This action might not be possible to undo. Are you sure you want to continue?
TOPICS IN C H E M I C A L E N G I N E E R I N G A S ERIES O F T E X TB O O KS AN D M ON OGRAPHS
SERIES EDITOR Keith E. Gubbins Cornell University
ASSOCIATE EDITORS + Mark A. Barteau, University of Delaware Edward L. Cussler, University of Minnesota Klavs F. Jensen, Massachusetts institute of Technology Douglas A. Lauffenburger, Massachusetts Institute of Technology Manfred Morari,ETH W. Harmon Ray, University of Wisconsin William B. Russel, Princeton University
ANALYSIS OF TRANSPORT PHENOMENA
WILLIAM El. DEEN
Massachusetts Institute of Technology
SERIES TITLES Receptors: Models for Binding, Traficking, and Signalling D. Lauffenburger and J. Linderman Process Dynamics, Modeling, and Control B. Ogunnaike and W. H. Ray Micmstructures in Elastic Media N. PhanThien and S. Kim Optical Rheometry of Complex Fluids G. Fuller Nonlinear and Mixed Integer Optimization: Fundamenlals and Applications C. A. Floudas Mathematical Methods in Chemical Engineering A. Varma and M. Morbidelli The Engineering of Chemical Reactions L. D. Schmidt Analysis of Transport Phenomena W. M. Deen
New York Oxford OXFORD UNIVERSITY PRESS 1998
OXFORD UMVERSITY PRESS
Oxford New York
Athens Auckland Bangkok Bogotii Bombay Buenos A k s Dar es Salaam Delh Florence Hong Kong Istanbul Karachi Kuala Lumpur Madras Madrrd Melbourne Mexico City Nairobi Paris Singapore Taipei Tokyo Toronto Warsaw
CrJcutta Cape Town
To Meredith and Michael
and associated companies in Berlin Ibadan
Copyright 8 1 9 9 8 by Oxford University Aess, Inc. Published by Oxford University Press, Inc., 198 Madison Avenue, New York, New York 10016 Oxford is a registered trademark of Oxford University Press.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recordmg, or otherwise, w i t h o u t the prior permission of Oxford University Press.
Library of Congress Catalo@nginPublication Data Den,Wiham M. (William Murray), 1947Analysis of transport phenomena ) William M. D a n . cm. (Topics in chemical engineering) p. Includes bibliographical references and index. ISBN 0 195084942 1 . Transport theory. 2. Chemical engineering. I. Title. 11. Series: Topics in chemical engineering (Oxford University Press) TF'156.nD44 1 9 9 8 6 6 0 '. 2 8 4 2 4 ~12 9731237
cm
Rinted in the United States of America on acidfree paper
CONTENTS
PREFACE
xiii
Chapter 1
DIFFUSIVE FLUXES AND MATERIAL PROPERTIES
1.1 Introduction 1.2 Basic Constitutive Equations 1.3 Difisivities for Energy, Species, and Momentum 1.4 Magnitudes of Transport Coefficients 1.5 Molecular Interpretation of Transport Coefficients 1.6 Continuum Approximation References Roblems
Chapter 2 CONSERVATION EQUATIONS AND THE FUNDAMENTALS OF HEAT AND M A S S TRANSFER
2.1 Introduction 2.2 General Forms of Conservation Equations 2 3 Conservation of Mass 2.4 Conservation of Energy: Thermal Effects 2.5 Heat Transfer at Interfaces 2.6 Conservation of Chemical Species 2.7 Mass Transfer at Interfaces 2.8 OneDimensional Examples 2.9 Species Conservation from a Molecular Viewpoint References Problems
Chapter 3
SCALING AND APPROXIMATION TECHNIQUES
3.1 3.2
Introduction Scaling
CONTENTS
Contents
ix
Reductions in Dimensionality Simplifications Based on Time Scales Similarity Method 3.6 Regular Perturbation Analysis 3.7 Singular Perturbation Analysis 3.8 Integral Appproximation Method References Problems
33 3.4 3.5
6.3
6.4
Steady Flow with a Moving Surface TimeDependent Flow 6.5 Limitations of Exact Solutions 6.6 Lubrication Approximation References Problems
Chapter 7
Chapter 4 SOLUTION METHODS FOR CONDUCTION AND DIFFUSION PROBLEMS
Introduction Fundamentals of the Finite Fourier Transform (FFT) Method Basis Functions as Solutions to Eigenvalue Problems 43 4.4 Representation of an Arbitrary Function Using Orthonormal Functions 4.5 FFT Method for Problems in Rectangular Coordinates 4.6 SelfAdjoint Eigenvalue Problems and SturmLiouville Theory 4.7 FET Method for Problems in Cylindrical Coordinates 4.8 FFT Method for Problems in Spherical Coordinates 4.9 PointSource Solutions 4.10 Integral Representations References Problems
CREEPING FLOW
7.1
4.1 4.2
7.2 7 . 3 7.4 7.5 7.6
Introduction General Features of Low Reynolds Number Flow Unidirectional and Nearly Unidirectional Solutions Stream Function Solutions PointForce Solutions Particle Motion and Suspension Viscosity 7.7 Corrections to Stokes' Law References Problems
Chapter 8

LAEllNAR FLOW AT HIGH REYNOLDS NUMBER
8.1 8.2 83 8.4 Introduction General Features of High Reynolds Number Flow Irrotational Flow Boundary Layers Near Solid Surfaces 8.5 Internal Boundary Layers References Problems
Chapter 5 FUNDAMENTALS OF FLUID MECHANICS
Introduction Fluid Kinematics Conservation of Momentum Total Stress, Pressure, and Viscous Stress Fluid Statics Constitutive Equations for the Viscous Stress Fluid Mechanics at Interfaces Dynamic Pressure Stream Function 5.10 Nondimensionalization and Simplification of the NavierStokes Equation References Problems
Chapter 9
FORCEDCONVECTION HEAT AND MASS TRANSFER IN CONFINED LAMINAR FLOWS
9.1 Introduction 9.2 Pkclet Number 9.3 Nusselt and Sherwood Numbers 9.4 Entrance Region 9.5 Fully Developed Region 9.6 Conservation of Energy: Mechanical Effects 9.7 Taylor Dispersion References Problems
Chapter 6
UNIDIRECTIONAL AND NEARLY UNIDIRECTIONAL FLOW
6.1
6.2
Introduction Steady Flow with a Pressure Gradient
CONTENTS
Contents
rd
Chapter 10
Appendix
FORCEDCONVECTION HEAT AND MASS TRANSFERIN UNCONFINED LAMINAR FLOWS
Introduction Heat and Mass Transfer in Creeping Flow Heat and Mass Transfer in Laminar Boundary Layers Scaling Laws for Nusselt and Sherwood Numbers References Problems
10.1 10.2 10.3 10.4
VECTORS AND TENSORS
A.l
A . 2
Introduction Representation of Vectors and Tensors A.3 Vector and Tensor Products A.4 Vector Differential Operators A.5 Integral Transformations A.6 Position Vectors A . 7 Orthogonal Curvilinear Coordinates A.8 Surface Geometry References
Chapter 1 1

MULTICOMPONENT ENERGY AND MASS TRANSFER
Introduction Conservation of Energy: Multicomponent Systems 11.3 Simultaneous Heat and Mass Transfer 11.4 Introduction to Coupled Fluxes 11.5 StefanMaxwell Equations 11.6 Generalized Diffusion in Dilute Mixtures 11.7 Transport in Electrolyte Solutions 11.8 Generalized ~ t e f k ~ a x w e Equations ll References Problems
INDEX
583
1 1 . 1 11.2
Chapter 12
TRANSPORT IN BUOYANCYDRIVEN FLOW
Introduction Buoyancy and the Boussinesq Approximation 12.3 Confined Rows 12.4 Dimensional Analysis and Boundary Layer Equations 12.5 Unconfined Flows References Problems 12.1 12.2
Chapter 13
TRANSPORT IN TURBULENT FLOW
Introduction Basic Features of Turbulence TimeSmoothed Equations 13.4 Eddy Diffusivity Models 13.5 Other Approaches for Turbulent Flow Calculations References Problems
1 3 . 1 13.2 13.3
Although the book tries to bridge the gap between introductory texts and the research literature. It is a byproduct of nearly 20 years of teaching a onesemester course that is one of three "core" offerings required of all firstyear graduate students in chemical engineering at MIT. the book goes beyond the bare essentials and also beyond what can be presented in a onesemester course. heat. Transport is one of the more mathematical subjects in engineering. where most chemical engineering students have not had a prior graduatelevel course in applied mathematics. diffusion of chemical species. a significant fraction (about 20%) is devoted to a presentation of the additional methods which are needed. the mathematical level is more advanced than in most undergraduate texts. and then they are used to obtain the governing equations for total mass. and to prepare students to solve a wider variety of problems and to understand the literature. any additional courses in transport are taken as electives. and viscous transfer of momentum. ORGANIZATION AND USE This book is based on the premise that there are certain concepts and techniques which apply almost equally to momentum. To provide a more rounded view of the subject and allow more teaching flexibility. and mass transfer and that at the graduate level these common themes deserve emphasis. and' chemical species. Thus. The conservation equations for scalar quantities are derived first in general form in Chapter 2.PREFACE SCOPE AND PURPOSE This book is intended as a text for graduatelevel courses in transport phenomena for chemical engineers. it is intended for all students at the graduate level. not just those who will pursue transportrelated research. energy. The scaling and orderofmagnitude concepts which are crucial in modeling have received . Chapter 1 highlights the similarities among the "molecular" or "diffusive" transport mechanisms: heat conduction. the selection of material for this course has represented an attempt to define the elements of the subject that anyone with a graduate degree in chemical engineering should know. For both master's and doctor's degree candidates. It is assumed that the reader has a good grounding in multivariable calculus and ordinary differential equations. To make the book as selfcontained as possible. This integration of mathematical methods with transport reflects the situation at MIT. but not necessarily any background in solving partial differential equations.
.
per unit mass.. Diameter. vectors are bold Roman (eg. Heat capacity at constant pressure.g. directed from 1 to 2 or A to B. M a s s flux of species i relative to the massaverage velocity. scalars are italic Roman (eg. Pseudobinary diffusivity for species i in a dilute mixture. Molar concentration of species i. Heat transfer coefficient. Molar flux of species i relative to fixed coordinates.. In general. Binary diffusivity for species i and j. L Mi n n . Unit (base) vector associated with coordinate i. based on molar concentrations. Molar flux of species i relative to the massaverage velocity.. Rate of energy input per unit volume from an external power source. Mass flux of species i relative to fixed coordinates. Enthalpy per unit mass. Many which appear only in one chapter are not included.f ).. Partial molar enthalpy for species i. . Total molar concentration. Molecular weight of species i. Unit vector normal to a surface (outward from a control volume). 7). Avogadro's number. Mass transfer coefficient for species i. for the x component of v). DamMhler number for reaction relative to diffusion. Vector and tensor components are identified by subscripts (e. Characteristic length. Ni Biot number. see the Appendix. Brinkman number. n.The magnitude of a vector or tensor is usually represented by the corresponding italic letter (e. NA. Grashof number.v). Thermal conductivity. and tensors are bold Greek (e. Unit vector normal to an interface (phase boundary).LIST O F SYMBOLS Following is a list of symbols used frequently in this book.g.v for v and T for 7). Roman Letters Bi Br C Ci d 6 Da Di Dij ei g Gr h I? Ri Hv ji Ji k kc. For more information on vector and tensor notation and on coordinate systems.g.. v. Gravitational acceleration.
 Greek Letters Thermal diffisivity. Rayleigh number. Thermodynamic pressure. Position vector. Orderofmagnitude equality. Surface (especially that enclosing a control volume). Velocity of a point on an interface (phase boundary). Massaverage velocity. Energy flux relative to the massaverage velocity. Mole fraction of species i in a mixture. V2 Laplacian operator. Vorticity. Sherwood number for species i. Latent heat per unit mass. Identity tensor. the outer velocity evaluated at the surface (Chapters 8 and 10). Kinematic viscosity. Volume (especially that of a control volume). V Gradient operator. Rate of formation of species i per unit area (heterogeneous reactions). Temperature. Viscous stress tensor. Rateofstrain tensor. Velocity of species i. Alternating unit tensor. in turbulence. Stream function. Prandtl number. k / dP.WST OF SYMBOLS LIST OF SYM~O& Nu Nusselt number. . Time. the fluctuating part of the velocity (Chapter 13). Radius (usually) or universal gas constant (Chapters 1 and 11). Mass concentration of species i. Mass fraction of species i in a mixture. Special Symbols D/Dt Material derivative. Partial molar volume for species i. CI/p. Volumetric flow rate. Internal energy per unit mass. In boundsry layer theory. Total mass density. Velocity of a point on the surface of a control volume. Sbress vector. P6cldt nymber. %SCOUS dissipation function. Reynolds number. Rate of formation of species i per unit volume (homogeneous reactions). Total stress tensor. Schmidt number. Viscosity. Penetration depth or boundary layer thickness. Characteristic velocity (often the mean velocity). Dynamic pressure.
and momentum each have two principal components: the convective transport. in that it implies that energy. along with the analogous relationship for vis . the same is true for gradients in velocity. these relationships remain largely empirical. Equilibrium states of matter are therefore characterized by the absence of spatial gradients in temperature and species concentration. and in some cases the numerical values of the coefficients. 1 INTRODUCTION If the temperature or the concentrations of individual chemical species in a material are perturbed so that they become nonuniform. and momentum tend to move from regions of higher to lower concentration. Although the forms of various constitutive equations. as well as a flux of a given species in a mixture which results from a gradient in the concentration of that species. Within a fluid. the establishment of suitable constitutive equations is one of the central aspects of the analysis of transport phenomena. The spontaneous dissipation of such gradients is a fundamental and farreaching observation. respectively. matter. can be rationalized using molecular models. The types of constitutive equations which arise in energy and mass transfer are summarized in Table 11. which automatically accompanies any bulk motion. The constitutive equations which are ordinarily used to describe these fluxes are the "laws" named after Fourier and Fick. the resulting gradients tend to disappear over time.Chapter 1 DIFFUSIVE FLUXES AND MATERIAL PROPERTIES I . The purpose of constitutive equations is to relate these fluxes to local material properties. the molecular transport processes are specific to a given material or class of materials and therefore require much more attention. Accordingly. These basic constitutive equations. matter. of course. The fluxes of energy. There is. and the molecular or difisive transport. an energy flux which results from a temperature gradient. which is related to smallscale molecwlar displacements. Whereas convection is relatively easy to describe in a general manner.
Although Fourier's taw and Fick's law are used to model energy and mass transfer throughout most of this book. then the flux normal to the surface is 1. Consider an imaginary surface which is perpendicular to a vector n of unit magnitude.scalar coefficient. His masterpiece was a paper submitted in 1807 to the institut de France. giving A schematic representation of the heat flux vector is shown in Fig. 1l(b). pressure. Materials for which this is true. modem idea concerning the dimensions and units physical quantities. and composition. 9399). the heat flux normal to the surface (i. and diffusion of a given species due to gradients in temperature. the energy flux relative to fixed coordinates has a contribution also from convection (bulk flow). n is the unit normal vector which describes the orientation of the surface. An example is wood. 5. In it he not only gave the fim comet statement of the differential equation for bansient conduction in a solid. the flux parallels the temperature gradient.2 through A. electrical potential. That is. these include energy transport due to concentration gradients. q only represents energy transfer relative to the local massaverage velocity. As shown in Table 11. and simplified molecular models are presented to provide a semiquantitative explanation for the magnitudes of the coefficients and their dependence on temperature. the scalar conductivity must be replaced by a conductivity tensor ( K ) . or the concentrations of other species. Schematic of the heat flux q at a surfacewith a unit normal a for (a) a surface with arbiwary orientation and (b) a surface perpendicular to the y axis. k.TABLE 11 Energy and Species Fluxes Resulting from Gradients in Various Quantities  Gradlent Energy flux Conduction (Fourier) Diffusionthermo effect (Dufour) Flux of species i Temperature Concentration of species i Concentration of species j (f i ) Electrical potential Pressure Thermal diffusion (Soret) Diffusion (Fick) Multicomponent diffusion (StefanMaxwell) Ion migration (Nern stPIanck) Pressure &ffusion cous m s f e r of momentum. Moreover. there are contributions to q beyond that shown in Eq. but also determined the temperature in seven1 p h l e m s using the technique of eigenfunction expansions (Fourier ~efies). the surface happens to be perpendicular to the y axis of a rectangular coordinate system.e. In a moving fluid. The constitutive equations are presented using the general vectortensor notation described in the Appendix. pp. and the integal symbol (Ravetz and rattanGuinness. where k is the thermal conductivity.' According to Fourier's law.4. Materials which are anisotmpic have an internal structure which makes the thermal conductivity depend on direction. (unit vector in y direction). it must be emphasized that there are many other possible gradientflux combinations. (a) (b) Figure 11. ' J O S C P ~ Fowier (17681830) was a French civil adrmnisrrator and diplomat who punued mathematical physics i n his spare time. and T is the temperature.21) implies that there is no preferred direction for heat conduction. Typical values of the transport coefficients are given. V is the gradient operator. Additional information on the stress tensor and constitutive equations for momentum transfer is in Chapter 5. are termed isotropic. The general description of energy and mass transfer in rnulticomponent. Notice that the subscript y identifies q. In a mixture. is presented in Chapter 11. where conduction parallel to the grain is often several times faster than that perpendicular to it. (1. This paper established rhe main concepts of linear boundary value pmblems7which dominated h e study of diffeferrndalequations for the next cenlury Among his other conlribut o mathematics and physics wem h e Fourier integral. are the focus of this chapter.. encompassing all of the entries in Table 11. including many solids and almost all fluids. as in Fig. For a surface of arbitrary orientation. as the y component of the vector q. nonisothermal systems.2 BASIC CONSTITUTIVE EQUATIONS Heat Conduction The constitutive equation which is used to describe heat conduction can be traced to the work of J.. which he had developed for that purpose (we Chapter 4). 1l(a). in Oillispie. in Eq. The use of a single . he showed how t o exploit coordinate sYs~m~s suited to particular geomebies. (121) (see Chapter 11). Vol. . 11. in this book. the heat flux q (energy Row per unit crosssectional area) is given by which is a onedimensional form of Fourier's law. in the direction of n) is If. so that n =e. For anisotropic materials. as shown in Fig. The reader who is unfamiliar with this notation should first see Sections A. Fourier in the early nineteenth century. 1972. subscripts are not used to indicate differentiation. pressure. Accordingly.
the molar flux relative to the massaverage velocity. pp. and each . 1971. PP.Diffusion of Chemical Species The molar flux of species i relative to fixed coordinates is denoted as N. v is defined merely by selecting a fixed reference frame. the fluxes and concentration gradients of all species are interdependent (see Chapter 11). just as a force represents an overall rate of transfer of momentum. (1.. Again relative to fixed coordinates.. the velocity viof species i is defined as TABLE 12 Flux of Species i in Various Reference Frames and U n i t s for a Mixture of n Components Definitions of fluxex: Reference velocity Molar units Mass units Flux rehrionships: n Ni=Civ+ Ji= c . Accordingly. respectively. A constitutive equation to describe diffusion in a binary mixture was proposed by A. v. TABLE 13 Fick's Law for Binary Mixtures of A and B Reference velocity Mass units Molar units . in Gillispie. As shown in Chapter 5. an equal and opposite stress is exerted by the fluid on the opposite side of the test surface. j . with reference to the orientation of that test surface. The mass and mole fractions are related to the total mass density ( p ) and total molar concentration (0by Stress and Momentum Flux The linear momentum of a fluid element of mass m and velocity v is mv. and pi are the concentrations of i expressed in molar and mass units. in a mixture.. is given by the stress vector. Another reference frame which is sometimes used is the volumeaverage velocity (see Problem 12). and j/w. a stress (force per unit area) represents a fiux of momentum. = DBA(Problem 1. C N. Consider some point within a fluid. the corresponding flux in mass units is ni= Mi N. and e]ecb~physiology. In a multicomponent mixture. Vol. 1620). v ( ~which . For a di~ussion of the ea]y work on diffusion of Fick and of Thomas Graham. and xi are the mass fiaction and mole fraction of species i. was an early proponent of the rigomus application of physics to medicine and physiology (RoUlschuh. except in special circumstances (e. . Accordingly.l). s(n). defining s(n) as the stress exerted by the fluid on a particular side of the surface establishes an algebraic The various diffusional fluxes which are defined using v or are as follows: J. Fick in the midnineteenth centurya2Four forms of Fick's law are shown in Table 13 (see also Problem 12).26). Based on a steadystate mass balance for a solute (oxygen) in a conh~ousflow system ( h blood cixulation). Two useful reference frames are the massaverage velocity. see Cussler (1984.molar flux relative to the molaraverage velocity. so that species velocities are unambiguous. All of these forms of Fick's law are equivalent. The relationships among these fluxes are summarized in Table 12. 614617). a Geman physiologist. Indeed. The part of the species flux which is attributed to bulk flow and the part which is due to diffusion depend on the definition of the mixture velocity. Any number of average velocities can becomputed for a mixture. Fick's law is not applicable to multicomponent systems. respectively. The local rates of momentum transfer in a fluid are determined in part by the stresses. denoted as 4 that 0 . Notice that vi is the same for any consistent choice of flux and concentration units. In a pure fluid. and the molaraverage velocity. It can be shown contains the same binary difisivity for species A and B..where Mi is the molecular weight of i. membrane biophysics. through which passes an imaginary surface having an orientation described by the unit normal n. The "Fick pinci~le'' for computing cardiac output is still taught in modem courses in physiology.g. there are as many indepen'Adolph Fick (18291901). dilute solutions). the mass flux relative to the molaraverage velocity. so that diffusion of any pair of species is characterized by only one diffusion coefficient. The stress vector is defined as the force per unit area on the test surface. depending on how one chooses to weight the velocities of the individual species. v ( ~ ) + J ~ ( ~ ) . bioenergetics. Moreover. i= I where C. The stress at that point. are defined as dent diffusion coefficients as there are pairs of species. v must be interpreted as the massaverage velocity. it was a remarkably advanced concept when conceived. the mass flux relative to the massaverage velocity. J ~ ( the ~ . as given by Eq. 4. exerted by the fluid towad which n points. In addition to his law far m s i o n (1855). = C V ' ~ . he made pioneering theoretical and experimental conoibutions in biomechanics. and n where oi is the number of chemical Species in the mixture.
the opposite sign convention (as used by Bird et al. Vv is the velocity gradient tensor. u.e. at constant pressure.. 12(b). and (Vv)' is the transpose of Vv.. neglecting the effects of pressure). The normal component represents a normal stress or "pressure" on the surface. >0 and s. and complex fluid mixtures (e. Gases and lowmolecularweight liquids are ordinarily Newtonian. The orientation of the reference plane is specified by the first subscript. acting in the x direction. and exerted by the fluid at greater y. 12(a). which specify a reference plane. where ?p is the heat capacity per unit mass. In that case I S seen to be the force per unit area on a plane perpendicular so that. is obvious. in that each of the fluxes is proportional to the gradient of some quantity. AND MOMENTUM The basic constitutive relations given in Section 1. A stress may be interpreted as a flux of momentum. s(n) can be resolved into components which are normal (s.) to the surface. SPECIES.2ll). in which case positive values of a . For constant p. Because we in terms of the force exerted on a plane of constant y by fluid at greater have defined uyx y. and it depends only on y. This contribution to the stress tensor is denoted as I. Note that the definition of uv contains three parts. the onedimensional form of Fourier's Iaw.ninstead of n 6 u. (1.2ll). microemulsions and foams) are nonNewtonian. we return to this subject briefly in Chapter 5. polymer melts).24) becomes cpand .. and Fick's law. It should be mentioned that some authors assign the opposite meanings to the subscripts in uu. The analogy becomes more precise if one considers the viscous shear stress in a unidirectiomlflow. and the direction of the force is indicated by the second subscript. This is illustrated most simply by considering a surface which is normal to the y axis of a rectangular coordinate system. correspond to momentum transfer in the + y direction. (1.. For that special case. In rhe lattn case the scalar components of s(n) equd components of the stress tensor u. evaluated locally in a fluid. As shown in Fig. respehively. the changes in the enthalpy per unit mass (fi) are given by mgum 12.(1. Eq. a viscosity can still be defined for a unidirectional flow. blood). The sign convention adopted here is the one used most commonly in the fluid mechanics literahue. Although nonNewtonian fluids do not satisfy Eq. gives the information needed to compute the stress vector for a surface with any orientation. (1. their flow. (1. it is related to velocity gradients by The stress tensor. by As shown in Chapter 5.sign convention for each of the components of this vector. In this context.212) and Eq.211) reduces to . The experimental and modeling studies which seek to establish constitutive equations for complex fluids comprise the field of rheology. 5. the part of the stress which is caused solely by fluid motion is termed the viscous stress.g. Fourier's law.24). uYx to the y axis. (1. 1. and a sign convention. In systems where thermal effects are more important than mechanical forms of energy. for example. The opposite sign convention ('exerted by the fluid at lesser y") is sometimes chosen.212) and the analogy would have been exact.. the stress vector is reIated to the stress tensor. The analogies among the constitutive equations are strengthened by expressing each as a flux that is proportional to the gradient of a concentration. where v=vx(y)ex. The correspondence between Eq. Tensile and compressive forces correspond to s . positive values of uyx correspond to transfer of x momentum in the . (1.3 DIFFUSIVITIES FOR ENERGY. a useful measure of energy is the enthalpy. 1960) emphasizes the analogy between momentum transport in simple flows and the transport of heat or chemical species.g. which may be represented in component form as where p is the viscosi@. There is a qualitative analogy among Eq.2 are similar in that each relates a flux to the gradient of a particular quantity. "concentration" is the amount per unit volume of any quantity. then a minus sign would have appeared in Eq. Eq. and also are influenced by. Various highmolecularweight liquids (e. and the tangential components are shear stresses. Considering variations in temperature only (i. as already noted. for a surface of arbitrary orientation. The proportionality constant in each case is a type of diffusivity. as shown in Fig. certain suspensions (e.y direction. That is. As described in more detail in Chapter 5. and (b) a normal to the y nmis. everyday examples are air and water. so that s = o . respectively. Representation of the sss vector s(n) for (a) a surface with arbitrary orientaton.g. c 0. For a Newtonian Jluid with constant density.) and tangent (sl. a direction. not only chemical species. Such fluids are characterized by structural features which influence. the only nonvanishing component of the velocity is in the x direction. If we had chosen the opposite sign convention for the stress.
.
.
.
.
.
.
.
.
.
.
.
Chapter 2 2. = O and both V and S are constant. conservation equations are based more on first principles and have certain universal forms. the fluid velocity relative to the surjbce is vv. A unit vector n defined at every point on the surface is normal to the surface and directed outward. In this chapter there is extensive use of vectortensor notation. *re 21. Differential elements of volume and surface area are denoted by dV and dS. as discussed in Chapter 1. its amount per unit volume).2 GENERAL FORMS OF CONSERVATION EQUATIONS The equations describing the conservation of scalar quantities are derived in this section using the concept of a contml volume. respectively.n. 21. is there no fluid flow across the surface. . Control volume. as discussed in Chapter 11. t ) denote the rate of formation of the quantity per unit volume. Heat transfer in pure fluids and mass transfer in isothermal mixtures are treated in parallel throughout most of this book because the differential equations and boundary conditions describing them are largely analogous. Whereas constitutive equations are largely empirical and materialspecific. One is the constitutive equation used to relate the diffusive flux of a quantity to the local material properties. t) dK The rate at which the quantity enters the control volume across a differential element of surface is . which relates the rate of accumulation of the quantity to the rates at which it enters. or is formed within. where r is time. The fluid velocity and the velocity of the control surface are denoted as v(r. respectively..v. In this chapter we begin by deriving a number of conservation equations which apply to any scalar quantity. To allow for chemical reactions and certain energy inputs from external sources. selected for its usefulness in formulating a desired balance equation. The rate at which the quantity is formed within a volume element is Bv m! Equating the overall rate of accumulation to the rates of entry and formation gives 2 . The size and shape of a control volume may vary with time. both the species conservation equation and the interpretation of the difisivity are reconsidered from a molecular viewpoint. For a given point on the control surface. let B . Nonetheless. 1 INTRODUCTION There are two major parts to any transport model.e. t) and vs (r. Examples include heat transfer in reactive mixtures and multicomponent mass transfer in gases. A control volume is any closed region in space. the minus sign is needed because n points outward. there are important situa'ions where transport processes are coupled in such a way that analogies are of little use. a spwified region. and let F(r..(Fmn) dS..) . isothermal mixture. energy. and the reader may find it helpful to review the entire Appendix before proceeding. The amount of the quantity contained in a differential volume element centered about the point r is b(r. The general results are then used to obtain the equations governing heat transfer in a pure fluid and mass transfer in a dilute.Only if v = v. Simple applications of those equations are illustrated using several examples. CONSERVATION EQUATIONS AND THE FUNDAMENTALS O F HEAT AND MASS TRANSFER Conservation Equations for Finite Volumes As shown in Fig.) velocities all vary with position on the control surface. t). We begin by considering a fued contml volume. Anally. including the amounts of mass. a control volume is assumed to have a volume V(t) and surface area S(t). and individual chemical species. The outward normal (n) and the fluid (v) and surface (v. Let b(r. t ) denote the concentration of some quantity (i. t ) be the total flux of that quantity (ie. and the component of the relative velocity that is perpendicular to the control surface and directed outward is (v . (r. and the control volume boundaries need not correspond to physical interfaces. the flux relative to some fixed reference point or origin). The other is the conservation equation. It is a region in which the rate of accumulation of some quantity is equated with the net rate at which that quantity enters by crossing the boundaries and/or is formed by internal sources. where r is the position veztor. in which v.
rcr
0
z
( I )
Q)
zi
4 3
Y
0
a
.d
0 Y
g
'3
3g
E
=F
a2
8
C
2 $
Ca2 '6 a g2 &$ cd
82
vlS
f!:
2
2
89
; 3
z
3
G
2: 3 8
4 !.
fi
.E
s
Conservation o f Mass
TABLE 21 General Conservation Equations for Interior Points and Interfaces
Combining Eqs. (2.215) and (2.216), the balance at an interface is
r
Interior points
Points at interfaces
(A)
ab=
Equations (2.217) and (2.218) are simply alternative ways to write the general balances, and contain no new physical information or assumptions,
at
v.F+B,
[ ( F  &)B

(F

bvl)A
1
'
= B~
(B)
Flux Continuity and Symmetry Conditions within a Given Phase
In formulating transport models it is often helpful to have a mathematical boundary pass through the interior of a given phase. For this reason, it is important to establish certain properties of the fluxes at interior points. Tbese are derived by again considering a control volume of vanishingly small size. Dividing each term of Eq. (2.24) by S and rearranging, we obtain nates, or independent of both angles 0 and 4 in spherical coordinates (see Fig. A3). Such problems are termed misymmetric or spherically symmetric, respectively. In these cases it is found that the radial component of the flux must vanish at the origin, or that
F,= 0 at r = 0
For a control volume of a given shape, V = cvL3 and S = C, L ~where , L is the characteristic linear dimension and c , and c, are constants which depend on the shape. Accordingly, for any control volume, V/S+O as L0. The bracketed term in Eq. (2.219) remains finite in this limit, indicating that
(axisymmehic or spherically symmetric).
(2.222)
This result is derived for the spherical case by considering a small, spherical control volume centered at r = 0. The outwardnormal component of the flux at the control surface is F,.= e, .F. If F, = F,(r, t) only, then Eq. (2.220) leads directly to Eq. (2.222). For the cylindrical case with Fr=Fr(q z , t) we choose a cylindrical control volume centered on the line r= 0 and recognize that the contributions of the ends become negligible as r+O. Thus, the integrand in the dominant surface integral is F, , and Eq. (2.222) is obtained once again. Equations (2.221) and (2.222) apply also to the diffusive fluxes, f , or f,.
The analogous result for the stress in a fluid is given in Section 5.3. One impmant implication of Eq. (2.220) is that F is a continuqus function of position at all interior points. This is seen by applying this relation to the control volume in Fig. 23, but with the interface replaced by an imaginary, fixed surface. For t + 0 , the edge contributions will vanish, leading to the conclusion that the normal component of F evaluated at one side of the imaginary surface equals the normal component evaluated at the other side. In other words, the normal component of I? must be continuous. [The same conclusion is reached by setting Bs=O and v,= 0 in Eq. (2.2IS).] Because the imaginary surface can have any orientation in space, all components of F must be continuous at interior points, and the flux vector must be a continuous function of position. Moreover, because b and v are continuous functions of position within a given phase, it follows that the diffusive flux f is also continuous at interior points. As shown in Chapter 5, analogous arguments lead to the conclusion that the stress in a fluid is a continuous function of position. The usual reason for placing a mathematical boundary in the interior of a phase is '0 exploit some form of symmetry which is present. If there is a plane of symmetry with unit normal n, then the flux component F, = n . F must change sign at the symmetry plane. The only way F,, can change sign yet still be continuous is if it vanishes, so that
Summary
The most important results of this section are the balances. derived for interior points or
points on interfaces, using either the total flux or the diffusive flux. These are summarized in Table 2 1.
2 . 3 CONSERVATlON OF MASS
Continuity
me results of Section 2.2 are applied first to total mass. The concentration variable in this case is the total mass density, or b = p . By definition, there is no net mass flow relative to the massaverage velocity, and thus there is no diffusive flux for total mass (i.e., f=O). In addition, there are no sources or sinks (Bv=O). Thus, Eq. (C) of Table 21 becomes
F, = 0 (symmetry plane). (2.221) Rotational symmetly is best described using polar coordinates. Of particular interest are situations where the field variables are independent of the angle 6 ' in cylindrical coordi
This statement of local conservation of mass is called the continuiry equation. It is used in this book to derive a number of other general relationships.
ermal Effects
39
If the fluid density is constant, the continuity equation simplifies to
Alternative Conservation Equations
(2.32)
.2
Va v= 0
(constant p).
The density of any pure fluid is related to the pressure and temperature by an equation of state of the form p=p(e T). Liquids are virtually incompressible, so that the effects of pressure on p can be ignored. Although gases are not incompressible in the them~odynamic sense, the effect of pressure on density in gas flows is often negligible. As discussed in Schlicting (1968, pp. 101 I), gases tend to behave as if they are incompressible if the velocity is much smaller than the speed of sound. In that the Mach number (Ma) is the ratio of the flow velocity to the speed of sound, this criterion for applying Eq. (2.32) may be stated as M a e l . (For dry air at O°C, the speed of sound is 331 d s . ) For both liquids and gases, spatial variations in density due to temperature may be quite important, in that they cause buoyancydriven flows (free convection). However, as discussed in Chapter 12, even in such flows Eq. (2.32) is usually an excellent approximation. For these reasons, this form of the continuity equation is employed when solving all problems involving fluid mechanics or convective heat and rnass transfer in this book.
6'
One useful equation involves the amount of any quantity per unit rnass, denoted as fi=blp. Changing variables from b to b, the lefthand side of Eq. ( C ) of Table 21 becomes
According to Eq. (2.3I), the first bracketed term in Eq. (2.36) is exactly zero. Thus, Eq. (C) of Table 21 is equivalent to
This is employed, for example, in Chapter 11 to obtain a conservation equation for entropy.
An especially useful form of conservation equation applies to fluids at constant density. In this case, it follows from Eq. (2.32) that V .(b v) = v V b b(V . v) = v  Vb. Using this to introduce the material derivative, Eq. (C) of Table 21 becomes
+
Material Derivative
Using conservation of mass, we can write the general conservation equations for scalar quantities in other forms. These are expressed most easily in terms of the differential operator, Db   ~ . f + B, Dt (constant p).
This result is employed in Sections 2.4 and 2.6 to obtain the most common forms of the energy and species conservation equations. If the diffusive flux is of the form f = 9Vb (Chapter 1) and if the diffusivity 9is constant, then Eq. (2.38) becomes
Db = 9 V 2 b + Bv
Dt (constant p and 9 ) .
which is called the mterial derivative (or substantial derivative). The material derivative has a specific physical interpretation: It is ihe rate of change as seen by an observer moving with the fluid. This interpretation can be understood by considering the rate of change of a scalar b(r,t) perceived by an observer moving at any velocity u, written as (dbldt),. At any fixed position r, the time dependence of b will cause a variation equal to (db/at)At over a small time interval At. An additional change is caused by movement of the observer, provided that there are spatial variations in b. In particular, a small displacement described by a vector Ar, tangent to 4 causes a change in b equal to (dr).(Vb) (see Section A.6). Given that o = Ar/At for small time intervals, the resulting change in b over the time A t is (u  Vb)A t. Adding the two contributions to the change in b and letting At10, the rate of change seen by the moving observer is
This is the conservation equation for a "constantproperty fluid."
2.4 CONSERVATION O F ENERGY: THERMAL EFFECTS
In many problems which require an energy equation, thermal effects are more important than mechanical ones. This is usually true when there are large temperature variations,
or when heats of reaction or latent heats are involved. An approximate equation for conservation of energy that is suitable for such applications is obtained from Eq. (2.38) by setting b=pC,,l: f =q, and B,= H v . Thus, assuming that p and tp are constant, we have
DT pcPE = V . q + Hv
A
(constant p and cp).
For an observer moving with the fluid, u = V and
as indicated above.
The source term Hv represents the rate of energy input from external power sources, per unit volume. The most common example is resistive heating due to the passage of an electric current. The heat flux in a solid or pure fluid is evaluated using Fourier's law,
q = kVT
(2.42)
Heat Transfer at Interfaces
41
Assuming that k is constant, Eqs. (2.41) and (2.42) are combined to give
2.5 HEAT TRANSFER AT INTERFACES
In this section we derive the conditions satisfied by the temperature and the heat flux at boundaries between pure materials (i.e., when mass transfer is absent). Either material may be a fluid or a solid, but the two phases are understood to be mutually insoluble or immiscible. These interfacial conditions involving temperatures and temperature gradients form the basis for the boundary conditions used most commonly with Eq. (2.43). The interfacial conditions for simultaneous heat and mass transfer are discussed in Chapter 11.
This energy equation is the one used most frequently in this book. Notice that it is of the same form as Eq. (2.39). Equation (2.43) is given in rectangular, cylindrical, and spherical coordinates in Table 22. W o kinds of mechanical effects are ignored in Eqs. (2.41) and (2.43), both of which would appear as additional source terms. One is viscous dissipation, which is the conversion of kinetic energy to heat due to the internal friction in a fluid. This is usually negligible, with the exception of certain highspeed flows (with large velocity gradients) or flows of extremely viscous fluids. The other term not included, which involves pressure variations, is related to the work done in compressing a fluid. It is identically zero for fluids which are thermodynamically incompressible; specifically, it vanishes if dplaT at constant pressure is zero. Accordingly, this term is ordinarily negligible for liquids. For gases, it will be negligible if AP<C&,,AT, where A P and AT are the magnitudes of the pressure and temperature variations. For air at ambient conditions, a temperature change of 8 K is energetically equivalent to a pressure change of 1 x lo4 Pa (i.e., 0.1 atm). Thus, this effect is negligible also for many applications involving gases. In Chapter 9 there is a rigorous derivation of conservation of energy for a pure fluid, including the mechanical terms just discussed. A general treatment of energy conservation in a mixture is given in Chapter 11.
TABLE 22 Approximate Forms of the Energy Conservation Equation in Rectangular, Cylindrical, and Spherical Coordinates (Thermal Effects Only)'
Rectangular: T = T(x, y, z, t )
Interfacial Energy Balance
Consider a point r, on the boundary between phases 1 and 2. As shown in Fig. 24, the orientation of the interface is described by the unit normal n directed from phase 1 to phase 2. The phases might be a solid and a liquid (as indicated'in the figure), two solids, or two fluids. If there is no bulk flow across the interface, then the normal component of the velocity in either phase equals the normal component of the interfacial velocity. That is, in both materials vn= v,,,, where v, = n  v and vln= n  v,. Consequently, with f = q and Bs = Hs, Eq. (D) of Table 21 becomes
where the subscripts 1 and 2 indicate the side of the interface on which the normal heat dux q,, is evaluated. The source term Hs represents the rate of energy input from an external power source, per unit surface area. Such energy inputs are rare in chemical processes, although examples are given in Problems 213 and 215. For the usual situation with Hs=O, Eq. (2.51) becomes
The arguments "(r,t)" are included in Eqs. (2.51) and (2.52) to emphasize that these
relations are valid instantaneously at each point on the interface. The same is true for all of the interfacial conditions to be presented, although the arguments are omitted hereafter to simplify the notation. Thus, Eq. (2.52) is written more simply as q,l, =
Cylindrical: T = T(I; 8, z, t )
4 . 1 2 .
n
Phase 2 (Liquid)
Spherical: T = T(r; 8 , & t)
Phase 1 (Solid)
*It is .?Sum4 that fi
a s 1 v l is ~
a=t~(~C More ~ ) general . energy equations are given in Section 9.6 for pure fluids and in
c,
and k are consunt and that certain mechanical effects are negligible.
me
9 A1
Figure 24. Normal components of velocities and heat fluxes at a solidliquid interface.
*lion
11.2 for mixtures.
a pe a,, 393
 9 8 G S
C)
3 3 3
';a
O
0 k Z g.2 a2 s k c 38
m ,
"4
0 w
5J
.
For gas mixtures at uniform temperature and pressure.39). For gasliquid systems.+ve a at ar r a0 c i + . the requirement for constant density is most easily realized in liquid solutions.~ rsinOd4 . Eq. once the reaction rate is known. .75) Jinl 2 ' kci(ci12 .5. (2. z. which is again like Eq. and composition of the two phases. Heterogeneous Assume now that phase 2 is a fluid and that there is no bulk flow across the interface. large deviations of Ki from unity (e. cylindrical.Mass Transfer at Interfaces TABLE 23 Species Conservation Equations for a Binary or Pseudobinary Mixture in Rectangular. such that RS=Rsi I&. and liquidsolid systems span many orders of magnitude. Interfacial Species Balance Once again consider the boundary between phase I and phase 2. = v.65) may be applied to both components. = 3 for hydrogen. and for a given solute it is expected to depend on the temperature.&).e. in either phase). The mass transfer coefficient (kc. Let t ibe the stoichiometric coefficient for substance i in the reaction.[~ )+ i. Assuming that there is no bulk flow across the interface (i.. For example. (2. In this book we will ordinarily assume that Kiis independent of the solute concentration. = 0 and (2. (2. where D.cib) . The partition coefficient is a thermodynamic quantity. with the unit vector n normal to the interface and directed from 1 to 2. The fluxes of the reactants and products of a heterogeneous reaction are related by the reaction stoichiometry. Species Equilibrium at Interface As with heat transfer. Eq. gassolid. nonzero values of RS are encountered frequently in chemical engineering and are of great practical impoflance. This equation. The advantage of defining stoichiometric coefficients in this manner is that a speciesindependent reaction rate can be employed. a~ Spherical: Ci = Ci(c 0. and Spherical Coordinatesa v R.5l). (2. constant total molar concentration (C) is a more accurate assumption than is constant mass density (p). If phase 1 is an impermeable solid.. with a reaction written as t) Cylindrical: Ci= Ci(c 0. I a at dr r de az r ar $+?I . d ~ . for solutes in water versus nonpolar organic solvents) form the basis for separations by solvent extraction. and 5.g. +ac. Ki is derived from the Henry's law constant. = 1 for cyclohexane (&HI. emphasizing the importance of including this factor in Eq. y. However. Kim For example. However. Instead. and spherical coordinates in Table 23.72) to be rewritten as q d2ci ]+~vi b S+. (2. a2ci + R ~ . (2. V . the fluxes of all reactants and products can be calculated.I a ( rac. A kinetic expression relating Rs to the concentrations at the surface is required to complete the formulation of a problem of this type.Accordingly. D.[: .. with ticO for reactants and ci>O for products. I OIt i s assumed that p and Di are constant.74). liquidliquid.+ . pressure. I . Thus. reactions often occur at solid surfaces which are catalytically active but impermeable to reactants and products. 4 r) 6. Unlike the conesponding energy source term in Eq. = 1 for benzene (C. 2. This allows Eq.d(r2$)+o r at 1 z(sin a 0 G)+=~ ac. 5. _ + v. z. then J. is the binary or pseudobinary diffusivity.it is determined by the relative solubilities of i in the two liquids. .72) JinL = Rsi for all reactants and products. it is usually assumed that the resistance of the interface itself to mass transfer is negligible.dC. is given for rectangular. ' Rectangular: Ci = C. Values of Ki for various gasliquid.D . Cylindrical.(x. whether or not one of them is present in abundance. which is true for ideal solutions.7 MASS TRANSFER AT INTERFACES The discussion here of interfacial conditions for mass transfer parallels that presented for heat transfer in Section 2. In a binary system. the concentrations of species i at the interface are related by an equilibrium partition coef/icient.74) Cil = Ki Ci12 .). in the analysis of diffusion across a phase boundary it is important to bear in mind that local equilibrium does not imply equal interfacial concentrations. For liquidliquid s y s t e m s . (D) of Table 21 becomes Convection Boundary Condition where Rs is the molar rate of formation of species i per unit surface area. ~+ + vAac. t ) aci. the normal component of the flux of species i relative to the interface is given by Jin.)for species i is defined as (2.
H v . Jim rather than Nin).515) are Ji. the temperature at the surface of the wire exceeds the ambient value by the amount HvRRh. It follows that Eq. and they are "onedimensional" in the sense that the temperature or concentration depends on a single coordinate. (2. where T. the temperature is found to be Thus. Because the mass transfer coefficient for a given species depends in part on its diffusivity. It is apparent now that all of the physical conditions can be satisfied by a temperature field which depends only on r Assuming now that T = T(r) only.88) is Ewritten now as .e.CONSERVATION EQUATIONS OneDimensional Examples where Cib is the bulk concentration of solute i in phase 2. Tbus. The analogy between heat and mass transfer coefficients is made evident by comparing Eq. = 0 Jir= 0 at r = 0 The symmetry condition gration yields m. and nothing is done to cause the temperature at the ends to differ. this analogy is best preserved if the mass transfer coefficient continues to refer only to the dlfisive part of the solute flux (i.8 ONEDIMENSIONAL EXAMPLES The examples in this section illustrate the application of conservation equations and interfacial conditions for heat or mass transfer in relatively simple situations. the value of kc. (2. is assumed to be independent of position. The problems all involve steady states.g.515) is applicable and that the second boundary condition in r is If the wire is very long. The behavior of the temperature is revealed more clearly by using dimensionless quantities defined as Cylindrical coordinates (r: 8. The local heating rate.75) imply that the fluxes being represented vanish when temperature or concentration gradients are absent. and the temperature at the center of the wire is elevated further by an amount ~ . 8 ranges from unity at the center of the wire to zero in the bulk air. Table 22 is used to rewrite Eq. the fluid to which kci applies. we assume that h is independent of position. these transport mechanisms require gradients in the corresponding field variable (temperature or concentration). Note that the driving force is based on concentrations only within phase 2. must be zero. a difference in mole fraction rather than molar concentration). Substituting this result into the convective boundary condition at the surface [Eq. For steady conduction in a solid with such a heat source.82) and (2. The temperature is determined by first integrating Eq.). and h all assumed to be constant. This is equivalent to assuming a uniform current density and elecbical resistance.77) where C2 is another constant.75) with Eq. When mass transfer is accompanied by bulk flow normal to the interface. Mass transfer coefficients are defined sometimes using other driving forces (e.81) as This secondorder equation requires two boundary conditions in q which are given already by Eqs.. For simplicity.85)) indicates that the constant C..81 Heat lkansfer in a Wire Consider the steadystate temperature in a cylindrical wire of radius R that is heated by passage of an electric current and cooled by convective heat hansfer to the surrounding air. (2. and not energy transfer by bulk flow.82)] gives 2.76) (axisyrnmetric or spherically symmetric). (2... Equation (2. The convection boundary condition at the surface of the wire is written as where T. (2. Restricting heat and mass transfer coefficients to the representation of conduction and diffusion. then there is also no reason for the temperature to depend on z. (2. is the ambient temperature. ~ ~ / 4 k . = T(0) is the temperature at the center of the wire and Bi is the Biot number By definition. (2. is logical for another reason. will'tend to be different for each component of a mixture. z) are the natural choice for this problem. This is because heat transfer coefficients have been defined to represent only the conduction heat flux (q. we conclude that the temperature field is axisymmetric. respectively.55) and (2.83). Equations (2.514) and (2. (2. and the Pklet number (Pe).55). (2. the Darnkohler number (Da). (2. With Hv. then is nothing to cause the temperature to depend on the angle 0. whereas convection does not. k. Namely.84) to give Symmetry Conditions The symmetry conditions for mass transfer which are analogous to Eqs. Three important dimensionless groups are introduced: the Biot number (Bi). Eq. (2. A second inte (symmetry plane).43) reduces to Solving for C. (2. The field variables are therefore governed by ordinary differential equations. Exmple 2.
we first see what can be learned from the continuity equation. and the temperature at the wire surface is very close to the ambient value. so that the total m Y=o CA= CAO L A. we conclude that the massaverage velocity is zero throughout the gas film. can be obtained from NAY and because CB was assumed to have no effect on the reaction kinetics. The reaction rate follows nthorder kinetics (n>O).. but not when using the massaverage velocity! Rearranging Eq.810). 8 1 There is no reaction term in Eq. This indicates that there is a convective flux here when using the molaraverage velocity. unless rn = l). (D) of Table 13. the Biot number represents the ratio of the heat transfer resistance within the wire to that within the surrounding air. 6 r/R 0 . we obtain Using Eq. is independent of y. Taking advantage now of the assumed constancy of C. so that evaluating the flux ratio at any location determines the ratio for all y.6 cannot be used here because p is not constant. However. Diffusion in a binary gas with a heterogeneous reaction. the total mass density ( p ) will not be constant under these conditions. (2.as a function of the Biot number.31) becomes Thus. ( ~ ) = NAy(lm)/C. Thus. which requires that we use the molaraverage velocity as the reference frame. From Tables 12 and 13.e. (2. No further consideration of species B is necessary.. Eq.2 tion (C) is constant.721. Thus. we can apply Eq. Figure 25 shows the dimensionless temperature profile for several values of Bi. we adopt Eq. because N . From @. . This equation indicates that both fluxes are independent of position. from this expression. A tmB.61) to both species. and the m drop is in the air. (2. It follows from the stated assumptions that 0 0 0.One~imensional Examples $ . A stagnant gas film of thickness L is in contact with a surface which catalyzes the irreversible reaction.. Before using species conservation or Fick's law. Eq. convective heat transfer in the air is so rapid that the external temperature drop is negligible. Because l e catalytic surface is assumed to be impermeable (i. as given by The convective flux of A in this reference frame is defined as CAvJM). In selecting a form of Fick's law it is advantageous to employ an expression which involves C rather than p . Most of the results in Section 2.(2. The system to be considered is shown in Fig. pv. The man consequence of this is that J . the molaraverage velocity does not vanish unless m = 1. is a constant. For this steady. (2. onedimensional system with variable p. (2. = Niy for both species. 26. and it also shows how the reaction rate can influence the boundary condition used at a catalytic surface.4 0 . Unless the molecular weights of A and B are identical (i.=Niy.814) to eliminate N Example 2. It is assumed also that the gas is isothermal and isobaric. (2. Temperature profile in an electrically heated w i r e . (2. For Bi>> 1.. . For Bi<c 1.816) implies that v . (2. the total flux of A is given by .2 0.SO that Eq. B Figure 26. v y = O at y = L). It is assumed that CAdepends on y only and that its value at y = 0 is fixed o l a r concentraat C. Using Eq. radial a i n temperature heat conduction in the wire is so fast that the wire is nearly isothermal.. because it is C which is assumed to be constant. .82 Diffusion in r Binary Gas with a Heterogeneous Reaction This example illustrates the use of Fick's law for a binary gas. .817) becomes Gas Catalytic Surface Y= L The flux of A at the catalytic surface is directly related to the reaction rate at the surface.813) because there is no homogeneous reaction. even though the massaverage velocity is zero for all stoichiomeuies. The significance of Biot numbers for heat or mass transfer is discussed further in Chapter 3.73) together with Ji.e. The plot is based on Eq.816) to solve for NAY gives where k . we obtain Figure 25. Thus. (2.
Converting to the dimensionless variables defined by Eq. or (in dimensionless form) For firstorder kinetics (n = I ) and Da> lo2. we introduce the dimensionless quantities The parameter Da is the Damkohler number. diffusioncontrolled. C. The governing equation in dimensionless form is then where xAois the (known) mole fraction of A at q = 0 . n =2) at several values of Da are shown ill Fig. the concentration at the surface approaches zero.823) reveals that 4 4 1 as Da+O. Concentration profiles for diffusion through a gas f i l m w i t h a heterogeneous reaction. (2. (2. in which case Eq. The accuracy of Eq. itis a measure of the intrinsic rate of reaction relative to that of diffusion. which is seen to be the ratio of a reaction velocity (k.819) by CA= 0 at y = L. Another key feature of the dilute liquid solution is that it is pseudobinary from a diffusional standpoint. secondorder reaction (m = 1. as Da00. the product B will have negligible influence on the diffusion of the reactant A for any stoichiometty. The previous conclusion that v. (2. Thus. the flux of A is given by This type of fastreaction boundary condition holds for any irreversible. (2.818) and (2. those for internal and external heat transfer (Bi) or those for diffusion and reaction (Da). (2. Because species B is assumed not to affect the reaction kinetics either.822) with m = 1 and n = 2 .=O remains valid for the liquid. the process is controlled entirely by mass transfer and 40. it need not be considered at all. The one which compares raks of reaction and diffusion is sometimes called 'LDamk6Mer group II. is denoted as &. Because the reactant flux (and therefore the reaction rate) varies as 1 . Example 2. Also noteworthy is the qualitative similarity between this plot and Fig. (2. (2. An extensive tabulation of named dimensionless groups pertinent lo chemical engineering is given in Catchpole and Fulford ( 1 966). which depend on the reaction stoichiometry.' The reactant concentration at the catalytic surface. heterogeneous reaction. (2. the reaction rate law does not enter into the problem. (2.820) to determine the concentration profile.@(I).64)." Baause it is the only type of Damkohler number used in rhis book. this implies that 8(1)<0. so that the reaction is the controlling step and the reactant concentration is nearly uniform throughout the film. the conservation equation for species A is 'Them are no fewer than five dimensionless groups named after DarnlrGhler. Thus.Because NAY has been shown to be independent of position. In this case the reaction is slow relative to diffusion. The foregoing results indicate that if we were interested only in diffusioncontrolled conditions (Da +). Da> lo4 is needed to maintain this small level of error. is linear in y for all values of m.82. Inspection of Eq. One consequence of having a liquid is that we can assume constant p .823) indicates that This may be contrasted with Eqs.816) or (2. it is found that the concentration of A at the catalytic surface is governed by . (2. 25. Accordingly..the simplified boundary condition will lead to an error of <I% under these conditions.CAon') to a diffusion velocity (D. From Table 23. In both situations the parameter can be interpreted as the ratio of two resistances in series. Eqs. That is. The transition from kinetic to diffusion control as Da is increased is evident. there is no cmnvection now. When this simple condition applies.83 Diffusion in a Diiute Liquid Solution with a Heterogeneous Reaction We reconsider the situation of Example 2.824) for the present problem is judged most easily for an ~uimolar reaction (m = I).821) and proceeding much as before. Concentration profiles for an equimolar.818) for the gasphase problem. The results. two of which involve reaction m b . but with a dilute liquid solution in place of the gas. At the other extreme. 27.01. Equation (2. Using this information and Eq. The curves were obtained by solving Eq. are Elgure 27.lL).822) is separable and can be integrated from q = O to q = 1 to obtain implicit expressions for $= 0(1). the surface concentration of the reactant. showing the effect of the Darnkohler number.819) can be equated to give Before integrating Eq. which is an unknown constant. then we could replace Eq. no other identifier is added to t h e symbol Da. For secondorder kinetics (n = 2).
Suppose that a00 and P+. the boundary conditions are This problem is characterized by two Damkohler numbers. which corresponds to increasing the reaction rate constants to very high values without altering the equilibrium constant. c'. as shown in Fig. the steadystate conservation equations (from Table 23) are With this choice of dimensionless concentrations.whereas p involves the reverse rate constant and DB.837) From the boundary conditions at y = 0. The result is OA=[q] ar+P + [&(Iy) a+p][~~~h(~~p~~h(~~)sinh(~~n) (2. It is informative to examine the concentration profiles for a very fast reaction.The third parameter determines the direction of the reaction in the liquid film: The net reaction is A + B for y< 1 and B+A for y> 1. Thus. Consider the reversible conversion of A to B in a stagnant liquid film at steady state. a and p. = 0.835) once yields Y=O . The concentrations at y = O are known (C.830).For the liquid. t where al is a constant. These results are valid for any values of the parameters.823)]. Inserting Eq.. we have a. =KCB. (2. For solutes A and B in a dilute liquid.sinh xe". From the boundary conditions at y = L. (2. and converting to dimensionless variables. Integrating again. with K the equilibrium constant defined by Eq. (2.831) and (2. tanh x+ 1 and cash x. as will be seen. (2. corresponding to equimolar counterdiffusion [see Eq.with pla fixed. (2.844) reduce to Liquid Inert Surface . 8.832).831) and (2.. Example 2.For the gas. Combining Eqs. Diffusion in a stagnant liquid film with a reversible homogeneous reaction. and the inert solid surface at y = L acts only as an impermeable barrier. 28.840) The dimensionless parameters which appear are With the specified concentrations at y = O and an impermeable and inert surface at y = L . The reaction kinetics are first order and reversible. a differential equation which is independent of the reaction rates is obtained by adding Eqs. However. + (2. em).843) and (2. (2.. DA CA DBCB = %. From the kinetics and stoichiometry.831) and (2.84 Difbsion in a Liquid with a Reversible Homogeneous Reaction This example illustrates the influence of homogeneous reactions on concentration profiles and also demonstrates a technique for dealing with reversible reactions.831) will yield a single differential equation involving only C. it is valid for all stoichiometries. so that Eqs.. cr involves the forward rate constant and DA. we have a.838) as mentioned above. = BB at equilibrium. (2. = RvB=O and C .840) and (2. (2. The solution to this set of equations is complicated by the fact that Eqs.LA oB CA= CAOCB= CBO Figure 28. (2. (2. (2.838) in Eq.841) for the concentration of species A is This completes the formulation of the problem. The solution is completed using dimensionless variables defined as At equilibrium.832) a r e coupled through the reaction terms. R . = DACAo + DBCm. Whether or not reaction equilibrium is actually achieved in any part of the liquid f i l m under steadystate conditions is determined by the relative rates of reaction and diffusion. gives d 2 e~ (a d+ + P I% . For large arguments. The solution to Eqs.843) l The corresponding result for species B is Integrating Eq.(ay+p). this result held only for m = I.
( y ) . For the low solidification rates which are characteristic of crystal growth. The solution of Eq. the thickness of the nonequilibrium layer merely diminishes. This equilibrium is typically modeled for dilute alloys by the linear relationship where K. and eBeach change from the surface value to the equilibrium value over a dimensionless distance Aq of order of magnitude (a + P)'12.. The meltcrystal interface is located at y = 0. based on Eqs. Near the exposed surface (q= 0).l. For a = f3= 1 the reaction A + B proceeds at significant rates throughout the film. (y) and v. (2. .844) for moderate and fast reaction kinetics and y=O are shown in Fig. there may be very large concentration gradients. eventually varying as the inverse square root of the sum of the Damkohler numbers. m e dimensional thickness S of this nonequilibrium layer is therefore At the edge of the stagnant layer the concentration must match the bulk value. Brown.. For most impurities. The bulk of the melt is assumed to be wellmixed with a dopant concentration C_. In this nonequilibrium region.is the equilibrium segregation coefficient for solute i. which affect the value of pa.Figure 29. Notice that the equilibrium concentration is determined in part by the relative diffusivities. the reactants never achieve equilibrium throughout the film. 29.the conservation equation for the dopant (Table 23) becomes Thus. Plots of Eqs. it is desired to predict C . so that concentration in the solid is C. We have neglected diffusion of the dopant in the solid because of the extremely low diffusivity in the solid phase. Solidification is assumed to occur at a constant rate. The material is a dilute alloy with dopant concentration C.Si)=C. y = 4 y=0 Over most of the film the exponential terms are negligible. and the dopant properties.843) and (2. Ki<O. A.853) is This can be rewritten as . That is. = C. 210. steadystate problem with C. It is assumed that there is a "stagnant film" of melt of thickness in which there is no Bow parallel to the interface.850) At the meltsolid interface. so that c i ( . Figure 210. Given values for C. (2. this implies that the melt and solid densities are equal.850) and (2. onedimensional model shown in Fig. whereas the dopant 'This example was suggested by R. (2.843) and (2. This condition is written as Thus.1=/3= 1) and fast ( a = p = 100) reaction kinetics. the flux of dopant in the melt must match the flux in the solid. even as the reaction rate constants reach arbitrarily high levels. such that the velocities of the melt and solid relative to the interface are both equal to U. Results are shown for moderate (. For this onedimensional.85 Directional Solidification of a Dilute Binary Alloy2 The partitioning of a dopant or impurity during the solidification of a melt in a uniaxial temperature field is fundamental to most methods for growth of singlecrystal semiconductors and metals. (2. Using Eq.. transport of b e dopant in this layer is only by diffusion and convection normal to the interface.852) the interfacial balance is rewritten as Example 2. 8. (2. according to the model. = U. whereas for a = p= 100 there are distinct reaction and equilibrium zones. Concentration profiles for diffusion in a stagnant film with a reversible homogeneous reaction. such as metals in silicon. the mlt and solid can be assumed to be in chemical equilibrium at the interface. (2. Onedimension model for dopant distribution in directional solidification.849) that satisfies Eqs. In both cases it is assumed that y = 0.844). as discussed. This situation is simply described by the steadystate. Si. where A and B are added to or removed f r o m the liquid. when the reaction is fast the two species are in equilibrium throughout most of the film.
. This result is the same as if we had assumed that there was no wellmixed region in the melt and had simply set 6. is based on the bulk rather than the interfacial concentration of the dopant. this requires that there be a large number of molecules in any pointsize volume under consideration. The following derivation concerns the motion of a single molecule (or Brownian particle) relative to the surrounding fluid. Stagnant F i l m Models The foregoing analysis is an example of the use of a stagnant film model to represent convective heat or mass transfer in what is in reality a very complex situation. (2. (1953) first used this analysis to deduce an expression for the effective segregation coefficient for the dopant between the bulk melt and crystal. (2. 8. Working backward from time t + T to time t. if kc. so that the species concentrations are deterministic quantities. in general.82 and 2. t + T ) is the product of p(rR. (2. K. In any stagnant film model. and 13.855) becomes C ~ Y ) (1 K. t) and the probability of a jump described by a specific vector R.= 8 for all i). convection is excluded.= .or threedimensional velocity fields in the melt. For example. Tbe time required for a single step is T. Whereas Ki is determined onIy by thermodynamics. I +?exp (z) (2.. removal of impurities during solidification relies on vigorous mixing of the melt. Thus.9.R . Si is used as an adjustable parameter.R.1) p(r. the equation describing the probabilrty of finding a diffusing molecule at a given position and time has the same form as a conservation equation. the Pklet number for mass transfer is a measure'of the importance of convection relative to diffusion. describing a stepwise stochastic . a Di. (2.. As discussed in Section 1.855) indicates that Ci (y) = C. t). At the other extreme. 10. In reality. and thereby to illustrate the relationship between the molecular and the more macroscopic viewpoints.Ki)exp (Pei) K i ' If Eq. 9 SPECIES CONSERVATION FROM A MOLECULAR VlEWPOINT . as reviewed in Brown (1988). = I and there is no segregation of dopant between the bulk melt and the crystal. such as in most experimental studies of directional solidification. The molecule is assumed to move independently of other solute molecules. Thus.859) to estimate K.e. . Hence. Thus.t)A(R) dR. the position at time t is r . (2. Likewise. Eq. One of the limitations of the stagnant film approach is that. R is a vector with length 4 and random direction. (2. then it is found that K . However.860).g. Suppose that after being "released" from some location at t = O .. Nonetheless. in efforts to relate molecularlevel phenomena to observed rates of transport. . >> 1. where Pe. The probability density governing R is denoted by A(R). (2. diffusion in the film of melt is sufficiently fast that no significant concentration gradient is able to develop.860) implies that kc. If it is assumed that the film thickness has the same value for all solutes (i. Chemical reactions are not considered. The next step terminates at some position r at time t + T. we recognize that p(r. Eq. solidification model. so that its calculation is a stochastic rather than a deterministic problem.. may be computed using Eq. under the proper conditions. t + ~ ) = b ( r . Using Eq. if one chooses the value of 6 .where n is a . it For Pei+O in the is a ratio of a flow velocity (U) to a characteristic velocity for diffusion (Di/S. 8.856).857) is used for the dopant concentration profile. 8. which restricts the derivation to dilute solutions. (2. The position of a given molecule can be described only in terms of probabilities. is the Peclet number for solute i. the molecule t t b s a number of discrete steps or 'Ijumps" of uniform length l in random directions. depends on the solidification rate and other process variables. For the situation illustrated in Fig.. Having a large number of molecules at every such point averages out the randomness in the instantaneous positions of individual molecules. stagnant film models as used in this example and in Examples 2. a constant.where Pe. Keff is like Ki. in this reference frame. The prediction of :'.:" constant which depends on the type of flow and which is typically< 1. Equation (2. As exemplified by Eq. for that dopant at some other velocity. . The probability density for finding the molecule at position r at time t is denoted by p(r. flow patterns in systems with forced convection (e. and Di are known. Equation (2. stirred reactors) are often quite complex. The objective of this section is to derive the equation for the probability distribution. it is natural to focus on the motion of a single molecule. The film thickness is related to the mass transfer coefficient and diffusivity of i by The conservation equations for chemical species developed in this chapter are based on the concept of the local concentration of species i in a mixture. convection and diffusion parallel to the interface are ignored. 12. the effec! tive film thickness needed to give the correct value of the mass transfer coefficient is different for each component of a mixture.. buoyancydriven flow will usually result in two. and all effects of mixing in the fluid are embodied in the effective film thickness. then Eq. In many applications.) I= c. .83 find frequent use as first approximations in engineering calculations. we refer to it as the case for "no bulk mixing. In directional solidification.855) it is found that Keff = Ki + (1 . 211.6.91) is an example of a Markoff integral equation." Burton et al.859) has found widespread use in research on crystal growth from melts.). K. .defined as 2 . n for various situations is discussed in Chapters 9. to match the observed whem 6(x) is the Dimc delta function and the integration is over all space.. In general. For this reason. n u s .. it is found that for given flow conditions we have kCixD7. Accordingly. Considering all possible jumps. except that K . precluding an analytical solution to the "real" mass transfer problem. (no bulk m i x i n g ) .
The diffusivity in Eq.613). compare with Eq. We conclude this discussion by showing how the diffusivity Di can be calculated ftom information on the movement of a single molecule or particle. (2. Using spherical coordinates to integrate A(R) yields This is the desired equation governing the probability density p(r. The coefficients appearing in Eqs.54).91) can be written as Taylor expansions about position r and time t. The first two terms for the expansion in time are then the molar concentration is related to the The significance of Eq.92) ensures that the only allowed jumps are of length 4. We conclude that md. (2. the probabilities in Eq. Equation (2.65) to give That is.910) governs the probability density for each of N independent solute molecules in a dilute solution.612) and (A.93)(2. Substituting these leading terms in Eq. that average vanishes. If there are N molecules of solute i in a volume probability densities by Equation (2.95) to evaluate the integrals. it is found that The similarities between Eqs.910) and (2. and the result is expressed in terms of the identity tensor 6. The Dirac delta function has the additional property that 16(x x. the probability is unity that the molecule is located somewhere in space. Two other integrals which will be needed are stated now. (2.913) is that it implies that Eq. (2. A(R) is normalized so that the probability is unity that a given jump will end somewhere on a spherical surface of radius t'. (2.9lo).911) must then be identical. (2. (1. using Eq. the probability density fi(r. one for each solute molecule.?I.66).) = 0 for x Z x. we find that Finally. because R has no directional bias. (2.95) gives the average value of the dyad RR.. Positions of a molecule malung randomly directed jumps of length t. (A.5. (2. With t' and 7 assumed to be small relative to the length and time scales of interest.91) gives where u = e / is ~ a 'Tump velocity" Like that used in the lattice model for diffusion in Section 1. applying the identity VVp:S= V Z pand rearranging yields process. Using the index j to denote a particular solute molecule.94) gives the average value of the singlestep displacement vector R. Consider the molecule to be at the origin at t = O .910) and (2. Using Eqs. Equation (2.) f ( x ) dr =f (xo). The overall distribution of solute molecules in position and time can be described using a sum of such probabilities or. (2.each in time 7.910) is obtained by setting v = O and RE= 0 in Eq. At later times its position is described by . A deterministic result comparable to Eq.911) are not coincidental. (2. Using Eqs. p(r. a concentration. the first three terms for the expansion in position are where all derivatives are evaluated at r and t. I . if N is large. (2. (A. After a large number of jumps. t) for finding a given solute molecule at position r at time f. t) is normalized such that T h a t is. t) may be approximated as a continuous function of position and time. Equation (2.911) could have been derived by summing N equations like Eq.Figure 211.914) is the same as the lattice result. '.because S(x x.
g. (2. Theoretical. Nye.V+. Brown.r ( t )= x(t) ex+ y(t) e . E. The usual units of i are C m2 or A m2. Burton. Brenner. (2. p+ 0 as r+ co. Fulford. Chem. Chem. R. and G. In such simulations the instantaneous positions of N molecules or particles are computed by solving Newton's law of motion with an interatomic or interparticle potential field which describes their interactions. D.912)J is . electrical potential 4 are related as i = K. Oxford University Press. (2. The distribution of solute in crystals grown from the melt. to measured molecule or particle positions (e. Radiation Heat Transfelfer. ( t ). Schlicting.C. Cess. we now use Eq.is the valence of ion i. Boston. and J. T.910) becomes .919). Sparrow. t) in the solution. J. Slichter. See Allen and Tildesley (1987) for a thorough discussion of molecular dynamics methods.. Clarendon Press. s' vanishes as a result of the spherical symmetry. .652 X lo4 C mol').. Prim. Eq. If the intitial position of the jth particle is denoted as r. M. the current density i and . and that there are no chargeproducing reactions talung place in the solution. g is the gravitational acceleration vector. Oxford.. ButterworthHeinemann. Flow in Porous Media: Darcy's Law A relationship often used to model fluid flow in porous media is Damy's law.) is given by Using Eq. (2. and D. z(t) e. The Darcy permeability has units of m2 and is a constant for a given mate . Tildesley.1987. (2. A. 1978. the normalization condition for p [similar to Eq. 58(3): 4640. 1953. 1968. : . corresponding to a radial coordinate r(t) = 1 r . McGrawHill. J. J. 1991. AIChE J. McGrawHill. The mean displacement of the molecule. J. Howell..9) Thus. The meansquare displacement is given by where u. References Taking as unity the probability that the molecule is located somewhere in space. 1966. where K is the Darcy permeability. Phys.rj(0)I2). second edition. r) only. 22. p = p(r. in the absence of + + 6Djt = ( 1 r. Dimensionless p u p s . (2.920) can be written for the collection of particles as in the solution? (b) Derive the partial differential equation which governs &r. M. The charge density in an electrolyte solution (p. A.r 1 In = (2 +y z2)'n In an isotropic medium. Problems 21.916)is (see Section 4. Eng. (a) What can be learned by applying conservation of charge to an arbitrary control volume ?his result can be used to evaluate Di by measuring displacements during many identical time intervals. P . and the summation is over all ions in the solution. The calculation of the meansquare displacement as a function of time is the basis for the evaluation of diffusion coefficients using molecular dynamics simulations for liquids. R. and 9 is the dynamic PRssure (Chapter 5). 1957. as obtained from the experimental visualization of particles or molecular dynamics simulations). P. Ind. that is. Except near charged surfaces. electroneu~ality( p e z O ) is a good approximation. Oxford. (2.McGrawHill. as expected. R. Siegel.917) to evaluate certain average properties of the position. 1988. (2.= (r 2 )/6t. New York. Edwards. . Allen. and W. 34: 881911. and therefore contains no information on Di. where the angle brackets indicate averages taken over all N particles. F. Cstchpole. J. . 1981. To relate D. and D. D. These are averages which would be observed by recording the results of a large number of random walks.(O). Wasan.917) in Eq. sixth edition. Theory of Wansport processes in single crystal growth from the melt. any directional bias. New York. Assume chat this holds.we find that where F is Faraday's constant (9. is the electrical conductance of the solution (a constant). and R. '  The solution to Eq. Part I.915) that satisfies Eq. BoundaryLayer Theory. Using spherical coordinates (Table 23) and Di = e2/(6r). Charge Conservation and Electric Current For an electrolyte solution of uniform composition (VCi=O). H. The average of the squares of these displacements will yield (r2). Computer Simulation of Liquids. 21: 19871991. New York. Inte@acial Transport Processes and Rheology. H. P. the probability density is spherically symmetric. Physical Pmperties of Crystals. allowing the diffusivity to be calculated as D. Thermal Radiation Heat Transfer. R. each of duration t. A. then Eq.
@ ( I ) U .LI 8 UJ C e. & g 5 U ..
. For algebraic simplicity. Oxygen Diffusion in Tissues Oxygen is consumed in body tissues. B CFICo." As shown in Fig. u . = aPA at equilibrium. (a) For the special case of no reaction.) (c) Determine the enhancement factor E. The sieving coeficienr.. consider steadystate O2diffusion in a sphere of radius r. One would expect that for a fixed value of the equilibrium constant (K).829) and (2. as in Example 2. As shown in Fig. (2. Determine the 0. The solubility of A in the liquid is a. and the reflection cue$?cient. Pressure monitoring mangefor vessel containing fluorine gas. the cells in the center are lulled by lack of 0. In other words. (It is helpful to first add the conservation equations for A and B. or by cells maintained in vitro.~ ) P R ' F 2 . In the liquid. x = 6 % U R x=O x= L m8. Vessel 5 Discuss the behavior of 4 3 for extreme values of a and the retentate and membrane Pklet numbers. transport of solute in the filtrate is assumed to be purely convective. (b) Determine the flux of A under reactive conditions. Onedimensional model for ultrafiltration. as the tumor grows. assume that DA= DB= D. The solute concentration in the retentate f i l m CR(x)is assumed to reach the bulk value Co at a distance 6 fmm the membrane. a liquid in a highly porous membrane) separates gas compartments containing different partial pressures (PA) of species A.830).e .u)ekR o(1 . If ro and the rate of 0.) is maintained constant at C..g. Determine when an oxygenfree central core will exist and a d an expression for r. it is determined by the solute flux and filtrate velocity. because no O2 is available to react. The diffusivity in the retentate is D. at a rate which is often nearly independent of the 0. (By ''liquidequivalent" it is meant that C. concentration profile.. the flux enhancement would be greatest for very fast reaction kinetics. P211. The rate and equilibrium expressions are as in Eqs. Why is it desirable to operate with a small Pe. (2. such that C.For an ideal semipermeable membrane. . 212.830). defined as the ratio of the flux of A with reaction to that without reaction. reactant A is introduced at x = 0 and reactant + .84. and Pe..The filtrate concentration is not known in advance. Bimolecular Reaction in a Liquid Film Assume that the bimolecular. For the central core the assumption of zerothorder lunetics is no longer valid.(y) and the reaction rate. the smaller the value of 6. u=0. Assume that the O2 concentration at the outer surface ( r = r. u= 1. In the membrane the solute flux (N) is given by velocity is v where Cdx) is a "liquidequivalent"concentration and DMis the corresponding solute diffusivity in the membrane phase.. A simple but useful model for steadystate ultrafiltration. C. A is converted reversibly to B. P212.) The intrinsic selectivity of the membrane toward the solute is expressed by the diffusivity ratio D.829) and (2. This situation occurs in certain solid tumors where. and a large Pe. @= (1 . even at the phase boundaries. defined by r < r. and for a completely nonselective membrane. Pe. 211. the more effective the agitation of the retentate.N 2 Purge v merit Figure E7. Such a reaction permits the solute to exist in either of two forms. 28. is shown in Fig. Show that 2 1 1 0 . and the filtrate . irreversible reaction A B+ C takes place at steady state in a liquid film of thickness L..such that C. based on the stagnantfilm concept. P28. Determine the steadystate concentration profiles CA(y)and C. : : j. which is nonvolatile and therefore does not escape. assume that a stagnant liquid film (e. concentration. relate the flux of A to the partial pressures and other system parameters. C(r).? 29. with no net consumption of that solute. As a model for a tissue region or aggregate of cells.. . This augmentation is termed "facilitated transport. Facilitated llansport A reversible chemical reaction can be used to augment steadystate transport of a solute across a liquid film or membrane. A steady state exists. Use rate and equilibrium expressions analogous to those in Eqs. both of which can diffuse.=N/v. with zerothorder consumption of 02. measures the overall effectiveness of the separation for a particular solute. no O2 will reach a central core. consumption are sufficiently large. Examine the behavior of E in this limit.P e ~ + ( l. refers to the liquid.(x) is defined so that the concentration variable is continuous./D. except assume that the reversible reaction beWeen A and B is heterogeneous (at y = L) instead of homogeneous. and thereby increases the effective concentration of the transported species. Reversible Heterogeneous Reaction in a Liquid Consider the situation depicted in Fig.
and at the top surface of plate 2 there is convective heat transfer to the ambient air (temperature T.. Determine x. In the limit k+= the reaction occurs at a plane x =xR. (b) Determine the rate of freezing at a given instant. Bimolecular reaction in a liquid film. respectively. (a) Determine the steadystate temperature profile in the plates when both are at rest. Cdx).(t). The heat of fusion of silicon is A. 216. assume that the tube wall is thin (R.. passing a current from the solid to the melt releases heat at the interface. Determine the effect of Hson the interface position. \ The layer of solid outside the tube is not necessarily thin. expressed as dR. Mass Fluxes for a Growing Bubble A vapor bubble of radius R(t) is growing in a supersaturated liquid. at a given instant.). B. The rate of frictional heating at the contact surface between two solids may be expressed as (a) Calculate the steadystate temperature profiles in both silicon phases. which forms a boundary between the part of the f i l m that contains A (but not B) and the part that contains B (but not A).e... this is called a pseudosteady appmximadon (Chapter 3). Assume that temperature variations in the x and y directions can be neglected.= Ri) and has negligible thermal resistance. To do this. heat transfer coefficient h).c T. . The internal heat transfer coefficient is h and the heat of fusion is i.(x). and the reaction rate. while plate 1 remains stationary.. Finally.ldt.) is between the temperatures of the graphite plates (i.Gas Liauid Gas Figure P211. P216. Graphite Plate T = TI B at x = L.. 1957). For simplicity. as shown in Fig. assume that the freezing occurs slowly enough that the time derivative in the energy equation is negligible. 215. Brown. The bottom surface of plate 1 is at constant temperature To. The volumetric rate of formation of C is given by 214. .. as shown in Fig.e.) and liquid density ( p L ) are constant. and c is the coefficient of dry friction for the materials involved. A. 213.(O) = R. The thermal conductivities of the melt and solid are k.. TI<T. The plates are held in contact only by gravity.Facilitated transport across a Figure P213. and the height H of the meltsolid interface. The liquid solution is diIute everywhere with respect to solutes A.. (c) Evaluate the mass flux in the liquid at a fixed position just outside the bubble.(x). choose a control volume of constant radius which. (b) Assume now that plate 2 moves horizontally at speed U. z Liquid Figure P212. Assume that transport is spherically symmetric and that the gas density (p. Peltier Effect * As shown in Fig.. normal to the interface). st k. C. assuming that R. coordinates are chosen so that the origin is at the center of the bubble). The rate of energy release at the interface is given by where U is the relative velocity of the solids. across the vaporliquid interface).. (c) Find R. a silicon Graphite Plate T = T2 Molten Silicon Solid Silicon >. A system with meltsolid interface. is the current density in the direction (units of N r n 2 ) . where k is the secondorder. and the inner wall of the tube is cooled by convective heat transfer to a gas at bulk temperature TG. *This problem was suggested by R. (b) Evaluate the mass flux into the bubble (i.. Assume further that TG is independent of axial position. and C.e. P215. y is the force per unit area holding the solids in contact (ie. Are the concentration gradients continuous or discontinuous at x =xR? (a) Show that the vapor in the bubble is at rest. where /3 is the Peltier coefficient (units of volts) and i. (a) Determine the pseudosteady temperature profile in the solid at a given instant. C. If the reaction kinetics are sufficiently fast. A and B cannot coexist in the liquid. molten and solid silicon are maintained in contact between two graphite plates separated by a distance L. h e z i n g of a Liquid Outside a %be A pure liquid is being frozen outside a refrigerated tube. P213. Both phases occur because the melting temperature of silicon (T. Determine the rate of interfacial energy generation and its effect on the steadystate temperature profile in the plates. The center of the bubble is stationary (i. The bulk liquid is at its freezing temperature (T. a phenomenon called the Peltier eflect (Nye. homogeneous rate constant. and k. H s = pi. and k . The plate dimensions in the x and z directions are sufficiently large that T = T ( y ) only. Heat Conduction in Solids with Friction TWO solid plates of differing thermal properties are in contact. (b) When a meh is an electrical conductor and its solid is a semiconductor (as is the case for silicon). is slightly smaller than R(t).).
assume that the melting occurs slowly enough that the time derivative in the energy conservation equation is negligible. The passage of a current through the heater material yields a constant rate of energy generation.Air h T A r2 plate 2 kz P2 U Figurn P217. (a) Find the pseudosteady temperature profile T(z.). Assume that the steadystate temperature is a function of x only.(L) ICo. Calculate the heating rate that is just sufficient to eliminate any heat loss from the inside of the container. and the candle begins to melt.) required for the candle to melt completely? 218. which represents radiant energy transfer from the flame minus convective losses. Vessel Wall with a Heater To prevent heat loss fmm a fluidfilled vessel.. Melting of a Candle After a candle is lit the wax next to the flame soon reaches its melting temperahre (T. Melting of a candle. respectively. Assume that the layer of melted wax on top of the candle is thin enough that the heat flux toward the candle on the liquid side of the meltsolid interfan is approximately equal to go. For simplicity. Assume that go is constant. assuming that the temperature is approximately independent of radial position. There is also convective heat transfer from the sides of the candle to the surrounding air (ambient temperature T. The base of the candle is at the ambient temperature. P218. The outside values are To and h. (b) Evaluate dL/dt at a given instant. . 219. assume that after a certain time t a candle of radius R has a length L(t). Consider the arrangement shown in Fig. Hv. where To<Ti. Membrane Reactor The enzymecatalyzed reaction A + B is to be carried out under steadystate conditions with the enzyme immobilized in a membrane. what is the time (1. (c) Assuming that the initial length is L ( O ) = h . and to help separate any residual reactant from the product. heat transfer coefficient h). this is a pseudosteady appmximation (Chapter 3). P217. such that C.. 217. The membrane has two purposes: to prevent loss of the expensive enzyme. t) in the candle. As shown in Fig. The inside bulk temperature and heat transfer coefficient are Ti and hi. One surface of the membrane (at x = L) has a thin coating that permits rapid diffusion of A from the feed stream to the membrane. it is decided to attach an electrical heater to the outside of the wall. tact. P219. as shown in Fig. There is a net heat flux go to the top surface of the candle. ~ Freezing of a liquid outside a tube. A key Liquid Solid Gas T = TG Liquid T = TF Tube ~ a ~ i R w P216.
1 INTRODUCTION This chapter has a dual purpose: to discuss the simplification of transport models and to describe methods for obtaining approximate solutions. discussed in Section 3. For simplicity. . there are systematic ways to approach the formulation process.*.4.8 employed various assumptions to yield steadystate. and &. A qualitative understanding of what is being modeled is essential. Indeed. and scaling is the process of identifying the correct orders of magnitude of the various unknowns which one confronts in a new problem. . (c) l'ko measures of reactor performance are the purity of the product stream. homogeneous reaction rate constant in the membrane is k. The concentrations shown for h e feed stream and product sueam are assumed to apply just inside the coating and membrane. and the fractional recovery of B in the product stream. z 0 y . SCALING AND APPROXIMATION TECHNIQUES Figure P219. the Rux of B across the coating is described by where kc is the permeability coefficient of the coating for B (similar to a mass transfer coefficient). The scale of a variable is an estimate of its maximum order of magnitude. and determine the limiting values of both quantities for very fast or very slow reactions (large or small A. depend strongly on a qualitative understanding of the system being modeled. 73 (b) Determine O. ". feature of the system is that the coating restricts (but does not eliminate) the passage of B.5. Use the following dimensionless quantities: 3 . 1:. respectively. and they are included here for that reason. Discuss conditions which will maximize and/or 6. perturbation methods are useful when a problem with scaled variables contains a small parameter. With negligible B present in the feed stream. The firstorder. The examples in Section 2. they are used to greater advantage in many of the convective transport problems later in this book. continues with perturbation methods in Sections 3. with fixed y). The similarity method applies only when a dynamically determined length scale is much smaller than the system dimensions.2.( 7 ) and eB (73).NBx(L)' Derive general expressions for 6.&. (a) State the equations governing the concentrations of reactant and product in the membrane. A+B /f' . DA= DB= D.. Several types of simplifications which can be made in steady or timedependent problems are illustrated in Sections 3. . Nonetheless. like the scaling process itself.. Certain solution methods are based on many of the same ideas. and concludes with integral approximation methods in Section 3.Feed Stream Membrane Product Stream C .6 and 3.8. The examples in this chapter involve conduction and diffusion only. and integral methods. assume that the concentrations of A and B in the product stream are maintained at low levels and that the difisivities in the membrane are equal. The discussion of solution techniques begins with the similarity method in Section 3.Membrane reactor. C. Finding the simplest differential equations and boundary conditions which adequately represent a real system is something of an art. and in complex situations some trial and error may be required.s 0 <\ .7. but the methods are more general. onedimensional problems. When are such assumptions valid? The key to model simplification is the concept of scales. N ~ (O) x N~~(0) . .3 and 3.
22) (3. then by definition. the distinction between the two types of scales may be important.1. The three main classifications for a ). Because we are concerned mainly with differential equations. When one quantity is a constant. " x  1" is not the same as "x= O(1)" (see Section 3. in this case pure numbers. diffusion. which is applied to both temporal and spatial derivatives. the actual second derivative may be much smaller than this estimate. (3. For these estimates we assume that. concentration. Ax. The scale of any variable may be influenced by the values of dimensionless parameters. each with the same number of variables and parameters.3 and 3. Thus.1 and x> 10 are reasonable starting . dimensionless quantity are that it is small (x<< 1). the temperature and length scales are A T = H . Using these scales in Eqs. (3.21). (3. (the temperature increment above the ambient) varies smoothly by an amount ~ " ~ ' 1 4 over k the distance R. Iff and x have been scaled. Scaling is a special type of nondimensionalization in which the independent and dependent variables are chosen so that they are order one. "" is used to denote an orderofmagnitude equality. For Bi >> 1. Scaled variables are created by dividing dimensional variables by their respective scales.Af dx2 Ax (Ax)' (3. the concentration scale for a solute in one region of a liquid may differ from that in another. and the scale of a dependent variable may differ in various domains of the independent variables(s). respectively. R ~ / and ~ Ar= R. Using this guideline. but x<0. the smoothness of a function is ordinarily taken for granted and the main problem is to identify Af and Ax. '  ing a complex problem. making the derivatives order one is the main priority. then Af.sets of such quantities can be constructed. we need to consider the scales for changes in the variables. Throughout this book. and the first and second derivatives are order one." In orderofmagnitude equalities algebraic signs are ignored. two or more alternative . viscous stresses) have the effect of smoothing the corresponding field variables (temperature.T. When this is done. large (x>> 1 What is effectively small or large dependson the problem. For example. as in Eq.6).23) ' Scaled Variables Putting the mathematical description of a physical process in dimensionless form ("nondimensionalization") is a standard procedure for minimizing the number of variables and parameters.23) assumes that the first derivative changes from roughly zero to AflAx over the interval Ax. (3. As indicated.    V The finitedifference expression in Eq. . Indeed. The factor "f' in the temperature change is of marginal significance for orderofmagnitude purposes and will be ignored. because we are usually hying to identify terms which can be neglected. a rough guideline is that a threefold discrepancy is tolerated in either direction.81. Thus. and this is where trial and error is sometimes involved. velocity). this is said as "x is of order y" or just "x is order y. this tendency to overestimate the second derivative is rarely a concern. Dimensionless variables also have scales. For a function with very little curvature. To scale transport equations we need reliable orderofmagnitude estimates of various first and second derivatives. the changes in variables are usually more important than their absolute magnitudes. will be accurate if f varies smoothly and monotonically over the interval Ax.points when no other information is available. just finding the proper scales can be the main goal in analyz where Hvis the heating rate and R is the radius of the wire. Thus. The scales in some transport problems are easily identified.' If x and y are variables. First. as already mentioned.1. or order one (x I).23).Orders o f Magnitude In discpssing which quantities are large enough to be important in a given problem and which are negligible.22) and (3. l'he scale of a variable is an estimate of its maximum order of magnitude. The heat and mass tmnsfer examples in Chapter 2 all involved some use of dimensionless quantities. but in others they are far from obvious. "x . the orders of magnitude of the first and second derivatives are given by df Af dx Ax' d2f ( A f I W . the magnitudes of various terms in an equation are revealed by the dimensionless parameters which appear as coefficients. Or. In estimating spatial derivatives we are aided by the fact that the molecular transport processes (conduction. for a function f ( ~ ) which varies grudually by an amount Af over an interval Ax. In all but the simplest problems. Length and time scales are also called chamcteristic lengths and chamcteristic times. the derivatives are estimated as 7 As Ucd in this book.0 . In situations where a variable changes by a small fraction of its absolute value. the expression for the temperature profile reduces to In scahng the variables in differential equations.22). Scales for Known Functions The concept of scales will be illustrated using three functions selected from the heat and m a s s transfer examples in Chapter 2. That is. not just for the variables themselves. we need a symbol to indicate order of magnitude. then "xy" indicates that the marimurn values of x and y differ by less than a factor of ten. consider the electrically heated wire of Example 2. the concentration scale at early times may differ from that at later times. Doing this may require considerable insight into which rate processes are important. we must find the scales for derivatives. It is seen that T. However. suggesting terms which may be neglected. If O and q are scaled dependent and independent variables.1" means that the maximum value of x is between 0. Eq.
In other words. (In this problem the key parameter is a = ~ . we need ways to find the scales for unknown functions. and r are divided by their respective scales to give O and 17. In convective transport problems.81.> 5. In other words. As a second case consider diffusion across a liquid film with an irreversible. L ~ />> D. Example 3. Thus.217) or (3.)IAT. we will pretend that no solutions are available. indeed. (2. Eqs. rather than by the geometry.84. derived from Example 2. . a closedform mathematical expression involving known functions). the temperature must vary gradually over the entire cross section. we again use two of the examples from Chapter 2. (3. The concentration of the reactant for this situation is obtained by letting a+. By setting AT=H. To illustrate this approach. The temperature scale is not obvious at the outset.21 Temperature Scale for an Electrically Heated Wire Again consider the physical situation in Example 2.843). especially. A surprisingly effective method for scaling unknown functions is to define the dimensionless variables in such a way that no physically important term in the governing equations is multiplied by a large or small parameter. the dimensional problem is stated as Here the concentration scale is C . the length scale must be R. we assume here that the reaction is irreversible and very fast. For simplicity. d m . R ~ I ~is the temperature scale found earlier from the solution. (3.24) becomes Scales for Unknown Functions 1 The first and second derivatives of this scaled function are i >N.. Sometimes.) The only parameter which remains is the dimensionless source term.T. The result expressed in dimensional variables is Given that the heat source is dismbuted evenly throughout the wire. using some prior knowledge of scales to simplify the governing equations is often the only way to obtain an analytical solution (i.83. recall that H . Thus. is the firstorder rate constant.T. For the dimensionless temperature we let O = (T.22) and (3. However. it is dynamic rather than geometric.(3.e. to confirm the scaling results. The concentration scale is again C . (3. SO that The length scale here is determined by the balance between two rate processes (reaction and diffusion).21) is due not to an error in scaling. and r)= r/R must be an adequately scaled coordinate.R'/~. where the surface temperature equals that of the surrounding air. The heat transfer problem expressed in terms of @(q) is where DA is the diffusivity and k. For Bi>>l. Thus. but the length scale is now Because most of the concentration it is irrelevant that the liquid f i l m actnally change occurs between y = 0 and y = extends much farther. and the length scale is L.It is easily confirmed that both estimates are within a factor of 2 of the true maximum values. There are no large or small parameters in Eqs. These exact values are consistent with Eq. then some means is needed to identify them before obtaining a solution. the presence of such parameters in strategic places usually means that the variables are improperly scaled. because the original set of equations was already simple enough to solve exactly. both derivatives are order one. That is. to y = L. The solutions are used only at the end. In these particular examples the scaling adjustments accomplish very little. and eliminating those parameters by embedding them in the variables corrects the scaling. the exact derivatives are consistent with Eq. (3. involves a homogeneous reaction in a liquid film.218). The scaled function and its derivatives for this case are d m . and that heat can be lost only from the surface. the film thickness. and y+O in Eq.21). For Da >>1. ~>>dm. we convert Eq. If T.23) provide the desired level of accuracy. (3. (3. to illustrate the thought process. heterogeneous reaction. where AT is to be determined. but to the peculiarity of the linear function.21). The third case.216) to Again. then Eq. the concentration profile is simply  If scales are to be useful in formulating and simplifying models. we have eliminated Ibe parameters altogether! This indicates that our choice of temperature scale is appropriate. it is the only practical path to a numerical solution. The scaled function and its derivatives are The discrepancy between the actual second derivative and the expectation from Eq.. As confirmation. as in Example 2. 1. B0.
.
d ?.2 M % c.a c.5 3 s 3 ~ g g z 0 0 9 4 s 8 O E :g 8 3 3 3 '3 c 'V!gw i a g 9 8 9 r 0g4 .'. 3 L d . .s 2 gj 5 91 Q.% = # 3 g2.a .E o . % .5 E 0 .o .4 a.m 83.3 32 boo E '5.a 0 S s.% m 3 2 3 c** . O g $2 0 0 0 & K ' C 3 ~ o p".2 m ag: = MSG 5 U o E 3E 2 g 5: 2 .
.
That is.317) we have used the boundary conditions in x and have approximated the surface temperature as The energy equation is now Thus. with Bi<< 1. + . Eq. (3.T. lead to the simplest form of solution. as expressed by Eq.322) becomes To obtain an energy equation involving section. (3. temperature variations in that direction have been neglected not because there is little or no heat transfer at the top and bottom surfaces of the fin. Following the approach of Example 3. rL . (3. if Bi is sufficiently small. However.. + . we obtain Thus. In terms of Z. This temperature variable is scaled but the coordinate is not.'P. The nonzero terms are we average each term in Eq.33). we choose m =i.324) where the constant m is to be determined. however. To. the energy equation for the fin is the internal heat transfer resistance is small.. A dimensionless temperature and zcoordinate are defined now as .310) over the cross To eliminate the small parameter from the differential equation. Estimating the order of magnitude of the derivative and rearranging the result as in Eq. a. we define the scaled coordinate as Z=Bim 6. the energy equation and boundary conditions are Assuming that the base temperature and heat transfer coefficient are uniform makes the temperature field symmetric about x = 0. is used now to show that temperature variations in the x direction are negligible. (3. Averaging in a similar manner the boundary conditions involving z.' .313).22. It is seen now that the scale for z is (wklh)In. (3. We conclude that the Notice that the tern representing conduction in the x direction ( d 2 ~ / d 2has ) not gone away. it has only changed form. For O(Z). the temperature variations over x are much smaller than the temperature scale. The scaled coordinate is then = (T h  T.319) and (3 320). The boundary condition at the top surface.).314).  . that neither W nor L is the correct length scale for temperature variations in the z direction. the tip will act as if it were insulated. incidentally. to good approximation.:* <. It will be shown. or the dimensional ones given above. (3. scding the z coordinate first will sharpen our insight and. The crosssectional average temperature is governed entirely by Eqs. Using these variables. Eq. can be solved for the average temperature. We have chosen W for the nondimensionalization because the crosssectional dimension of a thin object is often more representative than is the length.the boundary condition at the tip of the fin becomes h Eq. (3. This average is These equations. This dynamic length scale reflects the competition between heat conduction from the base to the tip and heat loss from the top and bottom. Given that the temperature field is approximately onedimensional. but rather because governing equations for O(Z) in the limit Bi 0 are . This indicates that T= T(z) only.85 From Table 22. . (3. the local value can be replaced by the crosssectional average (at constant 2 ) . we obtain The boundary conditions are .
The more general situation is discussed in Example 3. .21). t = O the concentration in one external solution is suddenly changed to C. The species conservation equation from Table 23 reduces to A The initial and boundary conditions for C(x. t) and D. we obtain 3. In effect. the geometric length is much less than the characteristic length.=T. in a liquid film (see Section 3. the imposed values of dG1d. No solute is present initially. .74. as shown in Fig. because the effects of one or more distant boundaries are never "felt" on the time scales of interest. In neither case is the aspect ratio. if T. These results resemble those for slow and fast homogeneous reactions. whereas the temperature in the long fin reaches the ambient value well before the tip.2).41H3. the controlling parameter. 34. In a stagnant medium the time involved is the characteristic time for conduction or diffusion. making this a difficult Wblem to solve in a completely general manner. which is modeled as a homogeneous material.(t).I.AND APPROXIMATION TECHNIQUES on Time Scales (b) Membrane The solutien is O ( Z )= cosh Z . we have Nxlx=o = .(t).(t).(r) and C. (To simplify the notation. Diffusion through a membrane separating wellmixed solutions. where K is the partition coefficient. . represents the solute flux in the x direction.47) are coupled through the concentration and flux conditions at x = 0 and x = L.4 SIMPLIFICATIONS BASED ON TIME SCALES If the temperature or concentration is suddenly perturbed at some location. . Evaluating the solute fluxes just inside the membrane. as shown qualitatively in Fig.D(o. The short fin is essentially isothermal. 35. . ax I ) . (a) Overall view. The limiting solutions for short and long fins are >. 0) = 0. and at where N. in that it determines how fast a system can respond to changes imposed at a boundary. with solute concentrations C. respectively. the specified function To(x) introduces an additional length scale which may nullify the use of the lowBiotnumber approximation near the base. This characteristic time is a key factor in formulating conduction or diffusion models. t) = KC. The two examples in this section involve transient diffusion across a membrane of thiclcness L. the subscripts usually used to identify individual species are omitted here. The behavior of the concentration profile in the membrane at short times is quite m e r e n t than that at long times.* Membrane x= 0 x=L $2 ' F e n 34. N. A fast response may justify the use of a steadystate or pseudosteadystate model. For Bit< 1 and y>> 1. is.(x)). t ) are <' C(x.) The external solutions are assumed to be perfectly mixed. respectively.respectively. (b) enlargement of the area indicated by tbe dashed rectangle in (a). y= LIW. C(0. t )= KC. the centration changes resulting from the step change at x = 0 spread over only a fraction .x may make it impossible to neglect temperature variations in the x direction in the vicinity of the base. Eq.tanh A sinh 2. a finite time is required for the temperature or concentration changes to be noticed a given distance away Erom the original disturbance. For small t. using the method of matched asymptotic expansions. The basic model formulation conunon to both is described first.D(L. Equations (3. that analysis serves as a prototype for other conduction and diffusion problems. As with the elementary model of a fin presented here. Because the base temperature is constant in the present problem. but with more general boundary conditions. A slow response may allow one to model a region as infinite or semiinfinite. . Conversely.74 is how to combine a twodimensional solution in one region with a onedimensional solution in an adjacent region. The solute concentration and diffusivity within the membrane are C(x. a c This completes the basic problem statement. (3. . a dynamically "long" f i n is one where A>> 1. the approach used here will yield a temperature profile which is a good approximation in most of the fin. but not near the base (z=O).331) is a good approximation throughout the fin. Each solution has a volume 1 ! and the exposed area of the membrane is A.331) A fin that is dynamically or functionally "short" is one where A<< I .=~= .%> $9 . Lumped models for the external compartments are derived from integral (controlvolume) statements of solute conservation based on Eq. (2..e.  We focus first on the membrane. The basic issue addressed in Example 3. (3. ax ac t). The objective is to determine how the various concentrations evolve over time. If the base temperature is not constant (i. Assuming that the stirred compartments are closed except for mass transfer to or from the membrane. or ~ c c ( W W h ) That '~.. C(L.
eventually would be achieved. The solution for C(x. 35(b).. it is the failure of Eq.46). (3...5 using the method of similarity.e.(t) remains very close to its initial value. depicted in Fig. The new feature here is that S increases with t. until the final state corresponding to t = t .e. it follows from Eq. so that this assumption eventually fails. Thus. Applying the second boundary condition at x = instead of x = L assumes that L>>6. (3.r to that for a steady state. (3.. and C.43) and (3. for diffusion over the distance L. During this second period the external concentrations gradually equilibrate. The original boundary conditions.22) and (3. S(t). Returning to the membrane diffusion example.49). such that t.( ~ f ) ' ~ .. which most severely limits the applicability of the penetration analysis. (3. This timedependent length scale is the distance over which significant concentration changes have occurred at any instant. (3. The characteristic length for small i is therefore the penetration depth. (3. As suggested above. Qualitative behavior of the solute concentration in a membrane after the solute is suddenly added to one of the external solutions. To ensure that S(t)ccL.48). Analogous arguments apply to transient heat conduction. as indicated by the curve for t = t . we wish to estimate the time interval for which Eq..<: . for which the characteristic time is where a is the thermal diffusivity. This shows that the penetration depth increases as the square mot of time. (3. Neglecting Be time derivative in Eq. The solution to Eq.23) to estimate the derivatives. for 6<<L). Concentration profiles'are shown for several distinct times. ac KC.49). rather than Eq..(t) and C.4I). Eqs. the membrane diffusion problem is now uncoupled from the external mass balances.(t) changes appreciably from Cw Thus. as evidenced by the cancelation of the factor KC. The second series of changes.416) for given external concentrations c. At any time t. this condition will be violated long before C. h p l e 3. for these short times. the concentration will change from zero to a significant fraction of KCo over the time r that has elapsed since the step change at the boundary.48). C. suppose that V is large enough that C..(t) is .44). were constant). If the external concentrations change slowly enough. .. the boundary condition applied at x = w.45) and (3.. a wellknown feature of diffusion (or conduction) problems. will ordinarily occur over a much longer time scale than the first.49) is much simpler than that stated originally.<t. is reached. It also provides an orderofmagnitude estimate of the time muired for diffusion to occur over a specified distance.413) that we must have Given that V is large. is a known constant. then the steady state corresponding to the curve labeled C. a 2 s2 (3.42). Example 3. (3. as in Example 3. of the membrane thickness. The purpose of the following examples is to establish the time scales for the two kinds of behavior and to show how the diffusion problem can be simplified when certain criteria are met. to zero over the distance 6. If the external compartments are large enough that their concentrations change slowly.41 Penetration Analysis for Short Times We begin with an approximation to the membrane diffusion problem which is valid for small t (i. The orderofmagnitude estimates of the derivatives are then . are then replaced by C(0. The penetration problem defined by Egs. (3.41 1) Substituting these estimates into Eq.410) d2C KC. (3.22..42 Pseudosteady Analysis for Long Times We now seek an approximation for the same membrane diffusion problem which is valid for long times. Moreover.4I). using Eqs. the  4iTbila. t ) is completed in section 3.41) and solving for S(t) gives 6. . (3. t) is Because C.<t. (3. This would occur after the changes had spread over the entire thickness L and after additional time had elapsed to allow the concentration profile to adjust to the presence of the righthand boundary. At any representative position x. (3.<t. then we expect the instantaneous profile in the membrane to resemble that for a steady state (i. (3. they are not affected by the concentration or temperature scales. in Fig. The dependence of the penetration depth on time is deduced now from orderofmagnitude estimates of the two terms in Eq. Eqs. and (3. at r ' ' (3. will in fact be achieved. these results can be used to estimate diffusion or conduction times in a wide variety of situations.' concentration will change from KC. Fjgure 35. These time scales for diffusion and conduction are insensitive to the problem geometry. the differential equation for C(x. Setting 6(t) = L and solving for t yields the characteristic time t .. will be valid. as if C.<t. t) = KC. 35(a).4I). If the external concentrations were to remain at their initial values. a curve very close to that for r = t.
.
The derivatives in Eq. Turning now to the initial and boundary conditions.52) and (3. including Abrarnowitz and Stegun (1970). it is evident that Eqs. With this choice.51) becomes where erf and erfc are the ermrfunction and complementary ermrfunction. (3.513) as the initial condition.g. involving a step change in concentration at x = 0 and t = 0. g(t) will have specific numerical values. where a is a constant. These functions are plotted in Fig. .41. respectively. (3. (3.4 is that. whereas erfc(7) decreases from 1 to 0.(Dt)'12in section 3. erf(q) increases from 0 to 1. If the assumption that 0= 8 ( q )is correct.517). Because g =g(t). The only difference between g(t) and the penetration depth G(t). such as x = L. (3. The technique is illustrated by an example in this section and by several other applications throughout the book. which is the number that can be accommodated by a secondorder.510) is nonlinear in g. this will be true Integrating Eq. the boundary conditions for Eq. 36. only two boundary conditions must be satisfied. Notice that the particular constant chosen in Eq.511). That is. in that this problem contains no fixed characteristic length. Values of the functions are tabulated in such books. (3. In this introductory example. ' A similarity solution is developed here for the shorttime.. (3. Although our knowledge of S(t) indicates that g(t) ( ~ r ) ' " . with Eg. the similarity method may be used with nonlinear as well as linear problems.512) and applying the boundary conditions yields where g ' =dgldt. To complete the solution.512). then neither t nor x can appear separately in the Chfferential equation and other conditions for 8.59) only if the product gg' is a constant. Eq.AND APPROXIMATION TECHNlQOES methods for solving partial differential equations which are discussed in Chapter 4.j. (3. Integrating Eq. One requirement of the similarity method is seen to be satisfied. (3. we will show how to derive g(t) without such prior information. .59) reduces to y. transient diffusion problem described in Example 3. but noticing that (g2)' = 2gg1. The properties of the error function and related functions are discussed in many mathematical handbooks. ordinary differential equation such as Eq. (3.510) has simplified the argument of the exponential in Eq. In terms of the similarity variable. As q varies from 0 to m.CI(KC. Using the dimensionless concentration 8 . The exact value of the constant is immaterial. The derivatives are given by Thus. The similarity transformation has evidently been successful. . except that a positive constant is needed to have g>O and g'>O. subscripts are used as reminders of which variable is being held constant in each partial derivative. t = 0 and x = w will both correspond to 77 = if Eq. there is not a boundary condition imposed at some finite value of x. the governing equations are for Eq. namely.511) once yields .). (3. (3.511) become and g(t) is a scale factor which is to be determined. spreadsheet programs for personal computers).513) is satisfied. (3. (3. Requiring this. where . Equation (3. (3. and they are available also from a variety of software packages (e. we need to determine g(r).54) will be equivalent to one another if That is. we obtain . it is a linear differential equation for g2. Assume now that 8 can be expressed as a function of a single independent variable r).' . . . Eq. . i. : . The simplest equation for O ( q ) is obtained by setting gg ' = 2 0 . whereas S(t) was used only to describe orders of magnitude.'* Integrating Eq.51) are expressed now in terms of q.
.
these definitions. Thus.[1 + a ( ~ T.2. i = K V/L. =4. so that the surface temperature is T.i.>> 1. (3. Solving for a.61). Eq.65) 1 ' where k .69: . The O(E) problem is I' : where H . we will compute only @. Improved approximations to the roots are obtained now by considering the O(E) t e r n in Eq. It is found that a . (3.. (the ambient temperature) and a is a constant.81 is modified now by assuming that the thermal conductivity and source terms in an electrically heated wire are both functions of temperature.= KV'/L~.612) is linear in a. (3. There appears to be no general solution which is valid for all E. ( ' ) = . just as Eq. it is assumed that Bi = hRlk. the source term in Eq.617) xl = .68) was linear in a. Eq. Essentially. i is the current density and K is the electrical conductivity of the wire. and = 0. such that e<<l.62 Electrically Heated Wire with Variable Thermal and Electrical Conductiviti@ The problem in Example 2. Note that Eq. It follows that H. O. (3. Substituting Eq. is the thermal conductivity at the reference temperature T . so that the first approximations to the desired roots are I Regular Perturbation Analysis 97 k = k.(r]) and @. Equation (3.616) is written as where xi denotes the ith root of Eq. (3.2 + O(E).623) into Eq. with the boundary conditions Continuing this procedure to 0(e2) gives the problem. The problem now is to determine the coefficient functions.613) ' Suppose now that the temperature dependence of k and Hv is relatively weak. The three roots of Eq. because the coefficient of dOldr) depends on O. which is a feature of all regular perturbation problems.61). The energy equation is Because O depends on the radial position as well as on the parameter E . greatly facilitates continuing the procedure to the desired level of accuracy. we adopt the view that the O(1) terms have already been satisfied. In other words. the roots of the original equation are now where the temperature scale is A T = H .63). ' ' )= . j (3.(17).68) is a linear equation which may be solved for the corresponding a.(r)). 2 8 (3.'').ÿ APPROXIMATION TECHNIQUES This first approximation is equivalent to setting E =O in the original equation. (3. (3. even though the original equation was not. A convenient set of dimensionless quantities is Because the three %") are already known.64) are a . (3.2 E+E~+O(E~).(3. the coefficients in the expansion are not constants. (3.We will be satisfied with the first correction to the temperature profile caused by the variable thermal properties. Because conduction of heat and electricity in metals is by the same mechanism. To simplify the analysis. The local rate of heat generation is given by Hv= i 2 / ~where . From Ohm's law. Accordingly. With . is a constant. Eq. (3. (3. 1 I .") yields XI= . Example 3.)]. we obtain A key feature of this example was that the governing equations for O(E") with n> I were Linear. we assume that K E k.620) is nonlinear.. R ~ I ~ . This linearity starting with O(e). as they were in the previous example. aJ2) = 0. Treating the temperature as a function of s and expanding the solution as a power series.616) is transformed .. where V is the applied voltage (assumed to be independent of r) and L is the length of the wire. and that the 0(e2) terms are still negligible.620) yields The boundary conditions are expanded in a similar manner to give Assume that over some range of temperatures the thermal conductivity can be expressed as a linear function of the temperature. and aOc3) = 2. .
O. see Bender and Orszag (1978). to derive successively better approximations to @.(r)). is f i These are the equations for constant values of k and Hv. However.72) or x = 1..71 Solution of an Algebraic Equation As in the discussion of regular perturbation methods. Consequently.26). (3.g. .22 and 3.62 is a good approximation to the temperature profile throughout the wire. (3. The fact that 8. Thus. The key characteristic of such problems is that each of two or more regions requires a different approximation.= I or a = 1.. which is mandatory. a reduction in the order of the governing equation for E = O is a hallmark of singular perturbation problems.71) have very different scales. Nayfeh (1973). the two roots of Eq. ' 3. The O(E)problem for 0](17)is ' . as discussed in Section 3. the success of the perturbation approach in solving a differential equation hinges on whether the first term in the perturbation expansion (e. Consider the roots of Equation (3.) can be expressed in terms of elementary functions. For information beyond the introduction given here. To correct the scaling problem.?qk . either or both of the other terms on the Efthand side might be important.was a simple polynomial greatly facilitated the determination of (3.) happened to be linear also.630) is a linear differential equation for el(?). Singular Prmrbation analysis is relatively new. The flaw in the reasoning leading to Eq. Using Eq. solved earlier. which is the same as that obtained by setting e = 0 in the original equations.". Singular perturbation methods are needed to treat problems where uniformly valid solutions cannot be found. .(3. Thus.6. inadvertently lost because setting E = 0 reduced the equation from second order to first order. This implies that E"."  The O(1) problem governing a . (3. Such problems are singular in the sense that the solution obtained by setting E = 0 is radically different than the asymptotic solution for e +O. Doing that in Eq. We will proceed as if the roots of this quadratic equation could not be found exactly. It was noted in the previous example that the perturbation procedure always yields a sequence of linear equations beginning at O(E). it is instructive to begin with an algebraic equation. . Kevorkian and Cole (1981). the first two terms of the expansion for O are . the solution for . .. . so that Eq. and use the exact answer only to check the results of the perturbation analysis. there is an immediate difficulty: what happened to the second root? It was Thus. c.633) can be extended indefinitely (as time and patience allow). The solution is found to be . with differential equations it usually leads to the inability to satisfy one or more boundary conditions for E = 0.72) was the implicit assumption that x = O(1) for both roots.. (3. Matched asymptotic expansions have been described in many texts on applied mathematics. E . in the present example the O(1) equation (that governing O. As a rule. From Eq. Example 3. .32. .7 SINGULAR PERTURBATION ANALYSlS An important feature of regular perturbation solutions to differential equations is that the expansions are uniformly valid for all values of the independent variable. Although not true in general. (3.71) gives ~1=0 (3. As will be shown. . This led to the expectation that ex2= O(E). . boundary layer problems in fluid mechanics.Resolving the difficulty requires that we pay careful attention to the scaling. with nonhomogeneous terms arising f r o m the O(1) solution. a cornmon feature of singular perturbation problems is that the small parameter. As already mentioned. As shown in the examples which follow. The procedure followed in obtaining Eq. a new variable is defined by setting . having been employed first in the 1950s for the similar to the transformations used to scale the coordinates in Examples 3. In the present p:. the solution in Example 3.. the first approximation to the solution is found by setting E = O in the governing equation. for example). through what is called a dominant balance. setting E = 0 reduces the order of the equation. or Van Dyke (1964). the highest derivative). in the limit E + O .73) in Eq.< . 8. ( 3.g. the solution is #<. This reduction of order is sufficient to invalidate a regular perturbation expansion. Aside from x2. where E<< I .71) yields What is needed now is to identify the most important terms for E+O. (3. However.74) becomes ' . multiplies the highestorder term in the equation (e. Out of this research developed what is now called the method of marched asymptotic expansions [see Chapter 4 of Van Dyke (1964)l.7. Let us assume that the dominant terms are those involving x 2 and X. Lin and Segel (1974). In regular perturbation analysis. :$ + B = O is not an acceptable starting point for constructing better approximations. The Constraints which determine a are that the governing equation be second order (allowing us to h d both roots) and that X = O(1) at most. not all singular perturbation problems have this feature (see Section 7. enample the singularity manifests itself in the loss of one root.
Thus.I/').. the small parameter is e =DaP'.719) but not Eq. Eq. This is made plain by the fact that @=0 is consistent with Eq.61. (3. Example 3. (3.. The Concentration boundary layer .cn. and x2. Consider a cylindrical catalyst pellet in which a fast.720). The pellet is modeled as a permeable medium with an effective difhsivity D for the reactant and an effective homogeneous rate constant k. To rescale the radial coorhnate for the boundary layer.is shown qualitatively in Fig. : : . the differential equation reduces to 8 = 0. (3. the singular perturbation analysis has provided the first terms in an expansion of the exact solution. as shown below.ed to be dominant in Eq. irreversible reaction leads to a sharp drop in concentration near the outer surface. which is small in the boundary layer. (3. C(r). The boundary conditions are = 1. c o' (3. The naive approach of setting E = O in Eq. this solution is obviously incorrect near the outer surface. showing the concentration boundary layer which results for Da>> I . The region in which the concentration declines sharply is called the concentration boundary layer: The dimensionless reactant concentration and radial coordinate are given by C 0 ~ The remaining term is now O(E. The O(1) problem is ' so that the roots at leading order are a.71) are.712) and (3. If we set E = 0.1 and a i 2 ) =0.77) and computing successive t e n s as in Example 3. respectively. x= . The exact roots of Eq. That is. This illustrates how trial and error is sometimes involved in determining scales. The roots X. (3.'<(. ".(')= . (3.(')= .717) becomes Expanding the numerator for E<< 1 gives It can be seen from Eqs. (3.72 Diffusion in a Cylinder with a Fast. making the stretching factor large will require that bcO. Schematic of a cylindrical catalyst pellet. Using cylindrical coordinates. This would have given reactant concentration at steady state. x. As seen now in Eqs.713). but missed x1 entirely.71) gave us an approximation to x. The different coordmate scalings will lead to different approximations for the reactant concentration in the two regions.(2! In terms of the original variable x. the species conservation equation is X= n=O x 0 0 a. a different variable is needed for the boundary layer. are then which gives a. we would have concluded that E'"' = 1 or a = 112.alysis The term that does not involve X is O(E).74). Homogeneous Reaction in this example the singular perturbation approach is applied to a boundaryvalue problem. (3. firstorder reaction occurs at sites which are uniformly distributed throughout the pellet.1* 4 1 + 4 & 2& ' The quantity (1 . is "stretched" using the factor cb to obtain a coordinate which is O(1). Changing to the new independent variable. We conclude that whereas t) is scaled adequately for the core region. accurate up to O(E').= O(1) but x1= O(E'). Because 6 is small. Equation (3..75) provides the basis for the perturbation solution. Equation (3. . iy r.716) R' where C. by definition. is the concentration at the outer surface and R is the pellet radius. + c.consistent with our assumption that the others are dominant. the roots are . . of course.. The fast. What if we had guessed the wrong set of dominant terms? If x2 and the term involving E were assum.717) is an immediate indication that the perturbation problem is singular. (3. (3. let .1 and a..q). as for the analogous problem involving a planar liquid film discussed in Section 3. .I' .715) that the "minus" and "plus" roots correspond to x. see Aris (1975) for a discussion of these parameters. As already mentioned. The O(E)problem is & ' Because the reaction is assumed to be fast. Although accurate for the "core" region in the center of the pellet.2. t=O(1) in the boundary layer.714) and (3. 37. . Both the diffusivity and the rate constant depend on the catalyst material. The fact that E multiplies the second derivative in Eq. .75) is solved by representing each root as a power series expansion in E . the difficulty came from assuming that x = O(1) for both mots. which contradicts the assumed dominant balance. Figure 37.
Accordingly.719) here.~ + o(eZb>]d 8 . the boundary layer coordinate is ho=e5.736) The boundary condition at the surface.@ = 0. which must be handled separately. Using ' &. (3. (3. which requires that a. Eq.((I) = 0.739) to calculate the flux into the cylinder gives .720).or b = . the first and last terms are balanced to give 26 + I = 0. Unlike problems involving distinct regions separated by a definite surface. (3.732) Thls completes the O(1) solution for the boundary layer. Instead.112. large values of the boundary layer coordinate 6 correspond to values of the original variable q near unity: In general. (3.735) te6 Substituting Eq. (3. The fact that E appears only as E ' in ~ Eq.728) that satisfies Eq.cb+l[l+ dt2 E & . (3. the O(1) problem for the boundary layer is * ' in either form of Eq.(m) =O. (3. it is necessary to match Eq.732) into Eq. TO evaluate ao. = 1 and Thus. The boundary condition at the center applies to the core region. as defined by expansions for each of these concentration variables in E. expressed as ' n the final expression comes fro? @ o ( [ )and is the same as the flux into a planar $lab of material.e.Singular PerturbatSon AnarySIS  Given that &Kb will be small in the boundary layer (i.723) Equation (3. because the boundary layer does not extend to the center of the cylinder. (3. is expanded in a similar manner. such as a phase boundary. the matching must be accomplished using the successive levels of approximation to 6(5) and 6(q). of order zero (see Section 4. (3.2 ' 6. = 0. and the reaction term is a key part of the physical situation. denoted by @=8(q). The curvature of the cylinder surface has no effect at O(1) because the boundary layer thickness is much smaller than the cylinder radius. the boundary layer is thin enough to make the surface look flat. rmative to compute the flux of reactant into the cylinder. In general.725) gives fies the (trivial) matching requirement that @. which will give the first correction due to the The 0(&In) curvature of the surface.737) 1 This completes Notice that Eq.7383 where a. T+I (3.N .734) To distinguish between the two regions.730) to the core solution. Ofimportance.731) gives @. \ ~ = ~ =dr D r =~ R l = o) E53e112  [ 1 112 & + ~ ( e ) ] . The composite solution for the present example is (3. The solution to Eq. an overall or composite solution can be constructed by adding the solutions for the two regions and subtracting the part of the answer that is common to both solutions: 8= &([) + .lim 8(rr). the matching is not accomplished at a fixed location. Eq.(3. (3. (3.733) and solving for =. (3. dS (3. it is not legitimate to use &.725) suggests an expansion for the boundary layer region of the form Substituting Eq. (3. That is. gives (3. is a constant which remains to be determined.. the convex shape of the cylindrical surface reduces the flux.() The required value of b is determined by the dominant balance. ple illustrates many features of problems which can be solved only by singular perturbation.726) into Eq. which is given by Dco dC .733) (3.7). In the present problem the matching requirement is trivial. Thus. From Eq. problem for the boundary layer. (3.728) is the same as the equation which would have been written in rectangular coordinates.731) is based on the observation that for E < < 1. because it is readily shown that 6(q) = 0 at all levels of approximation.61lirn B(g)= B(8 + 8(v) .736) corrects for the "double counting" which would otherwise occur in the matching region defined either by f +m or qt 1. The result is eZb+ d20 '. (3. The equation must remain second order. b<O). As can be seen. (3.(3. the concentration in the boundary layer is denoted now as 8 = 6(0. the matching must be done in an asymptotic sense.727)and that boundary condition. is given by and the corresponding species conservation equation is @. but it also happens to have an exact solution valid for the entire cylinder. The exact where Io((x) is the modified Bessel function of the first kind. The second term comes from el([) and is a first approximation for the effects of surface curvature.729) is the solution for the concentration field up to O(E). the coefficient of the first derivative can be simplified using an expansion of the form 141 x) = 1 + x + 0(2) for x+O.
it is straightforward to compute the rate of formation of the product (species C) from Eq. (3.753) apply to an outer region.753). (3. : ! . With time rescaled in this manner.(t). . The singular perturbation approach also applies to problems involving two or more divergent time scales. (3. * dC*= k.+k.746) to eliminate C.746).752) and (3. (3.(x)'is the modified Bessel function of the first kind.. Making the time dimensionless using k. while Eqs. I .740) is found to be identical to the result in Eq..743) :. as will be illustrated now using a problem in chemical kinetics. This fact motivates the choice of a new time variable o for small times. ! < ' Equation (3.(t).72 was the emergence of two very different length scales for concentration variations. for the boundary layer. . (3. given by Eq.744) become For a wellstirred batch reactor of fixed volume.CA fk. . (3. at concentration C. Consequently. (3. . . The QSSA is expected to work best when C. Eq.746) provides a suitable scale for C. but not necessarily changes in C. 1970): . . It is assumed here that only species A is present initially. (3. (3. . ' . the species concentrations are governed by the following firstorder differential equations and initial conditions: %#$ . (3. is equivalent to adopting a time scale representative of changes in C. Eq.AND ~ R O X I M A T ( O N TECHNIQUES e(r)5 Co ' (3. Eqs. abbreviated here as QSSA. .$! 1 1 . ~ 0 ) = 0 would . . (3.. with the result that += e.751) Example 3.. the ability to satisfy the initial condition for the intermediate. c. (3.747) (3.745) is not considered further because C.. (3. Evaluating CB(t)using that result and Eq.even for small 7. . of order one. . we introduce the dimensionless variables . . .750) (3. would then be replaced by this algebraic equation. (3. dt CJO) = CO. . and C .(t} in favor of C.743) and (3.744) gives One of the differential equations. The rescaling of time used here is analogous to the rescaling of distance used in steadystate .753). as seen in the previous example. It is evident then that the QSSA is not uniformly valid for all 7.Further reflection indicates that the QSSA fails at small r. To examine the limitations of this very useful approach.When the QSSA is even roughly correct. &. . the singular perturbation analysis yields an expansion of the exact solution. In terms of the dimensionless variables. (3.~ the ~ ) flux .:boundary layer problems..? . it is common practice in kinetics to assume that dCB/dt 0.738). A. If the second reaction is fast enough to maintain the intermediate (species B) at relatively ]OW concentrations. (3. The remaining parameter. it must be that E d@dr= O(1) for sufficiently small T. be lost.# '. . (or 8 and 4 ) can be obtained independently of C . ( E . and C .748) where I. Eq.:. the governing equations for small times are Using Eq.740) have the following asymptotic expansions for large arguments (Abrarnowitz and Stegun. This is called the quasisteadystate approximation. Eq.~ ~ ) / I . The Bessel functions in Eq..745). ( E .' Using these results to evaluate the first two terns of the expansion for I . Using this approach. adopting the QSSA would lead us to neglect the time derivative in Eq.756) and (3. . In singular perturbation parlance. With these definitions.754) ' .749) and the dimensionless parameters e rk L k. much smaller scale.753) is always 0(1).CB. as in the definition of T. Consider the sequential homogeneous reactions The definition of 8 is motivated by the fact that Cois the scale for CA. is not necessarily large or small. one for the core region and another. Because the righthand side of Eq. (3.G . 7 = kl t (3. when 4 increases rapidly from 0 to approximately 8. (3.<<C. . .73 QuasiSteadyState Approximation in Chemical Kinetics A key feature of Example 3. Thus. and so the analysis focuses on small values of E. Eqs.757) are the governing equations for an inner region.743) is readily integrated to obtain C.. thereby motivating the definition of 4.
72. is a constant to be determined later by matching. Because the same two constants are involved. Because these equations are valid for small times. Considering first the limiting '+". and using Eq.779). (3. (3. Because the solutions in neither region are identically zero.776). it is evident that the QSSA is simply a first approximation to the outer or longtime behavior. one of the a.. Eqs.764) is not valid for small times. (3. although the term e" was neglected. ( T )+ Eo ( ~E ~ ) .763) for n = O and Eq.daow~ of the form of the O(e) term in the outer expansion.761) that all higherorder (n 2 1) corrections in the outer solution are governed by lg 4. and a . the algebraic equations are Eq. $(u) = $.765) for n > 1.6 .778) reduces to N * the regmuping of terms caused by the variable change.767) and (3. the matching requirement is +. 780) up to 0(e2). because Eq. each pair of terms in the outer expansions happens to be governed by one differential equation and one algebraic equation.760) and (3. dependent variables 8 and 4). *:% ' 4'. Expanding B(T) and &T) in integral powers of E as gives the inner equations. (3. the I has not been W e d relative to hw. Consider now the inner problem. (3. To do this. Solving the n = 1 outer equations for 8.779) to match Eq. The solution to Eqs. The differential equations are Eq. ( w ) + E~z(o)+ E~ O(e3). (3. multiplied by where a. lim 70 &r)= lim &w). by using the initial conditions. To complete the solution through O(s) it is necessary to evaluate the constants a.762) and (3. (3. + &T) = A(@ + 6. Eq.(w) (3. w h i c h appear in the outer solution.770) . Choosing 4. (3. (3. where a.(T)EI~ e(7) = fi0(7) a1(7)& + +O ( E ~ ) . wbm (3.behavior of 4 and expanding the exponentials in Eq.769) (3. For Eq..20 expansions must be matched asymptotically. It would be incorrect to try to evaluate a .759) the governing equations for the outer region become . (3.777) . (3. (3. We begin with the outer problem. Solving again only the O(1) and O ( E )problems gives It follows that the O(1) outer problem is Because Eq. either r or o could be used. The two solutions are matched asymptotically. this provides a better illustration of the mechanics of asymptotic matching than does Example 3.763) is equivalent to the QSSA. In this case there are two differential equations at each successive level of approximation. sions to a common independent variable. (3. terms is now E.766) for na 1. . dependent variables 8 and 4 ) and for long times (the outer region. ~t is necessary that + &(o)E + ~ ( w )+ E O(e3) ~ . j where n 3 1. and 4.764) yields Recalling that T= WE. it is not necessary to work with both 0 and it is sufficient to match the expansions for either.762) for n = 0 and Eq. as shown.758) (3. Eq. the limiting behavior of 6 is 8 (o)= &(w) + ~ . (3. (3.763) is . (3. It follows from Eqs.\?rc p r ~ e c d now to solve for 0 and 4 for short times (the inner region. the inner and outer . it is now correct to invoke the initial conditions for 8 and 4.768). Expanding &o)and &a) in integer powers of e as Neglecting the exponential in Eq.768) for small T gives Thus. is another constant. The next step is to examine the limiting behavior of 4and 4and to convert the resulting expres.
32.1 +F[l + h(w. 38 for c = 0. (3. (3.32 it was mentioned that when the conhtions for the fin approximation are satisfied. as mentioned earlier.776). Eq. Eq. but is very close to the composite result for 0 > 4 . and the composite solution from the singular perturbation analysis.(x) may be z = . (3.(x). corresponding to A>> 1 in Example 3.E8(l. 0 Figure 38.782) += $+ 4.inner composite ' This notation is similar to that in Example 3. namely small Biot number and large aspect ratio. Outer solution. are compared in Fig. the fin approximation will not in general satisfy the boundary condition at r = O We have seen that the failure of an approximate solution to satisfy a boundary condition is typical of problems where singular perturbation analyses are useful. (3. Dimensionless concentrations of species B obtained using the following: the QSSA.the governing equations for the outer problem are Example 3. The dimensionless temperature and coordinates which will be used are . the outer expansion. so that To= T. :.108 SCALING AND APPROXIMATION TECHNIQUES Singular Perturbation Analysis 109 1 ' inconsistent with the fin approximation's neglect of temperature variations in the x direction.782). . the inner expansion. The temperature at the base of the fin (z=O) is assumed now to be a function of the transverse coordinate. l+h 4 where $ and are evaluated from Eqs. (3. it is assumed that To(x) is symmetric about x=o. and the rest of the fin as an outer region. The singular perturbation methodology has been applied successfully to several other reaction schemes. and that is true here.32 as Denoting the outer solution as @(q. expressed as &w). The various results for the dimensionless concentration of species B. The problem formulation for O(q. equivalent to Eq. Z ) . rather than the crosssectional average. this example demonstrates that the QSSA provides an approximation to the true "longtime" or "outer" solution that is accurate at O(1)..QSSA . The z coordinate appropriate for the outer region was identified in Example 3.5) is By analogy with Eq. The O(E) correction obtained here provides an explicit indication of the error involved for small (but nonzero) E . In Example 3. Z) and changing the z coordinate in Eq.776). The results are for s = 0. The difficulty is that a given function T.768) and (3. (1963) and Segel and Slemrod (1989).736). the approximate temperature profile will be accurate for most of the fin. the method of matched asymptotic expansions gives rigorous bounds on the accuracy of the fin approximation and provides a procedure for calculating corrections to the approximate solution obtained previously.1 and A = 2. In summary. Eq. For simplicity. The solution obtained using the QSSA is inaccurate for short times. (3.764).74 Limitations of the Fin Approximation We return to the problem of heat transfer in a thin object. Most accurate is the composite solution from the singular perturbation analysis. so that E =Bi = hWlk. This completes the solution through O(E). respectively.. In other words. but not necessarily near the base.768).32. :: The small parameter here is the Biot number. except that O is now the local temperature.. (3. Thus. the composite solution for 4 is (3. which interpolates smoothly between the inner and outer solutions. 33. By treating the part of the fin near the base as an inner (boundary layer) region. the aspect ratio y will not appear. for examples see Bowen et a]. the fin depicted in Fig..1 and A = 2. The analysis is simplified somewhat by assuming that the fin is dynamically long.785) from 5 to 2 .outer .l)] + 0(c2).
.
= Da) turns out to be exact.. (3. For consistency with Eq. although the concennation profile. The maximum error of about 6% is at Da = 1. but no approximations have yet been made. O =eDaq. 6. Substituting Eq. Thus. other than inferring (from scaling arguments) that the flux must be proportional to Da".84). because (as already mentioned) 6 = ~ a . the original differential equation has been reduced to an algebraic equation for &For specified values of Da and n.82 Approximate Tkansient Solution We return to the transient reactiondiffusion problem defined by Eqs. for special cases. is now a function of time. the assumed profile is made to satisfy Eq. The approximation comes in picking a concentration profile which will allow us to estimate the derivative and integral in Eg. Eqs.the exact concentration profile and flux are The only difference from Eq. Explicit expressions for 6 are obtained from Eq. in principle.89) will complete the approximate solution for @(q). The integral equation to be solved is now ' The key step. (3.88) is not unique. A reasonable choice is Thus. intractable differential equation. Eq. (3. (3. (3. (3. turn out to be exact.81 ApproKimate SteadyState Solution We begin with the steadystate version of the reactiondiffusionproblem described above: Integrating the timeindependent form of Eq.87).810) and (3.For n # 0.88) that 6' is equal to the dimensionless solute flux entering the film at 7= 0. Example 3. A result for Da+O and any n can be obtained using a regular perturbation scheme. (3.86). or Da>> 1 . as before. Further analytical progress can be made for fast reactions.85) has been used. Notice from Eq. (3. That is.e.88) is qualitatively correct.817) into Eq. Of course. which is TABLE 31 Comparison of Exact and IntegralApproximate Fluxes for a Steady.88). an iterative solution of Eq. Eq. 6<< 1.83H3. (3. . Equation (3.Although it cannot be solved analytically. and it is consistent with the boundary condition in Eq.81. (3. Eq. dO ldq = 6. (3. Although a singular perturbation approach applies. is the selection of a suitable functional form for the concentration. but'not for other values of n.. (3. (3.83) over the thickness of the film yields the desired integral equation. the governing equation for the O(1) boundary layer problem has the same form as the original. By requiring that 6 ( 0 )=0. no progress can be made in that case.lin that case for all Da. 6 (). (3. to the opposite extreme of Da+ co.85). respectively. With S>O. FirstOrder Reaction in a Filma Da Exact flux Approximate flux % Error "The dimensionless fluxes shown are the values of (a@/aq) at q=0.88) into Eq. the partial differential equation for 6(q. but also for the corresponding steadystate problem. although in general it will not satisfy Eq. . =+% tanh 6. for T= 0.for n = 0. The values of 6' are compared later w i t h exact results for the flux. the approximate flux for n = O (i. Da 0 and Da +=.86). (3. The penetration depth at steady state. in that many other functions (e.=o I where S is a constant. These results are Example 3. various polynomials) exhibit similar qualitative behavior. (3. This is true not only for the transient problem shown.87). and performing the differentiation and integration.g. S must satisfy ' A comparison of the exact and approximate fluxes for n = 1 is given in Table 31.89) in the limits of very slow or fast reactions. (3. Thus. (3. For n = 0.88) is that the dimensionless penetration depth. The accuracy of the integral solution is judged by comparisons with the exact analytical results obtainable for n = 0 and n = I . In this case.88) gives the necessary decrease in @ with increasing distance into the film. the approximate transient profile is chosen as There are no slow or fast reaction Limits for n = O . yields S = Da. Eq. is what was determined in Example 3. Substituting Eq.is not! For n = 1 the exact results are @=cosh d77 ' 6 7tanh 6sinh 6 ).8ll). (3.114 SCALING AND APPROXIMATION TECHNIQUES Integral Approximation Method 115 This problem can be solved exactly for n=O or n = 1 by the techniques presented in Chapter 4. whereas the asymptotic results for Da +0 and Da+ .816) leads to Thus. 6. r ) has been reduced to an ordinary differential equation for RT).818) can be integrated numerically using standard software. Thus.. (3.
irreversible. M. (3.811). The quasisteadystate assumption: A case study in perturbation. and J. SIAM Rev. In Point Defects in Solids. Arnold. A. Van Dyke. SIAM Rev. Mathematics Applied to Deterministic Problems in the Natural Sciences. and A. Lin. Vol. NJ. New York. a polynomial of a given order). 1961. A. ) The result is a firstorder. a discussion of these techniques is beyond the scope of this book. and L. A. J. U. Diffusion in serniconductors. The gas diffuses into the surrounding liquid. J . 1975. 14: 547571. National Bureau of Standards. Vol. we recover the result given in Eq. MA. Similarity Analyses of Boundary Value Problems in Engineering. 163255. Cole. 1989. Jr. I. In general. Derive expressions for the $ I A< c (tP L L . . and G. Handbook of Mathematical Functions. Kevorkian. is The two estimates of t. and 1. New York. an orderofmagnitude estimate for t. Dresner. Inherent also in weighted residual methods is a procedure for improving accuracy by adding more functions to the approximate solutions. H. from Eq. McGrawHill. Conduction Hear Transfer: AddisonWesley. Macmillan. Aris. where it is consumed by a firstorder. Kamke. (3. D. including orthogonal collocation and finite elements. and S.. A. Pearson. L. 1966.and V. Phys. linear equation for tj2. Slemrod.413). Consultants Bureau. Conceptually related to the integral technique is a class of methods based on weighted residual equations. Casey. t Problems 31. Bender. However. those methods are limited to problems which have a large or small parameter! M a n y examples of the integral method applied to heat conduction problems are given in Arpaci (1966). In this problem the dimensionless penetration distance is S(T) and the dimensional distance is S(7)L. = 0 far from the bubble and that the liquid is stagnant. J. Chem. It is permissible then to neglect all terms in Eq. 1972. Sci. C. Nonlinear Analysis in Chemical Engineering.116 SCALING AND APPROXIMATION TECHNIQUES Problems the upper bound for 8 is given by Eq.818) which are multiplied by e  for certain widely used numerical methods for solving differential equations. Eds. Unsteadystate vaporization and absorption. Absorption from a Bubble with Reaction Consider a spherical bubble of radius R(t) containing pure gas A. Washington. (3. Similarity. SpringerVerlag./n = 3. 1964.. 40: 361378. M. which may be very sensitive to the form of the trial function usedthat is. could be based on the characteristic time for diffusion over the maximum (steady state) penetration distance. Assume that C . weighted residual methods are more accurate than the integral method. Segel. resulting in (3. Hansen. provide (a) an orderly way of keeping track of the size of neglected terms and (b) a systematic procedure for increasing accuracy (adding another term to the series expansion). Oxford. New York. References This nonlinear differential equation is solved by first multiplying by S and then recognizing that 6 dud^) ~ ( 1 1 2 d(#)/dr. Finlayson. Clarendon Press. New York. Simplification and scaling. the assumed form of the concentration field (or other field variable). A. A.Perturbation Methods. New York. a pseudosteady analysis is justified for the liquid. V. Department of Commerce. New York. 1964. Slifkin. H. Reading. PrenticeHall. Nayfeh. Englewood Cliffs. The time required to reach a steady state (1. A. Segel.) is estimated from Eq. (3. by contrast. M. we set 2 Da r. Diflerentialgleichungen. even for problems which are intractable using other analytical methods.g. and M. 475. Leipzig. Segel.1... closely related limitation is that there is no systematic way to improve the accuracy of the approximate solution. are obviously similar. E. L. 1944. Academic Press. Weighted residuals form the basis A. WileyInterscience.Oppenheim. B. The liquid concentration of A at the gasliquid interface is C. Perturbation Methods in Applied Mathematics. Perturbation Methods in Fluid Mechanics. or Alternatively. Crawford. 1978. 1943. R. Arpaci. Chem. the underlying principles and numerous examples are provided by Finlayson (1980). L. SelfSimilarity. Among these methods are collocation and the Galerkin method of weighted residuals.E. 1981. M. Eng. S. and L. H. A. Studies in diffusion: 111. Vol. 1975. Solids 19: 3550.Ch. C. 1974. Barenblatt. G. One is that it is very difficult to predict the accuracy of the approximation. I. Plenum Press. H.811). pp. Slyozov. Accordingly. 31: 446477. homogeneous reaction (rate constant k). Lifshitz. Orszag. 1. C. The solution is Abramowitz.05 to be small enough. L. 1983. A.S. this approach has two very important limitations. Similarity Solutions of Nonlinear Partial Differential Equations.. DC. 1970. Acrivos. Advanced Mathematical Methods for Scientists and Engineers. M. The Mathematical Theory of Difision and Reaction in Permeable Catalysts.820) by seeing when the exponential term becomes small. 2. However. R. At steady state. The kinetics of precipitation from supersaturated solid solutions. Considering e3 =0. C. 1980. for a similar form of trial function (e.Ges. Jr. The integral method is impressive in that it yields approximate solu6ons with relatively little mathematical effort. 1963. 18: 177188. Boston.V. A. p. New York. Pitman. K. 1973. another estimate for t. J. Stegun. (a) If R(t) varies slowly enough and if enough time elapses following introduction of the bubble. Bowen. J. Tth *Z 5 $ Fbz 3 Gr i. 1979. Perturbation methods. emphasizing once again that the duration of transients can be estimated without knowing the transient solution. 1 . Akadernische Verlagsgesellschaft Becker & Erler Kom. and Intermediate Asymptotics. Trans. McGrawHill. Germany. G. Singular perturbation refinement to quasisteady state approximation in chemical kinetics. New York. The second.
assume that a solder wire of radius R is melting at a constant rate.) of As. and the flux of A in the liquid. The concentration profile is found to be like that shown in Fig. t)/r will be helpful. a metal piece which is to be joined to another). Rewrite the energy equation in terms of the local heat flux q. under these conditions. Are different sets of conditions required for ~<<R'/D and t >>R21D? (b) Show that when the conditions of part (a) are satisfied. .. After the reduction in the outside concentration.(K t ) 2 3/(r. Specifically.) This problem is based on an analysis given in Casey and Pearson (1975).118 SCALING AND APPROXIMATION TECHNIQUES  liquid concentration profile. P33. where x is distance from the exposed surface.. I ) rather than T(x. yielding a constant wire velocity v. k.. Determine the ratio of the flux with reaction to that without reaction. Use this approach to obtain an expression for T(x. Similarity Solution with a ConstantFlux Boundary Condition Consider a semiinfinite slab initially at temperature T. Assume that the conditions of part (b) are satisfied (i. can be evaluated by the method of similarity. Figure P33.) (b) In part (a). (same units as a mass transfer coefficient). as a function of the Damkahler number. T= T(z)only)? (b) Determine T(z) for the conditions of part (a).t).) (c) Use orderofrnagnitude reasoning to identify the range of times for which a pseudosteady analysis is justified. t) s E ( z ... The wire is long enough that it eventually reaches the ambient temperature far from the surface. Qualitative behavior of the concentration of As in solid Si.t).(x.C@dRldt..(x. convective transport caused by bubble shrinkage was neglected. (b) Describe how the result of (a) can be used to determine DA(CA) for 0 6 CA Co from a single concentration profile like that in Fig. t). (Hint: The transformation C.e. (R. An experiment is performed in which one surface of a sample of pure Si is contacted with As over a period 0 a t at. solute leaves the tube both by diffusion through the open ends and by diffusion across the tube wall. Under what conditions will the solute concentration in much of the tube be independent of z? Show that under these conditions the dimensionless concentration in the central part of the tube is given by h  Figure P35..) ? (Hint: Use an interfacial balance to relate Cb. 35. ConcentrationDependent Diffusivity Suppose that it is desired to determine the extent to which the diffusivity (DA)of As in solid Si depends on the concentration (C. C(2. terms of those quantities.. as a function of v. Transient Diflbion in a Permeable I'ube with Open Ends The openended cylindrical tube of radius R and length 2L shown in Fig. As shown in Fig. CA(r. t) = C. A constant heat flux q..' T.. Heat transfer to solder. P33. and other parameters. Under these conditions C(I. P35 is immersed initially in a fluid with solute concentration C. t). convection is negligible). the ambient temperature is T. and the latent heat of fusion is A. . (a) Identify the conditions under which radial variations in the internal concentration will be negligible. C. and show that q. is imposed at x = 0 beginning at t = 0.. NA. P32. The melting . r) with DA= DA(CA) can be transformed to an ordinary differential equation by introducing a similarity variable. at some time 1. and the flux of A at the gasliquid interface. At the end of the experiment the sample is sectioned and analyzed to determine C. 0) = c. Identify the conditions under which t h s is a good approximation. p . to).. t). Heat Tksnsfer to Solder Solder is made from various alloys designed to melt upon contact with a very hot surface (i. what must be the relationship between Co and the concentration of A in the gas bubble (C. ( ?i k L. Transient diffusion in a permeable tube with open ends. (Hint: Assume that the slope and the area under the curve can be calculated accurately. Air (c) Sketch qualitatively the concentration profiles for various values of t. (a) Under what conditions is a onedimensional analysis valid for the temperature profile in the wire (i. t) is the crosssectional average concentration. temperature of the solder is T There is a constant heat transfer coefficient h from the wire to the ambient air.<C.e. z. where c(z.. we obtain (a) Show that the conservation equation governing C. The diffusivity can be evaluated in Solder Wire Figure P32. (c) Determine the required rate of heat conduction from the surface to the wire.e. 32. the rate of transmural diffusion being governed by the solute permeability. The tube wall is permeable to the solute.(x. At t = O the concentration in the surrounhng fluid is changed rapidly to a new constant value. 34. Assume that L/R>> 1. i : 33. after the surface at x = 0 is contacted with As.
including terms of O(Da).] 1 1 3 I 39. are constants. determine the dependence of 6/L and C*/CAoon Da.C.. Assume that at any given time the precipitate consists of widely dispersed spherical particles of radius R(t). The small parameter is the Damkohler number. (a) Sketch qualitatively CA(x). In this region. show that the concentration of A is governed by [Hint: First combine the conservation where 8 = CA/CAo. assume that y is neither large nor small compared to unity./C. To do this. which do not interact with one another. Brown. and the rate of particle growth is controlled by diffusion. It is found that after an initial stage of this process that involves nucleation.r ) = xlL. (d) For fast (but not infinitely fast) reaction kinetics the reaction will be confined largely to a zone of thickness 6 at the center of the liquid film. Complete the solution. t). The volumetric reaction rate is given by Assuming that the reaction is slow relative to diffusion. and K. The objective here is to extend the analysis of Problem 212. This coarsening of particles. with B the abundant species. Assuming that the reaction is slow relative to diffusion.. precipitation may lead to the formation of grains or particles of a new phase.L21DA Obtain three terms in the expansion for @ = C. Da=kC. .. The objective here is to derive one of their simpler results. a similarity solution can be obtained for the entire tube. and Da = kCAoL2/~A. The relationship is (c) For slow reaction kinetics (Dacc 1). P212. A.. which is the result of Lifshitz and Slyozov. and use a new dimensionless concentration O(x. include terms which are O(Da2). Use orderofmagnitude considerations to determine S and C*.. Diffusion with a Slow SecondOrder Reaction Consider a secondorder. Da = k. with CA= Co at the surface (r = R). and therefore it depends on particle size. irreversible reaction occurring at steady state in a liquid film of thickness L.(x) in terms of CA(x). Scaling and Perturbation Analyses for a Bimolecular Reaction Consider an irreversible. In the second stage the degree of supersaturation is slight. Reaction and Diffusion with a Nonuniform Catalyst Distribution* !r .(x). include terms which are o(D~'). Obtain three terms in the expansion for @ = CAICo. for Da>> 1. The slab is to be used in a laboratory experiment involving the irreversible reaction A +B. * T h i sproblem was suggested by R. The second parameter which arises is y = K..). After deposition the catalyst concentration Cdx) within the porous slab is found to be nonuniform. use a regular perturbation scheme to determine C. it is sufficient to consider one representative particle for which CA= CA(ct) in the surrounding solution. a moderate value of k. The reaction is carried out in a long. An extensive analysis of the kinetics of the second stage of this process was given by Lifshitz and Slyozov (1961). from which the catalytic agent is deposited in the porous medium. 37. which follows the rate expression . and a large (but finite) value of k. r (a) Derive an expression for the growth rate of the particle. .. equations for A and B and solve for C. (b) For the special case of equal reactant diffusivities (DA= DB) and equal boundary concentrations of A and B (CAo = C.IC.Perhvbation Analysis of Dit'fusion and Reaction with Saturable Kinetics Assume that the reaction A+B follows the rate expression . dRldt. where k. The particles being formed by precipitation are composed of pure B. It is the interfacial energy term which provides the driving force for diffusion. define a new axial coordinate x such that x=O at one end of the tube. valid when growth is slow enough that diffusion in the solution is pseudosteady. Consider a solid solution consisting of components A and B. Specifically. a regular perturbation scheme can be used. and there is no transport across the surface at x = L. and show that as a consequence there is no convection. Assume that the catalyst concentration is given by 38.. known as Ostwald ripening. where L is the slab thickness and Cco is a constant. Far from the particle the concentration of A is constant at C . 310.liquid film at steady state. .' . has important effects on the physical properties of various alloys. there is a second stage in which the particles grow to much larger sizes. where the constant is proportional to the interfacial energy.) (b) Show that eventually R(t) t 1 I 3 . which was limited to "infinitely fast" reaction kinetics. A certain catalyst is prepared by immersing a thin sheet of inert porous material in a solution. Why does similarity work for the variable O but not for $?Why is it necessary to use x rather than z? 36. A key aspect of this problem is that the concentration of A at the particle surface is influenced by the interfacial energy. bimolecular reaction in a. Determine &r)) for slow kinetics. That is. and Cc(x) for a very small value of the reaction rate constant k. use a regular perturbation scheme to determine CA(x) and the flux of A at x=O. Accordingly. (Assume that the densities of the particle and solution are the same. as shown in Fig. Enzymatically catalyzed reactions often have rate laws of this form. (c) Use orderofmagnitude reasoning to determine when the pseudosteady assumption is valid. .That is. t ) = ex'"lbgx. These results show when the approximation used for infinitely fast kinetics (S= 0 ) is reasonable.Problems 121 (d) When the conditions of part (c) are satisfied. cylindrical region of radius R under steadystate conditions.. Ostwald Ripening' In a supersaturated solid solution..(r) and the flux of A at the surface. CA CB C*. The concentration of the reactant (species A ) at x = 0 is Co. The small parameter is the Damkohler number.J2/~AC.
when the characteristic length may be much smaller than R. . use a singular perturbation analysis to determine the steadystate flux of A entering the slab. nz) (a) Show that if axid conduction is neglected. the bulk gas temperature TG will be assumed to be constant. The variations in air temperatwe (TA) are expected to be of the form TA(t)= To+ T . assuming that the ampoule and its contents are moving at a constant velocity v. The conditions are such that Da will be very large. the heat transfer coefficient between the sphere and the air is given by h = kAIR. is governed by the Airy equation. Brown. (a) The thermocouple diameter is 0. P4445). Evaluate the first two terns in the expansion for the flux. TT(t). where T(r) is the catalyst temperature. Heat transfer in unidirectional solidification. so that the side with high catalyst concentration is placed next to the barrier.l PI) k (Btu hr.07 0. Time Response of a Thermocouple" A themocouple is to be used to measure temperature fluctuations in an air stream. 313. tive heat transfer from the furnace to the ampoule is negligible compared to convective heat transfer. h = 0. are constants and o is the frequency of the temperature oscillations. assuming that the ampoule and its contents are stationary. called Any functions. = 500 K. Aside from proper calibra . k = 0. use the singular perturbation approach to obtain the first two terms of a series expansion for the temperature field. The thermocouple bead can be modeled as a sphere of radius R. the opposite surface of the slab is to be placed next to an impermeable barrier. melt. Notice the effect of w on the magnitude and delay of the response. ft3) tp (BN 1b.I ft. sin 2 m t . h.) (b) Convert the energy equation of part (a) to dimensionless form. . P312. representative values are R = 1 cm. justifying a lumped model. = . For the purposes of this problem. Furnace Ampoule Furnace Entry of A will occur only at one surface. (c) Repeat part (b). (a) It is intended that the side with depleted catalyst (x= L) be placed next to the imperme I I I able barrier. = 400 K. [Derive a general expression. Explain why this will be true even for large w .The packing has an effective thermal conductivity k . A. Denoting the thermal conductivity of the air as k .010 in. (a) For orderofmagnitude estimates. 312. determine the axial temperature profile in the melt and solid. and identify the conditions under which there will be a thermal boundary layer. (b) Derive a general expression for the thermocouple temperature response. and heat transfer from the catalyst pellets to the gas is characterized by a heat transfer coefficient. releasing heat at a volumetrjc rate H . where the concentration of A will be maintained at CAO. (b) For the onedimensional conditions suggested by part (a). Heat h n s f e r in Unidirectional Solidification* One method for establishing the steep temperature gradients needed for directional solidification is to interface two heat pipes (constant temperature pieces) with t~mperaturesTI and T 2 which bracket the melting temperature T. 'This ptoblem was suggested by R.YrOblemS T Z J Figure P312. of the material. (Teniperature variations within a catalyst pellet have been neglected.015 500 0. Melt and solid are contained within a long ampoule of radius R.24 0. the coordinate r refers to radial position in the reactor. Show that the first term in a perturbation expansion for C. Other representative data are as follows: Parameter p(lb. The reactor wall is maintained at a constant temperature.U.1 50 ! " $ .] Find the location of the interface. which is of the form 't I I b Solid R I The solutions to this equation. Brown. Also show that radial temperature gradients within the melt and solid are negligible. A.5 W K' cm'. not using the specific numerical values of part (a). are given in Abramowitz and Stegun (1970. A component supplied by the flowing gas reacts exothermically at the surface of the catalyst pellets. packedbed reactor of radius R. and solid d l have the same thermal conductivity k. (b) Suppose that the slab is accidentally reversed. The heat transfer coefficient between the ampoule and furnace wall is h. and T. which is not necessarily at z = 0. TI= 600 K.') Air Thermocouple ' 0. ' Show that the thermocouple will be nearly isothermal at any given instant. a steadystate energy balance for the packing leads to 'This problem was suggested by R. For this arrangement. Show that radia ' 311. The latent heat of fusion is A. Assume that the melt and solid have the same density p. The catalyst surface area per unit volume is a. This arrangement is shown for a cylindrical geometry in Fig. . Assume that the thickness of the ampoule wall is negligible and that the ampoule wall. Tw. Heat Ikansfer in a Packed Bed Reactor A catalytic reaction is to be carried out in a long.05 W K.O F . T. Where are the large temperature m e n t s located? What are the physical reasons for the boundary layer? (c) For the boundary layer conditions of part (b). where To and T.cm'.
~ "e.5 * g g z gz 5 i.4 5 % : . 2.Qg 9 A'" 0 6 g z O b c a C 0 P.3 %yZ..f.e 8 % p g3ss..2 5.ij 5sb g' O 3 'S l s f 2 2 r 0 9 a s II $ 2 s..2 a a.2 5 Q).i'$ 3 e $ a s ~ ~ g A O z B & g. .5 3 0 2 g m % o 1) E Z .E g '= . 3: .s 3 :'q 8g ~a.z%g m '2 5 n ZE. = .e zy m ' ae .OE'S gs $ 2 .~ 2 g 4P g z a a g 8" Z Q 9 2 2 rn 5 03s **sg Y c . 8 0g.r 3 4 EoC. 2 gsz33$ g&II 322." cn a 5 a g t g g .g. " %0 00 2 .
A. The variables involved are qi= xl(2(cri t)' "). the interface location at time t is given by x=Mt). At one surface (x=O) the concentration of the reactant is maintained at C. ' 1 2 ( ~ F . Assume that the liquid and solid have the same density.. so that a true steady state will be achieved for t Let t. using the integral approximation method. (3. (Hint: erf x+ W& as x+O. respectively).!. objective here is to obtain a more specific estimate of the time to reach a steady state. (e) Consider the special case where the initial liquid temperature is at the freezing point..t).e. where i = S or L. (Hint: A reasonable fonn for the concentration profile is provided by an expression analogous to Eq. Show that a pseudosteady analysis involving TS(x.] 317. are held constant for t>>O. To define the proportionality constant. li . what is needed to make the thermocouplereading accurately reflect the air temperature? h his problem was suggested by R. As shown in Fig..4 it was suggested from orderofmagnitude considerations that a pseudosteady concentration profile would be achieved after a time ~ > > L ~ IThe D .) Use an orderofmagnitude analysis of the timedependent energy equation and boundary conditions to explain why a pseudosteady approximation is accurate when this requirement is met.To)erfc[ (&. . . the bulk temperature of the fluid above the wall is T. Complete the solution for TS(x. Also. obtain an orderofmagnitude estimate of the distance required for the surface temperature to decline from the heated value to T.(X. . t) and TL(x. (a) State the differential equations and boundary conditions governing the solid and liquid temperatures. approximations. show that the pseudosteady and exact results for 6(t) are in good agreement.Icy.e p d ~ F TO) ' 314. P315. *I . Assume now that the boundary concentrations. Fluid Metal Insulation Figure P314.T ~ ) ~ . As shown in Fig. Brown. P314. tion..whereas the other bounding surface (x= L) is inert and impermeable. occupies the space x>O.<. but different thermal conductivi :.. T . 6 ( t ) may be determined exactly. 316.. be the time required to approach the steady concentration profile to within 5%. at x = 0 and C = 0 at x = L. Solidification will begin.t) leads to Comparing this with the exact results of parts (c) and (d)..Y ~ ~ ~'" L yrr'I2h erf y k. where y is a numerical constant which remains to be determined. without invoking pseudosteady or othe. In this situation. respectively.! %*L. . Heat 'lkansfer f r o m a Heated Section of a Wall For a laboratory heat transfer experiment it is desired to heat one section of a flat wall in such a way that the heated section is nearly isothermal at steady state.?'.. (b) Tahng into account the fixed temperatures at x = 0 and x+but ignoring for the moment the interfacial conditions at x = S(t). the other dimensions of the metal sheet will greatly exceed W Resistive heating at a constant rate H v = C will be confined to a strip of width 2L by attaching a pair of electrical leads at that spacing (running parallel to the z axis). a reasonable approximation to the .. show that both the solid and liquid temperatures may be expressed in terms of error functions. which is below the freezing temperature TF of the liquid. t) and TL(x. : . t )"2. . 315.. :. Use the integral approximation method to estimate the flux of reactant at x = 0. Assume that convective heat transfer at the top surface of the metal sheet is characterized by a constant heat transfer coefficient h.)lfi<< 1. . Velocity of a Solidification Front Assume that a pure liquid.~ 2 kL aS L ~ ( T _ . Starting with the energy conservation equation for the metal sheet. provided that ~. ?"r. . e&~. use scaling arguments to identify ranges for the dimensionless parameters which will accomplish the aforementioned design objectives. At t=O the temperature at x=O is suddenly lowered to To. initially at temperature T..) 'I2]. it is proposed that the wall be a metal sheet of thickness W backed by a rigid material which provides good thermal: and electrical insulation. Time to Reach a Steady State in Transient Diffusion In the discussion of the membrane diffusion problem in Section 3.. and the solidliquid interface will advance into the region formerly occupied by liquid. . C = KC. use 6(t) = 2 (a. The latent heat of fusion is A. with sharp transitions in the temperature at the edges of the heated section. Determine the temperature in the central part of the heated section (i. I ) in terms of y. = TF. not too close to x = k L ) .88). concentration profile is . Heat transfer from a heated section of a wall. a .+? *. Integral Analysis of Diffusion and Reaction with Saturable Kinetics An enzymecatalyzed reaction with kinetics as in Problem 38 is being carried out at steady state in a liquid film of thickness L. T.ties (kL and k . (d) Use the interfacial energy balance to show that y is given implicitly by e . (c) Use the results of part (b) and the requirement that T= TF at the solidIiquid interface to show that SK tlR. (a) During the time period 0< t < tL when the concentration change at x = 0 has not yet penetrated over the entire thickness L of the membrane.
ignoring any initial transients). 'hmdent Diffusion in a Porous Medium with Reversible Adsorption Assume that species A is able to diffuse within some porous material. the solute concentration and the flux at the boundary are found to be (for y>O) where the concentrations C . 7).. and the concentrations at the boundaries are constant.5. Note: The transformation used here is given in Kamke (1943).. Write the differential equations and boundary conditions governing the first two terms of the expansion. t) = 0 . remains to be determined. T ~ porous F medium. which occupies the space initially contains no A or S. (b) Show that this may be approached as a regular perturbation problem involving the small parameter E . This difference in apparent diffusivity can be of great importance in systems designed to promote adsorption. (b) With Co(t) as stated in part (a) and rl = x l [ 2 ( ~ t ) ' ~ the ] . determjne tL. 318. U. 0) For t >tL.= 0. t) = Co and Ci(L. (a) Show that this approach will work only if Co(t)=atr. and p = 1/4.rp= CslKCo. Letting O = CiICoand X=x/L. where K>>1. so that D. (b) Show that for large 7. and its solutions (Whittaker functions) are expressible in terms of the confluent hypergeometric functions M and U (Abrarnowitz and Stegun. as usual. a pseudoequiIibrium is reached where 9s yc ( T h i s is not a true equilibrium because the adsorption rate is not zero. l ~ . K = [y+(1/4)]. 505). flansport in a Membrane with Oscillatory Flow A membrane of thickness L in contact with aqueous solutions is subjected to a pulsatile pressure on one side.e. Nt). governing equation becomes The solute concentration in the membrane is Ci(x. 320. Oo(X. the effective diffusivity of A is DA. C . but with a partition coefficient of unity (K= 1) and the boundary concentration a given function of time. identify the best choice for defining a scaled. Co(r). = Co at x = 0. where 0.. The penetration depth. Starting at t = 0. where a and y are constants. are based on total volume (fluid plus solid). 7.. and o are constants. r). ) ' ~ K = k.)? Use this assumed profile to calculate t. t). and C. dimensionless variables are 8 = CAICo. p.) and a(. The adsorbed .lk. W~th the porous material modeled as a homogeneous medium. where the flux is given by where r) = xlg(t) as : before. when changes in 8 are relatively slow. Suitable where .. (c) Determine the first term in the expansion. 319. such that CLO. and ~ = x ( k .Problems 127 x>O.. except that the apparent diffusivity is reduced from DA to D. = U(t)= Uo+ U. Let S denote the adsorbed form of A. one could assume a solution of the form v .)What is a sufficiently large value of f? Show that the functional form of 0 in this case is the same as for no adsorption. species is assumed to be immobile. where it is reversibly adsorbed on the solid phase.. a reasonable approximation to the concentration profile is Remaining to be determined is the function a(t). For consistency with part (a) and with the steadystate solution. T = tk. The rate of adsorption is given by where z=$. Consider convective and diffusive transport of some dilute solute i.(t) as a timedependent concentration scale. 1970. That is. leading to timeperiodic flow across the membrane. What is realistic about this assumed concentration profile and what is not? If the time for the concentration change to penetrete the full thickness of the membrane is defined by 6(tL)=L. dimen yields the differential equation sionless time. so that Show that the transformation It is desired to investigate solute transport after many pressure pulses have occurred (i. 4 0 ) = 0. One might attempt a similarity solution by using C. The transmembrane velocity is given by (a) Use 'an orderofmagnitude analysis to show that for short contact times the concentration profile for A is approximately the same as if there were no adsorption. The differential equation for w(z) is called Whittaker's equation.The pressure pulses are relatively slow. (a) Mthe various possible time scales.l(I + K). What is a sufficiently small value of T? Obtain expressions for 0 and q valid for short times. what must be the values of a(t. state the equations governing @(X. Similarity Solutions for TimeDependent Boundary Concentrations Consider a transient diffusion problem like that in Section 3.sin mot. Using this information.
Assume that the C10. how must the above results be modified? Derive the differential equation which governs 6 ( t ) for this case. r 2 ) = & exp(r2) eerfc 7. A. Additional quantities are as follows: S. 'This problem was suggested by K.] (c) Use the results of parts (a) and (b) to solve for S(t). That is.. and k. k. (d) If k2 is small but not identically zero. give an orderofmagnitude estimate for the amount of time which must elapse before the concept of a pseudosteady reaction front at x= S ( t ) becomes valid. (e) Obtain orderofmagnitude estimates of the time required to reach a steady state for D a c c l and Da>>l. which constitutes about 5% of the pulp. C10. Are the results consistent with the scales estimated in part (b)? (d) Repeat parts (b) and (c) for Da>> 1 . and k. species A is synthesized by the cells and enters the extracellular medium at x = O with a constant flux No. 255). irreversible reaction in the f i l m and bulk solution. Solution CA= C A O ' % '% ' '* ' '\ Iv ' Figure P322. As C102 diffuses in and reacts. concentration at x = O remains constant at CAD. and (e) Of course.respectively. e. Smith. The C102 reacts rapidly and irreversibly with lignin. the reactions during the bleaching process may be written as 322. 323. A c ~ .. is controlled by the interfacial balance from part (a). . The kinetics of the bleaching process may be studied under simplified conditions by exposing a waterfilled mat of pulp fibers to a dilute aqueous solution of C102. 321. Throughout the liquid. assuming that k. including the equations which govern the where A is ClO.The large magnitude of k. P322. Concentrations in a StagnantFilm Model of a Bioreactor The liquid phase in a stirredbatch reactor containing a cell suspension is to be modeled as a stagnant film of thickness 6 in contact with a wellmixed bulk solution. The film is assumed to be thin. B is unbleached lignin. Thus. The scales for C. and therefore the process time scale. also undergoes a slow. (c) Calculate the steadystate concentrations in the film and bulk solution for Da<< I .and C. Derive an qrderofmagnitude estimate for the thickness (() of the reaction zone. the volume of the stagnant film. Species A enters the extracellular solution with a constant flux at x=O and is consumed by a firstorder.+k2 =0. t .APPROXIMATION TECHNIQUES Problems 129 where r(x) is the gamma function (Abrarnowip and Stegun. the solutionmat interface is at x = 0. 1! I( I I 3 (a) Derive the conservation statement which applies at x = 6(t). because it is a component of the pulp fibers. solute concentration in the bulk solution as well as in the stagnant film. Also note that whereas Da and s are both small. and (b) For the conditions of part (a). . such that E<< 1. and EVGSS. [ H i n t :Assume that the overall rate. estimate the magnitudes of the concentration scales and then calculate the steadystate concentrations. < Use orderofmagnitude reasoning to estimate the concentration scales. For y=O the concentration at the boundary is constant. Bleaching of pulp fibers. is not truly infinite. no A is present initially. P represents products of the C10. and these results reduce to those derived in Section 3. their ratio may be large or small. Stagnantfilm model of a bioreactor. and are denoted by c:.. Notice that for y = the Bux is independent of time. the remainder of the pulp (predominantly cellulose) is inert with respect to C10. decomposition. C is bleached lignin. In addition to its reaction with lignin. r a = 6. 6 1 I CIO. the location of that boundary being denoted by x = S(t). p. ihis is seen by noting that U(&. <1. 4. (b) Consider the steady state with Da = ~ L ~ / D . and the thickness of the mat is taken to be infinite. as shown in Fig. Distinguish between situations where Da>>e and Da<<c. CS the volume of the bulk solution. the total surface area at x = 0 or x = S . P321. where it reacts to form species B according to The reaction rate is given by Figure P321.that the unbleached lignin concentration at x>&t) is . the boundary between brown (unbleached) and white (bleached) pulp moves away from the interface. are the homogeneous reaction rate constants.. spontaneous decomposition. (xv 0 I volume = V . is what maintains a visually sharp boundary. I I 1 Bulk Solution C*(V I I volume = EV = SS I I c. 1970. and that CAo<<Cm Note that. ACA. " Film . DifTusion and Reaction in the Bleaching of Wood pulp* Chlorine dioxide (C102) is commonly used to bleach the aqueous suspension of wood pulp which is produced by the digester in the h a f t process. (a) Formulate the complete transient problem. Absorption with a Nearly Irreversible Reaction Assume that gaseous species A is absorbed by a stagnant liquid.T(l)= 1.5 for constant Co and K = 1. A undergoes an irreversible reaction with firstorder rate constant k. lignin constant at C is immobile. (f) For large but finite k. Beginning at t=O. show that diffusion in the region O<x<S(t) can be modeled as pseudosteady. Note that the scales for CAand for the spatial variation in CA (written as ACA)are not necessarily equal. As shown in Fig.
despite the changing crystal radius. The system is at steady state and the liquid layer. (c) State the orderofmagnitudecriteria which must be satisfied for the pseudosteady analysis to be valid.C< 1. N. (Hint: Expand both of the concentration ' variables as regular perturbation series. but not quite to the extent that the &action can be considered irreversible. so that the crystal immediately bcgins to melt. with dimensionless concentrations defined as O(q)= CAICoand T ( q ) = CB/Co. 'This problem was suggested by K. = D. 4 is determined later [see part (d)]. assume equal diffusivities. R(r). and the gas occupies the space z>0. It will be of the The similarity solution for this problem was given originally by Arnold (1944).(O). which occupies the space x>O. 325 Melting of a Small Crystal* A small. (b) Assuming that the temperature field is pseudosteady. (a) State the differential equations and boundary conditions governing the liquidphase concentrations. UnsteadyState Evaporation A pure liquid composed of species A is brought into contact with a stagnant gas which initially contains only species B.A. =f (4). show that . The melting temperam T . For simplicity. so that the total molar concentration in the gas is constant. The system is at constant temperature and pressure and the gas mixture is ideal.where the equilibrium constant K has units of inverse concentration.5. (d) Derive an expression which could be used to compute C$ from xA. is governed by (a) Show that v = 0 throughout the liquid. implicit form. DA= D.(z. What is the proper length scale for converting x to the dimensionless coordinate q? What is needed for the liquid thickness to be considered infinite? (b) Use perturbation methods to calculate the flux of A into the liquid f i l m . . Assume that the bulk liquid is stagnant and that the solid and liquid densities are equal. The unknown constant (c) Determine O(77). derive expressions for R(t) and for the time t.) 324. and initial temperature To is placed in a large volume of the same liquid. Determine the first two tenns in the expansion for the flux. where Q(q)=xA/xA0. The reaction equilibrium strongly favors B.cT_. t). The latent heat for melting (per unit mass) is denoted as 1.. x . is effectively infinite in thickness. (b) Using a similarity variable (q) analogous to that in Section 3. is such that To<T. The gasliquid interface is maintained at z=0. required for the crystal to melt completely. Discuss the effect of the reaction reversibility on the flux. Srnilh. B is insoluble in the liquid and therefore remains in the gas phase. x. leading to a mixture of A and B in the gas. which is at temperature T_. spherical crystal of initial rad~us R. The liquidphase concentrations at the gasliquid interface ( x = 0) are C A (0)= Co and CB(0) =0 . for E=KC. The mole fraction of A in the gas at z = 0 is xAo. (a) Show that the mole fraction of A in the gas. Evaporation of the liquid begins at t = 0.
which is solved by standard methods. The FFI' method is one of many analytical or numerical techniques in which exact or approximate solutions to partial differential equations are found by expanding the solution in terms of a set of known functions. = at D. (2. It is mentioned in Carslaw and Jaeger (1959) and Butkov (1968). +Rvi. (2. Whereas the FFT method requires that at least one spatial dimension be finite. and permits a more direct attack on many problems. Then. the reader should not confuse this method with fast Fourier transform methods. Eq. The FFT method is more flexible. the partial differential equations in such problems are the simplest ones we will encounter. a precise definition of linearity is needed.. however.g. but only briefly.V~C. 4. In addition to their many applications. weighted residual methods). The methods presented here were chosen because of their broad applicability and because exposure to them also provides a helpful foundation for understanding modern numerical techniques for solving partial differential equations (e. The focus on conduction and diffusion in solids and stagnant fluids is continued from Chapter 3. The differential operator is said to be linear if One or the other of these equations is the starting point for almost all problems in this chapter. Green's functions are used most simply in problems with unbounded domains. but prior exposure to that method is unnecessary for what is presented here. This transform methodology was introduced to chemical engineering by N. With v = 0 and constant thennophysical properties. To clarify what is meant by that statement. Eq. and this chapter is concerned with methods which apply to this broad class of problems. Applying the FFI' method to partial differential "The use of finite Fourier transforms in the solution of conduction and diffusion problems has received little attention in standard texts. and Morse and Feshbach (1953).43) for conservation of energy reduces to equations like Eqs. a wide variety of partial differential equations can be represented as For species i in a dilute solution. Many problems in transport involve linear partial differential equations..65) gives ac.1 INTRODUCTION The preceding chapters were concerned mainly with the formulation of transport models. Carslaw and Jaeger (1959). . are also abbreviated as FFT. A. Most of this chapter is devoted to the finite Fourier transform (FFT) method.12) reduces the number of spatial variables until only a twopoint boundaryvalue problem or initialvalue problem remains. Morse and Feshbach (1953) and Carslaw and Jaeger (1959). Brown. Amundson. again with v = 0. Widely used in certain areas of engineering and physical science. and then determining the unknown coefficients in the expansion. Butkov (1968). for example. (4. Useful compilations of the solutions to many conduction and diffusion problems are provided by Carslaw and Jaeger (1959) and Crank (1975). R. called basis functions. The reader who is familiar with separation of variables is in a position to notice these improvements.11) and (4. (4. the two methods are complementary. which employs "pointsource" so1utions of the differential equations (Green's functions).2 FUNDAMENTALS OF THE FINITE FOURIER TRANSFORM (FFT) METHOD Linearity The FIT method is applicable to linear boundaryvalue problems in domains where at feast one of the spatial dimensions is finite. but they are approximate andor fairly specialized. see Arpaci (1966). The FFT method is basically equivalent to the technique of separation o f variables discussed in many traditional texts on mathematical methods in physics and engineering.12) where S(r) is a specified function of position. t) be the field variable (e. Thus. unfortunately. while teaching at the University of Minnesota. those are numerical methods for computing Fourier transforms of discretely sampled data. Incidentally. and let 3 be a differential operator which contains one or more spatial derivatives and perhaps also a time derivative. Some solution methods were introduced in Chapter 3. who learned it as a graduate student at Minnesota in the 1970s. making them ideal for illustrating unfamiliar methods. For discussions of other classical techniques see. The author was exposed to this approach by R.g. Space limitations preclude a discussion of the many other analytical methods applicable to linear problems.' After the FIT method is discussed in some detail. Let O = O(r. for example. temperature or concentration). The "ordinary" Fourier transform applies to an infinite domain. which.Fundamentals o f m n i t e Fouriei Transform Method Chapter 4 r SOLUTION METHODS FOR CONDUCTION AND DIFFUSION PROBLEMS 4. there is an introduction to another approach for solving linear problems. Churchill (1963).
(4.12) both reduce to Zaplace's equation. the corresponding operator is g= D. Eq. Consider. that is.) (4.)@ = h. (4. by definition. (4. Any differential operator containing products of derivatives will also be nonlinear. We conclude that t . and let I' t Because.= bC. T(r.1Z). (4. where 8 itself is a specified function of position r.25a). 41. or coordinate su$ices. . As an example. Indeed. There are only three types of linear boundary conditions for the secondorder differential equations with which we are concerned. For a secondorder reaction. which may also be written in terms of differential operators.k2Ci.C AND O DDIFFUSlON PROBN where a. and @ =$ . (4.21 Uniqueness Proof for Laplace's Equation For steady conduction or diffusion without volumetric source terms. t) and O. with R. and a. where the normal component of the gradient is specified. A boundaryvalue problem consists of a differential equation together with certain boundary conditions. so that the operator is seen to be nonlinear. (4. (A. Wodimensional regions represented by rectangular coordinates. .) (Dirichlet) (Neumann) (4. VO = g(r. It follows that 0 is also a solution to Laplace's equation and +2.2Z).11) and (4. the solution to Eq. For the conduction and diffusion problems governed by Eqs. (4. linearity demands that the source t e r m (Hv or Rvi) either must be independent of the field variable If any part of the source term is independent of or must be a linear function of T or Ci.kl.(r. (4. that the alternative expressions yield the same values of T(r. t) are any functions [not necessarily soIutions of Eq. solid rod for which it is desired to determine the  Uniqueness A very imponant characteristic of linear boundaryvalue problems is that.11) or (4.54)] and using V2$=0. for example. with minor exceptions. we assume that the variable t. T or Ci. I. (r. Putting Eq. t) once all summations and other opera Figure 41. . 0 =0 on S. t ) cannot be written in more than one fom..2Sb).abt. This requires that O be constant throughout K However. the dependence of % on the field variable (Ci) makes it impossible to satisfy Eq.) A linear boundaryvalue problem is one in which all of the differential operators are linear.2Sa) The integrand here cannot be negative. (4.. Suppose now that there are two different solutions to Eq. It is easily confirmed that this differential operator is linear.25b) n . say.2Sc) is a "mixed" or "Robin" condition. A similar approach yields proofs for the Neumann and Robin boundary conditions.V.26) must be unique. on a bounding surface. Thus.21)1. L n such problems the solution is unique only to within an additive constant. consider Eq. the other restriction of the FFT method is that the boundaries must correspond to constant values of the coordinates. an initial condition is also needed. the three regions depicted in Fig. and Eq. are any constants and O.21) gives Setting 4= t. b . Example 4. As an example of a uniqueness proof for the solution of a linear partial differential equation. we obtain so that 2 = DiVZ .(r. this chapter contains many examples of problems where the solutions can be written as two or more infinite series involving different sets of elementary functions.29) is satisfied only if I VO I = 0 everywhere.25c) Equation (4.21) will be nonzero. .29) =f (r. In this case. (4. (For timedependent problems.slat and S = 0.b in Green's first identity [Eq. (4. Eqs. and +2 cannot differ. . v 1i: 1.12) for the case of an irreversible. Each might represent the cross section of a long. 8 can be constant in V and zero on S only if 0 =0 everywhere. once a solution is found by any method it must be the only solution. is called an "essential" or "Dirichlet" condition. . so that Eq. the solutions are unique.(r. (4.b(r) is subject to Laplace's equation in a volume V with a nonhomogeneous Dirichlet condition on the bounding surface S: where kl is the homogeneous reaction rate constant.26). denoted by #. then the function S in Eq.23) in the form of Eq. The exceptions to uniqueness are steadystate problems intions have been volving Neumann (gradient) conditions on all boundaries. however. n . is termed a "natural'' or "Neumann" condition. ?he uniqueness property does require. This does not mean that a solution for. They are @ .v@ + h. (4.' (4.$2.) (Robin) (4. firstorder reaction. : Domains Aside from the requirement of linearity. the righthand side vanishes and 11~01'dV= 0.
see Chapters 2 and 3 for examples of problem formulation and nondimen Notice that each term in Eq. The limits of integration correspond to the range of x in the problem. What constitutes an appropriate set of basis functions is described in Section 4.(y) for all n is sufficient to solve the problem. (4. or h. then 0 =0 is indeed the solution for the problem. For now. or spherical coordinates.(y). is to use cylindrical coordinates.(y) is obtained by transforming the original set of equations as in Eq. (For simplicity." Equation (4.(x) specified by Eq. ' t o=f.213) is a product of two functions. in which the dimens~onlesscoordinates are x and y and the dimensionless temperature is Y). then the boundary corresponds to a constant value of r. and the results must be added to construct the desired overall solution.21) is said to be homogeneous in the dependent variable @ if S = 0. for transient problems. if the x boundaries were at locations other than x = 0 and x = 1 . as required. curvilinear coordinate systems have been developed for numerous geometries (see Section A. Equation (4.. (4.(x) 1 Homogeneity An additional consideration is whether a differential equation or boundary condition is homogeneous or nonhomogeneous. (4. of course. ( y ) is referred to simply as the "transformed temperature. and the result integrated over the finite interval 0 5 x 5 1. the initial condition) are all linear and homogeneous. The functions @. twodimensional temperature field. The number of subsidiary problems which must be solved equals the number of nonhomogeneities.22 Steady Conduction in a Square Rod Consider the problem depicted in Fig.. & where @ . (4. This feature is common to dl FFT and separation of variables solutions.214) F R Method The objective of the introductory FIT example which follows is to illustrate the mechanics of the method. Applying the FFT to both sides of Eq. Thus. namely C. 0 = 0. It will be shown that O. 137 steadystate.213). Whether or not this is a solution for the boundaryvalue problem will depend. y). y) is a2@ += a 2 0 ax" af 0. The best approach here. Steady heat conduction in a square rod with gpecified surface temperatures. In the rectangular domain of case (a) the boundaries each correspond to constant x or constant y. n = 1.25) result when the functions on the righthand sides are zero (i.21 1). Enm~le 4.(x)=+ : sin n m . (4. the main thing to notice is that each of the basis functions vanishes at x = O and x = 1. = 0 ) . two or more subsidiary problems must be solved.(y) is identical to the expansion coefficient C. Thus. To complete the solution it is necessary to determine the coefficients which multiply the basis functions in Eq. (4. (4. The problems in this book involve only rectangular. the convergence properties of the series solutions. This restriction is not as severe as it first seems. of course." The reader who is familiar with the separation of variables technique for solving partial differential equations will recall that it is directly applicable only to problems where the differential equation is homogeneous.215). which looks extraneous.210) gives . (4. because orthogonal. but neither x nor y is constant on the boundary. We start by assuming that the solution can be written in a series expansion as @.215) defines an integral transform in which the original function is multiplied by a basis function. Eq. and there is only one nonhomogeneous boundary (or initial) condition. on the boundary conditions. nothing happens.214).23) is an example of a linear. In other words. If the partial differential equation and auxiliary conditions (the boundary conditions and. homogeneous equation.7)..) The temperature field is nonuniform as a consequence of the specified temperature variations along the boundaries at y = 0 and y = 1.136 S O L U ~ ~F M OR CONDUCTION ~ T H ~ AND S DJFFUSlON PROBLEMS Fundamentals d the Finite Fourier ~ransforrn (FFT) Method Figure 42.. sionalization.e. and certain other issues relating to how and why the method works. In casq (b) the location of the circular boundary is still readily described using rectangular coordinates. has the effect of normalizing the basis functions. The problem statement for @(x. (4. There is no volumetric energysource term. g = 0. the FFT approach is more flexible. the definition of the transform would be modified accordingly. 2. so that Eq.213) will be valid are discussed in Section 4. and the conditions under whch an expansion such as Eq. . cylindrical. 42. 1' 9 (3. If the differential equation is nonhomogeneous andlor there is more than one nonhomogeneous boundary condition.3. many other important considerations are deferred until later in the chapter. the requirement that the boundaries be coordinate surfaces restricts us to certain regular shapes. The homogeneous forms of the boundary conditions in Eq.213) satisfies the homogeneous boundary conditions in Eq. the nonhomogeneous terms are sometimes called "forcing functions. Such topics include the identification of suitable sets of basis functions. expressed as T(x. if there are no nonhomogeneities.214) are the basisfunctions. so that determining O.4. The $ factor in Eq.(y). In case (c) there is no obvious coordinate system which would describe the boundary. For this problem the finite Fourier transform (FFT) of the temperature is defined as 1' I I d $5 j 'f . f = 0.. the advantage in doing this will be evident when we are finished. namely r = R. (4. most examples in this chapter begln with a problem already stated in dimensionless form.A partial differential equation which is homogeneous always has as one possible solution the trivial solution. A differential equation governing @. (4. as done in several of the examples of Chapters 2 and 3. each involving only one spatial variable. (4. As will be seen.
2I).This is done by first fTansfoming Eq. This situation contrasts with that encountered when using other integral transforms (e. . 3 BASIS FUNCTIONS AS SOLUTIONS : T O EIGENVALUE PROBLEMS Sets of functions which can be used as basis functions in the FIT method are the solutions to certain types of eigenvalue problems.225) where Y xis a secondorder differential operator involving x. (4.221) and (4. It is worth noting that obtaining a solution for nonhomogeneous boundary conditions at both y = 0 and y= 1 was not significantly more difficult than if there had been only one nonhomogeneous boundary condition. cosh na(cy)].225) would have equaled &. lems of interest here. sinh nvy (4.(y) = Cn(y)/2.212).I2 instead of am.223) sinh n.(y) to @(x.(a@/dx).. the integral in Eq. where c is any constant.. Eqs.3I). For the probi. of d21dr2in Xx must be positive. (4.(y) were identical. vanishes because of the homogeneous boundary conditions satisfied by 0.Notice that the factor included in the definition of @. (4.220) include [exp(n. (0)= @. sinh nrr(ly)+Bn sinh nm n=l sinh n r y I sin nmx. ' L .Basis Functions as Solutions to Eigenvalue Problems Using Eq.(y) and C..213) to give @. giving the less convenient result that O. and then it discusses the properties of eigenfunctions which make them suitable basis functions.. 1 4 .224) and (4.(y)... (4.. Eq. = 1 for m = n. the values of A (and all other constants) are real.216) is rearranged using two integrations by parts to give Y)=G z [A. but inverting the transform to recover the final solution may be extremely difficult. (4. the Laplace transform) to solve partial differential equations. This completes the information required to calculate an(y). (4. Of concern here are functions O(x) which satisfy differential equations of the form + .g. Thus.218). Eq. (4.The overall solution is OCX. (4. (4. and [sinh nny.225) indicate that Bm(y) = Cm(y). cosh nn YI. involving O(d@. it follows from the definition of the basis functions that d2@. (4. where A. Concerning the final term in Eq..215). and B. To maintain the correct algebraic signs in Eq. in this case going from @. one finds Y xby selecting only those terms in 3 which involve x. If the basis functions had not been normalized in this manner.( n ~ ) & ~ sin n m = (nn)' 8. The transformed problem was solved using elementary methods to obtain O.(4.218) becomes : i i / 1 These results are combined to give the differential equation for B.I&). the second term is rewritten in terms of the transformed temperature as where .218). the FFT method first involved transforming the original problem in x and y to one involving y only.220) requires two boundary conditions. With the partial differential equation written in the form of Eq. which are obtained by transforming the original boundary conditions involving y.sry). are constants. (4. Given that S. vanishes because cP..222). (4. termed characteristic values or eigenvalues. that involving @..rr What remains is to establish the relationship between @.l #= . The integral involving the product of the basis functions is evaluated as [@. exp(nvy)]. Lineatly independent solutions of Eq.y). /i 2 In summary. The first term in Eq.211). Choosing the last of these pairs (with c = 1) and satisfying Eqs. and h is a constant which can have only certain values. The first of the "boundary terms" in Eq..(y). = 0% for rn # n and S. The operator Zx 3: that corresponds to a given conduction or diffusion problem usually may be identified simply by inspection of the conservation equation. sinh nn(cy)].(~)@~(x) d*=2 sin rnrx sin n m dr=6. Accordingly. it was necessary only to recognize that O.(x) was defined such that a.. [cash nlry.(Y) = Eigenvalue Problems Eigenvalue problems arise when certain homogeneous differential equations are subject only to homogeneous boundary conditions.(y) and Cn(y). With the FJT method the inversion step.(I)= 0. is automatic! Equation (4.(x) was needed to make On(y) and Cn(y) identical. (4. where deriving and solving the transformed problem may be straightforward. the coefficient . a is the Kronecker delta. . Eq.. the transformed temperature is found to be A. This section begins by reviewing the formulation and solution of eigenvalue problems. sinh nn(1 y) B. These solutions are called eigenfunctions. (4.. [sinh nny.The second boundary term. To complete the overall series solution.
not included in Eq. any conduction or diffusion problem in rectangular coordinates. where a and b are constants. the energy conservation equatio? is A set of homogeneous equations such as that give.(x) = 0.38) always has the trivial solution a = b = 0. the differential equation for the eigenvalue problem is then where n is any integer. Setting A = O in Eq. For there to be nontrivial solutions.=(l)".(x) can be dropped from the set of eigenfunctions.315). because it is just the trivial solution @. which leads to the eigenvalues.32 Eigenvalue Problem with One Neumann and One Dirichlet Condition The problem to be considered is now The general solution is again Eq. (4. certain other allowable types of boundary conditions are discussed in Section 4.39) Equation (4. (4. These are given by The operator 2x obtained from Eq. is any constant. (4.38). sin n m . constant value for @ will satisfy @'(O) = 0 but not @(I) = 0. Checking the boundary conditions in Eq. The special case of A = O always requires separate consideration. (4. Applying the boundary conditions.It is shown later that the eigenfunctions corresponding to n<O are also not needed.35a). is A=O. where the constants now must satisfy Equation (4. Notice that <P. while the value of a is undetermined. In the following three examples. (4. (4. they may be any combination of the homogeneous forms of the three types of conditions given by Eq. which restricts the possible values of A .316) does in fact represent the complete set of eigenfunctions.37) yields @ = b. the characteristic equation is sin A = 0. (4. Including the possibility of a volumetric heat source. (4. This excludes the possibility that A = 0. these constants must satisfy a pair of algebraic equations. . Indeed.Referring to Eq. involving a stationary medium with constant thermophysical properties. it is seen that a nonzero. yields Eq.313) implies that a =O. (4. (4.311. Thus. it is seen from either of the algebraic equations that b = O for all n. The eigenfunctions are then @. the boundary conditions for Eq. is any constant.33) is just 2x = d21dr2. corresponding to the zeros of the function sin A . This requirement leads to the characteristic equation. (4.25).34). then the probIem statement is A cos A = 0. (4. The resulting eigenfunctions are The differential equation has the general solution @ = asin Ax+b cos Ax.32) is rewritten in the form of Eq.313) yields the characteristic equation.(x) = a.38) and noting that cos A.37). From Eq. (4. The value of b is undetermined. (4.6. I). so that Eq.21) as Accordingly. Example 4.1 1) The same eigenvalue equation is obtained for a transient conduction problem involving T(x.312). In general. (4. (4. (4.314) where n is any integer. Another possible eigenvalue. solutions are derived for eigenvalue problems of the type just discussed. h can only have certain discrete values A. Returning to Eq. The boundary conditions for eigenvalue problems must be of certain types. in that they are redundant.For example. In other words. the determinant of the coefficient matrix must be zero.. which can be written as a single matrix equation. Ignoring for the moment the possibility that A = O and noting that sin A. y). In each case x is a dimensionless variable defined so that the boundaries are at x = 0 and x = 1. Example 4. with temperature T(x. where primes are used to denote differentiation with respect to x and A is a constant. =cos(nn)=(l)". the solution corresponding to A = 0 is simply a constant. consider steadystate heat conduction in two dimensions.3. where b.31 Eigenvalue Problem with Two Dirichlet Conditions If both boundary conditions are of the form of Eq. (4.34) are usually some combination of where a. by Eq. (4. either algebraic relation in Eq.
.
4.34) and boundary conditions chosen from Eq.Although Example 4.(x) form a complete orthonormal sequence. In each of these cases there are simple. As mentioned below Eq. Because the eigenfunctions with ncO differ only in sign from those with n>O. note the crucial role that orthogonality played in relating Cn(y) to O. it left some fundamental questions unanswered. (4. The results presented here apply not just to series constructed using the trigonometric basis functions described in Section 4. (4.sin(x). Notice that in case IV one of the eigenfunctions is a constant.6. For what kinds of functions is such a series valid? As more and more terms are evaluated. but also to expansions employing any other basis functions derived from the general StumLiouville problem described in Section 4. (4. y). Notice also that in Eq.4 REPRESENTATION OF AN ARBITRARY FUNCTION USING ORTHONORMAL FUNCTIONS A cornerstone of the M method (and the separation of variables method) is the assumption that an arbitrary function.320) involves only sines and cosines. . can be represented as a series expansion involving some set of orthogonal functions. O(x.34)." "FourierBessel.214) are seen now to be normalized and linearly independent versions of the eigenfunctions derived in Example 4. it is sufficient that f (x) be piecewise continuous in that interval [see. f (x). It is helpful to bear in mind that the physical 1. el. because it gives a nontrivial solution. each consisting of a basis function @.7) and Legendre polynomials (Section 4.) Any such expansion involving orthonormal functions is referred to here as a Fourier series. the orthonormal sequences obtained from four common eigenvalue problems are given in Table 41.3." or "FourierLegendre" are sometimes used. With this additional information about basis functions. When one or more Robin (mixed) conditions is present. are called spectral coeficients. n = 0 is already excluded because it corresponds to the trivial solution. for example. A piecewise continuous function is one which is continuous throughout the interval. and it also presents a general procedure for evaluating the coefficients in such a series. we choose to exclude the negative integers. (4. @. coupled with the fact that Eq.e.311). note that sin@)= .35). it was assumed in Eq. Similar reasoning applies to the eigenfunctions given by Eq. but both limits are finite. so that once again we exclude those for ncO. Generalized Fourier Series : To address the main issues. (4. discontinuites.8). This.319) there is a "mirror image" negative root.22 illustrated the mechanics of evaluating the coefficients for one such series. sometimes it will be convenient to label the first term with n = 0. Those include Bessel functions (Section 4. b] in x (i. it is necessary only to ensure linear independence. the differential equation of the eigenvalue problem. where the functions @. and the constant coefficients c. it is sufficient to consider arbitrary functions of a single variable..3ll).. It is desired to represent such a function using a series of the form %ble 41 Orthonormal Sequences of Functions from Certain Eigenvalue Problems in Rectangular Coordinatesa Case Boundary conditions Basis hrnctions ' "All of the functions shown satisfy Eq. for a s x s b). Finally. in what manner will the series converge to the function it is supposed to represent? This section addresses those questions. Accordingly. (4.219) it was important for the basis functions to satisfy Eq.225). (4. as in Example 4. In this case we must retain n = 0. (4. (4. (4.316). Considering the eigenfunctions given by Eq. or which exhibits a finite number of jump . (4.(y).320). Consequently.224) and (4. The basis functions defined by Eq. as discussed in Sec. With regard to the eigenfunctions in Eq.22. tion 4. note that cos(x) = cos(x). such as the temperature field in a heat transfer problem.33. using Eqs. To use a Fourier series to represent a function f ( x ) over some closed interval [a. adjectives such as "Fouriersine. (4. the reader may find it helpful to review Example 4. Thus. For each positive root of Eq. explicit expressions for the eigenvalues.(x) multiplied by a coefficient Cn(y). are selfadjoint.(x) = 1 1 4 . processes of conduction or diffusion (and the second derivatives which are the mathe . At a jump discontinuity the function has different limits evaluated from the left and right. (4. (4. including negative as well as positive values of n would result in a set of functions which is not linearly independent. (The starting value of n has no special significance. Churchill (1963)l. cos(vxf2). could be expressed as the sum of an infinite number of terms. the eigenfunctions for n<O duplicate those for n>O.213) that a certain twodimensional temperature field.3. When reference is made to expansions involving specific basis functions. involving Eq.Representation of an Arbitrary Function Using Orthonormal Punctiofis cussed in this section. the eigenvalues must be found numerically. To ob$in an orthogonal set of basis functions from a set of eigenfunctions like that generated in any of the three examples in this section. indicates that we should discard all eigenfunctions corresponding to negative values of A. corresponding to A =O. Those eigenfunctions were useful because they satisfy hdmogeneous boundary conditions of the same type (Dirichlet) as the homogeneous boundary conditions in the heat transfer example. For convenient reference.31.34) in the interval 10. These four problems include all cornbinations of Dirichlet and Neumann boundary conditions applied at x = 0 and x = t .
(4. each multiplied by a corresponding base vector. b) where the function is continuous. The main difference is that the number of base vectors needed in a physical problem is ordinarily just three. f(x+) and f (x) represent the right. evaluated at x. (4. The spectral coefficients correspond to the components and the basis functions are analogous to the base vectors. A heuristic derivation of Eq. the constants An and B. the series converges to f(x). then N+ lim f.(x) =f(x+) +f(x) 2 :. (4.43) isolates the corresponding spectral coefficient. for a<x<b). but not at the endpoints themselves. immediately from Eqs. and then taking the square root to evaluate the magnitude of the vector.22. the basis functions in Eq. (4. 1976.41) are orthonormal. The scalar product selects the component of the second vector corresponding to the particular base vector.43) is like forming the scalar or dot product of a base vector with some other vector. y). The overall or average convergence properties of a Fourier series are best described using the concept of a norm. Accordingly. However.). defined by Eqs.. its basis in functional analysis is beyond the scope of this book.45) restates the definition of the FFT of @(x.42) is the Fourier Series Theorem (Naylor and Sell. (4. (4. but such discontinuities ordinarily consist of finite jumps. (while omitting e. an expansion for a function @(x.matical embodiments of those processes) tend to smooth the temperature or concentration profile. It is analogous to forming the scalar (dot) product of a vector with itself. it is impossible to accurately represent an arbitrary function using an incomplete set of basis functions. In Eq. or Equation (4. (4.(4.. usually ensuring that the field variable and its first derivatives are continuous within amgivenphase. Thus. The fundamental justification for Eq. written as Ik(x) ll. at a jump discontinuity. (4. Greenberg. A more general classification of functions which can be represented using Eq. For Fourier series it is found that . (4. Let f. (4. at each point in the open interval (a. It is seen that the transformed temperature.and lefthand limits of the function. given previously by Eq. (4. For example.221) and (4.e. is defined as ' : which involves the inner product of the function with itself. was used indirectly in Example 4. A subtle but important distinction is that the pointwise convergence described by Eq. Just as the base vectors for a given coordinate system are (usually) defined so that they are mutually orthogonal and normalized to unit magnitude. The inner product operation i n Eq.41) are given by Equation (4. and then using the orthonormality of the basis functions.41) and (4.321). is a spectral coefficient. The n o m of a function f(x).41) has been developed using linear operator theory [see Naylor and Sell (1982) and Ramkrishna and Arnundson (1985)].4l). (4. b) (i.42) is obtained by forming inner products of with both sides of Eq. whereas the number of basis functions in Eq.(x) represent the partial sum involving the first N terms in the series for f(x). without the inner product notation. A very readable introduction to the concept of orthogonal bases in finitedimensional (vector) space versus infinitedimensional (function) space is given by churchill (1963). the temperatures specified along two of the boundaries.41) is analogous to the more familiar reptesentation of an arbitrary vector as a sum of scalar components.(y). the strategy in the FFT method was to compute the spectral coefficients for the twodimensional temperature field. at all points in (a. b] and weighting function w(x). An and B.. whereas Eq. (4. which in turn depended on the spectral coefficients for certain boundary data (i. y) can be written as Calculation of Spectral Coefficients Assuming that a given function f(x) can be represented by a Fourier series. including those arbitrarily close to the endpoints. 1988).e. Thus. This inner product provides a measure of the magnitude of the function.44) is the same as the expansion for the temperature in Example 4. with respect to a specific interval [a.b] (Hildebrand. respectively. That is. by simply treating the additional variables as parameters.47). convergence is to the arithmetic mean of the limits. respectively. If that is true.215).41) is infinite.222) are the spectral coefficients for the functions fo(x) andfl(x). as defined by Eq. Discontinuities in the field variable may exist at internal interfaces or dong boundaries. (4.47) is guaranteed at all interior points. just as Eq. almost any conduction or diffusion problem will involve only piecewise continuous functions. The Fourier series representations of functions of two or more variables follow The extent to which these partial sums converge as N is increased may be described both in a local and in an average sense. The fact that there is an infinite number of basis functions in any set tempts one to think that it would not hurt to omit one. 1982). to give To discuss the convergence of Fourier series we return to functions of one variable. (4. An important result pertaining to local or pointwise convergence applies when both f (x) andf '(x) are piecewise continuous on the interval [a.). Convergence The inner product notation is used here. (4. the spectral coefficients in Eq.22.42). just as it is impossible to represent an ditrary threedimensional vector using only ex and e. It is noteworthy that Eq. O.42).
ld nw It follows that the desired sine series is 00 sin n g x nn 1 =2C [l (I)"]. any skepticism which the reader might have when first encountering an identity like Eq. and all harmonics conrribute. However. Although they were known to other prominent mathematicians. Lagrange. 43. "=I .412). Example 4. That situation is analogous to a vector which parallels one of the coordinate axes and which therefore has a single nonvanishing component. This illustrates how increasing the number of terms gives convergence in the mean and local convergence at interior points..choosing the coefficients in the Fourier series in the manner prescribed by Eq. An interesting result which is related to Eq. Lagrange was adamantly opposed to the idea that trigonometric series could be used to represent arbitrary functions. including points progressively closer to the ends of the interval.412).x are Again choosing the sines in case I of Table 2 ~ h contributions e in the remarkable paper submitted by Fourier in 1807 to the Institut de France were described in the first footnote of Chapter 1. because it would have been sufficient to set co= 1 and c .42 FourierSine Series form) = 1 x 41. I sin nrrx n * T T . for a fixed number of terms N. as is the case for any constant. after review by Laplace. This occurs whenever the function f(x) is symmetric about the midpoint of the interval. Although it may seem quite improbable that the righthand side equals unity. ~ Plots of this series are shown in Fig. What was not mentioned is that. This bkhavior is referred to as convergence in the mean (Churchill. when constructing a Fourier solution. In this case there is no symmetry about x = 3. which illustrate typical convergence behavior and other features of such series. with the interval [O. with biographical information. if an infinite series involving a set of basis functions has coefficients which do not exhibit this behavior. which concerns the spectral coefficients themselves.(4.42). no matter Example 4. f. 1963).: I case the series is able to give the exact value of zero at x = 1. a Fourier series containing a finite number of terms provides the best "least squares" approximation to the function f(x). 1963). (4.49) is that. (4. Following are some examples of Fourier series representations for simple functions. based on partial sums involving either 4 or 20 terms. This underscores the point that if f ( x ) happens to have the same functional form as one of the basis functions. because the basis functions vanish there and f(x) does not. the problem would have been exceptionally easy. The 1807 manuscript was rediscovered many Years after his death and. The exact function is also shown. I 972)! and the complete series is given as m 1x=2C. n=1 how many terms are used. this particular series fails at the endpoints. (4.In words.(x) is a much better approximation to f ( x ) = 1 than is f.42) has the effect of minimizing the norm of the error (Churchill. Monge. based on 4 or 20 terms in Eq. but in this . . the choice of basis funcseries for boundary or initial data as part of an tions is ordinarily dictated by other parts of the problem. ': . From Eq. but does not necessarily help at the endpoints themselves. the basis functions which must be used to expand some function f ( x ) are not necessarily those which yield the simplest Fourier series for that function. 44. The selection of examples was influenced by the fact that. and Lacroix. Fourier's results were not actually published until the appearance of hls book in 1822. X Figure 43. the coefficients for f (x) = 1 .41 FourierSine Series for f(x)=1 With the FIT method it is often necessary to represent a constant by a Fourier series. . the paper was rejected. the N+. a oneterm representation is exact and an infinite series is not required.(x). All of these examples employ trigonometric basis functions chosen from Table 4. (4. Suppose it is desired to represent f ( x ) = 1 using the sines given as case I in Table 41. The behavior near x = 0 is similar to that in the previous example. Although the other reviewers favored acceptance. published after a delay of 165 years (GrattanGuinness. Incidentally. Plots of this series are ahown in Fig. Thus. n+m norm of the error involved in approximating f(n) with fN(x) vanishes as Thus. = 0 for n>O. (4.412) is excusable. notice that only the oddnumbered "harmonics" (odd values of n) contribute to the sum in Eq. then it is not a legitimate Fourier series. Thus. 11. the coefficients are given by c . If the cosines in case IV of Table 41 had been chosen to represent f(x) = I. = ~sin n m &=* [I (I)"]. is that lim cn= 0. Another result. In other words. As expected. Fouriersine series representations of f ( x ) = 1.1. this result is ~ o r r e c t .
1976). to determine the average temperature of an object. Fouriercosine series representations of a step function.6 x 0 . It is sufficient for this that the function be :1 !: . then termwise differentiation of the Fourier series will yield a series which converges to Results from this series are plotted in Fig.493. to compute the heat flux from a representation of the temperature field. Convergence in the mean is not violated.. b) where f f ( x ) is continuous (Hildebrand. The exact function is also shown. This is called the Gibbs phenomenon.4 0. for example.. however. A computationally convenient representation of the series is obtained by using four sums involving a new index: Differentiation of a Fourier series may be needed. leaving E n Fig. These three '. An unfortunate property of Fourier series is that the amplitude of the overshoot near any discontinuity eventually fails to decay further as more and more terns are added. 43. (2) f'(x) is piecewise continuous in [a.47).. F i g w 44. Termwise integration of a Fourier series may be desired. 8 1 Figure 45. . Once again. but caution is needed because the resulting series does not always converge to the derivative. because the overshoots are confined to progressively thinner regions as the number of terms is increased. b]. and (3) f ( x ) satisfies the same (homogeneous) boundary conditions as @. 45 there is a distinct overshoot evident near the discontinuity at x=a. based on 8 or 40 terms in Eq. In the case of trigonometric Fourier series. . The exact function is also shown. for example.. A discontinuous function like this is handled by performing the integration for the inner product in a piecewise manner. 1963. termwise differentiation is sometimes valid even if condition (3) is not satisfied (Churchill. the arithmetic mean of the right. Similar overshoots are also seen in Figs.. 1976). Morse and Feshbach (1953). 6 . f r ( x ) at all points in (a.. 7. but they are not all necessary. Example 4. In this case the piece for 10.and lefthand limits at the discontinuity). b]. 5 . 45.(x). &I vanishes. Notice that Eq. With sines or cosines the overshoot reaches an asymptotic limit of about 9% of the size of the step. Fouriersine series representations of f(x)= 1 . the series evidently converges to f = i at x=a (i. based on 4 or 20 terms in Eq.x .2 0. the approximation fails at one of the boundaries ( x = I). (4. and 40 terms when the upper limit is 9. If. !: (1)f ( x ) is continuous in [a.e..417). (4. and 45 near the boundaries where pointwise convergence fails. for example. 1. 44. no matter how many terms are used. even for a series with as many as 40 terms. The actual computed values are fB(i)=0. Hildebrand.43 FourierCosine Series for a Step Function Consider the step function defined by which will be represented now using the cosines in case II of Table 41. The obvious approach is to differentiate each term. and the plus sign is for n = 2. Differentiation and Integration of Fourier Series where the minus sign applies for n =0. (4.464 and fa(&) =0. (4.Kepresentation of an Arbitrary FunctionVUsing Orthonormal Functions 151 0 0 0. 3.417) yields 8 tenns when the upper limit for m is 1. In accordance with Eq. conditions are sufficient for termwise differentiation.414). 4. see.
(4. we seek a solution of the form The first boundary term in Eq.510) are 0. For the general case depicted in Fig. there is no preferred "direction" for the expansion of @ ( x .(I) = B.510) and (4. are defined by Eqs.55) is evaluated as The other terms from the original differential equation. The remaining integral in Eq. choosing sinh nlry and sinh n v ( l y) as the independent solutions of the homoge. because 0.512) has the particular solution 8. the absence of homogeneous boundary conditions in the x Arection requires that one of the boundary terms in Eq. basis functions involving y would serve as well as ones involving x.(0) = @. which is a more complicated version of Example 4. H(x.5 FFT METHOD FOR PROBLEMS IN RECTANGULAR COORDINATES The examples in this section illustrate the application of the finite Fourier transform method to a variety of problems involving rectangular coordinates. (4. (4.(0) =A. (4. Equations (4. where the constants An and B.55) vanishes.. 46.51 Steady Conduction with Multiple .511) .51) is arbitrary. 1963). the transformed differential equation is i I It is seen that having one or more nonhomogeneous boundary conditions in the basis function coordinate (in this case.220)]. Nonetheless. @ = O at x = I .22. (4. tenn by term.22. no choice of basis functions will satisfy any of the boundary conditions exactly. Eq. nonzero values on all boundaries.152 SOLUTlON METHODS FOR CONDUCTlON A N D DIFFUSION PROBLEMS FFT Method for Problems in Rectangular Coordinates Applying the FFT to the first tenn we obtain. The energy conservation equation for this problem is written as (4. (4. The boundary values for 0.41). and H = O . . we choose as basis functions ones which satisfy homogeneous b o u n q conQtions of the same type (Dirichlet) as the nonhomogeneous conditions in the actual problem. (1) = 0. The dimensionless temperature has specified.512) remains nonhomogeneous because of the nonhomogeneous boundary condition at x=O. the second boundary term becomes Thus. Figure 46.513) are just the spectral coefficients for the I function f (x) = 1 (see Example 4.22. and there is a volumetric heat source term. Equation (4. = (2G)l (nrr). Using the ITT method. the general solution is . From case I in Table 41. ' neous equation.55) be retained..Nonhomogeneous Tenns Consider the steady.54).510) with Eq. so that the validity of termwise integration is rarely (if ever) an issue. (4. . The boundary conditions satisfied by Eq. Example 4. so that. with integration by parts. @ = 1 at y = O and y = l . are transformed in a straightforward manner as In the absence of any homogeneous boundary conditions. Included are transient as well as steadystate problems. Thus. (4.221) and (4. so that the coordinate preference expressed by Eq. and using the boundary temperatures specified at x = 0 and x = 1.511) become i & . (4. 4.. 46. we have : Collecting all of the terms. y). assume that 0 = 2 at x = O . Unlike the situation in Example 4. Evaluating the derivative of 0. @. the partial differential equation and all four boundary conditions are nonhomogeneous. because none of the conditions are homogeneous. x ) and/or a nonhomogeneous term in the original partial differential equation causes the differential equation for the transformed temperature to be nonhomogeneous [compare Eq. 153 piecewise continuous (Churchill. (4. '. Steady heat conduction in a square rod with specified surface temperatures and a volumetric heat source.222). As a special case.in Eq. y).Equation (4. nonzero values of these constants are the consequence of nonhomogeneous boundary conditions in the y coordinate. twodimensional heat conduction problem shown in Fig. As discussed in Example 4.. respectively.
.
For the temperature to bejnite at y =. we obtain The decaying exponentials indicate that for large y. the solution is rewritten more simply as m Example 4. Accordingly.x An. n = 1.5 is not necessarily valid. eventually requiring an impractjcal .53 'Ikansient Diffusion Approaching a Steady State We return to the membrane diffusion problem introduced in Section 3. (4.5I). There is a transient part which dec'ays exponentially with time. 0 = 0 at t =0.[Bi/(l +Bi)]x. In contrast to the situation for large times.= L ~ / D for small t the series in Eq. Accepting exp(h. In other words. 48. whereas the similarity result is valid at most for f 5 f2. there is no way to formulate an eigenvalue problem in t.526) and (4. The series in Eq.535) consists of two main parts. (4.x .(0)=0.533) is and the overall solution is then sin n n x @(x. the final form of the temperature field is @(x. It is sufficient to consider the most slowly decaying exponential. With reference to Fig. 3 ~ ~ where reach a steady state is of the same order of magnitude as the characteristic time for diffusion over (see Section 3. the temperature is a function of x only. or t=3/r2=0. The FIT procedure for transient problems is very similar to that for steadystate situations.517). transforms simply to @. It is straightforward to show that the solution of the resulting onedimensional problem is f ( x ) = 1.cos An sin An ) The initial condition. the thermal effects of the edge are "felt7' over about 1. it is necessary that cn=O. as mentioned in Section 3.5. . nnn=1 The solution in Eq. far from the lefthand edge of the object in Fig. (4.. Using this result to simplify Eq. The basis functions cannot involve time because the differential equation has only a first derivative in t. The corresponding dimensional time 1 ~ .4. (4. 2.03 (as stated above for Bi = I). We seek a solution of the form I Figure 48. t)=2 C ( I e(nm)2r). Thus. namely = Using Eqs. The first exponential reaches 5% of its initial value at r 2 t = 3 . (4. The basis functions needed here are those satisfying Dirichlet boundary conditions (case I of Table 4l). Thus. L is the membrane thickness and D is the solute diffusivity. (4.where b. In other words.y) =0. and cn are the unknown constants.536) converges very slowly. (The thickness is the dimension in the x direction.3. so that the time needed to reach a steady state can be estimated simply by looking at the first term. (4.527). .536) converges very rapidly for large t.(&)r 2 c ( An+Bi )e*~( Bi sin A.. we now consider times short enough that the concentrations in the external solutions remain approximately constant. (4. information on the mysterious fourth boundary is not needed. the membrane thickness. we wish to compute the solution for t 5 t. Thus. but long enough that the similarity solution derived in Section 3. Transient diffusion in a membrane. which is that containing A. with constant concentrations at the boundaries and an initial concentration Not only does Eq. The form of the solution far from the edge is identified by substituting O = f ( x ) in the original problem statement.4).524). This is a valuable guideline in modeling.3. t.536) more clearly reveal the form of the steadystate solution. edge effects in steady conduction or diffusion in a thin object become negligible after one to two thicknesses. Using Eq. we find that y = 1. The dimensionless problem for Q(x. 35(a). The main distinction is that in a onedimensional transient problem there is never a choice in the basis function variable. the complete transformed problem is (4. y) = 1 .) As a general rule. (4.05 as being close enough to zero and using A1 =2.5 thicknesses. the time to is 0 . as discussed in connection with Eq.. The solution to Eq. (4. the final solution for the transformed temperature is where the coefficients are now functions of time. which is used to make both coordinates dimensionless. 47 the solution is onedimensional. and a part which is independent of time and evidently represents a steady state achieved at large t. (4. it allows for terrnwise differentiation in x and provides an exact represention of the values of O at the boundaries. (4.. t) is depicted in Fig.413).530) Bi n= 1 sin Xnx I Transforming the time and space derivatives in the part~aI differential equation..518) in Eq. the overall solution is found to be $ sin n m .528) We wish now to estimate how near the lefthand edge the onedimensional solution is valid. Evaluating bn using Eq. the steadystate solution is identified as the Fourier expansion for the function f ( x ) = 1 .
number of terms as r+O. Alternative approaches are needed for calculations involving very small 2, such as the similarity method. Thus, the similarity and FFT methods may be applied to the same problem in a complementary manner.
0
The overall solution is
Example 4.54 'Mimieat Conduction with No Steady State We now consider a problem similar to the last example, but with Neumann (constant flux) conditions at both boundaries. Suppose that for r>O there is a constant radiant heat flux imposed on one surface of a solid slab, while the other surface is thermally insulated. The problem is illustrated in Fig, 49. With two Neumann conditions, the basis functions are those from case IV of Table 41:
Focusing now on the behavior of the solution at long times, we see that cos n m lirnB(x. t j = t + 2 C p t+f(x).
rfw
,=
1
( m 2
This is the first FFT example in which it has been necessary to include the constant basis function, a,, which corresponds to the eigenvalue &,= 0.The constant basis function requires special treatment and has an important effect on the structure of the solution, as will be seen. The expansion for the dimensionless temperature 8 is of the usual form,
the only difference being that the summation index now must start at zero. Transforming the spatial derivative in the usual manner, the integration by parts leads to
In contrast to Example 4.53, where the solution eventually became a function of x only, here there is no steady state, The physical reason for this is the imbalance in the heat fluxes at the two surfaces. Because the boundary conditions imply a continual net input of energy, the temperature increases indefinitely; the constant rate of temperature increase achieved at large times [the part of Eq. (4.545) proportional to t] is simply a reflection of the constant rate of heating at the x = 0 surface. (In reality, such temperature increases could occur only for a limited time, because a constant net heat flux could not be sustained once the surface became sufficiently hot,) The form of the function f ( x ) in Eq. (4.545) is inferred now from a combination of local (differential) and overall (integral) energy balances. We begin by stating some general relations, and then we apply them to this particular problem. Integrating the local conservation equation over x gives the integral balance for a control volume extending from x = 0 to x = 1:
Equation (4.546) is integrated over time to give Equation (4.539), the transformed time derivative [i.e., Eq. (4.53 I)], and the transformed initial condition yield the relations governing 63,. The results are written separately for n=O and n>O as
where 8 (t) is the average temperature at a given instant and s is a dummy variable. The foregoing relations are valid for all t. For a system with constant fluxes at both boundaries, we assume that the longtime solution is of the form Solving Eqs. (4.540) and (4.541) for 0, and 0,, respectively, gives lim @(x, t ) = Bt f ( x ),
I+
+
where B is a constant; compare this with Eq. (4.545). It follows that
1
&/dx =  1
I
I
I
a/& = $@/dx*
Figure 49. Transient heat conduction in a solid slab, with a constant heat flux at one boundary and the other boundary insulated.
Using Eq. (4.549) to evaluate the terms in the differential energy equation at large times, we obtain
a@ a +f(x)] = B, lim =[Bt , at at
a 2 0 dZ d 2f lim = ?[Bt + f (x)]= , a 2 ax dx
This confirms that Eq. (4.549) is the correct limiting form for the solution, provided that f ( x )
satisfies the differential equation,
Figure 410. Transient diffusion in a membrane, with a timedependent concentration a1 one boundary.
Using the initial and boundary conditions of the present problem in Eq. (4.547), we find that for all t,
Comparing this result with Eq. (4.550), it is seen that B = I [in agreement with Eq. (4.545)] and that
The basis functions are the same as those used in Example 4.53, and the approach is very similar. Equations (4.532) and (4.533) become
Equation (4.553) and the boundary conditions indicate that f(x) also must satisfy
At first glance it appears that with Eq. (4.555) in addition to the two boundary conditions in Eq. (4.556), the problem for f(x) is overspecified. That is not true, because the information gained h m the boundary condition at x = O turns out to be equivalent to that at x = I ; the integral constraint in Eq. (4.555) is needed to obtain a unique solution. The solution for f(x) is
Equation (4.562) differs from Eq. (4.533) only in that the nonhomogeneous term in the differential equation is now a function of time. This firstorder, linear differential equation has the general solution
Thus, the cosine series in Eq. (4.545) is found to be an expansion of the polynomial given by Eq. (4.557); an independent calculation of the Fourier expansion of Eq. (4.557) confirms that this is the case. The complete solution [Eq. (4.544)] is then rewritten more clearly as
Using the initial condition to evaluate the constant c,, the transformed concentration is found to be
Once again, it is only this last form of the solution which represents the boundary data exactly, and in which termwise differentiation of the Fourier series may be used to calculate The approach just described is useful also for transient problems which have steady states. the main difference being that for such cases B = 0. Thus, for Example 4.53,
Combining the transformed concentration from Eq. (4.564) with the basis functions from Eq. (4.530), the overall solution is
K
1
sin nnx
The solution obtained from Eq. (4.559).
f(x) = 1  x ,
is that already incorporated in Eq. (4.536).
:
which reduces to Eq. (4.535) for K = 0 (constant boundary concentration), as it should. Using the separationofvariables method to solve problems with timedependent boundary conditions, such as this one, it is necessary to employ a special technique called Duhamel superposition [see Arpaci (1966)). As seen here, the FFT method requires no modification for such The solution given by Eq. (4.565) provides insight into the pseudosteady approximation used for the membrane diffusion example in Section 3.4. If K is small, then the changes imposed on the boundary at x = O are relatively slow. In terms of Example 3.42. small K is similar to having L~/(DT)<< 1. In other words, the characteristic time for diffusion is much smaller than the external time constant which influences the boundary condition(s), so that the diffusional response of the system is comparatively rapid. From Eq. (4.565). a more precise criterion for slow changes at the boundary is */.rr2<< 1. When that condition is satisfied, the solution reduces to
Example 4.55 Tkamient Diffusion with a TimeDependent Boundary Condition We now consider a problem similar to that in Example 4.53, except that one of the boundary conditions is a function of time for r>O. As shown in Fig. 410, it is assumed that after the initial step change at t = O , the concentration at x = 0 decays exponentially with time, with a (dimensionless) time constant K  1 .
,
m
O(x, t ) = 2e ^'
n= 1
[l  e("")'']
sin n m = e"W(x, t), nv
The temperature is transformed first using the basis functions in x to give
where O*(x,t)~denotesthe solution obtained by specifying the constant concentration O = I at x=O; compare Eq.(4.566) with Eq. (4.535). For t>> l / v 2 (or t>>0.1), all of the transient terms in the summation are negligible and Eq. (4.566) reduces to
At this stage the expansion is written as
The last form of Eq, (4.567) is exactly what we would have obtained, much more simply, f r o m a pseudosteady analysis of the present problem (i.e., setting a@/& = O in the partial differential equation). As discussed in Section 3.4, it is seen that the pseudosteady approximation has two requirements: The boundary conditions must change slowly compared to the diffusional response of the system, and a certain amount of time must elapse before the concentration profile can achieve its pseudosteady form.
which is a straightforward extension of Eq. (4.44). Transforming a second time using the basis functions in y results in
b a
@m(Z)=10b*m(Y)@n(Y9 2)
o o @n(x)*rn(y)@(~, Y, L) dl. dy,
I
(4.576)
Example 4.56 Steady Conduction in Three Dimensions Problems involving three (or four) independent variables are solved by the application of two (or three) finite Fourier transforms. The steadystate heat conduction problem in three dimensions shown in Fig. 411 will illustrate the approach. It is assumed that 0= 1 on the top surface of a rectangular solid of dimensions a X b X c, and 0= 0 on all other surfaces; there is no volumetric heat source term. The full problem statement is
Clearly, the order in which the transforms were applied was immaterial. The problem is reduced now to that of computing the coefficient functions, On,(z). In transforming the boundaryvalue problem we apply both FFT's (x and y) at once, thereby proceeding directly to the governing equations for @,,(z). Applying both transforms to the x derivative in Fq. (4.568) gives
Doing the same for the y and z derivatives, we obtain
The expansion for B(x, y, z ) will be written using two sets of basis functions, one involving x and the other y. Because of the Dirichlet boundary conditions, we choose (from case I of Table 41) the basis functions
Combining Eqs. (4.578)(4.580), we obtain the differential equation for Onm(z),
d 20n, d2 + (m/b)2] rr20,, = 0. (4.581) Applying the two integral transforms to the nonhomogeneous boundary condition at z = c
'
gives
@= 1
Figure 4 1 1 . Steady heat conduction in a rectangular solid. There is a constant temperature, @ = 1, at the top surface (z = c) and O = 0 on all other surfaces.
The boundary conditions for Eq. (4.581) are then

The solution for the (doubly) transformed temperature is
@
Itm
= 2*
nmd
[I(
1)"][1 (I)"]
sinh ( [(n/a)2 + (m/b)'] 1'2?rr} sinh ( [ ( d+ ~ (m/b)'] ) ~ 1'2m]'
(4.584)
= d2/dr2, any combination of the usual three types of boundary conditions Thus, witb ZX gives a selfadjoint eigenvalue problem. It is shown now that a broad class of differential equations yields selfadjoint eigenvalue problems. Consider the general, secondorder, linear, homogeneous equation,
Using Eq. (4.577), and recognizing from Eq. (4.584) that only the odd values of n and m will contribute, the overall solution is found to be
Sening p(x) = exp( lf,(x) dx) and q(x) =p(x)fo(x),Eq. (4.64) is rewritten as
In general, if there are k independent variables in a partial differential equation, then k 1 finite Fourier transforms are needed to obtain the solution,
Motivated by the form of Eq. (4.65), we choose the operator weighting function w(x) as
Y,associated with some
4.6 SELFADJOINT EIGENVALUE PROBLEMS
AND STURMLIOUVILLE THEORY
It was mentioned in Section 4.3 that if an eigenvalue problem is selfadjoint, then the eigenfunctions are orthogonal (with the proper weighting function) and the eigenvalues are real. It was also stated that eigenvalue problems based on Eq. (4.34) are selfadjoint. To understand such statements and to extend the FFT method to a wider variety of problems, including conduction or diffusion in cylindrical or spherical coordinates, it is necessary to define what is a selfadjoint eigenvalue problem. An eigenvalue problem involving the independent variable x ordinarily consists of a secondorder differential equation in the form of Eq. (4.3I), along witb homogeneous boundary conditions of the types given in Eq. (4.35). That problem is said to be selfadjoint if the operator 2x is such that
where it is assumed that p, dpldx, q, and w are all continuous in the interval of interest, a 5 x 5 b, and that p>O and w > O for a<x< b. Assuming again that u(x) and v(x) satisfy any combination of the three types of homogeneous boundary conditions, the lefthand side of Eq. (4.61) becomes
where u(x) and v(x) are functions which satisfy the boundary conditions of the eigenvalue problem, but not necessarily the differential equation. Implicit in Eq. (4.61) is the specification of a certain interval [a, b] and weighting function w(x). To illustrate the application of Eq. (4.6I), consider the familiar operator d21dr2,for which w(x) = 1. Starting with the lefthand side of Eq. (4.61) and integrating by parts, we find that
The boundary terms vanish as before, provided that p(a) and p(b) are finite. Evaluating the righthand side of Eq. (4.61) gives
which confirms that problems involving this operator are selfadjoint. The boundary terms vanish as a consequence of the homogeneous boundary conditions satisfied by u and v. This is obvious when the boundary conditions are of the Dirichlet or Neumann types. If, say, there is a Robin boundary condition at x = 6, then uf(b)+Au(b)=vl(b) +Av(b) = 0 and
=I. $(, 2)&+
[quv *=(Xxu,
v).
Accordingly, eigenvalue problems based on the differential equation,
with boundary conditions as in Eq. (4.39, are selfadjoint with respect to the weighting function w(x). Eigenvalue problems of this form are called StunnLiouville problems [see, for example, Churchill (1963) and Greenberg (1988)j. The bounbry conditions given in Eq. (4.35) are not the only ones which yield selfadjoint eigenvalue problems. Also leading to such problems is the periodicity requirement,
coordinate. Taking advantage of the integration by parts performed for the SturmLiouville operator in Eq. (4.67), the transform of that tern is written generally as
Inspection of Eq. (4.67) indicates that if u and v satisfy Eq. (4.6lo), then the boundary terms will vanish in the integration by parts, as required for a selfadjoint problem. Equation (4.6 10) arises, for example, with problems in cylindrical coordinates involving the angle 8, in which the domain is a complete circle. One other way a selfadjoint eigenvalue problem is obtained is if p vanishes at one or both ends of the interval. If, for example, p(a) = O and @(a) and @'(a)are both finite, then the corresponding boundary term in Eq. (4.67) will again vanish. This type of condition arises in certain problems in cylindrical or spherical coordinates (see Sections 4.7 and 4.8). All SturmLiouville problems yield eigenvalues which are real and eigenfunctions which are orthogonal. A proof that A is real is given, for example, in Churchill (1963). The orthogonality of the eigenfunctions is a direct consequence of the fact that the SturmLiouville operator is selfadjoint, as is shown now. Considering two different eigenfunctions a, and @ ,, selfadjointness implies that
The eigenvalue problems introduced in Section 4.3 and applied in Section 4.5 correspond to w = 1, p = 1, and q = 0. The operators and eigenfunctions in some of the more common problems in cylindrical and spherical coordinates are discussed in Sections 4.7 and 4.8. Eigenvalue problems not encompassed by classical StumLiouville theory, such as those where w and p are discontinuous, are discussed in Rarnkrishna and Amundson (1974a,b, 1985) and Hatton et al. (1979). A few such situations are examined in the problems at the end of this chapter.
4.7 FFT METHOD FOR PROBLEMS I N
CYLINDRICAL COORDINATES
For the most general type of problem in cylindrical coordinates. 8 = 8(1; z, 0, 1). When there is no volumetric source term, the dimensionless energy or species consemation equation is
From the definition of the eigenvalue problem and the properties of the inner product (Section 4.3) we obtain
Subtracting Eq. (4.613) from Eq. (4.612) leads to
Because A, # A,, this implies that
We will be concerned only with steady twodimensional and unsteady onedimensional problems involving @(c z), @(r: O), or @(r, t). For @(< z), the eigenvalue problems in z will have the same differential operator as in rectangular coordinates, and thus they will be identical to those considered in Section 4.3. For O(r; 8), the eigenvalue problems in 9 again have the same differential operator; the only distinctive feature of these nonaxisymmetric problems has to do with periodic boundary conditions, as discussed at the end of this section. However, with @(c z ) or @ ( I ; t), the eigenvalue problems in r involve a differential equation not yet considered, Bessel's equation. Because they are the most unique aspect of cylindrical problems, the solutions of this equation are dis'cussed first.
,
Bessel Functions
Eigenvalue problems in r result in
In other words, the solutions to SturrnLiouville problems are orthogonal. In SturmLiouville problems with boundary conditions from Eq. (4.35). or with p = O at one of the endpoints, the eigenfunctions are unique. That is, there is only one linearly independent eigenfunction corresponding to a given eigenvalue. A proof of this, along with a discussion of the exception which occurs with periodic boundary conditions, is given by Churchill (1963). When applying the F I T method, the most awkward term to transform in the partial differential equation is that involving ie,, the secondorder operator in the basis function
'This general thcwy is due to C.s the generation folowing Fourier.
,
m (18031855) and J. Liouville ( 1 80918823, French mathematicians in
which is a SturmLiouville equation with w(r)=p(r) = r and q(r) = 0 [compare with Eq. (4.69)]. For problems involving annular regions, such that a 5 r 5 b with a 0, a selfadjoint eigenvalue problem results when any of the three types of boundary conditions from Eq. (4.35) are applied at r = a and r = 6. For domains containing the origin, all that is required at r=O is that @ and @' be finite, as discussed in Section 4.6. Equation (4.72) is a particular form of Bessel's equation, which is written more generally as
+
where a is a constant. The boundary condition at r = 1 yields the characteristic equation, Jo(hn)= 0. where rn is a ppameter and u is any real constant. In Eq. (4.72), m = A and v = 0. The properties of the solutions to Bessel's equation have been studied extensively (Watson, 1944). The two linearly independent solutions are usually written as J,,(mx) and Y,(mx), and are known as Besselfunctions of order u of the first and second kind, respectiveIy. Values of Bessel functions of integer order are available from numerous sources, including software written for personal computers, making calculations involving these functions quite routine. The solutions to Eq. (4.72) are the Bessel functions of order zero, J,(Ar) and Y,(hr). The first derivatives and the integrals of J,(Ar) and Yo(Ar) can both be expressed in terms of the corresponding Bessel functions of order one, J,(Ar) and Y,(Ar), so that we need be concerned here only with the properties of J,, J,, Yo, and Y,. Graphs of these functions are shown in Fig. 412. As the plots suggest, all of the functions have an infinite number of roots. Two values worth noting are Jo(0)= 1 and J,(O)=O. An important distinction between Bessel functions of the first and second kinds is that, whereas Jo(0) and J,(O) are finite, Yo(0) and Y,(O) are not. Of the many identities involving Bessel functions, ones which are particularly helpful to us in evaluating derivatives and integrals are
(4.77)
That is, the eigenvalues are the roots of Jo. The normalization condition for the eigenfunctions is
Notice the inclusion of the weighting function w(r) = r in the inner product. To determine a,, we need to evaluate the integral in Eq. (4.78). Integrals like this occur in any eigenvalue problem involving Jo, so that it is worthwhile to derive a general result for the interval [O,e], valid for any of the three types of boundary conditions ' . This is done by first recalling that J,(A,r) satisfies Bessel's equation with m = A, at r = k and v = 0,SO that
The next step is to multiply both sides of Eq. (4.79) by 2(dlddr), The lefthand side gives
Consider an eigenvalue problem involving Eq. (4.72) on the interval 0 5 r 5 1, in which one boundary condition is @(I)=0. The fact that Yo(0) is unbounded requires that the sorution Yo(Ar) be excluded, so that
? and using Eq. (4.74a) to evaluate the derivaIntegrating Eq. (4.710) from r = 0 to r = t tive of Jo, we obtain
The modified righthand side of Eq. (4.79) is rearranged as
.Integrating Eq. (4.712) by parts gives
Equating the results from Eqs. (4.711) and (4.713) leads to the desired identity, which is
X
FiF3rp 412. Besse] functions of orders zero and one.
Either form of Eq. (4.714) may be more convenient, depending on the boundary condition at r = 4. A second integral which may as well be evaluated now is one which arises when representing any constant as a FourierBessel series involving Jo(hnr).The integral is
719.e=8. we find from Eqs. k2e=s. Example 4. (4.74b). the orthonorma1 basis functions suitable for the Robin boundary condition in Eq. 1 7 3 .From case IIX of Table 42. the final solution becomes .81.72) on the interval [0. where r = 0 is not in the domain and the solution Yo(A. (1. The transformed temperature is defined as Notice again the inclusion of the weighting factor r: Using Eq. which is subjected to a constant rate of heating for t>O due to passage of an electric current. (4.721) and the graphs of the Bessel functions in Fig. ~. h24=7. (4. (4.so that the basis functions will contain only Jo(Anr). (4. (4.78).81. This is a transient problem leading to the steady state analyzed in Example 2.719). initially at ambient temperature.5a.719) are Case Boundary condition equation Basis functions 7 with eigenvalues given by the positive roots of III "All of the functions shown satisfy Eq. (4. with mf(0)=O. *The 6rst few eigenvalues for cases 1 and 11 are as follows: I: h.&5. It is found that the transformed temperature must satisfy The solution to Eq. the steadystate solution is !! Substituting @ . Heat transfer f r o m the wire to the surroundings is described using a convection boundary condition.717) containing the derivatives in r gives Equation (4. (4.717) and malung use of Eq.717) is one involving r.016.lsble 42 Orthonormel Sequences of Functions from Certain Eigenvalue Problems in Cylindrical Coordinatesa Characteristic The only eigenvalue problem derivable from Eq. we choose ~ @ / kand R as the temperature and length scales.715) makes use of the identity given as Eq.14) that The boundary terms vanish because of the homogeneous boundary conditions in Eq. For problems involving annular regions. )c&= 1 0 .4=3. for the timeindependent part of Eq.e=2. we obtain A very similar procedure is used to derive the orthonormal basis functions corresponding to Neumann or Robin conditions. Inspection of Eq. (4.76). the basis functions will be linear combinations of Jo(hnr) and Yo(Anr).727).832. The domain includes r = 0. 412 indicates that zero is not an eigenvalue for any Bi>O. Then. Table 42 summarizes the three types of basis functions arising in the analysis of conduction or diffusion inside a complete cylinder of radius 4.r) cannot be excluded. (4. and (4. the dimensionless problem formulation for @(r. Returning now to the eigenvalue problem for a full cylinder with a Dirichlet boundary condition at r = 1. (4. Transforming the mnhomogenwus term in Eq. (4. It: )bt=O. (4.616) to transform the term in Eq.7. and the Biot number is not necessarily large or small.71 'Itansient Conduction in an Electrically Heated Wire Consider a long. Refening to the dimensional quantities used in that example.654.725) is and the overall solution is ? The form of the overall solution is clarified by rewriting the steadystate part. h. cylindrical wire. (4. respectively. t ) is $ The transformation of the time derivative and the initial condition is straightforward. From Example 2.
Modified Bessel Functions Another differential equation which commonly arises in problems involving cylindrical coordinates is the modijied Bessel's equation. (4. Identities which are particularly helpful in evaluating derivatives and integrals are Example 4. The base is maintained at a different temperature than that of the top or the curved surfaces. ( m ) . I !: i Rgure 414. Assume that a hollow cylinder has a constant temperature at its base.730) and (4. and a different constant temperature at its top and its inner and outer curved surfaces (see Fig. The conservation equation and boundary conhtion for the reactant are written as .736) where a and b are constants.730) differ only in the sign of the m2*2term.73) and (4.735) is @(r)= alo(Da1f2 r) b ~ . 414). with m =DaIR. twodimensional conduction problem involving @(r. our concern is with the functions corresponding to v = 0 and v = 1. z X i Figure 413.72 Steady Diffusion with a FirstOrder Reaction in a Cylinder Consider a onedimensional. . in discussing the results of the singular perturbation analysis of this problem for Da+. ( D a ' ' r).73 Steady Conduction in a Hollow Cylinder Modified Bessel functions arise also in FlT solutions. The Darnkohler number (Da) is not necessarily large or small. and the reactant is maintained at a constant concentration at the surface. that solution is excluded. (4. Only half of the hollow cylinder is shown. Modified Bessel functions of orders zero and one. As with the "regular" Bessel functions.72.z( r d e r) ~ @ a =o. The solutions to Eq. Accordingly. respectively. Given that ~ . steadystate problem involving diffusion and a homogeneous reaction in a long cylinder.735) shows the latter differential equation to be a modified Bessel equation of order zero. the final solution is found to be This solution was stated without proof in Example 3. (4. and are called modijied Bessel functions of order v of the first and second kind. d0 (o)=o. ~ + (4. dr ~(I)=I.730) are written as Urn) and K . Graphs of these modified Bessel functions are shown in Fig. as this example will show. where rn is a parameter and v is any real constant. The reaction is first order and irreversible. I I I t K 1 1. 413. The steady. Steady. The most obvious difference between Bessel functions and modified Bessel functions is that the latter do not display oscillatory behavior or possess multiple roots. Equations (4. z) is . The limiting values of the modified Bessel functions are A comparison of Eqs. the general solution to Eq. ( ~ a " ~ is rnot ) finite at r = 0. twodimensional heat conduction in a hollow cylinder. written generally as Example 4. Using the boundary condition at r = 1 to evaluate a.
A peculiarity of these periodic problems is that both sin n8 and cos nt9 are eigenfunctions individually.n.5 . Solutions to Eq.. Basis functions in r will involve a linear combination of J. Nonaxisymmetric Problems It was mentioned earlier that eigenvalue problems involving the angle 8 employ the same differential equation as in rectangular coordinates. . the domain is a complete circle (e.749) in Eq. . Any attempt to use @. the constants in Eq. . the coefficients in the expansion for the temperature will involve sinh h. the preferred expansion is Transforming the term in Eq.743) and (4. so that both of the modified Bessel functions must be retained. c+27T].. (4.6. (4. there are two linearly independent eigenfunctions corresponding to each eigenvalue (each value of n). (4. .. 8) is that.. we can expand the temperature using basis functions in z. an= 1 The homogeneous form of this differential equation is seen to be the modified BesseI's equation of order zero.747) are found to be which are orthonormal with respect to the weighting function w(8) = 1 and any interval [c.751) yield a selfadjoint eigenvalue problem. That is the same as stating that the field variable is a singlevalued function of position. Equations (4. rather than in a certain linear combination..750) which satisfy Eq. n=1. = n. If.3. Thus.O and cos A. The eigenvalues will be solutions to a transcendental equation which involves the parameter K (the ratio of the inner radius to the outer radius).(8) obtained from Eq. unlike most other twodimensional problems (with both dimensions finite). That is. (4. (4. an integer. and of the boundary conditions in 1.738) containing the z derivative gives The transformation of the other term in the partial differential equation. Another distinctive feature of problems where O = O(r. it is found that the characteristic equation is The solution is completed by using Eqs.751) are sin A. 0 5 0 5 2v). there will be no true boundaries at planes of constant 8. 2 sin  @.744). with A. and Yo. and they will need to be determined numerically. (4.z (or the corresponding exponential functions of z). (4. is routine.= 1 + cos n0 y. In those situations it is sufficient to require that the field variable and its first derivative with respect to 8 be periodic in 8.746) is 7 6 (n+1)9 . As usual.g. For this reason. If the domain does not encompass a full circle. Thus. in that all of the eigenvalues are known explicitly. the general solution for Eq. Thus. the basis functions (from Table 41. however.We can construct an FFT solution using basis functions either in r or in z. there is no choice in the basis function coordinate. Including a particular solution to satisfy the nonhomogeneous tenn.746).752).4. the expansion must be constructed using @. then the boundary conditions will be any of the usual homogeneous types. case I) are then together with appropriate boundary conditions.747)(4..6. The eigenfunctions for problems which are periodic in 8 are 1 @o=. Applying the boundary conditions given in Eq. (4.. (4. @(@) satisfies and the basis functions are given by If these basis functions are used. with a period of 2.(r) fails because the resulting functions of 8 do not have the required periodicity.750) and (4. The governing equation for the transformed temperature is where c is an arbitrary constant. Given the Dirichlet boundary conditions in that coordinate.. n=2. (4. The domain does not include r=O or r = . adding 27r to 8 returns one to the same physical location. the negative eigenvalues are excluded to ensure orthogonality of the eigenfunctions. .750) are of the form T h i s set of basis functions has a major advantage. the latter solution cannot be excluded here because r = O is outside the domain. Alternatively. Indeed.z and cosh A. as discussed in Section 4. different numbering systems are employed for the eigenfunctions and eigenvalues.@. In such cases the conditions which accompany Eq.
the differential equation for O(r.72).82..86) q ( e ) + ~ @ . their properties are covered in discussions of Bessel functions of fractional order [see Watson (1944) or Abramowitz and Stegun (19 7 0 ) l .8 FFT METHOD FOR PROBLEMS I N SPHERICAL COORDINATES The discussion here is restricted to spherical problems which are nxisymmetric (i .84).84) differs from Eq. Modified Spherical Bessel Functions A differential equation closely related to Eq. X.is also expressible as C I1 q ( e )= 0 Ant = tan A. (4. (4.84). The solutions to Eq. Evaluating the integral in Eq. That is.82) where .86) when the domain includes r = 0.83) are based on . t) and no source term. (4. For the interval [O.l). 9.The eigenvalues and basis functions corresponding to the three types of boundary conditions are listed in Table 43. trigonometric functions of 8 have been eliminated in favor of polynomials in q. It is a particular form of the spherical Bessel S equation. (4. whereas those for the lattter give Legendre polynomials in q.zO. (4. & = 0.. Fortunately. Accordingly. with QJ(0)=O. (4. independent of the angle 4). may be expressed in terms of Bessel functions of order n +(112).88). for r+O.t). the dimqnsionless conservation equation is In problems involving complete spheres. 0 sin2 A. . and thus an. i sin h.176 SOLUTION METHODS FOR CONDUCTION AND DIFFUSION PROBLEMS 4. r d r Notice that the weighting function for this inner product is w(r)=?. the nth eigenfunction becomes CP. . only elementary functions are needed for conduction or diffusion problems involving O(r. . called spherical Besselfinctions.84) on the interval [O. (4. Changing variables in Eq. which is a S turnLiouville equation with w(r) =p(r) = r 2 and q(r) = 0 [compare with Eq. which is Bessel's equation of order zero. 7. el. ( e )= o x. t) is found to be .(r)= a n . We will be concerned only with problems involving @(q t) or a(< q ) . (4. The use of these basis functions in an FFT solution is illustrated in Example 4. Otherwise. It is usually advantageous in spherical problems to replace the coordinate 8 with q~ cos 8. Thus. Spherical Bessel Functions The eigenvalue problems in r which come from Eq. = 1=a 0 (A) 7 sin h r 2 r2 r r = a. The eigenvalue problems for the former yield spherical Bessel functions in r.85). The boundary condition used will determine the specific values of A . "All of the functions shown satisfy Eq.8. the solution to Eq. sin(hr)lrtA but cos(Ar)lr+w.2.A P ) tan hne where a and b are constants.t'] the normalization requirement is where 0 5 0 5 n.69)].1 5 q 5 1.e = ( 1 . bFot case 11.e. sin hr @(r)=a+b. only in the replacement of r by r2. Equation (4. we find that for any of the three types of homogeneous boundary conditions at the sphere sulfate. (4. the coefficient in Eq.. so that b=O in Eq.. which is written more generally as Table 43 Q r t h o n o d Sequences of Functions from Certain Eigenvalue Problems in Spherical Coordinatesa Characteristic Case Boundary condition equation Basis functions where m is any real constant and n is a nonnegative integer. r r .87) is given by Thus. which corresponds to n = 0. For @ = @(r. r cos hr r (4.( r sin Ant' n=1. Those functions are discussed now in turn. (4. (4. (4.. (4.Thus. the requirement that the temperature or concentration be finite everywhere eliminates one solution of Eq.85) is the modified spherical Bessel's equation.
818) was evaluated using Rodrigues' formula and n integrations by parts [see Arpaci (1966)l. Rectanelar problems yield either trigonometric or hyperbolic functions.812) transforms a problem for B(r. From Eq. The reactant concentration at the outer surface is constant.Example 4. they are standardized such that P n ( l ) = 1 for all n. @(l)=1.83. namely m. The application of Legendre polynomials to a diffusion problem is illustrated in Example 4.the eigenvalues must be given by The first four Legendre polynomials and corresponding basis functions are listed in Table 44. and it has solutions which are discussed in detail in Hobson (1955).813) has only one solution which is bounded at q = k 1. 11 in 11.81 Steady D m o n in a Sphere with a FirstOrder Reaction Consider a spherical particle of unit (dimensionless) radius. and q(q) =0. The psulting onedimensional problem for the reactant concentration O(r) is d0 .81. (4. For the special case of interest here.83) has the differential equation The integral in Eq. . Eqs. (The other linearly independent solution of Eq. One key finding is that for Eq. The application of Eqs. (4. for any problem on the interval 11. the solution is .: ple of a situation in which this transformation is useful is given in Problem 31. + Thus.e. and the system is at steady state. Legendre Polynomials The eigenvalue problem in q which is associated with Eq. (4. (4.815) leads to It is readily confirmed that Eq.#). where n = 0. and the remaining members of this orthogonal sequence can be generated using the recursion relation. From Eqs. An examspherical V2 operator into a problem for 8(. (4. cdled modifrd spherical B e s s e l ~ c r i o n s are . The first two Legendre polynomials are Po@)= 1 and P. spherical problems give the corresponding functions divided by s This underlies a wellknown transformation used in solving spherical conduction or diffusion problems. that solution is of no physical interest and will be disregarded. (4. the Legendre polynomials can be computed using Rodrigues' formula.85) and (4.811) to a diffusion problem is illustrated in Example 4.6). I cosh mx X Alternatively. Applying the normalization requirement to Eq. whereas the oddnumbered ones contain only odd powers of x. This is a SturmLiouville equation with w(q)= 1. (4. the orthonormal basis functions in q are which is Legendre 's equation.810) differ only in the sign of the m 2 2 term. (4. (4. where the Damkohler number is based on the sphere radius.730)].(x) =x. the even and oddnumbered Legendre polynomials are even and odd functions.. (4. Consequently. (4.811). The correspondence between the solutions to the ordinary or modified spherical Bessel's equations and the solutions to the analogous equations in rectangular coordinates is noteworthy. (4.1.) From SturmLiouville theory (Section 4. involves what are called Legendre polynomials of the second kind..8la).810) reveals that this is a modified spherical Bessel's equation.(O)=O.813). Eqs. Evaluating A using the boundary condition at the surface. The Legendre polynomials are orthogonal but not orthonormal.813) to have a solution which is bounded at q = 1 (corresponding to 8= 0 and v). The solutions to Eq. t) involving the t) involving the rectangular one. expressible in terms of modified Bessel functions of order n + (112). irreversible reaction.73) and (4. it is necessary that B=O. the eigenvalues are determined solely by the requirement that the solution be finite.815) and (4. (4. p(q) = (1 .9Da @=O. within which there is a firstorder. which is not bounded at q = +. The other key finding is that (4. A comparison with Eq. rZdr( d r ) d0 .810) and (4. the Legendre polynomials Pn(x)are orthogonal on the interval [1. I] with respect to the weighting function ' To keep the concentration finite at r=O (and to satisfy the noflux condition). (4.8lo).As with the corresponding equations for cylindrical problems [i. respectively.the general solution is expressed in elementary functions as sinh mx f(x) = A +B. the general solution is where the functions Pn(x) are Legendre polynomials. X W ( X ) = I. (4.817) that the evennumbered Legendre polynomials contain only even powers of x.816) or Eq. in which there is a change in the dependent variable given by It may be confirmed using either Eq.
1)" sin n n r @(rpt)=2 . with X. The overall solution is then n v ) ( . the steadystate solution for this problem is that obtained in Example 4. Using Eq. consisting of material B. q ( r . is Basis functions @o(n)= 3 Notice that the effect of the firstorder reaction is to give a second term multiplying On on the lefthand side of the differential equation.829) is 1 (n + Da ] { I . the steadystate temperature outside the sphere. (4.84. namely.616) to transform the r derivative term in Eq. (4.=. Specifically.((nn)2+ Da) I] 1. a r 'All of the functions shown satisfy Eq.ly [: ]exp[.72.832) The analogous problem for a cylinder was solved in Example 4. (4.831) As mentioned earlier.=. Noting that dTldz= G is equivalent to dTlar= Gq. Rewriting Eq.(4. A sphere of radius R.exp[. The resulting problem for 0. given by Eq.(r) = 0 o"(r)e(r. By contrast. A steady heat flux results from the imposition of a uniform temperature gradient G far from the sphere (parallel to the z axis). The smaller length scde shortens the diffusion time. The reason is that for large Da the relatively fast reaction prevents the reactant from penetrating beyond a small distance from the surface. setting Da=O in Eq..822).831) using Eq.829). r)? dr.TABLE 4 4 Legendre Polynomials and Corresponding Orthonormal Basis Functionsa Legendre polynomials Po(7))= 1 The remaining terms in the partial differential equation transform in a straightforward manner. (4. (4. the solution to the microscopic problem is used to derive a result useful on a more macroscopic scale. (4.81.814). The solution of this "microscopic" problem illustrates the use of Legendre polynomials and also shows how the FFT method . Example 4. (4.. 2[: nq~)~+D ]{le~~[((nrr)~+Da)t]). @(r. we obtain the relation for transient diffusion without reaction. In Example 4. we obtain the final form of the solution. even at steady state. r (4. The solution to Eq. is surrounded by material A. The expansion 1 @. the time to reach the steady state is shortened appreciably by the reaction. 415. nm) +Da sin n nr ((n~)~+Da)r].phase. I]. t) is then The problem statement for Notice that for Dacc a2. q).813) on the interval [I.83 Heat Conduction in a Suspension of Spheres This example concerns the steady temperature field in a system consisting of a sphere surrounded by some other stationary . is governed by The required basis functions are given by case I of Table 43. (4. t) = slnh(@ rs i r) n +25 nv)(. The microscopic problem is illustrated in Fig. the reaction has negligible effect on the time required to reach the steady state. all quantities here are dimensional. as does the initial condition. For convenience in later applications. In this case the duration of the transient phase is governed by the time required for diffusion over the entire radius of the sphere. assume that no reactant is present initially and that the mmensionless surface concentration is unity for t>O. Example 4.is applied to a situation involving two phases. even though Da may still be large enough to have an important influence on the steadystate concentration profile. The ratio of the thermal conductivities is denoted as y= k$k. for Da> n2. with and the transformed concentration are written as i ? = 1.82 'Ihnsient Diffusion in a Sphere with a FirstOrder Reaction Consider a transient problem wbich leads to the steady state of the previous example. The sphere and the surrounding material have different thermal conductivities. the effective conductivity of a suspension or composite solid containing spherical particles. and the temperature gradient far from the sphere is assumed to be uniform.823). we find that .
834). and (4. is transformed as with A(0. (4. (4. the transformed temperature in the surrounding material is Thus. The boundary condition at r = .(4. (4.814). and C . causes all terms to vanish except that for n = 1. no information is provided concerning A. (4. the internal and external temperature fields are coupled by the conditions The only eigenvalue problem associated with Eq. it is found that Transforming the individual terms in Eq. The gradient in the transformed temperature outside the sphere is evaluated from Eq.. as given in Table 44.846) vanish at r = . the differential equation for qn(r)is Using Eqs.833) is that involving q Accordingly. (4.q) Equations (4. @. (4. which have solutions of the form r*.849). the differential equation for the transformed temperature inside the sphere is dT/dz = G at r = Material A T = Y(r. A(r. (4.6.182 SOLUllON METHODS FOR CONWCnON AND DIFFUSION PROBLEMS 7)= 1 Figure 415.. From Eqs. Steady heat conduction in a sphere (material B) and surrounding phase (material A). Thus.844. using the boundary conditions at r = R and r = . but it is simplified immediately by recognizing that for the temperature at r=O to be finite. Eq.845) and (4..843). we obtain No conclusion is reached at this point concerning any of the B. we obtain .16). so that we cannot require that aNar=O at r=O. (4. and ~ = C O S 0.842) is similar. There is a uniform temperature gradient G far from the sphere. and solving for the constants B. because all of the corresponding terms in Eq.840)..840) made use of 4. (4. (4. (4.841) is I 1 q=1 0 The solution of Eq.(p). the roots of the resulting quadratic equation are found to be p = n and p = (n + 1). It follows that the solution inside the sphere has the form The temperatw inside the sphere.846). the general solution of Eq. (4. Eqs.848) and (4. Similarly. (4.) Requiring that the temperature and the normal component of the beat flux be continuous at the interface. Recalling that the weighting function for inner products involving an(?) is unity.842) are equidimensional equarions.q) finite. the fact that the temperature gradient at r = m is proportional to @. (4. (4. The last step in Eq.836) and (4. and C. The matching conditions at r = R.841) and (4.837). Substituting that trial solution into either of the differential equations. p).839) and (4. Likewise. B.847) to evaluate the terms in Eqs. must satisfy What remains is to determine the constants A..843) as Comparing Eqs. are transformed to Notice that the elimination of the boundary terms in the integration by parts requires only that the temperature and its gradient in 7 be finite at q = k 1.833) with the help of Eq. together with the orthogonality of the basis functions. all negative powers of r must be eliminated. (This problem is not spherically symmetric. the basis functions will be those obtained from Legendre polynomials.
't' .
Note in this regard that erf is an odd function. (4. This suggests that we can obtain a solution which has the desired character at x= 0 and r = 0 by using the x derivative of Eq. Second. O)=O for r # 0 . For convenience. and that C and aC/ax from h. (4. the fact that C. t)=O. t) satisfies Eq. as derived in Section 3. so that the solution for an instantaneous source at x = 0 and t = 0 is It is seen that the concentration profile resembles a normal or Gaussian probability distribution. (4.10. all variables are dimensional. In this section a series of diffusion examples is used to derive certain pointsource solutions.y. (4. (4. km. For a source at x = x' and t = t'. The key observation is that the values of aC1ax obtained fmm this solution approach a "spike" at x = 0 as t+O.. it is found that K=m. the generalization of Eq. or threedimensional.93). Thus. so that erf()= erf() = 1. t) be the corresponding solutions to onedimensional problems in y and z. The utility of the corresponding Green's functions in integral representations is discussed in Section 4. In this and subsequent examples in this section. applicable to a wide variety of linear problems.99) at rq. the solution for a sudden increase in concentration from 0 to Co at x=O is + y2 z2)'I2. The use of the x derivative in particular is motivated by the respective initial and boundary conditions of the two problems. Pointsource solutions are useful for two reasons.913) confirms that Eq.  \ where K is a constant.9I).96) is E m p l e 4. Let C. the pointsource solution is evidently of the f o m A termbyterm comparison of Eqs. (4. C(x. then dCBx (or any other derivative of the solution) also satisfies that differential equation. This approach relies on the fact that if a function C(x. Solutions for point sources of unit magnitude are called Green's finctions.91 Instantaneous Point Source in One Dimension The solute concentration C(x. As given by Eqs. all such source terms are referred to here as point sources.95) satisfies the initial and boundary conditions given by Eq./2. (4.(4. They may be either instantaneous or continuous in time. where m. (4.912) and (4.93)express the fact that the source is localized. Equations (4. (4. .94) are both continuous at x=O. (3. Moreover. Attention is given now to situations where the source is restricted to a plane. the concept of a point source provides a simple but effective model for physical situations'in which the actual source is highly localized. each of which is valid for an unbounded domain.Z.911) is a solution to Eq. It will be shown now that the desired concentration field can be obtained simply by differentiating the similarity solution for a sudden step change in the concentration at a boundary. (4. y.518). or a point within the domain. z. in which the overall solution is expressed as an integral or sum involving pointsource solutions. depending on whether the problem is one.91 I). It is also readily confirmed that Eq. It is straightforward to confirm that Eq. it is found that Although derived for the semiinfinite region x z 0. and that the total amount of solute for t>O is fixed ?a. this also represents the solution to a diffusion problem for an infinite domain. and let C2(y.5.92) and (4.99) is satisfied. with a peak centered at x = 0 and a variance of 2Dt. where m is the number of moles added at t = 0 and r = (2 The desired threedimensional solution is obtained from the onedimensional result of the previous example by recognizing that the problem is separable. (4. two. This is shown by substituting dC/ax for C in Eq. (4.98). (4. (4..92). (4.(x. as the similarity solution does. a line.92 Instantaneous Point Source in Three Dimensions The solution for an instantaneous point source of solute at the origin at t = 0 must satisfy C(km.SOUITION W H O D S FOR CONDUCTION AND DIFFUSION PROBLEMS throughout the region of interest.95). respectively. First. Now assume that + Evaluating the derivatives in Eq. t) be the solution given by Eq. Z.516) and (3. t) and C3(z. that no solute is present for tcO. y. t)=C(x.91) and interchanging the order of the partial derivatives.98) using Eq. is the amount of solute released (moleslarea).94). Using Eq. in which C = 2C0 at x = m and C = 0 at x=. t)=C(x. the singular solutions obtained in this manner provide the basis for certain analytical and numerical methods. t) resulting from an instantaneous source of solute at the plane x = 0 at t = 0 must satisfy Example 4.
(4. zt = L. L).917) was constructed by assuming that rn moles were released at each source (actual and image). T' L Image Source I =t Figure 416. Referring to the (x. where rir is equated with the total rate of solute diffusion through any spherical shell. z) and ( x ' . y'.and z = 0. The requirements of the real problem that C = 0 for x = t .918) is evaluated with the help of the variable change. Instantaneous solute source at a distance L below an impermeable boundary. Thus. Letting Att+O. in the integral. and then show how those lead to integral representations of the solutions to . It is found from Eq. With no flux across the plane z=0. as shown. for example.\r . 4. y = 0. and z= will continue to be satisfied.910) that Notice that the solution given by Eq. z<0. (4. However. indicates that the product C. The result for an instantaneous point source at the origin at t = 0 is then The form shown for spherical coordinates is the basis for Eq. z) coordinates in Fig.9. y.915) to a point source at position (x'. y = +. The generalization of Eq. how solutions to diffusion problems can be constructed by superposition of pointsource solutions.98) is linear and homogeneous.r'1214~(r1') 8 [ ~ ~ (r')] t 3i2 3n = C(r r I .and t' = 0. u = [r/2][D(t. The approach employed here to obtain a pointsource soh> tion for a partially bounded domain is an example of the method o f images. The purpose of this section is to show. t)= 8 [ ? r ~ (t t')] m(t1) dt'. namely the origin. respectively. We begin by defining the Green's functions which correspond to the basic pointsource solutions derived in Section 4.r ' ) 2 ] / 4 ~ ( d ) C(x. 416. the total amount of solute represented by Eq. 0.9 it was shown that solutions to certain diffusion problems could be obtained by adding pointsource solutions or integrating them over time. m moles will remain in the physically relevant region. in a more general way. suppose that a solute is released at the origin at a timedependent rate. where r and r' are position vectors with the components (x.t')]'n. Eq. (4. A fictitious image source of equal strength is placed an equal distance above the boundary. For the special case of a constant release rate. The impermeable boundary corresponds to the xy plane.917). the solute might be a nonvolatile contaminant buried in soil at a depth L below the surface. The linearity of the governing equations indicates that all such releases will make additive contributions to the solute concentration. because z>0 is outside the physical domain. (4.is not of interest. and C3 correspond to sources at the planes x = 0. The continuous release of solute can be approximated as a series of discrete releases. (4. z').93 Method of Images Consider an instantaneous point source of solute located a distance L from an impermeable boundary.188 SOLUTION METHODS FOR CONDUCTION AND DIFFUSION PROBLEMS C2.916) with x' = 0.=.t').5). the integral in Eq. Simply applying the pointsource solution for an unbounded domain. (2. 0. 'The steadystate concentration resulting from a constant release rate at the origin is simply m  It is easily confirmed that this solution satisfies the steadystate problem. must correspond to a source at the intersection of those planes. y'. t) = erfc 47iDr (2(Di) ] I 2 ) ' One way to obtain the steadystate solution is to let t + m and recall that erfc(0)= 1 (see Section 3. Accordingly. thereby ensuring that the noflux condition is met. t ) =  .. by Morse and Feshbach (1953.916 ) Example 4. The result for constant m is C(r. which was presented without proof. y. L). so that the sum of the contributions at the various release times becomes an integral. at t =O. for a total of 2m. respectively. each occuring during a small time interval At'. That C=O also for z = +. For example. 13Cldz = 0 at z = 0. t is treated as a parameter. 2. the solution for m moles released at the actual source is This expression represents a superposition of solute releases occuring at all times I' up to the current time t. y. pp.94 Continuous Point Source in Three Dimensions Solutions for continuous point sources may be expressed as integrals over time of the corresponding solutions for instantaneous point sources. so that the sum of any two of its solutions is also a solution.C2C. 812816). (4. t .917) is correct. we add to the problem a fictitious "image" source of equal strength at the position (0. Example 4. Part I.$)z+ ~ y y . y' = 0. a classical technique which is discussed. For example. the resulting concentration is found to be C(r. (4. The amount of solute released between times t' and t' +At t is given by m(tl)At'. (4.10 INTEGRAL REPRESENTATIONS In Section 4. Specifically. The sum of the two pointsource solutions will yield a concentration field which is symmetric about the plane z = 0. we can exploit the fact that Eq.) 2 + (z. does not yield a solution which satisfies the noflux condition at the boundary. the point source is located at (0. m(t) (moles/time). z') at t = t' [(x .
(1984).S(r .94. (4. This chapter concludes with two examples which illustrate the use of integral representations. Such effects are of interest in calculating rates of mass transfer in suspensions. The Dirac delta for three dimensons has the properties S(r) = 0 for r # 0. where r is the position vector.(rr) f(r. The Green's function for transient diffusion. Letting r +r. t. (4. Eq. such that the concentration is a known hnction of position on the surface. the transient result is C(r. t') F(r .105) . is defined as the solution for m = 1. the overall concentration is given by Green's Functions for Diffusion in Unbounded Domains For steady diffusion with a constant source of solute m at position r'.101 Multiple Expansion The objective here is to examine the effects of steady diffusion from an arbitrary particle on the solute concentration in the surrounding fluid.I .(r') as the only unknown. 1 (4. (4. This discussion is restricted to threedimensional domains in which C0 as (rl+ =. where the integration is over all positions r' in the unbounded domain V For timedependent sources. Including it in the Green's function. denoted as f(r r'). It is desired to determine the concentration field at positions r in the fluid such that I r I>> 1 r'. The approach is analogous to that used for the timedependent point source in Example 4.9. For example. the differential conservation equation is written as (4.1011) DV~F= aF at .r'. the Dirac delta with a scalar argument has propeaies like those stated above for &r).190 SOLUTlON MIZlWODS FOR CONWCllON AND DIFFUSION PROBLEMS problems which involve spatially distributed sources. From Eq.103) DVY= . From Eq.r'. I h(t)&i) dt = h(0). Integral Representations Solutions for spatially distributed sources may be written as integrals of Green's functions over position. The discrete values of m. far from the particle. Assuming that no solute is present at t = 0. satisfies where the integration is over all positions r' on the surface of the object. Thus. t) = 0 where h(r) is any function and the integration is over all space.10lo). v It is equally straightforward to describe the effects of solute sources at the surface S of an object immersed in an unbounded fluid.9.10) becomes mdr') f (r.6(r. Suppose now that there is a Dirichlet boundary condition. Example 4. (4. 417. t .1010) to evaluate the concentration at points not on the surface.10. jV h ( r ) & r ) d = ~ h(O). as done here. the solution is Whether to include D in the Green's function or to lump the difisivity with the source is a matter of taste.t'). which is a boundary integral equation involving the function m. I)r' .r')A V1. . Accordingly. it is necessary also to integrate over all times t' up to the present time t. the concentration field in the fluid is given by As mentioned in Section 2. the Green's function for transient diffusion is for t <t'. all variables are dimensional. (4.t'). Such problems are solved numerically by discretizing the integral to yield a set of algebraic equations involving values of m. r.tl)dV' dt'. The Green's function for this problem. Thus. For a steady problem with the source distribution r&(r I ) (moles/area/time).. (4. (4. (4. denoted as F(r . Eq. . .920) that .(r') at specific points. As in Section 4. As shown in Fig. 6(t)=0 for t f 0. provides the closest parallel with the approach for transient problems discussed below.107) :16 8 [vD(t  for t 2 t'.r') dS1. consider a steady problem with a volumetric source of solute mdr') (moles/volurne/ time).101) DV~C = m6(r . where S is the Dirac delta function. The formulation and solution of such problems is discussed by Brebbia et al. . the surface of the particle is described by the position vector r' (with maximum magnitude . I and there is a unit normal n directed toward the fluid.(rt) obtained by solving those equations may be used in Eq.r').916). The concentration increment at position r due to sources in a small volume A V' centered at r' is m. Letting A V' + O and integrating. rfzv(rf .r') and it is found from Eq. (4. except that there the integration was over time.r')S(t.101) reduces to Laplace's equation (no source) at all points except r = r'.
(4. (4. Example 4.84 by using the temperature gradient within a sphere. For higherorder terms. which is based on the form of the concenbation field far from a sphere. There is 30 specific consideration here of the rate processes occurring within the particle. pp. Specifically. where the solute source term on the surface (ms) has been rewritten as the outwardnormal component of the flux (naN). and K is the partition coefficient for the solute. Thus. the contribution to C. (4. the result is v apparent diffusivity within the sphere is KDs [see Eq. It is seen that the first.920)].1019) and (4. 6. This appmach is similar to that employed originally by Maxwell (1873). Multipole expansions are used to describe longrange effects in a variety of physical conkxts.101 indicates that the singleparticle contribution will be in the form of a dipole. To focus on the effects of the particles. is constant. (A.r') dS'. as shown by the second form of the equation. 12761283). results from net release of solute from the particle to the fluid. the integral in Eq. 0). Using Eq.8 . Equating Co from Eqs. Part 11.102 Effective DitTusivity in a Suspension The effective thermal conductivity of a dilute suspension of spheres was calculated in Example 4. the terminology comes from the use of such expansions in electrostatics. the steadystate solute concentration in the fluid is expressed as n C(r) = n N(r ') f (r . we obtain Is where CJr) is the concentration which would exist at position r if there were no particles present in the system. For convenience.66).104). . the total rate of release. It is assumed that the unperturbed part of the concentration corresponds to a uniform imposed gradient. as discussed in Chapter 7. . (4. such that V C . r') by c(r.39) and related discussion].1019) with the result obtained by assuming that the volume V is occupied by a single phase.1017).852) and by malung the substitutions R = a . Indeed. Erom each particle is that of an isolated sphere. the particle behaves as a point source of solute [compare with Eq. see Morse and Feshbach (1953. . A particle of arbitrary shape immersed in a fluid. Based on Eq.83. respectively.1014) in Eq.when there are no chemical reactions or solute reservoirs in the particles). Approximating c(r. using E . to b t approximation. whatever combination of diffusion and chemical reactions is present. (4. The integral is simplified by recognizing that at distant positions the distinction between r' and the origin (which is at the center of V) is insignificant. (4.(r) is the perturbation caused by the particles. The solute diffusivities in the sphere and the surrounding fluid are denoted by Ds and D. the result is perceived in the fluid only as a flux N(rf) at the surface.. the dipole is the dominant effect [see. Tills result was obtained by equating c with the dipole part of 9 in Eq. Accordingly. rq= z.852)]. (4. and the distinction between surface positions r' and the origin (which is within the particle) will be relatively unimportant. The next term is the "dipole. and n is the unit normal.. . the result is i. will disturb the concentration in the surrounding fluid by the amount w. or "monopole" term. the perturbation in the concentration far from V is given by f This expression resembles Eq. Positions within the fluid and on the particle surface are denoted by r and r'. Example 4.. Summing the effects of all particles. the Evaluating only the first two terms.: t'. it is found that Equation (4.e.. (4.1020). r') is not a Green's function. %: . with concentrations based on values in the fluid. The reason that K appears in the diffusivity parameter y is that the derivation of Eq.1012). defined as the spheretofluid concentration ratio at equilibrium. they are especially valuable in the analysis of momentum transfer in suspensions. (3. 2:. although c(r. ." which involves the first moment of the surface flux and which decays as rP2.Figure 417. (4. Suppose now that a suspension with a uniform number density of n spheres per unit volume occupies a volume C: which is large enough that it contains many particles. respectively. (4. T(c I))in (4. .1012) is evaluated by expanding the Green's function in a Taylor series about r' = 0. and C.For particles which do not act as net sources (i. a homogeneous sphere of radius R. where Keff and Deware the effective partition coefficient and effective diffusivity. which decays as r'. by analogy. For a dilute suspension.108). the first integral is just m. Assuming that there are no net sources of solute in the particles. where 4 is the volume fraction of spheres. K::. Using Eq. the concentration perturbation at r due to a single sphere of radius a centered at r' is ! . respectively. L : . The effective diffusivity in a suspension is derived here using a different method.1015) is an example of a multipole expansion.852) assumed that the field variable was continuous at the phase boundary. At distant positions in the Ruid the particle will appear almost as a point. (4. that dipole is evaluated using the results of Example 4. V is chosen to be a sphere of radius R. The effective diffusivity of the suspension is found by comparing Eq. Thus. it is found that Y. centered at the origin. and Gz =raVC.
1967. Advanced Calculus for Applic~tions. 1982. D. V. including both phases. Specifying KcH is the same as defining the concentration which is to be used in conjuncion w i t h Den There are two common approaches. A. D. Chem. Churchill. A twodimensional analysis of porous membrane transport.862). second edition. Washington. Sci. 1988. Sci. S. Problems 41. Chelsea. T . Transport properties of twophase materials with random structw. Watson. V. Eng. Rev. PrenticeHall. H. Steady Diffusion in a Square Rod with a FirstOrder Reaction Assume that a firstorder. E. MA. New York. N. Amundson. in the fluid surrounding V). y). 29: 14571464. A. I. Arnundson. 29: 13531361. Sci. as shown in Fig. Wrobel. Ramkrishna. is given in Batchelor (1974). Amundson. Eng. 1974. and Ken= 1. F. second edition. Eng. The lhe'O(@)" is a reminder that particleparticle interactions. and T. W. It is evidentrfrom Eq. 1973. tubular reactors. (4. Use the FFI' method to determine B(x. 30: 556562. Polym. a frequent source of confusion in the literature on sffusion in suspensions and porous media is the failure to state explicitly which choice was made. Boundary Element Techniques. Wlkbrmd. Transient diffusional interactions between solid bodies and isolated fluids. P41. Annu. 1972.Deen. Sci. Lightfoot. K. 1970. Proc. Stein. P. DC. A.and I. C. New York. NJ. F. A Treatise on t h Theory of Bessel Functions. P42. Chem. MA. Jaeger. Ind. at equilibrium. S. (4. G. Department of Commerce. me Theory of Spherical and Ellipsoidal Harmonics.. Methods of Theoretical Physics. W. Berlin. 1944. A. S. F. C. R. R. Conduction through a random suspension of spheres. Membrane mass spectrometer inlet for quantitation of nitric oxide. Clarendon Press. National Bureau of Standards. R. 1974a. N.I)+. . Advanced Engineering Mathematics. Chem. 1979. extending beyond heat conduction and diffusion. and N. and N. 1991. Ctank. R. 6: 227255. M.1976. and H. Biol. New York. D.. J. Deen. A Treatise on Electricity and Magnetism. In that case. London. Linear Operator Mefhods in Chemical Engineering. R. M. sionless. Polym. Mathematical Physics. P. Englewood Cliffs. heat conduction problem shown in Fig. Vol. and a more general introduction to their application in physical problems is found in Morse and Feshbach (1953). irreversible reaction takes place in a long solid of square cross section. C. Linear Operator 7heory in Engineering and Science. Oxford. NJ. Noble. 1955. 1892 [first edition published in 1873). Stirred pots. Conduction of Heat in Solids. M. = 14(K. were not included in the analysis. Chiang. is not obtained until Kc. Conduction Heat Transfer AddisonWesley. J. Steady conduction in a square rod with a constant heat source. 1963. J. 1984. M. third edtion. Batchelor. U. Reading. A review of the transport properties of systems where one phase is dispersed in another. 5: 773778. D. All surfaces are exposed to a uniform concentration of reactant.94 SOLUTION METHODS FOR CONWC~ON AND DIFFUSION PROBLEMS I : I hich is analogous to Eq. An important special case is a suspension of impermeable spheres. Ramlaishna. Joseph Fourier 17681830. Hatton. M. l\vo adjacent sides of a square rod are insulated. Transport in composite materials: reduction to a selfadjoint formalism. and J. Then is a constant volumetric heat source. Clarendon Press. second edition. Lewis. Morse. 1953. M. Soc. The difference between these expressions underscores the need to define K . Sell. The brief djscussion here of Green's functions and integral representations necessarily omits many aspects of these subjects. New York. MIT Press. Feshbach. Englewood Cliffs. B. D. Oxford. G . Lond. Naylor. 1967.. 1975. Consider the steady. Cambridge. The other common ipproach k to use the volumeaverage concentration in the suspension. S. Keller. Reading. 1974b. Handbook of Mathematical Functions. K. ~hich gives & . R. and W. Fluid Mech. Chem. One is to base the suspension concentration on hat in the fluid (continuous) phase. is pecified. Setting y=O in Eq. S. 42. Butkov. vhich would yield a # term. and L. R. J. A. and N. MA. Tannenbaum. Fourier Series and Boundary Value Pmblems. while the other two sides are maintained at the All quantities are dimenambient temperature. 1959. T. Res. PrenticeHall. NJ. 1993. Eng. and G. R. E. Arpaci. the suspension concentrahon will quai that in particlehee fluid (ie. Steady Conduction in a Square Rod with a Heat Source References Abramowitz. Jeffrey. Englewood Cliffs. GrattanGuinness. PrenticeHall. H. twodimensional.second edition. Brebbia. Use the FFT method to determine O(x. Unfortunately. Telles. Biosci. Equilibrium distribution of flexible polymer chains between a macroscopic solution phase and small voids. A 335: 355367. The Mathematics of Dzmsion. 1. SpringerVerlag. I. y). C.. and selfadjoint operators. Wishnok. 1966. Ramkrishna. Clarendon Press. Casassa. Math.S. Mass Spectrometry 22: 4552. McGrawHill. 1985. and J. Oxford. 1968. 34: 13391344. C. Stegun. Juhasz. Figure P41. Greenberg. Cambridge University Press. AddisonWesley. E. McGrawHill. Maxwell. A more extensive discussion of the use of Green's functions in conduction problems is given in Carslaw and Jaeger (1959). (4. R. Lett. SpringerVerlag.1021) that a definite value of D . 1: 421437.1021) gives Hobson. Effect of local Peclet number on mass transfer to a heterogeneous surface. S. Carslaw. N. second edition. and E.
.21 it was shown that for Laplace's equation applied within an arbitrary volume K with a Dirichlet boundary condition at the surface S. irreversible reaction. What is the relationship between this function and the Biot numbers for heat or mass transfer?] 48. (b) Repeat part (a) using the cosines given as case IV of Table 41. The top and bottom surfaces are insulated.71 1 5). Steady conduction with Neumann boundary conditions. the lefthand surface has a specified temperature f(x). Steady Conduction in a Lofig. Figure P48. How do local convergence and mean convergence compare with part (a)? (c) Does termwise differentiation of the series from (a) lead to a representation of ff(x)? How about the series from (b)? : 47. Eq. the solution is unique to within an additive constant.] solvent where the monomer is maintamed at negligible concentration. dO/dx = 0 !:. Three sides are insulated and there is a specified heat flux at the fourth. (4. G(t). (b) Calculate the average monomer concentration remaining in the sheet. . Make a plot which compares the exact function with the results from 4. Da. for t>O the reactant concentration is held constant at x=O.)6 . Uniquenesr Roofs for Laplace's Equation In Example 4. y). The righthand boundary is at ' . lkansient Diffusion in a Film with a FirstOrder Reaction . The surface at x = l is impermeable. Steady conduction in a long. Steady diffusion w i t h a firstorder reaction in a square rod. (b) Show that for the Robin boundary condition. Figure P44. Is there a unique solution? (b) What happens if the flux at the top is a constant? Suppose. ' Figure P42. .(rs)?O. It is proposed that the dimensionless problem statement be 44.x 13. ?he Darnkohler number.and 20term partial sums of the series. Use the FFI'method to determine O(x. homogeneous. (a) Use the FFT method to determine Wx. that de/dy = 1 instead of 8OBy = 4. 45. Fourier Series Representations (a) Derive the Fourier series for f ( x ) = x for 0 S x r 1.. . (4. . the coefficient function h. rectangular solid. [Hint: First show that for the situations of interest in heat or mass transfer. (3. the solution is unique. Eq. 8 Consider transient diffusion in a liquid or solid film with a firstorder. t. Compute the norm of the error for both cases. and that at t = O the sheet is immersed in a wellstirred container of Figure P43. 46. In this problem the Neumann and Robin boundary conditions are considered. P43. express the flux in dimensional form.25c). ' PC.x. y). P48. f d81av= 112 . P44. $'. As shown in Fig.'! . (a) Use the FIT method to determine B(x. ). [Hint: Compare your result with Eq. . and the righthand boundary is sufficiently distant that it can be modeled as being at y = 00. assume that there is no reactant present initially and that . (c) Determine the outward flux of monomer at the surface of the sheet. Steady Conduction with Neumann Boundary Conditions Consider steady. (a) Show that for the Neumann boundary condition. there is a unique solution. 'hansient V i s i o n from a Solid Sheet It is desired to characterize the rate of diffusion of a contaminant (unreacted monomer) out of a sheet of polymeric material. using the sines given as case I of Table 41. twodimensional conduction in the square rod shown in Fig. The model to be employed assumes that the monomer is initially at uniform concentration. Transient diffusion in a film with a firstorder reaction. is not necessarily large or small.25b). Rectangular Solid Consider steady conduction in the rectangular solid shown in Fig. for example.
i.  SOLUTION M H O D S FOR CONDUCllON AND DIFFUSION FWO6LEM3 c=c. t) is governed by then the average mass transfer coefficient. as shown in Fig. and determine Da*. method. the maximum value of Da for which the proposed problem formulation is valid.e. Mass lknsfer to a Partially Active Catalytic Surface Consider steady diffusion of a reactant across a Liquid film to a catalytic surface which has alternating "active" and "inactive" areas. (a) Use the FIT method to determine C(x. (c) Assuming that Da<Da*. Because of the symmetry it is sufficient to consider onehalf of a repeating unit of width b. Diffusion in a Liquid Film with a ZerothOrder Reaction A liquid film resting on a surface is brought into contact with a gas mixture at t =O. P413. which is the equilibrium ratio of the polymer concentration in the pores to the polymer concentration in bulk solution. Partitioning of Polymers Between Pores and Bulk Solution One of the simplest models for the configuration of a linear. A component of the gas which is not present initially in the liquid is absorbed and undergoes a zerothorder reaction (i. r). for the . The fibers are widely spaced. .both small and large Da. denoted as e = u/b. (kc}lk. The relative mass transfer coefficient depends on two dimensionless parameters. (b) Evaluate the steadystate concentration profile.72. parallel fibers of material B (conductivity k . @. It is proposed except with the reaction that the reactiondiffusion process be modeled as shown in Fig. (kc }. Certain configurations of long chains are unable to fit in small pores. so that it is sufficient to model a single fiber of radius R.e.. P49. where kc* = DIS is the mass I ~swnes that N is large. Co. are the mots of the Bessel function I.  . kc@). . r). which Evaluate the relative mass transfer coefficient. as a function of time.The variable r represents position along the cbain. e and K = 2Nb. the radius of gyration is given by (Nt2/6)1n. expressed as a fraction of the total contour length (i. 412. Steady Conduction in a Fibrous Material i I It is desired to characterize steady heat conduction through a composite which consists of h n g . In this model. determine @(I. with a constant rate whenever the reactant is present). (b) Evaluate the reactant flux at the surface. far from a given fiber. use the FFI' method to determine @(x. At the top surface of the f i l m the reactant concentration is a constant. The fraction of the sudace which is active. The probability distribution P(r. X C F I I I Fieure P49.. The extent of this steric exclusion is reflected in the partition coefficient K. Assume that. This result was derived originally by Casassa 41967)..is defined as w h e r e P(z 1) is the probability of finding one end of the chain at dimensionless position c once all chain configurations that intersect the pore wall have been excluded. flexible polymer is a "randomflight chain. leads to the steadystate problem described in Example 4. The reaction is confined to strips on the surface of width a  (a) Define the Damkohler number used here. Mass transfer to a partially active catalytic surface. y). For a randomflight chain and cylindrical pores the partition coefficient is given by (b) If the local mass transfer coefficient. Show that the partition coefficient is given by transfer coefficient for a completely active surface.(x). term written as Da rather than Da@. P48. (a) Assuming that there is no reactant within the cylinder initially. 410. For what combinations of E and K will (k. is given by . The position of each successive mass point is selected at random. The analogy between the resulting set of polymer configurations and the random walks of diffusing molecules allows the probability of certain chain configwations to be calculated using an equation like that used to describe transient diffusion.one t.*.as shown in Fig. b n s i e n t Diffusion in a Cylinder with a FirstOrder Reaction Consider a transient reactiondiffusion problem in a cylinder which. ) surrounded by a continuous phase of material A (conductivity kA). is not necessarily small.where A is the radius of gyration of the polymer divided by the pore radius. (a) Determine 6 ( x .. On the active areas of width a there is a constant flux No (directed toward the surface).)lk. with reference only to the position of the immediately preceding mass point. Compare the estimate from obtained itom orderofmagnitude and scaling considerations. there is a uniform temperahue @ent of magnitude G directed perpendicular to the fiber axis. 411. as indicated by the dashed lines. on the inactive areas the flux is zero. 49. t) using the (b) Use the exact solution from (a) to estimate the time required to reach solution a steady with state. Oats I).* + I? Discussions of mass transfer at surfaces of this type me given in Juhasz and h e n (1991) and Keller and Stein (1967). Accordingly." In this model the polymer is assumed to be a freely jointed chain consisting of N mass points connected by rectilinear segments of length l .. $$ where 8. :' +13. at large times. the shape of the chain is described by a random walk.
.
. (y. the inlet is designed so that the rate of removal of the species is small enough to have a negligible effect on thesystem being monitored. (a) Consider eigenvalue problems based on gZ. K using the . Assume that the initial concentration of dopant is uniform at C. and that solidification at velocity U begins suddenly at t = 0 (see Fig. t. is needed for solving certain problems involving uniform convection. What happens to the exponential factors? *This problem was suggested by R. C. 418. (d) Estimate the halftime of the response (i.202 SOLUTlON METHODS FOR CONDUCTlON AND DIFFUSION PROBLEM Fluid Mixture Figure P419. 8). where 8. or membranetofluid concentration ratio at equilibrium. An examination of the selfadjoint eigenvalue problems associated with ie. Equation (2. t). Ideally. Ttansient Effects in Directional Solidification* Suppose that it is desired to measure the Darcy permeability (K) in a small. (a) Use the F I T method to determine C(y. Time Response of a Detector with a Membrane.a inlet which is permeable to the species of interest. r). When the inlet device contacts the m vacuum maintained in the instrument causes the species to diffuse across the membrane and enter the detector. P419. For information on the inner product and eigenfunctions. d@/& + Bi 0= Bi x=o x= 1 420. the membrane thickness (L).Inlet One strategy for continuo~slymonitoring the concentration of a gaseous species or a volatile solute in a Liquid is to equip a mass spectrometer (or other suitable detector) with a membrane i x t u r e . the overall response is limited by mass transfer in the fluid and diffusion through the membrane. If the dopant level within (a) Use rhe FFT method to determine 8 ( 1 . t) in the melt. Detector with membrane inlet. Show that this is accomplished by setting @=exp(Pe x12)q and that the general solution is 419.e. (b) A useful trick for solving differential equations like that in part (a) is to modify the dependent variable so as to eliminate the first derivative. is denoted as K. Bi) be defined? Pertinent dimensional quantities include the bulk concentration in the fluid (C. together with any combination of the three types of homogeneous boundary conditions. see Problem 420. if Darcy's law is used to describe the flow of an incompressible fluid in a porous material.A. A known volumetric flow rate Q will be imposed. and Po will be measured. where 9 . An application of these results is in Roblem 421. for t>O). At the sphere surface. the pressure field is governed by (c) Determine the eigenvalues and orthonomal eigenfunctions for the Dirichlet boundary conditions. Inner Product and Eigenfunctions for Uniform Convection* Figure P418. for 0 5 O< y and 9 = 9. @(O) = @(I)= 0. The parution coefficient. Brown.85. C (y. The sphere will be held at the end of a circular tube by suction. Determine Cs dopant concentration in the solid is not uniform.are constants. (c) Evaluate the solute flux leaving the membrane at x = 0. that is. Flow in a Porous Sphere As discussed in Problem 22. Thus. (1993). the time required to reach onehalf of the final detector signal) for Bi = 1. (a) How must the dimensionless quantities (x. it is assumed that 8= 8.. 210). Consider what happens when the solute of interest is added suddenly to the fluid. the solute diffusivity in the membrane (D). A schematic of a membrane inlet is shown in Fig. for y< 0 5 w. the flux leaving the membrane at x=O). A differential operator which arises in problems involving conduction or diffusion accompanied by a uniform bulk velocity is where Pe is the P6clet number. " is needed in the inner product to make Show that a weighting function w(x) =eAPe such a problem selfadjoint. The objective here is to extend that model to describe transients associated with the startup of the process. porous sphere of radius R using the arrangement shown in Fig.849) provides an example of such an operator.e. and the pressures 8. (b) An important consequence of the timedependent concentration in the melt is that the = . P418. (. (b) Derive the p r e s s u ~ f l o w relationship which would pennit calculation of experiment described above. and 8. A stagnantfilm model for unidirectional solidification was presented in Example 2. (b) Use the FfT method to determine O(x. Assume that the detector itself responds instantaneously and that its signal is proportional to the solute throughput (i. assuming that diffusion witbin the solid is negligible. and the mass transfer coefficient (kc) in the fluid. 421. t). A porous sphere held by suction at the end of a circular tube. An example of such a system was described by Lewis et al.
as described in Ramkrishna and Amundson (1974b). onedimensional conduction in rectangular coordinrites.r What is the physical basis for these conditions at x = y? The results given here apply to a composite material consisting of two layers. t ) for the solid and V ( t ) g. for the Liquid. Inner Product and Eigenfunctions for NonUniform Material Pmperties r' Figure P424. the dimensionless energy equation iswritten as ' 424. When the heat capacity per unit volume (&) andlor thermal conductivity vary w i t h position. Thus. the solute concentration in the solid. show that the Green's function may be written as and inner product which yield a selfadjoint eigenvalue problem are Note that in this case it is not possible to satisfy the condition C+O as r+m. t). note that the form of the eigenfunctions changes at x = y. Also. For such problems it is helpful to write the overall solution as a vector. as shown in Fig. as when the region of interest is a layered composite consisting of two or more different materials. The reaction rate is sufficiently fast that C = 0 at the surface. Green's Functions for Diffusion in W o Dimensions (a) For steady diffusion in two dimensions. Tkansient Conduction in a Layered Composite Consider transient conduction in a composite solid consisting of two layers. with spatially varying properties. show that a selfadjoint eigenvalue problem is obtained with the inner product 426.) Particular care is needed when the dimensionless properties C(x) and K(x) are discontinuous.. [Hint: The characteristic equation for the eigenvalues comes from the matching conditions at x= y.(')(X) for 0 5 X< Y. O(x.(x) = O.t') ' for r <t'. t). which have different thermophysical properties. O(x. For transient.C. Conduction in a layered composite consisting of materials I and 2. and @"(x) = for y<x 5 1 . 423. Point Source Near a Boundary Maintained at Constant Concentration Assume that a reactive solute is released at a steady rate m at a point source located a distance L from a catalytic surface. show that the Green's function is (b) If C(x) and K(x) are both piecewise constant.< . it is assumed that the solute concentration in the liquid is zero at all times. Brown. except that the liquid volume is assumed to be finite. for t 2 t'. Transient Diffusion from a Solid to a Stirred Solution of Finite Volume In Problem 46. (a) If C(x) and K(x) are continuous functions for 0 a x S 1 and if the usual types of homogeneous boundary conditons apply at x = 0 and x = I. which involves transient diffusion from a solid to a stirred liquid.. show that the differential equation 425. the eigenvalue problems for conduction differ from those for constant properties. 4 d ( t. where r = (2 y2)1n. . Construct a solution for C(r). is given by If the usual types of boundary conditions apply at x = 0 and x = 1 . !. the solute concentration in the liquid increases as that in the solid decreases. In this problem the concentration field consists of two parts. + (b) For transient diffusion in two dimensions. until a steady state is achieved. 204 SOLUTlON MnHODS FOR CONDUCTION AND DIFFUSION PROBLEMS / the solid must be within 5% of the steadystate value to give acceptable material properties. P424. Material 1 Material 2 422. how much of the product must be discarded? *This problem was suggested by R.. Assuming that the mass transfer resistance in the liquid is negligible. . . It may be helpful to let @. Using the information on the inner product and eigenfunctions in Problem 423. A. is governed by provided that where *(t) denotes the concentration in the liquid. I. determine @(x. such that C". The physical situation considered here is similar. this approach may be extended to any number of layers. then the operator 3.
I ) and T(t).(1979).(t) is governed by The nth basis function is also written as a vector. are discussed in Ramhnshna and Amundmn (1974a) and Hatton et d.. but (as given below) a special form of inner product is needed to make the approach work. t) and q ( t ) are expressed as to complete the solution for cP. t) and 1Ir(t). (The normalization condition is applied later. so that the differential equation for the eigenvalue problem is expressed as where H is a constant. Bn(t). (g) Complete the solution for O(x. Also.6. in that both result in a sum of scalar terms. where a solid or fluid region with a spatially distributed concentration or temperature is in contact with a stimd Ruid.206 SOWTlON MFMODS FOR CONDUCTION AND D / where the second form is obtained by using the boundary condition at x = 1. . (c) Use the first line of the eigenvalue problem to determine cDn(x) to within a multiplicative constant. an(x). (a) Show that the liquid concentration is governed by How must 9(t) and y be defined? @) Based on the differential equations for @(x. prove that the eigenfunctions are orthogonal in H i n t :Use the approach in Section 4. the differential operator with 0 is written as and the corresponding eigenvalue problem is Show that if the inner product of the vector functions u(x) and v(x) is defined as then the eigenvalue problem is selfadjoint. Evaluate H and solve for O.procedure. A vector operator ZX is defined so that the differential conservation equations for B(x. The expansion for 0(x.1 the inner product defined in part (b). [ (f) Show that O. It is seen that the inner product of the solution and basis function vectors is a scalar. Note that this inner product of vectors is similar to the dot product of vectors. = n= 1 @(x.) (d) Use the second line of the eigenvalue problem to find the characteristic equation governing A. r) = n= 1 @n(r)@n(x) I r Problems of this general type. which represents the time dependence common to the two parts of the solution.(r). The overall approach closely parallels the usual Fm.. t) is written as 00 00 1 (a. (e) Use the normalization condition.
.
23) to compute w(r. and r2 are not. there is no need to specify an axis of rotation. The matrix in Eq. If w = 0 everywhere. In the situation illustrated in . The customary measure of rotational motion at any point in a fluid is the vorticity. [Some authors omit the factor from the definitions of the rateofstrain and vorticity tensors. as pointed out in Aris (1962. we find from Eq. but in general r. is considered first.22) in terms of w.21) 1 P 1 According to the righthand rule for cross (vector) products. Note that when we use Eq. The coordinate origin is point 0. R.v. Motion of material points in a fluid undergoing rigidbody rotation. an antisymmetric tensor can be constructed from h is related to the vorticity the three components of any vector.R. pp.610).Fig. the flow is said to be irmtational. are normal to the axis of rotation. like that of a rigid solid. there is a simple proportionality between w and a Taking the curl of both sides of Eq. and points 1 and 2 are material points on two of the circular streamlines.RigidBody Rotation Purely rotational motion. the vectors r. and r2 in the direction parallel to w or n do not contribute to the cross product. where E is the alternating unit tensor (see Section A. and AR = R2. (Letting t o represent a vector rather than a tensor deviates from our usual convention.r. This result is used below in interpreting the difference in velocity between neighboring material points in an arbitrary flow. (5. whereas the vectors R1and R Thus.. we focus first on the properties of a. Ar = r2. we find that V. (A. it is found that for rigidbody rotation. is a dyad whose nine elements in rectangular coordinates consist of all possible first derivatives of the three velocity components with respect to the three coordinates.Any antisymmetric tensor has only three independent. Using the identity Figure 51.3). streamline " : . Recognizing that R amounts to a twodimensional position vector.21) and using identity (6) of Table A1. and R2are parallel to the paper. For the special case of rigidbody rotation. The definitions used here are the ones more commonly employed. (5. The angular velocity is denoted as w = on..210) shows the relationship between the elements of and the rectangular components of w.] To complete the description of rigidbody rotation. (5.) The linear and angular velocities at any point are related by Although in rigidbody rotation the vorticity is a reflection merely of the (constant) angular velocity. The positions of these points relative to the coordinate origin are given by . 2425). in complex flows it provides a useful measure of the local rotational motion at any point in the fluid.2).R . the fluid is rotating counterclockwise about an axis which is perpendicular to the page. t ) . respectively. Thus. we find that Because w is a constant vector and R does not vary in the direction of n. where n is a unit vector parallel to the axis of rotation. The velocity gradient.26) that for any pair of material points in a fluid undergoing rigidbody rotation the relative velocity is U f. 51 requires that n be directed toward the reader. 51. It follows that where is the rateofstrain tensor and is the vorticity tensol: The advantage of this decomposition of Vv is that. and r2. the relative velocities and relative positions of the two points are related by The second equality is a consequence of the fact that any components of r. (5. (5. One point is labeIed on each of two circular srreamlines. ' which is valid for any vector a. t) from v(r. The vorticity tensor f vector w according to r (a) : + Rewriting Eq. . (5. by analogy with Eq. . all terms vanish except that containing V.27) would be written as Vv = g(I' a). . (5. nonzero elements. We now make use of the observation that any tensor can be written as the sum of a symmetric tensor and an antisymmetric tensor (see Section A.R = 2. the counterclockwise rotation shown in Fig. Vv. the velocity gradient can be expressed as ' T v=wXR. Letting Av = v. Thus. so that Eq. as their names suggest. the symmetric (r)and antisyrnmetric parts are uniquely associated with the local rates of deformation and rotation.
.
there is the possibility of a net increase or decrease of momentum in the control volume due to convective transport of momentum across its boundary. conservation equations for scalar quantities in Chapter 2.+O. . (5. from Eq. What is desired now is to rewrite the lefthand side of Eq.37). so that the integral represents the total momentum of the fluid in the material volume ahd the lefthand side gives the rate of change of momentum. will be very useful for obtaining a differential form of the momentum conservation equation.:+v (pv)1 [Z +p I vv .53). If the surface bounding the control volume is assumed to deform with the flow.31) in integral and differential forms applicable to fluids. In other words.32) by the local fluid velocity. As in Chapter 2.5. (2. Rewriting the remaining term using the material derivative. A material point. (ASlo). Accordingly. which are reduced eventually to differential equations valid at any point in a fluid. as done here. we let u=pv. (5. (5. both of which are informative and will be discussed. The additional factor which must be considered with an arbitrary control volume is that any mass crossing the control surface carries with it a certain amount of momentum. Such a control volume is called a material volume. Newton's second law for a material volume becomes Being able to write the rate of change of total momentum in terms of a material derivative inside a volume integral. (5.38) applies not just to a material volume. Eq. respectively. from Eq. (5. (5. Newton's second law is written as L Combining Eqs. its surface and volume are denoted as SM(t)and V d t ) . (5. is simply a material volume in the limit V. but to any control volume. Thus.3 CONSERVATION O F MOMENTUM The linear momentum of a solid body with mass rn and translational (centerofmass) velocity v is rn< a vector. The second task is to characterize the forces exerted on a body of fluid by its surroundings. :.(pvv)=v a +v [. but the approach that adheres most closely to Eq.31) has been replaced in Eq. the momentum input due to the force F is equated with the rate of accumulation plus the convective losses. For an arbitrary control volume. which involves the product of the momentum concentration with the .. 1 h : This identity. then no mass will cross any part of the surface. TWO integral transformations from the Appendix are used to obtain the desired result. (A.2. The Leibniz rule for differentiating the v&e integral of any vector u(r.32) and (5.32). gives where F is the net force acting on the object. and chemical species derived in Chapter 2. the rate of change of momentum equals the net force. Thus. such that the surface velocity vs always equals the local fluid velocity v. The convective transport of momentum is represented by the surface integral. is Combining these two results for a material volume. The objective of this section is to rewrite Eq. which holds also if u is replaced by a scalar. the momentum balance is expressed as . Eq.32) in more useful form. To apply it to Eq. This will be shown by adopting a viewpoint more like that used to derive the The product pv is the concentration of linear momentum (the amount of momentum per unit volume).. (5. the derivations begin with macroscopic (control volume) equations. using local velocities instead of the centerofmass velocity. From the continuity equation. Newton's second law of motion is stated as The divergence theorem applied to the dyadic product vu. energy. the first bracketed term is identically zero. where vs = v.3I). (5. There are two basic ways to do that. Eq. as discussed in Section 5. The resulting equations describing the conservation of linear momentum are comparable in most respects to the conservation equations for mass. the control volume will always contain the same material. For a body of constant mass. For a material volume. is .35) simplifies to Momentum in a Material Volume Conservation of momentum may be expressed using any choice of control volume. t). Momentum in an Arbitrary Control Volume Of importance. is called Reynolds' transport theorem. The first task is to describe the rate of change of%e momentum of a body of fluid. leading to an expression for F. The centerofmass velocity in Eq. 3 .31) makes use of a control volume that always contains the same mass of fluid. The integrand on the righthand side is expanded now to obtain at (pv)+V.
As discussed in Chapter 1. in special circumstances. where we have made use of the fact that (pv)(v*n)= n. Equation (5.' . respectively. This completes our first main task. Forces which act directly on a mass (or volume) of fluid are called body forces. r. This will become clearer when F is evaluated.Surface forces are described using the stress vector s(n).316) becomes where S refers to the total control surface. (5. are made arbitrarily small in the limit indicated. as shown in Fig. we obtain the same result as before.). Control volume of thickness 2 t centered on an imaginary.311) becomes . (5.. .313).Conservation o f Momentum 219 net outward velocity at any point on the surface. the stress changes sign when the direction of n is reversed. s(n) is the stress exerted on the control surface by the surroundings.. then + This key result indicates that the stresses on any vanishingly small volume are in balance. . so that F = F. when n is the usual outward normal from the control surface. and S. If gravity is the only body force. pg. (5. Integrating the stress over the surface of m Y control volume gives s(n)= s(n). and S. is the gravitational force per unit volume at any point in the fluid. what remains is to evaluate F. K v = v. (2. so that Eq. planar test surface within a fluid.For a sufficiently thin control volume (i.. V / S = O(t) as (0.39) does not involve any doublecounting of momentum transfer.312) and (5. (5. (5. and therefore it describes a stress equilibrium which holds in the vicinity of any point.e.. The key conceptual point is that the stresses which contribute to F correspond only to difSusive fluxes of momentum.220). 4?+0). The integrands in Eq. the convention used in this book is that s(n) is the stress exerted by the material that is on the side of the surface toward which n points.314) is rearranged first to obtain I For any control volume with a characteristic linear dimension t .39) is rewritten as Using eqs. In essence.. (5. this is just a manifestation of the actionreaction law of mechanics (Newton's third law of motion). which is a macroscopic (integral) statement of conservation of momentum for any control volume.315) remain finite in this limit. and therefore contribute to Fs. (5.32).?: . The volumetric or body force includes contributions from gravity and.1 8) Thus. Applying the Leibniz formula and divergence theorem once again.3. (5. where g is the gravitational acceleration.g.36). 53.39) reduces to Eq. [This amounts to a sign convention for s(n).. The unit n o d s directed toward sides 1 and 2 are n and n. electrostatic forces are discussed in Chapter 11. Evaluation of Forces The externaI forces acting on a body of fluid are of two general types.II test surface 1 I I . Thus. as shown below.:~: Figure 53. Fs.] Thus. The net body force and net surface force on the control volume are denoted as Fv and Fs. (5. Some important properties of s(n) are revealed by applying the integral form of conservation of momentum to a control volume of small size. The simplest of those is seen by considering a control volume of thickness 2 4 centered on an imaginary planar surface within a fluid.(pvv). Eq. from other fields (e. We confine our attention here to gravity.L . so that the control volume is a material volume. whereas those which act on surfaces are expressed in terns of stresses (forces per unit area). so that for a sufficiently small control volume the righthand side vanishes and we find that for any control volume. respectively. I..the stresses integrated over the edges will be negligible compared to the contributions from the top and bottom surfaces (S.. then Eq.. Eq. which are the same. 8. ensuring that the surface integral in Eq. The integrand.an electric field acting on a fluid which has a net charge). Viscous forces and pressure act on surfaces. which is the force per unit area on a surface with a unit norma] n. Expanding and simplifying the integrand as in Eq. so that we conclude that at any point in a Auid. The areas S. (5. Stress equilibrium at a point has a number of consequences. The analogous result for the flux of a scalar quantity is Eq.
Using Eq. whereas the fourth is oriented in a direction described by the unit normal n. That is. Thus.326). (5. Accordingly. Switching from (x. Dr (5. force per unit volume).326) Applied to this tetrahedron. z) to the indices (1. we conclude that p Dv =pg+V.. Eq. u .324) for linear momentum. Three of the surfaces are normal to the rectangular coordinate axes. the integral form of conservation of linear momentum becomes Applying the divergence theorem to the surface integral and collecting all terms in a single volume integral. (1. mass is expressed as a density and the forces are also expressed on a volumetric basis (i. we will have little further need for the stress vector. 2. it is the mass times the acceleration.314).) = s(ei) for i = 1 .e. Because of the pointwise nature of Eq. AND VISCOUS STRESS The main task which remains in stating the basic equations of fluid mechanics is to relate u t o the dynamic conditions and to the properties of a given fluid. ( 5 320) It follows that the term in square brackets must vanish at any point in a fluid. or 3. 54. Symmetry of the Stress Tensor An important characteristic of the stress tensor is that.) as uij. Conservation of angular momentum for a material volume is written as Using h s definition in Eq. (5. As introduced in Chapter 1 [see Fig. in the context of fluid statics. The lefthand side is the rate of change of momentum at a material point.. as shown in Fig. (5. provided that the only body force is gravity. Consider a small. and z surfaces are the projections of So given by '[ k L' t Because Eq. (5. we define the jth component of the vector s(e. namely. y. It is seen that a knowledge of u at a given point in a fluid is all that is required to calculate the stress for any specified orientation n. the stress vector s(n) is expressed as the dot product of n with a sum of three dyads of the form eis(ei).28).316) becomes This relation is valid for any fluid which behaves as a continuum.320) that 5.~. s(ei) = i .325) applies to a control volume of any size. .)s(. That is.5. Thus.) +(n e. y. tetrahedral control volume. which was stated previously (without proof) as Eq. we obtain Another major consequence of stress equilibrium concerns the calculation of the stress vector for a surface with an arbitrary orientation. we conclude from Eq.)s( lim S ex)+ (n. 3) and noting from Eq.)s( .6. with exceptions for a few special fluids. ej . (5. 12 and Eq. If the area of the arbitrarily oriented ("n") surface is S.e. (5. it is symmetric. as will be shown now.318) that s(e. The righthand side is a sum of forces. however small. uu=uji for all i and j. e.323) to evaluate the stress vector in Eq. the integrand must vanish at every point. Certain general characteristics of the stress are discussed in this section. S+O so [s(n)+ (n e.321) results in where a is the stress tensor: What we have just done is to derive the relationship between the stress vector and stress tensor. That is. in a reference frame moving with the fluid. then the areas of the x. Constitutive equations which describe the viscous stresses in particular types of fluids are presented in Section 5. (1. The rate of change of angular momentum is equated here to the net torque computed from the moments of the gravita . the symmetry of IT and the decomposition of the total stress into pressure and viscous contributions.4 TOTAL STRESS.2lo)]. 2. This symmetry is a consequence of the conservation of angular momentum.e.Figure Smss tetrahedron. (5. Pressure forces are explored further in Section 5.. (5. )]= 0. (5. PRESSURE. k k which is analogous to Eq.
.
(5. ~ a$ r 1 =Ppr$+ p [ ~ + v ~ ~ + v y ~ + v z  a~ sin @+[ 7 $(r2Trr) + rsin8 as r sin 0 a+ r 1 a a I cote r z component: ax ay a~ =pgzaP+ %+%+%] az [ 8 component: [ax ay az 1 a 1 a 1 ~T+o? =pg. (5. Z + V .51) to evaluate the pressure gradient. For example.326).I. Eq.52) In terms of the position vector r. Situation (b) has the advantage that we can apply the divergence theorem to the integral in Eq. we find that I 8 component: 2 k r\ /. in the direction parallel to the gravitational vector). The components of this equation in rectangular. is the (constant) density of the fluid. (5. (5.ldP+ . Eq. V . The net pressure force on this object. (5. Fp. (5.P ~ Z . and spherical coordinates are given in Tables 51. If the density is constant. o ) + ~ 8 a e ( ~ s s i8)+n r ae [?dr rs~nea+ r The general statement of conservation of linear momentum. as shown in Fig. (5.53) 5. &' .( r 2 ~ . 55(a). Example 5. Thus. if the + z direction is taken to be "up.+ vr ar r ae avr r sin 6 vrn a". conservation of linear momentum implies that T= 0 in r i. [Jcomponent: (5.5. consider an object of arbitrary shape submerged in a constantdensity fluid. (5. is written now as F i 4 component: =pgrI %+ 1 (r2Tr+)+d i +asw 1 a r ++~ T& 2 cot 8 r sin B am [r 2 t r d 9 rsinOd4 r r which is called the Cauchy momentum equation. the upward force on the object due to the pressure equals the weight of an equivalent volume of ffuid. cylindrical. the pressure in a constantdensity fluid at rest is given P(r) =P(O) + p g . . Setting v = O and Eq. Because the pres. The presswe force is evaluated most easily by considering the situation in Fig. In particular. where the solid has been replaced by an identical volume of fluid.is given by VP=pg.49).54). by (5. this replacement of the solid does not affect the pressure distribution in the surrounding fluid. P(z)= P(0) . where the minus sign reflects the fact that positive pressures are compressive (i.e. and TABLE 52 Cauchy Momentum Equation in Cylindrical Coordinates r component: .sure in the fluid depends on depth only. 52. this is Archimedes' law..' 1: . because P is a continuous function of position within the volume V.55) when p.e. dPldz = pg. To calculate this force.1) is integrated readily to determine P(r). where P(0) is the pressure at the origin.51) b This expresses the elementary fact that the pressure in a static fluid varies only with height (i.. this is not necessarily true for (a).51 Buoyancy A consequence of the fact that pressure increases with depth in a static ffuid is that the pressure exerts a net upward force on any submerged object.55) is that Fp is independent of the absolute pressure. That is. because we have said nothing about the meaning of P within a solid. r .Huid Statics TABLE 51 Cauchy Momentum Equation in Rectangular Coordinates TABLE 53 Cauchy Momentum Equation in Spherical Coordinates r component: y component: [$+ av dv." then g = ge. the pressure distribution on the fluid control surface in (b) must be identical to that on the solid surface in (a)..5 FLUID STATICS The only mechanical variable in a static fluid is the pressure. An important implication of Eq. pressure acts in the n direction).. respectively. adding any . and 53. 55(b).. Applying the divergence theorem and using Eq.
pp. can be proven formally by considering what is necessary to ensure material objectivity. 56.. but which do not depend directly on r.. or T = ~(l?). In this section we state the constitutive equation for a Newtonian fluid and also provide a few examples of constitutive equations for nonNewtonian fluids. This is the principle that the form of a constitutive equation must not depend on the position or velocity of an observer. Figure 56. and therefore must be invariant to tirnedependent translations or rotations of the coordinate system (Leal.312). when considering flow caused by spatial variations in fluid density.. S. (5. and setting n. responds instantaneously to changes in the local rate of strain.61) are indeed linear functions of r. equals the pressure force on S2 plus the weight of the intervening fluid.The divergence of the velocity (Vv) is the sum of the diagonal elements of I? (sometimes called the trace of I? and written as trr).e. Viscous stresses are frictional in character. constant to P does not affect the net pressure force on a completely submerged object.g= +g). (5. whereas K is the dilatationul viscosity or bulk viscosity. Pressure force on a submerged object. if P is constant. e.. 55 and noticing that the pressures acting normal to the top and bottom (or front and back) surfaces in the figure make no contribution to F. where it is desired to compute the x component of the pressure force on the solid surface. exg=O) or if the weight of the fluid is negligible.52 Pressure Force on Part of an Object: Projected Areas When only part of the surface of an object is in contact with a fluid. the buoyancy force is found to be ! 5. We return to the concept of buoyancy in Chapter 12. is the unit normal directed toward the surface. so that the x direction is not necessarily horizontal. and has a local viscous stress which is a linear function of the local rate of strain.55) to the control volume shown by the dashed lines in Fig. Denoting the desired force as F.510) indicates that no matter how complex the shape of S. is motivated by the demonstration in Section 5. and many examples of such fluids were mentioned. If x is directed downward (i.. Consider the situation shown in Fig. in Eq. If x is horizontal (i. Applying Eqs. As shown. The control volume in @) is used to calculate the pressure force on the object in (a). we have where n. in that they reflect the molecularlevel resistance to attempts to move one part of a fluid relative to another. The coefficients have been selected so that p is the shear viscosity discussed in Chapter 1. as is approximately true for a stagnant column of gas of moderate height. Thus. pp. relative to g is considered to be arbitrary..6 CONSTITUTIVE EQUATIONS FOR THE VISCOUS STRESS Constitutive equations are necessary to relate the viscous stress to the rate of strain and to material properties such as viscosity. the type of fluid motion which matters in this regard is deformationthe pulling apart or pushing together of material pointsand not uniform translation or rigidbody rotation. The idea that the viscous stress tensor should be a function only of the rateofstrain tensor. The defining characteristics of a Newtonian fluid are that it is homogeneous and isotropic.2 that I? measures the local rate of deformation of a fluid.so that both terms in Eq.. we obtain where (P)2 is the pressure averaged over the surface S. 105112). it is seen that the pressure force on S . .e. The buoyancy force is the net force on the object due to the static pressure variations and gravity. 3948). the most general expression for ~(r for ) a fluid with these characteristics is Identifying the first integral with F.. there is in general a net pressure force on that part of the surface.54) and (5. and not simply Vv. This fact greatly simplifies force calculations in statics problems involving curved interfaces. for example. Example 5. then Fp =0. the pressure force can be obtained simply by multiplying the projected area (S2) by the corresponding average pressure. Moreover. by Aris (1962. Use of the projected area to calcu Solid Fluid E i 6 ! I Equation (5.. The fact that T must depend on r.. The orientation of e. then where p and K are scalars which may depend on state variables such as temperature. (5. it is found that Newtonian Fluids The concept of a Newtonian fluid was introduced in Chapter 1. Evaluating the gravitational force on the solid body by setting p=p.= ex in the second integral. 1992.Constitutive Equations for the Viscous Stress 225 Fluid P = Pf = Pf Figure 55.
pp.= v. v = 2 (r2v$+i rsin (vg 1 a r ar % .28). 253255). 3 1 'c='r+=P[= avr G+ %( a e a8 sin @+. and it is generally believed that for other fluids K << Moreover. (5.e. 55.61) that is multiplied by K vanishes. Of interest is the extent to which P differs born (minus) the mean normal stress. Setting K = O in Eq. 1968. which may be used as a purely mechaniTABLE 54 Viscous Stress Components for Newtonian Fluids in Rectangular Coordinates Using Eqs.. Batchelor. = 0. (5.g.62) is given in component form for the three common coordinate systems in Tables 54. If defined in this way. (5. 141). monatomic gases. An analysis of dilatational viscosity in a liquid containing gas bubbles is given in Batchelor (1970.64) / where we interpret P as the thermodynamic pressure. (5. the viscous stress is simply cal definition of pressure (e.. the total stress is related to the pressure and viscous stress by a= PS+ T. the pressure (F) is given by Note from Table 54 that if v.64) to evaluate the normal stress components for a Newtonian fluid. spherically symmetric) expansion or compression. the constitutive equation for a Newtonian fluid is given by TABLE 55 Viscous Stress Components for Newtonian Fluids in Cylindrical Coordinates where r is defined by Eq. and 56. where V v = 0 from continuity.6I). The effects of K on fluid dynamics are difficult to detect and are usually ignored. for example. we find that.61) and (5. For these reasons we will be concerned only with p in the anaIysis of flow probIems. It is related to the rate of energy dissipation in an isothermal fluid under these conditions (Schlicting. the term in Eq. It has been shown that K = O for ideal.Constitutive Equations f o r the Wscous Stress 227 The dilatational viscosity measures the molecularlevel resistance to an isotropic (i.(y) and v. TABLE 56 Viscous Stress Components for Newtonian Fluids in Spherical Coordinates  r sin 0 +'+at$ r r . Equation (5. = v.63) reduces to Eq. p.r sln 8 a4 1 a v . pp. As discussed in Section 5. (1 212). Eq. 5760).4. (5. For an incompressible Newtonian fluid. for incompressible fluids. 1970. (5.
. (5.r +p dB lap a i a [a r ( r. The resistance of longchain polymers to orientation or elongation gives polymeric fluids an elastic character. C I i r Non. The dependence of the viscosity on the rate of strain is the major concern in many applications.69) and (5. Using Eqs.v for those coordinate systems is shown in Tables 54. Eqs.+ p aP ay [ ++a 2 ay2 a2vY a2v. incompressible.constitutive Equations for the Viscous Stress 229 Substitution of this and the analogous expressions for up and uz in Eq. 58.' This expression of conservation of momentum.K(Vv). whereas V . The NavierStokes equation is given in component form for the three common coordinate systems in Tables 57. isothermal fluid are just v and I! Counting the three scalar components of v. A dependence of viscosity on rate of strain may reflect the ability of the flow to orient particles or macromolecules. Stokes(18191903) in 1845. andlor to influence the conformation of individual polymer molecules. Among the distinguishing characteristics of various nonNewtonian fluids are a dependence of the apparent viscosity on the rate of strain. P and are identical for a stagnant fluid. Bird et al.4. it is apparent now why the factor was included in Eq. (1987). 55. (5. and "memory" ef' ' h s quation was reported independently by the French physicist L.49). Noting that the NavierStokes equation has three components. 409414). whicharediscussednext. is beyond the scope of this book. . certain polymer solutions.Theana1ysisofviscoelastic phenomena. Pearson (1985). which is manifested as an additional tension along streamlines (i. unusually large values of the normal components of the viscous stresses. an incompressible fluid. In general. ' ' . Because K is ordinarily quite small. G . The resulting equations describe what are calledgeneralizedNewtonianjluidr. (5. (5.411).a2v p . we find that fects in the relationship between the stress and the rate of strain.e.dr (me) )+2 do2 ' 7 z+$ I avr a2v z component: . (1987) is the source for most of the information presented below on generalized Newtonian fluids. T I z component: aP a 2 v .v + ~ ~ %y+ a.. For discussions of that and other aspects of polymer rheology see Bird et al. together with the continuity equation written as : provides the usual starting point for the analysis of the flow of gases at moderate velocities and of simple liquids. . ] .65) results in TABLE 57 NavierStokes Equation in Rectangular Coordinates x component:  r P = P . an elevated normal stress in the flow direction). TABLE 58 NavierStokes Equation in Cylindrical Coordinates r component: . 8 component: = p g . including normal stress and memory effects..610) are seen to provide four partial differential equations in these four unknowns. there is little distinction between P and P for Newtonian fluids. (5. and which in turn influence the relationship between the viscous stress and the rate of strain. such fluids have structural features which are influenced by the flow. y %+vT 21 . conservation of linear momentum is written as which is the Navierdtokes equation. aZv. and 59. Substituting this result in Eq. (5. muids which exhibit elastic characteristics and have finite relaxation times are called viscoelastic.p g ... or a fluid where K = 0. and suspensions of nonspherical andfor strongly interacting particles. there are a total of four unknown functions. The finite time required for such molecules to reach an equilibrium configuration is the source of memory effects.additional information on Stokes is given i n Chapter 7. (5. Examples include polymer melts. pp.z + p [ e x Z + d + ~ [da:: NavierStokes Equation An important special cace is that of a Newtonian fluid with a constant density and viscosity. Navia (17851836) in 1822 and the Anglolrish physicist G.63) and (A. or Tanner (1985). and 56.Newtonian Fluids Any fluid which does not obey Eq. .61): It is needed to ensure the (near) equality between P and y component: p ["d. Incidentally. As mentioned in Section 5.67) Thus. A discussion of Navier's contributions is in Dugas (1988. and it can be modeled using fairly straightforward modifications of the constitutive equation for a Newtonian fluid. the unknowns in an analysis of the flow of a pure. to break up particle aggregates. .+ v x ax ~ + v y ay~ + v dz ~ ~ ] = p g z .61) is nonNewtonian..
111 k . I el. I  d n 2 r h l L .. b .
At the other extreme of large rates of strain. Thus. This is because gas viscosities are relatively small. instead of the NavierStokes equation. (2. as shown in Fig. for an incom? pressible Newtonian fluid. "inert" means that the solid is not undergoing a phase change. this and all other interfacial conditions discussed in this section apply locally at each point on the interface. It is important to recognize. (5. This and the continuity equation for an incompressible fluid [Eq.24.v. Eq.6lo)] provide the governing equations for a generalized Newtonian fluid. followed by stresses at interfaces where surface tension effects are absent. If one is interested only in the liquid flow. n . unless the rates of strain in the . If t is any unit vector tangent to the interface. This will be referred to as the nopenetration condition.v. vn = O in the fluid contacting the solid surface. is that the velocity components tangent to a fluidsolid or fluidfluid interface are continuous. (5. the stress at the interface with a flowing liquid is not usually .7 FLUID MECHANICS AT INTERFACES In this section we consider the behavior of the velocity and stress at fluidsolid and fluidfluid interfaces.= t v n . Thus. !. Nonetheless.dynamic problems. In analyzing liquid flows it is often possible to neglect the shear stresses at gasliquid interfaces. f I I . ing flow in the gas. there is ordinarily no relative motion or "slip ' where the materials contact one another. where n is normal to the interface. (5. toliquid viscosity ratio being : : gas greatly exceed those in the liquid. whereas in mechanics it is viewed as a force per unit length. Tangential Velocity An empirical observation.tion occurs for 7 1 rO. some of the effects of surface tension are illustrated using examples involving static fluids. This uncoupling is discussed in Example 6. for a stationq. The equality of pressures is a good approximation also in many fluid. provided that surface tension is negligible. these interfacial conditions become 5. with a typical gas(Chapter 1). The matching of tangential velocities on the interface between materials 1 and 2 is written as 7 and is referred to as the noslip condition. the viscous stress is substituted into the Cauchy momentum equation to give Stress Balance At an interface where surface tension is negligible.72) : .72). (5. such that 2hr>> T . The equations satisfied by velocity and stress components which are tangential to or normal to an interface provide the boundary condtions for fluid mechanics problems. = s(n)I2. Accordingly. the imaginary test surface in Fig. the component of the total stress normal to the interface (unn = n . : . impermeable.smalI relative to other viscous stresses in the gas. (5. Eq. (Although not indicated explicitly. a number of flow problems involving generalized Newtonian fluids in simple geometries can be solved analytically. between two fluids. In thermodynarnics it represents the energy required to create new interfacial area. however. Whichever of the generalized Newtonian models is used. = t .16. that under these same conditions the shearfree interface is ordinarily nor a good approximation when analyz: . This result applies at any point on a flat or curved interface. In this context. then the corresponding tangential component of velocity is v. in that it uncouples the Aow problem in the liquid from that in the gas. . Notice that for static fluids. The dependence of p on I? makes the viscous term in Eq. In terms of the pressure and viscous stresses. (5. In particular.5lo).a) and any component of stress tangent to the interface (un. . Velocities are discussed first. That is. :. To satisfy Eq.=n. treating the interface as shearfree results in major simplifications. Applying conservation of linear momentum to this control volume gives Normal Velocity As shown by Eq.a ) must each balance. acting within the plane of the interface. r Stress Balance with Surface Tension The surface tension (y) at an interface between two immiscible fluids is a material property which has units of energy per unit area or force per unit length. .) At an interface between a fluid and a stationary solid.73) reduces simply to an equality of pressures. and inert solid. The effects of surface tension on the stress balance at an interface are derived by considering a control volume of thickness 2e centered on an interfacial area S. each component of the total stress must be in equilibrium. in that the normal viscous stresses are often small or even zero. the shear stresses exerted by the gas on the liquid will be small compared to the viscous stresses within the liquid. As will be shown.616) nonlinear. where n is the unit normal) is governed by conservation of mass. Finally.. the component of velocity normal to an interface (v. verified in countless situations. 53 being replaced by an interface. s(n). ' '. That is. it can affect the mechanics of a fluidfluid interface if one or both of the following conditions hold: if the interface is nonplanar or if the surface tension varies w i t h position. The proof of this is basically the same as that given for Eq.71) reduces to the frequently used requirement that v. (5. .=O in the fluid contacting the solid surface. and then the modifications to the stress bal• ances required by surface tension. the fluid becomes Newtonian with a viscosity k . 57.318). Problem 54). like those discussed in Chapter 2. the normal viscous stress at any solid surface is zero (see .
g a % 5 I II 2 a .aEo a E ' Z d L O C F "28x g' 2s 2 8 3 8 a7 2 I .* ( I ) g z .%  + r % Q 9 4 ~ T I @ I1 g 35  0 E.g 8 % 4 8 s O F 8 2.5S + g c 0rL &@g &'.
.. The solution (5. it is found that h(x) must satisfy Example 5.715) requires that. are constants within the respective phases.52). P.72 Free Surface Next to a Wetted Wall Consider a gasliquid interface near a wetted.711 ) Assuming that static pressure variations within either fluid are negligible on the length scale of the bubble or drop. Thus. (5. the maximum height of the interface for small e is &A.715) indicates that the characteristic length for this problem is which is called the YoungLaplace equation.721) h(x)= + O(e3).. and h. The nonlinearity of Eq. rather than being specified in advance.e. as expressed by Eq.71 1) can be satisfied for constant P. for weakly wetting surfaces. (5. Ignoring the normal viscous stresses. and n points outward. venical surface.711). h. the liquid pressure at the gasliquid interface is where the constant term (Po) has been evaluated from the requirement that the gas and liquid pressures be equal at x = m. =e. The height of the gasliquid interface is described by z = h(x). and P. . The contact angle (a) requires that the liquid rise near the surface.711).Fluid Mechanics at Interfaces "Wetting' 237 'Nonwetting" Liquid \ r Gas This is an example of a free surface problem. The coefficient on the righthand side of Eq. = 0. (5. (5. (5. it is found that h. It is assumed that the gas pressure is constant.824)] Solid Solid Figure 58. 59.79) reduces to with the boundary conditions PIP2= 2 X y. . h(x) be determined by numerical " integration. we obtain The slope of the gasliquid interface at the wall is dictated by the contact angle..E with E > 0 and e << I . = 0. h. Using Eq. an analytical solution is obtained fiom a regular perturbation analysis. The dashed line indicates the tangent at the threephase contact point.716) is expanded in a Taylor series about the angle ~ 1 to 2 give and the interface height is expressed as Gas Figure 59. The normal stress balance again reduces to Eq.Substituting the expressions for the curvature and pressures in Eq.71 Pressure in a Bubble or Drop The objective is to determine the pressure difference across the surface of a gas bubble or a liquid drop immersed in some other liquid. in whlch the location of an interface is determined by certain static or dynamic conditions. This is the reason that the equilibr i u m shape of a small bubble or drop is spherical. Thus. such that P. Example 5. which is called the capillary length. up to O(e3)is + 0. (5. Gasliquid interface near a vertical wall. for a sphere of radius R the mean curvature is a constant. As mentioned above. in general.n>O) and the surface is described as z = h(x) only. Notice that Eq. The interior of the bubble or drop is denoted as fluid 1 and the surrounding phase as fluid 2. where the interface is flat and h = 0. such that cr = ( n l 2 ). whereas the static pressure in the liquid varies with height. P. the curvature is given by [see Eq. (5.6. However. Eq. X = 11R.(5. = Po. neglecting any effects of fluid motion. Using the methods of Section 3. . When n is directed upward (i. as shown in Fig. for both wening and nonwetting behavior. (A. (5. and y only if the curvature is independent of position. (5. # 0. n. It is desired to determine the function h(x) under static conditions. The cotangent in Eq. Contact angle (a)at a gasliquidsolid boundary.716). where x is distance from the wall and h(m)=O. where now fluid 1 is taken to be the liquid and fluid 2 the gas.
Accordingly. Thus.g. this approach is extremely useful for analyzing both confined flows and flows around submerged objects. Conservation of momentum for a Newtonian fluid is ensured by rewriting the NavierStokes equation in terms of $(x.81) for a steady flow leads to The rationale for this choice is seen by substituting Eq. (5. 5. As illustrated by many examples in later chapters. This includes flows which have a planar character and flows which are axisymmetric. (5.51.94) by x and then subtracting one equation from the other. For instance (as shown in Example 6. When there are density variations or body forces other than gravity.9 STREAM.84). where 9 is constant. the stream function $(x. because the actual pressure is needed to determine the location of the gasliquid interface (as in Example 5. Using the dynamic pressure. In other words.l Stream Function 239 5. (5. The pressure force (Fp) on an object of volume V which is completely submerged in a static fluid was calculated in Example 5. without an): free surfaces) it is advantageous to combine the pressure and gravitational terms in the NavierStokes or Caucby momentum equations by using the dynamic pressure (sometimes called the modijied pressure or equivalent pressure). when possible. The usual practice in problem solving is to select a reference value of 9 = 0 at some convenient location. (5.the pressure at the origin. In such cases pressure and gravitational effects cannot be fully combined. The dynamic pressure (9) is defined as sure to the drag or lift forces on a submerged object are computed using 9.83) reduces to Eq. Reversing the signs in Eq. in that case 9 + c = P(O).91) into the corresponding continuity equation for an incompressible fluid. Thus. one solution suffices for all spatial orientations. The result is where S denotes the entire (closed) surface of the object. The first term (pgV) was derived previously from the static pressure variation. The velocity m s n t s in the first and last terms are related to the stream function by . whereas the integral involving 9 gives the additional pressure force due to fluid motion. Indeed. In this section the equations which govern the stream function are derived. the stream function is defined in such a way that if +(x. The value of P at any position r is determined only to within an additive constant. so it is not surprising that some authors use the opposite sign convention. y) to satisfy continuity.. It is seen that even if the absolute value of P is known at some location (e.93) by y and each term in Eq.53). leading to Archimedes' law.83). is that problems can then be solved without reference to the direction of gravity. the interpretation of 8 is less simple (see Chapten 1l and 12). the parabolic velocity profile in a fluidfilled. (5.82) are where c is a constant.72). the continuity equation will be satisfied automatically. (5. integration of Eq.2I). r = O). 1 i so that the NavierStokes equation becomes The stream function is a mathematical construct which is extremely useful for solving incompressible flow problems where only two nonvanishing velocity components and two spatial coordinates are involved. spatial variations in 9 are always associated with fluid motion. because 9 is defined only in terms of its gradient. adding any constant to P or 9does not change the values of the integrals in Eq. the x and y components of Eq. Eq. In addition to being a useful vehicle for deriving the velocity and pressure fields in such problems. and then the physical interpretation of the stream function is discussed. For a static fluid. (5. the stream function itself provides information which helps to visualize flow patterns. y) is defined by .51. For a twodimensional flow described by rectangular coordinates. such as those involving wave dynamics or flow in open channels. Using Eq. not P: As shown in Example 5.. y) can be determined for a given flow. the contributions of PITS . the arbitrary nature of c allows the value of 9 there to be chosen Freely. (5. y). ' It is evident from Eq.FUNCTION . the corresponding result for a Bowing fluid is ' The pressure is eliminated by differentiating each t e r m in Eq.91) does not alter the ability of $(x. (5. (5.8 DYNAMIC PRESSURE In solving flow problems involving incompressible fluids with known boundaries (ie. paraIlelplate channel is independent of whether the channel is horizontal or vertical. The modified pressure loses its utility in freesurface problems. (5. The advantage in combining the pressure and gravitational terms.82) that V 8 = 0 in a static fluid. Thus. when the NavierStokes equation applies. the reference value chosen for 9 has no effect on drag or lift calculations.
.
.
.'..107) and the viscous terms have a coefficient of unity. (5. The remainder of this section is devoted mainly to the significance of Re and Sr for the analysis of various types of flows. Approximations Based on the Reynolds Number: Steady Flows y=O The physical significance of the Reynolds number is clearest for steady flows. and pressure scales which are suitable for a given flow. His experimental observations on water flow in pipes (Reynolds.' 20sbome Reynolds (18421912) was a professor of engineering at Manchester. They may be interpreted also as representing convective transport of momentum (see section 5. 1883) established the importance of the dimensionless group which would be named in his honor and thereby provided the foundation for the concepts of dimensional analysis and dynamic similarity in fluid mechanics. The increment in # (or c) between any two streamlines is 0. . y. '.: . As discussed in later chapters. The dimensionless continuity equation. $ . Inertial (or convective) effects tend to be prominent when Re>> 1 and nearly absent when Re<< 1. Streamlines for the twodimensional flow in Examples 5.. I'! (5.:. are referred to as the viscous terms. <$ and the Stmuhal number. The pressure scales appropriate for various conditions are revealed by considering extreme values of Re. The characteristic velocity (LI) is representative of the maximum velocity.. England. '\? J. in tube flow. Alternatively. velocity.. .:' The continuity and NavierS tokes equations are nondimensionalized by introducing length.: To facilitate the discussion of approximations. which have their . Eq.104) reduces to x = 1 x=o x=i Figure 511. (5. :::'. Re is evidently a measure of the relative importance of inertial and viscous effects in steady flows. Because Re is the coefficient of the inertial terms in Eq. The terms containing the second derivatives (V2v).:.. ' ..$. (5. An approximation to Eq. The most natural choice for that coefficient is unity. viscous transport of momentum is essentially a difisive process. the material derivative of the velocity has been separated into its temporal and spatial parts. which descnbe the fluid acceleration in a material reference frame. 3 the continuity and NavierStokes equations become %\ 'q. . as discussed below. ~ 4 '2. L is the tube radius or diameter.. The characteristic time ( r ) and characteristic pressure or stress (n)depend on the particular dynamic conditions. b i . among other contributions. . The nonlinear terms (v*vv). :' ! ? .. The velocity terms in Eq. so that the pressure scale for small Re is PU n=.. .109) .92.. . origin in the description of the viscous stresses. . .103).104).107) fall into two categories. .Nondimensionalization and Simplification of the NavierStokes Equation "< 245 7 * . If all variables are scaled properly. . With dimensionless variables and differential operators defined as . Reynolds was also responsible for the basic theory of lubrication and for the averaging pmcess which underlies the statistical analysis of turbulence. (5.3). . (5.91 and 5. . time.. are called inertial tams.'. The characteristic length (L) is a distance over which the velocity changes appreciably. it is a measure of the relative importance of the convective and diffusive transport of momentum. has the same form as its dimensional counterpart and requires no further comment. . '. The dimensionless groups which : : T. Dimensionless Equations ' . Eq. for which Eq. As !' discussed in Chapter 1.107) for Re+O is 3.. are the Reynolds n ~ r n b e r . . then the coefficient of the pressure gradient cannot be large or small (see Section 3. appear now on the the lefthand side of the NavierStokes equation. I : . for example.2. L z. (5.' k?::' > .I.2).
.
Engineering Rheology.. i Show that for an incompressible. E. (5. However. one velocity component is given by where C and m are constants. New York. R. BoundaryLayer Theory. 1985. Philos.71. Rosensweig.h p o r t o f vorticity Soap contains surfactants which lower the surface tension of an airwater interface. as well as the YoungLaplace equation.. Brenner. Cambridge University Press. R.. 1 . Leal. E.) The questions below pertain to an incompressible. Eds. Eng. It is sufficient to prove this for planar. (Lond.1 . and 0. 1989). In Encyclopedia of Physics (Handbuch der Physik). and D. 1959. 1985. Oxford. L. 1. Tanner. Dover. Elsevier. I. D. ButteworthHeinemann. Armstrong. Soc. Scriven. No. Interfacial Transport Processes and Rheology. Normal Viscous Stress at a Surface : . Pearson. that is.) (b) Show that for any planar flow. Schlicting. An experimental investigation of the circumstances which determine whether the motion of water shall be direct or sinuous. Sci 12: 98108. Boston. R. R. Mathematical principles of classical fluid mechanics. Newtonian fluid. =l? (Hint: Use the stress tetrahedron shown in Fig. Tensors. G. (c) Derive the result in (b) from the stream function equations. Semn. S. where R is the inner radius. e. General consideration of angular momentum and polarization. ' ' (a) Assuming that the fluid is incompressible and that v. Laminar Flow and Convective Transport Pmcesses. Fliigge and C. h n d . Dynamics of Polymeric Liquids. Clarendon. threedimensional flow). / . 8.= P. . 1987. L. all are assumed constant. determine v. ButterworthHeinernann. if at some point in a fluid P. Newtonian fluid. Dahler. S. are the pressures acting on imaginary test surfaces normal to ex. 0) =Po. G.1020) indicates that no such simpIification is possible for flows at moderate or large Re when Sr = 1. Wiley.(x. Isotropic Property of Pressure Given that pressure is the only stress in a static fluid and that it acts normal to all surfaces. R. Bird. The wall of a soap bubble consists of a thin film of water between two such interfaces. R. . Soap Bubble '. k' is the thickness of the water film. 53. Fermhydrodynumics. PrenticeHall. Batchelor. and L. New Yo* 1968. the vorticity vector satisfies It is appwnt that in such situations creeping flow will remain pseudosteady. Reynolds. 54. can be obtained also by pelforming a force balance on a hemispherical control volume corresponding to onehalf of the bubble. cylindrical.(x.)A 275: 504527. I. and P is the pressure acting 0" a test surface with an arbitrary unit normal n. twohmensional flow in a region defined by xy 2 0. A. 0) = 0. SpringerVerlag. Vol. NJ. Note that this vorticity equation is like the conservation equation for any other scalar quantity. and the Basic Equations of Fluid Mechanics. K.(a) Show that in general (i.e.. Dugas. but with no homogeneous source term. London. Van Dyke. show that it must be isohopic. 1992. Theory of structured continua. and of the law of resistance in parallel channels. 1883. E. 1960.1970. I. the result in (a) reduces to References Aris.. That is. sixth edition. An lntroduction to Fluid Dynamics. respectively. The pressure inside the bubble is P . . show that P. A. (Hint: Use the continuity equation. (a) Using the pointwise normal stress balance at an interface. (5. B. C. A History of Mechanics. Cambridge. New York.79). Cam bridge University Press. (5. H. Vol. J. Hassager. New York. McGrawHill. 1985. 1963. Trans. Wasan. . [This is a reprint of a book originally published in 1955.. y). assuming that P(0. and the surface tension is y. A 174: 935982. Cambridge. Berlin.1 Edwards. Mechanics of Polymer Processing. CA. Stanford. Englewood Cliffs.. and it is assumed that t < < R . Eq. Dynamics of a fluid interface. Thus. 1988. 0. This problem is concerned with the equilibrium size of a spherical soap bubble.1019) will reduce once more to Stokes' equation. Eq. Scriven. Proc. 1962 (reprinted by Dover. the external pressure is Po. 1982.) 65. R. 9 ( x . (Hint: Use the results of Example 5. T. Soc. 1 (b) Show that the relation in (a). H. R. vorticity is created only at surfaces. Chem. Pr and P. Truesdell. pp. M.=P. 54. (Hint: Begin by taking the curl of each term in the NavierStokes equation.. solid surface. An Album of Fluid Motion. R. and e. . second edition. show that This approach is like that used in Example 5.) 52. y). and spherical surfaces. for a timedependent. Vectors.52. Velocity and Pressure Calculations '0 and Assume that for a certain steady. 1991. 125263. Boston. I ' Problems 51. . Parabolic Press. Eq. J. the normal component of the viscous stress vanishes at any impermeable. /' (b) For what values of m will the flow be irrotational? (c) For what values of m can the NavierStokes equation be satisfied? For those cases determine the dynamic pressure.
. As used here.. it is the number of directions which tends to be the k most crucial. Stokes equation. in general. In seeking solutions to flow problems involving incompressible Newtonian 6 (or generalized Newtonian) fluids. This result will be valid provided that L>7R. 'one or both of which is rotating (Couette flow). and timedependent flow due to either a pressure gradient or a moving surface. analytical solutions to the full Navier. Whether or not a flow is unidirectional depends not : just on the intrinsic properties of the motion but also on the coordinate system chosen. for instance. . t). b. and most of those are for situations in which the flow is unidirectional.) (c) If it is desired to derive corrections to the result in (a). y) is given in Section A. (5.. whereas the number of dimensions is the number of spatial coordinates needed to describe the flow. The surface tension is y and the contact angle is a.8. show that Chapter 6 UNIDIRECTIONAL AND NEARLY UNIDIRECTIONAL FLOW L = 27 cos a! P W ~ R ' where p. Let R be the radius of the tube and let L represent the height at the bottom of the meniscus. it is the use of cylindrical coordinates which makes the problem twodimensional (r. (a) Find the equations governing the streamline which passes through the point (1. The same is true for the number of dimensions. (The formula for the curvature of a surface with height expressed as F(x. is unidirectional in cylindrical but not in rectangular coordinates. The three main types of situations considered are steady flow due to a pressure gradient. 1) at time t. z) instead of threedimensional (x.56. Yt z). and c are constants. show that the general equations in (b) can be reformulated as a perturbation problem using the parameter Show that the answer in (a) corresponds to the O(1) solution to the pertubation problem. of course. There are relatively few exact. streamlines and panicle paths do not coincide. is whether the flow is steady or timedependent. 1. 1. @) Calculate the path of a particle which starts at (1.g. surface tension causes the water in the tube to rise above its level outside. : that is.(r). The first part of this chapter illustrates exact solutions which are obtainable for r unidirectional flow. steady flow due to a moving surface. With Poiseuille flow. 57. relative to the outside whter level.Particle paths are governed by = v(r. and radial flow directed outward from a cylindrical or spherical surface. Some examples are: pressuredriven flow in a tube (Poiseuille flow). compare with Eg. the flow in the gap between coaxial cylinders. there is only one nonzero velocity component. for example. CapiUary Rise When one end of a small tube made of some wettable material (e. Note that the other points included on this streamline change w i t h time. The remainder of the chapter is con. Another pertinent characteristic. Derive (but do not attempt to solve) the equations which govern the interface location for the general case. Couette flow. is the density of water. This illustrates the fact that. so tbat the expression in (a) is not a good approxjmation. 1) at t = 0. : Tbvo characteristics of fluid dynamics problems which are helpful in organizing solution strategies are the number of directions and the number of dimensions. where the one nonzero velocity component is v. the : number of directions is the number of nonvanishing velocity components. (c) Use the results of part (a) to show that unless a = b = c. (a) Assuming that static pressure variations near the meniscus are negligible. The objective here is to compare and contrast the streamlines in this flow with the paths of fluid particles. threedimensional flow has velocity components given by where a.. Streamlines and Particle Paths Assume that a transient. dr dt In contrast to the calculation of streamlines. ( . t is the variable used to trace the particle path. neither the streamline for t = 0 nor that for t = 1 passes through the point r(1) found in part (b). glass) is immersed in water.. denoted as r(1).. (b) Suppose now that the condition L>>R does not hold.910). Determine the particle location at t = 1.
~ I D I R E C T I O N A n L ow
Sttady Flow with a Pressure Gradient
253
cerned with the limitations of those results in modeling actual flows and is also concerned with the extensions of such models to a broader class of situations. The lirnitations arise from entrance and edge effects and also from stability considerations. The extensions involve viscous flows which are nearly unidirectional, in the sense that a second velocity component, while not identically zero, is small. An example is pressuredriven flow in a gradually tapered channel. The analysis of such problems proceeds much like that for unidirectional flow, by exploiting what is termed the lubrication approximation. Including in the scope of our analysis nearly unidirectional as well as exactly unidirectional flow greatly increases the number of real situations which can be modeled. The numbers of directions and dimensions and the presence or absence of timedependence are not the only characteristics which define classes of flow problems. As mentioned in Section 5.10, another key consideration is the relative importance of inertial and viscous effects. In most of the problems in this chapter, including the nearly unidirectional flows, the viscous terms in the momentum equations are dominant and inertia is unimportant. The general features of viscous and inertially dominated flows are the focus of Chapters 7 and 8, respectively.
:*:
3; + :
Figure 61. Flow in a tube of constant but arbitrary cross section. (a) Section normal to the direction of flow; (b) settion parallel to the direction of flow.
z
x
Flow
.:
,
Accordingly, if the axial component of the vorticity vanishes (ie., w,= 0), then 1(1 must satisfy a twodimensional form of Laplace's equation and must have a value that is independent of position on the boundary. It follows from Example 4.21 that $ is uniform throughout the tube cross section and, from Eq. (6.22), that v,,=v,=O. In other words, if there is no rotational (swirling) component of the velocity, a fully developed flow in a tube of any crosssectional shape will be unidirectional. The NavierStokes equation written in terms of the dynamic pressure is
6.2 STEADY FLOW WITH A PRESSURE GRADIENT
The flow of an incompressible fluid in a tube or other channel of constant cross section is said to be fully developed if the velocity does not vary in the direction of flow (i.e., it is independent of axial position). This occurs only beyond a certain distance from the tube inlet, called the entrance length. The entrance length depends on the tube diameter and Reynolds number, as discussed in Section 6.5. Fully developed flows are ordinarily unidirectional; indeed, they are the most important type of unidirectional Row. The general features of such Rows are discussed first, followed by specific examples. Consider the flow of an incompressible fluid in a tube of arbitrary crosssectional shape, as depicted in Fig. 61. The axial coordinate is x and the velocity field is assumed to be fully developed, so that v = v(y, z, t) only. It is desired to identify the conditions for which the flow will be unidirectional (i.e., for which v, = v,=O). Given that dv,l ax = O , the continuity equation for this threedimensional situation reduces to
With v, = v, =0, all terms in the y and z components of this equation vanish, indicating that #lay = d9ldz = 0 or that 8 = 9(x, t ) only. O f importance, the inertial terms in the x component also vanish. That is, v.Vvx= 0 for this fully developed flow. The x component of Eq. (6.24) is reduced then to a linear partial differential equation, which for unsteady flow is
Because vx=vx(y, z, t) and 9 = $ ( x , t), d9Iax can depend at most on time. In other words, none of the other terms in Eq. (6.25) depend on x, so that a9Idx cannot be a function of x either. For steady, fully developed flow we obtain
To examine the possibility of flow in the yz plane we define a stream function $(y, z, t) as
Here 9 = P(x) only, so that O l d x must be a constant. Equations (6.25) and (6.26) are the starting points for analyses of the fully developed flow of Newtonian fluids. The simplifications in the Cauchy momentum equation for fully developed flow are similar. Applications of Eq. (6.26) and the analogous momentum equation for cylindrical coordinates are illustrated by the examples which follow.
Example 631 Flow in a ParallelPlate Channel The objective is to determine the steady, fully developed velocity field for pressuredriven flow in a channel with flat, parallel walls, as shown in Fig. 6 2 . This is called plane Poiseuilleflow. It is assumed that the z dimension is large. so that v, = v,(y) only; see Example 6.52 for a discussion of edge effects in channels of rectangular cross section. Equation (6.26) reduces to
2 d V,=I @ 
An impermeable, solid surface is always a streamline (Section 5.9), so that at the tube
wall # is at most a function of time. Moreover, it follows from Eq. (5.91 3) that
d 3
(6.27)
P &
Steady Flow with a Pressure Gradient
255
Y 4

Rgure 62. Flow in a parallelplate channel with wall spacing 2H and mean velocity U.
where C, is a constant. A symmetry argument like that used to derive Eq. (2.222) leads to the conclusion that T, = 0 at r = 0,so that C , = 0 and
The fact that there will be a pressure drop in the direction of flow (i.e., &ldz<O) indicates that 7, Q 0. For a powerlaw fluid it is found from Eq. (5.613) and Table 510 that with noslip boundary conditions given by
The integration of Eq. (6.27) is simplified by the fact that &I& is a constant. Integrating twice and evaluating the two integration constants using Eq, (6.28) gives
The minus sign in Eq. (6.216) is required because T G 0, and ldv,ldrl is used because dv,ldr s 0. Combining Eqs. (6.215) and (6.216) and solving for dv,ldr gives
The great advantage in using 8 instead of P is that Eq. (6.29) is valid for a channel with any spatial orientation. If the channeI is horizontal (i.e., g, =0), then &Pldx=dP/dr; if the channel is vertical and there is no applied pressure (i.e., dP/dx= 0), then 0 1 & = pg. Note too that the same parabolic velocity profile results if one of the noslip boundary conditions is replaced with a symmetry condition at y = 0,namely ryx = 0 or dv,ldy = 0, The mean velocity in the channel (U) is related to the pressure drop by
f 'i>'
ifi
The solution is completed by integrating Eq. (6.217) and using the noslip condition, v,(R) =0, to evaluate the integration constant. The result is
so that Eq. (6.29) may be rewritten as
where U is the mean velocity. The shape of the velocity profile in Eq. (6.218) is determined by the powerlaw exponent, n. For n+O, the centerline velocity approaches the mean velocity. approximating plug flow. (As indicated in Chapter 5, polymeric fluids exhibit values of n as small as 0.2.) In general, the volumetric flow rate is given by
It is seen that the maximum (centerplane) velocity i s f ~The . volumetric flow rate per unit width of channel is
1.
:
1
' j ,
*
For a Newtonian fluid, where n = 1 and rn = p, Eqs, (6.218)(6.220) yield the wellknown
results for Poiseuille flow,
Example 622 Flow of a PowerLaw Fluid in a Circular lhbe Consider steady Row of a powerlaw fluid in a cylindrical tube of radius R. For this fully developed, axisymmetric flow we have v, = vz(r), v, = v, = 0, and 9 = 9 ( z ) . With such a velocity field, rrz (or r,,) is the only nonvanishing component of the rateofdeformation tensor. It follows from Eq. (5.63) that., for a Newtonian or generalized Newtonian fluid, T , (or %) is the only nonzero component of the viscous Smss. and that T,= r,(r) only. From the z component of the Cauchy momentum equation we obtain
?rlt4 cH' Q =  8p  dz'
With 9 =4P(z) and T~ = Eq. (6.213) is satisfied only if both sides y u a l a constant; that is, must be constant. Integrating, it is found that
: :

Thus, the velocity profile for a Newtonian fluid in a circular tube is parabolic, as for a parallelplate channel, but the maximum (centerline) velocity in thls case is 2U. Equation (6.223) is one form of Poiseuille' s law.
'
'This type of flow is named after Jean Poiseuille (17971869). a French physician whose interest in the mechanics of blood flow led him t o conduct an extensive set of pressure and flow measurements in Paris in the 1830s and 1840s. He employed water and other liquids in glass capillary tubes. and was extremely meticu
c
o
w

Steady Flow with a Pressure Gradient
257
Example 6.23Flow of %o Immiscible Fluids in a ParallelPlate Channel A simple type of twophase flow occurs when immiscible fluids occupy distinct layers in a parallelplate channel, as depicted in Fig. 63. The density and viscosity of fluid 1 (p, and p , ) may differ from those of fluid 2 (p, and k). It is desired to determine the steady, fully developed velocities of the two fluids, which are denoted as vL(y) and <(y), respectively, The NavierStokes equation for each phase reduces to
where
Integrating this twice gives
where aiand bi are constants. These four constants are determined by the conditions
ui and K are assumed to be known. What remains is to relate the pressures in the two fluids. In Examples 6.21 and 6.22 it was seen that for fully developed flow of a single, incompressible fluid in a channel of known dimensions, specifying the mean velocity was the same as setting the axial gradient of the dynamic pressure [see Eqs. (6.2lo), (6.219), and (6.222)]. Extra care is needed in the present problem, because O(i'ldr, while constant within each fluid, is generally not the same in the two phases. The constraints on the twophase flow are revealed by considering the actual pressure, P. For the general case of a channel inclined at an arbitrary angle, P(')= p)(x, y, z ) even in the absence of Row, because of static pressure variations. However, from the definition of the dynamic pressure, Eq. (5.8l), it follows that for this flow
:,,
Equations (6.226) and (6.227) are the usual noslip conditions at the solid surfaces, whereas Eqs. (6.228) and (6.229) express the matching of the tangential components of velocity and stress at the fluidfluid interface. It is convenient to introduce the constants
.
'i:
8
.
Thus, the constancy of #(')I& implies that aP(')ldx too is constant within each phase. The final piece of information needed comes from the normal stress balance at the fluidfluid interface, based on Eq. (5.79). Given that the interface is flat and that there are no normal viscous stresses (i.e., r,=O), the values of P there must match. We conclude that dP/ax has the same constant value throughout both fluids. To emphasize that there is only one independent pressure gradient, Eq. (6.230) is rewritten as
a
The need to consider actual pressure, and not just dynamic pressure, is typical of problems involving fluidfluid interfaces.
*'
where ui has units of velocity [compare with Eq. (6.29)J and K is dimensionless. The velocities in the two fluids are written as
Fluid 2
Fluid 1


4
H2
Example 6.24 Flow of a Liquid F h Down an Inclined Surface With reference to Fig. 64, i l m flowing down a surface which is oriented !, ,' the objective is to determine the velocity of a liquid f at an angle P relative to vertical. The film thickness (H) is assumed to be constant, making it !: i'.: possible to have steady, fully developed flow. Thus, it is assumed that v,=v,(y) only and v,,= v, = 0, from which it follows (as before) that 9 = P(x) only and that dP1d.x is constant. The falling liquid film can be viewed as a special case of the twofluid problem in Example 6.23, in which fluid 1 is now the liquid and fluid 2 the gas. Assuming that the gas occupies a
:
fi
HI
Figure 6 3 . Flow of two immiscible fluids in a parallelplate channel.
Figure 64. Flow of a liquid film down an inclined surface.
A
lous. At about the same time, the dependence of Q on was revealed also by a less extensive and precise fe' of mssurernents made in Berlin by a German engineer, Gotthilf Hagen (17971884). 7lus, Q. (6.223) 1s sometimes called the HagenPoiseuilleequation. The first derivation of this result horn the NavierStokes equarion is attributed to the physicist Eduard Hagenbach (18331910) of Basel in 1860, although several appear to have done this at about the same time. Incidentally, Poiseuille also invented the Utube mercury manometer, to measure arterial pressures in animals (Sutera and SkaIak. 1993).
Stcady Flow with a Moving Surface
259
space at least as thick as the liquid film, and recalling that a typical ratio of liquid to gas viscosities is lo2 (Chapter l), the parameter K defined by Eq. (6.231) will be extremely large. Assuming also that the gas pressure is uniform, the reasoning leading to Eq. (6.235) indicates that dP/ dx=O in the liquid. Noting that g , = g cos /3, it follows from Eqs. (6.232) and (6.235) that the liquid velocity is

(b)
f
,g:
. ..,.
r'.
i
where the subscript L denotes liquid properties. The velocity profile can be rewritten in terms of as the mean velocity (U)
: i.
Fluid

Figure 65. Couette viscometer. The outer cylinder is rotated at angular velocity
u=H2pd cos P
~ P L
Notice that Eq. (6.237) is exactly the same as the result obtained in Example 6.21 for flow in a parallelplate channel of halfwidth H. Assuming that K+, as done in deriving Eq. (6.236), is the same as neglecting the shear stress exerted by the gas on the liquid. As discussed in Section 5.7, this is a common approximation at gasliquid interfaces, and it has the effect of making the liquid velocity independent of the gas properties. It is readily confirmed that Eq. (6.236) is obtained also by solving
':,
w and the inner cylinder is fixed. (a) Overall view; (b) enlargement of area indicated by dashed rectangle in (a) for & < < I , with H=sR and U = wR.
fluidfilled gap is of thickness d?and the linear velocity at the wetted surface of the outer cylinder is wR. Noting the circular symmetry and neglecting end effects, it is assumed that v and 9 do not depend on e or z, Similar to fully developed flow in a channel, a velocity field of the form v,= v,(r) and v, = v, = 0 is consistent with continuity. From the B component of the NavierStokes equation it follows that
with the boundary conditions
'
,,
,
Integrating Eq. (6.31) twice yields terms in v, involving r' and r . Evaluating the integration constants using the noslip boundary conditions,
This is clearly the preferred approach if one is interested only in the liquid, in that it avoids having to determine the velocity field in the gas.
we obtain
6.3 STEADY FLOW WITH A MOVING SURFACE
Flow is induced not just by a gradient in dynamic pressure, but also by motion at a solid or fluid interface. Flows caused by the tangential movement of a surface provide the purest illustrations of surfacedriven flow, in that there need not be any pressure gradient in the flow direction. The two examples in this section, both of which entail robtion of a solid surface, fall in that category. The main feature that distinguishes these rotational flows from the other unidirectional flows in this chapter is that the swamlines are curved, implying that all material points are undergoing spatial accelerations. A presSue gradient normal to the direction of flow(i.e., t@l8r>O) is needed to sustain such accelerations.
E m p I e 6.31 Couette Flow Steady flow in the annular gap between long, coaxial cylinders, 0°C or both of which is rotated at a constant angular velocity, is termed Couefte flow. In a Couette dscomekr, as shown in Fig. 65(a), the inner cylinder is fixed and the outer one is rotated. The
. ,,".
t:..:
,
k:'... b : ,
.

A,
'
The r component of momentum serves only to determine P(r) horn v, and therefore it provides no additional information concerning the velocity. The z component only confirms that for the assumed form of v, 9 must be. independent of z. Couette viscometers are normally designed with narrow gaps, which leads to a simpler velocity profile. For E<<1 the effects of surface curvature will be negligible, as in the first approximation to the concentration field in Example 3.72. Reformulating the problem using local rectangular coordinates, as shown in Fig. 65@), conservation of momentum gives
: <'
The velocity profile for small gaps is then simply
,
.
;
where U = oR and He&. Couette flow for planar surfaces, as given by Eg. (6.35), is referred to as plane Couene pow. An identical, linear velocity profile is obtained (more laboriously) by
TimeDependent Flow
261
expanding the numerator and denominator of Eq. (6.33) in Taylor series about s= 0.Local rectangular coordinates find extensive use in lubrication theory (Section 6.6) and boundary layer theory (Chapter 8). As seen here, they are appropriate when the smallest radius of curvature of the bounding sufface(s) greatly exceeds the other length scales governing the flow. Measurements of the torque required to turn the outer cylinder (or to keep the inner cylinder stationary) provide an effective method to determine the viscosity of the fluid. The magnitude of the torque exerted on the fluid by the outer cylinder is G = ( 2 . r r R 2 ~ ) r r , ( where ~ ) , L is the wetted length of the cylinders. The shear stress, ~~,=pni(v,/r)ldr (Table 5  3 , is determined in general by differentiating Eq. (6.33). For narrow gaps, however, rr0 T, = pdvX/dycomputed from Eq. (6.35). An important property of plane Couette flow is that the shear stress is independent of position, or 7y,= wU/H. Using q,= T,, = pO/E, the torque for E c<1 is
Using Eq. (6.37) in Eq. (6.38) and integrating, the pressure is found to be of the form
where g(z) is an unknown function. Integrating Eq. (6.39). another expression for the pressure is

where f(r) is an unknown function. The two unknown functions are identified by comparing Eqs. (6.310) and (6.3ll), leading to the conclusion that
Example 6.32 Surface of a Liquid in RigidBody Rotation Figure 66 shows an example of a system with a gasliquid interface of unknown shape, consisting of a liquid in an open container of radius R that is rotated at an angular velocity o. If the container is rotated long enough, a
steady state is reached in which the liquid is in rigidbody rotation. It is desired to determine the steadystate interface height, h(r). assuming that the ambient air is at a constant pressure, Po. Because the viscous shess vanishes for rigidbody rotation, the analysis will apply to any liquid, Newtonian or nonNewtonian. The effects of surface tension will be neglected. The velocity of a fluid in rigidbody rotation is the same as that in a rotating solid, or
ve (r) = or
where c is an unknown constant. To determine the height of the interface, we note that with viscous stresses absent and surface tension assumed to be negligible, the normal stress balance requires that the liquid pressure at the interface equal Po. Setting z = h(r) and P = Po in Eq. (6.312) gives
)',
,
Thus, the interface is parabolic in shape, with the lowest point at the center. The additional condition needed to determine c is a specification of the total volume of fluid in the container, or V = 2 r h(r)r dr.
(6.37)
with v,=v, = 0.Because of the free surface, P will be used as the pressure variable instead of 9. The circular symmetry indicates that P = P(6 z) only; the 8 component of the Cauchy momentum equation (Table 52) confirms this. Given that g,= 0 and g,= g, the r and z components of conservation of momentum become
[
: , Using Eq. (6.313) in (6.314) to calculate c, the final result is p,'
,'
where ho= v/(TR~) is the liquid height under static conditions.
q,:
I
I
Figure 66. A liquid in rigidbody rotation in an open container.
6.4 TIMEDEPENDENT FLOW
Exact solutions for a number of timedependent, unidirectional flow problems can be found by using the similarity method (Chapter 3) or the finite Fourier transform method (Chapter 4), as illustrated by the two examples in this section.
Example 6.41 Flow Near a Flat Plate Suddenly Set in Motion A flat plate at y = 0 is in
contact with a Newtonian fluid, initially at rest, which occupies the space y>O. At t = O the plate is suddenly set in motion in the x direction at a velocity U, and that plate velocity is maintained indefinitely. The objective is to determine the fluid velocity as a function of time and position. This problem may be viewed, for example, as representing the early time (or penetration) phase in the startup of a Couette viscometer. Assuming that v, = v,(y, t) only and that dY/ax = 0,the problem formulation is
Liquid
TimeDependent Flow
263
the problem is rewritten as
pz=; &(%i)
where v&P is the kinematic viscosity. Setting @ = v,/U and replacing D with v, this is the same as the transient diffusion problem solved using the similarity method in Section 3.5. Thus, with the similarity variable defined as The parameter P is the ratio of the viscous time scale (R~IV) to the imposed time scale (wI). Pseudosteady behavior is expected for PC< 1 (see Section 5.10). In applying the finite Fourier transform method the solution is expanded as
00
a@
'
a@
+ 8 ( l + y sin 7).
the solution is
4 = n= C1 @.(~)@"(q),
The analogy between this problem and the transient diffusion problem of Section 3.5 underscores the fact that the kinematic viscosity is h e diffusivity for momentum transfer, Thus, as mentioned in the discussion of the Strouhal number in Section 5.10, the characteristic time for viscous momentum transfer over a distance L is
(6.4 13)
It follows that the time to achieve steady flow in the Couette viscometer of Example 6.31 is H~IV, or (ER)~Iu.
:
The Bessel function Jo satisfies the requirement that the velocity be finite at 1)=0 (see Section 4.7). The differential equation and initial condition are transformed to
Example 6.42 Pulsatile Flow in a Circular %be It is desired to determine the velocity of a Newtonian fluid in a circular tube of radius R, when the fluid is subjected to a timeperiodic pressure gradient. It is assumed that the fluid is initially at rest and that for t>O the axial pressure gradlent is given by
=
8;. which has the solution
1
89'
az
 p a ( l + y sin a t ) ,
Using Eqs. (6.414) and (6417) in Eq. (6.413), the overall solution is found to be
@(,,r)=
where a, y, and o are positive constants. The timeaveraged pressure gradient is pa, where a has units of m s  ~ ;y is dimensionless and o has units of s'.  . For a unidirectional flow w i t h v, = v,(,: I ) and v, = v,= 0, the dimensional problem to be solved is
=a(I
at
av*
+ y sin o f )

162 n=l
[ ( l  e  k ~ r l ~ ) + (  ) ( p  ~ ~ / ~ + ~
yh2,
A :
+ p2
sin
7/3cos 7)
I
JO(Xn7l) ~ ~ ( h ~ ) (6.4 1 8)
a
*
:
As will be seen, a sufficient condition at the tube centerline is that v,(O, t) be finite. It is advantageous here to use dimensionless variables. The length, time, and velocity scales chosen as the tube radius (R), the imposed time scale (a'), and the mean velocity comsponding to the timeaveraged pressure gradient (U,), respectively. From Eqs. (6.222) and (6.46) we find that W,,=R~~/(~U). With the dimensionless variables defined as
It is seen that the velocity contains transient contributions, which decay with time, and timeperiodic contributions, which continue indefinitely. Examining in more detail the individual parts of Eq. (6.418), we observe that the part that is independent of y (the amplitude of the p m s w pulses) must be the response to a step change in the pressure gradient, from zero to some final value. Thus, for a step change followed by a constant pressure gradient (i.e., for y = 0), the dimensionless velocity is
.
,.  , . ,
 ,
Based on the slowestdecaying term, an approach to within 5% of the steadystate solution for a stcp change requires that 7/P= 3/h12 = 3/(2.405)2= 0.52, or t = 0.52 R2/v= 0.52 tv. Thus, the transients disappear in less than one characteristic time for viscous momentum transfer. The steadystate part of Eq. (6.419) must correspond to the familiar, parabolic velocity profile in a tube.
% = 2 u a u. + &t=.* 2 Y 335: 4 8 5 % C L E $2 3asP" zs E ..3 Y X 2 2 2 =" O " .W a 5 8 .s5 39 3 d'Z Sue a o a a .5g 0 a zzg m $ j > a . gd)ct? 2 rz 3 Es 2s a m0 3$ ru 0 ' E s E ss 5 g g 0 s 5 9 3 8 0a "= 2 $z+ 92:s 0 S * S E 3 02 k3y % A G O 0 z *g .
8. 67 are borne out by measurements in developing flows (Atkinson et al. where V is the scale for v. which is discussed in more detail in Chapter 8. which has impoMnt implications for momentum transfer in the wall layer. u(Lv)= 2 U. where The term involving av. 67 and a comparison of Eqs. Developing flow in a parallelplate channel. (d) v. Eq.lax2.8).512). the x component of the NavierStokes equation for the core reduces to [ This relation. it is found that . (6.(c) v. Another implication of the smallness of 6/x is that Vm<< 1.59) and (6. It follows that a8/~3y<<@ldx. the fully developed velocity profile from Example For consistency with this.57) makes . it is assumed that the velocity in the developing wall layer is of the form where again 9= 9 ( x ) only. 1967).514) we infer that 3 u(0)= U. the d2v. (6. the continuity equation is integrated to give From Fig. In the c o n region it is assumed that the large value of Re makes the viscous terms in the NavierStokes equation negligible (see Section 5. (6.58) is obtained by integrating each term over the thickness of the wall layer. (b) v..56) and (6.. With v. @Lv)= H Using Eq.58) leads to 6 6 The smallness of v . In terms of the coordinates in Fig. Using these approximations. (6. (6. . It follows that 6/x<< 1 throughout most of the entrance region. and U. W~th regard to v .. (d) An approximate solution for the velocity in the developing region is sought now using the integral method (see Section 3. in the wall layer are x. 145)].=O a t y = O . (6. 4 0 ) = 0 . Because 9 does not vary appreciably with y.58) is evaluated using integration by parts. for x 3L . (A.56) can be used ' 0 evduak the pressure gradient in the wall layer in terms of u(x). p. Thus. One implication is that d2vxl a x 2 c < a ~ / This ~ .513) and (6. for x =0. (6. at x=Lv. The lefthand side becomes because d&dr>O.58) as P 6 : This is identical to the boundary layer approximation. (a) Coordinates. (6.51 1) is rewritten with the help of E q .55). As indicated by Eq. implies that dl terms in the y component of the NavierStokes equation will be much smaller than the dominant terms in the x component.uIx2 and #vxldy2. (6. is seen by noting that the scales for x. respectively. or that 9 is approximately a function of x only. so that d2v. for Ocx< L. An integral form of Eq. which has been used widely in boundary layer theory. the flow is fully developed. Equation (6. a large value of Re ensuns that hVLvc< 1. is called the momentumintegml equation or the K c i d n integral equation [see Schlicting (1968. and v. the x component of the NavierStokes equation for the wall layer is written as Noticing that the term in square brackets vanishes at both limits of the integral.Umibtfons o f Exad Solutions where Y is a dummy variable.=u(x)..57) to replace the pressure term and v.10 and also Chapter 8). (6.514) to evaluate the terms in Eq..ldy in Eq. (6. 67.lh2 term can be neglected in the x component of the NavierStokes equation. Using Eqs. y. Eq. which gives : Combining Eqs. in Eq.52) and as the results of this analysis will confirm. the momentum equation for the wall layer becomes (a) (4 (c) Figure 67. When the entire cross section has adjusted. (6. This is seen fmm the continuity equation.se of the fact that v.510) with the integrals of the other terms in Eq.U1a2. The qualitative features of the velocity profiles shown in Fig.respectively.
52 Edge Effects for Flow in a Rectangular Channel Consider fully developed flow in a channel with the rectangular cross section shown in Fig.052 versus 0. Thus. (6.517) becomes This firstorder differential equation for 6. when the integral involving fi has been evaluated using the substitution r = 1 (813). 8. the opposing edge is effectively at Z = . (6. (6. where the channel edges were ignored.052 corresponds to about 93% of the transition. The mmensional form of the momentum equation which applies here is Eq.21.. That is. The dimensionless problem statement becomes @(K 0) = 0. 1969).olumetric flow rate in a channel of constant cross section is independent of axial position.2lo)]. . the integral analysis provides a reasonable estimate of the entrance length for large Re. @. Channel of rectangular cross section: (a) overall view of cmss section.: It is assumed in the second boundary condition in Eq.(0) =0. 68(b). 68(a). we find that .526) . H This is very similar to the highRe asymptote of Eq. It is found from Eq. O(E: m) finite. the coefficient 0. 3 4 = (A n 1)". this implies that H I HIS where U is the mean velocity in a channel without edge effects [see Eq. Dimensionless variables are defined as Equation (6. a s shown in Fig.088). For an incompressible fluid the . The flow is in the x direction (not shown). The analysis focuses on a region near one edge. The finite Fourier transform method is applied with an expansion of the form so that Eq.26).088 corresponds to 99% of the transition from a uniform to a parabolic velocity profile and that the deviation from fully developed flow decays exponentially as x becomes large (Atkinson et al.() finite. (6. (b) enlargement of area indicated by the dashed rectangle in (a).5Z).517) provides one of the two relationships needed to determine the unknown functions u(x) and &x). Using the variable ~ = y / 6 ( n ) to integrate over the channel cross section. This relationship between u and 6 is used to eliminate u from the momentum result. Integrating over the entire entrance region. The final expression for the velocity entrance length is L ' The solution for the transformed velocity is V=0. although the numerical coefficient is smaller (0. The objective is to i (a) (b) Figure 68.: which yields the transformed problem d 2@n  d2? A:@. which is subject to the initial condition 8(0)=0. i Example 6.Limitations o f Exact Solutions 269 or. Another relationship between u(x) and 6(x) is obtained from conservation of mass. (6. in dimensionless form.526) that the channel is wide enough that the flow is influenced only by the nearest channel edge. determine how far one must be from the side walls to obtain the parabolic velocity profile of Example 6.052 Re. Given that the coefficient 0. (6.519) that . (6. is nonlinear but separable.
5 " E m q s Q) 3 $ 0 J> C U P ) 3 .$ . ~ 8 32g as .$.3 a rr. c .5?1$ 2u "913 S ~ " " F~ a BJU* ! i 9.5 g k W '3 cE! . 0 gcgu IS?. .o&3 z yma\ m a . 5 % E I h 0 ..ia v* P % S 2 5 3 8g:W E. e5. i s .a $ g T ' >* '9 x .2s fL~tb.* $ Bemy a .c 3 8 3 a n ' S s s M e 5 8 WC) 2 .d ! 3 0 " S E 83 34.
Ll(h. $ symmetry requires v. F or L. (neglecting the entrance region) is r!. i:' For h.= H. As one might expect. ru h dx hL L L.. and y we arbitrarily choose the downstream values of u and h. (For 'clarity in labeling..(x) and c.67) it is necessary to identify the appropriate length and velocity scales. like that for fully developed flow in a parallelplate channel. Eqs. it is determined by the magnitude of av. Flow in a tapered channel. analogous to Eqs. (6. the halfheight is and the total pressure drop is . For ho or hL to be suitable scales for y it is necessary that entrance effects be negligible (see Section 6. 69. as shown in Fig. viscous flows. f h(x)]=0. To proceed further it is necessary to specify the function h(x). (6. we recover the result for a parallelplate channel of constant halfheight H.. and h. (6.) are used as scales. It is readily confirmed that the same criteria result if the upstream values (u. an additional requirement is that L>> L. specifying q and h(x) determines u(x).) is not the channel length (L). may be estimated from Eq. respectively. y) we integrate the continuity equation to obtain dx du dx u d h =_ uL(bhJu. and h.5).= h. (6. Equations (6. (6. . it follows from conservation of mass that the product y(x)h(x) for an incompressible fluid must be a constant. Using Eq.612) where q is the volumetric flow rate per unit width. 6. We begin by defining the conditions under which the Iubrication approximation can be used.(x. (6.210) and ( 6 . L. and recognizing that rf.649.67) are the two conditions needed to make the lubrication approximation valid. the only difference being that the channel height and mean velocity vary with axial position. Indeed. (6. dhl dx. As the scales for v.(x) are unknown functions. To evaluate v.67) become 2: where Y is a dummy variable.= h. Follewing are four examples of the analysis of nearly unidirectional.. the pressure is determined by integrating Eq. Using Eq. to vanish at y = 0. Thus. which is . where L.61) twice to give where c..l l ) . changes by an amount comparable to u. and that distance will greatly exceed L if the fractional change in the channel height is small.66) and (6. (6. y) we integrate Eq. 2 . (6.A Lnow / ' lubrication ~ p ~ r o x i m a t i o n 273 is helpful but not absolutely necessary..h.) The walls are not necessarily planar.52) by using H = b. which are denoted by u. Finally. The total pressure drop i" :!.66) and (6. Accordingly.. To apply Eqs.). Given the volumetric flow rate and the local halfheight h(x). is defined as the distance over which v. it will be shown that the lubrication approximation yields a velocity profile which is parabolic in y. and this result for L. we find that dv. vy is seen to be directly proportional to the slope of the channel walls.the result is + k where Re = 2uhplp = ql v is independent of position along the channel. = . In other words. the channel height has been greatly exaggerated. =4P(L).614) to evaluate the integrand.61 Flow in a Tapered Channel Consider steady flow in a twodimensional channel with a wall spacing which decreases gradually in the direction of flow. 2 . where Po= P(0) and 9. Evaluating those functions using the boundary condi tions vJx. With Ly= h. Assuming planar walls.lax. Figure 69. Denoting the local mean velocity as u(x).66) and (6. such that To determine v. Notice that the Reynolds number is based on the crosssectional (smaller) dimension. we find that.(x. Example 6. a channel with gradually tapering walls yields a velocity profile like that in a parallelplate channel. Thus.This completes the set of conditions.613). the objective is to determine the velocity and pressure. The length scale for velocity changes in the x direction (L.
Figure 610. (6. t' ' Integrating Eq. . (6. The restriction due to the neglect of the entrance region is L>>L. where L. and forces for a class of twodmensional. with v. uo r$)r ~r$) ) = = Re. . we find that The mean axial velocity and wall velocity are related now by evaluating Eq. ). is to situations in which two solids in relative motion are ':?..627) in Eq. G ( . rather than a change in cross section. (6. may be estimated where &ldz is a function of z. the overall pressure drop is computed by integrating Eq. General lubrication problem in two dimensions. analogous to Eq. Integrating twice and applying the noslip condition at the tube wall (which usually remains valid despite the presence of flow normal to the wall)./uoLrlLz. It is assumed that the pressure is specified at two. (6..621) to be valid. the scale for r is Lr= R. + + y ' . As a find remark. No explicit consideration of L. originating with Reynolds (1886). note from Eq. (6. where v. assume that the fluid velocicy normal to the tube wall (the membrane) is a known constant. u must remain of the same order of magnitude as uw Thus.626) at r= R... v. . The main difference between this and the preceding example is that the axial variations in the velocity are caused by the loss of fluid through the permeable wall. 610. there is a volumetric flow rate q (per unit width). is the scale for vr and u.=vw. is the scale for v.6l). Accordingly.. vanish at r = 0. The coordinates are chosen so that the upper surface is stationary. As shown by the second term. = U... (6.63 Lubrication Theory The classical application of the lubrication approximation. was needed because the ratio of velocity scales was known directly. This additional restriction on the analysis is expressed as : The aspect ratio in the geometric condition has been replaced by the ratio of velocity scales.~RECRONAL now Lubrication ~~proxirnaGon 275 Example 6. Eqs. v. (6. without rotation.625) and requiring that v. machine parts with lubricating oil). For simplicity. whereas the lower surface has a tangential velocity. for the scaling assumptions embodied in Eq. z) from conservation of mass.221). the tube cannot be too long. The last result is analogous to Eq. The first term corresponds to the pressure drop in an impermeable tube of length L. ' two solid surfaces are assumed to be separated by a distance h(x). the continuity equation in cylindrical coordinates indicates that v. we obtain where u(z) is the local mean velocity. where v. and the dynamic condition has been reduced to the requirement that the Reynolds number based on the wall velociv (Re.624) with the continuity equation gives Example 6.67) become which is readily integrated (for constant v. problems in which one surface slides along the other. constant.65). (6.. The objective is to determine the velocity and pressure in the tube.) to give Substituting Eq. pressure.As a result of the motion of the lower surface andlor an applied pressure difference. steady ..<< 1. The objective in this : example is to determine the velocity.. ar ) l a 1d9' %= p dz ' 8= 8(z) only.locations. consider steady flow in a cylindrical tube of radius R and length L The mean axial velocity at the tube inlet is given as u. Analogous to Eq. Assuming that the excess pressure drop in the entrance region is negligible. The result is a differential equathn governing u(z).. 1 ' X = XI x = x. (6. is F.. the effect of the fluid loss through the wall is to reduce the axial pressure drop. The next step is to determine u(z) and VXI. (6. and they are frequently operated with a low external pressure to cause filtrate to pass outward through the walls of the fibers. (6..62 Flow in a Permeable Tbbe Membrane filtration modules are often configured as bundles of hollow fibers.separated by a thin film of fluid (eg. Neglecting entrance effects. (6.) be small.628) that the axial velocity will fall to zero at z= u p / : (2~~ However. The lubrication approximation for axisyrnmetric flow in cylindrical coordinates. To model the flow in one such fiber. Combining Eq.66) and (6.626) yields the final expression for v.623) to give << vw 1. As shown in Fig.
Integrating Eq.643) is dh/dx= o(E). This force (per unit width) will have a horizontal or drag component (F. it is this large ratio which leads to favorable lubrication performance. (6. (6.640) that 9 = O(E'). it is seen that v. strictly speaking.637) and (6. (6. that lubrication problems usually involve large pressure variations withm thin layers of fluid.).= rxy @vX/ay.. It is found using Eq. = O(1). (Note.9(x1)= 9 .84)]. 610. in lubrication problems the external pressure variations are very small compared to those in the narrow gap. and AY.  which resembles a linear. To focus on the forces due to fluid motion (i.. pressure. Thus.640). The widely different length scales lead to different dominant terms in the NavierStokes equation for the three regions. as suggested above.where gap which is the final expression for vx(x. As discussed in more detail in the next example. is 0(1). Eqs. we will use 9 in the normal stresses instead of P [see Eq. both of which are characterized by length scales much larger than the gap thickness. Eq. (6. which is an unknown function of x. is known.where L =x. Thus. similar to the shear stress. so that static pressure variations are negligible and the distinction between 9 and P is quantitatively unimportant. The interpretation of the pressure boundary conditions is discussed later.e. . and s. and H is the characteristic gap height. The velocity.640) at x = x . Let E HIL+O.. we obtain &? 6 p U 12pq dx .= 2@vxlax = O(E) and ryy = 2pdvylay= O(E). (6. To solve a problem of this type it is necessary to specify h(x) and two of the following three quantities: U. rYx= T.) .633) that Solving for HI&.W O L =/ Lubrication ~ ~ ~ r o x i m a t i o n 277 such that A 8 = 9(x. the noslip boundary conditions are applied to the integrated form of the lubrication equation. . The next step is to relate dP/dr. the lubrication problem is uncoupled from the external flow problems. y). (6. so that to good approximation 9 may be treated as a constant within each bulk fluid [see Leal (1992.) Thus. is The function g(x) serves to normalize n. to the constant flow rate q. then the small parameter is HIL.pUlH. The stress exerted by the fluid on the upper surface is s = n.638) indicates that T. (The argument of the function h is omitted now. Given that the coefficient of the O(EI) pressure in Eq. (6. whereas Eq. This is the justification for assuming that 9 is known at each end of the We consider now the force which the fluid exerts on the upper surface in Fig. the ratio of the pressure to the  . (6. Accordingly.T . the contribution of the pressure to s. gives a relationship among U. As the first step. pp. The velocity derivative of most interest is which is needed to calculate the shear stress. (6.633) becomes which allows the third quantity to be calculated.61. both of the normal viscous stresses are negligible.q. The derivative needed to determine v.. (6.9. if the gap length is also the external length scale.) and a vertical or lift component (F. the boundary conditions for the pressure require an asymptotic matching of the pressure variable used here with the pressures computed in the bulk fluid on either side of the narrow gap. 387396) for a more detailed discussion]. the components of s are calculated as from which it follows that As in Example 6. LIH.adhldr. so that the overall problem has a singular perturbation character.635). and A 9 . where the pressure is approximately that of a stagnant fluid. Thus. Again referring to the discussion following Eq. It To identify the dominant contributions to s is seen from the discussion immediately following Eq. which is the scale here. Eq. the x component of the stress vector reduces to .611). with the understanding that h = h(x) throughout the analysis. Fortunately. Thus.u. to exclude buoyancy).g. and viscous stresses are then computed as needed from the other relationships given above. however.635) to eliminate dPldr. For comparison.) The result is shear stress is large. combination of plane Couette Bow (the term containing LI) and plane Poiseuille flow (the term containing B l d r ) . In principle. Evaluating Eq.639) indicate that 2. (5. (6.x . the pressure is calculated as Tabng the term containing U as being representative of the righthand side. it is apparent that 9 huLIH2.F ' Using Eq. the lubrication problem is coupled to two external flow problems.
This is consistent with the pressure scale inferred from Eq. it is permissible to set g = 1. .64 the thin fluidfilled gap extended from x = 0 to x = L.644) reduces to Setting x . = 0 in Eq. (6. (6. the force components are at x = h&(ho+ h.the drag force on the top surface (per unit width) is found to be F. the av. Example 6. A differential element of length along the upper surface in Fig..65 Sliding Cylinder In each of the preceding examples the lubrication approximation was applied to a fluid region with a definite length. . the better.lhL = 2. An example is shown in Fig./ax term was neglected.e. using Eqs. For instance. . (6.618). In Eq. i < x f L < I .644) the shear stress and pressure contributions are O(E) and O(E'). this force ratio is in general.L6 for holhL=2.respectively.646).645) and (6. it is of interest sometimes to compute the torque (per unit width) exerted by the fluid on the upper surface. The main features of this problem are that the flow results entirely from the relative motion of the surfaces (i. 1. Figure 612. so that only the pressure needs to be considered in calculating the y component of the stress vector. (6. Equation (6. . we obtain P . at ~ xIL= 8. For h. Eq.647). Using Eqs.648) In addition to the drag and lift forces. =pu~lh. in which it is assumed F%ure 611. with the slider bearing of Example 6.= j Pdr. as given by Eq. 612. The main objective is to calculate the drag and lift forces.645).648)..641) is used to evaluate the flow rate as . It is assumed that P(O) = 9 ( L )= 0 and that the relative velocity U is known. It is seen that thin fluidfilled gaps can provide extremely favorable loadtodrag ratios. Given that O<hL<ho. A 9 =0) and that the gap height decreases monotonically in the direction of flow. in Eq. equals the load which can be supported by the bearing divided by the drag. Slider bearing.640). it is found that The pressure reaches a maximum value of To compute the corresponding forces. The ratio FJF. the lift force is (6. so that. (6. (6. and therefore it measures the effectiveness of the lubrication. the higher this number..it is seen that the pressure maximum always occurs in the downstream half of the gap. 611 provides a good illustration of the principles of lubrication. or   Example 6. without rotation. Cylinder moving parallel to a surface. Using Eq. In other situations the limits * to be imposed on x are not so clear. (6.Lubrication ~pproximation 279 I n evaluating r. Using Eq. (6. the stresses must be integrated over the surface.653) and (6.64 Slider Bearing The slider bearing shown in Fig.654).640) and performing the integrations. (6. (6. The torque about position ro is given by i In keeping with the lubrication approximation. The inclined surface is taken to be planar.). 610 is g(x) dr. Given that A 9 =0.
References .659) in a Taylor series about x=O.660). we conclude that Eq. 2. It is seen that the analysis will be limited more by the restriction in Eq. Card.. and W.658) is valid only when the gap is narrow. b(X). 1961. McGraw Hill. immersed in a liquid. I F References 28 1 where x.658). 0.665) in Eq. and H. the lubrication approximation becomes questionable for i r / ~ exceeding I only 0. The fact that 9= O outside the gap suggests that we set A 9 = O in Eq." Using Eqs. In other words.661). B. Soc. Leal. For example. with the asymptotic value of blc=4/3. Numerous applications of the lubrication approximation in polymer processing are discussed by Middleman (1977). S. Theoty of Hydmdynumic Lubrication. Lond. x = L 3P provides a practical definition of "far upstream" and "far downstream. Wiley. The asymptotes of b and c are closely approached already at X = 3. Noting that h(x) is symmetric about x=O. Cambridge University Press. Smith. B. is obtained by expanding Eq. ButterworthHeinemann.666) will be accurate if &lRS lop3. P= 0 at x = kx. McGmwHill. Oxford University Press.661) than by errors in Eq. B. C. it is found that Atkinson. . Sternlicht. A simpler expression. Pinkus. (6. Drazin. New York.640) and (6. R. (6. Of particular interest is the behavior of b and c. Adopting coordinates in which x=O is always the point of minimum separation. R. Fundamentals o . H. . AlChE 3. On the theory of lubrication and its application to Mr. (6. . (6. when x varies by an amount comparable to e. 1. Lernrnon. Reid. E. the result is which turns out to be an excellent approximation to h(x) even for fairly large x. (6. (6. 1969.660) as which suggests that e is the characteristic length in the flow direction.658)]. 4 . P. Pearson.. It is informative to rearrange Eq. (6. and 3d16. 0. then holR must be even For additional information on the fluid dynamics of lubrication see Cameron (1966) or Pinkus and Sternlicht (196 1). Kernblowski. In other words. the final result for the flow rate is Although it is tempting to choose xo= R. L. h changes by an appreciable fraction of h. the expression for q is rewritten as The functions a(X). Mechanics o f Polymer Processing. the cylinder is fixed and the surface moves at a constant velocity U. at X=. H. Atkinson. M. The minimum separation distance is b. 13: 1720. A.5 the gap height is underestimated by <7% for any value of ho/R. AlChE J. Low Reynolds number developing flows. which are the functions needed to compute 8 and q [see Eqs. 1992. which facilitates evaluation of the integrals in Eq. Smith.&<<R. G . (6. Thus. Beauchamp Tower's experiReynolds. A. Measurements of velocity profile in developing liquid flows. moves parallel to a planar surface without rotating. London. Chandrasekhar. J. (6. Cambridge.641) and integrate from a "far upstream" to a "far downstream" location to evaluate q. Philos. 1985. E. for pressure and Row calculations. and requiring that xo S 0. Using Eq.658). 1966. Trans. Luminar Flow and Convective Transport Processes.whe. M. care is needed because the lubrication approximation used to derive Eq. 1981. In other words. If Re exceeds unity. 1965. 1967. (6. at IXIRJ =0. Elsevier.. AlChE J. ments.Y UNIDIRECnONAL FLOW that a long cylinder of radius R. The Principles of Lubrication.Brocklebank. The distance between the surfaces is given by : i which is exact in the limit holR+O.1 to 0. 1886. Choosing ~ . (6. to good approximation. (6. 1961. the geometric requirement for the lubrication approximation is expressed as Thus.2. 1977. and B. including an experimental determination of the viscosity of olive oil. f Polymer Processing. Cameron. and I. Pearson (1985).664) and (6. 11 : 995999. A 177: 157234. but large enough that. Retaining only the first two nonzero terms. P.662) ' 0 evaluate the pertinent integrals. Hydmdynumic Stability. M. we want xo to be small enough that we can use the lubrication equations. making the dynamic requirement of Re lxoi~J c< 1 more restrictive than Eq.660). remains to be determined. the ratio b(3)lc(3)is within 1% of the value of b(m)l c(m). New York. 15: 548553. C. G. Entrance region flow. and Tanner (1985). I . Boston. Hydrodynamic and Hydromagnetic Stability. and c ( X ) range from zero at X = 0 to asymptotic values of ~ 1 2 d . = 3 t = 3 ( 2 h ~ to ~ )ensure '~ that P 0 at the ends of the region being modeled. It is desired to determine the flow rate q in the gap between the surfaces. Using Eq. London. Christiansen. (6. and J. New York. respectively. S. Middleman. (6.2 R for the lubrication approximation.
.
Figure P614.":. A cylindrical tank connected to an outlet pipe. .  .* .A large bubble rising in a liquidfilled tube. The densities of the cylinder and liquid are p... is specified. %be with Spatially Periodic Radius Consider steady flow in a circular tube in which the radius varies with axial position as R(z) = R. respectively.R. derive an expression which relates the viscosity to the cylinder velocity (Lr) and the other parameters. local rectangular coordinates can be used... i. . . determine v.'. as shown in Fig..  . .I I I I Air I 1 b Figure P612. (b) enlargement of area shown by dashed rectangle in (a). P612. Q. Assuming that the flow in the gap between the cylinder and tube is fully developed.'. ' Liquid . allowing the tank to drain. Consequently. .c<R. Bubble Rising in a lhbe Consider a large gas bubble rising in a liquidfilled tube. 614. P614(b).. 61 may be used.: . (b) What is required for the pseudosteady approximation to be valid? 3 (b) Determine v(c z ) and 8 ( z ) using the lubrication approximation. . 613. with the cross section taken to be a square of side length 2H. . Wall X . .. ' That is.R. and p. The instantaneous liquid level in the tank is h(t) and the initial level is h. as shown in Fig. '1 612..'. Q. .(y... 2). : ... as shown. (a) Overall view.:. U . .. I . .) 615. State any additional geometric or dynamic assumptions which are needed.. Assume that the bulk of the bubble forms a cylinder of radius RB and length L.:$. + R. then the bubble will be highly elongated. Assume that the only significant flow resistance is that in the pipe. The coordinates shown in Fig. The gap between the concentric cylinders is of thickness HER. (a) Assuming that @ldr is given.*+.) and tube radius (RT) are such that VB113>>~. If the bubble volume (V. . + . The liquid film between the bubble and tube wall is expected to be very thin (H . : . U.. The top of the tank and bottom of the pipe are open to the atmosphere. l. . Such a system is shown in Fig. Time Required to Drain a Thnk A pipe of radius Rp and length L is connected to the bottom of a cylindrical tank of radius R. (Hint: The bubble velocity is not uniform. .+ H I Figure P613. . Falling Cylinder Viscometer A possible (although not commonly employed) approach for measuring the viscosity of a Liquid is to determine the speed at which a solid cylinder falls through a closely fitting tube filled with the liquid. ? i  "' . It is convenient to fix the coordinate origin on the wall. The volumetric f l o w rate. . the mean value for the bubble. local rectangular coordinates can be used to describe the flow in the film. \I (a) Assuming that flow in the pipe is pseudosteady..I % .''. Flow in a Square Duct Consider fully developed flow in a channel of square cross section. so that the velocity at the gasliquid interface differs from 0.R. . 7 : 611. Assuming that the gap is thin (H/Rl << I).' ... (b) Relate #ldx to the volumetric flow rate. . s . P614(a). Assuming that the flow in the thin liquid film is fully developed and that the bubble dimensions are known. .FalIing i:y linder viscometer.). g 'q. For a tube of length . . :?.. respectively. . ' . and that the pipe Row is Laminar and fully developed.A. . (a) Under what conditions will the lubrication approximation be applicable? . . At r = O a valve is opened. derive an expression for the steady rise velocity of the bubble. sin (7). State any additional geometric or dynamic assumptions that are needed. the mean radius is Ro and the amplitude and period of the variations are R1 and 2 l . . . as shown in Fig.. P613. ' . determine the time required to &main the tank (but not the pipe).. i i 1 RT /.
. it is desired to derive expressions for the forces on the cylinder in the limit h.. Figure P619. an undesirable situation.. some of the liquid will recirculate instead of being drawn out with the wire. (b) If h. steady condensation of a vapor on a vertical wall results in a condensate f i l m of thickness Wx) which flows down the wall at a mean velocity u(x)... = 0). : A step in the manufacture of certain polymeric films is to pass the liquid polymer between a pair of calender rolls. normal to the vaporliquid interface.(x. is the hydraulic permeability of the membrane. The forces and torque referred to below are all expressed per unit length of cylinder. how will the overall pressure drop compare with that for a tube of constant radius.(x) is related to the transmembrane pressure drop AP(x) by v. A 618. assuming that S(0)=O..The pressure at the exit of the bottom chamber is p2.Wire coating process.. Each roll has a radius R and is rotated at a constant angular velocity o.? 619. We consider here a simplified model of this process for a Newtonian fluid. is given... F. the die dimensions... .m y UNIDIRECTIONAL n o w ' I(b) Relate dSl& to u(x) and v .. Determine Nx) and u(x). Flow in + Flow out P=9. (not known in advance). P617. find the equation for the streamline which separates the forwardflow and recirculating regions. expressed as the velocity v. derive the relationship between q and 8. the pressure difference maintained across the die ( A 9 = Po . The minimum halfheight of the gap is such that $/R<< 1. Assuming that h. Show that there is no net contribution of the viscous stresses to F. and sketch several streamlines to explain why.. As shown in Fig. the final coating thickness (h_) will depend on the wire velocity (U). and the liquid properties. the counterrotation of the rolls forces the liquid through the narrow gap between them. F. P620. Also.8. InLine Particle Filter Figure P616 shows the arrangement of a type of inline filter designed to remove occasional particulates from a liquid. assuming that the fluid dynamic conditions are such that v.Inline particle filter. Assume that the lubrication approximation is applicable. Calendering (a) State the relationship between u ( x ) and Wx). 1 P 617. 621). U. Assume also that 9 = 0 in the upstream reservoir and at x =x.giving a linear velocity oR at its surface.and the die geometry. as shown in Fig. G=O). The local transmembrane velocity v. (b) What orderofmagnitude criteria must be satisfied for the results of part (a) to be valid? I '. y) in the liquid is parabolic in y. To guide your sketch. y) and 9 ( x ) in the gap between the cylinder and the flat surface. sketch several streamlines for a case where ho>h. Condensation on a Vertical Wall (Constant Rate) As shown in Fig.. In general. and that the coating is thin compared to the radius of the wire.65.YL). and thereafter behaves as a solid. where k. and the dynamic pressure is constant there at 9. Assume that the f i l m separates from the rolls at some position x=x./R+O. Express these constraints in terns of 6 and u. (a) Evaluate v..e. Assume that the rate of condensation.(x. .=k. R = R. is too large. the volumetric flow rate per unit width. Wire Coating In one type of continuous coating process a wire is pulled from a reservoir of the liquid coating material t h u g h a tapered die. 616.AP.e. Figure P616. P619. (c) Identify the orderofmagnitude criteria which are sufficient to justify the fluid dynamic assumptions. to A??. i I t i . determine the maximum opening for which there will be no recirculation (h = h)... The flow resistance within the top chamber is negligible.. (a) Assuming that the lubrication approximation is valid for flow in the bottom chamber. so that rectangular coordinates can be used. It is desired to relate the overall pressure drop to q. (a) Relate h. : L>>t. is constant. (b) Show that there is no lift force (i. Sliding Cylinder For the physical situation of Example 6. (c) Evaluate the drag force. (d) Show that no applied torque is needed to prevent rotation of the cylinder (i.
). (b) It is suggested that if the top surface is a gasliquid interface. Rotating Cylinders As shown in Fig. and the final f i l m thickness (h. and replaced by liquid flowing normal to the bristles from the interior of the brush. = 1.lh. giving a linear velocity wR at its surface. In the transfer of a liquid to a flat surface by a brush. Each cylinder has a radius R and is rotated at a constant angular velocity w . Flow in a Thin Liquid Film Driven by Surface Tension Consider the twodimensional situation shown in Fig. Figure P620. A simplified model for this process. y) and P(x)in the gap. The nature of the top surface is (for the moment) unspecified. Liquid is removed from the brush by a tangential force at the surface acting parallel to the axis of the bristles. Show that this idea is consistent with an interfacial stress balance. y) and show that 9 = 0 everywhere. .226. A gradient in surface tension might result. Corotating cylinders separated by a Y n liquid film. Show that h. then the mysterious stress of magnitude K could be due to a gradient in surface tension. (a) Use the lubrication approximation to derive expressions for vx(x. then H cannot be constant. which side of the container must be hotter? (Continue to assume here that H is independent of x.. and evaluate dyldr.What is needed to ensure that the geometric and dynamic requirements for the lubrication approximation. in which the cylindrical bristles are replaced by flat surfaces of indefinite height. assume that two parallel cylinders separated by a thin f i l m of liquid are rotated in the same direction. (a) Assume that the liquid film is thin enough that the flow is fully developed (except very near the sides).Model for the brush application of a liquid onto a flat surface. the brush acts as a reservoir for the liquid. (a) Determine vx(x.Calendering. A thin liquid film with a constant shear stress at the top surface. 622. assuming that the mean film thickness is given as H. Determine dYldr and v.Figure P622. was i' t (a) (b) . Ignore wetting phenomena at the sides. in which there is a shallow layer of a Newtonian liquid in a rectangular container. 621. Figure P623. Liquid tr. [Hint: Show that the pressure gradient in the liquid found in part (a) is inconsistent with a constant pressure Po at the gasliquid interface.) (d) Use a lubrication analysis to determine H ( x ) . will appear as a parameter. The depththere is a constant shear stress (5 tolength ratio is such that HIL<< 1. If dyldT<O (as in any pure liquid) and K>O. P622. are satisfied? 623. x. (b) Calculate the torque (per unit length) which must be applied to either cylinder to maintain the rotation. The minimum gap halfheight is such that ho/R<< I. the brush bristles are parallel to the surface. In other words. Assume that the lubrication approximation is valid and that 9 = 0 in the bulk fluid. P621. ( d ~ / d r l < < 1 and ReldH/drj << 1 . Brush Application of a Liquid onto a Surface Figure P621. in the limit ho/R+O. for example. except that = K) and no normal component of velocity at y = H. from a temperature gradient in the x direction.(y).) (c) Show that if the top surface is in fact a gasliquid interface. @) Use conservation of mass to determine x. The fluid is bounded at the bottom and sides by solid surfaces.
The "brush" is regarded as stationary and the surface is moving at a velocity v. the behavior of colloidal dispersions (small L).as shown in Fig.. suspension rheology (small particle size). its failure to provide an adequate approximation to the flow far f r o m a sphere at arbitrarily small but nonzero Reynolds number. U. Several approaches for the solution of creeping flow problems are then discussed. z). ". The chapter closes by focusing on a rather unexpected aspect of Stokes' equation. starting witb examples of unidirectional and nearly unidirectional flows. 'Ldictated by . The use of singular perturbation theory to obtain corrections to Stokes' law for finite Reynolds number is described. The chapter begins with an overview of some of the distinctive properties of !*creeping flow. . and assume that far from the brush there is no motion in the liquid relative to the surface..  3 . & . Following that are derivations of some fundamental results pertinent to suspensions. also termed creeping Jow or Stokes flow. Chapter 7 CREEPING FLOW (a) Determine the fully developed velocity. including a derivation of Stokes' law for the drag on a sphere. #. '. 1 INTRODUCTION applications to numerous technologies and natural phenomena involving small length kr The theory of low Reynolds number flow. = U. v. which result from the virtual absence of inertial effects. Singular (or pointforce) solutions of Stokes' equation.:. The focus of this chapter is the creeping flow of . and the processing of polymer melts (large p).(y. including Faxen's laws for the motion of a single sphere and Einstein's result for the viscosity of a dilute suspension of spheres. Figure P623(a) is a view parallel to the axis of the model bristles.) {. which have found much use in describing the behavior of particulate systems. . P622(b). Newtonian fluids. @) Relate the final thickness L of the liquid film40 H. The application of the stream function to twodimensional problems is introduced. has 5. (Hint: Calculate the volumetric flow rate Q per channel. Neglect the width of the model bristles. and the fluid properties. which are spaced at a distance 2H. f I.(L) and velocity (U) scales and/or extremely viscous fluids. are discussed next. namely. \> . 1 : y .suggested by Taylor (1960). Examples include flow in porous media and in the microcirculation (small L and U).
5 d 3: 9 u o g .5 0 gZ& .g =Ns 3 '2 a . g ". 5 '2 Z 0 + "5 A " Z 8 5 x u M.t: n a sags mi3 PO6d 3 c z * .S f z .2 g " 23 h !z .g 82 3 0 Q g $2 gl = ?2 d a % a 2r m * .! 0 a V3 0 P) n 5a 4 3 .s.% 0 6 .g 5.S 8 8 .3 as.
216) leads to (7.218) which indicates that the force on SI is transmitted fully to S2. (7. (7. (7. in the fluid (e. (7. and let u 'be in Eq.214) and (7. One way in which Eq. (7. As two or more distinct parts.210) begins by using the constitutive equation for an incompressible.. pp. the proof of Eq. If u is known. and the boundary conditions. Now let v" and d the velocity and stress for creeping flow of the same fluid in the identical geometry. the Lorentz reciprocal theorem states that small Reynolds number.) One other general result for creeping flow which must be mentioned is the reciprocal theorem.29) may be used is in calculating the fluiddynamic force on a submerged object (fluid or solid). Following Happel and Brenner (1965). but if the shape of the object is irregular. (1965. Lorentz in 1906 [see Happel and Brenner. (7. Then. d:rf=2prw:r'.=O. but with different boundary conditions on S.214) (7.61.27) be valid. so that the pressure is written as 9 instead of P Forming the doubledot product of each term with I"' gives : with the rotational symmetry a b u t the z axis. :rt=d:rf.215) Interchanging the primes and double primes gives the analogous result. (This simplification is much like that discussed in Example 5. as mentioned above. As illustrated by Example 7. A. Newtonian fluid to write d as Gravitational effects have been excluded. reported originally by H.27). so that I ?' : r"= In : I" and. which emphasizes the simplifications inherent in the pseu This elegant result is used in many derivations in creeping flow theory. Let v' be a velocity field which satisfies continuity and Stokes' equation throughout some volume V bounded by the surface S (which may consist of ' be the corresponding stress tensor. (7. whose surface corresponds to S1.=O. imaginary sphere)."'f=0.31) It can be shown that 6 : rv=v. gravitational and other body forces are excluded. (7. The doubledot product of any two tensors is commutative. (7.which requires that Eq.29) provides the option of calculating the force by integrating n . from Eqs. the continuity equation. o over some more convenient surface S. Eq. 8587)].52. v4=Rosin0 atr=R.212) reduces to @I re 71. whereas the axisymmetric geometry implies v. (7.215). as above). a large. Rotation of a sphere about the z axis at : rW=2~rf : r. v. it leads to remarkable simplifications in certain force calculations. and integrating over K Eq. Using the identity . The rotation ensures that v+ will be important. so that Eq. in connection with pressure forces in static fluids.g.
ppliedto both disks causes each to move toward the midplane at velocity U(t). It is assumed .. (7. using the creepingflow velocity field in the r and 0 components yields It remains to be confirmed that this form for v satisfies Stokes' equation. then. Note from Table 59 that if Re were not small. 0). so that the torque on a differential element of surface U Sis given by . Because the leading terms do not . (7. SubThe differential equation for f(r) is equidimensional.ly Unidirectional solutions 297 are also consistent with the hypothesis that v.dHldt. to Eq. There is no exact. 8)R sin 8 dS.Iv<< 1 and H. (7. the r and 8 components of Stokes' equation reduce to aS/ar=a9/aO=O. dG = rr4(R. where (as noted above) 9 is constant throughout the fluid. d ~ R= (7. We conclude that Eq.=v. at least one of those unknown functions would have to depend on both r and 0. (7. The boundary conditions require that a = O and b = ~ ~ fluid velocity is given by v4(c O)= uR3sin 8 9 Of interest is the torque (G) exerted by the fluid on the sphere.314) and (7.313). (7.312) andor (7. In particular. analytical solution to the rotating sphere problem for moderate or large Reynolds number.32 Squeeze Flow As shown in Fig. which is in the G<O.a r 2 a 3 r Z ~ r ( ar ) +r 2 1 sm eao ( 4 sin 6 ' 3 ae) r2 2 : s eb (7. the simplest f o m for the velocity is evidentIy v. it is assumed that two disks of radius R . (7.don is 2H(t).3I).(8) and C. forcing out the incompressible fluid between them. Squeeze flow. the general solution is : SO othat . Why.35) leads to the conclusion that n= 1 or 2. we assume that Re = H. the inertial terms in the r and 0 components of the NavierStokes equation would require that 9 =P(r.The disk separa.37) to evaluate r. w stitution of this trial solution in Eq. To obtain creeping flow with negligible edge effects. 2 sin 6 dB d+.  This ability to obtain a differential equation and boundary conditions for f(r) confirms that the velocity is of the form given by Eq. (7. where the fluid exits.34).required by Eq.34) Substituting the assumed form for v+ into Eq. +6 direction.=O. is the initial (maximum) separation and Uo is the maximum value of U(t). so that U = . so that Ngure 72. A point on the surface is at a distance R sin 8 from the axis of rotation. (%33) reduces not just Stokes' equation. A normal force F(t) :. indicating (together with the symmetry) that 9 is constant throughout the fluid. was it necessary to assume that Re<< l ? The answer concerns the pressure and the other components of momentum. In other words. the assumed form of the velocity implies that the inertial terms in the &component of momentum vanish identically. thereby *elating Eqs. A force F(t) applied to each of two disks cause:s them U(t). to It is noteworthy that the use of Eq. The flow is assumed to be axisyrnrnetric. = 0. v e = 0. The 4 component of Stokes' equation becomes o=. Paying spccial attention to the dependence of v4 on 8.315) indicates that these two expressions for 9 cannot be reconciled using any choice of the functions C. the where a and b are constants.=O.:' ) : I .8re brought together. (7.IR<< I.further that 9 = 0 at the edges of the disks. with v e = O . ' ' h u n p l e 7.:. (7.match.squeezing out the fluid between them. all terms involving 0 cancel or factor out. Thus.37) satisfies all three components of conservation of momentum only for vanishingly small Re. With v. . Inspection of Eqs. 72. we find that Finally. integrating over the sphere surface gives The surrounding fluid resists the rotation of the sphere.= v.(r). leaving o=r2df .34) and the boundary conditions. (7.2 5 dr ( dr) d f(R)=oR. f(w)=O. i t h solutions of the form f= I".38) Using Eq.U. where 2H.33).33) full NavierStokes equation. but also the move together at velocity . v* =fir) sin 0.
Stokes' equation reduces to o=a +9 p _ d8ar 3pUr 2 H3 ' (7.320) implies that . Using Eqs.This 7. HZ8 9 2p a.cmmRmLow The small Reynolds number makes the problem pseudosteady. implies that 9 =Y(r t ) only and that we should focus on the r component of Stokes' equation. (7.323) at z =H.. z I $(m.U.325) and requiring that 9= 0 at r = R gives 3~ U P(")=.3. by analogy with the lubrication appmximation (Section 6.326) a2vr dz2' r (7.330) wheref(r) is to be determined. (7.317) vr(r.16) The boundary conditions for v. The applied force F(t) can be related now to the velocity U(t) and the halfspacing H(t).=O at z=H. and therefore a 9 / d c depends on offered by the fact that the boundary conditions for v. Thus. ~[(i) &) 1. In addition.316) and applying these boundary conditions gives ~(b) [I (. vz=O at z=0.HolR (see below). Vc= I. . gives The function f (t) is obtained now by evaluating Eq.325) Integrating Eq.4 H 4 1 ( : T I . Ignoring any gravitational or buoyancy contributions.318) and (7.=U at Z = H . (7. . and the assumption that HdR<<l implies that the flow will be almost unidirectional. t ) = .6). A flue is (7.320). (7.321) do not require that vz depend on r. we assume that the pressure gradient is of the fonn a9 =r f ( t ) 9 dr (7. r dr az (7. (7.I ) = Zntegrating Eq. the continuity equation requires that v.322) (7. o H3f z 1 z 3 = [(ZJ)T(ij) ~ 1.so that Uo may be interpreted as the initial disk velocity.. a'1ar are both ~ r o ~ o d o nto a l r. Accordingly. are evaluated by using the continuity equation. R ~ (7.322).(c z. t) only. That is. are %=o az at .319) over z leads to an expression for v. [l(ir].TI.0'2 v.320) 0 To the integral we need to know how v. (7. z. v. The smallness of v..) a a.318) he radial velocity depends on r through d9ldr (which remains to be determined) md depends on t through both H(t) and a9Idl: The pressure gradient and v.4 STREAM FUNCTION SOLUTIONS md3using ~ q (7. f ) = a vZ(z.327) (7. 3 1 2 3 (7. . . Assuming that vz= vz(z.322) in Eq. Eq.it is found that V.(Z.lv.319) Integrating Eq. = ...329) 1 d (rv. the force applied to either plate is given by (7.. where 3PU f ( t ) = 2 H~' . . (7. (7. (7. (7. (7.328) v. the relatively narrow gap will make d2vr/d? the dominant viscous term.323) For a constant force E the maximum value of U is obtained at r = 0 .)+3=0.
and Stokes' equation will be valid close to the comer.412) for g.41 Flow Near a Corner Consider the situation depicted in Fig. A flow like this might occur. One surface slides past the other at a velocity U.46) and (7. In such problems the stream function (Section 5. (7.415) simplifies to The velocity components for this special case are v.14) i B.( 0) sin B]. leaving Expressed in terms of J. for example. secondorder differential equations involving f(B) and another function g(0). After solving Eq. for a planar flow in cylindrical coordinates is defined as (Table 511) To determine the stream function we must solve Eq.47). it is found that all terms containing r factor out. (7.(a. (7. (7.2 I)COS R . which suggests that the distance from the comer (r) be used in estimating the relative importance of inertial and viscous effects. B) is known. (7.9) provides an effective approach for obtaining a solution.4l). There is no fixed length scale in this problem.41).4.410) to evaluate the four constants in Eq. (7.415) The velocity components for any angle a can be obtained now. (7. (7. 73. where C = bl2. general solutions of the streamfunction form of Stokes' equation are discussed briefly. D=a/2. Examining the boundary conditions in Eqs.(e)= .sin2 a [c? sin B . This problem is one of several involving sharp corners which are discussed in Moffat (1964).49) is solved by rewriting it as two coupled. The stream function. the strearnfunction form of Stokes' equation is simply This ability to obtain a problem for f(0) which is consistent with Stokes' equation and all boundary conditions confirms that the assumed form of the solution js correct. (7. (7. (7.414). by using Eq. @(c 0). and A and B are additional constants. One of these analyses leads to Stokes' law for the drag on a sphere. (7. Equation (7. solid surfaces which meet at an angle a.48) Substituting Eq. and in particular the noslip condition at the moving surface. Eq. (7. it appears that the solution may be of the form 9(r. @ = rS(0). Once +(r.411) becomes d2f+f= d82 a cos 0 b sin 8.. as L I: f(B)=A cos 8+Bsin 9+CBcos B+DBsin 8 . these boundary conditions are written in terms of ((.48) in Eq.42) with these boundary conditions. (7.26). the stream function is found to be Ur Z N c 0) = 2 . Using Eq. In addition to the examples. in which a fluid is bounded by two planar. the pertinent Reynolds number is Re(r)= Urlv. (7. if desired. The general solution for this nonhomogeneous equation is Using Eq. (7.(sin a)B cos 0.tude. 8=0 and B= a. the boundary conditions become : : 1 From Eq. a cos a)@ sin 01. Flow near a comer due to a moving surface. U I [(:. For perpendicular surfaces (a= d 2 ) .415) in Eq. the velocity components are calculated from Eq. It is desired to characterize the Row in this region.42). . Example 7. are expressed as . Eq. In other words. such that The noslip and nopenetration conditions at the solid surfaces. + where a and b are constants. Both surfaces are assumed to extend indefinitely. near the opening of a die in a coating or extrusion process.sin Figure 73..41). (7.
so that in writing Eq.' tion constant was set equal to zero. corresponding to Eq. are sin^. this is true for any angle a. e)=o. Substitution of Eq.428) into Eq. we conclude that $(. d2 sin 8 d E2 2 ar2 ( 1 r ae slneae' a) Equation (7.427).420) of the form (7.422) (7. in that it contains the first solutions to viscous flow problems. ' :.1. The stream function.42 Uniform Flow Past a Solid Sphere The slow motion of a solid sphere relative to a stagnant.428) is correct. (7. (7. n u s . and . (7. The absolute value of the stream function is arbitrary. are and the corresponding streamfunction form of Stokes' equation is 3 ae (.425) Figure 74. including fluid mechanics.Another distinctive aspect . and the drag on the sphere.424) and (7.427). with uniform velocity U far from the sphere.422). (7. first analyzed by Stokes (1851). the boundary conditions of no slip and no penetration at the sphere surface. E4 EZ(E~). (7. 8)1 ' r 2 u sin2 0 * (7. suggests that we seek a solution to Eq. (7.427) the integraY . v. the boundary conditions for f are ' " ' j Much like the preceding examples. neither v. although an infinite force is not allowed. (7. (7. solid mechanics.425) and (7. 8) coordinates. I. The boundary conditions for the stream function at the sphere surface. but the E~ operator defined in Eq. Flow around a stationary. It will be seen that this choice gives +h=O as the streamline corresponding to the sphere surface. The required behavior of $ far from the sphere. the pressure. nor v.' As shown in Fig.427) i . . The paper in which he reported his cdcdation of the drag on a sphere (Stokes 1851) is one of the most famous in the history of fluid mechanics.(R. so that their product is finite. $. for an axisymmetric flow in spherical coordinates is defined as (Table 511) e) =o. : From Eqs. solid sphere. depends on r. in Gillispie.420) results in f t t t t L m n ghis career at Cambridge University. (7. pp. the fact that all terms involving 8 could be factored out to obtain Eq.Surprisingly. (7. 7479). Although of this problem is that 9P and T the stresses singular at r =0. That is. This illustrates the fact that an infinite stress at a point is not necessarily unrealistic.430).423) vr(m. (7. Substitution shows that n can assume the values 4. It Prsedcd by some 10 years the first published derivstions of Poiseuillevsequation (Chapter 6). It is desired to determine the velocity. Integrating Eqs. 2. 8). with solutions of the form f = $. V&DO. assuming that Re = UR/v<<1. both vary as rI. 74.. 13. 8) =f(r)sin2 8.429). the surface area varies as r and the stresses as r'. George Stokes (181S1903) made major theoretical and experimental conMbutions to many areas of physical science.423). (7. $(r.429) is another example of an equidimensional equation. confirms that Eq.U sin 0.(. as given by Eq. (7. 1 a$ 1 Yo= a$ r sin 0 ar . . In spherical (r. Vol. and uniform flow far from the sphere. optics.420) is similar to the biharmonic equation. and therefore are singular at r=O. is the most widely known problem in low Reynolds number hydrodynamics.421) is not quite the same as the Laplacian (V2) involving r and 0. Stokes shares credit (with Navier) for deriving the momentum conservation equation for Newtonian fluids and was largely responsible for the concept of the noslip boundary condition at a solid Surface. and the use of fluorescence (Parkinson. The streamfunction conditions for the uniform flow far from the sphere. v = U = U e. are expressed as v. corresponding to Eq. . it is convenient to choose a reference frame in which the sphere (of radius R) is stationary and the fluid far from the sphere is moving at a uniform velocity. Equation (7. the force on an imaginary cylindrical surface of radius r remains finite in the limit r+O. He is known also for his work mathematics.426) and comparing the results. Example 7. and that all of the original boundary conditions are satisfied using Eq. (7. (7. 0) + r2u sin 0 cos 0.(R.428) $(1. 1976. 0) . 0) + U cos 0. viscous fluid.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
liquid films. Steady heating of a solid plate moving at a constant speed. In forced convection the flow is caused by an applied pressure or moving surface.. where it is the effect of temperature or solute concentration on the fluid density which causes the flow. which indicates the importance of convection relative to conduction or diffusion. A key parameter in any forcedconvection heat or mass transfer problem is the Pkclet number (Pe). (9. and examples are given to illustrate their evaluation for laminar flow 'Ubes. the upstream value is specified as To and the downstream value is determined by the plate velocity ' and heating rate. Because Pe is the coefficient of the convective terms. and the velocity and differential operators have been made dimensionless using the characteristic velocity U and characteristic length L. In free convection the fluid mechanics and heat and/or mass transfer problems must be considered simultaneously. Anention is given then to heat and mass transfer coefficients in confined flows. This is in contrast to free convection or buoyancydrivenflow. Pe ' can be interpreted as the ratio of a characteristic velocity for convection to a characteristic velocity for conduction or diffusion.e. It is assumed that T = T(x. and it is a known quantity in the energy or species conservation equation. the velocity can be determined first. and transport in turbulent flows is discussed in Chapter 13. as discussed in Chapter 12. Heat and mass transfer in unconfined laminar flows is the subject of Chapter 10. Although several of the derivations in this chapter apply to forced convection in general. The chapter concludes with an analysis of solute dispersion in tube flow. as depicted in Fig. the energy equation for a pure fluid is generalized to include terms associated with the compressibility of the fluid and with viscous dissipation of energy. it evidently indicates the importance of convection relative to the molecular transport processes. Certain general features of heat and mass transfer in confined flows are illustrated by a simple physical situation. A comparison of Eq. effects which were neglected in the approximate form of where @ is the dimensionless temperature or concentration. whereas the rest of the plate is effectively insulated (i. Sh) a . (5.lL both have units of velocity. The dimensionless heat transfer coefficient (Nusselt number. The dependence of Nu and Sh on axial position and on the type of boundary condition at the wall is discussed. The problem to be analyzed involves steady heat transfer to a moving solid. A plate of thickness L is assumed to be moving at a speed U in the x direction. Indeed. 91. Far "upstream" and "downstream" from the heated section the plate is assumed to approach constant temperatures.1015) reveals that Pe plays the same role in the energy or species equations as Re does in the NavierStokes equation.10. of course. In the first of the examples which I Insulated Region I I I Heated Region 90 I I I Insulated Region Figure 91. called Taylor dispersion. Nu) and mass transfer coefficient (Shewood number. After that.21) with Eqs. the examples involve only confined laminar flows of the types considered in Chapter 6. In mass transfer this ratio will differ. in attempting to draw meaningful conclusions from the magnitudes of either Pe or Re. Under those conditions both equations have the form 9 . and when distinctions are needed we will denote the P6clet number for species i as Pe. In the region 0s x 5 K the top surface receives an incident heat flux go. and other confined geometries. 1 INTRODUCTION This chapter is concerned with forcedconvection heat and mass transfer in tubes. noticing that d L and D. which is the situation of interest here.14) or (5. in general.Chapter 9 . as in Chapters M7 and is more or less independent of the heat or mass transfer process.2 PECLET NUMBER The significance of the P6clet number is clearest when one examines the dimensionless energy or species conservation equation for steadystate problems without volumetric source terms. as already mentioned. . y) only.. In such cases no other parameters are present in the differential equation. the heat fluxes at the surfaces are negligible). The chapter begins with a discussion of certain approximations which are justified when Pe>> 1. defined. 9. : for each species in a mixture.FORCEDCONVECTION HEAT A N D MASS TRANSFER IN CONFINED I LAMINAR FLOWS the energy equation presented in Chapter 2. Thus. The identification of the velocity and length scales which are appropriate for a given problem is crucial.
Steady heating of a solid plate moving at a constant speed. and examples are given to illustrate their evaluation for laminar flow in tubes. and it is a known quantity in the energy or species conservation equation.1014) or (5. whereas the rest of the plate is effectively insulated (i. After that. In the first of the examples which / Insulated Region I Heated Region 90 I I1 Insulated Region ! 1 Figwe 91. the energy equation for a pure fluid is generalized to include terms associated with the compressibility of the fluid and with viscous dissipation of energy. Sh) are defined. as discussed in Chapter 12. .e.1015) reveals that Pe plays the same role in the energy or species equations as Re ' does in the NavierStokes equation. A plate of thickness L is assumed to be moving at a speed U in the x direction. FORCEDCONVECTION HEAT AND MASS TRANSFER I N CONFINED LAMINAR FLOWS 9. F a r "upstream" and "downstream" from the heated section the plate is assumed to approach constant temperatures.Chapter 9 I i the energy equation presented Chapter 2. Under those conditions both equations have the form 9. It is assumed that T = T(x. the velocity can be determined first. In such cases no other parameters are present in the differential equation. the heat fluxes at the surfaces are negligible). and other confined geometries. The chapter concludes with an analysis of solute dispersion in tube flow. the upstream value is specified as To and the downstream value is determined by the plate velocity and heating rate.. as depicted in Fig. Nu) and mass transfer coefficient (Sherwood number. In mass transfer this ratio will differ.2. In forced convection the flow is caused by an applied pressure or moving surface. y) only.1) with Eqs. as in Chapters 68. it evidently indicates the importance of convection relative to the molecular transport processes. Although several of the derivations in this chapter apply to forced convection in general. where it is the effect of temperature or solute concentration on the fluid density which causes the flow. and transport in turbulent flows is discussed in Chapter 13. In free convection the fluid mechanics and heat andlor mass transfer problems must be considered simultaneously. The dimensionless heat transfer coefficient (Nusselt number. noticing that d L and DilL both have units of velocity. (9. This is in contrast to free convection or buoyancydriven flow.2 PECLET NUMBER The significance of the Pkclet number is clearest when one examines the dimensionless energy or species conservation equation for steadystate problems without volumetric source terms. and is more or less independent of the heat or mass transfer process. The identification of the velocity and length scales which are appropriate for a given problem is crucial. the examples involve only confined laminar flows of the types considered in Chapter 6. A key parameter in any forcedconvection heat or mass transfer problem is the Pkclet number (Pe). . 91. Attention is given then to heat and mass transfer coefficients in confined flows. t Certain general features of heat and mass transfer in confined Rows are illustrated t by a simple physical situation. In the region 0 s x 5 K the top surface receives an incident heat flux q. which indicates the importance of convection relative to conduction or diffusion. The chapter begins with a discussion of certain approximations which are justified when Pe>>l. in attempting to draw meaningful conclusions from the magnitudes of either Pe or Re. which is the situation of interest here. and when distinctions are needed we will denote the i Pkclet number for species i as Pe. Pe can be interpreted as the ratio of a characteristic velocity for convection to a characteristic velocity for conduction or diffusion. in general. effeots which were neglected in the approximate form of where @ is the dimensionless temperature or concentration. Indeed. Because Pe is the coefficient of the convective terms. Thus. called Taylor dispersion. as already mentioned. Heat and mass transfer in unconfined laminar flows is the subject of Chapter 10. The dependence of Nu and Sh on axial position and on the type of boundary condition at the wall is discussed. The problem to be analyzed involves steady heat transfer ' to a moving solid.. Liquid films. of course.. (5. A comparison of Eq. [ for each species in a mixture.1 INTRODUCTION This chapter is concerned with forcedconvection heat and mass transfer in tubes. and the velocity and differential operators have been made dimensionless using the characteristic velocity U and characteristic length L.
214H9.32. (9. the approximate results ("Appror. and that for the heated region is found from Eq. The result is i or. and that the temperature and axial heat flux be continuous where the regions join ([=0 and l= A). The form of the governing equations is simplified by using the dimensionless quantities The dimensionless average temperatwe is governed by (9. but it is unnecessary in the present problem to assume that temperature variations in the y direction are small. 0) = 0. Temperature profiles for a moving plate heated for 0 5 6 5 10. We focus now on the behavior of the temperature in the heated . as computed from Eqs. an exact solution is obtained for any value of Pe. An exact solution for 8(5) is obtained by dividing the plate into three regions. respectively.213). (9. 91 are Approx. ay JT c \ ' Although this linear boundaryvalue problem can be solved using the methods of Chapter 4. a less detailed solution will suffice here. ' Example 9 .21la). y) in the twodimensional problem of Fig. The simplified analysis in the second example illustrates the approximations which apply when Pe is large. . a crosssectional dimension is usually the appropriate length scale.(x.22 Heating of a Moving Plate at Large Pe The analysis just completed did not place any restrictions on Pe.T ANb MASS TRANSER Yeclef number follow. 2 : 1 Heating of a Moving Plate at Arbitrary Pe The energy equation and boundary conditions for T(x. (9.2. The general solutions for the upstream and downstream regions are obtained by integrating Eq. ' * In Eq.210) we selected as the characteristic dimension the plate thickness (L) rather than the length of the heated region (K).220). defined as Figure 92. (9. The "Exact" c w e s are based on Eqs.2llb). are shown by the solid curves in Fig. The ordinary differential equation goveming T ( x ) is obtained by averaging each term in Eq.22). L " ' ~ ' 1 " " " " ' ~ " ' 1 r " 1 " ' 4 . Suppose that it is desired to know only the crosssectional average temperature at a given x. Choosing L instead of K will enable us to examine situations where KIL>>l and the exact value of K is unimportant.216)with A = 10. The resulting solution is The integration employed here to obtain a onedimensional problem is similar to that used in the fin analysis in Example 3. For confined flows in general. 92. using the boundary conditions at the top and bottom surfaces. (9. (9. (9.212) and (9. Example 9.") are from Eq.214)(9.(9. .216). The integration constants are evaluated by requiring that the upstream and downstream solutions satisfy Eqs.1 la) The exact temperature profiles for rhree values of Pe.
21lb).. Using Eq.21. because there are no other nonzero terms in Eq. Agreement between Eqs. and thus it is essential at all values of Pe. the denominator equals the volumetric flow rate.the axial conduction term is ~ 1 0 % of the convection term over >90% of the heated region. 1972).215). As shown by the dashed curves in Fig. This temperature profile is an excellent approximation provided that Pe(h[)>>I and Pej>zl. but [as with Eq. Under these conditions. but that the temperature in the upseeam region will be vimally qmturbed by the heating. For the two insulated regions. then QC. are sometimes called "mixing cup" quantities. we expect not only that axial conduction will be negligible in the heated section. if axial diffusion of species i is negligible.217) indicates that convection is the dominant mode of axial energy transport in the heated region when Pe is large (except for l s A). conduction upstream from the heated section should be rendered ineffective by the motion of the plate. provided that Pe is large and that the heated section is not too short. Ptclet numbers much larger than this are common in applications.j ) >> 1. the bulk temperature (Tb) or bulk concentration of species i (Cib)in the i fluid are simply those constant values far from the interface. neglecting axial conduction. 0 . n u s . For reasons which will be made clear. (9. Shah and London. subsequent examples of convective heat and mass transfer in this chapter assume that Pe is large enough for these approximations to be valid. we suggest that Pe >10 is a reasonable guideline for engineering work. the temperature profile from Eq. In the transverse (y) direction. the preferred choice is a velocityweighted average. (9. but the region where there is significant error becomes smaller as Pe is increased. if Pe A>50. heat and mass transfer coefficients are frequently used to describe transport between an interface and the bulk of a given fluid. Eq. Indeed. If all of the fluid passing a given axial position were collected and mixed. (9. reveals good agreement when Pe(A. Anticipating that conduction in the heated (middle) region would be negligible for large Pe. rather than at the true inlet. the convection and axial conduction terms must be equal for all Pe.region when Pe is large. where Pe is based on the mean fluid velocity and the full crosssectional dimension (diameter or plate spacing). Q. (9. Eq. (9. provided that. What constitutes a large enough Pkclet number to apply any or all of these approximations depends on the physical situation and the degree of precision required.218) and (9. their ratio is found to be 1 convection Pe d@ldlek(fi conduction d28/d(Z 1 (1) Axial conduction (or diffusian) may be neglected relative to axial convection (provided there are other nonzero terms in the differential equations). L 1 9. If the length of the heated region is 10 times the plate thickness (i. (9. 92.22 are valid for many heat and mass transfer problems involving confined flow. in addition to large Pe. in contrast to the threeregion problem solved in Example 9. and 0." it is necessary to employ some sort of cross. .could have been obtained somewhat more easily by first neglecting d 2 0 / d { 2 in Eq. Unless noted otherwise. where none of the fluid is "at infinity. In summary.217) only compares axial conduction with convection. The simplifications made in Example 9. Papoutsakis et a]. We conclude that the existence of a downstream insulated region has linle effect on the temperature profile elsewhere. Thus. is the molar flow rate of i. (9. In confined flows. Accord 1 ingly. As expected. (9.215) for Pe = 5. 1970. At high Pe.e.215) to evaluate the terms in Eq. Heat and Mass Transfer Coefficients As introduced in Chapter 2. conduction is the only mechanism for energy transport. Hsu.214) for the upstream insulated region requires only that PeA>> 1. would be the resulting concentration. Based on several analyses of heat or mass transfer in circular tubes or parallelplate channels with axial conduction or diffusion (Ash and Heinbockel. then C . jt Note that because the actual secondorder differential equation was replaced by a firstorder equation in deriving Eq. this criterion for neglecting axial conduction requires only that Pe>5. the length of the active section greatly exceeds the crosssectional dimension. A comparison of Eq. approximate expressions for 0. This result is obtained fmm the original problem by considering only the heated region. Thus. A = lo).3 [ PIUSSET AND SHERWOOD NUMBERS V L t. accuracy is lost near both ends of the heated =tion.219). Eq. confined flows.(9.219) with the corresponding exact solution for the heated region. In other words. When the fluid extends an in: definite distance from the interface and eventually reaches a constant temperature or concentration.220) is the solution to a problem involving one region.(9. Tband C. 1971. 1978.b.= 0 to good approximation. That is. (9. The solutions which result are (3) The conditions at the outlet may be ignored. That is. 1980a. for large Pe: where the flow is assumed to be in the z direction and A is the crosssectional area. sectional average value for the bulk temperature or bulk concentration. The driving force used in the expression for the energy or species flux is the difkrencs in temperature or ! concentration between the interface and the bulk fluid. Accordingly. (2) The "inlet" temperature or concentration may be specified at the beginning of the "active" section.219)] the regions where the errors occur vanish as Pe+. and the above three simplifications i have been employed in the majority of published studies of heat and mass transfer in i. Tan and Hsu. (9.219) loses accuracy near the downstream end of the heated section ([+A). only the upstream boundary condition could be satisfied. It follows that. for example. however. Eq. (9. (9. It is important to keep in mind that Eq..220) is almost indistinguishable from that of Eq. for sufficiently large Pe. and applying the upstream boundary condition at C= 0 rather than at f = w.2lla).2llb) arising from convection and axial conduction. (9.
k. The internal Row is assumed to be fully developed. (9.35) and (9. then Eq..33). The concentration of solute i in the internal solution is Ci(z 2 ) . we obtain The concentration at the inner membrane surface is eliminated now by equating the two expressions for the Rux. this strategy depends on having experimental or theoretical values for kc. is treated as a known constant.3I).34). For luminar Row in tubes. They are given by Figure 93..31) is the one which arises naturally in a partially lumped model for solute transport in a tube.I). heat and mass transfer coefficients can be computed from first principles. Design methods for hollowfiber dialyzers are detailed in Colton and Lowrie (1981). The dimensionless forms of the heat transfer coefficient (h) and mass transfer coefficient (k.36).(z). The flux from the internal fluid to the wall is given by flux on the righthand side of Eq. As discussed later.g.35) can be evaluated using either Eq. as shown in this chapter. and (3) the djalysate. In summary. After rearrangement. depending on the functional form for kdz). z). (9. the dialysate. is the inlet concentration. The flux of solute i through the membrane (based on the inner radius) is where kmi is the permeability of the membrane to i (analogous to a mass transfer coefficient). Obviously.. Choosing the former and combining Eqs.31 HollowFiber Dialyzer To illustrate why bulk quantities in confined flows are defined as in Eq.39) gives ac. (9. (9. we wish to take advantage of the available information on kci by using a macroscopic (partially lumped) model. Integrating Eq. is independent of position. where axial diffusion has been neglected. respectively. where L is the characteristic length (e.. 0 Using Eq. The key point to notice is that the integration of the convective term leads naturally to the velocityweighted average used to define the bulk concentration. ( . The derivation is easily generalized to include a mass transfer resistance in the external fluid. Only the first two resistances are considered here.. (9. Eqs. is the only unknown. t )= RnS. The Dimensional Analysis: Nusselt and Sherwood Numbers Cid Dialysate Information on heat and mass transfer coefficients is correlated most efficiently using dimensionless groups. and C . for heat exchangers) are based on the approach illustrated here.The quantity kcikmil(kci + k. $(R. Equation (9. solutes can be added to or removed from the internal solution as desired. (9. Because many design calculations (e. ac.39) can be integrated using elementary analytical or numerical methods. the lefthand side is rewritten as where U is the mean velocity. we examine steady mass transfer in a dialysis device consistindf many hollow fibers arranged in parallel. tube diameter). by distinguishmg between the surface and bulk concennations of i in the dialysate.34) over r gives from which one can readily calculate the overall rate of removal of solute i from the fiber.32) and (9.32) or Eq. this gives The final differential equation for the bulk concentration is where it is assumed that the mass transfer coefficient is a known function of position.33). calculations of overall mass transfer rates become elementary. (9. we have shown that the bulk concentration defined in Eq. as is k..g. is the concentration in the dialysate at the outer membrane surface. For simplicity. one must rely on empirical correlations. " Z 'd'=RDi ac. it is assumed that the dialysate flow rate is sufficiently large that C. . equal to Q[C. In such a system then are three mass transfer resistances in series. we have reduced Eq. associated with (1) the fluid inside the fiber. (2) the membrane. (9.34) to an ordinary differential equation in which C.a ac. much of h e literature on convective heat and mass transfer concerns the evaluation of transfer coefficients.). Rather than attempting to solve Eq. The differential form of the species transport equation for the fluid inside the fiber is vz dz where C ..(R. as discussed in Chapter 13. C . The fiber wall is a membrane which permits diffusional transport of solutes between the liquid solution flowing inside the fiber and an external solution. (9. for turbulent flow in tubes.3.) is the overall mass transfer coeficienr.Cib(L)]. thus. Thus. If the mass transfer coefficient is a known function of position.. If kCi is independent of z. Although Nu and Shi are used throughout this book. (9.. By adjusting the composition of the dialysate.. another dimensionless transfer coefficient in the engineering litera .WECTION HEAT AND MASS TRANSFER Example 9. 93.) are the Nusselt number (Nu) and Shenvood number (Sh. the assumption of a constant heat or mass transfer coefficient in a tube is accurate only for very long tubes. (9. Dialysis in a hollow fiber. 2). =< D. (9. A single fiber of internal radius R and length L is depicted in Fig.
.
for example. (9. Qualitative behavior of Nu or Shi in confined flows with stepchange boundary conditions at z=o.) can be understood in terms of the distance 6(z) that temperature (or concentration) disturbances at the wall penetrate into the fluid. this is not true if axial conduction is included.320) (9. In general.Shi = Sh. which is the usual convention for circular tubes. (T.5. (9. The tube wall is assumed to be at temperature To for z<0 and T . z and Pe appear only as a ratio. even though the temperature may continue to depend on z. .l(or Shi 1) for large z. the length of the thermal or concentration entrance region differs from that of the velocity entrance region. Thus. . the region modeled. The noteworthy aspect of the independent variables is that Pe has been embedded in the axial coordinate.. Neglecting axial conduction (i. provided that the tube cross section or film thickness is constant. and the flow is assumed to be laminar and fully developed for z>0..e. Brown (1 960)l.L and Nu 1.' Pe=. These remarks apply to steady transport in tubes of any crosssectional shape and in liquid films. when the temperature changes have reached the center of the tube. z) is   I IogNu r \ A Entrance """On  I  . the dimensional problem involving T(r. whereas that of Section 9.. an orderofmagnitude estimate of the temperature gradient at the wall is (aTldn). we give an overview of the typical behavior of Nu and Shi for confined laminar flows. Sc. geometric ratios) Additional parameters are involved if there are chemical reactions (Damkiihler numbers) or permeable walls (dimensionless permeabilities).41) are made dimensionless. 8. (Re. given enough length (see Section 6. r z Pe' 5E R  BE. The trends in Nu (or Sh.For example. (9. Entrance regions and fully developed regions are considered in turn in Sections 9.4 ENTRANCE REGION Basic Problem Formulation Here and in Section 9. For long enough tubes or films. Nu. The parameters for these and other.5 are four examples which illust~ate various aspects of convective transport in confined flows. Nu approaches a constant. so that a loglog plot of Nu(z) is linear at small z. the wall temperature changes from the inlet value to some new value at z = 0. where only a thin layer of fluid near the wall has been affected. as indicated.To T . The focus of this section is the thermal entrance region. The Peclet number is based on the mean velocity and the diarneter. whose pioneering analysis was published in 1883 [see. The position at which Nu becomes essentially constant separates the thermal entrance region from the themtally fully developed region. Nu initially varies as some inverse power of z. l )are . are constants.T .)IS. after L.  Qualitative Behavior of the Nusselt and Sherwood Numbers Before analyzing any examples in detail. The equations which govern @(77. assuming large Pe). for zz0.5). . conduction normal to the wall causes progressive increases in 6. all involving heat transfer in circular tubes..3222) Shi= Shi (Pe.. With a suitable crosssectional dimension as the characteristic length. As shown qualitatively in Fig. Moving downstream. and it remains at the new value for all z>0. We begin by stating the problem formulation which serves as a common point of departure. geometric ratios). Considering heat transfer in a tube. This is called the Graetz problem. It is assumed that Pe is large and that the temperature (or concentration) boundary conditions undergo a step change at z = 0. in these situations Nu is very large near z = 0 and declines with increasing z. Re.. veve~oped Region where R is the tube radius and U is the mean velocity. Qpically. Eventually.T. It follows from Eq. Similar statements apply to mass transfer and Sh.T. more complex situations in heat or mass transfer can be identified by writing all of the differential equations and boundary conditions in dimensionless form and proceeding as above. The fluid enters at T = To. . Shi = Shi(Fw. Sc.4 and 9.16) that 9. Graetz.3. Nu is very large near z = 0. geometric ratios). This choice of 6 is motivated by the fact that when all terms in Eq. A useful set of dimensionless variables is q= z. where To and T. 94. In such flows the velocity profile becomes fully developed.321) (9. 2 UR a log2 mgure 94. Fully . [F Z V(Re)J.5 is the thermally fully developed region.
(9.we find that The equations governing @(s) are now I n Eq. which required little mathematical effort. one representing axial convection and the other representing radial conduction. the constant is chosen as 312. @ varies from zero (the inlet temperature) to one (the wall temperature) as . Equation (9. We are concerned with axial positions such that a<< 1. as will be shown. is negligible.m~ I .41 Nusselt Number in the Entrance Region of a %be with Specified Temperature The objective is to obtain an asymptotic approximation for Nu whlch is valid at small [.417) to the similarity variable. The boundary conditions for @(x. [) will apply to all conditions. are less convenient for calculations than any of the functions encountered in Chapter 4. the term in parentheses must be constant.323) that rhe Nusselt number varies as 51'3. At these axial positions the fluid remains at the inlet temperature. for which we need to determine the temperature profile. What remains is to evaluate the constant C. (9. Example 9. so that Eq.49). just as the analogous series representations for transient conduction problems converge very slowly for small t. The temperature can be expressed as an eigenfunction expansion.411). The main reason another approach is needed is that the resulting Fourier series converges very slowly for small l. it is advantageous to use different solution strategies for the entrance region and fully developed region. we will take advantage of the thinness of the nonisothermal region and develop a similarity solution. g ( [ ) . From Eqs. recognizing that 3g2g' = (g3)r and requiring that g(0) = 0. . for later convenience. where 6 is the penetration depth divided by the tube radius. Comparable variations in O occur over the axial distance 6. This linearization of the velocity profile near the wall is referred to as the Uv8que approximation after M. for the heat transfer problem just stated.47)(9.411) and to infer how the thermal penetration depth grows with distance along the tube.414) indicates that the a@/ax term. except near the wall. A comparison of Eqs. Retaining only the dominant terms in Eq.416). must be proportional to the thermal penetration depth. who published the similarity solution for the entrance region in 1928. Converting Eq. Accodngly.5. the energy equation is There is only a x term in the coefficient of aO/d[. A secondary reason is that the basis functions which arise in this problem. in principle. (9. or that Notice that all parameters have been eliminated from this set of equations. we obtain For s to be the only independent variable.413) we infer that It follows from Eq. An orderofmagnitude analysis w i l l be used to identify the dominant terms in Eq. (9.46) becomes Equation (9. represents most of what there is to know about heat transfer in the early part of the thermal entrance region. It follows that for the nonisothemd region near the wall. which is associated with the curvature of the tube wall. (9. so that it is useful to define a new radial coordinate which is based at the wall. (9. (9. a single solution for @(q. UvQue. We assume now that 63 = @(s) only. where The boundary conditions follow readily from Eqs.! ~ AND T MASS TRANSFER Entrance Rwion 3ms  The remaining two terms in Eq. is based on the fact that the tube center is many characteristic lengths from the wall. which replaces the centerline condition. (9. as described in Section 9. amount 8.413) and (9.420). (9. 5) are Using this new coordinate. reflecting the fact that the curvature in the parabolic velocity profile is not evident near the tube wall. instead of using the finite Fourier transform method.412) and (9. (9. so that. In actuality.y2<<x. The positiondependent Length scale. Eq. called Graetzfunctions. are evidently the dominant ones. but that type of solution is not very helpful for the early part of the entrance region. y varies by the Thus.415) implies that gal'".4ll). For small 3 the temperature changes will be confined to a region near the tube wall. Thus.412) we have used the fact that . (9. where C is a constant which is of order of magnitude unity.
firstorder differential equation for the one remaining unknown. 0= 0. (9.357. Integrating this separable equation from the initial condition to <=tTand F = 1 gives. This value of & or K is in reasonable agreement with the exact results from the eigenfunction solution of the Graetz problem.) for a tube is the axial position at which Nu approaches its constant.43 1) becomes Because the nonisothermal region is very thin and almost all of the fluid remains at the inlet temperature of zero.416). 1967). fardownstream value to within some specified tolerance. 5 ) is of the form (T.42 Thermal Entrance Length The thermal entrance length (L. The fourth and fifth conditions are obtatned by noticing that the lefthand side of Eq. if R is . p.434) in Eqs.E~O~H AND EMASS A TT ~ N S F E R which have the solution Entrance Region 385 where r(x) is the gamma function and r(=)= 2. These last two conditions for O are written as Example 9. where Nu is within 6% of the fully developed value for (= 0. the temperature remains at the inlet value. the entrance length must be of the form Using Eq. By definition. (9. this is a consequence of the noslip condition and the constancy of the temperature along the wall. After changing variables from q to s. and it shows that C = 1. (9.06 and within 1% for f =0. are This result confirms the form for Nu given by Eq.43). The thermal entrance length for a constant flux at the wall is similar to that for a constant temperature. it is assumed now that O(s. In this example we use an integral approximation to estimate K. the results for parallelplate channels are similar to those for tubes. and also its r) derivative. it is found that where K(= lT)is a constant. The first three.46) over the radial coordinate to give Substituting the resulting expression for O in Eq. but it has been extended by constructing solutions as powerseries expansions in g (Newman. The LRvQue approximation is highly accurate only for the early part of the thermal entrance region.435H9. Likewise. 1970. 1969. we have set @. which corresponds to the "core" region in the center of the tube.428) Five conditions must be imposed to determine the coefficient functions. The assumed form of the temperature profile makes it convenient to employ a new radial coordinate.41. must vanish at the tube wall. It is assumed that for s l I . the Nusselt number in the early entrance region is Thus. F(0) = 0 and F(J. Eq. The dimensionless form of the Graetz problem. the corresponding dimensionless distance is F ( l )= S(()IR. (9. which shows that Nu is a function of [ only. corresponds to a specific value of 5.= 0. constant). The required integral equation is obtained by first integrating Eq. finally. The final result is NU = 1. (9. For a circular tube with a specified heat flux at the wall. but the value of C is somewhat larger (see Problem 93). (9.10 (Shah and London. F ( l ) .433) and (with the help of symbolic manipulation software) performing the differentiation and integration involving s. The Nusselt number (based on the tube diameter) is calculated as which is analogous to the similarity variable used in Example 9. in this analysis the entrance region is assumed to end at the axial position where the temperature changes just reach the tube centerline. That is.357 (f)"3~e1'3 Similar to the KhhPohlhausen method for momentum boundary layers (Example 8. 1978). which was an assumption in the Graetz formulation. we obtain which is a nonlinear. implies that z= L. the functional form of Nu is the same as for a specified temperature.) = 1. Wors~eSchmidt.46). (9.67894 (Abramowitz and Stegun. which we denote as tTb Thus. This expression is valid only for Pe>> 1. The radial temperature profile is approximated now by assuming that all temperature changes occur within a dimensional distance 6([) from the tube wall. 253). which follow from the wall and core temperatures and the continuity of the radial heat flux.439) and solving for the coefficients. (9.
431) equals (114) dO. so that O = 1 at the inlet and 8= 0 at the wall.Tw)/(To . so that we obtain Integrating from l= = (where 69.53) for the velocity entrance length. Thus. it is found that c . All that is needed to calculate NU() is the first eigenvalue. This is illustrated in Example 9. Equation (9. it is sufficient to require that @(O) be finite.49) is that the dimensionless temperature is redefined as 8= (T. This result is valid for any axial position. Before returning to the eigenvalue problem. The expression for the fully developed Nusselt number is simply Using the inner product defined by Eq. The additive tern "1. The constantflux case for a circular tube is examined in Example 9.) is needed. (9. = a. and the finite Fourier transform method can be used to derive an eigenfunction expansion for the temperature. As with Bessel's equation.510) we see that.T. Now. the thermal entrance lengths for these laminar flows can be summarized as and the inlet condition becomes which is analogous to Eq.51 for a tube with a specified wall temperature. With a wall temperature of zero. Example 9.52). Graetz functions are not widely available in computationally convenient form. so that only the smallest eigenvalue (A. (9. even though the temperature continues to depend on axial position. 6. (9.54). In some problems. The eigenfunctions defined by Eq. The basis functions are written as @. (9. (9. Using these results to evaluate the temperature gradient and bulk temperature in Eq. where G(x) denotes the Graetz function that is finite at x = 0 and a.5 FULLY DEVELOPED REGION The calculation of Nu for z+usually requires the evaluation of at least one t e r m of the eigenfunction expansion for the temperature field. the energy equation is transformed to Thus. from Qs. Eq. (This yields a slight simplification in the form of the solution. we examine what is needed to compute the Nusselt number for 500. a transformation used by a number of authors (Ash and Heinbockel.51 Nusselt Number in the Fully Developed Region of a n b e with Specified Temperature It is desired to determine the fuIIy developed value of Nu for the Graetz problem. for {+. it is found that the eigenvalue problem is where t is a dummy variable. the exponentials.52. Except for the factor ( 1 .ldf. However... = 0)to a finite value of l gives and using the approach of Chapter 4. only one of the independent solutions is finite at q=0. The only change from Eqs. (9. Solving for @ . Papoutsahs et a]. Nu becomes constant at large 5.) The differential equation and boundary conditions are linear. . .52) a r e called cylindrical Graetz functions. (9.$)rl.46) as where c.57) and (9. however." which is a correction for small Pe. the mathematical problem can be made much simpler. 1970.58). Nu is given by The calculation of Nu is simplified by using an overall energy balance to relate the bulk temperature to the temperature gradient at the wall. (9.E m O N HEAT AND MASS 'il?ANSkR m y DeveloMd Region replaced by the channel halfheight (H).52) is a SturrnLiouville problem with w(q) = (1 .).53) is written now as 9. including those with a specified heat flux. As the second boundary condition.(q) =an G(Anq). (6. p(q) = 9 and q(q) = 0.the differential equation is identical to Bessel's equation. It is readrly shown that the lefthand side of Eq.$).46)(9. and the Graetz functions all cancel. is based on the typical extent of edge effects in conduction (Example 4. is a normalization constant. Rewriting Eq. The temperature is expanded as The most slowly decaying term eventually dominates.
49). (9.51. We conclude that cR+O as f +. becomes a n (9.1980a. b. we @(?I. Nusselt independently calculated the first three in 1910. 1970.57). 1970.526) (9. Adopting the superposition approach. Because both of the boundary conditions in 7 have been chosen so that they are homogeneous. (9. Integrating the differential equation for $ and applying the boundary condition at q = 1 indicates that C. the problems for cCl(q) and 445) become (9. Finally.45)(9. q(l$)&(~&).. 5) = 0. b.46) and rearranging gives (T. The extensive review of Shah and London (1978) is a good source of tabulated results for this and related problems. However. the solution for Q(q..= 4. q4 d+q4j+C. x)= ( a ) x" 2. q.b) facilitates caIculations for the cylindrical case that is of interest here. When digital computers became available it was possible at long last to calculate large numbers of eigenvalues. " = o (b). As with the constanttemperature boundary condition considered in Example 9. Comparing Eqs. alCt rl Kummer's function is evaluated from its powerseries representation (Abramowitz and Stegun.. so must the sum of @(q) and &f). (9. L. the Nusselt number is calculated as Nu = 3. we find that Suppose now that we define R such that a n 1 a (l$)=(qG).7044. (9. It will be shown that the other functions in Eq. in which the temperature is written as the sum of the solutions of three problems. Finding the smallest root of Eq.52 Nusselt Number in the Fully Developed Region of a Thbe with Spetified Flux It is desired now to calculate the fardownstream value of Nu for a circular tube with a specified heat flux at the wall.529) 0) = . Because a ( q . Sellars et al. b. l ) satisfies the original partial differential equation.O = f l ( q . 3>+ $(TI>+ # a . 1 d dlC. f ) will be of the same form as that given for @(?I.n! (o. (9. (9. (9. This is of the same form as Kummeri equation. although the eigenvalues will be different. only M (also called Kummer's function) is needed. Wq. x ) (Abramowitz and Stegun.525) do not vanish.657 Thus.523) *= dl As discussed in Brown (1960) and Shah and London (1978). x ) and U(a. p. both of those early workers obtained results sufficiently accurate to give Nu = 3. constant). 504) as M(a.516) and (9. (1956) were pioneers in this regard.528) (9. Because U(a. In particular.To)I(qw RIk). the variable changes s = A q2.527) ensures that the inlet condition is satisfied. the dimensionless temperature is defined here as O = (T. Explicit solutions are needed for only two of the problems. both of which involve only polynomials of a single variable. and Eq. G(A7) = ~ ( s ) e . so that no further consideration of 0 is needed in determining the temperature far from the inlet. R satisfies the original partial differential equation. (9. p. G(X) =0. is another constant." ' ~ 5) transform the differential equation in the eigenvalue problem for G to To make the temperature gradient at the wall unity.532) Determining cG(q) and &[) is straightforward. Substituting that sum into Eq.527) (9.517). The dimensionless problem formulation is as given by Eqs. Graetz determined the first two eigenvalues in 1883 and W. 504). thereby allowing the eigenfunction expansion to be extended to relatively early parts of the entrance region. 0) is unbounded.66.cl/(q). i ) + $ t 4 ) + 4 t i ) = where C. (9.&O). Denoting that constant as C. 5) in Eq. (9.519) using 20 terms of the series for M.525) the solutions to whtch are the confluent hypergeometric functions M(a.533) Example 9. b.530) both equal a constant. a complete solution to the constantflux problem can be obtained by using the finite Fourier transform method and Graetz (or Kummer) functions. the temperature field for l00 is obtained more easily by a linear superposition approach. except that the boundary condition at the wall is replaced by R . we obtain hl = 2. It is found that for l+. a6 rlall and the characteristic equation for the Graetz problem. (9.
Fully ~ e v z ~ Region e d It is not necessary to evaluate C. . 96. The continuous curves were calculated fiom Figure 95. respectively. and serve to bracket the other results.536) 600. In the first two cases.429). R in Nu and B is replaced by H. where Nu.533). The effect of B on the fully developed value of Nu is shown in Fig.) or constant temperature (T. (9. (2) mass transfer. for (9. The entranceregion results given by Eqs. (1964) by ~ 0 . 0 3whereas . if the wall is a membrane with a finite permeability (Example 9.538). The constantflux and constanttemperature cases correspond to the Limits B+O and B+m. (1971) and Colton and Lowrie (1981) is equivalent to 2B fiom the second expression.538) are satisfactory for most purposes when 4'>0. I l l I I L I I I I I I I I I . I . ~ v & t i n ~ the bulk temperature using Eq. if the resistance of the wall is not negligible. . The other results were obtained by calculating the first 121 terms in the eigenfunction expansions for the temperature (Shah and London. Nusselt number for fully developed laminar flow in a circular tube with a constant heat flux (q. and (9. is the heat transfer coefficient for the wall. for both geometries. we can use the fact that axial variations in the bulk temperature result only from the known heat flux at the tube wall.428) and (9. However. The wall conductance parameter for the reaction case equals the Damkohler number.536). Figure 95 shows plots of Nu(5) for both the constanttemperature and constantflux cases. are the respective constanttemperature and constantflux values. (9.51. the mixed or Robin boundary condition.31). for a channel with plate spacing 2H. Using the constantflux boundary condition in that relation gives [+m. The "wall thermal resistance parameter" used by Shah and London (1978) equals (28)' from the first expression. Comparing Eqs. Mixed Boundary Conditions and Other Conduit Shapes Thus far. whereas the "wall Sherwood number" employed by Colton et al.429) are quite accurate for 5 ~ 0 . The asymptotic results are based on Eqs. (9. but this will be done to complete the solution for the temperature field far from the inlet. even though the temperature continues to vary with 5. Eq. (9. kmiis the permeability of the wall (membrane) to solute i.428). with Nu 4 for all conditions.534) and (9. which is an asymptotic expression for large 5. B is similar to the inverse of a Biot number. irreversible reaction with a moderate rate constant. Shah and London (1978) Asymptotic results . = 7/24. (9. 95 for the Nusselt number. and Nu. is the rate constant for the heterogeneous reaction. I 1 1 1 1 1 1 . That is. 103 and 127).533). for ance.08. (9.) at the wall. where h.45. (9. . the results for the Sherwood number are entirely analogous to those shown in Fig. the discussion of Nu has focused entirely on circular tubes with constanttemperature or constantflux boundary conditions. and k. . to calculate Nu. Accordingly. Three situations where that type of condition arises are: (1) heat transfer. . we find that C. .523). . if the wall catalyzes a firstorder. where B is the wall conductance parameter The definitions of B for the three situations just mentioned are As in Example 9. (9. 1978). constant).364 (q. l ) 11 =%=4. Results are shown both for circular tubes and parallelplate channels. pp. The results for the two geometries are very similar. We cannot apply the inlet condition to Eq. For mass transfer between a dilute solution and a tube wall. . At any given value of 5. included are intermehate values taken f r o m the review by Shah and London (1978. The more general boundary condition is written in dimensionless form as which has the solution Ob= . Nu. it is seen that Nu becomes constant far from the inlet. we find that 391  In addition to the asymptotic results for large and small l.523) and (9. which is valid for all Another expression for the bulk temperature is obtained from the overall energy bal5 and any type of boundary condition at the wall. A more general form of boundary condition in heat or mass transfer involves a linear combination of the field variable and its gradient at the wall. the fully developed values in Eqs.59). and (3) mass transfer. This simple interpola . the constantflux results are about 20% larger than those for constant temperature. respectively. the difference being that the heat and mass transfer coefficients for the fluid are replaced by W R and DilR. 4 I I . (9. 3 % over the entire range of B (see Problem 912). m . . which differs from the exact results computed from the eigenvalues of Sideman et al.< Nu S Nu. The Nusselt number for large is calculated as Nu = 2 @b(l)@(l.
.
u ) dS. In previous chapters we used Hv to represent such external sources. It cannot appear in both places. (9.(n.a ) dS.61) allows for energy transfer d e to bulk flow through the control surface. This is just a statement of the first law of thermodynamics for a control volume. (9.66). The result is Assuming that gravity is the only body force.V $ and recognizing that in terrestrial applications 4 is independent of time. the gravitational work done by the system on its surroundings is Using Eq. Certain other types of work. as in the derivation of Eq. The lefthand side of the resulting differential equation is simplified by using the continuity equation and introducing the material derivative. We omit the Hv term here.g) dF! Accordingly. the first term on the righthand side of Eq.p(v. including that due to an electrical power source.610) confirms that gravity may be included either as a potential energy term on the lefthand side or as an energy s o m e term on the righthand side. . The rate of work done by the fluid on its surroundings has two parts. one due to surface forces and the other due to body forces. which is the approach taken in Bird et al. a)dS. pp. the surface force exerted by the Auid on the surroundings is . (1960.67). we convert the surface integrals to volume integrals and let V+O. Eq. (9. Whereas in elementary texts on thermodynamics the first law is generally written for a system of fixed mass. it follows from Eq. The result of these manipulations is where q is evaluated using Fourier's law.610) that A comparison of this with Eq. (9.37). and then to express H as a function of T and f? That is the approach used here. as in Section 2. The first steps in obtaining such a relation are to evaluate p D ( v 2 / 2 ) l ~and t to subtract that expression from Eq. The total rate of surface work on the surroundings ( w ~ )is then which is a general expression for conservation of energy at any point in a pure fluid.the force exerted by the surroundings on a surface element dS is (n. 314315). and w is the rate of work done by the fluid on its surroundings.69) is rearranged to WG= . (n. It will be more useful to have an energy equation that involves the temperature. Evaluating the heat transfer and work terms as just described.610) from Eq. there are two basic options. The contribution to the work from forces exerted on the surface S is evaluated in terms of the stress tensor (a). To obtain a differential form of Eq. H. pp.2. 314315). The rate of work done on the fluid just outside this differential surface element is v.24) to evaluate the lefthand side. decomposed as usual into pressure and viscous contributions: According to the definition of a. with the understanding that (if needed) it may simply be added to the other volumetric source terms in the final. (9. can also be included. Thus. so that Q represents conduction only. (9. an alternate form of the mechanical (or total) energy equation includes potential energy on the lefthand side.(n. (9.67) yields the desired equation involving only the internal energy. which are encountered more frequently in heat transfer problems in solids than in fluids. Defining the gravitational potential energy 4 by g = . The surface integral in Eq.' Subtracting Eq.Conservation of Energy: Mechanical Effects 395 where Q is fhe rate of heat transfer from the surroundings to the fluid inside the control volume. Once this is done. (8.61) becomes 'As explained in Bird et al. (2. (9.g) dK v I (9.67). which is The total rate of work done on the surroundings is wS+WG. (9. By forming the dot product of v with Eq. Eq. the rate of work done by the surroundings on a volume element dV is p (v .61) accounts for convective energy transport. (9. A relationship involving mechanical forms of energy is obtained from the Cauchy momentum equation. we obtain a scalar equation with terms having the same dimensions as those in Eq. The other is t o convert ? I to the enthalpy per unit mass. thereby obtaining an equation for the internal energy alone. where we have used the identity v . T )= (T(9. (9. One is to express ?i as a function of T and p.65) This is called the mechanical energy equation. differential form of the energy equation.68). (1960. The rate of heat transfer into the control volume by conduction is 9 v) n.
.
Taylor (18861975) made major contributions to several areas of applied mechanics and transport theory.71) has the same form as the conservation equation for onedimensional diffusion in a stagnant fluid.~ has applications as diverse as the analysis of peak broadening in chromatographic separations. to obtain results which will be useful for the Taylor dispersion analysis. involves fewer restrictive assumptions than the original analysis given by Taylor (1953). Viscous dissipation tends to be important mainly for polymeric fluids (large p) or highspeed flows (large a). Because we will be applying these results to dispersion in a long tube of radius R. as well as his experimental and theoretical investigation of instabilities in Couette flows. t).3. we obtain There are no boundary terms remaining from the integrations by parts. Z. namely KID.74). Re. it will be shown that the crosssectional average solute concentration C(z.. a sharp peak in solute concentration is created. P r . we in@r that Nu =Nu (F. we use Eq.5 all correspond to Br+O. the prediction of the degree of mixing of petroleum products pumped in sequence in a pipeline. The objective of the following analysis is to evaluate K for the laminar flow of an incompressible. To show how moments may be used to evaluate K. Given the ' ~ u r i nhis ~ career at Cambridge University. The results presented for Nu in Sections 9. called Taylor disper~ion. Because diffusion limits the rate at which the solute can move to distant locations.73) by l p and then integrating as in Eq. D.we first examine the relationship between the diffusivity of a solute and certain moments of the concentration distribution. If the initial concentration is given by c=A 'I where f([)+O as l+ f .76) and (9. It is seen that the various moments are functions only of time. expresses the relative importance of viscous dissipation and heat conduction. The dispersivity. and the measurement of molecular diffusivities.77). Using Eqs. described by Aris (1956). The conclusions are applicable to any onedimensional. As the peak travels along the tube it gradually broadens. (Some authors employ the Ecken number. in which case viscous dissipation is negligible. The pth moment of C about (= 0 is defined as where p is a nonnegative integer. will remain finite for finite values of T. (9. (9. Br. Newtonian fluid in a circular tube. The axial dispersion of solute which causes the peak broadening resembles molecular diffusion. (9.tube. nonuniform axial velocity in a. resulting differential equation for T ) is variables l= c(5. t) is governed by an equation of the form The dimensionless diffusivity in this equation is the coefficient on the righthand side. K. Among his bestknown contributions to fluid mechanics are his concepts of isotropic hubulence and turbulent energy spectra.7 TAYLOR DISPERSION If a soluble substance is injected rapidly into a fluid flowing in a tube. Ec = BrIPr.626) for steady flow. The Brinkman number. (9. For sufficiently long observation times. A differential equation for mP(r) is obtained by multiplying both sides of Eq. the precise form of f(5) is unimportant for our purposes. transient diffusion process in an unbounded domain. Let z represent axial position relative to such an origin. 9. Taylor dispersion concerns the evolution of a solute concentration peak as would be seen by an observer moving with the fluid.4 and 9. we introduce the dimensionless zlR and r = D ~ J R The ~ . it is remarkable that Eq. it does not in general equal the molecular diffusivity.71) as the starting point. because the assumed form of the initial concentration ensures that C and acldc vanish as (+ 5 rn. but is actually a combined effect of diffusion and convection. This approach. This makes it advantageous to employ a coordinate origin that moves at the mean velocity of the fluid.) Reasoning as in Section 9. The lefthand side gives CdA. we find that the moments must satisfy . Integrating by parts twice to evaluate the righthand side. where A denotes the crosssectional area. while avoiding the more difficult task of solving the full species conservation equation for C(7. Br. geometric ratios) (9. He had a remarkable ability to examine a phenomenon using a relatively simple experiment and was able to build on the experimental observations a thorough and imaginative theoretical analysis. G . (9. it follows that all m. then the initial values of all mp will be finite. The approach will be to calculate the first few moments of the concentration distribution. is a constant whose value depends on the shape of the tube cross section and the form of the fully developed velocity profile.ECTION HEAT AND MASS TRANSI%? Taylor Dispersion 399 where AT is the temperature scale used in defining @. This phenomenon. Other than the restriction just stated. I.
for large comes negligible. Thus. the solution is found to be 00 ~ .713) and the initial condition is of the form CO(V. Because the axial diffusion term has been retained. and U. the ability to determine a constant value of K from Eq. although the complete details of these lower moments are unnecessary.716) and (9. (9. (9.78).. (9. proportional to the total amount of solute present. The more general analysis given by Aris (1956) shows that where f(q) is computed from the initial concentration distribution. r ) is seen to be The boundary conditions are analogous to Eqs.712) and (9. Eventually.74) are based on the crosssectional average concentration. We turn now to the problem of evaluating the necessary moments in terms of the known quantities R. (9.hi 7 n= 1 . the governing equation for C. (9. In physical terns. and m . as implied by Eq. (9.711). we need not proceed beyond m2..716) over 77. 7)is obtained by integrating Eq. It will not be necessary to evaluate these constants.the constant (4) term beA partial differential equation governing CJq. m. and we find that 7. C . we assume also that the initial disbibution is axisymmetric. (9. The axial velocity used in Eq. The moments m. leading to where q = rIR and Pe = 2URID.T)= mo+ I a. 0) =f(q).74). in the moving reference frame the mean velocity is zero. remains constant because there is no directional bias in a diffusion equation such as Eq. and 2.The initia! values. integration of Eq. This equation is subject to the radial boundary conditions Equations (9. may be computed if desired by replacing by C by f(j)in Eq.711) over 5.711) is valid for any value of Pe. which does not change.73).711) was obtained by subtracting U from the usual parabolic velocity profile for a tube.78) indicates that the zeroth and first moments remain constant for all T. as expected if Eq. (9. Likewise m. The result is It is seen that the diffusivity (in this case.710) ultimately validates Eq. Inspection of Eq. Setting p = 0 in Eq. The utility of this approach stems from the fact that to calculate KID. For simplicity. .78) gives the value obtained for K does not depend on the initial concentration being axisymmetric.719) AS in the preceding discussion. as was done in deriving Eq. so that C is always independent of the cylindrical angle 8. (9..(r) is found by integrating Eq. Eq. which is related to the solute center of mass. m.71) is correct. (9. grows linearly with time. Jo(Anq)e . mo is constant because it is proportional to the total mount of solute present. from which it follows that C+O as 5t k o . the exact form of the initial concentration distribution out to be unimportant. (9. To obtain m. (9. For p = 2. 1.(q. is simply a constant. for all 7 and that the moments m. These moments are defined by The two types of moments are related according to Thus. we will need to find C. (9. As already noted.(r) are finite. The analysis requires that we work also with analogo~s moments of the local (radially dependent) concentration. ( q . In the moving reference frame the conservation equation for the solute (assumed to be dilute) is The ordinary differential equation which governs m.The analysis which will be presented for tube flow confirms that m2 grows linearly in time for that situation. where U is the mean velocity of the fluid relative to the tube. . M. Using the methods of Chapter 4. (9. KID) may be computed from the longtime behavior of the ratio m2/m0. It is only necessary to assume that the solute is initially confined to a finite length of tube.716). (9.71). (9.717) constitute a sequence of differential equations which can be used to compute progressively higher moments. D. We will be concerned only with the moments corresponding to p=O. defined by Eq. (9..
) is a function of q only. As discussed in detail by Taylor (1953). essary. h. That the leading constant in Eq.720) must equal nz. or ~>0. varies as DI. (9. (9. (9.(q.) = 0. The analysis of Taylor dispersion in tube flow is the prototype for a general class of convectivediffusion problems.2. Aris (1956) also showed how to compute K for any other crosssectional geometry and any fully developed flow. We expect the ~ ( e . shown that m.71) is the sum of the motecular diffusivity and a term which. Thus. The differential equation for C. we arrive at the desired ratio of moments: Finally.(q.732) in Eq. using Eq. he also provided several comparisons between theory and experiment. In this case the problem was to find the "effective diffusivity" (K) which could be used with the crosssectional average . as is easily confirmed. the convective contribution to dispersion vanishes and K = D. For a uniform velocity (plug flow). As already mentioned. The homogeneous form of Eq. (9.(q. the remainder of the integral is ~ ( e . it turns out that rn. 7 ) for large r is seen to be The constant (mo) tern in Co does not contribute to the integral in Eq. The governing equation for m.723) consists of a sum of terms corresponding to those in Eq. in which it is desired to calculate effective transport properties through spatial or temporal averaging. which in this case was parabolic. (9. 7) is Any more detailed consideration of the timedependent terms in C.720).(~)is 7) will be unnec Considering the behavior of m.718) and has solutions r ) = c j ~ ~ ( q)eh:r. such that Neglecting the terms which decay in time and then integrating. The elegant result given by Eq. As already noted. is proportional to the center of mass for the solute.. From Eq. surprisingly. The particular solution of Eq.723) is of the form Thus." " )as T+=.723) is the same as Eq. dm.722). the dispersivity K appearing in Eq. only for large 7. Because it is a linear. (9. m. (9.ldr+O fairly quickly. we evaluate the dispersivity as The remaining terms all have the form where bn is a constant and Each of the functions g. Adding the particular and homogeneous solutions. and rnl +m.733) is valid for conditions even more general than those assumed in the above derivation. (9. a constant. the complete solution to Eq.720). Taylor (1953) showed that the convective dispersion result applies equally to a step change in solute concentration. (=3. (9. Aris (1956) showed that the same result is obtained if the initial concentration distribution is not axisymmetric. While giving results for the special case of laminar flow in a cylindrical tube. after a short period of adjustment (depending on the initial concentration distribution).723) is the solution to the homogeneous equation added to a particular solution of the nonhomogeneous equation. the solute center of mass moves at the mean velocity of the fluid. (9. the second term arises from the combined effects of axial convection and radial diffusion. If the initial pulse is symmetric about c= 0. Although our derivation focused on a pulse input of solute..Eqs.730) indicate that which is seen to be nonhomogeneous because of the C. The boundary conditions and initial condition for C. term. the general solution to Eq.Accordingly. (9. (9.715) and evaluating the integral. are similar to those for C. (9. (9.~ : ? term to be negligible for A:T> 3. (9.729) and (9.. nonhomogeneous differential equation. The differential equation for m. The numerical coefficient (1148) depends on the shape of the velocity profile.(A). (9. is seen by using Eq. the nth term of the particular solution is denoted as It is readily (the part of the particular solution which corresponds to the constant.727) where cj is a constant that is determined by the initial condition.(q) must satisfy the homogeneous boundary conditions.Taylor Disknlon 403 where J I(A. where A.717).83) is the smallest positive root of J. is The behavior of C. (9.720) in Eq. (9.7lo). obtained by setting p = 1 in Eq. = O for all T.(r).
.
$ C 5 2 : rcr ' 3 .i T a 5.E! ~+ 3 5 s4 6 *z 2 : g z !E V) 3 . b 2 g 2 O z L % : .Y r (I n ) a . . 3 8 25 Y E dm . g z s i ~ z h s .Es.. 3 0 31 c .El ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
(11.=Bp in Eq. not all of the remaining coefficients are independent.62) and the GibbsDuhem equation. in Eq.. although not yet in the most convenient form. . (11.62). (1 1. 2. The physical interpretation of di suggests that the sum of all di should vanish. and the thermal conductivity is derived by applying Eq. it is necessary that We now have a basis for identifying all of the forcefl ux couplings which arise in energy and mass transfer. the Dufour energy flux also vanishes (i.810) allows us to identify forces and fluxes which are consistent with Eq.816) to a fluid where diffusional fluxes are absent (i. Eq.. (11.Generalized StefarrMaxwell Equations 467 It is convenient for applications to rearrange the energy and mass flux terms in Eq. (11. it is seen then that Bm = (kT)I.62). Based on the remaining terms in Eq. As suggested by the form of Eq. A comparison of Eq.=I n .v). or + i= 1 CB.v = 0).88) is rewritten now as where di is the diffusional driving force defined by Eq. As discussed in connection with Eq.23). are independent of the fluxes and forces. From Eq.= (kT)I and Bq. where the coefficients B. which involves the difference between the body force on species i and the total body force on the mixture. B. Summing Eq. Equation (11. Recalling that &/pi= v.8lo). the matrix of independent coefficients is indeed n xn.82) with Eq.815). An alternative approach is to write the forces as a linear combination of flux terms. Using Fourier's law in Eq. n I..v).e. (11. This interpretation of d. the Bu matrix is the inverse of the Lo matrix.4I). (11. which seems inconsistent with Eq. An additional set of relationships among the coefficients follows from Eg. which is like the StefanMaxwell equations (Eq.817) over all components. .815). (11. (11. Using B. including B . (1 1. (11.44) for a binary gas and as discussed below. three coefficients are needed to completely describe heat and mass transfer in a binary mixture (Table 11l). (1 1.817) from (n 1)2 to (n + 1)(n 2)/2.41) and are independent of x") and f ( ~ ) . (11.815). Instead of employing the full. These equations imply that the B. or This immediately reduces the apparent number of independent coefficients in Eqs. (11.816).=O.lpi. (11. Before arriving at the final form of the generalized StefanMaxwell equations. The transport properties of a multicomponent fluid are embodied in the phenomenological coeficients. . L11. Using Eq. the B0 m a t r i x must be symmetric.e. however. (11. then the corresponding This additional set of constraints reduces the number of independent coefficients to n(n+ 1)/2. The subscript 0 has been chosen to refer to tbe thermal terms. Equation (11.88).5. (11. . (q(c' tq(x)) +C .I)]. Because all of the phenomenological coefficients. (11. we will subtract from q the term representing diffusional transport of enthalpy.810) cannot be directly coupled with energy or mass fluxes. . (11. rz. Thus. matrix is (n + I) x (n+ 1). j = O .817) is a generalization of the StefanMaxwell equations. q")= 0). one way to express those relationships is to write the fluxes as a linear combination of force terms./ are related to the coefficients L..42) and the relationship between Bo and LU.815). In this case. As already mentioned. is the preferred form. (1 1. For Eq. vj.816) and (11. or fluxes must be (q(c) 15) becomes + q(x)) and j.810) from Eq. In obtaining Eq. The relationship between B .V in T= B. the fluxes may be viewed as a set of mathematically independent variables which determine the forces. (11. However.. is a force per unit volume tending to move species i relative to the mixhue. we find that + + one confirms that. (11. it is useful to establish certain properties of these coefficients and to relate them to more familiar quantities. (11. multicomponent energy flux q defined by Eq. (11. (11.814). as shown by Eq.88). this interpretation of B. According to the CuriePrigogine principle. respectively. This leaves the sum of the Fourier and Dufour fluxes. and six are needed for a ternary mixture. is perhaps most evident in the body force term in Eq. indeed.Equation (11.Specifically.816)' we find that the multicomponent Dufour flux is given by .v. we used the thermodynamic identities The utility of the quantity di comes from the fact that CRTd. remains valid when diffusion is present. (11.819) to hold for an arbitrary set of values for the fluxes (q@) 9'')) and (v. the viscous dissipation and chemical reaction terms in Eq. if the forces are chosen as V 1nT and CRTdi.8 . so that the coefficient relating the temperature gradient to the energy flux is written as B.. as shown below.(v. where the Fourier or conduction flux is denoted by q@).
it is evident that DiicO!This makes it clear that the Dii are not ordinary diffusivities.828). by replacing $Bij with Do.v. Although diffusivities depend to some extent on the solution composition.vi). are all positive numbers. d . for example. kb. the diagonal elements 9ii(or Dii)of the multicomponent diffusivity matrix are not needed. (11. that Thus. (1 1. With this choice.l of these equations are independent. " z'1 (v. This is demonstrated by noting. except that it has v in place of via In fact. ( 1 1. Roperties of B i .. n. The coefficients Bio (i # 0) in Eq.824). (11. (11. Moreover.5I).) in the multicomponent diffusion equations are not the selfdiffusivities which might be measured by studying the diffusion of a tracer form of species i relative to unlabeled i.)tCR v In T. .817) in a form more closely resembling the StefanMaxwell equations.830). It follows from Eq. = v. the tion than are the corresponding By. we obtain VX. Thus. evaluating q(') using Fourier's law.(or Dii) from the offdiagonal coefficients. as just done in obtaining Eq.X. (1 1.824).829). = DU. for gas mixtures..The following examples demonstrate the relationship between the generalized StefanMaxwell equations and the equations usually used to describe multicomponent diffusion in gases and in dilute mixtures. in Eq. The advantage of the coefficients 9. (11.822) are that xixjCRT 9 p. (1 1. .830) is identical to Eq. so that the solutes are identified by the indices i = 2.828) j= 1 9 . (1 1. we find that ! 1.. (1 1.824) to gases. that is.. CRT Finally. ( I I .818). Of greater interest in mass transfer are the coefficients B. Assuming also constant pressure and no body forces other than gravity.. (1 1.Thus..828) may be chosen at will. respectively. Example 11.5 for a relation analogous to Eq.Generalized ~iefanMaxwellEquations 469 q(")= * k~z BjO(vj n . (11. This confirms that the Dufour flux vanishes when diffusion is absent. (1 1. It must be emphasized that the aii (or D. (11. (1 1. The interpretation of $!hi/ for liquids and solids is not as straightforward.823) and (1 1. which follow from Eqs. (11.= 1. we define multicomponent d@usivities. (11. (11. j # 0). Eq. (11.822). applying Eq.. Although v. In summary. however. it is commonly the case that there is no measurable binary diffusivity which corresponds even approximately to 9 . we obtain As The summation term in Eq. d . = R. Using Eq. X.830) j j+i = l 9ij 1 CRT already mentioned.814)] and the two sets of relationships among the multicomponent diffusivities [(Eqs. (11. such that 3 (vjvk) % (q"' + 4'x'). V r p p i = R mIn xi. neglecting q(").821) " X. 'i i Setting V. any velocity other than vj may be substituted here for v. there was no need in Section 11. denoted by 1. To put Eq.62). we obtain . Example 11. (11.82 Multicomponent Diflusion in Dilute Mixtums A dilute mixture is one which consists mainly of a single abundant species.817) describe multicomponent thermal diffusion effects.820). Multicomponent diffusivities for the various solutesolvent pairs are similar to the corresponding binary diffusivities. and setting bi = B. n.. from Eq. (11.X.822) B. = v.X.. 3. The relations given by Eq. which provides a way to compute each of the 9.by i a n d j = 1 . j = lji' Because the mole fractions and the offdiagonal D.814) that only n .v). multicomponent mass transfer is described by the generalized StefanMaxwell equations. according to what is convenient for ( [ a particular problem.is that they are much less dependent on composi.. (11. Indeed.81 Multicomponeot Diffusion in Ideal Gases We consider first the diffusional driving force. i=... 9 multicomponent diffusivity in a gas is the same as the usual binary diffusivity of species i and j.. There is a subtle implication in choosing v. ( 1 1. (i # 0.824)J. the (original) StefanMaxwell equations.817) is rearranged to 1 1 . In a liquid solution. for gases we have 9 . let the solvent be denoted as component I .828) are called the generalized StefanMaxwell equations. and neglecting thermal diffusion.. Eq. defined by Eq. By. (11. Eq. it is usually equated with either v or vi. = C X ( v. .For solutesolute pairs. in Eq. This will be true if i and j are both solids or immiscible fi uids under the conditions of interest. it is true nonetheless that Bil 'D. and (1 1.825) now resembles that in the StefanMaxwell equations. together with the driving force constraint [Eq.825) is rewritten more generally as di= X. such that they cannot form a binary solution. For an ideal mixture. using Eq. 2 .
New York. J. Cussler. (1 1. E. surface. F. the summation is expanded as Hirschfelder. Lightfoot. so that v. The T h e m 1 D i m i o n Column.833) reduces to 111. K=(CBICA) the equilibrium constants at the top and bottom surfaces are related by lnA= "> ( I KO R To TL. S.component is the solvent. Toor. pp. R (a) Use interfacial mass balances to relate the fluxes of A and B to the reaction rate at the I r This relation is equivalent to Eq. 1961. and B. Derjaguin.61). from the generalized StefanMaxwell equations. . Electrochemical Systems.K. Wiley. Prater. Eng. Ji = .828) to give Assuming for the moment that all of the fluxes are collinear. Curtiss.. It follows from Eq. NJ. Wiley. New York.) are known. Combustion of a Carbon Particle i" 1 Within a certain range of temperatures the combustion of carbon occurs primarily through the reaction For this pseudobinary situation. R. Berlin. The temperature produced by heat of reaction in the interior of porous particles. From van't Hoff's equation. 1969. L. 1952. = (1 vj 11 1 v1( )vl . New York. 3: 198207. 260 and 269. Vol.832) that the summations on the righthand side of Eq. Newman. 1957. 1958. Assume that the gas contains only oxygen (species A) and carbon monoxide (species B) and that the mole fractions (xi_) and the temperature far from the particle (T. Wiley. Dukbin. Ibbs. Assume here that radiation heat transfer is negligible. and I? Mazur. and R. Eq. The objective here is to derive Eq. The fast. C. B. Transport Phenomena and Living System. S. PrenticeHall. 'and using v i = Ni/Ci in Eq. (f) Describe how the analysis would be changed if the gas contained an additional inert species (e. Sci. 8: 284286. (b) Determine the oxygen flux at the particle surface. ho. Also derive the relationship between the reaction rate and dR/dt. 7 . assuming that the particle and surroundings act as blackbodies. (1 1.g. 1976. . (c) Use an interfacial energy balance to derive the relationship betureen the heat flux at the particle. Ed. Matijevic. . the other solutes in a liquid).D. (e) Explain how to include radiation heat transfer in the analysis.L. Vasaru. when the mixture is a liquid solution. L. NonEquilibrium Thermodynamics. Bil =DL.. S. To. Pll2.D. pp. 1973. (d) Determine the particle temperature. Edwards. reversible reaction A t t B occurs at the surfaces. References Beris. State the new governing equations and outline the approach needed to calculate the reaction rate and particle temperature. V. provided that the ratios BilB. Effect of a Reversible Reaction on Energy Ransfer in a Gas A binary gas mixture consisting of species A and B is confined between parallel plates maintained at different temperatures. Cambridge University Press. Multicomponent Diffusion. Englewood Cliffs. Physical Chemistry of Macromolecules. and the particle is small enough that Pe<< 1 for both heat and mass transfer. 1954. Bird. the reaction rate. " '). Equilibrium double layer and electrokinetic phenomena. Diffusion in threecomponent gas mixtures.. J. (1 I. k t CC be the molar concentration of carbon in the solid.834) will be negligible. you may assume that the product CDAB is independent of position. D. or solids. liquids. C. O. the flux equation for a dilute mixture. For simplicity. A comparison of the thermal diffusion terns indicates that for a diIute mixture.. G.. 1974.for the conditions of part (c). Wiley. I ! kb. the abundant. Under these conditions. Thermodynamics of Flowing Systems.Civ. and I vj I/ I vI I are not large. 1962. and D. Cambridge.835) becomes 1 for a dilute . 522523. there is no homogeneous reaction.61). AIChE J. mixture and that Ji = Ni . Molecular Theory of Gases and Liquids.Recalling also that v. (11. Themal Difision in Gases.g. Fodor. H. and T. C. The reaction is very fast. NorthHolland. In Suqace and Colloid Science. N. New York. 1974.). New York. Miiller. E. 1994. A. 112. de ( h o t . Problems I where the diffusional interactions between species i and the abundant component have been separated from those between i and the other minor species (e. Reinhold.D. so that A and B are in equilibrium there. G.E. The equilibrium constant.I depends on temperature. pp.Cd. Chem. Consider a spherical carbon particle of radius R(t) suspended in a gas mixture. under pseudosteady conditions. (11.. h w . 157165. (11.Problems ' 47 1 Dilute mixtures may exist as gases.v and x. pp. V In T + at the carbon surface. Amsterdam. and the heat of reaction. Elsevier. Amsterdam. as shown in Fig. Oxford University Ress. 49272. = v. N. We begin by choosing v. and T. Eq. S. VEB Deutscher Verlag der Wissenschaften. I. E. Tanford. G. N. (11.
= 0 at r = Rf. R.I.. for r c R f . Rp and known quantities. The reaction is assumed to be extremely fast. as shown in Fig. thermal conductivity. and the heat of reaction (AHR=HB. is surrounded by a stagnant gas which contains oxygen (species B). as shown in Fig. Use the van? Hoff relationship between KL and KO. How (if at all) is the ratio of the energy fluxes affected by the algebraic sign of AHR? it ! 1 113. (d) By considering mass transfer in the gas inside the flame front (Ro<r<RJ). Assume that k . the bulk temperature (T= T.KL. [ ' where k . the droplet temperature. so that it may be modeled as occurring only at a position r=R. y=0 Gas 8.. Assume that the temperature (To) and liquidphase concentrations (CAo. assume that the following quantities are given: the mole fractions in the bulk gas (x. assumed constant. Pll4. throughout the liquid. Assume that the pressure is constant. The temperaure difference is small enough that the total molar concentration. E.). Burning of s Liquid Fuel Droplet The burning of a small droplet of liquid fuel is to be modeled as a pseudosteady. The function xAo(To). Species B. Burning of a liyuid fuel droplet. is the molar concentration of A in the pure liquid. Show that the only parameters involved are Gas Liquid A+B c.= k . which you may regard as a good approximation here. is the rate constant at temperature To. for r>R." The fast reaction ensures that x .where (a) Sketch qualitatively the expected temperature and concentration profiles. there is only B and P. and Npo. The second equality requires that IT. where it undergoes an irreversible. only A and P are present. rate constant depends on temperature according to . (d) Evaluate Oo(q) and O1(q). and the other parameters.ldt to NA. Species A diffuses from the gas into the liquid. Discuss the magnitude and direction of the error in the flux calculation which would result from assuming that k.P \ I Flame front . The system is at steady state.). \ <<)&. firstorder reaction.). and diffusivity are all constant to good approximation. = 0. =x. (b) Use an interfacial mass balance at the droplet surface to relate dR. compare the actual energy flux with that which would occur in the absence of chemical reaction.HA)are all known constants. (b) Derive an expression for the energy flux.(t). which is algebraically simpler than the general case. diffuses back to the gas.Reactive gas mixture confined between parallel plates. and temperature and composition depend on y only. E is the activation energy. is independent of position for d l i and j. the gas mixture is ideal. (a) Determine the species fluxes in terms of KO. the value of x . The combustion reaction is written as A+mB+nR where P represents all combustion products. and the combustion products do not enter the droplet. T=TL Gas (c) Assume now that E<< 1. Absorption with a temperaturedependent reaction rate. you may assume that the product CD. 114. Reformulate the result of part (b) as a perturbation problem. State the differential equations and boundary conditions which govern the first two terms in the series. = x. respectively. is found.%ENERGY AND MASS TRANSFER y= L * Problems 475  Figure P112. x. homogeneous. spherically symmetric process. The quantities which are to be determined in the analysis are To.. KL. next to the droplet. and the flux of fuel at the droplet surface. Figure P114. and determine the flux of A at the gasliquid interface at this level of approximation.C. =xh. with @(q) expressed as an expansion in E. relate NAu to x.x. Pll3. and the other parameters. and which describes the mole fraction the droplet radius at a given instant (R.. again in terns of KO. In addition to all thennophysical properties. CB= CBO Figure Pll3.. where CAois the gasphase concentration of A next to the droplet and C .ToIITo<<1. (c) In the limit TL+To. (It is sufficient to restrict your solution to Da>> 1.) (c) Use an interfacial mass balance at the flame front to relate the fluxes of the three species at r = RI Let N&) and ~ ' . T / . Oo(q) and Ol(q). A droplet of radius Ro(t). NAr(Ro)= NAw (b) Derive the differential equation and boundary conditions which govern the dimensionless reactant concentration O(r)). the product... the fluxes evaluated on the gas side. Absorption with a TemperatureDependent Reaction A stagnant liquid film of thickness L separates a gas from an impermeable and thermally insulated solid surface. For simplicity.) I where AHR is the heat of reaction.Thus.<cC. consisting of pure fuel (species A).(y).) at the gasliquid interface are also known and that the system is at steady state. is also given. however.p Gas ! A . the "flame front. Show that 1 N . f ) be the radial fluxes of i evaluated just inside and outside the front.. (Hint: Note that C. (a) Use conservation of energy to obtain an explicit relationship between T ( y ) and C.. is not known until TO. = c. and R is the gas constant. The firstorder.
v. PII6. is not known in advance. C) and T. B Bulk Solution Stagnant Film Ag Electrode I [AgCI]= T = To X~=XAO T=T_ IJ x A = x A _ y= 6 1 Cs ) ' II IAs+I= [K+]= c. The pressure in the gas is constant (atmospheric) and the gas mixture is ideal. q. such that one component (species A) condenses at the surface but the other (species B) does not. That is.) is a known function of the temperature at that interface (To)). DiffusionLimited Reaction at an Electrode Consider a stagnantfilm model for steady reaction and diffusion near a silver (Ag) cathode.) and the temperature and mole fraction of A in the bulk gas (T. a nonzero heat of reaction MR.. and C p respectively.=O for all r).) ?j. =x.).(~.xB.(r) and T = T(r)only.) are known constants throughout the film (i. the mass density ( p ) is not necessarily constant. to the rate of product formation at (a) Show that in the absence of mass transfer. and describe how to compute To. where the f. determines the apparent film thickness to use in this type of model.). Explain how to proceed. Effect of M a s s lhnsfer on the Heat Wansfer Coefficient The effect of mass transfer on the heat bansfer coefficient at a given interface can be approximated using a stagnant film model. and x. . That is. for $ 2 to. The molar latent heat for species A is X=RA(~. the thickness of the condensate layer is L. Changes in the gas composition and temperature are assumed to be confined to a stagnant film of thickness a. In general. (a) Relate each of the positiondependent fluxes. The interfacial temperature is To and the bulk fluid temperature (at x = 6) is T_. (h) State the energy conservation equations for the two gas regions and describe how to determine the remaining unknowns. with + t 1 I.are userspecified functions and the Yi are constants. impermeable. determine T(r) and Ts. Assume that these thicknesses are known. Is the result different for a gas than for a liquid? Discuss the conditions which will make hlh. [KCI]=Cp Y = L y= 0 [CI]=C. The reaction A B+C takes place at the particle surface. show how to calculate xi(r). B. Pll8. Film model for condensation. spherical particle of radius a is suspended in a gas mixture containing components A. consider a binary gas in contact with a cold. there are no chemical reactions in the fluid). (g) Use an interfacial energy balance at the flame front to relate the temperature gradients at r = R)) and R r ) to the flux of A and the heat of reaction. For simplicity.(r). (b) M v e an expression for hlh. Explain how to determine RSc. C2 Figure Pll8. Assume that the electrolyte solution contains AgCl and KC1 at bulk concentrations of C. due to variations in the mole fractions and temperature. assume that these fluxes (N.. AH.e.e. (e) Suppose now that you are given the function RS&xA. (c) Supposing that Rsc is known.assuming again that T o is known. Rsc. for the general case. and the gas is stagnant. Nonisothemal Diffusion and Reaction in a Ternary G a s A small. 117. the heat transfer coefficient is given by Thus. a measured or theoretical value of h. Assuming that To (and thus xAd are known. the surface.(Tf). Pll6. The temperature and composition of the gas are functions only of y.) and temperature (Ts) at the panicle surface are unknown.Far from the particle the mole fractions and temperature are given as x. and Rf: between N (f) Use an interfacial energy balance at the droplet to relate dRddt to the energy flux in the gas.do not react. B. 8 I I 1 Bulk Gas A... Use an interfacial energy balance to derive the additional equation which is needed. Solid Ag is formed at the surface by reduction of whereas K+ and C1. Diffusion to a silver electrode. Note that. G4. there will be nonzero fluxes of one or more species normal to the interface. either greater than or less than unity. (8) 115. and assume that you have software to solve sets of equations of the form I I 116. vertical surface. I 118. Assume that x. respectively. determine the rate of condensation. ' (e) By considering mass transfer outside the Rame front (r>R. as shown in Fig. Assume that there is a stagnant film of fluid of thickness 6 adjacent to an interface located at x = 0 . The mole fractions (xi. The mole fraction of A in the vapor contacting the condensate (x. The reaction at the electrode surface. along with the wall temperature (T. State the differential equations and boundary conditions which must be solved. You may assume that CDq is approximately constant.Problems 475 * obtain another relationship . You may assume that T(y) in the liquid is approximately linear. assume that the thermal conductivity in the gas mixture (k.). (d) Again assuming that Rsc is known.. which is driven by an external power source. Ni. T) which describes the reaction kinetics for this system at atmospheric pressure. and C. Recall that the heat transfer coefficient is defined so that it represents only the conduction (Fourier) energy flux at x = O .. (b) Show that the massaverage velocity is zero throughout the gas (i. ( i = A . . Film Model for Condensation As shown in Fig. find NAY (b) Determine T(y) in the gas f i l m .) is constant and that variations in AHR with temperature are negligible. is Gas Film A. (c) Suppose now that 7. including mass transfer.
.
and C. That is.) (b) Show that (c) Using the result from part (b). (d) Numerical methods are required to determine C.c< I. The rapid and irreversible reaction A +mB (with m # 1) occurs at the catalytic surface ( y = L). (a) Formulate the diffusion problem.(x) and $ ( I ) = &x)FIRT. but the filtrate concentrations of the small ions (C. x= 0 b PII11. and leads to a diffusion potential. and known constants. 1 I I I 1 c. /3= 010. isothermal Diffusion and Reaction in a Ternary Gas Consider steady diffusion' and reaction of a gas in the stagnantfilm geometry shown in Fig. A membrane is available which allows water and salts to pass freely into the filtrate. solutes are numbered as shown. Show that under these conditions the scale for C(x) is C$/C. state the equations which govern C. P1111. Musion Potential in Protein UltrafUtration i wheref is a function of the solute concenmtons.. C Gas c3. Assume that the bulk retentate has known concentrations of Kt. =? A.) are not known in advance.. I I Solvent VF K+ CIR cm prm C 3 ~ . termed "concentration polarization. The addtional parameter which will appear is the diffisivity ratio.(x) and the other concentrations. or D. Ns.Cl. and isobaric. As discussed in Problem 28. P1112. such that E = C.! h ENERGY AND MASS TRANSFER (a) Derive a differential equation involving the anion concentration. What is the scale for C+(x)? What was the advantage in choosing the anion? 1111.(DIL)(C~~IC. (c) Assume now that y = C. Accordingly. and it is desired to know whether a high protein charge will lessen or increase concentration polarization. which in turn suggests that Ns. conditions which could be used to determine x i ( y ) and N 1i ! . q. [Hint: Assume diffusional equilibrium of K t and C l across the membrane. p. B. The gas contains species A. to be determined later. the apparent diffusivity of the protein is p. and an anionic protein denoted as RPm (m>O). whereas species C is inert. The net protein charge can be manipulated by varying the pH. and C. as parameters. (The three Include any stoichiometric relations or auxiliary conditions which are needed.. The key feature of this problem is that the large size of the protein molecule makes D. Show that f = 1 for rn = 0. (c) Show that the differential equation for x. B. = D2./C...c1 and that the membrane is highly charged. (b) Solve for x d y ) in terms of NAYand known constants.. see part (b) of Problem 1110. Suppose that a dilute protein solution is to be concentrated using ultrafiltration.] 1112. suitable choices for the dimensionless variables and flux are (a) Using the NernstPlanck approach. show that the conservation equation for the protein takes the form Rewrite the differential equation and boundary conditions for the anion in dimensionless form. Identify each of those constants.<cD.= 0 x= S Figure P1112. state the differential equations and boundary . isothermal. Use this to relate C(0) to C. which is the same as the thermodynamic condition of uniform electmchemical potential in ideal solutions. (b) Show that the NernstPlanck equation implies that the equilibrium condition for species i is . and C(L) to C." can impede the separation process.) The effective film thickness ( 8 ) and the filtrate velocity (vF) are also given. and that f > 1 for m>O.). and the protein charge lessens concentration polarization. (d) Derive the first term in a perturbation expansion for 0 ( 7 7 ) and . (e) The analysis could have been done for the cation instead of the anion. Notice that the usual proportionality between the flux and the overall concentration difference does not hold. Thus. The mole fractions of all species are assumed to be known at y=O.Reaction and diffusion in a ternary gas. . A good approximation is that D. use this result to compute the flux. the salt flux..The gas m i x t u r e is ideal./C.and C . including how to determine CIFand C. In Ci+ zi$= constant. Stagnantfilm model for ultrafiltration of a charged protein. while keeping all protein in the retentate solution.(x) Cdx) c3fx) cIF =? c. C. where i = A . . . C(x).. and s are constants. Outline how to complete the solution.(y) can be put in the form Bulk Retentate I Film Filtrate where k. Consider the stagnant film model depicted in Fig. (Regard C. the tendency for the protein concentration at the upsheam surface of the membrane to exceed that in the bulk retentate. .. B.
.
.
is one of the few free convection problems which has a simple analytical solution.26) will be unsatisfactory. Equation (12.4X lo2 5 .' Based largely on the assumption that Ap/poc< 1 . 69).. Eqs. which concerns the stability of a layer of fluid heated from below. (12. It is convenient to rewrite the density as i/ /: ' I where /3 and pi are the thermal expansion coeficient and solutal expansion coeficient.1 m. 0 lol9 ~ 4. a pure fluid in a parallelplate channel). 1 ! where the reference condition has been chosen as T = To and Ci= C. which gives P= 3. The basis for the Boussinesq approximation is examined in detail in Mihaljan (1962). except in the term (ppo)g. 'Named after the French physicist J. the linearization of po/p in Eq.l 4 "Based on Eqs.1 x lo" 3.28) and (12. examines the conditions under which buoyancy effects are strong enough to cause motion in an otherwise static fluid. The analysis of Mihaljan (1962) assumed that density variations are linear in temperature. and To=200C. where it is concluded that two dimensionless parameters must be small. The second. Ordinarily P > O . It is seen that Eq. as in this case. whereas additional concentration terms can be added for multicomponent mixtures. The second part of the Boussinesq approximation is the assumption that the variable density p can be replaced everywhere by the constant value po. The Boussinesq approximation is examined also in Spiegel and Veronis (1960). Neglecting the viscous dissipation and compressibility effects..29). for either fluid. and e. The solutal coefficient Pi is negative in some systems and positive in others. L=0. This is crucial. The first. respectively. p. The concentration term is absent for a pure fluid. but one which is much more amenable to analysis. using a different approach. for air and water with AT= 10°C and L=0.25) and (12.4 X lo) K' at 20°C. so that p decreases with increasing I: For an ideal gas P = lITO.212) describe any combination of forced and free convection in a pure fluid. (1 2. the analysis of buoyancydriven flows may be extremely complicated.210) is the usual continuity equation for constant density. Hereafter.066 X K ' (Gebhart et al. this approach yields what is still a difficult set of coupled differential equations.211) differs from the NavierStokes equation only by the addition of the temperature term. Using the Boussinesq approximation. The effects of pressure variations on density are assumed to be negligible. For liquids. 946). where it is found that P<O for Oc Te4"C and P>O for T>4OC. the density is expressed as I TABLE 121 Values for the Parameters which Limit the Boussinesq Approximationa Fluid Water Air &I E2 2.3 CONFINED FLOWS In this section we consider two wellknown examples of confined (or internal) flows which are due to buoyancy. this is equivalent to requiring that Aplpo<< 1. (12.g. where Ap is the maximum change in density. /3= 2. (12. the sqlution to the general set of equations is so difficult that almost all published work has employed what is called the Boussinesq approximation.28) and (12. which involves fully developed flow in a vertical channel. p.2. (12..6 X 10.For water at 20°C. Values of E .1 rn are shown in Table 121.28) is by far the more restrictive requirement under these conditions. the continuity and momentum equations for a pure Newtonian fluid with constant p become ' Equation (12.25) is just the leading part of a Taylor series expansion about the reference state. Water exhibits unusual behavior near its freezing point.26). The necessary conditions are given as where AT and L are the characteristic temperature difference and length. 1988. so that the requirement for linearity is in addition to Eqs.Clearly. V. . Boussinesq (18421929). respectively. Even if one focuses on the simplest situations (e. The equations for an isothermal mixture are analogous.. the energy equation is the same as used previously. because it leads to the continuity equation for a constant density fluid and allows the viscous stress to be evaluated as in the NavierStokes equation. Allowing for variations in temperature and the concentration of species i . (12. Data for other common liquids are given in Bolz and Tuve (1970.29) with AT= IOeC. 12.1 0 x 12. (12. The first is the assumption that p varies linearIy with temperature andlor concentration. and Eq. we consider only temperature variations. When P changes rapidly with temperature. namely Within the limitations of the Boussinesq approximation. There are two parts to the Boussinesq approximation. as in Eqs.
(+ I ) =0.g. in Eq. Velocity in a vertical channel with free convection only (U=O) or combined free and forced cmvwtion(U=2.. the result for a fully enclosed channel is ' Figure 121. If @ldx is assumed to be known.211) that.. 123. With the temperature (and buoyancy force) independent of x. Returning to Eq.To) a Notice that the velocity is the sum of an odd function and an even function. . we see that the simplest expression for the velocity is obtained by choosing To= T. Flow in a vertical channel with walls at different temperatures. which is shown in Fig. and mean velocity.) is shown in Fig.36) reduces to dB d .e. and use the result in Eq. another motivation for choosing an intermediate temperature is that it will minimize the errors made in linearizing the function p(T). it is assumed that T2>T.35) to obtain + I Equations (1 2. As one might expect. As shown in Fig. Stability of a Layer of Fluid Heated from Below The system to be analyzed. consists of a static layer of fluid between horizontal surfaces main .5 0 Y/H 0.31 Flow in a Vertical Channel We first consider a vertical parallelplate channel of indefinite length. I ) 2 1 0. (12. 122. For example. Before choosing To.Example 12. = v. and U cannot all be specified independently. we assume that v. respectively.5 1 k Turning now to the momentum equation.35).With To= T.38) and (12. This velocity profile is antisymmeuic. as shown in Fig. Accordingly. yielding a thermally fully developed region with no streamwise temperature variations. it is informative to derive a general relationship involving the pressure gradient. We evaluate d 9 l d r again by setting To= T. with v. (12. reference temperature.. the channeI width i s 2H and the wall temperatures are T. Example 1232. To. [' Integrating and using the noslip conditions.9 must be independent of y. the flow is upward in the warm half of the channel (y>O) and downward in the cool half (ycO).). the mean velocity is evaluated as This illustrates the dependence of dP/dr on the reference state that was mentioned in Section 12. (12. (12.. the velocity is found to be 1 \ Only the even part of v. assume that the channel is closed at the ends. = 0 and g .. we note from the y component of Eq. with walls at different temperatures. it is feasible to have fully developed flow. and T. 121. Equation (12. In particular. the x component becomes 1! r' P Figure 122.pogP(Tm . = . v.(y) and vy= 0. In modeling a real system. With temperature a function of y only. where vo is the maximum velocity under these conditions. Using Eq.35). the energy equation reduces to which indicates that the temperature profile is linear.. 122. In this system the heat fluxes in and out of the fluid will come into balance. T.39) represent pure free convection and mixed free and forced convection. If there is a net flow (i. all that is needed to complete the solution is to specify the reference temperature. With g. Introducing a dimensionless coordinate and the arithmetic mean temperature T.36). = 0. This relationship shows that dP/dn.2. the linear profile is expressed as ..bascdonEq.39).. (12. contributes to U. the pressure gradient for U=O must vanish if To=T.(12. An example of the mixed case (with U = 2v. so that thae is no net flow and U=O. U 0) the pressure gradient will not vanish.
which are governed by Eqs. p. the ensuing flow consists of a regular pattern of hexagonal cells. This muchstudied physical situation is called the Binard problem.314) gives aT ab.211) in a similar manner gives . The growth of disturbances leads eventually to a new flow pattern. how it responds is what characterizes its stability. 9=bp+p. the approach used here does not predict tbe nature of the new flow. (1 2. and velocity. The temperature at the bottom (T2) exceeds that at the top ( T . VO. (1 2. (12. the energy equation reduces to where B and p represent small perturbations to the base case. The unknowns are 8.2. In the static fluid the temperature and pressure depend only on z. This "topheavy" arrangement tends to be unstable. The objective of the analysis is to predict the critical condition when flow begins. Expanding the various terms in the energy equation using Eq. Eq. is reproduced in Chandrasekhar (1961.19) and ( 12. whose width is close to 2L. As viewed from above. Using identity (7) of Table A1. (12. If the temperature difference (AT= T. With v=O. in order to balance the buoyancy force.Vv.. z= L 1 Fluid We are used to 9 being constant in a static fluid.323). In the Bknard problem it turns out that a given disturbance will either decay in time or grow exponentially. we conclude that the base case ([he static fluid) is stable. For conditions in which all possible disturbances can be shown to decay. from a classic study by H.33).322). As in the previous example. so that the vertical temperature gradient causes the density to increase with height. (12. and using continuity and identity (10) of Table A. it is sufficient to consider only the z component of Eq. where T. (12.21 0 ) . which eliminates p. the base case is evidently unstable. the result is The solution which satisfies the boundary conditions is where w = V x v is the vorticity.314) (12. W e will choose as the reference temperature the arithmetic mean of the surface temperatures. or v . and the three components of v.Using these expressions in Eq. because it involves a product of two perturbations. in this case the hexagonal cells described above. A photograph of this pattern. The third basic relationship is the continuity equation. if at least one disturbance can be shown to grow. but the viscosity of the fluid opposes any motion. The number of dependent variables is reduced by taking the curl of Eq. .12). This linearization is crucial to the methodology. ) .Figure 123. which is Using Eq.319) does not involve v. Here. The development here largely follows that in Chandrasekhar (1961). we obtain where G=AT/L is the magnitude of the static temperature gradient. is given by Eq. BCnard in 1900. (12.+iMa0 =at at at at' L 1 : t In Eq. the energy equation becomes Expanding Eq.320). The thickness of the fluid layer is L. pressure.1. 0 and v r .311) for 1: integrating. The other simplifications follow directly fiom the form of the static temperature profile. bT(z). On the other hand. (12.3. In the presence of disturbances. which is called a linear stabiliry analysis. With this choice. the dynamic pressure is evaluated as The original set of dependent variables has been reduced now to the primary unknowns. Static layer of fluid between horizontal surfaces at different temperatures. Taking the cud again. To=T. it is found that flow begins only when AT exceeds a critical value. the temperature and pressure are expressed as T = b T + 0. 10). because in this problem any nonzero v represents a perturbation. The assumption that the disturbances are small allows us to linearize the governing equations by neglecting all terms which involve products of disturbances. that is.TI) is increased gradually in an experiment.317) we have neglected v . the Imomentum equation reduces to Because Eq. the dynamic pressure must vary with height. (12. and letting 9 = 0 at z = 0. (12. The underlying idea is that any real system is subjected almost continuously to small vibrations or other upsets. In obtaining this form of the momentum equation we neglected v . We will first determine the temperature and pressure in the static fluid and then examine what happens when the fluid experiences small but otherwise arbitrary disturbances in temperature. and the z axis is directed upward. (12. ( 1 2. p.. No special symbol is used for the velocity disturbance..315) tained at different temperatures.
the six boundary condtions for Eqs.330). it is assumed that the disturbances are exponential in time. The disturbance(s) to a real system might have almost any functional form. and f = v . pp. two more boundary conditions are needed for the velocity. the boundary conditions for the temperature perturbation are * 8=0 at z = 0 and z = L .325). is associated with a particular wave number. Two alternative situations are considered. Hildebrand (1976. or mode.(z. what is needed is a general way to represent the possible disturbances in temperature and velocity. (12. In this representation the function is decomposed into fundamental contributions corresponding to a continuous range of spatial frequencies in x and y. The change from stable to unstable behavior occurs when s changes sign.335) and (12. so that the objective of our analysis is to define the conditions under which s =O. (12.329) Thus. Here we take advantage of the fact that an arbitrary function can be represented as Thus. Differentiating Eq. such that The usual nopenetration condition implies that v. (12. although the results for both situations are discussed at the end of the analysis. (12.333) and (12. and either (12. Using Eq.337) in Eqs. The more realistic one assumes that the fluid layer is bounded by solid surfaces at which the noslip condition applies. pp. r) be piecewise differentiable with respect to x and y in any finite interval and that the integral of la/ over x and y exist [see. Before proceeding. In applying Eq. It is not necesary to know the contributions of individual modes to any particular disturbance. Thus. but it yields a problem which is much easier to solve. In Eq. the solution is completed only for t h s case. This is difficult or impossible to achieve experimentally.336) gives Combining these two equations so as to eliminate the temperature variable. dz2 =0 at z = 0 and z = L (free surfaces). but it can be shown that in the Binard problem the imaginary part of s is zero (Chandrasekhar.With both surfaces assumed to be maintained at constant temperature. 234240)].330) to stability problems.324). the new temperature and velocity variables are @(z) and V(z).=O at z = 0 and z = L . given by The temperature and velocity are left in dimensional form. respectively. and v. Thus. for example. it follows that ! az at z = 0 and z = L (solid surfaces).329). we obtain The decomposition into modes is applied to the Benard problem by letting f = 8 and F = 8 : for temperature. for each wave number. (12.319) and (12.323). 7 (12.326) with respect to z and then reversing the order of differentiation.323) is fourth order. 2426). constant on both surfaces. (12. (12. it follows that a2v. z and 1 are treated as parameters. The only requirements are that the function a(x. Rearranging the continuity equation gives With v. y. disturbances will grow or decay with time.323) are provided by Eqs. 1961. (12.327) or (12. the constant s for a given wave number will be complex. = 0. we obtain With T~ =r . the amplitude of any disturbance may be viewed as a linear superposition of functions of the form Because Eq. (12. 2. it is convenient to introduce the dimensionless quantities i : This is one form of the Fourier integral.319) and (12. and F = V for velocity. Each contribution. Based on the foregoing. t) differs. The key point is that we are able to examine the stability of the system with respect to all disturbances by letting c vary from 0 to m. (12. (12. (12. as indicated earlier. in general. we obtain . which is to infinite intervals what the Fourier series is to finite intervals. (12.334) in Eqs. The function a. (12. Accordingly.327) I which correspond to particular modes. rather than oscillate. The various derivatives off are evaluated as In the other situation it is assumed that there is no shear stress at either surface. Using Eqs.325) In general.324) The corresponding wavelength is 2 d c .
344) (1 2. It is clearer now that the key physical parameter is Ra. this information is then used to identify the equations which govern boundary layers.. ($ . (12. the critical Rayleigh number is calculated to be Notice. (12. 12. the static layer of fluid is predicted to be stable with respect to any small disturbance. Accordingly. Using d Ra. To determine the conditions corresponding to neutral stability (i.351) indicates that the minimum Rayleigh number is %=0 dl Ra. four of the six boundary conditions for this equation are given by V=O at j = O and J= 1.h2)1 sin n*= ( n 2 d + h 2 f sin n v l = h2 Ra sin H T [ . The eigenvalue problem is simplified by expanding Eq.343) and (12. from Eqs. (12.. The solutions of the eigenvalue problems which arise when one or botb of the bounding surfaces are solid are discussed in Chandrasekhar (1961). For discussions of this and other experimental work with the BCnard system. which implies that the Rayleigh number at neutral stability is given by . the remaining two boundary conditions are found to be =o ? Because the differential equation and all of the boundary conditions are homogeneous. 69)]. PAT) is in the numerator of Ra. it appears that the solidsurface and freesurface cases each constitute an eigenvalue problem. Ra. The results for the three cases are summarized in Table 122. p. An experimental value for solid surfaces i s 1700+51 [Silveston (1958). we set u = 0 in Eq. is in .e. Eq. Proceeding largely by analogy with the derivations in Chapters 8 and 10. along with a discussion of the dimensionless parameters involved. is consistent with the expectation that a stationary boundary with no slip will resist flow more effectively than a boundary with zero shear stress. For fixed A.~lmensional%nal~sis a n y r  Layer kquations where Ra is the Rayleigh number: Notice that the density change which destabilizes the static fluid (i..342). as quoted in Chandrasekhar (1961.2. in exceIlent agreement with the theory.340). Using the velocity conditions stated earlier. Either of those books is a good general source of information on hydrodynamic stability. see Chandrasekhar (1961) and Drazin and Reid (1981). at (= 0 and j= 1 at {= 0 and l= 1 (solid surfaces).the denominator. Drazin and Reid (1981) has an introduction to nonlinear stability. Finally. (12. sin nwl.. The result is TABLE 122 Critical Values of the Rayleigh Number for a Layer of Fluid Heated from Below Qpes of surfaces Both free One solid and one free Both solid R% 657.e. the boundary conditions for the freesurface case are just v = d=2v= od 4 dc2 dC4 at {= 0 and 5 1 (free surfaces).348) It is straightforward to show that all even derivatives must vanish at the boundaries. and the viscosity. (12. . we obtain n = 1. which is expected to be stabilizing. The solution to Eq. s = 0). which is in fact the case.5 1101 1708 which is the final form of the differential equation which governs stability. (free surfaces). we now minimize Ra. The rest of our analysis is limited to the special case of free surfaces.339) (with u=O). as well as discussions of theoretical analyses of the cellular flow patterns.. Given that disturbances of any wave number might be present. Indeed. (12. there are nontrivial solutions only for specific values of Ra.345) What we wish to determine is the smallest value of Ra at which the limits of stability are reached. Combining the temperature conditions w i t h Eq.. with respect to A. it is unstable. For two solid surfaces. (12.343) (1 2. (12. For Ra<Ra. This suggests that stability will be lost when Ra exceeds some minimum value. In addition to linear stability analysis. (1 2.342) which has this property is V=a..4 DIMENSIONAL ANALYSIS AND BOUNDARY LAYER EQUATIONS This section begins with a discussion of the dimensionless forms of the differential equations for free convection. Substituting this into Eq. The trend in Ra. is increased to 1708. = (112 + h2)) k2 ' =o d2v dl2 I I which corresponds to n = 1.346) as I it is found that the critical value of A is Ac= TI*. that the righthand side vanishes.345). for Ra>Ra. for a given wave number (A).
geometric ratios).42). the Grashof number is equivalent to the square of a Reynolds number based on U. Introducing the dimensionless quantities into Eqs. the characteristic velocity is defined as f I O = O(T.43) in terms of the Grashof and Prandtl numbers gives The major distinction between forced and free convection lies in the velocity scale. To identify the velocity scale. the momentum and energy equations are written as 1 L ' . Equations (12. and the inertial pressure scale.What in forced convection would be the Peclet number is written in free convection as Accordingly. The time scale is assumed to be L/(r. Gr. we conclude that there will be a momentum boundary layer in free convection when GrlR>>1.Pr. (12. Boundary Layer Equations It was mentioned above that the Grashof number is like the square of a Reynolds number. We wish now to identify the factors which determine the Nusselt number in free convection. and viscous terms are all comparable.411) where e. In flows where buoyancy effects are prominent. is a unit vector which points in the direction of gravity. geometric ratios). denotes position on a surface.31. It is related to the Grashof and Prandtl numbers as Rewriting Eqs. and the characteristic temperature difference as AT.413) (12.42) as Grln. parallels that in Section 9. both quantities indicate the relative importance of the inertial and viscous terms. The dynamic pressure term. Setting that coefficient equal to one. Rewriting Eq. (12. contains no parameters and therefore adds no information at present. (5.43) is expressed as GrlRh'. This discussion.1020). Pr. Pr. these are the most general.Dimensional Analysrs and Boundary Layer Equations 495 Dimensional Analysis In defining dimensionless variables we denote the characteristic velocity for a buoyancydiven flow as U. defined as ' where r. which arose in Example 12.48) with Eq.42) and (12.48) and (12. as derived in Example 12. this implies that the coefficient of the buoyancy term is neither large nor small.410) and (12. emphasizes that ~ r " ' plays the same role in free convection that Re plays in forced convection. Another group that is commonly encountered in free convection is the Ray .38) in dimensionless form gives Only the Grashof number is involved here. (12. the corresponding momentum equation for forced convection. Within the restrictions of the Boussinesq approximation. we would expect this term to be of the same order of magnitude as the inertial and pressure terms.. which are the three terms in parentheses.410) (12. The dimensionless continuity equation. Whereas in forced convection the velocity scale is imposed and is therefore known in advance. (12. consider the velocity in a vertical parallelplate channel.the characteristic length as L. in free convection Ub must be inferred by balancing the appropriate terms in the governing equations. (12. we focus on the buoyancy term in Eq. Our main interest here is in the dimensionless parameters. A comparison of Eq. (12. Pr. (12.3. (12.411) that the local and average Nusselt numbers are of the form Nu = Nu(?s.2lo). within the boundary layer the inertial.32.49) indicate that the velocity and temperature in free convection are functions of the form T = V(f. Nu = Nu (Gr. Thus. Gr. which has the same form as Eq. That is. geometric ratios). Accordingly. because of the special nature of the temperature field in this flow. It follows from Eqs. employed. which is restricted to steady states. [ As a specific example.211) and (12. dimensionless forms of the momentum and energy equations for buoyancydriven Row. by analogy with the findings for forced convection in Chapter 8.. the dimensionless variables and differential operators are leigh number. geometric ratios). is retained in the momentum equations for As indicated by the second equality. buoyancy. (12. where Gr is the Grashof number.212). which may or may not be important (depending on the geometry).  (12. Outside the boundary layer the inertial and buoyancy effects are important.414) It is conventional to express the coefficient of the viscous term in Eq. G r . (12. Assuming that O is a properly scaled temperature. h ~ 2is . the coefficient of the conduction term in Eq.
.
me other dimensionless variables are defined as in Section 12.4, with L taken to be the vertical dimension of the plate. At the end of the analysis it is shorn that the local velocity and temperanue fields are independent of L That is, it turns out that the only pertinent length scale is the distance x from the leading edge (bottom) of the plate. It was shown in Section 8.4 that this is also a featwe of forced convection past a flat plate, as well as a feature of other wedge flows. Indeed, the similarity approach used here is much like rhat in Examples 8.41 and 8.42, and the reader may find it helpful to review those examples before proceeding. The similarity solution for the vertical plate was obtained first by E. Pohlhausen in 1930 (see Schlicting, 1968, pp. 300302). solution The fluid far from the surface is at the reference temperature. Thus, a satisfacto~ for the outer region is V=0, 8 = 0 , and 9 =constant; it is confirmed by inspection that this satisfies the general continuity, momentum, and energy equations. As the starting point for the boundary layer we choose Eqs. (12.420)(12.422), which contain variables that are appropriately scaled for an isothermal surface. The trivial nature of the outer solution greatly simplifies the boundary layer problem. In Eq. (12.421) we set dYI/d*=O (as determined by the outer solution) 2 Fig, 1 24) to obtain and 4 = ~ 1 (see
The boundary layer energy equation. Eq. (12.422). is transformed in a similar manner to give
It is assumed here that 8 = 6(q) only, so that the primes on O mean derivatives with respect to q. With constanttemperature boundary conditions at the wall and in the bulk fluid. O does not have to be modified with a positiondependent scale factor. Each quantity in parentheses in Eqs. (12.58) and (12.59) contains functions of x. If the similarity hypothesis is correct, all of those quantities must be constant. The two unknown scale factors give us two degrees of freedom in choosing those constants. As the first choice we set c = g3
i
(12.5 10)
to make the coefficient of O unity in Eq. (12.58). As the second choice we let
where Eq. (12.510) has been used to obtain the second equality. Noticing that c'g is proportional to (g4)', Eq. (12.511) indicates that As usual with twodimensional flows, it is advantageous to employ the stream function. The dimensionless stream function is defined here as (g4)' = 4, g(0) = 0. (12.512)
i
! !
1
The necessity for the condition g(0) = 0 is discussed later. Equations (12.510) and (1 2.512) imply that the scale factors are
Using this in Eq. (12.53) gives
The functional form of g indicates that the boundary layer thickness varies as xl". It is easily shown using Eq. (12.513) that cg' = 1. Thus, all of the terms involving functions of x in Eqs. (12.58) and (12.59) are indeed constants, as required.
Evaluating the scale factors as just discussed. the differential equations for the boundary layer become
which replaces the continuity and momentum equations. A similarity variable (t)) and mod fied stream function (0are defined now as
where the scale factor g varies as the boundary layer thickness. The motivation for the modified stream function is explained in Section 8.4. Briefly, the stream function in a boundary layer next to a solid surface scales as the boundary layer thickness (or g) times the velocity component
parPllel to the surface. The local velocity scale for the present problem is unknown, so the product of the two scale factors is denoted simply as the function c. For the change of variables. the streamfunction derivatives are calculated as
The boundary conditions for F and @ are
The boundary conditions for the stream function are obtained from the nopenetration and noslip conditions at the surface (q= 0) and the stagnant conditions in the outer fluid ( q = m). The thermal boundary conditions involve the specified temperatures at the surface, in the outer fluid, and in
the fluid approaching the leading edge at x=O. Our need to make the latter two equivalent to @(m)=O is what requires that g(0)=0, as given in Eq. (12.512). Equations (12.514) and (12.515) are coupled. nonlinear differential equations which must be solved numerically. Before discussing the results, we mention certain alternative ways to express the similarity variable and the velocity. Using Eqs. (12.55) and (12.513), the similarity variable is given by
a$ CF' a$
df
3
a2$ 2
CF"
9
3 g3
a34c~"
7
(1 2.56)
C'
aZ$ !9=CglllF'+c'F.g
afa9  (7;;) F'  % q ~ " , 8
cgl
( 22.57)
where primes are used to denote the derivatives of any function of a single variable. Using these relationships. Eq. (12.54) becomes
where Gr. is the Grashof number based on x, the distance from the leading edge; compare wirh Eq. (12.45). As is apparent fmm the second equality in Eq. (12.517). q is independent of the overall length of the plate, L. From the definition of the modified stream function, the velocity parallel to the surface is
This velocity has been made dimensionless using U,,which depends on L [see Eq. (12.44)]. However, Eq. (12.519) can be rearranged to
which is independent of the plate length. The characteristic velocity here is similar to U,, but it involves x instead of L. We conclude that the local velocity and temperature are both independent of the overall plate length. An extensive set of velocity and temperature calculations for the isothermal vertical plate, for 0.01 S R S 1000, is given in Ostrach (1953). Representative profiles for four values of Pr are plotted in Figs. 125 and 126 as F 1 ( q )and 6 ( q ) ,respectively. It is seen that the position of the maximum velocity (i.e., the maximum in F') moves farther from the surface as Pr is decreased and that the peak velocity also increases in magnitude. Likewise, the temperature variations extend farther into the fluid as Pr is reduced. A comparison of the two figures indicates that for large Pr the nonzero velocities extend much farther from the surface than do the temperature changes. This suggests that the thickness of the thermal boundary layer for Pr+m will be only a small fraction of that of the momentum boundary layer. The implications of the relative boundary layer thicknesses are discussed in Example 12.52, in connection with the factors which control Nu. Measured velocity and temperature profiles near vertical flat plates are generally in very good agreement with the above theory when the flow is laminar (Ostwh, 1953; Schlicting, 1968,
Figure 126. Temperature profiles for free convection past an isothermal vertical plate, from results tabulated in Ostrach (1953). The abscissa is defined by Eq. (12.517), and O is defined by Eq. (12.51).
f
1
,
p. 302). As with forced convection past flat plates, the Row becomes turbulent at a certain distance from the leading edge. For free convection at isothermal vertical plates the transition from laminar to turbulent Row occurs at values of Ra,=Cr, R of about lo9. Most data are for air or other fluids where Pr is neither large nor small, and the dependence on Pr implied by the use of Ra, may not be precisely correct. A detailed discussion of transition in free convection is found in Gebhart et al. (1988, Chapter 11). Similarity solutions for freeconvection boundary layers have been found for a variety of other boundary conditions on vertical surfaces (Sparrow and Gregg, 1956; Yang, 1960; Gebhart et al., 1988, Chapter 3). The constant heatflux condition on a vertical plate, studied by Sparrow and Gregg (1956). is important because it is often easier to achieve experimentally than constant temperature. Like the isothermal theory, the constantflux analysis gives results which are in good ageement with experiments. That analysis is discussed in Problem 123. Results for many geometries are reviewed in Gebhart et al. (1988). In addition to the results for Newtonian fluids, similarity solutions for freeconvection boundary layers have been obtained for powerlaw fluids (Acrivos, 1960). Information on the Nusselt number for Newtonian fluids is presented in the next example.
Figurn 125. Velocity profiles for free convection past an isothermal vertical plate, from results tabulated in Ostrach (1953). The abscissa is defined by Eq. (12.5.17), and F' is relsted to the velocity by Eg. (12.520).
Example 12.52 Effects of Pr on N u in Free Convection The rescaled coordinate and velocity , are appropriate for the mumenturn boundary layer near an defined by Eq. (12.419). 9 and 0 isothennal surface. If Pr 1, the thicknesses of the momentum boundary layer (S,) and thermal boundary layer (6d will be similar, and there is no need for further adjustment of the governing equations. Eqs. (12.420H12.422). If Pr is very large or small, however, 8, will not be comparable to a,, and the scaling will be incorrect. As shown for forced convection in Chapter 10, examining the scaling of the energy equation for extreme values of Pr reveals how R influences Nu. The objective here is to perform a similar analysis for free convection at an isothermal surface. The scaling analysis for a constantflux boundary condition is discussed in Problem 122. To identify the terms which must be balanced in the scaling procedure, we first review the characteristics of the various regions, as summarized in Fig. 127. In the momentum outer region.

I
buoyancy, inertial
Consider now what happens for h + O . In this case, buoyancy and inertial effects are dominant in virtually the entire thermal boundary layer, which is mainly in the momentum outer region (see Fig. 127). For the inertial and buoyancy terms to balance, Eq. (12.523) requires that n =O. Combined with the two general relationships, this indicates that
1 1
viscous
1
m=i,
n=O,
p=&
(Pr+O}.
(1 2.525)
For R+, buoyancy and viscous effects are expected to be important within the relatively thin thermal boundary layer; we cannot yet tell about the inertial effects. For the buoyancy and viscous terms to be comparable, Eq. (12.523) requires that 2m n = 0. Solving again for the three constants gives
t 6 , (Pr << 1)
A /
As a consistency check, we examine what this implies about the inertial terms. Using these constants in Eq. (12.523), we find that the inertial terms in the thermal boundary Layer are 0(Pr') as Pr+m. Thus. the ineaial terms are indeed negligible, as indicated in Fig. 127. To infer the functional form of the Nusselt number for heconvection boundary layers, we first recall from Section 10.4 that
Figure 127.Characteristics of boundary layer and outer regions in free convection. Indicated are the relative thiclmesses of the momentum and thermal boundary layers, along w i t h the terns in the momentum equation which are important in each region.
the inertial and buoyancy terms are dominant and the viscous terms are negligible. In the momen
t u m boundary layer, the viscous terms are important (by definition), and they are balanced by
some combination of inertial and buoyancy terms. The analysis assumes that the pressure term is no larger than the buoyancy term and does not consider the pressure gradient explicitly. Based on what we h o w about forcedconvection boundary layers, we expect 416, to become very large as R+O. Thus, the thermal boundary layer will reside largely in the momentum outer region. Conversely, we expect 446, to become very small as R+ , in which case the thermal boundary layer will occupy only a small fraction of the momentum boundary layer. In this region very near the surface, it will be shown that the inertial terms become unimportant. For extreme values of Pr, the new variables for the thennal boundary layer are defined as where A@ is the dimensionless temperature difference between the surface and the bulk fluid. Accordingly, the key piece of information is the relationship between the scaled coordinate Y and the original dimensionless coordinate jf. Based on Eq. (12.419) and the values of rn just determined, those relationships for the two extremes of Pr are
when m, n, and p are constants which are to be determined. For the isothermal surface considered here, the temperature need not be nscaled. Converting the continuity, momentum, and energy equations to the new variables, we obtain
What this means, for example, is that the thermal boundary thickness for large R varies as Gr1'4Pr114. Accordingly, the local Nusselt numbers for the two cases must be of the form
when the functions f,(Q and f2(9 are independent of Gr and Pr. The functional forms of the average Nusselt numbers are then (12.532) Nu = C1 ~ r pr1I2 " ~ (Pr +O),
=
c2e l f 4 pr1I4
(R+),
(12.533)
m e unknown constants will be chosen so as to eliminate R while preserving the proper balance of tarns in each equation. Certain relationships must hold for either small or large Pr.From Eq. (12.522) we infer &at m  p = n. This implies that the two inertial terms in Eq. (12.523) are comparable to one 'other, as are the two convection terms in Eq. (12.524). Similar conclusions were reached in the analysis of forced convection in Chapter 10. The other general statement is that, in the thermal h n d a r y layer, conduction and convection must be comparable. Equation (12.524) indicates then that 2m  1 = n. Thus, two of the three relationships needed to determine the constants are valid for either extreme of Pr.
where C1 and C2 are constants. The asymptotic behavior of the average Nusselt number is illustrated by the results of Le Eevre (1956) for the isothermal, vertical flat plate: (12.534) (Pr+ 0), Nu = 0.8005 GrU4Pr1'2
Nu = 0.6703 ~ r prLf4 " ~

(Pr +).
(12.535)
It is seen that the constants in Eqs. (12.532) and (12.533) for this case are C , =0.8005 and C2=0.6703. Kuiken (1968, 1969) confirmed these expressions and used matched asymptotic expansions to obtain additional terms in the respective series involving PI. An expression used by Le Fevre (1 956) to fit theoretical results for all R is
plates, vertical axisymmetric flows, inclined surfaces, horizontal surfaces, curved surfaces (eg., horizontal cylinders), spheres, and other shapes are found in Gebhart et al. (1988).
Mixed Convection
The presentation of convective heat and mass transfer in largely separate discussions for forced convection (Chapters 9, 10, and 13) and free convection (the present chapter) suggests that only one mechanism is important in any given situation. This is true for many, but not all, applications. Useful guidance on when to expect forced, fiee, or "mixed convection" regimes in laminar or turbulent tube flow, based on the values of Re and Ra, is provided by Metais and Eckert (1964). A general analysis of combined , , forced and free convection in laminar boundary layers is given in Acrivos (1966), in which it is shown that the controlling parameter is Gr/Re2 for Prc<1 and G r / ( ~ e ~ p r ' ' ~ ) ' for Pr>> 1. Numerous other findings for mixed convection are reviewed in Gebhart et : al. (1988, Chapter 10). A good source for additional information on buoyancy effects in fluids is Gebhart ' et al. (1988), which has already been cited a number of times. This is a very comprehen: sive text with an emphasis on engineering applications. Another general source, with more of a geophysical flavor, is Turner (1973).
I
F'igure 128.Avenge Nusselt number for free convection past an isothermal, vertical plate. The dashed lines are from Eqs. (12.534) and (12.535), the solid c w e is from Eq. (12.536), and the individual symbols are based on the results of Osrrach (1953).
References
As shown in Fig. 128, tlus agrees with the two asymptotic formulas and also matches the results of Ostrach (1953) computed at discrete values of Pr. Local values of the Nusselt number for vertical plates are usually expressed as Nu,=hrlk. As an example, for the isothermal, vertical plate at large Pr, Kuiken (1968) obtained (Pr+w, Tw constant). The corresponding result for a constantflux boundary condition (Gebhm et al., 1988, p. 89) is
Nu,= 0.6316 Gr:'I5 Pr1IS Nu, = 0.5028 Gr:l4 Pr1I4 Acrivos, A. A theoretical analysis of laminar natural convection heat transfer to nonNewtonian fluids, AIChE J. 6: 584590, 1960. Acrivos, A. On the combined effect of forced and free convection heat transfer in laminar boundary layer flows. Chem Eng. Sci. 21: 343352, 1966. Bole, R. E. and G. L. Tuve (Eds.) Handbook of Tables for Applied Engineering Science. Chemical Rubber Co., Cleveland, OH, 1970. Chandrasekhar, S. Hydrodynamic and Hydromagnetic Stability. Clarendon Ress, Oxford. 196 1. Cormack, D. E.. L. G. Leal, and J. Imberger. Natural convection in a shallow cavity with differentially heated end walls. Part 1. Asymptotic theory. J. Nuid Mech. 65: 209229. 1974. Drazin, P. G. and W. H. Reid. Hydrodynamic Stubiliv. Cambridge University Press, Cambridge. 1981. Gebhart, B., Y. Jaluria, R. L. Mahajan, and B. Sammakia. BuoyancyInduced Flows and TransportHemisphere, New York, 1988. Goldstein. S. Modem Developments in Fluid Dynamics. Clarendon Press, Oxford, 1938 [reprinted by Dover, New York, 19651. Hildebrand, F. B. Advanced Calculus for Applications, second edition. RenticeHall, Englewood Cliffs, NJ, 1976. Kuiken, H. K. An asymptotic solution for large Prandtl number free convection. J. Eng. Math. 2: 355371, 1968. Kuiken, H.K. Free convection at low Prandtl numbers. J. Fluid Mech. 37: 785798, 1969. Le Fevre. E. J. Laminar free convection from a vertical plane surface. Ninth International Congress of Applied Mechanics. B N S S ~ ~ 1956, S , Vol. 4. pp. 168174. Metais, B. and E. R. G . Eckert. Forced, mixed, and free convection regimes. J. Heat Transfer 86: 295296, 1964.
(Prr,
q, constant),
r is * Gr with AT=q,Llk. is Gr, with AT=q, xN;. The differences in exponents between Eqs. (12.537) and (12.538) indicative of differences in the scaling of the thermal boundary layer thickness. However. fie two expressions can be put in more comparable form by noting that Gr: = Gr, Nu,. Thus, Eq. (22.538) is rewritten as where
01:is the rnodijied Grashof number based on r Essentially. ~
"d
(Pr q , constant). This differs from Eq. (12.537) only in that the coefficient is 1 2 4 larger. Equation (12.540) is impractical for calculations because the value of AT, which is needed to determine Gr,, is not known in advance for the constant flux case. Equation (12.538) has only known constants on the righthand side, m&ng it the more appropriate expression. Numerous other results for vertical
00,
Nu,
= 0.5630
~r:'~ fill4
Mihaljan, I. M. A rigorous exposition of the Boussinesq approximations applicable to a thin layer of fluid. Astmphys, J. 136: 11261 133, 1962. Ostrach, S. An analysis of laminar freeconvection flow and heat transfer about a flat plate parallel * to the direction of the generating body force. NACA Tech. Rept. No. 11 1 1 , 1953. Schlicting. H. BoundaryLayer Theory, sixth edition. McGrawHill, New York. 1968. S~arrow, E. M. and J. L. Gregg. Laminar free convection from a vertical plate with unifom surface heat fiux. Trans. ASME 78: 435440, 1956. Spiegel, E. A. and G. Veronis. On the Boussinesq approximation for a compressible fluid. Astmphys. J. 131: 442447, 1960. Turner, I. S. Buoyancy Effects in Fluids. Cambridge University Press, Cambridge, 1973. Wooding. R. A. Convection in a saturated porous medium at large Rayleigh number or PCclet number. J. Fluid Mech. '1 5: 527544, 1963. Yang. K.T. Possible similarity solutions for laminar free convection on vertical plates and cylinders, . I Appl. Mech. 27: 230236, 1960.
where T, is the Ruid temperature far from the surface, assumed constant. For large Gr* and moderate Pr, the y coordinate, velocity components, and temperature are rescaled as
where a, b, c, and d are constants. For extreme values of Pr, these variables are adjusted further as
Y F Prm, ~
~,=i. hn, ,
v
H=G ~ r q ,
1
where m n, p, and q are additional constants. The objective of this problem is to determine the various constants and to infer the functional form of %.
(a) Use the heatflux boundary condition to show that a = d and (for extreme Pr) rn = q. (b) Using the continuity and momentum equations, show that a = , b = h , c = A, and
Problems
121. Flow in a Closed Horizontal Channel Heated at One End Consider the situation depicted in Fig. P121, in which the fluid in a parallelplate channel with closed ends is subjected to a horizontal temperature gradient. The temperatures at the ends are such that T,zT,, and the top and bottom are perfectly insulated. A buoyancydriven Row results. If the channel is sufficiently thin, the flow will be fully developed in a core region which
d = &.
(c) For Pr+O, show that m = f , n =
5 , P = $ ,q = i , and
$, p =
i
(d) For
show that m =
8, n=
8 , q = k, and
extends over most of the channel length; the ends of this region are indicated approximately by the dashed lines.
!
(a) Determine v,(y) in the core region. Assume that the end regions, where v, # 0 and the streamlines are curved, are a negligible fraction of the channel length. Assume also that convective heat transfer is negligible. (b) What dimensionless criteria must be satisfied to make this analysis valid?
A more complete analysis of this problem, including asymptotically matched solutions for the core and end regions, is given in Cormack et al. (1974).
123. Free Convection Past a Vertical Plate with a Constant Heat Flux Consider a boundary layer next to a vertical plate of height L with a constant heat flux q , at the surface. The objective is to derive a similarity transformation which is suitable for moderate values of R.The basic dimensionless quantities are as defined in Section 12.4, except that with AT q,L/k the modified Grashof number Gr* replaces Or. Using the results of Problem 122, the appropriate variables for large Gr* and moderate PTare
122. Scaling for a FreeConvection Boundary Layer with a Constant Heat F l u x The scaling for a freeconvection boundary layer at a surface with a constant heat flux
The similarity variable (q),modified stream function (F),and modified temperature (G) are defined as
differs from that for an isothermal surface, in that the surface temperatun depends on the boundlayer thickness. Thus, the temperature scale is not known in advance. As an example, consider a flat plate with vertical dimension L and constant heat flux q, at the surface. Setting AT= q&k, the dimensionless temperature and modified Grashof number are defined as
What is to be determined are the scale factors, g($, c ( 3 , and b(@.
(a) Show that the ratio b(X)/g(Z) must be constant. What is the advantage in choosing b(q/ g ( f ) =  l? (Hint: Examine the constantflux boundary condition.) (b) Determine the set of scale factors which gives
1
I
T = T,
jg Fluid
v
I
T = T,
G"+ Pr(4FG1  F'G) =0,
R ~ r P121. e Flow in a closed horizontal channel w f i ends at different temperatures.
This is the set of equations solved numerically by Sparrow and Oregg (1956) in their analysis of this problem. (The function G used here is equivalent to their 0.)
(c) Confirm that the local Nusselt number is given by
(d) Show that, according to this approach, Nu is given by
as stated in Sparrow and Gregg (1956).
124. Temperature Variations in a Static Fluid Assume that the temperature in a fluid is described by
Compare this with the exact results for the isothermal plate in Section 12.5.
127.BuoyancyDriven Flow in Porous Media
Buoyancyinduced flows may occur in saturated (i.e.. liquidfilled) porous media, thereby affecting, for example, heat transfer from underground pipes or storage tanks. According to Darcy's law (Problem 22), the velocity of a constantdensity liquid in a porous solid is proportional to V P  pg. Using Eq. (12.23). the generalization to include buoyancy effects is
where a and b are constants and the y axis points upward. For which combinations of a and b is o be static? Consider the possibility that either constant may be negative, it possible for the fluid t zero, or positive. 325. Integral Boundary Layer Equations for Free Convection Show that the integral forms of the momentum and energy equations for steady, twodimensional, freeconvection boundary layers are
1
k
'
;
where K is the Darcy permeability. When using Darcy's law to replace the momentum equation, the ability to satisfy noslip boundary conditions is lost. Accurate results are obtained nonetheless, provided that the system dimensions greatly exceed the microstructural length scale of the porous material. For steady conditions, the energy equation for porous media is
where a , , is the effective thermal diffusivity of the liquidsolid system.
(a) Define the dimensionless variables which convert Darcy's law and the energy equation
to
where i$,,= S,IL and $= 6dL are the dimensionless boundary layer thicknesses. Start with Eqs. (12.4 16H12.4 18). and assume that the outer fluid is stagnant and isothermal.
126.Integral Analysis for Free Convection Past a Vertical Plate
7 l e objective is to obtain an approximate solution for the boundary layer next to an isothermal vertical plate, using the integral equations given in Problem 125. As in the original analysis by Squire in 1938 [see Goldstein (1938, pp. 641643)], the functional forms suggested for the velocity and temperature are
where Ra is the Rayleigh number for porous media. What velocity and pressure scales are implied? (b) If Ra is large, a boundary layer will exist. Consider the twodimensional configuration shown in Fig. 124, but with the fluid replaced by a saturated porous material. In terms of the dimensionless stream function defined as
where 9 is defined in Eq. (12.419) and f and 8 are functions which are to be determined. No distinction is made here between the momentum and thermal boundary layers.
(a) Discuss the qualitative behavior of the functional forms suggested for Cx and 8, and
show that the boundary layer equations are
explain why these are reasonable choices. (b) Show that the system of equations which must be solved is
a2$ a@ = sin 4, ay2 ay
(c) For specified temperatures at the boundaries, the scaling of 63 is unaffected by Ra. In such cases, the equations of part (b) are rescaled by adjusting only the y coordinate and stream function as
(c) Determine 6 and f.(Hint:Assume that 6 = h A m and f = B in.)
j = j J Ra",
$= S/ ~a~
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
is given below. or vi= (llhi) a/&. z=z. the curl is written most compactly as a determinant: av.x.. Alternatively.725) is more convenient for derivations involving differentiation or integration because it involves only constant base vectors. which we have written as either (4J z) or (r.r2. In this case it is readily shown from Eq. The remaining term is expanded further as ( v V h = % ax c v q .. note that in cylindrical coordinates there is an irregularity in our notation..724) l h i s completes the general results for orthogonal curvilinear coordinates.713)].). Equation (A. the term on the far right vanishes [see Eq.v_dV"+_y+. . Whichever expression is used. f is any differentiable scalar function and v is any differentiable vector function. (A. are shown in relation to rectangular coordinates in Fig. v xv= [$$Iex+ [2$Ie. =ax 3 avz (Vv).720) When all components are included. Material Derivative The material derivative is given by Cylindrical Coordinates Circular cylindrical coordinates.73) that hx=hl = h. (vv). Expanding one component.. =ay av.. denoted as (r.4. = I. the pnncipal results are summarized in Table A2. 0. . such that lr(= (r 2+ z 2 ) l n f r: . A3(a). =ax (A. For convenient reference.~ = av az az av (A.+ze. where e.+ze.+ [$%Iec av ay a~ VY=s+$+2 a2f a2f a2f Again.nguIar coordinates. Rectangular Coordinates In rWta. y = r sin 0..722) " I n these relationships. (A.=L (VV). (A. The remainder of tlus section is devoted to the three most useful special cases. and it is found that tbe general expressions for the differential OWrators reduce to the forms given in Section A.718) by setting v = V. we obtain avv.=% a2 The result is (Vv). the position vector is expressed as r = r cos 0ex+r sin Be. = d"r (V v ) . vf'/=dfd +!Ye +!!fez ax i)y av ax az az Curl The curl of v is evaluated in a similar manner.61). Using x = r cos 8 . z). the position vector is given by Eq. Laplacian The Laplacian for curvilinear coordinates is derived from Eq. the unit vector in the radial direction. (A. (W.Orthogonal Curvilinear Coordinates a r~ a TABLE A2 Differential Operations in Rectangular Coordinatesa which is the general expression for the divergence.. (A. the position vector is given by r = re. =% (Q v ) .
=ez. Inverting the relationships in Eq. (A. and e..73) and (A.726). on 0 is shown in Eq. .~ RAND TENSORS S (a) (b) Figure A3.730) to find expressions for the rectangular base vectors... the expression for e. dr ae and e.=(l/h. + sin Be. the scale factors and base vectors for cylindrical coordinates are found to be er= cos Be. consider the 6 quantities in cylindrical coordinates.. Cylindrical coordinates (a) and spherical coordinates (b). To illustrate the derivation of scale factors and base vectors. a .= . V T. including V2v. e. From Eqs.r sin 8ex+r cos Be. 0 5 0 5 rand 0 5 4 5 2 2 ~ .+ cos Oe.78) as A summary of differential operations in cylindrical coordinates is presented in Table A3.=h o e. (A.730). = sin Be. (A. 0 5 8 5 2 ~spherical. (A. (A. confirms the equivalence of Eqs. . (A.) (dr/a@.Repeating the calculations for the r and z quantities. e. may be obtained from the tables in Chapter 5.77) and (A.725) and (A.725) it is found that . we obtain ex= cos Oe. The differential volume and surface elements are evaluated using Eqs.730a) The dependence of e. and v * VV. Several quantities not shown. The ranges of the angles are: cylindrical.sin Oe.
1 1 a ~ . . it may be necessary to reverse the order of A and B (thereby changing the sign of n) to make the unit normal point outward from a closed surface. dr 1 r aeee+ r sin af e a+e& 1 + cos be. is tangent to a "u curve" on the surface (i.. V 6 7." In general. differential operators. In this case the position vector at the surface is given by S ~ i f i c a l l yA .. This completes the information needed to understand the integral forms of the various conservation equations.sin +em. may be obtained from the tables in Chapter 5. Much of the material in this section is adapted from Brand (1947).739a) (A. their cross product is orthogonal The two tangent vectors are found to be .73).739b) Finally. As already mentioned. (A.. a r sin 0 130 sin o ) + . 2 dS+= r dr dB. Suppose now that u=x. (A. A compilation of scale factors. It follows that a unit normal vector is given by n=..736). (Vv)*. the expression for e. The position vector at the surface is denoted as r. The complementary expressions for the rectangular base vectors are ex= sin 8 cos +er + cos 0 cos 4e. From Eq. v = y.( ~ v ~ ) + . Many other orthogonal coordinate systems have been developed.738b) (A. (A. confirms the equivalence of Eqs. However.738).sin Be.( v .. two vectors that are tangent to the surface are Depending on how u and v are selected. (A. (A. cot 6 r A.sin +ex (1) Vf=e af r + .740) dS. including V2v. + cos 0 sin e4 = . to both. several other quantities.= r sin 9 dB dg.741) A summary of differential operations in spherical coordinates is presented in Table A4. and solutions of Laplace's equation in 40 such systems is provided in Moon and Spencer (1961).= r sin 8 dr d$.8 SURFACE GEOMETRY In Section A. and the surface is represented as the function z = F(x..A x B JAXB~' (A.= sin 0 cos +ex + sin 0 sin TABLE A4 Differential Operations in Spherical Coordinatesa 4ey+cos Oe.5 a number of integral transformations were presented involving vectors which are normal or tangent to a surface. and hence is normal to the surface.= I avo r sin 8 d 4 V.738a) (A. The objective of this section is to show how those vectors are computed and how they are used to define such quantities as surface gradients.738c) e. these vectors are not orthogonal to me another and they are not of unit length.735) and (A.VECTORS AND TENSORS Suriace Geometry s rt e.v =r ~ dr . (A.(u. and v Vv.+cos 4e+. dS. = cos 6 cos 4e. v). u and v.1 af $e. the differential volume and surface elements are evaluated as dV=r 2 sin 0 dr dB d+ (A.82) Normal and Tangent Vectors We begin by assuming that positions on an arbitrary surface are described by two coordinates. y).. e. a curve where v is held :Onstant) and B is tangent to a "v curve.e.=sin 0 sin #er+cos 0 sin +e.~ 0 d+ r sln av The dependence of the three base vectors on B and 4 is shown in Eq. "In these relationshipsf is any differentiable scalar function and v is any differentiable vector function. which are not necessarily orthogonal.
this operator is for surfaces what V is for threedimensional space. The special properties of such sets. they are obviously not coplanar. the surface gradient operator is (A.811). (A. denoted as V. B. (A. The volume of a parallelepiped which has the three vectors as adjacent edges will be nonzero only if the vectors are not coplanar. Evaluating the unit tangent using Eq. as given by Eqs. J z)=z. (A. (A. H (A. 294) It is readily confirmed that if G(x.F(x.86). righthanded sets. they satisfy + For a surface described by z = F(x. It is ~hghtforward to verify that these relationships will hold if where H is given by Eq. consider a planar surface corresponding to a constant value of z. However. we obtain From Eqs. 8.8. as discussed in Section A. make them very convenient for the representation of other vectors. n) and a second set (a. n).. Indeed. 1947). Then. a brief discussion of the properties of reciprocal bases is needed. The special properties of these base vectors leads them to be termed recipmcal bases.1 8) Notice that each base vector is orthogonal to two members of the recipmcal set. In this case H = 1 and the surface gradient operator reduces to a=B X n H * b=n x A H ' AXB c==n.f: Accordingly. we define the quantity in parentheses as V.3. this formula yields the same result as in Eq.=e. (A. denoted as e. Thus.71). (A. the result is (A.63).71) and (A. Suppose now that (A. To derive an expression for V. neither of which is orthogonal. Surface Gradient Reciprocal Bases Until now we have employed base vectors. compare Eqs. base vectors need not be orthonormal.+ and the unit normal is computed as An orthonormal basis is its own reciprocal. As a simple example.62) and employing the coordinates u and v.17) Because A and B are tangent to the surface and n is normal to it. B. that volume is equal to the scalar triple product. c) are reciprocal bases. (A. .89) and (A. y).. thus.810) V.89) it is found that Accordingly. the rate of change of a scalar function f along the curve is given by By analogy with Eq. Accordingly. As already mentioned. an arbitrary vector can be expressed in terms of any three vectors which are not coplanar or. we consider a sulface curve with arc length s. (A. Their scalar triple product is 1 I The key to describing variations of geometric quantities or field variables over a surface is the surface gradient operator. p. For a surface represented as z = F(x. any three vectors with a nonvhishing scalar triple product constitute a possible basis. by definiton. n may have to be replaced by n to give the outward normal. it is found that An alternate way to compute a unit normal is to represent the surface as G(x. or even orthogonal.7Z ) . y. not linearly dependent (Brand. it proves convenient to employ two complementary sets of base vectors. b. which form orthonormal.814) and (A. y). One basis of interest is the set of vectors defined above. equivalently. In describing the local curvature of a surface and other surfacerelated quantities. 1976. 2) = O and to use (Hildebrand. y ) .+e ax d Yay  d (z = F= constant). For an orthonormal' basis we have H = 1. H 0.
. Specifically. 1947. B. B.824) gives ye=  1 2R (cylinder). 1901. Moon. Vector and Tensor Analysis. Wiley. the curvature of cylinders and spheres is uniform (i. 1953. Hildebrand.y2)112. y).823) that 1 yf= R (sphere). Feshbach. (AS7) gives so that an alternative expression for V. New York. Morse. with the surface defined as z = F(x) = ( R ~ x 2 ) l R . Hassager.which is simpIy a twodimensional form of V involving the surface coordinates x and y. this is shown using Eq. Wilson. 1963. Mean Curvature The unit normal vector will be independent of position only for a planar surface.827) As discussed in connection with Fig.Cartesian Tensors. A2. which involves the surface gradient and surface curvature.. N. References For a surface with z = F(x) only. 1962 (reprinted by Dover. this simplifies to Aris. y). New York. Bird. McGrawHill. the mean curvature is given by This result is used in Chapter 5 in deriving the stress balance at a fluidfluid interface. P. For a general closed surface. y) = ( R ~x2 . 209). (A. P f For a surface expressed as z = F(x. the unit vector m= t X n is tangent to the surface S and outwardly normal to the contour C. 1976. Methods of Theoretical Physics. R. Hay. and E. (A. New York. NJ. Wiley. C. Intmduction to Mechanics of Continua. Bird. 1987. and the Basic Equations of Fluid Mechanics. and %e>0when n points toward the local center of curvature. For a cylinder of radius R. it is found from Eq. P. the gradient and surface gradient operators are related by Integral Transformation One additional integral transformation is needed. %?is negative or positive according to whether the surface is locally convex or concave. W. 1953. Dover. Cambridge. x<O x>O . New Haven. the more rapid the variation in n. F. Vector and Tensor Analysis. 1960. edition. Englewood Cliffs. 1961. G. Armsbong. PrenticeHall. New York. A measure of the local curvature of a surface that is useful in fluid mechanics is the mean curvature.. (A. it is found that when e. Dynamics o f Polymeric Liquids.820). second edition. New York. E.826) Unlike most surfaces. Wiley. PrenticeHall. B. B. As shown in Brand (1947. Cambridge University Ress. including the effects of surface tension. Vol. X. Brand. Eq. Berlin. points toward the local center of curvature. In other words. (A.825) For a sphere with z = F(x. using the symbol fox mean curvature. 1961 (reprinted by Dover. B. and H. Vector Analysis.e. E. is (A. R. W. points away from the local center of curvature (as in these examples) and when e. M. and 0. New York. R. Ginn. Advanced Calculus for Applications. Transport Phenomena. or. (A. respectively. Tensors. Setting v = n f in Eq.which is proportional to the surface divergence of n. Vectors. The rate of change of n along a surface is clearly related to the extent of surface curvature: The greater the curvature. Jeffreys. The lefthand side is rearranged by first noting that n x V = n x V. Stewart. L. second Two important special cases are cylinders and spheres. Prager. 1989). H. Field Theory Handbook. For any piece of a surface described by z = F(x. Yale University Press. it is found that Subtracting the term n n m Vfrom V has the effect of removing from the operator any contributions which are not in the tangent plane. After further rnanipulation of the lefthand side. 1 . and D. R. Spencer. Englewood Cliffs. p. Lightfoot. 1973). SpringerVerlag. %e<0when the outward normal n points away from the local center of curvature.. E. independent of position). New York. NJ. New York.
. R.. 226. 4. 417.. P. 166. bold numbers indicate where full citations appear. R. 225. S. C. 20. 493. A.. 323. C. See Hik. M.. 529. 212. Curtiss. See Hirschfelder. C. 470 Dahler. 580. 392. P. 133. 429 Brinkman. Bender. G.. G. W . 363 Chwang. 117 Beris. 117 Brand.. 397.. 404. J. L. A. 161. R. 444.. Carslaw. B.. 194 Crawford.248. Berk.. 388. 347. D. M . See Young.. M. R. 94.A. N. 26. 426. H . S. 194 Ash. 581.. M. 281. 342. 324 Chiang.. A. 404 Atkinson.. 63 Cess. M.. K. 0. J... J. C.P. 315. E. G. Ayyaswamy. 417. M. S. P. T. M . Brebbia. B. 108. E. 581 Breach. P. R. 99. 324. 316. Bolz. G. 324. 117 Batchelor. H . E. J . A. Crawford. M. R. 391. E. S. R. T. 506 Cornish. See also Sureshkumar. M. E.. 191. Brenner. 313. L. 545 Anderson. 484. M. 116. S.404. 387.. S.. 404. 505 Borg. M. 25. Brown. 281 Churchill. D.. 326 Brocklebank. 314. M .A.. 26. 551. 248. R. W. S. 63. 422. 194. R. M. 209.AUTHOR INDEX Page numbers are given only for first authors. 148. M. 194 Cameron. Brian. Chizhevskaya. 248 De Groot.488. D.. See Casey. V. 404 Acrivos. 395. R.. 551. 268.. See also Johnson. 119 Catchpole. 117. 265. Block. 563. 52. See Kevorkian. E. C.. 529. J.. Deissler. 367 Cole.. Christiansen.. 396. T. 194. 404. 505. E. Chandrasekhar. J. M. E. 309. N. F..581 Armstrong. B.. J. W. See Chapman. 248. J . J . 117. 324. 281..388. See Atkinson. 227. 450. H. E. R. 58.. H. P. See also Bungay.545 Bird. 63 Arnundson. 220. 551. J. 6. 470 Deen. R. See Hatton. 117. 505 See also Bowen.516. R. A.A. K. 512. See Kays. W.M. 470. W. 413. 501. 133. 10. V. Juhasz. B. D. 362 Bates. 194.T. C. 117. 100. Edwards. Colton.. 194 Churchill. 127.. Cussler. J. 309. G . Lewis.538. 545 Denbigh. J. 145. R. L. L. H. Bhatti. Jr. R. 505 Chapman. M. 26. See Atkinson. 429... R. J.. W. 26 Chester. 432.404. Jr. N. S. C.. 281 Aung. 133. A. 22. 58. S . See Bhatti. 363. See also Hirschfelder. S. 122. 194. 392.. R. A. I.. See Mason.. H. 194 Casassa. 428.. J. M. 20. 491. B. See Sparrow. See also Rarnkrishna.. 430 Cormack. 311. Allen. 464. S.. 443. 447. 353. 229. R. A. Crank. 342. 146..403.. I. S. P. D. 399. M. 194. See Johnson...517. A. See Colton. A. S. 199 Casey. 266. 26. B . 505.26.. 179. 17. 212.. 576. 0. 130 Arpaci. D.C.. J. 0. 324 Burton. 93.. J. 26 Aris. J. 63 Bungay. 25. 26 Bowen. See Glazov. K. 152. 404 . H. 375. 265. C. 133.. 330 Burgers. See Bud. 95. E. P. L. P Z I \ I / i 1 1 Abrarnowitz. 324. P. Andersen. 429 Barenblatt. E. R. 377. 133. 151. I. S. S. Brenner. 176. Happel. 404. D. M.417. C. Choi. D.. Cowling. N. B. Arnold.. G. H. 63 Butkov. R. See Skelland. 404 Brown. H.. K.. See Chester.. 324 Cochran. 265. Jr. H. 429. 117.. C. 400. 133. M . N. F. 392. 194. 281 Card. W . B. A.V..
Orszag. Petukhov.. O. P. 529. J .. 553. Falkner. J. See Christiansen. See Rogers. J. See Mason. J. T. 363 Imberger. M. D.. 461. 108.. Edwards.. See Catchpole. 548 Nye. 10. E. See Bhatti. S. See Kim... Drazin. 117.. C. B . 117 f i m . E. 363 Dukhin. 245. Reichardt. S. Lemmon. B. O. S. 350. A. 363. W. 99. H. 324.. 508. 913. See Semn. J. 430. 546 Rouse. M . M. 5 16. 212. 195. M.. 146. See Carslaw.. Mason. R. See Cormack. Nayfeh. 363. 295. C. B. 430 Lifshitz.. Scriven. C. S.. Goldstein.. 540.. 94. S. 353. 406 Oppenheim. 308. L. 4. 426. 375. Sellars. N . 546. 235. R. 185. A. 324. G . Ruckenstein. S. S. 551. 184.. 504. R. J. See Vasaru. V. E. C. L.569. K. W. 63. 422. 195 Mazur.363 Lance. 506 MilneThornson. S. R. 505 Kutateladze. Jain. 546 Kuiken. 551.. 513. 363 Hassager. 404. 27 Matijevic. 538. See Glazov. V. R. J . 309. 17. I.. 293. 35. H. 546 Jovanovic.. T . J. S. M. E. 195 Jeffreys. A. Imai. 116. 546. W.. 225. C . Segel. 220. 464. 545. Hatton.. G. 551. 195. N .. J. 349. 520. P. S. D. McComb. S . E. 324.. Kevorkian. D. 248. 248. L.. 503. Z . 568. R. See Grew. 328 Harper. See also Cor D. 581 Hinze. C. 255 Poling. J. 192.. J. R. P. 549. B . C. M . See Burton.. 195. See Papoutsakis... 435. 194.. Serrin. M. 544. W. 349. E . W . See Drazin. Reynolds. F. G. P. 13. 350.. I. 505 Middleman. A. 535. See also Tan. M. Laufer. 430 Ladyzhenskaya. D. See Tennekes. Kays. Lighthill. 0.. 316. 529. 347. Luss. M . B . S. 363. 13. H .. J. E. Lowrie. A. 363 Moffat. Y. 2. E. 581 Moser. W. J. 202. 166. 248 . 543. P. J. See Casey. 500. G. Keey. S... J.364 Jones. 363. 539. 193. 316. 512. 147. See Acrivos. 505. Rector. L. 463. 368 Rohsenow. 147.. 532.. R. 539. 324 Lamb. 546. 308. 198 Kakq... D. Ostrach. J. 63. Peck.. 545. B. 148 Fulford. B.. S . 504. Goldstein. 275. 38. E. E.. H. 198 Kemblowski. 430 Hartnett. See Anderson. G. See also Brenner. 281. 529.. Lumley. 336. A. S. MUller. 566.. 195 Newman. See Russell. 294. R. M . R.D. G. 552. 471. 529. Israelachvili. A. R. D. L. R. B. D. 518. C. A. 347. F. E. See Naylor. See also Proudman. 506.. C. See Dukhin. T. B. P. E. 323. 204. 581 Prandtl. N.. See also Bird. 426. W.. 295. D. R. K . 529. A. L. S.. A. H... W. 27. L. See Bowen. See also Papoutsakis. Petersen. R. M. See Russell.. See Brian. V. K..516. See Colton. 324 Esterman. S. M. 133. T. J. 324 Klein. M. 546 Kim. K. Poiseuille. S . 4 Finlayson. L. W. 505. J. J . 267. 515. C.. 546 Noble. See Morse. 388. Sell. H. N. Moody. G. M. Raithby. J. GrattanGuinness. 281.324. 540.. B. 517. 459. E. 300. 392. R. A. 528. 362. 11.532. B . I . G. 229. K . 430 Morgan. Notter. 27 Rosenhead. 151. 315. 581 Heinbockel. 226. 336. Hay. 404. F. L. See Dukhin. See Vasaru. 542. 501. See Colton. J. Metais. H. 324 Quinn. 417. H. 363. 228. 14. N. 25. 485. K. L. Lin. 413. 429. 308. 26... 387. Prager. J. See Reid. Mahajan. B . Schlicting. 25.. 545 Eckert. See Ash. L. R. S.. 0. J. See Kim. G. 26.. 546. 384. 501. 194. J. 430 Russel.. Jr... See Borg. B. J. D. S. B. E. C . M.. R... 167.. S. E.. F. B. Hatton. See Johnson. 404 Pearson. See Young. 281. T. See Shah. C. M. C. See Durst. Juhasz. A. C . 430 Morse. W. 13.. E. A. 94. 528. 363. Maxwell. Koury. 99. S . 324 Sammakia. H. 248. A. 581 Johnson. F. J.. 533. G.. 404 Sender.. 506 Papoutsakis.. I. See Atkinson. 0. 546. T. Levich. J. 500. L. N. Kodera. See Sideman. Hildebrand. L. Karnke. 505. 121 Lightfoot. M. I.R. O. H . S.. 504. See Hatton. 117 Naylor. 207 Howarth. 470 Durst. L.. G. 515. 2. Diqnes. R. See Metais. Raithby. S. J. H. E. 316. Fick. K . W. See Gebhart. 324. A. Fodor. 27 Kott. 27 Gnielinski. Fourier. C.. M. See Kim. H. A. See Flik.. 503. 194 Gregg. 505.. 536. W. J. A. 539. 505 Leal. E. 513. K . 364367. 178. J . S. 310.. 367 Derjaguin. L. M. E. See Garner. J. M. R. Marrero. J. R. 323. 392. A. Pearson. 363 Howell. 281. H . J. A. See Beris. V. 360. C . Glazov. 369.510 Gillispie. See Rohsenow. J. R. 313. 302. 336. P. S. W. J . B. 248. See Bender. 324 Karrila. See Sideman. 350352. 417. See also Dahler.Author Index Denn. See Gebhart. 545 Goddard. 282. C. 308.446. See Sellars. H. W. See also Lin. 26. 336.. 519. J.. 336 Prater. E. 11. F.. 430. S. H. 323. See Durst. 189. G. 265. A. 194 Greenberg. 63. G. W. 546 Memll. E.. See Vasaru.. J. P. S . L. 313. 26 Hiigge. 546 Hirschfelder. 428. 324. G . F. F. L. 540.. R. mack. M . 533. N. See Mason. 117 Happel.. 281. 20. 490. 127 Kaplun. 430 Gebhart. E. E. C. See also Jones. 117 London. 363 Lim. N . 471 Prausnitz.. 498.. 117.. 148. A. T. R.. A. R. C. M. 375. Einstein. 117 Fljk. R. See Colton.E. Prim. E. 78.. B. G. J. See Israelachvili. 281.431. W. See Sparrow. E. 470 Hales. See Gebhart. 349.... N. Z. 505 Dresner. 27 Jones. 146. S. W . A. 117 Dugas. 281 Piplun. S. B. W. Grew. E. C. E. 167.. P. 309. 505. 431 Lewis. H. R. G . W. 544. 305. 240. R. 528. 195. 315. 27. 546 Rogers. 235.. H. 533. 319. See Jones. 430. 547. 512. H.. B . 308310. Proudman. M. S. V. M.J. 0. F. A.. 18. 27 Reid. 326. Rosensweig. H . J. Moin. Hobson. See Reid. Edwards.. M . 363 Girifalco. Keller. J. 493. 563. J. M. w. 546 Moon. 512. M. E. C.. J.See Touloukian.. 426. 309. 484. 546 Schowalter. 25.. I. 248 Rott. 360... 329 Moffat. 428. 576. 463. J. 539. See Bird.55 1 . A. 471 Nikuradse. Y. H. 324 Ramkrishna. 281.. C. 544. 546 Rogallo. G. G . K. S. D. See Sleicher. 195. 546 Launder. See also Andersen. H. A. S. See Tadmor. Hansen. K . J. A. W. R. P. 27 Jaeger. Jeffrey. Garner. G.. See Morgan. Gogos. N. C . P. Jaluria.. 281 Mihaljan. . 79.. E. 471 Ho. J . 532. 277. 538. 429. 195 Le Fevre. 547 Reid. 512. 543. 444.. I . L. 430 Rallison. 539. See de Groot. 194. 248. D. P. B. 309. J.. H. K. Y. B ..545 Godson. E. 516. 248. R. C. J. J. 548 Plnkus. 357. C. A.. J. Saville. 26 Glagoleva. 305. G. L. B. See Siegel. 363 Feshbach. 404 Newman. 20. D. F.347.484. P. J. E..G.. G. V. 207. 117. C. C. F? G. 525. J. 581 Morgan. S. J. Reinhold. Hsu. M. See Andersen. J. 442. H.. D. 367. J. A. E. 248. 545. T. K. R.. V. Jr.. M.. I. P. B.. See Acrivos. Ibbs. H . E. 68 Olbrich..
Slyozov. T. See Spalart. Slichter. 523 Angular velocity. 497. S.. S. 349 for turbulent pipe flow. T. Skelland. Y S. C. 360 flow separation. 142 as solutions to eigenvalue problems. See Casey. See Colton. W. 334. 305 Bingham fluid. K. B. Tennekes. 350 flow in a planar jet. J . 488 Bernoulli's equation. E. P. L. 371. M . 529. 49. 545. R.. A. R. W.A. See Sellars..540. 249. 375. A. A. 223 Aspect ratio. See Keller. I.T. 507 vertical flat plate. D. 453. 319. P. R. See Chwang. 430 Sleicher. Yang. 485. 195 Weast. 117. Smith. Watrnuff. 398 Brinkman's equation. 99. 363 Tannenbaum. E. 357 matching of velocity with outer or inviscid Aow. B. 375. 428. 290. 451.. V. See Olbrich. 3W393. 518. 46 in selfadjoint eigenvalue problems. 5 6 5 4 2 . See Burton. L. 501. 471 Tani. 237 Carreau model. 100. 82. 325. C. 248. C. Jr. J. I. A. 422 temperature in creeping flow.324 Stroeve. 404. S. M . 143 orthogonality of. 507 energy equation. 82 Blasius equation for boundary layer flow past a flat plate. See Moon. 404 See also Bhatti. G. See Lifshitz. T. 230 Catalyst pellet. S. 177. S. R. Wasan. P. 404 Wrobel. 324. E. A. R. Slernrod. G. O. C. 316. 546 Spalding.I?. G. D. 540. 546 Tildesley. 170. K . 506 Young. 293. 0 . M . 337. V. 506. 471 Touloukian. D. 540. P. S. See Allen. T. 216. 82 modeling simplifications. 313 Arc length. W . Perturbation methods Averages in turbulent flow. E. 425 Bulk concentration. 266. See Free convection Buoyancy forces. A. G. 340 scaling considerations. 101 in laminar forced convection behavior of Nu and Sh.Shab. A. Wilson. M. See Pinkus. N. 506. 434. J. 450. See Spiegel. A. 356 flow past a wedge. D. H. 330 Ziugzda. 441 Tadmor. R. R. D. C. M. 385. 420.. M. G. 248. W . C . 340 momentum equation. M. 229. 329. 501. 446. W. M. 134 Boussinesq approximation.546 Spiegel. 549 Boundaryvalue problem. 63. H. 281. H. 226. 27 Tribus. Spalart. See Sutera. K. 430 Activity coefficient. C. H. 581 Wishnok. S. L. Wu. 419. 560 Analogies among diffusive fluxes. Zukauskas. 175 in rectangular coordinates. 167 Biharmonic equation. Sparrow. S. 282 Taylor. 225 Bulk temperature. G . 429. P.. See Brebbia. Truesdell. See Lewis. E. Turner. 408 Siegel.209.. 43. C. H. 403 Base vectors. 325. 11. C . 414 in laminar free convection behavior of Nu. 418. 41 for linear boundary value problems. 392. R. F? Toor. Slifkin. 483 Capillarity and capillary pressure. 27 Wild. 515 Biidewadt flow. 483 Brinkman number.. Telles. Smith. Stein. See Edwards. N.. See Acrivos. C . 48.F. 302. 122 Alternating unit tensor.. 384. Stokes. M. 232 for heat transfer. C. 506 T h e . K . See Serrin. H . 454 Boltzmann distribution. J. See Scaling. 471 Veronis. 132. E. 496 momentum equation. C. 250 Capillary length. 166 Boundaryintegral methods.. 139 from SturmLiouville problems. I. 144 in spherical coordinates. E. See Abramowitz. L. Y. 52 404 . K. 42. 392. A. I. B. See Timesmoothing. 451 Airy functions. R. 505. 425 energy equation.. J.See Morgan. V . A. L. Shah. 176. 391. 413. J. 83 as a ratio of resistances. E. 496 scaling considerations. E. 95. J... Sideman. G. 210. 329. 5 1% 532.546 Van Dyke. H.. T . 510 WorsgeSchmidt. See Segel. 231 Biot number. Stewart. R. R. W. 335 Bessel functions. See Launder. Tanner. B. 539. 403. 236.. 404 Tanford. 63 Skalak. B. G. 507. 429. 508 Spencer. 375 Bulk (dilatational) viscosity. 388. D. Z. D. 399.. 256. 569 Basis functions for conduction and diffusion problems in cylindrical coordinates. 559. I .. Von K h h .. G. See Falkner. 282. See Lewis. A. Warner.. 506 Stegun. I? R. 546. See Atkinson. See also Notter. R. K. Speziale. R. E. M . 134 for mass transfer. 552. T.. 7 between heat and mass transfer. R. See Colton. 164 Benard problem. 422 flow past a flat plate. A. 546 Sutera. M. M.501.. 503. P. J.. 364 Vasaru. J. Sternlicht. 497 in turbulent flow. M. See Bird. A.. 544. J. 341 flow in the wake of a flat plate. Wooding. S. See Zukauskas. I13 SUBJECT INDEX in fluid mechanics. 371. 27 Tan. 223. 460 Boundary conditions for fast reactions. A. 325. J. 326 Bubbles.. R. 11.. B . Turbulent flow Axial conduction or diffusion effects. 347... Skan. 191 Boundary layers for diffusion with a homogeneous reaction. J. P. 43. W. 168.H. R. 551. Watson. 336. 323.. 11. See Acrivos. See Brebbia. 235.M. 316.. 350.. R. 414. Watson.. 80 Asymptotic analysis.See Bolz. C. A. J.E. C. 282 Taylor. 363. 569 Archimedes' law. S .. 375 Buoyancydriven fldw. See Jones. A. 368 Body forces.. Sureshkumar. 180 linear independence of. Van Driest.
461. 222 turbulent flow. 2 19 for chemical species. 18. 458 Diffusionthermo effect. 44 turbulent flow. 526 Edge effects in modeling conduction and diffusion. 4. 268 Eigenfunctions. names of specific groups for multicomponent systems. 396. 460 Deformation. See Turbulent flow Coefficient of dry friction. 4 Concentration polarization in ultrafiltration. 97. 20. 297 uniform Bow past a sphere. 435 for linear momentum free convection. 455 Drag coefficient or drag force. 309 momentum equation. See also Dimensional analysis. 370. mass transfer. 100. 31. 11 3. 192. 55. 325 cylinder. 236. Heat transfer. 325. 464. 450. 403. 323. representative values. 304. 484 Dielectric permittivity. 463 . 42. 310 Dirac delta. 483 Eckert number. 455.465. 395. 134 Disk diffusion from. 444. 452 dilute solutions. 56. 238. 292 particle motion in. 431 rotating. 74. See also Basis functions. 396 Distributed model. 33. 549 lubrication flow. 35 mixed. 572 Current density. 82 Divergence. 388 Conformal mapping. 452 ChapmanEnskog theory. 308 reciprocal theorem. 39 thermal and mechanical effects.469 for viscous stress in generalized Newtonian fluids. 88. 347 Centrifuge. Simultaneous heat and mass transfer binary. 54.466 Drops. 461 Electrohnetic phenomena. 161 Dyads. 523 for energy thermal effects only. 300 general features. 231. 457 Electrical double layers. 470 Diffusion potential. 135 Coordinate systems cylindrical. 57. 435 Newtonian fluids. 456 Cylinders 1/ ' 1 b j I' 65 Concentration profiles. 61. 219. 444. 395. 518 Dimensionality. 483 general. 397 mixtures. Rateofstrain tensor Density. 232 integral foms. See also Friction factor bubble or droplet. 469 electrolyte solutions. 13 StokesEinstein equation. See Leibniz formulas Diffusion of chemical species. See Energy flux. 523 for enthalpy. 154 in modeling viscous flows. on surfaces. 407 Dissipation 545 of energy. 43.Cauchy momentum equation. 15 mass or molar. 173 in creeping flow. 34. 353 in irrotational flow. 12.425 Dufour flux.451. 119 as a ratio of velocities. See Fourier series Eigenvalue problems. 485 generalized Newtonian fluids. 215. 235 Contact resistance in heat transfer. 44. 52. 316. See Fluid. See Material derivative Coordinate surfaces. 378. See also Mass transfer general. 243. 4 thermal. 470 randomwalk models. velocity field. 437 Conduction. 52 Darcy's law. 442. See also Dufour flux Diffusivity binary. 13 dependence on temperature. 265. Dissipation function. 448. 104 stoichiometric coefficients. 347 Conservation equations for angular momentum. 249. 48. 575 Couette flow. 316. Free convection forced versus free. 25 multicomponent.438. 96. 464 Creeping flow axisymmetric extensional flow past a sphere. 456 Differentiation of integrals. 446. 543 for vorticity. 448. 521 for turbulent kinetic energy. 37 turbulent flow. Finite Fourier transform method Electrical conductivity. 170. 398. 180. 77. 336. 326 Debye length. 564 Double layers. 19. 571 Divergence theorem. 53. 512 flat plate. 37 Continuum approximation. 179. 101 Condensation. See also Diffusivity. 335 Constitutive equations basic forms of. 139. 8 kinetic theory for gases. 302 Curl. 522 D' Alembert's paradox. See Basis functions Eigenfunction expansions. 127. 512 use of dynamic pressure in calculation. 492. 17 Characteristic lengths. 81. 63. 451 hindered. 396. 305 integral representations. 258 Coupled fluxes. 493. 100. 434 turbulent flow. 452 reference frames. 19 potential Aow. 12 Dipole contributions. 452 for energy. 447. 580 Dimensional analysis. 192 tracer. 53. 367 Dispersivity. 31 Control volume. 435. See Vectors and tensors Dynamic pressure. 22 Control surface. 435. 435 for entropy. 76. 39 for internal energy. 78. 44. 222 CauchyRiemann equations. 47 Convection. 455 Chemical potential. 543. 46 quasisteadystate approximation in kinetics. 356. SO Dimensionless groups. 398. Contact angle. 247. 572 spherical. 468 mutual (or gradient). 413 rotating. 444 conduction or diffusion in. 446. 448. 508 mixtures. Thermal conductivity Coneandplate viscometer. 65 self. 311 pointforce solutions. pressure diffusion in.450. 526 effects of concentration. 69 Colburn equation.441. 89. 21 eddy. 9. 435. 306 flow near a comer. 45. 52. 442. definition. 294. 45. 323 heat transfer from. 344 Damkbhler number. 230 for viscous stress in Newtonian fluids. 464 general. 451. 292 general solutions for the stream function. 16 lattice model for solids. 454 effects of thermodynamic nonideality. 75. 56 Dirichlet boundary condition. 59 Directional solidification. 398 Eddy diffusivity. 74 Characteristic times. 524 interfacial species balance with. 505 Convective derivative. 31 Convection boundary conditions. 518. 569 rectangular. 341. Mass transfer. 455. 238 Driving forces for energy and mass transfer. 336 solid sphere. 469 pressure. 56 1. 83. 19 in a suspension. 267. 12 pseudobinary.467 Duharnel superposition. 76. 112. 460. 561. rate of. 449 homogeneous. 295 squeeze flow. 262 Charge density. electrical. 573 nonorthogonal. 328 Confluent hypergeometric functions. 16. 517 Concentration. See also Forced convection. 219. 173. 47 Closure problem. 465 Chemical reactions heterogeneous. 50. 428 in high Reynolds number flow. 450. 342 mean curvature of surface. 228 pure fluids. 5 binary. 20 for momentum. 42 Continuity equation. 579 orthogonal curvilinear. dependence on temperature and concentration. 34. 277 molecular. 8 multicomponent. 31 1 rotating sphere. 349. 450. 434 for kinetic energy. 437. 313. for mass. 2 for binary mixtures with coupled fluxes.
314 Jets. See Dynamic pressure Ergodic hypothesis. 136. 360 Ke model. 570 Graetz functions. 378 Fourier integral. Sherwood number dimensional analysis. 306 FalknerSkan equation. 158 at interfaces. 333 relationship to irrotational flow. 130. See also Nusselt number. 434 Inviscid flow. 7. Mass transfer. 162 square rod. See also Convection. 428 Flow work. 487 momentum and energy equations. 552 Gibbs phenomenon with Fourier series. Thermal conductivity Heat flux. 41 viscous. 112 for diffusion with an nthorder reaction. I 33 rectangular coordinates edge effects in steady conduction. 154 flow in a rectangular channel. 490 Fourier series. 341. 158 transient diffusion approaching a steady state. 454. 2 Ensemble. 226. 39 at interfaces. 386 . 386 transient conduction in an electrically heated wire. 3. 394 Fluid definition. 513. 507 effects on Nu. See also Basis functions. 382 fully developed region of a tube with specified wall flux. 147 differentiation and integration of. 347. 209. 191. 152 steady conduction in three dimensions. 208 kinematics. 545 K h m h integral equation. See Chemical reactions Hotwire anemometry. 384. 438 Heat sources electrical. 5 15 Fully developed flow. 173 temperature for laminar flow in a tube. 503. 469 Identity tensor. 412. 213 rigidbody rotation. Diffusion of chemical species. 256 Hamel flow. 435. 497 turbulent pipe flow. 151 sine and cosine series (examples). 503 effects on boundary layer thrcknesses. 460 Energy conservation. 200. 210 statics. 179 transient diffusion in a sphere with a firstorder reaction. 93 Evaporation. 534 Heat transfer coefficient. 336 general features. 181. Electrostatics Isotropic materials. 363 Homogeneity of differential equations. Fourier series cylindrical coordinates pulsatile flow in a tube. 75. 485. See Chemical reactions Homogeneous chemical reactions. 458 Electrostatics. 366 Heat capacity. method of. 448 Film theory. 296. 220 Fick's law. Heat transfer. 494. 43. 152 suspension. See also Basis functions. 184 transient. 108 hollow cylinder. 91. 533 Hydrodynamic stability. 311 Ferrofluids. 114. 35. 464 Equation of state. 142 Integral approximation method. 342 past wedges. 393. with no steady state. 456 Gegenbauer polynomials. 438 Extensional flow. 455. 483 dimensional analysis. 487. 252 Gamma function. Heat transfer. See Dissipation of energy Energy flux. 136 Heterogeneous chemical reactions. 208 rate of deformation. See also Heat transfer. 135. 397. 170 fin. 264. 24 Enthalpy. 381 Grashof number and modified Grashof number. Electroneutrality.Electroneutrality. 2 Free convection. 396. 190. 151 Gradient. 83. 35 1 Falling f i l m .N i k u r a d s e equation. 222 Fluxes. 267. 199 Equivalent pressure. 454 Faxen's laws. 180 Flat plate laminar flow. 230 GibbsDuhem equation. 24 Error function. 335 Irrotational flow. 1 73 rectangular solid. 58. 212. See Current density. 341.257 Faraday's constant. Thermal conductivity in multicomponent systems. 486 flow past a vertical flat plate. 517 Entropy. 151 Gibbs phenomenon. See Stagnantfilm models Fin approximation. 238. 353 for thermal entrance length. See Energy flux Heat of reaction. 497 stability of a layer of fluid heated from below. 447. 495 Boussinesq approximation. 303 Equilibrium partition (or segregation) coefficient. 146 convergence properties. 456. 430 Gauge functions. See Conservation equations Energy dissipation. general properties of. 170 fundamentals. 507 Gravitational force. 148151 Fourier's law. 246. 356. 156 transient diffusion with a timedependent boundary condition. 515 1 j. 162 transient conduction with no steady state. 188 Incompressible fluid approximation. Simultaneous heat and mass transfer conduction edge effects in steady conduction. 512 Ideal gas. 508 Integral representations. / 1 t I 1 j Gauss' law. 267. 216 Green's functions for diffusion. 396 Heat transfer. 388 fully developed region of a tube with specified wall temperature. 517 velocity. 96. 83. 14. 496 Free surfaces. 237. See Energy flux. 512. 493 examples flow in a vertical channel. 414 laminar free convection stability of a layer of fluid heated from below. Nusselt number boundary layers. 466 Gibbs equation. 487 vertical channel. 357 turbulent flow. 47. 38. 305 Generalized Newtonian fluids.435 Entrance length thermal or concentration. 38. 5. 160 spherical coordinates heat conduction in a suspension of spheres. See also Convection. 108 Finite Fourier transform method. 384 for velocity entrance length. 386 Graetr problem. 333 between intersecting planes. 483 Equidimensional equations. 36 Forced convection. 262 steady conduction in a hollow cylinder. 115 KhhPohlhausen method. 183. 353 K h b . 436 radiant. 440. 7 Heat conduction. 265 Integral momentum equation. 221. 181 steady diffusion in a sphere with a firstorder reaction. 48. 136. 42. 205 Green's identities. 385. See also Nusselt number HeleShaw flow. 397. 310 Intermolecular forces. 565 HadamardRybczynski equation. 64. 437. 326 HagenPoiseuille equation. 260 Friction factor. 380. 397 Inner product. 486 vertical flat plate. 221. 17 Internal energy. 431 Gas absorption. 352. 346 Ions. Finite Fourier transform method calculation of spectral coefficients. 95 sphere in creeping flow. 210 absence of drag. 173 steady diffusion with a firstorder reaction in a cylinder. 353. 4 1 laminar forced convection entrance region of a tube with specified wall temperature. 465 Gibbs notation (for vectors and tensors). 43 in single component systems. Nusselt number. 560 Images. 268 steady conduction in a square md. 261. 549 Flow separation. 77. 212. 561. 497. 154 electrically heated wire. 265. 82. 43. 345 past a cylinder.
25 Kronecker delta.KhhPohlhausen method. 270 in a permeable tube. Creeping flow. 256 instability of. 461 entrance region. 265 film on an inclined surface. 147 Numerical methods. 17 E v Q u e approximation. 566 LennardJones potential. 1 15 liquid film with a reversible homogeneous reaction. 225 Newton's second law. 386 effects on Nu and Sh. 274 for a slider bearing. Turbulent flow dimensionless forms. 135 uniqueness proof. 16 Moments of a concentration distribution. 178 Leibniz formulas for differentiation of integrals. 86. 390 tube. 261 parallelplate channel. 217 Nikuradse equation. 4 Mass conservation. 464 NonNewtonian fluids. 275 geometric and dynamic requirements. 38. 95 Orthogonality of functions. 278 for a sliding cylinder. 271 momentum equation. 214. See also Laminar flow. 65 Mixed convection. 399 Momentum conservation. 519 Knudsen flow. 509 turbulent flow in pipes. 395 Melting. 143 Oseen approximation. 228. 76 liquid film with an nthorder homogeneous reaction. 92. 179. See also Boundary layers.See Dimensional analysis Nonequilibrium thermodynamics. 59 Massaverage velocity. 143 of vectors. 417 Peltier effect. 160 ultracentrifuge (pressure effects). 539 Ohm's law. 268 circular tube. See Perturbation methods Material control volume. Simultaneous heat and mass transfer convection and diffusion. 97. 451 sphere with a firstorder reaction. 189 cylinder with a fast homogeneous reaction. 191 Nusselt number. 374. 274 permeability coefficient (solute). 72. 420 submerged objects at low PCclet number. entrance region. fully developed region. 130 continuous point source. 503. See also Conservation equations Kinetic theory of gases. 448 in a ternary gas mixture with a heterogeneous reaction. 457 Onsager reciprocal relations. 487 Linearity of differential equations. 82. 414 effects of boundary conditions and conduit shape. pulsatile or after a step change in pressure. 361. 4 Mass transfer. 376 pseudosteady approximation. 43 Membranes and membrane separation processes. 572 LaserDoppler velwimetry. See Conservation equations Mass flux. 454 Neumann boundary condition. 507 vertical flat plate. 417. 323 Oseen tensor. 262 circular tube. 457. 8 Kinematics. 384. See Surface tension Markoff integral equation. 376 filtration using hollow fibers. 308 Partition coefficient. 47. 225 Material point. Shewood number. 156 nonideal solution. 53 1 Nondimensionalization. transient. 38 Macroscopic balances. 244247 NernstPlanck equation. 358. See also Sherwood number Matched asymptotic expansions. 19 Molecular length and time scales. 33. Bulk temperature Mixing length. See also Conservation equations Marangoni flows. 87 Mach number. 4 Molecular diameter (or radius). 127. 133 Linear stability analysis. 379. See Equilibrium partition (or segregation) coefficient Peclet number. I 0 0 cylinder with a firstorder reaction. 383 Lewis number. 384 tube. 15 Kolmogorov scales in turbulence. stagnant film model concentration polarization in ultrafiltration. 448. 4 Mole fraction. 425 for thermal or concentration boundary layers. 458 binary gas with a heterogeneous reaction. 56 diffusion binary electrolyte. 54. See also Diffusion of chernical species. See Free convection NavierStokes equation. 279 general. 272 Lumped models. 387. 133 Lubrication flow drag and lift forces. 426 hollowfiber dialyzer. 375 effect on axial dispersion of solutes. See Bulk concentration. 180 ternary gas with a heterogeneous reaction. 16. 376 1 I I I I t Mass transfer coefficient. See also Porous media dialysis using hollow fibers. 5 Multicomponent diffusion in gases (StefanMaxwell equations). 68 1I . as a function of position. Lubrication flow channel with a rectangular crosssection. 444. 279 in a tapered channel. 452 at interfaces. 403 effect on the thermal entrance length. 62 Mechanical energy equation. 318. 76 liquid film with an irreversible homogeneous reaction. 562. 505 Mixingcup quantities. 186188 liquid film with a heterogeneous reaction. 390 laminar free convection scaling laws for boundary layers. 58. 528 Modified Bessel functions. 443. 74 Order symbols.441 Linear boundaryvalue problems. 537. 209 Mean curvature of a surface. 398 gasliquid interface in a stirred reactor. 23 Molecular velocity. 580 Mean free path. 177 Molaraverage velocity. 264 near a plate suddenly set in motion. 562. 66 transient diffusion. 469 in liquids or solids (generalized StefanMaxwell equations). 44. 228 Norm of a function. 208 Kinetic energy. 46 laminar forced convection axial dispersion in tube flow. 116. 3 10 Natural convection. 277. 92 ultrafiltration. 398. 377 dimensional analysis. 449 with timedependent boundary concentrations. 258 electroosmosis. twodimensional. 380 scaling laws for boundary layers. 469 relationship between StefanMaxwell equations and Fick's law. 65 directional solidification of a binary alloy. 37. 173 instantaneous point sources. 254 concentric rotating cylinders (Couette flow). 89 reflection coefficient. 8. 39CL392 qualitative behavior for confined flows. 16 Meansquare displacement. 495 laminar forced convection creeping flow past a sphere. Species flux Mass fraction. 334. 253 Laplace's equation. 503. 257 immiscible fluids in a parallelplate channel. See Conservation equations Momentum flux. 555. 86. 53. 50 binary gas with unsteady evaporation. 5 1 3. 449 Multipole expansion. 89. 388 L a m i n a r flow. 384. 555 Kummer's function. 437 Law of the wall Legendre polynomials. 7. 464 Orderofmagnitude equality. Free convection. 412. 569 Orthonormal functions. 371 in approximations for confined flows. 504. 1 73 Modified pressure. 572 Material objectivity. 135 Laplacian. 214 Material derivative. 134 Newtonian fluids. 517. 32. 532 Latent heat. 353 Kinematic viscosity. See also Diffusion of chemical species. 390. 214 Newton's third law. 191. 388. steady. 114. 78 membrane. See Dynamic pressure Modified spherical Bessel functions. 413 tube.
413 qualitative behavior for confined flows. 246. 422 in laminar freeconvection boundary layers. 209. 316 Stokes' paradox. 253. 247 in approximations to the NavierStokes equation. 13 Selfadjoint eigenvalue problems. 76. 443. 94. 299 S tefanMaxwell equations. 425 from kinetic theory for gases. 74 dynamic versus geometric. 209 Stream function. 215 Rheology. 235 in inviscid flow. 436 examples Dufour flux in a binary gas mixture.441. 512 in the lubrication approximation.91 flow near a plate suddenly set in motion. 221 wales. 145 Pipe flow. 81 Shear viscosity. 529 Reynolds analogy. 245 effect on momentum boundary layer thickness. 308 Pointsource solutions in diffusion. 414 temperature in the thermal entrance region of a tube. 347 flow past a wedge. 311 Reichardt eddy diffusivity model. 100 limitations of the fin approximation. 509 Position vectors. 420 effects on Sh. 537. 566 Potential energy. 245. 65 Stagnation flow. 357 free convection past a vertical flat plate. 341. 210 of a fluid above a solid surface. 437 . 344. 501. 246 in static fluids. 43 Rateofstrain tensor. 130 Simultaneous heat and mass transfer. 95 singular perturbation analysis asymptotic matching. 94 order symbols and gauge functions. See Multicomponent diffusion Stirred tanks. 297 Stability. 8. 316 heat or mass transfer from. 557 Perturbation methods. 302. See also Inviscid flow. 276 mechanical properties. 528 Prandtl number. 247 SturmLiouville theory. 418. 216 Strouhal number. 332. 378 StefanBoltzmann constant. 383 transient diffusion in a membrane. 308 Streaklines. 483 forces. 27 1 Reynolds stress. 429 mean curvature of. 241 Stress equilibrium. 292 in diffusion. 19 Stokes' equation. 224. 487. 89 in pulsatile tube flow. 114. 304. 56. 503. 417. 5. 426 StokesEinstein equation. 8 effects on boundary layer thicknesses. 244. 425. 539 from kinetic theory for gases. 578 Reciprocal theorem (Lorentz). 311. 441 thermal diffusion in a binary gas mixture. 264 Quasisteady approximation. 237 solution of an algebraic equation. 306. See Viscosity Sherwood number. 392 Slider bearing. 78 elimination of large or small parameters. 567 Species equilibrium at interfaces. 238. pipe flow. 360 flow in the wake of a flat plate. 164 Selfsimilar solutions. 501. 245. 92 transient diffusion with timedependent boundary concentrations. 226 Pressure diffusion. 413. 304. 323 Stokes' theorem. 338 effect on the velocity entrance length. 566 Stokeslet. 217 Stress tensor. 245. species 4. 419. 186189 Poiseuille flow. See Similarity method Separation of variables. 10. 99 Phase change. 79. 145. 134 Scaled variables. 452 Pseudosteady approximation in chemical kinetics. 16 representative values. 507 Schmidt number. 6. 104 relationshp to laminar boundary layer theory. 253 at inteifaces. 7 Rigidbody rotation. See also Fourier series Speed of sound. 364 Stagnation point. 527. 353 Stanton number. 497 temperature for creeping flow past a sphere (Pe I). 244. 264. 335 in lubrication flow. boundary conditions. 413 turbulent flow in pipes Sieving coefficient. 115. 539 for laminarturbulent transitions. 380 scaling laws for boundary layers. 230 Rayleigh number. 43 Stefan equation. See also Thermal diffusion Space curves. 74 Scales. 222 thermodynamic compared to mean normal stress. 428. 231. See Material deriva evaporation of a water droplet. 425. 58. 313. 522 Reynolds transport theorem. 75 identification for unknown functions. 218 symmetry of. 368 surface of a liquid in. 223. 350 flow in a planar jet. 515 Prandtl mixing length. 261 flow past a flat plate. 342. 239 in creeping flow. 184 in creeping flow. See Tubes Pointforce solutions in creeping flow. 77 in laminar forced convection i momentum boundary layers. 380 Permutation symbol. 8 effect on boundary layer thickness. 101 coordinate stretching. 254 Power requirement. 241 relationship to vorticity. 536 Reynolds number. 104 in creeping flow. 11 Pressure in boundary layer flow. 219 Stress vector. 336 thermal or concentration boundary layers. 211 magnitude. 339 i n creeping flow. 507. 334. See Chemical reactions Reciprocal bases. 16 representative values. 96 interface shape near a wetted wall. See Pseudosteady approximation Radiation. 74 in conduction or diffusion problems. 425 submerged objects at low Pkclet number. 306. 353. 335. 65 Similarity method. 446 Simultaneously developing velocity and temperature profiles. 108 quasisteadystateapproximation in chemical kinetics. 344. 47 Speciescal flux. 495 Reactions. 242 from the trajectory method. 238 in fully developed flow. 264. 56 Soret effect. 456 Porous media. 18 Spheres conduction or diffusion in.Penetration depth. 127 unsteadystate evaporation. 270. 176 Squeeze flow. 247 in approximations to the NavierStokes equation. 445 interfacial energy balance. 292 Stokes flow. 334 relationship to flow rate. 338. 377 dimensional analysis. 265 effects on Nu and Sh. 307. 467 Piecewise continuous functions. 77. See Creeping flow Stokes' law. Irrotational flow Powerlaw fluid. 414. 278 Solidification. 88. 77 Scaling. 260 Robin boundary condition. 460 Poisson's equation. 580 Spherical Bessel functions. 294. 133. 380 laminar forced convection creeping Aow past a sphere. 85. 43 Phenomenological coefficients. 63. 293 dynamic. 86 in the estimation of derivatives. 295. 437 nonisothennal reaction in a porous catalyst. 507 effects on Nu. 340 solution of an algebraic equation. 233. 43. 254 Poiseuille's law. 95 regular perturbation analysis electrically heated wire with variable properties. 334 in inviscid and irrotational flow. 326. 102 corrections to Stokes' law. 179181. 395 Potential flow. 491. See also Finite Fourier transform method Series resistances. 242 Streamlines. 316 diffusion in a cylinder with a fast homogeneous reaction. 497. See also Diffusion of chemiSpectral coefficients. 293 in irrotational Aow. 512 Stagnantfilm models. 75 identification for known functions. 255 PoissonBoltzrnann equation. 164 Substantial tive derivative. 352 from the stream function.
512 velocity measurements. 536 reaction rates. 4 4 in mass transfer. 18 kinetic theory for gases. 219. 210 Wakes. 21 1 transport of. 396 Viscous stress tensor. 184 Thermal diffusion. 540 Prandtl number for. 487. 519 transition to. 398. 388. 373. 513 intensity. 420. 335 vector. 342. 537 Tensors. 532 mixing length. 534. 36 in heat transfer. 96. 471 Vectors and tensors addition and scalar multiplication. 262 permeable. 235. 426 Suspensions effective diffusivity in. 542 direct numerical simulations. 470 in a binary gas mixture. 42 Thermocapillary motion. 258. 200 kinetic theory for gases. 265. 3.Surface area. 58 I products of. See Entrance length. 514 examples. 450. 554 representation. 536 Reynolds stress. 541 scales. 525. 236 Whittaker functions. 77. 386. 518. 11 of a suspension. 7. 264. 16. flow in. 313. 445 Thermal diffusivity. 543 laminar sublayer. 318 Wire. 328 Trace (of a tensor). laminar to turbulent flow. 569 Van Driest eddy diffusivity. 225 dependence on temperature. 528 models. 567 Temperature. classification of. 75. 391. 11 of a suspension. 329 wetting. 5 12. 225 Transition. 518. 520 fluxes. 534 velocity near a wall. 236 I . 570 Surface curvature. 233 dependence on temperature and surfactant concentration. 274 turbulent flow in. 545 enhanced crossstream transport. 428. 527 Wedge flows. 254. 517. 50. 97 of a fibrous material. 540. 444. differential elements of. 235. 3 for anisotropic materials. 295. 323. 27 species. 563. 527. 426 effect on interfacial stress balance. 545 energy dissipation. 352. 127 Whitehead's paradox. 357 Wall variables. 346. 326. 23 1 symmetry of. 560 integral transformations. 528 timesmoothed equations. 512 Transpose (of a tensor). 7. 532 Nu for Prl. 382. 500. 235. specific types ofJaw mixture averages. tensor. 532 velocity and thermal entrance lengths. 529 law of the wall. 221 Volume. 234. 530 fluctuations. See also Entrance length. 393. 5 14. 529 Van't Hoff 's equation. 517 Turbulent flow closure problem. 513. 328 bulk (dilatational). 231 young~aplace equation. 334 Velocity profiles 266. 211 Velocity potential. 342. 235 thennocapillary motion. 578 Velocity. 520 Torques. 18. 554 differential operators. 536. tube flow complete velocity profile. 170 Yield stress. 513. 501. 27 Volume fraction. 249. 398. 25 1 Unit normal vectors. 235 Surfactants. 2 11 in rigidbdy rotation. 9. 16 representative values. 526. 4. 260. 528 intennittency. 221 magnitude. 229 Viscometer. 7. 330 Thermal conductivity. 533 wall variables. 577 Unit tangent vectors. 537 laminar flow in. 43 Viscoelasticity. 543. 407 Taylor series (using position vectors). 380. 545 kinetic energy. 540 eddy diffusivities. 577. 524 Reynolds analogy. 27 in creeping flow. 235 Surface viscosity. electrically heated. 523. 5 12 KE model. 48 Taylor dispersion. 386. 192 effective thermal conductivity in. 16. See Vectors and tensors Terminal velocity. 544. 293 in irrotational flow. 10. 524 fundamental characteristics.428. 4 Velocity gradient tensor. 420. 329 Timesmoothing. differential elements of. 3 13 Viscous dissipation function. 184 effective viscosity of. 376. 16 representative values. 527 Unidirectional flow. 443. 31. 8 Thermal equilibrium at interfaces. 393. 341. 567. Heat transfer Temperature profiles. 553 surface normal and tangent. 350 Wetting. 216 Surface gradient. 522. 10. 513. 3 dependence on temperature. 313 Symmetry conditions general. 579 Surface tension. 225 definition. 532. 48. 305. 532534 View factors. 553 Tubes heat or mass transfer in. 570 volumeaverage velocity. 537 temperature profiles. 530. 23 1. in a simple shear flow. 234. 35. 580 Surface forces. 264.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.