PHYSICAL FLUID DYNAMICS

THE MODERN UNIVERSITY PHYSICS SERIES

This series is intended for readers whose main interest is in physics, or who need the
methods of physics in the study of science and technology. Some of the books will
provide a sound treatment of topics essential in any physics training, while other,
more advanced, volumes will be suitable as preliminary reading for research in the
field covered. New titles will be added from time to time.

Bagguley: Electromagnetism and Linear Circuits

Clark: A First Course in Quantum Mechanics
littlefield and Thorley: Atomic and Nuclear Physics (2nd Edn.)
Lothian: Optics and Its Uses
Lovell, Avery and Vernon: Physical Properties of Materials
Perina: Coherence of Light
Tilley and Tilley: Superjluidity and Superconductivity
Tritton: Physical Fluid Dynamics
Wolbarst: Symmetry and Quantum Systems
Physical Fluid
Dynamics

D. J. TRITTON

School of Physics
University of Newcastle-upon- Tyne

lriii5I VAN NOSTRAND REINHOLD COMPANY

~ New York - Cincinnati - Toronto - London - Melbourne
©D. J. Tritton, 1977

ISBN-13: 978-0-442-30132-3 e-ISBN-13: 978-94-009-9992-3

DOI: 10.1007/978-94-009-9992-3

Ubrary of Congress Catalog Card No. 7641890

All rights reserved. No part of this work covered by

the copyright hereon may be reproduced or used
in any form or by any means - graphic, electronic,
or mechanieal, ineluding photocopying, recording,
taping, or information storage of retrieval systems
- without written permission of the publishers.

Published by Van Nostrand Reinhold Company Ltd.,

Molly Millars Lane, Wokingham, Berkshire, England

Published in 1977 by Van Nostrand Reinhold Company

ADivision of Litton Educational Publishing, Inc.,
450 West 33rd Street, New York, N. Y. 10001, U.S.A.

Van Nostrand Reinhold Limited

1410 Birchmount Road, Scarborough, Ontario, MIP 2E7,
Canada

Van Nostrand Reinhold AustraUa Pty. Limifed

17 Queen Street, Mitcham, Victoria 3132, AustraUa

Library of Congress Cataloguing in Publication Data

Tritton, D. J.
Physical fluid dynamics.

(The Modem university physics series)

Bibliography: p. 346
Ineludes index.
1. Fluid dynamics. I. Tide.
QC151.T74 532 76-41890

Typeset in Great Britain by Preface Ltd., Salisbury

Preface

To classify a book as 'experimental' rather than 'theoretical' or as 'pure' rather than

'applied' is liable to imply umeal distinctions. Nevertheless, some Classification is
necessary to teIl the potential reader whether the book is for him. In this spirit, this
book may be said to treat fluid dynamies as a branch of physics, rather than as a
branch of applied mathematics or of engineering. I have often heard expressions
of the need for such a book, and certainly I have feIt it in my own teaching.
I have written it primariIy for students of physics and of physics-based applied
science, aIthough I hope others may find it useful. The book differs from existing
'fundamental' books in placing much greater emphasis on what we know through
laboratory experiments and their physical interpretation and less on the mathe-
matieal formalism. It differs from existing 'applied' books in that the choice of
topics has been made for the insight they give into the behaviour of fluids in
motion rather than for their practical importance. There are differences also from
many existing books on fluid dynamics in the branches treated, reflecting to some
extent shifts of interest in reeent years. In particular, geophysical and astrophysical
applications have prompted important fundamental developments in topics such as
conveetion, stratified flow, and the dynamics of rotating fluids. These developments
have hitherto been reflected in the contents of textbooks only to a limited extent.
Much of the book is based on lectures I have given to final year physics students
at Newcastle-upon-Tyne University, though I have substantially expanded parts of
the material. I have also been influenced by teaching fluid dynamics both at a more
elementary level and to postgraduate students of geophysics and planetary physics.
I have tried to learn which approaches to various topics students find most informa-
tive, and I hope this experience has led to improvements in what I have written.
I have had the final year physics students particularly in mind when deciding
what background knowledge to assume of the reader. The mathematical methods
used should all be famiIiar to such students (although substantial parts of the book
would be intelligible to a reader at an earlier mathematical stage). Also, it is not
really paradoxical that a book aimed at physics students should contain less
explanation of basic physics (such as related thermodynamics) than some other
fluid dynamics texts; knowledge of this can reasonably be assumed.
Nevertheless, I hope that the book will be of value to a variety of other type s of
reader and I have tried to extend its usefulness wherever this has been possible
without distorting the primary aim. Workers in widely various type s of applied
science need an understanding of the phenomena of fluid motion. I have in mind
particularly, but not solely, geo- and planetary physicists (meteorologists and
oceanographers as weIl as students ofplanetary interiors), ofwhose needs I
probably have the most immediate appreciation. Where fluid dynamical topics are
vi Preface
taught as part of a geophysics course, I believe that this should be a suitable text-
book. I believe also that postgraduate students and other research workers, faced
with a project involving fluid dynamics, should find this book of value. It is not
uncommon for otherwise interesting geophysical papers to be marred by the
employment of fluid dynamieal concepts in a way that shows serious misunder-
standing of their significance. But these misunderstandings are not surprising in the
absence of convenient sources of information. Such readers should not imagine that
there are easy short-cuts to an understanding of the phenomena of fluid motion,
but I trust that this book may make the long route a little less arduous.
The choice of topics in a book of this sort must be controversial. The size of the
subject necessitates many arbitrary omissions. The reader wishing to discover just
what is included and what is omitted may find it useful to refer not only to the
contents list but also to Section 1.2. The major limitation is the restriction to
incompressible flow. This will, I know, make the book inadequate for so me
courses. Compressible flow is, however, such a large topic that it really requires a
book of its own; just a chapter would perhaps do it more of an injustice than total
omission. Even within the limitation to incompressible flow, I am much aw are of
omissions that some readers will regret. What I hope I have achieved is a text giving
students sufficient knowledge of the basic concepts of fluid dynamics and sufficient
insight into the consequences of these concept s that they will be able to use other
(probably more advanced) sources to obtain information that is omitted here.
The systematic development of the subject occupies Chapters 5 to 22. After the
introduction in Chapter 1, Chapters 2 to 4 treat three particular topics in a
descriptive (though hopefully not too superficial) way. These topics have been
place d ahead of the systematic treatment for two reasons. Primarily they are
intended to give the reader some understanding of the type of phenomena with
which one is concerned; it is a long haul through the basic concepts of ehapters 5
to 8 if one does not know what it is all for. Additionally, I hope that Chapters 2
to 4 may make the book more accessible to students at an earlier stage of their
studies.
Chapter 23 considers experimental methods; it is convenient to discuss these all
together rather than alongside the experimental results. Chapter 24 illustrates
applications of fluid dynamics - geophysical, biophysical, technological, and so
forth. Its nature is explained mo re fully in Section 24.1. In a sense, this chapter is
included for 'entertainment', but some knowledge of its range of applications
should add to the reader's appreciation of the way in which fluid dynamics has
developed.
Much of the illustrative material throughout the book has been drawn from
research papers. It is a feature of fluid dynamics - probably more than of most
branches of physics - that the details of rather simply specified topics are complex
and still imperfectly understood. (Others must have shared my experience of having
difficulty in convincing a postgraduate student that the topic proposed for him had
not been fully elucidated long ago.) It is thus often appropriate to use even for
introductory purposes (e.g. Chapters 2 to 4) topics that are still the subject of
research. This feature of the subject also makes it easier for a book to serve both as
an undergraduate text and as a source of information for research workerso
The figure captions often contain details that students will wish to ignore; these
are intended for the more advanced reader who wants to know the particular
conditions to which the data refer. However, I hope that these details may some-
Preface vii

times convey to students something of the 'flavour' of experimental fluid dynamics.

The book is more fully refereneed than most undergraduate texts. Some of the
references indicate sources of material - of illustrations or of ideas. (Sometimes I
have thought a reference for a speeific point appropriate, because the simplification
justifiable for the newcomer to the subject may lead the mo re experienced reader
to ask, 'Just what does he mean by that?') Other references have been inc1uded for
the reader who is using the book as an information source and wishes to follow up
a topic in detail. No attempt at completeness has been made; that would involve far
too many references. I have tried to give an appropriate point of entry into the
literature of a topic, mo re often a reeent review or significant paper than the
pioneering research work. In general, the role of a reference should be apparent
from the context. In addition to speeific references, the re is a bibliography of
related books (and mms).
Writing the book has been a task of some years. Consequently, some parts are
more up-to-date than others in the choice of illustrative material and references. I
have tried to modify the parts written earlier in the light of any significant new
developments. But there comes a point when one has to say 'that is the final
version'; otherwise the process never converges!
The primary purpose of the book being pedagogical, I have compiled a selection
of problems for the student to work, although the emphasis on experimentally
based information me ans that not all topics lend themselves readily to problem
working. Some of the problems involve bringing together ideas from different
chapters, so they are all collected together at the end of the book. The problems
are very variable in difficulty and are intended more for selection by the teacher
than for use by the student on his own. A few of the problems amount to little
mo re than the substitution of numbers into equations; the quantities have been
chosen to be physically realistic and it is hoped that the student will attempt to
visualize the situation rather than just make the substitution mechanically. At the
opposite end of the spectrum, a few of the problems will be solvable only by
students with knowledge of mathematical methods not used in the text. Some of
the problems are based on examination questions set to physics students at the
University of Newcastle-upon-Tyne; I am grateful to the University for permission
to inc1ude these.
Many people have been most helpful with information and comments - too
many for me to list them individually; my gratitude is nonetheless sincere. I must
name my colleagues who have given particular help. Discussions with Drs. P. A.
Davies, A. Ibbetson, and C. W. Titman have frequently helped me to c1arify my
ideas. The material in the appendix to Chapter 14 was developed in collaboration
with Dr. D. C. Tozer.
Section 6.2 required speeific assistance. Professor S. C. R. Dennis provided
numerical details of the veloeity field. He kindly recomputed these more accurately
and in a more suitable form than in Ref. [95}; these computations were carried out
in the Data Handling Division at CERN, Geneva. The subsequent computations
leading to Figs. 6.3-6.5 were made in the Computing Laboratory of Newcastle-
upon-Tyne University by Mrs. S. Hofmann, for whose interest and skill I am most
grateful.
Acknowledgements to bodies who have given permission for the reproduction of
copyright material are listed separately. I am grateful also to the authors who have
agreed to my using their material and particularly to those who have provided
viii Pre/ace
original prints of their photographs or additional information. Again a blanket
'thank you' must suffiee. In many cases, I have presented material in a slightly
different form from its original publication and have therefore not needed to seek
permission; my debt to the authors is, of course, no less. The originators of all
photographs and sets of data are indicated by the referenees.
I am, of course, particularly indebted to my typists, Mrs. M. Bennison and
Miss M. Hopkinson, and I am equally indebted, for their skilful draughting of many
of the diagrams, to Mrs. D. Hewett and Mrs. E. R. Thompson. I am grateful also for
the serviees of the University Ubrary (and its inter-library loans service) and of the
Photography Department (who made excellent copies of photographs of whieh the
originals were unobtainable) at the University of Neweastle-upon-Tyne.
Finally my thanks go to former and present staff of Van Nostrand Reinhold Co.
Ltd., particularly Mr. D. J. Carpenter, Mrs. T. Mozley, Miss L. J. Ward and Mr.
D. Winsor, for their encouragement and cooperation.
In conclusion, let me say that I should mueh welcome any comments on the
book that readers may care to send.
Acknowledgements

I am grateful to the following bodies for permission to reproduce figures of which

they hold the copyright:

American Institute of Aeronautics and Astronautics (Fig. 22.2)

American Institute of Physics (Figs. 4.9 and 15.1 0)
American Meteorological Society (Fig. 24.1)
American Society of Mechanieal Engineers (Fig. 16.1)
Annual Reviews Inc. (Figs. 16.3 and 16.7)
Artemis Press (Fig. 24.3)
Barth J. A. Verlag (Fig. 19.9)
Boeing Scientific Research Laboratories (Fig. 16.8)
Cambridge University Press (Figs. 14.7, 14.11, 15.9, 16.14, 17.4, 19.2,22.13,
22.20 and 22.21)
Central Electricity Generating Board (Figs. 24.9 and 24.10)
Company of Biologists Ltd. (Fig. 24.18)
Gordon and Breach Ltd. (Figs. 16.4 and 16.6)
Her Majesty's Stationery Offiee, by permission of the Controller (Fig. 20.1)
Her Majesty's Stationery Offiee, by permission of the Controller, courtesy
Hydraulics Research Station, Wallingford (Figs. 24.5 and 24.6)
Indian Association for the Cultivation of Science (Fig. 4.15)
Macmillan Journals Ltd. (Fig. 24.2)
Microf1lms International Marketing Corp. Inc. (Figs. 4.5, 4.8 and 4.11)
National Aeronautics and Space Administration (Fig. 14.5)
Physical Society of Japan (Eigs. 3.4,11.10,19.7 and 19.8)
Royal Meteorological Society (Figs. 15.15 and 15.17)
Royal Society (Figs. 18.4 and 18.5)
Scienti[ic American (Fig. 24.22)
Springer-Verlag (Figs. 2.1O(a) and (b), 3.7 and 14.6)
Tel/us (Fig. 16.9)
University of Chicago Press (Fig. 24.4)
vm Verlag (Figs. 4.16, 14.1 and 14.4)
Vieweg und Sohn (Fig. 4.6)
Those silent waters weave for him
A fluctuant mutable world and dim,
Where wavering masses bulge and gape
Mysterious, and shape to shape
Dies momently through whorl and hollow,
And form and line and solid follow
Solid and line and form to dream
Fantastic down the eternal stream;
An obscure world, a shifting world,
Bulbous, or pulled to thin, or curled,
Or serpentine, or driving arrows,
Or serene slidings, or March narrows.

From "The Fish" by Rupert Brooke.

Contents

PREFACE v

ACKNOWLEDGEMENTS ix

1. INTRODUCTION
1.1 Preamble 1
1.2 Scope of book 2
1.3 Notation and definitions 4

2. PIPE AND CHANNEL FLOW 6

2.1 Introd uction 6
2.2 Laminar flow theory: channel 7
2.3 Laminar flow theory: pipe 9
2.4 The Reynolds number 11
2.5 The entry length 12
2.6 Transition to turbulent flow 13
2.7 Relationship between flow rate and pressure gradient 16
3. FLOW PAST A CIRCULAR CYLINDER 18
3.1 Introduction 18
3.2 The Reynolds number 19
3.3 Flow patte ms 19
3.4 Drag 27
4. CONVECTION IN HORIZONTAL LAYERS 30
4.1 The configuration 30
4.2 Onset of motion 31
4.3 Flow regimes 33

5. EQUATIONS OF MOTION 42
5.1 Introd uction 42
5.2 Fluid partides and continuum mechanics 42
5.3 Eulerian and Langrangian co-ordinates 44
5.4 Continuity equation 45
5.5 The substantive derivative 46
xii Con ten ts
5.6 The Navier-Stokes equation 48
5.7 BOlllldary eonditions 53
5.8 Condition for ineompressibility 56
Appendix: Derivation of viseous term ofdynamieal equation 59

6. FURTHER BASIC IDEAS 61

6.1 Streamlines, streamtubes, partide paths and streaklines 61
6.2 Computations for flow past a eireular eylinder 63
6.3 The stream funetion 67
6.4 Vorticity 67
6.5 Vortici ty eq ua tion 71
6.6 Cireulation 72

7. DYNAMICAL SIMILARITY 74
7.1 Introduction 74
7.2 Condition for dynamieal similarity: Reynolds number 75
7.3 Dependent quantities 77
7.4 Other governing non-dimensional parameters 79
8. LOW AND HIGH REYNOLDS NUMBERS 81
8.1 Physical signifieanee of the Reynolds number 81
8.2 Low Reynolds number 82
8.3 High Reynolds number 83
9. SOME SOLUTIONS OF THE VISCOUS FLOW EQUATIONS 88
9.1 Introduction 88
9.2 Poiseuille flow 88
9.3 Rotating Couette flow 89
9.4 Stokes flow past a sphere 90
9.5 Low Reynolds number flow past a eylinder 92

10. INVISCID FLOW 94

10.1 In trod uetion 94
10.2 Kelvin eireulation theorem 94
10.3 Irrotational motion 94
10.4 Bernoulli's equation 96
10.5 Dragin inviscid flow: d'Alembert's 'paradox' 97
10.6 Applications of Bernoulli's eq uation 98
10.7 Some definitions 100

11. BOUNDARY LAYERS AND RELATED TOPICS 101

11.1 Boundary layer formation 101
11.2 The boundary layer approximation 101
11.3 Zero press ure gradien t sol ution 103
11.4 Boundary layer separation 106
Contents xiii

11.5 Drag on bluffbodies 109

11.6 Streamlining 110
11.7 Wakes 111
11.8 lets 112
11.9 Momentum and energy in viseous flow 116

12. LIFT 119

12.1 Introduction 119
12.2 Two-dimensional aerofoils 120
12.3 Three-dimensional aerofoils 123
12.4 Spinning bodies 124

13. THERMAL FLOWS: BASIC EQUATIONS AND CONCEPTS 127

13.1 Introduction 127
13.2 Equations of eonveetion 127
13.3 Classifieation of eonveetive flows 130
13.4 Foreed eonveetion 130
13.5 Flow with eoneentration variations (mass transfer) 132

14. FREE CONVECTlON 135

14.1 Introd uetion 135
14.2 The governing non-dimensional parameters 136
14.3 The adiabatie temperature gradient 139
14.4 Free eonveetion as a heat engine 141
14.5 Conveetion from a heated vertical surfaee 142
14.6 Thermal plumes 148
14.7 Conveetion in fl uid layers 148
Appendix: The Boussinesq approximation in free eonveetion 155

15. FLOW IN ROTATlNG FLUlDS 162

15.1 Introduction 162
15.2 Centrifugal and Coriolis forees 162
15.3 Geostrophie flow and the Taylor-Proudrnan theorem 163
15.4 Taylor eolumns 165
15.5 Ekman layers 171
15.6 Intrinsie stability and inertial waves 176
15.7 Rossby waves 177
15.8 Convection in a rotating annulus 182

16. STRATIFIED FLOW 184

16.1 Basic concept s 184
16.2 Blocking 187
16.3 Lee waves 191
16.4 In temal waves 192
16.5 Stratification and rotation 199
xiv Contents

17. INSTABILITY PHENOMENA 201

17.1 Introduction 201
17.2 Surface tension instability of a liquid column 201
17.3 Convection due to internaI heat generation 203
17.4 Convection due to surface tension variations 204
17.5 Instability of rotating Couette flow 206
17.6 Shear flow instability 209

18. THE THEORY OF HYDRODYNAMIC STABILITY 213

18.1 The nature oflinear stability theory 213
18.2 Onset of Benard convection 214
18.3 Overstability 218
18.4 Rotating Couette flow 218
18.5 Boundary layer stability 220

19. TRANSITION TO TURBULENCE 225

19.1 Boundary layer transition 225
19.2 Transition in jets and other free shear flows 229
19.3 Pipe flow transi tion 233

20. TURBULENCE 238

20.1 The nature of turbulent motion 238
20.2 Introduction to the statistical description of turbulent motion 239
20.3 Formulation of the statistical description 242
20.4 Turbulence equations 244
20.5 Caleulation methods 248
20.6 Interpretation of correlations 248
20.7 Spectra 250
20.8 The concept of eddies 252

21. HOMOGENEOUS ISOTROPIC TURBULENCE 253

21.1 Introduction 253
21.2 Space correlations and the elosure problem 254
21.3 Spectra and the energy cascade 255
21.4 Dynamical processes of the energy cascade 259

22. THE STRUCTURE OF TURBULENT FLOWS 261

22.1 Introduction 261
22.2 Reynolds number similarity and self-preservation 261
22.3 Intermittency and entrainment 265
22.4 The strueture of a turbulent wake 269
22.5 Turbulent motion near a wall 275
22.6 Large eddies in a boundary layer 281
22.7 The Coanda effeet 284
Contents xv

22.8 Stratified shear flows 286

22.9 Reverse transition 290

23. EXPERIMENTAL METHODS 292

23.1 General aspects of experimental fluid dynamies 292
23.2 Velocity measurement 294
23.3 Pressure and temperature measurement 298
23.4 Flow visualization 298

24. PRACTlCAL SITVATlONS 301

24.1 Introduction 301
24.2 Cloud patterns 301
24.3 Waves in the atmospheric drculation 303
24.4 Continental drift and convection in the Earth's mantle 303
24.5 Solar granulation 306
24.6 Effluent dispersal 308
24.7 Wind effects on structures 310
24.8 Boundary layer control: vortex generators 313
24.9 Fluidics 315
24.10 Vndulatory swimming 318
24.11 Convection from the human body 320
24.12 The flight of a boomerang 323

NOTATlON 324

PROBLEMS 329

BIBLIOGRAPHY AND REFERENCES 346

INDEX 354
1
Introduction

1.1 Preamble

We know from everyday observation that liquids and gases in motion behave in very
varied and often complicated ways. When one observes them in the controlled
conditions of a laboratory, one finds that the variety and complexity of flow
patterns arise even if the arrangement is quite simple. Fluid dynamics is the study
of these phenomena. Its aims are to know what will happen in a given arrangement
and to understand why.
One believes that the basic physicallaws governing the behaviour of fluids in
motion are the usual well-known laws of mechanics - essentially conservation of
mass and Newton's laws of motion, together with, for some probIems, the laws of
thermodynamics. The complexities arise in the consequences of these laws. The
way in which observed flow patterns do derive from the governing laws is often by
no means apparent. A large theoretical and conceptual structure, built, on the one
hand, on the basic laws and, on the other hand, on experimental observation, is
needed to make the connection.
In most investigations in fluid mechanics, the physical properties of the fluid, its
density, viscosity, compressibility, etc., are supposed known. The study of such
properties and the explanation in terms of molecular strueture of their values for
different fluids do, of course, constitute another important branch of physics.
However, the overlap between the two is, in most cases, slight. This is essentially
beeause most flows can be described and understood entirely in macroscopie terms
without referenee to the moleeular structure of the fluid.
Thus, for the most part, fluid dynamical problems are concerned with the
behaviour, subject to known laws, of a fluid of specified properties in a specified
configuration. One might for example want to know what happens when oil flows
through a pipe as the resuIt of apressure difference between its ends; or when a
column of hot air rises above a heat source; or when a solid sphere is moved
through a tank of water - the whole of which might be place d on a rotating table.
Ideallyone would like to be able to solve such problems through an appropriate
mathematical formulation of the governing laws; the role of experiment would then
be to check that solutions so obtained do correspond to reality. In fact the mathe-
matical difficuIties are such that a formaI theory often has to be supplemented or
replaced by experimental observations and measurements. Even in cases where a
fairly full mathematical description of a flow has been obtained, this has often been
possible only after experiments have indicated the type of theory needed. The
subject involves an interplay between theory and experiment. The proportion each
contributes to our understanding of flow behaviour varies greatly from topic to
2 Physical Fluid Dynamics

topie. This book is biassed towards some of those topics where experimental work
has been partieularly important.
On the other hand, the book is primarily eoneerned with 'pure' fluid meehanies
rather than with applieations. It attempts to develop an understanding of the
phenomena of fluid flow by considering simple eonfigurations - simple, that is, in
their imposed eonditions - rather than more eomplieated one s that might be of
importanee in partieular applieations. This is not to deny the influenee of appllea-
tions on the development of the subjeet. Fluid dynarnies has numerous and
important applieations to engineering, to geo- and astrophysies, and to biophysics;
if this is not evident a glanee forward to Chapter 24 will make the point. The
topies chosen for investigation, even in fundamental studies, owe mueh to the
applieations eurrently considered signifieant. For example, it is doubtful whether
the intriguing phenomena that arise when the whole body of the fluid is rotating
would have reeeived mueh attention but for the importanee of such effeets in
atmospherie and oeeanic motions.

1.2 Scope of hook

Fluid dynamies has many faeets. It is neeessary in any book to make some
restrietions to the range of topies considered. The prineipal restrietions in this book
are the following six.
(i) The laws of classical meehanies apply throughout. Sinee other restrictions will
lirnit the flows considered to low speeds, the exclusion of relativistie effeets is not
signifieant. The exclusion of quantum meehanical effeets just means that we are not
dealing with liquid helium.
(ii) The length seale of the flow is always taken to be large eompared with the
moleeular mean-free-path, so that the fluid ean be treated as a eontinuum. More
precise meaning will be given to this statement in Seetion 5.2. It excludes the flow
of gases at very low pressures (rarified gases) from our eonsiderations.
(iii) We eonsider only ineompressible flow; that is flow in whieh the pressure
variations do not produee any signifieant density variations. In isothermal flows
this means that the density is a eonstant; in other flows that it is a funetion of the
temperature alone.
There are two ways in whieh fulfilment of this eondition ean eome about. The
fluid may have such a small eompressibility (such a large bulk modulus) that, even
if large pressure variations are present, they produee only slight density variations.
Or the pressure variations may be sufficiently small that, even if the eompressibility
is not so small, the density variations are small. Uquid flows ean usually be treated
as ineompressible for the former reason. More surprisingly, gas flows ean often be
similarly treated for the latter reason - whenever, as we shall see in Seetion 5.8, the
flow speed is everywhere low eompared with the speed of sound. Thus, although
this restriction excludes the many interesting phenomena of the high speed flow of
gases (eompressible flow), it still retains a wide range of important situations in the
dynamies of gases as well as of liquids.
(iv) We eonsider only Newtonian fluids. This is a statement about the physieal
properties that affeet the stresses developed within a fluid as aresult of its motion
and thus enter the dynamies of a flow. To see what is meant by a Newtonian fluid
we eonsider a simple eonfiguration, Fig. 1.1; the relationship of this to a general
Introduction 3

A --- 8

Figure 1.1 Sehematic diagram of viseous stress generated by simple veloeity variation, u(y).
Short arrows represent forees aeting in plane of AB, drawn on side of fluid on whieh force aets.

flow configuration, and thus a mo re rigorous definition of a Newtonian fluid, will

be considered in Section 5.6. In Fig. 1.1, all the fluid is moving in the sam e
direction but with a speed that varies in a perpendicular direction; Le. the only
non-zero component of the velocity is the x-component, u, and this is a function,
u(y), of the co-ordinate y. Across any arbitrary plane, perpendicular to y, within
the fluid, a stress will act. As drawn in Fig. 1.1, the faster fluid above the plane AB
will drag the fluid below forward, and the slower fluid below wiH drag the fluid
above baek. Equal and opposite forees wiH thus act on the fluid above and below
as shown (although the arrows are drawn on the sides of the fluid on which they
act, the lines of action of both forees are actually in the dividing plane AB). The
generation of this internal stress is known as viscous action. In a Newtonian fluid
the stress is directly proportional to the velocity gradient; if 7 is the foree per unit
area,
au
7=11- (Ll)
ay
11 is the coeffieient of viscosity of the fluid, often called just the viscosity.
Another way of stating the definition of a Newtonian fluid is to say that it is a fluid
of constant viscosity. Constant here means that it does not depend on the veloeity
fie1d; the viscosity of a Newtonian fluid may (and usually does) vary with temper-
ature OI it may (but usually does not) vary with pressure.
All gases and liquids with small molecules interacting in a simple way are
observed to be very c10sely Newtonian. Non-Newtonian behaviour can arise in
liquids with long molecules (polymers), in solutions of polymers, in liquids where
the molecules tend to gather in mOIe organized struetures than usual, and in
suspensions and emulsions (the former name being used when solid material is
dispersed through a liquid and the latter when drops of one liquid are dispersed
through another). Thus all the common fluids one is likely to use in the labOIatory
are Newtonian, as are many of the fluids oecurring in important applications. How-
ever, by restricting ourselyes to Newtonian fluids, we are exc1uding various fluids of
importance in biophysics (e.g., blood flowing through small capiHaries), industrial
applications (e.g., many paints), and chemical engineering.
4 Physical Fluid Dynamics

(v) We shall not (with two exceptions, Sections 17.2 and 17.4) consider flows
with fr ee surfaces. Topics such as waves on a water surface, flow in open channels,
and the dynamics of bubbles and drops are thus outside the scope of the book.
(vi) We shall not consider any problems in which electromagnetic effects are
important, either purely electrohydrodynamic problems (such as the behaviour of a
liquid of variable dielectric constant in an electric field) or magnetohydrodynamic
problems (such as the flow of an electrically conducting fluid in a magnetic field).
All the above restrictions mean, in varying degrees, that important and interesting
topics are being omitted. But the range of phenomena that remains within our scope
is wide and varied. We shall be dealing primarily with flows induced by imposed
pressure gradients, flows arising from the relative motion of boundaries or of one
boundary and ambient fluid, convection - that is flows induced by of associated
with temperature variations - and flows strongly affected by rotation of the whole
system or by density stratification. This may not at first sight seem a very broad
spectrum of topics, but each has many facets. A major reason for this is the
frequent occurrence of instabilities, leading to the breakdown of one type of flow
into another. The next three chapters will illustrate the variety and complexity
of the phenomena that occur, and illustrate in particular that the complexity can
arise even for very simple imposed conditions. We may also remark here that the
examples of applications of fluid dynamics in Chapter 24, drawn from widely
assorted branches of applied science, have all been chosen as cases that can be under-
stood, at least in part, within the limitations of this book.
Throughout most of this book, except where specific experimental arrangements
are concerned, we shall talk about fluids, without making any distinction between
liquids and gases. This is because the range of situations within the above limitations
is just the range in which liquids and gases exhibit the same phenomena (provided
the comparison is made in the correct quantitative way - see Chapter 7); circum-
stances in which any one of the limitations, except (vi), did not apply would
normally refer specifically either to a liquid or to a gas. Points (iii) and (iv) are
particularly important. The facts that in many flows gases behave as if they were
incompressible and that the same law of viscous behaviour applies to gases and
many liquids are central to the development of a common dynamical description for
the two phases.

1.3 Notation and definitions

A list of symbols used in this book is given on pages 324-328. This is intended as
an aid when a symbol reappears in the text sometime after it has been first intro-
duced.
Here we may just note the symbols for the basic fluid dynamical quantities
appearing throughout the book. Fluid velocity is denoted by u (with Cartesian
components (u, v, w) in directions (x, y, z)), the pressure by p, the temperature by
T, fluid density by p, the viscosity (already introduced in Section 1.2) by 11, the
kinematic viscosity CIl/p) by v, and time by t.
Many of the dynamical and physical quantities appearing in fluid dynamics are
standard quantities with accepted definitions familiar to a physicist. It has not been
thought necessary to de fine such quantities in the text, but their dimensions are
quoted in the list of symbols when this provides a useful rerninder of the definition.
Introduction 5

It is conventional to use a double letter symbol for most non-dimensional

quantities (e.g., Re for Reynolds number, Ra for Rayleigh number). Such symbols
will be printed non-italicized to distinguish them from products of two quantities.
Two terms that will be used frequently require definition. Both refer to partic-
ular elasses of flow that are often considered because of their relative simplicity.
Firstly, a steady flow is one which does not change with time. An observer
looking at such a flow at two different instants will see exactly the same flow
pattem, although the fluid at each position in this pattem will be different at the
two instants. Mathematically (using the Eulerian system to be introduced in Section
5.3), steady flow may be expressed
ajat = 0 (1.2)
where the derivative operates on any parameter associated with the flow. Steady
flow can occur only if all the imposed conditions are constant in time. This means
that a flow is steady only in the appropriate frame of reference (flow past a fixed
obstaele may be steady, but the same situation seen as the obstaele moving through
the fluid is not steady, even though the two cases are dynamically equivalent - see
Section 3.1). However, one would normally choose that frame for study of the flow.
A flow that is changing with time is, of course, called unsteady. An intrinsically
unsteady flow is one that is not steady in any frame of reference. Such a flow must
occur if there is no frame in which the imposed conditions remain fixed. We shall be
seeing that intrinsically unsteady flow also sometimes arises spontaneously even
when the imposed conditions are steady.
Secondly, a two-dimensionaI flow is one in which the motion is confined to
paralleI planes (the velocity component in the perpendicular direction is zero every-
where) and the flow pattem in every such plane is the same. Formally,
w = 0, a/az = 0 (1.3)
Such a motion may occur in an effectively two-dimensional geometry with the
ends in the third direction so distant that they have negligible effect on the flow in
the region of interest.
The significance of these concepts will become elearer through specific examples
in the following chapters.
2
Pipe and Channel Flow

2.1 Introduction

In this and the next two chapters, we take three geometrica1ly simple flow configur-
ations and have a look at the principal flow phenomena. These will provide a more
specific introduction than the last chapter to the character of fluid dynamics. We
consider these examples now, before starting on the formai development of the
subject in Chapter 5; we can then approach the setting up of the equations of
motion with an idea of the types of phenomena that one hopes to understand
through these equations. Although these chapters are primarily phenomenological,
the present chapter will also be used to introduce some simple theoretieal ideas.
The first topic is viscous incompressible flow through pipes and channels.
Consider a long straight pipe or tube of uniform circular cross-section. One end of
this is supplied by a reservoir of fluid maintained at a constant pressure, higher than
the constant pressure at the other end. A simple arrangement for doing this in
principle, using a liquid as the working fluid, is shown in Fig. 2.1; practical arrange-
ments for investigating the phenomena to be described require some refinement of
this arrangement. Fluid is pushed through the pipe from the high pressure end to
the low. We suppose that the gravitational force on the fluid is irrelevant, either
because the pipe is horizontal or because this force is small compared with the
forees associated with the pressure differences. Although the re are other
experiments that one can do with pipes, this configuration is usually known as pipe
flow.
Channel flow is the two-dimensional counterpart of pipe flow. Flow is supposed

~~
- - - - H --
- - - -~~ -

[Link]=::::~:L
Supply
-- - - - - - - - -
--
Overllow

-
Outlet to
atmosphere

Pipe
- --- ---

Figure 2.1 Simple pipe flow experimental arrangement. The pipe length is redueed in seale.
Pipe and Channel Flow 7

to occur between two paralleI planes c10se together. The pressure difference is
maintained between two opposite sides of the gap. The other two sides must be
walled but are supposed to be so far away from the working region that they have
no effeet there. This is obviously a more difficult arrangement to set up experiment-
ally, and the description of observations wiH be given in the context o(pipe flow.
However, a simple piece of theory can be developed about one possible flow pattern,
and it is convenient to consider this first for channel flow.

2.2 Laminar flow theory: ehaooel

Figure 2.2 shows the notation for this theory. The channel width is taken as 2a and
its length as 1; we are supposing that 1 is much larger compared with a than one can
show in the diagram. Pressuresp! and P2 (P! > P2) are maintained at the ends of
the channel. Co-ordinates are ehosen as shown with the zero of y on the mid-plane
of the channel.
It is observed (see Section 5.7) that the fluid immediately next to the walls
remains at re st - a fact known as the no-slip condition. The speed, u, with which
the fluid moves in the x-direction must thus be a function of y - zero at y = ± a,
non-zero elsewhere. This distribution u(y) is known as the velocity profile.
The remoteness of the other walls, in the z-direction, is taken to imply that the
flow is two-dimensional (see Seetion 1.3). We can thus consider all processes to be
occurring in the plane of Fig. 2.2.
For the present theory we also suppose that the velocity profile is the same at all
distances down the channel; that is at all x. Since the density p is taken to be
constant as discussed in Section 1.2(iii), this is obviously one way of satisfying the
requirement that the flow should conserve mass. The same velocity profile trans-
ports the same amount of mass per unit time past every station.
The pressure varies with x, obviously, but is constant across the pipe at each x.
We shall see below that apressure gradient in a given direction generates a force in
that direction; the re is nothing to balanee such a force in the y- or z-direction.
We consider now the forees acting on a small element of fluid of sides OX and oy
as shown in Fig. 2.2 and side oz in the third direction. There are two processes
giving rise to such forees - the action of viscosity as described in Section l.2(iv),
and the pressure.
The expression for the viscous force iHustrates an important general point that,
although the viscous stress depends on the first spatial derivative of the velocity,
the viscous force on a fluid element depends on the second derivative. The net force

Figure 2.2 Definition diagram for ehannel flow. Width of ehannel is shown exaggerated with
respeet to length.
8 Physical Fluid Dynamics
y

e ----t-------..,==;:-----,t-..--- D
õy
A--;---===----I-- --....t- S

Figure 2.3 Extension of Fig. 1.1 to show viscous stresses acting on fluid element.

on the element is the small difference of the viscous stresses on either side of it.
Figure 2.3 extends Fig. 1.1 to show this. Per unit area perpendicular to the y.
direction, forees p(aujay)y+oy and -p(aujay)y act in the x·direction on the region
between planes AB and CD. The net force on our element is

[ p(au) _p(au) ]oxoz=~(pau)OYOXOZ

ay y+liy ay y ay ay (when oy is small enough)
a2 u
=p-oxoyoz (when p is constant)
ay2
(2.1)
The viscous force per unit volume is pa 2 ujay 2 • In the present case, for which u is
independent of x and z, this may be written pd 2 u/dy 2 • From the general shape of
the velocity profile in Fig. 2.2, or from the physical expectation that viscous action
will oppose the flow, we anticipate that d2 u/dy 2 will be negative.
The pressure decreases as one goes downstream; the re will be slightly different
pressure forees acting on the two ends of the element. Since pressure is force per
unit area, that at the upstream end acting in the downstream direction on the
element is Pxoyoz (px denoting the value of p at x) and that at the downstream end
acting in the upstream direction is

Px+lixoyoz = (Px + :: ox) oyoz (2.2)

(for small enough ox). The net force in the downstream direction is
ap
--8xoy8z (2.3)
ax
or -apjax per unit volume. This will be positive.
Again the partial derivative (which has been written because we shalllater be
100king at this matter in a more general context) may be re place d by a total deriva-
tive, because p varies only with x. Further, the assumption of an unchanging velocity
profile makes the dynamical processes the same at all stations downstream; the
pressure force per unit volume - Le. the pressure gradient - must be independent
Pipe and Channel F/ow 9

ofx. Hence,
ap dp Pl - P2
- ax =- dx = I =e, say (2.4)

The momentum of the element oxoyoz is not changing; each fluid particle
travels downstream at a constant distance from the centre of the channel and so
with a constant speed. Hence, the total force acting must be zero,
a2 u ap
JJ. -a 2 oxoyoz - -oxoyoz = 0 (2.5)
!y ax
that is
d2 u
JJ.-=-e (2.6)
dy 2
With the boundary conditions
u =0 at y =±a (2.7)
this integrates to give
e
u=-(a2 _y2) (2.8)
2JJ.
We have aseertained that the velocity profile is a parabola.
The mass of fluid passing through the channel per unit time and per unit length
in the z-direction is

f
a
-a pudy = 2epa 3 /3JJ. (2.9)

2.3 Laminar flow theory: pipe

The corresponding flow in a pipe is usually known as Poiseuille flow (or sometimes,
in the interests of historieal accuracy, Hagen-Poiseuille flow). This case is margin-
ally more complicated because of the cylindrical geometry. Figure 2.4 shows a
cross-section of the pipe; the flow direction (the x-axis) is norm al to the page. The
velocity profile now represents the speed as a function of radius, u(r), and we again

Figure 2.4 Defmition sketch for pipe flow.

10 Physical Fluid Dynamics
consider the case when this is independent of x. We consider an element of fluid as
shaded in Fig. 2.4 and having length öx in the flow direction. The viscous forees on
the two faces of this now differ slightly not only because the velocity gradients
differ but also because the two faces have different areas. The force on one face is
p(ou /or)rörpöx (2.10)
and, by an argument paralleI to that in Section 2.2, the net viscous force on the
element is

(2.11)

The pressure force on one end of the element is

prörpör (2.12)
and the net pressure force
-(op/ox )röxörpör (2.13)
Arguing in the same way as for channel flow we get

p !(r :) = -Gr (2.14)

Integration gives

Gr
u=--+Alnr+B
2
(2.15)
4p
A must be zero for the velocity not to become infinite on the axis and B can be
evaluated from the fact that
u =0 at r =a (2.16)
giving
G
u = -(a2 - r2 ) (2.17)
4p
The velocity profile is a paraboloid with a maximum speed
Ga 2
u max = 4J.l (2.18)

The mass per unit time, or mass flux, passing through the pipe is

I o p2
a
1fTU
dr _ rrpGa 4 _ rrp(Pl - P2)a4
----
8p 8p[
(2.19)

This is a quantity of some im portance as it can readily be measured. Agreement

with observation (under circumstances to be delimited below) provides an important
check of the validity of underIying hypotheses, such as the no-slip condition and the
applicability of continuum mechanics (see Chapter 5). Alternatively, an unknown
viscosity can be determined from the rate of flow of the fluid through a tube under
a known pressure gradient. This is the principle of one important type ofviscometer.
Pipe and Channel Flow 11

An average speed can be defined as the mass flux divided by the density and
cross-sectional area
Ga 2
u av = 8J,J.
(2.20)

2.4 The Reynolds number

We have determined above one theoretieally possible flow behaviour. It is not the
only possibility. So me times the actual flow behaviour corresponds to the theory,
sometimes not. In order to specify the circumstances in which the different types
of flow occur, we need to introduce the concept of the Reynolds number.
There are several types of variable associated with the pipe flow configuration:
the dimensions of the pipe, the rate of flow, the physical properties of the fluid.
For present purposes, we will suppose that the situation is fully specified if we
know:
d (= 2a), the pipe diameter
u av , the average flow speed
p, the fluid density
J,J., the viscosity
Two omissions from this list require some comment. The pipe length is not in elude d
on the supposition that, provided the pipe is long enough, the type of flow is deter-
mined before the downstream end can have any influence. The pressure gradient is
not ineluded because it cannot be varied independently of the above parameters. To
produce the same average flow-speed of the same fluid through the same pipe will
require the same imposed pressure gradient - whether or not the Poiseuille flow
relationship applies. Hence, the pressure gradient need not be ineluded in the
specification of the situation. It is arguable that one should in elude the pressure
gradient and omit the average speed, since the former is the controlled variable in
most experiments. In Section 19.3 we shall see that there is one case in which this
would certainly be the better procedure. However, this has not been conventional
practice, and to adopt it would make for confusion. We notice that u av relates
directly to the total rate of flow through the pipe and must be the same at every
station along the length - unlike u max and other speed s one might define that
depend on the detailed velocity profile.
We are now concerned with the question: what type of flow occurs for given
values of d, u av , p, and J,J.? But these four parameters are dimensional quantities,
whereas the concept of a 'type of flow' does not have dimensions associated with it.
Just as it is meaningless to write down an equation, A = B, with A and B of differ-
ent dimensions, so it is meaningless to associate a type of flow with certain values
of any dimensional quantity. For example, one would not expect a partieular type
of flow to occur over the same range of U av for pipes of different diameter or for
different fluids. The values of u av specifying the range must be expressible (in
principle, whether or not in practice) in terms of the things that determine it. There
must be some (known or unknown) expression for it, and this expression must
bring in other dimensional quantities in order to be dimensionally satisfactory. Thus
we look for the factors that determine the type of flow in terms of dimensionless,
12 Physical Fluid Dynamics
not dimensional, parameters. Such dimensionless pararneters must be provided as
combinations of the speeifying dimensional parameters. (We have arrived by a
plausibility argument at a conclusion that we shall be exarnining more systematically
in Chapter 7.)
There is one dimensional eombination of the four parameters for the present
eonfiguration
Re =puavd (2.21)
JJ.
This is known as the Reynolds number of pipe flow.
It is one example of the general definition of the Reynolds number as
Re =pUL = UL (2.22)
JJ. Il

where U and L are veloeity and length seales; that is typieal measures of how fast
the fluid is moving and the size of the system. As the book proeeeds, we shall be
seeing the relevanee of forms of Reynolds number to a variety of situations.
Henee, it is appropriate to diseuss the different types of flow in a pipe in terms
of different ranges of Reynolds number.

2.5 The entry length

When the Reynolds number is le ss than about 30, the Poiseuille flow theory
always provides an aeeurate description of the flow. In faet Equation (2.21) was
first established empirleally by Hagen and Poiseuille and the theory was given later
by Stokes - a somewhat surprising historieal faet if one knows howeasy it is to fall
to verify the theory as aresult of having the Reynolds number too high!
At higher Reynolds numbers, the Poiseuille flow theory applies only after some
distanee down the pipe. The fluid is unlikely to enter the pipe with the appropriate
parabolie veloeity prome. Consequently, there is an entry length in whieh the flow
is tending towards the parabolie prome. At low Reynolds number, this is so short
that it ean be ignored. But it is found both experimentally and theoretically that,
as the Reynolds number is inereased, this is no longer true. The details of the entry
length depend, of eourse, on the aetual veloeity prome at entry, whieh in turn
depends on the detailed geometry of the reservoir and its eonneetion to the pipe.
However, an important case is that in whieh the fluid enters with a uniform speed
over the whole eross-section. Beeause of the no-slip eondition, the fluid next to the
wall must immediately be slowed down. This retardation spreads inward, whilst
fluid at the eentre must move faster, so that the average speed remains the same
and mass is conserved. One thus gets a sequenee of veloeity promes as shown in
Fig. 2.5. Ultimately, the parabolie prome is approaehed and from there onwards
the Poiseuille flow theory applies.
We ean use this ease to illustrate the dependenee of the extent of the entry
length on the Reynolds number. Defining X as the distanee downstream from the
entry at whieh u max is within 5 per ee nt of its Poiseuille value [233]
X Re
d ~ 30 (2.23)
Pipe and Channel Flow 13

o 0.5 1.0 1.5

uluav

Figure 2.5 Laminar velocity profiles in pipe entry length. Based on average of various
experimental and theoretieal profiles, as collected together in Ref. [233).

This means, for example, that for flow at a Reynolds number of 10 4 (chosen as a
high value at which this type of flow can occur) in a pipe of diameter 3 cm the entry
length is 10 m long. Evidently there will be many practical situations in which the
Poiseuille flow pattem is never reached. Even if it is reached, it will often not
occupy a sufficiently large fraction of the length for equation (2.19) to be a good
approximation to the relationship between pressure drop and mass flux.
Incidentally, the length at the other end over which the presence of the outlet
has an effect is always relatively small.

2.6 Transition to turbulent flow

The above is an important, but perhaps rather uninteresting limitation to the occur-
rence of Poiseuille flow. As the Reynolds number is increased further, there is a much
more dramatic change in the flow. It undergoes transition to the type of motion
known as turbulent, the flows considered so far being called laminar.
In laminar pipe flow the speed at a fixed position is always the same. Each
element of fluid travels smoothly along a simple well-defined path. (In Poiseuille
flow, this is a straight line at a constant distance from the axis; in the entry length
it is a smooth curve.) Each element starting at the same place (at different times)
follows the same path.
14 Physical Fluid Dynamics

t
Velocity

_Time

Figure 2.6 Example of velocity variations in turbulent flow.

When the flow beeomes turbulent, none of these features is retained. The flow
develops a highly random eharaeter with rapid irregular fluetuations of velocity in
both space and time. An example of the way in whieh (one eomponent of) the
velocity at a fixed position fluctuates is shown in Fig. 2.6. An element of fluid
now follows a highly irregular distorted path. Different elements starting at the
same place follow different paths, since the pattem of irregularities is changing all
the time. These variations of the flow in time arise spontaneously, although the
imposed conditions are all held steady.
Figure 2.7 shows the effeet of the ehangeover from laminar to turbulent flow on
dye introdueed continuously near the entry of a pipe in a streak thin compared
with the radius of the pipe. (This arrangement is, in its essentials, the one used by
Reynolds [207] in the experiments that may be regarded as the genesis of
systematic studies of transition to turbulence - although the re had been some
earlier work by Hagen.) When the flow is laminar, the dye just travels down the
pipe in a straight or almost straight line as shown in the upper picture. As the flow
rate is increased, thus increasing the Reynolds number, the pattem changes as shown
in the lower pieture. The dye streak initially travels down the pipe in the same way
as before, but, after some distance, it wavers and then suddenly the dye appears
diffused over the whole cross-section of the pipe. The motion has become turbulent
and the rapid fluctuations have mixed the dye up with the undyed fluid.
The distance fluid travels downstream before becorning turbulent varies with
time, and , at any instant, the re can be a laminar region downstream of a turbulent
region. This comes about in the following way. The turbulenee is generated initially

High pressure
reservoir /

;::::::=::====~~--;:=c;1.1/IIlJTlJ7I~77/ZmmnraT777Z

Figure 2.7 Dye streaks in laminar and turbulent pipe flow. (Pipe length compressed relative
to other dimensions.)
Pipe and Channel Flow 15

------~ ............ ~-----------------

Figure 2.8 Growth and transport of turbulent plug. Shaded regions are turbulent, unshaded
laminaroThe mean fluid speed is approximately midway between the speed s of the front and
rear of the plug.

over a small region. This region is actually localized radially (elose to the wall) and
azimuthally as well as axially. However, it quickly spreads over a cross-section of
the pipe, and the re is then a short length of turbulent flow with laminar regions both
upstream and downstream of it. This short length is known as a turbulent plug. (We
postpone till Section 19.3 further consideration of the origin of turbulent plugs.)
The turbulence then spreads - the laminar fluid next to each end of the turbulent
plug is brought into turbulent motion - and the plug gets longer. The fluid is, of
course, meanwhile traveHing down the pipe. Hence, the development of a plug is as
shown in Fig. 2.8. The shapes of the interfaces are indicated roughly, based on
photographs such as those in Fig. 2.9. As the plug grows the interfaces soon occupy
a very short length of the pipe compared with the plug itself, and so the laminar and
turbulent regions are well demarcated.
After a while another plug is born in a similar way. By this time the previous
plug has moved off downstream so there is again laminar fluid downstream of the
new plug. The plugs sometimes originate randomly in time, sometimes periodically;
the circumstances in which the two cases occur will be discussed further in
Section 19.3. When the front interface of one plug meets the rear interface of
another, as aresult of their growth, the two simply merge to give a single longer
plug.

Figure 2.9 Photo sequence showing passage of turbulent plug past fixed observation point,
similar to those in Refs. [167; 169). Flow is downwards; time increases from left to right. Plug
enters field of view at top of first frame (which also shows end of earlier plug at bottom) and
leayes it at bottom of seventh frame. Flow visualization by addition of small amount of
birefringent material - refractive index for polarized light affected by shear (Refs. [32, 166).
16 Physical Fluid Dynamics

a
b
e
Figure 2.10 Oscillograms ofvelocity fluctuations at the centre of a pipe. Traces (a) and (b)
(from Ref. (211)) show random plug production for Re = 2550, I/d = 322; trace (a) was given
by a.c. amplification of the signal and shows the velocity fluctuations in the plugs; trace (b)
was given by d.c. amp1ification of the same signal and show s principally the local mean velocity
change between laminar and turbulent flow. Trace (e) (from Ref. (194) is the counterpart of
traee (b) for periodie plug produetion for Re "" 5000, I/d = 290. Note: veloeity increases
upwards in traees (a) and (b) but decreases upwards in traee (e).

As aresult of these processes, a sensor at a fixed point in the pipe observes

llternately laminar and turbulent motion. Figure 2.10 shows oscillograms of
velocity fluctuations arising in this way, for random and for periodic plug
production. The abrupt changes between laminar and turbulent motion illustrate
again the sharpness of the interfaces.
The fraction of the time that the motion is turbulent - known as the inter·
mittency factor - increases with distance downstream as aresult of the growth of
the plugs. Far enough downstream, in a long enough pipe, the laminar regions have
all been absorbed and the flow is fully turbulent.
From the considerations earlier in this section, one would expect that there
should be a critical value of the Reynolds number, below which the flow is wholly
laminar, above which transition to turbulence occurs. In fact, the situation is
more compllcated than that. The transition process is extremely sensitive to the
detailed geometry of the entry from the reservoir and to the level of small
disturbances in the incoming fluid. As aresult, transition has been observed to
start at values of the Reynolds number ranging from 2 x 10 5 • (This will
be discussed further in Section 19.3.) The impllcation of this variation is not that
the reasoning establishing the role of the Reynolds number was erroneous, but that
dimensional parameters other than the four listed (d, u av ' p, and Jl) are relevant to
transition.

2.7 Relationship between flow rate and pressure gradient

The pressure difference needed to produce a given flow rat e through a pipe is
larger when the flow is turbulent than when it is laminar. This is shown in Fig. 2.11.
Since the abscissa is the Reynolds number, it is appropriate that the ordinate
should be a non·dimensional form of the average pressure gradient (Pl - P2)/I,
chosent here as (Pl - P211 3 p/Jl 2 1. The dotted lines in Fig. 2.11 show this parameter

tA more conventional choice would be (p 1 - P2)d/pu'iv1 (the above parameter divided by

(Re)2). However, because it does not contain u av , our choice is more convenient for certain
considerations in Seetion 19.3. It also shows the behaviour in a partieular pipe more directly.
Pipe and Channel Flow 17

10 ' r-----T---r-,_~----,_----~--,__r~----_,
S
6

10'
S
6
(p, p , )dlp
/J ~~ 4

10'
S
6

10'L-__ L-~~L-~~ ____~____~__~-L~____~

2.10' 4 6 S 10 ' 2 4 2
Ac

Figure 2.11 Variation of non-{jimensional average pressure gradient with Reynolds number
for pipe flow. Dotted lines: wholly laminaI flow and wholly turbulent flow. Full line: example
of actual case. Based on information in Refs. [38, 211).

plotted (logarithmically) against Reynolds number for Poiseuille flow and for
wholly turbulent flow. The continuous line shows the behaviour for an actual case.
At low Reynolds number it follows the Poiseuille flow line; as transition starts it
rises above this, and, ultimately, when transition is complete in a short fraction of
the totallength, approaches the turbulent flow line. (The case chosen shows
transition at a relatively low Reynolds number. With higher Reynolds number
transition in a pipe of practicable length, there is likely to be some departure from
the Poiseuille law before transi tion because of the importance of the entry length.)
3
Flow Past a Circular Cylinder

3.1 Introduction

Relative motion between some object and a fluid is a common occurrence. übvious
examples are the motion of an aeroplane and of asubmarine and the wind blowing
past a structure such as a tall building or a bridge. Practical situations are however
usuaUy geometrically complicated. Here we wish to see the complexities of the flow
that can arise even without geometrical complexity. We therefore choose a very
simple geometrical arrangement, and one about which the re is a lot of information
available.
This is the two-dimensional flow past a circular cylinder. A cylinder of diameter
d is placed with its axis normal to a flow of free strearn speed uo; that me ans that
Uo is the speed that would exist everywhere if the cylinder were absent and that still
exists far away from the cylinder. The cylinder is so long compared with d that its
ends have no effect; we can then think of it as an infinite cylinder with the same
behaviour occurring in every plane normal to the axis. Also, the other boundaries to
the flow (e.g. the walls of a wind-tunnel in which the cylinder is place d) are so far
away that they have no effect.
An entirely equivalent situation exists when a cylinder is drawn perpendicularly
to its axis through a fluid otherwise at rest. The only difference between the two
situations is in the frame of reference from which the flow is being observed.
(Aspects of relativity theory are already present in Newtonian mechanics.) The
velocity at each point in one frarne is given by the vectorial addition of Uo onto the
velocity at the geometrically similar point in the other frarne (Fig. 3.1). This
transformation does not change the accelerations involved or the velocity gradients
giving rise to viscous forces; it thus has no effect on the dynarnics of the situation .

•
u.
\7u,
V ~

Ae·
u.

Figure 3.1 Velocity veetors u, and u 2 at pointA in two frames of referenee.

Flow Past a Circular Cylinder 19

(These remarks apply only when Uo is constant; if it is changing then one does have
to distinguish between the cylinder accelerating and the fluid accelerating.)
It is, however, convenient to use a particular frame of reference for describing
the flow. Except where otherwise stated, the following description will use the
frame of reference in which the cylinder is at rest.

3.2 The Reynolds number

One can have various values of d and Uo and of the density, p, and viscosity, Il, of
the fluid. For reasons like those applying to pipe flow (Section 2.4) the important
parameter is the Reynolds number
puod
Re=- (3.1)
Il
This is the only dimensionless combination and one expects (and finds) that the
flow pattem will be the same when Re is the same. We thus consider the sequence
of changes that occurs to the flow pattem as Re is changed. We are concemed with
a very wide range of Re. In practice this means that both Uo and d have to be varied
to make observations of the full range. For example, Re = 10 -I corresponds in air
to a diameter of 10 Ilm with a speed of 0.15 m s -I (or in glycerine to a diameter of
10 mm with a speed of 10 mm S-I); Re = 106 corresponds in air to a diameter of
0.3 m with a speed of 50 m s -I. Thus experiments have been done with cylinders
ranging fromfine fibres to ones that can be used only in the largest wind tunneIs.

3.3 Flow patterns

The following description of the flow pattems is based almost entirely on experi-
mental observations. Only for the lowest Reynolds numbers can the flow as a
whole be deterrnined analytically (Section 9.5), although there are theoretieal
treatments of aspects of the flow at other Reynolds numbers. Some of the flow
pattems have also been studied computationally.
Figure 3.2 shows the flow when Re ~ 1. The lines indicate the paths of elements
of fluid. The flow shows no unexpected properties, but two points are worth noting
for comparison with higher values of the Reynolds number. Firstly, the flow is
symmetrical upstream and downstream; the right-hand half of Fig. 3.2 is the mirror
image of the left-hand half. Secondly, the presence of the cylinder has an effect
over large distanees; even many diameters to one side the velocity is appreciably
different from uo.
As Re is increased the upstream-downstream symmetry disappears. The partide
paths are displaced by the cylinder for a larger distance behind it than in front of it.
When Re exceeds about 4, this leads to the feature shown in the computed flow
pattem of Fig. 3.3. Fluid that comes round the cylinder dose to it moves away
from it before reaching the rear point of symmetry. As aresult, two 'attached
eddies' exist behind the cylinder; the fluid in these circulates continuously, not
moving off downstream. These eddie s get bigger with increasing Re; Fig. 3.4 shows
a photograph of the flow at Re ~ 40 just before the next flow development takes
place. (For further computed flow pattems in this regime see Section 6.2.)
20 Physical Fluid Dynamics

Figure 3.2 Low Reynolds number flow past a eireular eylinder.

The tendency for the most striking flow features to occur downstream of the
cylinder becomes even more marke d as one goes to higher Reynolds numbers.
This region is called the wake of the cylinder. The reason why the re should be a
distinctive velocity distribution in the wake will be considered in Section 3.4, and
in Section 17.6 we shall be seeing that this distribution is one of a kind prone to
instability. In consequence, for Re > 40, the flow becomes unsteady. As with
transition to turbulence in a pipe (Section 2.6), this unsteadiness arises

Figure 3.3 Flow past a eireular eylinder at Re = 10 (computed: Ref. [95) .

F/ow Past a Circular Cylinder 21

Figure 3.4 Attached eddies on circular cylinder at Re = 41, exhibited by coating cylinder
with dye (condensed milkl. From Ref. [245].

spontaneously even though all the imposed conditions are being held steady.
Figure 3.5 shows a sequence of pattems in a cylinder wake, produced by dye
emitted through a small hole at the rear of the cylinder. The instability develops to
give the flow pattem, known as a Karman vortex street, shown schematically in
Fig. 3.6. Concentrated regions of rapidly rotating fluid - more precisely regions of
locally high vorticity (a term to be defined in Section 6.4) - form two rows on
either side of the wake. All the vortices on one side rotate in the same sense, those
on op posit e sides in opposite senses. Longitudinally, the vortices on one side are
mid-way between those on the other.
The whole pattem of vortices travels downstream, but with a speed rather
smaller than uo. This means that for the other frame of reference, a cylinder pulled
through stationary fluid, the vortices slowly follow the cylinder. Figure 3.7 shows
a photograph taken with this arrangement. A moderately long exposure shows the
motion of individual partieles in the fluid. Because the vortices are moving only
slowly relative to the camera, the circular motion associated with them is shown
well.
Although, at Reynolds numbers a little above 40, the vortex street develops
from an amplifying wake instability [62,296] (Fig. 3.5), increasing Reynolds
number leads to an interaction with the flow in the immediate vicinity of the
cylinder [60,274] . When Re is greater than about 100, the attached eddies
are periodically shed from the cylinder to form the vortices of the street.
Whilst the eddy on one side is being shed that on the other side is re-forming.
Figure 3.8 shows a elose-up view of the region immediately behind the cylinder,
with the same dye system as in Fig. 3.5, during this process.
Figure 3.9 shows a side-view of a vortex street, again shown by dye released at
the cylinder. So me times the vortices are straight and elosely parallel to the cylinder
(indicating that the shedding occurs in phase all along the cylinder); sometimes they
are straight but inclined to the cylinder, as in the top part of Fig. 3.9 (indicating a
linear variation in the phase of shedding); and sometimes they are curved, as in the
bottom part of Fig. 3.9 (indicating a more complicated phase variation). The exact
behaviour is very sensitive to disturbanees and so it is difficult to predict just what
will occur at any given Reynolds number. One can, however, say that the lower the
Reynolds number the greater is the likelihood of straight, parallel vortices.
22 Physical Fluid Dynamics

t
r

e
, •
..• f •
f e.
..

f
I '

Figure 3.5 Cylinder wakes exhibited by dye introdueed through hale in eylinder. (CyIinder is
abseured by braeket but ean be seen abliqueIy, partieuIarly in (a) and (f).) (a) Re '" 30;
(b) Re '" 40; (e) Re = 47; (d) Re = 55; (e) Re = 67; (f) Re = 100.
Flow Past a Circular Cylinder 23

Figure 3.6 Relative positions of vortiees in Karman vortex street.

It should be emphasized that, in Figs. 3.5, 3.8, and 3.9, the introduction of dye
was continuous. The gathering into discrete regions is entirely a function of the
flow.
Figure 3.10 shows a sequence of oscillograms at two points fixed relative to a
cylinder in a wind-tunnel as the main flow velocity is increased. In Fig. 3.10(a) and
(b) it can be seen that the passage of the vortex street past the point of observation
produces an almost sinusoidal variation. The frequency , n, of this is usually specified
in terms of the non-dimensional parameter
St = nd/uo (3.2)
known as the Strouhal number. St is a function of Re (although a sufficiently
slowly varying one that it may be said that St is typically 0.2).
With increasing Re the st rong regularity of the velocity variation is lost (Fig.
3.10). An important transition occurs at a Reynolds number a little below 200.
Below this, the "\'ortex street continues to all distanees downstream. Above, it
breaks down and produces ultimately a turbulent wake, although, as we shall be
seeing below, the intermediate stages take different forms over different Reynolds
number ranges. Figure 3.11 is a photograph of a turbulent wake. The name turbulent
implies the existence of highly irregular rapid velocity fluctuations, in the same way
as in Section 2.6. The turbulence is confined to the long narrow wake region

Figure 3.7 Vortex street exhibited by motion of partides floating on water surfaee, through
which eylinder has beendrawn; Re = 200. From Ref. [266).
24 Physical Fluid Dynamics

Figure 3.8 Motion immediately behind cylinder; Re = 110.

downstream of the eylinder. The eharaeter of a turbulent wake will be diseussed in

mo re detail in Seetion 22.4.
Despite the transition to turbulenee, markedly periodie vortex shedding remains
a eharaeteristie of the flow up to the highest Reynolds numbers (~107) at whieh
observations have been made. When this is not immediately apparent from
oseillograms like Fig. 3.10, Fourier analysis reveals the presenee of a dominant

Figure 3.9 Side-view of vortex street behind cylinder; Re = 150. From Ref. [61).
Flow Past a Circular Cylinder 25
Re ~ 85; x/d = 8 (a)

Re = 85; x/d ~ 23 (b)

Re = 185; x/d ~ 8 (e)

Re = 185;x/d = 23 (d)

Re = 280; x/d = B (e)

Re = 270; x/d - 23

Re = 640; x/d ~ 8

Re ~ 640; x/d = 23 (h)

r~
Figure 3.10 Oscillograms of velocity fluctuations in cylinder wake. Traces are in pairs at
same Re, but different distanees, x, downstream from cylinder. For both positions, probe was
slightly off-eentre to be influenced mainly by vortices on one side of street. (Notes: relative
velocity amplitudes are arbitrary; time scale is expanded by factor of about 3 in traces (g)
and (h).)
26 Physical F/uid Dynamics

Figure 3.11 Turbulent wake exhibited by dye emitted from cylinder (out of picture to right).
From Ref. [118].

frequency n amongst the other more random fluctuations. There are two ranges,
200 < Re < 400 and 3 x 10 5 < Re < 3 x 106 , in which the regularity of shedding
decreases: in the former, the Strouhal number shows a lot of scatter; in the latter,
the periodicity is lost except perhaps very elose behind the cylinder [62,209].
However, at the top end of each range, another transition occurs which restores the
regularity; the Strouhal number becomes weIl defined again.
The change at Re ~ 200 arises from instability of the vortex street to three-
dimensional disturbanees [66]. Bends - such as that shown at the bottom of Fig.
3.9 for lower Re - increase in amplitude and irregularity as the vortices travel
downstream. The effect can be seen in Fig. 3.10 by comparing the oscillograms at
the two stations. UItimately the irregularity becomes dominant and the wake is
turbulent.
At Re ~ 400, a further instability occurs [66] , this time elose to the cylinder
in the region where the fluid is moving away from it to form the eddies. As a resuIt
the vortices in the street are turbulent. As noted above, the instability has,
curiously, a stabilizing effect on the Strouhal number. Although the motion
within the vortices involves rapid irregular fluctuations the vortex shedding occurs
with quite precise periodicity. However, as the vortices travel downstream the
turbulence spreads into the regions between them and disrupts the regular
periodicity, again leading finally to a fully turbulent wake.
This behaviour continues over a wide Reynolds number range, up to about
3 x 10 5 . There are sub-ranges over which the detaiis of the instability immediately
behind the cylinder are different, but we leave these complications aside [66, 113,
114] . At Re ~ 3 x 10 5 , a much more dramatic development occurs. To understand
this we must first consider developments at lower Re at the front and sides of the
cylinder.
There the phenomenon known as boundary layer formation occurs. This will be
the subject of a full discussion later in the book (Section 8.3 and Chapter 11). For
the moment, we may just say that the re is a region, called the boundary layer, next
to the wall of the cylinder in which all the changes to the detailed flow pattem
occur. Outside this the flow pattem is independent of the Reynolds number. For
these statements to mean anything, the boundary layer must be thin compared with
the diameter of the cylinder; this is the case when Re is greater than about 100.
(Boundary layer formation does not start by asudden transition of the sort we
have been considering previously; it is an asymptotic condition approached at high
enough Re. The figure of 100 is just an order of magnitude.)
Flow Past a Circular Cylinder 27

Laminar separation
a9
/
\ Turbulent separation
Turbulent reattachment
Turbulent separation

Larninar separation

Re < 3 x 10' 3 x 1 0' < Re < 3 x 10· Re > 3xl0·

Figure 3.12 Separation positions for various Reynolds number ranges (Refs. [209,248).

The change in the flow at Re ~ 3 x 10 5 results from developments in the

boundary layer. Below this Reynolds number the motion there is laminar. Above it
undergoes transition to turbulence. At first, this transition takes a rather compli-
cated form [209, 248] : laminar fluid elose to the wall moves away from it as if it
were entering the attached eddies; transition then occurs very quickly and the
turbulent flow reattaches to the wallonly a small distance downstream from the
laminar separation. (See Section 11.4 for more precise consideration of these
processes.) There are also complications due to the facts that the transition can
occur asymmetrically between the two sides of the cylinder [58] and non-
uniformly along its length [131].
At higher values of the Reynolds number, above about 3 x 10 6 , transition occurs
in the boundary layer itself, thus eliminating the laminar separation and turbulent
reattachment. The transition process is now similar to that described for pipe flow.
Whether or not it has previously undergone laminar separation and turbulent
reattachment, the turbulent boundary layer itself separates; the fluid in it moves
away from the wall of the cylinder and into the wake some distance before the
rear line of symmetry. This occurs, however, much further round the cylinder than
when the boundary layer remains laminar (Fig. 3.12). As a resuit the wake is
narrower for Re > 3 x 10 5 than for Re < 3 x 10 5 . When Re > 3 x 10 5 , the fluid
entering the wake is already turbulent and so the transition process immediately
behind the cylinder is eliminated.

3.4 Drag

An important quantity associated with the relative motion between a bodyand a

fluid is the force produced on the body. One has to apply a force in order to move
a body at constant speed through a stationary fluid. Correspondingly an obstaele
place d in amoving fluid would be carried away with the flow if no force were
applied to hold it in place. The force in the flow direction exerted by the fluid on
an obstaele is known as the drag. There is an equal and opposite force exerted by
the obstacle on the fluid .
At high Reynolds numbers, this can be thought of as the physical mechanism of
wake formation. Because of the force between it and the obstaele, momentum is
28 Physical Fluid Dynamics

o
Figure 3.13 Wake produetion: sehematie velocity profiles upstream and downstream of an
obstac1e.

removed from the fluid. The rate of momentum transport downstream must be
smaller behind the obstac1e than in front of it. There must thus be a reduetion in
the veloeity in the wake region (Fig. 3.13). (At low Reynolds numbers, when the
veloeity and pressure are modified to large distanees on either side of the obstac1e,
the situation is more eomplex.)
For the eireular eylinder, the important quantity is the drag per unit length; we
denote this by D. To show the variation with Reynolds number we need a non-
dimensional representation; this is provided by the quantity known as the drag
eoeffieient, defined to be
D
CD = IhpuÕ d (3.3)

(The faetor Ih has, of eourse, no role in the non-dimensionalization, but is

eonventionally inc1uded beeause Ihpu~ has a eertain physieal signifieanee - see
Seetion 10.7.)
Figure 3.14 shows the variation of the drag eoeffieient with Reynolds number.
The use of the non-dimensional parameters enables all eonditions to be covered
by a single eurve. The eurve is based primarily on experimental measurements, too
numerous to show individual points. At the low Reynolds number end, the experi-
ments ean be matehed to theory.
The eorresponding plot between Reynolds number and drag eoeffieient ean also
be given for a sphere, although with less precision beeause of the experimental
problem of supporting the sphere. The eurve, although different in detail, shows all
the same prineipal features as the one for the eylinder.
A few features of the euryes merit eomment - some of these being points to
whieh we shall retum in later ehapters.
At low Reynolds numbers,
1
C cx- (3.4)
D Re
(This is aetually an aeeurate representation for the sphere, an approximate one for
the eylinder - Seetions 9.4 and 9.5.) For a given body in a given fluid (flXed d, p,
and f.J.) this eorresponds to
D cxuo (3.5)
 F/ow Past a Orcular Cylinder 29
1o·r-__--__,-______.-______.-______,-____-,,-____-,,-____-,______-,

10

10 - '
L
l0~
-·----~------~
10~------
10L,~-----1~
0-3------1~0-·-------
10L·-------1~0-6----~1~0 '
Re

Figure 3.14 Variation of drag coefficient with Reynolds number for circuIar cylinder. Curve
is experirnental, based on data in Refs. [94,105,209,250,274,292].

Direet proportionality of the drag to the speed is a characteristic behaviour at low

speeds.
Over a wide range of Reynolds number (10 2 to 3 x 105) the drag co~fficient
varies little. For a given body in a given fluid, constant CD corresponds to
D cx: u~ (3.6)
Proportionality of the drag to the square of the speed is often a characteristic
behaviour at high speeds.
However, there is a dramatic departure from this behaviour at Re ~ 3 x 105 • The
drag coefficient drops by a factor of over 3. This drop occurs sufficiently
rapidly that there is actually a range over which an increase in speed produces a
decrease in drag. We have seen that this Reynolds number corresponds to the onset
of turbulence in the boundary layer. The consequent delayed separation of the
boundary layer results in a narrower wake. Since this in turn corresponds to less
momentum extraction from the flow, one might expect the lower drag. The result is
nonetheless a somewhat paradoxical one since transition to turbulence usually
prdduces an increased drag. In fact, on the cylinder and sphere, the force exerted
directly by the boundary layer does increase on transition. But this is more than
counteracted by another effeet - changes in the pressure distribution over the
surface (see Section 11.5).
4
Convection in Horizontal Layers

4.1 The configuration

Our third example of fluid motion in a simple configuration is taken from a

different branch of fluid mechanies, free conveetion. The cause of motion is the
action of a gravitational field on the density variations associated with temperature
variations - essentially the well-known fact that hot fluid rises. A systematie
discussion of free convection will be given in Chapter 14. Here we discuss the
phenomena to be observed in one particular arrangement [72, 145] .
A layer of fluid (Fig. 4.1) occupies the space between two horilOntal plates, a
vertieal distance d apart. Each plate is maintained at a steady uniform temperature,
the lower one hotter than the upper one (T2 > T 1 ). The walls limiting the extent of
the layer in the horilOntal direction are supposed to be so far apart that they have
little influence on the motion in most of the layer.

T,

:::::::::::::;':::::::::(~:::I 19
Figure 4.1 Definition sketch for Benard conveetion.

This is known as the Benard configuration. Historieally, the name is inaccurate;

Benard's pioneering observations, although for long believed to relate to this con-
figuration, were actually mostly of another phenomenon (Section 17.4) that gives
rise to similar effects. However, the name is so weil established that its use in this
way causes no confusion.
Benard convection has been extensively investigated both experimentally and
theoretically. The theoretieal work has provided, for example, the most accurate
information about when flow occurs (Section 4.2). On the other hand, most of our
knowledge about the detailed flow patterns (the sequence of events to be sum-
marized in Section 4.3) comes from experiment. Such experiments have been
carried out with a variety of fluids, as will be illustrated by the figures in this
chapter. Considerable ingenuity has been needed to overcome the practical diffi-
culties of approximating well to the theoretieal configuration whilst having a way of
observing the flow patterns.
Conveetian in Harizantal Layers 31

4.2 Onset of motion

The Benard configuration differs from the situations we have been considering
hitherto in that one cannot im media te ly say whether flow will oecur. Fluid at rest
in a layer of the so rt shown in Fig. 4.1 is in equilibrium. However, heavy cold fluid
is situated above light hot fluid; if the former moves downwards and the latter
upwards, there is a release of potential energy whieh ean provide kinetie energy for
the motion. There is thus a possibility that the equilibrium will be unstable. Flow
thus arises not because of the absenee of equilibrium (Le. the absenee of any
solution of the governing equations with the fluid at rest) but beeause the equi-
librium is unstable.
Thedistinetion may be clarified by contrasting the Benard situation with the
flow in a similar fluid layer between vertieal plates (to be considered more fully in
Seetion 14.7). In this the fluid next to the hot wall rises and that next to the cold
wall faUs (Fig. 4.2) because no balanee between the gravitational and pressure forees
exists with the fluid at rest.
In that ease, motion always oeeurs. In Benard conveetion, it oeeurs only when
eertain criteria are fulfilled. An obvious first criterion is that the temperature should
decrease upwards,
(4.1)
so that heavier fluid overlies lighter.t But this is not the only criterion. The instabi-
lity is opposed by the frictional action of viscosity. It is also opposed by the action
of thermal conduetivity, which tends to remove the temperature differenee between
the hat rising regions and the cold falling regions. Motion occurs only when the
destabilizing action of the temperature differenee is strong enough to overeome
these.
This depends on a quantity known as the Rayleigh number, whieh plays a role
for the present problem similar to that played by the Reynolds number in the
previous problems. The Rayleigh number is
Ra = gcx(T2 - T 1 )d3
(4.2)
VI(

T, T,

l.
Figure 4.2 Conveetion in a vertieal slot.

tThis assumes that the coeffieient of expansion is positive. In the few cases where it is negative,
sueh as water below 4°C, the eriterion (4.1) is of eourse reversed.
32 Physical Fluid Dynamics

//1 /1/////1/11////1//1

0000000
///1/17 ////1//7/1///17
Figure 4.3 Schematic representation of Benard convection at a Rayleigh number a little above
critical.

where d is the depth of the layer, (T2 - T 1 ) is the temperature differenee aeross it,
g is the acceleration due to gravity, and a, v, and K are properties of the fluid,
respectively its eoefficient of expansion, kinematic viscosity, and thermaI diffusi-
vity.t The role of the Rayleigh number will be diseussed in Seetion 14.2 and 18.2;
however we may notice here that the factors that drive the motion - the combina-
tion of the temperature differenee, the consequent expansion, and gravity - appear
in the numerator and the factors mentioned above as opposing it appear in the
denominator.
Instability oceurs when the Rayleigh number exeeeds a critical value of about
1700 (see Section 18.2). Below this the fluid remains at rest. Above it comes into
motion. The fluid establishes hot rising regions and cold falling regions with
horizontal motion at top and bottom to maintain continuity, as shown schemati-
cally in Fig. 4.3. The rising fluid loses its heat by thermal eonduetion when it gets
near to the cold top wall and can thus move downwards again. Similarly, the cold
downgoing fluid is warmed in the vicinity of the hot bottom wall and can rise again.
When the flow is established as a steady pattem, the eontinuous release of potential
energy is balaneed by viseous dissipation of mechanieal energy. The potential
energy is provided by the heating at the bottom and the eooling at the top. From a
thermodynamic point of view the system is thus a heat engine (ef. Seetion 14.4).

1.5
Nu
1.4

1.3

1.2

1. 1

1.0
•
•• • •
. .'

0.9
0 400 800 1200 1600 2000 2400 2800
Ra

Figure 4.4 Variation of Nusseit number with Rayleigh number for Benard conveetion, show-
ing increase of heat transfer with onset of motion. Data obtained by C. W. Titman.

tThermal diffusivity is defined as K = k/pCp (k = thermal conductivity; p = density;

Cp = speeific heat at constant pressure).
Gmveetion in Horizontal Layers 33

The onset of motion as the Rayleigh number is increased is most readily detected
by its effect on the heat transfer. When the fluid is at rest, this is due to conduction
(radiative heat transfer is frequently negligible, though it sometimes has to be sub-
tracted out, particularly in experiments with gases); hence, if H is the rate of heat
transfer per unit area

(4.3)

When the fluid comes into motion it carries heat with it and the heat transfer in-
creases above that given by conduction alone. Figure 4.4 shows the observed varia-
tion of the quantity
Hd
~=~~-~) ~~
with Rayleigh number. (Nu is known as the Nusselt number - see Sections 13.4 and
14.7.) Nu remains equal to 1 below the critical Rayleigh number, but increases
above 1 with the onset of motion.

4.3 Flow regimes

The detaile d stmcture of the convection and, in particular, the planform of the
rising and falling regions is a matter of some complexity. For values of the Rayleigh
number not too much in exceSs of the critical value, the convection always occurs
in a fairly regular pattem, as was implied by Fig. 4.3. (The individual elements in
such a pattem are known as convection eelis - or Benard eelIs - and the flow as a
whole is described as cellular.) The geometry of this pattem can, however, take a
variety of forms , as can be seen in Figs. 4.5 to 4.7. The matter is complicated by the
fact that the variation of fluid properties (particularly viscosity) with temperature
influences the pattem. When this variation is negligible the eonveetion first oeeurs
in long 'rolls' of rising and falling fluid (as in Figs. 4.5 and 4.6). Because these rolls
are long, the whole pattem is sensitive to the shape of the vertical walls (e.g., Fig.
4.6).

Figure 4.5 Benard convection at Ra = 3 x 10 3 , Pr = 367; aluminium powder flow visualiza-

tion (see Section 23.4). From Ref. (236) .
34 Physical Fluid Dynamics

Figure 4.6 Benard convection at Rayleigh number slightly above critical, Pr "" 800;
aluminium powder flow visualization (Section 23.4). From Ref. [143).

As the Rayleigh number is increased, a sequence of changes in the mode of con-

vection occurs. The values of the Rayleigh number corresponding to these changes
depend on a further parameter, the Prandtl number
Pr=v/K (4.5)
This is a property of the particular fluid. Even when this is allowed for, there are
considerable variations in the quantitative observations, associated with sensitivity
to disturbanees and hysteresis (transitions occurring at different Rayleigh number
when this is inereasing and when it is deereasing). Henee, no quantitative details will
be provided in the following aceount,t although values of the Rayleigh and Prandtl

Figure 4.7 Benard convection at Ra = 2.6 x 104 , Pr = 100; aluminium powder flow visualiza-
tion (Section 23.4). From Ref. [210).

tThe account is synthesized from observations in a variety of fluids, and seems to be the most
consistent description that can be given at the present time. However, this means that not all
parts of the account necessarily apply at every Prandtl number. Also, aimost certainly, future
work will reveal additional significant stages of development.
Conveetion in Horizontal Layers 35

Figure 4.8 Detailed view of eell in Benard conveetion, Ra = 1. 76 X 10' ; aluminium powder
flow visualization (Seetion 23.4). From Ref. (159).

numbers in the figure captions give some indications. In general, corresponding

developments tend to occur at a lower value of the Rayleigh number when the
Prandtl number is lower. As an extreme example, in mercury , with Pr = 0.025, it is
difficult to observe any form of convection other than turbulent.
The first development is a change in the planform of cellular convection from
the rolls shown in Figs. 4.5 and 4.6 to a three-dimensional pattem such as that in
Fig. 4.7 . The flow remains steady. Quite a variety of three-dimensional pattems has
been observed ; again , probably, temperature variation of viscosity has influenced
some of these. Figure 4.8 shows a elose-up of one convection cell in a three-
dimensional array. The fluid rises in isolated regions at the centre of polygons like
that shown and falls in a continuous network formed by the peripheries of the
polygons. The horizontal flow elose to the top produces the streaks seen in the
photograph.
Further developments involve the motion becoming unsteady. Figure 4.9 shows
observations of the temperature variations. At each Rayleigh number, one trace
shows the variation with time at a fixed position, the other the variation with posi-
tion observed by traversing a probe through the fluid. At low Rayleigh numbers,
the temperature at a fixed position is effectively constant, whilst the traverse
produces a roughly periodic variation associated with the passage of the probe
through the eelIs. At higher Rayleigh numbers unsteadiness is apparent. At first, the
time variations are a perturbation of the previous pattem, but with increasing Ra,
they become increasingly dominant.
Other experiments have elucidated some of the processes involved. The first un-
steadiness is in the form of periodic pulsations within the celIular flow. There is some
dispute as to the detailed form of the pulsations [150,294] - perhaps the re is more
36 Physical Fluid Dynamics

·0·4 ~1\
~V~/\j\C\ A.~A ",/,\/'\. /'\ " /\

(a)

_.:o~L-=~~---::;::::;==-====--=====-----=---===">. "",. ."."

3 6 ~ ,
12

(b)

3:~~
o I 2 3 4

'~ /JA. ~ I\. 1\

''{~YVV 0I\[Jr~
o Il /1 f\ MA 1'1

o 10 20 30 40 !l0 60
(e)

than one mode of oscillation - but their occurrence is weIl illustrated by Fig. 4.10.
This picture was obtained by having the fluid illuminated along onlyone line and
moving the camera so that the image of this line moved across the film. The result is
essentially a space-time representation of the flow, and the 'ribbed' pattem indi-
cates the occurrence of periodic fluctuations.
Further increase in Rayleigh number results in the periodic variations giving way
to much less regular ones. Convection cells can stili be observed (Fig. 4.l1),
although with greater variability in size and shape. The flow within the cells,
however, is turbulent. At Rayleigh numbers weIl above critical, only fluid passing
quite elose to the lower boundary is significantly heated. The resulting hot rising
regions are thus narrow and so, owing to a simHar effect at the upper boundary, are
the cold falling regions (Fig. 4.l2). Much of the fluid circulates approximately
isothermaIly. This is an example ofwhat, later in the book, we shall come to know
Conveetion in Horizontal Layers 37

(d)

. 2~
0 - - --

-.2~
0
~----------~1------------~2------- ~

(e l

Figure 4.9 Temperature variations in Benard eonveetion observed by moving and stationary
probes. (a) Ra = 5.0 x 10 3 , (b) 1.21 x 10 4 , (e) 3.03 x 10 4 , (d) 5.61 x 10 4 , (e) 1.55 x 10 6 ;
Pr = 0.72 (air) throughout. Abscissa seales represent x/d and time (minutes) for the two cases;
ordinate seales represent the differenee between the instantaneous temperature and the mean
temperature divided by (T, - T,). From Ref. (293).

as boundary layer development. The generation of the turbulenee within the

eonveetion eelis oeeurs in the narrow hot and cold regions. Beeause of the
turbulenee the eellular strueture is no longer readily apparent from observations of
the temperature field (Fig. [Link]).
At still higher Rayleigh number the eellular strueture disappears altogether. One
is now in the regime deseribed as turbulent conveetion. To understand the processes
oeeurring in this regime , it is helpful to look at the mean temperature distribution
aeross the layer. (Averaging is either over time or over horilOntal planes. The nature
and role of mean quantities in turbulent flows will be diseussed in Seetion 20.3.)
This takes the form shown sehematieally in Fig. 4.13, with large temperature
gradients elose to the boundaries and a nearly isothermal region in the interior, for
the following reason. In the interior the fluid motion transports a large amount of
heat ac ross the layer (Le. Nu ~ I), a process requiring negligiblevertieal temperature
38 Physical Fluid Dvnamics

17

o 10 20 30 40 50
POSITION (CM .)

Figure 4.10 Space-time flow visualization (see text) of Benard convection at Ra = 3.4 x 10',
PJ.= 57. From Ref. [149).

gradient. The boundaries constrain the motion so that it cannot provide the sam e
heat transport elose to them; large temperature gradients are needed there so that
the heat is transported by conduction. Throughout the sequence of developments
described above, the temperature profile tends away from the linear form that
occurs when the fluid is at rest and towards the form of Fig. 4.13. By the time the
motion is turbulent, the large gradient regions occupy only small parts of the layer.
In consequence, the most strongly unstable regions are those elose to the
boundaries and the dominant features of the flow originate there. We consider the
hot bottom wall; similar processes occur elose to the cold :,op wall, with the roles of
hot and cold fluid reversed. There are intermittent eruptions of hot fluid away from

Figure4.11 Counterpart of Fig. 4.5 at Ra = 1.3 x 10', Pr = 311. From Ref. [236).
Conveetion in Hon·zontal Layers 39

) e )

Turbulent Isothermal Turbulent

hat region circulation cold region

Figure 4.12 Principal features of eeliular eonveetion in boundary layer regime.

T
..

Figure 4.13 General form of mean temperature distribution for Benard eonveetion at high
Rayleigh number.

Figure 4.14 Thermals rising from a heated horizontal surface in water (Pr ~ 6); dye
produeed uniformly at surfaee by pH technique (see Seetion 23.4). From Ref. [237J.
40 Physical Fluid Dynamics

~.
. m"-
... -
. +~ ,.' I
.

Figure 4.15 Thermals rising from a heated horizontal surfaee in air (Pr = 0.7); flow visualiza-
tion using reeondensed water vapour evaporated at moist surfaee. From Ref. [206].

the bottom boundary. Colder fluid moves elose to the wall to re place the fluid in an
eruption. This is gradually warmed by conduction from the wall unt il there is a
thick enough layer of hot fluid for another eruption to be initiated. The process
then repeats itself.
Each eruption gives rise to one or more columns of hot fluid rising through the
interior region. These features are sometimes called thermals, sometimes plumes.
Over a range of Rayleigh number, the thermals penetrate right across the layer,
generating transient stable blobs of fluid elose to the opposite boundary [77].
Ultimately, however, at the highest Rayleigh numbers, there is such vigorous
mixing associated with the turbulent motion that the thermais lose their identity
before reaching the opposite boundary. The processes at each boundary now occur
independently of those at the other. They can be investigated by studying the
motion above a single heated plate at the bottom of a large expanse of fluid
(Figs. 4.14-4.16), corresponding in a sense to infinite Rayleigh number.
Two different types of motion have been observed, the relationship between
them being unelear. Both involve the above process of the generation of thermals.
However, in some cases, illustrated by Fig. 4.14, the eruptions occur with marked
periodicity and the hot rising regions are in the sam e places each time. In the more
usual case, illustrated by Fig. 4.15, the eruptions are not simultaneous over different
parts of the plate. At any instant they are randomIy distributed in position (Fig.
4.16 shows the random character of the instantaneous temperature fieId); at a fixed
Conveetion in Horizontal Layers 41

Figure 4.16 Shadowgraph (see Section 23.4) of convection around a heated horizonta! plate.
Dotted line shows position of shadow of plate in the absence of therrnal effects. From Ref.
[220].

position they oeeur randomly in time. Only in the latter ease ean the motion really
be ealled fully turbulent. However, almost eertainly, both types of flow involve
vigorous disorganized small-seale motion as well as the rather eoherent large-seale
motion of the thermals. The former oeeurs both within the thermals and in the
slowly downward drifting cold fluid between the thermaIs f2701.
5
Equations of Motion

5.1 Introduction

We turn now to the formulation of the basic equations - the starting point of the
theories that willlead, one hopes, to an understanding of phenomena such as those
described in Chapters 2 to 4. These equations are formulations appropriate to a
fluid in motion of the usuallaws of mechanics - conservation of mass and Newton's
laws of motion. In some situations other physical processes may be present, thermo-
dynamic processes for instance, and equations for these are similarly formulated.

5.2 Fluid partides and continuum mechanics

Before we can proceed with this formulation we need certain preliminary ideas, the
most important being the concept of a fluid partide.
The equations concern physieal and mechanieal quantities, such as velocity,
density, pressure, temperature, which will be supposed to vary continuously from
point to point throughout the fluid. How do we de fine these quantities at a point?
To do so we have to make what is known as the assumption of the applicability of
continuum mechanics or the continuum hypothesis. We suppose that we can
associate with any volume of fluid, no matter how small, those macroscopic proper-
ties that we associate with the fluid in bulk. We can then say that at each point there
is a partide of fluid and that a large volume of fluid consists of a continuous aggre-
gate of such partides, each having a certain velocity, temperature, etc.
Now we know that this assumption is not correct if we go right down to molecu-
lar scales. We have to consider why it is nonetheless plausible to formulate the
equations on the basis of the continuum hypothesis. It is simplest to think of a gas,
aIthough the considerations for a liquid are very similar.
The various macroscopic properties are defined by averaging over a large number
of molecules. Consider velocity for example. The molecules of a gas have hight
speeds associated with their Brownian motion, but these do not resuIt in a bulk
transfer of gas from one place to another. The flow velocity is thus defined as the
average velocity of many molecules. Similarly, the temperature is defined by the
average energy of the Brownian motion. The density is defined by the mass of the
average number of molecules to be found in a given volume. Other macroscopic

tHigh in the sense that, for the flows at speeds low compared with the speed of sound that are
considered in this book, the r.m.s. Brownian speed is much larger than the flow speed.
Equations of Motion 43

L, L,
Len9th sca le. L

Figure 5.1 Schematic variation of average energy of molecules with length seale. See text.

properties, such as pressure and viscosity, likewise resuIt from the average action of
many molecules.
None of these averaging processes is meaningful unIess the averaging is carried out
over a large number of molecules. A fluid partide must thus be large enough to
contain many molecules. It must stilI be effectively at a point with respeet to the
flow as a whole. Thus the continuum hypothesis can be valid only if there is a
length scale, L 2 , which we can think of as the size of a fluid partide, such that
(5.1)
where the meanings of LI and L 3 are illustrated by Fig. 5.1 . This figure uses the
example of temperature, rather than the natural first choice of velocity, because it
is easier to discuss a scalar. It shows schematically the average Brownian energy of
the molecules in avolume L 3 plotted against the length scale L (on a logarithmic
seale). The centre of the volume may be supposed fixed as its size is varied. When
the volume is so small that it contains only a few molecules, there are large random
fluctuations; the change produced by increasing the volume depends on the particu-
lar speeds of the new molecules the n induded. As the volume becomes large enough
to contain many molecules, the fluctuations become negligibly small. A temperature
can then meaningfully be defined. Liis proportional to, but an order of magnitude
or so larger than, the average distance from a molecule to its nearest neighbour.
At the other extreme, the volume may become so large that it extends into regions
where the temperature is significandy different. This will resuIt in an increase or
decrease in the average. L 3 is a typicallength scale associated with the flow; that is a
typical distance over which the macroscopic properties vary appreciably.
The applicability of the continuum hypothesis depends on the re being a signifi-
cant plateau between LI and L 3 as shown. One may regard L 2 as being an infini-
tesimal distance so far as macroscopic effects are concerned, and formulate the
equations (as differential equations implicitly involving the limit of small sepa ra-
tions) ignoring the behaviour on still smaller length seales.
44 Physical Fluid Dynamics

The same fluid partide does not consist of just the same molecules at all times.
The interchange of molecules between fluid partides is taken into account in the
macroscopic equations by assigning to the fluid diffusive properties such as viscosity
and thermaI conductivity. For example (again considering a gas for simplicity) the
physical process by which the velocity distribution of Fig. 1.1 generates the stress
shown is the Brownian movement of molecules across AB; those crossing in, say, the
+y-direction have on average less x-momentum and so tend to reduce the momentum
of the fluid above AB. The same fluid partide may be identified at different times -
once the continuum hypothesis is accepted - through the macroscopic formulation.
This specifies (in principle) a trajectory for every partide and thus provides meaning
to the statement that the fluid at one point at one time is the same as that at another
point at another time. For example, for a fluid macroscopically at rest, it is
obviously sensible to say that the same fluid partide is always in the same place -
even though, because of the Brownian motion, the same molecules wilI not always
be at that place.
However, for the continuum hypothesis to be plausible, it is evidently necessary
for the molecules within a fluid partide to be strongly interacting with one another.
If each molecule acted just as if the others were not there, there would be little
point in identifying the aggregate as a partide. Thus, if A. is the molecular mean free
path, continuum mechanics can be applied only if
(5.2)
so that each molecule undergoes many collisions whilst traversing a distance that
can stilI be regarded as infinitesimaI. Since A. can be large compared with LI as
defined above, this is an additional requirement to (5.1).
Once the continuum hypothesis has been introduced, we can formu1ate the
equations of motion on a continuum basis, and the molecular stmcture of the fluid
need not be mentioned any more. Hence, although the concepts developed above
underiie the whole formulation, we shall not have much occasion to refer back to
them. Velocity, henceforth, will be either a mathematical quantity or something
(hopefully equivalent) that one measures experimentally. So will all the other para-
meters. Their definitions as averages over molecules provide answers to the implicit,
but rarely explicit, questions: 'What is the real physical meaning of this mathematical
quantity?' 'What quantity does one ideally wish to measure?'
The continuum hypothesis is only a hypothesis. The above discussion suggests
that it is plausible, but nothing more. The real justification for it comes sub-
sequently - through the experimental verification of predictions of the equations
developed on the basis of the hypothesis.

5.3 Eulerian and Lagrangian co-ordinates

In setting up the equations governing the dynamics of a fluid partide, we evidently

need to decide whether we should use co-ordinates fixed in space or co-ordinates
that move with the partide. These two procedures are known respectively as the
Eulerian and Lagrangian specifications. The equations are much more readily for-
mulated using the former because the Lagrangian specification does not immediately
indicate the instantaneous velocity field on which depend the stresses acting between
fluid partides. Throughout this book we use only the Eulerian specification; Le.
Equations of Mo tio n 45

we write the velocity

u = u(r, t) (5.3)
where r is the position in an inertial frame of reference and t is time. Values of u
at the same r but different t do not, of course, correspond to the same fluid partide.
It is not always easy to relate Lagrangian aspects, such as the trajectories of fluid
partides, to an Eulerian specification. In the context of this book, this is particu-
larly relevant to the interpretation of flow visualization experiments in which dye
marks certain fluid partides. The relationship of the observed patterns to the
corresponding Eulerian velocity field may not be simple.

5.4 Continuity equation

We are now ready to start on the actual formulation of the equations. We consider
first the representation of mass conservation, often called continuity.
Consider an arbitrary volume V fixed relative to Eulerian co-ordinates and
entirely within the fluid (Fig. 5.2). Fluid moves into or out of this volume at points
over its surface. If dS is an element of the surface (the magnitude of dS being the
area of the element and its direction the outward normal) and u is the velocity at
the position of this element, it is the component of u paralleI to dS that transfers
fluid out of V. Thus, the outward mass flux (mass flow per unit time) through the
element is pu·dS, where p is the fluid density. Hence,

f
Rate of loss of mass from V = S pu· dS (5.4)

(This is, of course, negative ifthe mass in V is increasing.) We have also

Total mass in volume V = f v pd V (5.5)

Hence

i. d V = - f pu • dS
r pd V =fvapa (5.6)
dt lv ts
We are interested in the mass balance at a point, rather than that over an arbitrary
finite volume. Hence, we allow V to shrink to an infinitesimal volume; the integra-
tion in !(ap/at)d Vis redundant and we have
ap
(5.7)
-
at = -lim
v-+o
[!pu·dS/V]

dS

Figure 5.2 Definition sketch for derivation of continuity equation.

46 Physical Fluid Dynamics

That is
ap .
- = -dlV pu (5.8)
at
by definition of the operator div. This gives the general expression representing
mass conservation for a fluid in which both u and p are functions of position,
ap
- + v'. (pu) = 0 (5.9)
at
This is known as the continuity equation.
For the important special case when p is constant - see Section 1.2 (iii) and
Section 5.8 - the continuity equation reduces to the very simple form
v"u=O (5.10)
The velocity field is a solenoidal vector. We notice that this does not assume steady
flow. The time-variation does not appear expHcitly in the continuity equation of a
constant density fluid even when the flow is unsteady.

5.5 The substantive derivative

The next equation to be derived is the representation of Newton's second law of

motion; Le. the rate of change of momentum of a fluid partic1e is equal to the
net force acting on it. We need first of all an expression for the rat e of change of
momentum of a fluid partic1e. It would not be correct to equate the rate of change
of momentum at a fixed point to the foree, because different partic1es are there at
different times. Even in steady flow, for example, a fluid partic1e can change its
momentum by travelling to a place where the velocity is different; this acceleration
requires a force to produce it.
This is one example of a general problem. On occasion, one needs to know the
rat e of change of other quantities whilst following a fluid partic1e. In problems
where thermal effects are important, for example, various physical processes may
heat or cool the fluid. These determine the rate of change of the temperature of a
fluid partic1e, not the rate of change at a fixed point.
In this section, therefore, we examine the general question of rates of change
following the fluid. It is easier in the first place to consider a scalar quantity, and
we denote this by T, thinking of the example of temperature.
Quite generally, the small change oTproduced by a small change ot in time
and small changes OX, oy, OZ in Cartesian position co-ordinates is
aT aT aT aT
oT=-ot+-ox+-oy+-oz (5.11)
at ax ~ ~

and a rate of change can be formulated by dividing by ot

oT aT aT ox aT oy aT oz
-=-+- -+- -+-- (5.12)
ot at ax ot ay ot az ot
Ifnow we choose ox, oy, and oz to be the components of the small distanee moved
by a fluid partic1e in time ot, then (in the Hmit of -+ 0) this is the rate of change of
Equations of Marion 47

T of that particle. Also ox/ot, oy/M and oz/ot are then (in the same limit) the three
eomponents of the veloeity of the particle (u, v, and w). We thus have
DT aT aT aT aT
-=-+u-+v-+w- (5.13)
Dt at ax ay az
In general D/Dt denotes the rate of ehange (of whatever quantity it operates on)
following the fluid. This operator is known as the substantive derivative.
We ean re-write equation (5.13) as
DT aT
-=-+u'V'T (5.14)
Dt at
and the operator in general as
D
-=-+U'V
a (5.15)
Dt at
(exhibiting the physically obvious faet that it does not depend on the partieular
eo-ordinates used).
We see that the relationship eombines the two ways in whieh the temperature of
a fluid particle ean ehange. It ean ehange beeause the whole temperature field is
changing - a process present even if the particle is at rest. And it ean ehange by
moving to a position where the temperature is different - a process present even if
the temperature field as a whole is steady. As one would expeet, this latter process
depends on the magnitude of the spatial variations of the temperature and on the
velocity, determining how quiekly the fluid moves through the spatial variations.
Nothing in the above analysis restricts it to sealar quantities, and we ean similarly
write that the rate of ehange of a veetor quantity B following a fluid particle is
DB aB aB aB aB aB
-=-+u-+v-+w-=-+u'VB (5.16)
Dt at ax ay az at
Whereas U'V Tis the sealar produet ofveetors u and VT, u'VB eannot be
similarly split up. It is meaningful only as a whole. (u . V) operating on a veetor
must be thought of as a new operator (defined through its Cartesian expansion).
The partieular ease, B = u, gives the rate of ehange of veloeity following a fluid
particle,
Du au
- =- + U' Vu (5.17)
Dt at
u now enters in two ways, both as the quantity that ehanges as the fluid moves
and as the quantity that governs how fast the ehange oeeurs. Mathematically, how-
ever, it is just the same quantity in both its roles.
Returning finally to the information that we require for the dynamical equation,
we have that the rate of ehange of momentum per unit volume following the fluid
is
Du au
p -=p - + pU'Vu (5.18)
Dt at
Why, the reader mayask, is it pDu/Dt and not D(pu)/Dt? (The distinetion is
48 Physical Fluid Dynamics

important in the general case when both u and p are variabIes.) The only reason
why a particular bit of fluid is changing its momentum is that it is changing its
velocity. If it is simultaneously changing its density, this is not because it is gaining
or losing mass, but because it is changing the volume it oecupies. This change is
therHore irrelevant to the momentum change. Expressing the distinction verbally
instead of algebraically, we may say that '(the rate of change of momentum) per
unit volume' is different from 'the rate of change of (momentum per unit volume)',
and the former is the relevant quantity.

5.6 The Navier-Stokes equation

From above, the left-hand side of the dynamical equation, representing Newton's
second law of motion, is pDu/Dt. The right-hand side is the sum of the forees (per
unit volume) acting on the fluid partic1e. We now consider the nature of these
forees in order to complete the equation.
Some forees are imposed on the fluid externally, and are part of the specification
of the particular problem. One may need, for example, to specify the gravity field
in which the flow is occurring. On the other hand, the forees due to the pressure
and to viscous action - of which simple examples have been given in Section 2.2 -
are related to the velocity field. They are thus intrinsic parts of the equations of
motion and have to be considered here. Both the pressure and viscous action
generate stresses acting across any arbitrary surface within the fluid; the force on a
fluid partic1e is the net effect of the stresses over its surface.
The generalization of the pressure force from the simple case considered in
Section 2.2 is straightforward. We saw there that the net foree per unit volume in
the x-direction as a resuIt of a pressure change in that direction is -ap/ax. For a
general pressure field, similar effects act in all directions and the total force per
unit volume is -grad p. This te rm always appears in the dynamical equation; when
a fluid is brought into motion, the pressure field is changed from that existing when
it is at rest (the hydrostatie pressure). We ean regard this for the moment as an
experimental re sult. We shall be seeing that it is neeessary to have the pressure as a
variable in order that the number of variables matches the number of equations.
The general form of the viscous force is not so readily inferred from any simple
example. The mathematical formulation is outlined as an appendix to this chapter.
Here we shalllook at so me of the physical concepts underlying viscous action, and
then quote the expression for the viscous term in the dynamical equation that is
given by rigorous formulation of these coneepts. (Some further discussion of the
physical action of viscosity will be given in the con~ext of the particular example
of jet flow in Section 11.9.)
Viscous stresses oppose relative movements between neighbouring fluid partic1es.
Equivalently, they oppose the deformation of fluid partic1es. The difference
between these statements lies only in the way of verbalizing the rigorous mathe-
matieal coneepts, as is illustrated by Fig. 5.3. The change in shape of the initially
rectangular region is produced by the ends of one diagonal moving apart and the
ends of the other moving together. As the whole configuration is shrunk to an
infinitesimal one, it may be said either that the partic1e shown is deforming or that
partic1es on either side of AB are in relative motion. The rate of deformation
depends on the velocity gradients in the fluid. The consequence of this behaviour
Equations of Mo tio n 49

Figure 5.3 Deformation of reetangular element and relative motion of fluid on either side of
arbitrary line through the element.

is the generation of a stress (equal and opposite forees on the two sides) across a
surface such as AB; this stress depends on the properties of the fluid as well as on
the rate of deformation.
The stress can have any orientation relative to the surface across which it acts.
The special case, considered in Sections 1.2(iv) and 2.2, in which the stress is in the
plane of the surface is often thought of as the 'standard' case. We therefore look for
a moment at a simple situation in which viscous stresses normal to the surface
govern the behaviour. This is the falling column produced for example when a
viscous liquid is poured from a container (Fig. 5.4). In the absence of side-walls
transverse viscous stresses cannot be generated as they are in channel flow. The
reason the fluid at, say, P does not fall with an acceleration of g is the viseous
interaetion with the more slowly falling fluid at Q.
In the general case, the stress is a quantity with a magnitude and two directions-
the direction in which it acts and the normal to the surface - associated with it. It
is thus a second-order tensor. (The stresses acting ac ross surfaees of different orien-
tations through the same point are not, of course, independent of one another.)
The rate of deformation is also expressed by a second-order tensor - the rate-of-
strain tensor. From the eonsiderations above we expeet this to involve the velocity

Figure 5.4 Pouring of viseous liq uid.

50 Physical Fluid Dynamics

gradients. However, not all distributions of veloeity variation lead to deformation; a

counter-example is the rotation of a body of fluid as if it were rigid (see Section 6.4).
The rate-of-strain tensor selects the appropriate features of the veloeity field.
One expects the stress tensor to depend on the rate-of-strain tensor and on the
properties of the fluid. A Newtonian fluid (see Section 1.2(iv» can now be defined
rigorously as one in which the stress tensor and the rate-of-strain tensor are linearly
related.
The remaining ide as contained in the derivation of the viscous term of the
dynamicaI equation are simply symmetry considerations. For example, a mirror-
im age flow pattem must generate a mirror-image stress distribution. And the
analysis of a flow configuration using different co-ordinates must give the same
resulto
In Cartesian co-ordinates, the x-component of the viscous force per unit volume
(see appendix to this chapter) is

~ [2/1 au + AV"
ax. ax
u] + ~ay [/1 (auay + axav )] + ~az [/1 (awax + au)]
az
(5.19)

Similar expressions for the y- and z-components are given by appropriate permu-
tations.
/1 is the coeffieient of viscosity, defined through the speeial case considered in
Section 1.2(iv). Ais a second viscosity coeffieient. One would expect the re to be a
second such coeffieient, independent of the first, by analogy with the fact that
the re are two independent elastic moduli. This is a valid analogy. However, it has
often been the practice to introduce a relationship between /1 and A(A = -2/1/3).
This is done at the cost of redefining the pressure so that it is not the thermody-
namic pressure, and the second independent parameter then appears in the
relationship between the two pressures [23, 208] . A is difficult to measure experi-
mentallyand is not known for the variety of fluids for which there are values of /1.
Hence, the statement that a fluid is Newtonian usually means that /1 is observed to
be independent of the rate of strain and that Ais assumed to be so too.
However, for a fluid of constant density, the continuity equation (5.1 0) causes
the term involving A to drop out. If, additionally, /1 is taken to be a constant,
expression (5.19) reduces (with a further use of (5.10» to simply
(5.20)
The y- and z-components are correspondingly /1 V'2 v and /1 V'2 w, and the vectorial
viscous force per unit volume is /1 V'2 u.t
Because (5.20) is so much simpler than (5.19), one prefers to use the former
whenever possible; one tends to make this approximation even when the density
and/or the viscosity do vary appreeiably (see also appendix to Chapter 14). That
will be the procedure throughout this book. But it should be remembered that
situations may arise in the laboratory or in a practical application which are
properly described only by the full expression.
Collecting together the various contributions mentioned, we have
Du
p- = -V'P + /1\72 u + F (5.21 )
Dt

tThe meaning of v 2 operating on a veetor is defined through its eartesian expansion.

Equations of Motion 51

This is known as the Navier-Stokes equation. It is the basic dynamical equation

expressing Newton's second law of motion for a fluid of constant density.
The term F represents the contribution of those forces (such as gravity)
mentioned at the beginning of this section that have to be induded in the specifi-
cation of the problem. This is often known as the body force term, because such
forces act on the volume of a fluid partide, not over its surface in the way the
stresses between fluid partides act. The re action to a body force is remote from the
fluid partide concerned - usually outside the fluid region, although occasionally on
distant fluid partides.
We shall often be considering problems in which F =0, the cause of motion being
either imposed pressure differences or relative movement of boundaries. No body
forces are applied. It might be objected that, although one can well imagine all other
sources of body force being eliminated, almost every flow will take place in a gravity
field. It can be shown that gravitational body forces act significantly only on
density differences. If the density is uniform, the gravitational force is balanced by
a vertieal pressure gradient which is present whether or not the fluid is moving and
whieh does not interact with any flow. This hydrostatie balance can be subtracted
out of the dynamical equation and the problem reduced to one without body
forces. This assumes, of course, that the fluid region is supported at the bottom;
flow under gravity down a vertical pipe provides an obvious example where this is
not so and where the above remarks do not apply. We will consider the justification
for subtracting out the hydrostatic balance more mathematically in Section 13.2.
The Navier-Stokes equation may be re-written

au 1
- + U • Vu = - - Vp + vV
2
u
1
+- F (5.22)
at p p

v = /.1/ p and is a property of the fluid called the kinematic viscosi ty.
We notiee that the equation is a non-linear partial differential equation in u. The
non-linearity arises from the dual role of the velocity in determining the acceleration
of a fluid partide, as mentioned in Section 5.5. This non-linearity is responsible for
much of the mathematical difficulty of fluid dynamics, and is the principal reason
why our knowledge of the behaviour of fluids in motion is obtained in many cases
from observation (both of laboratory experiments and of natural phenomena)
rather than from theoretical prediction. The physieal counterpart of the mathe-
matieal difficulty is the variety and complexity of fluid dynamical phenomena;
without the non-linearity the range of these would be much more limited.
The continuity equation (5.10) and the Navier-Stokes equation (5.21) con-
stitute a pair of simultaneous partial differential equations. Both represent physieal
laws whieh will always apply to every fluid partide. Together they provide one
scalar equation and one vector equation - effectively four simultaneous equations -
for one scalar variable (the pressure) and one vector variable (the velocity) -
effectively four unknown quantities. The number of unknowns is thus correctly
matched to the number of equations. We see that the pressure must necessarily be
an intrinsic variable in fluid dynamical problems for there to be enough variables to
satisfy the basic laws of mechanies.
For many partieular problems it is convenient to use the equations referred to a
co-ordinate system rather than the vectorial forms. Most often one uses Cartesian
co-ordinates, but sometimes a curvillnear system is suggested by the geometry.
52 Physical Fluid Dynamics

Uste d below are the forms taken by the continuity equation (5.10) and the three
components of the Navier -Stokes equation (5.21) in Cartesian, cylindrical polar,
and spherical polar co-ordinates:
(i) Cartesian co-ordinates:
au av aw
-+-+-=0 (5.23)
ax ay az
au au au aU] = - -ap + /1 [a- 2u + -a2u + -a2u]
p [- + U - + v - + w - + Fx (5.24)
at ax ay az ax ax 2 ay2 az 2
together with similar equations for v and w.
(ii) Cylindrical polar co-ordinates (r = distance from axis, ep = azimuthal angle
about axis, z = distance along axis):

aUr + Ur +! ~ + au z = 0 (5.25)
ar r r aep az
p [aUr + U aUr + U</> aUr + U aUr _ U~] = _ ap
at r ar r aep z az r ar
+/1[a 2Ur +! aUr_u r +..!. a2ur+a2Ur_~ au</>J +F (5.26)
ar2
r r2 r2 ar aep2 az 2 r2 aep r

p [au</> + Ur aU</> + UrU</> + U</> au</> + Uz au</>J = _! ap

at ar r r aep az r aep

+ /1 [ a2u:t. +! au</> _ U</> +..!. a2u</> + a2u</> + ~ aUr] +F (5.27)

ar2 r ar r2 r2 aep2 az 2 r2 aep </>
p [auz + Ur auz + U</> au z + Uz auz J= _ ap
at ar r aep az az
a 2u 1 auz I a2uz a2uz ] F
+/1 [ - -Z+ - - + -2 - - + -2
ar2 r ar r aep2 az - + z
(5.28)

(iü) Spherical polar co-ordinates (r = distance from origin, e = angular displacement

from reference direction, ep = azimuthal angle about line e = 0):
aUr + 2ur +..!. aUIJ + ulJcot e+_1_ au</> =0 (5.29)
ar r r ae r r sin e aljJ

[ aUr + U aUr + UIJ aUr + ~ aUr _ uõ _ U$ ]

p at r ar r ae r sin e aep r r
= _ ap + /1 [a 2Ur +~ aUr _ 2u r +.!.. a2ur + cot e aUr
ar ar2 r ar r2 r2 ae 2 r2 ae
(5.30)
Equations of Motion 53

p [aulJ + U aUIJ + UrUIJ + UIJ aUIJ + ~ aUIJ _ u~ cot e ]

at r ar r r ae r sin e a</J r
1 ap [a 2 UIJ 2 aUIJ ulJ 1 a2 UIJ
=--; ae + fJ. ar2 +-;a-; - r2 sin 2 e + r2 ae 2
cot e aUIJ 1 a 2UIJ 2 aUr 2 cot e aU</> J
+T ae + r2 sin e a</J2 + r2 aB - r2 sin e a</J
2 + FIJ (5.31)

p [ au</>
au</> UrU</> UIJ au</> UIJU</> cot e
-+u -+-- +--+ +U</>
- -au</>
-
J
at r ar r r ae r r sin e a</J
=__I_ap+fJ.[a2u</>+[Link]</>_ U</> +~a2u</>+coteau</>
r sin ea</J ar2 r ar r2 sin 2 e r2 ae 2 r2 ae

+
1 a2u</>
- - + -2- -aUr --+
2 cot-e-aUIJ]
- +F (5.32)
r2 sin 2 e a</J2
r2 sin e a</J r2 sin e a</J </>
(One sometimes requires also the individual stress components in polar co-
ordinates. These will be introduced as required. For a more systematic treatment
the reader is referred to Refs. [12 and 21].)

5.7 Boundary conditions

Since the governing equations of fluid motion are differential equations, the speci-
fication of any problem must include the boundary conditions. We would expect
this on physical grounds; the motion throughout a fluid region is evidently
influenced by the presence and motion ofwalls or other boundaries. We examine
now the form taken by the conditions on the velocity field applying at boundaries.
There are obviously various types of boundary, giving rise to different possible
conditions. However, the only case that we need consider in any detail for the
purposes of this book is the most common type of boundary to a fluid region - the
rigid impermeable wall.
One condition applying at such a wall is obviously provided by the requirement
that no fluid should pass through the wall. If the wall is moving with velocity U and
the velocity of a fluid particle right next to the wall is u, the n this means that the
normal components of these two velocities must be the same;
U· n=U· n (5.33)
where nis the unit normal to the surface (Fig. 5.5). One often chooses a frame of
reference in which the boundaries are at rest, giving U = 0; this boundary condition
then becomes
u·n=O (5.34)
or in Cartesian co-ordinates with x and z in the local tangential plane to the wall and
y normal to it
v=O (5.35)
Another condition is provided by the no-slip condition, already mentioned in
Section 2.2, that there should be no ·relative tangential velocity between a rigid wall
54 Physical Fluid Dynamics
u

Figure 5.5 Velocity veetors of solid and of fluid particle immediately next to its surfaee.
(Note: U and u are shown different for definition purposes, although the text subsequently
shows them to be the same.)

and the fluid immediately next to it. Formally,

uxii=Uxn (5.36)
and, when U = 0,
uxn=O (5.37)
or
u=w=O (5.38)
It is apparent that some such condition must pertain. For example, without it, the re
wouId be no boundary condition on u (equations (2.7) and (2.16)) for pipe or
channel flow and the soIutions (equations (2.8) and (2.l7)) could not be obtained;
viscous action would place no limit on the amount of fluid per unit time that could
pass through a pipe under a given pressure gradient. However, it is not so apparent
that the condition should take this exact form. The notion underiying the no-slip
condition is that the interaction between a fluid partiele and a wall is similar to that
between neighbouring fluid partieles. Within a fluid the re cannot be any finite
discontinuity of veIocity. This would involve an infinite velocity gradient and so
produce an infinite viscous stress that would destroy the discontinuity in an infini-
tesimal time. If, therefore, the wall acts like further fluid, the action of viscosity
prevents a discontinuity in velocity between the wall and fluid; the no-slip condition
must apply . However, this concept that the wall acts like further fluid is itself an
assumption and the justification for the no-slip condition lies ultimately in experi-
mental observation. This experimental justification take s two forms. The first is
direet observation; dye or smoke introduced very elose to a wall does stay at rest
relative to the wall. The second is an a posteriori justification; the no-slip condition
is assumed and the solutions of the equations found in simple cases; agreement
between theoryand experiment then justifies the originaI assumption. The no-slip
condition is found to be violated only when the molecular mean-free path becomes
comparable with the distances involved; then the continuum equations are ceasing
to be applicable anyway.
Conditions (5.33) and (5.36) in combination do, of course, give
u=U (5.39)
Equations of Motion 55

or, in Cartesian co-ordinates on a wall at rest,

u = v = w = O. (5.40)
The total boundary condition is simply that there is no relative motion between a
wall and the fluid next to it. It is, however, important to note (and it will be ofsig-
nificance subsequently - Section 8.3) that the physical origins of the two parts of
the condition are quite .different. The no-slip condition depends essentially on the
action of viscosity, whilst the impermeability condition does not.
This is a convenient point to mention the forees exerted by amoving fluid on a
rigid boundary, a matter of obvious practical importanee. This again uses the notion
that the fluid acts on a wall in the same way as it ac ts on further fluid. However, no
assumption is involved here; the stresses must be continuous or a fluid particle at a
wall would experience infinite acceleration. Thus we may use the expressions in
Section 5.6 for stresses in the interior of the fluid. We use Cartesian co-ordinates as
above with the wall at y = O. Then the viscous stresses in the X-, y- and z- direction s
can be extracted from expression (5.19) and are given in the first place as

/1 au+av)
(- - . ( av)
2/1 - . /1
awav)
(- +- (5.41)
ay ax y=o' ay y=o' ay az y=o
Since, from the continuity equation,
av
-=----
au aw (5.42)
ay ax az
and since, also,

u = v =w =0 at y =0 for all x and z (5.43)

these reduce to

/1 - (au) .
ay y=o' (5.44)

The first and third quantities are the tangential forees per unit area on the wall.
There is no normal viscous foree, but there is a pressure force of p per unit area in
the -y-direction.
There are other important types of boundary besides the rigid impermeable wall.
The free surface of a liquid is an obvious example. And a rigid wall with suction or
injection of fluid through it has important practical applications, such as some air-
craft wings where the flow is controlled by suction of the air. However, we do not
need to formulate the corresponding boundary conditions for the purposes of this
book.
One other case does need mention - when the boundary condition is applied at
infinity. Often any boundaries are far from the region of interest, as in the examples
of an aeroplane weIl above the ground or a small model place d in a wind-tunnel.
The motion far from such an obstacle is the same as in the absence of the obstacle,
and one has a boundary condition of the form
u -+ Uo as r -+ oo (5.45)
56 Physical Fluid Dynamics

5.8 Condition for incompressibility

We have seen that both the continuity equation and the dynamical equation
simplify greatly if one can treat the density, p, as a constant. Fortunately, there are
many situations in which it is a good approximation to do so. Brief consideration
has been given to this matter in Section 1.2(ili) and we now examine more fully the
conditions in which one may make this approximation.
The status of the equation
p = const. (5.46)
is that of an equation of state. That is to say, in circumstances where it is not a good
approximation, one needs instead an equation of state giving the density as a func-
tion of pressure and temperature.
Correspondingly, when one does take the density as being constant, one is saying
that the density variations produced by the pressure and temperature variations are
sufficiently small to be unimportant. In this section we shall be principally
concerned with the effect of pressure variations, as we have already seen that such
variations are intrinsic to any flow. We shall derive a criterion for these to have a
negligible effect. Non-fulfilment of this criterion is the most famihar reason for
departures from equation (5.46); consequently this equation is called the incompres-
sibility condition, although the name is not a complete summary of the require-
ments for it to be applicable.
Temperature variations are also in principle intrinsic to any flow. (They may also
be introduced specifically - see Chapters 13 and 14 - but we leave to Section 13.2
and the appendix to Chapter 14 the corresponding considerations for that case.)
Firstly, the expansions and contractions as the fluid moves through the pressure
field involve temperature changes. The effect of these will, however, be covered by
the following discussion of the pressure effect; nowhere will it be assumed that the
changes are isothermal. Secondly, viscous action involves the dissipation of mechan-
ical energy (see Section 11.9), which reappears as heat; we will consider this briefly
at the end of this section.
Uquids are known to change their density very little even for large pressure
changes. One would expect to be able to treat these as incompressible. It is less
apparent that the re are important circumstances in which gases can be so treated -
although the resuIt is perhaps not wholly unexpected if one recalls that the
fractional change in the atmospheric pressure (and so the fractional change in the
air density) is small even when strong winds are blowing. The following derivation
of the criterion is thus of importance primarily for gases, although it is in fact quite
general.
We can write the density
p = Po + b.p (5.47)
where Po is a reference density - for example, the density at some arbitrarily
chosen point - and b.p is the local departure from this. If
b.p/Po ~ 1 (5.48)
then, for example, the term pDu/Dt in the dynamical equation can be approximated
by PoDu/Dt. A similar comment applies to the other places where p appears in the
Equations of Mo tio n 57

continuity and dynamical equations. Thus the equations with p const. may be used
when relationship (5.48) is satisfied.
Since the density changes under consideration result from pressure variations, in
order to estimate the typical size of l:i.p we need to know the typical size l:i.P of
these pressure variations. We get this infonnation from the requirement that the
pressure force must be balaneed by other terms in the Navier-Stokes equation and
thus will be of the same order of magnitude as at least one other term. (lf the flow is
produced by imposed pressure differences, then at the start of the motion the fluid
will accelerate until terms involving the velocity become comparable with the
pressure foree. If the flow is produced by imposed velocity differences, then the
flow will generate pressure differences of an appropriate size.) We confine attention
to steady flow without body forees, and so, either
'VP ~ pU''Vu (5.49)
or
(5.50)
or both (with the symboI ~ meaning 'is of the same order of magnitude as'). We
shall pursue the consequences of (5.49). We shall see in Chapter 8 that the only
circumstances when (5.50) applies whilst (5.49) does not are when the Reynolds
number is low, and the following analysis does not then apply.
Provided that the x-axis is chosen in a direction in which significant variations
oecur, (5.49) can be written
ap
-~
au
pu -= -p -
1 au 2 (5.51)
ax ax2 ax
This indicates that
(5.52)
where ill' and .Il (!P) are typical differences in p and u2 between points a distance
L apart. This means that, if one arbitrarily chose many such pairs of points, the
average difference in p would be of the general size l:i.P. Since l:i.P and l:i.(U 2 ) are
defined only as order of magnitude quantities they do not require more precise
definition than that.
If L is the generallength scale of the flow, ill' and l:i.(lj2) are the orders of
magnitude of the variations of p and u2 within the fluid; and are related by
(5.53)
We do not need to maintain the distinction between a typical difference in the
(square of the) velocity and a typical valu e of the (square of the) velocity itself; Le.
we can write
(5.54)
where U is a velocity-scale. The reason for this is that one always can (and normally
will) choose a frame of reference in which some points of the flow - for example
those at a boundary - are at rest. (It would be perverse to analyse the dynamics of
a low-speed aeroplane from the frame of reference of a high-speed aeroplane.) On
the other hand, it is necessary to maintain the distinction between the pressure
58 Physical Fluid Dynamics

difference scale and the pressure itself. Since the pressure appears only in the form
Vp in the governing equations, the absolute pressure can be increased indefinitely
without directlyt altering the dynamics; only pressure differences are relevant
Thus we have
(5.55)
which indicates the typical pressure variation in a flow of typical speed U. We now
use this to determine the typical density variation. This depends on the fluid and in
particular on its compressibHity, {3;
(5.56)
(For order of magnitude considerations it does not matter whether {3 is the isother-
maI compressibility, the adiabatic compressibHity of what.) The final result is given
in a convenient form if we now introduce the speed of sound, a, in the fluid;
2
a =
(ap) 1
ap s ~ p{3 (5.57)

(S = entropy). a is introduced here simply as a property of the fluid under consi-

deration, a measure of its compressibility.
Combining the various relationships, we have
t:..p t:..P (j2
p~ pa2~ a2 (5.58)

Thus criterion (5.48) is fulfilled if

(Ma)2 = (j2 la 2 ~ 1 (5.59)
Flows at speeds low compared with the speed of sound in the fluid thus behave as if
the fluid were incompressible.
The ratio U/a is known as the Mach number of the flow, and incompressible
flows thus occur at low Mach number. The fact that (Ma)2 is involved in relationship
(5.59) means that Ma does not have to be very small; when Ma is less than about
0.2, density variations are only a few per cent, bringing the accuracy of the incom-
pressibility assumption to within the so rt of accuracy attainable in many fluid
dynamical investigations.
Many important gas flows do occur at low Mach number. For example, the speed
of sound in air under atmospheric conditions is around 300 m s -I . Evidently, one
will often be concerned with speeds low compared with this.
The fact that liquids are much less compressible than gases is contained in this
analysis by the fact they have much higher sound speeds, thus giving lower Mach
numbers at the same U.
One general comment about the nature of the above argument may be made.
The way in which p enters the governing equations is important in deciding that
relationship (5.48) is a justification for treating p as a constant. Pressure provides an
immediate counter-example in the present context. The fractional pressure change

t It can do so indireetly by changing fluid properties.

Equations of Motion 59

may also be small; in fact, for a gas,

t::.P/p - I.::.p/ p (5.60)
This does not me an that the pressure can be treated as a constant. As we have
already noted, the pressure appears only in Vp and the absolute pressure can be
altered at will without changing the equations; comparison with it is thus irrelevant.
It was remarked that the above analysis does not apply at low Reynolds number.
Then one has to use (5.50) instead of (5.49). The corresponding analysis then gives
the criterion for incompressibility as
(5.61)
where Re is the Reynolds number - a somewhat academic result, rarely relevant to
real situations. (But see Section 24.4.)
A simiIar treatment of viscous dissipation shows that it also frequently has
negligible effeet. (It is, for example, not noticeably warmer swimming at the bottom
of a waterfall than at the top.) The details will not be given here; the corresponding
matter in the topic of free convection is discussed in the appendix to Chapter 14.
As in that case, the criterion for the resulting density changes to be negligible can be
expressed in a form not involving the viscosity coeffieient and related to the
criterion for incompressibility. This effeet can be ignored wherever (5.59) is fultilled
Hence, this provides an adequate criterion for the use of the incompressible flow
equations.

Appendix: Derivation of viscous term of dynamical equation

This appendix presents the main points of the mathematical formulation of the
ideas described physically in Section 5.6. For a more complete treatment, see, for
example Refs. [12 and 23] . We tirst examine the form of the rate-of-strain tensor
and then determine the consequences of a linear relationship between this and the
stress tensor.
Consider two material points separated by a distanee 01, having components OXi'
(01)2 = OXioxi (5.62)
with the summation convention for repeated suffixes applying.
D(01)2 D(oXi)
- - = 20x· - - = 20x·ou·
Dt I Dt I I

au· = ox/jXj
=20XiOX j_I ( --.1
au· +::::J..
au· )
(5.63)
OXj OXj OXi
It is thus the symmetrical combinations of the veloeity gradients that give rise to
distortions opposed by viscous action. (A motion in which
au· au·
= _ ::::::J. (5.64)
ax
_I
j OXi
for all i and j does not involve any relative movement between two neighbouring
points for any OXi') Moreover, these symmetrical combinations represent fractional
rates of change of the separation for various orientations; for example if
60 Physical Fluid Dynamics

1 D(ol) aUl
---=- (5.65)

They are thus rates of strain of the fluid. The rate-of-strain tensor ejj is formulated
as the symmetrical part of the velocity gradient tensor,

~1"'.1. -_ aaUj ejj =-21 (~jj + ~jj) (5.66)

Xj
[Parenthetically, the antisymmetric part
1
Tjjj =2" (~jj - ~jj) (5.67)

corresponds to the vorticity - see Section 6.4;

Tjjj =- 2"1 €jjkWk] (5.68)

The viscous stress tensor is given by defining Tjj as the force per unit area acting
in the i-direction across a surface normal to the j-direction. For a Newtonian fluid
this is linearly related to ej(
Tjj = Aijklekl (5.69)
The physical processes must be independent of the orientation and handedness
of the axes. A jjk1 must thus be an isotropic tensor and the most general form it can
take is [133]
A jjk1 = MjjO k1 + ~OjkOjl + XOilOjk (5.70)
This gives
Tjj = Mijekk + (~+ x)ejj (5.71)
(since ejj = ejD. There are thus two arbitrary constants involved; these are physical
properties of the particular fluid. From the particular case el 2 = Yzau/ay and all
other ejj equal to zero, we can identify that
~ + X = 2,u (5.72)
where,u is the coefficient of viscosity introduced in Section 1.2. Also,
aUk .
ekk =- = dlV u (5.73)
aXk

and so

T·· =,u
aU. + =.1
(aXj _I
aU.) + M·· div U (5.74)
Il aXj Il
The net viscous force on a fluid particle is given as before (Section 2.2) by the
differences in the stresses acting across opposite faces. The force per unit volume is
thus aTjj/aXj. Writing this in expanded form gives expression (5.19).
6
Further Basic Ideas

6.1 Streamlines, streamtubes, partide paths and streaklines

A streamline is defined as a continuous line within the fluid of which the tangent
at any point is in the direction of the velocity at that point. Its relationship to the
velocity field is thus analogous to the relationship of a line of force to an electric
field. Pattems of streamlines are useful (particularly in two-dimensional flow) in
providing a pictorial representation of a flow.
The streamlines for a known velocity field (u, v, w) are given as solutions of the
pair of differential equations
dx dy dz
-=-=- (6.1)
u v w
Two streamlines cannot interseet except at a position of zero velocity; otherwise
one would have the meaningless situation of a velocity with two directions.
A streamtube is a tubular region within the fluid bounded by streamlines.
Because streamlines cannot intersect, the same streamlines pass through a stream-
tube at all stations along its length.
Consider two stations along a streamtube of cross-sectional areas Sl and S2 as in
Fig. 6.1. We suppose that the cross-sections are small enough that there is negligib1e
variation of physical quantities over them and we can say that the densities and
speeds at the two stations are Pl and ql and P2 and q2 (q = 1u I; we can use the
scalar quantity, as the direction is by definition along the streamtube). The rate at
which mass is entering the volume between the two stations is P 1 q 1 Sl; the rate at
which it is leaving is P2q2S2. If the flow is either steady or incompressible (or, but
not necessarily, both) the mass in this region is not changing, and so
Plq1Sl = P2q2S2 (6.2)
If the flow is incompressible, Pl =P2 and
q2/ql = St/S 2 (6.3)
or, for a general station along the steamtube,
q 0: I/S (6.4)
The speed is inversely proportional to the cross-sectional area. Hence where the
streamlines are elose together the speed is high; where they are far apart it is low.
Convergence of streamlines indicates acceleration, divergence deceleration. These
facts are useful in the interpretation of streamline patterns. It is important to note
that we have derived them only for incompressible flow. Result (6.4) is, of course,
62 Physical Fluid Dynamics

Figure 6.1 Section of a streamtube.

directly related to the continuity equation, div u = 0, just as the parallel interpre-
tation that the electric field is large where there is a high density of lines of force
applies only when the re are no space charges and so div E = O.
One cannot directly observe a streamline by flow visualization. One can
synthesize a streamline pattem from pictures, such as Fig. 3.7, showing the change
in position of many partides during a short time interval.
In the most common types of flow visualization, dye (or smoke) is introduced
at a point. One then sees either a partide path or a streakline: the former if the dye
is introduced instantaneously and observed continuously or photographed with a
long exposure; the latter if the dye is introduced continuously and observed or
photographed instantaneously.
A partide path is effectively defined by its name. It is the trajectory of an
individual element of fluid.
A streakline is the locus of all fluid elements that have previously passed
through a particular fixed point.
In steady flow, streamlines, partide paths, and streaklines are all identical. A
partide is instantaneously moving along a streamline; if that streamline is unchanged
at a slightly later time, the partide will continue to move along it, and so it
traverses the streamline. Similarly, partides starting from the same point at different
times will all follow the same path, and so a streakline will consist of that path.
In unsteady flow, strearnlines, partide paths, and streaklines are all different; the
remarks in the last paragraph no longer apply. It is then that the interpretation of
pattems obtained by flow visualization calls for care; it may not be easy to infer
instantaneous flow pattems from the observations.
Thus, in Chapter 3, the steady flow pattems at low Reynolds number are quite
fully specified by the illustrations. Although the lines in Figs. 3.2 and 3.3 were the re
defined as partide paths, we can now say that these figures show strearnline pattems.
Further, Fig. 3.4, which is in principle a so rt of streakline pattem, can be used to
illustrate one of these flows. When the flow becomes unsteady, any individual
illustration teIls a more partial story. For example, the streakline pattems in Fig.
3.5 and the streamline pattem deducible from Fig. 3.7 complement one another.
The relationships between the streamlines, partide paths, and streaklines in an
unsteady flow - and the fact that they can look quite different from one another -
are best seen through an example. Section 6.2 wiIl consider one case in some detail,
and we wiIl take up the introduction of the basic ide as of fluid dynamics again in
Section 6.3.
Further Basic ldeas 63

6.2 Computations for flow past a circular cylinder

In this section we look at some computed flow patterns that illustrate the relation·
ship between streamlines, partide paths and streaklines. They concem a flow that is
steady in one frame of reference. However, we consider it also in another frame so
that the flow is then unsteady. (The computations thus also illustrate the relation·
ship between patterns in different frames, a matter mentioned in Section 3.1.) This
is, of course, a rather speeial type of unsteady flow, but it serves the present purpose
well. Illustration of the same ide as for a truly unsteady flow requires eine-film;
Ref. [141] provides a very effective example.
Figure 6.2 shows the resuIt of anumerieal solution [95] of the Navier-Stokes
and continuity equations for flow past a circular cylinder at a Reynolds number of
40. (In this and subsequent figures only half the flow is shown; the pattern is,
of course, symmetrical about the centre plane PQ.) This type of flow was the topic
of Chapter 3, where we saw that 40 is about the highest Reynolds number at
which the re is no unsteadiness due to wake instability. In Fig. 6.2 the cylinder is at
rest and the fluid far from it moves from left to right at speed uo. The figure shows
strearnlines, which, since the flow pattem is steady, are also partide paths and
streaklines. The dosed streamlines in the attached eddies behind the cylinder are
thus regions where fluid partides are moving on dosed paths; the eddies always
consist of the same fluid.
Detailed experimental comparison with this computed pattem is difficult. In
practice, a Reynolds number as low as 40 can be achieved only by working with
a smalllength-scale, at which observation of the flow details is not readily
achievable. However, the overall features of the flow are in agreement with the
observations described in Chapter 3. In addition, agreement between the value of the
drag coefficient given by the computations and the observed value suggests that the
details are correet.

--
------------
---~------

Figuf:e 6.2 Flow past a eireular eylinder at Re =40. Ref. [95).

64 Physical Fluid Dynamics

Figure 6.3 Streamlines for cylinder moving through stationary ambient fluid.

Figure 6.3 shows the corresponding streamline pattem in the frame of reference
in which the cylinder moves from right to left at speed uo, the fluid far from it
being at re st. It is thus related to Fig. 6.2 by the transformation of Fig. 3.1. The
regions where the fluid is now (Fig. 6.3) moving from left to right correspond to
regions where it was previously (Fig. 6.2) moving downstrearn faster than uo;
regions where it is now moving from right to left are those where it was previously
moving downstream more slowly than uo. The motion in the perpendieular
direction is the same at corresponding points of the two figures.
The pattem of Fig. 6.3 is an instantaneous one; at any other time the same
pattem is to be found displaced some distance to right or left.
The existence of the attached eddies is not apparent from this representation;
the reason is that the speeds relative to the cylinder in the attached eddies are so
small compared with Uo that the superposition of Uo swarnps them. The fact that
some of the fluid behind the cylinder moves towards it is implied in Fig. 6.3 by a
region (apparent only on dose examination) of fluid moving from right to left at
speeds slightly above uo.
Closed streamlines do appear in Fig. 6.3; indeed, on a figure of infinite extent, all
streamlines would be dosed (or begin and end on the cylinder). However, since the
pattem is not steady, this does not imply that fluid partides are moving on dosed
paths nor indieate the existence of permanent eddies like the attached eddies.
Figure 6.4 shows partide paths in the frarne of reference in which the cylinder
travels from right to left. A fluid partide is initially at re st with the cylinder far to
its right, is brought into motion as the cylinder approaches, follows some trajectory
as the cylinder passes, and finally settles down to rest again as the cylinder moves
away far to the left.
The single partide path in Fig. 6.4(a) is drawn to illustrate the relationship to the
streamline pattem of Fig. 6.3. At each instant the partide is moving with the velocity
indicated by the appropriate point on the strearnline pattem. For example, when
the partide has reached point A, the cylinder is directly below it (in the position
Further Basic ldeas 65

[b)

e
E

Figure 6.4 Particle paths for cylinder moving through stationary ambient fluid. See text for
significance of parts Ca) and Cb).

shown); the partide is thus at point A of Fig. 6.3. The partide path at A is seen to
I

be paralleI to the streamline through A'. At a later instant, the partide is at point B.
This corresponds to point B' of the streamline pattem, because, during the time in
which the partide has moved from A to B, the cylinder and the whole streamline
pattem have moved to the Ieft a distance of 1.5d (d being the cylinder diameter).
Similarly, when the cylinder was a distance 1.5d to the right of its illustrated
position (in Fig. 6.4(a»), the particle was at position C; it was thus at position C' of
the streamline pattem, and was moving accordingly.
Figure 6.4(b) shows a number of other partide paths for partides starting at
different distances from the plane of symmetry. Different partides, initially the
same distance from the plane of symmetry but separated in the direction of the
cylinder's motion, follow paths of similar shapes but at different times and in
different places. The paths in Fig. 6.4 have been drawn so that the partides
concemed are all immediately above the cylinder at the same instant (which we
may call t::: 0); Le. they are at the positions marked with crosses when the
cylinder is in the position shown.
These partide paths are of finite length. However, computation of their complete
length would require information extending outside the region for which the flow
pattem has been computed. Hence, the paths shown correspond to tuo/d (Le. time
in units of the time in which the cylinder moves one diameter) extending from -10
to 10. (For the cases where the edge of the diagram is reached first, points D, E, F
correspond to values of tuo/d of respectively -6.0, 3.7, and 7.3.) The motion
before tuo/d ::: -10 would be very slow and, for the partide paths near the top of
the diagram, the motion after tuo/[Link] 10 would be similarly very slow. Those
66 Physical Fluid Dynamics
particles which get caught up in the wake, on the other hand, are carried a long
way, and the path shown represents only the first part of the total path.
The above account covers the behaviour of all particles except those immediately
to the right of the cylinder at any time. These are in the attached eddies and so are
carried with the cylinder. The paths of such particles can be computed but are
difficult to illustrate. The particles move from right to left with a slow up-an d-down
motion corresponding to the circulation in the attached eddy; but the latter is so
sl ow that the path appears as a straight line in any diagram. These paths are, of
course, infinite ones. The existence of such paths shows the physical significance of
the attached eddies even though they are not apparent in Fig. 6.3.
Particle paths can eross, as is seen in Fig. 6.4. This means, of course, that the
particles concerned were at the crossing position at different times.

Figure 6.5 Streaklines for cylinder moving through stationary ambient fluid (and related
particle paths, shown as broken lines).

Figure 6.5 exhibits two streaklines (the two solid lines). Dye is supposed to have
been ernitted continuously from a point S at a distance d from the plane of
symmetry since tuo/d = -10 (t = 0 being the instant at which the cylinder is in the
position illustrated with its centre directly below the source). The lines show the
computed distributions of that dye at t =0 (curve SA) and at tuo/d = 10 (curve
SBe). Each streakline consists of particles that have passed through S at various
times - the further along the streakline from S, the earlier the time.
To demonstrate the relationship between streaklines and particle paths, Fig. 6.5
also shows - broken lines - two trajectories of particles originating at the source.
Line SAC originates at tuo/d = -10, reaches A at t = 0 and C at tuo/d = 10. Line SB
originates at t =0 and reaches B at tuo/d = 10. These particle trajectories can be
seen to be segments of paths like those in Fig. 6.4.
For reasons indicated above, these results are not of much value for direet
comparison with observation. They do demonstrate, however, the complexity of
the relationships between the various flow features in unsteady flow. They point
out the need for care in interpreting flow visualization experiments and illustrate
the type of reasoning that enters such interpretation.
Further Basic Ideas 67

6.3 The stream function

Sinee, for ineompressible flow,

div u = 0 (6.5)
one ean put
u = eurl A (6.6)
where A may be ealled a veetor velocity potential. Sinee this replaees one veetor
variable, u, by another veetor variable, A, this is not in general very useful. It is
useful, however, in a two-dimensional or axisymmetrie flow, where onlyone
eomponent of A is non-zero.
In a two-dimensional flow, for example, equation (6.5) beeomes
au av
-+-=0 (6.7)
ax ay
and this ean always be satisfied by introdueing t/I such that

u=-
at/l (6.8)
ay'
t/I is known as the stream funetion, beeause it is eonstant along a streamline -
verified as follows:
at/l at/l
dt/l =- dx_ + - dy = -vdx + udy (6.9)
ax ay
On a streamline
dx/u = dy/v (6.10)
and so
dt/l = 0 (6.11)

6.4 Vorticity

The vorticity, w, of a fluid motion is defined as

w = eurl u (6.12)
It is thus a veetor quantity defined at every point within the fluid. We sh all see in
later ehapters that some flow eonfigurations are most readily understood through a
eonsideration of the vorticity.
The definition of eurl by

ft'eurl u = lim ..!.fuodl (6.13)

8-+0 S
(where ii is the unit veetor normal to the surfaee S, round the edge ofwhieh the
line integral is taken) indieates that vorticity eorresponds to rotation of the fluid;
the line integral is non-zero only if fluid is going round the point under eonsidera-
tion. Flow without vortieity is ealled irrotational flow.
68 Physical Fluid Dynamics

Figure 6.6 Example of change of orientation (with translation subtracted out) in short time
interval of two materialiines instantaneously at right angles.

The physical meaning of 'rotation' and thus of vorticity when fluid partides can
change shape (Section 5.6) requires some consideration. A convenient way of seeing
howone may meaningfully talk about the rotation of a fluid partide in the presence
of deformation is the following. Consider a flow, such as a two-dimensional flow,
in which only the z-component of the vorticity is non-zero
av au
W =--- (6.14)
z ax ay
and consider two short materialIines instantaneously at right angles to one
another in a plane perpendicular to z (Fig. 6.6). A short time later these Iines of
fluid may have changed their orientation as shown. (We ignore the effeet of
second velocity derivatives in making them curved, as we are concerned with the
first order effeet at a point.) The rate at which the two Iines are turning will not in
general be the same, and it would not be sensible to de fine the rotation rate of a
fluid partide in terms of the angular velocity of one such line. However, the
average angular velocity of any two such lines at right angles is the same and is
equal to w z /2. This may be seen from the case in which the Iines are paralleI to the
x- and y-axes; then the angular velocity of the former is av/ax and that of the latter
is -au/ay. Since the definition of eur! is not related to any particular set ofaxes,
the resuIt is valid for other pairs of perpendicular lines in an xy-plane.
For further discussion of the physical significance of vorticity, it is convenient to
refer to three examples:
(i) The first is a fluid rotating as if it were a rigid body with angular velocity Q.
The velocity field is
u=Qxr (6.15)
giving
w=2Q (6.16)
The vorticity is the same at every point and equal to twice the angular velocity. For
comparison with the next example we write this also in cylindrical polar co-
ordinates:
U<fJ = nr, ur = U z = 0 (6.17)
Wz = 20., w r = WCjJ = 0 (6.18)
(ii) The second example again has every fluid partide moving on a circular path
about the z-axis, but with a different radial distribution of azimuthal velocity:
Further Basic /deas 69

K
u</> =- , ur =U z =0 (6.19)
r
This gives
W</> =0, Wr =0
1 a
= - - (ru</» = 0 for r *0 (6.20)
Wz
r ar
but leayes W z indeterminate on the axis. In fact, Wz becomes infinite on the axis, as
can be seen from Stokes's theorem

~u • dl = J(curl u) • dS =J li) • dS (6.21)

Carrying out the line integration round a circular path at constant r makes the left-
hand side equal to 2rrK independently of r. Hence, there must be a singularity in
W z such that the contribution wzdSz from the element at r = 0 to the surface
integral is 2rrK. Hence, this case corresponds to irrotational motion everywhere
except on the axis. Obviously, exactly this flow cannot occur in practice; however,
there are situations in which the vorticity is very high over a narrow linear region
and practically zero elsewhere.
(iii) The third example is the simple shear-flow
u = u(y) v =w =0 (6.22)
(in Cartesian co-ordinates) considered previously for different reasons (Sections 1.2
and 2.2). This has
Wx = w y = 0, Wz = -au/ay (6.23)
Examples (ii) and (iii) illustrate an important distinction. Rotation, as specified
by vorticity, c,orresponds to changing orientation in space of the fluid particle and
not to motion of the particle on a closed path. In example (ii) each fluid particle
moves round a circular path, but its vorticity is zero. In example (iii) each fluid
particle moves in a straight line but it has vorticity. The concept of an isolated rigid
particle moving on a circular path without rotating is readily illustrated, Fig. 6.7. It
is here contrasted with rigid body rotation in which the particle both moves on a
circular path and changes its orientation in space, so that the same side of it always
faces the axis of rotation. A fluid particle in the flow of example (i) behaves like

(a) (b)

Figure 6.7 Sueeessive positions of solid particle in Ca) rigid body rotation and (b) cireulation
without rotation. One side of particle is shown dotted to indieate its orientation.
70 Physical Fluid Dynamics

Figure 6.8 Sueeessive positions and shapes of initially square fluid partiele cireulating without
vorticity (size of partiele exaggerated relative to radius of eirele).

the particle in Fig. 6.7(a); its vorticity eorresponds to its changing orientation. The
behaviour of a fluid particle in example (ii) is le ss closely analogous to that of the
particle in Fig. 6.7(b), beeause it is also being deformed. Sueeessive eonfigurations
of an initially reetangular fluid particle are shown in Fig. 6.8. For small displaee.
ments, one ean see the meaning of the statements that the particle is not changing
its orientation and that perpendieular sides are rotating in opposite senses to give
zero vortieity.
Figure 6.9 iHustrates the interpretation of example (iii), where rotational flow in
straight lines oeeurs. Subsequent eonfigurations of an initially square fluid particle
are parallelograms. Similar geometries may be reached by a non·rotational distortion,
stretching along one diagonal and eompressing along the other (eorresponding
mathematieally to the prineipal axes of the rate-of-strain tensor). Rotation is then
neeessary, however, to give the aetual orien tation. Considering small displaeements,
we see that one of two instantaneously perpendieular side s is rotating, the other
is not, thus the average is non-zero and the re is vorticity.

(a)

D o
'"
Ib)

/
D
/

"
o
Figure 6.9 Square in (a) shear flow, equation (6.22), and (b) non-rotational distortion.
Further Basic ldeas 71

6.5 Vorticityequation

An equation for the vorticity in incompressible flow is obtained by applying the

curl operation throughout the Navier-Stokes equation (5.22) (with F = 0). We have,
as a vector identity,
v' x (u • v' u) = u • v' (V' x u) - (V' x u) . v' u + (V' • u)( v' x u) (6.24)
The continuity equation (5.10) makes the last te rm zero, and so
aw + u • v' IW -
- v' u = V V'2 W
W • (6.25)
at
The pressure has disappeared because it occurs in a conservative term, but, to offset
this simplification, the equation involves the velocity itself as weIl as the vorticity.
Equation (6.25) may be rewritten

(6.26)

giving an expression for the rate of change of the vorticity of a fluid partide. The
second term on the right-hand side shows that the action of viscosity produces
diffusion of vorticity down a vorticity gradient, like it produces diffusion of
momentum down a momentum gradient. This might be expected from the fact that
rigid body rotation with its uniform distribution of vorticity (example (i) above)
involves no viscous action.
The first term on the right-hand side of equation (6.26) represents the action of
velocity variations on the vorticity. The precise physical interpretation of this term
varies with the details of the velocity field, but may be illustrated as foIlows.
Consider a partide with its vorticity (instantaneously) in the z-direction and of
r
magnitude Then the z-component of the te rm under discussion
aw
(w • yo u)z = ~az- (6.27)

Positive aw/az indicates that the fluid partide is elongating in the direction of its
vorticity. To conserve mass it contracts in the x- and/or y-direction (notice that the
continuity equation entered into the derivation of equation (6.26»). It thus rotates
faster, as does a solid body of decreasing moment of inertia (e.g. a ballet dancer who
brings in her arms, or the Earth as its denser materials settle to the centre). The
vorticity inereases. This process is often referred to as vortex stretehing.
Just as one can draw streamlines everywhere in the direction of the local velocity,
so one can draw vortex lines everywhere in the direction of the local vorticity. An
important resuIt, that will form the basis of some of the physical arguments in
subsequent chapters, is that for an inviscid fluid (see Section 8.3 and Chapter 10) a
vortex line always consists of the same fluid partides.
This may be shown by considering a short segment AB of length € of a vortex
line instantaneously in the z-direction (Fig. 6.10)
Wt=o = (0,0, n (6.28)
For an inviscid fluid, the development of the vorticity field is governed by the
72 Physical Fluid Dynamics

Vortex line
=0
at t

Vortex line
at t = ot

Figure 6.10 Successive positions of vortex line and two points on it.

equation
Doo
- =00· V'u (6.29)
Dt

(6.30)

and so, after a small time Ot, the vorticity of the same fluid is

oot=ot=
au av aw)
( ~-Ot,~-Ot,~+~-Ot (6.31)
az az az
We now enquire about the orientation of A'B', A' being the position at t = ot of the
fluid that was at A at t = 0 and B' being similarly related to B. If the velocity
components at Aare (u, v, w), those at B are
au v + -e
(u + -e aw)
av w +-e
az' az' az
Thus
(6.32)

:>)
AA' = (uOt, vOt, wot)

BB' = [( u + :~ e ) Ot, ( v + Ot, ( w + ~: e ) ot ] (6.33)

and

AB =" (au-eot, -eot,

av e + -aw)
eOt (6.34)
az az az
The components of (6.34) are in the same ratios to one another as those of (6.31);
the new orientation of the segment of vortex line is the same as that of the material
line. Applying this result at each point along the vortex line, one concludes that it
always consists of the same fluid partides.

6.6 Circulation

The circulatiün rüund any arbitrary clüsed lüüp in the fluid is defined as

r = ~u • dl (6.35)
Further Basic /deas 73

This differs from the definition of the vorticity in equations (6.12) and (6.13) in
that the loop is nowa finite one. The cireulation is non·zero in example (ii) of
Seetion 6.4 as well as in examples (i) and (iii).
The eireulation is related to the vorticity by Stokes's theorem, equation (6.21).
There must be vorticity within a loop round whieh cireulation oeeurs.
7
Dynamical Similarity

7.1 Introduction

In only a minority of fluid dynamical situations can one determine the flow as an
exact solution of the equations of motion. The necessary mathematical methods
often do not exist. Even when an exact solution can be obtained it may not be
unique and so may not correspond to what actually oecurs. Hence, much of fluid
mechanics concems the development of bOth experimental and theoretical
proeedures for elucidating flows that cannot be rigorously calculated.
A useful starting point for this development is the following question: under
what conditions do similar flow pattems occur in two geometrically similar pieces
of apparatus? When these conditions are fulfilled, the two flows are said to be
dynamically similar.
It is easier to think in terms of a specific system and we choose the example
shown in Fig. 7.1. An elliptical obstacle of dimensions a and b as shown is plaeed in
a channel of width c through which fluid moves. Other lengths may be necessary
to specify the dimensions in the third direction. We consider two such pieces of
apparatus, which are geometrically similar; Le. one is just a scaled up version of the
other, so that
al/al =b 2/b l =e2/cI (7.1)
(where the suffIxes refer, of course, to apparatus 1 and apparatus 2). The two fluids
in the two sets of apparatus may be different; p and 11 are parameters of the problem.
Given geometrical similarity, there is then the possibility that the two flow pattems
will be similar; that is
UP2/UP! =VP2/VPl =WP2/WPl =UQ2/uQl =VQ2/VQl =WQ2/WQl (7.2)
n,
for all pairs of geometrically similarly located points PI, Ql, Q2 in the two
pieces of apparatus. However, this dynamical similarity will oceur only if further
conditions are fulft1led, and we have to investigate what these are.
Although the discussion will refer to this example, the analysis to be given
applies to all steady situations govemed by the incompressible continuity equation
and the Navier-Stokes equation without a body foree. The same general method is
applied with appropriate modifications to situations where other terms and/or
other equations come in, as we shall be seeing in later chapters.
There are several ways in which this question of dynamical similarity is important.
It is evidently of practical importance for experiments with models. Tests of, say,
an aeroplane or a ship using a small-scale model in a wind-tunnel or a towing tank
will not be useful if the flow in the model experiment is quite different from that
Dynamical Similarity 75

/////////L///////1/L

®t
.Q

~
-
?7777777777777777~
_b _ J e

Figure 7.1 Example of situation to which ideas in text can be applied.

which occurs on the full scale. To take an extreme example, they will be useless if
the flow with the model is larninar and that on the full scale is turbulent; less
extreme differences, such as two laminar flows with different strearnIine pattems,
are also undesirable. Dynamical similarity is likewise important in more fundamental
experimental work. If one has observed a certain flow behaviour, one needs to
know when this behaviour will occur and whether further experiments are necessary
to investigate the full range of behaviour. Dynamical similarity also plays an
important role in the theoretieal development; it leads to useful approximations to
the equations of mutIOn (see Chapter 8).

7.2 Condition for dynamical similarity: Reynolds number

The derivation of the condition for dynamical similarity starts by introducing

length and velocity seales, L and U. These are typical measures of the size of the
apparatus and the rate at which the fluid is moving in it. In our example, a possible
choice of L would be the width of the obstaele, L =a, and a possible choice of U
would be the speed at the centre of the channel well upstream of the obstaele,
U= Up. However, the choiees are arbitrary and the conelusion does not depend on
the details of them. No pressure difference scale is introduced as the pressure
differences cannot be varied independently of the velocity field. It would be
possible to introduce apressure difference scale instead of a velocity scale, but
again this would alter only the form in which the conclusion is expressed, not the
conclusion itself.
No time scale is introduced, because we are restricting attention to steady
imposed conditions. The analysis will inelude time-variation of the flow, because,
as we have seen in Chapters 2-4, unsteadiness can arise spontaneously. When this
happens, the unsteadiness does not provide any time scale of its own; the rapidity
of the fluctuations is related to the length and velocity seales. A flow that is
inherently unsteady because of variations in the imposed conditions does involve an
independent time scale, indicating the rapidity of these variations; this case is not
covered by the following analysis.
There are two routes by which we may reach the conelusion. We will consider
each in tum. The fust is the quicker. The second perhaps shows better what under-
lies the conclusion and illustrates the method used in more complicated problems
where the fust route is less easy to apply.
Dynarnieal sirnilarity pertains if the ratio of each of the velocity components at
geometrically sirnilar points of the flow to the velocity scale is the same in both
76 Physieal F/uid Dynamies

pieces of apparatus; that is if

u' =ulU, v' =v/U, w' =w/U (7.3)
are the same when
x' =x/L, y' =y/L, z' =ziL (7.4)
are the same. We focus attention on u'; exactly parallel considerations apply to v'
andw'.
u' is a dimensionless quantity. In any expression for it, dimensional quantities
must appear in non-dimensional combinations. We can thus write
, (b e " , p UL )
u =f 1: L' ... ,x ,y ,z '--;;- (7.5)

The quantities on the right-hand side are all the independent non-dimensional com-
binations of the physieal parameters on whieh u' can depend. Any further non-
dimensional combinations, for example pUb/Il, are also combinations of the above
and are thus impHcitly included.
If all the quantities on the right-hand side of (7.5) are the same for two systems,
then u' is the same. Equality of b/L, elL, etc. in the two cases is a restatement of
geometrical similarity. Equality of x', y' and z' indicates that geometrically similar
points are being considered. Hence, equality of pUL/1l is the significant criterion.
Dynamieal similarity of two geometrically similar systems of the type under
consideration exists if they have the same value of the Reynolds number
Re =pUL = UL (7.6)
JJ. v
The second route to this resuIt starts with the governing equations, in this case
equation (5.10) and equation (5.22) with F = O. The argument is clearer if we
expand these in Cartesian co-ordinates,
au av aw
- + - +-=0 (7.7)
ax ay az

au +u au +v au +wau= _ ~ ap +v(a 2u + a2 u + a2 u) (7.8)

at ax ay az p ax ax 2 ay2 az 2
with similar equations for v and w. [Link] and velocity seales, L and U, are intro-
duced as before and are used to give non-dimensional forms (denoted by primes) of
the variabIes:
x' =x/L y' =y/L z' =ziL t' =tU/L (7.9)
u' =ulU v' =v/U w' =w/U (~)' = ~/pU2
(~being the difference between the pressure and some reference pressure.)
Separate time and pressure difference scales are not introduced for the reasons
discussed earHer. This means that t and ~ are non-dimensionalized using appro-
priate combinations of the other seales. The ways chosen are not unique (e.g. IlU/L
has the dimensions of pressure as weIl as pU 2 ); however, different choiees would
make no difference to the final conclusion. Substituting (7.9) into equations (7.7)
Dynamical SimikIrity 77

and (7.8) gives

!!(au' + av' + aw' ) = 0 (7.10)

L ax' ay' az'
U 2 au' U2 (,au' , au' , au')
- - + - u-+v-+w-
Lat' L ax' ay' az'
= _ U 2 a(!:J.p)' + vU (a 2u' + a2u' + a2u') (7.11 )
L ax' L 2 ax'2 ay'2 az'2
with similar equations for v' and w'. Henee,
'V' ·u' = 0 (7.12)
au'
- + u'·V'u' = - V'(A~)' +-
1
V'2 U ' (7.13)
at' '-'f' Re
where
, ,a ,a ,a
V = x ax' + y ay' + z az' (7.14)

u' =u/U (7.15)

and Re is the Reynolds number, UL/v. The boundary eonditions ean be similarly
eonverted to eonditions in the non-dimensional variabIes. We see that if the
Reynolds number is the same for two geometrically similar situations, then the
equations for the non-dimensional variables are the same. Henee, they have the
same solutions and the same flow pattems oeeur. If the Reynolds number is differ-
ent, the equations are different and there is no reason to expeet the same flow
behaviour. We have the same result as before that the eondition for dynamieal
similarity is equality of the Reynolds number.
This eonclusion greatly reduees the amount of work needed to obtain full
information about a situation such as that in Fig. 7.1. Whereas it might seem that
one would need to investigate separately the effeet of varying eaeh of L, U, p and p,
in faet one need investigate only the variations with Reynolds number.

7.3 Dependent quantities

One may need to know the value of some quantity dependent on the flow. One
would expeet that when dynamical similarity pertains, then information about such
quantities ean be transferred between systems; for instanee, model experiments ean
give information not onlyabout the flow on full seale but also about quantities
associated with it.
We may use, as an example of such a quantity, the foree aeting on an obstacle
like that in Fig. 7.1. We denote this by D (treated as a sealar on the assumption that
its direction is known from symmetry; if not the argument ean apply to eaeh eom-
ponent). The drag eoeffieient, de fine d as
CD =D/YzpU2 L 2 (7.16)
78 Physical Fluid Dynamics

is a non-dimensional form of this (the 1/2 is ürelevant but introdueed for eon-
sisteney with eonventional notation - see also Chapter 3).t It is not the only non-
dimensional form, but a different choice would give an equivalent eonclusion.
The drag eoeffieient depends only on the Reynolds number
CD = [(Re) (7.17)
There are again two routes to this eonclusion. The first is adireet dimensional
argument: one says that the only dimensionally satisfaetory expressions of
D = [(L, b, e, ... , U, p, Il) (7.18)
are of the form

CD=[(~,~, ... ,Re) (7.19)

and when there is geometrieal similarity (7.19) beeomes (7.17).

The second route starts with the faet that forees aeting on the obstac1e are
pressure forees of !lp per unit area and viseous forees of the form /laujay per unit
area. In the non-dimensional variabIes, these are pU 2 (t:.p)' per unit area and
(pU 2 /Re )au'jay' per unit area. When Re is the same in two cases the solutions for
the primed variables are the same and the forees per unit area are proportionalt to
plJl. The total forees are thus proportional to plJl L 2 . Henee, when Re is the same,
DjplJl L 2 is the same; and one has equation (7.17).
This relationship enables information about forees to be transferred from one
apparatus to another geometrieally similar apparatus. If Re is the same, CD is the
same; one ean then infer the foree in one apparatus from measurements in the other.
We now have the full justifieation for the viewpoint adopted in Chapter 3 that Fig.
3.14 eontains eomp1ete information about the drag on a cireu1ar eylinder in steady
ineompressible flow.
It is well known that dimensional analysis sometimes gives a more explieit
expression for a dependent quantity than a relationship of the type of (7.17).
This oeeurs when the number of dimensional quantities involved is smaller. For
example, we shal1 see in Seetion 8.2 that when the Reynolds number is low, the
equation of motion simplifies to a form (equation (8.7)) not involving p.
Thus p drops out of (7.18) and so out of the relationship between Re and CD' This
requires
CD cx: I/Re (7.20)
When the other quantities are fixed this eorresponds to the drag being direetly pro-
portional to the speed.
If /l drops out instead of p, similar reasoning gives
CD = eonst. W.r.t. Re (7.21)
eorresponding to the drag being proportional to the square of the speed. This
resuIt relates in a rather eomplex way to the behaviour at high Reynolds numbers
(Seetions 8.3 and 11.5).

tThe appearance of L 2 in the denominator of (7.16) compared with d in (3.3) is because D is

now the total force on the obstacle whereas, in Chapter 3, it was the force per unit length.
:j:As one goes from one case to the other: not as one varies U.
Dynamical Similarity 79

Although this section has been almost entirely concerned with drag, the
discussion has been intended to illustrate general principles applying to any
dependent quantity. Another example is the frequency of a vortex street. We see
now how the statement (equation (3.2» that the Strouhal number is a function of
the Reynolds number is a consequence of dynamical similarity.

7.4 Other governing non-dimensional parameters

The above discussion of the Reynolds number illustrates a general procedure. When
other equations of other terms in the equations are applicable, other dimensionless
combinations of the parameters may be formulated in addition to or instead of the
Reynolds number. An early stage of the investigation of a new configuration is
usually the formulation of the governing non-dimensional parameters, which deter-
mine the dynamieal similarity. The use of the Rayleigh and Prandtl numbers to
specify the convection problem in Olapter 4 arises from such considerations. This
and other cases will be considered in detail in later chapters, but we look straight-
away at a few simple examples to illustrate further the principles:
(i) The first is a situation similar to that considered in Sections 7.1-7.3 but with
a free surface on which waves can develop. In certain circumstances, important in
the dynamics of ships, these waves are dominantly gravitational and in addition to
U, L, p and v, the acceleration due to gravity, g, is a parameter of the flow. There
are two independent non-dimensional combinations of these
Re = ULlv and Fr = UI(gL)1I2 (7.22)
the Reynolds number and the Froude number. Dynamical similarity requires that
(7.23)
(suffIxes 1 and 2 referring to the two geometrically simiIar systems being compared).
The drag coefficient, for example, has the dependence
CD = [(Re, Fr) (7.24)
(ii) Sometimes one is concerned primarily with effects of the gravitational waves
and the viscous flow below the surface is unimportant. The Froude number is then
the only relevant governing non-dimensional quantity, and
CD = [(Fr) (7.25)
(This applies, more precisely, when nearly all the energy, generated by the work
done by force D is carried away by the waves rather than dissipated by viscosity.)
(iii) For a situation again similar to our main example but with a speed high
enough -for the fluid to be compressible, dynamical similarity requires equality
between systems of both the Reynolds number and the Mach number, Ma = U/a
(a is the speed of sound; see Section 5.8). Thus
CD =[(Re, Ma) (7.26)
(iv) When the flow is unsteady as a resuIt of changes in the imposed conditions,
these changes will have a time scale 'It associated with them. In problems such as
the above there is then the additional non-dimensional parameter U'It IL, and
80 Physical Fluid Dynamics

dynamieal similarity throughout the development of the flow requires equaIity of

this in addition to the Reynolds number.
It should be noted that, in the context of model testing, the above diseussion of
dynamieaI similarity is the statement of an ideaI. It is often not possible in practice
to make all the governing non-dimensionaI parameters the same as on the full seaIe.
In ship model testing, for instanee, a reduetion in L requires an inerease in U to
keep the Reynolds number the same but a reduetion in U to keep the Froude
number the same (sinee there is little manoeuvreability of p, v and g). Henee, tests
have to be made without full dynamicaI sirnilarity, and speciaI attention must be
given to the errors arising in the transfer of information to the full scale.
8
Lowand High Reynolds Numbers

8.1 Physical signifieanee of the Reynolds number

The Reynolds number - introdueed in the last ehapter in the context of dynamieal
similarity - ean be given a physieal interpretation. This is useful in gaining an under-
standing of the dynamieal processes that are important in different Reynolds
number ranges, and in formulating eorresponding approximations to the equations
ofmotion.
To diseuss this we need a name for eaeh of the terms in the dynamical equation
of steady ineompressible flow
pu·V'u = -Vp + JlV'2 U (8.1)
The second and third terms are given the obvious names, pressure foree and
viseous foree. The first term is ealled the inertia foree. Physieally, it is not a foree,
but it has the dimensions of foree per unit volume and it is sometimes eonvenient to
think of the dynamieal equation in terms of a static balanee between forees. The
proeedure is analogous to the mare familiar use of the term eentrifugal foree to
represent the aeeeleration involved in cireular motion. No new idea is involved here,
just a new name.
In the non-dimensional form of equation (8.1)
, n' , _
A
u·vu--v~p
)' n'( + -1vn'2
U' (8.2)
Re
(ef. equation (7.13)), the primed quantities (possibly exeepting (~p)') may be
expected to be of order unity in magnitude. We shall see later that there are im-
portant qualifieations to that statement. However, as a starting point it is justified
so long as the length and velocity seales, U and L, have been chosen as typical
quantities. Then a general speed will be of order U and I u' I - 1; a general distanee
over which quantities vary signifieantly will be of order L and ajax' etc. will be of
order unity.
Henee the ratio of the first term to the third in equation (8.2) is of order Re.
The eorresponding terms in equation (8.1) are in the same ratio. This indicates a
physieal interpretation of the Reynolds number as
inertia forees
Re------ (8.3)
viseous forees
An alternative (entirely equivalent) formulation of this result, cited beeause we
82 Physical Fluid Dynamics
shall proceed in this way in subsequent chapters, is to write
I u· Vu l-lP/L, I vV 2 u 1- vU/L 2 (8.4)
Hence
.:. . 1u_·_v.:....u---.:I _ UL = Re
(8.5)
IvV2 ul v
The Reynolds number thus indicates the relative importance of two dynamical
processes. At a general point within the flow, the ratios of these two terms will not
be exactly equal to the Reynolds number, but their characteristic magnitudes will
be in this ratio.

8.2 Low Reynolds number

When the Reynolds number is much smaller than unity the viscous force dominates
over the inertia force so much that the latter plays a negligible role in the flow
dynarnics. One may use an approximate form of the equation of motion with the
inertia term dropped. Equation (8.2) becomes

O=-V'(Llp)'+~ VilU' (8.6)

Re
these terms being of order I/Re and the neglected term of order 1. The pressure
term must be retained since it is necessary to match the number of variables to the
number of equations (Section 5.6). Physically, this means that the size of the
pressure term is always govemed by the other dynamically important terms - in
this case by the viscous term.
Reverting to the dimensional form, equation (8.6) is
Vp = /.lV2 U (8.7)
At every point in the fluid there is an effective balance between the local pressure
and viscous forces. Equation (8.7) is known as the equation of creeping motion. It
is evidently much simpler than the full Navier-Stokes equation, and solutions have
been found for many cases for which the full equation has not yielded a solution.
One case will be discussed in Section 9.4. Such solutions are found to agree well
with the observed behaviour at low Reynolds number (see, e.g., Fig. 9.3), thus
justifying the procedure leading to the approximation.
It is a characteristic feature of low Reynolds number flows that viscous interac-
tions extend over large distances. For example, particles sedimenting at low
Reynolds number affect each other's motion even when their separation is large
compared with their size. Figure 8.1 illustrates this long range viscous action for
flow past a eircular cylinder. It re-presents the information of Fig. 3.2 to show the
velocity distribution across the mid-plane at a Reynolds number of 0.1. In the next
section we shall be looking at the corresponding figure for high Reynolds number
flow; comparison of the two provides a good illustration of the way in which
different dynamical processes dominate in different Reynolds number ranges.
Apart from Section 9.4, we shall give little further attention to low Reynolds
number flows. There are nevertheless many interesting low Reynolds number
phenomena, as is particulady weIl illustrated by the mm of Ref. [42].
Lowand High Reynolds Numbers 83

<4-- Uo ----

Figure 8.1 Veloeity distribution on eentre plane in flow past eireular cylinder at Re = 0.1.

8.3 High Reynolds number

Corresponding arguments for high Reynolds number flow indicate that the viscous
force is so small compared with the inertia force that it can be neglected.
Equation (8.2) then approximates to
u'· Vu' = - V(.:::lp)' (8.8)
of dimensionally
pu' VU = -Vp (8.9)
This is Euler's equation of inviscid motion. When it applies, the fluid at each point
has an acceleration directly related to the pressure gradient.
The argument applies also to unsteady flow (/1 V 2 u being negligible compared
with pu • Vu regardless of the size of pau/at), and a more general form of Euler's
equation is
Du
p Dt = - Vp (8.10)

The relationship of this equation to the actual behaviour at high Reynolds

numbers is much more complex than the relationship of the creeping motion
equation to low Reynolds number flow. Comparing Euler's equation with the
Navier-Stokes equation, we see that the discarded term is the highest order differ-
ential term - the onlyone involving second space derivatives. The approximation
thus reduces the order of the differential equation. A corresponding reduction must
be made in the number of boundary conditions.
We saw in Section 5.7 that the no-slip boundary condition is a consequence of
the action ofviscosity. One may thus expect that this is the condition that should
be discarded for mathematical consistency with the inviscid equation. This is indeed
84 Physical Fluid Dynamics

the case; we shall see in Chapter 10 that solutions of Euler's equation are obtained
by matching only to the impermeability condition (equation (5.33». Imposition of
the no-slip condition also would result in no solution being obtainable.
We now have a paradoxical situation. The statement that Euler's equation applies
at high Reynolds number means, more preeisely, that as the Reynolds number is
increased the viscous term becomes relatively smaller and smaller, although never
absolutely zero; Euler's equation becQmes a better and better approximation. A
boundary condition, however, cannot be similarly relaxed as the Reynolds number
increases. It either applies or it does not apply; there is no meaning to the state me nt
that the no-slip condition is present but to a negligible extent. On the other hand,
one would not expect there to be some Reynolds number at which the no-slip con-
dition suddenly 'switches off', and it is found experimentally that it continues to
apply no matter how high the Reynolds number.
Consequently, the viscous term in the dynamical equation must always remain
significant in the vieinity of a boundary, so that the equation remains of the order
appropriate to the boundary conditions. The region in which this happens is known
as the boundary layer. The reasoning (equations (8.2)-(8.5» that the viscous force
should be negligible breaks down in the boundary layer because the flow develops
an internallength scale much smaller than the imposed length scale, L. This is the
boundary layer thickness, ö. We shall see below, and in more detail in Chapter 11,
that the size of the viscous term Can be determined by ö whilst the size of the
inertia term is stiIl determined by L:
lu· Vul-U 2 /LlvV 2 ul-vU/ö 2 (8.11)
Inertia and viscous forees can thus remain of comparable order of magnitude if
U2 /L - vU/ö 2 (8.12)
that is if

LÖ '-_. (Uv'L) -1/2

= Re- 1I2 (8.13)

The difference between the two length scales must become more marked as the
Reynolds number inereases.
We consider first the simplest example of a boundary layer. Suppose that a flat
plate of negligible thickness is place d in a uniform stream, speed uo, parallel to it
(Fig. 8.2). Then, for a theoretical situation governed by Euler's equation and with-
out the no-slip condition, the flow could obviously continue as if the plate were not
there (Fig. 8.2(a». The speed would be Uo everywhere. In a real fluid, at large
Reynolds number, the speed remains very close to Uo over most of the flow. Close
to and behind the plate, however, there are regions in which a large change occurs
(Fig. 8.2(b». These are the boundary layers on either side of the plate and the wake
behind it. We shall not consider the wake further at the moment, but it can be
thought of as the extension of the boundary layers downstream.
The rapid variation of speed in the boundary layer gives rise to much larger
values of a2 u/ay 2 than would otherwise exist. This makes the viscous force much
larger and so makes it appropriate to use (8.11), rather than (8.4), for the orders of
magnitude.
It is often useful to have a preeise definition of the boundary layer thickness, ö
(shown in Fig. 8.2(b». There is, of course, no line beyond which the presence of
Lowand High Reynolds Numbers 85

(a)

(b)
-- - ----

Figure 8.2 Velocity prof11es in flow past thin plate: (a) imagined inviscid flow; (b) real fluid
at high Reynolds number. Dotted lines indicate edges of boundary layer and wake.

the plate has absolutely no effect; the velocity stilI asymptotes to uo, but very
rapidly. A common procedure is to choose 8 such that
u = O.99uo at y =8 (8.14)
(y is distance from plate); the boundary layer is taken to be the region in which the
velocity differs by more than 1 per cent from the free-stream velocity.
The longitudinallength scale L is provided in this example by the distance from
the leading edge. We shall see in Section 11.3 (equation (I 1.29» that

X~ (UVOX) -112
0::
(8.15)

- as expected from relationship (8.13).

The eage of the boundary layer is not a streamline. The only significance of the
line, y = 8, shown in Fig. 8.2(b) is that indicated by equation (8.14). Fluid crosses
this line. In the present example, fluid just outside the boundary layer at one value
of x is inside it at larger x.
For any obstacle other than a flat plate parallel to the free-stream, the situation
is more compHcated. The fluid is diverted past the obstacle and the solution of
Euler's equation is not just a uniform flow (which would not satisfy the impermea-
bility boundary condition). It would thus be meaningless to de fine the boundary
layer and wake as the regions in which the velocity departs significantly from uo.
They are defined instead as the regions in which the action of viscosity significantly
affects the velocity distribution. Suppose one has found a solution of Euler's
equation for the flow past such an obstacle (which we may call the inviscid flow
solution). This will not satisfy the no-slip condition. elose to the wall of the
86 Physical Fluid Dynamics

Figure 8.3 Inviseid (broken line) and high Reynolds number (fullline) veloeity distributions
aeross centre plane in flow past eireular eylinder (negleeting effeets of boundary layer
separation).

obstacle, viscous action will modify the flow so that the no-slip condition is obeyed.
The region in which this happens is the boundary layer.
We can thus generalize the specification of a boundary layer implied by
equation (8.14): the boundary layer is the region in which the velocity differs by
more than 1 per cent from the inviscid flow solution.
Figure 8.3 shows an example (ef. Fig. 8.1 for the corresponding example at low
Reynolds number). The inviscid flow solution for flow past a circular cylinder gives
a velocity profile across the mid-plane as shown by the broken curve. The speed at
the wall is twice the free-stream speed. The full curve shows a high Reynolds num-
ber velocity profile, satisfying the no-slip condition. The boundary layer is the
region in which the two profiles differ significantly.t
Chapter 10 will describe the elements of inviscid flow theory. Chapter 11 will
discuss boundary layers. We can now see the role of these two aspects in the develop-
ment of our understanding of flow at high Reynolds numbers. The first stage in
tackling a new problem is usually a solution of the inviscid flow problem. This will,
subject to qualifications made below, describe most of the flow. It does so, how-
ever, only because the region in which it applies is separated off from the no-slip
condition by the boundary layer. One of the pieces of information obtained from

tWe know from Chapter 3 that Fig. 8.3 involves some oversimplifieation. In the first place,
over a wide Reynolds number range, the separation of the flow from the wallleading to the
formation of attaehed eddies oeeurs upstream of the station shown. Secondly, even when it
oeeurs downstream, the existenee of the separation will modify the inviseid flow solution - not
directly through the extension of the viseous region but indireetly through a ehange of the
pressure distribution that makes a different solution of Euler's equation applicable. (See
Seetion 11.4.)
Lowand High Reynolds Numbers 87

the inviscid flow solution is the pressure distribution over the boundaries. This is
part of the input to the second stage, a treatment of the boundary layer. If this
indicates that the boundary layer undergoes separation (see Section 11.4), some
modification to the inviscid solution will be required; in principle an iterative pro-
cedure is then needed, though in practice it may be difficult.
Historically, Euler's equation is older than the Navier-Stokes equation, and it
was puzzling that it described some aspects of fluid behaviour very weIl whilst
failing totally to describe others. We can now see the reason: those aspects described
well were those unaffected by the presence of boundary layers. An important
feature that does depend on boundary layers is the force on an obstacle, and we
shall see that it was with this that failure of Euler's equation was particularly
dramatic. The introduction of the concept of boundary layers, by Prandtl in 1904,
was alandmark in the history of fluid dynamics; a very high proportion of subse-
quent developments stern directly from it.
We have seen in Chapters 2 and 3 that flow at high Re)fllolds number is prone to
instability. Above we have been thinking mainly of laminar flow. The discussion is,
however, by no means academic. Transition to turbulence occurs in regions such as
boundary layers and wakes, whilst laminar flow continues in inviscid regions. Hence,
the division into the two regions is still useful.
It is apparent that a mu ch wider range of phenomena occur at high Reynolds
number than at low. For this reas on alone one would wish to give the former
situation much more extensive consideration in a book of this sort. However, it is
worth noting that this emphasis coincides to a large extent with the type of situation
one meets most frequently in practical situations. The values of the kinematic
viscosity for water and air (at common values of the temperature and pressure) are
respectively 1.0 x 10-6 m 2 S-1 and 1.5 x 10-5 m 2 S-1. In either fluid one needs
only an object of a few centimetres in size moving at a speed of a few centimetres
per second to reach a moderately high Reynolds number.
9
Some Solutions of the Viscous Flow
Equations

9.1 Introduction

An obvious aim, once the equations of motion have been set up (in either the full
form of Otapter 5 or in an approximate form such as those in Otapter 8) is to fmd
solutions of them. Fluid dynamies thus constitutes an important branch of applied
mathematics, although the re are severe limitations to what can be learned by theory
alone because of mathematical complexity, non-uniqueness, and instability. In this
book, the mathematical aspects are somewhat underplayed to leave room for a full
development of the experimental aspects. In this chapter we do look briefly at a few
solutions of the Navier-Stokes equation and of the equation of creeping motion,
both to illustrate the mathematical aspects of the subject and to provide informa-
tion required elsewhere. For more systematic mathematical treatment see Refs.
[12,13,14,15] .
IneidentaUy, in the solutions considered in this chapter, the non-linear terms
play either no role at all or a secondary role. They are unimportant either for
geometrical reasons or, in the case of creeping motion, for the reasons discussed in
Section 8.2. These solutions are, in this respeet, untypical of much of fluid
dynarnics where non-linearity is responsible for both the mathematical difficulties
and for the distinctive phenomena. The solutions of the boundary layer equations
in Otapter 11 will provide examples of the mathematical handling of such probIems.

9.2 Poiseuille flow

We have already Iooked in Otapter 2 at soIutions for flow through a channeI and
through a pipe. For completeness we need to see that the equations set up there
from first prineiples are indeed speeial cases of the full continuity and Navier-
Stokes equations.
The assumption that there is onlyone non-zero component of the veloeity
reduces the continuity equation (5.10) to
ou/ox =0 (9.1)
which accords with the veloeity prome being the same at all values of x. It also
me ans that onlyone component of the Navier-Stokes equation is significant and
this is
ou op
pu - = - - +p.V u
2
(9.2)
ox ox
Some Solutions of the Viscous Flow Equations 89

The left-hand side is zero from (9.l), and the equation takes the form of either
(2.6) or (2.14) for respectively Cartesian and cylindrical co-ordinates. Equations
(2.8) and (2.17) are thus solutions of the equations of motion that we have now
established.

9.3 Rotating Couette flow

Another simple case of some importanee (as the principle of a device for
measuring viscosity, amongst other applications) is rotating Couette flow. Fluid is
contained in the annulus between two long concentric cylinders of radii aIand az
rotating about their common axis with angular velocities nl and n z (Fig. 9.1). We
want to know the velocity distribution within the annulus (and, in viscosity
measurements, the torque acting on the cylinders). A solution of the equations of
motion may be obtained by assuming that the velocity is everywhere in the
azimuthal (I/J) direction and that the velocity and pressure are independent of I/J and
z (cylindrical polar co-ordinates shown in Fig. 9.1). These assumptions are the
obvious ones suggested by the geometrical symmetry, but, as we shall consider in
detail in Section 17.5, instability can produce a changeover to a flow with a more
complex structure, to which the following theory does not apply.
The continuity equation aU.p/al/J = 0 is automatically satisfied by these assump-
tions, and the azimuthal and radial components of the Navier-Stokes equation
become
0= JJ. (dZU.p +.!. du.p _ u.p)
dr z , dr ,2 (9.3)

pu~ dp
,
--=--
dr
(9.4)

with the boundary conditions

U.p=nlal at ,=al; u.p = nzaz at ,=az (9.5)
The fust equation can be s01ved to give u.p and this is then put into the second
equation to give p. Thus, the distribution of the azimuthal velocity across the

Figure 9.1 Defmitian sketch for rotating Couette flow (z-axis is normal to paper).
90 Physical Fluid Dynamics

annulus is determined by the balance of viscous stresses, whilst the pressure

distribution is determined by the balance between a radial pressure gradient and the
eentrifugal foree associated with the circular motion.
The solution for uef> (obtained by working in terms of the variable uef>fr) is
~=k+~ ~~
where
A = (.Q2a~ - .Qla?)/(a~ -a?), B = (.Ql - .Q2)a?a~/(a~ -a?) (9.7)
The torque ~l acting on the inner cylinder (per unit length in the z-direction) is
given by the viscous stresst #l[ra(Uef>/r)/ar],=a l multiplied by the area 21Tal and by
the radius al; Le.
~l =41T#LO~a~(.Q2 - .Qd/(a~ - a~) (9.8)
Similarly, the torque on the outer cylinder
~2 = -41T#la? a~ (.Q2 - .Qd/(a~ - a~) (9.9)
We notiee that ~l and ~2 are equal and opposite, as they must be sinee the total
angular momentum of the fluid is not changing.

9.4 Stokes flow past a sphere

The most famous solution of the equation of creeping motion (8.7) applies to low
Reynolds number flow past a sphere. It leads to the relations hip between the
velocity and the drag used to detennine viscosity in the familiar procedure of obser-
ving the rate of fall of a sphere through a viscous fluid. This is often known as
Stokes flow.
In spherical polar co-ordinates with 8 = 0 in the flow direction in the frame of
referenee in which the sphere is at rest (Fig. 9.2) the equations are obtained from
eqöations (5.29)-(5.32) with the inertia terms (the left-hand sides of equations
(5.30)-(5.32)) and the body force tenns put equal to zero, and with uef> =0,
a/acp =0 by symmetry. The boundary conditions are
U, =u6 =0 at r =a (9.10)
U,--+Uo cos8, U6--+-uOsin8 as r--+ oo (9.11)
where Uo is the free-stream velocity and a is the radius of the sphere. The solution is

=u 0 cos8 [ 1 -3a a3 ]
2r- +2r-
U
3
(9.12)
,

3
U =-u sin 8 [ 1 - -3a - a-] (9.13)
60 4r4r3

P - Po =- 7.3 7
#luoa
cos 8 (9.14)

tThat transformation to polar co-ordinates gives an expression of this form is to be expected

from the fact that there will be no stress in rigid body rotation, uef> <x r.
Sam e Solutions of the Viscous Flow Equations 91

Figure 9.2 Definition sketch for Stokes flow pa st a sphere (</> is azimuthal angle about (J =0
axis).

Po being the ambient pressure. This may readily be shown by substituting the
solution into the equations; for the forward integration the reader is referred to
other sources (e.g. Refs. [12,15]).
The force per unit area in the flow direction acting at a point on the surface of
the sphere is the sum of the appropriate components of the viscous and pressure
forees; that ist

a = -11 (au e) sin e-

(p - PO)r=a eos e (9.15)
arr=a
Substituting the solution into this gives
a e: 3lluo/2a (9.16)
Since this happens to be independent of e the total force on the sphere is just a
multiplied by the surface area of the sphere;
D = 41Ta 2 a = 61Tllauo (9.17)
In terms of a drag coeffieient defined by (7.16) with L = 2a, this is
CD = 61T/Re (9.18)
(ef. (7.20)).
Equation (9.17) is the well-known result, due to Stokes, used in falling sphere
viscometry. Figure 9.3 shows a comparison of this result with experimental
observations. The good agreement is important, not only for viscometry, but also
because it demonstrates the validity of the reasoning leading to the equation of
creeping motion and so encourages one to apply similar reasoning to other probIems.
Figure 9.3 also shows the departures from Stokes's law that occur when the
Reynolds number is too high for the creeping flow equation to apply. It is
important to remember that equation (9.17) can be used only when Re is less than
about 0.5. (It is not suffieient, as is sometimes said, that the flow should be laminar.)
In applying this result it should also be remembered that, as remarked in
Section 8.2, the viscous effects extend a long way at low Reynolds numbers. Distant
boundaries may thus have a disturbingly large effect. In a falling sphere viscometer,
the container diameter must be more than one hundred times the sphere diameter
for the error to be less than 2 per cent [123].

tThe viscous stress in spherical polar co-ordinates reduees to this simple form for a boundary
at rest.
92 Physical Fluid Dynamics

100

50

20
D 'te

•
X
pIJ2
10 •e
~ ev (Re)2
2 ,e

x
/'"
."
2
,,/
/0
x
/

0.5

0.05 0.1 0.2 0.5 2 5 10

Re - pu.d/J.I

Figure 9.3 Drag on a sphere at low Reynolds numbers. Experimental points from Refs. [161)
(x) and [221) (e), both using the falling sphere method. The line represents equation (9.17).

9.5 Low Reynolds number flow past a cylinder

Although the equation of ereeping motion generally deseribes flow at low Reynolds
number very satisfaetorily, there is a eomplieation that arises in two-dimensional
flow, such as the flow past a eireular eylinder. No solution of the equation ean be
found that matches to the boundary eonditions both at the surface of the eylinder
and at infinity (for mathematieal details see, e.g., Ref. [12]). This faet is sometimes
referred to as Stokes's paradox. Its resolution again involves the faet that the effeet
of a boundary on the velocity field extends to very large distanees from that
boundary. If we eonsider amoving eylinder in stationary ambient fluid, the fluid
veloeity remains eomparable with the eylinder velocity to distanees so large eom-
pared with the eylinder radius that the Reynolds number based on the flow length-
seale is of order unity. There is thus a region in whieh inertia forees are signifieant.
The lower the Reynolds number the more remote this region is from the eylinder,
but it always exists.
A low Reynolds number theory allowing for the existence of this region has
Some Solutions of the Viscow Flow Equations 93

been developed [7]. It gives, for the dependence of the drag eoeffieient on
Reynolds number for a eireular eylinder, the expression (in the notation of
Seetion 3.4)
CD =81T/Re(2.002 -In Re) (9.19)
This provides the low Reynolds number end of the eurve in Fig. 3.14 (the rest of
whieh is eompiled from experimental data). There is always a small departure from
(7.20).
In real situations, of eourse, it may be other boundaries rather than a remote
inertial region that shouid be considered in resolving Stokes's paradox.
10
Inviscid Flow

10.1 Introduction

The relationship of inviscid flow theory to actual flow at high Reynolds number has
been considered in Section 8.3. We now look at some of its features. Although
Euler's equation is non-linear, we shall see that in many important cases it reduces
to a linear equation. Consequently, it yields solutions much more readily than the
full Navier-Stokes equation. This is one of the most mathematically developed
branches of fluid mechanics in which experimental work plays a more minor role.
It is very fully treated in many books e.g. Refs. [7, 10, 179, 262J. Consequently,
we shall confine attention here to the important basic ideas, without developing the
methods of application of these ideas to particular cases.

10.2 Kelvin circulation theorem

It may be shown that, for any flow governed by Euler's equation, the circulation
round a loop consisting continuously of the same fluid partides is conserved. One
may write this

~.f u·dl = 0 (l0.1)

Dt1
This result is known as the Kelvin circulation theorem.

10.3 Irrotational motion

Arelated, less general but for present purposes more useful, result concerns the
vorticity. The vorticity equation corresponding to Euler's equation is equation
(6.29). This has the property that
Doo
if oo = 0, then -=0 (10.2)
Dt
If a fluid partide has no vorticity at some initial instant then it can never acquire
any.
This result is known as the permanenee of irrotational motion. It corresponds
physically to the fact that, when Euler's equation applies, the only stresses acting
on a fluid partide are the pressure stresses. These act normally to the partide
surface and so cannot apply a couple to the partide to bring it into rotation. (Once
/nviscid Flow 95

it is rotating, its vorticity can change through the vortex stretching action described
in Section 6.5, but this cannot produce the initial rotation.)
Many interesting flows are initially irrotational - for example, a uniform flow
approaching an obstacle. Hence, the study of inviscid motion may for many pur-
poses be reduced to the study of irrotational motion. This does not cover all
solutions of Euler's equation, but in the others, the vorticity would have to be
introduced initially by some other process.
In irrotational motion
(ü= curl u = 0 (10.3)
throughout the flow. Hence, one may introduce IP such that
u = grad IP (10.4)
IP is known as the velocity potential (by analogy with other potentials; it is not,
however, associated with energy in the way electrostatic and gravitational
potentials are).
From continuity
div u = 0 (10.5)
and so
(10.6)
the velocity potential obeys Laplace's equation.
Two parenthetic comments about this theoryare worth making. Firstly, the
boundary condition corresponding to the imperrneability condition at a stationary
wall is
alP/an = 0 (10.7)
where n is the direction normal to the boundary. We can now see that ifwe had
also a condition on rJ> corresponding to the no-slip condition, the problem would be
over-specified and no solution would be available.
Secondly, the analysis leading to equation (10.6) applies to both steady and
unsteady flow. Yet any time variation does not appear explicitly in (10.6). This
means that the instantaneous flow pattem depends only on the instantaneous
boundary conditions and not on the history of the flow. Every point of the flow
respCjlnds immediately to any change in the boundary conditions (a change, for
example, in the speed at which an object is moving through a fluid). The physical
mechanism by which this response is brought about is a change in the pressure
distribution. Evidently there must be an upper limit to the rate at which a pressure
change can actually transmit through the fluid, and so alimit to the applicability of
the present theory. Small pressure changes transmit at the speed of sound; larger
ones transmit at a speed of the same order of magnitude. Hence, the criterion for
validity of the theory is closely related to the low Mach number criterion. If
L/aw < 1 (10.8)
(where L is the length scale, wis the time scale introduced in example (iv) of
Section 7.4 and a is the speed of sound), then pressure changes transmit a distance
large compared with L in the time in which significant changes in the flow pattem
occur. Hence, the response can be effectively instantaneous.
96 Physical Fluid Dynamics

Techniques for solving Laplace's equation are highly developed as aresult of its
importance in a variety of other physical contexts. The problem of inviseid flow
past an obstacle is directly analogous to that of the flow of electric current around a
cavity in a conductor, or to that of the electric field distribution in a region of high
dielectric constant around a cavity oflow dielectric constant. However, we shall
see in Section lIA that, because of the phenomenon of boundary layer separation,
many simple solutions do not correspond closely to real flows. The most important
solutions are for cases, such as the flow past an aerofoil, that have no physically
significant counterpart in other branches of physics. Consequently, there has been
a lot of work on solutions of Laplace's equation in fluid dynamical contexts. The
speeial methods developed are described in the references mentioned above.

lOA Bemoulli's equation

As weIl as the veloeity distribution, derivable straightaway from a solution for ep,
one may also want to know the pressure distribution. In deriving an equation for
this, we will now confine attention to steady flow. However, we will derive it in the
first place for any steady inviscid flow, irrotational or rotational. Then Bemoulli's
equation relates the variation of speed and variation of pressure along a strearnIine.
For steady flow we have a fixed streamline pattem. We consider one streamline and
denote the distance along it by 1. At any point on it the component of Euler's
equation in the streamline direction is
aq ap (10.9)
pq al = - al
q is the magnitude of the velocity, q = 1u I. Integrating equation (10.9),
l/zpq 2 + P = constant along a streamline (10.10)
This is called Bemoulli's equation.
At positions along the strearnIine where the veloeity is high the pressure is low
and vice versa. l/zpq 2 is the kinetic energy per unit volume, and the equation may be
interpreted in the following way. When the pressure is increasing in the flow
direction, a fluid particle is doing work against the pressure gradient and so loses
kinetic energy. When the pressure is decreasing it gains kinetic energy.
When the flow is irrotational, Bemoulli's equation can be extended. We start
with the vector identity
u· Vu::: l/zV(u· u) - u x (VX u) (10.11)
For irrotational flow, the second term is zero and
u • V u::: l/zVq 2 (10.12)
Euler's equation is then
V(l/zpq 2 + p) ::: 0 (10.13)
and so
l/zpq 2 + P ::: constant throughout the flow (10.14)
/nviscid Flow 97

The Bemoulli constant is the same for every streamline. The relationship between
high velocity and low pressure and its converse apply throughout the flow.
We may see the physical significance of this result by referring to two simple
examples in which q is constant. If, further, u is constant - the flow is entirely
uniform - there can be no pressure variations between streamIines as there is
nothing to balance the pressure gradient. The Bemoulli constant is constant
throughout the flow. The other example is flow in circular streamlines with u<t>
independent of r (a case intermediate between examples (i) and (ii) in Section 6.4,
but more difficult to generate in practice than either of those). This involves
vorticity. The pressure varies with radius, the pressure gradient balancing the centri-
fugal force of the circular motion. The Bemoulli constant is thus a constant only
on each strearnline.

10.5 Drag in inviscid flow: d'Alembert's 'paradox'

Bemoulli's equation indicates the pressure distribution over the surface of an

obstacle placed in a stream and thus the force that would act on that obstacle if the
boundary layers could be ignored. We consider this for an obstacle that is symmetri-
eal upstream and downstream, as in Fig. 10.1, If the flow were from right to left
instead of from left to right, the streamIine pattem would be the mirror image
pattem about the plane of symmetry AB. However, equation (10.6) has the property
that, if ~ =~l is a solution, then so is ~ = -~l . Thus the streamIine pattem for the
flow from right to left is the same as that for the flow from left to right; just the
direction of flow along the streamlines is reversed. Hence, considering now just the
flow from left to right, the flow pattem downstream of the plane of symmetry is
the mirror image of that upstream. The speed q is the same at symmetrically placed
points such as e and D (Fig. 10.1). From Bemoulli's equation the pressure is also
the same. The pressure distribution around the rear half of the obstacle is the same
as that round the front half, and there is thus no net force on the obstac1e. The
drag is zero.
This result is actually true whether or not the obstacle is symmetrical (although

lA
I

,
:8
Figure 10.1 Inviscid flow past a symmetric body, considered in discussion of d' Alembert's
paradox.
98 Physical Fluid Dynamics

it is not demonstrated so quiekly in the latter ease [12]). Irrotational flow ealeula-
tions give zero drag on any obstacle plaeed in a fluid stream.
At one time it was thought that theory of this type ought to produee the correet
drag. Henee, the above resuIt beeame known as d' Alembert's paradox. This name
lingers today, although the resuIt eeases to be paradoxieal onee one knows that the
region of applieability of the theory is separated from the surfaee of the obstacle by
a boundary layer. (See also Seetion 11.5.)
Indeed it would be mare paradoxical if eompletely inviscid theory did give a
drag. We ean see this most readily by switehing attention to the frame of referenee
in whieh the obstacle is moved through stationary fluid. Work would then be done
against any drag and energy would be fed into the fluid. But, in the absenee of
viseous action, there is no process by whieh this energy eould be dissipated, and no
steady state would exist. (Arguments of this eharaeter require some eaution. They
do not apply when there are waves present whieh ean earry energy away 'to
infinity' .)
The resuIt that inviscid motion produees no drag applies only for steady flow.
In unsteady flow, although the veloeity distribution is the same as that for the
instantaneously eorresponding steady flow, the pressure distribution is different.
The work done against the drag as an obstacle is aeeelerated through a fluid provides
the ehange in the kinetic energy of the fluid.

10.6 Applications of Bemoulli's equation

Although, in steady flow, inviscid theory prediets no foree on an obstacle in the

direction of relative motion between it and the fluid, it can predict a force at right
angles to this direction. Such a foree is ealled a lift. lift generation is obviously of
great praetieal importanee; Bemoulli's equation ean be used to understand in a
general way why an aeroplane ean fly. Consider an aerofoil such as that shown in
Fig. 10.2. We use the frame of referenee in which the aerofoil is fixed, so that the
flow is steady. Above the eonvex upper surfaee the streamlines are pushed together
and the fluid moves faster than the free-stream speed; the pressure is therefore
redueed, as indicated in the figure. Similarly, the speed is redueed below the eon-
eave lower surfaee, and the pressure is increased. One sees that these pressure
changes are such as to produee a net upward force on the aerofoil. In Seetion 11.2
we shall see that the pressure differenee across the boundary layers is small, and so
does not affeet this argument, unIess boundary layer separation (Section 11.4)
occurs. However, a fuller understanding of the processes involved does require
consideration of viscous effects, and we shall retum to the topie of lift in Chapter 12.

Figure 10.2 Pressure changes associated with Bernoulli's equation around an unsymmetrical
aerofoil.
Inviscid Flow 99

- --.-""'"

Figure 10.3 Venturi demonstration. (Flow is from left to right; fluid in side-tubes is dyed.)

Another system that is readily understood in terms of Bernoulli's equation is the

Venturi tube. Indeed, this provides the simplest eonvenient demonstration of the
relationship between speed and pressure. A pipe with a fluid passing through at high
Reynolds number (thin boundary Iayers) has a eonstrietion in it - a short Iength of
redueed diameter. Beeause of eontinuity, the speed increases at the eonstrietion
and so the pressure is redueed. This is demonstrated by the apparatus shown in
Fig. 10.3; the height the water rises up the vertieal open side tubes is a measure of
the pressure. The pressure is seen to drop markedly at the eonstrietion and to rise
again after it. Aeeording to Bernoulli's equation, the pressures on either side of the
eonstrietion should be the same, but, beeause of some viseous effeets (notably those
to be deseribed in Seetion 11.4), the downstream pressure is a little lower.
The Venturi tube is of importanee not merelyas a demonstration deviee. For a
earefully designed eonstrietion, the pressure differenees ean be related quantitatively
to the flow rate; this thus provides one method of measuring flow rates through
pipelines. The Venturi is also a eonvenient simple way of obtaining pressures below
100 Physical Fluid Dynamics

atmospheric; for example, the suction pumps, fitted to taps and used by chemists
to speed up fIltration, operate on the Venturi principle.
Bemoulli's equation is also the basis of one of the most important instruments
for measuring fluid velocities, the Pitot tube. This will be described in Section 23.2.

10.7 Some definitions

Since all the terms in Bemoulli's equation have the dimensions of pressure, the
following nomeelature has developed. lhpq 2 is known as the dynamic pressure and
(lhpq 2 + p) as the to tal pressure or sometimes (for reasons indicated below) as the
stagnation pressure. To distinguish it from these quantities, P is sometimes called
the static pressure, although this is a somewhat misleading name (and it is quite
different from the hydrostatic pressure to be introduced in Section 13.2). The
static pressure is, of course, the pressure physically existing in the fluid.

Figure 10.4 Inviscid flow past an obstacle showing front and rear stagnation points, S and Q.

The total pressure has the physical significance that it is the pressure at which
the fluid comes to rest. Let us consider, for example, the flow past an obstaele
placed in an otherwise uniform flow of velocity Uo at pressure Po (Fig. 10.4). The
streamlines pass on different sides of the obstaele. One streamline, in the middle,
ends on the obstaele at point S and the fluid at this point is at rest. According to
inviscid flow theory, this and a corresponding point Q at the rear of the obstaele
are the only points at which the velocity is zero. S and Q are known as the forward
and rear stagnation points. Applying Bemoulli's equation to the streamline ending
there shows that the pressure at S is
Ps=lhpu'f,+po (10.15)
Le. it is the total pressure associated with the undisturbed flow. In theory, the
same pressure exists at Q, but in practice the action of the boundary layer almost
always greatly alters this. At the forward stagnation point, however, the effect of
the boundary layer is slight, and the pressure PS is actually observed. This is the
highest pressure anywhere in the flow.
11
Boundary Layers and Related Topics

11.1 Boundary Jayer formation

The reason for the occurrence of boundary layers and their role in high Reynolds
number flows have been considered in Section 8.3. However, the fact that flow
outside the boundary layers is irrotational (Section 10.3) provides another way of
viewing the process of boundary layer formation. Fluid partides can acquire
vorticity only by viscous diffusion (Le. through the action of the term VV'2 W in
equation (6.26)). The action of viscosity comes in at the boundary through the
need to satisfy the no-slip condition. As aresult vorticity is introduced into the
flow at the boundary, and then diffuses away from it. The boundary layer can be
defined as the region of appreciable vorticity. The boundary layer is long and thin
(L ~ li) when the fluid travels a long distance downstream during the time that the
vorticity diffuses only a small distance away from the boundary. This happens when
the Reynolds number is large.

11.2 The boundary Jayer approximation [22]

Because of the difference in length scales in different directions, certain terms in

the equations of motion play a negligible part in the dynamics of boundary layers.
We now see in a systematic way how this can be used to formulate an appropriate
approximation to the equations. This wi1l provide further justification for the ideas
introduced in Section 8.3. Also, the resulting equations can sometimes be solved
when the exact equations cannot. We shall be 100king (Sections 11.3 and 11.8) at a
couple of solutions both for their own interest and as our examples of the mathe-
matieal methods used for fully non-linear probIems.
From the outset we confine attention to steady, two-dimensional boundary
layers - asevere restriction from apraetieal point of view, but one that still
allows us to see the general principles involved. We suppose that the boundary layer
is forming on a flat wall (with the x co-ordinate in the flow direction andy normal
to the wall). A free-stream velocity outside the boundary layer is prescribed as a
function of x. This could be achieved by making the wall one side of a channel
of variable width as in Fig. 11.1 (with the channel width always large compared to
the boundary layer thickness). In fact, however, it makes negligible difference if the
surface is curved, so long as the re are no sharp comers - more precisely, so long as
the radius of curvature of the surface is everywhere large compared to the boundary
layer thickness. Thus, the prescribed free-stream velocity could be a solution of
102 Physical Fluid Dynamics

• •
L

FiguIe 11.1 Boundary layer on fiat wall of channel: defmition sketch.

Euler's equation for flow past an obstacle (x then being a eurvilinear eo·ordinate in
the surfaee).
We denote the free-stream veloeity by Uo and the pressure associated with it by
Po·
We take the boundary layer to have length seales L and li in the x- and y-
direetions, as in Seetion 8.3. We may expeet that the velocity seales wiil also be
different in different direetions and we denote the seales of u and v by U and V.
Similarly the order of magnitude of the pressure differenees aeross the boundary
layer in the y-direetion may not be the same as the order of magnitude of the
imposed pressure differences outside the boundary layer; we denote the seale of the
former by T and the seale of the latter by il. We now eonsider eaeh of the equations
in turn, labelIing the terms with their orders of magnitude .
The eontinuity equation is
au av
-+-=0 (11.1 )
ax ay
U v
L 0
The two terms must be of the same order of magnitude; fluid entering or leaving
the boundary layer at its outer edges must be associated with variations in the
amount of fluid travelling downstream within the boundary layer. Henee,
V ~ Uo/L; (11.2)

the velocity eomponent normal to the wall is small eompared with the rate of
downstream flow when the boundary layer is tbin.
The x-component of the Navier-Stokes equation is
au au
u- + v- = - - -+v-+v-
1 ap a2 u a2 u (11.3)
ax 2ay p ax ax ay 2
~ VU U2 il vU vU
-~-

L {) L pL L2 {)2
Boundary Layers and Re/ated Topics 103

The second expression for the order of magnitude of v au jay has been written using
relationship (11.2). The two parts of the inertia term are comparable with one
another, the smallness of V/U compensating for the more rapid variation of u with
y than with x. The two parts of the viscous te rm are however of different sizes when
olL is small, and v a2u/ax 2 may be neglected.
The y-component of the Navier-Stokes equation is
av av
u ax + v ay -p1 ap
ay
a2 v
+ v ax2
a2 v
+ v ay2 (11.4)

uv u 20
-"-'--
V 2 u 20
--"-'-
vV vUo vV vU
'r
-"-'- -"-'-

L L2 0 L2
p8 L 2 L 3 8 2 L8
In both equation (11.3) and equation (11.4) the pressure term will be of the same
order of magnitude as the Iargest of the other terrns. Hence,
ll/pL"-' U 2/L "-' vU/8 2 (11.5)
'r/p8 "-' U 2Ö/L 2 ,,-, vU/ L8 (11.6)
and so
'rjll "-' 8 2jL 2 (11.7)
The pressure differences across the boundary Iayer are much smaller than those in
the x-direction. Hence, at any value of y the difference between (1/p)apjax and
(1/p)dpo/dx is much smaller than the significant terms in equation (11.3) and we
may re place the former by the latter, giving
au au 1 dpo a2 u
u-+v-=-- - + v - (11.8)
ax ay p dx ay 2
Outside the boundary layer there is no variation with y and
duo 1 dPo
Uo dx = -p dx (11.9)

aresult which could also be obtained from Bernoulli's equation. Hence

au au duo a2 u
u-+v-=uo - + v - (11.10)
ax ay dx ay 2
This equation together with
au av
-+-=0 (11.11)
ax ay
constitute the boundary Iayer equations - two equations in the two variables u
and v. uo(x) must be known for a solution to be found.

11.3 Zero pressure gradient solution

The simplest, and in a ,sense most fundamental, solution of the above equations is
the zero-pressure gradient boundary Iayer;
dpo/dx = 0 (11.12)
104 Physical Fluid Dynamics

and so
Uo = constant (11.13)
Such a boundary layer is readily observed on a thin flat plate set up parallel to the
free-stream; one wall of an empty wind-tunnel or water-channel is sometimes used.
The equations for this case are
au au 02U
u-+v-=v- (11.14)
OX ay oy2
au av
- + - =0 (11.15)
OX ay
with boundary conditions
u =v=0 at y =0
u -+ Uo as y -+ oo (11.16)
We look for a solution of the form
u =uog(y/tl.) (11.17)
where tl. is a function of x. That the solution should be of this form is an assump-
tion. It corresponds to the velocity profile having the same shape at all values of x,
although with a different scale in the y-direction, and is thus physically plausible.
tl. is directly proportional to the boundary layer thickness, but it is convenient to
define it slightly differently from li.
Equation (11.15) can be satisfied by introducing a stream function I/J such that
u = ol/J/3y v = -ol/J/ox (11.18)
as in Section 6.3. Equation (11.17) is then
I/J =uotl.J(y/tl.) (g =J') (11.19)
Substituting this in equation (11.14) gives
u~ dtl. ff " + VUo J'" =0 (11.20)
tl. dx 2 tl.
where the prime indicates differentiation with respeet to
1} = y/tl. (11.21)
If this is to reduce to a total differential equation in J as a function of 1}, as it must
if the solution is of the assumed form, then one must have

(11.22)

and so
tl. 2 cx vx/uo + const. (11.23)
It is convenient to choose the constant of proportionality and the origin of x so
that
(11.24)
 ~
;:
::s
1.0 f- IS-
q
t--
f' .....I<" ~
~
;;:j
s:::.
0 .81- /. --I ::s
I:l..
::tI
~
11
i:l
Il. ...~- a
0.6f- e... / --i ~
"ts
~.

11. X
II
0.4r- 9'
0 9.5 x 10'
.3.0 x 10'
x 1.1 x 10·

0.2
I ~

11 Y(Il. IIIx) '"

Figure 11.2 Theoretical Blasius profile and experimental confirmation from Refs. [96) and [162J. o
ut
-
106 Physical Fluid Dynamics

It is found experimentally that this choice of the origin of x corresponds fairly

elosely to the leading edge of a flat plate set up in an otherwise unobstructed flow.
Equation (11.24) is essentially the same result as equation (8.l3).
Equation (11.20) now becomes
[[" +2['" =0 (11.25)
The boundary conditions transform to
[=1'=0 at 'Y}=O
(11.26)
[' ~ 1 as 'Y} ~ oo
The salutian of this total differential equation has to be abtained numerically [21,
22] . The resulting variatian of [' with 'Y}, and so the velocity profIle is shown in
Fig. 11.2. This curve is known as the Blasius profIle.
It has the property that
I' = 0.99 when 'Y} =4.99 (11.27)
The boundary layer thickness as previously defined (Section 8.3) is thus
Ö= 4.99(vx/UO)1/2 (11.28)
Other ways of writing this are
[j/x =4.99 Re;1/2 and Re/} =4.99 Rei/2 (11.29)
(Re x = uox/v; Re/) = uoö/v). The boundary layer thickness is small when the
Reynolds number is large, as expected. This is, of course, a necessary condition for
the theory to apply. Also Re/} is large when Re x is large; there is no ambiguity in
talking about large Reynolds number.
Figure 11.2 ineludes experimental observations for several values of Re x (from
two separate experiments). The agreement with the theoretieal profIle is good,
providing support for the variaus approximations and assumptions made in the
course of the theory. The experimental resuits have been scaled to the co-ordinates
'Y} (= Y(UO/VX)1/2) andi' (= u/uo). One sees the way in which the profIle maintains
its shape with distanee downstream although the boundary layer thickness is
changing - as assumed in equation (11.l7).
At higher values of the Reynolds number, the Blasius profIle is unstable and the
boundary layer becomes turbulent. The transition process will be described in
Chapter 19, and the nature of the turbulent boundary layer in Chapter 22. The
instability depends on Re/} , which, as we see from equation (11.29), increases with
Re x ' Thus, any zero pressure gradient boundary layer undergoes transition if it
extends far enough dawnstream. However, provided that the disturbance level is
not too high, the range in which the Reynolds number is high enough for boundary
layer formation but low enough for laminar flow is significant.

11.4 Boundary layer separation

We shall not enter into a discussion of the very large amount of work that has been
done to extend the ideas of the last sectian to less simple distributions of Uo (x)
(and to three-dimensional and unsteady boundary layers). There is, however, an
important phenomenon, known as baundary layer separatian, which we must
Boundary Layers and Related Topics 107

consider physically, if not mathematically. It can occur when the pressure is

increasing with distance downstream; that is when
dpo/dx >0 (11.30)
and so
duo/dx <0 (11.31)
A pressure gradient of this sort is called adverse (the opposite is called favourable).
The phenomenon may thus occur, for example, on the walls of a diverging channel
and on the rear part of an obstacle place d in a flow.
In an adverse pressure gradient, all fluid particles are losing kinetic energy as
they travel downstream. Those close to the wall have less energy initially than those
in the free stream as a resuit of viscous action. The pressure gradient can thus bring
them to rest at a distance downstream where Uo is stili appreciable. Further down-
stream the pressure gradient pushes fluid close to the wall in the opposite direction
from the mainstream. The resuit is a sequence of veloeity profiles of the general
form shown in Fig. 11.3(a). The corresponding pattem of streamlines is shown in
Fig. 11.3(b). The point marked S in each figure is known as the separation point.
Figure 11.4 is a photograph of two streamlines (streaklines in steady flow) in a
region of separation, for comparison with Fig. 11.3(b). The streamlines originate at
dye sources close to the wall on either side of the separation point.
If the wall has a sharp comer, separation will normally occur there. Otherwise, it
occurs at a position without speeial characteristics; the wall on which the separation
was occurring in Fig. 11.4 was flat, the adverse pressure gradient being produced by a
protuberance some distance downstream (out of the picture to the right).
Both laminar and turbulent boundary layers can separate. (In the latter case, the
veloeity profiles and streamlines in Fig. 11.3 must be interpreted as means.) l..aminar
layers usually (depending on the initial velocity profile) require only a relatively
short region of adverse pressure gradient to produce separation. Turbulent layers
separate less readily.

(a)

s
Figure 11.3 (a) VeJocity profiles during boundary Jayer separation; (b) corresponding
streamline pattem.
108 Physical F/uid Dynamics

Figure 11.4 Streamlines in separating flow.

The overall flow pattem, when separation occurs, depends greatly on the
particular flow. The upstream flow behind the separation point is normally fed by
recirculation of the separating fluid, but little can be said in general beyond this. In
some cases of laminar separation, the consequences are only locaI. The separated
boundary layer is much more unstable than the previous attached one and
transition to turbulence may occur very quickly after separation. The turbulent
region may then reattach to the wall (for reasons to be discussed in Section 22.7).
The result is a mean streamIine pattem of the form shown in Fig. 11.5. Downstream
the boundary layer is turbulent. We saw in Chapter 3 (Fig. 3.12) that there is a
Reynolds number range - 3 x 105 to 3 X 10 6 - in which this sequence of events
occurs on a circular cylinder.
When 10caI reattachment does not occur, the consequences of separation are
much more extensive. Chapter 3 has again provided an example - the formation of
attached eddies at all values of the Reynolds number above about 4 (although at
the lower Reynolds number end it is hardly a boundary layer that is separating).
Figure 11.6 shows another example - the flow past an aerofoil (see Chapter 12)
sharply inclined to the flow direction. Separation occurs quickly on the upper side,
as shown by the dye streaks. Transition to turbulence follows, but does not
produce immediate reattachment. Instead there is a highly fluctuating recirculating
flow over the whole of the top of the aerofoil. The turbulent wake has a width
about the same as the projected width of the aerofoil across the flow.
In cases like these the region of rotational flow is not confined to thin layers
next to boundaries pIus a thin wake. Vorticity introduced in the boundary layers is

Figure 11.5 Mean streamline pattem in laminar separation followed by turbulent reattachment.
Boundary Layers and Related Topics 109

Figure 11.6 Flow over an aerofoil at large angle of attack (view is slightly oblique from above
and front).

earried by the separated flow into regions that one would suppose to be irrotational
in obtaining the inviseid flow solution. The presenee of separation thus signifkantly
modifies the inviseid flow, and the assumption that the inviseid flow ean be
analysed without referenee to the boundary layers (exeept as a reason for ignoring
the no-slip eondition) breaks down. As an example, eonsider the observation that,
over a wide Reynolds number range, separation on a eireular eylinder oeeurs
upstream of the maximum width (Fig. 3.l2). On the inviseid flow solution for flow
past a eireular eylinder this is a region of favourable pressure gradient (Fig. 11.7).
lnviseid flow is stili oeeurring outside the boundary layer and wake, but not that
particular inviseid flow solution.

11.5 Drag on bluff bodies

Figure 11.7 indudes ameasured pressure distribution over the surfaee of a eylinder.
It ean be seen that, as aresult of the boundary layer separation, the rear region
(around 8 = 180°) is at a mueh lower pressure than indieated by the inviseid flow
solution. The difference between this pressure and that at the front (around 8 = O°)
produees a drag on the eylinder. At high Reynolds numbers, this drag is mueh
larger than the one produeed by the viseous stress over the surfaee. Thus for bluff
bodies, such as a eireular eylinder, the inapplieability of d' Alembert's 'paradox'
arises prirnarily from the indireet effeet of viseous action on the pressure distribu-
tion and only seeondarily from direet viseous stresses.
Over a wide range of high Reynolds number, the drag eoeffieient for a bluff
body varies little. One example is Fig. 3.l4, and it is a useful general rule in a
variety of situations that the drag is roughly proportional to the square of the
speed. We have seen in Seetion 7.3 (equation (7.21) that eonstaney of the drag
eoeffieient is predieted by dimensional eonsiderations if one assumes that the
viseosity is not a relevant parameter. At first sight, d'Alembert's paradox renders
this result useless. However, we ean now see how it relates to observation. Over the
Reynolds number range concerned, the separation point moves little. The pressure
distribution,like that shown in Fig. 11.7, is thus always mueh the same. The
pressure differenees are proportional to the square of the speed, and, sinee these
generate most of the drag, that has the same proportionality. The drag eoeffieient
is not eompletely eonstant beeause of small ehanges in the pressure distribution and
110 Physical Fluid Dynamics

/
/
/
/
I
Or------T~--------------------------~/-------I
\ I
Un \ I
\ I
\ I
\ I
\
\ I
\ I
\ I
\ I
I
\ I
-2 \ I
\ I
\ I
I
\ I
\ /
" ~~/_____ L_ _ _ _ _ _~_ _ _ _~
_3L------L______~____
o 30 60 90 120 150 180
o
Figure 11.7 Pressure distribution on a eireular eylinder. Broken line: inviscid flow solution.
Solid line: distribution measured at Re = 1.9 x 10' (data obtained by Flaehsbart, given in Ref.
(8)).

beeause of the small eontribution of the viseous stress. When the separation point
moves substantially, as when the boundary layer beeomes turbulent, the pressure
distribution is ehanged markedly and equation (7.21) breaks down eompletely - as
already diseussed in Seetion 3.4.

11.6 Sneamlining

The faet that the drag on an obstacle at high Reynolds number is primariIy due to
the redistribution of the pressure from the inviscid flow distribution, as aresult of
separation, has important praetieal implieations. If separation ean be delayed till
near the rear, or eliminated altogether exeept right at the rear where boundary
layers from opposite sides meet, the drag will be very low; the pressure forees on the
front and rear will almost balanee. This ean be aehieved with two-dimensional and
axisymmetrie shapes sueh as that shown in Fig. 11.8. Sueh shapes are said to be
streamlined. The drag on a two-dimensional streamlined body ean be as low as
1/15 of that on the eylinder of the same thickness. The most important feature is
the slowly tapering tall. This is the reason why streamlining of a railway engine with
a train behind it makes only marginaI differenee. On the other hand, it is essential

Figure 11.8 Streamlined body.

Boundary Layers and Related Topics 111

to their perfonnance that the wings and fuselage of an aeroplane and the parts of a
submarine should have streamlined profiles. It is also apparent that Fig. 11.8
resembles the shape of many marine creatures - fish, dolphins, etc.

11.7 Wakes

The fluid that has become rotational in a boundary layer retains this property as it
travels downstream. The flow past an obstacle thus departs from irrotational theory
not onlyaround the obstacle but also behind it. The latter region of rotational flow
is called the wake.
Behind a flat plate or a strearniined body the initial width of this wake is the
thickness of the boundary layers at the downstream end of the body (Fig. 11.9).
Behind a body on which separation occurs the initial wake width is governed by the
separation process and is much more related to the width of the body itself rather
than to the boundary layer thickness. Beyond these remarks, the structure of the
wake immediately behind the obstacle (the near-wake) depends too much on the
details of the particular obstacle for any general discussion to be useful.
Mter some distance, the velocity profile always develops into the general shape
shown in Fig. 3.13, regardless of the geometry of the obstacle. As we saw in Section
3.4 the reduction of momentum transport is related to the drag. The properties of
far-wakes can be considered in terms of the properties of this profile. Many of the
features of the flow behind cylinders described in Chapter 3 are observed also in the
wakes of other obstacles; as an example, Fig. 11.10 shows a vortex street in the
wake of a flat plate.
When the Reynolds number is high, a wake is long and thin - for the same
reason as before that the fluid travels a long distanee downstream whilst the
vorticity diffuses only a small distance sideways. The boundary layer equations can
thus be applied to wakes - a fact that we shall use when considering turbulent
wakes in Section 22.4. The solution for steady laminar flow will not be considered;
the proeedure is similar to that for jets described below [22].

Figure 11.9 Boundary layer and wake veloeity proflles at tralling edge of streamlined body
(sehematie).
112 Physical Fluid Dynamics

Figure 11.10 Vortex street behind fiat plate at zero incidence with Reynolds number based
on plate length of 6600. Flow visualization is combination of dye released from plate and
aluminium powder suspended in fiuid. From Ref. [246).

11.8 Jets

Ajet is produeed when fluid is ejeeted from an orifiee. We are eoneerned here with
the ease when the ambient fluid into whieh the jet emerges is the same fluid as the
jet itself. Beeause of their simpIieity, the geometries that have been most investi-
gated are a eireular orifiee giving an axisymmetric jet and a long thin sIit giving a
two-dimensional jet. The former is obviously the easier experimental eonfiguration,
but experiments with the latter have been performed; eare is needed to make the
jet neady uniform along its length.
At low Reynolds numbers, the fluid from an orifiee spreads out in all direetions.
At high Reynolds numbers, with whieh we are eoneerned here, ajet, Iike a wake, is
long and thin; the equations of motion may be used in the form of the boundary
layer approximation.
Jets beeome unstable at too low a Reynolds number for the laminar flow solution
of these equations to be observed. We wi11, however, take a look at the solution for
a two-dimensional jet for three reasons: it is interesting to see a solution like the
Blasius solution but not requiring numerieal analysis; some results for turbulent jets
(Seetion 22.2) may usefuUy be eompared with a larninar jet; and in Section 11.9,
we can illustrate certain basie properties of viseous flows in this context. Nor is the
result wholly academic; it can be a starting point for investigations of the instability.
What actually happens in high Reynolds number jet flows will be considered in
Sectian 19.2.
The equations for a laminar two-dimensional jet are just as for the Blasius
boundary layer
au av
-+-=0 (11.32)
ax ay
au au
u - + v - =v -
a2u (11.33)
ax ay ay2
(with x in the jet direction andy across the sIit from which it emerges). There is no
Boundary Layers and Related Topics 113

pressure gradient for reasons paralleI to those considered in Section 11.2; on the
boundary layer approximation the effective pressure gradient is that imposed from
outside and this is zero sinee the ambient fluid is at rest. The ejection of the fluid
from the orifiee will normally have been achieVed by maintaining a higher pressure
behind it. This will provide the fluid with its momentum, but the pressure variations
do not extend significandy downstream of the orifice.
The differenee from the boundary layer solution Hes in the boundary conditions,
which are
u-+O as y-+oo
au/ay = 0 and v=0 at y =0 (11.34)
expressing the facts that the ambient fluid is at rest and the jet is symmetrical about
its centre plane. (See also footnote to Section 11.9, p. 116.)
The procedure for solution c10sely follows that for the boundary layer. It is
again assumed that the velocity proflle is a similar shape at different distances down-
stream. However, the velocity scale, as well as the jet width, can now vary with x;
the form of successive velocity proflles is shown schematically in Fig. 11.11. Hence,
the velocity profile is taken to be of the form
u = umaxg(y/~) = u max g(17) (11.35)
where both the maximum speed u max and ~ (proportional to the jet-width li) are
functions of x. Drawing on our experience with the boundary layer solution we
take them to be proportional to powers of x:
(11.36)
One relationship between m and n is obtained from the requirement that the
inertia te rm and the viscous term must vary in the same way with x.
uau/axcx:x 2m - 1; va 2 u/ay 2 cx:x m - 2n (11.37)
and so
2m-1=m-2n (11.38)

Figure 11.11 Development of ajet.

114 Physical Fluid Dynamics

(This is entirely paralleI to the derivation for the boundary layer that .:l a:: x 1/2, but
presented in a somewhat different way.)
A second relationship is obtained from the fact that the to tal momentum per
unit time passing every cross-section of the jet must be the same. This is a statement
of momentum conservation for the jet as whole, which must apply since the jet does
not exert any force on any external body. (Its relationship to equation (11.33) for
the momentum of individual fluid partides will be discussed in Section 11.9.) Since
the jet is two-dimensional we consider the momentum per unit lengtlMn the
z-direction. The momentum per unit volume is pu and the rate at which it is being
transported downstrearn is u. Hence, the totaI rate of momentum transport at any
x is
(11.39)

Mis conveniently used as a measure of the strength of the jet. Conservation of

momentum indicates that
dM/dx =0 (11.40)

f:
Substituting (11.35) in (11.39),

M= PU~ax.:l oo
g2d17 a::x 2m + n (11.41 )

Hence,
2m+n=0 (11.42)
From equations (11.38) and (11.42) together we have
1 2
m=-- n=- (11.43)
3' 3
The ways in which the maximum velocity and the jet-width vary with x have been
deterrnined.
The pararneters determining the velocity profile at any x are x itself, the jet
momentum, and the properties of the fluid:
u max =umax(x, M, p, Il); .:l =.:l(x, M, p, Il) (11.44)
DimensionaI anaIysis together with the power laws aIready deterrnined indicates
that

(11.45)
where Cis a numericaI constant (there is no need for a similar numerical constant in
the expression for .:l, as its definition was arbitrary to that extent).
Equation (11.32) is satisfied by deriving u and v from a stream function 1/1. The
form appropriate to equations (11.35) and (11.45) is
1/1 = C(MllxjP)1/3f (11.46)
with
t=g (11.47)
Boundary Layers and Related Topics 115

Substituting in equation (11.33) gives

~f''' +JJ" +f'2 =0 (11.48)

c
with the boundary conditions
J=J"=O at l1=O (11.49)
f' ~O aSl1~oo

This can be integrated directly to give

J= A tanh(A Cl1/6) (11.50)

and so

(11.51)

A appears as an arbitrary constant, but substitution in equation (11.39) indicates

that
(AC)3 = 9/2 (11.52)
Hence the solution for the velocity proflle is

u C~;;vx f3 2[Y( 48P~X2 f/3]

= sech (11.53)

Although, as remarked before, this flow is not observable because of instability,

it may be supposed that this result is to be interpreted in the following way.
u max =oo at x =0, and the solution will not be valid in this region. A real jet starts
off with a finite velocity but a velocity proflle different from that given by the
theory. As it travels downstream, the proflle tends to the theoretical one, which
will occur asymptotically with an origin of x displaced from the jet orifice (either
upstream or downstream depending on the initial proflle). The solution has the
property that

d oo
dx J_oo pudy > 0 (11.54)

Hence, the amount of fluid being transported downstream by the jet increases with
distarlce downstream. The jet draws fluid into itself from the sides - a process
known as entrainment. Far enough downstream, only a small proportion of the
fluid in the jet is fluid that came out of the orifice. We shall see in Section 11.9 that
this result follows from rather general considerations and is thus not a property of
the particular solution.
It is for this reason that M, and not the rate of efflux from the orifiee, is used as
a measure of the jet strength.
The reader may have notieed that no boundary condition on v at Iarge y was
applied. Onee one knows about the entrainment the reason is apparent; such a
boundary condition would overconstrain the system.
116 Physical Fluid Dynamics
11.9 Momentum and energy in viscous flow

The foregoing theory of a laminar jet provides a convenient context for a few
further re marks about the basic properties of the equations of viscous flow.
Equation (11.40) was established by applying the principle of momentum
conservation to the jet as a whole, a different type of procedure from those
used so far. Although the equation follows directly from the laws of mechanies,
one would expect it to be related to the equation for the momentum of a fluid
partide and we now see how it is derived from this.
dM
-=-
d Joo
pu 2 dy=2PJ
oo au
u-dy (11.55)
dx dx -oo -oo ax
From equation (11.33)

J oo
u -au dy + f oo
v -au dy = V J a u dy = v [ -au ]
oo
2
-2
oo
=0 (11.56)
-oo ax -oo ay -oo ay ay -oo
since the velocity gradient is zero far outside the jet. Also
au Joo av Joo au
J a dy= [uv):oo - u-dy=O+ u-dy
OO
v- (11.57)
-oo y -oo ay -oo ax
using continuity. The two terms on the left·hand side of equation (11.56) are thus
equal to one another, and so
dM/dx =0 (11.58)
This result is thus shown to be an integrated form of (an approximation to)
the Navier-Stokes equation.t Such integrated forms are used in a variety of con-
texts. In general (e.g. when the boundary layer approximation does not apply or
when the re is a boundary on which the flow exerts a foree), they are more compli-
cated, but we shall not consider the general case in this book.
It is interesting to discuss kinetic energy similarly. An equation for the energy of
a fluid partide is obtained in general by taking the scalar product of the Navier-
Stokes equation with u. In a case such as the jet (where the two-dimensional
boundary layer approximation applies and the re is no pressure gradient), onlyone
velocity component contributes significantly to the energy; the equation is obtained
by multiplying equation (11.33) by u:
..!.u a(u 2 ) +..!.v õ(u 2 ) =vu õ2u =_V(ÕU)2 + ..!.}2(U 2)
2 ax 2 ay ay 2 ay 2 õy2 (11.59)

The left-hand side is the rate of change of kinetic energy (per unit mass) following
a fluid partide. The right-hand side has two parts: one is essentially negative and so
represents dissipation of energy; the other can have either sign and has the nature of
a diffusion of energy down the energy gradient.

tIt may be thought surprising that the Iaminar jet solution uses both equation (11.33) and an
integrated form of it. The extra information involved in the latter is the boundary condition
au/ay ->- 0 asy ->- oo. The solution, equation (11.53), satisfies this, although it was not used
explicitly. Equation (11.40) could be omitted if this condition were used explicitly, but the
procedure adopted much simplifies the determination of the exponents m and n.
Boundary Layers and Related Topics 117

We can learn more from the integrated form. The rate of transport of kinetie
energy in the x-direction by the jet is
1
E=-p Joo
u 3dy (11.60)
2 -oo

By a procedure similar to that for dM/dx, it can be shown that

de
- = -11
dx
J (au)2
-ay dy
oo

-oo
(11.61)

and thus is negative, as could have been antieipated.

Alternatively the argument can be turned round to show that since dE/dx must
be negative, 11 must be positive. A fluid with a negative viscosity coeffieient could
gain mechanieal energy from heat in a way that would violate the second law of
thermodynamics.
The discussion has been formulated in terms of a two-dimensional jet, but, from
their physical significance, we expect the results to apply generally. For ajet of
arbitrary cross-section, the momentum and energy transports are

M= J pu 2 dS, E=~JpU3dS (11.62)

dS is an element of cross-sectional area perpendicular to the main flow direction,

and the integrations are carried out over the whole cross-section.M and E are respec-
tively constant and decreasing with distance downstream.
These results enable us to make some general remarks about the action of
viscosity on momentum and energyo The jet illustrates these points weil because the
complicating effects of boundaries and of a pressure gradient are absent. Viscous
action tends towards equalizing the momentum of different fluid partides. This is a
redistribution of the momentum which conserves the total. The effect on the
kinetic energy consists partly of a dissipation that reduces the energy at every point
and partly of a redistribution, similar to the momentum redistribution, with no net
effect. Some parts of the fluid can be gaining kinetic energy by viscous action, the
redistribution more than counteracting the dissipation, but on average there is aloss.
The situation is analogous to the collision of inelastic bodies: conservation of total
momentum together with a redistribution such that the momentum of different
bodies is more nearly equal results in a loss of total kinetic energy.
From the facts that

(11.63)

and

(11.64)

one may infer that

!f udS>O (11.65)
118 Physical Fluid Dynamics

TIris is a rigorous argument if the profIles are similar at different x (as in equation
(11.35)), a plausibility argument otherwise. Henee, the fact that ajet entrains
ambient fluid is a eonsequenee of the combination of momentumconservation and
energy dissipation.
12
Lift

12.1 Introduction

We have seen that the drag on an obstacle place d in a flow arises essentially from
viseous action (Seetions 10.5, 11.5 and 11.6), whereas the generation of lift on a
suitably shaped obstaele ean be understood, at least in a general sense, from inviscid
theory (Seetion 10.6). In faet viseous action also enters into the proeess of lift
generation in an essential, although more subtle, way. We now look at the reasons
for and eonsequenees of this.
Lift ean be obtained either by appropriate asymmetrical shaping as diseussed in
Seetion 10.6, or by having asymmetrical body inelined to the flow direction, or, of
eourse, by a eombination of asymmetry and inelination (Fig. 12.1).
The eurvature of the eentre line in Fig. 12.1(a) and (d) is ealled the eamber. The
inelination angle a is ealled the angle of attack. The simplest geometry giving lift is
an inelined flat plate (Fig. 12.1(b)). However, boundary layer separation at the
sharp leading edge results in a 'stalled' flow eonfiguration (see below) exeept for
very small values of a and this does not provide a good context for a diseussion of
the important ideas. We turn instead to those shapes (known as aerofoils) that have
been developed speeially to produee high lift; our understanding of the subjeet has
been strongly influeneed by the obvious applieation to aireraft wings. An aerofoil
is a slender body with a rounded leading edge and a sharp trailing edge (Fig. 12.1(a),
(e) and (d)); we shall see that these features help to provide a good lifting behaviour
as weIl as helping to reduee drag (Seetion 11.6). A well-designed aerofoil ean pro-
duee a mueh larger lift than drag, as is illustrated by Fig. 12.2.
It will be eonvenient in the following diseussion to use the standard names for
the different dimensions of an aerofoil, as shown in Fig. 12.3. It will alSO be eon-
venient to talk of the top and bottom of the aerofoil with the lift thought of as
upwards, although the direction of the vertieal does not enter into the dynarnies; in

(al (bl (el (dl

FiJUre 12_1 Examples of lifting conIJgUrations.

120 Physical Fluid Dynamics
1.4 0.28

X/X~
1.2
/
X
X", 0 .24

1.0 / 0 .20
X
X~
I
X CL

/
0.8 0.16

X
CL 0.6
/ •
0.12 CD

/
X

0.4 /
X
0.08

0.2 X
/ • 0.04

.-.-.-.-.-.-./
./

0 0
X

- 0 .2 - 0 .04
0 4 8 12 16 20 24
0:

Figure 12.2 Example of variation with angle of attack of lift coefficient and drag coefficient
for an aerofoil. Note different seales for CD and CL' (Aerofoil type RAF 34 at Re based on
chord of 4.5 x 10· ; measurements by Relf, Jones, and Bell, given in Ref. [8 J.)

general, a lift is any force perpendicular to the direction of relative motion between
a bodyand a fluid.

12.2 Two.<fimensional aerofoils

The first part of the following discussion applies to strictly two-dimensional flow.
This should obviously be relevant to the motion around a body, such as a wing, of
large span. However, the re are complications that arise in any three-dimensional
system, which we must consider later.
The explanation of lift generation in terrns of Bernoulli's equation, given in
Section 10.6, implies that the circulation
r = ,u"dl (12.1 )
around the aerofoil is non-zero. To see this, we may consider a thin (thickness ~
chord) aerofoil inclined at a small angle to the flow direction, so that nearly every
point of the surface is nearly paralleI to the flow direction (Fig. 12.4). The upward
Lift 121

Figure 12.3 Aerofoil terminology.

force (per unit length in the spanwise direction) on the element dx is (PB - PT) dx,
where PB and PT are the pressures below and above the aerofoil at this station
along the chord. Bernoulli's equation gives
PB - PT = ~p(u} - u1) = ~P(UT + UB)(UT - UB) (12.2)
For a thin aerofoil, the variations of the velocity from the free-stream velocity Uo
will be small, and we may approximate this by
PB - PT = PUO(UT - UB) (12.3)
Hence the totallift per unit span,

(12.4)

The circulation round a contour following the surface of the aerofoil (physically
round a contour just outside the boundary layer) may be approximated by

(12.5)

Thus
L = -puor (12.6)
r is the circulation round any loop enclosing the aerofoil (since the circulation
1u·dl = Jw-dS round any loop not enclosing it is zero for irrotational flow) .

•
y

__- - - e - - -..........
Figure 12.4 Definition sketch for discussion of lift and circulation of a thin aerofoil.
122 Physical Fluid Dynamics

----------------------------

(a)
~ (b)

Figure 12.5 Inviscid flow around aerofoil (sehematie): (a) flow without eireulation; (b) flow
with eireulation, in whieh rear stagnation point is located by boundary layer separation.

This is an example of a general exaet resuIt of inviscid irrotational flow theory,

the Kutta-Zhukovksii (or Joukovski) theorem. The foree per unit length on any
two-dimensional body in relative motion to the ambient fluid with velocity Uo and
with cireulation r round it is
(12.7)
aeting at right angles to the direction of uo. (The sign eorresponds to the eonven-
tion that Uo is positive when the fluid moves past a stationary obstacle from left to
right or the obstacle moves through stationary ambient fluid from right to left;
r is positive when the cireulation is anticloekwise; and L is positive upwards.)
This mueh ean be said without referenee to viseosity. However, the generation of
the cireulation in the first place requires viseous action. This is apparent from the
Kelvin cireulation theorem (Seetion 10.2) if we eonsider an aerofoil started from
rest in stationary fluid - so that there is no initial eireulation. There is always a
solution of the inviscid flow equations with no cireulation. For a eambered and/or
inelined aerofoil that would normally generate lift, this solution has the rear stag-
nation point some distanee ahead of the trailing edge on the upper surfaee
(Fig. 12.S(a)). This is the flow patternt whieh aetually oeeurs immediately after
motion starts and which would neeessarily oceur at all subsequent times in a truly
inviscid fluid. However, boundary layer processes ehange it. The boundary layer on
the lower side separates at the trailing edge; the eddy so produeed interaets with
the inviscid flow to make its stagnation point eoincide with the separation
point (Fig. 12.5(b)). The new inviscid flow, outside the boundary layers, now
involves eireulation, of an amount determined by the geometry of the aerofoil in
eonjunetion with the requirement that the rear stagnation point is at the traHing
edge.
Any foree on an obstacle must be balaneed by a rate of momentum ehange
integrated over the fluid as a whole. We ean see from the figure that the movement
of the stagnation point does eorrespond to a defleetion of the flow in the ap pro-
priate direction.
The starting process by whieh the cireulation is generated involves diseharge of
vorticity into the fluid. One ean see that this must happen by applying the Kelvin
cireulation theorem to a cireuit enclosing both the initial position and the present
position of the aerofoil; this cireulation must remain zero. An eddy, of opposite
sense to the cireulation round the aerofoil is left at the place from which it started.
In principle, if one ignores further viseous action, this eddy remains there for ever.

tThe pattern is drawn for the frame of referenee in which the aerofoil is at rest, although
the simplest way of producing the effeets deseribed is to start the aerofoil moving.
Lift 123

Figure 12.6 Eddy produced by changing angle of attack of aerofoil. Flow visualization: two
dye lines originating far upstream.

Figure 12.6 shows a demonstration of this effect. Instead of starting the motion,
the aerofoil is fixed in a steady flow but its angle of attack is suddently changed
(actually from one sign to the other to produce a large effect). The resulting change
in lift corresponds to a change in the eirculation and a 'starting eddy' must be
generated as described above. The photograph, taken when the disturbance pro-
duced by the rotation of the aerofoil has travelled some distanee downstream,
shows that the pattem, although somewhat compHcated, is dorninated by an eddy
eirculating in the opposite sense to the change in eirculation round the aerofoil.
(The Reynolds number was comparatively low to avoid complications due to wake
instabili ty .)
We retum now to considering an aerofoil in steady motion. When the angle of
attack becomes large, boundary layer separation occurs on the upper surface elose
to the leading edge, giving a flow pattem like that shown in Fig. 11.6. The flow is
said to have stalled. The pressure in the separated region is less than that below the
aerofoil (around the front stagnation point). There is thus still some lift, but much
less than before stalling; there is also a sharp increase in the drag (Fig. 12.2).

12.3 Three-dimensional aerofoils

The relationship between lift and eirculation gives rise to further complications in
three-dimensional situations, i.e. for an aerofoil of finite span. The existence of
eirculation in combination with Stokes's theorem,
1u'dl =f W'dS (12.8)
implies that there must be vorticity generated in the vicinity of the aerofoil (but
outside the boundary layer); the surface over whieh the right-hand integral is
carried out can extend over an end of the aerofoil as shown in Fig. 12.7. In other
words, the concept of eireulation in an entirely irrotational motion is meaningful
only for a multiply-connected geometry.
The vorticity is generated near the ends of the aerofoil. Because of the pressure
difference, PB - PT, fluid flows around eaeh end from bottom to top. Vortieity is
produced in the boundary layers associated with this motion and is then carried
into the wake by the main-stream, giving the pattem shown in Fig. 12.8.
The fact that div oo == 0 is in prineiple satisfied by the longitudinal vortiees from
124 Physical Fluid Dynamics

1 u .dJ

Figure 12.7 Loop and surface integrals at end of aerofoil

the two ends of an aerofoil extending back to the position where the aerofoil
started and there joining up through the starting vortex, as shown in Fig. 12.8.
However, as they go back a long distance, the vortices become so diffuse and weak,
through the action of viscosity, that there is no appreciable motion associated with
them.
Usually, an aerofoil tapers towards its ends and the circulation drops gradua1ly to
zero; the longitudinaI vorticity in the wake is then more spread out in the spanwise
direction. Lift on a body ofwhich the span is not long compared with the thickness
cannot, of course, be discussed in these terms at all. The detailed dynamics of the
boundary layer must then be considered, the generation of a wake with longitudinal
vorticity again being an important part of the process.

,-7<...L. - - - - - -L2+-f--- - - -- - - - - - - - --
b i
-

\TraiJing
lIortices

/
t::::t#.::======~&t=.====--------
Figure 12.8 Schematic diagram of circulation, trailing vortiees, and starting vortex.

12.4 Spinning bodies

It is well known that a spinning ball is deflected sideways. Studies haye been made
of the lift force generated both on rotating cylinders (the Magnus effect) and
rotating spheres (the Robins effect) [57J.
A cylinder rotating about its axis and moving through a fluid in a direction
perpendicular to its axis experlences a force perpendicular tö both the direction of
motion and the axis. Like the lift on an aerofoil, this phenomenon can be under-
Lift 125

0 .6 r--.----.---,-----,--r---.---....----r---~--.....,

0.5

0.3

0.2

0.1

O~--~~-------r------------~

- 0.1~--~--~----~---L----~_L _____L_ _ _ __L_ _ _ __L____~

o 0.2 0 .4 0.6 0.8 1.0 2 4 6 8 10
Vluo

Figure 12.9 Experimental variation of lift coefficient with relative rotation rate for spinning
sphere. Note change in abscissa seale by factor of 8 at V/u o = 1. Curve is average of data in
Reynolds number range 1.5 x 10 3 -1.1 x 10' from Refs. [57,88,173).

stood through Bernoulli's equation - by working in a frarne of referenee in which

the axis is at rest and so the flow is steady. On the side on which the cylinder moves
in the same direction as the flow, the fluid velocity is increased and so the pressure
is reduced; on the other side the velocity is decreased and the pressure inereased.
There is thus a force acting on the cylinder from the latter to the former. But,
again as with the aerofoil, it is necessary to invoke viscous action to understand why
the rotation of the cylinder should produce a corresponding circulation of fluid.
Indeed, but for the no-slip condition the flow would make no distinction between
a rotating cylinder and a stationary one. At relatively low rotation rates, complica-
tions, like those described below for a sphere, can occur.
A rotating sphere also experiences a force perpendicular to both the rotation
axis and its motion. As for an aerofoil, the three-dimensional problt:m is inherently
more complicated than the two-dimensional. Figure 12.9 shows observations of the
variation of the lift coefficient with the ratio of the circumferential speed of the
sphere (v) to the flow speed (uo). At large V/uo, the behaviour can be understood
qualitatively in the way described for a rotating cylinder, but any quantitative
information must come from experiment.
At low V/uo, the force reverses in direction (although this may not occur for all
Reynolds numbers). This effeet is probably the result of transition to turbulence
126 Physical Fluid Dynamics

• •

Laminar separation Turbulent separation

(a) (b)

Figure 12.10 Sketch of flow pattem (a) produced by asymmetrical boundary layer
separation, contrasted with flow pattem (b) associated with the 'conventional' Robins effect.

occurring in the boundary layer on the side where the relative velocity between the
sphere and the fluid is larger, but not on the other side [88] . Since transition
delays separation (Section 11.4), this will produce a flow of the general form shown
in Fig. 12.1O(a). The resulting upward (in the geometry of the figure) deflection of
the wake produces a downward force on the sphere - an effect of opposite sign
from the 'conventional' Robins effect (Fig. 12.1 O(b ».
13
Thermal Flows: Basic Equations and
Concepts

13.1 Introduction

We tum now from the isothermal flows that we have been considering in previous
chapters to flows in which temperature variations are introduced. The practical
importance of such flows needs little illustration. The use of moving fluids to trans-
port or remove heat is well known, ranging from the eirculation of coolant through
a nuelear reactor to the mounting of a power transistor on a block with cooling fins.
Equallyapparent is the importance of thermal processes in meteorology and other
branches of geophysics; indeed, the ultimate source of energy for almost all atmos-
pheric motions and much of the oceanic eirculation is solar radiation.
In these examples, and in the more fundamental situations to be considered in
this book, the temperature variations are introduced by some process independent
of the flow dynamics. We are not concemed here with flow induced temperature
variations, arising from adiabatic expansion or com pressi on or from viscous dissipa-
tion. (The conditions in which such effects are negligible are considered in Section
5.8 and in the appendix to Chapter 14.) Most commonly, the temperature variations
are introduced through temperature differences between boundaries, or between a
boundary and ambient fluid. OccasionaIly, they are produced by internaI heat
generation, which can arise from a variety of causes such as radioactivity, absorp-
tion of thermal radiation, and release of latent heat as water vapour condenses.
Heating and cooling at either the boundaries or within the fluid results in the tem-
perature being a continuous function of position.
The name convection is given to this general category of flows. The same name
is sometimes used for more speeific processes occurring within such flows; however,
we shall use it in this book only in the general sense. Another name for the general
topic is heat transfer, used particulady of course when the effect of the flow in
transporting heat is of importance.
A closely related topic is that of flow with concentration variations or mass
transfer. We shall not be considering this systematically in this book, but brief
discussion of the analogy will be given in Section 13.5.

13.2 Equations of convection [8,33]

The temperature variations within a convective flow give rise to variations in the
properties of the fluid, in the density and viscosity for example. An analysis inelu-
ding the full effects of these is so complicated that some approximation becomes
128 Physical Fluid Dynamics

essential. The equations are commonly used in a form known as the Boussinesq
approximation. Here, we shall make this approximation without considering just
when it is valid or demonstrating the fact that its different parts are internally se1f-
consis1ent. These matters will be considered in some detail for the particular case of
free convection in the appendix to Chapter 14; this will illustrate the procedure. In
part, the Boussinesq approximation is the counterpart for convection of the incom-
pressibility approximation for the flows considered hitherto.
In the Boussinesq approximation, variations of all fluid properties other than the
density are ignored completely. Variations of the density are ignored except insofar
as they give rise to a gravitational foree. Thus the continuity equation is used in its
constant density form
V·u=O (13.1)
Similarly, pDu/Dt is replaced by Po Du/Dt, with Po constant - chosen, for example,
as the density at one typical position. However, a body force term is included to
allow for the effect of gravity; that is in equation (5.21) one puts
F=pg (13.2)
We may antieipate that the density variations will be important here and write
p =Po + Äp (13.3)
Also the gravitational acceleration is derivable from a potential
g=-V<I> (13.4)
(In most cases g = -gl. and <I> = gz, where z is taken vertically upwards, but this is, of
course, not true on a global seale.) Hence
F = -(Po + Äp)V<I> = -Y'(Po<l» + Äpg (13.5)
Ifwe introduce
P=P+Po<l> (13.6)
the Navier-Stokes equation (5.21) becomes
Du
Po - = - VP + /lV 2 u + Äpg (13.7)
Dt
If Äp = 0, this is the same as the Navier-Stokes equation without a body force
except that P has replaced p. Provided that the pressure does not appear explieitly
in the boundary conditions, this change makes no difference. We have thus verified
the conclusion considered more physically in Section 5.6, that the gravitational
force is unimportant if the density is uniform. One merely has to correct for the
hydrostatic pressure (the difference between p and P) if the actual pressure is
required.
We are now concerned with cases in which Äp =f=. 0, because of the temperature
variations. As a further part of the Boussinesq approximation, the dependence of p
on T is linearized,
Äp =- apoÄT (13.8)
a is the coeffieient of expansion of the fluid. Hence, the Boussinesq dynamical
Thermal Flows: Basic Equations and Concepts 129

equation is
Du 1 2
- = - - Vp + ,,'il u - [Link] (13.9)
Dt p
(where we have reverted to writing p and p for the quantities temporarily denoted
by Po andP).
One requires in addition an equation for the temperature. In the spirit of the
Boussinesq approximation, it is supposed that the fluid has a constant heat capacity
per unit volume, pCp ; then pCpDT/Dt is equal to the rate of heating per unit
volume of a fluid particle. (The choice of Cp, the specific heat at constant pressure,
is physically sensible, since the pressure is not free to respond directly to the heating
process; for a proper justification, however, one needs to look at the detailed nature
of the Boussinesq approximation - see the appendix to Chapter 14.) This heating is
brought about by transfer of heat from neighbouring fluid particles by thermal con-
duction and sometimes also by internal heat generation. The corresponding terms in
the thermal equation are analogous respectively to the viscous term and the body
force te rm in the dynamical equation. The conductive heat flux
H = - k grad T (13.10)
where k is the thermal conductivity of the fluid. Thus
pCpDT/Dt = - div H +J (13.11)
where J is the rat e of internaI heat generation per unit volume. Taking k to be
constan t, equation (13.11 ) may be rewri tten
aT J
-+U'VT=Kv 2 T+- (13.12)
at pCp
where
K = k/pCp (13.13)
K is known as the thermal diffusivity or sometimes as the thermometric conduc-
tivity.
Equations (13.1), (13.9), and (13.12) constitute the basic equations of convec-
tion in the Boussinesq approximation. They are one vector and two scalar equations
for the one vector and two scalar variables u, p and flT. (Since T appears differen-
tiated throughout equation (13.12) this is effectively an equation in flT, thus
matching up with equation (13.9).)
It is useful to have names for the new terrns. The additional term in the dynami-
cal equation, -[Link], is known as the buoyancy forcet (even when flTis negative
and the term represents the tendency for heavy fluid to sink). The two terms on the
right-hand side of equation (13.12) have obvious names: the conduction term and
the heat generation term. u'VT, which represents the transport of heat by the
motion, may be called the advection term. (It is often called the convection term,
but it is convenient to have a separate name from the processes represented by the
equations as a whole.)

tFor a perfect gas et = 1fT, and in texts concerned prirnarily with gases, the buoyancy terrn is
often written -gt:..T/T.
130 Physical Fluid Dynamics

Equation (13.12) requires boundary conditions for the temperature field. The
commonest type specifies the temperature of a boundary wall; the fluid right next
to the wall must then also be at that temperature. There are, however, other types.
As an example, we may mention the case in which the heat transfer through a wall
is speeified; the temperature gradient in the fluid is then fixed at the wall.
It is worth noting that thermal conduction plays an integral role in conveetion.
For example, when heat is introduced into a fluid by heating a boundary wall, the re
is no advection of heat through the boundary; the fluid first gains heat entirely by
conduction, although further from the wall advection may be the prineipal process.
The traditional division of heat transfer processes into radiation, conduction, and
convection is not completely sharp. Whenever convection occurs, it and conduction
become parts of a single process.

13.3 Classification of convective flows

Evidently a wide range of dynamical behaviours is to be expected, depending on the

importance of the buoyancy force relative to the other terms in equation (13.9).
The extreme cases are the least complicated and have been most extensively studied.
One extreme is represented by situations in which the fluid would be at re st in
the absence of the temperature variations; the buoyancy force is the sole cause of
motion. This is known as free conveetion, or sometimes as natural conveetion. A
whole variety of new phenomena arise, and these form the topic of Chapter 14.
The other extreme corresponds to the buoyancy force being negligible. Tempera-
ture variations are introduced into a flow originating from some other cause, but
they are not large enough to modify the flow significantly. This is called forced
conveetion. For example, at high Reynolds number, forced convection will occur if
the buoyancy force is small compared with the inertia foree; that is if
gaeLIeP ~1 (l3.l4)
where L is a length scale, e a temperature difference scale, and U a velocity scale
introduced independently of e. (Antieipating the nomenclature to be introduced in
Section 16.1, relationship (13.14) says that the Richardson number must be low, or
the internal Froude number high, for forced conveetion.) Forced convection is
frequently of importance in heat transfer appHcations. However, no new flow
phenomena arise and the only matter needing consideration is the temperature
distribution resulting from the flow. This will be discussed briefly below.
Intermediate cases are known as mixed conveetion. We shall not be considering
these in general. However, an important special case occurs when the imposed flow
is predominantly horizontal so that the buoyancy force acts perpendicularly to it.
This situation is known as stratified flow and will be considered in Chapter 16.

13.4 Forced convection

In forced conveetion, the velocity field is unaffected by the temperature field and
can be determined in the same ways as before. One e the velocity field is known,
equation (13.l2) determines the temperature field. But, of course, it is frequently
necessary to turn to experiment to get required results. The transfer of information
Thermal Flows: Basic Equations and Concepts 131

from one system to another requires not only dynamieal similarity - Le. that the
flow pattems should be the same - but also thermal similarity - Le. that the
temperature fields should have the same pattems. The former is guaranteed by
equality of the Reynolds number, but the latter gives rise to an extra eondition.
We will confine attention to steady eonveetion without internaI heat generation.
Equation (13.12) beeomes
(13.15)
Analysis of this equation along the lines of the analysis of the Navier-Stokes
equation in Seetion 7.2 shows that there will be thermal similarity when the two
systems have the same value of
Pe = UL/I< (13.16)
Pe is ealled the Peclet number. It may be interpreted as a measure of the relative
importanee of the two terms in equation (13.15), analogously with the interpreta-
tion of the Reynolds number as the ratio of inertia forees to viseous forees:

adveetion of heat
Pe------- (13.l7)
eonduetion of heat
When Pe is small, equation (13.15) approximates to
1<'V 2 T= 0 (13.18)
the flow is having negligible effeet on the temperature distribution, which is deter-
mined by the same equation as in a material at rest. At high Pe, equation (13.15)
approximates at first to
u''VT=O (13.19)
but now eonduetion ean be important in thermal boundary layers for reasons
analogous to the origin of viseous boundary layers.
Full sirnilarity of foreed eonveetion situations requires equality of both Re =
UL/v and Pe = UL/I<. An equivalent statement is that it requires the equality of
both the Reynolds number and
Pr = v/I< (13.20)
This non-dimensional parameter, ealled the Prandtl number, is a property of the
fluid, not of the partieular flow. Henee, there is a restrietion on the transfer of
information from experiments with one fluid to those with another.
The Prandtl number is the ratio of two diffusivities, v being the diffusivity of
momentum and vorticity and I< that of heat. The meaning of this ean be illustrated
by considering irrotational fluid at a uniform temperature entering a pipe of whieh
the walls are maintained at a higher temperature. Both the Reynolds number and
the Ptklet number are taken to be signifieantly above I. We know from Seetion 11.1
that vortieity is introdueed into the flow in a viseous boundary layer that increases
in thiekness with distanee downstream. Similarly, heat spreads into the flow in a
thermal boundary layer of inereasing thiekness, leaving a dfulinishing eore of fluid
that is stiIl effeetively at its initial temperature. Figure 13.1 show s sehematically the
relative thieknesses of the two boundary layers for values of the Prandtl number
signifieantly greater than 1, around 1, and signifieantly less than 1. The relatively
132 Physical Fluid Dynamics

(a) (b) (e)

Figure 13.1 Relative positions of edges of velocity boundary layer (solid lines) and thermal
boundary layer (broken !ines) in pipe flow entry: (a) Pr significantly above 1; (b) Pr - 1; (e) Pr
significan tly below 1.

more rapid diffusion of heat than vortieity as the Prandtl number is deereased ean
be seen.
Gases have values of the Prandtl number around (but a little less than) 1. Most
liquids have values greater than 1 but by widely varying arnounts. Water is typical of
the lowend of the range with Prandtl number (at room temperature) around 6.
In general, the kinematie viseosity varies mueh more widely than the thermal
diffusivity, so the high Prandtl number liquids are very viseous ones. The important
exeeption to these statements is liquid metals with high thermaI diffusivities,
produeed by the free eleetrons, giving low Prandtl numbers; for example mereury
has a value around 0.025.
A quantity of frequent praetieal importanee in eonveetion problems is the heat
transfer through a surfaee into or out of the fluid. If we denote the rate of transfer
per unit area by H, a non-dimensional form of this is
Nu=HL/k9 (13.21)
where k is the thermal eonduetivity and L and 9 are length and temperature
differenee seales (e.g. 9 might be the temperature differenee between the surfaee
under consideration and the ambient fluid). Nu is the Nusselt number. The Nusselt
number may have both loeal and overall meanings depending on whether H is the
loeal heat transfer or an average value over the whole surfaee. We shalllook at this
distinetion in more detail in the context of free eonveetion (Seetion 14.5). For the
moment we are eoneerned prineipaUy with the latter meaning.
For foreed conveetion, dimensional eonsiderations indicate that
Nu = .ttRe, Pr) (13.22)
Neither Re nor pr involves 9. Henee, when all other quantities are being held
eonstant,
(13.23)
the heat transfer is proportional to the imposed temperature exeess. This result,
whieh eomes essentially from the linearity in AT of equation (13 .15), is Newton's
law of eooling. We shall see in Seetion 14.5 that it does not apply to free conveetion;
nor, in general, does it apply to mixed conveetion. The farniliar statement that
Newton's law of eooling applies in a strong draught ean be made more preeise: it
applies when the foreed eonveetion approximation is valid.

13.5 Flow with concentration variations (mass transfer)

We tum now to situations in which, in place of the temperature variations, there is a

variable arnount of some substance mixed with, dissolved in, or otherwise earried by
Thermal Flows: Basic Equations and Concepts l33

the fluid. The variable salinity (salt concentration) of the oceans can be dynamically
important; air pollution studies require understanding of the processes determining
pollutant concentration; chemical engineering processes often involve the mixing of
different substances. The interaction between concentration variations and a veloeity
field is closely analogous to that between temperature variations and a veloeity
field; the two situations are governed by very similar equations. Thus many of the
ideas to be described in this and subsequent chapters can be applied also to mass
transfer probiems.
The amount of a substance carried by a fluid may be expressed by a concentra-
tion c - say, the mass of the substance per unit volume. c is a continuously variable
function of position. The presence of the substanee increases the density by an
amount D.p above its value Po corresponding to c = O. One may take a linear
relationship between concentration and density,
D.p = poacc (l3.24)
where ac is a coeffieient. (poa c = 1 if the substanee is absorbed into the fluid with-
out increase in volume, but that is not necessarily the case. ac can be negative, as in
the case of a lighter gas mixing in the flow of a heavier gas.) Thus the dynamicaI
equation may be taken as
Du
Dt = - p1 Vp + vV u + gacc
2 (13.25)

The distribution of c is determined by its advection by moving fluid partides and

by its diffusion between fluid particIes. Hence,
Dc
- =K V2c (13.26)
Dt c
where K c is a diffusion coeffieient depending both on the fluid and on the diffusing
substanee.
The correspondence between equations (13.25) and (13.26)t and equations
(l3.9) and (IS.12) (with, of course, the continuity equation (13.1) applying to
both systems) allows one to transfer ideas and results between convective systems
and mass transfer systems.
The point at which the analogy is most likely to break down is in the boundary
conditions. Ways of introdueing concentration variations are usually not the
counterparts of the commonest ways of introdueing temperature variations as
considered in Section 13.1. In particular, it is difficult to maintain the concentration
at a fixed value at a boundary, the counterpart of the common fixed temperature
boundary condition. An example of a more likely way of introdueing concentration
variations - one that is more readily achieved in the Iaboratory - is for two meeting
streams to have different concentrations.
Consequently, the mass transfer counterparts of most of the convection problems
considered in this book are not of much interest. The general ideas are, however,
applicable and, in particular, the dassification of convective flows in Section 13.3
has its mass transfer counterpart depending on the relative importance of the gravi-
tationaI term in equation (l3.25). The non-dimensionaI parameters governing mass

t In the absence of chemical reactions, equation (13.26) contains no counterpart to the

internaI heat generation term in (13.12).
134 Physical Fluid Dynamics
transfer problems are analogues of those for conveetion. We may mention in parti-
cular the Schmidt number, V/K c ' which corresponds to the Prandtl number, and
which is a property of the substances involved and not of the flow. One is usually
concerned either with Schmidt number around 1 (the case for one gas mixed in
another) or with very large Schmidt number (most other cases, such as aqueous
solutions of electrolytes or small solid particIes carried by a liquid or gas flow).
One further point about the equations may now be made - partly to relate the
present approach to that sometimes adopted elsewhere. It is sometimes permissible
to ignore mass diffusion and so to approximate equation (l3.26) by
Dc/Dt = 0 (13.27)

Two considerations are involved here: the fact that diffusivities are frequently very
small, that is the Schmidt number is frequently large; and the fact that concentration
differences are often introduced in a way that does not give the diffusivity an essen-
tial role at boundaries. When equation (l3.27) applies, each fluid particle always
consists of the same material and thus conserves its density; that is
Dp/Dt= 0 (l3.28)
Substituting this into the general form of the continuity equation (5.9) reduces
that to
\7' u = 0 (13.29)
without further approximation. This fact is useful because the Boussinesq approxi-
mati on applied to concentration variations does not have so manyaspeets as the
Boussinesq approximation applied to thermal conveetion. There are no counterparts
of the neglected thermodynamic processes, such as viscous dissipation and density
changes in the pressure field (see appendix to Chapter 14). In fact, the only approxi-
mations in using the set of equations (13.25), (13.27), and (13.29) are the neglect
of diffusivity and the assumption that the viscosity does not vary with the concen-
tration (although it is usually necessary to make the further approximation, p = po
in handling equation (13.25».
We shall not be considering specific mass transfer topics in this book. The matter
has been introduced because useful general remarks can be made by analogy with
conveetion. In the somewhat different context of stratified flow (Chapter 16), we
shall be referring again to the fact that temperature variations and concentration
variations have similar effects.
14
Free Convection

14.1 Introduction [33]

A free convection flow is one produced by buoyancy forees. Temperature differ-

ences are introduced, for example through boundaries maintained at different
temperatures, and the consequent density differences induce the motion; hot fluid
tends to rise, cold to fall.
Such flows have already been place d in the context of general thermal flows in
Section 13.3, and the introductory remarks to Chapter 13 have indicated their
importance. Figure 14.1 shows one example of free conveetion, the flow produced
by a horizontal heated cylinder in a large expanse of cooler fluid, illustrating the
complexity of the phenomena that can arise from the simple fact that hot fluid
rises. One sees that the cylinder is surrounded by a boundary layer of hot fluid; from
this a relatively narrow column (or 'plume') ofhot fluid emerges, becoming unstable
and ultimately turbulent as it rises.
In this chapter we shall be examining the basic principles of the study of free
conveetion, and then applying these to several situations, which, because of their
geometrical simplicity, have been extensively investigated.
We shall use the Boussinesq approximation (Section 13.2). (An appendix to this
chapter considers the conditions in which this is valid and Section 14.3 deseribes a
method of extending its applicability - important in geophysical contexts.) Thus
for steady eonveetion without internal heat generation, we are dealing with the
equations

\7. u =0 (14.1)

1
u • Vu = - - Vp + vV 2 u - gallT (14.2)
p

(14.3)

Equations (14.2) and (14.3) must now be considered simultaneously, as both

involve both u and T. The velocity distribution is governed by the temperature
distribution, but the temperature distribution depends through the advection of
heat on the velocity distribution. There is no possibility of determining one inde-
pendently of the other, as could be done in forced conveetion. For this reason, free
convection is hard to treat theoretically and much of our information about it
comes from experimental investigations.
136 Physical Fluid Dynamics

Figure 14.1 Shadowgraph of eonveetion around a heated horizontal eylinder (in position
indicated by dotted line - ef. Fig. 4.16); approximately, dark regions are those of hot fluid.
From Ref. [220].

14.2 The governing non-dimensional parameters

In such experimental investigations one needs to know when dynamical similarity

pertains. No veloeity seale is provided by the specification of a free convection
situation. Henee, one eannot defme a Reynolds number using only parameters
involved in setting the problem up. Instead, a temperature difference scale, E>, will
govern how vigorous the motion is. If equations (14.1) to (14.3) are analysed by
the methods introduced in Chapter 7, it is found that the dynarnical similarity
depends on two non-dimensional parameters:
Gr = gOlE>L 3 Iv 2 (14.4)
pr = vI" (14.5)
These are respectively the Grashof number and the Prandtl number. Only when
both of these are the same in two geometrically similar situations ean the same flow
pattems be expected. As already mentioned in the context of forced eonvection
(Section 13.4), pr is a property of the fluid and the transfer of information from
one fluid to another is restricted. A full investigation of a given situation requires
ranging of the Grashof number for fluids of a variety of values of the Prandtl
number and thus is inherently longer than an investigation of a situation governed
by a single non-dimensional parameter such as the Reynolds number.
Free Conveetian 137

For a given fluid, the Grashof number indicates the type of flow to be expeeted
- whieh dynarnieal processes are dominant, whether the flow is laminar or turbu-
lent, and so on - as the Reynolds number does for foreed flow. The Grashof
number eannot be given a general simple interpretation as the ratio of two dynami-
eal processes. Nevertheless it indieates the relative importanee of inertia and viseous
forees, as the following diseussion shows.
Either the inertia foree OI the viseous foree, or both, must be of the same order
of magnitude as the buoyaney foree. The motion ean reaeh a steady state only when
other terms balanee the buoyaney foree; on the other hand, sinee the buoyaney
foree is the eause of the motion, these other terms eannot beeome large eompared
with it. We suppose in the first place that the inertia foree is of the same order of
magnitude as the buoyaney foree.
1 u • V u 1 ~ 1 gaLlT 1 (14.6)

that is
(14.7)
This enables a velocity seale to be written down indieating typically how fast the
fluid wiil move as a resuit of the temperature variations
U ~ (gaLEl)l/2 (14.8)
We ean now eompare the orders of magnitude of the inertia and viseous forees:
1 u • Vu 1 UL (gaElL 3 )1/2 = Gr ll2
'------' ~ - ~ - - (14.9)
1 VV 2 u 1 V v2
This teils us that, when the Grashof number is large, the viseous for ee is negligible
eompared with the buoyaney and inertia forees (subjeet to qualifieations diseussed
below). On the other hand, it tells us nothing in the ease of small Grashof number
as the apparent predietion that the inertia foree is small is in eontradietion with the
originaI assumption that the inertia foree is eomparab1e with the buoyaney foree.
To deal with that ease, we start with the alternative assumption that the viseous
foree is eomparable with the buoyaney foree
1 vV 2 U 1 ~ 1 gaLlT 1 (14.10)
The eorresponding proeedure gives
. U~gaElL2/v (14.11)
and so
lu·\7ul Gr
'----=----' ~ (14.12)
1 v\72 u 1

This analysis indieates that small Grashof number implies negligible inertia forees,
but is irrelevant to the ease of high Grashof number.
Henee, in general, the Grashof number is a measure of the relafive importanee of
viseous and inertial effeets; but, beeause of the different powers to whieh Gr
appears in relationships (14.9) and (14.12), one eannot write a general expression
for Gr as a ratio of effeets.
In eonveetion probIems, one needs to know not only whieh dynarnieal processes
138 Physical Fluid Dynamics
are important but also which processes are important in determining the tempera-
ture distribution.
Adveetion UL
(14.13)
Conduetion
and so, when Gr is large,
"
I U' VT I/I" \72 T 1- Gr 1 / 2 Pr (14.14)
and, when Gr is small,

(14.15)
When the Prandtl number is around 1 (as it is for gases and for some liquids, see
Seetion 13.4), dominanee of adveetion over eonduetion always oeeurs simuItane-
ously with dominanee of inertial forees over viseous forees. When the Prandtl
number is small (as in liquid metals) or large (as in viseous oils and many other
liquids), this eorrespondenee does not apply.
The quantity
Ra=GrPr=go:E>L 3 jv" (14.16)
appearing in relationship (14.15) plays a speeial role in studies of eonveetion in
horizontallayers (see Chapter 4 and Seetion 14.7) and is ealled the Rayleigh
number.
The quantities UL/v and UL/" appearing in (14.9) and (14.13) ean, of eourse, be
identified as a Reynolds number and a P6clet number. Their role, however, is
different from that in foreed flows, sinee they involve U which is a dependent seale,
not an independent one. Henee, Re (and similarly Pe) is a dependent non-
dimensional parameter and one ean write
Re = [(Gr, Pr) (14.17)
Another important dependent parameter is the Nusselt number, indieating the
heat transfer as for foreed eonveetion (Seetion 13.4). From dimensional analysis
Nu =[(Gr, Pr) (14.18)
The form of this dependenee will be considered further in the context of particular
examples later in this ehapter.
Large values of the Grashof number oeeur mueh more frequently than small
ones. For example, the eomparatively small temperature differenee and length seales
of 1°C and 10-2 m give Gr - 103 in water and -10 2 in air. This is related to the
faet that quite vigorous eonveetion eurrents often arise as aresult of stray tempera-
ture differenees in any large volume of fluid left standing (and ean be a nuisanee if
one is trying to investigate something else!).
When Gr is large (and pr is not too small), relationships (14.9) and (14.14) imply
the dominance of inertia forees over viseous and of adveetion over eonduetion.
However, this is based on the assumption that the only length seale is the imposed
one, L. This assumption will be invalidated by boundary layer formation for
reasons simllar to those explained in Seetion 8.3. (Notice that the eonduetion term
is the highest order differential term in equation (14.3) just as the viseous term is in
the Navier-Stokes equation.) In the commonest situation where the flow is
Free Conveetian 139

produced by maintaining the temperature differences between impermeable

boundaries, thermal conduction is responsible for the introduction of temperature
differences into the fluid. Without thermal conduction the fluid next to a wall
could remain at a different temperature from the wall (a state of affairs analogous
to the theoretieal possibility of a velocity differenee between fluid and wall in the
absence of viscosity). Omitting the action of conductivity would thus remove the
convection problem altogether. The correct inference when Gr ll2 and Gr 112 pr are
large-is that the flow will have a boundary layer character. The flow considered in
Section 14.5 will illustrate more explicitly the nature of such boundary layers.

14.3 The adiabatic temperature gradient

We digress here to consider a matter of little significance for laboratory fluid

dynamies but of importanee in the applieation of laboratory resuIts to almost any
geophysical situation. The matter concerns the effect of the hydrostatic pressure
on the conveetion: as fluid rises, it expands and so cools; as fluid falls, it is com·
pressed and so warrns. This effect is ignored in the Boussinesq approximation, but
it is both useful and possible to make allowance for it. (The point is discussed more
systematically in the appendix to this chapter.)
The matter is not confined to free convection, but is conveniently considered in
this context. Indeed, we consider it first in the context of the particular type of free
conveetion, already introduced in Chapter 4, in which the imposed temperature
differences are vertieal. Considering such a situation on the basis of the Boussinesq
approximation, a first requirement for instability to bring the fluid into motion is
that the temperature should decrease with height. This is inferred by considering a
fluid partide displaced a small distance vertically - upwards, say. Ignoring inter-
actions with other fluid partides, the particle will be lighter than its new environ-
ment when the temperature decreases upwards and heavier when it increases
upwards. In the former case, it will tend to move further upwards and the displace-
ment is amplified; in the latter case, it tends to be restored to its originallevel.
(When interactions, such as heat conduction between partides, are take n into
account, additional requirements for instability come in.)
This argument is valid when the effect of changes in hydrostatic pressure can be
neglected. If it cannot, then the criterion for instability must be modified to allow
for the cooling of an upward moving fluid partide. Only if the partide's new
temperature exceeds that of its newenvironment will it continue to be displaced
upwards. There is a certain vertical temperature distribution for which a fluid
partide moving vertieally always has just the local temperature. For instability,
the temperature must decrease with height more rapidly than this. We denote this
distribution by Ta(z).
To determine it, we again consider a fluid partide displaced vertieally in an
ideal inviscid non-heat-conducting fluid. Thermodynamically irreversible proeesses
and heat transfer between partides involve the action of viscosity and conductivity.
Hence, the displacement occurs at constant entropy, and
dTa =
dz
(aT) dp
ap s dz
(14.19)
140 Physical Fluid Dynamics

The vertieal pressure gradient is produeed by the weight of the fluid. Henee,
dp
-=-gp (14.20)
dz
Also, in eonventional thermodynamie notation (with p = I/V)

and so
(:;t = (::) p = (:;t (:;) p (14.21)

dTa gaTa
-=--- (14.22)
dz Cp
The temperature gradient given by (14.22) is known as the adiabatie gradient, or, in
geophysieal usage, the adiabatie lapse rate.
The eriterion for instability, -aT/az > 0, is thus more eorreetly written
aT gaT
-->-
az Cp (14.23)

The distinetion is frequently important in geophysieal situations, although rarely in

the laboratory. We may illustrate this by considering the adiabatie temperature
gradient for air, whieh, at T'" 300 K is approximately 10 -2 Km -1. A Rayleigh
number of 104 (see Chapter 4) in a layer of 3 x 10-2 ip (typieal in the laboratory)
eorresponds to a temperature gradient of around 10 2 Km -1 whereas over a depth
of 10 m (the small end of atmospherie seales) it eorresponds to 10-8 Km-I. In
natural situations it is thus usually the temperature gradient eriterion (14.23) that
governs whether eonvection oeeurs. The Rayleigh number eriterion (Seetion 4.2)
is rarely relevant.
We saw (Fig. 4.13) that, for situations in whieh the adiabatie gradient is
negligible, high Rayleigh number eonveetion has a negligible temperature gradient
exeept in thin boundary layers. Correspondingly, vigorous eonveetion in natural
situations reduees the temperature gradient to the adiabatie gradient. For this
reason, regions oeeur in whieh the temperature distribution is governed by an
equation of the form of (14.22) - a faet of partieular importanee in the theories of
both stellar and planetary interiors.
The foregoing eonsiderations suggest that problems of eonveetion may usefully
be treated in terms of
() = T - (Ta - To) (14.24)
a quantity known as the potential temperature. This is indeed the ease, not only for
the above type of stability eonsideration but quite generally. When temperature
ehanges due to motion in the hydrostatie pressure field are take n into aeeount, the
equations beeome the same as those derived ignoring such ehanges, provided that T
is replaeed by (). (The justifieation for this statement and the eonditions in whieh it
is true are both indieated in the appendix to this ehapter.) The main body of our
diseussion of free eonveetion is being developed in terms of laboratory situations
for whieh the distinetion between T and () is insignifieant. However, the appliea-
bility of these ideas is greatly extended by the concept of potential temperature.
Beeause of the adiabatic temperature gradient, it is not always satisfaetory to
deseribe flows without buoyaney forees as 'isothermal'. The alternative term is
 Free Conveetian 141

'barotropic'. Flows with buoyaney forees are eonversely deseribed as 'baroelinie'. In

general, a barotropie situation is one in whieh surfaees of eonstant pressure and
surfaees of eonstant density eoineide; a baroelinie situation is one in whieh they
interseet. When the prineipal eause of pressure variation is the hydrostatic effeet,
barotropy implies that eonstant density surfaees are horizontal and thus that
buoyaney forees are absent.

14.4 Free convection as a heat engine

A free eonveetive flow is thermodynamieally a heat engine. Heat enters the fluid at
hot boundaries, is transported by the flow, and leayes it at colder boundaries; during
the transport, forees are generated whieh feed kinetie energy into the flow. The faet
that this energy is usually dissipated again within the fluid, rather than beeoming
available as work done externally, eomplieates the overall thermodynarnies but does
not ehange the eharaeter of the kinetie energy generation process. (One ean do a
'thought experiment' in which the action of viseosity is simulated by a large number
of tiny propellers whieh do deliver external work.) One rnight expeet therefore that
eonsiderations involving the laws of thermodynamies would enter more explieitly
into the analysis of free conveetion. The aim of this seetion is to clarify this point.
From equations (14.1) to (14.3) one ean derive expressions for the rate of
kinetic energy generation and for the rate of heat transfer. In order of magnitude
the ratio of these quantities ist
W/Ql - gcxL/Cp (14.25)
where L is the length seale of the eonveeting system and the notation of the left-
hand side eorresponds to eonventional thermodynarnie us age as in Fig. 14.2. This
ratio is always very small in any situation to whieh the Boussinesq approximation

-t---- w= O2 - O.

O2

T 2 (>T.)

Figure 14.2 Definition diagram for heat engine, drawing heat Q2 from thermal reservoir at
temperature T2 and delivering heat Q. to thermal reservoir at temperature T. and doing
external work JV.

tThe derivation of this result involves methods outside the scope of this book. They are
somewhat similar to those mentioned in Section 11.9. The result is also c10sely related to
equation (14.78).
142 Physical Fluid Dynamics
applies (see appendix); that is such situations have very low 'efficiency',
W~Ql, Qz ~Ql (14.26)
This fact has two consequences. Firstly, the kinetic energy generation is a
negligible perturbation in the first law of thermodynamics. This therefore reduces
to a heat conservation law (strictly an enthalpy conservation law, corresponding
to the appearance of Cp , not Cv, in the equations). Such a law is expressed by an
integrated form of equation (14.3). (Note - this does not mean that mechanieal
effects on the heat transfer are in every way negligible; the heat transfer is usually
much larger than it would be if the fluid remained at rest.)
Secondly, the efficiency is so low compared with the Carnot efficiency,
(Tz - Td/T1, that the constraints imposed by the second law of thermodynamics
are automatically fulfilled and do not need explicit consideration. (It might be
remarked that the efficiency in (14.25) is, in order of magnitude, independent of
(T2 - T1)/T1, and that one could thus envisage a very weak convection produced
by very small temperature differences in which the actual efficiency did apparently
exceed the Carnot efficiency. The resolution of this apparent paradoxJies in the
fact that such temperature differences would be smaller than those associated with
the adiabatic temperature gradient - which, from equation (14.22), have order of
magnitude gexTL/Cp . The existence of the latter provides a Carnot efficiency at
least as large as the actual efficiency.)
Situations to which the Boussinesq approximation does not apply do require
much more explicit thermodynamic formulation than the situations being
considered here.

14.5 Cooveetioo from a heated vertical surfaee

We now start looking at particular free convective flows, bearing in mind the
remarks at the end of Section 14.2 about the importanee and character of high
Grashof number situations. The simplest example of such a flow is that produced
by an isolated heated vertical plate. This may be thought of as the counterpart in
free convection of the Blasius boundary layer (Section 11.3) in forced flow.
The plate, maintained at a constant uniform temperature T 1 is on one side of a
large expanse of fluid of which the temperature is To far from the plate (Fig. 14.3).
The distanee from the lower edge is denoted by x and the distanee from the plate
by y. The presence of the upper edge affects the flow alongside the plate only very
near the top. x thus plays the role of the imposed length scale in the way that the
distance from the leading edge did in Section 11.3. Above the upper edge the hot
fluid continues to rise in a plume, but it is the flow alongside the plate with which
we are concerned here. We are supposing that the plate is wide enough in the
z-direction for the motion to be considered as two-dimensional.
The fluid next to the plate is heated by thermal conduction. As aresult it rises
up the plate. When the Grashof number, now defined as
Gr = gex(T1 - To)X 3/V 2 (14.27)
is large enough (and considering at the moment a fluid with pr - 1), the speed
generated in this way is large enough that the heat is carried off in the x-
direction before it has penetrated far in the y-direction. The convection occurs
Free Conveetian 143

..r-; / Soundarv
,ayer
I
I
I

, I

I To

L
T,
I
X t
t
I
I
I
....L-.L>-_ _ Y

Figure 14.3 Definition diagram for free eonveetion from a heated vertical surfaee.

entire1y in a tbin boundary 1ayer (Fig. 14.4). Outside this the fluid remains almost
at rest at temperature To. It does drift slowly towards the plate, as the boundary
layer entrains fluid, the amount of fluid moving up the plate increasing with x.
The temperature is T 1 at the wall and To outside the boundary layer. The
velocity is zero both at the wall and outside the boundary layer. Their proftles must

Figure 14.4 Shadowgraph of eonveetion around a heated vertical plate - ef. Fig. 4.16. lnner
bright !ines are produeed by elosest negligibly refraeted light and indicate edge of boundary
layers; outer bright !ines are produeed by strongly refraeted light at plate surface and indicate
temperature gradient there. From Ref. [220).
144 Physical Fluid Dynamics

(a) 3.2 ,

[;
Oo.;)
2.8
r~ .
.)

2.4 #il \ v

e 1\ eP

2.0
\~
\
e
'" .." .)

UX P
--" 1.6
2u(GrI"
~o~

\
I

1.2 c>
0

\0
1\

.8
'\ o 0

----
.4

~
"-
o 1.0 4.0 5.0 6.0

thus be of the general forms shown in Fig. 14.5. 80th the boundary layer thickness
and the maximum velocity vary with x, whilst the temperature difference across the
boundary layer is of course fixed at (T1 - To).
The lines on Fig. 14.5 are actually the solution for Pr = 0.72, on the assumption
that the flow remains laminar. This is obtained by making the boundary layer
approximation to equations (14.1) to (14.3) and then following a procedure closely
similar to that for the Blasius boundary layer described in Section 11.3 (see Refs.
[8, 22]). The boundary layer thickness and velocity scaIes are
~ = x Gr- 1I4 , U= v Gr 1l2jx (14.28)
Thus in a given flow
Ö a: ~ a:x 1l4 , u max a: Ua:x 1l2 (14.29)
and the totaI amount of fluid moving up the plate is proportionaI to X 3/4 • A local
Reynolds number can be de fine d on the maximum velocity and the boundary layer
thickness and this is proportionaI to Gr 114 •
Free Conveetian 145

(bJ 1.2

1.0

.8
,
~
t~

T- To
.6
\~.
T, - To \ t\

.4
I\~
\ <tlJ
D

2
~~
~ D rl.

-......::..
~ r-
o 1.0 4.0 5.0 6.0

Figure 14.5 Velocity (a) and temperature (b) profiles for laminar free convection on a
vertical plate. 0 Grl / 3 ; 20.5, D 41,0 82, <l 143, t> 225. Co-ordinates are scaled to allow
representation by a single curve for all Gr. Experimental data in air (due to Schmidt and
Beckmann) for constant conditions; Gr'/3 is proportional to distance from lower edge. From
Ref. [192).

Similar results ean be obtained for other values of the Prandtl number [192].
When the Prandtl number is large, the velocity boundary layer is thick eompared
with the temperature boundary layer (as in foreed conveetion, Fig. 13.1). The heat
diffuses only slowly away from the wall, but the buoyaney of the thin hot layer
drags mueh more fluid into motion through the action of viseosity. On the other
hand, when the Prandtl number is small, the velocity and temperature boundary
layers are of similar thieknesses (in eontrast to the behaviour in foreed conveetion).
Any fluid that is heated by thermaI eonduetion moves under the action of
buoyaney, even if viseous effeets have not spread that far from the wall [177].
Laminar flow is observed over the lower part of a vertieal heated pIate. In this
region, good agreement is found between the ealeuIations of boundary layer theory
and experimental measurements (with the origin of x taken close to the lower
edge), as is illustrated by the data included in Fig. 14.5.
Further up the plate, the laminar flow beeomes unstable and undergoes a
transition proeess whieh results ultimately in a fully turbulent boundary layer (Fig.
146 Physical Fluid Dynamics

Figure 14.6 Transition in free conveetion boundary Jayer, exhibited by smoke introdueed
through a row of holes close to the lower edge. From Colak-Antic, P. 'Dreidimensionale
Instabilitätserseheinungen des laminarturbulenten Umsehlags bei freier Konvektion längs einer
vertikalen geheizten Platte', Sitzungsberichte der Heidelberger Akademie der Wissensehaften,
Mathem.-naturw. Klasse. Jahrg. 1962/64,6. Abhandlung. Berlin-Heidelberg-New York.
Springer 1964.

14.6) [81,112,244]. It is difficult to give a value for the Grashofnumber at which

the transition starts, for several reasons: it wiU vary with Prandtl number; it is
sensitive to disturbances; and, perhaps most important, the disturbances grow
slowly at fITst, so there may be a considerable length in which the transition process
has started but has not significantly altered the laminar flow. For fluids with pr
around 1, substantiaUy laminar flow can be observed up to a Grashof number of
an order of magnitude of 10 9 , even though it becomes unstable, in the sense to be
explained in Chapter 18, at a Grashofnumber [187] between 10 5 and 106 • (Such
high Grashof numbers do not require an enormous system, since x appears to a high
power in Gr. For the same reason, the difference between 106 and 109 represents
only a factor of 10 in distance up the plate, but it is nevertheless a strilang
difference.)
There are two mechanisms by which this type of boundary layer can become
unstable [187]. One is essentiaUy the same process as initiates transition in other
shear flows (see Section 17.6). Its onset after some distanee up the plate is
associated with the increase of the local Reynolds number. The buoyancy force
Free Conveetion 147

plays no direet role in this instability, although it does, of course, produce the flow
in the first place. The other instability involves the buoyancy force directly; that is
to say, it would not be exhibited by a similar velocity profile produced in a
different way. The buoyancy of a displaeed fluid particle modifies the velocity
field in a way that accentuates the displaeement.
The relationship of the two instability mechanisms to the transition to
turbulence is uncertain. However, the second type of instability starts lower down
the plate (except at very low Pr), but amplifies only slowly, so that probably the
first type actually gives rise to transition [111] (except perhaps at very high pr).
The later stages of transition involve the growth and spreading of turbulent spots
[276] as described for other flows in Chapters 2 and 19.
We shail not be considering the structure of the fully turbulent flow for this
configuration, but it is, of course, investigated by methods like those described in
Chapters 20 and 22.
The heat transfer per unit area from the plate by the combination of conduction
and convection is, of course, a function of the distanee up the plate, H(x). The local
Nusselt number is
NUL = Hx/k(T1 - To) (14.30)
The total heat transfer per unit area from a plate of length 1is

=fo Hdx/l
I
HT (14.31)

and the Nusselt number based on this is

NUT = HTl/k(T1 - To) (14.32)
If the dependence of the heat transfer can be represented by a power law
Nu = C Gr n (14.33)
(corresponding to

(14.34)
in a given flow), the total Nusselt number has a similar power law dependence on
the Grashof number based on the total plate length,

_ C [gCX(T
NuT-- 1- To )l3] n
3n ,,2 (14.35)

In laminar flow, both theory and experiment show a power-law dependenee of

this sort with n = 1/4; C depends on the Prandtl number.
In turbulent flow [76,288] ,a similar power-law representation, usuaily with
n = 1/3 (corresponding to uniform heat-flux), is often used to summarize the
experimental data. In this ease , however, it may be only an approximate
representation of a more complicated dependence.
It is interesting to note the implications of these results for the dependenee of
the total heat transfer from a given plate on the temperature elevation (T1 - To). If
the plate is short enough for the flow to remain laminar over it,
H T cx (T1 - TO)5/4 (14.36)
148 Physical Fluid Dynamics

If the plate is long enough for most of the flow to be turbulent, then approximately
(14.37)
Intermediate cases are more eomplieated beeause the transition region will move as
the temperature elevation is ehanged.
These results are typieal of free eonveetion in general. In eontrast with foreed
conveetion, Newton's law of eooling does not apply. The heat transfer always
inereases more rapidly than direet proportionality to the temperature elevation.
This is beeause of the dual role of the heating. The larger the temperature elevation,
the greater is the rate of heat transfer by a partieular flow rate; but also the larger
the temperature elevation, the greater is the flow rate.

14.6 ThermaI plumes [27,33,280]

In the same sense that the flow considered above is the eounterpart of the Blasius
boundary layer, a thermal plume is the eounterpart of ajet. It is produeed by a
loeal souree of heat with a large expanse of fluid above it. A eolumn of hot fluid
rises above the souree and further fluid is drawn in and heated. When Gr l/2 and
Gr l/2 Pr are large (as is almost always the case), this eolumn or plume extends
vertieally mueh further than it extends horizontally. Readily observable examples
are the plumes over tires and ehimneys.
Just as ajet far enough from its oritiee ean be considered to originate from a
point (or line) souree of momentum, so a plume ean be considered to originate
from a point (or Une) souree of heat (sometimes displaeed vertieally from the
aetual souree). The quantity eonserved with distanee along the plume (analogous to
momentum in ajet, equations (11.39) and (11.40)) is the heat transport;

!J wtlT dS = 0 (14.38)

z being vertieal. The momentum flux increases with distanee up the plume through
the action of the buoyaney foree. The mass flux also inereases; the plume entrains
ambient fluid.
As with jets, laminar plumes are very unstable [112] and it is the turbulent ones
that are of most physieal interest. The top part of Fig. 14.1 is, of eourse, a plume
and a further pieture will be shown in Seetion 22.7 (Fig. 22.23), where we shall be
using plumes to illustrate the Coanda effeet.

14.7 Conveetion in fluid Iayers

Conveetion in thin layers of fluid has already been considered in Chapter 4, with
particular referenee [Link]. The purpose of the present seetion is to
place that ehapter in context and to extend the diseussion to vertieallayers.
A layer of fluid is eonfmed between two parallel plates at different temperatures,
T I and T2 , and a distanee d apart. The other dimensions of the layer are large
eompared with d. It is usual to specify dynamieal similarity in terms of the
Rayleigh number (= GrPr) and Prandtl number. This originates from the speeial role
Free Conveetion 149

of the Rayleigh number when the layer is horizontal, but, in general, it is simply an
equivalent alternative to the Gr and pr speeifieation.
If the length seale Listaken to be d, the Nusselt number
Nu=Hd/k(Tz - T 1 ) (14.39)
has a special interpretation for this partieu1ar geometry. It is the ratio of the aetual
heat transfer through the layer to the heat transfer that wou1d oeeur by eonduetion
alone if the fluid remained at rest. (Figure 4.4 illustrated the use of the inerease of
Nu above 1 as an indication of the onset of motion.)
Conveetion in fluid layers has a special interest as the simplest example of the
faet that a fluid may eome into motion for two alternative reasons. Either there is
no equilibrium eonfiguration (Le. putting u = 0 in the equations does not resu1t in a
balanee of the remaining terms); or there is an equilibrium eonfiguration, but it is
unstable. (There is a third possibility, although it does not arise in the present
context: that there is a stable equilibrium eonfiguration and also a dynamie
solution of the equations, the latter oeeurring if sufficient initial disturbanee is
present.)
When the two plates are horizontal, with T1 the temperature of the upper, then
equations (14.1) to (14.3) are satisfied by
u = 0, T= T z + (T1 - Tz)z/d, P = pga(Tz - Tdz z/2d (14.40)
where z is measured vertieally upwards from the lower plate (and the referenee
density Po used in subtraeting out the hydrostatie pressure has been chosen as the
density at Tz ). When T1 > T z , this solution is always stable and the fluid remains at
rest. When T z > T1 , the stability depends on whether the Rayleigh number is below
or above its critical value. As will be seen in Seetion 18.2, this value is independent
of the Prandtl number, although for all other developments pr is a parameter of the
problem. The flow patterns that resu1t from this instability have been considered in
Seetion 4.3.
When the layer is not horizontal, there is no equilibrium solution. For the
remainder of this seetion we eonsider the layer between two vertieal plates. The
smaller gravitational foree on the fluid elose to the hot wall (Tz , say) eannot be
balaneed by the same vertieal pressure gradient as would balanee the larger foree on
the fluid elose the cold wall; but, with the fluid at rest, there ean be no horizontal
pressure gradient. Mathematieally, if the fluid were at rest, the temperature
distribution would be
(14.41 )
where x is the distanee from the hot wall. The buoyaney foree, gat::.T, then has
non-zero eurl in the other horizontal direetion,ga(Tz - Td/d. (l/p)Vp has zero
eurl and so these two terms eannot be in balanee.
The simplest eonsequenee of this non-equilibrium is a simple cireu1ation of the
sort shown in Fig. 4.2. However, this flow ean break up into more eomplieated
patterns, and we now look briefly at these.
Beeause the same eireulation ean extend from bottom to top as in Fig. 4.2, the
height h of the layer is arelevant parameter even when it is large compared with d.
(The other horizontal dimension should remain unimportant provided it is large
enough.) There are thus three governing non-dimensional parameters, Ra, Pr, and
h/d. Moreover, fewer experiments have been performed for this eonfiguration than
150 Physical Fluid Dynamics

for horizontallayers. It is unlikely that the full range of phenomena has been
observed [115]. Also, a summary of the principal observations must be rather
vague about the quantitative conditions in which the different flows may occur.
The following account is probably most accurate at high Prandtl number (around
103 ) [101,102]. When the Prandtl number is lower (e.g. around 1), the second
development (formation of boundary layers) is probably not to be observed because

()'2

I - ----<r--------/ (a)

ty

x/d

Figure 14.7 Distribution of vertical velocity at midheight across a vertical slot containing
paraffin for various values of the Rayleigh number: (a) 3.1 x 10'; (b) 2.95 X 10 5 ; (e) 6.6 x 10 5 ;
(d) 3.6 x 10 6 • From Ref. [101 J.
Free Convection 151

(f) Ce) Cd) Ce) Cb) Ca)

Figure 14.8 Sketehes of streamlines for eonveetion of silieone oi! of Pr "" 900 in a vertieal
siot for various values of the Rayleigh number: (a) 3.0 x 10 5 ; (b) 3.6 x 10'; (e) 4.0 x 10 5 ;
(d) 4.9 x 10 5 ; (e) 5.8 x 10 5 ; (f) 6.8 x lOs. From Ref. [101).
152 Physical Fluid Dynamics
of prior oeeurrenee of the third development (break up into smaller eonveetion
eells) [286].
At lower values of the Rayleigh number, the flow does have the general
eharaeter shown in Fig. 4.2. At first, all the fluid partieipates in the eireulation. As
the Rayleigh number is raised, however, the motion beeomes inereasingly eoneen-
trated in boundary layers on the hot and cold walls, leaving almost stationary fluid
in the central region. Figure 14.7 illustrates this with a series of velocity profiles
measured half-way up the apparatus; profiles at other heights are very siruilar. The
temperature in the central region is almost eonstant over eaeh horizontal plane but
is higher at the top of the slot than at the bottom.
Further inerease of the Rayleigh number produees a breakdown of the single
eonvection eell extending the full height of the slot into a number of shorter eelis, as
is illustrated by the sequenee of flow pattems shown sehematiealiy in Fig. 14.8 and
by the photograph of one eell in Fig. 14.9. When it first appears, this seeondary

Figure 14.9 Conveetion eeli in vertieal slot (Ra,," 4.5 x lOs, Pr "" 900, h/d = 20);
aluminium powder flow visualization (see Seetion 23.4). From Ref. [286).
Free Conveetian 153

flow is weakly superimposed on the primary cireulation, but at higher Rayleigh

numbers, the two interaet strongly.
In eaeh of the eells, the fluid is travelling up the hot wall and down the cold one.
The sense of cireulation is the same in neighbouring eells (in eontrast to the
behaviour in horizontallayers) and large shear must oeeur between them. The next
development is presumably a eonsequenee of this. Smaller eells of reverse cireulation,
not extending to the hot and cold walls appear between the main eells, giving a flow
pattem of the form shown in Fig. 14.10. The small eelIs aet in a sense as 'idling
wheels' reducing the shear between the main eonveeting eelIs.

Figure 14.10 Sketch of strearnlines for convection of sllicone oll in vertical slot at
Ra"" 3 x 10 6 • Based on Ref. [101].

At the highest values of the Rayleigh number the flow eeases to be steady. Waves
are generated in the lower part of the hot upgoing boundary layer and the upper
part of the cold downgoing boundary layer. These break down into turbulent
motion in a way similar to transition on a single heated surfaee. The result is a
disruption of the previous eelIular motion and the produetion of a turbulent eore in
the eentre of the slot. Thus the general strueture of the flow is as shown in Fig.
14.11.
154 Physical Fluid Dynamics

Waves
damped

),
'From' :(j\):\
.... \ .]·0·" :'·
·'.: n··
. . .' ...

__
.. .. \

, Breaking'
. (r~~:\

.~
'Hook' ...A
J06-
\ '.

-\ --
. , -
-
.

\
Origin
of waves

o x

Figure 14.11 Schematic representation of principal features of high Rayleigh number

(>-10 6 ) convection in a vertical slot. From Ref. [102].
Free Convection 155

Appendix: The Boussinesq approximation in free convection

In this appendix we examine systematically the conditions for the Boussinesq

approximation to be a good approximation in the speeifie context of steady free
conveetion. This is not the only context in whieh the matter is important (indeed,
some of the points, such as the effeet of the hydrostatie pressure on the density,
should in prineiple be considered even for barotropic flows) but a general
diseussion beeomes too eomplieated. The following analysis may serve as the
prineipal demonstration in this book of the methods used in such eonsiderations.
Some of the diseussion that follows the analysis has wider applieation. On the other
hand, partieular examples of free eonveetion may involve eomplexities not
covered by the general diseussion - for example, through having more than a
single length seale.
We have not made extensive use of the Boussinesq equations in this book.
However, it should be remembered that the Boussinesq approximation underlies the
statement that dynarnieal similarity of free eonveetive flows depends on the
Grashof and Prandtl numbers. When the approximation fails, other non-
dimensional parameters eome in. Thus, the eonsiderations of this seetion are
important not only for the mathematieal formalism but also for the whole question
of the transfer of results from one situation to another. This is of partieular
relevanee to attempts to model geophysieal situations in the laboratory.
We suppose that the flow is produeed by the introduction of temperature
differenees of order S into a system of length seale L. It is important for some of
the following eonsiderations that the vertieal dimension of the system should be
of order L, but we shall not usually make a distinetion between seales in different
direetions.
The prineipal eriteria for the applieability of the Boussinesq approximation are
A = aS ~ 1 (14.42)
B= gp(jL ~ 1 (14.43)
C=gaL/Cp ~ 1 (14.44)
((j = isothermal eompressibility; other symbols as before). There are some supple-
mentary eriteria eonneeted with the eonstaney of fluid properties.
If in addition to the above,
D = gaLTo/Cp S ~ 1 (14.45)
(where To is the absolute temperature), the Boussinesq equations apply directly. If
(14.42)-(14.44) are fulfilled but not (14.45), they apply with the variable T
replaeed by the potential temperature e (equation (14.24». In this ease S is the
seale of the imposed variations in e.
We shall eonsider the signifieanee of relationships (14.42)-(14.45) after we have
examined their role s in the Boussinesq approximation. We may remark, however,
that relationships (14.43), (14.44) and (l4.45) are likely to be violated only in
large-seale geophysieal situations and it is for these that the following eonsiderations
are most important.
We ehoose the two independent thermodynamie variables in terms of whieh
other thermodynarnie quantities may be expressed as the pressure and temperature,
156 Physical Fluid Dynamics
and write
P =Po + Ph +P' (14.46)
T=To+T' (14.47)
Po and To are the absolute pressure and temperature at some arbitrarily chosen
referenee point. Ph is the hydrostatie variation of the pressure and is thus a
funetion only of the vertieal eo-ordinate. p' and T' are the variations associated
with the motion, and are thus general funetions of position. T' eneompasses all
variations in T, whether associated with the boundary eonditions or arising from
internal effeets such as viseous dissipation or adiabatie volume ehange.
The density may now be written
P = Po + Ph + p' (14.48)
where Po is the density at Po and To, Ph is the departure from this in hydrostatie
eonditions, and p' is the further departure associated with the flow.
We now eonsider how the more general equations of motion reduee to the
Boussinesq approximation. For steady flow, the eontinuity equation is
(14.49)
The orders of magnitude of the two terms are (Ph + p') U/L and (Po + Ph + p') U/L.
To write the equation in its ineompressible form (14.1) we require
Ph/PO ~ 1, p'/Po ~ 1 (14.50)
When these are true (we may assume them and look for self-eonsisteney), Ph and p'
may be expressed
Ph=PO{jPh, p'=po(-aT'+{jp') (14.51)
and (14.50) is equivalent to
{jPh ~ 1 (14.52)
aT' ~ 1 (14.53)
(jp' ~ 1 (14.54)
Taking these in turn, eondition (14.52) depends on the size of Ph, whieh may be
estimated from the hydrostatie balanee
(14.55)
giving
Ph - PogL (14.56)
Henee, (14.52) is fulfilled when B is small.
One requirement for eondition (14.53) is evidently that A should be small.
However, it is not at this stage apparent that this is the only requirement as we have
yet to examine flow-indueed temperature variations.
We shall see below that eondition (14.54) is a mueh weaker eondition than
(14.52) and (14.53) and so does not need separate eonsideration.
Turning to the dynamical equation, the eonditions in (14.50) enable us to
replaee pu· V'u by Po u· V'u, and when equation (14.1) appHes the viseous term
Free Conveetian 157
may be redueed to Jl V2 u (for eonstant Jl) as considered in Seetion 5.6. Henee, for
steady flow
(14.57)
where the hydrostatie balanee of equation (14.55) has been subtraeted out as in
Seetion 13.2.
To put equation (14.57) into the form used in the Boussinesq approximation
(equation (14.2)) one also writes
p' = -PoOlT' (14.58)
whieh will be valid if
{3p' ~OI.T' (14.59)
The order of magnitude of p' is indieated by equation (14.57); the second term in
this wiil be of the same order of magnitude as the largest of the other terrns. For
free eonveetion
(14.60)
Henee, eriterion (14.59) is equivalent to (14.43) (small B).
In eireumstanees in whieh (14.42) and (14.59) are fulfilled, (14.54), ofwhieh we
postponed eonsideration above, will be mueh more strongly fulfilled.
The starting point for the development of the temperature equation (14.3) is the
first law of thermodynamies in the form
dE= dQ - pdV (14.61)
where E is the internaI energy (eonventionaily denoted by U in texts on thermo-
dynamies), dQ is the heat supplied, and Vis the volume. If the thermodynamic
quantities are taken to apply to unit mass
V= l/p (14.62)
Applying (14.61) to a fluid particle
DE DQ DV
-=--p- (14.63)
Dt Dt Dt
From the eonsiderations in Seetion 13.2, the heat supplied to a fluid particle
per unit mass is
DQ 1
- = - (k V2 T + J + <1» (14.64)
Dt P
where J is the internaI heat generation per unit volume due to external eauses, and
<I> is the heating due to viseous dissipation of mechanieal energy.
Using relationships derived by standard thermodynamic proeedures we may
write
DE = (aE) DT + (aE) Dp
Dt aT p Dt ap T Dt
= (c _ap)
P P Dt
DT +{3p - OI.T Dp
p Dt (14.65)

OI. DT {3 Dp
DV (av) DT (av) Dp =- ---- (14.66)
Dt = aT p Dt + ap T Dt p Dt p Dt
158 Physical Fluid Dynamics

Substituting (14.64), (14.65) and (14.66) into (14.63) gives

DT aT Dp 1 2
C - - - - = -(kv" T+J+<I» (14.67)
PDt pDt p
This beeomes the Boussinesq equation (14.3) if the terms (aT/p ) Dp/Dt and <I>/p
ean be negleeted.
The first of these has two eontributions, from the variation of the hydrostatic
pressure Ph and from that of the hydrodynamie pressure p'. The ratio of the former
to the first term of equation (14.67) is
a1\l·V'Ph
-----
pCpu·V'T
aTOPh
PoCpT'
---gaLTo
CpT'
(14.68)

Henee, the term may be negleeted when (14.45) is fuIfiUed (i.e. when D is
small). (14.45) is often a mueh more severe eriterion than (14.44), and it is there·
fore fortunate that this eriterion ean be dispensed with by working in terms of the
potential temperature
dTa gaTo
e =T - (Ta - To) wheret -=--- (14.69)
dz Cp
We see this as foilows. Ph and Ta vary vertically only. Henee,
aT DPh aTw dPh
- - = - -=-gawT
pDt p dz

=C -DTa -ga:w 1" (14.70)

p Dt

But
gawT' _gaL
(14.71)
CpDT/Dt Cp
and the seeond term of the last expression of (14.70) ean be negleeted when
(14.44) is fuIfiiled. Then equation (14.67) beeomes
De aT Dp' 1 2
C - - - -=-(k\l e +J+<I» (14.72)
PDt p Dt p
Sinee Ta does not vary horizontalIy, this ehange makes no differenee to the
dynamie equation (14.57).
For the remaining unwanted pressure effeet, the seeond term in (14.72), to be
negligibIe, one requires
a1\l· V'@' aTop'
--- - -- ~ 1 (14.73)
pCpu.V'T pCpT'

tA definition of Ta slightly different from equation (14.22) has been chosen, so that
V 2 Ta = 0 and the heat conduction term is unaffected by the change of variable; the following
considerations imply that Ta is significantly different from the physical adiabatic temperature
only in circumstances in which the approximation breaks down. There is also slight arbitrari-
ness, depending on the exact definition of To, but this is similarly unimportant.
Free Conveetian 159

p' can be estimated from (14.60) and the criterion is

gaL
-aTo ~ 1 (14.74)
Cp
Since aTo is around 1 for gases and is small for other fluids, (14.74) will be
fulfilled whenever (14.44) is fulfiIled.
We may note parenthetically that the above discussion provides the rigorous
justification for the choice of Cp as the speeific heat in Section 13.2. By using T
and Vas the independent thermodynamic variables one can analogously formulate
an equation for Cv DT/Dt, but an extra term arises that is negligible only when the
difference between Cv and Cp is itself negligible.
Finally, the reduction of (14.67) or (14.72) to the Boussinesq form requires
that the viscous heating should be negligible. Viscous energy dissipation has been
considered only in terms of the example of a laminar two-dimensional jet (Section
11.9). However, this indicates the essential information that we need for an order
of magnitude estimate - that the dissipation per unit volume is proportional to the
viscosity and to the square of the veloeity gradient (see equation (11.59)). In fact,
generalization of the ideas in Section 11.9 gives the resuIt that
aUj ( aUj
<I>=IL- - +aUj
-) (14.75)
aXj aXj aXj

(with the summation convention for repeated suffIxes applying). Thus

<I> - ILU 2/{j2 (14.76)
where {j is the length scale appropriate to viscous effects. The distinetion between L
and {j is made so that the diseussion applies even when there is strong boundary
layer formation.
In free conveetion, the balanee between the buoyaney foree and the viseous
foree gives
ILU/a 2 - pgaT' (14.77)
Henee,
<I> \ pgaT' U gaL
(14.78)
pCpu.'VT - pCpUT'/L - Cp = C
The eriterion for the viseous dissipation to have a negligible effeet on the thermal
balanee is that this should be small; Le. it is relationship (14.44). This applies
whether {j - L (viseous flow) or {j ~L (viseous dlssipation eonfined to boundary
layers).
We ean now retum to the one question that we left unsettled in eonneetion with
the eontinuity and dynamic equations, whether (14.42) (small A) is an adequate
representation of (14.53). When the temperature equation takes the Boussinesq
form it will be, but there remains the possibility that the largest temperature
differenee is that associated with the adiabatic gradient. The corresponding
fractional density ehange is
gcxL
allTa - -Ca T0 (14.79)
p
and, as with (14.74), this is always small when Cis small.
160 Physical Fluid Dynamics
In addition to the various effects considered above, the Boussinesq approxima-
tion, as designated in Section 13.2, requires effective constancy of the physical
properties of the fluid; that is of jJ., k, a, and ep. The temperature and pressure
variations must not be so large as to produce large fractional changes in these
quantities. Often fulfilment of this requirement will go along with fulfilment of the
requirement of small fractional density changes. However, in prineiple, any
parameter having a much larger temperature or pressure coeffieient than density
needs speeial consideration. In practice, the temperature variation of viscosity is
often the most serious problem, both in the Iaboratory and in studies of natural
situations.
What is the physical significance of the parameters, A, B, e, and D that
prineipally govern the applicability of the Boussinesq approximation?
The requirement of small A just says that the fractionaI density change
produced by expansion in the imposed temperature field must be negligible.
B and erelate to the 'scale heights' of the system - the heights over which
various pararneters change by a fraction of order unity. For example, in an
isothermaI perfect gas at rest
dp/dz = -pg, p = p/RT (14.80)
giving
p=poe-gZjRT, p=poe-gZjRT (14.81)
In this case the scale heights of pressure and density are both RT/g. More generally,
1/gp{3 and ep/ga are the scale heights of isothermal pressure and adiabatic tempera-
ture. B and e are the ratios of the Iength scale of the system to these scale heights.
The Boussinesq approximation applies only to systems that are small eompared
with the seale heights.
Another physical interpretation of e - in terms of thermodynamic 'efficiency' -
has been discussed in Section 14.4.
D is evidently related to e (although usually D ~ C), but it is more readily
thought of as the ratio of the adiabatic temperature gradient to a typical imposed
temperature gradient.
B and e (and for that matter D) are always exceedingly small on the laboratory
scale. Departures from Boussinesq conditions can occur owing to variations of fluid
properties - particularly temperature variation of viscosity and, in water,
temperature variation of expansion coefficient - and, in experiments with gases,
owing to A not being small enough.
The importance of B and e arises in Iarge-scale geophysical systems, where they
are not always small. Motions of the atmosphere extending throughout its depth,
and motions in the interiors of planets and stars, for exarnple, may involve processes
that cannot be modelled in the Iaboratory. The foregoing analysis shows that when
B is not small, one must additionally take into account (i) the effect of the hydro-
static pressure Ph throughout the equations of motion and (ii) the effeet of the
hydrodynamic pressure p' on the buoyancy foree. When e is not small, one must
take into account (üi) heat generation by viscous dissipation and (iv) other non-
Boussinesq effects in the thermal balanee (relationships (14.71) and (14.73».
B and e are in principle thermodynamically independent parameters. In practice
they tend to be of the same size. For a perfect gas
a=1/T, {3=1/pRT (14.82)
Free Conveetian 161

and so
B/C= Cp/R = "t/h -1) (14.83)
For real gases the ratio will siInilarly be of order unity. For eondensed phases
(liquids and solids) the ratio a/pfjCp is known as Grüneisen's ratio, G, giving
B/C= I/G (14.84)
G rarely differs greatly from 1. Water is something of an exeeption to this (its G
varies strongly with temperature but is typically 10 -1), but there would seem to be
no applieation of this; the density at the bottom of the sea is inereased only a
small amount by the weight of the water above. On the other hand, there are
applications of these ideas to eondensed phases in the study of planetary interiors
(see, e.g., Seetion 24.4).
These eonsiderations imply that all the effeets (i) to (iv) above are liable to
beeome signifieant siInultaneously and it is generally not useful to take one into
aeeount whllst ignoring the others. Thus situations in whieh the Boussinesq
approximation fails beeome very eomplieated theoretically. We have already seen
the problems of modelling them in the laboratory, so it is altogether diffieult to
aseertain the flow pattems.
15
Flow in Rotating Fluids

15.1 Introduction [35, 130, 163]

This chapter is concemed with the dynamics of fluids in rotating systems. This
braneh of fluid meehanics has developed rapidly in reeent years as an obvious
eonsequenee of interest in geophysieal flow probIems. Evaluation of the parameters
shows that the motions, partieularly on the large seale, of the Earth's atmosphere,
oeeans, and eore and of stars and galaxies will all exhibit the effeets diseussed in
this ehapter. The rotation gives rise to a range of new phenomena; here we eonsider
a small seleetion of these.
The whole subjeet eould be formulated as seen by an observer extemal to the
rotation. Sinee, however, all boundary eonditions will be specified in terms of the
rotating frame of referenee it is easier to modify the equations of motion so that
they apply in such a frarne.
If one takes a body of fluid and rotates its boundaries at a constant angular
velocity n, then at any time suffieiently long after starting the rotation, the whole
body is rotating with this angular velocity, moving as if it were a rigid body.
There are then no viseous stresses acting within the fluid. Any disturbanee -
Le. anything that would produee a motion in a non-rotating system - will produee
motion relative to this rigid body rotation. This relative motion ean be considered
as the flow pattem; it is the pattem that will be observed by an observer fixed to
the rotating boundaries.

15.2 Centrifugal and Coriolis forees

The effeet of using a rotating frame of referenee is weIl known from the meehanies
of solid systems; there are aeeelerations associated with the use of a non-inertial
frame that ean be taken into aeeount by introducing eentrifugal and Coriolis forees.
That statement may be expressed in a form appropriate to fluid systems by

(DU)
Dt
=(DU)
I Dt R
+.Qx(.Qxr)+[Link] (IS.I)

The subseripts I and R refer to inertial and rotating frarnes of referenee. {Du/Dt)r is
thus the aetual aeeeleration that a fluid partide is experiencing and so p{Du/Dt)I is
the quantity to be equated with the sum of the various forees aeting on the fluid
particle. {Du/Dt)R is the aeeeleration relative to the rotating frame and ean thus be
Flow in Rotating Fluids 163

expanded in the usual way:

( -DU) =aUR
- + (u' VU)R (l5.2)
Dt R at
Dropping the subscript R, as all velocities will be referred to the rotating frame
throughout the re st of this chapter, the equation of motion is
au 1
-+ U' Vu = - - VP - Qx (Qx r)- 2Qx U + vV 2 u (15.3)
at p
The second and third terms on the right-hand side of (15.3) are, of course, respec-
tively the centrifugal and Coriolis forces.
In many problems the centrifugal force is unimportant. This is because it can be
expressed as the gradient of a scalar quantity;
Qx (Qx r) = _V(~n2 r'2) (15.4)
where r' is the distance from the axis ofrotation (Fig. 15.1). Hence replacing the
pressure by (p - ~pn2 ,2) reduces the problem to one that is identical except that
the centrifugal force is absent. This is entirely analogous to the procedure of sub-
tracting out the hydrostatic pressure to remove the effect of gravitational forces, as
discussed in Section 13.2. The centrifugal force is balanced by a radial pressure
gradient which is present whether or not the re is any flow relative to the rotating
frame and which does not interact with any such flow. The limitations to this state-
ment are just the same as for the gravitational case. Firstly, the pressure must not
appear explicitly in the boundary conditions. Secondly (since p has been take n
throughV), the density must be constant; centrifugal force variations associated
with density variations do give rise to body forces that can alter or even cause a
flow; such effects are outside the scope of the present discliSsion, although a flow
in which they are present is considered briefly in Section 15.8.

Figure 15.1 Definition sketch.

It should be emphasized that the centrifugal force under discussion here is that
associated with the rotation of the frame of reference as a whole. In other contexts
(e.g. rotating Couette flow) it is sometimes convenient to talk about the centrifugal
force associated with circular motion relative to the frame of reference (either
inertial or rotating). This is then a way of discussing physically effects that are
contained mathematically in one or both of (u • vu) and the Coriolis term.

15.3 Geostrophie flow and the Taylor-Proudman theorem

We thus see that the eause of differenees between the dynamies of non-rotating and
rotating fluids is the Coriolis foree. It is instructive to consider flows that are domi-
164 Physical FluUl Dynamics
nated by the action of Coriolis forees and thus see these düferences in their extreme
form. Let us suppose therefore that the Coriolis effeet is large compared with both
the inertia of the relative motion and viseous action. Restrieting attention to steady
flow this means
I u • Vu I < In x u I (15.5)
and
I vV 2 u I < I n x u I (15.6)
Expressing these in terms of seales in the usual way
U2 /L <ilU and vU/L 2 <ilU (15.7)
or
(15.8)
The quantities U/nL and v/nL 2 are known respectively as the Rossby number and
the Ekman number; they indicate the ratios of respectively inertial to Coriolis
forees and viscous to Coriolis forees. As in a non-rotating fluid, viscous effeets may
beeome important, even when simple order of magnitude eonsiderations indicate
otherwise, in boundary layers close to surfaees. We are here considering flows out-
side such boundary layers; that is flow in a region for whieh Euler's equation would
apply if the system were not rotating.
When both the Rossby number and the Ekman number are small, the equation of
motion becomes
1
2nx u= -- \lp (15.9)
p
where the pressure has been modified by the eentrifugal pressure as discussed above.
Flows in whieh this balanee between Coriolis foree and pressure foree pertains are
ealled geostrophie flows.
An important property of such flows is immediately evident. The Coriolis
foree is always perpendieular to the flow direction. Henee, the pressure gradient is
also perpendicular to the flow direction. This means that the pressure is eonstant
along a streamiine - in marked contrast to the behaviour in non-rotating systems
where one is aecustomed to think of the pressure variations along a streamIine (e.g.
Bemoulii's equation).
This feature of geostrophie flows is farniiiar in the interpretation of weather
maps. These are usually eompiled primarily from information about the pressure
distribution. Isobars are drawn and, because of the dominating influenee of the
Earth's rotation, these are then taken to be also the lines along whieh the wind is
blowing (strietly speaking, the wind at some height rather than the wind near
ground level).
Another interesting property of geostrophic flow is diseovered by taking the
eud of equation (15.9)
vx (nx u)=o (15.10)
This expands to
n· Vu -u ·Vn+ u(V· n) -n(V· u) = 0 (15.11)
F10w in Rotating F1uids 165

g is not a function of position, so the second and third terms are zero. The con-
tinuity equation is unaltered by the rotation, so
V'·u=o (15.12)
and equation (15.11) becomes
g·V'u=O (15.13)
Ifaxes are chosen so that g is in the z-direction, this is
nau/az = 0 (15.14)
Le.
au/az = 0 (15.15)
There is thus no variation of the velocity field in the direction parallel to the axis
of rotation. This result is known as the Taylor-Proudman theorem. ln component
form

au av aw
-=-=-=0 (15.16)
az az az
If one is dealing with a system with solid boundaries perpendicular to the rotation
axis so that w = 0 at some specified value(s) of z then this implies
au av
- =- =0 w=0 everywhere (15.17)
az az '
and the theorem says that the flow is entirely two-dimensional in planes perpendi-
cular to the axis of rotation.

15.4 Taylor Columns

The Taylor-Proudman theorem has simple and striking consequences, illustrating

the fact that rotating fluids exhibit a range of phenomena not found in non-rotating
fluids. The principal of these is the fonnation of 'Taylor columns'. These occur
when there is relative motion between an obstacle and the fluid in a strongly rota-
ting system. We consider first the case in which this motion is perpendicular to the
axis of rotation (giving what is called a transverse Taylor column). The fluid is
deflected past the obstacle. Since the flow must be two-dimensional this deflection
also occurs above and below the obstacle (visualizing the axis of rotation as vertical~
There are thus columns of fluid, extending parallel to the axis from the obstacle,
round which the fluid is deflected just as if the solid walls themselves extended
there.
Although Taylor columns realizable in the laboratory involve complications not
described by this simple explanation, prediction of their occurrence has received
abundant experimental confirmation. Figure 15.2 "illustrates this with a sequential
pair of photographs showing the development of dye streaks originating inside and
outside a Taylor column. Two features are to be noted. Firstly, the dye streak
originating in the column is much shorter than those originating outside, showing
that the fluid there is moving very slowly (all the dye streaks were initiated at the
166 Physical Fluid Dynamics

same instant). Secondly, the dye streaks meeting the Taylor column are deflected
around it much as if it were a solid obstacle; it is important to realize that the dye
lines are well above the top of the obstacle and would just pass straight over it in
the absence of rotation.
Full appreciation of Fig. 15.2 requires some description of the method by which
it was obtained. The basic apparatus consists of a rotating cylindrical tank, with an

Figure 15.2 StreakIine development in and around transverse Taylor column; see text for
detaiis. Photos by C. W. Titman.
Flow in Rotating Fluids 167

~---'l.-- Tru neated eyl i nder

Field of view

Plug in top and exit

tube for relative flow
Rotating
tank

Figure 15.3 Key to Fig. 15.2.

arrangement for pumping water in at the periphery and out on the axis. This pro-
duces the relative flow between the body of the fluid and an obstacle fixed in the
tank. Because the tank is rotating, this flow is not radial but azimuthal, fluid
particles moving round on circles concentric with the tank; the pumping can be
thought of as producing a radial pressure gradient which is balaneed by the Coriolis
force associated with the azimuthal motion. The obstacle is a truncated cylinder on
the floor of the tank and extending about one-third of its depth. Dye can be
released (by the pHotechnique described in Section 23.4) in four places, of which
the positions in plan are shown in the sketch (Fig. 15.3); in elevation, these dye
sources are about midway between the top of the obstacle and the top of the tank.
One source is over the obstacle and thus in the Taylor column. The other three are
placed so that the dye from them meets and passes round the Taylor column.
This experiment is, in principle, a repetition of the pioneering experiment by
Taylor [256], first demonstrating the validity of the surprising predictions of the
Taylor-Proudman theorem. Subsequent work [126,127,284] has provided much
information about the quite complex detailed structure of flows of this type; brief
mention of further features will be made at the end of this section.
A different type of Taylor column, a longitudinal Taylor column, arises when an
obstacle moves along or paralleI to the axis of rotation. To understand this directly
in terms of the Taylor-Proudman theorem, one needs to consider the fluid as
extending to infinity in both direetions parallei to the axis of rotation. Then the
obstacle pushes a column of fluid mo ving with its own speed ahead of it and pulls a
similar column behind it. In practice, as one might expect, Taylor columns are very
long but not infinite when the Rossby number is very small but not zero. Moreover,
for reasons connected with the loe al non-geostrophic regions mentioned below, part
of a Taylor column can be observed even when the apparatus is relatively short.
Experimental observation of this type of Taylor column is thus possible. Figure 15.4
gives an example. A sphere is rising along the axis of a rotating tank. Water near the
bottom of the tank is dyed, and the sphere has risen into a region of clear water
168 Physical Fluid Dynamics

Figure 15.4 Longitudinal Taylor columns, shown by method described in text. The sphere is
seen as a bulge in the column. Ro = uo/nd = 0.136; Ek = vtnd' = 0.00134. From Ref. [176).

above. Dye carried into this region by the motion demonstrates the presence of
Taylor columns ahead of and behind the sphere. With this technique, not all the
fluid in the forward Taylor column is necessarily dyed, since some may have
started above the dyed region; but the presence of dye so far ahead of the sphere
clearly demonstrates the existence of the column, and is in marked contrast with
what would happen in the absence of rotation.
In developing an understanding of the behaviour of rotating fluids, it is evidently
going to be valuable to formulate a physical description of the processes underlying
the Taylor -Proudman theorem. This is not in fact easy, just as it is not easy to
develop a ready physical intuition about the behaviour of solid gyroscopes. A
reading of Taylor's original papers makes it evident that he was surprised by the
Flow in Rotating Fluids 169

theoretical predictions and perhaps half expected that his experiments would not
confirm them. However, some comments may be made.
As a starting point we note that the Coriolis force is independent of the actual
location of the axis of rotation. The phenomena are thus not aresult of rotation
about a partieular axis but rather the re sult of the flow occuring in the presence of
uniform background vorticity of value 2Q. At low Rossby number any vorticity
associated with the relative motion is small compared with this. The relative motion
cannot interact with the background vorticity in a way that would alter it so that it
no longer corresponded just to a uniform rotation. It is motions that would do this
that are exc1uded by the Taylor-Proudman theorem. (Note that the derivation of
the theorem - starting with taking the curl of the momentum equation - indicates
that it is a state me nt about the action of vorticity.)
One can say that there are two types of such motion: flows that violate
3u/3z = 3v/3z = 0 (15.16a)
and that would cause twisting of the background vorticity; and flows that violate
3w/3z = 0 (15.16b)
and that wou1d cause stretching of the background vorticity. (These ideas depend
on the resuIt - Section 6.5 - that vortex lines always consist of the same fluid.)
The two processes are not in general independent but it is convenient to consider
them separa tely.
Vortex twisting would occur for example if the fluid approaching an obstacIe
were slowed down but that approaching the region above it were not (Fig. 15.5 -
the sort of flow that would occur in the absence of the background vorticity). This
would produce a large vorticity component not in the z-direction, associated with
which would be relative motions large compared with the original relative motion.
The background vorticity thus resists the twisting and, in the case of low enough
Rossby number, prevents it altogether. This is closely analogous to the action of a
solid gyroscope in resisting turning, except that it is each individual vortex line
that has the gyroscopic action not the system as a whole.
Vortex stretching would occur if fluid in a tank of finite depth moved from above
an obstacle to some other station (Fig. 15.6). This would locally increase the
vorticity (by the process described in the discussion following equation (6.27»
again causing a relative motion large compared with the original relative motion.

D 1"

Figure 15.5 Flow violating Taylor-Proudman theorem through vortex twisting in regions of
non-zero ou/oz.
170 Physical Fluid Dynamics

Figure 15.6
r
Taylor-Proudman theorem violating motion of fluid from above obstacle.

Thus this is also resisted. The analogy with a gyroseope is not quite so simple here;
if a gyroseope were made of a deformable material eertain deformations would be
resisted by gyroseopie action. It should be noted that, although the above diseussion
has been given in terms of vortex stretching, it applies equally to vortex compression;
aloeal reduetion in the vorticity also eorresponds to a large relative motion.
The Taylor-Proudman theorem indieates that, under eireumstanees speeified
above, the flow is two-dimensional. It is natural to enquire next about the pattem
of this two-dimensional flow. For example, we have seen that, aeeording to the
theorem, there is no transfer of fluid between the inside and outside of a Taylor
eolumn; but there will be separate two-dimensional flows in eaeh of these regions.
However, no further information about these flows ean be obtained from the
geostrophie flow equation. This may be seen as follows. The equation now has two
components:

-2Qu=-! ap (15.18)
p ax
2Qu =_! ap (15.19)
p ay
u and u are linke d by the eontinuity equation
au av
-+-=0 (15.20)
ax ay
apparently providing three equations for the three unknowns u, v, and p. However,
differentiating (15.18) with respeet to y and (15.19) with respeet to x and then
eliminating a2 p/axay gives equation (15.20) again. Continuity, therefore, does not
give any independent information and there are effeetively only two equations for
the three variabies. Any solenoidal two-dimensional motion satisfies the geostrophie
equations. The Taylor-Proudman theorem is the limit of the information to be
obtained in this way.
How then is the motion, for example inside a Taylor eolumn, determined? The
answer is that the flow develops loeal regions where the geostrophie equations do
not apply. Both the Rossby number (UjnL) and the Ekman number (V/QL2) are
inereased if the length seale is decreased. If, therefore, there are regions in whieh
the flow parameters vary over a distanee small compared with the imposed length
seales, then inertia forees and/or viseous forees may be loeally important. There
Flow in Rotating Fluids 171

may be local violation of the Taylor-Proudman theorem. An obvious place for

such a development to occur is between the two flows which, according to the
theorem, do not interaet; that is at the edge of the Taylor column. Thin shear layers
are observed here. The whole structure of the detailed flow in the Taylor column
(such as the slow motion seen in Fig. 15.2) is deterrnined by these shear layers
together with the boundary layer on the obstaele's surface. Because the two regions
are separated by a layer in which the Taylor-Proudman theorem does not apply,
there is in fact some interehange of fluid between the interior and the exterior of a
transverse Taylor column. It is the shear layers also that govem the length of a
Taylor column (either transverse or longitudinai) so that, although it is long at low
Rossby and Ekman numbers, it does not extend to infinity as predicted by the
Taylor-Proudman theorem.
A discussion of the structure of the shear layers is beyond the scope of this
book [182]. However, it is interesting to note that we now have two elasses of
flows that spontaneously develop regions of smaIllength scale. In boundary layer
formation it happens in order that all the imposed conditions should be fulfilled;
the re is no solution of the equations when all terms that are small on the first
approximation are negIected. Here, it happens as a process that selects a particular
flow pattem; there is a multiplicity of solutions when all apparently small terms
are negIected.

15.5 Ekman layers

We eonsider nowa flow in a rotating fluid in whieh viscous forees are important but
which is much simpler than the shear layers mentioned above. This is the Ekman
layer, the boundary layer between a geostrophic flow and a solid boundary at which
the no-slip condition applies. This tu ms out to be actuaIly simpler than the corres-
ponding problem in a non-rotating fluid. The results have the added interest of rather
direet application to the atmosphere and the oceans.
We suppase that a body of fluid rotating about the z-axis has a boundary in the
xy-plane. A geostrophie flow is oecurring within the main body of the fluid and
this is taken to be a uniform unidirectional flow, uo, in the x-direction. In practice
this me ans that the length scale of variations in the geostrophic flow must be large
compared with the boundary layer thickness that emerges from the following
analysis. Associated with Uo is a uniform pressure gradient in the y-direction, the
equations ofmotion in the geostrophic region being
1 oPo
2Uuo = - - - (15.21)
p ay
0= _! opo (15.22)
p ox
Between this geostrophic flow and the boundary is a region in which, as usuai,
viscous forees are brought into play by the boundary condition. We look for a
solution for this boundary layer with the velocity field uniform in both the x-and
y-directions. This is best justified a posteriari: one finds such a solution. How-
ever, it is at first sight a rather surprising proeedure, since one knows that no
solution of this type exists in the absenee of rotation (Sectian 11.3). The reason for
172 Physical Fluid Dynamics
the difference is associated with the fact that a stress ac ts between the boundary and
the fluid. Without rotation, this stress extraets momentum from the flow; in zero
pressure gradient, the totaI flow momentum must decrease as one goes downstream,
Le. the boundary Iayer must grow. In a rotating fluid the stress can come into
equilibrium with the integrated Coriolis force associated with the motion; once the
boundary Iayer is formed no further growth need occur. Since the Coriolis force ac ts
at right angles to the motion, this explanation implies that the boundary Iayer
necessarily involves flow in both directions paralleI to the wall.
Accordingly, we take u and v both non-zero but
au au av av
-=-=-=-=0 (15.23)
ax ay ax ay
and, as is then required by continuity,
w=O (15.24)
The three components of the momentum equation become

(15.25)

(15.26)

ap =0 (15.27)
az
Equation (15.27) implies that
ap = apo = O. ap = apo (15.28)
ax ax ' ay ay
Hence, substituting (15.21), the equations to be soIved are
a2 u
-2nv=v-2 (15.29)
az
a2 v
-2n(uo - u) = v az 2 (15.30)

The boundary conditions are

u = v= 0 at z =0 (15.31)
u ~ Uo and v~0 as z ~ oo (15.32)
Eliminating u or v gives a soIuble fourth-order differential equation. However, the
equation can be kept second-order by working in terms of the complex variable
Z=(u+iv)/uo (15.33)
Multiplying equation (15.30) by i and adding it to equation (15.29) gives
a2
2in(u + iv) - v -2 (u + iv) - 2inuo = 0 (15.34)
az
Flow in Rotating Fluids 173

i.e.
d2 Z
v dz2 - 2in(Z - 1) = 0 (15.35)

with
Z=O at z=O (15.36)
and
Z-+-l as z-+-oo (15.37)
The solution of this is
[( 2iQ f/2 ]
Z =1 - exp \ ----;- Z (15.38)

or, taking real and imaginary parts,

u =uo [1 - e-z/~cos(z/A)] (15.39)
u = uoe-z/~ sin(z/A) (15.40)
where
A = (V/Q)1I2 (15.41)
Figures 15.7 and 15.8 show the nature of this solution. The former shows u/uo and
u/uo as funetions of z/ A. The latter presents the same information in the form of a
polar diagram, known as the Ekman spiral. Different points along this correspond
to different values of z/ A (not uniformly spaced) and the velocity at each height
is then given in magnitude and direction by the line from the origin to the corres-
ponding point on the spiraI. Noteworthy features are, firstly, that elose to the
boundary u/u = 1 and so the flow is at 45° to the geostrophic flow; and secondly,
that at z/ A = 'Tr the flow is in the same direction as the geostrophic flow but
approximately 1.1 times as fast.
We notice that, as Q -+- 0 then A ~ oo; in the absence of rotation this solution
collapses, as implied by the physical discussion above. In that case the problem of
physical interest is always the growth of the boundary layer.
Laboratory verification of the above theory in full is difficult because achievable
Ekman layers are rather thin. However, the main features can be demonstrated, an
example being given in Fig. 15.9. This shows traverses with hot-wire anemometers
(Section 23.2) in an air flow similar to the water flow discussed in connection with
Fig. 15.2. The geostrophic flow is thus azimuthal, but radial motion exists in the
Ekman layers on the top and bottom boundaries. Although the profile very elose to
the boundary is lost owing to probe size, maxirna in the azimuthal and radial
velocities are observed in relative positions that are in good agreement with Fig. 15.7.
A more important role of laboratory work on Ekman layers has been to reveal
two types of instability [103,253]. Both can be seen in Fig. 15.10, a dye pattem
generated from crystals on the boundary. One type produces the shorter waves at
the top left of the picture; the other produces the less regular, larger scale pattem
in the middle.
The Ekman spiraI is related to the observation that the wind at ground level and
the wind aloft are commonly in different direetions. At a generallatitude, it is the
vertieal component of the Earth's rotation that is important, and the theory indicates
174 Physical Fluid Dynamics

v
Ug

o 2 3 4 5
zlj.

Figure 15.7 Velocity distributions in Ekman layer.

Figure 15.8 The Ekman spiral: polar diagram of velocity veetor u in Ekman layer. Numbers
along spiral are values of Z/4.
Flow in Rotating Fluids 175

U,

c:
,9
~

.~
a.
m
u
'~

>'"

Hot wire voltage

Figure 15.9 Azimuthal and radial velocity component measurements in Ekman layer below
azimuthal geostrophic flow. From Ref. [253].

Figure 15.10 Ekman layer instabilities. From Ref. [103].

176 Physical Fluid Dynamics

that the wind elose to the ground will be at 45° to the geostrophic wind - and to
the left of it in the northem hemisphere. In practice, because the atmosphere is
usual1y turbulent, Reynolds stresses (see Section 20.4) replace viscous stresses and
the angle is less than 45° [243].
An Ekman layer below a geostrophic flow of variable velocity has an additional
important property, which leads to a significant interaction with the geostrophic
region. The process is illustrated schematically by Fig. 15.11. The solid arrows
indicate the geostrophic flow and the broken arrows the transverse motion in the
associated Ekman layer on a boundary in the plane of the page. This transverse
motion involves divergence (or convergence) aml thus gives rise to a flow out of (or
into) the geostrophic region. This process is known as Ekman layer suction (or
injection). The signs are such that the Ekman layer sucks when the relative vorticity
of the geostrophic flow has opposite sign to the basic rotation. The suction changes
the boundary condition experienced by the geostrophic region. Because of the
Taylor-Proudman theorem this has an effect throughout the geostrophic flow,
and, in certain circumstances, Ekman layers can exert a controlling influence on the
development of this. Transfer of fluid between boundary layers and the body of
the fluid in this way is, for instance, the process by which a tank of fluid is brought
to the angular velocity of its boundaries (as mentioned in Section 15.1). For most
geometries 'spin-up' in this way occurs much more rapidly than it would through
viscous diffusion [35].

.- - +
Figure 15.11 Sketch indicating reason for Ekman layer suction: see text.

15.6 Intrinsie stability and inertial waves

Rotating fluids have an intrinsie stability, in the sense that if a fluid partiele is
displaced the re is a tendency for it to retum in a way that would not occur in a
non-rotating fluid. Consider an isolated partiele of unit mass which, as aresult of
some disturbance, moves with a speed q in any direction perpendicular to the axis
of rotation. (We are not, of course, considering here a genuine fluid motion or even
a particular physical situation, but merely illustrating a general feature in the
simplest way.) A Coriolis force of magnitude 2r?q acts on it, always at right angles
to its direction of motion. It thus moves on a circular path of radius r given by
q2 jr =2r?q (15.42)
that is
r =qj2r? (15.43)
Flow in Rotating Fluids 177

It goes one e round the eircle in a time

t = 2rrr/q = rrln (15.44)
independent of q. It thus returns to its originaI position periodicalIy, twice during
every revoIution of the fluid. It is thus often said that a rotating fluid has an
intrinsie anguIar frequency, 2rr/t = 2Q.
In a compIete system, the effeet of this constraining tendency acting on every
fluid particle is that rotating fluids can support wave motions, known as inertial
waves, that wouId not arise in the absence of rotation. Waves can be generated, for
exampIe, by oscillating a disc in a rotating tank (Fig. 15.12). Both theoretically
and experimentally, waves are generated for any anguIar frequency beIow 2Q.
We shall not enter into a detaile d discussion of the properties of these waves.
Very similar waves occur in stratified fluids and will be described in Section 16.4.
Many of the features of those waves, such as the frequency being related to the
orientation but not the magnitude of the wave number and the group veIoeity
being perpendicular to the phase velocity, also apply to inertiaI waves [35,200) .
Aconsequence is the similar appearances of Figs. 15.12 and 16.11(a).

Figure 15.12 Inertial waves produced by oscillating disc in a rotating tank (lower part of
pattem is produced by reflection at walls). Aluminium powder flow visualization (see Section
23.4). From Ref. (35).

15.7 Rossby waves [202]

By way of a further illustration of waves resuIting from rotation of the system, we

consider Rossby waves. These occur in particular geometricaI arrangements. We
consider them partly because they are of some meteoroIogical importanee (see
Section 24.3) and partly because their origin can be seen rather directly in terms of
the Taylor-Proudman theorem.
The simpIest configuration in which Rossby waves occur is a Iayer of fluid
bounded by two pIanes both almost perpendicular to the axis of rotation but not
quite paralleI to one another so that there are slight variations in the depth of the
178 Physical Fluid Dynamics

layer. Wave motions can then arise from 'twanging' the constraint produced by the
Taylor-Proudman theorem. If a flow in the layer were exactly geostrophic, then it
would have to follow contours of constant depth (from a line of reasoning exactly
paralleI to that given for the formation of Taylor columns). Correspondingly, if a
fluid partic1e is displaced to a position of different depth, the re will be a tendency
for it to retum to its original station. If it has inertia it will overshoot and oscillate.
A complete flow pattem of such oscillations constitutes a Rossby wave.
The inertia involved here is that associated with the unsteadiness 't)f the wave,
au/at. That associated with u· Vu is still neglected - by restricting attention to
waves of small enough amplitude. Viscous effects are also assumed negligible. The
dynarnical equation is thus
au 1
-+2~hu=--Vp (15.45)
at p
The nature of Rossby waves can be shown by considering the simplest case
(Fig. 15.13) in which the depth, h, varies uniformly and in one direction only
(chosen as the y-direction):
ah dh
ax =0; dy ='Y, a constant (15.46)

'Y is taken to be small.

We suppose that the waves retain the properties of geostrophic flow that
au/az = av/az = 0 (15.4 7)
but that aw/az becomes non-zero. (This apparently arbitrary procedure can be
shown, a posteriari, to involve no inconsistency; the pressure variation obtained by
integrating the z-component of equation (15.45) produces negligible modiflcation
to the x andy components when 'Y is small.) Then the flrst two terms in the con-
tinuity equation
au av aw
-+-+-=0 (15.48)
ax ay az
are independent of z and so
aw
- = const. w.r.t. z (15.49)
az
It now simplifles the analysis to choose the z-axis to be exactly norm al to one of
the boundaries, z = 0 say (Fig. 15.13). If r is small, this makes negligible difference

h
"
z

Figure 15.13 Definitian diagram for simple Rossby wave theory.

Flow in Rotating FIuids 179
to the dynamical equations. At this boundary w =O. Hence
aw/az = wh/h (15.50)
where wh is the value of w at the top boundary, z = h. Here the flow must be
paralleI to this boundary (since it is supposed separated from the no-slip condition
by a thin boundary layer). Thus
Wh cosec lX = Vh see lX (15.51)

(15.52)
in view of equation (15.47). Combining equations (15.48), (15.50), and (15.52),
au av
-+-+-=0
"Iv
(15.53)
ax ay h
The Rossby waves are given by this form of the continuity equation taken in con-
junction with the x andy components of equation (15.45),
au
--2nv=-- -
1 ap (15.54)
at p ax
av
-+2Qu=-- -
1 ap (15.55)
at p ay
From the physlcal ideas discussed earlier we expect to find waves involving
oscillations up and down the depth gradient. We look for ones in which the velocity
is constant in this direction; Le. u and vare not funetions of y. In so doing, we
are losing generality even for the particular geometry considered. However, this
provides an adequate context in which to see the principal properties of the waves.
The pressure must stilI vary in the y-direction because of the Coriolis foree.
Cross-differentiation of equations (15.54) and (15.55) now gives
a2 v + 2n-
-
au
axat ax =0 (15.56)

Substituting from equation (15.53) (in which the second term is now zero),
a 2 v 2n'Y
- - -v= 0 (15.57)
axat h
For scales over which the depth variation is small enough, 2n'Y/h may be treated
as a constant, and we have a linear equation for which we may look for a solution of
the form
v = Vo ei(wt-kx) (15.58)
If this can be found with real w and k, the flow can be described as a wave motion.
Substituting gives
wkvo - 2n'Yvo/h = 0 (15.59)
and thus
(15.60)
180 Physical Fluid Dynamics

The phase and group velocities are respectively

cp = w/k =hw 2 /2ilr (15.61)

Cg = dw/dk = -hw 2 /2ilr (15.62)

These waves have the following properties:
(i) they are linear (the properties are independent of the amplitude so long as this
is small enough for the equations to apply);
(ii) they are dispersive (c p is a function of w);
(iii) they are essentially progressive (the solution applies for onlyone sign of cp,
and so there can be no superposition to give standing waves);
(iv) au/ax and vare in phase (equation (15.53) with av/ay = 0) and so u and vare
rr/2 out of phase.
The most direet experimental method of checking a theory of this sort is to
introduce a mechanieal oscillator into an appropriate configuration. Such experi-
ments have been performed [132], although not in a geometry to which the above
simplest theory can be applied, and show good agreement between theoretieal and
observed properties of Rossby waves.
Spontaneously occurring Rossby waves arise when an otherwise geostrophic flow
is deflected in a way that violates the Taylor-Proudman theorem. Figure 15.14
shows schematica1ly the pattem produced when a Rossby wave of the type analysed
above is superimposed on a uniform bulk flow (speed U) in the x-direction. The
wave causes Li.e flow to meander up and down the depth variations. The wavy
lines represent the instantaneous streamIine pattem produced by the combination of
the bulk flow and the instantaneous Rossby wave motions - indicated by the short
arrows. The streamlines have the same spacing at the crests and at the troughs; the
greater speed when u has the same sign as U is associated with each streamline then
being at its minimum depth.
The whole pattem in Fig. 15.14 progresses in the x-direction with the phase
velocity c p relative to the bulk flow. If U= -cp, the pattem remains stationary.

u
..
dh
dy

Figure 15.14 Diagram summarizing the properties of a Rossby wave superimposed on a mean
flow.
F/ow in Rotating F/uids 181

This is what happens in flow past an obstacle in the appropriate direction in a

rotating fluid of variable depth. The fluid is disturbed by the obstacle in a way that
'twangs' the Taylor-Proundman theorem and then oscillates as its travels down-
stream. The waves seleet the wavelength such that c p = -U, and so the pattem is a
steady flow relative to the obstac1e (which could be analysed as a steady flow by
replacing a/at by ua/ax in the above treatment). The waves propagate downstream
despite the 'freezing' by the bulk flow because the group velocity is different from
the phase velocity; they propagate at a speed Cg relative to the fluid and thus at a
speed Cg + U = 2U relative to the obstacle.
Figure 15.15 shows a wave pattem produced in this way. This again is a situation

Figure 15.15 Train of Rossby waves produced by flow past a bump - see text - and shown
by trajectories over a short time interval of suspended particles. Photo by A. J. Faller from
Ref. [202].
182 Physical Fluid Dynamics

to which the above simple analysis is not really applieable, but the processes
oeeurring are sirnilar to those deseribed above. The rotating annulus eontains fluid
of greater depth at the outer wall than at the inner. The obstacle is at the bottom
left of the pieture, obseured by its support, and the relative flow generating the
waves was produeed as a transient effeet by asudden small ehange in the rotation
rate.

15.8 Convection in a rotating annulus [108, 128]

We eonsider one example in whieh thermaI density variations, gravity, and rotation
are all present. The equations of motion for such a situation will not be exarnined,
but we willlook at a eonfiguration for which the re is a large amount of experimental
information - again partly because of meteorologieaI applieation.
The arrangement consists of an annulus of fluid that rotates about its axis of
symmetry, this being vertieal; the outer wall is heated and the inner wall is eooled
(Fig. 15.16). Conveetion sets up a flow in which hot fluid moves upwards and
inwards and cold moves downwards and outwards. Coriolis forees aet on this to
give an azimuthal flow in the same sense as the rotation near the top and in the
opposite sense near the bottom. This beeomes dominant when the rotation is fast
enough.
This flow ean then develop instabilities, such as that in Fig. 15.17, whieh shows
seetions through the same flow at three different depths. The development of a
number of regular eelIs (three in Fig. 15.17) eauses the azimuthal flow near the top
and bottom to meander in a wavelike fashion. As a non-linear development when the
amplitude of the instability becomes Iarge, this cireuIation is eoneentrated into a
narrow fast meandering stream with rather slow closed eddies in the regions between
this and the walls (as ean be seen in the first and third pietures of Fig. 15.17). The
number of waves round the annulus varies with the parameters eoneemed - the
geometry and the non-dimensional forms of the temperature differenee and
rotation rate. The waves ean also develop less regular and/or less steady forms,
either beeause the flow region is close to the transition from one wave number to
another or beeause the whole wave pattem itself beeomes unstable .

.Q
/
I
I
Thermally in sulating

.,
I T, T1 (> T,I
boundaries
I
I
I
I
.. vv
~Fluid
region
I
I
I AXIS of
symmetry

Figure 15.16 Arrangement for convection in a rotating annulus.

Flow in Rotating Fluids 183

Figure 15.17 Patterns produced by suspended partides at three levels in rotating conveeting
annuius: (a) near top, (b) in middle, (e) near bottom. From Ref. (99).
16
Stratified Flow

16.1 Basic Concepts [33,34]

The name, stratified flow, is applied to a flow primarily in the horizontal direction
that is affeeted by a vertieal variation of the density. Such flows are of eonsiderable
importanee in geophysical fluid meehanics. The obvious ease of the effeet of
vertieal temperature variations on the wind near the ground is onlyone of a
number of examples in the atmosphere, and the effeets of both temperature and
salinity variations play an important role in manyaspeets of dynamical
oeeanography.
The density may, in general, either inerease or deerease with height. The former
ease gives rise to an interaetion between the mean flow and the eonveetion that
would oeeur in the absenee of mean flow. One example is the alignment of Benard
eells by a mean shear [49, 184], illustrated in the laboratory by Fig. 16.1. This
shows an illuminated eross-seetion, perpendicular to the flow, of an air ehannel
with heated bottom and eooled top; the smoke has been introdueed a long way
upstream and so the pattem indieates the oeeurrenee of regular rolls with their
axes along the flow. Another example - turbulent motion originating partly from
a mean flow and partly from eonveetion - will be considered in Seetion 22.8.
However, in this ehapter we are primarily eoneemed with the ease of stable
stratifieation, that is to say the density decreases with height. Vertieal motions
then tend to earry heavier fluid upwards and lighter fluid downwards, and are thus
inhibited. This inhibition may take the form of modifying the pattem of the
laminar motion or of preventing or modifying its instability.
We require a quantitative eriterion for this to be astrong effeet. Sinee most of
the experiments on stratified flows have used salt rather than heat as the stratifying
agent (ef. Seetion 13.5) we shall retain the density variations explicitly, rather than
relating them to temperature variations. We eonsider the ease of flow outside
boundary layers at high Reynolds and Peclett numbers, so that both viseous and
diffusive processes are negligible. Thus we write the momentum and density
equations (for steady flow)
PU"VU = -Vp + pg (16.1)
U"Vp=O (16.2)

tFor brevity, we retain the names, Peclet number and (subsequently) Prandtl number,
although, when salt is the stratifying agent these now refer to UL/ "e and v/"c (the Schmidt
number), where "e is the diffusivity of the saIt.
Stratified Flow 185

Figure 16.1 Cross·section of convection eelIs in channel flow; Ra = 4.16 x 10 3 , Re = 8.3.

(Note: patterns at top are reflection of eelIs in channel roof.) From Ref. [49).

We take z vertically upwards and suppose that the basic stratification consists of a
uniform density gradient (-dPoidz). Because Po does not vary horizontally, the
balanee between Pog and the hydrostatic pressure can be subtracted out from
equation (16.1) just as it can for an entirely uniform density (Section 13.2).
We now consider, superimposed on this basic configuration, a flow with length
and velocity scales L and U, produced, for example, by moving an obstacle of size
L horizontally through the fluid at speed U. This will produce a modification of
the density field which we denote by p', related to the stratification by equation
(16.2) in the form
U' 'VP' + wdPo/dz = 0 (16.3)
In order of magnitude

p'~ ~Id~o I (16.4)

Wis now restricted by the fact that the flow cannot produce buoyancy forees
associated with p' that are larger than the other forees invoIved. Since the buoyancy
force does not contribute directly to the horizontal components of equation (16.1)
it is convenient to work in terms of the vorticity form of this equation:

a'
p(u''Vw-w'\7u) =-g ( xL a ')
_yL (16.5)
ay ax
Since the order of magnitude of w is UIL this indicates that the order of magnitude
of p' must remain not greater than
p' ~ PolP /gL (16.6)
Comparison of this with (16.4) indicates that

W/U~ polP/gL2! ~o I= (Fr)2 (16.7)

When (Fr)2 is small the horizontal motion has only much weaker vertical motion
associated with it.
Fr is called the internaI Froude number, Of, when as at present, there is no
danger of confusion with the Froude number associated with free surface effects
(Section 7.4), simpIy the Froude number. 1/(FrY is sometimes known as the
Richardson number (see also Section 22.8).
186 Physical Fluid Dynamics

Similar analysis can be given for flows in which viscous and/or diffusive effects
are strong. This is a matter of some complexity, since different detaile d treatments
are appropriate for low, intermediate and high Prandtl number. Thus we omit
consideration of it; when we talk below of low F roude number flows, it is assumed
that any other criterion for the flow to be strongly constrained by stratification is
also fulfilled.
Often low Froude number motion can be considered to be entirely two-
dimensional in horizontal planes. For example, in the relative movement between a
spherical obstac1e and a stratified fluid, neady all the fluid is deflected to the sides
of the sphere, little above and below it. Thus the flow pattem in a horizontal plane
has a c10ser resemblance to unstratified flow past a cylinder than to unstratified
flow past a sphere.
This is illustrated by Fig. 16.2, showing patterns produced by dye originating
on a sphere in this type of flow. The view from above (Fig. 16.2(a)) is very similar
to patterns observed in flow past cylinders (e.g. Figs. 3.4 and 3.8). In contrast the
sideview (Fig. 16.2(b)) shows none of the structure of such pattems; for it to do so,
there would have to be marked vertical motions.
An indirect consequence of this is that the instability in the wake of a sphere in
a stratified fluid also resembles that in the wake of a cylinder, as is illustrated by
Fig. 16.3 (ef. Fig. 3.5).

Figure 16.2 Plan·view (a) and side-view (h) of flow past a sphere in a stratified fluid;
Fr = 0.21, Re = 164. From Ref. [92).
Stratified Flow 187

Figure 16.3 Wake of a sphere in a stratified fluid; Fr =0.8, Re = 377. Photo by W. R. Debler.
Reproduced by permission from 'Periodie Flow Phenomena', by E. Berger and R. Wille,Ann.
Rev. F1uid Mechanics, Volume 4. Copyright © 1972 by Annual Reviews Inc. All rights reserved.

16.2 Blocking

Clearly a geometry for which the motion is two-dimensional in vertical planes will
lead to radieal differences between strongly stratified and unstratified flow.
Consider, for example, relative motion between the fluid and a horizontal cylinder
which extends right ac ross the flow. No fluid can be deflected round the cylinder
without vertieal motion. If this is prevented, all the fluid in front of or behind the
cylinder must be at rest relative to the cylinder no matter how far upstream or down·
stream one goes. F ormally, if v = 0 because of two-dimensionality and w = 0
because of strong stratification, the continuity equation becomes
au/ax = 0; (16.8)
then for any z at which u is zero at one value of x, u is zero at all x .
There is an evident similarity between this phenomenon, known as blocking, and
a longitudinal Taylor column (Section 15.4). Indeed a fairly detailed analogy can
be drawn between two-dimensional motion in stratified and rotating fluids, the
gravitational and gyroscopic constraints producing a range of comparable
phenomena [285]. The analogy is elosest when the stratified fluid has a Prandtl
number around unity.
Figure 16.4 shows an experimental demonstration of the blocking effeet. A
cylinder was traversed slowly from right to left through a very long channel con·
taining salt-stratified water. A dyed region was introduced initially ahead of the
cylinder, but at a position that, when the photograph was taken, was far to the
right. The interpretation is thus similar to that of Fig. 15.4. Not all the blocked
region is necessarily dyed, but the presence of dye far to the left of the cylinder
does demonstrate strong blocking action.
188 Physical Fluid Dynamics

Figure 16.4 Blocking ahead of a cylinder in a stratified fluid; see text for experimental
procedure. From Ref. [73).

As with Taylor columns, the simple theory provides a useful understanding of

why the phenomenon occurs but little information about the detailed motion. In
fact, low Froude number flows past obstac1es observed in the laboratory often
show marked asymmetry. Blocking occurs strongly ahead of the obstac1e, as seen in
Fig. 16.4, giving rise to a region sometimes known as an upstream wake, but the
downstream effect may be much weaker. Almost certainly this difference has been
particularly noticed because most experiments have been performed with salt in
water as the stratifying agent. This has a very low diffusivity, giving a very high
Prandtl (Schmidt) number. We can understand the behaviour for such a fluid in the
following way [73, 117] .
In any real situation, the disturbance produced by relative motion of fluid and
an obstac1e can extend only a finite, even if very large, distance ahead of the
obstac1e. Fluid initially far ahead of the obstac1e must u1timat~ly pass above or
be10w it into the downstream region. This requires terms neglected in the simple
prediction of blocking to become significant. We may anticipate that this comes
about through the common process of a difference in length scales in different
direetions. Ahead of the obstac1e it is not in fact necessary to have a smalllength
scale appearing; the required effect can come about through the difference between
the height of the obstac1e, L, and the length of the blocked region, A, as sketched
in Fig. 16.5. For the same reas on as with equation (16.5), we work in terms of the
vorticity equation; we make the simplifications appropriate to two-dimimsional
flow, but, for a reason which will become apparent, we now inc1ude the viscous
term.

(16.9)

Because we are considering high Prandtl number, it is consistent to continue to

Figure 16.5 Length scales L, /1., and 6 in stratified flow past an obstac1e (see text). I - blocked
region ahead of obstac1e; II - boundary layer region behind obstac1e.
Strati/ied Flow 189

negleet diffusion of the stratifying agent. Then every fluid partide eonserves its
density throughout its motion (the meaning of equation (16.2)). The horizontal
density differenees produeed by some fluid partides rising a distanee ~L will be
~L I dpo/dz I. Remembering also that 11 ~ U/L, we ean write down the orders of
magnitude of eaeh of the terms in equation (16.9); using the same method as for
the analysis of boundary layers in Seetion 11.2, the three terms are respeetively
~U2 /XL, ~VU/L3 and ~gL I dpo/dz l/pX. These are in the ratios (Fr)2 :X(Fr)2 /ReL: 1
(where Fr is defined by equation (16.7) and Re is the Reynolds number ULIv). We
see that no matter how large X/L is, the inertia foree remains small eompared with
the buoyaney force when Fr is small. But the viscous force can become comparable
with the buoyancy foree if
X/L ~ Re/Fr 2 (16.10)
This result serves two purposes. Firstly, it provides a quantitative estimate of the
length of the blocked region. As expected it is long compared with the obstacle
size when Fr is small (provided Re is not too small).
Secondly, it indieates the physieal meehanism by which fluid partides pass the
obstade. Small viscous effects eome into play far ahead of the obstade and very

Figure 16.6 Flow around a cylinder in a stratified fluid. The dark band from top left is a
streakline originating weil ahead of cylinder and above blocked region; the rapid dropping down
of this fluid irnmediately behind the cylinder can be seen. From Ref. [73].
190 Physical Fluid Dynamics

gradually lift fluid partides (or lower them in the ease of fluid partides starting
below the centre plane) so that, by the time they reach the obstade, they are able
to pass over it.
The downstream behaviour ean now be understood. The partides eoming over
the obstade have retained their originaI density, which is markedly different from
that appropriate to the level they now occupy. They thus tend to drop baek to their
originalievel very rapidly, an effectthat ean be seen in Fig. 16.6 (a dose-up of the
cylinder in the flow previously shown in Fig. 16.4). Inertia and/or viseous forees
eome into play here; the argument leading to equation (16.8) breaks down here
through the appearance of a locallength scale li, small eompared with L (Fig. 16.5).
The subsequent downstream development will be referred to in Seetion 16.3.
This sequence of events depends on density diffusion being weak. For a fluid
with Prandtl number around unity, for example, diffusion acts on the density whilst
viscous action is lifting fluid, and partides passing over the obstacle differ in density
only slightly from other partides at the same level. The dropping-back process does
not then oecur. The analogy with Taylor eolumns leads one to expect the down-
stream region to be mueh more like the upstream.

r----------------10.-------------------------------,
z
(arbitrary units)

o
- 0.8 - 0.4 - 0 .2 0.4 0 .6 1.0 u
(arbitrary units)

-2

-.......------ 4

L-------________ -l 0L-______________________________ ~

Figure 16.7 Example of theoretical velocity distribution in upstream wake of body moving
from left to right. Reproduced, with permission, from 'Finite Amplitude Disturbanees in the
Flow of Inviscid Rotating and Stratified Fluids over Obstaeles', by R. R. Long, Ann. Rev. F1uid
Mechanies, Volume 4. Copyright © 1972 by Annual Reviews Inc. All rights reserved.
 Stratified F/ow 191

....
, 0

• I
'0 II
\

I, ,

.-" .. .....
'\
" '" ...
... ...
" \ '

Figure 16.8 Pattem produced by suspended partic1es in region ahead of a zero-incidence fiat
plate (out of picture to right) moving through a stratified fluid. From Ref. [195].

The minimum Prandtl number at which there is negligible downstream blocking

and the maximum at which there is effectively upstream-downstream symmetry
both depend in complex ways on the other parameters [106] . By appropriate
choice of obstac1e size and speed, it has been possible to observe nearly symmetrical
flows in a salt-stratified system [155]. There is not, however, so much experimenta!
information available as for the asymmetric flow described aboveo
The detailed structure of upstream blocked regions has been the subject of
several investigations. An interesting feature is the appearance of alternations in the
flow direction, giving a velocity profile of the form shown in Fig. 16.7 and
illustrated experimentally in Fig. 16.8.

16.3 Lee waves

The dropping-down process illustrated by Fig. 16.6 win (except at low Reynolds
number) be inertial. Overshoot and oscillations about the equilibrium position may
then occur.
A similar process may occur at any Prandtl number when inertial and stratifica-
tion effects interaet; that is at values of the Froude number around unity. The
result is a pattern of waves, known as lee waves, downstream of the obstade. An
example is shown in Fig. 16.9; the obstade in this case is on the floor of the
channel. Although each fluid partide is oscillating up and down as it travels down-
stream, the overall flow pattern is steady in the frame of reference of the obstade
192 Physical Fluid Dynamics
(apart from the fact that flows of this type are rather unstable and frequently
develop turbulent regions~
Lee waves have been extensively studied, partly because of the important
meteorological application to flow behind hill ranges [27] . Both theory and
experiment have shown that a wide variety of different pattems can occur
depending on the Froude number and the geometry; the overall depth of the
channel as well as the height of the obstade influences the structure of flows like
that shown in Fig. 16.9 [34,90].

Figure 16.9 Example of lee waves produced by towing obstacie (from left to right) along
floor of tank of stratified fluid; the camera moves with the obstacie to give approximately
steady flow pattem. From Ref. [170 J .

16.4 InternaI waves

The tendency for a displaced fluid partide in a stably stratified fluid to retum to its
originallevel means that waves can be generated in a variety of situations, of which
the lee wave phenomenon is just one example. In this section, we examine some of
the properties of these intemal waves in the simplest context of a small·amplitude
wave in an expanse of fluid that is otherwise at rest.
An important parameter associated with this situation may be introduced (and
its physical significance illustrated) by considering first the behaviour of a small
fluid partide that is displaced vertically. Suppose its density - which it conserves -
is poeD). If it is displaced adistance Az upwards, the density of the fluid surraund-
ing it is poeD) + Azdpo/dz and the net gravitational force on it is -gAzdpo/dz.
Hence, its motion is govemed by the equation
d 2 Az dpo
Po 7 =g dz Az (16.11)

and it oscillates about its original position with an angular frequency,

N= (_.K dP O) 1/2
(16.12)
Po dz
(dpo/dz being negative).
Stratified Flow 193

N is ealled the Brunt-Väisälä frequeney. When the density variations are due to
temperature variations
N = (ga dTo/dz)l/2 (16.13)
When, in addition, the adiabatie temperature gradient (Seetion 14.3) is
significant
dTo gOtTo )] 1/2
N= [gOt ( - + - - (16.14)
dz ep
The above analysis does not, of course, describe any actual fluid dynamical
situation. For that we must turn to the full equations of motion. We consider a
wave of small amplitude in an inviscid, non-diffusive fluid. Henee, we require the
equations of unsteady motion, but we omit the non-linear terrns on the basis that
these must be negligible when the amplitude is small enough:
poau/at= -'VP +g~p (16.15)
yt. u = 0 (16.16)
a~/at + wdpo/dz = 0 (16.17)

~ is the departure of the density from its basic distribution, Po (z), and P is the
corresponding departure from the hydrostatic pressure. We note that, although the
problem is linearized, the term u· 'il p in the density equation stiIl enters through
the interaetion between the vertieal velocity component and the basic stratification.
We look for wavelike solutions, periodie in both space and time:
u=Uexpi(wt+kxx+kyy+kzz) (16.18)
p =P exp i(wt + kxx + kyY + kzz) (16.19)
~ = Q exp i(wt + kxx + kyY + kzz) (16.20)
(where, of course, the real parts correspond to the physieal quantities). Substitu-
tion of these into equations (16.15) to (16.17) gives
iwpoU= -ikxP (16.21)
iwpo V= -ikyP (16.22)
iwpo W = -ikzP - gQ (16.23)
ikxU+ikyV+ikzW=O (16.24)

iwQ+dpo W=O (16.25)

dz
where (U, V, W) are the components of U. Elirnination of U, V, W, P and Q from
these homogeneous equations shows that they are eonsistent when and only when
k 2 ) 1/2
(
W = N 1 - k2 + k~ + k 2 (16.26)
x Y z
This may also be written
w = N(k 2 - k;)1/2/k (16.27)
194 Physical Fluid Dynamics
or
w=Nlsin81 (16.28)
where 8 is the inclination to the vertieal of the vectorial wave number
k = Mx + yk y + zk z and k = 1k I.
Thus waves exist for any value of the angular frequency from zero up to the
Brunt-Väisalä frequency N. Above N there are no wavelike solutions. (Since, in an
unstratified fluid, N = 0, this gives confirmation that the waves are essentially a
consequence of the stratification.) When k z = 0, corresponding to a wave pattem
without vertical variation, w = N; an array of vertical columns, with the velocity
varying in the horizontal direction, oscillates at the Brunt-Väisälä frequency - as
might be expected from its simple derivation above. When k x = k y = 0, correspond-
ing to a wave pattem without horizontal variation, w = 0; this corresponds to the
blocking phenomenon in steady flow discussed in Section 16.2.
The phase velocity of the waves is a vector in the direction ofk with magnitude
w/k.
Of greater physieal importance is the group velocity, indieating the speed and
direction with which kinetic and potential energy are transmitted through the
fluid. The waves are dispersive in a rather unusual way; the frequency does not
depend on the magnitude of the wave number but it does depend on its direction.
The standard result that for a wave in one dimension the group velocity is dw/ dk
may be extended to three dimensions giving
_ aw _ aw _ aw
C =x-+y- +z- (16.29)
g ak x ak y akz
For the present situation
Nk
Cg = k 3 (k 2 _ ~;)1I2 [xkxk z + ykyk z - z(k 2 - k;)] (16.30)

Since the properties of the waves are obviously axisymmetric about the vertical, we
discuss them in terms of the case for whieh there is no variation in the y-direction;
that is k y = O. Then
Ncos8
Cg = - k - (i eos 8 - zsin 8) (16.31)

It is interesting to compare this with the corresponding phase velocity:

Nsin8
cp = - - (i sin 8 +z eos 8) (16.32)
k
The group and phase velocities are mutually perpendicular. Energy is thus trans-
mitted along lines in the planes of the wavefronts. It is this that gives the waves their
rather unfamiliar character - although, as noted in Section 15.6, analogous waves
can occur in a rotating fluid.
The other direction that is of interest in providing an understanding of the
structure of the wave motion is the direction of motion of the fluid particles.
Equation (16.24), which may be written
U'k=O (16.33)
Stratified Flow 195

z
k and Cp

Wavefronts (lines of
intersection of surfaces
of constant phase with
x the xz-plane)

Motion of
Cg fluid particles

Figure 16.10 Summary of properties of internaI wave in stratified fluid.

shows that this motion is always perpendicular to the wave number and thus to the
phase velocity. The waves are essentially transverse. Moreover, since when k y = 0,
then v= 0 (equation (16.22», the motion is along the same line as the group
velocity. (More generally, equations (16.21), (16.22) and (16.30) show that the
x and y components of U and Cg are in the same ratio kxl ky-)
Figure 16.10 summarizes the above properties of the waves.
The applicability of this theory has been demonstrated experimentally.
Equation (16.31) shows that the group velocity is directed at an angle 8 to the
horizontal, independently of k. Since 8 is directly related to w, the angular
frequency (equation (16.28», all energy associated with a single frequency is
transmitted at this angle. In the experiments a long horizontal cylinder was
oscillated in a stratified fluid in the horizontal direction perpendicular to its axis.
This produced a two-dimensional wave pattem which was observed by a schlieren
optical method (Section 23.4) working on the refractive index changes associated
with the density changes. (A special system was used to eliminate the effect
associated with the basic stratification and to showonly the changes produced by
the wave perturbation.) Since the source is localized and the frequency fixed,
energy radiates from the cylinder only in four narrow bands at angle 8 (above and
below) to the horizontal. This expectation was confirmed by the experiments, as is
illustrated by Fig. 16.11(a). Experimental confirmation of the relationship between
the oscillation frequency and the orientation of the rays, as given by equations
(16.28) and (16.31), is shown in Fig. 16.12. When w exceeds N, no waves are
produced, as is shown by the contrast between Fig. 16.11(a) and (b).
The light and dark bands in Fig. 16.11(a) are lines of constant density pertur-
bation and thus, in an instantaneous photograph, lines of constant phase. These are
seen to run along the ray as in Fig. 16.1 O. Observations of the changing pattem
showed that they moved across the ray in agreement with the theoretical direction
of the phase velocity.
No prediction about the wave numbers present is given by the theory. For a
single w, all values of k are possible and all transmit in the same direction. In the
experiments, there will have been a range ofwave numbers present that Fourier
Figure 16.11 Waves produced by Vlbrating cylinder in a stratified fluid (dark vertiealline is
cylinder support). (a) w/N = 0.90; (b) w/N = 1.11. Ref. (185).
198 Physical Fluid Dynamics

synthesize to give the narrow rays with undisturbed fluid eIsewhere. The detailed
structure depends on the motion in the immediate vicinity of the oscillating
cylinder.
Superposition of waves of various frequency and wave number can give rise to
a variety of pattems. Such patterns may be observed so long as the fluid velocities
remain small enough for non-linear interactions to be negligibIe. We shall not
consideT this in detail, but we willIook brit;fly at one exampIe. Figure 16.13 shows

1.0 .-------r-----....,....---~---__,_---___,.

0.8

0.6

sir! 8

0.4

0.2

0.6 0.8 1.0

w/N

Figure 16.12 Experimental verification of relationship between frequency and ray angle for
waves in a stratified fluid. Different symbols correspond to different amplitudes of the driving
oscillation. From Ref. (185).

the pattem produced in an apparatus similar to the one used for the above oscillat-
ing cyIinder ex~eriments when a horizontaI cylinder was moved uniformIy in a
direction at 10 to the horizontal. One sees a wake and a wave pattem. Theflow in
the immediate vicinity of the cylinder is govemed by the full non-linear equations
of motion. But the motion aIso produces a far-fieId disturbance that wouId not
exist in a system that did not support waves. This can be understood in terms of
the above linear wave theory.
Stratified Flow 199

Figure 16.13 Pattem produced by motion of a horizontal cylinder through a stratified fluid
(dark verticalline is cylinder support). Ref. [238].

16.5 Stratification and rotation

It is dear that the frequencies 2D and N, in respectively a rotating fluid and a

stratified fluid, play analogous roles. Both can be shown by a crude argument to be
the angular frequency with which a displaced partide oscillates; and both are the
upper limit of the angular frequency range for which waves can occur. The relation-
ship between the role of the Rossby number in a rotating fluid and that of the
Froude number in a stratified fluid is readily seen if we write the latter in the form
Fr= UjLN (16.34)
When both rotation and stratification are present, their relative importance is
indicated by the parameter
S=N/2D (16.35)
Such situations arise in various geophysical contexts, and although we shall not
consider them in any detail, it is useful to have a criterion that indicates the type of
problem. A value of S around 1 indicates that rotation and stratification have
comparable influences; when S is small, one is dealing essentially with a rotating
flow modified by stratification, and when S is large it is the other way round.
200 Physical Fluid Dynamics
70

60

50

40
'Ala
30

20

10

0 0.2 0.3 0.4

S

Figure 16.14 Length 'A of a Taylor column (non·dimensionalized by sphere radius a) as a

function of stratification parameter S. Ro = 1.76 X 10- 2 , Ek = 8.0 x 10- 4 • From Ref. (89).

Nevertheless the modifieations ean be substantial even when S is not elose to 1.

For example, it is found that Taylor eolumn formation in a rotating fluid is strongly
reduced as S is ehanged from 0 to around 0.1, as may be seen in the experimental
results in Fig. 16.14. This shows the length of the transverse Taylor eolumn on a
sphere as a funetion of S (the length being defined as the distanee in whieh an
average fluid veloeity in the eolumn relative to the ambient fluid falls to I/e of the
sphere's veloeity).
17
Instability Phenomena

17.1 Introduction

Frequent mention has been made in foregoing chapters of transitions from one
type of flow to another (or from a state of rest to a state of motion) occurring as a
result of instability of the former. We now come to a systematic consideration of
such phenomena.
This chapter collects together a variety of examples of instability, emphasizing
the experimental observations. Sections 17.2, 17.3 and 17.4 concern instabilities of
re st configurations; that is they are cases where the instability is the cause of
motion. Sections 17.5 and 17.6 concern flow instabilities causing transition from
one type of motion to another. In most cases, instability leads not to a single new
flow pattern but a whole sequence. Some of the further developments, after the
initial instability, are also described in this chapter, but the amount of detail varies,
depending partly on whether the topic is considered elsewhere in the book.
üften one can formulate a physical argument indicating the basic dynamical
processes involved in an instability. Where appropriate, such arguments are given in
the present chapter; they can add greatly to one's understanding of the phenomena.
However, they do not give any quantitative indication ofwhen the instability
occurs. The next chapter will introduce the type of theory that has been deve10ped
for that purpose.

17.2 Surface tension instability of a liquid column

The first example is taken from a branch of fluid mechanics - flows with free
surfaces - generally outside the scope of this book. It is included because it relates
to familiar observations and because the instability has a simple physical
explanation.
The basic state is a cylindrical column of liquid held together by the action of
surface tension. As time proceeds, such a column develops corrugations in its shape
and ultimately breaks into discrete drops [116,120].
Theoretically, one may consider the column to be initially stationary; we are
dealing with the instability of arest configuration. In practice, of course, such a
column cannot be supported. The instability is investigated experimentally in jets
of liquid emitted from a circular hole. If the stress exerted on the jet by its
surroundings is negligib1e its velocity is uniform both across and along the jet. The
instability develops as the fluid travels away from the hole, resulting in a break up
some distance downstream. (Such jets differ from those considered in Section 11.8,
202 Physü;al Fluid Dynamics

a
A 8

Figure 17.1 Instability of water column - schematic.

in that we were there concemed with jets emerging imo the same ambient fluid -
for example air into air, or water into water - whereas we are now concemed with
jets emerging into surroundings with which there is negligible interaction - for
example water into air.)
The destabilizing mechanism may be understood in terms of Fig. 17.1, showing
the type of disturbance that grows. The column retains its circular cross-section but
with diameter varying along its length. The changed surface curvature will produce
a change in the pressure within the jet need ed to balanee the action of surface
tension. Provided that the wavelength is long enough the azimuthal curvature
dominates over the longitudinal curvature; Le. the radius of curvature is
effectively the loe al radius a. Hence, the pressure is a maximum where the radius is
a minimum (at, e.g., Ain Fig. 17.1) and a minimum where the radius is a maximum
(at, e.g., JJ). The pressure gradient thus pushes fluid in direetions that amplify the
original disturbance. (More precisely, in an inviscid situation, it accelerates fluid in a
direction that increases the velocity associated with the disturbance. In a viscous
situation it works against the viscous force in a way that increases the displacement
associated with the disturbanee.)
An altemative, equivalent, formulation of the physical cause of the instability
may be made in terms of the fact that a disturbance of sufficiently long wavelength
that conserves the volume of the jet decreases its surface area, and thus its surface
energyo
Figure 17.2 shows an instability of this type developing in a water jet. The highly
regular wavelength has been achieved in this laboratory investigation of the
instability by deliberately introducing a disturbanee, periodic in time, at the orifiee.
The break up of the jet into discrete drops, when the instability amplitude has
become large, can be seen at the right-hand side of Fig. 17.2. The drops can form
various quite complicated, but repeatable, pattems, of which an example is shown
in Fig. 17.3.
Although, in these pictures, the instability was promoted by a periodic distur-
bance, this is not an essential feature; break up of any suchjet will occur
spontaneously. This relates to the break up of a column of water from a tap (even
when the flow is fast enough for the fractional changes in velocity produced by

Figure 17.2 Initial development of capillary instability ofwater jet. From Ref. [116].
Instability Phenomena 203

r r I l I
'I I
<;
I

Figure 17.3 Example of subsequent development of instability in Fig. 17.2. From Ref. (116).

gravitational aeeeleration to be negligible) and to the familiar observation that the

surfaee ofwater poured from a kettle is wavy. (In many 'domestie' examples, the
situation is eomplieated by the jet not having an initially eireular eross-seetion, in
whieh ease one does not start with an equilibrium eonfiguration. However, the flow
from a laboratory tap - with for example a nozzle for fitting rubber tubing - may
eorrespond fairly closely to the ideal situation.)

17.3 Convection due to internaI heat generation

Sinee the most important example of instability of arest eonfiguration, Benard

conveetion, has been extensively deseribed in Chapter 4, we eonsider a related
situation to represent this type of instability in this ehapter. A horizontallayer of
fluid of large horizontal extent is contained between a thermally insulating bottom
boundary and a eonstant temperature top boundary. Heat is generated uniformly
throughout the vplume of the fluid. In laboratory experiments an approximation to
this may be aehieved through the ohmic dissipation of an alternating eleetrolytic
eurrent.
The heat so released is extraeted from the layer through the top boundary and
the temperature decreases upwards. The fluid is thus liable to the same sort of
instability as a fluid layer heated from below - as in Benard conveetion. Motion
oeeurs when a modified Rayleigh number gald 5 /vKk (in the notation of Section
13.2) exeeeds a critical value. As the Rayleigh number is inereased above this, the
flow goes through a sequenee of ehanges of eomparable eomplexity to, but
different in detail from, that deseribed in Chapter 4. As examples, Figs. 17.4 and
17.5 show plan and eross-seetional views of eellular eonveetion at Rayleigh
numbers far above critical. The former illustrates a tendeney for the horizontal
size of the eells to beeome large eompared with the depth of the layer. The latter
shows that the falling regions beeome mueh smaller, and eonsequently faster
moving than the rising regions. (The horizontal elongation is also apparent in this
figure, but is more diffieult to interpret beeause the seetion euts different polygons
of the plan view in different ways.)
One ean see why differenees from Benard eonveetion must arise at these high
Rayleigh numberso The Benard flow involves some fluid partides travelling on
closed paths at eonstant temperature (e.g. Fig. 4.12). This cannot occur when every
fluid particle is being internally heated. The ehanges in eell geometry shown in
Figs. 17.4 and 17.5 allow every fluid particle to lose heat by eonduetion during
eaeh eireuit, the elongation by inereasing the time for a eireuit to the same order as
the eonduetion time [273] , the narrow downgoing region by bringing all fluid
partides into the thermal boundary layer. That it should be the downgoing
eurrents that are narrow and the upgoing that are broad is, of eourse, due to the
204 Physical Fluid Dynamics

Figure 17.4 CeUs in internaUy heated conveetion. Ra = 4 x 10 4 , Pr = 6. Note: field of view

is about 35d square, where d is the layer depth. Flow visualization using beads that drop out
of hot fluid and rise out of cold fluid (owing to difference in expansion coefficient from water);
edges of polygons are regions of upward flow, centres downward flow; streaks due to beads still
in suspension can also be seen. Ref. [278].

asymmetry between heating over avolume and cooling over a surface; only cold
thermal boundary layers are formed (again a contrast with Fig. 4.12).

17.4 Cooveetioo due to surface teosioo variatioos

Figure 17.6 shows a cellular pattem that looks superficially similar to those
considered in Sections 4.3 and 17.3. In fact, the flow is produced by a different
instability meehanism, operative in a fluid layer with a free surface and depending
on the temperature variation of the surface teosion.
Figure 17.7 illustrates the process. A layer of liquid is bounded by a solid wall
from which it receives heat afld a free surface at which it loses heat. The surface
tension of a liquid decreases with increasing temperature. Consequently, if part of
Instability Phenomena 205

Figure 17.5 Vertical section through intemally heated conveetion. Ra = 2.5 x 10 5 • Flow
visualization using suspended partides illuminated only in one plane. Ref [93).

the free surface should become 10ca1ly hotter than the rest, as aresult of some small
disturbance, fluid is drawn away from the region by the action of surface tension.
Other fluid comes in from below the surface to replace it. This fluid has been
closer to the hot solid boundary and will thus be hotter than the fluid at the
surface. The original disturbance in the temperature distribution is thus amplified.
A steady state can be established with hot fluid moving towards the free surface,
losing heat as it travels paralleI to and close to the surface and then moving away
again as cold fluid. In this way a distribution of surface tension can be maintained
that drives the motion against the retarding action of viscous forees. It is found

Figure 17.6 Hexagonal eells arising in surface tension driven conveetion; aluminium powder
flow visualization (see Section 23.4). From Ref. [144).

cooled free
A boundary

- ~ -- - ---

Heated rigid boundary

Figure 17.7 Mechanism of surfaee tension instability; effeet ofloeal increase in temperature
atA.
206 Physical Fluid Dynamics
experimentally (e.g. Fig. 17.6) that this process ean produee an array of hexagonal
cells of much greater regularity than is observed in gravitational conveetion. The
fluid moves towards the surface at the centre of eaeh of the hexagons and away
from it around their peripheries.
Benard's original experiments on eellular conveetian aetually mostly showed
surfaee tension driven patterns (ef. Sectian 4.1) [228]. This led to the belief-
now discarded - that a hexagonal array was the primary type of convection.
Of eourse, one ean get systems in whieh both surfaee tension and gravitational
effeets are significant, but we shall omit consideration of these [190].

17.5 In8tability of rotating Couette flow

We retum to the flow between two eoneentrie eylinders rotating at different rates,
considered in Sectian 9.3. There we looked at the salutian applieable when the
flow remains entirely azimuthal. However, this simple flow does not always oeeur;
instability may lead to a mare eomplex pattern developing.
Although this is the instability of a flow, not of arest eonfiguration, there is a
signifieant similarity between it and Benard type instability (Chapter 4 and
Sectian 17.3), with centrifugal foreet replacing buoyaney foree as the destabilizing
mechanism. The analogy extends to some features of the full mathematica1 treat-
ment as well as to the physieal ideas being diseussed here [91].
The eause of the instability may be understood by considering a toroidal
element of fluid (i.e. in eylindrical polar co-ordinates, the fluid between rand
r + dr and between z and z + dz but at all valu es of ep). Suppase this is displaeed to
a slightly larger radius. If it is now rotating faster than its newenvironment, the
radial pressure gradient associated with the basie flow will be insuffieient to
balanee the centrifugal foree associated with the displaced element. The element
will then tend to move stiIl further outwards. Similarly, an element displaced to a
slightly smaller radius will tend to move still further inwards. There is thus an
instability associated with some distributions of angular veloeity. (The argument
cannot be applied to a fluid partic1e loealized in the [Link], as the displaeement
of this would introduee azimuthal pressure variations.)
The argument may be put into a quantitative form to indicate the canditian for
instability in the absenee of the stabilizing action of viseosity; Le. the
canditian analogous to the requirement for Benard instability that the temperature
must deerease with height. An elemental toroid, initially at radius t, and cireulating
at angular veloeity nr, is suppo~ed displaeed to radius f/ without interaeting with
the remainder of the fluid. Its angular momentum is then conserved and its
angular veloeity after displaeement is
nr =nrt 2/f/2 (17.1)

tOnly if I nl - n, I « nl is this a problem in rotating fluids in the sense of Chapter 15. The
following discussion is thus phrased in terms of a non-rotating frame of reference. The centri-
fugai. force considered is that acting on a particle moving on a circular path relative to this
frame. The case of small relative rotation could be discussed using a rotating frame, in which
case the action attributed here to the centrifugal force would appear as the action of the
Coriolis force.
Instability Phenomena 207

Its cenfrifugal force will exceed that of undisturbed fluid at '1'/, circulating with
angular velocity S1 11 if
I S1~ I> I S111 I (17.2)
Hence, there is instability if
I S1r~2 I> I S1 11 '1'/21 when '1'/ > ~ (17.3)
that is if

(17.4)

This is known as the Rayleigh criterion for the instability of Couette flow [37].
When the two cylinders are rotating in the same sense the flow is either stable
everywhere or unstable everywhere. Substituting in equation (9.6), criterion (17.4)
then becomes
(17.5)
When the two cylinders are rotating in opposite senses, the region elose to the inner
cylinder is unstable and that elose to the outer cylinder is stable.
The Rayleigh criterion is modified in real cases by the action of viscosity (see
Section 18.4).
Figure 17.8 shows photographs of some of the phenomena that re sult from the
instability and its subsequent deve10pments. These phenomena are quite
varied [235, 240] , and a complete account ofthe observations is not possible
here. By way of typical illustration, Fig. 17.8(a), (b) and (c) show developments
for a case in which the gap between the cylinders is relatively small compared with
the overall radius and in which the outer cylinder is at rest [82, 226] .
The first instability (occurring when the critical curve shown in Fig. 18.5 is
crossed) gives rise to an axisymmetric flow pattem, as shown schematically in
Fig. 17.9. Vortices of altemate senses circulate between the two cylinders. Each
vortex extends toroidaily right round the annulus. These vortiees are known as
Taylor cells. Figure 17.8(a) shows a photograph of this flow.
Increase of the rotation rate of the inner cylinder causes the Taylor ceils to
become wavy in the azimu thal direction as illustrated by Fig. 17 .8(b). The number
of waves around the annulus varies with rotation rate, but exhibits hysteresis, the
exact pattem occurring for a particular set of conditions depending on how those
conditions were approached. Howevee; the wave pattem travels round the annulus
at the same angular velocity - about one-third the angular velocity of the inner
cylinder - for all wave numberso
Further speed increases lead to further complexities in and greater irregularity
of the flow pattem. As in Benard conveetion, the re is a long range in which both
the regularity of a cellular pattem and the irregularities of turbulent fluctuations
can be observed (Fig. 17 .8(c)). Ultimately, the latter predominates and fuily
turbulent motion occurs throughout the annulus.
In some circumstances, turbulent motion appears suddenly without the
preceding stages of development. This occurs principally when the two cylinders
are rotating in opposite senses, but the outer one faster, so that the unstable region
according to the Rayleigh criterion is confined to a small fraction of the annulus
208 Physical Fluid Dynamics

Figure 17.8 Flow pattems in rotating Couette flow - see text for detaiIs. Aluminium powder
flow visualization (see Section 23.4). From Ref. [82].
Instability Phenomena 209

8. I
• ,I
----l
8, I
I
I
I
I
I
I
n, I
I
I
I
I
I
I
I
I
I
I

Figure 17.9 Arrangement of Taylor eells.

elose to the inner cylinder. However, the turbulent region is not so confined, and it
is likely that this instability is associated more with the shear in the annulus than
with the action of the centrifugal forees. Its mechanics have thus more in common
with the effects to be described in Section 17.6 than with the instability giving
Taylor cells.
The stage preceding full turbulence is then intermittent turbulence, similar to
that observed, for example, in pipe flow (Sections 2.6 and 19.3). For certain
ranges of the imposed conditions, this intermittent turbulence is produced by a
elearly identifiable spiral structure (Fig. 17.8(d)), [82,283] . A spiral band of
turbulent motion and a siInilar band of larninar motion form a pattem !ike the
stripes on a barber's pole. This pattem rotates roughly at the mean angular velocity
of the two cylinders, so that, at a fixed position, altemately laminar and turbulent
motions are observed. The motion is also altemately laminar and turbulent if one
follows the trajectory of a fluid partiele, so that continuous transition from laminar
to turbulent motion and continuous reverse transition are involved in the mechanics
of the structure.

17.6 Shear flow instability

A variety of the most important cases of the instability of fluids in motion fall into
the general category of the instability of shear flows. A shear flow is one in which
the velocity varies principally in a direction at right angles to the flow direction
(Fig. 17.1 O(a)). The simplest example of this, although it does .not correspond
immediately to a physical situation, is a flow with a finite discontinuity in the
velocity as shown in Fig. 17.1 O(b). The fact that this is subject to instability - the
Kelvin- Helmholtz instability - may be seen as follows. We consider the develop-
ment of a disturbance in the frarne of reference in which the two velocities are
equal and opposite (Fig. 17.10(c) and (d)), so that, by symmetry, the disturbance
210 Physical Fluid Dynamics

(d) _ _ _ _--:.. ;::::=r=-___

Figure 17.10
(e)
A_
B+ +_
0- ..._ _ +--
To illustrate the cause of Kelvin-Helmholtz instability - see text.

does not travel with the flow. We now suppose that the boundary between the fluid
mo ving in one direction and that in the other becomes slightly wavy (Fig. 17.10(e)).
This will make the fluid on the convex side (at, e.g., A and D) move slightly faster
and that on the concave side (at, e.g., B and C) slightly slower. In a steady state,
Bemoulli's equation would apply. Pressure changes indicated by the + and - signs
in Fig. 17.1 O(e) will be produced by the disturbance. Thus the steady disturbed
state is not possible and the pressure gradients are in directions producing
amplification of the disturbance. More preeisely, the fluid cannot support a
pressure discontinuity and the unsteady flow must involve motions that counteract
the above action of Bernoulli's equation. These are an acceleration of fluid above
A and away from it and an acceleration below B and towards it. Such acclerations
amplify the disturbance.
The experimental observations most closely related to this model have been
made in stratified fluids (Chapter 16), primarily because the stratification provides
a convenient method of producing a shear flow with little variation in the flow
direction. A long horizontal channel was filled with liquid, dense at the bottom,
lighter at the top. It was then tilted slightly so that the heavier fluid flowed down
the sloping bottom wall and the lighter fluid up the sloping top wall, produeing a
steady stratified shear flow. Experiments were performed both with a sharp density
discontinuity between two immiscible fluids (with surface tension acting at the
interface) and with a continuous density variation in miseible fluids. Figures 17.11
and 17.12 show examples of the instability for the two cases respectively, whlm the
destabilizing mechanism described above is strong enough to overcome the
stabilizing action of the stratification. The regularity of the pattems du ring their
development is noteworthy, but this does subsequently break down giving a
turbulent motion.
Examples of shear flow instabilities in the absence of stratification appear in
various places in this book. The velocity profiles (Fig. 17.13) of pipe flow, a
boundary layer, a wake, ajet, and a free convection boundary layer all come into
this category. In practice all such laminar flows break down when the Reynolds
number based on the length scale over which the shear occurs, is high enough,
although not all cases are unstable to small disturbances in the sense to be explained
in Chapter 18.
Instability Phenomena 211

Figure 17.11 Shear instability at the interface of two immiscible liquids, the lower being
denser. From Ref. [264].

The deve10pment of the instability is very sensitive to the details of the ve10city
proftle. The critical Reyno1ds numbers of different flows may be widely different
and, as is illustrated by Figs. 2.9, 3.5,14.6,14.9,17.12,19.5,19.6 and 22.23(a), the
cünsequences of the instability are very varied. Clearly, any genera1izations about
the physical processes must be treated with caution. However, the velocity proftles
may be dassified into two broad groups (Fig. 17.13): those with a point (or points)
of inflection, such as a wake, ajet, or a free convection boundary layer; and those
with no point of inflection, such as pipe flow or a forced flow boundary 1ayer (in
zero or favourab1e pressure gradient) [21, 164].
For the first group, processes broadly similar to the Kelvin-He1mholtz
instability described above come into playand the flow is always unstable at high
enough Reynolds number. The role of viscosity is prirnarily stabilizing, preventing
the instability at low Reynolds numberso
Flows of the second type are not subject to this instability, and there is the
possibility (in theory rather than in practice) that they remain stable at all valu es of
the Reyno1ds number. However, they frequently exhibit a different kind of
instability that is essentially a consequence of viscous action. Viscosity now p1ays a

Figure 17.12 Shear instability in stably stratified fluid, exhibited by the lower denser fluid
being dyed. From Ref. [265].
212 Physical Fluid Dynamics

(a) (b) (c) (d) (e)

Figure 17.13 To illustrate that the velocity profiles of (a) pipe flow, (b) a boundary layer,
(c) a wake, (d) ajet, and (e) a free convection boundary layer are all shear flows.

dual role, providing the destabilizing mechanism at high Reynolds number, but stilI
damping out the instabitity at low Reynolds number.
The physical processes underlying these statements are subtle, and here we
consider just one point. The process transferring energy from the mean flow to the
disturbance to be described for turbulent flow in Section 20.4 can come into play
in any shear flow for a large enough disturbance of the right structure. It may often
come into play for a small disturbance, but when there is no point of inflection the
disturbance can acquire the appropriate structure only through the action of
viscosity.
We shall not dwell here on the consequences of the instability of shear flows.
Most of the important cases are described in other chapters, with detaile d
discussions ofpipe flow (Sections 2.6 and 19.3), the wake behind a circular
cylinder (Section 3.3), boundary layer trallSition (Section 19.1), and transition in
ajet (Section 19.2). In many cases, the final result far enough downstream is fully
turbulent motion, at any Reynolds number for which the laminar flow breaks
down. (There is often the complication that the Reynolds number increases with
distanee downstream.) A notable exception to this is the Karman vortex street
(Sections 3.3 and 11.7).
18
The Theory of Hydrodynamic Stability

18.1 The nature of linear stability theory

A large body of theoretieal work has been developed in attempts to understand

and predict phenomena of the sort described in the previous chapter. In most cases
the cause of the instabitity can be formulated in a way that gives some physical
understanding of the processes involved. However, this is usually opposed by some
stabilizing process, such as the damping action of viscosity, and the physical argu·
ment provides no precise information about when the instability will occur. It also
provides little indication of the detailed structure of the motion resulting from the
instability.
This chapter describes the simplest type of theory that has been developed to
make quantitative predictions about these matters -linear stability theory. We
shall see that the re are limitations to the information given by this and so one needs
to turn to the much more difficult non·linear stability theory for further informa·
tion [241]. However, that is beyond the scope of this book. Even the linear theory
will be presented without mathematical detail. Thus this chapter is essentially an
elementary introduction to the subject, formulated in a way that should provide
those who wish to apply the results of this type of theory with some understanding
of its nature and limitations.
The underIying notion is that transition from one type of flow to another results
from spontaneous amplification of disturbances present in the original flow. One
would then expect the occurrence of the transition to depend on the intensity and
structure of disturbances present, and this is frequently found to be the case.
ClearIy, a theory covering all possibilities would be unmanageably complex;t more·
over, comparison with experiment would be difficult since the nature of the distur·
bances is often not in complete experimental control.
Unear stability theory adopts a less ambitious objective - to ascertain when a
flow is unstable to infinitesimal disturbances. It thus gives no prediction about
transition promoted by sufficiently large disturbances; this may occur when the
theory indicates stability. onthe other hand, it should indicate conditions in which
the flow cannot remain in its postulated form and must undergo transition to
another type of motion. Infinitesimal disturbances are always present, even in the
most carefully controlled experiment, and if these tend to be amplified, the flow
will necessarily break down.

tThere is also the rather vexed question of how large a 'disturbancll' can be and stili be
regarded as a disturbance rather than a change in the basic flow.
214 Physical Fluid Dynamics

The approaeh is analogous to that used to understand instability of solid

mechanieal systems. A pin stood on its point with its eentre of gravity directly
above the point is in equilibrium. This equilibrium is, however, unstable beeause
even the smallest displaeement of the pin will be amplified. Sinee infinitesimal
disturbanees ean never be eliminated, the pin always falls.t
The linear stability theory for a partieular flow starts with a solution (or approxi-
mate solution) of the equations of motion representing the flow. One then eon-
siders this solution with a small perturbation superimposed, and enquires whether
this perturbation grows or deeays as time passes. All terms involving the square of
its amplitude are negleeted; it is this that limits the theory to infinitesimal
disturbanees.
This linearization of the equations provides ameans of allowing for the many
different forms that the disturbanee ean take. Any pattem of disturbanee may be
Fourier analysed spatially. Beeause of the linearity, there are no interaetions
between different Fourier eomponents. The equations may thus be broken down
into separate sets of equations for eaeh Fourier eomponent, indieating the stability
or instability of that component. All eomponents ean, of eourse, be treated in a
single mathematieal operation by retaining the wave number (indicating the
partieular Fourier eomponent) as a parameter.
The theory for a single Fourier component eonsiders a perturbation whose
general form is proportional to
exp[ikor + ut] (18.1)

The exaet meaning of this statement varies from problem to problem; two examples
will be considered below. In general, U is eomplex
U = U r + iUj (18.2)
The growth or deeay of the perturbation is determined by the sign of U r , the real
part. If U r is positive, the Fourier component under eonsideration is amplified; this
amplifieation willlead to the breakdown of the originaI flow. If U r is negative, the
Fourier component dies away and the originaI flow ean remain.
Henee, if U r is negative for all values of k, the originaI flow is stable to all infini-
tesimai disturbanees. This is a neeessary eondition for stability. If U r is positive for
some values of k, the eorresponding perturbation will be spontaneously amplified.
This is a sufficient eondition for instability.
The signifieanee of these remarks is most readily seen through particular
examples.

18.2 Onset of Benard convection [37]

The first ease whieh we shall eonsider in more detail is Benard eonveetion -
described in Chapter 4 and Seetion 14.7.
For an infinite expanse of fluid with the density inereasing upwards, the physi-
eally evident instability is eontained in the set of equations (16.1 5)-(16.17) for

t Buridan's ass -
which died of starvation because it was exactly rnidway between a bunch of
hay and an equally attractive pail of water - should have found a similar resolution to its
problerns.
The Theory of Hydrodynamic Stability 215

small disturbances in a stratified fluid. The square of the Brunt-Väisälä frequency

2 g dpo dTo
N =-- -=gOi- (18.3)
p dz dz
is negative in this case. Hence the frequency w of equation (16.28) is an imaginary
quantity.
w =ia (18.4)
and the amplitude of the disturbance is proportional to e± at, with a real and
positiveo The case e+atcorresponds to instability. (For instability one requires only
that there should be at least one unstable solution of the equations. The existence
of a second decaying solution, which would occur on its own only if one had an
initially large disturbance of just the right configuration, is irrelevant.)
For the Benard problem, one is concemed with the modiflcation of this by
viscosity and thermal conductivity, together with the boundary conditions. The
previously amplifying solution may now become a decaying one (whilst the
previously decaying one always remains so).
The initial condition is specified by the set of equations (14.40). A small per-
turbing velocity, temperature, and pressure field is superimposed. The perturbation
must, of course, satisfy the continuity equation and the boundary conditions. In
accordance with the ideas in Section 18.1 it is taken to bc of the form

u = U(z) exp[i(kxx + kyY) + at] (18.5)

t::.T= e(z) exp[i(kxx + kyY) + at] (18.6)

t::.p =p(z) exp[i(kxx + kyY) + at] (18.7)

(the physical quantities being, of course, the real parts of these complex quantities).
The Fourier analysis is thus performed in the two horilOntal directions. It is
necessary, in order to satisfy all the imposed conditions, to allow the perturbation
to have general distributions in the vertical direction; U(z), e(z), and p(z) are
determined as part of the solution.
The equations of motion are used to investigate the development of this pertur-
bation with time, as indicated by a. The results possess the following features.
Firstly, they depend only on the total wave number
k =(k; + k~)1/2 (18.8)
and not on k x and k y individually. This is to be expected as all horizontal direetions
are equivalent and the choice ofaxes x and Y is arbitrary. However, it implies that
there is a variety of flow pattems all ofwhich are simultaneously unstable.
Secondly, a is always real,
ai =0 (18.9)
The perturbation thus either grows continuously or decays continuously. Any
instability takes the form of an exponentially amplified disturbance.
The results on the sign of a may be presented on a graph of the non-dimensional
wave number,
~=kd (18.10)
216 Physical Fluid Dynamics

7r-----,,-----,------r-----,,----~----__,

0<0

~c"t - 1------------1
3

o 1000 12000 3000 4000 5000 6000

Raerot Ra

Figule 18.1 MarginaI stability eurve for Benard conveetion.

against the Rayleigh number

Ra = gcx.(T2 - T 1 )d 3 /VK (18.11)
as in Fig. 18.1. The locus of a = 0 separating the stable region with a < 0 from the
unstable region with a > 0 is independent of the PrandH number.t
For low values of Ra, a is negative for all ~. All disturbances tend to die away
and the rest configuration is stable. For higher values of Ra there is a positive a
solution for some values of ~. Hence, some disturbances are self-amplifying, and
the re is instability. A critical Rayleigh number Raerit is given by the lowest Ra at
which this is true (Fig. 18.l): Raerit = 1708.
This is one of the successes of linear stability theory. We saw in Section 4.2 that
fluid in a horizontallayer remains at rest when the Rayleigh number is le ss than
about 1700 and convects when it is greater. There haso been a number of experi-

trrhis result eould have been anticipated. On this locus, ou/ot is zero. u. Vu is also zero,
beeause there is no mean velocity field and second-order terms in the perturbation velocity
field are negleeted. Henee, the whole inertia term Du/Dt is zero. Dynamical similarity is then
governed by the Rayleigh number alone. The only balanee of terms to be considered is that in
the thermal equation, as indieated by (14.15) (no longer restricted to low Grashof number
when the inertia term is identieally zero).
The Theory of Hydrodynamic Stability 217

Amplitude

Figure 18.2 Schematic diagram of types of departure from exponential growth.

mental determinations of the Rayleigh number for the onset of motion, mostly
using the heat transfer method illustrated by Fig. 4.4. Two examples of the results
are: 1700 ± 50(Ref. [231)), 1790 ± 80(Ref. [263]). The agreement with the
theoretical critical Rayleigh number is good and the theory can be said to explain
the onset of motion.
One would hope that the theory might also make useful predictions about the
flow pattem resulting from the instability. However, there are complications. The
first is general to all problems in linear stability theory and concems the relationship
of a theory about exponentially amplifying disturbances to the steady or averagely
steady flow pattem resulting from the instability. As the exponential growth con-
tinues, the originaI assumption that the perturbation was small will break down.
Non-linear terms in the equations will become important and alter the exponential
dependence on time. This alteration may take the form either of a levelling off (as
in curve A of Fig. 18.2) or of a still faster growth (curve B). In the former case,
which applies to Benard conveetion, it is possible that the ultimate steady state
flow pattem resembles the unstable flow pattem oflinear theory. However, one
cannot know whether this resemblance should exist without reference to non-linear
theory. When the more catastrophic curve B applies, there must evidently be further
developments before a steady state is reached and there is likely to be little resem-
blance between the form of the first instability and its ultima te consequence.
Retuming specifically to Benard conveetion, a similarity between the unstable
flow patterns and the resulting convection may be expected. For values of the
Rayleigh number only a little above critical, there is only a small range of unstable
wave number. However, restriction to a single wave number does not uniquely
specify the flow pattem, for reasons connected with equation (18.8). Not only may
k x take a whole range of values for a given k, but also linear combinations of solu-
tions of the form (18.5) to (18.7) with different k x but the same k are single wave-
number solutions. In consequence, there are many single wave-number pattems,
some of a complexity that makes them not instantly recognizable as such [239].
All occur with equal probability on linear stability theory. One has to tum to non-
218 Physical Fluid Dynamics
linear theory or to experiment to discover that the actual pattem at slightly super-
critical Rayleigh number is a simple roll pattem such as that in Fig. 4.5.
Once the flow pattem is known one can compare the observed horizontallength
scale with that of unstable disturbances according to linear theory (i.e. that corres-
ponding to ~crit in Fig. 18.1). The two concur - the distance between the middle of
a rlsing region and the middle of a falling one is roughly equal to the depth of the
layer - indieating that the application of linear theory can be extended in this way.
These remarks apply only when the Rayleigh number is just a little above critical.
At higher values, the linear theory tells one little. Again many non-linear theories,
beyond the scope of this book, have been developed, but mu ch of our information
on this flow comes from experiment (Section 4.3).

18.3 OverstabiIity [37]

The result for Benard convection that ai = 0 does not apply to all systems. When
ai "*
0, the development of the instability in time takes the form of an amplifying
sinusoid, as sketched in Fig. 18.3. Correspondingly, the resulting motion ifnon-
linear effects limit the growth is an oscillatory motion.
This type ofbehaviour is frequently known as overstability (although the name
is not a good summary of the physical processes involved).

Perturbation

Figure 18.3 Overstability.

An example is convection in a horizontallayer rotating about a vertieal axis. This

exhibits overstability when the Prandtl number of the fluid is low enough. Figure
18.4 shows an oscillogram of the temperature difference between two level s
obtained from an experiment with mercury.
In general, overstability is liable to occur whenever the system has a feature that
can give rlse to wave motions. In the above example, the wave motions are those
associated with the rotation (Section 15.6).

18.4 Rotating Couette flow

We shall not discuss in any detail stability theory applied to rotating Couette flow -
whieh has been considered phenomenoIogicalIy in Section 17.5. However, it may be
mentioned that this is another of the major successes of linear stability theory
198, 257] and is historlcalIy important as the first case in which a detaile d compa-
The Theory of Hydrodynamic Stability 219

Figure 18.4 Temperature differenee oseiIlogram in rotating Benard conveetion. (Time

increases from right to left; at start of traee system it is stiIl warrning up and at end it is
switehed off.) From Ref. (109).

rison was made between the theoryand experiment. This was done by Taylor in
1923, observing the onset of cellular motion. Figure 18.5 shows his comparison for
one value of a2/al' The vertical axis corresponds to the case mainly considered in
Section 17.5 of only the inner cylinder rotating. Points to the right of this corres-
pond to both cylinders rotating in the same sense, and the dotted line represents the
Rayleigh instability criterion, equation (17.5), obtained ignoring the action of

4000 /
/
/

I
.- /
3300 I
/
Unnable /
/
I
/
/
/
/
/
.- /

Observed Points • /...-- n, ~ ~

Calculated Points 0 / nl a~
.-.-
/

Stable /
/
/
/

- 4000 - 3000 - 2000 - 1000 o 1000 2000 3000

nla~
II

Figure 18.5 Comparison of theoretieal and experimental eonditions for instability of rotating
Couette flow with al/al = 1.14. From Ref. (257).
220 Physical Fluid Dynamics
viscosity. The left-hand part of the figure corresponds to the two cylinders rotating
in opposite senses.
The sudden appearance of turbulent motion, not preceded by cellular motion,
mentioned in Section 17.5, occurs at much higher values of I n2a~ lv I and is not
predicted by the theory.

18.5 Boundary layer stability

As a second example for more detailed consideration, we look at the stability of a

two-dimensional zero-pressure-gradient boundary layer. This is typical of a shear
flow instability, and much of the following description applies also to the other
flows mentioned in Section 17.6. The boundary layer results have, however, been
tested experimentally in more detail than most other cases.
The breakdown of laminar flow in the boundary layer is supposed to arise essen-
tially from the instability of the Blasius profile (Section 11.3). This profile, with
given values of the free-stream velocity Uo and boundary layer thickness 0 and thus
of the Reynolds number
Re = uoolv {I 8.12)
is taken as the basic flow upon which the perturbation is superimposed. The varia-
tion of 0 with x is ignored for the stability calculation.
A two-dimensional disturbance of the form
!lu = U(y) exp(ikx + at) (18.13)
!lv = V(y) exp(ikx + at) (18.14)
!lp = P(y) exp(ikx + at) (18.15)
is considered (where the co-ordinates are the same as in Section 11.3). The fact that
the basic flow is two-dimensional is not immediately a justification for taking the
disturbance to be two-dimensional; a two-dimensional flow can undergo a three-
dimensional transition, and indeed, the later stages of development and the resulting
turbulent motion are strongly three-dimensional. However, there is aresult, known
as Squire's theorem, that in linear stability theory the critical Reynolds number for
a two-dimensional paralleI flow is lowest for two-dimensional perturbations [21,36].
We may thus restrict attention to these.
The perturbation expressed by equations (18.13)-(18.15) is a single Fourier
component in the flow direction. As for Benard conveetion, the perturbation must
have an unprescribed form in the normal direction - a feature that might be
expected from the fact that the growth of the disturbance will not extend appre-
ciably outside the boundary layer where there is no shear.
The disturbance will be carried downstream by the flow. Its periodicity in the x
direction will produce a periodicity in time. We may thus anticipate that
ai =1= 0 (18.16)
- for a reason different from that considered in Section 18.3.
As before, the stability or instability is indicated by the sign of aro The marginal
condition is given by the' locus of ar = O. For shear flows in general, this locus on a
Reynolds number versus non-dimensional wave number plot takes one of the forms
The Theory of Hydrodynamic Stability 221

kö

Re

Figure 18.6 General forms of stability loops for shear flows.

shown in Fig. 18.6. There is instability inside either loop A or loop B and stability
everywhere outside it. A loop of the form A, with the two limbs asymptoting
towards one another as Re ~ oo applies when the velocity profile has no point of
inflection at which the magnitude of the vorticity is a maximum. The Blasius profile
is of this kind. A loop of the form B applies for a flow, such as ajet, that does have
such a point of inflection. (See Section 17.6 for a discussion of the differences in
the destabilizing mechanism between these two types of flow.)

3 X 10 -4

o~~--~----J------L----~
I 2000 4000 6000 8000
Re : uo6
JJ

Figure 18.7 Stability loop for Blasius profIle from Ref. (229). (Note: two limbs asymptote
to one another at much larger Re.)
222 Physical Fluid Dynamics

For comparison with experiment, it is more convenient to specify the particular

Fourier component by its frequency rather than its wave number. The theory
determines the speed at which each component travels downstream, so the two
specifications are equivalent. Figure 18.7 shows the results for the Blasius profile in
this way, Ojv/uõ being the most appropriate non-dimensional form of the frequency
Oj. Ö is taken as the 99 per ee nt thickness (equation (8.14)). The curve shown gives
the critical Reynolds number below which no disturbances are amplified as
Re erit = 1200. (However, this result is sensitive to the detailed computational
procedure and different workers' results vary by around 20 per cent [183].)
The instability of a single Fourier component, as considered by the theory, gives
rise to a wave-like motion, periodic in both space and time. Such motions are called
Tollmien-Schlichting waves. Their occurrence in the spontaneously occurring
breakdown of laminar flow depends on the selective amplification of a particular
frequency - or narrow band of frequencies. A wide range of frequencies will be
present in a general disturbance. At any given Reynolds number, only a band of
The Theory of Hydrodynamic Stability 223

Figure 18.8 Amplified (a) and damped (b) disturbances in a free convection boundary layer.
See text for detaiis; note different horizontal and vertical seales. Disturbance is introduced in
the same place in eases (a) and (b) but frequency is different. From Ref. [203).

these will be unstable according to Fig. 18.7, and within this band the amplification
will vary. If a sufficiently narrow frequency range is amplified significantly more
than other frequencies, a nearly sinusoidal disturbance may be observed to develop
out of the initially random disturbances.
In practice, these remarks are observed to apply when the initial disturbance
level is very low. Frequently when this level is not low enough, Tollmien-
Schlichting waves are not observed as the first stage of boundary layer transition.
The best verification of the theory discussed above has been obtained from
experiments in which a periodic disturbance is deliberately introduced [225] . The
usual method in boundary layer studies is to stretch a fine metallic ribbon ac ross
the flow close to the wall. This is vibrated by passing an alternating current through
it in the field 9f a magnet, just the other side of the wall. The resulting disturbance
of the flow velocity can then be traced downstream.
It has been found that the disturbance increases or decreases its amplitude for a
given frequency in good agreement with the prediction of the sign of ar in Fig. 18.7.
224 Physical Fluid Dynamics
We illustrate this with the different example of the free convection boundary
layer on a heated vertieal plate (Section 14.5). The temperature variations make it
easier to exhibit the waves visually. Figure 18.8 was obtained using an interfero-
metrie technique such that the light and dark bands are essentiaIly isotherms of the
twp-dimensional flow. (It should be noted that the optical arrangement exaggerates
the horizontal scale relative to the vertieal by a factor of 6.4.) In this case, the
periodic disturbance was introduced into the outer part of the boundary layer by a
mechanieal vibrator. The two parts of the figure contrast an amplifying and a
decaying disturbance, occurring in the positive and negative ar regions for this flow.
When Tollmien-Schlichting waves are amplified, this process is only the first
stage of an elaborate sequence of events. This is a case represented schematically by
curve B of Fig. 18.2. The description is continued in Section 19.1 as part of a
discussion of transition to turbulence.
19
Transition to Turbulence [21,164,219,240,249]

19.1 Boundary layer transition

Our knowledge of the development from the first instability of a boundary layer,
described in Section 18.5, to the fully turbulent motion that ultimately results
comes mainly from experimental investigations. Many of the experiments have used
the vibrating ribbon technique (Section 18.5). Although this was originally
introduced to investigate the relevance of the linear stability theory, it proved a
useful way of controlling the boundary layer development and ensuring that the
same features were to be observed repeatedly at the same place. Its use was thus
extended to experiments on the essentially non-linear developments further
downstream. The relationship of information so obtained to the processes in
naturally occurring transition will be considered later in this section.
The breakdown of a Tollmien-Schlichting wave into a more complicated
motion occurs when its amplitude reaches a critical value of around 0.01 to 0.02
(depending on the non-dimensional frequency of the wave) in the ratio of the root
me an square velocity at its maximum position to the free-stream velocity [137].
This critical amplltude is not always reached. The Reynolds number of a laminar
boundary layer increases with distance downstream, in consequence of the increas-
ing boundary layer thickness. The linear development of a wave of a given
frequency is thus indicated by a horizontalllne on the stability loop, as shown in
Fig. 19.1. A wave that is initiaBy amplified may subsequently be attenuated as this
line passes out of the loop. Provided that this happens before the critical amplitude
is reached, the whole history of the wave is weB described by the linear theory.
If, however, the critical amplitude is reached, the subsequent development of
the disturbance is entirely different. The linear theory no longer applies and,
regardless of whether the stability loop indicates amplification or attenuation of
the initial wave, a sequence of events ensues leading finally to a fully turbulent
boundary layer.
This sequence starts with the waves becoming three-dimensional; their
amplitude varies in the lateral direction paralleI to the wall. Simultaneously, they
interact with the mean flow, so that the velocity profile also has three- dimensional
variations. Figure 19.2 shows one set of measurements illustrating this development.
The growth of these three-dimensionalities continues until there are transient
local regions of very high shear. Figure 19.3 indicates the variation of the velocity
profile inferred from measurements with hot-wire anemometers (Section 23.2),
through one cycle of the wave at the lateral position where the wave amplltude is
largest. The next development is initiated in the regions of high shear, apparent in
the figure for t'between 0.38 and 0.6.
226 Physical Fluid Dynamics

Q"<"~-n-A""t~tenuation

Re

Figure 19.1 Locus on stability loop of disturbance of given frequency.

0.02 (a)

0.01

- 0.01

- 0.02 '--_........._ . . . L . - _........_....L.._--L.._---L_ ____'_----'L....__

(b)

Ulu.

(e)

(d)

0.08

0.06

0 .04

0.02
O~_~_...L.__ ........_ _ L_ __L.._ ___L_ ____'_~L....__ _

2.5 2.0 1.5 1.0 0.5 o - 0.5 - 1.0 - 1.5

z (in.)

Figure 19.2 Example of variations in z-direction (where x is flow direction, y [Link] from
wall) of (a) mean velocity in z-direction, (b) mean velocity in x-direction, (c) r.m.s. velocity
fluctuation in z-direction, and (d) r.m.s. velocity fluctuation in x-direction, in nominally two-
dimensional transitional boundary layer. Solid lines: y/6 = 0.31 ; dotted line: y/6 = 0.11. From
Ref. [138].
 Transition to Turbulence 227
V
I; r' '" 0 0 .3 0.38 0.45 0.6 0.8 1.0
1.5

1.0

0.5

0 0.5 1.0
U

UQ

Figure 19.3 Instantaneous veIocity profIles in transitionaI boundary layer; t' is time in
units of oscillation period. From Ref. [147].

Hitherto, although the structure of the motion has become increasingly complex,
the time scale of the velocity fluctuations has remained of the same order as the
period of the initial wave. Now irregular fluctuations with much shorter time scales
appear. A small part of the boundary layer has become turbulent.
The remainder of the transition process consists essentially of the growth of
local regions of turbulent motion produced in this way, whilst they travel down-
stream [97]. The turbulence spreads fairly quickly over the thickness of the
boundary laYJ:r. Each turbulent spot - as it is usually called - is then surrounded
laterally and 10ngitudinaHy by non-turbulent fluid. The motion within the spot is
similar to that in a fuHy turbulent boundary layer. EIsewhere, the velocity
fluctuations are still much slower. A spreading process then occurs, non-turbulent
fluid next to the spot being continuously brought into turbulent motion. The
turbulent spots have a characteristic shape - shown in Fig. 19.4 - the spreading
rates in different direetions being such that they retain this shape as they grow
[224].
As the spots grow they merge into one another. EventuaUy, all the non-turbulent
regions have been absorbed and the boundary layer is whoUy turbulent.
The transition takes the detaile d form indicated by Figs. 19.2 and 19.3 only
when the boundary layer is quite closely two-dimensional. SmaU three-dimensional
variations in the boundary layer thickness, preceding the growth of Tollmien-
Schlichting waves, have been observed to change the location and orientation of

•
Laminar

Figure 19.4 Characteristic shape (in pIane paralleI to the wall) of a turbulent spot in a
boundary layer.
228 Physical Fluid Dynamics

the high-shear regions from which the spots develop. They occur in planes perpen-
dicular to the wall (and paralleI to the main flow) instead of paralleI to the wall
[142] .
Spontaneously occurring boundary layer transition (not promoted by a
vibrating ribbon) is more difficult to investigate. However, it is probable that
essentially the same processes occur. Tollmien-Schlichting waves are not always
observable. Probably then, the critical amplitude for full transition is reached
before the frequency selection process implied by the stability loop has proceeded
far enough for a singIe frequency to be apparent. Later stages of the transition
process can then stiil be similar to those described above.
In natural transition, the spots wiil generally originate more randomly in time
than when they arise at a given phase of artificially generated waves.

Figure 19.5 Turbulerrt spot generated in boundary layer in water channel by disturbance at
x = 50 cm. Flow visualization by dye crystals sprinkled on wall upstream of picture. Note: the
spot is the disturbed region at right of picture; clear patch at left is region that has been swept
clear of dye by the passage of the spot. From Ref. [1001.

A large localized disturbance can directIy produce a turbulent spot. In

experiments to investigate the growth and structure of spots, they have been
generated by a rod momentarily poked into the boundary layer and by a small
spark [100, 224] . Figure 19.5 shows such an artificially generated spot.
The investigations on which the above account has been based were made in
zero-pressure-gradient boundary layers. Similar processes often occur in the
presence of a pressure gradient. The effect of an adverse gradient is generally
destabilizing - it decreases the Reynolds number at which corresponding develop-
ments occur - and that of a favourable gradient is generally stabilizingo These
changes arise from changes in the laminar velocity profile [249].
A different transition process that can occur when the pressure gradient is
adverse has already been considered in Section 11.4: the sequence of laminar
separation, free shear layer transition, and turbulent reattachment. Although the
initial attached laminar layer might be stable, the final reattached layer remains
turbulent [249].
Transition to Turbulence 229

19.2 Transition in jets and other free shear flows

Some of the features of boundary layer transition, described above, occur only in
shear flows adjacent to a solid boundary. We now look at the transition process in
free shear flows; that is ones away from any solid boundary. As our main example
we consider transition in jets (of the type introduced in Section 11.8 with the
ambient fluid the same as that in the jet). We shall see that the principal differences
arise in the later stages of transition.
The breakdown of a laminar jet usually starts quite close to the orifice, where
the velocity proflle still depends on the details of orifice geometry and the flow
upstream of it. Hence, different experiments may not be directly comparable, and
there may similarly be difficulties of comparison between experiment and theory.
However, the broad features of the transition process are always much the same,
and, if we omit quantitative detail, can be described without reference to any
particular jet.
Disturbances first appear as approximately sinusoidal fluctuations, indicating a
selective amplification process like that in a boundary layer. This stage can be
satisfactorily related to linear stability theory - a stability loop of type B in
Fig. 18.6 being relevant. Free shear flows become unstable at lower Reynolds
numbers than shear flows next to walls; the order of magnitude of the critical
Reynolds number for ajet is typically 10 (compared with 10 3 for a boundary
layer) [218,255].
The ability of an applied periodic disturbance in the appropriate frequency
range to promote the instability can have rather spectacular results in jets - as is
shown in Fig. 19.6. In such experiments, the disturbance is usually provided as a
sound wave from a nearby loudspeaker. The acoustic wavelength is long compared
with the wavelength of the instability, and the fluid thus experiences a periodic
disturbance of negligible phase variation.t
Pictures (a}-(c) of Fig. 19.6 show a smoke-carrying jet when acoustic
disturbances were minimized. Figure 19.6(a) and (b) have general illumination,
(a) with a relatively long exposure so that details of the instability are blurred and
(b) as a flash so that an instantaneous pattem is seen. The close-up (e) is again an
instantaneous picture, but with illumination through a siit showing one part of the
jet. A spontaneously arising periodic instability can be seen. Pictures (d}-(f) show
the same jet with the frequency from a loudspeaker selected to have maximum

tThe observation that jets respond to sound is an old one, first made with flames. The following
introduction to an eady paper on the subject, written in 1858 by J. Leconte 1156), makes
pleasant reading: 'A short time after reading Prof. John Tyndall's excellent artiele, I happened
to be one of a party of eight persons assembled after tea for the purpose of enjoying a private
musical entertainment. Three instruments were employed in the performance of several of the
grand trios of Beethoven, namely, the piano, violin and violoncello. Two "fish-tail" gas-burners
projected from the brick wall near the piano. Both of them burnt with remarkable steadiness,
the windows being c10sed and the air in the room being very calm. Nevertheless it was evident
that one of them was under a pressure neady sufficient to make it flare. Soon after the music
commeneed, I observed that the flame of the last mentioned burner exhibited pulsations in
height which were exactly synchronous with the audible beats. This phenomenon was very
striking to everyone in the room, and especially so when the strong notes of the violoncello
came in. It was exceedingly interesting to observe how perfectly even the trills of this instru-
ment were reflected on the sheet of flame. A deaf man might have seen the harmony.'
230 Physical Fluid Dynamics
effeet; (d) and (e) are the eounterparts of (a) and (b). Like (e), (f) has siit
illumination, but this is stroboscopic at the same frequeney as the sound. Sinee
the exposure extends over about 1800 fiashes of the stroboseope, the elarity of
the pieture demonstrates the extreme regularity of the pattem in its initial stages.
It is elear from Fig. 19.6 and from other similar work [50,107] that the
instability can give rise to well-developed vortices whilst it is stiIl in the periodie

Figure 19.6 Jet instability at Re = 1690. See text for procedures for different pictures. From
Ref. [59).
 Transition to Turbulence 231

stage. However, the motion does not remain periodie for any great distanee down-
stream. Breakdown to turbulent motion oeeurs. The Karman vortex street in a
wake (Seetions 3.3 and 11.7) appears to be the only ease in whieh periodie motion
persists to all distanees downstream (and the n only over a eertain Reynolds number
range). Presumably, this is a ease in whieh the non-linear terms play a role o.f the
form shown sehematieally as eurve A in Fig. 18.2 whereas most examples of shear
flow are represented by a eurve of form B.
It is in the breakdown from periodie or roughly periodie motion to turbulent
motion that the most signifieant differenee between free and wall shear flows
oeeurs. Free flows have no eounterpart of turbulent spots. The randomness
eharaeteristic of turbulenee develops at a roughly equal rate throughout the
transition region. Correspondingly, the time seale of the fluetuations beeomes
eontinuously shorter; there is no sudden loeal appearanee of mueh more rapid
fluetuations.
Figures 19.7 and 19.8 show oseillograms of the velocity fluetuations at various
distanees downstream, for, respeetively, spontaneously arising fluetuations and

. . .. .
lC·30 ..... Y-2161 X'7OMt.I,Y'4MIA

X'30MM,Y'2M" lC·IOOMM. Y'-4 MM

X'IOOMM, [Link]

Figure 19.7 Oscillograms of naturally arising velocity fluctuations in a free shear layer. x is
distance downstream from separation;y indicates position across shear layer. From Ref. [216).
 ~~
• TK.
X020'"
- '1'00 x•.,o.., '1'02II1II

Mfv\
X035'" '1'01 ... Xo"O 1liii '1'0-2 1liii

.---...,
.
X'50 '1'03 ... Xo.,O'" Y.-2 ....

----~~".~~
XoSO .., yo-I .., Xo130.... '1'0-4..,

X070.., yoe ... X0200.... '1'00

Figure 19.8 Counterpart of Fig. 19.7 with acoustic excitation. From Ref. [216].
Transition to Turbulence 233

acoustically stimulated ones. These were actually obtained in the free shear layer
produced by boundary layer separation at a sharp comer, but essentially similar
observations can be made in ajet [217] . The difference from Fig. 2.10, which is
characteristic of oscillograms produced during transition involving spot growth, is
apparent. The flow becomes turbulent through the continuous deve10pment of
irregularities throughout the flow, instead of the sudden local appearance of
irregularities which subsequently spread.

19.3 Pipe flow transition [166, 175]

We retum to the topic of pipe flow; the description of transition in Section 2.6
needs filling out. Although this is in principle one of the simplest configurations and
although it played such an important role in the early history of transition studies,
the details of the transition process tum out to be relatively complex and are, quite
possibly, not all known. Pipe flow illustrates, in a more extreme way than other
flows, the limitations to linear stability theory (Chapter 18), in providing only a
sufficient condition for flow breakdown, not a necessary one.
Linear stability theory applied to Poiseuille flow indicates that the parabolic
velocity profile is stable at all values of the Reynolds number; Re erit = oo. (There is
still a slight question mark, but it is only slight, against this result because there are
mathematical difficulties in covering every possible type of small disturbance [242].)
Nevertheless pipe flow becomes turbulent at some Reynolds number between
2 x 10 3 and 10 5 . Almost certainly, both the departures from the parabolic profile in
the entry length and the response to non-infmitesirnal disturbances play a role in the
interpretation of this fact.
Different geometries result in different detaile d flows in the entry length, but
the entering fluid will often be irrotational. Then, an annular boundary layer grows
on the wall, and the flow starts approximating to Poiseuille flow when the thickness
of this becomes elose to the radius of the pipe (Fig. 2.5). When the thickness of this
boundary layer is small compared with the pipe radius, it will resemble a Blasius
boundary layer, which is known to be unstable at high enough Reynolds number.
We may anticipate that there will be a stable length right at the start of the pipe
(the Reynolds number based on boundary layer thickness being small here),
followed by the unstable region, this being followed in tum by another stable
region as the profile a pproaches Poiseuille flow. If disturbances are amplified
sufficiently in the unstable region, then transition to turbulence may be expected.
Evidently, this description of the start of transition is consistent with a high
degree of variability in the maximum Reynolds number at which larninar flow can
be maintained. In the first place, small variations in the velocity profile, due to
different entry geometry, may affect the extent of the unstable region. In the
second place, the maximum amplitude of velocity fluctuations reached at the end
of the unstable region will be sensitive to the level of small disturbances present at
its start; the appearance of local regions of turbulence probably depends on a
certain amplitude being reached.t

tThe distiction between a ehange in entry geometry and a disturbance may be somewhat
blurred, particularly if wall roughness enters the story. However, here the former is intended to
imply a ehange in the detailed pattem of the laminar flow that would oceur in the absence of
any instability. The latter is intended to imply a small uneontrolled unsteadiness in the flow.
234 Physical Fluid Dynamics
By taking great eare to minimize the disturbanee level, experimenters have
maintained laminar Dow up to values of Re = uavd/v of around 1.0 x 10 5
Ref. [198]. It is likely that in these experiments there was aloeal unstable region
in the entry boundary layer, but with subsequent deeay of the amplified
disturbances [233, 254] .
When turbulenee develops at Reynolds numbers above about 10 4 the region of
boundary layer instability is probably an important feature of the transition
process. Turbulenee has been observed [194] to appear first of all in small spots
restricted laterally as weIl as axially and presumably forming in the boundary layer.
These spots then spread in all directions and several of them merge to form a
turbulent plug.
However, the lowest values of the Reynolds number (around 2 x 10 3 ) at whieh
transition oecurs are weIl below the minimum value (araund 104 ) at which theory
[233,254] indieates the existence of an unstable region. In many experiments
[64,167, 194,211], the entry geometry has involved sharp edges or comers and
there was probably a region of separated flow. This would be unstable at a lower
Reynolds number than any attaehed boundary layer, and rapid amplifieation of
disturbances would oeeur as shown in Fig. 19.9.

Figure 19.9 Flow at sharp entry to pipe; Re = 5240. From Ref. [152).

However, this is probably not the only way in whieh transitian can be initiated
at the lower Reynolds numbers. Observations [167,168] have been made, same of
them with an arrangement unlikely to produee separation, of transition oceurring
at around 3000 in a manner simiIar to that described ab ave for higher Reynolds
numbers - turbulent spots originating close to the wall and then spreading to give
turbulent plugs. One ean only say that growing turbulent plugs ean originate and so
transition occur at Reynolds numbers down to about 2000, provided that there are
large enough disturbances. Probably there is no unique process for the origin of
plugs in this range; a large disturbance ean take such a variety of forms.
The triggering disturbance can be another turbulent region. Observations have
been made of one turbulent spot appearing in the boundary layer a little way
downstream of an older one, whose presence has caused sufficient disturbance
[167,168] .
Despite these complications in the detailed mechanics of transition, it is clear
what governs the lower limit in Reynolds number for transition. Onee turbulent
plugs are produced, the transition process is substantially the same at all values of
the Reynolds number. The turbulence propagates into the neighbouring non-
turbulent regions at eaeh end of a plug. The plugs thus grow in length at the
Transition to Turbulence 235

expense of the non-turbulent region. When the front interface of one plug meets
the rear interface of another, the two merge to give a single plug. Thus the
intermittency factor increases with distance downstream, until all the non-turbulent
regions have been absorbed and fully turbulent pipe flow results. Figure 19.10
shows the rates at which front and rear interfaces travel downstream as functions of
the Reynolds number. At high Re, the interface speeds are substantially constant
at about 1.5 and 0.5 times the mean flow speed. The difference between them is
le ss at lower Re and disappears when Re is about 2300. Below this plugs do not
grow and fully turbulent flow is not produced at any distance downstream. There
is some variability in the figure quoted for the rninimum Re at which transition can
be produced by large disturbances, valu es going down to 1800. Nevertheless, it is
clear that the lirniting factor is plug growth. Observations have been made of the
processes leading to plug production at much lower Reynolds numbers, but
transition does not then result [64 J.
The region of intermittently turbulent flow is often very long. This depends on
the rate of plug production, which is sensitive to the disturbance level. Quantitative
estimates thus relate only to particular experiments, but this region can be
hundreds of diameters long.
We saw in Section 2.6 that the production of turbulent plugs can be either
random or periodic in time, as illustrated by the oscillograms of Fig. 2.1 O. The
random production might be regarded as the normal state of affairs, periodic
production occurring through a rather special mechanism. Nonetheless, the latter
is a phenomenon of some interest and significance and we now consider how it
comes about.
The pressure gradient needed to produce a given flow rate is much larger when
the flow is turbulent than when it is laminar (Fig. 2.11). Conversely, the flow rate

1.6

UF 1.4

u av
1.2
UR
u av
1.0

0.8

0.6

0.4
2 3 4 5 6 7 8 9
Re/10 3

Figure 19.10 Ratios of front and rear velocities of plugs to the mean Dow speed as functions
of Reynolds number (based on experimental results in Refs. [169,194, 211)).
236 Physical Fluid Dynamics

-
Intake Outlet

-
Figure 19.11 Sequence of positions of larninar and turbulent rnotion (shaded regions
turbulent) during periodic plug formation. Length and diarneter are not to the sarne seale (and
interfaces are therefore shown plane).

produced by a given pressure difference between the ends is much higher for
laminar flow than for turbulent flow. In most experimental arrangements, it is the
pressure difference, not the flow rate, that is held constant. Hence, in transitional
flow, as a turbulent plug grows the flow rate reduces. This inhibits the instability of
the entry region, and the probability of another plug being produced becomes
negligible. Only when the front of the plug passes out of the far end of the pipe
does the fraction of the pipe length in laminar motion increaseo The flow rate then
starts to increase too. When it has increased sufficiently, another plug is produced
near the entry, and the cyc1e recommences. Since the front of a plug travels down-
stream at about three times the speed of the rear (Fig. 19.1 0), there can be
substantial variations in the fraction of the flow that is turbulent (Fig. 19.11) and
thus in the flow rate. If these variations pass through the criterion for instability
(such as the critical Reynolds number of the entry boundary layer), the instability
can be renewed at just the sam e phase of each cyc1e and the flow will pulsate with a
well-defined period.
This happens only over a limited Reynolds number range. At smaller Reynolds
numbers, the difference between laminar and turbulent flow rates is too small. At
higher, the flow becomes turbulent in only a small fraction of the totallength and
so the variations in flow rate are again too small. As one would expect, the
Reynolds number range depends on the lengthjdiameter ratio of the pipe (and
periodic transition does not occur at all when this ratio is less than about 60).
One can, therefore, quote only examples of this range, and two are given below
[194]. The Reynolds number ranges appear short, but this is deceptive; there is a
large change in the fraction of the fluid in the pipe that is turbulent as one goes
through the range, and so a small change in Reynolds number is associated with a
large change in the non-dimensional pressure drop. For I/d = 540, the range has
been observed to start t at a Reynolds number (based on a velocity averaged not
only over a cross-section but al so over a cyc1e of the pulsation) of about 4700; the
intermittency factor at the outlet became unity:j: at Re = 5900. Corresponding
values of the non-dimensional pressure gradient (p 1 - P2 )d3 / pv21were 2.1 x 105
and 5.0 x 10 5 • For I/d = 78, the Reynolds number range (with the same significance)
was 5500 to 5700 and the range of (Pl - P2)d 3 /pv 21was 4.5 x 105 to 5.9 X 10 5 •

t In these experirnents, the flow was larninar for Re below this.

:j:It is not certain whether this or a higher Re is the upper end of the pulsating regirne. It is in
principle possible for the pulsations to occur with fully turbulent flow at the outlet, so long as
there is stilI a substantial fraction of the pipe length with interrnittent turbulence.
Transition to Turbulence 237

In conclusion, it is now abundantly clear why the traditional view that there is a
single critical Reynolds number, below which the flow is laminar, above which the
flow is turbulent, cannot describe the facts. Instead we have to introduce aset of
Reynolds numbers relating to different events in the transition process. We start at
the high end:
(i) The theoreticallinear stability llmit ofPoiseuilIe flow is Re = oo. We may
guess that a pipe flow with an artificially induced parabolic profIle at the entry
could remain larninar to even higher Reynolds numbers than flows with a normal
entry length.
(ii) There may be a Reynolds number above which a pipe flow with a normal
entry length will always become turbulent through the amplification of small
disturbanees. If so, this Reynolds number is at least 105 •
(iii) Below this there is stilI a region of instability in the entry length but not a
sufficiently large one to promote transition when the disturbances are small. The
lowest Reynolds number at which this exists is somewhat uncertain but around
104 •
(iv) When the flow is spontaneously pulsating, the Reynolds number periodically
becomes larger than its average value making the entry length temporarily unstable.
The lowest average Reynolds number at which this provides an instability
mechanism is again uncertain but is around S x 10 3 •
(v) The lower limit to transition even for very large disturbanees is provided by
the growth or otherwise of plugs. The critical Reynolds number for this is better
defined at (2.1 ± 0.3) x 10 3 •
20
Turbulence

20.1 The nature of turbulent motion

Frequent references to turbulent motion have been made in previous chapters, but
detaile d discussion of the nature of that motion has always been postponed. We
now take this matter up [86] .
No short but complete definition of turbulence seems to be possible. Perhaps
the best brief summary of it is 'a state of continuous instability'. A more complete
specification becomes somewhat lengthy, but, as we shall see, no detinition can
eHminate the borderline examples that one may or may not choose to call turbulent.
A convenient starting point for a discussion of the nature of turbulent motion is
the following fact: each time a flow changes as the resuIt of an instability, one's
ability to predict the details of the motion is reduced. We examine the meaning of
this statement tirst of all through examples in which the consequence of the insta-
bility is not (immediately) turbulent motion.
Consider the motion in a Karman vortex street (Section 3.3) in the wake of an
obstac1e. The velocity at a point fixed relative to the obstac1e varies periodically
and roughly sinusoidally. The phase of this variation is arbitrary, depending on the
small disturbances at the time the flow commeneed. If, therefore, one asks for a
prediction of the instantaneous velocity, without making any observations, this
cannot be given within certain limits. This lack of predictability arises from the
instability producing the vortex street; in the steady flow, from which the vortex
street develops, such a prediction could be made. The degree of unpredictability is,
however, small. One requires only a single observation indicating the phase of the
fluctuations for all the details of the flow to be determined. When, with increasing
Reynolds number, a further instability causes loss of regularity in the array of
vortices, the unpredictability is increased. One can, for example, no longer say that,
if one has made an observation of the velocity, then the velocity will be the same
one period later. However, markedly systematic features may stiIl exist - regions of
high vorticity passing a point in a sequence, although not a completely periodic
one. The flow can stiIl be described with more emphasis on these systematic
features rather than random ones. Each subsequent instability reduces one's ability
to do this, until the random features are dominant.
Another example is provided by cellular convection (Sections 4.3, 17.3 and
17.4). In an arrangement such as Fig. 4.1, every point in a horizontal plane is a
priori similar. There is nothing to distinguish the points at which the fluid rises from
those at which it falls; one cannot predict whether it wiIl be rising or falling at a
particular place. The positions of the rising and falling currents are different on
different oecasions when one uses the same piece of apparatus, depending on
Turbulence 239

the small disturbanees present at the cümmencement .of each experirnent.

Nevertheless the flüw pattem can have a high degree .of regularity (as is
particulady exemplified by surface tensiün driven cünvectiün - Fig. 17.6), the
relative lücatiün .of rising and falling currents being repeatable from experirnent tü
experiment. Further iustability, in a sequence .of events such as that deseribed in
Sectiün 4.3, reduces the repeatable features .of the flüw and sü reduees the predict-
ability .of its details. At the end .of the sequence randüm features becüme düminant.
A flüw may be called turbulent when the level .of predictability is sü reduced
that it is appropriate tü give a statistical descriptiün .of it.
Oeady, that statement dües nüt prüvide an unambiguüus classificatiün .of flüws
intü turbulent and nün-turbulent. But .one wüuld nüt expect tü be able tü dü this.
Figure 19.8, für example, shüws velücity fluctuatiüns develüping cüntinuüusly
from a regular form tü a turbulent .one. At .one extreme the mütiün is unambig-
uüusly nün-turbulent; at the üther it is unambiguüusly turbulent; but there is nü
stage in between that .one can identify as the start .of turbulence. (Sümetimes there
is a sufficiently sudden develüpment in the transitiün sequence that it is meaningful
tü say that the turbulence first üccurs (lücally) there. An example is the appearance
.of turbulent spüts in büundary layer transitiün (Sectiün 19.1). The difference arises
through .one st age .of the sequence üccurring .on a smalllength scale; the time scale
is cürrespündingly small and the develüpments leading tü lücal randümness are
rapid cümpared with the üther stages .of transitiün.)

20.2 Introduction to the statistical description of turbulent motion

A statistical descriptiün is fürmulated in terms .of average quantities, since these

are repeatable früm experiment tü experiment. The detailed nature .of the averaging
prücess will be cünsidered belüw, but anüther example may indicate the general
meaning .of a statistieal descriptiün. Cünsider turbulent pipe flüw. A measurement
.of the instantaneüus velücity at süme püint in the pipe is, .on its üwn, .of little use.
It dües nüt indieate what the velücity will be at the same püint at anüther instant
ür what it will be at any üther geümetrically similar püint (Le. a püint at the
same distance from the axis). Nevertheless, .one may expect a greater similarity
between twü geümetrically similar püints than between twü geümetrically dissimilar
püints. This similarity is provided by the average quantities. The mean velücity, für
example, is the same at all püints at the same distance from the axis, but varies
with this distance.
The ideal aim .of a theory .of turbulent mütiün wüuld be the develüpment .of a
statistical mechanics analügüus tü those develüped in the kinetie theüry .of gases.
Such a theüry wüuld have tü be based .on the equatiüns .of fluid mütiün instead .of
thüse .of the dynamies .of mülecules interacting .only thrüugh elastie [Link]üns. It is
thus a very difficult task. Würk .on these lines (beyünd the scope .of this b.oük) has
been attempted, but variüus, nüt necessarily valid, assumptiüns have tü be made
befüre prügress is püssible.
Cünsequently, much .of .our [Link] .of turbulent flüws [Link] from experi-
ment. Süme .of the quantities that enter intü any statistical theory can be measured
experimentally. A physical description .of the principal [Link] .occurring within
a turbulent [Link] can be develüped from these. Generally, a rather cümplex inter-
[Link] between theüry and experiment results, theüry indicating which quantities
 (b)

Figure 20.1 Weather maps for (a) 24 Jan 1952 (b) 26 Jan 1952. Lines are contours oftheheight (at intervalsof 30.5 m) at which the pressure
is 500 mbar (approximately half ground level pressure); contours of pressure at a fixed height of about 5.5 km would look very simUaro From
Ref. [234) .
242 Physical Fluid Dynamics

might usefully be measured and the experimental results pointing the way for
further theoretieal developments.
The use of a statistical description does not imply that any counterpart of the
uncertainty principle in quantum mechanics is involved. The governing equations
are essentially deterministic; in principle, every detail of a turbulent motion should
be determinable if the initial conditions are completely defined. The analogy with
kinetic theory is thus with dassical Maxwell-Boltzmann statistics, not with
quantum statistics.
The reason that a deterministic treatmen t of turbulent flows is not in practice
possible is that important features of the motion (for example the large eddies to
be described in Sections 22.4 and 22.6) develop from entirely insignificant pertur-
bations. An example of this may be taken from a context in which it is rather
familiar - atmospheric motions on the scale that governs the prineipal features of
the weather. (This type of motion is in the bordedand that one may or may not
choose to call turbulent, but it shows sufficient of the characteristics of turbulence
for the present purpose.) One of the difficulties of long-term weather forecasting is
that very fine details at one time can govern major pattems at a later time. Figure
20.1 shows two weather maps, separated by two days. In the first, there are two
sirnilar looking distortions of the pattern associated with the large cyclone over
Canada, around SSoN, 62°W and around 43°N, 7SoW. In the period between the
two maps, the former distortion just faded out whereas the latter amplified and
rnigrated to become the significant cyclonic region at the south of Greeniand in the
second map. Subsequently, it merged with the other original cyclone to the north
of Norway with the elimination of the high pressure ridge over the Atlantic.
It is sometimes said that one should not try to understand why flows are
turbulent but rather why they are ever larninar; science seeks to interpret order in
nature, not disorder. This remark certainly adds to our understanding of the nature
of turbulence. However, in the absence of a full statistical theory, it is difficult to
translate the remark from the rnetaphysical to the physical.

20.3 Formulatioo of the statistical descriptioo

The statistical description of a turbulent flow starts by dividing the velocity and
pressure field into mean and fluctuating parts. We may consider the procedure for
one component of the velocity; the other components and the pressure are freated
in just the same way. At each point, the velocity component is written as U + u,
where, by defmition ü = 0 (an overbar denoting averaging). For theoretieal
purposes it is sometimes convenient to think of the average as an ensemble average;
i.e. one considers a large number of identical systems and takes the average of
the velocity at corresponding instants over all these systems. In practice, the
average is usually a time average; one observes and averages the velocity at a point
over a period long enough for separate measurements to give effectively the same
result. Procedural difficulties can arise when the imposed conditions are unsteady,
but we need not consider such situations here.
Thus throughout the following the average of any quantity T signifies
- = -1
T JS Tdt (20.1)
2s -s
Turbulence 243

where s is Iarge compared with any of the time scales involved in the variations of
T.
U indicates the mean motion of the fluid. Information about the structure of
the velocity fluctuations is given by other average quantities, the first being the
mean square fluctuation, u 2 • (u 2 )1/2 is known as the intensity of the turbulence
component, and
(20.2)
as the intensity of the turbulence. It is directly related to the kinetic energy per
unit volume associated with the velocity fluctuations,
1 -
L = - pq2 (20.3)
2
The same intensity field can in principle be produced by many different pattems
of velocity fluctuation. Before we look at the average quantities most often used to
examine the more detailed structure of the velocity field, we consider briefly an
alternative representation. This is in a sense the most fundamental statistical
representation, although it is not the most convenient for the development of
models of turbulent structure based on experimental observation. The probability
distribution function p(u) of a velocity component at one point is defined so that
the probability that the fluctuation velocity is between u and u + du is p(u)du. One

f:
thus has

oo p(u)du = 1 (20.4)

The intensity is related to this,

u2 =foo u 2 p(u)du (20.5)

-oo

but the probabHity distribution function contains more information than the
intensity. Relationships between velocity fluctuations at different points (or times)
are indicated by joint probabHity distribution functions. For example a second-
order function,p(ul, U2), may be defined so that the probability that the velocity
at one point Hes between Ul and Ul + dUl and that at the other point simultaneously
Hes between U2 and U2 + dU2 is P(UI, U2)dUl dU2' In principle, for a complete
representation of the turbulence, this process has to be continued to all orders.
Probability distribution functions are sometimes determined experimentally, but
much more frequently further average quantities are measured. Fuller information
than is given by u~bout the fluctuations at a single point can be obtained by
measurements of u 3 , u 4 , etc.
Information about velocity fluctuations at different points (or times) is given by
correlation measurements. The correlation between two velocity fluctuations Ul
and U2 is defined as Ul U2 and the correlation coefficient as

jR = UlU2 / ( uiz4 )~ (20.6)

Ul and U2 are for the moment quite general quantities; but as examples, they
could be simultaneous values of the same component of the velocity at two
244 Physical F/uid Dynamics

different points, or two different components of the velocity at a single point. If

the fluctuations Ul and Uz are quite independent of one another, then their
correlation is zero. However, any turbulent flow is governed by the usual equations
and these do not allow such complete independence, particularly for fluctuations
at points elose to one another. Hence, non-zero correlations are observed.
The concept of correlations, like that of probability distributions, can be
extended to higher orders, by defining quantities such as Ul Uz U3. A complete
specification of the turbulence again requires one to consider all orders up to
infinity. In practice, detailed attention is usually confmed to double cor-
relations (Ul uz) with briefer investigation of triple correlations.
Experimental studies of turbulent flows often involve the interpretation of
correlation measurements. One of the reasons for working particularly with cor-
relations is that those of loworder lend themselves satisfactorily to physical
interpretation, in ways to be discussed in Section 20.6. We shall also be introducing
later (Section 20.7) the spectrum functions which are the Fourier transforms of
correlation funetions. However, we now have enough material to examine the way
in which the equations of motion are developed for turbulent flows.

20.4 Turbulence equations

In the interests of coneiseness and convention, it is necessary to use here the suffix
notation for vector equations which has been avoided elsewhere in this book
(except in the appendix to Chapter 5 and in equation (14.75)). For a full ex-
planation of this notation see Refs. [23,38,133] . Its basic features are as follows:
each suffix can take values 1, 2 or 3, corresponding to the three co-ordinate
directions; a vector equation can be read as any one of its component equations by
substituting the appropriate value for the suffix common to every term; and the
repetition of a suffix within a single term indicates that that term is summed
over the three values of that suffix.
With the velocity divided into its mean and fluctuating parts, the continuity
equation (5.10) is
div(U + u) = 0 (20.7)
that is
(20.8)
Averaging this equation (the processes of averaging and differentiation are inter-
changeable in order),
auj/aXj = 0 (20.9)
Subtracting this result from the originaI equation,
au;/aXj = 0 (20.10)
The mean and fluctuating parts of the velocity field thus individually satisfy the
usual form of the continuityequation.
The same division applied to the Navier-Stokes equation (equation (5.22) with
Turbulence 245

F = 0) gives
a( Ui + Ui) + (U,. + U·) a( U
-o--=--~
j + Ui) -_ - -1 a(p + p) + v _a2-..:(,--u'..!...i:;-+_U!-<-i) (20.11)
õt I I aXj p aXi axJ
Carrying out the averaging process throughout this equation
aUi aUi ---au;
1 ap a2Ui
-+U,.- +U·-=-- - + v - - (20.12)
at I aXj I aXj p aXi axJ
which, with the aid of the continuity equation (20.10), may be rewritten
aUj
Uj -
1 ap a2Ui
= ---+ V - - 2 - -
a(UjUj)
- (20.13)
aXj p aXi aXj ax,
where, additionally, attention has been restricted to steady mean conditions by
making the fust term of (20.12) zero.
Equation (20.13) for the mean velocity Uj differs from the laminar flow
equation by the addition of the last term. This term represents the action of the
velocity fluctuations on the mean flow arising from the non-linearity of the Navier-
Stokes equation. It is frequently large compared with the viscous term, with the
resuIt that the mean velocity distribution is very different from the corresponding
laminar flow.
The character of this interaction between the mean flow and the fluctuations
can be seen most simply in the context of a flow for which the two-dimensional
boundary layer approximation applies. The turbulent fluctuations are always three-
dimensional, but if the imposed conditions are two-dimensional, the re is no
variation of mean quantities in the third direction and terms such as a(uw)jaz
(that would otherwise appear in the next equation) are zero. Omitting such terms
and terms that are small on the boundary layer approximationt in equation (20.13)
gives the turbulent flow counterpart of equation (11.8); that is
au au 1 ap au 2 a_
u aX + v ay =-; aX + v ay2 - ay (uu) (20.14)

This equation is applied to turbulent boundary layers, jets, wakes, etc.

Writing the last two terms of (20.14) as
1 a (au
; ay J1 ay - puu
_) (20.15)

shows that the velocity fluctuations produce a stress on the me an flow. A gradient
of this produces a net acceleration of the fluid in the same way as a gradient of the
viscous stress. The quantity (-puu), and more generaUy the quantity (-PUiUj), is
called a Reynolds stress.
The Reynolds stress arises from the correlation of two components of the
velocity fluctuation at the same point. A non-zero value of this correlation implies
that the two components are not independent of one another. For example, if

tThe boundary layer approximation is used here to its fullest extent. In some studies of
turbulent flows, some further terms (e.g. o(u')/ox) are retained because measurements indieate
that they are not so very small.
246 Physical Fluid Dynamics

(a) (b) (e) (d) (e)

Figure 20.2 Geometrical interpretation of Reynolds stress: if pa tterns of velocity fluctuations

shown in (a) and ili occur mo re frequently than those in (e) and (d), giving negative uv, then
V'2 is Jarger than U'2 as indicated by (e).

uv is negative, then at moments at which u is positive, v is more likely to be

negative than positive - and conversely when u is negative. Transferring attention
to co-ordinates at 45° to the x and y direetions shows that this corresponds to
anisotropy of the turbulence-different intensities in different direetions. Putting
u' = (u + v)/y2 v' = (v - u)/y2 (20.16)
gives

uv =.! (U'2 - V'2) (20.17)

2
Figure 20.2 shows the geometrical significance of this.
One can readily see how a correlation of this kind can arise in a mean shear flow.
We may consider the case of positive au/ay as shown in Fig. 20.3. A fluid partide
with positive v is being carried by the turbulence in the positive y-direction. It is
coming from a region where the mean velocity is smaller and it is thus likely to be
moving downstream more slowly than its newenvironment; i.e. it is more likely
to have negative u than positive. Similarly negative v is more likely to be associated
with positive u. The process is in general - but not in detail - analogous to the
Brownian motion of molecules giving rise to fluid viscosity.
Further understanding of the interaction between the mean flow and the
fluctuations is obtained from the equation for the kinetic energy of the turbulence.
Subtracting equation (20.12) from equation (20.11) gives
au; au;
-+ u·-+u·-+u·--u·-=---+v--
au; au; au; 1 ap a2 u; (2018)
at I aXj J aXj J aXj J aXj p ax; axj .

Figure 20.3 To illustrate the generation of a ReynoJds stress in a mean veJocity gradient.
Turbulence 247

Multiplying this by Uj and averaging

I a(uf) 1 a(u~) au'· 1 a - 1 a a 2Uj

---'- + -1Ij --'- = -UjUj - ' - - - (u;Uj) - - - (jiii;) + VUj --2
2 at 2 aXj aXj 2 aXj p aXj aXj
(20.19)
(where the rearrangement of terms has made use of the continuityequation
(20.10)). Since the summation convention is being applied, the mathematics
invo1ves multiplying each component of the dynarnical equation (20.18) by the
corresponding velocity component and then adding the three resulting equations.
For steady mean conditions the first term of equation (20.19) is zero, but it
indicates the physical significance of the equation; in view of the summation
convention

U; =q2 =2"'i:./p (20.20)

and so each te rm in the equation represents some process tending to increase or

de ere ase the kinetic energy of the turbulence
With the boundary layer approximation applied to a flow which is steady and
two-dimensional in the mean, equation (20.19) becomes

1 a(q2) 1 a(q2) _ au a (1 -yo 1 _ ) a 2 Uj

- u - - + - V - - = - u v - - - -q v+-pv +VUj-- (20.21)
2 aX 2 ay ay ay 2 p axJ
The left-hand side and the second term on the right-hand side are terms that become
zero when integrated over the whole flow. They represent the transfer of energy
from place to place, respectively transfer by the mean motion and transfer by the
turbulence itself. As in a larninar flow (Section 11.9), the viscous term can be
divided into two parts: one is essentially negative and thus represents viscous
dissipation; the other (usually small) integrates to zero and so is another energy
transfer process. The input of energy to compensate for the dissipation must be
provided by the only remaining term, ( -uvilU/ay). We have seen that uv is likely to
be negative where au/ay is positive, giving this term the required sign. Although
local regions of positive (uva u/ay) can occur, they cannot occupy the majority of
the flow or the turbulence cannot be maintained.t
The equation for the energy of the mean flow contains a corresponding term of
0pposite sign. The term thus represents a transfer of energy from the mean flow to
the turbulence. One can therefore say that the Reynolds stress works against the
mean velocity gradient to remove energy from the mean flow, just as the viscous
stress works against the velocity gradient. However, the energy removed by the
latter process is directly dissipated, reappearing as heat, whereas the action of the
Reynolds stress provides energy for the turbulence. This energy is ultimately
dissipated by the action of viscosity on the turbulent fluctuations. Frequently, the
loss of mean flow energy to turbulence is large compared with the direet viscous
dissipation.

tThis statement need not be true for systems described by a dynamical equation with additional
terms to those in (20.11); for example a buoyaney foree. These can give further terms in the
energy equation whieh may represent alternative turbulenee generating meehanisms.
248 Physical Fluid Dynamics

20.5 Calculation methods

In many praetieal situations, only the mean flow development is of real interest.
The strueture of the turbulenee is relevant only in so far as it determines the
Reynolds stress. Attempts have thus been made to relate the Reynolds stress to
the mean flow in ways that all ow the mean flow development to be ealeulated
(for example, for boundary layers in various pressure distributions) without detaile d
study of the turbulenee.
A simple example of such a proeedure is the concept of eddy viseosity. This
supposes that the analogy between the turbulent fluetuations and moleeular
Brownian motion ean be made quantitative in the form
(20.22)
VT may be taken as a eonstant or as a funetion of position - guessed from some
model of the turbulenee or empirieally determined. The proeedure is open to the
objeetion that sinee (as we shall be seeing in Chapter 22) the turbulenee normally
involves large-seale eoherent motions, the Reynolds stress at any point depends on
the whole veloeity profile, not just the local gradient.
More refined ealeulation methods have been developed [69,70,140] (although
we shall not be considering them here). The previously large gap between basie
studies of turbulent motion and the needs of applied seientists has diminished greatly
in reeent years as aresult of these methods. However, it remains true that mueh of
the deseription of turbulenee of the type given in this book has not yet found applie-
ation and that empirieal methods with little support from fundamental studies are
stiIl in extensive use.

20.6 lnterpretation of correlations

Correlation eoeffieients (Seetion 20.3) play an important role in both theoretieal

and experimental studies of turbulenee. To illustrate how they ean indieate the
seale and strueture of a turbulent motion, we now look at typieal properties of
double correlations. Some of the ideas introdueed rather vaguely here wiIl be used
more specifieally in Seetions 21.2, 22.4 and 22.6.
When Ul and U2 are veloeities at different positions but the same instant, UIU2
is known as a space eorrelation. Its partieulars may be speeified by a diagram such
as Fig. 20.4(a). Most attention is usually given to eorrelations of the same

(a) (b)

Figure 20.4 Schematic representation of double velocity correlations.

TurbuJence 249

Figure 20.S Typical correlation curves.

component of velocity at points separated in a direction either paralleI to that

velocity component (Fig. 20.4(b)) or perpendicular to it (Fig. 20.4(c)). We may
call these respectively longitudinal and lateral correlations.
The correlation will depend on both the magnitude and direction of the
separation, f. Different behaviours in different direetions may provide information
about the stmcture of the turbulence, a point that will be taken up again in Section
22.4. Here we pay more attention to the variation with distanee, r = I f I. When
r =0, Ul =U2 (provided they are in the same direction) and the correlation co-
efficient R of equation (20.6) is, by definition, equal to 1. At large r the velocity
fluctuations become independent of one another and R asymptotes to o. In
consequence, the dependence of R on r takes typically one of the forms shown in
Fig. 20.5. (R has a maximum value of 1 at r =0 and so CdR/ar)r=o = O. However,
the curvature at r = 0 is usually large and experimentally measured correlations
often appear to have finite slope at r = 0.)
A negative region in the correlation curve implies that Ul and U2 tend to be in
opposite direetions more than in the same direction. For a longitudinaI correlation
this would imply dominant converging and/or diverging flow pattems. There is
often no reason to expect such pattems and one may expect that longitudinal
correlations will usually give a curve such as A (Fig. 20.5). On the other hand,
lateral correlations may be expected to have a negative region, like curve E, since
continuity requires the instantaneous transport of fluid across any plane (by the
fluctuations) to be zero. Such expectations are not always fulfllled; but, when they
fail, this may itself be informative about the stmcture of the turbulence.
A correlation curve indicates the distanee over which the motion at one point
significantly affects that at another. It may be used to assign a length scale to the
turbulence; a length can be defined for example as JöRdr, Of as the distance in
which R falls to I/e, or, if the curve has a negative region, the value of r at which R
is a minimum. We shall see in Sections 20.8,21.3 and 22.4 that this concept is
extended to associate a variety of length scales with the turbulence.
The correlation of the same velocity component at a single point at different
instants is known as an autocorrelation. Such a correlation depends, of course, on
the time separation s in a similar way to the dependence of a space correlation on r.
It can be used to de fine a typical time scale of the turbulence.
250 Physical F/uid Dynamics

When the turbulent motion is occurring in a flow with a large mean velocity, it
is possible for the turbulence to be advected past the point of observation more
rapidly than the pattem of fluctuations is changing. An autocorrelation will then
be directly related to the corresponding space correlation with separation in the
mean flow direction, the same curve applying for both when one puts s = rjU. This
transformation is cal1ed Taylor's hypothesis. The extent to which it applies varies
greatly between different flow situations.
A correlation between velocities measured at both different positions and
different instants is eal1ed a space-time correlation. Such measurements are very
useful in indicating the trajectories of eertain features ('eddies' - Section 20.8)
associated with the turbulenee, and have played an important role in some of the
descriptions of the motion to be mentioned in Sections 22.4 and 22.6. However, we
shall not discuss them further in this book.
In principle, of course, correlations involving the pressure fluctuations as well as
the velocity fluctuations may be formulated. Equation (20.21) has already involved
pv - a pressure-velocity correlation. However, such quantities are difficult to
measure and so have received less attention.

20.7 Spectra

Another method of discovering the time scales associated with a turbulent motion
is, evidently, Fourier analysis of the velocity fluctuations into component
frequencies. The reader is referred to other sources for the full theory [65] ; here
just sufficient theory will be given to show the result of passing a turbulence signal
through a frequency filter before the usual squaring and averaging. Suppose that
the velocity signal is u(t). Then the output from the filter is

X(t) = Joo u(t - t')'Ir(t')dt' (20.23)

o
where 'Ir(t) is the response function of the filter; i.e. 'Ir(t) is the output at time t
when the input is a delta function at t = O. X(t) is a fluctuating funetion and one
can measure its mean square, which is

x2 = Joo Joo u(t - t')u(t - t")'Ir(t')'Ir(t")dt'dt" (20.24)

o 0

where the average is over t, by the procedure of equation (20.1), and so may be
taken inside the integration with respeet to t' and t". But
u(t - t')u(t - t") = u 2 R(t' - t") (20.25)
where R(s) is the autoeorrelation coefficient for time delay s. We now introduce the
the Fourier transform cp(w ) of the autocorrelation such that

u 2 R(s) = Joo cp(w)eiwsdw (20.26)

o
(R(s) being an even funetion). Substituting in (20.24)

X2 = Joo Joo Joo cp(w)'Ir(t')'Ir(t")eiw(t'-t")dwdt'dt" (20.27)

000
Turbulence 251

But

A(w) = f'" eiwt'w(t')dt'

o
(20.28)

is the amplitude of the output when the input is sinusoida1 with angu1ar frequency
w. Thus

x2 = Joo <f>(w)A(w)A*(w)dw (20.29)

o
Ifthe filter is a good one, A(w)A*(w) is mueh larger over a narrow range of
frequencies, centred on wo, than elsewhere and equation (20.29) reduees to
X2 = C<f>(wo) (20.30)
where C is a ealibration eonstant.
Thus this proeedure measures the Fourier transform of the autoeorrelation
funetion. Putting s = 0 in equation (20.26)

u2 = Joo <f>(w)dw (20.31)

o
showing that <p(w) may be interpreted as the eontribution from frequeney w to the
energy of the turbulenee. (We have here considered one veloeity eomponent, but
equation (20.2) provides immediate extension from the energy associated with
eomponents to the energy as a whole.) <f>(w) is thus known as the energy speetrum.
Since R(s) and <f>(w) are Fourier transforms of one another (both real), either is
in principle ealculable from the other. In practice, however, mueh greater aeeuraey
is obtained by adireet measurement of the quantity required than by a ealeulation
from measurements of the other quantity. For example, the whole trend of <f>(w) at
large wis determined by the detailed shape of the R(s) eurve elose to s = 0, whieh,
as we have seen, is diffieuit to determine. Henee, correlation measurements and
speetrum measurements tend to make separate eontributions to the understanding
of a turbulent flow.
As weil as the frequeney speetrum, one ean de fine wave number speetra- Fourier
transfürms of the space eorrelations . This is more eomplieated because one is now
dealing with three dimensions, and the reader is again referred to other sourees for
the theory [38,39]. Suffiee it to say here that one ean defme a quantity E(k),
where k is the magnitude of the wave number, such that

L/p 1
=_q2 = Joo E(k)dk (20.32)
2 0

E(k) indicates the distribution of energy over different length seales. It is an

important parameter in many theoretieal treatments of turbulent motion. However,
it eannot be measured experimentally; one would need simultaneous information
from every point of the flow.
When applieable, Taylor's hypothesis (Seetion 20.6) ean be used to derive a
spatial speetrum from an observed time speetrum. However, this is a one-
dimensionaI speetrum with respeet to the component of the wave number in the
mean flow direction, and so is not in general a eomplete determination of the
speetral eharaeteristies or of E(k).
252 Physical Fluid Dynamics

20.8 The concept of eddies

As we shall see in the following ehapters, the division of a turbulent motion into
(interaeting) motions on various length seales is useful beeause the different seales
play rather different roles in the dynamies of the motion. This is usually expressed
by talking of 'eddies of different sizes'. A turbulent 'eddy' is a rather ill-defined
concept - but a very useful one for the development of deseriptions of turbulenee.
On the smaller seales, one eannot identify individual eddies, and the expression 'small
eddies' means no more than those parts of the motion that are eoherent only over
short distanees. on the large seales, one ean often identify eharaeteristic features of
the motion (Seetions 22.4 and 22.6) and these may be ealled 'large eddies'.
An eddy differs from a Fourier eomponent in the following way. A single
Fourier component, no matter how small its wavelength (that is how large k),
extends over the whole flow. An eddy is loealized - its extent is indieated by its
length seale. (The use of the word eddy does not, however, neeessarily imply a
simple eireulatory motion.) However, small eddies eontribute to larger wave·number
eomponents of the speetrum; the speetrum eurve is often interpreted loosely in
terms of the energy associated with eddies of various sizes.
For any separation, r, the eorrelation eoefficient is determined by all eddies
larger than ~r. Only the largest eddies ean thus be related directly to eorrelation
measurements. Conversely, the observable speetrum funetion has a value at wave·
number k x influeneed by all eddies smaller than ~l/kx (for reasons arising from
the faet that only one·dimensional speetra ean be observed [85]). Henee, it is
usually most eonvenient to use eorrelation measurements to provide information
about the larger seales and speetrum measurements for the smaller seales.
21
Homogeneous Isotropic Turbulence

21.1 Introduction

Many theoretieal investigations of turbulence have been developed around the

concept ofhomogeneous, isotropic turbulence - turbulence ofwhich the statistical
properties do not vary with position and have no preferred direction. An approxi-
mation to such a motion can be obtained behind a grid, such as the one shown in
Fig. 21.1, in a wind-tunnel. The theories have thus been supplemented by observa-
tion. However, we will postpone consideration of experimentally based ideas about
the stnicture of turbulence to the different contexts of Chapter 22. Here, without
detail, we use the context ofhomogeneous isotropic turbulence to illustrate one of
the basic difficulties of understanding turbulence and to develop some ideas relevant
to all turbulent flows.

Figure 21.1 Typical turbulence generating gIid.

The energy production term in equation (20.19) is zero in isotropie turbulence,

and so the motion must decay through viscous dissipation. In theoretieal work, the
turbulence is supposed to be generated at an initial instant and then to decay as
time proceeds. Behind a grid, the re is strong turbulent energy production for the
first ten or so grid mesh-Iengths along the tunnel; the turbulence then becomes sub-
stantially isotropie and decays with distance down the tunnel. (This, in principle,
implies inhomogeneity, but the decay is slow enough for this to be negleeted.) We
shall eonsider eertain features of the motion applying at any stage of this deeay
process.
254 Physical Fluid Dynamics

21.2 Space correlations and the closure problem

The assumption of homogeneity and isotropy much simplifies the formulation of

space correlations. These depend only on the distance between the two points and
not on their location or the orientation of the line joining them. Moreover, it may be
shown that the general correlation, as in Fig. 20.4(a), may be expressed in terms of
the longitudinal and lateral correlations of Fig. 20.4(b) and (c). Thus only these two
funetions of r (denoted respectively by ffr) and g(r» are needed for complete speci-
fieation of the double veloeity eorrelations. When, additionally, the eontinuity
equation is introdueed a relationship between these is found,

(21.1)

Henee, only a single funetion gives the eomplete specification.

The mathematies behind these statements runs as follows. The eorrelation
eoeffieient between velocity eomponent ui at one point and eomponent Uj at a
point r away is a second-order tensor Riir). When there is isotropy [39,133]
Rij(r) = ~(r)rirj + TJ(r)öij (21.2)
For this to take the appropriate forms in the partieular cases oflongitudinal and
lateral correlations
nr) =(I(r) - g(r»Jr2 ; TJ(r) =g(r) (21.3)
Continuity gives
aU;laXj = 0 (21.4)
Substituting this in )_
R ij =Ui(0)u,{ryu 2 (21.5)
gives
(21.6)
whieh reduees to equation (21.1).
Figure 21.2 shows an experimental eheek of equation (21.1). Direet measure-
ments of g are eompared with ones ealculated from measurements of f.
One would now like to determine f as a function of r. No use has been made so
far of the dynamieal equation nor of any statistieal mechanieal principle. These
should eomplete the specifieation of the problem. However, it is at this stage that
diffieulties arise. We shall not eonsider the mathematies in any detail, but the nature
of the problem will be briefly deseribed as the simplest illustration of the diffieulties
eneountered by theories of turbulent motion. Efforts to overeome these diffieulties
are still an important aetivity [191].
A dynamieal equation for the double eorrelations ean readily be formulated
from the Navier-Stokes equation. However, beeause of the non-linear terrns, this
involves triple eorrelations. One form of this equation, known as the Karman-
Howarth equation, is
a (2n +
-U
(U 2 )3/2 a (4.) _ 2vu 2 a (4 af)
---r/----r- (21.7)
at r4 ar r 4 ar ar
Homogeneous Isotropic Turbulence 255

1.0,- I I I

0.8 \~\\

0.6 r
\\ '\
f,g
0.~ -
0.4 fo- \ x

O'.~X
-------X
0,2
o

'C?.~
i~ ~O --- x___"
Or---------------------~~~=&.-~-~-=~O~.----~'~e·====-ir-~~v--~

o 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8

r/M

Figure 21_2 Correlations in grid turbulenee: (X) [(r), measured; (0) g(r), ealculated from[(r);
(e) g(r), direetly measured. From Ref. (258).

Figure 21.3 Representation of triple velocity eorrelation denoted by i; ef. Fig. 20.4.

j is the correlation between two velocity components at one point and one at
another as shown schematically in Fig. 21.3; it contains information about all two-
point triple correlations in the same way as f contains information about all double
correlations. Similarly, an equation for the triple correlations involves fourth-order
ones. Whatever order one goes to, there is always one more unknown than the
number of equations, and the system remains insoluble. This situation is known as
the elosure problem in turbulence, and a variety of suggestions for an additional
hypothesis to provide a solution have been advanced. However, our knowledge of f
as a function of r remains primarily experimental (Fig. 21.2).

21.3 Spectra and the energy cascade

An equation in terms of spectral functions instead of correlation functions may be

obtained by taking Fourier transforms throughout equation (21.7). Again omitting
 256 Physical Fluid Dynamics
mathematical details, this gives
3E(k, t)/3t = F(k, t) - 2vk 2 E(k, t) (21.8)
The problem of the additional unknown remains, of course, but this equation
provides a convenient starting point for further discussion. E is the energy spectrum
of equation (20.32); the dependence on t is shown as areminder that we are dealing
theoretically with a time-dependent situation. F is related to the Fourier transform
of the triple correlation. However, it can be given its own physical interpretation
through its role in equation (21.8).
The left-hand side represents the rate of change of the energy associated with
wave number k. The second term on the right-hand side is a negative term involving
the viscosity and is thus the energy dissipation. It can be shown that

J"o Fdk =0 (21.9)

and so the first term on the right-hand side of equation (21.8) represents the trans-
fer of energy between wave numbers.
Figure 21.4 shows graphs of E and k 2 E for a typical situation.t The latter has
large values at much higher k than the former. Hence the viscous dissipation is
associated with high wave numbers; Le. it is brought about by small eddies. This is
a consequence of the fact that turbulent flows normally occur at high Reynolds
number. The action of viscosity is slight on a length scale of the mean flow (e.g. a
grid mesh-Iength). Yet much more dissipation occurs than in the corresponding

E
k'E
(arbitrary
seales)

20 30 40 50 60 70
kM

Figure 21.4 Energy and dissipation speetra in turbulenee behind a grid (distanee downstream
from grid = 48M, where M = grid mesh-length). Data from Ref. [282).

tFigure 21.4 is based on experimental data. For homogeneous isotropie turbulence, a relation-
ship can be derived between E(k) and the measurable one-<limensional spectrum, thus
overeoming the non-measurability of E(k) mentioned in Seetion 20.7. However, the conversion
involves differentiation of experimental eurves and so E(k) is determined with rather poor
aeeuracy.
Homogeneous Isotropic Turbulence 257

laminar flow. This requires the development of local regions of high shear in the
turbulence; that is the presence of smalllength seales.
The small dissipative eddies must be generated from larger ones. The effect of
this on the energy spectrum is contained in the second term of equation (21.8); Le.
one expects the transfer of energy to be primarily from low wave numbers to
high. This inierence is confirrned experimentally by observations in grid turbulence
of changes in the spectrum with distance downstream.
This interpretation of the behaviour of the terrns of equation (21.8) allows the
development of amodel of turbulence which has relevance not just to homogeneous
isotropic turbulence but to most turbulent flows.
Energy fed into the turbulence goes primarily into the larger eddies. (In grid
turbulence this happens during the initial generation; in other flows to be considered
in Chapter 22 it happens throughout the flow.) From these, smaller eddies are
generated, and then stilI smaller ones. The process continues until the length scale is
small enough for viscous action to be important and dissipation to occur. Th~
sequence is called the energy cascade. At high Reynolds numbers (based on (U 2)1I2
and a length scale defined in a way indicated in Section 20.6) the cascade is long;
Le. there is a large difference in the eddy sizes at its ends. There is then little
direet interaction between the large eddies governing the energy transfer and the
small dissipating eddies. The dissipation is deterrnined by the rate of supply of
energy to the cascade by the large eddies and is independent of the dynamics of the
small eddie s in which the dissipation actually occurs. The rate of dissipation is then
independent of the magnitude of the viscosity. An increase in the Reynolds number
to a stiIl higher value - conveniently visualized as a change to a fluid of lower
viscosity with all else held constant - only extends the cascade at the small eddy
end. StiIl smaller eddies must be generated before the dissipation can occur. (Since,
as we shall see, the total energy associated with these small eddies is small, this
extension has a negligible effect on the total energy of the turbulence.) All other
aspects of the dynamics of the turbulence are unaItered.
This inference, that the structure of the motion is independent of the fluid
viscosity once the Reynolds number is high enough, has important implications and
will be considered again in Section 22.2. It is given the (sornewhat misleading) name
.of Reynolds number similarity.
The dynamies of the energy cascade and dissipation may be supposed to be
governed by the energy per unit time (per unit mass) supplied to it at the large eddy
(low wave number) end. This is, of course, equal to the energy dissipation, e. This
implies that for all wave numbers large enough to be independent of the energy
production processes, the spectrum function E depends only on the wave number,
the dissipation, and the viscosity; that is
E == E(k, e, v) (21.10)
If the cascade is long enough, the re may be an intermediate range (the inertial sub-
range) in which the action of viscosity has not yet come in; that is
E==E(k, e) (21.11)
Dimensional analysis then gives
E = Ae2l3 k- S/3 (21.12)
where A is anumerieal constant.
258 Physical Fluid Dynamics

This famous re sult, known as the Kolmogoroff -5/3 law, was not verified experi-
mentally for some years, beeause the Reynolds numbers of turbulent flows under
investigation were not high enough. Curiously, whilst experimental progress was
being made towards verifying the result, doubts were arising about its theoretieal
validity. At any instant, energy dissipation is not oeeurring roughly uniformly
throughout the turbulenee; there are patehes of intense small eddies involving high
dissipation and other pa tehes where there is little dissipation. This fits in with a
model of the energy transfer process to be mentioned in Seetion 21.4; there is also
experimental evidence that a velocity signal, fIltered at a frequeney cõrresponding
to the small eddies, shows periods of aetivity and periods of quieseenee [215] . This
implies that there is a length seale - the size of the dissipating patehes - that is not
governed solely by € and II and thus raises doubts about the dimensional argument.
The implieations of this for the theory have reeeived some diseussion [122].

10-]

10 -'

0;- 10- 5
~E

e
.2 10-'
ti
e
.2
E
"
B 10-'
'"
!:l-
õi
e
0
.~ 10- 8
'"
.~
"0
ei.
e 10- 9
0

10- 10
+

10- 11

+
10- 12
10 10'
Wave number (m - Il

Figure 21.5 One-dimensional spectrum measured in a tidal channel;!ine has slope of -5/3.
Data from Ref. (119).
Homogeneous Isotropic Turbulence 259

However, there is nowabody of experirnental evidence supporting the -5/3

power spectrum. Good results were first obtained in an oceanographic channel flow,
produced by tides, with a Reynolds number of 4 x 10 7 • Figure 21.5 shows a
spectrum obtained from this, plotted logarithmically, with a line of slope -5/3.
(This is actuallY the measurable one-dimensional spectrum, but it can be shown that
if this is proportional to k -n, then so is the energy spectrum.)

21.4 Dynamical processes of the energy cascade

The discussion in Section 21.3 of the energy cascade did not consider the mecha-
nism by which the transfer of energy from large scales to small scales occurs. It is
instructive to consider first the related but more readily understood process of the
mixing of a scalar contaminant [85, 86] . Suppose a blob of fluid, as shown in Fig.
21.6(1), is marked in some way, for example by being heated or dyed. If this blob is
in a turbulent flow, its distribution will change with time in the way indicated
schematically by successive configurations in Fig. 21.6 (see also Fig. 22.7). Its distri-
bution in space becomes more and more contorted by the velocity fluctuations, but
so long as molecular diffusion plays no role, the marked fluid is always just the same
fluid. If the PecIet number is high, diffusion is negligible at first. However, the
highly contorted patterns involve very steep gradients of the marker and so ulti-
mately diffusion becomes significant, causing previously unmarked fluid to become
marked. The higher the PecIet number, the more contorted the pattem must
become before thiii happens.
This process is analogous to the cascade because it involves the appearance of
smaller and smaller length scales until this is limited by molecular effects (diffusion
or viscosity). However, the energy cascade is a more complicated matter because it
involves the interaction of the velocity field with itself instead of with a quantity
not involved in the dynamies. Any brief account of it must involve oversimplifica-
tion, but three (not wholly independent) processes may be identified.

2 3

4 5 6

Figure 21.6 Schematic representation of successive configurations of marked blob of fluid

within turbulent motion.
260 Physical Fluid Dynamics

The first is the process of repeated instability considered in Section 20.1. Each
stage may give rise to motions not only of greater complexity but also involving
smaller scales than the previous stages. For example, one stage may produce local
regions of high shear that can themselves be unstable.
Secondly, turbulence of a smaller scale may extract energy from larger scale
motions in a way analogous to the extraction of energy from a mean flow by the
turbulence as a whole (Section 20.4).
Thirdly there is vortex stretching [259] . The random nature of turbulent motion
gives a diffusive action; two fluid particles that happen to be close together at some
instant are likely to be much further apart at any later time. The turbulence will
have carried them over different paths. This applies to two particles on the same
vortex line. Th~ process of vortex stretching, considered in Section 6.5, will thus be
strongly present - although occurring in a random fashion. This increases the mag-
nitude of the vorticity, but because of continuity also reduces the cross-section of
the vortex tube. There is thus an intensification of the motion on a smaller scale;
that is a transfer of energy to smaller eddies.
This process may also be seen as a cause of the patchy distribution of dissipation
mentioned in Section 21.3. The vorticity intensification process will be strongest
where the vorticity already happens to be large. At any instant the production of
small eddies is thus occurring vigorously in some places and only weakly in others.
22
The Structure of Turbulent Flows

22.1 Introduction

Figure 22.1 shows four photographs of a turbulent boundary layer. Hydrogen

bubble markers were released in small squares (as deseribed in Seetion 23.4) from a
fine wire paralleI to the wall and perpendieular to the mean flow. The different
pietures were taken with the wire at different distanees from the wall, ranging from
elose to the wall to elose to the outer edge of the boundary layer. It is apparent
that the flow has eonsiderable 'strueture': different parts of the boundary layer
show marked differenees in the dye pattems; in eaeh part eertain features ean be
pieked out, e.g. the longitudinal streaks in Fig. 22.1(a) and the large irregular patehes
into whieh dye has eolleeted in Fig. 22.1 (b) and (e).
All turbulent flows exhibit strueture of this kind. There is no situation in whieh
the motion ean simply be deseribed as 'totally random'. In this ehapter we look at
aspeets of the strueture of turbulent flows, partieularly of turbulent shear flows.
Seetions 22.2-22.6 diseuss and illustrate the main eoneepts that have been
developed to deseribe and interpret this strueture. Mueh ean be said that applies to
all shear flows; such ide as will be illustrated by a variety of flows (depending largely
on the best illustrations available). Some ideas have to be diseussed in the context
of a partieular flow; more detailed attention is given to the turbulent wake
(exemplifying shear flows away from a solid boundary - sometimes known as free
shear flows) and to the turbulent boundary layer in zero pressure gradient
(exemplifying shear flows next to a solid wall).
Seetions 22.7-22.9 stand more on their own. Eaeh deals with a separate aspeet
of turbulent flows: the tendeney for flows to attach to walls or eaeh other (the
Coanda effeet); the effeet of stratifieation; and the tendeney for some flows to
revert to laminar motion.

22.2 Reynolds number similarity and self-preservation

As always, it is neeessary to know the range of applieability of any measurement -

whether it is relevant to other similar flows or whether it is peeuliar to the
partieular situation investigated. Sinee the number of measurements need ed for a
reasonably full understanding of any turbulent flow is large, it is highly desirable to
rnake observations of general applioability. Two ideas help here.
The fust is the concept of Reynolds number similarity already introdueed in
Seetion 21.3. The larger eddies and the mean flow development are independent of
the viseosity (so long as this is small enough to make the Reynolds number large).
262 Physical Fluid Dynamics

Figure 22.1 Turbulent boundary layer structure at different distances from wall. See text
for procedure; flow is from left to right. (a) yur/v = 4.5; (b) yur/v = 50.7; (e) yur/v = 101;
(d) yur/v = 407 (y/6 "" 0.85) (notation of Section 22.5). Ref. [139).
The Structure of Turbulent Flows 263
35

~
'A1\
30

\
25
,.. t
w 2 1/J;jf(w)

211u 20 \
'\
ff 9~~
w'l/[Link])

Ir \
211üii 15

(s -, ) 9uv

\
'"
10
\

5
O~
~
~ I---
o 0
o SOO 1000 2000 2500 3000 3500 4000 4500 SOOO

~ (s - ')
211

Figure 22.2 Comparison of intensity and Reynolds stress spectra in channel f1ow, showing
former extending to much higher frequencies. (Detailed form of co-ordinates is unimportant
except that ordinate is proportional to (frequency)' x spectrum function.) From Ref. (154).

This is true of the mean flow because the last term of equation (20.14) normally
dominates the last-but-one term and because the Reynolds stress is produced by
the larger eddies. (The contribution of different length-scales to the Reynolds stress
may be investigated by applying spectral analysis not only to the energy as in
Section 20.7, but also to the Reynolds stress [84,154]. Figure 22.2 makes a
comparison between the energy and Reynolds stress (one-dimensional) spectra in
turbulent channel flow; the much more rapid fall-off of the latter at high wave
numbers is apparent.) Hence, experiments at any (sufficiently high) Reynolds
number provide information applicable to all values of the Reynolds number. We
shall see in Section 22.5 that some qualification of this concept is needed for the
motion in the vicinity of a solid boundary.
The second idea is that of'self-preservation' [40]. This is the counterpart for
turbulent flow of the occurrence of similar velocity proflles at different distances
downstream in larninar flow (as discussed in Sections 11.3 and 11.8). However,
self-preservation requires not only that the mean velocity distribution should be
similar, but also that all the pararneters associated with the turbulence should have
similar distributions at different stations.
The equations of motion permit self-preservation only in particular cases. For
example, it can occur in a wake only sufficiently far downstream for the mean
velocity deficit at the wake centre to be small compared with the total mean
velocity. The experimental evidence suggests that, when self-preservation is possible,
then it occurs. This enables a description ofthe turbulent motion developed from
measurements at one station to be applied to the whole flow.
264 Physical Fluid Dynamics

Figure 22.3 Two-dimensional turbulentjet; schlieren (see Section 23.4) picture in water,
obtained by having jet at slightly different temperature from ambient. Photo by C. W. Titman.

The ideas of Reynolds number similarity and self-preservation taken in con-

junction allow certain features of turbulent flows to be derived rather simply. We
may illustrate this most readily by considering the development of the mean flow
of a two-dimensional jet. As for a laminar flow, the jet is characterized, far enough
downstream from the orifice, by M, the momentum transported per unit time over
each cross-section. Ignoring a small contribution from the non-linear effects of the
velocity fluctuations,

M= (oo peP dy (22.1)

Putting, analogously with (11.36),

Umax a:x m ; Aa:x n (22.2)
(the origin of x being adjustable) conservation of momentum again gives
2m+n=O (22.3)
Secondly, the concept of Reynolds number sirnilarity implies that II does not enter
into the counterpart of relationship (11.44) and so
A = A (x,M, p) (22.4)
Dimensional requirements now imply that
Aa:X (22.5)
that is
n=1 (22.6)
The Structure of Turbulent Flows 265

Equation (22.3) now gives

1
m=-- (22.7)
2
These results have been obtained more simply than the corresponding results for
laminar flow. In contrast, however, the treatment of the larninar flow can be
continued to indieate the detailed fonn of the velocity profile (equation (11.53)),
whereas nothing more can be derived about the turbulent flow without additional
assumptions or experimenta1 observations.
Figure 22.3 shows an experiment verifying the result (equation (22.5)) that a
turbulent jet spreads linearly. This resuIt is true also ofaxisymmetric jets, just the
same derivation applying (although now m = -1). Thus such jets spread out in a
conieal shape.

22.3 Intennittency and entrainment

In many situations turbulent motion occurs in only a lirnited region - that region
in which high shear has been generated. Turbulent jets and wakes are usually
surrounded by n0n-turbulent fluid; a turbulent boundary layer usually OCCuIS
beneath an inviscid irrotational flow. In such cases the interface between the
turbulent and non-turbulent regions is sharp. It has however a highly irregular
shape with bulges and indentations ofvarious sizes, as shown in Fig. 22.4. The
bulges and indentations are carried downstream by the flow. At the same time the
detaile d shape of the boundary is changing; each bulge and indentation can be
identified over only a limited time.
At a fixed point (such as Ain Fig. 22.4), random alternations between turbulent
and non-turbulent motion occur. Figure 22.5 shows oscillograms of the velocity
fluctuations at different distances from the centre line of a wake. It is possible to
define quite accurately the fraction of the time that the motion is turbulent. This
quantity is called the intennittency factor 'Y. Figure 22.6 shows its distribution in a
two-dimensional wake. At the centre of the wake 'Y is 1; the motion is always
turbulent. Outside the wake 'Y is 0; turbulent motion never penetrates there. But
over a substantial fraction of the wake width, turbulent and non-turbulent motion
alternate. Similar interrnittency distributions can be determined for other flows .

Figure 22.4 Sketch of a wake,..illustrating sharp irregular interfaces between turbulent and
non-turbulent fluid.
266 Physical Fluid Dynamics

(al

(bl

(el

(dl

Figure 22.5 Oseillograms in turbulent wake at x/d = 28, showing inereasing intermitteney at
edge. y/d = (a) 0.87, (b) 2.25, (e) 3.4, (d) 4.2. (Note: these traees were obtained with eleetronie
boosting of high frequencies; this improves the eontrast between laminar and turbulent periods
by showing a quantity related to the vorticity.)

The effeet of intermittency on flow visualization in the outer part of a boundary

layer is apparent in Fig. 22.1(d). Regions can be seen where the dye patches move
irregulady and are highly distorted; and others where they rernain regular both in
spacing and in shapeo
At the interface between turbulent and non-turbulent fluid, the turbulence
spreads. Fluid just in the non-turbulent region will a short time later be in turbulent
motion. This spreading is part of the process of entrainment for a turbulent flow.
Entrainment by laminar flows - the fact that ajet, for example, draws fluid into
itself from the sides - has been discussed in Sections 11.8 and 11.9. In partieular,
through relationships (l1.63}-(11.65), the entrainment was related to energy
dissipation. Turbulent flows have much higher entrainment rates than the
corresponding laminar flows, a fact that may be related to the additional
dissipation by the velocity fluctuations. Associated with the entrainment of new
fluid into a flow such as ajet there must be a process by which new fluid becomes
turbulent. Otherwise a decreasing fraction of the flow would be turbulent, and, for
example, self-preservation could not occur. This process is the spreading at the
interface.
We can now see the reason why the interface is sharp. Associated with the
velocity fluctuations of any turbulent motion are intense vorticity fluctuations.
(The importanee of vorticity in the dynamies of turbulence is apparent from the
The Structure of Turbulent Flows 267

.... .....
.....
", ,
0.8
""
",
",,,
0.6
,,
Vo - v ,,
Vo - Vmin
,,
0.4
,,
,,
,
0.2
",,,
" .......
O~ ____ ~ ______ ~ ______- L______ ~~ ____ ~ ____ ~

o 0.1 0.2 0.3 0.4 0.5 0.6

y/[(x - xo)d] y,

Figure 22.6 Mean velocity profile (solid Une) and interrnittency distribution (broken line) in
a two-dimensiol!.al wake (Re = 1360; x/d> 500). Abscissa is distance from centre of wake,
non-dimensionalized in the way appropriate to self-preservation. Both curves are experimental,
although data points are not shown. From Ref. (268).

discussion in Section 21.4.) We will suppose that the flow in the non-turbulent
region is irrotational, as is usually the case in the examples that have been
mentioned. The acquisition of vorticity by initially irrotational fluid can only be
brought about through the action of viscosity (Section 10.3). Thus the spreading
of the turbulence at the interface involves the action of viscosity and must be
effected by those eddies for which viscosity is significant; Le. the small eddies.
The length scale over which the change from turbulent to non-turbulent (rotational
to irrotational) motion occurs is the size of these eddies, and so the interface
appears very sharp on the scale of the flow as a whole.
The shape of the interface is, on the other hand, influenced by eddies of all sizes.
The irregular and changing corrugations occur with a wide range of length seales. In
particular, the largest bulges and indentations are produced by the large eddies to be
described in Sections 22.4 and 22.6; these eddies are responsible for the fact that
the region in which '1 is non-zero but less than I extends over a sizeable fraction
ofthe flow (see, e.g., Fig. 22.6).
The flow outside the interface, although called non-turbulent, does involve
velocity fluctuations [199]. These are produced by the neighbouring turbulent
region. Evidently the motion of a bulge in the interface changes the pattem of
irrotational motion outside it. However, these velocity fluctuations are entirely
irrotational and are dynamically quite different from turbulent fluctuations. They
268 Physical Fluid Dynamics

attenuate rapidly with distance from the interface. We now see that the inter-
rnittency factor is most appropriately defined as the fraction of the time that
vorticity fluctuations are occurring.
The processes at the interface constitute a subject of study in their own right
[146]. In particular, one wants to understand the way in which the turbulence
spreads. We shall not consider this topic here except to make the following point.
Although the action of viscosity is essential to the spreading process, the concept of
Reynolds number similarity impHes that the spreading rate does not depend on the
magnitude ofthe viscosity. Experimentally this appears to be the case. However,
the earHer discussion of Reynolds number similarity (Section 21.3) does not extend
to this process. The detailed dynarnics are uncertain, but, presumably, the situation
is analogous with the energy cascade; although the spreading is brought about by
the small eddies its rate is governed by the larger eddies. The total area of the inter-
face, over which the spreading is occurring at any instant, is determined by these
larger eddies [212].

Figure 227 Jet in concurrent flow and neighbouring dye streak: (a) jet also dyed; (b) jet
not dyed.

Figure 22.7 illustrates and summarizes the entrainment process described in this
section. Two sources, side by side, ernit water into the uniform flow in a water
channeI. The water from the lower source is ernitted sufficiently faster than the
main flow that it forms a turbulent jet. The water from the upper source is
ernitted at the same speed as the main flow; there is thus no shear and no
turbulence generation. This source merely provides a way of marking fluid with
minimum dynarnical effeet. In Fig. 22.7(a), both sources ernit dyed water. The jet
is visible (it has a rather different structure from jets considered elsewhere in this
book because of the motion of the surrounding fluid). The way in which its
spreading brings the dye from the other source into turbulent motion can be seen.
In Fig. 22.7(b), only water from the passive source is dyed, although the jet is still
present; this shows more clearly the effeet of entrainment on the initially larninar
fluid. The figure also illustrates the processes shown schematically in Fig. 21.6; the
longer the dye streak has been inside the turbulent region, the wider is the Tfmge of
length scales on which it has been distorted.
The Structure of Turbulent Flows 269

22.4 The strueture of a turbulent wake

The understanding of the dynamics of turbulence in shear flows depends primarlly

on the measurement and interpretation of the pararneters introduced in Chapter 20.
Further discussion thus has to focus attention on particular flows, and we consider
the two-dimensional wake to exemplify the procedures and the ide as that eome
from them. Nevertheless we shall frequently refer in this section to ideas that have
more general applicability.
As usual the Cartesian co-ordinate system used to specify directions has x in the
mean flow direction, y across the wake, and z parallel to the axis of the body
producing the wake.
_ OilYious first measurements are q2 , the separate contributions to this
(u 2 , v2 , and w2 ), and the Reynolds stress -uv. Figure 22.8 shows the distributions
of these across a wake; Fig. 22.9 shows the same quantities divided by the inter-
mittency factor, indicating approximately (not exactly, because of the irrotational
fluctuations mentioned in Section 22.3) the average values du ring the intervals
when the point of observation is inside the turbulent region.
The Reynolds stress is zero right at the middle of the wake, by symmetry, and,

0 .12 ,....---- y-----,....--- - - - , - - - - , -- - -y-------,

0.10

0.08

;}(I< - I<o)
0.06

elC.

0.04

0 .02

0 .1 0 .2 0 .3 0.4 0 .5 0.6
yll (I< - 1(0)d ] 'I>

Figure 22.8 Distributions of intensities and Reynolds stress in a two-dimensional wake

(Re = 1360, x/d> 500). Co-ordinates are normalized in the way appropriate to self-preservation;
y = 0 corresponds to wake centre. All curves represent experimental data (Ref. [268 J).
270 Physical Fluid Dynamics
0.1 2 r------r----===:------r----,---~--__.,

0 .10

0 .08

.?" (x - xo)
1 0 .06
"'(Und
etc

0.04
"Q'n
~

0.02 ;;
w1
-uv
O~--~~----~----~-----L----~----~
o 0.1 0.2 0.3 0.4 0.5 0.6

y/[(x - xo)d) \OI

Figure 22.9 Curves of Fig. 22.8 divided by the interrnittency factor.

of eourse, it falls to ?:ero outside the wake. Its maximum is in the vieinity of the
maximum ofaU/ay (ef. Fig. 22.6) and its general, although not its detailed, effeet
on the mean veloeity is similar to that of a viseous stress. -uv is mueh larger than
vaUjay. It eauses the wake to inerease in width and deerease in veloeity defieit with
distanee downstream.
The interpretation of these observations is assisted by information on the energy
balanee; whieh processes are supplying energy to the turbulenee at eaeh point and
whieh are removing it? Figure 22.10 shows distributions of eaeh of the terms in
equation (20.21) measured in a wake.
The maximum energy produetion is elose to the position of maximum Reynolds
stress. The turbulence is being kept going by the working of this against the mean
veloeity gradient. At the centre of the wake, the produetion is zero (sinee both uv
and aUl ay are zero there). The dissipation, on the other hand, has a maximum at
the centre. This is balaneed primarily by the adveetion term - IhU3 q 2 jax. As the
mean [Link] defieit decreases with x, so, in accordance with self-preservation, the
seale of q2 decreases; thus the fluid at the centre of the wake is losing turbulent
energy as it travels downstream.
At the outer edge of the wake, on the other hand, the advection term has the
The Structure of Turbulent Flows 271
opposite sign. This corresponds to the supply of turbulent energy to previously non-
turbulent fluid as the wake spreads. Figure 22.1 0 shows that this energy is supplied
by the outward transport of energy by the fluctuations themselves from the region
where the production is large.
When measurements such as those described above are supplemented by
measurements of correlations and spectra and by flow visualization experiments,
some ide as may be developed about the role of eddies of different sizes in the
dynamies of the turbulence. We have already seen that the smaller eddies contribute
relatively less to the Reynolds stress than to the energy (illustrated for a different
flow by Fig. 22.2). More generally, the smaller eddies have a structure that is less
characteristie of the partieular flow than the larger eddies and the description of
their behaviour developed from observations in grid turbulence (Chapter 21) may
be applied. The later stages of the energy cascade are similar in all flows. It is
because of this that observations in shear flows may be used for experimental
investigations ofthe Kolmogorofflaw, equation (21.12).

(i ii)

GAIN

o ~------------------~-------=----~~

LOSS

o 0.1 0 .2 0.3 0.4 0.5

Y/! (x - Xo Idi II

Figure 22.10 Experimentally determined turbulent energy balanee in a two-dimensional

wake. Abseissa seale as in previous graphs, ordinate seale arbitrary (Re = 136Q. x/d> 500).
(i) Pr~uetion, -uvõU/õy (il) dissipation,vujÕ 2ujl õxJ (üi) advection, -lIzUõq' fõx (note
V.Võq2fõy is negligible in a wake); (iv) turbulent transport, -õ(l/zq2V + Pü/p)fõy. From
Ref. [268].
272 Physical Fluid Dynamics

The larger eddies, that do play ~ role in the generation of the Reynolds stress,
must be more eharaeteristie of the partieular flow, sinee the Reynolds stress
distribution neeessarily varies from flow to flow. SO long as the eddies are stilI
relatively small eompared with the length seale of the mean flow (e.g. wake width),
their features are adequately deseribed by the eonsiderations of Seetion 20.4. The
anisotropy associated with Reynolds stress produetion (Fig. 20.2) arises essentially
in the way summarized by Fig. 20.3. Usually, the largest eontribution to the energy
of the turbulenee is made by eddies large enough to be oriented in this way but
small enough to be within one part of the mean velocity proftle.
Stilllarger eddies extend aeross mueh of the flow and there will be appreciable
variation ofthe mean shear within their length seale. The ideas of Fig. 20.3 are then
too great an oversimplification. One ean identify eertain patterns in the turbulenee
on this seale that oeeur over and over again; one says that the large eddie s have a
rather definite strueture and orientation. This strueture varies between different
flows mueh more than the motion on smaller seales. The large eddies play a role out
of proportion to their eontribution to the turbulent energy, both in the interaction
between the mean flow and the turbulenee and in the turbulent energy transfer
process involved in Fig. 22.10. It is beeause ofthis that the ealeulation of the
development of turbulent flows is so diffieult.
Extensive experiments have been made for some flows - inc1uding the two-
dimensional wake - in attempts to elueidate their large eddy strueture. Models of
the strueture are usually based on a eombination of flow visualization experiments
and measurements of various eorrelation funetions. Numerous measurements are
needed before one can be reasonably sure that the model eorresponds to real
features ofthe flow. We eannot eonsider here all the data for any one flow, but an
example may be given simply to illustrate the type of reasoning on which
descriptio ns of the large eddies are based. Figure 22.11 shows measurements in a
two-dimensional wake of the eorrelation of the y-eomponents ofthe velocity at
two points separated firstly by a distanee r x in the x-direction and seeondly by a
distanee r z in the z-direetion (both measured at a distance from the eentre plane
corresponding to (Uo - U)j(Uo - Umin ) = 0.65, ef. Fig. 22.6). In isotropie
turbulenee these two eurves would be identieal. It is seen that this is far from
the ease; in partieular Ryy(rx ) has a marked negative region, whereas Ryy(rz ) is
always positive. One feature of the large eddy strueture must be that the return
flow for veloeity fluetuations towards or away from the eentre of the wake oeeurs
dominantly in the upstream and downstream direetions and little in the transverse
(z and -z) direetions.
The aeeumulation of a large amount of evidenee of this type together with
visual observations has led to the suggestion that two dominant large-seale
struetures oeeur in wakes [118, 271] . These probably originate in rather different
ways.
The first eddy type is shown in a highly schematic way in Fig. 22.12. In each
half of the wake the eddy consists of two paralleI vortex tubes of opposite sense.
There is some flow along these tubes, as well as arounll them, so that a spiralling
motion results. The tubes may not be straight, but one does not have sufficient
information to assign any other shape to them. Each tube has its axis in an xy-plane,
with the part near the edge of the wake further downstream than the part near the
centre, and the two tubes are separated in the z-direction. Such motions oecur
The Structure of Turbulent Flows 273

1.0 r - -- - - - , , - - - - - - , - - - - - - , - - - - - - r - - - . , - - - - - - , . -- - - - - ,

0.8

0.6

0.4

0.2

o -x x
o~----~~-----------------=~====~~~
o ~O
~o-__~o-----o
- O.2~--~~--~---~--------L---~--~
o 2 4 6 8 10 12 14
,Id

Figure 22.11 Correlation curves in a two-dimensional wake for the y-component of the
velocity fluctuation and points separated in the x( 0) and z(x) direetions (Re = 1300, x/d = 533,
y/d = 4). Data from Ref. [118).

Figure 22.12 Schematic representation of double roller eddy in a wake.

274 Physical Fluid Dynamics

".----- ... , ....

,
I
I
I
I

------
,,
,
,
,

- - - _... -'
~CYLlNDE.R

Figure 22.13 Sketch of 'jet-type' large eddies in a wake. From Ref. [118).

simultaneously on both sides of the wake so that the eddy as a whole has the shape
indicated in the figure.
It has been shown that motions of this type can arise as the resuIt of the action
of the mean shear on initially isotropic fluctuations - these being generated, it is
assumed, in the process of non-turbulent fluid becoming turbulent at the interface.
One can readily see that shear will have a selective action on different fluctuational
vorticity components, some vortex lines being stretched and the vorticity
intensified, others being shortened and the vorticity reduced. However, the details
of the structure so selected cannot be inferred without a mathematical theory
[271] .
The second type of large eddy motion in a wake occurs simultaneously at
several (typically five) fairIy evenly spaced positions along the wake in the
streamwise direction. The motion consists of a jet-like flow from near the centre
of the wake outwards, producing a bulge on the interface between the turbulent
and non-turbulent fluid. As it travels outwards, the fluid tends to curve round in the
flow direction. The retum flow towards the centre of the wake is more diffuse.
Again the re is some correlation between the motion on the two sides of the wake,
so that the motion as a whole has the general form indicated by Fig. 22.13.
It is these eddies that relate particularIy to the example of the interpretation of
correlation measurements given above.
Also, these eddies are particularly apparent in experiments with dyed wakes.
One needs cine-mm to leam much from such flow visualization, but the eddies can
be seen in Fig. 3.11.
The probable origin of this type of motion is aloeal instability of the mean
velocity profile, somewhat analogous to the instability of a laminar wake that
produces aKanmin vortex street (Section 3.3). There is a major difference in the
The Structure of Turbulent Flows 275

result of the instability in that the large eddies are of very limited extent in the third
direction. Also, beeause it is probably triggered by the intense smaller-seale
fluetuations, the oeeUITenee of the instability is intermittent and irregular.

22.5 Turbulent motion near a wall [40, 212J

The general ideas of the last seetion, although made speeific in the context of wake
flow, apply to all turbulent shear flows. However, they provide a very incomplete
story when the flow is adjacent to a solid boundary.
At such a boundary (at rest), the boundary condition that the fluid veloeity is
zero applies at every instant. Thus it applies to the mean velocity and to the
fluetuations separately,
Uj=O; Uj =0 (22.8)
The faet that the fluctuations drop to zero at the wall has the particular implication
that the Reynolds stress is zero
-uv =0 (22.9)
The only stress exerted directly on the wall is the viscous one.
Away from the wall, on the other hand, the turbulence generates a Reynolds
stress, large eompared with the viscous stress, in the usual way. The total stress
T = pi}Uj3y - puv (22.10)
(we are stili considering a two-dimensional flow for which the boundary layer
equations apply). This eannot vary rapidly withy without producing a very large
mean acceleration and thus requiring an improbable mean flow distribution. (For
example, in channel flow with no variation of mean quantities in the x-direetion, T
varies linearly aeross the channel as it does for larninar motion - the flISt integral
of equation (2.6).) Consequently, the viscous stress elose to the wall must match
up with the Reynolds stress further out. Although T varies only slowly, /J.3Uj3y
and -puv eaeh vary rapidly, the former being much larger at the wall than in
laminar flow but beeorning very small away from the wall, the latter being large
away from the wall and zero at the wall.
Figure 22.14 iliustrates this by showing distributions of T and of its two
contributions-measured in a zero-pressure-gradient boundary layer. The region
where the viscous stress makes a large contribution is a small fraction of the total
boundary layer thickness (note the change in abseissa scale). To generate this
distribution of viseous stress, the mean velocity proftle must rise steeply at the
wall and then become comparatively flat. Figure 22.15 compares a typical
turbulent boundary layer profile with a corresponding laminar one.
ParalleI considerations apply to turbulent pipe flow. Figure 22.16 compares
laminar and turbulent me an veloeity profiles, firstly for the same flow rate and
secondly for the same pressure gradient.
It is now elear that the presence of the wall causes the fluid viscosity to enter in
a much more important way into the dynarnics of turbulent motion than it does for
free flows. The concept of Reynolds number similarity is no longer so useful.
276 Physical Fluid Dynamics
yu,/v
20 40 60 80 100 120 140

0.03 0.04 0.05 0.2 0.4 0.6 0.8 1.0 1.2

y/8
Figure 22.14 Distributions of the total stress, Reynolds stress, and viseous stress aeross a
boundary layer (Re ö = 7 x 104 ). Note 30-fold ehange in abscissa seale at ylo = 0.05. Based on
data in Refs. [136, 223 J.

This effect extends to the fluctuations as well as ~the mean flow. Figure 22.17
shows distributions of q2 and of its components u 2 , v2 , and ~ across a boundary
layer. (The reason for the alternative co-ordinates will be seen later.) Large varia-
tions are to be observed with strong maxima close to the wall. (Figure 22.17 shows
also the distribution of the intermittency factor and it is evident that division by
this - as was considered for wake flow, Figs. 22.8 and 22.9 - would affect only
the shape of the tails at the larger values ofyjo.)

U
Uo

OL-__ ~ ____ ~ __ ~~ __ ~ ____L-__ ~ ____~__~____~

o 2 4 6 8 10 12 14 16 18
y(Uo/vx)'h

Figure 22.15 Comparison of larninar and turbulent boundary layer velocity promes.
(Laminar: Blasius prome; turbulent: experimental prome at Rex = 9.2 x lOs, with tripping at
x = 0 to promote transiüon.)
The Structure of Turbulent Flows 277

This observation ean be interpreted in connectian with the turbulent energy

balanee, Fig. 22.18. The rate of energy produetion, -uv au/ay, also has a large
peak elose to the wall. The reasan is related to the above diseussion of the stress.
Very elose to the wall-uv is small and so there is little energy produetion; far
from it, au/ay is small with the same eonsequenee. Mathematieally,

-uv
ay =.!.p
au (r _" ayau)auay
,... (22.11)

(a)

(b)

Figure 22.16 Comparison of (i) laminar and (ü) turbulent velocity profiles in a pipe for
(a) the same mean veloeity and (b) the same pressure gradient. (The diagrams eorrespond to a
turbulent flow Reynolds number of about 4000; for higher Re the eontrast is more rnarked -
ef. Fig.-2.11.)
 278 Physical Fluid Dynamics
yuTI/)
20 40 60 oo 100 120 140

16
,,,.-"',,,
X 10- 3 ,
,
I

10 14 I

I
I
12 I
8 I
10 1.0
Il -;;.
6
U T2 U~ 8 0.8
etc. etc.

6 0.6 'y
4

4 0.4
2
,I "
2 I v2 0.2
_ ........
I
""~... .,.

0.01 0.02 0.03 0.04 0.05 0.2 0.4 0.6 0.8 1.0 1.2
ylfJ

Figure 22.17 Distributions of total and component intensities and of intermittency factor
across a boundary layer (RefJ = 7 x 104 ). Based on data in Ref. [136].

and, if 7 is treated as a constant, this is a maximum when

/lau/ay =7/2 (22.12)
Thus the energy production is largest in the vicinity of the changeover from a
predominantly viscous stress to a predominantly turbulent one.
The energy balance diagram, Fig. 22.18, shows that over the region where the
production is high the other important term is the dissipation; the remaining terms
in equation (20.21) are much less significant (apart from some rather complicated
transfer processes very close to the wall). The inner part of the boundary layer is
thus sometimes said to be in 'local equilibrium' - meaning that there is aloeal
balance between the process supplying energy to the turbulence and that removing
it.
Towards the outer edge of the boundary layer the quantities in equation (20.21)
are much smaller, but all the terms are now important. The dissipation somewhat
exceeds the production and there is also a loss by advection - corresponding, as in
a wake, to the supply of energy to newly turbulent fluid. These losses are compen-
sated by turbulent transport. The turbulence in the outer region is maintained by
that in the inner region.
More detaile d consideration of these and related ideas has led to adivision of the
turbulent boundary layer into inner and outer regions. The dynamical processes
occurring in the latter are not dissimilar from those in a wake, but the inner region
requires its own model. The ideas that have been developed for this region can be
applied to most turbulent flows adjacent to a wall- for example, to turbulent
The Structure of Turbulent Flows 279

YUT/V
o 20 40 60 80 100 120 140
50

GAIN (i) GAIN

(iii)

I
I
I
I
LOSS I LOSS
I
I
I
I
I
I
I
-50L-____~____ _ L_ _ _ _~_ _ _ _~~ _ _~1_L_L_ _~_ _~_ _~~~-1

o .01 .02 .03 .04 .05 0.4 0.6 0.8 1.0

y/6

Figure 22.18 Energy balanee diagram (prineipa1 features) for a turbulent boundary layer
(Re6 = 7 x 10'). (i) Produetion; (ü) dissipation; (iü) adveetion; (iv) transport aeross boundary
layer (primarily by turbulenee, but inc1uding viseous eontribution at low y /6). (lnner part
based on Ref. [212]; outer part based on Ref. [136] but modified to reduee diserepaneies
with Ref. [212] and with requirement that integrated transport should be zero.)

ehannel and pipe flow (although not, for example, to a boundary layer elose to
separation). Beeause of the loeal equilibrium, the strueture of the inner region is
substantially independent of the outer flow.
We thus want to specify the flow in the 'wall-region' by a velocity seale
eharaeteristic of this region. This is provided by TW, the value of T at the wall, which
has the dimensions of density x (velocity? Henee we define
UT =(TW/p)1I2 (22.13)
as our veloeity scale. UT depends on the flow as a whole, but onee it is specifled,
the structure of the wall region is specifled. It is conflrmed experimentally that the
turbulent intensity distributions scale with UT. For example, the maximum value of
u2 is always about 8u;.
The relationship between UT and the external veloeity Uo for a boundary layer
depends (rather weakly) on the Reynolds number. Under typicallaboratory
conditions, UT/UO is in the range 0.035 to 0.05.
Conditions in the wall region are now speeifled by the three pararneters: Up the
distanee from the wall y, and the kinematie viseosity II. The mean velocity U is a
280 Physical Fluid Dynamics
function of these
U= [(uT,y, v) (22.14)
from which dimensional considerations give
U/u T =[(yuT/v) (22.15)
It is again confirmed experimentally that various turbulent wall flows have a
common proflle of this form.
We have seen that the viscosity is important only very close to the wall. With
increasing y it ceases to play a role long before parameters other than those in
(22.14) have an influence. One can then say that the mean velocity gradient depends
only on UT and y,
au/ay = [(uT,y) (22.16)
although one cannot say the same for U because it is separated from its origin by a
region in which v is important. Dimensional analysis applied to (22.16) gives
au/ay = uT/Ky (22.17)
where K is a universal constant (the Karman constant) of turbulent wall flows. It is
found exper:mentally that K = 0.41. Integration of (22.17) gives, in a form
corresponding to (22.15),

~ =k[ln(Y~T) +A] (22.18)

where A is another constant.

This logarithmic proflle is one of the most famous results in the study of
turbulent flows. In Fig. 22.19, the velocity proflle of Fig. 22.15 is plotted with
log-linear co-ordinates and the straight line corresponding to equation (22.18) is
evident.
The proflle departs from the logarithmic form whenyuT/v < 30, because of the
importance of viscosity in this region. Very elose to the wall viscous action is
dominant. At the wall
U=v=w=O (22.19)
and, taken in conjunction with the continuity equation
aU av aw
-+-+-=0 (22.20)
ax ay az
this gives
av/ay =0 at y =0 (22.21)
Hence,
a(Uv) = a2 (uv) = 0 at y =0 (22.22)
ay ay 2
and treating 7 as a constant in equation (22.1 0)
a2 u =-=0
-
a3 u at y=O (22.23)
ay2 ay 3
The Structure of Turbulent Flows 281

v/ 8
0.004 0.01 0.04 0.1 0.4
25rr----r-_r_r-r----r---_r--~~r_--_r----~_.~~~1.0

20 0.8

15 0.6
U U
Uo
10 0.4

5 0.2

4 10 40 100 400 1000

Figure 22.19 Example of a turbulent boundary layer mean velocity profIle plotted on log-
linear co-ordinates (Rell =2.3 x 10'). Solid lines correspond to equations (22.18) and (22.24);
broken line shows trend of data.

We thus expect there to be a significant region right next to the wall in which the
velocity proftle .is Jinear
U/u T = yuT/v (22.24)
The curve corresponding to this is included in Fig. 22.19 (the Jogarithmic pJotting
now serving to expand this region) and it is seen that the experimental points fall
on it for yuT/v < 8.
This thin region is known as the viscous sub-Iayer. Sometimes it is called the
laminar sub-Iayer, but that is less accurate as it contains large veloeity fluctuations;
u2 and w 2 are free to rise more rapidly from the wall than uv. Although the viscous
sub-Iayer is very thin, a substantial fraction of the total mean veloeity change across
the boundary layer occurs within it.
The upper limit in y of the logarlthmic profile occurs where dynamieal processes
relati~g to the boundary layer as a whole become significant - where, for example,
the large eddy structure (see Section 22.6) is influenced by the outer part of the
layer. This is observed experimentally to occur around y/c5 = 0.2, where 15 is the
total boundary layer thickness (defined, say, so that U= 0.99Uo aty = 15). The
corresponding value ofyuT/v (about 200 in Fig. 22.19) depends on the Reynolds
number Uoc5/v.

22.6 Large eddies in a boundary Jayer

The fact that any turbulent flow involves large eddies with a structure characterlstic
of that flow was introduced in Section 22.4. Wall flows have some speeial features,
282 Physical Fluid Dynamics

which we illustrate with a brief discussion of the large eddie s of a boundary layer.
This is currently a topic of much research activity and an account of it may
become rather quickly out of date. At present, the interpretation is complicated
by the fact that, in contrast with the wake (Section 22.4), correlation measure-
ments and flow visualization experiments apparently reveal different aspects of the
large eddy structure; full synthesis of the different observations has not yet been
made.
Correlation measurements indicate that the selective action of the mean shear,
described in Section 22.4, again operates in the outer part of a boundary layer.
Indeed, it may be expected to operate in any shear flow with an interface with
non-turbulent fluid. The result, in a boundary layer, is the occurrence of large
eddies with a structure like one half of that shown in Fig. 22.12. Ref. [271].
However, the correlations are not in all respeets similar to those in a wake,
indicating that other large eddies are different. Information on the structure of
these has come primarily from flow visualization experiments, such as those in
Fig. 22.1. Again, stiIl pictures can convey only a limited impression of what is seen
in motion, but Fig. 22.20 shows photographs of a particularly informative type; the
method of flow visualization is the same as in Fig. 22.1 , but the boundary layer is
seen in side-view (the dye release wire is perpendicular to the wall) instead of plan-
view.
Two types of large-scale motion are apparent in such experiments, an eruption
of slow moving fluid away from the wall and an inrush of rapidly moving fluid
towards the wall.
The former originates in the viscous sub-layer elose to the wall. The streaky
structure seen there (Fig. 22.1 (a)) is produced by regions, separated in the
z-direction, of fluid moving faster than average downstream and of fluid moving
more slowly than average. Fluid in the latter regions intermittently erupts, in a
motion with a highly repeatable structure [134] , into the main body of the
boundary layer. The sub-layer appears to 'burst', and, for this reason, large eddies
of this type are often known as bursts. Figure 22.20(a) shows a boundary layer
during this process; the motion away from the wall is apparent and the spacing of
the dye patches shows also that this fluid is moving downstream more slowly than
average. The eruptions vary in size, but some of them penetrate right through the
boundary layer to contribute to the intermittency at its outer edge.
The second type of motion - the inrush - is in its main features just the reverse
of an eruption. Fluid moving faster than average downstream is carried in towards
the wall. This gives it an eve n more marked excess longitudinal velocity and this
structure is particularly apparent in the inner half of a boundary layer. Figure
22.20(b) was taken during an imush; the slight variation of the longitudinal
velocity ac ross most of the boundary layer, as indicated by the spacing of dye
patches, implies that fluid elose to the wall is moving much faster than average.
The processes generating the eruptions and inrushes and their relationship to the
shear selected eddies remain open questions. It is often suggested that the eruptions
are produced by an instability in the sub-Iayer region. Certain features of the region
resemble markedly features of transition to turbulence in a boundary layer - for
example the streaks resemble the development of three-dimensionality described in
Section 19.1. However, this interpretation faces the difficulty that essentially
identicallarge-scale motions occur in boundary layers on rough walls, for which the
flow structure elose to the wall is quite different [68, 121].
The Structure of Turbulent Flows 283

Figure 2220 Turbulent boundary layer (a) during eruption, (b) during inrush. Flow is from
left to right, wall is at bottom; see text for procedure. From Ref. (121).

The eruptions and the inrushes both have struetures generating a large Reynolds
stress. In faet, between them, they are probably responsible for almost the whole
observed Reynolds stress; this implies that they are also the main sourees of energy
for the turbulenee. Orientation of smaller seale motions plays little part in Reynolds
stress generation in a boundary layer.

1.0

0.8 X

0.6 \
X
R"J<

0.4
\ X

0.2

O~---L----~--~-- __ ~
x_X
-------x
____L __ __ L_ _ _ _ ~ _ _~_ _ _ _~_ _~

o 0.2 0 .4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2 .0

,,,lõ

Figure 22.21 Variation of correlation coefficient of longitudinal velocity components with

longitudinai separation, 'x' in wall region of turbulent boundary layer, Ylo = 0.097. Note that
correlation remains significant at values of,x of order of boundary layer thickness Ei and
large compared with distance from wally. From Ref. (277).
284 Physical Fluid Dynamics

A eurlous feature of the large-eddy motion close to the wall is that not all
quantities associated with it seale in the way implied by the diseussion of Seetion
22.5 [188] . For example it involves a length seale whieh is large eompared with the
distanee from the wall (illustrated by the eorrelation funetion shown in Fig. 22.21).
Evidently this seale is being imposed on the wall region by the (less intense) fluetua-
tions further out. At first sight, this invalidates the argument leading to equation
(22.18). To provide a reeoneiliation, it has been suggested [68,269] that the large
eddies near the wall involve a 'universal' or 'aetive' motion - with a length seal e
proportional to the distanee from the wall - and an 'irrelevant' or 'inaetive'
motion - with a mueh larger length scale. The idea requires some modifieation to
fit in with eurrent models of the large eddies, but some such separation of the roles
of different length seales must apply; the oeeurrenee of the logarlthmie velocity
profile is well established.

22.7 The Coanda effeet

Suppose a two-dimensional jet emits into a region bounded by a side wall, as shown
in Fig. 22.22. It is found that the jet is pulle d down on to the wall giving the type
of flow shown in the figure. This behaviour is known as wall attaehment and is an
example of what is ealled the Coanda effeet.
The reason for it ean be understood as follows. In the ease of an uneonfined jet,
entrainment (Seetion 22.3) produees an inflow of fluid from the sides. In the
presenee of a wall, the jet eannot draw fluid into itself in this way. Instead it is
drawn down on to the wall.
Beeause of the higher entrainment rate, the Coanda effeet oeeurs mueh more
strongly when the flow is turbulent; for example, a turbulent jet will attach to a
more distant wall than a laminar jet. Similarly, two-dimensional flows exhibit the
Coanda effeet more strongly than three-dimensional; in the latter ease fluid for
entrainment ean enter from the sides.
The Coanda effeet ean oeeur in a variety of flows. Another example is the
reattaehment after transition of a separated boundary layer (Seetion 11.4). It ean
also take the form of the attaehment of two flows to one another.
We may illustrate the Co anda effeet with the example of two turbulent thermaI
plumes (Seetion 14.6) attaehing in this way. The photographs in Fig. 22.23 were
obtained by having two long thin parallel heating elements close to the floor in a
tank of water. Dye eould be introdueed into the inflow to the plumes from two

77n~~
<;L;))));;;:
Figure 22.22 Schematic diagram of wall-attaching jet.
 The Structure of Turbulent Flow!! 285

Figure 22.23 (a) Single plume; (b) plumes showing Coanda effeet; (e) the same flow as in (b)
but with dye on onlyone side.
286 Physical Fluid Dynamics

sources, also close to the floor, a little way to the side of each element remote from
the other element. In Fig. 22.23(a) only the right-hand element is switched on and
the single plume can be seen. In the other two pictures both elements are being
heated. Each plume would normally entrain fluid from the space between them.
However, since there is no way in which this fluid can be replaced, this entrainment
is not possible. Instead, the two plumes are drawn together into a single plume
rising above the line midway between the two elements. Figure 22.23(b) shows this
flow. Figure 22.23(c) shows the same situation, but only the right-hand dye source
is present; the change in direction of motion of fluid from the right as it meets
fluid from the left can be seen.

22.8 Stratified·shear flows

Turbulent flow of stratified fluids is a topic of some importance for its geophysical
applications. It has been the subject of some laboratory studies, but not enough for
any very full account to be based on these aloneo Many of the most important ideas
have been developed in a meteorological context, principally the study of the lowest
layers of the atmosphere as the wind blows over hot or cold ground [172,205,
243] . This is too large a subject to be diseussed here. Hence, this seetion just
introduces some of the basic ideas without attempting to show their origin in any
detail.
We are coneerned with the predominantly horizontal mean flow of a fluid whose
mean density varies vertically. The mean velocity also varies vertically, and we shall
confine attention to two-dimensional flow. Henee, the speeification of the situation
is primarily in terms of the two gradients, dU/dz and dpo/dz.
The veloeity gradient can lead to generation of turbulence in the usual way
through the action ofinertia forees (Section 2004). The role of the density gradient
depends on its sign. If the density increases upwards, then buoyaney forees provide
an additional source of energy for the turbulence. If the density decreases upwards,
then the turbulence must do work against buoyancy forees, which therefore
produce a loss of turbulent energy additional to viscous dissipation; turbulence
cannot persist when the density gradient is too large.
From the considerations of Chapter 16, we expect the quantitative formulation
of the relative importance of inertia and buoyancy forees to be made in terms of
some form of the Froude number (equation (16.7)). The role of the veloeity
gradient in the dynamies of turbulent flow suggests that U/L should be replaced by
dU/dz. Also it is convenient to work in terms of the reeiprocal of the square of this
Froude number; that is in terms of
Rl = _ .:::.g(-,-dp:....;o,,-Id--,z)'-="
(22.25)
Po(dU/dz)2
This is called the Richardson number (sometimes the gradient form of the Richard-
son number to distinguish it from other forms defined somewhat differently). As
was to be expected, it depends on the sign of the density gradient but not on that
of the veloeity gradient.
Negative Rlchardson number corresponds to a destabilizing density gradient;
both shear and buoyancy give rise to turbulence generation. When - Ri is small, the
former is dominant and the motion is essentially of the type we have been consider-
The Structure of Turbulent Flows 287
ing hitherto. When -Ri is large, the latter is dominant and the turbulence may be
more like the free convection turbulence described in Section 4.3.
Positive Richardson number corresponds to a stabilizing density gradient;
turbulent motion cannot be sustained when Ri becomes large.
Various experiments have been carried out to discover the more detailed
dynamies of these processes. The simplest configuration in principle (although not
in practice) is a flow with uniform velocity and density gradients. Such a configura-
tion has been achieved using a special wind-tunnel with graded heated element s and
flow resistance grids at its entry [290]. It was used for investigations with positive
Richardson number. The numerical details of the results may not be generally
applicable, since the flow was stiil developing in the downstream direction at the
observing station (and the Reynolds number may not have been high enough for Ri
to be the only parameter); but the results should indicate weil the general trends.
The damping action of the stratification on the turbulence is illustrated by Fig.
22.24, which shows the variation of the temperature fluctuations relative to the
temperature gradient as a function of Richardson number (since the geometry was
constant, the fact that the ordinate is not non-dimensional is unimportant). The
turbulence is almost completely suppressed when Ri reaches 0.45.
Changes in the structure of the turbulence also occur. Figure 22.25 shows the
variations of the correlation functions _uw/(u 2 w 2 )1I2 and _w8/(W 2 (J2)1/2 (where 0
is the temperature fluctuation); the former is the normalized Reynolds stress,
whilst the correlation -wO plays a role in the transport of heat by the turbulence
corresponding to the role of the Reynolds stress in momentum transport. The latter
falls off more rapidly, indicating that the turbulence changes in a way that makes it
relatively less efficient as a heat transfer mechanism than as a momentum transfer
mechanism. This can be understood physically in this way: a fluid particle that is

Figure 22.24 Variation of intensity of temperature fluctuations with Richardson number in

a stratified shear flow. From Ref. [290).
288 Physical Fluid Dynamics
0.6 .------~-----,----.__---r_____,
x

X
0.5
x

- uw (!J.
((71 .;;r IV. 0.4

0.2

0.1

OL-_ _ ~~ __ ~~ __ ~ ___ ~_~

o 0.1 0.2 0.3 0.4

Ri

Figure 22.25 Variation of normalized Reynolds stress (x) and heat transfer (0) with
Richardson number in a stratified shear flow. From Ref. [290J.

displaced but then falls back to its original position without any mixing with its
newenvironment does not transfer any heat, but it can transfer momentum
through the action of pressure forees. At the higher values of the Richardson
number, the turbulence may be thought of as a random superposition of internal
waves (Section 16.4).
In other situations, complexities may arise from variations in the Richardson
number, as defined above, from place to place. As an example of the consequences
of this we consider flow close to a horizontal upward-facing heated of cooled
surface - sufficiently close that the velocity scale is determined entirely by the
wall stress and is UT as discussed in Section 22.5. The temperature scale is then
simHarly determined by the vertical heat transfer, which, like the stress, varies little
in the region under consideration. lf H is the rate of heat transfer per unit area
(taken as positive if the transfer is upwards, for example from hot ground to cooler
air), this scale is
(22.26)
UT enters this expression because the temperature variations needed to produce a
given heat transfer are smaller when the turbulence is more vigorous. Specification
of the problem in terms of UT and OH is particularly useful in application to the
The Structure of Turbulent Flows 289

atmospheric boundary layer, as these are often known when the dynamies of
the whole system are not.
If stratification is having little effect on the dynamies of the turbulence, we know
that the veloeity gradient in the region under consideration is given by equation
(22.17),t
au/az =uT/Kz (22.27)
Similar considerations applied to the temperature field lead to+
aT/az = -0 H/Kz (22.28)
Whether the turbulence is indeed unaffected by the stratification is indieated by the
Richardson number
Ri =gOi dT/dz =_ gaKO H Z =_ grxKHz (22.29)
(dU/dz)2 u; pCpu;
The Richardson number and thus the relative importance of the stratification
increase with height. It is useful to introduce the height \ L \ at whieh \ Ri \ = I
L = -pCpu;/gcxKH (22.30)
This is the Monin-Obukhov length; like the Richardson number it is defined to be
positive when the stratification is stabilizing and negative when it is destabilizing.
For either sign, when z ~ \L \, the effect of stratification is slight; for example,
provided that z can stiIl be large enough compared with V/UT' the logarithmic
profile will be observed. This is therefore known as the forced convection layer. As
z is inereased, the changes due to stratification become more important and finally
dominant.
With destabilizing stratification, this me ans that when z ~ -L buoyancy is the
main source of turbulence generation. This is therefore a free convection layer. A
prineipal feature of the flow as a whole may then be the following [272] : large
eddies of the eruption type originate elose to the wall as described in Section 22.6;
as they move away from the wall they become increasingly influenced by
buoyancy, until, at large heights, they resemble the free convection plumes
described in Section 4.3.
With stabilizing stratification the primary effect of the temperature variations
again appears far from the wall, but, because this is nowa damping effect, it can
alter the whole structure of the flow. If the outer part of a boundary l~er ceases
to be turbulent, it loses its Reynolds stress (note that, because u 2 and w 2 decrease,
the Reynolds stress falls much more rapidly than the correlation function shown in
Fig. 22.25). But it is this stress that keeps the fluid eloser to the wall movingo Thus
if the outer turbulence is damped this fluid is much slowed down by viscous
friction at the wall. In terms of the above analysis, UT is reduced and the 'forced
conveetion' layer is affected.

t Notation: it is conventiona! to denote the direction normal to the wall by y except when this
is explicitly vertical, and then to denote it by Z.
:j:It is a moot point whether the constant here should be the same as that in equation (22.27),
but the two are found to be sufficiently close that we need not make any distinction for the
present purpose.
290 Physical Fluid Dynamics

U
Uo

OL-________________ ~ __________________ ~ ________ ~

o 2
z (cm)

Figuu 22.26 Velocity profiles of boundary layer below heated surface at 24 cm (x) and
74 cm (e) downstream from start ofheated section. Data from Ref. (189).

Experimental observations of this effeet were made in the boundary layer

under the top wall of a wind-tunnel, one seetion of whieh was heated [189] .
Turbulenee established before the heated seetion deeayed in the stably stratified
region. Figure 22.26 shows velocity proftles at two distanees downstream from the
start of the heated seetion. The deeeleration of fluid elose to the wall is apparent.
It is this effeet which is responsible for the eommon observation of the wind
'dropping' on a cIear evening as the ground eools by radiation.

22.9 Revene transition

In the last seetion, we saw an example of the reversion of turbulent motion to

laminaro This process, known as reverse transition or relaminarization, ean oeeur in
a variety of eonfigurations. In addition to a boundary layer entering a region of
stable stratifieation, flows in whieh it has been observed inelude boundary layers
that enter a region of strongly favourable pressure gradient [53] ; pipe and ehannel
flows in which the Reynolds number is redueed by a ehange of geometry [52,230];
and pipe flow in whieh the viseosity is inereased and/or the density is deereased by
heat transfer [56].
It has reeently beeome apparent that reverse transition involves its own
eharaeteristie meehanisms; it is not just a matter of the viseous dissipation beeoming
The Structure of Turbulent Flows 291
0.15 ,.----.-----,.----.,,----..,----.----,-----,

0.10 0.4

[iV

(;?)'h/U o1 (t?~)y,
X 0.3

0.05
o~ 0.2
X~
X
0.1

0 0
0 20 40 60 80 100 120 140
x/D

Figure 22.27 Variation of longitudinal intensity (x) and normalized Reynolds stress (0) with
distanee downstream during reverse transition in ehannel flow. Both quantities plotted are
maxirna with respeet to position aeross ehannel. (Re2a = 1730). Data from Ref. [52).

too large for the turbulent energy production to sustain the turbulence. As an
example, we may consider observations [52] made in a channel of which the
dimensions gradually change with distance downstream from 12.7 x 76 mm to
12.7 x 228 mm; the consequent drop in speed reduces the Reynolds number (based
on the smaller dimension) from a value at which the flow is turbulent to one at
which reverse transition occurs. Figure 22.27 shows maximum values, with respeet
to position across the channel, of the intensity oflongitudinal velocity fluctuations
and of the correlation coeffieient that provides the Reynolds stress, as funetions of
the distance downstream from the end of the cross-section change. One sees that, as
the turbulence decays, the velocity components become less correlated. The structure
of the turbulence is thus changing in a way that produces a faster approach to laminar
motion.
Various suggestions have been made about the dynamical causes of reverse
transition, of which we may mention two. One, applying to all wall flows, is that
the formation of the large eddies erupting from the viscous sub-Iayer (Section 22.6)
is inhibited [222]. The other, applying to boundary layers in favourable pressure
gradients, is that the region of intermittent turbulence (Section 22.3) becomes so
large that it extends to the wall [104].
23
Experimental Methods

23.1 General aspects of experimental fluid dynamics

We do not have space for a full description of all the experimental techniques used
in obtaining the results discussed in this book. This chapter can give only a general
survey, intended to place the various methods in some perspective. The techniques
appropriate to different branches of fluid mechanics - and to different experiments
within one branch - are diverse, and decisions about those to be used in a particular
project require detailed consideration of the successes and failures of those used in
previous related projeets. One must add to this that the limitations to an experiment
often lie in the performance of transducers, and the experimentalist should always
be on the alert for new possibilities.
The purpose of an experiment in fluid mechanics may range from direet verifi-
cation of a theory (as in Figs. 11.2, 16.l2 and 18.8) to a general exploration of the
phenomena that occur in a given situation. Most of the experiments that have
contributed to the ideas described in this book fall somewhere between these two
extremes - although the proportions contributed by theory and by experiment to
the final story are very variable. Most often one is dealing with a situation for which
mathematical difficulties preclude a full theory, but in which it is stilI useful to refer
to the equations of motion in deciding what measurements to make and how to-
interpret the results. Studies of the energy balance in turbulent flow, as in
Figs. 22.l0 and 22.18, [Link] straightforward examp1e of this. Even in topics for
which there is a wholly adequate theory, primari1y exp1oratory experiments may
have p1ayed an equally important role. For example, hydrodynamic stability is now
one of the more highly developed theoretieal branches of the subject - Chapter 18 -
but the need for this type of treatment of the equations of motion wou1d not have
been apparent without experimental observations of instabilities.
The design of an experiment involves careful attention to the requirements of
dynamical similarity. One must ask what range of phenomena one wishes to study
and thus what values the relevant non-dimensional parameters should have. Then
one must consider how these can be achieved, or, if they fall outside the practicable
range, what departures are least unacceptable.
In this process of design, there are three questions that frequently arise: (i) will
the work consist primarily of quantitative measurements or of observations of flow
pattems? (ü) will the work be done with an existing installation or with specially
built apparatus? (iii) what fluid will be used?
The distinction implied by the first question - transducing versus flow visualiza-
tion - is not complete. One can, for example, measure velocities by timing the
movement of dye. However, it does provide a useful general classification of experi-
Experimental Methods 293
ments. The decision depends partly on the previous state of knowledge of the topic
under investigation -- one is most likely to opt for flow visualization in a preliminary
exploration - and partly on the efficacy of available transducers in the particular
situation. However, the two approaches are often complementary. No purely
qualitative study is likely to answer all questions about a flow. On the other hand,
measurements are. often difficult to interpret without the assistance of flow visuali-
zation. Within this book, rather frequent use of the results of visualization experi-
ments has been made because the photographs provide ready illustrations of the
flows under discussion. The reader should not conelude that experimental fluid
mechanics is primarily a matter of 'look and see'; one always aims to express results
as quantitatively- as possible.
There is a third category of experiment, quantitative but not involving detailed
probing of the flow. In these experiments, some buIk quantity associated with the
flow is measured. Examples are the mass flux in pipe flow (Sections 2.3 and 2.7),
the torques acting on the cylinders in rotating Couette flow (Sections 9.3 and 17.5),
and the heat transfer in convection experiments (Section 4.2 - particularly Fig.
4.4 - and Sections 13.4 and 14.5). The experimental methods involved depend very
much on the particular experiment, so they will not be discussed further in this
chapter.
Some fluid mechanics laboratories consist mainly of standard flow systems into
which different experiments can be introduced. Others consist mainly of equipment
that has been built for particular experiments. This depends primarily on the branch
of fluid mechanics being studied. The investigation of the flow past obstaeles or of
boundary layers requires a uniform flow with minimal velocity fluctuations. To
obtain this is of itself a complicated matter. Hence, such experiments are normally
carried out in a laboratory permanendy equipped with a wind-tunnel (the name for
any system providing a working air stream), a water flume or channel (similar
systems with water), or a towing tank (a large tank of stationary water through
which an obstaele can be moved). A few of the many examples of results obtained
with such equipment are Figs. 11.2, 11.7, 19.5 and 22.8. Readers interested in more
details of such techniques are referred to Refs. [21,26]. In contrast, many experi-
ments, for example in convection, are difficult to fit into standard systems, and
every experiment involves the constmction of a special piece of apparatus.
Examples of observations made in such experiments inelude Figs. 4.5, 14.5, 15.2
and 18.5.
The choice of working fluid, unIess dictated by the available standard equipment,
is usually influenced by two considerations, the achievement of the desired values of
the governing non-dimensional parameters and the performance of flow transducers.
Whenever possible, of course, either air or water is used. The choice between these
two often depends on the type of experiment; more successful techniques for
velocity measurement have been developed for air than for water, whilst flow
visualization is generally more successful in water than in air. Other considerations
may, however, enter. For example, one might face a situation in which the fact that
water more readily gave the desired values of the governing parameters (its lower
kinematic viscosity is often advantageous in this respect) had to be set against the
practical difficulties of containing it in an arrangement with movable probes. Other
fluids are sometimes used because they give better values of the governing para-
meters of because they have properties particularly appropriate to an experiment.
For example, silicone oils are widely used in convection experiments: they can be
294 Physical Fluid Dynamics
obtained with different viseosities, but otherwise similar properties, which provides
a eonvenient method of varying the Prandt! number; and the viseosity varies with
temperature mueh less than for many fluids.

23.2 Veloeity measurement

Obviously the most important quantity to be measured in most flows is the fluid
velocity. Here we shalllook at the prineiples by which various methods of
measuring veloeity work; the reader interested in the arrangements of a full working
system and the proeedures for operating it should follow up the references. The
general name for any instrument for measuring fluid veloeity is 'anemometer'.
Two instruments - the Pitot tube and the hot-wire anemometer - are used
widely in many different type s of experiment. A third - the laser-Doppler
anemometer - seems likely to move into this eategory in the eoming years. The
remaining teehniques have not given rise to general purpose instruments but have
proved useful in particular experiments.
The Pitot tube [214] is illustrated in Fig. 23.1. The inner tube with a hole at
the nose, S, of the instrument is entirely sealed from the outer tube. Several holes
(typieally five), of whieh two are shown in the figure, round the periphery of the
tube alllead into the outer annulus. The theory of the operation of a Pitot tube is
contained essentially in equation (10.15), Bernoulli's equation applied to the
streamline that ends at the forward stagnation point of an obstacle plaeed in a
stream. If the Pitot tube points into the flow, the pressure Ps will oeeur at the
point S and will be measured by a manometer eonneeted to oudet I (the mano-
meter must, of eourse, block the tube, so that there is no flow through it). Caleula-
tion of the speed, u, from the relationship, Ps = 1hpu 2 + p, requires information
about the pressure p. This is provided by the peripheral holes, P. These are so me
distanee downstream from the nose so as to be beyond the region of pressure
variation. Sinee the pressure differenee aeross the boundary layer is negligible, the
pressure at these holes is the statie pressure p. The average pressure from several
holes is observed, as the pressure at a single hole would be more sensitive to mis-
alignment of the tube than the ps-reading. A manometer eonneeted between
oudets I and 2 reads the pressure 1hpu 2 , thus providing adireet measurement of u.
Provided that the Pitot tube is small eompared with the length seale of variations of

( p
s

Figure 23.1 A Pitot tube.

Experimental Methods 295

the flow, u and pean be interpreted as the speed and pressure that would exist at
the position of the Pitot tube in its absenee.
Pitot tubes have been most widely used in air, where they ean measure speeds
from about 1 m S-1 upwards. With greater diffieulty they ean be used in water to
measure speeds down to about 3 cm S-1 (Ref. [204]). Sinee one applies inviscid
theory to determine the velocity, it is essential for the Reynolds number of a
Pitot tube to be high. However, exeept for speeially made tiny instruments, this
requirement is fulfilled at all measurable speeds. The main limitations to the use of
Pitot tubes are their size and the slowness of their response; they eannot be used in
flows of very smalllength seale, nor to measure rapid velocity fluetuations. In most
turbulent flows, for example, only the mean velocity ean be measured with a Pitot
tube.
The principle of operation of a hot-wire anemometer is virtually indieated by its
name. An eleetrieally heated wire is eooled by the flow, the rate of eooling depen-
ding on the velocity. In the simplest mode of operation, the eurrent through the
wire is maintained eonstant; its temperature and thus its resistanee, measured by
the voltage aeross the wire, depend on the veloeity. In an alternative, widely used,
mode, a feed-baek eireuit maintains the wire at a eonstant resistanee and so tempera-
ture; the eurrent needed to do this is a measure of the fluid velocity.
Hot-wire anemometers have been most widely and sueeessfully used in gas
flows. In general, hot-wires are more sensitive at low speeds than high; exeept that
if the speed is too low, free eonveetion heat transfer takes over from foreed convee-
tion, making the eooling insensitive to veloeity. Hot-wire anemometers ean be used
in air readUy down to about 30 cm S-1 ; below this, measurements are more diffieult
although not impossible.
In reeent years the use of these instruments in liquids has been mueh developed.
Here particular use has been made of the hot-film anemometer, a device working on
the same principle but geometrically different from the hot-wire. The heated
element consists of a thin metallic film on the surface of a wedge-shaped thermally
and eleetrically insulating probe.
Hot-wire (and hot-film) anemometers operate best in just those conditions where
Pitot tubes faiI. They are small and rapidly responding. They have thus beeome the
principal instruments for studying fluctuating flows, in particular the phenomena
of transition and turbulence. They have the additional advantage of giving the
information in eleetrical form, which can be processed electronically to give
quantities such as intensities, correlations, and spectra. Neady all the measurements
presented in Chapters 19 to 22 were made with hot-wire anemometers.
Although a hot-wire anemometer is simple in principle, its actual use is a matter
of some complexity. A large body of information [29,38,197] - and a certain
amount of 'folklore' - has grown up around hot-wire anemometry. This covers
topics such as: the calibration of hot -wires (this is always necessary - hot-wires are
not absolute instruments); the geometrical combinations of hot-wires needed for
me asu ring different components of velocity fluctuations (a particulady important
arrangement being two wires in the form of an X, as shown in Fig. 23.2); the
particular problems in achieving accuracy that arise when the velocity fluctuations
are not small compared with the mean velocity; the problems associated with
velocity measurements in the presence of temperature variations, to which a hot-wire
probe is also sensitive; and much else.
The laser-Doppler anemometer measures velocity by measuring the Doppler
296 Physical Fluid Dynamics

L
Mean velocity Velocity
fluctuation
components measured

Figure 23.2 An 'X-wire', which, when arranged so that the mean velocity is in the platle of
the X, can be used to measure both components of velocity fluctua tion in that plane.

shift of light seattered within the moving fluid. With the advent of lasers, sufficiently
monoehromatic light beams are available for this effeet to be measurable. The
seattering eentres are the tiny partides of dust that are present in any liquid or gas;
these are small enough that they are always moving effeetively with the instan-
taneous fluid velocity.
Figure 23.3 shows sehematieally the prineiple of operation of a laser-Doppler
anemometer. (The full optieal systems are, of eourse, mueh more eomplex than
this, and take a variety of forms [196].) The incident and seattereri beams are
inelined to one another so that the point of interseetion loeates the point at which
the velocity is being measured. The seattered beam interferes at the photomultiplier
with an attenuated beam direet from the laser to give beats, the frequeney of which
is a measure of the Doppler shift.
Like hot-wire anemometers, laser-Doppler anemometers have the advantage that
they respond to rapid fluetuations in the velocity and thus ean be used to study the
details of transitional and turbulent flows. They have the additional advantage that
there is no probe to disturb the flow. On the other hand, in some situations there
are severe praetieal diffieulties in mounting a preeision optieal system around a flow.
(Also, laser-Doppler anemometers are costly.)
Almost every effeet of fluid motion must at some time have been tried as a
means of measuring velocities. We now look briefly at various deviees that have
proved useful on oeeasion. The list does not pretend to be exhaustive.
Mechanieal systems have the ad van tage of simplieity. They are widely used on

Light attenuation

silvered silvered
mirror mirror

-
F low

Figure 23.3 Schematic diagram of laser-Doppler anemometer.

Experimental Methods 297
the large scale, a famiHar example being the cup anemometer used in everyday
meteorological applications. They are more difficult to adapt to the small scale of
laboratory work. However, two methods that have been used are: observation
through a telescope of the deflection of the free end of a cantilevered quartz
fibre [148] ; and mounting a vane on the extended arm of amoving coil millia-
meter [124] . Such arrangements can give high accuracy in low-speed laminar flows.
They are much less satisfactory, but still sometimes useful, in turbulent flows [27 S] .
The frequency of the vortex street behind a cylinder (Section 3.3) is a measure
of the velocity. This is particularly convenient as ameans of calibrating a wind-
tunnel or water-channel below the Pitot tube range (once the Reynolds number -
Strouhal number relationship has been established). The technique has, however,
also been used in a speeial probe [290].
In liquids, use may be made of the 'electrokinetic' or 'strearning potential'
method, although, so far, this has been tried only in water [281]. At the interface
between a liquid and a solid, as at the interface between two solids, a double layer
is formed, a layer of positive charge on one side and a layer of negative on the
other. If the liquid is moving, its charge may be advected introdueing measurable
potential differences depending on the velocity.
Heat-transfer devices may take other forms than the hot-wire anemometer. For
example, the faster a fluid is moving the smaller the temperature rise produced by
a given rate of heat input. This fact is used in the arrangement shown in Fig. 23.4
[261] . The middle wire is heated at a fixed rate. The temperature difference be-
tween the outer wires is then, over certain ranges, inversely proportional to the
velocity component in the direction shown. The wires are made long and set in
different direetions, so that this remains true in the presence of other veloeity com-
ponents. This device can operate in flows with their own temperature variations.
Finally, it should be emphasized that methods depending on the simple timing
of the passage of marked fluid from one point to another are not to be seomed.
Many flow visualization experiments can be made quantitative in this way. Heat is
sometimes used instead of a visible marker. A good way of determining the speed of
a wind-tunnel, when this is low, is to put an altemating current through a fine wire
stretched across the tunnel and to measure the wave-Iength of the resulting tempera-
ture variations. A probe - a thermocouple, for instance - is traversed along the
tunnel, and the phase variations between the input to the heated wire and the out-
put from the probe are observed. This is tedious, but it does provide an absolute
calibration, below the Pitot tube range, on which other devices can be based. Prohes
for local veloeity measurement, based on the timing of the passage of a heat pulse
from a source wire to a detector wire, have also been developed [67,267] .

.. <.- -------:===
-e·::------:.-==
..

Figure 23.4 Two perpendicular views of the wire arrangement in a temperature rise
anemometer.
298 Physical Fluid Dynamics

23.3 Pressure and temperature measurement

We need not consider at any length the methods used for measuring flow parameters
other than the velocity, since these methods do not differ signifkantly from those
used in other branches of physics.
The pressure at a point on a wall may be measured through a small hole, similar
to the ones in a Pitot tube, linked to a manometer. Alternatively, a small diaphragm
may be set in the wall, its movement changing a capacity or compressing a piezo-
electric crystal. The diaphragm method has the advantage of much more rapid
response, so it can be used to measure the pressure fluctuations at a bounding wall
of a turbulent flow.
The measurement of pressure in the interior of a flow has been considered above
in connection with Pitot tubes. There is at present no satisfactory device for
measuring rapid pressure fluctuations within a flow.
An important aspect of pressure measurements, as weIl as of velocity measure-
ments with Pitot tubes, is the design of sensitive manometers. The various type s are
described in Ref. [26].
It should be added that, whilst pressure measurements are often very useful in
wind-tunnel and similar studies, the pressure variations occurring in many experi-
ments, for example in the field of free conveetion, are too small to be measured.
In experiments on conveetion, particulady free conveetion, measurements of the
temperature field [148] are usually both easier and more accurate than measure-
ments of the velocity field. The former, therefore, often provide the main source
of information about the structure of the flow. However, we can consider the
methods used very briefly, since they are, at least in principle, wholly straight-
forward. The transducers most widely used are resistance thermometers and thermo-
couples [63]. A resistance thermometer often conveniently takes the form of a
'hot-wire anemometer' with a very low current through it. Thermistor beads are
also often used as the sensitive elements of resistance thermometers. The junction
of a thermocouple can also be made very small, so all these deviees can be made to
respond rapidly.
Optieal systems can also sometimes give quantitative information about tempera-
ture fields; however, these will be considered under the heading of flow visualiza-
tion.

23.4 Flowvisualization [32,80,291]

The variety of methods of flow visualization is illustrated by the photographs

throughout the book. Here we colleet the different methods into a systematie list,
referring back to other chapters for examples. Again no claim is made that the list
is complete.
Conceptually the simplest procedure is the introduction of smoke or dye. The
introduction of smoke into a gas flow without unduly disturbing it is rather difficult
except in wind-tunnels specially designed for the purpose. Also, the time scale is
often too rapid for the eye to follow the phenomena; high-speed photography or
some other special technique may be needed. Hence, dye-in-liquid experiments are
often preferred. Nevertheless very good results can be obtained with smoke [26],
as is illustrated by Figs. 14.6, 16.1 and 19.6. A system widely used in smoke
Experimental Methods 299

tunnels produces a smoke of fine droplets of paraffin, by evaporating the liquid

and allowing it to recondense in the flow. A particularly effective dense white
smoke is produced by exposing titanium tetrachloride to the atmosphere, but this
is both toxic and corrosiveo A convenient way of producing 'instant smoke' for
simple exploratory experiments is the reaction between hydrochloric acid and
ammonia vapours, but again its extensive use has its dangers.
Many of the photographs in this book illustrate the use of dye; a selection,
illustrating different dyes and different ways of using them, consists of Figs. 3.4,
3.5, 3.9, 12.6, 15.4, 17.11, 19.5 and 22.7. Convenient and effective dyes for intro-
ducing into water flows inelude potassium permanganate, gentian violet, and
methyl blue. The first has the advantage of producing particularly strong contrast,
but at the cost of a sometimes inconveniently large density increaseo
As alternatives to introducing dye, there are several ways in which marker can be
produced electrically - the hydrogen bubble technique, the tellurium technique,
the electrolytic precipitation technique, and the pH technique. The first works best
at high speeds, such as those in many water channel experiments; the second and
third at intermediate speeds; and the last at the low speeds likely to be encountered
in convection and rotating fluid experiments.
In the hydrogen bubble method a wire stretched across the flow is the cathode
of an electrolytic system, the anode being, for example, a wall of the channel.
Typically 10 volts are needed with tap water, larger voltages with deionized water.
Provided the wire is_very fine (typically 10-100 J,Lm), the hydrogen bubbles pro-
duced at it are small and numerous enough to move with the flow and give the
appearance of a white dye. By insulating portions of the wire and/or pulsing the
voltage, patches of dye can be produced, giving most effective pictures like those in
Figs. 22.1 and 22.20.
A cathodemade of, or coated with, metallic tellurium releases a dense brown dye,
which is effective for flow visualization [295] . Care is needed to avoid toxic
effects.
In a similar way a white colloidal eloud is produced at an anode made of certain
materials [247] . Those containing tin (in water containing sodium carbonate) are
particularly effective; solder is thus a convenient material. This method - known as
the e1ectrolytic precipitation method - requires rather larger voltages than the
others, typically 30-50 V.
The last of these electrical methods [55] makes use of the fact that there is a
change in pH (the measure of acidity or alkalinity) in an electrolyte in the vicinity
of an electrode. Hence, if the working fluid contains an indicator and is titrated to
be elose to the end-point of that indicatOf, the application of a voltage between a
fine wire and so me other point in the flow can produce a local colour change at the
wire. This method has two marke d advantages: the production of the dye produces
no density change of displacement of the flow; and the dye gradually reverts to its
previous state, so the system can be run continuously without becoming filled with
dye. It has the disadvantage that it can be used only in elosed systems, not in flows
where fluid is continuously entering and leaving. The most usual indicator is thymol
blue; this changes from amber (neutral) to dark blue (basic) at a cathode and so
gives elearly visible marker. Figures 4.14 and 15.2 were obtained in this way. With
both this method and the tellurium method, it is important that hydrogen bubbles
should not be produced. For both methods, voltages in the range 1-10 Vare
satisfactory, and this gives no trouble in deionized water.
300 Physical Fluid Dynamics
Other flow visualization experiments make use of suspended particles (e.g.
Figs. 15.17,16.8 and 17.5). Of the various particles available, polystyrene beads
are partieularly useful in water, as their density is only slightly greater than that of
water. In silleone oils and other highly viseous fluids, aluminium powder has been
very widely used for flow visualization ('silver' paint is a eonvenient souree of
suffieiently fine particles). The alurninium particles are small flakes whieh align
with the shear in a flow; when light enters from one direction and is viewed from
another, bright and dark patehes are seen depending on the loeal orientation.
In aD flow visualization experiments it is neeessary to give some thought as to
what features of the flow are shown up. Smoke and dye experiments usuaDy show
partiele paths or streaklines (Seetion 6.1). The eleetrieal methods allow other
possibillties; one ean, for example, release dye all along a wire at a given instant
and obtain information from its subsequent locus. Sirnilarly, suspended particles
may be used in various ways. Photographing them for an interval gives portions of
particle paths, whieh, if short enough, ean be used to synthesize a streamiine pattem
even in unsteady flow (e.g. Fig. 3.7). Other teehniques ean give overall information
about the strueture of the flow; this is partieularly true of the aluminium powder
method (e.g. Figs. 4.5,14.9 and 15.12) and Fig. 17.4 shows polystyrene beads used
in a speeial way for this purpose.
The last group of flow visualization methods to be considered consists of optieal
techniques making use of refraetive index variations in the fluid [129]. Such varia-
tions oceur as aresult of density variations associated with a temperature or eon-
eentration field. They are thus partieularly appropriate for studies of eonveetion
and stratified flow. However, even in nominaDy eonstant density flows it may be
possible to introduee density differenees whieh are krge enough to all ow the use of
one of these teehniques whilst keeping the intemal Froude number high enough
for the flow to be unaffeeted.
In the shadowgraph method, parallellight enters the fluid and is defleeted where
there are refraetive index variations. If the second spatial derivative of the refraetive
index is non-zero the amount of defleetion varies giving a pattem of bright and dark
regions related to the flow strueture. Figures 4.16, 14.1 and 14.4 are shadowgraphs.
In the sehlieren method, parallellight is again used, but is brought to.a foeus
after passing through the fluid. A knife-edge or other stop at this foeus bloeks off
some of the light. If the light has been defleeted in the fluid the amount passing this
stop will be inereased or decreased. Changes in intensity thus relate to the first
spatial derivative of the refraetive index. By having an optieal arrangement in whieh
the stop is at a eonjugate point of the souree whilst the observing sereen is at a eon-
jugate point of the fluid, one ean obtain an image of the flow pattem with these
intensity variations. There are many variations in the details of the arrangement
[129]. Figures 16.11, 16.13,22.3 and 24.20 are sehlieren photographs. (It should
be noted that the name 'sehlieren' is oeeasionaIly used to refer to optieal methods
in general rather than this specifie one, whieh is then known as the Töpler-
sehlieren method.)
In an interferometer [251], light that has passed through the fluid interferes
with light from the same souree that has not passed through the fluid. Whether this
interferenee is eonstruetive or destruetive depends on the optieal path differenee,
and so the pattem of bright and dark bands refleets the pattem of isotherms in the
fluid. Again there are various optieal systems. Figure 18.8 shows two interferograms.
All these teehniques work best in two-dimensional eonfigurations, as the observed
pattem is the integrated effeet of the passage of the light through the fluid.
24 Practical Situations

24.1 Introduction

The main body of this book has been concerned with developing a basic under-
standing of the phenomena of fluid motion. Applications have been ignored, apart
from passing references to illustrate a particular point. One does not capture the
full flavour of the subject without some emphasis on the fact that moving fluids
occur in a wide variety of practical situations. Some of these applications have had
a profound effect on the direetions of advance of the basic studies.
It would not be possible in the space - nor perhaps very interesting - to give a
systematic survey of applied fluid mechanics. lnstead, we look at, and discuss
briefly, a selection of particular topies. The selection has been made with two
purposes in mind: to illustrate the variety of branches of applied science in which
fluid dynarnics arises; and to show applications of different topics in the main body
of the book. The applications discussed are thus not intended to be those of greatest
importance or topicality. They are, however, all topics of current or reeent research.

24.2 Cloud pattems

The very varied patterns formed by clouds refleet the variety of dynamical processes
that occur in the atmosphere, and are often a useful immediate indication of these
processes [201,227]. Clouds form when moist air is cooled so that the saturation
vapour pressure faUs below the actual vapour pressure. Although cooling can occur
in a variety of ways, the most common is the cooling associated with expansion as
air rises; that is cooling associated with the adiabatic temperature gradient
(Section 14.3). One may thus observe situations in which cloud is forming in rising
air and evaporating in descending air to give a pattem of cloudy and clear patehes
related to the flow.
Of the many possible illustrations of this, we choose 'cloud streets'. One can
fairly frequenUy observe clouds in long parallellines, quite evenIy spaced. These
are usually either convection rolls aligned by the mean flow as considered in
Section 16.1 (ef. Fig. 16.1), or billows generated by shear instability in stably
stratified air as considered in Section 17.6 (ef. Fig. 17.12). The two processes can
be distinguished, even if the stratification is unknown, by the orientation of the
cloud lines with respeet to the wind direction; convection cells are aligned along the
wind, billows across it. Lee waves (Section 16.3) can also cause such patterns if the
ridge producing them is sufficiently long and straight.
Figure 24.1 is a satellite photograph of aligned clouds. (It was taken over the
south-east USA; the coastline of Georgia can be seen, the clouds being over land.)
302 Physical Fluid Dynamics

The lines are at a height of about 1 km and have a spacing of about 3 km, so the
pattem would not be seen by an observer on the ground. Such large-scale pattems,
the occurrence of which has been appreciated only since satellite observations have
been available, probably involve their own distinctive dynamical processes. It has
been suggested [74] that they arise from instability of the atmospheric Ekman
layer; i.e. they are the counterpart of the laboratory patterns in Fig. 15.1 O.

Figure 24.1 Apollo photo of Georgia eoast (10 a.m. local time, 4 April1968) from height of
200 km. From Brown, R. A., 'A Secondary Flow Model for the Planetary Boundary Layer',
J. Atmas. Sci., 27,742 (1970). By permission of the American Meteorological Society.
Practical Situatians 303

However, eonveetive instability also plays a role [151,158], so the overall effeet
may be an interaetion between the types ofmotion represented by Figs. 15.10 and
16.l.

24.3 Waves in the atmospheric circulation

The large-seale cireulittion of the atmosphere frequently involves wave motions.

These are prominent at mid-latitudes in both the northem and southem hemi-
spheres. Figure 24.2 illustrates this. It shows the trajeetory of a balloon designed
to float in the atmosphere at positions of eonstant air density; that is effeetively at
a eonstant height, of, in this ease, about 12 km. The balloon was released in New
Zealand, and its position observed through radio signals for 102 days. Wavelike
oseillations in lati tud e as it travels round the earth in the prevailing westerly winds
are apparent. SimiIar features in the northem hemisphere ean be seen in the isobar
pattems shown in Fig. 20.1, although the development of strong eyclones within
the wave pattem makes this situation rather eomplex.
The longest waves (three or four wavelengths round the Earth) are dynamically
simiIar to Rossby waves [202] . The theory of Rossby waves for a spherieal shell is
slightly different from that given in Seetion 15.7, but the faet that they may oceur
may be understood through the faet that the depth of the atmosphere paralleI to the
axis of rotation varies with latitude. Sometimes the waves form a fixed pattem
arising from disturbanee of the general eireulation by mountains; i.e. they are a
natural eounterpart of the flow illustrated by Fig. 15.15. Waves progressing relative
to the earth may also be observed; their origin is less apparent.
Shorter waves are probably mo re often associated with temperature variations
between pole and equator - they are baroelinic waves rather than barotropie. These
temperature variations are responsible for the general westerly flow on whieh the
waves are superimposed; the Coriolis foree tums the meridional cireulation that
would arise on a non-rotating earth. This proeess is modelle d by the laboratory
experiments deseribed in Seetion 15.8. The generation of the waves can now be
understood as an instability. Despite the differenee in geometry, the atmospheric
waves may be compared [108] quite closely with some of those observed in the
Iaboratory (Fig. 15.17). The non-linear development mentioned in Section 15.8 has
its counterpart in the development of the jet-stream, a simiIar narrow region in
which much of the momentum of the atmospheric circulation is concentrated.

24.4 Continental drift and convection in the Earth's mantle

The important role of fluid mechanics in meteorology and oceanography is obvious.

Its role in the geophysics of the Earth's interior is somewhat le ss famiIar, and we
consider here an example of this.
The main divisions of the Earth (discovered through the reflection and refraction
of elastic waves originating in earthquakes) are a solid crust (the top few tens of
kilometres), a solid mantle (extending from the bottom of the erust to aradius a
little over one half the Earth's radius), a liquid eore, and, probably, a solid inner
core. One would expect fluid dynamical problems to arise in the core; motions
 •

Figure 24.2 Trajectory over

102 days of a ba1loon released
from New Zealand and circulat-
n"ü Of LAUNCH ing at height of about 12 km.
From Ref. [174].
Figure 24.3 Reeonstruetion of eontinental distribution around 200 million years ago. (At the bottom Madagasear and India lie between Afriea
and Antaretica, and a portion of Australia is seen to the right of Antaretica.) After A. G. Smith, from Understanding the Earth, Artemis Press Ltd.,
1972.
306 Physical Fluid Dynamics
there generate the Earth's magnetic field. More surprisingly, there are fluid dynami-
cal problems associated with the mantle.
It is known that soüds under stress undergo not only an elastic deformation but
also a continuous creep process. If this continues for a very long time, it may resuIt
in a complete change of shape of the soüd, which is then behaving as a fluid. Thus,
over time scales of geological development, it may be appropriate to treat the
mantle as a fluid, and indeed possibly as a Newtonian fluid.
It is also now widely accepted that the Earth's crust is undergoing continuous
changes; that, for example, the distribution of the continents over the surface of
the Earth has changed [252]. Figure 24.3 shows a suggested reconstruction of this
distribution around 200 million years ago [110] . The relative motions of different
portions of the crust occur in narrow linear zones, the regions in between moving as
rigid plates. This has led to the topic being given the name 'plate tectonics'. The plate
movements are responsible not only for continental drift, but also for the formation
of mountain ranges, rift valleys, mid-ocean ridges, and ocean trenches and for
earthquakes and voleanic activity.
Various suggestions have been made about the cause of the movements; the
majority involve some aspect of thermaI convection [273,279]. It is a matter of
discussion whether the plates themselves play a basic role in the dynamics or
whether they are driven by a flow occurring in the mantle below. It is the latter
view that would make concept s from this book more directly applicable. Thermal
convection could weU occur in the mantle. Convection due to chemical differences,
rather than temperature differences, is also a possibility, if the separation of heavier
materials to the centre which has led to formation of the core is stiU continuing.
Much of the discussion has been based on our knowledge of Benard convection
(Chapter 4) and on its counterpart with internaI heat generation (Section 17.3).
Simple cellular flow, boundary layer type flow, and quasi-turbulent flow have all
been considered. However, it may be an oversimplification to apply Benard-type
flows to the mantle. The only summary that can be made at present is that it is
highly probable that mantle flow occurs but its details remain very uncertain.
(It is worth noting that mantle convection is an exception to two statements
eIsewhere in this book that certain ideas are rarely relevant to applications. The
exceptions arise because one is concerned with an extremely viscous fluid - the
Prandtl number is around 10 24 • Firstly, compressibility effects - and other non-
Boussinesq effects - do enter marginaUy into the problem. This might seem very
surprising when the velocities are tiny, but it is an example of the fact that at low
Reynolds number relationship (5.61), and not (5.59), is the criterion for neglect of
compressibility. Secondly, the Rayleigh number criterion is often invoked into
discussions ofwhether convection occurs - despite the statement in Section 14.3
about its irrelevance to natural situations.)

24.5 Solar granulation [71, 157]

Observations of the Sun's photosphere - the thin layer elose to its surface from
which the visible radiation comes - show a granular structure, as illustrated by
Fig. 24.4. Discrete bright patches, often polygonal in shape, are separated by a
network of narrower dark regions. The patches vary in size, but are typically
103 km across, so that the re are of the order of 106 granules over the surface of the
Practical Situations 307

Figure 24.4 Solar granulation. (Photo obtained by Project Stratoscope of Princeton

University sponsored by National Science Foundation, National Aeronautics and Space
Administration, and Office of Naval Research.) From Bahng, J. and Schwarzschild, M. (1961)
Astrophys. J. 134,312, published by the University of Chicago Press. (Copyright 1961 by the
University of Chicago.)

Sun. The pattem is a continuously changing one, the lifetime of an individual

granule being typically ten minutes. Doppler shift measurements of velocities
indicate that the bright (hot) regions are ascending and the darker (colder) regions
are descending.
The pattem is thus almost certainly a convection pattem. Its resemblence to
Benard convection (Chapter 4) in the regime in which a cellular structure is retained
although the flow within the cells is turbulent (Figs. 4.11 and 4.12) is noteworthy.
It may be that the motion is basically of this type, although the re is some difficulty
in making the comparison quantitative; the Rayleigh number must be very much
larger (by a factor of perhaps 10 5 ) than ones at which such patterns are observed
in the laboratory.
It is plausible that the convection may be little affected by the Sun's rotation
and by magnetic fields; for example, the Rossby number based on observed velo-
cities and granule sizes is around 10 2 • Of various causes of differences from the
308 Physical Fluid Dynamics

laboratory situation, two are certainly important. Firstly, non-Boussinesq effects

(appendix to Chapter 14) will be very marke d ; the length scale of the convection is
probably deterrnined by the scale height rather than by the depth of the unstable
region [232]. Secondly, the convecting region is overlaid by a stably stratified
region and possibly most of the photosphere is in this; the pattem may be a mani-
festation of a convecting region below rather than in the observed region. A marked
periodieity has been observed in the velocity fluctuations; this is presumably due
to waves in the stably stratified region, stirtlUlated by the convection below.

24.6 Effluent dispersal

The problems ofboth air poUution and water poUution evidently involve fluid
dynamical considerations amongst their many other aspects. In principle, the type
of question posed is usually the same - where is pollutant, emitted from a source,
subsequently found and in what concentration? In practice, the flow situations in
which pollution may occur are very varied, and different branches of fluid dynamics
are involved in answering the question in different cases. We thus consider a
particular example.
This is the discharge of sewage or other effluent into the sea or an estuary. The
design problem is, eVidently, to ensure that the contaminant has become sufficiently
dilute to be unobjectionable before reaching places where it could be harmful - an
aim that has, of course, not always been fulfilled in the past. Sewage has approxi-
mately the same density as fresh water. It is therefore buoyant when discharged
into salt water. The problems involved are thus broadly the same as those associated
with the discharge ofhot water into cold, a matter of practical importance in con-
nection with the retum of cooling water from power stations. Excessive heating can
be biologically damaging (the problem sometimes known as thermal poUution).
Also it is important that the warm water should not retum too directly to the power
station intake with consequent loss of thermodynamic efficiency.
A major aspect of the investigation of any system or proposed system is the
deterrnination of the pre-existing flow pattems, due, for example, to tides or ocean
currents. Field observations are usually supplemented by experiments with models
of the site; the model may be either a physical one in a laboratory or anumerieal
one in a computer. To illustrate laboratory modelling, Fig. 24.5 shows amodel of
the Tees estuary on the north-east coast of England, and Fig. 24.6 shows a simula-
tion of tidal flows in such amodel made during an investigation of the best location
for a sewage outfall [180].
Despite the fact that each site has its own characteristics, there are many fluid
dynamical topics of general relevanee. To illustrate this, we consider the example
of buoyant effluent being emitted horizontally into stationary ambient water [181] .
The principal features of the flow, which may be expected to be turbulent, are
shown schematically in Fig: 24.7. The effluent at first travels horizontally as ajet
(Section 11.8). The action of buoyancy causes this jet to curve upwards, and in
time its motion becomes neady vertieal, its properties then being essentially those
of a thermal plume (Section 14.6). How quickly the upward curving takes place
depends on the intemal Froude number of the initial jet. Throughout this jet/plume
flow the processes of turbulent entrainment and rnixing (Section 22.3) are taking
Pracrical Situations 309

Figure 24.5 Model of Tees estuary at the Hydraulics Research Station, Wallingford.

Figure 24.6 Trajectories of candle floats in Tees-side model at Hydraulics Research Station,
Wallingford.
310 Physical Fluid Dynamics

Figure 24.7 Sketeh of buoyant jet below free surfaee.

place, leading to dilution of the effluent. When the plume reaehes the surfaee, the
buoyant fluid spreads out in a surfaee layer. This is stably stratified and so mixing
is inhibited (Seetion 22.8). Further dilution may oeeur here but is a mueh le ss
efficient process. Thus the eontaminant eoneentration where the effluent rea ehe s
the surfaee is an important quantity - determined by the char ac te risti es of turbu-
lenee in stratified jets and plumes.

24.7 Wind effeets on struetures

During reeent years greatly inereased attention has been given to the effeet of the
wind in the design ofbuildings and other struetures [193,213]. Both the direet
effeet of possible wind damage to the strueture and the indireet effeet of possible
undesirable changes to the surrounding wind pattem due to the presenee of the
strueture are matters of importanee. In this seetion we look at two examples of
the former. One is an eeonomieally trivial example, chosen beeause the fluid
dynamies is straightforward, yet interesting. The other - the eollapse of the
Ferrybridge eooling towers - was a major ineident that illustrates the eomplexity
of a mo re typieal situation. It should be emphasized that, although in both these
examples the trouble was not foreseen, the majority of the work in this field is, of
eourse, eoneerned with prevention not cure; wind-tunnel experiments of the type
mentioned below are now frequently performed at the design stage.
Figure 24.8 shows the 'erowning feature' of the Civie Centre at Neweastle-upon-
Tyne; three 'eastles' are mounted on prongs in a form eorresponding to the city
eoat-of-arms. Sometime after ereetion the two outer easdes developed an oseilla-
tion, moving in antiphase in the plane of the 'trident'; their supporting prongs were
behaving like a tuning fork. Cine-film of the oseillation shows the distanee between
the eastles varying by up to 5 per eent, but larger oseillations almost eertainly
oeeurred. The wind at the time was roughly perpendieular to the plane of the
trident and was not unusually strong (about 15 m S-i). This behaviour resuIted
from resonanee between the frequeney of eddy shedding by the eylindricaI easdes
and the natural frequeney of the tuning fork mode of oseillation. We saw in
Seetion 3.3 that, beeause of eddy shedding, the speetrum of the velocity fluetua-
Practical Situafions 311

Figure 24.8 The crowning feature of the Civic Centre, Newcastle-upon-Tyne.

tions and consequently that of the force fluctuations on the cylinder, has a sharp
peak at a frequency corresponding to a Strouhal number of about 0.2, even at very
high Reynolds numbers. In this case, the behaviour was sufficiently simple that it
could be diagnosed without any experiments, but it illustrates the point that, when
instabilities occur, it is not necessarily the strongest winds that are most dangerous.
It was cured by adding a load inside the castles to change the resonant frequency
and by filling the support tubes with sand to damp the oscillations.
On 1 November 1965, three cooling towers at Ferrybridge, Yorkshire, collapsed
in a gale (Fig. 24.9). The three were members of a group of eight in two rows of
four. All three were in the leeward row, indicating that the interaction between the
towers had been important. Elucidation of the mode of failure was quite a compli-
cated story, but essentially it was a quasi-static tensile failure, Le. a tensile failure that
would have occurred in a steady wind of the highest speed actually reached tempor-
arily, of the shell under the wind loading [75] . There were two main stages in this
elucidation. Firstly, wind-tunnel tests on a model of the complex (Fig. 24.10)
determined pressure distributions over the towers; it was not possible to achieve
the full-scale Reynolds number as is ideally required (Chapter 7), so it was hoped
that the highest possible value would serve. Secondly, the results of these were fed
into membrane theory computations of the stresses in the shell produced by the
combination of the wind pressures and the shell's own weight. The results were
very sensitive to the details of the pressure distribution. Even though the total wind
force on the leeward towers was le ss than that on an isolated tower, the redistribu-
tion of the forees by the presence of the other towers could lead to failure. In this
case, it was not thought that any dynamic effect (such as that in the example above)
312 Physical Fluid Dynamics

played a role; the leeward towers will have experleneed larger velocity fluctuations
due to the presence of the front row (and these might have contributed to the
velocity temporarlly reaching the critical value), but the frequencies were not
approprlate to any resonance. Clearly the subtlety of a situation like this is such
that either very extensive model testing is necessary at the design stage or large
safety margins must be allowed.

Figure 24.9 Collapse of third tower at Ferrybridge. From Ref. (75).

Practical Situations 313

Figure 24.10 Wind-tunnel model of Ferrybridge power-station. From Ref. [75).

24.8 Boundary layer control: vortex generators

We have seen in Chapter Il, partieularly Seetions 11.4 and 11.5, that the effeet of
boundary layer behaviour often extends beyond the boundary layer itself. Whether
a boundary layer is laminar or turbulent, whether and where separation oeeurs may
have a marke d or even eontrolling influenee on the performance of a whole flow
system. Attempts to ehange or eontrol the eharaeter and development of boundary
layers may thus be worthwhile. Such boundary layer eontrol [153] has found
applieation prineipa11y in aeronautieal engineering. There are, however, other
applieations; for example, vortex generators, the deviees to be described below, have
been used to suppress wind-excited oseillation of a bridge [287] (ef. Seetion 24.7).
In aeronauties, the purpose of boundary layer eontrol varies from one situation
to another [153] ; aims have been as diverse as attempting to keep the entire boun-
dary layer on a wing laminar (so as to rninimize the viseous stress) and promoting
transition to turbulence right at the leading edge (to prevent separation). We eannot
here mention all the variety ofmethods used in boundary layer eontrol, so we will
look at one example, the use of vortex generators, illustrated by Fig. 24.11.
This photograph shows the wing of a Trident 1 aireraft. (It was taken during
314 Physical Fluid Dynamics

- '-

Figure 24.11 Trident wing with vortex generators. Photo provided by Hawker-Siddeley
Aviation Ltd.

development flying and one can see the tufts attached to the wing to show the flow
pattem over it.) The vortex generatars are the line of specially shaped protruber-
ances elose to the leading edge.
In general, vortex generators are introduced to prevent or delay separation in a
situation in which the fact that the boundary layer is turbulent is insufficient to do
so. Each vortex generator produces a longitudinal vortex extending downstream
from it - by a process somewhat analogous to the generation of wing-tip vortices
(Fig. 12.8). These enhance the mixing across the boundary layer, bringing rapidly
moving fluid from autside the boundary layer in elose to the wall. This supplements
the mixing due to the turbulence. It is apparent from Fig. 11.3 that the change in
velocity profile will inhibit separation.
VOTtex generators were introduced on the Trident wing (Fig. 24.11) to cape
with a particular problem. The suppression of separation is important du ring take-
off and landing when the leading edge droop is down. The purpose is partly to delay
stall (Section 12.2) but more particularly to avoid instability of the aircraft when
the angle of attack is elose- to the stalling angle. Flight and wind tunnel tests showed
that separation started on one part of the wing and spread progressively in such a
way as to produce a nose up pitching moment which tended to further increase the
angle of attack. Introduction of the vortex generators at the critical parts of the
wing changed this unstable behaviour into a stable one by delaying the outer wing
separation until the flow on the inner wing had separated.
Practical Situalions 315

24.9 Fluidics

A significant technological development of reeent years has been the introduction

of 'fluidic' devices [51, 83, 135, 186] . These serve purposes similar to those
normally served by electronic devices but operate on certain features of fluid
motion. The most successful fluidic elements have been logic elements for use in
control systems, which are essentially small-scale computers intended for a speeific
purpose, such as the control of a machine tool through a programme d sequence
of operations. In such elements the signal takes an off/on form, corresponding to
the binary digits 0 and I; in both types of device considered below, the signal is 'on'
if ajet of air is issuing through a certain orifice. These devices cannot normally
compete in smallness, speed, or cost with electronic devices, but they can operate
in certain circumstances where electronic systems are unsatisfactory: in extremes of
temperature; in the presence of ionizing radiation; where there is a fire risk; and
(surprisingly) in the presence of severe vibration.
We illustrate this with brief descriptions of two types of device, chosen for the
interesting fluid dynamical prineiples on which they operate.
The first type operates through the phenomenon ofwall attachment (the Coanda
effeet - Section 22.7). Consider a system like that in Fig. 22.22, but with the jet
situated symmetrically between two walls. Despite the symmetry, the jet always
attaches to one wall; which depends on the small disturbances present. It can be
located on a particular wall if fluid for entrainment is supplied on the opposite side.
Thus in the arrangement shown schematically in Fig. 24.12, if the re is a signal at
input I, the power jet is located as shown and the re is a signal at output I but not
at output 2. This signal will remain even when the input signal is removed. The jet
will switch to output 2 only when a signal is put on to input 2. Thus the arrange-
me nt constitutes a fluidic bistable, a device which, because it has a memory, is the
building block of storage and counting systems.
Altematively, the power jet can be placed asymmetrically between two walls so
that it always attaches to a particular wall unIess fluid is supplied from that side
(Fig. 24.13). In this arrangement the jet retums to its original position as soon as
the signal is removed. Thus this provides a logic element. If the output is take n

///1///(((

Output 2
rr7/'--:r--r-

Power jet ----

- Output1

Input
(control iet) 2

Figure 24.12 Sehematic arrangement (ef. Fig. 22.22) of a wall-attaehment bistable deviee.
316 Physical Fluid Dynamics

P
uuIUU'UU'UUU'E. 0,,,,,,, (ORI

ower
Jet ~

Output 2 (NORl

Inputs
(centrel jets)
1 2 3

Figure 24.13 Schematic arrangement of a wall-attachment OR/NOR device.

from side 1, it is an OR element, there being an output signal if the re is asignaI on

one or more inputs. If the output is taken from side 2, it is a NOR element.
The diagrams illustrate only the principles involved. The geometrieal details of
aetual deviees are, of eourse, more eomplieated; for example, the asymmetry in an
OR/NOR element may be provided by inlets from the room of different sizes on
the two sides. Figure 24.14 is a photograph of a wall-attaehment bistable (with two
inputs on eaeh side) and an OR/NOR element (with three inputs).
Another type of deviee is the turbulenee NOR element (often known as the
'turbulenee amplifier'), shown sehematically in Fig. 24.15 and photographed, with
cover plate removed, in Fig. 24.l6. Its operation is based on the faet that there is a

Figure 24.14 Wall-attachment devices, bistable and OR/NOR, manufactured by British

Fluidics and ControIs Ltd. of Ilford.
 Practical Situations 317
I nputs

Z?Zi??<Z'==- - ----..c:===
Power jet ___ __Output
??ä??????"
=

Inputs

(a) (b)

Figure 24.15 Sehematic arrangement of NOR unit ('turbulenee amplifier'). (a) No input:
laminar jet; (b) Input: turbulentjet.

wide Reynolds number range over which ajet will remain laminar if sufficiently
undisturbed but will become turbulent in the presence of a large disturbance (ef.
Section 19.2). Thus, in the absence ofa signal on any of the inputjets, the power
jet remains laminar; a laminar jet spreads only slowly and so most of the air leayes
through the output (Fig. 24.15(a)). If the re is a signal on one or more of the inputs,
transition is promoted, and the mu ch more rapid spreading of a turbulent jet results
in so little of the air leaving through the output that it gives no signal (Fig. 24.15(b)).
This is consequently a NOR element, the re bei ng a signal at the output only if there
is no signal at any input. Other logic components can be built up from combinations
of NOR units.
However, this type of device is currently finding its prineipal applications in
interface equipment, providing the primary information for fluid control systems.
For example, the presence of a component on a production line can be detected by
the interruption of an input. The sensitivity of jets to sound (Section 19.2) is useful
in this connection; the input can be an acoustic signal rather than a bulk flow.

Figure 24.16 Turbulenee amplifier (and eover plate) manufaetured by CompAir Maxam Ltd.
of Camborne.
318 Physical Fluid Dynamics
24.10 Undulatory swimming

An obviously important range of biophysical applications of fluid dynamics con-

cerns the propulsion of animaIs - the flight of birds and insects, the swimming of
fish and other creatures. There are various types of propulsion. Here we consider
just one, related to topics in earlier chapters, illustrated by Figs. 24.17 and 24.18.
(Both these pictures are of organisms being used for research into this mode of
swimrning.) Various species of long thin animaIs swim by sending waves along their
bodies. Figure 24.17 shows amphioxus (a primitive fish, also known as a lancelet)
swimrning in this way; it can do so either head-first or tail-first [289]. Other
examples are snakes, leeches, eels, and some species of marine worm. In some cases,
there may be several wavelengths within the body length. Various micro-organisms
propeI themselves by sending a wave down a flagellum attached to the body;
Fig. 24.18 shows the spore of a water mould [178] , and other examples indude
spermatozoa and some types ofbacteria.
The physical principle of such swimming may be understood by considering the
force on each portion of the body or flagellum as if it were in isolation. A cylinder
moving through a fluid in a direction perpendicular to its axis experience s a larger
drag than one moving parallel to its axis. Correspondingly, the drag on a cylinder in
oblique motion is not directly opposite to its velocity. Thus it is possible for an
animal swimming with every portion of the body moving forward nevertheless to
have some portions producing a propulsive foree. We may illustrate this by con-
sidering an inextensible body in a perfectly sinusoidal wave motion [78, 260] .
Figure 24.19 shows two consecutive positions of one wavelength of the body for a
case in which the forward swimming speed is one-third the speed at which the
wave travels backwards relative to the body. The short lines joining corresponding
points of the body indicate approximately the direction in which each portion is

Figure 24.17 Amphioxus swimming, /:lead first from Ieft to right. Photo provided by
J. E. Webb, WestfieId College.
Practical Situations 319

Figure 24.18 Spore of Blastocladiella emersonii. From Ref. [178].

instantaneously moving through the water. The resulting forees on the body may
then be in the direetions indicated by the double-headed arrows. It is seen that
enough of these may have a forward eomponent to overeome the drag in the regions
where they have a baekward component.
An interesting feature of this swimming mode is that it ean be effeetive over all
Reynolds number ranges. For the examples in Figs. 24.17 and 24.18, the Reynolds
number (based on body diameter and swimming speed) is respeetively around 103
and around 10-4. It is clear that the differenee in drag associated with different
direetions of motion is essential to the mechanism. At high Reynolds numbers this
differenee may be large (a eylinder in transverse motion developing a large drag as

Figure 24.19 Conseeutive positionsAB and CD of length of body swimming from right to
left by sending wave from left to right. Single arrows indicate motion of portions of body;
double arrows indicate possible resulting forees on these portions.
320 Physical Fluid Dynamics

discussed in Section 11.5), and so this swimming mode can be quite effective. How-
ever, at high Reynolds numbers, there are many other modes available [165], and
this one is not usually found in species highly adapted to swimming. At low Rey-
nolds numbers, the drag differences are much smaner (because motion of a body in
any direction brings a lot of fluid into motion with it - Section 8.2). They are
sufficient for swimming to be possible, but a vigorous wave is necessary to produce
only slow forward motion. The swimming is thus ineffieient, but since most other
swimming modes faH altogether at low Reynolds number, the existence of this
mode is important.

24.11 Convection from the human body

Because the ambient air is usuaUy cooler than the human body, convection currents
are set up around the body. In the absence of other causes of appreciable motion,

Figure 24.20 Schlieren photograph of free convection from man's head. Photo provided by
R. S. Clark, National Institute for Medical Research.
Practical Situations 321

free convection boundary layers, broadly simiIar to those described in Section 14.5,
surround the body.
In reeent years it has been realized that this flow may be of considerable medical
significance. The concentration of airbome micro-organisms is markedly higher in
the boundary layers than in the surrounding air. In particular, the body is con-
stantly losing skin scales (about 10 10 per day) throUgh the rubbing actions of
limbs and clothes, and many of these scales have micro-organisms attached. It has,
for example, been suggested [160] that convection may produce the observed
connection between skin disease and respiratory disease (e.g. eczema and asthma),
by transporting organisms from the skin to the nose; and that it may account for
the increase in respiratory infections after a fall in air temperature, which would
resuIt in more vigorous conveetion. An understanding of the role of the boundary
layer in transporting skin scales is important for the reduction of infection during
orthopaedic operations. It enters into the design of air flow systems in operating
theatres, infant incubators, and industrial clean rooms, and into the design of
protective clothing such as operating gowns.
As a starting point for investigating these phenomena, detailed studies have been
made of the convection from a naked standing man and from a full-scale heated
model of a man [87, 160] . The distribution, thickness, and general behaviour of
the boundary layers were investigated by the schlieren method (Section 23.4),
giving photographs such as that in Fig. 24.20. Hot-wire anemometers (Section 23.2)
and thermocouples were used to obtain mo re information about the velocity and
temperature distributions. These are similar to the distributions for a heated vertical
plate (Section 14.5), and transition simBarly occurs; the flow is laminar on the
legs but becomes turbulent further up the body. On a dothed man, the Grashof
number may be low enough for the flow to remain larninar over the whole body.
The next stage is, of course, the extension of this work to the more complicated
situations related to the applications. An important additional aspect is the develop-
ment of instruments to deteet small airbome partides; a successful method has
used a modified hot-wire anemameter system, a pulse being generated by the
sudden cooling of the wire on the impact of a partide [79].

Figure 24.21
o
Sketch ofboornerang with 'breaks' to show cross-sections of arms.
322 Physical Fluid Dynamics

Figure 24.22 (a) Flight of boomerang with light. (b) Computer simulation of view of
boomerang flight. From Ref. [125).
Practical Situations 323
24.12 The flight of a boomerang

The inventors of the boomerang somehow chaneed on a design that requires aerofoil
theory for an understanding of its performance. Whilst the name boomerang is
commonly associated with a device that returns to its thrower, there are also
boomerangs which fly almost straight. These were used for hunting food and in war,
having the advantage over an ordinary projectile of travelling much further and
strildng harder; the good 'aerodynamic design' gave small viscous_energy losses and
stability of flight. The differences between re turning and nonoreturning boomerangs
are outwardly rather slight; small changes in the detailed shaping have a large effect
on the trajectory. The essential features of a boomerang are shown in Fig. 24.21.
Each arm has a cross-section of aerofoil shape, with the leading edges on opposite
sides of the two arms. There will thus (Section 11.6) be little resistance to rotation
of the boomerang in its own plane (anticlockwise in Fig. 24.21). Actual boomerangs
are usually rather more complicated than this, often being slightly twisted so that
the two arms are not in the same plane. The flight paths can be correspondingly
complicated, with changes in the orientation of the boomerang producing different
detaile d dynamics over different parts of a flight.
However, the basic mechanism by which a boomerang returns can be understood
fairly simply in the following way [125]. Areturning boomerang has its aerofoil
sections shaped or oriented so that a sideways force is generated (as discussed in
Sections 12.1 and 12.2). The bend in the middle is not essential to the performance,
although it may be important for ease of throwing. Consider a boomerang flying in a
vertical plane, so that an observer positioned appropriately sees it travelling from
right to left and rotating anticlockwise. Sideways forees act on the two arms, but
unequally. The arm instantaneously nearer the top is moving through the air faster
than the other, because the linear motion adds to the rotational motion for the
upper arm and subtracts from it for the lower. The consequent difference in side-
ways forees produces a couple on the boomerang. In the absence of rotation, such
a couple would twist the boornerang out of the vertical plane. But gyroscopic action
associated with the rapid spin of the boomerang leads to it turning about a vertical
axis. The boomerang thus travels on a curved path whilst remaining vertically
oriented.
Computations based on refinements of this basic idea have provided detailed
information about the behaviour of boomerangs of various designs (125] .
Figure 24.22 shows a comparison of an actual flight path shown by a boomerang
with a light on one tip and the result of such a computation (the programme includ-
ing a perspective correction so that the pattern corresponds directly to the
photograph).
Notation

The following list omits some items which appear in onlyone section of the book
and are defined there.
Where a symbol has more than one meaning (either all Iisted below or some
defined only in the text), the use in any place should be apparent from the context.
Dimensions of quantities below are indicated where this provides a useful
reminder of the definitions. Primary dimensions are denoted: M = mass; L = length;
T= time; e = temperature.t
Subscripts: Symbols with a subscript speeific to the symbol are listed below.
Other subscripts have general uses. Subscripts corresponding to a co-ordinate
indicate the corresponding component of a vector (e.g. Ut/J = azimuthal component
of veloeity; k x = x-component of wave number). The subscript 0 indicates a
reference or ambient value; a quantity with this subscript is usually a constant, but
is occasionally a basic variable speeifying a situation (e.g. uo(x) = velocity outside
boundary layer; Po (z) = basic density stratification). Subscripts I, 2 indicate values
at boundaries. Subscripts av and max indicate average and maximum values.
Superscripts: An overbar, -, indicates an average value. A cap, A, indicates a
unit vector (e.g. i = unit vector in x-direction; ii = unit vector normal to surface).

a half-width of channel
radius (pipe, cylinder, sphere, liquid column)
length speeifying geometrical similarity
speed of sound
b length speeifying geometrical similarity
c chord of aerofoil
length specifying geometrical similarity
concentration
Cg group veloeity
cp phase veloeity
d diameter (pipe, cylinder)
thickness of fluid layer
f similarity form of stream function (boundary layer, jet)
isotropic turbulence longitudinaI correlation function
f( ) general function of quantity(ies) in brackets

tWhen heat is involved its dimensions (ML' T-') are written as a separate group. The reason is
that, when interactions between mechanieal energy and thermaI energy are not physieally
important (as in the Boussinesq approximation), heat ean sometimes be treated as a further
primary dimension.
Natatian 325

g Igl
similarity form of velocity profile (boundary layer, jet)
isotropie turbulence transverse correlation function
g acceleration due to gravity
h height (vertieal slot, Rossby wave layer)
V-I
j isotropic turbulence triple correlation function
k thermal conductivity ((ML 2r 2 )L -1 r l e-I)
wave number (L -1 )
horizontal wave number, (k; + k~ )112
k wave number (L -1 )
I length (channel, plate)
co-ordinate along streamline
dl element of length
m exponent (u max a: x m )
n co-ordinate normal to boundary
frequency
exponent (Ll a: x n , Nu a: Gr n )
P pressure (ML -1 T- 2 )
pressure fluctuation in turbulent flow
Ph hydrostatic pressure
PS stagnation pressure
q lul
total velocity fluctuation in turbulent flow
r polar co-ordinate
space separation (correlations)
r general position
space separation (correlations)
s period of integration
time separation (correlations)
t time
U x-component of velocity
x-component of velocity fluctuation in turbulent flow
UT wall flow velocity scale
u velocity
velocity fluctuation in turbulent flow
v y-component of velocity
y-component of velocity fluctuation in turbulent flow
w z-component of velocity
z-component of velocity fluctuation in turbulent flow
x Cartesian co-ordinate
distance in main flow direction
Xo origin of x for self-preserving development
y Cartesian co-ordinate
distance perpendicular to main flow direction
z Cartesian co-ordinate
co-ordinate perpendicular to planes of two-dimensional motion
vertical co-ordinate (positive upwards)
co-ordinate paralleI to axis of rotation
326 Notation
A eonstant of integration
non-dimensional quantity in theory of Boussinesq approximation
B eonstant of integration
non-dimensional quantity in theory of Boussinesq approximation
B general veetor quantity
e numerieal eonstant of proportionality
non-dimensional quantity in theory of Boussinesq approximation
drag eoeffieient (dimensionless)
lift eoeffieient (dimensionless)
speeifie heat at eonstant pressure ((ML 2 r 2 )M-l e-I)
drag (MLr 2 )
drag per unit length (MT- 2 )
non-dimensional quantity in theory of Boussinesq approximation
D/Dt substantive derivative
E rate of kinetic energy transport (M L 2 T- 3 in three dimensions, MLT- 3 in
two)
energy speetrum (with respeet to wave num ber) (L 3 r 2 )
internal energy
F speetral transfer funetion (L 3 r- 3 )
F body foree per unit volume (ML -2 r 2 )
G pressure gradient (ML -2 T- 2 )
H heat transfer per unit area per unit time ((ML 2 T- 2 )L -2 T- 1 )
H heat flux ((ML 2 r 2 )L - 2 r 1 )
J rate of internal heat generation per unit volume ((ML 2 T- 2 )L -3 T- 1 )
K eonstant of proportionality
Karman constant
L length seale
Monin-Obukhov length
lift (ML r 2 )
lift per unit length (MT- 2 )
M rate of momentum transport (MLT- 2 in three dimensions, MT- 2 in two)
N Brunt-Väisälä angular frequeney
P differenee between pressure and hydrostatie pressure
pressure amplitude (wave or perturbation)
me an pressure in turbulent flow
probability distribution funetion
Q heat (in thermodynamie eonsiderations)
density amplitude (wave or perturbation)
R gas eonstant
eorrelation eoeffieient
eorrelation eoeffieient between x-eomponents of veloeity at two
positions
s general surfaee
eross-seetional area of streamtube
entropy
N/2n (dimensionless)
dS element of surfaee
T temperature
mean temperature in turbulent flow
Notation 327
adiabatie temperature
veloeity seale
x-eomponent of veloeity amplitude (wave of perturbation)
x-eomponent of mean veloeity in turbulent flow
bulk velocity superimposed on wave
veloeity of front of turbulent plug
veloeity of rear of turbulent plug
veloeity of boundary
veloeity amplitude (wave of perturbation)
me an veloeity in turbulent flow
v transverse veloeity seale
y-eomponent of veloeity amplitude (wave or perturbation)
y-eomponent of me an veloeity in turbulent flow
general volume
l/p
dV element of volume
W vertieal veloeity seale
z-eomponent of veloeity amplitude (wave of perturbation)
work done by heat engine
Ek Ekman number
Fr Froude number
internaI Froude number
Gr Grashof number
Ma Maeh number
Nu Nusselt number
Pe Pec1et number
pr Prandtl number
Ra Rayleigh number
Re Reynolds number
Rea Reynolds number in whieh the length seale is a
Ri Riehardson number
Ro Rossby number
St Strouhal number
a eoeffieient of expansion (e -1)
angle of attack
angle between boundaries
eoeffieient of density variation with eoneentration
eompressibility (M- 1 LT2 )
ratio of speeifie heats
gradient of layer depth
intermitteney faetor
transverse length seale
boundary layer thiekness (99 per eent thiekness)
ox, etc. small ehange in x etc.
fJ jj Kroneeker delta
€ small distanee
turbulent energy dissipation per unit mass (L 2 T- 3 )
alternating tensor
z-eomponent of vOftieity, W z , when W x = w y = 0
328 Notation

r/ similarity form of distance across boundary layer or jet

y-component of vorticity, w y , when W x = W z = 0
e polar co-ordinate
angle between wave number and vertical
potential temperature
temperature fluetuation in turbulent flow
walllayer temperature seale
thermal diffusivity (L 2r l )
eoneentration diffusivity (L 2 r- I )
moleeular mean free path
length of bloeked region, Taylor eolumn
second viseosity eoeffieient (ML -1 r- I )
JJ. viseosity (ML -I rl)
v kinematie viseosity (L 2 r- l )
~ non-dimensional wave number
P density (ML -3)
Ph hydrostatie density field
a foree per unit area in main flow direction on surfaee (ML -I r- 2 )
amplifieation eoeffieient in stability theory (rl)
real and imaginary part s of amplifieation eoeffieient
tangential stress in shear flow (ML -1 r- 2 )
value of T at wall
polar eo-ordinate
veloeity potential (L 2 r l )
energy speetrum (with respeet to angular frequeney) (L 2 r- l )
stream function (L 2 r- l )
'"
W
w
angular frequeney
vortieity (r- I )
r eirculation (L 2 r- l )
Ll length proportional to thickness of boundary layer, jet, etc.
Ekman layer thickness
.:lx etc. finite difference in x etc .
M' pressure difference scale
e temperature difference scale
temperature amplitude (wave or perturbation)
longitudinal pressure variation scale
torque per un it length (NLr-2 )
turbulence kinetic energy per unit volume (ML -I r- 2 )
transverse pressure variation scale
r-
gravitational potential (L 2 2 )
energy dissipation per unit volume (ML -1 r- 3 )
time scale
angular velocity (local)
angular velocity of reference frame
angular veloeity of reference frame.
Problems

For the purposes of the following probIems, air may be taken as a perfect gas with
density 1.2 kg m -3 , kinematic viscosity 1.5 x 10 -5 m 2 s -I, and speeific heat at
constant press.!lre 1.0 x 103 J kg -I K -I; water may be taken to have density
1.0 x 103 kg m- 3 , kinematic viscosity 1.0 x 10-6 m 2 S-I, and coefficient of
expansion (at 20°C) 2.1 x 10-4 K- I •

1. A layer of fluid (viscosity Il, density p) of depth d flows under the influence of
gravity down a wide plane inclined at an angle 0 to the horizontal. Assuming that d
is constant and that all conditions are steady, find the veloeity as a function of the
normal distance y from the plane. Determine also the ratio of the average velocity
to the maximum velocity, and the mass flux per unit width of the plane. (The free
surface supports no viscous stress and the pressure above it may be supposed uniform.)

2. A vertical tube of diameter 2 mm has its upper end open to the atmosphere. At
the lower end water is maintained at a pressure of 104 N m -2 above atmospheric.
What is the longest length of tube for which this pressure will produce a flow
through the tube? (Ignore surface tension.)
What volume of water will pass through the tube per second if the length is half
this? (Assume that the flow is laminar and ignore the entry length.)

3. A two-dimensional channel of constant width d, containing fluid of density p

and viscosity Il, has one wall at rest and the other moving in its own plane with
speed U. The fluid moves under the action of a negative pressure gradient, -G,
applied paralleI to U. In what circumstances will the flow speed exceed U in some
region?
What can you say about the stress on the moving wall when the conditions for
this are just not fulfiUed?

4. A fluid flows under apressure gradient G down a channel with parallel walls of
effectively infinite extent. The width of the channel is d. The temperature of the
fluid varies linearly with the distance y from one wall,
T= T I + (T2 - Tdy/d
The viscosity of the fluid varies exponentially with the temperature,
Il = llo ek (T-To)

Supposing that the temperature and velocity proftles are the same at aU stations
downstream, determine the distributions ofvelocity and vortieity.
330 Problems

Under what circumstances do you think the supposed temperature distribution

would be a realistic one?

5. Ajet is specified by the diameter of the orifice, the momentum per unit time of
of the fluid issuing from the orifice, and the density and viscosity of the fluid. What
will be the form of the criterion determining whether the jet is laminar or turbulent?

6. For flow of nitrogen in a pipe of diameter 10 mm at a temperature of 300 K,

estimate how low the pressure may be without violation of criteria (5.1) and (5.2)
for the applicability of continuum mechanies. (Avogadro's constant = 6 x 1023
mol- 1 ; collision cross-section of nitrogen molecule =6 x 10- 19 m 2 .)

7. Show that the continuity equation (5.9) may be rewritten

Dp/Dt + p V· u = 0
Derive the equation in this form from first principles, starting with the fact that
the mass of fluid in avolume V moving with the flow is conserved; that is
D(p V)/Dt = O.

8. Show that a change offrame of reference as in Fig. 3.1 results in a change of

each of au/at and u· Vu, but leaves Du/Dt unchanged.

9. Verify relationship (5.61) as the criterion for compressibility effects to be

negligible in a low Reynolds number flow.

10. The water supply in the apparatus of Fig. 2.1 is suddenly tumed off.
Supposing that the instantaneous flow pattem throughout the pipe is that of
Poiseuille flow, derive an expression for the time in which the height of the water
above the pipe entry falls to one half of its original value.
Formulate criteria for the negligibility of the inertial effects associated with
(a) the acceleration of the water into the pipe, and (b) the changing velocity
within the pipe.
Evaluate the above time for a pipe of diameter 1 mm and length 10m leading
out of a tank of horizontal cross-section 10-2 m 2 . Check whether the above
criteria are fulfilled if the initial height is 100 mm.

11. Show that the streamlines of the motion

u = -ny v = nx W=Wo
(where n and Wo are constants) are
x 2 + y2 =a2 x =a sin(S1zjwo + a)
(Note: do this by forward integration, not by substitution of the result.)

12. In each of cases (i) and (ii) below, a flow is described in a Lagrangian way by
giving, as a function of time t, the co-ordinates (x, y, z) of the fluid partide that is
at (xo,Yo, zo) at t = O. For each flow, (a) formulate the equations for the path of
this partide, (b) derive the Eulerian equations (u = u(x,y, z, t) etc.), (e) consider
whether the flow is steady, (d) consider whether the flow satisfies the incompres-
Problems 331

sible continuity equation, and (e) derive a general equation for a streamline. What is
the nature of the flows?
(i) x = xoe- 2t / s y = yoe t / s z = zoe t / s
(ii) x = xoe- 2t / s y = Yo (1 + tjS)2 Z = zoe 2t / s(1 + tjS)-2
(t> 0; s = constant > 0)

13. Derive the general equation for a streamline in the two-dimensional flow (an
approximate form of Rossby wave)
u = Uo v = Vo cos(kx - cxt)
where Uo, Vo, k, and cx are constants. At t = 0, what is the equation for the stream-
line passing through x = O,y :: O?
Derive also the equation for the path of the partide which is at x = 0, y = 0 at
time t = O.
Comment briefly on the comparison of the streamline and the partide path in
the two limiting cases, cx =0 and k =O.

14. Show geometrically that for cylindrical polar co-ordinates

ai/atP = .p and a$/atP = -i
Hence, by writing
[Link] Aa
u=urr+uq,.p+uzz y=r-+- - + z -
ar r atP az
show (i) that the inertia term take s the form given in equations (5.26), (5.27), and
(5.28); and (ii) that, for the case when u y =U z =0 and uq, depends on ralone, the
viscous term takes the form given in equation (9.3).
Note: the order of operations is indicated by writing U' YU = (u'v)u and
v 2 u = ( Y' v)u.

15. (i) Show that, for an unsteady flow in which only the x-component of velocity
is non-zero and this varies only in the y-direction, the Navier-Stokes equation
reduces to a form analogous to the equation ofunsteady one-dimensional heat
conduction in a solid.
(ii) An effectively infinite flat plate bounding a semi-infinite expanse of fluid
oscillates in its own plane with velocity
U= Uo sin wt
Supposing that the induced fluid motion is an oscillation of the same frequency,
how do the amplitude and phase vary with distance from the plate?
(iii) For a geometry similar to that in (ii), the plate is suddenly brought into
motion at time t = 0 and then moves in its own plane with constant velocity Uo.
Both the plate and the fluid were at re st for t < O. Show that, for t > 0, the fluid
velocity is
u = Uo erfc[y/2(vt)1/2]
Find the force per unit area (as a function of time) needed to produce this motion.
332 Problems
Henee, find the total work done after any given time, and determine what propor-
tion of this work has appeared as kinetie energy and what proportion has been
dissipated.
2 oo 2
2 - V2 ]
[ erfex=-f (erfexi dx=~
fo
vrr
oo
e-X dx; erfeO= 1;
x

16. Show that for an axisymmetrie vorticity distribution w = (0, 0, t(r)) in

eylindrieal polar eo-ordinates (r, ep, z), the vorticity equation reduees to

at=v- -a (rat)
- -
at r ar ar
If there is a coneentrated line vortex along the z-axis at time t = 0 in an otherwise
irrotational fluid, show that the vorticity distribution at any subsequent time is

t =~ exp( _r2 /4vt)

4rrvt
What is the eorresponding veloeity distribution?

17. If, for turbulent flow through a pipe, it is observed that the pressure gradient is
proportional to Q7!4 (where Q is the volume rate of flow), prediet how the pressure
gradient would vary with the viseosity and density of the fluid (for fixed Q).

18. The figures below represent observations of the power needed to propeI two
objects (e.g. toy submarines) through a fluid at various speeds. The objeets are
geometrieally similar, the second being five times as large as the first (linear
dimensions). Show that, although the relationship between speed U and power P is
different in the two cases, all the figures are consistent and ean be eombined in an
appropriate way. (The other relevant quantities are the density and the viseosity of
the fluid.)
Comment on the interpretation of the different relationships between U and P.
Small object (L = 1 length {mit):
U 1 2 3 5 10 speed units
P 1.0 4.5 12 50 400 power units.
Larger object (L = 5 length units):
U 1 2 3 5 10 speed units
P 10 80 270 1250 10000 power units.

19. A cireular dise is set spinning with angular veloeity n in an incompressible

viseous fluid. Use dimensional analysis to find the general funetional form of the
expression for the time t in whieh the angular velocity falls to n/2. Assume that
this depends only on n, the radius of the dise a, the density of the dise a, the
density p and the viscosity Il of the fluid.
Could you designate any of the parameters involved as a Reynolds number?
In an experiment of this type using a fluid having p = 1.0 x 10 3 kg m -3 and
Il = 1.0 X 10-3 kg m -1 S-I, t was observed to be 100 s when n was 20 rad S-I.
Problems 333

What ean you conelude from this about the motion of the same dise (a) in a fluid
withp= 1.0 x 103 kgm- 3 ,1l=2.0x 10-3 kgm- 1s-1 ,and(b) ina fluid with
p = 0.75 X 10 3 kgm- 3 ,Il = 3.0 x 10- 3 kgm- 1 s-1?
Suppose now that one argues on physical grounds that t must be directly
proportional to a (beeause this enters the problem only through the moment of
inertia of the dise and not through the resisting fluid motion). How ean the general
expression for t be simplified?
What ean you now conelude in cases (a) and (b) above?

20. A string of density a, diameter d, length I is held in tension aeross a ehannel

through which fluid of kinematic viseosity v flows. The string vibrates when its
fundamental natural frequeney eoincides with the Karman vortex street frequeney.
The following observations were made of the tension F at whieh resonanee oeeurred
for various flow speeds U of a fluid with v = 1.0 X 10- 5 m 2 S-1:
0.2 OA 0.6 0.8 1.0 1.2 lA
0.096 0043 1.03 1.92 3.00 4.3 5.8
Suppose now that the fluid is ehanged to one with v = 5.0 X 10-5 m 2 S-1. At whieh
of the following speeds ean the resonanee tension be predicted and then what is it?
U(m S-1) OA 1.0 5.5
(You may assume that the only properties of the st ring that enter the problem are a,
d, and I, and that ehanges in the density of the fluid have no effeet (exeept through
the kinematie viseosity); and that no harmonies are generated.)

21. A plate bounding a region containing fluid of density p and kinematie viseosity
v is subjeeted to a tangential oseillatory foree, F per unit area, of frequeney n so
that
F = Fo sin 2rrnt
As aresult the plate oseillates with frequeney n and with amplitude A o. Experi-
ments show that, when the resulting motion of the fluid is laminar,A o is
proportional to Fo (all other quantities being held eonstant) and, when it is fully
turbulent, A o a: F~/2. How would you expeet A o to ehange (a) for laminar motion,
(b) for fully turbulent motion, and (e) in the transition region between the two, for
the following cases?
(i) when the frequeney is doubled (at eonstant Fo, p, and v)
(ii) when the frequeney is doubled and the fluid is ehanged to one of half the
original viseosity (at eonstant Fo and p).
You may suppose that the other boundaries of the fluid region are distant (so that
no length eharaeteristie of the region enters the problem) and that the inertia of the
plate is negligible (so that Fo is entirely balaneed by the fluid resistanee to the
motion). .
Note: iUs not neeessarily possible to make a predietion in every ease.

22. An effeetively two-dimensional jet of fresh water is emitted horizontally from a

slit of width a with an average speed U elose to the top of a tank containing salt
water. The density of the salt water exeeeds the density, Po, of the fresh water by a
334 Problems
controllable arnount, Ap. Ap is sufficiently small that its only effeet on the
dynarnics of the jet is in changing the gravitational force on fluid partides. In a
series of experiments the lowest value of U at which the jet became turbulent was
observed for various values of Ap. The results were:
a= 5.0 mm; Po = 1000 kg m -3 ; J.L(viscosity) = 1.0 x 10- 3 kg m- l S-I

Ap (kg m -3) 024 8 12 16

U(mms- l ) 10 25 33 43 51 58
Similar experiments can be done with changed slit-width and with different fluids
and contarninants. Will the flow be larninar or turbulent in each of the following
cases?
(i) (ii) (iii)
a(mm) 10 5.0 50
Po (kg m- 3) 1000 1.0 2.0
J.L(10-3 kgm-ls- l ) 1.0 0.010 0.020
Ap (kg m- 3) 2.0 0.020 0.0010
U(mms- l ) 25 400 20

(You may assume that the wall s of the tank have no effeet on the motion; and that
the diffusivity of the contaminant may be ignored. g may be taken as 10m s -2.)

23. Determine the conditions for dynarnical similarity of steady incompressible flow
of an electrically conducting fluid in a magnetic field, governed by the equations=
V· u=O
V'. B = 0
1 1
u • VU = - - v' P + - (V x B) x B + vV 2 u
P pJ.L
1
u·VB=B·Vu+-V 2 B
UJ.L
Briefly make any comments on the results that you consider to be of interest.
(Notation: u = velocity; B = magnetic field;p = pressure; p = density; v = kinematic
viscosity; J.L = magnetic permeability; u = electrical conductivity.)

24. (i) Determine the vorticity distribution in rotating Couette flow (Section 9.3).
Determine the circulation round any path containing the inner cylinder when the
radius of the outer cylinder becomes infinitely large.
(ii) For fixed n 2 jn l and a2/al' compare the magnitudes of the tangential
force per unit area acting on either cylinder and the difference between the normal
forees per unit area on the two cylinders in terms of the Reynolds number.

25. (i) What is the 1argest size ofwater drop (assumed spherical) for which the rate
offree fall in air can be calculated using Stokes's formu1a (9.17) and what is its rate
offall?
(ii) How small must a water drop be for its fall under its own weight to be
Problems 335

negligible when it is in a 1 m s -1 wind generating turbu1ence of intensity equal to

10 per cent of the mean velocity?

26. (i) Is the motion incompressible for the flows given by the following velocity
potentials? lf so, determine the corresponding stream functions.
(a) ep = c(x 2 + y2)
(b) ep = C(}C2 _ y2)
(ii) Is the motion irrotational for the flows given by the following stream
funetions? If so, determine the corresponding velocity potentials.
(a) 1/1 = C(x 2 + y2)
(b) 1/1 = C(x 2 _ y2)
Sketch the strearnlines for all cases and the lines of constant ep where possible.

27. Show that if the effeet of the hydrostatic pressure is significant, Bernoulli's
equation becomes
p + I,6,pq 2 + pgz = constant
where p is the true pressure and z is the vertieal co-ordinate (positive upwards).
Hence, show that the velocity ofliquid emerging from a small hole in a tank with
a free surface is the same as the velocity that would be acquired by free fall from the
level of the surface to that of the hole.

28. Find how the constant in Bernoulli's equation varies with radius for each of the
two velocity fields, equation (6.17) and equation (6.19).

29. Fluid flows out of a reservoir (so large that the fluid in it can be considered
stationary) through a circu1ar tube with a Venturi constriction in it. The radius of
the tube is 10 mm, reducing to 5 mm in the constriction. lfthe pressure (above
atmospherie) is 200 N m -2 in the reservoir and 190 N m -2 in the main part of the
tube, what is it in the constriction? Suppose that the flow is incompressible and
inviscid (low Mach number, thin boundary layers).

30. (i) Sensitive liquid level manometers can be read with an accuracy of about
± 10- 2 mm ofwater. What is the lowest air velocity that can be measured to
within 5 per cent with a Pitot tube?
(ii) It is found empirically that buoyancy effects upset the calibration of a hot-
wire anemometer when Gr 1/ 3 > Re/2. (Why is this of a different form from equation
(13.14)?) What is the lowest air velocity that can be measured when the
ratio of the absolute temperature of the wire to that of the air is 1.5?

31. Show that all the conditions for inviscid incompressible flow past a fixed
circular cylinder of radius a (with uniform velocity at large distances) are satisfied
by a velocity potential in cylindrieal polar co-ordinates of the form

Uo ( r+ a: )eos ep
336 Problems

Show that the corresponding expression in spherical polar co-ordinates for flow
past a sphere is

What are the pressure distributions over the surfaces of the cylinder and sphere?
32. In spherical polar co-ordinates (r, (J, ep), the velocity potential
<I> = -kr2P2(COS (J) = -W,,2(3 cos 2 (J - 1)
represents the axisymmetric flow produced by the confluence of two equal and
opposite streams. Sketch the streamline pattem in any plane through the axis of
symmetry.
If nowa solid sphere with surface , = a is placed in this flow, how is the velocity
potential modified? Hence, determine the distribution of velocity and pressure over
the surface of the sphere.
State briefly the principal ways in which a real flow would be expected to depart
from this ideal one.

33. (i) For (a) inviscid flow past a sphere and (b) Stokes flow past a sphere,
determine the distance to the side of the sphere (measured in sphere radii) at which
the difference from the free-stream velocity falls to 1 per cent.
(ii) For high Reynolds number flow past a cylinder, estimate the distance at
which the difference from the free-stream velocity falls to 1 per cent (a) to the side
of the cylinder and (b) directly downstream of the cylinder.
Note: refer to the solutions of Question 31 and to Fig. 22.6.

34. Compare the two-dimensional boundary layers on plane and curved surfaces
below the same inviscid velocity distribution uo(x), where x is the curvilinear
co-ordinate in the surface. Find the order of magnitude of the pressure variation
across the boundary layer for the latter case, by considering the balance between
the pressure gradient and the centrifugal force associated with the curved flow.
Hence, show that the boundary layer equations are the same for the two cases
provided that 0 IR ~ I, where 0 is the boundary layer thickness and R is the
radius of curvature of the surface.

35. In the two-dimensional flow away from a stagnation point on a flat wall, the
inviscid velocity at the wall is
Uo =ax
where x is the distance along the wall from the stagnation point (ef. the inviscid
flow in Question 26(i)(b )). Show that the boundary layer below this flow, in the
region where the Reynolds number ax2 lv is large, has constant thickness. If the
stream function in the boundary layer is written
1/1/= ka x [(Ylo)
show that it is appropriate to put
k = 0 = (vla)1I2
Problems 337

and that the governing differentia1 equation is then

["'+[('-['2+1=0
Formu1ate the boundary conditions for this.

36. The growth of a boundary 1ayer can be inhibited by sucking some of the fluid
through a porous wall. In appropriate circumstances, this can give a boundary 1ayer
of which both the thickness and the ve10city profIle remain constant with distance
downstream. Consider this situation for a two-dimensiona1 flat p1ate boundary
1ayer in zero externa1 pressure gradient. Determine the ve10city profIle, by con-
sidering first the continuity equation and then the momentum equation with
boundary conditions:
u = 0 and v = - Vo at y = 0
u = U oo at large y.
Whereabouts on a wall of finite extent wou1d you expect this solution to app1y?
How does it re1ate to the solution when Vo =O? Mention another situation in
which the introduction of an extra parameter gives rise to a simp1er solution in a
similar way.

37. Show that the ve10city profIle (11.53) satisfies each of (11.63), (11.64), and
(11.65). (Note: by use of the appropriate variab1e this can be done without
detaile d integrations.)

38. (i) Use a procedure similar to that in Section 11.9 to show that, in the wake
behind a two-dimensiona1 obstac1e fixed in a stream of ve10city uo,

dMjdx=O, whereM=p J: oo u(uo -u)dy

Interpret M as the rate of transport of the momentum deficit produced by the

drag on the obstac1e.
(ii) Corresponding1y, the rate oftransport of kinetic energy deficit is
1
E =- p
2
Joo
-oo
u(u5 - u 2 )dy

Show that
1
E= uoM --p
2
J-oo
oo
u(uo - u)2dy

Hence, for a se1f-propelled body (Le. a body ho1ding itself fixed against the flow)
for which M = 0, show that the wake invo1ves transport of kinetic energy away from
the body.

39. (i) The disp1acement thickness and the momentum thickness of a boundary
1ayer are defined respective1y by

0* = Joo
o
(1 - ~)dY;
Uo
0** = Joo ~ (l-~)dY
o Uo Uo
338 Problems

Show that for Blasius flow eaeh of these is a eonstant proportion of the 99 per eent
thiekness.
(ii) Show, also for Blasius flow, that

u5 do**/dx =v(au/ay)y=o
(Hint: relate eaeh side to an integral in terms of the non-dimensional parameters 1/
and J, and then prove that the two integrals are equal.) Interpret this result in terms
of Newton's second law.

40. Show that the velocity field (in two-dimensional polar eo-ordinates (r, ep))

ur = -uo (1 _ar 2
2
) eos ep

ul/J = Uo
a2) sin ep + -r
(1 +"2
r 21fr
represents a solution of the equations of inviseid irrotational motion around a
cireular eylinder of radius a, tending to a free-stream velocity Uo far from the
eylinder and with cireulation r round the eylinder. Determine the pressure
distribution over the surfaee of the eylinder, and henee show that the lift per unit
length is given by the Kutta-Zhukovskii result, L = -puor.

41. From the information that an aireraft of mass 104 kg has wings of average
ehord 3 m and total span 30 m, estimate the magnitude of the air speeds associated
with eireulation around the wings during level flight at a speed of 100 m S-I.

42. Consider a smooth ball of diameter 40 mm and density 1.2 x 10 3 kg m- 3

travelling (instantaneously) horizontally at a speed of 20 m S-I. Cheek that the
experimental data in Fig. 12.9 are applieable and use them to determine (a) the
radius of eurvature in a horizontal plane of its trajeetory when it is spinning about a
vertieal axis at an angular velocity of 600 rad S-I, and (b) how rapidly it would
need to spin about a horizontal axis in order to eounteraet 20 per eent of the
gravitational foree.

43. (i) Consider foreed eonveetion in a cireular pipe of radius a with the Poiseuille
velocity profile, equation (2.17). Suppose that the temperature of the pipe wall
increases linearly with distanee along the pipe, and that, in eonsequenee, the same
eonstant axial temperature gradient aT/ax exists everywhere in the pipe. Determine
the radial temperature distribution. Determine also the heat transfer from the pipe
wall. Express the latter result in the non-dimensional form, equation (13.22)
(defining the temperature differenee seale as the temperature ehange in a length of
the pipe equal to its diameter, e = 2aaT/ax).
(ii) Show how the above solution ean be extended to apply to Poiseuille flow
oeeurring with uniform internaI heat generation in a pipe with a thermally
insulating wall. Explain why, for this flow to establish a uniform axial temperature
gradient, there must be a temperature differenee between the fluid at the wall and
that at the eentre, and derive an expression for this differenee.
Problems 339

44. Write the equations of free eonveetion in the approximate form appropriate to
a two-dimensional boundary layer. Henee, verify the relationships (14.29) for the
variation of boundary layer thiekness and maximum veloeity in free eonveetion
from a vertieal plate.
By making the substitutions

where
1/ = [ga(T I - TO )/v 2 x] 1I4 y
show that the differential equations governing this flow are
4(" + 3//" - 21'2 + 40 = 0
40" + 3Pr/0' =0

Write down the boundary eonditions for these equations.

Show that the heat transfer dependenee on (T I - To) takes the fonn of
relationship (14.36).

45. Conveetion in a layer between two vertieal walls at different unifonn

temperatures (as deseribed in Seetion 14.7) of large enough aspeet ratio (hld) ean
have a region in the middle vertieally in which the veloeity and the temperature
depend only on the horizontal eo-ordinate and not on the vertieal. Show that, in
this region, the temperature profile is unaffeeted by the flow and detennine the
veloeity profile. Express the latter resuIt in terms of the dependenee of the
Reynolds number, based on the maximum veloeity and the layer width, on the
Grashof number.

46. (i) Movements of the Earth's mantle are inferred with speeds around 10- 9 m S-I
(see Seetion 24.4). Supposing these are produeed by free convection in a
homogeneous fluid of kinematie viseosity 10 17 m2 S-I and eoeffieient of expansion
10 -5 K -I and that the length seale of both the veloeity and temperature fields is
106 m, what are the temperature differenees associated with the flow?
(ii) If the thermal diffusivity of the mantle is 10-7 m2 S-I, what is the thiekness
of the thermal boundary layers associated with the above flow? If the temperature
variations are eonfined to such boundary layers, how is the answer to (i) modified?

47. A tank of fluid rotates in rigid body motion. Superimposed on this is a 'jet-
stream' type of motion in whieh some of the fluid is moving azimuthally with a
different angular veloeity. Show that an observer at rest in the laboratory regards
the modifieation to the foree field by this stream as a ehange in the eentrifugal
foree; that an observer rotating with the tank regards it as either a Coriolis foree or
a Coriolis foree pIus a eontribution to u· V'u (depending on whether the Rossby
number is low or not); but that the two are agreed on the radial pressure gradient
associated with the motion.

48. A demonstration of the action of the Coriolis foree in fluid dynamies ean be
made by introdueing a capiliary tube into a uniform flow in a rotating ehannel. The
basie flow, the tube, and the axis of rotation are all mutually perpendieular. Explain
340 Problema
briefly why there is a flow through the tube, and calculate the ratio of the mean
speed of this flow to the speed of the basic flow when the angular velocity of
rotation is 0.5 rad s -1, the radius of the capillary tube 0.2 mm, and the kinematic
viscosity of the fluid 1 mm 2 S-1.
To an order of magnitude, how much higher is the water level on the French
coast than on the English coast when there is an eastwards tidal flow through the
English Channel at a typical speed of 1 m s -1 ?

49. Apply the laminar theory of the Ekman spiral to the case of a semi-infmite
expanse of fluid with a boundary moving in its own plane at a constant velocity U
with respeet to a frame of referenee rotating with angular velocity il about an axis
perpendicular to the boundary. The fluid far from the boundary is at rest in this
frame. How does the flow vary with distanee from the boundary? What is the
smallest distanee at which the flow direction is exactly opposite to that of the
boundary?
This theory may be applied to the motion of the upper layers of the ocean under
the action of the wind stress, with the difference that the viscous stress at the
surfaee, not the velocity, is prescribed. Show that the surface moves in a direction
at 45° to the direction of the stress, and the net mass flux over the whole flow is at
90° to the direction of the stress. (It may be assumed that only the vertical
component of the Earth's angular velocity need be considered.)

so. An Ekman layer becomes unstable at a Reynolds number (based on the

geostrophic velocity and the layer thickness) of about SS. Experiments are
performed in water in a rotating armuIus of inner and outer radii 0.45 and 0.55 m.
Ekman layers are observed on boundaries normal to the rotation axis (a) by moving
the boundary relative to the annulus at a speed of 30 mm s -1 ; (b) by producing a
transient geostrophic flow by suddenly changing the rotation rate by 10 per cent;
and (e) by produeing a geostrophie flow by maintaining a pressure differenee
between the inner and outer walls 10 per eent greater than that needed to balanee
the eentrifugal foree. In eaeh ease, what is the maximum or minimum rotation rate
at whieh the instability eould be observed? Also, at this rotation rate, is the
assumption that the Ekman layer is thin eompared with other relevant dimensions
fulfUled?

Sl. For Rossby waves of the type analysed in Seetion 15.7, show that the eriterion
for non-linear inertial effeets to be negligible is that the Rossby number based on
the velocity in the direction of the depth gradient and on the wavelength should be
small.
52. (i) A eyc10ne has pressure differenees of about 10 mbar (10 3 N m -2) over
horizontal distanees of about 10 3 km. What is a eharaeteristie wind speed?
(ü) A tornado has pressure differenees of about SO mbar over horizontal
distanees of about 10 2 m. What is a eharaeteristie wind speed?
(Briefly justify your choice of proeedure in eaeh case.)

53. (i) For free eonveetion in a rotating fluid, determine the non-dimensional
parameter that indicates the relative importanee of Coriolis and inertia forees
(ignoring viseous forees).
Problems 341

(ii) For free conveetian in a fluid rotating about a vertical axis sufficiently
rapidly that the Coriolis force is dominant, show that the Taylor-Proudman
theorem still applies to the vertical component of the velocity, but is modified for
the horizontal companents. mustrate the physical significance of the new form by
discussing briefly the vorticity balanee associated with the increase in speed with
height of westerly winds at mid-latitudes.

54. For a region c10se to the ocean surface in which the temperature decreases from
25°C to 15°C and the salinity increases from 3.50 to 3.55 per cent by weight over
100 mincrease in depth, which of the temperature stratification, the salt stratifi-
cation, and the Earth's rotation will have the most significant effeet on the
dynamical behaviour?

55. In Sectian 16.1, it is shown that when the Reynolds and Pec1et numbers are
both large, the criterian for stratification to affect a flow strongly is that Fr 2 should
be small. Show that the corresponding criteria in the cases of (a) small Reynolds
number but large Pec1et number and (b) small Reynolds number and small Pec1et
number are respectively that Fr 2 /Re and Fr 2 /RePe should be small.

56. It is desired to investigate the lee waves associated with the flow of a wind of
10 m S-I past a hill in an isathermaI (and therefore subadiabatie) atmosphere at
300 K, by towing al: 10 4 scale model of the hilLat 50 mm s -1 in a channel in
whieh a uniform vertical salt gradient has been established. If the channel depth is
200 mm and the water at the top is fresh, what should be the salt concentration
(expressed as weight of salt per unit weight of water) at the bottom, assuming that
the water volume daes not change when salt is dissolved in it? (Assume that
viscosity and diffusion may be neglected.)

57. A horizontal cylindrical body is oscillated periodically but non-sinusoidally in a

stratified fluid at a frequency equal to 0.3 times the Brunt-Väisälä frequency. For
each of the internaI wave modes generated, determine the direction of the group
velocity, the direction of the phase velocity, and the ratio of the group velocity to
the phase velocity.

58. Derive the principal properties of inertial waves in a rotating fluid (as shown in
Fig. 15.12) by the procedure applied to internaI waves in a stratified fluid in
Sectian 16.4. (Start with the linearized, inviscid, but time-dependent equations for
mati on in a rotating fluid and substitute velocity and pressure fields of the form of
equations (16.18) and (16.19).) In particular, show that
w = 2Q I eos eI
where e is the angle between k and Q, and that the group velocity is perpendicular
to k in the plane of k and .Q. Consider the significanee of the limiting cases w = 0
and w= 2Q.

59. The eapillary instability of a stationary cylindrical column of liquid may be

investigated by introducing a disturbance of a controlled wavelength. The table
below gives the amplification rate (reciproeal of e-foldillg time), in units of S-I, as a
function of the radius of the eolumn and of the wavelength, for a liquid with
342 Problems

density 1.5 x 10 3 kg m -3, surface tension 6 x 10 -2 N m -1 and kinematic viscosity

1.0 x 10 -4 m 2 S-I. For a 1 mm radius column of another liquid with density
1.0 x 10 3 kg m -3, surface tension 2 x 10 -2 N m -1 , and kinematic viscosity
0.5 x 10-4 m 2 S-I, what would be the amplification rate of alO mm wavelength
disturbance and what wavelength would give the maximum amplification rate?
Radius (mm)
0.5 1 2
5 660 0 0
10 750 310 0
15 580 340 85
Wavelength (mm) 20 480 300 135
30 340 230 135
40 290 180 120
60 180 140 90
60. Show that the non-dimensional parameter governing the onset of the variable
surface tension instability described in Section 17.4 is XATd/pvl<. (the Marangoni
number) where X is the negative rate of change of surface tension with temperature,
AT is the horizontally averaged temperature difference across the layer, and other
symbols are as in the text.
The critical Marangoni number in the absence of buoyancy effects is 80. The
critical Rayleigh number (for free upper surface with no horizontal heat transfer) in
the absence of surface tension effects is 670. For a layer of oil with p = 1.0 X 10 3
kgm- 3 ,v= 1.0 x 10-4 m 2 S-I,1<. = 1.1 X 10- 7 m 2 S-I,o:= 1.0 x 10-3 K- 1 ,and
X = 6 x 10-5 N m -1 K -1 , estimate the depth at which the predominant destabilizing
mechanism changes. What is the minimum temperature difference necessary to
produce motion at depths of each of 1/10 and 10 times the above depth?

61. Write down the velocity distribution in rotating Couette flow for the two
speeial cases of (a) two cylinders rotating in the same sense with the same angular
veloeity, and (b) a single cylinder rotating in an infinite expanse of fluid. Hence,
show that according to the inviseid stability criterion (the Rayleigh criterion) the
former is stable and the latter is neutrally stable.
Show that the Rayleigh criterion may be reformulated as indicating that the
motion is unstable if the vortieity has the opposite sign to the fluid angular veloeity.

62. The tendency for buses on a frequent service to form bunches can be under-
stood as follows: if one bus is slightly delayed, more passengers will accumulate at
the next stop and the delay will be increased; another bus that gets slightly early will
siInilarly become earlier stilI. A simple quantitative model of this can be formulated
by considering an infinite homogeneous bus route. If the nth bus is a time tn late,
its additional delay at a stop may be taken proportional to the excess time since the
last bus; that is to (tn - tn-I). After AN stops
Atn =A(tn - tn_dAN
If bOth the buses and the stops are sufficiently elose toge the r , tn can be treated as a
continuous function of n and N. Then one may write
3tn /3N=A(t n - tn-d
Problems 343
Or; approximating by the first two terms of a Taylor expansion,
atn/aN=Aatn/an - YzAa 2 t n/an 2
Consider the evolution in time (i.e. with respeet to N) of a disturbanee that is
periodie in space (i.e. with respeet to n).
Diseuss the extent to whieh this treatment is analogous to the methods used for
studying hydrodynamie instability.

63. Consider whether any signifieant part of the boundary layers might be laminar
in the cases of (a) a ship oflength 100 m sailing at a speed of 10m s -1 , (b) a fish of
length 0.5 m swimming at 2 m s -1 , and (e) the aireraft of Question 41.

64. It is planned to earry out a vibrating ribbon experiment to verify the boundary
layer stability eurve shown in Fig. 18.7 for Reynolds numbers up to 5000, using a
wind-tunnel with speeds ranging from 1 to 50 m s -1 . Owing to limitations in the
eleetrieal and mechanieal system, the ribbon ean be vibrated only in the range
15-300 Hz. Also, so thai the ribbon does not signifieantly block the flow, it is
desired that the boundary layer thiekness should be at least 5 mm. Show that the
experiment ean be performed within these eonstraints, but only by varying the
distanee of the ribbon from the leading edge. Determine the range of variation
needed.

65. The flow in a pipe is in the regime deseribed in Seetion 19.3 in whieh turbulent
plugs are produeed periodieally. The pressure differenee is such that, if the flow
were entirely laminar, the Reynolds number would be 9000. A plug is generated
elose to the pipe entry whenever the Reynolds number rises above 6000. Use the
informatioll in Figs. 2.11 and 19.1 0 to estimate the fraetion of the flow that is
turbulent (a) when a plug is just being generated, (b) when the rear of a plug is
passing out of the pipe, and (e) when the front of a plug is passing out of the pipe.
Estimate also the range over whieh the Reynolds number oseillates during the eyele.
(Ignore entry length effeets on laminar and turbulent flow properties.)

66. Formulate an equation for the energy of the mean motion in a turbulent flow
(by multiplying equation (20.13) by Uj ) and rearrange the term representing the
interaetion with the fluetuations to exhibit (a) a loss term equivalent to the pro-
duetion term in the turbulenee energy equation, and (b) a te rm representing energy
transport by the turbulenee (integrating to zero over the whole flow).

67. The solar radiation falling on the planet Venus has an intensity of
2.6 x 10 3 W m -2. About 70 per eent of this is refleeted and about 1 per eent
reaehes the planet's surfaee. The remainder is absorbed in a c10ud layer, 15 to 30 km
thick. Cellular eonveetion is observed in the sub-solar region of this c10ud layer.
Suppose that such eonveetion ean oeeur only if the Rayleigh number of Seetion
17.3 is in the range 10 3 to 10 6 , and that it is brought into this range by small-seale
turbulence producing an effeetive kinematie viseosity and thermal diffusivity (equal
to one another). In what range must this eddy viseosity lie? (For order ofmagnitude
purposes, suppose that the atmosphere has the properties of CO2 at 300 K and
10 5 N m -2, and that g is the same as for the earth.)
344 Problems

68. Show that equation (21.1) for the eorrelation funetions in isotropie turbulenee
implies that

f~ gr dr=O
provided that f -+ 0 more rapidly than r -2 as r -+ oo. Indicate why one would expeet
this resuIt from eontinuity eonsiderations.

69. (i) The energy dissipation in isotropic turbulenee (denoted by € in Seetion

21.3) depends only on the Lm.s. veloeity fluetuation (U 2)1/2 and a length seale lof
the large eddies. Use dimensional analysis to infer the form of this dependenee.
(ii) The length seale 1/ of the dissipating eddies (the Kolmogoroff length seale)
depends only on the energy dissipation and the kinematie viseosity. Use dimensional
analysis to infer the form of this dependenee.
(iii) Supposing that one requires 1//1 < 10 -3 for the inertial sub-range to be
observable (so that there is a deeade in wave number between 10/1 and 1/101/),
express this as a eondition on the Reynolds number based on (U 2)1I2 and I.
(Assume that the proportionality eonstants involved in parts (i) and (ii) are ~ 1.)
(iv) With the observations that (Ui )1I2 ~ U/30 and I ~ M/3, where U is the
mean veloeity and M is the grid mesh length, determine the minimum UM/v for an
experiment to observe the inertial sub-range. Show that, with the eondition that
the flow must be ineompressible, the experiment requires a very large wind-tunnel.

70. Determine the pattem of two-dimensional, irrotational, ineompressible motion

pro du ee d when a spatially sinusoidal normal veloeity is imposed on the boundary
of a semi-infinite region of fluid; that is the boundary eonditions are
v= Vo eos kx at y = 0
U, v -+ 0 asy -+ oo
This analysis is the starting point for a theory of the irrotational velocity
fluetuations in fluid adjaeent to a turbulent region. Explain why analysis of the
response to a sinusoidal disturbanee is relevant to this, and show that the resulting
motion does not generate a Reynolds stress.

71. (i) If the velocity and length seales of an axisymmetrie jet are of the form

what are m and n for (a) laminar flow (b) turbulent flow?
(ii) If the velocity, length and temperature differenee seales of an axisymmetrie
plume are of the form
u max cx: x m ; (Tmax - To) cx: x P ; t:. cx: x n
what are m, p, and n for (a) laminar flow (b) turbulent flow?

72. The ratio UT/UO of the wall stress velocity to the free-stream velocity for a
turbulent boundary layer on a smooth wall varies with Reynolds number sufficiently
slowly that it may be taken as typically 1/30. Estimate the total skin friction (Le.
Problems 345

that part of the drag due directly to viscous forces on the surface) acting on the
wings of the aircraft in Question 41.
If the engines have a power of 2000 kW, compare the above drag with the total
drag on the aircraft.
For the assumption that the wings are smooth to be valid, any roughnesses must
be within the viscous sub-Iayer. How small does this require them to be?

73. In turbulent flow with a boundary that supports no stress (e.g. a free water
surface), the stress varies linearly with distanee from the boundary,
r=pay

Use arguments analogous to those in Section 22.5 to show that the counterpart of
the logarithmic profile (22.18) is
Us - U=A(ay)1/2 + B(av)1/3
where Us is the velocity of the surface and A and B are numerical constants.
(Hint: show that the velocity profile given by assuming that au/ay depends
only on a and y gives an absurd viscous stress at y = 0; hence, infer that the region
of applicability may have its velocity origin shifted by an amount that depends on
a and v.)

74. Evaluate the Monin-Obukhov length L for the atmospheric boundary layer
over smooth fiat ground for conditions in which the wall stress velocity
UT =0.3 m s -1 and the vertieal heat flux H =200 W m -2 . What are the gradients
of the mean velocity and mean temperature at a height of L/lO? (Check that the
adiabatic gradient may be neglected in these considerations.)

75. (i) Formulate the equation for the mean temperature in a turbulent boundary
layer type flow (Le. apply to equation (13.12) - with J = 0 - the procedure leading
from the Navier-Stokes equation to equation (20.14». Show that the turbulence
produces a heat transfer across the flow proportional to w8, in the notation of
Seetion 22.8. Consider the physieal signifieanee of non-zero w8.
(ii) Rederive equation (20.21) for a flow in which buoyaney forees are
signifieant. Show that for a horizontal mean flow, there is an additional term,
proportional to the vertieal heat transfer, representing a produetion or removal of
turbulenee energy depending on whether the flow is unstably or stably stratified.
(üi) Show that if the Reynolds stress is assumed proportional to the mean
veloeity gradient and the heat transfer to the mean temperature gradient, then the
ratio of the additional term in (ii) to the produetion term,- uwau/az, is essentially
th~ Riehardson number.
Bibliography and References

I Introductory reading

This list indudes books which give a good elementary aeeount of fluid dynamieal
topies in the context of an applieation.
I Cottrell, A. H. (1964) The Mechanieal Properties of Matter (Wiley).
2 Sutton, O. G. (1965) Mastery of the Air (Hodder & Stoughton).
3 Hidy, G. M. (1967) The Winds (Van Nostrand).
4 Goody, R. M. and Walker, J. C. G. (1972) Atmospheres (Prentiee-Hall).
5 Little, N. C. (1967) Magnetohydrodynamics (Van Nostrand).
6 Sh apiro, A. H. (1961) Shape and Flow (Heinemann).

II elassical tex ts

Referenees [71 and [81 together still serve as soureebooks for work done up to the
1930s, the former primarily for theory and the latter primarily for experimental
work. The other books, of whieh Refs. [9] and [11] are also primarily
experimental, are less eomprehensive but full of useful insights.
7 Lamb, H. (1932 - 6th edn.) Hydrodynamics (Cambridge/Dover).
8 Goldstein, S. (ed.) (1938) Modern Developments in Fluid Dynamics, 2 vols
(Oxford/Dover).
9 Prandtl, L. (1952) Essentials of Fluid Dynamics (Blaekie).
10 Prandtl, L. and Tietjens, O. G. (1934) Fundamentals of Hydro- and Aero-
mechanics (MeGraw-Hill/Dover).
II Prandtl, L. and Tietjens, O. G. (1934) Applied Hydro- and Aeromechanics
(MeGraw-Hill/Dover ).

III Theoretical texts

Of the books on fluid dynamies written for applied mathematieians or theoretieal

physicists, the following may be the most useful to readers of the present book.
12 Batehelor, G. K. (1970) An Introduction to Fluid Dynamics (Cambridge).
13 Curle, N. and Davies, H. J. (1968) Modern Fluid Dynamics, Vol. 1: Incom-
pressible Flow (Van Nostrand Reinhold).
14 Hunt, J. N. (1964 )Incompressible Fluid Dynamics (Longmans).
15 Landau, L. D. and Lifshitz, E. M. (1959) Fluid Mechanics (Pergamon).

IV Engineering texts

Many books on fluid dynamies have been written for engineers; the following are
among those giving extensive treatment of basie fluid dynamies.
Bibliography and References 347

16 Daily, J. W. and Harleman, D. R. F. (1966) Fluid Dynamics (Addison-

Wesley).
17 Duncan, W. J., Thom, A. S. and Young, A. D. (1970 - 2nd edn.) Mechanics
of Fluids (Arnold).
18 Li, W.-H. and Lam, S.-H. (1964) Principles of Fluid Mechanics (Addison-
Wesley).
19 Massey, B. S. (1975 - 3rd edn.) Mechanics of Fluids (Van Nostrand
Reinhold).

V Other general texts

(References [2.11 and [221 are broader in their scope than is suggested by their
titles.)
20 Richardson, E. G. (1961 - 2nd edn.) Dynamies of Real Fluids (Arnold).
21 Rosenhead, L. (ed.) (1963) Laminar Boundary Layers (Oxford).
22 Schllehting, H. (1968 - 6th edn.) Boundary Layer Theory (McGraw-Hill).

VI Books on partieular branehes of [luid dynamies

The books in this section contain introductory material and could be read alongside
the present book. (See also Refs. [211 and [22].)
23 Aris, R. (1962) Veetors, Tensors, and the Basie Equations of Fluid Meehanies
(Prentice-HalI).
24 Pankhurst, R. C. (1964) Dimensional Analysis and S eale Faetors (Chapman
and Hall).
2$ Massey .. B. S. (1971) Un its, Dimensional Analysis, and Physical Similarity
(Van Nostrand Reinhold).
26 Bradshaw, P. (1964) Experimental Fluid Meehanics (Pergamon).
27 Scorer, R. S. (1958) Natural Aerodynamics (Pergamon).
28 Hess, S. L. (1959) Introduction to Theoretieal Meteorology (Holt/Constable).
29 Bradshaw, P. (1971) Introduction to Turbulence and its Measurement
(Pergamon).
30 Tennekes, H. and Lumiey, J. L. (1972) A First Course in Turbulence (M.I.T.
Press).
31 Monin, A. S. and Yaglom, A. M. (1971) Statistical Fluid Mechanies: Meeh-
anics of Turbulenee (M.I.T. Press).
32 Merzkirch, W. (1974) Flow visualization (Academic Press).

VII Further books on partieular branehes

The books here are more speciallzed than those aboveo

33 Turner, J. S. (1973) Buoyaney Effeets in Fluids (Cambridge).
34 Yih, C.-S. (1965) Dynamies of Nonhomogeneous Fluids (Macmillan).
35 Greenspan, H. P. (1968) The Theory of Rotating Fluids (Cambridge).
36 Lin, C. C. (1955) The Theory of Hydrodynamie Stability (Cambridge).
37 Chandrasekhar, S. (1961) Hydrodynamic and Hydromagnetie Stability
(Oxford).
38 Hinze, J. O. (1975 - 2nd edn.)'Turbulence (McGraw-Hill).
39 Batchelor, G. K. (1953) The Theory of Homogeneous Turbulenee
(Cambridge).
348 Bibliography and References
40 Townsend, A. A. (1976 - 2nd edn.) The Strueture of Turbulent Shear Flow
(Cambridge).

VIII Films

Many fluid dynamiea1 phenomena are, of eourse, mueh better illustrated by cine-
films than by still photographs. Of the va1uable series of films produeed by the
National Committee for Fluid Meehanies Fllms, and distributed by Eneyclopaedia
Brittaniea Edueational Corporation, the following are the most relevant to this book.
Manyextraets from these films are also avallable as short film loops.
41 Shapiro, A. H. Vorticity.
42 Taylor, G. I. Low Reynolds Number Flows.
43 Abernathy, F. H. Fundamentals of Boundary Layers.
44 Fultz, D. Rotating Flows.
45 Long, R. R. Stratified Flow.
46 Mollo-Christensen, E. L. Flow Instabilities.
47 Stewart, R. W. Turbulenee.
48 Kline, S. J. Flow visualization.

IX Specifie referenees

49 Akiyama, M., Hwang, G. J. and Cheng, K. C. (l971)J. Heat Transfer 93,335.

50 Anderson, A. B. C. (1956)J. Aeoust. Soe. Amer. 28,914.
51 Angrist, S. W. (1964) Sei. Amer. 211,6,81.
52 Badri Narayanan, M. A. (1968) J. Fluid Mech. 31,609.
53 Badri Narayanan, M. A. and Ramjee, V. (1969) J. Fluid Meeh. 35, 225.
54 Bahng, J. and Sehwarzsehild, M. (1961) Astrophys. J. 134, 312.
55 Baker, D. J. (1966) J. Fluid Meeh. 26,573.
56 Bankston, C. A. (1970) J. Heat Transfer 92,569.
57 Barkla, H. M. and Auehterlonie, L. J. (1971) J. Fluid Meeh. 47,437.
58 Bearman, P. W. (1969) J. Fluid Meeh. 37, 577.
59 Beeker, H. A. and Massaro, T. A. (1968)J. Fluid Meeh. 31,435.
60 Berger, E. (1964) Z. Flugwiss. 12,41.
61 Berger, E. (1964) Jahrbueh 1964 WGLR, 164.
62 Berger, E. and Wille, R. (1972) Ann. Rev. Fluid Meeh. 4,313.
63 Billing, B. F. (1964) Thermoeouples: Their Instrumentation, Seleetion, and
Use (Inst. Engng. Inspeetion).
64 Binnie, A. M. and Fowler, J. S. (1947) Proe. Roy. Soe. A 192,32.
65 Blackman, R. B. and Tukey, J. W. (1959) The Measurement of Power Spectra
(Dover).
66 Bloor, M. S. (1964) J. Fluid Meeh. 19, 290.
67 Bradbury, L. J. S. and Castro, I. P. (1971) J. Fluid Meeh. 49,657.
68 Bradshaw, P. (1967)J. Fluid Meeh. 30,241.
69 Bradshaw, P. (1969) Nat. Phys. Lab., Aero. Rep. 1287.
70 Bradshaw, P., Ferriss, D. H. and Atwell, N. P. (1967) J. Fluid Meeh. 28,593.
71 Bray, R. J. and Loughhead, R. E. (1967) The Solar Granulation (Chapman
and Hall).
72 Brind1ey, J. (1967) J. Inst. Maths. Applies. 3,313.
73 Browand, F. K. and Winant, C. D. (1972) Geophys. Fluid Dynam. 4,29.
74 Brown, R. A. (1970)J. Atmos. Sci. 27, 742.
75 Central Eleetrieity Generating Board, Report of the Committee of Inquiry
into Collapse of Cooling Towers at Ferry bridge 1 November 1965.
Bibliography and Referenres 349

76 Cheesewright, R. (1968) J. Heat Transfer 90, 1.

77 Chu, T. Y. and Goldstein, R. J. (1973) J. Ffuid Mech. 60, 141.
78 C1ark, R. B. and Tritton, D. J. (1970)J. Zool. 161,257.
79 C1ark, R. P. (1973)J. Hyg. 71,577.
80 9ayton, B. R. and Massey, B. S. (1967)J. Sci., Instrum. 44,2.
81 Co1ak-Antic, P. (1964) Sitzungsberichte der Heidelberger Akademie der
Wissenschaften, Mathem.-naturw. Klasse, Jahrg. 1962/4, 6. Abhandlung
(Springer),313.
82 Coles, D. (1965) J. Fluid Mech. 21,385.
83 Cooper, B. J. (1968) Sci. J. 4, 12, 52.
84 Corrsin, S. (1949) J. Aero. Sci. 16, 757.
85 Corrsin, S. (1959) J. Geophys. Res. 64, 2134.
86 Corrsin, S. (1961) Amer. Sci. 49,300.
87 Cox, R. N. and C1ark, R. P. (1973) Rev. Gen. Therm. 12, 11.
88 Davies, J. M. (1949)J. AppI. Phys. 20,821.
89 Davies, P. A. (1972) J. Fluid Mech. 54, 691.
90 Davis, R. E. (1969) J. Fluid Mech. 36, 127.
91 Deb1er, W. R. (1966) J. Fluid Mech. 24, 165.
92 Deb1er, W. and Fitzgerald, P. (1971) Uni!'. Michigan, Dept. Engng. Mech.,
Tech. Rep. EM-71-3.
93 De la Cruz Reyna, S. (1970) Geofis. Int. 10,49.
94 Delany, N. K. and Sorenson, N. E. (1953) Nat. Adv. Comm. Aero., Tech.
note 3038.
95 Dennis, S. C. R. and Chang, G.-Z. (1970) J. Fluid Mech. 42,471.
96 Dhawan, S. (1952) Nat. Adv. Comm. Aero., Tech. note 2567.
97 Dhawan, S. and Narasimha, R. (1958) J. Fluid Mech. 3,418.
98 Donnelly, R. J. and Fultz, D. (1960) Proc. Roy. Soc. A 258,101.
99 Douglas, H. A., Hide, R. and Mason, P. J. (1972) Quart. J. Roy. Met. Soc. 98,
247.
100 Elder, J. W. (1960)J. Fluid Mech. 9,235.
101 Elder, J. W. (1965) J. Fluid Mech. 23,77.
102 Elder, J. W. (1965)J. Fluid Mech. 23,99.
103 Faller, A. J. and Kaylor, R. (1967) Phys. Fluids 10, S 212.
104 Fiedler, H. and Head, M. R. (1966)J. Fluid Mech. 25,719.
105 Finn, R. K. (1953) J. AppI. Phys. 24, 771.
106 Freund, D. D. and Meyer, R. E. (1972)J. Fluid Mech. 54,719.
107 Freymuth, P. (1966)J. Fluid Mech. 25,683.
108 Fultz, D. (1961) Adv. Geophys. 7, 1.
109 Fultz, D. and Nakagawa, Y. (1955) Proc. Roy. Soc. A 231,211.
110 Gass, I. G., Smith, P. J. and Wilson, R. C. (eds.) (1971) Understanding the
Earth (Artemis).
111 Gebhart, B. (1969) J. Heat Transfer 91,293.
112 Gebhart, B. (1973) Ann. Rev. Fluid Mech. 5,213.
113 Gerrard, J. H. (1965) J. Fluid Mech. 22, 187.
114 Gerrard, J. H. (1966)J. Fluid Mech. 25,401.
115 Gill, A. E. and Davey, A. (1969) J. Fluid Mech. 35,775.
116 Goedde, E. F. and Yuen, M. C. (1970) J. Fluid Mech. 40,495.
117 Graebe1, W. P. (1969) Quart. J. Mech. AppI. Math. 22,39.
118 Grant, H. L. (1958) J. Fluid Mech. 4, 149.
119 Grant, H. L., Stewart, R. W. and Moilliet, A. (1962) J. Fluid Mech. 12, 241.
120 Grant, R. P. and Midd1eman, S. (1966) Amer. Inst. Chem. Engrs. J. 12,669.
121 Grass, A. J. (1971)J. Fluid Mech. 50,233.
122 Gurvich, A. S. and Yaglom, A. M. (1967)Phys. Fluids 10, S 59.
123 Happe1, J. and Brenner, H. (1965) Low Reynolds Number Hydrodynamics
(Prentice-HalI).
350 Bibliography and References
124 Head, M. R. and Surrey, N. B. (1965)J. Sci. Instrum. 42, 349.
125 Hess, F. (1968) Sci. Amer. 219,5,124.
126 Hide, R. and Ibbetson, A. (1966)Icarus S, 279.
127 Hide, R., Ibbetson, A. and Lighthill, M. J. (1968)J. Fluid Mech. 32,251.
128 Hide, R. and Mason, P. J. (1975) Adv. Phys. 24,47.
129 Holder, D. W. and North, R. J. (1963) Nat. Phys. Lab., Notes AppI. Sci. J1.
130 Howard, L. N. (1963)J. Appi. Mech. 30,481.
131 Humphreys, J. S. (1960)J. Fluid Mech. 9,603.
132 Ibbetson, A. and Phillips, N. (1967) Tellus 19,1.
133 Jeffreys, H. (1961) Cartesian Tensors (Cambridge).
134 Kim, H. T., Kline, S. J. and Reynolds, W. C. (1971) J. Fluid Mech. SO, 133.
135 Kirschner, J. M. (ed.) (1966) Fluid Amplifiers (McGraw-Hill).
136 Klebanoff, P. S. (1955) Nat. Adv. Comm. Aero., Rep. 1247.
137 Klebanoff, P. S. and Tidstrom, K. D. (1959) Nat. Adv. Comm. Aero., Tech.
note D-195.
138 Klebanoff, P. S., Tidstrom, K. D. and Sargent, L. M. (1962) J. Fluid Mech.
12,1.
139 KIine, S. J. et al. (1967) J. Fluid Mech. 30, 741.
140 KIine, S. J. et al. (eds.) (1969) Proc. 1968 AFOSR-IFP-Stanford Conf.
Computation of Turbulent Boundary Layers, Mech. Engng. Dept., Stanford
Univ.
141 KIine, S. J. and Shapiro, A. (1965) Film loop FM-48 Pathlines, Streaklines
and Streamlines in Unsteady Flow (extract from Ref. [48J).
142 Komoda, H. (1967) Phys. Fluids 10, S 87.
143 Koschmieder, E. L. (1966) Beitr. Phys. Atmos. 39, 1.
144 Koschmieder, E. L. (1967) J. Fluid Mech. 30,9.
145 Koschmieder, E. L. (1974) Adv. Chem. Phys. 26, 177.
146 Kovasznay, L. S. G., Kibens, V. and Blackwelder, R. F. (1970) J. Fluid Mech.
41,283.
147 Kovasznay, L. S. G., Komoda, H. and Vasudeva, B. R. (1962) Proc. 1962
Heat Transfer and Fluid Mech. Inst., Stanford, 1.
148 Kraus, W. (1955) Messungen des Temperatur- und Geschwindigkeitsfeldes bei
freier Konvektion (Braun, Karlsruhe).
149 Krishnamurti, R. (1970) J. Fluid Mech. 42,309.
150 Ktjshnamurti, R. (1973) J. Fluid Mech. 60, 285.
151 Kuettner, J. P. (1971) Tellus 23, 404.
152 Kurzweg, H. (1933) Ann. Phys. (5) 18, 193.
153 Lachmann, G. V. (ed.) (1961) Boundary Layer and Flow Control, 2 vols (Pergamon).
154 Laufer, J. (1950) J. Aero. Sci. 17,277.
155 Laws, P. and Stevenson, T. N. (1972) J. Fluid Mech. 54,745.
156 Leconte, J. (1858) Phil. Mag. (4) 15,235.
157 Leighton, R. B. (1963) Ann. Rev. Astron. Astrophys. 1,19.
158 LeMone, M. A. (1973)J. Atmos. Sci. 30, 1077.
159 Leontiev, A. I. and Kirdyashkin, A. G. (1968)Internat. J. Heat Mass Transfer
11, 1461.
160 Lewis, H. E. et al. (969) Lancet 1, 1273.
161 Liebster, H. (1927) Ann. Phys. (4) 82, 541.
162 Liepmann, H. W. (1943) Nat. Adv. Comm. Aero., Wartime Rep. ACR JHJO.
163 Lighthill, M. J. (1966) J. Fluid Mech. 26,411.
164 Lighthill, M. J. (1968) in McDowell, D. M. and Jackson, J. D. (eds.) Osborne
Reynolds and Engineen'ng Science To-day (Manchester Univ. Press) 83.
165 Lighthill, M. J. (1969) Ann. Rev. Fluid Mech. 1,413.
166 Lindgren, E. R. (1957)Ark. Fys. 12, 1.
167 Lindgren, E. R. (959)Ark. Fys. 15,97.
Bibliography and References 351
168 Lindgren, E. R. (1959) Ark. Fys. 15,503.
169 Lindgren, E. R. (1969) Phys. Fluids 12, 418.
170 Long, R. R. (1955) Tellus 7, 341.
171 Long, R. R. (1972) Ann. Rev. Fluid Mech. 4,69.
172 Lumiey, J. L. and Panofsky, H. A. (1964) The Structure of A tmospheric
Turbulence (Interscience).
173 Maccoll, J. W. (1928) J. Roy. Aero. Soe. 32, 777.
174 Mason, B. J. (1971) Nature 233,382.
175 Mattioli, E. (1959) Atti Acad. Sci. Torino 94, 855 (trans1ated as Adv. Group
Aero. Res. Dev., Rep. 263).
176 Maxworthy, T. (1970) J. Fluid Mech. 40,453.
177 Merk, H. J. (1958) AppI. Sci. Res. A 8, 100.
178 Miles, C. A. and Holwill, M. E. J. (1969)J. Exp. Biol. 50,683.
179 Milne-Thompson, L. M. (1962 - 4th edn.) Theoretieal Hydrodynamics
(Macmillan).
180 Ministry of Technology (1968) Hydraulics Research 1967, Rep. Hyd. Res.
Statian (HMSO) p. 24.
181 Ministry of Techno1ogy (1970) Hydraulics Research 1969, Rep. Hyd. Res.
Station (HMSO) p. 11.
182 Moore, D. W. and Saffman, P. G. (1969) Phil. Trans. Roy. Soe. A 264,597.
183 Moore, F. K. (ed.) (1964) High Speed Aerodynamics and Jet Propulsion,
Vol. IV Theory of Laminar Flows (Oxford).
184 Mori, Y. and Uchida, Y. (1966)Jnternat. J. Heat Mass Transfer 9,803.
185 Mowbray, D. E. and Rarity, B. S. H. (1967)J. Fluid Mech. 28, 1.
186 Moylan, M. J. (1971 - 2nd edn.) Fluid Logic in Simple Terms (Machinery
Publishing Co.).
187 Nachtsheim, P. R. (1963) Nat. Adv. Comm. Aero., Tech. note D-2089.
188 Narahari Rao, K., Narasimha, R. and Badri Narayanan, M. A. (1971) J. Fluid
Mech. 48,339.
189 Nicholl, C. I. R. (1970) J. Fluid Mech. 40, 361.
190 Nie1d, D. A. (1964)J. Fluid Mech. 19,341.
191 Orszag, Z. A. (1970)J. Fluid Mech. 41,363.
192 Ostrach, S. (1953) Nat. Adv. Comm. Aero., Rep. 1111.
193 Owen, P. R. (1971) Quart. J. Roy. Met. Soe. 97,396.
194 Pantulu, P. V. (1962) [Link]. thesis, Dept. Aero. Engng., Indian Inst. Sci.,
Banga1ore.
195 Pao, Y.-H. (1967) Boeing Sci. Res. Labs., Flight Sci. Lab. Rep. 102 (document
Dl-82-0488).
196 Penner, S. S. and Jerskey, T. (1973) Ann. Rev. Fluid Mech. 5,9.
197 Perry, A. E. and Morrison, G. L. (1971)J. Fluid Meeh. 47, 577.
198 Pfenninger, W. (1961) in Ref. [153], p. 970.
199 Phillips, O. M. (1955) Proc. Camb. Phil. Soe. 51, 220.
200 Phillips, O. M. (1963)Phys. Fluids 6, 513.
201 Pilsbury, R. K. (1969) Clouds and Weather (Batsford).
202 Platzman, G. W. (1968) Quart. J. Roy. Met. Soe. 94,225.
203 Po1ymeropoulos, C. E. and Gebhart, B. (1967) J. Fluid Mech. 30, 225.
204 Preston, J. H. (1972)J. Phys. E, Sci. Instrum. 5,277.
205 Priestley, C. H. B. (1959) Turbulent Transfer in the Lower Atmosphere
(Univ. Chicago Press).
206 Ramdas, L. A. and Malurkar, S. L. (1932)Jnd. J. Phys. 7, 1.
207 Reyno1ds, O. (1883) Phil. Trans. Roy. Soe. 174,935.
208 Rosenhead, L. (1954) Prac. Roy. Soe. A 226, 1.
209 Roshko, A. (1961)J. Fluid Meeh. 10,345.
210 Rossby, H. T. (1969) J. Fluid Mech. 36,309.
352 Bibliography and References
211 Rotta, J. (1956)Ing. Arch. 24,258.
212 Rotta, J. C. (1962) Prog. Aero. Sci. 2, 1.
213 Sachs, P. (1972) Wind Forces in Engineering (Pergamon).
214 SaIter, C., Warsap, J. H. and Goodman, D. G. (1965) Aero. Res. Counc., Rep.
and Memo. 3365.
215 Sandborn, V. A. (1959)J. Fluid Mech. 6,221.
216 Sato, H. (1959)J. Phys. Soc. Japan 14, 1797.
217 Sato, H. (1960) J. Fluid Mech. 7, 53.
218 Sato, H. and Sakao, F. (1964)J. Fluid Mech. 20,337.
219 Savulescu, St. N. (1968) Tranzitia de la Scurgerea Laminara la cea Turbulenta
(Ed. Acad. Rep. Soc. Romänia).
220 Schmidt, E. (1932) Forsch. Geb. IngWes. 3, 181.
221 Schmiedel, J. (1928)Phys. Z. 29, 593.
222 Schraub, F. A. and KIine, S. J. (1965) Stanford Univ., Dept. Mech. Engng.,
Thermosciences Div., Rep. MD-12.
223 Schubauer, G. B. (1954)J. AppI. Phys. 23, 191.
224 Schubauer, G. B. and K1ebanoff, P. S. (1955) Nat. Adv. Comm. Aero., Rep.
1289.
225 Schubauer, G. B. and Skramstad, H. K. (1947)J. Res. Nat. Bur. Stand. 38,
251.
226 Schultz-Grunow, F. and Hein, H. (1955) Z. Flugwiss. 4,28.
227 Scorer, R. (1972) Clouds of the World (LothianjDavid and Charles).
228 Scriven, C. V. and Sterniing, L. E. (1964) J. Fluid Mech. 19, 321.
229 Shen, S. F. (1954)J. Aero. Sci. 21,62.
230 Sibu1kin, M. (1962) Phys. Fluids S, 280.
231 Silveston, P. L. (1958) Forsch. Geb. IngWes. 24,29 and 59.
232 Simon, G. W. and Weiss, N. o. (1968) Z. Astrophys. 69,435.
233 Smith, A. M. O. (1960) J. Fluid Mech. 7,565.
234 Smith, C. V. (1959) Met. Office, Met. Rep. 21.
235 Snyder, H. A. (1970)Internat. J. Non-linear Mech. S, 659.
236 Somerscales, E. F. C. and Dropkin, D. (1966)Internat. J. Heat Mass Transfer
9,1189.
237 Sparrow, E. M., Husar, R. B. and Goldstein, R. J. (1970) J. Fluid Mech. 41,
793.
238 Stevenson, T. N. (1968)J. Fluid Mech. 33,715.
239 Stuart,J. T.(1964)[Link]. 18,481.
240 Stuart, J. T. (1965) AppI. Mech. Rev. 18, 523.
241 Stuart, J. T. (1971) Ann. Rev. Fluid Mech. 3, 347.
242 Suhanek, P. C. and Lawrence, R. L. (1972) Z. angew. Math. Phys. 23,969.
243 Sutton, O. G. (1953) Micrometeorology (McGraw-Hill).
244 Szewczyk, A. A. (1962) Internat. J. Heat Mass Transfer S, 903.
245 Taneda, S. (1956) J. Phys. Soc. Japan 11, 302.
246 Taneda, S. (1965) J. Phys. Soc. Japan 20, 1714.
247 Taneda, S., Honji, H. and Tatsuno, M. (1974) J. Phys. Soc. Japan 37, 784.
248 Tani, I. (1964) Prog. Aero. Sci. S, 70.
249 Tani, I. (1969) Ann. Rev. Fluid Mech. 1, 169.
250 Tanida, Y., Okajima, A. and Watanabe, Y. (1973)J. Fluid Mech. 61,796.
251 Tanner, L. H. (1966) J. Sci. Instrum. 43,878.
252 Tarling, D. H. and Tarling, M. P. (1971) ContinentaI Drift (BelljPeIican).
253 Tatro, P. R. and Mollo-Christensen, E. L. (1967)J. FIuid Mech. 28,531.
254 Tatsumi, T. (1952)J. Phys. Soc. Japan 7,489 and 495.
Bibliography and References 353

255 Tatsumi, T. and Kakutani, T. (1958) J. Fluid Mech. 4, 261.

256 Taylor, G. I. (1923) Prac. Roy. Soe. A 104, 213.
257 Taylor, G.1. (1923) Phil. Trans. Roy. Soe. A 223,289.
258 Taylor, G. I. (1937) J. Aero Sei. 4,311.
259 Taylor, G. I. (1938) Prac. Roy. Soe. A 164,15.
260 Taylor, G. I. (1952) Prac. Roy. Soe. A 214, 158.
261 Taylor, R. J. (1958)J. Sei. Instrum. 35,47.
262 Temple, G. (1958) An Introduction to Fluid Dynamies (Oxford).
263 Thompson, H. A. and Sogin, H. H. (1966) J. Fluid Meeh. 24,451.
264 Thorpe, S. A. (1969) J. Fluid Meeh. 39,25.
265 Thorpe, S. A. (1971)J. Fluid Meeh. 46,299.
266 Timme, A. (1957)Ing. Areh. 25,205.
267 Tombach, I. H. (1973) Rev. Sci. Instrum. 44, 14l.
268 Townsend, A. A. (1949) Aust. J. Sci. Res. A 2, 451.
269 Townsend, A. A. (1957) in Görtler, H. (ed.) Boundary Layer Research,
Prac. lUTAM Symp., Freiburg (Springer) p. 1.
270 Townsend, A. A. (1959) J. Fluid Meeh. 5,209.
271 Townsend, A. A. (1970) J. Fluid Meeh. 41, 13.
272 Townsend, A. A. (1972) J. Fluid Meeh. 55,209.
273 Tozer, D. C. (1967) in Gaskell, T.F. (ed.) The Earth's Mantle (Academic
Press), Ch. 11.
274 Tritton, D. J. (l959)J. Fluid Meeh. 6,547.
275 Tritton, D. J. (963) J. Fluid Meeh. 16,269.
276 Tritton, D. J. (1963) J. Fluid Meeh. 16,417.
277 Tritton, D. J. (1967)J. Fluid Meeh. 28,439.
278 Tritton, D. J. and Zarraga, M. N. (1967)J. Fluid Meeh. 30,21.
279 Turcotte, D. L. and Oxburgh, E. R. (1972) A nn. Rev. Fluid Mech. 4, 33.
280 Turner, J. S. (1969) Ann. Rev. Fluid Meeh. 1,29.
281 Turpin, J. L. (1972) Phys. Fluids 15,968.
282 Uberoi, M. S. (1963)Phys. Fluids 6,1048.
283 Van Atta, C. (I966)J. Fluid Mech. 25,495.
284 Vaziri, A. and Boyer, D. L. (1971)J. Fluid Meeh. 50,79.
285 Veronis, G. (1970) Ann. Rev. Fluid Mech. 2, 37.
286 Vest, C. M. and Arpaci, V. S. (1969) J. Fluid Mech. 36, 1.
287 Wa1she, "D. E. and Is1es, D. C. (1969) Nat. Phys. Lab., Aero. Special Rep. 023.
288 Warner, C. Y. and Arpaci, V. S. (1968)Internat. J. Heat Mass Transfer 11,
397.
289 Webb, J. E. (1973)J. Zool. 170,325.
290 Webster, C. A. G. (1964) J. Fluid Meeh. 19,221.
291 Werle, H. (1973) Ann. Rev. Fluid Mech. 5,361.
292 Wiese1sberger, C. (1921) Phys. Z. 22,321.
293 Willis, G. E. and Deardorff, J. W. (1965) Phys. Fluids 8, 2225.
294 Willis, G. E. and Deardorff, J. W. (1970)J. Fluid Mech. 44,661.
295 Wortmann, F. X. (1953) Z. angew. Phys. 5,201.
296 Zdravkovich, M. M. (1969) J. Fluid Mech. 37,491.
Index

Acceleration of fJuid partide, 47 Baroclinic waves, 182-183, 303

Acoustic promotion of instability, 229-230, Barotropic fJow, 141
317 Benard eelis, 33 - see also Benard convection
Active motion in turbulent boundary layer, Benard conveetion, 30-41, 148-149, 203,
284 206,214-218,238-239
Adiabatic temperature gradient, 139-141, applications, 306, 307
142,158,160,193 in shear, 184-185, 301-303
Advection term in temperature equation, 129 motion whenjust supercritical, 217-218
Ad verse pressure gradient, 107 rotating,218
Aerofoil, 96, 98,108-109,110-111,119- stability theory, 214-218
124,323 turbulent, 37-41, 287, 289
terminology,119-121 Bernoulli's equation, 96-100, 103,210,294
three-dimensional, 123 -124 Billows, 210-211, 301
two-dimensional, 120-123 Bistable, fJuidic, 315-316
Aeroplane, 18,98,111, 119,313-314 Blasius profile, 105-106
Aluminium powder fJow visualization, 300 stability, 220- 222
Ammonium chloride fJow visualization, 299 Blastocladiella,319
Amphioxus, swimrning, 318 Blocked region,
Arnplification of disturbance, 213-224 length,188-189
Arnplifier, turbulence, 316-317 structure, 190-191
Anernometers,294-297 Blocking,187-191
cup,297 Bluff bodies, drag, 109-110
electrokinetic, 297 Body force, 51, 128
heat-transfer, 297 Boomerang, 323
hot-film, 295 Boundary, force on, 55
hot-wire, 295, 321 Boundary conditions, 53-55, 130, 133
laser-Doppler,295-296 Boundary layer, 26-27, 84-86, 101-111,
Pitot tube, 294-295 131-132,138-139
quartz-fibre, 297 adverse pressure gradient, 106-109
vane, 297 approximation, 101-103,245,247
Angle of attack, 119 control,313-314
Anisotropy of turbulence, 246 equations, 103, 104-106, 111, 112-115
Annulus, convection in rotating, 182-183, favourable pressure gradient, 290
303 fIat plate, 103-106
Atmospheric circulation, 240-242, 303 free conveetion, 37, 142-148, 210-212,
Attached eddies, 19-21, 108 321
Attachment, wall, 284, 315 instability, 210-212, 220-228
Attack, angle of, 119 laminar two-dimensional, 101-106
Auto-correlation, 249-250, 250-251 reverse transition, 290
Averaging in turbulent fJow, 239,242-243 rotating fJuid, 171-176
ensemble, 242 separation, 27, 106-109, 228, 314
time, 242-243 stability theory, 220- 224
Averaging over molecules, 42-44 thermal, 37, 131-132, 142-146, 204
thickness, 84-85,106,131-132,
Background vorticity, 169 144-145
Baroclinic fJow, 141 transition, 27,225-228
 Index 355
transition, effeet of pressure gradient, 228 ineompressible flow, 46
turbulent, 245, 261-262, 275-284, 313 spherical polar eo-ordina tes, 52
turbulent stratified, 288-290 turbulent flow, 244, 254
vorticity in, 101 Continuum hypothesis, 2, 42-44
zero pressure gradient, 103-106 Conveetion, 127-161
Boussinesq approximation, 128-130, 134, Benard - see Benard eonveetion
135,139,155-161,306 eelis, 33-35, 203-206, 238-239, 307
eonditions for in free conveetion, 155 -161 eireular eylinder, 135-136
Brownian motion, 42 Earth's mantle, 306
Brunt-Väisälll frequeney, 192-195, 199, 215 free, 30-41,130,135-161 - see also
Buildings, wind effeets, 18,310-313 Free eonveetion
Bulk modulus, 2 foreed, 130-132
Buoyaney foree, 129, 135, 160,286 horizontallayers - see Benard eonveetion
Bursts in turbulent boundary layer, 282, 291 horizontal surfaee, 40-41
human body, 320-321
Caiculation methods, turbulent flow, 248 internai heat generation, 203-204, 306
Camber,119 mixed, 130, 132
Cartesian eo-ordina tes, 52 natural - see Free eonveetion
Cascade, energy, 256-260 rolls, 33, 184-185, 301
Cells, conveetion, 33-35, 203-206, rotating annulus, 182-183
238-239,307 - see also Benard rotating horizontallayer, 218
eonveetion surfaee tension variations, 204-206
in shear, 184-185, 301-303 verticallayer, 31, 149-154
Taylor, 207-209 vertical surfaee, 142-148, 222-224, 321
Centrifugal foree, 162-163 Cooling
Channel, water, 293, 297 foreed conveetion, 132
Channel flow free conveetion, 147-148
laminar flow, 6-9, 88-89 Newton's law, 132
reverse transition, 290-291 Cooling towers, 311-313
Chord of aerofoil, 121 Co-ordina tes
Cireular eylinder - see Cylinder, cireular Cartesian, 52
Cireulation, 72-73,120-122 eylindrical polar, 52
generation of, 122-123 Eulerian, 44-45
Kelvin theorem, 94, 122 Lagrangian, 44
Classical mechanies, 2 spherical polar, 52-53
Closure problem, 254-255 Coriolis foree, 162-164, 169
Cloud patterns, 301-303 Correlation, 243-244, 245-246, 248-250
Coanda effeet, 284-286, 315 auto, 249-250, 250-251
Coefficient of expansion, 32, 128, 129 boundary layer, 283-284
Coeffieient of viscosity, 3, 50, 60 - see also eoefficient, 243
Viseosity double, 243-244, 245-246, 248-250,
Column, liquid, instability of, 201-203 254,272-273,283-284
Compressibility, 2, 58, 306 grid turbulenee, 255
Compressible flow, 79 interpretation, 248-250,252
Coneentration variations, 132-134 isotropic turbulenee, 254
Conduetion, thermal, role in conveetion, 130 measurement, 295
Conduetion term in temperature equation, 129 pressure-velocity, 250
Conduetivity space, 248-249, 254-255, 272-273,
thermal, 32,129 283-284
thermometrie, 129 spaee-time, 250
Conservation of mass, 45, 61 - see also temperature-velocity, 287
Continuity equation triple, 244, 254-255
Constrietion, flow through, 99 wake, 272-273
Continental drift, 303-306 Couette flow, rotating, 89
Continuity equation, 45-46, 51 flow patterns, 207 - 209
boundary layer, 102 instability, 206-209, 218-220
Boussinesq approximation, 128, 156 iaminar flow theory, 89-90
Cartesian co-ordinates, 52 stability theory, 218-220
cylindrical polar co-ordinates, 52 Creeping motion, 82-83, 90-93
356 Index
Critical Rayleigh number, 32-33, 203, attached, 19-20,63,66,108
216-217, 306 dissipative, 257,258
Critical ReynoIds number large, 257, 272-275, 281-284, 289, 291
boundary layer, 222 small, 257, 258, 271
jet, 229 turbulent, 252 - see also Eddies, large;
pipe flow, 233, 235, 237 Eddies, small
Cup anemometer, 297 Eddy - see also Vortex
Cyclones, 240-242 shedding, 21, 24, 310-311
Cylinder, circular viscosity, 248
drag, 28-29,93 Effieiency, thermodynamic, 142
flow past, 18-29,63-66,82-83,86, Effluent dispersal, 308-310
92-93 Ekman layer, 171-176
heated, convection from, 135-136 injection, 176
inviscid flow, 86,109-110 instabitity, 173-175, 302
low Reynolds number flow, 19-20, suetion, 176
82-83, 92-93 thickness, 173
pressure distribution, 109-11 0 Ekman number, 164, 170
rotating, lift on, 124-125 Ekman spiral, 173-174
stratified flow, 187-190, 198-199 Electrokinetic veloeity measurement, 297
Cylinders, rotating, flow between - see Electrolytic preeipitation flow visualization,
Couette flow, rotating 299
Cylindrical polar co-ordinates, 52 Energy
balance, turbulent, 270-271, 278-279,
d'Alembert's paradox, 97-98,109 286
Deformation of fluid particle, 48-49, 59-60 cascade, 255-260
Density, 2, 56 dissipation - see Dissipation
effeet of concentration, 133 equation, 116
effect of pressure, 56-59 equation, turbuIent, 246-247
effeet of temperature, 56, 59,128 spectrum, 251, 256, 263
effeet on dynamics, 51, 56,128 transport, 11 7
Density variations, flows with, 128-129, turbulent, 243
184-186 - see also Free conveetion; Ensemble average, 242
Stra tified flow Entrainment, 115, 117-118, 266-268, 284
Diffusivity Entry length, 12-13, 233, 237
coneentration, 133, 134 Equation of creeping motion, 82, 90-92
thermal, 32, 129, 131 Equation of state, 56
Dirnensionless parameters, 11-12,76-80 Equations of motion, 42-60
Dissipation, 116-117, 159, 160 conveetion, 127-130
turbulent, 247,256-258 turb ulent , 244-247
Disturbance, amplification of, 213-214 Equilibrium, 31,149,214
Double correlation, 243-244, 245-246, IocaI,278
248-250,254,272-273,283-284 Eulerian co-ordinates, 44-45
Drag, 27-28, 77-79, 318-320 Euler's equation, 113, 87, 94
bluffbodies, 109-110 Expansion coeffieient, 32, 128, 129
circular cylinder, 28-29,93 Experirnental methods, 292-300
coeffieient, 28,77-79,109
crisis, 29
inviseid flow, 97 -98 FavourabIe pressure gradient, 107, 290
sphere, 28,91-92 Fish, 111,318-320
streamIined bodies, 110 Flame, instabitity of, 229
Dye, 298-299 Flat pIate
Dynamic pressure, 100 convection from, 40-41,142-148,
Dynamical equation - see Navier-Stokes 222-224, 321
equation flow past, 84-85, 103-106
Dynamical sirnilarity, 74-80,131,136,155, wake, 111-112
292 Flow visualization, 292-293, 298-300
Fluctuations in turbulent flow, 243
Earth's mantle, 303-306 Fluid control devices, 315-317
Eddies - see also Vortex; Vortiees Fluid particle, 42-44, 46
Index 357

aceeleration, 47 Heat transfer, 33, 127,132,138,147-148,

deformation, 48-49, 59-60 288,293
Fluidics, 315-317 anemometer, 297
Flume, water, 293 Heat transport, 148, 287, 288
Flux, mass, 9,10,61,293 - see also Mass High Reynolds number flow, 83-87
transport Homogeneous turbulence, 253-260
Foree Horizontallayer, eonveetion in - see Benard
body,51,128 eonveetion
buoyancy, 129, 135,160,286 Horizontal surfaee, eonveetion over, 40-41
eentrifugal, 162-163 Hot-film anemometer, 295
Coriolis, 162-164, 169 Hot-wire anemometer, 295-296, 321
inertia, 47-48, 81 Human body, eonveetion from, 320-321
on obstac1e - see Drag; Lift Hydrodynamie stability, theory of,
pressure, 8,48,81 213-224,233
viscous, 7-8, 48-50, 59-60, 81 Hydrogen bubble flow visualization, 299
Foreed conveetion, 130-132 Hydrostatie pressure, 48, 51, 128, 139, 156,
layer, 289 160
Frame of referenee, ehange, 18-19,63-64
Free conveetion, 30-41, 130, 135-161 lmpermeability eondition, 53
as heat engine, 141-142 Inaetive motion in turbulent boundary layer,
boundary layer, 37, 142-148, 210-212, 284
321 Ineompressibility, eondition for, 56-59
horizontallayers - see Benard eonveetion Ineompressible flow, 2
horizontal surfaee, 40-41 Inertia force, 47-48, 81
human body, 320-321 Inertial sub-range, 257 -258
internaI heat generation, 203-204, 306 Inertial waves, 177, 194
layer, 289 Infleetion in shear flow profile, 211, 221
verticallayer, 31,149-154 Imush in turbulent boundary layer,
vertical surfaee, 142-148,222-224,321 282-283
Free surfaee, 4, 79, 202, 204 Instability, 31, 149,201-224
Frequeney Benard, 31-33, 214-218
Brunt-Väisäll!, 192-195, 199,215 boundary layer, 210-212, 220-228
intrinsie of rotating fluid, 177, 199 Ekman layer, 173-175, 302
vortex street, 23, 79, 297, 310 flame,229
Frietion - see Viseosity internally heated layer, 203-204
Froude number, 79, 80 jet, 201-203, 210-212, 229-233
internaI, 130, 185-186, 199, 286,300, Kelvin-Helmholtz,211
308 liquid eolumn, 201-203
pipe now, 210-212, 278
rotating Couette now, 206- 209,
Gentian violet flow visualization, 299
218-220
Geometrical similarity, 74
shear now, 209-212
Geostrophie flow, 163-171
equation, 164, 170 surfaee tension varjations, 204- 206
vertieal plate conveetion, 145-147,
Granulation, solar, 306-308
210-212,222-224
Grashof number, 136, 142, 321
wake, 20-21, 111-112, 210-212
high, 138-139, 142
wake in stratified now, 186
physical significanee, 137-139
Intensity, turbulent, 243
Gravitational foree, 51, 128
boundary layer, 276-278
Grid turbulenee, 253, 255, 256
measurement, 295
Group velocity,
wake,269
inertial waves, 177
Interfaee, laminar-turbulent, 265-268
internaI waves, 194-195
Rossby waves, 180-181 Interferometers, 300
Intermitteney, 16,209,235-237,265-268,
Grllneisen's ratio, 161
291
factor, 16,236,265,267,269-270,278
Hagen-Poiseuille flow - see Poiseuille flow small-scale, 258
Heat engine, 32, 141-142 Internai Froude number, 130,185-186,
Heat generation, internaI, 129,203-204 199,286,300,308
358 Index
Internal heat generation, 129, 203-204 Logarithmie velocity profile, 280, 284, 289
InternaI waves, 192-198, 288 Low Reynolds number flow, 19-20, 28, 59,
Intrinsie frequency of rotating fluid, 177, 82-83,90-93
199
Inviscid flow, 71-72, 83, 94-100 Mach number, 58-59, 79, 95
Irrelevant motion in turbulent boundary Magnus effeet, 124-125
layer, 284 Manometers, 298
Irrotational flow, 67,94-96 Mantle of Earth, 303-306
permanence of, 94 Mass
Irrotational velocity fluctuations, 267-268 conservation, 45, 61 - see also Continuity
Isotropie turbulenee, 253-260 equation
flux,9,10,61,293
Jet, 112 transfer, 132-134
instability, 210-212, 229-231 transport, 115, 117-118
laminar two-dimensional, 112-118 Mean free path, 44
transition, 229-233, 317 Mean quantities - see Averaging in
turbulent, 245, 264-265, 268, 284, turbulent flow
308-310,315-317 Mean velocity equation, 245
width, 113-114, 264-265 Mereury, conveetion in, 35
with free surfaee, 201-203 Methyl blue flow visualization, 299
Jet-stream, 303 Micro-<>rganisms, 318, 321
Joukovski theorem, 122 Mixed conveetion, 130, 132
Model testing, 74-75, 80, 308-309,
Karman eonstant, 280, 289 311-313
Karman-Howarth equation, 254 Moleeular mean free path, 44
Karman vortex street, 21-26,111-112, Moleeules, averaging over, 42-44
212,231,238,297 Momentum, rate of ehange, 46, 47 -48
Kelvin cireulation theorem, 94,122 Momentum equation - see Navier-Stokes
Kelvin-Helmholtz instability, 211 equation
Kinematie viscosity, 51,131-132 Momentum transport, 28,114-117,264
Kinetie energy - see Energy Monin-Obukhov length, 289
Kolmogoroff law, 257 - 25 9, 271 Motion, equations of - see Equations of
Kutta-Zhukovskii theorem, 122 motion

Lagrangian eo-<>rdinates, 44
Laminar flow, 13,88-93 Natural conveetion - see Free eonveetion
boundary layer, 103-106 Navier-Stokes equation, 48-53, 88-90,
ehannel, 7-9, 88-89 102-103, 128
jet, 112-115 Cartesian eo-<>rdinates, 52,102-103
pipe, 9-11,88-89 eylindrical polar co-<>rdinates, 52
rotating Couette, 89-90 derivation, 48-50,59-60
Laneelet, swimming, 318 properties, 51,116-118
Laplaee's equation, 95-96 solutions, 88-90
Lapse-rate, adiabatie, 140 spherical polar co-<>rdinates, 52-53
Large eddies of turbulent flow, 257, turbulent flow, 244-245, 254
272-275,281-284,289,291 Newtonian fluids, 2-3, 50, 60
Large Reynolds number flow, 83-87 Newton's law of eooling, 132
Laser-Doppler anemometer, 295-296 Newton's second law, 46, 51
Lee waves, 191-192, 301 Non-dimensional pararneters, 11-12,76-80
Length-scales Non-Newtonian fluids, 3
flow,12,43,75,84,101-103,171, NOR elements, fluidie, 315-317
188-190 No-slip eondition, 7,53-55,83-84
fluid partide, 43, 44 NusseIt number, 33, 132, 138, 147,149
moleeular, 43, 44
turbulent flow, 249, 252, 284 Obstade, flow past, 18,74,85-87,95,100,
Lift, 98,119-126 101-102,109-111 - see also
Line vortex, 69 Cylinder; Sphere
Linear stability theory, 213-224, 233 in rotating fluid, 165-171
Local equilibrium, 278 in stratified fluid, 186-192
Index 359

One-dimensional speetrum, 251, 252, 256, Pressure, 7,48,51,95

259,263 dynamie,100
Onset of motion, 31-33, 201-206, hydrostatie, 48, 51,128,139,156,160
214-218 measurement,298
Optical methods of flow visualization, 300 stagnation, 100, 294
OR element, fluidie, 315-316 statie, 100, 294
Overstability, 218 total, 100,294
Pressure differenee scales, 57,102-103
Paraffin droplets flow visualization, Pressure foree, 8, 48,81
298-299 Pressure gradient, 7-11,16-17,48
Particle, fluid - see Fluid particle adverse,107
Particle path, 62, 65-66, 300 effeet on transition, 228
Pa thline - see Particle path favourable, 107,290
Peclet number, 131-132, 138 Probability distribution funetion, 243
Permanenee of irrotational motion, 94 Proudman-Taylor theorem, 164-171,176,
Perturbation, amplification of, 213-224 177-178
Phase velocity
inertial wa yes, 177 Quartz-fibre anemometer, 297
internai waves, 194
Rossby waves, 180-181
Photosphere of Sun, 306-308 Rate-of-strain tensor, 49-50, 59-60
pH-teehnique of flow visualization, 299 Rayleigh eriterion (Couette flow), 207, 219
Pipe flow, 6-7, 9-17, 88-89, 233-237 Rayleigh number, 31, 79,138,148-154
entry length, 12-13 critical, 32-33, 203, 216-217,306
laminar flow, 9-11, 88-89 Re-attaehment, 27,108,228,284
reverse transition, 290 Relaminarization,290-291
stability,233 Resistanee thermometers, 298
transition, 13-17,233-237 Resistanee to flow - see Drag; Viseosity
turbulent, 14, 16-17,275-277 Reverse transition, 290- 291
Pitot tube, 294-295 Reynolds number, 11-12, 19,75-80,131,
Planetary waves, 303-304 138,210-212,317,319-320
Planets, interiors of, 140, 160, 303-306 critical - see Critical Reynolds number
Plate, flat - see Flat plate high, 83-87
Plate tectonies, 306 low, 19-20,28,59,82-83,90-93
Plug, turbulent, 15, 234-237 physical signifieanee, 81-82
growth,235 role in dynamical similarity, 75-80,
periodie produetion, 236-237 131,311
Plume Reynolds number similarity, 257,261-265,
from point or line souree, 135-136, 148, 275
284-286,308-310 Reynolds stress, 245-246,247
in eonveetion above horizontal surfaee, reverse transition, 291
38-41,289 shear flow instability, 212
Poiseuille flow, 9-11,12,88-89 spectrum,263
departures from, 11-13 stratified flow, 287 - 288, 289
stability, 233 wake,269-270
Polar eo-ordinates, 52-53 wall flow, 275-276, 283, 289
Pollution,308-310 Riehardson number, 130, 185,286,289 -
Polystyrene beads for flow visualization, see also Internai Froude number
300 gradient form, 286, 289
Potassium permanganate flow visualization, Rigid body rotation, 68-70, 162-163
299 Robins effeet, 124-126
Potential Rolls, conveetion, 33, 184-185, 301
veetor veloeity, 67 Rossby number, 164, 170-171, 199
velocity,95 Rossby waves, 177 -182, 303
Potential flow, 94-97 Rotating annulus, eonveetion in, 182-183
Potential temperature, 140, 158 Rotating Benard conveetion, 218
Prandtl number, 34, 79,131-132, Rotating bodies, lift, 124-126
136-139,145,148-149,186 Rotating Couette flow - see Couette flow,
typical values, 132 rotating
360 Index
Rotating eylinders, flow between - see isotropic turbulence, 255-259
Couette flow, rotating measurement, 295
Rotating fluids, 162-183 one-dimensional, 251, 252, 256, 259, 263
analogy with stratified flow, 187 Reynolds stress, 263
stratified,199-200 tidal channel, 258-259
Rotation Speed - see Velocity
offluid particle, 67-70 Sphere
of frame of referenee, 162-163 drag, 28, 91-92
rigid body, 68-70,162-163 low Reynolds number flow, 90-92
rotating, lift on, 124-126
Salt, dissolved in fluid, 132-134, 184 stratified flow, 186
Sca le Spherical polar co-ordinates, 52-53
length - see Length-scales Spinning bodies, 124-126
pressure difference, 57,102-103 Spin-up, 176
temperature difference, 132, 136, 288 Spiral turbulence, 209
time,79,95,249,250 Spot, turbulent, 147,227-228,234
velocity, 12,57,75,102,279 Spreading of turbulence, 15,227,235,
Scale height, 160, 308 266-268
Schlieren, 300, 320-321 Squire's theorem, 220
Schmidt number, 134, 184 Stability, theory of, 213-224, 233 - see also
typical values, 134 Instability
Self-preservation, 263-264 Stable stratification, 184-200,210-211,
Separation, 19,27,106-110,123,228, 286-290,301,308-310
313-314 Stagnation points, 100
bubble, 108,228 Stagnation pressure, 100,294
consequences, 108-110, 123, 228 Stall, 108-109, 123, 314
prevention, 110,313-314 Stars, 140, 160, 306-308
Shadowgraph, 300 Starting vortex, 122-123, 124
Shear flow, 3, 69 State, equation of, 56
effeet on Benard conveetion, 184-185 Static pressure, 100,294
instabiJity, 209-212 Statistical description of turbulence, 239- 244
stratified, 210-211, 286-290, 301 Steady flow, definition, 5
turbulent, 245-247,261-291 Stokes flow, 90-92
viseaus action, 3,7-8 Stokes's paradox, 92
vorticity, 69-7 0 Stratification
Shear layers, free stable, 184-200, 210-211,286-290,301,
in rota ting fluid, 171 308-310
transition, 229-233 unstable, 184,286-287,289
Silicone oi!, 293-294 Stratified flow, 184-200
Similarity analogy with rotating fluids, 187
dynamical, 74-80,131,136,155,292 rotating, 199-200
geometrical, 74 shear flow instability, 210-211, 301
Reynolds number, 257,261-265,275 turbulent, 286-290, 308-310
thermal, 131, 136, 138, 155 Streakiine, 62, 66, 300
Small Reynolds number flow, 19-20,28,59, Stream function, 67,104,114
82-83,90-93 Streaming potential velocity measurement, 297
Smoke, 298-299 Streamiines, 61, 63-64
Solar granulation, 306-308 spacing of, 61-62
Sound, speed of, 2,58,95 Streamlining, 11 0-111
Space correlation, 248-249, 254-255, Streamtube,61-62
272-273,283-284 Stress
Space-time eorrelation, 250 Reynolds - see Reynolds stress
Span of aerofoil, 121 tensor, 49, 60
Specific heat, 129, 158-159 viseaus, 2-3, 7,48-50,59-60
Spectrum of turbulence, 250-251,255-259 wall, 55, 279
channel,263 Strouhal number, 23, 79
energy, 251,256,263 Structures, wind effects, 310-312,313
grid turbulence, 256 Sub-layer, viscous, 281,282
interpretation, 252 Substantive derivative, 46-48
Index 361

Suffix notation, 244 jet, 229-233, 317

Sun, 306- 308 - see also Stars pipe flow, 13-17,233-237
Surface tension instabilities, 201-203, wake, 23-26
204-206 Triple correlation, 244, 254-255
Swimming,318-320 Turbulence, 14, 238-291
amplifier, 316-317
Taylor eelis, 207-209, averaging, 239, 242-243
Taylor columns, 165 -171 correlations - see Correlation
effect of stratifieation, 199-200 energy, 243
longitudinal, 167-168, 187 equations, 244-247
transverse, 165 -167 grid, 253, 255,256
Taylor-Proudman theorem, 164-171, 176, homogeneous, 253-260
177-178 intensity, 243, 269, 276-278
Taylor's hypothesis, 250, 251 isotropic, 253- 260
Tellurium flow visualization, 299 nature of, 14,238-239
Temperature spectrum - see Spectrum
adiabatic gradient, 139-141, 142, 158, spiral,209
160,193 spreading, 15, 227,235,266-268
difference seale, 132, 136, 288 statistieal description, 239-244
equation, l29, 157-159 stress - see Reynolds stress
measurement, 298 Turbulent boundary layer, 245, 261-262,
potential, 140, 158 275-284,288-290,313
rate of change, 46-47 Turbulent convection, 37 -41,287, 289
Thermal, 38-41, 289 Turbulent flow, 14,238-291
Thermal conduction, role in convection, 130 ealculation methods, 248
Thermal conductivity, 32,129 energy balance, 270-271, 278-279,286
Thermaldiffusivity, 32, 129, 131 mean velocity equation, 245
Thermal flows, 127 -161 near wall, 275-284
Thermal similarity, 131,136, 138, 155 stratified, 286-290, 310
Thermocouples, 298, 321 Turbulentjet, 245, 264-265, 268, 284,
Thermodynamic efficiency, 142 308-310,315-317
Thermodynamics Turbulent pipe flow, 14, 16-17,275-277
first law, 142, 157 Turbulent plug, 15, 234-237
second law, 117,141-142 Turbulent shear flow, 245-247,261-291
Thermometers, 298 Turbulent spot, 147,227-228,234
Thermometric cond uctivity, 129 Turbulent wake, 23-26, 245,263,265-267,
Thickness 269-275
aerofoil,121 Two-dimensional flow, definition, 5
boundary layer, 84-85,106,131-132,
144-145 Undulatory swimming, 318-320
jet, 113-114,264-265 Universal motion in turbulent boundary
Thymol blue flow visualization, 299 layer, 284
Tidal channel, 259 Unstable stratifieation, 184, 286-287,289
Time-seale, 79, 95 Unsteady flow, 5, 62, 63-66, 95,98
turbulent flow, 249, 250 Upstream wake,188-191
Titanium tetrachloride flow visualization, 299
Tollmien-Schlichting waves, 222- 224, VlIislUli-Brunt frequency, 192-195,199, 215
225-228,229-233 Vane anemometer, 297
Tõpler-schlieren, 300 Variations of fluid properties, 160,290
Torque, 90, 293 Velocity
Total pressure, 100, 294 average over flow, 11
Towing tank, 293 Eulerian, 45
Trailing vortex, 123-l24 fluctuations in turbulent flow, 14,
Transition, reverse, 290- 291 242-244
Transition to turbulence, 225-237 fluid partide, 42 - 44
boundary layer, 27, 225-228 group - see Group velocity
free convection boundary layer, 145-147, maximum, 10, 113,264
321 mean in turbulent flow, 242- 243,245
free shear layer, 108,229-233 measurement, 294-297
362 Index
phase - see Phase velocity definition, 67
wall-stress, 279, 288 diffusion, 71, 101, 131-132
Velocity potential, 95 eq uation, 71- 72
veetOI,67 fluctuations, in turbulence, 260,266-268
Velocity profile physieal sjgnificance, 68-70
bloeked region, 190-191 production, 101
laminar boundary layer, 104-106
laminar ehannel flow, 9
laminar jet, 115 Wake,20, 108, 111-112, 198
laminar pipe flow, 10 formation, 27-28, 111
logarithmie, 280-281, 284, 289 instability, 20-21,111-112,210-212
stratified boundary layer, 289, 290 instability in stratified flow, 186-187
transitional boundary layer, 225 - 227 periodic, 21-26,111-112,186-187,212,
turbulent boundary layer, 275-276, 238,297
280-281,284 strueture, 111
turbulent pipe flow, 275-277 turbulent, 23-26, 245, 263, 265-267,
turbulent wake, 267 269-275
Veloeity seale, 12,57,75,102,279 upstream,188-191
Venturi tube, 99 Wall
Vertieallayer, eonveetion in, 31, 149-154 boundary eonditions at, 53-55
Vertieal surfaee, eonveetion from, 142-148, force on, 55
222-224,321 turbulent flow near, 275-284
Vibrating ribbon experiments, 223, 225 Wall attachment, 284, 315
Viscosity Wall region, 278-281
action of, 3, 7-8, 48-50, 59-60, Wall stress velocity, 279, 288
116-118,122-123 Water channel, 293, 297
eoeffieient, 3,50, 60 Water flume, 293
destabilizing action, 211-212 Water-mould,318
eddy,248 Wave-number
kinematie, 51,131-132 internai waves, 194--195
measurement, 10,89,90 Rossby waves, 179
second eoeffieient, 50, 60 turbulenee, 251, 252, 256-259
Viscous energy dissipation - see Dissipation Waves
Viscous flow equations, 48-53, 88-93 - see atmospheric circulation, 303-304
also Creeping motion; Navier-Stokes barodinic, 182-183,303
equation inertial, 177, 194
Viseous foree per unit volume, 8,50,60,81 internai, 192-198,288
Viscous stress, 3,48-50,60 lee, 191-192, 301
Viscous sub-layer, 281, 282 Rossby, 177-182, 303
Visualization, flow, 292-293, 298-300 swimming organism, 318-320
Vortex Tollmien-Schlichting,222-224,
generators,313-314 225-228,229-233
line,.68-70,71-72 Weather maps, 164,240-242,303
shedding, 21,24,310-311 Wind direction, variation with hejght,
starting, 122-123, 124 173-176
street, 21-26, 111-112, 186-187, 212, Wind effects on structures, 310-312, 313
231,238,297 Wind tunnel, 293, 297
stretching, 71, 169, 260 Wing, 111, 119, 313-314 - see also Aerofoil
trailing, 123-124
twisting, 169
Vortices, in jet instability, 230 Zero-pressure gradient boundary layer,
Vorticity, 60, 67-73,101,123,169,185,188 103-106
background, 169 Zhukovskii theorem, 122