Professional Documents
Culture Documents
BY
DAVID L. COTRELL
THESIS
Urbana, Illinois
Abstract
Flow between a circular cylinder and an inner shaft (with or without topography)
driven by an axial pressure gradient or boundary rotation is considered computationally.
For spiral Poiseuille flow (SPF) where the inner shaft is a cylinder, complete linear
stability boundaries are found for a wide range of parameters. This analysis extends the
previous ranges of Reynolds number (Re) studied, and accounts for arbitrary disturbances
of infinitesimal amplitude over the entire range of Re for which SPF is stable. For counter-
rotating cylinders or the ratio of the angular velocity of the outer cylinder to that of the
inner cylinder does not exceed the square of the ratio of the radius of the inner cylinder to
flow becomes linearly unstable. For the remaining co-rotating cases, it is shown there is no
instability for Re less than some minimum value at which there is a turning point or vertical
asymptote. For some ratios of the radii and angular velocities, disconnected neutral curves
lead to a multi-valued stability boundary (i.e., multiple critical Taylor numbers for fixed
Re). It is also shown that the smallest radius ratio for which annular Poiseuille flow is
linearly unstable is about 0.12, below which no Tollmien-Schlichting-like instability occurs.
Computations are also reported for steady flows driven either by a rotating shaft with
axisymmetric axially-periodic radius variation, or by a coaxial screw. For the axisymmetric
case and a fixed value of the mean radius to the outer cylinder radius, the flow is computed
as a function of Taylor number for several dimensionless amplitudes of the radius variation.
For small amplitude and sufficiently small Taylor numbers, only a "modified" Couette flow
is found, with the axial and radial velocity components much smaller than the azimuthal
component. As the amplitude of the radius modulation increases, solution structure and
iii
bifurcation behavior are significantly affected. Beyond a critical modulation amplitude,
only a single solution branch is found.
For flow between a fixed outer circular cylinder and a coaxial rotating screw with a
helical groove of rectangular cross section, solutions of the incompressible Navier-Stokes
equations that are helically fully-developed (i.e., invariant along a helical coordinate) are
computed by a finite-element technique. In a frame rotating with the screw, the simplest
flow at each Taylor number is steady, and is characterized in terms of a "meridional"
pseudo-streamfunction in a plane passing through the screw axis, and the azimuthal
velocity component, normal to that plane. The dependence of the resulting flow on
Ta and the geometric parameters is presented. At small Ta, the flow in the clearance
between the cylindrical barrel and the screw flight is nearly axial (i.e., has axial and
azimuthal components, a very small radial component except near the mouth of the groove,
and is separated from one or more helical vortices in the groove). As Ta increases, the
meridional flow in the groove intensifies, and penetrates out into the clearance, leading to
significant radial flow in the clearance. For all but the narrowest grooves, a scouring
flow develops at sufficiently high Ta, in which flow from the clearance penetrates to the
hub of the screw. The mean axial flow generated by screw rotation is characterized in
terms of a Reynolds number Re, and its dependence on Ta is discussed for each geometry
considered. The nearly linear variation of the dimensionless torque per unit length with Ta
for all geometries, and the strongly nonlinear variation of Re and Ta for some geometries,
are related to overall features of the flow. The implications of the computations for
understanding radial mixing in screw-driven flows are also discussed.
iv
Dedicated to the two people who got me here:
Norman and Patricia Cotrell,
and the one who saw me through it:
Kelli J. Killen.
v
Acknowledgments
First and foremost I wish to thank my advisor, Professor Arne J. Pearlstein, for
his guidance throughout this research. I have benefited greatly from his knowledge and
experience not only in my research but also in my writing and communication skills.
I wish to thank my lover, Kelli J. Killen, for her love, understanding, and support.
Also, I want to thank my mother Patricia G. Cotrell, father Norman F. Cotrell, and brother
Lonnie D. Cotrell for their unending support and love.
I would like to thank my committee membersProfessors Richard C. Alkire, S.
Balachandar, and Charles L. Tucker, III, and S. P. Vankafor their time, effort, and
suggestions.
I would like to acknowledge my colleagues in 249A MEB, especially Dr. Hanjie Lee,
Dr. Joe Mantle, and Dr. Lifeng Wang with whom I had many great discussions. Also, I
would like to acknowledge assistance of Dr. Sarma L. Rani during preliminary development
Professor Lothar Koschmieder and Anne E. Staples are also acknowledged. This work
was supported in part by NSF Grants CTS-9422770 and CTS-9613241, and DOE Grant
DE-FG02-96ER45607. Some of the computations were performed using the facilities of
the National Center for Supercomputing Applications.
vi
Table of Contents
Page
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviii
Chapter
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 The Connection Between Centrifugal Instability and
Tollmien-Schlichting-like Instability for Spiral Poiseuille Flow . . . . . . . . . . . . . . . . . . . . . 5
2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2. Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3. Numerical approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
vii
3.2. Complete stability boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1. Stability boundary for = 0.77 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.2. Stability boundary for = 0.95 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3. Comparison to experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2.1. Nonrotating outer cylinder. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .65
4.2.2. Counter-rotating cylinders ( < 0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2. Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.3. Numerical solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
viii
5.4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.4.1. = 0 (Taylor-Couette flow) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.4.2. = 0.01 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.4.3. = 0.03 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.4.4. = 0.07 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.4.5. = 0.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.5. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.5.1. General dependence of the flow on . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
9 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
ix
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
x
List of Figures
Figure Page
2.1. For = 0 and = 0.5: (a) critical Ta, (b) critical m, (c) critical k, (d)
critical c versus Re. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
2.2. For = 0.2 and = 0.5: (a) critical Ta, (b) critical m, (c) critical k, (d)
3.1. For = 0 and = 0.77: (a) critical Ta, (b) critical m, (c) critical k, (d)
critical c versus Re. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
3.2. For = 0.2 and = 0.77: (a) critical Ta, (b) critical m, (c) critical k, (d)
critical c versus Re. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
3.3. For = 0.5 and = 0.77: (a) critical Ta, (b) critical m, (c) critical k, (d)
critical c versus Re. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
3.4. For = 0 and = 0.95: (a) critical Ta, (b) critical m, (c) critical k, (d)
critical c versus Re. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
xi
3.5. SPF stability boundaries for = 0.77 and the values of Re (shown adjacent
to each curve) and investigated experimentally by Mavec (1973). The
dashed line corresponds to = 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
3.6a. Nature of transition in -Re plane for = 0.77 data of Mavec
4.1. For = 0 and = 0.1: (a) critical Ta, (b) critical m, (c) critical k, (d)
critical c versus Re. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
4.2. Critical Ta versus Re for = 0 and = 0.1, 0.5, 0.77, and 0.95. . . . . . . . . . . . 169
4.3. Neutral curves for Re = 100, = 0, and = 0.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . .170
4.4. For = 0.25 and = 0.1: (a) critical Ta, (b) critical m, (c) critical k, (d)
critical c versus Re, (e) expanded view of (d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
4.5. For = 0.5 and = 0.1: (a) critical Ta, (b) critical m, (c) critical k, (d)
critical c versus Re, (e) critical c versus Re near the mode transition B, (f)
4.7. For = 1 and = 0.1: (a) critical Ta, (b) critical m, (c) critical k, (d)
critical c versus Re, (e) expanded view of (d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
xii
4.8. For = 0.2 and = 0.1: (a) critical Ta, (b) critical m, (c) critical k, (d)
critical c versus Re, (e) expanded view of (d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
5.1. Dimensionless torque per iunit length, T , versus Ta for = 0 and = 2. . . . 181
5.2. contours on branch 2 for = 0.5, = 0, = 2, and Ta = 120. . . . . . . . . . . . . 182
xiii
5.25. contours on branch 2c for = 0.5, = 0.07, = 4, and Ta = 120. . . . . . . . . . 205
5.26. contours on branch 2c for = 0.5, = 0.07, = 4, and Ta = 150. . . . . . . . . . 206
5.27. contours on branch 3c for = 0.5, = 0.07, = 4, and Ta = 120. . . . . . . . . . 207
5.28. contours on branch 3c for = 0.5, = 0.07, = 4, and Ta = 150. . . . . . . . . . 208
6.2. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.2. (a) Ta = 0,
(b) Ta = 25, (c) Ta = 50, (d) Ta = 100, (e) Ta = 150, (f) Ta = 200.
Positive and negative streamfunction values correspond to solid and dashed
lines, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
6.3. and v contours for = 2, = 0.5, = 0.8, and = 0.2. (a) Ta = 0, (b)
Ta = 200. Positive and negative streamfunction values correspond to solid
and dashed lines, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
6.4. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial
mixing () for = 2, = 0.5, = 0.8, and = 0.2. (a) Re versus Ta, (b)
T versus Ta, (c) versus Ta. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .221
6.5. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.5. (a) Ta = 0,
(b) Ta = 25, (c) Ta = 50, (d) Ta = 100, (e) Ta = 150, (f) Ta = 200.
xiv
6.6. and v contours for = 2, = 0.5, = 0.8, and = 0.5. (a) Ta = 0, (b)
Ta = 200. Positive and negative streamfunction values correspond to solid
and dashed lines, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
6.7. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial
mixing () for = 2, = 0.5, = 0.8, and = 0.5. (a) Re versus Ta, (b)
T versus Ta, (c) versus Ta. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .226
6.8. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.8. (a) Ta = 0, (b)
Ta = 25, (c) Ta = 50, (d) Ta = 100, (e) Ta = 150, (f) Ta = 200.Positive
6.11. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.2. (a) Ta = 0,
(b) Ta = 25, (c) Ta = 50, (d) Ta = 100, (e) Ta = 150, (f) Ta = 200.
Positive and negative streamfunction values correspond to solid and dashed
lines, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
6.12. and v contours for = 2, = 0.5, = 0.5, and = 0.2. (a) Ta = 0, (b)
Ta = 200. Positive and negative streamfunction values correspond to solid
and dashed lines, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
6.13. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial
mixing () for = 2, = 0.5, = 0.5, and = 0.2. (a) Re versus Ta, (b)
T versus Ta, (c) versus Ta. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .236
xv
6.14. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.5. (a) Ta = 0, (b)
Ta = 10, (c) Ta = 15, (d) Ta = 25, (e) Ta = 50, (f) Ta = 100, (g) Ta = 150,
(h) Ta = 200. Positive and negative streamfunction values correspond to
solid and dashed lines, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
6.15. and v contours for = 2, = 0.5, = 0.5, and = 0.5. (a) Ta = 0, (b)
Ta = 200. Positive and negative streamfunction values correspond to solid
and dashed lines, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
6.16. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial
mixing () for = 2, = 0.5, = 0.5, and = 0.5. (a) Re versus Ta, (b)
T versus Ta, (c) versus Ta. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .242
6.17. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.8. (a) Ta = 0, (b)
Ta = 4, (c) Ta = 6, (d) Ta = 25, (e) Ta = 50, (f) Ta = 100, (g) Ta = 150,
6.20. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.2. (a) Ta = 0,
(b) Ta = 25, (c) Ta = 50, (d) Ta = 100, (e) Ta = 150, (f) Ta = 200.
Positive and negative streamfunction values correspond to solid and dashed
lines, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
6.21. and v contours for = 2, = 0.5, = 0.2, and = 0.2. (a) Ta = 0, (b)
Ta = 200. Positive and negative streamfunction values correspond to solid
and dashed lines, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
xvi
6.22. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial
mixing () for = 2, = 0.5, = 0.2, and = 0.2. (a) Re versus Ta, (b)
T versus Ta, (c) versus Ta. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .253
6.23. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.5. (a) Ta = 0,
(b) Ta = 25, (c) Ta = 50, (d) Ta = 100, (e) Ta = 150, (f) Ta = 200.
Positive and negative streamfunction values correspond to solid and dashed
lines, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
6.24. and v contours for = 2, = 0.5, = 0.2, and = 0.5. (a) Ta = 0, (b)
6.28. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial
mixing () for = 2, = 0.5, = 0.2, and = 0.8. (a) Re versus Ta, (b)
T versus Ta, (c) versus Ta. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .263
xvii
List of Tables
Table Page
2.1. Discretization convergence = |T acrit,Mp T acrit,Mp,max | for Re = 100,
= 0, and = 0.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
2.2a. Minimum neutral values of Re (and corresponding values of k and m) for
(in parentheses) for = 0.5 and selected values of and Ta. . . . . . . . . . . . . . . . . 140
3.1. Comparison of experimental and computed values of Ta crit for = 0.
Parentheses denote values of differently defined Taylor numbers read from
figures of other authors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
xviii
5.2. T as a function of the number of elements in the radial and axial directions
for Ta = 300, = 4, and = 0.25; dashes denotes combinations of Nz and
Nr for which Newton iteration would not converge. . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.3. Comparison of computed values of T to the experimental results of Donnelly
xix
Chapter 1
Introduction
Flow between an outer circular cylinder and an inner shaft has been of interest since
the work of Couette in the nineteenth century. Since then, considerable effort has been
devoted to understanding such flows driven by either boundary rotation or an axial pressure
gradient.
When the inner shaft is also a circular cylinder, applications include rotary
fractionation columns (Willingham et al. 1947; Macleod & Matterson 1959), heat transfer
(Gazley 1958; Kaye & Elgar 1958; Becker & Kaye 1962; Kataoka, Doi & Komai 1977),
electrochemical reactors using rotating cylinder electrodes (Gabe 1974; Legrand, Dumargue
& Coeuret 1980; Coeuret & Legrand 1981; Gu & Fahidy 1982; Gabe & Walsh 1983;
Eklund & Simonsson 1988; Gabe et al. 1998; Gao, Scheeline & Pearlstein 2002), membrane
oxygenation and filtration devices (Strong & Carlucci 1976; Moore & Cooney 1995),
and so-called "vortex flow reactors" (Giordano et al. 1998, 2000; Resende et al. 2001).
Other applications include continuous-flow photochemical reactors (Haim & Pismen 1994;
Sczechowski, Koval & Noble 1995), and reactors for certain classes of chemical reactions in
which a nearly uniform residence-time distribution is desired (Kataoka et al. 1975; Cohen
& Marom 1983; Haim & Pismen 1994; Giordano et al. 1998).
When the inner shaft has either an axisymmetric axially-periodic radius variation or
has helical symmetry (e.g., a screw), applications include screw extruders (Griffith 1962;
Kroesser & Middleman 1965; Tung & Laurence 1975; Choo, Neelakantan & Pittman 1980;
Booy 1981; Choo, Hami & Pittman 1981; Elbirli & Lindt 1984; Bruker, Miawl, Hasson
& Balch 1987; Bohme & Broszeit 1997; Broszeit 1997) used in polymer processing, and
viscoseals or labyrinth pumps and seals (Stoff 1980; Rhode et al. 1986). Other applications
include "through-hole" Schwartz et al. (1992) and "blind-hole" plating processes, and
microfluidic applications of static mixers with helical elements (cf. Bertsch et al. 2001).
1
This thesis considers several representative flows of this type. First, linear stability
of spiral Poiseuille flow (SPF) in an annulus driven by an axial pressure gradient and
rotation of an inner circular cylinder is investigated. In chapter 2, complete linear stability
boundaries for the radius ratio Ri /Ro = 0.5 and several values of the rotation
rate ratio o /i are computed, where Ri and Ro are the inner and outer cylinder
radii, respectively, and i and o are the (signed) angular speeds. The analysis extends
the previous ranges of Reynolds number (Re) studied computationally by more than
eightyfold and the range studied experimentally by seventyfold, and accounts for arbitrary
disturbances of infinitesimal amplitude over the entire range of Re for which SPF is stable
for some range of the Taylor number (Ta). For < 2 , we show how the centrifugally-
driven steady or azimuthally-traveling-wave bifurcation of circular Couette flow (Re = 0)
connects to the Tollmien-Schlichting-like instability of nonrotating axial flow at sufficiently
high Re. The transition from centrifugal to shear instability is quite abrupt, with SPF
being unstable at all Ta beyond the value of Re at which nonrotating annular Poiseuille
flow (Ta = 0) becomes linearly unstable. For > 2 , it is shown that there is no instability
for Re below some minimum value at which there is a turning point or vertical asymptote.
In the case of a turning point, a range of Re exists for which there are two critical values
of Ta, with SPF being stable below the lower one and above the upper one, and unstable
in between.
In chapter 3, complete linear stability boundaries are shown for = 0.77 and 0.95, and
extensive comparison is made to experimental data for these and other radius ratios. For
each , there is a wide range of Re and for which the critical Ta is nearly independent
of Re, followed by a precipitous drop to Ta = 0 as seen for = 0.5. Computed critical
values of Ta for all the nonzero-Re experimental data for = 0.77, 0.9497, and 0.959
(covering a range of ), and for most of the = 0 data for = 0.8 and 0.955. For
= 0 and = 0.955, agreement with experimental data from an annulus with aspect ratio
(length divided by gap) greater than 570 is very good (less than 3.2% difference) at each
2
Re up to 325, suggesting that no subcritical instability occurs over a wide range of Re
for = 0 at this . For = 0 and smaller , comparison to experiment (with smaller
aspect ratios) is consistent with subcritical instability at small Re and insufficient length
for development of secondary flow at large Re. For = 0.77 and a range of positive and
negative , agreement with experiment is very good for Re < 135 except at the most
positive or negative (where insufficient development length appears to be significant),
whereas for Re 166 insufficient development length is important for all but the most
positive (where subcritical instability apparently occurs). For = 0.9497 and 0.959,
flow is linearly unstable is about 0.12, so that for = 0.1 there is no Tollmien-Schlichting-
like instability at high Re. In each case, there is a range of Re for which disconnected
neutral curves are found. For each 6= 0, the disconnected neutral curves lead to a multi-
valued stability boundary, corresponding to two disjoint ranges of stable Ta. For each
counter-rotating ( < 0) case considered, there is a finite range of Re for which there
exist three critical values of Ta, with one branch originating in the Re = 0 instability of
Couette flow. For the co-rotating ( = 0.2) case, there are two critical values of Ta for
each Re in an apparently semi-infinite range of Re, with neither branch of the stability
boundary intersecting the Re = 0 axis. Also, there are cases for which the direction of the
disturbance wave propagation switches direction.
In chapter 5, computations are reported for steady flows driven by a rotating shaft
with axisymmetric axially-periodic radius variation. For a fixed value of the mean radius to
the outer cylinder radius, the flow is computed as a function of Taylor number for several
dimensionless amplitudes of the radius variation. For small amplitude and sufficiently
small Taylor numbers, only a "modified" Couette flow is found, with the axial and radial
3
velocity components much smaller than the azimuthal component. As the amplitude of the
radius modulation increases, solution structure and bifurcation behavior are significantly
affected. Beyond a critical modulation amplitude, only a single solution branch is found.
Finally, chapter 6 considers flow between a fixed outer circular cylinder and a
coaxial rotating screw with a helical groove of rectangular cross section, solutions of the
incompressible Navier-Stokes equations that are helically fully-developed (i.e., invariant
along a helical coordinate) are computed by a finite-element technique. In a frame rotating
with the screw, the simplest flow at each Taylor number is steady, and is characterized
(i.e., has axial and azimuthal components, a very small radial component except near the
mouth of the groove, and is separated from one or more helical vortices in the groove).
As Ta increases, the meridional flow in the groove intensifies, and penetrates out into
the clearance, leading to significant radial flow in the clearance. For all but the narrowest
grooves, a scouring flow develops at sufficiently high Ta, in which flow from the clearance
penetrates to the hub of the screw. The mean axial flow generated by screw rotation is
characterized in terms of a Reynolds number Re, and its dependence on Ta is discussed
for each geometry considered. The nearly linear variation of the dimensionless torque per
unit length with Ta for all geometries, and the strongly nonlinear variation of Re and Ta
for some geometries, are related to overall features of the flow. The implications of the
computations for understanding radial mixing in screw-driven flows are also discussed.
4
Chapter 2
The Connection Between Centrifugal Instability and
Tollmien-Schlichting-like Instability for Spiral Poiseuille Flow
2.1. Introduction
The stability of steady, axisymmetric, incompressible flow driven by differential
rotation of coaxial circular cylinders has been investigated extensively since the work
of Taylor (1923). Spiral Poiseuille flow (SPF), driven by cylinder rotation and an axial
pressure gradient, has been of interest since the experimental work of Cornish (1933) and
theoretical work of Goldstein (1937). Papers by Takeuchi & Jankowski (1981) and Ng
& Turner (1982) marked the first appearance of correct theoretical results concerning the
stability of SPF with respect to nonaxisymmetric disturbances, and the beginning of a
more complete understanding of its stability. The best treatments of the stability of this
flow are found in their papers, where previous work is discussed extensively.
For the radius ratio Ri /Ro = 0.5 and rotation rate ratios o /i = 0.5,
0, and 0.2, Takeuchi & Jankowski (1981) investigated the stability of SPF experimentally
for 0 Re 150 and computationally for 0 Re 100, where Ri and Ro are the radii
of the inner and outer cylinders whose respective constant (signed) angular speeds are i
and o , the Reynolds number Re is defined by V Z (Ro Ri ) /, and V Z and are the
mean axial speed of the base flow and the kinematic viscosity, respectively. They found
2
that the critical value of the Taylor number Ta i (Ro Ri ) / is strongly affected
by axial flow and the rotation rate ratio. For = 0 and 0.2, Ta crit initially increases
as Re increases from zero, with the critical disturbance remaining axisymmetric. This
stabilization continues until a (globally) maximum value of Ta crit is reached at some Re.
Beyond the maximum, for = 0, Ta crit decreases but still exceeds its Re = 0 value over the
entire Re range considered by Takeuchi & Jankowski (1981), while for = 0.2, Ta crit falls
below its Re = 0 value. For = 0.5, Ta crit decreases as Re increases from zero (i.e., SPF
is destabilized by sufficiently slow axial flow) with a critical azimuthal wavenumber (m crit )
5
of 1, and continues to decrease until a local minimum is reached. As Re increases further,
the critical Ta increases until mcrit changes to 2, beyond which point Ta crit decreases with
Re until reaching a second (lower) local minimum. For still larger Re, Ta crit increases
monotonically with Re over the range considered by Takeuchi & Jankowski. For = 0
and 0.2, experimental and computational results are in excellent agreement up to Re = 40,
with experimental values of Ta crit at higher Re exceeding computed values to an extent
that increases with Re. For = 0.5, the experimental Ta crit exceeds the computed
value over the entire range of Re. (Use of the symbols Re and Ta in this work to refer
to Reynolds and Taylor numbers defined differently by others implies conversion to the
definitions of this work.) For each , however, the overall characters of the experimental
and computed stability boundaries are qualitatively similar.
In contemporaneous work, Ng & Turner (1982) performed computations for = 0.77
& Turner found that the Ta at which SPF becomes unstable with respect to axisymmetric
disturbances decreases rapidly beyond Re = 6000. Their results agreed well with the = 0
experimental results of Mavec (1973) up to Re = 400 for = 0.77 and Snyder (1962, 1965)
up to Re = 200 for 0.95.
Since the investigations of Takeuchi & Jankowski (1981) and Ng & Turner (1982), work
on SPF has focussed on its use at low Re as an open system for the study of convective
and absolute instabilities and noise amplification and noise-sustained structures (Babcock,
Ahlers & Cannell 1991, 1994; Babcock, Cannell & Ahlers 1992; Lucke & Recktenwald 1993;
Recktenwald, Lucke & Muller 1993; Tsameret & Steinberg 1991a, 1991b, 1994a; Swift,
Babcock & Hohenberg 1994; Tsameret, Goldner & Steinberg 1994), pattern formation
and the structure of supercritical modes (Buhler & Polifke 1990; Raffa & Laure 1993;
6
Tsameret & Steinberg 1994b; Buchel, Lucke, Roth & Schmitz 1996; Wereley & Lueptow
1999; Moser et al. 2000; Moser, Raguin & Georgiadis 2001), and secondary instabilities
(Lueptow, Docter & Min 1992). Closely related filtration flows in which one cylinder
is porous (Min & Lueptow 1994; Kolyshkin & Vaillancourt 1997; Johnson & Lueptow
1997), and flows in which the axial velocity component is time-periodic (Hu & Kelly 1995;
Weisberg, Kevrekidis & Smits 1997; Marques & Lopez 2000) have also attracted interest.
More recently, Meseguer & Marques (2002) reported computations for = 0.5.
Instead of parametrizing the problem in terms of and Ta, those authors used
Besides fundamental interest in the stability and dynamics of SPF, there are important
applications. Chief among these are those involving heat or mass transfer or chemical
reaction. The development of vigorous radial flow is important in a number of systems,
since absent secondary flow resulting from instability, radial transport in circular Couette
axial pressure gradient provides a finite residence time (Gabe 1974; Legrand, Dumargue
& Coeuret 1980; Coeuret & Legrand 1981; Gu & Fahidy 1982; Gabe & Walsh 1983;
Eklund & Simonsson 1988; Gabe et al. 1998; Gao, Scheeline & Pearlstein 2002), membrane
7
oxygenation and filtration devices (Strong & Carlucci 1976; Moore & Cooney 1995), and
so-called "vortex flow reactors," in which considerable mixing occurs within each vortex,
but transport across interfaces between axially propagating Taylor-like vortices occurs
only by diffusion, with mean axial flow providing a controllable finite residence time
(Giordano et al. 1998, 2000). In RCE reactors, mass transfer at the working electrode
can play a dominant role in determining the overall current density (i.e., reaction rate
per unit area), as well as the degree of axial and azimuthal macroscopic uniformity of
the microstructure induced by surface electrochemical processes. Vortex flow reactors are
thought to have considerable potential in systems involving cells and other shear-sensitive
biological components (Resende et al. 2001). Spiral Poiseuille flow and its stability have
also been considered in the context of continuous-flow photochemical reactors (Haim &
Pismen 1994; Sczechowski, Koval & Noble 1995). Axially-propagating Taylor-like vortices
are particularly attractive for certain classes of chemical reactions in which a nearly uniform
residence-time distribution is desired (Kataoka et al. 1975; Cohen & Marom 1983; Haim
& Pismen 1994; Giordano et al. 1998).
Despite considerable interest in SPF and the transitions it undergoes as Ta and Re
increase, the complete linear stability boundary has not been determined for SPF for any
combination of and (or Re ,o and in the parametrization of Meseguer & Marques
(2002)). The only computational work found on the stability of SPF for any combination
of and since the work of Takeuchi & Jankowski and Ng & Turner is that of Recktenwald
et al. (1993), limited to axisymmetric disturbances over the range 0 Re 20, and that
of Meseguer & Marques (2002).
Although Reid (1961) conjectured, in the context of the narrow-gap limit, that the
initial low-Re stabilization of SPF by axial flow must give way to an instability of Tollmien-
Schlichting (TS) type at higher Re, that connection has been made only for = 0 and
= 0.95, and then only for the restricted class of axisymmetric disturbances (Ng &
Turner 1982). (As shown in chapter 3, for these values of and , transition from a
8
centrifugal instability to a TS-like instability occurs at Re = 7716, with nonzero azimuthal
wavenumbers for both modes, in a range (6000 Re 7739.5) in which the computations
of Ng & Turner were restricted to axisymmetric disturbances.)
Also, to date the only = 0.5 results available for > 2 , for which range Synge
(1938) showed that the Re = 0 flow is stable for arbitrary disturbances of infinitesimal
magnitude, are those of Joseph & Munson (1970) (see also Joseph (1976)), and Meseguer
& Marques (2002), both for = 1. As shown in 2.4.2, the computations of Meseguer
& Marques (2002) are fundamentally flawed by failure to consider disturbances with a
sufficiently broad range of azimuthal wavenumbers, and for = 1 disagree with the results
of Joseph & Munson.
Here, complete stability boundaries for SPF are reported for = 0.5, the radius
ratio considered by Takeuchi & Jankowski (1981) and Meseguer & Marques (2002). The
results account for arbitrary infinitesimal disturbances and extend the range of Re studied
computationally by more than eightyfold and experimentally by seventyfold, covering the
entire range of Re for which SPF is linearly stable for some range of Ta. Here = 0 is
considered, in which case the outer cylinder is fixed, as well as > 0 and < 0 (co- and
= 0.5 are presented for = 0 (2.4.1), > 0 (2.4.2), and < 0 (2.4.3). This is followed
by a discussion in 2.5, and some conclusions in 2.6. Results for the larger values of
pertinent to most experiments are reported in chapter 3, where extensive comparison is
made to experimental data for a wide range of Reynolds numbers and rotation rate ratios.
An account has been submitted for publication (Cotrell & Pearlstein 2003).
9
2.2. Formulation
For a constant-property fluid, flow between concentric circular cylinders driven jointly
by a constant axial pressure gradient and rotation of one or both cylinders is considered.
An inertial frame is chosen in which the cylinders have no axial velocity. The governing
equations for the velocity V and pressure P are nondimensionalized using the dimensionless
variables r = R/(Ro Ri ), z = Z/(Ro Ri ), = tVZ /(Ro Ri ), v = V/VZ , and
2
p = P/(VZ ) employed by Takeuchi & Jankowski (1981).
The steady, axisymmetric, axially fully-developed base-flow velocity field (denoted by
subscript b) satisfying no-slip conditions on the rigid inner and outer walls at r = /(1 )
and r = 1/(1 ), respectively, is
vrb = 0 (2.2.1a)
" #
Ta r 2 2 (1 )
vb = + 3 (2.2.1b)
Re (1 2 ) r (1 ) (1 + )
" #
1 r 2 (1 )2 ln (1 2 ) ln [r(1 )]
vzb =2 . (2.2.1c)
1 2 + (1 + 2 ) ln
With this nondimensionalization, the base flow for fixed consists of an axial component
whose profile and magnitude are constant, and an azimuthal component whose profile
depends on and whose magnitude is proportional to Ta/Re.
The disturbance velocity and pressure fields (denoted by 0 ) are written in normal mode
form according to
where k, m, and are the axial wavenumber, (integer) azimuthal wavenumber, and
temporal eigenvalue, respectively, and take k to be real and to be complex. Neglecting
nonlinear terms, one is left with a system of homogeneous linear ordinary differential
10
equations in the radial coordinate for the disturbance velocity components and pressure,
satisfying homogeneous boundary conditions, and equivalent to equations (6a)-(6d) of
Takeuchi & Jankowski (1981).
Note that stability of SPF has been reported in terms of three different
number versus an outer-cylinder Taylor number for selected values of Re. Finally, the
computations of Meseguer & Marques (2002) are presented in terms of a critical inner-
cylinder Taylor number versus Re for selected values of an outer-cylinder Taylor number.
For the parametrization of Takeuchi & Jankowski and Ng & Turner (1982) chosen here,
the profiles of the azimuthal and axial velocity components are independent of Ta and
Re, thus allowing centrifugal and axial shear effects to be assessed independently of profile
changes at fixed , as for the limiting cases of annular Poiseuille flow (no rotation) and
circular Couette flow (no axial pressure variation).
11
in the left half-plane for all m and k. As discussed by Takeuchi & Jankowski (1981), Ng
& Turner (1982), and Meseguer & Marques (2002), the analysis can be restricted to k 0
without loss of generality.
To compute critical values of Ta for given Re, , and , begin with Mk discrete axial
wavenumbers (153 Mk 202) nonuniformly distributed over 0 < k 100. For each
m considered (vide infra), then do "axial wavenumber traverses" in which, for fixed Ta,
the eigenvalues with largest growth rate r are identified at each of the Mk wavenumbers.
For each m, locate two initial values of Ta, one (Ta s ) stable for all such k, and one
(Ta u ) unstable for at least one k. For each Ta s and Ta u , identify the axial wavenumber
ki , among the Mk considered, at which r assumes its maximum value. For each Ta, a
quadratic polynomial is fitted through the points (r , k), with k = ki1 , ki , and ki+1 .
Differentiating, and estimating the axial wavenumbers maximizing r at Ta s and Ta u ,
max max
this process is used to iteratively refine k at each Ta until | 1 r,n+1 /r,n | < 1 . Once
maximum values of r (negative for Ta s and positive for Ta u ) and the corresponding
values of k have been found, secant iteration between Ta s and Ta u is used to estimate a
new value Ta 1 at which rmax vanishes. A wavenumber traverse is performed at this new
Ta, and the sign of the maximum r determines whether Ta s or Ta u is replaced. Secant
iteration is continued until | 1 T al+1 /T al | < 2 . (Tests in which a convergence criterion
was enforced on | 1 T as /T au | showed that the stability boundaries were unchanged.) At
this juncture, for each m there are one or more extremal values of Ta and associated values
of k. For each m, such a (k,Ta) pair corresponds to an extremum on a k-Ta neutral curve
(i.e., a curve dividing portions of the k-Ta plane in which no temporal eigenvalue lies in the
right half-plane (RHP) for a given m, from portions in which one or more eigenvalues lie in
the RHP for that m). The critical azimuthal wavenumber mcrit is determined as follows.
12
lie in the RHP for any combination of m and k at the extremal Ta. The corresponding
critical axial wavenumber and wavespeed are denoted by kcrit and ccrit = i,crit /kcrit .
For nonzero Re, the initial estimate of mcrit is obtained from results at nearby values of
Re.
For > 2 , the procedure is identical, except that mcrit cannot be estimated by
beginning with the value of mcrit for circular Couette flow at Re = 0, which is stable with
respect to all infinitesimal disturbances. Instead, first finding a (positive) Re for which one
or more unstable values of Ta exist for some m. Extremal values of Ta are then computed
for each m in a range centered about m = 0. After mcrit is found for this initial Re, it is
used as the starting point to compute the remainder of the stability boundary as outlined
above.
For each , a range of m was examined (sometimes not identical to the initial range),
sufficiently large that it was clear that the critical Ta varies monotonically for m beyond
the limits of the range. As shown in 2.4, the computed values of mcrit were nonnegative
unless > 2 . Nonetheless, for each < 2 , spot checks were performed for m < 0, and
in no case was a neutral Ta found to lie below the value of Ta crit computed for m 0,
consistent with the results of Takeuchi & Jankowski (1981). The importance of considering
a sufficiently large range of m cannot be overemphasized (see 2.4.2).
For given , an upper limit on the range of kcrit over 0 Re Re AP was estimated
by computing critical values of Ta, m, and k at each of a small number (5-10) of Re values
As shown in 2.4.2, there are values of for which ranges of Re exist in which
there are two or three critical values of Ta. In these cases, closed disconnected neutral
curves (cf. Lopez, Romero & Pearlstein 1990 and references cited therein) exist whose
13
extrema must be found. While systematic techniques, developed to do this for stability
problems in which the constant-coefficient differential equations allow closed-form solution
(Terrones & Pearlstein 1989), can be generalized to systems requiring approximate spatial
discretization, rapid growth of computational complexity with matrix size makes that
approach impractical for the level of resolution required for most nonzero base flows.
Thus, after first recognizing the existence of a range of Re in which multiplicity occurs, the
procedure described above is then used with several pairs of initial Taylor numbers. In each
case, the topology of the computed stability boundary is consistent with the assumption
figures shown by them. For = 0.77, the results agree with those tabulated by Ng &
Turner (1982) over the entire Re range they considered (0 Re 6000), except for the
computed values of mcrit and ccrit at Re = 300 and 500. For these two cases, the critical
azimuthal wavenumbers are smaller by one. The values of T acrit agree to within one part in
50,000, and the values of kcrit differ by less than 0.1%, even at the two values of Re at which
computed values of mcrit are discrepant. For = 0.95, the values of Ta crit agree to at least
three significant figures with those of Ng & Turner (1982) over the range 0 Re 6000.
Agreement is considerably better than that except at Re = 10, 100, and 2000, where the
values of mcrit are one larger, one smaller, and one larger, respectively, and there are small
differences in kcrit . For = 0.95, larger differences over the range 6000 < Re < 7739.5
between the results of Ng & Turner (1982) for axisymmetric disturbances and ours for
arbitrary disturbances occur, and are discussed in chapter 3. Comparison to the low-Re
results of Recktenwald et al. for = 0 is also excellent. For = 0.5, the root-mean-square
(rms) difference between the computed values of Ta crit at Re = 1, 2, . . . , 20 and the values
corresponding to the fitted form of Recktenwald et al. is 1.2 103 , with Ta crit 60 over
14
that range. The rms differences between computed values of kcrit and ccrit are 1.7 105
and 1.6 104 , respectively, compared to values of kcrit and ccrit on the order of 3 and
1, respectively. Also, the results are compared to previous stability analyses for annular
Poiseuille flow, and found excellent agreement between the critical Re values computed here
and the tabulated results of Sadeghi & Higgins (1991) (less than one percent difference for
= 0.5), as well as the graphical results of Mahadevan & Lilley (1977) and Garg (1980)
at smaller , for which agreement was also good.
To ensure that the critical values are fully resolved, convergence tests are performed for
the largest value of Mp used in this case. In this work 1 = 106 , 2 = 2 107 were
used, and achieved convergence with Mp,max 40 for each combination of Re, , and
. The small differences between the results of this work and those of Ng & Turner
(1982) in the few cases cited above are believed to be due to the use of somewhat more
2.4. Results
The complete linear stability boundaries are reported in the Re-Ta plane, accounting
for three-dimensional disturbances, for = 0.5 and each rotation rate ratio considered by
Takeuchi & Jankowski (1981). The results cover the entire range from Re = 0 (the Taylor-
Couette limit) to Re AP (the -independent Re beyond which the base flow is unstable for
every Ta, corresponding to onset of TS-like instability in the nonrotating annular Poiseuille
flow). Over the range of Re investigated by Takeuchi & Jankowski, the computed results
15
2.4.1. Nonrotating outer cylinder
As discussed in 2.1, the maximum Re values considered in the computations and
experiments of Takeuchi & Jankowski, and in the computations of Meseguer & Marques
(2002) were 100, 150, and 125, respectively. Here, a brief review of the results is given for
shown by Takeuchi & Jankowski. The scalloped stability boundary is associated with
the "stair-step" behavior of mcrit , as shown in figure 2.1b and discussed by Takeuchi &
Jankowski. As noted by those authors, shear generally destabilizes SPF for 61 Re 150.
Figure 2.1a shows that this destabilization continues to about Re = 400.
On a plateau over the range 400 < Re < Re = 9916, Ta crit ( 88) is greater than its
nonrotating value of 68.19. The existence of this plateau indicates that the magnitude of
the axial component of the base flow does not significantly affect the centrifugal instability
in this range of Re. At Re , Ta crit begins a precipitous fall, corresponding to a transition
from centrifugal instability to a parallel shear instability (associated with a critical layer) of
TS-type. The flow is unstable for all Ta beyond Re AP = 10359, a value agreeing with the
graphical results of Mahadevan & Lilley (1977) and Garg (1980) for nonrotating annular
Poiseuille flow as well as they can be read.
Figure 2.1b shows that mcrit increases by one unit at each of seven Reynolds numbers
in the range 0 < Re < 270. (The jump of two in mcrit between Re = 40 and 50 shown in
table 1 of Takeuchi & Jankowski is clearly due to the gap between consecutive Re values.)
The critical azimuthal wavenumber drops directly from 7 to 2 as Re passes through Re .
16
monotonically with Re, while for Re 41, kcrit decreases monotonically over each constant-
mcrit range of Re. For Re Re AP , kcrit increases discontinuously each time mcrit
increases, leading to the "wavenumber fan" shown. For 260 Re Re , for which
mcrit = 7, kcrit decreases, nearly inversely with Re, to very small values (kcrit = 0.0422
at Re = 9914). The very small axial wavenumbers are consistent with flow visualization
studies of the secondary flow (Nagib 1972; Joseph 1976). As Re passes through Re , kcrit
increases more than thirtyfold. Between Re and Re AP on the TS-like branch, kcrit varies
between 1.481 and 1.479, reflecting the fact that the TS-like instability is driven by axial
known result for Taylor-Couette instability with = 0. Since mcrit = 0 for small Re,
this corresponds to transition from a steady axisymmetric disturbance flow at Re = 0 to a
time-periodic axisymmetric flow at nonzero Re, as Taylor-like vortices propagate axially.
As with kcrit , discontinuities in ccrit at higher Re correspond to jumps in mcrit . For
0 < Re < 260, ccrit decreases monotonically over each range of Re for which mcrit remains
unchanged. Over 260 < Re < Re , for which mcrit = 7, ccrit 1.86. As Re increases
beyond Re , ccrit discontinuously drops to values characteristic of instability for annular
Poiseuille flow.
to Re AP . Results are also presented for = 0.5, beyond the Rayleigh criterion. Finally,
the results of Meseguer & Marques (2002) are reconsiderd for > 0.
17
The complete stability boundary for = 0.2 is presented in Figure 2.2a. At Re = 0,
Ta crit is found to be 124.7, corresponding to onset of Taylor vortices absent mean axial
flow. The critical Ta increases to a maximum of about 170 near Re = 36. For larger Re,
Ta crit decreases monotonically, first to a broad plateau (Ta crit 79), and then, beginning
relative to the Re = 0 situation. This differs from the = 0 cases discussed above, and
is consistent with the work of Takeuchi & Jankowski, whose results do not extend to the
plateau region. This trend is opposite to that predicted for = 0, where Ta crit was higher
on the high-Re plateau than at Re = 0.
Figure 2.2b shows that the variation of mcrit with Re is qualitatively similar to the
= 0 case, with the largest value of mcrit (on the plateau) being 6 rather than 7.
Figure 2.2c shows that for 30 Re 100, where mcrit has a staircase variation,
kcrit varies piecewise continuously with Re and approximately linearly with log(Re) for
each mcrit , in excellent agreement with the results of Takeuchi & Jankowski. Outside this
range, mcrit = 0 for Re 30, and mcrit = 6 for 100 < Re Re = 9979. Qualitatively,
the variation of kcrit with Re is similar to that found for the = 0 case, as is that of ccrit
(figure 2.2d).
Figures 2.3(a-d) show the stability boundary, mcrit , kcrit , and ccrit for = 0.5. In
this case > 2 , so that according to Synges viscous extension of the Rayleigh criterion
(Synge 1938), no linear instability is possible for Re = 0. In fact, in this case SPF is linearly
stable for Re < Re min = 70.2, as shown in figure 2.3a. To the right of the turning point at
Re min , one branch of the stability boundary corresponds to a critical Ta increasing rapidly
with Re, and the other to a branch decreasing to a plateau (Ta 70), on which Ta crit
remains nearly constant until falling precipitously at Re = 10049, reaching Ta crit = 0
18
at Re AP = 10359. Thus in a finite range Re min Re Re AP , SPF is linearly stable
below the lower critical Ta and above the upper critical value, and unstable in between.
(Computations on the upper branch are difficult beyond Re = 1000 due to rapid growth
in the number of expansion functions required to maintain adequate resolution.)
The multi-valuedness of the stability boundary shown in figure 2.3a is associated with
existence of disconnected neutral curves in the k-Ta plane at fixed Re > Re min . Consistent
with the work of Synge, for = 0.5 there are no neutral curves if Re < Re min . Figure
2.4 shows that for Re = 100 the neutral curves for 14 m 3 and 3 m 9 are
closed and that each extends over only a finite range of Ta. In fact, at Re min , the first
neutral curve (for m = 6; see figure 2.3b) appears as a point, and grows as Re increases.
For each m, the process is essentially identical to the appearance of closed disconnected
neutral curves in buoyancy-driven stability problems (Pearlstein 1981; Pearlstein, Harris
& Terrones 1989; Terrones & Pearlstein 1989; Lopez, Romero & Pearlstein 1990; Ali &
Weidman 1990), and as in those cases, the result is a stability boundary (figure 2.3a) that
is not single-valued. Between Re = Re min and Re = 100, the other neutral curves shown
in figure 2.4 first appear as points, and grow as Re increases. The neutral curves shown
for Re = 100 in figure 2.4 were determined by a fine-grained grid search for instability in
the k-Ta plane using increments of 0.01 and 10 for the wavenumber and Taylor number,
respectively. At Re = 100, these are the only neutral curves found for 20 m 11 over
the ranges 0.1 k 10 and 0 Ta 1000.
on the upper branch, followed by unit decreases as Re increases. (Vertical line segments
with no filled circles between their endpoints in figures 2.3(b-d) denote jump discontinuities
19
in mcrit , kcrit , and ccrit , and are provided to clarify variation of these quantities along the
stability boundary.) No Re for which an axisymmetric disturbance is critical was found.
Figure 2.3c shows that kcrit is again about 1.48 on the TS-like branch, and suffers a
jump decrease to 0.026 as Re decreases through Re . As Re decreases further on the upper
branch of the stability boundary, kcrit increases roughly like 1/Re (204 kcrit Re 235
for 8.6 Re 9000), until mcrit jumps from 6 to 6, at which point kcrit jumps from
2.41 to 1.53. On the mcrit = 6 portion of the lower branch of the stability boundary,
kcrit increases as the turning point at Re = 70.2 is approached, and continues to increase
on the upper branch to a value of 1.70 at Re = 70.3. Beyond that point, kcrit decreases
as Re increases on the mcrit = 6 portion of the upper branch, reaching a value of 1.69 at
Re = 72.8. At that point, mcrit decreases to 7, and kcrit jumps to 1.97 before decreasing
monotonically to 1.79 at Re = 118.6. As Re increases further, kcrit exhibits "fan-like"
behavior, decreasing monotonically as Re increases for each value of mcrit , and increasing
discontinuously each time mcrit suffers a jump decrease.
On the TS-like branch, ccrit varies between 0.40 and 0.41 (figure 2.3d), and jumps
discontinuously to about 2.50 as Re decreases through Re . It remains nearly constant on
Note that the negative values of ccrit on the upper branch and on part of the lower branch
of the stability boundary correspond to disturbances propagating upstream against the
axial flow, and that ccrit does not vanish at any point on the stability boundary. This
point is discussed in 2.5.6.
The special case = 1, corresponding to a base flow with constant angular velocity,
has been considered by Meseguer & Marques (2002). Their figure 13 shows the critical
Ta approaching 70.69 (with mcrit = 5) at high Re, and the critical Re approaching a
20
vertical asymptote at 96.14 (with mcrit = 6) at high Ta. Unfortunately, the latter result
disagrees with the computations of Joseph & Munson (1970), who showed that for = 0.5,
2Re crit 165 (Joseph 1976, figure 46.1) in the Ta limit.
In fact, the results presented in tables 2.2a and 2.2b show that the asymptotes reported
by Meseguer & Marques (2002) are both incorrect, with the correct horizontal and vertical
asymptotes being near Ta crit = 63.86 (with mcrit = 6) and Re crit = 82.33 (mcrit = 8),
respectively, the latter value being very close to that (2Re crit 165) reported by Joseph
(1976). Tables 2.2a and 2.2b show that the values reported by Meseguer & Marques
(2002) correspond very closely to the smallest Ta (71.13) and Re (96.16) obtained when
only positive values of m are considered, providing convincing evidence that the incorrect
results of Meseguer & Marques (2002) are due to consideration of an insufficient range of
azimuthal wavenumbers.
For = 0.5, the stability boundary partially determined by Takeuchi & Jankowski
(1981) are completed. As shown by those authors, beginning from the Re = 0 onset
at Ta crit = 111.3, axial flow destabilizes SPF with respect to centrifugal instability up
to Re 13. The critical Ta then rises slightly to a local maximum near Re = 21,
before falling and reaching another minimum near Re = 32. Axial flow then stabilizes
SPF, with Ta crit increasing monotonically until about Re = 200. The scalloped nature
of the stability boundary, corresponding to step changes in mcrit , is evident. The critical
Ta ( 124) remains approximately constant for 200 < Re < Re = 9683, beyond which a
Re (discussed in 2.5.5).
Figure 2.5b shows that mcrit again increases in unit steps with increasing Re, starting
from 1 for 0 Re < 20 and increasing to 8 at Re . As for = 0.5, there is again no Re
21
for which mcrit = 0. As Re increases through Re , mcrit jumps directly from 8 to 2. For
Re Re Re AP , mcrit remains at 2.
As is the case for other values of , figure 2.5c shows that kcrit is again a piecewise
continuous function of Re. For = 0.5, crit > 0 and mcrit = 1 at Re = 0
2.5. Discussion
2.5.1. Dependence of the stability boundary on
Figure 2.6 shows that at Re = 0, Ta crit is higher for the co- and counter-rotating base
states than for = 0. (The = 0.5 case, for which > 2 , is linearly stable for all Ta
at Re = 0.) Like the = 0.2 case, for which = 0.8 2 is only 20% below the Rayleigh
criterion, the = 0.5 case is also stabilized, relative to the = 0 situation, by rotation of
the outer cylinder. As Re increases, the = 0.2 co-rotating base flow becomes less stable
than both the = 0 and = 0.5 cases. For each < 2 , Ta crit approaches a nearly
Re-dependent plateau value beyond about Re = 500 (with the lower branch of the multi-
valued = 0.5 stability boundary also approaching such a plateau). For the values of
considered, Ta crit on the plateau decreases monotonically with increasing . At values
22
asymptote (for = 1). For = 1, it is known (Joseph & Munson 1970) that there is
an asymptotic Re of about 82.5 as Ta , consistent with the computed values of
Re crit = 82.36 and 82.33 at Ta = 2000 and 3000, respectively. The results in table 2.3
show that differential rotation ( 6= 1) stabilizes SPF at high Ta for = 0.9 and 1.2, and
suggest that Re min is a turning point for > 2 , except when = 1, in which case there
is no azimuthal shear in the base flow and Re min corresponds to a vertical asymptote.
more stringent convergence criteria used here, the computed results for < 2 are
essentially identical to the computational results tabulated by Takeuchi & Jankowski
(1981) over the range of Re considered by them. The results are very similar to their
experimental data at low Re, and differ in the higher part of their experimental range
(0 Re 150) in precisely the same way that Takeuchi & Jankowskis computations
differed from their own experiments. Also, as stated in 2.3, comparison of the computed
values of the dimensionless drift velocity Vdrif t /VZ = ccrit to the ratio c /kc of fitted
functional forms computed by Recktenwald et al. (1993) for 0 Re 20, = 0, and
= 0.5 shows excellent agreement, with the rms error for Re = 1, 2, . . . , 20 being
1.6 104 , where Vdrif t is the axial drift speed of the vortical structures.
For = 0, Meseguer & Marques (2002) reported results (their table 1 and figure 2)
very close to the computations and experimental data of Takeuchi & Jankowski for small
Re. For counter-rotating cylinders ( < 0), Meseguer & Marques (2002) asserted that
"When the cylinders are rotating with opposite signs of angular speeds, the centrifugal
mechanism is dominant over the axial shear, as already concluded in Meseguer & Marques
(2000), where the shear was induced by a relative sliding between the cylinders." As has
been shown for = 0.5 (figure 2.5a), there is in fact a transition from centrifugal to
TS-like ("shear") instability at higher Re.
23
Consideration of an insufficient range of azimuthal wavenumbers, identified as the
source of the discrepancy between the = 1 results of Meseguer & Marques (2002) on the
one hand and ours and those of Joseph & Munson (1970) on the other hand (see 2.4.2),
affects other stability boundaries reported by Meseguer & Marques (2002). Figure 4 of
Meseguer & Marques (2002) shows a stability boundary in the Re-Ta plane for Re ,o
o Ro (Ro Ri )/ = 450 (corresponding to Ta = 225 for = 0.5), beginning at Re = 0
just above Ta = Re ,i = 900, with just below 2 = 0.25. For sufficiently large Re (say,
Re > 40), Ta decreases monotonically, corresponding to increasing values of for Re ,o
fixed. As one continues to smaller Ta on the stability boundary of Meseguer & Marques
(2002), one reaches a turning point near Re = 107, with Re now decreasing as Ta decreases.
Near Re = 103, the stability boundary passes through Ta = 450, corresponding to = 0.5.
For = 0.5, figure 4 of Meseguer & Marques (2002) shows SPF to be stable for Re up to
about 103, with mcrit = 5 (figures 4 and 10 of Meseguer & Marques (2002)).
In contradistinction to those results, the computations for = 0.5 show that the
stability boundary passes through Ta = 450 near Re = 86 (see figure 2.3a), at which point
mcrit = 7 (figure 2.3b). Over the range of Re considered by Meseguer & Marques (2002),
figure 2.7 shows the correct stability boundary in the Re-Ta plane for Ta Re ,o = 225
(corresponding to Re ,o = 450 and figure 4 of Meseguer & Marques (2002)), along with
the variation of mcrit along the curve. Results in figure 2.7 were obtained by selecting
Ta, calculating from Ta = 225, and then computing a critical Re, or by selecting Re,
Ta < 0) cases for Ta = 225 is consistent with Re = 0 results shown in figure 37.1.b of
Joseph (1976). Steady axisymmetric SPF is stable in a region lying between the upper
and lower portions of the stability boundary. It is conjectured that the stability boundary
24
crosses the Ta = 0 axis at Re AP = 10359, with o (Ro Ri )2 / = Ta maintaining its
constant value of 225 while the sign of Ta (and i ) changes. For Ta = 225 and Re lying
between the two turning points (at approximately 75 and 88), there are three disjoint
ranges of Ta = i (Ro Ri )2 / for which SPF is unstable, and two in which the base flow
is linearly stable. Note that at Re = 0, Ta crit is invariant with respect to the sign of m,
so that there is a positive critical azimuthal wavenumber, 2, in addition to the negative
value shown. Also note that between Re = 0 and the point near Re = 39.5 at which
= 2 = 0.25, Ta (and hence ) are nearly constant, with Ta varying between 900 (near
Re = 39.5) and a maximum of about 905.3 (near Re = 30), and intersecting the Re = 0
axis near Ta = 903.3.
Comparison to figure 4 of Meseguer & Marques (2002) shows that the correct stability
boundary differs significantly from that shown by them, and that even on the upper portion,
the dependence of mcrit on Re is entirely different. The reason for the discrepancies on
the upper ( > 0) portion is evident from this works figure 2.4, which shows all of the
neutral curves in the k-Ta plane for = 0.5 and Re = 100. In that case, SPF is stable
below the minimum of the m = 6 neutral curve, and above the maximum of the m = 7
neutral curve (see also figure 2.3b). Consistent with the explanation identified in 2.4.2
for the discrepancies in the = 1 results, it is conjectured that the results shown in figure
4 of Meseguer & Marques (2002) were also obtained without adequate consideration of
negative values of the azimuthal wavenumber, which are critical for sufficiently large .
When the correct asymptotic Re (82.36) is used in Eq. 5.9 of Meseguer & Marques (2002),
the value of Re
eff for SPF in their table 2 is reduced to 66.18, about 20% and 22% (rather
than 7% and 9%) below the rotating Hagen-Poiseuille and spiral Couette values (82.88 and
85.11) cited, respectively. The lower ( < 0) portion of the stability boundary in figure
2.7 shows that for o (Ro Ri )2 / = 225, Ta crit for the counter-rotating case is nearly
independent of Re up to at least Re = 125.
25
Also, note here that the neutral curves shown in figures 5(a-d) of Meseguer & Marques
(2002) are necessarily incomplete, since the critical azimuthal wavenumber shown on the
middle ("hidden") branch of the stability boundary in their figure 4 is m = 5, whereas
only the m = 3 and m = 6 neutral curves are shown in their figures 5(a-d) for Re values in
the range of multiplicity. It is clear that disconnected neutral curves must exist for several
values of m, as has been shown in figure 2.4 for = 0.5 and Re = 100.
of a paper by Chandrasekhar (1960), Reid (1961) conjectured that the initial (low-Re)
instability of SPF predicted in the narrow gap limit ultimately must connect to a TS-like
instability at high Re. Prior to the present work, such a connection had been made (for
= 0 and = 0.95) only when the disturbances were restricted to axisymmetric ones (Ng
from below (Gage & Reid 1968; Kelly 1994). In that case, the mean flow has no effect
on the onset of Rayleigh-Benard convection. For SPF, the Reynolds number affects the
critical Ta for the onset of centrifugal instability (weakly on the plateau), and Ta affects
the critical Re for the TS-like instability (weakly). On the portion of the SPF stability
boundary consisting of the plateau and the TS-like portion beyond, the similarity to plane
Poiseuille flow heated from below is quite striking, with coupling between centrifugal and
shear effects being very weak.
For each , mcrit decreases from its value on the high-Re plateau to 2 as one passes
through Re . For Re < Re Re crit , mcrit = 2, so that this instability is refered
to as "Tollmien-Schlichting-like", since the "true" TS instability (as 1) would be
26
axisymmetric. Note that the nonzero value of mcrit at Re AP (both for = 0.5 and
for > 0.5, as shown in chapter 3) is at variance with the expectation that the critical
disturbance is axisymmetric as Re approaches Re AP (Ng & Turner 1982, p. 101). The
critical axial wavenumber, kcrit , increases considerably as Re increases through Re .
predicted for < 2 , with Ta crit approaching a plateau value, and falling rapidly towards
zero between Re and Re AP . On the upper branch, however, Ta crit increases rapidly
with Re. This is interpreted as follows. For > 2 , the Rayleigh criterion requires that
Couette flow (Re = 0) be linearly stable for all Ta. Figure 2.3a shows that SPF is indeed
stable for all Ta in the range 0 Re < Re min , and is destabilized by axial shear in a
progressively wider range of Ta as Re increases beyond Re min . For = 0.5, it is also
apparent that for Re > Re min , SPF can be destabilized by decreasing Ta across the upper
branch of the stability boundary. For > 2 (and 6= 1), it is unclear whether the upper
branch exists for all Re > Re min , or has a vertical asymptote at some finite Re.
In interpreting figure 2.3a, it is understood that crossing the two branches of the
stability boundary at a single Re > Re min does not correspond to the same base flow
being unstable at two different values of Ta. This is because, for fixed , the magnitude of
the azimuthal component of the base flow (2.2.1b) depends on Ta/Re, while the magnitude
and profile of the axial component (2.2.1c) are invariant. Hence, the multi-valued stability
boundary in figure 2.3a is interpreted as meaning that SPF, with fixed profiles of the axial
and azimuthal velocity components and a fixed magnitude of the axial component, is stable
for two ranges of the magnitude of the azimuthal velocity component and unstable in the
intermediate range. Only if the stability boundary shown in figure 2.3a twice intersected
27
a line corresponding to Ta = Re ( constant) would it be possible to state that the same
base flow becomes unstable at two different values of Ta crit . There has been no such case
found.
Next, consideration is given to experimental verification of the double-valued stability
boundary predicted for = 0.5. The existence of a multi-valued stability boundary for
SPF is particularly significant, since SPF is one of only two steady isothermal flows for
which multiple ranges of stability have been predicted, the other being spiral Couette
flow (Meseguer & Marques 2000), for which the nominal base flow necessarily ceases
to exist at a finite time (regardless of the Reynolds and Taylor numbers) for cylinders
of finite length. (For the two-phase parallel shear flow considered by Blennerhassett
(1980), the disconnected neutral curves shown (his figure 5) result from a two-dimensional
linear stability analysis. As discussed by Pearlstein (1987) and Renardy (1989), Squires
transformation is applicable to that flow but Squires theorem is not, so that the number
of critical values of the Reynolds number cannot be determined on the basis of a two-
dimensional analysis.) For no flow in which multi-valued stability boundaries have been
predicted is there any experimental verification (or attempted verification) of the existence
does not require establishment and maintenance of one or more concentration gradients,
which is difficult in systems with the rigid impermeable boundaries used in the analyses.
To locate two values of Ta crit at a fixed Re beyond the turning point at Re min , a
simple approach would seem to be to start at two Taylor numbers (each with the same
), greater than and less than Ta(Re min ), and to increase Re from Re = 0, where circular
Couette flow is known to be stable according to the Rayleigh criterion. If in both cases
one could proceed to an Re in excess of Re min without destabilizing the base flow, one
28
could then stop at that Re and increase Ta for the smaller Ta and decrease Ta for the
larger Ta, until the two branches of the stability boundary are encountered. If there is
no pair of Taylor numbers bracketing Ta min for which it is possible to maintain the base
flow from Re = 0 to Re > Re min , then one must conclude that finite-amplitude instability
occurs along the path(s) chosen. (In this work, the term "subcritical instability" is used
in this context, since, for > 2 , SPF can be linearly unstable at Taylor numbers below a
linear critical value.)
Takeuchi & Jankowski (1981), whose figure 1a shows that for = 0, Ta passes through a
local maximum near Re = 60, corresponding to the existence of multiple critical values of
Re over some Ta range. For = 0, figure 2.1a of the present work shows that for each Ta
in the range 68.19 < Ta < 104.4, SPF is stable for a finite Re range not including Re = 0.
For Ta = 100, figure 2.8 shows that the stable Re range lies between the minimum of a
"banana-shaped" m = 5 neutral curve, and the maximum of a closed and disconnected
m = 2 neutral curve. The other neutral curves shown (m = 0 and 1) pass through the
Re = 0 axis. Since the instability mechanism must be independent of the directions of
the axial and azimuthal flows, each neutral curve must be symmetric about the Re = 0
axis, from which it follows that the m = 0 and m = 1 Re-k neutral curves are closed when
viewed in the k 0 half-plane.
Invariance of the instability mechanism with respect to the directions of cylinder
rotation and axial flow can be used to represent the stability boundary in the full Re-Ta
plane instead of its positive quadrant. For = 0, Ta crit is a unimodal function of Re for
Re 0, and figure 2.1a shows that for sufficiently small Ta, SPF is linearly stable over
29
|Ta| > Ta 2 there is no stable range of Re. In each intermediate Ta range, Ta 2 Ta
Ta 1 and Ta 1 Ta Ta 2 , there are two disjoint finite ranges of stable Re, corresponding
to the gaps between the primary neutral curves and the extrema of disconnected neutral
curves lying between them in the k-Ta plane. One, for positive Re, was discussed in the
previous paragraph. The other, for negative Re, is its mirror image. At Ta 2 , the minimum
and maximum of the disconnected neutral curves meet the maximum of the lower primary
neutral curve and the minimum of the upper primary neutral curve, respectively.
Figure 2.5a shows that for = 0.5, Ta crit is not a unimodal function of Re,
with scalloping being associated with changes in mcrit shown in figure 1c of Takeuchi
& Jankowski (1981). When viewed in terms of a critical Ta versus Re in the positive
quadrant of the Re-Ta plane, the stability boundary is single-valued. In this case, there
are no closed disconnected neutral curves in the k-Ta plane at fixed Re. On the other hand,
figure 2.5a shows that there can be one (e.g., at Ta = 105), three (e.g., at Ta = 107), or five
(e.g., at Ta = 108) positive critical values of Re for fixed Ta. Existence of multiple critical
values of Re at fixed Ta must correspond to existence of one or more closed disconnected
neutral curves in the k-Re plane, each with at least two extrema.
sense. For = 0.5, however, figures 2.3b and 2.3d show that on the upper branch, mcrit
and ccrit assume negative values, with the magnitudes of both growing as Re increases.
Thus, for = 0.5, linear stability theory predicts that as Ta increases and crosses the upper
branch of the stability boundary, or decreases through the lower branch of the stability
boundary sufficiently close to the turning point, SPF becomes unstable with respect to a
disturbance propagating upstream against the mean axial flow, but with the same helical
30
sense as the base flow. It is clear from figure 2.3d that the sign of ccrit changes
discontinuously on the lower branch at the transition from mcrit = 6 to 6 near the
turning point, and that there is no point on the stability boundary at which ccrit = 0.
2.6. Conclusions
The complete linear stability boundaries for spiral Poiseuille flow presented in 2.4
show that for each rotation rate ratio < 2 considered, the connection is made between
the onset of instability for circular Couette flow with no mean axial flow to the onset of
instability in annular Poiseuille flow absent rotation. For > 2 , no instability is possible
on a linear basis for 0 Re < Re min . For > 2 and 6= 1, there is a turning point at
Re min , beyond which there exists a range of Re for which the stability boundary is multi-
valued and the flow is stable in two disjoint ranges of Taylor number. For the special
case = 1, there is a vertical asymptote in the Re-Ta plane at Re min = 82.33, a value
which differs from that found by Meseguer & Marques (2002), apparently due to their
consideration of an insufficient range of the azimuthal wavenumber m.
For < 2 , a small mean axial pressure gradient generally stabilizes the flow with
respect to centrifugal instability, as shown by earlier investigators. In each case, Ta crit
reaches a plateau before falling precipitously to zero as Re approaches Re AP , the critical
Reynolds number for annular Poiseuille flow. The transition from centrifugal instability
31
Chapter 3
Computational Assessment of Subcritical Instability and
Apparent Transition Delay in Spiral Poiseuille Flow Experiments
3.1. Introduction
In chapter 2 complete stability boundaries have been presented for spiral Poiseuille
flow (SPF) for the radius ratio Ri /Ro = 0.5 and several combinations of the rotation
rate ratio o /i over the entire range of Re V Z (Ro Ri ) / for which the flow
is linearly stable at any Taylor number Ta i (Ro Ri )2 /, where and VZ are the
kinematic viscosity and the mean of the axial component of the base-flow velocity, and
i and o are the (signed) angular velocities of the inner and outer cylinders, whose
radii are Ri and Ro , respectively. Their work, for the same radius ratio considered in the
experiments and computations of Takeuchi & Jankowski (1981), as well as in the more
recent computations of Meseguer & Marques (2002), extends the range of Re investigated
in previous computations and experiments by eightyfold and seventyfold, respectively, and
axisymmetric disturbances over 6000 < Re 7739.5 for = 0.95. For both radius ratios,
their results showed that Ta crit initially increases as Re increases from zero, before reaching
a broad plateau. For = 0.95, Ng & Turner found that the Ta at which the flow becomes
unstable with respect to axisymmetric disturbances decreases rapidly beyond Re = 6000.
32
Their results agreed well with the = 0 experimental results both of Mavec (1973) up to
Re = 400 for = 0.77, and of Snyder (1962, 1965) up to Re = 200 for 0.95.
Here, we compute complete stability boundaries for the radius ratios considered by Ng
& Turner, which are equal or close to those for which most of the experimental work has
all of the combinations of Re, , and considered by Snyder (1965) and Mavec (1973).
The stability boundaries extend the earlier work of Ng & Turner for = 0.77 and
0.95 to the full range of Re for which the flow is linearly stable at any Ta, and account
for arbitrary disturbances of infinitesimal amplitude over that range. Here it is shown
that in the only SPF case ( = 0, = 0.95) for which a connection of the Re = 0
centrifugal instability to the high-Re shear instability had been made (Ng & Turner
1982), the transition actually occurs from a nonaxisymmetric centrifugal instability to
a nonaxisymmetric TS-like instability, at a Reynolds number in the range where Ng &
is the length of the annulus. Takeuchi & Jankowski and Ng & Turner noted that beyond
some -dependent Re, there is an apparent transition delay, in which the experimental
values of Ta crit found by flow visualization lie above the critical values computed by linear
stability theory, with the discrepancy increasing with Re. Beyond concluding that "the
linear theory has an even greater range of applicability than demonstrated here," Takeuchi
& Jankowski proposed two mechanisms for the systematic underprediction of Ta crit (or
"apparent transition delay") at higher Re. The first is associated with use of experimental
33
annuli insufficiently long to allow secondary flow to develop to detectable amplitudes. The
second is applicable to experiments in which dTa crit /dRe was negative and a constant-
head pump was used, in which case formation of weak vortical structures would have the
effect of reducing the mean axial velocity.
potentially responsible for failure to detect instability at and above critical Taylor numbers
predicted by linear theory are also unimportant. Thus, for at least some combinations
of and , there is good reason to believe that instability sets in through infinitesimal
disturbances up to Reynolds numbers in excess of several hundred. The results also allow
stability boundaries for SPF are presented for = 0.77 and 0.95 and several values of .
In 3.3, we present results for specific combinations of Re, , and studied experimentally,
to which no previous comparison to theory has been made, along with detailed comparison
to experiment, and a discussion of implications for interpretation of the experiments.
Additional discussion follows in 3.4, and some conclusions are presented in 3.5. An
account of this work has been submitted for publication (Cotrell, Rani & Pearlstein 2003).
for the two radius ratios (0.77 and 0.95) considered by Ng & Turner at several values
of the rotation rate ratio , accounting for arbitrary three-dimensional disturbances of
infinitesimal amplitude. Convergence tests, as described in chapter 2 for = 0.5, were
34
performed for = 0.77 and 0.95 at each for which results are reported below. In general,
the number of radial expansion functions required to achieve convergence decreases with
increasing . Except where otherwise indicated, these results are in excellent agreement
with those tabulated by Ng & Turner over the ranges of Re they investigated. The results
cover the entire range of Re for which the flow is linearly stable, from Re = 0 (the Taylor-
Couette limit) to Re AP (beyond which the base flow is unstable for all Ta, corresponding
to onset of TS-like instability in nonrotating annular Poiseuille flow). We note that Re AP
is independent of , since it corresponds to the nonrotating limit.
Extending the analysis to higher Re, we see in figure 3.1a that Ta crit continues on a
plateau (Ta crit = 58.6) up to Re = 8677, and falls rapidly to zero over the range
Re < Re Re AP = 8883.3. This value of Re AP is in good agreement with previous
graphical results (Mahadevan & Lilley 1977; Garg 1980) for the stability of nonrotating
by axial flow over the entire range of the centrifugal instability contrasts with that found
for = 0.5, where increasing Re stabilized and destabilized SPF in different parts of the
pre-plateau range 10 < Re < 400. For = 0.77, Ta crit is nearly constant in a plateau
range (200 < Re < 8000) that starts at smaller Re values than for = 0.5. The value of
Re AP (8883.3) is also smaller than the value (10359) for = 0.5. As discussed in 3.4.1,
35
the dependence of Re AP on is part of a systematic variation from Re AP = 5772 as 1
to Re AP as from above, where < 0.15 (Mahadevan & Lilley 1977; Garg
1980).
As for = 0.5, figure 3.1b shows that the stepwise increases in mcrit over the range
0 Re Re are by one unit. Our results agree with the tabulated values of Ng & Turner
except at Re = 300 and 500 where, as per the discussion in chapter 2, we find mcrit = 19
and 20, respectively, in contrast to their values of 20 and 21, apparently due to tighter
control of convergence. Figure 3.1b shows that mcrit = 21 over a significant portion of
the high-Re plateau, compared to the maximum value of mcrit = 7 for = 0.5. As Re
passes through Re , mcrit decreases abruptly from 21 to 2. On the TS-like branch, mcrit
suffers a final step decrease to 1, its value at Re AP . This contrasts to the constant value
of mcrit = 2 on the TS-like branch for = 0.5.
Figures 3.1b and 3.1c show that, as for = 0.5, the piecewise continuous dependence
of kcrit on Re is associated with jumps in mcrit , as described by Ng & Turner. There is
again a value of Re ( 42) below which kcrit increases monotonically with Re for each
mcrit , and above which kcrit decreases monotonically with Re for each mcrit . Note that
Turner (1982) at their 18 values of Re in the range 0.01 Re 6000, except at Re = 300
and 500, where the computed values of mcrit differ from theirs by one. As for = 0.5
(chapter 2), the almost constant value of ccrit on the mcrit = 0 branch corresponds to
a dimensional frequency that increases nearly linearly from zero as Re increases. For
mcrit < 15, ccrit decreases monotonically with Re for each mcrit , while for mcrit 15, ccrit
increases piecewise continuously up to mcrit = 21. This contrasts with the results for =
0.5, for which ccrit decreased monotonically with Re for each mcrit . For 800 < Re Re
36
(mcrit = 21), ccrit remains nearly constant (ccrit 2.36). As Re passes through Re , ccrit
decreases abruptly to about 0.39, and then decreases slightly to its Ta = 0 value of 0.38
as Re increases from Re to Re AP .
The stair-step behavior of mcrit and the associated discontinuous dependence of kcrit
and ccrit on Re indicate that Ta crit is a continuous but only piecewise differentiable
function of Re, as was the case for = 0.5. That the slope discontinuities are less apparent
than in the = 0.5 case reflects the fact that the values of mcrit are considerably larger for
= 0.77, so that (m + 1)/m and (m + 1)2 /m2 , which express the ratios of the m-dependent
terms in the disturbance equations for consecutive azimuthal wavenumbers, are closer to
unity at the larger . Note also that the "width" of the steps (i.e., the Re range for which
mcrit is constant) is considerably smaller for = 0.77 than for = 0.5, corresponding to
the larger range of critical azimuthal wavenumbers for = 0.77 (0 mcrit 21) compared
shows that mcrit increases to 20 in unit steps for 0 Re 900, and remains constant for
900 Re < Re = 8712. As one passes through Re , mcrit jumps directly from 20 to
2. As Ta decreases below about 10, the critical m on the nearly vertical TS-like branch
again decreases from 2 to 1, the value it maintains all of the way to Re AP = 8883.3, where
Ta crit = 0.
The critical k increases piecewise continuously with Re until reaching a global
maximum near Re = 60 (figure 3.2c), beyond which kcrit decreases sharply until Re .
The variation of kcrit with Re is similar to that found for = 0.5 and = 0.2. Figure
3.7d shows that the dependence of ccrit on Re is very similar to that for = 0 (figure
3.2d).
37
For = 0.5, there are significant differences between the = 0.5 and = 0.77
stability boundaries. First, figure 3.3a shows that for = 0.77, SPF is stabilized with
increasing Re, with Ta crit increasing monotonically from its Re = 0 value of 32.83 to about
69 on a broad plateau between about Re = 700 and the precipitous drop at Re = 8579.
This behavior contrasts with that for = 0.5, where SPF is alternately destabilized and
stabilized for 0 < Re < 1000. Second, for = 0.77, the nearly constant value of Ta crit
on the plateau ( 69) is greater than the value at Re = 0, unlike the = 0.5 case, for
which Ta crit on the plateau lies below the Re = 0 value. Finally, the scalloped behavior
critical azimuthal wavenumbers are mcrit = 23 and 24 over relatively wide Re ranges on
the high-Re plateau. As Re passes through Re , mcrit decreases directly from 24 to 2.
For Re < Re < Re AP , mcrit again decreases from 2 to 1. As shown in figures 3.3c and
3.3d, kcrit and ccrit are piecewise continuous functions of Re, whose dependence on Re is
6000 Re 7739.5. Here it is shown that Re AP is indeed 7739.5 (where mcrit = 0),
and completed the stability boundary by considering nonaxisymmetric disturbances up to
that Re. Comparison to the results of Ng & Turner over the range 0 Re 6000 (see
chapter 2) is excellent. For Re > 6000, the results agree with theirs only in the narrow
38
at Re = 7000, Ng & Turner report a Taylor number for the onset of axisymmetric instability
which, using the present scaling, corresponds to Ta = 367.68, as opposed to the computed
value of 47.56 (for mcrit = 149).
Qualitatively, the stability boundary (see figure 3.4a) is similar to that for = 0.77.
For = 0.95, Ta crit remains nearly constant for 1000 < Re < 7716, with its value ( 47)
on the high-Re plateau being greater than at Re = 0. The values of Ta crit for = 0.95
are less than for = 0.5 and 0.77.
As for = 0.5 and 0.77, figure 3.4b shows that mcrit increases by unit steps (from
the computed mcrit values and those found by Ng & Turner for Re 6000. For = 0.95,
the computed values of kcrit and ccrit are essentially identical to theirs over that range,
except at Re = 10, 100, and 2000, where differences in ccrit are due to unit differences in
the computed values of mcrit . The dependence of kcrit and ccrit on Re shown in figures
of the kcrit and ccrit plots at high Re. For = 0.95, the constant-mcrit ranges of Re
are so narrow, and mcrit is so large, that the step discontinuities in kcrit and ccrit would
not be graphically apparent beyond about Re = 80 (corresponding to mcrit of about 25)
39
regardless of the number of points shown. This is a consequence of the fact that for the
centrifugal instability, mcrit grows rapidly as 1 and behaves much like a continuous
wavenumber. Also, note that the values of mcrit are fully resolved. Inadequate resolution
(particularly in choosing insufficiently small tolerances for the k- and Ta-iterations) can
lead to spurious nonmonotonic variation in the computed values of mcrit , since the minima
of neutral curves for large m occur at extremely similar Ta values.
compared their computations at = 0.77 to experimental data of Nagib (1972) and Mavec
(1973) up to Re = 400 at the same , and computations at = 0.95 to data of Snyder
(1962, 1965) up to Re = 200 at = 0.959. These comparisons showed generally good
in the = 0 experimental work of Becker & Kaye (1962), Sorour (1977), Gravas & Martin
(1978), Sorour & Coney (1979), Grosvenor (1981), and Greaves, Grosvenor & Martin (1983)
are presented in 3.3.1. In 3.3.2, computational results are presented for and detailed
comparisons are made to data for all values of considered in the Re 6= 0 experiments of
Mavec (1973) for = 0.77 and Snyder (1965) for = 0.9497 and 0.9590. A discussion
of the implications for interpretation of the experimental data, including an assessment of
subcritical instability, and the mechanisms proposed by Takeuchi & Jankowski as possible
explanations for Ta expt comp
crit > Ta crit , follows in 3.4.1.
40
to computation has previously been made. Here, we consider the experimental data, and
present computations for relevant values of Re and that shed light on interpretation
of the former. Unless identified by uppercase superscripts, Reynolds number and Taylor
number values reported by authors using different definitions of those quantities than used
= 0.734; no comparison was made for = 0.820. In both cases, Ta crit initially increased
with Re, reaching its maximum value between Re = 400 and 550 for = 0.734, and
between 350 and 400 for = 0.820. For larger Re, Ta crit decreased monotonically to zero
at values of Re near 1000 and 900 for = 0.734 and 0.820, respectively. For both radius
ratios, the decrease in Ta crit with Re was nearly linear. For = 0.734, table 3.1 shows the
11 smallest values of Re read from figure 13 of Kaye & Elgar, along with the corresponding
critical values of their Taylor number Ta KE
crit (in parentheses). Also shown are values of
Ta comp
crit corresponding to each Re, and values of Ta expt KE
crit = Ta crit [2(1 + )/(1 )]
1/2
calculated from the reported Taylor numbers. Over the range of Re shown in table 3.1,
experimental and computed results are in generally reasonable agreement. The difference
(2%) between the experimental and computed values of Ta crit at Re = 0 is comparable
to the 1% difference between the experimental value of the critical angular velocity and
"the theoretical value predicted by Taylors theory" cited by Kaye & Elgar, and provides
a measure of the error in reading values of Ta KE from the authors plot. For the larger
Re shown, the experimental values of Ta crit lie 4-20% above the computed values. In the
range of Re where Ta comp expt
crit exhibits plateau behavior, Ta crit increases slowly with Re until
reaching a maximum at some Re between 400 and 550, beyond which it decreases. For
Re values in figure 13 of Kaye & Elgar beyond those shown in table 3.1, Ta expt
crit decreases
41
falling off rapidly to zero near Re = 104 via a TS-like instability. The implications of this
comparison for interpretation of these experimental results, and quite similar experimental
results for = 0.820 (figure 14 of Kaye & Elgar), are discussed in 3.4.1.
For = 0, Yamada (1961, 1962) obtained qualitatively similar results for = 0.971,
0.981, and 0.987. For a number of Re values between 0 and 1100, he reported critical
dimensionless rotation rates at which either the ratio of the torque coefficient to its base-
flow value (Yamada 1961; figure 16) or the dimensionless axial pressure drop (Yamada
1962; figure 18) increased "sharply." His results, while differing somewhat among the
small-gap radius ratios considered and between the two diagnostics of transition, generally
showed that Ta crit is a unimodal function of Re, and reaches its maximum value in the
range 400 Re 700. Beyond the maximum, Ta crit values reported for both transition
diagnostics decreased rapidly with Re. For example, at = 0.987, the reported Ta crit
decreased by more than a factor of three between its maximum near Re = 650 and the
largest Re for which Yamada reported results, 1100. For each , extrapolation of the
reported critical Taylor numbers to zero (i.e., the annular Poiseuille limit) gives an intercept
near Re = 1200.
For = 0 and 0.81 , heat transfer measurements by Becker & Kaye (1962)
showed that the ratio of the Nusselt number to its base-flow value undergoes a very well-
defined transition at a Ta crit that increases with Re over 0 Re 1430. (Beyond
Re = 1430, no clear transition is evident in the results of Becker & Kaye.) At the eight
smallest Reynolds numbers considered by those authors, table 3.1 shows values of the
Taylor numbers Ta BKcrit (in parentheses) at which Becker & Kaye reported transition, along
1/2
with critical values Ta expt BK
crit = 2Ta crit (1 )/(1 + ) and Ta comp
crit computed from this
linear stability analysis at the same . (Although transition is quite distinct at Re = 1007.5
and 1430, the Nusselt numbers before transition at these two Re values differed from the
nominal base-flow values by about 10% and 50%, respectively. Note also that the Taylor
Based on the cylinder radii given, = 0.8097 0.0008. Based on the reported gap
between cylinders and the outer cylinder radius, = 0.8055 0.0010.
42
number defined by Kaye & Elgar (1958) differs from that of Becker & Kaye, and that the
Reynolds number defined by both Kaye & Elgar and Becker & Kaye differs from ours by
a factor of 2.) Parenthetical values in table 3.1 were read from figure 4 of Becker & Kaye,
and are thought to be in error by no more than one or two units in the third significant
the two values corresponding to the radii and gap given by Becker & Kaye.
In general, the results are in good agreement with those of Becker & Kaye. At the
four smallest Re, the experimental Ta crit values lie 6-17% below predictions of linear
stability analysis, while at the next four Re, the experimental values are 11-20% higher than
predicted. The latter discrepancy is consistent with inadequate axial development length
for the disturbance flow in the apparatus (L/(Ro Ri ) 172) of Becker & Kaye, identified
later by Takeuchi & Jankowski as an explanation for a similar discrepancy between their
experimental results (for which L/(Ro Ri ) 115) and their own computations. Since
dTa crit /dRe 0 (see table 3.1) up to Re = Re for 0.77 (see also figures 3.1a and
3.4a), the alternate constant-head mechanism proposed by Takeuchi & Jankowski is not
applicable in this case.
Gravas & Martin (1978) used hot-wire anemometry to investigate onset of secondary
flow at = 0 for = 0.576, 0.81, and 0.9 over the range 43 Re 1000. For each , they
found that Ta crit increased monotonically with Re. Here, we make comparisons to the
current results for the similar radius ratios of 0.5, 0.77, and 0.95, as well as to additional
are less well defined. At = 0.576, however, there is no evidence of plateau behavior, while
plateau behavior is predicted for = 0.5 (figure 2.1a of chapter 2). To establish that
differences between the experimental and computational results are not simply due to the
43
different radius ratios in the experiments and computations, we performed computations
at the experimental values of . At = 0.576, Ta expt
crit increases monotonically with Re
over 70 Re 255 (see figure 5 of Gravas & Martin), whereas computations for the same
radius ratio show (consistent with results for = 0.5) that Ta comp
crit decreases monotonically
with Re over the same range. Some of this discrepancy might be associated with annuli
of small aspect ratio L/(Ro Ri ) (varying between 32 and 175 in the experiments of
Gravas & Martin), in which vortical structures have insufficient opportunity to develop to
a detectable amplitude. The discrepancy might also be partly attributable to significant
geometric imperfection (i.e., axial and azimuthal variations of the gap), as discussed in
later work by Greaves, Grosvenor & Martin (1983).
For = 0, Sorour (1977) and Sorour & Coney (1979) used hot-wire anemometry
to determine Ta crit for = 0.8 in the range 26 Re 468 and for = 0.955 in
the range 26.4 Re 595. For each Re, Sorour & Coney reported critical values of
Ta SC = 2Ta 2 2 /(1 2 ) at nine uniformly spaced radial locations in the annular gap, to
which a curve was fitted. Table 3.2 shows in parentheses the minimum value of Ta SC
on that curve as read from either figure 1 of Sorour & Coney or similar figures in Sorour
for most Re, so that even for these very similar radius ratios, we have performed specific
computations for the experimental .) Parenthetical values of Ta SC up to 104 are thought
to be correct to three significant figures, while larger values are thought to be in error by
no more than one or two units in the fourth significant figure.
First consider the = 0.955 results, for which the aspect ratio is more than four
times its value for = 0.8. As shown in table 3.2, the results of Sorour & Coney are in
excellent agreement with the current computations up to Re = 325, with the maximum
difference in that range being less than 3.2%, and the mean of the absolute value of
the relative difference being 2.2%. The differences between computed and experimental
44
values are roughly comparable to uncertainties in Ta expt
crit associated with readability of
experimental Taylor numbers as propagated through the square root relationship between
Ta and Ta SC . The aspect ratio in the = 0.955 experiments of Sorour & Coney exceeded
570, compared to 114 and 186 for Kaye & Elgar, 172 for for Becker & Kaye, 115 for Takeuchi
& Jankowski, and a range of 32-175 for Gravas & Martin. From the close agreement of the
current computed critical Taylor numbers with the data of Sorour & Coney, we conclude
that finite-amplitude instability either did not occur in their experiments, or occurred
only slightly below the critical values predicted by linear stability theory. It is furthered
conjectured that the larger aspect ratio of Sorour & Coney provided sufficient streamwise
length for development of vortical structures detected by hot-wire anemometry, at least up
to Re = 325. As Re increases, table 3.2 shows that the value of Ta at which instability
could first be detected in an annulus of fixed aspect ratio continued to increase, while the
linear stability analysis predicts a plateau in the Re-Ta plane. This is consistent with the
hypothesis of insufficient axial development length at high Re.
For = 0.8 and Re 320, table 3.2 shows that agreement between the experiments
of Sorour & Coney and computation is still good, but less close than for = 0.955. For
= 0.8, the maximum difference and mean absolute value of the relative difference over this
range of Re are 13% and 6.0%, respectively, with the experimental values of Ta crit lying at
or slightly below the computed values, except at the two lowest Re and at the highest Re.
(For Re = 36.5, the nearly identical experimental values of Ta SC at each radius seem to be
anomalously high.) These results are taken as evidence that for Re 320, either there is
no subcritical instability, or subcritical instability sets in at Ta values only slightly below
those predicted by linear theory. The relatively small differences between experiment
and computation at the highest Re suggest that the aspect ratio (L/(Ro Ri ) = 130) is
sufficient to allow development of the disturbance to detectable levels for = 0.8 and
45
Re 320. (Beyond Re = 320, the experimental values of Ta crit continue to grow with
Re, while the computed values have essentially reached their plateau, as shown in table
3.2.)
Grosvenor (1981) and Greaves, Grosvenor & Martin (1983) used hot-wire anemometry
to detect the onset of instability in SPF for = 0 over essentially the same range
(43.5 Re 1112) previously considered by Gravas & Martin (1978) for similar radius
ratios. These authors reported Taylor numbers at four azimuthal positions in a facility
with considerably less axial and azimuthal variation of the gap (maximum 3.9% variation
from the mean gap for = 0.909) than in the work of Gravas & Martin. Each parenthetical
value in table 3.2 was taken directly from table A2 of Grosvenor, and corresponds to the
lowest value among those for the four azimuthal positions. For = 0.909 (with aspect ratio
310), the computed Ta crit at Re = 43.75 lies less than 5% below the experimental value,
with the discrepancy increasing to about 9% at Re = 87.5, 131, and 262. The fractional
discrepancy decreases (to 6%) at Re = 435.5, before the expermental values of Ta crit
fall increasingly below the computed values at higher Re. These results are interpreted
in terms of sufficient axial development length at low Re, and subcritical instability at
higher Re. For = 0.565, computations at Re = 69.35, 198.9, and 300.45 show that
the computed values lie 40-54% below the experimental values. Also, these results are
interpreted as evidence of a profoundly insufficient axial development length, consistent
with the annular aspect ratio of less than 50.
For = 0 and = 0.8, Buhler & Polifke (1990) reported that the onset of instability
occurred for mcrit = 1 over 2.7 < Re < 4.6, and for mcrit = 0 for both larger and smaller
Re. They also reported a "scalloped" Re-Ta crit stability boundary. Their results differ
from ours at = 0.77 (figure 3.1a), and from specific computations for = 0.8, in which
46
Wavespeeds. From the definition ccrit = i,crit /kcrit (see chapter 2), we can show
that Vdrif t /VZ = ccrit , where Vdrif t is the axial drift speed of the vortical structures. For
= 0, current results can be compared to two previous reports of dimensionless axial drift
speed. At Taylor numbers somewhat above critical, Donnelly & Fultz (1960) measured
dimensionless drift speeds of 1.25 , 1.07 , and 1.36 at Re = 3.13, 4.40, and 5.70 for = 0.9497 ,
and Howes & Rudman (1998) obtained Vphase /VZ = 1.16 0.005 in slightly supercritical
computations at Re = 2.61, Ta = 12.5, and = 0.9524. These results compare very well
to current computed values of ccrit for = 0.95, which vary between 1.169 and 1.170 for
0 Re 8.
Figure 5 of Sorour & Coney (1979) shows experimental values of Vdrif t /VZ at seven
values of Re over 48 < Re 500 for = 0.80, and at nineteen values of Re over
16 < Re < 610 for = 0.955. For = 0.80, the dimensionless drift velocities decreased
which point ccrit jumps to 1.54, coincident with mcrit jumping to 3. At larger Re, the
computed ccrit undergoes a series of discontinuous increases, separated by progressively
shorter Re intervals of monotonic decrease. Computations performed for = 0.80 show
similar behavior. For = 0.955, the experimental dimensionless drift velocity increases
from about 1.6 near Re = 16 to about 2.8 near Re = 100, before decreasing to about
1.8 near Re = 610. For = 0.95, figure 3.4d shows that for Re = 10, ccrit is about 1.3,
and that for the experimental Re increments, ccrit increases monotonically and approaches
an asymptotic value of about 3.3 slightly beyond the highest Re considered by Sorour &
Coney. These results, which differ only slightly from those of the computations performed
for = 0.955, are in good agreement with the data of Sorour & Coney. A later paper by
Simmers & Coney (1980) states that the wavespeed measurements of Sorour & Coney (for
47
which Taylor numbers were not reported) were performed "with the flow just critical,"
from which it can be inferred that over the range of Re where the critical Ta values are in
excellent agreement, the wavespeed should be close to the critical values.
Figure 2 of the experimental paper by Wereley & Lueptow (1999) shows that for = 0
and = 0.83, mcrit changes from 0 to 1 near Re = 8. Our computations show that this
transition occurs at Re = 16 and 8 at = 0.77 and 0.95, respectively, suggesting that
the experimental Re for transition is a bit low. Also, invariance of Ta crit with respect to
the direction of the axial flow (see 3.5.5 of chapter 2) requires that dTa crit /dRe vanish
Values of (NR )i and (NR )o have been read for each data point in Mavecs figure
4, and calculated = (NR )o /(NR )i and the corresponding critical Taylor number,
Ta expt comp
crit = (NR )i (1 )/2. These, along with values of Ta crit and mcrit computed at
the calculated values of , are shown in table 3.3. Readability errors in (NR )i and (NR )o
are small enough that the calculated values of and Ta expt
crit are thought to be accurate to
within 1%. For the 18 cases shown in table 3.3 for which > 2 = 0.5929, we expect that
48
there is a range of Re for which there are two values of Ta crit , as discussed for = = 0.5
(see chapter 2). For purposes of comparison to Mavecs results in these cases, we have
computed only the smaller Ta crit . In what follows, we consider the results in two ranges
of Re: 24 Re 106, and Re 134.75.
For each Re, the experimental and computed values of Ta crit are unimodal functions
of , providing that the two = 0 experimental points at Re = 33, 49, and 330 are
averaged. At small Re, agreement between experimental and computed values of Ta crit
is generally excellent. In the range 24 Re 106, there are 30 combinations of and
Re for which the experimental and computed results differ by less than 2%, including
ten of the twelve rotation rate ratios at Re = 106. This level of agreement is considered
to be an indication that the random errors in Mavecs experimental data are generally
very small. Nonetheless, at small Re, there are still some systematic differences between
experiment and computation. For each Re in the range 24 Re 106, the experimental
value of Ta crit exceeds the value predicted by linear theory for each 0.7 and for each
0.35. In the intermediate range 0.7 < < 0.35, we have calculated the mean and
root-mean-square (rms) of the difference = Ta comp expt
crit Ta crit between the experimental
and computed values of Ta crit , along with the variance of , at each Re. For two Re,
the mean and rms values are much larger than the variance. At Re = 49, the mean
and rms differences are 3.3 and 3.4, respectively, and the variance is 0.45. (These values
are not significantly reduced by averaging the two experimental values at = 0.) At
Re = 63.5, the corresponding values are 3.1, 3.1, and 0.27. By contrast, the mean and rms
differences and the variance are 0.25, 0.43, and 0.15, respectively, at Re = 24, and 0.18,
0.41, and 0.16, respectively, at Re = 106. These results suggest small systematic errors in
the experiments at Re = 49 and Re = 63.5, discussed below.
For Re 134.75, the situation is quite different. First, in this range, the experimental
and computational results agree within (the arbitrary) 2% at only eight points, four of
which are at Re = 134.75, the smallest Re in this range. Second, in this range of larger
49
Re, the difference at each Re varies nearly monotonically with , beginning with < 0
for the most negative values of , and ending with > 0 for the most positive values.
Linear least-squares fits of the form = a + b using only the values 0.7 < < 0.35
give values of the slope a = 2.6, 3.1, 3.8, 4.3, and 4.7 at Re = 134.75, 166, 244, 330, and
403.5, respectively, compared to 0.67, 1.7, 1.4, 1.4, 1.3, 1.1, and 0.50 at Re = 24, 33, 49,
63.5, 65.5, 82.5, and 106, respectively. When all of the values of are included, the slopes
are 8.9, 2.5, 1.4, 6.1, 0.4, 1.2, and 0.2 at the lower Re values, and 6.6, 6.7, 9.2, 14.9,
and 21.7 at the higher values. These results are discussed in 3.4.1.
In figure 3.5, we plot the computed results presented in table 3.3 using axes
o (Ro Ri )2 / Ta and i (Ro Ri )2 / Ta, as is conventional for Re = 0 (cf. figure
6.2 of DiPrima & Swinney 1985). This representation is useful for assessing the effects
of Re and understanding behavior near the Rayleigh line ( = 2 ). Figure 3.5 shows
that for = 0.77, the value of o (Ro Ri )2 / at which the critical i (Ro Ri )2 /
attains its minimum value initially shifts to more negative values as Re increases. At
an Re apparently lying between 49 and 82.5, the location of this minimum shifts towards
positive values of o (Ro Ri )2 /, with the o (Ro Ri )2 / = 0 axis being crossed between
Re = 106 and 134.75. The high-Re plateau behavior discussed in 3.2.1-3.2.2 is reflected
in the near-coincidence of the results for Re = 330 and 403.5. (The points in figure 3.5 are
connected by third-order splines, which is responsible for the slight apparent divergence of
the curves for Re = 330 and 403.5 at sufficiently positive and negative .)
3.5, two values of Ta crit correspond to a constant- straight line intersecting the stability
boundary for a given Re at two different values of o (Ro Ri )2 /.) For Re = 106 and
= 0.81, for which table 3.3 shows Ta expt comp
crit = 78.79 and Ta crit = 77.32, we computed
50
a second value, Ta crit = 463. None of Mavecs results suggest that he encountered two
critical values of o for a single , so in all other cases we have computed only the smaller
Ta crit for comparison to his results.
0.95. For = 0.9497 and 0.959 and seven nonzero values of Re up to 200,
Snyder (1965) measured critical values of 1 for fixed values of 2 , using water as the
experimental fluid in annuli with aspect ratios between 285 and 349. The results shown
in dimensionless form in his figures 1-2 (for = 0.959) and figure 3 ( = 0.9497) cover
a range of spanning negative and positive values for 0 Re 200. For Re > 0, we
1/2 1/2
have read values of Ri i (Ro Ri )3/2 / and Ri o (Ro Ri )3/2 / from these figures,
and calculated the values of shown in table 3.4 from their ratio. Values of Ta expt
crit
shown in table 3.4 were obtained by multiplying ordinates read from Snyders data by
[(1 )/]1/2 . Computed values Ta comp
crit at each combination of Re, , and are shown
Snyders estimates of the systematic (1.5%) and statistical (1%) errors in his measurements
of the critical angular velocity of the inner cylinder. For 34 combinations of Re and ,
computed and experimental values of Ta crit agree within 2%. In 20 cases, agreement is
within 1%. These generally small differences between computed and experimental values
of Ta crit imply that errors in the values of Ta and due to the reading of his figures 1-3
are also small. For 0 < Re 40, the mean and rms values of (0.41 and 0.43, 0.48
and 0.50, 0.64 and 0.78, and 0.08 and 0.26, for Re = 5, 10, 20, and 40, respectively) do
not exceed 5% of Ta crit . For Re = 80, agreement is within 2% for 13 of the 17 values of
51
negative values of than at intermediate values. Second, there are values of Re for which
the variance of is much smaller than its rms value. For example, for Re = 5 and 10, the
variance of (0.017 and 0.021 for Re = 5 and 10) is about 4% of the root-mean-square
value of , which strongly suggests that the random contribution to the differences is very
small compared to the already small rms differences. This suggests, at least for Re = 5 and
10, that the primary contribution to the difference between experiment and computation
is a systematic one. As for the data of Mavec, we have calculated linear least-squares fits
= a + b to the differences as a function of for each Re. For Re = 5, 10, 20, 40, 80,
120, and 200, the calculated slopes are a = 0.036, 0.033, 0.20, 0.31, 0.90, 0.95, and
0.036, respectively. These values are much smaller than the corresponding values for the
= 0.77 data of Mavec. (When only data in the range 1.5 < < 0.74 are included, we
find a = 0.036, 0.033, 0.10, 0.31, 0.32, 0.95, and 1.3 for the same Reynolds numbers.)
Unlike the situation for = 0.77, table 3.4 shows that Ta comp
crit significantly exceeds
Ta expt
crit for only a few combinations of Re and . Specifically, /Ta crit exceeds 0.04 for
Re = 80 at = 1.91 (/Ta crit = 0.08), for Re = 120 at = 0.392 (/Ta crit = 0.18), and
for Re = 200 at = 0.146, 0.293, and 0.417 (/Ta crit = 0.05, 0.07, and 0.07, respectively).
The relatively large discrepancy for Re = 120 at = 0.392 appears to be an isolated one,
with no obvious explanation. For Re = 200, is quite small (|| 0.82) for < 0, and
then increases monotonically to a maximum of 2.36 at = 0.417 before falling rapidly to
5.16 at = 0.746. This trend is consistent with subcritical instability and insufficient
development length being unimportant at negative , and with the latter effect becoming
increasingly important as increases beyond zero.
3.4. Discussion
3.4.1. Implications for interpretation of experiment
As shown in 3.3.1 and 3.3.2, there is a broad range of Re and for which SPF
loses its stability at critical values of Ta very close to those predicted by linear theory.
Sometimes, however, there are systematic differences between experimental and computed
52
values of Ta crit . In this section, we assess the roles of finite-amplitude instability and
mechanisms of apparent transition delay in contributing to these differences, and identify
regions in the parameter space in which they occur.
For the cases considered in 3.3.2, finite-amplitude instability would correspond
to experimental values of Ta crit lying below those predicted by linear theory. (Note
that when > 2 , finite-amplitude instability on the upper branch of the stability
boundary (see chapter 2) would manifest itself through onset of secondary flow at a
Ta crit above the value predicted by linear theory.) To date, little is known about the
nonlinear stability of SPF (Joseph & Munson 1970; Joseph 1976) or even nonrotating
annular Poiseuille flow. For the latter, results are limited to a perturbation analysis for
axisymmetric disturbances (Strumolo 1983), and to computational simulations at three
Reynolds numbers (13,000, 15,500, and 20,000) at = 0.7 with initial disturbances having
one nonzero axial wavenumber (Shapiro, Shtilman & Tumin 1999). When the axial flow
between rotating cylinders is exactly a quadratic function of the radial coordinate (as
can be arranged by choosing a relative axial velocity difference between the cylinders so
as to exactly cancel the logarithmic term in (2.1c) of chapter 2) and = 1, Joseph &
Munson (1970) have shown that subcritical instability cannot occur, and have argued
that this result should carry over approximately to the case in which 6= 1. Based on
what is known about subcritical instability in related flows (e.g., circular Poiseuille flow),
we conjecture that when subcritical onset is possible in SPF at high Re, the amplitude
values for all Re > 0. They attributed this apparent delay in transition to either an
insufficient axial development length for the vortical structures, or to reduced axial flow
associated with vortical structures when dTa crit /dRe < 0 and the flow is driven by a
53
constant-head pump. They suggested that the linear stability analysis appears to be valid
for Re values in excess of 100. Note that Ta expt comp
crit exceeding Ta crit cannot be regarded, by
delay might be the dominant effect, with subcritical instability becoming observed only
at larger aspect ratios. On the other hand, when Ta expt comp
crit and Ta crit are nearly identical
over a range of Re, the evidence for absence of subcritical instability is much stronger.
For = 0, comparison of the experimental results of Kaye & Elgar (1958) and
Becker & Kaye (1962) to the current computations shows that agreement is generally very
good at small Re. As Re increases, the experimental values of Ta crit initially exceed the
computed values, apparently due to insufficient axial length for development of detectable
secondary flow. (The other mechanism for apparent transition delay suggested by Takeuchi
& Jankowski (1981), namely that use of a constant-head pump might contribute to Ta expt
crit
exceeding Ta comp
crit when dTa crit /dRe < 0 (e.g., for = 0.2, 0, and 0.5 with = 0.5; see
Takeuchi & Jankowski and chapter 2), is not applicable to the experiments of Kaye &
Elgar or Becker & Kaye (1962), for which current results show that dTa crit /dRe 0 for
It seems likely that, over some range of Re, there is competition between the effects of
insufficient development length (which tends to increase the apparent Ta crit above values
predicted by linear theory), and finite-amplitude instability. At sufficiently large Re, finite-
The results of Kaye & Elgar shows that at sufficiently high Re, subcritical instability
overwhelms the mechanism(s) responsible for apparent delay of transition, and leads to
instability (at all Ta) at Re values close to those associated with subcritical instability of
54
Poiseuille flow (corresponding to 1 with no rotation) near Re = 1000 (Carlson,
Widnall & Peeters 1982). Linear analysis would predict that Re 5772 (Orszag 1971) as
1, consistent with the monotonic decrease of the computed values of Re AP as 1
(figure 3.6). The results of Kaye & Elgar, showing that Ta crit = 0 near Re = 1000 for
= 0.734 and Re = 900 for = 0.820, are consistent with values of Re at which finite-
amplitude instability can be expected to set in for annular Poiseuille flow. Extrapolation
of the Ta crit values of Yamada (1961, 1962) to Ta crit = 0 gives a value of Re near 1200,
is much less than the 1 plane Poiseuille nonrotating limit, and comparable to the
value for onset of finite-amplitude stability in that limit. This strongly suggests that
finite-amplitude instability occurred in Yamadas experiments.
On the other hand, the essentially perfect agreement between current computations
and the experimental data of Sorour & Coney (1979) for = 0 and = 0.955 in an annulus
of high aspect ratio (> 570) convincingly shows that in their experiments, finite-amplitude
instability did not occur and the length for development of detectable disturbances was
sufficient for Re up to 325. Since the disturbance level in the experiments of Sorour &
Coney, particularly at the entry to the rotating test section, would seem to have been
significant, the degree of agreement strongly suggests that either subcritical instability
does not occur for Re < 325, or if it does, the amplitude threshold is quite high or the
range of subcritical Ta is very small. The degree of agreement between experiment and
linear theory provides evidence for the absence of subcritical instability that would seem
to be stronger than any previously available for SPF, and quite likely for any open flow.
For spanning a range of negative and positive values, agreement between the data of
Snyder (1965) and Mavec (1973) and current computations is excellent over a wide range
of Re and . However, even for small Re, there are values of Re and extreme values of
for which systematic deviation occurs. Here, we discuss the implications of the agreement,
as well as of the discrepancies, for interpretation of the experimental data.
55
Based on the generally excellent agreement found between Mavecs experiments and
current computations at essentially all "intermediate" values of (0.70 < < 0.35) at
seven Re values in the range 24 Re 106, we conclude that in this regime, transition-
delaying mechanisms were unimportant, and that either no finite-amplitude instability
occurred, or if it did, the Taylor number at onset was only very slightly less than the linear
critical value.
For Mavecs data at Re = 49 and 63.5, comparison of the mean and rms values of
= Ta comp expt
crit Ta crit to its variance strongly suggests the presence of systematic errors,
which while small, are considerably larger than at other Re in the range 24 Re 106. It
is conjectured, from the systematically positive values of at these two Reynolds numbers
and from the stability boundaries shown in figure 3.5, that the actual Re was somewhat
lower than the reported Re for these two cases.
in at values of Ta crit lying progressively below the predictions of linear theory as and Re
increase. Of the eight combinations of Re 134.75 and for which the experimental and
computed values of Ta crit agree within 2%, four are at the lowest Re (134.75). At least
two of the remaining four appear to correspond to values of at which the experimental
and computational results "cross over," corresponding to a for which the competing
effects of subcritical instability and insufficient development length nearly cancel. This
contrasts to the situation at Re = 106, for which experimental and computed values of
Ta crit agree within 2% at nine consecutive values of .
Figure 3.6a shows a map of the experimental values of and Re considered by Mavec,
with each symbol corresponding to the assignment of the nature of the transition at that
point. Note that there is not a one-to-one correspondence between points for which linear
56
theory underpredicts the experimental value of Ta crit ( < 0) and those identified as
"apparent delay of transition," or between those for which > 0 and those identified
as "subcritical onset." Rather, we have evaluated and its mean, variance, and rms,
and the dependence of these quantities on Re and , and on this basis, characterized the
transition at each point. This characterization is somewhat subjective. For example, for
Re = 49 and 63.5, we characterize the transition as "linear onset" over a broad range
of , since the small variance of compared to its mean and rms values suggests that
the consistently positive values of are associated with small errors in Re rather than
subcritical onset. For several values of Re, it appears (as discussed above) that finite-
amplitude instability and the effects that lead to apparent delay of transition compete
with each other, giving values at consecutive values of with opposite signs, and with
magnitudes considerably greater than the low-variance of in the low- and intermediate
instability was not observed for Re < 106, and that at higher Re, subcritical instability
occurred only at sufficiently positive values of , with the required value of decreasing
with Re. On the other hand, the map suggests that apparent delay of transition occurred at
sufficiently large values of even for small Re. At larger Re, competition between apparent
transition delay and finite-amplitude instability seems to have occurred at rotation rate
ratios intermediate between the large values at which onset is subcritical and the smaller
values at which linear theory accurately predicts onset. For Re 244, this transition
map suggests that linear onset was not observed in Mavecs experiments, with onset being
subcritical for sufficiently positive , and being apparently delayed for smaller .
57
In general, increasing at fixed Re leads to a transition from delayed onset to linear
onset, to competition between delay and subcritical instability, and finally to subcritical
instability.
In Snyders experiments with 0.95, the extremely small variance of for some
Reynolds numbers suggests that the difference between the computed and experimental
values of Ta crit for these Re values is largely due to small systematic errors. In particular,
we note the 4% variance of the already small rms values of for Re = 5 and 10 As for
similar cases in the comparison of Mavecs data to computation, we conjecture that this
is associated with a (small) difference between the reported Re and the actual value, as
discussed above for Mavecs data at Re = 49 and 63.5. Since is consistently negative
for Re = 5 and 10, we conjecture that the actual Re was larger than the reported value in
these cases.
Figure 3.6b shows a transition map for the experiments of Snyder (1965). Compared
to results for the experiments of Mavec shown in Figure 3.6a, probably the most dramatic
difference is the wider range of Re and over which subcritical instability does not occur.
The transition map indicates that subcritical instability was not observed in Snyders
experiments for Re < 120, except at one point (Re = 80, = 1.91), where assumes
the most negative value in any of the experiments of Mavec or Snyder. For larger Re,
subcritical effects appear to be manifested in a range of whose width increases with
increasing Re. With the exception of the Re = 80, = 1.91 point, increasing at
fixed Re leads to a change from linear to subcritical onset (when Re is high enough), to
a competition between apparent transition delay mechanisms and subcritical effects, and
ultimately to significant apparent delay of transition.
58
Figures 3.6a and 3.6b show a broad qualitative similarity in the regions of the -Re
plane in which linear onset of instability in SPF occurred in the experiments of Mavec
and Snyder. In both cases, onset is closely predicted by linear theory up to at least
Re = 166 (200 for Snyders experiments) over a significant range of . For Mavecs = 0.77
onset appears to be linear for a somewhat wider range of negative in the experiments of
Snyder than those of Mavec. In Mavecs experiments, subcritical instability occurs over a
progressively broader range of positive as Re increases. In contrast, in the experiments
of Snyder for a larger radius ratio and larger aspect ratios, departures from linear onset at
over a smaller range of Re and , or has a higher amplitude threshold, or occurs over a
narrower range of Ta lying below the Ta crit of linear theory) in the higher radius ratio
experiments of Snyder.
high-Re instability of annular Poiseuille flow had been made prior to the present work only
for = 0 and = 0.95, subject to the limitation that the disturbances were axisymmetric
(Ng & Turner 1982). As shown in 3.2.2, for = 0 and = 0.95 the transition occurs
59
TS-like instability of nonrotating annular Poiseuille flow at Ta = 0 and Re = Re AP occurs
over a very small range of Re for each combination of and considered.
For each combination of and , mcrit decreases from its value on the high-Re plateau
to 2 as one passes through Re . Beyond Re , mcrit is nonincreasing, and at Re AP assumes
the values of 1 and 0 for = 0.77, and 0.95, respectively. Note that the nonzero value of
mcrit at Re AP for = 0.77 is at variance with the expectation that the critical disturbance
is axisymmetric as Re approaches Re AP (Ng & Turner 1982, p. 101).
is progressively reduced as 1, with Re 0 = 24, 16, and 8 for = 0.5, 0.77, and 0.95,
respectively. This suggests that in the narrow-gap limit, the initial range of Re for which
instability sets in as axisymmetric Taylor-like vortices propagating downstream will be
small. There is no prior theoretical work accounting for nonaxisymmetric disturbances for
6= 0 in the narrow-gap limit other than that of Chung & Astill (1977), errors in which
have been discussed by Takeuchi & Jankowski and Ng & Turner.
The only experimental work for 6= 0 in the narrow-gap limit is that of Snyder (1965),
for 0.95 over the ranges 0 Re 200 and 2 0.92, with the upper bound
on being approximately the value beyond which the Couette flow without axial flow is
linearly stable (Synge 1938). For = 0.2, it is evident from points near the = 0.2 line in
Snyders figure 1 that Ta crit increases monotonically with Re over the range investigated.
This is consistent with the trend suggested by current results at = 0.5 and 0.77 (figure
2.2a of chapter 2 and figure 3.2a, respectively), where we see that as increases, a) there
is an increase in the Re beyond which axial shear destabilizes the base flow with respect
to centrifugal instability, b) the magnitude of that destabilization decreases, and c) the
plateau begins at higher Re. Together with current results for = 0.5 and 0.77, the
results of Snyder thus suggest that at least for = 0.2, axial flow does not reduce the
critical Ta for onset of centrifugal instability in the limit 1.
60
For the Taylor number definition used by Takeuchi & Jankowski and in the present
work, Ta crit vanishes for Re = 0 as 1. Figure 3.7 shows the = 0 results of 3.2
plotted in terms of a modified Taylor number defined by
1/2
= Ta
Ta , (3.4.1)
1
the critical value of which approaches 1707.762 . . . as 1 for Re = 0 and = 0. It
grows without bound as 1. Note that Re AP
appears that the plateau value of Ta
approaches the plane Poiseuille limit Re = 5772 from above as 1. Finally, we note
that mcrit = 0 at Re AP for = 0.95 is consistent with the expected behavior as 1,
based on the two-dimensionality of the critical TS disturbance in plane Poiseuille flow.
3.5. Conclusions
The complete linear stability boundaries for spiral Poiseuille flow show that for
= 0.77 and 0.95 and each rotation rate ratio < 2 considered, we can connect the onset
of instability for circular Couette flow with no mean axial flow to the onset of instability in
annular Poiseuille flow absent rotation. For > 2 , no instability is possible on a linear
basis for 0 Re < Re min , where Re min denotes the location of a turning point beyond
which the stability boundary is multivalued and the flow is stable in two disjoint ranges
of Taylor number. For < 2 , an axial pressure gradient stabilizes the flow with respect
to centrifugal instability up to high Reynolds numbers, as shown by Ng & Turner. In
each case, we find that Ta crit reaches a plateau before falling precipitously to zero as Re
approaches Re AP , the critical Reynolds number for annular Poiseuille flow. The transition
from centrifugal instability at small Re to a shear instability of Tollmien-Schlichting type
occurs at Re (slightly smaller than Re AP ), at which the critical azimuthal wavenumber
drops from its value on the high-Re plateau to mcrit = 2 in each case considered.
Comparison to experimental results for = 0 suggests that for each and aspect
ratio, there is a range of Re for which finite-amplitude instability does not occur, and for
61
which the annulus is long enough to allow development of secondary flow to detectable
amplitudes. Data for a narrow-gap ( = 0.955) annulus of large aspect ratio (> 570), for
which agreement between experiment and computation is essentially exact up to Re = 325,
strongly suggest that finite-amplitude instability does not occur over a wide range of Re
when the outer cylinder is fixed. Similar comparison to 6= 0 data for = 0.77 and
0.95, both up to Re > 100, suggests in both cases the existence of a substantial range
of Re and in which finite-amplitude instability does not occur. That computed values
of Ta crit lie below experimental values at higher Re, especially when is very positive or
negative, is taken as evidence of insufficient development length for the secondary flow in
the experimental facilities used.
62
Chapter 4
Linear Stability of Spiral Poiseuille Flow for
Small Radius Ratio
4.1. Introduction
The stability of spiral Poiseuille flow (SPF) between coaxial cylinders, driven by
rotation of one or both cylinders and an axial pressure gradient, has been of interest
for many years (Takeuchi & Jankowski 1981; Ng & Turner 1982; Meseguer & Marques
2002). In chapters 2 and 3, complete linear stability boundaries have been presented for
several radius ratios and rotation rate ratios, identified the connection between centrifugal
The "small-" case (as distinguished from the so-called "large-gap" or "wide-gap"
= 0.5 case) is of significant potential interest in applications, since the stable range
Ta 1 Ta Ta 2 for steady axisymmetric Taylor vortex flow (with no mean axial flow)
increases as decreases (Debler, Funer & Schaaf 1969; Snyder 1970; DiPrima, Eagles &
Ng 1984), with Ta 2 /Ta 1 seeming to be between 10 and 100 for = 0.5, compared to much
smaller values in the narrow-gap ( 1) limit. Here, Ri /Ro is the radius ratio,
Ta i (Ro Ri )2 / is the Taylor number, Ri and Ro are the radii of the inner and
outer cylinders, respectively, i and are the angular speed of the inner cylinder and the
kinematic viscosity, respectively, and Ta 1 and Ta 2 are the critical values at which steady
Couette flow and steady Taylor vortex flow, respectively, lose their stability. If axially-
propagating axisymmetric Taylor-like vortices remain stable in a large range of Ta when
a small mean axial pressure gradient is superimposed, then flow in a wide-gap or small-
annulus driven by cylinder rotation and an axial pressure gradient will be of interest
where laminar heat or mass transfer rates in excess of the diffusive rate associated with
63
Couette flow are desired. The laminar nature of the flow in mixing applications is especially
important in a number of biomedical and biotechnology processes, where turbulent shear
is associated with cell damage (Strong & Carlucci 1976; Resende et al. 2001).
To date, the only published results for the stability of SPF with < 0.5 appear to be
those of Chung & Astill (1977), Hasoon & Martin (1977), and Recktenwald et al. (1993),
all of which are restricted to o /i = 0, where o is the angular speed of the outer
cylinder. Chung & Astill graphically showed critical values of Ta for Re = 0, 50, and
100 at = 0.1 and 0.25, and for Re = 150 at = 0.25. (Here, Re V Z (Ro Ri )/
is the Reynolds number, where V Z is the mean axial speed, and have converted other
authors Reynolds numbers to values based on the definition used here.) As discussed in
4.3.3, their numerical results are incorrect. In the same year, Hasoon & Martin (1977)
reported computations of the critical Taylor number up to Re = 1000. They used an
axial velocity profile uniform across the annular cross-section, which approximation was
said to be validated by comparison to results at = 0.9 for the correct axial profile. The
correctness of the results of Hasoon & Martin has been questioned by DiPrima & Pridor
(1979), who have also identified an error in the governing equations used by the former
authors. In the more recent work of Recktenwald et al. (1993), = 0.1 was one of the
radius ratios for which the stability of SPF with respect to axisymmetric disturbances was
investigated numerically over the range 0 Re 20, and for which a quadratic function
of Re 2 was fitted to the results.
Here, the SPF stability computations of chapters 2 and 3 have been extended to
= 0.1. For the five values of investigated, the Reynolds number range extends more
than two orders of magnitude beyond the largest Re considered in the = 0 analysis of
Recktenwald et al. (1993). The results differ qualitatively from those at small Re, and
from those at larger . In particular, for both positive and negative values of , regions of
Re for which closed neutral curves give rise to two disjoint ranges of stable Ta are found.
64
Also, unlike the larger- cases previously studied, there is no transition from centrifugal
instability to a Tollmien-Schlichting-like instability at high Re.
The chapter is organized as follows. The results are presented in 4.2, followed by
a discussion in 4.3 of the direction of disturbance wave propagation, implications for
experiment, and the relationship to other work. Some conclusions are presented in 4.4.
4.2. Results
Basic code validation was accomplished by comparison to previous tabulated results
for 0.5, as described in chapter 2. Comparison of computations at smaller to the
graphical results of Mahadevan & Lilley (1977) and Garg (1980) for annular Poiseuille flow
(Ta = 0), and to the tabulated small-Re results of Recktenwald et al. (1993) (see 4.3.3)
revealed excellent agreement. In contrast to the 40 terms that always provided adequate
resolution convergence for larger (chapter 2), up to 70 terms were sometimes needed to
and 0.2) have been presented for = 0.5 by Takeuchi & Jankowski (1981) and in chapter
2, and allow the effects of smaller radius ratio to be clearly identified. The other two
counter-rotating cases ( = 1 and 0.25) provide additional information on the effects
of rotation rate ratio on the stability of counter-rotating flow.
For = 0, critical values of the Taylor number Ta, azimuthal wavenumber m, axial
wavenumber k, and wavespeed c are shown in figures 4.1a, 4.1b, 4.1c, and 4.1d, respectively.
Figure 4.1a shows that the stability boundary is a single-valued function of Re (i.e., there
is a single critical Ta for each Re). For all combinations of and satisfying < 2 that
were considered in chapter 2, the stability boundary was single-valued for 0 Re Re AP ,
beyond which the flow was unstable at all Ta. The Re = 0 intercept of the stability
65
boundary at Ta crit = 1264.43 corresponds to the onset of Taylor vortex flow. As Re
increases, Ta crit increases until reaching a global maximum (near Re = 46). As for larger
, scalloping occurs due to integer jumps in mcrit (Takeuchi & Jankowski 1981; Ng &
Turner 1982), with the pronounced slope discontinuities near Re = 46 (point B) and 87
2 and 3), for which Ta crit decreases sharply at Re , where a transition occurs from a
centrifugal instability to a Tollmien-Schlichting-like instability. The = 0.1 behavior is
consistent with that reported for the nonrotating (Ta = 0) annular Poiseuille case at small
by Mahadevan & Lilley (1977) and Garg (1980), who found that the critical Reynolds
number (Re AP ) increases rapidly as decreases below 0.15. For Ta = 0, table 4.1 shows
that as decreases below 0.2, Re AP increases monotonically, and that there appears to be
a critical value < 0.14 below which no critical Re AP exists. Extrapolation of linear and
quadratic least-squares polynomials fitted to the dependence of 1/Re AP for the three
and four smallest values of for which Re AP was computed gives values of = 0.1199 and
0.1178, respectively, suggesting that the critical radius ratio is about 0.12.
Figure 4.3 shows neutral curves in the k-Ta plane for several azimuthal wavenumbers.
Neutral curves for m = 1 and 2 each consist of a single "primary" neutral curve (not
displayed for m = 2, since it lies far above the Ta range shown) and one closed disconnected
neutral curve (CDNC). Each primary neutral curve has a vertical asymptote at k = 0,
with Ta tending to infinity as k , while the closed neutral curves are said to be
disconnected in the sense that they are not connected to other neutral curves at bifurcation
points (Pearlstein 1981; Pearlstein, Harris & Terrones 1989). For each other value of m,
there is only a primary neutral curve. In contrast to results for larger (cf. figure 4 of Ng
& Turner 1982), the primary neutral curves for some m (m = 0, 1) are the "envelopes"
66
of two intersecting branches, across each of which one temporal eigenvalue (or possibly
a pair) crosses into the right half-plane (RHP). At the intersections, the slopes of these
primary neutral curves are discontinuous. The parts of these branches not shown (i.e., the
smooth continuation of each branch through the junction) correspond to curves on which
one or more additional eigenvalues cross into the RHP, in addition to the one that crossed
at a lower Ta on the primary branch. For = 0, the existence of CDNCs in the k-Ta
plane does not lead to multiple ranges of stable Ta for fixed Re. For a CDNC to lead
to a multi-valued stability boundary, either there must be no primary neutral curve (as
for = 0.5 and > 2 in chapter 2), or there must be some range of Ta lying between
the maximum Ta on a CDNC and the minimum Ta on a primary neutral curve, through
which no other neutral curve passes. Neither situation obtains for = 0 and = 0.1.
Figure 4.1b shows that mcrit increases stepwise, from 0 for 0 Re 46, to 1 for
increasing Re over each range for which mcrit is a constant. This behavior leads to the
three-part "wavenumber fan" shown. To an excellent degree of approximation, k crit varies
inversely with Re over the range 250 Re 105 .
Figure 4.1d shows that the dimensionless wavespeed of the critical disturbance,
ccrit crit /kcrit , which is the ratio of the dimensionless phase velocity to the mean axial
speed V Z , is nearly constant over the range of Re for which mcrit = 0. Near Re = 46,
where mcrit increases from 0 to 1, ccrit jumps by about 50%, and then falls over the range
where mcrit = 1. It then increases discontinuously again near Re = 87, where mcrit jumps
67
4.2.2. Counter-rotating cylinders ( < 0)
For = 0.25, 0.5, and 1, the results differ qualitatively from those for = 0
(4.2.1), and from those for = 0.5 at 0.5 (Takeuchi & Jankowski 1981; chapters
2 and 3). The three values of are discussed in order, with a detailed description of the
neutral curves being given for = 0.5, the case studied at larger .
= 0.25. For = 0.25, figure 4.4a shows that over the range 0 Re 94, Ta crit
is single-valued and decreases monotonically from its value of 2759.3 at Re = 0. Near
Re = 60 (point B), the slope on the high-Ta branch changes discontinuously, corresponding
to mcrit increasing from 0 to 1. More strikingly, for 94 < Re 125.5, there exist three
values of Ta crit and two disjoint ranges of stable Ta, one below the low-Ta branch and
one between the intermediate- and high-Ta branches. A second slope discontinuity occurs
where the high-Ta (low-Re) branch joins the intermediate branch near Re = 125.5 (point
A). The intermediate-Ta branch continues downward to the turning point C, where it
smoothly connects with the low-Ta (high-Re) branch at Re = 95. For large Re, Ta crit
approaches an asymptotic value (Ta crit 244).
This triple-valued stability boundary is a direct consequence of the existence of
CDNCs, previously found in stability analyses of quiescent fluid layers in which the
density depends on two or more stratifying agencies with different diffusivities (Pearlstein
1981; Pearlstein, Harris & Terrones 1989; Terrones & Pearlstein 1989; Lopez, Romero &
Pearlstein 1990), in a buoyancy-driven flow in an inclined layer (Chen & Pearlstein 1989),
boundaries found for larger (chapter 2), the existence of multiple values of Ta crit for some
range of Re at fixed and does not correspond to one base flow becoming unstable at
several different Taylor numbers. Rather, it should be interpreted in terms of a base flow
68
with a particular axial velocity (depending only on ) becoming unstable at two different
magnitudes of the azimuthal velocity (whose profile depends only on and ), whose
magnitude depends on Ta/Re.
Figure 4.4b shows that mcrit again increases from 0 to 1 (at point B), and from 1 to 2
(at point A) over the range of Re considered. Comparison to the = 0 case (figures 4.1a
and 4.1b) shows that these transitions occur at similar Reynolds numbers, and that the
intermediate-Ta branch AB for = 0.25 corresponds to the m = 2 branch in figure 4.1a,
which begins at A and continues to large Re. It is clear that = 0.25 is more negative
than the value of for which the mcrit = 2 branch has an infinite slope at its junction
(point A) with the mcrit = 1 branch.
Figure 4.4c shows that kcrit is again a piecewise continuous function of Re, which
in this case is triple-valued for Reynolds numbers between the turning point C and the
junction A. Unlike the behavior observed for = 0, the dependence of kcrit on Re is not
fan-like. On the high-Ta branch, terminating at A, kcrit is essentially independent of Re,
with a small discontinuity at B/B0 , where mcrit increases from 0 to 1. (The prime denotes
a second point on the mcrit , kcrit , or ccrit plots, at the same Re as the unprimed point,
Figure 4.4d and the enlargement 4e show that for = 0.25, ccrit is positive and
essentially constant on the m = 0 portion of the high-Ta branch, as in the = 0 case
considered above (figure 4.1d) and in all of the < 2 cases considered in chapter 2.
The critical wavespeed increases by approximately a factor of two at the first azimuthal
wavenumber transition (B/B0 ), and then decreases monotonically until the junction is
reached at A. At that point, ccrit jumps discontinuously to a negative value (about 32)
at A0 , corresponding to a traveling-wave disturbance propagating upstream against the
69
axial component of the base flow. As one moves down along the intermediate branch of the
stability boundary in figure 4.4a, the magnitude of ccrit rapidly decreases, corresponding to
a reduction in the speed of the backward-propagating neutral disturbance as Re decreases.
At the turning point C, ccrit = 3.59. As Re increases on the low-Ta branch, ccrit
continues to increase, and near Re = 201 smoothly passes through zero, corresponding to
a reversal in direction of the traveling-wave disturbance, and the existence of a single Re at
which the disturbance is stationary. Finally, as Re increases beyond 300, ccrit approaches
its asymptotic value of about 0.1.
(point A), where it joins the intermediate-Ta branch (I) with a slope discontinuity. The
intermediate branch continues downward, through a small slope discontinuity at point B,
to the turning point C, where it smoothly connects with the low-Ta (high-Re) branch near
Re = 83.3. Thus, in the multi-valued range 83.3 Re 172.0, the base flow is stable
below the low-Ta branch, as well as between the intermediate- and high-Ta branches.
For two values of Re in the triple-valued range, the neutral curves in figures 4.6a and
4.6b show that a gap exists between the maximum Ta on the CDNCs and the minimum
Ta on the m = 0 neutral curve, which is the lowest-lying primary neutral curve (with
slightly higher Re is joined by the m = 3 CDNC, which also makes its first appearance
at a point. At Re = 90, the gap between the minimum Ta on the m = 0 neutral curve
and the maximum Ta on the m = 2 CDNC is about 80% of the former. The flow is also
70
stable for Ta below the minimum of the m = 2 CDNC. For Re = 165, figure 4.6b shows
that several other CDNCs have appeared, and the gap between the minimum Ta on the
m = 0 neutral curve and the largest Ta on any CDNC (on m = 3) has been reduced. For
Re = 165, the minimum still occurs on the m = 2 CDNC. (For Re = 90 and 165, all
of the CDNCs and the m = 0 neutral curve are shown, and no part of any other neutral
curve lies below the minimum of the m = 0 neutral curve.)
The junction at point A corresponds to the Re (172.0) at which the gap disappears
between the maximum on the m = 3 CDNC and the minimum on the m = 0 primary
neutral curve. As one continues down the intermediate branch (I) of the stability boundary,
mcrit jumps from 3 to 2 at point B (Re = 99.5), between the values of Re for which
neutral curves are shown in figures 4.6a and 4.6b, and each CDNC shrinks and ultimately
coalesces at a point in the k-Ta plane. The last of these coalescences (for the m = 2
CDNC at Re = 83.3) corresponds to the intermediate- and low-Ta branches of the Re-Ta
stability boundary joining smoothly at the turning point C in figure 4.5a, near Ta = 480.
The behavior of the neutral curves at each end of the triple-valued range is qualitatively
identical to that found in earlier studies of the onset of buoyancy-driven convection in
horizontal fluid layers in which the density depends on two or more independent stratifying
agencies (Pearlstein 1981; Pearlstein, Harris & Terrones 1989; Terrones & Pearlstein 1989;
Lopez, Romero & Pearlstein 1990).
Figure 4.5b shows that mcrit = 0 on the high-Ta branch, jumping directly to 3 at
point A. This value of mcrit is maintained as one moves downward on the intermediate-
Ta branch to B, at which point mcrit jumps to 2, which is the critical m up to at least
Re = 105 . For = 0.5, there is no Re for which mcrit = 1. The maximum mcrit occurs
at intermediate Re values, which differs from the = 0 and 0.25 cases above and the
< 2 cases considered in chapter 2, for which mcrit increased by unit steps along the
arclength of the stability boundary.
71
Figure 4.5c shows that kcrit is a piecewise continuous function of Re, and is triple-
valued for Reynolds numbers between the turning point C near Re = 83.3 and the junction
A near Re = 172.0. Like the behavior predicted for = 0.25, the dependence of k crit on
Re is not "fan-like". On the high-Ta, m = 0 branch, terminating at A, kcrit is essentially
Figure 4.5d shows that on the high-Ta branch, ccrit is positive and essentially constant,
as in the = 0 and 0.25 cases considered above (figures 4.1d and 4.4d) and in all > 2
cases considered in chapter 2. The high-Ta branch terminates at A, and ccrit jumps
discontinuously to a negative value at A0 on the intermediate branch, as for = 0.25.
As one moves down along the intermediate branch of the stability boundary toward B in
figure 4.5a, the magnitude of ccrit rapidly decreases, corresponding to a neutral disturbance
propagating less rapidly upstream against the base flow as Re decreases. After a small
discontinuity at B (see figure 4.5e) corresponding to mcrit jumping from 3 to 2, ccrit
about 0.1.
= 1. For = 1, figure 4.7a shows that Ta crit is single-valued and decreases
monotonically over 0 Re 79 from its Re = 0 value of 9414.4. For 79 < Re < 305, there
again exist three values of Ta crit and two disjoint ranges of stable Ta, one below the low-Ta
branch and one between the intermediate- and high-Ta branches. As for = 0.25 and
0.5, a slope discontinuity occurs where the intermediate- and high-Ta branches join near
the junction at Re = 305 (point A), whereas the connection between the intermediate-
72
and low-Ta branches near Re = 79 is smooth. For large Re, Ta crit is single-valued and
approaches its asymptotic value (Ta crit 56). Note that the range of Re for which
multiple critical Taylor numbers exist is larger than for = 0.25 or 0.5.
For = 1, figure 4.7b shows that over the entire range 0 Re < 4 10 4 , there is
no Re for which mcrit = 0, unlike the = 0, 0.25, and 0.5 cases, so that there is no Re
for which the onset of instability occurs through an axisymmetric disturbance.
The dependence of kcrit on Re (figure 4.7c) differs from that for less negative values
of , since mcrit changes on the intermediate-Ta branch below the turning point. Again,
the junction is approached from the intermediate branch, ccrit decreases for = 1 (figure
4.7e) and increases for = 0.5 (figure 4.5f).
up to Re AP = 10359 (chapter 2). For = 0.2 and = 0.1, the Re = 0 Couette flow is
linearly stable according to the > 2 criterion (Synge 1938), and the stability boundary
cannot intersect the Re = 0 axis. The results differ qualitatively from those shown above
for < 2 , as well as from those for = = 0.5.
For = 0.2, figure 4.8a shows that up to about Re = 66.8, the flow is stable for all Ta,
which differs from the stability boundaries for 0 (figures 4.1a, 4.4a, 4.5a, and 4.7a), in
which for small Re, SPF is stable for only a finite range of Ta. For Re > 66.8, the stability
boundary has two branches, and there are two disjoint ranges of stable Ta, the first being
finite and lying below the low-Ta branch, with the second being semi-infinite and lying
above the high-Ta branch. As Re increases on the low-Ta branch from the turning point
73
near Re = 66.8, Ta crit monotonically decreases to its asymptotic value (Ta crit 241),
while on the high-Ta branch, Ta crit continues to increase with Re up to at least Re = 103 .
This contrasts to the single branch found for large Re when = 0.1 and 0, on which
Ta crit approaches an asymptote as Re .
Figure 4.8b shows that mcrit = 3 at the turning point and on both branches for
sufficiently small Re. On the low-Ta branch, mcrit jumps to 2 near Re = 105, and
remains unchanged at larger Re. On the high-Ta branch, mcrit falls to 4 near Re = 162,
and to 5 near Re = 622. For = 0.2, there is no value of Re for which the onset of
Figure 4.8c shows that as Re decreases on the low-Ta branch, kcrit increases like
1/Re, as for the other values considered, while on the high-Ta branch, kcrit decreases
monotonically for each range of Re for which mcrit is constant. At point B, where
mcrit jumps from 2 to 3, kcrit increases discontinuously. As Re decreases towards the
turning point, kcrit passes through a maximum, and then decreases continuously through
the turning point until mcrit jumps from 3 to 4 at point A on the high-Ta branch.
Beyond point F, kcrit decreases monotonically with Re until the discontinuity at D, where
it increases, followed by monotonic decrease to the largest Re considered. Note that the
critical wavenumbers for the high- and low-Ta branches intersect at point E near Re = 520,
and that this intersection corresponds to two different values of Ta crit and two different
values of mcrit .
For = 0.2, figures 4.8d and 4.8e show that at large Re on the low-Ta branch, c crit
is positive. As Re approaches the turning point (Re 66.8) along the mcrit = 2 low-
Ta branch, ccrit decreases, passing smoothly through zero near Re = 160. The critical
wavespeed continues to decrease as one moves through the turning point, with the fall-off
74
being nearly linear in log Re as Re increases along the high-Ta branch. The discontinuities
of ccrit at the Reynolds numbers where mcrit jumps are barely discernable on the scale of
figure 4.8d.
4.3. Discussion
4.3.1. Relationship to other work
For spiral Poiseuille flow with = 0.1, the only previous results appear to be those
of Chung & Astill (1977), Hasoon & Martin (1977), and Recktenwald et al. (1993), all for
= 0. Chung & Astill showed critical values of Ta (their figure 5) for Re = 0, 50, and 100,
and stated that mcrit = 0 for Re 100. For = 0.1, the Taylor number defined by them is
exactly one-ninth of Ta used here, and their critical Taylor number at Re = 0 corresponds
to Ta crit 3 104 , more than twenty times the value computed. At Re = 0, the value
Ta crit = 1264.43 is in excellent agreement with that of Walowit, Tsao & DiPrima (1964),
to whose results Chung & Astill made no comparison or reference. Errors in the results of
Chung & Astill at larger have been discussed by Takeuchi & Jankowski (1981) and Ng &
Turner (1982). The computations of Hasoon & Martin (1977), which predict that Ta crit
increases monotonically with Re over the range 0 Re < 1000, are in disagreement with
the results shown in figure 4.1a. How much of the discrepancy is due to use of a uniform
axial velocity profile by Hasoon & Martin, and how much is attributable to an error in
their governing equations (DiPrima & Pridor 1979), is not known.
For several values of , Recktenwald et al. used a shooting method to compute the
where Ta 0crit , a2 , and a4 were determined by a least-squares fit. As shown in 4.2.1, the
critical disturbance for = 0 and = 0.1 is indeed axisymmetric up to Re = 46. The
75
agreement between the approximation (4.3.1) and the computations of this work is better
than 6 parts per million at Re = 0, 1, 2, , 19, and 14 parts per million at Re = 20,
suggesting the correctness of both sets of results. Comparison of (4.3.1) to these numerical
results shows that the difference is less than 0.3% at Re = 40, a value twice the highest Re
at which Recktenwald et al. used direct computational results to determine the coefficients
in the fitted form (4.3.1). Good agreement also obtains between the computed values of
kcrit and the fitted form of the critical wavenumber
kcrit = 3.3393[1 (Re/133.1)2 + (Re/74.76)4 ] given by Recktenwald et al. has an rms error
300 times larger. For = 0.5, however, the functional form for kc given by Recktenwald et
al. fits the computed results very well, and is a much better fit than the analog of (4.3.2).)
The rms difference between the computed wavespeeds and the values of Recktenwald et al.
(using (4.3.2) instead of the expression given by them for kcrit ) is 2.4 104 for the same
values of Re. The range of Re considered by Recktenwald et al. was too small for the
qualitative differences between results for = 0.1 and the larger values of investigated
by those authors to be apparent.
The results shown for = 0 in Table 4.1 strongly suggest that for annular Poiseuille
flow (i.e., absent rotation), no linear instability occurs for 0 < < 0.12, consistent
with the earlier work of Mahadevan & Lilley (1977) and Garg (1980), who showed that
the critical Re increased very rapidly as approached 0.15. This shows that the apparent
linear stability of circular Poiseuille flow ( = 0) at all Re (Salwen, Cotton & Grosch
1980) is not an isolated case, and that Poiseuille flow is linearly stable for all Re from the
= 0 (circular) case up to about 0.12. Although the relationship of the stability of
narrow-gap ( 1) annular Poiseuille flow to the stability of plane Poiseuille flow has been
discussed (Mott & Joseph 1968; Mahadevan & Lilley 1977; Garg 1980; Sadeghi & Higgins
1991), and the linear stability of circular Poiseuille flow at all Re has been identified as
76
representing a "limit result" for the axial profiles as 0 in annular Poiseuille flow (Mott
& Joseph 1968), there are no previous discussions connecting the lack of a critical Re for
circular Poiseuille flow to the lack of a critical Re for small- annular Poiseuille flow over
a nonzero range of .
Finally, for = 0, the Ta versus Re plots in figure 4.2 show that the stability
boundary for = 0.1 has a high-Re asymptote very close to the plateau values for the
larger- cases considered in chapters 2 and 3. Like the = 0.5 case, the = 0.1 stability
boundary has a global maximum at an Re at which transition occurs between critical
values of m. Unlike the = 0.5 case, the results for = 0.1 show that Ta crit on the high-
Re plateau lies below the Re = 0 value for Taylor-Couette flow. That the Ta scaling in
figure 4.2 nearly "collapses" the plateau behavior for 0.1 0.95 shows that for a wide
range of Re and , the critical angular velocity is given by crit = 44.7/Ri Ro , accurate
within 3%.
stability analysis for 0, and for all considered for 0.5 in chapter 2, figures
4.4e, 4.6f, and 4.8e show that for each counter-rotating = 0.1 case considered, c crit
passes smoothly through zero on the intermediate-Ta branch. Thus, for = 0.1, the
assertion of Goldstein is consistent with the linear stability analysis everywhere except at
the single Re at which ccrit passes through zero. Linear stability theory thus predicts
that there is a single Reynolds number Re s for which one can decrease Ta from the
stable range between the intermediate- and high-Ta branches, and transition from a
steady axisymmetric z-invariant SPF base flow to more complicated flow through a steady,
nonaxisymmetric, axially-periodic disturbance. For Re slightly less than Re s , the direction
of propagation of the disturbance flow would be upstream against the axial component
77
of the base flow, while for Re slightly greater than Re s , the disturbance structure will
propagate downstream. Note that for = 0 and = 0.8, the experimental work of Buhler
& Polifke (1990) shows that the direction of propagation of axially traveling waves with
m = 1 can be reversed by changing Re.
The prediction of a single Re at which the onset frequency is zero for neutral
disturbances of infinitesimal amplitude with = 0.1 can be contrasted to the existence of a
range of conditions in which experiment shows that steady helical vortices exist for larger
(Buhler & Polifke 1990; Lueptow, Docter & Min 1992; Tsameret & Steinberg 1994).
Also, note that for six small values of Re (0.11 Re 1.15) and = 0.677, figure 16 of
Giordano et al. (1998) shows that the dimensionless axial phase velocity, V phase /VZ (which
is the ccrit used here at Ta = Ta crit ), decreases nearly linearly to zero with increasing Ta
in some range above Ta crit . Results for at least one Reynolds number suggest that the
fixed. As shown by the latter authors, the results of Meseguer & Marques (2002) are
fundamentally incorrect, due to restriction of the disturbances to an insufficient range of
azimuthal wavenumbers. For fixed values of 6= 1, the multi-valued stability boundaries
found for = 0.5 (chapter 2) are double-valued in the Re-Ta plane for Re > Re min > 0.
For 0 Re < Re min , SPF is linearly stable for all Ta, consistent with stability at Re = 0
according to the Rayleigh criterion. As Re increases through the turning point at Re min ,
closed neutral curves emerge as points in the k-Ta plane. The work of chapter 2 shows
that for = 0.5 and several rotation rate ratios > 2 , SPF is unstable in the range
between the lowest Re on any of these neutral curves and the highest Re on any of them,
and linearly stable for Re lying above or below this range. For the cases considered, the
78
upper and lower limits of the unstable range corresponded to negative and positive values
of the azimuthal wavenumber m, respectively.
For = 0.1 and = 0.2 > 2 , the multi-valued stability boundary differs
from that found in chapter 2 in that there is no finite Re at which a transition from
that the stability boundaries extend over the entire range of Re, and are triple-valued in
a finite range of Re. In that range of Re, SPF is stable below the lowest critical Ta and
between the intermediate and highest critical Ta, and unstable for all other Ta.
One is thus led to consider experiments to confirm the existence of the multi-valued
stability boundaries predicted for = 0.1 and < 0, with the goal of confirming the
two disjoint ranges of stable Ta and the two disjoint ranges of unstable Ta predicted
over a range of Re. For SPF with = 0.1, figures 4.4a, 4.5a, and 4.7a show that for
each counter-rotating case considered, there are three critical values of Ta for Re 1 ()
Re Re 2 (), where Re 1 and Re 2 denote the minimum and maximum values of Re for
which multiplicity is predicted. The values of Ta on the low-, intermediate-, and high-Ta
branches are denoted by Ta L , Ta I , and Ta H , respectively. As discussed in 4.2.2, the
analysis predicts that in the multi-valued range of Re, SPF is linearly stable for Ta Ta L
79
been seen that 0.25 < < 0 for = 0.1. On the other hand, from the work of Synge
(1938) it is known that the Re = 0 base flow is linearly stable for > 2 = 0.01, which
strongly suggests that at Re = 0, Ta crit grows without bound as 2 from below.
For < , the stability boundaries shown in figures 4.4a, 4.5a, and 4.7a suggest that
f Re 2 (), the three critical values of Ta might be found experimentally
for Re 1 () Re
as follows. The lowest value, Ta L , might be reached by increasing Re at Ta = 0 until
f and then increasing Ta at fixed Re until Ta L is reached. The intermediate
reaching Re,
and upper Taylor numbers, Ta I and Ta H , respectively, might be reached by starting at a
f Then
Taylor number in the range Ta I Ta Ta H at Re = 0, and increasing Re to Re.
Ta could be increased until Ta L is reached, or decreased until Ta I is reached.
One key issue in determining if SPF becomes unstable as predicted by linear analysis
in the range of multi-valuedness is whether instability can set in through disturbances
of finite amplitude in one or both ranges of stable Ta. First, consider < and the
possibility of reaching Ta L . For the cases considered, the computed values of Re 2 (all
less than about 300) are not very large, so that there is good reason to believe that the
nonrotating annular Poiseuille flow is either globally stable, or at least stable with respect
to a large class of finite-amplitude disturbances. This expectation is based on the fact that
the narrow-gap limit of annular Poiseuille flow ( 1), plane Poiseuille flow, is globally
stable up to about 1000 (Carlson, Widnall & Peeters 1982), while for = 0, global stability
obtains for Re up to about 2000. This suggests that it should be possible to achieve the
only in a very narrow range of stable Ta just below the linear Ta crit . These comparisons
are consistent with the conclusion of Takeuchi & Jankowski (1981) that finite-amplitude
instability does not occur for Re 40 when = 0 and = 0.5.
80
For = 0.1 and each negative considered, the analysis here predicts that SPF is
linearly stable for 0 Ta < Ta
crit as Re . Computations reported in 4.2.1 strongly
suggest that this will be true for < 0.12. Thus, for < , there would appear to be
an - and -dependent Ta
crit such that SPF is linearly stable for all Re if Ta < Ta crit , and
axial shear apparently has no effect on the onset of centrifugal instability via infinitesimal
disturbances. Moreover, figure 4.2 strongly suggests that the value of Re at which the
transition from centrifugal to Tollmien-Schlichting-like instability occurs (see chapter 2)
originates with the mcrit = 2 branch assuming an infinite slope at point A, where the
transition from mcrit = 1 to mcrit = 2 occurs as Re increases. As one approaches from
below the rotation rate ratio at which multi-valuedness sets in, the slope of the mcrit = 2
branch must become infinite and the width of the range of multiplicity must vanish (i.e.,
4.4. Conclusions
The linear stability of spiral Poiseuille flow is quite different for = 0.1 than for
0.5. One key difference is the absence of a transition from centrifugal to Tollmien-
Schlichting-like instability at high Re. For = 0.1, there is no critical Re beyond which
the nonrotating flow is unstable. Like circular Poiseuille flow, annular Poiseuille flow for
= 0.1 is linearly stable for all Re, so that at each rotation rate ratio, SPF is linearly
stable below an asymptotic value of Ta crit as Re .
In addition, for = 0.1 and each rotation rate ratio considered, a range of Re is found
for which closed disconnected neutral curves exist in the k-Ta plane. For each negative ,
these neutral curves give rise to a triple-valued Re-Ta stability boundary over some finite
81
range of Re, corresponding to two disjoint ranges of stable rotation rates and two disjoint
ranges of unstable rotation rates. For > 2 , there is a finite range of Re beginning at
0 in which the flow is linearly stable for all Ta, and a semi-infinite range of Re in which
a double-valued stability boundary separates two disjoint ranges of stable rotation rates.
For each case considered in which the outer cylinder rotates, there is a single nonzero Re at
which the axial wavespeed of the critical disturbance vanishes, corresponding to a reversal
of the axial direction of the propagating disturbance.
82
Chapter 5
Effect of Axially-Periodic Inner-Cylinder Radius on
Steady Axisymmetric Couette and Taylor-Couette Flow
5.1. Introduction
Flow between an inner rotating cylinder and a concentric fixed outer cylinder has
been of interest for many years. Several comprehensive reviews are available (cf. Chossat
& Iooss 1994). One of the simplest geometric perturbations of the "standard" flow is
when one of the axisymmetric boundaries is allowed to have an axially-periodic radius.
This problem is of interest in several contexts.
First, flow between a circular cylinder and an inner shaft with an axially-periodic
radius rotating about their common axis is of direct interest in the design and operation of
labyrinth seals. Experimental and computational work in this area (cf. Stoff 1980; Rhode
et al. 1986; Yucel & Kazakia 2001; Eser 2002, and references cited therein) has focused on
the effects of rotation on the relationship between the axial pressure drop and the mean
axial flow rate, largely in the turbulent regime.
Second, wavelength selection is important in a number of spatially periodic flows
in which no extrinsic wavelength is imposed. Spatial periodicity can either arise from
1978b; Benjamin & Mullin 1981, 1982; Mullin, 1982; Ikeda & Maxworthy 1994) for such
selection. These situations include cases in which chemical reaction occurs either in the
bulk of the fluid, with axial mass transfer being largely due to convective transport from
83
one toroidal vortex to its counter-rotating neighbor (Swinney et al. 1990; Iosilevskii et al.
1993; Moore & Cooney 1995 and references cited therein), or heterogeneously at one of
the cylinders (Eisenberg, Tobias & Wilke 1954; Morrison, Striebel & Ross 1986; Madore
et al. 1992; Gabe et al. 1998; Podlaha 2001; Gao, Scheeline & Pearlstein 2002) at the
University of Illinois. The importance of controlling vortex size in these flows is due to
the strong dependence of mass transfer rates on the velocity field. Maintaining vortex
size independent of end effects is especially critical in a free-surface Taylor-Couette reactor
recently developed and demonstrated (Gao, Scheeline & Pearlstein 2002) at the university
of Illinois. Using an axial variation of radius to select wavelength has the added advantage
that the phase can be prescribed independent of end conditions. Note that in Rayleigh-
Benard convection, application of a horizontally-periodic temperature on one bounding
wall leads to a situation in which the phase is determined by the flow (Kelly & Pal, 1978).
a constant-radius shaft provides no mixing beyond diffusion. This, and the absence of
sharp edges, may make this approach attractive for applications where cell damage must
be avoided.
84
For small-amplitude radius variations, one might expect the following results.
a) At small angular velocities, one would expect the flow to be generally similar to Couette
flow except for the loss of axial invariance and the appearance of nonzero axial and
radial velocity components, whose axial periodicity is expected to be determined by
experimental work by Koschmieder (1975), Rotz & Suh (1976), Ikeda & Maxworthy (1994),
Painter and Behringer (1995, 1996, 1997, 1998a,b), Zimmermann, Painter & Behringer
(1998a-c), Smits, Auvity & Sinha (2000), and Staples & Smits (2001). (Related work on
the effect of localized radius variation has been reported by Eagles & Eames (1983), Riecke
(1988), and Ning, Ahlers & Cannell (1990).) The most detailed of these investigations are
those of Koschmieder, Rotz & Suh, and Ikeda & Maxworthy.
Koschmieder (1975) performed flow visualization experiments using aluminum flakes
in the space between a fixed outer circular cylinder and a coaxially rotating inner cylinder
fitted with a set of O rings placed along its axis. The ratio of cylinder radii was 0.727.
(Note that the inner and outer cylinders of Koschmieders apparatus had radii of 4.572 and
6.285 cm, respectively, rather than the diameters of 4.527 and 6.285 cm given in his paper)
The O-ring thickness was 3 mm, and they were placed along the axis in a nominally periodic
fashion. The aspect ratio (test section length divided by cylinder gap) was 51. At the
largest dimensionless wavelength (O-ring spacing divided by the gap between cylinders)
of 4.09, vortices were discernible along the entire column at rotation rates as low as 3%
85
of the critical value for Taylor-Couette flow (Ta T C ) at this value of . "Sinks," at which
flow near the outer cylinder was radially inward, were separated by one wavelength, and
were located one-quarter wavelength below each O ring. "Sources," at which flow near
the outer cylinder was radially outward, were not visually distinct. At higher rotation
rates the number of vortices doubled, with the newly formed pairs being initially smaller
and weaker than the larger pairs formed earlier. At 48% of the critical rotation rate, the
ratio of axial extents was 1.53, while at the critical rotation rate, the vortex sizes were
indistinguishable. The vortices remained axisymmetric with the same wavelength up to
about 20 times the critical rotation rate, at which point four additional pairs formed at
axial locations where errors in O-ring mounting made single-pair vortices too long to be
stable. For dimensionless wavelengths ranging from 1.98 to 2.51, Koschmieder found only
a single pair of vortices well into the supercritical regime.
Maxworthy & Ikeda (1994) performed a flow visualization study using guanine flakes
in a system where the shaft radius varied axially according to
where and are the amplitude and axial wavelength, respectively, and Ri is the mean
inner radius. The radius ratio Ri /Ro was 0.959, and the dimensionless amplitude and
wavelength of the radius variation were /(Ro Ri ) = 0.11 and /(Ro Ri ) = 2.8,
respectively, with an aspect ratio of 87, where Ro is the cylinder radius. The shaft and
wavelength based on the the mean gap. (Note that Ikeda & Maxworthys statement (p.
5219) that the wavelength "was 126% of twice the minimum gap" should instead refer to
the maximum gap.) When either i or o was zero, only one axisymmetric flow was
86
observed. If both cylinders were rotated, hystereis was observed and the critical conditions
depended on which cylinder underwent a quasi-steady speed change while the other cylinder
rotated at constant speed.
Here, a computational investigation of the flow driven by a rotating shaft with the
harmonic axial radius variation (5.1.1) is presented, focusing on the multiplicity and
bifurcation of the solutions as a function of Taylor number for several different amplitudes.
The ratio of the mean radius of the shaft to the outer-cylinder radius is held constant at
0.5, corresponding to the "wide-gap" case, in which axisymmetric Taylor vortices in the
standard case persist to at least ten times the first critical Taylor number for a constant-
radius inner cylinder (Debler, Funer & Schaaf 1969; DiPrima, Eagles & Ng 1984; DiPrima
& Swinney 1985).
The paper is organized as follows. In 5.2, the formulation for the full Navier-
5.2. Formulation
The flow is driven by an axisymmetric shaft with mean radius R i rotating steadily
and coaxially with respect to a fixed outer circular cylinder of radius Ro . The radius of
the shaft is taken to vary harmonically in the axial direction according to (5.1.1). The
constant angular speed of the inner cylinder is denoted by i .
Axisymmetric steady flows in this geometry are determined by solving the full Navier-
VR VR VZ
+ + =0 (5.2.1a)
R R Z
!
VR VR V2 P 2 VR 1 VR VR 2 VR
VR + VZ = + + + (5.2.1b)
R Z R R R2 R R R2 Z 2
2
V V VR V V 1 V V 2 V
VR + VZ + = + 2+ (5.2.1c)
R Z R R2 R R R Z 2
87
2
VZ VZ P VZ 1 VZ 2 VZ
VR + VZ = + + + . (5.2.1d)
R Z Z R2 R R Z 2
No-slip boundary conditions are enforced on the rigid impermeable shaft and cylinder
The flow is assumed to be axially periodic with the following inflow and outflow boundary
conditions
V(R, 0) = V(R, ) (5.2.3a)
V(R, 0) V(R, )
= (5.2.3b)
Z Z
P (R, 0) = P (R, ) . (5.2.3c)
the dimensional coordinates R and Z with Ro , while the velocity and pressure are
2
nondimensionalized using i Ri and i2 Ri , respectively. The system (5.2.1) can be
written in dimensionless form as
vr vr vz
+ + =0 (5.2.4a)
r r z
vr vr v2 p (1 )2 2 vr
vr + vz = + vr 2 (5.2.4b)
r z r r Ta r
v v vr v (1 )2 2 v
vr + vz + = v 2 (5.2.4c)
r z r Ta r
vz vz p (1 )2 2
vr + vz = + vz , (5.2.4d)
r z z Ta
where
21 2
= (r ) + 2 , (5.2.5)
r r r z
and the Taylor number is Ta = i (Ro Ri )2 /, where is the kinematic viscosity.
88
The boundary conditions (5.2.2) become
v(r, 0) v(r, )
= (5.2.7b)
z z
Note that satisfaction of the two two-point boundary conditions (5.2.7a) and (5.2.7b)
does not generally imply satisfaction of two one-point boundary conditions of the form
used by Fasel and Booz (1984) (their equations 12 and 13) to compute Taylor vortices with
= 0. Those authors correctly point out that conditions of the form (5.2.9) capture the
appropriate solution only if the surface on which the axial velocity component vanishes is
a plane perpendicular to the symmetry axis. Here, note that this will be true for solutions
symmetric about z = /2, but not for other axially-periodic solutions. The existence of at
least one solution symmetric about z = /2 is guaranteed by the following: the number of
89
steady solutions of (5.2.4), (5.2.6), and (5.2.9) must be odd (Benjamin 1976); the boundary
conditions (5.2.9) are symmetric about z = /2; and the number of solutions asymmetric
about z = /2 must be even. From this, it follows that the number of solutions symmetric
about z = /2 must be odd, and hence nonzero. Computations by Fasel & Booz for = 2
are said to confirm this point, as do ours. Note that if one uses the boundary conditions
(5.2.9) for 6= 0, they must be applied at an axial position corresponding to an extremum
of the axial radius variation on the shaft (i.e., at an axial location about which the radius
variation is an even function of the axial coordinate).
methods (Reddy & Gartling 1994). A consistent penalty method (as opposed to the
reduced integration penalty model) was used to satisfy incompressibility. Quadrilateral
elements (using an isoparametric mapping) with quadratic velocity and discontinuous
linear pressure interpolation (cf. Chen, Pritchard & Tavener 1995; Goodwin & Schowalter
1996), and 5 5 Gaussian quadrature for numerical intergration were used. For the
calculation, simplicity of the physical domain and the fact that this work is interested in
the case where < 1 (i.e., the minimum gap is larger than the amplitude of the waviness),
allow us to use a simple boundary-fitted mesh obtained by transformation of the uniform
the form
f (v) = 0 , (5.3.2)
which is solved by Newton iteration. As with all iterative schemes, one must decide when
to stop the iteration. Unless otherwise noted, convergence is accepted when the maximum
point-wise residual is less than 106 . For = 0, the analytical approximation of Taylor-
90
Couette flow for the small gap (Chandrasekhar 1961, 71) in a circular annulus was used
as the initial iterate at the lowest resolution for Ta values above the first bifurcation.
Solutions at each Ta were used as the initial iterate for the next highest Ta. For > 0,
the procedure differed only in that v = 0 was used as the initial iterate at the lowest
elements (Nz and Nr , respectively) is shown in tables 5.1 and 5.2 for = 0 and 0.25
(a value larger than any considered here), respectively, in each case for the value of
for which results are reported in 5.4. For = 0, = 2, and Ta = 300 (the highest
Ta considered), table 5.1 shows that the torque converges monotonically as N z increases
of the shaft radius larger than any considered here. Convergence is still monotonic with
increasing Nr at fixed Nz , and monotonic with increasing Nz at fixed Nr provided Nz
is large enough. As for = 0, = 2, and Ta = 300, convergence is still more rapid in
the axial direction than in the radial direction. Note that for Nz = 32 and Nr = 64, T
changes by less than 0.2% as the number of axial and radial elements is increased further.
In general, the spatial resolution needed to ensure a specified degree of accuracy increases
with increasing and Ta. Unless otherwise indicated, this work uses Nz = 32 and Nr = 64
for all computations.
To further validate the code, a comparison between our computed values of the
dimensionless torque per unit length
T
T = , (5.3.3)
2
91
and previous computational results for = 0 (Taylor-Couette flow) by Fasel & Booz (1984)
for = 2 and Ta = 300 is shown in table 5.1, where T is the dimensional torque per unit
length. The torque is computed at the outer cylinder by numerical integration of the
velocity field using a five-point Gaussian quadrature rule. For Ta 300, our results agree
with both sets of theirs (computed on the inner and outer cylinders) to within about
0.2%. Also, note that there is excellent qualitative agreement between our computed
streamfunction and velocity contours, and the analogous plots (figures 7(a-e)) of Fasel &
Booz (1984). Comparison is also made to the experimental torque data of Donnelly &
Simon (1960) for = 0. As shown in table 5.3 and discussed in 5.4.1, there is good
agreement between computed and experimental results.
5.4. Results
The results in this section pertain to = 0.5.
A single scalar measure is used to characterize the solutions, namely the dimensionless
torque per unit length T . Also, a dimensionless pseudo-streamfunction is defined
1
vr = and vz = (r) , (5.4.1)
z r r
satisfying (5.2.4a).
Couette flow for which torques have been reported beyond the first bifurcation from circular
Couette flow are those of Fasel & Booz (1984) for = 2. The computational wavelength
chosen ( = 2) is close to the wavelength of Taylor vortices in annuli of high aspect ratio,
92
and also provides for direct comparison to results for nonzero and elucidation of the
effects of radius variation. As discussed in 5.3, our results and theirs are in excellent
agreement.
Experimentally, comparison is made to the work of Donnelly & Simon (1960). (Note
that Debler et al. (1969) measured torques for = 0.5, but reported only critical values of
Ta (and corresponding T values) at that radius ratio.) Table 5.4 shows that the relative
difference between our computations and values of the dimensionless torque per unit length
calculated from the tabulated dimensional torques and reported viscosities, densities, and
length of Donnelly & Simon ranges from about 0.06% at Ta = 10.2 to 4.7% at Ta = 107.
The overall level of agreement suggests that experimental artifacts were unimportant in
the work of Donnelly & Simon. Note that the wavelength in the experiments of Donnelly
& Simon is determined in part by end effects, and is generally not equal to the fixed value
well with those of Donnelly & Simon up to Ta = 192, suggesting the limitations of the
perturbation calculations of Kirchgassner & Sorger at high Ta.
93
Figure 5.1 shows that below a critical Ta of 68.19 (point A), there is a single branch
of solutions (branch 1), corresponding to circular Couette flow. On branch 1, the torque
increases linearly from zero as Ta increases. Branch 1 continues indefinitely beyond the
first bifurcation (point A), since for = 0 circular Couette flow is an exact solution of the
Navier-Stokes equations for all Ta. On this branch the streamfunction is identically zero.
A second branch of solutions (branch 2) bifurcates at point A (Ta = 68.19) from
the circular Couette flow, and corresponds to well-known steady axisymmetric Taylor
vortex flow (DiPrima & Swinney 1985; see also Chapter 2). As expected, the torque
is considerably higher than for circular Couette flow. The torque on branch 2 increases
superlinearly with Ta, with the curvature being significant. The meridional streamlines
on branch 2 for Ta = 120 (figure 5.2) correspond to a counter-rotating vortex pair. For Ta
slightly above the bifurcation A, the vortex cores are centered. As Ta increases beyond A,
vortex cores shift away from center. Note that because the partial differential equations
and wall boundary conditions are axially invariant, and the computational domain is taken
to be axially periodic, the phase is indeterminate.
The bifurcation shown at point A in figure 5.1 is consistent with calculations of the
eigenvalues of the Jacobian matrix J, which show that there is a sign change in |J| between
Ta = 60 and 70. This is consistent with the passage of a temporal eigenvalue (computed
using the stability code described in Chapter 2) into the right half-plane (RHP).
At point B (Ta 98.60), a third branch of steady axisymmetric solutions (branch 3)
bifurcates from circular Couette flow. The torque on this branch initially increases more
rapidly with Ta than on branch 2. This leads to a cross-over of branches 2 and 3 near
Ta = 207. Beyond that point, branch 3 corresponds to the solution with higher torque.
Computation of the lowest-lying neutral curve for Couette flow, using the code described
in Chapter 2, shows that the Taylor number at point B corresponds to the neutral Ta
at an axial wavelength = 1, precisely one-half the wavelength used in the Navier-Stokes
computation of the present base flow. A countably infinite set of such bifurcations from
94
branch 1 corresponds to axial wavelengths equal to 1/N of the forcing wavelength, where
N = 1, 2, . . ., and N = 2 corresponds to point B. We believe that these solutions and
their counterparts with oppositely-rotating vortices (corresponding to a 180 phase shift)
are the only steady axisymmetric solutions when = 0. Figure 5.3 shows meridional
streamlines on branch 3 for Ta = 120. Unlike the single vortex pair observed for solutions
on branch 2 (figure 5.2), solutions on branch 3 consist of two vortex pairs, each having half
the wavelength of those found on branch 2. For Ta = 120, the solution shown in figure 5.3
does not vary significantly from what one would expect to see just above the bifurcation
B, and thus the vortex cores are centered. As one moves beyond the bifurcation B, vortex
cores begin to shift away from center as observed for solutions on branch 2.
The bifurcation at point B in figure 5.1 is consistent with the computed eigenvalues of
J, which show that |J| changes sign between 90 and 100, consistent with the passage of two
additional temporal eigenvalues (computed using the stability code described in Chapeter
2) into the RHP.
5.4.2. = 0.01
Here, and for all other results presented below, is the wavelength of the axially-
periodic variation of the shaft radius, and is the dimensionless wavelength of the
of solutions (branch 1c ), here refered to as modified Couette flow, in the sense that one
expects the axial and radial velocity components to be small and to scale with . (Note
that for = 0, the only nonzero component of the small-Ta solution is v .) On this
branch, T increases nearly linearly with Ta up to point A0 (near Ta = 70, just above the
first critical Ta for = 0), where a bifurcation apparently similar to that at Ta = 68.19
(for = 0 at point A in figure 5.1) occurs. Unlike the = 0 case, for solutions on branch
95
1c there are three nonzero velocity components and thus the meridional streamfunction
is not identically zero. Figure 5.5 shows streamlines for Ta = 50. For = 0.01, inner
radial boundary curvature is enough to cause nonzero variation in the axial and radial
velocity components, but not enough to cause significant change in the vortices observed
on branch 2 of figure 5.1. The azimuthal velocity is very much stronger than the other
two components.
The bifurcation at point A0 in figure 5.4 is consistent with the computed eigenvalues of
J, which show that |J| changes sign between 68 and 70. Note that the bifurcating branch
2c corresponds to one of two solution branches that apparently bifurcate from branch 1 c at
point A0 . The second solution branch (with similar topology) has vortices rotating in the
direction opposite to those on branch 2c , and has slightly higher values of T . For fixed Ta,
one would expect the difference in T to increase as increases. For Ta = 80, figures 5.6
and 5.7 show meridional streamlines for the branch 2c solution and its oppositely rotating
counterpart, respectively.
These streamline contours are qualitatively similar to those observed on branch 2
(figure 5.2) for = 0, except in this case the geometry fixes the phase within one-half
These solutions show that there a single vortex pair, and that the wavelength is twice
that of the Taylor Couette branch for = 0. At these Ta, the vortex cores are shifted
considerable off center. The separating streamsurface between voritces of a single pair
96
coincides with the inner wall extrema. At B0 (thought to be near Ta = 80, below the
Ta corresponding to point B in figure 5.1 for = 0), the results suggest that a second
bifurcation may occur, this time from branch 3c (the smooth continuation of the low-
Ta branch of modified Couette flow). The bifurcating branch 4c rapidly approaches a
linear variation of T with Ta, having essentially the same slope and intercept as branch
1c . On branch 4c , figures 5.12 and 5.13 (for Ta = 120 and 150, respectively) show that
solutions consist of a single vortex pair (as found on branchs 3c and 1c ). These solutions
are qualitatively similar to those found on branches 1c (figure 5.4) and 2 for = 0 (figure
5.1).
The bifurcation at point B0 in figure 5.4 is consistent with the computed eigenvalues of
J, which show that |J| changes sign between 76 and 80. Again, note that the bifurcating
branch 4c corresponds to one of two solution branches that apparently bifurcate from
branch 3c at point B0 .
Farther along branch 3c , at point C0 (near Ta = 135), results suggest that a third
bifurcation may occur, with the new branch of solutions (branch 5c ) showing a rapid
increase of T with Ta. The bifurcation at point C0 in figure 5.4 is again consistent with
the computed eigenvalues of J, which show that |J| changes sign between 135 and 136. On
branch 5c , figure 5.14 shows meridional streamlines for Ta = 150. On this solution branch
there are three vortex pairs, with two similar pairs occupying the downstream region, and
one larger pair lying upstream. The two downstream pairs extend about one-fourth of the
forcing wavelength, while the single upstream pair is about one-half as long as the forcing
wavelength (similar to branches 2c (figure 5.4) and 2 for = 0 (figure 5.1)).
5.4.3. = 0.03
For = 0.03 and = 4, figure 5.15 shows that for Ta less than about 75, there is
97
that of the geometric forcing for modified Couette flow branches at smaller (figure 5.5).
Streamsurfaces separating vortices are attached to the wavy wall at the extrema of the
inner boundary, as for branch 1c solutions at = 0.01 (figure 5.5). The bifurcation at
point A0 in figure 5.15 is thought to be located between Ta = 70 and 75, which is the
interval in which |J| changes sign. On the bifurcating solution branch 2c , T increases
superlinearly with Ta and remains the high-torque branch throughout the range of Ta
considered. This behavior is analogous to that of branch 2c for = 0.01. For branch 2c ,
figure 5.17 shows meridional streamlines for Ta = 120. For this solution branch, there
are two sets of vortex pairs, each with one-half the forcing wavelength, which is also the
wavelength for the standard Taylor-Couette case. Again, each separating streamsurface
attaches at an inner wall extremum.
Beyond point A0 , the continuation of branch 1c (branch 3c ) develops noticeable
curvature as Ta increases. This was also the case for the analogous = 0.01 solution
branch. Figures 5.18 and 5.19 show branch 3c meridional streamlines for Ta = 120 and
150, respectively. On this branch, solutions have the forcing wavelength. The separating
streamsurface again attaches at the inner wall extrema, and vortex cores are shifted towards
the inner-wall maximum radius. As Ta increases along branch 3c beyond point A0 , figure
5.15 suggests that a second bifurcation occurs near Ta = 80 (point B0 ). The bifurcation
at point B0 in figure 5.15 is thought to be located between Ta = 75 and 80 according to
the sign change of |J| between these two values. On the bifurcating solution branch 4c ,
T rapidly approaches a linear variation with Ta having slope similar to branch 2c . Note
that branch 4c bifurcates in the direction of increasing T , which was not the case for
the second bifurcating branch (4c ) in the = 0.01 case. For Ta = 120, figure 5.20 shows
meridional streamlines on branch 4c , which consist of two vortex pairs, one smaller than
the other. The smaller pair, whose separating streamsurface is attached at the inner-wall
maximum radius, has vortex cores that are nearly centered, while the cores of the larger
98
pair, whose separating streamsurface attaches at the inner-wall minimum radius, has its
cores shifted discernably away from center.
Further along branch 3c , figure 5.15 suggests that a third bifurcation occurs. This
bifurcation is inconsistent with the computed eigenvalues of J, which show that |J| on
branch 3c does not change sign beyond Ta = 80. At this point, it is not clear if branch
5c bifurcates from branch 3c or is completely disconnected from any of the branches found
so far (which is the case for = 0.07 discussed in th next section). The new branch of
solutions (branch 5c ) apparently approaches a linear asymptote with slope nearly identical
to that of branch 1c for low Ta, as was the case for = 0.01, where the slope of branch 4c
approached the slope of branch 1c (figure 5.4). On branch 5c , figures 5.21 and 5.22 show
meridional streamlines for Ta = 120 and 150, respectively. Solutions have the forcing
wavelength and are qualitatively similar to those of branch 1c . As one increases Ta on
5.4.4. = 0.07
For = 0.07 and = 4, figure 5.23 shows T as a function of Ta. For sufficiently
small Ta, there is again a single branch of solutions (branch 1c ), but unlike the smaller-
amplitude cases discussed above, this branch develops noticeable curvature before the
radius there is radial inflow. The bifurcation at point A0 in figure 5.23 is consistent with
the computed eigenvalues of J, which show that |J| changes sign between 70 and 80. On
the bifurcating solution branch 2c , T again increases superlinearly with Ta and remains
the high torque branch over the range of Ta considered. On branch 2c , figures 5.25 and
5.26 show meridional streamlines for Ta = 120 and 150, respectively. Solutions on this
branch have the forcing wavelength.
99
Beyond point A0 , the continuation of branch 1c (branch 3c ) continues to increase
superlinearly with increasing Ta. As Ta increases along branch 3c , results suggest a second
bifurcation occurs near Ta = 100 (point B0 ). The bifurcation at point B0 in figure 5.23
is consistent with the computed eigenvalues of J, which show that |J| changes sign again
between 90 and 100. On branch 3c , figures 5.27 and 5.28 show meridional streamlines
for Ta = 120 and 150, respectively. On this branch, there is a single vortex pair whose
wavelength is that of the forcing, consisting of one large and one small vortex. Again, the
separating streamsurfaces attach at the inner-wall extrema. On this bifurcating solution
branch 4c , T rapidly approaches a linear variation with Ta. On branch 4c , figure 5.29
shows meridional streamlines for Ta = 150. Solutions on this branch are qualitatively
similar to solutions on branches 2c and 3c . Note that on solution branches which have
bifurcated from the "primary branch" (i.e., branch 1c and its continuation branch 3c ), the
values of T are much closer to those on the primary branch than was the case for smaller
. Branches 2c , 3c , and 4c have similar slope at high Ta.
For = 0.07, a solution branch (5d ) apparently not connected to the primary branch
is found. This is the only case considered for which this result has been found. As Ta
increases beyond the turning point on branch 5d , T increases with nearly equal slopes on
its upper and lower branches. For Ta = 150, figure 5.30 shows meridional streamlines on
the lower portion of branch 5d . Solutions on this portion of the branch consist of one large
vortex pair whose separating streamsurface attaches at the inner-wall maximum radius,
and two smaller (weaker) vortex pairs stacked vertically above the inner-wall minimum.
The lower of these two pairs has a separating streamsurface that attaches at the inner-wall
minimum radius. For Ta = 150, figure 5.31 shows meridional streamlines on the upper
portion of branch 5d . Solutions on this portion of the branch differ from those
on the lower portion in that there is only a single pair of smaller (weaker) vortices, whose
separating streamsurface again attaches at the inner wall minimum radius.
100
5.4.5. = 0.1
For = 0.1, only a single solution branch at each Taylor number considered has been
found, as shown in figure 5.32. For this case, torque varies nearly linearly with Ta over
an initial range (up to about 60), before developing noticeable curvature as Ta increases
further. Solutions on this branch have the forcing wavelength. Figures 5.33-5.35 show
meridional streamlines for Ta = 50, 120, and 150, respectively. For Ta = 50 (figure
5.33), vortex cores are centered and the separating streamsurfaces attach at the inner-wall
extrema. As Ta increases to 120 and then to 150, vortex cores begin to shift towards
the separation streamsurface attached at the inner-wall maximum radius. As can be seen
in figures 5.33-5.35, there is radial outflow at the inner-wall maximum radius and radial
inflow at the inner-wall minimum radius.
Results computed for = 0.25 are qualitatively similar to those for = 0.1. In
5.5. Discussion
5.5.1. General dependence of the flow on
For = 0, two bifurcations from circular Couette flow have been found. The first
is the standard Taylor-Couette flow, while the second corresponds to the neutral Ta at a
dimensionless axial wavelength = 1, precisely one-half the wavelength used in the Navier-
Stokes computation of the present base flow. A countably infinite set of the latter such
bifurcations from branch 1 corresponds to axial wavelengths equal to 1/N of the forcing
torque, while beyond about Ta = 210, branch 3 has the highest torque. This crossing is
thought to be a result of the metric used, and do not suggest that branch 3 becomes the
stable branch.
101
For = 0.01, there is a significant change in the bifurcation behavior and solution
structure as compared to the = 0 case. For small Ta, there is a single branch of
solutions corresponding to a modified Couette flow (i.e., nonzero axial and radial velocity
components and much smaller than the azimuthal component). For this case, three
bifurcations form the primary branch 1c and its continuation 3c have been found. The first
(branch 2c ) corresponds to the high torque branch (analogous to branch 2 for = 0), while
the second (branch 4c ) corresponds to the low torque branch at high Ta, and the third
branch has torque values lying between those of the high torque branch and the primary
branch. Solutions on the primary branch and branch 4c have the forcing wavelength,
while solutions on the high torque branch 2c have one-half the forcing wavelength. Finally,
solutions on branch 5c are made up of three vortex pairs, with two pairs having one-fourth
the forcing wavelength, and a larger pair having one-half the forcing wavelength.
For the case of = 0.03, there is again a single branch of solutions for small Ta,
corresponding to modified Couette flow. There are again three bifurcations form the
primary branch 1c and its continuation 3c . The first (branch 2c ) corresponds to the high
torque branch as was the case for = 0.01, while the second (branch 4c ) corresponds to a
branch of solutions with torque values between those found on branches 2c and 3c . The
third branch (5c ) of solutions has torque values lying below values found on the other three
branches, with a slope very similar to that of branch 1c . Solutions on the primary branch
and branch 5c have the forcing wavelength, while solutions on the high-torque branch 2c
have one-half the forcing wavelength. Finally, solutions on branch 4c are made up of two
vortex pairs, one smaller than the other.
For large-amplitude radius variations ( 0.1), a single pair of vortices at each Ta
have been found. Since for 6= 0, solutions with two or more vortex pairs (found at smaller
) necessarily have vortex pairs of different sizes, this suggests that beyond some critical
value of (apparently between 0.07 and 0.1) either a) such solutions become disallowed;
102
b) the Ta at which such solutions appear becomes very large; or c) the domain of
convergence of our solution technique becomes very small for such solutions.
= 0.959, for which axisymmetric Taylor vortices in the standard = 0 geometry are
stable over only a very small range of Ta (extending only about 10% above the critical
value at which circular Couette flow becomes unstable). The primary emphasis in their
work was on development and evaluation of unsteady, 3-dimensional wavy Taylor vortex
flow and the effects of boundary topography. However, they did provide some images of
the flow in the range of Ta for which the steady axisymmetric flow is stable, so that some
qualitative comparison to their results is possible.
In the experiments of Ikeda & Maxworthy (1994), the form of the radius variation
on the inner cylinder is identical to ours, with geometric parameters = 0.959, = 2.8,
and = 0.112. In the steady, axisymmetric regime, their visualization results show that
there is radial inflow above the inner-wall minimum radius, which is precisely what our
computations show for = 0.5, = 4, and = 0.1. These narrow-gap experimental results
103
Chapter 6
Computation of Helical Flow Driven by Rotation of a
Screw in a Coaxial Outer Circular Cylinder
6.1. Introduction
Flow in or through a circular annulus driven by rotation of the inner cylinder has been
of interest since the work of Taylor (1923). One of the simplest ways to influence this flow
is to change the geometry of either the inner or outer cylinder. The work of chapter 5
shows that for the case of axisymmetric axially-periodic inner radius variation, one can
expect the flow to be significantly modified compared to that for two cylinders. In this
chapter we consider the case of an inner cylinder having a helical groove (i.e., a screw).
Most fluid dynamical studies in rotating screw geometries have focussed on either
screw extruders used in polymer processing, or viscoseals or labyrinth pumps and seals.
In the extrusion context, the early work of Carley et al. (1953a,b) described the basic
concepts of flow through screw extruders, and has provided the basis for industrial design
in this area for many years. However, this and most subsequent work (Griffith 1962;
Kroesser & Middleman 1965; Tung & Laurence 1975; Choo, Neelakantan & Pittman 1980;
Booy 1981; Choo, Hami & Pittman 1981; Elbirli & Lindt 1984; Bruker, Miawl, Hasson &
Balch 1987; Bohme & Broszeit 1997; Broszeit 1997) in the area has been for non-Newtonian
(and in some cases nonisothermal) flows. There has been some work done for isothermal
Newtonian fluids (Hami & Pittman 1980; Booy 1981, Bohme & Broszeit 1997), but only
Stokes flow solutions (i.e., inertial terms neglected) have been reported.
Labyrinth seals (i.e., viscoseals or screw seals) are used as dynamic seals for rotating
shafts. With small axial pressure gradients, infiltration of oil, water, or other fluids can
be greatly reduced. Common seal shaft geometries include a single helical groove of
104
Taylor numbers considered were well into the turbulent regime. The bulk of the analytical
and computational work on flow in seals (cf. McGrew & McHugh 1965; Meric & Macken
1974) uses the parallel plate model, which approximates the flow between an outer circular
cylinder and a rotating inner cylinder into which a helical groove of rectangular cross-
rotating about the axis of a cylindrical hole of overall aspect ratio (length divided by
diameter) eight can give an electrodeposited copper layer with a thickness that is essentially
uniform, compared to significant axial nonuniformity that occurs when convective mass
transfer is due to a time-periodic Poiseuille flow driven by oscillatory motion of the hole
smaller temporal period (in the rotating lab frame) than the helical passage time.
Finally, for many years there has been significant interest in the use of static helical
inserts to enhance heat or mass transfer in flow in a circular pipe driven by an axial pressure
gradient. Until recently, interest in such devices has largely been driven by their lack of
moving parts and ease of maintenance, particularly in corrosive environments (cf. Martin &
Galey 1994). Recently, however, static mixers with helical elements have attracted interest
in microfluidic applications, in which steady three-dimensional flow offers the possibility
of good mixing with no moving parts at extremely low Reynolds numbers (cf. Bertsch et
al. 2001).
105
Here, for a single-helix screw, we present a computational investigation of laminar
helical flows driven solely by screw rotation. Here, geometric (i.e., aspect ratio, radius
ratio, screw flight clearance, and trough width) and dynamic (i.e., Taylor number) effects
on the velocity field are considered. To the best of our knowledge, these are the first
analytical or computational results for this flow for a Newtonian fluid at nonzero Taylor
number. The case of an axial pressure gradient (with and without screw rotation) is dealt
with separately.
6.2. Formulation
The geometry of interest is that of a fixed outer circular cylinder and an inner coaxial
screw. In a plane passing through the symmetry axis of the outer cylinder, figure 6.1 shows
a projection of the screw geometry. Here, = Ro /L is the overall aspect ratio, = Ri /Ro
is the hub ratio, = (Ro R1 )/(Ro Ri ) is the clearance ratio, and = l/L is the groove
width ratio, where Ri and Ro are the radii of the screw hub and outer cylindrical barrel,
respectively, R1 is the radius of the flight, l is the groove width, and L is the axial screw
pitch length. The screw rotates at constant angular speed about the z axis.
Use of a helical coordinate system in a frame rotating with the screw reduces
calculation of the simplest flow to a steady one, fully-developed in the helical direction.
Helical coordinate systems can be classified into two groups, according to whether or not
they are orthogonal. A helical coordinate system seems to have been used first by Waldron
(1958) to solve electromagnetic problems. Subsequently, helical coordinate systems have
been used to investigate flow driven by single-screw extruders (Nebrensky, Pittman &
Smith 1973; Tung & Laurence 1975; Hami & Pittman 1980; Choo, Hami & Pittman 1981;
Bohme & Broszeit 1997; Yu & Hu 1997), and flow in helical ducts and pipes (Wang 1981;
Germano 1982; Song & Yoo 1987; Choi, Song & Yoo 1988; Wang & Andrews 1995).
The nonorthogonal helical coordinate system discussed by Waldron (1958) and first
used in a fluid mechanics context by Bohme & Broszeit (1997) was used to compute the
106
velocity and pressure fields. For a left-handed helix, the mapping between this helical
coordinate system and a (right-handed) cylindrical coordinate system (r, , z) is
% = r/Ro (6.2.1a)
= (6.2.1b)
= z/L + . (6.2.1c)
2
Note that the development by Waldron (1958) is for a right-handed helix. The pressure
and velocity are scaled according to
P
= (6.2.2a)
2 Ri2
V
v= , (6.2.2b)
Ri
where P is the dimensional pressure variation associated with screw rotation. Note that
flow driven solely by screw rotation gives rise to no mean axial pressure gradient, just
like plane Couette flow and "sliding Couette flow" (Joseph 1976, vol. 1), which like the
flow of interest are driven strictly by boundary motion. For flows that are helically fully-
developed and driven solely by screw rotation, there is no mean pressure variation in the
axial direction. On the other hand, when an external pressure gradient is applied (e.g., in
a static mixer), the flow is driven by a mean axial pressure gradient.
Flows in this geometry will be computed using a mixed formulation (i.e., cylindrical
vr 1 1 vr vr v2 2 %
vr + v + vz v 2 (6.2.3a)
% 2 % %
107
1 h 2 vr 1 1 v i
= + vr 2
% Ta % %2
v 1 1 v v vr v 2
vr + v + vz + + vr (6.2.3b)
% 2 % %
1 1 1 h 2 1 1 v v i
= + v + 2
2 % Ta %2 %
vz 1 1 vz vz 1 2
vr + v + vz = + vz , (6.2.3c)
% 2 % Ta
1 1 1 v vz
(%vr ) + + =0 , (6.2.3d)
% % 2 %
where
1 1 1 2
2 2
= (% ) + + , (6.2.4)
% % % 4 2 %2 2
where Ta = Ri Ro / is a Taylor number. The final terms on the left-hand sides of (6.2.3a)
and (6.2.3b) arise from the Coriolis acceleration, while the penultimate term on the left-
hand side of (6.2.3a) arises from the centrifugal acceleration, which was not incorporated
into the dimensionless pressure. The density and viscosity are taken to be constant. The
no-slip boundary conditions on the rigid impermeable inner and outer walls are
108
on the outer cylinder. The angle = 1/(2%) in (6.2.5e) is characteristic of the helix
and lies in the range 0 < /2. Axial periodicity requires
v(%, 0) v(%, 1)
= 1 (1 ) % 1 (6.2.6b)
advantageous.
Discretization of (6.2.3) leads to a quadratically nonlinear algebraic system of the form
f (v) = 0 , (6.3.1)
resolution can be interpolated to get an initial iterate for the next-higher resolution.
To validate the code, we compare our results to existing computations and
experiments. No Navier-Stokes calculations for Newtonian fluids in screw geometries at
nonzero Ta ahve been found. Thus, as an initial check, we compare our computational
results to flows which are simplifications of the helical formulation (6.2). The
incompressible Navier-Stokes equations in helical coordinates (6.2.3) reduce to equations
109
in axisymmetric cylindrical coordinates in the limit = . The appropriate terms have
been set to zero in (6.2.3) in order to compute axisymmetric Taylor-Couette flow (Fasel &
Booz 1984) and an axisymmetric geometric modification of Taylor-Couette
flow (chapter 5; Chossat & Iooss 1994) as initial checks of the code. In both cases there
is excellent agreement between previous results and solutions computed using the present
code.
Our streamline and axial velocity contours can also be compared to the Newtonian
Stokes-flow computations of Bohme & Broszeit (1997) for = 1.515, = 0.7, = 0.1, and
= 0.8788. Although those authors provide no numerical values of the torque, velocity,
or streamfunction to which we can make direct comparison, their streamline and axial
velocity contours are indistinguishable from our results for the same parameter values.
Comparison is also made to their dimensionless volumetric flow rates, which shows that
our computed values agree with the results of Bohme & Broszeit as well as the latter can
be determined from their figure 7. The slight asymmetry of their axial velocity contours
with respect to the symmetry line = 1/2 of the domain is discussed in 6.5.
6.4. Results
This work investigates how the helically fully-developed flow depends on the geometric
parameters defined in table 1, and on Ta. In this chapter, we consider the case in which
flow is driven solely by screw rotation. An axial Reynolds number Rez = V z (Ro R1 )/
is defined based on the mean axial speed of the flow driven by the rotating screw and the
gap between the cylindrical barrel and the flight, a dimensionless torque per unit length
T
T = , (6.4.1)
2
110
proportional to the mean dimensionless radial kinetic energy per unit mass, which provides
a measure of radial mixing using only the velocity field, where T is the dimensional torque
per unit length. The mean axial speed is computed by transforming an integral in a plane
normal to the screw axis into an integration over the computational domain (figure 6.1).
provides a simple (if somewhat crude) overall measure of convective radial mixing due to
the flow. Also, a dimensionless "meridional" pseudo-streamfunction is defined
1
v% = v = , (6.4.3a, b)
% % %
characterizing flow in planes containing the screw axis, which satisfies (6.2.3d), where the
helical and cylindrical components of v are related by
v% = v r (6.4.4)
v = vz + v tan . (6.4.5)
In what follows, we hold two geometric parameters (the aspect ratio and the hub
ratio ) constant and investigate the effects of the groove depth () and width ( ) and
Ta on the velocity field, radial mixing (as measured by ), and the dimensionless torque
per unit length T . In this work, reference to and contours of constant are referred
to simply as the meridional streamfunction and meridional streamlines, respectively; the
region 1 (1 ) < % < 1 (R1 < r < Ro ) is referred to as the "clearance"; and (R1 Ri )/
l = (1 )(1 )/ is referred to as the groove aspect ratio. The results shown pertain
to = 2 and = 0.5. Note that for fixed values of the hub and barrel radii (consistent
with fixed ), Ta is proportional to the rotation rate.
111
6.4.1. Shallow grooves ( = 0.8)
For large values of , corresponding to shallow grooves, there is a broad range of groove
widths and Taylor numbers for which annular Poiseuille flow is a good approximation to
the meridional flow in most of the clearance. In this section, we consider flows for = 0.8,
for which the radial component of the meridional flow is small except in and immediately
adjacent to the groove.
Narrow groove. For a shallow ( = 0.8) narrow ( = 0.2) groove, figures 6.2(a-f)
show meridional streamlines (solid and dashed lines) and vr2 contours (color) over the
surface generated by a line segment perpendicular to the screw axis whose ends move
along the screw axis and a helix having the same pitch as the screw.) For Ta = 0, the
streamlines in figure 6.2a show that the meridional flow is essentially axial except in the
groove. In the clearance, there is little meridional streamline curvature except where the
axial flow interacts with the primary (e.g., dominant) vortex in the groove, separated from
the clearance flow by a separation streamsurface attached to the groove walls just near
the lips. (At Ta = 0, two very small secondary vortices are confined to the corners of the
groove.) That there is a single dominant vortex in the groove is not surprising, since the
groove aspect ratio is 1. Four distinct regions of high vr2 are located in and just outside
the groove. Radial flow and mixing are largely confined to the groove and to a region
just beyond in which the meridional flow is sensibly affected by the groove. The two
upstream well-mixed regions are separated by a thin layer, that includes the separation
streamsurface, in which vr2 is much smaller, as are the two downstream well-mixed regions.
For small Ta, radial mixing is relatively poor over most of the domain.
112
For Ta = 25 (figure 6.2b), the meridional streamlines are slightly asymmetric in and
near the groove, but are otherwise qualitatively similar to the Stokes-flow case (figure
6.2a). The vr2 contours show significant asymmetry, the degree of which is much greater
than that of the meridional streamlines, since the values of vr2 are orders of magnitude
separation streamsurface, is both inclined and curved. As Ta increases (figures 6.2c and
6.2d), the primary vortex in the groove continues to deform, becomes more asymmetric,
and accommodates an increasingly large secondary vortex in the downstream corner of the
groove. The upstream interface between regions of good radial mixing in the clearance
begins to depart from the vertical, and the downstream interface has become increasingly
distorted. Figures 6.2e and 2f (Ta = 150 and 200) show the beginnings of interaction
between the meridional flow in the clearance and in the groove, with the primary vortex
extending out of the groove and into the clearance, thus leading to increased meridional
streamline curvature in the clearance. Figures 6.2(c-f) show that the regions of high v r2
grow both axially and radially in the groove, and especially in the clearance.
Figures 6.3a and 6.3b show contours of the azimuthal velocity at Ta = 0 and 200. At
Ta = 0, the contours are symmetric about = 1/2, as expected for Stokes flow. Figure
6.3a also shows that except in and near the groove, v has very little axial variation. Its
radial variation is essentially that of circular Couette flow. Comparing the v contours
in figures 6.3a and 6.3b, we see much less change than in the meridional streamlines,
with azimuthal flow outside the groove being still essentially Couette-like, except near the
flights, where departures from Couette behavior are confined to a small region whose radial
extent increases with Ta. The v contours for Ta > 0 are qualitatively similar to those of
113
Stokes flow (figure 6.3a), with the main difference being the degree of axial nonuniformity
(i.e., contour curvature) near the groove.
Figure 6.4a shows that as Ta increases, the mean axial flow rate, parametrized by Re,
increases a bit more than linearly with Ta. Figure 6.4b shows that the increase in the
torque T is essentially linear with Ta over the range shown. Finally, figure 6.4c shows that
the integral measure of radial flow, , initially increases quite rapidly with Ta. Computed
values of (0.168, 0.255, 0.363, and 0.494 at Ta = 250, 300, 350, and 400, respectively)
show that continues to increase superlinearly for Ta beyond the range of figure 6.4c.
computational plane is again essentially axial in the clearance except near the groove,
where there is significant meridional streamline curvature. Meridional streamlines from
the clearance penetrate deeply into the groove, nearly to the hub. Two prominent corner
vortices are accommodated by the groove, whose aspect ratio is now only 0.4. This differs
from the = 0.2 case at Ta = 0 (figure 6.2a), for which the groove had unit aspect ratio
and accommodated a single vortex, and the meridional streamlines from the clearance
penetrated only slightly into the groove. Figure 6.5a also shows that there are two distinct
regions of high vr2 , each penetrating into the groove. This contrasts to the narrow groove
case, in which two regions of good radial mixing lied wholly within the groove and two
were entirely outside, with the inner and outer mixed regions separated by a thin layer
in which the radial velocity was small. The wider groove gives rise to much better radial
transport into the groove (note the larger red regions in figure 6.5a compared to figure
6.2a, as well as the different color maps.) In addition, there are two smaller regions in
which weak radial mixing occurs in the upstream and downstream corners of the groove,
corresponding to the helical corner vortices.
114
As Ta is increased to 25 (figure 6.5b), the upstream corner vortex grows at the expense
of the downstream one, occupying a larger axial and radial portion of the groove. The
upstream corner vortex is bounded by a (helical) separation streamsurface over which flow
from the clearance passes, whose projection onto the meridional plane is reminiscent of
flow over a two-dimensional backward-facing step. At this Ta, the upstream vortex is
still confined to the groove. Two regions of high vr2 lying largely in the clearance have
deformed significantly, while also shifting downstream. Of the two regions of weak radial
mixing in the groove for Stokes flow (figure 6.5a), the upstream one, associated with the
remaining groove vortex, has intensified, while the downstream one has disappeared. As
Ta continues to increase (figures 6.5(c-e)), the groove vortex grows radially and begins to
extend into the clearance. As Ta increases from 50 to 200 (figures 6.5(c-f)), the region of
high vr2 in the upstream corner of the groove grows axially and radially (e.g., figure 6.5e),
until encompassing about a third of the groove and a large portion of the upstream part
of the clearance. At Ta = 50, separation occurs very close to the upstream lip of the
groove, whereas at Ta = 100, the separation streamsurface attaches upstream of the lip
onto the flight itself. A significant "scouring" flow has developed in the downstream half
of the groove, with the outer flow penetrating to the hub, despite protrusion of the groove
vortex into the clearance. The high vr2 region near the upstream lip at Ta = 0 (figure
6.5a) has moved significantly downstream, with its inner portion centered in the groove
and its outer portion stretched in the downstream direction (e.g., figure 6.5d). As Ta
increases still more, figures 6.5e and 5f show that the groove vortex extends well into the
clearance, and that the upstream attachment of the separation streamsurface has moved
farther upstream. At the highest Ta, figure 6.5f shows that the groove vortex affects the
meridional flow far into the clearance. For Ta = 200, meridional streamlines in the groove
are qualitatively similar to those for narrower grooves (see figure 6.2f), although the groove
vortex protrudes farther into the clearance. At Ta = 200 (figure 6.7f), there are now three
distinct regions of high vr2 instead of the four found at lower Ta. This behavior contrasts
115
to the narrow-groove case (figure 6.3f) in which four distinct regions of high v r2 persist at
large Ta.
Figures 6.6a and 6.6b show v contours for Ta = 0 and 200. Again, the azimuthal
velocity contours are symmetric about = 1/2 at Ta = 0 and nearly symmetric at
Ta = 200, and show little curvature except near the groove. The radial extent for which
there is significant contour curvature is slightly larger for = 0.5 than for = 0.2, with
the azimuthal velocity component in the clearance being essentially independent of the
axial coordinate, as found for = 0.2.
Figure 6.7a shows that as Ta increases, Re increases superlinearly, with the departures
from linearity being much stronger than for = 0.2. Note that for each Ta, Re exceeds
its value for the = 0.2 case. This is expected, since absent an imposed axial pressure
gradient, all of the mean axial flow is driven by helix rotation, and is directly attributable
to transfer of axial momentum from the groove to the clearance. This process is clearly
facilitated by increasing the groove width. Figure 6.7b shows that T again increases
nearly linearly with Ta over the range shown, with its value for each Ta being less than for
the = 0.2 case. This is associated with the fact that the average gap between the screw
and barrel is higher as increases, thus leading to lower average radial shear. Finally,
figure 6.7c shows that increases quite rapidly with Ta, and that the apparent asymptotic
value is larger than predicted for = 0.2. This reflects the better radial mixing associated
with the wider groove.
Wide groove. For a shallow ( = 0.8) wide ( = 0.8) groove, figures 6.8(a-f) show
meridional streamlines and vr2 contours over the range 0 Ta 200. As Ta increases
from 0 to 200, we see that the meridional streamlines and v contours progress through a
sequence similar to that for = 0.5. The groove aspect ratio is now 0.25, and at Ta = 0,
the helical vortices in the corners of the groove occupy only a small part of the groove, and
fluid from the clearance "scours" a large part of the hub. The wider groove contributes
to more radial mixing over more of the domain than occurs for similar Ta values when
116
= 0.2 and 0.5. Note that for Ta = 200, the vortex occupying a large portion of the
groove protrudes farther into the clearance than for the = 0.5 case.
The azimuthal velocity contours and meridional streamfunctions shown in figures 6.9a
and 6.9b demonstrate that the v contours are much less affected by Ta. The principal
differences between the v contours at Ta = 0 and 200 are greater axial variation in the
clearance and less fore-aft symmetry (especially in the inner part of the clearance, near
the groove) at Ta = 200.
Figure 6.10a shows that as Ta increases, Re increases more rapidly than previously
seen for = 0.2 and 0.5, with the value at Ta = 200 being considerably larger than for
the previous two cases. Figure 6.10b shows that the increase in T with Ta is noticeably
superlinear, which differs from the cases thus far discussed. Note that the value of T at
Ta = 200 is smaller than for the narrow- and intermediate-width grooves. Finally, figure
6.10c shows that increases more rapidly with Ta than in the = 0.2 and 0.5 cases,
and appears to approach an asymptote whose value is larger than those for the narrower
grooves considered.
6.11(a-f) show meridional streamlines and vr2 contours over the range 0 Ta 200. For
Ta = 0 (figure 6.11a), meridional streamlines in the clearance show little curvature, and
penetrate only marginally into the groove. In this case, the groove has an aspect ratio
of 2.5, and accommodates two vortices stacked radially, which differs from the previous
cases in which there was a single major groove vortex, or one vortex in each corner of
the groove. Figure 6.11a shows that at Ta = 0, there are four distinct regions of high
vr2 , with two confined to the outer part of the groove, and two in the clearance, all near
the lips. This differs from the shallow-groove ( = 0.8) case, in which there was good
radial mixing in most of the narrow groove. As Ta is increased to 25 (figure 6.11b), the
meridional streamlines in the groove develop significant asymmetry. The outer vortex,
117
which is approximately trapezoidal in the meridional plane and whose core is nearly
undisplaced, is stretched towards the hub on the upstream side and compressed outwards
toward the groove lip on the downstream side. The regions of good radial mixing in
the clearance are generally displaced downstream, with contours of the more downstream
region extending beyond the downstream end of the spatial period. At Ta = 50 (figures
6.11c), the outer groove vortex, whose core is displaced slightly upstream, continues to
deform until it extends to the upstream part of the hub. At Ta = 100, figure 6.11d shows
that both groove vortices extend from nearly the hub to the clearance, unlike the Stokes-
flow case (figure 6.11a). At Ta = 100, the region of good radial mixing that for Ta = 0 was
located in the groove near the downstream lip, has merged with the upstream well-mixed
region in the clearance to form a single well-mixed region extending over almost all of the
clearance opposite the groove, and into the groove itself. For Ta = 150 and 200, figures
6.11e and 6.11f show that there is only one recirculation region in the groove, and that the
meridional flow in the clearance now penetrates to the downstream end of the hub.
Figures 6.12a and 6.12b show v contours for Ta = 0 and 200. Again, the azimuthal
velocity contours are symmetric about = 1/2 at Ta = 0 and nearly symmetric at
Ta = 200. For Ta = 0, we see little curvature in these contours except in and near the
groove, while for Ta = 200 there is significant contour curvature halfway from the flight
to the barrel. This contrasts to the Ta = 200 behavior seen for a narrow shallow groove,
where significant curvature of the v contours does not penetrate nearly as far into the
clearance. The v contours for 0 < Ta < 200 are qualitatively similar to those of figures
6.12(a-b), with the main difference being the degree of axial nonuniformity (i.e., contour
curvature) near the groove.
Figure 6.13a shows that as Ta increases, Re increases superlinearly. Figure 6.13b
shows that the variation of the dimensionless torque T is essentially linear with Ta over
the range shown. Finally, figure 6.13c shows that increases rapidly with Ta, possibly
approaching an asymptote.
118
Intermediate-width groove. For a groove of intermediate depth ( = 0.5) and
width ( = 0.5), figures 6.14(a-h) show meridional streamlines and vr2 contours over the
range 0 Ta 200. For Ta = 0 (figure 6.14a), meridional flow in the clearance is
essentially axial, showing curvature only near the mouth of the groove. The groove aspect
ratio is again 1, so that at Ta = 0 the groove accommodates a single vortex. From figure
6.14a, we see that there are again four regions of high vr2 , with the two in the groove
being much weaker than the two in the clearance. As Ta is increased to 10 (figure 6.14b),
the meridional streamlines develop noticeable asymmetry in and near the groove. The
secondary groove vortex in the downstream corner grows, and the dominant groove vortex
undergoes significant deformation. The attachment of the streamsurface separating the
primary groove vortex from the clearance flow has moved downwards from the lip on the
downstream groove wall, and the clearance flow penetrates somewhat more deeply into the
groove, especially in the downstream portion. The Ta = 0 regions of good radial mixing in
the downstream part of the groove and in the upstream part of the clearance have merged,
leaving three regions of high vr2 .
For Ta = 15, figure 6.14c shows an intensification of the interaction between the
meridional clearance flow and the downstream groove vortex, concentrated near the
downstream groove lip. The secondary groove vortex, confined to the downstream corner
at Re = 10, has grown and joined with the clearance flow, with the separation helix on the
downstream groove wall giving way to a saddle just upstream of the wall. At Ta = 25,
the clearance flow is seen to penetrate all the way to the downstream part of the hub,
and the size of the primary groove vortex is clearly diminished. At Ta = 50, figure 6.14e
shows that the primary groove vortex begins to extend into the clearance, as it continues to
diminish axially. At Ta = 100, the upstream separation helix has moved from the lip to the
upstream part of the flight. At this Ta, figure 6.14f shows that the groove vortex is as wide
at the groove mouth as at the hub, and the meridional flow in the clearance is decidedly less
axial. At Ta = 150, the primary groove vortex extends into the clearance about 40% of the
119
radial distance from the flight to the barrel (figure 6.14g), and the meridional streamlines
penetrating into the downstream portion of the groove show signs of incipient "pinching
off. " Figure 6.14h shows that at Ta = 200, the primary groove vortex has grown radially
outward toward the barrel wall, and its most downstream portion extends much closer to
depth (figure 6.12a) there is significantly more contour curvature near the groove mouth
and well into the clearance, for both Ta = 0 and 200.
Figure 6.16a shows that as Ta increases, Re increases superlinearly to about 2.5 at
Ta = 200. This contrasts to the narrow ( = 0.2) groove case of the same depth (figure
6.13a), for which Re is about 0.23 at Ta = 200. Figure 6.16b shows that the increase in
T is essentially linear with Ta over the range shown. Finally, figure 6.16c shows that
increases rapidly with Ta, apparently approaching an asymptote whose value is two orders
of magnitude greater than that found for = 0.2.
considerably wider, especially in the groove, where the radial velocity is very low, even far
from the hub. The two regions with the best radial mixing are in the clearance, near the
upstream and downstream lips of the groove, and have nearly circular cross sections in
the computational plane. Figure 6.17b shows that at Ta = 4, the meridional streamlines
are highly asymmetric. The primary and secondary groove vortices are contained entirely
within the groove, and the secondary vortex in the downstream corner of the groove at Ta =
0 has grown considerably. In a significant volume of fluid in that corner, the magnitude
120
of the radial velocity is smaller than in the Ta = 0 case. Finally, we see that a region
of good radial mixing is beginning to develop in the groove, just inwards of the upstream
lip.
For Ta = 6, figure 6.17c shows that the dominant groove vortex no longer extends to
the downstream wall, and attaches to the hub well upstream of the downstream corner.
The separation streamsurface attaches at or near the upstream groove lip. The (smaller)
downstream groove vortex has given way to a "scouring" flow in which fluid from the
clearance penetrates to the hub. The regions of good radial mixing have cross sections
considerably less circular than at smaller Ta. The upstream region of good radial mixing in
the groove has grown, and radial mixing in the central portion of the groove has intensified
due to penetration of the main flow into the groove.
Figure 6.17d shows that at Ta = 25, a significant intensification of the scouring flow
has occurred, with the separated flow region in the upstream portion of the groove having
smaller axial extent and larger radial extent, respectively. The groove vortex penetrates
slightly into the clearance and radial mixing has intensified near the downstream lip, in the
central part of the groove, and near the upstream groove wall, at the expense of mixing in
everywhere, especially near the upstream and downstream lips, and near the groove mouth.
As Ta increases still further, to 150 and 200 (figures 6.17g and 6.17h, respectively), the
groove vortex penetrates even farther into the clearance, and extends farther downstream
(and closer to the downstream lip), with the separation helix on the upstream flight and
the reattachment on the hub remaining nearly fixed. The downgoing part of the scouring
flow in the downstream portion of the groove now has a substantial "upstream" velocity
121
component, which grows as the scouring flow is channeled through an increasingly narrow
gap between the groove vortex and the downstream lip.
Figures 6.18a and 6.18b show that the very different meridional flows for Ta = 0 and
200 give rise to smaller differences in the azimuthal flow, with the primary effects being
more axial variation in v in the central part of the clearance at Ta = 200, along with
somewhat higher azimuthal velocities in the groove.
Figure 6.19a shows that as Ta increases, Re increases superlinearly from about 0.80
at Ta = 50 to about 2.1 at Ta = 100, and to about 7 at Ta = 200. Figure 6.19b shows
that T increases superlinearly with Ta over the range shown, with its value at Ta = 200
being higher than for either the narrow- or intermediate-width grooves considered at this
groove depth. Finally, figure 6.19c shows that increases rapidly with Ta, with the value
at Ta = 200 ( 100) exceeding the corresponding values for = 0.2 and 0.5.
For deep grooves (i.e., small clearance between the flight and barrel, corresponding to
small values of ) flow in the clearance has relatively little effect on fluid deep in a narrow
groove, but strongly affects flow deep in a wide groove, especially at high Ta. Here, we
consider the effect of groove width and Ta for = 0.2.
Narrow groove. For = 0.2, figure 6.20a shows that the Ta = 0 meridional flow
consists of a nearly axial flow in the clearance, with three recirculating eddies "stacked"
in the groove, reminiscent of the two-dimensional Stokes flow in a driven cavity computed
by Joseph & Sturges (1978). The clearance flow is separated from the outer groove vortex
by a streamsurface attached to the upstream and downstream groove walls near the lips.
This streamsurface is slightly depressed at its center, = 1/2, relative to the attachment
helices on the groove walls. The outer groove vortex is separated from the central one by
another streamsurface having greater curvature than the first. There is very little radial
mixing in the inner part of the groove, and radial mixing in the clearance is weak upstream
and downstream of the groove.
122
For Ta = 25, figure 6.20b shows that the streamsurface separating the clearance flow
from the outer groove vortex has deformed considerably, with the maximum depression
now being located downstream of the midline ( = 1/2). The locations at which this
surface attaches to the groove walls have not moved significantly. The weak innermost
groove vortex has shifted to the downstream corner, and the vortex previously confined
to the center of the groove now penetrates inward to the upstream part of the hub. The
streamsurface separating the outer groove vortex from the flow closer to the hub has become
highly asymmetric and nearly planar over most of its width, with its maximum depression
now being very close to the upstream groove wall. Radial mixing in the groove remains
quite weak beyond about one clearance width from the mouth. Figure 6.20c shows that
as Ta increases to 50, the outer groove vortex extends deeper into the groove, and the
meridional flow in the clearance beyond the groove is less symmetric than at Ta = 25.
For Ta = 100, figure 6.20d shows that the outer groove vortex has become even
more distended, and that the streamsurface separating it from the clearance flow is now
attached to the upstream groove wall very close to the lip. This streamsurface dips sharply
downward just upstream of the downstream groove wall. The upstream region of good
radial mixing in the downstream part of the groove has merged with the one in the upstream
part of the clearance. The result is three regions of good radial mixing: one largely in the
clearance near the downstream lip; one largely in the groove near the upstream lip; and
one between the first two, extending from the clearance upstream of the first region of
upstream wall, and protruding increasingly into the clearance flow as Ta increases. The
clearance flow now penetrates deeply into the downstream side of the groove. Some of
the fluid passing over that vortex in the clearance enters the groove, penetrates more than
123
halfway to the hub (as far as the dominant vortex), and rejoins the clearance flow near the
downstream lip. Between this flow and the dominant vortex, on the one hand, and the
hub on the other, lie two relatively weaker counter-rotating vortices.
Figures 6.21a and 6.21b show that there is remarkably little difference in the azimuthal
velocity between the Ta = 0 and Ta = 200 cases. The distribution, which is exactly
symmetric about = 1/2 for Ta = 0, is nearly symmetric for Ta = 200, notwithstanding
significant asymmetry in the meridional flow, both in the clearance and near the mouth of
the groove.
Figure 6.22a shows that as Ta increases, Re increases superlinearly from about 0.049
at Ta = 25 to about 0.2 at Ta = 100, and to about 0.42 at Ta = 200. Figure 6.22b shows
that T is nearly linearly proportional to Ta over the range shown. Finally, figure 6.22c
shows that increases rapidly with Ta, with the value at Ta = 200 being 0.14.
helical vortex near the hub. The streamsurface separating the clearance flow from the
outer groove vortex is again attached to the sidewalls near the lips. In contradistinction
to the narrow groove case for Ta = 0, the outer separation streamsurface near the groove
mouth is neither convex nor concave. Radial mixing is again concentrated in two regions
of the clearance near the lips, and in two regions of the groove, also near the lips.
Figure 6.23b shows that the meridional flow for Ta = 25 is quite different from the
Ta = 0 flow. In particular, the flow in the groove now consists of a single separated region,
attached to the upstream wall of the groove very close to the lip, and to the hub near its
center. This vortex penetrates slightly into the clearance flow, and extends nearly to the
downstream wall. The clearance flow passes into the groove through the gap between the
attached groove vortex and the downstream groove wall. Compared to the Ta = 0 case,
124
radial mixing is enhanced near the center of the groove mouth and in the upstream part
of the groove near the lip, but is suppressed in the part of the groove near the downstream
lip.
As Ta increases to 50 and 100, figures 6.23c and 6.23d show that the attached
groove vortex penetrates more deeply into the clearance flow, and that the gap between
its downstream end and the downstream wall widens, allowing for considerably stronger
"scouring" of the downstream half of the groove. Also, note that the locations at which
the separating streamsurface attaches (i.e., close to the upstream lip, and near the center
of the hub) move very little. The upstream and downstream regions of good radial mixing
become much less asymmetric about = 1/2 as Ta increases.
Figures 6.23e and 6.23f show that at even higher Ta (150 and 200, respectively),
the attached vortex penetrates still further into the clearance (some 45% of the distance
from the flight to the barrel at Ta = 200), but that the gap between that vortex and the
downstream lip has ceased to narrow. Good radial mixing obtains in a very large part of
the domain, with the vr2 contours being approximately symmetric about = 1/2.
Figures 6.24a and 6.24b show v contours for Ta = 0 and 200. The results are
qualitatively similar to the narrow groove case (figures 6.21a and 6.21b) except that the
azimuthal velocity contours show much more curvature near the groove mouth and into
the clearance. As was the case for narrow groove, v contours for Ta values between those
shown are qualitatively similar to those of figures 6.24(a-b), with the main difference being
the degree of axial nonuniformity (i.e., contour curvature) near the groove mouth and into
the clearance.
Figure 6.25a shows that as Ta increases, Re increases superlinearly to about 3.1 at
Ta = 200. This contrasts to the = 0.2 case (figure 6.22a) in which Re is about 0.42 at
Ta = 200. Figure 6.25b shows that the increase in T is essentially linear with Ta over
the range shown, reaching a value of about 1.77 104 at Ta = 200. Finally, figure 6.25c
shows that again increases rapidly with Ta.
125
Wide groove. For a deep ( = 0.8) wide groove ( = 0.8), figure 6.26a shows that
the meridional streamlines for Ta = 0 consist of a primarily axial flow in the clearance,
and a primary helical groove vortex, separated by a streamsurface attached to the groove
walls. There are four regions of good radial mixing, two in the groove beginning near the
lip and extending inward toward the hub, and two smaller regions in the clearance near
the lips.
As Ta increases to 25, figure 6.26b shows that the primary groove vortex is largely
confined to the upstream part of the groove, and a scouring flow reaches the hub
downstream. The primary vortex penetrates into the clearance, and attaches very close
to the upstream lip. The upstream region of good radial mixing in the clearance has
weakened, while radial mixing in the upstream part of the groove has strengthened. Two
other regions of good radial mixing are apparent. One is just upstream of the downstream
lip, extending from the middle of the clearance inward along the downstream groove wall.
The other occupies the central part of the domain, extending from the clearance just
beyond the mouth well inwards toward the hub.
Figures 6.26(c-f) show that the large upstream groove vortex occupies a progressively
smaller portion of the groove at Ta = 50 and 100, attaching to the hub somewhat upstream
of the Ta = 25 attachment. This vortex extends farther into the clearance at Ta increases.
The regions of good radial mixing are much less asymmetric, and strengthen considerably
as Ta increases. At sufficiently high Ta, the streamsurface separating the groove vortex
from the clearance flow attaches to the flight just upstream of the lip. Again, a large
fraction of the domain is subject to good radial mixing at larger Ta.
Figures 6.27a and 6.27b show v contours for Ta = 0 and 200. The contours show
that in the groove there is considerably less axial variation near the hub at Ta = 200 than
at Ta = 0, while near the mouth, there is more axial variation at Ta = 0. At any given
distance from the hub, the azimuthal velocity on the symmetry line = 1/2 is higher at
Ta = 200 than at Ta = 0.
126
Figure 6.28a shows that as Ta increases, Re increases superlinearly to about 7.6 at
Ta = 200. This contrasts to the = 0.2 case (figure 6.25a) in which Re is about 3.1
at Ta = 200. Figure 6.28b shows that T increases superlinearly with Ta over the range
shown, reaching a value of about 1.47104 at Ta = 200. Finally, figure 6.28c shows that
increases rapidly with Ta, apparently approaching an asymptote whose value is two orders
of magnitude greater than that found for = 0.2 (i.e., narrow groove).
6.5. Discussion
6.5.1. Trends with geometric variation
In this section, we describe the main features of how the Ta dependence of the
meridional and azimuthal flow, the mean axial flow (characterized by a Reynolds number),
and the dimensionless torque per unit length and radial mixing depend on geometry.
Meridional components of the helical flow. For Ta = 0, the Stokes flow is fore-
aft symmetric about the = 1/2 plane. With the exception of the two widest grooves
( = 0.5 and 0.8) in the shallow ( = 0.8) groove case, the flow at Ta = 0 always consists
of a nearly axial meridional flow in the clearance, separated from a vortical flow occupying
all but the outermost part of the groove. When the groove aspect ratio is on the order
of unity, flow in the groove is dominated by a single primary vortex with much smaller
secondary corner vortices. For the two exceptional cases ( = 0.5 and 0.8, with = 0.8),
the groove aspect ratio is significantly less than unity, and the Ta = 0 flow in the groove
consists of two relatively large corner vortices, with the clearance flow penetrating to or
almost to the hub.
With the exception of the narrow ( = 0.2) groove cases, the symmetric Ta = 0
flow gives rise at Ta = 25 to a decidedly asymmetric flow in which a single groove vortex
attaches at or near the upstream lip and to the hub. This vortex occupies the upstream
part of the groove. In the narrow groove cases, meridional flow in the clearance at Ta = 25
is separated from the groove flow by a streamsurface that attaches on the groove walls near
the lips and dips slightly more into the groove near the downstream lip than anywhere else.
127
(For the narrow groove, such a flow develops by Ta = 150, except for the deep ( = 0.2)
groove considered. In the latter case, reattachment occurs on the upstream groove wall
rather than on the hub.)
The most prominent characteristic of the primary vortex is its penetration into the
clearance as Ta increases. In each case, the separation streamsurface is nearly flush with
the flight at Ta = 25, and deflects increasingly into the clearance as Ta increases. In all
cases except for the narrow ( = 0.2) groove, this deflection and the resulting penetration
of the groove vortex into the clearance ultimately lead to significant radial flow in a large
fraction of the clearance (figures 6.5f, 6.8f, 6.14f, 6.17f, 6.23f, and 6.26f).
In all but the two deepest ( = 0.2 and 0.5) of the narrow ( = 0.2) groove cases, the
hub downstream of the reattachment of the primary vortex is scoured by fluid originating
from and returning to the clearance. The strength of this flow depends on the width
of the gap between the primary vortex and the downstream groove wall. Note that the
width of this gap sometimes increases with increasing Ta, corresponding to a reduction
in the axial extent of the primary vortex (figures 6.2e and 6.2f, 6.21e and 6.21f, 6.23(b-f),
6.26(b-d), and 6.11e and 6.11f). In other cases, the gap between the primary vortex and
the downstream groove wall narrows as Ta increases (figures 6.5(b-f), 6.8(b-f), 6.14(b-f),
and 6.17(c-f)).
Finally, we observe that in several cases, the upstream attachment moves upstream of
the lip and onto the flight as Ta increases. This movement is associated with significant
penetration of the primary groove vortex into the clearance. However, we note that once
the attachment has moved from the groove wall to the lip, it is not displaced significantly
upstream regardless of the extent to which the primary vortex extends into the clearance
(figures 6.5(d-f), 6.8(d-f), 6.14(d-f), 6.17(e-g), 6.20(c-f), 6.23(b-f), and 6.26(b-f)).
128
is approached. The azimuthal component of the flow in the groove is somewhat more
complex since the no-slip condition must be satisfied on the groove walls. Only when the
groove is relatively wide and deep (figures 6.18a, 6.24a, and 6.27a, corresponding to the
wide and intermediate-depth case, the deep and intermediate-width case, and the wide
and deep case, respectively) does the azimuthal velocity anywhere in the groove exceed
approximately 10% of the maximum value (at the barrel).
The deviations from circular Couette flow in the clearance increase as the mouth is
approached and increase with increasing groove width. For the wide and shallow groove
( = 0.8, = 0.8), the azimuthal flow in the groove has very little variation except near
the groove walls. This is clearly associated with the fact that for this geometry the narrow
flights serve to perturb what would otherwise be circular Couette flow between the hub
and the barrel. Azimuthal flow in the clearance at Ta = 200 differs from the Ta = 0 flow
primarily in the absence of symmetry about = 1/2, and in the larger departures from
circular Couette flow, especially when the groove vortex extends far into the clearance.
With the exception of the two deep ( = 0.2 and 0.5) wide ( = 0.8) groove cases,
the azimuthal flow in the groove is remarkably similar at Ta = 0 and 200. The primary
difference is a reduction in the axial variation of the azimuthal component of the helical
flow near the hub at Ta = 200 compared to the Stokes flow case, especially for grooves
of sufficient depth and width (see figures 6.18a and 6.18b, 6.24a and 6.24b, and 6.27a and
6.27b).
While the azimuthal velocity contours at Ta = 200 are always asymmetric about
= 1/2, the asymmetry is much greater in the clearance (especially near and just beyond
the mouth) than in the groove. This is to be expected since the velocities in the groove are
generally much smaller than in the clearance, corresponding to relatively smaller departures
from the (linear) Stokes flow that gives rise to symmetric behavior at Ta = 0.
129
Radial mixing. In this section we discuss how , the global measure of radial
mixing, depends on Ta and the geometric parameters, and relate this dependence to the
corresponding variations in the pointwise measure vr2 .
We begin by noting that is defined as the volumetric average of the square of the
dimensional radial velocity component, normalized with a viscous velocity scale. Thus,
it vanishes as Ta 0. In each case, initially increases rapidly with increasing Ta.
Measured by , the poorest radial mixing occurs with the narrow and shallow ( = 0.8,
= 0.2) groove, followed closely by the narrow and deep ( = 0.2, = 0.2) groove. The
third worst case is the narrow groove of intermediate depth ( = 0.5, = 0.2). For each
of these cases, lies between approximately 0.1 and 0.2 at Ta = 200. Compared to for
other geometries, this relatively low value is associated with relatively shallow penetration
of the primary vortex into the clearance and weak meridional flow in most of the groove,
as reflected in the vr2 contours and meridional streamlines. For each groove depth, at
each Ta increases monotonically with groove width. For all but the narrow ( = 0.2)
grooves, increases with groove depth at each Ta. This corresponds to regions of good
radial mixing occupying a larger fraction of the domain as decreases.
temporal complexity and lower spatial symmetry as Ta increases. This will give rise to
different radial mixing behavior than that which would be predicted by an extension of
the present analysis to higher Ta.
Mean axial flow. We characterize the mean axial flow by a Reynolds number based
on the mean axial velocity and the gap between the barrel and flight. For = 1/2 and
fixed Ta, comparison of velocities for different geometries requires that Re be divided by
130
. To compare volumetric flow rates at = 1/2, one should multiply Re by (2 + /
)/(2) to account for the dependence of the area ratio on the geometric parameters.
In each case, Re increases superlinearly from zero as Ta increases. For each groove
depth, Re increases monotonically with groove width at each Ta. Also, Re increases
transfer occurs. We note that the degree of nonlinearity increases with increasing groove
width at each groove depth.
Dimensionless torque per unit length. The dimensionless torque per unit length
T (6.4.1) is defined using a scaling that allows direct comparison of results for different
geometries. For each geometry, the increase in T with increasing Ta is at least slightly
superlinear. This reflects the increased radial variation of the azimuthal velocity as Ta
increases, as well as the increased axial variation of the azimuthal velocity component
on the sidewall of the groove. At each groove depth, T decreases monotonically with
groove width at each Ta. At each groove width, T increases rapidly with increasing
groove depth at each Ta. This is a direct consequence of the greater radial gradient of the
azimuthal velocity component between the rotating barrel and the stationary flight over a
progressively smaller gap.
One notable feature of the meridional flow in certain regimes is its apparent
resemblance to two-dimensional flow over a rectangular cavity. For deep narrow grooves
(6.4.3), figure 6.20a shows that at Ta = 100, flow in the groove consists of a "stack"
of three helical vortices, separated from the main clearance flow by a nearly cylindrical
separation streamsurface that attaches to the groove walls near the lips. From the
standpoint of the meridional flow, this streamsurface is much like a moving lid in a two-
131
dimensional cavity flow. The flow is qualitatively similar to the Stokes-flow calculation of
Joseph & Sturges (1978).
In other regimes, the meridional flow superficially resembles flow over a two-
dimensional backward-facing step (Armaly et al. 1983). The obvious differences between
these two flows include the fact that the present flow is driven by rotation of a screw
about an axis lying in the meridional plane, where as flow over a backward-facing step
is driven by a streamwise pressure gradient. For the two-dimensional geometry, the two-
dimensional flow that is observed for sufficiently small Reynolds numbers (Gresho et al.
1993) ultimately loses its stability, and becomes three-dimensional and unsteady (Armaly
et al. (1983)). For the rotating screw case, however, the azimuthal flow perpendicular
to the meridional plane (and hence the three-dimensionality of the flow) is central to the
development of the helical vortex that attaches near the upstream lip.
Another limiting geometry is that in which the pitch vanishes. In that case, the
groove is axisymmetric. This is precisely the case of interest for axisymmetric labyrinth
seals. We are aware of no previous computations on flows in this geometry in the laminar
regime. The closest related work of which we are aware is that of Raspo & Crespo del
Arco (2003), who computed axisymmetric flows between a rotating inner circular cylinder
and a rotating outer cylinder with a single groove of rectangular cross section.
profiles are qualitatively similar to ours, but do not allow for direct comparison because
of the different geometry and the relatively small length of the flow facility, which lead to
counter-current flow in the test section. (Note that for a double-helix screw with the two
grooves one hundred and eighty degrees out of phase, there is a flow fully-developed in the
helical direction such that in any plane perpendicular to the screw axis flow at each point
is identical to the flow at the same distance from the center line on a opposite diameter.
132
However, the double-helix flow can not be related to the single helix flow because in the
double-helix case, both grooves affect the flow.)
of geometries and rotation rates for applications. The details will, of course, depend on
the precise objective. For example, when enhancement of mass transfer from the main
flow in the clearance to the hub is desired, one would like to generate a flow in which
a large fraction of the hub is "scoured." On the other hand, if one wants to perform
transfer problems of interest involve streamwise depletion or buildup along the length of
the passage, so that entry effects are invariably important. Whether or not hydrodynamic
entry effects, not accounted for in this helically fully-developed analysis, are important will
depend on the application.
6.6. Conclusions
The results presented 6.4 and discussed in 6.5 show that a variety of helically fully-
developed flows can be generated by coaxially rotating a screw-like shaft with a single
133
helical groove of rectangular cross section in a circular cylinder. The computed flows
suggest that a wide variety of mass transfer regimes can be achieved by varying the
geometry and the Taylor number.
We note that the flows shown were computed over a relatively limited range of Taylor
number, with two of the four dimensionless geometric parameters held fixed. Consideration
of a wider range of Taylor numbers and of additional geometries associated with variation of
the two other dimensionless geometric parameters can be expected to significantly increase
the variety of accessible flows and mass transfer regimes.
134
Chapter 7
Summary
of Goldstein in the 1930s and Chandrasekhar and others beginning in the 1950s. Until
now, the stability of this flow has not been understood over the full range for which the
base flow is linearly stable.
In chapter 2, we consider the stability of SPF for = 0.5 and several rotation rate
ratios . For < 2 , we have made the connection between the onset of instability for
circular Couette flow with no mean axial flow, and the onset of instability in annular
Poiseuille flow absent rotation. In each case, Ta crit reaches a plateau before falling
precipitously to zero as Re approaches the critical value for (nonrotating) annular Poiseuille
flow. On the other hand, for > 2 , no instability is possible on a linear basis for
0 Re < Re min , where Re min denotes the location of a turning point (for the special case
= 1, there is a vertical asymptote whose value is shown to differ from that previously
found) beyond which the stability boundary is multi-valued and the flow is stable in two
for each and experimental aspect ratio, there is a range of Re for which finite-amplitude
instability does not occur, and for which the annulus is long enough to allow development
of secondary flow to detectable amplitudes. Data for a narrow-gap ( = 0.955) annulus
of large aspect ratio (> 570), for which agreement between experiment and computation
is essentially exact up to Re = 325, strongly suggest that finite-amplitude instability does
not occur over a wide range of Re when the outer cylinder is fixed. Similar comparison
135
to 6= 0 data for = 0.77 and 0.95, both up to Re > 100, suggests in both cases the
existence of a substantial range of Re and in which finite-amplitude instability does not
occur.
In chapter 4, it is shown that the stability boundary for = 0.1 is quite different from
those for 0.5. One key difference is the absence of a transition from centrifugal to
Tollmien-Schlichting-like instability at high Re. For = 0.1, there is no critical Re beyond
which the nonrotating flow is unstable. Like circular Poiseuille flow, annular Poiseuille flow
for = 0.1 is linearly stable for all Re, so that at each rotation rate ratio, SPF is linearly
the bifurcation behavior and solution structure compared to the smooth shaft case. For a
corrugated inner cylinder and small Ta, there is a single branch of solutions corresponding
to a modified Couette flow (i.e., nonzero axial and radial velocity components much smaller
than the azimuthal component). For large-amplitude radius variation, a single pair of
vortices is found at each Ta. Since solutions with two or more vortex pairs (found at
smaller ) necessarily have vortex pairs of different sizes, this suggests that beyond some
critical corrugation amplitude, such solutions either a) become disallowed; b) appear at a
Ta that is very large; or c) are not found by our solution technique, possibly because its
136
computed flows suggest that a wide variety of mass transfer regimes can be achieved
by varying the geometry and the Taylor number. We note that the flows shown
were computed over a relatively limited range of Taylor number, with two of the four
dimensionless geometric parameters held fixed. Consideration of a wider range of Taylor
numbers and geometries associated with variation of the two other dimensionless geometric
parameters can be expected to significantly increase the variety of accessible flows and mass
transfer regimes.
137
Chapter 8
Tables
Mp Mp
10 1.78 102 20 5.63 107
12 7.77 103 22 1.11 107
14 1.24 103 24 1.5 108
16 8.06 105 26 1.0 109
18 1.39 106
138
< m < m0
Ta Recrit kcrit mcrit Rem0 km0 mm0
100 115.5 0.9839 6 151.1 1.016 5
225 87.65 0.8087 8 103.2 0.8844 6
300 85.18 0.6265 8 99.97 0.6968 6
400 83.88 0.4778 8 98.23 0.5371 6
1000 82.55 0.1944 8 96.41 0.2201 6
2000 82.36 0.09740 8 96.16 0.1108 6
Table 2.2a. Minimum neutral values of Re (and corresponding values of k and m) for
different ranges of m at selected T a with = 1 and = 0.5.
< m < m0
Re T acrit kcrit mcrit T am0 km0 mm0
150 79.82 0.9200 6 100.7 1.018 5
200 71.52 0.7550 6 84.51 0.8695 5
300 66.82 0.5310 6 76.16 0.6257 5
500 64.70 0.3263 6 72.55 0.3891 5
1000 63.86 0.1640 6 71.13 0.1975 5
Table 2.2b. Minimum neutral values of T a (and corresponding values of k and m) for
different ranges of m at selected Re with = 1 and = 0.5.
139
Ta 0.9 1.0 1.2
300 80.74 (-8) 85.18 (-8) 92.33 (-6)
1000 86.31 (-6) 82.55 (-8) 132.9 (-5)
3000 144.2 (-5) 82.33 (-8) 264.0 (-5)
Table 2.3. Reynolds numbers on the stability boundary and azimuthal wavenumbers
(in parentheses) for = 0.5 and selected values of and T a.
140
Authors Re T aexpt
crit T acomp
crit
Table 3.1. Comparison of experimental and computed values of T acrit for = 0. Parentheses
denote values of differently defined Taylor numbers read from figures of other authors.
141
Authors Re T aexpt
crit T acomp
crit
142
T aexpt
crit T acomp
crit mcomp
crit T aexpt
crit T acomp
crit mcomp
crit
Re = 24
-0.37 33.64 33.70 1 0.33 39.58 39.97 0
-0.26 32.52 32.29 1 0.41 46.20 46.18 0
0 31.05 31.86 1 0.46 55.17 53.01 0
0.20 34.64 34.86 1 0.48 76.06 57.03 0
Re = 33
-0.52 38.58 39.10 2 0.17 36.93 38.65 1
-0.37 36.64 36.89 2 0.26 40.61 41.62 1
-0.19 35.22 35.67 2 0.43 55.64 55.07 1
0 33.61 35.97 1 0.50 75.12 70.82 1
0 34.58 35.97 1
Re = 49
-1.01 54.73 54.10 4 0.11 42.81 46.36 4
-0.68 43.49 46.58 4 0.20 45.02 49.03 4
-0.36 40.81 43.14 4 0.25 46.93 51.12 5
-0.13 40.40 43.15 4 0.33 51.73 55.65 5
0 40.81 44.34 3 0.50 77.23 76.96 8
0 41.78 44.34 3
Re = 63.5
-1.01 73.41 59.17 6 0 45.73 49.33 7
-0.94 55.38 55.13 6 0.17 48.99 52.62 8
-0.41 45.79 47.85 6 0.21 50.20 53.79 8
-0.26 44.76 47.60 6 0.24 51.49 54.80 9
-0.17 44.73 47.89 6 0.27 53.11 55.86 9
-0.10 45.37 48.29 7 0.52 77.23 74.13 11
Re = 65.5
-1.05 60.50 58.10 6 0.14 50.46 52.31 8
-0.48 48.26 48.77 6 0.19 51.64 53.59 9
-0.27 47.11 48.15 7 0.28 56.05 56.51 9
-0.20 47.37 48.26 7 0.39 62.82 61.88 10
-0.11 47.52 48.72 7 0.44 69.14 65.36 11
0 47.87 49.88 7 0.49 72.82 69.76 11
0.090 48.99 51.25 8
Re = 82.5
-1.13 65.03 62.90 7 0 51.23 53.01 10
-0.58 51.88 53.31 8 0.13 52.70 54.58 11
-0.41 51.05 52.20 8 0.17 52.70 55.23 11
-0.30 51.05 51.86 9 0.21 54.58 56.00 11
-0.24 50.90 51.84 9 0.24 54.94 56.64 12
-0.13 50.61 52.17 10 0.44 63.85 62.65 13
-0.072 50.61 52.45 10 0.56 67.67 68.81 14
143
T aexpt
crit T acomp
crit mcomp
crit T aexpt
crit T acomp
crit mcomp
crit
Re = 106
-0.93 65.03 62.18 9 0.095 56.11 55.91 13
-0.74 59.35 59.13 10 0.15 56.64 56.31 14
-0.51 56.88 56.70 11 0.24 56.94 57.02 14
-0.26 56.35 55.39 12 0.37 58.97 58.65 14
-0.14 55.17 55.23 12 0.59 64.85 63.33 15
0 55.20 55.47 13 0.81 78.79 77.32 15
Re = 134.75
-0.78 70.47 63.36 12 0.56 57.88 58.99 16
-0.37 60.02 58.81 14 0.65 61.64 60.82 16
-0.26 59.29 58.04 14 0.87 70.03 75.10 15
-0.19 58.61 57.67 14 0.94 76.21 86.55 15
-0.054 57.52 57.09 15
Re = 166
-0.85 77.53 67.65 14 0.14 58.11 56.88 17
-0.67 69.14 64.79 15 0.23 57.97 56.49 17
-0.39 63.94 61.24 16 0.34 57.76 56.24 17
-0.23 63.11 59.60 16 0.48 56.79 56.49 17
-0.15 62.32 58.90 16 0.71 57.17 59.85 16
-0.083 60.52 58.38 16 0.84 67.70 66.67 15
0 60.41 57.75 17 0.96 69.59 80.01 15
0.091 58.29 57.15 17
Re = 244
-0.79 82.50 71.27 18 0.15 59.82 56.69 18
-0.70 74.65 69.52 18 0.29 57.88 55.50 18
-0.39 71.50 63.99 19 0.53 54.82 54.97 17
-0.25 66.94 61.78 19 0.77 54.73 59.02 16
-0.13 64.00 60.05 19 0.93 63.64 68.82 15
-0.058 63.05 59.09 19 1.01 67.82 77.81 14
0 62.38 58.36 19
Re = 330
-0.82 81.94 74.48 21 0.15 61.35 56.53 19
-0.67 79.29 71.05 21 0.27 57.67 55.32 19
-0.35 69.79 64.45 20 0.45 52.96 54.27 18
-0.18 67.29 61.35 20 0.67 50.17 55.48 17
-0.071 65.82 59.58 20 0.80 51.79 58.99 16
0 64.44 58.54 20 0.96 56.35 68.90 15
0 65.26 58.54 20 1.11 61.94 89.43 13
0.079 62.53 57.43 19
Re = 403.5
-0.63 79.65 71.05 22 0.20 60.17 55.87 19
-0.39 75.76 65.70 21 0.72 47.82 56.04 16
-0.29 71.29 63.67 21 0.84 48.85 60.30 16
-0.13 69.59 60.72 20 1.01 55.32 72.65 14
0 66.70 58.55 20 1.24 61.64 96.71 11
0.097 63.64 57.18 20
144
T aexpt
crit T acomp
crit mcomp
crit T aexpt
crit T acomp
crit mcomp
crit
Re = 5
-0.738 12.19 11.78 3 0.238 9.628 9.169 0
-0.436 10.17 9.653 1 0.425 10.47 10.01 0
-0.243 9.436 9.026 0 0.668 13.45 12.94 0
0 8.936 8.809 0
Re = 10
-0.731 12.30 11.78 3 0.234 9.790 9.416 1
-0.428 10.37 9.847 1 0.414 10.76 10.22 0
-0.234 9.794 9.243 1 0.656 13.70 13.05 0
0 9.263 9.053 1
Re = 20
-0.924 16.21 15.33 5 0.204 12.08 11.56 1
-0.710 14.10 13.69 4 0.369 13.26 12.38 1
-0.393 12.48 11.86 2 0.620 15.99 15.61 1
-0.220 11.72 11.34 2 0.750 17.87 17.49 1
0 11.35 11.18 1 0.798 22.18 20.37 0
Re = 40
-1.06 16.78 16.30 9 0 14.74 14.79 3
-0.863 17.30 17.01 8 0.157 15.15 15.19 3
-0.631 15.86 15.88 7 0.299 16.28 15.95 2
-0.331 14.94 14.95 5 0.680 19.71 20.11 2
-0.168 14.84 14.73 4
Re = 80
-1.91 23.46 25.50 18 0.117 20.99 20.84 29
-1.46 22.82 22.18 19 0.204 21.67 21.49 31
-1.09 20.40 20.33 20 0.383 23.28 23.45 36
-0.894 19.95 19.69 21 0.531 25.28 26.12 42
-0.687 19.48 19.28 22 0.628 28.65 28.85 48
-0.459 19.40 19.15 23 0.709 32.35 32.23 55
-0.233 19.55 19.40 25 0.823 40.88 40.79 69
-0.119 19.87 19.71 26 0.887 58.58 50.80 80
0 20.14 20.18 28
Re = 120
-1.32 25.39 24.49 30 0.185 25.11 26.08 54
-0.932 24.00 23.39 33 0.392 23.26 28.47 63
-0.755 23.69 23.17 35 0.464 28.92 29.63 67
-0.565 23.63 23.15 38 0.573 31.34 31.90 73
-0.384 23.73 23.36 41 0.673 34.14 34.78 81
-0.195 24.00 23.85 44 0.801 42.06 40.44 92
-0.085 24.27 24.30 47 0.901 58.31 48.68 98
0 24.60 24.74 49
Re = 200
-1.12 29.89 29.42 55 0 30.33 31.15 83
-0.748 29.89 29.17 62 0.146 30.53 32.08 89
-0.595 29.89 29.27 66 0.293 31.01 33.28 95
-0.448 29.79 29.50 69 0.417 32.19 34.55 101
-0.306 29.76 29.86 73 0.520 34.38 35.81 105
-0.150 30.06 30.42 78 0.603 37.71 37.00 109
-0.068 30.27 30.79 81 0.746 44.76 39.60 112
145
Recrit kcrit mcrit ccrit
0.14 42286 0.55488 1 0.344
0.15 28660 0.67428 1 0.366
0.16 21277 0.77912 1 0.385
0.17 16900 0.87593 1 0.400
0.18 14170 0.95879 1 0.412
0.19 12432 1.0331 1 0.421
0.2 11326 1.0988 1 0.426
0.3 11475 1.4350 1 0.393
0.4 14552 1.6307 1 0.354
0.5 10359 1.4786 2 0.404
0.6 10296 1.7175 2 0.385
0.77 8883.3 1.9974 1 0.383
0.8 8548.4 2.0204 0 0.386
0.95 7739.5 2.0399 0 0.395
146
Nr Nz
4 8 16 32 64
4 11663 11857 11881 11887
8 11670 11212 11273 11278 11279
16 11050 10563 10573 10570 10570
32 10880 10418 10412 10406 10406
64 10852 10397 10388 10382 10382
128 10379 10378
Table 5.1. T as a function of the number of elements in the radial and axial
directions for T a = 300, = 2, and = 0; the dash denotes combinations
of Nz and Nr for which Newton iteration would not converge.
147
Nr Nz
4 8 16 32 64 128
4 18928 18980 19024 19032
8 21397 15666 14315 14332 14342
16 19620 14856 13633 13623 13625
32 19130 14587 13487 13468 13467
64 14531 13465 13446 13444 13443
128 13441 13440
Table 5.2. T as a function of the number of elements in the radial and axial
directions for T a = 300, = 4, and = 0.25; dashes denote
combinations of Nz and Nr for which Newton iteration
would not converge.
148
Ta Texpt
T
149
Symbol Definition Name
Ro /L aspect ratio
Ri /Ro hub ratio
(Ro R1 )/(Ro Ri ) clearance ratio
l/L groove width ratio
150
Chapter 9
Figures
120 5
100
4
80
3
T acrit
kcrit
60
2
40
1
20
rag replacements
151
a) 0 c) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
8 2
6
1.5
4
mcrit
ccrit
1
0.5
0
b) -2 d) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 2.1. For = 0 and = 0.5: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
180 5
150
4
120
3
T acrit
kcrit
90
2
60
1
30
rag replacements a) c)
0 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
152
8 3
2
4
mcrit
ccrit
2
1
b) -2 d) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 2.2. For = 0.2 and = 0.5: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
105 3
104 2
T acrit
kcrit
103 1
102 0
rag replacements a) 10 1 c) -1
101 102 103 104 105 101 102 103 104 105
Re Re
153
8 20
10
4
0
0
mcrit
ccrit
-10
-4
-20
-8
-30
b) -12 d) -40
101 102 103 104 105 101 102 103 104 105
Re Re
Figure 2.3. For = 0.5 and = 0.5: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
replacements
700
6 7 8
9
5 10
11
500
4
12
13
Ta
3 14
3
300 4
5
7 8 9
154
100
0 2 4 6
Figure 2.4. Neutral curves for Re = 100 and = = 0.5, corresponding to azimuthal
wavenumbers in the ranges 14 m 3 (dashed) and 3 m 9 (solid).
140 5
120
4
100
3
80
T acrit
kcrit
60
2
40
1
20
rag replacements a) c)
0 0
100 10 1
10 2
10 3
10 4
10 5
100 101 102 103 104 105
Re Re
155
10 10
8 8
6 6
mcrit
ccrit
4 4
2 2
b) 0 d) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 2.5. For = 0.5 and = 0.5: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
3
10
0.5
0.2
0.5
T acrit
102
0
replacements
1
10
1 2 3 4 5
10 10 10 10 10
Re
Figure 2.6. Critical T a versus Re for = 0.5 and = 0.5, 0.2, 0, and 0.5.
156
1000
0 1 2
3
4
replacements
800 5
600
7
400
T acrit
8
7
6
200 5
3 4
2
-200
-400
0 25 50 75 100 125
Re
157
4
10
3
10
m=5
2
Re
10
m=0
1
10 m=2
replacements m=1
0
10
0 2 4 6 8
158
120
2
60 1
0
Ta
159
-60
-120
-25 -12.5 0 12.5 25
PSfrag replacements
Re1/3
Figure 2.9. Stability boundary for = 0 and = 0.5, for positive and negative T a and Re.
70 5
60
4
50
3
40
T acrit
kcrit
30
2
20
1
10
rag replacements a) c)
0 0
100 10 1
10 2
10 3
10 4
10 5
100 101 102 103 104 105
Re Re
160
25 3
20 2.5
15 2
mcrit
ccrit
10 1.5
5 1
0 0.5
b) -5 d) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 3.1. For = 0 and = 0.77: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
60 5
50
4
40
3
T acrit
kcrit
30
2
20
1
10
rag replacements
a) 0 c) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
161
25 3
20
15 2
mcrit
ccrit
10
5 1
b) -5 d) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 3.2. For = 0.2 and = 0.77: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
80 5
4
60
3
T acrit
kcrit
40
20
1
rag replacements a) 0 c) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
162
30 2
25
1.5
20
15
mcrit
ccrit
1
10
5
0.5
b) -5
d) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 3.3. For = 0.5 and = 0.77: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
50 6
5
40
4
30
T acrit
kcrit
3
20
2
10
1
rag replacements a) 0 c) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
163
160 4
120
3
80
mcrit
ccrit
2
40
1
0
b) -40 d) 0
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 3.4. For = 0 and = 0.95: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
100
330
403.5
rag replacements
49
330
75 65.5
244 403.5
i (Ro Ri )2 /
166
82.5 134.75
106
65.5
49
164
50
63.5
33
24
25
-100 -50 0 50 100 150
o (Ro Ri )2 /
Figure 3.5. SPF stability boundaries for = 0.77 and the values of Re (shown adjacent to each curve)
and investigated experimentally by Mavec (1973). The dashed line corresponds to = 2 .
450
375
300
Re
225
150
165
75
PSfrag replacements
0
-1.5 -1 -0.5 0 0.5 1 1.5
Figure 3.6a. Nature of transition in -Re plane for = 0.77 data of Mavec (L/(R o Ri ) = 160):
@ linear onset; subcritical onset; O apparent transition delay; competition
between subcritical instability and apparent transition delay.
250
200
150
Re
100
166
50
PSfrag replacements
0
-2 -1.5 -1 -0.5 0 0.5 1
Figure 3.6b. Nature of transition in -Re plane for 0.95 data of Snyder (285 L/(R o Ri ) 349):
@ linear onset; subcritical onset; O apparent transition delay; competition
between subcritical instability and apparent transition delay.
250
0.95
200
1
T acrit [/(1 )] 2
150
0.77
100
0.5
frag replacements
50
0
1 2 3 4 5
10 10 10 10 10
Re
1
Figure 3.7. Critical T a[/(1 )] 2 versus Re for = 0 and = 0.5, 0.77, and 0.95.
167
2000 7
B B0 A0
6
1500 5
4 A
T acrit
rag replacements
kcrit
1000 A 3 B
500 1
a) 0 c) -1
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
168
3 2
B0
0
2 A 1.75
A0
mcrit
ccrit
1 B0 A 1.5
A
0 B 1.25 B
b) -1 d) 1
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 4.1. For = 0 and = 0.1: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
200
150 0.1
T acrit
100
0.5
replacements 50
0.77
0.95
0
1 2 3 4 5
10 10 10 10 10
Re
Figure 4.2. Critical T a versus Re for = 0 and = 0.1, 0.5, 0.77, and 0.95.
169
20000
16000 m = 1
m=0
m = 2
12000
Ta
8000
m=1
replacements
4000
m=2
m=1 0
0 10 20 30
170
3000 102
B
2500
A B
1 A
10 B0
kcrit
1500
A0
-1
10
1000
10-2
500 C
a) 0 c) 10 -3
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
171
3 10
B0
A
A0 0 B
2 C C
-10
mcrit
ccrit
1 B0 A
-20
0 B
-30
A0
b) -1 d) -40
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 4.4. For = 0.25 and = 0.1: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
2
B0
1 A
B
L
ccrit
replacements
-1
m=2
m=3
m=0 -2
0 1 2 3 4 5
10 10 10 10 10 10
Re
Figure 4.4e. Critical wavespeed for = 0.25 and = 0.1.
172
rag replacements 5000 102
H A
H
A
4000 10
1
3000 10
0
I
T acrit
kcrit
C B
I
B0 A0
2000 10-1 L
B
1000 10-2
C L
a) 0 c) 10-3
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
173
4 50
H L
I A
0
3 B A 0
C
I L -50
B
2 C
B0
mcrit
ccrit
-100 I
1
-150
0 A
H -200
A0
b) -1 d) -250
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 4.5. For = 0.5 and = 0.1: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
rag replacements
-56 0.6
-57
m=2 0.3
-58 L
ccrit
ccrit
0
-59
B0 m=3
-0.3
-60
e)
174
-61 f) -0.6
98 99 100 101 100 101 102 103 104 105
Re Re
Figure 4.5. (contd) For = 0.5 and = 0.1: (e) critical c versus Re near the mode
transition B, (f) critical c versus Re near ccrit = 0.
104
m=0
3
Ta
10
m=3
m=2
eplacements
2
10
10-1 100 101 102
175
104
7
4
103
Ta
replacements
6
2
10
10-2 10-1 100 101 102
Figure 4.6b. Neutral curves for Re = 165, = 0.5, and = 0.1, corresponding
to azimuthal wavenumbers shown to the left of each curve.
176
105 102
A
1
10
104 A
rag replacements
10
0 B
T acrit
0
C B
kcrit
103
10-1 A0
C
102 B
10-2
a) 101 c) 10 -3
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
177
4 200
B0 A
0
B C B
3 C A0
-200
mcrit
ccrit
2 B0 -400
-600
1 A
-800
A0
b) 0 d) -1000
100 101 102 103 104 105 100 101 102 103 104 105
Re Re
Figure 4.7. For = 1 and = 0.1: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
4
2
ccrit
A
L
0
replacements -2
0 1 2 3 4 5
10 10 10 10 10 10
Re
Figure 4.7e. Critical wavespeed for = 1 and = 0.1.
178
rag replacements
106 1.5
B
105 D 1
H
B0
C
T acrit
kcrit
104 0.5
A0
D0
A E D
3
10 0
C
L
B
a) 102
c) -0.5
101 10 2
10 3
10 4
10 5
101 102 103 104 105
Re Re
179
200
B0 L
-2
L
A 0
C B
H D
-4 H
mcrit
0
A
ccrit
-200
H
0
D
-6
-400
b) -8 d) -600
101 102 103 104 105 101 102 103 104 105
Re Re
Figure 4.8. For = 0.2 and = 0.1: (a) critical T a, (b) critical m, (c) critical k, (d) critical c versus Re.
0.2
0.1
ccrit
-0.1
-0.2
1 2 3 4 5
replacements 10 10 10 10 10
Re
Figure 4.8e. Critical wavespeed for = 0.2 and = 0.1.
180
10000
8000
6000
T
4000
181
2
3
2000
B
PSfrag replacements 1 A
0
0 50 100 150 200 250 300
Ta
Figure 5.1. Dimensionless torque per unit length, T , versus T a for = 0 and = 2.
1
r
182
PSfrag replacements
0.5
0 1
z
r
183
PSfrag replacements
0.5
0 z 1
5000 3c
4000 2c
5c
4c
3000
T
C0
184
2000
PSfrag replacements
B0
1000 1c
A0
0
0 50 100 150 200 250 300
Ta
Figure 5.4. Dimensionless torque per unit length, T , versus T a for = 0.01 and = 4.
1
r
185
PSfrag replacements
0.5
0 z 2
r
186
PSfrag replacements
0.5
0 2
z
r
187
PSfrag replacements
0.5
0 2
z
Figure 5.7. contours on the oppositely rotating counterpart of branch 2c for = 0.5,
= 0.01, = 4, and T a = 80.
1
r
188
PSfrag replacements
0.5
0 2
z
r
189
PSfrag replacements
0.5
0 2
z
r
190
PSfrag replacements
0.5
0
z 2
r
191
PSfrag replacements
0.5
0
z 2
r
192
PSfrag replacements
0.5
0
z 2
r
193
PSfrag replacements
0.5
0
z 2
r
194
PSfrag replacements
0.5
0
z 2
3c
3000
4c
5c
2000
T
C0
2c
195
B0
PSfrag replacements
1000 A0
1c
0
0 40 80 120 160
Ta
Figure 5.15. Dimensionless torque per unit length, T , versus T a for = 0.03 and = 4.
1
r
196
PSfrag replacements
0.5
0
z 2
r
197
PSfrag replacements
0.5
0
2
z
r
198
PSfrag replacements
0.5
0
z 2
r
199
PSfrag replacements
0.5
0
z 2
r
200
PSfrag replacements
0.5
0
z 2
r
201
PSfrag replacements
0.5
0
z 2
r
202
PSfrag replacements
0.5
0
z 2
3000 5d
B0
2000
T
203
A0
1000
PSfrag replacements 1c
0
0 50 100 150 200
Ta
Figure 5.23. Dimensionless torque per unit length, T , versus T a for = 0.07 and = 4.
1
r
204
PSfrag replacements
0.5
0 2
z
Figure 5.24. contours on branch 1c for = 0.5, = 0.07, = 4, and T a = 50.
1
r
205
PSfrag replacements
0.5
0
2
z
Figure 5.25. contours on branch 2c for = 0.5, = 0.07, = 4, and T a = 120.
1
r
206
PSfrag replacements
0.5
0
2
z
Figure 5.26. contours on branch 2c for = 0.5, = 0.07, = 4, and T a = 150.
1
r
207
PSfrag replacements
0.5
0
2
z
Figure 5.27. contours on branch 3c for = 0.5, = 0.07, = 4, and T a = 120.
1
r
208
PSfrag replacements
0.5
0
2
z
Figure 5.28. contours on branch 3c for = 0.5, = 0.07, = 4, and T a = 150.
1
r
209
PSfrag replacements
0.5
0
2
z
Figure 5.29. contours on branch 4c for = 0.5, = 0.07, = 4, and T a = 150.
1
r
210
PSfrag replacements
0.5
0
2
z
Figure 5.30. contours on branch 5cl for = 0.5, = 0.07, = 4, and T a = 150.
1
r
211
PSfrag replacements
0.5
0
2
z
Figure 5.31. contours on branch 5ch for = 0.5, = 0.07, = 4, and T a = 150.
4000
3000
2000
T
212
1000
0
0 40 80 120 160
Ta
PSfrag replacements
Figure 5.32. Dimensionless torque per unit length, T , versus T a for = 0.1 and = 4.
1
r
213
PSfrag replacements
0.5
0
2
z
r
214
PSfrag replacements
0.5
0
2
z
r
215
PSfrag replacements
0.5
0
2
z
Ro 1
Periodic
Periodic
R1 1 (1 )
PSfrag replacements
Ri
216
0 0
0 (1 )/2 (1 + )/2 1
0 (L l)/2 (L + l)/2 L
1E-14
1E-15
1E-16
a) PSfrag replacements b)
replacements a)
Figure 6.2. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.2. a) T a = 0,
b) T a = 25. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
1E-13
1E-14
218
1E-15
1E-16
c) PSfrag replacements d)
replacements c)
Figure 6.2. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.2. c) T a = 50,
d) T a = 100. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
1E-13
1E-14
219
1E-15
1E-16
e) PSfrag replacements f)
replacements e)
Figure 6.2. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.2. e) T a = 150,
f) T a = 200. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
-1E-05
-0.0001
-0.001
-0.01
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1
-1.2
-1.4
-1.6
220
-1.8
-2
a) PSfrag replacements b)
replacements a)
-1
10
0.05
Re
10-2
0.025
-3
10
a) 0 c) 10 -4
3000
221
2500
2000
T
1500
1000
500
b) 0
0 50 100 150 200 250
Ta
Figure 6.4. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial mixing ()
for = 2, = 0.5, = 0.8, and = 0.2: (a) Re versus T a, (b) T versus T a, (c) versus T a.
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
222
1E-13
1E-14
1E-15
1E-16
a) PSfrag replacements b)
replacements a)
Figure 6.5. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.5. a) T a = 0,
b) T a = 25. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
223
1E-13
1E-14
1E-15
1E-16
c) PSfrag replacements d)
replacements c)
Figure 6.5. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.5. c) T a = 50,
d) T a = 100. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
224
1E-13
1E-14
1E-15
1E-16
e) PSfrag replacements f)
replacements e)
Figure 6.5. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.5. e) T a = 150,
f) T a = 200. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
-0.0001
-0.001
-0.01
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1
-1.2
-1.4
-1.6
-1.8
225
-2
a) PSfrag replacements b)
replacements a)
0
10
0.4
Re
10-1
0.2
-2
10
a) 0 c) 10 -3
3000
226
2500
2000
T
1500
1000
500
b) 0
0 50 100 150 200 250
Ta
Figure 6.7. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial mixing ()
for = 2, = 0.5, = 0.8, and = 0.5: (a) Re versus T a, (b) T versus T a, (c) versus T a.
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
227
1E-11
1E-12
1E-13
a) PSfrag replacements b)
replacements a)
Figure 6.8. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.8. a) T a = 0,
b) T a = 25. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
228
1E-11
1E-12
1E-13
c) PSfrag replacements d)
replacements c)
Figure 6.8. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.8. c) T a = 50,
d) T a = 100. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
229
1E-11
1E-12
1E-13
e) PSfrag replacements f)
replacements e)
Figure 6.8. and vr2 contours for = 2, = 0.5, = 0.8, and = 0.8. e) T a = 150,
f) T a = 200. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
-0.0001
-0.001
-0.01
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1
-1.2
-1.4
-1.6
-1.8
230
-2
a) PSfrag replacements b)
replacements a)
1
10
1
Re
100
0.5
-1
10
a) 0 c) 10 -2
2500
231
2000
1500
T
1000
500
b) 0
0 50 100 150 200 250
Ta
Figure 6.10. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial mixing ()
for = 2, = 0.5, = 0.8, and = 0.8: (a) Re versus T a, (b) T versus T a, (c) versus T a.
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
1E-13
1E-14
232
1E-15
1E-16
1E-17
1E-18
a) PSfrag replacements b)
replacements a)
Figure 6.11. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.2. a) T a = 0,
b) T a = 25. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
1E-13
1E-14
233
1E-15
1E-16
1E-17
1E-18
c) PSfrag replacements d)
replacements c)
Figure 6.11. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.2. c) T a = 50,
d) T a = 100. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
1E-13
1E-14
234
1E-15
1E-16
1E-17
1E-18
e) PSfrag replacements f)
replacements e)
Figure 6.11. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.2. e) T a = 150,
f) T a = 200. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
-1E-06
-1E-05
-0.0001
-0.001
-0.01
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1
-1.2
-1.4
235
-1.6
-1.8
-2
a) PSfrag replacements b)
replacements a)
-1
10
0.04
Re
10-2
0.02
-3
10
a) 0 c) 10 -4
7000
236
6000
5000
4000
T
3000
2000
1000
b) 0
0 50 100 150 200 250
Ta
Figure 6.13. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial mixing ()
for = 2, = 0.5, = 0.5, and = 0.2: (a) Re versus T a, (b) T versus T a, (c) versus T a.
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
237
1E-11
1E-12
a) PSfrag replacements b)
replacements a)
Figure 6.14. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.5. a) T a = 0,
b) T a = 10. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
238
1E-11
1E-12
c) PSfrag replacements d)
replacements c)
Figure 6.14. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.5. c) T a = 15,
d) T a = 25. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
239
1E-11
1E-12
e) PSfrag replacements f)
replacements e)
Figure 6.14. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.5. e) T a = 50,
f) T a = 100. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
240
1E-11
1E-12
g) PSfrag replacements h)
replacements g)
Figure 6.14. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.5. g) T a = 150,
h) T a = 200. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
-0.0001
-0.001
-0.01
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1
-1.2
-1.4
-1.6
-1.8
241
-2
a) PSfrag replacements b)
replacements a)
101
0.5 100
Re
10-1
0.25 10-2
10-3
a) 0 c) 10 -4
7000
242
6000
5000
4000
T
3000
2000
1000
b) 0
0 50 100 150 200 250
Ta
Figure 6.16. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial mixing ()
for = 2, = 0.5, = 0.5, and = 0.5: (a) Re versus T a, (b) T versus T a, (c) versus T a.
0.03
0.01
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
243
1E-10
3E-11
1E-11
1E-12
a) PSfrag replacements b)
replacements a)
Figure 6.17. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.8. a) T a = 0,
b) T a = 4. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.03
0.01
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
244
1E-10
3E-11
1E-11
1E-12
c) PSfrag replacements d)
replacements c)
Figure 6.17. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.8. c) T a = 6,
d) T a = 25. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.03
0.01
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
245
1E-10
3E-11
1E-11
1E-12
e) PSfrag replacements f)
replacements e)
Figure 6.17. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.8. e) T a = 50,
f) T a = 100. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.03
0.01
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
246
1E-10
3E-11
1E-11
1E-12
g) PSfrag replacements h)
replacements g)
Figure 6.17. and vr2 contours for = 2, = 0.5, = 0.5, and = 0.8. g) T a = 150,
h) T a = 200. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
-0.0001
-0.001
-0.01
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1
-1.2
-1.4
-1.6
-1.8
247
-2
a) PSfrag replacements b)
replacements a)
102
1.5
101
0
10
Re
-1
10
-2
10
0.5
-3
10
a) 0 c) 10 -4
6000
248
5000
4000
T
3000
2000
1000
b) 0
0 50 100 150 200 250
Ta
Figure 6.19. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial mixing ()
for = 2, = 0.5, = 0.5, and = 0.8: (a) Re versus T a, (b) T versus T a, (c) versus T a.
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
249
1E-13
1E-14
1E-15
a) PSfrag replacements b)
replacements a)
Figure 6.20. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.2. a) T a = 0,
b) T a = 25. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
250
1E-13
1E-14
1E-15
c) PSfrag replacements d)
replacements c)
Figure 6.20. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.2. c) T a = 50,
d) T a = 100. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
1E-11
1E-12
251
1E-13
1E-14
1E-15
e) PSfrag replacements f)
replacements e)
Figure 6.20. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.2. e) T a = 150,
f) T a = 200. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
-1E-08
-1E-07
-1E-06
-1E-05
-0.0001
-0.001
-0.01
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1
252
-1.2
-1.4
-1.6
-1.8
a) -2
PSfrag replacements b)
replacements a)
0.04
-1
10
0.03
Re
10-2
0.02
-3
10
0.01
a) 0 c) 10 -4
25000
253
20000
15000
T
10000
5000
b) 0
0 50 100 150 200 250
Ta
Figure 6.22. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial mixing ()
for = 2, = 0.5, = 0.2, and = 0.2: (a) Re versus T a, (b) T versus T a, (c) versus T a.
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
254
1E-11
1E-12
1E-13
1E-14
a) 1E-15
PSfrag replacements b)
replacements a)
Figure 6.23. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.5. a) T a = 0,
b) T a = 25. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
255
1E-11
1E-12
1E-13
1E-14
c) 1E-15
PSfrag replacements d)
replacements c)
Figure 6.23. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.5. c) T a = 50,
d) T a = 100. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
1E-10
3E-11
256
1E-11
1E-12
1E-13
1E-14
e) 1E-15
PSfrag replacements f)
replacements e)
Figure 6.23. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.5. e) T a = 150,
f) T a = 200. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
-1E-05
-0.0001
-0.001
-0.01
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1
-1.2
-1.4
-1.6
257
-1.8
-2
a) PSfrag replacements b)
replacements a)
1
10
0.3
0
10
Re
0.2
10-1
0.1
10-2
a) 0 c) 10 -3
20000
258
15000
T
10000
5000
b) 0
0 50 100 150 200 250
Ta
Figure 6.25. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial mixing ()
for = 2, = 0.5, = 0.2, and = 0.5: (a) Re versus T a, (b) T versus T a, (c) versus T a.
0.03
0.01
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
259
1E-10
3E-11
1E-11
1E-12
a) 1E-13
PSfrag replacements b)
1E-14
replacements a)
Figure 6.26. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.8. a) T a = 0,
b) T a = 25. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.03
0.01
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
260
1E-10
3E-11
1E-11
1E-12
c) 1E-13
PSfrag replacements d)
1E-14
replacements c)
Figure 6.26. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.8. c) T a = 50,
d) T a = 100. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
0.03
0.01
0.003
0.001
0.0003
0.0001
3E-05
1E-05
3E-06
1E-06
3E-07
1E-07
3E-08
1E-08
3E-09
1E-09
3E-10
261
1E-10
3E-11
1E-11
1E-12
e) 1E-13
PSfrag replacements f)
1E-14
replacements e)
Figure 6.26. and vr2 contours for = 2, = 0.5, = 0.2, and = 0.8. e) T a = 150,
f) T a = 200. Positive and negative streamfunction values correspond to solid and dashed lines, respectively.
-0.0001
-0.001
-0.01
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
-1
-1.2
-1.4
-1.6
-1.8
262
-2
a) PSfrag replacements b)
replacements a)
2
10
0.6
1
10
Re
0.4
100
0.2
10-1
a) 0 c) 10 -2
20000
263
15000
T
10000
5000
b) 0
0 50 100 150 200 250
Ta
Figure 6.28. Dimensionless mean flow rate (Re), torque per unit length (T ), and radial mixing ()
for = 2, = 0.5, = 0.2, and = 0.8: (a) Re versus T a, (b) T versus T a, (c) versus T a.
Bibliography
Ali, M. & Weidman, P. D. 1990 On the stability of circular Couette flow with radial
heating. J. Fluid Mech. 220, 53.
Armaly, B. F., Durst, Pereira, J. C. F. & Schonung, B. 1983 Experimental and
theoretical investigation of backward-facing step flow. J. Fluid Mech. 127, 473.
Babcock, K. L., Ahlers, G. & Cannell, D. S. 1994 Noise amplification in open Taylor-
Couette flow. Phys. Rev. E 50, 3670.
Becker, K. M. & Kaye, J. 1962 Measurements of diabatic flow in an annulus with an
inner rotating cylinder. J. Heat Trans. 84, 97.
Benjamin, T. B. & Mullin, T. 1982 Notes on the multiplicity of flows in the Taylor
experiment. J. Fluid Mech. 121, 219.
Bertsch, A., Heimgartner, S., Cousseau, P. & Renaud, P. 2001 Static micromixers
264
Bohme, G. & Broszeit, J. 1997 Numerical flow simulation for Bingham plastics in a
single-screw extruder. Theoret. Comput. Fluid Dyn. 9, 65.
Booy, M. L. 1981 The influence of non-Newtonian flow on effective viscosity and
channel efficiency in screw pumps. Polymer Eng. Sci. 21, 93.
Broszeit, J. 1997 Finite-element simulation of circulating steady flow for fluids of the
memory-integral type: flow in a single-screw extruder. J. Non-Newtonian Fluid
Mech. 70, 35.
Bruker, I., Miaw, C., Hasson, A. & Balch, G. 1987 Numerical Analysis of the
Press.
Chen, Y. M. & Pearlstein, A. J. 1989 Stability of free-convection flows of variable-
viscosity fluids in vertical and inclined slots. J. Fluid Mech. 198, 513.
265
Chen, J. H., Pritchard, W. G. & Tavener, S. J. 1995 Bifurcation of flow past a cylinder
between parallel planes. J. Fluid Mech. 284, 23.
Chen, M. M. & Whitehead, J. A. 1968 Evalution of two-dimensional periodic Rayleigh
convection cells of arbitrary wave numbers. J. Fluid Mech. 31, 1.
Choi, H. C., Song, J. H. & Yoo, J. Y. 1988 Numerical simulation of the planar
contraction flow of a Giesekus fluid. J. Non-Newtonian Fluid Mech. 29, 347.
Choo, K. P., Hami, M. L. & Pittman, J. F. T. 1981 Deep channel operating
characteristics of a single screw extruder: finite element predictions and
experimental results for isothermal non-Newtonian flow. Polymer Eng. Sci. 21, 100.
Choo, K. P., Neelakantan, N. R. & Pittman, J. F. T. 1980 Experimental deep-channel
velocity profiles and operating characteristics for a single-screw extruder. Polymer
Eng. Sci. 20, 349.
Chossat, P. & Iooss, G. 1994 The Couette-Taylor Problem. Springer-Verlag, New York.
Chung, K. C. & Astill, K. N. 1977 Hydrodynamic instability of viscous flow between
coaxial cylinders with fully developed axial flow. J. Fluid Mech. 81, 641.
Coeuret, F. & Legrand, J. 1981 Mass transfer at the electrodes of concentric
cylindrical reactors combining axial flow and rotation of the inner cylinder.
Electrochim. Acta 26, 865.
Cohen, S. & Marom, D. M. 1983 Experimental and theoretical study of a rotating
annular flow reactor. Chem. Eng. J. 27, 87.
Cornish, R. J. 1933 Flow of water through fine clearances with relative motion of the
boundaries. Proc. Roy. Soc. London A 140, 227.
Cotrell, D. L. & Pearlstein, A. J. 2003 The connection between centrifugal instability
and Tollmien-Schlichting-like instability for spiral Poiseuille flow. J. Fluid Mech.,
266
Cotrell, D. L., Rani, S. L. & Pearlstein, A. J. 2003 Computational assessment of
subcritical instability and apparent transition delay in spiral Poiseuille flow
experiments. J. Fluid Mech., submitted for publication.
Debler, W., Funer, E. & Schaaf, B. 1969 Torque and flow patterns in supercritical
circular Couette flow, M. Hetenyi and W. G. Vincenti, eds., Proc. 12th International
Congress of Applied Mechanics, p. 116, Springer-Verlag, Berlin.
Debler, W., Funer, E. & Schaaf, B. 1969 Torque and flow patterns in supercritical
circular Couette flow, M. Hetenyi and W. G. Vincenti, eds., Proc. 12th Int. Congress
stability of Couette flow and Taylor vortex flow. Phys. Fluids 27, 2403.
DiPrima, R. C. & Pridor, A. 1979 The stability of viscous flow between rotating
concentric cylinders with an axial flow. Pro. Roy. Soc. London A 366, 555.
DiPrima, R. C. & Swinney, H. L. 1985 Instabilities and transition in flow between
cylinders. II. Visual observations. Proc. Roy. Soc. London A 258, 101.
Donnelly, R. J. & Simon, N. J. 1960 An empirical torque relation for supercritical flow
between rotating cylinders. J. Fluid Mech. 7, 401.
Drazin, P. G. & Reid, W. H. 1981 Hydrodynamic Stability. Cambridge University
Press.
Eagles, P. M. & Eames, K. 1983 Taylor vortices between almost cylindrical boundaries.
J. Eng. Math. 17, 263.
267
Eisenberg, M., Tobias, C. W. & Wilke, C. R. 1954 Ionic mass transfer and
concentration polarization at rotating electrodes. J. Electrochem. Soc. 101, 306.
Eklund, A. & Simonsson, D. 1988 Enhanced mass transfer to rotating cylinder
electrode with axial flow. J. Appl. Electrochem. 18, 710.
Gage K. S. & Reid, W. H. 1968 The stability of thermally stratified plane Poiseuille
flow. J. Fluid Mech. 33, 21.
Gao, H., Scheeline, A. & Pearlstein, A. J. 2002 Spatially-controlled microstructural
variation using a free-surface flow driven by a rotating cylinder electrode: Growth of
268
Giordano, R. C., Giordano, R. L. C., Prazeres, D. M. F. & Cooney, C. L. 1998
Analysis of a Taylor-Poiseuille vortex flow reactorI: Flow patterns and mass
transfer characteristics. Chem. Eng. Sci. 53, 3635.
Giordano, R. L. C., Giordano, R. C., Prazeres, D. M. F. & Cooney, C. L. 2000
Griffith, R. M. 1962 Fully developed flow in screw extruders. Ind. Eng. Chem. Fund.
1, 180.
Grosvenor, R. I. 1981 Axisymmetry of spiral Taylor vortex flow in annuli and the
effect of gap width variations. M.Eng. Thesis, Univ. Wales Inst. Sci. Tech., Cardiff.
Gu, Z.-H. & Fahidy, T. Z. 1982 Electrolytic mass transport at a rotating outer
cylinder electrode with developing axial flow in the annulus. J. Appl. Electrochem.
12, 659.
Haim, D. & Pismen, L. M. 1994 Performance of a photochemical reactor in the regime
269
Hasoon, M. A. & Martin, B. W. 1977 The stability of viscous axial flow in an annulus
with a rotating inner cylinder. Proc. Roy. Soc. London A 352, 351.
Howes, T. & Rudman, M. 1998 Flow and axial dispersion simulation for traveling
axisymmetric Taylor vortices. AIChE J. 44, 255.
Hu, H.-C. & Kelly, R. E. 1995 Effect of a time-periodic axial shear flow upon the onset
of Taylor vortices. Phys. Rev. E 51, 3242.
Ikeda, E. & Maxworthy, T. 1994 Spatially forced corotating Taylor-Couette flow. Phys.
Rev. E 49, 5218.
Iosilevskii, G., Brenner, H., Moore, C. M. V. & Cooney, C. L. 1993 Mass transport
and chemical reaction in Taylor-vortex flows with entrained catalytic particles:
application to a novel class of immobilized enzyme biochemical reactors. Phil.
Trans. Roy. Soc. Lond. A 345, 259.
545.
Joseph, D. D & Sturges, L. 1978 The convergence of biorthogonal series for biharmonic
and Stokes flow edge problems: Part II. SIAM J. Applied Math 34, 7.
Kataoka, K., Doi, H., Hongo, T. & Futagawa, M. 1975 Ideal plug-flow properties of
270
Kelly, R. E. & Pal, D. 1978 Thermal convection with spatially periodic boundary
conditions: resonant wavelength excitation. J. Fluid Mech. 86, 433.
Kirchgassner, K. & Sorger, P. 1969 Branching analysis for the Taylor problem. Quart.
J. Mech. and Applied Math 22, 183.
inner electrode of a concentric cylindrical reactor with axial flow. Electrochim. Acta
25, 669.
Lopez, A. R., Romero, L. A. & Pearlstein, A. J. 1990 Effect of rigid boundaries on the
onset of convective instability in a triply diffusive fluid layer. Phys. Fluids A 2, 897.
Mahadevan, R. & Lilley, G. M. 1977 The stability of axial flow between concentric
cylinders to asymmetric disturbances. AGARD CP-224, 9-1.
271
Marques, F. & Lopez, J. M. 2000 Spatial and temporal resonances in a periodically
forced hydrodynamic system. Physica D 136, 340.
Martin, N. & Galey, C. 1994 Static mixer for oxidation and disinfection by ozone.
Ozone Sci. Eng. 16, 455.
Mavec, J. A. 1973 Spiral and toroidal secondary motions in swirling flows through an
annulus at low Reynolds numbers. M.S. Thesis, Illinois Inst. Tech., Chicago.
McGrew, J. M. & McHugh, J. D. 1965 Analysis and test of the screw seal in laminar
and turbulent operation. ASME J. Basic Eng. 87, 153.
Meseguer, A. & Marques, F. 2002 On the competition between centrifugal and shear
instability in spiral Poiseuille flow. J. Fluid Mech. 455, 129.
Min, K. & Lueptow, R. M. 1994 Circular Couette flow with pressure-driven axial flow
and a porous inner cylinder. Exp. Fluids 17, 190.
Moser, K. W., Raguin, L. G., Harris, A., Morris, H. D., Georgiadis, J., Shannon, M. &
Philpott, M. 2000 Visualization of Taylor-Couette and spiral Poiseuille flows using
a snapshot FLASH spatial tagging sequence. Magn. Reson. Imaging 18, 199.
272
Mott, J. E. & Joseph, D. D. 1968 Stability of parallel flow between concentric
cylinders. Phys. Fluids 11, 2065.
Mullin, T. 1982 Mutations of steady cellular flows in the Taylor experiment. J. Fluid
Mech. 121, 207.
13, 209.
Ng, B. S. & Turner, E. R. 1982 On the linear stability of spiral flow between rotating
cylinders. Proc. Roy. Soc. London A 382, 83.
Ning, L., Ahlers, G. & Cannell, D. S. 1990 Wave-number selection and traveling vortex
waves in spatially ramped Taylor-Couette flow. Phys. Rev. Letters 64, 1235.
Orszag, S. A. 1971 Accurate solution of the Orr-Sommerfeld stability equation. J.
Fluid Mech. 50, 689.
Painter, B. D. & Behringer, R. P. 1996 Localization of rolls near onset in a Taylor-
Couette cell with spatial perturbations. Bull. Am. Phys. Soc. 41, 1791.
Painter, B. D. & Behringer, R. P. 1997 Effects of spatial inhomogeneities on the
transion to Taylor vortex flow in the Taylor-Couette system. Bull. Am. Phys. Soc.
42, 2139.
2031.
273
Painter, B. D., Olafsen, J. S. & Behringer, R. P. 1995 Effects of spatial perturbations
on pattern selection at the onset of Couette-Taylor flow. Bull. Am. Phys. Soc. 40,
2018.
Pal, D. & Kelly, R. E. 1978 Thermal convection with spatially periodic nonuniform
heating: nonresonant wavelength excitation. Proc. of the 6th Int. Heat Transfer
Conf. 2, 235.
Pearlstein, A. J. 1987 The onset of stability via three-dimensional disturbances in
parallel shear flows. Bull. Am. Phys. Soc. 32, 2087.
Raspo, I. & Crespo del Arco, E. 2003 Instability in a rotating channel-cavity system
with an axial through-flow. Int. J. Heat and Fluid Flow 24, 41.
Recktenwald, A., Lucke, M. & Muller, H. W. 1993 Taylor vortex formation in axial
through-flow: Linear and weakly nonlinear analysis. Phys. Rev. E 48, 4444.
Reddy, J. N. & Gartling, D. K. 1994 The Finite Element Method in Heat Transfer and
Fluid Dynamics. CRC Press, Boca Raton.
Reid, W. H. 1961 Review of "The hydrodynamic stability of viscid flow between
coaxial cylinders" by S. Chandrasekhar (Proc. Nat. Acad. Sci. U.S.A. 46, 141143,
274
Resende, M. M., Tardioli, P. W., Fernandez, V. M., Ferreira, A. L. O., Giordano, R.
L. C. & Giordano, R. C. 2001 Distribution of suspended particles in a Taylor-
Poiseuille vortex flow reactor. Chem. Eng. Sci. 56, 755.
Rhode, D. L., Demko, J. A., Traegner, U. K., Morrison, G. L. & Sobolik, S. R. 1986
Prediction of incompressible flow in labyrinth seals. ASME J. Fluids Eng. 108, 19.
Riecke, H. 1988 Imperfect wave-number selection by ramps in a model for Taylor
vortex flow. Phys. Rev. A 37, 636.
Ross, C. A. 1994 Electrodeposited multilayer thin-films. Ann. Rev. Mat. Sci. 24, 159.
Rotz, C. A. & Suh, N. P. 1976 New techniques for mixing viscous reacting liquids.
Part I. Mechanical means to improved laminar mixing. Polym. Eng. Sci. 16, 664.
Sadeghi, V. M. & Higgins, B. G. 1991 Stability of sliding Couette-Poiseuille flow in
an annulus subject to axisymmetric and asymmetric disturbances. Phys. Fluids 3,
2092.
Salwen, H., Cotton, F. W. & Grosch, C. E. 1980 Linear stability of Poiseuille flow in a
circular pipe. J. Fluid Mech. 98, 273.
Schwartz, M., Li, W., Phan, N. H., Nobe, K. & Pearlstein, A. J. 1992 Plating of
copper in tubular holes with rotating screw electrodes. Surf. Coat. Technol. 52,
269.
Sczechowski, J. G., Koval, C. A. & Noble, R. D. 1995 A Taylor vortex reactor for
heterogeneous photocatalysis. Chem. Eng. Sci. 50, 3163.
Shapiro, I., Shtilman, L. & Tumin, A. 1999 On stability of flow in an annular channel.
Phys. Fluids 11, 2984.
Silveston, P. L. 1958 Warmedurchgang in waagerechten Flussgkeitsschichten. I. Forsch.
Ing. Wes. 24, 29, 59.
275
Smits, A., Auvity, B. & Sinha, M. 2000 Taylor-Couette flow with a shaped inner
cylinder. Bull. Am. Phys. Soc. 45(9), 37.
Snyder, H. A. 1962 Experiments on the stability of spiral flow at low axial Reynolds
numbers. Proc. Roy. Soc. London A 265, 198.
Snyder, H. A. 1965 Experiments on the stability of two types of spiral flow. Ann.
Physics 31, 292.
Snyder, H. A. 1970 Waveforms in rotating Couette flow. Int. J. Non-Linear Mech. 5,
659.
transfer, in a concentric annulus having an inner rotating wall. Ph.D. Thesis, Leeds
University.
Sorour, M. M. & Coney, J. E. R. 1979 An experimental investigation of the stability of
spiral vortex flow. J. Mech. Eng. Sci. 21, 397.
Couette flow with low axial Reynolds number. Can. J. Chem. Eng. 54, 295.
Strumolo, G. S. 1983 Perturbed bifurcation theory for Poiseuille annular flow. J. Fluid
Mech. 130, 59.
Sullivan, T. & Middleman, S. 1985 Factors that affect uniformity of through-holes in
276
Swift, J. B., Babcock, K. L. & Hohenberg, P. C. 1994 Effects of thermal noise in
Taylor-Couette flow with corotation and axial through-flow. Physica A 204, 625.
Swinney, H. L., Kreisberg, N., McCormick, W. D., Noszticzius, Z. & Skinner, G. 1990
Spatiotemporal patterns in reaction-diffusion systems, D. K. Campbell, ed., Chaos:
intrinsic noise in the Couette-Taylor system with an axial flow. Phys. Rev. E 49,
1309.
Tsameret, A. & Steinberg, V. 1991a Convective vs. absolute instability in Couette-
Taylor flow with an axial flow. Europhys. Lett. 14, 331.
49, 1291.
Tsameret, A. & Steinberg, V. 1994b Competing states in a Couette-Taylor system
with an axial flow. Phys. Rev. E 49, 4077.
277
Tung, T. T. & Laurence, R. L. 1975 A coordinate frame for helical flows. Polymer
Eng. Sci. 15, 401.
Vukasinovic, J. 2001 Flow in the gap between a rotating screw and a co-axial
stationary outer cylinder. M.S. Thesis, Georgia Inst. Tech., Georgia.
278
Yu, Q. & Hu, G.-H. 1997 Development of a helical coordinate system and its
application to analysis of polymer flow in screw extruders. Part I. The balance
equations in a helical coordinate system. J. Non-Newtonian Fluid Mech. 69, 155.
Yucel, U. & Kazakia, J. Y. 2001 Analytical prediction techniques for axisymmetric
flow in gas labyrinth seals. J. Eng. for Gas Tur. and Pow. 123, 255.
Zimmermann, W., Painter, B. D. & Behringer, R. P. 1998a Effects of spatial disorder
on the transition to Taylor vortex flow. Europhys. Lett. 44, 599.
Zimmermann, W., Painter, B. & Behringer, R. 1998b Pattern formation in an
279
Vita
David L. Cotrell was born in Alton, Illinois on October 22, 1968 to Patricia G. and
Norman F. Cotrell. David dropped out of high school his senior year to work as a pipefitter.
After about a year, he returned home to get his GED. He then enrolled at Illinois College
in Jacksonville, Illinois in a 3-2 Physics/Engineering program with the University of Illinois
Bachelors of Science degrees in Mechanical Engineering and Physics from the University
of Illinois and Illinois College, respectively.
Upon completion of his undergraduate studies, he enrolled in a masters program at
the University of Illinois at Urbana-Champaign in the Department of Mechanical and
Industrial Engineering working with Professor Arne J. Pearlstein. While working on his
Masters of Science in Mechanical Engineering, he was supported by a NASA GSRP.
During his time as a graduate student, he taught undergraduate laboratory sections in
Design for Manufacturability, Engineering Materials, and Introductory Gas Dynamics.
Also while a graduate student, he had the privilege of being an instructor for Introduction
to Fluid Mechanics and Thermodynamics.
During his time at Illinois College David met the love of his life, Kelli J. Killen, who
he has been with ever since.
280