a number
of problems in which turbulent flows along rigid walls are
encountered. The universal equation (5.22) was first deduced by von
Karman (1930) and Prandtl (1932b) using completely different
(from that presented above and from each other) arguments. Later, a
whole series of proofs was found for it; some of them are of
considerable interest, and will be reproduced below. A simple, purely
dimensional deduction of this equation was published for the first
time in 1944, in the first edition of Landau and Lifshitz’ book

(1963); see also Squire (1948).
Instead of dimensional analysis in the deduction of Eq. (5.22), we
may also use the invariance o f the dynamic equations o f an ideal
fluid, which describe the flow for z > 6/, under the similarity
transformations x - ^ k x , y ^ ky, z ^ k z , and t ^ k t . Since these
transformations change the half-space z > 0 into itself, it is natural
to think that all the statistical characteristics o f turbulence in this
half-space (insofar as they are independent of the viscosity) will also
be invariant under them. However, as we have already seen, if we
ignore the viscosity (i.e., neglecting for the time being the boundary
condition m = 0 when z = 0), we can consider only the relative
velocities it (2i) — u (22) • Since the friction velocity clearly does
not change under similarity transformations, then in accordance with
the assumption on invariance, the dimensionless ratio “
for values of Z\ and Zi that are not too small, can depend only on the
ratio
(5.23)
From the definition (5.23) of the function it follows that
+ the latter relationship, by virtue of the
identity ^ ~ ^ ‘ ^ may also be written in the form
^ ( 5 , - y = g(?i) + g ( y . (5.24)
It is not difficult to show that the only continuous solution of the

functional equation (5.24) is the logarithmic function =A\nl;
thus, we arrive at the equation
u { z ,) ~ u { z , ) = A a , l n ^ . (5.22')
which is equivalent to Eq. (5.22).

From dimensional arguments it is clear that the height flj of the
lower boundary of the logarithmic boundary layer must be defined
by the equation \ — where is another universal dimension-

less constant, determined to approximately the same degree of
accuracy as the constant a„. The data of Fig. 25 show that it is
permissible to take a (= 3 0 . From a comparison of Eqs. (5.22) and
(5.13) it follows that the constant Ai must be of the formA^ — Au,
in Bu , where S is a dimensionless universal constant; therefore
V ' *
Eq. (5.22) may be rewritten as
«(2) = a, I n w h e n (5.25)
Thus
f(z^) = A \ n z ^ - \ - B when 2 + > «|. (5.25')
Here it must be pointed out that instead of the coefficient A the

traditionalX = is used frequently; * is usually called von Kdrmdn’s
constant. Replacing A by :ând denoting by p, Eq. (5.25) may

also be rewritten as
u{z) = ^ \ n ^ . (5.25")
The numerical values of the constants ( o r < = l/>4) and B (or

fj = e-xB) £qg (5 .2 5 )-(5 .2 5 ") may be determined from the
experimental data. As indicated in Sect. 5.2, the measurements may
be carried out in smooth tubes, in rectangular channels with smooth
walls, and in the boundary layer on smooth plates. The first useful
measurements for this purpose, i.e., precise measurements of the
mean velocity distribution and the shear stress distribution, were
made by Nikuradse (1932) in water flows in smooth, straight tubes of
different radii with Re = — varying between 4 X 1 0 ^ and 3.2 X
10*. His data show that for a considerable part of the flow, beginning
at a distance of about 30 ~ from the wall and continuing almost to
the center of the tube, the mean velocity may be described with
good precision by an equation of the form of Eq. (5.25). For the
coefficients A and B in this equation, Nikuradse gave two choices of
values: with A = 2.4, B = 5.8 (i.e., with x = 0.417, p 0.09

1/11) the best agreement of Eq. (5.25) with the experimental data
was obtained for the region 30 v/«, < 2 < 1000 v/«„ which does not
by any means extend to the center of the tube, while the values A =
2.5, B = 5.5 (i.e., x = 0.40, p =5^ 0.11 0.11 1/9) prove to be the
best for the application of Eq. (5.25) to the whole region from 2 =
30 v / « , to the center of the tube. Later, similar measurements were
repeated many times, both for flows in tubes (or in plane channels),
and for boundary layers on flat plates; while Eq. (5.25) in all cases is
quite reliably confirmed, in the values obtained for A and especially
B, there is a small amount o f scatter [see, for example, the survey by
Hinze (1962), which contains a detailed analysis of a large amount of
data on the values of A and B] .* Many data referring to the three
types of flow under consideration are collected in Fig. 25 (borrowed
from Kestin and Richardson (1963), and presented here by way of
example). We see that for ^ < 5 v/k, all the observed values of«+ =
«/a, lie fairly closely on the curve«+ = ru,/v of Eq. (5.20), while for
500 v/u, > 2 > 3 0 v/u, they lie on the curve«+ = 2.5 logzuj^
corresponding to Eq. (5.25) with coefficients/I = 2.5, B = 5.1 (i.e.,
X = 0.4, p = 0.13) recommended by Coles (1955) which are close to
the mean of the values of the coefficients proposed by all the other
investigators (both earlier and later than 1955).*
In the intermediate region 5 v/i<, < 2 < 30 v/ k. ,the experimental
values o f the ratio plotted in Fig. 25 clearly deviate from both
the values given by Eq. (5.20) and from those obtained from Eq.
^We reca ll th a t, a c co r d in g to th e th e o r e tic a l d e d u c t io n , A an d B m u s t b e universal

c o n s ta n ts o n ly fo r an id e a liz e d p lan e-parallel flo w w ith c o n s ta n t shear stress 'Zq. T h u s for
real la b o r a to r y e x p e r im e n ts , in w h ic h also th e reg io n o f a p p lic a b ility o f Eq. ( 5 . 2 5 ) is
d e te r m in e d “b y e y e , ” th e sm all sca tte r o f th e em p ir ic a l v a lu e s o f th e s e c o e f fic ie n ts w ill b e
e x p la in e d c o m p le t e ly , e v e n if th e u n a v o id a b le errors o f m e a s u r e m e n t and p r o c ed u r e are
ig n o r e d .
^ F o r e x a m p le , C lauser ( 1 9 5 6 ) p r o p o s e d fo r b o u n d a r y -la y e r flo w s th e valu es A = 2 .4 4 , B =
4 .9 ; T o w n s e n d ( 1 9 5 6 ) u se d th e valu e s A = 2 .4 4 , B = 5 . 8 5 ; in th e t e x t b o o k b y L o n g w e ll
( 1 9 6 6 ) slig h tly d iffe rin g v alu es A = 2 .7 1 , B = 3 .5 are r e c o m m e n d e d ; S p ald in g ( 1 9 6 4 ) and
E scu d ier a n d N ic o ll ( 1 9 6 6 ) a s su m e d th a t A = 2 .5 , 5 = 4 . 7 ; th e m e a s u r e m e n ts o f C o an tic
( 1 9 6 6 ) in circular tu b e f lo w an d o f C b m te -B e llo t ( 1 9 6 5 ) in p la n e c h a n n e l f lo w gave th e
resu lts A = 2 . 5 0 , 4 . 4 0 < ^ < 5 . 5 0 an d A = 2 .7 a n d 4 .5 < 6 . 0 , r e s p e c tiv e ly , e tc . T h e
s u g g e stio n o f a p o s sib le slight d e p e n d e n c e o f A an d B o n R e (i.e ., o n v is c o s ity V) has b e e n
m a d e [H in z e ( 1 9 6 2 ) ; C o m te -B e U o t ( 1 9 6 5 ) ] . F o r a c o n s ta n t-str e s s layer su ch a d e p e n d e n c e is
o b v io u s ly in c o m p a tib le w ith p h y sic a lly c o n v in c in g d im e n sio n a l a r g u m e n ts; h o w e v e r , it can
be e x p la in e d e asily b y th e in f lu e n c e o f stress varia tio n . T h e data o n th e d e p e n d e n c e are
e x tr e m e ly u n ce r ta in [e .g ., th e d e p e n d e n c e o n R e is la c k in g e n tir e ly in th e q u ite a ccu rate
r e c e n t da ta b y L indgren ( 1 9 6 5 ) ] ; th e r e fo r e a t p r e se n t it see m s rea so n a b le to disregard it.
(5.25). According to the experimental data, the values of f(z+) in

this region may be represented by a smooth curve, which passes
easily into the curve (5.20) at the point z+= 5 and into the curve
(5.25') with A = 2.5, 5 = 5.1 at the point z+ = 30 (the dotted line
in Fig. 25). Further, we note that the deviation of the dotted curve
from the broken curve, sohd line in Fig. 25, is not very great. Thus,
in cases when great accuracy is not required, it is permissible to
assume that right up to the value z+= 11.1 (the abscissa of the
intersection point of the curves «+ = 2 + and «+ = 2.5 lnz+4- 5.1) the
mean velocity profile is given by Eq. (5.20) while for z+ > 11.1 it is
given by Eq. (5.25). This assumption [accepted, e.g., in the papers of
Prandtl (1919; 1928) and G. I. Taylor (1916)] means, of course, that
du
we ignore the intermediate zone in which pv and — pu'w' are of
the same order of magnitude, and assume that outside the viscous
sublayer, which we now consider as having thickness 8^, = 8, =11.1
v/w,, immediately follows the logarithmic layer, in which the viscous
stresses are negligibly small in comparison with the Reynolds stresses.
Further, we note that from the viewpoint of the turbulent
viscosity K [see Eq. (5.5)] we may define the viscous sublayer as the
sublayer in which it is permissible to assume that K = 0 (i.e., that
the effective viscosity is equal to v = const). In exactly the same way
the logarithmic layer may be defined as the layer in which the
molecular viscosity is negligibly small in comparison with the
turbulent viscosity given by the equation
= * « 0 .4 . (5.26)
In these terms the dotted hne in Fig. 25 will correspond to a layer in

which the effective viscosity varies smoothly in some way from a
value close to v (for 2 = 5v/u,) to 30xv « 12v (for 2 = 3 0 v / m ,).
So far, we have discussed only the mean velocity profile of a
turbulent flow; however, the physical considerations which have led
us to a universal law of the wall, and to the concept o f a logarithmic
layer, may be applied equally well to the investigation o f any other
one-point velocity moments close to a rigid plane wall. All these
moments, within the constant-stress layer will, of course, depend
only on the parameters z, x, v, and p, i.e., they must be represented
by the product of some power of the friction velocity and of a
particular function of the dimensionless distance 2 + = 2 u,/v. For
sufficiently large 2 +, the statistical regime of turbulent fluctuations
can no longer depend on the viscosity v; thus in the case of central

moments (independent of the mean velocity u) the corresponding
universal functions will tend to a constant a s 2 +-> oo. As an example
we take the one-point second moments of the fluctuations v' and
w'. There are six such moments in all; however, two of them
(namely, u'a' and v'w') are identically zero due to the symmetry o f
the turbulence with respect to the plane Oxz. Therefore there remain
only four nonzero moments: u”‘ = <s\, x)'* = 4 , = and u'w'i
Therefore, in addition to Eq. (5.13), we shall also have four equa
tions o f the form
containing four new universal functions. Instead of we may

also use the function which describes the height
variation of the correlation coefficient between u' and w'
Within the logarithmic layer the function must be identically

equal to unity, while the functions /i, /a, fs, and fs assume constant
values
fi(oo) = Ai, [2 ( 0 0 ) = ^ 2, fs(°°) = A 3 and fs(°o) = ^ 5 ==
As we approach the wall, the' functions fi, ■ ■ ■, fi tend to zero,

while their Taylor-series expansions at the point z+= 0 take the
form:
A (^+) = ^ 1^+ + * 1^+ + • • ••

/2(^+) = V + + M + + •••«
/3(2^ +) = «3< + ^ ^ + 4 - •••.
/4(^+) = a4 < + V t + •••
(since u' = 1/ = w ' = = 0 when z = 0). The coefficients 04 and

&4 are related in a simple manner to the coefficients C4 and C5 of Eq.
(5.20') since close to the wall — l — f'(z+) [from Eq. (5.11)].

The exact form of the functions fi......... fi cannot be determined
theoretically, but their approximate form may be found from the
data of Laufer (1954), Klebanoff (1955), Comte-Bellot (1965),
Coantic (1966; 1967) and others (see, for example. Figs. 26 and 27
based on the data of Laufer and Klebanoff). Other examples of these
two graphs were published by Dumas and Marcillat (1966); the
agreement o f their data with that of Comte-Bellot is not very close,
but must be considered satisfactory. The data for tube flow and for
boundary-layer flow agree well with each other. In particular, for
constants Au A 2 , and A 3 in both cases approximately identical values
are obtained:
2.3, 1.7, 0.9.
(Values of At and A 3 relatively close to these were also obtained by

other authors including those who made measurements in the surface
layer of the atmosphere; see Sect. 8.5 below.) Processing the existing
data on the velocity fluctuations in proximity to a wall, we may also
obtain an approximate estimate (although still with fairly low
accuracy) o f the values of the coefficients Oi, . . . . 04 ; namely,
^ 0.3, 02 ~ 0.07, » 0.008, » 0.001
[see Laufer (1954); Klebanoff (1955); Townsend (1956); Coantic

(1965)].
J.0 K “Tr“
•
IS V U.
•v L
10 X
} -IT " TJ----- r r
1.5 -
tja r / U,
OJ) 1 / \lw
\j\
to ^ 30 dO 50 M 70 dO
f
zu,
24- —
FIG . 2 6 . U niversal d im e n sio n le ss p r o file s o f th e
in t e n s it y o f flu c t u a tio n o f th e th r e e v e lo c ity
c o m p o n e n t s in a f lo w c lo s e t o a s m o o t h w a ll
a c co r d in g t o th e d ata o f K le b a n o ff ( x ) and
L au fer ( • ) .
10
0.3 u'w
/ "~ui
0.6
0.^ " l\
/ u'wl'
0.2
0
J W 20 30 ^0 50 60
VP I p
70 80 90 WO
zu,
2“ V
FIG. 27. U niversal f u n c t i o n s /4 ( z + ) a n d f ; (z + ) ac

c o r d in g t o L a u fe r ’s m e a s u r e m e n ts.
For higher-order moments of the velocity fluctuations (of order

three and higher) there is considerably less data than for the
second-order moments defined in Eq. (5.27). Nevertheless, Comte-
Bellot (1963; 1965) constructed some preliminary empirical graphs
of the universal f u n c t i o n s = ( and f^(z^) = u'^ I

which define the skewness and excess of the longitudinal component
u'.
Similarity laws related to the law of the wall may be formulated
not only for the mean velocity profile and the profiles of one-point
velocity moments, but for general many-point probability distribu
tions of the velocity components at points in the constant stress
layer. All these distributions must depend only on i/, u+, and the
coordinates of the points; moreover, the dependence on the viscosity
V must vanish if we consider the probability distribution of the
simultaneous values of the velocity fluctuations u', v', and w' at points
..., within the constant stress layer satisfying
the following conditions: 1) z,- » e* = u/u* for all i = 1, . . . , n(i.e.,
all the points are located within the logarithmic layer); and 2) the
distances between any two of the points are much greater thanz*.
Under these conditions, the probability density o f the dimensionless
variables u'/m, v'/um, w'/u* can depend only on the dimensionless
parameters
‘'l ^2
This general similarity hypothesis for the probability distributions

within the logarithmic layer can be generalized also to the many-
point and many-time probability distributions (cf. Sect. 7.5, where
the more general case o f a density-stratified turbulent boundary layer

will be considered). However, at present no experimental data exist
even for the simpler many-point probability distributions at a single
time, suitable for the confirmation of the hypothesis, and the data
on many-point velocity moments within the constant-stress layer are
extremely poor.
In certain cases, primarily in the case of turbulent heat- and mass-transfer through a fluid
(see below, Sect. 5.7), the solid line of Fig. 25 will clearly be insufficiently exact, while the
smooth broken line will be quite inconvenient. In these cases, it is desirable to have an
analytical expression which describes the behavior of the mean velocity profile in the
intermediate layer 6u < 2 < 6/ and which makes a smooth juncture with the limiting
expressions (5.20) and (5.25) for 2 6u = ayV/wHc and 2 > 6 / = a/v/w*, respectively. A
whole series of such expressions for functions of varying complexity has been put
forward at various times by different authors. One of the first of these was von Kdrmdn’s
expression (1934; 1939), which uses the approximation / ( i + ) = /4i In z+ B\ for the
transition region a,, < 2+ < ai, where the coefficients y4i and B\ (different from A and B)
are such that the sum /li In z+ + B\ takes at z+ = ctv and at 2+ = the same values as
the functions (5.20) and (5.25), respectively; in other words, it corresponds to the joining
of the points D q and D.\ in Fig. 25 by a straight line. Later, Hofmann (1940), Reichardt
(1940) and Levich (1962) Sect. 4, proposed other interpolation formulas for the values of
f( z+), a „ < 2 + < a / , which give a somewhat better correspondence with the results of the
measurements. A different idea was used in the works of Squire (1948), Loitsyanskiy
(1958), Frank-Kamenetskiy (1947), and in the very similar works of Rotta (1950),
Hudimoto (1941; 1951) and Miles (1957); here the region of flow is divided into two layers
only: z^ < a„ and z + > a v . In the first of these regions, the usual linear equation (5.20) is
used; in the second reg io n /( 2 +) is given by a more complicated analytical expression than
Eq. (5.25'), which tends asymptotically to Eq. (5.25') as 2 + - > 0 0 , and makes a smooth
juncture with Eq. (5.20) at 2 + = a „ . On the other hand, Ribaud (1940), Hama (1953),
Deissler (1955; 1959), Rannie (1956), Levich (1959) Sect. 25, Ts’ai Ko-yeng (1961) and
Tien and Wasan (1963) divided the flow into two layers 2 + < a / and 2 + ai and used the
usual logarithmic equation (5.25') in the second of these, and a more complicated
expression for f( z+) than Eq. (5.20) in the first. The works of Lin, Moulton and Putnam
(1953), Loytsanskiy (1960; 1962a) and Carr (1962), used a three-layer model (i.e., it was
assumed that f(z+) was given by different equations for 2 + < au, au<^ and for
24- > ^ i ) , but unlike von Kdrmdn’s work, in the first of these layers, a nonlinear expression
for f(z +) was adopted (different such nonlinear expressions being used in all cited works).
In addition, Loytsyanskiy (1960; 1962a) also proposed a single, but very complicated
expression for f(z+) which is applicable for all values of 2 +. Simpler, approximate
expressions, describing the behavior of f(z+) over the whole constant-stress region and
which are in fairly good agreement with all the data were proposed by Reichardt (1951a),
van Driest (1956), and Spalding (1961); one of Spalding’s two equations was also derived
later by Kleinstein (1967). A survey of work along these lines is given in Hinze (1961), Sect.
7, Longwell (1966), Sect. 8.7, and in the articles of Loitsyanskiy (1962a), Rotta (1962b)
and Kestin and Richardson (1963); a very detailed review (containing about forty different
equations) is included in the thesis of Coantic (1966).
The question of the detailed behavior of f( z+) as 2 + - > 0 has evoked widespread
discussion. In some of the above-mentioned works, it is assumed that f(z+) = z+ for
2 + < ; tto ; i.e., all derivatives of f ( 2 + )higher than the first are assumed equal to zero a t 2 + =
0; this assumption was later criricized by a number of authors. According to the equation
proposed by Ranney, /(< ?+ ) = - f - ___as 2 + -> 0, where 0; this result
clearly c o n tr a d ic ts th e e x a c t e q u a tio n ( 5 .2 0 ') . A t th e sam e tim e, the e q u a tio n s o f R e ich ard t,

R ib au d , Lin, M o u lto n and P u tn am , Carr and T ien and Wasan lead to an e x p a n s io n o f Eq.
( 5 . 2 0 ') w ith €4^ 0, w h ile a c co r d in g to H am a, D eissler, L evich , L o y ty a n s k iy and T a ’ai
K o -y e n g , C4 = 0 in th is e x p a n s io n . F in a lly , S p alding s im u lta n e o u s ly in tr o d u c e d tw o
d iffe r e n t e x p r e s s io n s fo r f ( z + ) w ith C4 ^ 0 and w ith C\ = 0. T he c o n sid e ra b le a tt e n tio n
given to th e q u e s tio n o f th e b eh a v io r o f f ( z + ) as z + - > 0 is e x p la in e d b y th e fa c t th a t for
s u ffic ie n t ly large Prandtl (or S c h m id t) n u m b e r s, th is b eh avior w ill h ave a grea t e f f e c t o n
tu rb u le n t h e a t-tran sfer (o r m ass transfer); th is wUl b e d is c u s se d in greater d e ta il in S ect. 5 .7 .
In th e m a jority o f th e w o r k s w e see a lso th at it is n o t th e fu n c tio n f ( z+) th at is c h o s e n as
th e b asis of t^ d is c u s sio n , but th e dependence on 2+ of th e e f f e c t iv e v is c o s ity
V - | - /C = x/p = V [ / ' (<2r^.)]“ \ T he p r o p o s e d fo r m o f this d e p e n d e n c e in s o m e cases,
is s e le c te d sim p ly o n th e basis o f the d ata, b u t m o r e o f t e n it is m o t iv a te d th e o r e tic a lly w ith

th e aid of som e so rt of q u alita tiv e a r g u m e n ts, u su a lly s u p p le m e n te d by a special
s e m iem p ir ic a l h y p o th e s is . H o w e v er , n o n ^ o f th e h y p o t h e s e s u se d has b e e n s tr ictly p r ov ed ; at
th e sam e tim e , th e s c a tte r o f th e curves = f ( z + ) p r o p o s e d b y th e variou s a u th o r s has th e
sam e order o f m a g n itu d e ( i f n o t greater) an d has th e sam e cha ra cter as th e sca tte r o f the
e x is t in g e x p e r im e n ta l p o in ts . C o n s e q u e n tly , at p r e se n t, th ere is n o ju s tific a tio n for c h o o s in g
any o n e “ b e s t ” e q u a t io n fo r / ( z + ) , an d in c h o o s in g b e t w e e n th e d iffe r e n t variants o f t h e s e
e q u a tio n s , o n e has o n ly to c o n s id e r w h ic h is m o s t su ita b le fo r solvin g an y arbitrary p r o b le m .
T h e p r e se n t situ a tio n c o n c e r n in g th e e q u a t io n s for the fu n c tio n s / i . . . / 4 , e tc ., is
c o n s id e r a b ly less sa tis fa c to r y . T h e se fu n c tio n s d e p e n d o n th e q u ite m y s te r io u s sta tistica l
regim e o f th e v e lo c ity flu c t u a tio n s very c lo se to a rigid w a ll. O n e o f th e m o s t gen eral ideas
on this regim e is th a t th e relative role o f th e n o n lin e a r term s in th e d y n a m ic e q u a tio n s
d ecreases w ith d im in ish in g d ista n ce fr o m th e w a ll. T h e re fo r e it see m s r ea son ab le th a t th e
f lu c t u a tio n s u \ v \ and w ' i n th e v is c o u s su b la y e r c o u ld be d escr ib e d b y lin earized e q u a tio n s .
O ne o f th e first c o n c r e te d e d u c t io n s from this h y p o th e s is w as m a d e b y L andau and L evich
early in 1 9 4 0 [c f. L andau and L ifsh itz ( 1 9 6 3 ) and L ev ich ( 1 9 6 2 ) ) ; th e y c o n je c tu r e d th at
th e linear e q u a tio n s m u st im p ly a c o n s ta n t tim e-sc a le fo r th e e n tir e v is c o u s su b la y e r , and
w ith this r eason in g, o b t a in e d th e p r o p o r tio n a lity o f - u ' w ' to y ^ . H o w e v er , this d e d u c t io n
is n o t r igorou s and th e m ore r e c e n t a n a ly tic d eriv a tio n o f the sam e resu lt b y L yatk h er
( 1 9 6 8 ) has even m o r e d o u b t f u l p o in ts in it. S o m e o th e r w o r k s (c ite d a b o v e ) o n th e m ean
v e lo c ity p r o file near a w all m ay also b e in c lu d e d in th e fa m ily o f in v e s tig a tio n s r elated to a
v is c o u s su b la yer; h o w e v e r , w e d o n o t th in k th at fu rth er d is c u s sio n o f th em h ere w ill b e
u se fu l.
More d e ta ile d in v e s tig a tio n s o f th e v e lo c ity flu c t u a tio n s in th e v ic in ity o f a w a ll w ere
m a d e b y S ternb erg ( 1 9 6 2 ; 1 9 6 5 ) and S c h u b e rt an d C o rcos ( 1 9 6 7 ) . It w as a ssu m ed in all
th ese w o r k s th a t tu r b u le n t flu c t u a tio n s are in itia te d in th e o u t e r parts o f a tu r b u le n t layer
and are tr a n sp o r te d to th e v is c o u s su b lay er fr o m th e o u tsid e ; also , th e s e w o r k s are all b ased
o n lin ea r iz e d d y n a m ic e q u a tio n s . T h e d iffe r e n c e b e tw e e n e a ch w o r k is in th e d iffe r e n t
sim p lific a tio n s o f th e e q u a t io n s and d if fe r e n t s e le c tio n s o f th e b o u n d a r y c o n d it io n s o n th e
o u te r b o u n d a r y o f th e su b la y e r . In all ca ses th e a u th o r s carried o u t prelim in ary c a lc u la tio n s
o f th e s ta tistic a l c h a r acteristics o f th e flu c tu a tio n s w ith in th e su b la y e r and o b t a in e d r esu lts,
w h ic h b y and large agree w ith th e data o n th e v e lo c ity f lu c t u a tio n s fo u n d b e fo r e 1 9 6 5 . It is
in ter e stin g to n o t e th a t E in ste in and Li ( 1 9 5 6 ) c o n s id e r e d a c o m p le t e ly d if fe r e n t p h y sic a l
m o d e l o f th e p ro c ess o f in itia tio n o f flu c t u a tio n s w ith in th e v is co u s su b la y e r (a ssu m in g th e
p ro c esses at th e w all to b e m o s t im p o r ta n t); n e v e rth e le ss, th eir e q u a t io n s tu rn o u t to be
id e n tic a l w ith th o se o f S ternb erg ( 1 9 6 2 ) [ th o r o u g h c o m p a r is o n o f th e th e o r ie s o f E in ste in
and Li and Sternb erg can b e fo u n d in t h e p ap er b y K istler ( 1 9 6 2 ) ] .
H o w ev er, all th e e x is t in g th eo r ie s m u st be c o n s id e r e d at p r e se n t as c o m p le t e ly
prelim in ary sin c e th e y d o n o t tak e in t o a c c o u n t th e la te s t p r o fo u n d e x p e r im e n ta l fin d in g s
o f K lin e, R e y n o ld s e t al. ( 1 9 6 7 ) , C o a n tic ( 1 9 6 7 a ; b ) , W illm arth an d B o Jang T u ( 1 9 6 7 ) ,
B a k e w e ll and L u m le y ( 1 9 6 7 ) , and o th e r s . E sp e c ia lly im p o r ta n t se e m to b e th e r esu lts o f
Kline an d his c o -a u th o r s, s h o w in g v ery clearly m a n y n e w featu re s o f th e flu c t u a tio n s w ith in

the v is c o u s su b lay er (e .g ., th e g reat part p la y e d b y v f lu c t u a tio n s and b y sharp variation s o f
th e f lo w in th e sp a n w ise d ir e c tio n ) , w h ic h w ere n o t ta k en in t o a c c o u n t in all previou s
th eo r ie s.
5.4 Flow Along a Rough Wall; Roughness Parameter and

Displacement Height
We now consider a turbulent flow near a rough wall with

protrusions of mean height /jothat are not small in comparison with
the friction length In this case the mean velocity profile will
be given by Eq. (5.15) which contains the universal function f ofz+
and a number of other parameters. Since functions of many'variables
are quite inconvenient for practical applications, the use of dimen
sional reasoning in this case is not very fruitful. In fact, however, the
situation is by no means as bad as this.
Let us assume that Once again, we consider the cases of
small and large values of z separately. For small 2 , comparable to the
height ho, we must expect that the mean velocity u(z) will depend
considerably on the form and mutual spacing of the irregularities of
the wall and will be different above the protrusions and over the
hollows between them. Thus here it is impossible to hope for any
simple general rules. However, in many cases we are interested
primarily in the mean velocities not too close to the wall, at
distances z which are large in comparison both with the friction
length 2 , = - ^ and in comparison with the mean height ho of the
protrusions. Naturally, it is considered that at such distances from
the wall neither the viscosity nor the local properties of the surface
will have any effect; here the essential feature will be only the
presence of a constant flux of momentum in the negative direction
o f the O2 axis with a “ sink” on the plane z = 0. Therefore, both the
deductions o f Eq. (5.22), applied to a flow in a half-space bounded
by a smooth wall, for 2 ^ Aq > still hold in the case o f a rough
wall. Consequently,
u{z) = A u, In z-{-'Ai for (5.22a)
Since the constant A = —in Eq. (5.22a) gives the value of the mean
velocity gradient corresponding to a given momentum flux

it must have the same value 2.5) for a rough wall, as for a smooth
wall. (The same must be true also for all the characteristics o f the
velocity fluctuations in the logarithmic layer above a rough wall, e.g.,
for the values of the ratios = A i , a^lu^ = A 2 , and o^lu^, =
A 3 ). However, as far as the coefficient A i is concerned, this is
determined by the boundary conditions on the lower boundary of
the region of applicability of Eq. (5.22a); consequently, it depends
on the variation o f the mean velocity in the immediate neighborhood
of the wall, which differs considerably depending on whether the
wall is rough or smooth. We note that the origin of the distances z
for z ho may be chosen arbitrarily between the bottom and top of
the protrusions o f the wall, without perceptibly changing the results;
in practice, it is chosen usually from the condition of best agreement
of the observed profile m(2 ) with the logarithmic equation (5.22') [we
shall return to this later in the subsection].
Equation (5.22a) for the velocity profile above a rough surface
may be written in dimensionless form in various different ways. We
may, of course, use Eq. (5.25), which is applicable in the case of a
smooth wall, i.e., putting
u{z) = u , ( A \ n for (5.25a)
However, here B will no longer be a universal constant, but will

depend on the dimensionless parameters which determine the sizes,
forms and positions of the irregularities of the wall: B — b ( - ^ ^ ,
a. P, •••)• Moreover, writing Eq. (5.22a) as Eq. (5.25a), although
convenient for comparison of the mean velocity profiles close to
smooth and rough walls, is not really natural (in any case, whenAo;^>
v/«.). In fact, the right side of Eq. (5.25a) contains v, while, as we
would expect with /io ^ v /« * , the mean velocity profile (and
therefore the value of the coefficient/li in Eq. (5.22a) as well) is, in
general, independent of the viscosity and will be determined solely
by the size, form and relative position of the irregularities of the wall
which completely determine the nature of the flow in the lowest
layer. Thus for it is more convenient to represent the
mean velocity profile as
u (z ) = u, [ a In i - ^ + B ' y (5.28)
We may now expect that for sufficiently great the value of B' =
A \n will not change with variation of the parameter ,

V
containing the viscosity v (i.e., it will be a function of the parameters
a, P, . . . which describe the form and relative position of the
irregularities only). It is also possible to rewrite Eq. (5.25a) in the
form
zu^
u (z) = Ain — + B o , (5.25b)
where B q ^ 5 is the universal constant of Eq. (5.25) for a velocity

profile above a smooth wall and - v exp {{Bq - B) I A) \ may be
called the equivalent viscosity [it obviously depends on the wall
roughness, namely, Vq = v -fi {hoU^lv, a j3, . . .)]. Finally, we can try
to express the co n sta n ts, in Eq. (5.22a) directly in terms of the skin
friction characteristics of the fluid on the wall, which depend in
some unknown manner on the geometrical properties of the wall. As
a dimensionless skin friction characteristic we normally use the
friction coefficient (or friction factor)
(5.29)
where U is the characteristic velocity of the flow. However, for an

idealized flow in an infinite half-space, there is no characteristic
velocity. Thus here we may only form the dimensionless coefficient
c^{z) = 2
which depends on height, and is consequently of little use. But if we

substitute
u{ z) =
into Eq. (5.22') we obtain the following simple law:

= .4 In ^ .
]/^ 4 V
Hence we see that within the logarithmic layer
- K l v w ,
Thus the quantity
(5.30)
which has the dimension of length, is independent of 3. This quantity

is an objective characteristic of the dynamic interaction of the flow
with the wall, dependent, naturally, on the irregularities of the wall;
it is called the roughness parameter, the roughness length, or simply
the roughness. Now substituting
-1
into
U{Z):
we obtain
consequently,
According to Eq. (5.31) the roughness parameter may also be

defined as the height at which the mean velocity of the flow would
become zero if the logarithmic equation for u (z) were applicable
down to this height; in fact, of course, the logarithmic equation
ceases to apply at much larger values of z .
To find the dependence of and Zo on the height o f the
protrusions, we need to have results from experiments in which only
their size but neither the shape nor the relative position is varied.
Very careful experiments of this kind were carried out by Nikuradse
(1933) in circular tubes. The walls of the tubes were coated with
sand grains of a given size, which varied from experiment to
experiment, and were glued on as close together as possible. The
dependence relationship o f the coefficient B' in Eq. (5.28) on
obtained in these experiments is shown in Fig. 28. (The light and
dark dots here correspond to the two different means of defining fi':
the first, based on direct comparison o f the measured velocity profile
with Eq. (5.28), the second described in Sect. 5.5.) For a
homogeneous sand roughness with log < 0 .6 , i.e., w i t h - ^ ^ <
4, and equation of the type
B' = A In where A = 2.5, fi = const( « 5.5)
will hold. Thus, for Aq < 4 we have Ve = and the velocity profile
will be completely independent o f ho, i.e., the wall may be considered
to be dynamically smooth'^ ; we observe that with these values of /to ,
the protrusions of the wall will be immersed entirely in the viscous
sublayer; this explains the fact that they exert no effect whatsoever
on the flow in the logarithmic layer. With 0.6 < In < 1.7, i.e.,
with 4 <; 60, a transition regime occurs in which the peaks

of the protrustions project from the viscous sublayer and create
additional disturbances, leading to a specific dependence of both the
coefficients B and 8 ' on •V
the wall in this case must be
*^The te r m s hydraulically smooth o r aerodynamically smooth are also fr e q u e n tly u s e d in

th e liter a tu re . H o w e v er , th e term a d o p te d in th e t e x t is a d v a n ta g e o u s sin c e it is u n re la te d to
th e s p e c ific n a tu r e o f th e flu id . T h e sa m e rem ark is a lso true regarding th e te r m s d escrib in g
all o th e r r o u g h n e s s reg im es.
lo g
FIG . 2 8 . D e p e n d e n c e o f th e c o e f f ic ie n t b ' o n ^°^*~ a c c o r d in g t o N ik u r a d se ’s d a ta ( 1 9 3 3 ) .
considered to be dynamically slightly rough. Finally, with l o g - ^ ^ >

1.7, i.e., with > 60, the viscous sublayer practically ceases to
exist, while the flow in immediate proximity to the wall consists
entirely of eddies, generated by the flow around the individual
protrusions; here the mean velocity profile u (2 ) is independent of the
coefficient of viscosity v, and B' is a constant (according to the data
of Fig. 28, 8.5). In this latter case the wall may be called
dynamically completely rough. O f course, whether the wall is
dynamically smooth, slightly rough or completely rough depends not
only on the nature of its surface, but also on the values of u, and v
(i.e., on the Reynolds number of the flow). The values of the three
parameters B, B', and Zo for dynamically smooth and dynamically
completely rough walls according to Nikuradse’s data (which refer
only to a homogeneous sand roughness) are equal to
B ^ 5.5. B' « 2.5 in + 5.5, 2 o~ i i r smooth wall
B ■x. — 2.5 in + 8.5, B' as 8.5, = ^ completely rough wall
If we consider the data on the logarithmic layer above a

homogeneous sand-roughened wall to be sufficiently complete, we
may associate with roughness of any other type the height o f the
equivalent sand roughness ha, which corresponds to the same

logarithmic velocity profile for the same t o . For a number of
artificial completely rough surfaces covered with geometrically
regular protrusions, this height hs was determined experimentally by
Schlichting (1936) [see also Schlichting (1960), Chapt. X, Sect. 7].
In Schlichting’s book, we may also find additional data and
bibliographical references on the heights of the equivalent sand
roughness both for artificial rough surfaces and for a number of
ordinary surfaces used in engineering (concrete, cast-iron, steel, etc.)
which in many cases prove to be dynamically completely rough.
Similar data can be found in the book by Longwell (1966), papers by
Chamberlain (1966; 1968) and some other sources.
Another procedure for describing the logarithmic velocity profile
close to walls of differing roughness consists in determining the
corresponding decrease in mean velocity u in comparison with the
flow above a smooth wall with the same shear stress to. The presence
of irregularities always leads to the smoothing of the velocity profile
in the immediate vicinity of the wall, i.e., to the slowing down of the
rate of increase of the mean velocity for very small values of z.
Therefore, in the case of a rough wall, the value of the velocity on
the lower edge of the logarithmic layer, and hence throughout the
layer, is less than that for a smooth wall. Using Eqs. (5.25) and
(5.28) we may write
+ (5.32)
where A = const « 2.5; B = B q = const « 5. Thus - ^ i s independent

of z, and for a dynamically completely rough wall will depend
linearly on I n T h e dependence between and In
will become nonlinear for the case of a slightly rough wall,

while for values o f /lo where the wall becomes dynamically smooth,
the difference 6 m is equal to zero. A considerable amount of
experimental data on i.e., on the values of 8 ' , for different types
of roughness in zero-pressure-gradient boundary layers and flumes
were collected by Hama (1954) [see also Clauser (1956), Robertson
(1957) and Perry and Joubert (1963)]. Figure 29, borrowed from
Clauser’s survey paper (1956), gives some of the values obtained for
We see that for sufficiently large values of will, in fact,
be a linear function of In which corresponds to B' = const (case

of a completely rough surface). However, transition to _an all
intermediate regime, with deviation o f the dependence of on In
from linearity, for different types of roughness occurs at
different values of —^ (generally, between 30 and 1 0 0 , according to
the data of Fig. 29); moreover, the character of the dependence of
^ on is completely different for different types of slightly
rough surfaces. We note, further, that Fig. 29 allows us to determine
very easily, for any roughness represented on it, the height AjOf the
equivalent sand roughness, i.e., the sand roughness corresponding to
the same value of For completely rough walls (but only for
them) the height hs will obviously be proportional to the height ho;
FIG. 29. E ffe c t o f w a ll ro u g h n ess o n th e rela tiv e d e c re a se o f th e m ea n

v e lo c ity 6 m/m^ [a c c o r d in g to C lauser ( 1 9 5 6 ) ] . T h e v a r io u s s y m b o ls o n
th e figure in d ic a te da ta referring to d iffe r e n t ty p e s o f r o u g h n ess. T h e
d a sh e d lin e c o r r e sp o n d s to da ta fo r h o m o g e n e o u s sand r o u g h n ess.
hence the roughness parameter = will in this case also be

proportional to ho.
We have already mentioned that the logarithmic dependence of
the mean velocity on the height must also hold for the wind velocity
in the lower layer of the atmosphere (thickness of the order of
1 0 - 1 0 0 m) for neutral (or close to neutral) thermal stratification
[this fact, apparently, was first pointed out by Prandtl (1932a)] .The
experimental verification of this was obtained by Sverdrup (1936),
Paeschke (1936) and many other researchers. [See, e.g., Sutton
(1953) and Priestly (1959a)]. in a number of cases observations on
the wind profile were accompanied by direct measurements of the
shear stress to; for a discussion of the methods of such measurement
see below. Sect. 8.3. Since y m a y be determined from the
slope of the logarithmic profile, assuming x as 0.4, >1 = ^^ « 2.5 is
known from laboratory experiments, independent measurement of to

allows the logarithmic equation (5.31) to be verified far more
reliably than is possible when only observations of the vertical
variation of wind velocity are available. It is only necessary to keep
in mind that the accuracy of each individual measurement of to is
usually not very high, thus sufficiently reliable results can only be
obtained on the basis of considerable data [see, e.g., Perepelkina
(1957)].
All methods of determining the friction velocity show that its
values normally lie between 10 and 1 0 0 cm/sec in the atmosphere.
Since for air v 0.15 cm^/sec, it is clear that the underlying land
surface of the atmosphere is almost always completely rough. The
logarithmic profile of the wind velocity may conveniently be put in
the form (5.31), using the concept of the roughness parameter zo-
Moreover, in the atmosphere, the correct choice of the origin of
height 2 quite often is important. In fact the sizes ho of the
irregularities of the ground here may have comparatively large values
(for example, for tall grass, or fields covered with crops); at the same
time increasing the height of the measurements z often proves quite
frustrating, since as z increases, the role played by buoyancy effects
also increases rapidly (see Chapt. 4 of this book). It is important here
to consider the question of the form of the velocity profile at heights
only slightly greater than the mean height z of the protrusions of the
ground.
For z ^ ho, the parameter ho (of the dimension of length) will also
influence the mean flow; as already seen, the viscosity v for a
completely rough wall may be ignored. Hencedu/d^ will be a function
of T o , p, 2 , and ho, from which we can form one dimensionless
combination halz. It then follows that instead of Eq. (5.21) for the
mean velocity gradient, the more general equation
du(z) _ u, (5 3 3 )
dZ 7.Z ^ \ 2 I
will hold for z ^ Hq, where x « 0.4, and ^ ( ■ ^ ) is a correction

function. The correction to the usual equation (5.28), of course,
must cease to play a role a t z ^ / i o , where it follows that ^f(0 ) = 1 .
Taking into account - ^ < 1, we expand the function as a
power series in ~ , and obtain
du
dz
We now introduce a new zero level, putting z = Zi + z ', where z\ is

fixed and has the same order of magnitude as ho, and z' is greater
than both Zi and h». Let us now replace z in the right side of Eq.
(5.34) by z ' , and instead of expanding in powers of use an
expansion in powers of It is easy to see that the leading terms of
this new expansion will take the following form;
du __ M,
dz ~~
We now choose Zi in such a way that the first-order terms in this

expansion become zero (i.e., we p u tz i = aho = g ' { 0 ) h o ) . Then with
accuracy to second-order terms we obtain
dz r.(z “*
— z{) ’ (5.36)
where
u(z) = ^ \ n ^ ^ . (5.37)
We see that when z is not too small, the effect of the finite height ho
in the first approximation leads to the velocity profile being

logarithmic, and the heights being measured from the given level
z = 2 i and not from 2 = 0. The height Zi may be called the
displacement height or zero displacement (by analogy to the concept
of the displacement thickness in boundary layer theory).
Data on the values of the roughness parameter zo and the
displacement height z, for various types of natural surfaces may be
found widely in the literature [see, for example, Paeschke (1936);
Deacon (1949); Sutton (1953); Ellison (1956); Priestly (1959a);
Tanner and Pelton (1960); Lettau (1967); Zilitinkevich (1970)].
Unfortunately, these data do not agree very well with each other
(probably because both parameters zn and Zi depend on fairly fine
details o f the underlying surface). As is obligatory for a completely
rough surface, the roughness parameter Zq., determined according to
the wind profile, may often be assumed approximately proportional
to the height hr, of the protrusions of the ground. However, for both
ordinary grass cover and for agricultural crops, the proportionality
coefficient is considerably greater than the coefficient 1/30 (which,
according to Nikuradse’s data, corresponds to a homogeneous sand
roughness) and, generally, is close to 1/10 or even to 1/5. (Tanner and
Pelton, for example, recommend Zq ^ h ^ ll.S as a mean estimate for
different grasses, crops, and bushes.) It is interesting that the results
on /?o/zo obtained by Chamberlain (1966; 1968) for several forms of
artificial roughness elements in a wind-tunnel are very close to the
existing results for land with natural vegetation. O f course, the ratio
ho/zQ must not be a strict constant for a set of irregularities of
different forms; thus, it is not surprising that one of the empirical
relationships obtained for different plant populations has the form
Zo / 1 1.7 [see Lettau (1967)]. The displacement height Zi,
for vegetation o f not too great a height, may generally be taken equal
to zero; however, for high vegetation it must often be taken between
ho and -j-.
Further, we note that for high grass cover, the roughness
parameter may also depend on the mean velocity of the wind, which
bends the stalks and hence changes the form of the surface. For
example. Deacon (1949), found that for tall grass (about 60 cm
high), Zo may vary from 9 cm in a very slight wind to about 4 cm in a
strong wind. This is similar for water waves, where both the height
and form o f the irregularities also are clearly dependent on the wind
velocity. If, in studying wind-waves, the viscosity and surface tension
of the water are neglected and the roughness of the sea is defined
entirely by the local atmospheric conditions (i.e., by the local value
of the shear stress only), then from dimensionality arguments we
obtain for the roughness coefficient an equation of the form
Here g is the gravitational acceleration, ul — tolp (where p is the air

density), and the constant b may, in particular, depend on the ratio
of the densities of air and water. The observations of the wind
velocity profile over the sea obtained till now are still very
incomplete and inexact; moreover, they do not leave the impression
that the relationship (5.38) is well justified experimentally. (We note,
for example, that Chamock (1958b) and Ellison (1956) who used
this equation obtained very different values of b\ according to
Chamock, 6 » 1/80 and according to Ellison 6 1/12. Later,
Kitaygorodskiy and Volkov (1965) [see also Kitaygorodskiy (1968)
and Zilitinkevich (1970)] concluded that there is a very wide range
of conditions in which Eq. (5.38) is approximately valid with b ^
1/30. However, careful laboratory experiments by Hidy and Plate
(1966) and Wu (1968) imply the estimate b ^ 1/90 and the data of
field observations collected by Phillips (1966) are in agreement with
the value b ^ 1/40.) In this connection, many researchers are
inclined to consider that on the sea Zq does not depend just on the
local meteorological conditions [but, for example, on the wind fctch,
which characterizes the path of the wind over the sea; cf., e.g., Hino
(1966)]. At the same time, all the existing data indicate that the
surface of the sea is dynamically considerably smoother than the
majority of land surfaces on earth; even for a fairly strong wind on
the sea 2 o< 0.1 cm [according to the data of several authors, the
surface of the sea may even be taken to be dynamically smooth
provided the wind is not very strong; see Sutton (1953); Deacon
(1962)]. Therefore, in the study of winds over the sea, the concept
of the roughness parameter is used comparatively rarely; more often,
instead of Zo, in this case we use the values of the friction coefficient
Cj, referred to the velocity U = u{z) at some fixed height z [see, for
example, Wilson (1960); Deacon (1962); Deacon and Webb (1962);
Roll (1965); Phillips (1966)]. In any case the question of the
roughness of sea surface is, at present, far from being completely
clear.
5.5 Turbulent Flows in Tubes and Channels;

Skin Friction Laws
The general laws of turbulent flow along a rigid wall discussed
above, refer to a wide class o f flows. These include, in particular,
flows in channels and tubes. We have already used some data on
turbulence in channels and tubes in the two preceding subsections.
However, flows in channels and tubes also possess some specific
features lacking in the idealized flows in the half-space 2 > 0 with
which we have dealt so far. First, in flow in a channel or tube unhke
a flow in a half-space, there is a characteristic length A/i (the
half-width of the channel) or R (the radius of the tube) and a
characteristic velocity Uo (maximum mean velocity in the middle of
the channel or on the axis of the tube) or Um (average bulk velocity
defined by
fit
^ fu { z )d z
or
ju (r)rd r
Thus, it is clear that the theory of flows in a constant stress layer

close to a wall does not exhaust the whole theory of flows in
channels and tubes. We shall consider this below.
We begin with general similarity considerations. For simplicity, let
us assume first that the walls are dynamically smooth. We shall
consider only the flow at sufficient distance from the intake into the
channel or tube, for the intake conditions to have no effect.
Moreover, we shall assume that the flow is steady and completely
turbulent. In this case only the longitudinal component of the mean
velocity u = u* will be nonzero, and all the statistical characteristics
of the turbulence will depend on the single coordinate 2 , the distance
from the wall of the channel or tube (for a tube z = R — r, where r is
the distance from the axis). For a channel or tube of a given size and
fluid of given density and viscosity, we shall have a one-parameter set
of flows, defined by the value of the “ pressure-head” —the constant
longitudinal pressure gradient. The pressure gradient will also define

uniquely such characteristics of the flow as the velocities Uq and Um
and the shear stress at the wall to, or, otherwise, the friction velocity
Thus, the values of the statistical characteristics of the
flow at a distance z from the wall may depend on the following
parameters: p; v; 2 ; A/, (for a tube we shall now denote the radius by
Hi, i.e., we put Hi = R) and one o f the velocities t/o. Um or «*. From
these parameters we may formulate two dimensionless combinations:
the Reynolds number 5 = (it will be convenient to choose as
the characteristic velocity) and the dimensionless distance 71 = .

ni
Therefore we have
« (z) = . - ^ ) = a.cp (I 7)): (5.39)
and this more general equation now replaces the law of the wall
(5.13).«
Generally speaking, the mean velocity profile of a flow in a
channel or tube is defined by a function of two variables (which, of
course, can be different for both a channel and a tube). However,
there are two im portant hmiting cases in which the function <p(|, ti)
can be expressed in terms of a function of one variable.
The first of these cases (considered in detail in Sects. 5 .2 -5 .4 ) is
that of small 2 , i.e., tq ~ <C 1 • In this case the dependence of the
velocity u{z) on the length//i becomes negligible, so that Eq. (5.39)
is transformed into
for (5.40)
(where is some constant considerably less than one) which is Eq.

(5.13). The second hmiting case (which we have not met before) is
that of large values of rj corresponding to the “ turbulent core”
of a flow, close to the center of the channel or the axis of a tube. In
this region of large values of r], remote from the walls, we may
^ E q u a tio n (5 .3 9 ) is, of c o u r se , e q u iv a le n t to th e e q u a tio n u (z) = u ^ f{z u ^ /v )

( 2 // i- z ) w * /i^ ) , w h ic h w a s fo r m u la te d in a f o o t n o t e in S ect. 5 .2 fo r a sp ecia l ca se o f a
c h a n n e l flo w .
assume that the arguments used in Sect. 5.3 for the justification of
the logarithmic velocity profile will apply. That is, since the
turbulent shear stress here is many times greater than the viscous
stress, there is reason to think that in the “ turbulent core” the
viscosity will have no direct effect on the flow (although it will affect
it indirectly, since the boundary conditions on the boundary of the
“ core” depend on it, and also the value of «„ which by Eqs. (5.17)
and (5.17') determines the distribution o f the turbulent stresses
through the whole flow). In the absence of a characteristic length
scale, from here we obtain Eq. (5.21), leading to a logarithmic
profile; when we have such a scale//i, we obtain a more general
equation of the form P’^tting the unknown
function (pi in the form T i(^ )= — f ' M where f i ( l ) = 0 , and
integrating the expression for from some value of 2 to z = H\,
we find
(5.41)
Equation (5.41) is usually called the velocity defect law and the
region of its applicability is therefore sometimes called the defect
layer. The law (5.41) was first formulated by von Karman (1930) for
tube flow, on the basis o f the experimental data of Fritsch (1928).
Later, the correctness of this law was carefully verified many times
FIG . 3 0 . V e r ific a tio n o f th e v e lo c ity d e f e c t la w for

tu rb u le n t flo w in a tu b e a c co r d in g to L a u fe r ’s d ata
(1 9 5 4 ).
14
12
10
%-u(zf
> IV
k ,
u, 8 f t
-^ 4
2
0
0.01 0.02 0.04 0.05 0.10 0.20 0.40 0.60 1.0
zjH,
FIG. 31. V e r ific a tio n o f th e v e lo c it y d e f e c t la w fo r tu r b u le n t

f lo w in a r ectan g u la r c h a n n e l a c c o r d in g t o th e d a ta o f L a ufer
(A) a n d (+ ).
and all the results were affirmative (see Figs. 30 and 31). Now this
law is usually considered as a special case o f the general principle of
Reynolds number similarity which holds with considerable accuracy
for a wide class of turbulent flows. According to this principle, for
sufficiently large Reynolds numbers (with v < ^ /C ~ UL, where U and
L are characteristic scales of velocity and length) for a relatively large
region of turbulent flow (generally including almost the entire flow
with the exception of comparatively thin layers adjacent to the
walls), the mean motion will not be dependent directly on the
viscosity coefficient (i.e., on the Reynolds number) which will affect
the flow only through the boundary conditions and the value of U
[see Townsend (1956)]. We must also note here that this Reynolds
number similarity will occur not only for the mean velocity, but also
for all the statistical characteristics o f the flow that are not
connected directly with the viscous dissipation of kinetic energy, i.e.,
for example, it holds for or u'w', but not f o r ^ ^ - j^ . We shall
return to this im portant assertion at the end of the present
subsection.
Now we return to the mean velocity distribution and consider the
im portant argument of Izakson (1937) [later developed in the works
of Millikan (1938) and von Mises (1941)], which permits us in
certain cases to establish the exact form of the functions / and f i. Let
us assume thatrii < tio, so that there exists some “ overlap interval”
1< Ti < Tio of values of tj , for which the general equation

(5.29) may be put simultaneously in the form (5.40) and in the form
(5.41) [this assumption, of course, cannot be justified theoretically,
but must be verified by experimental d a ta ]. In this case, adding Eqs.
(5.40) and (5.41), we obtain
/(5^) + /i ii) = / 2 (?) for ■'ii < < Tô. (5-42)
where [2 (1 ) denotes the value of — which, clearly, can depend only
on the Reynolds number ^ , but not on vj — . From the

functional equation (5.42) it follows easily that all three functions
f, fI, and f 2 are logarithmic functions. In fact, differentiating Eq.
(5.42) first with respect to I and then with respect to ri, we obtain
thus it follows that fUr]) = A In |ri + B, where A and B are

constants of integration. Thus we arrive once again at Eq. (5.25),
starting this time from different assumptions. Substituting the
expression for /(|ri) into Eq. (5.42), we confirm that
/ j ("/)) = — v 4 I n T j - I - B 2, B,^ — Bi = B,
Thus the velocity defect law in the overlap region t?j < tj < tJq must
have the form
-^ o - “ (^). ^ _ l | n 4 - + B^. (5.43)
If we assume, in addition, that Eq. (5.43) may be applied right up to

the value z = H\ (i.e., t] = 1), then obviously Bi = 0 (this last
deduction, of course, does not follow automatically from the
preceding arguments since they apply only to the region tji < t i <
tio). Finally, the logarithmic equation for f 2 gives the following
dependence of the r a t i o o n the Reynolds number;
- ^ = 1 In (5.44)
Introducing the friction coefficient (or friction factor)
instead o f w e can rewrite the last equation in the form
(5.45)
(Later, we shall see that this simple derivation of the logarithmic

equation and friction law may be applied to a number of problems;
see Sect. 5.7 and the end of Sect. 6 .6 .)
Until now, we have always assumed that the walls of the channel
or tube are smooth. However, it is easy to see that the whole
argument may easily be transferred to the case of a channel or tube
with rough walls. In this case q)(|, ti) in Eq. (5.39) and filt]) in Eq.
(5.40) may also depend on the additional arguments Aqu./n (or/io///|),
a, p, . . . , which define the sizes, forms and relative positions of the
irregularities of the wall. Nevertheless, it is natural to think that in
the “ core” of the flow the presence of roughness will be felt only
through the boundary conditions and the value of the turbulent
shear stress (depending on the value of the friction on the wall), and
not directly; if this is the case, then the same equation (5.41) will be
true for both smooth and rough walls. However, assuming that the
regions in which Eqs. (5.40) and (5.41) are satisfied, partially overlap,
we obtain again the functional equation (5.42), with the only
difference being that now f and fa niay also depend on additional
parameters characterizing the roughness. Hence, as previously, it
follows that for the overlap region tji < t] < t)o all three functions
/, fi, and fi must be logarithmic with common coefficienti4= 1/x for
the logarithm; consequently, this coefficient must be a universal
constant (just as the coefficient S, is). For B and B2, these may
contain some common summand depending on the size and nature of
the roughness. If we assume that Eq. (5.43) is applicable right up to
z = H\, then B, = 0 and = B; thus on this assumption the data on
the friction factor Cf allow the value of B (or B' of Eq. (5.28) which
is simply connected with B ) to be determined at once. The values of
B' for different represented by the dark dots in Fig. 28 were

obtained in exactly the same way.
In the special case of a dynamically completely rough wall, the
function
must be independent of the viscosity v, i.e., it must be a function

only of ^ —
U ^/I q AZq
a, B,. . . . Therefore, it follows that the function
foil, — a, S, . . . ) will also depend only on ^ = ^ ,

a, p , ___Consequently, in this case Eqs. (5.44) and (5.45) take the
form
f/o 1 //|
+ (5.44')
In (5.45')
Y~c, r.yT *0 )/^2 ’
where the coefficient may depend on the form and relative

position o f the irregularities of the walls.
Considerable experimental data confirming the correctness of Eq.
(5.40) [which is the same as Eq. (5.13)] for turbulent flows in tubes
and channels was collected in Fig. 25. Of course, if it is to be possible
for the distances z . = - ^ to be transformed into values of the
variable r i = - ^ , we must know the number f = which is a

'M l ^ V
(J H
complicated function of the Reynolds number Re = ■ , given
implicitly by Eq. (5.45). As a guide, the table below is given to
indicate the dependence of the lower boundary 8, = 30 of the
logarithmic layer, for flows in smooth tubes, on the most widely
U^D
used Reynolds number Re^, = — — . (We shall discuss the connection
between the maximum velocity Uo and the bulk velocity Um later.)

With increase o f the Reynolds number, the ratio StM, naturally

decreases, but not unboundedly. For every real tube, as the Reynolds
number increases, an instant finally occurs when it appears to be
completely rough. Then the lower boundary of the logarithmic layer
begins to depend only on ho and does not vary with further increase
of the mean velocity of flow.
Re^, 5-103 10^ 105 106
0.2 0.1 0.012 0.0016
For the upper boundary of the values of z for which the

logarithmic equation (5.25) is applicable, it was already pointed out
that, according to Nikuradse’s data, this value, for both smooth and
rough tubes, over a wide range of Reynolds numbers is quite close to
the value z = R. (Thus the logarithmic layer can be extended
without gross error to the center of the tube.) However, for the
logarithmic equation to be apphced to the whole tube flow, it was
necessary for Nikuradse to change the coefficients A and B
somewhat in comparison with the values which gave the best
agreement for the part of the flow next to the wall. It is clear,
therefore, that the logarithmic equations proposed by Nikuradse for
the whole flow cannot be considered to be in agreement with the
theoretical equation (5.25) which is correct for 6 ;///i < !!< % . In fact,
the coefficients A and B in the theoretical equation (5.25) must be
defined on the basis of the extensive data relating only to the part of
turbulent flows adjacent to the wall; using only these coefficients, an
attem pt can be made to elucidate whether the equation for the
velocity profile relating to the logarithmic layer will break down in
the central part of the tube. Millikan (1938) was the first to
attem pt such an undertaking. He used almost all the scant data
on flows in tubes and channels known at the time. Especially
noteworthy is that for the values of A and B in Eq. (5.25), Millikan
obtained A = 2.5, B = 5.0, which differ very little from what are
now considered to be the most reliable values (see above. Sect. 5.3).
Further, according to his estimates, the deviations of the mean
velocity profile in a circular tube from the values given by Eq. (5.25)
are completely unobservable up to approximately rj = 0.1 (i.e., up to
a 10 % change in x), while for all t] they do not exceed 1 0 % of the
corresponding value ofti(z). In spite of the fact that, at present, the

quantity of data on the velocity profile in circular tubes has
increased considerably since 1938 [see, for example, the survey
article of Hinze (1962)], Millikan’s deductions remain entirely valid
even now.
From Nikuradse’s results it is also clear that if both A — Ijt. and Si
in the logarithmic form (5.43) of the velocity defect law are
considered as empirical constants, then we may assume that this
equation will apply right up to z = Hi, i.e., it may be in good
agreement with the experimental data even forBi = 0 . Theoretically,
however, A cannot be considered as an arbitrary constant, but must
be defined on the basis of measurements of the flow close to the
wall. To obtain the best agreement, here we must take B ^ 0 [see,
for example. Fig. 30 which is borrowed from Hinze (1959) and
plotted on the basis of Laufer’s measurements (1954); the value A =
2.44, i.e., X ==» 0.41 is chosen here in accordance with Clauser’s
recommendations (1956)]. Similar results are obtained for flows in a
rectangular channel (see Fig. 31, borrowed from Schubauer and
Tchen (1959), where A is put equal to 2.5); the deviation from the
logarithmic equation (5.43) in both cases begins at about ii= 0.25.
The problem of selecting an analytic equation forfi(Ti), which gives
good agreement with the data over a large range of values of t], is
considered in a large number of works [see, e.g., Gosse (1961)].
However, in practical problems usually it is possible simply to
consider that the mean velocity profile in a tube is described by the
logarithmic equation right up to the axis. Such a profile is very
different from the parabolic profile of
laminar Poiseuille flow. In fact due to
the far more powerful radial mixing in
turbulent flow, the velocity profile is
everywhere (except in the thin layer on
the walls) considerably more uniform
than in laminar flow (see Fig. 32).
Further, we note that with t) > 2 .5 ,In
differs only slightly from 2.03tiV2; thus
it is not surprising that in the first
careful measurements of the turbulent
velocity profile close to the axis of a
tube carried out by Darcy (1858), an
FIG. 32. Comparison of the mean . . , ^
velocity profiles for laminar (L) em pirical form ula was obtamed w hich in
and turbulent (7^ flows in a tube. OUT n o ta tio n haS the form
U ,-u (z)
= 5 .0 8 (l— ^ ) ^ (5.46)
Taking the velocity profile to be logarithmic right to the axis of

the tube, and ignoring the thin layer on the walls of thickness 8, =
30 i'/m*, to which the logarithmic equation is not applicable, we may
also formulate a simple relationship between the maximum velocity
on the axis o f the tube and the averaged bulk velocity t/ni-
Integrating Eq. (5.43) with i5 i= 0 , multiplied by 2n (//i — z) from z =
0 to 2 = //i, and dividing the result b y it//i, we obtain
u .-u „
. — _— I" (1 — 7))ln7)rf'ii = const s»3.75.
According to Eq. (5.44), for smooth tubes
- ^ = ^ l n ^ + 5 ; , B', = B , - 3 . 7 5 . (5.47)
This equation is also well confirmed by experiment.

Equation (5.47) allows us to formulate the skin friction law, i.e.,
the dependence on the Reynolds number R e D = = - ^ —of the new
dimensionless friction coefficient X= — = 8 ( 7^ ? , which differs
from the coefficient Cf in Eq. (5.45) not only by the unim portant
numerical multipHer, but primarily by substituting the more easily
measured bulk velocity Um for the maximum velocity Uq- If in Eq.
(5.47) we write in the form of a product ^ , we
obtain
B a — 3 . 7 5 ----- - \ n 4 V J
1 /T
— W • In (RCyj
xK F
(Re„o yVT)
y T^ + - | - B;
- r ) 5,, 5 ,3 = --------^ y-g-
^
(5.48)
Here K is determined easily from the known pressure drop A,/? =

along a section of the tube of length / and from the discharge
Q= —^ Uja; thus the data on the dependence of X on Reo is

considerable. Nikuradse (1932) showed that all these data are in very
good agreement with Eq. (5.48) with the following values o f the
coefficients; — 0. 87 (corresponding to y. ~ 0.41), 8^= - 0 .8
64
[see Fig. 33, where together with the curve (5.48), the line X =
is also plotted corresponding, by Eq. (1.26) to the skin friction law

for laminar flow in tubes]. According to Eq. (5.48), % decreases
without limit as Rep increases; of course, this is not actually the case,
since as Re^ increases, every tube will finally be completely rough.
Once this happens, the friction coefficient X will be independent of
the Reynolds number and will be defined with good precision as
(5.48')
which follows from the assumption that Eq. (5.28) with B '= const is
appUcable to all z for 0 < z< H i. This behavior of % is illustrated
clearly by the data of Fig. 34.
If great accuracy is not required, we may assume that for
turbulent flow in a plane channel of width 2 //i, the logarithmic form
of the velocity defect law (5.43) will also hold right up to z = //i,
which corresponds to the center o f the channel (in this case, of
course, Bx must equal zero). The same equation is often used in the
case of flow in an open plane channel (or a flume) of depth H\,
bounded above by the free surface of the fluid; in a channel of this
type, the maximum velocity Un will clearly be attained on the upper
surface, so that here also, S i= 0. In addition, in the case o f an open
channel, sometimes instead of Eq. (5.43) the following equation of
von Karman (1930) is used:
Even better agreement with the experimental data may be obtained

with the more general equation
FIG . 3 4 . T h e skin fr ic tio n c o e f f ic ie n t X as a fu n c t io n o f t h e R e y n o ld s n u m b e r R e ^ fo r

tu b e s o f v a r io u s ro u g h n ess.
U o -u (2 )
1 + Cln (5.49')
where Q is a new numerical coefficient, which, according to H unt’s

data (1964), is somewhat greater than unity. Finally, Ellison (1960),
on the basis of certain theoretical considerations, proposed the use of
an equation of the form
Uq- u (z)
In '+ ('+ x f (5.49")
w.
for open channels, instead of Eq. (5.49) or (5.49'). The right side of
Ellison’s equations for certain values of 6 , proves to be very close to
the right side o f Eq. (5.43) [with Bi = 0] and to the right side of Eq.
(5.49) or (5.49'). According to Ellison’s indirect estimate of b (which
will be discussed in greater detail in Chapt. 5), b ^ \ A , which does
not contradict the existing data on the mean velocity profile in open
channels.
Let us consider briefly, the application of Reynolds number

similarity to the moments of the velocity fluctuations in tube and
channel flows. According to this similarity, e.g., all nonzero second
moments, i.e., the four quantities u'^, v'^, w'"^, and u'w', must be
represented within almost all the flow, with the exception only of
the wall layer which is quite thin for sufficiently large Re, by a
formula of the form where?/ = z/H^ and v>,-, * = 1, 2 , 3, 4, are
universal functions (particular to each moment). There are extensive
data on the four universal functions i/f,-, for example, in Townsend
(1956) and Hinze (1959), in Comte-Bellot (1 9 6 5 )^ o a n tic (1967a)
and elsewhere. In particular, the value of the ratio u'^/ui on the tube
axis must be a constant for all sufficiently great Re according to
Reynolds number similarity; the correctness of this assertion is now
confirmed by numerous data [the value of the constant is 0.788 +
0.002 according to Coantic (1 9 6 7 a)]. Thus Eq. (5.44) shows that the
relative intensity of the turbulence on the tube axis (which can be
characterized quantitatively by the value o f ( u'^ ) ^ / U q) must
decrease continuously as Re increases; this fact is also known from
experimental data (cf. the cited paper of Coantic).
Reynolds number similarity must apply also to the higher-order
moments of the velocity fluctuations; however, at present, the
experimental data on such moments are very scarce. Nevertheless,
Comte-Bellot (1965) found that the dimensionless skewness and
flatness factors of all three velocity components can be represented
for almost all the flow region in a plane channel as universal
functions of ?; = (in complete agreement with the principle of
the Reynolds number similarity).
'F o r a fin ite range o f va lues o f R e y n o ld s n u m b e r , th e g en era l e q u a t io n ( 5 . 4 8 ) m a y

a p p r o x im a te ly b e r e p la c e d b y c o n s id e r a b ly sim pler p o w e r skin fr ic tio n la w s o f th e fo r m
X = c ( R e ^ ) - '" . (5 .5 0 )
S u ch p o w e r la w s are o b t a in e d , in pa rticula r, w ith th e a p p r o x im a tio n o f th e v e lo c ity p r o file

u ( z ) b y a p o w e r e q u a t io n o f th e fo r m
,5 ,5 ,1
(w h ic h has o f t e n b e e n d o n e in th e p a st). It is e a sy to see th a t Eq. ( 5 . 5 1 ) im p lie s th e

e q u a tio n s
1
Um 2 U,
t/o ~ (n+l)(/J + 2) ’ ) • '” = 7 r r r -
T he b e s t-k n o w n p o w e r fr ic tio n la w is B la siu s’ em p ir ic a l la w ( 1 9 1 3 )
^ = 0.3164(Rep) (5.52)
w h ic h w a s fo r m u la te d on th e basis o f da ta referring to R e y n o ld s n u m b e r s R e o not

e x c e e d in g 5 « 10^ (se e Fig. 3 3 ) . A c c o r d in g to Eq. ( 5 . 5 l \ B la siu s’ la w c o r r e sp o n d s to th e
“ o n e -s e v e n th la w ” fo r th e v e lo c ity p r o file , b y w h ic h u ( z ) ^ ; th is “ o n e -s e v e n th la w ”
(w h ic h agrees w e ll w ith e x p e r im e n t at Re « 10^) has b e e n u se d w id e ly in m a n y
in v e s tig a tio n s sin c e th e 1 9 2 0 ’s.^ H o w e v er , n o n e o f th e p o w e r la w s is un iversal (i.e ., su ita b le
fo r all R e y n o ld s n u m b e r s); fo r larger R e y n o ld s n u m b e r s, sm aller v a lu e s o f n. m u s t b e u sed .
O f c o u r se , in a d d itio n to p o w e r e q u a tio n s , w e m a y also c h o o s e o th er e q u a tio n s for
X (R e ), w h ic h a p p r o x im a te th e im p lic it rela tio n sh ip ( 5 . 4 8 ) o ver a large range o f va lu es o f R e.
0.221
In particu la r, N iku ra d se p r o p o s e d th e u se o f an a p p r o x im a te e q u a tio n X = 0 .0 3 2 + — ;
Re^
o th er in v e stig a to r s p r o p o s e d m a n y o th e r e q u a tio n s o f th is ty p e (se e, e .g ., K o lm o g o r o v s
critica l n o t e ( 1 9 5 2 ) w h ic h is d e v o te d t o several su ch e q u a tio n s ).
In c o n c lu s io n w e m u s t stress th a t E qs. ( 5 . 4 8 ) an d ( 5 . 4 8 ') are also n o t r ig o ro u s b e c a u se
th e y are b a se d o n th e a s su m p tio n th a t th e lo g a r ith m ic e q u a tio n fo r th e v e lo c ity p r o file is
ap p lic a b le right up to th e axis o f th e tu b e . M ore p r e cise e q u a t io n s ca n b e o b ta in e d w ith th e
aid o f r e fin e d la w s fo r th e v e lo c ity p r o file w h ic h ta k e in t o a c c o u n t th e v a riability o f th e
shear stress. O n e o f th e m e th o d s fo r c o r r e c tin g th e a p p r o x im a te lo g a r ith m ic fo r m o f th e
v e lo c ity d is tr ib u tio n w as d e v e lo p e d b y T o w n s e n d ( 1 9 6 1 ) and A . J. R e y n o ld s ( 1 9 6 5 ) b a sed
o n th e gen era l e q u a t io n ( 5 . 3 9 ) s u p p le m e n te d b y s o m e se e m in g ly natural b u t m o r e s p e c ific
h y p o th e s e s . A s a resu lt, th e y o b t a in e d an e q u a tio n o f th e fo r m
( 5 . 2 1 ')
dz y,2 y T dz f
Here w* - u sual fr ic tio n v e lo c it y , y, -- 1 /A is th e v o n Kdrmdn c o n s ta n t, B is
a n e w d im e n sio n le ss c o n s ta n t, a n d \ d T / d z \ for tu b e s and p la n e c h a n n els is e q u a l to
a c co r d in g to E qs. ( 5 . 1 7 ) and ( 5 .1 7 ') . E q u a tio n ( 5 . 2 1 ' ) is th e r efin ed fo r m o f th e u sual
e q u a tio n ( 5 . 2 1 ) [va lid in th e c o n sta n t-stre ss la y er o n ly ] w h ic h d iffe rs fr o m th e original
e q u a tio n b y a c o r r e c tin g s e c o n d term o n th e right side. It im p lie s a sp ecia l fo r m o f th e
v e lo c ity p r o file in th e r eg io n o u t s id e th e im m e d ia te p r o x im ity o f th e w a ll (i.e ., in th e region
w h e r e th e m o le c u la r v is c o s ity d o e s n o t play a c o n s id e r a b le p a r t), w h ic h tra n sfo r m s in t o th e
u sual lo g a r ith m ic p r o file in th e th in c o n s ta n t-str e s s la y e r sa tis fy in g th e c o n d it io n
z — « tq [s e e T o w n s e n d ( 1 9 6 1 ) ] . A c c o r d in g to th e e x p e r im e n ta l d a ta c o lle c t e d and
dz
p r o c e s s e d b y A. J. R e y n o ld s , B ^ 1.1 fo r circular tu b e s and B ^ \ . l fo r tw o -d im e n s io n a l
ch a n n els; th e r e fo r e th e valu e B - 1.5 c o u ld b e u se d s im u lta n e o u s ly fo r b o t h ty p e s o f flo w s
as a r ea so n a b le first a p p r o x im a tio n . (R e y n o ld s o b t a in e d a lm o s t th e sa m e v a lu e B ^ 1 .4 7
a lso fo r a b o u n d a r y -la y e r f lo w w ith z er o pressure g r a d ie n t.)
A r ela te d (b u t m o r e c o m p lic a t e d ) th e o r y w a s d e v e lo p e d b y K le in ste in ( 1 9 6 7 ) w h o
s h o w e d th a t th e e f f e c t o f shear-stress va ria tio n c o u ld b e in c lu d e d q u ite e a sily in th e
s e m iem p ir ic a l a rg u m e n ts le a d in g to S p a ld in g ’s e q u a t io n for th e u n iversal la w o f th e w a ll
^M ore r e c e n tly . B u r to n ( 1 9 ^ 5 ) p r o p o s e d th e fo llo w in g sim p le g e n e r a liz a tio n o f th e

“ o n e -s e v e n th la w ” : = 17+ + ( m+ / 8.74)'^. T his g e n e ra lize d la w has s a tis fa c to r y accu ra c y in
th e w h o le la w -o f-th e -w a ll la y e r an d ca n b e u s e d fo r a r e fin e m e n t o f B la siu s’ fr ic tio n
e q u a tio n ( 5 . 5 2 ) .
[wQ have alread y m e n t io n e d th a t this e q u a tio n w a s a lso d eriv ed b y K le in ste in

in d e p e n d e n tiy fro m S p a ld in g ) . H o w e v er , th e d e v ia tio n s o f th e lo g a r ith m ic v e lo c ity p r o file
fro m all th e refin e d v e lo c ity d is tr ib u tio n s o b ta in e d b y in c lu d in g th e e f f e c t o f shear-stress
va ria tio n arc cu r re n tly s till w ith in th e range o f th e sca tte r o f th e e x is tin g data.
5.6 Turbulent Boundary Layer on a Flat Plate

Now we consider a turbulent boundary layer on a long flat plate
with constant velocity U o f the incident flow (i.e., with zero pressure
gradient). The mean flow in this boundary layer will be steady and
almost plane-parallel; in many ways, it will be close to the flow in a
tube, the radius o f which is equal to the thickness 6 o f the boundary
layer, and the velocity of the axis equal to the velocity U outside this
layer. However, there are at least two reasons for a breakdown in the
analogy between the flow in a tube and in a boundary layer. First,
the physical conditions on the outer edge of the boundary layer are
quite different from the conditions at the center of a tube; outside
the boundary layer (i.e., for z > 6) there is usually no turbulence
while the flow in a tube will be turbulent everywhere. Second, the
characteristics of the boundary-layer flow depend not only on the
coordinate normal to the plate but also (even if comparatively
weakly) on the longitudinal coordinate a:, reckoned along the plate.
This leads to the fact that, theoretically, the flow in a boundary layer
is considerably more complex than flows in channels or tubes.
The difference between flows in a boundary layer and in tubes and
channels does not affect the thin layer of fluid in the immediate
vicinity of the plate in which the universal “law of the wall” (5.13)
apphes. As clearly seen from Fig. 25 where all the points lie on the
same curve, this law is identical for tubes, channels and boundary
layers. The value of u* [which is found in Eq. (5.13)], in a boundary
layer may, o f course, vary with variation of x; however, this variation
proves to be fairly slow (see below, at the end of this subsection), so
that in any section x = const, the flow has time to become adjusted
to local conditions, and depends only on the value of u.* for given x.
The relative thickness of the layer to which the law (5.13) applies
UX
also varies with increase of x: for small Re*= — not exceeding 10®,
Eq. (5.13) may be used approximately for all z almost up to the very
UX
edge of the turbulent boundiiry layer; for Re.x= — of the order of
1 0 ’ - 1 0 ®, the total thickness of the viscous sublayer and logarithmic
layer will no longer exceed 10—20% of the thickness 6 . This last fact,
of course, distinguishes boundary-layer flow from the flow in tubes
and channels, where for all Re > Rccr, the logarithmic equations
may be used without great error right up to the axis of the tube or
the middle of the channel.
The mean velocity profile in the outer part of the boundary is
naturally characterized by the dependence on z of the velocity
defect U — u(z). If we start from the assumption (which is
confirmed well experimentally) that for an unperturbed boundary
layer the velocity profile in every section x = const depends only on
the local conditions which refer to this section, the difference
U — u{z) must be determined by U, 6, the parameters of the fluid v
and p and (in the case of a rough plate) the roughness parameter for
given X. According to the general principle of Reynolds number
similarity, we must expect that in the outer part of the flow, which is
sufficiently distant from the wall, the viscosity v and the roughness
parameter will exert an effect only by means of the value of the
shear stress on the wall io, or the friction velocity a, = l / ^ " ’
local friction coefficient . Thus we obtain the following

general form of the velocity defect law for a boundary layer:
U - a (z) ^
- / . ( I - 17) (5.53)
[see Rotta (1962a, b ) ] . To verify this law, it is convenient to plot a

graph o f ^ as a function of ~ and to attempt to explain the
character o f the scatter of the experimental points by dependence on
the value of the ratio However, it is found that this scatter is
masked completely by the ordinary scatter of experimental points.
Therefore, within the limits of measurement errors, all the values of
^ lie with sufficient accuracy on a single curve [see, for
example, Fig. 35, borrowed from Clauser’s survey article (1956)].
Thus the dependence of the ratio — on (which varies
slowly with variation of Re^) even if it occurs, is quite weak. 1 0
^ ^ C ro cco ( 1 9 6 5 ) a tte m p te d to u n m a sk su ch a d e p e n d e n c e in th e e x p e r im e n ta l d ata and

to e x p la in it b y a sp ecia l s e m iem p ir ic a l th e o r y b a sed o n a r efin e d fo r m o f C o le s ’ “ w a k e la w ”
(this la w w ill b e d isc u sse d la te r ). H o w ev er, his a tt e m p t w a s n o t c o m p le te ly su c c e s sfu l and
w ill n o t b e d isc u sse d in th is b o o k (c f., h o w e v e r , th e f o o t n o t e o n C o le s ’ “ w a k e la w ” later in
this s u b s e c tio n ).
Consequently, in practice, the velocity defect law for a boundary

layer on a flat plate may be written as
U -u (z)
(5.54)
which is completely analogous to the law used above for flows in

tubes and channels.
20
u.
X Data o f Freeman
o — ' — K le b a n o ff and Diehl
• — "— Schultz-Grunow
— Hama
__ Johns Hopkins University
(for a rough wall)
Hama (for a rough wall)
— Moore (for a very rough
wall)
0.2 0.^ 0.6 0.8 W
FIG . 3 5 . V e r ific a tio n o f th e “ v e lo c ity d e f e c t la w ” fo r a tu r b u le n t b o u n d a r y la yer,

a c co r d in g t o th e d a ta o f d iffe r e n t a u th o r s.
In comparison with the velocity defect law for tubes and channels
(5.41), the law (5.54) has only one deficiency: whereas in Eq. (5.41)
we have the precisely defined length//i, in Eq. (5.54) we have the
thickness of the boundary layer 6 which has no precise definition
and is determined only very approximately from the experimental
data. At the same time, the ratio of 6 to more precisely defined
thicknesses of the type, e.g., of the displacement thickness 6 * in a
turbulent boundary layer (as distinct from the laminar case) is
variable (dependent on u ,lu ). In particular, if the law (5.54) holds,
then from the very definition (1.53) of the displacement thickness 6 *
it follows that
8* = C - ^ 8 , C= f / , ( 7j)</7) = const. (5.55)

6
(We ignore here the thickness of the layer o f nonnegligible viscous
stresses to which the law (5.54) is inappHcable.) Taking this
relationship into account, R otta (1951a) proposed to use in Eq.
(5.54) instead of the argument rj = z/i, the dimensionless argument
= which does not contain 6 ; by Eq. (1.53), this argu
ment will also be convenient for the normalization condition
uu
/
0
cpi(7)i) = 0 for = C-, then, clearly, l = and cp, (t)j) = / i

Consequently, in reducing the observations it is completely per
missible to replace 6 by 8* , and if ^ is a single-valued
function of r)i = this means that the law (5.54) is also satisfied.
O f course, Eq. (5.54) cannot be satisfied right up to 0. This is

because it does not contain the viscosity, and does not take the
boundary conditions «(0) = 0 into consideration. In the immediate
vicinity of the wall the universal law (5.13) applies, and, as we have
seen, in the overlap region where both these laws apply simulta
neously, the functions / and fi are bound to be logarithmic. Thus it is
not surprising that after transfer of the data of Fig. 35 to a new
graph, where a logarithmic scale is used for the abscissa, the function
fi(Ti)from Ti= 0.01 to approximately ti= 0.15 will be a straight line
—A In Ti + Si(see Fig. 36, where only a part of the points given in Fig.
35 are taken into account).* * We note that here the coefficient A —
1 /x has exactly the same value as in Fig. 3 1 for plane channels as is
required by the universality of the logarithmic law (5.25); however.
Si differs for channels and for boundary layers. Moreover, comparing
Figs. 36 and 31 makes it clear that for t] > 0.15, /i (tj) for a boundary
layer diverges far more sharply from the straight line /I In than
the corresponding function for plane channels. According to Hama
(1954), the value of fi(Ti)for boundary layers in the region ti > 0 .1 5
" S i m i l a r graphs o f th e fu n c t io n for b o u n d a r y la y e r s o n s m o o t h a n d ro u g h fla t

p la tes ca n b e f o u n d , e .g ., in th e p a p ers o f R o tta ( 1 9 6 2 a , b ) and F u r u y a a n d F u jita ( 1 9 6 7 ) .
is well described by the empirical formula /i( t) ) = 9.6 (1 — ti)^; for

flows in plane channels, however, /i (t)) for ti > 0.15 may be described
reasonably well by the equation / j (tj) — A' I n r j - j - w h e r e A' and B'
differ only slightly from A andB,.
FIG. 3 6 . Logarithmic form o f the v elocity d efect law for a

boundary layer.
The considerable difference between the velocity profile in a

turbulent boundary layer close to its outer edge and the velocity
profile in tubes and channels close to the axis can easily be explained
by the difference between the nature of the turbulent flow itself in
the outer part of the boundary layer and the central part of the tube
or channel. In this respect, we observe that, unlike a laminar
boundary layer which has no sharply defined outer edge, a turbulent
boundary layer normally has a definite edge, outside of which there
is no turbulence (i.e., the flow is actually irrotational). This edge (or
interface) has an irregular outline and varies in a disordered manner
with time. This is clearly visible in photographs of a turbulent
boundary layer obtained by the shadow-graph method, which
renders the perturbed region o f flow visible (see, e.g.. Fig. 37). The
general diagram of the turbulent boundary layer corresponding to
this photograph is given in Fig. 38. According to the experimental
data of Klebanoff (1955), Corrsin and Kistler (1955), Fiedler and
Head (1966) and others, the upper edge of a turbulent boundary
layer at any given point will fluctuate between approximately
0.36 —0.46 and 1.26 (where 6 is the distance from the wall at which
F IG . 3 7 . S h a d o w graph o f th e tu r b u le n t b o u n d a r y lay er o n a c y lin d e r [fr o m R o t t a ’s

pap ers ( 1 9 6 2 a , b ) ) . D ir e c tio n o f flo w is fro m le f t to right.
the mean velocity equals 0.99U) almost according to the normal dis
tribution, with mean value 0.785 —0.806. The fraction of time y ( 2 )
during which turbulence will be observed at a distance z from the wall
decreases with increase of z, and with z = 6 equals only about 0.06.
Consequently, it is clear why the logarithmic profile characteristic
for developed turbulence may be used with reasonable accuracy
almost to the center o f a tube or channel. However, in a boundary
FIG . 3 8 . G en era l diagram o f a tu r b u le n t b o u n d a r y la y e r o n a fla t p la te .

layer with large Re, it is satisfied comparatively accurately only for a

thin layer of fluid close to the wall. Outside this region of flow,
periods of the presence and the absence of turbulence follow each
other alternately in rapid succession. Thus the mean velocity
obtained by averaging over a sufficient time interval is actually a
mean between values of the velocity corresponding to both laminar
and turbulent flow. (By assuming that the logarithmic wall law in
fact appHes in the outer part of the boundary layer, but only when
the turbulence is present, and that the irrotational velocity is that of
the free stream, B. G. J. Thompson (1963; 1965) was able to obtain a
two-parameter family of standard velocity profiles which gives a
good agreement with measured boundary layer velocity profiles [cf.
also Dvorak and Head (1967)].) Since the exchange of momentum
between neighboring layers of fluid in laminar flow is considerably
less than in turbulent flow (hence, the velocity gradient is sharper), it
is clear that in the outer part of a turbulent boundary layer the mean
velocity, with increase of distance from the wall, must increase more
rapidly than by the logarithmic law. This is confirmed by the data of
Fig. 36. Also, for a boundary layer occurring as a result of
high-intensity turbulent flow along a flat plate, the difference
between the conditions on the outer edge of this layer and at
the center of a tube must be less considerable. In fact, the ex
periments of Wieghardt (1944) and others, who passed the flow
first through a grid producing a high turbulence level, show that the
form o f / i (ti) in Eq. (5.54) depends on the turbulence level of the
incident flow and as this level increases, it approximates the form
characteristic of flows in tubes and channels.
An interesting attem pt to combine the universal law of the wall
(5.13) with the velocity defect law (5.54) was made by Coles (1956).
On the basis of Eq. (5.13) the general form of the mean velocity in
every section x - const of the boundary layer may be written as
U(z) = u \ f [ ^ ' ) + h { z ) \ (5.56)
where f{z+) is the function represented in Fig. 25 (and transformed

into A in z+-f-B for 2 + > 30) and ft (2 ) is a correction to the universal
law o f the wall (5.13), which becomes zero in immediate proximity
to the plate (i.e., for 2 < 0.16).However, in this case — = ^4 In ^
+ A(8) — h{z) for 2 > 3 0 -^ , and, hence, for the universal velocity
318 STATISTICAL F LUIO MECHAN ICS
defect law to be satisfied, A(2 )must depend only on y . On this basis,

Coles proposed that
i ^ = / ( i ^ ) + S .(£ ), (5.57)
where ~ 0.4 is von Karman’s constant, IT is a new constant and

w (ti) is a function of a single variable normalized by the condition
lei (1) = 2 . However, a certain indeterminacy still remains connected
with the fact that the thickness 6 is not uniquely defined; using this
fact, Coles added to ay(Ti) the additional normalization condition
which, as is easily seen, is equivalent to the condition 8* — = ^ 8.

Consequently, by this condition he gave a definite form to the
coefficient C in Eq. (5.55), i.e., he derived an exact connection
between the scale 6 and the displacement thickness 6*, which may
easily be found from experiment, the value o f-^ and the parameters
X and n. On the basis of considerable experimental data on velocity
profiles in boundary layers on a plate, for both constant pressure and
in the presence of a pressure gradient, Coles found that for a broad
class of two-dimensional turbulent boundary layers, the function
ay (11) is the same. Thus, according to his data, the external conditions
of flow, including the pressure distribution in a free flow, are
reflected only in the value of the factor II. In the case of a
compound pressure distribution, this factor can be assumed to be
dependent on the x-coordinate; in the simplest case of flow past a
plate with zero pressure gradient (i.e., at constant velocity) II = const
» 0.55 for all not too small Reynolds numbers.*^ The universal
function to(Ti), obtained by Coles by evaluating the experimental
'^ A c c o r d in g t o an u n p u b lish e d rep o rt b y C o le s in 1 9 6 2 (r e v ie w e d b y C r o cc o ( 1 9 6 5 ) ] ,

n = c o n s t at R c j , , = V S * * / v ^ 6 0 0 0 , w h ile fo r v a lu e s c f R e 5 , , l e s s th an a b o u t 6 0 0 0 , fl
d e c re a ses rap id ly a n d b e c o m e s z er o at R e j , , = 5 0 0 . A c c o r d in g to C r o c c o , II is slig h tly
d e p e n d e n t o n R e j , , ( i . e . , o n u * / U ) a t all v a lu es o f R c j , , .
data is shown graphically in Fig. 39. It possesses an almost exactly

skew-symmetric, S'-shaped form, and is fairiy well approximated by
the section of the sinusoidal curve w — I + sin = i _ cos
for 0 <T i < 1. (This form of w(tj) was used extensively, e.g., by
Spalding (1964) and Escudier and Nicoll (1966) for many calcula
tions related to a broad class o f flows with one plane rigid
boundary.)
FIG. 39. U niversa l w a k e fu n c tio n w (t]) a c co r d in g t o C o le s ( 1 9 5 6 ) .
In Eq. (5.57), we may proceed to the limit formally as u* ^ 0

(i.e., as To 0); with the aid of Eq. (5.55) and the condition a; ( 1) =
2, we obtain the simple result
Thus the function ay (ri) gives the form of the mean velocity profile at
a point with to= 0, i.e., at a point of separation of the boundary
layer, at which ^ = 0 (see the schematic representation in Fig. 8).
On this basis. Coles called the function tw('n) the wake function, and
the law by which the deviation of the mean velocity profile in the
boundary layer from the universal law of the wall (5.13) is strictly
proportional to the universal wake function, the wake law. The
general form of the mean velocity profile u{z) satisfying this wake
law is shown schematically in Fig. 40. Here the dashed line represents
the distribution of the mean velocity in the upper half of a
hypothetical turbulent wake. This is produced basically by the
interaction of the turbulent core of this wake with the outer laminar
flow, and by the fluctuations of the interface between the turbulent
and nonturbulent parts of the flow in accordance with Fig. 38; the
difference o f the mean velocities at the center and on the boundary
FIG . 4 0 . S c h e m a tic r ep r e s e n ta tio n o f th e v e lo c ity p r o file sa tisfy in g Coles*

w a k e la w .
211
of this wake is equal to — «,. If we place a rigid plate at the center
of this wake, then the velocity distribution in the wake is changed
considerably. This is due to the fact that close to the surface, the
flow begins to be strongly constrained by viscous effects caused by
the no-slip condition u { z ) = 0. This additional limitation, in pure
form, generates the “ universal wall profile” shown in Fig. 40 by the
dotted line. As a result of superimposing this “ wall profile” on the
wake profile, the real velocity profile in a turbulent boundary layer,
represented by the solid line in Fig. 40, is obtained.
Another approach to the problem of the velocity profile in the
boundary layer was developed by B. G. J. Thompson (1963; 1965).
We have mentioned it earlier in this subsection; it is clearly related to
Coles’ approach, although it is based on quite different assumptions.
Similar to flows in a tube, from the existence of a section of the
flow in which both the wall law (5.13) and the velocity defect law
(5.54) are satisfied (with logarithmic functions f and fi), it follows
that the local friction coefficient
must satisfy an equation of the form
= ^ (5.58)
[cf. the deduction of Eq. (5.45) in Sect. 5.5]. Just as with the law
(5.54), the presence in Eq. (5.58) o f an inexactly defined boundary-
layer thickness 6 makes this equation unsuitable for practical
applications; thus it is desirable to replace 6 by another more easily
defined scale of length. Most frequently, one uses Eq. (5.55), which
allows Eq. (5.58) to be rewritten in the form
A comparison of this relationship with the data of direct measure

ments of the values of the local friction coefficient at different
points o f the boundary layer on a smooth flat plate, obtained by
Schultz-Grunow (1940) and Smith and Walker (1959), is given in

Fig. 41; in the figure, the experimental values follow the right
V c,
side o f Eq. (5.58) fairly closely with x = 0.40 and 2.6.
FIG . 4 1 . D e p e n d e n c e o f th e lo c a l skin fr ic tio n c y fo r th e b o u n d a r y lay er o n a fla t p la te

o n th e R e y n o ld s n u m b e r R e g * a c co r d in g to th e d a ta o f S c h u ltz -G r u n o w and o f S m ith
and W alker.
Replacing 6 simply by the length x (reckoned from the origin of

the plate) is somewhat more complicated. For this purpose we must
express the thickness 6 in terms of x, which may be carried out using
the following approximate considerations. Since the streamlines of
rotational turbulent flow in the boundary layer do not pass outside
the layer, the mean outer edge of the boundary layer must coincide
with a streamline of the mean motion at a distance 6 from the plate.
Therefore, it follows that the tangent of the angle of inclination of
this edge to the Ox axis must be equal to the ratio of the mean
vertical velocity (in the direction of the axis Oz) to the mean
horizontal velocity at points of the edge. However, the mean
horizontal velocity on the edge of the boundary layer is equal to [/;
the mean vertical velocity is independent of U and is determined
only by the relative fluctuating motion connected with the transfer
of momentum in the Oz direction. Consequently, by dimensionality

considerations, this vertical velocity must be proportional to the
friction velocity «*, formulated with the value of -to for given x. Thus
follows the relationship
§ = (5.60)
where bi is a numerical coefficient (of the order o f unity). We now

assume that the boundary layer is turbulent, originating, practically
speaking, from the leading edge o f the plate itself. Thus Eq. (5.60) is
satisfied for all x beginning with x = 0 (where 6 = 0). Generally
speaking, the quantity u ^ / u is a function of x, but since it varies very
slowly, by integrating Eq. (5.60), we may take it outside the integral
sign without considerable error (changing the coefficient bi some
what at the same time if need be). Therefore, we obtain the
approximate equation
Z= b i ^ x . (5.61)
Strictly speaking, Eq. (5.61) follows from Eq. (5.60) only if
U ~~ V 2
may be represented as a simple power function of x. However, it will

be seen from the following discussion that such a representation
possesses fairly high accuracy; see, for example, Eq. (5.66), a
comparison of which with experimental data is given in Fig. 42.
Substituting Eq. (5.61) into Eq. (5.58), we obtain
This result was given first by von Karman (1930; 1934). He obtained
it by means of somewhat different arguments and it is in good
agreement with K em pfs (1929) older experimental data obtained
with the aid of direct dynamometer measurements of the local skin
friction coefficients. Von Karman’s result is also in good agreement
with the later experimental data of Schultz-Grunow (1940) and
Dhawan (1952) [see Fig. 42, where the coefficients = 1.8, i.e.,
FIG. 42. D e p e n d e n c e o f th e sk in fr ic tio n c o e f f ic ie n t c f f o t t h e

b o u n d a r y la y e r o n a fla t p la te o n th e R e y n o ld s n u m b e r Re^c
a c co r d in g t o th e da ta o f K e m p f, S c h u ltz -G r u n o w , and D h a w a n .
C urve 1: v o n K drm dn’s sk in fr ic tio n la w ( 5 . 6 2 ) w i t h x = 0 . 3 9 a n d
f l s = 1 .7 ; C urve 2: S c h u ltz -G r u n o w la w ( 5 . 6 4 ) ; C urve 3: F a lk n e r
la w ( 5 . 6 6 ) .
x = 0.39 and B ^ = 1.7, are chosen in order to obtain the best

agreement with K em pfs d a ta ]. We note further, that from the
comparison of the numerical values o f the coefficients B 4 and B5 in
Eqs. (5.59) and (5.62), leading to the best comparison with the
experimental data, it is also possible to estimate the value of b in Eq.
(5.61). In particular, if we take the values of these coefficients used
in Figs. 41 and 42 (neglecting the fairly small divergence in the
chosen values of x), for the thickness 6 , defined by Eq. (5.55) with
1.55
C = 0.4 3.9 [i.e., with the same value of C as used by Coles
(1956)], we obtain 0.3.
For rough plates, the above equations must be changed in a
manner similar to that for rough tubes; for example, Eq. (5.58), for a
completely rough tube must be written in the form (5.45') [but with
6 instead of / / J . However, we shall not discuss this at length here

[see, for example, Schlichting (1960), Chapt. 21, Sect. 4 ].
A c c o r d in g t o Eq. ( 5 . 6 2 ) , C/ is a s lo w ly d e c re a sin g f u n c tio n o f x. H o w e v er , th e e x p lic it

d e fin itio n o f C/ as a fu n c tio n o f Re* w ith th e aid o f E q. ( 5 . 6 2 ) is fairly c o m p lic a t e d a n d , in
p r a c tic e , sim p le r in te r p o la tio n o r e m p ir ic a l e q u a t io n s are n o r m a lly u s e d . T h u s, fo r e x a m p le ,
S c h lic h tin g ( 1 9 6 0 ) [p r o c e e d in g , in fa c t, n o t fr o m v o n K drm dn’s e q u a t io n ( 5 . 6 2 ) b u t fr o m
th e m o r e gen era l r esu lts o f P randtl, d e s c r ib e d , fo r e x a m p le , in S c h lic h tin g ’s b o o k ( I 9 6 0 ) ]
p r o p o s e d for C/ th e c o m p u ta tio n a l fo r m u la
Cf = ( 2 In R e ^ — 0 . 6 5 ) - * - “ , (5 .6 3 )
w h ic h g ives r esu lts c lo s e to th o se o b t a in e d fr o m E q. ( 5 . 6 2 ) . L ater, S c h u ltz -G r u n o w ( 1 9 4 0 )

u s e d th e e q u a tio n
Cf = 0 .3 7 0 ( In R e ^ ) - 2 “ ^, (5 .6 4 )
th e resu lts o f w h ic h are c o m p a r e d w ith th e d a ta in Fig. 4 2 . F in a lly , i f w e a ssu m e th a t th e

m e a n v e lo c ity p r o file in th e b o u n d a r y la yer in all s e c tio n s is g iv en b y th e “ o n e -s e v e n th la w ”
(se e th e c lo sin g rem ark s o f S e c t. 5 .5 ) , th e n th e very s im p le fo r m u la
C/ = c ( R e ^ > ® (5 .6 5 )
m a y b e o b ta in e d . T his fo r m u la , w ith c = 0 . 0 5 7 6 , a lso g iv es a fairly g o o d d e s c r ip tio n o f th e

m e a s u r e m e n ts fo r th e range o f R e y n o ld s n u m b e r s 5 x 10^ < Re;^ < 1 0 ^ . W ith fu rth er
increa se o f th e R e y n o ld s n u m b e r Re^^, E q. ( 5 . 6 5 ) b e g in s to give a r e d u c e d v a lu e o f c / , so
th a t t o o b ta in th e b e s t p o s sib le a g r e e m e n t it is n e cessa ry to rep la ce th e p o w e r in d e x 1 /5 in
th e fo r m u la b y a sm a ller v a lu e . In pa rticula r, fo r v a lu e s o f Re^^ up t o 1 0 ^ , g o o d a g r ee m e n t
w ith th e da ta m a y b e o b t a in e d b y u sin g F a lk n e r ’s fo r m u la ( 1 9 4 3 )
C /= 0 . 0 2 6 2 (R e^> . (5 .6 6 )
In a d d itio n t o th e lo c a l fr ic tio n c o e f f ic ie n t Cf, w e m a y also c o n s id e r th e to ta l fr ic tio n

c o e f f ic ie n t C / o f a p la te o f le n g th /, given b y
I i
Cf = — -- -------- = \ - C f{x )d x , F = 2 f to ( x ) d x (5 .6 7 )
'i h'
[c f. E qs. ( 1 . 5 1 ) - ( 1 . 5 2 ) ] . T his to t a l fr ic tio n c o e f f ic ie n t h a s b e e n m e a su r ed m u c h m o r e

fr e q u e n tly th a n th e lo c a l c o e f f ic ie n t C/, a n d th e r e is a grea t d e a l o f d a ta referring t o it, th e
ea rliest o f w h ic h g o e s b a c k to 1 7 9 3 [s e e S ch u b a u e r a n d T c h e n ( 1 9 5 9 ) ; G o ld ste in ( 1 9 3 8 ) an d
S c h lic h tin g ( I 9 6 0 ) ) . U sing E q. ( 5 . 6 2 ) as his sta r tin g -p o in t, S c h o e n h er r ( 1 9 3 2 ) o b ta in e d fo r
Cf th e rela tio n sh ip
^ 4 .1 3 In ( R e , . C / ) . R e, = i ^ . (5 .6 8 )
ycf
w h ic h gives good a g r ee m e n t w ith e x p e r im e n ta l d a ta . S c h lich tin g , u sin g th e m o re

c o m p lic a t e d c a lc u la tio n s o f Prandtl in s te a d o f E q. ( 5 .6 2 ) , p r o p o s e d th e sim p le in te r p o la tio n
fo rm u la
Cf = 0.455 (In (5.69)
w h ic h g ives resu lts th a t d iffe r lit tle fr o m th o s e o f E q. ( 5 . 6 8 ) . S c h u ltz -G r u n o w ( 1 9 4 0 ) u se d a

sim ilar fo r m u la
Cf = 0.427 ( In Re, — 0.407)" 2 ®'’ . (5.70)
I f w e start fr o m th e “ o n e -s e v e n th la w ,” th e n as w e ha v e a k e a d y s e e n w e o b ta in a
d e p e n d e n c e fo r C / o f th e fo r m C / ' ^ ( R e / ) b e tt e r a g r e e m e n t o ver a w id e range o f

R e y n o ld s n u m b e r s, h o w e v e r , m a y b e o b t a in e d if, fo llo w in g E q . ( 5 . 6 6 ) w e p u t
C / = 0.0306 Re/ ^ (5.71)
C o m p a r iso n o f th is resu lt w ith E q. ( 1 . 5 2 ) fo r th e lam in a r fr ic tio n c o e f f ic ie n t o f a p la te

s h o w s th a t in th e tu r b u le n t c a se , th e fr ic tio n is c o n s id e r a b ly greater th a n in th e lam in a r case
(fo r e x a m p le , w ith R e / = 5 X 10^ it is a lm o s t 2 .5 tim e s grea ter) a n d d e c re a ses far m o r e
s lo w ly as th e R e y n o ld s n u m b e r in crea ses.
All th e p r e c e d in g c o n s id e r a tio n s refer to th e ca se in w h ic h th e b o u n d a r y la y er m a y b e
a ssu m ed fu lly tu r b u le n t fr o m p r a c tic a lly th e le a d in g e d g e o f th e p la te . H o w e v er , i f tr a n sitio n
to a tu r b u le n t regim e o c cu rs o n ly a t th e p o in t x = Xq, w h ic h is s u ffic ie n t ly far fr o m th e
lea d in g e d g e , th e n w e m u st in t r o d u c e in t o th e skin fr ic tio n la w a c o r r e c tio n fo r th e lam inar
p o r tio n o f th e b o u n d a r y la y er. I f w e u se P ra n d tl’s a s su m p tio n ( 1 9 2 7 ) th a t a fter tr a n sitio n to
th e tu r b u le n t reg im e , a b o u n d a r y la y er b e h a v e s a p p r o x im a te ly as i f it h a d b e e n tu rb u le n t
fr o m th e le a d in g e d g e o f th e p la te , th is c o r r e c tio n w ill b e r e d u c e d to th e fo llo w in g : fr o m the
value o f th e fr ic tio n fo r c e F c a lc u la te d a b o v e w e m u st su b tra ct th e d iffe r e n c e b e tw e e n th e
fr ic tio n fo r c es fo r tu rb u le n t and lam inar reg im es fo r th e p o r tio n o f th e p la te fr o m th e
leadin g ed g e to th e p o in t o f tr a n sitio n ;: = ;: o .I f w e ta k e th is c o r r e c tio n in t o a c c o u n t, th e
fr ic tio n c o e f f ic ie n t w ill e q u a l
O/ — [^1, (-ô) ^2 (-ô)] —

_
^ 9U^-‘21
= --------------- Rii------------------------ R ^ *
w here C / is g iv e n , sa y , b y Eq. ( 5 . 6 8 ) [o r b y s o m e o f th e o th e r e q u a tio n s g iv en a b o v e ] and

Cl (jfo) a n d C^^Xq) are th e fr ic tio n c o e f f ic ie n t s fo r a p la te o f le n g th / w a sh ed b y tu rb u le n t
and la m inar flo w s , r e s p e c tiv e ly . T his c o r r e c tio n p e r m its us to d e scr ib e th e tr a n sitio n a l region
b e tw e e n th e fr ic tio n la w ( 1 . 5 2 ) fo r a lam in a r flo w an d th e fr ic tio n la w s fo r a pu rely
tu rb u le n t f lo w w h ic h w e h ave b e e n d isc u ssin g . T he c o e f f ic ie n t A(xo ) in Eq. ( 5 . 7 2 ) w ill, o f
c o u r se , d e p e n d o n th e c ritica l R e y n o ld s n u m b e r Re^^ = Rejc cr a t w h ic h th e tr a n sitio n fr o m
lam inar to tu r b u le n t f lo w o c cu rs; a c c o r d in g to P ra n d tl’s d a ta , fo r Rcxq ~ 5 X 1 0 ^ , for
e x a m p le , A (j:o) -= 1 7 0 0 [t h e valu es o f A (jcq) fo r s o m e o th e r v a lu es o f Re^^j m a y b e fo u n d
in S c h lic h tin g ’s b o o k ( I 9 6 0 ) ] .
T h e c ite d results referring to th e law o f va riatio n o f Cf and C, w ith va ria tio n o f th e

R e y n o ld s n u m b ers Re^^ and R e / m a y also be u sed in in v e stig a tin g th e va ria tio n w ith increase
of X o f th e m ean v e lo c ity p r o file , w h ic h d e p e n d s o n = ^/ | / and 5 a c co r d in g to Eq.
( 5 .5 7 ) . T h e d e p e n d e n c e o f c / o n x has b e e n d isc u sse d a b o v e , and th e v a ria tio n o f 6 w ith
increase o f x m a y b e d e te r m in e d fr o m E q. ( 5 . 6 1 ) w h ic h o n c e again c o n ta in s ^
Sin ce Cf d e crea ses very s lo w ly w ith in crea se o f x , th e th ic k n e ss 6 increa ses a lm o s t as th e first
p o w e r o f x ( i f w e use Eq. ( 5 . 6 6 ) , 6 is p r o p o r tio n a l to It is im p o r ta n t th a t th is

increase b e c o n s id e r a b ly m o r e rapid th an th e in crease in th ick n e ss o f a la m inar b o u n d a r y
la yer, w h ic h , a c co r d in g to E q. ( 1 . 3 3 ) is p r o p o r tio n a l to (c f. Fig. 4 , S e c t. 1 .4 , w h ic h

c o n ta in s results o f th e d ir e c t m e a s u r e m e n t o f th e th ic k n e ss 6 in b o t h the lam inar and the
tu rb u le n t parts o f th e b o u n d a r y la y e r ).
5,7 Profile of Concentration of a Passive Admixture Close

to a Wall; Mass- and Heat-Transfer in a Turbulent
Boundary Layer
We have discussed the questions of the mean velocity distribution
and o f the friction in turbulent flows close to a wall. It is found that
similar considerations may also be applied to the investigation of
turbulent mass- and heat-transfer. Below, we shall give some basic
facts about this problem; a more detailed description may be found,
for example, in the engineering books of Grober and Erk (1955),
MacAdams (1954), and Eckert and Drake (1959), in Chapt. 14 of the
monograph edited by Howarth (1953), Chapts. 3 - 4 of Levich
(1962), in the survey articles of Deissler (1959), Kestin and
Richardson (1963), and Spalding and Jayatillaka (1965).
As in Sects. 5 .2 -5 .3 , we consider a plane-parallel flow of fluid in the
half-space z > 0 , bounded by a rigid smooth wall, directed along the
Ox axis in the absence of a longitudinal pressure gradient. Let us
assume that on the bounding surface z = 0, a constant value Oo of the
concentration of the passive admixture is maintained. Then a
constant flux j of the admixture will occur in the fluid, directed
outward from the wall, i.e., in the positive Oz direction, and Eq.
(5.7') will have the form
J (2) = - P X ^ + P^' v d ' = J o = const. (5 .7 3 )
The profile of the mean concentration will depend on the

statistical characteristics of the velocity field (defined by the

parameters v and «*) and also on the molecular diffusivity %, the
density p and the admixture flux /o. Introducing the special
dimension 0 for and denoting the dimension of mass by M, we
have
[Jo ]-M S L -^T -\ [pI = m - ^ K \ = L T -\ [v] = IxI = Z^7’- ‘.
Consequently, by dimensional considerations, we have
» (2 )-» (0 ) = ^ c p ( ^ * . = (5.74)
where
(5.75)
is a constant of dimension 0 , giving a natural scale of the

concentration [von Karman’s dimensionless constant %= ^ ^ 0.4 is
included here in the definition of 6*, since this will sometimes be
found to be convenient in the subsequent argument] ,and(p(2+, Pr)is
a new universal function o f two variables, satisfying the condition
q)(0, Pr) = 0. If is the temperature T, then Eqs. (5.73)—(5.75)
must, of course, be written as
df
q{z) = - CpPx aF + = ^0 = const, (5.73')
f ( 2 ) - f (0) = - ^ < p ( ^ . = P')’ (5.74')
xa, Cpp (5.75')
where the function (p(z+, Pr) is the same as in Eq. (5.74). We note
that for the case of temperature, the assumption of passivity is more
doubtful than for a material admixture. This is due both to the
presence of buoyancy forces in a nonuniformly heated fluid (this will

be discussed in greater detail in Chapt. 4), and the temperature-
dependence of the fluid density, viscosity, and thermal diffusivity
[for the problem o f taking into account these dependences which are
different for gases and for liquids, see, for example, the articles of
Deissler (1959) and van Driest (1959)]; in addition, the heating due
to energy dissipation is not always negligible in laboratory flows.
Nevertheless, below, as a rule, will be called, for definiteness, the
temperature, and equations of the form (5 .7 3 ')-(5 .7 5 ') will be used,
but with ■& replacing T. (As was explained in Sect. 1.5, the use of the
symbol # indicates that we are concerned throughout with a
completely arbitrary passive admixture.)
Equation (5.74') which contains the unknown function q>, may be
simplified considerably for sufficiently large values of z. By
sufficiently large we mean values o f z which, above all, satisfy the
condition z = ^ , so that the molecular viscosity v has no effect
on the distribution of the mean velocity and hence (for the same
reason) no effect on the distribution of the mean temperature.
Moreover, for the values of z under discussion, it is required that
the turbulent heat flux = Cp be much greater than the
molecular heat flux Cppx \ under this condition the coefficient %
will also have no effect on the variation of ^ with height. Since the
turbulent heat flux is produced by the same eddy motion as the
turbulent momentum flux t(')= —pu'w', the eddy viscosity K and
the eddy thermal diffusivity are assumed naturally to be of the
same order of magnitude. Thus, if Pr 1, i.e., then with
•2 > 2 * = TP the eddy thermal diffusivity will be far greater than the
molecular diffusivity x- Therefore, with z > 2 *, both necessary condi
tions will be fulfilled for the values of z to be considered sufficiently
large. However, if P r < I, the eddy thermal diffusivity [which, for
2 > 2 * according to Eq. (5.26) will be of the order of «*2 ] will be far
greater than the molecular diffusivity x only when « > ^ = P r “' • 2 ,.

Thus, generally speaking, we may ignore the effect of both v and %
on the variation of the mean temperature with height only if the two
conditions are satisfied; 2 ^ 2 * and 2 > P r ' ' ’ 2 # (the second of
which, clearly, is only im portant in the case Pr < 1).
With 2 > 2 * and 2 > P r ‘ • 2 *, the mean temperature gradient must
be determined only by the parameters qolcp^y, «*, and 2 . Thus, by
dimensional reasoning, we have

^ 9 ____ £o (5.76)
dz aCp^%U^Z az
where a is a new dimensionless constant of the order of unity.

Consequently,
»(2) — &(0) = -^ln z-l-v 4 , (5.77)
(cf. Landau and Lifshitz (1963), Part 1, Sect. 54), while the
dimensional constant y4i [the same as in the case of Eq. (5.22)] here
must be determined from the matching condition of the profile
(5.77) with the mean temperature profile in the lower layer, to
which Eq. (5.77) is inapplicable.
From Eq. (5.77) it is clear that
<p(2 ^, Pr) = l l n 2 ^ + C for 2+»m a x ( l , P r"'). (5.77')
Thus, for sufficiently large values of 2 +, (p will differ from the

function f of Eq. (5.13) only in the value of the numerical
coefficients (only one of which, C, can depend on the Prandtl
number). Further, we note that by Eq. (5.9) the profile (5.77)
corresponds to the following value of the eddy thermal diffusivity:
AT»= ax«,2. (5.78)
Thus it is clear that the constant a has the meaning of the reciprocal
of the turbulent Prandtl number for a logarithmic boundary layer;
«= pi = T ' = <5.79)
Unfortunately, the data on the concentration profiles of a passive

admixture in real flows close to a wall are considerably poorer than
the data on the mean velocity profile; consequently, until now, the
experimental verification of these equations has remained very
incomplete. The often-repeated measurements by meteorologists of
the mean temperature profile in the surface layer of the atmosphere
are of very little use for this purpose, since in the atmospheric
surface layer with the air temperature dependent upon height (i.e.,
with nonneutral thermal stratification), the buoyancy forces play a
considerable role and do not allow the temperature to be considered

as a passive admixture; for a more detailed discussion, see Chapt. 4.
More promising would be data o f careful measurements of the
humidity profile (i.e., the concentration of water-vapor) in the
surface layer of the atmosphere. In fact, although buoyancy will
affect the vertical distribution of humidity also in the presence of
vertical temperature gradients (see Chapt. 4), the humidity profile
may be observed also in “neutral conditions,” i.e., for an isothermal
regime, whereas the temperature profile will be of interest only when
the conditions are nonneutral. So far, only very little experimental
data exist on this subject; nevertheless, we may mention here the
results of Pasquill (1949), Rider (1954) and other investigators,
which confirm that when the temperature stratification is close to
neutral, the humidity profiles are well described by the logarithmic
equations (see below. Sect. 8.2). In laboratory shear flows ac
companied by intensive forced convection, the temperature may be
considered as a passive admixture with greater justification than in
the atmosphere; however, the published measurements of the
temperature profiles, (or, moreover, on the admixture concentration
profiles) in laboratory flows close to a wall are extremely scanty, and
most of them possess very low accuracy. Thus all that may be said of
them is that they do not contradict the law (5.77). However, as far
back as the end of the 1920’s, Elias showed that in the boundary
layer on a heated flat plate, the temperature profile for z > 0.056 is
approximately logarithmic, and similar to the mean velocity profile
[see Elias (1929)]. Later, Nunner (1956) also obtained in a
nonisothermal flow of air along a rough tube, an almost exactly
logarithmic profile of the mean temperature, commencing with z =
0.02^ and continuing almost to the center of the tube. The
measurements of the temperature profiles by Reynolds, Kays, and
Kline (1958), Brundrett, Baines, Peregrym, and Burroughs (1965)
and Perry, Bell and Joubert (1966) are also in good agreement with
the supposition that a comparatively thick layer having a logarithmic
temperature profile exists. Finally, careful measurements by Johnk
and Hanratty (1962) and by Hishida (1967) of the temperature
profiles for air flow in a smooth heated tube with low values of heat
flux which permit consideration of fluid properties as constant and
also reduce all other sources of errors, show quite definitely that in
the fully developed thermal regime and at all Re ^ 2 5 , 0 0 0 ,
temperature profiles have a logarithmic form from = 30 to =
200-300 (see, e.g.. Fig. 43 taken from Johnk and Hanratty’s paper).
The results of Johnk and Hanratty have rather low scatter; theiâver-
aged data on dimensionless temperature T^.(2 ) = Cppu*[T(z) - T(0)]/
when K = 0.4, can be represented by the equation
= 0.9 Inz^ + 1.3
(i.e., according to the data Pr^ = 0.9, a = 1.1,C(0.7) = 1.3. Almost

the same results were obtained by Hishida; according to his data Pr^ =
0.87, a = 1.15, C(0.7) = 1.4. Also, the considerably more scattered
experimental data of Reynolds, Kays and Kline (1958), and Perry,
Bell and Joubert (1966), were treated by the latter who obtained
quite close mean estimates: Pr^ = 1, « = 1, C(0.7) = 1.6.
25
20
15
n T+ = 3 . 7 2 + 4.9 6!nZ+
10 ^^ 0
^ o
0
oâ
...
10 10^ 10^
FIG . 4 3 . T h e d is tr ib u tio n o f th e d im e n sio n le ss m ea n te m p er a tu re T+ ( z ) = Cp p u ^ [ T ( 0 ) ] I qq

in a lay er o f f lo w in a tu b e a c co r d in g to tw o o f J o h n k a nd H a n r a tty ’s m e a su r em en ts: a)
Re = 7 1 ,2 0 0 ; b ) R e = 7 0 ,2 0 0
In the outer part of the nonisothermal flow in a tube, channel or

boundary layer, an extended Reynolds number similarity principle
(namely, a Reynolds and Piclet number similarity principle) appar
ently must be valid. According to this principle, the statistical
characteristics o f the temperature field are not dependent directly on
the molecular viscosity and thermal diffusivity (i.e., on dimensionless
combinations Re, Pe, and Pr = Pe/Re) for a region of the flow

distant from rigid walls, if both Reynolds and Peclet numbers are
sufficiently large. The most im portant result of the principle for the
mean temperature distribution is a general temperature defect law of
the form
«?i - ^(2) \
- 4 ^ - ■ 'fl*’' ' ' <5,80)
where is a tube radius, channel half-width, or boundary-layer

thickness and the function (piiij) can o f course be
different for different types of flow. The data of Johnk and Hanratty
for flow in a tube at various Re (but at only one Pr = 0.7) support
the validity o f the temperature defect law; according to these data,
for a tube flow the function cpiirj) from ri = 0 .0 2 and to 7 = 1 can be
approximated by the equation 9 ^(77) = 2.9 [or by the equation
(fiirj) = - rj^/12)]. Some applications of the temperature defect
law for a tube flow can be found in the paper by W. Squire (1964).
Some preliminary measurements of the function for a
nonisothermal boundary-layer flow were published by Zukauskas,
Slanciauskas, PediSius, and Ulinskas (1968); however, their data do
not agree with the assumption o f independence of q»ôn Pr (perhaps
because the values of Re and Pe were not high enough in the
measurements). We must also stress that in the overlap layer where
both the temperature wall law (5.74') and the temperature defect
law (5.80) apply, the functions and must both be logarithmic
(cf. Sect. 5.5); however, the definition of the form of the
functions outside of the logarithmic layer and the accurate experi
mental verification o f the temperature defect law strongly require
further careful measurements of temperature profiles in all three
types of flow at high values of Reynolds and Peclet numbers, and for
a wide range o f Prandtl numbers.
The existence o f an overlap layer where both the temperature wall
law and the temperature defect law have a logarithrriic form permits
a quite general heat transfer law to be obtained. In fact, the sum of
two logarithmic laws can easily be written in the form:
5 ,-^ 0 1 ^ i “*
_ 1 -------^ = 1 In J _ + C ,(P r)
where ^(0). (Pr) = C(Pr) + , and Cj is the constant term in

the logarithmic form of the temperature defect law (it is clearly
independent of Pr, but may differ for tubes, channels and boundary
layers). The last equation gives us the following general law for the
main dimensionless characteristic of the heat-transfer process the
heat-transfer coefficient (Stanton number)
Cu =
where U is some typical velocity scale of the flow:
1 In 2
c. = ------------- i------------- , C„ = i c , (Pr) - — > (5.80')
ln(R ev^)/»<a + Cj(Pr) “ 2xa
where Re = UH^/v and Cf = 2(u*/U)^ is the friction coefficient

(Kader and Yaglom (1970); cf. also the analogous derivation of Eq.
(5.45) in Sect. 5.5).
For the case of heat transfer in air (Pr = 0.7) we know th at C(0.7 =
1.3; moreover, for the case of flow in a circular tube the constant
Cj can be roughly estimated from the data o f Johnk and Hanratty
(1962) on the temperature defect law as Cj « 0.3. Therefore all the
coefficients in Eq. (5.80') for heat transfer in an air flow in a circular
tube can be considered as known. However, comparison of the
computations using this equation with the heat-transfer data is not
straightforward since almost all the existing data on heat transfer in
tubes relate not to the tube axis temperature ), but to the
bulk temperature (i.e., to the average temperature of the fluid
flowing through tubes)
Hj / Hj
Mz)u{z)(H^ - z)dz uizXH^ - z)dz,
and as the typical velocity the buld velocity
u( 2 )(Hj - z)dz
R2 J
is generally used. In other words, the existing data give us not the
values of the coefficient ch, but the values of the modified heat-
transfer coefficients
Ch = % /C ppU Jô - = c ^ /a - A)
where
A = —----- 3 - •
^0 -
In estimating the correction A for practical purposes it is possible,
at least for air flow in a tube, to neglect the thin wall layer o f the
direct influence of the molecular viscosity and thermal conductivity
and to consider the velocity and temperature profiles as purely
logarithmic ones. In this approximation it is easy to show that
3.75 c* - 1/2
when « 1
(cf. the analogous derivation of Eq. (5.47) in Sect. 5.5). Using this
equation for A, equation (5.80') may be placed in the form
----------- , C4 (Pr) = Cg(Pr) - ln2/xa - 3.4

ln(Re£)VC/)/xa + C4 (Pr) 80”)
where Re^ = DU^/u = 2 H ^ J v , Cf = 2(u»/i;^)2. Further, making

use of the data from Sect. 5.5 on the dependency o f C, = A/4 orR e^
(see Fig. 33) we can calculate from Eq. (5.80") the dependence of
the heat-transfer coefficient Cf,, or, what is the same, of the so-called
Nusselt number Nu = C;jRePr, on Reynolds number Re^. The
comparison of the results with data from various sources is shown in
Fig. A, where simultaneously the widely used purely empirical
relation Nu = 0 . 0 1 8 [cf. for example, McAdams (1954)] is
presented.
The comparison of the computations using Eqs. (5.80') and
(5.80") with data on heat and mass transfer in tube flows with other
values o f Prandtl number will be considered later in this subsection.
Both the temperature law o f the wall and the temperature defect
law can be generalized for the characteristics of temperature
fluctuations quite similar to the corresponding results for velocity
fluctuations. However, at present, there are almost no reliable
experimental data suitable for a verification of the relationships

predicted by these generalized laws.
0.018Reo ■ X
^ 10^
i ♦ - Cogian (1940)
CJ
f - Barnet, Kobe (1941)
II 4 - Drexel, McAdams (1945)
♦ f « Gukhman, Ilyukhin
I ^ Gondelsman, Naurits '
♦ J r ^-Nunner (1956)
'^-Lebchuk, Dyadyakin (1958)
▲ rp-Sleicher (1958)
^-Ede (1961)
7-Hishida (1967)
Hasegawa, Fujita (1968)
0 “ Dyban, Epik (1968)
10
5-10^ to* 5 - 1 0 ‘^ 10^ 5 -1 0 ^
Reo
FIG. A. C o m p a r iso n o f d a ta o n tu r b u le n t h e a t transfer in air f lo w in tu b e s fr o m various
so u r ce s w ith th e o r e tic a l d e p e n d e n c e im p lie d b y E q. ( 5 . 8 0 ' ) [c o n t in u o u s lin e ) a n d w ith th e
em p irica l fo r m u la N u = 0 . 0 1 8 Re£)® * [d a sh e d lin e] a c co r d in g to K ader and Y a g lo m ( 1 9 7 0 ) .
The lack of data considerably complicates also the reliable

estimation of the coefficient a (or Pr^ = 1). In the majority of apphed
heat-transfer calculations concerning both engineering devices and the
atmosphere, until now, following O. Reynolds (1874), it was usually
assumed that a = 1. However, as a rule, these calculations lead to
fairly good agreement with the data from direct measurements.
Nevertheless, the integral heat-transfer characteristics which are of

special interest for heat technology, are usually quite insensitive to
not too great variations of the parameter a and hence do not allow
reliable estimates of its value to be made. Perhaps the best existing
estimate of a and Pr^ for a logarithmic layer is that following from the
data of Johnk and Hanratty (1962), and Hishida (1967), and which
was given briefly above. Attempts to estimate a directly from the
measurements of mean velocity- and temperature-profiles for flows
in tubes and channels, also result generally in values of a that are
somewhat greater than unity, of the order of 1.1 —1.4 [i.e., to the
deduction that Pr^ « 0.9—0.7 < 1; see, for example, Corcoran et al.
(1952); Cavers, Hsu, Schlinger and Sage (1953); Hsu, Sato and Sage
(1956); Ludwieg (1956); Sleicher (1958); Venezian and Sage (1961);
Zukauskas et al. (1968); and also the book by Longwell (1966) and
the survey articles of Deissler (1959), Kestin and Richardson (1963),
and Blom and De Vries (1968)]. However, sometimes (principally, in
a number of experiments on heat-transfer in liquid metals) values of a
slightly smaller than 1, i.e., Pr^ > 1 are also obtained [see, for
example, Subbotin, Ibragimov, Nomofilov (1963); Dwyer (1963) and
Longwell (1966)]. The determination of a from the wind and
temperature profiles in the atmosphere may be carried out only in
the case of temperature stratification other than neutral; however,
here, it is possible to attem pt to find the limit to which a tends as the
stratification approaches neutral, and to equate this limit to the ratio
of the coefficients and K for the logarithmic boundary layer. Such
attempts have been carried out many times, but with diverse results.
Thus, for example, Rider (1954), Priestley (1963—1964) and
Swinbank (1964) found that a ^ 1, but in the works of Swinbank
(1955) and Gurvich (1965), the value a < 1 was obtained. The results
collected by Panofsky et al. (1967) give the impression that a is very
close to unity (see below. Sect. 8.2). In general, we must bear in mind
that, so far, all the existing data on turbulent heat-transfer (and
mass-transfer are largely contradictory, and great care must be taken
in using them. Thus, for example, even in works which give
approximately the same mean value of a, we find obvious contradic
tions in the assertions regarding the variation of a with variation of
the distance from the wall z. (According to Ludwieg (1956) a in
creases slightly with increase of z, while, on the other hand, according
to Corcoran et al. (1952) and Sleicher (1958) a decreases with increase
of 2 .) These facts may easily be explained as a result of low accuracy
of all the existing measurements. This is especially dangerous because

the experimental graphs must be differentiated to obtain a orPr^. In
this connection, it is worth noting once again that according to the
general considerations given above, in the logarithmic region of fully
developed flow, where both the shear stress and the turbulent heat
flux are constant, the coefficient a must also be constant (i.e.,
entirely independent o f 2 ). Moreover, since we are dealing here with
a region of flow in which both the molecular viscosity and thermal
diffusivity play no part, a must be independent of them, and hence
must also be independent of the molecular Prandtl number Pr. This
last assertion is in agreement with the surprisingly small variations of
a values obtained for fluids ranging from machine oil (investigated by
Zukauskas et al.) to liquid metals, i.e., having an enormous range of
Pr-values. Further, if we believe that conditions of forced convection
in which the temperature is transferred as a purely passive admixture
are possible, then the values o f a for heat- and for mass-transfer must
be considered as completely identical. However, if these two values
differ, then this can only mean that some physical mechanism
(unknown to us) exists that produces a different effect on the
transfer of heat and of mass.
For very small values o f = z u j v , for a smooth wall, naturally,
the equation
,?(e) - ,?(0) = - , cp(3^, Pr) = KPr - z . , (5.81)
must hold (this is analogous to Eq. (5.20). Further terms of the

expansion of the function <p(2 +, Pr) as a Taylor series in powers of
in the neighborhood of the point = 0 may be obtained by
differentiating Eq. (5.73') at the point 2 = 0 . Since in the case of a
wall at constant temperature r?' = 0 at 2 = 0, then, for Eq. (5.20'),
with accuracy to fourth-order terms in 2 ^, we will have
.3 , 4
t>(2 ) = d(0) - + c; ^ 0^ ^ Pr) ^ KPr( 2 ^ - c\z\)
(5.81')
*w
From the fact that the second and third derivatives o f <p(z^, Pr)
with respect to become zero at z^ = 0, it follows that the change
in temperature will be close to linear over a considerable range of
values of thus we may introduce the concept of a sublayer of
molecular thermal diffusivity, in which the values of the function
(p(z^,Pr) practically do not differ from K ? r . 2 ^. However, the
thickness of this sublayer will now be determined by the three
parameters v, x» and u* so that, generally speaking, we can only state
that = tlr(Pr)v/u^. Directly beyond the sublayer of molecular
thermal diffusivity there must follow some transitional region,
intermediate between the region of applicability of Eq. (5.81) and
the logarithmic layer. Beyond this layer begins the logarithmic
temperature layer, described by Eq. (5.77). Unfortunately, at
present, there are no reliable direct measurements o f the temperature
profile, or the profile o f admixture concentration, close to a wall
which would permit us to verify all these general deductions and to
trace the complete source of the function (p(2 ^.,Pr). Therefore, we
must infer the values o f this function for small z^ on the basis of
indirect data, obtained in the study of heat- and mass-transfer in
turbulent flows.
T h e se in d ir e c t d a ta ( o f w h ic h th ere are a la ig e n u m b e r ) c o n s is t o f d e d u c t io n s o n th e
v a lues of th e d im e n sio n le ss in teg ra l h e at-tra n sfer ch a r a cter istic s, th e N u s s e lt n u m b e r
Nu = ^ c o e f f ic ie n t (t h e S ta n to n n u m b e r ) C/, =
= — , r , a ° ------ 5 -T fo r d iffe r e n t v a lu es o f R e s = a n d Pr = — . H ere

Reg P r Cppt/ ( » , — »o ) ^ X
* ^We n o t e th a t u n lik e th e e x a c t e q u a t io n ( 5 . 2 0 ') , E q . ( 5 . 8 l ’) fo r th e m e a n te m p er a tu re

p r o file (b u t n o t fo r th e c o n c e n tr a tio n p r o file o f a m ateria l passiv e a d m ix tu r e ), is o n ly
a p p r o x im a te . In fa c t, th e in itia l e q u a t io n ( 5 . 7 3 ' ) fr o m w h ic h E q. ( 5 . 8 1 ' ) is d e d u c e d , is it s e lf
a p p r o x im a te , s in c e in d e d u c in g it w e d r o p th e term d e scr ib in g th e h e a tin g o f th e flu id b y
th e d is s ip a tio n o f k in e t ic e n e r g y , w h ic h o c cu rs in E q . ( 1 . 7 0 ) [S e c t. 1 . 5 ] . I f w e reta in th is
te r m , th e n in th e T a y lo r e x p a n s io n fo r th e fu n c t io n (jp(z+), term s o f th e ord er o f z\ and
occu r, w ith c o e f f ic ie n t s d e p e n d in g on th e sta tistic a l p r o p e r tie s o f th e v e lo c ity
flu c tu a tio n s . T h e w o r k o f T ien ( 1 9 6 4 ) is d e v o te d t o th e e s t im a tio n o f t h e s e te r m s; h e s h o w s
th a t in th e la y e r 1 0 , th e y m a y p la y s o m e r o le o n ly i f > 10~® (w h e r e is n o w
a ssu m ed t o b e e x p r e s s e d in u n its w ith d im e n sio n s
i9 >o= 0 ( 0 ) is th e te m p er a tu re o f th e w a ll (g e n e r a lly , a fla t p la te or a tu b e w a ll) w h ic h is

a ssu m ed c o n s ta n t, is th e te m p er a tu re o f th e f lo w o u tsid e th e b o u n d a r y la y e r or o n th e
axis o f th e tu b e , 6 is th e th ic k n e ss o f th e b o u n d a r y la y e r or th e radius o f th e tu b e , U is th e
v e lo c ity o f th e u n p e rtu r b e d f lo w or th e v e lo c ity o n th e a x is o f th e tu b e.^ ^ T h e sam e
d im e n sio n le ss c h a ra cteristics are d e te r m in e d u su a lly also in c a lc u la tio n s o n m ass-transfer; in
this case i t is o n ly n ec essa ry t o r ep la c e b y /o , o m it Cp in th e d e n o m in a to r s o f N u and Cp
and ta k e 'ô a n d to d e n o t e th e c o rr e sp o n d in g c o n c e n tr a tio n s . D eissler ( 1 9 5 1 ; 1 9 5 4 ) ,
H a tto n ( 1 9 6 4 ) and several o th e r in v e stig a to r s, s h o w e d th a t v e lo c ity , te m p er a tu re and
c o n c e n tr a tio n p r o files, as w e ll as all in teg ra l c h a ra cteristics o f h e a t- and m a ss-transfer in
tu b e s an d b o u n d a r y la y ers, are r ela tiv e ly in se n sitiv e (a t le a st, w h e n th e R e y n o ld s n u m b e r is
n o t t o o large and th e Prandtl n u m b e r n o t t o o s m a ll) t o v a r ia tio n o f t a nd q a lo n g th e Oz
axis. T h u s th e m o d e l o f a c o n s ta n t stress an d c o n s ta n t f lu x la y e r m a y u su a lly b e u sed as a
rea so n a b le (o r e v en e x c e lle n t ) first a p p r o x im a tio n . I f w e n o w r ew rite E q . ( 5 . 7 7 ') as
»(^)_9„ = e . ( i | n ^ + c ) , (5.77")
th en to d e te r m in e th e d iffe r e n c e Oi — Oo it is o n ly n ecessa ry to k n o w th e pa ra m eters x and

a a n d t o e stim a te th e c o n s ta n t C (w h ic h d e p e n d s e ss e n tia lly o n th e v a ria tio n o f th e
te m p er a tu re in th e th in la y e r c lo s e t o th e w a ll). A w h o le series o f th eo r ie s h a s b e e n
p r o p o s e d fo r th e e s tim a tio n o f th is c o n s ta n t. T h e y are c o n fir m e d e x p e r im e n ta lly w ith
v a ry in g d eg rees o f ac cu ra c y .
T h e s im p le s t h y p o th e s is p e r m ittin g th e e s t im a tio n o f th e v alue C w as p r o p o s e d b y
O. R e y n o ld s ( 1 8 7 4 ) . H e p r o c e e d e d fr o m th e a s su m p tio n th a t th e m e c h a n is m s o f th e transfer
o f h e a t a n d m o m e n tu m in a tu rb u le n t f lo w are id e n tic a l; th u s th e tu r b u le n t h e a t flu x q
a c c o r d in g to R e y n o ld s ’ th e o r y m u st b e c o n n e c te d w ith th e shear stress t b y a r ela tio n sh ip o f
th e fo r m
(5 .8 2 )
w h e r e U\, 6 ^2 , and i0'2 are values o f th e m e a n v e lo c ity and te m p er a tu re at tw o arbitrary

le v e ls. A c c o r d in g to th is r e la tio n s h ip , th e m e a n v e lo c ity and m e a n te m p er a tu re p r o files in
th e w h o le c o n s ta n t flu x la yer (c o n s ta n t T and q ) m u st b e e x a c tly p r o p o r tio n a l (w ith
c o e ffic ie n t of p r o p o r tio n a lity equal to C p z jq ). T h e r e fo r e , it f o llo w s th a t
q )( 2 +, P r ) = x / ( z + ) [ s o th a t q) is in d e p e n d e n t o f Pr] a n d C = ycB. C o m p a rin g E q. ( 5 . 8 1 )
w ith E q. (5 .2 0 ), and E q . ( 5 . 7 7 ' ) w ith E q. ( 5 . 2 5 ) , i t is e a sy to ascerta in th a t this
p r o p o r tio n a lity w ill o ccu r o n ly i f / = v (i.e ., i f Pr = 1 ), a = 1 (i.e ., Pr^ = 1 in th e
lo g a r ith m ic la y e r s), an d , m o r e o v e r .
dz dz
(i.e ., Pr^ = 1 a lso in th e e n tir e tr a n sitio n a l reg io n b e t w e e n th e v is c o u s su b la y e r and th e
^ ^We h a v e a lready n o t e d th a t in pra ctica l stu d ies o f h e a t transfer in tu b e s, th e b u lk

v e lo c ity and th e b u lk te m p er a tu re are g en era lly u se d in ste a d o f U a n d . H o w e v er ,
w e shall n o t d w e ll o n th is p o in t, s in c e i f w e k n o w th e va lu es o f th e criteria N u and c ^ , b a sed
on a n d t /, it is e a sy to o b ta in th e v a lu e s o f th ese criteria b a sed c n and q u ite sim ilar
to th e d e v e lo p m e n t p r e se n ted in S e c t. 5 .5 [c f. th e d e r iv a tio n o f E q. ( 5 . 4 7 ) ] and in th is
s u b s e c tio n in c o n n e c t io n w ith th e d a ta o f F ig. A.
lo g a r ith m ic la y e r ). We n o t e th a t th e c o n d it io n Pr = 1 is o n ly a p p r o x im a te ly sa tisfie d (w ith

a ccu racy n o t e x c e e d in g 2 0 - 3 0 % ) fo r air and fo r th e m a jo rity o f o th e r ga ses, and is n o t
sa tisfie d at all fo r an im p re ssiv e m a jo rity o f liq u id s; th e c o n d it io n Pr^ = 1, o n th e o th er
h a n d , im p o se s severe r estr ic tio n s o n th e d iso r d e re d v e lo c ity a n d te m p er a tu re flu c tu a tio n s
w h ic h a lso rarely ever h o ld e x a c t ly . N e v er th ele ss, R e y n o ld s ’ a s su m p tio n [o r as m o r e
c o m m o n ly c a lle d , the Reynolds analogy, in v ie w o f th e a n a lo g y b e t w e e n th e flu x o f h e a t
and m o m e n tu m e x p r essed b y E q . ( 5 . 8 2 ) ] w r itte n in th e fo rm
C/, = i Cf, (5 .8 3 )
is a surprisingly g o o d d e s c r ip tio n o f m a n y e x p e r im e n ta l r esu lts o n tu rb u le n t h e a t e x c h a n g e in

gases, a n d , h e n c e , is still u se d fairly w id e ly in en g in e e r in g c a lc u la tio n s.
A n o b v io u s d e fe c t o f th e R e y n o ld s a n a lo g y is th a t it ig n o r e s th e e f f e c t o n th e h e a t
e x c h a n g e o f th e m o le c u la r Prandtl n u m b e r , w h ic h in a n u m b e r o f cases p la y s a d e fin ite part.
A s im p le g e n e ra liza tio n o f E q. ( 5 . 8 3 ) in te n d e d to ta k e in t o a c c o u n t th e e f f e c t o f Pr in s o m e
m a n n er, w a s p r o p o s e d in d e p e n d e n tly b y P ra n d tl ( 1 9 1 0 ; 1 9 2 8 ) , a n d G. I. T a y lo r ( 1 9 1 6 ) .
A c c o r d in g to th eir p o s tu la te s, it m a y b e a ssu m ed th a t a = 1 in th e lo g a r ith m ic la y e r , b u t in
V
th e v isco u s su b la y e r (e x te n d in g right up to ^ = ) w e m u st use th e str ict e q u a tio n s
( 5 .2 0 ) and ( 5 . 8 1 ) , i.e ., w e m u st ta k e in to a c c o u n t th e true valu es o f th e c o e f f ic ie n t s v and x-

F o r th e tr a n sitio n a l reg io n , th is is n e g le c te d in th e P rand tl-T a y lo r th e o r y , i.e ., it is a ssu m ed ,
V
in e ff e c t , th a t fo r z > — th e lo g a r ith m ic e q u a tio n s ( 5 . 2 5 ) and ( 5 . 7 7 ) h o ld . T h u s th e
fu n c tio n c p (z+ , P r ) is d e fin e d b y E q. ( 5 . 8 1 ) fo r z+ < and b y E q. ( 5 .7 7 ') w ith a = 1

fo r > OLo\ th u s, in p articular, it f o llo w s th a t in th is th e o r y , C = a t , x P r - l n a „ . H o w e v er ,
it is ea sy to v erify th a t th is v alue o f C (an d a = 1) c o r r e sp o n d s t o th e heat-tra n sfer
c o e f f ic ie n t
1
Ch --- ( 5 . 8 4 )
and this is th e m a in resu lt o f th e P ra n d tl-T a y lo r th e o r y . F o r Pr = 1, E q. ( 5 . 8 4 ) b e c o m e s

id e n tic a l t o E q. ( 5 .8 3 ) ; h o w e v e r , i f Pr 1 (b u t d o e s n o t d iffe r g r ea tly fr o m u n it y ) th e n ,
w ith a r ea so n a b le c h o ic e o f (P ran dtl, fo r e x a m p le , t o o k a „ = 5 . 6 ) it giv es s o m e w h a t
b e tte r a g r ee m e n t w ith th e e x p e r im e n ta l d a ta th a n R e y n o ld s* fo r m u la ( 5 . 8 3 ) .
A fu rther im p r o v e m e n t to th e Prand tl-T a y lo r th e o r y w a s p r o p o s e d b y v o n Kdrmdn
(1 9 3 4 ; 1 9 3 9 ) w h o in tr o d u c e d th e “ tra n sitio n a l la y e r ” a v < z + < a i in t o th e c a lc u la tio n
(w h ere t t u = 5 , a / = 3 0 ; se e a b o v e S e c t. 5 .3 ) . A s w e h a v e a lready o b serv ed (at th e e n d o f
S e c t. 5 .3 ) , v o n Kdrmdn a ssu m ed th a t in th is tr a n sitio n a l la y er th e v e lo c ity p r o file w a s also
given b y a lo g a r ith m ic e q u a tio n o f fo r m ( 5 .2 5 ) , b u t w ith d if fe r e n t valu es o f th e c o e f fic ie n ts
2 r.
A an d in particular, h e t o o k ^ a n d , th e r e fo r e , K = “ 0 . 2 v z + fo r 5 <
< 30. In th e c a lc u la tio n o f th e m e a n te m p e r a tu r e p r o file [i.e ., o f th e fu n c tio n

q )(z + , P r ) ] it w as a ssu m ed th a t in th e r eg io n z+ < 5 , th e rela tio n sh ip ( 5 . 8 1 ) is sa tisfie d
(i.e ., o n ly m o le c u la r th erm a l d iffu s iv ity is o p e r a tin g ), in th e reg io n 5 < z + < 3 0 , t h w e is
b o th m o le c u la r and tu rb u le n t th erm a l d iffu s iv ity w h ile /(» « = /(, th a t is, =

dz
C p (x j-°0 2u z) ' ^ c o e f f ic ie n t x m a y b e ig n o r e d and w e m a y a ssu m e

th a t = - --------I t ------- (PTf in th is th eo r y fo r 2 > 5 v / u * is ta k en t o b e e v er y w h e r e

dz Cp^ • 0 .4 a , *
eq u a l to u n ity ). O n th is basis, it is e a sy to o b ta in th e e n tir e c o u r se o f th e fu n c tio n , <p(2+, P r )
[d e f in e d b y th re e d iffe r e n t a n a ly tic a l e x p r e s s io n s ] a n d , in p articular, to fin d th e valu e o f C;
fo r Cf, w e th en o b ta in th e e q u a tio n
1
c, = ------------------------------- ^---------------------------- = • (5 .8 5 )
1 + 5 [(Pr - 1) + In ] /^
F o r Pr = 1, th is e q u a t io n o n c e again b e c o m e s Eq. ( 5 . 8 3 ) b u t w ith values o f Pr o f th e order

o f several u n its o f o n e or t w o te n th s (i.e ., fo r ordin ary liq u id s), it gives co n sid e ra b ly b e tter
a g reem en t w ith e x p e r im e n ta l d a ta th an d o e s th e Pran d tl-T aylor eq u a tio n ( 5 .8 4 ) . H o w ev er,
for vary large or very sm all valu es o f Pr, Eq. ( 5 . 8 5 ) also lead s to sharp d e v ia tio n from
e x p e r im e n t.
Later, various a u th o rs p r o p o s e d a co n sid era b le n u m b e r o f n e w sem iem p irica l (or sim p ly
e m p irica l) e q u a tio n s fo r Ch (or N u ); see, e.g ., th e w o r k s c ite d at th e end o f S e c t. 5 .3 and th e
papers b y R ib a u d ( 1 9 6 1 ) ; K a p itza ( 1 9 4 7 ) ; S h e r w o o d ( 1 9 5 0 ) ; L evich ( 1 9 5 1 ) ; R eich a rd t
( 1 9 5 1 b ) ; C h apm an an d K ester ( 1 9 5 3 ) ; D eissler ( 1 9 5 1 ; 1 9 5 4 ) , P e tu k h o v and Kirillov ( 1 9 5 8 ) ;
van D riest ( 1 9 5 9 ) ; S p ald in g ( 1 9 6 3 ; 1 9 6 4 ) ; W. Sq uire ( 1 9 6 4 ) ; Sp ald in g an d J ayatillak a
( 1 9 6 5 ) ; K ish en ev sk y ( 1 9 6 5 ) ; and D il’m an ( 1 9 6 7 ) . In th ese w o r k s, n u m e ro u s a d d itio n a l
r eferen ces m a y b e fo u n d . A lso , in a c o n sid era b le part o f th ese w o rk s, th e basic im p ro v e m e n t
t o th e th eo r y co n sists in th e c h o ic e (o n th e b asis o f variou s in tu itiv e or em p irical
c o n sid e r a tio n s ) o f a m o r e c o m p lic a t e d e x p r ess io n fo r th e fu n c tio n (p (z + , P r ) w h ic h is given
g enerally in th e fo r m o f an e q u a tio n for /(^ = --------------------------- y . C o n s e q u e n tly , th e

Cp^ dhjcLz
c a lc u la tio n o f th e p r o file ^ ( e ) [i.e ., th e fu n c tio n (p (z + , P r )] a n d th e d e te r m in a tio n o f th e
c o e f f ic ie n t Ch r ed u c e to an in te g r a tio n (w h ich fo r m o r e c o m p lic a te d th eo r y m u st b e carried
o u t n u m e r ic a lly ). T h e n u m e ric a l p r o c ed u r es fo r s o lv in g th e various h e a t- and m ass-transfer
p r o b le m s are d isc u sse d , e .g ., b y H a tto n ( 1 9 6 4 ) , D vo ra k and H ead ( 1 9 6 7 ) , D o n o v a n , H anna
and Y era zu n is ( 1 9 6 7 ) , and Patnakar a n d Sp ald in g ( 1 9 6 7 ) . We shall n o t d w e ll h ere o n th ese
results, w h ic h are b a sic a lly o f a p u re ly en g in e e r in g character, and shall c o n sid e r o n ly th e
p h y sic a lly in ter e stin g q u e s tio n o f th e sp e cia l fea tu res o f h e a t transfer fo r very large or very
sm all Prandtl n u m b e r s.
In b o t h th ese lim itin g c a ses, th e a s su m p tio n m a d e in th e Pra nd tl-T a y lo r and v o n Kdrmdn
th eo r ie s th a t th e th ic k n e ss o f th e m o le c u la r th erm al d iffu s iv ity la y e r m a y b e e stim a te d as
5^ = , w h e r e o„ is fo u n d fro m th e in te r s e c tio n o f th e t w o a s y m p to t e s o f th e m ea n
v e lo c ity p r o file a n d h e n c e is in d e p e n d e n t o f Pr, is clearly u n ju stified . L et us tak e first th e

ca se o f very sm a ll Prandtl n u m b e r s, w h ic h are prim arily c h ara cteristic o f liq u id m e ta ls (i.e .,
o f m er cu ry a n d m o lt e n m e ta ls ). S in ce in th is case x is very large, th e m o le c u la r th erm al
d iffu s iv ity la y e r is c o n sid e ra b ly th ick er th a n th e v isco u s su b la y er; th u s, n e g le c tin g th e
c o e f f ic ie n t x in th e w h o le r eg io n o f th e lo g a r ith m ic m e a n v e lo c ity p r o file h ere m u st lead to
serio u s errors. T o a v o id th e m , m a n y in v e stig a to r s, e sp e c ia lly th o se w h o stu d y th e th eo r y o f
tu r b u le n t h e a t e x c h a n g e in liq u id m e ta ls [fo r e x a m p le , M artinelh ( 1 9 4 7 ) ; L y o n ( 1 9 5 1 ) ;
L y k o u d is and T o u lo u k ia n ( 1 9 5 8 ) e t a l . ] , did n o t ig n o r e th e m o le c u la r th erm a l d iffu siv ity in
a n y f lo w r eg io n . T h is is n o t , in fa c t, n e c essa ry , at lea st in th e ca se o f an id e a liz e d flo w in the
w h o le h a lf-sp a c e 2 > 0 ; b u t th e th ic k n e ss 6-, o f th e la yer w ith in w h ic h th e c o e f fic ie n t X
still p la y s a r o le (i.e ., is c o m p a r a b le w ith th e e d d y d iffu s iv ity K e ) for Pr 1 m u st be
e stim a te d w ith th e aid o f th e r e la tio n s h ip s x (s in c e K ^ 2 for ^ v / a ^ and
/<8 bas th e sam e o rder o f m a g n itu d e a s/C ). T h u s, fo r e x a m p le , in th e sim p lifie d “ tw o -la y e r ”
P r a n d tl-T a y lo t m o d e l, fo r Pr 1 w e m u s t a ssu m e th a t< p (^ + , P r ) is g iv en b y E q. ( 5 . 8 1 )
fo r
*/ 0 (w h e r e is n o w in d e p e n d e n t o f Pr and is o f th e sa m e o rd er as th e
d im e n sio n le ss th ic k n e ss o f th e v isco u s su b la y er a ^ ) a n d is g iv e n b y E q . { S . l l ' ) fo r >

O Q
SO th a t C = — a In O n th is basis it is e a sy to c a lc u la te th e heat-tra n sfer
c o e f f ic ie n t Ch. S u ch a c a lc u la tio n w a s carried o u t b y L e v ic h ( 1 9 6 2 ) , ta k in g P„ = 1 1 . 7 , a =

1 .2 . In sp ite o f th e r o u g h n e s s o f th e a s su m p tio n s u sed in th e c a lc u la tio n , it is f o u n d th a t th e
e q u a tio n th u s o b ta in e d c o rr e sp o n d s fa irly w e ll t o m a n y o f th e d a ta o n h e a t transfer fo r
tu r b u le n t f lo w s o f liq u id m e ta ls in tu b e s.
T h e e x p r e s s io n fo r C (P r) = C i( P r ) - C 2 ^ C i( P r ) - 0 .3 can b e a lso s u b s titu te d d ir e c tly
in to Eq. (5.8 0 ^ ) for . S u ch a p r o c ed u r e w a s u s e d b y K ader a n d Y a g lo m ( 1 9 7 0 ) [w h o
to o k th e value a = 1 . 1 and a c o m p a r a tiv e ly lo w va lu e o f jSy]; a fter in tr o d u c in g th e
c o r r e c tio n c o n n e c te d w ith th e u se o f in s te a d o f (i.e ., th e tr a n s fo r m a tio n fr o m C;, to
Ch) th e y o b ta in e d fr o m E q. (5 .8 0^ ) a rather g o o d d e s c r ip tio n o f n u m e r o u s and rela tiv ely
accu ra te d a ta o n h e a t transfer in tu b e flo w s o f m er cu ry and s o d iu m -p o ta s s iu m a llo y (b o th
havin g Pr = 0 . 0 2 5 ) .
T h e s e c o n d lim itin g ca se Pr 1 is e n c o u n t e r e d in th e s tu d y o f h e a t transfer in
te c h n ic a l o ils , or in th e d if fu s io n o f m a ter ia l a d m ix tu r e s in liq u id s; it is c o n s id e r a b ly m o r e
c o m p le x . In th is case th e th erm al d iffu s iv ity su b la y er is c o n sid e ra b ly th in n e r th an th e
v isco u s su b la y er a n d , c o n s e q u e n tly , e v en fo r z < ^ th e transfer o f h e a t (o r th e
a d m ix tu r e ) is d e te r m in e d m a in ly b y th e e d d y d iffu s iv ity . A t th e sa m e tim e , th e q u a n tita tiv e
u n d e rsta n d in g o f tu r b u le n c e is m o s t la c k in g ju st in th e reg io n z ^ v / w * , in w h ic h th e
m o le c u la r v is c o s ity still p la y s a sig n ific a n t r o le . S in ce as y e t n o a p p ro a ch is k n o w n to th e
p r o b le m o f d e te r m in in g in d e p e n d e n tly th e b e h a v io r o f th e tw o f u n c t i o n s / ( 2 + ) and 9 ( 2 + )
in th is r eg io n , it is g e n e ra lly a ssu m ed th a t th e c o e f fic ie n ts K an d Kb fo r v / « * e ith e r
agree w ith e a ch o th e r e x a c t ly or e lse d iffe r o n ly b y a c o n s ta n t fa c to r a (c h o s e n m o r e or less
arbitrarily o r r o u g h ly e stim a te d fr o m th e d a ta ). T h e re fo r e th e s tu d y o f tu r b u le n t h e a t
transfer fo r Pr 1 red u c e s to th e in v e s tig a tio n o f th e a s y m p to t ic b e h a v io r o f th e tu rb u le n t
shear stress — p u ' w ' [o r , e q u iv a le n tly , o f o n ly o n e fu n c tio n f ( z + ) ] fo r z v /w * (i.e .,
fo r 1), to w h ic h a n u m b e r o f w o r k s has b e e n d e v o te d (se e th e e n d o f S e c t. 5 .3 )
fo r th is very rea so n . L et us n o w a ssu m e th a t K a s 2 - > 0 (i.e ., th a t I u ' w ' I^
lO 'o;'! ^ ^ ’" » a n d [ / ( ^ ^ ) — Pr ) — x P r . S in ce th e t h ic k
ness 8^ o f th e v isco u s su b la y e r is d e fin e d b y th e c o n d it io n / ( ( 6 v) ^ v , w h ile th e th ic k n e ss
o f th e th erm a l d iffu s iv ity su b la y er is d e fin e d b y th e c o n d it io n o b v io u s
th a t in th e case u n d er c o n s id e r a tio n
T h e p r o file o f Q ( z ) m a y b e d e scr ib e d sa tisfa c to r ily b y th e lin ear e q u a t io n ( 5 . 8 1 ) w h e n
z > 6v o n ly ; in th e la y er ft' < r < th e s o lu tio n o f th e e q u a t io n - Q q = c ^p K^ ^
2 "” ^ gives a rea so n a b le first a p p r o x im a tio n t o th e p r o file (s o th a t ^ ( 2 ) = c o n s t -
w h ile for 2 > 5^ it is tr a n sfo r m e d in t o a lo g a r ith m ic p r o file ( 5 . 7 7 ) . A n a ly tic a lly th ese

resu lts m a y b e o b ta in e d fr o m a sim p lifie d tw o -la y e r , P ran d tl-T a y lo r m o d e l i f w e a ssu m e th a t
fo r 2 ^ < and = aKz^ fo r ( w h e r e i s d e fin e d b y th e
m a tc h in g c o n d it io n = a K ^ ^ ) and ta k e th e m o le c u la r d iffu s iv ity x in t o a c c o u n t w h e n
2, < (i.e ., 2 < 5^^). T h e e q u a t i o n - ^ 0 ^ c v /ith z < th e n m a y b e

dz
tr a n sfo r m e d to
? = r1 *- PrK^
^ =—
Kd, •
In fa c t, w e shall b e d e a lin g w ith a th ree-la y er m o d e l b e c a u s e th e e x p a n s io n o f th e s o lu tio n

o f th is e q u a t io n in p o w e r s o f Pr h as a d if fe r e n t fo r m in th e reg io n s Pr K^ = a^ 2 'J*Pr < 1
and P rK ^ = ^ S o lv in g th is e q u a t io n fo r b o t h la y ers and
> (ajP r)"^ ^'” , jo in in g to g e th e r th e t w o s o lu tio n s a t 2 ^ = (a ^ P r )" ^ '” a n d th e s e c o n d
s o lu tio n w ith th e lo g a r ith m ic p r o file at a n d k e e p in g o n ly th e le a d in g term s in th e
e x p a n s io n s in p o w e r s o f Pr, w e o b ta in ea sily th e fo llo w in g a s y m p to t ic e q u a t io n fo r th e
h ea t-tra n sfer c o e f f ic ie n t fo r P r » 1:
- <Prt " (5 .8 6 )
[c f. L ev ich ( 1 9 6 2 ) ; th e m e t h o d o f d e r iv a tio n o f th e e q u a t io n s k e tc h e d is th a t o f K ader

( 1 9 6 6 ) ] . Here is a d im e n sio n le ss c o n s ta n t p r o p o r tio n a l t o ; a c co r d in g to K ader
00
-l/m
B1 = “1 (-!)•
imk)^ - 1
H o w e v er , w ith r esp e c t to th e valu e o f th e in d e x m th er e is a t p r e se n t s o m e d iv e r g e n c e in

v ie w s, as w e h a v e alrea d y o b se r v e d . W e h a v e seen th a t b y th e c o n tin u ity e q u a t io n w e m u st
h a v e m > 3 . I f w e ta k e m = 3 , th e n a c co r d in g t o E q. ( 5 . 8 6 )
Ch (P r) ® at Pf 1. (5.86')
T h e a s su m p tio n th a t m = 3 , p e r m itte d M urphree ( 1 9 3 2 ) t o o b ta in an e q u a tio n fo r Cf^

w h ic h gave a g o o d d e s c r ip tio n o f all th e d ata o n h e a t transfer in flo w s o f gas, w a te r and
te c h n ic a l oils (w ith Prandtl n u m b e r s u p to 1 0 0 0 ) ; la te r , R ib a u d ( 1 9 4 1 ) c a m e to sim ilar
c o n c lu s io n s . P urely e m p ir ic a lly , th e rela tio n sh ip ( 5 . 8 6 ' ) w a s fo u n d b y C h ilto n and C o lb u rn
( 1 9 3 4 ) w h o s tu d ie d m ass tra n sfer in liq u id s (w it h o u t a n y r e fer e n c e t o th e w o r k o n h e a t
tra n sfer); la ter, it w a s c o n fir m e d b y a n u m b e r o f o th e r in v e stig a to r s o f w h o m w e m u st
m e n t io n , L in, D e n to n e t al. ( 1 9 5 1 ) w h o c a re fu lly m e a su r ed th e rate o f m ass transfer in
e le c tr o c h e m ic a l r e a c tio n s (w ith Prandtl n u m b ers v a ry in g fr o m 300 to 3 0 0 0 ) . A t th e
b e g in n in g o f th e 1 9 5 0 ’s, th e a s su m p tio n th a t m = 3 o n c e again b eg a n t o b e w id e ly u se d in
th e o r e tic a l w o r k s [e .g ., R e ich a rd t ( 1 9 5 1 a ) , an d l i n , M o u lto n and P u tn a m in th eir stu d y o f
tu r b u le n t m ass tra n sfer ( 1 9 3 3 ) ] ; th e sa m e a s su m p tio n w a s u sed in th e w o r k s o n h e a t and
m ass tra n sfer b y P e tu k h o v an d K irillov ( 1 9 5 8 ) , W asan an d W ilke ( 1 9 6 4 ) , D il’m a n ( 1 9 6 7 )
and s o m e o th er s. N e v er th ele ss, R a n n ie ( 1 9 5 6 ) u se d s u c c e s sfu lly th e rela tio n sh ip s m = 2 and
~ (P r )~ ^ 2 for a fairly g o o d d e s c r ip tio n o f th e e x p e r im e n ta l d a ta o n h e a t tra nsfer w ith
P r < 120» a n d K ish in ev sk y ( 1 9 6 5 ) m a d e an a tt e m p t t o s h o w th a t th e s e rela tio n sh ip s can b e
u s e d e v en fo r th e d e s c r ip tio n o f m ass-tra n sfer d a ta in a co n sid e ra b ly w id e r range o f Pr
v a lu e s ; F ra n k -K a m e n e tsk iy (1 9 6 7 ) [in th e second e d itio n of his book of 1947]
a ssu m ed e v e n th a t th e e q u a t io n 'x. (Pr)"^ [w h ic h c o r r e sp o n d s to th e m o d e l o f a p u rely
lam inar su b la y e r w ith K = = 0; c f. E qs. ( 5 . 8 4 ) and ( 5 . 8 5 ) ] can b e a p p lie d fo r Pr o f th e
order 10^ (h o w e v e r , h e u se d d a ta fo r o n ly th re e va lu es o f th e Prandtl n u m b e r as a basis for
his a s su m p tio n ). A t th e sa m e tim e , m a n y o th er s [fo r e x a m p le , L ev ich ( 1 9 4 4 ; 1 9 5 1 ; 1 9 5 9 );
D e issle r ( 1 9 5 5 ; 1 9 5 9 ) ; L o its y a n s k iy ( 1 9 6 0 ) ; S o n a n d H anratty ( 1 9 6 7 ) ] are in c lin e d to th e
p r o p o s itio n th a t m = 4 a n d , c o n s e q u e n tly ,
C ft~ (P r) ^ for P r > > l : (5 .8 6 " )

th ey s h o w e d th at th is r esu lt a lso g ives a c o m p le te ly sa tis fa c to r y d e s c r ip tio n o f m a n y e x istin g

d ata o n tu r b u le n t h e a t and m ass transfer w ith va lues o f Pr up to 4 0 0 0 . We h a v e already
n o t e d in S ect. 5 .3 th a t in w o r k s b y S p a ld in g ( 1 9 6 1 ; 1 9 6 4 ; and o th e r s ) t w o d iffe r e n t
e q u a tio n s w e re p r o p o s e d : o n e w ith m = 3 and th e s e c o n d w ith m = 4 . In th e o ld litera tu re
o n th e s u b je c t, it is also p o ssib le to fin d e q u a tio n s c o rr e sp o n d in g to m = 5 [R e ic h a r d t
( 1 9 5 1 b ) ] a n d ev en to m = 6 [H o fm a n n ( 1 9 3 7 ) ] .
T h e very great n u m b e r o f c o n tr a d ic to r y th e o r e tic a l e q u a tio n s all h av in g see m in g ly
sa tis fa c to r y e x p e r im e n ta l v e r ific a tio n m a y b e e x p la in e d in several w a y s. First, e a ch p r o p o se d
e q u a tio n c o n ta in s a d d itio n a l unknow n param eters se le c te d so as to o b ta in th e b e s t
a g r ee m e n t w ith e x p e r im e n ta l d a ta u sed ; at th e sam e tim e th e sca tte r o f all th e e x istin g d a ta
is q u ite g reat b e c a u s e m a n y fa c to r s m a k e a c cu ra te e x p e r im e n ts o n h e a t and m ass transfer
rather d iffic u lt. S e c o n d , m o s t o f th e researchers h a v e u se d d a ta for a co m p a r a tiv e ly
r estr ic te d range o f Prandtl n u m b e r s an d a p p ly d iverse d a ta p r o c essin g te c h n iq u e s w h ic h in
s o m e ca ses ha ve o b v io u s d e fe c ts . F in a lly , th e e x p o n e n t s 2 / 3 and 3 / 4 in th e m a in c o m p e tin g
e q u a tio n s ( 5 .8 6 ') and (5 .8 6 " ) d iffe r b y o n ly 1/12, w h ic h b e in g very sm a ll, is d if fic u lt to
d e te c t reliably in la b o r a to r y e x p e r im e n ts . T h u s it is n o t surprising th a t th e sa m e d a ta w e re
o f t e n u sed b y d if fe r e n t researchers to s u p p o r t various th e o r e tic a l e q u a tio n s . It is clear th a t
th e fin a l s o lu tio n o f th e p r o b le m u rg en tly req uires e x tr e m e ly carefu l e x te n s iv e m ea su r e
m e n ts u n d e r fu lly c o n tr o lle d c o n d it io n s and w ith a w id e range o f th e va lues o f Pr,
p e r m ittin g u n ifo r m an d p r ecise s ta tistic a l tr e a tm e n t.
A sp e cia l e le c tr o c h e m ic a l te c h n iq u e h a s b e e n d e v e lo p e d b y H a n ra tty a n d h is c o -w o r k e rs
at th e U n iversity o f Illinois t o m ea su re th e various tu r b u le n c e c h a r a cter istic s very c lo se t o a
w a ll [s e e , e .g ., M itch ell a n d H a nratty ( 1 9 6 6 ) ; a nd H anratty ( 1 9 6 7 ) , w h e r e a d d itio n a l
refer e n c e s can a lso b e f o u n d ] . T h e te c h n iq u e w a s u se d in particular to m ea su re th e rate o f
m ass tra nsfer fro m a w a ll t o flu id . A ll th e resu lts o b ta in e d in th e range 4 0 0 - 2 0 0 0 o f Prandtl
(or, m o r e p r e cise ly , S c h m id t) n u m b e r s w e re d e scr ib e d b y S o n a n d H an ratty ( 1 9 6 7 ) w ith fair
a c cu ra c y b y th e e q u a tio n = K,^/v = 0 . 0 0 0 3 2 z \ (w h ic h c o r r e sp o n d s t o th e a s su m p tio n
th a t m = 4 ) . H o w ev er, th e m o s t e x te n s iv e an d a c cu ra te h o m o g e n e o u s d a ta o n m ass transfer
o ver a very w id e range o f S c h m id t n u m b e r s can b e fo u n d in th e th ree d isser ta tio n s b y H am il
t o n ( 1 9 6 3 ) [s e e a lso, H arriott an d H a m ilto n ( 1 9 6 5 ) ; H ub bard ( 1 9 6 4 ) ; H ub bard and L ig h tfo o t
( 1 9 6 6 ) ; K ader ( 1 9 6 9 ) ; and G u k h m a n a n d Kader ( 1 9 6 9 ) ] . In th e s e c o n d o f th e s e w o r k s
m ass tra nsfer w a s s tu d ie d via d iffu s io n -c o n tr o lle d r e d u c tio n o f sp e c ific io n s at a c a th o d e
d u rin g an e le c tr o c h e m ic a l r e a c tio n , i.e ., an e le c tr o c h e m ic a l te c h n iq u e w a s u s e d w h ic h w a s
d e v e lo p e d b y L in, D e n to n e t al. ( 1 9 5 1 ) and la ter w a s im p ro v e d c o n s id e r a b ly b y H an ra tty e t
al. T h e m e a su r em en ts w e r e m a d e a t S c h m id t n u m b e r s fr o m 1 7 0 0 t o 3 0 , 0 0 0 and th e results
o b ta in e d a greed v ery w e ll w ith E q. ( 5 . 8 6 ) fo r m = 3 [i.e ., w ith E q. ( 5 .8 6 ) ] . In th e w o r k s o f
H a m ilto n a n d o f K ader, a d if fe r e n t m e t h o d w as u se d , n a m e ly , m e a s u r e m e n t o f th e rate o f
s o lu tio n o f sp e cia l s o lu b le tu b e s e c tio n s m a d e fr o m b e n z o ic a cid or r ela te d su b sta n c e s w h e n
w a ter or a a g ly c e rin e-w a te r s o lu tio n w a s c ir cu la te d in th e tu b e ; H a m ilto n ’s o b se rv a tio n s
w ere m a d e a t S c h m id t n u m b ers fr o m 4 3 0 to 1 0 ^ , a n d K a d e r ’s o b s e r v a tio n s , a t S c h m id t
n u m b ers fr o m 5 0 0 t o 1 0 ^ . T h e resu lts o f b o t h a u th o r s are in o b v io u s c o n tr a d ic tio n w ith E q.
(5.86^^) and agree m u c h m o r e c lo s e ly w ith E q. ( 5 .8 6 ') .
F o llo w in g H ub b a rd ( 1 9 6 4 ) a nd H a m ilto n U 9 6 3 ) , D o n o v a n , H an n a , and Y e ra z u n is ( 1 9 6 7 )
c o n c lu d e d th a t m = 3 m u st b e ta k en in E q. ( 5 .8 6 ) . A d e ta ile d s ta tistic a l tr e a tm e n t o f all e x is t
ing d a ta o n th e su b je c t w a s m a d e b y Kader ( 1 9 6 6 ; 1 9 6 9 ) and G u k h m a n and K ader ( 1 9 6 9 ) .
T h e se au th o rs h ave d e te r m in e d th e n o n lin e a r least-sq u a res e stim a to r s o f th e t w o p a ram eters
B \ and m in E q. ( 5 . 8 6 ) [i.e ., th e va lues o f B and m m in im iz in g th e su m o f th e squares o f th e
d iffe r e n c e s b e t w e e n th e right a n d th e m id d le o f E q. ( 5 . 8 6 ) ] a t ev ery p a rticular va lu e o f
R e y n o ld s n u m b e r (or p r e cise ly ov er m a n y sm all intervals o f R e-v a lu es) u se d in th e d if fe r e n t
e x p e r im e n ts . T h e results o f su c h a tr e a tm e n t are c o lle c t e d in F ig . la and b (in F ig. la fo r th e
th re e m o s t e x te n s iv e stu d ies c ite d and in F ig. Ib fo r th e 26 d iffe r e n t w o r k s in w h ic h th e
nu m erica l d a ta o n h e a t a nd m ass transfer at va rious Pr a n d R e can b e fo u n d ). We see th a t
th e results o b ta in e d d e fin ite ly su p p o r t th e a ssu m p tio n th a t m = 3 .
346 ST A T IS TIC A L F L U ID M ECHANICS
c
— • — eg o
t3 <d
♦ 1 <2 8
A
• n
-1 ■? C
2
A E oi
1 V
-
-a 2
S o
? 6
cd
A ^ "o
1
r •-i
11
r~ •-n i l
A
9 *
S3
73 O
S 42
3 3
« 2
00 CNJ n
c ^
il
8-g
vp
o^ O
■§■1
!l
S^
8 ^j2
« § S
I I ^
^11
lo j
•S 3 S
^ -S &
'B ^ ^
C M
4> 5
•§<§*=
(UVO
o .g
3a 3 e
B ô ô
O) ^-M C
(U
i5 i - s
u- cIj o
CM
E q u a tio n ( 5 . 8 6 ) w ith m = 3 and = 1 2 .5 is an a s y m p to t ic versio n ( f o r P r » 1) o f th e

m o r e a ccu ra te e q u a tio n ( 5 . 8 0 ' ) w ith CgCPr) = 1 2 . 5 + (ln P r ) /K a - 5 .0 - In 2 / 2 x 0 .
(T h e last e q u a tio n c o r r e sp o n d s t o a sim p le tw o -la y e r m o d e l and th e c o n s ta n t term in the
e x p r e s s io n fo r Cg is c h o s e n in a c co r d a n ce w ith th e d ata fo r Pr = 0 .7 o n w h ic h th e
c o n tin u o u s curve in F ig. A is b a s e d ). T h e r esu lts o f th e c o m p u ta tio n s u sing E q. ( 5 . 8 0 ' ) w ith
su c h C ^ iP r ) (c o r r e c te d b y th e c o n s id e r a tio n o f th e tr a n sfo r m a tio n fro m th e c o e f f ic ie n t
to th e c o e f f ic ie n t ; i.e ., fo r th e c h a n g e fr o m E q. ( 5 . 8 0 ) to Eq. ( 5 . 8 0 ^ \ w h ic h is o f so m e
im p o r ta n c e at Pr < 1 0 o n ly , a nd p r e se n ted in th e fo r m o f th e d e p e n d e n c e o f N u = C;, Re^, Pr
o n R e ^ ) are s h o w n in Fig. B fo r an e x te n s iv e range o f v a lu e s o f Prandtl n u m b e r and R e y n o ld s
n u m b e r in c o m p a r iso n w ith n u m e r o u s d a ta o n h e a t a n d m ass transfer in tu rb u le n t tu b e flo w s
ta k en from 3 2 d if fe r e n t s o u r ce s. T h e very g o o d a g r e e m e n t b e tw e e n th e c a lc u la tio n s and
th e d a ta d e m o n str a te s s im u lta n e o u s ly th e v a lid ity o f th e a s y m p to t ic e q u a t io n ( 5 . 8 6 ) w ith
= 1 2 .5 a n d m = 3.
W hen th e a s su m p tio n m = 3 is u se d , th e least-sq u a res e s tim a te o f th e th e c o n s ta n t B^ in
E q . ( 5 . 8 6 ) fo r all th e e x is tin g d a ta a c co r d in g to K ader is B^ = 1 2 .5 . T his p relim in ary
e stim a te o f B^ c o rr e sp o n d s to = 0 .0 0 1 a c co r d in g to th e e q u a t io n rela tin g B^ to a^. We
see th a t th e value o f o b ta in e d c o in c id e s p r e cise ly w ith th e e stim a te o f p ra ctica lly the
sa m e c o e f f ic ie n t g iv e n o n an a b s o lu te ly d if fe r e n t basis at th e e n d o f S e c t. 5 .3 . T his
c o in c id e n c e e v id e n tly enlarges th e lik e lih o o d o f b o t h e stim a te s .
O f c o u r se , all th e a b o v e c o n s id e r a tio n s rela te o n ly to the ca se o f h e a t and m ass transfer
c lo s e t o a s m o o th w a ll. In th e case o f th e h e a t (o r m a ss) transfer c o e f f ic ie n t (i.e ., the
S ta n to n n u m b e r ) = N u /R e Pr req uires th e e v a lu a tio n o f th e d iffe r e n c e to w h ic h
th e ch a n g e o f d in t h e lo w e r la y e r ( o f th ic k n e ss c o m p a r a b le w ith th e m e a n h e ig h t Hq o f th e
irregu larities o f th e w a ll) m a k es a co n sid e ra b le c o n tr ib u tio n . T o d e scr ib e th is c o n tr ib u tio n
O w e n and T h o m p s o n ( 1 9 6 3 ) p r o p o s e d to r ep re sen t th e ch a n g e o f i^( 2 )in th e lo w e s t r ou g h n ess
la y er (i.e ., in th e la y er o f th e d ir e c t in flu e n c e o f w all r o u g h n ess) o f a c o m p le te ly ro u g h flo w as
= kS^B
w h e r e B is a d im e n sio n le ss q u a n tity d e p e n d e n t o n ly o n r o u g h n ess R e y n o ld s n u m b e r h ^ u j v
a nd Prandtl (o r S c h m id t) num ber Pr fo r giv en fo rm and a r ra n gem en t o f th e w a ll
irregularities. T h e d iffe r e n c e ^ ( 2 ) - i9q for 2 in th e lo g a r ith m ic la yer m a y b e e stim a te d in
this ca se b y th e e q u a tio n
J(z) - - — In— xB
w h e r e 2 ^ is th e usual ro u g h n ess p a ra m eter. U sin g th e d a ta b y N u n n er ( 1 9 5 6 ) , D ip p re y [se e

D ip p re y and S ab ersky ( 1 9 6 3 ) ] , th eir o w n da ta and th a t o f o th e r s , O w e n and T h o m p s o n
o b ta in e d th e em p irica l r ela tio n
w h e r e th e v a lues o f th e c o n s ta n ts c , k, and n, w h e n is r ep la c ed b y th e e q u iv a le n t sand

ro u g h n ess are a p p r o x im a te ly 0 .5 , 0 .4 5 , and 0 .8 , r e sp e c tiv e ly . A d d itio n a l in fo r m a tio n
o n th e values o f th e fu n c tio n B can b e fo u n d in th e w o r k s o f C ham b erla in ( 1 9 6 6 ; 1 9 6 8 ) ,
w h o h as d o n e specia l w in d -tu n n e l e x p e r im e n ts a n d h a s also c o lle c t e d d a ta ( s o m e o f it his
o w n ) o n m ass transfer u n d e r fie ld c o n d it io n s .
5.8 Free Turbulence

So far, we have considered turbulent flows in channels, tubes and
boundary layers, i.e., flows close to rigid walls, the friction on which
o
O
leads to the continuous generation o f vorticity and exerts a consider

able effect on the whole flow. However, in nature and in technology,
we often encounter turbulent flows of a completely different kind,
in which there is no direct effect of any rigid walls, and which are
therefore called free turbulent flows. The most im portant kinds of
free turbulent flows are turbulent wakes behind rigid bodies placed
in a constant velocity flow (or moving through fluid at rest),
turbulent jets, and turbulent mixing layers, occurring on the
boundary between streams of different velocities w ithout any rigid
walls between them.
Free turbulent flows play a large role in many natural phenomena
and in a number o f engineering devices. Thus it is not surprising that
a considerable number of theoretical and experimental investigations
has been devoted to their study. These investigations have led to a
number of results that are useful for practical calculations but which
have added little to our understanding of the physical nature of
turbulence. As a result, we shall limit ourselves here to a brief
investigation of some general deductions relating to the basic types
of free turbulent flows mentioned above.
The known methods o f describing free turbulence are based, first,
upon certain hypotheses of self-preservation of corresponding flows,
and, second, on the use of more special semiempirical hypotheses.
The self-preservation hypotheses may usually be justified with the
aid of general similarity and dimensional arguments; consideration of
these is of definite physical interest. However, the semiempirical
theories of free turbulence use, in addition to general laws of physics,
some additional hypotheses of a more speculative character; the
deductions thus obtained are important, primarily, for practical
applications. Henceforth, we shall confine ourselves in this section
almost exclusively to the analysis o f the self-preservation hypotheses
for free turbulent flows.
We shall begin by attempting to explain the general idea of the
self-preservation of free turbulence using the special example of a
three-dimensional turbulent jet issuing in the Ox direction from the
end of a thin tube of diameter D (but of arbitrary section) into an
infinite space filled with the same fluid. Let us compare the fluid
dynamical variables at different sections x = const. The observations
show that, in general, a turbulent jet may be considered to be
comparatively narrow. Thus the longitudinal component of the mean
velocity in a jet considerably exceeds the transverse component,
while the longitudinal variations of the statistical characteristics of
the fluid dynamical variables prove to be much smaller than their

transverse variations. On this basis it is natural to expect that the
properties of the flow field in a given section o f the jet will
reproduce to some extent the corresponding properties in sections
upstream. In other words, it must be expected that the statistical
characteristics of the fluid dynamical variables of a jet in different
sections of it (i.e., with different values of x) will be similar to each
other. This means that in every section x = const, we may introduce
characteristic length and velocity scales L(x) and U(x), respectively,
such that the dimensionless statistical characteristics obtained by
using these scales will be identical in all sections. Of course, this
similarity will not occur immediately after the orifice from which the
jet is issuing; it will become probable only at sufficiently large
distances (in comparison with the diameter D) from this orifice,
where its dimensions and form cease to influence the flow in the jet.
Moreover, for similarity to occur it is also necessary that the
Reynolds number Re = in the section of the jet under
consideration be sufficiently large to ensure the presence o f fully
developed turbulence (and likewise the viscosity-independence o f the
statistical regime of the large-scale components of turbulence which
cause interaction between the flow fields in neighboring sections of
the jet, which is the physical basis of the similarity of these fields).
This postulate of similarity is yet another apphcation of the
general principle of Reynolds number similarity discussed in Sect.
5.5. It may also be derived with the aid of simple considerations
related to those used in Sect. 5.3 in the deduction of the logarithmic
equation for the mean velocity profile close to a wall (but beyond
the edge o f the viscous sublayer). In fact, in the present case of a
three-dimensional submerged jet, the fiow depends on the orifice
diameter D, the initial velocity o f efflux o f the jet Uo and the fiuid
parameters v and p. Thus the statistical characteristics of th e flow,
for example, the mean velocity u or the Reynolds stress pu'w' (where
w' is the radial fluctuating velocity in cylindrical coordinates (r, <p, x)
with axis Ox) by dimensional considerations must be given by
equations of the form
(5 87)
— p a 'w ' = p t / o g i ( - ^ . tp, R e o ) , where R eo = - ^ ^ .
At a sufficiently great distance from the orifice, the dependence of

these characteristics on xjD practically ceases to exist. This means
that when xjD is sufficiently large, we may simply write x!D = oo in
Eq. (5 .8 7); it is a corollary of the postulate that the right side of
Eq. (5 .8 7 ) tends to a limit asjc/D->-oo. However, it is natural to
suppose that the “initial conditions” Uo and D for sufficiently large
x/Dwill, in general, be obtained only through the value of the “initial
m om entum” of the jet which determines the integral characteristics
of the flow, slowly varying with increase of x, at the section x =
const [i.e., the characteristic velocity Ujaax=U{x) on the axis o f the
jet and the half-width of the jet L —L(x) which is equal to the value r
For which u (r, 0, x ) = y ^max (-«)1; the whole statistical flow in
the neighborhood of the plane x = const wUl in some way be
adjusted to the values of the scales U and L.. It follows, therefore,
that for large values of x/D, the jet may now be considered as
axisymmetric since the q>-dependence can be determined only by the
form of the outlet orifice) and Eq. (5.87) in this case may be written
as
Re, (5.88)
• (•«)
where
Re
Finally, for sufficiently large Re, we may put Re* = oo in Eq. (5.88)
since for a developed turbulent regime the eddy viscosity always
considerably exceeds the molecular, and the parameter v will not
play any part; this is precisely the principle o f Reynolds number
similarity. Consequently, for sufficiently large values of x/D and Re;^
u Û {x)f L (x) LL( a:) (5.89)
where f{r) and g{r) are universal functions. These relationships

express the hypothesis of the self-preservation of a flow in a jet.
Further, we note that unlike turbulent flows in the presence of rigid

walls, self-preservation will occur in this case for the entire region of
turbulent flow. (But, as in the case of “wall” turbulence, it relates
only to large-scale components of motion unconnected with the
viscous dissipation of energy.)
Important connections between the functions L{x),U (.jc), f (r)and
g(r) may be obtained by substituting Eq. (5.89) into the Reynolds
equations. In the case of steady axisymmetric motion, and neglecting
terms containing the viscosity, the Reynolds equations may easily be
shown to reduce to a single equation
= (5.90)
[see, for example, Townsend (1956)]. The term ~ (a'° — here

describes the normal turbulent stress and the action of the
longitudinal pressure gradient; usually it proves to be much smaller
than the other terms (in any case, in the central part of the flow) and
very frequently is simply neglected. Consequently, we obtain
-~ du
ou . — uu
du 1I c)
CJ y —
which is the basis of almost all theoretical investigations relating to

turbulent jets. With the aid of the continuity equation for the
averaged motion
= “ <5.91)
the radial velocity w may be ehminated from Eq. (5.90'). However,

we must note first, that by multiplying Eq. (5.90) by r and
integrating with respect to r from 0 to c», we obtain,
OO CO
d _2 /»_2
/ u r d r = Q, i.e., j u r dr = M = const. (5.92)
dx
0
(This equation expresses the law of conservation of momentum, i.e.,

the equality of the momentum flux through any sufficiently large
sphere with the center dose to the outlet orifice). Substituting now
from the first of Eqs. (5.89), we find that
f /^ri)dri= const.
Thus it follows that UL (and, consequently, also, the Reynolds

number Re* = UL/v) cannot be dependent on x. Thus l/L.
Considering this result and substituting Eq. (5.89) into Eqs. (5.90')
and (5.91), it is easy to show that these equations cannot be valid if
- ^ 7 ^ const; the same result, of course, follows from the more
general equations (5.90).
Thus, we find that self-preservation is possible only when
L (x)^x-x0 , U ^(x-X o )-\ (5.93)
where Xo is the constant of integration. Here, x = Xq plays the part of

a virtual “origin o f the je t.” Since we are interested only in
asymptotic laws for very large values of x, here and below we may
simply assume that Xo= 0. According to Eq. (5.93), at sufficiently
great distances from the orifice, the breadth of the jet increases in
proportion to the distance, that is, the jet takes the form of a right
circular cone, while the mean velocity of flow in the jet decreases in
inverse proportion to the distance. We note further that the amount
oo
of fluid passing through the cross section of the jet is 2ic J u r d r , i.e.,
0
it increases in proportion to x. This indicates that the jet is

continuously taking in fluid from the surrounding, nonmoving,
medium.
Under the conditions (5.93), the system of equations
(5.90')—(5.91) possesses an infinite number of solutions o f the form
(5.89). In fact, this system allows us to obtain only a single
relationship connecting the two unknown functions /(ti) and g ( t ] )
which is found to be of the form
1
= ^ = co n s t
[see, for example, Squire (1948)]. The position is even worse if we

use the more general equation (5.90). Here we obtain only one
relationship containing four unknown functions [see Townsend
(1956)]. A complete determination of all the unknown functions
requires the introduction of still further hypotheses. In many works
various semiempirical theories of turbulence are used for this
purpose; see, for example, the classical papers of Tollmien (1926)
and Gortler (1942), and the books of Shih-i Pai (1954), Hinze
(1959), Abramovich (1963), and Vulis and Kashkarov (1965).
Of course, from the Reynolds equations it follows only that for
self-preservation (5.89) to occur, the scales U and L must satisfy Eq.
(5.93), but it does not follow that self-preservation must occur.
Self-preservation seems probable because of the general considera
tions given above (which, as we shall show below, may also be
obtained from standard dimensional reasoning); however, it may
definitely be established only on the basis of an analysis of the data.
For jets issuing into a space filled with fluid at rest, Eqs. (5.89) and
(5.93) have been verified experimentally by many investigators [see,
for example, Reichardt (1942); Corrsin (1943); Hinze and Van der
Hegge Zijnen (1949); Corrsin and Uberoi (1950); Corrsin and Kistler
(1954); Forstall and Gaylord (1955), and surveys by Hinze (1959);
Abramovich (1963); and Vulis and Kashkarov (1965)]. It was found
that even at fairly small distances from the orificc (x>8D), Eqs.
(5.89) and (5.93) give a good description of the profile of the mean
longitudinal velocity u{x, r). Complete self-preservation, however, of
the mean velocity profiles and of the second moments of the velocity
fluctuations is definitely established only at more distant cross
sections of the jet (x>50Z)). A dependence of the profile on the
Reynolds number Re = is also observed, which becomes weaker
as Re increases and for Re ^ 10^ ceases to be noticeable. For
example. Fig. 44 gives (according to Reichardt’s data ) the velocity
profile in three sections o f the jet, normalized by division by f/max =
U and related to dimensionless distances obtained with a length scale
L such that u{L) = ^ We see that the similarity of the profile is
fulfilled in a completely satisfactory manner.
Similar hypotheses on self-preservation may be formulated also for
other specific types o f free turbulent flows. However, we shall
consider here only a plane turbulent jet issuing into the space x > 0
FIG. 44. Dependence of the normalized mean velocity uj U on 7?=>'/Z, in three different
sections of an axisymmetric jet [according to Reichardt’s data (1942)].
(which is filled with the same fluid) in the Ox direction from an

infinitely long slit lying in the Oijz plane along the Oy axis; the
three-dimensional turbulent wake behind a finite rigid body placed
close to the origin of coordinates in a uniform stream in the Ox
direction; the plane turbulent wake behind an infinitely long cylinder
with its axis along the Oy direction, placed in a stream of fluid in the
Ox direction; finally, the mixing layer between two plane-parallel
flows in the Ox direction in the half-spaces 0 and z < 0 , having
constant but different velocities (say, Ui for z ^ > 0 and (Afor z < 0 )
in the initial section Oyz. In the case of the wakes behind rigid
bodies, it is natural to transform to a moving inertial system of
coordinates which moves with the undisturbed flow (i.e., to consider
only the deviations of the velocity in the wake from the undisturbed
velocity, that are damped at great distances from the body); however,
for the mixing layer, it is useful to select the coordinates in such a
manner that the condition b'2 = — Ui is satisfied. In this case, for all
the flows mentioned, it must be expected that for sufficiently large
Reynolds numbers the mean velocity profile and the turbulent stress
profile at sufficiently large distances x may be put in the form

(5.89), where r is the transverse coordinate (the distance from theOjc
axis for three-dimensional flows and from the plane 2 = 0 for plane
flows), and w is the transverse velocity. The scales of length L(x)and
velocity U{x) will always be proportional to some power of the
longitudinal distance from the virtual origin of x:; that is,
L ( x ) ^ x " ‘, U {x)-^x-" (5.94)
(so that the Reynolds number Re^ = - is proportional to

x'"“"). The indices m and n may be determined, as above, from the
requirements for the functions to be exact solutions of the Reynolds
equations (neglecting the molecular viscosity). In the case of wakes
behind obstacles, it is convenient to make a preliminary simplifica
tion of the Reynolds equations by using the fact that at large
distances from the body, the mean longitudinal velocity u within the
wake will be only slightly less than the constant velocity L'o o f the
flow (equal to the undisturbed velocity outside the wake). Thus u =
Uo — Ml where Uq= const, U\ <c Uo and the mean transverse velocity w
wUl be of an order not exceeding that o f Jii. Therefore, for example,
in the case of a three-dimensional wake, the Reynolds equation at a
large distance x may be written approximately as
dUi 1 d
= (5.95)
and, consequently, in this case
oo
J = A/=: const. (5.96)
Further considerations for all these types of free turbulent flow are
completely analogous to those for a free turbulent jet. The results
obtained are given in the table below. [See also, Squire (1948);
Schlichting (1960); Schubauer and Tchen (1959).] As seen from the
table, the most im portant difference between the types of free
turbulence we have mentioned consists in the types of dependence
on the distance x of the Reynolds number Re^ = ^ . In a
plane wake, as in a three-dimensional jet, Re does not vary with

distance; however, the flow in a plane wake differs from that in a
three-dimensional jet in the speed of self-adjustment [according to
Townsend (1956), complete self-preservation in a plane wake behind
a cylinder is established only at distances x > (500—1000)D, where D
is the cyhnder diam eter]. In a plane jet and in a mixing zone Re
increases with increase of x. Finally, in a three-dimensional wake, Re
decreases with distance, so that for sufficiently large x the turbulence
must decay; at such distances the assumption of self-preservation is,
of course, no longer applicable. For further discussion of such flows,
see, for example, Shih-i Pai (1954); Townsend (1956); Birkhoff and
Zarantonello (1957); Hinze (1959); Abramovich (1963); and Vulis
and Kashkarov (1965).
Three- Two- Three- Two-

Mixing
dimensional dimensional dimensional dimensional
layer
jet jet wake wake
m 1 1 1/3 1/2 1
n 1 1/2 2/3 1/2 0
m — n 0 1/2 -1 /3 0 1
Attention must now be given to yet another variant of free

turbulence, vertical turbulent jets of thermal origin, which occur due
to the action of buoyancy above a heated body. The postulates of
self-preservation of such jets were first considered by Zel’dovich
(1937). If, as above, the direction o f flow in the jet (i.e., the vertical
direction) is taken along the Ox axis, then these postulates consist in
the existence of the possibility, in every section of the j e t .«= const,
of introducing scales of length L{x), velocity [/(jc)and temperature
0(.r) such that the dimensionless mean characteristics of the flow
field obtained by the use of these scales will be identical in all
sections. In particular, Eqs. (5.89) and similar equations for the mean
temperature and the radial turbulent heat flux Cppw'T'
T = e(x)A . w ' r = -U {x)Q {x)g, L(x)

(5.97)
will apply (where w' is the radial fluctuating velocity in the

cylindrical coordinate system with axis Ox). Substituting Eqs. (5.89)
and (5.97) into the Reynolds equation obtained by averaging the

system of Boussinesq equations of Sect. 1.5, and neglecting terms
containing the molecular viscosity and thermal diffusivity (but not
terms describing the buoyancy!) we obtain the result [using a
deduction similar to that of Eq. (5.93)], that
U (x)^x-\ e (jc )~ x -V (5.98)
Similar considerations may be applied to a two-dimensional con

vective jet arising above a heated cylinder lying in the horizontal
plane x = 0 along the Oy axis. In this case, we obtain
L{x) — X, t/(x ) = const, 0 (x ) — x~^. (5.99)
(These equations were also obtained by Zel’dovich.)

We note further, that a large part of the results collected in the table
above, as well as Eqs. (5.98) and (5.99), may also be obtained simply
from certain dimensional considerations. For example, for a three-
dimensional jet, all the averaged characteristics at large distances x
may depend, in addition to the distance x, only on the density p and
on the total momentum 2npM of the mass of fluid issuing from the
orifice per unit time. Therefore, it follows that
= « . 100)
[which is in complete agreement with Eqs. (5.93)]. For a plane jet,

instead of 2npM, it is necessary to consider the momentum o f the
mass of fluid issuing per unit time from unit length of the slit,pAl, =
oo
p J u dz. Consequently, here

—oo
- AfV* / ^ \ --------- Ml ! z \
(in accordance with the second column of our table). Even more
simple is the application of dimensional reasoning to the plane
mixing zone; here, the conditions at the section x = const can
depend only on x and on the given difference of the velocities
— t /2 «= 2Ui = Ut). Thus, for example,
(5.102)
In the case of free convective jets, all the averaged characteristics

may depend on the total heat flux along the jet
Q = 2«Cpp j u'T'rdr,
0
or the heat flux per unit length of a heated cylinder
oo
Qi = f WV'dz
—oo
and the buoyancy parameter gp (for an ideal gas equal to glT^). Here
Q and Qi, naturally, may only occur in combinations Q/cpp and
Qi/cpp. Hence, it follows that
= (ir). ( i ) (5.103)
for a three-dimensional convective je t and
for a plane convective jet. As is clear, these results coincide with Eqs.
(5.98) and (5.99).
Only in the case of wakes is the position somewhat more
complicated, since, in addition to the total drag force acting on the
body
/■=2icpt/oN = 2upt/o J U i r d r
or the drag force per unit length of the cylinder
/=■, = pU qN^ = pUo J U jd z ,
which determines the “momentum defect” in the wake, in the

conditions of this problem there occurs a further dimensional
quantity, namely, the undisturbed velocity Uq. Thus in this case,
equations of the type (5.100)—(5.102) cannot be deduced on the
basis of dimensional considerations only. However, with the aid of
somewhat more special considerations, these may be obtained,
without necessitating recourse to the exact equations. Specifically,
we use the fact that the tangent of the angle of inclination of a
streamline of the averaged flow in the wake to the Ox axis is equal to
Therefore, we may conclude that if L{x) is the
u Uo — u, (Jo
characteristic breadth of the wake, and U =U {x)is the characteristic
scale of the velocities «i and w , then
(5.105)
But since
oo
= J u^rdr
0
is constant in the case o f a three-dimensional wake and

Of;
~ J dz
— oo
is constant in the case of a plane wake, it follows that UL"^ ~ N =const

in the first case, and U L ^ N i = const in the second. These results,
together with the relationship (5.105) show that
that is,
NUl
f (5.106)
for a three-dimensional wake and
that is,
/ (5.107)
for a plane wake.

Of course, the results concerning free convection jets must not be
confused with the analogous results for free flows with forced
convection, the transfer of a passive admixture t? (which may also be
heat) from a steady continuous source by free turbulent flows of
purely dynamic origin. In this case, the initial system of equations
will consist of the ordinary equations of fluid dynamics, without any
buoyancy terms, and the mass (or heat) transport equations (1.72).
The length and velocity scales will naturally be the same as for the
corresponding purely dynamic problem (i.e., they will be given by
Eqs. (5.94) with the values of the indices shown in the table), but
there is also an additional scale of concentration (or temperature)
0(x) .This new scale may be determined with the aid of an additional
Reynolds equation, obtained by averaging Eq. (1.72) with x = 0.
However, it is simpler to take as a starting point the fact that the
mass of the admixture (or the heat flux) passing through any section
X = const per unit time is constant. This immediately gives the
precise value of the index k in the equation which
depends on the type of the flow and the type of the source
(two-dimensional or three-dimensional). The corresponding results
will be discussed in detail in Chapt. 5 in connection with the study of
turbulent diffusion. Here we shall note merely that knowii^g the
distribution of the mean velocity u and the mean temperature d , and
the values of the turbulent fluxes u'w' and d'to', we can also
determine the values of the corresponding eddy diffusivities K and
Kd and find their ratio a = = Measurements permitting this
Ff/ A
to be established have been performed many times by different
investigators for different kinds of flows [see, in particular, the works
cited earlier in this subsection, primarily, Hinze (1959), Abramovich
(1963), and Vulis and Kashkarov (1965)]. On the whole, all these
results are in fairly good agreement with each other and show that to
a first approximation the eddy diffusivities K and /(» may be
assumed to take constant values independent of the molecular
coefficients v and x in almost every cross section of the jet or wake,
and that the ratio a = K^!K for free turbulence is greater than unity
(namely, a ~ 1.8—2 for a cylinder wake, and for a plane jet, and a ~
1.2—1.4 for an axisymmetrie jet). It was also found that the values of
/Co for heat and mass transport are exactly the same (but different
from the eddy viscosity K), so that Prf = Scj # 1 [cf. Kiser (1963)].
The independence of a and Pr^ = a"' on the molecular fluid
properties was clearly demonstrated in the experiments of Sakipov
and Tcmirbayev (1965) [see also Vulis and Kashkarov (1965)], who
measured K and /C» in free jets of technical oil, water and mercury
(i.e., at molecular Prandtl numbers in the range from 10* to 10'^!)
and obtained identical values of a and Pr^.
In conclusion, we must stress that all free turbulent flows possess
one very important feature, the region of space occupied by the
rotational turbulent flow will possess at every instant a clear but very
irregular boundary (strictly speaking, this is not a surface but a very
thin layer, the so-called “Corrsin’s superlayer” ) outside which the
motion is irrotational. As already explained in Sect. 2.2, fluid may be
drawn into the turbulent region from outside, but cannot escape
from it. Only large-scale turbulent velocity fluctuations can penetrate
into the region of irrotational motion; these are damped at distances
of the order of the transverse scales L{x) of the region of rotational
flow. These large-scale fluctuations cause the irregular form of the
interface between the turbulent and the irrotational motion.
As an illustration, Fig. 45 gives an excellent shadowgraph of the
turbulent wake behind a bullet in flight through the air, taken from
the work of Corrsin and Kistler (1954).
Thus, every free turbulent flow has, in the words of Townsend
(1956), a “ double structure” : it consists of intensive small-scale
turbulence and relatively less intensive large-scale vortices, which
FIG. 4 5 . Shadowgraph o f the turbulent wake behind a bullet in flight.
produce irregular distortion and motion of the turbulent-irrotational

interface. Due to this motion of the interface at any point of a free
turbulent flow, not too remote from its axis (or, in the case of a
two-dimensional flow, from its plane of symmetry), small-scale
turbulence will sometimes be present and sometimes be absent, i.e.,
the turbulence will be intermittent.** In Sect. 5.6, we have already
discussed the intermittency of the turbulence in the outer parts of a
turbulent boundary layer. In free turbulent flows, the zone of
intermittency of turbulence is, however, considerably wider than in
boundary layers; if in a boundary layer, the intermittent region is
observed to lie at a distance o f 0 .46-1.26 from the wall, then, for
example, in a plane turbulent wake, according to Corrsin and Kistler
(1954), it will be quite distinct in a zone from 2 = 0.4L(a:) [where z
is the transverse coordinate] at least to 2 = 3.2L{x). In addition to
the cases of free turbulence and the outer part of a turbulent
boundary layer (which is similar in many respects to free turbulent
flows), intermittency of turbulence and quite definite irregular
moving interfaces between the turbulent and nonturbulent regions of
the flow are also observed. For example, in tubes and boundary-layer
flows at the “ transitional” Reynolds numbers which favor the
formation of turbulent “ slugs” and “ spots” (see above. Sect. 2.1),
and also, in certain other cases [see, e.g., the survey of Coles
*®lt is often assumed that interm ittency will occur only b eyo n d som e d efinite distance
from the axis (or plane o f sym m etry) o f the flow. However, there are som e indications, e.g., in
the axisym m etric wake n ot too close to the b od y, that the irregular meanderings o f the wake
will generate considerable interm ittency even on the wake axis [see Hwang and Baldwin
(1 9 6 6 ); or Gibson, Chen and Lin ( 1 9 6 8 ) ] .
(1962)]. We may also expect that similar phenomena will, in general,

be characteristic of a wide range of turbulent motion, and will play a
considerable role in the initiation and development of turbulence.
However, so far, the theoretical investigation o f these phenomena has
met with considerable difficulties and only first steps in this
direction have been made (see, e.g.. Coles (1962); Liepmann (1962);
Townsend (1966); and Kovdsznay (1967), where additional refer
ences can also be found). We shall not dwell here further on these
questions; let us only note that the related subject of intermittency
of small-scale turbulence will be of importance for some discussions
in volume 2 of this book.
5.9 Semiempirical Theories of Turbulence
We ha ve already s h o w n in S e c t. 5 .1 th a t in tu r b u le n t flo w s th e la w s o f m e c h a n ic s are
e x p r e s s e d b y th e R e y n o ld s e q u a t io n s , in w h ic h th e n u m b e r o f u n k n o w n s e x c e e d s th e
num ber of e q u a tio n s . T h u s, th e R e y n o ld s e q u a t io n s o n ly a llo w us t o o b ta in so m e
c o n n e c t io n s b e t w e e n th e various ch aracter istic s o f tu r b u le n c e , an d t h e y c a n n o t b e
“ s o lv e d ” in th e usual s en se o f th e w o r d . T h e r e fo r e , in s e le c tin g s o lu tio n s o f the R e y n o ld s
e q u a tio n s w h ic h h a ve a p h y sic a l m e a n in g , cer ta in fu n c tio n s d escrib in g th e tu r b u le n c e m u st
n ecessarily b e given in d e p e n d e n tly o f th e s e e q u a tio n s . In certain cases, th e fo r m o f th ese
fu n c tio n s m a y b e fo u n d (a c cu rate t o a fairly sm all n u m b e r o f em p iric a lly d e te r m in e d
c o n s ta n ts ) o n th e basis o f d im e n sio n a l c o n s id e r a tio n s ; m o r e o f t e n , h o w e v e r , th e use o f
d im e n sio n a l c o n s id e r a tio n s o n ly le a d s to r e la tio n s h ip s w ith u n k n o w n fu n c tio n s th at m u st
th en b e fo u n d from e x p e r im e n t. In gen era l, th e n u m b e r o f th ese u n k n o w n fu n c tio n s n e e d e d
to d e scr ib e th e various tu r b u le n t flo w s e n c o u n t e r e d in nature or in en g in e e r in g a p p lic a tio n s ,
is very great. T h u s it is u n d e rsta n d a b le w h y m a n y in v e stig a to r s h a ve tried to red u c e th e
d e te r m in a tio n o f th e fu n c tio n s t o fin d in g a sm all n u m b e r o f sta tistica l c h a ra cteristics o f
tu r b u le n c e , w h ic h w o u ld b e ap p lic a b le s im u lta n e o u s ly to m a n y d if fe r e n t flo w s . T h e th eo r ie s
o f tu rb u le n c e w h ic h u se , in a d d itio n to th e strict e q u a tio n s o f flu id m e c h a n ic s , s o m e
a d d itio n a l rela tio n s fo u n d pu rely em p ir ic a lly fr o m e x p e r im e n t, or d e d u c e d w ith th e aid o f
q u a lita tiv e p h y sic a l c o n s id e r a tio n s an d th e n verified e x p e r im e n ta lly , are c a lled semiempirical
theories. O f c o u r se , fr o m th e v ie w p o in t o f “ p u r e ” th e o r e tic a l p h y sic s , all th ese th eo ries
m u st b e c o n s id e r e d as n o n r ig o r o u s , b u t th e y are q u ite ty p ica l o f in v e stig a tio n s in the
tu rb u le n t regim e. In th e d e v e lo p m e n t o f o u r u n d e rsta n d in g o f t u r b u le n c e , th e s em iem p irica l
th eo r ie s h ave p la y e d a c o n s id e r a b le part, and m a n y o f th em e v e n n o w are still w id e ly used in
t e c h n o lo g y . C o n s e q u e n tly , it w o u ld se e m u s e fu l h ere to g ive at lea st a b r ie f o u tlin e o f th e
fu n d a m e n ta l id eas o f th e m o s t im p o r t a n t sem iem p ir ic a l th e o r ie s p r o p o se d b y B o u ssin esq
( 1 8 7 7 ; 1 8 9 9 ) , Prandtl ( 1 9 2 5 ) , G. I. T a y lo r ( 1 9 1 5 ; 1 9 3 2 ) and v o n K drman ( 1 9 3 0 ) . Fu rth er
d e v e lo p m e n t o f th is a p p roa ch to th e th e o r y o f tu r b u le n c e and s o m e actu al e x a m p le s o f th e
u se o f s e m iem p ir ic a l th eo r ie s w ill b e c o n s id e r e d in th e fo llo w in g s e c tio n .
T h e s y s te m o f R e y n o ld s e q u a t io n s fo r th e m e a n values o f th e flo w variables is n o t c lo se d
d u e to th e p r e se n c e in th e m o f a d d itio n a l term s c o n ta in in g tu rb u le n t (R e y n o ld s ) stresses.
T h e r e fo r e , it is clear th a t th e s im p le s t m e a n s o f c lo su r e o f t h e s y ste m o f e q u a t io n s c o n sists
o f e sta b lish in g c o n n e c t io n s b e t w e e n th e m e a n f lo w variables, o n th e o n e h a n d , and th e
R e y n o ld s stresses, o n th e o th e r . T his p e r m its th e la tter va lu es t o b e e x p r e s s e d in term s o f
th e for m e r . B e lo w , w e shall e n u m e r a te several o f th e s e k in d s o f c o n n e c tio n s p r o p o se d b y
d if fe r e n t s c ie n tists a t d iffe r e n t tim es. We shall c o n s id e r o n ly th e s im p le s t plan e-parallel flo w s
in w h ic h o n ly th e x - c o m p o n e n t o f th e m ean v e lo c ity Ux = u is d iffe r e n t fr o m z e r o , w h ile
this d e p e n d s o n ly o n th e z c o o r d in a te , or, w h ic h is a lm o s t th e sam e, flo w s in a circular tu b e
in w h ic h th e m ean v e lo c ity u is d ir e c te d e v er y w h e r e a lo n g th e O x axis and d e p e n d s o n ly o n

the d ista n ce z = R - r from th e w all o f th e tu b e . In this c ase, th e R e y n o ld s e q u a t io n w ill
take th e form ( 5 . 1 6 ) [o r E q. (5 .1 6 ^ ) ]; h e n c e th e v e lo c ity p r o file w ill b e d e te r m in e d in this
case b y a single e q u a tio n [th e first e q u a tio n o f E q. ( 5 . 1 6 ) or E q. ( 5 .1 7 ^ ) ], c o n ta in in g a
single a d d itio n a l u n k n o w n x<'>=— p u ' w ' . C o n s e q u e n tly , to d e te r m in e th e fu n c tio n w (z), it is
o n ly n e cessa ry to b e a b le to e x p r ess x<'> in term s o f u { z ) .
T h e s im p le s t p o ssib le a s su m p tio n o n th e c o n n e c t io n b e t w e e n th e tu r b u le n t stress
— a nd th e p r o file u ( z ) is B o u s s in e s q ’s a s su m p tio n ( 1 8 7 7 ; 1 8 9 7 ) o n th e e x is t e n c e o f a
rela tio n sh ip o f th e fo r m ( 5 . 5 ) c o n ta in in g s o m e c o e f f ic ie n t o f e d d y v is c o s ity K . S trictly
sp ea k in g , E q. ( 5 . 5 ) is n o t an a s su m p tio n and d o e s n o t fo r m u la te a n e w c o n n e c tio n ; it o n ly
,(i)
rep laces th e u n k n o w n b y th e n e w u n k n o w n K = ------=—
p (du/dz). H o w e v e r , i f it is a ssu m ed
th at K d e p e n d s o n th e c o o r d in a te in s o m e d e fin ite m a n n er, th e n w e im m e d ia te ly arrive at

a sem iem p ir ic a l th e o r y b a se d o n a h y p o th e s is and c a p ab le o f e x p e r im e n ta l v e r ific a tio n .
T h e s im p le s t a s su m p tio n a b o u t th e va lu e o f K is th a t it is a c o n s ta n t. T his a s su m p tio n
has b e e n a p p lie d to free tu r b u le n t flo w s b y m a n y a u th o r s an d tu rn s o u t t o b e a r ea son ab le
first a p p r o x im a tio n fo r su c h flo w s ; h o w e v e r , as easily see n , w h e n a p p lie d to f lo w s in plan e
c h a n n els or circular tu b e s, it le a d s to c o m p le te ly in c o r r e c t results. In fa c t, w ith K = c o n s t,
th e e q u a tio n s for th e m e a n v e lo c ity o f tu r b u le n t flo w w o u ld d iffe r from th e co rr e sp o n d in g
eq u a tio n s for la m in ar f lo w o n ly in th e n u m e ric a l value o f th e v is c o s ity . C o n s e q u e n tly , it
w o u ld f o ll o w th a t th e tu b e or c h a n n e l v e lo c ity p r o file in this case w o u ld b e parabo lic.
H o w e v er , it is k n o w n th a t su c h a p r o file d o e s n o t c o r r e sp o n d to e x p e r im e n t (see F ig. 3 2 ,
Sect. 5 .5 ) .
T h ere is, o f c o u r se , n o t h in g rem ark able in th is c o n tr a d ic tio n ; it is w e ll k n o w n th a t as w e
ap p ro a c h th e w all th e c o e f f ic ie n t K m u st te n d to z er o (b u t n o t m o r e s lo w ly th an th e third
p o w e r o f th e d ista n ce fr o m th e w all; see S e c t. 5 .3 ) . T h u s th e a s su m p tio n th at K is c o n s ta n t
is q u ite in ap p lic a b le to th e case o f f lo w in th e p r e se n c e o f rigid w alls. H o w e v er , in certain
p r o b le m s , relatin g, fo r e x a m p le , t o tu r b u le n t je ts or w a k e s p r o p a g a ted in an in f in it e sp a c e,
or to tu rb u le n c e in th e fr e e a tm o sp h e r e , th e a s s u m p tio n K = c o n s t is n o t e n tir e ly invalid. In
th ese ca se s, s u b s titu tio n o f E q. ( 5 . 5 ) w ith c o n s ta n t K in th e R e y n o ld s e q u a tio n lea d s t o a
sem iem p ir ic a l e q u a tio n w h ic h c o n ta in s th e unknow n pa ra m eter K, w h ic h m u st be
d e te r m in e d from o b s e r v a tio n (w h ic h in various cases le a d s to e x tr e m e ly diverse values).
F o r flo w s b o u n d e d b y w a lls, it is im p o ss ib le to assum e th at K = c o n s t, b u t here in a
n u m b e r o f cases, w e m a y in tr o d u c e o th er r ea son ab le h y p o th e s e s o n th e c o e f f ic ie n t o f
tu rb u le n t v is c o s ity K w h ic h p e r m it us to m a k e th e R e y n o ld s e q u a tio n s d e te r m in a te . T h u s,
for e x a m p le , in th e case o f a plane-parallel flo w c lo se t o a plan e w all (b u t n o t t o o
c lo s e , b e y o n d th e lim it o f th e v is c o u s su b la y e r ) th e h y p o t h e s is th at K is p r o p o r tio n a l to th e
d ista n ce from th e w all lead s to g o o d results; fr o m it w e fin d th e lo g a r ith m ic e q u a t io n ( 5 . 2 2 ) ,
w h ic h agrees w ith th e d e d u c t io n s from d im e n sio n a l c o n s id e r a tio n s . F o r f lo w s in tu b e s,
c h a n n e ls and b o u n d a r y lay e r s, a c o n s id e r a b le n u m b e r o f h y p o th e s e s have b e e n p r o p o s e d o n
th e d e p e n d e n c e o f K o n th e d is ta n c e from th e w all z ; a s tw o sim p le e x a m p le s , w e m a y
m e n t io n th e n o t e o f G osse ( 1 9 6 1 ) , in w h ic h fairly g o o d a g r ee m e n t w ith e x p e r im e n t in tu b es
and b o u n d a r y layers is o b t a in e d b y th e a ssu m p tio n th a t /T is p r o p o r tio n a l t o th e fu n c tio n 1
- (1 - Tj)^ w h ere r\ = z / / / , and H i is th e radius o f th e tu b e or th e th ic k n e ss o f th e
b o u n d a r y layer, and th e w o r k o f S z a b le w sk i ( 1 9 6 8 ) , w h o o b t a in e d fair a g r ee m e n t fo r flo w s
in tu b e s an d pla n e c h a n n els w ith th e a s su m p tio n th a t K - K U ^ z \ / \ - r ] for 0 ^ t 7 < w
and K = K { m H x ) = c o n s t fo r m < 1 w here m ^ 0 .8 . In th e im m e d ia te p r o x im ity o f th e
w all th e in te r a c tio n o f th e p r o c esses o f m o le c u la r an d tu r b u le n t fr ic tio n lea d s t o a
d e p e n d e n c e o f th e fo r m / C « x u * z q ) ( 2 ii* /v ) ; s o m e sem iem p ir ic a l h y p o th e s e s o n th e form o f
th e f u n c t io n <p w e re p r o p o s e d b y D eissler ( 1 9 5 5 ) , R a n n ie ( 1 9 5 6 ) , L evich ( 1 9 6 2 ) and oth er s
(see a b o v e . S ec ts. 5 .3 a n d 5 . 7 ) . In th e surface la y e r o f th e a tm o sp h e r e , in th e case o f stable
th erm al str a tific a tio n , a c co r d in g t o all th e e x istin g d a ta as th e h e ig h t z in creases, th e
c o e f f ic ie n t first in crea ses a lm o s t lin early , and th e n its rate o f in crease b e c o m e s slow er.
w h ile fo r s u ffic ie n t ly large z , th e e d d y v isco sity m a y b e a ssu m ed p ractically c o n s ta n t. In this

c o n n e c t io n , several for m u las fo r K {z ) have b e e n p r o p o s e d , w h ic h agree w ith th e general
rules w h ic h w e have in d ic a te d [s e e , for e x a m p le , Y u d in and Sh vets ( 1 9 4 0 ) ; D o r o d n it s y n
( 1 9 4 1 ) , B erlyan d ( 1 9 4 7 ) 1 . E ach o f th ese e q u a tio n s c o r r e sp o n d s to a variant o f th e
sem iem p irical th eo r y . T h u s, a fter d e te r m in in g all th e p a r a m eters o c cu rrin g in th e c h o s c n
fo rm u la for K{z) from ob se rv a tio n a l data, w e can c a lc u la te , to s o m e degree o f a c cu ra c y , th e
variation o f th e w in d v e lo c ity w ith h e ig h t, an d s o lv e s o m e o th e r p r o b le m s o f m e te o r o lo g ic a l
in terest.
In th e in v e stig a tio n o f m ass and h e a t transfer in a tu rbu le n t flo w th e r o le o f th e
tu rb u le n t stress is p la y e d b y th e tu rb u le n t flu x o f m ass or h e a t </<*> = CppO'u;',
and th e r ole o f th e e d d y v is c o s ity K is p la y e d b y th e e d d y d iffu s iv ity d e fin e d b y Eq.
( 5 .9 ). F o r this c o e f f ic ie n t w e m a y r ep ea t all th a t has b e e n said ab o v e c o n c e r n in g K (s e e , in
particular. S e c t. 1 0 b e lo w , w h e r e w e shall discu ss a tm o sp h e r ic d iffu s io n and c o n sid e r certain
scm iem p irical th eo r ie s w h ic h u se a d e fin ite form o f th e d e p e n d e n c e o f o n z ).
O fte n , h o w e v e r , th e c o n c e p t o f the e d d y v isco sity (or d if fu s iv ity ) d o e s n o t h in g to
sim p lify th e tu rb u le n c e p r o b le m since th e s e le c tio n o f an app rop riate a s su m p tio n
c o n c e r n in g this q u a n tity is very d if fic u lt and it is n o t clear w ith w h ic h su ch c h o ic e o n e
sh o u ld w o r k . T o fa c ilita te this s e le c tio n , s o m e o th er sem iem p ir ic a l th eories have b e e n
fo r m u la te d . In m a n y o f th ese th eo r ie s, an e sse n tia l part is p la y e d b y th e '"mixing length"'^
in tro d u c ed by Prandtl ( 1 9 2 5 ) [th is id e a w as p u t forw ard in sligh tly d iffe r e n t form earlier by
G. I. T a ylor ( 1 9 1 5 ) 1 .
U se o f th e c o n c e p t o f the m ix in g le n g th is b e s t illu strated b y using th e e x a m p le o f the
tu rb u len t transfer o f a con ser v a tiv e passive a d m ix tu r e Let th e m ean c o n c e n tr a tio n
d e p e n d o n ly o n th e z c o o r d in a te . As a resu lt, w e are c o n c e r n e d m ain ly w ith th e transfer o f
this a d m ix tu r e a lo n g th e Oz axis. L et us assum e th at the transfer is e ffe c t e d o n ly by
tu rb u le n t flu c tu a tio n s o f v e lo c ity . C learly, this m a y o c c u r in su ch a w ay th at th ere arise in
the m e d iu m sm all ra n d om je ts w h ic h transfer th e a d m ix tu r e from o n e level z = c o n s t to
a n oth er. L et us assu m e th at every su ch j e t travels a lo n g th e Oz axis a d is ta n c e /', and th a t
o n ly after this is it m ix e d w ith th e su r rou n d in g m e d iu m ; in this case V w ill b e th e m ix in g
le n gth . T h e r e fo r e , in a c co r d a n ce w ith o u r a s su m p tio n , o n ly je ts p r o d u ce d at th e le vel z - V
and m o v in g up w ard s, or p r o d u c e d at z + /' and m o v in g d o w n w a r d s w ill im p in g e u p o n th e
level z. If w e n o w assum e th a t th e m ix in g o f th e j e t takes place n o t gradually b u t s u d d e n ly ,
then the je t p r o d u c e d at th e le v el z - V w ill transfer to th e level z its in itial c o n c e n tr a tio n
[e q u al, o n th e average to t?(z - / ' ) 1 w h ile th e je t p r o d u c e d at th e level z + /' w ill transfer a
c o n c e n tr a tio n e q u a l, o n th e average, to i5(z + /'). T h u s, arriving at the level z , th e je ts w ill
c o n ta in c o n c e n tr a tio n s o f th e a d m ix tu r e d iffe r e n t from the m e a n c o n c e n tr a tio n at th at
level, i.e ., this w ill lead to flu c tu a tio n s o f c o n c e n tr a tio n eq u a l to H (^ 4 : /') — d ( z ) ^ j.
/' • It is clear th at th e je ts m o v in g u p w ard s c o r r e sp o n d to a p ositiv e vertical
flu c tu a tin g v e lo c ity w ' and th o s e m o v in g d o w n w a r d s t o a n e g a tive value o f w ' . H e n c e , i f w e

tak e /' to have a p o s itiv e sign for je ts m o v in g u p w a r d , a n d a n e gativ e sign fo r je ts m o v in g
d o w n w a r d , th en
/'> = p 5 7 F = - p F 5 7 - ^ . (5 .1 0 8 )
T hus, as is easily see n , th e tu r b u le n t d if fu s io n c o e f f ic ie n t m a y b e e x p r e s s e d very sim p ly

in term s o f the m ix in g le n g th (w h ic h is, g e n erally sp e a k in g , a ran d o m q u a n tit y )
K^ = l ' w \ (5 .1 0 9 )
We m ay a tt e m p t to use E q. ( 5 . 1 0 8 ) n o t o n ly for c a lc u la tin g flu x e s o f h e a t or m aterial

a d m ix tu r e s, b u t also to d e scr ib e th e m o m e n tu m flu x . T his is th e basis o f P ran d tl’s
"'momentum transfer theory"' ( 1 9 2 5 ) w h ic h p r o p o se s th at for th e lo n g itu d in a l c o m p o n e n t

o f m o m e n tu m pw th ere also e x ists a d e fin ite m ix in g le n g th /', an d , c o n s e q u e n tly ,
du
= , K ^ l'w '. (5 .1 1 0 )
H ere, th e le n g th /' in s o m e sen se is a n a lo g o u s to th e le n g th o f th e m e a n free path o f the

m o le c u le s in th e k in e tic th e o r y o f gases; it d e fin e s th e path alo n g w h ic h so m e sm all flu id
e le m e n t (lu m p o f flu id ) passes b e fo r e it is m ix e d w ith th e r em ain in g flu id particles and
transm its its m o m e n tu m to th em . In this sen se , E q. ( 5 . 1 1 0 ) fo r K is a n a lo g o u s to the
e x p r e s s io n fo r th e c o e f f ic ie n t o f m o le c u la r v is c o s ity (w h ic h derives fr o m th e k in e tic th eo r y
o f gases) V ^ where is th e le n g th o f th e m e a n free p ath , and is th e v e lo c ity o f
th erm al m o t io n o f th e m o le c u le s . O f c o u r se /' is m a n y orders o f m a g n itu d e greater than \
th er e fo r e, in sp ite o f th e fa c t th a t u,n gen erally is greater th an w \ E q. ( 5 . 1 1 0 ) gives values
o f K th a t are c o n sid e ra b ly greater th an v.
We n o w n o t e th a t a lu m p reach in g th e la yer z fr o m th e layers z — V and e - h / ' w ill have
m e a n lo n g itu d in a l v e lo c itie s o f u { z — I') and u ( e - l - / ' ) , r es p e c tiv e ly . T hus, in b o t h cases
th e y w ill p r o d u c e a flu c t u a tio n u' in th e lo n g itu d in a l v e lo c ity w h ic h is c lo s e in a b s o lu te
valu e t o . T h e n , a ssu m in g th a t th e flu c t u a tio n o f th e vertical v e lo c ity w ' is c lo se in

dz
a b s o lu te value to th e flu c t u a tio n o f th e lo n g itu d in a l v e lo c ity u ' ,
Prandtl e x p r e s s e d th e tu r b u le n t stress t<‘> in th e fo r m
w h ere / is a n e w le n g th o f th e sam e order as th e r o o t m ean square o f th e r an d o m le n g th /' (/
is n o t , o f co u r se , sim p ly e q u al to ( / ' ^ ) 2 , sin c e w ' is o n ly o f th e sam e order as / ' , and
d o e s n o t have to b e e x a c t ly eq u a l to / ' fo r every lu m p ). T he a b s o lu te va lu e sign in Eq.
( 5 . 1 1 1 ) is c h o s e n so th a t th e stress a lw a y s has th e sam e sign as . T h u s the m o m e n tu m
is a lw a y s tr a n sm itte d from th e fa s ter-m ovin g layers o f th e flu id to th e s low er . T h e le n g th / in

E q. ( 5 . 1 1 1 ) , w h ic h is n o lo n g e r a r a n d o m q u a n tit y , usua lly is also c a lled th e m ix in g le n g th .
E q u a tio n ( 5 . 1 1 1 ) a llo w s th e e d d y v is c o s ity to b e w r itte n in th e form
(5 .1 1 2 )
w h ere / is a le n g th w h ic h d e p e n d s, generally sp e ak in g, o n th e c o o r d in a te s and ch aracterizes

th e scale o f tu rb u le n c e (th e m e a n d im e n sio n o f the tu r b u le n t flu c t u a tio n s ) at a given p o in t.
T h e n th e d e te r m in a tio n o f th e m e a n v e lo c ity p ro file req uires o n ly th e fix in g o f th e e x p lic it
d e p e n d e n c e o f I o n th e c o o r d in a te s . F o r f lo w s c lo se to an in fin ite p lan e w a ll w h ic h are
c h aracterized b y a c o n s ta n t value o f T, in th e reg ion b e y o n d th e v isco u s su b lay er, th ere is n o
le n g th scale. T h e re fo r e , every q u a n tity w ith d im e n sio n s o f le n g th m u st b e p r o p o r tio n a l to
th e d ista n ce from the w all. P u t t i n g / = x2: and x<'>= t = p u l = c o n s t, from E q. ( 5 . 1 1 1 )
we o b t a i n ^ / \ . q . , u = u J k In z-\- c o n s t [ s e e P r a n d t l ( 1 9 3 2 b ) ] .T h u s , again w e o b ta in
a lo g a r ith m ic la w fo r th e v e lo c ity p r o file , w h ic h , as w e ha v e s e e n , is c o n fir m e d very w e ll b y

e x p e r im e n t. F o r flo w s in a p la n e c h a n n e l or in a circular tu b e , as a first a p p r o x im a tio n w e
m a y e v e n a ssu m e th a t l = y,z r em ain s c o r r e c t right u p to th e m id d le o f th e c h a n n el or th e
axis o f th e tu b e . S u b stitu tin g this r ela tio n sh ip in t o E q. ( 5 . 1 1 1 ) and r ep lacin g th e le f t side b y
E q. ( 5 . 1 7 ) or (5.17^) [i.e ., o n c e again id e n t ify in g w ith t = -|-v , w e o b ta in a

d iffe re n tia l e q u a tio n w ith r esp e c t t o u ( z ) , w h ic h w ith an a p p rop ria te c h o ic e o f th e c o n s ta n t
X, lea d s t o a v e lo c ity p r o file w h ic h c o rr e sp o n d s fairly w e ll w ith e x p e r im e n ta l d ata fo r a
c o n sid e ra b le part o f th e c ross s e c tio n o f th e c h a n n el or tu b e [se e G o ld ste in ( 1 9 3 8 ) ] .
H o w e v er , in im m e d ia te p r o x im ity to th e c en te r o f th e c h a n n e l or tu b e , th is a p p r o x im a tio n
is clearly inappH cable; h ere it is c o n s id e r a b ly b e tt e r to tak e th e m ix in g le n g th to b e
a p p r o x im a te ly c o n s ta n t. T a k in g , for e x a m p le , / = for a tu b e o f radius R, th e n b y E qs.
( 5 . 1 7 ') and ( 5 . 1 1 1 ) , w e have
In tegratin g th is e q u a t io n , w it h th e b o u n d a r y c o n d it io n u (R) = Ut., w e o b ta in
i I,
— -------•
T h e la tter r esu lt, as w as s h o w n as far b ack as D arcy ( 1 8 5 8 ) , w ith p » 0 .1 3 is in g o o d
a g r ee m e n t w ith th e d a ta in th e reg io n 0 .2 5 < - ^ < 1 [se e E q. ( 5 . 4 6 ) , S e c t. 5 . 5 ] . F in a lly , to
o b ta in a g r ee m e n t w ith e x p e r im e n t fo r th e flo w in a circular tu b e for all values o f o u tsid e

th e lim its o f th e visco u s su b la y e r , w e m u st p u t
/ = /? [0 .1 4 -0 .0 8 (l _ ^ y _ 0 . 0 6 ( l — (5 .1 1 3 )
[N ik u r a d se ( 1 9 3 2 ) ] . T h e la tter e q u a t io n , for z/R 1, tak es th e form / = 0 . 4 z - 0 . 4 4
0 . 4 z , w h ile as z/R - > 1 it b e c o m e s / « 0 .1 4 /? , i.e., it m a y b e c o n s id e r e d as an
in te r p o la tio n fo r m u la b e tw e e n th e v a lu es for / c lo s e to th e w a ll and at th e c en te r. A n o th e r

e q u a tio n for / = /(z ) fo r f lo w in a tu b e (or c h a n n e l) w as p r o p o s e d b y S z a b le w s k i ( 1 9 6 8 ) . A
similar d istr ib u tio n o f I over th e cross s e c tio n o f th e tu b e w as p r o p o s e d also b y O b u k h o v
( 1 9 4 2 ) , w h o a t th e sam e tim e p u t forw ard a h y p o th e s is o n th e le n g th / p e r m ittin g its
d e p e n d e n c e o n th e c o o r d in a te s to b e fo u n d for tu b e s o f an y cross s e c tio n . In th e im m e d ia te
p r o x im ity o f th e w all (w ith in th e lim its o f th e v is c o u s su b la y e r ) / m u st d ecre a se o b v io u sly
w ith de c re ase o f z m o r e rapidly th an a c co r d in g t o th e lin ear law ; h y p o t h e s e s o n th e b e h av ior
o f / in th is r eg io n o f flo w hav e b e e n p r o p o s e d b y R o tta ( 1 9 5 0 ) , H am a ( 1 9 5 3 ) , van D riest
( 1 9 5 6 ) and o th er s. A th ree-layer d is tr ib u tio n o f / across a p lan e tu r b u le n t b o u n d a r y layer
w as str o n g ly r e c o m m e n d e d b y Sp ald in g ( 1 9 6 7 ) as b e in g surprisingly sa tisfa c to r y for m o s t
p ractical p u rp o se s; this d is tr ib u tio n is a c o m b in a tio n o f van D r ie s t’s d is tr ib u tio n c lo s e to the
w all, a lin ear d is tr ib u tio n / = kz (w it h k = 0 . 4 3 5 ) in th e lo g a r ith m ic la y e r b e lo w z = 0 . 0 9 6 / «
and a u n ifo r m d is tr ib u tio n / = 0 .0 9 6 fo r z > 0 . 0 9 6 /k .
L ater, Prandtl m ad e s o m e g e n e r a liz a tio n s o f th e m o m e n t u m transfer th e o r y th a t are
ap p lic a b le , fo r e x a m p le , t o th r e e -d im e n sio n a l f lo w s [c f. G o ld ste in ( 1 9 3 8 ) , S e c t. 8 1 ] , or
w h ic h ta k e in t o a c c o u n t th e p o s sib ility th a t fo r = 0 , th e tu r b u le n t v is c o s ity , in
c o n tr a d ic tio n t o E q. ( 5 . 1 1 2 ) is, n e v e r th e le ss n o n z e r o [P ran dtl ( 1 9 4 2 ) ] . H o w e v e r , w e shall

n o t d isc u ss th ese g e n e r a liz a tio n s in d e ta il b u t p r o c e e d im m e d ia te ly to T a y lo r ’s vorticity
transfer theory, th e s e c o n d fu n d a m e n ta l sem iem p ir ic a l th e o r y .
T h e fo r m u la tio n o f this th e o r y d erived fro m an a tt e m p t t o c o n sid e r th e e f f e c t o f

pressure flu c tu a tio n s o n m o v in g flu id p articles, w h ic h lead to variation in th eir m o m e n tu m ,
th ereb y p re v en tin g m o m e n tu m fr o m b e in g c o n s id e r e d as a c o n ser v a tiv e a d m ix tu r e as th e
e le m e n ts o f flu id are d isp la c e d . O n this basis, G. I. T a y lo r ( 1 9 1 5 ) , in tr o d u c e d fo r th e first tim e
the c o n c e p t o f th e “ m ix in g le n g t h .” H o w e v e r , u n lik e Prandtl, h e a ssu m ed th a t a “ m ix in g
le n g t h ” m u st e x is t fo r th e v o r tic ity b u t n o t fo r th e m o m e n tu m ; la ter, T a y lo r d e v e lo p e d this
idea in greater detail [s e e G. I. T ay lo r ( 1 9 3 2 ) ) .
T a y lo r ’s id e a is b e s t ju s tifie d fo r a tw o -d im e n s io n a l flo w in w h ic h , as is k n o w n , in the
ab se n c e o f v is c o s ity , th e v o r tic ity is transferred w it h o u t ch a n g e as th e flu id particles m o v e .
T h u s th e single c o m p o n e n t o f th e v o r tic ity

dil dw
= - ^ ------^ o f a tw o -d im e n s io n a l flo w w ith
v e lo c ity fie ld u = { u , w ] is a c o n ser v a tiv e q u a n tity e v er y w h e r e o u t s id e th e v is co u s su b layer.

T h e tu r b u le n t shear stress x — — p u ' w ' o f a tw o -d im e n s io n a l f lo w w ith m e a n v e lo c ity
a = { a , 0 ) e v er y w h e r e parallel t o th e Ox a x is, satisfie s th e e q u a t io n
1 dx d
Tdi = - -
here th e c o n tin u ity e q u a t io n for the v e lo c ity flu c t u a tio n s = 0 is u se d . In a
h o m o g e n e o u s flo w a lo n g th e Ox a xis, th e derivatives o f t h e averaged q u a n titie s w ith r e s p e c t

to X w ill b e e q u al to z er o ; c o n s e q u e n tly .
^ = (5 .1 1 4 )
___
I n tr o d u c in g th e m ix in g le n g th for th e v o r tic ity / ( , such th a t coy = /,'-^ , pw'<Oy =
p w 7 J - ^ ^ . A s in P ra n d tl’s th e o r y , w e ta k e w ' in th e fo r m w '= w h e r e /' is th e
r a n d o m m ix in g le n g th fo r th e v e lo c ity . In th is ca se , w e o b serve th a t Wy = ^
dû
an d , t h e r e f o r e , - p i - = - r - r ‘ F in a lly , w e o b ta in
az dz*
_I
where /,= (-/V j)* is th e c h a racteristic le n g th w h ic h p lays th e sam e part in T a y lo r ’s
t h e o r y as d o e s th e le n g th / in P r an d tl’s th e o r y . O f c o u r se , i f w e c o n sid e r the fa c t th a t
a ctu ally th e v e lo c ity d istu rb a n c es in a plan e-parallel flo w m a y a lso b e th re e -d im en sio n a l,
th en th e d e d u c t io n g iv e n for E q. ( 5 . 1 1 5 ) lo s e s its fo r c e , in this case, E q. ( 5 . 1 1 5 ) m u s t b e
regarded as an em p irical c o n n e c t io n o n ly s u sc e p tib le t o e x p e r im e n ta l v e rific a tio n . E q u a tio n
( 5 . 1 1 5 ) is also su sc e p tib le to fu rther g e n e ra liza tio n [s e e , fo r e x a m p le , G . I. T a y lo r ( 1 9 3 5 c ) ,
or G o ld ste in ( 1 9 3 8 ) , S e c ts. 8 4 - 8 5 ] ; h o w e v e r , w e shall n o t d w e ll o n th is p o in t.
F o r th e case / i = c o n s t, E q. ( 5 . 1 1 5 ) tu rn s o u t t o b e strictly e q u iv a le n t to P r a n d tl’s
e q u a t io n ( 5 . 1 1 1 ) . In fa c t, i f / i = c o n s t, th e n b y in teg r a tin g E q . ( 5 . 1 1 5 ) w ith r e s p e c t t o z w e
o b ta in t = -^ p/J • A ls o , th e last e q u a t io n agrees w ith P r an d tl’s e q u a t io n i f w e a ssu m e
th a t 11 = l \ / 2 . In g en eral, h o w e v e r , th e results o f T a y lo r ’s th e o r y d iffe r slig h tly fr o m th o s e

o f P r a n d tl’s th e o r y . W hen a p p lie d t o pressure f lo w s in plan e c h a n n els a nd circular tu b e s th e
T a y lo r th e o r y w ith = X |Z (w h e r e it is c o n v e n ie n t t o ta k e x i « 0 . 2 3 for a c h a n n el f lo w and

X i « 0 . 1 9 fo r a tu b e f lo w ) p e r m its o n e to o b ta in a fo r m o f th e m e a n v e lo c ity p r ofile w h ic h
gives fair a g r ee m e n t w ith th e e x is t in g d ata a lm o s t up t o th e m id d le o f th e c h a n n el or th e
axis o f th e tu b e [s e e , e .g ., G o ld s te in ( 1 9 3 8 ) , S e c ts. 1 5 6 - 1 5 7 ] .
U sin g an a r g u m e n t sim ilar to th e d e d u c t io n o f E q . ( 5 . 1 1 1 ) , and s u b s titu tin g
in to E q. ( 5 . 1 0 8 ) w e m a y d e d u c e th e e q u a tio n
= (5 -1 1 6 )
w h e r e h is y e t a n o th e r m ix in g le n g th , w h ic h n o w c h a r acterizes th e transfer o f a passive

c o n ser v a tiv e a d m ix tu r e 0 . T h e r e la tio n sh ip b e tw e e n th e le n g th s h and / (or / i ) d e te r m in e s
the value o f th e tu r b u le n t Prandtl n u m b e r Pr^ = (or th e value o f a = T h u s, for
e x a m p le , in a regio n o f flo w in w h ic h w e m a y p u t / = c o n s t, th e value k = l c o rr e sp o n d s to

P r ^ = a = 1, w h ile fo r k = l i w e o b ta in P r , = 1 /2 , a = 2. T h e o r e tic a lly , o f c o u r se , th e le n g th
k is n o t b o u n d to b e sim ilar to e ith e r / or / | , h e n c e it is im p o ss ib le to assu m e (as is
s o m e tim e s d o n e ) th at from th e m o m e n t u m transfer th e o r y fo ll o w s the value P r / = 1, an d
from th e v o r tic ity transfer th e o r y , th e valu e Pr^ = 1 /2 , in fa c t, th e r e la tio n sh ip b e tw e e n th e
le n g th s k , /, a n d h can o n ly b e s o u g h t o n th e basis o f e x p e r im e n t.
N e ith e r P ra n d tl’s m o m e n tu m transfer th e o r y n o r T a y lo r ’s v o r tic ity transfer th eo r y
c o m p le te ly s o lv e s th e p r o b le m o f th e c o n n e c t io n b e t w e e n th e R e y n o ld s stresses and th e
m ean v e lo c ity fie ld . This is b e c a u s e a n e w q u a n tit y , th e m ix in g le n g th , is in tr o d u c e d in th ese
th eo r ie s, and to d e te r m in e this at ev ery p o in t o f th e f lo w a d d itio n a l h y p o t h e s e s are required.
A n e x tr e m e ly g eneral h y p o th e s is w h ic h a llo w s us, in particular, to esta b h sh a general
c o n n e c t io n b e t w e e n th e le n g th / an d th e m e a n v e lo c ity fie ld w as p r o p o s e d b y v o n Kdrmdn
( 1 9 3 0 ) - r / z e hypothesis o f local kinematic similarity o f th e field o f tu r b u le n t v e lo c ity
flu c tu a tio n s. A c co r d in g t o this h y p o th e s is , th e field o f tu r b u le n t v e lo c ity flu c t u a tio n s in the
n e ig h b o r h o o d o f every p o in t o f a d e v e lo p e d tu r b u le n t f lo w (e x c lu d in g o n ly th e p o in ts o f
th e th in v is co u s s u b la y e r c lo se t o a rigid w a ll, in w h ic h th e a c tio n o f th e v is c o s ity is revealed
d ir e c tly ) are sim ilar to e a ch o th e r and d iffe r o n ly in th e le n g th and tim e scales, or, w h ic h is
th e sa m e th in g, le n g th and v e lo c ity . T o fo r m u la te this h y p o t h e s is m a t h e m a tic a lly , w e
in tr o d u c e in t o th e n e ig h b o r h o o d o f every p o in t o f_ A e f lo w Jfo a m o v in g c o o r d in a te s y s te m ,
m o v in g w ith a v e lo c ity eq u a l to th e m e a n v e lo c ity u (ô) at th e given p o in t; th e c o o r d in a te s
in this s y ste m w ill o b v io u sly agree w ith th e c o m p o n e n t s o f th e vQcioi x — — u (Xq) t .
A lso a c c o r d in g to this h y p o th e s is , a t every p o in t Xq m a y be d e fin e d a scale o f le n g th / (jc„)
and a scale o f v e lo c ity U {Xo) su c h th a t in th e n e ig h b o r h o o d o f this p o in t, th e field o f
tu rb u len t v e lo c ity flu c tu a tio n s w ill ta k e th e form
a ' ( j f ) = t /( j c „ ) ® ( 5 ) , (5 .1 1 7 )
w h ere the fu n c tio n V ( i ) i s universal, i.e ., in d e p e n d e n t o f th e p o in t Xq [an d c o n s e q u e n tly o f

th e m ean v e lo c ity fie ld ii( J C ) ]. T h u s, in v o n K drm dn’s th e o r y , th e c o n n e c tio n b e tw e e n th e
c h a ra cteristics o f tu rb u le n t flu c t u a tio n s and th e m e a n v e lo c ity fie ld w ill o c c u r o n ly b e c a u se
o f th e d e p e n d e n c e o f th e scales / an d U o n th e fie ld u{x).
T o d e te r m in e th e d e p e n d e n c e o f th e scales / and U o n th e m ean v e lo c ity o f flo w in a
p lane-parallel ste a d y flo w w ith m e a n v e lo c ity U(z) d ir e c te d e v er y w h e r e a lon g th e Ox axis
and d e p e n d e n t o n ly o n z , w e m ay use Eq. ( 1 . 7 ) for th e v o r tic ity , th e >»-com p onent o f w h ic h ,
ignorin g th e v is c o s ity , w ill tak e th e form
+ = (5.118)
S u b stitu tin g in to this e q u a t io n the value o f th e to ta l v e lo c ity u ^ u (Zq) I {Zo)v(\),

w here / is th e u n it v e c to r o f th e Ox a xis and
is th e v e cto r o f th e relative d im e n sio n le ss c o o r d in a te s (an d it is a s su m ed th a t jro=*yo = 0 ) , w e

m a y pass from Eq. ( 5 . 1 1 8 ) to an e q u a tio n in th e u n k n o w n s a = 1, 2, 3, w ith
c o e f fic ie n ts c o n ta in in g th e q u a n titie s U ( Z q ) , 1 ( zq) , U{ zo) and th eir deriv atives w ith resp e c t
to Zq. Sin ce acco r d in g to o u r h y p o th e s is (J) is a universal fu n c tio n , it c a n n o t b e
d e p e n d e n t o n z. T hus w e m a y o b ta in the relatio n sh ip s
= const. = const
t/{io) <''0 ‘‘A
[se e, fo r e x a m p le , G o ld ste in ( 1 9 3 8 ) , Sect. 1 5 8 , or S c h lic h tin g ( 1 9 6 0 ) , (Thapt. 1 9 , S e c t. 5 ,
w h ere th e c o rr e sp o n d in g d e d u c t io n is giv en in d e tail for th e special case o f a
tw o -d im e n s io n a l fie ld « ' ( j r ) ) . T h e re fo r e , it f o llo w s th at
s /# - <>■'’ )
w h ere x and B are universal c o n s ta n ts w h ic h , g en era lly sp eak in g, m a y b e g iv en any value,
sin ce th e scales / and U are d e fin e d o n ly up to an arbitrary m u ltip lier .
T h e first o f th e rela tio n sh ip s o f Eq. ( 5 . 1 1 9 ) m a y also b e d e d u c e d d ir e c tly from
d im en sio n a l c o n s id e r a tio n s i f it is s tip u la te d th at th e le n g th / d e p e n d s o n ly o n ^ and
T h e s e c o n d r ela tio n sh ip o f Eq. ( 5 . 1 1 9 ) is o b ta in e d in the sa m e m an n er from th e
a s su m p tio n th at U also d e p e n d s o n ly o n and - ^ (o r o n / and ^ ) .
T h e le n g th / is clearly a n a lo g o u s to th e m ix in g le n g th s / and l\ in b o t h P ra n d tl’s and

T a y lo r ’s th e o r y . In fac t, b y Eqs. ( 5 . 1 1 7 ) and ( 5 . 1 1 9 )
= — p a ' w ' = p6^^x;i(0)'i;3(0) = c o n s t • , (5 .1 2 0 )
w here c o n s t = B 2 y j ( 0 ) t ; 3 ( 0 ) . Sim ilarly,
— ( 5^121)
w h ere n o w th e c o n s ta n t has a d iffe r e n t value. O f c o u r se , i f w e rep lace th e c o n s ta n t by u n ity

in E q. ( 5 . 1 2 0 ) or E q. ( 5 . 1 2 1 ) , i.e ., i f w e id e n tify th e le n g th / w ith th e m ix in g len g th for
m o m e n tu m or v o r tic ity , th e n th e c o n s ta n t x in th e first e q u a t io n o f E q. ( 5 . 1 1 9 ) c a n n o t be
c h o s e n arbitrarily b u t m u st b e d e te r m in e d o n th e basis o f e x p e r im e n t.
^ ^Sin ce the e q u a tio n s o f m o t io n have G alilean inva rian ce, th e ch ara cter istic le n g th scale
o f th e flu c tu a tio n s / w ill clearly b e in d e p e n d e n t o f th e a b s o lu te value o f th e m e a n v e lo c ity
w (z), an d w ill d e p e n d o n ly o n th e variation o f u in th e n e ig h b o r h o o d o f th e g iv e n p o in t, i.e.,
o n th e derivatives o f w (z). T h e a s su m p tio n th a t / d e p e n d s o n ly o n th e first tw o d erivatives
dujdz and cfluldz'^w ill in this case b e th e sim p le st h y p o th e s is relating to this q u a n tit y .
V o n K drm dn’s h y p o th e s is , e x p r essed b y E q. ( 5 . 1 1 7 ) , ta k e n lite r a lly , im p o s e s e x c e s siv e ly

rigid c o n str a in ts o n th e tu r b u le n t v e lo c ity flu c tu a tio n s w h ic h d o n o t agree w ith th e natural
id ea o f th e irregularity o f th e v ariation s o f th e flu c tu a tin g v e lo c ity in space and tim e. As
easily see n from th e fo llo w in g d is c u s sio n , th e h y p o th e s is o f lo c a l self-p reserv ation p roves to
be a p p lic a b le n o t to in d ivid u al r ea liza tio n s o f th e fie ld o f flu c tu a tin g v e lo c ity , b u t o n ly to
th e statistic a l c h a r a cteristics o f such a field (see C hapt. 8, V o lu m e 2 w h ic h is d e v o te d to th e
sim ilarity h y p o th e s e s p r o p o s e d b y A. N . K o lm o g o r o v ). It is im p o r ta n t, h o w e v e r , th a t th e
fu n d a m e n ta l results ( 5 . 1 1 9 ) o f v o n K drm dn’s th e o r y m ay be d e d u c e d even o n th e basis o f
far w e a k e r a ssu m p tio n s; as w e have already s ee n , in a certain sen se th e y are natural
c o n s e q u e n c e s o f d im e n sio n a l c o n c e p ts . F u rth er, w e shall n o t e th a t as L o its y a n s k iy ( 1 9 3 5 )
s h o w e d , for th e d e d u c t io n o f Eqs. ( 5 . 1 1 9 ) it is s u ffic ie n t to app ly th e h y p o th e s is o f loc a l
self-p reserv ation t o th e m e a n v e lo c ity fie ld , w ith th e r e q u ir e m e n t th at at every p o in t JC„ =
(jfo, I/o. ô) there w ill be d e fin e d a scale 1 {zq) such th a t fo r z o w ith accu racy to
sm^ll q u a n titie s o f th e third order i n / , th e fo llo w in g c o n d it io n h o ld s:
In fa c t, to this a c cu ra c y , w e have
a (i) - u ( ij) ^ di ). + ).
c+ 4c*f/(
Sin ce, a c co r d in g to this h y p o th e s is , the right side o f th e ab o v e e q u a tio n m u st b e

in d e p e n d e n t o f Zq an d / , th e first e q u a tio n o f E q. ( 5 . 1 1 9 ) m u st also b e sa tisfied ; th e n , from
th e c o n d it io n o f Zq-in d e p e n d e n c e o f th e ratio
(ô) _ ^(ô) ,
P1“(^0+ 0- w(ô)l^ ^—j*
Eq. ( 5 . 1 2 0 ) is also d e d u c e d .
F o r plane-parallel flo w w ith zero pressure g r a d ien t a lo n g an in fin ite p lan e 2 = 0 , w ith
c o n s ta n t shear stress t = th e v o n Kdrmdn r e la tio n sh ip ( 5 . 1 1 9 ) fo r / in c o n ju n c tio n w ith
P ran d tl’s e q u a t io n x = [se e E q. ( 5 . 1 2 0 ) ] gives
<=>23)
T h u s, again w e o b ta in a lo g a r ith m ic e q u a t io n fo r th e m ean v e lo c ity
"W = - ~ l n ( z - z „ ) + fl; (5 .1 2 4 )
h is to r ic a lly , th is d e d u c t io n w a s th e first w h ic h led to a lo g a r ith m ic e q u a tio n . F or pressure

f lo w in a plan e c h a n n el or circular tu b e , in a d d itio n to th e e q u a tio n s t = — x and x =
p /^ /— ) it is also n e c essa ry to u se th e r e la tio n sh ip ( 5 . 1 7 ) or (5.1 7 ')* We m ay th en ob ta in

\dz/
e q u a tio n s for th e p r o file o f m ( z ), w h ic h , w ith an a p p rop riate c h o ic e o f th e c o n s ta n t x
(d iffe r in g fo r a c h a n n el an d a tu b e ), m a y lea d t o s a tis fa c to r y a g r e e m e n t w ith th e e x istin g
data [se e, e .g ., G o ld ste in ( 1 9 3 8 ) ] .
T h e sem iem p ir ic a l th eo r ie s o f Prandtl, T a y lo r , and v o n K i i m i n are th e classical e x a m p le s
o f ap p ro a c h e s to th e tu r b u le n c e p r o b le m b a sed o n a s su m p tio n s o n th e e x is t e n c e o f a
r e la tio n b e t w e e n th e lo c a l R e y n o ld s stress and th e lo c a l m e a n v e lo c ity fie ld . A tte m p ts to
fin d an im p ro v e d fo r m o f su ch a r ela tio n are still c o n tin u in g ; see , for e x a m p le , L ettau
( 1 9 6 7 ) , w h e r e o th er refer e n c e s to w o r k s o f this a u th o r m ay also b e fo u n d [a critical
d isc u ssio n o f L e t ta u ’s ap p ro a c h w as g iven b y L u m le y an d S tew a r t ( 1 9 6 5 ) ] . T h e classical
th eor ie s s e e m e d t o b e sa tis fa c to r y in th e early d a y s m a in ly b e c a u se th e y w e re co m p a r e d
a lm o s t e x c lu siv e ly w ith m e a s u r e m e n ts of th e m ean v e lo c ity d is tr ib u tio n , and this
d is tr ib u tio n is rather in se n sitiv e to th e h y p o t h e s e s a d o p te d (e sp e cia lly i f s o m e u n d e te r m in e d
c o n s ta n ts w ere in c lu d e d in th e th e o r y w h ic h m u s t b e d e te r m in e d fr o m o b s e r v a tio n ). W hen
m o r e d e ta ile d e x p e r im e n ts w e re p e r fo r m e d it b e c a m e c le a i th a t a n y th e o r y o f this ty p e is
e r r o n e o u s in p rin cip le and m a y b e u se d for th e d e s c r ip tio n o f very lim ite d tu rb u le n c e
p h e n o m e n a o n ly . F o r e x a m p le , th e m e a s u r e m e n ts o f Sch w a rz and C osart ( 1 9 6 1 ) , B^guier
( 1 9 6 5 ) and s o m e o th er s, s h o w d e fin ite ly th at in f lo w s w ith n o n s y m m e t r ic v e lo c ity p r o files
the p o in t o f van ish in g shear stress d o e s n o t agree very o f t e n w ith th e p o in t at w h ic h th e
v e lo c ity grad ien t v a n ish es; o b v io u s ly , th is fa c t c o n tr a d ic ts all m ix in g le n g th and e d d y
v is c o s ity th eo r ie s. It fo llo w s also fr o m th is fa c t th a t in n o n s y m m e tr ic a l flo w s , th e e d d y
v is c o s ity m u s t ta k e n eg a tiv e va lues at s o m e p o in ts ; th is m a k es th e w h o le c o n c e p t o f an e d d y
v is c o s ity see m m o r e th an a h t tle strange as a p p lie d to su ch flo w s . In fa c t, if w e d e fin e th e
e d d y v is c o s ity K in a. p u re ly fo r m a l m a n n er by K = — , th e n K p ro ves to b e n e gative

p(du/dz)
rather o f t e n , e sp e c ia lly w h e n large-scale natural tu r b u le n c e is c o n s id e r e d (se e, fo r e x a m p le , a
sp ecial m o n o g ra p h by Starr (1 9 6 8 ) c o n c e r n in g th is s u b je c t). A t p r e se n t, a lm o s t all
in v e stig ators agree th a t th e R e y n o ld s stress is n o t a lo c a l p r o p e r ty o f th e m o t io n b u t
d e p e n d s o n th e w h o le f lo w fie ld (in clu d in g its tim e h is to r y ). T h ere are n u m e r o u s d e d u c t io n s
related to this id e a , s u gge stin g th a t a tu r b u le n t flu id m u st b e c o n s id e r e d as a special
c o n tin u o u s m e d iu m w ith n o n -N e w t o n ia n v is c o e la s tic b e h a v io r [R iv lin , ( 1 9 5 7 ) ; M o ffa tt
( 1 9 6 7 ) ; C row ( 1 9 6 8 ) , an d o t h e r s ] . S u ch a d e s c r ip tio n o f tu rb u le n c e s h o w s v ery clearly th e
lim ita tio n s a n d in a d e q u a c ie s o f th e classical s em iem p ir ic a l th eo r ie s; u n fo r tu n a t e ly , b y its e lf,
it d o e s n o t give a u n iq u e a ccu rate m a th e m a tic a l m o d e l o f shear tu rb u le n c e. Several very
d iffe r e n t a tte m p ts to fo r m u la te an im p r o v e d n o n -N e w t o n ia n sem iem p ir ic a l th e o r y o f th e
R e y n o ld s stresses (i.e ., a th e o r y a c c o r d in g to w h ic h th ere is a n o n lin e a r r ela tio n b e tw e e n the
stress and th e rate o f strain o f th e v e lo c ity fie ld ) w e re m a d e b y L u m le y ( 1 9 6 7 a ) , P h illip s
( 1 9 6 7 ) and s o m e oth er s [se e also K ov d sz n a y ( 1 9 6 7 ) ] . H o w e v er , th e results o b ta in e d b y
th e s e a u th ors are o f a p relim in ary cha ra cter o n ly and w ill n o t b e d is c u s se d fu rth er h e r e.
6. THE ENERGY BALANCE EQUATION AND

ITS CONSEQUENCES
6.1 Equation for the Reynolds Stress Tensor
Earlier, we saw that as a result of the occurrence of additional
terms containing the Reynolds stress -cf/j = — in the Reynolds
equations for the mean motion, the system of these equations is not
closed. It is natural to attem pt to close this system of equations by

supplementing the Reynolds equations with new ones, which
describe the variation in time of the stresses themselves. These
equations for -ciy will be deduced in this subsection; we shall see that
they in turn, also contain several additional unknowns and, hence,
again, will not form a closed system. Nevertheless, these equations
for do impose new dynamic constraints, which are of definite
interest, on the statistical characteristics of turbulence since they
allow us to make a number of quahtative deductions on the
properties of turbulent flows. Particularly useful is the equation of
balance of turbulent energy, which describes the time-variation of
the density of kinetic energy of the fluctuating motion (or, in short.
of the turbulent energy) E { = -2 ; a detailed consideration of
this equation, every term of which has an explicit physical meaning,
shall be made in several subsequent subsections of this section.
For the deduction of equations for the Reynolds stresses, we may
use the general method of formulating the equations of moments
proposed by Keller and Friedmann (1924) [see also, Keller (1925)].
Let uu «2, . • •, un be some N different or coincident fluid dynamic
variables of a turbulent flow of compressible fluid, and aTi, Jf2- • • • >
be some N different or coincident points of the flow region. In this
case, the derivative with respect to time o f the A^th-order moment
... X2y • • •) Jf/vTi t) — til (-^l' 0 ^2 {^2< 0 • • • 0 (6.1)
(since the order of averaging and differentiation may be reversed)

may be written as
. . . + «, {X,, t) {X,. t ) . . . _ ( 62)
It is now sufficient to eliminate all time derivatives on the right side

of Eq. (6.2) with the aid o f the corresponding fluid dynamic
equations. We obtain the equation of balance for the moment flu,. . .
•••' X n , t) which express (jCi, x ^ , t) in the
form of a combination of the moments of the flow variables and

their space derivatives.
This method, o f course, may also be appUed to nonaveraged
quantities. Thus, for example, substituting the Navier-Stokes equa-
d du] dui
tions (1.6) into the equation —P « » + we obtain
~ r + + (/’«<•V + ~ + "/•'’<•«)! =
= (P«;^y + p«;^/) + Z' ( - ^ + ~ (6.3)
/ dui duj\ , , . .
where o.^ = p. is the viscous stress tensor in an incom
pressible fluid. Thus, in particular, for the kinetic energy density we
obtain
# + ^ - P^* (6.4)
where
tq Z \ UJCo
denotes the specific dissipation of kinetic energy (per unit time and
unit volume of fluid). The expression in parentheses on the left side
of Eq. (6.4) gives the energy flux density, which is due both to the
direct transfer o f energy by the displacement of fluid particles, and
to the action of the pressure forces and the molecular forces of
internal friction. The right side of Eq. (6.4), however, indicates that
the total kinetic energy of an arbitrary volume of fluid varies not
only because of the mechanical influx or efflux of energy through its
boundaries, and the action of the forces of pressure and viscous
friction applied to its boundaries, but also due to the action of the
body forces and the effect of energy dissipation which leads to the
transformation of parts o f the kinetic energy into heat.
Now instead of using the Navier-Stokes equations, we use the
Reynolds equations (5.1),
+ + = (6.5)
and take into account that p = const and 0; we obtain in a

similar manner, the following equation for the tensor puûf.
dÛiUj d -------
I F - + IT , [p“ i“ j+ +
+ (piifija + pufiia) — («/»ya + j] =
= (.PUiXj + pUjXt) + p + -^ ) —(a;. 4-
+ + <6.6)
In particular, for the density of the kinetic energy of the averaged

motion = ^ pMpttp we obtain
dEs
dt
= p « A -p ® .+ p « > ;-S -* <6.7)
where
«.P
denotes the rate o f dissipation of energy of the mean motion under

the action of molecular viscosity. The physical meaning of all the
terms of this equation, after elimination of the last term on the right
side which we shall consider in detail below, is analogous to that of
the corresponding terms of Eq. (6.4). We note only that the energy
flux density of the mean motion, in addition to the term
describing the transfer of energy molecular viscosity forces, contains
the term pw'a'pwj which describes the related process of energy
transfer due to the action of “ eddy viscosity.”
By averaging Eq. (6.3) and then subtracting Eq. (6.6) from it
termwise, we obtain the required equation for the Reynolds stress
tensor:
dû[u'j
dt
du, du'i
- w . +«>;.)]=9u\x-+ ,u)x‘+ / 4 J
This equation may also be obtained by first formulating the dynamic

equation for the velocity fluctuations u'i=ui — ui (equal to the
difference of the ith Navier-Stokes and Reynolds equations)
du, d - _ _____
P “S r + ( P “ i“ « + P“ a“ i + P “ i“ a — p“ i«a + = Pî- (6-9)
and then applying Eq. (6.2) to the central moment u[u'j = —

However, we observe immediately that in addition to the mean
velocity «i and the Reynolds stresses pu\u'j, Eq. (6.8) contains several
new unknowns. These new unknowns are, first, the third-order
central moments second, multiples by v of the second
moments of the velocity fluctuations and their space derivatives,
occurring in the members and and not expressed directly
in terms of the Reynolds stresses, and third, the joint second moments
_____ du'i
of the pressure and velocity fields of the form /?'«' and P ' [which,
with the aid of Eq. (1.9') may be written in the form of integrals of
two-point, third-order moments of the type
u\{x, t)u'j{x', t)u'^{x', /)]. Thus, the Reynolds equations (6.5) and
the equations for the Reynolds stresses (6.8) once again do not form
a closed system. Now if we try to complete this system with the aid
of equations for the new unknowns occurring in Eq. (6.8) and begin
with the equations for the third-order moments then these
equations in their turn introduce many other new unknowns [for
example, the fourth-order moments ph^h'.k^?/,' and third-order mo-
_____ d u '~ ^ ,
ments of the type p'ti'ji',.
J «
or -r—
OXf fi O X f i
a l ; see Chou (1945a)] and the
difference between the number of unknowns and the number of
equations becomes even greater. Thus formulation of the equations
for the higher moments does not permit us to obtain at any stage a
closed system of equations describing the turbulent motion.
It is also easy to derive the equations for the turbulent fluxes <7,- =
p^'ui (or = CpP&'ui) which have appeared in Eq. (5.7) for it is
only necessary to use an equation for t?' = together with Eq.
(6.9). These equations for the flux vector qt are quite similar to Eq.
( 6 .8 ); they also contain a number of new unknowns and do not allow
us to obtain a closed system.
O f c o u r se , w e m a y a tt e m p t t o c lo se th e s y ste m o f e q u a tio n s ( 6 . 5 ) and ( 6 . 8 ) w ith th e aid

o f a d d itio n a l h y p o th e s e s o f s o m e k in d o r a n o th e r , w h ic h p e r m it th e e x p r e s s io n o f th e “ n e w
u n k n o w n s ” in E qs. ( 6 . 8 ) in term s o f th e first an d s e c o n d m o m e n ts lit and Su ch

a tte m p ts , in fa c t, c o n s t it u t e n e w variables o f th e “ sem iem p ir ic a l th e o r y o f tu r b u le n c e ,”
w h ic h d iffe r fr o m th e th eo r ie s c o n s id e r e d in S ect. 5 .9 o n ly in th e fa c t th a t th e h y p o th e tic a l
c o n n e c t io n s he r e are o f a s o m e w h a t m o r e c o m p le x stru ctu re. T h ese m ore c o m p le x ,
sem iem p ir ic a l th eo r ie s have b e e n p r o p o s e d b y m a n y au th o r s at d if fe r e n t tim es; here w e shall
c o n s id e r o n ly a fe w o f th e m . O n e o f th e first w o r k s d e v o te d to th e p r o b le m o f th e c losu re
o f th e s y s te m o f e q u a tio n s ( 6 . 5 ) and ( 6 . 8 ) w as th a t o f Z agustin ( 1 9 3 8 ) ; th en c a m e a sh o r t
n o t e b y K o lm o g o r o v ( 1 9 4 2 ) in w h ic h tu r b u le n c e is ch a r a cter iz e d b y tw o b a s ic p aram eters,
the in t e n s it y and scale p aram eters, in term s o f w h ic h all th e term s o f th e en ergy b a lan ce
e q u a tio n are th en e x p r e s s e d . T his latter e q u a tio n is d e d u c e d fr o m E q. ( 6 .8 ) . [A sim ilar id e a
w as s u g g e ste d s o m e w h a t later b y Prandtl ( 1 9 4 5 ) .] F o llo w in g K o h n o g o r o v , N e v z g ly a d o v
( 1 9 4 5 a ; 1 9 4 5 b ; e t c .) p r o p o s e d a n u m b e r o f d if fe r e n t h y p o t h e s e s fo r th e c losu re o f th e
s y ste m o f e q u a tio n s ( 6 . 5 ) and ( 6 . 8 ) . Sim ilar in v e s tig a tio n s w e re carried o u t in th e 1 9 4 0 ’s b y
C hou ( 1 9 4 5 a ; 1 9 4 5 b ; 1 9 4 7 ) . C h o u , in particular, w as th e first to a tt e m p t to u se, in a d d itio n
to E qs. ( 6 . 5 ) and ( 6 . 8 ) , th e e q u a tio n s o f th e th ird -order m o m e n ts th e fou rth -o r d e r
m o m e n ts w h ic h th e n arise are e lim in a te d w ith th e aid o f th e h y p o th e s is a b o u t

th e van ish in g o f th e fo u rth -o r d e r v e lo c ity c u m u la n ts , o r s o m e related h y p o th e s is . F in a lly , in
th e 1 9 5 0 ’s and 1 9 6 0 ’s th e w o r k s o f R o tta ( 1 9 5 1 b ; 1 9 5 1 c ) and D av id o v ( 1 9 5 8 ; 1 9 5 9 a ;
1 9 5 9 b ; 1 9 6 1 ) app eared w h ic h w e shall discu ss in s o m e w h a t greater d e ta il, so as to d isp lay
b y c o n c r e te e x a m p le s th e gen eral c h aracter o f th is w h o le g r o u p o f in v e stig a tio n s.
R o tta c o n s id e r e d E qs. ( 6 . 5 ) and ( 6 . 8 ) fo r a ste a d y tu rb u le n t flo w , w ith m e a n v e lo c ity
U\(Xz) d ir e c te d e v e r y w h e r e a lo n g th e O xi axis and in th e ab se n c e o f e x te r n a l fo r c es . F or
s im p lic ity , h e a ssu m ed th a t th e term s in b r ac k e ts o n th e le ft sid e o f E qs. ( 6 . 8 ) describ in g th e
“ transfer d if fu s io n o f R e y n o ld s str e s se s ,” m ay in th e first a p p r o x im a tio n b e ig n o r e d ,
c o n sid e rin g o n ly th o s e r egion s o f th e f lo w w h e r e th e tu r b u le n c e is m o r e or le ss h o m o g e n e o u s .
C o n s e q u e n tly , th e transfer o f tu r b u le n t q u a n titie s fro m n eig h b o rin g r egion s p la y s a sm all
part. F u rth er, as in th e w o r k s o f K o lm o g o r o v ( 1 9 4 2 ) a nd Prandtl ( 1 9 4 5 ) it w a s a ssu m ed
th a t tu r b u le n c e m a y b e d e scr ib e d b y its in t e n s it y (s p e c ific e n e rg y ) and
th e scale /, a variable o f th e d im e n s io n o f le n g th d escr ib in g th e m e a n size o f th e tu rb u le n t

d istu rb a n c es, an d fo r m in g an a n a lo g to P ra n d tl’s “ m ix in g le n g th .” T o e valu ate th e q u a n tity
/ du\ duj \
\ ~dx i / ’ fo llo w in g C h o u ( 1 9 4 5 a ; 1 9 4 5 b ) , used th e fa c t th at th e pressure
flu c tu a tio n s s a tisfy th e P o isso n e q u a t io n ( 1 . 9 ) , and , c o n s e q u e n tly , m a y b e p u t in a form

related t o E q. ( 1 . 9 ') , w h e r e th e in tegran d is th e sum o f tw o s u m m a n d s, o n e , d e p e n d e n t o n
the m ean v e lo c ity U\{xz) and th e o th e r c o n ta in in g o n ly v e lo c ity flu c tu a tio n s . T he
contribution of the first of these summands in /?' I -r----- -f- — I may be put in the form
o f a scries o f su c ccssivc d erivatives o f th e m e a n v e lo c ity p ro file Wi(.V3 ) w ith c o e f fic ie n ts th at

are integrals o f th e tw o -p o in t, sec o n d -o r d e r v e lo c ity m o m e n ts . T h e se c o e f fic ie n ts satisfy
algebraic relatio n sh ip s w h ic h are d e d u c e d fr o m th e c o n tin u ity e q u a t io n and m a y b e
ex p r essed a p p r o x im a te ly in term s o f th e R e y n o ld s stresses a nd th e le n g th /. R o tta id e n tifie d
th e c o n tr ib u tio n o f th e s e c o n d su m m a n d , w h ic h is d e p e n d e n t o n th e t w o -p o in t, third-order
/ du \ du'j
m o m e n ts w ith th e value o f p ' I ,------- 1I- - 3, - ^ in h o m o g e n e o u s tu r b u le n c e , w ith c o n s ta n t
^ \ ax
Ox jf ' x1
dx:
0
m e a n v e lo c it y . A lso , h e p r o p o s e d to e v a lu a te it w ith th e aid o f th e h y p o t h e t ic a l relatio n sh ip

(ju stifie d b y certain q u alita tiv e c o n s id e r a tio n s ; see b e lo w , th e f o o t n o t e in S e c t. 6 . 2 )
/ dll': da. \ [ - - 2 \
p' ( + id j) - 3
w h e r e f i is a n u m erica l c o e f f ic ie n t , d e te r m in e d fr o m th e d a ta . R o t t a p r o p o se d to use a
similar h y p o th e tic a l r ela tio n sh ip o f th e form
, r)//- , dUj
w h e r e C2 an d 63 are n e w n u m e ric a l c o n s ta n ts , fo r th e e v a lu a tio n o f t h e “ d is s ip a tio n te r m s ”

o n th e right side o f E q . ( 6 . 8 ) c o n ta in in g th e v is c o s ity . U sing th ese fo r m u la s, E q s. ( 6 . 5 ) and
( 6 . 8 ) p e r m it us t o d e te r m in e th e ratio o f t h e various R e y n o ld s stresses t o th e m e a n v e lo c ity
g r a d ie n t, and to o b ta in certain o th e r r esu lts w h ic h m a y b e verified e x p e r im e n ta lly . W ith an
a p p rop riate c h o ic e o f th e param eters C2 yCz all th e s e results prove to b e in fairly g o o d
a g r ee m e n t w ith th e m e a s u r e m e n ts o f th e tu r b u le n c e c h a ra cteristics in a c h a n n el, carried o u t
b y R e ich a rd t ( 1 9 3 8 ) and L a u fer ( 1 9 5 1 ) . In his s e c o n d pap er, R o tta ( 1 9 5 1 c ) , in a d d itio n to
th e r ela tio n sh ip s ( 6 . 1 0 ) and ( 6 . 1 1 ) , a lso p r o p o s e d a d iffe r e n tia l e q u a tio n for th e t im e - a n d
s p ace-variations o f th e len g th /, and th e r e b y o b ta in e d a m o r e c o m p le te e x p la n a tio n o f th e
e x p e r im e n ta l facts.
A m o r e d e ta iled s c h e m e for th e sem i em p ir ic a l c lo sin g o f E qs. ( 6 . 5 ) and ( 6 . 8 ) w a s
d e v e lo p e d b y D a vid o v . In his w o r k s, u n lik e R o t t a ’s w o r k s , o n th e le f t sid e o f E q. ( 6 . 8 ) o n ly
th e term s ja~\~ w h ic h d e scr ib e th e transfer o f R e y n o ld s stresses du e
t o th e pressure flu c tu a tio n s are n e g le c te d . T h e term s o f E q . ( 6 .8 ) c o n ta in in g th e v is c o s ity m

— — , du\ du\
m a y b e tran sfo r m e d in t o th e fo r m -I- 2 ;j ; th e first term is e x p r e s s e d in
^ ^ ^ dx^ dx^
term s o f th e R e y n o ld s stresses, w h ile D a v id o v p r o p o s e d to ta k e th e s e c o n d term e q u a l to
— (d u e to th e is o tr o p y o f sm all-scale m o t io n , w h ic h w e shall d isc u ss in d e ta il in V ol-
duo du[
u m e 2 o f th is b o o k ) , w h e r e £i = v is a n e w fu n d a m e n ta l ch a r a cter istic o f turbu-
dXr^ Ox^
le n c e th at is su b ject to d e fin itio n . T o d e te r m in e th e term s o n th e right side o f E q. ( 6 . 8 ) c o n
tain ing th e pressure, D a vid ov u se d th e e q u a tio n
w h ere Ci is an em p ir ic a l c o n s ta n t [rela te d to th e c o n s ta n t C\ in E q . ( 6 .1 0 ) ] and B i j is a

fairly c o m p le x a d d itio n a l te n s o r , e x p r e s s e d in te r m s o f th e m e a n v e lo c it y , its space
derivative and th e R e y n o ld s stresses; th is te n so r d e scr ib e s th e a n is o tr o p y o f th e v e lo c ity
flu c t u a tio n s in th e b o u n d a r y la y e r c lo se t o a rigid w all.^ ^ F u rth er, fo r th e th ird -order
m o m e n ts e q u a tio n s sim ilar to E q. ( 6 . 8 ) are d e d u c e d w h ic h are th e n s u b je c te d to

even m o r e radical s im p lific a tio n s in w h ic h all te rm s c o n ta in in g th e v is c o s ity , and a lso th e
fo u rth c u m u la n ts o f th e v e lo c ity are ig n ored ; m o r e o v e r , th e sy m m e tr ic a l c o m b in a tio n o f
s u m m a n d s o f th e fo rm is rep la c ed b y th e e x p r essio n w h e r e c<i is
a n o th e r em p ir ic a l c o e f f ic ie n t . F in a lly , to d e te r m in e Ki , D a vid ov first p r o p o s e d a special

“ transport e q u a t io n ” c o n ta in in g , in a d d itio n to e i, o n ly th e first, s e c o n d and third m o m e n ts
o f th e v e lo c ity fie ld ; later, h o w e v e r , h e r e fu te d th is [s e e D av id ov ( 1 9 6 1 ) ] and rep la c ed it b y
an e q u a t io n o f th e form
0 ^ d \ dyu e, ~ r~ 7 du e?
(6 .1 3 )
, duo du^
w h e r e C3 is a th ird em p ir ic a l c o e f f ic ie n t , an d 7 ^^ = , ^ = 1 , 2 , 3 , are th ree
fu rth er a d d itio n a l u n k n o w n s . N o w , fo r th e c lo su re o f th e s y s te m it is n e cessary to fo r m u la te

e q u a tio n s fo r yh \ th is m a y b e carried o u t o n th e basis o f E qs. ( 6 . 9 ) fo r u\ w ith th e aid o f
th e gen era l rule (6 .2 ). A fte r co n sid e ra b le s im p lific a tio n , using once again certain
h y p o th e tic a l r e la tio n s h ip s, th e s e last e q u a t io n s are r e d u c e d to th e fo llo w in g form :
(i dx^ } dx^ + dx^ + 9 "1 + *^4 6 fft -
T h u s, fin a lly , w e o b ta in a very c o m p le x b u t c lo s e d s y ste m o f 23 q u asi-linear d iffe r e n tia l
eq u a tio n s in 23 u n k n o w n s , u^, u \ u j , Sj and yu c o n ta in in g fo u r em p ir ic a l c o e f fic ie n ts , th e
values o f w h ic h m a y b e e s t im a te d , fo r e x a m p le , o n th e basis o f L a u fe r ’s m e a s u r e m e n ts. A

m ore e x a c t v e r ific a tio n o f th is s y ste m requires sp e cia l e x p e r im e n ts , in w h ic h all th e
unknow ns in v o lv ed w o u ld have to be care fu lly m e asu red ; so far, a p p a r e n tly , such
m e a su r em en ts have never b e e n carried o u t .
^'^For e x a m p le , in th e case o f a f lo w a b o v e a plan e rigid w a l l a s = 0 , it is r ea son ab le to

assu m e th a t B q = d peî^Sj^ w h e r e e = e7 is th e m ean rate o f d is s ip a tio n o f tu rb u le n t energy
(w h ic h it is u su a lly c o n v e n ie n t to use in ste a d o f e 1 ) a n d d is a d im e n sio n le ss c o n s ta n t. S u ch
a s u p p o s itio n w as u sed in particular b y M o n in ( 1 9 6 5 ) for a b o u n d a r y -la y e r flo w o f a
th er m a lly stra tifie d flu id . M on in has c o n s id e r e d all th e e q u a tio n s for th e seven n o n z e r o
sec o n d m o m e n ts , u'2 ^ , u \t\ a n d (b a sed o n th e B o u ssin esq

e q u a tio n s ; in th ese e q u a tio n s all th e term s w e re n c g le c te d w h ic h d e scrib e tu r b u le n t transfer
( “d if f u s io n ” ) and th e term s o f th e fo r m p\dT^/dxi)v/QTe e lim in a te d in th e e q u a t io n s for th e
m o m e n ts w / r V i t h th e aid o f th e fo llo w in g h y p o th e tic a l fo r m u la , rela ted to E q. ( 6 .1 2 ) :
6.2 Equation of Turbulent Energy Balance

Just as Eq. (6.4) for the total kinetic energy E is derived from Eq.
(6.3), we may obtain from Eq. (6.8), after summation over / = /, the
following equation for the mean density of the kinetic energy of the
fluctuating motion E , ~ ^
where
is the mean rate of viscous energy dissipation of the fluctuating

motion. The expression in parentheses on the left side of this
equation gives the turbulent energy flux density connected with the
transfer of turbulent energy by the mean motion, the pressure
fluctuations, the forces of internal friction (molecular viscosity),
and, finally, the “ turbulent viscosity,” i.e., the fluctuating compo
nent of velocity (the term ^ p«^«'«'). The term
du^
occurring on the right side of the equations for E,, and Et with
opposite signs describes the mutual exchange of energy of the mean
and fluctuating motion. We shall discuss the question of the value of
this term in detail later.
Equation (6.15) is a general energy balance equation of turbulent
flow. It shows that the density of turbulent energy at a given point
of the flow may vary due to the transfer of turbulent energy from
different parts of the fluid (i.e., by the diffusion of turbulent
energy), the action of the fluctuations of the external forces, the
viscous energy dissipation and, finally, the transformation of part of

the energy of the mean motion into turbulent energy, or, conversely,
the transformation of part of the turbulent energy into energy of the
mean motion. The turbulent energy in this equation, of course, may
also be replaced by the turbulent intensity (i.e., by the mean kinetic
energy of the turbulence per unit mass o f the fluid)f> = - ^ = y « 'a '.
If we now denote by the symbol + “o. the first
derivative with respect to the mean motion, then, since == 0, Eq.

(6.15) may be reduced to the form
Db dUff - ^ d
Ot ~ dx^ dx^ 2 p “ a “1“
duo
The energy balance equation (6.15) or (6.17) completes the Reyn

olds equations in the sense that it imposes a further im portant con
straint on the statistical characteristics of turbulence. True, it con
tains new unknowns p'u[ and e, which do not occur in the
Reynolds equations, therefore, a more complete description of the
turbulent flow is required. However, the physical meaning of the
terms in Eq. (6.17), containing these new unknowns is fairly clear.
This is a great help in attempts to express them approximately in
terms o f simpler characteristics. Moreover, in a number o f cases, for
example, if the turbulence is almost spatially homogeneous, the terms
describing the diffusion of turbulent energy can be ignored to a first
approximation because the spatial transfer of energy here plays only
a small part. The use of the energy balance equations to complete
the Reynolds equations was first proposed by Kolmogorov (1942).
Proceeding to a more detailed consideration of individual terms in
the turbulent energy balance equation, let us begin with the terms
containing pressure fluctuations. In the general energy balance fo r£ t
these terms are unimportant; as Eqs. (6.15) and (6.17) show, for an
incompressible fluid, the pressure fluctuations lead only to an
additional transfer o f turbulent energy from some parts of the fluid
to others. Hence, if we consider a volume of fluid, through the
boundary of which there is no inward or outward flux of turbulent
energy, the presence of pressure fluctuations will generally have no
effect on the variation of the total turbulent energy of this volume.
Moreover, it seems reasonable to suppose that the contribution of

the pressure fluctuations to the turbulent energy flux density is in
many cases relatively small. Nevertheless, these fluctuations play a
very im portant part in a turbulent flow. Specifically, consider the
1 2
quantities E i = - g - ( n o sum over i ) , which are the mean densities
of the fluctuating kinetic energy along the individual coordinate
axes. For these quantities, on the assumption that X\ = 0 , Eq. ( 6 . 8 )
may be reduced to the form
where
the quantity pej may be defined as the “dissipation” o f the energy

(These “ partial dissipations” will all be equal to p ^ / 3 if the Reynolds
number is large enough, due to the isotropy of small scale motion;
see Chapt. 8 in Volume 2 of the book.)
Let us now assume, for definiteness, that the mean motion is
parallel to the Oxi direction (so that U2 = = 0 ) and the turbulence
is homogeneous in the Oxi and Oxz directions (i.e., all mean
characteristics o f the turbulence depend on Xs only). Using the
du^
continuity equation -33^ = 0 , we can rewrite Eqs. (6.18) in the form
dE,
dt dx
du ^ duo - — — du
= - ^ ^
(6.19)
dE,
dt dx,
dEo d
_?__|___
dt ^ dx,
du2 dui
Equations (6.19) show that p ;5— and p ' - ^ describe the mutual
I'-ta 0 X3
interchange of the energy £1 of the longitudinal fluctuations with the
energies £ 2 and £3 of the transverse fluctuations. Moreover, it is clear
from these equations that the energy o f the longitudinal fluctuations
may be maintained from the energy o f the mean flow [this process is
described by the term — p a ' w '- ^ of the first equation of Eqs.
(6.19)], but that the energy of the transverse fluctuations may be
increased only by the energy o f the longitudinal fluctuations. In fact,
if we consider a steady regime, i.e., we assume = 0 , and integrate
all the equations of Eqs. (6.19) over a space region V, across the
boundary of which there is no flux of turbulent energy, we obtain
V V V V
(6.20)
Thus the energy of the mean motion, in fact, is transmitted directly

only to the longitudinal fluctuations u[, while the transverse
fluctuations and Uj obtain energy from the longitudinal velocity
fluctuations due to the action o f the pressure fluctuations. Conse
quently, the latter produce a redistribution of energy between the
fluctuating motion in different directions and create a tendency
towards isotropy of the fluctuating motion.
To obtain a clear picture of the energy-transfer mechanism from
the longitudinal velocity fluctuations to the transverse fluctuations,
we note that the negative sign of
( du'^ du'^ \
P dJ^= ~ \P J^,^P -^1
denotes the presence of a positive correlation between the positive

(negative) pressure fluctuations and the convergence (divergence) of
the longitudinal velocity fluctuation. In other words, that the
convergence of u[ leads to a predominance of positive pressure
fluctuations, while the divergence of u[ leads to a predominance of
negative pressure fluctuations. At the same time, the positive
pressure fluctuations produce, predominantly, a divergence of the
transverse fluctuations ii'^ and Wj, while negative fluctuations produce
a convergence. However, this must be the case, if, for example, two
neighboring elements of fluid are moving toward each other along
the direction of the mean motion. Then in the region between them
there will be formed at the expense of their energy a positive
pressure fluctuation and this increase of pressure will lead to efflux
of fluid in the transverse direction in accordance with the condition
_ _
of incompressibility — 0. In general, if a'^ > and
so that the velocity fluctuations along the Oxi axis are predominant,
then even in the absence of a mean velocity, the sign of the pressure
fluctuations will basically 'be determined by whether neighboring
elements of fluid converge or diverge in the Oxi direction; conse
quently, p ' - ~ w i l l here be negative and the pressure fluctuations
will tend to equalize the energy of the various components of the
velocity [see, e.g., Batchelor (1949a) where this role of the pressure
fluctuations is worked out, using the example of homogeneous
turbulence]
* * T h e se arg u m e n ts j u s tify th e a s su m p tio n th a t p ' m u st in c lu d e th e te r m
p r o p o r tio n a l t o — (u'y — c o n ta in e d in Eqs. ( 6 . 1 0 ) an d ( 6 . 1 2 ) . W e n o t e
dll', \
fu rth er, th a t d u e to th e ten so rial character o f p ' ^ j fo ll o w s a u to m a tic a lly
th at
/ dll': dll': \ -r-r
as is sta te d in E qs. ( 6 . 1 0 ) and ( 6 . 1 2 ) . T h u s it is natural t o th in k th a t th e pressure
f lu c tu a tio n s in tr o d u c e a c o n tr ib u tio n to w here i ^ /, p r o p o r tio n a l t o —
i.e., th e y le a d to a d e c rease in th e a b s o lu te value o f all R e y n o ld s stresses , i j.

du„
Now let us consider the term^l = — ^ which, in the energy
balance equation describes the exchange of energy between the mean
and the fluctuating motion. If at a given point of space A > 0 , then
the turbulent energy density at this point increases at the expense of
the energy o f the mean motion. If, on the contrary, A<Q, then the
energy density of the mean motion increases at the expense of the
energy o f the fluctuations. The latter possibility, at first glance,
seems paradoxical; however, this requires more careful analysis.
Equation (6.15) shows that for flows of incompressible fluid in a
field of nonfluctuating body forces, the only possible source of
turbulent energy within a volume which has no influx of turbulent
energy across its boundary is the transformation of the energy of the
mean motion. Under these conditions, the initiation and develop
ment of turbulence or the maintenance of a stationary state in the
given volume are possible only on the condition that the integral of
A over the whole volume is positive [see, for example, Eq. (6.20)].
We encounter these conditions, in particular, in usual laboratory
flows of incompressible fluid in tubes, channels and boundary layers
(with low “initial turbulence” of the incident flow) where direct
measurement of ?«•«'■ and show that, as a rule, A is positive at all
points of the turbulent flow (which agrees well with the classical
semiempirical theories of Sect. 5.8).
However, if the turbulence has some “external” source of energy,
for example, if it is caused by artificial mixing of the fluid, or, in the
case of a compressible fluid, if it arises because of the presence of
density fluctuations produced by influx of heat, then the possibility
of the transformation of the turbulent energy into energy of the
mean motion, i.e., the possibility that A < 0 , is not excluded. This is
exactly the case for atmospheric turbulence on the scale of the
general circulation of the atmosphere of the earth. In this case, by
turbulence we must understand so-called macroturbulence, i.e., the
totality of irregular large-scale motion of cyclone and anticyclone
type, superimposed on the regular atmospheric circulation; the idea
of the statistical description o f this macroturbulence was first
advanced by Defant (1921). In conditions of general circulation, the
individual “turbulent disturbances” (cyclones and anticyclones) can
arise entirely due to the energy introduced by local influx o f heat,
while later, some part of their energy may also be transferred to the
mean flow of the general circulation. A long time ago, geophysicists
concluded from studying the data on the energy, momentum and

angular momentum balance of the whole atmosphere, that the
observations cannot be explained without the assumption that in
certain atmospheric regions, disturbance energy is transformed into
mean flow energy [see, for example, the survey article by Rossby
(1948)]. The possibility of such energy transformations was then
analyzed theoretically [see, for example, the work of Lorenz (1953)
in which conditions determining the sign of A are established].
Finally, we note that for the flows which make up the general
circulation of the atmosphere, the actual values of A have been
calculated directly by a number o f authors on the basis of Eq. (6.16),
according to the data of meteorological observations. Also, in many
cases, it was actually found that ^4< 0; on this point, see, for
example, the works of Monin (1956c), Gruza (1961), and especially
the monograph by Starr (1968), all of which contain extensive
additional bibliographies.
Similarly to the derivation of Eq. (6.15), we can derive the balance
equation for mean square fluctuation of the concentration of a
passive admixture (or o f the temperature), i.e., for the variance =
_ ,j)2 ^ y2. This equation has the following form:
+u„ ~ + ^ ^ = 2 ^ - 2 ^ -2N
dt d x „ \ “ ^ dx„ j ax„
(6.15')
where
2N = 2x
m
is the mean rate of diminution of mean square fluctuation due to

the action of molecular diffusivity (this quantity is often called the
concentration or temperature dissipation) and/?'is the fluctuation o f
the intensity of external sources of i?at a given point (e.g., o f radiative
influx and of temperature increase due to change of water phase
state when i? is a temperature of the atmosphere). The term A =
-2 UgO' in this equation describes the production of mean square
fluctuation by the interaction o f turbulent flux of concentration
(or temperature) and mean concentration (temperature) gradient; the

,o'2 --
term describes the advection of by mean motion, and the
expression in parentheses gives the flux density of due to

molecular and turbulent diffusion.
6.3 General Concept of the Viscosity and
Thermal Diffusivity
The eddy viscosity was introduced in Sect. 5.1 for the case of a
plane-parallel flow; it was used several times later in Sect. 5 in
describing simple laboratory turbulent flows (see, especially, Sect.
5.9). We shall now show how similar coefficients may also be defined
for arbitrary three-dimensional flows.
Let us assume that the turbulence is generated only as a result of
the transfer o f part of the energy of the mean flow into the energy of
small-scale disturbances, i.e., due to A being positive. It is reasonable
to think that in this case all the statistical characteristics of
turbulence (in particular, the Reynolds stresses) must depend on the
mean velocity field. However, with respect to the mean motion, the
Reynolds stresses are analogous to the viscous stresses in the laminar
motion of a fluid. Hence, if the mean motion has the character of the
overall motion of the fluid as a rigid body, i.e., if it is not accom
panied by any strain, then it is natural to assume that the Reynolds
stresses acting on any selected surface element in the fluid will be
directed along the normal to this element. But in that case the tensor
___ I ___2
pu'iu'j will be isotropic: pu'iu'j = , where t = y p « '« '= -g-pf>.
Consequently, the turbulent energy Et = pb is in a definite sense,
analogous to the pressure. However, if the mean motion results in
relative displacement o f the fluid particles, then the stresses xW=
— pu'.u'j must apparently depend on the mean velocity derivatives.
Restricting ourselves to the case of fairly small mean velocity
gradients, it is possible in the first approximation to take into
account only the first derivatives and to assume that the
dependence o f ti/ on these derivatives is linear. This idea is
fundamental to attempts to introduce the general concept of eddy
coefficients. __
Since the stresses form a symmetric tensor, their dependence
on the derivatives must be of a tensor nature. The quantities
do not form a symmetric tensor, but from them a symmetric rate of

du[ duj ^
strain tensor may be formed, > which characterizes
the deviation of the mean fluid motion from the motion of a rigid
body. Consequently, it is natural to consider the tensor as a
2
linear function of the tensor which becomes for ^,^ = 0.
The coefficients of this linear function will be coefficients o f “eddy
viscosity.”
It is known that the molecular viscosities v and are connected by
the relationships |x~pMm/m with the molecular mean
velocity and mean free path Im- We may assume that for the
coefficients of eddy viscosity there will exist similar relationships, in
which the role o f «m and /«, will be played by the corresponding
characteristics o f disordered turbulent motion. It is clear that the
root mean square of the velocity fluctuations, i.e., will be the
characteristic analogous to Um. Instead of Im, however, we must use
the length scale of turbulence I, which describes the mean distance to
which turbulent formations can be displaced, conserving their
individuality [i.e., the Prandtl “mixing length,” which is of the same
order of magnitude as the “correlation length,” determined from the
space correlation function with the aid of an equation of form
(4.75)]. The turbulence may have different length scales in different
directions because it may be of course nonisotropic. Hence, strictly
speaking, at every point o f the flow there must be defined an
ellipsoid of length scales, i.e., there must be given a symmetric tensor
of second rank/,-; (the scale tensor) of the dimension of length. Then
the values M u = pKij — p Y b lij will be coefficients of eddy
viscosity. Using the tensor hj and taking the symmetry of the stress
tensor into account, we may suppose that
= Yb w - (6.21)
This equation was proposed by Monin (1950b); it may be considered

as one more semiempirical hypothesis, similar to those introduced in
Sect. 5.9. Of course, the most general linear dependence between the
tensors and Oij would have the form
pajwy — -g pbhij p^/yo,p*I*ap>

390 S T A T IS TIC A L F L U ID M ECHANICS
whereKtjop is a fourth-rank tensor, which is symmetrical with respect

to (, /and to a, p, and which satisfies the condition 0. The
use o f Eq. (6.21) means that the tensor /Cijapis assumed to be o f the
specific form:
However, if we make no additional assumptions on the scale

tensor then Eq. (6.21) [which is analogous to Boussinesq’s
equation (5.5) defining the scalar coefficient of eddy viscosity K\
may be considered not as a hypothetical connection, but simply as
the definition of new turbulent characteristics Uj introduced instead
of the Reynolds stresses tI;-. In fact, Eq. (6.21) gives a system of six
equations in six unknowns /ii, l\ 2 , /is. I 22 , I 23 , kz- Transforming to a
coordinate system in which (D,j is a diagonal tensor, and taking into ac
count that (Dll + <1)22 + = 0 , it is easy to show that the determinant
of this system is proportional to (d et|| 0 ij||) 2. Thus, provided that det
llOijll 0 , i.e., if (t>ij is not a degenerate tensor, then the quantities
are determined uniquely in terms of pu'iu'j and, conversely, Eq. (6.21)
can always be satisfied. In this case, the arguments used above are
simply an aid in the physical interpretation of and K.tj= V b l ip
which permit, if necessary, certain hypotheses to be formulated as to
their nature. If, however, det l|d>ij|| = 0, then Eq. (6.21) imposes
additional restrictions on the form of the Reynolds stress tensor;
these restrictions may sometimes be only a fairly rough approxima
tion to reality. For example, if the mean motion is plane-parallel, so
thatMi = ui(x 3), m2 = W3 = 0 and, accordingly, only the components O 13
and O 3, of the tensor (Djj are different from zero, then it follows from
Eq. (6.21) that /,y = 0 when / # / and u'^ =u'i =u'i =2bl3 ; however,
the last equations u? =u'î = u'i = 0 are not, generally speaking,
satisfied in real plane-parallel turbulent flows (see above. Fig. 26,
Sect. 5.3).
Nevertheless, it will still be meaningful in some cases to use b and
lij instead of the tensor since the tensor/jj has a more obvious
geometrical meaning and its principal axes may sometimes be
preassigned from geometrical considerations. In many cases it is even
possible as a first rough approximation simply to assume that
is an isotropic tensor, i.e., to use the hypothetical equation
= (6 .22)
(We note that here the tensor u\u'^ is not isotropic, but its principal
axes are the same as those of the strain rate tensor ;y, which is not
quite true in real flows.) The hypothesis (6.22) is close to that used
by Boussinesq; the quantity l Y b ~ K which appears in it is the scalar
eddy viscosity. This hypothesis is used and discussed in many works;
as examples, we can mention the papers by Harlow and Nakayama
(1967), and Lumley (1967b) [in his paper, Lumley uses the time
scale t* = l/\fb instead of the length scale /].
According to the hypothesis (6.22), .4 has the form
A = P/C<1>,3 ^ (6.23)
which is analogous to the expression for the rate o f energy

dissipation of the mean motion
but with the molecular viscosity fi replaced by the eddy viscosity K-

The condition /1> 0 is equivalent to K> 0. Thus the formal use of
the concept of eddy viscosity f o r /l < 0, indicates the introduction of
“negative viscosity” (cf. the end of Sect. 5.9).
Equations analogous to Eq. (6.21) or Eq. (6.22) may also be
formulated with turbulent heat and mass transfer. If is the
admixture concentration, then the turbulent flux density of the
admixture in the Oxi direction is equal to Qi>'u'. In the general case
of anisotropic turbulence, we may write
(6.24)
where the components of the ten so r/(»;„ are the admixture eddy
diffusivities; when ij is the temperature, the coefficient Cp must be
inserted on both sides of Eq. (6.24) [cf. Eqs. (5.9) and (5.9'), Sect.
5.1). When using the scale tensor/,j, it may be assumed that the
anisotropy of the eddy diffusivities is^related to the anisotropy of the
length scales, i.e., that /('8(,= aop Y b lij, where a»is a dimensionless
parameter. If we ignore the anisotropy of the scale tensor, i.e., put
/<^f»/6ij-,then Eq. (6.24) becomes
= = (6-25)
We note that even for plane-parallel turbulent flows, for which Eq.
(6.25) may be considered simply as the definition of a new quantity
= this relationship becomes hypothetical (and requries
observational verification) as soon as we assume that a» = const (i.e.,
is space- and time-invariant). In a purely formal manner, we may also
write &= f t '= i.e., we may use Eq. (6.25) to calculate the
third-order velocity moments occurring in Eqs. (6.15) and (6.17)
where is a numerical coefficient, which need not be the same as

[Eq. (6.26) is itself, of course, yet another semiempirical hypoth
esis]. Similarly, we may put instead o f t? and evaluate the
third-order moments in Eq. (6.15') as follows:
= -a ^K ^ (6.26')
aX;
where is a new dimensionless coefficient [see, for example, the

works o f Csanady (1967a; 1967b)].
Adopting the hypothetical relationships (6.22) and (6.26) we
introduce into the energy balance equation, instead of the variables
pWjtty and , the characteristics / (which can also be replaced by
t* = l/\/b) and b (or K = l V b ) ; i.e., we reduce the number of
unknowns quite considerably. The next step in this direction will be
to establish a connection between the rate o f energy dissipation e.t
which also occurs as an unknown in Eq. (6.17), and / and b (or / and
K, OT t* and fc). If we assume that there is such a connection, then its
form may be found uniquely by dimensional considerations
- <~ cH ~ c't*
^ cH* (6.27)
where c* is another nonnegative dimensionless parameter (the reason

why we have written it as the fourth power o f another number will

be seen in Sect. 6.6). The relationship (6.27) once again may ê
considered simply as the definition of a new quantity c , replacing ej
(assuming that K and / are given, for example, with the aid of =
p K and K — l V ^ ) - However, if we assume that c = const, then
Eq. (6.27) becomes at once an additional semiempirical hypothesis.
Analogously, we shall obtain a semiempirical balance equation for
the mean square concentration (or temperature) fluctuation if we
replace the expressions u'd^' ^nd in Eq. (6.15') by semiempiri
cal relationships (6.25) and (6.26') and use simultaneously the
following hypothetical equation for N :
A '2 u l/ 2 A 'l K
2N = ^ (6.27')
where is a new dimensionless constant [cf. Csanady (1961a;

1967b)].
6.4 Energy Balance in a Compressible Fluid

Taking into consideration the compressibility of the fluid, and, in
particular, the presence of density fluctuations p', a considerable
complication is introduced into t h ^ a b o w evaluations. The mean
momentum density is now equal to pa, = p«, + p«p where p«, = p' m,
is the mean momentum density o f the fluctuating motion; in the
absence of density fluctuations it becomes zero. Similarly, the mean
kinetic energy density here takes the form
E = \ P«,«. = j + pl^ X + T = ^st + •(6.28)
where £« is the energy density o f the mean motion,
= \ \ ( P « X + P ' « X )
is the energy density of the fluctuating motion, and £'^< = p«' • «„ =

p' m' • is an additional part of the fluctuating energy, connected
with the transport of the momentum by the mean motion.
In formulating the moment equations in a compressible fluid, we
must take as our starting point the general equation of continuity
(1.1) and the equations o f motion (1.3), which we rewrite here as
TT + / = 1, 2, 3. (6.29)
In Eqs. (6.29) 6/., as usual, denotes the unit tensor and is the
general viscous stress tensor, given by the equation
/ dui duj 2 \ rd U a . .
the remaining notation does not differ from that used in Eq. (1.3).
Averaging Eqs. (1.1) and (6.29), we obtain
(p«/+ p«;)+ ^ + p“ ^“ «+ p“ > .-+ p«i«^+ - 0,-j =
(6.31)
which in this case play the same role as the Reynolds equations (5.1)
for turbulence in an incompressible fluid. With the aid of Eq. (6.30)
and the equation
I ,7 I - Y 1 I •
dt dx^ + “ « P d x i ^ ^ dx^ ’
which is a corollary of Eq. (1.4) [and hence is applicable only when

the viscosity fluctuations are ignored], we may also find an
expression for the d e r i v a t i v e . We then obtain
T + ^ ( P ^ . + P ^ : + pK - = 9X, + cp,. (6.32)
where
is a quantity which becomes zero for an incompressible fluid.

Comparison o f Eqs. (6.32) and (6.31) shows that
Thus the <Pt have a clear physical significance: they describe the
exchange o f momentum between the mean and fluctuating motion.
From the dynamical equations (6.29) and the continuity equation
(1.1), it is easy to obtain also an equation for the tensor pUjMj, which
differs from the corresponding equation for an incompressible fluid
[Eq. (6.3), Sect. 6.1 ] only in the fact that an must now be taken to
mean the viscous stresses in a compressible fluid. In particular, the
kinetic energy density E — ^ in a compressible fluid will satisfy
the equation
W+ {£u,-\-pu, — MpO^p) = — pe, (6.35)
which outwardly differs from Eq. (6.4) only by the presence on the
right side of a further term p which describes the kinetic energy
variation due to compressions and rarefactions o f the fluid elements
(which are accompanied by changes in the internal energy). Taking
the mean value of all the terms o f this equation, we obtain an
equation for the mean kinetic energy density E. However, this
equation will not be formulated here but instead we shall write down
three equations for the three parts Esi and Et of the density E
dE, du.
dt
- - I 1 -------- ( ----- - du^
+ + (6.36)
dEs,
dt dx. p«:«p«p)=
- - , 1 - - ^P"a du„ Du
= u y x '- «a<Pa+2
^ + T P“ X “ ?+ P 'K — —
________ _______ du ( ______ dUo ____ D u 1
= - f , + p' - 1 p< » ; + K -o f)■
where
D d - d - - du, — ,
In these equations the expression in brackets describes the mutual

transformation of energies £« and Est and the expressions in braces
describe the mutual transformations of E, and £< or Egt and Et.
Equations (6.30)—(6.34) and (6.36) contain several new variables
which are not contained in the corresponding incompressible
equations. The most im portant of these new variables are the
quantities the components of the mean momentum density of
the fluctuating motion, or, which amounts to the same thing, the
mean density of the turbulent mass flux. If there is no mean
turbulent mass flux in the direction of the mean motion, then =
p/i' • z/, — 0 and the fluctuating energy will be described by Et only.
The most important difference between the general equation for Et
and the incompressible equation (6.15) consists of the presence of
the term
5 = (6.37)
which we shall now consider in somewhat greater detail.

In most turbulent flows encountered in practice (both in nature
and in technology), the main body force acting on the fluid and
performing work is the force of gravity. Taking the Oxs axis vertically
upward, we may write X i= Y i — g 6 i3, where S’ is the gravitational
acceleration and V, are the components of the acceleration of the
other body forces. However, since the gravitational acceleration is, in
general, considerably greater than the individual acceleration of the
particles in the mean motion and also the acceleration of all
other body forces Yt, then in the majority of problems we may write
(6.38)
The meaning o f this quantity may be explained by observing that in

the presence of density fluctuations q', a buoyancy force — p'g acts
on the turbulent elements, so that B is the mean work done by the
buoyancy force in turbulent displacements of fluid elements. Thus,

B describes the mutual transformation o f the kinetic energy of
turbulence and the potential energy of a vertical column of fluid of
variable density in a gravitational field. These mutual transformations
generally play a considerable role in the fluid with vertical density
stratification, for example, in the atmosphere with nonneutral
thermal stratification or in the sea with salinity stratification. If the
vertical stratification o f the fluid is stable, then the vertical
displacements o f the fluid elements are accompanied by consump
tion of energy in performing work against the buoyancy force, so
that B < 0; we note that for stable stratification, the density
decreases with height, and the density fluctuations and the vertical
velocity will be positively correlated. On the other hand, for
unstable stratification during vertical displacements of turbulent
elements, the buoyancy forces will perform work due to the
potential energy of stratification, leading to an increase of the
turbulent energy; in this case B > 0 (the correlation of u'^ and p 'is
negative).
For atmospheric turbulence and several other important cases, to
calculate we may use the Boussinesq equations and in accordance
with Eq. (1.73) assume that p '= — ppF, where p is the thermal
expansion coefficient of the medium. For definiteness, we shall
consider the case o f a gas, and, according to the equation of state of
an ideal gas, we shall assume that p = l / f , where T is the mean
temperature. In this case p « '« — T' /t' , i.e., the turbulent mass
flux proves to be proportional to the turbulent heat flux. Hence, we
obtain the following expression for B :
= = where q = Cp^T%. (6.39)

/ Cpl
To calculate B approximately, we can often use the semiempirical

equation
q = - CpPK^ - CpM
which together with Eq. (6.39) gives
B = —a - ^ p K - ^ . (6.40)
T 0X3
However, sometimes we must consider the fact that as a fluid

element moves vertically, its temperature T, generally speaking,
varies, i.e., it is not a strictly conservative quantity. But since the
radiant and molecular heat exchange between fluid elements is
almost always very small, the entropy or the potential temperature
T-i
functionally connected with it (see the closing remarks of Sect. 2.4)

may be assumed to be conservative. Due to the fact that the pressure
fluctuations are small in comparison with the temperature fluctua-
7’/ 0/
tion (which is the basis of the Boussinesq approximation) - y
h e n c e 6 ' « ' . T a k i n g this as our starting point, when Eq. (6.40)

is incorrect, due to the nonconservative nature o f the temperature
the semiempirical relationship (6.25) can usually be applied to the
potential temperature 0, writing
B= (6.40')
6 0 X3
(Here once again a is the ratio of the eddy diffusivities for tempera
ture and momentum.)
6.5 The Richardson Number and the Eddy Viscosity in a

Thermally Stratified Medium
In the preceding subsection we considered the energy balance
equation for an arbitrary compressible medium. Henceforth, how
ever, of all the effects connected with compressibility, we shall
consider only the mutual transformation o f kinetic energy and
potential energy of density stratification, while, in connection with
the Boussinesq approximation, the density will be assumed to
depend only on the temperature fluctuations (but not on pressure
fluctuations). Then the fluid may once again be assumed incompres
sible; i.e., the equation ^ ] ^ ~ 0 may be used; however, in the
equation for the vertical velocity, the buoyancy force must also be
taken into account, which in the case of a gaseous medium is written
in the form o f the additional term ~gPT' = — y T' on the right side
(cf. Eq. (1.75), Sect. 1.5). Completely analogous results may be
obtained for an incompressible fluid of variable density p; here it is

necessary only to replace — T' by p' in all subsequent equations.
Thus we shall assume that the force of gravity is the only external
body force, and we may proceed on the basis of the system of
dynamical equations in the Boussinesq approximation, i.e., we may
assume that satisfies Eq. (1.75) with p = l / f . In this case, again
repeating the deduction o f the energy balance equation, we obtain
dEj^
dt
1 -
where Ei — -^ P«p«3- This equation differs from its incompressible
form (6.15) only by the replacement of on the right side by
B= ^ T'u'^, the meaning of which was explained at the end o f the
previous subsection. [If we consider the corresponding equations of
the form (6.19) for the “partial energies” £■,.E j.a n d £ 3 , then the
term B appears in the third of the equations.] Equation (6.41) is
applicable, in particular, to atmospheric turbulence, since the air
motions in the atmosphere are described usually by Boussinesq
equations to high accuracy.
We now note that in developed turbulence, the viscous stresses are
negligibly small in comparison with the turbulent Reynolds stresses
(excluding the viscous sublayer in immediate proximity to the wall,
which we shall not consider here). Thus it is natural to assume that
the energy transfer due to the viscous forces, i.e., disordered
molecular motions, is very small in comparison with the energy
transfer by turbulent velocity fluctuations, i.e., that the last term in
parentheses on the left side of Eq. (6.41) is negligibly small in
comparison with the second term. Let us suppose that the turbulence
is homogeneous in the direction of the Oxi and 0x2 axes. In this case,
all the statistical characteristics of turbulence will depend only on Xs
while, by the continuity equation ~ = 0, i.e., W3 = 0 . In addition to
the notations:,, Ui for the coordinates and the velocities, we shall also
use the notation x, y, z and «, v, w; then ignoring the small term
, we may write Eq. (6.41) in the form:
db 1 d u 'u 'w ' \ dp w — - du — — - dv g —— _

_ = — ------ — —7-------- u'w' ------- v 'w ' — ~>r— T'w' — t,
(6.42)
where « '« ' = v ' ^ w ' ^ , and = -i « '« ' is the turbulence
intensity (i.e., the mean kinetic energy of the fluctuations per unit
mass of the fluid).
If the mean velocity of flow has everywhere the same direction
(e.g., along the Ox axis), then Eq. (6.42) becomes even more
simplified; here
dt
The latter equation may also be written as
where
g J ’w' _ g_ q_
T dll CpT du
The dimensionless quantity Rf clearly determines the relative role of

buoyancy in the generation of turbulent energy, by comparison with
the dynamic factors (transfer o f enetgy from the mean motion); it is
called the flux Richardson number. It is clear that Rf < 0 for 0
(i.e., with unstable thermal stratification); R f> 0 fo r^ < 0 (stable
stratification); for neutral stratification Rf = 0.
In Eq. (6.43), we have already ignored the viscous flux of energy.
Also, it is often assumed that the transfer of energy by the pressure
forces is srrtali in comparison with its transfer by the velocity
fluctuations, and, hence, in the energy balance equation one often
ignores terms containing pressure fTuctuations. inis auppoeition
seems quite plausible but it has no strict justification. In the case of a

steady flow without any heat transfer (i.e., with neutral thermal
stratification) close to a plane wall, a logarithmic layer is formed,
within which the turbulence intensity b is constant (equal to cul,
where 5 in accordance with the data of Fig. 26, Sect. 5.3), and
the diffusion terms and p'w'lp both differ from m2 only by a
constant multiplier according to dimensional arguments. Hence the
energy transfer in this layer is identically equal to zero, while in Eq.
(6.44) only the terms — u'w' and — e,,which must compensate
each other, are nonzero. However, outside the logarithmic layer, and
in the presence o f a turbulent heat flux, the diffusion terms may in
come cases be comparable in value with other terms (see below, end
of Sect. 8.5). Nevertheless, they are more often than not ignored,
since these terms are usually supposed to be not too large, and,
principally, because they are almost always unknown. Then the
energy balance equation takes the comparatively simple form
db 1 d u A iw ---- da
^ - 2^ ^ (1 - Rf) - e,. (6.46)
Sometimes instead o f dropping the diffusion term, it is assumed that

this term is proportional to y T'w' [because the two become zero
together; see, for example, Klug, (1963), Takeuchi and Yokoyama
(1963)]. In this case, instead of Eq. (6.46) we obtain
— u 'w '-^ (\ — — (6.46')
which has as simple a form, but which contains the additional

dimensionless parameter o; this allows better agreement with the
data to be obtained.
Let us begin with the consideration of Eq. (6.46). Since t t > 0
always, and — u'w' > 0 practically always, then, by Eq. (6.46)
stationary (undamped) turbulence is possible only if
Rf<l. (6.47)
Equation (6.47) is in fact the criterion for transition to turbulence in

a stratified medium, obtained using the energy balance of the
disturbances (see Sect. 2.9). Like the other criteria obtained by the
energy method, it is apparently fairly rough, i.e., it gives an elevated
value of the critical Richardson number. Thus, on this basis, we can
only assert that stationary turbulence is possible for R f< ^ , where
/? = Rfcr is likely to be less than unity (in all probability, considerably
less). In a purely formal manner, the value \IR= 1/Rfcr may be
identified with that of the parameter a in Eq. (6.46'); with such a
choice o f a, the energy criterion obtained from this latter equation
will give the true value of Rfcr-
Now applying the semiempirical relationships (5.5), (6.26), (6.27),
and (6.40) and ignoring the term containing pressure fluctuations, we
may transform Eq. (6.44) into the following “semiempirical energy
balance equation” :
db \ ^ r, k ' f6 48'l
Here K is the eddy viscosity i.e., eddy diffusivity for momentum,

and a K = K r andab/(=/(6 are eddy diffusivities for heat and energy,
respectively. Ignoring the energy diffusion (i.e., the last term on the
right side), we obtain the following criterion [which is equivalent to
Eq. (6.47)] for the possibility of existence of undamped turbulence
*= (6.49)
Here
(6.50)
T (du /dzy a
is the ordinary Richardson number [see Eq. (2.3') at the end of Sect.
2.5 in which, instead of was used in accordance
with Eq. (6.40') which is somewhat more exact than Eq. (6.40).
The criterion (6.49) was obtained first by Richardson (1920) [on the
assumption that a = 1]; it shows that Ricr < but does not define
the value of R i c r more precisely. In Sect. 2.8 we have already seen
that according to linear stability theory, a flow of stratified fluid will
be stable to infinitely small disturbances if at all points of it R i> x ;
consequently, R icrm in<-|-; o f course Ricmiinhere plays the same

role as R ec rm a x in the usual hydrodynamic stability theory since
large values of Ri correspond to stability of a flow. On the other
hand, however, there is some justification for assuming that perhaps
a ^ 0 as R f-» - Rfcr, and that turbulence may exist for as large a value
o f Ri as desired (so that Ricr = oo, see below, Sects. 7.3 and 8.2).
Since by definition f ( - ^ = ^ = u \ Eq. (6.48), for steady
conditions and ignoring the diffusion of turbulent energy, may be
put in the form
(1 — a Ri) = - ^ (1 — Rf) = .
Consequently, for the eddy viscosity K we obtain the equation
— (6.51)
According to Eq. (6.51) it seems that K would become zero (i.e.,

turbulent exchange would cease) for R f = 1, that is, Rfcr = 1, Ricr =
1/ a , which contradicts what has been said above. However, use o f the
semiempricial relationship (6.27) with c = const for Rf close to Rfcr
becomes meaningless, and therefore Eq. (6.51) will also no longer be
applicable. We now note that in the absence of thermal stratification
within the logarithmic layer cl=y.z (since by Eq. (6.48) here
c/ = whi l e K = û^z, - 0 - — ^ ) ; thus for arbitrary
stratification, it is convenient to write
c/r=-/2X,(Rf), A:=tt*xzXi(Rf)(l - R f ) ' \ (6.52)
The function Xi (Rf) must satisfy the conditions A,t(0)= 1 and

Xi(R)= 0, where R = Rfcr. To emphasize the fact that Ki(R)= 0,
instead of ( R 0 > we may introduce into our discussion the function
X(Rf) such that
= «= (6.53)
with the aid of this new function the expression for K is written in
the form
A r= tt* x 2 X (R f)(l-o R f)'/\ (6.54)
which is analogous to Eq. (6.52) but with the change of Rf into aRf,
a = l / R , i n the last set o f parentheses. The same result (6.54) may be
obtained if we assume that cl=xzK (Rf). However, instead o f ignoring
the diffusion of turbulent energy we consider this diffusion to be
proportional to -i- T'w'; i.e., instead of Eq. (6.46) we use Eq. (6.46')
with a= l/R . In Eq. (6.54) we must suppose that X(0) = 1, but now
we have automatically /C (/?)=0 and therefore we may assume that
A ,(R 0 has no zeros. In Sect. 7.4 we shall see, in particular, that results
that are reasonable in many respects may be obtained if we simply
put X (R O = 1. It should be stressed, however, that this choice of the
function A(RO has no theoretical justification whatsoever and it is
completely permissible to change it in order to obtain a better
agreement with experiment; we shall discuss this in greater detail in
the following section.
The balance equation for the mean square temperature fluctua
tions in a plane-parallel, thermally stratified flow contains no
additional terms and may be written as
2 w'T' — - 2N (6.55)
dt dz\ } dz
[the terms X 2 T ' R ' of Eq. (6.15') are ignored in Eq.

(6.55)]. With the aid o f the relationships (6.25), (6.26'), and (6.27')
we may transform the latter equation into the “semiempirical
equation o f the mean square temperature fluctuation balance” o f the
form
6.6 Turbulence in the Planetary Boundary Layer of

the Atmosphere
H ere an d in th e fo llo w in g s u b s e c tio n w e shall d isc u ss tw o e x a m p le s o f th e use o f th e
s e m iem p ir ic a l e q u a t io n fo r th e tu r b u le n t en e rg y b a la n c e in a c tu a l p r o b le m s , w h ic h illu strate
th e t y p ic a l fe a tu re s o f th e s e m iem p ir ic a l a p p ro a c h t o t h e s tu d y o f tu r b u le n c e .
A s ou r first e x a m p le , le t us c o n sid e r th e p r o b le m o f th e b o u n d a r y layer s et up in th e

a tm o sp h e r e c lo se t o th e earth d u e t o c o m b in e d a c tio n o f fr ic tio n o f th e air o n th e
u n d e r ly in g surface a n d o f th e C oriolis fo r c e c a u sed b y r o ta tio n o f th e earth. T h e la yer in
w h ic h th e fr ic tio n a l fo r c e s a pp ear t o b e s ig n ifica n t is c a lled th e planetary boundary layer (or
th e fr ic t io n layer, or th e Ekman layer). We shall c o n s id e r o n ly th e p la n eta ry b o u n d a r y layer
over a p la n e h o m o g e n e o u s u n d e r ly in g s u r fa c e , w h ic h w e shall ta k e as th e p la n e z = 0 , fo r
s te a d y e x te r n a l c o n d it io n s an d shall, as a r u le, assu m e th a t th e th er m a l s tr a tific a tio n m a y b e
a ssu m ed n e u tral. In a d d itio n , w e u se th e fa c t th a t w ith in th e p la n eta ry b o u n d a r y lay e r w e
m ay put c o n s t; th u s, th e c o m p r e ss ib ility o f th e air fo r th is p r o b le m is n eg lig ib le . S in ce
all th e sta tistic a l c h a r acteristics o f tu r b u le n c e in th is lay e r d e p e n d o n ly o n z , w e m a y u se th e
fo r m ( 6 . 4 2 ) o f th e en e rg y b a la n c e e q u a t io n . In th is e q u a t io n , w e m a y ig n o r e th e term o n th e
g ______
le ft (d u e t o s ta tio n a r ity ) a n d t h e te r m T ' w ' on the right side, s in c e w e h a v e assu m ed th e
1
s tr a tific a tio n t o b e neu tral; m o r e o v e r , as usual w e also ig n o re th e term — — ------ w h ic h
p oz
descr ib e s t h e transfer o f tu r b u le n t e n ergy d u e t o pressure flu c tu a tio n s ; w e m a y a lso a ssu m e
th a t th is te r m is in c lu d e d in th e rig h t sid e o f E q . ( 6 . 5 7 ) b e lo w w ith t h e aid o f a ch an ge o f
th e c o e f f ic ie n t . A s a resu lt, w e o b ta in
____ du ____ dv ^ \ d u 'u 'w *

’ (6 .5 6 )
w h ic h , a fter a p p lic a tio n o f th e sem iem p ir ic a l h y p o t h e s e s ( 6 . 2 2 ) , ( 6 . 2 6 ) an d ( 6 . 2 7 ) [w ith

= c o n s t ] , an d th e d iv isio n o f all term s b y p , ta k es th e fo r m
T h is e q u a t io n m u st b e c o n s id e r e d to g e th e r w ith th e R e y n o ld s e q u a t io n s ( 6 . 5 ) w h ic h in th is
case m a y b e w r itte n as
dou,xv d p _
1 - 1 . 2, 3. (6.i8)
(We ig n o r e th e v is c o u s vertical m o m e n tu m flu x in c o m p a r is o n w ith th e tu rb u le n t

m o m e n t u m f lu x pw'a;'.) T h e e x te r n a l fo r c e s X i a c tin g in th e b o u n d a r y la y e r are, first, th e
gravity fo r c e (v e r tic a lly d o w n w a r d s ) an d , s e c o n d , th e C o r io lis fo r c e Y = 2 u X «o (w h ic h is
p e r p en d icu lar t o th e v e lo c ity u an d h e n c e d o e s n o w o r k ), w h e r e w is th e angular v e lo c ity
v e c t o r o f th e r o ta tio n o f th e e arth an d th e sign X s y m b o liz e s th e v e c to r p r o d u c t. W e n o w
s e le c t t h e a x e s O x i —_Ox an d 0 x 2 —Oy in su c h a w a y th a t Ox is parallel t o th e surface
w in d . T h e n X\ a i ^ X 2 w ill b e ^ u d t o th e c o n e s p o n d in g c o m p o n e n t s o f th e C o r io lis
a c c e le r a tio n , i.e ., X i = ^ 1 = 2 a)zV, X 2 ™ 5^2= — 2 d)zU, w h e r e coz = wsincp is th e vertical
c o m p o n e n t o f th e v e c to r to, and (p is a la titu d e o f th e o b s e r v a tio n p o in t. U sin g th e
s em iem p ir ic a l e q u a t io n ( 6 . 2 2 ) t o e v a lu a te , a n d d iv id in g t h e R e y n o ld s e q u a t io n s b y p ,
w e m a y w r ite th e first t w o o f t h e s e e q u a t io n s (w h ic h are all th a t are n e e d e d ) in th e fo r m
Io “ I
- K
dp
- —
d
=
dv p, - —
\ dp
K - -------- ( 6 . 5 9 )
dz dz p dx dz dz p dy
T h e h o r iz o n ta l pressure gr a d ie n t j p r o b le m w ill b e g iven and w e shall

assu m e it to b e in d e p e n d e n t o f th e c o o r d in a te s x , y, z; th e z -in d e p e n d e n c e o f th e
pressure grad ien t is th e usua l a s su m p tio n o f b o u n d a r y -la y e r th e o r y . O u tsid e th e fr ic tio n
layer th e first te rm s in E qs. ( 6 . 5 9 ) [w h ic h d e scr ib e th e fr ic tio n a l fo r c e ] m a y b e ig n o r e d , and
th e m o t io n w ill b e d e te r m in e d b y th e “ g e o str o p h ic w in d e q u a t io n s ”
u — ------- -- ------ = G c o s a, V = ------------- ^ = — G s in a, (6 .6 0 )

2po)^ d y 2po)^ d x
w h ere
G= - i -
2po)^
is t h e g e o s t r o p h ic w in d v e lo c ity and a is th e angle b e t w e e n th e v e c to r w ith c o m p o n e n ts
^------^ and th e surface w in d (th e angle o f to t a l w in d r o ta tio n in th e p lan etary
b o u n d a r y lay e r ) w h ic h is su b je c t t o d e te r m in a tio n . _
_ E q u a tio n s ( 6 . 5 7 ) and ( 6 . 5 9 ) p e r m it us to d e te r m in e th e d is tr ib u tio n o f th e w in d u{ z) ,
v( z) in th e fr ic tio n layer i f th e e d d y v is c o s ity K ( z ) is giv en as a fu n c tio n o f z or i f th e
fu n c tio n l ( z) is a ssu m ed to b e giv en in s o m e w a y [th is last a s su m p tio n is a d v a n ta g e o u s in
s o m e r esp e c ts relative t o th e a s su m p tio n o n th e fo r m o f K ( z ) ] . D iffe r e n t a s su m p tio n s o f
b o t h th e s e ty p e s have b e e n used in a n u m b e r o f w o r k s; for e x a m p le , b y M on in ( 1 9 5 0 a ) ,
B lackadar ( 1 9 6 2 ) , L etta u ( 1 9 6 2 ) and A p p le b y and O h m ste d e ( 1 9 6 5 ) , b a se d o n d iffe r e n t
a s su m p tio n s a b o u t / ( z ) , and th e d e ta ile d surveys b y Z ilitin k e v ic h , L a y k h tm a n and M onin
( 1 9 6 7 ) and Z ilitin k e v ic h ( 1 9 7 0 ) , c o n ta in in g e x te n s iv e b ib lio g r a p h ie s. L et u s b e g in b y
e stim a tin g th e v a riation w ith h e ig h t in th e a tm o sp h e r ic surface la yer o f th e ;c -c o m p o n e n t
of tu r b u le n t shear stress x^ = p /C -^ = pw*. Integratin g th e first e q u a t io n o f E q.
( 6 . 5 9 ) w ith r esp e c t to h e ig h t, w e h ave
//o _ //o
ul (0) — ul ( H q) =
J
I (
\
— -L — ) d z <
^ dx! J
f I— ~
\ ^ dxj
d z < 2o>JioG.
H ere w e use th e fa c t th a t o m it tin g th e term 2o)zV leads t o th e s tr en g th e n in g o f th e

in e q u a lity , sin ce th e C o r io lis fo r c e partially c o m p e n s a te s th e a c tio n o f th e pressure gradient.
We m a y s ele c t Ho i n su c h a w a y th a t th e relative varia tion o f u i in th e layer o f th ic k n e ss Ho
d o e s n o t e x c e e d th e to le r a n c e a, i.e ., so th a t th e c o n d it io n
ul(Q)
w ill b e fu lfille d . For th is, it is s u ffic ie n t to req uire th a t th e fo llo w in g in e q u a lity
a iii (0)
H. < (6.61,
b e satisfie d . A c c o r d in g to th e data lu J G is o f th e order o f 0 . 0 5 . M oreov er, for m o d e r a te

u- (0)
la titu d e s , 2o)2 ^ 10'^ sec~^, so th a t fo r G ^ 1 0 m /s e c , w e o b t a i n ^ 2 5 0 m . W ith a
2o)^G
to le r a n c e a = 20% , w e o b ta in an e stim a te o f th e th ic k n e ss o f th e surface layer 50 m.
W ithin th is lay er w e m a y ig n o r e th e a c tio n o f th e C oriolis fo r c e , and h e n c e th e r o ta tio n o f

th e w in d w ith h e ig h t c a u sed b y th is fo r c e , and a ssu m e to b e c o n s ta n t, i.e ., w e m a y use
th e o rd in a ry lo g a r ith m ic e q u a t io n for th e w in d v e lo c it y . C o n s e q u e n tly , in th is layer
K = = = 6 = Ci“ * = const,
w h e r e Ci is a c o n s ta n t c o e f f ic ie n t w h ic h m a y b e e s t im a te d , for e x a m p le , fro m th e data o f

Fig. 2 6 , a n d , fin a lly , b y E q. ( 6 .5 7 )
Cl = xz, ^ ^ (6 .6 2 )
We n o w tu rn t o E qs. ( 6 . 5 9 ) . D iffe r e n tia tin g th e s e e q u a tio n s w ith r esp e c t to z , w e m ay

w r ite th e resu lt in th e fo r m o f a single e q u a tio n :
= 2/(0^ ^ , ( 6 .6 3 )
/ ^11 dif
w here f = K 4“ ^ j ^ c o m p le x q u a n tity ch aracterizin g b o t h c o m p o n e n ts o f th e
shear stress (d iv id e d b y th e d e n s ity p ). In th e surface layer / ta k es th e value u l. F u rth er, w ith

th e aid o f / , th e tu r b u le n t e n e rg y b a la n c e e q u a tio n ( 6 . 5 7 ) m a y b e w r itte n as
l/l^ (6 64)
~K "dz^dzl^' ^ ’
In a d d itio n to t h e p aram eter w,, w h ic h d e te r m in e s th e r egim e o f tu r b u le n c e in th e surface
lay er o f th e a tm o sp h e r e , E qs. ( 6 . 6 3 ) and ( 6 . 6 4 ) also c o n ta in th e C oriolis p aram eter 2o)z.
F r o m th e s e t w o p aram eters w e m a y fo r m u la te th e le n g th scale
(6 .6 5 )
2o)^
(T h e v o n Karmdn c o n s ta n t x is ad d ed here for c o n v e n ie n c e .)
U sin g d im e n sio n a l c o n s id e r a tio n s , w e m a y n o w w r ite
(6 .66)
w h e r e th e d im e n sio n le ss fu n c tio n s f (C ), ^ ( C ) . req uire fu rth er d e fin itio n , w h ile for

£ = 0 th e y are all equal to u n ity . A fte r th e s u b s titu tio n o f t h e r e la tio n sh ip s ( 6 . 6 6 ) in t o E qs.
( 6 . 6 3 ) - ( 6 . 6 4 ) , w e o b ta in t w o e q u a tio n s w ith th re e u n k n o w n s ; h e n c e a fu rther e q u a tio n is
n e e d e d t o c lo se th e s y s te m . In th e a b se n c e o f reliable da ta o n h e ig h t-d e p e n d e n c e o f th e
le n g th scale /, M o n in ( 1 9 5 0 a ) has a ssu m ed , for s im p lic ity , th at 1. U n d e r th is
a s su m p tio n Eqs. ( 6 . 6 3 ) - ( 6 . 6 4 ) [after s u b s titu tio n fr o m E q. ( 6 . 6 6 ) ] ta k e th e form
= (6 .6 7 )
w h e r e d=--auK-c^ is a n u m b e r o f th e order o f u n ity . T h e fu n c tio n s F a n d w h ic h o c c u r in
th is e q u a t io n m u st, b y r ea so n o f th eir p h y sic a l m e a n in g , te n d to zero as C ^ - > co..

n
It m a y b e s h o w n th a t th e s y s te m ( 6 . 6 7 ) has a u n iq u e d a m p e d s o lu tio n i f 6 > 0 . H o w e v er ,

fo r 6 = 0, it has n o s o lu tio n s w h ic h are d a m p e d as S -> cx>, b u t a s o lu tio n d o e s e x is t w h ic h
te n d s t o z er o as i te n d s t o s o m e fin ite v a lu e. In o th e r w o r d s, th e tu r b u le n t b o u n d a r y la y er
in th is case p o s se s se s a fin ite th ic k n e ss . E q u a tio n s ( 6 . 6 7 ) w e re in teg r a te d n u m e ric a lly b y
M o n in fo r 6 = 0 and 6 = 0 .1 6 . F o r 6 = 0 , th e th ic k n e ss o f th e tu rb u le n t la y er is c lo se
t o 1 0 / / , i.e ., it is v ery large. W ith z < 2 / / th e f u n c tio n s F( l ) a n d (D (S ) fo r 6 = 0 a nd fo r
6 = are v ery c lo s e t o e a ch o th er . ____
W hen f (S ) h a s b e e n d e te r m in e d , th e v e lo c ity v e c to r o f th e w in d (w, v) a t d iffe r e n t
he ig h ts m a y b e fo u n d fr o m E qs. ( 6 . 5 9 ) , w h ic h fo r th is p u rp o se m a y b e w r itte n c o n v e n ie n tly
in th e fo r m
(6.68)
T h e param eters w* a n d a are u n k n o w n ; w e recall th atw * also o c cu rs in th e e q u a tio n fo r th e

s c a l e / / . T o d e te r m in e th e m , w e u se th e fa c t th a t th e w in d v e lo c ity b e c o m e s z er o at so m e
h e ig h t zo ( “ ro u g h n ess h e ig h t”) w h ic h m a y b e a ssu m ed k n o w n . T h e n , eq u a tin g E q. ( 6 . 6 8 ) for
z = 2 o t o ? ero , w e o b ta in
<f (Co) = (6.69)
w h e r e Co = ^ and w e k n o w n . T h e fu n c t io n cp(So) m a y b e a ssu m ed k n o w n ,
sin c e /^ (£ ) is a lrea d y d e te r m in e d . C o n s e q u e n tly ,
a = ^ -a r g /^ '(C o ). ( 6 .7 0 )
M oreov er, s in c e il = |(p (£ o )l, £ o = ^ ( n ) > w h e r e ^ is th e inv erse fu n c tio n o f |(p|. T h e latter
r e la tio n s h ip m a y b e r e w r itte n as
(6 .7 1 )
a (i)>
w h ic h c o m p le te s th e d e te r m in a tio n o f th e u n k n o v m para m eters. T h is d e te r m in a tio n is
sim p lifie d c o n s id e r a b ly b y th e fa c t th a t th e ro u g h n ess h e ig h t zq is u su a lly sm all e n o u g h to
m a k e it p o s sib le t o p u t K( zq) = ku^Zq, i.e ., (fo) = 1* T h e r e fo r e th e f u n c tio n F ( f ) can b e
fo u n d fo r f o f th e ord er f o fr o m th e first e q u a t io n ( 6 . 6 7 ) w ith <D = 1, w h ic h im p lie s,
to g e th e r w ith th e b o u n d a r y c o n d it io n F( 0) = 1 , th a t F = 1 + a t + /^ log f + . . . , i.e ., th a t
C M ^ ^ lo g K fo + b w h e r e b = A + i B is 2l c o m p le x c o n s ta n t [c f. K a za n sk iy a n d M onin
( 1 9 6 1 ) ] . S in c e rjfo = w h e r e % = u J G y E q. ( 6 . 6 9 ) m a y n o w b e r e w r itte n in th e
fo r m
1/2
lo g J L = B - \o t( * ( — - >42^ . 8 in a = (6 .7 2 )
y.2 / n.
w h e r e f = u J G , T h e c o n s ta n t b is fo u n d b y so lv in g th e c o m p le te e q u a tio n s ( 6 . 6 7 ) in th e
e n tir e b o u n d a r y la y er. W ith 6 = 0 , w e o b t a in th e v a lu e 6 = - 1 . 6 9 + 2 .4 9 /; w ith 6 = 0 . 1 6 ,
w e o b ta in 6 = - 1 . 8 1 + 1 .7 1 i.
We m u st fu rth er n o t e th a t w h e n w e a p p r o x im a te th e w in d d is tr ib u tio n th r o u g h o u t th e
e n tir e fr ic tio n la y er it is n e c essa ry to ta k e th e v a lu es o f th e “ ro u g h n ess p a r a m e te r ” zq
c o n sid e ra b ly larger th an th o s e u se d in th e a p p r o x im a tio n o f th e w in d p r o file in th e surface
lay er. T h is is n a tural s in c e as th e vertica l sca les o f th e p h e n o m e n o n u n d er c o n s id e r a tio n
in crea se, it is n e c essa ry t o in crea se th e h o r iz o n ta l sca les a ls o , a n d , c o n s e q u e n tly , to
in tr o d u c e a “r o u g h n e s s ” c h a ra cteristic o f large areas o f th e u n d e rly in g su rface.
T a k in g th is rem ark in t o a c c o u n t, c o m p a r iso n o f th e c a lc u la tio n u sin g E q . ( 6 . 7 2 ) w ith th e
d ata o f L etta u ( 1 9 6 2 ) , B lackadar ( 1 9 6 2 ; 1 9 6 7 ) , B y so v a an d M a sh k o v a ( 1 9 6 6 ) , and
K u rp a k o v a a n d O r le n k o ( 1 9 6 7 ) s h o w e d th a t th e v a lu e s o f ? and a c a lc u la te d fo r 6 = 0 are
clearly in d isa g r ee m e n t w ith th e e x p e r im e n ta l v a lues. H o w e v e r , w ith 6 = 0 .1 6 th e y are in
q u a lita tiv e a g r e e m e n t w ith e x p e r im e n t (se e F ig. 4 6 , C urve 2 , w h e r e th e Rossby number R o
= r?/y.^ = GjliOzZo is u se d in s te a d o f th e para m eter 17).
V e ry sim ilar r esu lts w e re a lso o b ta in e d b y m a n y o th e r s , u sin g slig h tly d iffe r e n t
sem iem p irica l h y p o th e s e s . F o r e x a m p le , B lin o v a and K ib el ( 1 9 3 7 ) u se d th e a s su m p tio n th a t
K {z ) = fo r all z (i.e ., th a t (f) = l ) ; th eir r esu lts are r ep re sen te d in F ig . 4 6 as C urve 1.
B lackadar ( 1 9 6 2 ) used th e sam e a p p ro a ch as M o n in ( 1 9 5 0 a ) b u t ig n o r e d th e vertica l
d iffu s io n o f tu r b u le n t e n e rg y (i.e ., h e a ssu m ed th a t a& = 6 = 0 ). H o w e v e r , u n lik e M o n in , in
a c co r d a n ce w ith th e fa irly rou g h da ta o f L e tta u ( 1 9 5 0 ) , B lackadar a ssu m ed th a t th e scale /
in crea ses lin ea rly w ith h e ig h t o n ly in th e lo w e s t la y e r o f th e a tm o sp h e r e , b u t th a t its rate o f
increa se th e n falls, a n d , as 00 , / te n d s a s y m p to t ic a lly t o s o m e c o n s ta n t le n g th Xjc w h ic h
h e c h o s e in th e fo r m \ T h e f u n c tio n cl (z) w a s th e r e fo r e c h o s e n in th e fo r m
cliz) =
1 f I +v.2coz2//3G
T h e n u m e ric a l para m eter p w a s c h o s e n so th a t th e v a lu e o b ta in e d fo r th e ang le o f r o ta tio n

c o in c id e d w ith th e o b se rv a tio n a l d a ta o b ta in e d at B r o o k h a v e n (lo c a te d near N e w Y o r k ). It
w as fo u n d th a t th is c o n d it io n w a s s a tisfie d b y p== 0 . 0 0 0 2 7 . A t Ihe sam e tim e , L e tta u
( 1 9 6 2 ) has c o n sid e re d a n o th e r variant o f th e s e m iem p ir ic a l th e o r y ; h e also ig n o r e d th e
v ertica l en e rg y d if fu s io n (i.e ., s u p p o s e d th a t = 0 ), b u t d e fin e d th e e d d y v is c o s ity K w ith
th e a id o f th e e q u a tio n
n i / 2
[sim ilar to P rand tl’s fo r m u la ( 5 .1 1 2 ) 1 an d t o o k th e le n g th scale / in th e fo r m
X2
/ =
1 + 33. 63
L et us a lso refer t o th e w o r k o f A p p le b y and O h m ste d e ( 1 9 6 5 ) w h o u sed an e q u a tio n o f th e

fo r m
1 - exp
(-?)
to g e th e r w ith th e P ra n d tl-L ettau e q u a tio n for th e e d d y v is c o s ity K , and c o n sid e re d
2 (1) ^X/ G to be a d e fin ite fu n c tio n o f th e R o ss b y num ber R o. F in a lly , B o b y le v a ,
Ziuiuirttcîch and L a y k h tm a n ( 1 9 6 7 ) h ave ta k e n in t o a c c o u n t th e in flu e n c e o f th e th erm a l
s tr a tific a tio n o f th e a tm o sp h e i^ a d d e d th e te r m - l a K ^ o n th e U f t s l d i o f
r 9i
o
I « g
« °-g
S "
tt!
o ^ SN
Q: ^ 2
• v t .ap
fe
^ ^
s>. :S2 ’PQ§
||o
I s 'i
5 fj
I« ?^ W
2
‘="P; 8
•S >»
c o «
• S ||
-M c« 45
2 o
•a r2
1^1
O C ^ .
^ ed « o
^ &2
s^ a§ ^
< +II
VOCo
"« a. IaI ^ o“
-’O "S
«X <N Ia.I
l l l l 1^ /mf sa ^ o
J •Sf ^ ^ 2
fe § § ^
1 l l 3* - • 0«« ■ ^ I “ '.I,
•E|||ig 2o^S2 f if
^ Its
-^3 D
^ ^ "a."a1. Ia.+
^ ^ ^ 1 "a.+
•=i+
"3. / a MM*
BBi^3 • \0 | ! | _Tts VO
ON
1 1• ^
lCs. !o° io" ;>»o
w
V t | :t|
S 2 rJ »
i tJ ^§2 «
/ f>
u K 7- * sS &g
§I
u. rt 'S .3
a*|o
410
E q. ( 6 .5 7 ) ; th e le n g th scale / th e y d e fin e d w ith th e aid o f a sp ecia l m o d ific a tio n o f th e v o n

Kdrmdn h y p o th e s e s ( 5 . 1 1 9 ) [th is m o d ific a tio n w ill b e d isc u sse d in S e c t. 7 .4 in c o n n e c tio n
w ith Z ilitin k e v ic h and L a y k h tm a n ( 1 9 6 5 ) ] . T h e r esu lts o b ta in e d w e re d e p e n d e n t, in ad d i
tio n t o th e R o ss b y num ber R o , o n th e n o n d im e n s io n a l str a tific a tio n p a r a m e te r ju =
-r? (T h is n e w p a ra m eter is th e r a tio o f th e le n g th scale H o f E q. ( 6 . 6 5 ) to
a n o th e r le n g th L = -c^pTu%/Kgq w h ic h w ill p la y a very im p o r ta n t part in th e fo llo w in g

s e c tio n .)
A s w e ha v e s e e n , i f th e la w o f va ria tio n o f scale / w ith h e ig h t is k n o w n , fr o m th e
R e y n o ld s e q u a tio n s an d th e e n e rg y b a la n c e e q u a tio n w e can o b ta in a c lo s e d s y ste m o f th ree
e q u a tio n s in th re e u n k n o w n fu n c tio n s ; th e fa c t th a t t w o o f th e s e f u n c tio n s w e r e c o m b in e d
a b o v e in t o a single c o m p le x fu n c tio n F , d o e s n o t , o f c o u r se , m a k e a n y d iffe r e n c e . T h e
c o rr e sp o n d in g e q u a tio n s , w h ic h w e re w r itte n in d iffe r e n t w o r k s in s o m e w h a t d iffe r e n t
d im e n sio n le ss fo r m s, w e re so lv ed n u m e ric a lly b y all th e a u th o r s c ite d a b o v e . S o m e o f th e
resu lts o b ta in e d are c o lle c t e d in F ig . 4 6 to g e th e r w ith th e c o rr e sp o n d in g da ta ; th e sca tter o f
th e e x p e r im e n ta l p o in ts is q u ite str o n g , b u t th e g en eral a g r ee m e n t b e t w e e n th e th e o r e tic a l
p r e d ic tio n s and th e da ta can b e c o n sid e re d as sa tisfa c to r y .
T h e m ain d e d u c t io n s fr o m th e p r e ce d in g a n a ly sis are th e gen era l e q u a t io n s ( 6 .7 2 ) ; th e
particular sem iem p ir ic a l th e o r ie s are n e e d e d o n ly for th e d e te r m in a tio n o f th e u n k n o w n
c o n s ta n ts A and B. (W hen th e th er m a l str a tific a tio n is ta k en in to a c c o u n t th ese c o n s ta n ts
b e c o m e fu n c tio n s o f th e sta b ility param eter /i; s ee , for e x a m p le , Z ilitin k e v ic h ( 1 9 7 0 ) w h ere
th e q u e s tio n o f th e d e p e n d e n c e o n ju is d isc u sse d in d e ta il.) F o llo w in g C sa n a d y ( 1 9 6 7 c ) , le t
us n o w s h o w th a t th e s e e q u a tio n s ca n e a sily b e o b ta in e d fr o m q u ite s im p le d im e n sio n a l
r ea so n in g . L et us d iv id e th e tu rb u le n t p la n eta ry b o u n d a r y lay er in t o th e “ w a ll la y e r ” (w h ere
th e C o rio lis fo r c e an d th e pressure g ra d ien t h a v e n o in f lu e n c e o n th e flo w ) an d th e “ o u ter
la y e r ” (w h e r e th e m o le c u la r v is c o s ity a nd th e su rfa ce p a ram eters are u n im p o r ta n t). In th e
w a ll la y er th e usual la w o f th e w a ll w ill b e valid w h ic h , fo r a d y n a m ic a lly r o u g h w a ll, can b e
w r itte n in th e fo r m
id
w h ere is th e r o u g h n ess pa ra m eter. In th e o u te r la yer it is r ea so n a b le to s u p p o s e th a t th e
v e lo c ity d e f e c t la w m u st b e valid; th is la w m u st clea rly h a v e th e fo r m
Ge
H ere Gc"'® is th e c o m p le x v a lu e o f th e g e o str o p h ic (o r “ fre e-stre a m ” ) w in d v e lo c ity an d th e

arg u m e n t o f th e c o m p le x fu n c tio n o n th e right side is in fa c t th e r a tio o f th e h e ig h t z to
th e c h a ra cteristic le n g th scale H o f E q. ( 6 . 6 5 ) w h ic h d e te r m in e s th e th ic k n e ss o f o u r tu rb u
le n t b o u n d a r y la y er; th e a rg u m e n ts in favor o f su ch a la w are r ela ted to th o s e fav o rin g th e
o rd inary v e lo c ity d e f e c t la w o f S e c ts. 5 .5 and 5 .6 . N o w i f w e s u p p o s e th a t an o verlap la y er
b e tw e e n th e t w o lay ers e x is ts in w h ic h b o t h th e la w s a p p ly , th e n it is ea sy t o s h o w th a t in
th is la y er b o t h th e f u n c tio n s { and m u s t b e lo g a r ith m ic , i.e ..
n(z)
w h ere (7 = / i - iA is a c o m p le x c o n s ta n t, and x and z^^ h a v e th e sam e m e a n in g as in th e

p r e v io u s s u b s e c tio n s o f th is se c tio n . E q u a tin g th e tw o e x p r e s s io n s fo r H(z) in th e o v erlap
layer, we obtain the following equation:
irio g j!!--c ‘
'L .
D iv id in g th e la st e q u a tio n in t o real a n d im a g in a ry parts, w e o b ta in th e tw o e q u a tio n s ( 6 .7 2 ) .
6.7 Distribution of Suspended Particles in a Turbulent Flow

L et us n o w c o n s id e r th e p r o b le m o f th e m o t io n o f s u sp e n d e d particles in a tu rb u le n t
f lo w o f in c o m p r e s sib le flu id . T h e th e o r y o f th is p h e n o m e n o n w a s d e v e lo p e d b y B aren b la tt
( 1 9 5 3 ; 1 9 5 5 ) [se e a lso K o lm o g o r o v ( 1 9 5 4 ) ] . T h e fu n d a m e n ta l p o s tu la te o f th e th e o r y is th e
a s su m p tio n th a t th e size o f th e s u sp e n d e d p a r tic le s is sm all in c o m p a r iso n w ith th e le n g th
scales o f th e tu r b u le n c e . T h is p e r m its u s to assum e th a t th e y fo r m , as it w e re , a c o n tin u o u s
d is tr ib u tio n o f a d m ix tu r e in th e b a sic flu id m e d iu m . T h e to ta l d e n s ity o f th e a d m ix tu r e m a y
b e w r itte n in th e fo r m
p= po (1 — 5 ) Pi5 = Po + (pi — Po) 5, ( 6 .7 3 )
w h e r e po a n d pi are th e d e n s itie s o f th e flu id and o f th e p a rticles and s is th e relative v o lu m e

o f th e p a rticles; in w o r k s o n silts s is ca lled th e “t u r b id ity ” ; po and pi are p h y sic a l
c o n s ta n ts , b u t th e “ tu r b id ity ” s flu c tu a te s as a resu lt o f tu rb u le n t m ix in g so th a t
p' = (pl— Po) s'.

T h e v e lo c ity o f th e m o t io n o f th e m ix tu r e at a giv en p o in t (i.e ., th e v e lo c ity o f th e cen te r
o f g ravity o f an in fin ite s im a l v o lu m e o f th e m ix tu r e su rro u n d in g th e giv en p o in t) w ill b e
d e fin e d as th e m a ss-w eig h te d m e a n o f th e v e lo c ity o f th e b a sic flu id Woi and th e v e lo c ity o f
th e a d m ix tu r e Uu
= «o.- + -y-«u- (6.74)
T h e e q u a t io n o f m o t io n fo r a m ix tu r e m a y b e w r itte n as
^ ^ [po(1 — s) « 0i«0« + PiS«u“ ic. + (6.75)
w h e r e p is th e to t a l pressure at a g iv en p o in t o f th e m ix tu r e , is th e sum o f th e v isco u s

stress te n so r in th e b a sic flu id in th e p r esen ce o f a d m ix tu r e s, and th e ten so r o f th e
a d d itio n a l stresses arising o n a c c o u n t o f th e in te r a c tio n o f th e su sp en d e d p a rticles. O f th e
b o d y fo r c e s , w e ta k e in t o a c c o u n t o n ly th e g ravity fo r c e , w h ic h is d ir e c te d a lo n g th e Ox^
c o o r d in a te in th e d ir e c tio n o f A'3 d ecrea sin g .
T h e e q u a tio n s o f m ass b a la n c e fo r th e b a sic flu id a n d th e a d m ix tu r e ta k e th e fo r m
^Po(l — S) , <^Po(l— s)« 0« _ n t^PiS , _n

m dx, Ot dx,
A d d in g th e e q u a tio n s , w e o b ta in th e e q u a t io n o f m ass b a la n ce fo r th e m ix tu r e
O n th e o th e r h a n d , d iv id in g th e s e e q u a tio n s th r o u g h o u t b y po and p i a n d th en a d d in g , w e
o b ta in th e c o n tin u ity e q u a tio n fo r th e m ix tu r e
[(1 —s)«oa + s«i.l = 0 - ( 6 .7 7 )

dx.
T a k in g in t o a c c o u n t th e fa c t th a t th e pa rticles are sm a ll, and a d o p tin g th e a d d itio n a l

p o s tu la te th a t th e flu id pa rticle a c c e le r a tio n s in a tu r b u le n t f lo w are sm a ll in c o m p a r iso n
w ith th e g r a v ita tio n a l a c c e le r a tio n gy w e m a y a ssu m e th a t th e h o r iz o n ta l v e lo c ity o f th e
b a sic flu id a n d th e a d m ix tu r e are th e sam e w h ile th e v ertica l v e lo c itie s d iffe r b y s o m e
q u a n tity a , th e rate o f g r a v ita tio n a l s e d im e n ta tio n o f th e p a rticles. H e n c e fo r th , w e shall
lim it o u r selv e s t o th e d is c u s sio n o f th e case o f lo w “ tu r b id ity ,” i.e ., w e shall a ssu m e th a t
s < C l 1- T h e n a m a y b e a ssu m e d in d e p e n d e n t o f s, th a t is, c o n s ta n t (e q u a l to th e rate o f
g r a v ita tio n a l s e d im e n ta tio n o f a single pa rticle in an in f in it e flu id w h ic h w e a ssu m e to b e th e
sam e fo r all p a r tic le s). W e th e n h a v e
( 6 .7 8 )
Po (1 — s) UolUo^ 4 - P l S « li« l« =
PoPl s(1—s)
T h e e q u a t io n s o f m o t io n ( 6 . 7 5 ) n o w tak e th e fo r m
dt + ■ = — P 3. (6 .7 9 )
p p
a n d th e c o n tin u ity e q u a tio n s ( 6 . 7 7 ) m a y b e w r itte n as
du, d s ( l — s )ffl
(6 .8 0 )
dx, = - ( P . - P o ) ^ - - - - - - - - ^- - - - - - - ■
F r o m n o w o n , w e shall use o n ly E qs. ( 6 . 7 3 ) , ( 6 . 7 6 ) , ( 6 . 7 9 ) , and ( 6 . 8 0 ) , and th e r e b y
c o n sid e r th e p r o b le m o f th e d is tr ib u tio n o f s u sp en d e d p a rticles in a s te a d y f lo w o f flu id
fillin g th e h a lf-sp a c e ^ 0 , a n d , in th e m e a n , h o m o g e n e o u s a lo n g th e O x ^ 0 x 2 a x e s. In
th is ca se it m a y _ b e a ssu m ed th a t all th e m e a n c h a ra cteristics d e p e n d o n ly o n th e c o o r d in a te
z = x ^ a n d th a t U2 = 0 .
W riting U3 = w a n d averaging th e c o n tin u ity e q u a t io n , w e o b ta in

dz
c o n s t. S in c e , fo r a ste a d y reg im e th ere is n o o v erall vertica l m ass tra nsfer, w e m u st p u t
p w = p w + p ' w ' = 0 . R e ca llin g E q . ( 6 . 7 3 ) , w e m a y r ew r ite th e la tter r ela tio n sh ip as
(6 .8 1 )
Pi — Po F r o m th e very e x is t e n c e o f a ste a d y d is tr ib u tio n o f th e su sp en d e d

P
p a rticles, it f o llo w s th a t th e co rr e la tio n b e tw e e n w ' a n d 5 ' m u st b e p o s itiv e sin c e in a ste a d y
reg im e th e regular g r a v ita tio n a l s e d im e n ta tio n o f th e p a rticles m u st b e b a la n c e d b y Uieir
tu r b u le n t tra n sfer u p w a r d s, w h ile th e tu r b u le n t f lu x of m ass is equal to p'a>' —
(p i — Po) C o n s e q u e n tly , w < 0 , i.e ., th e m e a n m o t io n o f th e m ix tu r e h a s a d o w n w a r d
c o m p o n e n t; w e recall th a t th e v e lo c ity o f th e m ix tu r e is d e fin e d as th e m a ss-w eig h te d m e a n
o f th e v e lo c itie s o f th e b a sic flu id a nd th e p a rticles, and n o t th e v o lu m e -w e ig h te d m e a n .
A veragin g E q . ( 6 . 8 0 ) w e o b ta in
5 ( 1 — 5 ) /Z *
w h e n c e , a fter in teg r a tio n w ith r e s p e c t to z w e fin d
w = — a s (1 — 5) a -f- c o n s t.
T h e c o n s ta n t term m th is e q u a tio n m a y b e ta k en eq u a l to z e r o , sin c e, fo r large 3 , s , w's'

a n d , c o n s e q u e n tly , Wy te n d t o z e r o . H e n c e , c o m p a r in g th e last e q u a tio n w ith ( 6 . 8 1 ) w e h a v e
w ' s ' = 5 (1 — s ) a as. ( 6 .8 2 )
I f w e u se a sem iem p ir ic a l e q u a tio n o f th e ty p e ( 6 . 2 5 )
W s' = — (6 .8 3 )
th en th e la tter r e la tio n sh ip ta k es th e fo rm
asK = - a. (6 .8 4 )
A s th e m ea n e q u a tio n o f m o t io n o f th e m ix tu r e , w e use th e R e y n o ld s e q u a tio n in th e

d ir e c tio n o f th e Ox\ a xis, w h ic h fo r valu es o f z th a t are n o t t o o large, ta k es th e fo llo w in g
form :
K = c o n s t. (6 .8 5 )
E H m inating K fr o m E qs. ( 6 . 8 4 ) and ( 6 . 8 5 ) w e o b ta in
dz dz ’
w h e r e o) = ^ is a d im e n sio n le ss pa ra m eter. In teg ra tin g th is e q u a t io n , w e o b ta in
5 = sê , (6 .86)
w h e r e Sq is th e value o f s a t s o m e virtual “ r o u g h n ess h e ig h t” Zq at w h ic h u (z ) b e c o m e s
zer o .
T a k in g th e a p p r o x im a te va lu e o f — o n th e right sid e o f E q. ( 6 . 7 9 ) , w e see th a t
th is e q u a t io n , to g e th e r w ith th e m ass b a la n ce e q u a t io n (6 .7 6 ) c o in c id e s w ith th e
c o r r e sp o n d in g e q u a t io n s fo r a c o m p r e ss ib le flu id . H e n c e th e e n e rg y b a la n ce e q u a tio n h a s th e
sam e fo r m as fo r a c o m p r e ss ib le flu id in a gravity fie ld . Ig n oring th e e f f e c t o f d if fu s io n o f
th e tu r b u le n t e n e rg y a n d o th e r sm all te r m s, w e w r ite th is e q u a t io n in th e fo r m
= (6.87)
w h ere I \ b — K and B = — is th e w o r k o f su sp en d in g th e particles b y th e tu rb u le n t

flo w per u n it v o lu m e o f th e m ix tu r e . U sin g E qs. ( 6 . 7 3 ) a n d ( 6 . 8 2 ) w e o b ta in
B = — Q ^ g s ' w ' — — c p ^ s ( l — s) a. (6 .8 8 )
T h is e x p r e s s io n w a s first o b ta in e d b y V e lik a n o v ( 1 9 4 6 ) . E x p r essin g s ' w ' b y m e a n s o f E q .

( 6 . 8 3 ) , w e o b ta in
B = \H >
E q u a tio n ( 6 . 8 7 ) m a y th e n b e r e d u c e d to th e fo r m
w here
{du/dzY
is a d e m e i^ io n le s s n u m b e r a n a lo g o u s t o th e R ic h a r d so n n u m b e r . E lim in a tin g th e v e lo c ity
g r a d ie n t b y u sing E q. ( 6 . 8 5 ) w ith K = / V~~^ w e o b ta in
b = cûl y 1 — a^Ri (6 .9 0 )
In th e a b s e n c e o f s u sp e n d e d pa rticles R i = 0 a n d b = cû». E q u a tio n ( 6 . 9 0 ) s h o w s th a t th e

p r e se n c e o f s u sp e n d e d p a r tic le s in th e flo w le a d s t o a lo w e r in g o f th e tu r b u le n t e n e r g y , i.e .,
it in f lu e n c e s th e d y n a m ic s o f th e flo w . T h is r esu lt is c o n fir m e d b y d ir e c t e x p e r im e n ts . T h u s
th e c a lc u la tio n o f th e m o t io n o f silt, o n th e a s su m p tio n th a t s u sp e n d e d p a r tic le s h a v e n o
e ffe c t on th e d y n a m ic s o f th e flo w (th is is c a lle d th e “ d if fu s io n a p p r o x im a tio n ” in
s e d im e n ta tio n t h e o r y ), is p o s sib le o n ly w ith s u ffic ie n t ly sm all R i.
T h e d is tr ib u tio n s o f m e a n v e lo c ity u { z ) in a flo w carryin g s u sp e n d e d p a rticles m a y b e
fo u n d w ith th e aid o f E qs. ( 6 . 8 4 ) , ( 6 . 8 5 ) , and ( 6 .9 0 ) . I f w e in tr o d u c e d im e n sio n le ss
q u a n titie s , w r itin g
n" = - ^ /=■ ( 0 . /c = x a . ^ $ ( 0 , c/ = w ' r ( C ) , 1 = 5 (0 , ^ = X ’
w here L = — — , th e n th e e q u a tio n s ta k e th e fo r m
C $ 5 ' = — (oS, ?*/='' = !, + (6 .9 2 )
T h e fu n c tio n S ( ^ ) is o f s p e cia l in ter e st; th is d e scr ib e s th e v e rtic a l d is tr ib u tio n o f th e
s u sp e n d e d p a rticles. E lim in a tin g F a n d O fr o m E q. ( 6 . 9 2 ) for T'(C) = 5 ”( ^ o b ta in
2 \V<
B a re n b la tt in v e stig a te d th e s o lu tio n o f th is e q u a tio n in general fo r m , w ith o u t giv in g a

c o n c r e te fo r m to th e fu n c tio n 'F. H e a ssu m ed o n ly th a t it satisfies th e o b v io u s c o n d it io n
^ ( 0 ) = 1, and is a n o n in c re a sin g fu n c tio n o f th e R ich a rd so n n u m b er sin c e th e le n g th scale
I m u st d ec re a se d u e to th e p r e se n c e o f th e s u sp e n sio n . T h e resu lts o f th e a n a ly sis s h o w th at
for (0 < l an d 0 ) > 1 th e n a tu re o f th e s o lu tio n is q u ite d iffe r e n t. W hen c o > l (lo w flo w
v e lo c ity or large p a rticles) th e transfer o f pa rticles ta k es p la c e prim arily c lo se to th e b o t t o m
o f th e flo w . H o w e v er , in th e b e d reg io n , th e th e o r y w e h a v e d isc u sse d (b a se d o n th e
a s su m p tio n th a t s is sm a ll) is in a p p lic a b le and m u st b e r ep laced b y a m ore e x a c t th e o r y .
O u tsid e th e b e d r eg io n R i ^ 0 , so th a t th e d is tr ib u tio n o f th e su sp en d e d particles
a s y m p to t ic a lly te n d s to th e d is tr ib u tio n o b ta in e d a c co r d in g t o th e d iffu s io n a p p r o x im a tio n .
W ith 0 < 1 (h ig h flo w v e lo c ity or sm a ll p a r tic le s), th e transfer ta k es p la c e in th e b o d y o f
th e flo w . T h e d istr ib u tio n o f p a rticles a t g rea t h e ig h ts ^ n d s a s y m p to t ic a lly to s o m e lim itin g
self-preserving d istr ib u tio n (R i - > c o n s t ) fo r w h ic h s is in v ersely p r o p o r tio n a l to z , a nd
- u.
II ( r ) = —- lo g -f- c o n s t. T h u s th ere e x ists so m e lim itin g sa tu r a tio n o f th e flo w , su ch
th a t a t still greater sa tu r a tio n , th is th e o r y b e c o m e s in a p p lic a b le in th e b e d reg io n .

4 TURBULENCE IN A
THERMALLY STRATIFIED
MEDIUM
7. GENERALIZATION OF LOGARITHMIC LAYER THEORY

TO THERMALLY STRATIFIED FLOWS
7.1 Thermally Stratified Turbulent Boundary Layer as a

Model of the Atmospheric Surface Layer
In Chapt. 3 we considered in detail turbulent flow in the boundary
layer above an infinite flat plate. Our deductions were compared
with both data from laboratory flows and observations of the air
motions in the surface layer of the atmosphere. However, it was
noted that for this purpose it is only valid to use those observations
relating to neutral stratification, that is, to cases when the air
temperature in the lower layers is practically constant with height.*
^ G en eta lly sp e a k in g , n e u tra l s t ia tif lc a tio n is ta k en t o m e a n a te m p er a tu re d is tr ib u tio n in

w h ic h th e vertical g r a d ie n t d T j d z is id e n tic a l w ith th e a d ia b a tic g r a d ie n t Ga d e fin e d in
S e c t. 2 .4 . In fa c t, it is o n ly u n d er th ese c o n d it io n s th a t th e v ertica l d is p la c e m e n t o f a flu id
e le m e n t w ill b e a c c o m p a n ie d b y n e ith er lo ss n o r gain o f p o te n tia l e n e rg y . H o w e v e r , sin ce
th e ty p ic a l valu es o f th e v e r tic a l te m p er a tu re g r a d ie n t in tKe fe w te n s o f m e te r s o f th e earth
(C o n t ’d o t t p . . 4 1 8 ) .
417
However, neutral stratification is fairly rare in nature. In fact,

during the day the temperature over the earth generally decreases
considerably with height, but at night, as a rule, it increases with
height (i.e., so-called temperature inversion takes place). Thus neutral
stratification occurs only for fairly short periods before sunset and
after sunrise.^ Consequently, the motion in the surface layer cannot,
for most cases, be classified as belonging to the simple scheme of the
turbulent boundary layer discussed in Chapt. 3.
Therefore, in practice, when studying turbulence in the surface
layer of the atmosphere, we must take into account the presence in
the atmosphere of temperature stratification which produces a
systematic height variation of the density. We know that the
presence in a gravitational field of density inhomogeneities leads to
the appearance of buoyancy forces, producing upward displacement
of the less dense fluid particles and downward displacement of the
more dense fluid particles.^ As a result, the particles that are less
dense (denser) than the surrounding medium acquire additional
energy during their upward (downward) motion, due to the work
done by the buoyancy forces, while during downward (upward)
motion, they expend part of their energy in overcoming the
buoyancy forces. The potential energy of a density-stratified medium
in a gravitational field can thus be directly transformed into
turbulent energy, and, conversely, turbulent energy can be trans
formed into potential energy of the medium. This chapter will be
devoted to a quantitative study of the effect of such conversions of
energy on turbulence.
In Sect. 6.4 we calculated the contribution made by the work of
the buoyancy forces in the balance of turbulent energy. This
a tm o sp h e r e are u su a lly o f th e order o f s o m e h u n d r e d s o f tim e s greater th an G a . i n m o s t
cases w e m a y ta k e Ga t o b e z e r o . T h u s it is u su a lly p o ssib le to m a k e n o d is tin c t io n b e tw e e n
th e o r d in a ry te m p er a tu re T a n d th e p o te n tia l te m p er a tu re d o f E q . ( 2 .4 ) . T h e r e fo r e , for
s im p lic it y , w e shall h e n c e fo r th a lw a y s sp ea k o f th e ordin a ry te m p er a tu re a lth o u g h , str ictly
sp ea k in g , in m o s t cases th e e q u a t io n s w h ic h w e use w ill b e e x a c t o n ly fo r th e p o te n tia l
te m p er a tu re . A ls o , in S e c t. 8 , w e shall m a k e n o s p e cia l m e n t io n o f th e fa c t th a t in certain
cases c o n c e r n in g o b se rv a tio n s carried o u t in a rela tiv e ly th ic k lay er o f air o f th e order o f
several te n s o f m e te r s, th e data u sed refer in fa c t n o t to th e ordin a ry b u t to th e p o te n tia l
te m p er a tu re .
Ôver th e sea th e s itu a tio n is d iffe r e n t an d str a tific a tio n co m p a r a tiv e ly c lo se t o neu tra l is
o b serv ed m o r e o f t e n . H o w e v e r , w e shall n o t c o n sid e r th e m a rine a tm o sp h e r e in d e ta il in th is
book.
^Here an d h e n c e fo r th , th e b u o y a n c y fo r c e w ill b e ta k e n to m ean th e d iffe r e n c e b e tw e e n
th e A r ch im e d e a n fo r c e (d esc rib ed b y th e A r ch im e d e s la w ) and th e b o d y fo r c e o f g r a v ity . In
a h o m o g e n e o u s (n o n str a tifie d ) m e d iu m , th ese t w o fo r c e s b a la n c e e a ch o th e r e x a c t ly , and
h e n c e th e g ravity fo r c e may sim p ly b e ig n o r e d ; s e e th e o p e n iilg paragraph o f S e c t. 1 .2 .
TURBULENCE IN A THERMALLY STRATIFIED MEDIUM 419
contribution (per unit volume) is given by the equation
(7.1)
where w' is the fluctuation of the vertical velocity [see Eq. (6.38),
Sect. 6.4]. From Eq. (7.1) it is clear that density stratification will
affect turbulence only if the density fluctuations p' are correlated
with the fluctuations w '. Considering the meteorological appUca-
tions, we shall find it convenient to change from working with
density fluctuations (which are very difficult to measure in the
atmosphere) to temperature fluctuations, which are readily suscep
tible to direct measurement. Here, as in Sect. 6.5, we shall take into
account only density fluctuations connected with fluctuations of
temperature, and the considerably smaller density fluctuations pro
duced by the fluctuations of atmospheric pressure will be ignored. In
the case of not too large temperature fluctuations, we may write p '~
— (plT )T ', where T is the absolute temperature, and, therefore, in
this approximation
5 = = (7.2)
where q = C p p w ' T ' is the vertical turbulent heat flux (see Sect. 6.4).
Thus it is clear that the buoyancy effect on turbulence in the
atmosphere is connected directly with turbulent heat-transfer from
the underlying surface to the atmosphere and vice-versa. This leads
once again to the conclusion that this effect must vanish when the
vertical temperature gradient is equal to zero; in fact, according to
the fundamental equation of the semiempirical theory
g= (7.3)
where Kris the eddy thermal diffusivity and, consequently,
(7.4)
dT
so that 5 = 0 when = 0.
In the theoretical analysis of turbulent phenomena in the surface
layer of the atmosphere, we must take into account the presence of
vertical temperature stratification and the associated vertical turbu
lent heat flux. On the other hand, the horizontal inhomogeneity of
420 S T A T IS TIC A L F L U ID MECHANICS
the underlying surface, which, to some extent, is always present in

the real atmosphere, may naturally be neglected at first. In fact, for a
comparatively flat underlying surface, the general character of which
does not change over a comparatively large region, this inhomogene
ity will not play a very im portant part, while taking it into
consideration makes the theoretical analysis considerably more
complicated. Thus we shall consider only a simplified model of
turbulence in a fluid filling the half-space above an infinite
homogeneous plane surface 2 = 0 o f constant roughness Zo (the
displacement height ho, which may be taken into account by a simple
change of origin for z, will also, for simplicity, be ignored). As a
result, we shall always assume in this chapter that all one-point
averaged characteristics of the flow variables depend only on the
vertical coordinate z.
If we assume that all one-point moments are dependent only on 2 ,
we assume implicitly that these moments may be determined, i.e.,
that the values of all flow variables in the surface layer of the
atmosphere possess a determinant statistical stability. Generally
speaking, this assumption may give rise to doubt, since the
atmospheric conditions depend considerably upon the time of day
and the time of year, while in addition to the regular diurnal and
annual variations, the values of any meteorological variable at a given
point of the atmosphere also undergo irregular fluctuations with very
diverse periods. These irregular fluctuations may be considered to be
due to the occurrence of turbulence of different three-dimensional
scales from the very small (of the order of centimeters and fractions
of a centimeter) to the very large (of the order of the dimensions of
cyclones and anticyclones, or even o f the scale of inhomogeneities of
the general circulation of the atmosphere). Thus, for example, the
time means of temperature or wind velocity at a given point of the
atmosphere prove first, to depend considerably on the size of the
averaging interval, and, second, for a given scale of averaging, to
fluctuate from sample to sample under the action of components of
turbulence with periods comparable to or greater than the size of the
averaging interval. This phenomenon called the “drift of the mean”
o f meteorological variables, considerably complicates attempts to
determine their statistical characteristics. Nevertheless, experience
shows that if we confine ourselves to observations relating to a
specified season of the year, a specified time of day, and specified
synoptic conditions (i.e., specified “ weather”), then averaging over a
time interval t which is considerably longer than the characteristic
TURBULENCE IN A T H E R M A L L Y S T R A T IF IE D M E D IU M 421
period of the “energy-containing eddies” (turbulent formations

which contain the principal fraction of the turbulent energy), the
mean values of the meteorological variables will be relatively stable.
But, in this case, the relevant observations may be taken to
constitute a “statistical ensemble,” permitting us to carry out
probability averaging. In the surface layer o f the atmosphere, the
time scale of the energy-containing eddies may be estimated to be of
the same order o f magnitude as the ratio Lo/L/, where U is the
characteristic value of the wind velocity and L q is the characteristic
horizontal length scale of the energy-containing eddies, measured in
tens or a few hundreds of meters. Hence LolUis of the order of some
tens of seconds, and, averaging over a time interval of the order of
ten to twenty minutes, the mean values of wind velocity, tempera
ture, etc., prove to be relatively stable and may be considered as
approximate values of the probability means for corresponding
random fields. When the averaging period is extended substantially to
intervals of the order of several Jiours or even more, the mean values
change considerably, and once again may even prove to possess very
low stability due to the effect of long-period “synoptic fluctuations”
relating first to medium-scale turbulence and then to macroturbu-
lence; however, we shall not deal with this type of turbulence here.
In addition to the assumption of horizontal homogeneity, we shall
also adopt several simplifying assumptions relating to the dynamic
equations which describe the turbulence under investigation. First, as
already stated, we shall ignore variations of density produced by
pressure fluctuations, and shall confine ourselves to equations which
have been linearized with respect to deviations of the density,
temperature and pressure from the corresponding “standard values”
po, To and po (which are dependent only on z and satisfy the equation
of statics dpoldz = —pog and the equation of state po= RqqTo).
Henceforth, the values po=p and To will be regarded simply as
constants since their variation with height in the surface layer (which
is some tens of meters thick) is negligibly small. In this case, the
equations o f motion reduce to the well-known “free convection
equations” or “ Boussinesq equations” of the form
du . du . du . du 1 dp' , .
dv . dv . dv ^ dv 1 dp' , . (7.5)
^ ^ ^ ^ _ 1 ^
dt ' djc ^ dy ^ dz po dz +' VA®, - f r .
In these equations the term — T' describes the “buoyant

*0
acceleration” of the fluid particles depending on the gravity force.
However, the Coriolis force is neglected, since, according to the
estimate given in Sect. 6 .6 , this force has no perceptible effect on the
mean motion in the bottom 50 m (or thereabouts), and, moreover,
can have no effect on the fluctuating motion, the velocity of which
in the lower atmosphere is approximately one order o f magnitude
less than the mean velocity. Of course, at heights of more than a few
tens o f meters above the surface of the earth, Eqs. (7.5) w ithout the
Coriolis forces no longer apply; this must be borne in mind when
deductions from these equations are compared with observation.
The continuity equation for the Boussinesq approximation may be
written just as in the case of an incompressible fluid:
W + (7.6)
It follows from this and the horizontal homogeneity that w = 0. Thus

the mean motion is plane-parallel and is given by the mean velocity
vector u(z) = {U(z), V(z), 0}. It is also clear that in the absence of
the Coriolis force and with horizontal pressure gradient constant
with height, the direction o f the vector « ( 2 ) must be 2 -independent.
We now take the Ox axis along the vector u (z). In this case the
equation v = 0 will hold; thus, the probabiUty distribution for the
flow variables may naturally be assumed invariant not only under
parallel displacements in the Oxt/ plane, but also under reflections
with respect t o th e Oxz plane. Consequently, in particular, it follows
that Xxy = —p o « V = 0 and Xyz = —pov'w'= 0.
If we take into account that the mean wind velocity in the surface
layer varies during the 24-hour cycle by an am ount of the order of
1 0 m/sec, it is not difficult to estimate that under ordinary
conditions, in which there are no particularly rapid weather changes,
the term containing ~ in the averaged first equation of Eqs. (7.5)
proves to be far smaller even than the term containing the mean
pressure gradient ^ or than the mean Coriolis force. Hence, when
we average Eqs. (7.5), the time derivatives may also be discarded, and
we may assume that in the surface layer
— Po«W + p y v X = const (7 .7)

TURBULENCE IN A T H E R M A L L Y S T R A T IF IE D M E D IU M 42 3
(see Eq. (5.11), Sect. 5.2J. Thus it seems probable that the mean
flow in the surface layer over a homogeneous and flat earth or sea
surface may be assumed not only plane-parallel but also steady, and
the shear stress x = (the vertical momentum flux) constant with
height.
In addition to Eqs. (7.5) and (7.6), we must also use the heat
equation, which determines the time variation of the temperature T.
According to the results of Sect. 1.5, this equation may be written
as;
v ( ^ + - | f + » f + - ^ ) = ^ „P xV r+ Q , (7.8)
where Q is the influx of heat due to radiation and phase changes of

moisture contained in the air (other forms of heat influx and, in
particular, heating due to the kinetic energy dissipation in the surface
layer may be ignored completely). If we also ignore the influx of
heat Q, then we obtain the ordinary thermal conductivity equation
, dT , dT , dT
+ + + = (7 .8 )
the averaging of which leads to the conclusion that in steady

conditions
q^C pP ow 'T '— c ^ P o X ^ = (ônst, (7.9)
i.e., that the mean vertical heat flux caused by molecular and
turbulent thermal conductivity is constant [cf. Eq. (5.73'), Sect.
5.7]. However, if we also take Q into account, the principal part of
which usually consists of the radiant heat flux, then instead of Eq.
(7.9) we obtain the more complex equation
^+ = const, (7.9')
where Q = — so that qs can usually be identified with the mean

vertical flux of radiant heat. According to the preliminary data of
Robinson (1950) [see also Haltiner and Martin (1957), Chapt. 15],
significant variation of qs with height usually takes place only in a
thin layer of air close to the surface of the earth, the thickness of
which does not exceed (or at least only slightly exceeds) 1 m. If this
is true, then when we consider higher layers we may put qs = const,
i.e., we may ignore the radiant influx o f heat Q. This is confirmed
indirectly by the fact that the theory based on Eq. (7.9), as will be
seen later, gives satisfactory agreement in the majority of cases with
the data of observations in the surface layer.'* The relative smallness
of the contribution of the variability of to the height variation of q
was confirmed also by crude estimates of Mordukhovich and Tsvang
(1966). In general, however, the question of the validity of the
approximate equation q^ = const in the surface layer of air still can
not be regarded as definitely proved, either theoretically or experi
mentally. Particularly doubtful from this viewpoint are conditions of
very stable stratification (observed, for example, during the night),
when the turbulent heat flux q is very small, and the variation with
height of the flux qs may be considerable [cf. Funk (I9 6 0 )]. Hence
the attempts to consider the effect of radiative heat transfer on
turbulence more exactly, which were begun by Townsend (1958)
and Yamamoto and Kondo (1959) should be considered most
opportune. In addition, we must also note that there must exist
vertical variations of the turbulent fluxes r and q caused by nonzero
horizontal gradients of mean wind velocity and temperature which
may be considerable even in sites usually considered to be quite
homogeneous [cf. Mordukhovich and Tsvang (1966); and Dyer
(1968)]. However, since only very preliminary estimates have been
made of the vertical variations of the turbulent fluxes, we shall
confine ourselves here to a consideration only of the rougher theory
which follows from the assumptions (7.7) and (7.9), the formulation
of which, in any case, must be regarded as a necessary first step in
the elucidation of the mechanism o f turbulence in the surface layer.
Finally, we can now give a definite description of the idealized
theoretical model to be studied in this chapter. We shall consider a
plane-parallel steady fluid flow in the half-space 2 > 0 above a flat
homogeneous surface 2 = 0 (characterized by given roughness 20 )
described by the Boussinesq equations (7.5), (7.6) and (7.8') and
satisfying conditions (7.7) and (7.9) with given values and q; these t
conditions act as boundary conditions on the upper boundary of the

layer. According to our previous remarks, we may expect that this
model will correspond comparatively well to many actual flows in
‘*We must point out, however, that the logarithmic profile of the mean velocity, which
fo llo w s fr o m th e th e o r y o n ly w h e n T = c o n s t, is s a tisfie d fairly w e ll in tu b e s a lm o s t to th e
c en te r o f th e tu b e w h e r e T = 0 . T h u s th is in d ir ec t p r o o f can in n o w a y be regarded as very
reliable.
the lower layer of the atmosphere (of the order of several tens of
meters) above a comparatively flat and homogeneous underlying
surface. We may also assume that this model will also be applicable
to some other types of flow, for example, to turbulence in the
bottom or surface layers of the sea, in isothermal conditions but with
considerable sahnity stratification (and hence also, density strat
ification) and to some artificial turbulent flows set up in the
laboratory. (In the case of density stratification, o f course, we simply
have to replace by —— and 7’' / f by — p'/p everywhere.)
However, henceforth, for definiteness we shall always speak of
turbulence in the surface layer of the atmosphere (since most of the
available data relate to this), taking “surface layer” everywhere to
mean the idealized model of a constant flux layer of a density-
stratified medium which we have described.
7.2 Application of Dimensional Reasoning to Turbulence
in a Stratified Medium
The turbulence characteristics in a thermally stratified medium
described by Eqs. (7.5), (7.6) and (7.8') and conditions (7.7) and
(7.9) can depend clearly only on a fairly small number of physical
quantities; namely, on the parameters po, v, and x which occur in
^0
these equations, on the values of -t (or «*) and q which give the
momentum flux or heat proceeding from infinity to the surface 2 =
0 (or vice-versa) and characterizing the dynamic and thermal
interaction of the flow with the underlying surface, and on the
roughness parameter zo, which completely describes the geometrical
properties of the underlying surface. However, not all of the
parameters play an equally significant part. First, we know that in
regions with sufficiently developed turbulence (i.e., almost every
where except in a very thin sublayer contiguous to the underlying
surface) the fluxes o f heat and momentum due to molecular motion
are always very small in comparison with the turbulent fluxes of heat
and momentum (see, for example, Sects. 5.1, 5.3 and 5.7). Thus in
these regions the terms of Eqs. (7.5) and (7.8') which contain the
molecular coefficients v and x may generally simply be ignored. But
then Eqs. (7.7) and (7.9) may be rewritten in the simpler form
— Pqu' w ' = x = const, CpPoT'w' = q = const,
which indicates that the characteristics of turbulence must be

independent of v and x- As for the roughness parameter this

determines the boundary conditions on the underlying surface and,
through these boundary conditions, also affects the absolute value of
the mean velocity u{z) and the difference of mean temperatures
T{z) — f(0 ) at considerable distances from the underlying surface.
Ho)vever, the vertical variation of the mean velocity and the mean
temperature at distances z at which the direct effect of the
underlying surface is no longer perceptible, can not depend on zo, but
must be determined only by the values t and q of the flux of
momentum and heat. In other words, variations of the roughness
parameter Zo can only lead to a shift of the curves u = u{z) and f =
f { z ) by some constant amount, but can not affect the form of the
profiles u{z) and T{z). Therefore, the dependence on height of the
mean velocity gradient and temperature gradient and other related
characteristics of a developed turbulence in the surface layer of air
must be determined by the following four parameters only: the
density po, the “buoyancy parameter ” gjTo (describing the buoyancy
effect), the turbulent shear stress x (or k,. ^ | / ) and the vertical
turbulent heat flux q (or 9 /Cppo).® Using this simple postulate, we
must remember that when ^ = 0 (i.e., in the case of neutral
stratification, when = 0 ) no buoyancy effect on the turbulence
should be apparent, that is, the dependence on the parameter
i0
must vanish. In this case, we have returned to the case of a
^ H o w ev er, w e sh o u ld n o t e th a t th is natural p o s tu la te , w h ic h is fu n d a m e n ta l fo r th e
f o llo w in g th e o r y , s o m e tim e s e v o k e s c erta in o b je c tio n s. T h u s, o n o c c a s io n , th e id ea is
a d v a n ced th a t th e te m p er a tu re and w in d p r o files in a str a tifie d a tm o sp h e r e m u st d e p e n d o n
th e m o le c u la r c o n s ta n ts v a n d x» or at le a st o n th eir r a t i o - t h e Prandtl n u m b er Pr = v / x
[se e , for e x a m p le , T o w n s e n d ( 1 9 6 2 a ) ] . In several ca ses, th e b asis fo r th is id ea is M alk u s’
c o n tr o v e rsia l th e o r y o f tu r b u le n t c o n v e c t io n ( 1 9 5 4 b ) d isc u sse d , for e x a m p le , b y T o w n s e n d
(1 9 6 2 b ), S p iegel ( 1 9 6 2 ) , and L in d z e n ( 1 9 6 7 ) ; a c co r d in g to this th e o r y th e m o le c u la r
c o n s ta n ts pla y a c o n sid e ra b le r o le fo r all z. T h e e f f e c t o f th e Prandtl n u m b e r Pr o n
tu r b u le n t c o n v e c t io n is also th e su b je c t o f K r a ic h n a n ’s w o r k ( 1 9 6 2 a ) , th e resu lts o f w h ic h ,
for m o d e r a te va lu es o f Pr, d o n o t c o n tr a d ic t th e c o n c lu s io n s o b ta in e d in th is c h ap ter. O n
th e o th e r h a n d , B usinger ( 1 9 5 5 ) , for e x a m p le , a ssu m ed th a t th e ro u g h n ess para m eter Zu
a ffe c ts th e fo rm o f th e w in d a n d te m p er a tu re p r o file s e x p lic itly , w h ic h m a k es all his
e q u a tio n s c o n s id e r a b ly m o r e c o m p lic a t e d . L a y k h tm a n ’s a s su m p tio n s ( 1 9 4 4 ; 1 9 4 7 a ; 1 9 6 1 )
g o ev en further: in ste a d o f th e o rd in a ry ro u g h n ess pa ra m eter, h e used a le n g th p a ram eter Zq
w h ic h a ffe c ts th e fo r m s o f th e p r o files e x p lic itly and d e p e n d s o n th e size s and fo r m o f th e
irregularities o f th e su rfa ce and o n th e th erm al str a tific a tio n in a c o m p lic a te d m an ner.
homogeneous medium, for which the logarithmic layer theory

developed in Sect. 5.3 is valid.
Thus, when z ^ z o , the turbulence characteristics at the height 2
will depend only on five quantities: z, Po, a n d - — . In this
/0 ^p?o
case, since there are four independent dimensions—length, time, mass
and temperature, we can formulate only one independent dimension-
less combination (to within a numerical multiplier) from these five
quantities. Following Obukhov (1946), Monin (1950c) and Monin
and Obukhov (1953; 1954), we shall choose this dimensionless
combination to be
where
-------- ------- (7.12)

7.
is the length scale formulated from the parameters X , ir. and —— .

‘ 0
(The dimensionless von Karman constant x was first introduced into

the equation for L by Obukhov in 1946 and has usually been
preserved in all subsequent works by tradition and the sign of L is
chosen so that L > Q for stable thermal stratification, when 0.)
Then we may affirm that the dependence on height of any mean
turbulence characteristic / in the surface layer of air which is
independent of the properties of the underlying surface for not too
small e may be written in the form
,7.13)
where fo is a combination with the dimensions of /, formulated from

the parameters -i-, p,,, and ----, while Fit) is a universal function.
*0 ^/7?0
As the velocity scale we take ^ and for the temperature scale we
take
T = - — - ^ (7.14)
Cppo
[cf. Eq. (5.75')] where the constant x is also introduced by tradition

and the sign ofT^ is chosen so that in the stable case 7 * > 0. Finally,
as the natural scale for the eddy exchange coefficient K we shall take
the quantity xu^\L\.
According to Eq. (7.13), we can describe the dependence of the
turbulence characteristics on height by means of universal functions
of S. Thus, for example, for the vertical gradients of the mean
wind velocity and temperature, we obtain
§ = (7.15)
(7.16)
where g{Z) and gi(S) are two universal functions of Equations

(7.15) and (7.16) were given in Monin and Obukhov (1953; 1954);
they may be considered as an immediate generalization of the
fundamental equations (5.21) and (5.76) of the logarithmic layer
theory to the case of a thermally stratified medium. Substituting
these equations into Eqs. (5.5) and (5.9), which define the eddy
exchange coefficients K and K t for momentum and heat, we find
that
K = - ^ = ^ , K (7.17)
du jd z ^ (C) ’ ^ d T' l dz g i (Q
The ratio of the exchange coefficients Kt and/( is equal to
i = = (7.18)
generally speaking, it is some universal function of £;. The Richardson

number Ri, defined in Sects. 2.4 and 6.5 [see Eqs. (2.3') and (6.51)]
by reason of Eqs. (7.12)and (7 .1 4 )-(7 .1 6) equals
Ri — ^ (7 19)
T, {dujdzY [^ (C )] ^ a (0 ^ (C )‘
This is also a universal function of Similarly, the flux Richardson

number Rf of Eq. (6.45) defined by the heat flux q, the momentum
r and the velocity profile u (2 ) will equal
dz
The flux Richardson number Rf obviously is connected simply with

the eddy viscosity K'.
A := x « .I - R f . (7.21)
From the general form analysis of the functions g(C) andgi(C),

which will be discussed later in this section (i.e., Sect. 7.3), it may be
deduced that both Ri (C) and Rf (£;) are monotonic functions and
hence have a single-valued inverse function. Thus it follows that Ri,
Rf and ^ may all be used with equal validity as parameters
characterizing the thermal stability of the air.
The situation is similar for the characteristics of the concentration
field of a passive admixture in a stratified medium, when there is a
constant admixture flux / = pow''&' across the boundary z = 0 (in the
surface layer of the air we may take this passive admixture to be,
e.g., water-vapor; then j will denote the mean evaporation from unit
area of the underlying surface per unit time). Here it is only
necessary to add / to the set of parameters defining the mean
turbulence characteristics. However, we also have the additional
independent dimension of the quantity 0 . Consequently, once again
we have a single dimensionless characteristic l = zjL. In particular, for
the vertical gradient di^jdz of the mean humidity (in the future we
shall consider only this passive admixture), we obtain
-0, - L (7 22)
***~ ’
which is completely analogous to Eqs. (7.15) and (7.16); the

exchange coefficient for moisture (eddy diffusivity) is then equal to
ATs = — . (7.23)
(njdz g2 (C)
The actual profiles of the wind velocity, temperature and humid

ity in the surface layer can be obtained by integrating Eqs. (7.15),
(7.16), and (7.22)
u - Ti {z,) ^*-[f f 'j
f {z,) - r \ z , ) 7\ [ f , ( f -) - /. ( -^) (7.24)
ft ( Z , ) ~ d (Z,) = K [ f 2 ( ^ ) - /2 ( x )
where
/( Q = f g (^) dW A (C) = / g, (I) dV, h (Q f g 2 (^) d l (7.25)
7.3 The Form of the Universal Functions Describing

Turbulence in a Stratified Medium
From the previous discussion in the surface layer of a thermally
stratified medium, two qualitatively different turbulent regimes are
possible. They correspond to the case of stable stratification
(downward flux of heat; i.e., t/< 0 and, accordingly, L > 0 an d 7 \,>
0) and unstable stratification (q > 0, L < 0, T,. < 0). These two
regimes must approach each other as the stratification approaches
neutral (as q 0). Accordingly, all universal functions describing
turbulence are divided into two distinct branches; for C > 0 and for
C <0.
In this subsection we shall discuss only the simplest turbulence
characteristics, namely, the profiles of the mean velocity u, the mean
temperature T and the mean concentration of admixture (hu
midity) 0 . Thus we shall be interested primarily in the functionsf (^),
/i(^), a n d / 2 (C) and their derivatives g ( Q = f ' ( 0 , gi(C) = / ( (C), and
Let us begin by considering the functions f{'Q) and g (0- It will be

convenient to rewrite Eq. (7.15) in the form
az
We now fix the values of 2 , w,, and g/To, and let the value of the
turbulent heat flux q decrease without limit, thus approaching the
conditions of neutral stratification. In this case L will increase in
itbsolutc value without bound, so that C = tends to zero. In the

limit, as q ->■ 0 , we must obtain the ordinary “logarithmic” equation
— ~ , which contains neither q nor g/7'o- Consequently
9 ( 0 ) - - I im C /'( C ) = l. (7.26)
c-> o
Since cp(^) is naturally assumed continuous, for fixed q and«<, but

sufficiently small z (i.e., when 2 <c|i.|), the value of q)(^) will be close
to unity. This means that for z<^\L\ the turbulent exchange will
differ very little from the exchange in a temperature-homogeneous
medium, so that the turbulence in the layer z < |L | is produced
primarily by dynamic factors. Thus the scale L determines the
thickness of the layer in which thermal factors do not play an
im portant part; it may therefore be called the height o f the dynamic
sublayer.^ Since thermal factors may be ignored in the very lowest
layer, naturally it follows that we must take into account the
properties of the underlying surface in the case of a stratified
medium, exactly as in the case of a homogeneous medium. In other
words, the roughness parameter Zo, defined as in Sect. 5.4, is in all
cases an objective characteristic of the dynamic influence of the
underlying surface on the flow. Since the roughness parameter zq
may be determined according to the wind distribution only in the
dynamic sublayer, it is clear that this parameter cannot depend on
the thermal stratification. In practice it is most convenient to
determine this parameter by using a logarithmic approximation of
the velocity profiles in the vicinity of the underlying surface; in the
atmospheric case, it is desirable to use the cases of neutral or
near-neutral stratification, for which \L\ is large and the dynamic
sublayer is of considerable thickness. Taking into account the fact
that the logarithmic velocity profile vanishes at height ?oand that for
z -€.\L\ the velocity profile may always be assumed logarithmic, the
first equation of Eq. (7.24) may be written as
În usin g this n a m e, w e m u st n o t , o f c o u r se , fo r g e t th a t th e th ic k n e ss o f th e lay er in

w h ic h th erm al fa c to r s play n o e sse n tia l part is in fa ct e q u a l to th e le n g th L m u ltip lie d b y a
c o n s ta n t f a c t o r - w h i c h a c co r d in g to th e d ata q u o te d in th e n e x t s e c tio n is o f th e order o f
several h u n d r e d th s.
Thus it is clear that the function f{z), together with the roughness
parameter Zo, defines the velocity profile uniquely.
The dominant role of the dynamic factors for z<^\L\ must
naturally be connected with the fact that the Richardson number Ri
is small for such z. In fact, limiting ourselves to the more convenient
flux Richardson number Rf = aRi, it is easy to deduce from Eqs.
(7.20) and (7.26) that
l i mRf( 2 ) = 0 and : (7.27)

z -> o \ /z=n
thus within the dynamic sublayer
lRf|«TLT<^l-
For fairly small values of = the function cp(^) = ^ '(^ )m a y be

expanded in a power series
cp(C)=l + p,C+ p.,C’ - f . . . . (7.29)
Therefore, for the function /(C) close to neutral stratification, we

obtain
/ ( C ) - c o n s t - f In |C| + pi(; + -y C 2 + •••: (7.30)
similarly, the number Rf here is represented as a power series
= .... ( 7 .3 1 )
Since the data do not as yet permit reliable estimation of

several coefficients p„, only the linear approximation of the function
q) ( 0 can be of practical value
cp(C)«l + pC, (7.32)

/(C) ^ const + In |C| + pC, ^ +
" ^ ^ ’ (7.33)
Rf«C-K ^ = f - p ( - l) ' (7.34)
[here p is the coefficient Pi of Eqs. (7 .2 9 )-(7 .3 1)].

With regard to Eqs. (7.29)—(7.34) it must be kept in mind that

since ? ; > 0 and correspond to two qualitatively different
turbulent regimes, we have no grounds for assuming that the
function cp(^) is analytic at the point C = 0 (although it is continuous
at this point, since as 0 , the two regimes of turbulence coalesce).
Consequently, ^ > 0 and S < 0 may correspond to different values of
the coefficients o f the expansions (7.29)—(7.31) and, in particular,
to different values of the constant p. Moreover, the coefficients
must be considered as universal constants, of the same type as the
von Karman constant x, only if they are regarded as exact values of
the coefficients of the Taylor series of the universal function (p(0
[limiting ourselves to one definite sign of . However, at present, we
cannot determine the exact analytic form of(p(^) w ithout recourse
to some more or less arbitrary hypotheses. At the same time, this
function is not determined with sufficient accuracy by measurements
for the values o f its derivatives at the point ^ = 0 to be defined with
any great precision. Thus, in practice, the coefficient (3 is found
generally by approximating the empirical function cp(C) over some
range of values of ^ by means of Eq. (7.32) or by some other
analytical expression. Consequently, the value of p obtained depends
considerably on the choice of the approximation interval and on the
equation for q)(?); hence it is not surprising that different investi
gators give widely varying values o f this coefficient (see below, the
end of Sect. 8.2). However, all these discrepancies have no effect on
the question of the sign of fJ, the answer to which follows directly
from elementary physical considerations. In fact, for stable stratifica
tion (i.e., q< 0 , L > 0 and S> 0 ), turbulent exchange is impeded,
and hence the mean velocity profile must be “steeper” than the
logarithmic profile corresponding to neutral stratification. On the
other hand, for unstable stratification (^ > 0 , L < 0 , ? < 0 ), the very
intensive turbulent mixing must lead to equalization of the mean
velocity, so that here the wind velocity must increase more slowly
with height than in the case of neutral stratification. Thus p
must be positive for L > 0 and negative forL < 0 ; i.e., p > 0 both for
C > 0 and C < 0 .
As we have already pointed out, Eqs. (7 .3 2 )-(7 .3 4 ) are significant
only for fairly small absolute values of t. Turning now to the
problem of the behavior o f q)(^) and /(C) for very large values of
IC| = - j ^ , we shall begin with the case of great instability; i.e., large
negative values of The asymptotic behavior of cp(^) and f(^) as

—oo may be studied by considering large values of 2:;3 g>l L \ for given
q> 0 , M* and g/To (i.e., for given L < 0 ) or by studying the limiting
process 0 (that is, L < 0) for given g > 0, g/To and 2 . The latter
limiting process clearly corresponds to the approach to purely
thermal turbulence in conditions of “free convection,” characterized
by the presence of unstable stratification with q > 0 and the absence
of mean horizontal velocity and the associated friction on the
underlying surface. In this case, the turbulence obtains its energy
not from the mean motion, but from the temperature instability. In
addition, it has the character of an assemblage o f thermal jets arising
at individual points o f the underlying surface, and rather weakly
intermixed. Further, if the existing horizontal mixing nevertheless is
sufficient to ensure the invariance of the probability distributions of
the flow variables with respect to shifts and rotations in the
horizontal plane, then such turbulence is a particular case of the
turbulent flows considered in this chapter. These flows are character
ized by only two parameters: g/To and qlcppo (since in “pure”
convection, u *= 0). However, from the similarity viewpoint, the
turbulent regime at heights z~^\L\ for given cannot differ from the
regime when 2 is fixed and not too great but m* is very small. Thus in
the unstable case, turbulence at great heights is always determined
primarily by thermal factors; i.e., the statistical characteristics of the
temperature field will depend in this case on g/To and q/Cppo but not
on From this viewpoint, in the limiting case it is
simplest to consider first, the behavior of the functions /i(^) and
gi{^) = f{iV> which characterizes the mean temperature distribu
tion, and only then turn to the functions [orq ) ( 0 = ^g(C)] and
/(g) [since in the study of the horizontal velocity distribution we
cannot appeal to the limiting case of “pure” thermal convection, in
which, generally speaking, there is no mean horizontal velocity at
all].
Since it is impossible to formulate any length scale from the
parameters g/To and q/cppo, the turbulent regime in conditions of
free convection must be self-preserving (see, for example, the
investigation of convective jets in Sect. 5.9). In particular, it is easy
to verify that Eq. (7.16) and the second of Eqs. (7.24) will not
contain ««only if gi(C) and /i(g) const. Thus
g, (C) = — . A (C) = const + c,x/’r (7.35)
when !:<C — 1
(the factor x‘ ‘ is added here for convenience). In other words,
o r- c, / ,
dz 3 \ C pPo
(7.36)
when 2 ;^>|L|, L < 0 . Equations (7.36) were first pointed out by

Prandtl (1932a) and later, independently, by Obukhov (1946). They
are completely analogous to the equations deduced by Zel’dovich
(1937), Eqs. (5.103)-(5.104) for the temperature distribution in the
center of convective ascending jets and differ from them only in that
here we are concerned not with convection above a heated sphere or
cylinder, but with convection above a large heated plane surface.
Later, these equations were discussed in detail in several works [see,
for example, Monin and Obukhov (1953; 1954), Priestley (1954;
1955; 1956), Kazanskiy and Monin (1958)]. From them it follows,
in particular, that in unstable stratification, the temperature distribu
tion at great heights tends to isothermal; this is due to the presence
of very intensive mixing (which increases with height as a result of
the appearance of larger and larger eddies with length scales
comparable with the distance from the underlying surface), leading
to equalization of the temperature. The eddy thermal diffusivity in
conditions of free convection from Eq. (7.36) is equal to
with increase of the distance from the earth’s surface, this increases
rapidly.
It is now possible to present arguments which give quite definite
asymptotic form to the functions g(^), (p(^) = t|r(^) and f(^) a s ? - ^
— oo. Since the limiting process — oo is equivalent to
(with the remaining parameters and z fixed), the turbulence regime
with 1 cannot depend on u,.. However, it is impossible to
formulate any dimensionless combination from (//cppo, gITn, poand z;
thus all the dimensionless characteristics defined by these parameters
must have a constant (universal) value. In particular, it is natural to
assume that one such dimensionless parameter is the ratio of the
eddy diffusivities for heat and momentum a = KrIK. Thus, in free

convection conditions
'h . - = const. /c = f 7^ (7.38)

/v ■ ' «-.ô Vc,,Po 7o/
[where a_ -- lim a (C )l. However, in this case, it is clear that

I -> - to
g(C) = — «f (C )^ — when C<C — 1. (7-39)
/(C) = + const when C C — 1, (7.39')
where Co ~ 'a_ In other words,
— = a-
dz 3 v . * \c^Po ^ 0 /
when q > 0 , z ^ ' ^ \ L \ , z->''^\L\. Equations (7.40) were also obtained

first (in slightly different form) by Prandtl (1932a) [see also
Obukhov (1946), Monin and Obukhov (1954)]. With increase of z,
the mean velocity u(z) approaches a constant value for the same
reasons which, in free convection, cause the approach to isothermal
conditions as o o . We note that u* now appears in Eqs. (7.40),
since u{z) = 0 when = 0 , and therefore the function u{z) cannot
be independent o f The Richardson numbers Ri and Rf will also
depend on in free convection conditions these numbers, by Eqs.
(7.39), (7.19), and (7.20), are equal to
'^This p la ces th e d e r iv a tio n o f Eqs. ( 7 . 3 8 ) - ( 7 . 4 1 ) o n a slig h tly w e a k e r fo o t in g th a n th e

d eriv a tion o f E qs. ( 7 . 3 5 ) - ( 7 . 3 7 ) . In fa c t, th er e is n o vertica l m o m e n tu m transfer under
id e a liz e d free c o n v e c t io n c o n d it io n s , a n d , th e r e fo r e , str ictly sp e a k in g , th ere is also n o e d d y
v is c o s ity K. C o n s e q u e n tly , it is th e o r e tic a lly p o s sib le , for e x a m p le , -^ « » a s
0 . H o w e v e r , n o p h y sic a l r ea so n s e x is t to s u p p o s e th a t th e c o n tr ib u tio n to th e e d d y v is c o s ity
o f th e v e lo c ity flu c tu a tio n s p r o d u c e d b y p u rely th erm a l fa c to r s w ill v a nish. T h u s, at p r e se n t,
it see m s rea so n a b le to c o n sid e r E q. ( 7 . 3 8 ) w ith 0 < a _ ^ < ® ° a s a pla u sib le h y p o th e s is w h ic h
requires e x p e r im e n ta l v e r ific a tio n .
TURBULENCE I N A T H E R M A L L Y S T R A T IF IE D M E D IU M 437
Thus it is clear that with increase of z, both Ri and Rf increase in

absolute value without bound (remaining negative the entire time).
Let us turn now to the second limiting case ^ ^ + c » . The
investigation of the asymptotic behavior of the functions (p( 0 and
f(^) for large positive ^ corresponds to the examination of the mean
velocity profile z7(2)for large 2 in the case of stable stratification
(fixed L > 0 ) or to the examination of the case o f very small positive
L for fixed 2 (i.e., very sharp temperature inversions). However, in a
sharp inversion with a vanishingly small wind, the turbulence
becomes degenerate and the motion of the medium takes on a very
special character. In fact, in conditions o f highly stable stratification,
the existence of large-scale turbulent fluctuations becomes impos
sible (since these fluctuations would have to expend too much
energy in performing work against the gravity forces), and turbulence
can exist only in the form of small-scale eddies (this can also be
explained by the statement that in this case large-scale waves are
stable and do not undergo transition to turbulence). In the case of
still greater stability, even small-scale turbulence will become
practically impossible, and the fluctuating motion of the medium
will probably appear in the form of a random collection of internal
waves. In any case, it is clear that high stability considerably im
pedes turbulent exchange between different layers of the fluid and
hence turbulence will take on a local character; the characteristics of
turbulent exchange (for example, the eddy viscosity and the number
Rf = ) forg ^ l(i.e., for fixed L > 0 but large height z ' ^ L , or
for fixed height e but very small L > 0 ) cannot depend expUcitly on
the distance z from the underlying surface. The latter deduction can
also be confirmed by the following argument: according to Richard
son’s deduction (see Sect. 6.5), the flux Richardson number Rf
cannot exceed unity under any steady conditions; i.e., the maximum
possible value Rfcr of this number 1 (apparently Rfcr is
considerably less than unity, but, at present, this is not im portant for
our purposes). On the other hand, it is natural to assume that the
variation of Rf as C= increases (i.e., for fixed 2 , with increase of
stability) must be monotonic: physical principles which could lead to
a situation where Rf began to decrease once again with increase of
stability, are difficult to imagine. However, if Rf(^) increases
monotonically with increase of and at the same time cannot exceed
some value Rfcr, as ^ - > 00 , it must approach some limiting value R
(which it is natural to identify with Rfcr).
Thus, in the case o f stable stratification, a universal value R must

exist such that
= const, K ~ ‘*-u^LR when L. (7.42)
But from this it follows that
g (^) = 4 = const, cp(g = = CaC,

t (7.43)
/(C) = c o n s t+ = c o n s t+ C3C when
where C3 = \/R = 1/Rfcr> and, o f course,
u (z) = 2 + const - C3 ^ + const. (7.44)
Thus, at great heights, in a stable medium, the mean velocity

increases linearly with height, while the constant gradient of this
velocity is determined uniquely by an “external” parameter, which
varies from case to case, e.g., the parameter =
CpT which in
fact is the same as the ratio of the turbulent heat flux to the
turbulent shear stress. This result was pointed out by Obukhov
(1946) [see also Monin and Obukhov (1954)].
According to Eqs. (7.29), (7.30), (7.39), (7.39'), and (7.43), the
universal functions tp(C) and / ( C ) —/ ( ^ ± y ) [the term —/ ( ± y )*®
added here to avoid indeterminacy in the choice of the reference

level of the function f(?)] must have a general form close to that
given in Figs. 47 and 48. In the following section this conclusion is
well supported by direct measurements in the surface layer of the
atmosphere.
Now we shall discuss briefly the general form of the functions
g'i(S). fi{^) and g 2 { l) ,h { ^ ) which describe the vertical profiles of
temperature and concentration of a passive admixture (humidity).
According to Sect. 5.7
/ i ( C ) « ^ In |C |+ c o n s t when |C|<C1: (7.45)
here ao = a ( 0 ) is a constant close to unity (which was denoted

T U R B U L E I. IN A T H E R M A L L Y S T R A T IF IE D M E D IU M 43 9
simply by a in Sect. 5.7). On the other hand, as we have seen, at

large negative S in “free convection” conditions, the behavior of the
function /i(^) is determined by Eq. (7.35); in this region the function
t\{Q differs from /(?;)only by a constant multiplier l/a_«, (the exact
value of which is still unknown, see below. Sect. 8.2). Later, we shall
see that when 5 ^ 0 , the transition from the “forced convection”
conditions, characterized by logarithmic profiles o f velocity and
temperature, to the “free convection” conditions defined by Eqs.
(7 .3 5 )-(7 .4 1 ) is comparatively sharp; i.e., it occurs over a small
range of values of Thus the behavior of the function /i(^) on the
whole positive semiaxis is described fairly completely by the values
of the constants x, Ci and ao (or x, Cj = Cix‘%-oo, ao and a-oo). In
particular, whether or not it is possible to consider the function fi(^)
as similar to the function /(^) when ^ < , 0 is probably determined
first by how close the values of the constants ao and are to each
other. The general form /i(C )—/ i ( — -j) 0, however,
certainly cannot differ from that shown in Figs. 47 and 48 as the
form o f the functions /(C) — / ( ~ 4 ) '
FIG . 4 7 . S c h e m a tic fo r m o f th e u n iversal FIG. 4 8 . S c h e m a tic fo r m o f th e universal

fu n c tio n cp (Q .
fu n c t io n / ( 0 — / ^ ±
The behavior of the function /, ( 0 when S > 0 is somewhat more

complicated. Since
/i(^ )~ y * co n st, (7.46)

the problem of the form fi (?) is closely connected with that of the
dependence of a(?) on the stability parameter (or on the
numbers Rf and Ri). If we assume as is done most often in
meteorological literature that - ^ = a = const, then the function
fi(^) will agree with f(Q up to a constant multiplier; if a = 1 (as
is also assumed), then /i (? )= /(? ), up to additive constants which
play no part. For small positive the function ^(?) is given by the
logarithmic formula (7.45); for somewhat larger C. by the “logarith
mic + linear” equation
A (Q - const f 1 (In C+ p,C) (7.47)
[with coefficient pi which may differ from the corresponding

coefficient in Eq. (7.33)]. If we assume that a(?)varies comparatively
little over a fairly wide range of positive values of including also
part of the region in which there is a linear velocity profile, then, for
sufficiently large values of ? within this range, the temperature
profile also will be described fairly well by a linear equation of the
form
A (Q ==c o n s t+ r ( 2) = ^ ( ^ f - f ^ + c o n s t , (7.48)
where a is a constant of the order of unity. However, observations of

turbulence in a highly stable sea and associated measurements in
laboratory water flows definitely show that in a very stable medium
cc(^) takes very small values (see below. Sect. 8.2, and, in particular.
Fig. 69). In other words, for very great stability, the eddy diffusivity
for heat K t is considerably less than the eddy diffusivity for
momentum K. According to Stewart (1959), physical arguments
explaining the reason for this are difficult. The case of an extremely
stable medium may be represented by a layer of heavy fluid (for
example, water), above which there is a considerably lighter fluid
(for example, air). Then turbulent motion in the lower fluid will lead
to disturbances of the free surface and to the appearance of
individual “splashes,” penetrating into the upper fluid, and then
falling again under the action of the gravity forces. The penetration
of “water” into the “air” will set up fluctuations of pressure in the
“air” causing a definite exchange of momentum between the two

media; at the same time, heat exchange is, practically speaking,
absent (we recall that we have agreed to ignore molecular thermal
conductivity). Thus we may assume that for very high stability
(possibly greater than that which is actually observed in atmospheric
inversions), the eddy viscosity K will have a finite value, while /(rwill
be close to zero. However, if this is true, then it follows that at very
large positive the temperature profile f { z ) will be consider
ably steeper than the velocity profile S( 2 ); it follows from the fact
that K t -*- 0 as S -> o o ,th at the steepness of the temperature profile
increases without bound as zjL increases. Consequently, the general
form of the functions / , —/ j j and 9 , (C) ^ C/| (C), at very large
positive ^ will differ from the general form of the functions /(C) —
/ ( ^ ) and 'f(C) [see Figs. 49 and 50]. Also connected with this is the
Rf
fact that the Richardson number Ri = —- in conditions of high
stability, may apparently assume very large values (since if a -> -0 as
Rf Ript, then Ri o o ) ; thus there is no limiting “ critical” value
of Ri (unhke the case of Rf).
FIG . 4 9 S c h e m a tic fo r m o f th e universal FIG. 5 0 . S c h e m a tic fo r m o f th e u n iversal

fu n c tio n (C).
fu n c tio n / i (C) -
We consider now the functions / j (Q and gj — f'i (*^)- As long as it

was widely held that K = K t (i e., that a = 1 ), it was completely
natural to assume also thatKi} = K = KT, that is, the function /'(C) did
not differ from / ' (C) and f \ (C). However, abandoning the assump
tion that eddy diffusivities for momentum and for heat are equal, we
are immediately faced with the problem of the value of the eddy
diffusivity for an admixture K^, which determines the form of the
functions fziO and Below, we shall see that the data on the
values of /Caand are even poorer than those on the values of A'/
and Therefore, it is impossible to draw any truly reliable
conclusions from them. Thus it only remains to have recourse to
physical intuition. In this connection, we may note that in the paper
by Priestley and Swinbank (1947), one o f the first in meteorological
literature in which nonequality of the different eddy diffusivities was
discussed seriously, it was proposed that in view of the special role of
temperature fluctuations in convective motions (expressed, for
example, by the fact that particles that are warmer than the
surrounding medium prefer to move upward), the eddy thermal
diffusivity Kr may have one value, while the eddy diffusivities for
momentum K and humidity /(:> will coincide at another value.
However, the reasoning o f these investigators is not very clear, and
their viewpoint is far from being shared by all [see, for example,
Robinson (1951)]; their arguments seem especially unconvinc
ing with regard to the surface layer of the atmosphere. The more
convincing view is that by reason of the similarity of the physical
mechanism of heat and humidity transfer (or the transfer of some
other passive admixture), which takes place only by means of direct
mixing of the air masses, the eddy diffusivities Kr and /<» will be
equal while the eddy viscosity K is also affected by the pressure
fluctuations and may differ from /(r = /<». This latter view is held at
present by the majority of specialists [see, for example, Ellison
(1957), Charnock (1958a)]. Here, we should also take into account
that for not too great temperature inhomogeneity, the selective
effect of the buoyancy on the fluid elements will produce no great
effect on the turbulent heat transfer. Therefore, so long as we do not
know of any other physical exchange mechanism that would affect
heat and humidity transfer in different ways, there is no reason to
reject the assumption that and, consequently, that ^ 2 (S) =
/ 2 (£) = / i ( S ) + const. Of course, this assumption needs careful
verification, using the data of sufficiently accurate special observa
tions in nature and in the laboratory, which only began to appear
recently.
7.4 Further Discussion of the Universal Functions;

Interpolation Formulas and Semiempirical Formulas
In Sect. 7.3 we analyzed the general behavior of the universal
functions (p{ 0 = ^/'(O an d / , (C), = V,'(Q> and displayed
the asymptotic formulas which determine the behavior of these
functions for values of the arguments close to zero, large and
positive, and large and negative. However, in many cases, it is
convenient to have explicit formulas giving the values of the
functions for all Such expUcit formulas may be obtained most
simply by interpolation, using the known asymptotic laws, and also
taking into account the experimental data on the functions in the
intermediate range of moderate values of For this purpose we
may also use one of the variants of the semiempirical theory of
turbulence, generalizing it to the case of a stratified medium, or we
may simply attem pt to select expressions for the turbulence
characteristics on the basis of data only, w ithout any reference to the
theory. A considerable amount of literature has been devoted to
these three approaches to the problem o f describing the turbulence
characteristics in a stratified medium. The literature comprises at
least several hundred papers, but here we shall deal only briefly with
certain of the relevent results, paying particular attention to more
recent works.
First, we consider the behavior of the functions f ( Qa n d fi{^) in
the region of negative Z- As we have already observed, for small
negative both these functions must be approximated by the
logarithmic function. For values of S somewhat greater in absolute
value, we may use the “logarithmic + linear” formula (7.33) or
(7.47); while for very large negative the functions/(^) and fi{l)
will approach a constant value, their deviation from the correspond
ing constants being damped Hke [see Eqs. (7.35) and (7.39')].
Above, we also recalled that according to the data, the transition
from the “forced convection” regime described by the logarithmic
functions/(^) and/i(^), to the “free convection” regime correspond
ing to Eqs. (7.35) and (7.39') apparently takes place without any
considerable intermediate region (and, moreover, for quite small
values of \^\ = z/\L\, of the order of a tenth or several hundredths; see
below. Sects. 8.1 and 8.2). Following this, Kazanskiy and Monin
(1958) assumed that the values o f f (^) on the whole negative semiaxis
will be described fairly well by the simplest interpolation formula:
ln|C| when C, < C < 0,

/(C) (7.49)
In Ic, I + C2 (C~ - cr ■'’) when C< C,.
Similarly, for
ln|Cl
“0
(7.50)
where Cix'*= C2/a_oo. The values of the dimensionless constants cto,

a-COand which determine the extent of difference between the
functions /(^) and /i(^) must be refined further on the basis of the
data. At present, it is possible to assert only that the coefficient Oq is
close to unity (but the assumption which is often allowed that all
these constants are equal to unity is not justified). As for the
constants C2, C2/a_oo=Cix*/^ and ^1 which occur in Eq. (7.49), their
values will be discussed in greater detail in the following section.
Applied to the functions <p(!;) = C/' (C) , and cp^(!;)==
a„C/'(C) = aotp, (C) , Eqs. (7.49)—(7.50) clearly lead to the following
relationships:
1 when C, < C< 0 ,

(7.51)
and similarly
1 when C2 < C < 0,

c' (7.52)
. _ 2_r-v» when C < !: 2,
where C,x%o = ^2 = —^ C j. Naturally, these functions will be

00
discontinuous since the functions (7.49) and (7.50) are continuous

but not differentiable at the points Ci and C2, respectively.
In spite of the presence of a discontinuity in the derivative, in
many respects the functions (7.49) and (7.50) are quite sufficient to
describe the wind and temperature profiles in the surface layer, with
arbitrary, unstable stratification. However, when it is desirable to

have functions /(C) and fi(?) with continuous derivatives, we can, for
example, following Priestley (1960a), reject the determination ofî
and from the data, and instead write C, = — ) and Cj = — | j
[the functions (7.51) and (7.52) will then, o f course, be continuous].
The same result may also be obtained if, following Kazanskiy and
Monin (1958), we retain the hnear term in the equations for /(£:) and
/i (^) for small values o f i.e., if we assume, for example, that/(C) =
l n | q + pc, cpO:)= l + pc for 0 > c > c ; „ a n d / ( O ^ i n l C i l + pC.H-
CjCC"''’ — cp(!;) = — — for Here p may be chosen in
such a way that the function 1 + p? gives a good approximation to
the experimental values of q>(^) for small and is then determined
from the conditions 1 K i=
— o whi ch ensures the
continuity of cp(C) at the point ? = Ci. On the other hand, following
Priestley (1960a) and taking both constants p and to be
undetermined, we may choose these constants in such a way that the
function
1 -t- pC when C, < C< 0,

(7.53)
when ; < Cj
will have a continuous derivative at the point . Then
f{ q = f
5
will have two continuous derivatives for all ? . In general, writing
cp(C) =
we can ensure that the function (7.54) has n continuous derivatives

and, consequently, that f(^) has an ( n + O t h continuous derivative.
However, to do this, it is necessary only to equate the values of the
functions i-(_ p ,C + a n d ---- and their first n
derivatives to each other at the point S=Si, and then evaluate the
n+ 1 unknowns Pi, . . . , p„from the system of ra + 1 equations thus
obtained. It is not difficult to obtain explicit formulas for the
solution of this system of equations for any n; according to these
formulas, if n - ^ o o , then o o , but the coefficients p* all tend to
finite values
J l'" . = (3»+T)‘t l c f ■ ^ ^2.7 a.S5)
[see Priestley (19 6 0 a)]. The hmiting values (7.55) of the coefficients
P;, correspond to an infinitely differentiable and, moreover, analytic
limiting interpolation function
T(C) = ^ / e>‘dt. (7.56)
Completely analogous interpolation formulas may also be formulated

for the function (pf, the sole difference being that the coefficient C2
must now be replaced by C '2 — — CjX''’ag. Thus, giving n the
values 1 , 2 , . . . and <» we obtain an infinite number of interpolation
formulas of the form (7.54) for the functions (p(C) and <Pr(^)- It is
found that these interpolation formulas [including the limiting
formula (7.56)] differ very Httle from each other and describe the
existing data (which are characterized by considerable scatter) in
practically the same manner. To illustrate the closeness of the various
interpolation formulas to each other. Fig. 51 shows a diagram
borrowed from Priestley (1960a). This diagram represents the
dimensionless heat flux
• 9
dT V2
1 g
dz
(which becomes constant in conditions of free convection), obtained

by using interpolation formulas of the type (7.54) for (pr(0 with n =
0, 2 and 5. In the calculations used as a basis for Fig. 51, it was
assumed that «o= 1 (and, in general, a(C) = 1), 0.4 and
C'2 ~ C‘2 — 0.95 (this last value corresponds to certain data on
temperature profiles in the atmosphere); here, Ri = Rf = ^ is
-R i
FIG. 5 1 . D e p e n d e n c e o f th e d im e n sio n le ss h e a t flu x q o n th e

R ic h a r d so n n u m b e r Ri. T h e b r o k e n Une c o r r e sp o n d s t o th e f u n c
tio n o f fo r m ( 7 . 5 2 ) w ith f 2 = - ( C a / S ) ^ ; th e d o t t e d a nd c o n
tin u o u s cu r v es c o rr e sp o n d to in te r p o la tio n fo r m u la s o f th e fo r m
( 7 . 5 4 ) for <P7 ' ( f ) w ith n = 2 and « = 5 .
plotted on the abscissa (if, instead of this, we used the argument

C= ^ , the divergence between the curves would be even less). In
general, with Ca= 0.95, the divergence between the values ofcp(^)
corresponding to the “ zero-order” interpolation formula (7.51)
[with Ci = — and the “limiting” formula (7.56), nowhere
exceeds 15%, while a deviation of close to 15% is observed only for C
close to Cl and is basically connected with the presence on the curve
(7.51) of the jum p discontinuity at C=Ci- The first smoothing, with
the aid of Eq. (7.53), immediately results in values of (p(S) which
nowhere differ from the values of Eq. (7.56) by more than 5%. Thus
it is clear that when dealing with observations in the atmosphere, the
use o f interpolation formulas of the type (7.54) that are more
complicated than Eq. (7.53) is not greatly significant. From this
viewpoint, there is hardly sufficient justification for Webb’s proposal
(1960) to use an interpolation formula of the form
( 1+ for 0,
<p(g= (7.57)
for c<c„
where the coefficient €■>, is determined from the data, and p, C, and
are chosen from the condition that the function (7.57) should
have two continuous derivatives at the point t = h (or else is
found from observation and P,C2 , and C are chosen according to the
aforesaid condition).
Until now, th e case of positive f has been studied very little
experimentally. Therefore, it seems reasonable to use the simplest
interpolation formula for the fu n ctio n s/(f) and(p(5) = f/'(f). 0 < ? < ° °
of the form
/(f) = ln? + /3;r,(p(C)=l+îf (7.58)
which gives correct results both close to zero and for large positive
values of f.
A different type of interpolation formula may be obtained if,
following Ellison (1957), we take C= 0 (i.e., at neutral
stratification) R f -> 0 and K~y.u,^, while for 1 ^ 1 ^ 1, (i.e., for free
convection,—Rf and Rf [if we assume
that 0 < a .„< « > ; see Eq. (7.38) and (7.41)], i.e., » (_aRf)'\
where oris some constant [equal, as may easily be seen, to ( 3 /C 2)^].
Starting from this, we may expect that for the whole range 0 > Rf
> —oo(i.e., along the semiaxis 0 > C > —oo) the dependence o f/(
on Rf will be described fairly well by an equation of the form
K = m ,z{\ (7.59)
which gives correct results for both small and large values of Rf. At
the same time, it is found that when 0 = 1/Rfcr, Eq. (7.59) gives
asymptotically correct results in the limiting case of very stable
stratification as well. In fact, it is clear from Eq. (7.59) that as Rf
increases, it can never exceed the value I / 0 and that K l z - ^ 0 as
R f-> 1/(T (i.e., as C= ^ -> + c « ), in complete accordance with Eqs.
(7.42). Thus, it may be expected that by using Eq. (7.59), which
contains a single empirical constant a, we shall be able to give an
approximate description of the mean velocity profile in a turbulent
stratified medium under any stratification. It is difficult to depend
on the value of a = l / R f ^ being identical to the value
which leads to true asymptotic behavior of K (Rf) at large negative
values of Rf. However, since at present the values of Rfcr and of C,
are known only very imprecisely, we do not as yet have enough
justification for using different values of a in Eq. (7.59) for Rf < 0
and for Rf > 0.
Equation (7.59) may also be obtained, beginning with the simplest

attem pt at generalizing the semiempirical theory o f turbulence to a
stratified medium. In fact, as we have seen already in Sect. 6.5, the
semiempirical equations for the turbulent energy balance in a
stratified medium, may be transformed to
K = u J ,(\-c R l)''\ (7.60)
where /i = x^X(Rf) is the “length scale of turbulence,” suitably

defined. If it is now assumed that A,(Rf)= 1, i.e., that the scale U for
any stratification is given by one and the same equation li = xz
(which is true in the case o f neutral stratification), then Eq. (7.59)
follows immediately. In precisely this manner, this equation was first
deduced by Obukhov (1946), who used it to investigate the
asymptotic behavior of the functions and (p(S) at large positive
and large negative values of C. In recent years the various methods of
deducing Eq. (7.59) have been discussed in detail by a number of
researchers [see, for example, Yamamoto (1959), Panofsky (1961a),
Priestley (1961), Sellers (1962), Syono and Hamuro (1962), Klug
(1963), and Rijkoort (1968)]. However, since all the deductions pro
posed use certain arbitrary hypotheses, they add little to the views of
Ellison (see above), who considered Eq. (7.59) simply as an
interpolation formula for AT(Ri), which gives correct results for large
negative values, values close to zero and large positive values of
Since K = and Rf = Eq. (7.59) may be rewritten as the

following simple algebraic equation for the function cp(?)
cp4_o;(p3_i = 0 . (7.61)
This equation contains a single unknown constant o, which may be

eliminated by introducing, instead of C= -|- as an independent
variable, the quantity nj = o C= w h e r e = —.

0
Then the equation
takes the form
<p4_.^(p3_i^0. (7.62)
Equation (7.62) was solved numerically by Yamamoto (1959) and

then also by Klug (1963). Yamamoto has given a table of values of
the function 9 (11) [which is in good agreement with the schematic of
Fig. 47] , and some fairly simple analytical approximations of this
function on the positive and negative semiaxes; graphs are plotted
giving the dimensionless wind velocity xu/«. as a function of the
dimensionless height 2 /zo for different values of the stratification
parameter tio = azolL. Later, Syono and Hamuro (1962) and Oka-
moto (1963) formulated the solution of the fourth-degree equation
(7.61) in radicals, and carried out a more detailed investigation into
the form of the function (p(ti) and the wind-velocity profiles
corresponding to it. However, we shall not dwell on the detailed
study o f the solution of Eq. (7.61), which is of Httle interest to us; in
any case, it is only one o f the possible interpolation formulas for cp
(and, moreover, is clearly not a very exact one).
Let us also note that for f < 0 , the arguments in favor of Eq. (7.59)
may also be used to justify the equation
K = Ku^z(l-a'Ri/^ (7.59')
where a' is another constant (namely, a ' = a_„a). If we now replace

the length L by Z,'= w^(a«/3 z)//<(^/7’o) (dr/az) = a(f)/-, and the variable
f = z/Z,'by f' = z/L', then Eq. (7.59') may be rewrittei; in a form
similar to Eq. (7.61), but with the replacement of f by f' [and<p(f)
by <p(f')l and of a by o'. This equation for <p(f') was used by
Panofsky, Blackadar and McVehil (1960) and by several others (see
Sect. 8.1).
T h e fa c t th a t E q . ( 7 . 6 1 ) c a n n o t b e e x a c t fo llo w s , in p articular, fr o m th e fa c t n o te d
a b o v e ; i.e ., th is e q u a t io n c o r r e sp o n d s t o th e p h y sic a lly u n lik e ly a s su m p tio n th a t th e scale o f
tu rb u le n c e l\ a t h e ig h t z is in d e p e n d e n t o f th e te m p er a tu re str a tific a tio n . H o w e v er , i f w e
assu m e th a t
(7 .6 3 )
'■- " ‘(f)
th e n w e o b ta in fo r (p th e e q u a t io n
<p4_o(;<p« — i = 0. (7.64)
w h ic h c o n ta in s a fu rther u n k n o w n universal fu n c tio n [c f. Y o k o y a m a ( 1 9 6 2 b ) J .

In th e case o f sta b le str a tific a tio n , tu r b u le n t e x c h a n g e b e c o m e s m o r e d if fic u lt , and large
tu rb u le n t fo r m a tio n s c a n n o t e x is t (and at R f > R f c r = l a th e tu rb u le n c e c a n n o t e x is t at all).
rhcrefore, it is natural to cxpect that for positive f the function \ will decrease
nionotonically from \(0 ) = 1 to zero as f increases from zero to infinity (or as Rf increases
from zero to Rfcr). As a concrete example of the function X(Rf) Kazanskiy and Monin
(1956) considered the case when X (Rf) = (1 — a Rf) where 7 > -^ is some number.
With this choice of /.(Rf), Eq. (7.60) becomes
/( = 7 .(v (i-o R n ^- (7.65)
(This relationship, with 7 = 1/2 was used earlier by Holzman (1943) as an empirical formula
for K.) Expressing K and Rf once again in terms of the function (f (^}, we find that Eq.
(7.65) is equivalent to the equation
_1
1 — (7.66)
> (0
Equation (7.66) shows that when r] = a£ > 0 (i.e., in stable stratification) rp > 1, while
(p(0) == 1 and (p(v)) ->00 as For small r\, we may put cp(T]) = 1 + 6 , where
6 1; we find easily that 6 = \r\; i.e., q) ~ 1 + ayS Thus the product ay = |3 must be
identical to the coefficient /3 in Eq. (7.32). Further, for large r) = af, the asymptotic
relationship cp ^ will be satisfied, showing that a = ~ (7-43)].
/<
Kazanskiy and Monin evaluated (p(T^) and /()]) for the special case 7 = 0.6; the dependence
of the dimensionless eddy viscosity ol<l'ui,,L on the dimensionless height r\ = Oz/Ly
obtained on the basis of this calculation, is very close to the exponential dependence
— 1— which is used in a number of meteorological works [Dorodnitsyn
(1941), Matveyev (1960) and oth ers]. The case 7 = 6 was considered by Borkowski (1964)
and the general equation (7.66) was investigated later in detail for several values of 7 by
Rijkoort (1968).
The approximation (7.65) is most suitable for positive ^ (and Rf). In fact, when applied
to negative f (for unstable stratification), it gives the apparently valid free convection laws
(7.38)-(7.41) as — co and Rf — co only if 7 = 1/4. Thus, with S < 0 it is
necessary to use either a special case (7.61) of Eq. (7.66) [i.e., to assume in advance that 7 =
1/4] or else reject the assumption on the existence of a finite limit cc _oo. It is also easy to
obtain a general form of the dependence of the scale of turbulence l\ on the stratification in
unstable conditions, which is compatible with the free convection laws. In fact, if we assume
that li y.zK{zlL) [so that \( f ) = 1 for approximation (7.61)], then by Eqs. (7.64) and
(7.39) we obtain
27 \
as C > — 00 (7.67)
(Here /.(c^ ) = const, i.e., l\ is proportional to z follows naturally from the absence in the
conditions of free convection of any fixed length scale other than 2.) For small values of z,
using Eq. (7.32), we obtain
(7.67’)
Thus, if ? > , then as f varies from zero to — 00 the function = IJy.z will first
in crease lin e a r ly , th e n its r a te o f g r o w th w ill d e c r e a s e , a n d as t, ->— co, it w ill te n d t o a

c o n s ta n t.
We m a y a lso b e g in b y c h o o s in g in s o m e w a y th e fu n c tio n X ( ^ ) ; .t h e n fr o m th e
fo u rth -d e g r e e e q u a tio n ( 7 . 6 4 ) it is ea sy to d e te r m in e th e fu n c tio n cp(^), i.e ., to fin d th e
w in d d is tr ib u tio n w ith h e ig h t, c o rr e sp o n d in g to a g iv en scale o f tu rb u le n c e li = x z K ( z l L ) ;
o n e e x a m p le o f s u c h a p ro ced u re m a y b e fo u n d in th e w o r k b y T a k e u c h i and Y o k o y a m a
( 1 9 6 3 ) , w h ere s o m e fairly ro u g h d ata w a s u sed fo r a cru de e stim a te o f o n e o f th e p o ssib le
“ sca le s” o f tu rb u le n c e an d th e n th e c o n e s p o n d in g fu n c tio n cp( J) w a s e v a lu a te d .
E q u a tio n s ( 7 . 6 1 ) [o r E q . ( 7 . 6 4 ) fo r giv en X i ^ ) ] and ( 7 .6 6 ) are particular e x a m p le s o f
th e fo r m u la s fo r rp(^) o b ta in e d fr o m s e m iem p ir ic a l th eo r ie s. A s w e have a lready seen in
C h a p t. 3, e v en fo r tu rb u le n c e in a h o m o g e n e o u s m e d iu m th ere e x is t a w h o le series o f
d iffe r e n t variants o f th e sem iem p ir ic a l th e o r y , w h ic h lead to slig h tly d iffe r e n t results.
T h e re fo r e , in th e c o n sid e ra b ly m o r e c o m p le x c o n d it io n s o f n o n n e u tr a l s tr a tific a tio n , th e
n u m b e r o f p o ssib le s e m iem p ir ic a l th eo r ie s is very large, a n d th ese th eo r ie s lead to very
diverse e x p r e s s io n s fo r cp(£) a n d th e o th e r universal fu n c tio n s . O f c o u r se , all th ese th eo r ie s
c o n ta in certain arbitrary a s su m p tio n s, w h ic h req uire fu rth er v e rific a tio n ; h o w e v e r , the
e x p r e s s io n s o b ta in e d fo r (p (£) in m a n y cases d o n o t e v en sa tisfy th e gen era l a s y m p to t ic
r ela tio n sh ip s d e d u c e d in th e p reviou s s u b s e c tio n a n d , th e r e fo r e , can be u sed o n ly fo r so m e
lim ite d range o f v a lu e s o f T h u s, fo r e x a m p le , in th e first sem iem p ir ic a l th eo r y o f
tu rb u le n c e in a s tr a tifie d m e d iu m p r o p o s e d b y R o ss b y and M o n tg o m er y ( 1 9 3 5 ) , it w as
a ssu m ed th a t K — w */i, a n d a s p e cia l h y p o th e s is w as used t o d e fin e th e d e p e n d e n c e o f th e
scale 11 o n th e h e ig h t a n d s tr a tific a tio n ; as a resu lt, th e a u th o rs c o n c lu d e d th a t
w h e r e a = c o n s t, i.e ., th a t
cp3 _ (p — aC = 0 (7 .6 9 )
if a ( S ) = 1 [c f. th e d e d u c t io n o f E q . ( 7 . 6 1 ) fr o m E q . ( 7 . 5 9 ) ] . I f \ l\ 1, E qs. ( 7 . 6 8 )
an d ( 7 . 6 9 ) lea d t o th e r ea so n a b le “ lo g a r ith m ic + lin ea r ’* e q u a t io n fo r th e w in d p r o file (w ith
/3 = a /2 ) ; h o w e v e r , th e a s y m p to t ic b e h a v io r o f cp(^) b o t h as ^ — o o and as ^ co
pro v es t o b e in c o r r e c t h e r e. In a n o th e r s e m iem p ir ic a l th e o r y d e v e lo p e d b y O gura ( 1 9 5 2 a ) ,
th e in itia l p rem ises are th e e q u a t io n o f tu r b u le n t en erg y b a la n c e and th e h y p o th e s e s th a t
uj\
K =■■ - ■ and a = 1. A fo r m u la w a s th e n o b t a in e d fo r th e w in d v e lo c ity , w h ic h in our
n o t a tio n m a y b e w r itte n as
+ + (7 .7 0 )
w h e r e a is an e m p ir ic a l c o n s ta n t. It is clear th a t w h e n 1^1 1, th is fo r m u la also is

e q u iv a le n t t o th e “ lo g a r ith m ic + lin e a r ” fo r m u la (w ith /3 = a / 2 ) . H o w e v e r , as ^ - > — o o , o n c e
again w e d o n o t o b ta in th e r e la tio n sh ip ( 7 . 3 9 ) , w h ic h c o rr e sp o n d s to th e “ 1 /3 -p o w e r la w ”
fo r th e w in d -v e lo c ity p r o file (a n d th e te m p er a tu re p r o file ) in c o n d it io n s o f free c o n v e c tio n
(in th e o th e r lim itin g case £ - > - | - o o th e co rr e c t a s y m p to t ic rela tio n sh ip is n o w sa tisfied i f
w e a ssu m e th a t o = 1 / / ? = l/R f^ f) - O gura’s e d d y v is c o s ity h y p o th e s is w a s also used by
B ussinger ( 1 9 5 5 ) b u t in c o n ju n c tio n w ith th e R o ss b y -M o n tg o m e r y h y p o th e s is o n th e scale /,
(in ste a d o f usin g th e s e m iem p ir ic a l e q u a tio n o f e n e rg y b a la n c e ). T h e result w h ic h he
o b ta in e d m ay b e w r itte n as fo llo w s :
9 (0 = 1 ----------------- ^---------------------. ( 7 .7 1 )
j l l - f ( 1 _ 4 0 'A ]_ c
A c co r d in g to E q. ( 7 . 7 1 ) , (f (5 ) - 1 + 2 ^ w h e n 1^ < C 1, b u t th e a s y m p to t ic b e h a v io r o f
p ( £ ) for large va lu es o f \t\ o n c e again p ro v es to b e in c o r r ec t for e ith e r sign o f (In th e case
o f stro n g s ta b ility , it is fo u n d th a t (p (£) e v en ta k es c o m p le x v a lu es.)
W ith th e aid o f a sp e cia l, q u ite a r tificia l, h y p o th e s is , S w in b a n k ( 1 9 6 0 ; 1 9 6 4 ) o b ta in e d
th e very s im p le fo r m u la
?( 0 = I J ^ - c ’ <'^•'^2 )
w h ic h , lik e E q . ( 7 . 7 1 ) c o n ta in s n o e m p ir ic a l c o n s ta n ts a t all. T his fo r m u la o n c e again has th e

c o rr e c t a s y m p to t e s [ o f th e fo r m o f E q s. ( 7 . 3 2 ) a n d ( 7 .4 3 ) , r e s p e c tiv e ly ] fo r very sm all and
very large p o sitiv e valu es o f ^ (w ith c o e f fic ie n ts p = 0 . 5 , /? = C3 = 1 ), b u t as £ — cxdit
d o e s n o t b e c o m e th e “ 1 /3 -p o w e r la w .” We n o t e fu rth e r th a t E q . ( 7 . 7 2 ) le a d s to a sim p le
e x p o n e n t ia l h e ig h t d e p e n d e n c e o f th e e d d y v is c o s ity K [p r o p o r tio n a l t o 1 - e x p ( - Z I L \
w h ic h see m s p ro b a b le in th e case o f sta b le s tr a tific a tio n . A fu rth er variant o f th e
s e m iem p ir ic a l th e o r y w a s p r o p o s e d b y K a o ( 1 9 5 9 ) , th e m ain resu lt o f w h ic h m a y b e w r itte n
as
9(0 = +
w h e r e a is an em p ir ic a l c o n s ta n t. Sim ilar t o all th e p r e c e d in g fo r m u la s, E q. ( 7 . 7 3 ) r ed u c es to
th e “lo g a r ith m ic + lin e a r ” p r o file in near n e u tra l c o n d it io n s , b u t w h e n £-> + 00 it has an
in c o r r e c t a s y m p to t e . A m o r e c o m p lic a t e d r e la tio n s h ip fo r q)(S) , w h ic h m a y b e w r itte n as
Vi
V(C)-aC = (7 .7 4 )
2<p( 0
w a s o b ta in e d in th e sem iem p ir ic a l th e o r y o f B usin g er ( 1 9 5 9 ) . T h is r e la tio n s h ip is

a d v a n ta g e o u s in th a t th e co rr e sp o n d in g fu n c tio n q>(g) e x h ib its th e c o rr e c t a s y m p to t ic
b eh a v io r b o t h fo r C- ► + 00 (w ith Rfcr = 1 /(T , th a t is, C 3 = a) a n d fo r £ — o o . x h e sam e
adva n ta g es are also s h o w n b y th e fu n c tio n
/« ) - /
fo u n d b y Z iU tin k ev ich a n d L a y k h tm a n ( 1 9 6 5 ) , startin g fr o m th e sem iem p ir ic a l en erg y

b a la n ce e q u a tio n ( 6 .4 8 ) [ w it h = 0 , i.e ., ig n o rin g th e d if fu s io n o f e n e r g y ] s u p p le m e n te d b y
th e fo llo w in g g e n e ra liza tio n o f v o n K drm dn’s h y p o th e s is ( 5 . 1 1 9 ) o n th e “ m ix in g le n g th ” /:
g f(r
dz
I = — Xl
du dû g d^T
2
~dz I P
(w h ere Xi a n d a are e m p ir ic a l c o n s ta n ts , a = K j j K ) . A m o r e g en era l sem iem p ir ic a l th e o r y

ta k in g th e e f f e c t o f th e e n e rg y d if fu s io n in t o a c c o u n t (i.e ., a ssu m in g th a t ^ 0 ) was
d e v e lo p e d n u m e ric a lly b y V ager ( 1 9 6 6 ) ; it a lso le a d s t o a d e fin ite fo r m o f th e fu n c tio n s </>(f)
an d / ( f ) . A t th e sa m e tim e , b y s e m iem p ir ic a l a r g u m e n ts Y a m a m o t o a n d S h im a n u k i ( 1 9 6 6 )
a d v o c a te d th e r e p la c e m e n t o f E q . ( 7 . 6 2 ) b y a pair o f e q u a t io n s
“ hkp^ - 2 q>^ + 1 = 0 fo r th e u n sta b le case (tj < 0 )

(7.75)
- h l^ - 2 cp^ H- 1 = 0 fo r th e sta b le case (77 > 0 )
w h ere p is a sm a ll n u m e r ic a l c o n s ta n t; th e y ca lc u la te d th e universal fu n c tio n s c o rr e sp o n d in g

to E q . ( 7 . 7 5 ) w ith p = 1 /6 . T h e s o lu tio n o f th e first e q u a t io n ( 7 . 7 5 ) is q u ite c lo se to th e
so lu tio n o f E q . ( 7 . 6 2 ) a t t? < 0 (a n d h as th e sa m e a s y m p to t ic b e h a v io r w h e n 17 - > 0 or t j
-®®), b u t th e s o lu tio n o f th e s e c o n d e q u a t io n ( 7 . 7 5 ) b e h a v e s lik e 1 + fo r sm a ll 17
and h l^ fo r large t j , i.e ., has a usual b e h a v io r w h e n p = 0 o n ly .
A n o th e r a p p ro a c h t o th e d e te r m in a tio n o f th e universal fu n c tio n s o f th e w in d and
te m p er a tu re p r o file s in th e lo w e r a tm o sp h e r e fo r u n sta b le str a tific a tio n w as u sed b y
P a n d o lfo ( 1 9 6 6 ) . H e a ssu m ed th a t th e w in d v e lo c ity sa tisfie s th e “lo g a r ith m ic + lin ea r ”
e q u a tio n in th e ran g e 0 > f f i o f values o f f , w h ile th e te m p er a tu re p r o file sa tisfies th e
“ 1 /3 -p o w e r la w ” ( 7 . 3 5 ) - ( 7 . 3 6 ) fo r all f 1 . T his im p lie s th a t fo r th e fu ll d e te r m in a tio n o f
th e fu n c tio n s (p(f) a n d ( p i ( f ) o n th e w h o le sem ia x is f < 0 it is n ecessa ry to k n o w o n ly th e
fu n c tio n K t I K = a ( f ) . T h e la st fu n c t io n w a s d e te r m in e d b y P a n d o lfo w ith th e aid o f th e
q u e s tio n a b le h y p o th e s is th a t f = R i fo r all f < 0 . It f o llo w s fr o m th e h y p o th e s is and th e
sta te d a s su m p tio n s o n th e w in d a n d te m p er a tu re p r o files th a t a ( f ) = 1 / ( 1 + /3f) w h e n 0
> f 1 and « ( f ) If 1^/^ w h e n f 1 . N o w it is q u ite e a sy to w r ite d o w n th e fin a l e q u a tio n s
fo r th e fu n c tio n s (p (f) a n d ( p i ( f ) fo r all v a lu e s f < 0. S in ce a (f)-> » o o a sJ | -> -- 0 0 a c co r d in g to
th e P a n d o lfo th e o r y it is clear th a t th e co rr e sp o n d in g v e lo c ity p r o file u ( z ) d o e s n o t sa tisfy
th e “ 1 / 3 -p o w e r la w .”
O n e m o r e set o f e q u a t io n s fo r v e lo c ity a n d te m p e r a tu r e p r o file s in th e u n sta b le case (i.e .,
at f < 0 ) w as p r o p o s e d b y B usinger ( 1 9 6 6 ) , b a sed o n n o t very relia b le pu rely e m p irica l
a rg u m en ts. T h e se e q u a tio n s h a v e th e fo r m
(P(f) = (1 - fp i ( f ) = (1 - « ( ? ) = (1 - . (7-76)
We see also th a t here S' = R i fo r all f < 0 a n d th e “ 1 /3 -p o w e r la w ” is n o t valid e ith e r for

v e lo c ity or fo r te m p er a tu re p r o files.
Still fu rth er e x a m p le s c o u ld b e g iv e n o f th e use o f s e m iem p ir ic a l h y p o th e s e s fo r th e
e x p lic it d e te r m in a tio n o f th e fu n c tio n s (p(^) and f ( ^ ) [se e , fo r e x a m p le , E llio tt ( 1 9 5 7 ;
1 9 6 0 ) , K ra v c h e n k o ( 1 9 6 3 ) , N a ito ( 1 9 6 4 ) a n d e sp e c ia lly R ijk o o r t ( 1 9 6 8 ) , w h e r e a lo n g list
o f th e h y p o th e tic a l e q u a t io n s fo r 9 (S) a nd f ( ^ y is c o n s id e r e d ] . It m u st b e b o r n e in m in d ,
h o w e v e r , th a t all th e e x p r e s s io n s fo r th e s e fu n c tio n s o b ta in e d in su c h a m a n n er m u st be
c o n sid e re d o n ly as a p p r o x im a te fo r m u la s, th e a c cu ra c y o f w h ic h in a g iv en range o f values
o f f (o r o f th e n u m b e r s R f, o r R i) m u st b e e sta b lis h e d fu rth er o n th e basis o f c o m p a r iso n
w ith o b s e r v a tio n . In th e n e x t s e c tio n , w e shall see th a t th e sca tter o f th e e x istin g
o b se rv a tio n a l d a ta is still fairly great; h e n c e c h o o s in g b e tw e e n th e v a rious sem iem p irica l
fo r m u la s is rather d iffic u lt.
In a d d itio n to s e m iem p ir ic a l fo r m u la s for th e w in d p r o file , w h ic h are d im e n sio n a lly
c o rr e c t [i.e ., th e y r e d u c e t o certain a s su m p tio n s c o n c e r n in g th e fu n c tio n s cp(^) or / ( ^ ) ] w e
m a y also fin d in th e m e te o r o lo g ic a l litera tu re a great n u m b e r o f w o r k s (as a r u le , m o r e th an
fifte e n yea rs o ld ) w h ic h p r o p p se sem iem p ir ic a l or p u rely e m p ir ic a l fo r m u la s fo r th e p r o files
o f th e m e te o r o lo g ic a l variables th a t are n o t b a sed o n d im e n sio n a l c o n s id e r a tio n s (very
fr e q u e n tly c o n ta in in g , in s te a d o f em p ir ic a l c o n s ta n ts , so m e em p irica l fu n c tio n s o f th e
R ic h a r d so n n u m b e r or o th e r str a tific a tio n ch a r a cter istic s). T h u s, for e x a m p le , fo r several
y e a rs; th er e w a s a rd en t d is c u s sio n in th e s c ie n t ific litera tu re as to th e p o s sib ility o f using th e
“ g e n e ra lize d lo g a r ith m ic la w ” (w it h p a ra m eters d e p e n d e n t o n th e sta b ility ) t o d e scr ib e th e
w in d p r o file in th e surfa ce la y er o f th e a tm o sp h e r e fo r d iffe r e n t c o n d it io n s o f te m p er a tu re
s tr a tific a tio n . A m o n g th e a u th o r s w h o a d v o c a te d th is la w , w e can m e n t io n , fo r e x a m p le ,
R o ss b y a n d M o n tg o m e r y ( 1 9 3 5 ) , S u tto n ( 1 9 3 6 ; 1 9 3 7 ) , B u d y k o ( 1 9 4 6 ; 1 9 4 8 ) , B jorgum
(1 9 5 3 ) an d m a n y o th er s. A n o b v io u s d e f e c t o f all fo r m u la tio n s o f th e “ g e n e ra lize d
lo g a r ith m ic la w ” is th a t th e y d o n o t ta k e in t o a c c o u n t th e a c tu a l s y s te m a t ic d e v ia tio n s o f
the h e ig h t-d e p e n d e n c e o f th e w in d v e lo c ity fr o m th e lo g a r ith m ic e q u a t io n in c o n d it io n s o f
n o n n e u tr a l s tr a tific a tio n . M o reo v er, e v en in th e lo w e s t la y e r , w h e r e te m p er a tu re str a tific a
tio n p la y s a very sm all part, th e g e n e ra lize d lo g a r ith m ic p r o files d o n o t te n d as a rule to
th o se w h ic h ^ e o b se rv e d in th e a tm o sp h e r e (in particular, e x tr a p o la tio n o f th e w in d pro file
to th e value u ( zq) = 0 h ere le a d s to a v a lu e zq w h ic h d e p e n d s o n th e s ta b ility ; i.e ., is n o t
an o b je c tiv e c h a ra cteristic o f th e u n d e rly in g su rfa ce.
O th er a u th o r s, in c lu d in g S c h m id t ( 1 9 2 5 ) , B e s t ( 1 9 3 5 ) , L a y k h tm a n ( 1 9 4 4 ; 1 9 4 7 a ),
F r o st ( 1 9 4 8 ) , D e a c o n ( 1 9 4 9 ) and T a k e d a ( 1 9 5 1 ) , su g g ested a p p r o x im a tin g th e w in d p r o file
in th e a tm o sp h e r ic su rface la y e r b y p o w e r fu n c tio n s o f z (i.e ., b y a s o -c a lle d “g e n e ra lize d
p o w e r la w ” )* S c h m id t, B est a n d F r o st used a sim p le fo r m u la o f th e u ( z ) w h ile
L a y k h tm a n , D e a c o n a n d T a k e d a d e v e lo p e d a th e o r y in w h ic h
(7 .7 7 )
î — ô
th a t is,
w h ere e is a p a ram eter d e p e n d e n t o n th e th er m a l str a tific a tio n (p o sitiv e for inver

sio n s, z er o fo r n e u tra l str a tific a tio n , and n e g a tiv e in th e u n sta b le c a se ). E q u a tio n ( 7 . 7 7 )
e n a b le s us to d e te c t th e g en era l character o f th e d e v ia tio n s o f th e w in d -v e lo c ity p r o file fr o m
th e lo g a r ith m ic law ; as e - > 0 it tra n sfo rm s in t o th e ordin a ry lo g a r ith m ic p r o file. H o w e v er ,
if, fo llo w in g L a y k h tm a n , we a ssu m e th a t th e para m eter zq a lso depends on th e
s tr a tific a tio n , th e n th e e f f e c t o f th e s tr a tific a tio n w ill still b e e v id e n t as z - > 0 , w h ic h
c o n tr a d ic ts th e d e d u c t io n s fr o m sim ila rity th e o r y . In a d d itio n , th e n u m b e r o f free
p ara m eters available in s u c h a th e o r y fo r m a tc h in g th e o b se rv a tio n a l d a ta in e a ch ca se, is so
large th a t it is d iffic u lt t o fin d th eir v a lu e, a n d th e a c cu ra c y o f th e c a lc u la tio n s is r ed u c ed .
H o w e v er , if, fo llo w in g D e a c o n , w e ta k e zq to b e c o n s ta n t, th e n w e m u st a ssu m e th a t e is
also h e ig h t-d e p e n d e n t [s e e , fo r e x a m p le , D a v id so n a n d Barad ( 1 9 5 6 ] , w h ic h is e n tir e ly
u n sa tisfa c to r y . W e m u st a lso p o in t o u t th a t fo r m u la ( 7 . 7 7 ) c ^ n o t b e c o n s is t e n t w ith th e
d im e n sio n a l c o n s id e r a tio n s o f S e c t. 7 .2 e ith er; in p articular, h ere d e p e n d s e x p lic itly o n
th e fu n c tio n e = e ( R i) [w h e r e R i is ta k e n at a fix e d h e ig h t] and o n th e ro u g h n ess

p aram eter zq (w h ic h in L a y k h t m a n ’s s ch em e also d e p e n d s o n th e str a tific a tio n ).
O n th e b asis o f th e a n a ly sis o f d a ta , it w a s p r o p o s e d , in p articular, b y H a lstead ( 1 9 4 3 )
a n d P a n o fs k y ( 1 9 5 2 ) , to ta k e in to a c c o u n t th e s tr a tific a tio n e f f e c t o n th e w in d p r o file b y
in tr o d u c in g a lin ear c o r r e c tio n t o th e lo g a r ith m ic la w , i.e ., b y using a fo r m u la o f th e fo r m
u(^) = ^ ( l n - f + bA, (7 .7 8 )
w h e r e b i s a. pa ra m eter d e p e n d e n t o n th e s ta b ility (p o sitiv e fo r in v e r sio n s an d n e g a tiv e in th e

u n sta b le c a se ). F o r m u la ( 7 . 7 8 ) can b e r e c o n c ile d w ith d im e n sio n a l c o n s id e r a tio n s ; w e n e e d
B
o n ly p u t ^ = - ^ w h ic h tra n sfo rm s in t o th e M o n in -O b u k h o v fo r m u la ( 7 . 3 3 ) .
A ll th e rem ark s o n th e w in d p r o file m a y b e r ep ea ted fo r th e te m p er a tu re and h u m id ity

p r o file s , p r o v id e d th a t th e str a tific a tio n is u n sta b le , n e u tra l or slig h tly sta b le . H o w e v e r , fo r
very h ig h s ta b ility , as w e h a v e alrea d y o b s e r v e d , a sp e cia l s itu a tio n arises; th er e are g r o u n d s
fo r b e lie v in g th a t u n der th ese c o n d it io n s th e e ^ d y d iffu s iv ity K t = /(a is c o n ^ d e r a b ly less

than K, a n d h e n c e th e p r o files o f te m p er a tu re T a n d a d m ix tu r e c o n c e n tr a tio n 0 in th is case
d iffer in fo r m fr o m th e m ea n v e lo c ity p r o file (c f. F igs. 4 8 a n d 5 0 ) . T h e th e o r e tic a l a n alysis
o f th is p h e n o m e n o n p resen ts very g reat d iffic u ltie s a n d , s o far, has n o t h a d a n y c o n s p ic u o u s
success; h o w e v e r , E llis o n ’s a tt e m p t ( 1 9 5 7 ) t o e stim a te th e d e p e n d e n c e o f a — - j ^ o n th e
R ic h a r d so n n u m b e r R f = —^ is w o r th y o f m e n t io n , sin c e it led to results w h ic h are in
u n e x p e c te d ly g o o d a g r ee m e n t w ith th o s e o f th e later e x p e r im e n ts o f E lliso n an d T urner

( 1 9 6 0 ) [se e th e e n d o f S ect. 8 .2 b e l o w ] . T o ca lc u la te a , E lliso n u sed th e tu r b u le n t energ y
b a la n ce e q u a t io n fo r a s tr a tifie d m e d iu m and r ela ted e q u a t io n s for th e s e c o n d m o m e n ts
w''^ a n d w ' T' . A s w e have a lready seen in S e c t. 6 .1 , th e s e e q u a tio n s co n ta in
u n fo r tu n a t e ly , a large n u m b e r o f n e w u n k n o w n term s. T h e re fo r e , in order to o b ta in his
d e d u c tio n s , E lliso n h a d to m a k e a n u m b e r o f a d d itio n a l rou g h a s su m p tio n s ( o f th e ty p e o f
th e sem iem p ir ic a l h y p o th e s e s ) b y w h ic h certain term s w ere a ssu m ed n e g lig ib ly sm a ll and
certain c o n n e c tio n s w e re p o s tu la te d b e tw e e n th e n o n n e g lig ib ly smaU term s and th e s e c o n d
m o m e n ts un der c o n s id e r a tio n . U sin g th ese a s su m p tio n s, w h ic h im p lic itly c o n ta in e d th e
a s su m p tio n th a t th e h e a t e x c h a n g e is sm all in co m p a r iso n w ith th e m o m e n tu m e x c h a n g e in
the case o f h igh s ta b ility , E lliso n arrived at a rela tio n sh ip w h ic h m a y b e w r itte n as fo llo w s :
_ Kt “o< p(C )[/?¥(C )-q
w here a o an d R are s lo w ly vary in g fu n c tio n s o f w h ic h in th e first a p p r o x im a tio n , m a y

sim p ly b e a ssu m ed t o b e e m p irica l c o n s ta n ts . It is clear th a t h ere a o is id e n tic a l w ith th e
v alue o f K t I K fo r neu tra l str a tific a tio n (i.e ., in th e lo g a r ith m ic la y er) a n d /? is th e critical
value R fcr w h ic h R f can never e x c e e d an d fo r w h ic h K t b e c o m e s z er o . V e r y ro u g h in tu itiv e
e stim a te s m a d e b y E lliso n le d h im to c o n c lu d e th a t th e p a ram eter R m u st b e co n sid e ra b ly
less than u n ity (fo r e x a m p le , c lo s e to 1 /7 o r 1 /1 0 ) ; h o w e v e r , th is c o n c lu s io n still c a n n o t b e
ta k en as fin al. I f w e k n o w th e w a y in w h ic h th e n u m b e r R f a p p ro a c h e s its lim itin g value R -
R fc r , i.e ., th e order o f th e sm all term o ( l ) in th e a s y m p to t ic e x p a n s io n (C) = - ^ [ 1 - | - 0 (1 )]
o f th e fu n c tio n cp (t) fo r large £ ; th e n , p r o c e e d in g fr o m E q . ( 7 . 7 9 ) it is p o s sib le to
d e te r m in e also th e a s y m p to t ic b e h a v io r o f th e fu n c tio n (pi (£ ) [or (C ) = — (C)l as

I Oq
£ H o w e v er , a t p resen t, w e still d o n o t h ave th e n e c essa ry d a ta fo r th is.
7.5 General Similarity Hypothesis for a Turbulent

Regime in a Stratified Boundary Layer and Its
Application to the Characteristics of Turbulent
Fluctuations
So far, we have considered only profiles of the mean velocity,
temperature and humidity in the surface layer of the atmosphere,
and have applied the dimensional considerations discussed in Sect.
7.2 to them only. However, these considerations have a general
character, and the results of Sects. 7 .3 -7 .4 far from exhaust all the
applications of similarity theory to turbulence in a stratified
boundary layer. We shall now consider some further appHcations of
this theory, relating to statistical characteristics other than the mean
values of the basic flow variables.
First, we formulate the fundamental postulates on similarity in

the most general form. In Sect. 7.1 we noted that for plane-parallel
turbulent flow of a stratified fluid above a plane homogeneous rough
surface (which is a natural model of the “atmospheric surface
layer” ), all the one-point moments of the flow variables wUl depend
on the vertical coordinate z only. However, the restriction to
one-point moments was introduced only because we were interested
primarily in them. In fact, in such a model, all the probabihty
distributions for the values of the flow variables at several points will
be invariant under arbitrary translations of this set of points in the
plane Oxy and its reflections in the vertical plane Oxz passing
through the direction of the mean velocity u{z), and will also be
stationary (independent of the time shifts). In other words, in this
model, the probability distribution for the values of arbitrary flow
variables at the points (;ci, t/,, z,, <i), . . . , (x„, y„, z„, t„) can
depend only on the parameters ^2 — jc,, . . . ,Xn — Xt; yz — yu- ■ ■ ,
Hn — yi\ Z\..........Zn\ h — tu • • • , tn — ti and will not change if the
direction of the Oy axis is reversed.
As flow variables we shall consider only the three components of
velocity u, v, w and the temperature T (the further inclusion of the
admixture concentration d leads only to the appearance of obvious
additional formulas, which are of little importance in the discussion.
These variables may be represented as
u = u(z)-\-u\ v = v', w = w', T = T {z)-\-T ', (7.80)
where u{z) and T{z) are ordinary (nonrandom) functions of 2 which

were discussed in detail earlier. Thus it is sufficient to consider only
th e probabihty distributions for the random fluctuations
u'{x, y, z, t); v'{x, y, z, /); w'{x, y, z, t) and l''{x, y, z, i). In this
model of a stratified medium, the general hypothesis of similarity of
the turbulent regime, apphed to the surface layer o f the atmosphere,
may be formulated as follows: the joint probability distribution for
the values o f the dimensionless fluctuations u'lu^, v'/u^, and
r / i r * ! at the points (x,. t/i, Zi, /i), ... , {x„, i/„,z„, t^)can depend
only on the dimensionless parameters
X2 — Xi x„ — xx^ yt — y\ yn — y \ . .
| Z. | > •••• \L] ' \L \ ' \ L\ • L L'
\ L\ ••••• \ L\ '
provided that the following two conditions are satisfied: 1 ) the

heights Zi, . . . , Zn are not too great (they lie within the layer in
which it may be assumed that « ♦ = const, q= const and the effect of
the Coriolis force may be ignored) and not too small (all much
greater than the “roughness height” Zo); and 2 ) the distance between
any two different points {xu yu Zj) and {Xj, y,, Zj) and all nonzero
differences \ ti — \ are n ot too small (so that the interactions
between fluctuations at the space-time points under consideration
show no influence o f molecular effects, determined by the molecular
viscosity and thermal conductivity of the air) and are n ot too great
(so that the conditions of horizontal homogeneity and steadiness do
not break down). Throughout this chapter, L and T ^ denote the
same length and temperature scales as in Eqs. (7.12) and (7.14). The
justification of this hypothesis is, in fact, already included in the
considerations of Sects. 7.1—7.2.
As an example, we shall consider in greater detail the joint
probability distribution for fluctuations (u', v', w', T') at a fixed
space-time point (a:, y, z, i). According to the similarity hypothesis,
this distribution can depend only on the vertical coordinate z, and its
probability density v', w', r')m a y be written as
<ir,| «. «, ir.i Ll
where 'F is a universal function of five variables. Of course, a

function of five variables is a very complicated characteristic; thus it
is essential that in the limiting cases o f great instability
and great stability ^ neutral
stratification formula (7.81) can be simplified. In the case

of neutral stratification q ^ O and the buoyancy has no effect on the
turbulence. Thus, here, the parameter g/To must drop out of Eq.
(7.81). Therefore, in this case, there is no dependence on zjL (L
contains gITo)', i.e., the probability distribution is height-
independent. Moreover, in neutral stratification there are no temper
ature fluctuations, so that the dependence of O on T' is described by
the multiplier 6{T') [where 6 is the Dirac delta-function]. As a
result, we obtain
0 ( a ', w'. -— •— as ^^0 , (7.82)

\u^ I ' L
where 'I^ is a universal function o f three variables, describing the

probability distribution of the velocity fluctuations at a point of the
logarithmic layer. The limiting case o f very great instability (free
convection) is obtained for q>0, 0 , so that here the parameter
«*must disappear from Eq. (7.81). Consequently, it follows that
’' z - y > x
V\ ^0 ^p9o / \ ^0 ^p9o i
w' r
(7.83)
(c/po)
where is another universal function. Finally, it may be assumed

that in the case of very great stability the turbulent fluctuations
acquire a local character, i.e., that their statistical properties cease to
depend on the distance z from the underlying surface. Thus it must
be expected that as ^ ^ + oo, the parameter z will disappear from
Eq. (7.81), i.e.,
i ) as i (7.84)
Unfortunately, Eqs. (7.82)—(7.84) are still all very difficult to

verify, since multidimensional probability distributions are very
difficult to determine reliably from experimental data. Therefore, we
shall further restrict ourselves to the investigation of only the
simplest characteristics of the distribution (7.81), namely, to the
lowest moments o f the fluctuations u \ v', w', T at the fixed point
{x, y, z, t). According to the general formula (7.13), any one-point
moment o f these fluctuations may be represented in the form of
some combination of the parameters and multiplied by a
universal function of ' = [the same result, o f course, follows from

Eq. (7.81)1. From the definition of the fluctuations, their mean
values are zero; thus we can proceed at once to the second moments.
In all there are ten such moments. However, three of these, namely,
u 'v', v'w' and. v ' T art identically zero. This is due to the symmetry
o f the turbulence with respect to the direction o f the mean wind
(this has already been discussed in relation to the moments u'v' and
v'w' in Sect. 7.1). The moments u'w' — — u? and w 'T ' = have
* <^p9o
constant values; thus only five moments remain to be considered—
the variances of u', v', w', T (in place of which it will be convenient
to consider the standard deviations Ou, Ov, a„„ Ot) and the mixed
second moment u'T'. By Eq. (7.13), these quantities may be written
as
(7.85)
Thus, to determine all the second moments o f the wind-velocity and

temperature fluctuations, it is sufficient to determine the parameters
«*and q and to know the five universal functions /s, / 4, /s, /e, and /t.
These may be used, in particular, to express such interesting
turbulence characteristics as the anisotropy coefficients a jo u and
a j o u and the correlation coefficients r„r and fuT
_ f± r
• 11 w
= _____ -
/ s ’ / a ’ / a /s ’
(7.86)
(in the last two equations, the upper sign relates to unstable stratifi
cation and the lower to stable stratification).
The functions • • • , felC) are clearly nonnegative by defini
tion for all values of I. The moment TTT' defined by the function
describes the turbulent heat transfer in the direction of the
mean wind; this transfer usually plays no essential part and often
may be neglected. With neutral stratification and with extremely
great instability u'T' = 0 (see below); however, for other values of
the function fiCO takes positive values, because with unstable
stratification, the inequality T' > 0, as a rule, will hold when w' >
0. Therefore the u' fluctuations will more often than not be negative,
while for stable stratification, the situation is reversed. Moreover,
the ratio -uY'/w'T’’ = x / 7 (?) may even take values greater than 1 , since
u' lluctuations arc usually considerably greater than w ' fluctuations

Isee Sect. 8.5 below and Yaglom (1969)].
Tlie asymptotic behavior of the functions fs, . . . , as ^->-— 0 0 ,
^ ^ + 00 and as Itl-Ô may be established, starting with Eqs.
(7.82)-(7.84), as was done above for the functions/, fi, fz describing
the mean profiles. With increase o f instability (i.e., as 00 ), «*
must disappear from all the equations of (7.85); consequently,
Functions /s, A- and /s will increase asymptotically in the manner of
I C 1'^’, fe tends to zero similar to \ S and will tend to some
constant. Moreover, it must be borne in mind that as 0, u (2 )
0 also, i.e., the turbulence approaches a regime in which in the plane
Oxy, no direction is especially distinguished, and the fluctuations of
u' and v' play identical roles (the idealized regime of “true free
convection” in the absence of mean horizontal velocity). Thus we
must expect that as — °o,
/3(g «/,(g /5(g (7.87)

/cCg^Ceicr'/-. / 7 (C)«o,
or, in other words,
„ ~ r'( ^ sz \'l' „ (I \'l*

nvo ^0 ) ' To)
(7.87')
[formula (7.87') for o«, was actually known to Prandtl (1932a) but
was then forgotten; later, it was obtained independently by Obukhov
(1960), together with the formula for ay]. The universal constants
C4 — C s= and C e = Ce*'* in Eqs. (7.87) can be
determined from observations. From Eqs. (7.87) it also follows that
in conditions o f free convection ~ 1, = const, r^j- = const,
'‘uw— and r „ r « 0 . In the other limiting case as 00 (i.e.,
with unbounded increase of stability), the turbulence characteristics
by Eq. (7.84) cannot depend explicitly on z. Therefore as^ -^ 00 , all
the functions o f Eq. (7.85) [and their combinations (7.86)] must
tend to constants. Finally, as 1^1 ^ 0, the fluctuations of temperature
vanish [see Eqs. (7.82)] and Consequently, fe(0)and ^7 (0 )
can only be interpreted as limits; in fact, the functions /e( 0 a n d / 7 (^)
must be considered as consisting of two separate branches (for ?; > 0

and for ^ < 0 ). For the constants ^3 (0 ) =A^, f^(0) =A^ and / 5 (G) =A^,
these describe the intensity of velocity fluctuations in the logarith
mic layer of a homogeneous fluid and have well-defined values (see
above, Sect 5.3 where these constants are denoted hy AuAiUxvdAz)-
For small values of 1^1, /a, / 4, and U may probably be described
approximately by two terms of the Taylor series
/^(C) = A + P ^ i = Z, 4, 5, (7.88)
where the coefficients Pi are all negative (since turbulent exchange

becomes weaker as stability increases).
Equations similar to Eq. (7.85) [and containing new universal
functions] may be written for higher-order moments of the
fluctuations u' v', w', and T . Thus, for example:
(7.89)
Functions U and /g describe the skewness of the probability

distribution for jo'and T':
(7.90)
oi /I’
while functions fio and /u describe the vertical diffusion of turbulent

energy and o f the intensity of temperature fluctuations. On the basis
o f the concept of thermal convection, as a set of ascending motions
of warm air in the form of comparatively intensive narrow jets and
descending motions of cool air in the form of the slow sinking of
considerable masses, we may expect that fg, • • • , fn will be positive,
at least for negative According to Eqs. (7.82)—(7.84), as the
instability increases (as — 0 0 ) the functions /g and fio must
increase asymptotically similar to 1^1, /g must decrease asymptotically
similar to !Sh' (so that the skewness factors / g / / | and must tend
to constants), and fn(S) decreases similar to jC|~'^’. With increase of
the stability (as S ^ 00 ) all functions /g, . . . , fa and also the ratios
/g/Zj and fg jfl must tend to constants.
T URBULE NCE IN A T H E R M A L L Y S T R A T IF IE D M E D IU M 463
The statistical characteristics of the velocity and temperature

derivatives at a fixed point {x, y, z, t), generally speaking, will now
also depend on the molecular viscosity and thermal diffusivity v and
X (recall that in the formulation of the similarity hypothesis it was
required that the differences \ti — and the distances between
different points (xu yu Zi) and (Xj, yj, Zj) should not be too small).
Thus applying dimensional reasoning to such characteristics leads to
more complicated formulas which now contain universal functions of
several variables. However, there are two im portant exceptions to
this rule, relating to the quantities
•= rs (^ + ^ r
and
In fact, these q u a n titi^ occur in the turbulent energy balance

equation (6.43), where e is denoted by the symbol ?(, and in the
related equation (6.55) for the balance of temperature fluctuation
intensity, and with the aid of these equations can be expressed in
terms of the ordinary one-po^nt moments o f the velocity and
temperature fluctuations. Thus e and N should not depend explicitly
on the molecular constants v and %. This is no fo rtu i^ u s fact; as we
shall show in Chapt. 8 in Volume 2 of this book, e and N play a
considerable role in the laws o f relatively large-scale motion and can
be determined from the probability distributions for the fluctuations
in velocity and temperature at two points sufficiently far apart.
Thus, to describe the dependence of e and N on the height z, we
may use the ordinary similarity formula (7.13) which does not
contain v and x- That is, we may put
where (p, and q>ivare new universal functions. As |^| - ► 0, cp, and (pjvwill
tend to finite limits (p, ( 0 ) = 1 (since here e = «2 - ^ , see above. Sect.
6.5), and cpjv(0)= const = 1/oq ; moreover (Pe (C) ~|C|and(pN(S) ~ |

a s ^ - ^ —oo. The formulas for the movements o f fluctuations at several
points are considerably more complicated; these will contain, in all
cases, universal functions of several variables. Thus, for example, the
two-point moments o f the simultaneous turbulent fluctuations at the
points (xi, «/i, zi) and (xz, y^, z^) will depend, generally speaking, on
the four variables and • Practically speaking,
the experimental determination o f functions of four variables is
hopeless at present; thus conditions in which the number of variables
in the formulas for two-point moments can be reduced are of great
interest. In particular, one such condition is that — 5 > Si where is a
positive number defining the lower bound of the range of values of |C|
for which the turbulent regime will be a regime o f free convection
(we shall discuss the order o f magnitude of in the following
section). Under this condition, we may take «* = 0, so that no finite
length scale can be formulated from the parameters o f the problem.
Moreover, in this case there is no favored direction in the Oxy plane,
thus all scalar (i.e., independent o f the orientation of the coordinate
axes) two-point moments can depend only on ro= [{x^ — x^Y +
(i/2 — y \ W ‘, but not on X2 — x\ and «/2 — y\ separately. From 2 i, z^
and To, two dimensionless combinations may be formulated, for
example, z j z i and r o /Y Z1Z2 (or Z2/Z1 and r/V ZiZz, where r =
[rl+{z 2 — Z\Y]i^). Thus in conditions of free convection, all scalar
two-point moments will be equal to some combinations of the
parameters gITo, ^/Cppo and z = V Z\Z2 , multiplied by a universal
function of z 2lz\ and ro/z (or of zj/zi and r/ 2 ). In particular.
t ) w ' { X 2 , V2, Z 2 , t) =
7 k )'
U'(Xi, y i, Zj, 0 « '(-< 2 > y 2 . ^ 2. 0 = 'V'(Xi, yi, Z ,, i ) ‘V'(X2, y2. ^2, 0 =
■ F fe )’ 0.94)
T 'ix i, yi, Zi, t) T { x 2 , y 2, Zj. 0 =
where R\, Rz, and Rz are universal functions of two variables [these
formulas were presented by Obukhov (I9 6 0 )].
The experimental determination of functions of two variables is

also fairly complicated, but not hopeless. However, it is noteworthy
that in certain cases the form of the dependence on one of the
variables has been predicted successfully in a theoretical manner,
while for a number o f statistical characteristics it has proved possible,
in general, to reduce all the indeterminacy existing in the theoretical
formulas to an indeterminacy in the choice of a numerical
coefficient. For this it is only necessary to use some additional
similarity considerations. These considerations are related to a
completely different class of turbulent flows, which includes
atmospheric turbulence as a special case. Chapter 8 in Volume 2 will
be devoted mainly to this type o f similarity; therefore, we shall
postpone any further analysis of formulas of the type o f Eq. (7.94)
to Volume 2 of this book.
A s in th e case o f f, f\, f i , a n u m b e r o f a d d itio n a l n o n r ig o r o u s r esu lts c o n c e r n in g th e

fu n c tio n s /a, . . /, y, cp, a n d (Pn m ay be o b ta in e d using d iffe r e n t variants o f th e
sem iem p ir ic a l th e o r y o f tu r b u le n c e . T h u s, fo r e x a m p le , K a z a n sk iy and M o n in ( 1 9 5 7 ) [se e
also M o n in (1 9 5 9 a )], on th e basis o f th e s im p lifie d en e rg y b a la n c e e q u a tio n ( 6 . 4 6 )
s u p p le m e n te d b y s o m e fu rth er s e m iem p ir ic a l h y p o th e s e s , o b ta in e d fo r / 5 (^) an a p p r o x i
m a te e q u a tio n o f th e fo r m
1 V,
I- (7 .9 5 )
/ ' (0 J
w h e r e j 4 s = f 5 ( 0 ) is a n e m p ir ic a l c o n s ta n t. T h e sa m e c o n s id e r a tio n s , a p p lie d t o th e b a la n c e
e q u a tio n in th e fo r m ( 6 . 4 6 ') , lea d s t o th e r e la tio n sh ip
'U
( 7 . 9 5 ')
/'(C)
w h ic h c o n ta in s a fu rth er e m p irica l c o n s ta n t a . L a te r, P a n o fs k y a n d M cC o r m ic k ( 1 9 6 0 )
a d v a n ce d a h y p o th e s is w h ic h le d th e m to c o n c lu d e th a t
/5(C) = ^ [V '( 0 - (7 .9 6 )
w h e r e f l is a n e w e m p ir ic a l c o n s ta n t, w h ile B usin g er ( 1 9 5 9 ) a n d M u n n ( 1 9 6 1 ) , u sing t w o

d iffe r e n t a s su m p tio n s o n th e m ix in g le n g th in a str a tifie d m e d iu m , o b ta in e d th e fo r m u la s
V.
(7 .9 7 )
A(C) V (0
[w h e r e a is th e sa m e c o e f f ic ie n t as in E q. ( 7 . 7 4 ) ] a n d
A(C)^------ A
/"(C) (7 .9 8 )
‘ [/'(C)]»’
r e s p e c tiv e ly . E q u a tio n ( 7 . 9 8 ) w ith A ^ = 1 (w h ic h c o n tr a d ic ts th e d a ta ) w a s a lso u sed b y

Y okoyam a (1962a). For M l), P r ie stle y (1 9 6 0 b ) and P a n o fs k y (1 9 6 1 b ) u se d th e
ie m iem p irica l r ela tio n sh ip
/ 6 W = >»6“oVU'^)- ^6 = / 6 (0)- (’ -99)

M o n in ( 1 9 6 5 ) c o n sid e re d th e c o m p le te s y ste m o f d y n a m ic e q u a tio n s fo r th e o n e -p o in t,
sec o n d -o r d e r m o m e n ts o f v e lo c ity a n d te m p er a tu re flu c tu a tio n s , and n e g le c te d th e te r m s in
th is sy ste m w h ic h d e scr ib e th e vertica l transfer o f th e s e q u a n titie s (i.e ., in p articular, all
te r m s c o n ta in in g th ird -order m o m e n ts ) , w h ile h e e lim in a te d th e pressure flu c t u a tio n s w ith
th e aid o f s e m iem p ir ic a l h y p o th e s e s o f th e t y p e ( 6 . 1 2 ) . H e th u s o b ta in e d th e a p p r o x im a te
r e la tio n sh ip
[ /^ ( 0 - /l( 0 ] /i( t ) « 2 , (7 .1 0 0 )
c o n n e c tin g th e va lu es o f th e fu n c tio n s /a, fiy and fs (fo r ^ = 0 , it is su p p o r te d b y th e da ta

g iven in S e c t. 5 .3 ) . F u rth er, usin g certa in a d d itio n a l s e m iem p ir ic a l h y p o th e s e s , M o n in w a s
a b le to e x p r ess th e a n is o tr o p y c o e f fic ie n ts an d f s i Q / f A i Z ) in te r m s o f cp(^) =
I f ' i ' Q a n d th e c o n s ta n ts A ^ = faCO), M O ) and A ^ = fs(0) w h ic h are k n o w n fr o m
e x p e r im e n t (w h ile it w a s s h o w n th a t o n th e s e a s su m p tio n s, th e in e q u a lity C 2 = 1//? 5 .5
m u st b e v a lid ). A fu rther s e m iem p ir ic a l r ela tio n sh ip w a s p r o p o s e d b y P a n d o lfo ( 1 9 3 6 ) ,
c o n n e c tin g th e fu n c tio n s cp(£), / ( J ) , a n d fsC S ). B y u sin g s o m e fa irly r o u g h h y p o th e s e s ,
P a n ch ev ( 1 9 6 1 ) o b ta in e d th e fo llo w in g c o n n e c tio n s b e tw e e n th e c o e f fic ie n ts C j , C g , and
C g o f Eqs. ( 7 . 3 6 ) and ( 7 .8 7 ') :
(fro m w h ic h , in particular, it f o llo w s , th a t in free c o n v e c t io n r ^ r — 1, w h ic h s e e m s q u ite

d o u b tfu l).
T h e c o n n e c t io n b e tw e e n th e un iversal f u n c tio n s cps ( ^ ) , f ( ^ ) a n d îo(S) m a y b e o b ta in e d
fro m th e e n e rg y b a la n c e e q u a tio n ( 6 .4 3 ) ; a c co r d in g to th is e q u a t io n , it s ee m s p r o b a b le th a t
? e (0 = C /'(C )-!:-Y V ;o (';) (7 .1 0 2 )
[see M o n in ( 1 9 5 8 ) ] . E q u a tio n ( 7 . 1 0 2 ) is e x a c t to th e deg ree in w h ic h it is p e r m issib le

to ig n o r e th e v ertical transfer o f tu r b u le n t e n e rg y d u e to th e pressure flu c tu a tio n s and
v isco u s fr ic tio n (it b e c o m e s q u ite e x a c t i f w e in c lu d e th e su m m a n d I p ' w ' l p ^ u l in th e
d e fin itio n o f th e f u n c t i o n / i q ) . I f w e c o m p le te ly ig n o r e th e v ertica l d if fu s io n o f e n e rg y [i.e .,
w e a d o p t E q. ( 6 . 4 6 ) | , th e n w e o b ta in
? ,(C )^ C /'(C )-C , (7 .1 0 3 )
g q
w h ile i f w e c o n s id e r th is d if fu s io n t o b e p r o p o r tio n a l t o -f. ------;;— (i.e ., i f w e a d o p t E q.
^0
( 6 .4 6 ') , th e n
cp^(C) = C /'(C )-a C . ( 7 . 1 0 3 ')
O f c o u r se , b o t h E qs. ( 7 . 1 0 3 ) an d ( 7 . 1 0 3 ' ) are, in g e n e ra l, less e x a c t th a n E q. ( 7 . 1 0 2 ) [b u t

th e y are e q u iv a le n t to E q. ( 7 . 1 0 2 ) in th e lim itin g case o f n eu tra l s tr a tific a tio n a n d E q.
( 7 .1 0 3 ' ) is e q u iv a len t to E qs. ( 7 . 1 0 2 ) a n d ( 6 . 4 3 ) in th e ca se o f free c o n v e c t io n , sin c e th e n
/ 1 0 (f) R e m e m b e r in g th e a s y m p to t ic fo r m u la s ( 7 . 3 3 ) an d ( 7 . 3 9 ' ) fo r th e fu n c t io n f ( C )
and ta k in g as starting p o in t, for e x a m p le , E q. ( 7 .1 0 3 ' ) w e fin d
9, (C) = l 1^1 ■ (7.104)

h + (P-«)C for Kl<Ci-
I?
T h u s, a c c o r d in g t o E q. ( 7 . 1 0 3 ' ) in nearly n e u tra l str a tific a tio n , e « ; and in th e case o f
in te n se in s ta b ility e , i.e ., it c ea ses to d e p e n d o n th e h e ig h t (th e sam e

i 0
resu lts, b u t w ith u n d e te r m in e d n u m erica l c o e f f ic ie n t s in b o t h ca ses are o b ta in e d b y
d im e n sio n a l c o n s id e r a tio n s o n ly ) .
Sim ilar rela tio n sh ip s m a y a lso b e d e d u c e d fo r cp^r (S ) i f w e u se as o u r starting p o in t th e
b a la n ce e q u a tio n fo r in t e n s it y o f te m p er a tu re flu c t u a tio n s ( 6 .5 5 ) . U sin g th is e q u a tio n w e
e a sily o b ta in th e rela tio n sh ip :
y w (0 = f:/;(C )-^ i:/;i(0 . (7 1 0 5 )
M oreover, if w e also ig n o re th e vertical d if fu s io n of th e in te n s ity o f te m p er a tu re

f lu c t u a tio n s [first term o n th e right side o f E q. ( 6 . 5 5 ) ] , th e n w e o b ta in
= (7-106)
F u n c tio n ( 7 . 1 0 6 ) p o sse sse s th e p ro p e r tie s
---- 1I I aa V
( 7 .1 0 7 )
-
C o n s e q u e n tly , in th is a p p r o x im a tio n fo r near-neutral s tr a tific a tio n , V ------------w h ile for
_ ^0“
^
great in s ta b ility , N d e c re a ses w ith h e ig h t a c co r d in g t o th e la w
T h e resu lt fo r near-neutral str a tific a tio n d o e s n o t d e p e n d o n ig n o rin g th e d if fu s io n term ,

w h ile th e e f f e c t o f d if fu s io n m a y lead to a ch a n g e o f th e n u m erica l c o e f fic ie n t in Eq.
( 7 . 1 0 8 ) . T h e b e h a v io r o f q).v ( t ) w h e n £ - > + 0 0 c a n n o t b e sta ted w it h o u t k n o w in g h o w th e
fu n c tio n a ( ^ ) = / ' ( S ) / / i (£ ) b e h a v e s fo r large p o s itiv e £.
O f c o u r se , w e m a y r ep ea t fo r fo r m u la s ( 7 . 9 5 ) - ( 7 . 1 0 1 ) , ( 7 . 1 0 3 ) , ( 7 . 1 0 3 ') , a n d ( 7 . 1 0 8 ) all
th a t has b e e n said p r e v io u sly a b o u t th e a n a lo g o u s sem iem p ir ic a l fo rm u la s rela tin g to /,
an d f 2 \ t h e y are all o n ly a p p r o x im a tio n s , th e a ccu ra c y o f w h ic h and th eir r eg ion s o f
a p p lic a b ility m u st b e fo u n d b y c o m p a r in g th e m w ith g o o d -q u a lity e x p e r im e n ta l resu lts. A t
p r e se n t, th e da ta o n flu c tu a tio n s are still very in c o m p le te ; n e v e r th e le ss, th e sem iem p ir ic a l
fo r m u la s c a n b e o f c o n sid e r a b le v a lu e, sin c e in m a n y ca ses th e y can h e lp us to e stim a te w ith
fair a c cu ra c y th e order o f m a g n itu d e o f th e c o rr e sp o n d in g s ta tistic a l c h a ra cteristics, and
serve as a gen era l g u id e in w o r k in g u p th e r esu lts o f th e e x istin g (c o m p a r a tiv e ly r o u g h )
o b se rv a tio n s.
8 . COMPARISON OF THE THEORETICAL DEDUCTIONS

WITH THE DATA
8.1 Wind-Velocity Profiles in the Surface Layer
of the Atmosphere
Here we compare data from meteorological measurements in the
atmospheric surface layer with the theoretical deductions o f Sect. 7.
We take first the simplest measurements of the mean wind velocity at
different heights z, repeated many times at different points on the
surface of the earth with the aid of various types of anemometers
[many o f which are described, for example, in the books of
Middleton and Spilhaus (1953); Kedrolivanskiy and Stemzat (1953)
and Lettau and Davidson (1957)]. It was already noted in Sect. 5.4
that when the underlying surface is comparatively flat and the
temperature stratification is close to neutral, the data on the
wind-velocity profile in the lower atmosphere are described fairly
well by the logarithmic formula (5.31).
However, in the diabatic case (i.e., with
considerable temperature gradients),
the dependence of the wind velocity
u{z) on In 2 [data on which may be
found, for example, in Thomthwaite
a n d K aser (1943), Shcherbakova
( 1 9 4 9 ) , Deacon (1949), Pasquill
(1949), Monin (1953) and in many
dozens of more recent works] exhibits
regular deviations from the simple lin
ear relationship. For unstable stratifica
tion, the increase of u{z) with increase
o f In 2 is always slower than linear,
while for stable stratification it is faster
than linear. As a typical example, Fig.
0.01 0.57 24 875 52 (borrowed from Monin (1953)],
^ (m eter s) gives mean wind-velocity profiles ob
F IG . 5 2 . E m pirical p r o files o f th e
tained in 1951, from meteorological
m e a n w in d v e lo c ity in th e surface measurements in the Kazakhstan steppe
la yer fo r d iffe r e n t th er m a l str atifi at heights of 0.5, 1, 2, 4, 8, and 15 m.
c a tio n s .
The six curves in Fig. 52 were obtained
by averaging 61 individual wind-
velocity profiles in six groups, each characterized by nearly constant
values o f the stability parameter
Tq [5(1 )]2
where To is the standard temperature, and the figures in parenthesis

indicate the height (in meters) of the observation point.
The six stability groups indicated by the numbers +2, + 1 ,. . . , —3
in Fig. 52, correspond to stabihty classes ranging from moderate
stability (group +2) through slight stability (+1), near-neutral
stratification (0), slight instability ( —1) and moderate instability
( - 2 ) to intense instability ( —3).
We see that only in the neutral stratification does the logarithmic
formula give a good representation o f the whole wind-velocity profile
(i.e., it is represented by a straight line in Fig. 52). In all other cases,
at sufficiently great heights, deviations from the logarithmic law are
observed and are in qualitative agreement with the theoretical
predictions of the preceding section. However, at the same time, at
low heights the wind-velocity profiles for all groups will be given
approximately by a formula of the type u{z) — A \n — with the same
'2'o
value of the “roughness height” zo (which is, of course, estabUshed
most reliably using profiles with near-neutral stratification); in the
conditions o f Fig. 52, this height is close to 1 cm.
According to the similarity theory of Sect. 7, the height variation
of the wind velocity in all cases must be determined by the universal
function /(^) o f the dimensionless height where Z ,=
— verify this result and to establish the exact form of
the function /(^), in addition to the values ofM( 2 ), we require the
values o f « * = and q, which permit us to evaluate the length
scale L. However, direct measurement of the turbulent momentum
and heat fluxes t and q in the surface layer of the atmosphere (which
is the subject of a special discussion in Sect. 8.3) began to be carried
out only comparatively recently, and, so far, are fairly infrequent
and rather inaccurate. Thus the first empirical verification o f Eq.
(7.24) containing the function/(^), which permitted a typical graph
of this function to be drawn for the first time, was carried out by
Monin and Obukhov (1953; 1954) by a quite different method. As a
basis for this verification, they used extensive data (collected during
four expeditions in 1945, 1947, 1950 and 1951 in different regions
of the U.S.S.R.) relating to wind and temperature profiles under
different meteorological conditions. Further, it was assumed that

formulas (7.24) hold and also some additional assumptions were
adopted as to the form of the functions f(t,) and The
satisfactory agreement between the results of data processing on the
basis o f these assumptions, confirmed their admissibility as a
reasonable first approximation, and showed that the general form of
the function/(^) must be consistent with its empirical values found
in this manner.
The additional assumptions on the wind and temperature profiles
made by Monin and Obukhov consisted of assuming the validity of
the “logarithmic + linear” approximation (7.33) of f(S)in describing
the wind profile in the lowest four meters, and of assuming the
similarity of the wind and temperature profiles within this layer. On
this basis, all these wind profiles, within the lowest four meters, were
approximated by a formula o f the form
— ■îiôgz-\-D)-^CiZ, (8.2)
where i is the number of the profile, and 4,, C,, and D are empirical
coefficients (determined by the least squares method). Since all
measurements refer to heights z;^>Zo, the coefficient D is assumed to
be the same for all profiles of the same observation point and equal
to — log Zo, where Zq is the roughness parameter. The values of the
coefficients At and C, then allow us to find for every profile the
values o f - ^ a n d ^ according to the formulas
P _ Cf InlO (g 3)
* ~ InlO ' L — Ai
Further, for each profile, the empirical value of the stability

parameter B-^ = ^ is evaluated which, by Eq. (7.33) and
To
the assumption of the similarity of the u {z) and T (z) profiles, is taken
to be equal to the quantity
g ,.T . ‘" “ + I T _ I
Knowing that û= 10 ", we may also define p as the regression

coefficient of the set of values j [where ~ is taken from Eq.
(8.3)] on the empirical values o f Si. It is then found that p «= 0.6,
while the correlation between the values of fl'^-^jand 5 i was fairly
high. The values o f — and L = together with all the data onu (2 )
were used to formulate the empirical function
where the plus sign denotes stable stratification and the minus sign,
unstable stratification [the value of u |^^-^jis obtained by interpola
tion of the measured values of u (2 )]. The obtained results shown in
Fig. 53 allow us to formulate the empirical function f(C) for the
range - 3 < ^ < 6. In spite of lack o f precision in measuring the wind
velocity and the comparatively rough approximate method used to
determine the values of and L, the empirical points lie on a
smooth curve (consisting of two branches) with small scatter (except,
perhaps, for cases of large positive for which the scatter becomes
fairly considerable). The graph thus obtained confirms the existence
of the universal function which determines the dependence of
the wind velocity on height. Also, the empirical function/(^), taken
as a whole, is in agreement with the schematic graph in Fig. 48, i.e.,
it is found to have a number of asymptotic properties predicted in
the previous section: in near-neutral stratification (small |^|) it is close
to In 1^1 + const, for great instability ( f ^ -1) it tends asymptotically
to a constant, while for great stability ( f > l ) , it increases ap
proximately similar to a linear function (in Fig. 53, these three
asymptotes are represented by dotted lines). As later investigations
showed, the value of the function/(^) found by Monin and Obukhov
in 1953-1954 proved to be fairly accurate for ^ < 0 , but for ^ > 0
they were substantially too low (in fact, the graph of f(t) for Z
increasing apparently rises considerably more steeply than the graph
shown in Fig. 53).
Graphs similar to that in Fig. 53 have been plotted frequently by
different authors on the basis of other experimental evidence. Thus,
l[ a ,„ - u ( f) l
M on in a n d O b u k h o v ( 1 9 5 3 ; 1 9 5 4 ) . T h e vario u s s y m b o ls o n th e graph d e
n o t e data o b t a in e d in d iffe r e n t e x p e d itio n s .
Perepelkina (1959b) used the data on the wind and temperature

profiles for 2 m, by Pasquill (1949) and Rider (1954), for an
empirical determination o f the values of /(C) — ± for [C| < - | .
Later, she also worked out the data on the wind and temperature
profiles up to a height of 70 m, carried out in the summer and
autumn o f 1960—1961, from a 70-m meteorological mast in the
south Russian steppe close to the town of Tsimlyansk. Still later,
Shiotani (1962) published a graph of the function /(C) —/ ( ± for
1 3
^ < |C| similar to Fig. 53, and also a separate detailed graph of
/(C) — ± for 1^1 < 0.3, obtained using a method similar to that
described above from the data of observations on wind and
temperature profiles carried out in 1960—1962 on a 45-meter
television tower near Tokyo. As soon as data o f the direct
measurement of and q appeared, attempts were made to plot a
graph similar to Fig. 53, in which values o f L would be determined
according to the results o f such measurements (made simultaneously
with profile measurements), without recourse to any special hypoth
eses. Such graphs were plotted, for example, by Gurvich (1965) and
by Zilitinkevich and Chalikov (1968a) with reference to the data
obtained from 1962 to 1965 in the region of Tsimlyansk (see below.
Figs. 54 and 55).
u (z)-u (/L //2 )
to
^■^ccm s ty
-7.0
~Z.O A A
r / - ^ ^-hcon.s i
-3.0 4
-4.0
aooi 0.002 0M0 5 aorao2 0.05 o j 0.2 a s to 2.o_^s.o
FIG. 54. E m p irical graph o f th e f u n c tio n f (^) ~ f ~ for

C < 0 a c c o r d in g t o th e data o f G urvich ( 1 9 6 5 ) .
The graphs of the function / ( C) — given in Fig. 53 are

intuitively clear and give a good description of the general pattern of
the dependence of the wind velocity on height for unstable and
stable stratification. However, for the quantitative estimation o f the
_ u(z) -u(L/?0)
7.0 const
6.0 /
5.0 i
4.0
3.0
2.0 / .
10
A
1 1
0 Q 'oodm zm QOY O.f0.20.3 0.7 2,03,0
-to
-2.0
-3.0
-AO ^ const
FIG. 55. E m p irica l graph o f th e f u n c tio n / ( C ) — / ^

fo r C > 0 a c c o r d in g t o th e data o f G urvich ( 1 9 6 5 ) .
deviations of the function /(g) from the logarithmic function for

small ISI and its deviations from the constant value for large negative
these graphs are not very suitable, since corresponding parts of the
empirical curves on such graphs are compressed greatly. As a result,
several researchers have either used modifications of the graph o f Fig.
53, or modifications o f the empirical function /(g) itself, which are
more convenient in certain respects. Thus, for example, Priestley
(1959a) modified the section o f Fig. 53 corresponding to negative g
(unstable stratification) by replacing the linear scale of the dimen-
sionless heights by a logarithmic scale. The deviations of/(g)
from the logarithmic function for small |g| then became much more
marked, and, in particular, it became evident that even for g —0.05,
these deviations are fairly considerable (although in Fig. 53 it seems
that the continuous and dotted lines do not differ from each other
for Igl < 0.5). Similar graphs [but with replaced
which in many cases is determined more easily from the measured
values of u (z)] were later reported by R. J. Taylor (1960a), who used
the measurements o f Rider (1954) and Swinbank (1955), and by
Takeuchi (1961), who obtained extensive data from observations
made on a prairie near O ’Neill, Nebraska (U.S.A.), in 1953 and 1956,
and collected in the monographs of Lettau and Davidson (1957) and

of Barad (1958). R. J. Taylor determined the values of and L
according to direct measurements of t = a^d <? made by Rider
and Swinbank, while for this purpose Takeuchi used an approxima
tion o f the wind and temperature profiles by functions o f the form
(8.2) and formulas (7.33) and (7.47) [with «„ = 1, pi = p]. At the
same time, Takeuchi compared the empirical values of L obtained by
this method with the values obtained by the Monin-Obukhov method
(1954) which is similar in principle, but different in certain details,
and uses measured values of u* and q. According to his results, all
three methods of determining L agree satisfactorily with each other.
Later, Gurvich (1965) plotted the same graphs using the data of
simultaneous measurements of the wind profile, turbulent heat flux
and shear stress made in the summer o f 1962 and 1963 in the steppe
close to the town of Tsimlyansk (see Figs. 54 and 55, the points of
which correspond to data averaged over a series o f measurements).
Additional results of treatment of the wind profile data based on the
similarity theory o f Sect. 7 may be found in scores of works
including those by Panofsky, Blackadar and McVehil (1960),
Panofsky (1963; 1965), Lumley and Panofsky (1964), Panofsky,
Busch, Prasad et al. (1967), Webb (1960; 1965), J. J. O ’Brien (1965),
Swinbank (1964; 1966; 1968), Swinbank and Dyer (1968), Busch
(1965), Busch, Frizzola and Singer (1968), Bernstein (1966),
Chamock (1967a,b), Paulson (1967), Businger, Miyake, Dyer and
Bradley (1967), Zilitinkevich and Chalikov (1968a), Rijkoort (1968),
Fichtl (1968), and others. Similar results have also been obtained re
cently by investigators who have made laboratory measurements in
special wind-tunnels with a portion of the floor that can be heated or
cooled to produce thermal stratification [cf., for example, Malhotra
and Cermak (1963); Cermak, Sandborn et al. (1966); Cermak and
Chuang (1967) and Chuang and Cermak (1967)]. Although the
scatter o f the experimental points is considerable in all the
measurements, the values of f(Q found by different investigators are,
on the whole, in satisfactory agreement with each other. (Some of the
discrepancies discovered will be pointed out in the subsequent
discussion o f the results.)
In several cases, when the experimental results are obtained, the
basis adopted is not the wind profile u (z) itself, but the values of the
wind-velocity gradient According to Eq. (7.15'), to obtain the
universal relationship, we must consider the dimensionless quantity
^ 7 § = 'p ( t ) ’ <p (C) = c/ ' ( ; ) , (8.5)
or, which is the same, the quantity
^ dz
However, since the evaluation of requires the values o f the

turbulent fluxes ■^ = pul and ?,(on the latter of which there exist, so
far, relatively few reliable data), the values of K have often been
determined according to the Richardson number
d f // du V 1
« ( 0 / ' (0
which is a single-valued function of f. Graphs of the function /((Ri)

were reported by Rider (1954), Deacon (1955), Priestley (1959a),
Ellison and Turner (1960), Gurvich (1962), Deacon and Webb
(1962), and Webb (1965). Figure 56 shows the graphs of Gurvich
(1962) relating to the range -0 .0 0 5 > Ri > - 4 , and also taking into
account Deacon’s data (1955). The broken line in the graph gives the
relationship
corresponding to the asymptotic “ 1/3-power law” (7.39) for

conditions o f free convection, with coefficient a = 0.97, found by
the least squares method, applied to the aggregate of data with R
< —0.05. The continuous line represents the asymptotic value
^ ( 0 ) = X 0.4 of the function K(Ri) in the logarithmic sublayer,
which is determined very accurately from the graph of ^(R i) plotted
using a linear scale o f Ri numbers (in exactly this way, an extremely
reliable demonstration was given that in the atmosphere also x «
0.4). There are no data for Ri > 0 (stable stratification) in Fig. 56,
but, in general, observations with large positive Ri are very rare in

meteorology. Therefore, the most extensive data on the values of
K(Ri) for Ri > 0 were obtained by Ellison and Turner (1960) in
laboratory experiments in which the role of the temperature gradient
was played by a density gradient produced by the salinity gradient of
water (see below, Sect. 8.2). The data of Ellison and Turner are quite
scattered, but overall they demonstrate convincingly that on the
positive semiaxis /((Ri) decreases with increase of Ri, as would be
expected.
I''
0.005m 0.02 QOS0.7020.3as W 2.0 so 70

FIG. 5 6 . Empirical graph of the function K (Ri)
according to Gurvich (1962).
Let us consider now in more detail the results for unstable

stratification. The dependence of 9 = ^ on the parameter {J .-=
~ [where Z.' = « , =:a(C)Z, so that f' = Ri 0(f')] which is

also connected single-valuedly with was studied empirically by
Panofsky, Blackadar and McVehil (1960) among others. According
to their results, for all negative I' a good approximation to 9 ( ^ 0 is
given by the solution of an equation of the form (7.61) with the
coefficient a' = aa_„ = 18 [i.e., by the relationship 9 (f) = (1 - a'Ri)"^
with o' - 18; cf. Sect. 7.4, Eq. (7.59')]. Later, the same equation was
tested on extensive data from quite different locations in several
other works, including those by Kondo (1962), Lumley and
Panofsky (1964), Panofsky (1965), Swinbank (1966), Panofsky,
Busch et al. (1967), Bernstein (1966), Chamock (1967a), Paulson
(1967), Fichtl (1968) and Rijkoort (1968). The results show some
small discrepancies between them, but it is generally concluded that
an equation of the form 9 (f) = (1 - a'Ri)"^describes the existing data
with sufficient accuracy for all negative values of Ri if a' is chosen
between 10 and 20 (for example, Paulson considers a' = lia s the best
guess, while Kondo recommends the value o' = 13 and according to
the works o f Panofsky and his group, a '< » l 8 for all geographical
locations). Nevertheless, some investigators prefer the original Ellison
equation (7.59) [or, what is the same thing, Eq. (7.61)], which is
equivalent to Eq. (7.59') only if a(f)is considered constant; however,
in such an approximation of 9 (f) the coefficient a is usually found to
be considerably smaller than o' (with the exception o f the works by
Panofsky et al. (1960), and Bernstein (1966) where the value a = 14
was recoinmended, and by Zilitinkevich and Chalikov (1968a), where
it was found that 0 = 12 for a restricted range of Ri values). Thus
Panofsky (1965) and Klug (1967) recommend the value a = 7 ,
Pasquill [see Swinbank (1966)] finds that a value in the range from 4
to 6 is appropriate and Chamock (1967a) uses the value a = 4 in his
reexamination o f Bernstein’s results (in all these works the data of
Swinbank’s observations in Australia were used). The relative
smallness of a agrees well with the finding that a_„ is apparently
considerably greater than unity (see below, Sect. 8.2). Let us also
note that both Eqs. (7.59) and (7.59') are in agreement with the
“ 1/3-power law” (7.39) and a = (3 /C2)^, a ' = aa_„ = (a/K)"* where a'is
the Gurvich constant of Eq. (8.7). Therefore the data of Fig. 56
correspond to a value of a o f the order of 30 (while a ' =18
corresponds to a 0.82).
Webb (1960) approximated the empirical values of (p(f') on the
semiaxis f ' < 0 by an equation of the form (7.57) which leads to
values o f <p(f') close to those obtained from the equation 9 (Ri) =
(1 - 18Ri)"^" for all not too great values of - f ' [see Panofsky (1963)
or Lumley and Panofsky (1964)], but in the limit as
corresponds to a “ 1 / 3 -power law” of the form 9 (?')«“ - (30f')”‘^^
(i.e., to the equation 9 =» ( -30Ri)”‘/^). Later, Webb (1965) proposed
another form of the universal wind-profile equation which disagrees
with the “ 1/3-power law” in the free convection limit. Some
transformations and an approximation of the most widely used
equation 9 (Ri) = (1 - a'Ri)"*/^ may be found in Fichtl (1968). In
Zilitinkevich and Laykhtman (1965) it was stated that their
semiempirical equation for 9 (f) gives a satisfactory description of the
data (however, the data used in this work were rather early).
Swinbank (1964; 1966) found that his exponential equation (7.72)
[which contradicts the “ 1/3-power law” ] approximates the existing
data quite well, but some other authors do not share this opinion
[see, for example, Barad (1963); Panofsky (1965); Swinbank (1966);
TURBULENCE IN A THERMALLY STRATIFIED MEDIUM 47 9
cf. also Bernstein (1966) and Charnock (1967a)]. The discrepancy

with the theoretically very attractive “ 1/ 3 -power law” for the
velocity profile makes doubtful the empirical equations forq>(f) and
9 i(f) proposed by Businger (1966) [and independently also by
Dyer] and the equations of Pandolfo (1966) [see Sect. 7.4; let us
recall that both these sets o f equations prescribe that f = R i ] .
Nevertheless, both these sets of equations found some satisfactory
empirical justification and were used in certain applications [see, for
example, Paulson (1967); Businger, Miyake, Dyer and Bradley
(1967); Deardorff (1968); Krishna (1968); Yoshihara (1968); cf. also
the discussion in Kapoor and Sundararajan (1968)]. Purely empirical
graphs o f the function 9 (f) or q>(f') were plotted by Gurvich (1962;
1965) [see also Fig. 57, taken from the latter of these works; Busch
(1965), Swinbank (1968)]; in this work, the dependence of f ‘/®<p(f)
on Ri is given. All these graphs are in agreement with the assumption
of a universal dependence, but show certain small deviations from
each other. Knowing the function 9 (f) [or /(?)], it is easy to find the
dependence of the flux Richardson number Rf = f/ 9 (f) on f as well.
In exactly the same way, if we k n o w 9 (f'), where C'= z jU = tJaiX),
it is easy to find the dependence o f the ordinary Richardson number
Ri = Rf/a = on Finally, if we know the functions a (^)[its
determination will be discussed later] and 9 (f) [or 9 (f')], then we can
find the dependence o f both Richardson numbers R f and Ri on any
of the variables f and f'. As an example. Fig. 58 shows the
dependence o f f on Ri which follows from the data of Mordu-
khovich and Tsvang (1966). Empirical graphs of the dependence of
FIG . 5 7 . E m p irica l graph o f th e fu n c tio n

9 (C) a c c o r d in g to th e data o f G urvich.
Ri on f (or f') were also plotted by Panofsky (1963) [cf. Lumley

and Panofsky (1964), Gurvich (1962), Webb (1965), Pandolfo
(1966), where an artificial method is used to show that Ri = f,
Swinbank( 1968) and some other au th o rs]. An empirical graph o f the
dependence o f on f was constructed by Cramer (1967). As f->0
and RiÔ , of course, R f « f and Ri<«f'asf/ao [see Eq. (7.28)];
therefore it follows from the data o f Fig. 58 (and from all the other
data on dependence o f Ri on f) that a o is close to unity.
FIG. 5 8 . T h e e m p irical d e p e n d e n c e o f f o n R i a c co r d in g t o th e d ata o f M o rd u k h o v ich and

Tsvan g ( 1 9 6 6 ) .
We see that the precision o f the existing data on profiles and

turbulent fluxes at present is insufficient to reliably distinguish
between the various expressions proposed for universal wind-profile
functions. Therefore it does not seem worthwhile to discuss in detail
the pros and cons of various formulations. At present, it only seems
reasonable to use the physically justified laws describing the
asymptotic behavior of the universal functions and to verify their
agreement with experimental results; if possible it is also desirable to
obtain estimates of the numerical coefficients in these laws. This
program was carried out by Zilitinkevich and Chalikov (1968a) [see
also Zilitinkevich (1970)] using extensive data obtained by the

researchers at the Moscow Institute o f Atmospheric Physics during
the summers of 1963 to 1965 in the steppe near Tsimlyansk. As to
unstable conditions (the stable case will be considered later), it was
assumed that the simple interpolation formula (7.53) is valid, i.e.,
that
_-_)lniri+0r whenr. < f < 0

w hen f< f,
where the unknown constants oi and may be eliminated by the

continuity conditions of the functions /(f) and 9 (f) = f/'(?) at the
point f = fi. Therefore only two undetermined coefficients Q and
were included in this formulation of the universal wind-profile
functions on the negative semiaxis f < 0. Further, the values of these
coefficients (and of the constant k entering into the definition o f the
universal functions) were determined by the least squares method
from all the data of the simultaneous measurements of wind and
temperature profiles and turbulent fluxes and q. As a result, the
t
following least squares estimates were obtained: C2 ^ 1.25, -0.16,

K 0.43 which imply that ^ » 1.45 and fli 0.24. The 95% confidence
intervals of all the estimates proved to be rather narrow and the
measured points are concentrated along the curve (8.8) with
relatively small scatter; thus the asymptotic behavior o f the functions
/(f) and 9 (f) = f/'(f) given by Eq. (8.8) agree with the experimental
results treated by Zilitinkevich and Chalikov (unfortunately it is
restricted to the range 0 > f > - 1 . 2 only). It was also stated by
Zilitinkevich and Chalikov that the same data may also be described
without noticeable increase in the scatter by the equation
j In If I when - 00 ..0077<< ff < 0

) 1.2 f - * / ^ + 0 . 2 5 when f < -0.07
of the form (7.49), or by Eq. (7.61) when a = 1 2 , or by a

semiempirical equation related to one proposed by Zilitinkevich and
Laykhtman (1965).
An empirical verification of the “ 1/3-power law” for wind velocity
profile was also given (based on various meteorological data by R. J.
Taylor (1960a,b)and by Gurvich (1962; 1965) [cf. Figs. 54 and 56].
In all cases the data confirm the law, but the estimates obtained for
the coefficient m display some scatter; R. J. Taylor gives values ofC 2

ranging from 0.9 to 1.5, while Cj *«0.93a!.(f according to the data of
Fig. 56 and C2 1.4 according to the data of Fig. 54. Let us also
recall that C2=3/a*^^; thus the value a = 7 [recommended by
Panofsky (1965)] corresponds to C2 **1.55.
It is im portant to note that all existing data show that the
“ 1/3-power law” (which refers theoretically only to the case of very
large negative f ) begins to be valid at unexpectedly small values of f
(or f', or Rf, or Ri) of the order of -0 .1 (or even of several
hundredths). Since R f is equal to the ratio of the rates of production
of turbulent energy by thermal and by dynamical factors, this latter
result apparently shows that convection produces vertical turbulent
mixing much more effectively than wind shear. As a result the
thickness o f the dynamic sublayer (in which the thermal factors play
no significant role) in fact comprises only a small part o f |£ |.®
The measurements o f velocity profiles and turbulent fluxes in
thermally stratified flows generated in special meteorological wind-
tunnels are, at present, even less reliable than meteorological field
measurements. However, the data presented by Cermak, Sandborn
et al. (1966) and by Chuang and Cermak (1967) are quite consistent
with the deductions o f the similarity theory of Sect. 7 and show that
there are no significant discrepancies between the universal relation
ships observed in wind-tunnels with artificial thermal stratification
and in the atmospheric surface layer (both in unstable and in stable
conditions).
Now we consider the case of stable stratification. The observations
in stable conditions are much less numerous than those in the
unstable case, and the corresponding data are, in general, rather
scattered and do not agree so well with the assumption of universal
similarity, defined by only one dimensionless height (Hence,
Gurvich’s graph given in Fig. 55 must be regarded as merely tentative
without additional confirmation.) It is possible that the scatter of the
experimental points for positive values of f is explained partially by
the considerable intermittency of turbulence in stable conditions,
leading to great variability for the time-means, in view of which it is
necessary here to increase considerably the averaging period in order
® From th is v ie w p o in t , t h e tra d itio n a l in c lu sio n , in th e d e n o m in a to r o f e x p r e s s io n ( 7 . 1 2 )

for th e scale L , o f t h e v o n Kdrmdn c o n s ta n t % - l e a d i n g to an a d d itio n a l in crease in this scale
o f 2 .5 tim e s , is very u n fo r tu n a te .
to obtain reliable results. Some part may also be played by the fact
that strong inversions in the surface layer of the atmosphere are
generally accompanied by strong radiation cooUng, during which the
meteorological conditions are quite unsteady and q varies consider
ably with height. Nevertheless, we may note that even at the
beginning of the 1950’s Rider and Robinson (1951) and Halstead
(1952) found from observations that in the case of very stable
stratification, the wind profile is usually approximately linear. The
same deduction may be made from the observations of Liljequist
(1957), in the Antarctic, where strong inversions are customary.
According to McVehil, (1964) and Webb [see Lumley and Panofsky
(1964), p. 117)] for 0 < C ' c < 0 . 3 , where 1^' = 2 /L '(see above) the
formula 'p(*^')= 1 +PJC gives a farily good representation of(jp(S')
where PJ = 7 according to the data of McVehil (who worked out the
results of observations made near O’Neill, Nebraska and at the
“ South Pole” station in the Antarctic) and p j« 4.5 according to
Webb’s data. Finally, the treatment of the various meteorological
observations in temperature inversions carried out more recently by
McVehil (1964), Webb (1965) [see also Lumley and Panofsky
(1964), p. 117], J. J. O ’Brien (1965), Gurvich (1965), Busch (1965),
Panofsky, Busch et al. (1967), Hogstrom (1967b), Chalikov (1968),
Hoeber (1968), and Zilitinkevich and Chalikov (1968a) lead to very
similar conclusions which tends to lend them weight. According to
all the data where is no abrupt change of regime when the stability
increases (similar to transition from the forced to the free convection
sgime in the unstable case) and the wind profile can be described by
he unique “ logarithmic + linear” equation (7.58) in an extensive
range of positive values of f (or f ', or Rf, or Ri). As an example, we
give in the figure (A) below, comparison of the prediction from the
equation /(f') = In f ' + 7 f '+ const or, what is practically the same
thing, 9 ( f ') = l + 7 f' (the continuous line on the figure) with
McVehil’s observations (made near O ’Neill, Nebraska and at the
“ South Pole” station in the Antarctic). The figure shows very good
agreement; according to Panofsky et al. (1967) the equation 9 (f') =
1+ ' in this stability range is approximately equivalent to the
equation q>(f) = 1 + lOf of the form (7.58). The same value |3' = 10 of
the coefficient in Eq. (7.58) was obtained by Gurvich (1965) [see
Fig. 55] and by Zilitinkevich and Chalikov (1968a) [see also
Zilitinkevich (1970)] who treated by the least squares method the
extensive data of several expeditions in the steppe near Tsimlyansk.
At the same time J. J. O’Brien (1965) obtained the averaged estimate
Ug - «x, m/sec
FIG. A. The comparison o f the predictions from the “logarithmic + linear” law with = 7
related to the difference - i7i o f wind velocities at 1 m and 2 m with McVehil’s observa
tions (1964) for stable stratifications (at 0 < z / £ ' < 0 .3 ) .
|3" 5.0 (with great scatter) of the coefficient in the equation<p(f') =

1 Fichtl (1968) used a two-layer equation of the form <p(f')
= 1 + 4.5f' when 0 < Ri < 0.01 and = 1 + 7f' when 0.01 < Ri < 0.1 and
Chalikov (1968) found that /3', » 7 according to data of the Soviet
Antarctic Expedition. The value <=»1 was also obtained by Cermak
et al. (1966) from laboratory measurements in a wind-tunnel with a
stable thermal stratification. Thus we must conclude that Eq. (7.58)
with Pi close to 1 0 satisfactorily describes almost all existing data
from stable stratifications. The coefficient C3 in the ^y m p to tic
equations (7.43) and (7.44) apparently must be of the same order
(near 10). It is interesting to note in this connection that the data o f
Ellison and Turner (see below, Sect. 8.2) also lead to the conclusion
that the value of the constant R = Rfcr apparently lies between 0.01
and 0.15, so that C3 = l / R ^ 7 - 1 0 . However, for a more precise
determination o f this constant [and for a more complete elucidation
of the conditions under which the “linear law” (7.44) holds] it is
still necessary to organize special careful meteorological observations
in strong inversions and to carry out additional laboratory tests.
It is easy to see that the equation
<p(f)= 1 is equivalent to the rela
tionship 9(Rf) = ( 1 -/3 'iR 0 '‘- The de
pendence of f on R f in this case also
takes a very simple form: here f =
Rf/(1-0'iRf), Rf = r/(l+ /3 ',n [cf. Eq.
(7.20)]. Similarly if q»(f')= 1 +/3"r'then
r = Ri/(l - ^ R i ) , Ri = f7(l + r r ' ) . l f a ( 0
is considered constant (and is known),
then these relations also give the de
pendence of Ri on f. The empirical
graph of the last dependence is shown
in Fig. 59 based on the data of Gurvich F IG . 5 9 . Empirical dependence of
(1965). Ri on f in the case of stable strati
Finally, the value of the constant p fication [according to Gurvich’s data
in formulas (7.32)-(7.33), which deter ( 1 9 6 5 ) ] .
mines the behavior o f q)(C) an d /(Q for
small 1^1, proves to be very indeterminate, it depends to a great
extent on the selection of the range of values of C for which this
value is found. Let us consider the case f < 0 only (since the stable
case was discussed above). We have pointed out already that Monin
and Obukhov (1954) found that p ~ 0.6 leads to a fairly good
approximation of the wind profile in the lowest four-meter layer by
the “ logarithmic + linear” function. This result was also confirmed
later by Panofsky, Blackadar and McVehil (1960), J. J. O’Brien (1965),
Busch (1965), Bernstein (1966), and others who used an approxima
tion of the form (7.32) for a comparatively wide range of negative
values o f g. At the same time, certain researchers who determined p
from data referring to a more restricted range of values of ^ or even
with the aid o f the formula
' d ( du y
c=o
obtained considerably higher values of p, of the order of several units

[cf. the value j3 ->• 1.45 obtained by Zilitinkevich and Chalikov

(1968a)] or even of the order of 10 [see, for example, Webb (1960;
1965); Taylor (1960a); or Kondo (1 9 6 2 a)]. Thus the determination
of p from the condition o f the best approximation of the values of ^
over a limited range is dependent upon the range selected and does
not coincide with the determination o f p from the derivative of q5(C)
at zpro. It is not surprising, therefore, that in R. Taylor (1960a, b),
Takeuchi (1961), Kondo (1962a), and O’Brien (1965) a whole series
of different values of p was obtained. Consequently, it follows that
the “ logarithmic + linear” approximation must be used, in general,
only with considerable caution, while for ^ < 0, it is not advisable to
apply it widely. As we have seen in the case of unstable stratification,
the transition from the logarithmic law to the limiting “ 1/3-power
law” of the theory of free convection takes place in a very thin layer;
thus formulas (7.32) and (7.33) which in fact refer only to this
transitional layer, have little meaning here.
8.2 Data on Temperature and Humidity Profiles
In addition to observations on the wind-velocity profile u{z), the
most common meteorological observations also include observations
on the mean temperature profile T{z) and, to a lesser extent, of the
mean humidity ^>(2 ) in the surface layer of the atmosphere. For
temperature measurements, various types of thermometers are used.
They are fitted with special devices to shield them from the direct
action of solar radiation. Electrical instruments are most convenient
for measuring the profile T{z). These include, for example, resistance
thermometers and thermistors, which register the current variation
due to the variations of resistance of a conductor or of a special
semiconductor (thermistor), induced by variations of the air temper
ature, and thermocouples, with one junction maintained at a
constant temperature, and the others exposed to the atmosphere
[see, for example, Laykhtman and Chudnovskiy (1949), Mclbroy
(1955), Krechmer (1957), Lettau and Davidson (1957)]. The
electrical instruments permit the direct measurement of small
temperature differences T{z\) — f (22 ), which allows the precision of
such measurements to be increased to a few hundredths or even to
one hundredth of a centigrade degree. To measure the humidity, the
most common method is the comparison of the readings of wet- and
dry-bulb thermometers. At various times, many other principles have
also been suggested as a basis for these measurements [see, for
example, Laykhtman and Chudnovskiy (1949), Middleton and
Spilhaus (1953), Lettau and Davidson (1957); International Con

ference, 1965]; so far, however, none of these principles has led to
the development of a simple and reliable device which would give
values of 0 with high relative accuracy. Thus it is still very difficult to
obtain high-quahty data on the profiles of i&(2 ) .
The general form of the temperature profile
f (2)for both stable and unstable stratification
is well known at present from many observa
tions. A typical example illustrating this form
is the data o f Monin (1962) in Fig. 61,
obtained by averaging 61 individual tempera
ture profiles observed during the summer of
1951 in open-steppe conditions (the averaging
carried out over the same six stability-
homogeneous groups mentioned at the begin
ning of Sect. 8.1 in connection with Fig. 52).
From Fig. 61 it is clear that in unstable
stratification, the rate o f decrease o f f{ z ) as In
z increases is slower than the linear law, while
for stable stratification, the rate of increase of
T{z) with increase o f In z is faster than linear.
A similar rule has already been noted above
with regard to the stratification-dependence o f
the law of increase of wind velocity with height 0.5 1 2 4 d 12
(see Fig. 52). Comparing Figs. 52 and 61, we z (m eter s)
may even assume that in all stratifications the FIG. 61.t E m p irica l p r o
files o f th e m e a n te m
profiles of wind velocity and temperature are
perature in th e surface
similar to each other, i.e., that for any stability la yer for d iffe r e n t strati
of the atmosphere the ratio fic a tio n s.
(8.9)
for any two heights zi and zo within the surface layer will assume the
same value (positive for inversions, zero for neutral stratification, and
negative for unstable stratification). To verify this assumption it is
sufficient to compare the two shape functions
u ( z ) — u {Za) T (z)-T (z,)
and ( 8 . 10)
“ (•?!) —W(^0) T (z{)-T (z,)
where the heights Zo and zi are fixed, and 2 is a variable, and to check
t O u e t o A u th o r ’s r evision F ig. 6 0 w a s d e le te d .
whether or not they are the same; another method consists of the
verification o f the constancy of the temperature to velocity gradient
ratio with z. Both types of verifications have been carried out by
many researchers at various times and to various degrees of accuracy
[in particular, by Pasquill (1949), Rider and Robinson (1951),
Panofsky (1961b), Swinbank (1964; 1968), McVehil (1964), Gurvich
(1965), Busch (1965), Swinbank and Dyer (1967), and Charnock
(1 9 6 7 b )]. In the older works it was always found that the wind and
temperature profiles in the surface layer were particularly similar
when the deviations from neutral stratification were not too great;
often it was even asserted that this similarity held exactly for any
stratification, or at least for any unstable stratification. However,
the recent results of the Australian group (which are not yet reliably
confirmed by the data of other researchers) indicate the existence of
a considerable shape difference of the wind-velocity and temperature
profiles for unstable stratifications (cf., for example. Fig. 62 below).
-R i
F IG . 6 2 . Sh a p e fu n c tio n s - [ X{ A) - X ( l ) ] l [ X { \ 6 ) - A"(l)] for m e a n v e lo c ity
ill), tem p er a tu re ( D , and h u m id ity (t?) vs. R ic h a r d so n n u m b e r at 2 m w ith corre
s p o n d in g standard errors o f th e m ean [a c c o rd in g to S w in b a n k and D y e r ( 1 9 6 7 ) ] .
Thus, we must conclude that the profiles o f ii{z) and T{z) are not
similar to each other although a completely reliable quantitative
determination of the degree of dissimilarity for a wide stability range
requires further careful investigation (cf. the discussion of the
function a(f) below). _
The situation is even worse for humidity profiles The
comparatively old measurements o f these profiles, for example, those
by Pasquill (1949), and Rider (1954), all showed that within the
lowest few meters, the shape function jg dose to both
the functions (8.10). In other words, the early measurements of
humidity profiles i&(z) for various stratifications did not contradict
the statement that these profiles are similar to those of the wind and
the temperature; let us recall that the wind and the temperature
profiles are quite similar according to early data. However, more
recent measurements by Crawford (1965), Hogstrom (1967a, b), and
Swinbank and Dyer (1967; 1968) [see also Dyer (1967)] show that
humidity profiles are apparently similar to the temperature profiles,
but, in general, are dissimilar to wind profiles. A more complete
discussion of these results will be given at the end of this subsection;
here we reproduce Fig. 62 [taken from Swinbank and Dyer (1967)]
where the dependence of the three shape functions Sj^= [A^(4) - 5( 1)] /
[X(\ 6 ) - A'(])]on the Richardson number at 2 m is shown (for negative
Ri only) according to all the data from the four Australian
expeditions. The arguments in the definition of the shape function
are the heights in meters; the mean wind velocity 17, the mean
temperature f and the mean humidity 5 are used as the quantity X
(the corresponding functions are indicated as Su, Sj, and S^)\ the
standard errors of the mean values over the stability-homogeneous
groups presented in Fig. 62 are also shown (except, however, the
case of the wind-function over the middle of the stability range
where the errors are too small to show). According to the data in Fig.
62, the wind and humidity shape functions take approximately the
same value, namely, 0.5, which corresponds to a logarithmic profile
for close to neutral stratification; there are no values for S t with
|Ri|< 0.01 in the diagram because the accuracy of temperature
difference measurements is clearly insufficient for determination of
S t in almost isothermal conditions. However, with increasing
instability the shape functions for all three variables depart increas
ingly from the neutral value; moreover, S t and S& are practically
indistinguishable over the whole range of Ri, but they both depart
considerably from Su values. The results shown in Fig. 62 are very

significant and agree completely with the deductions from the
intuitive physical reasoning on the similarity o f the mixing mechan
ism for heat and passive material admixture advanced at the end of
Sect. 7.3. We must keep in mind that the accuracy of humidity
profile measurements is rather poor at present and that even the data
of the temperature measurements by the Australian group show
certain regular departures from some data from other sources [e.g.,
Zilitinkevich and Chalikov (1968a)]. Hence the results in Fig. 62
must be regarded at present as merely preliminary and thus require
further careful verification.
By the general formulas of similarity theory (7.24), the ratio (8.9)
may be written as
^ / i ( C i ) - / i (Co) .g
U(Z,)-U{Z0) M, /( f :i ) - /( C o )
where = /(C) and /i(C) are universal functions

describing the dependence of the wind and the temperature profiles
on the dimensionless height. Thus it is clear that the constancy of
this ratio for all 2i and zo means that the functions /(Q and fi(?)
differ only by a constant multiplier; but
^ K ’
thus the similarity of the wind and temperature profiles is equivalent

to the height-independence, and independence of Richardson num
ber of the ratio a of the eddy diffusivities for heat and momentum.
The data shown in Fig. 62 and other similar data give definite
grounds for assuming that the ratio of the eddy diffusivities KWK is
not strictly constant. However, as yet no well-established data exist
which provide a completely reliable quantitative determination of its
dependence on C = zjL-, for a more detailed discussion of this point
see below. With regard to the eddy diffusivity for humidity, at
present we can say only that all existing data on the humidity
profiles are in agreement with the assumption that /C»//Ct= const
[and even with the assumption that = 1)].
A more complete investigation of the temperature profile from the
viewpoint o f similarity theory requires the formulation of the
empirical dependence of the quantity [f(z) — T{a\L\)]jT^
on the argument C= Here a is a given coefficient, for example,

1/2 or 1/10, a n d ! and must be determined from the values o f
and q, obtained from direct measurement or by estimation using
some rehable indirect method, on the basis Eqs. (7.12) and (7.14).
The function [T{z) — f (a|L|)]/7, was formulated by R. J. Taylor
(1960a) [using the data of Rider (1954) and Swinbank (1955)],
Takeuchi (1961) [from the data of extensive observations o f the
wind and the temperature profiles, carried out in 1956 near O’Neill,
Nebraska], and Gurvich (1955), and Zihtenkevich and Chalikov
(1968a) [using the data of extensive measurements near Tsimlyansk
(U.S.S.R.), in 1962-1965; see Figs. 63 and 64 taken from Gurvich].
Since Takeuchi determined the values of L and 7", using an ap
proximate indirect method on the assumption t h a t = his data
cannot be used for determination o f a(C)- The old data treated by
R. J. Taylor show that even for —0.03, a transition occurs in the
atmosphere to a “ free convection regime,” characterized by the
“ 1/3-power law.” Moreover, these data were also used by Taylor for
an approximate estimation ofa = j ' ( Q / / i (U- He found that for 1^1
0.03, i.e., in conditions of “ forced convection,” a(?) = ao ~ 1.2
according to Rider’s data, and ao ~ 0.8 according to Swinbank’s data;
for L, < —0.03 (for free convection), a{l) = a-oo « 1 .7 according to
T U )-T (^ )
h
to
-1 0 r
/a
- 2.0
- 3 .0 r
{ ’ *c ,ms6
- 4 ,0
•
'-ln^+cor>
1
0.0070.0020.0030.070.02 0.05 0.7 0.2 0,5 7.0 2.0 50
FIG. 63. E m p irical graph o f th e fu n c tio n / i (C) —f\ ^--- for

C < 0 [a c c o rd in g to G u r v ic h ’s d ata ( 1 9 6 5 ) ] .
rrv-rfwJ
T*
zo
6.0
c o n st
5.0 f
^.0 t
/
3.0
2.0 /
10
1 1 1
0
0.0050.o /o .o 2 a g y u ^ ^'0 2 0 3 0 .3 0 .7 1 0 2 .0 3 0 5 .0
-to
- 2.0
-3.0
-t c o n s t
-40
FIG . 6 4 . E m pirical graph o f th e fu n c tio n / , (C) —
/ i (0 , 1) for C > 0 [a c c o rd in g to th e da ta o f Gur-
v ic h (1 9 6 5 )].
Rider, and a_oo = 1.3 according to Swinbank [see also, Deacon

(1959)]; finally, in inversions (for ^ > 0 ) , a == 1.8 according to
Rider and a « 1.0 by Swinbank’s data. The considerable discrepancy
between the data of Rider and Swinbank is entirely natural if we
consider the poor accuracy of their turbulent flux measurements and
that the quantity of observations which could be used was very small
in both cases. Later, numerous attempts were made to estimate the
values o f the function a(^) =KTlK from various existing observations
in the atmospheric surface layer. For example, Swinbank (1964),
Gurvich (1965), Busch (1965), Dyer (1965; 1967), J, L. H. Sibbons
[see Swinbank (1966)], Mordukhovich and Tsvang (1966), Pandolfo
(1966), Record and Cramer (1966), Charnock (1967b), Zubkovskiy
(1967), Businger, Miyake et al. (1967), Prasad and Panofsky [see
Panofsky, Busch, Prasad et al. (1967)], Kapoor and Sundararajan
(1968), Zilitinkevich and Chalikov (1968a), and some others.
However, the scatter of all the existing estimates is very great at
present, as can be seen, for example, in Fig. B where the data
obtained in AustraUa, U.S.S.R., and the U.S. are shown. The dotted
line in the figure gives the results of Charnock’s treatment (1967b) of
Australian data, the broken and continuous lines were proposed by
Zubkovskiy (1967) and Zilitinkevich and ChaUkov (1968a) based on
the observations near Tsimlyansk covering a much more restricted
stability range (according to the last authors, a(f)ss 1 and tp(f)®“
-4 4 -4 0 -2 0 - 1.6 - 1.2 - 0.8 -0 .4

K-z/L
FIG . B. D ata o n th e d e p e n d e n c e o f a = K ^ / K o n • - K e r a n g , A ustralia [ S w in b a n k ( 1 9 6 4 ) ] ;
o - T s i m l y a n s k , U .S .S .R . [M o r d u k h o v ic h and T svan g ( 1 9 6 6 ) ] ; X - R o u n d Hill, M ass., U .S .A .
[P a n o fs k y et al. ( 1 9 6 7 ) ] . T h e d o t te d lin e w as p r o p o se d b y CTiarnock ( 1 9 6 7 a ) , th e b r o k e n
lin e b y Z u b k o v sk iy ( 1 9 6 7 ) , th e c o n tin u o u s lin e b y Z ilitin k e v ic h and C halik o v ( 1 9 6 8 a ) .
<pi(?) for all values of z/Z, observed in Tsimlyansk). The same degree of
scatter is also shown by all the other data on a(^).The scatter may be
explained naturally by the necessity to differentiate empirical wind
and temperature profiles of low accuracy and to use values o f the
turbulent fluxes t and q which are now measured with quite low
precision; we must also remember that the time and space variability
of these fluxes is almost unknown for the atmospheric surface layer;
however, the preliminary data of Mordukhovich and Tsvang (1966),
and Dyer (1968), on height variability of x and q show that it is quite
large. Thus the general deduction from all the existing data is that a
reliable estimate of a(^) requires considerably more accurate mea
surements of the wind and temperature profiles and of the simultan
eous values of x and q at the same observation point than those
presently available. On the whole the existing data show only that
a(0) = Go is close to unity, i.e., they are in agreement with the best
laboratory measurements mentioned in Sect. 5.7, according to which
tto *= 1.1; with increase of instability, the ratio a increases and with
increase of stability it seems to be sUghtly decreasing. However, the
estimates of the limiting value a_«, are presently quite uncertain: the
Australian observations imply the value a-co^ 3 -3 .5 . Nevertheless,

some investigators are incUned to use considerably lower estimates
(close to 2 or even between 1 and 1.5). The data for atmospheric
inversions are even more scattered, and much less numerous than
those for unstable stratification; according to the treament by Prasad
and Panofsky [Panofsky et al. (1967); see also Busch and Panofsky
(1968)], the mean value of the eddy diffusivity ratio a in surface
layer inversions is close to 0.7; the data from nonatmospheric
sources on values o f a in very stable conditions will be discussed at
the end of this subsection.
Many of the existing determinations of a(f) were based on the
evaluation of the universal function = connected with
o(f) by the relationship a(^) =(p(?)/ 9 i(^). The graphs 9 1 (f) were
plotted, for example, by Gurvich (1965); see Fig. 65, Dyer (1965;
1967), Chamock (1967b), and Swinbank (1968); here the graph of
f 9 i(f) is given. In the discussion of these results the question o f the
validity of the “ 1/3-power law” (7.35)—(7.36), i.e., of the relation
9 i(f) ~ , and of the range of its applicabihty to unstable
atmospheric surface layers, are of primary importance. We have
already mentioned the data of R. J. Taylor (1960a) which shows that
in an unstable atmosphere the transition to the “free convection
regime,” i.e., to the regime characterized by the “ 1/3-power law,”
occurs at unexpectedly small values of - f o f the order of several
hundredths. However, the first indications o f the vahdity of this were
FIG . 6 5 . E m p irical graph o f th e universal

fu n c tio n cpi (C) [a c c o r d in g to th e data o f G ur
vich ( 1 9 6 5 ) ) .
obtained even earlier, in the works of Priestley (1955; 1956) and R. J.

Taylor (1956a) where the “dimensionless heat flux” I was first
introduced; we shall define this quantity later. For very small
negative Richardson numbers and values o f the transition from the
“ forced convection regime” characterized by a near-logarithmic
temperature profile to the “ free convection regime” with its
“ 1/3-power law” follows from Webb’s result. Webb (1958) obtained
data on temperature-profile observations made near Edithvale,
Australia, over a period of several years. Thus, for a series of
observations with unstable stratification, he calculated the quantity
__1__ T(8) — T(2)

1.5 log 4 iof\
[the numbers in parentheses denote the height in meters] and plotted

the dependence of this quantity on the Ri number at a height o f 1.5
m (see Fig. 66). The horizontal continuous lines in Fig. 66 denote
the values of T = 1 , corresponding to the logarithmic p r o f ile r (2 ) ~
const —In 2 , and T = 0.73, corresponding to the “ 1/3-power law”
T ( z ) ^ const-f 2 3. For -Rii.s, the smallest observed values of
which are close to 0.015 or even less, F is actually close to unity,
however, it then begins to decrease, and for Ri as -0 .0 3 it takes a
value close to 0.73. This value is maintained thereafter for a
considerable range o f Richardson numbers. The dotted line in Fig. 66
denotes the gradual transition from T = 1 to F = 0.73, which would
have been observed if observation of the relationship
FIG . 6 6 . D e p e n d e n c e o f F o n R i in th e case o f u n sta b le s tr a tific a tio n ,

a c co r d in g t o m e a su r em en ts.
which corresponds to the logarithmic profile, had continued right up

to the height z at which Ri = —0.03, while above this, the
relationship ~ ^ ^ was satisfied.
An interesting and unexpected pecuUarity of Fig. 66 is that
beginning from approximately Rii.s = 0.4, the value o f F once again
begins to decrease. If this result is correct and has a universal
character (this is still quite doubtful) then it follows that with
increasing instability, but fixed height, or with increasing height, but
fixed unstable stratification, an instant occurs when the dimensional
arguments which imply the “ 1/3-power law,” cease to operate. This
agrees somewhat with the fact that the “ 1/3-power law” is not
confirmed by the rather rough laboratory convection experiments of
Thomas and Townsend (1957) and Townsend (1959), where the
horizontal velocity u ( 2 ) was equal to zero. However, Townsend’s
results may be explained by the single fact that with zero (or very
small) horizontal velocity, horizontal mixing is greatly reduced, and
as a result, the turbulence at a given height z is horizontally
homogeneous only over a very large averaging time. There are also
some other explanations for the apparent inapplicability of the
“ 1/3-power law” to Townsend’s experiment. For example, Deardorff
and Willis (1967) concluded that existing laboratory data neither
contradict nor confirm the assumption that at sufficiently large
Rayleigh numbers artificial convection in a box in the complete
absence of a horizontal mean velocity will lead to a —1/3-power
temperature profile with a coefficient Ci close to that observed in
the atmosphere. On the other hand, there are some slight grounds for
assuming that in the absence of horizontal mixing due to a horizontal
mean velocity, the condition of generation of rising convective jets
within the sublayer of molecular thermal conductivity will affect the
whole turbulent regime. Thus the statistical characteristics of turbu
lence at all heights will depend in this case on the molecular diffusivity
X [cf. Malkus’ convection theory (1954b)]. This question was dis
cussed at some length by Townsend (1962a), Priestley (1962) and
Webb (1962), who, however, did not obtain completely definite con
clusions. Meanwhile, new measurement data were obtained by Dyer
(1965; 1967) and Swinbank (1968); their results also suggest that
with very unstable stratification (for f < — 0.6 approximately) some
deviations from the free convection “ 1/3-power law” may occur in
the atmosphere, but these deviations now appear to be considerably
smaller than those found earlier. (Dyer and Swinbank suggest that
•Pi(f) ~ for f < - 0 .6 and that the law <Pi(n ~ iri'® '''* is in
excellent agreement with all the data where —0.1 > f > —5 in
contrast with the law z(97’/3z) ~ z"' which follows from Malkus’
theory and was advocated in the 1962 discussion.) The same problem
was discussed in the above-mentioned paper by Deardorff and Willis
(1967), but their deductions are also quite uncertain. Because of the
obscurity of the question, in further consideration of unstable
stratification, we shall, as a rule, have in mind only a layer with
modest values of Ifl, where the “ 1/3-power law” undoubtedly is well
satisfied, beginning from a height z of the order o f several
hundredths of |Z,|.
To determine the numerical value of the coefficient C, in the
“ 1/3-power law” (7.36), it is convenient to formulate the dimension-
less heat flux
Vo I
which is a single-valued function of t, and of Ri. In conditions of free

convection, this quantity will take a constant value, namely,
£
^-oo ’ ^ts asymptotic behavior for small S and Ri is given by
the formula
* J -1 --
a 2 x2 I c I 2 =a^x2|Ri | 2.
*
The schematic form of the dependence of q (Ri) on Ri is shown in
Fig. 51, Sect. 7.4, with logarithmic scales on both axes. This
dependence was also investigated empirically by Priestley (1955;
1956), R. J. Taylor (1956a), Perepelkina (1959a), Busch (1965),
Mordukhovich and Tsvang (1966), and Swinbank (1968), using
various data from simultaneous measurements of T { 2 ), ^7(^), and q .
Figure 67 gives an empirical graph of q (Ri), plotted according to
Swinbank (1955), R. Taylor (1956a), Perepelkina (1959a) and
Mordukhovich and Tsvang (1966); the vertical segments in Fig. 67
indicate the mean square deviations within various stability-
homogeneous groups, with the exception of the data of Perepelkina,
who gave only the mean values. The individual values of q are
characterized in all cases by considerable scatter, which is connected
with the comparatively low accuracy of all the existing
1.0
-Ri
FIG . 6 7 . T h e d e p e n d e n c e o f q o n Ri a c co r d in g to th e data o f S w in b a n k and R . J. T a y lo r ( o ) ,
P e rep elk in a (A ) , and M o rd u k h o v ich an d T svan g ( • ) .
measurements; however, the mean values in Fig. 67 agree well with

the assumption of the existence of two limiting regimes q — |R i| ^
and q = q__^ = const, denoted by continuous lines in the figure. The
transition region between the limiting regimes is extremely narrow
according to the early data o f Swinbank and R. J. Taylor; the data of
Perepelkina, Mordukhovich and Tsvang, and all the other existing
data on q and <Pi(?) [including those of Panofsky, Blackadar and
McVehil (1969); Gurvich (1965); Busch (1965); Dyer (1965; 1967);
and Zihtinkevich and Chalikov (1968a)] indicate a somewhat
smoother transition, but confirm that it is quite sharp in comparison
with other transitional regimes encountered in physics. According to
all the data, with the exception of the recent results of Swinbank
(1968) which will be mentioned a little later, the free convection
regime and the “ 1/3-power law” begin to be valid at unexpectedly
small values of |Ri| and l?l in the interval 0.02—0.05; this fact was
already noted above. The empirical value of the proportionality
_'
coefficient in the asymptotic relationship <7( R i ) ~ |R i | 2 is close to
0.17 [Deacon (1959)], which agrees well with the generall;/ accepted
values X ~ 0.4, ao ~ 1. However, the data on the constant </-=care not
in satisfactory agreement with each other. According to the early
Australian data o f Swinbank and R. J. Taylor, after introducing an

instrumental correction, 0.9 [Priestley (1959a); Deacon
(1959)1; according to Perepelkina, q-oo^ 0.8; according to Gurvich,
and Zilitinkevich and Chalikov (who determined the value of Ci =
3/(l-oo)^(f from measurements in a restricted range of negative values
of Ri), </ 0 .7 -0 .7 5 ; according to fairly rough data of Cramer and
Record, and of Panofsky, 0.95 and 7 -» ~ 0.75, respectively,
while according to Mordukhovich and Tsvang, 0.8. The
agreement among all these estimates clearly is quite good. Unfortu
nately the more recent Austrahan measurements of Swinbank and
Dyer (1968) do not support the results of the other investigations.
According to Dyer (1967) the best value of q-oo [corrected for the
error caused by the use o f the difference ratio AT/Az instead of the
derivative dT/dz] is close to 1.15 for the range 0.02 < \z/L\ < 0.6. (In
his eariier paper where no corrections was made. Dyer (1965) gave an
even slightly greater estimate; a greater value of q also corresponds to
Dyer’s data for lz/L| > 0.6.) The analysis of the same data by
Swinbank (1968) showed that in fact q (Ri) does not take on a
strictly constant value according to the Australian measurements of
1962-1964. Dyer’s estimate corresponds in Swinbank’s graph to a
flat minimum o f q (Ri) at Ri —0.1; however, with increase of —Ri
the value of q also increases and at the greatest value of —Ri
considered by Swinbank (close to 1) ^ shows only a slight tendency
to “ flatten-out” to a constant value (slightly greater than 1.5), but a
strictly constant value has not been established even at such strong
instability. The explanation of the obvious discrepancy among these
results is still unclear.
Knowing q-oa, it is easy to evaluate the coefficient C, =
and vice versa. Therefore all the determinations of Ci and of q-.^. are
closely connected with each other. For $ . 00= 0.75, 1, 1.15, and 1.5,
we obtain the following values of C^: C i« 3.6, 3.0, 1.7, and 2.3,
respectively.
Of course, many dimensionless combinations of bfldz,duldz.u^,
q/cp8 , g/To, and z also exist, different from q (and A'; see Eq. (8.6),
Sect. 8.1); they may all be represented by the principal wind velocity
and temperature functions /(?) and /i(f) [or<p(f) and<Ti(f)] and may
also be determined from experimental data. Thus, for example,
Pasquill(1949), Rider (1954) and Perepelkina ( 1959a) considered the
empirical formulation of the function
If /(?) is known, then the values of f i(Ri) also allow us to determine

the universal function fi (C) which describes the temperature profile.
However, the accuracy of the data on Fi (Ri) in the comparatively
early works cited is completely insufficient for use in this manner.
Swinbank(1968) constructed the empirical graphs of the functions
a z dT C
F A O = F4(R1) = - - ------------ .
To du ^2 ,p(^)
His graphs are characterized by fairly small scatter; the data onFs(f)
permit the conclusion that Kj/K = a ( f ) « 2.7 If]'/** over a wide range of
negative stabilities. However, we have already noted that there are
some unusual discrepancies between the data of Swinbank (1968)
and those of certain other investigators.
The function analogous toFi(Ri), but for the humidity profile.
was formulated empirically by Pasquill (1949) and Rider (1954)

from measurements of the wind and humidity profiles and measure
ments o f evaporation with the aid of weighing lysimeters. Clearly,
since
da
~dz
[cf. Eq. (8.6) in Sect. 8.1 ], by comparing the values of Fz, F\ and K,
we could in principle also draw definite conclusions about the
behavior of the ratios K^/K t and K d/K. However, due to the
considerable inaccuracy of the data of Pasquill and Rider on
humidity profiles and evaporative fluxes, they are useless for such
a comparison.
Later, Charnock and Ellison (1959) found a strong correlation

between fluctuations of humidity and temperature above sea level
and considered this to be rather uncertain indirect confirmation of
the equality between the eddy diffusivities Ks) and K t for water vapor
and heat. However, much more definite (although also not quite
rigorous) proofs o f the hypothesis that K» = K t were obtained still
later by Crawford (1965), Dyer (1967), Swinbank and Dyer (1967)
and Hogstrom (1967b). One of the proofs consists of the diagram
shown in Fig. 62, which was discussed above. Another method is
based on the investigation of the dependence on thermal stability of
the dimensionless flux o f water vapor
* /
PQ(g/TQ)^^^\dT/dz\^^^\dd /dz\z^ C^\f\( 0 \'^^^\f2 ( 0 \
which is quite analogous to the dimensionless heat flux q. The first

empirical graph of the function /(R i) in the range -0 .0 0 0 3 > Ri >
—2.5 was constructed by Crawford (1965). He used the data on
evaporation and on the profiles of wind velocity, temperature and
absolute humidity, obtained at a special site during 1962—1963 from
measurements with a weighing lysimeter, an infrared hygrometer (the
air was drawn through it from different heights), and a thermo-
couple-anemometer mast. The resulting graph is quite similar to that
in Fig. 67; for - R i < 0.025 the points are concentrated near the
straight line / = 0.2|Rir*^^, with the coefficient 0.2 close to the
coefficient 0.17 in the similar relation for q in the forced convection
regime, whereas for - R i> 0.025 they correspond with relatively small
scatter to a constant value / =» 1.4. The last value is slightly greater
than the majority of estimates of q for the free convection regime,
but is quite close to the estimate q ^ 1.32 obtained by Dyer (1965).
Later, Dyer (1967) calculated the quantities q and / and also (pi(0
= z/r^97’/9z and <l)2 (?) 9^/9z from the data of the 1962—1964
Australian expeditions where the evaporative water-vapor flux was
measured by a method described in the next subsection, and the
humidity profile by a special crystal dew point hygrometer with air
sampled from various heights. It was found here that the functions
fpi(f) and 9 2 (f) are practically identical over the whole range of
unstable stratifications investigated, although the 92 data are some
what more scattered. In the range - 0 .0 2 < - 0 .6 both functions
satisfy the “ 1/3-power law” very well and the value * evaluated for
this range is close to 1 .1 0 when corrected for the small error due to
the approximation of the derivatives bT/dz and 9t?/92 by the

corresponding finite difference expressions; this error, which leads to
an overestimate of f and q, was neglected by Crawford and Dyer in
1965. The last estimate practically coincides with the estimate
1.15 obtained by Dyer in the same work. Finally, Hogstrom (1967b)
used the comparatively rough data of meteorological and hydro-
logical observations in the plains of southern Sweden during
1961-1963, where humidity fluctuations were measured by an
Australian instrument and the humidity profile was measured with
poor accuracy by the usual meteorological hygrometers. He obtained
an estimate / « 1.25 (without a finite difference correction), which
agrees with those of Crawford and Dyer within the limits of the
experimental errors. Since f/q = K ^ I K j , all the above-mentioned
data clearly support the hypothesis that K ^ = K j for all neutral and
unstable stratifications.
So far, we have considered only data referring to negative (or, at
least, only small positive) values of Ri and Data referring to strong
inversions have until now been obtained only rarely in atmospheric
measurements and display great scatter. Consequently, further
careful investigation of this subject is necessary. Nevertheless,
Liljequist’s observations (1957) in the Antarctic seem to confirm in a
number of cases the presence of a linear temperature profile in strong
inversions, as predicted by similarity theory under some additional
assumptions (see Sect. 7.3). On the other hand, if we adopt the
hypothesis that the eddy diffusivities for heat and for a material
admixture are the same also for stable stratifications, then in the
investigation o f values of a = Kr/K = Ki,/K in extremely stable
conditions, we may use the data from oceanographic observations
and laboratory experiments, for flows of liquid with very stable
density stratification caused by a negative vertical salinity gradient.
The first and best-known data of this type were obtained by G. I.
Taylor from oceanographic observations in the Kattegat. These data
showed that considerable turbulent exchange may exist even when
Ri is of the order of 10, and that a = K^JK is then so small that Rf =
a Ri remains much smaller than unity [see Proudman (1953) and the
discussion by Stewart (1959)]. The evaluation of these data in
Proudman’s book show that a ~ 0.03 - 0.05 for 4 < Ri < 10 (black
squares in Fig. 68). Values of a = Kr/K ^ 0.02- 0.05 in close
agreement were established by direct measurements of K and Kr in a
layer of water under the ice on Lake Baikal [see Kolesnikov (I9 6 0 )].
Later, comparatively rough data on the dependence of a on Ri for Ri
> 0 were obtained in the laboratory experiments of FJlison and
Turner (1960). In these experiments a turbulent flow of salt water

with negative vertical salinity gradient was set up in a tube of
rectangular cross section (with glass side walls; in one section of the
tube the profiles of velocity and density were measured, and also, by
indirect methods, the eddy diffusivities K and were estimated.
Figure 68 gives the results obtained and the oceanographic data of
Taylor and Proud man. In this figure, the scatter of the experimental
points is very considerable, and, moreover, there is a systematic
deviation of the data referring to e = 1 cm and z = 1.5 cm; however,
the general tendency of a to decrease with increase of Ri can be seen
fairly easily. The continuous Unes in Fig. 68 correspond to the values
of a obtained on the basis of the semiempirical formula
a o (l-R f//? )
a(Rf): Rf = aRi, (8.14)
(1 —Rf)*
proposed by Ellison (1947). It is assumed here that ao = 1.4 which is

apparently too high; see Sect. 5.7; the values of /? = Rfcr which
correspond to each curve are written above it. The experimental
points of Fig. 68 more or less agree with the deductions drawn from
Eq. (8.14); according to these data, the best estimate o f R = Rfcr lies
somewhere between 0.10 and 0.15. Another rough estimate of the
value of Rfcr was obtained by Webster (1964) from measurements in
a wind-tunnel. In these measurements, the velocity shear and thermal
FIG . 6 8 . C o m p a r iso n o f th e data o f G. I. T ay lo r and o f

E lliso n and T urner c o n c e r n in g th e d e p e n d e n c e o f a o n
Ri w ith th e fo r m u la o f E lliso n ( 8 .1 4 ) : • o - d a t a o f E lli
so n an d T urner, relating to h e ig h t z = 1.5 cm ; ^ ^ - d a t a
r ela tin g t o z = 1 cm ; ■ - T a y l o r ’s data. T h e o p e n s y m
b o ls refer to le ss reliable e x p e r im e n ta l data.
Stratification were produced by two grids, one with variable heating

of the bars and the other with variable bar diameter. According to
Webster’s result, Rfcr 0.35. Some other estimates o f Rfcr were
reported by HOgstrom (1967b) and by Webster (1964); none of them
is very reliable and most o f them assert that Rfcr is somewhere
between 0.08 and 0.4.
8.3 Methods of Measuring Turbulent Fluxes of Momentum,
Heat, and Water Vapor
A s alread y o b s e r v e d , a c o m p le te v e r ific a tio n o f th e d e d u c t io n s fr o m s im ila r i^ th e o r y
d isc u sse d in S e c t. 7, req uires a k n o w le d g e o f n o t o n ly th e p r o files /7 ( z ) , f ( - ? ) , a n d 0 ( z ) , b u t
also o f th e tu r b u le n t m o m e n t u m flu x (shear stress) t — o //f , th e h e a t flu x ^7 and th e w ater
v ap or flu x (rate o f e v a p o r a tio n or c o n d e n s a t io n ) /. M o reov er, sin c e th e q u a n titie s x (o r w«),
q and / are very im p o r t a n t c h a r a cteristics o f a tm o sp h e r ic tu r b u le n c e , d e scr ib in g m o s t
d ir e c tly th e in t e r a c tio n b e t w e e n th e a tm o sp h e r e an d th e u n d e rly in g su r fa c e, it is natural
th a t th eir d e te r m in a tio n b e o n e o f th e c en tral p r o b le m s in th e p h y sic s o f th e a tm o sp h e r ic
surface layer. T h u s, at this p o in t it is relevan t to give at lea st a b r ie f d e s c r ip tio n o f the
princip al m e t h o d s o f m easuring tu r b u le n t flu x e s .
T h e m o s t universal d ir e c t m e th o d o f m easuring q, and j is th e flu c t u a tio n (o r e d d y
c o rr e la tio n ) m e t h o d , w h ic h c o n s ists in r ec ord in g th e tim e variation o f th e flu c tu a tio n s
11', T \ an d \Y at a fix e d p o in t, a n d s u b s e q u e n tly e v a lu atin g th e tim e -m e a n s o f th e
p r o d u c t s u 'w \ w ' T \ and S in ce
___ 1 __ _
= q — Cj,pQw'T\ y = p,jtc;'D', (8 .1 5 )
th e s e m e a n s d e te r m in e all th e q u a n titie s o f in te r e st t o us. H o w e v er , th e in s tr u m e n ts u se d for

flu c t u a tio n m e a s u r e m e n ts m u st r ecord w it h o u t d is to r tio n all d eta ils of th e m icro-
m e te o r o lo g ic a l flu c t u a tio n s , w h ic h c o n tr ib u te su b s ta n tia lly to th e values o f u^,,q, and /. T h e
la tter r e q u ir e m e n t im p o se s im p o r t a n t lim ita tio n s o n th e p aram eters o f th e in s tr u m e n ts
w h ic h are d if fic u lt t o d e te r m in e a c cu ra tely w it h o u t ad va n ce k n o w le d g e o f th e d eta ils o f th e
flu c t u a tio n s th a t a ff e c t th e tu r b u le n t flu x e s . In f a c t th e in f o r m a tio n req u ired c o n c er n s th e
spectral c o m p o s itio n o f th e tu r b u le n t flu x e s w h ic h still is s tu d ied very in a d e q u a te ly ; w e
shall d iscu ss this in greater d e ta il in C h ap t. 8 o f V o lu m e 2 o f this b o o k . M o reov er, w h e n th e
flu c t u a tio n m e t h o d is u se d to d e te r m in e tu r b u le n t flu x e s , w e n e e d to k n o w h o w to c h o o s e
th e tim e-averagin g interval in s u c h a w a y th at th e m e a n values o b ta in e d w ill b e d o n e c lo s e to
th e m e a n ta k en a c co r d in g to s o m e sen sib ly d e te r m in e d s ta tistic a l e n s e m b le , a n d so th a t th e
gradual “ d r ift o f th e m e a n ” m e n t io n e d in S e c t. 7 .1 w ill n o t o c cu r. A ll th ese fa c ts m a k e the
m e a s u r e m e n t p r o c ess far m o r e c o m p lic a t e d , and in th e fin al a n a lysis lead t o r esu lts th a t are
n o t a lw a y s s u ffic ie n t ly a c cu ra te. N e v e r th e le s s, a t th e p re se n t tim e , th e flu c t u a tio n m e th o d
rem ain s th e o n ly g en eral m e a n s o f d e te r m in in g all th ree q u a n titie s x , q , and / d ir e c tly .
F o r m easu rin g th e fin e d e ta ils o f th e w in d -v e lo c ity f lu c t u a tio n s m' and w \ u n til 1 9 5 8 ,
m o s t researchers used th e h o t-w ir e a n e m o m e t e r , w h ic h h a d ahready b e e n s u c c e s sfu lly used
for m a n y y ea rs in th e m e a s u r e m e n t o f v e lo c ity flu c t u a tio n s in w in d -tu n n e l flo w s . G O edecke
( 1 9 3 5 ) w as o n e o f th e first to s u c c e e d in m e a su rin g in th e a tm o sp h e r e using th e h o t-w ir e
te c h n iq u e . L ater, fr o m 1 9 4 7 t o 1 9 4 9 , s u ita b le in s tr u m e n ts o f th is ty p e w ere d e v e lo p e d in
th e U .S .S .R . b y O b u k h o v an d K rech m er [s e e , fo r e x a m p le , K r ech m er ( 1 9 5 4 ) ] . A t a b o u t th e
sam e tim e , h o t-w ir e a n e m o m e te r s also b eg a n to b e u sed fo r a tm o sp h e r ic research in
A ustralia and G reat B ritain [s e e , fo r e x a m p le , S w in b a n k ( 1 9 5 1 a ; 1 9 5 5 ) ; D e a c o n ( 1 9 5 5 ) ;
M cllro y ( 1 9 5 5 ) ; J o n e s and P asqu ill ( 1 9 5 5 ) ] . A p articularly sen sitive h o t-w ir e a n e m o m e te r
for a tm o sp h e r ic m e a s u r e m e n ts w as d e v e lo p e d b y M cC ready ( 1 9 5 3 ) . A c o n sid e ra b le n u m b e r
o f w o r k s o n m e a su r em en ts by a tm o sp h e r ic h o t-w ir e a n e m o m e te r s w as p u b h s h e d in th e la te
1 9 5 0 ’s and in the 1 9 6 0 ’s; h o w e v e r , w e shall n o t e v en a tt e m p t to m e n t io n all o f th em here.
T h e g eneral p rincip les o f h o t-w ir e a n e m o m e t r y are w e ll k n o w n , fo r e x a m p le , in w o r k s b y
Corrsin ( 1 9 6 3 ) and K ova szn a y ( 1 9 6 6 ) , and th er e fo r e w ill n o t b e r ep ea ted here in d etail.
T h e sen so rs o f all h o t-w ir e a n e m o m e t e r s are th in m e ta l w ires w h ic h are h e a te d b y a
current to a te m p er a tu re o f several h u n d r e d d eg r ee s du rin g th e m e a s u r e m e n ts. T h e
te m p er a tu re variation s o f th e w ir e, a n d c o n s e q u e n tly th e v ariation s o f its resista n c e , are
d e te r m in e d b y th e h e a t transfer t o th e air. T h is h e a t transfer d e p e n d s o n th e f lo w v e lo c ity
c o m p o n e n t n o r m a l to th e w ir e, an d is p r a c tically in d e p e n d e n t o f sm a ll air-tem p eratu re
v a riatio n s, w h ic h are g e n erally o f th e order o f fr a c tio n s o f a d e g ree. B y m easu rin g th e
flu c t u a tio n s o f th e w ire r e s is ta n c e , th a t is, th e cu rren t th ro u g h th e w ir e, th e flu c t u a tio n s o f
th e w in d -v e lo c ity c o m p o n e n t n o r m a l to th e w ir e m a y b e d e te r m in e d . W hen th e w ire o f th e
h o t-w ir e a n e m o m e t e r is p la c e d v e r tic a lly , th e current flu c t u a tio n s c o r r e sp o n d t o flu c tu a
tio n s o f th e h o r iz o n ta l w in d v e lo c ity . I f w e n o w set up t w o w ires in th e vertical plan e
passing th ro u g h th e d ir e c tio n o f th e m e a n w in d , w h ic h is fo u n d fr o m an in ertial w in d vane,
and in c lin e th em at an g les o f + 4 5 and - 4 5 t o th e vertical, th e n th e d iffe r e n c e in th e
resista n ces o f th ese w ires w ill b e p r o p o r tio n a l t o th e vertica l c o m p o n e n t o f th e w in d
v e lo c it y , w ith th e p r o p o r tio n a lity c o e f f ic ie n t d e p e n d e n t o n th e m e a n w in d v e lo c ity .
H ot-w ire a n e m o m e te r s fa c ilita te th e rec o r d in g o f f lu c t u a tio n s o f w in d v e lo c ity w ith an
a c cu ra c y o f up to 1 c m /s e c , w ith a tim e-lag ( t im e -c o n s ta n t) o f th e order 0 .0 1 sec;
in s tr u m e n ts o f th is ty p e havin g sm aller tim e-lag a n d w h ic h are c o n s id e r a b ly m o r e s en sitiv e,
are also p o ssib le .
T h e n o n lin e a r ity o f th e d e p e n d e n c e o f th e resista n c e and th e c u r ren t o f th e w ire o n th e
w in d v e lo c ity is a great d isad v a n ta g e . F o r e x a m p le , w h e n a n a n a lo g c o m p u te r is used in ste a d
o f a d igital c o m p u te r , th e c a lc u la tio n s o f th e m e a n values o f th e flu c t u a tio n p r o d u cts
b e c o m e very c o m p h c a t e d . O n e m e t h o d o f lin ea r iz in g a h o t-w ir e a n e m o m e te r is to pass th e
cu rren t flu c t u a tio n s th ro u g h an a d d itio n a l n o n lin e a r d e v ic e h avin g a c h a r a cteristic w h ic h
c o r r e sp o n d s t o th e a n e m o m e t e r ca lib r a tio n curves in su ch a w a y th a t th e o u t p u t signal w ill
b e p r o p o r tio n a l to th e m easu r ed v e lo c ity flu c t u a tio n s [M cC read y ( 1 9 5 3 ) , R. J. T a y lo r ( 1 9 5 8 ) ,
D y e r an d M aher ( 1 9 6 5 a ) ] . T o a v o id su ch c o m p lic a t io n s , B o v sh ev e r o v an d G urvich [G u rvich
(1 9 5 9 ) ; B o v sh ev e r o v and V o r o n o v ( I 9 6 0 ) ] c o n s tr u c te d a linear in s tr u m e n t at th e M o sc o w
I n stitu te o f A tm o s p h e r ic P h y sic s in 1 9 5 8 for m easu rin g fln e-sca le v e lo c ity flu c tu a tio n s.
T heir in s tr u m e n t, an u ltra so n ic m ic r o a n e m o m e te r , w a s b a sed o n th e d e p e n d e n c e o f th e
transit tim e o f s o u n d fr o m tr a n sm itter to m ic r o p h o n e o n th e v e lo c ity o f air in its p a th .
I n stru m e n ts o f th e sam e ty p e b u t less p recise w e re
also d e sig n e d in th e U .S . a n d Japan b y B usinger an d
S u o m i. [S e e L etta u and D a v id so n ( 1 9 5 7 ) , Barad Si
(1 9 5 8 ), K
r v aim
d i i i i al
u i and
a iiu B
u uusinger
a i i i g c i ( 1 9 6 3 ) , jK
vaaim
i i i i ial,
u , C ram er - t -----------j -
et al. ( 1 9 6 4 ) , and K aim al, W yngaard and H au gen | '

( 1 9 6 8 ) .] T h e m o s t im p o r ta n t parts o f th e B ovsh ev - 2 .5 c m |
erov-G urvich in s tr u m e n t are th e m in ia tu r e c y lin d ric a l I •
s o u n d tran sm itters and th e m ic r o p h o n e s , w h ic h are 2 * »
m m in d ia m eter and a b o u t 5 m m lo n g . T h e in s tr u m e n t Ri
has tw o tran sm itters (s o u r c e s ) S j and S 2 , an d tw o
receivers (m ic r o p h o n e s ) R i and R 2 ; th e pair S i and FIG. 69. S c h e m e o f th e arrange-
R 2 is 2 .5 cm fr o m th e pair R j an d S 2 (se e Fig. 6 9 ) . m e n t o f e m itte r s an d receivers
T h e s o u n d , th at is, th e u ltra s o u n d , is p r o p a g a ted a lo n g in th e a c o u s tic m ic r o a n e m o m e te r ,
p r actica lly id e n tic a l pa th s from S i to R i and fr o m S 2
to R 2 in the o p p o s it e d ir e c tio n . I f th e tr a n sm itter signals have id e n tic a l p h ases, th e phase
d iffe r e n c e o f th e m ic r o p h o n e signals w ill b e p r o p o r tio n a l t o th e w in d -v e lo c ity c o m p o n e n t in
th e d ir e c tio n o f th e base o f th e in s tr u m e n t and a lm o s t in d e p e n d e n t o f th e air te m p er a tu re .
T his in s tr u m e n t also e n a b le s th e v e lo c ity flu c tu a tio n s to b e m e a su red w ith an accu ra c y o f
up to 1 c m /s e c , w ith a tim e lag o f less th an 0 . 0 0 1 sec.
In a d d itio n to h o t-w ir e a n e m o m e te r s a nd a c o u s tic a n e m o m e t e r s , s o m e tim e s propeller

a n e m o m e te r s and sp ecial m ic r o v a n e s, c o n s tr u c te d o n th e p rin cip le o f an ordinary w in d vane,
are used to m easure fin e-sca le flu c t u a tio n s o f w in d v e lo c ity [se e, e .g ., K aim al, Cram er e t al.
( 1 9 6 4 ) and D y e r , H ick s and King ( 1 9 4 7 ) ) . M icrovanes are e sp ec ia lly usefu l for recordin g
flu c tu a tio n s o f th e w in d d ir e c tio n w h ic h , to g e th e r w ith data o b ta in e d b y th e h ot-w ir e
a n e m o m e te r , p erm it in s ta n ta n e o u s values o f all th ree c o m p o n e n ts o f v e lo c ity u \ v\ and w '
to b e d e te r m in e d (se e, fo r e x a m p le , C ram er and R e c o r d ( 1 9 5 3 ) , and L etta u and D avid son
( 1 9 5 7 ) S e c t. 5 . 2 ) . H o w e v e r , th is ty p e o f app aratus has greater tim e lag th an th e in str u m en ts
d isc u sse d a b o v e , and th e p o s sib ility o f a u to m a tin g th e m e a su r em en ts and th eir s u b s e q u e n t
pr o c essin g is sm aller.
M easu rem en t of th e te m p er a tu re flu c tu a tio n s V m a y b e m ad e using resistan ce-
th e r m o m e te r in s tr u m e n ts, w ith c o n d u c t o r or s e m ic o n d u c to r sen sors, or w ith th e aid o f a
m in iatu re th e r m o c o u p le . A ty p ic a l e x a m p le o f a r esistan ce th e r m o m e te r fo r m easuring T ' is
K r e ch m e r ’s m ic r o th e r m o m e te r ( 1 9 5 4 ) or sim ilar A ustralian in str u m en ts [M c llr o y ( 1 9 5 5 ) ,
D y e r ( 1 9 6 1 ) , D y e r an d M aher ( 1 9 6 5 a ) ] . W hen th e resistan ce e le m e n t is used to m easu re
tem p er a tu re f lu c t u a tio n s , a very w e a k current is p assed th ro u g h it, and th e o v e rh ea t relative
to th e surro u n d in g air d o e s n o t e x c e e d 0 . 0 1 ° . U n d e r th ese c o n d it io n s th er e is pra ctica lly n o
h e a t transfer in t o th e air, and th e w ire te m p er a tu re , and c o n s e q u e n tly th e w ire resistan ce,
d e p e n d o n ly on th e air-tem p eratu re flu c t u a tio n s b u t are in d e p e n d e n t o f th e w in d -v e lo c ity
flu c t u a tio n s . A s a result, th e flu c tu a tio n s o f th e current passing th ro u g h th e resistance
c le m e n t are p r o p o r tio n a l to th e air-tem p era tu re flu c tu a tio n s, w ith th e p r o p o r tio n a lity
c o e f fic ie n t d e te r m in e d o n ly b y th e param eters o f th e in s tr u m e n t.
T o m easure th e h u m id it y flu c t u a tio n s d ', w e t - a n d dry-b ulb r esistan ce th e r m o m e te r s , or
w e t an d dry th e r m o c o u p le s , m a y b e used. T h e y give th e in s ta n ta n e o u s v a lu es o f th e tw o
te m p eratu res T and T \ , fr o m w h ic h 0 ' is d e te r m in e d w ith the aid o f th e “ p s y c h r o m e tr ic
e q u a t io n ” [s e e , for e x a m p le , M id d leto n and S p ilh au s ( 1 9 5 3 ) , or K ed ro liv a n sk iy and
S ter n z a t ( 1 9 5 3 ) ] . T h e n T{ m ay b e fo u n d n u m e r ic a lly fr o m th e values o f 7*'and T{ [M c llr o y
( 1 9 5 5 ) , or b y fe e d in g th e flu c t u a tio n s T'sind 0 ' in t o a sp ecia l electric a l a n alo g d e v ic e w h ic h
a u to m a tic a lly carries o u t an o p e r a tio n e q u iv a len t to s o lv in g th e p sy c h r o m e tr ic e q u a tio n
[S w in b a n k ( 1 9 5 1 b ) ; R. J. T a y lo r ( 1 9 5 6 b ) ; T a y lo r and D ye r ( 1 9 5 8 ) ; D y e r an d M aher (1 9 6 5 a ) ;
cm/sec
60
BO
40>
...- -
* .*• T*
* • •
•• -20- • • •
-40'- . • •
'-60-
• •
• •
-80-
F IG . 7 0 . E x a m p le o f a c o r r e la tio n graph for th e

flu c tu a tio n s o f te m p er a tu re a n d vertical v e lo c ity .
Polavarapu and M unn ( 1 9 6 7 ) ] . In eith e r case, e stim a tin g th e accu racy o f th e o b ta in e d values
o f 0 ' a n d , e sp e c ia lly , / = p w 'd ', is con sid e ra b ly m o r e d if fic u lt th a n d e te r m in in g T and
[se e R. J. T ay lo r ( 1 9 6 3 ) 1 ; c o n s e q u e n tly , the p o ssib ility o f using th e flu c t u a tio n m e th o d to
o b ta in c o m p le te ly r eh ab le values o f j still has a bright fu tu r e . P rospects in this d ir e c tio n are
en h a n c e d b y su b s titu tin g fo r th e tw o th e r m o m e te r s , an o p tic a l infrared h y g r o m e te r w h ic h is
p r actically free o f tim e lag; th is m e th o d o f h y g r o m e tr y is based o n th e d e p e n d e n c e o f th e
air-refraction c o e f fic ie n t o n th e w ater-vapor c o n t e n t [s e e E lagina ( 1 9 6 2 ) ] . H o w ev er, th e
infrared h y g r o m e te r has not yet been d e v e lo p e d to a stage w h ere it can b e used
sy ste m a tic a lly un der fie ld c o n d it io n s .
In p rin c ip le, flu c t u a tio n m e a s u r em en ts p erm it sim u lta n e o u s values o f ii' and w \ or o f oi*'
and T ' or w ' and 0 ' to b e d e te r m in e d for a series o f su cce ssiv e in sta n ts. T o illu strate the
results o b ta in e d w ith th is m e t h o d , Fig. 7 0 [w h ich is b o r r o w e d from M onin ( 1 9 5 3 ) ] s h o w s a
c o rr e la tio n graph w h ic h ch aracterizes th e d e p e n d e n c e b e tw e e n th e flu c tu a tio n s u:*' and T'
from m e a s u r e m e n ts at a h e ig h t o f 1.5 m under in version c o n d itio n s . T h e graph c o rr esp on d s
t o a c o rr e la tio n c o e f f ic ie n t b e tw e e n th e values o f c:' and T ' e q u a l to - 0 . 2 4 ; the
tu r b u le n t h e a t flu x is e q u a l here to q = - 0 . 0 3 c a l/c m ^ m in . F igure 7 1 , also tak en from
M o n in , sh o w s a c o rr e la tio n graph w h ic h characterizes th e relatio n b e tw e e n th e flu c tu a tio n s
ii' a n d u}' fro m m e a su r em en ts also at a h e ig h t o f 1.5 m . Here th e corre la tio n c o e f fic ie n t r,,,,-
is e q u a l to - 0 . 6 5 and = 0 .3 3 m /s e c .
w,'m/sec
• • •
.• •
•
< •• • •
•
0.2 * *^
•
0 •• ^^••
% ’ .* • i/m /se c
...
- 0.2 •
•
-OA
••• . •
-06
•
-O.d • •
-1 0 __ •
-1 1 •
FIG. 7 1 . E x a m p le o f a c o rr e la tio n graph for th e flu ctu a-
tio n s o f th e h o r iz o n ta l a n d vertical c o m p o n e n ts o f the
w in d v e lo c ity .
O f c o u r se , th e n u m e ric a l c a lc u la tio n o f th e m e a n values o f th e flu c t u a tio n p r o d u cts fr o m

traces o f the ty p e s h o w n in F ig. 1 is very la b o r io u s and te d io u s . H o w e v e r , i f linear
in s tr u m e n ts are u se d , or in str u m en ts m a d e a lm o s t linear b y a sp ecial lin ea rizin g circu it, th en
th e m ean p r o d u cts (or squares) o f th e flu c t u a tio n s m ay b e o b ta in e d a u to m a tic a lly b y a
special c o rr e lo m e te r . T h is is an analog in teg r a tin g d e vic e w h ic h c a lc u la te s th e integral o f the
p r o d u c t o f th e tw o in p u t signals. S u ch a u t o m a tic co rre lo m e te r s w e re d e v e lo p e d in th e
U .S .S .R . b y B o v sh ev e r o v , G u rvich , T atarsk iy and T svang ( 1 9 5 9 ) ; see also B ov sh ev e r o v ,
G u rv ich , M o rd u k h o v ich and T svang ( 1 9 6 2 ) . T h e y w e re used to d e te r m in e th e tu r b u le n t h e a t
flu x [B o v sh e v er o v , G u rv ic h , and T svang ( 1 9 5 9 ) ] . In A u stralia a sim ilar app a ratus w as
d e v e lo p e d b y T a y lo r and W ebb ( 1 9 5 5 ) and D y e r ( 1 9 5 8 ) . [S ee also D y e r and M aher ( 1 9 6 5 a ,
b ) and D y e r, H ick s, and King ( 1 9 6 7 ) .] F lu c tu a tio n s m ay also b e r ec o r d e d o n analog
m a g n e tic ta p e, th en c o n v e r te d to digital fo r m a n d , fin a lly , c o m p u te d b y digital c o m p u te r
[c f. K aim al, H augen and N e w m a n (1 9 6 6 )]. B usin ger, M iyak e e t al. (1 9 6 7 ) m ade a
c o m p a r is o n o f d iffe r e n t h e a t- flu x m e a s u r e m e n t m e t h o d s , and s h o w e d th a t tlie h o r iz o n ta l

variability o f tu r b u le n t h e a t f lu x is q u ite large e v e n fo r h o m o g e n e o u s u n d e rly in g su rfaces.
In a d d itio n to th e flu c t u a tio n m eth o d fo r m e a su rin g tu r b u le n t shear stress, th e
d y n a m o m e tr ic m e th o d m a y a lso b e used . This, w as first p r o p o s e d b y S h ep p ard ( 1 9 4 7 ) and
th e n u s e d b y P asquill ( 1 9 5 0 ) , R ider ( 1 9 5 4 ) , several A m e r ica n researchers [se e L e tta u and
D a v id so n ( 1 9 5 7 ) , S e c t. 3 .2 , and th e r efer e n c e s in B rad ley ( 1 9 6 8 ) , G urvich ( 1 9 6 1 ) , and
B rad ley ( 1 9 6 8 ) ] . T h is m e th o d c o n s ists in th e u se o f a b lo c k o f soil, w ith a grass cover
c h a r a c t e r is tic .o f th e lo c a lity . T his b lo c k is s e t flu sh w ith th e surface o f th e g r o u n d o n a
m o v in g p la tfo r m ; su sp e n d e d b y w ires in th e e x p e r im e n ts o f G urvich and B rad ley, and
flo a tin g o n flu id in th e e x p e r im e n ts o f o th e r researchers; this p la tfo r m is h e ld in p lace b y a
sp ecia l spring. T h e w in d drag o n th e p la tfo r m is d e te r m in e d b y m e a su rin g e ith e r th e te n s io n
in th e spring, as d o n e b y S h ep p a r d , Pasqu ill e t al., or th e d is p la c e m e n t o f th e p la tfo r m . T h e
in str u m en ts u se d to m ea su re x an d b y th is m e th o d are v ery c a p r ic io u s, and th eir use
requires great c a u tio n . H o w e v e r , th e a c cu racy o f th e b e st o f th ese in str u m en ts is c o m p a r a b le
to th a t o f th e b e st e x istin g flu c t u a tio n in s tr u m e n ts d e sig n e d fo r th e sam e p u rp o se .
In th e fie ld o f m e t e o r o lo g y , th e tu r b u le n t w a te r vapor f lu x (o r th e e v a p o r a tio n r a t e ) / is
m o s t o f t e n m e a su red b y m e a n s o f various t y p e s o f w e ig h in g ly sim e te r s (so il e v a p o r a to r s). In
th e s e in s tr u m e n ts, a large b lo c k o f s oil is c u t, w ith care ta k en to preserve its str u c tu r e . T h e
h o le m a d e w h e n this b lo c k is r e m o v e d , is d e e p e n e d an d sp e cia l scales are p la c e d in sid e . T h e
b lo c k o f soil is th e n retu r n e d to its origina l p o s itio n and w e ig h e d at regular intervals. In
c o m p a r is o n to th e s e su p erior ly s im e te r s, th e use o f o th e r ty p e s req uires a sp e cia l transfer o f
th e b lo c k o f so il for e a c h w e ig h in g . In e ith e r c ase, th e rate o f e v a p o r a tio n o f m o is tu r e fr o m
th e b lo c k ca n b e d e te r m in e d ; c f., for e x a m p le , M cllro y and A n g u s ( 1 9 6 3 ) , and C raw ford
( 1 9 6 5 ) . U n fo r tu n a te ly , it is still d if fic u lt t o d e te r m in e th e e x t e n t t o w h ic h th e rate o f
e v a p o r a tio n o f an e x c is e d b lo c k c o rr e sp o n d s to th e e v a p o r a tio n rate o f th e u n d istu r b e d so il.
A s a r esu lt, s o m e researchers are in c lin e d t o th in k th a t all th e data o b ta in e d fr o m w e ig h in g
ly s im e te r s are q u ite q u e s tio n a b le . T o r efu te th is o p in io n , it is n e cessary fo r c o m p a r is o n to
h ave a b u n d a n t m ater ia l fr o m s im u lta n e o u s m e a s u r e m e n ts o f th e rate o f e v a p o r a tio n using
ly s im e te r s an d t h e flu c t u a tio n m e t h o d . T h e flrst a tt e m p ts o f this ty p e w e re m a d e b y
S w in b a n k and D y e r ( 1 9 6 7 ; 1 9 6 8 ) . S o far, h o w e v e r , c o m p le te ly reliable m ateria l o f this
n atu re d o e s n o t e x is t . T h is is b e c a u s e th e p r e se n t flu c t u a tio n m e a s u r e m e n ts o f th e
e v a p o r a tio n rate are n e ith e r w e ll d e v e lo p e d n o r p articularly accu ra te.
T o d e te r m in e th e tu r b u le n t h e a t flu x q an d e v a p o r a tio n rate y, w e m ay also u se th e
energy b a la n ce e q u a t io n , i.e ., th e h e a t b u d g e t e q u a t io n , o f th e u n d e rly in g surface w h ic h has
th e fo r m
B = q^rQ -\-m j, ( 8 .1 6 )
H ere B is th e r ad iation b a la n ce o f th e surface o f th e ea rth , s p e c if ic a lly , th e su m o f th e

in c id e n t lo n g - and sh o r t-w a v e r a d ia tio n , fr o m w h ic h th e o u tg o in g r a d iation is s u b tr a c te d , Q
is th e h e a t flu x in th e s o il, and m is th e la te n t h e a t o f v a p o r iz a tio n . In certain cases o f very
dry s o il, fo r e x a m p le , in a d e ser t, th e last term o n th e right sid e m a y b e a s su m ed to b e sm all
in c o m p a r is o n w ith th e o th e r t w o ter m s. T h e n , m easu rin g B and 0 w e can c a lc u la te q.
H o w e v e r , w h e n q is m e a su red b y th e flu c t u a tio n m e t h o d , E q. ( 8 . 1 6 ) an d th e data from
m e a s u r e m e n ts o ^ q . B, and Q, can b e used t o fin d th e e v a p o r a tio n r a t e / . I f q is u n k n o w n , b u t
th e values o f / o b t a in e d b y using ly s im e te r s are c o n sid e re d s u ffic ie n t ly r eliab le , th e n Eq.
( 8 . 1 6 ) m a y b e u se d t o fin d th e value o f q\ sec R ider ( 1 9 5 4 ) . O n th e o th e r h a n d , i i q and j arc
m easu red d ir e c tly , th e n E q . ( 8 . 1 6 ) m a y b e used as an a d d itio n a l v e rific a tio n o f th e data
o b t a in e d [D y e r ( 1 9 6 1 ) ; D y e r and Maher ( 1 9 6 5 a , b ) ] . F in a lly , i f n e ith er ^7 nor / is m easured
d ir e c tly b u t w e have d a ta o n th e te m p e r a tu r e an d h u m id it y p r o files, th e n w e m a y use the
a s su m p tio n th a t th e e d d y d iffu s iv itie s fo r h e a t and m o is tu re are eq u a l. It f o llo w s , th er e fo r e,
th a t th e so-c a lle d Bowen ra tio y = q / m j \ a d im e n sio n le ss q u a n tit y , satisfie s th e e q u a t io n
T= mj = m d , — d o (8.17)
w h ere T\ and To are th e m ean tem p er a tu re s at t w o f ix e d h e ig h ts Zi and Zo, a n d an d do

are th e m ean s p e c ific h u m id itie s at th e sam e h e ig h ts. F r o m E qs. ( 8 . 1 6 ) a nd ( 8 . 1 7 ) it fo llo w s
th at
, = (8 .1 8 )
th ese last e q u a tio n s all^ w us t o ca lc u la te th e values o f q and / fr o m th e m easu r ed values o f

B, Q, Ti, fo , , an d do- Sim ilar e q u a t io n s m a y b e d e d u c e d o n th e basis o f certain o th e r
p o s tu la te s c o n c e r n in g th e r e la tio n b e t w e e n th e d if fe r e n t e d d y d iffu s iv itie s [s e e , for e x a m p le ,
L etta u and D a v id so n ( 1 9 5 7 ) , S e c t . 7 . 3 ] . It m u st b e n o t e d , h o w e v e r , th a t Q an d — do can
o n ly b e m e a su r ed w ith very large error; th e r e fo r e th e a c cu ra c y o f th e e n e rg y b a la n ce
m e th o d is c o n sid e ra b ly le ss, fo r e x a m p le , th an th a t o f th e flu c t u a tio n m e th o d .
8.4 Determination of Turbulent Fluxes from Data

on the Profiles of Meteorological Variables
A b o v e w e gave several m e th o d s o f m ea su r in g u , , q, a n d / . H o w e v e r , all th e s e m e th o d s
are fairly c o m p lic a t e d a n d req uire sp e cia l e q u ip m e n t, w h ic h is o n ly availab le at p r e se n t in a
f e w research o r g a n iz a tio n s. A t th e sam e tim e , it w o u ld b e o f in ter e st t o m a k e a s y s te m a tic
d e te r m in a tio n o f tu r b u le n t flu x e s a t a large n u m b e r o f p o in ts o n th e e a r th ’s surfa ce. It is
q u ite u s e fu l th er e fo r e to give m e th o d s of d e te r m in in g m*, q, and / fr o m sim pler
m e a s u r e m e n ts, prim arily fr o m th e o rdinary m e a s u r e m e n ts o f th e p r o files o f th e m e te o r o
lo g ic a l variables. T h e p r o file m e a s u r e m e n ts are o f t e n called g r a d ien t m e a s u r e m e n ts b e c a u se
th e y d iffe r fr o m th e standard m e te o r o lo g ic a l m e a s u r e m e n ts a t a single f ix e d h e i ^ t . In th ese
m e a s u r e m e n ts, w e also d e te r m in e th e values o f th e g r a d ien ts o f th e m e te o r o lo g ic a l variables.
T h e r e fo r e , th e m e t h o d o f d e te r m in in g flu x e s fr o m p r o files is o f t e n called “th e m e th o d o f
in ter p r e tin g grad ien t m e a s u r e m e n ts ” or “ th e e s t a b lis h m e n t o f th e flu x -g r a d ie n t r e la tio n .”
We shall n o t d w e ll he r e o n all th e o ld s p e cu la tiv e sem iem p ir ic a l fo r m u la s w h ic h e x p r ess
th e flu x e s in term s o f th e p r o files, b e c a u se fe w o f th e m have b e e n c o m p a r e d w it h th e fairly
r ec e n t d a ta fr o m d irect m e a s u r e m e n ts, an d th e y are very u n reliab le. I n ste a d , w e shall use th e
sim ilarity th e o r y d e v e lo p e d in S e c t. 7 for a tu r b u le n t r egim e in th e surface lay e r o f th e
a tm o sp h e r e . A c c o r d in g to th is t h e o r y , th e w in d -v e lo c ity p r o file and th e te m p er a tu re p r ofile
(o m ittin g , for n o w th e h u m id it y ) are d e te r m in e d b y th e general fo r m u la s ( 7 . 2 4 ) w h ic h
c o n ta in th e param eters q, Zq [o n th e latter o f w h ic h d e p e n d s th e m e a n v e lo c ity u ( z ) ] ,
f o = f (zo) [ f ( z ) d e p e n d s o n f o ] v o n Kdrm ^n’s universal c o n s ta n t x an d th e t w o universal
fu n c tio n s /(E ) and / i ( E ) . Here x is c lo se t o 0 .4 ; s o m e in f o r m a tio n o n / an d is given in
S e c ts . 7 , a n d 8 . 1 - 8 . 2 . F o r th e tim e b e in g , w e shall assu m e th a t th e s e fu n c tio n s are e x a c t ly
k n o w n . In th is case an y fo u r m e a s u r e m e n ts o f th e valu es o f th e v e lo c ity and te m p er a tu re
w ill p e r m it fo r m u la tio n o f fou r e q u a tio n s w h ic h are s u ffic ie n t in p r in c ip le, fo r th e
d e te r m in a tio n o f th e fo u r p aram eters w*, q, z q , and f o . T h e r eq u ired n u m b e r o f
m e a s u r e m e n ts m a y also b e r e d u c e d w it h o u t grea t d if f ic u lt y , fo r e x a m p le , b y n o t con sid e rin g
th e values f ( z ) a t all, b u t o n ly th e d iffe r e n c e s f ( 2 2 ) — f ( z \ ) w h ic h are in d e p e n d e n t o f f o .
H o w e v er , th is is c o m p lic a t e d b y th e fa c t th a t at p r e se n t, / an d U are k n o w n o n ly
a p p r o x im a te ly an d are n o t g iv en b y a ny sim p le a n a ly tic a l fo rm u las; data given b y d iffe r e n t
researchers are c o n tr a d ic to r y , and th e m e a s u r e m e n t results a lw a y s c o n ta in s o m e errors.
T h e r e fo r e , in p r a c tic e , a p p r o x im a te e x p r e s s io n s m u st b e used for / a n d fi, and d iffe r e n t
m e th o d s , using d iffe r e n t sets o f original d a ta , w ill lead to s o m e w h a t d if fe r e n t results.
T h e M o n in -O b u k h o v m e th o d in S e c t 8.1 o f treatin g th e m e a su r em en ts o f th e w in d and
te m p er a tu re p r o files in c lu d e d as a c o m p o n e n t part, th e d e te r m in a tio n o f th e values o f
and L (w h ic h also p e r m it us to fin d <7 ), in te r m s o f th e values o f Ti ( 2 ) and T{ z ) . In s p ite o f

th e fa c t th a t th is m e th o d w as b a se d o n th e use o f th e la ter-criticized “ lo g a r ith m ic + lin ea r ”
fo r m u la fo r th e p r o files an d o n th e a s su m p tio n th a t a z = 1, its a ccu ra c y proves to b e higher

than m ig h t b e e x p e c t e d . T his w a s s h o w n b y c o m p a r in g th e values o f w* an d o b ta in e d b y
this m e t h o d , w ith th e results o f d ir e c t m e a s u r e m e n t. H o w e v e r , it is im p o ss ib le n o w to
r e c o m m e n d th is m e th o d for d ir e c t practical use in its original fo r m , sin ce it is rather rough
and very c u m b e r s o m e ; in particular, it in c lu d e s t h e d e te r m in a t io n o f s o m e param eters b y
th e least squares m e th o d fr o m th e r esu lts o f several m e a s u r e m e n ts.
T a k e u c h i ( 1 9 6 1 ) an d S h io ta n i ( 1 9 6 2 ) p r o p o s e d c lo s e ly r elated m e th o d s o f d e te r m in in g
th e values o f w* an d q fr o m w in d - an d v e lo c ity -p r o file m e a s u r e m e n ts. T h e se m e th o d s also
use th e “ lo g a r ith m ic + lin ea r ” a p p r o x im a tio n o f th e universal f u n c tio n s , and th e p o s tu la te
th a t K t — K - S p e c ific a lly , T a k e u c h i gave p ractical r e c o m m e n d a tio n s for sim p lify in g th e
c a lc u la tio n , an d a lso c o m p a r e d th e c a lc u la te d valu es o f w* an d q w ith s o m e r esu lts o f d irect
m e a s u r e m e n t o f th e flu x e s . N e v e r th e le s s, th e m e th o d s o f fin d in g w* and q are still
u n ju stifia b ly c o m p lic a te d ; at th e sam e tim e , t h . y use o n ly a sm all part o f th e available
in fo r m a tio n o n / ( ^ ) .
M e th o d s o f d e te r m in in g th e tu rb u le n t flu x e s a cco r d in g t o p r o file m e a s u r e m e n ts w ere
a lso d isc u sse d b y m a n y research ers, fo r e x a m p le : K aza n sk iy and M on in ( 1 9 5 6 ; 1 9 5 8 ; 1 9 6 2 ) ,
P riestley ( 1 9 5 9 b ) , W ebb ( 1 9 6 0 ; 1 9 6 5 ) , P a n o fs k y , B lackadar, an d M cV e h il ( 1 9 6 0 ) , M onin
( 1 9 6 2 a ) , K o n d o ( 1 9 6 2 b ) , D e a c o n and W ebb ( 1 9 6 2 ) , P a n o fsk y ( 1 9 6 3 ) , K lu g ( 1 9 6 3 ; 1 9 6 7 ) ,
D y e r ( 1 9 6 5 ; 1 9 6 7 ) , B usin ger, M iyak e e t al. ( 1 9 6 7 ) , Z ilitin k e v ic h a nd C h a lik o v ( 1 9 6 8 b ) and
B u sch e t al. ( 1 9 6 8 ) .
F o llo w in g K azan sk iy and M o n in , a gen eral m e th o d is g iven here w h ic h rep re sen ts in a
su itab le fo r m th e d a ta o n th e universal f u n c tio n s fo r p ractical c a lc u la tio n o f th e tu rb u le n t
flu x e s ; first, h o w e v e r , a fe w gen eral rem ark s shall b e m ad e.
T h e in itial p r o b le m is t o s e le c t th e m o s t s u ita b le q u a n titie s fo r d e te r m in in g th e f lu x e s . In
p rin c ip le, th e values o f th e w in d v e lo c ity at th ree arbitrary h e ig h ts s h o u ld b e s u ffic ie n t to
fin d th e p aram eters ^7 , and Zq \ c o n s e q u e n tly , th e im p re s sio n m a y b e g iven th a t it is q u ite
su p e r flu o u s t o use te m p er a tu re o b se r v a tio n s . H o w e v e r , as P riestley ( 1 9 5 6 b ) s h o w e d , e v en
w ith th e use o f th e values o f th e w in d -v e lo c ity u (z) a t m o r e th an th ree h e ig h ts, z d o e s n o t
a llo w us t o d e te r m in e q r elia b ly . T his is b e c a u se th e values th u s o b ta in e d are q u ite d iffe r e n t
fr o m th e data o f d ir e c t m e a s u r e m e n ts and d iffe r c o n sid e ra b ly for d if fe r e n t m e th o d s o f
tr e a tm e n t. T h e r e fo r e , th e s u p p le m e n ta r y use o f f ( z ) w o u ld appear to h ave g r eat pra ctical
s ig n ifica n ce . T his d e d u c t io n in fa c t f o llo w s a lso fr o m P a n o fs k y , B lackadar and M cV e h il
( 1 9 6 0 ) , w h o e x p e r im e n te d w ith a m e th o d o f d e te r m in in g q fr o m the values o f u ( z ) a t th ree
h e ig h ts (an d th e value o f z q ) o n ly , b u t d id n o t o b ta in g o o d r esu lts. N e v er th ele s s, it is w o r th
n o t in g th at K lug ( 1 9 6 7 ) o b ta in e d sa tis fa c to r y a g r e e m e n t b e tw e e n th e data o f th e d ir e c t flu x
m e a s u r em en ts o f S w in b a n k ( 1 9 6 4 ) and th e values c o m p u te d from w in d o b s e r v a tio n s at six
h e ig h ts b y a rather c o m p lic a t e d le a st squares m e th o d based o n th e fo r m o f th e universal
w in d f u n c tio n / ( f ) p r o p o s e d b y P a n o fsk y ( 1 9 6 5 ) .
S in ce th e values o f and q are o f prim ary in te r e st, it m ig h t b e c o n sid e re d c o n v e n ie n t to
e lim in a te Zq at th e very b e g in n in g , and to c o n sid e r o n ly th e d iffe r e n c e s /7 (z 2 )— m ( 2 i),
igno rin g u (z) its e lf. H o w e v e r , in th is case it is ea sy t o see th a t th ere c o u ld b e a great loss in
ac c u r a c y , s in c e th e d iffe r e n c e s in w in d v e lo c ity are d e te r m in e d c o n sid e ra b ly less a c cu ra tely
th a n th e w in d its e lf, e s p e c ia lly i f th e h e ig h ts z j and Z2 are co m p a r a tiv e ly c lo s e to g eth er.
M oreov er, it m ay b e s h o w n th a t w h e n Zq is e lim in a te d , e v en a fairly sm all error in th e
v e lo c ity d iffe r e n c e s w ill p r o d u c e a g reat e ff e c t o n ti:^ an d q. S in ce Zq at a given p o in t and
tim e o f year is to b e d e te r m in e d o n ly o n c e (it d o e s n o t v a ry), an d sin c e it is n o t d if fic u lt to
fin d , it is m o r e c o n v e n ie n t to fin d Zq first, and th e n use this value th ro u g h o u t.^ A t th e sam e
tim e , To = T(zo) is n o t a c o n s ta n t and th er e fo r e m a y b e e lim in a te d ; th u s it is r eason ab le to
g
O f c o u r se , s o m e tim e s it is very d if fic u lt to d e te r m in e Zq, for e x a m p le , in th e m arine
a tm o sp h e r e . T h e r e fo r e , w ith this in m in d , Z ilitin k e v ic h and C h a lik o v ( 1 9 6 8 b ) gave
n o m o g ra m s for d e te r m in in g and q fr o m th e valu es o f bu = u{2) - ?7(0.5) an d 6 7’ = T{2)
- 7 ’( 0 . 5 ) , w h e r e th e n u m b e r s in d ic a te the h e ig h t in m ete r s.
use th e te m p er a tu re -p ro file data o n ly in th e fo r m o f th e d iffe r e n c e s f (Z2 ) — f ( 2 i ) . T h e n

the data o f o n ly tw o m e a su r em en ts are req uired to d e te r m in e and q.
C o n s e q u e n tly , it is r e c o m m e n d e d th a t, to fin d a n d q fro m d ata o n m (z) and f ( z ) ,
o n e s h o u ld begin b y d e te r m in in g th e ro u g h n ess p a ra m eter Zq, and th e n p r o c e e d fr o m th e
w in d v e lo c ity U at s o m e fix e d h e ig h t H and fr o m th e d iffe r e n c e 67* b e tw e e n th e m ean
te m p eratu res at tw o d if fe r e n t h e ig h ts, for e x a m p le , at 2 H an d H j l . A c c o r d in g to th e
sim ilarity h y p o th e s is fo r a tu r b u le n t regim e in th e surface layer o f th e a tm o sp h e r e , th e
q u a n titie s U = h ( H ) 5 T = r (2 //) — f ( y j m a y b e e x p r e s s e d in term s o f a * ,
- , and Zq. C o n s e q u e n tly , an d an d also L , w h ic h d e p e n d s o n th e m , m a y , in tu rn ,
'0
b e e x p r e s s e d in te rm s o f / / , X , U, I Ty an d z q . F r o m th ese five term s w e m ay fo r m u la te
tw o in d e p e n d e n t d im e n sio n le ss c o m b in a tio n s , w h ic h are w r itte n as
K -'r r v - '-.-i-
w h ere a is an arbitrary d im e n sio n le ss c o e f f ic ie n t ; w e shall say m o r e a b o u t th e s u itab le
c h o ic e o f a later. T h u s w e o b ta in
^ = = C,,). ^ = (8 .2 0 )
w h ere f , , F 2, F 3 are s o m e universal fu n c tio n s o f B w h ic h also d e p e n d o n I f graphs o f

th ese fu n c tio n s w e re available, th e y c o u ld b e u sed to d e te r m in e q, and L fr o m th e data
o f g rad ien t m e a su r em en ts. G raphs o f / = 1, 2 , 3, m a y b e p lo t te d eith e r e m p ir ic a lly using
th e data o f in d e p e n d e n t sim u lta n e o u s m e a s u r e m e n ts o f U, 6 T, an d q, or else th e o r e tic a lly
b y fix in g th e form o f th e universal fu n c tio n s /( ^ ) and / i ( S ) . T h e s e c o n d o f th ese m e th o d s
p erm its a progressive im p r o v e m e n t in th e a c cu racy o f d e te r m in in g flu x e s co m p a r e d to th e
a c cu ra cy o f ou r in f o r m a tio n o n th e universal fu n c tio n s . T h e r e fo r e , w e shall d iscu ss this
m e th o d in s o m e w h a t greater d e ta il.
A s o u r starting p o in t, w e tak e th e general fo r m u la ( 7 . 2 4 ) w h ic h w as used to d e te r m in e
th e universal fu n c tio n s f ( ^ ) and T h e n th e first e q u a t io n o f E qs. ( 8 . 1 9 ) and th e first
t w o e q u a tio n s o f E q. ( 8 . 2 0 ) m ay b e w r itte n as
D_ C 2 / F ^ ) - / , ( \/2F3)
F, ( / ( 17/^3) -/(C o//^3)I» ’
P _ (8 .21)
f O l F ^ ) - /(Co//â)"’
F —
2 a [ / (1/F ,) - / (Co//-'a)l [ / , {2/F,) - / , 0/2F,)] ’
T h e first o f th ese e q u a tio n s gives the form o f F 3 (B, ^ ) im p lic itly . I f th e fu n c tio n s /(C ) an d
/i(C ) are k n o w n , th en , for e x a m p le , this e q u a tio n m ay be so lv ed grap h ically fo r F 3 and the
tw o rem ain in g e q u a tio n s o f E q . ( 8 . 2 1 ) w ill give F i ( B , Co) and F 2 (B, Co) p a ra m etrically.
K aza n sk iy and M onin (1 9 6 2 ), and Z ilitin k e v ic h and C h alik ov (1 9 6 8 b ) a ssu m ed
th a t th e w in d -v e lo c ity p ro file and th e te m p eratu re profile are sim ilar, i.e., th at
1 K
/ , (C) = - / ( C ) , where a = = c o n s t. U n d e r th is a s su m p tio n , th e c o e f f ic ie n t o in E qs.
( 8 . 1 9 ) and ( 8 . 2 0 ) also m a y c o n v e n ie n tly b e ta k en e q u a l t o K t I K . T h e n E q . ( 8 . 2 1 ) ta k es th e

sim pler fo r m
a 1 fC2/F,) - / { l / 2 F , )
^ ^ (7 0
/=•.=
' /(l//= 'j)-/(C ./f> ) ■ ( 8 .2 1 ')
l/( l/f » ) - /( W /'.) lI /( 2 //'.) - /( l/ 2 f .) l •

In fa c t in th e w o r k s c ite d , an e v e n m o r e sp e cia l a s su m p tio n w as used , n a m e ly , th a t a = 1.

H o w e v er , r etain in g th e c o e f f ic ie n t a in ^ an d F 2 d o e s n o t ch a n g e th e c a lc u la tio n s c h e m e at
all; th u s n o sp ecial p u rp o se is served b y assu m in g a particular n u m erical v alu e. It is also
p o s sib le t o a ssu m e in th e n e x t a p p r o x im a tio n th a t a is d iffe r e n t, fo r e x a m p le , in neu tral,
u n sta b le , and stab le str a tific a tio n s; under th is a s s u m p tio n E q s. ( 8 . 2 1') can a lso b e u sed.
O n th e basis o f E q . ( 8 . 2 1 ' ) , K a z an sk iy and M o n in ( 1 9 6 2 ) , a nd Z ilitin k e v ic h a n d C h alik ov
( 1 9 6 8 b ) d r e w up n o m o g r a m s o f th e fu n c tio n s Fu F t , an d F^ b a se d o n tw o d iffe r e n t
a s su m p tio n s c o n c er n in g th e fo r m o f th e universal f u n c tio n s / ( f ) = / i ( f ) . F igu res 1 2 - 1 A give
th e Z ilitin k e v ic h -C h a lik o v n o m o g r a m s, w h ic h c o rr e sp o n d to th e fu n c t io n / ( f ) s a tisfy in g
E qs. ( 7 . 5 8 ) an d ( 8 . 8 ) w ith th e n u m e ric a l p aram eters r e c o m m e n d e d b y Z ilitin k e v ic h and
Clhalikov ( 1 9 6 8 a ) , and t w o s u p p le m e n ta r y a ssu m p tio n s: x = 0 .4 3 ; and a = 1. T h ese
n o m o g r a m s w ere check ed by Z ilitin k e v ic h an d C h a lik o v using e x te n s iv e d ata o f fie ld
m e a su r em en ts in th e U .S .S .R ., A u stralia an d th e U .S .; th e c o rr e sp o n d in g va lu es o f th e flu x e s
w ere also c o m p a r e d w ith results o b t a in e d fr o m s o m e o th e r a s su m p tio n s c o n c e r n in g th e fo r m
o f / ( f ) [see Z ilitin k e v ic h and C h a lik o v ( 1 9 6 8 b ) , a n d Z ilitin k e v ic h ( 1 9 7 0 ) ] .
O f c o u r se , th e n o m o g r a m fo r F 2 m a y n o t o n ly b e used t o d e te r m in e th e tu r b u le n t h e a t
flu x q, b u t also to d e te r m in e th e e v a p o r a tio n r a t e / , pr o v id e d th a t, in a d d itio n t o U an d bT,
th e d iffe r e n c e b e t w e e n th e s p e c if ic h u m id it ie s a t tw o h e ig h ts 8 d = d ( 2 / y ) —
m e a su r ed . A c c o r d in g to th e gen era l fo rm u la s (7 .2 4 ) and th e e m p ir ic a lly d isc o v er e d

sim ilarity b e t w e e n th e p r o files T ( z ) an d 0 ( ^ )
y/?o 1/(p?o
lU U = ( 8 .22)
w h e r e ai = /CJ /C (= a i f = Kx).
With th e aid o f E qs. ( 8 . 2 1') it is n o t d if fic u lt t o esta b lish th e fo r m o f t h e fu n c tio n ( 8 . 2 0 )
fo r nea r-neutral, very u n sta b le and very stab le s tr a tific a tio n s. F o r near-n eu tral str a tific a tio n
1^1 is sm a ll and IF^l is large w h ile in E q. ( 8 . 2 1') w e m a y a ssu m e th a t / ( £ ) « l n |C|. T h e n E qs.
( 8 . 2 1 ') b e c o m e c o n sid e ra b ly sim pler, and h ave e x p lic it s o lu tio n s o f th e fo r m
ln (l/C o)’ In 4 . I n ( 1 / C o) ’ B '
In th e case o f e x tr e m e in s ta b ility , B w ill b e large and n eg a tiv e and Fz sm all and n e g ative.
The a s y m p t o t ic b eh a v io r of th e fu n c tio n s (8 .2 0 ) is fo u n d here by in tro d u c in g
/ (C) « C2— C 2 IC I ^ in E qs. ( 8 . 2 1 ' ) . T h e n w e o b ta in
1
|B|“
A
QQ
QQ
_o
4)
lo E
''B
i
I I
I
I
I
00
I
B
a
CO
rs
CD
o .
s
V
QQ
OQ
•§
a
OQ
feO
II
I
s
I«
703)
E
I
B
o
F in a lly , fo r grea t s ta b ility , B w ill b e large an d p o sitiv e an d sm all an d p o sitiv e . T h e

a s y m p t o t ic b e h a v io r o f th e f u n c tio n ( 8 . 2 0 ) in th is case m a y b e d e te r m in e d b y in tr o d u c in g
f(0 - C3 + in E q. ( 8 .2 2 ') . T h e n w e fin d th a t B c a n n o t increase w it h o u t b o u n d , b u t
w ill te n d to th e lim it as - > 0 . T h u s fo r stab le str a tific a tio n , B c a n n o t

ZC3 (1 —Wo)
e x c e e d 3l2Cz. W hen is sm a ll, fo r Fi an d F 2 w e o b ta in
2 y? F \
( 8 .2 5 )
3 C ^ (1 -C ,)
T h e flu x -g r a d ie n t r e la tio n a ssu m es an e s p e c ia lly sim p le fo r m for great in s ta b ility . In this

ca se , i f all th e m e a s u r e m e n t h e ig h ts e x c e e d g iL , w h ic h is th e lo w e r b o u n d a r y o f th e
“ 1 /3 -p o w e r -la w ” layer (fo r th is it is s u ffic ie n t th a t H j2 ! > C iZ ,) , th e n to in terp ret th e
m e a s u r e m e n t results w e can use th e form u la s w h ic h are corr e c t in th e lim itin g case o f free
c o n v e c t io n . T h e c rite rio n / / / 2 > Ci^ w h ic h en su r es th is p o s sib ility m ay b e r ew r itten as
I^ 3 1< r)-| I . U sin g fo r m u la ( 8 . 2 4 ) fo r F 3 , w e m a y w r ite th is c riterion in th e fo r m

^ I <•! I
! ( 2 >— l ) | C, | * 0.01
\B \>
w h e r e, as in th e fo r m u la tio n o f th e n o m o g r a m , it is a ssu m ed th a t C 2 1 .2 5 an d lîl ^ 0 . 0 3 .

T h e value o f th is criterio n , in r o u n d figu res, varies fr o m 0 . 0 0 0 5 fo r £ 0 = 0 . 0 0 5 , to 0 . 0 0 4 for
So = 0 . 0 5 . T h u s th e rig h t sid e o f this in e q u a lity m a y b e w r itte n a p p r o x im a te ly in th e fo r m
0.1 £0 for this range o f ô. T h e p r o p o s e d c riterio n th e n red u c es t o th e fo r m
IT \ 3^0
//2 (8 .2 6 )
w h e r e Zq and H are m e a su r ed in m e te r s, U in m /s e c , and in °C .

I f th e c riterion ( 8 . 2 6 ) is sa tisfie d , th e n fo r / / / 2 , w e have a regim e o f free
c o n v e c t io n , w h e r e
dT (8 .2 7 )
C2 / dz
[see E q s. ( 7 . 3 6 ) and ( 7 . 3 9 ) ] . It is e asy to s h o w fr o m th is th a t h ere
1 1
~ “ -0 0 2
1
|57-|2
(
(8.28)
2 ^ —2 0 .
w h ic h m a y also b e o b ta in e d fr o m E qs. ( 8 . 2 0 ) a n d ( 8 . 2 4 ) . S u b stitu tin g th is value o f q in t o
E q. ( 7 . 3 8 ) fo r th e e d d y v is c o s ity K = K r at h e ig h t z o b t a in
3x2
K(H): hT
(8.29)
-1 i i 1
T h e n q an d K { H ) differ fr o m 1 5 7 ' | ^ and | 5 7 ^| 2 ^ r e s p e c tiv e ly , o n ly b y universal,
b u t n o t d im e n sio n le ss , m u ltip lier s. E x p r e s s in g / / in m e te r s, 6 7 in d eg rees, ^7 in cal/cm ^ m in ,
and K {H ) in m ^ /s e c , and p u ttin g x 0 . 4 3 , a_^_, ^ 1, C 2 ^ 1 .2 5 as r e c o m m e n d e d by
Z ilitin k e v ic h and C h a lik o v ( 1 9 6 8 a , b ) , an d T q ^ 3 0 0 ° K , w e o b ta in th e a p p ro x im a te
e q u a tio n s
A :(/o = o . i 2 w ^ /^ | 6 r | ' / * . (8 .3 0 )
T h e se fo r m u la s e n ab le q an d K {H ) to b e e s t im a te d very q u ic k ly fr o m tem p eratu re

d iffe r e n c e s in th e case o f u n sta b le str a tific a tio n . H o w e v er , th e n u m erica l c o e f fic ie n ts in th e
last e q u a t io n are n o t c o m p le te ly reliable at p r esen t. T h is is b e c a u se an u n ex p la in e d
d isc re p a n cy e x ists b e t w e e n th e variou s data; fo r e x a m p le , a c co r d in g to D y e r ( 1 9 6 7 ) , th e
± A
first o f E q s. ( 8 . 3 0 ) has th e fo r m q « » 0 .2 6 | ST" |^ .
8.5 Characteristics of Wind Velocity and Temperature

Fluctuations in the Atmospheric Surface Layer
The turbulent fluxes of momentum and heat are only special cases
of the statistical characteristics of wind velocity and temperature
fluctuations. Many other characteristics of these fluctuations have
been analyzed from the viewpoint of similarity theory in Sect. 7.5;
they may all be found experimentally with the same apparatus used
in the fluctuation method of determining u,, and q. If data exist
concerning simultaneous measurements o f «« and q and of any
statistical characteristic of the fluctuations, or if we have data on the
measurements of this statistical characteristic, and estimate «,vand q
by the method described in Sect. 8.4, the deduction of similarity
theory relating to this characteristic may be verified and the
corresponding universal function of C= -^ formulated empirically.
Thus the data o f fluctuation measurements permit us, in principle, to
determine all the functions ....... /, ,, f, and qp.v introduced in Sect.
7.5.
Data for moments of the turbulent fluctuations in the atmospheric
surface layer can be found in Swinbank (1955), Deacon (1955),
Perepelkina (1957; 1962), Lettau and Davidson (1957), Barad
(1958), Gurvich (1960), Tsvang (1960), Panofsky and McCormick
(1960), Monin (1962b), Zubkovskiy (1962; 1967), Lumley and
Panofsky (1964), Klug (1965), Panofsky and Prasad (1965), Mordu-
khovich and Tsvang (1966), Panofsky, Busch et al. (1967), Cramer
(1967) and in many other works. Unfortunately, all these data are
still fairly inexact, strongly scattered and do not agree very well with
each other; in addition, they are far from being complete. Both the
scatter and the disagreement among the data can possibly be
explained by the fact that in all cases, there is considerabre error in

determining the turbulent fluxes which have quite large variability,
both vertical and horizontal [cf., for example, Mordukhovich and
Tsvang (1966), and Businger, Miyake et al. (1967)], even above a
relatively very homogeneous surface. However, many o f the measure
ments were carried out above a clearly inhomogeneous terrain. In
every case the existing data allow us to draw only certain preUminary
conclusions about the statistical characteristics of the fluctuations. It
is important, however, that in all cases these conclusions agree fairly
well with the predictions of the similarity theory and give a definite
representation of the general form of certain universal functions.
Data on the dependence of the dimensionless standard deviations
of the velocity fluctuations av/«*>and on thermal
stratification, that is, on the universal functions /a ,A , and /s are
given in Swinbank (1955), Deacon (1955), Gurvich (1960), Panofsky
and McCormick (1960), and in all the above-mentioned works later
than 1960. For example. Fig. 75 shows the functions A(t), and
/s(f) obtained by Cramer (1967) from observations at Round
Hill, Mass., at heights of 16 m (circles) and 40 m (triangles)
for averaging time of 1.2 min; such comparatively small averaging
times may be advantageous for eliminating the low-frequency
fluctuations which are dependent on large-scale features of the
terrain and do not follow similarity theory. A survey of other data
on and ay^,/u^ can be found in Panofsky and Prasad
(1965), and Panofsky, Busch et al. (1967). The graph o f the function
fsCt) in Fig. 75a shows almost no dependence on the stability
parameter f although may increase somewhat with in
creasing instability. More definite indications concerning the slight
monotonic increase of with decrease of f along both semi
axes are seen on the plots given by Panofsky, Busch et al. The
values of / 3 (G) a n d / 4 (G), corresponding to the values of the ratios
ou/u^ and Oy/u^ at neutral stratification, can be determined from
atmospheric observations only with relatively low precision, but in
general they agree quite well with the estimates = A ,^ 2.3 and
=A 2 ^ 1.7 obtained from measurements in the logarithmic layer
of laboratory flows [see the end o f Sect. 5.3] .The relationship/ 3(f) =
•A(f) ~ IfI* predicted by similarity theory for the “ free convection
regime,” or for large negative values of f, is neither confirmed nor
unconfirmed by the existing data.
Since there is no doubt that the large-scale features are unimpor
tant, the data on /s(f) are more numerous and more reliable than
-0 .8 -0 .6 -0 .4 -0 .2 0 0 .2 0 .4 0 .6 0 .8 1 .0
^=Z/L
b
FIG. 7 5 . D a ta o f H. E . C ram er ( 1 9 6 7 ) o n th e universal fu n c tio n s : a - o „ / u ^ = / 3 ( f ) ; b -

o iv /« * = / 4 ( f ) ; c - o w / u , = / s ( f ) .
those on / 3(f) and / 4 ( f ) - In addition to Cramer’s figure (Fig. 75c), we

give here also the summary figure (Fig. C) of Panofsky, Busch et al.
(1967), where observations from many sites are collected. The
eq uations/s(f)= 1.25 [.^(f)- 1.8 and /;(?)= 1.25 [I
(where c^(f) is taken as recommended by Panofsky, Busch et al.) of
the forms (7.96) and (7.95) are also plotted in Fig. C; they turn out
to be almost indistinguishable from each other over the whole range
of values of f in Fig. C, and agree with the data quite well. The
problem of the exact value of /s(0) according to observations in the
atmospheric surface layer has stimulated considerable discussion. The
data of the comparatively old Russian observations of Gurvich
---- MEAN OF POINTS

-----FROM THE EQUATION
1.25 [(jp (f)-r.S r]‘/3
•-y^FROM THE EQUATION
__ /.25[r-f/(p(r)]i/4
t)
-L .
-1 .0 - 0 .7 -0 .5 - 0 .4 -0 .3 -0 .2 - 0 .1 0 0.1 0.2
UNSTABLE f STABLE
FIG . C. T h e ratio a ^ j u ^ as a f u n c tio n o f f a c c o r d in g to m a n y sou r ce s. T h e s eg m en ts repre
sen t th e standard d e v ia tio n o f th e m e a n s, assu m in g ail o b se rv a tio n s are in d e p e n d e n t o f each
o th er .
(I960) [see also Perepelkina (1957)] gave the estimate /s( 0 ) = / l 3

0 . 7 - 0 .8 , which is slightly below the laboratory estimate 0.9 of
Sect. 5.3, but agrees very well with the measurements o f Cermak,
Sandborn et al. (1966), and Cermak and Chuang (1967), in a
meteorological wind-tunnel at Colorado State University in the U.S.
However, most of the other investigators [for example, Panofsky and
McCormick (1960), Pasquill (1962a), Panofsky and Prasad (1965),
Kaimal and Izumi (1965), Mordukhovich and Tsvang (1966),
Businger, Miyake et al. (1967), and Cramer (1967)], came to the
conclusion that /s(0) is actually slightly greater than unity. The mean
value of all the estimates of /s(0) =y4a from atmospheric observations
lies in the range from 1.2 to 1.3 (the value 1.25 is used in the
summary Fig. C); although this value is higher than the laboratory
estimate, the discrepancy is not serious. The theoretical prediction
(7.87), according to which fs (S )^ i.e., (qgzlcpPoToY^^
at large negative f, agrees fairly well with the data; this was also
confirmed by the special measurements of Myrup (1967) from an
instrumented aircraft flying over a desert-dry lake bed in California,
at heights of 7.5 m to 93 m. His results showed that the equation
Ovi, = cz'/® is valid in strong convection with rather high accuracy.
The values of the coefficients C5 and C's = C sk ”*/^ were estimated by
Perepelkina (1962) from Gurvich’s fluctuation measurements in the
steppe near Tsimlyansk, on the assumption that Eq. (7.87) is correct
for Ri < —0.04. Perepelkina found that C's 1.4, i.e., Cs 1.9.
However, bearing in mind that Gurvich’s data on Ow are probably
iiiKlcrcstiinatccl, it is not surprising that Perepelkina obtained a

higher estimate C's ^ 1.8, or Cs 2.4, on the basis of the data from
Swinbank’s old and quite rough measurements (1955). The treat
ment of the raw data of Mordukhovich and Tsvang (1966) gives an
estimate which is even slightly higher than the last one: C's ^ 2.1,
Cs 2.8 if all the points with Ri < —0.04 are used. However, a
similar treatment of the Mordukhovich-Tsvang measurements corre
sponding to Ri < —0.3 leads to a lower estimate: C s «« 1.6, Cs ^ 2.2.
Figure 76 gives data on otIITJ = from various sources.
Although the scatter is considerable, especially for near-neutral
conditions where the ratio otI\TJ becomes indeterminate because
both the numerator and the denominator become very small, it is
possible to plot a smooth mean curve on the basis of the most
reliable measurements. According to this curve, the function /e(f)
takes a value close to unity for near-neutral stratification (it is clear
of course that /6(0) can be defined only as a limit of /e(f) as f ^ 0
either from positive or negative values), is almost constant under
stable conditions and drops off somewhat, taking values less than
unity, with increasing instability. The question of the validity o f the
limiting 1/3-power law” (7.87), according to which ^(f)®*
oT^C'eiqlcpPof^îgz/To)'^^^ was specially examined by
Priestley (1 960b) according to data from Australian measurements,
those in the U.S. at O’Neill, Neb., and by Myrup (1967) from
measurements above a desert-dry lake bed in extremely unstable
conditions. Both Priestley and Myrup found that the predicted
relation ot = cz~Î'^ is confirmed rather well by existing data.
Nevertheless, the large scatter of points in Fig. 76 indicates that
additional careful investigation will be very useful. Priestley also
- I -----------1-----------1 -
Panofsky et al. (1967) f Monin (1962b) Tsimlyansk
Round Hill O'Neill o ^ \ Cramer (1967)
112 /; Round Hill
*/am ^ « I
Australia Tsimlyansk *
2
M^0m 9 Vm
o im
I?
-0.7 -0.6 -0 .5 -0.4 -0.3 -0.2 -0.1 0 0 .1 0 .2 0 .3 0.4 0.5 0 .6 0.7 0.8 0.9
^=Z/L
F IG . 7 6 . D a ta o n th e universal fu n c tio n o t H t J = /e C f ) .
made an attem pt to estimate the values of the numerical coefficients

C 6 and Cfi = C from observations. According to his quite rough
estimate, ^ 0.4, i.e., Cg 1.3-1.4. Later, Perepelkina (1962) used
the data o f Tsvang’s measurements o f temperature fluctuations at
Tsimlyansk in 1958-1959 and independently from Priestley obtained
a fairly close estimate C g ^ l . l , i.e., Ce 0 .3 -0 .3 5 . Finally,
Mordukhovich and Tsvang (1966) evaluated the constant Ce Cain
their notation from extensive measurements at Tsimlyansk in 1964
at heights of 1 m and 4 m, and also found that C e ** 1 .0-1.1, that is,
Cfi 0.3. The data from Swinbank’s old measurements (1955) were
improved by Perepelkina (1962); they led to the conclusion that
c ; « 1.2,Ce « 0.35.
The dependence of the correlation coefficients ruw = and
= ixl/s/el on the stability was studied by Swinbank (1955),
Mordukhovich and Tsvang (1966), and Zubkovskiy and Tsvang
(1966). The existing data show that the typical absolute values of
both correlation coefficients and ry^T in the atmospheric surface
layer are in the range 0 .3 -0 .5 . Moreover, increases considerably
with increase of instabihty from about 0.35 for near-neutral
conditions, to about 0.6 for Ri in the range —0.3 to —0.8. On the
other hand, ruw is approximately independent o f stability, and close
to 0.4 for all neutral and moderately unstable conditions.___
The horizontal heat flux measurements qh-CpPouT' in the
atmospheric surface layer led to interesting results. Apparently the
first measurements of this type were carried out by Shiotani (1955)
who published results of three measurements of the value of qh, for
an averaging time of 1 min; the values of q^ obtained in all three cases
turned out to be greater than the simultaneous values of vertical heat
flux q. However, these results were forgotten, or overlooked, by
almost all scientists and for many years most researchers assumed,
and even explicitly formulated, that horizontal flux qhmust be equal
to zero or at least be very small. However, the data of the direct
measurements of Zubkovskiy and Tsvang (1966), and Zubkovskiy
(1967) [see Fig. D] show that the ratio -qhlq = «/?(?) is close to 3 for
near-neutral conditions and even grows with increasing stability. With
increasing instability the ratio drops off but not too rapidly, taking a
value near 1.5 at Ri «= —0.8. These data also agree with those of
similar measurements made in England, above the surface of a lake,
under the direction of P. A. Sheppard (unpublished). The absolute
value of the correlation coefficient r u T ^ u l'J a u O f is close to the
absolute value of the coefficient r^T (^ 0.35) for near-neutral
conditions, and even exceeds lrH»rl in inversions; however, with

increasing instability IrwrI grows and |r „ 7’| decreases, so that,
0.6 and - 0 .2 when Ri = —0.7 according to the data of
Zubkovskiy and Tsvang. The theoretical deductions of Sect. 7.5
imply also that u T ' and 0 as Ri ^ —«>. O f course, it must
be borne in mind that it is natural to compare the horizontal
turbulent heat flux qh with the horizontal advective heat flux CpPouT
and not with the vertical heat flux q. In comparison with the
advective flux, the flux q^ will usually be rather small and often can
be neglected. However, some problems exist which quite definitely
require due consideration of the flux q^ [the semiempirical approach
of Monin (1965) is one example; cf. also Yaglom (1969)];
moreover, the nonvanishing of qh impUes im portant results on the
form of the eddy diffusivity tensor, which will be considered in
Sects. 10.3 and 10.5.
R;
The data on higher-order moments of the turbulent fluctuations in

the atmospheric surface layer are so scanty that they are clearly
insufficient for a graph of any of the corresponding universal
functions to be plotted. However, all the available data are in
agreement with the qualitative predictions of Sect. 7.5. Thus,
according to Deacon (1955), the values of the vertical velocity
skewness w'^/a^ for unstable stratification lie between 0.24 and 0.81,
that is, are strongly positive as was predicted in Sect. 7.5; moreover,
these values increase considerably with increase of instability.
Similarly, Gurvich (1960) obtained an estimate of the order 0 .4 -0 .6
ioT~i^la\v at heights of 1 m and 4 m above the steppe at midday in

the summer. Finally, Pries and Appleby (1967), and Holland (1968),
analyzed the skewness of w, together with that of u, v, and T, for a
series of observations in the U.S. at Round Hill, Mass., over a
relatively inhomogeneous surface. They also found that w' has
positive skewness in unstable conditions, increasing with the instabil
ity parameter - f = z/lZ,l. Even fewer data exist for the values of
temperature skewness , but the data of Pries and Appleby, and
that of Holland, indicate that these values are positive under both
unstable and stable conditions which agrees also with the very
preliminary results o f Perepelkina (1957) for the unstable case. Some
data on the skewness of the horizontal velocity components u and v,
and on the flatness factors 6, i.e., on the ratios of the fourth
moments to the square of variances, or, what is the same, on the
coefficients of excess 6 - 3 of the fluctuations u', v' w', and T' in the
atmospheric surface layer (based on Round Hill observations), were
analyzed by Pries and Appleby (1967) and Holland (1968); however,
all the results obtained are very preliminary and will not be dis
cussed here.
From the dependence of the correlation coefficients r«vv. ’’uT, and
ry,T on f, we may conclude that the multidimensional probability
density for the normalized fluctuations u'fou, v’lov,w’/aw, and r ' / a j ’
varies with change in the stratification, specifically, that the form of
the corresponding probability distribution is not universal. Even the
form o f the one-dimensional distributions of w'/ctw and T 'joj is not
universal, being essentially non-Gaussian. In addition, the probability
distributions of uja^ and vjoy, in general are not strictly Gaussian,
but they are probably close to the Gaussian distribution in many
cases. Some data on one-dimensional probability distributions of
turbulent fluctuations in the atmospheric surface layer are given in
Pries and Appleby (1967), and Holland (1968). Holland (1967;
1968) has also developed a special computational technique for evalu
ating a great variety of different statistical characteristics of the four
dimensional distribution of u , v', w', and T', and o f the correspond
ing filtered variables for several digital filters. Among these are the
two-dimensional joint probability density functions for many pairs of
variables, the joint moments, and the conditional mean values of
different variables at fixed values of a pair of other variables.
However, the possibility of analyzing all these results within the
framework of the general similarity theory for turbulence in a
stratified fluid is very remote.
Several methods exist of determining the values of the variables e

and N, but they will be discussed in Chapt. 8 of Volume 2 of this
book. Nevertheless, at present, almost no data exist on the values of
N in the atmospheric surface layer and on the corresponding
universal function ‘PMS). There are considerable data on e in the
surface layer and on the universal function iê(f) [for example: R. J.
Taylor (1952); Takeuchi (1962); Panofsky (1962); Lumley and
Panofsky (1964); Hess and Panofsky (1966); Record and Cramer
(1966); Zubkovskiy (1967); Myrup (1967); Panofsky, Busch et al.
(1967); Busch and Panofsky (1968);and Fichtl (1968)]. However, all
the available empirical estimates of display a very great scatter;
moreover, some of the most extensive measurements were carried
out in insufficiently homogeneous locations (Brookhaven, Round
Hill, and Cape Kennedy) and at relatively great heights. Therefore all
attempts to verify the turbulent energy balance equation (7.102) are
presently inconclusive and lead to contradictory results. We must also
bear in mind that all the existing direct measurements of turbulent
energy diffusion and the corresponding universal functions/io(f) and
?/io(?) [Record and Cramer (1966); Myrup (1967); Panofsky et al.
(1967)] are extremely rough, do not include the pressure term which
is usually assumed without proof to be small, and are so scattered
that it is almost impossible to use them. There are two main
hypothetical simplifications of the balance equation (7.102) which
have been suggested (and advocated) by different authors: 1) the
assumption of an approximate balance between the rate of energy
dissipation ? and the rate of production of mechanical energy
-u'v'(duldz), for example, on the approximate relationship iê(f) =
^( 0 = f/'(f)» which implies for nonneutral conditions that the
diffusion of turbulent energy be approximately equal to the rate of
buoyant energy production or absorption w'T'(g/To) in absolute
value, but has the opposite sign, so that f + (x/2)f/'io(0 0 [Lumley
and Panofsky (1964); Hess and Panofsky (1966); Record and Cramer
(1966); Fichtl (1968)]; and 2) the assumption of the negligible
smallness of the energy diffusion term and the validity of the
approximate equation of the form (7.103) [Takeuchi
(1962); Panofsky, Busch et al. (1967); Busch and Panofsky (1968)].
The contradiction between these two opinions is not yet explained
satisfactorily. It is also worth noting that Myrup (1967) has found
some preliminary indications that in extremely strong convection,
where the mechanical energy production plays no substantial role and
the term {/'(f) may be neglected in Eq. (7.102), energy diffusion is
important and cannot be neglected (that is, Eq. (7.105) here is

incorrect and a # 1). However, this result also needs further
confirmation.
PARTICLE DISPERSION
IN A TURBULENT FLOW
9. THE LAGRANGIAN DESCRIPTION OF TURBULENCE

9.1 The Lagrangian Dynamic Equations of an
Incompressible Viscous Fluid
In the preceding chapters the Eulerian description of the fluid
motion is used throughout. In such a description, the motion of an
incompressible fluid (to which we shall also limit ourselves in this
chapter is characterized at time t by the velocity field u { X , /), that is,
by the values of the velocity vector at all possible space points
X = (iVi, Xt, Xs). Here, for reasons that will be clarified from what
follows, it will usually be convenient to designate the coordinates as
Xi instead of Xi as was done in foregoing chapters. In this case, the
dynamic equations, from which pressure may be excluded with the
help of Eq. (1.9), permit at least in principle, determination of the
values o f the Eulerian variables u {X, t)at any time / >/o in terms of
the given initial values « (X, to) = «o i^)- However, it is inconvenient
to use Eulerian variables to study such phenomena as turbulent
527
transport, that is, the spread of admixtures in a turbulent flow, or

the deformation o f material surfaces and lines in a turbulent flow,
which consist of fixed fluid elements. It is more convenient to use
the Lagrangian description of the flow which consists of following
the motion of fixed “ fluid particles” beginning from some initial
time L= /q rather than the velocities o f the fluid at fixed space points
X. By fluid particle we mean a volume o f fluid having linear
dimensions which are very large compared to the average distance
between molecules. Therefore, for such volumes it is reasonable to
speak o f their velocity, and remain within the framework of the
mechanics of a continuous medium. Nevertheless, the linear dimen
sions are so small that the velocity and pressure inside the volume
may be considered practically constant, and during the time interval
under discussion these volumes may be considered moving “ as a
whole,” that is, without noticeable deformation. In other words, a
“fluid particle” is an identifiable “ point” o f the volume of fluid
which is moving within this volume according to the equations of
fluid mechanics. The Lagrangian description is related most directly
to the motions o f the individual fluid elements, producing in sum the
fluid flow; therefore, it is physically more natural than the Eulerian
description. At the same time the use of Lagrangian variables, which
refer to individual particles of the fluid, turns out to be much more
awkward analytically than the use o f Eulerian variables u { X , t ) . As a
result, the fluid dynamic equations in Lagrangian form are rarely
used for specific calculations. Although for an ideal, nonviscous fluid
such equations are presented in some well-known textbooks [see for
example. Lamb (1932)], for a viscous fluid the corresponding
equations are comparatively unknown at present. A derivation of
these equations which permits them to be written in a relatively
compact form was first reported only rather recently by Gerber
(1949) [see also Monin (1962c)]. Although the use of the viscous
Lagrangian equations in turbulence theory is still a matter for the
future [cf. Lyubimov (1969)], we shall nevertheless present here the
derivation of these equations, and at the same time introduce several
concepts and symbols that will be used constantly later in the
chapter.
The complete Lagrangian characteristic of an incompressible fluid
flow is the function X { x , t) which gives, for any time t, the
coordinates X o f all possible “ fluid particles” identified by the values
of some parameter x . In principle, the fluid dynamic equations
permit the value o f A'(jf, /) to be determined for any / > ôin terms
PARTICLE DISPERSION IN A TURBULENT FLOW 52 9
of the given initial values of the “ fluid particle” velocities V{x, t) =

that is, in terms of the values of
d X ( x . t)
dt
The relationship between the Lagrangian and the Eulerian character

istics is given by the expression
= t]. (9.1)
The transformation from the Eulerian dynamic equations to the

Lagrangian equations includes replacement o f the independent
variables ( at, t) by ( X , 0 and transformation from the unknown
function it (X, t) to the new unknown X (jc, t) related to u { X , t) by
Eq. (9.1).
Furthermore, in this section, we shall always use as the Lagrangian
parameters of the fluid particles x , the initial values of their spatial
coordinates X at the time / = ; that is, we assume
x = X { x , to). (9.2)
In this case the function o f two variables X ( x , t) describes the

family o f trajectories of “ fluid particles” found at the initial time
/ = /o at all possible points x of the volume occupied by the fluid.
Thus at any time t > ô. the points X = X {x, t) corresponding to all
possible values of x continuously fill the entire volume occupied by
the fluid. Therefore, the Lagrangian description consists of assigning
a fluid flow by a family of trajectories, differing from each other by
the values of x , and in each of which the role of the parameter is
played by the time t. According to the above, the “ fluid particles”
which correspond to these trajectories are actually mathematical
points flowing along with the fluid.
Let us now derive the Lagrangian dynamic equations. We shall use
the Cartesian components of the vectors X and x , and designate
them as (^,, Xz, Xs)and (a:i, Xi, Xs), respectively. As indicated above,
the transformation from the Eulerian dynamic equations to Lagrang
ian equations includes first the replacement of the independent
variables (Xi, X 2 , X 3 , t) by(A;i, X2 , Xs, t). In making this change o f
variables, we emphasize the transform from Cartesian coordinates to

nonstationary curviHnear and nonorthogonal coordinates which
follow the motion of the fluid. Actually, at all times each coordinate
surface Xi = const consists of the same “ fluid particles” ; initially
such surfaces are planes, but as time passes they become distorted
due to mixing with the fluid.
Furthermore, with respect to the variables (xi, x-z, jcs), we shall use
for the Jacobians the abbreviated notation
d (A, B. C) _ 3^
d ( ^ ,. X2, X3)
and without further comment the fact that the quantity [A, 3, C]
does not vary under an even permutation of the variables A, B,C, and
changes sign under an odd permutation.
In transforming from Eulerian coordinates to Lagrangian
coordinates a major role is played by the matrix
dX^
M=
dXa
According to Eq. (9.2), at the initial time
matrix M is a unit matrix and |M| = 1. The variables - ^ a r e elements

dx,
of the inverse matrix Af-', or the co-factors of the e l e m e n t s - ^ in the
matrix M, divided by |yW|. Consequently, we obtain the following

formula for calculation of the derivatives with respect to the Eulerian
coordinates
^ = (»-5)
Here and henceforth by (/, /, k) we mean an even permutation of the

indexes ( 1 ,2 , 3). It is easy to prove the correctness o f formula (9.5),
by writing the left side as where the repeated index a, as
usual, denotes summation; the same expression is obtained on the
right side of the formula by expansion of the determinant which
enters into it with respect to the elements of the column /.
PARTICLE DISPERSION IN A TURBULENT FLOW 531
Prom Eqs. (9.1) and (9.5) is derived the following expression for
the velocity divergence:
dll, () t)X~
~ dx: i w
Y 2 \r I d\ M \
-f-
^1’ ~W ’ 4- \M\ dt
where are the Cartesian components of the velocity field. For an

incompressible fluid the velocity divergence is exactly equal to zero;
thus - s= 0 and consequently \M\ does not vary with time. Since
at the initial time \M\ = 1 , this equation remains correct also for any
time. Recalling expression (9.4) for|yW|, we have
[ X „ X „ X , ] = \. (9.6)
This equation also acts as the continuity equation for an incom

pressible fluid, either viscous or ideal, in the Lagrangian formulation.
Furthermore, we shall use formula (9.5), setting \M\ = 1 on its
right side. Using this formula twice, we obtain the expression for the
Laplacian with respect to the Eulerian coordinates A’.:
f __ ^ —
^ x J — dX, dX, —
= [;^2. ^3. [^2. ^3-/1 ] + 1 ^ 3 . ^1.1^3. 1 + [X„ X „ [X„ X , J \ I .
(9.7)
Now we can easily proceed to the Lagrangian formulation of the

Navier-Stokes equations
dui 1 dp
(9.8)
Using Eqs. (9.1), (9.5), and (9.7), we obtain
V V rV V dXi
+ ^3> -^1' -^3' -^1' dt
. (9.9)
Equations (9.6) and (9.9), in the unknowns X, (jr, t),i = 1 , 2, 3 and

p { x , t ) form the complete system of Lagrangian dynamic equations

of an incompressible viscous fluid.
Forces exist corresponding to the nonlinear terms in the equations
of motion. These forces describe the interaction between the
components of a mechanical system. Thus, in the Navier-Stokes
equations ( 9 .8 ) nonlinear, that is, quadratic terms in the variables w,
are found in the expression for the acceleration these represent
the forces of the inertial interaction between the spatial inhomogene
ities of the velocity field tt{X, t) in terms of which the pressure
gradient may be expressed with the help of Eq. (1.9'). (Let us
emphasize that the viscous forces are described in Eq. (9.8) by a
linear expression.) However, the inertial interactions have a relative
nature; they are eliminated in the transformation to the particle
attached reference system. In the Lagrangian equations of motion
(9.9) the expressions which are nonlinear in the variables describe
only the real forces of interaction between the fluid particles, that is,
the pressure and viscous forces; the viscous interactions are described
here by nonlinear terms of the fifth degree in the variables Xj. The
ratio of the typical values of the nonlinear and linear terms in the
equations of motion may be called the interaction constant. Thus,
for the Navier-Stokes equations (9.8), the inertial interaction
constant is the ratio of the typical values of the inertia forces and the
viscous forces, that is, the Reynolds number Re; for sufficiently large
Re, which is characteristic of developed turbulence, the inertial
interactions are quite strong. In the Lagrangian description the
viscous itneraction constant is the ratio of the typical values of the
viscous forces to the typical acceleration, for example, 1/Re; for
large Re the viscous interactions turn out to be quite weak.
Let us now formulate the Lagrangian dynamic equations for a
two-dimensional, plane-parallel flow. Let the motion of the fluid
take place only in the planes ;c3 = const, so that ^ 3 = ^ : 3 , where Xi
and X 2 do not depend on X3. Then, using designations of the following
type for the Jacobians with respect to (jci, X2):
d(xi. X2) (9.10)
Eqs. (9.6) and (9.9) may easily be reduced to the form

PARTICLE DISPERSION IN A T U R B U L E N T FLOW 533
d’A", 1 , V,, I )r r V, dA-, dX,

+ A-2, ^2.2’ dt (9.11)
^’■^2_ *FY nllv^fvrv dXi
;i:2. X.2’ ~ W 1-
Here the viscous forces are described by nonlinear expressions o f the

third degree in the variables X,-.
In concluding this subsection, let us note that in addition to the
use o f expression (9.1), another means exists o f relating the
Lagrangian and Eulerian descriptions o f fluid flow. This is done by
investigating arbitrary conservative characteristics of “ fluid parti
cles,” that is, characteristics that have values for a fixed fluid particle
which do not vary during its motion. In the Lagrangian description
each such characteristic may be written in the form 'F(at) since for
fixed X it does not depend on time t. In the Eulerian description,
however, the values of such a characteristic at a fixed space point
X may vary with time t, that is, t). The relationship
between the Lagrangian and Eulerian descriptions will be given by
the expression
t)> ^1- (9.12)
Since = 0, then the total derivative with respect to time o f

the right side must also be equal to zero, that is.
dt dt "1“ dX, dt M
dt I “ a fty
dX„ —
where we have used Eq. (9.1). Using also the continuity equation for
an incompressible fluid 0A„ — 0, we obtain
^4-
dt dX,
= 0. (9.13)
This equation is the differential analog of expression (9.12). Thus,

the Eulerian field o f any conservative characteristic of the fluid
particles satisfies the advection equation (9.13).
Consequently, the conservative characteristic'F(at) = 6 ( x — JCo)

which differs from 0, and is equal to infinity only for one fluid
particle, namely, the initial coordinate Xo, corresponds to the
Eulerian field
<Sf(X, 0 = 01, (9.14)
which is the solution of the advection equation (9.13) for the initial
condition
to)==o{X-x,;). (9.15)
The general notion o f a “ conservative characteristic” represents, of

course, a mathematical idealization. To make it possible to actually
follow the motion o f the fluid particle it must be “marked,” that is,
have some properties distinguishable from those o f the surrounding
medium. Such marking is achieved most simply by the addition of
some distinctive passive admixture which is transported by the fluid
motion without exerting an influence on the flow. In such a case the
admixture concentration >?(X, O acts as the Eulerian conservative
characteristic ^(X, <). However, the real concentration field is also
subjected to molecular diffusion; consequently, it satisfies the
diffusion equation
= xW . (9.13')
dt dX„
This equation differs from the advection equation (9.13) for the
idealized conservative characteristic ^(X, f) by the proportionality of
the right side to molecular diffusivity. Now the Lagrangian form of
the diffusion equation may be deduced in a manner quite similar to
that o f the deduction of the Lagrangian dynamic equations [Corrsin
(1962); cf. also Okubo (1967)]. Here the Lagrangian concentration
field
d(x,0 = ,?[X(x,«] (9.12')
would be the main variable; this variable depends on / and satisfies

the equation
dOb = y f c . X g , [X2,Xg,,>]] + [Xg,Xl,[X3,Xl,i>]] +

dt ^
(9.9')
+ [X i,X 2,[X i,X 2,.?]]|’
which implies that i? does not depend on t when )<: = 0, where

X.(x, t), / = 1, 2, 3 are the solutions o f the dynamic equations (9.6)
and (9.9).
9.2 Lagrangian Turbulence Characteristics
For a statistical description of turbulence it is necessary first to
indicate which of its characteristics are assumed to have probability
distributions, that is, which are realizations of some random field. In
the discussion o f turbulent flows in previous chapters, we always
assumed the random field to be the Eulerian velocity field u (X, t).
However, in that case, the Lagrangian variables
t) = u \ X { x , t), (] and X ( x . t) = x + f V{x, t')d t'
will also be random functions of the arguments x and /. Moreover,

for any finite number n o f fluid particles, identified by the fact that
at the time t = 0 they are at the points X i......... x„, multidimen
sional joint probability densities will exist for their coordinates X
and velocities V at arbitrary times .. ,tm- These multidimensional
densities are functions of 3« + /n variables xu, «i2, ^ 13, • •., Xn\, Xn2, Xnz
t \ , , tm , where x n , Xa, xa are three components of the vector JC/,
and are the basic Lagrangian statistical characteristics o f turbulence.
Let us note also that joint probability distributions will exist for sets
of variables, that is, for both the Lagrangian coordinates or velocities
of fixed fluid particles, and the Eulerian flow velocities at fixed
points; the densities o f these “mixed probability” distributions are
sometimes also o f interest.
We shall designate probability densities by the following type
symbols:
p{X, V, . . . \Xi, X 2, tu ti, ...) ,
where those values to the lefr o f the vertical line in parentheses

536 STATISTICALFLUIDMECHANICS
)
denote the random variables in question, designated by the same
letters as the corresponding variables themselves, and those values
after the vertical line, the parameters on which these variables
depend. If the parameters corresponding to several variables coincide,
and if there is no danger o f confusion, we designate them following
the straight line only once.
The various Lagrangian and mixed statistical characteristics of
turbulence satisfy many general relationships, some o f which will be
pointed out here. We shall begin with the relations resulting from the
“ advection equation” (9.13). By taking expression (9.14) to be the
solution of this equation, and according to Eq. (SJ. 1)
/ ) 8 [ A ' - A - ( J » : , /)! =
= u^X(x, t), t ]h\ X-X{x, t)] = VAx, t ) H X - X { x , 01.
we obtain
[ A - - A -(x . 01 , d V , (X. t ) 8 [ X - X { x . 01 ^
dt dXl
However, from the definition of probability averaging [see Eq.

(3.12)]
h \ X - X ( x , t ) \ ^ p { X \ x , t),
/ ) 8 I ^ - ^ ( ^ . 0] = / VaP{X, V \ x , t)d V.
Consequently, averaging the equation for 6 [A"— A' {x, /)], we have
+ ^ t)d V = 0 , (9.16)
which is the statistical analog o f the advection equation (9.13).

Another type o f relationship between the statistical characteristics
may be obtained along with the Lagrangian velocity V (x, t), which
represents the value of the random function « (A, t) at the random
point X { x , t), by introducing the random variable V(A',, x , t)
depending on the fixed point X ^ = X { x , t). The values o f this
variable are the velocities o f those fluid particles which at the time
t = to were found at the point x, and at the time t, turned out to be
PARTICLE DISPERSION IN A TU R B U LE N T FLOW 537
at the fixed point Ai; its probability density p { V \ X i , x , t) is the

conditional probability density of the variable V ( x , t), under the
condition that X { x , t) = Xi- Therefore, strictly speaking, the
probabiUty distribution of V(A’i, x , t) is neither purely Lagrangian
nor purely Eulerian. However, it may be anticipated that with an
increase in.^ — to, the dependence of this probability distribution on
the value of will generally become less and less essential. For
sufficiently large values oT t = to, by comparison with the Lagrangian
correlation time, or the Lagrangian integral time scale,
,00
T^ f V',{x.t)VUx.f,)
J [ r ^ x . t) v ^ x . ’
this dependence may often be completely neglected. Thus, the

random variable V(X\, x , t ) , for such t — to, may frequently be
considered equivalent to the Eulerian random variable u{Xi, t).
Nevertheless, let us emphasize that replacement of V(Xi, x , t) by
u { X u /) may not always be proper for any values of^ — fo- For
example, we shall see later that in a turbulent boundary layer along a
plate at = 0 , Vz{X^, at, 7 ) '> 0 for all Xi and t, particularly in the
logarithmic layer V'3 (A’,, x , t) = const =« bu^, where is a
universal numerical constant, while Ui(Xu t) — 0.
It is not difficult to see that the joint probability density for the n
fluid particle velocities V(jCi, /|), . . . , /„), taken at different
times t i , . . . , tn), may be represented in the form
P (î» • *• > I*^1, . • .» » ^/i)

~ f ’''f P (^l> • • •> 1-^1’ • • • ’ "^1’ ■ • • ’ ^1* • • • ’ ^
X p (Aî. • • •, I Jfj, . . . , x„\ ti, . . . , t„)dXi . . . dX„,
(9.17)
where the first factor under the integral sign is the conditional probabil
ity density for the indicated velocities of the fluid particles under the
condition that their coordinates at the corresponding times assume
fixed values X{X u / , ) = Xi, . . . , X{x „, t„) = X„. If all the fluid
particles are different, that is, among the initial points Xi, no
two are identical, then this first factor is the joint probability density
for the random variables V(X^, Xi, ti), V(X„, x„, t„) intro
duced above. The second factor under the integral sign in Eq. (9.17)
538 STA T IS TIC A L F L U ID MECHANICS
is the joint probability density for the Lagrangian random variables

X{Xi, U), X{Xn, <„). Equation (9.17), which like its corol
laries is taken from Monin (1960), is a special case of the gen
eral “ theorem of total probability” o f probability theory. If all
the fluid particles under discussion are different and all the
differences h — to, , tn — to are sufficiently large, then the
dependence o f the first factor under the integral sign in Eq. (9.17) on
the arguments Xi..........x„ may be neglected in several cases. This
factor may then be identified with the probability density o f the
velocities Vi = u(Xi, ti), . . . , V„ = u{X„, tn)at fixed points so that
Eq. (9.17) here becomes an approximate relation between the
Lagrangian and the Eulerian statistical characteristics of turbulence.
Let us give special consideration to the particular case of Eq.
(9.17) when Xi = X 2 = • • ■ = x „ ( = x). Here the formula contains
the probability densities for the coordinates and velocities of a single
fluid particle at different times. When n = 1 we obtain
p { V \ x , 0 = / P { V \ X , X, t ) p { X \ x , t)d X . (9.18)
As already noted, for sufficiently large t — U the function

p { V \ X , x,t) may sometimes be identified with the probabiUty
density of Eulerian velocity V — u{X, <) at a fixed space-time point
( X , t);in this sense Eq. (9.18) may be considered as the statistical
analog o f the basic expression (9.1) relating the Lagrangian and
Eulerian velocities. Now substituting Eq. (9.18) into the expressions
for the mean value of V{x, t) which is derived from the general
equation (3.12), we obtain
V{x, t) = f V { X ~ 1 ^ ) p { X \ x , t) dX. (9.19)
When the dependence V{^X, x , /) on jc disappears for t — <x>, for

sufficiently large t — /othis equation may be rewritten in the form
V{x, t) = f u J x T J ) p i X j x , t) dX. (9.20)
For the joint probability distribution o f the velocities V^ — V{x, to)

and V2 = V (x, t) o f a single fluid particle at two sequential times /o
and t > /o, Eq. (9.17) leads to the expression
PAR TICLE DISPERSION IN A T U R B U L E N T FLOW 539
V2\x, = f p{V ^,V ,\X -,x,t„ t)p{X \xJ)dX , (9.21)
where the first factor under the integral sign is the joint probability
d e n s ity f o r th e v a ria b le s V^=V{x, to)= u{x, to) and
= V{Xy X, t). Thus from this, for the Lagrangian velocity
correlation function, that is, the mixed second moment VuVij of the
distribution (9.21), where the subscripts i and / are the numbers of
components of the vectors Vj and Vj, we obtain the following
relation;
V,{x, t^)Vj{x, t) = f u,(x, to)Vj(X, X, t) p(,X \x, t) d X . (9.22)
For sufficiently large values of t — /o> that is, in comparison to the

Lagrangian correlation time, the velocity correlation function under
the integral sign may be identified with the simpler Eulerian
space-time correlation function W; (jc, t^ U j{X, t) on the basis of the
above arguments. However, for these/ — /q. the values ofV'((Jc, to)
and Vj { x , t), and also the values o f Ui { x , /o) and Uj{X, t) may
already be considered practically uncorrelated; therefore, in this case
the asymptotic form of the equation is not of great interest. The left
side o f Eq. (9.22) is not the most general Lagrangian velocity
correlation function since at one o f the times, namely, to, the
coordinate o f the corresponding “ fluid particle” has a fixed value
equal to x . The Lagrangian velocity correlation function
Vi(x, ti) Vj{x, (2) where t 2 > t i > to is more general; using Eq.
(9.17), we can express it as
V7(4f, t, ) V j ( x , t^) - / / V ,{ X „ X , t,)V j{ X ,, X 2, x, t,, t^) X

y^p^X^, X 2\x, ti, t<^dX^dX2, (9.23)
where V{Xx, X2, x , t\, <2) indicates the random velocity of the fluid
particle at the time <2 under the condition that at the times to, t\, and
/j, this particle is found at the fixed points jt, X,, and Xj, respectively.
For sufficiently large t^ — to, the value of K(A’i, x, /,)m ay often be
considered approximately equal to « (^"1, /,); in exactly the same
way, for sufficiently large t^ — tu the value of K(A'i, Xj, x , tu ^2)
frequently may be considered approximately equal to«(A ' 2, ti)-
540 S T A T IS T IC A L F L U ID M ECHANICS
9.3 Displacement Characteristics of a Single Fluid Particle;

the Case of Homogeneous Turbulence
The motion o f a fluid particle, that is, one which at the time
is at the point x, is described completely by the vector function
X { x , t) which gives the position of this particle at an arbitrary time
t. In the place of the vector X{x, t) one may also use the
displacement vector o f the particle during a time interval x:
Ô+'C
K (x) = {X, + 1) - jf = J V { x , t ) d t ; (9.24)
this usually proves quite convenient. In this subsection we shall

study the statistical characteristics of the random vector K(t).
A complete description of the variable y (t) requires assignment of
its three-dimensional probability density p { Y \x\ x, t o ) , which
depends on t, x, and to as parameters. It is easily seen that for very
small t, compared with the time scale T given by the typical
Lagrangian correlation time, this probability density may be ex
pressed in terms o f the Eulerian statistical characteristics of
turbulence. Indeed, when t <C T, the Lagrangian velocity practically
does not vary in time t; therefore Eq. (9.24) may be rewritten here as
K (t)« V(x. to)x = tt{x, to)x.
Therefore, the probability density for V (t) in this case may be

transformed to
p{Y\x-, X, to) « I JC, ô). (9.25)
where P{u\ x, to) is the probability density of the Eulerian velocity

u{x, to) at the fixed point x at time t o- If the turbulent flow is
steady, then the density P{u\x,to) may easily be determined by
analyzing velocity observations at the point x during a long time
interval. Thus, for example, in turbulent flow in a wind-tunnel having
a square-mesh grid at its opening, thereby creating strong turbulence
downstream, the measurement data of Simmons and Salter (1938)
and Townsend (1947) show convincingly that the distribution
P{u\x), and consequently also/? (K| x, x) when t < 7", is quite close
to normal.
For not very small t the distribution p{Y\-^\ x , t o ) may not be

expressed in terms of the Eulerian statistical characteristics. How
ever, if then one may use the fact that here the right side of
Eq. (9.24) may be represented as a sum of integrals taken over
nonintersecting time intervals of length o f order T, and that are
weakly dependent random variables. Therefore, it is possible to apply
to this sum the so-called central limit theorem fo r weakly dependent
random variables, according to which the probabihty distribution of
the sum of a large number of such variables, under some broad
conditions, turns out to be very close to normal. Recently, the
central limit theorem was also proved directly, although with some
mild restrictions, for integrals of the type (9.24); see, for example,
Rozanov (1967) where an integral of a stationary random function is
considered; similar theorems occur also for integrals of several
nonstationary random functions. Unfortunately, direct use o f these
proofs is still impossible since the mild conditions imposed on the
random functions which figure in them cannot be directly verified
for real processes.* Nevertheless, these conditions are so natural that
it would be extremely remarkable if the probability distribution for
the displacement Y(x) when differed essentially from a
normal distribution. In some cases the distribution for Y {%), or at
least for some components of this vector, may be determined
approximately by measuring the concentration distribution at
various cross sections o f the “ mantle” of contaminated fluid created
by a source of admixture placed in the flow, for example, the
distribution of temperature in various cross sections of the thermal
wake behind a hot body. Thus, experiments have successfully shown
that in many turbulent flows the distribution Y (t) for large x actually
is very close to normal, while in the special case of wind-tunnel
turbulence behind a grid, it turns out that it is almost normal for all
values of t [see, for example, Collis (1948); Townsend (1951);
Uberoi and Corrsin (1953)]. The fact that the distribution of K(T) is
almost normal for all values of x is not surprising, since as we have
seen, it must be normal both for small and for large values of -r; its
closeness to normal for all x indicates only that for intermediate
' T h u s, fo r e x a m p le , several p r o o fs use th e so-ca lled strong mixing conditions-, as a p p lie d

t o th e f u n c tio n V ( x, t), th is r e d u c e s , r o u g h ly sp e a k in g , t o t h e r e q u ir e m e n t o f th e e x is t e n c e
o f a p o s itiv e fu n c tio n a ( t ) , su ch th a t a ( t ) - > 0 as T - > c o a n d th a t th e c o rr e la tio n
V ( x , t ' ) , — <x>< t ' < to, an d
c o e f f ic ie n t b e t w e e n a n arbitrary fu n c tio n a l o f th e values
th a t an arbitrary fu n c tio n a l o f th e values o f V ( x , t " ) , to + x < I " < c o , b e le ss th an a ( r )
in m o d u lu s .
542 S T A TIS TIC A L F L U ID MECHANICS
values of x no sharp readjustment of the form of the distribution

occurs. In general, however, there is no reason to expect that the
distribution of V (t) will also be close to normal for moderate values
of t.
Let us now investigate the most im portant numerical character
istics of the random vector V(r), its first and second moments. The
mean value o f this vector equals
/o+'C
F ( ^ = J v { x , t)dt. (9.26)
According to this, it may be shown that in several important special

cases Y (t) = Ux where U is the suitably defined average flow
velocity (see examples below). Then it remains only to investigate the
characteristics o f the displacement fluctuations
/o-+
r (^) = Y{x)~ ? 1 T ) == f V'(X, t)dt.
The second moment tensor of this random vector will be given by

the equation
(t) = r'i (X) Y'j (t) = J

D ij
/ fJ V- (X, /,) v ; (X, t 2) dtx dh\ (9.27)
it is called the fluid particle displacement covariance (or variance)

tensor. Let us note that when the probability distribution for K(t)
may be considered normal, the vector (9.26) and the tensor (9.27)
characterize this distribution completely.
For sufficiently small T,for which Y ' { \ ) ^ u ' ( x , /o)": >from Eq.
(9.27) we obtain
The coefficients B.. — u'.(x, (x, t^) for steady flows will de
pend only on x and, with the additional assumption of homogeneity
of the random field u { x , t), will be strictly constant. Specific
information about the form of the functions for not too small
T may be obtained, however, only for some special turbulent flows.
PARTICLE DISPERSION IN A T U R B U LE N T FLOW 543
We consider first the idealized case of stationary homogeneous

turbulence in which all flow variables are homogeneous random
fields in three-dimensional space and at the same time, stationary
random functions of t. In this case the mean velocity « is constant
both in space and in time, and consequently
V(jc, t ) ^ u \ X ( x , t), t\--=u, K(x) = « t.
Furthermore, here the fluctuating velocity of the fluid particle

V' (x, t) will have the same statistical characteristics for all x and will
be a stationary random function o f /, so that
V'i {X, h) V j {X, h) Bf} {h - h) = { u f u j y R\^> {(2 - / , ) , (9.29)
where the letter L indicates that the corresponding functions are

Lagrangian correlation functions. Changing now to the new variables
s = t2 — t\ and t — Eq. (9.27) may be rewritten as
Dij (t) = / / [ is) + (s)] dt ds =
^ {u 'iu fy f f (s) + r !!-} (s)] dt ds

(9.30)
or after integrating with respect to t
D u (x) ^ J (x - s) [ is) + 5'// is)] ds =
= (a f ) J (x — s) [R u (s) + Rfi (s)l ds. ( 9 .3 0 ')

0
A result similar to Eq. (9.30) for the variance of one component of

V (t) was first obtained in the classical work of G. I. Taylor (1921); it
was presented by Kampe' de Feriet (1939) in the form of Eq. (9.30')
for the case where j = y, and by Batchelor (1949b) in the general case.
When t = j, Eq. (9.30') becomes
Du(^) = 2 f (x-s)B^/i>(s)ds = 2 u ' / f ( z - s ) / ^ ^ ' ( s ) d s . (9.31)

0
(Here, just as below, summation with respect to i is not assumed!)

Let us now make the natural assumption that the Lagrangian
correlation function approaches zero when s->-oo, and so
rapidly that the following correlation time will exist:
7^ = (9.32)
0
As the Lagrangian time scale T mentioned above, the maximum or

average value o f the three variables Tu i = 1, 2, 3 may be taken.
Assuming also that the integral
CO
f sR'/i\s)ds = Si (9.33)
is finite, it is possible for sufficiently large that is, when i ^ T i , to

replace Eq. (9.31) with the asymptotic expression
_____ CO ____
D u { z ) ^ 2 u ? f { z - s ) R \f{ s ) d s = 2 u'i\T iX -Si). (9.34)
For very large x, the basic role in the right side of Eq. (9.34) will be
played by the term which is linear with respect to x. Thus, Eq. (9.34)
may be rewritten as
Z ) ^ ,( x ) « 2 < 7 > , (9.35)
which, as is easily seen, may also be obtained without assuming

finiteness of the integral (9.33). Similarly, for very large z the
following expression is obtained for Dij{i):
Dii (x)« ( « ; « ; ) 2 r , , x; Tij = f [M;’ (s) + (^)] ds. (9.36)

PARTICLE DISPERSION IN A T U R B U L E N T FLOW 54 5
Equation (9.35) indicates that the variance of particle displace

ment after sufficiently large time t becomes proportional to -t. This
result is completely analogous to the basic law of Brownian motion
according to which the mean square displacement of the Brownian
particle, or any particle that participates in the molecular diffusion,
is proportional to the time of motion (diffusion). For very small t
the displacement variance according to Eq. (9.28) depends quad-
ratically on t, which must be the case for any motion having finite
velocity. For intermediate values of x the dependence of D»('t) on-c
turns out to be more complex and depends on the form o f the
correlation function /?(V(s); see, for example, the graphs of the
functions £)<y('t) corresponding to several special forms of the function
plotted by Frenkiel (1952; 1953) and by Pasquill (1926b).
Also, according to the data of Frenkiel and Pasquill, large variations
in the values o { D u ( x ) , for variation in the form o f the function
M /’ (s), occur only if we assume that the function R u \ s ) may take on
negative values and change sign frequently with increase in the
argument s. For functions Rû (s) which remain positive everywhere,
the dependence of (x) on the specific form of the function R'u (s)
is very weak and the asymptotic formula (9.28) here is satisfied
rather well for all t < 7,-, and the asymptotic formula (9.35), for all
t > 5 7 ’,. Thus, for nonnegative functions Mi* (s) the variance Di,-( t )
turns out to be greatly dependent on the intensity of the turbulence
u f and its time scale Ti, but only very weakly dependent on the
specific form of the correlation function Rft^ (s).
It may be assumed that the results arising from the model of
stationary homogeneous turbulence have no practical significance.
This is because homogeneous turbulence in an unbounded space is
generally only a mathematical idealization, and the assumption of
stationarity worsens the matter since the energy dissipation implies
that a stationary flow of viscous fluid must have external sources of
energy, and therefore may not be homogeneous. Indeed, however, it
is easy to see that the above derivation for formula (9.31) requires
only that the flow be homogeneous with respect to the Oxi axis.
Consequently, we may indicate several completely realistic flows to
which the results obtained above may be applied. In particular,
following G. I. Taylor (1954a) [whose work will be discussed in detail
la te r], Batchelor noted that these results may be applied directly to
the simplest turbulent flow in a sufficiently long straight tube of
arbitrary constant cross section [Batchelor and Townsend (1956);
Batchelor (1957)]. Indeed, if the direction of the tube coincides
with the Ox\ axis, then along this direction the flow will be
homogeneous, although here the distribution o f the mean velocity
ui{x 2, X3) in the 0 X2X3 plane may be quite complex. Let us now
investigate the component yi('r)of displacement of the fluid particle
in the time x in the Oxi direction. The corresponding Lagrangian
velocity Vi (x, /„ + x) will, generally speaking, be a nonsta-
az ^
tionary random function of t which depends on the initial position
of the particle x in the plane 0x>x3. It is natural to expect, however,
that at some time after the starting time to of the particle under
discussion, the influence of the initial position a: will almost cease to
be felt. Thus the function V{{x, /o + t) may be considered independ
ent of X and stationary. Here, the mean longitudinal particle velocity
1/] (jc, /„ + '') = ^ 1 for sufficiently large t will depend on neither t nor
X, that is, it will be constant with respect to time and the same for all
fluid particles which occupy a fixed position at the time U or earlier.
It is obvious, and may be strictly proved with the help of the
arguments in Sect. 9.5, that Ui will also coincide with the mean
longitudinal velocity of all the fluid which intersects the fixed cross
section of the tube during the given time interval; that is, with the
averaged bulk velocity (7ave defined as the ratio of the volume o f the
fluid which comes from the tube in a unit time to the area of its
transverse cross section. Thus,
y, (x) « ^/ave- ^ (9-37)
for sufficiently large 1 . Furthermore, on the basis of Eq. (9.35)
r.T (9.38)
for sufficiently large t ; for slightly smaller values of t it is possiblejg

use the more exact equation (9.34) with i= 1. The characteristics
and Ti in Eq. (9.38) must be determined by the parameters on which
the statistical regime of turbulence in a tube depends. Since in a
round tube, the viscous sublayer fills an insignificant part of the cross
section for a sufficiently large Reynolds number (see above, Sect.
5.5), it follows that in this case and Ti will depend only on the
tube radius R and the friction velocity — | / x„/p, where to is the
shear stress on the wall. On the basis of dimensional arguments
u'iTi = cR u„ (9.39)
where c is a universal constant the value of which may be estimated

from data (see below, Sect. 10.4). A formula of this type will be
correct also for turbulent flow in a straight tube or channel of
noncircular cross section for sufficiently large Re; here, for R it is
only necessary to use a typical linear dimension of the cross section
of the tube or channel, and «... is determined by the mean value of the
wall shear stress.
9.4 Fluid Particle Displacements in Grid Turbulence and

in Turbulent Shear Flows
The Lagrangian Characteristics o f Self-Preserving Turbulent Flows
The derivation of Eqs. (9.37)—(9.39) depends considerably on the
fact that in a tube flow the motion of the fluid particles in the
transverse plane 0 x>x3 proceeds at all times only within the limits of a
fixed bounded part thereof. It is clear that this latter condition is
satisfied only for certain special turbulent flows. For example, in the
case of a turbulent boundary layer along a plane wall or a turbulent
jet, the average distance of a particle from the wall or from the jet
axis will increase w ithout Hmit with an increase in t; in such a flow,
therefore, the Lagrangian velocity cannot be considered a sta
tionary random function of time for any t. Thus, the class of real flows
to which Eqs. (9 .30)-(9.36) can be applied directly is quite narrow.
However, Batchelor (1957) noted that there is also a class of
flows, including several practically im portant ones, to which the
above formulas may be applied after some simple modifications. This
class consists of the stationary, self-preserving flows in which the
mean velocity is directed mainly along the Ox, axis, and the
turbulence structure for various values of the coordinate Xi is similar,
that is, differs only by the values of the characteristic length scale
L{xi) and velocity scale (7(xi). In other words, in such flows all the
Eulerian statistical characteristics of turbulence in the plane a;, =
const, reduced to a dimensionless form by division by the
corresponding combination of scales L and U, do not depend on the
values of x,. In such a case a particular fluid particle will be found at
all times under practically identical conditions, but with a variable
velocity scale U{x) = U[ Xy( x, /o-Ft)] and a variable time scale
T (.\
~U[X,(x, <o + ^ ) ] ’
54 8 S T A T IS TIC A L F L U ID MECHANICS
Therefore, to study the motion of this particle it is expedient to

measure time by the scale T (i), that is, to use in place of the time i
the variable T ) ( t ) related to -r by the expression d r i — Y ^ y and
velocity by the scale L/{i). Here it is natural to assume that for
sufficiently large values o f x, such that the initial positions: has
ceased to have an influence, the dimensionless Lagrangian velocity
fluctuation V (x, to + i)IU{'z) will be a stationary random function
F (ti) o f the variable ti;
True, the strict proof o f this proposition, and generally the fact that
self-preservation o f the Eulerian properties of turbulence implies
self-preservation o f its Lagrangian properties, is still missing. How
ever, it is a very plausible hypothesis, the results o f which agree well
in a number of cases with the existing, although still very incomplete
data concerning the Lagrangian statistical characteristics. Also, when
self-preservation o f the Eulerian turbulence characteristics is estab
lished with the help of dimensional considerations, the same
considerations will generally be used to form the basis for self-
preservation of the Lagrangian characteristics; the results thus
obtained agree in all cases with the conclusions from the assumption
in italics.
One o f the simplest cases to which the hypothesis formulated here
can be applied is that o f turbulence in the central part of a
wind-tunnel behind a turbulence-producing grid. In this case, the
mean velocity a = U q/„ where ii is the unit vector along the axis Oxi,
is strictly constant and consequently
r,(x) = t/oT, K2(x) = r3 (.) = o.
However, the turbulence here still is not completely homogeneous

but is decaying since because of viscosity the intensity of the velocity
fluctuations decreases slowly with increase in the distance from the
grid. As the measurements of grid turbulence show [see Chap. 7,
Volume 2 of this b o o k ], during an initial period of decay,
specifically, at comparatively small distances from the grid, the
turbulence behind the grid usually is approximately homogeneous in
the planes xi — const, and differs for various values of Xi only by the
velocity scale, proportional to (-«i — and the length scale

proportional to where xj is some virtual origin of
reference on the Ox\ axis. Since the mean velocity Uo in such a flow
will always be large compared to the velocity fluctuations, one may
assume t h a t^ j ( t) — = where t i s the time of motion of the
fluid particle reckoned from the virtual time to where it intersected
the plane x^ — a:®. For the velocity scale U (t) in the plane jCj = +
U qX, it is possible to assume any constant velocity multiplied by
for example, it is convenient for investigating the motion of a fluid
particle in the plane Xi = a at the initial time to assume
where Ua is the characteristic value of the velocity fluctuations when

a
x^ = a, and ■ . In this case the variable ti o f Eq. (9.40) is
Uq
equal to In t + const; therefore, one may take i^ = ln — .I t follows
from this that proposition (9.40) here reduces to the assumption that
the Lagrangian velocity correlation function has the form
V i ( x , /o + 'î) y j { x , /o + '^2) =
- Ulâ 5/y (in ^ - In ^ ) . (9-41)
where is the cross-correlation function of the stationary processes
Substituting Eq. (9.41) into the general equation (9.27) and

replacing the integration with respect to h and /a by integration with
respect to t i and t 2 between the limits to and t , in a manner similar to
the derivation of Eq. (9.30') we obtain
In —
‘'a
If we make the natural assumption of sufficiently rapid decrease of

the correlation function (9.41) with an increase in ti — tz, from Eq.
(9.42) it follows that for sufficiently large values of x
OO Q
D i) (t) = = (0) -h S j i (6)1 e " 2 dO. (9 .4 3 )

0
At the same time we have shown that the asymptotic law D ij(i)
may also be approximately applied to decaying turbulence behind a
grid in a wind-tunnel.
A similar argument may be made for a turbulent wake produced in
a free stream o f uniform velocity by a long cylinder of arbitrary cross
section located along the Oxz axis, or by some bounded solid body
with its center at the origin. As already seen in Sect. 5.9, at a
sufficiently large distance from the wake-producing body, where the
turbulent velocities in the wake become small by comparison with
the free-stream velocity Uo, the turbulent structure in the wake may
be considered self-preserving and different for various values o f the
longitudinal coordinate Xi, only in the length scale L{xi) and the
velocity scale U{xi). Moreover,
_i_
for a two-dimensional wake behind a cylinder, and
for a three-dimensional wake behind a bounded body. In the case

under discussion Xi (t ) » Lôxfor the appropriate choice of the origin
of the time coordinate, and sufficiently large -t; consequently,
for both the two- and the three-dimensional wakes. Since the
turbulent fluid does not leave the wake, at all times a given fluid
particle will wander within the limits of the volume occupied by the
wake, not leaving its boundaries. Therefore, it is natural to expect
(hat for siiiriciently large i, hypothesis (9.40) here will also be

correct with
1 ^
(with = or = respectively).
respectively. However, this indicates that
T
<n
D ,j(,) = u U f ( f ^ / “ - ^ - e '^ ) f 5 ,; ( e ) + 5;,(6)lrfO (9.44)
6 ^
for a fluid particle in a two-dimensional turbulent wake and
In —
2 0
[5o(e) + 5;^(6)]rf9 (9.45)
for a three-dimensional wake where Sij(G) in both cases has the same
meaning as in E q . (9.42). Thus, for sufficiently large values of t
for a two-dimensional wake

1 (9.46)
for a three-dimensional wake.
In the case of a two-dimensional wake behind a long cylinder located

along the 0 x 2 axis, 7 2 (1 ) = 0. In regards to the variable X3 ( t) for a
two-dimensional wake, and the variables Xzit) and Xsiz) for a
three-dimensional wake, at a large distance from a wake-producing
body, the two-dimensional wake becomes practically symmetric with
respect to the plane Oxtxz, and the three-dimensional wake becomes
practically axisymmetric. Thus, one may anticipate that for suffi
ciently large t , they will be close to zero for any initial position of
the given fluid particle.
We now investigate the motion of a fluid particle in a plane or
round turbulent jet which extends along the Oxt axis in a fluid-filled
space. Here also for sufficiently large xi, the statistical state may be
considered completely characterized by the length scale L(xi) and
55 2 S T A TIS TIC A L F L U ID MECHANICS
the velocity scale U (Xi) as was done in Sect. 5.9, where
L{xi)'^xi, U { x i) ^ x x ^
for a two-dimensional plane jet and
Z,(a:i) ~ X i , t/( A :i) ~ x r '
for a three-dimensional, axisymmetric jet. However, in contrast to

the case of a turbulent wake, no basic motion with a constant
velocity U o'û'\ exists in a turbulent jet; therefore, it is impossible
here to assume that A’i(t) f/o-t. However, since the mean flow in
the jet is also self-preserving, one may hope that in this case
hypothesis (9.40), with
{ /(.) = U[ X, ( i ) l L(x) = I[X .( T ) ]
for sufficiently large t will be applicable not only to Lagrangian

velocity fluctuations, but also to the Lagrangian velocity itself
V ( x , to + t). Consequently,
where F, = F, (t]) = const, since Fi(ti) is a stationary random

process. Taking into account that U(xi) ~ x r “ where a = 1/2 or 1,
respectively, we obtain from this that for an appropriate selection of
the time origin
for a two-dimensional je t /o 47 \
___ I _1 J. ^
X ( x ) — x 2 , ( J (z) — X 2^ ^(x ) — for a three-dimensional je t.
Consequently, in both cases, rfTj = — , tj = In —.Furthermore, similar

to the derivation o f Eqs. (9.42), (9.44), and (9.45), we find
In —
[•5iy(^04--S;7(6)]i/e (9.48)
PARTICLE DISPERSION IN A T U R B U LE N T FLOW 55 3
for a tw o -d im e n s io n a l je t a n d
In -
D,j{x)=ûW af (9 .4 9 )
for a three-dimensional jet where -S/y(6) = (■»)). For very

large values of i these equations become th.e asymptotic expressions
for a two-dimensional jet (9 50)

I X for a three-dimensional jet.
The reasoning used for the investigation of fluid particle motion in

turbulent jets is also applicable to the two-dimensional plane and
three-dimensional convective jets over heated bodies, and also to the
turbulent mixing layer between two plane-parallel flows of different
velocities Vi and Uz. As is known from Sect. 5 . 9 , turbulent motion
both in the convective jets and in the mixing layer is self-preserving
for sufficiently large Xi, where the Oxi axis is considered to be
vertical for a convective jet and parallel to the velocity o f both flows
for a mixing layer. The length scale L = L(xi) will be proportional in
all cases to xu the velocity scale V , for a two-dimensional convective
jet over a heated cylinder and of a mixing layer, turns out to be the
same in all cross sections xi = const; for a three-dimensional
convective jet U{xi) ^ x \ ' ' \ It follows from this that the average
longitudinal Lagrangian velocity l/j = in the two-dimensional
convective jet and in the turbulent mixing layer is constant; in the
case of the mixing layer it is equal to (Ut + Uz)l2, and in the
three-dimensional convective jet, proportional to XT''\ Therefore,
hypothesis ( 9 . 4 0 ) , with U(^) = C/[A:i(t)] and L ( t ) = L [ X i ( t ) ] f o r a
two-dimensional convective jet and a mixing layer reduces to the
simple statement that V'{x, to H-1 ) is a stationary random function
of the variable tj — In — . In the case o f a three-dimensional
_
convective jet, for which ®, that is,
55 4 STA T IS TIC A L F L U ID MECHANICS
this hypothesis indicates that the variable
V \x, +
is a stationary random function ofiri = ln — . In the usual way, we

find from this that
In —
.51)
for the two-dimensional convective jet and the mixing layer and
In —
3 3
D,j(z) = ^ U a^ l f (9.52)
for the three-dimensional convective jet. In particular, for very large

values of t
for a two-dimensional convective jet and a mixing layer
3 (9.53)
X2 for a three-dimensional convective jet.
[The asymptotic equations (9.43), (9.46), (9.50), and (9.53) first

appeared in Batchelor and Townsend (1956), Batchelor (1957) and
Yaglom (1965).]
Let us now show that Eqs. (9.50) and (9.53) may be derived also
from purely dimensional considerations without any use of special
hypotheses, if we accept the general Lagrangian similarity hypothesis
which states that the physical parameters on which the Eulerian
statistical characteristics of turbulence depend, also determine the
Lagrangian characteristics; in other words, they completely deter
mine the entire turbulent state. Indeed, according to Sect. 5.9, fo ra
three-dimensional jet of dynamic origin the defining physical
parameters are the fluid density p and the total momentum of the
fluid injected per unit time2jipM; for a two-dimensional dynamic jet.
the density p and the momentum pMi of the fluid injected per unit
time from unit length of the slit; for the mixing layer between two
plane parallel flows, p and the velocity Uq = — Ui, for the
three-dimensional convective jet, p, Cp, the total heat flux along the
jet Q and the buoyancy parameter g/To', for the two-dimensional
convective jet, p, cp. g/To and the specific heat flux Qj per unit length
of the heated cylinder. If, for example, the probability distribution
for displacement Y ( t ) of the fluid particle in the. time t , for a
sufficiently large t , depends only on these parameters and on t as
implied by the Lagrangian similarity hypothesis, then on the basis o f
dimensional arguments the corresponding probability density
p( K) = p{Yi, Yz, Ka) will have the form;
in the case of a three-dimensional dynamic jet;
p O') = 1 ^ - - J ^ ) (9-5''')
in the case of a two-dimensional dynamic jet;
(9.54")
'^0
in the case o f a plane mixing layer;
____ L ___ Y2 Y, \
p< y)= T ^
UpPô/ V lv ^ o / Up?^0/ \CpP ?’o/ /
(9.54"')
in the case of a three-dimensional convective jet; and
,p(S) Y2 >'3 (9.54”")

(9i . ’ ( 9 i ± \ ‘’ 7 19i
V Ta VV^pP ^0 / \^p? ^0 / ^0 / /
for a two-dimensional convective jet. Here ..., are five

universal functions characterizing the five indicated self-preserving
turbulent flows. Expressions (9.50) and (9.53) follow immediately
from Eqs. (9 .5 4 )-(9 .5 4 '"'). In general, asymptotic expressions for
arbitrary moments of the random vector V (x) = (Ki, Y2, F3) are
easily obtained from them. Equations similar to (9.54)—(9.54"'')
may be written for the probabiUty density of the Lagrangian velocity
V (x) other Lagrangian variables; however, we shall
not discuss this here.
Particle Dispersion in a Shear Flow with Constant Velocity Gradient
It is possible to determine the asymptotic behavior of the
variances and covariances D ,,(t) in idealized turbulent flow in the
entire unbounded three-dimensional space such that its Eulerian
velocity fluctuations u' {x, t) are_stationary and statistically homoge
neous, and the mean velocity u{x) does not vary with time but
depends linearily on the spatial coordinates. The latter condition is
necessary so that the field of velocity fluctuations may be considered
homogeneous, since the gradient of the mean velocity has an
essential influence on the structure of turbulent flow; however, in
homogeneous turbulence, the velocity gradient must be constant. Let
us assume that the mean velocity is directed along the Oxi axis and
varies only with respect to the Ox^ direction so that, for example,
i7i = rA's, (72 = «3 = 0 where Y = const. Without loss of generahty, it
may be assumed that at the time t = 0 the fluid particle was at the
point a: = 0. We designate the coordinates of this particle at the time
i asXi ( 0 and its velocity as V {t). In this case,
V { t ) ^ u [ X ( t ) \ + u'[X{t), t\ (9.55)
or in the projections on the coordinate axis
K ,(0 = r;s '3 (0 + W (^ ),

V2{t)=^V2{t), 1/3 (/):= 1/ 3 (/), ( 9 .5 5 ')
and
t
AV0 = / [ I ’^ 3 ( 0 + W ( O l dt,
t t (9.56)
{t) ^ f v : (0 dt. {t) = f w (0 dt.
Usinn these equations, it is not difficult to also express D,j(T) in

terms of the statistical characteristics of the field u '{ x , t), where in
coiistrast to the cases examined above, no special hypotheses are
required concerning Lagrangian self-preservation.
Since u '( x , /) 0, then also V' (0 = 0; consequently, also
X (t) = 0 on the basis of Eq. (9.56). Since the Eulerian velocity
u ' ( x , t) is homogeneous and stationary, the Lagrangian velocity
V (t) will also be a stationary random function; its correlation tensor
will have the form
Vi (^l) Vj {(2) -= ^17 (A -- t'2)-
Now let us investigate the covariance tensor D, i(x) = Ai(T)Xj(T). The

presence of the mean velocity directed along the 0 ,V| axis in no way
influences the particle displacement along the 0 x 2 and Oxi directions.
Therefore, here the variances D22(':), £>33(t) and Dzsiz) take the form
(9.30'), which is usual for homogeneous turbulence, and their
asymptotic behavior for small and large values of t is described by
Eqs. (9.28) and (9.35). This is not the case for the most interesting
components of the covariance tensors D h ( t ) and D isl':). For the
former, we may obtain from the first formula of Eq. (9.56) the
expression
Dn{^) = Xl{.) = f
0 0
+ r [;^3 (î) V,' ((2) + X 3 (h) v l (/,)] + (/, - / 2)} dh.
On the basis of the third equation of Eq. (9.56), the first term in the
braces may here be transformed into the following;
Xi (/,) A3 {(2) == f f Vs (6 ,) 1/3 (62) ddi d h =

0 0
t\ t\
f M3’(61-62) i/0iJO2=j'
00 0
+ f(h -o )B ^\o )d o - f ( | ^ , _ / 2 i_ e )fi(^ » ( 0)rf 0 .

0 0
The second term in the braces is calculated similarly:

-^3 (^1) (^2) + ^ 3 (^2) Vi {t\) =
= f f M3’( ^ .- e ) r f 6 =
0 0
A /, U1-/ 2I
= f Bit* (0) dB + f (6) + f [Mi’ (9) - Ms’ (6)] dO.
Using these expressions, the formula Dn(r) after repeated integration

by parts may be reduced to the form
T T
= y / ( 2 x 3 - 3 x 2 0 + 63) 5 ^ ' ( 9 ) ^/9 + r J'( , _ 6 ) 2 M i ’ (0) dd +

0 0
T X
+ ry*(x2 — 02) M i’(0) + 2 / (x — 0) 5'.i’(9) dB. (9.57)

0 0
Quite similar calculations lead to the following formula forDij:

T T
D,3 (x) = Tx f (x - 9) 5 ^ ' (9) + r (, _ 0) [5 'i’ (9) + B<3?(0)] dB.

“ ® (9.57')
From Eqs. (9.57) and (9.57') the following asymptotic equations are
obtained for large t :
D„ (x) « I . x3 ; D,3 (x)« (9.58)
Thus, the variance of the fluid particle displacement along the Oxt
axis in the direction of the mean flow for large x is asymptotically
proportional to that is, it increases with time significantly more
rapidly than the variances of the transverse displacements, asymptot
ically proportional to t. In addition, the displacements along the Oxi
and 0 x 3 axes turn out to be mutually correlated. Using the
asymptotic equation (9.35) for DasiT), we obtain the following
limiting value for the correlation coefficient between the variables
X,(x) and X3 (x):
= A s ( x ) l D „ ( x ) D 3 3 ( x ) f ^ - > ^ = 0 . 8 6 6 . . . . (9.59)
This limiting value turns out to be universal, that is, it does not
depend on the parameters T, Msând T3 .
The asymptotic equations (9.58) and (9.59) were first reported by
Corrsin (1959b). However, the equivalent expressions were known
even earlier from the semiempirical theory of turbulent diffusion
which uses parabolic partial differential equations to describe the
fluid particle dispersion. We shall consider this problem in the
following section (see Sect. 10.4).
The result by Hogstrom (1964) was quite similar to the first
equation (9.58). He considered particle dispersion in a flow with
strictly constant velocity along the axis OX2 , which makes it
possible to use the coordinate X^, = u i t instead of time t, a linear
profile U i = r X 3 of the component «i, and turbulent velocity
fluctuations u '3 in the direction of the OX 3 axis alone. Moreover,
assuming D 33(X 2 ) = CXl^, Hogstrom found the relation
£>11(^ 2 ) = r"£>33(^2)î/2(a+ 1)
which coincides with Eq. (9.58) if a = 1/2; he then used this

relationship in an attem pt to explain some atmospheric diffusion
phenomena. Later, Smith (1965) [see also the discussion of this
paper] investigated the same problem. He considered all three
components of the velocity fluctuation and gave special attention to
the conditional characteristics for the particles to reach the given
height ^3 at a given time t(= X 2 I U1 ) . However, his results will not be
dwelled upon here.
The Lagrangian Characteristics o f the Turbulent Boundary Layer

Let us now consider the problem of the fluid particle displace
ments in a turbulent boundary layer. As in Chapts. 3 and 4, the flow
will be considered as filling the half-space 2 > 0 where, specifically,
the wall z = 0 will be taken to be dynamically completely rough,
with a roughness parameter Zo. The Ox axis will be taken in the
direction of the mean flow. Without loss o f generality, one may
assume that the given fluid particle at the initial time t = to will be
found at the point with coordinates x = (0, 0, H). Let [X (x),
y(T),Z(T)] be the coordinates of this particle at the time +
and V (t) = li/('t), V'(-t), lF (t)], its velocity at this time. The random
function V (x) obviously is not stationary; thus, for example, for a
sufficiently large time interval 1 , the particle will most probably rise
to a significant height Z ( t ) , thereby strongly increasing its horizontal

velocity U{i). Generally speaking, there is no reason to expect that
the function V (t) may be transformed into a stationary function
with the help of a simple transformation to new scales of length and
time. However, it is natural to assume that in addition to the
parameters x and H, the Lagrangian statistical characteristics of
turbulence in a boundary layer will depend only on a small number
of “external” parameters which determine the turbulent state; that
is, those which enter into the expressions for the Eulerian statistical
characteristics. This assumption, essentially simpHfying the study of
the Lagrangian characteristics, was used in implicit form by
Kazanskiy and Monin (1957) [see also Monin (1959a)] to calculate
the form of a smoke plume in the atmospheric surface layer under
various stratification conditions. Later, it was precisely formulated
and investigated in detail by Ellison (1959) and Batchelor (1959)
[see also Batchelor (1964) and Chatwin (1968)] for the special case
of a neutral or nonstratified boundary layer. Still later, Gifford
(1962), supplementing this assumption by some semiempirical
hypotheses, and Yaglom (1965) derived from it a series of
corollaries concerning the general case of thermally stratified fluids.
Gifford’s conclusions were compared with the existing data by
Gifford himself, Malhotra and Cermak (1963) and Cermak (1963).
Mention should also be made of the works o f Panofsky and Prasad
(1965), Pasquill (1966), E. E. O’Brien (1966), Mandell and O’Brien
(1967), and Klug (1968). These works were all devoted to subsequent
development of the Lagrangian similarity treatment of boundary-
layer flows and its applications to the problem of diffusion in the
atmospheric surface layer. These apphcations will be considered in
Sect. 10.5.
Since the proposition stated earlier includes the statement that the
boundary-layer flow is described by a few parameters, when using it,
it is expedient to limit ourselves to cases where the fluid-particle
motion does not exceed the upper boundary of the constant stress
region in which z = pal = const where i is the shear stress, or in the
case of a thermally stratified fluid of the constant flux region in
which T = const and q = const, where q is the vertical heat flux.
Following Ellison (1959) and Batchelor (1959; 1964), we begin with
the case of a neutral boundary layer for which q = 0 . To be specific,
it may be considered, for example, that we are dealing with the
atmospheric surface layer under conditions of neutral stratification.
In this case, the turbulent state is defined by the parameters u, and
^(1. However, it is essential that outside the very thin fluid layer,
which has a thickness of the same order of magnitude as 2 „, and
directly adjacent to the wall 2 = 0 , only one parameter plays an
important role. Variation in the value of Zo, that is, replacement of it
by leads only to additional horizontal displacement of the entire
mass of tluid along the Ox axis with a constant velocity of
—* I n ~ , where z « 0 .4
is von Karman’s constant (see above, Chapt. 3, Sect. 5.4). In

addition, it is obvious that after a sufficiently large time t the
particle must “ forget” its initial height W, that is, for a large t the
value of N almost ceases to influence the statistical characteristics of
the fluid-particie motion. It follows from this that although the
statistical characteristics of the random vector V (-t) may generally
depend on four parameters, t , / / , u ^ , a n d z o , the effect of the second
and fourth is quite limited. The influence o f the initial height H will
be felt only during a finite time, the duration of which, on the basis
of dimensional considerations, must be of the order of the ratio ///«».
Also, the value of the roughness parameter 2o will really be essential
only if // :< Zo, and even in this case only during a time interval of
o r d e r o r H-.<Zo b u t t ^ then z„ will influence the
statistical characteristics of K (t) only through an additional constant
summand of the form — y lnz(, in the expression for U{i).
Let us now investigate the fluid particle mean velocity
V{ x) = ( ( 7 ( t ) . V ( t) . U 7 (t))
at the time tn+x. The symmetry of the boundary-layer flow with

respect to the Oxz plane implies that K(t) = 0 for all i and
consequently V (t) = = 0. For the components U (t) and W (t),
according to the above when or when x ' ^ Z dIu;,, if // = 0

or, generally speaking, H :< Zo, the second of these components may
depend only on and x, and the first must equal the sum of some
function of and t, and the constant velocity — ^ In z^. On the
basis of dimensional considerations, we obtain
a„, cu ~.
= V.^ I n - — (9.60)
. dZ(i)
bu. (9.61)
dz
where b and c = e'' are dimensionless universal constants. Integrating

these equations with respect to t , we find that for sufficiently large x
- - ^7 I n
, ez.
(9.60')
Z{-z)^bu,x. (9.61')
With an increase in the initial height H, Eqs. (9.60') and (9.61')

become applicable later, but the values of the parameters b and c for
any H remain the same. The fact that the particle experiences an
asymptotically constant average vertical velocity W (-t) in spite of the
zero value of the average vertical Eulerian velocity w ( X ) at all space
points of the flow, so that, in particular, also W^(0) = tiy(0, 0, H) =
0, is related to the inability of the particle to drop below this level
due to the wall at z = 0, while nothing prohibits its unlimited ascent.
Therefore, the probability distribution for Z(x) with an increase in t
extends upward more and more. Consequently, the average displace
ment of the particle Z{x) increases, which means 6 > 0 . It follows
from this that for any layer 0 < z < /? of fixed thickness h we have a
constant outflow of mass to the upper layers of the fluid. This must
be compensated for by an equivalent influx of mass from the upper
layers downward. (Let us recall that for the Eulerian velocity w { X )
the inequality w ( X ) 0 would mean that conservation of mass is
violated!) It seems natural to assume, however, that the velocityU^(t)
will be less than the characteristic value of the Eulerian vertical
velocity fluctuations, for example, less than the value ofo^ = (l^) ^
Therefore the constant b will apparently be appreciably less than
unity; several values in the range 0.1 to 1 have in fact been proposed
by several researchers; see below. An effort may be made to evaluate
the constant c beginning with the approximate equation
(9.62)
where u{Z) is the mean horizontal Eulerian velocity at a height Z.

Let us emphasize that in Eq. (9.62) it is not possible to substitute the
sign of exact equality. This is because the average on the left side is
taken with respect to the set of fluid particles which at the tim e t =
0 are found at the given height H, and on the right side, with respect
to quite another set of all possible fluid particles which regardless of
the time, are at a fixed height Z(t). If the fluid particles in question
have started at a given height H, but then H will have no
significant effect on the value ofU(x). Even in this case, however, the
mean Lagrangia'n horizontal velocity will not be exactly equal to the
mean Eulerian horizontal velocity at a height Z(i), since W{'z) > 0.
Consequently in the expression
u(^) = u [ x { x ) , y(T), z(x)]
averaging mainly considers particles which come to the height Z( t)

from below. In addition, even if it could be considered that
U(^) = i ^ h Y { z ) , z { ^ y \
where for fixed values of X(x), y (t), and Z(t) the average over the
values of u on the right side is the usual average over the values of the
Eulerian velocity, Eq. (9.62) still would not be exact. In fact, «(Z)
increases more slowly than Z, and consequently
Let us also note that both of these reasons must lead to the left side
of Eq. (9.62) being less than the right side. However, it is hoped that
when —, the difference between both sides of Eq. (9.62) will be
relatively small, so that the expression c ^ b which is derived from
Eqs. (9.62), (9.60), and (9.61') could be used as an acceptable or at
least a rough first approximation. Nevertheless, in some cases, it may
be appropriate to consider c as being somewhat less than b. (See
Sect. 10.5 where a semiempirical estimate of c will be given,
according to which c«s 0.66.)
Using dimensional considerations, it is possible also to derive
expressions for the second- and higher-order moments of the vectors
V'ix) - V(x) - V { z ) = iU'ix), K'(x), 1T'(t))
and
X '(^ ) = X ( x ) - X ( ^ ) = {X'{x),Y'{x),Z'(':)).
However, it is simpler to write down immediately the general

equations for the probability densities o f these vectors, from which
the expressions for all their statistical characteristics follow. As we
have seen, when — and —, the probability distribution for
may depend only on the parameters and x. Consequently,
the corresponding probability density must have the form
= (9.63)
\u u u ,
where P i( u ,v ,w ) is a universal function of three variables. In a

similar manner the probability density of the vector X ' i i ) is
determined by a formula o f the type
P(X') = U
4 Xt P21—
\u ' —. -)• (9.64)
* V ♦ t U•X u♦ z /]
where Pi(x, u, z)is another universal function; in particular, it follows

immediately from here that in the logarithmic boundary layer
namely, = where J,y are universal dimensionless
constants. According to Eq. (9.61') the length of M.t in Eq. (9.64)
may be replaced by the length Z = Z (t) proportional to it; therefore,
for example, the probability density for the vector JT(t) may be
written as
i . 5 ^ ) , ,9.65)
where the function P3 is obtained from P 2 by a simple change in scale

on the coordinate axes.
Let us now proceed to the more complex case of the fluid-particle
motion in a thermally stratified boundary layer; for example, the
atmospheric surface layer with nonneutral stratification. Here, in
addition to u,, we must consider among the “external parameters” of
the problem, the parameters qlcppo and gjTo. From these three
variables it is possible to make up a combination having the

dimension of length
which plays an im portant role in the deductions of Chapt. 4. The

roughness parameter zo also in this case will be essential only if
H ^ Z q and simultaneously If H ^ Z q Ot fi:^Zo,hut
then replacing Zo by zj only leads to an additional motion of the
entire mass of air in the Ox direction with a constant velocity
u
_*
HL f (?)-/(?)]
where f{%) is a universal velocity profile function in Chapt. 4.

H
Consequently, when and we must have the expres
sions
(9.67)
where i|3i and at)2 are universal functions of the variable 5 = ^ .

Integrating the second of these expressions, we obtain
(9.68)
where
1'3 (0 = / 1'2 (■^) d-n.
and W is the function which is the inverse of With the help of Eq.
56 6 ST A T IS TIC A L F L U ID M ECHANICS
(9.68), Eqs. (9 .6 6 )-(9 .6 7 ) may be rewritten as
(9.69)
dZ ( X )
d'z (9.70)
where cpi and cp are two universal functions, and the constant b is
introduced so that we may consider (f)(0) = 1; under this condition, h
will have the same value as in Eq. (9.61). On the basis of the
approximate equation (9.62) we may also expect that (pi(?) /(^);
therefore, as a first approximation we may assume
r m - m (9.71)
However, if greater accuracy is desired, it must be considered that

indeed ipi (C) < l i t ) for all t, ; but as yet, there are no data on the value
of the difference / ( ^ ) — (pi(^). Expressions (9.70) and (9.71) were
used by Gifford (1962), and later by Malhotra and Cermak (1963),
and Cermak (1963). In addition, they all propose without firm proof
that the function (p(^) differs only by a constant factor from the
function o f Eq. (7.85) which describes the root mean square
value of the Eulerian vertical velocity. Subsequently, the function
cp(C), according to the semiempirical formula (7.95) of Kazanskiy
and Monin, was assumed to satisfy the equation
Of course, this equation is not exact, but its use may be justified
somewhat by the fact that the condition cp(0) = 1, and the
asymptotic laws describing the behavior of <p(^) as are fulfilled
here.
For the probability density /?(A') of the three-dimensional vector
X{x) = ( X ( x ),y (T ),Z (t)) for and

wc must have on the basis of dimensional considerations, a formula

of the type
f I ) , (9.72,
where ^ = X(x) and Z = Z ( t ) are determined from Eqs. (9.69), or

(9.71), and (9.70), and P (x,y ,z\l,) is a universal function of four
variables, the last of which behaves as a parameter determining the
form of the probability distribution. Gifford, and Malhotra and
Cermak cited above, used in a simpler formula in place of this:
(9.72')
that is, it was implicitly proposed that the dependence of the form of
the probability distribution on the stratification parameter C= y is
quite weak, and to a first approximation, may be neglected.
However, this assumption does not follow from dimensional con
siderations; it is justified only by some very preliminary experimental
results [see Cermak (1963)].
When q >Q, that is, for unstable stratification, it is also possible to
obtain several simple asymptotic results for the variables W{x) and
Z (t) based on the limiting laws of the free convection regime [cf.
Yaglom (1965)]. Indeed, when and x]^> —, and quite
possibly even when 0-MÎ ^ follows from some of the

11^
empirical results o f Sect. 8, Chapt. 4, not only will the initial height
H cease to influence the vertical motion of the fluid particle, but also
the variable«*. In fact, under these conditions, the particle will spend
most of its time in the region of flow in which the condition of
“ pure” free convection is predominant, and which does not depend
on the friction velocity. Therefore if ^ > 0, or L < 0, then with
sufficiently g r e a t p e r h a p s even with —^ > 0.1, the functions
^■id) and i|);i(S) of Eqs. (9.67) and (9.68) must have the following
asymptotic form:
and ~ ( - 1) ’^
In exactly the same way, the function q)(g) of Eq. (9.70) probably
568 STA T IS TIC A L F L U ID M ECHANICS
differs little from its asymptotic expression (p(^) ~ (— even for

—C > 0 . 1 . In this case, when
t> > 0. 1| L | K and
the following relationships will be valid:
- 1“( i ^ * I 0 '
where a is a universal constant the value of which, in principle, may

be determined by experiment. Equations (9.73) and (9.74) are quite
analogous to the equations for Xi(t) and (tJ obtained above for
the two- and three-dimensional convective jets. The right side of Eq.
(9.74) is also very similar to the right side of Eq. (7.87') for Ow, but
this is explained simply by the fact that the variables and H^(t)
have the same dimensions and depend on almost the same dimen
sional parameters, except for replacement of the coordinate z by Z(i).
Consequently, the analogy of Eqs. (9.74) and (7.87') hardly implies
that the coefficients C' and (3/2) will be close in value. On the
contrary, it is natural to expect that the second of these coefficients
will be noticeably less than the first, that is, that a will have a value
of several tenths.
With respect to the mean horizontal Lagrangian velocity
dX(z)
U(X): dz
under free convection conditions, the situation is more complex.

This is because this velocity may not be independent of the
parameter when a* = 0 obviously also t/(x) = 0 . However, it is
clear that the function
in E q . ( 9 . 6 9 ) in this case will approach a constant value of / ( — oo)

with an increase in C, that is, with an increase in r . This is explained

by the fact that with time the particle rises into the fluid layer where
the horizontal velocity is practically constant. For a more precise
evaluation of U{x) the approximate formula (9.71) may be used with
+ const; it is also possible that in some cases it will be
expedient to assume that ?>i(C) = C2C''^*+ const, where C2 assumes a
smaller value than the coefficient C2 in Eq. (7.39). However, it is
natural to assume that the parameter when Z ( t ) > 0 . 1 | L | will
influence only the average horizontal displacement velocity of the
fluid particles, and not the turbulent velocity fluctuations. There
fore, we may expect that for sufficiently large t , for unstable
thermal stratification, we will have the expressions
p { U \ V', W') =
I U' V' W' \
\Cppo To ) \\Cppo T q ) \CpPo To ) \CpPo T q )

U' V' W'
(9.75)
J--Lz
Cp?o T'o
fJLXzV'*’
\\<^ppoô / \<^pPo ^0 /
(J-l-zV'Y
l^pPo ^0 } /
and
p { X \Y ',Z ')= ^ ^ P ,l^ , I). (9.75')
which are related to the expressions (9.54 ) and (9.54 ) which are
satisfied for the turbulent convective jet. From Eqs. (9.73) and
(9.75'), in particular, it follows that asymptotically, as -t-x*,
for unstable stratification or, more precisely, /),/('') =
5^y(^s/^'pPo'to)'^^ where 6,y are universal constants. (It is clear that
all these conclusions assume that the conditions for the validity
of the “ 1/3-power law” for the Eulerian mean velocity and
temperature are retained at least up to the height Z(x).)
9.5 The Lagrangian Velocity Correlation Function and Its

Relationship to the Eulerian Statistical Characteristics
In the previous section, we have seen that when investigating the
motion of a given fluid particle, the Lagrangian velocity correlation
function plays a very important role:
M /’Ui, t2, X) = Vi (Jf, /i) V j {X, h).
In the following section, data relating to this function will be

discussed. Meanwhile, let us note only that these data are quite
sparse and inaccurate because of the absence of reliable methods for
measuring the Lagrangian statistical characteristics of turbulence.
Consequently,. it is worthwhile to discuss at least briefly the
problem of the possible methods of determining these characteristics
theoretically, either directly or beginning with their relationship to
the Eulerian statistical characteristics of the random field u {X, t)
which have been studied more thoroughly.
The similarity considerations of the present subsection can also be
apphed to the statistical characteristics of the relative motion o f two
fluid particles having fixed coordinates Xi and at the initial time
The relative two-particle statistical characteristics are closely related
to the process of relative diffusion of a cloud of marked particles in
turbulent flow, relative to one particle or to the cloud center of
mass, which will be discussed in detail in Sect. 24 of Volume 2 of
this book. Here, we shall restrict ourselves to several remarks
concerning the characteristics of the two-particle relative displace
ment vector l(x) = X 2 {X2 ,to+ \)-Xi{xx,t(y+ 'i) for a boundary-layer
flow. The probabihty distribution o f the random vector/(t) depends
on parameters of the turbulence regime, time t and the initial
coordinates xx and x^, more precisely, on /q - x ^ - x i and zj only.
However, it is natural to expect that the statistical dependence on
the details o f the initial position o f the particles, i.e., on /q and zi,
will be lost if x is sufficiently large. Thus the asymptotic probability
density function and the moments of the vector /(t) as t ->• «> will
satisfy the same equations as the vector = [AT'(t), y'(x), Z'C-t)]; in
particular, in the nonstratified boundary layer, and
////~((jfg/cpPo7’o)'tîn the unstable boundary layer [cf. E. E. O’Brien
(1966) and Mandell and O’Brien (1967)].
Mandell and O ’Brien (1967) have also considered some higher
approximations to the asymptotic results obtained with the aid of
Taylor series expansions of the universal functions introduced in this
subsection. However, the possibility o f Taylor expansion of the
universal functions about the zero value of their arguments does not
follow from dimensional considerations and depends on the selection
of a specific form of the arguments. Therefore, these results cannot
be considered to have the same rigor as purely dimensional
deductions.
The problem of determining Lagrangian statistical characteristics

from Kulerian statistical characteristics will clearly not be simple.
Indeed, from the basic equations
V{x, t), t\, X{x, V{x, (9.76)
which are actually a definition of the variables V ( x , t ) andA'(JC,/).

the Lagrangian variables at time t depend on the values of the
Eulerian field « {X, t ) at all points of the random trajectory X { x , t ' ) ,
to < t ' < t. This trajectory is defined as the solution of a system of
integral equations
X iix , n = - ^ i + f u , \ x { x , t'% i = l, 2, 3, (9.77)
in turn containing the field « (A", /). Therefore, generally speaking,

the probability distribution for each Lagrangian variable will be
dependent on the entire infinite-dimensional probability distribution
for the values of u { X , t) in a function space of all possible vector
fields. Moreover, even the relationship between the function space
distributions of the fields a (A", /)andV(Ar, t), ox u { X , t ) m d X { x , t ) ,
turns out to be very complex and efforts to explicitly describe it do
not lead to results that may be applicable to any specific turbulent
flows [see Lumley (1 9 6 2 b )].
Investigation of the probability distributions in function space
may be avoided, assuming that the functions u, ( A", t),i = 1, 2, 3 are
analytic in all variables and, therefore, may be expanded in Taylor
series. It may be concluded from this that all the Lagrangian variables
may be expressed in terms of the Eulerian velocity and all its partial
derivatives at a single point in space-rime. Expressing A(Jf, Oand
V{x, t) as power series
IS i , , {t - ô)" =
n=0
n=:l
V(x A - V 1 a - n " -
n=i
it is possible using Eq. (9.76) to transform the coefficients of these

series to the form
d ” A -(;c. t )
n = \,2, ...
dt"
'='<- (9.79)
d ^ V i x , t)
d t ”-
[ ^ + u,(X. M X . I)
X = x , /=C„
n = 0, 1, 2, . . . ,
that is, to express them in terms of the Eulerian variables at the point
{x, U). From Eqs. (9 .7 8 )-(9 .7 9 ) some important general conclusions
follow; for example, as was noted by Lumley (1962a), it follows
directly that if the random field u (X, t) is statistically homogeneous,
then the random fields Y{x, t) = X ( x , t) — x a n d V { x , t ) are also
statistically homogeneous', this fact has actually already been used in
Sect. 9.3. However, if we use these formulas to determine the
Lagrangian statistical characteristics, the results turn out to be of
little importance. For the moments of the fields K(jc, <) and V{x, t),
very complex representations are obtained as sums of an infinite
number of single-point Eulerian statistical characteristics multiplied
by various powers of t — to- The use o f such representations for
specific calculations of the Lagrangian characteristics is possible only
for very small values o f t — tu, for which all but the first few terms of
the series (9.78) may be neglected. For example, as applied to the
Lagrangian velocity correlation function, only terms of order no
higher than (t — to)^ have been evaluated successfully, and even for
terms of the order (/ — 4)2, the estimate obtained turns out to be
quite complex.
Only in the case of terms of zero order with respect to t — to, that
is, at the time to, is the situation simple since by definition
V{x, to) = u ( x , ô).Therefore, for example,
M v {to, to, X) = Ui{x, t ^ J l x , ta) = B i J (Co, to\ x).
Thus, for an incompressible fluid and a statistically homogeneous

field u{X, I), and consequently also the field V ( x , /), a similar result
occurs also for all single-point and single-time characterisitcs of the
Lagrangian velocity. Here let us briefly outline its proof according to
Lumley (1962a). In the space of points x, we select the volume R
and investigate an integral of the type
where
( p ( V ) = ^( Vu ^2, Vs)
is some function of three variables. In this integral we transform

from the variables Xu X2, xs to the new variables Xj, X 2, X3.where A’=
X{x, t). In the case of an incompressible fluid, and only in this case,
such a transformation is considerably simplified by the fact that the
corresponding Jacobian
^ (•'^1, ^ 2, ^ 3) IY V VI
d(x,,x,,x,) ^ 2. A 3I ,
as we know, is identically equal to unity [see Eq. (9.6)]. Conse

quently, in this case
f <p(V(x, 0 ) d x = f t))dX, (9.80)
where Rt is the space region which is filled at the time t by the fluid
filling the region to at the time R . In addition, if the fields V (x , /)
and « (X, t) are statistically homogeneous, then the probability mean
values of the functions under the integral sign on both sides of Eq.
(9.80) do not depend on the coordinates. Nevertheless, we cannot
simply average both sides of Eq. (9.80) and take the mean values
of the function (p from under the integral sign. This is because the
region o f integration Rt on the right side is random and depends on
the field a (X t). However, since the space filled by the liquid is
unbounded (otherwise it would not be homogeneous), for any fixed
t we can select as R , for example, a sphere which has as large a radius
as desired. In particular, we can select its radius so large that the
57 4 S TA T IS TIC A L F L U ID M ECHANICS
region Rt differs from R only in regions of irregular shape located

comparatively close to the boundary of R and having a volume
which, with high probability, will not exceed a sufficiently small part
of the total volume of the sphere R. Thus, with very high
probability, we shall introduce only a very small relative error by
replacing the integral with respect to Rt on the right side of Eq.
(9.80) by a similar integral over the fixed region R. Now, averaging
both sides of the approximate equation obtained and dividing the
result by the volume R, we find
= t)), (9.81)
where we write the sign o f exact equality, since here the relative
error may be made as small as desired by selecting a sufficiently large
region R. In particular, if (j)(V) = then (p(V(ji:, 0 is the
characteristic function of the Lagrangian velocity V(x, t), and
<p(« (X, t) is the characteristic function of the Eulerian velocity
X{Xt t). Therefore, for homogeneous turbulence in an incompressible
fluid, the probability distributions for the Eulerian and the Lagran
gian velocities will coincide with each other at all times. Thus, in
homogeneous turbulence, paradoxes of the type encountered in the
case of a boundary-layer flow where
Uz{x, t ) ^ Q , but V^{x,t)==hO
are impossible.
Clearly, all o f our arguments refer only to one-point and one-time
probabihty distributions; for two-time probability distributions, the
statement in italics will, generally speaking, be invalid. Therefore, for
the Lagrangian velocity correlation function Jf) we may
conclude only that for homogeneous turbulence
{t, t\ X ) = B i j ( 0 = - t i i { X , t ) U j { X , t),
but nothing follows regarding the dependence of B^'l'^ t^) on

4 — ti. However, if the velocity field u { X , not only homogen
eous but also stationary, then the function V(jc, /)will be stationary
with respect to we have actually also used this circumstance above.
Indeed, representing
V j(x,t^'z)Û j[X {x,t+ x ),

in the form of a Taylor series (9.78)-(9.79), replacing to by t, and x

by/ + T, it is possible to write the correlation function as the sum of
single-point Eulerian characteristics evaluated at the Lagrangian point
and multiplied by powers of t. Therefore, if all the
Eulerian single-point characteristics do not depend on t due to the
stationarity and homogeneity of the field u {x, t), then
Equation (9.81) was proved above for the idealized case o f

stationary homogeneous turbulence in unbounded space and is
approximately valid for atmospheric surface layer turbulence; cf.
Lumley and Panofsky (1964), pp. 142-144. However, an equality of
the same type may also be established for a turbulent flow in an
infinitely long straight tube. Thus it is necessary only to take as ^ a
sufficiently long segment of the tube between the cross sections
xi = a and Xi=b. For any fixed -c=/ — we can select b — a so
large that the region Rt differs from R only by small irregularly
shaped regions close to the edges x i = a and Xi — b, the net volume of
which is very small compared to the total volume R. After dividing
both sides of Eq. (9.80) by the volume R and averaging, we obtain a
value on the right side which is essentially the volumetric average of
the function (p(«(A', O)over the cylindrical volume bounded by the
cross sections xi = a and Xi = b and the tube walls. On the left side
will be the variable q>( V(jf, /)) which for sufficiently larged — to may
generally be considered as independent of x (compare with the end
of Sect. 9.3). In particular, with <p (u) = Ui in this way we obtain
Vi = (Ja.v which was discussed before Eq. (9.37).
Let us again consider the simple case of homogeneous and
stationary turbulence. In this case u = U = const, and w ithout loss of
generality, one may even consider, as we shall do in the future, that
U = 0 ; this is equivalent to a simple transformation to a new inertial
system of coordinates. If B f j { x ) = Vi { x , t ) V j { x , / H-t) is the
Lagrangian velocity correlation tensor of this turbulence, and
E2. 13, t) =
= Uj(Xi, X'l, X3, t)u j{Xi + ^ 1, X2 + ^2, X3 4- Is, t + t)
is the time-space Eulerian correlation tensor, then as we know
Mv {0) = Bij (0, 0, 0, 0). However, the values of for t > 0,
generally speaking, cannot be expressed in terms of the values o fB a
57 6 STA T IS TIC A L F L U ID M ECHANICS
(si, I 2, h , T).Due to the absence of reliable data on the function

Bfji'z), some investigators have tried to assume that =
Bij (0, 0, 0, x) [see, for example, Baldwin and Mickelsen (1962) who
also compared the results of this assumption with experim ent];
however, this assumption has no theoretical basis. It is more natural
to assume that the Lagrangian correlation coefficients
[not summed with respect to /!] which describe the correlation

between the /th component of the velocity of the given fluid particle
at time t, and at t+ x will decrease with an increase in t more slowly
than the Eulerian correlation coefficients
/2
«;
describing the correlation between the velocities of the various fluid
particles which at time t and at time t + t turn out to be at the same
point in space. If this is so, then when > 0, the following inequality
t
must be fulfilled for all t = 1, 2, 3:
From the last inequality, in particular, it follows that the

Lagrangian correlation time, or the Lagrangian time macroscale, or
the Lagrangian integral time scale
must be greater than the Eulerian correlation time, or time

macroscale, or integral time scale
71= f
and the Lagrangian time microscale, or differential time scale
0 ,= - 2/ { 3? x=Oj
ill exactly the same way must be greater than the Eulerian time
microscale, or differential time scale
Unfortunately, a direct test of these conclusions is not yet possible

due to the absence of data from simultaneous measurement of the
Eulerian and Lagrangian time scales under conditions close to
homogeneity and stationarity. However, as Corrsin (1963) noted,
comparing the measurements of the Eulerian time-space velocity
correlations by Favre, Gavigho and Dumas (1953) in a wind-tunnel
behind a turbulence-producing grid, with data from the diffusion
experiments of Uberoi and Corrsin (1953) carried out under similar
conditions, makes it reasonable to assume that under such conditions
0;
« 0.7. When the mean velocity \u\ — U is different from zero and
comparatively large, the Eulerian correlation coefficient
Rii{z) = 0, 0, x)lu'i^
may be expected to decrease, with an increase in x, considerably

faster than when U = 0. The former describes the relationship
between the velocities of fluid particles which are carried with great
velocity at times t and t + z past a given space point, that is, which
are actually at large distance from each other. Actually, assuming
that
/?22(x) = M2’(px) (9.82)
where the direction of the 0 x 2 axis coincides with the horizontal

direction perpendicular to the mean wind. Hay and Pasquill (1959)
found from the data on horizontal diffusion in the atmosphere that
in the first, quite rough, approximation, p ~ 4. Subsequent investiga
tions in this area by many researchers [see, for example, Lumley and
Panofsky (1964), pp. 197—201; N. Thompson (1965); and Haugen
(1966)] show that the empirical estimates of P are extremely
scattered, but confirm its mean value [see below, following Eq.
(10.11)]. From Eq. (9.82) it follows that
7’2 = p7'2 and O2 — PO2 ;
therefore, the value of p = 4 corresponds to the case in which the

578 S TA T IS TIC A L F L U ID M ECHANICS
Lagrangian time scales significantly exceed the Eulerian time

scales.
Equation ( 9 . 8 2 ) , concerning coincidence of the form of the
Lagrangian and Eulerian time correlation functions, is frequently
used in investigations of turbulent diffusion. It may be considered
permissible only because the exact form of the functions Rfi (i) in
many cases turns out to be not particularly important; see, for
example, the discussion following Eq. ( 9 . 3 6 ) . In fact, there is no
theoretical foundation for this proposition. Moreover, in Chapt. 8,
Volume 2 o f this book, it will be shown that for very large values of
the Reynolds number
for ^ V T i < ^ r < ^ 7 ' r ,
while the Eulerian time correlation function /?,, (t) for very large Re
and t which were satisfied in the experiments of Hay and
Pasquill, for such t has the form
« 1 — C'iX^
but again is linear with respect to t i f « = 0 ; compare Corrsin ( 1 9 6 3 ) ,

where from the dependence of the various time-scale ratios on the
R eynolds, number is evaluated approximately. Thus, when
Eq. ( 9 . 8 2 ) cannot be satisfied for all t. In general,
nothing can be said concerning the behavior of the Lagrangian and
Eulerian correlation functions for values of t which are comparable
to the corresponding correlation times. Therefore, the theoretical
determination of the form of the functions (t) is possible only
with special purely empirical or semiempirical hypotheses.
S a ffm a n ( 1 9 6 3 ) m a d e a n a p p r o x im a te c a lc u la tio n o f th e values o f th e fu n c t io n

for a s p e c ific fo r m o f th e E ulerian tim e-sp a ce c o r r e la tio n fu n c tio n R i j ( ^ \ , ^2 , 5a, t ) ,
assu m in g th a t in e x p r e s s io n ( 9 . 2 2 ) th e v e lo c ity V j ( X , JC, t) fo r all t m ay b e r ep laced b y th e
E ulerian v e lo c ity Uj( X, t). H o w e v e r , sin c e w e hav e already n o te d th a t th is a s su m p tio n is
u n fo u n d e d , w e shall n o t d iscu ss th is fu rther. I n o u e ( 1 9 5 0 - 1 9 5 1 ) and O gura ( 1 9 5 2 b )
in v e stig a te d a m o d e l o f tu r b u le n t m o t io n in th e fo r m o f a set o f d istu rb a n c es o f th e v e lo c ity
fie ld o f various scales in w h ic h th e Lagrangian c o rr e la tio n fu n c tio n s y?[A)(x), / = 1, 2, 3, all
tu rn ed o u t to be given b y th e fo llo w in g universal form u la:
(t) = 1— ^ fo r 0 < t < 2 7 ’,. an d /? W (x )= 0 for t> 2r„
H o w e v er , I n o u e (1 9 5 2 ; 1 9 5 9 ) later c a m e t o th e c o n c lu s io n th a t a n o th e r m o d e l, in w h ic h
^ c o n v e n ie n t. G rant ( 1 9 5 7 ) used a sp e cia l m ix in g -le n g th h y p o th e s is

PAR TICLE DISPERSION IN A T U R B U L E N T FLOW 579
lo c a lc u la te the fu n c tio n and o b ta in e d a m o r e c o m p le x , b u t also universal,

forin u la
M / ' 4^ 4^ fo r O < T < 4 7 '^ and (x ) = 0 fo r t > 4 7 ’,.
T he sam e r esu lt w as th e n o b t a in e d b y M a tsu ok a ( 1 9 6 0 ) w ith a slight m o d ific a tio n o f I n o u e ’s

first m o d e l m e n t io n e d a b o v e . A n o th e r m o d e l o f tu r b u le n c e w as p r o p o s e d b y W andel and
K o fo e d -H a n s e n ( 1 9 6 2 ; 1 9 6 7 ) . In th eir m o d e l, a un iversal fo r m u la w as n o t o b ta in e d for th e
Lagrangian corr e la tio n f u n c t io n , b u t th is f u n c tio n tu rn ed o u t t o b e e x p r e s s e d q u ite sim p ly
via th e E ulerian tim e c o rr e la tio n fu n c tio n . S till a n o th e r h y p o th e s is fo r th e rela tio n b e tw e e n
E ulerian a n d L agrangian characteristics w as p r o p o s e d b y Philip ( 1 9 6 7 ) . O th er rela tio n s
f o ll o w also fr o m a s im p lifie d v o r te x m o d e l o f tu r b u le n c e as a s y s te m o f circular e d d ie s
s u p e r im p o s e d o n th e m e a n f lo w [c f. fo r e x a m p le , J o n e s ( 1 9 6 6 ) ] . I n d e e d , h o w e v e r , th e
Lagrangian corr e la tio n fu n c t io n d o e s n o t h ave th e sam e fo r m in all h o m o g e n e o u s an d
sta tio n a ry tu rb u le n t f lo w s an d m a y n o t b e u n iq u e ly ex p r essa b le via th e E ulerian tim e or
sp ac e-tim e co rr e la tio n f u n c tio n . T h e r e fo r e , all th e resu lts e n u m e r a te d here m a y b e
c o n sid e re d o n ly as a p p r o x im a tio n s t o th e real c a se , th e a c cu ra c y o f w h ic h still c a n n o t b e
e stim a te d d u e to th e a b s e n c e o f reliable data.
10. TURBULENT DIFFUSION

10.1 Problem of the Description of Turbulent Diffusion
In the preceding section the motion of individual “ fluid particles”

in turbulent flow was investigated. By “ fluid particle” we mean here
simply a volume o f fluid which is so small that within the framework
of continuum theory it may be identified with a point moving along
with the surrounding fluid. However, it is clear that the results
obtained cannot be compared with observation if there are no
particles in the fluid which are somehow “ tagged” so that their
motion may be followed. In other words, the theory developed in
the preceding section will be relevant only when certain elements of
the fluid are marked; that is, have some properties which are
distinguishable from the properties of the surrounding fluid. This
difference in properties will occur most frequently due to a
difference in chemical composition, or to the presence in the
individual volume elements of “ foreign substances” which differ
from the fluid itself; however, it may also be as simple, for example,
as a temperature difference. In any case, to provide “ tagged”
particles we shall say that there is some admixture in the flow. Here,
by admixture we mean those particles of the fluid having special
properties which permit their motion to be traced.
If we introduce an admixture only into specific parts of the
turbulent flow, then as a result of its transport in disordered
580 S T A TIS TIC A L F L U ID M ECHANICS
intermixed filaments, the ensemble of which constitutes such a flow,

it rapidly spreads to the entire volume occupied by the fluid. This
phenomenon, called turbulent diffusion, is the definitive character
istic of turbulent flows; it is not by chance that in Reynolds’ classical
experiments the occurrence of turbulence was defined precisely as
the phenomenon whereby, for the addition of a small quantity of
dye, the entire fluid rapidly becomes colored. It is clear that in
addition to turbulent diffusion the admixture usually will participate
also in molecular diffusion not related to turbulence. However, this
process is comparatively much slower and therefore in the turbulent
flows, it plays only a relatively small role. Turbulent diffusion
dominates such all-important and well-known phenomena as the
spreading of plant pollen in the atmosphere, the spread o f bacteria
and viruses, radioactive substances, volcanic ash and oceanic salt, air
pollution especially in cities due to industrial and transportation
smoke and gases, the transfer of moisture evaporated from the
surface of the earth, and from all types of water reservoirs, the
dispersion of floating objects on the surface of water reservoirs, etc.
Thus, it is not surprising that turbulent diffusion has been studied
widely; see, for example, Sutton (1953), Frenkiel and Sheppard
(1959), and Pasquill (1962b), the review articles by Batchelor and
Townsend (1956), Ellison (1959), Monin (1959c), and others.
For the marked fluid particles to be taken as “fluid particles” in
the sense of Sect. 9 above, it is only necessary that the marking
admixture be passive, or have no influence on the fluid motion, and
move with a velocity that is practically coincident with the
instantaneous flow velocity at the corresponding point. In particular,
it follows that the admixture particles must be sufficiently fine, or
smaller in Hnear dimension than those distances over which the
velocity u (X,t) may change appreciably, and so close in density to
the surrounding fluid that neither gravitational settling of the
admixture nor its buoyant floating up will be significant. Even here
the admixture particles still can not be completely identified as ideal
“fluid particles.” Indeed, any admixture may be dispersed also by
molecular diffusion or Brownian motion due to the thermal motion
of the fluid molecules while the molecular motion has no influence
on the “ fluid particles,” which are actually “mathematical points” of
a continuous medium subject to the fluid dynamic equations.
Although this is discussed in more detail in Sect. 10.2, it is not of
primary importance and in most cases may be neglected completely.
If the admixture particles may be observed individually, then it is
possible by tracing their motion, to determine the individual

Lagrangian trajectory X = X{t). Attempts may then be made to
evaUiate the Lagrangian statistical characteristics of turbulence by
averaging the data obtained from a series of such trajectories. This
method has gained wide popularity in meteorology in connection
with the use of the so-called no-lift balloons, of specially selected
weight which enables them to float in the air without going up or
down, and the usual air balloons. [See, for example, Pasquill (1962b)
where references to the original work may also be found. Experi
ments with other individual tracers, e.g., soap bubbles, thistledown,
etc., have also been carried out; cf. the surveys by Pasquill (1962b),
and Lumley and Panofsky (1964).] However, the results thus
obtained permit only very preliminary estimates of the Lagrangian
velocity correlation function and other Lagrangian characteristics of
atmospheric turbulence. This is because the tracers usually can only
be followed for a short time and frequently do not follow the air
motion precisely. In this last respect, the balloons are most suspect.
As a rule, the admixture is introduced into the flow in the form of
a fluid or gaseous additive, or as a large number of fine soUd
particles. Usually, it may be considered with complete confidence as
continuously distributed in space and characterized by the Eulerian
field of volumetric concentration ^ ( X ,t ) . (In a compressible fluid a
more convenient characteristic would be the mass specific concentra
tion 0/q, but here we shall consider only diffusion in an incom
pressible fluid.) By the description of turbulent diffusion we shall
understand the statistical description of the field for given
initial and boundary conditions including assignment of all the
sources of the admixture. In the presence of admixture sources the
concentration field {)'(A",/) will, generally speaking, be inhomogen-
eous,_and its mathematical expectation, the mean concentration
ii( 'X, t), will be some function of X and t. The determination of this
function is the most important, although not unique, problem of the
theory of turbulent diffusion.
To describe turbulent diffusion one may begin with the fact that
in each realization of turbulent flow the concentration field O’(A", ^)
in regions not containing admixture sources, satisfies the molecular
diffusion equation
with the given boundary conditions. Since the admixture is passive,

that is, the field « does not depend on the variable i[>, Eq. (10.1) is
linear with respect to 0. As a rule, the boundary conditions are also
linear with respect to usually they have the form
Z ■ £ + ?» = / ( ' ) . 0 0 -2 )
where n is the normal to the boundary, and p is some constant. For

solid walls which bound the flow, the boundary conditions are
homogeneous, namely, — here the value of |3=oo corresponds
to the wall completely absorbing the admixture, 0 corresponds
to the wall being completely impenetrable to the admixture, and
values of 0 < p < oo correspond to the case of partial absorption and
partial repulsion of the admixture at the boundary. In the recent
works o f W. Feller, A. D. Wentzel and several other mathematicians
on the general theory of Markov random processes, more general
boundary conditions or, in a certain sense, the most general ones,
have been investigated. These conditions consider the possibility of
the temporary stopping of the admixture as it reaches the boundary,
and its diffusion along the boundary [see, for example, Dynkin
(1965)]. However, since effects of this type have almost no real
significance in the spread of admixtures, we shall not discuss the
corresponding boundary conditions which are also linear. For a flow
which is unbounded in any direction, the boundary conditions at
infinity are usually taken in the form of the requirement # ^ 0 ; that
is, they again have the form of Eq. (10.2), w ith /(0 — 0 and P — oo.
The instantaneous sources of the admixture are usually described by
initial conditions for d (A ’>0; inhomogeneous boundary conditions
of the type (10.2) with f{t) - ^ 0 correspond to continuously active
sources. (More details on these conditions for various types of
sources will be presented below.)
For homogeneous boundary conditions the evolution of the field
■&{X,t)in the given region will be determined exclusively by the
transport of the admixture by the velocity field u { X , t ) and
molecular diffusion. In principle, the velocity field is uniquely
determined by the initial field u {X ,to) = Uq{X) with the help of the
fluid dynamic equations. Consequently, the solution of Eq. (10.1)
for the given initial concentration field /q) = ô(-^) may be
written as
(10.3)
where A is some operator which depends on the initial velocity field

«,i(A'), the parameter t and the form of the boundary conditions. Let
us emphasize that due to the linearity of Eq. (10.1) and the
boundary conditions, this operator is linear.
For a statistical description of turbulence the initial velocity field
« o ( ^ ) i s considered as random, that is, it is assumed that some
corresponding probability distribution exists in the function space of
its realizations (all possible solenoidal vector fields). However, in this
case, the operator/I[«o(-^), <]. which depends on a random field
tt^{X), will also be a random operator determined by some
probability distribution in the space of the linear operators which are
permissible reahzations of the operator A. Consequently, for a fixed
the concentration field ^ ( X . ^)when i > to will be random,
since it depends on A , that is, on Uo(X). Here the mean value of the
field /)will be defined by the equation
^ ( X , t) = A\Uo{X), (10.4)
where the operator A{t) = A \Uo ( X ) , t] is obtained from the random

operator i4[»o(^)> i] by means of probabihty averaging, or integration
with respect to the probability measure given in the space of the field
Ko(A’). Since the operators A[ tta{X), /] are linear for any Uo{X), and
averaging is a linear operation, the resultant mean operator iT(/) is
also linear. Thus, for a fixed initial concentration field A (t) the mean
concentration 0(A’, satisfies some linear equation. Using Eq.
(10.3), equations may also be obtained for the correlation function
and for higher-order moments o f the field Oo(X); however, these
equations will be nonhnear and much more complex, and, moreover,
they are rarely investigated [cf., however, Csanady (1967a,b) where
the variance of ■0'( A’, Ois studied]. Most of the existing methods for
describing turbulent diffusion reduce to the construction of the
linear equation for the mean concentration d(A', /)only or, equiva
lently, to finding the theoretical, semiempirical or purely empirical
linear operator J ( / ) .
When it is assumed that to determine the mean concentration we
may neglect molecular diffusion by comparison with turbulent
diffusion (the conditions under which this assumption is valid will be
investigated in the following subsection), the operator may be
represented in much more specific form than has been done
previously. Indeed, let us first investigate the case where at the initial
time / = to, the entire admixture (of quantity Q) is concentrated in
one “ fluid particle” which is at the point x. In this case
Since molecular diffusion is negligible, the admixture wUl at all times

remain in the same fluid particle as at the beginning; that is, at time t
it will be at the point X { x J ) . Consequently,
{^(X. t ) = Q6[A-— A'Cat.O].
Considering that
t ) \ = p { X \ x , t),
the probability density for the coordinates of the indicated “ fluid

particle” at the time t, we obtain
A { t ) b { X — x ) ^ p { X \ x , t).
Due to the linearity of the operator Z (/), for an arbitrary initial field
#o(A") we can now use the principle o f superposition; consequently,
it follows that
= = / P { X \x , t ) h { x ) d x . (10.5)
Thus, finding the mean concentration O’(A",/), while neglecting the

molecular diffusion, reduces to determining the probability density
for the coordinate X ( x , t ) o f one “ fluid particle.” Let us note that
the function p{ X x, t), on the basis of E q. (10.5), may itself be
interpreted as the concentration field d ' i X j ) , for the presence at the
initial time t = to at the point x of an instantaneous point source of
admixture o f unit concentration. Thus derives the possibility of
empirically determining p { X x , t) from the data of diffusion
experiments which was mentioned in Sect. 9.3.
If the operator A{t) is known, that is, if it is known how to
calculate the mean concentration field 0( A”, /) corresponding to a
given initial field ), it is also easy to determine the mean
concentration corresponding to the different types of admixture
sources encountered in practice. Again, let us examine, for example,
the case of “ purely turbulent” diffusion defined by Eq. (10.5) where
the velocity field u ( X , t ) will, for simplicity, be considered
stationary; this corresponds to steady turbulent flow. In this case,

the probability density p { X \ x , U + t ) for the coordinates of the
“ fluid particle” at the time to + x under the condition that at the
time h it was at the point x , may conveniently be designated by the
symbol p i { X \ x , t ) ; then in contrast to p\, the function p will not
depend on the parameter U. The mean concentration from the
instantaneous point source of output Q, i.e., producing Q units of
mass of admixture, will equal Q/?i (A’ljt, t — to). For a continuously
active stationary point source at the point x o f output Q, that is,
producing Q units of mass of the admixture per unit time, the mean
concentration ^ { X , t ) = ^{X), will not depend on time and will be
given by the equation
In practice it is often necessary also to deal with diffusion in a flow

having constant mean velocity u = U directed along the OXi axis; in
this case, it is convenient first to transform to a system of
coordinates which moves together with the mean flow. Here the
mean concentration from the continuously active stationary point
source at a point JCassumes the form
^ { X ) = Q f Pi (Xi - Ux, X.,, x ) d -, (1,0.6')
where pi{Xu X 2, XsIjc,t) is the probability density for the coordi

nates of the fluid particle in the absence o f a mean velocity. In the
case of a linear source located along the OX^ axis, the mean
concentration 0(A', /) is obtained by integration, with respect to^:2,
of the concentration from the point sources at (0, X2, 0). Conse
quently, for stationary turbulence, which is homogeneous in the
direction of the OX 2 axis, and in the presence of a mean flow velocity
U along the OXi axis, the mean concentration from an instantaneous
linear source on the OX2 axis will have the form
;? 3, o = Q j p y { X , - U z , x , , X 3 I0 , t ~ t o ) d X 2 =
—00
= Q p ^ X ^ - U z , X^\Q, t - t o ) . (10.7)
Here Q is the amount of admixture produced per unit length of the

0 X 2 axis and is considered constant, that is, independent ofA' 2, and
p,(A '„X 3|0. t) = fp,(X ^,X 2, m x)dX2
is the two-dimensional probabihty density for the coordinates and

Xsof the fluid particle at the tim e/o+x, under the condition that at
the time 4 it was at the point x = 0 , in the flow with a zero mean
velocity. In exactly the same way, for a continuously active
stationary linear source on the OX 2 axis of output Q, that is,
producing Q units of mass o f admixture per unit time for each unit
of length, the mean concentration d ( X i,^ 3) equals
X , ) = ^ Q f p , ( X , - U z , ^^310, x)d^. (10.8)
Using equations for the moments X/ and X , X j o f the coordinates

of the “ fluid particle” presented in the preceding section, we can
now express via the Lagrangian characteristics of turbulence the
values of
f A',0 (X, t) dX, J X i X ^ J x 7 T ) d X
and other similar quantities. On the other hand, these q uantities may
be determined also using the mean concentration ^ { X , t) found
from diffusion experiments. This makes it possible to obtain definite
empirical information about the Lagrangian characteristics. As an
example, let us consider the simplest case of a continuously active
linear source on the OX 2, axis in a flow with mean velocity U which
substantially exceeds the typical values o f the fluctuating velocity. In
this case the basic contribution to the integral with respect to d t o n
the right side of Eq. (10.8) will be made by the values of t close to
Therefore, with high accuracy we may assume here that
b{X„ x ,) = cox^si-p,[x,\0, ^ ) , (10.9)
where
OO
A(;^3|0, t ) = / a ( A 'x , ^ 3 |0 , ^)dX,
is the one-dimensional probability density for the coordinate X 3 of

the fluid particle. On the basis of Taylor’s equation (9.31) we may
obtain in particular that for homogeneous turbulence with
Xi/[/
:lX3(Xi)l^==2u!j' (10.10)
f » ( ^ l . ^ 3) dX,
where, as in Sect. 9,
«3 Ms’ (x) = K 3 (0, 0 V's (0, / + t) .
Consequently, the average width of the wake behind a stationary

Unear admixture source, which naturally is identified with(X|)'/»), at
small distances from the source, increases in proportion to this
distance since
x i for x , < ^ u T ^ = u j y?î>(T)dx.

0
This is well confirmed by the data from the diffusion experiments of

Kalinske and Pien (1944), Uberoi and Corrsin (1953), Townsend
(1954), and others. Similarly, it has been shown that under the same
conditions the wake behind a stationary point source must first have
the form of a cone, and then a paraboloid of revolution. This has also
been well known for a long time from laboratory experiments and
from observations of smoke plumes from factory smokestacks in a
strong wind [cf. G. 1. Taylor (1921) and also, the more recent experi
ments by Orlob (1959)].
According to Eq. (10.10),
cPxl
dX\
SO that in principle it is possible also to obtain the function (t)

from the empirical values of X\ (Xi). Indeed, however, the accuracy
of this determination o f Rzz (x) turns out to be extremely low. This is
due to the necessity of differentiating the experimental curve twice,
which, in practice, is always quite poorly defined. Nevertheless, some
researchers, for example, G. I. Taylor (1935b), Collis (1948), Barad
(1959), and Panofsky (1962) have tried to obtain in this way at least
approximate values of the Lagrangian velocity correlation function.
Other investigators, specifically, Kalinske and Pien (1944), Uberoi
and Corrsin (1953), Mickelsen (1955), Hay and Pasquill (1959),
Baldwin and Mickelsen (1962), Hogstrom (1964), N. Thompson
(1965), and Haugen (1966), used the measurements of the variable
X i in the wake behind a stationary linear or point source primarily
for obtaining more detailed information on the Lagrangian statistics,
and to estimate the Lagrangian integral time scale
7- = / / ? ,
Thus, assuming that the Eulerian and Lagrangian correlation func

tions of the lateral wind component are similar to each other. Hay
and Pasquill, by this method, deduced the crude estimate p 4 for
the corresponding proportionality coefficient p in the atmospheric
surface layer [see text following Eq. (9.32) where this was already
mentioned]. Kalinske and Pien (1944), and Uberoi and Corrsin
(1953), made laboratory measurements of the values of the function
^3 (Xi) in the wake behind a stationary linear source. From this they
estimated the time scale T and some other parameters of the function
and then used these estimates to refine the values of Riz
found by Eq. (10.11); however, even then the form of the function
^M’(T)was poorly defined. Let us also note that in most of the
experiments mentioned, the measurements refer to the diffusion of
heat (Taylor, Collis, Uberoi, and Corrsin) or of a foreign gas
(Mickelsen, Barad, Thompson, Hauger) in air, so that neglecting
molecular diffusion here requires special justification, which is absent
in all of these works. Similarly in the experiments of Hay and
Pasquill, and also in those of Hogstrom where the atmospheric
diffusion of lycopodium spores and of smoke were investigated, the
Brownian motion of the particles could, in principle, play some part.
In the Kalinske-Pien experiments where diffusion of fine drops of a

liquid admixture in water (of the same density as the water) was
studied, molecular diffusion could hardly have had any influence,
but the assumption of homogeneity of the turbulence appears not to
have been very accurate in either case.
As already indicated in Sect. 9.3, for homogeneous turbulence
there are substantial reasons for assuming that the probability
distribution for the coordinates of the “fluid particle” for any
diffusion time will be quite close to a Gaussian distribution. In other
words, if we take as coordinate axes the principal axes of the
covariance tensor of fluid-particle displacement
then for any t we may take
P\ '^) ~ 3 X
(2«)2 [ D „ ( t ) D2, ( x) D33(t)]'/*
V r v n I - * 2) ^ (-^3 -*^3) * ) 1n 1^1

x ^ ^ p t— 2 d ; t ^ ) -------------2 5 ^ 7 ^ ^ —
where the variances D,f ( t ) are expressed with the help of Eq. (9.31)
in terms of the normed Lagrangian velocity correlation function
Substituting this equation into the general expressions
(10 .5 )-( 10.9), under the not always completely justified assumption
that the direction of the mean velocity coincides with one of the
principal axes of the tensor D,j(T), it is possible to make these
expressions substantially more specific, to consider their asymptotic
behavior for various limiting cases, to obtain for some of them
simpler approximate formulas, and to carry out detailed calculations
for specific choice of functions R u \ t) [Frenkiel (1952; 1953),
Fleischman and Frenkiel (1954); compare also Hinze (1959), Sect.
5.5 and the beginning of Sect. 10.4 of this b o o k ].
O ft e n th e a d m ix tu r e m a y b e c o n s id e r e d n o t o n ly as a s u b s ta n c e c o n t in u o u s ly d istr ib u te d
in s p a c e, b u t a lso as a set o f disc re te particles; s o m e tim e s th is la tter v iew turns o u t t o b e
e v en m o r e c o n v e n ie n t. E ven th o u g h w e shall n o t use such a m o d e l in th e fu tu r e , for
c o m p le te n e s s let us pr e se n t h e r e s o m e results w h ic h t o u c h o n th e m a th e m a tic a l d e s c r ip tio n
o f th e fie ld o f a d isc re te a d m ix tu r e . L et us d e s ig n a te th e to t a l n u m b e r o f a d m ix tu r e particles
as TV, a n d th eir to ta l m ass w ill b e ta k en as u n ity . We shall c o n s id e r particles w h ic h d o n o t
d iffe r fr o m e a ch o th e r a n d are sta tistic a lly e q u iv a le n t in th e sen se th a t fo r a n y N, a n y
spatial reg io n s Vi , . . a nd a n y various n u m b e r s /i, 1 ^ th e
probability
w here X d e s ig n a te s th a t th e p o in t X b e lo n g s t o th e reg io n V, d o e s n o t d e p e n d o n th e
n u m b e r s i\, . . U . F r o m this it f o ll o w s , in particular, th a t all th e r a n d o m v e cto rs X i are
id e n tic a lly d istr ib u te d a n d th a t
is a s y m m e tr ic f u n c tio n o f t h e r eg io n s l/i, . . K*.

The d is tr ib u tio n o f th e d isc re te a d m ix tu r e at a f ix e d tim e w ill b e ch a ra cterized
c o m p le t e ly b y th e r a n d o m f u n c t io n o f th e r eg ion m-(K), th e v a lu e o f w h ic h is e q u a l t o th e
m ass o f th e a d m ix tu r e c o n ta in e d at th e giv en tim e in th e spatial region V. T h is r a n d o m
fu n c tio n p la y s th e sa m e role h ere as th e r a n d o m fie ld -O’(A') fo r a c o n t in u o u s ly d is tr ib u te d
a d m ix tu r e , m o r e e x a c t ly , as an a n a lo g o f th e variable / O (A ') d X ) . T h e fu n c tio n
l-i(V ) o b v io u s ly is an a d d itiv e f u n c tio n o f th e reg io n in th e sen se th a t \ i ( V \ + V 2 ) =

J a (^ i) + \ i ( V 2 ) for n o n o v e r la p p in g r eg io n s Vi a n d V 2 . In stead o f th e m u ltid im e n sio n a l
p r o b a b ility d is tr ib u tio n s w h ic h d e te r m in e th e ra n d o m fie ld d ( A') w e n o w hav e th e
p r o b a b ilitie s
••• = = = (1 0 .1 3 )
g iven fo r all p o s sib le fin ite sets o f th e spatial r eg io n s Vi , .. I//1 . In th e rem a inder o f this
s u b s e c tio n we shall, for s im p lic it y , c o n sid e r o n ly p r o b a b ilitie s (1 0 .1 3 ) fo r regions
V I ........... Vh w h ic h are n o t n o n o v e r la p p in g . In th e s p e cia l case o f in d e p e n d e n t s ta tistica lly
e q u iv a le n t a d m ix tu r e particles th e d is tr ib u tio n (1 0 .1 3 ) d e g e n e ra tes in t o th e so -called
p o ly n o m ia l d is tr ib u tio n
.......=
x [i-o ,(K ,)- ...
w h ere
" i - I ..
T h u s, in th e case o f in d e p e n d e n t particles th e f u n c tio n is g iv en c o m p le t e ly b y th e

v alue o f th e pro b a b ility Q i(V ) for all r eg io n s V. In th e general case o f d e p e n d e n t
s ta tistic a lly e q u iv a le n t particles th e d is tr ib u tio n ( 1 0 . 1 3 ) w ill rep resen t s o m e g e n e r a h z a tio n
o f th e p o ly n o m ia l d is tr ib u tio n ; t o assign it, it is n e c essa ry t o k n o w all th e p r o b a b ilitie s
.......^ ) =
= [A'l e V € I",:
, ................. . • • : , + 1 e V, ................. 4 } .
(1 0 .1 4 )
It is p ossib le t o s h o w th a t a n y m o m e n t o f th e r a n d o m variables M-( V i ) , .. î( K /,) m a y be

e x p r e s s e d in term s o f th e p r o b a b ilitie s ( 1 0 . 1 4 ) w ith th e h e lp o f t h e fo r m u la
= s ..... (-......
5=1 n.-{- ... + riuKm
(1 0 .1 5 )
w h e r e m = m, -f- • . . -f- is th e order o f th e m o m e n t , a n d are th e so -called Stirling

n u m b e r s w h ic h d iffe r fr o m z er o o n ly w h e n 0 < n m a n d satisfy th e d if fe r e n c e e q u a t io n
= ^ ,7 -V "i* ^ b o u n d a r y c o n d it io n s aj" = a' ” = 1. T h u s, in particular,
it f o ll o w s th at th e m th -o r d e r m o m e n t o f th e r a n d o m fu n c t io n | i ( K ) is d e fin e d c o m p le t e ly
b y th e j o in t p ro b a b ility d is tr ib u tio n fo r th e c o o r d in a te s o f m particles, in o th e r w o r d s , th a t
th e p r o b a b ilitie s ( 1 0 . 4 ) w ith ny + . . . + n/i < m d e fin e th e r a n d o m f u n c t io n pi ( V ) w ith
a ccu ra c y up t o m o m e n t s o f th e w t h o rder.
When m = 1, fr o m E q. ( 1 0 . 1 5 ) w e o b ta in
(] l(K ) = Q , ( V) , (1 0 .1 6 )
w h ic h is a n a lo g o u s t o E q . ( 1 0 . 5 ) , m o r e e x a c t ly , to th e e q u a tio n o b t a in e d fr o m E q. ( 1 0 . 5 )
after in teg r a tio n in A" o f b o t h sides o f it over th e r egio n V. W hen m = 2 , fr o m Eqs. ( 1 0 . 1 5 )
and ( 1 0 . 1 6 ) w e o b ta in
1 (1 0 .1 7 )
b- {V) - (K)l^ = Q A V ) - [Q, (K)]2 + i ; r 1
[i^ (K,) - (I",)] If^( Kj) - Kj)] = Q„ (K,. Kj) - Q, (K,) 0, ( V,)-
(1 0 .1 8 )
T h e first o f th es e e q u a t io n s s h o w s th at in th e p r e se n c e o f sta tistical in te r a c tio n b e t w e e n

particlcs, or w h e n Q 2 ( V) [Q i ( t he flu c t u a tio n s o f th e a m o u n t o f a d m ix tu r e in th e
fix e d v o lu m e V d o n o t d e crea se in d e f in it e ly (in q u a d r a tic m e a n ) e v e n w h e n N - > c o . In
o th e r w o r d s, in th e case o f m u tu a lly d e p e n d e n t p a r tic le s, e ven for a very large n u m b e r o f
particles N, th e values o f |li(K ) tu rn o u t to b e q u ite d if fe r e n t for d if fe r e n t r ea liza tio n s.
E q u a tio n ( 1 0 . 1 8 ) s h o w s th a t, in a d d itio n , in th is case th e flu c t u a tio n s in th e a m o u n t o f
a d m ix tu r e in t w o d if fe r e n t , n o n o v e r la p p in g r eg io n s are m u tu a lly co rr e la te d . S ig n ific a n tly ,
m u tu a l d e p e n d e n c e b e t w e e n th e particles is t y p ic a l o f tu r b u le n t d if fu s io n ; in fa c t, he r e th e
m o t io n s o f n eigh b o rin g particles are in flu e n c e d c o n s id e r a b ly b y th e sam e d is tu rb a n c es in th e
v e lo c ity fie ld , and th er e fo r e are sta tistica lly rela ted to e a ch o th er . T h u s for tu r b u le n t
d iffu s io n th e c o n c e n tr a tio n o f th e a d m ix tu r e in e a c h v o lu m e flu c tu a te s n o t ic e a b l y , an d for
a d ja ce n t, b u t n o n o v er la p p in g v o lu m e s m a y turn o u t t o b e d e p e n d e n t o n e a ch o th e r for a n y
N and e v en w h e n N = c o . T his last case is also d e scr ib e d b y th e m o d e l o f a c o n tin u o u s ly
distrib u ted a d m ix tu r e w ith a c o n c e n tr a tio n 0 (A') w h ic h is c o n tin u o u s ly d e p e n d e n t o n X.
10.2 Interaction Between Molecular and Turbulent Diffusion
The admixture propagation in a turbulent fluid due to its transport

by the moving fluid particles, that is, turbulent diffusion, usually takes
place more rapidly than its spread caused by molecular motion, that
is, molecular diffusion. Thus, for example, in the atmospheric surface
layer the virtual “eddy diffusivity” for passive admixtures, namely.
59 2 S T A TIS TIC A L F L U ID MECHANICS
the ratio of the turbulent flux of the admixture to the concentration

gradient, usually has values on the order of 10"^-10^ cm^ sec"*. In
other words, it turns out to be 10^-10^ times greater than the
molecular diffusivity, which for most gases is on the order of
10"^ cm^ sec"^ Due to turbulent diffusion, cigarette smoke in
concentrations sufficient to affect the human sense of smell, is
spread around a room in several seconds, while the effect of only
molecular duffusion would lead to the same result after only several
days. However, does this indicate that for a description of turbulent
diffusion it is possible to completely neglect molecular diffusion?
The transport of the admixture by the moving fluid particles
obviously leads to a situation in which the volume first occupied by
the admixture is extended in certain directions, and in others is
compressed, being distorted with considerable confusion. (This will
be discussed in more detail in Chapt. 8 of Volume 2.) However, the
total volume of admixture does not change with this deformation. As
a result, in each realization of turbulent flow the specific concentra
tion 0(A^, 0 at each point and any time will be equal to either the
density of the admixture Po or 0, depending on whether or not at the
initial time t = to the corresponding fluid particle contains the
admixture. [See Fig. 79 for a schematic diagram borrowed from the
review article by Corrsin (1959a).] Thus turbulent diffusion leads to
the formation in the fluid of highly distorted and twisted layers with
sharply varying values of admixture concentration. The equalization
=>
0
b
FIG. 7 9 . a - V a r i a t i o n in fo r m o f th e v o lu m e o c c u p ie d by
an a d m ix tu r e d u e to tu rb u le n t d if fu s io n ; b - s c h e m a t i c fo r m
o f d is tr ib u tio n o f a d m ix tu r e c o n c e n tr a tio n a lo n g a straight
line in te r s e c tin g th is v o lu m e .
of the concentration in adjacent layers, which is accompanied by an

increase in the volume occupied by the admixture, and a smoothing
of the concentration field, as a result of which the values o f the
function as a rule turn out to be intermediate between p o and
0, occurs only as a result o f molecular diffusion, and is the slower,
the smaller the diffusivity X- It is clear from this that for the
description of the small-scale statistical structure of the field ^ ( X , t)
it is impossible to neglect molecular diffusion; otherwise when t is
large, we obtain a rather unnatural distribution of concentration of
the type depicted on the right side of Fig. 79. However, the problem
of whether it is impossible to neglect the molecular diffusion, in
comparison with turbulent diffusion, when finding the mean
concentration as was done in the preceding subsection for
derivation of Eq. (10.5), is quite complex. We shall now deal with
this problem.
To investigate the relative role of molecular and turbulent
diffusion in the formation of the mean concentration field ■ ^ ( X , t),
for simplicity, we shall limit ourselves to the problem of diffusion of
an admixture from an instantaneous point source of unit output in a
field of homogeneous turbulence with zero mean velocity. We shall
consider this source as being placed at the time i = the point X =
0. In other words, we shall consider the concentration d(A^, /)
corresponding to the solution of Eq. (10.1) for the initial condition
ft (A", ô) =6(-^)- Let us restrict our consideration to the admixture
diffusion in some single fixed direction, for example, in the direction
of the 0X^ axis. As the characteristic o f the role of molecular
diffusion, we shall use the relative contribution of this diffusion to
the dispersion of the mean concentration distribution ■&{X, t). Here
it is essential to distinguish, on the one hand, the dispersion o f the
mean concentration distribution with respect to the location o f the
sources X — 0
fX iH ^)d X (10.19)
and, on the other, the dispersion (also with respect to the location of
the source X = Q) o f the center o f gravity
Xc{t) ^ j K i ^ { X , t ) d X
o f the concentration distribution d(A ', t), that is, the variance
In addition, the dispersion o f the concentration distribution ^ ( X, t )

with respect to its center o f gravity is also of interest:
D?(0 = / IA T i-^,(O P »(A ', t)d X . <10-21)
However, using the normalization condition
f ^ {X ,t)d X = 1
which expresses the law of conservation of the admixture mass, it is

not difficult to prove that
so that it is sufficient for us to calculate only Do(t) a n d X d t ) .

Further, it will be convenient for us to transform from the
stationary system of coordinates X to the nonstationary, noninertial
system of coordinates Y = X — X (0, /), the origin of reference of
which, at each time coincides with the point where the fluid particle
is found, which at the time to was at the point X = 0. For the
concentration field in this new system o f reference, we shall adopt
the notation
6(K. 0 = t), ^] = &(^, t). (10.23)
It is clear that
ae(K. t) _ d%(X. t)
dYi ~ dXi Jf=K+Jf(0, 0
On the other hand.
^ ^ . ^ n r + x i o , ,). < i = f + v .(o ,
where V(0, = is the Lagrangian velocity of the fluid

particle which at the time /o was at the point = 0. Consequently,

Eq. (10.1) in this mobile system of coordinates takes the form
§ + ( « , l K + ^ ( 0 , t), t)\-^ = x V \ (10.25)
where is the Laplace operator with respect to the variables

(I'l. i'2. Ya), and V (0, t) is the velocity o f the coordinate origin of
our mobile system with respect to the initial stationary system of
coordinates. The concentration 6(J', i) also satisfies the normaliza
tion condition
/ 6(K. t ) d V = 1
which we shall use without further comment. Furthermore, Eqs.

(10.19) and (10.20), after transformation to the new system of
coordinates, assume the forms, respectively,
= XUO, t) + 2X,{0, t) f / y l H V , t) d Y , ^ jq 2 6 )
and
2
= A'?(0, 0 + 2A',(0, t) f y ^ d l Y O d Y + l f Y i B ( Y , O d Y
■ (10.27)
We shall limit ourselves to the calculation of the values of Do (0 and

X l (t) for small values of t — to. We shall assume that all the moments
( K f = J r f q{ y , t)d Y ( 10 .2 8 )
(where ki, k^, kz are nonnegative exponents) may be represented as

Taylor series in integral powers of t — U. Then the dispersions D\(t)
and (/) will also be representable in the form of a series in powers
of t — U, where for calculation o f the first few terms of the series, it
is necessary to know only the first few terms of the series for the
moments (Ki) and {Vf} which enter into Eqs. (10.26) and (10.27).
Let us use Eq. (10.25) to calculate the moments (10.28). Here we
have «, [A'(0, t), /] = Vi(0, t), and, consequently, the velocity dif
ference in the braces on the left side of Eq. (10.25) may be
represented in the form
« , [ K + ^ ( 0 , t), / ] - K,(0. t) =
V I ^V V I ^ V ir V I
~ « dX, 2 <■ P d X ^ dA’j ' 6 T d X ^ d X ^ d X ., + ' *•> (10.29)
where all the derivatives of Ui{X, t) are taken at the point X =

X (0, i), that is, are functions only of t. Let us now substitute the
expansion (10.29) into Eq. (10.25), then multiply all of its terms by
and integrate with respect to Y. Taking into account that
0, after integration by parts of the terms containing the
spatial derivatives of 0 ( K, t), we obtain
^ ik, - 1)(
<=i '
, V A I \ , 1V . \ ,
d x ,\ Vi l~ ^ 2 Z i dx, dx^ \ r, / '
/=1 irr\
~^' 6 ^ ^ ‘ 'dX,dX,dX~^ Yt )+•••. (10.30)

i=i ^ '
where the moments which contain negative powers of K, are taken

identically equal to zero.
It is obvious from Eq. (10.28) that the zero-order mom ent (with
ki = kn = ks = 0) is exactly equal to one; since 0( K, /(,) = fi( K) all
the remaining moments approach zero a s t- ^ to - Therefore, terms
linear in t — /owill appear only in the expressions of the moments for
which the right side of Eq. (10.30) contains a zero-order moment. It
is not difficult to see that the only moments which have this
property are the moments (K?) and their leading terms have the form
{Yf) = 2x{t - Q + ■■■ ■ (1 0 .3 1 )

However, to determine the subsequent terms of the power series for

the moment (Ki) which we need below, we must also write out all
the moments which are of order(/ — /o)* when/ -> /o-
It is clear that those moments of nonzero order, different from
(yj) for which the moments may be found among the
summands on the right side of Eq. (10.30), will be of second order in
/ — (o- It is easy to prove that only the moments
(K,). ( r t ) and (r? r5 ) for i ^ /
are of this sort. Moreover, their leading terms have the form
= X + - ^ ) (t — tof + • • •,
(r^) = i 2 x ^ / - / o f + ....
(r?r5 ) = 4 x ^ ( / - / o f +
where all the derivatives of u { X , /)are taken at the space-time point

X=0,t = U. _____
Let us consider the evaluation of D\{t) and X l(t), with accuracy
up to terms of order (t — to) Since
^ ( 0 , 0 = «(0, ^ o )(^ -^ o )+ •••. (10.33)
for this we must know the moment (Ki) with accuracy up to terms of
order — and the moment (K?) with accuracy up to tenns of
order (t — ô)^- Thus, Eq. (10.32) for (Ki) is sufficient for our
purposes, but the expansion (10.31) for (K?) must be refined. To
determine the missing terms, let us write out Eq. (10.30) separately
for ^1 = 2, ki = ki = Q:
<^«l /U K \ I
■■■■
Considering on the right side of this formula only terms of order not
greater than (t—ô)^>with Eq.(10.32) we obtain
dux
I - < K i> = 2 X + 2 -IJ ;- < K ? ) + 2 (K ,K ,> + 2 ^ (K ,K ,> +
+ T 4 ^ < f 'i > + - ^ W > + ^ ^ < K ? K |) + .

3 dX\ dX^ dX\ dX^ dX\
dui dui du,
= 2x + 2 <>^?> + 2x dX^ \ d X , ' d X ,}^
dUx dui , du
dX^ \ dX^ ^ dX,
I d dUx
Smce-gjf^ = -5 ^
<KR= 2X«-'.) + 2x|J ^ ( ^ -< ,) '+ | x {2 w ^ + 2(57)’ +
(10.34)
where all the derivatives of tt(X, t) are taken at the point X = 0,

t = h.
Further, we need only the mean value of the moment (K?). Let us
recall that we are considering homogeneous turbulence where mean
values of all variables which depend on only a single space-point X
are constant. In particular,
d dUx d dui dUx

W JX ; ~ ^ ~dT + ^■ «a = 0.
Thus, in homogeneous turbulence
(Ki) = 2X (/ - « + 1 * { 2 ( A J -)' + I - ( IJ L + U . ) +
The expression in braces may easily be converted to the following

form:
and we obtain finally
^ = 2 x ( /- / o ) + |x W ( ^ - ^ o ) ^ + •••• (10.35)
Let us also note that according to Eqs. (10.22), (10.26), (10.27), and
(10.32), the variable D\{t) with accuracy up to terms of order
{t — ta)^ coincides with^K?); therefore
D?(0 = 2x ( < - / o) + | x ( W ( ^ - < o) ^ + • • • • (10.36)
In an unmovable fluid, molecular diffusion transforms unit mass

of admixture, concentrated at a point at time t — U, into a cloud
described by a spherically symmetrical concentration distribution
with dispersion 2%{t— to). It is obvious from Eq. (10.36) that when
t — to is so small that the second term on the right side of this
equation may be neglected by comparison with the first, the average
dispersion of the admixture distribution t) about its center of
gravity Xc{t) may also be assumed to be equal to2x(/ — to). In other
words, one may consider (in any case, to the extent to which this has
to do with dispersion) that in the first stage of turbulent diffusion
from a point source, spreading of the admixture under the action of
molecular diffusion is superimposed on its transport by the corre
sponding fluid particle, but does not interact with this transport.
However, the case is different for somewhat larger values of / — <o
because of the presence o f the second term on the right side of Eq.
(10.36). As a result, the dispersion O l(t) begins to increase more
rapidly than for molecular diffusion in an unmoving fluid; the
additional term here depends not only on the fluid molecular
diffusivity X, but also on the mean square velocity gradient of the
turbulent motion. Thus, an interaction between molecular and
turbulent diffusion occurs which causes acceleration of the molecular
diffusion; that is, more rapid expansion of the volume occupied by
the admixture, and, correspondingly, more rapid decrease in the
maximum concentration. The actual cause of this phenomenon is
obviously the sharp deformation of the volume occupied by the
admixture during turbulent flow (of the type shown in Fig. 79),
essentially increasing the role of molecular diffusion. Therefore, it is
not surprising that the “ acceleration of molecular diffusion” is not
felt immediately, but only some time after the ordinary molecular
diffusion transforms the point mass of admixture into a cloud, the
volume of which is sufficient for deformation to occur under the
influence of the instantaneous velocity inhomogeneities character

ized by the gradient of u {X, t ) .
The effect of molecular diffusion acceleration due to turbulence
was first noted by Townsend (1951). He gave preliminary estimates
for this effect and presented results from special experiments
(measurement of the gradual drop in the maximum temperature of
thermal spots produced by a pulse discharge of current in a turbulent
flow) confirming its existence. Moreover, Townsend indicated that
equations of the type (10.36) provide an adequate description for
significantly larger values of t — to than was supposed earlier. More
complete equations for diffusion from an instantaneous linear source
[containing some minor errors in the values of the numerical
coefficients; see Saffman (I960)] were obtained later by Townsend
(1954), and Batchelor and Townsend (1956). Finally, the second
term on the right side of Eq. (10.35) strongly resembles Eq. (9.58)
for the variance o f the longitudinal coordinate of a fluid particle in a
homogeneous turbulent shear flow with a constant mean velocity
gradient. This similarity is not accidental; for small t — to the entire
admixture is within a small neighborhood of the source, in which the
velocity u {X, t) may be assumed to be linearily dependent on the
coordinate X, limiting ourselves to the first term of the corre
sponding Taylor series, and Eq. (10.35) corresponds to this repre
sentation of the velocity field.
So far, we have considered only the dispersion D\{t) of the
distribution t) about its center of gravity. Proceeding to the
consideration of the dispersion D\{t) of an admixture cloud about
the position of the source, coinciding with the mean center of gravity
location Y j J ) and X \ {t) of the center of gravity itself, then we find
that here the interaction between molecular and turbulent diffusion
leads to the opposite effect of “ deceleration of diffusion.” Indeed, to
calculate the quantities D\ (t) and X l (t) according to Eqs. (10.26) and
(10.27), it is necessary only to determine the mean value
(0, t){Y\). Using Eqs. (10.33) and (10.32) for Xi(0, t) and (Ki), we
obtain
or, since in homogeneous turbulence = — (V«i)^,
X ,{Q, .... (10.37)

From this and from Eqs. (10.26), (10.27), (10.35), and (10.32) it
follows that
Dl (0 - X \ (0, t) + 2x ^ X ( W {t - t , f + .... (10.38)
3 ^ = ( 0 , 0 - X - tof + .... (10.39)
These equations were first obtained by Saffman (1960); see also

Okubo (1967) who deduced them with the aid of the Lagrangian
diffusion equation (9.9'). The second of these equations shows that
the variance of the admixture center of gravity for small t — ?oturns
out to be less than the variance X i (0, t) of the coordinate of the fluid
particle, determined almost exclusively for small t — to hy the
variance of the initial velocity Vi(0, to) = «i(0, to) of the fluid
particle, that is, close to«i {( — ô)^. In other words, due to molecular
diffusion, the admixture particles on the average stray away from
the fluid particle with which they first coincided, and, therefore, the
molecular diffusion slows down the turbulent diffusion. The mechan-
ism for this deceleration is easily seen: the quantity A'?(0, ^) is
determined by the variance of the “ average rate of transport of the
fluid particle”
and molecular diffusion leads to spread of the point fluid particle

into the entire admixture cloud. Therefore, here, the rate of
transport is the velocity Ui(X, t') averaged with respect to some
volume, which as a rule is less than the instantaneous velocity at the
point X {0, t'). Since X l (t) turns out to be less than A'? (0, t), the
total dispersion Dl{t) turns out to be less than the sum Ai(0, t)-\-
2x(^ — ô) of the individual contributions from the turbulent and
molecular diffusion without taking into account their interaction.
The deceleration of turbulent diffusion due to molecular diffusion
may also be described with the help of the “admixture correlation
f unct i on” — to) introduced by Saffman (1962a). The function
may be defined as the average covariance of the fluid
velocities at the points occupied by the admixture at the times / > to
and /o*.
(t - Q = {u, (0 ,t , ) u , \ Y + X (0,t). t \ y (10.40 )

60 2 ST A T IS TIC A L F L U ID MECHANICS
where the angular brackets, as above, symbolize averaging with

respect to the values o f V with the weight function V, t ) . Saffman
showed that the dispersion Dl{t) may be expressed in terms of
_ /o)with the aid of the following equation:
Dl{t) = 2 x { t - t o ) + 2 f (10.41)
Equation (10.41) is similar to Taylor’s equation (9.31) which has the

form
X ! (0, = (x)dx. (10.42)

/c
In addition, just a s — <o)may be represented as
Ba>0 = f «i(0, t m X , t)dX, (10.43)
where (A", 0 = S [A’— A’( 0 , /)| is the solution of the advection

equation = initial condition ’];(A, /o) = ^ W .
g(»)(; — to) obviously may be written as
B(8> (^ - <o) = / «i (0, Q Hi (A, t) (A, t) dX, (10.44)
where i]^(*)(A, t) is the solution of the diffusion equation
for the same initial conditiorn|; {X, to) = if we assume

to be nonnegative for all t , the effect of deceleration of the
turbulent diffusion as applied to the function is expressed by
the expectation that for all t > 0. In other words,
one may say that as a result of molecular diffusion the admixture
particle “ forgets” its initial velocity more quickly than the fluid
particle because of the additional averaging with respect to K in Eq.
PARTICLE DISPERSION IN A TURBULENT FLUW 60 3
(10.40). On the basis of this, Hinze (1959) made the following

purely qualitative proposal:
B W ( t ) = B W ( t ) . / ( t )
where f{x) is a monotonically decreasing function which depends on

the parameter % such that f{0) = 1. Following a suggestion of
Burgers, Hinze made a special investigation of the model / ( t )
(where a = ^ , c is a dimensionless constant, and L is a turbulence
length scale. However, this assumption has no theoretical basis,
except perhaps for the special case of turbulence for very small
Pecl6t number
Pe = A (Z t;
see Saffman (1962a). For very small t = t — to for which the

solution of the diffusion equation may be approximated by the
linear term of a series in powers of -r, it is possible to deduce from Eq.
(10.44) that the function is representable in the form
5(*,(^) = - e , " , ( x ) - x ( v « i ) 'x + •••,
from which Eq. (10.38) follows directly. It is natural to expect that

with an increase in <— to, or with an increase in volume occupied by
the admixture, the effect of deceleration of the turbulent diffusion is
amplified. Namely, that the ratio
[BW(t ) - BW(x)]/SW( t )
increases with an increase in t. Assuming that whenT-»- c», this ratio

approximates some constant A, Saffman (1960; 1962a) obtained the
asymptotic formula
Dlit) = Xl{0, t ) ^ 2 x { t - t o ) - A X U O , t), (10.45)
which is valid for large t — to- Saffman also presented some

arguments according to which
A ~ (Pr-Re)-> = (Pe)-'
where Re is the Reynolds number o f the turbulence constructed, for

example, of the root-mean-square velocity fluctuation («'?)'^’ and the
so-called Taylor length microscale
V.
X=
This prediction turned out to agree qualitatively with Mickelsen’s

data (1959), in which the distribution of the helium and carbon
dioxide mean concentration was measured downstream from a point
source in a grid turbulence. However, neither accurate quantitative
confirmation of Saffman’s hypothetical asymptotic equation (10.45)
nor qualitative confirmation o f Eq. (10.38) for small values of< — ta
has yet been obtained.
To understand the reason for this latter situation, one must
estimate the relative role of the molecular and turbulent diffusion,
and the interaction between them, in the formation of the mean
concentration field OCA",/) - For this it is necessary to compare the
orders of magnitude of the three terms on the right side of Eqs.
(10.38) and (10.45). For very small t — U Eq- (10.38) must be used,
and on its right side, the largest term is the second [of order t — <ol,
and the smallest, the third [of order( / — /o)®l • However, with an
increase in t — to the contribution of turbulent diffusion increases
more rapidly than the contribution o f molecular diffusion, and for
values o f t — to which exceed some value of ti , the contribution of
turbulent diffusion exceeds the value of — to)- Considering t) to
be small by comparison with the Lagrangian integral time scale Ti
(i.e., the time of significant variation of the fluid particle velocity), it
is possible to evaluate ti by means o f Eq. (10.33) forA'?(0, /); in this
case, we find that = 2x/«j. in most real turbulent flows, this value
of Ti is infinitesimally small; for example, in the atmospheric surface
layer u\ ~ 10^ to 10^ cm^ • sec“^, so that t i ~ 10”^ sec when % ~
10“ ‘ cm^ • sec. For large values of / — to, the value of A'?(0, <)may
be estimated with the help of the asymptotic formula (9.35);
however, the ratio of the contributions of the molecular and
turbulent diffusion here is of order
2r,
PAR TICLE DISPERSION IN A T U R B U L E N T FLOW 60 5
In other words, this ratio may be estimated as having the order of

(P r-R e ^ )'‘ since it is possible to show that usually ri~A .V v; both
these estimates show that under real conditions this ratio is always
extremely small.
Moreover, for sufficiently small t — U, such that the contribution
to the dispersion Dl{t) of the terms deriving from the interaction
between the molecular and turbulent diffusion may be considered
proportional to (t — this contribution will be much less than the
contribution of the molecular diffusion if
However, for stationary homogeneous turbulence
r-2
where
is the Lagrangian time microscale. Thus, when ^ < xx, the ratio
of the contribution to dispersion due to the interaction between the
molecular and turbulent motion to the contribution of the molecular
—
diffusion is of small magnitude of order On the other
(-
hand, when t — the asymptotic formula (10.45) may be
used to prove that this ratio has the order
Pe-x, ’
that is, it is not necessarily small. However, the ratio of the

contribution made by the interaction to the contribution o f the pure
turbulent diffusion is of order/I ~ (Pe)'*; that is, it is very small for
large Re and Pr ~ 1.
In summary, we may state that molecular diffusion does change
the dependence of the average diameter o f the diffusing admixture
60 6 ST A T IS TIC A L F L U ID M ECHANICS
cloud on time t — U and, therefore, makes it more difficult to use the

functions = X] (t) in determining the Lagrangian velocity correla
tion function by equations o f the type (10.11). However, the relative
change in the value of Dl (/) itself, caused by this diffusion, for/ — to
which is not very small and sufficiently large Reynolds numbers,
usually is negligibly small. Therefore, we may conclude that for
practical purposes, molecular diffusion may be neglected in calcu
lating the mean concentration d ( X, 1). Thus we shall proceed in this
manner in the subsequent subsections of this section.
10.3 Semiempirical Equation of Turbulent Diffusion
Let us return now to the investigation of turbulent diffusion of a

continuously distributed admixture, neglecting the molecular diffus
ion. Here the mean admixture concentration &(A', t) at the time t is
related to the initial concentration do W by Eq. (10.5). It is obvious
from this that if the probability density for the coordinates of the
fluid particle p { X \ x , t) satisfies some linear equation in the variables
X , t, then the same equation will be satisfied by the function
■O'CX, t) . Let us now try to understand what form this latter equation
may have.
Neglecting molecular diffusion, the admixture concentration
^ { X , t ) satisfies the advection equation
d% , ^
which is obtained from the diffusion equation (10.1) by equating its

right side to zero. Averaging all the terms of this equation, we obtain
where 5 = (Si, S 2, 5a) is the turbulent flux vector of the diffusing

admixture. The simplest and earliest theory of turbulent diffusion
proposed by G. L Taylor (1915) and Schmidt (1917; 1925), but in
essence belonging to Boussinesq (1877; 1887), is based on the
assumption that the flux 5 is proportional to the gradient of the
mean admixture concentration; that is.
(10.47)
where K is the eddy diffusivity /(»of Chapts. 3 and 4. (In the present
chapter we shall be dealing only with it, and, therefore, the subscript
dwill conveniently be omitted.) In the more general anisotropic case,
instead of the assumption (10.47), the existence of a linear
dependence between the vectors Si and is assumed:
(10.48)
where the eddy diffusivity tensor Kij is in general a function of X

and t.
From the semiempirical hypothesis (10.48) the semiempirical
equation of turbulent diffusion is obtained directly
^ (10.49)
dt ^ dX^ dX^ ^ 0 ? dX^
Since this equation will play a major role in the remainder of this
section, it may be well to begin by examining a few additional
considerations which are relevant to the derivation of this equation,
and which make more evident its range of applicability. We begin
with the simplest case of diffusion in a field of stationary
homogeneous turbulence, all the statistical characteristics of which
are invariant with a change in the time origin or a shift in the zero
point of the spatial coordinate system. In this case the covariance
tensor of fluid particle displacement in time x , that is, the tensor
K}(x),
is determined by the generalized Taylor fojmula ( 9 . ^ ') . Morôve^r,

the mean displacement is equal to Yi ( t ) = «iT, where « = (wi, uz, ih)
is a constant (because of stationarity and homogeneity) mean tlow
velocity. In addition, as we know from Sect. 9.3, substantial grounds
exist in this case for assuming that the joint probability distribution
for the variables (Vi, Y2, Y3) both for small and large t, by comparison
with the typical Lagrangian time scale T, will be very close to a
Gaussian distribution and apparently will, in general, not be
markedly different from Gaussian for all values of x. Therefore, the
probability density p { X \ x , t) here may be approximately rep
resented as a three-dimensional Gaussian density of the form (4.23)
608 ST A T IS TIC A L F L U ID MECHANICS
with mean values a, = jc ,-+ — ô) and covariance matrix

inverse to the matrix
||D jj(T )ll, in the expression for the
density, determined by Eq. (9.30'). (In particular, in the absence of a
mean velocity and under the assumption that D , j ( t ) = 0 when/ ^ /,
we again obtain Eq. (10.12).) It is not difficult to verify that such a
function p(A'|x, t) is a solution of Eq. (10.49) with coefficients KtjOf
the type
= = 4 7 W m + S‘/<’W1 * (10,50)
[see Batchelor (1 9 4 9 b )]. This same equation will also be satisfied in

this case by the mean concentration0 (A', t). Thus, the assumption of
an equation of type (10.49) [but with the coefficients of Ka defined
by Eq. (10.50), that is, depending on the time t = / — /o of
admixture propagation] for diffusion in a field of stationary
homogeneous turbulence is exactly equivalent to the assumption that
the probability distribution for the displacement vector of the fiuid
particle for any t will be normal (Gaussian). If, however, in the spirit
of the semiempirical hypothesis (10.48) we wish the distribution of
the mean concentration for all ^ t to be described by an
equation of type (10.49) with c o e f f ic ie n ts in d e p e n d e n t of t, then
according to Eq. (10.50), it is necessary to limit ourselves to values
of t which are large compared to the Lagrangian time scale T. Here,
on the basis of Eq. (10.50)
1 “
K i) = T i, = f [R'f/ ( ^ ) + R 'ji( x ) l dx (10 .51 )
in complete agreement with the semiempirical representation of the

eddy diffusivities as the product of a turbulent velocity scale by a
length scale [or, what is the same thing, the square of a velocity scale
multiplied by a time scale].
We have seen that for diffusion in a field of homogeneous
stationary turbulence the semiempirical equation (10.49), with
constant diffusion coefficients Kt.,, is satisfied only when / > /o + ?',
but for these t may be justified quite convincingly according to the
Gaussian form o f the probability distribution for Y (t) which is very
Ukely on the basis of the central limit theorem; see above, Sect. 9.3.
Let us note, however, that m this case the usefuhiess of Eq. (10.49)
becomes quite limited since the general expression for D (Xy t)here
may be written out directly, independently of this equation, for
example, beginning with Eqs. (10.5) and (10.12). Therefore, the
basic value of the semiempirical theory consists in the possibility of
its application to the more general case of inhomogeneous, or
nonstationary, turbulence which we shall consider.
First, let us show that, in general, Eq. (10.49) may also, strictly
speaking, be apphcable only if the diffusion time t = / — /o
significantly exceeds the Lagrangian time scale. For this we consider
the probabilistic meaning of this equation. Equation (10.49) is a
first-order equation in consequently, its solution is uniquely
determined by the initial value la-CA', ô) = where the
boundary conditions will be considered fixed in advance. Let us
designate temporarily the probability density p ( X l x , t) by the
symbol p {X , t x Jo) in order to emphasize its dependence on the
initial time to; in this case expression (10.5) may be rewritten as
T{Xri) = f H ^ ) Î ô)
Assuming here that ■&(X,/o) = 6 ( X —x), we obtain

p(X, t X, to) for any t > to. We shall use this result for the time ti < t,
and then again solve Eq. (10.49) for the time interval from the time
to t, for the initial condition ■0'(A’,/i) — p{X ,ti\x,U ). Then we
obtain
p { X , t \ x , to) = j p { X , t \ X „ t,)p{X,, t , \ x , (10.52)
where to < < t. However, this equation for the transition probabil
ity density p ( X , t\x,to) of the random function A' (/) will be correct
only if it is a Markov process. This means that the conditional
probability distribution for the value of the function at the time t,
under the condition that its values are known at arbitrary times <
t„-i < . . . < to (where to < t), will depend only on the last of these
values X (to), but not on the values of X (ti) with t{ < to- (In the
theory of Markov random processes, Eq. (10.52) is often called the
Smoluchowski equation.) Indeed, for an arbitrary but not necessarily
Markov random function X (t), when t o < t \ < t , the following
obvious equation will occur:
p { X , t \ X , to) = j P { X , ( \ X „ X, to )p { X „ t y\ X, Q d X y ,
where the first factor under the integral sign is the conditional
probability density for X (t), under the condition that the values of
X ( ( j ) — X, and X ( i o ) = x are fixed; this equation is a special
example of the theorem of total probability. The latter formula can
be transformed into Eq. (10.52) only if the indicated probability
density does not depend on the value of A" (ô) = x , and this
indicates that the random function X (t) is Markovian. Let us
reemphasize that our conclusion does not use the explicit form of
Eq. (10.49), but is based only on the fact that this equation is of first
order in i.
On the other hand, if the random function A '(jf,/)is Markovian,
then under very general conditions a differential equation of the type
(10.49) may be obtained for the probability density
p ( X l x , t ) = p ( X , t l x , to)
and, therefore, according to Eq. (10.5), also for the mean concentra
tion ■&iX,t). This important mathematical theorem was proved by
Kolmogorov (1931; 1933) [his special cases were established still
earlier by the physicists A. Einstein, A. D. Fokker, and M. Planck].
Specifically, Kolmogorov proved that under some broad regularity
conditions, imposed on the transition probability p ( X \ x , t ) and
ensuring that the Markov random function X (t) in question will be
continuous in a certain sense, the following derivatives exist:
V{x, t o ) ^
(10.53)
2Ku(x, to)= -dt^ r ' d x , t)Vj(Jc, t)
/=/o
where as usual Y{x, t) = X { x , t) — x and Y ' = Y — Y, and the

probability density p { X \ x , /) satisfies the equation
^ + = (10.54)
[See also the textbook by Gendenko (1962).] This equation

obviously coincides with Eq. (10.49) if we assume Vi —
This latter equation means that the variable F,, being the mean
velocity of the admixture particle, according to the first formula of
Eq. (10.53), is generally equal to the sum of the convection velocity
dK,j
w, and the additional velocity-^^.7 , related to the fact that in a field
of inhomogeneous turbulence the admixture particles may have a
tendency to move in a certain direction even in the absence of a
mean flow. Let us emphasize that here only the concept of mean,
and not instantaneous, particle velocity is introduced; the instantane
ous velocity in the “Markov model” does not exist, indeed, the limit
lim
K(x, t)
^->/o
which must denote the mean square instantaneous velocity of the

particle at the time ô, cannot be finite when Ku( x, to) 0 on the
basis of the second of Eqs. (10.53). This is because the mean square
path length traversed by the particle in the time between to and
ti to + T, when T -> 0, according to Eqs. (10.53), is of order x and
not Let us also note that the second of Eqs. (10.53) gives a
statistical interpretation of the “ eddy diffusivities” : Kij differs only
by the factor of 1/2 from the rate of change of the particle
displacement covariance which we considered in Sect. 9.3.
The nonexistence of the instantaneous velocity indi
cates that the random function X { x , t ) i s not differentiable. Thus, in

the semiempirical theory of turbulent diffusion, the trajectory
X { x J ) of a given “ fluid particle” is interpreted as a Markovian
function which is not differentiable anywhere. The trajectories of the
real fluid particles in a turbulent flow clearly do not have such
properties; the function A '(jc, /) is everywhere differentiable with
respect to / and, moreover, satisfies the Lagrangian dynamic
equations (9.6) and (9.9) which contain the first and second
derivatives of X { x , t) with respect to Xi and with respect to t.
Knowing the statistical properties of the derivative
dt
t)
or, the Lagrangian velocity, is important for many problems

61 2 ST A T IS TIC A L F L U ID MECHANICS
concerning particle dispersion in turbulent flow; see, for example,

Eq. (9.27) for the covariance tensor of fluid particle displacement.
The fact that the variance of fluid particle displacement in small time
-r actually is proportional not to i but to x* [see Eq. (9.28)] is a
related phenomenon. Finally, the function X { x , t ) also is not a
Markovian random function; for example, the probability densities
p ( X , /|jf,/o )an d p{X, t\x,Vo,to) for the variable X (t) for a fixed
value of X (to) = x and fixed values of
which for a Markovian function must coincide, will indeed be

different, particularly for small t = ^ — 4 where the second of these
probability densities will be essentially different from zero only
when X ^ x - { - Vo-t, but the first will not have this property.
However, if we consider the functior t) only at a discrete set
of times t = tn — to + m where the time step t is large compared to
the Lagrangian integral time scale T, then it is possible in practice to
regard the random sequence X { x , t n ) as Markovian. The increments
of the function X { x , t ) in nonoverlapping time intervals of length
t ' ^ T are, in fact, practically uncorrelated, and it is natural to expect
that they will also be almost independent; but a random sequence
with independent increments is certainly Markovian. Let us also note
that the variances of the increments of the functionX(jut, Oil' time
intervals of length ■z '^ T , according to Eq. (9.35), are proportional
to 1 , as should be the case for a Markovian function according to
Eqs. (10.53). Thus, if we considered instead of the differential
equation (10.49) a related difference equation in time describing the
Markovian sequence X { x j n ) , it would closely correspond with the
real properties of fluid particle motion in a turbulent flow.
Consequently, the semiempirical equation of turbulent diffusion
(10.49) may be expected to be significant, but only as the
differential analog of a difference equation in time with a step
Thus, Eq. (10.49) may be used to describe the field iT{X,t) at a
fixed time t where t — t o ^ T . in other words, equations for the
mean admixture concentration ‘Q { X ,t) obtained from the solution
of Eq. (10.49) may be considered asymptotic and correct when
t In practical problems related, for example, to admixture
diffusion in the atmosphere, it is usually of interest to describe the
diffusion on scales which significantly exceed the Lagrangian integral
time scale T, which in the atmospheric surface layer usually has a

value on the order of a second. Therefore, it is usually possible in
these problems to use the semiempirical equation of turbulent
diffusion. Incidentally, even for large values of ^ — ô, the semiempiri
cal equation may lead to incorrect results at very large distances from
the admixture source. This is because this theory actually assumes
that the rate of propagation of the admixture is infinitely large (see
below. Sect. 10.6).
In the following we shall usually deal with the case of stationary
turbulence homogeneous in planes X 3 = Z = const with a mean
velocity u = f/directed everywhere along the axisO î = OX. in this
case, the eddy diffusivities K a , like all other turbulence character
istics, may depend only on the coordinate Z. In addition, from a
cursory examination of the problem, it seems natural to assume that
the coordinate axes OXy = OX OX 2 = 0 7 and OX 3 = OZ coincide
with the principal axes o f the tensor Kjj, since OX, OY, and OZ are
clearly the preferred axes of the flow. Thus it may be assumed that
the semiempirical equation of turbulent diffusion (10.49) in this case
takes the following form:
^ + m Z ) § = K .A Z )S ^ + K „ ( Z ) ^ + ^ K , A Z ) § -
(10.55)
This form is quite standard, it is expounded and used in many

textbooks and monographs [see, for example, Sutton (1953),
Haltiner and Martin (1957), or Pasquill (1962b)] and in hundreds of
papers. However, there have been some recent arguments that
indicate that the standard and generally adopted form of the
diffusion equation (10.55) cannot be exact [see, e.g., Calder(1965)
and Yaglom (1969), where additional references can be found]. We
shall discuss these arguments in Sect. 10.5 below; however, we shall
first consider some results obtained with the aid of Eq. (10.55).
if gravitational settling of the relatively heavy diffusion particles
must be considered, which takes place with a constant velocity w,
d^
then it is necessary to add the term — w to the left side of Eq.
(10.55). in exactly the same way, the possibility of radioactive or
some other decay of the admixture [with a half-life of (ln2)/a] leads
to the occurrence in the left side of an additional term aO. if the
flow region extends to infinity in some direction, then the
corresponding boundary condition usually will be that •d(X ,t)
decreases sufficiently rapidly as the point X becomes infinitely

distant from all the admixture sources. The boundary conditions on
the solid walls must be selected from an analysis of the physical
processes which take place on these walls, but the solution obtained
must correspond in every case to the Markov random function
X { x ,t) [the enumeration of all these conditions is the subject of the
works by Feller and Wentzel mentioned after Eq. (10.2)]. In accord
with statements made at the beginning of Sect. 10.1, a boundary
condition on a solid wall Z = 0, sufficiently general to describe
practically all situations encountered, is a condition of the following
type;
-^ + ~ when Z — 0, (10.56)
where p is a constant having the dimensions of velocity. (For

generality, here we even assume the presence of gravitational settUng
with a velocity w which leads to the addition of the flux — iiy# to
the vertical turbulent admixture flux — A T^^-^^The case p = 0
corresponds to reflection of the admixture from the wall, the case
(J = oo, admixture absorption, and the case 0 < p < c», the inter
mediate situation of partial reflection and partial absorption. For
more detailed discussion of the condition (10.56) see Monin (1956a;
1959b) and Calder (1961).
10.4 Diffusion in a Field of Homogeneous Turbulence and in

Simple Shear Flows
Diffusion from Sources in Homogeneous Turbulence

We shall now consider the simplest case of diffusion in an idealized
model of homogeneous and stationary turbulence with a constant
mean velocity u = U, which, as usual, we shall assume to be directed
along the axis OXi = OX. Let us direct the coordinate axes
OXi = OX, 0 X 2 = OY and OX 3 = OZ along the principal axes of the
covariance tensorD,j('t) = Y\{'^)Y'j{'^) considering the OXiaxis as one
of the principal axes, and neglect molecular diffusion by comparison
with turbulent diffusion. Finally, we use the experimental fact that
the probability density p ( X \ x , t ) , in the case of homogeneous
turbulence, is close to the multidimensional Gaussian density for all
X = t — to, where, in addition, this circumstance has a theoretical
justification for large values of -t; thus, we can write the equation
» (X, t) - f f (x. y, z) X
— CO
(Y -yy {Z -zY
y exD I - ~2 D ^^ ~
,(x ) 2Dyy(x)
2 D,,(z)
~
2D^,(
2 D , , ( it))
1 dx dV d z
(10.57)
for the field of mean concentration i9'(A’, t), corresponding to the

initial value of •&{X, to) = ^p(.Y). Consequently to obtain a solution
for the mean concentration ■&(X,t) in any problem, it is necessary to
know only the variances D xx{t:), Dyyit) and Dzz{x) of the compo
nents of fluid particle displacement in the time x, which are
expressed by Eq. (9.31) in terms of the corresponding Lagrangian
velocity correlation functions (x), 5yy’(t)and In particular,
when
^ - ^ o = ^ » 7 ’ = max(7'^, T,),
where 7’*, Ty and Tt are the three Lagrangian integral time scales, it is
possible to assume that
D,,(x)^=2te»'Y,x = 2/C,,x.
Then the right side of Eq. (10.57) will coincide with the solution of
the paraboUc partial differential equation
dt ' dX dX'^ ' ' ^ y y 0 Y‘ dZ‘
which is a special case of Eq. (10.5 5), for the initial condition /o) =
^>o(AT). If ■do(^) = Q6(A') = Q6(X)6(y)S(Z) (the case of an instan
taneous point source at the point X =0), then the distribution (10.57)
will describe an ellipsoidal admixture cloud which decreases in concen
tration with distance from the center according to the Gaussian law.
The center of this cloud is at the point (t/x, 0, 0); that is, it is displaced
with the mean velocity of the flow, and the maximum concentration
at the center is proportional to
(Specifically, when it decreases with increasing -t, propor-

tional to x” ^.) If ô(A’) = Q6(X)6(Z), the case of an instantaneous
linear source along the 0 7 axis, then for any z the admixture cloud
will have the form of a cylinder of elliptical cross section with an axis
on the line X = Ux, Z = 0. In this case the maximum concentration
is proportional to )]-'/», or, for the large values of -r, it
decreases proportional to t ' ‘. For a steady continuously active point
source of constant output Q at the point A" = 0, Eqs. (10.6') and
(10.12) must be used, in which case, when X ^ U T , we can replace
Diiiz) by 2KiiX in the latter without substantial error. With this
substitution the integral with respect to t in Eq. (10.6') is easily and
explicitly carried out, and as a result we obtain
f u \,yk(X ^ , \k v ll
X { 2KxX ( Kxx + Kyy
(10.59)
Equation (10.59) describes the admixture jet which has the form of
an elliptical paraboloid oriented along the OX axis. When
Y this formula may be transformed to
Equation (10.60) shows that for fixed Y and Z at a large distance

from the source, the admixture concentration decreases approxi
mately proportional to .X*', and for fixed X and increasing distance
from the jet axis, it decreases exponentially. In exactly the same
way, for a stationary linear source on the OY axis of output Q (per
unit length), Eqs. (10.8) and (10.12) for X ^ U T , that is, when it is
possible to assume that
D xx = 2 K xxT, D „ = 2 /< „ t,
lead to the equation
< ->
where Ko{z) is the so-called Basset (or Macdonald) function (the

modified Bessel function of the third kind). When UX':^Kxx and
Z X, Eq. (10.61) may be simplified, taking the form
W T X T Z ) ^ ------- ^ ---- i r e x p / ------ (10.62)

^ ’ 2 {kK,,UX)'I‘ 4K,,X )
It is obvious from this that for a steady linear source, the admixture
concentration in a field o f homogeneous turbulence with a constant
mean velocity decreases when X ^ o o (for fixed Z) approximately
proportional to X-'Ik
Longitudinal Admixture Dispersion in a Straight Tube or Channel
Let us now proceed to the significantly more complex problem of
turbulent diffusion in a flow with a velocity gradient. Interesting
results may be obtained here only for a few specific types of flows.
We begin with the practically important problem of longitudinal
diffusion o f the admixture in a long straight tube {or straight
channel). The importance of such diffusion is connected in particular
with the fact that measurement o f the velocity of the longitudinal
transport of the admixture in a tube frequently proves to be the
most acceptable method of measuring the average flow velocity for
turbulent flows and even for laminar ones. Thus longitudinal
dispersion came under scrutiny rather long ago; for example, the
study of dispersion during laminar flow in blood vessels and in glass
capillaries by G. I. Taylor (1953), and the comparatively early work
by Allen and E. A. Taylor (1923) on dispersion during turbulent
tube flow. The experiments described in these works were such that
at some time 4 , a certain mass of admixture was introduced into the
tube at some point, or, in other words, an instantaneous point source
was simulated. Then at some distance X downstream from this point
the change with time in the average admixture concentration over a
cross section of the tube, fl'mCO was measured. Naturally, for some
time after the introduction of the admixture into the tube its
concentration at the point X will equal 0. Let t'« = /,,+ be the time
at the point X at which the admixture is first detected; then the ratio
«niax = will be equal to the maximum rate of flow in the tube,
or, in the case of axially symmetric tubes, the flow velocity on the
axis of the tube. The average rate o f flow Wav may be defined by the
Ugy = X/t) where /o + ti = is the time corresponding to some
preferred point of the function
According to data obtained in straight circular tubes, the
longitudinal admixture dispersion has the following distinctive
features: 1) the curves ^m{t) which correspond to sufficiently large
distances X, by comparison with the tube radius R, are found to be
approximately symmetric with respect to their maxima, although the
spatial distribution of the longitudinal flow velocity Ux{r) [where ris
the radial coordinate] over the cross section of the tube is not
symmetrical with respect to the average velocity «av; 2) if the time
tx = to~\----- corresponds to the maximum of the curve then
Wav
coincides with the bulk average velocity t/av defined as the ratio of
the volume of fluid flowing out of the tube per unit time to the area
of its transverse cross section. In other words, the center of gravity of
the longitudinal concentration distribution is displaced downstream
with a velocity equal to the average bulk velocity t/av; 3) the width
of the curve defined as the square root of the ratio
increases with an increase in the distance X, or, equivalently, the

time Ti = XlUav , in proportion to (or ). That is, the variance
has the form a? =2K\X\, where Ki is some constant; 4) the
’ m m
variable K = KiUly which obviously plays the role of an “effective
longitudinal eddy diffusivity” has a value which exceeds many times
that of the radial (turbulent or molecular) diffusivity of the
admixture. The second feature is of particular interest since it shows
that the definition of the maximum of the curve #,„(/) enables the
measurement of the bulk average velocity U.^y, which is most
significant for most hydraulic and other engineering applications;
also the third feature is important since it shows that the length of
the tube segment, symmetric with respect to the point X = U^vXi and
containing, e.g., 90% (or 95%, or 99%) of the diffusing admixture.
increases with time as x'/* so that its ratio to the average distance
X = t/avTi from the admixture source continuously decreases. Since
the section filled by the admixture moves with an average velocity
Uav, the fluid upstream from this section on the axis of the tube, and
consequently moving with a velocity exceeding overtakes and
passes this section with time.
The special features of longitudinal admixture dispersion in tubes
were theoretically analyzed and experimentally checked in the
brilliant works by G. I. Taylor (1953; 1954a, b), two of which are
devoted to laminar flows, and the other (1954a), to turbulent flows.
These works stimulated interest in these phenomena and promoted
several new theoretical and experimental investigations [cf. Batchelor,
Binnie and Phillips (1955); Aris (1956); Elder (1959); Ellison (1960);
Saffman (1962b); Philip (1963); Tyldesley and Wallington (1965),
and others] which refined Taylor’s results and expanded their range
of applicability. In particular, it was discovered that the above-
mentioned general properties (1)—(4) of longitudinal admixture
dispersion may be explained simply according to the general
properties of the fluid particle motion in striaght tubes and channels.
As noted at the end of Sect. 9.3, longitudinal dispersion of a fluid
particle in a straight tube, or a straight channel, with sufficiently
large time of motion t — to, is subject to the same laws as are in
force for homogeneous turbulent flow with a constant mean velocity

u = Uav. Consequently, the mean value of the longitudinal coordi
nate X { t o + t ) of the fluid particle which at the time to leaves from
the point X = 0 will be determined quite accurately for large x by the
expression X ( t o + t ) ~ f/avT, and the variance of the coordinate
X { t o + t ) will have the form
(X — x y ~ 2Ki, where K — const.
(In Sect. 9.3 this constant was written in the form wf r,.) For the
probability density pi(X|t) for values ofA’(/o + 'c), the central limit
theorem of probability theory gives substantial grounds for assuming
that for large x this probability density will be close to normal. Thus
it follows that the longitudinal distribution of the admixture cloud
from an instantaneous point source after a long time t will have a
bell-shaped Gaussian form with a mean value Lâv't and a variance
2/ ( t . If instead of ^ = 1 — to, we fix the value of X, and r(X) is the
time of passage of a fluid particle through a cross section at distance
X from the source, then -jt-----, where V^.y, is the mean
Lagrangian velocity of the particle in the time from U to/o + 'c(X).

Since, with an increase in K, the fluctuations in the passage time t(X)
become relatively less important, and x(X) approximates XjUsy, for
sufficiently large X the probability distribution of the mean
Lagrangian velocity 'r(X)will, with high accuracy, coincide with the
probabihty distribution of the mean Lagrangian velocity
V'x/£/av = ^ ( ^ o + - ^ ) -
which is averaged over the time interval XlUay. However,
^ (/o + 7 ^ ) = that is x ( t o - \ - - ^ ) = X + X';
consequently, for sufficiently large X
( X' \2
having a mean value of order 1/X,
and the subsequent higher-order terms, we find that for large values
of A’
where X' is a normally distributed random variable with zero mean

value and variance 2 K - ^ - Consequently, the variable t ( X ) will also
— X
have a normal probability distribution, with a mean value t = -rj—
C'av
_
and a variance 2K\^ where K\ = — These deductions are clearly in
” av
full agreement with the empirically estabhshed properties ( l ) - ( 3 ) of
longitudinal admixture dispersion. Property (4) mentioned above
may also be explained simply. In this case the longitudinal dispersion
of the admixture cloud is due largely to the influence of a transverse
mean velocity gradient; specifically, it is generated by the velocity
differences u (r) — t/av which substantially exceed both the turbulent
. m h t i c l e d is p e r s io n i n a t u r b u l e n t f l o w 621
fluctuations of the longitudinal velocitym' = u(/-) — u(r), producing

turbulent diffusion, and, even more, molecular fluctuations which
cause molecular diffusion.
However, in estimating the numerical values o f the coefficients /<
and A'l, the arguments above concerning the Lagrangian description
of the flow are of very little help. Indeed, the eddy diffusivity K is
expressed in terms o f the Lagrangian integral time scale Ti; however,
this scale is determined by the Lagrangian velocity correlation
function which is very difficult to relate to the directly measured
Eulerian statistical characteristics, and is usually evaluated only from
the data of diffusion experiments. In the case of longitudinal
admixture dispersion in a tube or channel, the value of K may be
obtained quite accurately using the semiempirical diffusion equation
for the Eulerian concentration field ^ /); this method was also
used in the influential works of Taylor. Since we are interested in the
study of asymptotic laws for large t (or X), we may expect that the
semiempirical equation will be quite accurate. Its use also has the
important advantage that the same theory may be applied to
dispersion both in turbulent and in laminar flows; for the latter, the
approximate semiempirical equation is replaced by an exact molecu
lar diffusion equation (10.1). Following I. Taylor (1953,- 1954a), we
assume that the mean concentration 0 does not depend on the
angular variable q), and we may write Eq. (10.55) or (1 0 .1) for
diffusion in a straight round tube in the form
di) I ~ / \ I ^ ^ / \ { \n
4 - Ujc (f) j x ~~ 'dJO 17 ■ (10-65)
Furthermore, it will be convenient to transform to a system of

coordinates which moves with the mean velocity of the flow, that is,
to replace the longitudinal coordinate X by the coordinate x =
X — U^y {t — to)', in addition, instead of r we shall introduce the
dimensionless radial variable z = -^ where R is the tube radius. On
the basis of the fourth feature above concerning longitudinal
admixture dispersion in tubes (established by experiment), we can
attem pt to neglect in Eq. (10.65) the term describing the
axial diffusion of the admixture. (Consideration of this term hardly
changes the arguments, but leads only to a small correciton in the
final results which will be indicated below.) Then Eq. (10.65) in the
variables jc, 2 , i assumes the form
^9,1-/. /7 1 l i d ,, / X
dt
We shall use this equation to describe the concentration distribution

d(jc, z, t) which corresponds to the presence at the time ^ = ô, at the
point a: = z = 0, of an instantaneous point source of admixture. Let
us first compare the time intervals %c and xa during the course of
which the admixture concentration at a given point in the tube will
change appreciably due to the mean rate of flow only (convection),
and to transverse diffusion alone. We can put ~ a,^ where (/)
' m rn
is the standard deviation of the admixture concentration distribu
tion Om averaged over the cross section of the tube. On the other
hand, Xd is the time during the course of which transverse diffusion
smoothes the inhomogeneities in the concentration distribution over
f^2
the cross section of the tube, and one may a s s u m e whereKo
is a typical value (a scale) of the transverse diffusivity. Clearly, the
value of 0^^^ and, therefore, tc increases with the propagation time of
the admixture t , according to the theoretical arguments and
experimental data, asymptotically proportional to ; however, xa
does not vary with time. Therefore, for a sufficiently large x, or x,
the following inequahty will be satisfied: that is,
a, (10.67)
m Ao
Let us limit our investigation to sufficiently large t for which the

inequality (10.67) is fulfilled, and calculate the value Qi of the total
admixture flux through a tube cross section ,v = const. (We must
recall that x = X — so that the cross section a: ^ const moves
along the tube with a velocity of /7av ) Obviously,
Qi = / / K ( r ) - t / a v ] ^ ^ / 5 = 2u/?2 f (10.68)
where S is the cross section of the tube; consequently, Qi may be

different from zero only in the presence of radial inhomogeneities of
PARTICLE DISPERSION IN TURBULENT FLOW 62 3
the concentration 0. Under condition (10.67) these radial inhomoge

neities are smoothed out quite strongly; therefore, only very weak
inhomogeneities may exist after a sufficiently long time, that is,
those which are constantly created by the inhomogeneous convective
transfer of the admixture, and which adapt to this transfer in the
sense that the effects of convection and transverse diffusion are
balanced for them. Such inhomogeneities satisfy the equations
. (10.69)
(Here on the left side we have replaced by since the radial

inhomogeneities 0 — are quite small.) From Eq. (10.69) we
obtain
»W - » ( 0 ) 7 ^ I l“ v I*"')- 1
0 0
Using this equation, (10.68) is easily reduced to the form
where
0
1
. (10.70)
Substituting the expression obtained for Qi into the equation
_n2 I _f)
which expresses the law of admixture mass conservation, and which

may also be derived, for example, by integration of all the terms of
the diffusion equation (10.66) over the cross section of the tube, we
obtain
dt — A dx^ •
Thus it is obvious that except for a diffusion coefficient K defined

by Eq. (10.70), the longitudinal dispersion of the admixture with
respect to a plane which moves along the tube with an average
velocity U^v is quite similar to molecular diffusion, and therefore
leads to a Gaussian distribution of the concentration in the Ox
direction. Consequently, the spatial variance of the longitudinal
admixture distribution will equal 2 K{t — to), which means
m = 2 K \ ' ^ , where K i — Kj Uiy,, ■z — t — ta.
In addition, considering that according to Eq. (10.70) K is of order

ifi
—^ — , we can reduce condition (10.67), for which our arguments
Ao
are correct, to the form
(10.71)
It only remains to calculate the values of K for both laminar and

turbulent flows. In the former case in Eq. (10.70) for K, according to
Eq. (1.23), we must assume that Ux(z) = 2L/av(l — and equate
Krr = X where % is the molecular diffusivity of the particular
admixture. Then Eq. (10.70) for K is easily reduced to the form
0 0 -7 2 )
so that
AC _ / l RU^y\
7. ~ I 7 V,
With the exception of extraordinarily slow flows in especially fine

1 Rl/av
capillaries, the number y ------ significantly exceeds unity, so that
is not difficult to see that by also taking into account

molecular diffusion, that is, preserving the term x in Eq.
(10.65), instead of Eq. (10.72) we would obtain
48x
which usually is practically indistinguishable from Eq. (10.72).
Clearly, if the velocity «(z)w ere constant, then the longitudinal
diffusion would be determined only by the molecular diffusivity %;
thus, the presence of the transverse velocity gradient causes a sharp
increase in the longitudinal admixture dispersion. It is interesting to
note that, on the other hand, the transverse diffusion delays the
longitudinal admixture dispersion; according to Eq. (10.72), with an
increase in % the value of K decreases. If there were no transverse
diffusion, then the distribution of the admixture first concentrated
in a pipe section of very small length 6 X, and uniform in the cross
section of the tube, in a time x is converted into a uniform
distribution for which const in a section of length f7max "c
and dml-Y) = 0 outside this section. [This result follows easily from
the advection equation (9.13).] In the presence of transverse
diffusion it is converted into a Gaussian distribution in which almost
all the admixture is concentrated in a tube section of length L —
for sufficiently large t , which is very small compared with
Since the value of K is not difficult to determine experimentally,

using measurements of the concentration fl-,„(/)or OmCX), Eq. (10.72)
creates the possibility o f obtaining values o f the molecular diffusivi-
ties X from such data. This method for measuring the value o f %was
tested by G. I. Taylor (1953), who made experimental studies of the
diffusion of potassium permanganate in glass capillaries, and thus
obtained a value of the molecular diffusivity % of this substance
which agrees well with the data from other measurements. Taylor
(1954b) also gave an extensive discussion of this method for
measuring x-
In turbulent flow with sufficiently large Reynolds numbers that
the thickness of the viscous sublayer is a negligibly small fraction of
the radius R, the average velocity may be given by the velocity
defect law (5.41);
where u* is the friction velocity defined by the shear stress on the

tube wall. Considering that the turbulent shear stress in a round tube
has the form x = [wee Eq. (5.17')], and assuming that the
eddy diffusivity is exactly equal to the eddy viscosity, using the
Reynolds analogy in the terminology of Sect. 5.7, it is possible to
represent the radial eddy diffusivity in the form
dr
Using the values of the function /i(ii)obtained by processing the data

from Nikuradse’s measurements and those of other investigators,
G. i. Taylor ( 1954a) using Eq. (10.70) obtained a value of K = 10.06
[so that c ~ 10 in Eq. (9.36)]. It is not difficult to estimate the
correction to this value of K produced by longitudinal turbulent
diffusion by calculating the corresponding admixture flux through a
cross section of the tube
For a rough estimate one may make the simple assumption that
K..V (r) = Krr ( r ) then Q1 = - ^R^K' ^ ,
where for K' numerical integration gives a value/(' = 0.052^«.,. The

quantity K' is quite small compared to K, but since only an
approximate numerical value may be obtained in any case forK/^««,
it is reasonable to add K' to the value of K obtained above, that is, to
use the formula
K = \ 0 . \R a , . (10.74)
(The inaccuracy in the hypothesis used for calculating K' will have
little significance due to the relative smallness of the entire summand
K'.)
The result (10.74) is in good agreement with the data for
longitudinal propagation of an admixture in a sufficiently long
straight round tube. Thus, for example, the data of Allen and E.
Taylor (1923) concernmg the diffusion of salt in a water flow in a
straight, round, but not very long tube, lead to the value o f K/Ru:. =
10.6 for one series of experiments and K/Rii,,: - 11.7 for another [see
G. I. Taylor (1 9 5 4 a)]. Taylor in collaboration with Ellison conducted
similar experiments in which a significantly longer smooth tube was
used. A value of K/Rn,, = 10.0 was obtained under almost exactly the
same conditions as in the above theory. When using a tube with very
rough walls, it turned out that KIRii;, = 10.5. The dependence of the
concentration on time in all these experiments, at a sufficiently
large distance from the admixture source, corresponds closely to the
Gaussian curve; see, for example. Fig. 80 in which the data from one
of Taylor’s experiments is reproduced. In the case of bent tubes or,
for example, industrial oil pipes, which are usually not exactly
straight, the values of /<//?«- naturally turn out to be somewhat
different from the theoretical values given above for straight pipes; in
several cases here, KIRu-:, assumes values closc to 20 [G. 1. Taylor
(1954a)].
FIG. 80. Salt concentration in a fixed cross scction

o f a straight tube as a function o f time from Taylor’s
data (1954a).
Longitudinal admixture dispersion in a wide, straight channel is

bounded by the walls ZÔ and Z--//. (The role of the upper wall
may be played by the free surface of the liquid.) This problem was
investigated by Elder (1959), Ellison (1960), Saffman (1962b), and
Tyldesley and Wallington (1965). Elder used Taylor’s method
(discussed above), while Saffman used the elegant method of Aris

(1956). Tyldesley and Wallington solved the semiempirical diffusion
equation (10.55) numerically.^ In Aris’ method one first derives the
equations for the moments from the semiempirical diffusion equa
tion (10.55), or the equation of molecular diffusion (10.1):
0„„(Z, 0 = / / t)X ''Y '^ d X d Y . (10.75)

--00 —CO
For example, when m = 0 , n = 0, 1 or 2, and Eq. (10.55) is used, it is

easy to see that these equations will have the following form:
aO„o d
dt dZ
(10.76)
(10.76')
dt dZ
the equations for the moments 0oi, 0n, 602 and higher-order moments
look the same. The resulting equations may be solved sequentially
for the initial and boundary conditions obtained from the initial and
boundary conditions for the concentration ( /). The asymptotic
behavior of the solutions obtained will describe the mean admixture
distribution for large values of t in the planes Z= const, or
specifically, in a thin layer which surrounds the height Z, with
accuracy to moments of the same order as that at which the system
of equations for the variables (10.75) was cut off. Similarly, the
asymptotic behavior of the variables
^ T y ld e s le y an d W allin gton o b ta in e d nu m erica l s o lu tio n s for an e q u a t io n o f the form

(1 0 .5 5 ), w ith an a d d itio n a l term v{ z) d^! dy o n th e le ft side, b u t w it h o u t th e term
p r o p o r tio n a l to T h e y d e scr ib e d d if fu s io n fr o m a c o n t in u o u s ste a d y p o in t or linear
s o u rce an d fr o m an in s ta n ta n e o u s linear so u r ce in a flu id la yer 0 < / / fo r several s p e cific
m o d e ls o f th e vertical v e lo c it y a nd d iffu s iv ity d is tr ib u tio n s . T h e results o b t a in e d w ere
c o m p a r e d w ith data fr o m various a tm o sp h e r ic d if fu s io n e x p e r im e n ts . H o w e v e r , w e shall n o t
give fu rther a tt e n tio n to th eir w o r k here.
II
will, with the same accuracy, describe the mean admixture distribu
tion in the entire three-dimensional volume occupied by the fluid.
The final results obtained in this way will be the same as those when
applying the Taylor method. According to these results, if the
admixture is introduced into the flow at the time ^=/n at the point
(0, 0, Z) where 0 2 < //, and is not absorbed by the walls, then
the center of gravity of the admixture cloud at the time f=/o + 't
asymptotically, for large values of -t, will have the coordinates
that is, it will be horizontally displaced along the X axis with a

velocity i/av. Along the vertical the admixture will be distributed
asymptotically uniformly, in the sense that any fluid layer of fixed
thickness AZ for large t will contain the same amount o f admixture,
where the center of gravity of the admixture in each layer also will
be shifted with the asymptotically constant velocity t/av • However,
for different layers the coordinates of the center o f gravity of the
admixture will, for large -c, differ by a constant which depends on
the corresponding values of Z and the form of the profiles «(Z)and
Kzz{Z)- The admixture distribution in the planes X = const at the
time ^0 + T will, for large t , be close to a Gaussian distribution with
zero mean value and variance
n
a2 = 2 A ;x + c o n s t, = Kyy{Z)dZ. (10.77)
The admixture distribution in the planes Y = const will be close to a

Gaussian distribution with mean value X ^ t/avT variance
= K , = K + (K „ )^ . (10.78)
where
H p W
0 Lz
H
(^ .A v = -^ / (10-79)
0
[see Eq. (10.70)]. The dispersion of the admixture distribution in

the OX direction in a thin layer close to a fixed height Z will also
have the asymptotic form (10.78), but here the constant component
will, in general, depend on Z.
Saffman investigated longitudinal diffusion in a two-dimensional
channel between two solid walls, considering it as a possible model of
admixture propagation in the atmospheric surface layer beneath an
inversion layer which prohibits further rise of the admixture at some
fixed height H. Since this model is quite rough, no attem pt was made
to describe wind velocity profile u{Z) more than schematically; in
addition to the simplest case of laminar Couette flow where
« = -7 ^ = X= const
and consequently according to Eqs. (1 0 .7 8 )-( 10.79),
Saffman analyzed only the case in which
(Z) = (a + l)^-^j where a = const,
and, according to the Reynolds analogy,
d u ld Z
In this last case the integral entering into the first formula of Eq.
(10.79) may be carried out explicitly. We obtain
,3 Ul,H
Elder analyzed diffusion in a channel with a free surface at a height

Z = H and compared the results obtained with the data from
experiments in which he investigated the spread of a drop of
potassium permanganate solution on a slightly inclined channel along
which water was freely flowing. Here it was required that a form of
the mean velocity profile be selected which would correspond well to
a real flow on an inclined channel; the result turned out to be quite
sensitive to even small variations of this form. Elder proposed that
the velocity profile in an open channel be given by the logarithmic
form of the velocity defect law (5.43) with von Karman’s constant
x = 0.41; then, if Ka is defined using the Reynolds analogy and it is
assumed that Kxx = K„,i—Kzz. one obtains
(A'n')av(A'^^)av=0-068//W,. K = 5.86//«., ( i q 82)

a : , (5.86 + 0.068)//«, = 5.93//«,.
K {K )
According to the data from Elder’s m e a s u r e m e n t s , ~
0.23, which exceeds approximately twice the corresponding value in
Eq. (10.82); one explanation of this discrepancy may be that the
assumption of equality of the three eddy diffusivities Kxx, Kyy and
Kzz is incorrect. If, instead of this, we assume that Kxx = Kyy>K 2z^
and recall also that usually turns out to be less than and
[see the discussion following Eq. (5.27)], then the effective
longitudinal diffusivity becomes
Kx= (5.86 + 0.23)//w. = 6.1//««.
This value agrees quite well with that of 77^ = 6.3 obtained from
experiment. In fact, the agreement is even better than one might
expect, considering that the experiments were carried out for a
comparatively small value of the Re number, small t , and under
conditions which do not permit high accuracy in the data. According
to Ellison (1960), the agreement obtained must be considered largely
accidental since near a free surface the logarithmic form of the
velocity profile cannot serve as a good approximation. Consequently,
Ellison repeated Elder’s calculations, assuming the velocity profile to
be given by Eq. (5.49"). According to his results, the value oiKIHu,,
depends strongly on the value of the coefficient b which enters into
Eq. (5.49"); the value of b = 1.4 corresponds best to Elder’s

experimental data.
Diffusion in Free Turbulent Flow

The case just examined, of longitudinal diffusion of an admixture
in a straight tube or straight channel, is rather special since several
properties are retained which are characteristic of diffusion in a field
of homogeneous turbulence. From this viewpoint, diffusion in
turbulent jets or turbulent wakes behind solid bodies are more
typical examples of diffusion in a shear flow. However, theoretical
study of turbulent diffusion in jets or wakes using the semiempirical
diffusion equation is impossible without adoption of additional
semiempirical hypotheses to determine the dependence of the eddy
diffusivities, and also the mean velocity u, on the spatial coordinates.
These additional hypotheses are usually based on the mixing-length
approach discussed in Sect. 5.9. The results obtained in this way
frequently turn out to be quite satisfactory from a practical
viewpoint [see, for example, Hinze (1959); Abramovich (1963)], but
theoretically are not very well founded, and thus shall not be dwelled
upon here.
In investigating an admixture distribution in a plane X = const
located downstream from a continuous steady point or linear source
at a small distance from it, compared with the length scale L=----UT
where U is the typical mean velocity scale and T is the Lagrangian
integral time scale, several results may be obtained by transforming
to Lagrangian variables, if we expand the Lagrangian velocity in a
Taylor series (9.78). In particular, as noted by Hinze (1951; 1959),
in many cases the first approximation turns out to be useful,
according to which V{ x, i ) — a { x , (x, to)- Thus, X { x , t) ■■=
x -\-{ u -\-u ') ■ {t — (q) and Z{x, t) = w' • (t — to) where the values
of u, u' and w' refer to the space-time point (x, /o). It follows
from this that in a steady flow with a mean velocity directed along
the OX axis, the coordinate Z ( / ) o f the fluid particle which starts
from the point jc = 0 at the time will be equal in first
approximation to
Z ( X ) ^ ^ ^ X (10.83)
u-j-u'
at the time that it reaches the plane X = const, where X<cL. If

PARTICLE DISPERSION IN T U R B U L E N T FLOW 63 3
then according to this equation Z{X) « (w'!u)X\ therefore,

the probabiHty distribution for Z (^) differs here by a constant factor
from the probability distribution for ay'(0). However, if the values of
u'are less than m, but not negligibly small, then
Z (X ) __ w ' u'w ' I u'^w' (10 84)
Consequently, it is easy to express the moments of the random

variable Z{X) in terms of the moments of the Eulerian velocity
fluctuation » '= ( « ', w') at the point x = 0 , where usually we may
limit ourselves to only the first few terms o f the series (10.84). When
the equations obtained have fourth-order moments o f u', apparently
it is also permissible to replace them by combinations of second
moments, assuming that the corresponding cumulants are negligibly
small [see Batchelor and Townsend (1956), and Hinze (1959)]. In
spite of the obvious roughness of such approximations they
satisfactorily describe the main characteristic peculiarities of the
temperature distribution in the planes X = const of a turbulent jet
along the OX direction, into which a heated conductor is placed in
the fixed plane X = 0 [Hinze and Van der Hegge Zijnen (1951), and
Corrsin and Uberoi (1950; 1951)]. As an example, Fig. 81 shows the
temperature distribution obtained by Hinze and Van der Hegge
Zijnen. They introduced a thin wire heated by an electric current to
a temperature of 400°C into a plane jet flowing from a narrow
nozzle, at a distance of 100 mm from the nozzle and 6 mm below
the axis of the jet. The OX axis in Fig. 81 was taken parallel to the
direction of the mean velocity at the wire, and passing through the
wire from which the distance X was measured; different symbols
were used to designate the data from the measurements for various
values of X. In contrast to analogous data for a homogeneous
turbulence, here the temperature distribution is asymmetrical, or
more elongated toward the region of positive values of Z, which
correspond to greater values of the mean velocity than negative Z.
The results obtained agree with Eq. (10.84) since in this case one
may take w'= 0 and u'w '< 0. The mean value of the distribution
~uTis/
shown in Fig. 81 is equal to 0.036; in this experim ent-----=j- has the
same value. Thus, to calculate the mean value
_____ fs (X,Z) z dz
Z{X) --------- —
J a (X.z) dz
we may neglect the third term on the right side of Eq. (10.84). Let
us note also that if we try to describe the temperature distribution in
the jet by a semiempirical diffusion equation with a constant
coefficient/(j;, then for the conditions of Fig. 81 we will obtain
which contradicts the data.
as %
OA
0.2
^=0.036
1
-0 .6 -O A -0 .2 0.2 0.4 0.6 0.8 Z /X
FIG. 8 1 . T e m p e r a tu r e d is tr ib u tio n in several cross s ec tio n s o f a

tw o -d im e n s io n a l tu r b u le n t j e t in w h ic h a h e a te d w ire has b e e n
p la ced (a c c o rd in g to H in ze and V an der H egge Z ijnen).
In the other extreme case, that is for very large X, or, in other
words, a very large diffusion time x = t — some results may be
obtained using the self-preservation arguments discussed in Sects. 5.8
and 9.4. From these arguments it follows immediately that for a
stationary continuous admixture source located near the body
producing the wake, or th^ nozzle emitting the jet, the mean
concentration distribution 0 in any two planes const, for
sufficiently large values of X, where X is calculated from the source
in the direction of the mean flow velocity, will be similar to each
other. In exactly the same manner, for an instantaneous source also

located near the body or nozzle and acting at the time /=/o, the
mean concentration distribution in three-dimensional space will be
similar at two times t sufficiently removed from the time ta. On the
basis of symmetry arguments it is natural to assume that the
corresponding distributions will be symmetrical, their maxima being
located on the axis of the jet or wake in the plane X = const for a
continuous source located sufficiently close to the axis or near the
point (>Y(t), 0, 0), where X(t) is given by the equations from Sect.
9.4 for an instantaneous source. Here the results of Sects. 5.8 and 9.4
permit simple determination of the dependence of the corresponding
maximum concentrations Om(-Y) or on X or t although they do
not offer the possibility of evaluating the exact form of the
distribution 0(y, Z )ord(A ', Y, Z).
Let us begin with the case of a continuously active steady
admixture source, and use the fact that, due to the stationarity of
the diffusion process, the admixture flux through any plane X =
const must be constant. In a turbulent wake the velocity of
admixture transfer along the OX direction far from the body will
obviously be nearly equal to the free stream velocity. At the same
time the dimensions of the area in the plane X = const where a
noticeable admixture flux occurs from a point source at the point
x = 0 will increase in proportion to U ^ X 'i ^ in the case of a
three-dimensional wake behind a finite body, and in proportion to
L(X) ■X
for a two-dimensional wake behind a cylinder.

In a third case of a wake behind a cylinder and a linear admixture
source parallel to the axis of the cylinder, the admixture flux per
unit length of the axis OY must be constant; in the plane X =const a
unit segment of the OY axis corresponds to a rectangle in which the
concentration 0 differs significantly from 0, having an area propor
tional to L (X) X'Ik However, since the admixture concentration in
the plane X = const is proportional toi%,(A'), we obtain
for a p o in t so u r ce in a th re e -d im en sio n a l w a k e ,
K, {X) ■ X~^ for a p o in t so u r ce in a tw o -d im e n s io n a l w a k e ,
A" ~ fo r a linear s o u rce in a tw o -d im e n s io n a l w a k e .

Similar arguments are also applicable to diffusion in a plane

mixing layer o f two plane parallel flows and in turbulent jets,
including the diffusion of a passive admixture in jets of convective
origin. However, now the mean rate Ti of admixture transfer through
the plane X = const will not be equal everywhere to the fixed
velocity /7o, but will be a function of X, Y, and Z. It is significant,
however, that when the parameter X is changed, the function
y, Z) of Y and Z remains similar, or self-preserving, its
maximum value remains constant for a plane mixing layer and
two-dimensional convective jet, and decreases in proportion toX"* for
a dynamical three-dimensional jet (issuing into a space filled with the
same fluid), in proportion to X-V2 for a dynamical two-dimensional
jet, and to /V"'» for a three-dimensional convective jet. Furthermore,
the area of that part of the plane X = const where the concentration
Y, Z) is noticeably different from zero increases in proportion
to L^{X) for a point admixture source in a three-dimensional jet, in
proportion toL (A ) for a linear source in a two-dimensional jet or
plane mixing layer, and in the case of a point source in a
two-dimensional jet, in proportion to L{X)[D 22 {'ixW^ where is
defined by Eqs. (9.47) or by related equations with the aid of the
equality A^(tx)=X. Since the admixture flux is proportional to the
product of the concentration, the velocity, and the area, the
arguments above, leading to Eqs. (10.85), now lead to the following
relationships:
for a linear so u r ce in a tw o -d im e n s io n a l je t,
for a p o in t s o u rce in a th r e e -d im e n sio n a l je t and for a
linear s o u r c e in a m ixin g layer or a tw o -d im e n s io n a l
c o n v e c t iv e jet,
{^) X~^^^ for a p o in t so urce in th e tw o -d im e n s io n a l je t, ^ 10.85 )

X~^^^ for a p o in t s o u rce in the th re e -d im en sio n a l c o n v e c t iv e je t,
for a p o in t s o u r ce in a m ix in g la y er or a tw o -d im e n s io n a l
c o n v e c t iv e jet.
Here and below, if there is no indication that the jet is convective,

then we assume it to be an ordinary dynamical jet which issues into a
space filled with the same fluid.
In the case of an instantaneous admixture source, the total mass of
admixture must remain constant, or if the source is linear, the mass of
the admixture per unit length of axis OY must remain constant. How
ever, in thetime i , the dimension along the 0 X^ axis of the admixture
cloud from an instantaneous source is characterized by the length scale

[0;,(t)]'/s naturally coinciding with L[.Y(t)| to within a constant.
According to the results of Sect. 9.4, we obtain from this the
following equations for the maximum concentration 0,u(t) of the
admixture cloud from an instantaneous source formed in the time t
after release:
“ * for a p o in t so u r ce in a th re e -d im en sio n a l w a k e an d a linear

so urce in a tw o -d im e n s io n a l w a k e ,
" ‘*'3 for a linear so u r ce in a tw o -d im e n s io n a l je t,
for a p o in t s o u rce in a tw o -d im e n s io n a l w a k e and in a three-

d im e n sio n a l je t,
^ for a p o in t so urce in a tw o -d im e n s io n a l j e t a n d for a linear
( 10.86)
so urce in a tw o -d im e n s io n a l c o n v e c t iv e je t an d in a m ix in g
layer,
for a p o in t so u r ce in a th r e e -d im e n sio n a l c o n v e c t iv e je t.
-3
for a p o in t s o u r ce in a tw o -d im e n s io n a l c o n v e c t iv e je t and
in a m ix in g layer.
Most of the results of Eqs. (10.85)-( 10.86) belong to Batchelor

(1957); the remainder were formulated by Yaglom (1965).
Diffusion in an Unbounded Homogeneous Turbulent Shear Flow

We conclude this subsection by considering admixture diffusion in
an unbounded shear flow with a constant transverse mean velocity
gradient dux(Z)ldZ = r and a homogeneous and stationary field of
fluctuations u'{X, t). In Sect. 9.4 we have seen that the interaction
of a velocity gradient with transverse “vertical” turbulent diffusion
leads to a qualitative change in the laws of longitudinal dispersion
instead of a simple increase in the effective horizontal diffusivity
caused by such interaction in tube or channel flows. Specifically, the
interaction implies that the longitudinal dispersion becomes
asymptotically proportional to t^, and not to t, as is usually the case.
Therefore, an initially spherical admixture cloud in such a flow will
assume the form of an elUpsoidal spindle elongated along the OX
axis, the major axis of which is slightly inclined with respect to the
plane 0. Since we are assuming the field o f the fluctuations «' to
be homogeneous and stationary, there is no difficulty here with the
use of the semiempirical diffusion equation concerning the depend
ence of the eddy diffusivities on the coordinates; these diffusivities
63 8 S T A T IS TIC A L F L U ID M ECHANICS
are naturally considered constant in space and in time. Therefore,

one has reason to expect that in this case all the basic features of the
diffusion process will be explained by investigating the solution of
the differential equation
ci» I p 7 0'^ I /r I hr (1 0 ^ 7 )
for the initial condition A’, U) — 6 {X). It is not difficult to see that
the desired solution of Eq. (10.87) at the moment t = to + t will be a
three-dimensional Gaussian distribution with zero mean and second-
order moments
( 10.88 )
i.e., that it will have the form
i?(X, <0 + x) =
exp -----------(,(,.88.)
I 4K , tr^ K „ ,S /3 4K„, 4 K „ ,j
This result and some of its generalizations were obtained by Novikov

(1958); later, it was also discussed in somewhat more detail by Elrick
(1962). The moment method of Aris (1956), based on examination
of Eqs. (10.76), (10.76'), (10.76"), etc., was applied to the same
problem, but with = 0, by Smith (1965). He found that the
admixture distribution at a given height Z has, for larger = t - I q , its
center at the point ( FZr/2.0) and the variances ^ = r^K^^-c^/Gand
<7^ ^ 2KyyT; these results are in full agreement with Eq. (10.88').
Equations (10.88) show the great difference between F = 0 and
r ^ 0;they also agree with Eqs. (10.53), according to which the dif-
fusivities Ka are equal to one-half the derivative of the covariances
Dij (t) at t = 0. Comparison of these equations with the exact Lagrang-
ian equations (9.57), (9.57'), and (9.58) demonstrated quite clearly
that our new results are an asymptotic form of the exact equations
for the variances Z)n=Z);r* and Di 3 = Dxt when t;^>7’3,where, as usual.
Ku = w'^T3,Kxx = u'^Ti. However, the semiempirical diffusion theory

does not permit evaluation of the covariances Da for moderate values
of T. Thus, once more we have verified that this theory describes
only the asymptotic laws which are valid for large diffusion times.
The maximum of the multidimensional Gaussian density function
is obviously inversely proportional to the square root of the
determinant of the covariance matrix. Thus the concentration
in the center of an admixture cloud created at the time / = /oby an
instantaneous point source will decrease asymptotically in propor
tion to with an increase in t = t — to, in a field of homogeneous
turbulence with a velocity gradient, and not in proportion to as
was the case for homogeneous turbulence with constant mean
velocity. In exactly the same way, the maximum admixture
concentration from an instantaneous linear source on the OY axis
will decrease in proportion to instead of x"'. Knowing the solution
corresponding to the instantaneous point source, and using Eqs.
(10.6) and (10.8), it is also possible to find the concentration
distribution corresponding to stationary point and linear sources;
however, we shall not discuss this here.
We now consider an unbounded turbulent flow having a more
complex mean velocity profile u(Z) than the linear profile, and
having eddy diffusivities that are dependent on the coordinate Z . In
this case, the exact solution of the diffusion equation (10.55), which
corresponds to an instantaneous point source, cannot be given
explicitly. However, if the dependence o f the mean velocity and the
diffusivities on Z is given by sufficiently simple equations, or if the
functions u(Z),KXX(Z), Kyy{Z) and Kzz(Z) are power functions, then
the basic features of the diffusion process described by the
semiempirical equation (10.55) may be investigated using Eqs.
(10.76), (10.76'), (10.76"), etc., for the moments 6„m(Z, t). Here
also the interaction of the mean velocity gradient with the vertical
turbulent diffusion, described by the diffusivity Kzz, leads to
horizontal dispersion which, as a rule, for large diffusion time
x = t — to substantially exceeds the ordinary horizontal turbulent
diffusion in the absence of shear. Moreover, compared to the
horizontal dispersion in tubes and channels, in an unbounded space
the additional horizontal dispersion usually does not reduce to a
simple increase to some new value K>Kxx in the horizontal
diffusivity, but leads to the proportionality of the horizontal
variance to a higher power of t than the first. This will be
discussed in detail in the following subsection which investigates the
practically more important case of diffusion in a half-space Z > 0

where all the special features of this type of diffusion in an
unbounded space are retained.
10.5 Diffusion in the Atmospheric Surface Layer
The most important examples of turbulent diffusion mentioned at

the beginning of Sect. 10.1 are undoubtedly those connected with
diffusion of admixtures in the atmosphere. It is not surprising then,
that in turbulent-diffusion literature the works devoted to atmos
pheric diffusion are predominant; among such works, in particular,
are all the references mentioned at the beginning of Sect. 10.1.
Of greatest interest in the study of atmospheric diffusion is
diffusion in the surface layer of the air most directly related to the
hfe and vital activity of man. Here, the problem consists mainly in
the fact that the diffusion takes place in a turbulent boundary layer
that is thermally stratified in general, and fills the half-space over a
solid or fluid underlying surface, which we shall assume to be
homogeneous and take as the plane Z = 0. Therefore, the results of
the preceding subsection in which it was supposed that the fluid
either flows between walls in a tube or a channel, or fills all
unbounded space, are not directly applicable to atmospheric dif
fusion. However, in passing let us note that the model of diffusion in
a plane channel between the walls Z = 0 and Z = H may with some
justification be applied to diffusion in the atmospheric surface layer
in the presence of a strong inversion at a height H. In this inversion
the temperature increases with height, thereby leading to a sharp
decrease in the vertical exchange at H; see, for example. Rounds
(1955); Saffman (1962b); Tseytin (1963); Tyldesley and Wallington
(1965). Clearly, our main interest in the theory of atmospheric
diffusion is not in this model, but rather in diffusion in the entire
half-space Z > 0 which we shall investigate below.
Solutions o f the Semiempirical Diffusion Equation

for the Case o f Uniform Wind
We begin with the simplest approach using the semiempirical
diffusion equation. As usual we shall make the nonrigorous assump
tion that the axes of the coordinate system OXYZ [where the
OX-direction coincides with the direction of the mean wind u = «(Z)]
are the principal axes of the eddy diffusivity tensor AT,j==/(ij(Z). In
this case, the semiempirical diffusion equation assumes the form of

Eq. (10.55); consequently, to reduce all the basic problems of
diffusion theory to clearly stated problems o f mathematical physics,
we need only give somehow, the height dependence of the wind
velocity n and of the diffusivities Kv.v, Kvti and/Cjj.
Equation (10.55) may be solved explicitly only for a few special
assumptions regarding the height dependence of its coefficients. The
qualitative features of the solutions corresponding to different types
of admixture sources may be found in the simplest example of Eq.
(10.58), that is, an equation of the form (10.55) with constant
coefficients », /C.vv Kim, and Kz:. This simplest semiempirical
diffusion equation was investigated by Roberts (1923). Here the
solution for an unbounded space corresponding to the presence of an
instantaneous point source of output Q at the point = is
obtained from Eq. (10.12) on replacing X-by — (7cand Du(x) by
2K ht . Using this fact it is easily seen that for diffusion in the
()^
half-space Z > 0, with a boundary condition of “ reflection” = 0
when Z = 0, the solution which corresponds to an instantaneous
point source at the point (0, 0, H) at the time t=tn is given by the
formula
9 (A, K, Z, t) - X
{Z-Hf (Z+Hf -]
X exp{-
(10.89)
For the boundary condition of “ absorption,” that is, when

0(A', K, 0, t) = 0, the solution of Eq. (10.58) will differ from Eq.
(10.89) only in that the second component in the brackets will have
a minus sign and not a plus sign. Similarly, Eqs. (10.59) and (10.61)
may be used to obtain the solutions o f Eq. (10.58) in the half-space
Z > 0,with absorption or reflection conditions at Z > 0, correspond
ing to a stationary point or linear source at a height Z = //.Usually,
these solutions may be simplified by using the fact that u
and therefore the convection term u almost always significantly
exceeds the corresponding diffusion term
____________/7%v •
-t^xx - ~ oX “ ’
see, for example, the estimates o f Karol’ (1960), Walters (1964) and
Yordanov (1967), concerning certain special diffusion problems.
Therefore, A', , usually be neglected in Eq. (10.55). In Eqs.
(10.59) and (10.61) this corresponds to the limit as K,,, -^0. Under
this condition the solution corresponding to a stationary point
source of output Q at the point (0, 0, H) and the reflection boundary
condition at Z = 0, assumes the form
k}(X. K, Z ) = ------S = = e +e
4.tA- I' KyyK,,
(10.90)
For an absorption boundary condition Z = 0 , again we need only

replace the plus sign in the brackets of Eq. (10.90) by the minus sign.
According to Eq. (10.90), the admixture distribution along the line
X = const, Z = const transverse to the direction of the wind, for any
X and Z, is given by a Gaussian function with standard deviation
l2 K yyX \'l,
I — =— proportional to and inversely proportional to u When
\ U / _
X increases and Y and Z are fixed, the concentration decreases
asymptotically in proportion to X”', similar to diffusion in a field of
homogeneous turbulence in an unbounded space. For reflection of
the admixture, the maximum surface concentration
KZ Z
neuH^XKyyJ
is inversely proportional to the square of the source height; it is

reached when y = 0 at a distance from the source A' directly
proportional to H^. For the absorption boundary condition, the rate
of the admixture absorption by the underlying surface
z=o
as is easy to prove, decreases asymptotically in proportion to X \ In

this case the maximum rate o f absorption
_ 4Q5„ j j ^ y /2
■ r.eûH3 \ K y y l
is inversely proportional to the cube of the source height and is

achieved when K= 0 and The solution for a stationary
linear source along the straight line X=Q, Z = //p ro d u c in g per unit
time a mass Q of admixture per unit length, for the reflection
boundary condition, will have the form
2 (nuXKzz)'''- (10.91)
(In the case of absorption of the admixture by the earth’s surface we

must again replace the plus sign by the minus sign.) According to Eq.
(10.91), the surface concentration from a linear source decreases in
proportion to at large distances. The solutions of Eq. (10.58)
which correspond to the more general boundary conditions
may also be given explicitly, but they have a more complex form,
and hence are not discussed here.
At first glance the above-enjamerated results obtained using Eq.
(10.58) for the concentration ^ appear quite likely; however, they
do not agree at all with the data obtained from various field
investigations of diffusion in the atmospheric surface layer. Thus, for
example, experiments show that the surface concentration of
admixture, produced by a stationary point source in a neutrally
stratified atmosphere, decreases with distance from the source
approximately proportional to [according to Sutton (1952),
proportional to X"' , but in no way proportional toA'"'; the surface
concentration from a stationary linear source in a neutrally stratified
atmosphere decreases in proportion to X-' or X-°-^ [Sutton (1953);
Pasquill (1962b)], but not to X-'h The data on stable or unstable
stratification are fewer than for neutral stratification, but they also
show that in no case does the decrease in the surface concentration
follow the simple rules derived from Eqs. (10.90) and (10.91) [see,
for example. Deacon (1949)]. Thus, one must conclude that Eq.
(10.58) is unsuitable for quantitative description of atmospheric
diffusion in the surface layer.
The lack of correspondence obtained may be explained in
principle by the fact that the semiempirical diffusion equation
cannot be rigorously derived and is not exact. Clearly, this is

insufficient basis for concluding that the semiempirical diffusion
equation is inapplicable to atmospheric admixture dispersion. Indeed,
all the preceding conclusions were derived on the basis of a very
rough model in which the height dependence of both the wind
velocity and the eddy diffusivities was completely neglected.
However, it is well known that in the real atmosphere both the wind
velocity and the diffusivities increase with height (see above, Chapt.
4). Therefore, it is natural first to try to generalize the model,
making some reasonable assumptions concerning the form of the
functions m(Z),K*3c(Z), Kyy(2 )and/C 2z(2 ). (Another type of generali
zation connected with the inclusion of nondiagonal components of
the tensor /v,j will be considered later.) These efforts have been made
by many investigators; since the solution of the diffusion equation
with variable coefficients encounters significant analytical difficul
ties, these efforts have generated extensive literature, chiefly of
applied interest.
L et us n o t e first th a t n o single e x p lic it a na ly tic a l s o lu tio n o f a n o n sta tio n a r y d if fu s io n

p r o b le m in a ha lf-sp a ce Z > 0 , w h ic h is easily red u c ed to fin d in g e q u a t io n s fo r th e d if fu s io n
fr o m an in s ta n ta n e o u s p o in t s o u r ce , has a p p a re n tly y e t b e e n o b t a in e d in a n y case o f a w in d
v e lo c ity w h ic h varies w ith h e ig h t u ( Z ) . T h e re fo r e , to solv e th e p r o b le m o f d if fu s io n o f th e
a d m ix tu r e c lo u d fro m an in s ta n ta n e o u s so u r ce , s o m e a p p r o x im a te a s su m p tio n s are a lw a ys
u sed . U su a lly , o n e first c o n s id e r s o n ly th e vertical d if fu s io n p ro cess, th at is, th e tim e
d e p e n d e n c y o f th e to t a l c o n c e n tr a tio n at a h e ig h t Z is d e te r m in e d , w h ic h is e qua l to
[see a b o v e , E q. ( 1 0 . 7 5 ) ] and o n ly th e n is th e h o r iz o n ta l d if fu s io n also a p p r o x im a te ly

c o n s id e r e d . T h e fu n c t io n 0 o o (^ . 0 » w e k n o w , sa tisfies E q. ( 1 0 . 7 6 ) as
d
dt dZ dZ
T h e c o r r e sp o n d in g in itia l a n d b o u n d a r y c o n d it io n s , fo r an in s ta n ta n e o u s s o u r ce at a h e ig h t
Z = H SLt th e tim e t = to, and r e fle c t io n o f th e a d m ix tu r e at th e b o u n d a r y Z = 0 , w ill have
th e form
= 0, Ooo ( Z , 0 >0 for Z - > cx). ( 1 0 . 9 2 )

dZ~ z=o
T hus, th e a d m ix tu r e d is tr ib u tio n b e t w e e n th e various h o r iz o n ta l p la n es Z = c o n s t d e p e n d s
o n n e ith e r th e w in d p r o file no r th e h o r iz o n ta l d iffu s iv ities , and is d e te r m in e d u n iq u e ly b y
th e f u n c t io n K n { Z ) and th e s o u r ce h e ig h t H. S o lu tio n s o f Eqs. ( 1 0 . 7 6 ) un der c o n d it io n s
PARTICLE DISPERSION IN A TU R B U LE N T FLOW 64 5
(1 0 .9 2 ) for f u n c tio n s K t i { Z ) , su itab ly a p p r o x im a tin g th e p r o file o f th e v ertical e d d y

d iffu siv ity in th e a tm o sp h e r ic surface la yer un der various m e t e o r o lo g ic a l c o n d it io n s , w ere
fo u n d , for e x a m p le , b y L a y k h tm a n in th e 1 9 4 0 ’s; later, t h e y w e re a lso in v e stig a te d b y m a n y
oth er s [s e e , fo r e x a m p le , M onin ( 1 9 5 6 a ) ) . W hen Kzz = K \ Z this s o lu tio n has th e for m
Z+H
2 V'HZ (1 0 .9 3 )
K T lt- lo ) J
w h e r e lo d e n o t e s a B essel fu n c t io n o f an im aginary a r g u m e n t, or a m o d if ie d B essel f u n c tio n

o f th e first k in d o f z er o order. S in ce fo r neutra l th erm al str a tific a tio n Kzz ^ Z,
where a = = const 1.1 [s e e E q . ( 5 . 2 9 ) and S e c t. 5 . 7 ] , it is natural t o u se Eq.
( 1 0 . 9 3 ) in th e a b s e n c e o f a n o t ic e a b le vertical te m p er a tu re g rad ien t. In th e ca se o f a surface

so u rce, or w h e n / / = 0 , Eq. ( 1 0 . 9 3 ) is greatly sim p lifie d . Here
Z
(1 0 . 9 3 )
In th e m o r e g eneral case o f a p o w e r -la w h e ig h t d e p e n d e n c y o f th e e d d y d iffu s iv ity w h e n

Kzz = K \ Z ^ , w h e r e 0 < n < 2,^ th e s o lu tio n o f th e p r o b le m ( 1 0 . 9 2 ) for E q. ( 1 0 . 7 6 ) is
given b y th e fo llo w in g :
1 -/2
0Oo(Z, 0 = ------------- ----------^ '(2-n^Kr{t-to)x

2-n
2(ZH) (1 0 .9 4 )
2-n
W hen / / - > 0 w e o b ta in
,t-n
(1 0 .9 4 )
2-n
E ven m o r e a w k w a r d fo r m u la s for Qoo(Z, t ) are o b t a in e d w h e n using fu n c t io n s Kzz( Z)

c o n ta in in g a “b r e a k ,” th a t is, m a d e u p o f t w o a na ly tic a l e x p r e s s io n s co v er in g d if fe r e n t
h e ig h t in tervals. N e v er th ele ss, w h e n
(K,Z for
K,z{Z) (1 0 .9 5 )
I const for
th e c o m p lic a t e d a n a ly tica l r ep r e s e n ta tio n o f th e f u n c tio n 0 o o ( Z , /), in th e fo r m o f an
Ît can b e sh o w n th a t Eq. ( 1 0 . 7 6 ) has n o s o lu tio n s a tisfy in g th e c o n d it io n Oqo ( Z , t ) - >

6{Z — H) as t - > t o w h e n Kzi — K\ Z^ y ^ > 2. H o w e v e r , this fa c t is n o t im p o r ta n t in
a tm o sp h e r ic d iffu s io n th eo r y because th e case n'> 2 is n o t m et in a tm o sp h e r ic
a p p lic a tio n s.
64 6 STA T IS TIC A L F L U ID M ECHANICS
integral o f Bessel f u n c tio n s w as p o in te d o u t a n d e x a m in e d nu m e ric a lly b y M o n in ( 1 9 5 6 a ) .

Y o r d a n o v ( 1 9 6 6 ; 1 9 6 8 a ) in v e stig a te d tw o -la y er m o d e ls o f K i z ( Z ) , c o n s istin g o f t w o p o w e r
fu n c tio n s , th a t w ere e v en m o r e c o m p lic a t e d .
A c c o r d in g to th e sim ilarity th eo r y o f C hapt. 4 for th e a tm o sp h e r ic surface layer, b e fo r e
sele c tin g an a na ly tic a l fo rm for th e f u n c tio n ( ^ ) j it is e x p e d ie n t to transform to th e
d im e n sio n le ss variables
'^ = TTT’ " = T M ’ ^ “ TTT’ v:i^|TT’ Q

w here Z, is a natural h e ig h t scale in th e surface layer d e fin e d b y Eq. ( 7 . 1 2 ) . If, as usual, w e
a ssum e th a t th e e d d y d iffu s iv ity for h e a t e q u a ls th at fo r a n y passive a d m ix tu r e , th e n Eq.
( 7 . 2 1 ) e v id e n tly im p lie s th a t th e q u a n tit y K in Eq. ( 1 0 . 9 6 ) d iffe rs fro m th e R ic h a r d so n
n u m b er Ri, or in th e case o f u n sta b le s tr a tific a tio n , fr o m its m o d u lu s , o n ly b y th e fa c to r
ct2 = ^ ap p a re n tly , in m an y cases is n o t very far fr o m u n ity . H o w e v er , th e cases
o f e x tr e m e in sta b ility or s ta b ility are th e e x c e p t io n s . T h e f u n c t io n H (^, t ) T |n ) w ill

n o w be d e fin e d b y th e c o n d it io n s
d- ~ di: c/c ’ 6»c C=o (1 0 .9 7 )

e->o for c->co, e(C,0) = B(c —T)),
w h ic h are universal, th at is, d o n o t c o n ta in a n y m e te o r o lo g ic a l pa ram eters. T h e o r e tic a lly
speaking, here w e m u st c o n s id e r o n ly th e t w o cases o f u n sta b le and stable str a tific a tio n
co rr e sp o n d in g to t w o d iffe r e n t fu n c tio n s K ( ^ ) , and w e o b ta in fo r m u la s w h ic h are ap p licab le
in all real cases. F r o m a practical v ie w p o in t , it is c o n v e n ie n t to also add a third case o f
neutral str a tific a tio n , a lth o u g h in princip le it m a y b e o b t a in e d from cith e r o f th e t w o
p reced in g o n e s as \L\-> oo. In this third case, as w e have already n o t e d , it is e x p e d ie n t to
a ssum e th a t A ( s ) = or, m o r e p r e cise ly , / ( ( t ) = and to ap p ly Eqs. ( 1 0 . 9 3 )
and ( 1 0 . 9 3 ). A s for th e sta b le and u n sta b le str a tific a tio n , if th e c o rr e sp o n d in g universal
fu n c tio n s K { ^ ) w ere k n o w n w ith s u ffic ie n t a ccu ra c y , it w o u ld b e p o ssib le to solve th e tw o
p r o b le m s ( 1 0 . 9 7 ) n u m e ric a lly o n c e an d fo r aU w it h o u t a ny use o f a na ly tic a l a p p ro x im a tio n s
for K ( i ) , s e ttin g up tab les o f c o rr e sp o n d in g fu n c t io n s 0 ( ^ , r h i ) o f three variables.
H ow e v er , w e k n o w fr o m Se c t. 8 that th e fu n c tio n K { ^ ) is k n o w n o n ly in general fe a tu re s
for un sta b le and e sp ec ia lly fo r stable s tr a tific a tio n . M oreover, th e use o f sim ilarity th eo r y
u n der real c o n d it io n s is a lw a y s a c c o m p a n ie d b y co n sid e ra b le sca tter o f e x p e r im e n ta l data;
c o n s e q u e n t ly , w e are a lm o s t lim ite d t o using a p p r o x im a te e x p r e s s io n s for w h ic h can
be s e le c te d to p e r m it us to o b t a in e x p lic it fo r m u la s fo r 11 i]). In particular, fo r w e a k
stable str a tific a tio n or w h e n L is large a n d o n ly relatively sm all va lu es o f £ are o f in ter e st, it
is p o ssib le fo ll o w i n g L a y k h tm a n ( 1 9 4 4 ) and D e a c o n ( 1 9 4 9 ) , to a p p r o x im a te 7 ((^ ) b y a
p o w e r fu n c t io n o f th e fo r m K \ V ' w ith an in d e x n s o m e w h a t less th an o n e . F o r strong
sta b ility , or w h e n w e are in te r e ste d in a broad range o f values o f f , w e m a y use th e fa c t that
K i t ) = aoS for ^ - > 0 a n d /C ( C ) = c o n s t for s ^
(see S e c t. 7 ) and use a f u n c tio n K i O o f th e form ( 1 0 . 9 5 ) . Sim ilarly, for w e a k u n sta b le

s tr a tific a tio n it is p o ssib le t o a s su m e a p p r o x im a te ly th at /((g ) = w h e r e n is
s o m e w h a t greater th an u n ity . F or stronger in sta b ility a tw o -la y e r fo r m u la o f th e f o llo w in g
form
for =
/C(C) = „ (1 0 .9 8 )
for - C > C
and w h ic h is derived from Eq. ( 7 . 5 1 ) is m o r e su itab le: this case w as e x a m in e d by Y o r d a n o v

( 1 9 6 6 ; 1 9 6 8 a ) . F in a lly , fo r e sp e c ia lly stro n g in s ta b ility , w h e n values o f I ^ I w h ic h exceed
0 .0 5 are of special in ter e st, one m ay a ssum e that /((J ) = K \ l ’ and use Eqs.
( 1 0 . 9 4 ) - ( 1 0 . 9 4 ) w ith « = 4 / 3 .
T h e a b o v e -o b ta in e d results p e r m it q u ite a ccu ra te s em iem p irical d e s c r ip tio n o f vertical
d if fu s io n in th e a tm o sp h e r ic surface layer for a n y th er m a l s tr a tific a tio n . H o w e v er , for
h o r iz o n ta l d if fu s io n , th e sim p le st (b u t very ro u g h ) m e t h o d here is to rep lace a p p r o x im a te ly
th e w in d v e lo c ity u { Z ) and th e c o e f f ic ie n t s K x x ( Z ) and K u n ( Z ) b y the c o n s ta n ts (7av,
( K x x ) a v and (/(y.,/)av» equal to th e average va lu es o f th e c o r r e sp o n d in g fu n c tio n s o f Z
w ith in th e la y er b e t w e e n th e u p p e r and lo w e r b o u n d s o f th e a d m ix tu r e c lo u d . (T his m a y be
^ Ic u la ted in ad v a n ce fr o m th e f u n c t io n 0oo ( Z , t ) . ) T h e n th e c o n c e n tr a tio n
i9’(X , Y, Z, t) from an in s ta n ta n e o u s p o in t s o u rce o f o u t p u t Q at th e p o in t (0 , 0, H)
occu rrin g at th e tim e t = to, w ill be rep re sen te d in first a p p r o x im a tio n as
»(x, y, z, t) « -------------- 5— J-,----------X
X exp 4 (t - t o ) ' 4 ( Ky y ) ^ y ( t -- h/„ )) } 0„o(Z, 0. (10.99)
w h ere Goo [ Z, t) is th e s o lu tio n o f th e c o rr e sp o n d in g e q u a t io n ( 1 0 . 7 6 ) un d er th e c o n d it io n s

( 1 0 . 9 2 ) . S in ce th e c o e f f ic ie n t s K x x { Z ) and / ( , , , / (Z ) usually are rather p o o r ly k n o w n , th e
practical a p p lic a tio n o f Eq. ( 1 0 . 9 9 ) fr e q u e n tly turns o u t t o b e q u ite d iffic u lt and m a y lead,
for various values o f th ese c o e f fic ie n ts , to q u ite d iffe r e n t results. Even m o r e sig nifican t,
h o w e v e r , is th e fact that w h e n using e q u a t io n s o f th e fo rm ( 1 0 . 9 9 ) for any s e le c tio n o f Kxx
and Kyy w ith a c o n s ta n t average w in d v e lo c ity , w e c o m p le t e ly lose the im p o r ta n t e f f e c t o f
in ter a c tio n b e t w e e n th e vertical m ix in g and th e w in d shear. We have already seen ab o v e that
in tu rb u le n t shear flo w s in tu bes and c h a n n els and in an u n b o u n d e d space, this c f f e c t w ith
s u ffic ie n tly large d if fu s io n tim e, p lays a m ain part in th e h o r iz o n ta l a d m ix tu r e dispersion .
T hus it is clear th a t Eq. ( 1 0 . 9 9 ) m a y be a p p lica b le o n ly for c o m p a r a tiv e ly small values o f
/ = /o in th e a tm o sp h e r ic surface layer to o .
The Effect o f Wind Shear on Horizontal Atmospheric Dispersion

Saffman (1962b) used Aris’ method for an approximate account
ing o f the influence of the wind shear on the horizontal dispersion.
Here, the admixture distribution in the planes Z— const is described
by several first moments Qnm(Z, t) given by Eq. (10.75). We obtain
the famihar equations (10.76) for the moments 9oo. 0io, 620, etc., from
the semiempirical diffusion equation in which, however, the bound
ary conditions used in Sect. 10.4, == 0, on the upper bound
Z=H
of the channel Z = H must now be replaced by the condition

%nm (2, ^ 0 when Z oo. Since we are interested primarily in the
asymptotic behavior o f the admixture cloud from an instantaneous
source at time to when t — to-> oo,v/e can limit ourselves to a surface
source at the coordinate origin, i.e., the point ( 0 , 0 , 0 ), since it is
natural to suppose that the source height H will influence the
dispersion process only for a limited time. To examine the basic
qualitative peculiarities resulting from the interaction between the
wind shear and the vertical mixing, first let us consider the simplest
case where u ( Z ) = r Z and K^^{Z)=K= const. (The case of u(Z) =
Uo + TZ, Kzz = const which is of interest for a source at finite height
H, is easy to obtain from this with the help of a transformation to a
new inertial system of coordinates.) Consequently, the desired
solution of Eqs. (10.76) for 0 oo(Z, () will be
0oo(Z, n = 0 o o ( Z , /o + t ) =
(The symbol x here and henceforth will designate the difference

t — to.) The solution of Eq. (10.76') for the function 0io (Z, t) and
thus 0oo(Z, /) w h en /(„ = /(, il(Z) = T 2is more complex;however, it is
clear that it must be proportional to QF and, consequently, on the
basis o f dimensional considerations must have the form
6,0 (Z, t) = (z, ^0+ ^) =- Qr^ • ?io ( ^ ) • ( 10. 100)
Here(pio(E) is a universal function of one variable such that (pio(g) 0

when I oo; with the help of Eq. (10.76') this function is expressed
simply in terms of exp (—|^) and the so-called parabolic cylinder
functions of mathematical physics [see Saffman (1962b)]. This
holds also for the most important function 02o(Z, t), which describes
the variance of the admixture distribution in the plane Z = const in
the direction of the OX axis. According to Eqs. (10.76"), this
function depends also on the horizontal eddy diffusivity Kxx\ more
exactly, it is the sum of two terms, the first of which depends only
on u{Z) and Kzz = K but not on Kxx, and the second, on Kxx(Z) and
Kzz = K, but not on u{Z). The first o f these terms is obviously
proportional to Q P and, therefore, on the basis of dimensional
considerations must be written in the form o f the product
Q r 2/(-M^T)'/’(P2o(Z/(/(T)‘/0 ;
the general form of the second term may also be written down easily
if we assume that Kxx is a power function of Z. In particular, if
K x x — K i = const or K x x = K i Z where /('I = const, then the solution
of Eqs. (10.76) which we need can be written as
for =
k ,,= k :z.
Here (?), (^) and (^) are new functions which are damped at
infinity and which, it turns out, may also be expressed in terms of
the function exp (— and the parabolic cylinder functions. Similar
but more complex expressions may be obtained also for the
higher-order moments 0 nm(2 , t) [with n + m > 2 ] .
According to Eqs. (10.100), (10.101), etc., the moments o f the
admixture distribution in the plane Z = const at time t = to+'z are
power functions of t multiplied by functions of $ — which, as
V AX
is easily proved, assume finite values when | = 0 . Therefore, in any
layer 0 < Z < H of fixed thickness H, the moments of the admixture
distribution in all the planes Z= const have asymptotically identical
form when t^ o o . The total admixture distribution between the
various planes Z = const is described by the function
This function does not depend on u (Z) and Kxx(Z);it shows that for
a fixed value o f t the thickness of the admixture cloud is on the
order o f J/ /Cx. Moreover, if / / , then the admixture is
distributed vertically almost uniformly within the layer of thickness
H, with a constant vertical density 9(h) = - 7 = = -. According to
Saffman, the equations for the moments 9io(^. / a + t ) and
ha{Z, /()+t) at Z < ^ y /C t, obtained after substitution into Eqs.
( 1 0 . 1 0 0 ) and ( 1 0 . 1 0 1 ) of the explicit expressions for the correspond
ing transcendental functions f , „ ( 0 <p2o(^)> passage
to the limit | - ► 0 , have the following form;
Ojo = , 020 = — ^ 4-
4 “ 30K« '
2 Q/C, / X\V.
( 10 . 102)
+
for K .x^ K \Z .
Thus it ! ollows that the coordinate X of the center of gravity o f the

admixture distribution in the lower layer, of thickness
and the variance of this distribution in the OX direction, have the
following form:
2/C,x for = ATi,

I — ~.W2/('x3-u| K* ____ (10.103)
130 __.)r2/Cx3+ < K xx = K\Z.
We see that the center of gravity of the admixture cloud in the lower
layer is shifted by the mean flow along the OX axis with an
ever-increasing velocity r which is proportional to the
velocity gradient F and the square root of the vertical eddy
diffusivity. The variance of this distribution is made up of two
components. The first is proportional to or distinguished only
by a constant factor from the horizontal variance of an admixture
cloud diffusing in an unbounded shear flow; obviously it is produced
by the interaction of the vertical turbulent mixing with the velocity
gradient. However, the numerical coefficient of this component
7/30—7t/16 ^ 0.036 is equal to only about one-fifth of the
corresponding coefficient and one-sixth for the case of an unbounded
flow region; see the previous subsection. As a result, the presence of the
wall greatly suppresses the horizontal dispersion. The second
component in the expression for describes the ordinary horizontal
dispersion due to turbulent fluctuations of horizontal velocity, that
is, turbulent diffusion with coefficient Kxx- This second component
has the ordinary form 2Kir when = Ku but when f(xx = K*Z, it
also depends on Kzz= K, that is, it expresses the interaction of
horizontal and vertical diffusion, and is proportional to However,
in any case, it is essential that for sufficiently large values of t, it
turns out to be negligibly small compared with the first component
oF order x‘. Let us also note that the equation for Q,n(Z, /)will differ
from Kq. (10.76") by the absence of a term with u{Z), under the
assumption that the mean velocity for all Z is directed along the OX
axis. Therefore, the variance will be described only by the second
component of Eq. (10.103) for ol^, with the obvious replacement of
Kx.x by Kyy', that is, for large values of t it will b^m uch smaller than
The maximum surface concentration dm=&m(T), achieved ob
viously at the point (X), 0, will decrease asymptotically in propor
tion to when oo if /(yy= const and in proportion to if
Kyy'^Z. Clearly, all these facts in no way agree with the simplified
equation (10.99). In addition, this equation turns out to be incorrect
also in the sense that the admixture distribution in the plane Z =
const, for sufficiently large values o f r = t — to, does not resemble a
Gaussian distribution in form. Indeed, using expressions that are
similar to Eqs. (10.10 0 )-( 10.102) for the moment 630, it is possible to
show that
for sufficiently large x. Consequently, it follows that the asymmetry

of the admixture distribution in the OX direction, for a fixed Z and
T->- 00 , does not approach zero, but unity.
These results refer to the model in which uiZ) =VZ,K^^ = K =
const. However, it is natural that in other cases also, if the wind
velocity is height-dependent, a similar situation will exist. It would
be of considerable interest to investigate the case
u(Z) = In A K, = ,
" Zo
that is, diffusion in the logarithmic layer, and also the cases in which
u(Z) and are defined by the equations derived in Chapt. 4 for
a thermally stratified boundary layer. However, such an investiga
tion, based on the semiempirical equation of turbulent diffusion, has
serious analytical difficulties. Therefore, we shall limit ourselves
briefly to considering diffusion in the neutrally stratified boundary
layer [following Chatwin (1968)].
Let us write d(X ,Z ,t) = f X, Y, Z, t) dY, where 5 (X, Y, Z, t) is
— 00
the mean concentration from an instantaneous point source of

output Q at the point (0,0 ,H) at the time d(X,Z ,t) is also the
mean concentration from an instantaneous linear source along the
line X = 0, Z = H at the time Iq of output Q per unit length. Then the
semiempirical diffusion equation, in which the longitudinal turbulent
diffusion term will be neglected as being small compared with the
convection term, implies the following equation for ^(X, Z, t):
— + — lo g -------- = — xu*Z — , x ' = a%. (10.104)

dt ^ Z q dX dZ dZ
It is clear that the function 6 qq(Z, <) = / i?(X, Z, rt dX must have the
form (10.93), with = ku . Thus
ooôoCZ, OdZ
Z^0 0 (Z, OdZ J
0
= - <0 •
(10.104')
The simplest derivation of Eq. (10.104) is based on the equation

diZ)
------ = which follows from Eq. (10.104) and on the initial
dt
condition <Z> = H when x = 0. When H = 0, the formula for
<)is considerably simpHfied [see Eq. 10.93')], and the result
(10.104') may be quite easily and immediately obtained.
Henceforth, for simplicity, H will usually be taken as zero and Q
as unity. Then Eq. (10.93') easily leads to the result
00
<7^ 2 = <2 - <z»2 = J (Z -
° (10.104")
The center of gravity o f the admixture cloud from an instantaneous

source in the boundary layer rises with the constant velocity -a and
the cloud thickness, defined as twice its vertical standard deviation, is
equal to twice the distance from its center of gravity to the earth.
The examination of the horizontal dispersion of the admixture
cloud is more difficult. Nevertheless, if we consider the quantity
00 00
<Z> =
“
J I
I
- C00
O 0
0
\ XMX,Z,t)dXdZ,
then according to Eqs. (10.93') and (10.104)
d{X) 1 --------
= -In (X) = In - 11 (1 0 .1 0 5 )
_r
dt »
Zoe Zr7q By
where
= - I e ^ In xdx = lim [1 + 1 /2 +• • • + l / n - Inn] ~ 0 .58

« n-x»
is the so-called Euler constant. For H ^ 0 the computations are more

complicated; however, for this case Chatwin also obtained the exact
asymptotic result, when « ^
'in
(1 0 .1 0 5 ')
The evaluation ot the quantity
00 00
Js
-0 0 0
(X - {X))^MX,Z,t)dXdZ
is considerably more complex and requires tedious algebra; neverthe

less, Chatwin’s final result for this problem is also quite simple:
> ~ 0-8 ^ - 2«*t. (1 0 .1 0 5 " )
All the above-mentioned results are in full agreement with the

predictions of the Lagrangian characteristics in a nonstratified

boundary layer established in Sect. 9.4 by similarity arguments and
dimensional analysis. We see also that the approximate semiempirical
diffusion theory permits one to estimate the dimensionless universal
constants entering the general equations (9.60), (9.61), and so on;
with the same accuracy as that to which the semiempirical diffusion
equation is valid, we must take the estimates: fe = » '= ax ~ 0.45,
c = be-^ ^ 0.56 6 and Dg3 (t) = that is, = * '2 % q.2.
that is.
The degree of accuracy of these estimates will be discussed at the end

of Sect. 10.6.
Let us also note that it is not justified to make a direct
comparison of these results for diffusion in a boundary layer with
Eqs. (10.103) because these equations deal with the characteristics of
the admixture distribution in the bottom layer only. However, Eq.
(10.104) may also be used for the computation of the Z-dependent
characteristics of the admixture distribution at a given height Z:
00 00
<Xi Z, t ) ) = j x ^ d x /j MX ,
00 00
a, = J* - <X{Z,t)))^MX/ j MX.
It is clear that as « ~ any fixed height Z gets into the bottom layer
and therefore the asymptotic results for large t at any Z are
independent of Z. According to the computations of Chatwin, as
t = OO for any fixed Z
{X(Z,t)) = (X) —— ~ 0.6 ---- , 105'”)
where <X> is given by Eq. (10.105); these results can be compared

with Eq. (10.103). Equations (10.105"') show specifically that the
center of gravity o f the admixture at any given level lags behind the
center of gravity of the whole cloud. This fact is quite expected
since, for any Z, there is a time after which nearly all the admixture
is at a height greater than Z, and, consequently, has a horizontal
velocity exceeding the wind velocity at height Z .
G eneral r esults o f a sim ilar ty p e fo r d if fu s io n in a th er m a lly stratified b o u n d a r y layer are

u n k n o w n at p resen t. H o w e v er , w h e n c o n sid e rin g vertica l d if fu s io n un d er very u n sta b le
c o n d it io n s w e can u se an e d d y d iffu s iv ity o f th e fo r m ( 7 . 3 7 ) ; th e r e fo r e th e e x a c t s o lu tio n o f
th e c o r r e sp o n d in g e q u a t io n fo r 0 o o (Z , / ) can b e f o u n d fr o m Eqs. ( 1 0 . 9 4 ) a n d ( 1 0 . 9 4 Y It is
easy t o s h o w th a t
is th e m e a n h e ig h t o f th e a d m ix tu r e c lo u d d e s cr ib e d b y th e s o lu tio n ( 1 0 . 9 4 ) c o r r e sp o n d
ing t o th e d iffu s iv ity K J o f th e fo r m ( 7 . 3 7 ) . T h e r e fo r e th e sem iem p ir ic a l d if fu s io n
th e o r y im p lie s th e value
32------------- 3 . 5C - ; 3/2
3/2
for th e c o e f f ic ie n t a in Y a g lo m ’s e q u a t io n ( 9 . 7 3 ) ; in fa c t th is va lu e is a p p a re n tly hig h er th an

th e true o n e . T h u s w e shall b r ie fly d iscu ss S a f fm a n ’s resu lts ( 1 9 6 2 b ) fo r a m o d e l o f a
b o u n d a r y -la y e r f lo w sa tis fy in g th e e q u a tio n s
a(Z) = WiZ^, Kzz{Z) = KZ^

K rx(Z)==/<iZ\ Kyy{.Z)^KzZ\ (1 0 .1 0 6 )
w h ere U\, K, /Ci, K 2 are c o n s ta n ts and n < 2. Here, th e g eneral fo r m o f th e s o lu tio n s o f

Eqs. ( 1 0 . 7 6 ) , ( 1 0 . 7 6 ' ) , (1 0.76 ^ '), e tc ., fo r an in s ta n ta n e o u s surfa ce p o in t s o u r ce o f o u t p u t Q
at th e c o o r d in a te origin at th e tim e t — to, m a y again b e e st a b lish e d o n th e basis o f
t)' we n o w h a v e a u n iq u e
d im e n sio n a l c o n s id e r a tio n s . In particular, in E q. ( 1 0 . 7 6 ) fo r 6 0 0 ^ / ,
d im e n sio n a l para m eter K ; th u s fr o m K, 2 , an d x = t — /q , it is p o ssib le t o c o n s tr u c t a
u n iq u e d im e n sio n le ss c o m b in a tio n 8 = Z (/C t) ^ . We o b ta in
------L . ^ (1 0 .1 0 7 )
000 ( Z . t ) = Q(Kx) ^ „ „ (z
w here (poo (S ) is a d im e n sio n le ss universal f u n c tio n , th e e x p lic it for m o f w h ic h is given b y

Eq. ( 1 0 . 9 4 ). A c c o r d in g t o Eqs. ( 1 0 . 7 6 ) and ( 1 0 . 1 0 6 ) , H |o (Z , i ) m u st b e p r o p o r tio n a l to
Q w i, and in a d d itio n a lso d e p e n d s o n ly o n K, Z , a n d x ; c o n s e q u e n t ly ,
m —\
6 ,0 (Z . t) = (/Cx) <p,o ( Z (1 0 .1 0 7 ')
where cpio(£) again is a un iversal fu n c tio n . F in a lly , o n th e basis o f Eqs. (10.76^^) and
( 1 0 . 1 0 6 ) , 0 2 0 m a y b e r ep re sen te d as th e su m o f a c o m p o n e n t w h ic h is p r o p o r tio n a l to Qu\ y
and a c o m p o n e n t w h ic h is p r o p o r tio n a l to Q K\, w h ile 0o2 is p r o p o r tio n a l to Q
th er e fo r e,
2m-1
0 2 O (Z , t) = (ACx) ^0) ( 2 «))_ !_
k-\
+ O/C.X ( / C x ) 2 - « <f(2)(Z
(10.107 )
Oo2 { Z. t ) = Q K 2 - (/Cx) 2 ^ 902 i z ( /i C x ) - ’ ''(2-'»).
Sin ce 0 ( Z . / ) - > 0 w h e n Z -> c v o , all th e fu n c tio n s ( p o o (|) , ? io (^). ? 2 0 ( 0 >

decrease w h e n E > *•. T hus o n th e basis o f E q. ( 1 0 . 1 0 7 ) it f o ll o w s th a t th e th ic k n e ss o f th e
J_
a d m ix tu r e c lo u d at th e tim e ^) + T m u s t have th e order o f . In a lay er o f
__ i _
th ick n e ss / / < C I ^ ^ th e a d m ix tu r e d is tr ib u tio n w ill b e p ra ctically in d e p e n d e n t o f
th e h e ig h t and w ill have a d e n s ity
n __ «00Q
(/Cx)V(2-/*) ’
w here
_n__ _^
«oo = 9 o o (0 ); th at is, 0 ( , o ( 2 — n)
o n th e basis o f E q. (1 0 .9 4 ^ ). M oreover w e h av e m a d e th e a s su m p tio n th a t th e v e lo c it y « ( Z )

is parallel to th e O X a x is at all levels, w h ic h , in c id e n ta lly , m a y ea sily b e r ela x ed , i f
n e c essa ry, b y in tr o d u c in g a n o th e r c o m p o n e n t s ( Z ) = SiZ'"*); this im p lie s th a t th e c en te r
1
o f gravity o f th e a d m ix tu r e d is tr ib u tio n in th e lay er o f th ic k n e ss H w ill have
c o o r d in a te s
m
(A') = 6u,x (/C x)^. (K) = 0. 6= (10.108)
9oo
With th e sam e a s su m p tio n , th e v ariances o f th e a d m ix tu r e d is tr ib u tio n in th e p la n e Z = c o n s t
w h ere Z ^ w ill h av e th e for m

2m k I
4^ - { K - .) ^ 4 -h f < \ ' (10.108')
w h e r eA ], h^, an d 6 3 are n e w c o n s ta n ts , and th e j o in t m o m e n t
^00 ^00
w ill b e e q u a l t o zer o since Ogi = 0 ,, = 0. T h e first term o f th e first e q u a t io n o f (1 0 .1 0 8 ^ )

again d escr ib e s th e h o r iz o n ta l disp e r sion c a u sed b y th e in te r a c tio n o f th e w in d shear w ith
th e v ertical tu r b u le n t m ix in g , and th e s e c o n d term o f th is e q u a tio n , th e ordinary h o r iz o n ta l
tu r b u le n t d if fu s io n . Q e a r ly , th e first o f th ese e f f e c t s w ill b e p r e d o m in a n t fo r large t o n ly if
2m 2 — n > k. In th e a tm o sp h e r ic surface layer, th e latter c o n d it io n a p p a re n tly m a y
a lw a y s b e c o n s id e r e d as fu lfille d . (S in c e for stab le or n eutra l str a tific a tio n <; 1, it m a y
p r o b a b ly b e a ssu m ed th a t /i + /j 2, w h ile fo r str o n g in sta b ility n is c lo s e to 4 / 3 , and th e
c o e f f ic ie n t K x x ( Z ) / \ i it ca n b e u se d here at all, w ill s h o w a ba rely p e r c e p tib le increase w ith
h e ig h t .) T h e r e fo r e , it m a y b e c o n c lu d e d th a t th e p r e se n c e o f a w in d shear a lw a y s p la y s a
basic ro le in h o r iz o n ta l d isp ersion in th e a tm o sp h e r e over c o m p a r a tiv e ly large tim e intervals.
S in ce it is natural to s u p p o s e th a t K, j , f ( Z) is a p p r o x im a te ly p r o p o r tio n a l to K x x { Z ) , th en
n e g le c tin g th e tu rn in g o f th e w in d w ith h e ig h t, th e d isp e r sio n in th e d ir e c tio n o f th e m e a n
w in d fo r large t w ill c o n s id e ra b ly e x c e e d th e d isp e r sio n in th e p e rp en d icu la r d ir e c tio n . In
2
a d d itio n , th e standard d e v ia tio n here is q u ite large relative t o th e average th ick n e ss o f
the cloud, which is o f the order (/< t ) con sequ en tly, the admixture cloud for large T
is exten d ed largely in the direction o f the mean wind.
T h e e f f e c t o f w in d shear o n h o r iz o n ta l a d m ix tu r e d isp e r sio n in th e a tm o sp h e r e is
d isc u sse d also in HdgstrOm ( 1 9 6 4 ) , S m ith ( 1 9 6 5 ) , a nd T y ld e sle y and W allin gton ( 1 9 6 5 ) .
T h e se in vestiga tors u se d n u m e ric a l m e t h o d s and Lagrangian analysis in c o m p a r in g their
results w ith e x p e r im e n ta l data. H o w e v er , w e shall n o t d iscuss th e m in d e ta il here.
S u tto n 's Equations f o r Atm ospheric Dispersion
T h e e s tim a te s p r e se n ted a b o v e fo r h o r iz o n ta l d isp e r sio n in th e a tm o sp h er ic surface layer

crea te new p o s sib ilities for th e m a th e m a tic a l an alysis of th e p ro cess o f a d m ix tu r e
p r o p a g a tio n fr o m in s ta n ta n e o u s sources. It is clear, h o w e v e r , th at this a n alysis w'ill
u n a v o id a b ly b e q u ite c o m p le x ; th e r e fo r e , v arious rather r ou g h b u t sim p le a p p r o x im a te
p ro c ed u r es h a ve f o u n d a p p lic a tio n in th e p ractical d e s c r ip tio n o f a tm o sp h e r ic d if fu s io n . In
particular, in E ngland an d in th e U n ite d S ta te s, th e a p p r o x im a te e q u a tio n s p r o p o s e d b y
S u tto n ( 1 9 3 2 ; 1 9 4 9 ; 1 9 5 3 ) are fr e q u e n tly u se d to c a lc u la te a d m ix tu r e d if fu s io n in th e
a tm o sp h e r e . A c c o r d in g to th ese e q u a tio n s , th e a d m ix tu r e d is tr ib u tio n fr o m an in sta n ta n e
o u s p o in t s o u r ce is a ssu m ed to h a v e a G aussian fo r m ( 1 0 . 1 2 ) in th e c o o r d in a te s y ste m
d isp la c e d w ith th e m e a n w in d w ith a c o n s ta n t v e lo c ity w, b u t w ith variances D a { x ) w h ic h
increase m o r e rapidly th an th e first p o w e r o f i . (T his is in fu ll a g r ee m e n t w ith Eqs.
( 1 0 . 1 0 8 ) and w ith th e fa c t th a t th e d ec re a se in th e su rface c o n c e n tr a tio n c o rr e sp o n d in g to
th e variances D , , ( t ) = 2 K a X in p ra ctice is t o o s lo w .) T o d e te r m in e th e fu n c tio n a l fo r m o f
th e variances D a ( t ) , S u tto n p r o p o s e d th a t for d if fu s io n in th e a tm o sp h e r ic surface lay er, it
is p o s sib le t o u se a p p r o x im a te ly T a y lo r ’s classical fo r m u la ( 9 . 3 1 ) for strictly
o b t a in e d o n ly un d er th e a s su m p tio n o f h o m o g e n e it y o f th e tu rb u le n c e, assu m in g in th is case
th a t th e Lagrangian v e lo c ity c o rr e la tio n fu n c t io n s (x ) hav e th e for m
= • i = l . 2, 3. (1 0 .1 0 9 )
Here t i are th e c o n s ta n t tim e scales, and n a c c o r d in g t o S u tto n assu m es v alues o n th e order

o f 0 . 1 - 0 . 3 , d e p e n d in g o n str a tific a tio n . Sin ce it is assu m ed th at n < 1, th e Lagrangian

integral tim e scale T o f Eq. ( 9 . 3 2 ) turns o u t t o b e in f in it e ly large; th er e fo r e , S u t t o n ’s th eo r y
d o e s n o t lead to an a s y m p to t ic a lly linear in crease in th e v ariances fo r an in crea se in tim e t .
T h e re fo r e , it f o ll o w s in particular th a t, str ictly sp eak in g, th is th e o r y is n o t a v e rsion o f th e
ordinary sem iem p ir ic a l th e o r y o f tu r b u le n t d if fu s io n w h ic h is a p p lic ab le , as n o t e d a b o v e ,
o n ly w h e n T ^ ^ ^ T/. H o w e v er , sim ilar t o th e o rd in a ry sem iem p ir ic a l th e o r y , in S u t t o n ’s
th e o r y a lso o n ly s u ffic ie n t ly large values o f r are in v e s tig a te d , th a t is, large n o t b y
c o m p a r is o n w it h Ti , h ere e q u a l t o in f in it y , b u t b y c o m p a r is o n w ith t j . A c c o r d in g t o Eqs.
( 1 0 . 1 0 9 ) an d ( 9 . 3 1 ) w h e n t th e variance D a ( x ) is a s y m p to t ic a lly p r o p o r tio n a l to
; S u t t o n ’s c o r r e sp o n d in g a s y m p t o t ic fo r m u la is w r itte n as
w h e r e u is th e m e a n w in d v e lo c ity in th e a tm o sp h e r ic la yer un der d is c u ssio n , an d Cf are
c o n s ta n t c o e f fic ie n ts ; w h e n ^ th is in d ic a te s th a t ( u ) ~ ^. In th e case o f
a d m ix tu r e d if fu s io n fr o m e le v a te d so u r ce s, S u tto n r e c o m m e n d s o n th e basis o f e x p e r im e n t
th at th e c o e f f ic ie n t s C , b e c o n s id e r e d d e p e n d e n t o n th e so u r ce h e ig h t H a n d d ec re a se w ith
an inc r e a se in H. L et us also n o t e th a t w ith variances D a ( x ) d e fin e d b y E q. ( 1 0 . 1 1 0 ) , th e
f u n c tio n ( 1 0 . 1 2 ) is a s o lu tio n o f a d if fu s io n e q u a t io n o f th e fo r m ( 1 0 . 5 5 ) , b u t w ith th e
d iffu siv ities
w h ic h d e p e n d n o t o n th e h e ig h t Z , b u t o n th e d if fu s io n tim e t . H o w e v er , in th is ca se, Eq.

( 1 0 . 5 5 ) c a n n o t b e derived fr o m th e statistica l a s su m p tio n s o f th e sem iem p ir ic a l d if fu s io n
th e o r y w h ic h require [s e e Eq. ( 1 0 . 5 3 ) ] th a t th e d iffu s iv itie s b e d e te r m in e d o n ly b y th e
v e lo c ity fie ld a (AT, / ) , and n o t b y w h e n an d w h e r e th e a d m ix tu r e so u r ce s are in tr o d u c e d
in t o th is fie ld . (A s G. I. T a y lo r ( 1 9 5 9 ) n o t e d , th e fact th a t th e e d d y d iffu s iv itie s in the
sem iem p ir ic a l th e o r y cannot depend on t f o ll o w s fr o m th e s u p e r p o sitio n prin cip le,
c o rr e sp o n d in g t o th e a s s u m p tio n o f lin ea r ity o f th e d if fu s io n e q u a tio n : i f K a d e p e n d s o n t ,
K i i at th e f ix e d p o in t (A', /) w o u ld b e m u ltiv a lu e d in th e p r e se n c e o f several in s ta n ta n e o u s

a d m ix tu r e so u r ce s w h ic h d o n o t e m it s im u lt a n e o u s ly .)
K n o w in g th e c o n c e n tr a tio n d is tr ib u tio n fr o m an in s ta n ta n e o u s p o in t s o u r ce , it is e asy to
c a lc u la te th e c o n c e n tr a tio n s fr o m sta tio n a ry p o in t and linear so u rces; th is can b e d o n e
e x a c t ly as a b o v e for R o b e r ts ’ e q u a t io n s , w h ic h c o r r e sp o n d t o c o n s ta n t d if fu siv ities. L et us
c o n sid e r, fo r e x a m p le , d if fu s io n fr o m a sta tio n a ry p o in t so u r ce o f o u t p u t Q at th e p o in t (0 ,
0 , / / ) , u n d er th e c o n d it io n o f a d m ix tu r e r e fle c tio n b y th e surface o f th e earth. In th is case,
S u t t o n ’s th e o r y gives th e f o llo w in g fo r m u la rela ted t o E q. ( 1 0 . 9 0 ) :
iZ-Hf
( 10.111)
F or a linear s o u r ce o n th e straight lin e X = 0 , Z = / / , it is n ecessa ry o n ly to drop the

fac to r w h ic h d e p e n d s o n Y fr o m th e right side a n d r ep la ce Q b y
\^^QCyX 2 ,
A c co r d in g to Eq. ( 1 0 . 1 1 1 ) , th e a d m ix tu r e d istr ib u tio n a lo n g th e straight lin e X = c o n s t, Z =

const is Gaussian with a variance
and the distribution of the surface concentration in the direction of the mean wind has the
form
T.CyCjilX^
that is, it decreases asymptotically in proportion to A''* , but in the case of a linear
^-1
source, in proportion to . The maximum surface concentration from Eq. (10.112)
equals ------- -- • inversely proportional to the wind velocity and

TzeuH^ Cy
the square of the source height, and is achieved at a distance from the source
which, in contrast to the case of Roberts’ theory, does not depend on the wind velocity u.
A comparison of Sutton’s equations with the data from atmospheric diffusion
experiments, may be found, for example, in Barad and Haugen (1959), Drimmel and Reuter
(1960), Haugen, Barad and Antanaitis (1961) Brummage (1968), Csanady, Hilst and Bowne
(1968) and many others; the comparison of Eq. (10.109) with data on the Lagrangian
velocity correlation functions is made by Munn (1963). The material presented in the works
cited indicates that Sutton’s equations give a very inaccurate description of the actual
Lagrangian velocity correlation functions. They can be made to correspond more or less
satisfactorily to the existing diffusion data only if the range of allowable values of n is
considerably expanded, and if «, as well as the parameters Ci, may assume different values
for diffusion in different directions, that is, in Eq. (10.110) the parameter n should be
replaced by rii. If this is done, however, the number of adjustable parameters in the
equations of Sutton’s theory become so large that the good agreement of these equations
with the results of field experiments is fully explained; moreover, the dependency of the
parameters C, and rii on the meteorological conditions is quite complex and not very well
studied. Therefore, it is natural that other means are sought for the approximate description
of the atmospheric diffusion data, which would be sufficiently simple, but theoretically
better founded than Sutton’s equations.
Stationary Solutions o f the Semiempirical Diffusion Equation

In the special case of transverse dispersion Dyy(x), in the cross-wind direction, new
approximate equations were proposed by Byutner and Laykhtman (1963); however, these
equations are quite complex and require empirical testing. A number of practical equations
for calculating atmospheric diffusion may be found also in the monograph by Pasquill
(1962b). Here we shall discuss only a few of these equations derived from the semiempirical
equation of turbulent diffusion (10.55). Above, we investigated only the solution of this
equation corresponding to the case of an instantaneous admixture source; it is specifically to
this problem the remark applies that, as yet, no exact solutions have been obtained for the
diffusion equation iwth height-dependent wind velocity. However, in practice, the distribu
tion of concentration from a continuous stationary admixture source frequently is of pri
mary interest, and this distribution is consider^ly easier to obtain by mathematical analysis.
Let us begin with the concentration 0(X , Z) from a stationary linear horizontal
cross-wind source, producing Q units of admixture mass per unit time per unit length;
obviously it is also equal to the quantity
%(X. Z ) = j %{X. Y. Z) dY,
where Y, Z) is the concentration from a stationary point source of productivity Q. In

d d
this case, in Eq. (10.55), the terms containing and - ^ y can obviously be omitted. If, in
addition, we neglect also the term —K ^x ^ ^2 == which describes longitudinal
(streamwise) diffusion by comparison with the convection term u (Z ) (for stationary

sources this neglect hardly ever leads to appreciable error), then Eq. (10.55) becomes the
so-called two-dimensional diffusion equation:
u(Z) (10.113)
dX dZ
In the case of admixture reflection by the surface of the earth, and of a source at a height 7/,
the desired solution must satisfy the following conditions:
^ 0 for X CO or Z -> 00; Kzz

d Z z=o
- 0;
r= 0 for X = 0. Z ^ H ; J uJdZ^Q for X > 0 (10.114)
the last two of these conditions may also be written as a ( / /) d ( 0 , Z) = Q6(Z — //). Thus,
we obtain an ordinary boundary-value problem with initial conditions for the parabolic
partial differential equation (10.113), in which the role of time is now played by the
variable X. In the special case, when ii{Z) = u = const, K i i ( Z ) = Kzz — const, the
solution of the problem (10.11 3 )-(1 0 .1 14) obviously will be given by Roberts’ equation
(10.91), which, as is known, does not agree with the data. The case of constant wind
velocity w, but linearly increasing vertical diffusivity Kzi(Z) = KiZ, was investigated by
Bosanquet and Pearson (1936). For the special case of a surface source with / / = 0, they
obtained the very simple formula
» -(X < ■« » «
This formula is already in comparatively good agreement with the observational data from
nearly neutral temperature stratification. Furthermore, several investigators also studied the
more general case o f an arbitrary power law height dependency o f the w ind, and the
diffusivities
(1 0 .1 1 6 )
In the special case o f /i = 1 — m , or for the so-called Schmidt conjugate power law, Eq.
(1 0 .1 1 3 ) w’as solved comparatively long ago by Sutton ( 1 9 3 4 ) with these coefficients, but
for boundary conditions which differ from Eq. ( 1 0 .1 1 4 ), in con n ection with an
investigation o f the evaporation in the atmospheric surface layer; a problem similar to this
was also exam ined by Laykhtman (1 9 4 7 b ). The solution o f Eq. ( 1 0 .1 1 3 ) under the
conditions (1 0 .1 1 4 ) corresponding to a surface source, for the coefficien ts ( 1 0 .1 1 6 ), was
obtained, in particular, by O. F. T. Roberts in an unpublished paper [see Calder ( 1 9 4 9 ) ] ,
Frost ( 1 9 4 6 ) [under the assumption that n — \ — m ] and Laykhtman (1 9 6 1 ; 1963 ); it has
the follow ing form:
m+1
(m — n + 2) Q tn—n\-2
Z) = X
p/ tn -\- \ \ ^ m - n + 2y KiX
•V//Z- n+2
MiZ
X exp \m — n-^2YK^X (1 0 .1 1 7 )
(under the assumption that m — + 2 > 0). The more general solution o f this equation
which corresponds to an elevated source at any height H may be found, for exam ple, in
Laykhtman (1 9 6 1 ; 1 9 6 3 ), Rounds (1 9 5 5 ), Monin ( 1 9 5 6 b ) and Smith (1 9 5 7 ); in this case
1-/I
Q(HZ)
(m — « + 2 )/C ,A '
•X
(I, ( H Z j.(m-zn-SW ( 1 0 .1 1 8 )
X exp
(m—rt + 2)2 KxX P V ( m — /I + 2)2/C ,x )•
w'here / p is a Bessel function o f imaginary argument, o f order p ——

1—^ n
considerably more awkward solution o f Eq. ( 1 0 .1 1 3 ) with the coefficien ts ( 1 0 .1 1 6 ) under
boundary condition
d\)
dZ z=o z=o
which corresponds to the general case o f partial or com plete absorption o f the admixture,
was published by Laykhtm an_(1961; 1963); he, and also Tseytin (1 9 6 2 ), discussed the case
in which the wind velocity u and the vertical diffusivity K n satisfy the power equations
( 1 0 .1 1 6 ) only up to som e fixed height Z = being constant at greater heights. More
general and more com p lex two-layer power-law m odels o f the coefficients u{Z) and
Ki t i Z) in Eq. ( 1 0 .1 1 3 ) were studied, in particular, by Yordanov (1 9 6 8 a, b, c).
In th e ca se u ( Z ) = a , lii - - - , (^) — w h ic h c o rr e sp o n d s to th e b o u n d a r y
"o
layer u n d e r n eutra l s tr a tific a tio n , th e p r o b le m ( 1 0 . 1 1 3 ) - ( 1 0 . 1 1 4 ) has n o t b e e n solved
a n a ly tic a lly w ith suc ce ss, b u t clearly it m a y b e solved n u m e ric a lly . B erlya n d , G e n ik h o v ic h ,
e t al. ( 1 9 6 3 ; 1 9 6 4 ) o b t a in e d th e n u m e ric a l s o lu tio n fo r several values o f H. H o w e v e r , th ey
a ssu m ed th a t th e linear increase in th e d iffu s iv ity K z z ( Z ' ) is reta in ed o n ly up to s o m e
“virtual h e ig h t o f th e surface la y e r ” h, and th at b e y o n d this h e ig h t, K z i { Z ) r em ains
c o n s ta n t, or e v en ch a n g es w ith h e ig h t in an arbitrary m a n n er so th at th e graph o f th e
f u n c tio n K z z ( Z ) has several breaks. O th er e x a m p le s o f n u m e ric a l s o lu tio n o f th e d if fu s io n
p r o b le m fo r p o in t a n d lin ear s te a d y so u r ce s, w e re d e scr ib e d in K lu g and W ipp erm an n ( 1 9 6 7 )
a nd in B er ly an d and O n ik u l ( 1 9 6 8 ) . T y ld e sle y also o b t a in e d a n u m erica l s o lu tio n o f th e
tw o -d im e n s io n a l d if fu s io n e q u a t io n for a lo ga r ith m ic w in d p r ofile and a linear d iffu s iv ity
pr o file. H e fo u n d g o o d a g r ee m e n t w ith th e data fr o m e x p e r im e n ts in n eutra l c o n d it io n s [se e
th e d is c u s sio n o f Pasq u ill’s p ap er ( 1 9 6 6 ) ] . F in a lly , in Y a m a m o t o a n d S h im a n u k i ( 1 9 6 0 ) , a
nu m e ric a l s o lu tio n o f Eq. ( 1 0 . 1 1 3 ) w as o b t a in e d un der th e a s su m p tio n th a t x = =
c o n s t, an d th a t a c co r d in g t o th e R e y n o ld s a n a lo g y , th e c o e f f ic ie n t K n e v e r y w h e r e eq u a ls
th e tu rb u le n t v is c o s ity , th a t is, is d e fin e d b y th e e q u a t i o n w h i c h fo r a
lo g a r ith m ic w in d p r o file lead s to th e result A'zz = x w « Z . H o w ev er, h ere in ste a d o f the

lo g a r ith m ic fo r m u la fo r w ( Z ) , th e g eneral sim ilarity e q u a t io n ( 7 . 2 4 ) w as u sed for th e w in d
p r o file w ith a universal f u n c tio n (p(C) = U ' ( ^ ) w h ic h satisfies th e fou rth -d e g r e e e q u a tio n
( 7 . 6 1 ) . T h e p ara m eter a o f E q. ( 7 . 6 1 ) w as e x c lu d e d fr o m th e p rob lem w ith th e h e lp o f a
sim p le tr a n s fo r m a tio n to a n e w le n g th scale = L / o . C o n s e q u e n tly , the p r o b le m o f
a d m ix tu r e d if fu s io n fr o m a linear surface so u r ce at a r ou g h n ess h e ig h t Z Z q , for the
c o n d it io n o f r e f le c tio n at th e earth su rface, w a s r ed u c ed to th e s o lu tio n o f o n e universal
e q u a t io n for a f u n c tio n o f t w o variables d e p e n d in g o n th e single pa ram eter Cq — ,
w h ic h arises in c o n n e c t io n w ith th e d e p e n d e n c y o f w ( Z ) o n Z q . It is clear th a t a po sitiv e

value of ^0 c o rr e sp o n d s to sta ble s tr a tific a tio n , and negative values to un sta b le
s tr a tific a tio n . Y a m a m o t o and S h im a n u k i o b t a in e d n u m erica l s o lu tio n s o f th e c o r r e sp o n d in g
e q u a t io n fo r a series o f values o f Co, and p r e se n ted th e m in th e fo rm o f several graphs. T h e
r e p r o d u c tio n o f o n e o f th e graphs in Fig. 8 2 s h o w s th e d e p e n d e n c y o f th e in d ex /;, in th e
law for th e de c re a se in th e surface c o n c e n tr a tio n w h e n Z Z . j , w ith increase in X, o n the
sta b ility p a ra m eter and th e d im c n sio n le ss d ista n ce fro m th e s ou rce A / Z o - T h e param eter
p h e r e w a s d e te r m in e d fro m th e e q u a t io n
w h ere b o t h va lues o f X 2 and A'l w e re ta k e n c lo se to th e fix e d c o o r d in a te X. We can see

th a t fo r n eutra l s tr a tific a tio n , or for a va lu e o f ^0 c lo se to 0 , th e param eter p up to very
large v alu es o f Xj Zo rem ains a p p r o x im a te ly c o n s ta n t and c lo se to 0 .9 . T his is in c o m p le te
a g r ee m e n t w ith th e em p irica l fa c ts d isc u sse d a fter Eq. (1 0 .9 1 ). H ow ev er, n o n n e u tr a l
s tr a tific a tio n , as a rule, d e p e n d s e ss e n tia lly o n th e d is ta n ce X.
T y ld e s le y , in so lv in g th e tw o -d im e n s io n a l d iffu s io n e q u a t io n n u m e ric a lly [se e th e
d isc u ssio n o f P a sq u ill’s paper ( 1 9 6 6 ) ] u sed several fo r m s o f the universal w in d p rofile,
n a m e ly , lo g a r ith m ic plus h n ear p r o file s w ith various values o f j3 in stable c o n d it io n s , pro files
s a tisfy in g E q. ( 7 . 6 1 ) w ith va rio us values o f a in u n sta b le c o n d it io n s , and S w in b a n k ’s
e x p o n e n t ia l w in d pr o file. The results o b ta in e d w ere com pared w ith field data on
a tm o sp h e r ic d if fu s io n an d s h o w s o m e u n e x p la in e d d iscrep an cies.
"I—-r;r
70^ 2 ^
FIG. 8 2 . T h e in d e x p in th e la w fo r th e d ec re a se in a d m ix tu r e su rface c o n
c e n tr a tio n as a f u n c t io n o f d is ta n c e fr o m th e s o u r ce a nd o f t h e s ta b ility
p aram eter.
If, w h e n u sing th e p o w e r -la w e q u a t io n s ( 1 0 . 1 1 6 ) , w e a ssu m e th a t /C z z iZ ^ m a y d iffer
fr o m th e tu rb u le n t v is c o s ity a t m o s t b y a c o n s ta n t f a c t o r a , th at is, th a t — c o n s t,
th en o n e m u s t tak e n = \ — m , or u se S c h m id t ’s c o n ju g a te p o w e r la w . In th e sp ecia l case

o f n eutra l s tr a tific a tio n t h e lo g a r ith m ic p r o file o f th e w in d v e lo c ity is a p p r o x im a te d
c o m p a r a tiv e ly w e ll ov er a sig n ifica n t h e ig h t range b y a p o w e r fu n c tio n w ith in d e x m = 1/7
[se e a b o v e , S e c t. 5 . 6 ] ; th e r e fo r e , it is e x p e d ie n t h ere to assu m e th a t m = 1/7, n = 6 /7 .
A c tu a lly , E qs. ( 1 0 . 1 1 7 ) - ( 1 0 . 1 1 8 ) for th e s e m a n d n, and th e c o rr e sp o n d in g s e le c tio n o f th e
c o e f fic ie n ts Ui and K i, c o r r e sp o n d q u ite w e ll to m a n y e x istin g d a ta o n d if fu s io n un der
neutral s tr a tific a tio n [s e e , for e x a m p le , Calder ( 1 9 4 9 ) ] . It is o b v io u s fr o m E qs. ( 1 0 .1 1 7 )
and (1 0 . 1 1 8 ), in particular, th at for such m and n th e surface c o n c e n tr a tio n d ecrea ses
a s y m p to t ic a lly in p r o p o r tio n t o X " ® ® , w h ic h agrees w e ll w ith e x p e r im e n t and w ith th e
data fr o m Fig. 82 fo r very sm all So. With n o n n e u tr a l s tr a tific a tio n , m a n y inve stig a to r s have
also a tt e m p te d to ap p ly E qs. ( 1 0 . 1 1 7 ) - ( 1 0 . 1 1 8 ) in w h ic h th e in d ic es m a n d n are ta k en to
depend on th e m e te o r o lo g ic a l c o n d it io n s [se e, for e x a m p le . D e a c o n ( 1 9 49 ), V au gh an
( 19 61 ), a n d G ee ( 1 9 6 6 ) ] , b u t he r e th e resu lts o b t a in e d d e p e n d q u ite stron g ly o n th e
m e t h o d o f sele c tin g th e values o f m an d n, and th e r e fo r e are less c o n v in c in g .
P r o c e ed in g n o w to d if fu s io n fr o m a s te a d y p o in t so u r ce o f p r o d u c tiv ity Q at th e p o in t
(0 , 0 , / / ) , w e again n e g le c t d if fu s io n in th e d ir e c tio n OX. In c o m p a r is o n w ith th e c o n v e c t io n
o f th e a d m ix tu r e b y th e m e a n flo w , th e sem iem p ir ic a l e q u a t io n for th e m e a n c o n c e n tr a tio n ,
the so -called th e r e -d im en sio n a l d if fu s io n e q u a t io n s , w ill have th e fo r m
(1 0 .1 1 9 )
dZ
T h e s o lu tio n o f Eq. ( 1 0 . 1 1 9 ) in w h ic h w e are in te r e ste d m u s t sa tisfy th e c o n d it io n s
•0 fo r X^+y^+Z^->oo, Kzz-^ = 0.
z=o
(1 0 .1 2 0 )
8 (0 , V, Z ) :
Q 5 (K )8 (Z — //)
u{H)
[c f. Eq. ( 1 0 . 1 1 4 ) ] . Increasing th e n u m b e r o f in d e p e n d e n t variables fr o m t w o to th ree in th e

tran sition fr o m Eq. ( 1 0 . 1 3 ) t o ( 1 0 . 1 1 9 ) s ig n ifica n tly c o m p lic a t e s th e p r o b le m . D avies
( 1 9 5 0 ) a tt e m p te d t o fin d an e x a c t s o lu tio n o f th e p r o b le m ( 1 0 . 1 1 9 ) - ( 1 0 . 1 2 0 ) fo r th e case
o f a surface s o u r ce , u n d e r th e a s s u m p tio n th a t u { Z ) ^ Z ^ , K z z ( ^ )
Kyy (Z) ^ h o w e v e r , h e s u c c e e d e d o n ly in th e sp ecial case o f a = m, th a t is, w h e n
^yy (2 ) = c u ( Z ) w h e r e c = c o n s t. In th e la tter case it is e asy to p r o ve th a t
y, Z) =— y L = exp / - — 'I • ¥(X,

2/w X \ 4cxl
Z).
w h ere ^ { X , Z ) is th e s o lu tio n o f p r o b le m ( 1 0 . 1 1 3 ) - ( 1 0 . 1 1 4 ) . T h u s, w h e n a = it is n o t
d if f ic u lt _ t o c o n s id e r th e d if fu s io n a lo n g th e O Y a xis; th e r e fo r e , th e a s s u m p tio n th a t
Ky y ^ u, w a s m a d e in a w h o le series o f w o r k s o n tu r b u le n t d if fu s io n . T h e s o lu tio n o f th e
c o rr e sp o n d in g p r o b le m fo r an e le v a te d p o in t s o u r ce u n d e r th e s e a s su m p tio n s is m o r e
c o m p le x b u t it also can b e f o u n d in e x £ lic it fo r m [se e Walters ( 1 9 6 5 ) ] . U n fo r tu n a te ly , th e
e q u a tio n s o b t a in e d fo r c o n c e n tr a tio n Y, Z) in m a n y ca ses clea rly c o n tr a d ic t th e
data o n a d m ix tu r e p r o p a g a tio n fr o m ste a d y p o in t so u r ce s in th e a tm o sp h er ic surface layer.
In th is c o n n e c t io n , S m ith ( 1 9 5 7 ) r ein v e stig a te d th e c a ^ o f th e genera l in d e x a , b u t w as
u n a b le t o o b ta in an e x a c t d e s c r ip tio n o f th e fu n c t io n K, Z ) ; in ste a d , h e o b ta in e d
th e t w o first n o n z e r o m o m e n t s o f th e a d m ix tu r e d is tr ib u tio n a lo n g th e straigjit lin e X =
c o n s t, Z = c o n s t:
oo oo
00{X, Z ) = j ? ( X . y, Z)dv, (X, Z ) = f y n ( x , y, z)dy. ( 10.121)
T h e fu n c t io n 0 o ( X , Z ) w ill o b v io u s ly sa tisfy th e sa m e e q u a t io n ( 1 0 . 1 1 3 ) [w it h th e
c o n d it io n s ( 1 0 . 1 1 4 ) ] w h ic h d e scr ib e s th e c o n c e n tr a tio n d istr ib u tio n fo r a s te a d y linear
s ou rce, th a t is, it m a y b e d e te r m in e d fr o m E q. ( 1 0 . 1 1 8 ) w ith n = 1 — m . F o r th e f u n c tio n
02 (X , Z ) , a d iffe r e n tia l s e c o n d -o r d e r e q u a t io n , w h ic h also c o n ta in s th e c o e f f ic ie n t K y y { Z ) ,
is o b t a in e d fo r it fr o m Eq. ( 1 0 . 1 1 9 ) . T his e q u a t io n m a y b e s o lv ed e x p lic itly for a n y / / f o r
s o m e particular v a lu es o f m a n d ct; h o w e v e r , for arbitrary m and a its e x p lic it s o lu tio n ,
e x p r e s s e d in te r m s o f c o m p le x tr a n sc e n d e n ta l fu n c tio n s , has b e e n f o u n d o n ly fo r th e case o f
a surface s o u r ce , th a t is, w h e n / / = 0 . It is natural to s u p p o s e , h o w e v e r , th at fo r n o t very
large H an d large X , th e s o lu tio n w h e n H > 0 w ill d iffe r little fr o m th e s o lu tio n fo r a
surface so u r ce . In _a d d itio n , i f o n ly t h e su rface d istr ib u tio n o f th e a d m ix tu r e is o f in ter e st,
th a t is, values o f i9’(X , K, 0 ) , th e n th e s o lu t io n for an e le v a te d s o u r ce m a y b e derived fr o m
th e general s o lu tio n o f a su rfa ce s o u r c e w ith th e h e lp o f an e le g a n t du a lity th e o r e m p r o ve d
by S m ith . T his th e o r e m e x p r esses th e sta tistic a l r eversibility o f th e d if fu s io n p r o c ess
d e scr ib e d b y th e s e m iem p ir ic a l e q u a t io n ( 1 0 . 5 5 ) . A c c o r d in g t o th e t h e o r y , f o r tu rb u le n t
d iffu sio n m th the reflection co n d ition on the surface Z = 0 , the value o f -^(X, Y, 0 ) o f the
fu n c tio n 0 ( X , K, Z ) which corresponds to an elevated adm ixture source a t j h e p o in t
X = Y = Oy Z = H coincides w ith the value o f ' t { X , K, M ) o f the fu n c tio n i>{X, Y , Z )
which corresponds to a surface source a t the p o in t X = K = Z = 0 (so th a t fo r (32(^, Z )
it is p o s sib le t o w r ite an e x p lic it fo r m u la fo r a n y H ). K n o w in g 0 o ( X , Z ) and B2 (X , Z ) , it
is p o ssib le t o d e te r m in e th e f u n c t io n i^ (X , Y, Z) a p p r o x im a te ly b y assu m in g , a c co r d in g t o
in tu itiv e p h y sic a l a s s u m p tio n s o n transverse d iffu s io n and e x ist in g d ata, th a t th e a d m ix tu r e
d istr ib u tio n a lo n g th e straight lin e X = c o n s t, Z = c o n s t, for all X and Y, d iffe rs very
little fr o m a G a ussian d is tr ib u tio n . C o n s e q u e n tly , fo r a n y X, Y and Z
8 { X, V, Z ) X e x p [— y y ^ Y H X Z ) ] ^ ^ ^ (1 0 .1 2 2 )
' ' [2 tcK 2 (A ', Z ) ] ‘/« ■
w h ere
02 (X Z)
U sing th is fo r m u la , S m ith , in particular, fo u n d th e rate o f d ec re a se in th e surface

c o n c e n tr a tio n in th e O X d ir e c tio n o n th e a x is o f th e a d m ix tu r e c lo u d , th a t is, w h e n K = 0 ,
for m = 1 /7 an d « = a = 6 / 7 . In th is c a se , A '“ a s y m p to t ic a lly w h ic h m o r e or less
agrees w ith th e da ta fo r ste a d y p o in t a d m ix tu r e so u r ce s in th e n early neu tra l a tm o sp h e r e .
In th e g eneral ca se o f arbitrary str a tific a tio n , Y a m a m o t o a n d S h im a n u k i ( 1 9 6 4 )
a tt e m p te d t o ev a lu a te th e cro ss-w in d (lateral) d iffu s iv ity b y c o m p a r in g th e d if fu s io n da ta
fr o m p o in t so u r ce s w ith th e resu lts o f n u m e ric a l s o lu tio n o f th e p r o b le m ( 1 0 . 1 1 9 ) - ( 1 0 . 1 2 0 )
fo r va lues o f K y y ( Z ) c o n ta in in g an adju sta b le p a ram eter. H ere, as also in th eir p r e ce d in g
w o r k , it w a s a ssu m ed th a t th e w in d is given b y E q. ( 7 . 2 4 ) , w h e r e (p{£) = C / ' ( g ) satisfies E q.
(7 .6 1 ), and th a t Kzz = M # xZ /(p(g). A s a first a p p r o x im a tio n , it w as a ssu m ed th a t th e
c o e f f ic ie n t K y y { Z ) d e p e n d s lin early o n h e ig h t Z fo r a n y s tr a tific a tio n , th a t is, it is given b y
a fo r m u la o f th e t y p e
w h e r e a (to ) is an u n k n o w n f u n c t io n o f So, or, o f str a tific a tio n . T h e m a in a d van tage o f th is

s e le c tio n of Kyy is th a t it is e a sy to e lim in a te th e pa ram eter a w ith th e h e lp o f a
tr a n s fo r m a tio n t o d im e n sio n le ss c o o r d in a te s
X i = X/Zo, r , = K/a'/^Zo. Zi = Z/Zo,
w h ic h d e p e n d o n a . W ith th is tr a n s fo r m a tio n th e p r o b le m o f d if fu s io n fr o m ^ s t e a d y surface

p o in t s o u r ce r ed u c es t o th e s o lu tio n o f a universal e q u a t io n fo r a f u n c t io n ^ ( X i , Kj, Z ,) o f
th ree variables w h ic h depends on a single param eter £o. T his e q u a t io n w a s solv ed
n u m e ric a lly for a series o f v alues o f £o, an d th e a d m ix tu r e d is tr ib u tio n alo n g th e OYi a x is,
c a lc u la te d for X i = 3 2 ,0 0 0 and Z i = 2 4 0 , w a s c o m p a r e d w ith c o r r e sp o n d in g d a ta fr o m
field d if fu s io n e x p e r im e n ts . The results of th is c o m p a r is o n p e r m itt e d a p p r o x im a te
d e te r m in a tio n o f th e u n k n o w n va lues o f a(Jô); th e values o f Zq and L * w e re e stim a te d
fr om th e data of s im u lta n e o u s w in d and te m p er a tu re p rofile m e a s u r e m e n ts. W ith
c o m p a r a tiv e ly small scatter, th e v alu es of a (So) o b t a in e d lay on a sm ooth curve
m o n o t o n ic a lly dec re a sin g w ith an increase in So; a c c o r d in g to th is curv e, a ~ 13 w h e n So =
0 , th a t is, fo r n eutral s tr a tific a tio n , 1 0 0 w h e n S o = - 0 . 0 1 4 , and a ~ 3 w h e n So =
0 . 0 1 . F u rth e r m o r e , t o c h e c k th e v a lid ity o f th e a s s u m p tio n m a d e c o n c e r n in g Ky y ( Z ) y th e
v alues o f a (So) fo u n d w ere u se d fo r th e o r e tic a l c a lc u la tio n o f th e c o n c e n tr a tio n
d istr ib u tio n a lo n g th e OY ax is fo r s o m e o th er values o f X a nd Z , th e r esults o f w h ic h again
w ere c o m p a r e d w ith th e o b s e rv a tio n s. A lth o u g h th er e w as in genera l s a tis fa c to r y a g r ee m e n t
b e t w e e n th e th e o r y and e x p e r im e n ts , th e scatter o f th e e x p e r im e n ta l p o in ts w a s so large
th a t it w as im p o ss ib le to draw a reliable c o n c lu s io n . For th e d ecrease in surface
c o n c e n tr a tio n a lo n g th e a xis o f th e a d m ix tu r e c lo u d , w itl^ neutral s tr a tific a tio n s, th e
c a lc u la tio n b a se d o n K y y = 13 yui Z le d t o th e e x p r e s s io n w h ic h agrees w e ll
w ith th e e x ist in g data.
T h e data also s h o w th a t th e h o r iz o n ta l d if fu s io n a lo n g th e OK d ir e c tio n b e c o m e s faster
w ith an in crease in th e h o r iz o n ta l e x te n s io n o f th e c lo u d , so th a t th e c o e f f ic ie n t Ky y
increases w ith an in crea se in th e d if fu s io n tim e x ^ t — to or th e c o o r d in a te X. T his is
e x p la in e d b y th e very im p o r t a n t e f f e c t o f acceleration o f the relative d iffu sio n o f th e
a d m ix tu r e c lo u d or a c ce le r a tio n of th e ch a n g e in d is ta n c e b etw een th e ind iv id u al
c o m p o n e n t particles o f th e c lo u d , w it h an inc r e a se in th e size o f th e c lo u d . T h e e f f e c t is
related to th e fa c t th a t, as th e size o f th e c lo u d in crea ses, m o r e and m o r e large-scale v e lo c ity

d istu rb a n ces b e g a n t o p a r tic ip ate in th e d isp e r sio n ; w e shall d iscu ss th is in C hapt. 8, V o lu m e
2 o f this b o o k . H ow e v er , as n o t e d a b o v e , w ith in th e fr a m e w o r k o f th e sem iem p ir ic a l th e o r y
o f tu rb u le n t d if fu s io n , it is im p o ss ib le t o a ssu m e th a t th e d iffu s iv ities d e p e n d o n th e tim e x .
N ev er th ele ss, D avies ( 1 9 5 4 a ) tried t o c a lc u la te th is e f f e c t , f o r m ^ y a ssu m in g th a t in Eq.
( 1 0 . 1 1 9 ) th e c o e f f ic ie n t s u and Kzi d e p e n d o n ly o n Z , sp e cific a lly but
Ky y ^ Z®' I K |i‘. (T h a t is, Kyy a c tu a lly d e p e n d s o n th e p o s itio n o f th e a d m ix tu r e so u r ce sin ce
this is th e o n ly m e a n s b y w h ic h th e Y origin is id e n t ifie d .) In th e sp ecia l case a = m,
P = 1 — 2 m , D avies fo u n d an e x a c t s o lu tio n o f E q. ( 1 0 . 1 1 9 ) ; h o w e v e r , it is clear th a t his
basic a s su m p tio n c o n c e r n in g th e d e p e n d e n c e o f Kyy o n Y sharply c o n tr a d ic ts th e in t e n t o f
th e sem iem p ir ic a l th e o r y . F r o m th is v ie w p o in t , th e p r o p o s itio n o f L a y k h tm a n ( 1 9 6 3 ) is
m o r e relevant. A c c o r d in g to L a y k h tm a n , th e c o n c e n tr a tio n Y, Z ) in th e case o f a
p o in t s o u r ce m u st b e f o u n d in th e fo r m ( 1 0 . 1 2 2 ) w h e r e 0 o ( ^ , 2 ) is g iv e n as th e s o lu tio n o f
Eq. ( 1 0 . 1 1 3 ) ; to d e te r m in e Y^(X, Z ) , it is r e c o m m e n d e d th a t th e Lagrangian e x p r e s s io n s
be ta k e n as th e basis [e .g ., th e T a y lo r fo r m u la ( 9 . 3 1 ) ] , rather th an th e sem iem p ir ic a l th e o r y .
Consideration o f the Possible Deviation o f the Principal Axes

o f the Eddy Diffusivity Tensor from the OX, O Y
and OZ Axes
The preceding conclusions were all based on the assumption that
the semiempirical equation of atmospheric diffusion has the form
(10.55). Let us now recall that even the most general semiempirical
diffusion equation may only be obtained by using certain nonexact
approximations; however, Eq. (10.55) contains also the additional
assumption that the OZ, OX and OY axes directed vertically, along
the mean wind and perpendicular to the wind, are the principal axes
of the eddy diffusivity tensorK<j. We have introduced this assump
tion since these directions are the preferred ones from the viewpoint
of the geometric and kinematic conditions of the flow; however, it
must be kept in mind that this argument is not rigorous. Therefore,
we must not exclude the possibility that in fact the semiempirical
equation of atmospheric diffusion should contain also some ad
ditional terms which are omitted in Eq. (10.55). Moreover, in
Sect. 7.5 we have already noted that in the atmosphere above
a homogeneous underlying surface, a horizontal turbulent heat
flux must exist in th e mean wind direction which is described by the
joint moment u ' T '= u ^ T ^ j ( Z l L ) . This moment is positive for
positive temperature gradients and negative for negative gradients,
and rather great; at nearly neutral stratification it is about three
times as great in absolute value as the moment w'T' and in stable
conditions the ratio is even greater [see Sect. 8.5, especially Fig. C].
Within the framework of the semiempirical diffusion theory, the
moment u'T' in plane-parallel turbulent flow must be represented in
the form w T '= Kxz (where Kxt < 0), while w'T' = - K z ^
The presence of a nonzero coefficient Kxz dearly indicates that the
coordinate axes do not coincide with the principal axes of the tensor
Ka; moreover, we see that in a nonstratified boundary layer, the
nondiagonal component Kxz of the tensor/C<i is about three times as
great as the diagonal Kzi. Under the usual assumption that the eddy
diffusivities for heat and for neutral mass admixture are equal, we
write the semiempirical equation for the turbulent admixture flux in
the mean wind direction as
(10.123)
where Kxz is a negative coefficeint and in neutral conditions is

approximately three times as great in absolute value as K^z. (The
third possible component of u ^ \ namely, —Kxy , is equal to zero
since K^y =K^y =Kyx =Ky^ = 0 in a two-dimensional turbulent flow
due to the symmetry of the flow relative to the mean velocity
direction.) O f course the large value of compared to Kzr does not
mean that the corresponding term — A - KxziZ) K (Z) — of the semi-
dX ’ dZ
empirical diffusion equation is always significant. In fact, this term is
due to the turbulent admixture flux in the mean wind direction
which is often small in comparison to the convection flux 77^;
consequently, it may often be taken as small compared withw .
However, if -----is taken into account, then there is no reason to
omit the term K ------ in the diffusion equation. Moreover, if

dXdZ
K x z ^ O , then also no reason exists for supposing that Kzx = 0 .
Qualitative physical reasoning on the mechanism of admixture
transfer in a turbulent flow with d = d (JQ, that is, independent of the
coordinates Y and Z, at the initial time t = makes it natural to
expect that K^x will be negative in a two-dimensional flow; the
approximate equality of the correlation coefficients ruT and r^T in
absolute values in a nearly neutral atmospheric surface layer also is
partly responsible for the supposition that Kzx * —t^xx!^ in such a
layer [see Yaglom (1969)]. In every case the general form of the
semiempirical diffusion equation in a plane-parallel turbulent flow
must, strictly speaking, differ from Eq. (10.55) by two additional
terms, and must be written as
5(Z )-^ = (Z) — + (Z )-^+ K (Z) — +

dt dx ax2 dxdz ay2
. A (k ^ A Z ) + a /k „ (Z ) — \ . (10.55')
dZ V dx) dz\^^ dz/
Here the coefficients Kxx, Kyy, and K^z are taken as positive, and K^z
and Kzx, as negative. The term containing K^z is usually o f the same
importance as that containing and often they can both be
omitted without considerable error; similarly in many cases the term
containing Kzx must be comparable with the usual vertical diffusion
term containing Kzz • However, all the qualitative predictions of the
semiempirical diffusion theory can hardly depend on the presence or
absence of the terms with coefficients Kxz and K.zx\ therefore all the
general deductions obtained above using Eq. (10.55) are apparently
correct. The last conclusion is confirmed also by the results of the
few existing calculations based on the diffusion equation with
nonzero coefficients Kxz and Kzx\we shall now consider these results.
Lettau (1952) [see also Priestley (1963)] was apparently the first
to draw attention to the possibility that Kxz # 0 in the semiempirical
equation of atmospheric diffusion. Later, Davies (1954b) tried to
estimate the possible influence of the corresponding term o f the
diffusion equation on the diffusion from a steady point source. For
his estimate he compared the solution of the equation
dX dX\ dZj dZ\ dZ)
for some special assumptions on the functional form of the

coefficients u(Z), Kxi{Z)and Ktt(Z), based partly on semiempirical
reasoning, with the solution of the ordinary two-dimensional
diffusion equation (10.113); as expected, the difference in solutions
turned out to be quite small. Later, Gee and Davies (1963) evaluated
the influence of the t e r m o n the values < X > and
calculated by Saffman [see Eq. (10.103)], and found that for certain
assumptions concerning Kxt (2 ) the corrections obtained may be of

the order of 15—20%. At almost the same time, Matsuoka (1961;
1962) reported that apparently 0 also, or that the diffusion
equation must have the form (10.55'); he presented some nonrigor
ous semiempirical arguments in favor of the equality K^x = Kxz and
supposed that K^x =Kxz = -(KxxKzzy/t. Matsuoka used his diffusion
equation to compute the admixture distribution from a linear steady
source. He found the corresponding values of the surface concentra
tion, but not of the concentration at other heights, under his
assumptions to agree exactly with the values given by the solution of
the corresponding “ classical” diffusion equation (10.55). Gee and
Davies (1964) also used Eq. (10.55') with^;,cz =A^zjcand some special
assumptions concerning all the diffusivities to compute horizontal
dispersion from an instantaneous source; their results show that the
influence of the two additional terms in Eq. (10.55') may be quite
significant on the diffusion from an instantaneous source. Also in
Gee and Davies, a semiempirical argument was presented showing the
equation Kzx = Kxz/2 to be more reasonable than Matsuoka’s equa
tion Kzx = Kxz- The relation Kzx —Kxzl'^> together with some others of
the same type, was used by Gee (1967) to compute diffusion from
an instantaneous point source in a flow with u(Z) = Uq +aZ,
v{Z) = Vo +bZ; in this case the tensor Kjf will have in general four
nonzero off-diagonal components. The results of Gee’s computations
with Kjf = const for all i and / show that the additional terms have
httle effect on diffusion. This contradicts the results of Gee and
Davies using other vertical profiles of the diffusivities, and illustrates
the sensitivity of the above-mentioned effect to the forms chosen for
these profiles. Since at present these profiles are still relatively
unknown, it is pointless to go into further detail here.
The Use o f the Lagmngian Similarity Hypothesis in

Surface Layer Diffusion Theory
In the preceding, we always began with the semiempirical equation
of turbulent diffusion. However, it is of interest to consider what
conclusions may be drawn concerning diffusion in the atmospheric
surface layer without using the semiempirical hypothesis of linear
dependency of the turbulent admixture flux on the average
concentration gradient. Thus, of significant help here may be the
proposition, which appears quite natural, on the similarity of the
Lagrangian turbulence characteristics in the surface layer which was
introduced in Sect. 9.4. We shall now proceed to some corollaries of

this proposition.
To begin with, let us recall that neglecting the effect of molecular
diffusion, the mean admixture concentration from an instantaneous
point source of unit output, at the point x = (x, y, z)ai the time h,
is equal to the probability density function of the Lagrangian
coordinates (X, Y, Z) of a fluid particle which at the time io is at the
point X (see above, Sect. 10.3). It follows from this, in particular,
that for an instantaneous surface source of output Q at the
coordinate origin, the mean concentration at the time t = /q + t for
neutral thermal stratification and sufficiently large t , will be
determined by a formula o f the type
1 { X , V, Z, + = (10.124)
where P 3 is a universal function o f three variables
and b, c and e = 2.718.. . are universal constants [see Eqs. (9.65),

(9.60'), and (9.61')]. It is possible to use the same expression for a
yy
source at height H only if — . For not very large values of t but
which nevertheless appreciably exceed — , somewhat more accurate

results are obtained if we assume that the statistical properties of the
Lagrangian velocity of a fluid particle at time t after its release at a
given point at height //, are the same as those of a particle released at
the ground (// = 0 ) at the instant to -X i where t i is of the order
specifically, xi =^H/u^, where ]3 is a universal constant; in other
words, if we assume that Eqs. (9.60') and (9.61') for Z(x) and ^(t)
are valid if the time t is replaced by t + Ti = t + [ Batchelor
(1964)] This correction is significant only in a comparatively small
time interval. However, when processing data from real experiments,
especially laboratory experiments, in which it is frequently necessary
În this s u b s e c tio n w e h av e already seen th at th e a p p r o x im a te sem iem p ir ic a l d iffu s io n

th eo r y lea d s t o th e e stim a te /3= l / x = l / a x ^ 2 . 2 , an d also to th e va lu es !? = */.’ ^ 0 . 4 5 ,
c = e~^b ^ 0 . 5 6 b . A s o m e w h a t m o r e a c cu ra te e s t im a te o f b w ill b e given at th e en d o f th e
n e x t s u b s e c tio n .
to limit oneself to not very large values of t , it sometimes turns out

to be useful [see, for example, Cermak (1963)]. In the case of
unstable or stable stratification, Eq. (10.124) is replaced by the more
general expressions
r, Z , + = (10.125)
where P is a new universal function of four variables Z = Z(x) and

X - X { x ) are determined by Eqs. (9.69) and (9.70), and L is a natural
length scale in the stratified boundary layer introduced in Chapt. 4.
For T which are not very large it is also possible here to introduce a
correction for a source height H, but a discussion of this will not be
presented here.
It follows from expression (10.124) in particular that the
variances of the admixture cloud from an instantaneous point source,
determined by the expressions
= Q -'J J f ( X — X f '^(X, V, Z ) d X d Y d Z ,
0* = Q " 'J f f r n ( x , y, Z ) d X d Y d z ,
= Q - ' J f f ( Z ~ Z ) n ( X , V, Z ) d X d V d Z ,
for neutral stratification, are asymptotically proportional to Z l

According to Eq. (9.61'), this indicates that the variances along all
the three axes of the admixture cloud from the instantaneous point
source with neutral stratification must be asymptotically pro
portional to that is, to and not to x as in the case of
homogeneous turbulence, and not to as in the case of homogene
ous shear flow. In the presence of thermal stratification, strictly
speaking, it is impossible to ensure that these variances (even for very
large x) will be strictly proportional to Z^, since here the form of the
distribution d{X, Y,Z, to + f) may depend on Z/L, that is, may vary
with variation in x. It is natural to suppose, however, that the
dependence of the distribution on Z/L in many cases will not be very
strong; thus in the first approximation here also the variances may be
considered proportional to Z l In the case of strong instability the
latter conclusion seems rather convincing and has additional support
[see Eq. (9.75')]; therefore, under these conditions, for values of x
which are not very small, we must have the expressions
,? 2 72 ^ 3
(The most exact of the last-mentioned expressions probably will be

that for .) In principle, all these conclusions may be compared
with data from experiments with instantaneous point admixture
sources. However, since there is still such little data, for experimental
testing it is more convenient to use the results from Lagrangian
similarity hypotheses concerning the concentration distribution from
steady admixture sources. These results will be discussed below.
According to Eq. (10.125), the surface admixture concentration
from a steady point source, at a sufficiently large distance X from
this source in the direction of the mean wind, will be determined by
the equation
00
^{X, 0, 0) = Q f 0. - 1 ; f ) | ^ - (10.126)
For a linear source this equation must be replaced by the following:
MX.O) = q [ (10.127)
6-00 '
The last two equations may obviously also be rewritten as
z
b{X, 0, 0) = Q
dz
r 0. - 1; A )
V ^ x ~ x \ (10.126')
= Q X —X d Z dX I [ z I
Z dz dz
and analogously
pi^ - ^ z - 1-
&(;r,o) = Q
X — X dZ . d X -\ \ z I z (10.127)
dz + di
For approximate calculation of the integrals on the right sides of

Eqs. (10.126') and (10.127'), following Batchelor (1959; 1964) and
EUison (1959), let us use the fact that for large X, the admixture
cloud from the instantaneous point source at the coordinate origin
must flow past the plane X = const in a time much less than the time
T required for the cloud to reach this plane. This may be expressed as
follows: since obviously ZjX -> 0 w h e n t ^ oo, the range o f values X
for which
r,(X — X Y . Z\
" ! • - '• T j
remains markedly different from zero, will make up a relatively small

fraction of X itself for large X. Therefore, for sufficiently large X it is
permissible to consider that
-u | ) ~ » ( ^ )
for any fixed -y . This assumption is quite similar to neglecting

longitudinal diffusion in the mean wind direction, compared with the
admixture convection by the mean flow, which is done in almost all
applications of the semiempirical diffusion equation. It follows from
this that Eqs. (10.126') and (10.127') may be replaced without any
great error by the approximate expressions
oo _
0 , 0) ^ - ^ f p ( i, 0 , - 1 : (10.126")
~dT
and
dz — oo —oo
where Z = Z(x) and X — X{-z) now are taken for a value of t such
that X (x) = X. If we replace the height zero on the left sides of Eqs.
(10.126") and (10.127") by an arbitrary height Z, then the
argument —1 of the functions P must only be replaced by Z/Z —1.
First, let us investigate the simplest case of diffusion under neutral
stratification. In this case the function P in Eqs. (10.126") and
(10.127") does not depend onZ/Z., that is, for all Z it is the same,
andX(T)and Z ( t ) satisfy Eqs. (9.6 0')-(9.61 '). Therefore,
(10.128)
and
b { X ,0) ^ (10.129)
Thus it is seen that in neutral conditions, the admixture concentra

tion from a steady linear source must decrease approximately
proportional to X~\ and from a steady point source, somewhat more
slowly than X-^; as we know, the data agree quite well with this
conclusion [see above, following Eq. (10.91)].
The index p in the expression &— X~'^ma.y be determined more
accurately from the formula p — — this was done by Cermak
(1963), keeping in mind also the above-mentioned correction toX(x)
and Z (t) related to the nonzero source height H. With this the
theoretical values o f p= p{X) calculated from Eqs. (10.128) and
(10.129) under the assumption that 6 = c = 0.1, were in even better
agreement with the existing data of both laboratory experiments and
field observations for neutral stratification than would have been
expected. Later, however, Pasquill (1966) advocated the value b - k
*=» 0.4 which, as we know, follows from the semiempirical diffusion
equation, under the assumption that a = 1. With this in mind, he
treated the data from field observations under neutral conditions of
the vertical admixture distribution at a given distance X from a
steady source. It was necessary to accept several nonrigorous
assumptions in order to derive the relationship between the observed
characteristics of the vertical distribution and the Lagrangian

statistical characteristics X and Z ; however, under these assumptions,
Pasquill found that the field data agree with the estimate b 0.4
with greater accuracy than that of the assumptions used in the
theoretical derivation o f this estimate.
The case for nonneutral stratification is even more compHcated.
First, the integrals on the right sides of Eqs. (10.126") and (10.127")
may now depend on Z/L, that is, on X. However, this dependency
will be weaker than the dependency on X of the factors before the
integrals. Thus, as a first approximation, it is possible also in this case
to use the expressions
0 , 0 )-^------- ^ ------- . scA', 0 ).

rfx
X(T)=X X(t) = X
It is more important that the functions X(x) and Z(t) here be defined
by Eqs. (9.69) and (9.70) which contain the unknown universal
functions f(^), (pi(^) and only their asymptotic behavior may be
considered known with accuracy up to indeterminate numerical
coefficients. Thus, for strong instability Z (x) ~ X (x) ~ x and
consequently
0 (A'. 0 , 0 ) ^ X - \ 0 ,) ^X-^! k
Of course, we may hope (and the data o f Sect. 8 give some

justification for this) that only the asymptotic behavior o f these
functions will play a basic role in the calculation of concentration. If
this is true, then, in first approximation, we can substitute for the
unknown functions almost any functions that behave properly for
small and for large values of the arguments. Such a possibility is
partially confirmed by the calculated results of Gifford (1962). In
accordance with Eq. (9.71), but in disagreement with the deductions
of the semiempirical theory, Gifford proposed that q)i (Q = /(^ ) so
_ _
that « (Z). Then he selected as / (^) a function of the form
lnC + 6C for C > 0,

Iniq for 0 < C < — 0.03,
/(C) = V ^ (10.130)
C 3 + ( 0 . 0 3 ) 3 + I n 0.03 for C < — 0.03
and assumed [according to Kazanskiy and Monin (1957)] that
1
1 /' (0
—
According to the dependency obtained from these assumptions,

where the dimensionless concentration & for a steady point
V
admixture source depends on the dimensionless distance -/ft -j-,

Gifford calculated the index p in the expression p for different
thermal stratification and different distances X. The values of p he
obtained turned out to agree comparatively well with the data
gathered in the United States during Project Prairie Grass at O’Neill,
Nebraska, and pubUshed by Barad (1958) et al. Later, Malhotra and
Cermak (1963) compared the results o f Gifford’s calculations, again
when b = 0 . 1 , with the data from several diffusion experiments
carried out in a special meteorological wind-tunnel with a heated
lower wall, producing an unstable temperature stratification; the
results obtained also were satisfactory. Cermak (1963) made an
additional verification of Gifford’s theory for the case of slightly
nonneutral conditions; he used Swinbank’s exponential function f(^)
[cf. Eq. (7.73)] in his calculations and obtained satisfactory
agreement between theoretical predictions and the field and labora
tory data. Later Pasquill (1966) revised Gifford’s calculations, using
the value b = 0.4, of the dependence of Z on A'. He also analyzed the
estimate of the same dependence by Panofsky and Prasad (1965);
according to his conclusions, the similarity theory cannot explain the
severe effect o f thermal stratification on the vertical spread of
admixture which is apparent in the published data of the O’Neill
experiments. Still later, Klug (1968) repeated and expanded Gif
ford’s calculations, using the same assumptions but with b = 0.4 and
a slightly different form of universal wind profile. He also found that
some features of the real concentration measurements are inconsist
ent with the theoretical predictions obtained. Consequently, no clear
solution has yet been obtained and additional careful investigations,
both experimental and theoretical, are greatly desirable.
10.6 Diffusion with a Finite Velocity
The fact that the semiempirical diffusion equation (10.55) is

parabolic leads to the result that the admixture introduced into the
fluid at the time to spreads immediately throughout the entire space.

Consequently, by the time / o+ t where t is as small as desired, the
admixture, even in a very small amount, can already be detected at
an arbitrarily large distance from the source [see, for example, Eq.
(10.89)]. This clearly contradicts the actual boundedness of the
velocities of the fluid particles. It is directly related to the fact that
in the semiempirical theory, the instantaneous velocity of the fluid
particles, as we have seen in Sect. 10.3, turns out to be infinite. In
practice, however, the appearance of an infinite velocity in the
theory usually does not play a large role, since the volume inside
which the concentration is not negligible, is always bounded, and the
distribution of the concentration inside this volume as a rule is
described satisfactorily by the parabolic diffusion equation for
diffusion times which are not too small. Nevertheless, close to the
real boundaries o f the admixture cloud, the use of the parabolic
diffusion equation may lead to essential errors. For example, smoke
coming from a stack at a height H indeed reaches the surface of the
earth no closer than a distance from the stack equal to — H where u
is the minimum velocity of the wind, and v is the maximum velocity
of the vertical motion of the smoke particles; at the same time,
according to the semiempirical parabolic diffusion equation, smoke
may be detected on the surface of the earth as close to the stack as
desired. Therefore, in some cases it is desirable to have a more
general semiempirical diffusion theory in which the boundedness of
the fluid particle velocities would be taken into consideration. Such a
generalization of the ordinary theory of diffusion, although not
necessarily turbulent, was briefly outlined, with no details, by G. 1.
Taylor (1921) and later developed at different times by many investi
gators; in this subsection, we shall discuss it principally according to
M onin(1955; 1956b).
We shall assume that the random function A" (Jf, /) which describes
the fluid particle motion is differentiable with respect to t, and that
its derivative, the velocity o f the fluid particle V{x, t), is bounded. In
this case, the function X ( x , t) cannot be considered a Markov
function of t. In fact, we have already seen that for a Markov
function A" (jf, /), the velocity will always be infinite provided
Kij + 0. If we assume that Kn = 0, then the velocity will not be
infinite, but, as is easily seen, the function X (.v, t) in this case will
not be random. Thus, the condition of boundedness of the velocity
of a fluid particle does not agree with the assumption that its
coordinate AT(jc, ^ is a Markov random function of /.
A similar situation arises in the theory of molecular diffusion

when attempting to take into account the inertia of a diffusing
particle; due to inertia its trajectory X { x , t) turns out to have a finite
derivative V (jc, t) = — everywhere. Thus, it follows that the
function X (j«r, 0 cannot be a Markov function. It is known that in the
theory of molecular diffusion, that is, Brownian motion, it is possible
to consider inertia of the particles by assuming that, not the function
^), but the six-dimensional function 0 } is
Markovian [see, for example, Uhlenbeck and Omstein (1930) or
Chandrasekhar (1943)]. We shall proceed in a similar manner.
For simplicity, let us limit ourselves to investigating a one
dimensional diffusion problem in the direction of the OZ axis.
According to what has been said above, let us try to construct a
generalization o f the semiempirical theory o f turbulent diffusion on
the basis of a new semiempirical assumption that is broader than the
assumption o f the Markovian nature of the function X (at, t). In
particular, let us assume that the two-dimensional random function
{Z{z, t), W{z, /)} is Markovian where Z{z, t) is the coordinate of a
fluid particle, and W{Z, = is its velocity. Since we are
only interested in the effects of boundedness of the velocity W, we
can limit ourselves to a very rough description of the possible values
of W. Let |U^|<C; let us divide the entire range of possible values of
the velocity — C < W < C into a finite number N of intervals An, and
instead of the specific values of W we shall indicate only a number n
of intervals A„ into which these values fall. In other words, instead of
the random function W {z, t), we shall consider the random function
n{t) which assumes discrete values of 1 , 2 , . . . also, we shall make
the assumption that the random function {Z{z, t), n{t)} is Markov
ian.
Let us now investigate the conditional probabilities
P { Z { z , t ) < Z - , n(t) = i \ Z { z , Q = z} = f p , { Z \ z , t ) d Z . (10.131)
The probability density function for the coordinate of a fluid

particle Z(z, t) which plays a basic role in the theory of turbulent
diffusion is expressed in terms of the functions P i { Z\ z , t) by the
equation
p { Z \ z , t ) = ' ^ P i { Z \ z , t). (10.132)
The basic differential equations of Markov random processes,

indicated by Kolmogorov (1931), reduce here to the following
system of equations for the functions Pi ( Z \ z , t), replacing the
ordinary semiempirical diffusion equation:
dPi _ V _ „ dWiPi
~ d f~ 2j dZ ' (10.133)
Here the following notations are used:
T =0
aji = aj i {Z, t ) - = Y ^ P [ n { t + i) = i \ Z { z , t) = Z, « ( 0 = y)
T=0
where
Yi { x \ Z , t) = Z ( 2 , t + x ) - Z { z , t)
is the particle displacement under the condition that the values of

Z(z, t ) = Z and n { t + x ) = n { t) = i are fixed. Obviously, Wi is the
mathematical expectation of the particle velocity W under the
condition that W belongs to the interval A,; thus, the value of
Wi (Z, t) for all Z and t lies somewhere inside the interval A{. The
meaning of the variables aj, may be clarified in the following way.
Since the function n{t) assumes only integral values, it may vary only
in jumps, making the transition from the value n{t — 0 ) = j at the
time which directly precedes /, to some other value n(/ + 0 ) = / at a
time which immediately follows t. The values of an when i 4- j are
the average frequencies of transitions from n (/ — 0 ) = / to
n(< + 0 ) = /, since ajiM is the probabiUty of a jump from n = j to
n = i during the time interval (t, t + At) under the condition that
n{t) = /.
Since
2 P { « (^ + x) = / |Z ( 2 , t) = Z , n i t ) = j } = \,
it is clear then that = consequently ajj = — 2 is equal

to the average frequency o f transitions from n = j to some other
value of n taken with the minus sign.
Equations (10.133) form a semiempirical system of diffusion
equations corresponding to the propositions formulated above. These
equations must be solved under the initial conditions
Pi(Z\z, t ) ^ e i ( z ) h { Z ~ z ) for
ei (z) = P { n {(q) = i \ Z (z, ô) = z].
(10.134)
The variables &i give the probability distribution forn(/o), that is,
actually, for values of the initial velocity of a fluid particle.
Let us investigate in more detail the simplest scheme in which the
entire region of possible values of the fluid particle velocity
— C < W < C is divided into two intervals At and A2. We select as Ai
and A2 the intervals 0 <H ^<C and — C < W < 0 ; in this case the
random function n(t) indicates the direction of fluid particle motion.
For determinacy, assume that both directions are equivalent, so that,
in particular, W = 0 ; in this case we may assume that W i —— W z = W,
where W is the mean absolute velocity of the particle, and
012=021 = —011= —0,12= 0 , where a is the frequency of a change in the
direction o f motion. Then Eqs. (10.133) assume the following form:
dp d Wp
- g r - \ - - § z ^ = a ( P 2 — Pi)^
(10.135)
dp2 d Wp
-w —
Equations of this type, or the telegraph equation which derives

from them (see below), were first obtained to describe the
one-dimensional diffusion o f photons by Fock (1926). Afterwards
they were used in describing molecular diffusion, taking into account
the finiteness o f the velocities of the diffusing particles, by Davidov
(1935), Cattaneo (1948—1949; 1958), R. Davis (1954) Venotte
(1958), and others. Such equations were proposed to describe
turbulent diffusion by Lyapin (1948; 1950), Goldstein (1951; here
the remarks of G. I. Taylor (1921) were developed into a mathemat
ical theory), and Gupta (1959); similar equations for two- and three-
dimensional turbulent diffusion were considered by Bourret (1961).
Assume that the diffusion takes place in a field of stationary

turbulence. In this case the values of a and W do not depend on t and
consequently may depend only on Z. Introducing new unknowns
p = pi + p 2, the probability density function for the coordinate o f a
diffusing particle, and q = W (p i — p^), the flux density of the
diffusing particles, we obtain
dp dq
+ 77T — 0 ;
^ + 2a , = - r - j / .
The first o f these equations, which is correct also for nonstationary

turbulence, expresses the law of mass conservation for the diffusing
admixture. According to this equation, one may assume
P= -W ’ ‘J = - - d F ’ (10.137)
where
denotes the probability distribution function for the coordinate

Z ( 2 , /) o f the diffusing particle. Substituting Eqs. (10.137) into the
second of Eqs. (10.136), we obtain for the function 'F the so-called
telegraph equation^
(10.138)
* T h is t y p e o f e q u a t io n [E q . 1 0 .1 3 8 )1 is c a lle d a telegraph e q u a t io n since it also

d escrib es th e la w o f tim e variation o f th e cu rren t in a c o n d u c t o r , fo r e x a m p le in a
telegraphic cab le , w ith r esista n ce R, c a p a c ita n c e C and self-in d u c ta n c e L, i f w e a ssu m e th a t
a = -^ a n d W = F ro m this a n a lo g y , it m a y b e c o n c lu d e d th a t th e e le c tr o n s
s o m e h o w c o m p le te a r a n d o m w a lk in th e c o n d u c t o r w ith a m e a n v e lo c ity
c h an g in g th eir d ir e c tio n o f m o t io n o n th e average w ith a fr e q u e n c y o f Rj 2 L \ in th is case, th e
\
value o f /C = is a n a lo g o u s t o th e d if fu s io n c o e f fic ie n t.
This equation is of the hyperbolic type; it describes the propagation

of a diffusing admixture having a finite velocity not exceeding U^;in
fact it is even supposed here that the velocity is always equal to in
absolute value. Also, this equation is only approximate since it is
based on a nonrigorous hypothesis concerning the Markov nature of
the function {Z(/), n(/)}; nevertheless, it is more accurate than the
ordinary, one-dimensional, semiempirical equation of turbulent
diffusion. Indeed, it is not difficult to show that our new diffusion
equation is a generalization of the parabolic diffusion equation; the
latter is obtained from Eq. (10.138) as a limiting case. This is easiest
to demonstrate in an example o f diffusion in a field of stationary
homogeneous turbulence where the coefficients a and W depend on
neither t nor Z, that is, are constant. In this case, p [and also pi and
P2] satisfies the same equation as ^F, namely, the telegraph equation
^ + (10.139)
If the parameters a and W are allowed to increase without bound in

such a way that the ratio WÎ2 a approaches a finite limit, which we
shall designate as K, then the telegraph equation (10.139) will be
converted into the parabolic diffusion equation ^ = K , which
corresponds to the case of homogeneous stationary turbulence.
The solution of Eq. (10.139), for a bounded space — o o < Z < o o
under the initial conditions (10.134), has the form
/ /
1 n~\ ' + ('1 — ' 2) - ^ / r-
' -iw /o [a-^ y 1 - w ) + - - 7 = 1# - /. (ax | / 1
(10.140)
where x = t ~ ta, t = Z /„ and I, are symbols of the Bessel

functions of an imaginary argument, and the function p is assumed
equal to zero when ici>ll^T. Equation (10.140) indicates that in this
theory the admixture distribution from an instantaneous point
source has a sharp edge, that is, a wave front b = ± \Vx in which a
finite fraction of the total amount of diffusing admixture is
concentrated, where this fraction decreases exponentially with time.

Behind this front there is a continuous wake formed by admixture
particles which have experienced multiple scattering, or a change in
direction of motion. In the center of this wake, namely, when
\t\<SiWx, it is not difficult to see that the right side of Eq. (10.140)
differs little from the function
■ _
2 n\V^z e
for T^ -a and any ei and 62; this is the solution of the parabolic
equation = K . K= Thus, the theory of diffusion with

a finite velocity leads to a noticeable discrepancy in the conclusions
from the parabolic diffusion equation only either for a very small
diffusion time 4 -,
d or close to the edge of the admixture cloud,
where the concentration obtained from any theory is very small. As
an illustration, graphs are presented in Fig. 83 for the continuous
part of the distribution p(Z\z, /)w hen ei = 62 = 1/2 and ei = 1, 62 =
0 for two values of ax, together with the graphs of the admixture
distribution corresponding to the parabolic diffusion equation.
The mean value and variance of the distribution (10.140) equal
— w/
(10.141)
W'X
{ z~ zr= -^ [i
they do not differ essentially from the values of
when a t 1. According to the model assumed here for the

diffusion process, the probability of scattering in a small time
interval equals ax + o{x) and scattering in nonintersecting time
t
intervals may be regarded as essentially independent random events.

Therefore, the number v ( ) of scatterings in the time interval is a
t t
random variable with a Poisson distribution having the parameter ax:

FIG . 8 3 . A c o m p a r is o n o f th e c o n t in u o u s part o f th e distri

b u t io n p { Z\ z, t) w h e n a x = 2 andocx = 5 fr o m d if fu s io n th e o r y
w ith a fin ite v e lo c ity w ith th e G aussian d is tr ib u tio n w h ic h f o l
lo w s fr o m th e pa rab o lic d if fu s io n e q u a tio n . C urve 1 co rre
sp o n d s t o th e right sid e o f E q. ( 1 0 . 1 4 0 ) , w it h o u t th e ter m s
c o n ta in in g th e 6 - fu n c t io n s , w h e n € i = € 2 = 1 /2 ; C urve 2, th e
sa m e, w h e n Ci = 1, € 2 = 0 ; th e d a sh e d curve is th e G aussian
d istr ib u tio n .
If we designate as Pm{%, t) the conditional probability density

function for Z{z, t) = Z { z , Iq+i) under the conditions rt(fo) = l,
V ( t ) = m, then obviously
This equation gives the expansion of the distribution (10.140) with

respect to the multiplicity of scatterings; using Eq. (10.140) it can be
se en th at t ) = 6 ( t ; Wr), and w h e n m 1 th e f u n c t io n s t )
are eq u a l t o z e r o w h e n |;;| > U”t, an d have th e f o l l o w i n g form ;
r \ ^ '( '" '^ 2 ) 1 /,

(10.143)
1172
Ibi < When a -> 0 0 , W —>co, the distribution
(10.140) asymptotically approaches the normal distribution with
mean value zero and variance-----

a = 2 /Cx which is the solution of the
parabolic diffusion equation ^ = K . The same limit, as we have
seen, is reached if a and W, and also - ^ — K, are fixed and S is

fixed, but t tends to infinity. Thus, for sufficiently large t at the
points t, far from the wave front ± Wx, this procedure for the
description of turbulent diffusion gives practically the same results as
the ordinary semiempirical theory based on the use of the parabolic
equation. This again indicates that this procedure is actually a
generalization of the ordinary semiempirical theory, in the sense that
it contains the latter as a limiting case.
As an example of the application of the theory of diffusion with a
finite velocity to the case of inhomogeneous turbulence, consider the
appearance of this theory as applied to vertical diffusion in the
atmospheric surface layer. We begin with the simplest case of a
nonstratified boundary layer. Within the limits of a logarithmic layer,
from dimensional considerations we have: = Xm*, a =/ju,/Zwhere X
and M are universal constants. It is easy to see that Eq. (10.138) with
such a and W becomes the usual diffusion equation dp/dt = 9/3Z K
dp/dz as a or (1 ^ 0°, W OT W'^/a -> 2K. Since H'*/2a has the
meaning o f eddy diffusivity K, we must have the relation ju = X^/x.
(For simplicity, we shall suppose now that a = 1; otherwise ■*. must be
replaced by x, =a% in all the subsequent equations.) Transforming to
dimensionless variables f = Z/L, x = hi^(t-to)lL where L is an arbitrary
length scale, we may rewrite Eq. (10.138) in the form
2n d'i> <?2 W .„ . ..
----- + ---------= ------ (10.144)
*2 C dx ^^2
where n = A/2k = const. Here, it is convenient to select the source

height H as the scale L used to measure the height Z, so that the
dimensionless variables (S, t) are defined by C = - ^ , and t =
Moreover, if // = 0, the case of a surface source, then it
is most convenient to use the dimensional variables f = Z,
T= The initial conditions which we wish to place on the
solution 'K(t, t ) can be formulated first in terms of the conditional
density functions />i and P2 which satisfy Eqs. (10.135);
Pi { Z \ I f , t ) -> i l l { Z — H ) , ( Z I / / , / ) - > C2?j ( Z — H ) for t -> ( q,
here £i and t j are the probabilities of the positive and negative initial
velocity of the diffusion particle, determined from physical consider
ations. (For example, for smoke particles from a stack, it is natural
to assume that ci = 1 , C2 = 0 , and for gas particles being spread from
the point of rupture o f a shell, that = sj = 1/2.) However, since
Eqs. (10.137) show that
z
d \'
/ ( P i + P-i) d Z and (/>, — P i) ,
the initial conditions for the function 'F may be written as
(C. 0)t =
=0
£ (C - 1). ^ = ( s2 _ 3 , )
where £(^) is the probabihty distribution function concentrated at

the point ? = 0 , that is, a function equal to zero when S < 0 , and
unity when S > 0 .
The solution of Eq. (10.144) in which we are interested and which
satisfies the condition of reflection of the admixture on the earth
surface Z = 0 will obviously be identical to the solution of the
equation
. 2n c)>r
in the entire space — c o < 5 < o o corresponding to the presence of a

symmetrical source at the point — H in addition to the source at the
point H. Therefore, we may assume
w a , T ) = v ,( s , t ) + ' f , ( - s . t ) .
where t ) is the solution of Eq. (10.146) in the whole space
which satisfies the initial conditions (10.145). In the special case of a

surface source when / / = 0 , it is obviously necessary to assume that
Cl = 1, E2 = 0 in the conditions (10.145); the s o l u t i o n t ) here
becomes the function Mô(C. t) which may depend only on the ratio
of its arguments g = — = , that is, has the form
X Ku^ {t — to)
'Ko(s. t ) =F(%). Here Eq. (10.146) becomes an ordianry second-order
differential equation for f ( |) , the solution of which, corresponding
to the conditions of our problem, has the form
1— for 0 < 5 < 1 ,

x) = F(5) =
0 for 5>1 or 5 < 0 ( 1 0 .1 4 7 )
[Monin (1956b)]. Consequently, the admixture concentration » =

is given here by
¥ (Z , / 10) = ~ . 0 .< Z < X h ,-. (1 0 .1 4 8 )

(l + Z /K x ) " + ‘ ’
Let us note that /*(1 — Q) = 1 according to Eq. ( 1 0 . 1 4 7 ) , that is, the

function F(l) here turns out to be continuous at the point 1 . in
other words, the fraction of admixture which is found at the wave
front Z = ku^r will be equal to zero at all times. (This is explained by
the fact that the frequency of scattering ^ = infinitely large
at the point Z = 0 , so that the particle which leaves the source at the
point Z = 0 must experience at least one scattering.) The behavior of
the admixture concentration near the front depends strongly on the
values o f the numerical constantn = -^ .W h e n n < 1 , or X< 2 )c« 0 .8 ,
the concentration tends to infinity on approaching the front, so that
the latter has the nature of a strong discontinuity. On the other
hand, when n > 1 , or X, > 2 x » 0 .8 , the concentration approaches
zero on approaching the front, and when n = 1 it approaches some

finite value. The experimental data of Kazanskiy and Monin (1957)
[see also Monin (1959a)] correspond best to a value of n close to
unity. These investigators measured the vertical profiles of the
admixture concentration in a smoke jet from a stationary, linear,
cross-wind source at a distance of 1 0 0 meters from the source.
Afterwards, the profiles observed were compared with the theoretical
curves / (6) = (1 — 5)"“ *(1 -f- S)"""* for different n ; the curve for n
= 1 gave the best fit. This deduction does not seem very convincing
since it does not take into account the effect of wind shear on
dispersion; however, it does agree well with other existing estimates.
Thus in one of the experiments of Kazanskiy and Monin, the
temporal rate of increase in the vertical diameter c? of a single puff of
smoke was studied with the aid of cinematography. The experi
mental curve d{t) was found to be a good approximation of the
straight line d = 2 W(t~to) where the value of \ = turned out to
be equal to 0.75, which corresponds to a value of n = \j2K close to
unity. On the other hand, it is natural to suppose that the root mean
square velocity of diffusing particles, which is equal to = Xm*
according to the theory of this subsection, must be of the same order
of magnitude as the typical Eulerian vertical velocity Oy; of the
boundary-layer flow. Since 0.9 to 1.2 [see Sects. 5.3 and
8.5], we again obtain the estimate X/2x 1.
To compare diffusion theory with finite velocity with the ordinary
semiempirical theory of turbulent diffusion, the function (10.148)
should be compared with the function (10.93') which is the solution
of a parabolic diffusion equation with eddy diffusivity of the form
K=x«,Z=/CiZ; the difference between x and K,'=ax,can evidently
now be neglected. This comparison, under the assumption that n=
1.25, is presented in Fig. 85, taken from Pasquill (1962b). We can see
that for each fixed = t-to, the profile of the mean concentration
t
d(Z) assumes very similar forms in the theory of diffusion with a

finite velocity, and in the ordinary semiempirical theory within the
range of values of Z corresponding to relatively large values of the
concentration. The essential difference of the profiles is observed
only near the front of the admixture cloud. Here, according to
diffusion theory, the concentration suddenly drops to zero with a
finite velocity, while according to the ordinary theory, it continues
to drop smoothly, remaining different from zero for values of Z as
large as desired.
Znxu^T
FIG. 8 5 . A c o m p a r is o n o f th e v ertica l d is tr ib u tio n o f c o n c e n tr a
tio n a c co r d in g t o d if fu s io n th e o r y w ith fin ite v e lo c ity (d a sh e d
lin e) w ith th e d istr ib u tio n o b t a in e d fr o m a pa rab o lic d if fu s io n
e q u a tio n w ith an e d d y d iffu s iv ity w h ic h inc r e a se s lin early w ith
h e ig h t (s o lid lin e).
It is interesting to compare the estimates of the Lagrangian

statistical characteristics obtained in Sect. 10.5 using the ordinary
semiempirical theory with the corresponding estimates from diffu
sion theory with a finite velocity; such a comparison may give an
idea of the accuracy of the semiempirical theory. For simplicity, let
us suppose that n = 1 and disregard the difference between y. and
x' = ax. According to Eq. (10.148),
00
<Z> =
=J J Z d ( Z ,t) d Z ^ J d(Z,t)dZ
0 0
f = (10.149)
J (1 + 1
= 4 ( l n 2 - 0 . 5 ) x «^t % 0.77xM^t w henn = 1, that i s , A = 2x.
We see that the theory of diffusion with a finite velocity leads to an

estimate of the constant b in the Lagrangian equation Z = bu^z [see
Eq. (9.61')] which is about 13% lower than the estimate from the
ordinary diffusion theory when a reasonable value o f vertical velocity

W is chosen. On the other hand, if we estimate the value of
= <Z - <Z»^ in a similar way, we obtain the following result:
1 1
=J Z ^ d ( Z , t ) d Z d i Z , t ) d Z - <Z>2=
0
= [ 8 ( 1.5 - 2 1 n 2) - 16 ( l n 2 - 0.5)2]x2«2^2 ~ Q .32x2u 2^ 2
. , (10.149')
when R = 1 .
We see that the estimate of the constant dgg in the Lagrangian

equation 0 3 3 (1 ) = is only about 30% of the corresponding
estimate obtained with the aid of ordinary diffusion theory, and the
estimate of the standard deviation about 55%, when the same
value of W is chosen. Therefore we must conclude that the ordinary
semiempirical diffusion theory may be used for an approximate
estimate of the position of the center of gravity of the admixture
cloud in a boundary layer, but it will give a seriously incorrect
estimate of the dispersion of the cloud.
W hen X - X D o t h e f u n c tio n ( 1 0 . 5 4 ) n a turally a p p ro a c h e s th e f u n c tio n (1 0 .9 3 ^ ) w h ic h is a

s o lu tio n o f th e p a rab o h c d iffu s io n e q u a t io n
df dZ * dZ ‘
In th e case o f a surface p o in t s o u r ce w W ch b e g in s t o p r o d u c e a d m ix tu r e , at tim e f = 0 , w ith

c o n s ta n t (u n it) rate, th e c o n c e n tr a tio n Oi (Z , /) is f o u n d by in teg r a tio n w ith r e sp e c t t o tim e
o f th e s o lu tio n o f E q. ( 1 0 . 1 4 8 ) ; w h e n / - > o o , it is easy t o o b ta in fr om th is th e fo llo w in g
e x p a n s io n in p o w e r s o f \ / t :
M Z . „ - 5 - , (Z„ f
1 In ^ ^ ( 10 . 150 )
This e q u a t io n p erm its o n e to ev alu ate th e rate o f a p p ro a c h o f th e c o n c e n tr a tio n pro file to

the log a r ith m ic profile w h ic h c o rr e sp o n d s to a s te a d y surface so urce.
In th e case o f an e le v a te d source at an arbitrary h e ig h t / / , the s o lu tio n o f th e
c o r r e sp o n d in g initial v alue p r o b le m fo r Eq. ( 1 0 . 1 4 6 ) m ay b e f o u n d u sin g th e general
R ie m a n n m e th o d for th e s o lu tio n o f s e c o n d -o r d e r h y p e r b o lic d iffe re n tia l e q u a tio n s . In
general, th is s o lu tio n is q u ite aw kw ard; it is in th e form o f an integral o f a c o m p le x
c o m b in a tio n o f h y p e r g e o m e tr ic fu n c tio n s ; see M onin ( 1 9 5 6 b ) . H o w e v er , in s o m e ca ses it
m ay b e su b sta n tia lly s im p h fie d . E x a m p le s o f this ty p e m a y b e fo u n d in M onin ( 1 9 5 5 ;
1 9 5 6 b ) ; h o w e v e r , w e shall n o t c o n s id e r th e m in d e ta il in th is b o o k .
Let us c o n s id e r b r ie fly d if fu s io n w ith a fin ite v e lo c ity in a th er m a lly stratified b o u n d a r y
~
layer. Here, th e c o n c e n tr a tio n d istr ib u tio n m u st b e d e fin e d a c co r d in g to d =
w h e r e ^F = M^(Z, 0 is th e s o lu tio n o f th e c o rr e sp o n d in g telegraph e q u a t io n ( 1 0 . 1 3 8 ) for

suitab le b o u n d a r y c o n d it io n s ; b e lo w w e shall use o n ly th e c o n d it io n o f r e fle c tio n o f th e
a d m ix tu r e at Z = 0 and th e initial c o n d it io n w h ic h c o rr e sp o n d s to th e p r e se n c e at th e tim e
t — tQ o f an in s ta n ta n e o u s p o in t a d m ix tu r e s o u r ce o f u n it o u t p u t at th e p o in t (0 , 0 , H). O n
the basis o f th e sim ilarity princip les d e v e lo p e d in C hapt. 4 w e m a y e x p e c t th a t th e
c o e f f ic ie n t s W, w ith v e lo c ity d im e n sio n s, and a, w ith fr e q u e n c y d im e n sio n s, m a y b e
r ep resen ted here in th e form
(1 0 .1 5 1 )
w h ere I is a le n g th scale d e fin e d b y Eq. ( 7 . 1 2 ) , ffi and cp2 are d im e n sio n le ss universal
fu n c tio n s , and 7. is a n u m erica l c o n s ta n t in t r o d u c e d for c o n v e n ie n c e , w h ic h a llo w s o n e to
im p o se th e a d d itio n a l r e s tr ic tio n fpi (0 ) = 1 o n th e fu n c tio n cpi (^). (In th is case X has th e
sam e m e a n in g as a b o v e , th at is, it d e n o t e s th e lim itin g value o f as
ratio o f Wjn.^_ for neutral s tr a tific a tio n .) S in ce our d iffu s io n th e o r y is already a p p r o x im a te ,

it is reason a b le to refer to a d d itio n a l a p p r o x im a te e x p r e s s io n s o f a s em iem p ir ic a l nature to
d e te r m in e th e fu n c tio n s (pi ( u ) and (pii(S)- T^îs m a y b e d o n e , for e x a m p le , as fo llo w s . We
take as a basis th e s e m iem p ir ic a l e q u a tio n / ( = w^ / ( l — a R f ) ^ [s e e S e c t. 6 ) ; if, in a d d itio n ,
w e assu m e th a t K ' ^ W l , th e n w e o b ta in
1 1
‘I (1 0 .1 5 1 )
'- 7 'W
w here / ( l ) is a universal fu n c tio n o f th e w in d p r o file in th e a tm o sp h e r ic surface layer. N o w
w e use th e fa ct th at in th e statio n a ry case Eqs. ( 1 0 . 1 3 6 ) have the fo rm
const, 2aq = — W —^ 2 ~y (1 0 .1 5 2 )
w here is a vertical a d m ix tu r e fiu x . F or n eutral s tr a tific a tio n , U'' = Xu,. = c o n s t; th e r e fo r e,
th e right side o f th e s e c o n d o f Eqs. ( 1 0 . 1 5 2 ) m a y b e w r itte n h ere in th e fo r m — .
Strictly speaking, in th e general case th is w ill n o t b e so; hoN ^ ver, it m a y b e h o p e d th a t here
also th e c o rr e sp o n d in g c o r r e c tio n to th e term — w ill b e r elatively sm all and

0
in sign ifican t. T h u s, as a first a p p r o x im a tio n , th e fo llo w in g m a y b e a ssu m ed :
2aq _ 2?___92 (0 _
dZ
O n th e o th e r h a n d , n e g le c tin g th e p o ssib le d iffe r e n c e s b e t w e e n th e e d d y a d m ix tu r e

d iffu s iv ity a n d th e e d d y v is c o s ity , th e grad ien t o f a s ta tio n a ry c o n c e n tr a tio n d istr ib u tio n ,
a cco r d in g t o sim ila rity t h e o r y , w ill eq u al
dZ r.u^Z
C om parin g th e t w o e x p r e s s io n s fo r and using E q. ( 1 0 . 1 5 1 ), w e ob ta in
V 3 = s V 'K > (1 0 .1 5 3 )
/'(Q
E q u a tio n s ( 1 0 . 1 5 0 ) , ( 1 0 . 1 5 1 \ a nd ( 1 0 . 1 5 3 ) w ith a k n o w n fu n c tio n f i Z) and a k n o w n

c o n s ta n t a , w h ic h m a y b e c o n s id e r e d as a ssu m in g d iffe r e n t values for stable and u n sta b le
s tr a tific a tio n , c o m p le t e ly d e te r m in e th e fo r m o f th e c o e f f ic ie n t s o f the telegraph e q u a tio n
( 1 0 . 1 3 8 ) for d if fu s io n in th e a tm o sp h e r ic surfacc layer w ith arbitrary str a tific a tio n . Here the
U^2
lim it a - > o o , W - > o o , — ►K = c o n s t, w h ic h c o r r e sp o n d s to r ep lacing th e telegraph
2a
e q u a t io n b y an ordinary p ara b o h c d if fu s io n e q u a t io n , w ill o b v io u sly b e eq u iv a len t t o the
lim it w h e n }-> o o .
U sing th e e q u a t io n s a b o v e and tra n sform in g to d im e n sio n le ss variables y and t —
l u , {t - t o )
w e can r ed u ce Eq. ( 1 0 . 1 3 8 ) to th e fo rm
(1 0 .1 5 4 )
w here n ~ and th e variable g an d th e fu n c tio n ^ ( g ) are d e te r m in e d b y th e fo llo w in g :
(10.155)
=/ ' 7 'Q
T he fu n c tio n s f ( ^ ) and and th e variable ^ = ^ (f) have d iffe r e n t fo r m s for neutral,

stable and u n sta b le s tr a tific a tio n . T h e case o f neutral stratifica tio n w as c o n s id e re d ab o v e;
here in all th e p r e ce d in g e q u a t io n s th e lim it L -> o or C — - > 0 m u st be tak en . B ut,
/ ' (C) ^ , I and g ^ ^ fo r sm all | C i; th e r e fo r e , Eq. ( 1 0 . 1 5 4 ) here a ssu m es th e for m

( 1 0 . 1 4 4 ) . U n d er stable stra tifica tio n , w e can use, for e x a m p le , for / ( S ) , the in ter p o la tio n
formula
where C2 = 1/Rfcr = o. In the case o f unstable stratification for large ^ we have

■ 4
/ ' (C) ^ . However, instead o f dwelling here on the analysis o f the situation for a
stratified boundary layer, w e remark only that the main difference from the neutrally
stratified case consists in the appearance o f the function ^ ( 5 ) in the equation, which
depends on stratification.
BIBLIOGRAPHY*
Abramovich, G. N., (1963) The Theory o f T urbulent Jets. Trans, by Scripta

Technica. Techn. ed. L. H. Schindel. Cambridge, Mass. M.I.T. Press.
Akhiezer, N. I., (1 9 6 5 ) The Classical M o m e n t Problem and som e R elated
Q uestions in A nalysis. Trans, by N. Kemmer. Edinburgh, Oliver and Boyd.
Allen, C. M., and E. A. Taylor, (1923) The salt velocity m eth o d o f water
measurem ent. Trans. A m e r Soc. M ech. E n g rs.,^ 5 , 2 8 5 - 3 4 1 .
Appleby, J. F., and W. D. O hmstede, (1965) Numerical solution o f the
distribution o f wind and turbulence in the planetary b oundary layer, iVoc.
A rm y Sci. C o n f, 1964, Vol. 1, Washington, 8 5 —99.
Aris, R., (19 5 6 ) On the dispersion o f a solute in a fluid flowing through a tube,
Proc. R o y. Soc., A235, No. 1200, 6 7 - 7 7 .
Arnol’d, V. 1., (1965a) Conditions for the nonlinear stability o f the plane
stationary curvilinear flows o f an ideal fluid, D oklady A N S S S R , 162, No. 5,
9 7 5 - 9 7 8 . (1965b) Variational principle for three-dimensional stationary
flows o f an ideal nuid,P rikl. Mat. iM e k h ., 29, No. 5, 8 4 8 - 8 5 1 . (1966a) On
one a priori estimate o f the theory o f hydrodynam ic stability, Izvestiya
Vuzov, Ser. Mat., No. 5(54), 3 —5. (1966b) Sur un principe variationnel
pour les ecoulements stationnaires des liquides parfaits et ses applications
aux problemes de stabilte non lineaires,/. Mecan., 5, No. 1, 2 9 - 4 3 .
Bakewell, H. P., and J. L. Lumley, (1967) Viscous sublayer and adjacent wall
region in turbulent pipe flov/, Phys. Fluids, 10, No. 9, 1880—1889.
Baldwin, L. W., and W. R. Mickelsen, (1962) T urbulent diffusion and
anem om eter measurements, J. Eng. M ech., 88, No. 1, 3 7 —69; discussion,
151-153.
Barad, M. L., (1958) Project Prairie Grass, a field program in diffusion, Vols.
1 - 2 , G eophys. Res. Pap., No. 59, Cambridge, Mass; (1959) Analysis o f
diffusion studies at O ’Neill, A dv. G eophys., 6 (Atmospheric diffusion and
air pollution), 3 8 9 - 3 9 8 ; (1963) Exam ination o f a wind profile proposed by
S w in b an k ,/. A ppl. M eteor., 2, No. 6, 7 4 7 - 7 5 4 .
Barad, M. L., and D. A. Haugen (1959) A preliminary evaluation o f S u tto n ’s
hypothesis for diffusion from a continuous point source, 7. M eteorol. 16,
No. 1, 12 20.
Barenblatt, G. I., (1953) Motion o f suspended particles in a turbulent flow,
*Many o f the Russian journals have been translated cover-to-cover during the last decade;
the reader interested in reading a particular article in translation should enquire o f his
librarian.
695
69 6 BIBLIOGRAPHY
P rik l M atem . M ekh., 17, No. 3, 2 6 1 - 2 7 4 ; Motion o f suspended particles in

a turbulent flow occupying a half-space or plane channel o f finite depth,
Prikl. M atem . M ekh., 19, No. 1, 6 1 - 6 8 .
Barnes, H. T., and E. G. Coker, (1905) Experiments on streamline m otion and
the measurements o f critical velocity, Proc. R o y . Soc. L ondon, 74, No. 503,
341-356.
Batchelor, G. K., (1949a) The role o f big eddies in homogeneous turbulence,
Proc. R o y . Soc., A 195, No. 1043, 5 1 3 - 5 3 2 ; (1949b) Diffusion in a field o f
homogeneous turbulence, A ustr. J. Sci. Res., A2, No. 4, 4 3 7 - 4 5 0 ; (1957)
Diffusion in free turbulent shear flows, J. F luid M ech., 3, No. 1, 6 7 —80;
(1959) Diffusion from sources in a turbulent boundary layer (unpublished
manuscript); (1964) Diffusion from sources in a turbulent boundary layer,
A rch. Mech. Stosow anej, 16, No. 3, 6 6 1 - 6 7 0 ; (1967) A n In tro d u ctio n to
F luid Dynam ics. Cambridge, the University Press.
Batchelor, G. K., and A. E. Gill, (1962) Analysis o f the stability o f axisymmetric
F luid M ech., 14, No. 4, 5 2 9 - 5 5 1 .
Batchelor, G. K., and A. A. Townsend, (1956) T urbulent diffusion. Surveys in
Mechanics, (G. K. Batchelor and R. M. Davies, eds.) Cambridge, the
University Press, 3 5 2 - 3 9 9 .
Batchelor, G. K., A. M. Binnie, and O. M. Phillips, (1955) The mean velocity of
discrete particles in turbulent flow in a pipe. Proc. Phys. Soc., B68, No. 432,
1095-1104.
Beguier, C., (1965) Mesures des tensions de Reynolds dans un dcoulement
dissym^trique en regime turbulent incompressible, J. M ^can., 4, No. 3,
319-334.
Belyakova, V. K., (1950) Stability o f a viscous fluid flow in a straight circular
tube,PnA:/. M atem. M ekh., 14, No. 1, 105—110.
Benney, D. J., (1961) A nonlinear theory for oscillations in a parallel f lo w ,/.
F luid M ech., 10, No. 2, 2 0 9 - 2 3 6 ; (1 9 6 4 ) Finite-amplitude effects in an
unstable laminar boundary l a y e r . Fluids, 7, No. 3, 3 1 9 - 3 2 6 .
Berlyand, M. E., (1947) T heory o f variation o f wind with height, Trudy
Nauchno-Issl. Uchr. Glavn. Upr. Gidrometeluzhby (Proc. H ydrom eteorolog
ical Sci. Institutes), ser. 1, No. 25, 16—67.
Berlyand, M. E., E. L. Genikhovich, V. P. Lozhkina, and R. 1. Onikul, (1963)
Numerical solution o f equations o f turbulent diffusion, and calculations o f
atmospheric contam ination near industrial enterprises, Trudy Gl. Geoflz.
Observ. (Proc. Main Geophys. Observ., Leningrad), No. 138, 3 - 1 7 ; (1964)
Numerical investigation o f atmospheric diffusion under normal and ab
normal stratification conditions, Trudy Gl. Geoflz. Observ. (Proc. Main
Geophys. Observ., Leningrad), No. 158, 2 2 - 3 1 .
Berlyand, M. E., and R. L Onikul, (1968) Physical basis o f the calculation of
atmospheric dispersion of industrial air contaminations, Trudy Glavn.
Geofiz. Observ. (Proc. Main Geophys. Observ., Leningrad), N o. 234, 3 - 2 7 .
Bernstein, A. B., (1966) A n examination o f three wind profile hypotheses, 7.
BIBLIOGRAPHY 697
A ppl. M eteor., 5, No. 2, 2 1 7 - 2 1 9 .

Best, A. C., (1935) Transfer o f heat and m o m entum in the lowest layers o f the
atmosphere, Geophys. M em ,, 7, No. 65.
Betchov, R., (1960) On the mechanism o f turbulent transition, Phys. Fluids, 3,
No. 6, 1 0 2 6 - 1 0 2 7 .
Betchov, R., and W. 0 . Criminale, Jr., (1967) Stability o f Parallel Flows,
Academic Press, New Y ork-London; (1966) Spatial instability o f the inviscid
jet and wake, Phys. Fluids, 9, No. 2, 3 5 9 - 3 6 2 .
Betchov, R., and A. Szewczyk, (1963) Stability o f shear layer between parallel
streams, Fluids, 6, No. 10, 1 3 9 1 -1 3 9 6 .
Birikh, R. V., (1966) On small disturbances o f the plane-parallel flow with cubic
volocity profile, Prikl. Mat. M ekh., 30, No. 2, 3 5 6 - 3 6 1 ; (1965) Spectrum
o f small perturbations o f a plane-parallel Couette flow , Prikl. Mat. M ekh.,
29, No. 4 , 7 9 8 - 8 0 0 .
Birkhoff, G., (1962) Helmholtz and Taylor instability,/>oc. Sym p . A ppl. M ath.,
13 (Hydrodynam ic instability), 5 5 —76.
Birkhoff, G., and J. Kampe de Fdriet, (1962) Kinematics o f homogeneous
turbulence, D. Metric transitivity, 7. Math. M ech., 11, No. 3, 3 1 9 - 3 4 0 .
Birkhoff, G., and E. H. Zarantonello, (1957) Jets, Wakes and Cavities, Academic
Press, N. Y.
Bisshopp, F. E., (1962) Nonlinear effects in thermal convection, 7. M ath, and
M ech., 11, No. 5 , 6 4 7 - 6 6 3 .
Bjorgum, O., (1953) On the velocity, tem perature and hum idity profiles, Quart.
J. R oy. Meteor. Soc., 79, No. 339, 8 4 - 9 3 .
Blackadar, A. K., (1962) The vertical distribution o f wind and turbulent
exchange in a neutral atmosphere, /. G eophys. Res., 67, No. 8, 3 0 9 5 - 3 1 0 2 .
Blackadar, A. K., (1967) External parameters o f the wind flow in the barotropic
boundary layer of the atmosphere, in: Global A tm ospheric Research
Programme (Rep. Study Conf., Stockholm , 28 J u n e -11 July 1967),
Stockholm , IV. 1 I V. l l .
Blasius, H., (1913) Das Ahnlichkeitsgesetz bei Reibungsvorgangen in F’liis-
sigkeiten. Forschungsarbeiten des Ver. Deutsch. Ing. No. 131, Berlin.
Blinova, E. N., and I. A. Kibel, (1937) The application o f the turbulence theory
to the height distribution o f the temperature and wind velocity, in:
Dynam ical M eteorology (ed. by B. I. Izvekov and N. E. Kochin), Vol. 2,
G idrometeoizdat Press, Leningrad-Moscow, 2 8 - 3 6 .
Blom, J., and D. A. DeVries, (1968) On the value o f turbulent Prandtl num ber,
in: Heat- and Mass Transfer, Proc. 3rd All Soviet Union Conf. (ed. by A. V.
Lykov and B. M. Sm ol’skiy), Vol. 1, Energiya Press, Moscow, 1 4 7 - 1 5 4 .
Bobyleva, 1. M., S. S. Zilitinkevich, and D. L. Laykhtman, (1 9 6 7 ) Turbulent
exchange in the thermally stratified atmospheric planetary boundary layer,
in: A tm ospheric Turbulence and R adio Wave Propagation (ed. by A. M.
Yaglom and V.I. Tatarskiy), Nauka Press, Moscow, 1 7 9 - 1 9 0 .
Bochner, S., (1933) M onotone F unktionen, Stieltjessche Integrale und
698 BIBLIOGRAPHY
harmoniche Analyse, Math. A n n ., 108, No. 3, 3 7 8 - 4 1 0 . (For English

translation, see Bochner, 1959.) (1959) Lectures on Fourier Integrals,
Princeton Univ. Press, Princeton, N. J.
Borkowski, J., (1964) The wind profile and turbulent diffusion in a stable
surface layer o f the atmosphere, A cta G eophys. Polonica, 12, No. 4,
199-205.
Bosanquet, C. H., and J. L. Pearson, (1936) The spread o f smoke and gases from
chimneys, Trans. Faraday Soc., 32, No. 8, 1 2 4 9 -1 2 6 3 .
Bourret, R. C., (1961) Turbulent diffusion in two and three dimensions by the
random-walk model with m em ory, Canad. J. Phys., 39, No. 1, 1 3 3 - 1 4 0 .
Boussinesq, J., (1877) Essai sur la theorie des eaux courantes, A/ew. pres, par
div. savants a VAcad. ScL, Paris,23, No. 1, 1 - 6 8 0 ; (1897) Theorie de
I’ecoulement tourbillonnant et tum ultueux des Hquides dans les lits
rectilignes a grande section, I - I I , Gauthier-Villars, Paris.
Bovsheverov, V. M., and V. P. Voronov, (1960) An acoustic anemometer,
Izvestiya A N S S S R , ser. g e o f iz .,^ o . 6, 8 8 2 - 8 8 5 .
Bovsheverov, V. M., A. S. Gurvich, M. I. Mordukhovich, and L. R. Tsvang,
(1962) Instruments for measuring tem perature and wind-velocity fluctua
tions, and statistical analysis o f the results o f measurements. Collection:
Atmospheric Turbulence, T rudy Inst. Fiz. A tm o s. A N SS S R Proc. o f Inst.
Phys. A tmosph. Acad. Sci. USSR), No. 4, 2 1 —29.
Bovsheverov, V. M., A. S. Gurvich, V. I. Tatarskiy, and L. R. Tsvang, (1959)
Instruments for the statistical analysis o f turbulence, T m d y Soveshch.
Issled. M ertsaniya Z vezd (Proc. o f Conference on Stellar Scintillation),
Acad. Sci. Press, Moscow, 2 6 - 3 2 .
Bovsheverov, V. M., A. S. Gurvich, and L. R. Tsvang, (1959) Direct measure
ments o f turbulent heat flow in the atmospheric surface XdiyQx. D oklady A N
S S S R , 125, No. 6, 1 2 4 2 - 1 2 4 5 .
Bradley, E. F., (1968) A shearing-stress meter for micrometeorological studies.
Quart. J. R o y . M eteor. Soc., 94, No. 401, 3 8 0 - 3 8 7 .
Browand, F. K., (1966) An experimental investigation of the instability of an
incompressible separated shear layer, J. F luid M ech., 26, No. 2, 2 8 1 - 3 0 7 .
Brummage, K. G., (1968) The calculation o f atmospheric dispersion from a
sidiok, A tm o sp h . E nvironm ent, 2, No. 3, 197—224.
Brundrett, E., W. D. Baines, J. Peregrym, and P. R. Burroughs, (1 9 6 5) Inner and
outer law descriptions o f temperature and velocity in two- and three-
dimensional boundary layers, AGARDograph 97, Part 2.
Brushlinskaya, N. N., (1965) Behavior o f solutions o f hydrodynam ical equations
when the Reynolds num ber crosses the critical one, D oklady A N S S S R ,
162, No. 4, 7 3 1 - 7 3 4 .
Budyko, M. I., (1946) D istribution o f meteorological variables in the atm os
pheric surface layer, Izvestiya A N S S S R , ser. geogr. i geofiz., 10, No. 4,
3 1 7 —329; (1948) Evaporation Under N atural C onditions, G idrometeoizdat
Press, Leningrad.
BIBLIOGRAPHY 69 9
Burns, J. G., W. H. J. Childs, A. A. Nicol, and M. A. S. Ross, (1959)

Development and use o f a vane device for boundary-layer m e a s u r e m e n ts ,/
F lu idM ech., 6, No. 1 , 9 7 - 1 1 2 .
Burton, R. A., (1965) A simple universal velocity profile equation, y4./.y4.>4.y., 3,
No. 4, 7 8 4 - 7 8 5 .
Busch, N. E., (1965) A note on the similarity hypothesis for wind profiles,
Danish A to m ic Energy C om m ission, Ris0 R eport No. 100, 1 - 2 6 .
Busch, N. E., J. A. Frizzola, and I. A. Singer, (19 6 8 ) The micrometeorology of
the turbulent flow-field in the atmospheric surface boundary layer, A cta
P olytechn. Scandinavica, Phys. incl. Nucleonics Ser., No. 59, 1 - 4 5 .
Busch, N. E., and H. A. Panofsky, (1968) Recent spectra o f atmospheric
turbulence. Quart. J. R o y. M eteorol. Soc., 94, No. 4 0 0 , 132—148.
Businger, J. A., (1955) On the structure o f the atmospheric surface layer, J.
M eteor., 12, No. 6, 5 5 3 - 5 6 1 ; (1959) A generalization o f the mixing-length
concept, / M eteor., 16, No. 5, 5 1 6 - 5 2 3 ; (1966) T ransport o f m o m entum
and heat in the planetary boundary layer, Proc. Sym pos. A rctic H eat B udget
and A tm o sp h . Circul. (Memo R M - 5 2 3 3 —NSF, Rand Corp.), 3 0 5 —332.
Businger, J. A., M. Miyake, A. J. Dyer, and E. F. Bradley, (1967) On the direct
determination o f the turbulent heat flux near the ground, J. A p p l
M ete o ro l, 6, No. 6, 1 0 2 5 - 1 0 3 2 .
Busse, F. H., (1967a) N onstationary finite amplitude convection, J. Fluid
Mech.,2%, 2, 223 - 239; (1967b) On the stabiUty o f two-dimensional
convection in a layer heated from below, / M ath, and Phys., 46, No. 2,
1 4 0 - 1 5 0 ; (1967c) The stability o f finite amplitude cellular convection and
its relation to an extrem um principle, J. F luid M ech., 30, No. 4, 6 2 5 —649.
Bysova, N. L., and G. B. Mashkova, (1966) Dimensionless characteristics o f the
wind velocity according to measurements in the lower 300-meter
atmospheric layer, Izvestiya A N SS S R , ser. fiz. atm. okeana, 2, No. 7,
681-687.
Byutner, E. K., and D. L. Laykhtm an, (19 6 3 ) Dispersion o f passive particles
from a point source in an inhomogeneous medium, Trudy Leningr.
G idromet. In-ta (Proc. Leningr. H ydrom eteor. Inst.) No. 15, 130—136.
Calder, K. L., (1949) Eddy diffusion and evaoporation in flow over
aerodynamically sm ooth and rough surfaces: a treatm ent based on
laboratory laws o f turbulent flow w ith special reference to conditions in the
lower atmosphere. Quart. J. Mech. A ppl. M ath., 2, No. 2, 1 5 3 - 1 7 6 ;( 1 9 6 1 )
Atmospheric diffusion o f particulate material, considered as a
boundary-value problem, J. M eteor., 18, No. 3, 4 1 3 - 4 1 6 ; (19 6 5) On the
equation o f atmospheric diffusion, Quart. J. R o y. M eteor. Soc., 91, No.
3 9 0 ,5 1 4 -5 1 7 .
Capriz, G., G. Ghelardoni, and G. Lombardi, (1966) Numerical study o f the
stability problem for Couette i\o ^ ,P h y s . Fluids, 9, No. 10, 1 9 3 4 - 1 9 3 6 .
Carr, A. D., (1962) The transport o f heat in a fluid in turbulent pipe flow.
Reactor, 1 , 6 - 1 4 .
70 0 BIBLIOGRAPHY
Carrier, G. F., and F. D. Carlson, (1946) On the propagation o f small

disturbances in a moving compressible fluid, Quart. A ppl. M ath., 4, No. 1,
1- 12.
Case, K. M., (1960a) Stability o f inviscid plane Couette flow, P/zys. Fluids, 3,
No. 2, 1 4 3 - 1 4 9 ; (1960b) Stability o f an idealised atmosphere. I. Discussion
o f results, Phys. Fluids, 3, No. 2, 1 4 9 - 1 5 5 ; (1960c) Taylor instability o f an
inverted atmosphere, Phys. Fluids, 3, No. 3, 3 6 6 - 3 6 8 ; (1961)
H ydrodynamic stability and the inviscid limit, J. F luid M ech., 10, No. 3,
4 2 0 - 4 2 9 ; (1962) H ydrodynamic stability and the initial value problem,
Proc. S ym p . A ppl. M ath,, 13 (Hydrodynam ic instability), 2 5 - 3 3 .
Cattaneo, C., ( 1 9 4 8 - 1 9 4 9 ) Sulla conduzione del calore, A t t i sem. Mat. Fis.
Univ. M odena, 3, 8 3 - 1 0 1 ; (1958) Sur une forme de I’equation de la chaleur
eliminant le paradoxe d ’une propagation instance, C om pte Rend. Acad. Sci.,
247, No. 4, 4 3 1 - 4 3 3 .
Catton, I., (1966) Natural convection in horizontal liquid layers, Fluids, 9,
No. 1 2 , 2 5 2 1 - 2 5 2 2 .
Cavers, S. D., N. Y. Hsu, W. G. Schlinger, and B. H. Sage, (1953) Tem perature
gradients in turbulent gas streams. Behavior near boundary in
two-dimensional flow, Ind. Eng, Chem,, 45, No. 9, 2 1 3 9 —2145.
Cermak, J. E., (1963) Lagrangian similarity hypothesis applied to diffusion in
turb u len t shear flow, J. F luid M ech., 15, No. 1, 4 9 —64.
Cermak, J. E., and H. Chuang, (1967) Vertical-velocity fluctuations in thermally
stratified shear-flows, in: A tm o sp h eric Turbulence and R adio Wave
Propagation (ed. by A. M. Yaglom and V. I. Tatarskiy), Nauka Press,
Moscow, 9 3 - 1 0 4 .
Cermak, J. E., V. A. Sandborn, E. J. Plate, G. H. Binder, H. Chuang, R. N.
Meroney, and S. I to, (1966) Simulation o f atmospheric m otion by
wind-tunnel flows. Fluid Dynamics and Diffusion Lab., Colorado State
U niv.,T echn. Rep. CER66JEC-VAS-EJP-GJB-HC-RNM-SI17, 101 pp.
Chalikov, D. V., (19 6 8 ) On wind-velocity and temperature profiles in
atmospheric surface layer with stable stratification, Trudy Glavn. Geofiz.
Observ, (Proc. Main Geophys. Observ., Leningrad), No. 207, 1 7 0 - 1 7 3 .
Chamberlain, A. C., (1966) Transport o f gases to and from grass and grass-like
surfaces, Proc. R o y . Soc., A 290, No. 1421, 2 3 6 - 265; (1968) Transport o f
gases to and from surfaces w ith bluff and wave-like roughness elements,
Quart. J. R o y. Meteor. Soc., 94, No. 401, 3 1 8 - 3 3 2 .
Chandrasekhar, S., (1943) Stochastic problem in Physics and astronom y,
M od. Phys., 15, No. 1, 1 - 8 9 ; (1953) The stability o f viscous flow between
rotating cylinders in the presence o f a magnetic field, Proc. R o y. Soc.,
A216, No. 1126, 2 9 3 - 3 0 9 ; (1954) The stability o f viscous flow between
rotating cylinders, M athem atika, 1, 5 - 1 3 ; (1958) The stability o f viscous
flow between rotating cylinders,Proc. R o y. Soc., 246, No. 1246, 3 0 1 - 3 1 1 ;
(1960) The stabiHty o f inviscid flow between rotating cylinders, J. Indian
M ath. Soc, (N, S.), 24, 2 1 1 —221; (1961) H ydrodynam ic and H ydrom ag-
netic S ta b ility, Clarendon Press, Oxford.
BIBLIOGRAPHY 701
Chandrasekhar, S., and D. D. Elbert, (1962) The stabiHty o f viscous flow

between rotating cylinders. II. Proc. R o y. Soc., A268, No. 1333, 1 4 5 - 1 5 2 .
Chapman, D. R., and R. H. Kester, (1 9 5 3 ) Measurements o f turbulent skin
friction on cylinders in axial flow at subsonic and supersonic velocities, J.
A eronaut. Sci., 20, No. 7 , 4 4 1 - 4 4 8 .
Chapman, S., and T. G. Cowling, (1952) The M athem atical Theory o f
N on-uniform Gases, Cambridge Univ. Press.
Charnock, H., (1958a) Meteorology, Sci. Progress, 46, No. 183, 4 7 0 - 4 8 7 ;
(1958b) Physical oceanography, Sci. Progress, 46, No. 183, 4 8 7 - 5 0 1 ;
(1967a) Com m ents on “ An examination o f three wind-profile hypotheses,”
J. A ppl. M eteor., 6, No. 1, 2 1 1 - 2 1 2 ; (1967b) Flux-gradient relations near
the ground in unstable conditions. Quart. J. R o y. M eteor. Soc., 9 3 , No. 395,
97-100.
Charnock, H., and T. H. Ellison, (1959) Fluctuations o f tem perature and
hum idity at 134 m over the sea, Deutsch. Hydrograph. Zs., 12, No. 4,
171-180.
Chatwin, P. C., (1968) The dispersion o f a p u ff o f passive contam inant in the
constant stress region. Quart. J. R o y. M eteor. Soc., 94, No. 401, 3 5 0 - 3 6 0 .
Chen, M. M., and J. A. Whitehead, (1968) Evolution o f two-dimensional periodic
Rayleigh convection cells o f arbitrary wave-numbers, J. F luid M ech., 31,
No. 1, 1 - 1 5 .
Chen, T. S., D. D. Joseph, and E. M. Sparrow, (1966) Evaluation o f Tietjens
function in stabiHty calculations, P/zjv's. Fluids, 9, No. 12, 2 5 1 9 —2521.
Chen, T. S., and E. M. Sparrow, (1967) Stability o f the developing laminar flow
in a parallel-plate channel, J. F luid M ech., 30, No. 2, 2 0 9 - 2 2 4 .
Chilton, T. H., and A. P. Colburn, (1934) Mass transfer (absorption) coefficients,
Ind. Eng. Chem,, 26, No. 11, 1 1 8 3 - 1 1 8 7 .
Chou, P. Y. (1945a) On velocity correlations and the solutions o f the equations
o f turbulent fluctuation. Quart. A ppl. M ath., 3, No. 1, 3 8 —54; (1945b)
Pressure flow o f a turbulent fluid between parallel planes. Quart. A ppl.
M ath., 3, No. 3, 1 9 8 - 2 0 9 ; (1947) Turbulent flow along a semi-infmite
plane. Quart. Appl. M ath., 5, No. 3, 3 4 6 - 3 5 3 .
Christopherson, D. G., (1940) N ote on the vibration o f membranes, Quart. J.
M ath., 11, No. 1 , 6 3 - 6 5 .
Chu, B. T., and L. S. G. Kovasznay, (1958) NonUnear interactions in a viscous
heat-conducting compressible g a s ,/. F luid M ech., 3, No. 5, 4 9 4 - 5 1 4 .
Chun, D. G., and W. H. Schwartz, (1967) Stability o f the plane incompressible
viscous wall jet subjected to small disturbances, Phys. Fluids, 10, No. 5,
911-915.
Clauser, F. H., (1956) The turbulent boundary layer. A dv. A ppl. M ech., 4, N. Y.,
1-51.
Coantic, M., (1965) Remarques sur la structure de la turbulence a proximite
d ’une paroi, C om pte R end. Acad. Sci., 260, No. 11, 2 9 8 1 - 2 9 8 4 ; (1966)
C ontribution a I’etude de la structure de la turbulence dans une conduite de
702 BIBLIOGRAPHY
section circulaire, These doctorat d ’etat es sciences physiques, Univ.

d’Aix-Marseille, Marseille; (1967a) Evolution, enfonction du nombre de
Reynolds, de la distribution des vitesses moyennes et turbulentes dans une
conduite, C om pte R end. Acad. ScL, 264, No. 19, A 8 4 9 - A 8 5 2 ; (1967b)
Quelques resultats nouveaux sur la structure de la turbulence dans la
sous-couche visqueuse, C om pte R end. Acad. ScL, 264, No. 20,
A 9 6 7 -A 9 7 0 .
Coles, D., (19 5 5 ) The law o f the wall in turbulent shear flow, 50 Jahre
Grenzschichtforschung, Vieweg, Braunschweig, 1 5 3 - 1 6 3 ; (1956) The law
o f the wake in turbulent boundary layer, /. F luid M ech., 1, No. 3,
1 9 1 - 2 2 6 ; (1962) Interface and interm ittency in turbulent shear flow,
M ecanique de la turbulence (Coll. Intern, du CNRS a Marseille, Paris, Ed.
CNRS, 2 2 9 - 2 4 8 ; (1965) Transition in circular Couette flow, J. F luid
M ech., 21, No. 3 , 3 8 5 - 4 2 5 .
Coles, D., and C. W. Van A tta, (1967) Digital experiments in spiral turbulence,
Phys. Fluids, 10, Supplem ent, S 1 2 0 -S 1 2 1 .
Collis, D. C., (1948) The diffusion process in turbulent flow ,/lw sm C ouncilSci.
Ind. Res., Div. Aero., Rep. A55.
Comolet, R., (1950) Recherches sur la genese de la turbulence dans les conduite.s
en charge, Publ. Sci. Techn. Min. A ir, No. 246, Paris.
Comte-Bellot, G., (1963) Coefficients de dissymetrie et d ’aplatissement, spectres
et co rd a tio n s en turbulence de c o n d u ite ,/. Mecanique, 2, No. 2, 1 0 5 - 1 2 8 ;
(1965) Ecoulement turbulent entre deux parois paralleles, PwM Sci. Techn.
Min. Air, No. 419, Paris.
Corcoran, W. H., F. Page, W. G. Schinger, and B. H. Sage, (1952) Gradients in
turbulent gas streams, Ind. Eng. Chem., 44, No. 3, 4 1 0 —430.
Corcos, G. M., and J. R. Sellars, (1959) On the stability o f fully developed flow
in a pipe, J. Fluid Mech., 5, No. 1, 9 7 - 1 1 2 .
Corrsin, S., (1943) Investigation o f flow in an axially symmetric heated jet of
air, Nat. A dv. Com. A eronaut., Wartime R eport W-94; (1959a) Outline of
some topics in homogeneous turbulent flow, J. G eophys. Res., 64, 12,
1 2 3 4 - 2 1 50; (1959b) Progress report on some turbulent diffusion research.
A dv. G eophys., 6 (Atmospheric diffusion and air pollution), 1 6 1 - 1 6 4 ;
(1962) Theories o f turbulent M echanique de la turbulence (Coll.
Intern, du CNRS a Marseille), Paris, Ed. CNRS, 2 7 - 5 2 ; (1963) Estimates of
the relations between Eulerian and Lagrangian scales in large Reynolds
number turbulence, J. A tm o sp h . Sci., 20, No. 2, 1 1 5 - 1 1 9 ; (1963)
Turbulence: E xperim ental M ethods, H andbuch der Physik, Vol. V III/2,
Springer, Berlin-Gottingen-Heidelberg, 5 2 5 - 5 9 0 .
Corrsin, S., and A. L. Kistler, (1955) The free-stream boundaries o f turbulent
flows, A dv. Com. A eronaut., Rep. No. 1244.
Corrisin, S., and M. S. Uberoi, (19 5 0 ) F urther experiments on the flow and heat
transfer in a heated turbulent air jet, Nat. A dv. Com. A eronaut., Rep. No.
998; (1951) Spectra and diffusion in a round turbulent jet, A dv. Com.
A eronaut, Rep. No. 1040.
BIBLIOGRAPHY 703
Cramer, H. E., (1967) Turbulent transfer processes for quasi-homogcneous flows

within the atmospheric surface layer, Phys. Fluids, 10, Supplement,
S 2 4 0 -S 2 4 6 .
Cramer, H. E., and F. A. Record, (1953) The variation with height o f the vertical
flux o f heat and m om en tu m , / M eteor., 10, No. 3, 2 1 9 - 2 2 6 .
Crawford, T. V., (1965) Moisture transfer in free and forced convection. Quart.
J. R o y. M eteorol. Soc., 91, No. 387, 1 8 - 2 7 .
Criminale, W. O., Jr., (1967) Interaction o f the laminar boundary layer with
free-stream turbulence, Fluids, 10, Supplem ent, S 1 0 1 -S 1 0 7 .
Criminale, W. 0 . , Jr., and L. S. G. Kovasznay, (1962) The growth o f localized
disturbances in a laminar boundary la y e r ,/. F luidM ech., 14, No. 1, 5 9 - 8 0 .
Crocco, L., (1965) T urbulent boundary layer and mixing coefficient, / Soc. Ind.
A ppl. M ath., 13, No. 1, 2 0 6 - 2 1 5 .
Crow, S. C., (1968) Viscoelastic properties of fine-grained incompressible
turbulence, / F luid Mech., 33, No. 1, 1 - 2 0 .
Csanady, G. T., (1967a) Concentration fluctuations in turbulent diffusion, /
A tm o sp h . Sci., 24, No. 1, 2 1 - 2 8 ; (1967b) Variance o f local concentration
fluctuations, Phys. Fluids, 10, Supplem ent, S 7 6 - S 7 8 ; (1967c) On the
“ resistance law” o f a turbulent Ekman layer, / A tm osph . Sci., 24, No. 5,
467-471.
Csanady, G. T., G. R. llilst, and N. E. Bowne, (1968) T urbulent diffusion from a
cross-wind line source in shear flow at Fort Wayne, Indiana, A tm osp h.
Environm ent, 2, No. 3, 2 7 3 - 2 9 2 .
Curie, N., (1956) On hydrodynam ic stability in unlimited fields o f viscous flow,
Proc. R o y. Soc., A 2 3 8 ,N o . 1 2 1 6 , 4 8 9 - 5 0 1 .
Darcy, H., (1858) Recherches experimentales relatives aux mouvement de I’eau
dans les tuyaux,M ^m . Sav. Etrang. Acad. Sci., 15, 2 6 5 - 3 4 2 .
Davey, A., (1962) The growth o f Taylor vortices in flow between rotating
c y l in d e r s ,/ F luid M ech., 14, No. 3, 3 3 6 - 3 6 8 .
Davey, A., R. C. Di Prima, and J. T. Stuart, (1968) On the instability o f Taylor
vortices, / Fluid M ech., 31, No. 1, 1 7 - 5 2 .
Davidson, B., and M. L. Barad, (1956) Some comments on the Deacon wind
profile, Trans. A m . Geophys. Union, 37, No. 2, 1 6 8 - 1 7 6 .
Davies, D. R., (1950) Three-dimensional turbulence and evaporation in the lower
atmosphere, I - I I , Quart. J. Mech. A ppl. M ath., 3, No. 1, 5 1 - 6 3 , 6 4 - 7 3 ;
(1954a) On diffusion from a continuous point source at ground level into a
turbulent atmosphere, Quart. J. Mech. A ppl. M ath., 7, No. 2, 1 6 8 - 1 7 8 ;
(1954b) A note on the two-dimensional equation o f diffusion in the
atmosphere. Quart. J. R o y. M eteor. Soc., 80, No. 345, 4 2 9 - 4 3 4 .
Davies, R. W., (1954) The connection between the Smoluchowski equation and
the Kramers-Chandrasekhar equation, Rev., 93, No. 6, 1169—1171.
Davies, S. J., and C. M. White, (1928) An experimental study of the flow o f
water in pipes o f rectangular section, Proc. R o y. Soc., A 119, No. 781,
92-107.
704 BIBLIOGRAPHY
Davis, S. H., (1968) C orvection in a box: on the dependence o f preferred

wave-number upon the Rayleigh num ber at finite a m p litu d e,/. F luidM ech.,
32, No. 3, 6 1 9 - 6 2 4 .
Davis, S. H., and L. A. Segel, (1968) Effccts o f surface curvature and property
variation on cellular c o n v e c t i o n , Fluids, 11, No. 3 , 4 7 0 —476.
Davydov, B. I., (1935) Diffusion equation with allowance for molecular velocity,
D oklady A N SS S R , 2, No. 7, 4 7 4 - 4 7 8 ; (1958) Phenomenologic equations
o f statistical dynamics o f an incompressible turbulent fluid, Z/i. Eksp. Teor.
Fiz., 35, No. 2(8), 5 2 7 - 5 2 9 ; (1959a) Statistical dynamics o f an incompres
sible turbulent fluid, D oklady A N S S S R , 127, No. 4, 7 6 8 - 7 7 1 ; (1959b)
Statistical theory o f turbulence, D oklady A N SS S R , 127, No. 5 , 9 8 0 - 9 8 2 ;
(1961) Statistical dynamics o f an incompressible turbulent ^\x\d, D oklady
A N S SS R , 136, No. 1 , 4 7 - 5 0 .
Deacon, E. L., (1949) Vertical diffusion in the lowest layers o f the atmosphere,
Quart. J. R o y. M eteor. Soc., 75, No. 323, 8 9 - 1 0 3 ; (1955) The turbulent
transfer o f m o m entum in the lowest layers o f the atmosphere, C. S. I. R. O.,
Div. M eteor. Phys., Techn. Pap. No. 4, Melbourne; (1959) The measure
ments o f turbulent transfer in the lower atmosphere. A dv. G eophys., 6
(Atmospheric diffusion and air pollution), 2 1 1 - 2 2 8 ; (1962) Aerodynamic
roughness o f the sea, 7. G eophys. Res., 67, No. 8, 3 1 6 7 - 3 1 7 2 .
Deacon, E. L., and E. K. Webb, (1962) Interchange o f properties between sea
and air. Small-scale interaction. The Sea (ed. by M. N. Hill), Vol. 1,
Interscience Publ., N. Y .- L o n d o n , 4 3 - 8 7 .
Deardorff, J. W., (1963) On the stability o f viscous plane Couette flo w ,/. Fluid
M ech., 15, No. 4, 6 2 3 - 6 3 1 ; (1964) A numerical study o f two-dimensional
parallel-plate convection, J. A tm o sp h . Sci., 21, No. 5, 419 - 4 3 8 ; (1968)
Dependence o f air-sea transfer coefficients on bulk stability, J. Geophys.
R es., 73, No. 8, 2 5 4 9 - 2 5 5 7 .
Deardorff, J. W., and G. E. Willis, (1967) The free-convection temperature
profile. Quart. J. R o y. M eteorol. Soc., 93, No. 396, 1 6 6 - 1 7 5 .
Debler, W. R., (1966) On the analogy between thermal and rotational hydro-
dynamic s ta b ility ,/. F luid M ech., 24, No. 1, 1 6 5 - 1 7 6 .
Defant, A., (1921) Die Zirkulation der Atmosphare in den gemassigten
Breiten der Erde, Geografiska Annaler, 3, 2 0 9 —265.
Deissler, R. G., (1951) Investigation of turbulent flow and heat transfer in
sm ooth tubes, including the effects o f variable fluid properties, Trans.
A m er. Soc. M ech. Engrs., 73, No. 1, 1 0 1 - 1 0 7 ; (1954) Heat transfer and
fluid friction for fully developed turbulent flow o f air and supercritical
water with variable fluid properties, Trans. A m er. Soc. Mech. Engrs., 76,
No. 1, 7 3 - 8 6 ; (1955) Analysis o f turbulent heat transfer, mass transfer and
friction in sm ooth tubes at high PrandU and Schmidt numbers, Nat. Adv.
Com, A ero n a u t., Rep. 1 2 1 0 ;(1 9 5 9 ) Convective heat transfer and friction in
flow o f liquids. T urbulent F low and H eat Transfer (ed. by C. C. Lin),
Princeton Univ. Press, 2 8 8 - 3 3 8 .
BIBLIOGRAPHY 705
Dhawan, S., (1952) Direct measurements o f skin friction, Nat. A dv. Com.
A eronaut., Tech. Note, No. 2567.
Dikiy, L. A., (1960a) Stability o f plane-parallel flows o f an inhomogeneous
fluid, P rikl M atem . M ekh., 24, No. 2, 2 4 9 - 2 5 7 ; (19 6 0 b ) Stability o f
plane-parallel flows o f an ideal fluid, D oklady A N S S S R , 135, No. 5,
1 0 6 8 - 1 0 7 1 ; (1960c) The roots o f the Whittaker functions and the
Macdonald functions with a complex index, Izvestiya A N S S S R , Ser.
matem ., 24, No. 6, 9 4 3 - 9 5 4 ; (1964) Stability o f plane-parallel Couette
flow, Prikl. M atem . M ekh., 28, No. 2, 3 8 9 - 3 9 2 ; (1965a) Contribution to
the nonlinear theory o f hydrodynam ic stability, Prikl. Mat. i. M ekh., 29,
No. 5, 8 5 2 - 8 5 5 ; (1965b) On nonlinear theory o f stability o f zonal flows,
Izvestiya A N S S S R , Ser. fiz. atmosf. i okeana, 1, No. 1 1 , 1 1 1 7 - 1 1 2 2 .
Dil’man, V. V., (1967) On the theory o f heat- and mass-transfer in turbulent
flow, Teor. O snovy K him . Tekhn., 1, No. 4, 4 3 8 - 4 4 5 .
Dipprey, D. F., and R. H. Sabersky, (1963) Heat and m om en tu m transfer in
sm ooth and rough tubes at various Prandtl numbers, Int. J. H eat and Mass
Transfer, 6, No. 5, 3 2 9 - 3 5 3 .
Di Prima, R. C., (1955) Application o f the Galerkin m ethod to the calculation o f
the stability of curved flows. Quart. A ppl. M ath., 13, No. 1, 5 5 - 6 2 ; (1961)
Stability of nonrotationally symmetric disturbances for viscous flow
between rotating cylinders, Phys. Fluids, 4, No. 6, 7 5 1 - 7 5 5 ; (1963)
Stability o f curved flo w s,/. A ppl. M ech., 3 0 , 4 8 6 - 4 9 2 .
Donnelly, R. J., (1962) Experimental determ ination o f stabihty limits, Proc.
S ym p . A ppl. M ath., 13 (Hydrodynam ic instabiUty), 4 1 - 5 3 ; (1963) Experi
mental confirmation o f the Landau law in Couette flow , Phys. Rev. Letters,
10, No. 7 , 2 8 2 - 2 8 4 .
Donnelly, R. J., and D. Fultz, (1960) Experiments of the stability o f viscous
flow between rotating cylinders, II. Visual observations, Proc. R o y. Soc.,
A 2 5 8 ,N o . 1292, 1 0 1 - 1 2 3 .
Donnelly, R. J., and K. W. Schwarz, (1963) The approach to equilibrium in
nonlinear hydrodynamics, Phys. Letters, 5, No. 5, 3 2 2 - 3 2 4 ; (1965)
Experiments on the stability o f viscous flow between rotating cylinders. IV.
Finite-amplitude experiments (with an appendix by P. H. Roberts), Proc.
R o y. Soc., A283, No. 1395, 5 3 1 - 5 5 6 .
Donovan, L. F., O. T. Hanna, and S. Yerazunis, (1967) Similar solutions of
turbulent boundary layer heat and mass transfer problems, Chem. Engrg.
Sci., 22, No. 4, 5 9 5 - 6 1 0 .
Doob, J. L., (1953) Stochastic Processes, Jo h n Wiley and Sons, N. Y.
Dorodnitsyn, A. A., (1941) Theory o f diurnal temperature variation in the
mixing layer, D oklady A N S S S R , 30, No. 5, 4 1 0 - 4 1 3 .
Drazin, P. G., (1958) The stability o f a shear layer in an unbounded
heterogeneous fluid, J. Fluid Mech., 4, No. 2, 2 1 4 - 2 2 4 ; (1962) On the
stability o f parallel flow o f an incompressible fluid o f variable density
and viscosity,/>oc. Cambr. Phil. Soc., 58, No. 4, 6 4 6 - 6 6 1 .
70 6 BIBLIOGRAPHY
Drazin, P. G., and L. N. Howard, (1962) The instability to long waves o f

u nbounded parallel inviscid flow, J. F luid Mech., 14, No. 2, 2 5 7 - 2 8 3 ;
(1966) H ydrodynamic stability o f parallel flows o f inviscid fluid, Advan.
A ppl. Mech., 9, 1 - 8 9 .
Drimmel, J., and H. Reuter, (1960) Der Exponent n in S u tto n ’s Ansatz fiir die
Lagrangesche K o r r e l a t i o n s f u n k t i o n , M eteor., Geophys. un d B ioklim .,
A 1 2 ,N o . 1 , 2 4 - 3 9 .
Dry den, H. L., (1959) Transition from laminar to turbulent flow. T urbulent
F low and H eat Transfer (C. C. Lin, ed.), Princeton University Press,
Princeton, N. J., 3 —74.
Dryden, H. L., G. B. Schubauer, W. C. Mock, and H. K. Skramstad, (1937)
Measurements o f intensity and scale o f wind-tunnel turbulence and their
relation to the critical Reynolds num ber o f spheres, Nat. Adv. Com.
A erona ut., Rep. No. 581.
Dumas, R., and J. Marcillat, (1966) Repartition en amplitudes de la vitesse dans
une couche limite turbulente, C om pte R end. Acad. Sci., 262, No. 12,
A 7 0 0 -A 7 0 3 .
D uty, R. L., and W. H. Reid, (1964) On the stability o f viscous flow between
rotating c y l in d e r s ,/ F luid M ech., 20, No. 1, 8 1 - 9 4 .
Dvorak, F. A., and M. R. Head, (1967) Heat transfer in the constant property
turbulent boundary layer, Int. J. H eat Mass Transfer, 10, No. 1 , 6 1 —81.
Dwyer, O. E., (1963) Eddy transport in liquid-metal heat transfer, A .I.C h E .
Journal, 9 , ^ 0 . 2 , 2 6 1 - 2 6 ^ .
Dyer, A. J., (1958) An improved electromagnetic integrator, J. Scient. Instr., 35,
No. 7, 240; (1960) Heat transport anem om eter o f high stability, J. Scient.
Instr., 37, No. 5, 1 6 6 - 1 6 9 ; (1961) Measurements o f evaporation and heat
transfer in the lower atmosphere by an automatic eddy-correlation
technique. Quart. J. R o y Meteor. Soc., 87, No. 3 7 3 , 4 0 1 - 4 1 2 ; (1965) The
flux-gradient relation for turbulent heat transfer in the lower atmosphere.
Quart. H. R o y . M eteor. Soc., 91, No. 388 151 - 1 5 7 ; (1967) The turbulence
transport o f heat and water vapour in an unstable atmosphere, Quart. J.
R o y. M eteor. Soc., 93, No. 3 9 8 , 5 0 1 - 5 0 8 ;( 1 9 6 8 ) An evaluation o f eddy flux
variation in the atmospheric boundary layer, J. Appl. M eteor., 7, No. 5,
845-850.
Dyer, A. J., B. B. Hicks, and K. M. King, (1967) The Fluxatron-A revised
approach to the m easurem ent o f eddy fluxes in the lower a tm o sp h e re ,/.
A ppl. M eteor., 6, No. 2 , 4 0 8 —413.
Dyer, A. Y., and F. Y. Maher, (1965a) The “ E vapotron.” An instrum ent for the
the measurem ent o f eddy fluxes in the lower atmosphere, C .S.I.R.O ., Div.
M eteor. P hys., Techn. Paper No. 15; (1965b) Automatic eddy-flux
m easurem ent with the Evapotron, / A ppl. M eteor., 4, No. 5, 6 2 2 —625.
Dynkin, E. B., (1965) M arkov Processes, Vols. 1 - 2 , Springer-Verlag, Berlin-
Heidelberg-New York.
Dyson, F. J., (1960) Stability o f an idealised atmosphere, II. Zeros o f the
confluent hypergeometric f u n c t i o n , Fluids, 3, No. 2, 1 5 5 - 1 5 8 .
BIBLIOGRAPHY 707
Eckert, E. R. G., and R. M. T>xdkQ,{\9 59) H eat and Mass Transfer, McGraw-Hill
Book Co., New York.
Eckhaus, W., (1965) Studies in Non-linear Stability Theory, Springer-Verlag,
Berlin-Heidelberg-New York.
Eiffel, G., (1912) Sur la resistance des spheres dans I’air en m ouvement, C om pte
R end. Acad. Sci., 155, 1 5 9 7 - 1 5 9 9 .
Einstein, H. A., and H. Li, (1956) The viscous sublayer along a sm ooth
boundary, Proc. A m er. Soc. Civ. Engrs., Paper 945.
Ekman, V. W., (1911) On the change from steady to turbulent m otion o f
liquids, Mat. A stron. Fys., 6, No. 12, 1 - 1 6 .
Elagina, L. G., (1962) Optical device for measuring turbulent moisture-content
f l u c t u a t i o n s , S e r . geofiz.,N o. 8, 1100—1107.
Elder, J. W., (1959) The dispersion o f marked fluid in turbulent shear flow, J.
F luid M ech., 5, No. 4, 54 4 —560; (1960) An experimental investigation o f
turbulence spots and breakdown to turbulence, J. Fluid M ech., 9, No. 2,
235-249.
Ehas, P., (1929) Die Warmeiibertragung einer geheizten Platte an stromende
Luft. I. Anlage und Ergebnisse der Versuche, Zs. angew. M ath. M ech., 9, No.
6 ,4 3 4 -4 6 5 .
Eliassen, A.. \i. H0iland, and E. Riis, (1953) Two-dimensional perturbation o f
flow with constant shear o f a stratified fluid, Publ. Inst. Weath. Clim. Res.,
Oslo, No. 1.
Elliott, W. P., (1957) A comparison o f some approaches to the diabatic wind
profile, Trans. Am er. Geophys. Union, 38, No. i‘, 2 1 - 2 4 ; (1960) A
hypothesis for the diabatic mixing length, J. M eteor., 17, No. 6 , 6 8 0 - 6 8 1 .
Ellison, T. H., (1956) Atmospheric turbulence. Surveys in M echanics (ed. by G.
K. Batchelor and R. M. Davies), Cambridge Uqiv. Press, 4 0 0 - 4 3 0 ; (1957)
Turbulent transfer o f heat and m o m entum from an infinite rough p l a n e , /
Fluid M ech., 2, No. 5, 4 5 6 - 4 6 6 ; (1959) Turbulent diffusion, 5 d . Progress,
47, No. 187, 4 9 5 - 5 0 6 ; (1960) A note on the velocity profile and
longitudinal mixing in a broad open channel, J. Fluid M ech., 8, No. 1,
33-40.
Ellison, T. H., and J. S. Turner, (1960) Mixing o f dense fluid in a turbulent pipe
flow. Parts 1 - 2 ,7 . F luid M ech., 8, No. 4, 5 1 4 - 5 4 4 .
Elrick, D. E., (1962) Source functions for diffusion in uniform shear flow,
Australian J. Phys., 15, No. 3, 2 8 3 - 2 8 8 .
Elrod, H. G., (1957) Note on the turbulent shear stress near a w all,/. A eronaut.
Sci., 24, No. 6, 4 6 8 - 4 6 9 (see also the am endm ent to this work: J.
A eronaut. Sci., 27, No. 2, 145, 1960).
Esch, R. E., (1957) The instability o f shear layer between two parallel streams,
J. Fluid Mech., 3, No. 3, 2 8 9 - 3 0 3 .
Escudier, M. P., and W. B. Nicoll, (1966) The entrainm ent function in turbulent
boundary-layer and wall-jet calculations, J. F luid M ech., 25, No. 2,
337-366.
708 BIBLIOGRAPHY
Page, A., and H. C. H. Tow nend, (1932) An examination o f turbulent flow with
an ultramicroscope, Proc. R o y. Soc., A 135, No. 828, 6 5 6 - 6 7 7 .
Falkner, V. M., (1943) The resistance o f a sm ooth flat plate witH turbulent
boundary layer, A in r . Eng., 15, 6 5 - 6 9 .
Favre, A., J. Gaviglio, and R. Dumas, (1953) Quelques mesures de correlation
dans le temps et I’espace en souffleiie. La R ech. A ero ., No. 32, 2 1 - 2 8 .
Fichtl, G. H., (1968) Characteristics o f turbulence observed at the NASA 150-m
meteorological to w e r ,/. A ppl. M eteor., 7, No. 5, 8 3 8 - 8 4 4 .
Fidman, B. A., (1959) New data on the viscous sublayer o f the turbulent flow,
in coll.: Hydraulics o f th e Structures and D ynam ics o f R iver Beds, Acad.
Sci. Press, USSR, Moscow, 1 8 3 -1 8 8 .
Fiedler, H., and M. R. Head, (1966) Interm ittency measurements in the
turbulent boundary layer, / F luidM ech., 25, No. 4, 7 1 9 - 7 3 5 .
F j0rtoft, R., (1950) Application o f integral theorems in deriving criteria of
stability o f laminar flow and for the barociinic circular vortex, Geofys.
P u b l, 17, No. 6, 1 - 5 2 .
Fleischman, B. A., and F. N. Frenkiel, (1954) Diffusion o f matter em itted from a
line source in a non-isotropic turbulent flow, J. M eteor., 12, No. 2,
141-145.
F o ck ,V . A., (1926) Solution o f a problem in the theory o f diffusion by the finite
difference m ethod and its application to the diffusion o f light, Trudy Gos.
Optich. In-ta, 4, No. 34.
Forstall, W., and E. W. Gaylord, (1955) M om entum and mass transfer in a
submerged water j e t , / A ppl. M ech., 22, No. 2, 1 6 1 - 1 6 4 .
Fox, J. A., M. Lessen, and W. V. Bhat, (1968) Experimental investigation o f the
stability of Hagen-Poiseuille Oow,P/2V5. Fluids, 11, No. 1, 1 - 4 .
Frank-Kamenetskiy, D. A., (1947) D iffusion and Heat Fxchange in Chemical
Kinetics, Acad. Sci. Press, USSR, Moscow-Leningrad (English translation:
Princeton Univ. Press, 1955).
Frenkiel, F. N., (1952) Application o f the statistical theory o f turbulent
diffusion to micrometeorology, / M eteor., 9, No. 4, 2 5 2 - 2 5 9 ; (1953)
Turbulent diffusion: Mean concentration distribution in a flow field o f
homogeneous turbulence, A ppl. M ech., 3, 6 1 - 1 0 7 .
Frenkiel, F. N., and P. A. Sheppard [ e d .] , (1959) Atmospheric diffusion and air
pollution, in G eophys., 6, Academic Press, N. Y.-London.
Freym uth, P., (1966) On transition in a separated laminar boundary l a y e r , /
F luid M ech., 25, No. 4, 6 8 3 - 7 0 4 .
Fritsch, W., (1928) Einfluss der Wandrauhigkeit au f die turbulente Gesch-
windigkeitsverteilung in Rinnen, Zs. angew. Math. M ech., 8, No. 2,
199-216.
From m , J. E., (19 65 ) Numerical solutions o f the non-linear equations for a
heated fluid layer, Fluids, 8, No. 10, 1 7 5 7 - 1 7 6 9 .
Frost, R., (1946) Turbulence and diffusion in the lower atmosphere, Proc. R oy.
Soc., A 186, No. 1004, 2 0 - 3 5 ; (1948) Atmospheric turbulence. Quart. J.
R o y . Meteor. Soc., 74, No. 321, 3 1 6 - 3 3 8 .
BIBLIOG RA PHY 709
Funk, J. P., (1960) Measured radiative flux divergence near the ground at night,
Quart. J. R o y. M eteor. Soc., 86, No. 369, 3 8 2 - 3 8 9 .
Furuya, Y., and H. Fujita, (1967) Effect o f surface roughness on the velocity
defect \ 2LyN,Phys. Fluids, 10, Supplem ent, S I 5 5 —S I 56.
Gallagher, A. P., and A. McD. Mercer, (1962) On the behavior of small
disturbances in plane Couette flow, J. F luid M ech., 13, No. 1, 9 1 - 1 0 0 ;
(1964) On the behavior o f small disturbances in plane Couette flow. Part 2.
The higher eigenvalues,/. F luidM ech., 18, No. 3, 3 5 0 —352..
Caster, M. A., (1962) A note on the relation between temporally increasing and
spatially increasing disturbances in hydrodynam ic stability, J. F luid Mech.,
14, No. 2, 2 2 2 —224; (1965) On the generation o f spatially growing waves in
a boundary la y e r,/. F luid M ech., 22, No. 3, 4 3 3 —441.
Gee, J. H., (1966) An approxim ate treatm ent o f the effect o f thermal stability
on turbulent diffusion, Quart. J. R o y . Meteor. Soc., 92, No. 392, 3 0 1 - 3 0 5 ;
(1967) A note on the effect o f directional wind shear on medium scale
atmospheric diffusion, Quart. J. R o y . M eteor. Soc., 93, No. 396, 2 3 7 - 2 4 1 .
Gee, J. H., and D. R. Davies, (1963) A note on horizontal dispersion from an
instantaneous ground source. Quart. J. R o y. Meteor. Soc., 89, No. 382,
5 4 2 - 5 4 5 ; (1964) A further note on horizontal dispersion from an
instantaneous ground source. Quart. J. R o y. M eteor. Soc., 90, No. 386,
478-480.
Gel’fand, 1. M., and N. Ya. Vilenkin, (1 9 6 4 ) Generalized F unctions, Vol. 4;
A pplications o f H arm onic A nalysis, Academic Press, N. Y.
Gerber, R., (1949) Sur la reduction a un principe variationnel des equations du
m ouvem ent d ’un fluide visqueux incompressible, A nn. Inst. Fourier, 1,
1 57-162.
Gibson, C. H., C. C. Chen, and S. C. Lin, (1968) Measurements o f turbulent
velocity and tem perature fluctuations in the wake o f a sphere, i4./.i4.i4./, 6,
No. 4, 6 4 2 - 6 4 9 .
Gifford, F. A., (1962) Diffusion in the diabatic surface la y e r,/. G eophys. Res.,
67, No. 8 , 3 2 0 7 - 3 2 1 2 .
Gill, A. E., (1965a) On the behaviour o f small disturbances to Poiseuille flow in
a circular pipe, /. F luid M ech., 21, No. 1, 1 4 5 - 1 7 2 ;( 1 9 6 5 b ) A mechanism
for instability o f plane C ouette flow and o f Poiseuille flow in a pipe, /.
F luid M ech., 21, No. 3, 5 0 3 - 5 1 1 .
Gille, J., (1967) Interferom etric measurem ent o f tem perature gradient reversal in
a layer o f convecting a i r ,/. F luid M ech., 30, No. 2, 3 7 1 - 3 8 4 .
G nedenko, B. V., (1962) The Theory o f Probability, Chelsea Publishing Co.,
N. Y.
G ddeke, K., (1935) Messungen der atmospharischen Turbulenz in Bodennahe
mit einer H itzdrahtm ethode, A nn, H y d r o g r .,^ o . 10, 4 0 0 - 4 1 0 .
Goering, H., (1959) Mathematische Untersuchungen zum Umschlag laminar-
turbulent, Afor/iem. N achrichten, 17, No. 3 - 6 , 3 5 8 - 3 9 3 .
Gortler, H., (1942) Berechnung von Aufgaben der freien Turbulenz au f Grund
eines neuen N ^ e ru n g san satzes, Zs. angew. Math. M ech., 22, No. 5,
710 BIBLIOGRA PHY
2 4 4 - 2 5 4 ; (1959) Uber eine Analogic zwischen der Instabilitaten laminarer

Grenzschichtstromungen an konkaven Wanden und an erw arm ten Wanden.
In g .A rc h ., 28, Nr. 1 , 7 1 - 7 8 .
Gortler, H., and W. Velte, (1967) Recent m athem atical treatm ents o f laminar
flow and transition problems, Fluids, 10, Supplem ent, S 3 -S 1 0 .
Goldstein, S., (1931) On the stability o f superposed streams o f fluids of
different densities, Proc. R o y , Soc., A 132, No. 820, 5 2 4 - 5 3 9 ; (1938)
M odern D evelopm ents in F luid D ynam ics, Vols. 1—2, O xford Univ. Press;
(1951) On diffusion by discontinuous m ovem ent and on the telegraph
equation. Quart. J. Mech, A p p l M ath., 4, No. 2, 129—156.
G or’kov, L. P., (1957) Stationary convection in a plane layer o f fluid near
conditions o f critical heat transfer, Z K Eksp. Teor. Fiz., 33, No. 2(8),
402-411.
Gosse, J., (19 6 1 ) Sur I’ecoulem ent turbulent des fluides, C om pte R end. Acad.
Sci., 253, No. 1 , 6 1 - 6 3 .
G otoh, K., (1968) The equilibrium state o f the finite disturbance in free flows,
J. Phys. Soc. Japan, 24, No. 5, 1 1 3 7 - 1 1 4 6 .
Graebel, W. P., (1966) On determ ination o f the characteristic equations for the
stability o f parallel flo w s,/. F luid M ech., 24, No. 3 , 4 9 7 —508.
Grant, A. M., (1957) Corrected mixing-length theory o f tu rbulent diffusion, 7.
M eteor., 14, No. 4, 2 9 7 - 3 0 3 .
Greenspan, H. P., and D. J. Benney, (1963) On shear-layer instability,
breakdow n, and tra n sitio n ,/. F luidM ech., 15, No. 1, 1 3 3 - 1 5 3 .
Grenander, U., (1950) Stochastic processes and statistical inference, Mat.,
l , N o . 3 , 195-277.
Grober, H., and S. Erk, (1955) Die Grundgesetze der Warmeiibertragung,
Springer, Berlin-Gottingen-Heidelberg.
Grohne, D., (1954) lib er das Spektrum bei Eigenschwingungen ebener Lamin-
arstromungen, Zs. Angew . M ath. M ech., 34, No. 3, 3 4 4 - 3 5 7 .
Grosch, C. E., and H. Salwen, (1 96 8 ) The stability o f steady and time-dependent
plane Poiseuille flo w ,/. Fluid M ech., 34, No. 1, 1 7 7 - 2 0 5 .
Gruza, G. V., (1961) Macro turbulence in General A tm ospheric Circulation,
G idrom eteoizdat Press, Leningrad.
G ukhm an, A. A., and B. A. Kader, (1969) Mass transfer from a wall o f a tube to
turb u len t fluid flow at high Schmidt numbers, Teor. O snovy Khim .
T e k h n o l, 3, No. 2, 2 1 6 - 2 2 4 .
G upta, H. C., (1959) Diffusion by continuous m o v e m e n ts,/. Math. Phys., 38,
No. 1 , 3 6 - 4 1 .
Gurvich, A. S., (1959) Acoustic m icroanem om eter for investigating the
microstructure o f turbulence, A kustich. Zhurnal, 5, No. 3, 3 6 8 - 3 6 9 ;
(1960) Experimental investigation o f frequency spectra and probability
distribution functions for the vertical com ponent o f wind velocity, Izvestiya
A N S S S R , Ser. geofiz.. No. 7, 1 0 4 2 - 1 0 5 5 ; (1961) Measurements o f the
stress in the atmospheric surface layer, Izvestiya A N S S S R ,S er. geofiz.. No.
BIBLIOGRAPHY 711
3, 4 5 8 - 4 6 6 ; (1962) T urbulent m om entu m transfer during unstable stratifi

cation and under near neutral conditions, Collection: Atmospheric tu rb u
lence, T rudy Inst, Fiz. A tm o s f A N S S S R , No. 4, 8 1 - 1 0 0 ; (1965) Vertical
wind-velocity and tem perature profiles in the atmospheric surface layer,
Izvestiya A N S S S R , Ser. Fizika atmosfery i okeana, 1, No. 1.
Hagen, G., (1839) Uber die Bewegung des Wassers in engen zylindrischen
Rohren, Pogg. A n n ., 46, 4 2 3 —442.
Hains, F. D., (1967) Stability o f plane Couette-Poiseulle flov/, Phys. Fluids, 10,
No. 9, 2 0 7 9 - 2 0 8 0 .
Halstead, M., (19 4 3 ) A stabiUty term in the wind-gradient equation, Trans.
A m er. Geophys. Union, 24, No. 1, 2 0 4 —208; (1952) The relationship
between wind structure and turbulence near the ground, G eophys. Res.
Pap., No. 1 9 , 9 7 - 1 2 6 .
Haltiner, G. J., and F. L. Martin, (1957) D ynam ical and Physical M eteorology,
McGraw-Hill Book Co., New Y ork-Toronto-London.
Hama, F. R., (1953) On the velocity distribution in the laminar sublayer and
transition region in turbulent shear flows, J. A eronaut. Sci., 20, No. 9,
6 4 8 —650; (19 5 4 ) Boundary-layer characteristics for sm ooth and rough
surfaces, Trans. Soc. Naval A rch. Marine Engrs., 62, 3 3 3 - 3 5 8 .
Hama, F. R., J. D. Long, and J. C. Hegarty, (1 9 57 ) On transition from laminar to
turbulent flo w ,/. A ppl. Phys., 28, No. 4, 3 8 8 - 3 9 4 .
Hama, F. R., and J. N u tan t, (1963) Detailed flow-fleld observations in the
transition process in a thick boundary layer, Proc. H eat Transfer F luid
Mech. Inst., Stanford Univ. Press, 11-93.
Hamel, G., (1911) Z um T urbulenzproblem , Nachr. Ges. Wiss. G ottingen,
Math.-Phys. Kl., 2 6 1 - 2 7 0 .
Hamilton, R. M., (1963) Solid-liquid Mass Transfer in T urbulent Pipe F lo w , Ph.
D. Thesis, Cornell University, Ithaca, N. Y.
H anratty, T. J., (1967) Study o f turbulence close to a solid w a ll,P / 2VS. Fluids,
10, Supplem ent, S126—S133.
Hansen, M., (1928) Die Geschwindigkeitsverteilung in der Grenzschicht an einer
eingetauchten Platte, Z. angew. M ath. M ech., 8, No. 2, 185—199.
Harlow, F. H., and P. I. Nakayama, (1967) Turbulence transport equations,
Phys. Fluids, 10, No. 1 1 ,2 3 2 3 - 2 3 3 2 .
H arriott, P., and R. M. Hamilton, (1965) Solid-liquid mass transfer in turbulent
pipe flow, Chem. Engng. Sci., 20, No. 1 2 , 1 0 7 3 - 1 0 7 8 .
H atton, A. P., (1964) Heat transfer through the tu rbulent incompressible
boundary layer on a flat plate. Int. J. H eat and Mass Transfer, 7, 8 7 5 —890.
Haugen, D. A., M. L. Barad, and P. Antanaitis, (1961) Values o f parameters
appearing in S u tto n ’s diffusion models, 7. M ete o ro l, 18, No. 3, 36 8 —372.
Haugen, D. A., (1966) Some Lagrangian properties o f turbulence deduced from
atmospheric diffusion ex p e rim e n ts,/. A ppl. M eteor., 5, No. 5, 6 4 6 —652.
H aupt, O., (1912) Uber die Entwicklung einer willkiirlichen F unk tio n nach den
Eigenfunktionen des Turbulenzproblems, Sitzungsberichte d.K. Bayer.
Akad. Wiss., 2 8 9 - 3 0 1 .
71 2 B IBLIOG RA PHY
Hay, J. S., and F. Pasquill, (1959) Diffusion from a continuous source in relation
to the spectrum and scale o f turbulence, Adv. G eophys., 6 (Atmospheric
diffusion and air pollution), 3 4 5 - 3 6 5 .
Heisenberg, W., (1924) Uber Stabilitat und Turbulenz von Fliissigkeitsstromen,
A nn. P hys.(4), 74, 5 7 7 - 6 2 7 .
Helmholtz, H., (1868) Uber discontinuierliche Flussigkeitsbewegungen,M6>«ars-
ber Kon. A kad. Wiss., Berlin, 23, 2 1 5 - 2 2 8 .
Herring, J. R., (1963) Investigation o f problems in thermal convection, J.
A tm o sp h . ScL, 20, No. 4, 3 2 5 - 3 3 8 . (1964) Investigation o f problems in
thermal convection: Rigid boundaries, J. A tm o sp h . ScL, 21, No. 3,
2 7 7 - 2 9 0 ; (1966) Some analytic results in the theory o f thermal convection,
J. A tm o sp h . ScL, 23, No. 6, 6 1 2 - 6 1 1 .
Hess, G. D., and H. A. Panfosky, (1966) The budget o f turbulent energy near the
ground, Quart. J. R o y. Meteor. Soc., 92, No. 392, 2 7 7 - 2 8 0 .
Hidy, G. M., and E. J. Plate, (1966) Wind action on water standing in a
laboratory ch a n n el,/. F luidM ech., 26, No. 4, 6 5 1 - 6 8 8 .
H0iland, E., (1953) On two-dimensional perturbation o f linear flow, Geofys.
P u b l, 18, No. 9, 1 - 1 2 .
Hino, M., (1966) A theory on the fetch graph, the roughness of the sea and the
energy transfer between wind and wave,Prac. Tenth C onference on Coastal
Eng.
Hinze, J. O., (1951) T urbulent diffusion from a source in turbulent shear flow,
J. Aeronaut. ScL, 18, No. 8, 5 6 5 - 5 6 6 ; (19 5 9) Turbulence. A n In trodu ction
to Its M echanism and Theory, McGraw-Hill Book Co., N. Y.; (1962)
T urbulent pipe-flow, d e la turbulence (Coll. Intern, du CNRS a
Marseille), Paris, Ed. CNRS, 1 3 0 - 1 6 5 .
Hinze, J. 0 ., and B. G. Van der Hegge Zijnen, (1949) Transfer o f heat and
m atter in the turbulent mixing zone o f an axially symmetric je t, >lpp/. Sci.
Res., A l , 4 3 5 - 4 6 1 ;(1 9 5 1 ) Local transfer o f heat in anisotropic turbulence,
Inst. Mech. Eng., ?xoc. Gener. Discus, on Heat Transfer, 1 8 8 - 1 9 7 .
Hirschfelder, J. O., C. F. Curtiss, and R. B. Bird, {\9 S A ) M olecular Theory o f
Gases and L iquids, John Wiley and Sons, N. Y.; Chapman and Hall, London.
Hishida, M., (1967) T urbulent heat transfer and tem peratrue distribution in the
thermal entrance region o f a circular pipe. Bull. Jap. Soc. M ech. Eng., 10,
No. 37, 1 1 3 - 1 2 3 .
Hocking, L. M., (1968) Long wavelength disturbances to non-planar parallel
flo w ,/. F/w/dMec/z., 3 1 , N o . 4 6 2 5 - 6 3 4 .
Hoeber, H., (1968) Einige Ergebnisse von Windprofilmessungen iiber See
bei stabiler Schichtung, Tellus, 20, No. 3, 4 9 5 - 501.
Hogstrom, U., (1964) An experimental study o f atmospheric diffusion, Tellus,
16, No. 2, 2 0 5 - 2 5 1 ; (1 967a) A new sensitive eddy flux instrum entation,
Tellus, 19, No. 2, 2 3 0 - 2 3 9 ; (1 9 6 7 b ) Turbulent water-vapor transfer at
different stability conditions, Fluids, 10, Supplem ent, S 2 4 7 -S 2 5 4 .
Hofm ann, E., (1937) Der Warmeiibergang bei der Strom ung im Rohr., Zs.
BIBLIOGRA PHY 713
Kdlteind., 44, 99 107; (1940) Uber die Gesetzmassigkeiten der Warmc- iiiid
Stoffubertragung auf Grund des Stromungsverganges im Rohr, Forsch.
G ebiete Ingenieurwesens, A l l , 159 - 1 6 9 .
Holland, J. Z., (1967) Joint density functions o f turbulent variables in the
atmospheric boundary layer, Phys. Fluids, 10, Supplem ent, S 2 2 0 -S 2 2 2 ;
(196 8 ) An application o f some statistical techniques to the study o f eddy
structure. R eport TID-24585 (Meteorology T lD -4500), U. S. Atomic
Energy Commission, Washington, D. C.
Holzman, B., (1943) The influence of stability on e v a p o r a t i o n , / I N e w York
Acad. 5c/., 44, 1 3 - 1 9 .
Hopf, E., (1952) Statistical hydromechanics and functional calculus, J. Rat.
M ech .A n a l., 1, No. 1, 8 7 - 1 2 3 .
Howard, L. N., (1959) H ydrodynamic stability o f a jet, J. M ath, and Phys., 39,
No. 4, 2 8 3 - 2 9 8 ; (1961) Note on a paper o f Jo h n W. M il e s ,/ Fluid M ech.,
10, No. 4, 5 0 9 - 5 1 2 ; (1963a) Neutral curves and stability boundaries in
stratified f lo w ,/. F luidM ech., 16, No. 3, 3 3 3 - 3 4 2 ; (1963b) Heat transport
by turbulent c o n v e c t io n ,/ F luidM ech., 17, No. 3. 4 0 5 —432.
Howarth, L., [ e d .] , (1953) Modern developments in fluid dynamics. High Speed
Flow, Oxford University Press.
Hsu, N. T., K. Sato, and B. H. Sage, (1956) Tem perature gradients in turbulent
gas streams. Effect of flow condition upon eddy conductivity, Ind. Eng.
C/zem., 4 8 ,N o . 1 2 , 2 2 1 8 - 2 2 2 3 .
Huang, H., and J. E. Cermak, (1967) Similarity o f thermally stratified shear
flows in the laboratory and atmosphere, Phys. Fluids, 10, Supplem ent,
S 2 5 5 -S 2 5 8 .
Hubbard, D. W., (1964) Mass Transfer in Turbulent Flow at High S c h m id t
N um bers,?\v. D. Thesis, The University o f Wisconsin, Madison.
Hubbard, D. W., and E. N. Lightfoot, (1966) Correlation o f heat and mass
transfer data for high Schmidt and Reynolds numbers, Industr. and Engng.
Chem. F undam ent., 5, No. 3, 3 7 0 - 3 7 9 .
H udim oto, B., (1941) On the turbulent boundary layer, Tra/i5. Japan Soc. Mech.
Eng., 7; (1951) A brief note on the laminar sublayer o f the turbulent
boundary layer, Mem. Fac. Eng. K y o to Univ., 13, 1 7 4 - 1 7 9 .
Hughes, T. H., and W. H. Reid, (1965) The stability o f laminar boundary layers
at separation, / Fluid M ech., 23, No. 4, 7 3 7 - 7 4 7 .
Hunt, J. N., (1954) The turbulent transport o f suspended sediment in open
channels,Pm c. R o y. Soc., A 2 2 4 ,N o . 1 1 5 8 ,3 2 2 - 3 3 5 .
Huppert, H. E., (1968) Numerical evaluation of the Tietjens function, Phys.
Fluids, 11, No. 3 , 6 7 3 - 6 7 4 .
Hurle, D. T., E. Jakem an, and E. R. Pike, (1967) On the solution o f the Benard
problem with boundaries of finite co n d u c tio n ity , Proc. R oy. Soc., A296,
No. 1 4 4 7 , 4 6 9 - 4 7 5 .
Hwang, N. H. C., and L. V. Baldwin, (1966) Decay o f turbulence in
axisymmetric wakes, / Basic Engineering, 88, No. 1, 2 6 1 - 2 6 8 .
714 BIBLIOGRA PHY
Inoue, E., (1950)—(1951) On the turbulent diffusion in the atmosphere, I - I l . J.

M eteor. Soc. Japan, 28, No. 12, 4 4 1 - 4 5 6 ; 29, No. 7, 2 4 6 - 2 5 3 ; (1952) On
the Lagrangian correlation coefficient for turbulent diffusion and its
application to atmospheric diffusion phenom ena, Geophys. Res. Pap., No.
19, 3 9 7 - 4 1 2 ; (1959) The effects o f thermal stratification on turbulent
diffusion in the atmospheric surface layer. A dv. G eophys., 6 (Atmospheric
diffusion and air pollution), 3 1 9 —330.
International Conference, (19 6 5 ) Humidity and Moisture, Proc. 196 3 Intern.
Confer, in Washington, Vols. 1—3, Reinhold Publ. Corp., New York.
lordanskiy, S. V., and A. G. Kulikovskiy, (1 9 6 5 ) The absolute stability o f some
plane-parallel flows at large Reynolds numbers, Z/z. Eksper. Teor. Fiz., 49,
No. 4(10), 1 3 2 5 - 1 3 3 1 .
Isserlis, L., (1918) On a formula for the product-m om ent coefficient in any
num ber o f variables, 12, No. 1 - 2 , 1 3 4 - 1 3 9 .
Ivanilov, Yu. P., (1966) Secondary regimes in convective flows, Izvestiya A N
S SSR , Ser. Zhidk. i Gaza, No. 3, 9 3 - 9 4 ; (1968) Secondary regimes in a
Couette flov/, M ekh. Z hidk. i Gaza, No. 1, 1 1 0 - 1 1 7 .
Ivanilov, Yu. P., and G. N. Yakovlev, (1966) On the bifurcation o f the fluid
flows betw een the rotating cylinders, Prikl. Mat. i M ekh, 30, No. 4,
768-773.
Izakson, A., (1929) On the definition o f turbulence, Zh. R ussk. Fiz.-K him .
O bshch., 61, No. 3, 2 5 7 - 2 9 3 ; (1937) Form ula for the velocity distribution
near a wall, Z/z. Eksper. Teor. Fiz., 7, No. 7 , 9 1 9 - 9 2 4 .
Jeffreys, H., (1926) The stability o f a layer o f fluid heated from below, P M .
M a g .(l), 2, 8 2 2 - 8 4 4 ; (1 9 2 8 ) Some cases o f instability in fluid m otion,
Proc. R o y . Soc., A 1 1 8 ,N o . 779, 1 9 5 - 2 0 8 .
Johnk, R. E., and T. J. H anratty, (1962) T em perature profiles for turbulent flow
o f air in a pipe—I. The fully developed heat-transfer region, Chem. Engng.
Sci., 17, No. 1 1 , 8 6 7 - 8 7 9 .
Jones, J. 1. P., (1966) C. R. T. displays o f wind eddies and some Lagrangian
deductions from a vortex model o f turbulence, 7. A ppl. M eteor., 5, No. 6,
816-823.
Jones, J. I. P., and F. Pasquill, (1959) An experimental system for directly
recording statistics o f the intensity o f atmospheric turbulence. Quart. J.
R o y . M eteor. Soc., 85, No. 365, 2 2 5 - 2 3 6 .
Joseph, D. D., (1965) On the stability o f the Boussinesq equations. Arch. R at.
Mech. A n a l, 20, No. 1, 5 9 - 7 1 ; (1966) Nonlinear stability o f the
Boussinesqu equations by the m eth od o f energy, Archiv. R at. Mech. A n a l,
22, No. 3, 1 6 3 - 1 8 4 .
Joseph, D. D., and S. Carmi, (1966) Subcritical convective instability. Part 2.
Spherical shells, / F luid M ech., 26, No. 2, 7 6 9 - 7 7 7 .
Joseph, D. D., R. J. Goldstein, and D. J. Graham, (1968) Subcritical instability
and exchange o f stability in a horizontal fluid layer, Phys. Fluids, 11, No. 4,
903-904.
BIBLIOG RAPHY 71 5
Joseph, D. D., and C. C. Shir, (1966) Subcritical convective instability. Part 1.

Fluid l a y e r s ,/ F luidM ech., 26, No. 1 , 7 5 3 - 7 6 8 .
Kader, B. A., (1966) On the structure o f viscous sublayer o f incompressible
fluid, Izvestiya A N SSSR , Ser. Mekh. Zhidk. i Gaza, No. 6, 1 5 7 - 1 6 4 ;
(1969) The structure o f the turbulence in a viscous sublayer o f turbulent
boundary layer, Candidate Dissertation, Moscow Institute o f Mechanic.
Engineering for Chemical Industry, Moscow.
Kader, B. A., and A. M. Yaglom, (1 9 70 ) Universal law o f turb ulen t heat and
mass transfer from the wall at large Reynolds and Peclet num bevs, D oklady
A N S S S R , 190, No. 1 , 6 5 - 6 8 .
Kafoed-Hansen, O., and C. F. Wandel, (1967) On the relation between Eulerian
and Lagrangian averages in the statistical theory o f turbulence, Ris0 Report
No. 50, Danish Atomic Energy Commission, Ris0, Denmark.
Kaimal, J. C., and J. A. Businger, (1963) A continuous wave sonic anemometer-
th erm om eter, y. A ppl. Meteor.^ 2, No. 1, 156—164.
Kaimal, J. C., H. E. Cramer, F. A. Record, J. E. Tillman, J. A. Businger, and M.
Miyake, (1964) Com parison o f bivane and sonic techniques for measuring
the vertical wind com ponent. Quart. J. R o y. M eteor. Soc., 9 0, No. 386,
467-471.
Kaimal, J. C., D. A. Haugen, and J. T. Newman, (1966) A com puter-controlled
mobile micrometeorological observation system, J. A ppl. M eteor., 5, No. 4,
411-420.
Kaimal, J. C., and Y. Izumi, (1965) Vertical velocity fluctuations in a nocturnal
low-level jet, / A ppl. M eteor., 4, No. 5, 5 7 6 - 5 8 4 .
Kaimal, J. C., J. C. Wyngaard, and D. A. Haugen, (1968) Deriving power spectra
from a three-com ponent sonic anem om eter, J. A ppl. M eteor., 7, No. 5,
827-837.
Kahnske, A. A., and C. C. Pien, (19 4 4 ) Eddy diffusion,//id. Eng. Chem ., 36, No.
3 ,2 2 0 -2 2 3 .
Kampe de Feriet, J., (1939) Les fonctions aleatoires stationnaires et la theorie
statistique de la turbulence homogene, A nn. Soc. Sci. Bruxelles, 59,
1 4 5 - 1 9 4 ; (1951) Averaging processes and Reynolds equations in atm o s
pheric turbulence, J. M eteor., 8, No. 5, 35 8 —361; (1953) Fonctions
aleatoires et theorie statistique de la turbulence (in the book: A.
Blanc-Lapierre, R. F o rte t, Theorie des fo n c tio n s aleatoires, Paris, Masson,
568—623); (1 9 5 6) Problemes m athem atiques poses par la mechanique
statistique de la turbulence, Proc. Intern. Congress M ath. A m sterdam , 1954,
3, 2 3 7 - 2 4 2 , Amsterdam, North-Holland Publ. Co.
Kao, S. K., (1959) T urbulent transfer in the boundary layer o f a stratified fluid,
J. M eteor., 16, No. 5 , 4 9 7 - 5 0 3 .
Kapitza, P. L., (1947) Theoretical and empirical formulas for the heat transfer in
two-dimensional tu rbulent flow, D oklad y A N S S S R , 55, No. 7, 59 5 —602.
Kapoor, R. K., and A. Sundararajan, (1968) Com m ents on “ E ddy conductivity
and eddy viscosity ratio in lapse conditions” (with a reply by J. P.
P a n d o l f o ) , / A tm o sp h . Sci., 25, No. 3, 5 22 —526.
71 6 BIBLIOGRAPHY
Karman, T. von, (1924) Uber die Stabilitat der Lam inarstromung und die
Theorie der Turbulenz, Proc. 1st Intern. Congr. A p p l M ech., Delft,
9 7 —112; (1930) Mechanische Ahnlichkeit und Turbulenz, Nachr. Ges. Wiss.
Gottingen, Math.-Phys. Kl., 5 8—76; (1934) Some aspects o f the theory o f
turbulent m o tion , Proc. Intern. Congr. A p pl. M ech., Cambridge; (1 9 3 9 ) The
analogy between fluid friction and heat transfer, Trans. A m er. Soc. Mech.
FMgrs., 6 1 , 7 0 5 - 7 1 0 .
Karol’, I. L., (1960) Effect o f turbulent diffusion in the direction o f the mean
wind on the distribution o f a substance diffusing through the atmosphere,
D oklady A N S S S R , 131, No. 6, 1 2 8 3 - 1 2 8 6 .
Kazanskiy, A. B., and A. S. Monin, (1956) Turbulence in surface layer
inversions, Izvestiya A N S S S R ,S e i. geofiz.. No. 1, 7 9 - 8 6 ; (1957) Shape o f
smoke plumes, Izvestiya A N S S S R , Ser. geofiz.. No. 8, 1020—1033; (1958)
T urbulent regime in th e atmospheric surface layer during unstable stratifica
tion, Izvestiya A N S S S R .S qx. geofiz., No. 6, 7 4 1 - 7 5 1 ; (1961) Dynamic
interaction between the atm osphere and the earth’s surface, Izvestiya A N
S S S R , Ser. geofiz., No. 5, 7 8 6 —799; (1962) D eterm ination o f turbulent
transfer o f angular m o m en tum , heat, and moisture from profile data.
M eteor, i G idrologiya, No. 1 2 , 3 - 8 .
Kedrolivanskiy, V. N., and M. S. Sternzat, { \ 9 S3) M eteorological Instrum ents,
G idrom eteoizdat Press, Leningrad.
Keldysh, M. V., (1951) Eigenvalues and eigenfunctions o f some classes o f
nonself-adjoint Qq\x2ii\o m , D oklady A N S S S R , 77, No. 1, 1 1 - 1 4 .
Keldysh, M. V., and V. B. Lidskiy, (1963) Spectral theory o f nonself-adjoint
operators. Transactions o f the F ourth All-U nion M athem atics Congress, Vol.
1, Acad. Sci. Press, Leningrad, 101 —120.
Keller, L. V., (19 2 5 ) Uber die Aufstellung eines Systems von Charakteristiken der
atmospharischen Turbulenz, Zh. Geofiz. M eteor., 2, No. 34, 2 7 5 —290.
Keller, L. V., and A. A. Fridm an, (1 9 24 ) Differentialgleichung fiir die turbulente
Bewegung einer kompressiblen Fliissigkeit, Proc. 1st Intern. Congr. A ppl.
M ech., Delft, 3 9 5 - 4 0 5 .
Kelly, R. E., (1967) On the stability o f an inviscid shear layer which is periodic
in space and time, J. F luid M ech., 27, No. 4, 6 5 7 - 6 8 9 ; (1968) On the
resonant interaction o f neutral distances in tw o inviscid shear flows, /. F luid
M ech., 31, No. 4, 7 8 9 - 7 9 9 .
Kempf, G., (1 92 9 ) Neue Ergebnisse der Widerstandforschung, Werft, Reederei,
H afen, 1 0 , 2 3 4 - 2 3 9 , 2 4 7 - 2 5 3 .
Kestin, J., and P. D. Richardson, (19 6 3 ) Heat transfer across turbulent,
incompressible boundary layers, Intern. J. H eat Mass Transfer, 6, No. 2,
147-189.
K irchg^sner, K., (1961) Die Instabilitat der Strom ung zwischen zwei rotieren-
den Zylindern gegeniiber Taylor-Wirbeln fiir beliebige Spaltbreiten, Z.
angew. M ath. Phys., 12, No. 1, 1 4 - 3 0 .
Kiser, K. M., (1963) Material and m om entum transport in axisymmetric
turbulent jets o f v/?iiQx,A.I.Ch.E.J., 9, No. 3, 3 8 6 - 3 8 9 .
BIBLIOGRAPHY 717
Kishinevsky, M. Kh., (1965) Heat transfer at high Prandtl numbers and mass
transfer to a turbulent flow from the wall at a sm ooth iubQ, Int. J. Heat and
Mass Transfer, 8, 1 1 8 1 - 1 1 8 6 .
Kistler, A. L., (1962) The structure o f the laminar s\xh\2iyQx, M ^canique de la
turbulence (Coll. Intern du CNRS a Marseille), Paris, Ed. CNRS, 2 8 7 - 2 9 7 .
Kitaygorodskiy, S. A., (1968) On the calculation o f the roughness parameter o f
sea surface, Izvestiya A N SSSR , Ser. Fiz. Atmosf. Okeana, 4, No. 8,
870-878.
Kitaygorodskiy, S. A., and Yu. A. Volkov, (1965) On the roughness parameter
o f sea surface and the calculation o f m o m entum flux in the surface layer o f
the atmosphere above the water, Izvestiya A N S S S R , Ser. Fiz. Atmosf.
Okeana, 1, No. 9 , 9 7 3 - 9 8 8 .
Klebanoff, P. S., (1955) Characteristics o f turbulence in a boundary layer with
zero pressure gradient, Advis. Com. A eronaut., Rep. No. 1247.
Klebanoff, P. S., and K. D. Tidstrom, (1959) Evolution o f amplified waves
leading to transition in a boundary layer with zero pressure gradient, Nat.
Aeronaut. Space A d m ., Tech. note D-195.
Klebanoff, P. S., K. D. Tidstrom, and L. M. Sargent, (19 6 2 ) The three-
dimensional nature of boundary-layer instability, / Fluid M ech., 12, No. 1,
1-3 4 .
Kleinstein, G., (1967) Generalized law o f the wall and eddy-viscosity model for
wall boundary layers, A IA A 5, No. 8, 1 4 0 2 -1 4 0 7 .
Klenshow, C. W., and D. Elliot, (1960) A numerical treatm ent o f the
Orr-Sommerfeld equation in the case o f a laminar jet, Quart. J. Mech. A ppl.
M ath., 13, No. 3 , 3 0 0 - 3 1 3 .
Kline, S. J., W. C. Reynolds, F. A. Schraub, and P. W. Runstadler, (1967) The
structure o f turbulent boundary la y ers,/. F luid M ech., 3,0, No. 4, 7 4 1 —773.
Klug, W .,(1 9 6 3 )Z u m vertikalen Windprofil, Phys. A tm o sp h ., 36, No. 3 - 4 ,
2 2 6 - 2 5 3 ; (1965) Diabatic influence on turbulent wind flucations. Quart. J.
R o y. Meteor. Soc., 91, No. 388, 2 1 5 - 2 1 7 ; (1967) D eterm ination o f
turbulent fluxes o f heat and m o m en tu m from the wind profile, Quart. J.
R oy. Meteorol. Soc., 93, No. 395, 1 0 1 - 1 0 4 ; (1968) Diffusion in the
atmospheric surface layer: comparison o f similarity theory w ith
observations, Quart. J. R o y, M eteorol. Soc., 94, No. 402 , 5 5 5 - 5 6 2 .
Klug, W., and F. Wippermann, (1967) Numerische Integrationen einer Gleichung
fiir die T urbulente Diffusion in der Atmosphare, A rch. M eteor., Geophys.
u nd B ioklim atol., 16, No. I, I - I I .
Kochin, N. I., I. A. Kibel’, and N. V. Roze, (1964) Theoretical H ydrom echanics,
Vol. 1. Transl. b y D. Boyanovitch; J. R. M. Radok (ed.) Interscience, New
York. Vol. II available from ASTIA as A D I 29210.
Kolesnikov, A. G., (1960) Vertical turbulent transfer in stably stratified sea,
Izvestiya A N S S S R , Ser, geofiz.. No. 11, 1 6 1 4 -1 6 2 3 .
Kolmogorov, A. N., (1931) Uber die analytischen M ethoden in der Wahrschein-
lichkeitsrechnung. Math. A nn., 104, 4 1 5 - 4 5 8 ; (1933) Zur Theorie der
718 BIBLIOG RA PHY
stetigen zufMligen Prozesse, M ath. A n n ., 108, 1 4 9 - 1 6 0 ; (1941a) Local

structure o f turbulence in an incompressible fluid at very high Reynolds
numbers, D oklady A N S SSR , 30, No. 4, 2 9 9 - 3 0 3 ; (1941b) Energy
dissipation in locally isotopic turbulence, D oklady A N SSSR , 32,
No. 1, 1 9 - 2 1 ; (19 4 2 ) Equations o f turbulent motion of an incompressible
fluid, Izvestiya A N S S S R .S qx. fiz., 6, No. 1 - 2 , 5 6 - 5 8 ; (1952) Skin friction
and velocity profiles o f turbulent flow^s in tubes, D oklady A N SSSR ,
84, No. 1, 2 9 - 3 0 ; (1954) New variant o f the gravitational theory of
m otion o f suspensions, Vestn. Moscov^ State Univ. No. 3, 41 -45; (1962) A
refinem ent o f previous hypotheses concerning the local structure of
turbulence in viscous incompressible fluid at high Renolds number. J. Fluid.
M ech., 13, No. 1 , 8 2 - 8 5 .
Kolmogorov, A. N., (1935) La transform ation de Laplace dans les espaces
\\ni^\xQ ^,C om ptes R cn dus Acad. S e t, 200, No. 21, 1 7 1 7 -1 7 1 8 .
K omoda, H., (1967) Nonlinear development o f disturbance in a laminar
boundary layer, Fluids, 10, Supplem ent, S87 -S 9 4 .
K ondo, J., (1962a) Observations on wind and tem perature profiles near the
ground, Sci. R eports T o hu ku Univ., Ser. 5, Geophys., 14, No. 2, 41--56;
(1962b) Evaporation from extensive surfaces of water., .SV/. Rep. T dhoku
Univ., Ser. 5. Geophys., 14, No. 3, 1 0 7 - 1 1 9 .
Konstantinov, A. R., (1952) Calculation o f evaporation from agricultural fields,
taking into account bands o f aforestration, Trudy Gos. Gidrologieli. In-ta,
No. 34(88), 1 5 - 6 5 .
Koppel, D., (1964) On the stability o f flow o f a thermally stratified flow under
the action o f g ravity,7. M ath. Phys., 5, No. 7, 9 6 3 - 9 8 2 .
Koschmieder, E. L., (1966) On convection on a uniformly heated plane, Beitr. z.
Phys. d. A tm o sp h ., 39, No. 1, 1 - 1 1 ; (1967) On convection under an air
su rfa ce ,/. F luid Meeh., 30, No. 1 , 9 - 1 5 .
Kovasznay, L. S. G., (1953) Turbulence in supersonic flow, / Aeronaut. Sei.,
20, No. 10, 6 5 7 - 6 7 4 , 682; (1965) Structure detaillee de I’ecoulement
pendant la transition, Conipt. rend. 8 Journees hydraul. Soe. hydroteehn.
France, Lille, 1964, 1, Grenoble, 1965, 21 42; (1966) Turbulence
measurements. A p p lie d M echanics Surveys, Spartan Books, Macmillan and
Co., 9 7 1 - 9 8 2 ; (1967) Structure of the turbulent boundary layer, Phys.
Fluids, 10, Supplem ent, S 2 5 -S 3 0 .
Kovasznay, L. S. G., H. K omoda, and B. R. Vasudeva, (1962) Detailed tlow field
in transition, Proc. H eat Transfer and F luid Mech. In st., Stanford Univ.
Press, 1 - 2 6 .
Kraichnan, R. H., (1959) The structure o f isotropic turbulence at very high
Reynolds numbers, J. F luid M ech., 5, No. 4, 4 9 7 - 5 4 3 ; (1962a) T urbulent
thermal convection at arbitrary Prandtl num bers, P /2V5. Fluids, 5, No. 11,
1 3 7 4 - 1 3 8 0 ; (1962b) The closure problem o f turbulence theory, Proc.
S ym p . A ppl. M ath., 13 (H ydrodynam ic instability), 1 2 9 - 2 2 5 .
Krasnosel’skiy, M. A., (19 6 4 ) Topological M ethods in the Theory o f N onlinear
Integral Equations, Pergamon Press, New York.
BIBLIOGRAPHY 719
Kravchenko, I. N., (1963) Wind profile in the surface \ 2iyQ xJzvestiya A N S S S R ,

Ser. geofiz., No. 3, 4 9 8 - 4 9 9 .
Krechmer, S. I., (1954) Method o f measuring m icrofluctuations in wind velocity
and tem perature o f the atmosphere, T rudy Geofiz. Inst. A N S S S R , No.
24 (151), 4 3 —111; (1957) Field instrum ent for measuring the tem perature
lapse-rate in the atm osphere, T rudy Tsentr. Aerolog. Observ., No. 22,
96 -9 9 .
Krishna, K., (1968) A numerical study o f the diurnal variation o f meteorological
parameters in the planetary boundary h y e r , M o n th ly Weather R ev., 96 , No.
5 ,2 6 9 -2 7 6 .
Kropik, K., (1 9 64 ) Beitrage zum Stabilitatsproblem der Poiseulleschen
A cta Phys. A ustr., 17, No. 4, 3 5 1 - 3 7 7 .
Krueger, E. R., (1962) The Stability o f C ouette and Spiral F low , Ph. D. Thesis,
Rensselaer Polytechnic Institute, T roy, New York.
Krueger, E. R., and R. C. Di Prima, (1964) The stability o f viscous fluid between
rotating cylinders w ith an axial flow, J. F luid Mech., 19, No. 4, 5 2 8 - 5 3 8 .
Krueger, E. R., A. Gross, and R. C. Di Prima, (1966) On the relative im portance
o f Taylor-vortex and non-axisymmetric modes in flow between rotating
cy lin d ers,/. F luid M ech., 24, No. 3, 5 2 1 - 5 3 8 .
Krylov, A. L., (1 96 3 ) P ro o f o f the instability o f a particular flow o f a viscous
incompressible fluid, D oklady A N S S S R , 153, No. 4, 7 8 7 - 7 8 9 ; (1964) On
the stability o f a Poiseuille flow in a plane c\\2inne\, D oklady A N S SSR , 159,
No. 5 , 9 7 8 - 9 8 1 .
Kulikovskiy, A. G., (1966) On the stability o f the Poiseuille flow and some other
plane-parallel flows in th e plane w ind-tunnel o f large finite length at large
Reynolds n u m b e r s , Mat. M ekh., 30, No. 5, 8 2 2 - 8 3 5 .
Kuo, H. L., (1961) Solution o f the nonlinear equations o f cellular convection
and heat t r a n s p o r t , / F luid M ech., 10, No. 4, 6 1 1 - 6 3 4 .
Kurpakova, T. A., and L. R. Orlenko, (1967) On the laws for th e tem perature
and wind velocity distributions in the boundary layers, Trudy Glavn.
Geofiz. O b s e r v .,^ 0 . 205, 1 3 - 2 4 .
Kuwabara, S., (1967) Nonlinear instability o f plane C ouette ^ \o ^ ,P h y s. Fluids,
10, Supplem ent (No. 9, Part II), SI 1 5 - S l 16.
Lamb, H., { \9 3 2 ) H ydrodynam ics, 6 th ed., Dover Publ., N. Y.
Landau, L. D., (1944) Turbulence, D oklady A N SSSR , 44, No. 8, 3 3 9 - 342.
Landau, L. D., and E. M. Lifshitz, (1 95S) Statistical Physics. Trans, by E. Peierls
and R. F. Peierls. Pergamon Press, London; (1963) F luid Mechanics,
Pergamon Press London.
Laufer, J., (1951) Investigation o f turbulent flow in a two-dimensional channel,
Nat. Advis. Com. A ero n a u t., Rep. No. 1033; (1954) The structure o f
turbulence in fully developed pipe flow, Nat. Advis. Com. A ero n a u t., Rep.
No. 1174.
Laykhtm an, D. L., (1944) Wind profile and mixing in the atmospheric surface
layer, Izvestiya A N SS S R , Ser. geogr. i geofiz., 8, No. 1, 1 - 5 ; (1947a) Wind
profile in the atmospheric surface layer under stationary conditions, Trudy
Nauchno-Issl. Uchr. Glavn. Upr. G id ro m etslu zh b y, ser. l , N o . 3 9 , 5 8 - 7 6 ;

(1947b) T ransform ation o f the air mass under the influence o f the
underlying surface, Mateorol. i G idrol, No. 1, 1 7 - 2 1 ; (1961) o f the
atm ospheric boundary layer, G idrom etizdat Press, Leningrad; (1963)
Diffusion o f a contam inant from point sources in the atmospheric surface
layer, T rudy Leningr. G idrom et. In-ta, No. 15, 3 - 9 .
Laykhtm an, D. L., and A. F. Chudnovskiy, (1 949) Physics o f th e atm ospheric
surface layer, G ostekhizdat Press, Leningrad-Moscow.
Leite, R. J., (1 9 5 9) An experimental investigation o f the stability o f Poiseuille
flo w ,/. F luidM ech., 5, No. 1 , 8 1 - 9 6 .
Lessen, M., (1950) On stability o f free laminar boundary layer between parallel
streams,TViOfr. A dv. Com, A eronaut., Rep. No. 979.
Lessen, M., J. A. Fox, W. V. Bhat, and Ting-Yung Liu, (1964) Stability of
Hagen-Poiseuille ^[o^,P hys. Fluids, 7, No. 8, 1 3 8 4 - 1 3 8 5 .
Lessen, M., and Sung-Hwan Ko, (1966) Viscous instability o f an incompressible
fluid half-jet ^\ov^,Phys. Fluids, 9, No. 6, 1 1 7 9 - 1 1 8 3 .
Lessen, M., S. G. Sadler, and T. -Y. Liu, (1968) Stability o f pipe Poiseuille flow,
Phys. Fluids, 11, No. 7, 1 4 0 4 - 1 4 0 9 .
Lettau, H., (1950) A reexam ination o f the “ Leipzig wind profile” considering
some relations between wind and turbulence in the friction layer, Tellus, 2,
No. 2, 1 2 5 - 1 2 9 ; (1952) On eddy diffusion in shear zones, G eophys. Res.
Pap., No. 19, 4 3 7 - 4 4 5 ; (1959) Wind profile, surface stress and geostrophic
drag coefficients in the atmospheric surface layer. A dv. Geophys. 6
(A tm ospheric diffusion and air pollution) 2 4 1 —257; (1962) Theoretical
wind spirals in the boundary layer o f a barotropic atmosphere, Beitr.
Phys, A tm o sp h ., 35, No. 3 - 4 , 1 9 5 - 2 1 2 ; (1967a) Problems of
micrometeorological measurements, in The Collection and Processing o f
Field Data (ed. by E. F. Bradley and O. T. Denmead), Interscience Publ.,
New York-London-Sydney; (1 96 7 b ) New hypothesis for the relationship
between eddy and mean states, Fluids, 10, Supplem ent, S 79—S83.
Lettau, H., and B. Davidson, (1957) E xploring the A tm o sp h ere's First Mile,
Vols. 1 - 2 , Pergamon Press, London-New York-Paris.
Levich, V. G., (1944) Theory o f concentrational polarization, Zh. Fiz. Khim ii,
18, No. 9, 3 3 5 - 3 5 5 ; ( 1 9 5 1 ) T heory o f diffusion processes in a moving fluid
(hydrodynam ics at high Prandtl numbers), D oklady A N SSSR , 78, No. 6,
1 1 0 5 - 1 1 0 8 ; ( \9 6 2 ) P hysicochem ical H ydrodynam ics (Trans, o f Russian
2nd ed.), Prentice-Hall, Englewood Cliffs, N. J.
Lewis, D. J., (1928) An experimental study o f the m otion o f a viscous liquid
contained between two coaxial cylinders, Prc;c*. R o y . Soc., A 117, 3 8 8 —407.
Liepmann, H. W., (1962) Free turbulent flows, M ecanique de la turbulence
(Coll. Intern, du CNRS a Marseille), Paris, Ed. CNRS, 211 - 2 2 6 .
Lighthill, M. J., (1952) On sound generated aerodynamically. I. General theory,
Proc. R o y . Soc., A 211, No. 1107, 5 6 4 - 587; (1954) On sound generated
aerodynamically. II. Turbulence as a source o f sound, Proc. R o y, Soc.,
A 2 2 2 ,N o . 1148, 1 - 3 2 .
BIBLIOGRA PHY 721
Liljequist, G. H., (1957) Energy exchange o f an Antarctic snow-ficld, Fart I,

Norsk Polarinstitut, Oslo.
Lin, C. C., (1945) On the stability o f two-dimensional parallel flows. Parts I - I I I ,
Quart. A p p l M ath., 3, No. 2, 1 1 7 - 1 4 2 , No. 3, 2 1 8 - 2 3 4 , No. 4, 2 7 7 - 3 0 1 ;
(1955) The Theory o f H ydrodyna m ic S ta b ility , Cambridge Univ. Press;
(1961) Some mathem atical problems in the theory o f the stability o f
parallel flows, J. F luidM ech., 10, No. 3, 4 3 0 - 4 3 8 .
Lin, C. C., and D. J. Benney, (1962) On the instability o f shear flows, Proc.
S ym p . A ppl. M ath., 13 (Hydrodynam ic instability), 1—24.
Lin, C. S., E. B. D enton, H. S. Gaskill, and G. L. Putnam , (1951)
Diffusion-controlled electrode reactions, Ind. Eng. Chem., 43, No. 9,
2136-2143.
Lin, C. S., R. W. M oulton, and G. L. Putnam , (1953) Mass transfer between solid
walls and fluid streams, Ind. Eng. Chem ., 45, No. 3, 6 3 6 —646.
Lindgren, E. R., (1 9 57 ) The transition process and other phenom ena in viscous
flow. A rk. F ysik, 12, No. 1 - 2 , 1—169; (1959) Liquid flow in tubes l-IIl,
A rk. Fys., 15, No. 2, 9 7 - 1 1 9 ; No. 6, 5 0 3 - 5 1 9 ; 16, No. 1, 1 0 1 - 1 1 2 ;
(1961) Liquid flow in tubes W - V , A r k . Fys., 18, No. 5, 4 4 9 - 4 6 4 ; No. 6,
5 3 3 - 5 4 1 ; (1965) E xperim ental study on turbulent pipe flows of distilled
water. Dept. Civil Eng., Oklahom a State Univ., Rept. 1AD621071; (19 6 7)
T urbulent convection—Malkus theory, Nat. Cent. A tm o sp h . Res. Techn.
N o te s ^ ^ o . 24, 1 5 1 - 1 7 5 .
Lindzen, R. S., (1967) T urbulent co n v ection-M alkus theory, Nat. Cent.
A tm o sp h . Res. Techn. N o tes, No. 24, 1 5 1 - 1 7 5 .
Lock, R. C., (1955) The stability o f the flow o f an electrically conducting fluid
between parallel planes under a transverse magnetic field, R o y. Soc.,
A 2 3 3 ,N o . 1192, 1 0 5 -1 2 5 .
Loeve, M., (1960) Probability Theory, 2nd edition, Princeton Univ. Press; D. van
Nostrand.
Loitsyanskiy, L. G., (1935) Some applications o f the m ethod o f similarity to the
theory of turbulence, Prikl. M atem . M ekh., 2, No. 2, 1 8 0 - 2 0 6 ; (1941)
Aerodynam ics o f the B oundary Layer, Gostekhizdat Press, Leningrad-
Moscow; (1958) Localization hypothesis for the turbulent flow o f a fluid in
the presence o f viscosity, Prikl. M atem . M ekh ., 22, No. 5, 6 0 0 - 6 1 1 ;(1 9 6 0 )
Transport o f heat in turbulent m otion, Prikl. M atem . M ekh., 24, No. 4,
6 3 7 - 6 5 1 ; (1962a) Semiempirical theories o f interaction between the
processes o f molecular and molar exchange in turbulent fluid flow, Trans, o f
the All-U nion Congress on Theoretical A p p lied Mechanics, I 9 6 0 , Acad. Sci.
Press, Moscow-Leningrad, 1 4 5 - 1 4 6 ; (1962b) Lam inar B oundary Layer,
Fizmatgiz Press, Moscow.
Longwell, P. A., (1966) Mechanics o f Fluid Flow, McGraw-Hill Book Co., New
York.
Lorentz, H. A., (1907) Uber die Entst^hung turbulenter Fliissigkeitsbewegungen
und uber den Einfluss dieser Bewegungen bei der Strom ung durch Rohren,
A bh. theor. Phys., Leipzig, 1, 4 3 - 7 1 .
72 2 BIBLIOGRA PHY
Lorenz, E., (1953) The interaction between a mean flow and random
disturbances, Tellus, 5, No. 3, 2 3 8 - 2 5 0 .
Low, A. R., (1929) On the criterion for stability for a layer o f viscous fluid
heated from below, Proc. R o y. Soc., A 125, 1 8 0 - 1 9 5 .
Ludwieg, H., (1956) Bestimmung des Verhaltnisses der Austauschkoeffizienten
fiir Warme und Impuls bei turbulenten G renzschichten, Zs. fHr Flugwiss,, 4,
No. 2 , 7 3 - 8 1 .
Ludwieg, H., and W. Tillmann, (1949) Untersuchungen iiber die Wandschubspan-
nung in turbulenten R e i b u n g s s c h i c h t e n , A rch., 17, No. 4, 2 8 8 —299.
Lumley, J. L., (1962a) The mathematical nature o f the problem o f relating
Lagrangian and Eulerian statistical functions in turbulence,MecamV/we t/e /a
turhulence (Coll. Intern, du CNRS a Marseille), Paris, Ed. CNRS; (1962b)
An approach to the E ulerian-Lagrangian problem, / Math. Phys., 3, No. 2,
3 0 9 - 3 1 2 ; (1967a) Rational approach to relations between motions o f
differing scales in turbulent flows, Phys, Fluids, 10, No. 7, 1405—1408;
(1967b) The applicability o f turbulence research to the solution o f internal
flow problems, in F luid Mechanics o f Internal F low (ed. by G. Sovran),
Elsevier Publ. Co., Amsterdam, 1 5 2 - 1 6 9 .
Lumley, J. L., and H. A. Panofsky, (1964) The Structure o f A tm osp heric
Turbulence, Interscience Publ., New Y ork-London-Sydney.
Lumley, J. L., and R. W. Stewart, (1965) On a conjecture o f L ettau, J.
A tm o sp h . Sci., 22, No. 5, 5 9 2 - 5 9 4 .
Lyapin, Ye. S., (1948) T urbulent mixing o f air in the atmosphere, M eteor, i
Ciidrol, No. 5, 1 3 - 2 3 ; (1950) Hyperbolic equation for vertical turbulent
transfer in the atmosphere, Trudy Glavn. Geofiz. Observ., No. 19(81),
175-184.
Lyatkher, V. M., (1968) Turbulent pulsations in the viscous sublayer, D oklady
A N S S S R , 180, No. 2, 300 -303.
Lykoudis, P. S., and J. S. Touloukian, (1958) Heat transfer in liquid metals,
Trans. A m er. Soc. Mech. Engrs., 80, No. 3, 6 5 3 - 6 6 6 .
Lyon, R. N., (1951) Liquid metal heat-transfer coefficients, Chem. Eng.
Progress, 47, No. 2, 7 5 - 7 9 .
Lyubimov, B. Ya., (1969) Lagrangian description o f turbulence dynamics,
D oklady A N SS S R , 184, No. 5, 1 0 6 9 -1 0 7 1 .
McAdams, W. H., (1954) Heat Transmission, 3rd edition, McGraw-Hill Book Co.,
New Y ork-Toronto-London.
MacCready, P. B., (1953) Atmospheric turbulence measurements and a n a ly s is ,/
M eteor., 10, No. 5, 3 2 5 - 3 3 7 .
Mcllroy, I. C., (1955) The atmospheric fine structure recorder, C.S.I.R.O ., Div.
M eteor. Phys., Techn. Pap. No. 3, Melbourne.
Mcllroy, I. C., and D. E. Angus, (1963) The Aspendale multiple weighed
lysimeter installation, C .S.I.R.O ., Div. M eteor., Techn. Paper No. 14.
McVehil, G. E., (1964) Wind and tem perature profiles near the ground in stable
stratification. Quart. J. R o y . M et. Soc., 90, No. 384, 1 3 6 - 1 4 6 .
BIBLIOG RAPHY 723
Malhotra, R. C., and J. E. Cermak, (1963) Wind-tunnel modeling o f alniospiicric

diffusion, / Geophys. Res., 68, No. 8, 2 1 8 1 - 2 1 8 4 .
Malkus, W. V. R., (1954a) Discrete, transitions in turbulent convection, Proc.
R o y. Soc., A 225, No. 1161, 1 8 5 - 1 9 5 ; (1954b) The heat transport and
spectrum of thermal turbulence, Proc. R oy. Soc., A225, No. 1161,
196-212.
Malkus, W. V. R., and G. Veronis, (1958) Finite amplitude cellular convection,
J. F luidM ech., 4, No. 3, 2 2 5 - 2 6 0 .
Mandell, D. A., and E. E. O ’Brien, (1967) Single point and relative diffusion in
unstable wall layers, Fluids, 10, Supplem ent, S 2 5 9 -S 2 6 2 .
Martinelli, R. S., (1947) Heat transfer to molten metals, Trans. A m er. Soc.
Mech. Engrs., 69, 9 4 7 - 9 5 9 .
Matsuoka, H., (196 0 ) A corrected turbulon model theory o f turbulent diffusion
and its application, / M eteor. Soc. Japan, 38, No. 3, 1 4 8 - 1 5 3 ; (1961) Note
on two-dimensional diffusion in the atmospheric surface layer, / Meteor.
Soc. Japan, 39, No. 6, 3 2 4 - 3 3 0 ; (1962) On the turbulent diffusion in the
atmosphere, Mem. Coll. Sci. Univ. K y o to , Ser. A, 30, No. 2, 2 4 5 - 2 7 2 .
Matveyev, L. T., (1960) Dependence o f the turbulence coefficient on height in
the atmospheric surface layer, Izvestiya A N S S S R , Ser. geofiz., No. 1,
83-88.
Meksyn, D., (1946) Stability o f viscous flow between rotating cylinders, I - I l l ,
Proc. R o y . Soc. A 1 8 7 ,N o . 1008, 1 1 5 - 1 2 8 , No. 1 0 1 1 , 4 8 0 - 4 9 1 , 4 9 2 - 5 0 4 ;
(1964) Stability o f laminar flow between parallel planes for two- and
three-dimensional finite disturbances, Z5. Phys., 178, No. 2, 1 5 9 -1 7 2 .
Meksyn, D., and J. T. Stuart, (1951) Stabihty o f viscous m otion between parallel
planes for finite disturbances, Proc. R o y . Soc., A208, No. 1095, 5 1 7 - 5 2 6 .
Mellor, G. L., (1966) The effects o f pressure gradients on turbulent flow near a
sm ooth wall, / F luid M ech., 24, No. 2, 2 5 5 - 2 7 4 .
Meyer, K. A., (1967) Time-dependent numerical study of Taylor vortex flow,
Phys. Fluids, 10, No. 9, 1 8 7 4 -1 8 7 9 .
Michael, D. H., (1961) Note on the stability of plane parallel flows, / Fluid
M ech., 10, No. 4, 5 2 5 - 5 2 8 .
Michalke, A., (1964) On the inviscid instability o f the hyperbolic tangent
velocity profile, / Fluid Mech., 19, No. 4, 543 - 5 5 6 ; (1965a) V ortex
form ation in a free boundary layer according to stability theory, / Fluid
M ech., 22, No. 2, 3 7 1 - 3 8 3 ; (19 6 5 b) On spatially growing disturbances in
an inviscid shear layer, / Fluid Mech., 23, No. 3, 5 2 1 - 5 4 4 ; (1968) The
instability o f mixing regions, lecture read at a meeting Transition fro m
laminar to Turbulent Flow, L ondon, July 1968; (1969) The influence of
the vorticity distribution on the inviscid instability o f a free shear layer,
Fluid D ynam ics Transactions, Vol. 4 (ed. by Fiszdon, W., Kucharczyk, P.,
Prosnak, W. J.), Panstwowe w ydaw nictw o naukowe, Warszawa, 7 5 1 - 7 6 0 .
Michalke, A., and P. F rey m u th , (1966) The instability and the formation of
vortices in a free boundary layer, A G A R D Conference Proceedings No. 4
72 4 BIBLIOG RA PHY
(Proc. Spec. Meeting held in Rhode-Saint-Genese, Belgium, 1 0 - 1 3 May

1966), 5 7 5 - 5 9 5 .
Michalke, A., and H. Schade, (19 63 ) Zur Stabilitat von freien Grenzschichten,
Ingenieur-Archiv, 33, No. 1, 1 - 2 3 .
Michalke, A., and A. Timme, (1967) On the inviscid instabiUty o f certain
two-dimensional vortex-type flows, J. F luid M ech., 29, No. 4, 6 4 7 - 6 6 6 .
Mickelsen, W. R., (1955) An experimental comparison o f the Lagrangian and
Eulerian correlation coefficients in homogeneous isotropic turbulence,
A dv, Com. A e r o n a u t, Techn. Note No. 3570 ; (1959) Measurements o f the
effect o f molecular diffusivity in turbulent d iffu sio n ,/. F luid M ech., 7, No.
3, 3 9 7 - 4 0 0 .
Middleton, W. B. K., and A. F. Spilhaus, (1953) M eteorological Instrum ents,
T o ro n to Univ. Press.
Mihaljan, J., (1962) A rigorous exposition o f the Boussinesq approxim ations
applicable to a thin layer o f fluid. A strophysics Journal, 136, 1 1 2 6 - 1 1 3 3 .
Miles, J. W., (1957) On the velocity profile for tu rbulent flow near a sm ooth
wall, J. A eronaut. S e t, 24, No. 9, 704; (1961) On the stability o f
heterogeneous shear flows, J. F luid M ech., 10, No. 4 , 4 9 6 - 5 0 8 ; (1963) On
the stability o f heterogeneous shear flows. Part 2, J. F luid M ech., 16, No. 2,
209-227.
Miller, J. A., and A. A. Fejer, (1964) Transition phenom ena in oscillating
boundary-layer flo w s,/. F luid M ech., 18, No. 3 , 4 3 8 - 4 4 8 .
MiUikan, C. B., (1939) A critical discussion o f turbu lent flows in channels and
circular tubes. Proc. 5 th Intern. Congr. A ppl. M ech., Cambridge (USA),
386-392.
Millionshchikov, M. D., (1939) Decay o f homogeneous isotropic turbulence in
viscous incompressible ^\\l\^s,D okla dy A N S S S R , 22, No. 5, 2 3 6 - 2 4 0 .
Mises, R .v o n , (1941) Some Remarks on the Laws o f Turbulent M otion in Tubes,
T. von Kdrmdn anniversary volum e, Calif. Inst. Techn. Press, Pasadena,
317-327.
Mitchell, J. E., and T. J. H anratty, (1966) A study o f turbulence at a wall using
an electrochemical wall shear-stress m eter, /. F luid M ech., 26, No. 1,
199-221.
Mitsuta, Y., (1966) Sonic anem om eter-therm om eter for general use, /. Meteor.
Soc. Japan, 44, No. 1, 12—23.
Moffatt, H. K., (1967) The interaction o f turbulence with strong wind shear, in
A tm o sp h eric Turbulence and R adio Wave Propagation (ed. by A. M. Yaglom
and V. I. Tatarskiy), Nauka Press, Moscow, 1 3 9 - 1 5 6 .
Monin, A. S., (1950a) Dynamic turbulence in the atmosphere, Izvestiya A N
S S S R , Ser. geogr. geofiz., 14, No. 3, 2 3 2 - 2 5 4 ; (1950b) Characteristics of
anisotropic turbulence, D o klady A N S S S R , 75, No. 5, 6 2 1 - 6 2 4 ; (1950c)
Turbulence in the atmospheric surface layer. Inform . Shorn. G l. Upr.
G idrom etsluzhby (Coll. Sci. Inform. H ydrom et. Service USSR, Moscow),
No. 1, 13—27; (1953) Mechanism responsible for the heating o f air over an
open steppe. Collection: Climatic and M icroclim atic Studies on the Caspian
BIBLIO G RA PHY 72 5
Plain, Acad. Sci. Press, Moscow, 1 0 0 - 1 2 3 ; (1955) Diffusion with finite

velocity, Izvestiya A N S S S R , Ser. geofiz.. No. 3, 2 34 —248; (1956a)
Semiempirical theory o f turbulent diffusion, T rudy G e o f Inst. A N S S S R ,
No. 33 (160), 3 —47; (19 5 6 b ) T u rbulent diffusion in the atmospheric
surface layer, Izvestiya 555'/?,Ser. geofiz.. No. 12, 1461 —1473; (1956c)
M acroturbulent exchange in the ea rth ’s atm osphere, Izvestiya A N 5 5 5 /? ,Ser.
geofiz.. No. 4, 4 5 2 - 4 6 3 ; (1958) Structure o f atmospheric turbulence, Teor.
Veroyat. i Y eyeP rim en., 3, No. 3, 2 8 5 —317; (1959a) Smoke propagation in
the surface layer o f the atmosphere, A dv. G eophys., 6 (Atmospheric
diffusion and air pollution), 3 3 1 - 3 4 3 ; (1 95 9 b ) On the boundary condition
on the earth surface for diffusing pollution. Adv. in G eophys., 6
(Atmospheric difffusion and air pollution), 4 3 5 —43 6; (1959c) Atmospheric
diffusion, Usp. Fizich. N a u k , 67, No. 1, 1 1 9 - 1 3 0 ; (19 6 0 ) Lagrangian
characteristics o f turbulence, D oklady A N S S S R , 134, No. 2, 3 0 4 —307;
(1962a) Structure o f wind velocity and tem perature fields in the
atmospheric surface layer. Collection: A tmospheric Turbulence, Trudy In-ta
F iziki A tm o sf. A N S S S R , No. 4, 5 - 2 0 ; (1962b) Empirical data on
turbulence in the surface layer of atmosphere, J. Geophys. R es., 67, No. 8,
3 1 0 3 - 3 1 0 9 ; (1962c) Lagrangian equations in the hydrodynam ics o f an
incompressible viscous fluid, Prikl. M atem , M ekh., 26, No. 3, 3 2 0 - 3 2 7 ;
(1965) Sym m etry properties o f turbulence in the atmospheric surface layer,
Izvestiya A N S S S R , Ser. fizika atmosfery i okeana, 1, No. 1 , 4 5 - 5 4 .
Monin, A. S., and A. M. Obukhov, (1953) Dimensionless characteristics o f
turbulence in the atmospheric surface layer, D oklady A N S S S R , 93, No. 2,
2 2 3 - 2 2 6 ; (1954) Basic turbu len t mixing laws in the atmospheric surface
layer, Trudy Geofiz. Inst. A N S S S R , No. 24(151), 1 6 3 - 1 8 7 .
Mordukhovich, M. I., and L. R. Tsvang, (1966) Direct measurements o f
turbulent fluxes at two heights in atmospheric surface layer, Izvestiya A N
S S S R , Ser. fiz. atmosf. okeana, 2, No. 8, 7 8 6 - 8 0 3 .
Munn, R. E., (1 9 6 1 ) A theoretical expression for the root mean square vertical
eddy fluctuation (w '^ )‘/ ^ J. M eteor., 18, No. 1, 3 8 - 4 2 ; (19 6 3 ) A reap
praisal of S u tto n ’s parameter n, J. A p p l M eteor., 2, No. 4, 4 4 0 —445.
Murphree, E. V., (1932) Relation between heat transfer and fluid in c iio n , Ind.
Eng. Chem., 24, No. 7, 7 2 6 - 7 3 6 .
Myrup, L. O., (1967) Tem perature and vertical velocity fluctuations in strong
convection, Quart. J. R o y . M eteor. Soc., 93, No. 39, 3 5 0 —360.
Nachtsheim, P. R., (1964) An initial value m e th o d for the numerical treatm ent
o f the Orr-Sommerfeld equation for the case o f plane Poiseuille flow,
N .A .S .A . Tech. N o te D-2414.
Naito, K., (1964) Some remarks on the Monin-Obukhov function in the
atmosphere near the ground, J. M et. Soc. Japan, 42, No. 1, 5 3 - 6 3 .
Narayanan, M. A., and T. Narayana, (1967) Some studies on transition from
laminar to tu rbulent flow in a two-dimensional channel, Zs. angew. M ath.
Phys., 18, No. 5, 6 4 2 - 6 5 0 .
726 B IBLIOG RA PHY
Nevzglyadov, V. G., (1945a) Phenomenologic theory o f turbulence, D oklady A N

SS S R , 47, No. 3, 169—173; (1945b) Statistical theory o f turbulence,
D oklady A N S S SR , 47, No. 7, 4 8 2 - 4 8 5 .
Nield, D. A., (1967) The therm ohaline Rayleigh-Jeffreys problem, J. F luid
M ech., 29, No. 3, 5 4 5 —558; (1968) The Rayleigh-Jeffreys problem with
bo u nd ary slab o f finite conductivity, J. F luid M ech., 32, No. 2, 3 9 3 —398.
Nikuradse, J., (1932) Gesetzmassigkeiten der turbulenten Strom ung in glatten
Rohren, V D I'Forschungsheft No. 356; (1933) Stromungs gesetze in rauhen
R ohren, VD I-Forschungsheft No. 361.
Nissan, A. H., J. L. Nardacci, and C. Y. Ho, (1 9 6 3 ) The onset o f different modes
o f instability for flow between rotating cylinders,/I./.C/z.^. Journal, 9, No.
5, 6 2 0 - 6 2 4 .
Novikov, Ye. A., (1958) T urbulent diffusion in a shear flow, P rikl M atem,
M ekh., 22, No. 3, 4 1 2 - 4 1 4 ; (1964) Convergence o f the functional Taylor
series for the characteristic functional o f a random field, Uspekhi M atem.
N auk., 19, No. 6(120), 1 9 5 - 1 9 7 .
N unner, W., (1956) W ^m eiibertragung und Druckabfall in rauhen Rohren,
VD I-Forschungsheft No. 455.
O’Brien, E. E., (1966) Diffusion o f clouds o f contam inant in a turbulent
boundary layer, / A tm o sp h . S e t, 23, No. 4, 3 8 7 - 3 8 9 .
O ’Brien, J. J., (1965) An investigation o f the diabatic wind profile o f the
atmospheric boundary la y e r ,/. Geophys. R es., 70, No. 10, 2 2 7 7 —2290.
O bukhov, A. M., (1941) Energy distribution in the spectrum o f a turb u len t flow,
Izvestiya A N S S S R , Ser. geogr. geofiz.. No. 4 - 5 , 4 5 3 - 4 6 6 ; (1942)
D istribution o f turbulent length scale in tubes o f arbitrary cross section,
P rik l M a te m M ekh,, 6, No. 2 —3, 109—220; (1946) Turbulence in thermally
inhom ogeneous atmosphere, T rudy In-ta Teoret. Geofiz. A N SS S R , 1,
9 5 - 1 1 5 ; (1954) Statistical description o f continuous fields, Trudy G eofiz.
In-ta A N S S S R , No. 2 4(151), 3 - 4 2 ; (1960) Structure o f tem perature and
velocity fields under conditions o f free conwection, Izvestiya A N S S S R ,S e i.
geofiz., No. 9, 1392—1396; (1962) Some specific features o f atmospheric
tu rb u le n c e ,/. F luidM ech., 1 3 , No. 1 , 7 7 —81.
Ogura, Y., (1952a) Note on the wind velocity profile in the non-adiabatic
atmosphere, J, M eteor. Soc. Japan, ser. II, 30, No. 10, 3 2 9 - 3 4 2 ; (1952b)
The theory o f turb u len t diffusion in the atm osphere. I—I I I , / . M eteor. Soc.
Japan, 30, No. 1, 2 3 - 2 8 ; No. 2, 5 3 - 5 8 ; No. 11, 3 8 6 - 3 9 7 .
Ohji, M., (1 96 7 ) Statistical theory o f wail turbulence, Phys. Fluids, 10,
Supplem ent, S I 5 3 —S I 54.
O kam oto, M., (1963) A note on the wind and tem perature profiles in the
diabatic atm osphere near the ground, Geophys. Mag. (Japan), 31, No. 3,
5 0 5 -514.
O kubo, A., (1967) Study o f turbulent dispersion by use o f Lagrangian diffusion
equation,P/zj's. Fluids, 10, Supplem ent, S 7 2 -S 7 5 .
Orlob, G. T., (1959) Eddy diffusion in homogeneous turbulence,/V oc. A m er.
Soc. Civil. Engrs., /. H ydraul. Div., 86, No. 9, 7 5 —101.
B IBLIOG RA PHY 727
Orlov, V. V., (1966) Experimental study o f wall turbulence in the channel, Z/i.
P rikl Mat. Tekhn. Fiz., No. 4, 1 2 4 - 1 2 6 .
Orr, W. M., ( 1 9 0 6 - 1 9 0 7 ) The stability or instability o f the steady m otions o f a
liquid, Proc. R o y. Irish A cad., A 2 7 , 9 - 2 7 , 6 9 —138.
Ovchinnikova, S. N., and V. I. Yudovich, (1968) Calculation o f a secondary
steady flow between rotating cylinders, P rikl Mat. M ekh., 32, No. 5,
8 5 8 -868.
Owen, P. R., and W. R. Thom pson, (1963) Heat transfer across rough surfaces,/.
F luidM ech., 15, No. 3, 3 2 1 - 3 3 4 .
Paeschke, W., (1936) Experimentelle Untersuchungen zum Rauhigkeits- und
Stabilitatsproblem in der bodennahen Luftschicht, Beitr. Phys. fr. A t-
m osph., 24, No. 3, 1 6 3 - 1 8 9 .
Pai, Shih-i, (1954) F luid D ynam ics o f Jets, D. van Nostrand Publ., Toronto-N ew
York-London.
Palm, E., (1960) On the tendency towards hexagonal cells in steady convection,
/ F luidM ech., 8, No. 2, 1 8 3 -1 9 2 .
Palm, E., and H. 0 ia n n , (1964) Contribution to the theory o f cellular thermal
convection, / F luidM ech., 19, No. 3, 3 5 3 - 3 6 5 .
Palm, E., T. ElHngsen, and B. Gjevik, (1 96 7 ) On the occurrence o f cellular
motion in Benard convection, 7. F luidM ech., 30, No. 4, 6 5 1 - 6 6 1 .
Panchev, S., (1961) Some quantitative relationships between the statistical
parameters o f purely thermal turbulence in the atmosphere, D oklady Bolg.
A kad. Nauk, 14, No. 8, 7 9 1 - 7 9 4 .
Pandolfo, J. P., (1963) A formula relating various nondimensional parameters o f
turbulence fields in atmospheric surface layer under diabatic conditions, /.
G eophys. Res., 68, No. 10, 3 2 4 9 - 3 2 5 6 ; (1966) Wind and tem perature
profiles for constant-flux b o undary layers in lapse conditions w ith a variable
eddy conductivity to eddy viscosity ratio, J. A tm o sp h . Sci., 23, No. 5,
495 -502.
Panofsky, H. A., (1952) A comparison o f different types o f stress measure
m ent, Quart. J. R oy. M eteor. Soc., 78, No. 337, 4 1 1 - 4 1 4 ; (1961a)
An alternative derivation o f the diabatic wind profile. Quart. J. R o y.
Meteor. Soc., 87, No. 371, 1 0 9 - 1 1 0 ; (1961b) Similarity theory and
temperature structure in the lower atmosphere. Quart. J. R o y. M eteor. Soc.,
87, No. 374, 5 9 7 - 6 0 1 ; (1962) The budget o f turbulent energy in the lowest
100 meters, J. Geophys. Research, 67, No. 8, 3 1 6 1 - 3 1 6 5 ; (1962) Scalc
analysis o f atmospheric turbulence at 2 m. Quart. J. R o y . Meteor. Soc., 88,
No. 375, 5 7 - 6 9 ; (1963) D eterm ination of stress from wind and
tem perature measurements, Quart. J. R o y. Meteor. Soc. 89, No. 37, 8 5 —93;
(1965) Reanalysis o f Swinbank’s Kerang observations, Final Rep. C ontract
No. A F(604)-6641, Pennsylvania State Univ., p. 67.
Panofsky, H. A., A. K. Blackadar, and G. E. McVehil, (19 6 0 ) The diabatic wind
profile, Quart. J. R oy. M eteor. Soc., 86, No. 369, 39 0 —398.
728 B IBLIOG RA PHY
Panofsky, H. A., and R. A. McCormick, (1960) The spectrum o f vertical velocity

near the surface, Quart. J. R oy. Meteor. Soc., 86, No. 370, 4 9 5 - 5 0 3 .
Panofsky, H. A., and B. Prasad, (1965) Similarity theories and diffusion, 7/zr. J.
A ir Water Poll., 9 , 4 1 9 - 4 3 0 .
Panofsky, H. A., N. Busch, B. Prasad, S. Hanna, E. Peterson, and E. Mares,
(1967) Properties o f wind and tem perature at R ound Hill, D artm outh,
Mass., Penn. State Univ. Res. and Develop. Tech. Rept. ECOM-0035-F.
Pasquill, F., (1949) Eddy diffusion o f water vapour and heat near the ground,
Proc. R o y. Soc., A198, No. 1052, 1 1 6 - 1 4 0 ; (1950) The aerodynamic drag
o f the grassland, Proc. R o y. Soc., A 202, No. 1068, 1 4 3 - 1 5 3 ; (1962a)
Recent broad-band spectral measurements o f turbulence in the lower
a tm o sp h e re ,/. Geophys. Res., 67, No. 8, 3 0 2 5 -3 0 3 1 ,( \9 6 2 b ) A tm ospheric
diffusion. A S tu d y o f the Dispersion o f Windborne Material fro m Industrial
and O ther Sources, D. van Nostrand Publ., London-Toronto-New
York-Princeton; (1966) Lagrangian similarity and vertical diffusion from a
source at ground level. Quart. J. R o y. Meteor. Soc., 92, No. 392, 185- 195
(also Disscussion, Quart. J. R o y. M eteor. Soc. (1967), 93, No. 395,
1 4 2 - 1 4 4 ).
Patankar, S. V., and B. D. Spalding, (1967) H eat and Mass Transfer in Boundary
Layers, Morgan-Grampian, London.
Paulson, C. A., (1967) Profiles o f wind speed, temperature and hum idity over
the sea. Dept, of Meteorology, Univ. o f Washington, Sci. Rep. NSF
GP-2418.
Pedley, T. J., (1967) The stability o f rotating flows with a cylindrical free
s u r f a c e ,/ F luidM ech., 30, No. 1, 1 2 7 -1 4 7 .
Pekeris, C. L., (1948) Stability o f a laminar flow through a straight pipe of
circular cross section to infinitesimal disturbances which are symmetrical
about the axis o f the pipe, Proc. Natl. Acad. Sci. U.S.A., 34, 285 -295.
Pekeris, C. L., and B. Shkoller, (1967) Stability o f plane Poiseuille flow to
periodic disturbances o f finite am phtude in the vicinity of the neutral curve,
J. F luid M ech., 29, No. 1, 31 -38.
Pellew, A., and R. V. Southwell, (1940) On maintained convective m otion in a
fluid heated from below, Froc. R oy. Soc., A 176, No. 966, 3 1 2 - 3 4 3 .
Perepelkina, A. V., (1957) Some results on trubulent temperature and vertical
wind-velocity fluctuations, Izvestiya A N SSSR , Ser. geofiz., No. 6,
7 6 5 - 7 7 8 ; (1959a) D etermination o f turbulent heat flux, Izvestiya A N
S S S R , Ser. geofiz.. No. 7, 1 0 2 6 - 1 0 3 5 ; (1959b) Statistical characteristics of
turbulent exchange in the atmospheric surface layer. Candidate Dissertation,
Inst. Phys. A tm osph.^ Moscow Acad. Sci. USSR; (1962) Characteristics o f
turbulence in the atmospheric surface layer under the conditions o f free
co n w ec iio n jzve stiy a A N S S S R . S qt. geofiz.. No. 2, 2 7 1 - 2 7 4 .
Perry, A. E., and P. N. Joubert, (1963) Rough-wall boundary layers in adverse
pressure g rad ien ts,/. F luid M ech., 17, No. 2, 1 9 3 - 2 1 1 .
Perry, A. E., J. B. Bell, and P. N. Jou b ert, (1966) Velocity and tem perature
BIBLIOGRAPHY 729
profiles in adverse pressure gradient turbulent boundary layers, J. Fluid

M ech., 25, No. 2, 2 9 9 - 3 2 0 .
Petrov, G. I., (1 9 37 ) On the stability o f vortical layers, T rudy T s/A G I (?roc.
Centr. A ero-Hydrodyn. Inst.), No. 304; (1 9 3 8 ) Propagation o f oscillations in
a viscous fluid and transition to turbulence, Trudy T s/A G I (Proc. Centr.
Aero-Hydrodyn. Inst.), No. 345.
Petukhov, B. S., and V. V. Kirillov, (1958) On the question o f heat transfer in
tu rbulent fluid flow in tubes, Teploenergetika, No. 4, 6 3 —68.
Pfenninger, W., (1961) Transition experiments in the inlet length o f tubes at
high Reynolds numbers, in B oundary L ayer and F low Control, Vol. 2 (ed.
by G. V. Lachmann), Pergamon Press, O xford-London-New York-Paris,
970-980.
Philip, J. R., (1963) The theory o f dispersal during laminar flow in tubes, I—II,
A ustr. J. Phys., 16, No. 3, 2 8 7 - 2 9 9 ; 3 0 0 - 3 1 0 ; (1967) Relation between
Eulerian and Lagrangian statistics, Phys. Fluids, 10, Supplem ent, S 6 9 -S 7 1 .
PhiUips, O. M., (1 95 5 ) The irrotational m o tion outside a free turbulent
boundary, Proc. C am br P h il Soc., 51, No. 1, 2 2 0 - 2 2 9 ; (1966) The
D ynam ics o f the Upper Ocean, Cambridge Univ. Press; (1967) The
maintenance o f Reynolds stress in turbulent shear f lo w ,/. F luidM ech ., 27,
No. 1, 1 3 1 - 1 4 4 .
Platzman, G., (1965) The spectral dynamics o f laminar convection, J. F luid
Mec/z., 2 3 , N o . 3 , 4 8 1 ~ 5 1 0 .
Plows, W. H., (1968) Some numerical results for two-dimensional steady laminar
Benard c o n v e c t i o n , Fluids, 11, No. 8, 1 5 9 3 -1 5 9 9 .
Polavarapu, R. J., and R. E. Munn, (1967) Direct m easurem ent o f vapor pressure
fluctuations and g rad ients,/. A p p l M e te o r, 6, No. 4, 6 9 9 - 7 0 6 .
Ponom arenko, Yu. B., (1965) On the “ hard self-excitation” o f the stationary
m otions in fluid dynamics, P rikl Mat. i M ekh., 29, No. 2, 3 0 9 - 3 2 1 ;
(1968a) On the generation o f space-periodic m otions in the fluid dynamics,
Prikl. Mat. i M ekh., 32, No. 1, 4 6 - 5 8 ; (1968b) Processes o f the
establishment o f hexagonal convection cells, Prikl. Mat, i M ekh,, 32, No. 2,
2 4 4 - 2 5 5 ; (1968c) On stabiUty o f the plane Couette flow, P rik l Mat. i
M ekh., 32, No. 4, 6 0 9 - 6 1 4 .
Popovich, A. T., and R. L. Hummel, (1967) Experimental study o f the viscous
sublayer in turbulent pipe flow, A .I.C k E . J., 13, No. 5, 8 5 4 - 8 6 0 .
Potter, M. C., (1966) Stability o f plane Couette-Poiseuille flow, / . F luidM ech.,
24, No. 3, 6 0 9 - 6 1 9 ; (1967) Linear stability theory of symmetrical
parabolic flows, Phys. Fluids, 10, No. 3, 4 7 9 - 4 8 9 .
Prandtl, L., (1910) Eine Beziehung zwischen Warmeaustausch und Stromung-
widerstand der Fliissigkeiten, Phys. Zs., 11, 1 0 7 2 - 1 0 7 8 ; (1914) Uber den
Luftwiderstand von Kugeln, Nachr. Ges. Wiss. G ottingen, Math.-Phys. K l.
1 7 7 - 1 9 0 ; (1921) Bemerkungen iiber die Entstehung der Turbulenz, Zs.
angew. Math. M ech,, 1, No. 6, 4 3 1 - 4 3 6 ; (1925) Bericht iiber Unter-
suchungen zur ausgebildeten Turbulenz, Zs. angew. Math. M ech., 5, No. 2,
730 B IBLIOG RA PHY
1 3 6 - 1 3 9 ; (1927) Uber den Reibungswiderstand strom ender Luft, Ergebn.

A erodyn. Versuchsanst., Gottingen, 3, 1—5; (1928) Bemerkung iiber den
Warmeiibergang im Rohr, Phys. Zs., 29, No. 14, 4 8 7 - 4 8 9 ; (1932a)
Meteorologische Anwendungen der Stromungslehre, Beitr. Phys. fr. A t-
m osph., 19, No. 3, 1 8 8 - 2 0 2 ; (1932b) Zur tu rbulenten Strom ung in Rohren
und langs Flatten, Ergebn. A ero dyn. Versuchsanst, G ottingen, 4, 1 8 - 2 9 ;
(1942) Bemerkungen zur Theorie der freien Turbulenz, Zs. angew Math.
M ech., 22, No. 5, 2 4 1 - 2 4 3 ; (1945) Uber ein neues Formelsystem fiir die
ausgebildeteTurbulenz,A'flcZ/r. Ges. Wiss. G ottingen, Math.-Phys. K l., 6 - 1 9 ;
(1949) Fiihrer durch die Strom ungslehre, 3 ed., Braunschweig, F. Vieweg.
Pretsch, J., (1941a) Die Stabilitat einer ebenen Laminarstromung bei Druck-
gefalle und Druckansteig, Jahresber. deutsch. L u ftfa h rtf., 1, 1 5 8 - 1 7 5 ;
(1941b) Uber die StabiUtat einer Lam inarstromung in einem geraden R ohr
mit kreisformigem Q uerschnitt, Zs. angew. Math. M ech., 21, No. 3,
204-217.
Pries, T. H., and J. P. Appleby, (1967) The distribution o f eddy velocities and
tem perature fluctuations in the first 100 meters. Technical R ep o rt
ECOM -6027, U.S. Electr. Com m and, F ort Huachuca, Arizona.
Priestley, C. H. B., (1954) Convection from a large horizontal surface, J.
Phys., 7, No. 2, 176—201; (1955) Free and forced convection in the
atmosphere near the ground, Quart. J. R o y . Meteor. Soc., 81, No. 348,
139—143; (1956) Convection from the E arth ’s surface, Proc. R o y . Soc.,
A 238, No. 1214, 2 8 7 - 3 0 4 ; (1959a) T urbulent Transfer in the L ow er
A tm osp here, Chicago Univ. Press; (1959b) Estimation o f surface stress and
heat flux from profile data. Quart. J. R o y . M eteor. Soc., 85, No. 366,
4 1 5 —418; (1960a) A determ inant hypothesis for the superadiabatic wind
and tem perature profiles. Quart.. J. R o y . M eteor. Soc., 86, No. 368,
2 3 2 - 2 3 6 ; (1960b) T em perature fluctuations in the atmospheric boundary
layer, J. F luid M ech., 7, No. 3, 3 7 5 - 3 8 4 ; (1961) An alternative derivation
o f the diabatic wind profile, Quart. J. R o y . M eteor. Soc., 87, No. 373,
4 3 7 - 4 3 8 ; (1962) Free convection in a wind. Quart. J. R o y. M eteor. Soc.,
88, No. 375, 1 0 0 - 1 0 1 ; (1963) Eddy diffusion in shear zones. Quart. J.
R o y . M eteor. Soc., 89, No. 380, 2 8 7 - 2 8 8 ; (1 9 6 3 —1964) Recent flux and
profile measurements in Australia. R eport at a symposium at Berkeley, 30
August 1963, and private correspondence 7 April 1964.
Priestley, C. H. B., and W. C. Swinbank, (1947) Vertical transport o f heat by
turbulence in the a t m o s p h e r e , R o y . Soc., A 1 8 9 ,N o . 1019, 5 4 3 - 5 6 1 .
Prokhorov, Yu. V., (1961) The m eth o d o f characteristic functionals, Proc. 4th
B erkeley S ym p. Math. Stat. Probab., 2, Berkeley-Los Angeles, Univ. Calif.
Press, 4 0 3 - 4 1 9 .
Proudm an, J., (1953) D ynam ical O ceanography, Methuen Publ., London.
Rannie, W. D., (1956) Heat transfer in turbulent shear flow, J. Aeronaut. Sci.,
23, No. 5 , 4 8 5 - 4 8 9 .
Rayleigh, (1880, 1887, 1895) On the stability, or instability, o f certain fluid
motions, Papers, Cambridge Univ. Press, 1, 4 7 4 - 4 8 7 ; 3, 1 7 - 2 3 ;
BIBLIOGRAPHY 731
4, 203 -2 0 9 ; (1892) On the question o f the stability o f the flow o f fluids, in

Sci. Papers, Cambridge Univ. Press, L ondon and New York, 3 , 575 - 5 8 4 ;
(1894) The T heory o f S o u n d , MacMillian Publ., London; (1913) On the
stability o f the laminar m otion o f an inviscid fluid, Sci. Papers, Cambridge
Univ. Press, 6, 1 9 7 - 2 0 4 ; (1916a) On the dynamics o f revolving fluids, ^c/.
Papers. Cambridge Univ. Press. 6, 4 4 7 - 4 5 3 ; (1916b) On convection
currents in a horizontal layer o f fluid when the higher tem perature is on the
under side, Sci. Papers, Cambridge Univ. Press, 6 , 4 3 2 - 4 4 3 .
Record, F. A., and H. E. Cramer, (1966) Turbulent energy dissipation rates and
exchange processes above a non-homogeneous surface. Quart. J. R o y.
M eteor. Soc., 92, No. 394, 5 1 9 - 5 3 2 .
Reichardt, H., (1938) Messungen turbulenter Schwankungen, 26, No.
2 4/25, 4 0 4 - 4 0 8 ; (1940) Die Warmeiibertragung in turbulenten Reibungs-
schichten, Zs. angew. Math. M ech., 20, No. 6, 2 9 7 —328; (1942) Gesetz-
massigkeiten der freien Turbulenz, VDl-Forschungs, No. 414; (1951a)
Vollstandige Darstellung der turbulenten Geschwindigkeitsverteilung in
glatten Leitungen, Zs. angew. Math. Mech., 31, No. 7, 2 0 8 - 2 1 9 ; ( 1 9 5 lb )
Die Grundlagen des turbulenten Warmeiiberganges, Arch. Ges. Wdrmetech.,
2, 1 2 9 -1 4 2 .
Reid, W. H., (1965) The stability o f parallel flows. Basic D evelopm ents in F luid
M echanics (ed. by M. Holt), Academic Press, New Y ork-London, 2 4 9 - 3 0 7 ;
(1966) Asympotic approxim ations in hydrodynam ic stabiUty.
N on-equilibrium Therm odynam ics, Variational Techniques, and Stability
(ed. by R. J. Donnelly, R. Herman, and L Prigogine), Chicago Univ. Press,
115-123.
Reid, W. H., and D. L. Harris, (1958) Some further results on the Benard
problem, Phys. Fluids, 1, No. 2, 1 0 2 - 1 1 0 ; (1964) On the stability o f
viscous flow betw een rotating cy lin d ers,/. F luid M ech., 20, No. 1, 9 5 —101.
Reynolds, A. J., (1965) Wall layers with non-uniform shear stress, J. Fluid
Mech., 22, No. 3 , 4 4 3 - 4 4 8 .
Reynolds, O., (1874) On the ex ten t and action o f the heating surface o f steam
boilers, Proc. Lit. Phil. Soc. Manchester, 14, 7 - 1 2 ; (1883) An experimental
investigation of the circumstances which determine whether the m otion of
water shall be direct or sinuous, and the law o f resistance in parallel
channels, Phil. Trans. R o y. Soc. L ondon, 174, 9 3 5 - 9 8 2 ; (1894) On the
dynamical theory o f incompressible viscous fluids and the determ ination of
the criterion, P M Trans. R o y. Soc. L ondon, 186, 123 -1 6 1 .
Reynolds, W. C., and M. C. Potter, (1967) Finite-amplitude instability o f parallel
shear flows, J. F luid M ech., 27, No. 3, 4 6 5 - 4 9 2 .
Reynolds, W. C., and W. G. Tiederman, (1966) Stability o f tu rbulent channel
flow, with application to Malkus’ theory, J. F luid M ech., 27, No. 2,
253-272.
Reynolds, W. C., W. M. Kays, and S. J. Kline, (1958) Heat transfer in the
tu rbulent incompressible boundary layer 1-Constant wall tem perature,
N A S A Tech. M em. 12-1-58 W.
732 BIBLIOG RA PHY
Ribaud, G., (1940) Courbe des vitesses au voisinage de la paroi en regime

d ’ecoulement turbulent, deduite des mesures de convection,C om pre Rend.
Acad, ScL, 211, No. 22, 541 543; (1941) Nouvelle expression du
coefficient de convection de la chaleur en regime d ’ecoulement tu r b u le n t,/.
Phys. R adium (8), 2, No. 1, 1 2 - 2 5 .
Richardson, L. F., (1920) The supply o f energy from and to atmospheric eddies,
Proc, R o y . Soc., A97, No. 686, 3 5 4 - 3 7 3 ; (\9 2 2 )W ea th er Prediction by
Num erical Process, Cambridge Univ. Press; (1926) Atmospheric diffusion
shown on a distance-neighbour graph, Proc. R o y. Soc., A llO , No. 756,
709-737.
Rider, N. E., (1954) Eddy diffusion o f m om entum , water vapour and heat near
the ground, Phil. Trans. R o y . Soc., A 246, No. 9 1 8 , 4 8 1 - 5 0 1 .
Rider, N. E., and G. D. Robinson, (1951) A study o f the transfer o f heat and
water vapour above a surface o f short grass, Quart. J. R o y . Meteor. Soc., 11
No. 333, 3 7 5 - 4 0 1 .
Riis, E., (1962) The stability o f Couette flow in non-stratified and stratified
viscous fluids, Geofys. Publik. (Oslo), 23, No. 4, 3 - 3 7 .
Rijkoort, P. J., (1968) The increase o f mean wind speed with height in the
surface friction layer, Afecfed en Verhand. No. 91, Royal Netherl. Meteor.
Inst., 115 pp.
Rivlin, R. S., (1957) The relation between the flow o f non-Newtonian fluids and
turbulent Newtonian fluids, Quart. A ppl. M ath., 15, 2 1 2 - 2 1 5 .
Roberts, P. H., (1965) The solution o f the characteristic value problems, Proc.
R oy. Soc., A 283, No. 1395, 5 5 0 - 5 5 5 (appendix to Donnelly and Schwarz,
1968); (1966) On nonlinear Benard convection, in N on-equilibrium
Therm odynam ics, Variational Techniques, and Stability (ed. by R. J.
Donelly, R. Herman, I. Prigogine). The University o f Chicago Press, Chicago,
1 2 5 -1 6 2 .
R obertson, J. M., (1957) The turbulent velocity distribution in rough pipe, Proc.
5 th M idwestern Conf. F luid Mech., Univ. o f Michigan Press, Ann Arbor.
Robinson, G. D., (1950) Two notes on temperature changes in the troposphere
due to radiation, Centenn. Proc. R o y. Meteor. Soc., 2 6 - 3 3 ; (1951) The
vertical convective heat flux in the atm osphere—a critical essay, Quart. J.
R o y. Meteor. Soc., 77, No. 331, 6 1 - 7 3 .
Roll, H. U., (1965) Physics o f the Marine A tm osp here, Academic Press, New
York-Lx>ndon.
Rosenbluth, M. N., and A. Simon, (1964) Necessary and sufficient condition for
the stability o f plane parallel inviscid flow, Phys. Fluids, 7, No. 4, 5 5 7 - 5 5 8 .
Rosencrans, S. I., and D. H. Sattinger, (1966) On the spectrum o f an operator
occurring in the theory o f hydrodynam ic stability, J. Math, and Phys., 45,
No. 3 , 2 8 9 - 3 0 0 .
Rossby, C. G., (1948) On the nature o f the general circulation in the lower
atmosphere. The A tm ospheres o f the Earth and Planets (ed. by G. P.
Kuiper), Chicago Univ. Press, chapt. 2.
BIBLIOGRAPHY 733
Rossby, C. G., and R. B. Montgom ery, (1935) The layer o f frictional influence
in wind and ocean currents,Pap. Phys. Oceanogr. M eteor., 3 , No. 3.
Rota, G. C., (1960) Spectral theory o f smoothing operations, Proc. Nat. Acad.
Sci. USA 46, No. 6, 8 6 3 - 8 6 8 .
R o tta, J. C., (1950) Das in W a n d n ^ e giiltige Geschwindigkeitsgesetz turbulenter
Strom ungen, Ing. A rch., 18, No. 4, 2 7 7 - 2 8 0 ; (1951a) Beitrag zur
Berechnung der turbulenten Grenzschichten, Ing. A rch., 19, No. 1, 3 1 - 41;
(1951b) Statistische Theorie nichthom ogener Turbulenz, 1, Zs. Phys., 129,
No. 5, 5 4 7 - 5 7 2 ; (1951c) Statistische Theorie nichthom ogener Turbulenz,
2, Zs. Phys., 131, No. 1, 5 1 - 7 7 ; (1956) Experimentcller Beitrag zur
Entstehung turbulenter Strom ung im Rohr,//i^. Arch., 24, No. 4, 258 281;
(1962a) Incompressible turbulent boundary layers, M ecanique de la
turbulence (Colloque Intern, du CNRS a Marseille), Paris, Ed. CNRS,
2 5 5 - 2 8 3 ; (1962b) Turbulent boundary layers in incompressible flow,
Progr. Aeronaut. Sci., 2, Pergamon Press, Oxford-London-New York-Paris,
1-2 1 9 .
Rounds, W., (1955) Solutions o f the two-dimensional diffusion equations, Trans.
A m er. Geophys. Union, 36, No. 2, 395 -405.
Rozanov, Yu. A., (1967) Stationary R a n d o m Processes, Holden-Day (tr. by A.
Feinstein), San Francisco, Cal.
Saffman, P. G., (1960) On the effect of the molecular diffusivity in turbulent
diffusion, J. F luid M ech., 8, No. 2, 2 7 3 - 283; (1962a) Some aspects o f the
effects o f the molecular diffusivity in turbulent diffusion, M ecanique de la
turbulence (Coll. Intern, du CNRS a Marseille), Paris, Ed. CNRS, 5 3 - 6 2 ;
(1962b) The effect o f wind shear on horizontal spread from instantaneous
ground source. Quart. J. R oy. Met. Soc., 88, No. 378, 382 - 3 9 3 ;( 1 9 6 3 ) An
approxim ate calculation o f the Lagrangian auto correlation coefficient for
stationary homogeneous turbulence, Appl. Scient. Res., A l l , No. 3,
2 4 5 -255.
Sakipov, Z. B., and D. Zh. Temirbayev, (1965) On the ratio of eddy diffusivities
for m o m en tu m and heat in a free turbulent jet, in Heat- and Mass Transfer
(Proc. 2nd All Soviet Union Conf., ed. by A. V. Lykov and B. M.
Sm ol’skiy), Minsk, Publ. House “ Nauka i Tckhnika,” 407 413.
Salwen, H., and C. E. Grosch, (1968) Stabihty o f Poiseuille How in a circular
pipe. Bull. A m er. Phys. Soc., 13, No. 5, 814.
Sani, R., (1964) On the non-existence o f subcritical instabilities in fluid layers
heated from below, / F luid Mech., 20, No. 2, 3 1 5 - 3 1 9 .
Sato, H., (1959) F urther investigation on the transition o f two-dimensional
separated layer at subsonic s p e e d s ,/. Phys. Soc. Japan, 14, 1 7 9 7 - 1 8 1 0 ;
(1960) The stability and transition o f a two-dimensional j e t , / F luid Mech.,
7, No. 1 , 5 3 - 8 0 .
Sato, H., and 0 . Okada, (1966) The stabihty and transition o f an axisymmetric
wake, / F luid M ech., 26, No. 2, 2 3 7 - 2 5 3 .
73 4 BIBLIOG RA PHY
Sato, H., and F. Sakao, (1964) An experimental investigation o f the instability

o f a two-dimensional jet at low Reynolds numbers, J. F luid M ech., 20, No.
2, 3 3 7 - 3 5 2 .
Scanlon, J. W., and L. A. Segel, (1967) Finite amplitude cellular convection
induced by surface t e n s i o n , / F luidM ech ., 30, No. 1, 1 4 9 - 1 6 2 .
Schade, H., (1962) Z um Stabilitatstheorie axialsymmetrischer Parallel-
stromungen, Ing.-Arch., 31, No. 5, 3 0 1 —316; (1964) C ontribution to the
nonlinear stability theory o f inviscid shear layers, Phys, Fluids, 7, No. 5,
623-628.
Schensted, I. V., (1960) C ontributions to the Theory o f H ydrodynam ic
Stability, Ph. D. Thesis, University of Michigan.
Schiller, L., (1932) Strom ung in Rohren, H andbuch der E xperim entalphysik, 4,
No. 4; (1934) Neue quantitative Versuche zur T urbulenzentstehung, Zs.
angew. Math. M ech., 14, No. 1, 3 6 —48.
Schlichting, H., (1933a) Zur Entstehung der Turbulenz bei der Plattenstrom ung,
Nachr. Ges. Wiss. G ottingen, Math.-phys. K l., 160—198; (1933b) Berech-
nung der Anfachung kleiner Storungen bei der Plattenstrom ung, Zs. angew.
M ath. M ech., 13, No. 3, 1 7 1 - 1 7 4 ; (1935a) Amplitudenverteilung und
Energiebilanz der kleinen Storungen bei der Plattengrenzschicht, Nachr.
Ges. Wiss. G ottingen, Math.-phys. K l., 1, 4 7 - 7 8 ; (1935b) Turbulenz bei
W ^m eschichtung, Z. angew. M ath. M ech., 15, No. 6, 3 1 3 - 3 3 8 ; (1936)
Experimentelle Untersuchungen zum Rauhigkeitsproblem,/wg.-ylrc/z. 7, No.
1, 1—34; (1 9 59 ) E ntstehung der Turbulenz, H andbuch der P hysik, Berlin;
(1960) B oundary L a yer Theory (4th ed.), McGraw-Hill Book Co., New York.
Schliiter, A., D. Lortz, and F. H. Busse, (1965) On the stability o f steady finite
am plitude co n v e ctio n ,/. F lu id M ech ., 23, No. 1, 1 2 9 - 1 4 4 .
Schmidt, W., (1917) Der Massenaustausch bei der ungeordneten Strom ung in
freier L uft und seine Folgen, Sitzungsber. A kad. Wiss. Wien, Math.-nat. K l.
(2a), 126, No. 6, 7 5 7 - 8 0 4 ; (1925) Der Massenaustausch in freier Luft und
verwandte Erscheinungen, Hamburg, Grand.
Schneck, P., and G. Veronis, (1967) Comparison of some recent experimental
and numerical results in Benard convection, Phys. Fluids, 10, No. 5,
927-930.
Schoenherr, K. E., (1932) Resistance o f flat surfaces moving through a fluid,
Trans. Soc. Nav. Arch. Marine E n g , 40, 2 7 9 - 3 1 3 .
Schubauer, G. B., and P. S. Klebanoff, (1956) Contributions on the mechanics
o f boundary-layer transition, A dv. Com, A eronaut. R ep. N o. 1289.
Schubauer, G. B., and H. K. Skramstad, (1947) Laminar boundary-layer
oscillations and stabiHty o f laminar flow, J. A eronaut. Sci., 14, No. 2,
6 9 -78.
Schubauer, G. B,, and C. M. Tchen, (1959) T urbulent flow. T urbulent F low and
H eat Transfer (ed. by C. C. Lin), Princeton Univ. Press, 7 5 —195.
Schubert, G., and G. M. Corcos, (1967) The dynamics o f turbulence near a wall
according to a linear m o d e l,/. F lu id M ech ., 29, No. 1, 1 1 3 - 1 3 5 .
B IBLIOG RA PHY 735
Schultz-Grunow, F., (1940) Neues Reibungswiderstandgesetz fiir glatte Flatten,

L u ftfahrtforsch., 17, No. 8, 2 3 9 - 2 4 6 .
Schultz-Grunow, F., (1967) The stability o f circular Couette flow w ith respect
to two-dimensional perturbations. F luid D ynam ics Transactions, Vol. 3 (ed.
by W. Fiszdon, P. K ucharczyk, W. J. Prosnak) Panstwowe wydawnictwo
naukow e, Warszawa, 8 3 —93.
Schwarz, K. W., B. E. Springett, and R. J. Donelly, (1964) Modes o f instabiHty in
spiral flow between rotating cylinders, J. F luid M ech., 20, No. 2, 2 8 1 - 2 8 9 .
Schwarz, W. H., and W. P. Cosart, (1961) The two-dimensional turbulent
wall-jet, J. F luidM ech., 10, No. 4, 4 8 1 —495.
Segel, L. A., (1962) The nonlinear interaction o f tw o disturbances in the thermal
convection problem , J. F luid M ech., 14, No. 1, 9 7 —114; (1965) The
nonlinear interaction o f a finite num ber o f disturbances to a layer o f fluid
heated from below, J. F luid M ech., 21, No. 2, 3 5 9 - 3 8 4 ; (1966) Non-linear
hydrodynam ic stability theory and its applications to therm al convection
and curved flows. N on-equilibrium Theorm o-dynam ics, Variational
Techniques, and S ta b ility (ed. b y R. J. Donnelly, R. Herman, and I.
Prigogine), Chicago Univ. Press, 1 6 5 - 1 9 7 .
Segel, L. A., and J. T. Stuart, (1962) On the question o f the preferred
mode in cellular therm al convection, / . F luid M e c h , 13, No. 2,
289-306.
Selig, F., (1964) Variational principle for Rayleigh-Taylor instability, Phys.
Fluids, 7, No. 8, 1 1 1 4 - 1 1 1 6 .
Sellers, W. D., (1962) A simplified derivation o f the diabatic wind profile, J.
A tm o sp h . Sci., 19, No. 2, 1 8 0 -1 8 1 .
Serrin, J., (1959) On the stability o f viscous fluid motions, R at. Mech.
A n a l, 3, No. 1, 1 - 1 3 .
Sexl, T., (1927a) Zur Stabilitatsfrage der Poiseuilleschen und Couetteschen
Strom ung, A nn. P hysik, 83, 8 3 5 - 8 4 8 ; (1 9 27 b ) Uber dreidimensionale
Storungen der Poiseuilleschen Strom ung, P hysik, 84, 8 0 7 - 8 2 2 .
Sexl, T., and J. Spielberg, (1958) Zum Stabilitatsproblem der Poiseuilleschen
Strom ung, A cta Phys. A ustr., 12, No. 1, 9 - 2 8 .
Shcherbakova, L. F., (1949) Some m ethods o f characterizing tu rbulent exchange
in terms o f observational data for the lower atmosphere, T rudy Glavn.
Geofiz. O b serv.,N o. 16(78), 2 5 - 5 1 .
Shen, S. F., (1954) Calculated amplified oscillations in plane Poiseuille flow and
Blasius flows, / . A eronaut. Sci., 21, No. 1, 6 2 - 6 4 ; (1964) Stability o f
laminar flows. Theory o f Lam inar Flow s (ed. by F. K. Moore), Princeton
Univ. Press, 719 - 8 5 3 .
Sheppard, P. A., (1947) The aerodynamic drag o f the ea rth ’s surface and the
value o f von K arm an’s constant in the lower layer, Proc. R o y . S o a , A188,
No. 1 0 1 3 , 2 0 8 - 2 2 2 .
Sherwood, T. K., (1950) Heat transfer, mass transfer and fluid friction
relationships in turbulent flow, Ind. Eng. Chem ., 42, No. 10, 2 0 7 7 - 2 0 8 4 .
736 BIBLIOGRA PHY
Shiotani, M., (1955) On the fluctuation o f the tem perature and turbulent
structure near the ground, J. M eteor. Soc. Japan, 4 0 , No. 3,
1 1 7 - 1 2 3 ; (1962) The relationship between wind profiles and stabilities
o f the air layer in the outskirts o f the city, J. M eteor. Soc. Japan, 40,
No. 6, 3 1 5 - 3 2 9 .
Sibulkin, M., (1962) Transition from turbulent to laminar pipe flow, Phys.
Fluids, 5, No. 3, 2 8 0 - 2 8 4 .
Silveston, P. L., (1958) Warmedurchgang in waagerechten Fliissigkeitsschichten,
I - I I , Forsch. G ebiete Ingenieurwes., 24, 2 9 —32; 5 9 —69; (1963)
Convectional heat transport b etw een rigid horizontal boundaries, Phys.
Fluids, 6, No. 2, 3 1 3 - 3 1 4 .
Simmons, L. F. G., and C. Salter, (1938) An experimental determ ination o f the
spectrum o f turbulence,/V oc. R o y. Soc., A 165, No. 920, 7 3 - 8 9 .
Sleicher, C. A., Jr., (19 5 8 ) E xperim ental velocity and temperature profiles for air
in turbulent pipe flow, Trans. A m er. Soc. Mech. Engrs., 80, No. 3,
693-704.
Slutskiy, E. E., (1938) Sur les fonctions aleatoires presque periodiques et sur la
decom position des fonctions aleatoires stationnaires en com posantes.
Actual. Scient. Industr., No. 738, 3 3 —55.
Smith, D. W., and J. H. Walker, (1959) Skin-friction measurem ents in
incompressible flow, A eronaut. Space A d m ., R eport R-26.
Smith, F. B., (1957) The diffusion o f smoke from a continuous elevated point
source into a tu rbulent a tm o sp h e re ,/. Fluid M ech., 2, No. 1, 4 9 - 7 6 ; (19 6 5)
The role o f wind shear in horizontal diffusion o f am bient particles, Quart. J.
R o y . M eteor. Soc., 91, No. 389, 3 1 8 - 3 2 9 (see also Discussion, Quart. J.
R o y. Meteor. Soc. (1966), 9 2 , No. 3 9 3 , 4 2 7 - 4 3 0 ) .
Snyder, H. A., (1968a) Stability o f rotating Couette flow, I. Asymmetric
waveforms. Phys. Fluids, 11, No. 4, 7 2 8 —734; (1968b) Stability o f rotating
Couette flow. II. Comparison w ith numerical Phys. Fluids, 11, No.
8, 1 5 9 9 - 1 6 0 5 .
Snyder, H. A., and S. K. F. Karlsson, (1965) N onaxisymmetric modes o f
secondary flow, Bull. A m . Phys. Soc., 10, 24.
Snyder, H. A., and R. B. Lam bert, (1966) Harmonic generation in Taylor
vortices between rotating cylinders. J. F luid M ech., 26, No. 3, 5 4 5 —562.
Son, J. S., and T. J. H anratty, (1967) Limiting relation for the eddy diffusivity
close to a ^ 2\\,A .I.C h .E . J., 13, No. 4, 6 8 9 - 6 9 6 .
Soprunenko, 1. P., (1 9 6 5) The stability o f je t flows, Izvestiya A N S S S R , Ser.
M e k h .,N o .4 ,3 1 -3 5 .
Sorger, P., (196 6 ) Uber ein Variationsproblem aus der nichtlinearen
Stabilitatstheorie zaher inkompressibler Stromungen. Zs. angew. Math.
Phys. 17, No. 2, 2 0 1 - 2 1 6 ; (1 96 7 ) Ein hinreichenes Stabilitatskriterium fiir
C ouette-Strom ungen, Zs. angew. Math, u nd M ech., 47, Sonderheft,
T 12 6 -T 1 2 8 .
Sorokin, V. S., (1 9 5 3 ) Variational principle in the theory o f convection, Prikl.
Mat. M ekh., 17, No. 1, 3 9 - 4 8 ; (1954) Stationary m otions o f a fluid heated
BIBLIOGRAPHY 737
from below, Prikl. Mat. M ekh., 18, No. 2, 1 9 7 - 2 0 4 ; (19 6 1 ) Nonlinear

effects in bounded flows at the Reynolds numbers near to the critical one,
Prikl. Mat. M ekh., 25, No. 2, 2 4 8 - 2 5 8 .
Southwell, R. V., and L. Chitty, (1930) On the problem o f hydrodynam ic
stability, I. Uniform shearing m otion in a viscous fluid, P hil Trans. R o y.
Soc. London, A 229, 2 0 5 - 2 8 3 .
Spalding, D. B., (1961) A single formula for the “ law o f the wall,” J. A ppl.
M ech., 28E, No. 3, 4 5 5 - 4 5 7 ; (1963) Heat transfer betw een a plate and the
tu rbulent flow o f a fluid, Inzh.-Fiz. Zhurn., 6, No. 3, 2 1 - 3 3 ; (1964) A
unified theory o f friction, heat transfer and mass transfer in the turbulent
boundary layer and wall jet, A ero. Res. Counc.y Rep. No. 25, 925; (1967)
Theories o f the turbulent boundary layer, Appl. Mech. R ev., 20s; No. 8,
735-740.
Spalding, D. B., and C. L. V. Jayatillaka, (1968) A survey o f the theoretical and
experimental inform ation on the resistance o f the laminar sublayer to heat
and mass transfer. Heat- and Mass Transfer (Proc. 2nd All Soviet Union
Conf., ed. by A. V. Lykow and B. M. Sm ol’skiy), Vol. 2, Nauka i T ekhnika
Publ. House, Minsk, 2 3 4 - 2 6 4 .
Sparrow, E. M., R. J. Goldstein, and V. K. Jonsson, (1964) Thermal instability in
a horizontal fluid layer: effect o f boundary condition and nonlinear
temperature profile, / F luid Mech., 18, No. 4, 5 1 3 - 5 2 8 .
Sparrow, E. M., W. D. Munro, and V. K. Jonsson, (1964) Instability o f the flow
between rotating cylinders: the wide-gap problem , J. F luid M ech., 20, No.
1 ,3 5 -4 6 .
Spiegel, E. A., (1962) On the Malkus theory of turbulence, M ecanique de la
turbulence (Colloq. Intern, du CNRS a Marseille), Paris, Ed. CNRS
1 8 1 -2 0 1 .
Squire, H. B., (1933) On the stabiUty o f the three-dimensional disturbances of
viscous flow between parallel walls, Proc. R o y. Soc., A142, No. 847,
6 2 1 - 6 2 8 ; (1948) Reconsideration o f the theory o f free turbulence, P/z//.
Mag. 39, No. 288, 1 - 1 4 .
Squire, W., (1964) T urbulent heat and mass transfer in sm ooth pipes, Int. J.
Heat and Mass Transfer, 7, No. 10, 1 0 6 9 - 1 0 7 6 .
Starr, V. P., (1968) Physics o f Negative Viscosity Phenom ena, McGraw-Hill
Book Co., New York-St. Louis-San Francisco-Toronto-London-Sydney.
Steinman, H., (1956) The stabiUty o f viscous flow between rotating cylinders,
Quart. A ppl. M ath., 14, No. 1, 2 7 - 3 3 .
Sternberg, J., (1962) A theory for the viscous sublayer o f a turbulent f lu id ,/.
F luid M ech., 13, No. 2, 2 4 1 - 2 7 1 ; (1965) The three-dimensional structure
o f the viscous s>\xh\dyQx,AGARDograph 97.
Stewart, R. W., (1956) Irrotational m otion associated with free turbulent flows,
J. F luid M ech., 1, No. 6, 5 9 3 —606; (1959) The problem o f diffusion in a
stratified fluid, A dv. G eophys., 6 (Atm ospheric diffusion and air pollution),
3 0 3 -311.
73 8 BIBLIOG RA PHY
Stuart, J. T., (1958) On the nonlinear mechanics o f hydrodynam ic stabiUty, J.

F luid M ech., 4, No. 1, 1 - 2 1 ; (1960) On the nonlinear mechanics o f wave
disturbances in stable and unstable parallel flows. Part 1. The basic
behaviour in plane Poiseuille flow, /. F luid M ech., 9, No. 3, 3 5 3 - 3 7 0 ;
(1962) On three-dimensional nonlinear effects in the stability o f parallel
flow, A dvances A eronaut. Sci., Vol. 3, Pergamon Press,
Oxford-London-N. Y.-Paris, 1 2 1 - 1 4 2 ; (1963) Hydrodynamic Stability,
Lam inar B oundary Layers (ed. by L. Rosenhead), O xford Univ. Press,
L ondon, 4 9 2 - 5 7 9 ; (1964) On the cellular patterns in thermal convection, 7.
F luid M ech., 18, No. 4, 4 8 1 - 4 9 8 ; (1965) H ydrodynam ic stability,
Mech. Revs., 18, No. 7, 5 2 3 - 5 3 1 ; (1967) On finite am plitude oscillations in
laminar mixing la y ers,/. F luid Mech., 29, No. 3 , 4 1 7 - 4 4 0 .
S ubbotin, V. I., M. Kh. Ibragimov, and Ye. V. Nomofllov, (1963) Determinalion
o f the relation between the heat and m om entum transfer coefflcients,
A to m n a y a Energiya, 14, No. 4, 4 1 4 - 4 1 6 .
S utton, O. G., (1932) The theory o f eddy diffusion in the atmosphere,
R o y . Soc., A 135, No. 826, 1 4 3 - 1 6 5 ; (1934) Wind structure and evapora
tion in a turbulent atmosphere, Froc. R o y . Soc., A146, No. 858, 7 0 1 - 722;
(1936) The logarithmic law o f wind structure near the ground. Quart. J.
R o y. M eteor. Soc., 62, No. 271, 1 2 4 - 1 2 6 ; (1937) The logarithmic law of
wind structure near the ground. Quart. J. R o y. Meteor. Soc., 63, No. 275,
1 0 5 - 1 0 7 ; (1949) The application to micrometeorology of the theory of
turb u lent flow over rough surfaces. Quart. J. R o y . M eteor. Soc., 75, No.
326, 3 3 5 - 3 5 0 ; (1950) On the stabihty o f a fluid heated from below,
R o y . Soc., A 204, No. 1078, 2 9 1 -3 0 9 ',(1 9 5 3 ) M icro m eteo ro lo g y, McGraw-
Hill Book Co., N. Y.-Toronto-London.
Sverdrup, H. U., (1936) The eddy conductivity o f the air over a sm ooth snow
field, Geophys. Pub., 11, No. 7.
Swinbank, W. C., (1951a) The measurem ent o f vertical transfer o f heat and
w ater vapour and m o m en tu m in the lower atmosphere with some results, J.
M eteor., 8, No. 2, 1 3 5 - 1 4 5 ; (1951b) A sensitive vapour pressure recorder,
J. Scient. Instr., 28, No. 3, 8 6 - 8 9 ; (1955) An experimental study o f eddy
transports in the lower atmosphere, C.S.LR .O ., Div. M eteor. Phys., Tcchn.
Pap. No. 2, Melbourne; (1960) Wind profile in thermally stratified flow.
Nature, 186, No. 4723, 4 6 3 - 4 6 4 ; (1964) The exponential wind profile.
Quart. J. R o y . M eteor. Soc., 90, No. 384, 119 - 1 3 5 ; (1966) Discussion on a
paper: “ The exponential wind profile,'’ Quart. J. R o y. M eteor. Soc., 92, No.
393, 4 16- 4 26; (1968) A comparison between predictions o f dimensional
analysis for the constant-flux layer and observations in unstable conditions,
Quart. J. R o y . M eteorol. Soc., 94, No. 402, 460- 467.
Swinbank, W. C., and A. J. Dyer, (196 7 ) An experimental study in
micrometeorology, Quart. J. R o y . Meteor. Soc., 93, No. 398, 4 94- 500;
(1968) Micrometeorological Expeditions 1 9 6 2 - 1 9 6 4 , C .S.LR.O ., Div.
M eteor. Phys., Techn. Pap. No. 17, Melbourne.
B IBLIOG RA PHY 730
Synge, J. L., (1933) The stability o f heterogeneous liquids, Trans. R o y. Soc.

Canada, 27, No. 1, 1—18; (1938) H ydrodynsim ic sidbiliiy, Sem icentennial
publications o f the A m er. M ath. Soc., 2 (Addresses), 2 2 7 —269.
S yono, S., and M. H amuro, (1962) Notes on the wind-profile in the lower layer
o f a diabatic atmosphere, / M eteor. Soc. Japan, ser. II, 40, No. 1, 1 - 1 2 .
Szablewski, W., (1968) Turbulente Parallelstrdmungen, Zs. angew. M ath. M ech.,
48, No. 1 , 3 5 - 5 0 .
Takeda, K., (1951) On the atmospheric turbulence, II. On the generalized
exponential fo rm u la ,/. Meteor. Soc. Japan, 29, No. 7, 2 4 3 - 2 4 5 .
Takeuchi, K., (1961) On the structure o f the turbulent field in the surface
b o undary layer, J. Meteor. Soc. Japan, 39, No. 6, 3 4 6 - 3 6 7 ; (1962) On the
nondim ensional rate o f dissipation o f turb u len t energy in the surface
boundary layer, J. M eteor. Soc. Japan, 40, No. 3, 1 2 7 - 1 3 5 .
Takeuchi, K., and O. Y okoyam a, (1963) The scale o f turbulence and the wind
profile in the surface boundary layer, J. M eteor. Soc. Japan, 41, No. 2,
108-117.
Tam, C. K. W., (1967) A note on disturbances in slightly supercritical plane
Poiseuille flo w ,/. F luidM ech ., 30, No. 1, 1 7 - 2 0 .
Tam, K. K., (1 9 6 8 ) On the asym ptotic solution of the Orr-Sommerfeld equation
by the m eth o d o f multiple-scales,/. F luid M ech., 34, No. 1, 1 4 5 - 1 5 8 .
Tani, I., (1967) Review o f some experim ental results on boundary-layer
transition, Phys. Fluids^ 10, Supplem ent (No. 9, Part II), SI 1 -S 1 6 .
Tani, I., and H. K om oda, (1962) Boundary layer transition in presence of
streamwise vortices, J. A erosp. Sci., 2 9 , 4 4 0 - 4 4 4 .
Tanner, C. B., and W. L. Pelton, (1960) Potential evapotranspiration estimates
by the approxim ate energy balance m eth o d o f Penman, J. G eophys.
Research, 65, No. 10, 3 3 9 1 - 3 4 1 3 .
Tatsumi, T., (1952) Stability o f the laminar inlet-flow prior to the form ation of
Poiseuille regime, I - I I , J Phys. Soc. Japan, 7, 4 8 9 - 4 9 4 , 4 9 5 - 5 0 1 .
Tatsumi, T., and K. G otoh, (1960) The stabiUty o f free boundary layers between
two uniform s trea m s,/. Fluid. M ech., 7, No. 3, 4 3 3 - 4 4 1 .
Tatsumi, T., and T. Kakutani, (1958) The stabihty o f a two-dimensional laminar
j e t , / F luid M ech., 4, No. 3, 2 6 1 - 2 7 5 .
Taylor, G. I., [All references by Taylor, with the exception o f 1959 may be also
found in the book, G. I. Taylor, Scientific Papers, Vol. 2, Cambridge
University Press, I960.] (1915) E ddy m o tio n in the atmosphere, Phil.
Trans. R o y. Soc., A 215, 1 - 2 6 ; (1916) Conditions at the surface o f a h o t
body exposed to the wind, Brit. A dv. Com. Aero. R ep. and M em or., 272;
(1921) Diffusion by continuous movements, Proc. L o n d o n Math. Soc. (2),
20, 196—211; (1923) Stability o f a viscous liquid contained between two
rotating cylinders, Phil. Trans. R o y. Soc. L ondon, A223, 2 8 9 - 3 4 3 ; (1931)
Effect o f variation in density on the stability o f superposed streams o f fluid,
Proc. R o y . Soc., A132, No. 820, 4 9 9 - 5 2 3 ; (1932) The transport o f
vorticity and heat through fluids in turbulent m otion, Proc. R o y . Soc.,
A 135, No. 828, 6 8 5 - 7 0 6 ; (1935a) Statistical theory o f turbulence, I - l l l ,

Proc. R o y. Soc., A 151, No. 874, 4 2 1 - 4 6 4 ; (1935b) Statistical theory of
turbulence. IV. Diffusion in a turbulent air stream, Proc. R o y. Soc., A 151,
No. 874, 4 6 5 —4 78; (1935c) Distribution o f velocity and tem perature
between concentric rotating cylinders, Proc. R o y . Soc., A 151, No. 874,
4 9 4 - 5 1 2 ; (1936a) Statistical theory o f turbulence. V. Effect o f turbulence
on boundary layer. Proc. R o y . Soc., A 156, No. 888, 3 0 7 —317; (1936b)
Fluid friction between rotating cylinders. I. T orque measurements, Proc.
R o y. Soc., A 157, No. 892, 5 4 6 - 5 6 4 ; (1938) Some recent developments in
the study of turbulence, Proc. 5 th Int. Congr. A ppl. M ech., Cambridge
(USA), 2 9 4 - 3 1 0 ; (1953) Dispersion o f soluble m atter in solvent flowing
slowly through a tube, Proc. R o y . Soc., A 219, No. 1137, 1 8 6 - 2 0 3 ; (1954a)
The dispersion o f m atter in turbulent flow through a pipe, Proc. R o y. Soc.,
A 223, No. 1155, 4 4 6 - 4 6 8 ; (1954b) Conditions under which dispersion o f a
solute in a stream o f solvent can be used to measure molecular diffusion,
Proc. R o y . Soc., A 2 2 5 ,N o . 1163, 4 7 3 —4 77; (1959) The present position in
the theory of tu rbulent diffusion, G eophys., 6 (Atmospheric diffusion
and air pollution), 1 0 1 - 1 1 2 .
Taylor, R. J., (1952) The dissipation of kinetic energy in the lowest layers o f the
atmosphere. Quart. J. R o y. Meteor. Soc., 78, No. 336, 1 7 9 - 1 8 5 ; (1956a)
Some measurements o f heat flux at large negative Richardson num ber,
Quart. J. R o y . M eteor. Soc., 82, No. 351, 8 9 —91; (1956b) An electronic
analogue o f the saturated vapour pressure curve o f water, J. Scient. Instr.,
33, No. 11, 4 3 6 - 4 3 8 ; (1958) A linear unidirectional anem om eter o f rapid
response, J. Scient. Instr., 35, No. 1, 4 7 —52; (1960a) Similarity theory in
the relation between fluxes and gradients in the lower atmosphere, Quart. J.
R o y . M eteor. Soc., 86, No. 367, 6 7 —78; (1 9 6 0 b ) The wind profile in the
atmospheric surface layer, Izvestiya A N S S S R , Ser. geofiz., No. 12,
1821 —1827; (1963) The effect o f instrum ental inertia on measurem ent o f
the turbulent flux o f water vapour, A ustr. J. A ppl. Sci., 14, No. 1, 2 7 - 3 7 .
Taylor, R. J., and A. J. Dyer, (1958) An instrum ent for measuring evaporation
from natural surfaces. Nature, 181, No. 4 6 0 6, 4 0 8 —409.
Taylor, R. J., and E. K. Webb, (1955) A mechanical com puter for micrometeor-
ological research, C .S.I.R.O ., Div. Meteor. P hys., Techn. Pap. No. 6,
Melboume.
Tem pel’man, A. A., (1 9 6 2 ) Ergodic theorems for homogeneous generalized
random fields and hom ogeneous random fields on groups. Lithuan. Mat.
Shorn. 2, No. 1.
Thomas, L. H., (1953) The stability o f plane Poiseuille i\o ^ ,P h y s . R ev., 91, No.
5, 7 8 0 - 7 8 3 .
Thomas, D. B., and A. A. Townsend, (1957) T urbulent convection over a heated
horizontal surface, J. F luid M ech., 2, No. 5, 4 7 3 —492.
Thom pson, B. G. J., (1963) The Calculation o f Turbulent B oundary Layers, Ph.
D. Dissertation, Cambridge University; (1965) A new two-parameter family
BIBLIOGRAPHY 741
o f mean velocity profiles for incompressible turbulent b ou ndary layers on

sm ooth walls, Res. Counc. R ep. No. 26, 830.
Thom pson, H. A., and H. H. Sogin, (1966) Experiments on the onset o f thermal
convection in horizontal layers o f gases, J. F luid M ech., 24, No. 3,
4 5 1 -479.
Thom pson, N., (1965) Short-range vertical diffusion in stable conditions. Quart.
J. R o y . M eteor. Soc., 91, No. 388, 1 7 5 - 1 8 3 (also Discussion, Quart. J.
R o y . Meteor. Soc. (1967), 93, No. 397, 3 8 3 - 3 8 6 .
T hornthw aite, C. H., and P. Kaser, (1943) Wind-gradient observations, Trans.
A m er. Geophys. Union, 24, No. 1, 166—182.
Tien, C. L., (1964) A note on distributions o f tem perature and eddy diffusivity
for heat in turb u len t flow near a wall, Zs. angew. Math. Fhys., 15, No. 1,
6 3 -66.
Tien, C. L., and D. T. Wasan, (1963) Law o f the wall in turbulent channel flow,
Phys. Fluids, 6, No. 1, 1 4 4 - 1 4 5 .
Tietjens, O., (1925) Beitrage zur E ntstehung der Turbulenz, Zs. angew. Math.
M ech., 5, No. 3 , 2 0 0 - 2 1 7 .
Tippelskirch, H., (1956) Uber Konvektionszellen, insbesondere in fliissigem
Schwefel, Beitr. Phys. A tm o sp h ., 29, No. 1, 3 7 —54.
Tollmien, W., (1926) Berechnung tu rbulenter Ausbreitungsvorgange, Zs. angew.
M ath. M ech., 6, No. 6, 4 6 8 - 4 7 8 ; (1929) Uber die Entstehung der
Turbulenz, Nachr. Ges. Wiss. G ottingen, Math.-Phys. K l., 2 1 - 4 4 ; (1930)
Beitrag zur Theorie der Turbulenz, Proc. 3rd Int. Congr. A ppl. M ech.,
Stockholm, 1, 105—108; (1935) Ein allgemeines Kriterium der Instabihtat
laminarer Geschwindigkeitsverteilungen, Nachr. Ges. Wiss. G ottingen,
Math.-Phys. K L , 7 9 - 1 1 4 ; (1947) Asymptotische Integration der Storungs-
differentialgleichung ebener laminarer Strom ungen bei hohen Reynolds-
schen Zahlen, Zs. angew. Math. M ech., 2 5 /2 7 , 3 3 - 5 0 , 7 0 —83.
Tollmien, W., and J. Pretsch, (1946) Instabilitat laminar Grenzschichten,
G ottinger Monogr. Uber die deutschen L uftfa hrtforsch , 3, Gottingen.
Tolman, R. C., and P. C. Fine, (1948) On the irreversible production o f entropy,
Rev. Mod. Phys., 20, No. 1, 51—77.
Townsend, A. A., (1947) The measurem ent o f double and triple correlation
derivatives in isotropic turbulence, Proc. Cambr. Phil. Soc., 43, No. 4,
560 570; (1951) The diffusion o f heat spots in isotropic turbulence,
R o y. Soc., A 209, No. 1098, 4 1 8 - 4 3 0 ; (1954) The diffusion behind a line
source in homogeneous turbulence, Proc. R o y . Soc., A 224, No. 1159,
4 8 7 - 512; (1956) The Structure o f T urbulent Shear F lo w , Cambridge Univ.
Press; (1958) The effects o f radiative transfer on turbulent flow o f a
stratified fluid, J. F luid M ech., 4, No. 4, 3 6 1 —375; (1959) Tem perature
fluctuations over a heated horizontal surface, J. Fluid M ech., 5, No. 2,
2 0 9 - 2 4 1 ; (1961) Equilibrium layers and wall tu rb u le n c e ,/. F luid M ech.,
11, No. 1, 9 7 - 1 2 0 ; (1962a) Natural convection in the earth ’s boundary
layer, Quart. J. R o y. M eteor. Soc., 88, No. 375, 5 1 - 5 6 ; (1 9 6 2 b ) Remarks
742 BIBLIOGRA PHY
on the Malkus theory o f tu rbulent flow , M ecunique de la turbulence (CoW.

Intern, du CNRS a Marseille), Paris, Ed. CNRS, 1 6 7 - 1 8 0 ; (19 6 6) The
mechanism o f entrainm ent in free turbulent flows, / F luid M ech., 26, No.
4 , 6 8 9 715.
Ts’ai, Ko-yeng, (1961) Heat transfer in a turbulent flow o f a fluid at high
Prandtl numbers, Inzh.-Fiz. Z h u rn .,4 , No. 8, 2 0 - 2 9 .
Tsvang, L. R., (1960) Measurement o f frequency spectra o f tem perature
fluctuations in the atmospheric surface layer, Izvestiya A N S S S R , Ser.
g e o f iz .,N o .8 , 1 2 5 2 -1 2 6 2 .
Tseytin, G. Kh., (1963) Propagation o f an im purity from a high-altitude source,
Trudy Leningr. Gidromet. In-ta, No. 15, 10 -3 1 .
Tyldesley, J. B., and C. E. Wallington, (1965) The effect o f wind shear and
vertical diffusion on horizontal dispersion. Quart. J. R o y , Meteor. Soc., 91,
No. 388, 1 5 8 - 1 7 4 (also Discussion: Quart. J. R o y. Meteor. Soc. (1966),
92, No. 393, 427 -43 0 ).
Uberoi, M. S., and S. Corsin, (1953) Diffusion o f heat from a line source in
isotropic turbulence,TVar. A dv. Com, A eronaut. R e p ., No. 1142.
Uhlenbeck, G. E., and L. S. Ornstein, (1930) On the theory o f Brownian
m o tio n, Phys. Rev., 36, No. 3, 8 2 3 - 8 4 1 .
Ukhovskiy, M. R., and V. I. Yudovich, (1963) Equations of stationary
c o n v e c t i o n , M atem . M ekh., 27, No. 2, 2 9 5 —300.
Vager, B. G., (1966) The effect o f turbulent energy diffusion in the
semiempirical model o f the atmospheric surface layer, Izvestiya A N S S S R ,
Ser.fiz. atmos. okeana, 2, No. 9, 9 2 0 - 9 2 7 .
Van Atta, C. W., (1966) Exploratory measurements in spiral turbulence, J.
Fluid. M ech., 25, No. 3 , 4 9 5 - 5 1 2 .
Van Driest, E. R., (1956) On turbulent flow near a wall, J. Aeronaut. ScL, 23,
No. 11, 1 0 0 7 - 1 0 1 1 ; (1959) Convective heat transfer in gases. T urbulent
Flow s and H eat Transfer (ed. by C. C. Lin), Princeton Univ. Press, 3 3 9 - 4 2 7 .
Vasudeva, B. R., (1967) Boundary layer instability experiment with localized
disturbance, J. F luid M ech., 29, No. 4, 745 764.
Vaughan, L. M., (1961) The prediction o f atmospheric diffusion by using an
eddy diffusivity based on the vertical transfer of h e a t , / . M eteor., 18, No. 1,
4 3 -49.
Velikanov, M. A., (1946) D ynam ics o f River-Bed Flow s, G idrom eteoizdat Press,
Leningrad-Moscow.
Velte, W., (1962) Uber ein Stabilitatskriterium der H ydrodynam ik, Rat.
Mech. A n a l, 9, No. 1, 9 - 2 0 ; (1964) Stabilitatsvcrhalten und Verzweigung
stationarer Losungen der Navier-Stokesschen Gleichungen, Arch. R ational
Mech. Anal., 16, No. 2, 9 7 - 1 2 5 ; (1966) Stabilitat und Verzweigung
stationarer Losungen der Navier-Stokesschen Gleichungen beim
T?iy\orpvob\Qm, A rch R ational Mech. Anal., 22, No. 1, 1 - 1 4 .
Venezian, E., and B. H. Sage, (1961) T em perature gradients in turbulent gas
streams: effect o f viscous dissipation on evaluation o f total conductivity,
A .L C h.E . / , 7, No. 4, 6 8 8 - 6 9 2 .
BIBLIOGRAPHY 743
V enotte, P., (1958) Les paradoxes de la theorie continue de Tequation de la

chaleur, C om pte R endus Acad. ScL, 246, No. 22, 3 1 5 4 - 3 1 5 5 .
Veronis, G., (1966) Large-amplitude Benard convection, 7. F luidM ech., 26, No.
1 ,4 9 -6 8 .
Vulis, L. A., and V. P. Kashkarov, (1965) Theory o f Viscous F/w/c? Nauka
Press, Moscow.
Walowit, J., S. Tsao, and R. C. Di Prima, (1964) Stability o f flow between
arbitrarily spaced concentric cylindrical surfaces including the effect o f a
radial temperature gradient, / A p p l M ech., 31, No. 4, 585 593.
Walters, T. S., (1 96 4 ) The effcct o f diffusion in the mean wind direction in
atmospheric t u r b u l e n c e , 11, 119—124; (1965) Diffusion from
an elevated point sources into a turbulent atmosphere, M athem atika, 12,
4 9-57.
Walz, A., (1969) B oundary Layers o f F low and Temperature, M.I.T. Press, Cam
bridge, Mass.
Wandel, C. F., and 0 . Kofoed-Hansen, (1962) On the Eulerian-Lagrangian
transform in the statistical theory o f turbulence, J, Geophys. Res., 67, No.
8, 3 0 8 9 - 3 0 9 3 .
Wasan, D. T., and C. R. Wilke, (1964) Turbulent exchange o f m om entum , mass,
and heat between fluid streams and pipe wall. Int. J. H eat and Mass
Transfer, 7, 8 7 - 9 4 .
Wasow, W., (1953) On small disturbances o f plane Couette flow, J. Res. Nat.
Bur. Standards, 51, 195—202.
Watson, J., (1960a) On the nonlinear mechanics o f wave disturbances in stable
and unstable parallel flows. Pt. 2. The development o f solution for plane
Poiseuille and for plane Couette flow, J. F luid M ech., 9, No. 3, 3 7 1 - 3 8 3 ;
(1960b) Three-dimensional disturbances in flow between parallel planes,
Proc. R o y. Soc., A 254, No. 1279, 5 6 2 - 5 6 9 ; (1962) On spatially growing
finite disturbances in plane Poiseuille flow, J. Fluid M ech., 14, No. 2,
211 - 22 1 .
Wazzan, A. R., T. T. O kamura, and A. M. Smith, (1967) Stability o f laminar
boundary layers at s e p a r a t i o n , Fluids, 10, No. 12, 2 5 4 0 - 2 5 4 5 .
Webb, E. K., (1958) Vanishing potential temperature gradients in strong
convection. Quart. J. R oy. Meteor. Soc., 84, No. 360, 1 1 8 - 1 2 5 ; (1960) An
investigation o f the evaporation from Lake Eucum bene, C .S.I.R.O ., Div.
M eteor. Phys., Techn. Pap. No. 10, Melbourne; (1962) Thermal convection
w ith wind shear, N ature, 193, No. 4818, 8 4 0 - 8 4 2 ; (1965) Aerial
microclimate. Mereoro/. Monographs, 6, No. 28, 27 -5 8 .
Webster, C. A. G., (1964) An experimental study o f turbulence in a
density-stratified shear flow, / Fluid. M ech., 19, No. 2, 2 2 1 - 2 4 5 .
Wieghardt, K., (19 4 0 ) Uber die Wirkung der Turbulenz au f den Umschlagspunkt,
Zs. angew. M ath. M ech., 20, No. 1, 5 8 - 5 9 ; (1944) Uber die turbulente
Strom ung in R ohr und langs der Platte, Zs. angew. Math. Mech., 24, No. 4,
294-296.
Willis, G. E., and J. W. Deardorff, (1967) Confirmation and renumbering o f the
744 BIBLIOGRAPHY
discrete heat tlux transitions o f Malkus, Phys. Fluids, 10, No. 9,

1861-1866.
Willmarth, W. W., and B o -JangT u, (1967) Structure of turbulence in the
boundary layer near the v^'dW,Phys. Fluids, 10, Supplem ent, S I 3 4 - S 137.
Wilson, B. W., (1960) N ote on surface wind stress over water at low and high
wind sp e e d ,/. Geophys. Res., 65, No. 10, 3 3 7 7 - 3 3 8 2 .
Witting, H., (1958) Uber den Einfluss der Strom linienkrum m ung au f die
Stabilitat laminarer Strom ungen, Arch. R at, M ech. A n a l, 2, No. 3,
2 4 3 -283.
W ortmann, F. X., (1955) U ntersuchung instabiler Grenzschichtschwingungen in
einem Wasserkanal mit der Tellurm ethode, 5 0 Jahre G renzschichtfor-
schung, Vieweg, Braunschweig, 4 6 0 - 4 7 0 .
Wu, J., (1968) Laboratory studies o f wind-wave in terac tio n s,/. F luid M ech., 34,
No. 1 , 9 1 - 1 1 2 .
Yaglom, A. M., (1948) H omogeneous and isotropic turbulence in viscous
compressible fluids, Izvestiya A N S S S R , Ser. geograf. i geofiz., 12, No. 6,
5 0 1 - 5 2 2 ; (1962) A n In tro d u ctio n to the Theory o f Stationary R andom
Functions, Prentice-Hall, Englewoods Cliffs, N. J.; (1965) Lagrangian
turbulence characteristics in a diabatic atmospheric surface layer and in
convective jets, Izvestiya A N S S S R , Ser. fiz. atmos. i okeana, 1, No. 2,
1 5 7 - 1 6 6 ; (1969) Horizontal turbu len t transport o f heat in the atmosphere
and the form o f the eddy diffusivity tensor. F luid D ynam ics Transactions,
Vol. 4 (ed. by W. Fiszdon, P. Kucharczyk, W. J. Prosnak), Panstwowe
w ydaw nictw o naukowe, Warszawa, 8 0 1 - 8 1 2 .
Yamada, H., (1964) On the stability to long waves o f two-dimensional laminar
] ti,R e p ts . Res. Inst. A p p l M ech., 12, No. 43, 1 - 1 5 .
Y am am oto, G., (1959) Theory o f turbulent transfer in nonneutral c o n d itio n s,/.
M eteor. Soc. Japan, 37, No. 2, 6 0 - 6 9 .
Y am am oto, G., and J. K ondo, (1 9 59 ) Effect o f surface reflectivity for long wave
radiation on tem perature profiles near the bare soil surface, Sci. R eports
T dhoku U niv., ser. V, Geophysics, 11, No. 1, 1—9.
Y am am oto, G., and A. Shim anuku, (1960) Numerical solution o f the equation
o f atmospheric turbulent diffusion, Sci. R ep o rts T dhoku Univ., ser V,
Geophysics, 12, No. 1, 2 4 - 3 5 ; (1964) The determ ination o f lateral
diffusivity in diabatic conditions near the ground from diffusion experi
m e n ts , / . A tm o sp h . Sci., 21, No. 2, 1 8 7 - 1 9 6 ; (1 9 6 6 ) T urbulent transfer in
diabatic co n d itio n s,/. M eteor. Soc. Japan, 44, No. 6, 3 0 1 —307.
conditions, / M eteor. Soc. Japan, 44, No. 6, 3 0 1 - 3 0 7 .
Y okoyam a, O., (1962a) On the thickness o f smoke plume in diabatic surface
layer, /. M eteor. Soc. Japan, ser. II, 40, No. 2, 8 3 - 9 2 ; (19 6 2 b) On the
contradiction and modification o f the equation o f diabatic wind p ro file,/.
M eteor. Soc. Japan, ser. 11,40, No. 6, 3 5 9 - 3 6 0 .
Y ordanov, D. L., (1966) On diffusion from a point source in the surface layer of
the atm osphere, Izvestiya A N SSSR , Ser. fiz. atmosf. okeana, 2, No. 6,
BIBLIOGRA PHY 74 5
5 7 6 - 5 8 4 ; (1967) On diffusion in the wind direction and some asym ptotic

formulae for surface layer atm ospheric diffusion, Izvestiya A N SSSR , Ser.
fiz. atmosf. okeana, 3, No. 8, 8 1 8 —8 26; (1968a) On some asym ptotic
formulae describing diffusion in the surface layer o f the atmosphere,
A tm o sp h . Environment^ 2, No. 2, 167—180; (19 6 8 b ) On turbulent
diffusion from a steady line source in the surface layer under unstable
stratification, Rev. Pure A ppL G eophys., 69, No. 1, 2 2 9 - 2 3 6 ; (1968c)
A dm ixture diffusion from linear source in the surface layer under unstable
stratification, Izvestiya A N S S SR , Ser. fiz. atmosf. okeana, 4, No. 6,
677-679.
Yoshihara, H., (1968) Sea-air interaction: A simplified m o d e l , / A tm o sp h . S e t,
25, No. 5, 7 2 9 - 7 3 5 .
Yudin, M. I., and M. Ye. Shevets, (1940) Stationary model of the height
distribution o f wind in a turbulent atmosphere, T rudy G I. G eofiz. Observ.,
No. 3 1 , 4 2 - 5 2 .
Yu, Yun-Sheng, and Sun Dah-Dhen (1 96 4 ) Stability o f viscous flow between
tw o rotating coaxial cylind ers,/. Franklin Inst., 277, No. 2, 1 4 0 - 1 4 9 .
Yudovich, V. I., (1965a) Stability o f steady flows o f viscous incompressible
D oklady A N S S S R , 161, No. 5, 1 0 3 7 - 1 0 4 0 ; (1 9 65 b ) An example of
generation o f the secondary stationary or periodic flow while losing the
stability by laminar flow o f viscous incompressible fluids, P n M Mat. M ekh.,
29, No. 3, 4 5 3 - 4 6 7 ; (1966a) The secondary flows and the instability o f the
fluid in the annulus between the rotating cylinders, Prikl. Mat. M ekh., 30,
No. 4, 6 8 8 —6 98; (1 9 6 6 b ) On the origin o f c o n v e c t i o n , Mat. M ekh.,
30, No. 6, 1 0 0 0 - 1 0 0 5 ; (1967a) Free convection and branching, Prikl. Mat.
M ekh., 31, No. 1, 1 0 1 - 1 1 1 ; (196 7 b ) On the stability o f convective flows,
Prikl. Mat. M ekh., 31, No. 2, 2 7 2 - 2 8 1 ; (1967c) An example o f the loss o f
stability and o f generation o f secondary fluid flow in closed vessel, Marem.
Sbornik, 74(1 1 6 ), No. 4, 5 6 5 - 5 7 9 .
Zagustin, A. I., (1938) Equations for the turbulent m otions of fluids, T rudy
Voronezh. Univ., 10, No. 2, 7 - 3 9 .
Zel’dovich, Ya. B., (1937) Limiting laws for turbulent flows in free convection,
Zh. Eksp. Teor. F iz., 7, No. 12, 1 4 6 3 - 1 4 6 5 .
Zilitinkevich, S. S., (1970) D ynam ics o f the A tm o sp h eric Planetary B oundary
L ayer, G idrom eteoizdat Press, Leningrad.
Zilitinkevich, S. S., and D. V. Chalikov, (1968a) On the determ ination o f the
universal wind and tem perature profiles in the surface layer o f the
atmosphere, Izvestiya A N S S S R , Ser. fiz. atmosf. i okeana, 4 , No. 3,
2 9 4 - 3 0 2 ; (1968b) On the com putations o f the vertical turb ulen t fluxes in
the surface layer o f the atm osphere from data o f profile observations,
Izvestiya A N S S S R ., Ser fiz. atmosf. i okeana, 4, No. 9, 9 1 5 - 9 2 9 .
profile observations, Izvestiya A N S S S R ., Sex fiz. a tm o s f i okeana, 4, No. 9,
915-929.
746 BIBLIOG RAPHY
Zilitinkevich, S. S., and D. L. Laykhtm an, (1965) Turbulent regime in the

atmospheric surface layer, Izvestiya A N S S S R ,S qx. atmosf. i okeana, 1, No.
2, 1 5 0 - 1 5 6 .
Zilitinkevich, S. S., D. L. L aykhtm an, and A. S. Monin, (1967) Dynamics o f the
atmospheric boundary layer, Izvestiya A N SSS R , Ser. fiz. atmosf. okeana, 3,
No. 3, 2 9 7 - 3 3 3 .
Zubkovskiy, S. L., (1962) Frequency spectra o f the horizontal wind-velocity
fluctuations in the atmospheric surface \diyQx, Izvestiya A N S S S R ,S qx.
geoiiZ ., No. 10, 1425—1433; (1967) Experimental study o f turbulence
characteristics in the atmospheric surface layer above the earth and the sea,
Candidate Dissertation, Inst. Phys. Atmosphere. Acad. Sci. USSR, Moscow.
Zubkovskiy, S. L., and L. R. Tsveng, (1966) On the horizontal tu rbulent heat
f l u x , 555/ ?, Ser.fiz. atmosf. okeana, 2, No. 12, 1 3 0 7 -1 3 1 0 .
Zukauskas, A. A., A. A. Slanciauskas, A. A. Pedisius, and R. V. Ulinskas,
(1968) Mass and m om en tu m transfer in a viscous boundary layer flow on a
Y\2iit,in H eat~ and Mass Transfer (Proc. 3rd All Soviet Union Conf., ed. by A.
V. Lykov and B. M. Sm ol’skiy), Vol. 1, Energiya Press, Moscow, 1 3 3 - 1 4 0 .
AUTHOR INDEX
Abramovich, G. N., 354, 357, 362, 632 Bobyleva, I. M., 409,410

Akhiezer, N. I., 225 Bochner, S., 219
AUen, C. M., 617,627 Bo Jang Tu, 283
Angus, D. E., 508 Borkowski, J., 451
Antanaitis, P., 659 Bosanquet, C. H., 660
Appleby, J. F., 406, 409, 410, 524 Bourret, R. C., 680
Aris, R., 619 ,6 2 8 ,6 3 8 Boussinesq, J., 59, 262, 264, 265, 606
ArnolM, V. I., 158,160 Bovsheverov, V. M., 505, 507
Bradley, E. P., 475, 479, 492, 507, 508,
Baines, W.D., 331 510, 518, 520
Bakewell, H. P., 283 Browand, F. K., 1 96,197,198
Baldwin, L. W., 3 6 3 ,5 76,5 88 Browne, N. E., 659
Barad, M. L., 455, 475, 478, 505, 517, Brummage, K. G., 659
588 ,659 ,67 6 Brundrett, E., 331
Barcnblatt, G. I., 412 Brushlinskaya, N. N., 168
Barnes, H.T., 75 Budyko, M. 1., 454
Batchelor, G. K., 27, 31, 143, 144, 385, Burgers, I. M., 48, 77
544, 545, 547, 560, 580, 600, 608, Burns, J.G ., 134
6 1 9 ,6 3 3 ,6 3 7 ,6 7 0 ,6 7 3 Burroughs, P. R., 331
Beguier, C , 373 Burton, R. A., 310
Bell, J. B., 331,332 Busch, N.E., 3 3 7 ,4 7 5 ,4 7 7 ,4 7 9 ,4 8 3 ,
Belyakova, V. K., 145 485, 488, 492, 493, 494, 497, 498,
Benney, D. J., 96, 121, 128, 201, 202 5 1 0 ,5 1 7 ,5 1 8 ,5 1 9 ,5 2 1 ,5 2 5
Berlyand,M. H., 366,662 Businger, J. A., 426, 452, 453, 454, 465,
Bernstein, A. B., 475, 477, 478, 479, 485 475, 479, 492, 505, 506, 507, 510,
Best, A. C , 455 518,520
Betchov, R., 73, 113, 123, 124, ) 28, 129, Busse, F. H., 180,181,1 85,1 86,1 89,1 90
130, 131, 132, 133, 134, 135, 1^6, 137, Bysova, N. L., 4 09 ,410
139, 142, 201, 202 Byutner, E. K., 659
Bhat,W. V., 146
Binder, G. H., 520 Calder, K. L., 6 1 3 ,6 1 4 ,6 6 1 ,6 6 3
Binnie, A. M., 619 Capriz, G., 102
Bird, R. B., 53 Carlson, F. D., 64
Birikh, P. V., 129,136 Carmi, S., 157
Birkhoff, G., 90, 254, 357 Carr, A. D., 282
Bisshopp, I'. E., 179 Carrier, G. F., 64
Bjorgum, O., 454 Case, K. M., 96, 120, 121, 125, 126, 138
Blackadar, A. K., 406, 409, 410, 450, 475, Cattaned, C., 680
4 7 7,485,498, 510 Catton, I., 110
Blasius, H., 310 Cavers, S. D., 337
Blinova, E. N., 40 9,410 Cermak, J. E., 475, 482, 484, 520, 560,
Blom,J., 337 5 6 6 ,5 6 7 ,6 7 1 ,6 7 4 ,6 7 6
747
748 AUTHOR INDEX
Chalikov, D. V., 4 7 3 ,4 7 5 ,4 7 8 ,4 8 0 ,4 8 3 , Davidson, B., 455, 468, 475, 486, 487,

484, 486, 490, 491, 492, 493, 498, 510, 505, 506,508, 509,517
5 1 1 ,5 1 2 ,5 1 7 Davies, D. R., 664, 668, 669
Chamberlain, A. C., 290, 294, 347 Davies, S. J., 147
Chandrasekhar, S., 19, 59, 72, 97, 100, Davis, R.W., 680
107, 109, 110, 111, 124, 170,678 Davis, S. H., 187 ,189,190
Chapman, D. R., 342 Deacon, E. L., 294, 295, 455, 468, 476,
Chapman, S., 53 492, 498,499, 5 0 4 ,5 1 0 ,5 1 7 ,5 1 8 ,
Charnock, H., 2 9 5 ,4 4 2 ,4 7 7 ,4 7 8 ,4 7 9 , 5 2 3 ,6 4 3 ,6 4 6 ,6 6 3
488, 492, 4 9 3 ,4 94,5 01 Deardorff, J. W., 1 11 ,129 ,168 ,181 ,
Chatwin,P. C , 560,651 4 7 9 ,4 9 6 ,4 9 7
C h e n ,C .C , 363 Debler,W. R., 107
Chen, M. M., 187 Defant, A., 386
Chen, T. S., 128,136 Deissler, R. G., 272, 282, 327, 329, 337,
Childs, W. H., 134 340, 342, 344, 365
Chilton, T. H., 344 Denton, E.B., 344, 345
Chitty, L., 128 De Vries, D. A., 337
Christopherson, D. G., I l l Dhawan, S., 323
Chou, P. Y., 377, 378 Dikiy, L. A., 9 6 ,1 2 0 ,1 2 1 ,1 2 4 ,1 2 5 ,
Chu,B. T., 6 5 ,6 9 129, 138, 160
Chuang, H., 4 7 5 ,4 8 2 ,5 2 0 Dil’man, V. V., 342, 344
Chudnovskiy, A. F., 486 Dipprey, D. F., 347
Qauser, F., 2 9 0 ,2 9 1 ,2 7 7 ,3 0 4 ,3 1 2 DiPrima, R.C., 9 8 ,1 0 2 ,1 0 4 ,1 0 5 ,1 0 7 ,
Coantic, M., 270, 277, 280, 282, 283, 309 176
Coker, E.G., 75 Donnelly, R. J., 103, 104, 168, 172, 173,
Colburn, A. P., 344 175,176
Coles, D., 77, 97, 101, 103, 106, 168, 176, Donovan, L. F., 342, 345
270, 277, 317, 318, 319, 324, 363, 364 D o ob,J.L ., 245,256
Collis, D. C , 541,588 Dorodnitsyn, A. A., 366, 451
Comolet, R., 76 Drake, R. M., 327
Comte-Bellot, G., 270, 277, 280, 281, 309 Drazin, P.G., 73,96, 117, 121, 122,
Corcoran, W. H., 337 123, 125, 126
Corcos,G. M., 145,283 Drimmel, J., 659
Corrsin, S., 315, 354, 362, 363, 505, 534, Dryden, H. L., 72, 78, 79, 85, 136
541, 559, 577, 578, 587, 588, 592, 633 Dumas, R., 280, 577
Cosart, W. P., 373 Duty, R. L., 102
Cowling, T. G., 53 Dvorak, F. A., 317,342
Cramer, H. E., 480, 492, 505, 506, 517, Dwyer, O.E., 337
518,519, 5 2 0 ,5 21,5 25 Dyer, A. J., 4 2 4 ,4 7 5 ,4 7 9 ,4 8 8 ,4 8 9 ,
Crawford, T. V., 489, 501, 508 492, 493, 494, 496, 498, 499, 501,
Criminale, W. O., Jr., 73,113, 123, 124, 5 0 5 ,5 0 6 ,5 0 7 ,5 0 8 ,5 1 0 ,5 1 7 ,5 1 8 ,
128, 129, 130, 131, 132, 133, 134, 135, 520
136, 137, 139, 142, 201 Dynkin, E. B., 582
Crocco, L., 312,318 Dyson, F .J., 125
Crow, S. C., 373
Csanady, G. T., 392, 393, 411, 583, 659 Eckert, E. R. G., 327
Curie, N., 139 Eckhaus,W., 165,192
Curtis, C .F ., 53 Eiffel, G., 82
Einstein, A., 610
Darcy, H., 304,368 Einstein, H. A., 283
Davey, A., 169, 170, 171, 175, 176, 192 Ekman, V. W., 76
Davidov, B. I., 3 7 8 ,3 8 0 ,6 8 0 Elagina, L. G., 507
AUTHOR INDEX 74 9
Elbert, D. D., 102 Gibson, C. H., 363

Elder, J.W., n , 6 1 9 , 6 2 1 Gifford, F. A., 5 60 ,5 6 6 ,6 7 5
Elias, F., 331 Gill, A. E., 143, 144, 145, 165
Eliassen, A., 125, 126 GUle,J., 182
Ellingsen, T., 189,190 Gjevik,B., 189,190
Elliott, D., 139 Gnedenko, B. V., 610
Elliott, W. P., 454 G6deke,K., 504
Ellison, T. H., 294, 295, 308,442, 448, Goering, H., 128
456, 476, 477, 501, 502, 503, 560, 580, Goldstein, R. J., 108,111,157
6 1 9 ,6 2 7 ,6 3 1 ,6 7 3 Goldstein, S., 20, 27, 37, 45, 47, 52,
Elrick, D. E., 638 64, 126, 138,325,368, 369,370,
Elrod, H.G., 272 37 1 ,3 7 3 ,6 8 0
Erk, S., 327 Gor’kov, L. P., 179
E sch,R .E ., 141 G0rtler,H., 7 3 ,1 0 7 ,1 6 8 ,3 5 4
Escudier, M. P., 277,319 Gosse,J., 304,365
Gotoh, K., 141,196
Page, A., 270 Graebel, W. P., 128
Falkner, V. M., 325 Graham, D .J., 157
Favre, A., 577 Grant, A.M., 578
Fejer,A. A., 202 Greenspan, H. P., 202
Fichtl, G. H., 475, 477, 478, 484, 525 Grenander, U., 254
Fidman, B. A., 270 Gr6ber,H., 327
Fiedler, H., 315 Grohne,D., 128,13 1,13 7
Fine, P. C , 53 Grosch,C. E., 1 30,131,146
Fj^rtoft,R., 122 Gross, A., 104, 105
Fleischman, B. A., 589 G ruza,G .V ., 387
Fock,V . A .,680 Gukhman, A. A., 345
Fokker,A .D ., 610 Gupta, H. C., 680
Forstall,W., 354 Gurvich, A. S., 3 3 7 ,4 7 3 ,4 7 4 ,4 7 5 ,
F o x ,J. A., 146 476, 477, 479, 480, 481, 483, 485,
Frank-Kamenetskiy, D. A., 282, 344 488, 491, 492, 494, 498, 505, 507,
Frenkiel, F. N., 545, 580, 589 5 0 8 ,5 1 7 ,5 1 8 ,5 2 0 ,5 2 3
Freymuth, P., 123 ,124,196
Friedmann, A. A., 9, 374 Hagen, G., 37,73
Fritsch,W., 298 H ains,F.D ., 129, 131, 195
Frizzola, A. J., 475,5 10 Halstead, M., 455,483
Fromm, J. E., 181 Haltiner, G. J., 423,613
Frost, R., 455,661 Hama, F. R., 198, 203,282, 290, 314, 368
Fujita, H., 314 Hamel, G., 152
Fultz, D., 103,104 Hamilton, R. M., 345
Funk, J. P., 424 Hamuro, M., 4 4 9,45 0
Furuya, Y., 314 Hanna, O .T., 342,345
Hanna, S., 3 3 7 ,4 9 2 ,4 9 3 ,4 9 4 ,5 1 7 ,
Gallagher, A. P., 129 5 1 8 ,5 1 9 ,5 2 1 ,5 2 5
Gaskill, H. S., 344, 345 Hanratty, T. J., 331, 334, 337, 344,
Gaster, M., 124,134 345
Gaviglio, J., 577 Hansen, M., 49, 77
Gaylord, E. W., 354 Harlow, F. H., 391
Gee,J. H., 6 6 3 ,6 6 8 ,6 6 9 Harriott, P., 345
Gel’fand, I. M., 221 Harris, D.L., 102,110
Genikhovich, E. L., 662 Hatton, A. P., 340, 342
Gerber, R., 528 Haugen, D. A., 505, 507, 577, 588,
Ghelardini, G., 102 659
75 0 AUTHOR INDEX
H au pt,0 ., 127 Jonsson,V.K ., 102,108,111

Hay, J. S., 577, 588 Joseph, D. D., 128, 152, 154, 156, 157,
Head, M. R., 315 ,3 1 7 ,3 4 2 167
Hegarty, J. C , 198,203 Joubert, P. N., 2 90,3 3 1 ,3 3 2
Heisenberg, W., 127
Helmholtz, H., 90 Kader, B. A., 334, 336, 343, 344, 345,
Herring, J. R., 181 348
Hess,G. D., 525 K aim al,J.C , 5 0 5 ,5 0 6 ,5 0 7 ,5 2 0
Hicks, B. B., 506, 507 Kakutani,T., 139
Hidy,G. M., 295 Kalinske, A. A., 587, 588
Hilst,G. R., 659 Kampe de Feriet, J., 208, 214, 254,
Hino, M., 295 543
Hinze, J. O., 80, 214, 277, 282, 304, 309, Kao, S. K., 453
354, 357, 362, 589, 603, 632, 633 Kapitza, P. L., 342
Hirschfelder, J. O., 53 Kapoor, R. K., 479, 492
Hishida, M., 331,337 Karlsson, S. K. F., 104, 106
Ho, C.Y., 104,106 Karol’, I. L., 642
Hocking, L. M., 147 Kaser, P., 468
Hoeber,H., 483 Kashkarov, V. P., 354, 357, 362
Hofmann, E., 282,345 Kays,W. M., 331,332
HOgstrOm, U., 483, 489, 501, 502, 504, Kazanskiy, A. B., 408, 435, 443, 445,
55 9 ,5 8 8 ,6 5 7 4 5 1 ,4 6 5 ,5 1 0 ,5 1 1 ,5 1 2 , 560,676,
Holland, J. Z., 524 688
Holzman, B., 451 Kedrolivanskiy, V. N., 468, 506
Hopf, E., 18,219 Keldysh, M. V., 94
Howard, L. N., 73, 96, 117, 118, 121, 122, Keller, L. V., 9, 374
123, 126, 139, 156, 167 Kelly, R .E., 197,198
Howarth, L., 47, 55, 64, 327 Kempf,G., 323
Holland, E., 122, 125, 126 Kester, R. H., 342
Hsu,N. T., 337 Kestin, J., 273, 277, 282, 327, 337
Hubbard, D. W., 345 KibelM. A., 2 7 ,3 1 ,4 5 ,4 0 9 ,4 1 0
Hudimoto, B., 282 King, K. M., 506, 507
Hughes, T.H., 136 Kirchgassner, K., 102
Hummel, R .L., 270 Kirillov, V. V., 342, 344
Hunt, J.N ., 308 Kiser, K.M., 362
Huppert, H. E., 128 Kishenevsky, M. Kh., 342, 344
Hurle, D .T., 108,111 Kistler, A. L., 2 8 3 ,3 1 5 ,3 5 4 ,3 6 2 ,
Hwang, N. H. C , 363 363
Kitaygorodskiy, S. A., 295
Ibragimov, M. Kh., 337 Klebanoff, P. S., 114, 198, 199, 203,
Inoue, E., 578 270, 280,315
lordanskiy, S. V., 138 Kleinstein, G., 282, 310
Isserlis, L., 233 Klenshow, C. W., 139
Ito, S., 520 Kline, S. J., 2 7 0 ,2 8 3 ,3 3 1 ,3 3 2
Ivanilov, Yu. P., 168, 170 Klug,W., 4 0 1 ,4 4 9 ,4 5 0 ,4 7 8 ,5 1 0 ,
Izakson, A., 208. 299 517, 560, 662,676
Izumi, Y., 520 Ko Sung-Hwan, 141
Kochin, N. 1., 2 7 ,3 1 ,4 5
Jakeman, E., 108,111 Kofoed-Hansen, O., 579
Jayatillaka, C. L. V., 327, 342 Kolesnikov, A. G., 502
Jeffreys, H., 107,110 Kolmogorov, A. N., 12, 14, 17, 219,
Johnk, R. E., 3 3 1,334,337 237, 310, 378, 382, 412, 610, 679
Jones, J. I. P., 504,579 Komoda, H., 198,199,202
AUTHOR INDEX 751
Kondo.J., 4 2 4 ,4 7 7 ,4 8 6 ,5 1 0 Lombardi, G., 102

Koppel, D., 124 L o n g ,J.D „ 198,203
Koschmieder, E. L., 187 LongweU, P. A., 27, 47, 64, 277, 282,
Kovdszay, L. S. G., 6 4 ,6 5 , 69, 137, 198, 290, 337
199, 202, 364, 373, 505 Lorentz, H. A., 152
Kraichnan, R. H„ 17,426 Lorenz, E., 387
Krasnosel’skiy, M. A., 168 Lortz, D., 180,184, 185
Kravchenko, I. N., 454 Low,A. R., 110
Krechmer, S. I., 486, 504, 506 Lozhkina, V. P., 662
Krishna, K., 479 Ludwieg,H., 270,337
Kropik,K., 145 Lumley, J. L„ 214, 283, 373, 391,475,
Krueger, E. R„ 1 0 4,10 5,10 7 477, 478,480, 483, 517, 525, 571,
Krylov, A. L., 102,130 572, 5 7 3 ,5 7 5 ,5 7 7 ,5 8 1
Kulikovskiy, A. G., 138 Lyapin, Ye. S., 680
Kuo, H. L„ 179, 181, 182 Lyatkher, V. M„ 272,283
Kurpakova, T. A., 4 0 9 ,4 1 0 Lykoudis, P. S., 342
Kuwabara, S., 195 Lyon, R. S., 342
Lyubimov, B. Ya., 528
Lamb,H., 2 7 ,9 0 ,5 2 8
Lambert, R. B., 172, 175, 176 MacAdams, W. H., 327, 335
Landau, L. D., 27, 29, 31, 53, 54, 56, 81, Maher, F. Y„ 505, 506, 507, 508
87, 90, 91, 116, 161, 272, 275, 283, 330 Malhotra, R. C., 475, 560, 566,676
Laufer, J„ 270, 279, 280, 298, 304 Malkus, W. V. R., 1 8 ,1 1 1 ,1 6 7 ,1 7 9 ,
Laykhtman, D. L., 406, 409, 410, 411, 426, 496
426, 4 5 3 ,4 5 5 ,4 7 8 ,4 8 1 , 486, 646, MandeU.D.A., 560,570
659, 661,666 Marcillat, J., 280
L e ite ,R .J „ 145 Mares, E., 3 3 7 ,4 9 2 ,4 9 3 ,4 9 4 ,5 1 7 ,
Lessen, M., 141, 146 5 1 8 ,5 1 9 ,5 2 1 ,5 2 5
Lettau, H., 294, 373,406, 409,410, 468, Martin, F .L ., 423,613
475, 486, 487, 505, 506, 508, 509, 517, MartineUi, R. S., 342
668 Mashkova, G. B., 40 9,41 0
Levich, V. G., 272, 282, 283, 327, 342, Matsuoka,H., 579,669
343, 344, 365 Matveyev, L. T., 451
Lewis, D. J., 104 McCormick, R. A., 465, 517, 518, 520
Li, H., 283 McCready, P. B., 504, 505
Lidskiy, V. B., 94 Mcllroy, I. C , 486, 504, 506, 508
Liepmann, H., 364 McVehil, G. E., 4 5 0 ,4 7 5 ,4 7 7 ,4 8 3 ,
Lifshitz, E. M„ 116, 161, 272, 275, 283, 485,488, 4 98 ,510
330 Meksyn, D., 100,102, 191, 192, 193,
Lightfoot, E. N., 345 195, 201, 202
Lighthill, M. J„ 71 Mellor, G. L., 270
Liljequist, G. H., 483, 502 Mercer, A. McD., 129
Lin, C. C , 73, 96, 100, 107, 109, 110, Merony, R. N., 520
113, 119, 121, 123, 128, 130, 133, Meyer, K. A., 169
135, 136, 141, 158, 201 Michael, D. H., 114
Lin,C. S., 282,3 44,3 45 Michalke, A., 123, 124, 141, 145, 147,
Lin, S. C , 363 196,197
Lindgren, E. R., 7 4 ,7 6 ,2 7 7 Mickelsen, W. R., 576, 588,604
Lindzen, R. S., 426 Middleton, W. B. K., 468, 486, 506
Liu Ting-Yung, 146 Mihaljan, J., 59
Lock, R. C , 130 Miles, J.W., 1 2 5,12 6,28 2
Loeve,M„ 245 MUler,J.A., 202
Loitsyanskiy, L. G„ 20, 45, 47, 282, 344, Millikan, C. B„ 299,303
372 Millionshchikov, M. D., 214
752 AUTHOR INDEX
Mitchell, J. E., 345 Orlenko, L. R., 4 09,410

Miyake, M., 475, 479, 492, 505, 506, 507, Orlov, V.V., 270,587
510 ,5 1 8 ,5 2 0 Ornstein, L. S., 678
Mock, W.C., 85 Orr,W. McF., 1 2 1 ,1 2 7 ,1 4 9 ,1 5 2
Moffatt, H. K., 373 Ovchinnikova, S. N., 169
Monin, A. S., 280, 287, 289, 406, 408, Owen, P. R., 347
409, 410, 427, 428, 435, 436, 438, 443,
445, 451, 465, 466, 468, 469, 475, 485, Paeschke,W., 292,294
487, 507, 510, 511, 512, 517, 521, 523, Page,F., 337
528, 538, 560, 580, 614, 645, 646, 661, Pai, Shih-i, 354, 357
6 7 6 ,6 7 7 ,6 8 8 Palm,E., 1 8 7 ,1 8 8 ,1 8 9 ,1 9 0
Montgomery, R. B., 452, 454 Panchev, S., 466
Mordukhovich, M. I., 424, 479, 492, 493, Pandolfo, J. P., 454, 466,479, 480, 492
497, 5 0 7 ,5 1 7 ,5 1 8 ,5 2 0 ,5 2 1 ,5 2 2 Panofsky, H. A., 214, 337, 449, 450, 455,
Moulton, R. W., 282,344 4 6 5 ,4 6 6 ,4 7 5 ,4 7 7 ,4 7 8 ,4 8 0 ,4 8 2 ,4 8 3 ,
Munn, R. E., 465, 507, 659, 687, 691 485, 488,492, 493,494, 498, 510, 517,
Munro,W.D., 102 518, 519, 520, 521, 525, 560, 575, 577,
Murphree, E. V., 272,344 5 8 1 ,5 8 8 ,6 7 6
Myrup, L.O., 520,5 21,5 25 Pasquill, F., 331, 468, 472, 488, 489, 499,
5 0 0 ,5 0 4 ,5 0 8 ,5 2 0 ,5 4 5 , 560, 577, 580,
Nachtsheim, P. R., 130 5 8 1 ,5 8 8 ,6 1 3 ,6 4 3 ,6 5 9 , 662, 674, 676,
Naito, K., 454 688
Nakayama, P. I., 391 Patmaker, S. V., 342
Narayana, T., 76 Paulson, C. A., 4 7 5 ,4 7 7 ,4 7 9
Narayanan, M. A., 76 Pearson, J. L., 660
Nardacci, J. L., 104,106 Pedisius, A. A., 333,337,338
Nevzglyadov, V. G., 278 Pedley,T. J., 147
Pekeris,C.L., 1 2 7 ,1 4 5 ,1 9 2 ,1 9 3
Newman, J. T., 507
Nicol, A. A., 134 Pellew,A., 1 0 0 ,1 0 2 ,1 0 9 ,1 1 0 ,1 1 1
Pelton,W. L., 294
Nicoll,W. B., 277,319
Nield, D. A., I l l Peregrym,J., 331
Nikuradse, J., 47, 49, 276, 288, 289, 306, Perepelkina, A. V., 292,472, 497,499,
5 1 7 .5 2 0 .5 2 2 .5 2 4
368
Perry, A. E., 2 9 0 ,3 31,3 32
Nissan, A. H., 104, 106
Nomofilov, E. V., 337 Peterson, E., 337, 492, 493, 494, 517,
Novikov, E. A., 242,638 5 1 8 .5 1 9 .5 2 1 .5 2 5
Nunner,W„ 331,347 Petrov, G. I., 139,152
Nutant, J., 198 Petukhov, B. S., 342,344
Pfenninger, W., 76
Philip,!. R., 579,619
O’Brien, E. E., 560, 570 Phillips, O. M., 8 1 ,2 9 5 ,3 7 3 ,6 1 9
O’Brien, J. J., 4 7 5 ,4 8 3 ,4 8 5 ,4 8 6 P ie n ,C .C , 587,588
Obukhov, A. M., 12, 15, 17, 368, 427, Pike,E. R., 108,111
428, 435, 436, 438, 449, 461, 464, Planck, M., 610
46 9 ,4 7 5 ,4 8 5 Plate, E. J., 295, 520
Ogura, Y., 452, 578 Platzman, G., 156, 167
Ohji, M., 272 Plows, W.H., 181
Ohmstede, W. D., 406, 409, 410 Poiseuille, J., 37
0iann,H., 188,189 Polavarapu, R. J., 507
0 k a d a ,0 ., 144 Ponomarenko, Yu. B., 129, 165, 179,
Okamoto, M., 450 184,190, 191
Okubo, A., 534, 601 Popovich, A. T., 270
Onikul, R. I., 662 Potter, M.C., 131,192,195
AUTHOR INDEX 753
Prandtl, L., 10,45, 82, 125, 127, 266, Sabcrsky, R. H., 347
274, 278, 292, 326, 341, 364, 366, Sadler, S. G., 146
36 7 ,3 6 8 ,3 7 8 ,4 3 5 ,4 3 6 ,4 6 1 Saffman, P. G., 578, 600, 601, 603, 619,
Prasad, B., 337, 475, 492, 493, 494, 627,640, 647,648, 668
5 1 7 ,5 1 8 ,5 1 9 ,5 2 0 ,5 2 1 ,5 2 5 ,5 6 0 , Sage, B. H., 337
676 Sakao, I-., 140
Pretsch,J., 117,132,145 Sakipov, Z. B., 362
Pries, T.H ., 524 Salter, C , 540
Priestley, C. H. B., 292, 294, 337, 435, Salwen, H., 130, 131, 146
442, 445, 446, 449, 466, 474, 476, Sandborn, V. A., 4 7 5 ,4 8 2 ,5 2 0
495, 496, 497, 499, 510, 521, 668 Sani, R., 156,167
Prokhorov, Yu. V., 221 Sargent, L. M., 114, 198, 199
Proudman, J., 502, 503 Sato, H., 123, 140, 144, 196, 197, 198,
Putnam, G. L., 282, 344, 345 337
Sattinger, D. H., 121
Schade, H., 123, 143, 144, 145, 195
Rannie,W. D., 28 2,34 4,36 5 Schensted, I. V., 127
Rayleigh, 97, 109, 117, 120, 123, 143 Schiller, L., 3 6 ,5 2 ,7 4
Record, F. A., 492, 505, 506, 525 Schlichting, H., 20, 45, 47, 49, 64, 72, 77,
Reichardt, H., 282, 342, 344, 345, 79, 124, 125, 128, 135, 138, 290, 325,
3 5 4 ,3 5 5 ,3 7 9 326,356,371
Reid, W. H., 73, 102, 110, 119, 128, Schlinger, W. G., 337
132, 133, 136 Schluter, A., 180, 184, 185
Reuter, H., 659 Schmidt, W., 2 64,455,606
Reynolds, A. J., 310 Schneck, P., 181
Reynolds, O., 6, 73, 148, 149, 207, Schoenherr, K. E., 325
258, 336, 340 Schraub, 1'. A., 270, 283
Reynolds, W. C , 21, 192, 195, 270, Schubauer, G. B., 85, 133, 135, 198, 304,
28 3 ,3 3 1 ,3 3 2 325, 356
Ribaud,G., 2 8 2 ,342 ,344 Schubert, G., 283
Richardson, L. F., 12, 402 Schultz-Grunow, 1-., 102, 322, 323, 325,
Richardson, P. D., 273, 277, 282, 327, 337 326
Rider, N. E., 331, 337, 472, 474, 476, 483, Schwarz, K.W., 16 8,1 7 2 ,1 7 3 ,1 7 6
488,489, 491,499, 500, 508 Schwarz, W. H., 373
Riis, E., 125,1 26,1 29 Segel, L. A., 73, 108, 180, 181, 182, 184,
Rijkoort, P. J., 449, 451, 454, 475, 477 188,189, 190
Rivlin, R. S., 373 Selig, F., 124
Roberts, O. F. T., 641,661 Sellars, J. R., 145
Roberts, P. H., 102, 105, 175, 179, 181 Sellers, W. D., 449
Robertson, J. M., 290 Serrin,J., 149,150
Robinson, G. D., 423, 442, 483 Se\l,T ., 145
Roll, H. U., 295 Shcherbakova, L. I ., 468
Rosenbluth, M. N., 123 Shen,S. F., 73,97, 119, 123, 128, 129,
Rosencrans, S. I., 121 130, 131, 133, 135, 136
Ross, M. A. S., 134 Sheppard, P. A., 50 8 ,5 2 2 ,5 8 0
Rossby, C.G., 1 8 7 ,3 8 7 ,4 5 2 ,4 5 4 Sherwood, T. K., 342
Rota, G. C., 208 Shimanuki, A., 4 5 3 ,6 6 2 ,6 6 5
Rotta, J. C , 74, 270, 282, 312, 314, 316, Shiotani, M., 472, 510, 522
368, 378,379 Shir,C. C., 157
Rounds, W., 640,661 Shkoller, B., 192,193
Rozanov, Yu. A., 2 45,256,541 Shvets, M. E., 366
Roze, N. V., 27, 31,45 Sibbons, J. L. H., 492
Runstadler, P. W., 270, 283 Sibulkin, M., 76
75 4 AUTHOR INDEX
SUveston, P. L., 187, 191 Szablewski, W., 365,368

Simmons, L. F. G., 540 Szewczyk, A., 142
Simon, A., 123
Singer, I. A., 47 5,51 0
Skramstad, H. K., 85, 133, 135
SianCauskas, A. A., 333, 337, 338 Takeda, K., 455
Slciclier,C. A., 337 Takeuchi, K., 401, 452, 474, 486, 491,
Slutskiy,E. E., 251 510,525
Smith, D.W., 322 Tam,K. K., 128, 138
Smith, F. B., 559, 638, 657, 661, 664 Tani, I., 7 2 ,1 9 8 ,1 9 9 ,2 0 2
Snyder, H. A., 103, 104, 106, 172, 175, Tanner, C. B., 294
176 Tatarskiy, V. I., 507
Sogin, H. H., I l l Tatsumi,T., 139 ,141,165
S o n,J.S ., 344,345 Taylor, E. A., 617,627
Soprunenko, I. P., 139 Taylor, G. I., 9, 10, 12, 13, 83, 97, 99,
Sorger, P., 150,151 104, 125, 133, 171, 252, 278, 341, 364,
Sorokin, V. S., 152, 155, 156, 167, 168, 366, 369, 543, 545, 587, 588, 606, 617,
179 619, 621, 625, 626, 627, 658, 677, 680
SouthweU, R. V., 100, 102, 109, 110, Taylor, R. J., 474, 481, 486, 491, 494,
111, 128 495, 497, 505, 506, 507, 525
Spalding, D. B., 2 7 7 ,2 8 2 ,3 1 9 ,3 2 7 , Tchen, C M ., 304,3 25,3 56
342, 345, 368 Temirbayev, D. Zh., 362
Sparrow, E. M., 10 2 ,1 0 8 ,1 1 1 ,1 2 8 , Tempel’man, A. A., 254
136 Thomas, D. B., 496
Spiegel, E. A., 18,426 Thomas, L. H., 130
Spielberg, J., 145 Thompson, B. G. J., 317,321
Spilhaus, A. F., 468, 487, 506 Thompson, H. A., I l l
Springett, B. E., 168, 176 Thompson, N., 577, 588
Squire, H.B., 113,2 7 5 ,2 8 2 ,3 5 4 , Thomson, W. R., 347
356 Thornthwaite, C. H., 468
Squire, W., 333, 342 Tidstrom, K. D., 114 ,198,199
Starr, V. P., 373, 387 Tiederman, W. G., 21
Steinman, H., 102 Tien, C. L., 282, 339
Sternberg, J., 283 Tietjens, O., 127
Sternzat, M. S., 468, 506 Tillman, J. E., 505,506
Stewart, R. W., 8 1 ,3 7 3 ,4 4 0 ,5 0 2 Tillmann,W., 270
Stokes, J., 37 Timme, A., 147, 197
Stuart, J. T., 73, 77, 100, 108, 109, 110, Tippelskirch, H., 187
111, 119, 128, 136, 161, 170, 171, 175, Tollmien, W., 117, 122, 128, 144, 160,
176, 179, 183, 188, 189, 191, 192, 193, 354
195, 196, 198, 199, 202 Tolman, R. C., 53
Subbotin, V. I., 337 Townend, H. C. H., 270
Sun Dah-Dhen, 102 Townsend, A. A., 1 8 ,2 7 2 ,2 7 7 ,2 8 0 ,
Sundararajan, A., 479, 492 2 9 9 ,3 0 9 ,3 1 0 ,3 5 2 ,3 5 4 ,3 5 7 ,3 6 2 ,
Sutton, O. G., 108, 111, 292, 294, 295, 364, 424, 426, 496, 540, 541, 545,
4 5 4 ,5 8 0 ,6 1 3 ,6 4 3 ,6 5 7 , 661 5 5 4 ,5 8 0 ,5 8 7 ,6 0 0 ,6 3 3
Sverdrup, H. U., 292 Ts’ai Ko-yeng, 282
Swinbank, W. C , 337, 442, 453, 474, 475, Tsao, S., 102
477, 478, 479, 480, 488, 489, 491, 492, Tseytin, G. Kh., 640,661
493, 494, 496, 497, 498, 499, 500, 501, Tsvang, L. R., 424, 479, 492, 493, 497,
504, 5 0 6 ,5 0 8 ,5 1 7 ,5 1 8 ,5 2 1 ,5 2 2 , 5 0 7 ,5 1 7 ,5 1 8 ,5 2 0 ,5 2 1 ,5 2 2 ,5 2 3
Synge, J. L., 97, 102 Turner, J. S., 4 5 6 ,4 7 6 ,4 7 7 ,5 0 2
Syono, S., 449, 450 Tyldesley, J. B., 6 1 9 ,6 2 7 ,6 4 0 ,6 5 7
AUTHOR INDEX 755
Uberoi, M. S., 354, 541, 57 /, 587, 588, White, C. M., 147

633 Whitehead, J. A., 187
Uhlenbeck, G. E., 678 Wieghardt, K., 83,317
Ukhovskiy, M. R., 156, 167 Wilke, C. R., 344
Ulinskas, R. V., 333, 337, 338 Willis, G.E., I l l , 16 8,496,497
WUmarth, W. W., 283
Vager, B. G., 453 Wilson, B.W., 295
Van Atta, C. W., 176 Witting, H., 102
Van der Hegge Zijnen, B. G., 49, 77, Wortmann, E. X., 134
354, 362,633 Wu, J., 295
Van Driest, E. R., 282, 329, 342, 368 Wyngaard, J. C , 505
Vasudeva, B. R., 137, 198, 199, 202
Vaughan, L. M., 663 Yaglom, A. M., 64, 68, 245, 334, 336, 343,
Velikanov, M. A., 415 348, 461, 523, 554, 560, 567, 613, 637,
Velte, W., 73, 150, 168 667
Venezian, E., 337 Yalovlev, G. N., 168
Venotte, P., 680 Yamada, H., 140
Veronis,G., 1 79,181,182 Yamamoto, G., 4 2 4 ,4 4 9 ,4 5 0 ,4 5 3 ,6 6 2 ,
Veronov, V. P., 505 665
Vilenkin, N.Ya., 221 Yerazunis, S., 342,345
Volkov, Yu. A., 295 Yokoyama, O., 4 0 1 ,4 5 0 ,4 5 2 ,4 6 6
von Karmdn, T., 10, 22, 152, 274, 282, Yordanov, D. L., 642, 646, 647, 661
298, 3 0 6 ,3 2 3 ,3 4 1 ,3 6 4 ,3 7 0 Yoshihara, H., 479
von Mises, R., 299 Yu Yun-Sheng, 102
Vulis, L. A., 354, 357, 362 Yudin, M. I., 366
Yudovich, V. I., 102, 156, 160, 167, 168,
Walker, J.H ., 322 169
Wallington,C.E., 6 1 9 ,6 2 7 ,6 4 0 ,6 5 7
Walowit,J., 102 Zagustin, A. I., 378
Walters, T. S., 642,664 Zarantonello, E. H., 357
Wandel, C. F., 579 Zel’dovich, Ya. B., 347, 435
Wasan, D. T., 282,344 Zilitinkevich, S. S., 294, 295, 406, 409,
Watson, J., 1 1 4 ,1 7 1 ,1 8 3 ,1 9 1 ,1 9 4 , 410, 411, 453, 473, 475, 478, 480, 481,
199 483, 486, 490, 491, 492, 493, 498, 510,
Webb, E. K., 295, 447, 475, 476, 478, 480, 511, 512,517
4 8 3 ,4 8 6 ,4 9 5 ,4 9 6 ,5 0 7 ,5 1 0 Zubkovskiy, S. L., 492, 493, 517, 522, 523,
Webster, C. A. G., 503, 504 525
Whippermann, F., 662 Zukauskas,A. A., 3 3 3,33 7,33 8
SUBJECT INDEX
Absorption, boundary condition of 641, Atmospheric diffusion, field data on 662

642 semiempirical equation of 666, 668
Acceleration, bouyant 422 theory of 640,645
Acoustic mode 68, 69, 70 Atmospheric diffusion experiments 628,
Admixture, longitudinal dispersion of 624, 659
626 Atmospheric diffusion theory 640, 645
passive 15 Atmospheric dispersion, Sutton’s equations
Admixture cloud 650, 657 for 657
acceleration of relative diffusion of 665 Atmospheric measurements 502
asymptotic behavior of 648 hot-wire anemometer for 504
center of gravity of 650 Atmospheric observations 518, 520
horizontal dispersion of 652 Atmospheric surface layer, diffusion in 640,
mean height of 655 643
problem of diffusion of 644 Atmospheric turbulence 2
thickness of 656 characteristics of 504
Admixture concentration, gradient of 11 Lagrangian characteristics of 581
Admixture dispersion, atmospheric 644 Australian data 499
horizontal 647 Autocorrelation function 228
longitudinal 617, 619, 620, 621, 625, Averaging operations 208
627 Averaging rule 207
Admixture mass, law of conservation of 594 Axisymmetric disturbances, linear theory of
Admixture particle 602 165
Admixture propagation, data on 664
Admixture sources 641
instantaneous 636 Banach spaces 168
Advection equation 533, 534, 536, 602, Barenblatt’s theory 412
606 Basset function 617
Amplitude equations 183 Benard cells 11 0,111,187
Anemometers, atmospheric hot-wire 505, Benard experiments 187
506 Bernoulli equation 79, 80, 89
acoustic 506 Benney’s results 201
linearizing hot-wire 505 Bernstein’s results 478
propeller 506 Birkhoff-Khinchin ergodic theorem 256
Angular momentum, law of conservation of Birkhoffs ergodic theorem 215
96 Blackadar’s theory 410
Anticyclones 386 Blasius approximate law 307
dimensions of 420 Blasius boundary layer 135, 136,137
Appleby-Ohmstede theory 410 Blasius empirical law 310
Archimedean forces 58,418 Blasius friction equation 310
Archimedes law 418 Blasius profile 132
Aris’ method 628, 647 Blasius solution 72
ArnoPd’s works 160 Blinova-Kibel theory 410
757
758 SUBJECT INDEX
Bobyleva-Zilitinkevich-Laykhtman theory Characteristic functionals, use of techniques

410 of 18,19
Boundary conditions, time-dependent 8 Charnock’s treatment, results of 492, 493
Boundary layer, logarithmic 274, 275 Chatwin’s computations 654
planetary 40 5,410 Chatwin’s result 653
separation of 80 Chebyshev’s inequality 250
transition in 166 Clauser’s recommendations 304
Boundary layer equation 45 Closure problem 9
Boundary layer flows, Lagrangian similarity Coefficient of intermittcncy 74
treatment of 560 Coles’ approach 321
Boussinesq approximation 108, 187, 398, data 318
399 results 176
continuity equation for 422 terminology 103
for fluid mechanics equations 59 wake law 320
Boussinesq equations 59, 108, 153, 156, Complex variable, theory of functions of
179, 180, 186, 380, 390, 397, 399, 120
421,424 Compressible fluid, equations of mechanics
Busse’s computation of 186 of5 2
cellular solutions of 179 flows of 53
of free convection 107, 187 Compressible gas, isotropic turbulence in 64
nonlinear 168, 184 Compressible medium, turbulence in 52
system of 358 Constant density rules 124
two-dimensional solutions of 187 Constant stress layer, theory of flows in 296
Bovsheverov-Gurvich instrument 505 Continuity, equation of 114
Bowen ratio 508 Continuous medium, mathematical points
Brownian motion 678 of 580
basic law of 545 Convection, cellular 111
of particles 588 forced 57
Budget of heat, equation of 52, 54 free 58, 59, 62, 448,459, 516
Buoyancy effects 107, 292, 426 Boussinesq equations of 187
Buoyancy forces 329, 330, 396, 397, 398, conditions of 436, 439, 446, 451, 452,
418 4 7 6 ,4 9 7 ,5 6 8
Buoyancy parameter 426 eddy thermal dissusivity in 435, 436
Buoyant acceleration 422 equations of 421
Burton’s generalization of one-seventh law laws of 451
310 (see also, one-seventh law) pure 567
Businger’s equations 454 regime of 443,464, 483,491, 494,498,
Businger’s semiempirical theory 453 516,518
Busse’s computation of Boussinesq equa limiting laws of 567
tions 186 (see also, Boussinesq equa true 461
tions) free-free case of 188
in geometrically similar flow 62
Case’s results 121,123 pure 434
Cellular convection 111 Convection cells, hexagonal 189, 191
finite amplitude 187 two-dimensional 187
stable form of 187 Convection equations, hnearized 180
hexagonal 187 nonlinear 179
Cellular motions, stability of finite ampli Convection theory, nonlinear 111
tude 185 Convective flows, stability of 154
Central limit theorem 541, 608, 619 Convective motions, stability theory for
Central moment 223 cellular 182
Chandrasekhar’s calculation 102 temperature fluctuations in 442
Characteristic functions 218, 219 Coriolis force 405, 406, 4 0 7 ,4 1 1 ,4 2 2 ,4 5 8
SUBJECT INDEX 769
Coriolis parameter 407 Diffusion

Correlation functions 9, 228 acceleration of relative 665
higher-order 229 coefficient of 11
Correlation length 389 molecular 57
Correlation time 252 deceleration of 600
Correlation volume 252 purely turbulent 584
Correlometers, automatic 507 theory of 685
Corrsin’s superlayer 362 Diffusion equations 606
Couette component 195 parabolic 682, 683, 684, 689
Couette now 78, 96, 97, 99, 112, 120,122, semiempirical 667, 673, 676, 679, 680
128, 131, 147, 150, 151, 165, 166, parabolic 677
170 solution of 602
between cylinders 103 three-dimensional 663
circular 38, 99, 101, 103, 104, 151, 166, two-dimensional 660, 662
167, 168, 169, 171, 175, 176, 177, Diffusion theory 688, 689, 690, 691
178, 193 problems of 641
axisymmetric disturbances of 103, 104 semiempirical 666, 668, 670, 677, 690
discrete transitions in 168 surface layer 669
mathematical analysis of behavior of Diffusivity, eddy admixture 692
104 Dikiy’s results 121,122
nonaxisymmetric disturbances of 104 Dirac’s delta function 206, 220, 249, 458
stability of 107 Displacement, zero 294
loss of 106 Displacement fluctuations, characteristics of
increase of Reynolds number in 106 542
laminar 167,1 71,6 30 Displacement height 294
nonlinear instability effects in 166 Displacement thickness, concept of 294
plane 34, 121, 127, 128, 129, 137, 151, Displacement thickness 6 * 50, 51, 79
156 ,165,195 Distribution, horizontal velocity 434
nonlinear stability of 195 mean temperature 434
plane-parallel 143, 145 functions, multiple 9
Couette flow stability problem 100 Disturbance energy, equation of balance of
Couette-Poiseuille flow 195 170
plane 131 Disturbance theory, hydrodynamic 103
Covariance function 228 third approximation of 189
Co variance tensor, fluid particle displace Disturbances
ment 542 axisymmetric 106
Cramer’s data 519 finite 93
Cross-correlation function 228 behavior of 161, 163
Cumulants 224, 230 free-stream 133
fourth-order velocity 378 infinitesimal 93, 185, 186, 190
Curie’s method 139 instabihty of 103
Cyclones 386 interaction of 198
dimensions of 420 nonaxisymmetric 106, 107
behavior of 104
critical Reynolds numbers for 105
Davey’s calculations 172 small 92
Davey’s values 175 method of 97, 127, 128
Davidov’s transport equation 380 application of 111
Davidov’s works 379 small-scale 388
Deacon’s data 476 space-periodic 134
Defect layer 298 theory of 124
Degrees of freedom, concept of 86 three-dimensional 112, 113, 114, 160,
Devices, low-inertia 1 198, 199,201,202
760 SUBJECT INDEX
Disturbances (cont.) Ellison’s semiempirical formula 503

time-periodic 124, 134 Energy, rate of dissipation of 376
turbulent 386 viscous dissipation of 352
two-dimensional 112, 113,114, 134, 152, Energy balance, turbulent 449, 525
160, 169,194,201 equation of 452
unstable 166 Energy balance method 509
wave-like 192 Energy density, kinetic 375
three-dimensional 192 Energy diffusion, turbulent 525
two-dimensional 192 Energy dissipation, kinetic 423
Donnelly-Schwarz experiments 175 rate of 57
Drag coefficient 33 Energy method 148, 152, 156, 157
Drag crisis 82 application of 167
Drag law 32 variational technique of 156
Dyer’s data 499 Ensemble mean 215
Dynamic sublayer, height of 431 Entropy mode 68, 69, 70
thickness of 482 Ergodic hypothesis 215,216
Dynamical equations, system of 27 Ergodic theorems 216, 248, 249, 251
Dynamical problems, viscous effect in 63 Euler constant 653
Dynamometric method 508 Eulerian characteristic 529
single-point 575
Eulerian concentration field, semiempirical
Eckhaus eigenfunction expansion method diffusion equation of 621
192 Eulerian conservative characteristic 534
Eddies, cascade process of breaking down Eulerian correlation coefficients 576, 577
12 Eulerian description of fluid motion 527,
energy-containing 421 528
hairpin 203 Eulerian dynamic equations 529
hierarchy of 12 Eulerian field 533,5 34,5 71
small-scale 437 of volumetric concentration 581
Eddy correlation method 504 Eulerian properties of turbulence, self-
Eddy diffusion coefficients 264 preservation of 548
Eddy diffusivity 429 Eulerian random variable 5 37
longitudinal 618 Eulerian space-time correlation function
power-law height dependency of 645 539
Eddy diffusivity tensor 607, 640 Eulerian statistical characteristics 538, 540,
Eddy exchange coefficients 428 5 4 1 ,5 4 7 ,5 5 4 ,5 6 0 ,5 7 0 , 572, 621
Eddy thermal conductivity 264 Eulerian time scale 577, 578
Eddy thermal diffusivity 329 Eulerian time-space velocity correlations,
Eddy viscosity 692 measurements of 577
coefficients of 389 Eulerian turbulence characteristics, self-
Eddy viscosity theory 373 preservation of 548
Eigenmodcs, spacc-periodic 137 Eulerian velocity 9, 538, 540, 563, 571, 574
Eigenvalue problem, asymptotic analysis of vertical 562, 566
105 Eulerian velocity field 5 35
Einstein’s summation convention 28 Excitation, hard 158
Ekman layer 405 soft 158
Elder’s calculations 631
Electrochemical reactions, rate of mass
transfer in 344 Falknerlaw 324
Elementary solutions, system of 96 Feynman measure 19
Ellison-Turner data 477 Finite disturbances, nonlinear theory of 157
Ellison’s equations 308,478 Fj(^toft-H0iland condition 122, 123, 144
Ellison’s relationship 456 Flatness factor 224
SUBJECT INDEX 761
I'lows wind-velocity 504, 506

between cylinders 38 Fluid, equations of heat conduction of a
boundary-layer 132 moving 59
unstable disturbances in 137 Fluid, temperature-inhomogeneous 58
convective 154 motion of 58
geometric conditions of 666 Fluid dynamic fields, homogeneity of 248
geometrically similar 40 Fluid dynamics, equations of 27, 217, 260,
inviscid 3 6 1 ,580 ,582
instability of plane-parallel 144 differential 207
proof of stability of 120 system of 157
stability theory of plane-parallel 119 Fluid elements, displacement of 397
in a jet 351 turbulent 397
kinematic conditions of 666 vertical 417
Lagrangian description of 528, 529 Fluid flow, Eulerian description of 533
laminar Lagrangian description of 533
steady 8 6 ,9 6 ,1 6 4 Fluid mechanics 4 ,5
nonconvective 154 Boussinesq approximation for equations
plane-parallel 114, 119, 121, 122,124 of 59
instability of 144 equations of 7, 8, 36, 41, 72, 179, 188,
inviscid stability theory of 126 201
stability problem for 112 nonlinear 160
theory of stability of 119, 128, 132 system of 31, 57
quasi-parallel Galilean invariant equations of 160
stability of 132 laws of 7
self-preservation of 351 statistical 3 ,1 0 ,2 2 ,2 3
hypothesis of 351 development of 10, 17
similar Fluid motion, Eulerian description of 527,
statistical ensemble of 209, 258, 263 528
theory of 48, 296 Fluid particle displacement, covariance
in channels and tubes 296 tensor of 612
past a flat plate Forced convection conditions 439,491
experimental verification of 48 Formations, turbulent 421
turbulent Fourier components 3, 17, 18, 67
dissipation of energy in 17 Fourier-Stieltjes integral 86
energy balance equation of 381 Free convection 58, 59, 62, 448, 459,
fluid mechanical fields of 9 516
free 349, 361 Boussinesq equations of 187
diffusion in 632 conditions of 436, 439, 446, 451, 452,
self-preservation hypothesis for 349 4 7 6 ,4 9 7 ,5 6 8
high-in tensity 317 eddy thermal diffusivity in 435, 436
L agrangian characteristics of self- equations 421
preserving 547 laws 451
plane-parallel 390, 392 pure 567
properties of 2, 264 regime 443, 464,476, 483,491, 494,498,
spread of admixtures in 528 516,518
structure of hydrodynamical fields of 8 limiting laws of 567
transition from 73, 75, 76 true 461
Fluctuation method 504, 509 Free turbulence 257
Fluctuations, air-temperature 506 methods of describing 349
density 419 Free turbulent flows 349, 361
micrometeorological 504 diffusion in 632
synoptic 421 self-preservation hypothesis for 349
temperature 419 Freymuth’s experimental findings 124
76 2 SUBJECT INDEX
Friction, coefficient of 33,286,301 Hagen-Poiseuille flow 37

internal 375 laminar 307
molecular forces of 375 Hagen-Poiseuille profile, parabolic 165
pressure forces o f 375 Hamilton’s data 346
Friction forces 38 Hamilton’s observations 345
Friction law 301 Hanratty’s electrochemical technique 345
Friction layer 405 Hansen’s measurements 50
Friction length 266 Heat conduction, equation of 59, 63
Friction velocity 266, 292 Heisenberg’s result 127
Friedmann-Keller method 9 Helmholtz instability 90
Functional, characteristic 220, 221 Holland’s computational technique 524
positive-definite 221 Homogeneous field 248
Functional derivatives 238 Hopfs equation 19
higher-order 239, 241 solution of 19
theory of equations in 19 Howard’s semicircle theorem 144
Function space integrals, mathematical tech Hubbard’s data 346
niques of 19 Hunt’s data 308
Hydrodynamic stability 157
inviscid 121
theory of 94 ,95, 101, 107
Hydrodynamic equations, linearized system
of 65,66
Galerkin’s mathematical method 102, 195 Hydrodynamics, equations of 27
Galilean invariance 371 Hygrometers, meteorological 502
of equations of mechanics 274 optical infrared 507
Gases, kinetic theory of 4, 9, 53, 54, 367 Hygrometry, method of 507
Gaussian density, multidimensional 614 Hypergeometric functions 691
function, multidimensional 639
three-dimensional 607
Gaussian density function, multidimensional
639
Gaussian field 231,247 Ideal tluid, plane-parallel flows of 116
Gaussian law 615 stability of 116
Gaussian processes 253 Ideal (nonviscous) fluid, mechanics of 195
Gaussian random fields 234, 237, 242 Incompressible fluid, diffusion in 581
homogeneous 254 simple flows of 31
Gaussian random functions 236, 237, 245 stability problem of 107
Gaussian random process 253 Incompressible vorticity mode 68
stationary 254 Incompressibility, equation of 58
Gee-Davies results 669 Inertia forces 38, 39
Gee’s computations 669 Inoue’s model 578, 579
Gifford’s calculations 676 Instability, absolute 96
results of 675,676 Instability criterion 115, 116, 117
Gifford’s conclusions 560 Instability effects, nonlinear 191
Gdrtler’s stability investigation 107 Integral time scale 252
Gotoh’s result 196 Integral volume scale 252
Gravitational field, nonuniformly heated Interactions, inertial 532
fluid in 58 Intermittency, coefficient of 74
Gravity forces 437 Internal friction, molecular forces of 375
Green’s function 19 pressure forces of 375
Grohne’s results 129,131 Interpolation formulas 445, 446, 447, 448,
Gurvich constant 478 449,450
Gurvich’s observations 519, 520 zero-ordor 447
SUBJECT INDEX 76 3
Inviscid flow, instability of planc-parallcl of turbulent boundary layer 559

144 Lagrangian concentration field 534
proof of stability of 120 Lagrangian correlation coefficients 576
stability theory of plane-parallel 119 Lagrangian diffusion equation 601
Inviscid stability problem 120 Lagrangian dynamic equations 529, 532,
Inviscid stability theory 119 534,611
Isserlis’ rule 233 Lagrangian form, fluid dynamic equations in
528
Lagrangian integral time scale 537, 576,
Johnk-Hanratty data 333, 334, 337 5 8 8 ,6 1 2 ,6 1 3 ,6 1 5 ,6 2 1 ,6 3 2 ,6 5 8
Johnk-Hanratty results 332 Lagrangian properties of turbulence, self-
Joining conditions, natural 123 preservation of 548
Lagrangian random variables 538
Lagrangian self-preservation 557
Kader’s data 346 Lagrangian similarity hypothesis 554, 555,
Kader’s method 344 669, 672
Kader’s observations 345 results from 672
Kalinske-Pien diffusion experiments 587, Lagrangian similarity treatment 560
589 Lagrangian statistical characteristics 548,
Kao’s scmiempirical theory 543 5 6 0 ,5 7 2 ,6 7 5 ,6 8 9
Kaplan’s machine computations 139 of turbulence 535, 536, 538, 560, 570,
Kazakhstan Steppe, meteorological measure 581
ments in 468 Lagrangian statistics 588
Keldysh theorem 94 Lagrangian time correlation function 578,
Kellcr-Friedmann method 374 579
K em pfsdata 323,324 Lagrangian time macroscale 576
Kinematic similarity, hypothesis of local Lagrangian time microscale 605
370 Lagrangian time scale 544, 577, 578, 607,
Kinetic energy, dissipation of 12, 339, 375 608,609
transformation of 398 Lagrangian trajectory 581
viscous dissipation of 299 Lagrangian turbulence characterisitcs, simi
KirchgSssner’s result 169 larity of 669
Klenshow-Flliot numerical calculations 140 Lagrangian variables 528, 535, 556, 571,
Kleinstein’s theory 310 632
Kolmogorov-Obukhov theory 13 Lagrangian velocity correlation function
Kolmogorov’s second main hypothesis 15 539, 549, 569, 572, 574, 581, 588,
Kolmogorov’s theory 16, 18 589, 606, 615, 621, 657, 659
Kolmogorov’s theory of locally isotropic Laminar sublayer 270
turbulence 14, 21 Landau-Stuart equation 179
Kolmogorov’s two-thirds law 15,17 Landau’s approach to nonlinear instability
Kraichnan’s theory 18 165
Kraichnan’s work 426 Landau’s coefficient 196
Krasnosel’skiy’s topological method 169 Landau’s constant 196
Krechmer’s microthermometer 506 Landau’s equation 166, 170, 172, 183, 191,
Krueger-Gross-DiPrima results 106 194,196 ,200
Kuwabara’s result 195 Landau’s ideas on developed turbulence 216
Landau’s schema 179
Landau’s theoretical prediction 173
Lagrangian analysis 657 Landau’s theory 160, 165, 166, 172, 173,
Lagrangian characteristics 528, 529, 586, 197
654 Laplace integral representation 121
of atmospheric turbulence 581 Large numbers, law of 251
of turbulence 24, 554 Laufer’s data 298, 299
764 SUBJECT INDEX
Laufcr’s measurements 281, 304, 380 problems of 641

Laykhtman’s assumptions 426 Matsuoka’s equation 669
Laykhtman’s proposition 666 Mean values, definition of 206
Laykhtman’s scheme 455 Measurement methods, heat-flux 508
Least squares method 4 8 1 ,4 8 3 ,5 1 0 Measurements, atmospheric 502
Lessen’s results 141 fluctuation 507
Lettau’s approach 373 humidity profile 490
Lettau’s theory 410 interferometric 182
Lewis’ observation 104 turbulent flux 492
Liljequist’s observations 483, 502 Mechanical energy, rate of production of
Lin-Benney results 201 525
Linear disturbance theory 110, 114, 134, Meksyn-Stuart method 195
147, 1 5 6 ,1 6 1 ,1 7 0 ,1 7 1 ,1 7 9 ,2 0 2 Meksyn’s analysis 202
Linear instruments 507 Meteorological variables 421
Linear stability theory 156, 160, 165, 166, drift of mean of 420
182, 185,186, 191,195,201 profiles of 454
Lin’s calculations 129,130 measurement of 509
Lin’s method 130 Meyer’s numerical solution 169
Liquid metals, experiments on heat transfer Michalke-Timme model 197
in 337 Michalke’s calculations 124
Lock’s calculation 130 Mickelsen’s data 604
Logarithmic layer 274 Microanemometer, acoustic 505
theory 427,4 28 ultrasonic 505
Logarithmic wall law 317 Millikan’s deductions 304
Lyapunov, stability in the sense of 157, Miles’ result 125,126
158,1 59,1 60 Mixing length, concept of 366, 369
Lyapunov-Schmidt analytical method 169 Mixing-length hypothesis 578
Lycopodium spores, atmospheric diffusion Moisture, exchange coefficient for 429
of 588 Molecular collisions 259
Lysimeters, weighing 508 Molecular diffusion 534, 580, 583, 584,
591, 593, 599, 601, 602, 604, 605,
606, 621,624, 625
Macdonald function 617 acceleration of 599
Macroturbulence, statistical description of effect of 600
386 effect of 670
McVehil’s observations 483,484 theory of 678
Malkus’ convection theory 4 96,497 Molecular diffusivity 329, 332
Malkus’ theory 18, 21 Molecular ensemble, statistical mechanics of
Malkus’ theory of turbulent convection 426 4
Many-particle systems 4 Molecular fluctuations 621
Markov function 677, 678 Molecular motion 591
Markov model 611 Molecular thermal diffusivity, concept of a
Markov nature of a function 682 sublayer of 339
Markov random function 609, 610, 612, Molecular transport coefficients 28
677 Molecular viscosity 329, 335, 343, 358,
Markov random processes, differential equa 37 6,381,411
tions of 679 forces of 261
theory of 582, 609 Molecules, mean free path of 10
Markovian function 611 root-mean-square velocity of 10
Mass transport equation 361 Moment tensor, three-dimensional 226
Mathematical analysis, methods of 206 Moments, central 230, 232, 234
Mathematical physics, parabolic cylinder higher-order 2 2 5 ,232 ,253
functions of 648 joint 226
SUBJECT INDEX 765
Moments (cont.) self-exciting 158

lowcr-ordcr 225, 230 softly excited 179
sccond-order 228, 231 subharmonic 197
spacc 221 origin of 197
time 227 theory of 89
Momentum, eddy diffusivities for 442 nonlinear 183
equation of conservation of 28 Oscillators, nonlinear 88
law of conservation of 352
Momentum defect 360
Momentum flux, equality of 352 Pair interactions 188
Momentum thickness 51 Palm’s theory 189
Momentum transfer theory 368, 370 Pandolfo equations 479
Momentum transport, coefficient of 262, Pandolfo theory 454
263 Parabolic cylinder functions 649
Monin-Obukhov formula 455 Parabolic diffusion equation 685, 688,690,
Monin-Obukhov method 475,509 692
Monin’s equation 389 Pasquill data 500
Monin’s theory 410 Pasquill’s results 331
Mordukhovich-Tsvang estimates 424 Passive admixture, eddy diffusivity for 591
Mordukhovich-Tsvang measurements 521 Passive admixture assumption 57
Myrup’s measurements 520 Photons, one-dimensional diffusion of 680
Physical theory, statistical approach to 4
Plane-parallel flows 119
National Bureau of Standards 133, 134 inviscid 124
Navier-Stokes equations 30, 52, 58, 59, 148, stability of 121, 122
153, 160, 177, 210, 260, 375, 377, two-dimensional disturbances of 114
531,532 Poiseuille flows 52, 112, 127, 128, 131,
Lagrangian formulation of 531 147, 165, 166, 192, 193, 194,200
Nikuradse’s data 50, 276, 289, 294, 303 circular 37, 143, 145, 147, 165
Nikuradse’s results 304 laminar 146
Nissan-Nardacci-Ho results 106 parabolic profile of 165
Nonconvective flows 154 stability of 142
Nonlinear processes, analysis of 198 laminar, parabolic profile of 304
Nonlinear theory 160 plane 35, 101, 129, 130, 137, 166, 191,
Non-Newtonian fluid, replacement o f fluid 194, 200,201
by 107 stability of 127, 138
Non-Newtonian viscoelastic behavior 373 Poisson distribution 683
Poisson’s equation 30, 378
Potential temperature 91, 92
Oceanographic observations, data from 502 Power friction law 310
Ogura’s eddy viscosity hypothesis 452 Power-law models, two-layer 661
Ogura’s semiempirical theory 452 Prandtl-Lcttau equation 409
Orr-Sommerfeld equation 115, 117, 118, Prandtl mixing length 378, 389
124, 126, 127, 128, 131, 132, 139, turbulent 370
140, 141, 142, 160 Prandtl-Taylor equation 342
eigenfunctions of 137 Prandtl-Taylor model, two-layer 342, 343
linear 192 Prandtl-Taylor theory 341,342
Orthogonal functions 86 Prandtl’s boundary layer equations 45
Oscillations, damped 129 Prandtl’sdata 326
fundamental 197 Prandtl’s equation 369, 372
periodic 162 Prandtl’s formula 409
steady 163 PrandtPs momentum transfer theory 366,
quasi-periodic 164 367,370
766 SUBJECT INDEX
Prandtl’s results 325 application of 309

Prandtl’s scniiempirical theory 373 extended 332
Prandtl’s theory 369,371 principle of 312,351
Probability, mathematical theory of 203 Reynolds numbers
Probability averaging, definition of 536 concept of 39
Probability density function 211 critical 115, 116
Probability distribution 211,213 large 38,39
moments of 222 flows with 41
Probability theory 7 ,2 1 1 ,2 5 0 ,2 5 1 monotonically increasing 104
central limit theorem of 619 small 38,39
techniques of modem 215 supercritical 114
Profile measurements 510 theory of 12
Profiles, longitudinal velocity 47 transitional 363
vertical velocity 4 7 ,4 8 Reynolds-Peclet number similarity principle
wind-velocity 450 332
Proudman’s oceanographic data 503 Reynolds-Potter computations 192, 193,
Psychrometric equation 506 194,195
solving 506 Reynolds stress tensor 261, 373, 376, 390
Reynolds’ similarity law 40,41
Richardson numbers 91, 92, 125,402,416,
Quantum field theory 4, 19 428, 429, 432, 436, 437, 441, 447,
4 5 6 ,4 7 6 ,4 7 9 , 488,490, 495, 646
empirical functions of 454
Random fields, homogeneous 250, 252 flux 400
theory of 5 Richardson’s assumptions 12
Random functions, stationary 245 Rider’s data 4 9 1 ,4 9 2 ,5 0 0
mathematical theory of 245 Rider’s results 331
theory of 203 Riemann method 691
Random processes, stationary 245, 246, 251 Roberts equations 658, 660
ergodic theorem for 251 Roberts theory 659
Random variable, defined 211 Rolls, two-dimensional 186, 187, 190
Rayleigh equations 117, 118, 120, 123, Rosencrans’ results 123
124,143,144 Rossby number 4 0 9 ,410 ,411
Rayleigh model 182 Rotta’s measurements 74
Rayleigh-Taylor instability theory 124 Rotta’s works 379
Rayleigh theorem 119, 120, 121, 122 Roughness length 287
Rayleigh’s criterion of instability 97 Round Hill observations 524
Regularity conditions 217
Reichardt’s data 354, 355
Resistance thermometers, wet- and dry-bulb Saffman’s admixture correlation function
506 601
Resistance-thermometer instruments 506 Saffman’s results 655
Reynolds analogy 341, 626, 630, 631, 662 Sato-Sakao experimental data 140
Reynolds conditions 207, 209 Sattinger’s results 123
Reynolds criterion 74 Scanlon-Segel theory 187
Reynolds equations 11, 260, 261, 263, 265, Schade’s analytical result 196
268, 269, 352, 354, 356, 358, 361, Schlichting’s interpolation formula 326
364, 365, 373, 374, 375, 377, 382, Schliiter-Lortz-Busse method 179, 180
3 9 4 ,4 0 5 ,4 1 1 ,4 1 4 Schmidt conjugate power law 661, 663
Reynolds experiments 75, 580 Schmidt number 60, 61, 283, 345, 347
Reynolds investigations 75 Schubauer-Skramstad experiments 133,
Reynolds number of flow 38 134, 135,198
Reynolds number similarity 299 Schubauer-Skramstad transformation 135
SUBJECT INDEX 7 67
Schultz-Grunow data 322, 323 Stability theory 187

Schultz-Grunow law 324 linearized 178
Schwinger equations 19 Stationary process, multidimensional 246
Sedimentation theory 415 Statistical ensemble 7, 210, 421
diffusion approximation in 415 Steady sources, linear 662
Segel’s calculations 182 Stirling numbers 591
Self-excitation, hard 88 Stratification, near-neutral 431, 467, 469,
soft 88 471,5 1 2 ,5 2 1
Self-preservation hypotheses 349 neutral 91, 417, 418, 4 3 0 ,4 3 1 ,4 3 2 ,4 3 3 ,
local 372 448, 449, 456, 458, 460, 466, 487,
Semicircle theorem 118, 126 488, 489, 502, 518, 560, 646, 662,
Semiinvariants 224, 230 663, 665,666, 6 7 4 ,6 9 1 ,6 9 2
Serrin-Velte-Sorger results 154 nonneutral 564, 663, 675
Shadow-graph method 315 conditions of 452, 455
Shape assumption 179 stable 91, 92, 430, 433, 437, 438, 448,
Shear flow, diffusion in 632 450, 451, 460, 473, 476, 482, 487,
particle dispersion in 556 502, 512, 516, 643, 646, 661, 671,
Shear stress, turbulent 426 692
viscous 262 temperature 4 1 8 ,4 5 0 ,4 5 4 ,4 5 5
Silveston experiments 191 thermal 431, 455, 468, 475, 482, 484,
Similar flows (see Flows) 5 0 4 ,5 1 8 ,5 6 9 ,6 4 5 ,6 4 7 ,6 7 1
Similarity criteria 3 8 ,4 1 ,5 9 stable 569
Similarity equation 662 unstable 91, 92, 430, 433, 435, 454,
Similarity theory 455, 456, 469, 482, 490, 460, 468, 473, 474, 487, 488, 494,
502, 504, 509, 517, 518, 524, 646, 495, 496, 502, 512, 523, 567, 569,
676, 692 6 4 3 ,6 4 6 ,6 7 1 ,6 9 3
general formulas of 490 Stream function, equation for 118
verification of deductions from 504 Streamline, displacement of 51
Skewness factor 224,462 schematic form of 50
Skin friction coefficient 33, 34, 36, 43, 49 Streamwise-vortex system, development of
dimensionless 35,48 199
dynamic measurements of 323 Stress tensor, turbulent 11
values of 49 viscous 261
Skin friction law 257 Stuart-Watson analysis 192
Small disturbances, method of 92, 97, 112 Stuart-Watson expansion procedure 197
Small disturbance theory 66, 69, 142, 147, Stuart’s calculations 172
148,156 Superposition, principle of 584
Smith’s duality theorem 664 Surface layer, atmospheric diffusion in 643
Smoluchowski equation 609 Suspended particles, interaction of 412
Snyder-Karlsson results 106 motion of 412
Snyder-Lambert results 175 vertical distribution of 415
Soprunenko’s numerical results 139 Sutton’s theory 658, 659
Sorger’s results 150 Swinbank’s data 491 ,492
Space-time points 7, 8 Swinbank’s exponential functional 676
Spalding’s equation 282, 310 Swinbank’s exponential wind profile 662
Splashes 440 Swinbank’s formula 453
Spots, formation of turbulent 363 Swinbank’s graph 499
Stabilities, principle of exchange of 101, Swinbank’s measurements 521
116,126,129
Stability, quantitative investigation of 200
Stability problem, equations of neutral 196 Taylor length microscale 604
mathemalical formulation of 92 Taylor microscale of turbulence 84
Stability surface, neutral 195 Taylor numbers 105, 173, 175, 176
768 SUBJECT INDEX
Taylor series 225, 242, 272, 33 8,5 7 1 ,5 7 5 , large-scale 10, 15, 16, 17
595,632 small-scale 10, 15, 16, 17
coefficients of 433 in compressible medium 52
loss of stability of 104 dynamics of 9
nonlinear development of 176 effect of radiative heat transfer on 424
stability of 178 Eulerian properties of 548
unstability of 178 Eulerian statistical characteristics of 538,
Taylor’s calculations 102 540, 541 ,5 4 7 ,5 5 4
Taylor’s classical formula 657 history of 5 ,6
Taylor’s data 171,627 homogeneous 13, 249
Taylor’s equations 587, 602 initiation of
Taylor’s experiments 627 mechanism of 166
Taylor’s method 10 1,627,629 nonlinear theory of 148
Taylor’s oceanographic data 502, 503 isotropic 13,249
Taylor’s result 83,619 concept of 13
Taylor’s scheme 133 mathematical technique of 14
Taylor’s semiempirical theory 373 kinetic energy of 397
Taylor’s theory 8 9 ,3 6 9 ,3 7 0 ,3 7 1 transformation of 397
Taylor’s vorticity transfer theory 368, 370 Lagrangian characteristics of 24, 554
formulation of 369 statistical 535,536
Taylor’s works 621 length scale of 449
Temperature defect law 333 marine 208
Temperature dissipation 387 mathematical theory of 258
Temperature-inhomogeneous fluid, motion problem of 7 , 8 , 9
of 58 analytical formulation of 9
Temperature profile, linear 502 semiempirical theories of 4, 10, 11, 12,
Temperature-profile data 511 354, 3 6 4 ,3 7 8 ,4 4 3 ,4 4 9 ,4 6 5
Temperature-profile observations 495 formulation of 10
Telegraph equation 680, 681, 682, 691, 692 statistical description of 535, 583
Thermal conductivity equation 423 in stratified medium 452
Thermal convection, concept of 462 semiempirical theory of 452
nonlinear theory of 110 theory of 3, 4, 5, 6, 7, 8, 9, 10, 18, 19,
Thermal convection problems, exchange of 22, 206, 207, 209, 214, 215, 216,
stabilities in 156 225, 243, 249, 251, 258, 259, 261,
Thermal expansion, coefficient of 58 266,528
Thermal stratification (see Stratification) applications to 250
Thermometric conductivity, coefficient of 55 statistical approach to 209
Time scale, integral 252 transition to 191
Tollmien-Schlichting-Lin theory 198 Turbulent diffusion 580, 593, 599, 601,
Tollmien-Schlichting theory 132, 133 602, 6 0 4 ,6 0 5 ,6 4 0 ,6 6 4
Tollmien’s calculations 132 description of 685
Total probability, theorem of 538, 610 one-dimensional
Townsend’s results 496 semiempirical equation of 682
and 1/3-power law 496 semiempirical equations of 606, 607, 612,
Transition, laminar-turbulent 167 6 1 3 ,6 5 1 ,6 6 0
Transition phenomena, mathematical theory semiempirical theory of 559, 611, 658,
of 168 666 ,6 7 8 ,6 8 8
Tsvang’s measurement, data of 522 study of 361
Turbidity, defined 412 theoretical 632
low 413 theory of 24
Turbulence vertical 637,639
atmospheric 2, 208, 210 Turbulent elements, vertical displacements
components of 10 of 397
SUBJECT IN D E X 769
I'lirbnlcnt energy 418, 421 Volume scale, integral 252

diffusion of 381 Volumetric concentration, Kulerian field of
cffect of 414 581
vertical 409 Von Karman-Pohlhausen theory 83
equation of balance of 374 Von Karman’s constant 276, 310, 318,407,
Turbulent fluxes, methods of measuring 504 4 3 3 ,4 8 2 ,5 6 1 , 631
Turbulent jets 352 diniensionless 328,427
Turbulent medium, wave propagation in 64 Von Karman's equation 306, 325
1 urbiilent motion 3 Von Karman’s expression 282
theory of small-scale 13 Von Kiirman’s hypothesis 3 7 0 ,3 7 2 ,4 1 1 ,4 5 3
Turbulent phenomena, theoretical analysis Von Karman relation 372
of419 Von Karman’s skin friction law 324
Turbulent regime, similarity hypothesis for Von Karman’s theory 342, 370
51 1 Von Karman’s transitional layer 341
Turbulent shear stress, fluctuating method Von Karman’s universal constant 509
for measuring 508 Vorticity transfer theory 370
Turbulent slugs 74
formation of 363 Wake function 319
Turbulent spots, development of random Wake law 320
199 Wakes, three-dimensional 360, 361, 550,
Turbulent stress tensor 11 5 5 1 ,63 5,637
Turbulent stresses 272 two-dimensional 361, 550, 551, 635, 637
Turbulent viscosity, coefficients of 10, 11 Wall, law of 266, 311,317,411
Turbulent wake, formation of 79 Wandel and Kofoed-Hansen turbulence
model 579
Yager’s semiempirical theory 453 Watson’s method 183
Van Driest’s distribution 368 Waves, neutral 116
Variational calculus, rules of 159 three-dimensional 198
techniques of 155 nonlinear amplification of 198
Velocity defect law 298, 300, 304, 312, Webb’s data 483
313, 317, 318, 321,411, 625 Webb’s proposal 447
logarithmic form of 315,631 Webb’s result 495
verification of 313 Wieghardt experiments 317
Velocity lluctuation, fine-scale 505 Wind, power law height dependency of 661
Velocity profiles, hyperbolic-tangent 142, Wind equations, geostrophic 406
196 Wind profile, semiempirical formulas for
nonquadratic 136 454
Vibrating-ribbon tech^nique 199 Swinbank’s exponential 662
Viscoelastic behavior, non-Newtonian 373 universal 662
Viscosity, coefficient of 28, 29 Wind-profile functions, universal 480,481
eddy 262,263 Wind-tunnels, roughness elements in 294
coefficient of 262 Wind-tunnel measurements 16
effective 263 Wind velocity, formula for 452
equivalent 286 geostrophic 406,411
kinematic 28 mean values of 421
moleciilur 263 profile 452
total 263 vertical component of 2
turbulent 262
coefficient of 262 Yaglom’s equation 655
Viscosity forces, transfer of energy molecu Yamamoto-Shimanuki equations 453, 454
lar 376
Viscous fluid, general theory of 126
Viscous interaction constant 532 Zel’dovich’s equations 435
Viscous sublayer 270, 272, 278, 288 Zilitinkevich-Chalikov nomograms 512

Statical Fluid Mechanics MONIN

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Statical Fluid Mechanics MONIN

Uploaded by

Copyright:

Available Formats

STATISTICAL

A. S. Monin and A. M. Yaglom

This is a translation of the first Russian edition of CxaTHCTHHecKaa

The theory o f turbulence discussed in this book is based on the usual

Authors’ Preface to the English Edition iii

Chapter 1 Laminar and Turbulent Flows 27

Cluipter 2 Miithematical Description of Turbulence.

6.4 Energy Balance in a Compressible Fluid 393

Chapter 4 Turbulence in a Thermally Stratified Medium 417

Chapter 5 Particle Dispersion in a Turbulent Flow 527

10. Turbulent Diffusion 579

This book is devoted to the mechanics of turbulent flows o f liquids

The indicated properties of turbulent flows may evidently be very

FIG. 1. Typical example of a recording of fluctuations of temperature, wind

In particular, the various motions o f the air in the earth’s

principle impossible. Moreover, such description would not be useful

turbulence there are only partial successes; consequently, the major

Let us now turn to a brief history o f turbulence. We shall

shall try to explain why it seems worthwhile to us to write such a

deductions still remain valid, since the properties o f averaging which

narrower (i.e., a less complete) formulation of the problem o f

theory o f turbulence to be devoted not to the problem of turbulence

im portant results of great practical value have been obtained in the

viscosity, thermal conductivity and diffusion. Using the stated

The semiempirical theories of turbulence are valuable for solving a

Big whorls have little whorls,

Richardson put forward these general ideas only in qualitative

flows (three-dimensional homogeneity assumes, in particular, that

viscosity, which is concentrated principally in the range of minimum

deductions obtained concerning the statistical properties of small-

components o f the velocity field play a considerably greater part

functional derivatives, and its most remarkable feature is linearity.

We conclude this introduction by discussing briefly the contents

complicated statistical characteristics. We shall confine ourselves

less-known applications o f the semiempirical theories and considered

this to give the reader a sufficiently complete picture of the present

Therefore, a knowledge of these two subjects will greatly assist the

The second volume of the book will begin with a mathematical

1. EQUATIONS OF FLUID DYNAMICS AND

Here and henceforth, we shall always adopt Einstein’s summation

same meaning as ^ The fundamental dynamic equations

expressing Newton’s second law apphed to a small volume of fluid,

which are the fundamental dynamical equations o f a viscous

are the components of the vorticity vector. In principle we may

which is obtained by applying the operation to Eqs. (1.6) [with

4it ./ dx^dx^ \x — x'\

where the integration is taken over the whole volume occupied by

We note first, that in the case of a steady flow the- total

where n, is the component of the unit vector of the outward normal

Examples o f exact solutions o f the fluid-mechanical equations

1. We begin with the simplified case o f a two-dimensional flow of

where u (2 ) = «i (2 ) is the only nonzero component of the velocity of

Thus the flow is described by a linear velocity profile, while the

On the plane 2 = 0 this force is directed along the Ox axis, while on

where Cf is the skin friction coefficient, then

The steady plane flow of fluid between infinite parallel planes

2. Let us now consider a steady two-dimensional flow between

It is clear therefore that in this case the pressure will be independent

- ^ = const , u{z) = — - ^ - ^ z ‘^ -\-a z-\-b, (1.18)

where Aip is the drop in pressure between the planes x = xo and x =

Hence, the dimensionless skin friction coefficient C/ of Eq. (1.15)

(because in the present case

The idealized flow described by the exact solution (1.19) is

3. We now turn to the case of a steady flow of fluid in a straight,

this equation after transformation to dimensionless form. Conse

where X — is the coefficient of thermal diffusivity (or thermo