Advances in Applied Mechanics

Volume 27
 Editorial Board
T. BROOKEBENJAMIN
Y. c. FUNG
PAULGERMAIN
RODNEYHILL
L. HOWARTH
C . 4 . YIH(Editor, 1971-1982)

Contributors to Volume 27
JOHN P. BOYD
RU LINGCHOU
C. K. CHu
V. I. FABRIKANT
JORG IMBEROER
JOHN C . PATTERSON
CHANG-LIN TIEN

VIGGO
TVERGAARD
KAMSIZ VAFAI
 ADVANCES 1N
APPLIED MECHANICS
Edited by
John W. Hutchinson Theodore Y. Wu
DIVISION OF APPLIED SCIENCES DIVISION OF ENGINEERING AND APPLIED SCIENCE
HARVARD UNIVERSITY CALIFORNIA INSTITUTE OF TECHNOLOGY
CAMBRIDGE, MASSACHUSElTS PASADENA, CALIFORNIA

VOLUME 27

ACADEMIC PRESS,INC.
Harcourt Brace Jovanovich, Publishers
Boston San Diego New York Berkeley
London Sydney Tokyo Toronto
THIS BOOK IS PRINTED ON ACID-FREE PAPER. @
COPYRIGHT @ 1990 BY ACADEMIC PRESS, INC.
ALL RIGHTS RESERVED.
NO PART OF THIS PUBLICATION MAY BE REPRODUCED OR
TRANSMITTED IN ANY FORM OR BY ANY MEANS, ELECTRONIC
OR MECHANICAL, INCLUDING PHOTOCOPY, RECORDING, OR ANY
INFORMATION STORAGE AND RETRIEVAL SYSTEM, WITHOU?
PERMISSION IN WRITING FROM THE PUBLISHER.

ACADEMIC PRESS, INC.

1250 Sixth Avenue, San Diego, CA 92101

United Kingdom Edition published by

ACADEMIC PRESS LTD.
24-28 Oval Road, London NWl 7DX

LIBRARY
OF CONGRESS CATALOG CARD NUMBER:48-8503

ISBN 0-12-002027-0

PRINTED IN THE UNITED STATES OF AMERICA

89909192 9 8 7 6 5 4 3 2 1
 Contents
CONTRIBUTORS ix

PREFACE xi

New Directions in Solitons and Nonlinear Periodic Waves

John P. Boyd

I. Introduction 2
11. Polycnoidal Waves 4
111. Periodic (“Cnoidal”) Waves as Exact Imbricate Series of Solitons 17
IV. Numerical Boundary Value Algorithms for Direct Computation of Solitons 24
V. Weakly Nonlocal Solitary Waves 49
VI. Summary: Quo Vudis? 74
Acknowledgments 76
References 76

Material Failure by Void Growth to Coalescence

Viggo Tuergaard

I. Introduction 83
11. Basic Equations 86
111. Continuum Models of Porous Ductile Solids 89
IV. Localization of Plastic Flow 96
V. Cell Model Studies 102
VI. Formation and Growth of Cracks 113
VII. Effect of Yield Surface Curvature 120
VIII. Strain Rate Sensitive Material 127
IX. Creep Failure by Grain Boundary Cavitation 134
X. Discussion 144
References 147

Complete Solutions to Some Mixed Boundary Value Problems in Elasticity

V. 1. Fabrikant

I. Introduction 153
11. Theory 155
111. Illustrative Examples 182
V
vi Contents

IV. Discussion 197

V. Conclusion 201
Appendix A 202
Appendix B 204
Appendix C 206
Appendix D 211
Appendix E 212
Appendix F 215
Appendix G 218
Acknowledgement 222
References 223

Convective and Radiative Heat Transfer in Porous Media

Chang-Lin Tien and Kambiz Vafai

I. Introduction 226
11. Forced Convection in Porous Media 228
111. Natural Convection in Porous Media 236
IV. Multiphase Transport in Porous Media 252
V. Radiative Heat Transfer in Porous Beds 260
Acknowledgments 27 1
References 27 1

Solitons Induced by Boundary Conditions

C. K . Chu and Ru Ling Chou

I. Introduction 283
11. Experiments 284
111. Numerical Results 286
IV. Theory 291
V. Conclusions 301
References 302

Physical Limnology
Jorg Imberger and John C . Patterson

I. Introduction 303
11. Seasonal Behavior 306
111. Surface Fluxes 321
IV. The Surface Layer 334
V. Upwelling 353
VI. Differential Deepening 370
VII. Differential Heating and Cooling 380
 Contents vii

VIII. outflow 391

IX. Inflow 405
X. Mixing below the Surface Layer 413
XI. Modeling 422
XII. Reservoir Destratification by Bubble Aerators 440
XIII. Summary 45 1
Acknowledgements 455
References 455

AUTHORINDEX 477
SUBJECT
INDEX 489
This Page Intentionally Left Blank
 List of Contributors

Numbers in parentheses indicate the pages on which the authors’ contributions begin.

JOHNP. BOYD(l), Department of Atmospheric, Oceanic and Space

Sciences and Laboratory for Scientific Computation, University of
Michigan, 2455 Hayward Avenue, Ann Arbor, Michigan 48109
Ru LINGCHOU(283), Lamont Doherty Geological Observatory, Colum-
bia University, New York, New York 10017 and NASA Goddard
Institute for Space Studies, Greenbelt, Maryland 20771
C. K. CHU(283), Department of Applied Physics and Nuclear Engineer-
ing, Columbia University, New York, New York 10017
V. I. FABRIKANT (153), Department of Mechanical Engineering, Concor-
dia University, Montreal, H3G 1M8, Canada
JORG IMBERGER(303), Centre for Water Research, The University of
Western Australia, Nedlands, Western Australia 6009
JOHNC. PATTERSON (303), Centre for Water Research, The University of
Western Australia, Nedlands, Western Australia 6009
CHANG-LIN TIEN(225), Department of Mechanical Engineering, Univer-
sity of California at Berkeley, Berkeley, California 94720
VIGGOTVERGAARD (83), Department of Solid Mechanics, The Technical
University of Denmark, Lyngby , Denmark
KAMBIZVAFAI(225), Department of Mechanical Engineering, 206 West
18th Avenue, Ohio State University, Columbus, Ohio 43210

ix
This Page Intentionally Left Blank
 This volume of Advances in Applied Mechanics contains six expository
articles that are surveys of active development in several major fields of
applied mechanics. John Boyd elucidates the limitations of the celebrated
advances in mathematical methodology made in the modern theory of
solitons and nonlinear waves as they apply to the determination of
solutions to more general problems, such as those involving boundaries at
finite distances. New directions in research are delineated for this field of
great and increasing importance, with refreshing ideas furnished on how
to cope with challenging problems that may arise from applications to
physical oceanography and other soliton-bearing systems.
Viggo Tvergaard presents a micromechanical study on the fracture
mechanism of material failure due to coalescence of microscopic voids in
ductile and polycrystalline metals. This is a problem of significant basic
value to the rapidly developing field of micromechanics as well as one of
practical importance.
V. I. Fabrikant’s article describes how a general solution can be
derived for various punch and crack problems involving an isotropic
elastic half-space, with or without axisymmetry, here exemplified with
results available for some immediate and future applications. This
method is of value not only to the study of elasticity, but to other
branches of engineering science where potential theory is being used.
Chang-Lin Tien and Kambiz Vafai bring forth an authoritative essay
on convective and radiative heat transfer and multiphase transport
processes in porous media, with or without phase change. This is a
subject that is known for its complexities and for its wide range of
applicability in high technology.
In discussing the solitons induced by boundary conditions, C. K. Chu
and Ru Ling Chou bring to our attention the urgent need for analytical
methods for resolving boundary-value problems associated with the
Korteweg-de Vries and other nonlinear evolution equations. This field is
undoubtedly emerging as an area of great importance.
xi
xii Preface

The article on physical limnology by Jorg Imberger and John C.

Patterson is an in-depth study of the hydrodynamics of lakes, which
involves physical processes no less complicated than those pertaining to
the ocean. Yet as a discipline, physical limnology has received far less
attention until only the last few years. In this field, we may find problems
of significance to environmental issues of today.
We are indebted to these authors for their interest and enthusiasm in
preparing these articles with great care to share with our readers their
expertise and scholarly views.

T.Y.W. and J.W.H.

 ADVANCES IN APPLIED MECHANICS, VOLUME 21

New Directions in Solitons and Nonlinear

Periodic Waves: Polycnoidal Waves, Imbricated
Solitons, Weakly Nonlocal Solitary Waves, and
Numerical Boundary Value Algorithms
JOHN P. BOYD
Department of Atmospheric, Oceanic, and Space Sciences
and
Laboratory for Scientific Computation
University of Michigan
Ann Arbor, Michigan

I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
11. Polycnoidal Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
A. Definitionandoverview ............................................ 4
B. Variational Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
C. Nonlinear Fourier Transform (“Hill’s Spectrum” or “Generalized Inverse
Scattering”) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
D. Theta Functions . . . . . . . . ................. ............... 10
E. The Phase Variable Bound 12
F. Polycnoidal Waves in Two Space Dimensions . . . . . . . . . . . . 16
111. Periodic (“Cnoidal”) Waves as Exact Imbricate Series of Solitons . . . . . . . . . . . 17
A. Background: Constructing Periodic Solutions from Soliton Trains . . . . . . . . 17
B. Imbricate Series and the Poisson Summation Formula . . . . . . . . . . . . . . . . . . 20
C. Generalizations and Open Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
IV. Numerical Boundary Value Algorithms for Direct Computation of Solitons . . 24
A. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
B. A Catalogue of Direct Computations of Solitons . . . . . . . . . . . . . . . . . . . . . . . 25
C. The Newton/Pseudospectral/ContinuationPolyalgorithm: A Closer Look 32
D. The Nonlinear Richardson (“Newton Flow”) Iteration and Other Artificial
TimeMethods ..................................................... 39
E. Suggestions and Guidelines . . . . . . . . . . . . . . . . . . , . . . . . . . . . , 48
V. Weakly Nonlocal Solitary Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
A. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
B. Far Field Analysis . . . . ... ... .. ..... ............... 52
C. PerturbationTheo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
D. Numerical Methods: Ignorance, the Radiation Basis, and Cnoidal
Matching ......................................................... 57
E. Physical Illustrations: Rossby Waves, Higgs Bosons, and the Slow
Manifold . . . . . . . . . . . . . . . . . . . . , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
F. Summary: The Generalization of the Concept of a “Solitary” Wave . . . . . . 73

1
Copyright 0 1990 Academic Ress, Inc.
All rights of reproduction in any form reserved.
ISBN 0-12-002027-0
2 John P. Boyd
VI. Summary: Quo Vadis? ................................................. 74
Acknowledgments ..................................................... 76
References ........................................................... 76

I. Introduction

Until recently, the study of solitary waves has been dominated by three
methodologies: (i) the inverse scattering transform, (ii) multiple scales
perturbation theory, and (iii) numerical initial value codes. For a time,
inverse scattering was widely regarded as the magic bullet of applied
mathematics, the philosopher’s stone that would turn all mystery into
truth. Even after it became clear that this method was limited to a very
special class of (mostly one-dimensional) wave equations, the enchant-
ment continued. Almost all books on solitons, even now, are treatises on
inverse scattering.
The method of multiple scales is important because it is the justifica-
tion for inverse scattering. The perturbation theory shows that when
waves are only weakly nonlinear and satisfy certain structural assump-
tions, such as a narrowly peaked Fourier spectrum or a very long length
scale, their dynamics is approximately described by one of those very
special wave equations that can be solved by inverse scattering. The
combination of inverse scattering with multiple scales perturbation theory
has been a high-powered microscope for viewing the hidden structure of
nonlinear waves. Unfortunately, the real world is three-dimensional, and
the collapse of three coordinates into one is legitimate only for weak
nonlinearity. For overcoming the limits of perturbation theory and
inverse scattering, the main tool has been numerical integration of the
initial value problem.
Time-marching is a flexible and very general strategy because it
requires no a priori assumptions about amplitudes, structure, or anything
else. The emergence of solitons from very nonsolitonic initial conditions
in one-, two-, and three-dimensional systems has been a compelling
motive for the study of solitary waves.
Unfortunately, time-marching is also clumsy and inelegant because the
numerical solution is always a potpourri of transients, dispersing wave-
trains, and solitons. Time-marching is a kind of numerical zoology, trying
to understand the beast by observing it in life. This is both useful and
necessary, but there is still a compelling need for a kind of computational
biochemistry to examine solitons in isolated, purified form.
 New Directions in Solitons and Nonlinear Periodic Waves 3

In this review, we will describe four new strategies for exploring

solitary and nonlinear periodic waves. All break away from the limita-
tions of the “Big Three” methodologies to open new horizons for what is
still a very lively field.
The first topic is the theory of polycnoidal waves. These are the
spatially periodic generalizations of multiple solitons. Their theory
explores what happens when solitary waves are not solitary, but instead
form periodic wave trains.
In this review, we will emphasize four different methodologies for
polycnoidal waves. The Nonlinear Fourier Transform and theta function
formalism are both extensions of inverse scattering. In contrast, the
Stokes series and the Newton/Fourier/continuation polyalgorithm are
direct methods that bypass inverse scattering completely.
The second topic is superimposing solitons to generate spatially
periodic waves, the method of “imbricate series.” This is an offshoot of
polycnoidal theory that has acquired a life of its own. Remarkably,
inverse scattering-soluble equations often allow an exact nonlinear
superposition principle. Imbricate series show that each soliton may
behave as an independent dynamical system even when very close to
other solitary waves.
The third topic is the direct numerical computation of multi-
dimensional solitary waves. New methodmontinuation in physical and
artificial parameters, rational Chebyshev pseudospectral schemes, and
preconditioned iterations-have enormously improved our capabilities.
Instead of marching for thousands of time steps, waiting for the transients
to disperse, the new iterative schemes, which can be interpreted as
integrations in an artificial time variable, compute the soliton directly in
only ten or twenty steps. Crude finite difference discretizations have been
replaced by exponentially accurate pseudospectral algorithms. Truncated
domains have been rendered obsolete by new spectral basis sets, such as
rational Chebyshev functions, which compute on the infinite domain.
The fourth topic is an important generalization of the traditional
concept of a solitary wave: the “weakly nonlocal” soliton. These are
coherent structures that are stable and long-lived, but violate the classical
definition of a soliton because they leak radiation to infinity. For the class
known as “nanopterons,” the leakage is exponentially small in the
amplitude, so the lifetime of the nonlocal soliton may be longer than the
age of the universe! Because of this exponential smallness, special
analytical and numerical tools are needed. Section V shows that the
4 John P. Boyd

necessary tool-making is well-advanced and gives seven examples of

nonlocal solitons.

II. Polycnoidal Waves

AND OVERVIEW
A. DEFINITION

The N-polycnoidal wave is an exact, spatially periodic solution to an

integrable nonlinear wave equation. The special case N = 1 is the
ordinary cnoidal wave. This is defined to be a nonlinear, spatially
periodic wave that translates at a constant phase speed c. The N -
polycnoidal wave for N > 1 is similar except that it is characterized by N
independent phase speeds. The polycnoidal wave is the spatially periodic
generalization of the cnoidal wave in the same way that the N-soliton
solution generalizes the single, isolated solitary wave.
In this review, we shall concentrate on the polycnoidal waves of the
Korteweg-deVries (KdV) equation
u, + uu, + u,,, = 0.

This is the simplest, most studied, and best understood case. However, it
is known that polycnoidal waves exist for other wave equations including
the cubic Schroedinger (Ma and Ablowitz, 1981), the Toda lattice
(Ferguson et al., 1982), the sinh-Gordon equation (Forest and
McLaughlin, 1983), and the sine-Gordon equation. The latter has been
heavily studied: Forest and McLaughlin (1982), Zagrodzinski (1982,
1983), Zagrodzinski and Jaworski (1982), and Jaworski and Zagrodzinski
(1982) are but a partial list.
The simplest nontrivial polycnoidal wave is the case N = 2, which was
dubbed the “double cnoidal” wave by Hyman (1978). Figure 1 illustrates
its four regimes. In the limit of very small amplitudes,

U(X, t) = a cos[k,(x - c , t ) + 4
4 + b cos[k,(x - c,t) + 42], (2.2)
where a and 6 are amplitudes, kl and k2 are wavenumbers, and Cpl and
& are phase constants. To O(a2,b2), the two cosine terms are
independently propagating linear waves. The shape of u ( x , t ) changes
with time as the crests and troughs of the two components, each traveling
at its own speed, reinforce and cancel in turn.
 New Directions in Solitons and Nonlinear Periodic Waves 5

a
- Double Cosine Wove - Perturbed Wavenumber
One Cnoidal Wave
- I\
- -
I I
-
- -

- I I I I I I

X X
*

FIG.1. Regimes of the double cnoidal wave. All 2-polycnoidal waves have these same
four regimes, but the schematics show the shapes for the “principal” branch. Each diagram
illustrates an interval that is three times the width of the spatial period.

In the opposite limit of very large amplitudes, the double cnoidal wave
may be approximated by two solitary waves of different sizes:

u(x, t) = 12s; sech2[sl(x - c l t ) + &] + 12s; sech2[s2(x- c2t)+ g52], (2.3)

where s1 and s2 are the pseudowavenumbers and where (2.3) applies only
when the solitons are separated. The peaks are very narrow in com-
parison to the spatial period (sir s2 >> l), so the pair of solitons on one
period interacts very weakly with the pair on the spatial period to the left
or to the right. However, when the peaks overlap, one should replace
(2.3) by the well-known KdV exact solution for the double solitary wave.
When s1 = s 2 , the double cnoidal wave degenerates into an ordinary
cnoidal wave with half the period; near equality between the two
pseudowavenumbers gives the “perturbed wavenumber two cnoidal
wave” shown in Figure 1. The fourth regime is a cnoidal wave that is
weakly perturbed by a sine wave.
6 John P . Boyd

FIG. 2. The superposition of wavenumber one and two components to generate the
small amplitude double cnoidal wave. The dashed and dotted curves are the individual
components; the solid curve is the 2-polycnoidal wave at a time when the two cosine waves
are in phase. Because each wave is travelling at its own phase speed (wavenumber two is
roughly four times faster than wavenumber one for the KdV equation), the shape of the
double cnoidal alters. However, the in-phase pattern shown here will periodically recur
forever.

Figure 2 illustrates how the two Fourier components reinforce and

interfere to create two peaks for small amplitudes. The relationship
between the nonlinear normal modes and the two crests changes
dramatically with amplitude. At large amplitudes, the nonlinear normal
modes are the crests themselves. When the amplitude is small, however,
the tall peak is the sum of the two normal modes (cosine waves) while the
smaller crest is the diference between the two independently propagating
waves.
It is striking, however, that the shape of u(x, t ) does not change with
amplitude (along the diagonal in Figure 1) in the sense that there is
always one tall peak and one short crest on each spatial period during at
least part of each recurrence cycle. Because of this, the small amplitude
and large amplitude approximations (Equations (2.2) and (2.3), respec-
tively) strongly overlap. For intermediate amplitude, one may describe
the double cnoidal wave as either the superposition of sine waves or as
the sum of two solitons. The error is small for both because both
approximations predict two unequal crests.
Strictly speaking, Figure 1 and Equation (2.3) describe only the
 New Directions in Solitons and Nonlinear Periodic Waves 7

“principal” branch of the double cnoidal wave. This branch has k z = 2k,
in (2.2) and has precisely two solitons, one tall and one short, on each
spatial period. Mathematically, one may have spatially periodic solutions
with m tall solitons and n short solitons on each period, where rn and n
are arbitrary integers. Haupt and Boyd (1989) and Hyman (1978) have
computed explicit examples where m is 1 and n = 2, 3, or 4. If the ratio
k z / k , is irrational, then the double cnoidal wave is almost-periodic in
space. Forest and McLaughlin (1982) discuss possible applications of
almost-periodic polycnoidal waves. Whenever we discuss a specific
solution branch, however, it will be the principal branch as shown in
Figure 1.
Polycnoidal waves have a bewildering variety of names. Mathemati-
cians have used “finite band,” “finite gap,” “waves on a circle,” and
“N-phase wavetrains” as synonyms for “polycnoidal.” The last of these is
very descriptive because the N-polycnoidal wave can always be written in
terms of N independent phase variables

[;ES k;(x - c;t) + @i, i = 1, . . . , N, (“phase variables”) (2.4)

each with its own independent wavenumber and phase speed. However,
we shall always refer to these solutions as “N-polycnoidal” waves.
One major issue is that polycnoidal waves are obviously very special.
How relevant are these special waves to the general spatially periodic
solution? The answer is that any u(x, t ) may be approximated to
arbitrarily high accuracy for an arbitrary, finite time interval by an
N-polycnoidal wave of sufficiently large N and appropriately chosen
parameters. Thus, to understand the polycnoidal special solutions is to
understand the general solution, too.
Polycnoidal methodologies include the following:

(i) Variational Principle

(ii) Nonlinear Fourier Transform
(iii) Theta Function Residual Equations
(a) Numerical solutions
(b) Perturbative solutions
(iv) Phase Variable Boundary Value Problem
(a) Stokes (Poincart-Lindstedt) series
(b) Newton/Fourier/continuationcomputations.

In what follows, we shall discuss each in turn.

8 John P . Boyd

B. VARIATIONAL
PRINCIPLE
The Korteweg-deVries equation has an infinite number of independ-
ent, conserved integrals (Ablowitz and Segur, 1981). Lax (1976) showed
that these provide a variational characterization of the N-polycnoidal
wave. If the first N - 1 of these invariants are fixed, then any solution
which minimizes the N-th invariant is an N-polycnoidal wave. The
minimizing solution is not unique because the polycnoidal wave is a
function of N phase variables; as t increases, each phase variable increases
linearly, too, and the shape of u(x, t) alters. Thus, the minimizing
solution is unique except for N arbitrary phase parameters.
Hyman (1978) translated Lax’s principle into FORTRAN. He dis-
cretized the functionals and performed a constrained minimization of the
N-th functional. This crude solution was refined by solving an ordinary
differential equation, the usual Euler equation of variational calculus.
Unfortunately, all his solutions had either four or five solitons on each
spatial period-typically one large solitary wave and three or four
identical shorter, broader peaks. He missed the “principal” branch and
did not thoroughly explore the parameter space.
Nevertheless, this approach has interesting possibilities. Its strength is
that one may compute N-polycnoidal waves for large N by solving an
ordinary differential equation (albeit one of order 2N + 1). However,
Hyman’s work has never been continued.

FOURIER
C. NONLINEAR TRANSFORM
(“HILL’SSPECTRUM”
OR
“GENERALIZED
INVERSE
SCATTERING”)
The early work on polycnoidal waves was the exclusive province of
pure mathematics. The result was many beautiful and intricate formulas,
existence proofs, and absolutely no attempt to actually calculate any-
thing. Very recently, however, Osborne and Bergamasco have developed
a fast algorithm for both the inverse scattering transform on the infinite
interval and its generalization for spatial periodicity. They call the latter
the “Nonlinear Fourier Transform.”
Unfortunately, a full explanation of their algorithm requires a heavy
dose of spectral theory, far beyond what is feasible to include in this
review. The computer program, however, is not particularly long or
complex. Osborne and Bergamasco’s two numerical articles have not yet
been published, but see Osborne and Bergamasco (1985, 1986, 1987) and
Tracy, Larson, Osborne, and Bergamasco (1988, 1989).
 New Directions in Solitons and Nonlinear Periodic Waves 9

In brief, their algorithm computes a set of amplitudes from an arbitrary

initial condition. When [u(x, f = O)l<< 1, these amplitudes are simply the
coefficients of the Fourier series of the initial condition. For larger
solutions, however, some or all of the amplitudes become (in effect) the
soliton pseudowavenumbers (the siin (2.3)) that determine the width and
amplitude of each solitary wave. Thus, the Osborne-Bergamasco algor-
ithm is a procedure for computing the amplitudes of normal modes-but
it is a generalization of ordinary Fourier analysis because the normal
modes are nonlinear. When a mode is sufficiently strong, it is not a
single, steadily propagating sine wave but rather a single, steadily
propagating soliton.
Of course, the word “amplitude” must be used with care because all
the amplitudes must be simultaneously fed to the algorithm to compute
the polycnoidal wave when two or more solitons are strongly interacting.
Nonetheless, the Nonlinear Fourier Transform is still very powerful
because one may predict the solitary waves (and smaller, quasi-sinusoidal
components) that will emerge from an arbitrary, spatially periodic initial
condition.
An amusing illustration is Osborne and Bergamasco’s (1986) reanalysis
of Zabusky and Kruskal (1965), the now-legendary note that first
described the elastic collisions of Korteweg-deVries solitary waves and
coined the word “soliton.” Zabusky and Kruskal’s initial sine wave broke
up into nine solitons. By carefully tracing each solitary wave before,
during, and after each soliton-soliton collision in the numerical output,
they were able to plot the x-t trajectory of each solitary wave. In
contrast, Osborne and Bergamasco (1986) were able to predict the
number of solitions and the whole time-evolution directly from the initial
condition. No Runge-Kutta or leapfrog for them!
Osborne and Bergamasco are currently applying their algorithm to
deduce the spectrum of ocean long waves from surface height measure-
ments. Another extension (Tracy et al., 1988) is to examine the
relationship between the Nonlinear Fourier Transforms of the Korteweg-
deVries and Cubic Schroedinger equations: one equation passes into the
other and vice-versa in the appropriate asymptotic limit (Johnson, 1976).
Thus, there is no doubt that this line of attack will continue. The
studies by Osborne, Bergamasco, and their collaborators are the direct
and logical extension of the vast amount of research on the inverse
scattering transform for the infinite interval.
Unfortunately, the algorithm does have some drawbacks. First, it is not
10 John P . Boyd

a good tool for exploring the parameter space of the double cnoidal or
triple cnoidal waves. An arbitrary initial condition is an infinite genus
polycnoidal wave (that is, an N-polycnoidal wave with N - + m ) . The
Nonlinear Fourier Transform does not provide any obvious way to
specialize the initial condition to a polycnoidal wave of finite N.
A second limitation is that the method is restricted to those very
special wave equations which can be solved via the inverse scattering
transform: the Kortweg-deVries, the sine-Gordon, and so on. It seems
probable that there are quasi-polycnoidal waves for nonintegrable wave
equations, too, but the Nonlinear Fourier Transform is useless for these.
The third and most serious limitation is that the theory behind the
generalized inverse scattering method is very complicated. Few applied
mathematicians or fluid dynamicists will have the stamina to master
dozens of theorems and new terms to understand a technology whose
range of applicability is limited-limited to the narrow parameteric range
where perturbation theory can justify the Korteweg-deVries equation.

D. THETA
FUNCTIONS

Another remarkable property of N-polycnoidal waves is that they can

be expressed in terms of the N-dimensional Riemann theta function (Its
and Matveev, 1975). Although the theta series are infinite, all the Fourier
+
coefficients may be expressed in terms of just N(N 1)/2 independent
parameters. These are the elements of the symmetric N-dimensional
“theta matrix.” Because an infinite number of coefficients has been
replaced by a small finite set of unknowns, theta functions permit some
remarkable numerical and perturbative algorithms.
Theta functions do not themselves satisfy the KdV equation, but rather
give the KdV solution through the transformation
u = 12(ln[e]), . (2.5)
This transformation was introduced by Hirota (1974) because it provides
a remarkably compact way of writing the N-soliton solution to the KdV
equation. He subsequently extended this formalism to most of the other
inverse scattering-soluble wave equations as reviewed by Matsuno (1984).
Nakamura (1979) and Boyd (1982b) independently showed that the
theta function satisfies Hirota’s transformed bilinear version of the KdV
equation. When the theta function is substituted into the Hirota-KdV
 New Directions in Solitons and Nonlinear Periodic Waves 11

equation, however, one is confronted by a paradox: an infinite number of

Fourier terms must vanish, but the N = 2 theta solution contains only
four parameters! The Boyd-Nakamura Theorem shows that when four
independent terms in the residual function vanish, this implies that all the
rest are zero, too: the theta formalism reduces a partial differential
equation to a small set of algebraic equations (“theta residual” set).
For the double cnoidal wave, for example, the unknowns are the two
phase speeds, c1 and c2, plus the off-diagonal element of theta matrix,
&, and a physically irrelevant constant of integration. The diagonal
elements of the theta matrix are parameters that are equivalent to the
amplitudes of the two independent wave components or solitons.
One technical complication is that the Boyd-Nakamura Theorem
reduces the Hirota-KdV equation to 2N residual equations whereas the
N-polycnoidal wave has only [1+ (N + 1)N/2] parameters-more equa-
tions than parameters when N > 2. Hirota and Ito (1981) tried the bold
expedient of solving 7 of the 8 residual equations for N = 3 and then
verifying that the remaining equation was also satisfied to machine
precision. The numerical evidence is that the “theta residual” equations
are always soluble, but a rigorous proof is lacking.
A second technical complication is that for large amplitudes, the
theta-Fourier series converges very slowly. Via the Poisson summation
formula, however, the theta function may be alternatively expressed as a
“Gaussian” or “imbricate” series. The lowest term is the KdV double
soliton and the series converges more rapidly as the polycnoidal wave
becomes larger. Boyd (1982b) shows that the Boyd-Nakamura Theorem
applies to the Gaussian series, too. In the seven years since the review of
Dubrovin (1981), polycnoidal wave theory has been extended in three
directions. First, Nakamura (1980) and his collaborators (Nakamura and
Matsuno, 1980) have generalized the theta formalism to many other wave
equations besides the KdV equation (Matsuno, 1984).
Second, several groups of mathematicians have filled in the theoretical
gaps, especially for equations other than the KdV. Tracy et af. (1989),
Ferguson, Flaschka, and McLaughlin (1982), and Forest and McLaughlin
(1982, 1983) are especially interesting. Zagrodzinski (1982, 1983),
Zagrodzinski and Jaworski (1982), and Jaworski and Zagrodzinski (1982)
apply the theta formalism to the sine-Gordon equation. Other interest-
ing discussions are given in the reviews by Date and Tanaka (1976),
Dubrovin (1981), and Zagrodzinski (1984).
Third, Boyd (1982b, 1984b, 1984c, 1984d) has computed perturbative
12 John P . Boyd

solutions to the KdV theta-residual equations for both the small

amplitude (Fourier) and large amplitude (“imbricate-Gaussian”) theta
series. These articles discuss a number of subtle details that must be
omitted here. The beauty of the theta formalism is that one may express
the complete answer via just three perturbation series, one for each of
the following: cl, c2, and the off-diagonal theta matrix element. Each
expansion is a double series in the two amplitudes (or equivalently, in the
diagonal theta matrix elements).
This simplicity in the final answer is one of the great strengths of the
theta formalism. The other is beauty; the multi-dimensional theta
functions, conserved integrals, trace formulas, logarithmic transforma-
tions, and hyperelliptic functions fit together as grandly and as astonish-
ingly as a painting of Jan Vermeer.
The great weakness of the theta formalism is that the simple answer is
obtained only at the price of great complexity in the derivation, a
weakness shared with the inverse scattering transform. This is not
unusual in applied mathematics; the method of multiple scales, for
example, is notorious for generating simple answers from long and
tedious calculations.
The difference is that multiple scales is a very general tool, applicable
to boundary layers and limit cycles as well as inviscid waves. The theta
formalism, alas, is only applicable to a handful of differential equations
which, on an infinite interval, can all be solved via inverse scattering. If
one likened these tools to languages, the method of multiple scales would
be a major tongue like French or English; the theta function theory
would be Basque or the dialect of a bush tribe in New Guinea.
Because of its intricacy and narrow scope, the theta method and the
Osborne-Bergamasco algorithm described in the previous section are
likely to remain the province of specialists. In the next subsection, we
shall discuss a much less elegant but more general approach to poly-
cnoidal waves.

E. THE PHASEVARIABLE BOUNDARY VALUEPROBLEM

AND THE STOKESSERIES

The double cnoidal wave is a function only of the phase variables

X = kl(x - c,t) + + 1 and Y = k,(x - c2t)+ G2. By substituting u ( X , Y)
into the KdV equation and using the definitions of the phase variables,
 New Directions in Solitons and Nonlinear Periodic Waves 13

one finds that the double cnoidal wave must solve the Phase Variable
Boundary Value Problem (PVBVP):
(U - C,)Ux + 2(u - ~ 2 ) +~ y + 6uXm + 1 2 ~ - + 8 ~ - = 0.
u X ~ X (2.6)
The numerical coefficients in (2.6) are those for the “principal branch”
with k, = 1, k2 = 2, and a spatial period of 2n, but Haupt and Boyd
(1989) also discuss the branch with k2 = 3.
Equation (2.6) is rather unusual in that it is a nonlinear problem with
two eigenvalues c , and c2. Nonetheless, the problem is well-posed on
both the periodic and infinite plane.
Haupt and Boyd (1989) discuss two ways of solving (2.6). The
numerical Newton/Fourier/continuation polyalgorithm will be explained
in the next section. The second method is the Stokes expansion, which is
known in celestial mechanics as the PoincarC-Lindstedt series.
Stokes observed that in nonlinear wave problems, one must expand not
only the wave height but also the phase speed-for the double cnoidal
wave, both phase speeds-as functions of the amplitude a :
u = au,+ a2u2+ a3u3+ - - - , (2.7a)
c, = c10+ acll + aZcl2+ . - - , (2.7b)
c2 = cZ0+ ac2, + a 2c22+ - - - , (2.7~)
The lowest order solution is the sum of two sine waves of different
wavenumbers ( k , # k2):
u1= c o s X + b cos Y. (2.8)
The amplitude parameters a and b are independent so that the double
cnoidal wave is a two-parameter family of solutions. However, the most
interesting case is when the two amplitudes are the same order of
magnitude; otherwise, the double cnoidal wave is merely an ordinary
cnoidal wave with a small perturbation. Consequently, we assume that b
is 0(1) and use the amplitude of cos X as the order of magnitude for the
amplitude of COSY,too. The perturbation series may be computed by
substituting (2.7) into (2.6) and matching powers of a.
At each order, we solve
-ClOui,x - 2C,Ui,Y + ui,xXx + 6Ui,xxv+ 12Ui,xw+ 8 u i . w ~ =E ( X , Y),
(2.9)
where
i-I

4 ( X , Y) = - 2 [ ( U j - C1j)Ui-j,x + 2(Uj - C2j)Ui-,.Y]. (2.10)

j=1
14 John P. Boyd

To solve (2.9), we expand both the solution and &(X,Y) as Fourier

series:
u,(x, Y ) = C x
i

m=On=-j
i
u ~cos(mX +
, ~ n~ ~ ) , (2.11)

2 x h,,,
i i
&(x,Y ) = sin(rnX + nu). (2.12)
m=O n = - i

One may prove recursively that each Fourier series in (2.11) and (2.12) is
finite as expressed by the upper limits on the sums. Matching trigono-
metric factors gives
u ~ , =fi,mn/(clOm
~,, + czo2n+ m3+ 6m2n + 12rnn2+ 8n3). (2.13)
When rn = 0 and n = 1 or m = 1 and n = 0, the denominator of (2.13)
vanishes. The phase speed corrections are chosen so that f;.,lo and f,ol are
0. In the language of the method of multiple scales, the phase speeds are
found by imposing the “nonsecularity” conditions. Because these condi-
tions set ui,lo= ui,ol= 0, the amplitudes of cos X and cos Y are a and a6
to all orders in perturbation theory.
At ninth order, the denominator of the (rn = 5 , n = -4) term is also
zero. Numerical solutions show that this trigonometric term, which
would is expected to be O(a’), actually varies proportionally to a’. Haupt
and Boyd (1987) show, using a simple one-dimensional wave equation,
how to modify the Stokes series to cope with such a resonance.
The basic idea is to assume that the resonant term occurs at lower
order (in this case, seventh order instead of ninth order) with an
undetermined coefficient. At ninth order, this extra degree of freedom is
chosen so as to make f9,5,--4zero. This strategy of carrying undetermined
parameters to higher order and then choosing them to eliminate
resonance is precisely what is done with the phase speed corrections.
Table 1 gives the first few terms in the Stokes series for the principal
branch of the double cnoidal wave. It is less compact than the
theta-Fourier series, but in compensation, one does not need to take the
logarithm and differentiate twice: the terms in Table 1 give u(x, t )
directly.
Like all small amplitude expansions, the Stokes series fails when the
soliton peaks are too narrow in comparison to the spatial period. Unless
one employs the theta formalism, there is no large amplitude perturba-
tion theory to switch to. However, Pad6 approximants, as shown for a
one-dimensional Stokes series in Boyd (1986a), can double the range of
usefulness of a power series.
New Directions in Solitons and Nonlinear Periodic Waves 15

TABLE 1
PHASE SPEEDS AND FOURIERCOEFFICIENTs
FOR THE KdV DOUBLE WAVE AS
CNOIDAL
COMPUTED VIA STOKES' EXPANSIONS"
at
CI = -1 +-24
a2b2
c*=-4+-
96

Fourier term Coefficient

cos x a
cos Y ab
a2
-
cos 2 x
12
a3
-
cos 3 x
192
a2b2
-
cos 2Y
48
a3b3
cos 3Y ~

3072
a2b
-
cosx+ Y
12

cosx-Y _ a2b
_
12
25a3b2
-
cosX+2Y
10368
a3b2
cos x - 2Y -~
128
a3b
cos 2 x +Y -
162
cos2x- Y 0

X = x - c,t + @, and Y = 2(x - c,r + @2),

where 4, and @, are phase constants. These
definitions imply that the series are for the
"principal branch" of the double cnoidal
wave (k2=2k,). Taken from Haupt and
Boyd (1989), which also gives the expansion
for the k, = 3k, branch.
16 John P. Boyd

Stokes expansion is very general because its derivation is purely

mechanical. It is unnecessary that the evolution equation be integrable;
Stokes’ algorithm merely manipulates finite sums of trigonometric
functions. The algebra is easily delegated to a computer, especially using
an algebraic manipulation language such as REDUCE, MACSYMA, or
Mathematica.
The weakness of Stokes’ method is that precisely because it is so
mechanical, it will always generate an answer even if none exists.
Generation of a few terms, or even many terms, is not an existence
proof. Vivid examples are given by the series for nonlocal solitons, which
are discussed in Section V.
As the amplitude increases, the accuracy of the Stokes series deterior-
ates to uselessness, even with Pad6 improvement. The remedy is to solve
the Phase Variable Boundary Value Problem numerically via the
Newton-Kantorovich/Fourier/continuation polyalgorithm (Haupt and
Boyd, 1987, 1989). Such numerical methods are the theme of Section IV.
Haupt (1987, unpublished Ph.D. thesis) has applied the Stokes series
to the wave equation of Peregrine (1966) and Benjamin et a f . (1972). This
equation is known to be nonintegrable. Nevertheless, when the Stokes
series is used to initialize a time integration, the numerical solution
remains close to the Stokes series for several recurrence periods at least
in some parameter ranges. The applicability of the concept of polycnoidal
waves to the Benjamin-Bona-Mahony equation and other nonintegrable
wave equations remains an intriguing open question.

F. POLYCNOIDAL
WAVESIN Two SPACEDIMENSIONS

The Kadomtsev-Petviashvili equation

(u, + 6uu, + u,,,), + 3uyy= 0 (“KP” equation) (2.14)
is a two-dimensional generalization of the Korteweg-deVries equation.
(Kadomtsev and Petviashvili, 1970). It, too, has polycnoidal wave
solutions. At small amplitudes, the double cnoidal is again the sum of
two noninteracting sine waves. At large amplitudes, the KP double
cnoidal wave is the superposition of solitons of two different sizes.
The difference from the Korteweg-deVries equation is that because
the KP equation has two space dimensions, the two independent
components of the double cnoidal wave are free to propagate at an angle
 New Directions in Solitons and Nonlinear Periodic Waves 17

to each other. In the special case when this angle is zero, the KP reduces
(after a rotation of coordinates, if needed) to the Korteweg-deVries
equation. Thus, the KP double cnoidal wave includes the KdV double
cnoidal wave as a special case.
Once again, however, the double cnoidal wave is a “genus 2”
hyperelliptic function, which may be expressed in terms of two-
dimensional Riemann theta functions. Once again, there are two
independent phase speeds and amplitudes.
Segur and Finkel (1985) give a good treatment of the KP double
cnoidal wave. In particular, they show that the theory predicts patterns
that qualitatively resemble the “X-shaped” patterns often seen in real
water waves. Hammack (1989) reviews laboratory experiments that have
confirmed the theta function theory for shallow water waves. Pierini
(1986) has performed initial value experiments.

HI. Periodic (“Cnoidal”) Waves as Exact Imbricate

Series of Solitons

A. BACKGROUND:
CONSTRUCTING
PERIODIC
SOLUTIONS
FROM SOLITON
TRAINS

As noted earlier, one obvious approximation to a large amplitude

spatially periodic wave is a train of evenly spaced, identical solitons. If
the solitary wave is denoted by U ( x ) , then the function
m

u(x)= C
n=--m
~(x-2nn) (3.1)

is manifestly periodic with a period of 2;n. To show that this is a very

good approximate solution whenever the solitons are tall and narrow in
comparison to the periodicity interval, an example is helpful.
The solitary wave of the Korteweg-deVries equation is
U ( x ;s) = 12s’ sech’(sx), (3.2)
where s is the “pseudowavenumber” and x is the spatial coordinate in a
frame of reference that is traveling with the wave. Denote the individual
terms in (3.1) by
U,(x) = U ( x - 2nn). (3.3)
18 John P. Boyd

If the soliton series is substituted into the KdV equation, one obtains

where the residual function r ( x ) would be identically zero if u ( x ) were an

exact solution. Since each Un(x)is a solution of the KdV equation, the
terms in braces { } in (3.4) are identically 0. The cross-terms in (3.4)
alas, are not zero-but they are exponentially small. The maximum
overlap of the soliton centered at x = 0 with its nearest neighbors

4
2
0
-2
-4
-6

$q
-8

-2

-4

X
Ro.3. (a) Solid curve: exact KdV cnoidal wave for s = 4 and period 2n. Dotted curve:
u,, = 1.03908 cos X. Dashed curve: us,, = -1.9099 + 3 sech2(x/2). (b) Solid curve: exact
cnoidal wave, same as in (a). Dashed: constant-plus-soliton approximation u,, as in (a).
The horizontal dividing line is at u = -6/n = -1.9099 to show that the soliton asymptotes
to this negative number rather than to zero. Although the soliton is not a spatially uniform
approximation to the cnoidal wave, it is accurate on x E [-n, n].
 New Directions in Solitons and Nonlinear Periodic Waves 19

centered on x = i 2 n is
= max lu(x; s) - U ( x ;s ) l = s >> 1, (3.5)
x s [ - n . n]

which is attained at x = fn. Thus, the residual r ( x ) -O(exp[-2sn]) so

that the error of the soliton series is exponentially small in the
amplitude-and-width parameter s.
All this is very straightforward and obvious. It is also irrelevant! Toda
(1975) showed that the cnoidal wave of the Korteweg-deVries equation
is given exactly by
m
12.9
- la2
u ( x ; s) = - - + sech2[s(x - 2nn)l. (3.6)
n n=-m

The constant -12FIn has been added to the sum of solitons so that the
spatial mean of u ( x ; s) is zero, which is the conventional normalization.
The phase speed c of the cnoidal wave is different from that of its
constituent solitons, but only by an exponentially small amount when
s >> 1. Remarkably, however, (3.6) is error-free for all s, even when s is
very small and u ( x ; s) = 24 exp(-n/s) cosx. Figure 3 illustrates how the
individual terms in (3.6) superpose to give the cnoidal wave.

TABLE 2
SUMMARY OF ARTICLES
ON IMBRICATE SERIES OF SOLITONS

Reference Wave equation

~~

Toda (1975) Toda Lattice, Korteweg-deVries

Korpel and Banerjee (1981) Korteweg-deVries
Zaitsev (1983) Benjamin-Ono, Joseph, Cubic Schroedinger,
Modified Korteweg-deVries
Kadomtsev-Petviashvili“
Whitham (1984) Korteweg-deVnes, Modified Korteweg-deVries,
Boussinesq and Burger’s’’
Boyd (1984a) Korteweg-deVnes, Cubic Schroedinger,
Modified Korteweg-deVries
Miloh and Tulin (1989) Benjamin-Ono
Boyd (1988e) Quartic Korteweg-deVries‘
Boyd (1989b) Resonant Triad

a Zaitsev alone has succeeded for an equation in more than one space dimension, but his

“pattern function” is restricted to the rational solitons of the KP equation.

” Burger’s equation does not have soliton solutions, but Whitham shows that it does have
periodic solutions that are the exact superposition of shocks.
Boyd shows that the superposition is nor exact even though the solitons and periodic
waves are hyperelliptic functions. All other articles prove the superposition principle is exact
for the wave equations treated.
20 John P. Boyd

Toda’s Theorem-published eighty years after Korteweg and deVries

had discovered and named the steadily translating, spatially periodic
solution to the differential equation-has spawned a kind of cottage
industry. Table 2 summarizes extensions of Toda’s work. The conclusion
is that the superposition of solitons is exact for all periodic waves and for
all solitons that are hyperbolic or rational functions.
In the next subsection, we shall give a general methodology for proving
and understanding such soliton-superposition principles.

B. IMBRICATE
SEFUES
AND THE POISSON
SUMMATION
FORMULA

Theorem 3.1. (Imbricate Series). Any periodic function f(x) with

period L has two series representations. If the usual Fourier series is

f ( x ) = a(2n)-’” 2
n=-m
g(m)ei2nxnll, (3.7a)

then the alternative expansion is the imbricate series

m
2Jr
f@)= 2
m=-m
G ( IX~ [ X - m 4 (3.7b)

where G ( k ) is the Fourier transform of g ( n ) , that is,

m

~ ( k=)(2n)-1/2 J
-m
g(x)eih h, (3.8)

and where IX is an arbitrary positive constant.

Definition 3.1. The expansion (3.7b) is said to be an imbricate

Fourier series or simply imbricate series.

Definition 3.2. G ( k ) is said to be the pattern function for the

imbricate series.

Definition 3.3. The Fourier series and the function defined by it are
both said to be the imbrication of the pattern function G ( k ) .

These rather odd-sounding names come from a Latin word, still used
by geologists, which means “to overlap like tiles” (Boyd, 1986a). The
term is apt because various copies of the pattern function do overlap.
 New Directions in Solitons and Nonlinear Periodic Waves 21

Many functions also have a second species of imbricate series that is

equally important for solitons and elliptic functions.

Theorem 3.2 (Alternating Imbricate). Zf f ( x ) is the sum of an odd

Fourier series of period 2L, that is, a series whose terms are restricted to
only the sines and cosines of odd degree, i.e.,
m

f ( x ) = n(2n)-'/2 2 g(a[n+ tl)ei(&+')"/~,

n=-m
(3.9a)

where (Y is an arbitrary scaling constant, then f ( x ) also has the alternating

imbricate series

where G ( k ) is again the Fourier transform of g(x).

Theorem 3.3. The pattern function G(x) is not unique.

The reason is that the Fourier coefficients are defined only by the
values of g(n) for integer n (or half-integer n for (3.9)). Consequently,
there are an infinite number of different functions g(n) that yield the
same Fourier coefficients. For example, one may add w(n)sin(nn) to
g(n), where w(n) is abritrary. Different choices for g(n) will yield
different pattern functions; however, the differences between various
pattern functions cancel out when the imbricate series is summed.

Theorem 3.4. The imbrication of a given G(x) is always a unique

function f (x).

It is only the reverse, function-to-pattern relationship, f ( x ) + G ( x ) ,

which is ambiguous.
The proofs of all these theorems are given by the Poisson summation
formula (Morse and Feshbach, 1953) together with other elementary
Fourier transform theorems. (Some Fourier monographs actually use
(3.7) as the definition of the Poisson summation formula.) The scaling
constant (Y has been included in the theorem to show that the narrower
the pattern function, the more slowly the Fourier coefficients decrease as
In[-+m and vice-versa.
22 John P. Boyd

TABLE 3
EXAMPLES
OF IMBRICATE
SERIES

Function" Fourier termb Imbricate termb

UKDV +
12F
6n cosech( g)e" 12s2sech2[s(x - 2nn)]

s sech(s[t - nn])
(sn) s( - 1)"tanh(s[t - nn])

&pd)
(cn) s(-1)" sech(s[t - nn])
e-(n/s)nz 2im s1/2 -(s/n)(x-rnP
4
"The Korteweg-deVries (KdV) cnoidal wave uKdV is normalized to have a zero
mean. The Benjamin-Ono cnoidal wave has the closed form expression
- ss/2
where S = tanh s.
UBenjsrnin-Ono - 1- (1 - s2)'/z cos x '
The triad expressions are those for unit frequency, but Boyd (1989b) shows that it is
trivial to construct the general triad solution from the functions displayed here. Each
triad component is proportional to the elliptic function which is shown in parentheses;
the more complicated formulas for the dn, sn, and cn functions themselves are given in
Boyd (1984a). Note that a1has a period of x while the other components have periods
of 2n so that the triad as a whole repeats only every 2n units of time.
The summation is over n = --m to m.

Happily, the Fourier coefficients of the elliptic functions and of simple

rational functions of cosx are known in analytical, closed form-and so
are their transforms. Table 3 is a collection of typical imbricate series and
the corresponding Fourier sums. The expansions for the second and third
triad components (elliptic functions sn and cn) are alternating imbricate
series.
The resonant triad solutions have been included to illustrate that
imbricate expansions are useful not only in space, but also in time. These
unit frequency triad solutions are special, but Boyd (1989b) shows that
the general triad solution can be obtained from these special solutions
merely by rescaling and a phase translation.
Each triad component is proportional to a different elliptic function;
the dn, sn, and cn functions (in the space coordinate) are also the
spatially periodic solutions of various wave equations such as the Cubic
 New Directions in Solitons and Nonlinear Periodic Waves 23

Schroedinger, Modified Korteweg-deVries, and so on. Lack of time

precludes cataloging all these relationships. The crucial point is that all
the hyperbolic and rational solitons can be “imbricated” to form exact,
spatially periodic solutions.
No exceptions are known. However, the phase speed of the periodic
wave is almost always different from that of the constituent solitons.
Imbricate series like those in Table 2 have been useful in many
branches of physics and applied mathematics. Besides the soliton/cnoidal
wave theory, other applications include the following: (i) constructing
adaptive coordinate grids for periodic problems (Boyd, 1989e), (ii)
comparing Fourier and Chebyshev domain truncation for spatially
decaying solutions (Boyd, 1988f), (iii) crystal lattice sums (Wimp, 1981),
(iv) periodic solutions of the heat equation (via theta functions like O3 in
the table), (v) asymptotic approximations to Mathieu, spherical har-
monic, and prolate spheroidal functions (Morse and Feshbach (1953) and
Boyd (1989d)), and (vi) periodic solutions of Burger’s equation as the
imbrication of shocks (Whitham, 1974).

AND OPENPROBLEMS
C. GENERALIZATIONS
An immediate question is, Does the exact soliton superposition
principle extend from ordinary cnoidal waves to N-polycnoidal waves for
N z ~ ? The same reasoning given earlier shows that, at least
approximately, the answer is yes. The double soliton formula, for
example, approximates the 2-polycnoidal wave to within an error that is
exponentially small in the inverse of the widths of the two solitary waves.
Boyd (1982b, 1984b, 1984c, 1984d) shows that the superposition is
exact for the N-dimensional Riemann theta functions whose second
logarithmic derivative (for the KdV equation) is the N-polycnoidal wave.
However, it is still an open problem whether the imbricate-soliton series
is exact or merely approximate for the hyperelliptic functions that
describe u(x, t ) itself.
Boyd (1988e) shows that the superposition is not exact (although it is a
good approximation even for very small amplitudes) for the steadily
translating, spatially periodic solutions of the quartically nonlinear KdV
equation. The cnoidal waves of this one-dimensional wave equation are
hyperelliptic functions. Clearly, merely being a special function-and
hyperelliptic functions are very special-is insufficient for the imbricate
series of solitons to solve a wave equation exactly.
24 John P. Boyd

The theme developed in this section-that the crests of a periodic wave

behave as solitary waves, dynamically independent of one another-has
been known to water wave specialists for a long time (Peregrine, 1983).
Munk (1949) noted, simply by observing the shapes of waves running up
a beach, that the waves become steeper and narrower as they approach
the shore until they look like solitons. Stiassnie and Peregrine (1980)
gave a more rigorous demonstration that such shoaling waves interact
very weakly. Ocean wave crests are not identical in either shape or
spacing, but it is the narrowness of the crests in comparison to their
spacing that is important. Yasuda and Tsuchiya (1982Fwith good
success-fit observed water wave spectra by a sequence of solitons with
variable spacing and amplitude.
It is plausible that similar ideas should work in more than one
dimension. For example, McWilliams (1984, 1989) has shown that
two-dimensional turbulence and three-dimensional, quasi-geostrophic
turbulence decay into a sea of small, intense vortices, which, apart from
occasional collisions, interact very weakly.
Lastly, the resonant triad solutions show that imbricate series may be
useful for time-dependence. Lorenz (1960) showed that the triad equa-
tions are the simplest model for instability and nonlinear vacillation. The
triad series are not the imbrication of solitary waves, but rather are the
superposition of isolated growth/equilibration/decay events. It remains
to be seen whether similar imbricate expansions will be useful for other
examples of time-periodic or time-repetitive vacillating instability.

W . Numerical Boundary Value Algorithms for

Direct Computation of Solitons

A. INTRODUC~ION

Numerical solutions to initial value problems have been an important

tool for understanding nonlinear waves. However, IVP codes are crude
instruments for investigating solitary waves themselves. Because the
solitons are unknown, the initial conditions will necessarily be crude
approximations or guesses to the solitary wave. The modeller is then
forced to try to distill the pure soliton from a sea of transients.
A new direction is to compute solitons directly by solving a nonlinear
boundary value problem. The soliton is assumed to be steadily translat-
 New Directions in Solitons and Nonlinear Periodic Waves 25

ing, which reduces the time-dependent wave equation to a nonlinear

boundary value problem in the spatial coordinates with the phase speed c
as the nonlinear eigenparameter.
In this section, we shall describe and catalog a rather diverse collection
of direct, solve-the-boundary-value-problem attacks on solitons. Each
investigation is numerically characterized by the three parts of its
computer code: (i) initialization (generating the first guess), (ii) iteration
(usually Newton’s), and (iii) discretization (finite difference, spectral, or a
special scheme).
In the next subsection, we give an overview of fifteen representative
studies. Subsection C is an in-depth analysis of one particular numerical
polyalgorithm: the continuation/Newton-Kantorovich/pseudospectral
strategy of Boyd (1986a, 1986b, 1988b, 1988d). The final subsection is
focused on iteration methods, which are crucial to efficiency.

B. A CATALOG
OF DIRECT
COMPUTATIONS
OF SOLITONS

1. Initialization (Generating the First Guess for the Iteration)

One special curse of nonlinear problems is that it is almost always

necessary to iterate. This in turn requires a first guess to initialize the
iteration. If the first guess differs too greatly from the true solution, the
iteration will diverge, so a good initialization is essential. Initialization
strategies include the following four: (i) an analytical solution, (ii)
continuation (“homotopy”), (iii) residual inhomogeneity and other
artificial parameter techniques, and (iv) a one degree-of-freedom numer-
ical solution.
Small amplitude perturbation theory is a fertile source of analytical
solutions. Boyd (1988d), for example, applies the method of multiple
scales to approximate equatorial waves as Korteweg-deVries solitons.
For nonlinear spatially periodic waves, an even simpler approximation-
the infinitesimal cosine wave-is sufficient. For small amplitude, the
perturbative approximation will be sufficiently close to the exact answer
so that the iteration will converge.
Unfortunately, numerical solutions are most desirable precisely for
those regions in parameter space where the analytic approximations are
terrible. The need to have a crude answer in order to compute an
accurate answer is very restrictive. The second initialization scheme, the
26 John P. Boyd

“continuation” or “homotopy” method, transcends this limitation. Con-

tinuation is so important that it will be given “star billing” in the next
subsection. The basic idea is very simple, however.
If one solves the wave equation for small steps in the amplitude a (or
+
any other parameter), then u(a 6a) = u ( a ) + O(Sa). This implies that
+
one may use the solution for u ( a ) as a good first guess for u(a 6a),
+
which in turn furnishes the initialization for u(a 26a), and so on. The
parameter march begins at a small amplitude where the perturbative
approximation is the initialization for the first computed amplitude. For
all larger amplitudes, the converged result of the iteration for each a is
the first guess for the next amplitude. One may thus march from very
small amplitude to very large amplitude in small, smooth steps.
Unfortunately, there are some interesting solitons that do not have
small amplitude limits! The geophysical and plasma vortices known as
“modons” (Flierl, 1987) and the solitons of the Kuramoto-Sivashinsky
equation (Chang, 1986) and the Flierl-Petviashvili monopole (Boyd,
1988d) exist only for finite amplitude. There are two popular strategies
for generating a first guess when analytical solutions are unavailable.
The first is to solve the wave equation using a very low order finite
difference or spectral scheme. Boyd (1988d), for example, faced the
problem
w, -r1wr - w - w2= 0.
+ (4.1)

Neglecting the nonlinear term gives an equation whose only solution is

the trivial one, so perturbation theory fails. Instead, Boyd (1988d)
truncated the rational Chebyshev pseudospectral series to the lowest
basis function:

Substituting (4.2) into (4.1) and demanding that the residual function
should vanish at the single interpolation point r = 2 gives a linear
equation for a, whose solution is a, = -1. This one-degree-of-freedom
approximation has a maximum error of 16% of the maximum in IuI, but
it is nonetheless a successful initialization for more accurate calculations.
Boyd (1986b) is another example.
The mysteries of choosing basis functions and collocation points will be
explained in the next subsection. The important thing is that the
analytical solution was obtained merely by specializing the numerical
procedure to a small number of unknowns. (In this case, one!)
 New Directions in Solitons and Nonlinear Periodic Waves 27

It is not necessary that the low order system be analytically soluble.

For example, if we attacked (4.1) with two basis functions, the
pseudospectral procedure would yield two quadratic equations in two
unknowns. Each equation is the equation of a conic section, so one can
determine all four solutions of the algebraic system by graphing both
hyperbolas (or ellipses) and identifying the points of intersection.
Polynomial systems of higher degree can be solved through library
software (“black boxes”) described by Watson and Scott (1987) and
Morgan (1987). Most of the roots of the algebraic system do not
correspond to differential equation solutions, but one can test all low
order roots by inserting them into the high resolution code. The “good”
solutions to the low order system are those that initialize convergent
iterations when the numerical resolution is increased.
The fourth strategy for initialization is to modify the homogeneous
boundary value problem by adding an artificial inhomogeneity. Usually,
one knows the general shape of the solitary wave; most one-dimensional
solitons qualitatively resemble the hyperbolic secant function, for ex-
ample. (The functions sech2(x) and exp(-x2) are alternatives of similar
shape.) Let us denote such a qualitative guess by uo(x). If Newton’s
method diverges when initialized with u o , we can artificially modify the
original problem.
Suppose that the goal is to solve L(u) = 0, where L is a nonlinear
operator. The “residual inhomogeneity” method modifies this wave
equation to
Lb) = (1 - t)L(uo), (4-3)
where t is an artificial parameter. When t = 0, the initialization uo is, by
construction, the exact solution of the modified equation (4.3). Obvi-
ously, Newton’s iteration will converge! After solving (4.3) for t = 0, we
can then apply the continuation method to march in small steps in t until
we reach t = 1. This technique of introducing an artificial parameter t so
that an arbitrary first guess is the exact solution for t = 0 is very popular
for solving nonlinear systems of equations. It is often called the
“homotopy method” with (4.3) as the “homotopy mapping”.
Boyd (1988d) also solved (4.1) using this residual inhomogeneity
method. The first guess W = exp(-r) is a poor approximation-in polar
coordinates, W ( r ) has a cusp at the origin! This does not matter,
however, because the continuation march in the dummy parameter t will
continuously deform the solution from the initial guess to the true
solution W ( r ) .
28 John P. Boyd

2. Iterations

The basic algorithm for any nonlinear problem is Newton’s iteration. If

the discretization is applied first, then the wave equation is transformed
into N nonlinear algebraic equations for the values of the soliton at each
of the N grid points. The iteration for a system of algebraic equations is
usually called the Newton-Ralphson method. Alternatively, as discussed
at greater length in the next subsection, one may apply the iteration
directly to the nonlinear differential equation. This tactic is the Newton-
Kantorovich method; one must solve a linear boundary value problem at
each iteration. Either way, the convergence is second order in the sense
that, close to a root, the number of correct digits doubles at each iteration.
Unfortunately, the Newton iteration is rather expensive because one
must solve a system of N linear equations in N unknowns at each step.
For this reason, quasi-Newton variations are very common. The simplest
tactic is to compute and factorize the Jacobian matrix only once and
merely backsolve to compute all the later iterates. The second order rate
of convergence is reduced to geometric convergence, that is, e,+l = e,/A,
where ei is the error of the i-th iterate and A is a constant > 1. The cost
per iteration is greatly reduced, however.
Many other quasi-Newton schemes are explained in texts on optimiza-
tion. However, there have been several successful applications of
non-Newton methods to solitons.
Eydeland and Turkington (1987) recast the quasi-geostrophic equation
for Rossby solitons in an infinitely long channel into a variational
problem. Their iteration has only a geometric rate of convergence, but
there are two major rewards for this (relatively) slow convergence rate.
First, they are able to rigorously proue that their iteration will converge
from an arbitrary first guess, eliminating the initialization problem and
simultaneously placing their work on a firm mathematical footing.
Second, Poisson’s equation is solved at each iteration instead of a more
complex partial differential equation. Since this is separable, the solution
requires only O(N2log, N) operations, where N is the number of grids
points in each coordinate. In contrast, a standard Newton-
Kantorovich/finite difference solution of the quasi-geostrophic equation
would require solving a nonseparable partial differential equation at each
step, which is O(N) more costly. Ingersoll and Cuong (1981) and Verkley
(1989) used similar iterations. In Subsection D, we show how these
schemes can be interperted as time-marching solutions to artificial
diffusion problems.
 New Directions in Solitons and Nonlinear Periodic Waves 29

Finally, a third strategy is to employ a nonlinear generalization of

Richardson’s iteration (“preconditioned Newton flow”). This, too, has
only a geometric rate of convergence. However, the nonlinear Richard-
son iteration combines the high accuracy of spectral algorithms with the
low cost of finite difference methods. This very promising tool will
receive “star billing” in Subsection D.

3. Dkcretizations : Finite Difserence, Spectral, and Special

Second-order finite differences have been popular with many authors

including Tung, Chang, and Kubota (1982), Ingersoll and Cuong (1981),
and Eydeland and Turkington (1987, 1988). Since these are discussed in
every numerical analysis text, we merely note that these algorithms are
easy to program but give only moderate accuracy unless one uses a
prohibitive number of grid points.
Special numerical discretizations are quite common because most
nonlinear wave specialists are good (and ingenious) mathematicians.
Deem and Zabusky (1978) and later papers reviewed by Dritschel (1988)
employ “contour dynamics”. The Euler equations for a fluid are
simplified by assuming that the flow consists entirely of patches of
uniform vorticity surrounded by irrotational motion. With this simplifica-
tion, the two-dimensional dynamics is reduced to computing the shapes
of the one-dimensional contours that bound the patches of constant
vorticity. Most of this work has focused on vortex merging, stability, and
other time-dependent processes. However, Deem and Zabusky (1978)
show some interesting vortex pairs that are independent of time except
for a uniform translation. Pierrehumbert (1980), Wu et al. (1984), and
Tanveer (1986) also give solutions for pairs of uniform vorticity patches.
Free surface water waves are commonly attacked via complex variable
techniques, which reduce the dimensionality by (effectively) eliminating
the vertical coordinate. Chen and Saffman (1980) solved an integro-
differential equation via sixth-order differences, then checked their code
by solving a different but equivalent problem via pseudospectral
methods.
Vanden-Broeck (1986) solved a complex free surface condition by
expanding the complex velocity W in terms of the modified Fourier
expansion
1+
W ( f )= (1- /3e-2nif)1/3(
m

n=O
>.
dne-2ninf (4-4)

where /3 is, along with the phase speed c and a finite number of the
30 John P . Boyd

coefficients d , , one of the unknowns. For the highest wave, Stokes

showed that the crest has a corner with an angle of 120°, which implies
p = 1 and W ( f )-f"'. The variable allows the expansion to mimic the
near-singularity for very steep waves that are just short of maximum
steepness.
The last special method discussed here is the Stokes perturbation series
in powers of the wave amplitude. This technique was illustrated for the
double cnoidal wave at the end of Section 11. The order-by-order
calculation of terms is usually regarded as a sequence of symbolic
computations rather than a numerical iteration. Nevertheless, there is an
initialization: the linear cosine wave is the lowest order approximation.
Like the quasi-Newton iterations, the Stokes series converges geometri-
cally: an error that is proportional to a" for some a is the definition of
geometric convergence. The perturbation theory closely resembles the
quasi-Newton scheme which computes and factorizes the Jacobian just
once. At each order in perturbation theory, one solves a linear
differential equation with the same linear operator; only the in-
homogeneous terms vary with n. The perturbation theory also increases
the accuracy by one additional power of the amplitude u at each order. In
contrast, each Newton iterate is a rational function of the amplitude. If
the denominators are expanded in powers of the amplitude u to convert
this into a polynomial in u like the perturbative approximation, one finds
that the n-th iterate is accurate to the 2"-th power of a. The degree (in
amplitude) doubles at each Newton iteration. Thus, perturbation theory
is a quasi-Newton iteration.
Since the Stokes series has a finite radius of convergence (or perhaps is
asymptotic), it has a smaller range than other algorithms. However, the
series often converges well into the soliton regime, and PadC ap-
proximants constructed from the Stokes terms usually double the
amplitude range where a given order of approximation is accurate.
Schwartz (1974), Boyd (1986a), Andersen and Geer (1982), and especi-
ally the reviews by Van Dyke (1975, 1984) illustrate PadC and other
series-extension methods.
The final discretization strategy is to expand the unknown in a series of
global expansion functions. The choice of basis functions is determined
partly by the geometry and partly by personal preference.
When the geometry is periodic in a given coordinate, the natural basis
functions are the sines and cosines of a Fourier series. Examples include
Haupt and Boyd (1987, 1989), Boyd (1986a), and most applications of
 New Directions in Solitons and Nonlinear Periodic Waves 31

Stokes series to water waves including Schwartz (1974). Meiron, Saffman,

and Yuen (1982) computed free surface waves by expanding the velocity
potential in a two-dimensional Fourier series with coefficients that vary
with height so that each term satisfies the three-dimensional Laplace
equation.
When the domain is a channel, i.e., finite in one coordinate with
nonperiodic boundary conditions at the sidewalls, the best basis functions
for the cross-channel coordinate are Chebyshev polynomials. Numerical
computations of the geophysical solitary waves known as “modons” also
use Chebyshev polynomials even though the domain is unbounded. The
reason is that modons have nonzero vorticity only within a finite region
bounded by a closed curve-an ellipse in Boyd and Ma (1989). Because
the flow outside this ellipse is analytically known, the computational task
is confined to the interior of the closed curve. Boyd and Ma solved their
interior problem in elliptical coordinates by combining Fourier functions
in the quasi-angular variable with Chebyshev polynomials in the quasi-

FIG.4. Streakfunction for an elliptical modon from Boyd and Ma (1989). (The
streakfunction is the streamfunction as seen in a coordinate system moving with the wave.)
Positive contours are solid, while negative values of Y are dashed. The boundary between
the interior and exterior flow is the contour Y = 0, an ellipse that is shown with a bold line.
By exploiting the four-fold symmetry (symmetric in n and antisymmetric in y), one may
compute this case to six decimal place accuracy with only 200 basis functions.
32 John P. Boyd

radial coordinate. Figure 4 shows the streamfunction (including the

analytically evaluated exterior flow) for a typical case.
On an infinite interval, one has a variety of good choices: rational
Chebyshev functions, sinc series, and domain truncation with either
Chebyshev or Fourier functions. All give spectral accuracy, but the
rational Chebyshev functions (Boyd, 1982, 1987a, 1987b) are the most
flexible and will be discussed at greater length in Subsection C. Boyd
(1988b, 1988d) gives examples.
For all these two-dimensional problems, these one-dimensional basis
functions are combined as tensor products. For example, a good choice
for the infinite channel is to define @,,,,,(x, y) = TB,(x)T,(y), where the
TB, are the rational Chebyshev functions (infinite interval in x ) and the
T, are the Chebyshev polynomials (for the cross-channel coordinate y ).

C. THENEWTON/PSEUDOSPEC~RAL/CONTINUATION
POLYALGORITHM:
A CLOSERLOOK

To avoid the review-as-a-list-of-references disease, this section will give

a detailed description of how to compute a solitary wave using particular
choices for the initialization, iteration, and discretization. The range of
options is so great that one cannot possibly cover all the possibilities in
depth. The particular choices we shall discuss here, however, represent
the mainstream of solving nonlinear boundary value problems.
The Newton-Ralphson and Newton-Kantorovich algorithms are
equivalent and require the same work. We will discuss the latter-
linearizing first and then discretizing the partial derivatives-because this
makes it easier to understand important techniques such as matrix
preconditioning and the nonlinear Richardson iteration explained below.
To illustrate these ideas, we shall solve the particular wave equation

2 YU“
u,,+u,,-y u----=o (EMR equation) (4.5)
c c

on the doubly infinite interval, x E 1-03, m], y E [-a,m]. The physical

background and the reason for the name “EMR equation” are given in
Boyd (1988d). Here, it will suffice to note that the EMR equation is an
approximate model for planetary-scale Rossby waves which are trapped
by Coriolis forces in a narrow band around the equator, y = 0.
 New Directions in Solitons and Nonlinear Periodic Waves 33

The Newton-Kantorovich iteration begins by writing

= (0
- u ( x , y ) + A(x, y ) , IAI << 1,
(4.6)
where the superscript is the iteration number. Strictly speaking, the
correction A ( x , y ) should have a superscript ( i ) also, but this has been
omitted for notational simplicity. The key step is to substitute (4.6) into
(4.5) and then linearize by neglecting terms of O ( A 2 ) . This gives the
linear, inhomogeneous boundary value problem that must be solved at
each step:

(4.7)
One then iterates (4.6) and (4.7) until A(x, y) is negligibly small.
To solve (4.7), we employ the rational Chebyshev functions defined by
TB,,(x) = cos(nt),
where t ( x ) is defined by
x = L cot t,
where L is a user-choosable constant (“map parameter”), which allows
one to adjust the scale of the basis functions to that of u(x, y ) . The
x-derivatives of the basis functions can be calculated by exploiting the
fact that the TB,,(x; L) are merely the terms of a Fourier cosine series in
disguise. The first six derivatives are given in Boyd (1987a). For the
present application, it suffices to note that
sin2t
TBn,x =-- [ - n sin(nt)],
L
(4.10)
sin4t 2 cos t sin3t
TB,,, = -[-n’cos(nt)]
L2
+ L2
[ - n sin(nt)].

When the basis functions are substituted into the differential equation,
H A =f, one obtains the residual function
cc

~ ( xd ;l , * * * , dN) ~ f ( x-) C d n [ ~ + n ( x ) I , (4.11)

n=l

where the d,, are the spectral coefficients of A ( x ) and the +,,(x) are the
basis functions. The goal of all spectral methods is to determine the a,, so
as to make the residual function as small as possible in some sense.
R ( x ) = 0 for the exact solution.
34 John P . Boyd

Galerkin’s method obtains N conditions by multiplying (4.11) by each

of the first N basis functions in turn and integrating. It is usually simpler
and almost equally accurate to apply the “pseudospectral” or “colloca-
tion” method, which imposes the condition
R(x,;d l , . . . , d N )= 0, i = 1, . . . , N , (4.12)
where the “interpolation” or “collocation” points in x are the mapped
images of an evenly spaced grid in the trigonometric coordinate t:

xi = L cot (y; I)), i = 1, . , . , N . (4.13)

The grid (4.11) does not include the endpoints (t = 0, n *-*x = kw), so
(4.9) and (4.10) are well-defined and nonsingular at all xi.
Both Galerkin’s method and the pseudospectral algorithm convert the
differential equation into an N x N matrix problem whose solution is the
vector of basis coefficients. For the latter, the linear algebra equation is
Jd = r, (4.14)
where d is the column vector of the spectal coefficients of the correction
A ( x ) ; r is the column vector of residuals of u(’)(x)at the i-th interpolation
point
ri = R(xi; al, . . . , uN), i = l , . . . ,N , (4.15)
where the ai are the coefficients of d i ) ( x ) ; and J is the Jacobian matrix
whose elements are
Jij = ( H @ j ) l x = x ,
3 (4.16)
where H is the linear operator acting on A in the linearized differential
equation. For (4.5),

Unlike finite difference matrices, the pseudospectral Jacobian matrix is

dense rather than sparse, but the reward is that to obtain moderate
accuracy (say 5%), the number of degrees of freedom is smaller-
typically by at least a factor of two in each coordinate. The rational
Chebyshev functions are so efficient that Boyd (1988d) computed the
solitons of (4.5) to three correct digits using only a total of 120 basis
functions in two dimensions. Although there are exceptions such as the
solitons of the Kuramoto-Sivashinsky equation, solitary waves are
 New Directions in Solitons and Nonlinear Periodic Waves 35

usually rather smooth functions. Good two-dimensional solitons from 100

basis functions are common.
Because the pseudospectral matrix is small, one can afford to employ a
simple, direct method to solve the linear algebra problem: Gaussian
elimination (Crout reduction). Factoring a 100 x 100 matrix requires only
about 60 seconds on an I B M A T and less than 10 seconds on a work
station. For difficult problems (N >> loo), one can drastically reduce both
operations and storage by using the preconditioned nonlinear Richardson
iteration described in the next subsection.
Two other tricks also greatly improve efficiency. First, solitons often
have symmetries. The most interesting solitons of (4.5), for example, are
symmetric in x and antisymmetric in y. One may halve the size of the
basis set in each coordinate by keeping only those basis functions that
match the symmetry of the solution. For the rational Chebyshev
functions, it is easy to show that all the even degree functions are
symmetric about the origin, while TBI, TB3, and all the other odd
functions are antisymmetric. When the basis is halved, one must
simultaneously halve the interpolation grid by replacing the 1 / ( 2 N ) in
(4.13) by 1/(4N).
The second trick is that since solitons vanish at infinity, one may
modify the basis by defining combinations of the TB, that vanish at
infinity. (This is strictly optional, but does make the computation more
efficient and is especially helpful when computing a low order spectral
solution to initialize the Newton iteration.) We may take

(4.17)

The new basis functions retain the even or odd symmetry of the TB,(x)
from which they are constructed.
For simplicity, we have explained the key ideas in one dimension, but
the extension to two dimensions is trivial. Normally, one employs a
tensor product basis. That is,
&,(x, y ) = @i(x)c$j(y), i = 1, . . . ,M, j = 1, . . . ,N , (4.18)
for a total of M N basis functions. The collocation points are of the form
(xi, yj), where xi and yj are both given by (4.13). The number of
collocation points and the map parameter L may be chosen independ-
ently for each coordinate.
The third major component of the polyalgorithm is the continuation
36 John P. Boyd

method. This, as briefly described earlier, is a strategy for tracing an

entire solution branch from a single starting point. It is still necessary to
use the other initializations described above-analytical solutions, per-
turbative approximations, or low order spectral solutions-to obtain a
decent first guess for one point on the solution branch. After u(E,,)has
been computed, however, one may apply continuation to compute the
solution for all E.
Zeroth-order continuation means using the solution for U ( E , - ~ ) as the
first guess for u(E,). A better procedure-after one has computed u for a
second value of E via zeroth-order continuation-is to fit a straight line
through the two previous values:

Note that one may use a variable step size in the parameter E. This
“first-order” continuation may be generalized by fitting polynomials of
higher degree. Boyd (1988d), for example, used a library routine that fit
a polynomial of up to fourth degree. It was called twice for each value of
E: once at the top of the loop to generate a first guess and once at the
bottom of the loop with a different flag to store U ( E , - ~ ) (along with
previously computed solutions) in preparation for the next extrapolation.
The only difficulty is that continuation fails whenever the Jacobian
matrix is singular. This occurs wherever the discretized, linearized
differential equation has a zero eigenvalue. One such breakdown occurs
at a bifurcation point where two independent solution branches intersect.
However, Keller (1977) showed that it is possible to “shoot the
bifurcation point” by trying various parameter steps in E until the Newton
iteration converges on the far side of the bifurcation point. Seydel (1988)
and Chen and Saffman (1980) describe how to switch branches at a
bifurcation point to trace the second branch, too.
The other class of breakdown is a “limit point” or “fold” where the
solution branch curves back upon itself so that no solutions exist for
E > climit,while there are two solutions for E > The classic example
is a small amplitude periodic solution. For the Korteweg-deVries cnoidal
wave of period 2n,for example, two solutions exist for all c > -1, while
there are no solutions with smaller phase speed. The two solution
branches that join at c = -1 are approximately given by
a*
u(x;a)=flalcos(x-ct), c=-1+-, lal<<l. (4.20)
24
 New Directions in Solitons and Nonlinear Periodic Waves 37

For all amplitudes, these two branches differ only through a phase shift
of n.
Parenthetically, we note that the Stokes expansion differs from other
quasi-Newton iterations because of the nonsecularity condition that gives
the corrections to c(a). This arises because the Stokes series is an
expansion about a limit point (a = 0) rather than one where the Jacobian
matrix J is nonsingular.
The standard remedy for a "fold" is to use the arclength of the curve as
the marching variable instead of E (Seydel, 1988). Arclength continuation
will smoothly trace the solution around the limit point from the lower
branch to the upper branch; the price is that the linear algebra problem is
increased in size from N to N + 1 and more programming is needed.
For many wave problems, however, a simpler strategy is to use the
amplitude rather than the phase speed as the marching parameter.

-3.0 -20 -10 0 1.0 2.0 1

30 3.O

2.0 20

I .o I.o

0 0

-1.0 -1.0

-2.0 -2.0

-3.0 -3.0
-2.0 -1.0 0 1.0 2.o 3.0
FIG.5 . A composite showing the dipole vortex of the EMR equation in one quadrant
for each of four different cases: c = 1 (upper right), c+m (lower right), c+ --oo (lower
left), and c = -1 (upper left). Each case may be reconstructed from the quarter-plane since
u(x, y) is symmetric about x = 0 and antisymmetric ( u ( x , y ) = -u(x, - y ) ) with respect to
y = 0.
38 John P. Boyd

Equation (4.20) and the double cnoidal Stokes series (Section 11) are
examples. Both define the amplitude parameter to be the lowest Fourier
coefficient, that of cosx. This choice is easy to program, but other
definitions of amplitude are mathematically and computationally accept-
able. For some examples such as the fifth-degree Korteweg-deVries
equation (Boyd, 1986a), this change-of-parameter is sufficient to avoid
limit points except for the trivial one at zero amplitude. (Because the
lowest order approximation (4.20) gives a good initialization for the
branches that meet at zero amplitude, arclength continuation is unneces-
sary at that fold.) Another virtue of the fixed-amplitude numerical
solutions is that they are easier to compare with Stokes series, which also
fix a. However, if there are no limit points, then one may use the phase
speed c as the marching parameter over all of parameter space. The
author uses whichever of c and a is most convenient for a given problem.
As noted earlier in discussing the residual inhomogeneity initialization,
one is always free to march in an artificial parameter when a physical
parameter is inconvenient. The “four-eyed” contour plot in Figure 5 is a
good illustration. The lower two cases are c+ fw. The obvious way to
compute them is to simply march in small steps from small negative c
(where a perturbative solution is accurate) to large Ic[. However, this is
very costly because the limit of infinite c is approached very slowly, as
O( UC).
A much better strategy is to think. First, a simple scale analysis shows
that the - u / c term is negligible in comparison to the others as 1c(+ m,
which reduces (4.5) to

u, + uyy- y2u - yu2

-- 0
C
(“Large-Jcl EMR equation”). (4.21)

Second, it is trivial to show that the solution of (4.21) for all c is given
exactly by
u(x, Y ) = cv(x, y ) , (4.22)
where v ( x , y ) is a single, universal function that solves (4.21) for the
special case c = 1. Equation (4.22) further shows that the soliton for
c = -m is merely the negative of that for c = m: the two branches connect
through I c l = w, and it is necessary to compute only a single function
v(x, y) to obtain both cases in the lower half of Figure 5 .
This parameter-free function was computed by marching in c to c = -1
and then solving
u,, + uyy-y2u + y u 2 + (1 - t)U = 0. (4.23)
 New Directions in Solitons and Nonlinear Periodic Waves 39

When z = 0, (4.23) is simply the EMR equation for c = -1. Marching to

z = 1 then gives u = -v(x, y). Instead of calculating the limit ( c J + t~ by
marching hundreds of steps in c, we compute the limiting solution
through (4.23) in just four or five steps in the artificial parameter z.
The four-fold symmetry of the solutions to this wave equation is very
helpful. This symmetry not only reduces the number of basis functions by
a factor of four, as described above, but also allows equal economy in
displaying the answer: four cases are combined in a single graph.
Clearly, the Newton/pseudospectral/continuation polyalgorithm is very
powerful and general. However, Werner Rheinboldt’s quote is still
relevant: “Applied mathematics is mathematics-plus.” In the next
subsection, we discuss some methods for reducing the cost of what is
almost always the rate-determining step: solving the matrix problem
(4.14).

D. THENONLINEAR (“NEWTONFLOW”)
RICHARDSON ITERATION
AND
TIMEMETHODS
OTHERARTIFICIAL

Newton’s iteration for the nonlinear system r(u) = 0 is equivalent to

applying the Euler forward marching scheme with unit time step to the
system of differential equations

du
(“Newton flow” equation), (4.24)
dT

whre T is a fictitious time variable and 1/J is the inverse of the Jacobian
matrix. Converting the problem from an algebraic to a differential system
seems crazy at first, but it is a very powerful conceptual and computa-
tional tool. All the theorems and all the algorithms of differential
equations can now be brought to bear to solve (4.24), which is known as
the ‘Newton flow” equation. Chu (1988) is a good review of such
“continuous realizations of iterative processes,” as his paper is titled. See
also Peitgen and Richter (1986), who show that the attractor basin
boundaries for the Newton flow are smooth while those for the Newton
iteration are usually fractal.
If we expand r(u) in a multi-dimensional Taylor series centered on a
40 John P . Boyd

root q,, (4.24) becomes

(4.25)

= -(u - q,) + O([u - ug]’) (4.26)

Thus, every root of r(u) = O is a local attractor. However, the rate of
convergence is geometric rather than quadratic except for Euler’s method
with unit time step. The residual decays with the artificial time T as
r(u( T ) )= e-’r(u(0)) (4.27)
for all initial conditions, even those far from a root. However, (4.27) is
not sufficient to guarantee success because the solution could diverge to
infinity or converge to the trivial solution u = 0.
The Newton flow concept opens up intriguing computational pos-
sibilities because one may replace the inverse Jacobian matrix by an
arbitrary matrix without altering the fact that each root of the residual is
a critical point of the differential equation. This freedom is enticing
because the pseudospectral Jacobian matrix is dense. Factorizing the
Jacobian to compute the Newton correction may be prohibitively
expensive for multi-dimensional problems. If we use N basis functions in
each coordinate, then the cost is O(2N6/3) operations for each iteration
in two dimensions. Ouch!
The simplest replacement for the inverse Jacobian is the identity matrix,
but unfortunately there is no guarantee (or even likelihood) of conver-
gence. If, however, the inverse Jacobian is replaced by a preconditioning
matrix, that is, a matrix whose inverse approximates the inverse Jacobian,
then it may be possible to prove fast convergence.
Orszag (1980) showed that an excellent pseudospectral preconditioning
matrix is that of the second order finite difference approximation to the
same problem on the same grid. Let the finite difference matrix be
denoted by H. Orszag demonstrated that for a second-order differential
equation, the matrix (1/H) J has eigenvalues that all lie on the interval
a;, E [l, 2.471. If the Newton flow equation is modified to
du
-= - (A).(.) (“Preconditioned Newton flow”), (4.28)
dT
then Taylor expansion near a root gives
du = -(f)J(u
- -~ 0 )+ O([u - UO]’). (4.29)
dT
 New Directions in Solitons and Nonlinear Periodic Waves 41

Orszag’s theory implies that each root is still a local attractor. If the error
u - ~0 is decomposed into the eigenvectors of (l/H)J, then the error in
different components will decay with T at various rates ranging between
exp( - T) and exp( - T/2.47).
Because of these varying rates of decay, it is not possible to find a
magic time step that will retain the quadratic convergence of Newton’s
method. However, the Euler forward scheme does give very rapid
geometric convergence: with a time step of $, one can guarantee that
each component of the error will decay by at least a factor of 2.33 at each
iteration. Since the factorization of the finite difference matrix is cheaper
than that of the pseudospectral Jacobian by a factor of O ( N 2 )in both one
and two dimensions, the thirty or so preconditioned iterations that are
needed to achieve spectral accuracy are orders of magnitude cheaper
than the corresponding four or five Newton iterations. When N = 100 per
coordinate, the preconditioned iteration may be cheaper by a factor of
2000!
Orszag developed this preconditioning for linear equations; in this
context, the forward Euler scheme is known as the Richardson iteration.
Thus, the “Preconditioned Newton Flow Iteration”, which follows from
applying the Euler method to (4.28), may alternatively be dubbed the
“Nonlinear Richardson Iteration”:

Hun+’= Hun - m(un) (“Preconditioned Newton Flow Iteration”),

(4.30)
where t is the time step.
Although the author devised (4.30), it is such an obvious idea that
others have independently invented it, too. This iteration is briefly
mentioned in the book by Canuto et al. (1987, p. 390) and also in a
conference paperftechnical report by Streett and Zang (1984), but the
author has not yet found it in a regular journal article. Nonetheless, it is a
very powerful technique.
The finite difference preconditioning requires that the unknowns be the
values of u ( x ) on the collocation grid, rather than the Chebyshev or
Fourier coefficients. However, this requires only the trivial modification
of replacing the Chebyshev polynomials or functions by the correspond-
ing “cardinal functions” (Boyd, 1989d).
For expository purposes, we shall use the simplest of cardinal
functions: the sinc or “Whittaker cardinal” functions (Stenger, 1981)
42 John P. Boyd

functions: the sinc or “Whittaker cardinal” functions (Stenger, 1981)

defined by
sin(nx)
sinc(x) = ~, (4.31)
JGX

which have the property that

sinc(0) = 1 and sinc(j) = 0 for all integers j Z 0. (4.32)
It follows that if we write
m

u(x)= 2
j=--m
ujr&(x; h ) , (4.33)

where h is the spacing of the grid, a constant, and

x-jh
@j(x;h) = s i n c ( k ) , j =O , f,f2,.. . , (4.34)

then the values of u ( x ) on the evenly spaced grid

xi = jh (4.35)
are simply
u(x,) = uj , (4.36)
so that the spectral coefficients are the grid point values.
The mechanics of differentiating the sinc functions, computing second-
order centered differences on a uniform grid, and solving (4.30) with a
band-solver are described in Boyd (1989d). The infinite interval is
truncated by assuming that all uj with 1jl> N are negligibly small and can
be replaced by zero. Figures 6 and 7 show the computation of the solitary
wave of the second order form of the Korteweg-deVries equation.
The accuracy-a maximum pointwise error which is 200,000,000 times
smaller than the maximum of the soliton-clearly illustrates the power of
pseudospectral methods. Although the Fast Fourier Transform cannot be
combined with sinc functions so that the cost of computing r(u) is O(N2)
(versus O(N log, N) for Chebyshev and Fourier pseudospectral algor-
ithms), each preconditioned Newton flow iteration required only about 4
seconds on a 6 Mhz IBM AT, even with 201 grid points. In contrast, the
LU factorization of a 201 X201 dense matrix would cost about 400
seconds. Although we need about 20 iterations to achieve spectral
accuracy instead of the 4 to 5 iterations of Newton’s method, the
 New Directions in Solitons and Nonlinear Periodic Waves 43

10-8 '
0 5
I I

10
I

15
I

20 25
t

30
Iteration Number
FIG.6. The maximum pointwise error as a function of iteration number for the
preconditioned Newton flow iteration. The differential equation is the once-integrated
Korteweg-deVries equation, u, + ( 4 2 - c)u = 0 for a phase speed of 4, subject to the
boundary conditions u ( * m ) = 0. The time step is 0.6 and the error decreases by a factor of a
little over 2.5 at each iteration until plateauing at about 5 X lo-*, which is the maximum
pointwise error of the spectral calculation with 201 grid points and a uniform grid spacing of
1
-
10'

preconditioned Newton flow is about 20 times faster overall. Better yet,

memory storage is also reduced by a factor of N/3. (Note that one may
compute r(u) without calculating and storing the Jacobian matrix.)
It is possible to solve the third derivative form of the Korteweg-
deVries equation, too. The crucial trick is to use a staggered grid in
which the finite differences and pseudospectral derivatives are evaluated
at
xj=h(j+$), j=-N,. . ., N , (4.37)

while the sinc coefficients, i.e., the unknowns, are the 2N 1 values of +
u ( x ) on the unstaggered symmetric grid (4.35). The nonlinear Richardson
iteration converges more slowly than that for the second order
equation-Orszag (1980) explains that the condition number of the
preconditioned matrix is poorer as the order of derivatives increases-but
it is still fast and efficient.
44 John P . Boyd

FIG. 7. Same case as the previous figure: u ( x ) for the first guess, the first three iterates,
and the final converged result, which is indistinguishable from the exact soliton. The
computational interval is [-lo, 101,but only a subinterval is shown because the soliton is
symmetric about x = 0 and exponentially small for large 1x1.

The book by Boyd (1989d) explains the use of other cardinal functions
with extensive differentiation tables, etc. However, the Fourier and
Chebyshev cardinal functions are very similar in shape to the sinc
function and are also given by simple analytical expressions.
Canuto et al. (1987) thoroughly discuss preconditioning and iterations.
Incomplete LU factorizations are a popular alternative to strict finite
difference preconditioning for multi-dimensional equations because even
finite difference matrices are expensive to invert in more than one
dimension (unless the equation is separable). Pseudospectral multigrid
may be the ultimate weapon for direct computation of solitons. This, too,
however, fits into the framework of the “Newton flow.”
The iterations of Eydeland and Turkington (1987, 1988) and Petviash-
vili (1976) and Petviashvili and Tsvelodub (1978) may also be justified by
applying Euler’s forward marching method to a differential equation in
artificial time. (These authors justified their schemes via variational
principles, however.) For example, Eydeland and Turkington solve
vxx + v y y + F(3 + CY - P I =O, (4.38)
 New Directions in Solitons and Nonlinear Periodic Waves 45

where c and p are eigenparameters and F ( z ) is a known function. If write

this as the generalized diffusion equation
v T = v x x + v y y + F(3 + CY - P I , (4.39)
where T is the artificial time coordinate, the Euler one-step method is
equivalent to the “power” method for solving eigenvalue problems.
If F were linear (as for analytically soluble modons (Flierl, 1987)), then
one could solve (4.39) via separation of variables. The high order modes
are (approximately) proportional to exp(ikx + imy) and decay with time
as exp(-[k2 + m 2 ] T ) .The gravest mode, that is, the eigenmode with the
simplest spatial structure, may either grow or decay, depending upon
F ( z ) . In any event, this mode will decay (perhaps at a negative rate!)
more slowly than any of the higher modes. Consequently, for sufficiently
large T, the solution of (4.39) is dominated by the lowest eigenmode.
Thus, the structure of v converges geometrically to that of the gravest
eigenfunction of the operator on the right hand side of (4.39). The
eigenvalue (if there is only one) may be determined by monitoring the
growth or decay rate of lvl for T >> 1.
When the problem is nonlinear, the qualitative behavior is exactly the
same. The power method is a very simple and effective way to determine
the gravest mode for both linear and nonlinear problems. The only
complication is that when F ( z ) is nonlinear, the changes in the amplitude
of )t will alter the shape of F, and this in turn will change both the
eigenvalue and eigenfunction. To specify a unique solution, one must
demand
ll Wll = a , (4.40)
where a is some amplitude parameter and the 11 11 denotes some norm
or measure of amplitude, and then enforce this constraint at each
iteration. Actually, most texts recommend applying (4.40) even for linear
problems (where it is not absolutely necessary) to avoid roundoff
problems.
To impose this constancy-of-amplitude condition, note that the Euler
forward scheme gives
lp+1= lp‘+ t{qix+ lpyy + F ( v j + d + l y - pj+l)}, (4.41)
where z is the artificial time step. If lp is known, then taking the norm of
both sides of (4.41) gives a nonlinear equation that can be solved for P ’ .
For Eydeland and Turkington’s problem, it is necessary to impose two
constraints like (4.40) to specify a unique solution. Applying these
46 John P . Boyd

(independent) conditions to (4.41) gives two nonlinear equations in the

two unknowns d+’, pj+’. The principle is the same, however, independ-
ent of the number of constraints.
One simplification is that it is not necessary that the “norm” be a norm
in a strict mathematical sense. For example, Eydeland and Turkington
prove that the vorticity is positive definite for their problem, so they use
the area integral of the vorticity as one of their constraints. This
definition of amplitude does not satisfy the mathematical definition of a
norm, but has the virtue of being easy to impose on the right hand side of
(4.41).
There is one remaining problem: the rapid decay of the high modes in
(4.39) demands that the time step t in (4.41) should be very, very small.
Eydeland and Turkington eliminate this by replacing 3 T in (4.39) by
-VuT - V y y pIf we define the vorticity w (strictly speaking, the negative
of the vorticity) and also the Green’s function for the Poisson equation G
via
-(VXX+ VYy) w * 3 = Go, (4.42)
then the pseudotime equation becomes
wT=-w + F(Gw + cy - p ) . (4.43)
The high order modes, which are approximate eigenfunctions of the
Laplacian operator, now all decay like exp(-T) due to the first term on
the right hand side in (4.43)). Consequently, a unit time step is stable and
the iteration becomes simply
,.-,j+l= F ( G , - ,-
~ &+Iy - pj+l), (4.4)
which is the form Eydeland and Turkington actually use. Verkley (1989)
independently discovered and used the same iteration to compute vortex
pairs on the sphere.
The philosophy of eliminating the huge spread in decay rates is similar
to the preconditioning of the Newton flow algorithm. However, the
“preconditioning” here is not intended to approximate the full Jacobian
matrix but only the part that controls the high order modes. (Here, this
part is the discretization of the Laplacian operator.) The reward for this
cruder preconditioning is that at each iteration, one only needs to invert
the Laplace operator-not a general, nonseparable partial differential
equation. The cost of solving the Poisson equation on an N x M grid is
only O(10 N M ) operations via multigrid (Briggs, 1987).
 New Directions in Solitons and Nonlinear Periodic Waves 47

The danger of “imperfect” preconditioning is that (4.44) may have

modes that might diverge if the iteration were applied with fixed c and p.
The beauty of the nonlinear power method is that c and p are n o t f i e d ,
but rather are adjusted at each iteration so that the norm of o always
stays the same.
Petviashvili (1976) used an iteration very similar to (4.44) except that
he chose to fix the phase speed rather than the amplitude with the result
that (4.44) k unstable and divergent. However, he devised a simple fix.
In his case, the right hand side of (4.44) is quadratic in the unknown.
This implies that if the first guess is too large by a factor of two, the next
iterate will be four times too large and the iteration diverges. The
behaviour is the same as that for the simple iteratiodordinary differential
equation
uT = u2 - t, u j + l = (uj)2. (4.45)
Equation (4.45) has a fixed point at u = 1, but it is unstable.
Petviashvili’s remedy is to multiply the nonlinear term in (4.45) by the
a-th power of the norm of the unknown divided by the a-th power of the
nonlinear term where a is a user-choosable parameter. For ( 4 . 4 9 , this
gives

UT = (5) n
u* - u t, uj+l= (uJ)’-n. (4.6)

+
If we write u = 1 A where A << 1, then near the fixed point u = 1, (4.46)
becomes
A T = ( l + A ) 2 - m - ( l + A ) z ( l - a)A+0(A2). (4.47)
This shows clearly that the fixed point is unstable for a < 1, but when
a > 1, A decreases exponentially fast with the pseudotime T. The Euler
one-step marching scheme converts (4.47) into a stable, geometrically
converging iteration. The best choice is a = 2 , whcih gives the exact
answer in one iteration.
When u is a vector of spectral coefficients, a = 2 is still the best choice
for quadratic nonlinearity. Equation (4.44) is replaced by

{ IIF(Gw‘‘Iu’’- CY - CL II
JnF(Gw‘- cy - p). (4.48)

We have omitted the superscripts on c and p because one can specify

these quantities with Petviashvili’s iteration. In his scheme, the two
48 John P. Boyd

amplitude constraints are variable and (c, p ) are fixed; Eydeland and
Turkington’s method fixes the amplitude constraints and computes c and
p during the iteration. These two alternatives of fixed amplitude/variable
phase speed versus fixed clvariable amplitude are almost always available
in soliton studies. The choice is strictly a matter of convenience.
The “nonlinear power” method of Eydeland and Turkington (and
Verkley), the “iteration-with-stabilizing-factor” of Petviashvili, and the
“nonlinear Richardson iteration” of the author and Streett and Zang
(1984) all have geometric rates of convergence in contrast to the
quadratic convergence of Newton’s method. The reward for slower
convergence is much lower cost per iteration. Furthermore, all three of
these algorithms require only O ( N M ) storage where NM is the total
number of grid points, and all can be combined with any type of
discretization.

E. SUGGESTIONS
AND GUIDELINES

The pseudospectral method with a Fourier, Chebyshev, or rational

Chebyshev basis is the recommended discretization. Finite differences
and finite elements work just fine, but are limited to rather modest
accuracy. In contrast, the pseudospectral error decreases exponentially
fast with N so that one can often obtain six decimal place accuracy with
only twenty or thirty grid points in each coordinate. For computing the
very small radiation coefficients of weakly nonlocal nanopterons (next
section), pseudospectral methods are almost essential.
For solving the linear algebra problem, we have always endorsed this
maxim: Never use anything but Gaussian elimination (Crout reduction)
except when absolutely necessary. The iterative schemes are much faster
and use less storage, but a few seconds’ savings is not worth the bother.
Furthermore, all iterations, even the robust schemes described above,
have limits. Eydeland and Turkington, for example, prove their scheme
converges for vortex pairs, but not necessarily for vortex quadrapoles.
Nonetheless, it is very comforting to know that there are fast, low
storage alternatives to Gaussian elimination. There is no longer any
good excuse for crude, low resolution calculations of multi-dimensional
solitons. A mainframe is not needed. A microcomputer and a couple of
hundred lines of code will do just fine.
 New Directions in Solitons and Nonlinear Periodic Waves 49

V. Weakly Nonlocal Solitary Waves

A. I N T R O D U ~ I O N

The classical definition of a soliton is: A “solitary wave” is a localized,

finite amplitude, steadily translating disturbance of permanent shape and
form and constant phase speed c. Unfortunately, this definition is too
narrow. Weakly nonlocal solitons are nonlinear waves that violate one or
more of the essential conditions in this definition, and yet closely
resemble classical solitary waves by being long-lived, stable, and
coherent.
The adjective “nonlocal” means that the quasi-soliton fails to meet the
test of “localization” because it asymptotes not to zero, but to a constant
amplitude oscillation as 1x1 increases. The adjective “weakly” means that
the wave is similar to a solitary wave, in spite of its nonlocal character,
because the amplitude of the “far field” radiation is very small in
comparison to the maximum amplitude of the solitary wave. It is
implicitly assumed that the “weakly nonlocal” wave cannot be altered into
a strict soliton through any small changes or tuning.
It is useful to subdivide “weakly nonlocal” waves into two sub-
categories, depending on precisely how the strict definition of a soliton is
violated. A “radiatively decaying soliton” is a nonlinear solution that has
the following characteristics:
(i) It is strictly localized in the sense that u(x, y , t)+O as 1 x 1 4 a.
(ii) It satisfies all other requirements of a solitary wave except that it
decays very slowly with time through radiation of energy to large
1x1.
(iii) The radiative decay is very slow because there is an almost pet$ect
cancellation between nonlinearity and dispersion.
(iv) It violates the strict definition of a soliton because the wave is not
steadily translating and permanent, but slowly disperses as it
propagates.
When a localized initial condition is given to a wave equation that
allows only nonlocal solitons, radiatively decaying solitons will dominate
the wave field for t >> 1. The nonsolitonic transients will disperse quickly,
leaving behind a coherent structure that also disperses, but at a much
slower rate. The nonlocal solitary wave is a structure in which most but
not all of the dispersion is balanced by nonlinearity.
50 John P. Boyd

This subcategory can be subdivided into “algebraically slowly” and

“exponentially slowly” decaying solitons, depending on whether the rate
of decay is an algebraic or exponential function of the appropriate small
parameter. We shall elaborate on these distinctions below.
The second subcategory, that of “steadily translating, nonlocal”
solitons may be similarly divided into two subclasses. A “nanopteron” or
micropteron” is a nonlinear solution on x E [-00, 031 that has the following
characteristics:

(i) It is permanent and steadily translating at a constant phase speed c.

(ii) It satisfies all other requirements of a solitary wave except that it
is not strictly localized because the wave asymptotes to a small
amplitude oscillation when 1x1 >> 1.
(iii) The amplitude of the far field oscillation is an exponentially small
function of the maximum amplitude of the wave for the nanop-
teron. The far field oscillation of the micropteron is an algebraic
function of some small parameter 6 << 1, that is, the oscillation is
proportional to 6“ for some n.

Figure 8 (top pair) compares these two species of weakly nonlocal

solitary waves. The radiatively decaying soliton is localized but unsteady;
the nanopteron (or micropteron) is steadily translating but nonlocal.
The theme of Section I11 was the close relationship between solitons
and spatially periodic (cnoidal) waves. Similarly, one may superimpose
nonlocal solitons with even spacing to generate a periodic “nanop-
teroidal” wave. This has an infinite number of evenly spaced peaks or
troughs, each differing only slightly from the corresponding features of
the nanopteron. The space between the large amplitude peaks is filled
with very weak oscillations whose amplitude and wavenumber match the
far field of the nanopteron. The bottom half of Figure 8 compares the
nanopteron and nanopteroidal wave. Such periodic generalizations of
nonlocal solitons are discussed at length in Boyd (1988b, 1989a, 1989d),
but the need for brevity prohibits a longer discussion of nanopteroidal
waves here.
In this rest of this article, we concentrate on nanopterons. Because
these have trivial time-dependence-steady propagation at a constant
phase speed-ne may write u(x, y , t ) = u(x - ct, y ) and compute the
waves by solving a boundary value problem (BVP). This is the same BVP
strategy as we applied numerically to strict solitons in Section IV. The
 New Directions in Solitons and Nonlinear Periodic Waves 51

X X
FIG. 8. Top left: schematic of a “radiatively decaying soliton” for I = 0 (dashed) and
some later time (solid). The dispersion need not be symmetrical. Top right: schematic of a
nanopteron. This is nonlocal, but is steady in a coordinate system travelling with the wave.
Bottom left: schematic of a nanopteron. Bottom right: the corresponding nanopteroidal
wave with a spatial period of 16.

only complication is that one must modify the discretization to cope with
the far field oscillation as described in Subsection C below.
The “radiatively decaying solitons” are much harder to analyze
because of their complex variations in both time and space. Boyd (1988~)
shows that the ordinary differential equation deldt = exp(-l/e), which is
the simplest model for the decay of a nanopteron, has multiple
logarithmic time scales. The time evolution of an actual nonlocal wave is
likely to be even more complex. Nonetheless, it is necessary to define
them because radiatively decaying solitons are what will be observed in a
numerical model with a localized initial condition. Williams and Wilson
(1988) is a good example.
Unfortunately, it is difficult to quantitatively measure the far field
radiation while the disturbance, initially confined to small 1x1, spreads
wider and wider through the emission of small amplitude waves.
Consequently, we shall concentrate on nanopterons instead. The two
species of weakly nonlocal solitons are closely related, however. One
52 John P. Boyd

may in principle use energy arguments to infer the decay rate of the
radiatively decaying soliton from the magnitude of the far field ripples of
the nanopteron (Segur and Kruskal, 1987).
The unfamiliar terminology is borrowed from zoology (Boyd, 1989a).
In classical Greek, “micropteron” means “small-wing” while “nanop-
teron” is “dwarf-wing”. The “wings” of the nanopteron, that is, the very
small ripples surrounding the large amplitude core, are marked in Figure
8 (top right).
It is unpleasant to wade through so many definitions, but this is the
price of assaulting a scientific frontier. Nonlocal solitons are still a very
new concept even though we shall discuss eight examples below. The
terms and the schematic, Figure 8, are needed to give a language to
describe these new ideas. The reward is the excitement of the pioneer,
splashing ashore onto undiscovered land.
In the next subsection, we shall show through example how to analyze
the far field oscillations. The following two subsections describe strategies
for coping with such oscillations: perturbation theory (which fails) and a
special numerical method (which spectacularly succeeds). Subsection E is
a catalogue of physical examples of both micropterons and nanopterons.

B. FARFIELDANALYSIS

The “far field” is defined as the region 1x1 >> 1 (with the soliton peak at
x = 0) where the solitary wave has decayed to such small amplitude that
its dynamics becomes linear. Because of the linearity, it is easy to prove
the following crucial result.

Theorem 5.1 (Far Field Analysis). Zf the phase speed of a nonlinear

coherent structure lies outside the range of the linear phase speeds of all
waves in the system, then it is possible that the soliton decays exponentially
as 1x1 +CQ. If c is equal to the speed of some linear wave mode for some
real wavenumber k, then the nonlinear coherent structure will be
oscillatory rather than decaying in the far field, and cannot be a classical,
localized soliton.

Proof -by-Example. Pomeau, Ramani, and Grammaticos (1988) con-

sider the generalized Korteweg-deVries equation which, after one
 New Directions in Solitons and Nonlinear Periodic Waves 53

integration, is
,,u + u,, + c 1
- - c u = 0,

where c is the phase speed and x is the spatial coordinate in a frame of

(5.1)

reference traveling with the wave. For conformity with other examples,
we have made some slight changes in notation. For brevity, we shall refer
to (5.1) as the “PRG” equation.
In the absence of the fourth derivative, Equation (5.1) reduces to the
KdV equation, which has the exact soliton
u ( x ) = 1 2 sech2(E[x
~ ~ - ct]) and c =4 2 . (5.2)
When E << 1, it is trivial to show that (5.2) is also a good approximate
solution of the PRG equation (5.1) because the fourth derivative is O ( E ~ )
smaller than the other terms in the equation.
In the far field where 1x1 >> 1, the nonlinear term is exponentially small
(at least to the extent that (5.2) can be believed), so the PRG equation
reduces to
u, +
u,, - cu = 0 (“Far field PRG”). ( 5 -3)
The solutions of this constant-coefficient ordinary differential equation
are exponentiais, each proportional to exp(ikx); the wavenumbers are
given by the linear dispersion relation
k2 = 4 f i(1+ 4~)’”. (5.4)
When the phase speed c > 0, as it is for the soliton, then one root for k2
is negative, implying that k 2:fic’”. This root is approximately the same
as for the KdV equation and corresponds to the exponential decay of the
sech2(Ex) in (5.2). The other root, however, implies that in the far field
u(x) - a+sin(kfx) + /?+ cos(kfx), 1x1 >> 1, (5.5)
where
kf = 1+ O(c) (“Far field wavenumber”) (5.6)
and where the subscripts on the constants ct., #? indicate that these may be
different for x + co and x + -a. Unless all the a’s and #?’sare zero, the
PRG wave flunks the localization test because it asymptotes to an
oscillation.
The PRG soliton is nonlocal because its positive phase speed matches
that of a free, traveling oscillation with wavenumber k = kdc). In
contrast, the Kortweg-deVries soliton is strictly localized because all the
54 John P. Boyd

linear sine waves of the KdV equation have negative phase speeds while
the soliton has a positive phase speed.

Strictly speaking, this proof of the Far Field Theorem is only half a
proof. (The other half is to prove that a* # 0 and/or /3* # 0.) Asymptotic
analysis as IxI--*w is nonetheless a very simple and powerful tool for
identifying candidates for nonlocal solitons.
The concept of the solitary wave as a robust, long-lived, and
concentrated disturbance may still apply even when the soliton is
nonlocal provided that the amplitude of the far field oscillations is
sufficiently small. For some micropterons, such as the long-lived but
radiating two-layer modons of Flier1 (1984), the smallness is directly
proportional to a small parameter with a small proportionality constant.
For other classes of nonlocal waves, a much stronger statement is
possible.

Theorem 5.2 (Exponential Smallness). (i) (General) If the solitary

wave is derived via the method of multiple scales with the assumption that
the pseudowavenumber E is small and if the zonal scale of the far field
oscillations is 0(1), then the amplitude of the far field radiation is
exponentially small as a function of 1 / ~ .
(ii) (Restricted) Zf the soliton is proportional to some positive integral
power of sech(Ex), then the amplitude (Y of far field radiation of zonal
wavenumber kfis

where Y is either a constant or an algebraic function of E.

Proof. See Boyd (1988b, 1989d). The key is the separation of length
scales. The core of the solitary wave generates the small amplitude waves
that propagate to infinity. This is clear for the radiatively decaying
solitary wave: if u(x, t = 0 ) decays exponentially with 1x1, then this initial
crest is the source of whatever waves appear at large 1x1 for large times.
However, the same is true for the nanopteron: insofar as the oscillation
for large 1x1 is concerned, the core is merely a stationary antenna.
In turn, the Fourier spectrum of the nanopteron core determines how
much energy is available in waves with k = kf to be radiated to infinity.
The Fourier transform of any smooth function, such as the core of a
 New Directions in Solitons and Nonlinear Periodic Waves 55

nanopteron, decays exponentially fast with wavenumber k. The broader

the soliton, the more rapidly its tranform falls off. The Fourier transform
of sech2(a), for example, is proportional to exp( - k n / [ 2 ~ ] )This
. implies
that the excitation of the waves with k = kfwill be exponentially small in
&-small proportional to exp( - k p / [ 2 ~ ] ) .
The separation-of-length scales between the nanopteron core and the
nanopteron far field is what makes the large 1x1 oscillation so small. The
slowly varying core and the rapidly varying far field wings are coupled
exponentially weakly because of their differing scales, 0 ( 1 / ~ versus)
O(1).

The arguments in Boyd (1988b, 1989c) are more systematic than this
brief verbal explanation but still short of a rigorous proof. Nevertheless,
for each of the physical applications in Subsection E including the PRG
soliton, (5.7) has been checked by explicit calculation.
The nanopteron is long-lived only when the radiation coefficient a << 1.
The theorem shows that this condition is met, perhaps to an almost
ridiculous degree, when E is only moderately small. For the PRG
nanopteron, for example, E = & implies a = 0(10-~)!

C. PERTURBATION
THEORY

It is easy to calculate a perturbation series in powers of the amplitude

and scale parameter E. For the PRG equation, for example, the
coefficients are polynomials in sech’x and the lowest order is the
Korteweg-deVries soliton (5.2). Unfortunately, the perturbation series
diverges.
One reason for the divergence is that a function like exp( - n k f / [ 2 & ] ) ,
which is smaller than any finite power of E as E - O , cannot possibly be
represented by a power series in E. A second reason is that a polynomial
in sech2(Ex) cannot approximate an oscillation for 1x1 >> 1 because each
term in the polynomial is exponentially small in the far field. Boyd
(1988a) shows, however, that even if the sign of the fourth derivative in
the PRG equation were changed so that its solutions are localized, the E
power series still diverges. The real villain is the assumption of multiple
length scales, which almost always yields divergent but asymptotic series.
Often, divergent power series can be summed via PadC approximants.
Boyd (1988a) shows that for nonlocal solitons, even PadC methods fail.
56 John P. Boyd

However, it is possible to calculate exponentially small quantities like the

radiation coefficients by applying the method of matched asymptotic
expansions in the complex plane. The pioneers were Pokrovskii and
Khalatnikov (1961) and Meyer (1980), who solved the linear problems of
WKI3 reflection without a turning point and quasi-trapped oscillations
around islands. Segur and Kruskal (1987) and Pomeau, Ramani, and
Grammaticos (1988) have extended this method to nonlocal solitons.
The key idea is to deform the contour of integration into the complex
x-plane. For the PRG, for example, the zeroth-order nanopteron has
simple poles at x = f i n / ( 2 ~ ) The
. neighborhood of these poles is the
“inner region”; in this disc-shaped boundary layer, the far field oscilla-
tion is not transcendentally small in comparison to the rest of u ( x ) .
Matching this inner solution to the outer solution, which vanishes as
Jx1+a along the real axis, gives the radiation coefficients (the a’s and
P ’ s in (5.5)) in the limit E- 0.
This improved perturbation theory is creative and deepens our
understanding of nonlocal solitons. However, it does have drawbacks.
First, as illustrated below, the method seems to generate rather inaccur-
ate approximations unless E is so small that the radiation coefficients are
below the limits of machine precision. In contrast, the numerical
strategies below can calculate the far field oscillation for all E , not just in
the limit E +0.
A more serious failing is that contour deformation implicitly assumes
that no singularity lies between the new and old integration paths. For
the linear, second order differential equation of Pokrovskii and Khalat-
nikov (1961), the singularities of u ( x ) are the known poles and branch
points of the coefficients of the differential equation. Unfortunately, no
such theory is available for nonlinear problems. Segur and Kruskal (1987)
and Pomeau, Ramani, and Grammaticos (1980) are forced to assume-
without proof-that the singularities of the perturbed nanopteron are
those of the unperturbed, zeroth order soliton. The numerical calcula-
tions support their conjectures, but it is not clear that future workers will
be so lucky!
Thus, their theories, which predict that the radiation coefficients are
nonzero, strengthen the argument that the G4 breather and PRG soliton
are weakly nonlocal. However, the matched asymptotics method is not
rigorous. A pure mathematician can still find some challenging problems
in nonlocal soliton theory!
 New Directions in Solitons and Nonlinear Periodic Waves 57

D. NUMERICAL
METHODS:
IGNORANCE,
THE RADIATION
BASIS,AND
CNOIDAL
MATCHING

Nanopterons can be calculated by solving a nonlinear boundary value

problem on x E [-a,031. The pseudospectral rational Chebyshev method
of Section IV is very effective when u(x) decays to 0 as IxI-cQ, but
nonlocal solitons do not. There are three strategies for coping with this
problem.
The first is to simply ignore it! Boyd (1988a) shows that the coefficients
of the rational Chebyshev functions, TB,(x), decrease exponentially until
they are somewhat smaller than ct-and then level off. Thus, this strategy
of “ignorance” is as effective as taking the multiple scales series to
optimum order: the error is O ( a ) but no better.
The second strategy is to augment the rational Chebyshev basis by
adding one or more special “radiation basis functions.” These are chosen
to mimic the desired far field behavior. For example, if
u(x) - a(&)sgn(x) sin(kx), (5.8)
which is appropriate for computing symmetric (with respect to x = 0)
nanopterons of all the examples discussed in the next subsection, then a
good radiation basis function is
@ r a d (E~) ;= tanh(tx) sin(kx). (5.9)
-
In the far field (1x1 >> l), tanh(tx) sgn(x), so this clearly reproduces
the asymptotic behavior of u ( x ) . However, this choice is not unique and
tanh(a) could be replaced by any smooth, antisymmetric function.
This special basis function replaces the N-th Chebyshev basis function
in the series for u ( x ) . Otherwise, the pseudospectral method is applied
exactly as in Section IV.
Boyd (1988a) shows that the mixed basis is a great improvement.
Because the Chebyshev functions are only required to do what they do
best-represent the exponentially decaying core of the nanopteron-the
coefficientsdo not level off at O( a).Figure 9 shows a PRG soliton and its
two constituent parts; one part is the sum of the rational Chebyshev
functions and the other is
For the linear problem of Boyd (1988a), a single radiation function is
sufficient. For nonlinear problems, however, the self-interaction of the
far field wave will give an O(a‘) correction to the far field u ( x ) . (This
58 John P. Boyd

0 5 10 15 20 25 30
X
FIG. 9. Solid: the PRG nanopteron for E = 0.16. The core (1x1 < 10) is approximately
~ ) . dashes: the sum of the rational Chebyshev functions in the
equal to 1 2 ~ * s e c h * ( ~Long
numerical solution. Short dashes: (Y +,&; E ) , where (Y is the “radiation coefficient” and
is the radiation basis function.

correction is faintly visible in Figure 9 as the oscillations in the sum of the

Chebyshev functions.)
The third strategy is to correct for this far field self-interaction by
matching the core to the appropriate cnoidal wave, that is, to a
nonlinear, spatially periodic oscillation. Because (Y << 1 if the nanopteron
is long-lived and only weakly non-local, Stokes’ perturbation method can
calculate the far field to as great an accuracy as one pleases. The twist on
the procedure described in Subsection 1I.E is that it is the far field
wavenumber k, rather than the phase speed c, which is expanded:
k - ko + &kz+ * * * . (5.10)
Turning the crank for the PRG equation gives
u,,(x) = a sin(kx) + az[ulo+ u12cos(2kx)]
- a3[ u 12
162k: - 18kg - 2~
]sin(~t), (5. l l a )

- u10
+
ko = { f [1+ (1 4 ~ ) ” In,
~]} k2 =
U1ZD
4k: - 2ko ’
(5. l l b )
 New Directions in Solitons and Nonlinear Periodic Waves 59

1 1
u10=-&, u12 = (5. l l c )
64k: - 16ki - 4~ ’
i =x + phase constant. (5.11d)
These Stokes expansions must be separately matched to the core as
x + --co and as x 03. However, if the nanopteron is symmetric about the

origin (see below), we can combine the cnoidal waves for large positive
and large negative x into a single basis function by multiplying sin(kx)
(and at higher order than (5.7), all antisymmetric terms in the cnoidal
wave) by tanh(Ex).
This gives the basis function
#cn(x; a)= d [ u l o + u ~ Z C O S ( ~ ~ ) I

Then, assume the expansion

a2u12sin(3kx)
I-
162k: - 18k; - 2~
(5.12)

(5.13)

Although #,.&; a) is a nonlinear function of its coefficient a, this does

not cause any real complication. The PRG equation is nonlinear, and we
must iterate no matter what the basis.
This third strategy is the most general because the Stokes series €or the
far field cnoidal wave may be extended to any order. This series
converges rapidly if a << 1 and gives an explicit analytic representation of
the basis function.
There is one subtlety: (5.8) assumes that u ( x ) is symmetric with respect
to x = 0. In reality, the symmetric, “minimal radiation” nanopteron is
only a special case of a much larger family. To the extent that linear
theory is accurate in the far field, i.e., to O(a),we have the freedom to
add sin(k,x) to the far field solution. Boyd (1988b, 1989c) shows that this
same freedom exists for the nonlinear solution, too. For example, it is
possible to compute a “one-sided” nanopteron that asymptotes to

(5.14)

In other words, we may force the nanopteron to vanish at one endpoint

at the price of twice as much oscillation as x + w . It is quite trivial,
however, to modify the radiation basis functions to compute whatever
member of the nanopteron family is desired.
60 John P. Boyd

Boyd (1988b, 1989c) gives a much fuller discussion of the far field.
Here, we choose for simplicity to compute only nanopterons that are
symmetric about the origin and have “minimal” far field radiation. Then
the /3’s in (5.5) are zero and a+ = fa,so there is only a single radiation
coefficient a.

E. PHYSICAL
ILLUSTRATIONS:ROSSBY
WAVES,HIGGSBOSONS,
AND THE SLOWMANIFOLD

1. Rossby Waves: Equatorial and Otherwise

Rossby waves are the planetary-scale free oscillations of the atmos-

phere and ocean. Pedlosky (1979) and Holton (1979) are good reviews of
the importance of these modes in geophysical fluid dynamics.
When the Rossby waves are confined to a channel, one may apply the
method of multiple scales to derive the Korteweg-deVries equation.
However, the KdV equation describes only the (lowest order) depend-
ence on x and t. The structure in the cross-channel coordinate y (latitude)
is the solution of a linear eigenvalue problem. For each latitudinal mode,
there is a distinct branch of solitons that have most of their energy in that
mode.
Several different mechanisms may confine Rossby waves to a latitudi-
nal channel including ocean topography, variations in the mean zonal
flow, and Coriolis refraction. The latter, which creates so-called “equa-
torial” waves that are confined to a narrow band of latitude centered on
the equator, gives the simplest class. However, the analysis of top-
ographically trapped and wind shear-trapped Rossby nanopterons is very
similar.
The soliton phase speed for the n-th equatorial Rossby is, letting E
denote the wave amplitude,
1
c = - -- 0(&2), (5.15)
2n +1
where n must be an odd integer (Boyd, 1980). The linear phase speed for
a Rossby wave of x-wavenumber k is

(5.16)
 New Directions in Solitons and Nonlinear Periodic Waves 61

For the gravest, n = 1 mode, the soliton phase speed lies outside the
range of all linear equatorial waves. Theorem 5.1 then implies that this
solitary wave should be a localized, long-lived soliton. This has been
confirmed by the initial value experiments of Kindle (1983), Greatbatch
(1985), and Williams and Wilson (1988).
In contrast, the higher mode Rossby solitons have nonlinear phase
speeds that match the speeds of linear waves of lower mode number. For
example, the n = 3 soliton has c = -+,which matches the phase speed of
n = 1 Rossby waves with k = 2. The consequence is that the n = 3 Rossby
soliton is weakly nonlocal, leaking energy along the equator in the form
of very small amplitude n = 1 Rossby waves with the far field wavenum-
ber kf= 2.
Although quantitative comparisons between nanopteron theory and
numerical experiments are not yet possible, the qualitative agreement is
excellent. Williams and Wilson (1988) find that the n = 1 mode is
long-lived and stable even for very large amplitude. However, the higher
modes are long-lived only for small amplitude (and then only because the
far field radiation is exponentially small in the amplitude). For moderate
amplitudes, the solitons radiatively decay through emission of Rossby
waves of lower latitudinal mode number, just as predicted. See Boyd
(1989a) for more details. Figure 10 shows the amplitudes of the n = 1 and
n = 3 Rossby modes for a simple two-mode model of equatorial waves.

2. The 9 Breather (Higgs Boson)

The so-called ‘‘94”model has been intensely studied by elementary

particle physicists (Boyd, 1988b). The partial differential equation is
u,, - u, - 2u - 3u2 - u3= 0. (5.17)
The method of multiple scales gives the perturbative solution (Dashen,
Hasslacher, and Neveu, 1974)
u(x, t ) - 0 . 8 1 6 ~sech(Ex) cos(wt), E << 1, (5.18)

(5.19)

Because this nonlinear solution is a standing oscillation that rises and falls
without propagation like the chest of a person, this species of soliton is
called a “breather”.
62 John P. Boyd

-0.150
-0.175
0.0 2.5 5.0 7.5 10.0
X
FIG. 10. The amplitudes of the n = 1 and n = 3 Rossby waves in a two-mode model of
the n = 3 equatorial nanopteron. The East-West dependence of the n = 3 mode, A3(x), is
approximately proportional to sech2(Ex), where ~ = f This . is the usual form for a
Korteweg-deVries soliton; the complete contribution of this term to u(x, y) is A3(x)q3(y),
where the latitudinal structure function is the sum of two Hermite functions.
The longitudinal structure function for the other latitudinal mode has a core proportional
to ~ ~ s e c h ~ (which , forced by the self-interaction of the n = 3 mode. The nanopteron is
~ x ) is
non-local because the n = 1 mode has oscillatory wings for large 1x1. For clarity, the
amplitude of the n = 1 mode has been exaggerated.

Segur and Kruskal (1987) showed that the E power series is only
asymptotic because the breather is weakly nonlocal. Their argument
employs the ansatz
m

U(X, t ) = C A J X ) cos(nwt), (5.20)

n=O

which is equivalent to the assumption-very likely, but not rigorously

proved-that the breather is periodic in time. Substituting (5.20) into
(5.17) gives a coupled set of nonlinear, ordinary differential equations in
x for the x-dependent coefficients of the Fourier series.
In the far field, 1x1 >> 1, the breather has decayed to such small
amplitude that the nonlinear terms may be neglected. The coupled,
nonlinear ODE'S reduce to the uncoupled, linear set
+
An,+ ( n ' d - 2)An = 0, n = 0, 1, . . . . (5.21)
 New Directions in Solitons and Nonlinear Periodic Waves 63

Because w’(E) 5 2 (Segur and Kruskal, 1987), one finds

(i) Ao(x) decays as
exp(-P2 Ixl), 1x1>> I; (5.22a)
(ii) A , @ ) decays as
exp(-E 1x0, 1x1 >> 1; (5.22b)
(iii) A2(x)oscillates as
a(&)sgn(x) sin(kx), 1x1 >> 1, (5.22~)
where the far field wavenumber is
(4w2- 2)ln 61n + O ( E ~ ) , 1x1 >> 1; (5.23)
(iv) All higher harmonics oscillate in the far field, too, but with larger
wavenumbers.
This situation is similar to that for Rossby waves. The dominant
component A,(x) is localized, but the wave is nonlocal because of higher
modes. These higher modes are forced by the self-interaction of the
dominant mode. Ao(x)and A&) enter at O ( E ~A3 ) , at O ( E ~ and
) , so on.
For the breather, “n-th mode” means “coefficient of cos(not)”; for
equatorial waves, “n-th mode” is the component whose latitudinal shape
is that of the n-th Rossby wave. Nevertheless, even though the nonlocal
modes are only weakly excited, the breather and the n = 3 Rossby wave
are both nanoptetons.
Segur and Kruskal show
-
a(&) Y exp(-[3/2]’”n/~), E << 1, (5.24)
exactly as predicted by Theorem 5.2(ii). The far field contributions of all
the higher harmonics, A3(x),A&), . . . , can be neglected because the
larger wavenumbers imply, according to the theorem, that their ampli-
tudes will be exponentially small in comparison to that of A2(x).
The constant Y, alas, can only be computed numerically. Segur and
Kruskal combined shooting and boundary value methods (in the complex
plane) to find (in our notation)
~ ( 0=) -0.018 f 0.004. (5.25)
Note the rather large uncertainty of 22%. They calculated only in the
limit E +0.
Boyd (1988b) computed Y( E ) for the symmetric, minimally nonlocal
nanopteron using the mixed Chebyshev/radiation basis algorithm de-
64 John P. Boyd

1.00
I
i.25
I
1.50
I
1.75
I
2.00
,
2.25
'.
2.50
1/€
FIG.11. Solid curve: numerically computed radiation coefficient a as a function of 1 / ~
for the r # ~breather
~ from Boyd (1988b). Theory asserts that for small E , a- V ( E ) exp(-
[3/2]"%/~), where v vanes algebraically (i.e., slowly) with E. Dashed curve: the radiation
coefficient as predicted by Segur and Kruskal (1987). The prediction is linear on this
log-reciprocal graph because these authors calculated only v(0).

scribed earlier. Figure 11 compares his a(&)with the extrapolation of

(5.24) and (5.25). Boyd's curve is clearly asymptoting to a line with the
same slope as that of Segur and Kruskal, corresponding to decay
proportional to exp( - 3 . 8 4 / ~ ) as predicted.
However, Boyd's Y(E) for E = 0.4 is roughly three times as large as
Segur and Kruskal's. Possibly the two curves are tangent in the limit
E+ 0; the graph ends at the relatively large value of E = 0.4. What is
certain is that the extrapolation of the Segur-Kruskal formula is accurate
only in order-of-magnitude for E L 0.4.
Figure 12 illustrates u(x, t) for E = 1. The maximum and minimum are
of equal amplitude in the perturbative approximation (5.18), but for
E = 1, the minimum is twice as deep as the maximum. Nonetheless, the
far field oscillation is still small (a= 0.0274 versus a maximum of
lu(x, t)l > 1) so that the concept of the breather is still meaningful. If we
chopped off the far field oscillations and used the core of the nanopteron
for E = 1 as the initial condition, the solution (after the radiation of initial
 New Directions in Solitons and Nonlinear Periodic Waves 65

0.5

0.3
0.1
-0.1
A
c
2 -0.3
Y

3 -0.5
-0.7
-0.9
-1.1

0- 7r
FIG. 12. A surface plot of the amplitude of the q54 breather as a function of x and t.
Only a quarter of the x-1 plane is shown because the breather is symmetric about the origin
in both space and time. Although the perturbation parameter E is not small ( E = l), the
amplitude of the wings is only 0.027, almost invisible on the scale of the graph.

transients) would lose only a small fraction of its energy during each cycle
of the breather.
Indeed, numerical experiments done in the 1970s (see the references in
Boyd, 1988b) show that for small to moderate amplitude, the @“ breather
is a stable and long-lived species. The mathematical proof that the soliton
does not exist is irrelevant to the longevity and coherence of 9“ solutions.
Physicists often criticize mathematical existence proofs as being un-
necessary demonstrations of what physicists already know. Here, how-
ever, the nonexistence proof is not merely irrelevant but extremely
misleading, almost the complete opposite of reality.

3. The “Slow Manifold” of Numerical Weather Prediction

The “slow manifold” is a hypothetical state of atmospheric or oceanic

motion that is completely devoid of high frequency gravity waves. If a
66 John P. Boyd

numerical weather prediction model contains 3N degrees of freedom,

then the “slow manifold” is the N-dimensional manifold within the
3N-dimensional phase space in which the motion consists entirely of
slow, low-frequency Rossby modes.
As a description of the observed atmosphere, the slow manifold is a
fiction: the amplitude of high frequency gravity waves is very small but
nonzero. More than forty years ago, however, it was recognized that the
slow manifold is an essential fiction for numerical weather prediction.
The reason is that the gravity waves oscillate too rapidly to be measured
by existing observational networks. If raw data is used to initialize a
prediction model, the amplitudes of the gravity waves will be mostly
noise. Unfortunately, the very high frequency of these waves means that
this noise can dominate predicted changes even though the amplitude of
the noise is very small.
The earliest remedy was to use a filtered set of equations, the so-called
quasi-geostrophic system, which eliminated the gravity waves completely.
This is too drastic. The nonlinear interaction of the Rossby waves will
drive low-frequency, forced oscillations in the gravitational modes, too.
A more realistic strategy is to modify the raw observations so that the
numerical forecast is initialized on the slow manifold and remains there
throughout the forecast.
The current state of the art is the so-called “nonlinear normal mode”
initialization (Baer and Tribbia, 1977). The method of multiple scales is
used to exploit the order-of-magnitude separation between gravity wave
and Rossby wave time scales. Nonlinear normal mode initialization has
evolved into a whole family of closely related algorithms; operational
forecast models now use the first or second order schemes with great
success.
The only difficulties are that (i) the slow manifold does not exist and
(ii) the perturbation series is divergent. (At least for some models, this is
true; the question of whether these problems are shared by all feasible
weather models is still unknown.) The mathematics of both the success
and failure of nonlinear normal mode initialization is very closely related
to that of our previous examples. The main difference is that the
localization requirement breaks down as t + m instead of as 1x1 +m.
To show this connection, we shall use the simplest possible model, the
set of five ordinary differential equations we shall dub the “L-K Quintet”
(Lorenz and Krishnamurthy, 1987):
 New Directions in Solitons and Nonlinear Periodic Waves 67

K=UW-bUz,
U, = -VW + bVz,
w,= -vu, I Rossby Triad
1
J
L-K Quintet
(5.26a)
(5.26b)
(5.26~)
x,=-z,
z, = x + bVU. I Gravity Dyad

The five unknowns are the time-dependent coefficients of five spatial

basis functions in this drastically truncated forecasting model. To simplify
(5.27a)
(5.27b)

normal mode initialization, the basis functions are not spherical har-
monics or Chebyshev functions but rather are the spatial structure factors
of the linear free oscillations of the model. The first three equations
describe the interaction of three Rossby waves. The second pair of
equations predicts the changes in the amplitude of the two gravity waves.
Our notation is that of Lorenz and Krishnamurthy (1987) except that we
capitalize the Rossby coefficients and write the gravity wave amplitudes
as lower case variables to emphasize that the Rossby modes are of much
larger amplitude.
Because the gravity wave dyad is linear in x and z, these two equations
may always be reduced to the single equation
z, + z = b(VU), . (5.28)
This is of the same form as that solved in Boyd (1988a) except that the
coordinate is time instead of space. The variables have been nondimen-
sionalized so that the frequency of the homogeneous solutions of (5.28),
i.e., that of the gravity wave free oscillations, is unity. Since Rossby
modes have frequencies small in comparison to those of the gravity
waves, it follows that V(f) and U ( t ) should vary with f only on a slow
time scale, equivalent to nondimensional frequencies much smaller than
one. Thus, (5.28) is a differential equation with rapidly varying homo-
geneous solutions and slowly varying forcing.
The resemblance becomes even closer if we exploit the fact that the
nondimensional parameter b << 1. Neglecting the b-dependent terms in
(5.26) gives a closed set of three equations in three unknowns. Lorenz
and Krishnamurthy chose to consider the special triad solution
U = E sech(&f), V = - E tanh(&f), W = --E sech(Ef). (5.29)
As noted in Section 111, the general triad solution is merely the
imbrication of this solution.
68 John P. Boyd

For this special case, one is once again in the realm of hyperbolic
functions, just as in our previous examples! Lorenz and Krishnamurthy
solve (5.28) exactly to show that while it is always possible to impose the
condition that z ( t ) and x(t)+ 0 as t + -CQ, the opposite limit gives
z(t) - -2b~re-~’(*~)
sin t as t+ 03 for E << I. (5.30)
The high frequency tail (5.30) implies that the slow manifold does not
exist. Even though the initial state (for large negative t) contains no
gravity waves and even though the gravity wave dyad is forced only by
the low-frequency interaction of the Rossby waves among themselves, a
high-frequency component persists even after the Rossby wave interac-
tion has faded into history.
T. Warn apparently was the first to suspect the nonexistence of the
slow manifold and the divergence of the nonlinear normal modes

nonexistence
is not true. Modifying (5.28) to
-
initialization. However, the equivalence relation
divergence (5.31)

2, - 2 = b(VU),, (5.32)
solely to make a mathematical point, gives a differential equation that
does have a slow manifold in the sense that its solution tends to 0 as
It[+ 03. Nevertheless, the method of multiple scales still diverges. The
reason is that the Fourier transform of the hyperbolic functions in (5.29)
and (5.32) does not have a sharp cutoff frequency w,. Instead, the
transform decays smoothly with IwI as I w I + m . This implies that the
“low-frequency” forcing terms in (5.29) and (5.32) have nonzero
high-frequency components, contrary to the separation of time scales,
which is the key assumption of multiple scales perturbation theory.
Fortunately, the exponential smallness of the Fourier transform for
(a( = 1 does permit the series to be asymptotic, and extremely successful
in operational forecasting.
Two questions remain. First, do these conclusions for the L-K Quintet
apply to more complicated models? J. Curry (private communication via
J. McWilliams) has shown that if one makes the L-K system more
consistent by including four additional gravity modes, the slow manifold
exists. This suggests that the multiple scales series may converge. What
happens for more complex models, with and without dissipation, is still a
mystery.
 New Directions in Solitons and Nonlinear Periodic Waves 69

Second, may one modify the L-K Quintet so that it does have a true
slow manifold? Boyd (1989~)shows that by applying the "band-limiting"
operator in frequency to the nonlinear terms in the gravity wave dyad,
one may suppress the exponentially small wing (5.30). It is feasible to
solve the band-limited L-K Quintet, but is is not yet clear how this
strategy could be applied to a more realistic forecasting model.
One thing is clear. Nonlocalization in time may have as much physical
significance as nonlocalization in space.

4. The Pomeau-Ramani-Grammaticos ( P R G ) Nanopteron

Since we have used the PRG equation (5.1) as a prototype throughout

most of this section, our discussion will be brief. Although this equation
is a generalization of the important Korteweg-deVries equation,
Pomeau, Ramani, and Grammaticos (1988) observe correctly that the
PRG model is more interesting as a mathematical example than as a
physical illustration. The reason is that extending the method of multiple
scales to higher order (beyond the Korteweg-deVries equation) usually
adds in additional cross-channel modes rather than merely tacking on an
extra term to the differential equation. However, at least in special
parameter ranges, the PRG equation does have applications to water
waves, electrical circuits, and other phenomena (Boyd, 1989). In
particular, Hunter and Scheurle (1988) have shown that (5.1) is a
consistent model for capillary-gravity waves for Bond number = 1/3.
Pomeau, Ramani, and Grammaticos (1988) solve their model using an
improved version of the matched-asymptotics-in-the-complex-plane
method of Segur and Krriskal (1987). One striking conclusion is that
when they compute a series that is accurate near the poles of the
nanopteron at x = f i n / ( 2 s ) , the nonlinearity drops out of the recursion
for large n. One finds that the high order coefficients are
an-K(-1)"(2n-l)! n>>l. (5.33)
The form of the coefficients is determined by the linear terms; only the
proportionality constant K is influenced by the nonlinear part of the PRG
equation.
If one applies their method to the linear, forced problem
uxxxx + u x x =f(=), (5.34)
where EX) has poles of the same order and location as the nonlinear
70 John P. Boyd

term in the PRG equation, one finds that the coefficients are given by
(5.33) exactly. It does not matter that (5.1) is nonlinear while (5.34) is
linear. For both the PRG equation and the linear model (5.34), it is these
coefficients a, that determine the radiation coefficient CY.
Thus, insofar as the far field is concerned, the PRG equation behaves
like (5.34). The nonlinear core of the nanopteron is merely an effective
forcing that excites the sinusoidal oscillations in the wings. The dynamics
of the far field is linear (at least to within 0(a2)); the nonlinear
interactions of the core merely determine the effective forcing function,
EX) in (5.34).
We have applied the mixed Chebyshev/radiation basis pseudospectral
algorithm to numerically compute the PRG nanopteron, and found an
error in Pomeau ef al. (1988). Their Equation (18) is missing a factor of 4.
The correct radiation coefficient is given by (in our notation and for the
form of Equation (5.1))
a(E ) - Y ( E)e-n’(2E), E << 1, (5.35)
where
~ ( 0=) 376.41. (5.36)
Fortunately, this typographical error has no effect on the main conclu-
sions of their article. Fitting a straight line though our computed
radiation coefficients for E = 0.05 and E = 0.075 gives
Y(0)
Y ( E ) =- << 1, (5.37)
1 +6 . 2 ~ ’ E

and also confirms Pomeau ef al.’s (corrected) value for ~ ( 0 to ) within

0.7%. However, it is not possible to check (5.36) to higher accuracy or to
compute the linear term in (5.37) to more than two decimal places
without heroic measures. The problem is that for E = &, CY = 6.5 x
This is sufficiently small so that roundoff error is significant even in
double precision. Yet even when the radiation coefficient is this tiny, Y ( E )
differs from ~ ( 0 by
) more than 30%.
As is also suggested by Figure 11, the matched asymptotic radiation
coefficients are quantitatively accurate only when so small that they
cannot be accurately confirmed even by double precision computations.
With the special pseudospectral method, however, one may compute the
wings of the nanopteron even when E is not small, as shown in Figures 9
and 12.
This algorithm could be extended to very, very small E by deforming
 New Directions in Solitons and Nonlinear Periodic Waves 71

the contour of integration into the complex plane as described by Boyd

(1985). Segur and Kruskal (1987) employed numerical integration in the
complex plane to complete their matched asymptotics calculation of ~(0)
for the @ 4 breather.

5 . Plasma Modons in Magnetic Shear

Modons are dipolar vortices that solve the quasi-geostrophic equation

of oceanography and the Hasegawa-Mima equation of plasma physics.
Meiss and Horton (1983) studied plasma modons in a magnetic shear. In
the absence of shear, modons are solitons in the classical sense, decaying
exponentially for large JxI.
Meiss and Horton show that in the spatially varying environment
created by magnetic shear, the index of refraction changes sign at large
distances from the modon center so that the flow is oscillatory in the far
field. Because the modon has exponentially decayed to very small
amplitude at the turning radius, however, the amplitude of the far field
radiation is exponentially small in the shear strength-a nanopteron. This
mechanism of nonlocalization via spatial inhomogeneity is probably
important in many other physical applications.
Because the decay-to-oscillation transition occurs in the far field, one
may compute the radiation by performing a standard WKB turning point
analysis. It is not necessary to deform the integration contour into the
complex plane in order to apply matched asymptotics at complex WKB
critical points, as was true for the PRG nanopteron and @ 4 breather.

6. Quasi-Geostrophic Monopoles

When /3, the y-derivative of the Coriolis parameter, is zero, any

radially symmetric vortex is an exact nonlinear solution of the quasi-
geostrophic equation. McWilliams and Flierl (1979) show that when the
(weak) @-effectis added, all one-signed (monopole) vortices slowly decay
with time by radiating Rossby waves. Flierl, Stern, and Whitehead (1983)
prove a theorem that shows that this fate is inevitable. Only when the
area integral of the vorticity is zero, as it can be for two-signed (dipole)
vortices, is a steadily translating soliton possible.
Nevertheless, the rate of dispersion of a strongly nonlinear monopole is
quite weak. At least in the initial value experiments, monopoles are
“algebraically decaying” because the rate of decay is directly propor-
tional to /3 instead of being exponentially small.
72 John P. Boyd

Gulf Stream rings are the observational motive for the study of
monopoles. These quasi-solitons are strong ocean eddies as much as
200km in diameter. When the observed shapes are used to initialize
quasi-geostrophic models that have artificially suppressed the nonlinear
terms, the vortex disperses in less than two months (Mied and Lin-
demann, 1982). When the nonlinear terms are retained, the same initial
condition gives a vortex that radiates slowly, but has a lifetime of several
years-just like observed Gulf Stream rings.
McWilliams (1989) reviews monopoles in more detail. Steadily trans-
lating but nonlocal monopoles, i.e., micropterons, may well exist.
However, there have been no explicit calculations of micropterons for
quasi-geostrophic monopoles or for either of the next two examples.

7 . The Flied-Petviashvili Monopole

Flierl (1979) and Petviashvili (1981) independently derived Equation

(4.1) as a model for radially symmetric Rossby solitary waves. Swenson
(1986) pointed out that although Flierl’s analysis is correct to lowest
order, the first order perturbation equation is insoluble. Boyd (1988d)
shows that the Flierl-Petviashvili equation (4.1) is also a consistent
lowest order model for the EMR equation, but again fails at first order.
In both cases, the mathematical reason for the perturbative breakdown
is that the lowest order model is translationally invariant. This implies
that the y-derivative of the monopole is an eigenfunction of the first (and
higher) order perturbation problems. Unless the perturbation is or-
thogonal to this eigenfunction, the perturbation theory fails.
The interesting difference is that Swenson’s perturbation is weak mean
shear whereas Boyd’s is the variation of the Coriolis parameter with
latitude. However, Petviashvili (1981) successfully solved his model to
first order including weak y-variations; his first order terms are different
from those of Swenson and Boyd. It is clear that higher order effects can
be very subtle, and that small inconsistencies in deriving models from the
hydrodynamic equations can have a profound impact on the existence of
true solitons.
Nevertheless, Swenson notes, “If no appropriate stationary solutions
exist, it is still possible that the Flierl solution is close to stationary. That
is, it might be more long-lived that a similarly strong vortex which does
not nearly meet the nonlinear dispersion relation .”
 New Directions in Solitons and Nonlinear Periodic Waves 73

8. Two-Layer Modons with a Weakly Radiating Lower Layer

Flierl (1984) studied modons in a two-layer model with a very deep

lower layer (as is true of the ocean). He showed that to lowest order in 6,
the relative thickness of the two layers, the solution is a modon in the
upper layer. However, the modon will force untrapped Rossby wave
radiation in the lower layer. Since there is no separation of scales
between the modon and the far field radiation, this nonlocal solution is
“algebraically decaying”: the amplitude of the far field oscillation is
proportional to 6, not to the exponential of the inverse of 6.
Nevertheless, Flierl shows that this quasi-soliton is long-lived: 6 << 1
over most of the ocean and the proportionality constant is also small. As
he writes in Flierl (1987): “. . . even nonisolated structures can be long
lived if the nonlinear advection is sufficiently strong and properly
organized. ”

F. SUMMARY:
THE GERERALIZATION
OF THE CONCEPT
OF A
“SOLITARY”
WAVE

One major theme is that for a soliton, existence is overrated. An

analogue is the property of convergence for an infinite series. Abel’s
remark that “divergent series are the invention of the devil” is now
regarded as nonsense; with a careful definition of asymptoticity, diver-
gent series may be very useful indeed. It was, however, necessary to
generalize “converges to” to “asymptotic to” in order to exploit
divergent series. In the same way, the concept of “soliton” is still useful
for nonlocal waves provided that we generalize “soliton” to allow
exponentially small far field oscillations.
The list of open problems is extremely long. One goal is to investigate
more systems with weakly nonlocal solitary waves; the probability is that
nanopterons are just as ubiquitous as true solitons. Another is to explore
the evolution of general initial conditions into radiatively decaying
solitons or nanopterons. A third is to find examples of micropterons; the
effects of perturbations on geophysical monopoles (Examples 6 and 7)
are poorly understood. A fourth goal is to replace the plausibility
arguments by rigorous proofs. Hunter and Scheurle (1988) prove the
existence of nonlocal solitons to the PRG equation, but even Segur and
Kruskal (1987) and Pomeau et al. (1988) have mathematical loose ends.
74 John P. Boyd

Weakly nonlocal solitons have now become respectable, but a vast

amount of polishing and refinement is still needed.

VI. Summary: Quo Vadis?

The history of solitons over the last two decades has been a tale of
three heroes: inverse scattering, the method of multiple scales, and
numerical initial value codes. This review has described four up-and-
coming newcomers.
The theory of polycnoidal waves is a mathematical poem with deep
roots in the past. The Nonlinear Fourier Transform algorithm of Osborne
and Bergamasco is the numerical realization and generalization of inverse
scattering. The Nakamura and Boyd theta function methods are the
periodic extension of Hirota’s bilinear transform for computing N -
solitons.
Landau once dreamed that turbulence was merely an N-polycnoidal
wave with N >> 1, but this now seems as illusory as the Seven Cities of
Gold. In the era of dynamical systems and chaotic attractors, polycnoidal
theory looks backward to nineteenth century studies of hyperelliptic
functions.
And yet, there is indisputable evidence that turbulence is not merely a
sea of randomness, but rather has strong, coherent, long-lived structures,
as stable and as conspicuous as icebergs. Understanding this interplay of
randomness and coherence requires not only a deep understanding of
solitons, but also of the multiply periodic coherent flows, the polycnoidal
waves. As Landau recognized, an N-polycnoidal wave of sufficiently large
N can mimic many features of true chaotic motion. Polycnoidal waves
and chaotically perturbed polycnoidal waves will still be useful models.
To reinforce this assertion, Hammack (1989) reviews laboratory
experiments that have observed two-dimensional Kadomtsev-Petviashvili
double cnoidal waves in surface water waves. The Torino group of
Osborne, Bergamasco and their collaborators are actively applying the
Nonlinear Fourier Transform to ocean time series.
Polycnoidal waves will probably never be as fashionable as dynamical
systems theory because their range of applicability-integrable
differential equation-is relatively narrow. Nevertheless, polycnoidal
wave theory will continue as a small but active area. More than 150 years
after John Scott Russell first observed solitary waves on a Scottish canal,
 New Directions in Solitons and Nonlinear Periodic Waves 75

we still have much to learn about water waves, and their counterparts in
optics, plasma physics, and many other fields.
Imbricate series theory is also a concerto rather than a symphony, but
it, too, will continue as a minor but steady line of research. It is quite
amazing that one may imbricate the soliton to generate the exact periodic
solution to many important wave equations. However, this nonlinear
superposition principle is more than a mathematical curiosity. The
concept of the periodic disturbance as a repetition of independent,
distinct pulses is often much closer to intuition-and sometimes
observation-than the model of the disturbance-as-sine-wave.
For example, it is very difficult to teach hydrodynamic stability theory
to meteorology undergraduates because the thought-model of a single,
growing Fourier component is irreconcilably different from what they
have already learned from weather maps. The naive stability theory
predicts a ring of six equally strong storms with equidistant spacing
around a circle of latitude. (Wavenumber six is the most rapidly growing
linear mode.) On the maps, however, each storm is as unique and
individual as a fingerprint, and the spacing between the storms is
irregular.
Unfortunately, this growing sine wave is the starting point for the most
sophisticated nonlinear theories now available. Is it possible to develop
an instability model that is pattern-function-based instead of Fourier-
based? The students see pattern functions on the maps, not sine waves. A
similar question may be asked in the time domain, too, as posed by the
imbricate series for the resonant triad, which is the simplest model for a
vacillating (time-periodic) instability.
On a long time scale, the dynamics of the atmosphere is chaotic and
unpredictable. On a time scale of several days, however, it is possible to
make a good forecast because each individual storm is predictable. It is
mostly the amplitude and phase of each storm which is irregular and
chaotic. Is it possible to generalize imbricate series to explicitly display
this pattern of order within disorder, and to handle phenomena that are
not simply periodic in time?
Direct numerical calculations of solitary waves, especially in two and
three dimensions, will be a major activity of the 1990s because we finally
have the algorithmic and electronic technology to solve multi-dimensional
nonlinear boundary value problems with comparative ease. Currently,
this capability is badly under-exploited in spite of the fifteen studies
reviewed in Section IV. The reason is that the community of nonlinear
76 John P. Boyd

wave theorists has almost no overlap with the tribe of supercomputers;

the few exceptions tend to be scorned by both groups. (I have the bruises
to prove it!) Nevertheless, as number-crunching has migrated from the
university computer center to the desktop, even theorists have learned a
little bit of programming. When the right algorithms are used, multi-
dimensional soliton calculations are simple and fast.
Lastly, as the study of solitons matures, there will be an increasing
need to generalize the concept. “Nonlocal” solitary waves will be only
one of these generalizations. Breathers, excited states, solitons-with-
dissipation, and other extensions will also be important. Some work has
already been done on most of these generalizations, but only to scratch
the surface.
At the moment, solitons are the “old” theory of nonlinearity, and
chaos/dynamical systems is the new and the trendy. Scientific fashions
aside, the reality is that we will not be able to truly understand either
until we know how to blend solitons and chaos, short-range structure and
long-range unpredictability. This is turn will require a much deeper
understanding of solitons and their generalizations. An intellectual alloy
is only as sound as the metals that make it.

Acknowledgments

This work was supported by grants OCE8509923, OCE8812300, and

DMS8716766 from the National Science Foundation. I thank C.-S. Yih
and the editor, T. Y. Wu, for suggesting this review. I thank Elsevier for
permission to reprint Figs. 8 and 10 from Boyd (1989a).

References

Ablowitz, M. J., and Segur, H. (1981). “Solitons and the Inverse Scattering Transform.”
SIAM, Philadelphia.
Andersen, C. M., and Geer, J. F. (1982). Power series expansions for the frequency and
period of the limit cycle of the van der Pol equation. Frequency and period of the limit
cycle of the van der Pol equation. SIAM J. A p p l . Math. 42, 678-693.
Baer, F., and Tribbia, J. J. (1977). On complete filtering of gravity waves through nonlinear
initialization. Mon. Wea. Rev. 105, 1536-1541.
Benjamin, T. B., Bona, J. L., and Mahony, J. J. (1972). Model equations for long waves in
nonlinear dispersive systems. Philos. Trans. Roy. SOC. London Ser. A 272, 47-73.
Boyd, J. P. (1980). Equatorial solitary waves, Part I: Rossby solitons. J . Phys. Oceangr. 10,
1699-1717.
 New Directions in Solitons and Nonlinear Periodic Waves 77

Boyd, J. P. (1982a). The optimization of convergence for Chebyshev polynomial methods

in an unbounded domain. J . Cornput. Phys. 45, 43-79.
Boyd, J. P. (1982b). Theta functions, Gaussian series, and the spatially periodic solutions of
the Korteweg-deVries equation. J . Math. Phys. 23, 375-387.
Boyd, J. P. (1984a). Cnoidal waves as exact sums of repeated solitary waves: New series for
elliptic functions. SIAM J . A p p l . Math. 44, 952-955.
Boyd, J. P. (1984b). The double cnoidal wave of the Korteweg-deVries equation: An
overview. 1. Math. Phys. 25, 3390-3401.
Boyd, J. P. (1984~).Perturbation theory for the double cnoidal wave of the Korteweg-
deVries equation. J. Math. Phys. 25, 3402-3414.
Boyd, J. P. (1984d). The special modular transformation for polycnoidal waves of the
Korteweg-deVries equation. J. Math. Phys. 25, 3415-3423.
Boyd, J. P. (1985). Complex coordinate methods for hydrodynamic instabilities and
Sturm-Liouville eigenproblems with an interior singularity. J. Cornput. Phys. 57,
454-471.
Boyd, J. P. (1986a). Solitons from sine waves: Analytical and numerical methods for
non-integrable solitary and cnoidal waves. Phys. D 21, 227-246.
Boyd, J. P. (1986b). An analytical and numerical study of the two-dimensional Bratu
equation. J. Sci. Cornput. 1, 183-206.
Boyd, J. P. (1987a). Spectral methods using rational basis functions on an infinite interval.
J. Cornput. Phys. 69, 112-142.
Boyd, J. P. (1987b). Orthogonal rational functions on a semi-infinite interval. J . Cornput.
Phys. 70, 63-88.
Boyd, J. P. (1988a). A comparison of numerical and analytical methods for the reduced
wave equation with multiple spatial scales. J. Cornput. Phys., submitted.
Boyd, J . P. (1988b). Numerical calculation of a weakly nonlocal soliton: The 44 breather.
Nonlinearity, submitted.
Boyd, J. P. (1988~).An analytical solution for a nonlinear differential equation with
logarithmic singularity. A d u . in Appl. Math. 9, 358-363.
Boyd, J. P. (1988d). Monopolar and dipolar vortex solitons in two dimensions. Wave
Motion, submitted.
Boyd, J. P. (1988e). Exact and approximate nonlinear waves generated by the periodic
superposition of solitons. Z. Angew. Math. Phys., in press.
Boyd, J. P. (19889. Chebyshev domain truncation is inferior to Fourier domain truncation
for solving problems on an infinite interval. J. Sci. Cornput. 3, 109-120.
Boyd, J. P. (1989a). Weakly non-local solitary waves. In “Mesoscale/Synoptic Coherent
Structures in Geophysical Turbulence” (J. C. J. Nihoul and B. M. Janart ed.) Elsevier,
Amsterdam, 103-112.
Boyd, J. P. (1989b). Nonlinear equatorial waves. In “Nonlinear Topics in Ocean Physics:
Proceedings of the Enrico Fermi School” (A. R. Osborne and L. Bergamasco eds.).
North-Holland, Amsterdam, in press.
Boyd, J. P. (1989~).Weakly non-local solitary waves. In “Nonlinear Topics in Ocean
Physics: Proceedings of the Enrico Fermi School” (A. R. Osborne and L. Bergamasco
eds.). North-Holland, Amsterdam, in press.
Boyd, J. P. (1989d). “Chebyshev and Fourier Spectral Methods.” Springer-Verlag, New
York, in press.
Boyd, J. P. (1989e). The arctan/tan mapping for periodic solutions with a region of rapid
variation. J . Cornput. Phys., submitted.
Boyd, J. P. and Haupt, S. E. (1989). Polycnoidal waves: spatially periodic generalizations
of multiple solitons. In “Nonlinear Topics in Ocean Physics: Proceedings of the Enrico
Fermi School” (A. R. Osborne and L. Bergamasco eds.). North-Holland, Amsterdam,
in press.
78 John P. Boyd
Boyd, J. P., and Ma, H. (1989). Lamb-Batchelor vortex pairs (modons) bounded by an
ellipse. J. Fluid Mech., submitted.
Briggs, W. L. (1987). “A Multigrid Tutorial.” SIAM, Philadelphia.
Canuto, C., Hussaini, M. Y., Quarteroni, A., and Zang, T. A. (1987). “Spectral Methods
in Fluid Dynamics.” Springer-Verlag, New York.
Chang, H.-C. (1986). Traveling waves on fluid interfaces: Normal form analysis of the
Kuramoto-Sivashinsky equation. Phys. Fluids 29, 3142-3147.
Chen, B., and Saffman, P. G. (1980). Numerical evidence for the existence of new types of
gravity waves of permanent form on deep water. Stud. Appl. Math. 62, 1-21.
Chu, M. T. (1988). On the continuous realization of iterative processes. SIAM Rev. 30,
375-387.
Dashen, R. F., Hasslacher, B., and Neveu, A. (1974). Particle spectrum in model field
theories from semiclassicalfunctional integral techniques. Phys. Rev. D 10, 4130-4150.
Date, E., and Tanaka, S. (1976). Periodic multi-soliton solutions of Korteweg-deVries
equation and Toda lattice. Progr. Theoret. Phys. 59, 107-125.
Deem, G. S., and Zabusky, N.J. (1978). Vortex waves: Stationary “V states,” interactions,
recurrence, and breaking. Phys. Rev. Lett. 40,859-862.
Dritschel, D. C. (1988). Contour dynamics. J. Comput. Phys. 77,240-266.
Dodd, R. K., Eilbeck, J. C., Gibbon, J . D., and Morris, H. C. (1982). “Nonlinear Wave
Equations.” Academic Press, New York.
Dubrovin, B. A. (1981). Theta functions and non-linear equations. Uspekhi Mat. Nauk 36,
11-80.
Dubrovin, B. A., Matveev, V. B., and Novikov, S. P. (1976). Nonlinear equations of
Korteweg-deVries type, finite zoned linear operators, and Abelian varieties. Russian
Math. Surveys 31, 59-146.
Eydeland, A., and Turkington, B. (1987). A computational study of Rossby solitary waves.
Stud. Appl. Math. 76, 37-67.
Eydeland, A., and Turkington, B. (1988). A computational method of solving free-
boundary problems in vortex dynamics. J . Comput. Phys. 78, 194-214.
Ferguson, W.E., Flaschka, H., and McLaughlin, D. W. (1982). Nonlinear normal modes
for the Toda chain. J. Comput. Phys. 45, 157-209.
Flierl, G. R. (1979). A simple model for the structure of warm and cold core rings. J.
Geophys. Res. 84 (C2),781-785.
Flierl, G . R. (1984). Rossby wave radiation from a strongly nonlinear warm eddy. J. Phys.
Oceangr. 14,47-58.
Flierl, G . R. (1987). Isolated eddy models in geophysics. Ann. Rev. Fluid Mech. 19,
493-530.
Flierl, G. R., Stern, M. E., and Whitehead, J. A., Jr. (1983). The physical significance of
modons: Laboratory experiments and general physical constraints. Dyn. Atmos.
Oceans 7, 233-263.
Forest, M. G., and McLaughlin, D. W. (1982). Spectral theory for the periodic
sine-Gordon equation: A concrete viewpoint. J . Math. Phys. 23, 1248-1277.
Forest, M. G., and McLaughlin, D. W. (1983). Modulations of sinh-Gordon and
sine-Gordon wavetrains. Stud. Appl. Math. 68, 11-59.
Greatbatch, R. J. (1985). The response of the equatorial ocean to impulsive wind stress. J .
Geophys. Res. 90,9097-9109.
Hammack, J. L. (1989). Laboratory experiments with nonlinear water waves. I n
“Nonlinear Topics in Ocean Physics: Proceedings of the Enrico Fermi School” (A. R.
Osborne and L. Bergamasco eds.). North-Holland, Amsterdam, in press.
Haupt, S . E., and Boyd, J. P. (1987). Modeling nonlinear resonance: A modification to the
Stokes’ perturbation expansion. Wave Motion 10, 83-98.
 New Directions in Solitons and Nonlinear Periodic Waves 79

Haupt, S. E., and Boyd, J. P. (1989). Double cnoidal waves of the Korteweg-deVries
equation: A boundary value approach. Physica D , submitted.
Hirota, R. (1974). A new form of Backlund transformations and its relation to the inverse
scattering transform. Progr. Theoret. Phys. 52, 1498-1520.
Hirota, R., and Ito, M. (1981). A direct approach to multi-periodic wave solutions to
nonlinear evolution equations. 1. Phys. SOC.Japan 50, 338-342.
Holton, J. R. (1979). “An Introduction to Dynamic Meteorology.” Academic Press, New
York.
Hunter, J. K., and Scheurle, J. (1988). Existence of perturbed solitary wave solutions to a
model equation for water waves. Phys. D 32, 253-268.
Hyman, J. M. (1978). Time evolution of almost periodic solutions of the KdV equation.
Rocky Mountain J. Math. 8, 95-104.
Ingersoll, A. P., and Cuong, P. G. (1981). Numerical model of long-lived Jovian vortices.
J. Atmospheric Sci. 38, 2067-2076.
Its, A. R., and Matveev, V. B. (1975). Hill’s operator with finitely many gaps. Functional
Anal. Appl. 9, 65-66.
Jaworski, M., and Zagrodzinski, J. (1982). Quasi-periodic solutions of the sine-Gordon
equation. Phys. Lett. A 92, 427-430.
Johnson, R. S. (1976). On the modulation of water waves on shear flows. Proc. Roy. SOC.
London Ser. A 338, 101-114.
Kadomtsev, B. B., and Petviashvili, V. I. (1970). On the stability of solitary waves in
weakly dispersing media. Soviet Phys. Dokl. 15,539-543.
Keller, H. B. (1977). Numerical solution of bifurcation and nonlinear eigenvalue problems.
In “Applications of Bifurcation Theory” (P. Rabinowitz ed.), pp. 359-384. Academic
Press, New York.
Kindle, J. C. (1983). On the generation of Rossby solitons during El Nino. In
“Hydrodynamics of the Equatorial Ocean” (J. C. J. Nihoul ed.), pp. 353-371.
Elsevier, Amsterdam.
Korpel, A., and Banerjee, P. P. (1981). Exact decomposition of cnoidal waves into
associated solitons. Phys. Lett. A 82, 113-115.
Lax, P. (1976). Almost periodic solutions of the KdV equation. SIAM Rev. 18, 351-365.
Lorenz, E. N. (1960). Maximum simplification of the dynamic equations. Tellus U,
243-254.
Lorenz, E. N., and Krishnamurthy, V. (1987). On the nonexistence of a slow manifold. J .
A m o s . Sci. 44, 2940-2950.
Ma, Y.-C., and Ablowitz, M. J. (1981). The periodic cubic Schroedinger equation. Stud.
Appl. Math. 65, 113-158.
Matsuno, Y. (1984). “The Bilinear Transformation Method.” Academic Press, New York.
McWilliams, J. C. (1984). The emergence of isolated, coherent vortices in turbulent flow. 1.
Fluid Mech. 146, 21-43.
McWilliams, J. C. (1989). Geostrophic vortices. In “Nonlinear Topics in Ocean Physics:
Proceedings of the Enrico Fermi Summer School” (A. R. Osborne and L. Bergamasco
eds.). North-Holland, Amsterdam, in press.
McWilliams, J. C., and Flierl, G. R. (1979). On the evolution of isolated, nonlinear
vortices. J . Phys. Oceangr. 9, 1155-1182.
Meiron, D. I., Saffman, P. G., and Yuen, H. C. (1982). Calculation of steady
three-dimensional deep-water waves. J . Fluid Mech. W, 109-121.
Meiss, J. D., and Horton, W. (1983). Solitary drift waves in the presence of magnetic shear.
Phys. Fluids 26, 990-997.
Meyer, R. E. (1980). Exponential asymptotics. SIAM Rev. 22, 213-224.
Mied, R. P., and Lindemann, G. J. (1982). The birth and evolution of eastward
propagating modons. J. Phys. Oceangr. 12, 213-230.
80 John P. Boyd
Miloh, T., and Tulin, M. P. (1989). Periodic solutions of the DAB0 equation as sum of
repeated solitons. J. Phys. A 22, 921-923.
Morgan, A. P. (1987). “Solving Polynomial Systems Using Continuation for Engineering
and Scientific Problems.” Prentice-Hall, Englewood Cliffs, New Jersey.
Morse, P. M., and Feshbach, H. (1953). “Methods of Theoretical Physics,” Vol. 1, p. 395.
McGraw-Hill, New York.
Munk, W. H. (1949). The solitary wave and its application to surf problems. Ann. New
York Acad. Sci. 51, 376-424.
Nakamura, A. (1979). A direct method of calculating periodic wave solutions to nonlinear
evolution equations. I. Exact two-periodic wave solution. J . Phys. SOC. Japan 47,
1701-1710.
Nakamura, A. (1980). A direct method of calculating periodic wave solutions to nonlinear
evolution equations. 11. Exact one- and two-periodic wave solution of the coupled
bilinear equations. J . Phys. SOC.Japan. 48, 1365-1371.
Nakamura, A., and Matsuno, Y. (1980) Exact one- and two-periodic wave solutions of
fluids of finite depth. J. Phys. SOC.Japan 48,653-657.
Osborne, A. R., and Bergmasco, L. (1985). The small amplitude limit of the spectral
transform for the periodic Korteweg-deVries equation. Nuovo Cimento B 85,
229-242.
Osborne, A. R., and Bergamasco, L. (1986). The solitons of Zabusky and Kruskal revisted:
Perspective in terms of the periodic spectral transform. Phys. D 21, 26-46.
Osborne, A. R., and Bergamasco, L. (1987). Numerical implementation of the Nonlinear
Fourier Transform for the Korteweg-deVries equation. J . Comput. Phys., submitted.
Orszag, S. A. (1980). Spectral methods for problems in complex geometries. J . Comput.
Phys. 37, 70-92.
Pedlosky, J. (1979). “Geophysical Fluid Dynamics.” Springer-Verlag, New York.
Peitgen, H . - 0 . and Richter, P. H. (1986). “The Beauty of Fractals.” Springer-Verlag, New
York.
Peregrine, D. H. (1966). Calculations of the development of an undular bore. J . Fluid
Mech. 25, 281-300.
Peregrine, D. H. (1983). Breaking waves on beaches. Ann. Rev. Fluid Mech. 15, 149-178.
Petviashvili, V. I., and Tsvelodub, 0. Yu. (1978). Horseshoe-shaped solitons on a flowing
viscous film of fluid. Soviet Phys. Dokl. 23, 117-118.
Petviashvili, V. I. (1976). Equation of an extraordinary soliton. Soviet J . Plasma Phys. 2,
257-260.
Petriashvili, V. I. (1981). Red spot of Jupiter and the drift soliton in a plasma. JETP Lett.
32,619-622.
Pierini, S . (1986). Solitons in a channel emerging from a three-dimensional initial wave.
Nuovo Cimento C 9, 1045-1061.
Pierrehumbert, R. T. (1980). A family of steady, translating vortex pairs with distributed
vorticity. 1. Fluid Mech. 99, 129-144.
Pokrovskii, V. L., and Khalatnikov, J. M. (1961). On the problem of above-barrier
reflection of high-energy particles. Soviet Phys. JETP W, 1207-1210.
Pomeau, Y., Ramani, A., and Grammaticos, B. (1988). Structural stability of the
Korteweg-deVries solitons under a singular perturbation. Phys. D 31, 127-134.
Schwartz, L. W. (1974). Computer extension and analytic continuation of Stokes’ expansion
for gravity waves. J. Fluid Mech. 62, 553-578.
Segur, H., and Finkel, A. (1985). An analytical model of periodic waves in shallow water.
Stud. Appl. Math. 73, 183-200.
Segur, H., and Kruskal, M. D. (1987). On the nonexistence of small-amplitude breather
solutions in +4 theory. Phys. Rev. Lett. 58, 747-750.
 New Directions in Solitons and Nonlinear Periodic Waves 81
Seydel, R. (1988). “From Equilibrium to Chaos: Practical Bifurcation and Stability
Analysis.” Elsevier, Amsterdam.
Stenger, F. (1981). Numerical methods based on Whittaker cardinal, or sinc functions.
SlAM Rev. 23, 165-224.
Stiassnie, M., and D. H. Peregrine (1980). Shoaling of finite amplitude surface waves on
water of slowly varying depth. 1. Fluid Mech. 97, 783-805.
Streett, C. L., and Zang, T. A. (1984). Spectral methods for solution of the boundary-layer
equations. AIAA Paper 84-0170, 9 pp.
Swenson, M. (1986). A note on baroclinic solitary waves with radial symmetry. Dyn.
Armos. Oceans 10, 243-252.
Tanveer, S . (1986). A steadily translating pair of equal and opposite vortices with vortex
sheets on their boundaries. Stud. Appl. Math. 74, 139-154.
Toda, M. (1975). Studies of a nonlinear lattice. Phys. Rep. 18, 1-124.
Tracy, E. R., Larson, J., Osborne, A. R., and Bergamasco, L. (1988). The relationship
between the spectral theories for the periodic Korteweg-deVries and nonlinear
Schroedinger equations. Phys. D 21, 83-106.
Tracy, E. R., Larson, J., Osborne, A. R., and Bergamasco, L. (1989). The relationship
between the spectral theories for the periodic Korteweg-deVries and nonlinear
Schroedinger equations. In “Nonlinear Topics in Ocean Physics, Proceedings of the
Enrico Fermi Summer School” (A. R. Osborne and L. Bergamasco eds.). North-
Holland, Amsterdam, in press.
Tung, K.-K., Chan, T. F., and Kubota, T. (1982). Large amplitude internal waves of
“permanent form”. Stud. Appl. Math. 66, 1-44.
Vanden-Broeck, J.-M. (1986). Steep gravity waves: Havelock’s method revisted. Phys.
FIuidr 29, 3084-3085.
Van Dyke, M. (1975). Computer extension of perturbation series in fluid mechanics. SlAM
J. Appl. Math. 28, 720-734.
Van Dyke, M. (1984). Computer extension of perturbation series in fluid mechanics. Ann.
Rev. Fluid Mech. 16,287-309.
Verkley, W. T. M. (1989). A variational principle for modons. J. Atmospheric Sci.,
submitted.
Watson, L. T., and Scott, L. R. (1987). Solving Galerkin approximations to nonlinear
two-point boundary value problems by a globally convergent homotopy method. SIA M
J . Sci. Statist. Comput. 8, 768-789.
Whitham, G. B. (1974). “Linear and Nonlinear Waves,” pp. 109-110. John Wiley, New
York.
Whitham, G. B. (1984). Comments on periodic waves and solitons. IMA J. Appl. Math. 32,
353-366.
Williams, G. P., and Wilson, R. J. (1988). The stability and genesis of Rossby vortices. J .
Atmospheric Sci. 49, 207-240.
Wimp, J. (1981). “Sequence Transformations and Their Applications.” Academic Press,
New York.
Wu, H. M., Overman, E. A., and Zabusky, N. J. (1984). Steady-state solutions
of the Euler equations in two dimensions: Rotating and translating V-states with
limiting cases. I. Numerical algorithms and results. J. Comput. Phys. 53, 42-71.
Yasuda, T., and Tsuchiya, Y. (1982). Theory of soliton spectra and its application to waves
in shallow water. Trans. JSCE 14, 428-431.
Zabusky, N. J., and Kruskal, M. D. (1965). Interaction of solitons in collision-less plasma
and the recurrence of initial states. Phys. Rev. Lett. 15,240-243.
Zagrodzinski, J. (1982). Dispersion relations and a comparison of different quasi-periodic
solutions of the sine-Gordon equation. J . Phys. A 15, 3109-3118.
82 John P. Boyd
Zagrodzinski, J. (1983). Direct approach to the periodic solutions of the multidimensional
sine-Gordon equation. J. Math. Phys. 24, 46-52.
Zagrodzinski, J. (1984). Solitons and wavetrains: Unified approach. J . Phys. A 17,
3315-3320.
Zagrodzinski, J., and Jaworski, M. (1982). Mixed solutions of the sine-Gordon equation.
Z . Phys. B 49, 15-77.
Zaitsev, A. A. (1983). Formation of stationary nonlinear waves by superposition of
solitons. Soviet Phys. Dokl. 28, 120-722.
 ADVANCES IN APPLIED MECHANICS, VOLUME 27

Material Failure by Void Growth to Coalescence

VIGGO TVERGAARD

Department of Solid Mechanics

The Technical University of Denmark
Lyngby, Denmark

I. Introduction .......................................................... 83
11. Basic Equations.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 86
111. Continuum Models of Porous Ductile Solids . . . . , . . . . . . . . . . . . . . . . . . . . . . 89
A. Approximate Yield Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
B. Nucleation and Growth of Voids . . . . . . . . . . . . . . . . . . . . . . . , . . . . .
C. Stress-Strain Relations . . . . . . . . . . . . . . .. . .. . . ... .. . ,. ... .
IV. Localization of Plastic Flow . . . . . . . . . . . . . ... ... . . . ... . .. ... .... .... . 96
V. Cell Model Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
A. Nucleation of Voids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
B. Growth and Coalescence of Voids . . . . . . ... . . . . .. . .. ... ... . ... . . 107
C. Two Size-Scales of Voids . . . . . . . . . . ... ... . ..... . ............... 109
VI. Formation and Growth of Cracks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
A. CrackFormation ................................................... 113
B. Crack Tip Fields . . . . . . ............ ......... . ... ... ......... 116
VII. Effect of Yield Surface Curvature . . . . . . . . . . . , . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
A. Kinematic Hardening Por .. . .. . . . . ... ... . .. , ... . . 120
B. Localization Predictions .. . .. . .. . ... .. . .. . . ... . . 123
VIII. Strain Rate Sensitive Material . . , . . . . . . . .. . . . . . . . . . . . , 127
A. Viscoplastic Material Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
B. The Charpy V-Notch Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
IX. Creep Failure by Grain Boundary Cavitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
A. Creep with Grain Boundary Cavity Growth . . . . . . . ...... ... ... 135
B. Creep Rupture in Structural Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
X. Discussion............................................................ 144
References ........................................................... 147

I. Introduction
Failure by coalescence of microscopic voids is an important fracture
mechanism in ductile metals. The voids nucleate mainly at second phase
83
Copyright 0 1990 Academic Press, Inc.
All rights of reproduction in any form reserved.
ISBN 0-12-002027-0
84 Viggo Tvergaard

particles, by decohesion of the particle-matrix interface or by particle

fracture, and subsequently the voids grow due to plastic straining of the
surrounding material. The first micromechanical studies of these pheno-
mena focussed on the growth of a single void in an infinite elastic-plastic
solid (McClintock, 1968; Rice and Tracey, 1969), and the results were
used to estimate critical strains for coalescence. Most of the subsequent
work has focussed on incorporating the effect of the interaction between
neighbouring voids, and there has also been some interest in micromech-
anical modelling of nucleation and of void deformations in the final stages
prior to failure by coalescence.
The most widely known porous ductile material model is that de-
veloped by Gurson (1977a) based on averaging techniques similar to
those used by Bishop and Hill (1951). The characteristic volume element
considered by Gurson (1977a) is an aggregate of voids and rigid plastic
matrix material rather than a polycrystalline aggregate, and approximate
upper-bound solutions on the micro-level have been used to derive a
macroscopic yield condition for the material. Alternative approximate
yield conditions have been proposed by a number of other authors, e.g.,
Shima and Oyane (1976), who considered powder metallurgical materials
on a purely empirical basis.
A number of micromechanical models, including that employed by
Gurson (1977a), make the simplifying assumption that voids are spherical
(or cylindrical). This gives a reasonable approximation in some cases,
dependent on the shapes of particles from which voids nucleate; but near
final failure this is usually a very poor approximation, as is illustrated by
the micrograph in Figure 1. This micrograph shows a local failure region
in the center of a notched tensile test specimen machined from a sintered
iron powder compact, which was initially fully dense (Becker ef al.,
1987a). It is seen that some voids have already coalesced to form larger
voids and that voids can link up by rather localized bands of smaller voids
as well as by necking of the ligament.
The porous ductile material models accounting for the nucleation
and growth of voids are discussed in some detail in Section 111, while cell
model studies that are a major tool in understanding material behaviour
are considered in Section V. Such model studies have been used to
investigate the nucleation of voids and void coalescence, as well as the
interaction between different size-scales of voids.
Applications of the material model to predict fracture mechanisms in
structural components or test specimens is one of the main purposes of
 Material Failure by Void Growth 85

FIG. 1. Local failure region in sintered iron powder compact, with no initial porosity
(Becker er al., 1987a).

developing these constitutive relations. Localization of plastic flow in

shear bands leading to final shear fracture by a void-sheet mechanism is
one such application discussed in Section IV, and ductile crack growth by
void coalescence is another discussed in Section VI. Extensions of porous
ductile material models to account for the effect of a rounded vertex on
the yield surface or material strain rate sensitivity are presented in
Sections VII and VIII, respectively. The viscoplastic material model used
to represent rate sensitivity is illustrated by analyses of the fracture mode
transition in the Charpy V-notch test.
Void growth and coalescence is also the dominant cause of failure in
polycrystalline metals subject to creep at elevated temperatures. How-
ever, at high temperatures the voids appear mainly in the grain
boundaries, and diffusion gives a significant contribution to the growth
rate. Due to the interaction of diffusion and dislocation creep, the rate of
void growth can here be constrained by creep. This rather different mode
of material failure by void coalescence is described in Section IX in terms
of a material model proposed by Tvergaard (1984b). Applications of the
model to analyses of creep rupture in structural components are included
in the discussion.
86 Viggo Tvergaard

11. Basic Equations

The growth of voids to coalescence by plastic yielding of the surround-

ing material involves so large geometry changes that finite strain
formulations of the field equations are a necessary tool. This applies to
micromechanical studies that focus on the local stress and strain fields
around each single void as well as to studies based on continuum models
that describe the average macroscopic effect of porosities in ductile
solids.
The finite strain formulations in this paper are based on a Lagrangian
framework, in which the reference configuration is usually identified with
the initial undeformed configuration. Relative to a fixed Cartesian frame,
the position of a material point in the reference configuration is denoted
by the vector r, and the position of the same point in the current
configuration is F. The displacement vector u and the deformation
gradient F are given by
u=?-r, F=-.
ar
ar
A convected coordinate formulation of the governing equations is used
(see, for example, Green and Zerna (1968) and Budiansky (1969)).
e
Convected coordinates are introduced, which serve as particle labels.
The convected coordinate net can be visualized as being inscribed on the
body in the reference state and deforming with the material. The
displacement vector u is considered as a function of the coordinates e
and a monotonically increasing time-like parameter t. Latin indices range
from 1 to 3, and the summation convention is adopted for repeated
indices.
Covariant base vectors ei and bi of the material net in the reference
configuration and the current configuration, respectively, are given by

The metric tensors in the reference and current configurations, gij and
G,, are given by the dot products of the base vectors
.
gij = ei ej, G..
fJ
= 5.
* .c.
I' (2.3)
The determinants of gij and Gij are denoted g and GI respectively, and
the inverse of the two metric tensors are denoted by gij and G'J.
 Material Failure by Void Growth 87
Components of vectors and tensors on the embedded coordinates are
obtained by dot products with the appropriate base vectors. Thus, the
displacement components on the reference base vectors satisfy

ui= ei * u, -
u' = e' u, u = uiei. (2-4)
Substituting (2.4~)into (2.1) and using this in (2.2) gives
gi = ei + u:ek, (2.5)
where ( ) , i denotes the covariant derivative in the reference frame. The
Lagrangian strain tensor qii = a(Cii- gii), expressed in terms of dis-
placement components, is then found using (2.3) and (2.5)
7..
11 = a(ui,j + +
uj,; U5uk.j). (2.6)
The true stress tensor u in the current configuration (the Cauchy stress
tensor) has the contravariant components di on the current base vectors,
where
= dJs.a.
1 I'
a'j = gi .u. g i . (2.7)
The contravariant components ti' of the Kirchhoff stress tensor on the
current base vectors are defined by

where a = dV/dV = p / p , expressed in terms of the volume element

dV and the density p.
The requirement of equilibrium can be specified in terms of the
principle of virtual work

where V and S are the volume and surface, respectively, of the body in
the reference configuration, and the specified surface tractions T = Tiei
per unit area in the reference configuration have the components on
reference base vectors
T' = (ti'+ tkJ,fk)n,. (2.10)
Here, n = niei is the surface normal in the reference state.
For the solution of boundary value problems, incremental equilibrium
equations are needed due to the material path dependence. When the
88 Viggo Tvergaard

current values of all field quantities, e.g., stresses zii, displacements ui ,

and tractions Ti, are assumed known, an expansion of (2.9) about the
known state gives to lowest order

= At I
S
pi6ui dS - [ ti’6vii dV - I,
T i 6ui d S ] . (2.11)

Here, (*)= a( ) / a t at fixed r,

and At is the prescribed increment of the
“time” t, so that ATi =Ati.’ are the components of the prescribed
traction increments. The terms bracketed in (2.11) vanish according to
(2.9) if the current state satisfies equilibrium. However, in linear
incremental analyses the solution tends to drift away from the true
equilibrium path, due to incrementation errors, and here the bracketed
terms in (2.11) can be included to avoid such drifting.
In the constitutive relations to be discussed here, the total strain rate is
taken to be the sum of the elastic and plastic parts, fiij = fit + fi;. Thus,
with an elastic relationship of the form ziJ
= %ijk/rjzl, the constitutive
relations can be written as
zij = gp/(
fikl - $/), (2.12)
where the Jaumann (co-rotational) rate of the Cauchy stress tensor zij is
related to the convected rate by
zij = #, + ( G i k Uj / + Gyd‘)fikl. (2.13)
For the elastic moduli, Budiansky (unpublished work, see also Hutch-
inson (1973)) has suggested using the finite strain generalization

where E is Young’s modulus and Y is Poisson’s ratio. The elastic

incremental stress-strain relationship with moduli (2.14) is actually
hypo-elastic, since it cannot be derived from a work potential. However,
it is noted that in the limit of small stresses relative to Young’s modulus
the relationship reduces to Hooke’s law on the current base vectors. In
elastic-plastic analyses the elastic contribution to the total straining is
usually very small so that the use of this hypo-elastic relationship rather
than a truly elastic one is a reasonable approximation.
An incremental stress-strain relationship in terms of i?“ is needed in
 Material Failure by Void Growth 89

(2.11). This expression is obtained from (2.12) by substituting (2.13) and

the incremental form of (2.8)

(2.15)

For time-independent plasticity, where t is a loading parameter since

time does not enter the formulation, the resulting constitutive relations
are of the form
+ii = L i i k l l j k , . (2.16)
On the other hand, viscous material models are time-dependent, so that
here t denotes the time. In visco-plasticity, where fiz represents non-
linearly viscous behaviour, the resulting constitutive relations are of the
form
-iji = L i j k l .
* q k l + +$, (2.17)
where +$ represents part of the viscous terms.

III. Continuum Models of Porous Ductile Solids

Early micromechanical studies of void growth in ductile materials

focussed on the behaviour of a single void in an infinite block of plastic
material (McClintock, 1968; Rice and Tracey, 1969). It was found both
for a cylindrical void and a spherical void that the rate of growth is
strongly dependent on the level of hydrostatic tension, so that fracture by
coalescence of voids would be promoted by a high level of triaxial
tension. This influence of the triaxial tension was also found in a series of
tests on notched tensile specimens, carried out to study the stress state
dependence of ductile failure in high strength steels (Hancock and
Mackenzie, 1976; Hancock and Brown, 1983).

A. APPROXIMATE
YIELDCONDITIONS

For a ductile material containing a certain volume fraction f of voids,

Gurson (1977a,b) has suggested the use of an approximate yield
condition of the form @(d, aM,f) = 0. Here, aii is the average macro-
scopic Cauchy stress tensor, and 0, is an equivalent tensile flow stress
representing the actual microscopic stress-state in the matrix material.
90 Viggo Tvergaard

Gurson (1977a) carried out a number of rigid-plastic upper bound

analyses to evaluate the yield function = 0. Some of these analyses
focussed on a cylindrical volume element, containing a concentric
cylindrical void, some focussed on the analogous spherical problem, and
in some of the cases the effect of elastic unloading regions around the
void was investigated as well. The approximate yield condition derived
on the basis of a spherical model problem with no unloading is of the
form

for q1 = q2 = 1 and f * =f. Here, a, = ( 3 ~ ~ ~ d j / 2is) 'the

~ macroscopic
effective Mises stress, and S" = a '' - Gij4/3 is the stress deviator. The
implication of this analysis is that the voids are assumed to be randomly
distributed, so that the macroscopic response is isotropic. Furthermore, it
is noted that (3.1) reduces to the standard Mises yield condition when
f =o.
Micromechanical studies for materials containing periodic distributions
of cylindrical voids or spherical voids have been carried out by Tvergaard
(1981,1982a) and compared with predictions based on the yield condition
(3.1). This comparison led to the conclusion that the approximation
obtained by (3.1) is improved by taking q1 = 1.5 and q2 = 1.
Final failure of a porous ductile solid occurs by void coalescence.
According to the Gurson model, loss of material stress carrying capacity
occurs when the voids have grown so large that (3.1) has shrunk to a
point (see Figure 2), i.e., at the ultimate value f = l/q,. However, this
value is unrealistically large. Based on a number of experimental studies

h
n

FIG.2. Yield surface dependence on the hydrostatic tension for various values of the
functionf*(f) in (3.1).
 Material Failure by Void Growth 91

Brown and Embury (1973) have found that after nucleation cavities
elongate along the major tensile axis and that two neighbouring cavities
coalesce when their length has grown to the order of magnitude of their
spacing. This local failure occurs by the development of slip planes
between the cavities or simply necking of the ligament. Using a simple
model, they estimated a critical void volume fraction for coalescence of
about 0.15, much smaller than l / q , . A micromechanical model study by
Anderson (1977) for the growth of a spherical void in the stress field in
front of a crack tip gave coalescence at f = 0.25.
Some ways of incorporating the effect of coalescence into the Gurson
model have been discussed by Tvergaard (1982~).The function f*(f) in
(3.1) was introduced by Tvergaard and Needleman (1984) to model the
complete loss of material stress carrying capacity at a realistic void
volume fraction. The function was chosen as
for f 'fc 7

where the void volume fraction at final fracture is denoted by fF , so that

f*(fF) =f: = l/q, (see Figure 2). According to (3.2) the modification of
the yield condition due to the effect of coalescence starts when the void
volume fraction reaches a certain critical value fc . Based on experimental
results (Brown and Embury, 1973) and numerical model analyses
(Anderson, 1977) the values fc = 0.15 and fF = 0.25 were chosen.
Alternative yield conditions for porous ductile solids have been
developed for the purpose of studying powder compacted metals. In most
cases, a quadratic expression in terms of the Mises stress and the mean
stress has been assumed, and the coefficients of the different terms have
been chosen to fit experimental results for uniaxial, or uniaxial plane
strain, tension or compression. Shima and Oyane (1976) have used this
type of procedure to fit experiments for specimens with known void
volume fractions, varying from 0.02 to 0.38. Written in the same notation
as (3.1), the yield function of Shima and Oyane (1976) is

The expression (3.3) was fitted with experiments for sintered copper, but
reasonable agreement was also found with some results for sintered iron
and sintered aluminium.
92 Viggo Tvergaard

Experiments with specimens of known porosity, produced by powder

metallurgy, have been used to test the material model based on (3.1),
developed with the main focus on ductile fracture. Hancock (1982) has
made comparisons with uniaxial compression tests for compacted steels
with a wide range of porosities, O<f<0.18, and has found good
agreement with (3.1) for q l = 1.5. Uniaxial compressive tests for a
powder compacted aluminium alloy have been carried out by Doraivelu
et af. (1984), who have compared the results with a number of different
yield criteria for ductile porous solids, finding rather good agreement
with (3.3). They find poor agreement with the Gurson model, but, in
fact, their comparison only relates to a yield function that approximates
(3.1). A reexamination shows that the agreement is good if q 1 = 2 is
chosen, at least for f up to 0.15. Becker and Needleman (1985) mention
that experimental uniaxial tension data obtained by Spitzig and Rich-
mond for porous iron specimens are well represented by taking q1 = 2.38
and q2 = 0.748. Quite recently, a comparison with a series of experiments
on notched tensile specimens made of sintered iron has been carried out
by Becker et al. (1987a), and good agreement with predictions based on
(3.1) has been found, using q1 = 1.25 and q2 = 0.95.
On the basis of all the experiments mentioned above, for sintered
materials with known porosity, it appears that using a value of q1 in the
range from 1.25 to 2 and q2 = 1 gives a reasonably accurate repre-
sentation of ductile porous solids. This agrees with the micromechanical
studies of Tvergaard (1981,1982a) for periodically distributed voids,
where the main emphasis was on localization predictions.
The rigid-plastic upper bound analyses of Gurson (1977a) were based
on expanding the displacement fields in terms of a few smooth functions.
More accurate predictions would require a larger number of functions, or
a finite element solution with a reasonably fine mesh. The latter approach
has been used by Guennouni and Francois (1987) for a planar array of
cylindrical voids. For some stress states, these numerical plane strain
solutions give rather localized yield zones, and the resulting yield
function has a slightly different shape than that found by Gurson (1977a).
The numerically found yield points have been fitted with a quadratic
expression in terms of a, and &3, analogous to (3.3), but with
different coefficients.

B. NUCLEATION
AND GROWTH
OF VOIDS

During plastic yielding, the void volume fraction changes, partly due to
void growth and partly due to the nucleation of new voids. Experimental
 Material Failure by Void Growth 93

investigations show that the nucleation of new voids occurs mainly at

second phase particles, by decohesion of the particle-matrix interface or
by particle fracture (see, for example, Puttick (1960), and Goods and
Brown (1979)). In some cases, nucleation appears to be controlled by the
plastic strain, whereas in other cases nucleation occurs when the
maximum normal stress on the particle-matrix interface reaches some
critical value.
The rate of growth of the void volume fraction is written on the form
!= (!)growth + (!)nucleation. (3.4)
Since the matrix material is plastically incompressible, the increment due
to growth is given by

where Q: is the plastic part of the macroscopic strain increment. Various

nucleation models have been discussed by Gurson (1977b). The model to
be used here is of the form

proposed by Needleman and Rice (1978).

The effective plastic strain E L , representing the microscopic strain-
state in the matrix material, is related to uMby the incremental equation

where E, is the slope of the uniaxial true stress-natural strain curve for
the matrix material at the current stress level a,. Thus, a material for
which void nucleation is controlled by the plastic strain can be modelled
in terms of (3.6) by taking d > 0 and B = 0. Chu and Needleman (1980)
have suggested a normal distribution for nucleation, so that d and 92 are
given by

B=O for &=(EL),, and i.L>O.

Here, E~ is the mean strain for nucleation, s is the corresponding
standard deviation, and fN is the volume fraction of void nucleating
particles.
For materials, in which nucleation is controlled by the maximum
normal stress on the particle-matrix interface, Needleman and Rice
94 Viggo Tvergaard

(1978) have suggested using the sum aM + a i / 3 as an approximate

measure of this maximum stress, thus taking d = 9. Then, by analogy
with (3.8), d and 9 are given by

for 5, + 4 / 3 = (a, + 4/3),,, and (5M + 033). > 0,

where is the mean stress for nucleation. If both strain-controlled and
stress-controlled nucleation takes place in the material, then d is the sum
of the values given by (3.8) and (3.9). Furthermore, the non-zero values
of d and 93 are only used if the conditions (3.8b) or (3.9b) are satisfied.

C. STRESS-STRAIN
RELATIONS

Assuming that the total strain rate is the sum of the elastic and plastic
parts leads to an incremental stress-strain relationship of the form (2.12)
or (2.16).
In the model of Gurson (1977a), the plastic part of the macroscopic
strain-rate for the porous ductile material is taken to be given by

(3.10)

since normality of the plastic flow rule for the matrix material implies
macroscopic normality (Berg, 1970; Bishop and Hill, 1951). It will be
assumed here that (3.10) also applies during the nucleation of new voids.
The microscopic equivalent tensile flow stress uM and equivalent plastic
strain ELare averages that do not specify the actual microscopic fields in
the matrix material surrounding the voids. However, it is assumed that
the rate of equivalent plastic work in the matrix material equals the
macroscopic rate of plastic work
d'fiF= (1 - f ) D M & L . (3.11)
Thus, using (3.7), the following expression is found for the rate of the
equivalent tensile flow stress in the matrix material

(3.12)

Now, the value of the parameter A in (3.10) is determined from the

 Material Failure by Void Growth 95

consistency condition 6 = 0, required during plastic loading. Substituting

(3.4)-(3.7) and (3.12) into this condition gives

(3.13)

where zklis defined by (2.13), and

(3.14)

(3.15)

(3.16)
Plastic yielding initiaties when @ = 0 and d, > 0 during elastic deforma-
tion, and continued plastic loading requires CP = 0 and m ~ i ~ k f2/0.H
After substituting (3.13) into (2.12), a few standard transformations
(Tvergaard, 1982d) lead to an incremental stress-strain relationship of
the form (2.16), with the instantaneous moduli specified by
(3.17)

M& = r&7kimz, M$l= m Frs 9 r s k l J (3.19)

for elastic unloading,
'=[E,,+mrs9 F rskimki]
G -1
(3.20)
for plastic loading.

It is noted that the symmetry 9'Jk'= B k i i j , satisfied by the tensor of

elastic moduli (2.14), is not satisfied by the present instantaneous moduli

L'Jki.This is partly due to the last term in (3.18), which has a noticeable
effect for the dilatant material considered here, and partly due to
M& # MZ in cases where 93 # 0.
The incremental constitutive relations derived here could also directly
be used in connection with other approximate yield conditions than (3.1),
e.g., with (3.3). Some of the expressions (3.14)-(3.16) would have to be
changed according to the different yield condition, but otherwise the
96 Viggo Tvergaard

constitutive relations would have the same form. It is noted that Shima
and Oyane (1976) did discuss the incremental stress-strain relations
associated with (3.3).
The uniaxial true stress natural strain curve for the matrix material,
referred to in (3.7), is generally obtained from a uniaxial tensile test. In
much of the following discussion, it will be assumed that the curve can be
represented by a piecewise power law

where ay is the uniaxial yield stress and n is the strain hardening

exponent.

IV. Localization of Plastic Flow

During the deformation of ductile solids, it is frequently observed that

at some stage a smoothly varying deformation pattern develops into a
pattern involving highly localized deformations in the form of shear
bands. Once localization has taken place, the strains inside the band
become very large without contributing much to the overall deformation
of the body. This can lead to shear fracture at an overall strain only
slightly larger than that at the onset of localization.
For a homogeneously deformed rate-independent solid, the state at
which bifurcation into a shear band mode is first possible coincides with
loss of ellipticity of the equations governing incremental equilibrium (Hill,
1962; Rice, 1977). Thus, the analysis of this material instability follows
the theoretical framework due to Hadamard (1903).
The value of the critical strain for loss of ellipticity is very sensitive to
the constitutive law (Rudnicki and Rice, 1975; Rice, 1977). The classical
elastic-plastic solid with a smooth yield surface and normality of the
plastic flow rule is quite resistant to localization, but deviations from the
classical model can have a strong effect. Thus, localization at a realistic
 Material Failure by Void Growth 97

strain level is predicted for a solid that develops a vertex on the yield
surface, as arises in physical polycrystalline models based on the concept
of single crystal slip. Also dilatational plastic flow and non-normality of
the plastic flow rule, as induced by the porous ductile material model
discussed in Section 111, have a significant destabilizing effect (Rice, 1977;
Needleman and Rice, 1978). Furthermore, the onset of localization is
very sensitive to small material inhomogeneities.
Simple model studies for shear localization can be carried out by a
procedure analogous to that developed by Marciniak and Kuczynski
(1967) for the problem of plane stress sheet necking. In these analyses,
an initial material inhomogeneity, such as a higher concentration of void
nucleating particles, is assumed inside a plane slice of material, and the
stress-states inside and outside this slice of material, respectively, are
assumed to remain homogeneous throughout the deformation history.
In the model, illustrated in Figure 3, the principal directions of the
stresses and strains outside the band are assumed to remain fixed, parallel
to the x'-axes in the Cartesian reference coordinate system. The major
principal stress outside the band is taken to be in the x'-direction, and the
slice of material containing the initial inhomogeneity is assumed parallel
with the x3-axis, with the initial angle of inclination and the unit
normal vector n, in the reference state. Then, the current angle of
inclination I) of the band, at any stage of the deformation, is given by

tan I) = exp(e'; - E ; ) tan qr. (4.1)

Here, E, are principal logarithmic strains, and ( )" denotes quantities
outside the band while ( )b will denote quantities inside the band.
The field quantities outside the band are specified by the external
loading, while the quantities inside the band have to satisfy compatibility
and equilibrium over the band interface. Compatibility requires con-
tinuity of the tangential derivatives of the displacement components ui

FIG.3. Shear band in a homogeneously strained solid.

98 Viggo Tvergaard

over the interface, so that the displacement gradients can be expressed by

uipi = uzj + cini, (4.2)
where ci are parameters to be determined. Equilibrium requires balance
of the nominal tractions on each side of the interface
(Ti)b= (Ti)O, (4.3)
where the nominal traction components T i on a surface with reference
normal nj are given by (2.10). Now, a set of incremental equations for ti
are obtained by substituting the incremental constitutive relations (2.16)
into the incremental form of (4.3), using (4.2) to express the strain
increments fit..
In a case where there is no material inhomogeneity, the incremental
equations for di are homogeneous, allowing only the trivial solution until
bifurcation occurs. The first such bifurcation into a shear band (for any
angle of inclination q of the band) marks the loss of ellipticity of the
governing differential equations (Hill, 1962; Rice, 1977).
For porous ductile materials, described by the Gurson model, Yama-
mot0 (1978) has analysed the bifurcation into shear bands, as a function
of the initial void volume fraction fr. In the limit of f r = O , where the
material model reduces to &flow theory, the critical strain for localiza-
tion is essentially infinite, but the critical strain decays very rapidly for
increasing fr, so that just a few percent porosity is enough for the
material instability to be predicted at a realistic strain level.
A detailed micromechanical study of shear band bifurcation that
accounts for the interaction between neighbouring voids and the strongly
nonhomogeneous stress distributions around each void has been carried
out by Tvergaard (1981). A power-hardening elastic-plastic solid was
considered, containing a doubly periodic array of circular cylindrical
voids as shown in Figure 4. Due to symmetries, only the region hatched
in Figure 4 needs to be analysed numerically, when this model material is
subjected to plane strain tension in one of the coordinate directions.
Bifurcation into another periodic pattern was also analysed by consider-
ing only the hatched region in Figure 4, making use of several symmetry
properties of the repetitive pattern. This bifurcation mode was used to
represent the deformations inside a shear band, as illustrated in Figure 5 .
The equilibrium conditions (4.3) on the band interface were satisfied
exactly, while compatibility was only satisfied on the average, since the
two periodic patterns are not locally compatible along the characteristic
direction.
 Material Failure by Void Growth 99

ololo

FIG. 4. Doubly periodic array of circular cylindrical voids.

FIG.5 . Shear band bifurcation mode found numerically for model material containing a
periodic array of cylindrical voids.
100 Viggo Tvergaard

By comparison with the full numerical shear band bifurcation analyses

described above, it was found that the Gurson model gives somewhat too
large localization strains, but that reasonably good agreement is obtained
by using q1 = 1.5 in (3.1). It is noted that no loss of ellipticity of the
governing differential equations happens in the full numerical bifurcation
analysis, where the matrix material is described by isotropic hardening
&flow theory, but still there is good agreement with loss of ellipticity
according to the approximate continuum model. This agreement with the
more detailed micromechanical study gives good confidence in localiza-
tion predictions obtained by the continuum model of a porous ductile
solid, and such predictions do play an important role in studies of ductile
fracture.
When there is a material inhomogeneity, the incremental equations for
ti obtained from (4.2) and (4.3) are inhomogeneous. Then the rate of
deformation inside the band will gradually increase relative to that
outside, and localization is defined by the onset of elastic unloading
outside the band. For a given size of the inhomogeneity, the localization
strain depends on the initial angle of inclination q1of the slice containing
the imperfection, and the variation of the localization strain with q, must
be determined to find the most critical angle of inclination. Such analyses
have been carried out by Yamamoto (1978) for inhomogeneities in the
form of a higher void volume fraction inside the band, and Saje, Pan, and
Needleman (1982) have included the effect of an additional constant
density of void nucleating particles.
Even if there is no initial porosity, f P = f;=O, so that the initial
response is that of a homogeneous bar, an initial inhomogeneity can be
represented by a non-uniform distribution of void nucleating particles.
Such cases have been studied by Tvergaard (1987a), with the in-
homogeneity specified by AfN, so that the volume fractions of void
nucleating particles inside and outside the band are related by
fk =f"N + AfN (4.4)
In all cases, the value of the initial angle of inclination qI that leads to
first localization tends to be much smaller for small imperfections than for
larger imperfections. But independent of the value of the localization
strain, the corresponding value of the current angle of inclination at
first localization is always about 43" in uniaxial plane strain tension,
rather close to the critical inclination for shear band bifurcation in the
homogeneous solid.
 Material Failure by Void Growth 101

2 W =26.9'- - W I =28.6'
&O I

W = 30.L' /WI=30.L'
I
1
0 0.1 0.2 0.3 0.4
E0
FIG. 6 , Maximum principal strain E~ inside shear band vs. strain EO outside. Initial
voidage fp = O.OOO1, f; = 0, and stress controlled nucleation fN = 0.04, uN = 2.10,, s =
0.40, (Tvergaard, 1982d).

The onset of localization is usually considered an indication of failure,

since the subsequent severe straining inside the band contributes little to
the overall strain. Several shear band computations have been continued
beyond the onset of localization (Tvergaard, 1982c), and these results
confirm that failure occurs by coalescence in a void-sheet inside the band,
while the material outside the band remains elastically unloaded.
However, the results shown in Figure 6 for a material with stress-
controlled nucleation show that the behaviour can be somewhat more
complex (Tvergaard, 1982d).
In Figure 6, the material is characterized by fN = 0.04, uN = 2.1ay, and
s = 0 . 4 in
~ (3.9),
~ with a very small inhomogeneity fi = O.OOO1, f: = 0, so
that the results are representative of the post-bifurcation behaviour
corresponding to a completely homogeneous solid. For 3, = 30.4", the
band has rotated to the critical current angle = 42.8" for bifurcation at
the corresponding critical strain E O = 0.237; however, elastic unloading
occurs only briefly, and subsequently the deformations remain nearly
uniform, -- 1. For a smaller initial angle, 3, = 28.6", localization
E ~ / E O

occurs at a strain beyond the elliptic-hyperbolic interface, but then the

material outside the band yields again, until final failure takes place at a
secondary localization. A somewhat similar saturation of localized flow
has been found by Hutchinson and Tvergaard (1981) for a material that
forms a vertex on the yield surface. For an even smaller initial angle,
102 Viggo Tvergaard

v, = 26.9", localization occurs later, but leads to direct failure. Thus, with
stress controlled nucleation, where non-normality of the plastic flow rule
(3.13) occurs, the localization leading to final failure may be a secondary
localization.
For materials with stress controlled nucleation, Needleman and Rice
(1978) and Saje, Pan, and Needleman (1982) have shown that a sudden
burst of nucleation (e.g., a very small standard deviation s in (3.9))
causes very early localization. However, a computation continued beyond
such an early onset of localization has shown that the localized strain
inside the band increases only a little (during the burst of nucleation),
and subsequently the material yields both inside and outside the band in
a rather uniform manner (Tvergaard, 1982d). Thus, the early localiza-
tions due to a burst of nucleation appear not to have a strong effect on
final material failure.

V. Cell Model Studies

Micromechanical investigations of the behaviour of ductile porous

solids have mostly focussed on a characteristic volume element containing
a single void. This includes an early numerical study by Needleman
(1972) for a square array of cylindrical voids (Figure 4), the analyses of
Gurson (1977a) for a spherical volume with a concentric spherical void, a
numerical analysis by Nemat-Nasser and Taya (1977), and the localiza-
tion studies by Tvergaard (1981, 1982a). Here, a few more recent cell
model studies will be discussed that focus on void nucleation and on void
growth to final failure.

OF VOIDS
A. NUCLEATION

The most obvious consequence of void nucleation in plastically

deforming metals is the generation of damage, but another important
consequence is a reduction of the macroscopic strain hardening capacity
during the nucleation process. This softening due to nucleation has a
dramatic effect on the prediction of localization and thus on the
occurrence of ductile fracture.
As discussed in Section IV, a burst of stress controlled nucleation can
result in very early localization according to the Gurson model (Need-
 Material Failure by Void Growth 103

leman and Rice, 1978; Saje et a f . , 1982), but strain-controlled nucleation

or less sudden stress-controlled nucleation can also have a significant
influence in reducing the critical strain for localization. The degree of
nucleation-induced softening predicted by this material model is a result
of assuming the nucleation model (3.6) and also assuming that (3.10)
remains valid during nucleation.
A number of void nucleation studies have been carried out by
Hutchinson and Tvergaard (1987) to get a more detailed understanding
of the softening behaviour. Results for nucleation of an isolated spherical
void in an infinite matrix under triaxial remote stressing were used to
predict the overall stress-strain behaviour in cases where interaction
between voids is negligible. Nucleation of the void, at stresses Z , that
have been applied proportionally, causes a redistribution of stress and
additional straining of the matrix, above the straining that occurs in the
absence of nucleation. This additional macroscopic straining during the
nucleation of the void volume fraction f is denoted by f A E , , such that
ZijfAEijis the extra work done by the remote stresses due to nucleation
(here the macroscopic components refer to a small strain analysis on
current Cartesian coordinates). Thus, if Miikl are the incremental com-
pliances for f = 0, so that kij= Mjjk&/ in the absence of nucleation, the
macroscopic strain increments during first void nucleation are

kij = Mijk$kl +f h E i j . (5.1)

For the nucleation of a spherical void in an isotropic hardening Mises
material an excellent approximation of the computed results is

Here, 2, , 2, , and 2; are the macroscopic mean stress, Mises stress, and
stress deviator, respectively. The functions F and G also depend
implicitly on E J E and on the total strain increment during nucleation,
but this dependence is weak. The variation of F and G with the stress
triaxiality Z,/Z, found in the computations, is illustrated in Figure 7.
For the Gurson model the strain contribution due to nucleation can be
calculated from (3.4)-(3.12). For the first voids nucleated (when f = 0) at
Z,, the result is
104 Viggo Tvergaard

I
I6 -
-
14-
-
12 -
-
10 -
-
8-
-
6-
-
4-
-
2-
-

FIG.7. Dependence of F and G in (5.2) on stress triaxiality (Hutchinson and Tvergaard,

1987).

assuming E , << E. Then, in terms of (5.2), the Gurson model gives

which is also plotted in Figure 7. It is seen that in the low triaxiality

range, the computed values of F a r e well represented by (5.4) for q1 = 1.5
and q2 = 1, while q , = 1.25 gives a better approximation at higher
triaxialities. The Gurson model neglects a dilatational contribution to
AE,, but this contribution is inherently small anyway when E, << E.
Nucleation is included in the Gurson model in a highly coupled
manner, so that the quantitative effects of nucleation on macroscopic
 Material Failure by Void Growth 105

ks

FIG. 8. Cylindrical cell model for periodic array of spherical void-nucleating particles
(Hutchinson and Tvergaard, 1987).

behaviour are not transparent. However, the fact that the model is in
good agreement with the micromechanical calculation of AE, increases
the confidence in this model.
The effect of a uniform distribution of spherical particles that nucleate
voids simultaneously has been investigated by numerical solution of the
axisymmetric model problem illustrated in Figure 8 (Hutchinson and
Tvergaard, 1987). Results for nucleation at constant marcoscopic stress
are shown in Figure 9, and the same cases have been analysed for
nucleation at constant macroscopic strain. The results for AE, obtained
by these cell model calculations have been compared with (5.1) and (5.2),
interpreting Eij as logarithmic strain increments and gij as true stress
increments. The values of F calculated from Figure 9 are not too much
higher than the corresponding value shown in Figure 7, whereas the
values found for G are a great deal larger. This agreement is quite reason-
able since the nucleated void volumed fraction, ! =0.01, in the cell
model studies is not very small and the matrix material is assumed
power-hardening while the tangent modulus E, used in (5.2) is that at
the onset of nucleation.
Both the calculations for an isolated spherical void in an infinite matrix
and the cell model calculations have been repeated with a kinematic
hardening description of the matrix material. These calculations show
that the values of both F and G in (5.2) are significantly increased due to
the yield surface translation. This is mainly an effect of the nonpropor-
tional stress history in the matrix material near the particle, due to
106 Viggo Tvergaard

2 - nucleaf ion

FIG. 9. Overall stress-strain behaviour for nucleation at constant stress, f = 0.01,

according to cylindrical cell model (Hutchinson and Tvergaard, 1987).

nucleation. In such circumstances, elastic-plastic material response is

quite sensitive to the high curvature, or rounded vertex, that often
develops at the loading point of the subsequent yield surface, and
kinematic hardening was applied to represent this effect.
The reduction of the macroscopic hardening rate represented by the
additional term f AE, in (5.1) was estimated by Hutchinson and
Tvergaard (1987). It was found that the hardening rate can even become
negative, for a relatively small amount of void nucleation, if nucleation
occurs over a relatively small range of straining.
Model studies of the nucleation process have also been carried out by
Needleman (1987). This work has focussed on the process of interfacial
decohesion, which has been described approximately by assuming that
the interfacial tractions are derived from a potential function 4. In terms
of the normal and tangential separations u, and u, , a characteristic length
scale 6, and two other parameters a,,,,, and a,the potential function was
defined as

with the normal and tangential tractions given by T, = a$/&, and

 Material Failure by Void Growth 107

T, = a@/&, , respectively. This procedure has made it possible to follow

in detail the gradual separation of the interface, dependent on the stress
state in the material. The procedure has also been used by Nutt and
Needleman (1987) to study the separation between matrix and fibers in
an aluminium alloy reinforced by Sic whiskers. The predicted patterns of
separation at the fiber end are virtually identical to those found by direct
observations using an electron microscope.

B. GROWTH
AND COALESCENCE
OF VOIDS

The axisymmetric model problem illustrated in Figure 8 has been used

in a number of studies of the growth of an initially spherical void under
various stress-states. Thus, Anderson (1977) has used the model to study
void coalescence in an ideally plastic material in front of a crack tip, and
Tvergaard (1982a) has used the model to study localization in a material
containing spherical voids.
Quite recently Koplik and Needleman (1987) have made several
analyses of this type to study void coalescence in an isotropically
hardening material. The voids are assumed present from the beginning
and void growth is studied for a fixed ratio of the radial and axial
macroscopic true stresses. The computations are carried out for two
different initial void volume fractions, fi = 0.0013 and fi = 0.0104, respe-
ctively. Initially, the rate of growth of the void volume fraction and the
rate of macroscopic radial straining are rather constant, as functions of
the macroscopic effective strain; but gradually the former increases while
the latter decreases, until radial straining stops. Subsequently, the
deformation takes place in a uniaxial straining mode, where all deforma-
tion is localized in the ligament between radially adjacent voids. During
this final stage of localized flow, the void volume fraction grows rapidly,
while the macroscopic true stress level decays rapidly.
Koplik and Needleman (1987) have compared the predictions of their
cell model studies with the modification (3.2) in the Gurson model,
introduced by Tvergaard and Needleman (1984), to account for final
failure. Rather good agreement was achieved by using the parameter
values q1 = 1.25, q2 = 1, and fF = 0.13, while taking fc = 0.03 for fi =
0.0013 and fc = 0.055 for fl = 0.0104. The value of fF was chosen to get a
good approximation to the void volume fraction versus macroscopic
effective strain curves at f >fc. It was also found that q1 should be
increased for decreasing strain hardening.
108 Viggo Tvergaard

1
fc

/
P
/
0.10 - / -
/
/
/
/
'0
/

0.05- >// -
d

0 0.05 f, 0.10
FIG. 10. Dependence of the critical value fc in (3.2) on the initial void volume fraction
fr , according to various cell model studies.

Other cell model studies of the same type have been carried out by
Becker et al. (1987a). Here the void volume fracture at final failure was
chosen as fF = 0.25 and the values of fc were found to be 0.12, 0.07, and
0.04 for the initial void volume fractions 0.06, 0.026, and 0.004,
respectively. Koplik and Needleman (1987) found that the value of f c
varies slowly with stress triaxiality and matrix strain hardening, but
clearly there appears to be a rather strong dependence on the initial void
volume fraction. The values of fc, as a function of fi , found in the two
investigations are plotted in Figure 10. Thus, on the basis of these
micromechanical investigations, it appears that the most realistic predic-
tions are obtained by using an f c value that depends on fi (or on the
volume fraction fN of void nucleating particles). Clearly, the values in
Figure 10 are smaller than the value fc = 0.15 suggested by Tvergaard and
Needleman (1984), based on the work of Brown and Embury (1973).
The results of these cell model studies have been used by Becker et al.
(1987a) in a detailed comparison of numerical and experimental results
for void growth and ductile failure in the non-uniform multiaxial stress
 Material Failure by Void Growth 109

fields of notched bars. The notched tensile specimens were machined

from partially consolidated and sintered iron powder compacts, which
had been made specifically to study the effects of porosity on mechanical
behaviour. The pores were fairly random and nearly spherical in the
higher density materials. Furthermore, the initial void volume fraction
was large compared to the volume fraction of particles, so that these
experiments allowed for a study of void growth and coalescence,
independent of void nucleation issues. Quite good agreement between
experiments and computations was obtained for several combinations of
notch shapes and initial void volume fractions, which gives good
confidence in the continuum model, and supports the use of the variable
fc indicated in Figure 10.
The values of fc found by the micromechanical studies (Figure 10) are
particularly interesting because the void (or particle) volume fractions in
structural alloys are usually quite low, so that the relevant fc value is
significantly smaller than 0.15. This could be part of the explanation for
the very low void volume fractions measured in the material near the
fracture surface in the centre of the neck of round tensile test specimens
(Cialone and Asaro, 1980; MOUSSY, 1985).

C. Two SIZE-SCALES
OF VOIDS

Several structural alloys contain two size-scales of particles, i.e., a

population of relatively large particles with low strength and a population
of much stronger small particles. Thus, structural steels tested by
Hancock and Mackenzie (1976) contain two populations of particles with
average diameters around 100 pm and 1pm, respectively, and aluminium
alloys tested by Hahn and Rosenfield (1975) contain particles with
average diameters of about 5 pm and 0.1 pm. In such materials,
nucleation first occurs at the large weak particles, and during the
subsequent growth of the large voids, a second population of small scale
voids starts to nucleate, primarily in the regions of stress and strain
concentrations near the larger voids.
A cell model study of the interaction between two size-scales of voids
has been carried out by Tvergaard (1982b). The model material initially
contains a periodic array of circular cylindrical voids, as shown in Figure
11, so that, due to symmetries, only the rectangular region DEFG in
Figure 11 needs to be analysed numerically. The matrix material around
110 Viggo Tvergaard

0 0
0

0 0 0
FIG. 11. Periodic array of circular cylindrical holes representing large-scale voids
(Tvergaard, 1982b).

the cylindrical voids is described by the Gurson model, and thus the
cylindrical holes represent the large-scale voids, while the small-scale
voids are represented in terms of the continuum model of a porous
ductile material. The larger voids are present from the beginning, while
the small voids nucleate either by a strain criterion or a stress criterion.
The material is subjected to generalized plane strain conditions, so that
the strain in the x3-direction is uniform.
The mesh used for this numerical analysis is chosen so that it allows for
localization of plastic flow in a shear band between two larger voids. In
the early stage of the deformation, nucleation starts to occur in the highly
stressed and strained material near the cylindrical holes, and the void
volume fraction varies smoothly along the surface of the holes. Subse-
quently, the nucleation and growth of voids starts to concentrate in a
narrow band that grows into the material, and a shear band forms. Final
fracture involves the formation of a void-sheet between two larger voids,
as has been observed in a number of experiments (e.g., Cox and Low
(1974) and Stone and Psioda (1975)).
This model analysis was used to predict final failure under a number of
different macroscopic stress states including plane strain, E , ~ ,= 0, or
axisymmetric conditions approximated by taking either E~~~ = E , or
SIII= S,. Here, E , and S, are the macroscopic logarithmic strain and true
stress in the xl-direction, etc., while S,, S,, and E, are the macroscopic
mean stress, Mises stress, and effective strain, respectively, and S,, is the
major principal stress. Values of the stress triaxiality parameter S,/S, vs.
 Material Failure by Void Growth 111

2.0
o fB0.1 over 35% of ligament
Sm’se
CD fB0.1 over 60%
1.6 I, X=0.5--& 0 f >0.1 over 100% I4

n f ~ 0 . 2over 35%
m.X = O . 5 - 0 4
1.2
be = SI’s*
0.8
= 0.25
=0.5
-Qel
’ I: Em=O
II: Em=E1
= 0.25 -04
=o
/0-a. rn: sm=s1
0.4
= 0.25 V0
o-aa
=o
0
0 0.02 0.04 0.06 0.08 0.10 0.12 C, 0.14
FIG.12. Stress triaxiality SJS, vs. effective strain E, near final failure in stress and
strain fields ranging from plane strain to axisymmetry (Tvergaard, 1982b).

E , at stages near final failure are plotted in Figure 12 for the different
stress states considered, corresponding to one particular set of material
parameters.
In experimental investigations by Hancock and Mackenzie (1976) and
Hancock and Brown (1983) for notched tensile specimens made of high
strength steels, it has been found that the results can be represented
approximately by a single failure locus in a plot of stress triaxiality vs.
effective plastic strain at fracture initiation. This experimental observa-
tion does not agree with localization predictions for uniformly strained
specimens, where is is known that materials are generally far more
resistant to shear band formation when strained under axisymmetric
conditions than under plane strain conditions (Needleman and Rice,
1978; Saje et al., 1982). However, the interaction of two size-scales of
voids does lead to agreement with the experimental observations, as
shown in Figure 12, even though localization is also here an important
part of the failure mechanism.
It should be noted that a reasonable agreement with the experimentally
observed failure locus has also been found by Needleman and Tvergaard
112 Viggo Tvergaard

(1984a) in a numerical analysis of notched bars, where the interaction of

different void populations was not accounted for. Part of the reason for
this agreement is that the notch geometries tend to suppress localization.
In a couple of more recent cell model studies, nucleation at both the
large particles and the small particles is incorporated into the analysis, by
representing both types of particles in terms of the Gurson model
(Tvergaard, 1987c,d). The small scale particles are assumed to be
uniformly distributed, with nucleation governed by a strain controlled
criterion, while the larger inclusions are represented as “islands” of
increased density of the amplitude of the void nucleation function.
Nucleation in the “islands” is taken to be stress controlled, so that
nucleation starts here, leading first to a soft-spot and subsequently to a
large void. For a plane strain analysis (Tvergaard, 1987c), Figure 13
shows deformed meshes and curves of constant void volume fraction at

FIG. 13. Deformed meshes and curves of constant void volume fraction at two stages for
material in uniaxial plane strain tension. Larger inclusions represented as “islands” of
increased particle density (Tvergaard, 1987~).
 Material Failure by Void Growth 113

two stages. At the first stage, the large cylindrical void has formed and
localization is starting, although the mesh is still rather uniform, while at
the second stage void-sheet failure is about to happen.
The same type of analysis has been carried out for a three-dimensional
array of “islands” representing larger inclusions, where a full 3D
numerical computation is necessary to study the final failure mode
(Tvergaard, 1987d). This analysis includes a numerical representation of
a 3D shear band, initiating in complex non-uniform stress and strain
fields.

VI. Formation and Growth of Cracks

The analysis of material failure by shear localization between two

larger voids, described in the previous section, was brought close to final
failure, as indicated in Figure 12. However, predicting the development
of an open crack requires some special modelling, which was not included
in this void interaction analysis.

A. CRACKFORMATION

A full description of final failure has been included in a numerical finite

element analysis of ductile shear fracture at the free surface of a
specimen subject to uniaxial plane strain tension (Tvergaard, 1982d).
Here, loss of material stress carrying capacity as the voids start to
coalesce was modelled by an additional term in (3.4), as suggested by
Tvergaard (1982c), and an element vanish technique was introduced to
describe the final formation of an open crack. When the failure condition
is met in an element (the yield surface (3.1) has shrunk to a point), the
element vanishes in that it no longer contributes to the virtual work
integral (2.11). For numerical stability, the elements are actually taken to
vanish slightly before the failure condition is met, and the nodal forces
arising from the small remaining stresses in the nearly failed elements are
gradually released in subsequent increments.
The surface region analysed corresponds to a semi-infinite solid
occupying the half-space x 2 5 0, with a slight initial surface waviness of
the form w = - 46, cos(nx’/lo). The x2-axis is normal to the surface, and
the x’-axis is parallel to the surface. For this solid under tension in the
114 Viggo Tvergaard

x'-direction, solutions are available that satisfy symmetry conditions at

the planes x1 = 0 and x 1 = l o , and such solutions are investigated by
numerical analyses for a rectangular region of width lo and depth h a , so
that hallo>> 1. The material has no porosity initially, but voids nucleate,
and at some stage localization of plastic flow occurs. The voids keep
growing inside the shear bands, leading to void-sheet failure. The
formation of open shear cracks predicted by the numerical analysis is
illustrated in Figure 14. It should be noted here that the accurate
prediction of plastic flow localization in a numerical analysis requires a
careful mesh design, so that the mesh agrees with the critical orientation
of shear bands at the strain where localization wants to occur (Tvergaard,
Needleman, and Lo, 1981; Needleman and Tvergaard, 1984b).
The ductile fracture process in a round tensile test specimen has been
analysed by Tvergaard and Needleman (1984). Here, a more realistic
model of the void volume fraction evolution during coalescence was used,
by introducing the function f * ( f ) defined in (3.2). Initially the bar has
uniform thickness with no porosity; then necking occurs, and sub-
sequently voids nucleate and grow rapidly inside the neck region, where
increased triaxial tension develops. Figure 15 shows the predicted void
volume fraction distribution at three stages, i.e., just before coalescence

'.;
starts in the centre of the neck, at a later stage when an open

open

.:
crack

1=01

'\
(a1 (bl (Cl
FIG. 14. Contours of constant maximum logarithmic strain E and void volume fraction f
in a surface region, where the top is the surface. Nucleation is strain controlled, with no
initial voids. The average strain E , along the surface is (a) E , = 0.268, (b) E , = 0.307, (c)
E, = 0.332 (Tvergaard, 1982d).
 n--
Material Failure by Void Growth 115

f = 01 f= 005

,c15

F r
la1

open crack

I bl

(CJ

FIG.15. Contours of constant void volume fraction in the neck of a round bar tensile
test. The load P decays with increasing crack growth: (a) P/P,,, = 0.731, (b) P/P,,, =
0.521, (c) P/Pmnx= 0.032 (Tvergaard and Needleman, 1984).

penny-shaped crack has developed, and finally when zig-zag growth of

the crack has given the conical void-sheet fracture known as the
cup-cone.
Figures 14 and 15 illustrate the significant difference between the
fracture mode in a plane strain tensile test and that in a round tensile test
specimen, as is also observed experimentally (Speich and Spitzig, 1982).
The analysis leading to Figure 14 did not account for necking. But in the
neck region of a plane strain tensile test specimen, final fracture occurs in
a shear mode induced by localization (Saje, Pan, and Needleman, 1982;
Tvergaard, Needleman, and Lo, 1981), while the void volume fraction
outside the band is still quite small. On the other hand, in the round bar
tensile test specimen, fracture by void coalescence in the centre of the
neck is predicted prior to localization (Figure 15).
116 Viggo Tvergaard

B. CRACK
TIPFIELDS

The ductile fracture process at a crack tip differs significantly from that in
more homogeneously strained circumstances, such as the plane strain or
axisymmetric tensile test specimens described above (Rice, 1976). The
large gradients of the stress and strain field near the crack tip typically
result in critical conditions for failure being reached in part of the
near-tip material at loads far less than the fracture load. Here, a
characteristic length scale of the material plays a role, such as the particle
spacing.
Rice and Johnson (1970) have studied the interaction of the crack-tip
with a neighbouring void, using a slip line field analysis together with
results of Rice and Tracey (1969) for the growth of an isolated spherical
void to estimate the critical crack tip opening displacement for crack
growth. An analogous study has been made by McMeeking (1977) on the
basis of a full finite element solution for the blunting crack tip fields. The
results of both analyses tend to overestimate critical crack tip opening
displacements observed experimentally.
The interaction between a single cylindrical void and a plane strain
tensile crack has been analysed numerically by Aoki et al. (1984) and
Aravas and McMeeking (1985), accounting also for small scale voids in
the surrounding material. The small scale voids are represented in terms
of the porous ductile material model as suggested by Tvergaard (1982b)
in the computation illustrated in Figures 11 and 12, and the crack growth
criteria used refer to failure of the porous material.
In a similar analysis, also based on small scale yielding at a mode I
plane strain crack, Needleman and Tvergaard (1987) consider the
influence of a whole array of uniformly spaced larger inclusions in the
material around the crack tip. In this investigation, no pre-existing voids
are assumed present, and the large weak inclusions are represented
by an array of “islands” of increased density of the amplitude of the
void nucleation function, with stress-controlled nucleation inside the
“islands.” During crack tip blunting, self-similar stress and strain fields
sweep over the near-tip material (McMeeking, 1977), so that an inclusion
at initial distance R from the tip is reached by the stress peak when
J/uoR= 0.5, while the strains are still very small (J and a, denote the
J-integral and the yield stress). Thus, large voids close to the crack tip
nucleate rather early, when the stress peak sweeps by, whereas the small
voids nucleate much later in the material close to the blunting crack tip,
where large strains develop.
 Material Failure by Void Growth 117

Ibl
I01

IC1

FIG. 16. Contours of constant void volume fraction near the tip of a blunting crack in a
material containing two size-scales of voids. (a) J/unDn= 0.856, (b) J/UoDo = 1.89, (c)
J/unDn= 3.89 (Needleman and Tvergaard, 1987).

Four different distributions of the larger inclusions near the crack tip
have been investigated by Needleman and Tvergaard (1987). Contours of
constant void volume fraction in the near tip region are shown for one of
the cases in Figure 16, where Do is the initial inclusion spacing. In this
case, the inclusion nearest the crack tip is off the crack line. The large
void at about 45" from the crack tip develops most rapidly, and crack
growth involves localized shearing leading to void-sheet fracture in the
material between this large void and the crack tip. Subsequently, the new
crack tip interacts with the next larger void, and thus the crack keeps
growing (Figure 16c).
The numerical results have been used to estimate the tearing modulus,
which was introduced by Paris et al. (1979) to evaluate the limits of stable
crack growth. The tearing modulus is defined in terms of the J-resistance
 Viggo Tvergaard

3- DATA FROM FRACTURE TOUGHNESS TESTS

bf 10
0
. /
HIGH STRENGTH STEEL ( M n S INCLUSIONS),
RICE AND JOHNSON 11970), PELLISSIER (1968)
u ALUMINUM 2 0 0 0 SERIES VAN STONE
ALUMINUM 7 0 0 0 ' S E R I E S ) E T AL.(1974)
0 A l S l 4 3 4 0 STEEL (MnS INCLUSIONS)

.
2 - IB-NI 2 0 0 MARAGING STEEL
lTi(C.N) INCLUSIONS) LOW(1974
MILDSTEEL (SPHEROIDIZED Fe,C INCLUSIONSI.
RAWAL AND GIJRLIND (1976)
DATA FROM CRACK-GROWTH INITIATION TESTS
OEnlA MILD STEEL
I - 0PRE-STRAINED EnlA GREEN AND

FIG.17. Crack-tip opening b, at fracture initiation, related to particle spacing D and

particle sue 2r0, plotted among the experimental results summarized by McMeeking (1977).

curve J(Aa) as
T = - E- *d J
da
Subsequently, using the numerically calculated slopes d l d a , the critical
value JI, for the onset of crack growth was estimated in three of the
computations. Thus, the numerical analyses discussed here have made it
possible to relate the microstructure, represented by the particle distribu-
tion and nucleation criteria, to fracture mechanics parameters such as J,,,
T, or the crack opening displacement. The values found for the crack
opening displacement at fracture initiation bf are compared with ex-
perimental results summarized by McMeeking (1977) in Figure 17. This
figure shows the ratio of the crack-tip opening displacement and the
particle spacing vs. the ratio of the particle spacing and diameter. It is
seen that the computed values, marked a, b, and d in Figure 17, are in
reasonable agreement with the experimental results.
A rather different mechanism of crack growth by void coalescence has
been studied by Becket et al. (1987b), both experimentally and theoreti-
cally. The material considered is an AI-Li alloy with coarse grain
boundary particles, which give rise to void nucleation at the grain
boundaries, so that the final fracture is predominantly intergranular. This
phenomenon of ductile grain boundary fracture is readily identified by
 Material Failure by Void Growth 119

the presence of microvoid dimples on the fracture surface and is distinct

from the brittle grain boundary fracture observed in other cases.
In the numerical analysis, the grains near the crack tip are modelled as
roughly circular regions free of void nucleating particles, while the
nucleation and growth of voids in the bands between these grains is
represented in terms of the ductile porous material model. Figure 18
shows grain distributions ahead of the crack tip for two of the cases
analysed, corresponding to the grain sizes 200pm and 60pm, respec-
tively, where the grains are outlined by curves of constant volume
fraction of void nucleating particles. If the voids develop as a string of
pearls along the grain boundary, then the most realistic initial width of
the grain boundary porous zone, in the present type of model, is of the
order of the void spacing (see comparisons made by Tvergaard (1982a)).
This width is chosen somewhat too large in the distributions shown in
Figure 18 in order to avoid the use of an extremely fine mesh. The initial
part of the crack growth has also been analysed with a more realistic
small width of the grain boundary porous zone.
The development of damage and crack growth found in the numerical
analyses for the different grain diameters show that voids nucleate at the
grain boundaries in the near vicinity of the crack tip, and that crack
growth by void coalescence occurs on a zig-zag path following grain
boundaries near the crack line.
Also these computations have been used to draw the J-resistance
curves, J vs. Aa, and the slopes of these curves are used to estimate the
tearing modulus T according to (6.1) for each case. Then, the critical
value J,, for the onset of crack growth is estimated by back extrapolation
using the slope d l l d a , and the critical value K I , of the stress intensity

+---- M (b)
IOOpm lOOpm

FIG.18. Material with void nucleating particles in the grain boundary modelled in terms
of porous material model, by considering the grains as roughly circular regions free of
particles (Becker et al., 1987b).
120 Viggo Tvergaard

factor is obtained from the plane strain small scale yielding relation

The predicted fracture toughness decreases significantly as the width of

the grain boundary porous zone is decreased. However, in spite of the
rather crude description of the grain boundary porous zone used in these
analyses, the quantitative agreement between the model predictions and
the experimentally measured values is quite reasonable (within a factor of
two), both for K I , and for the tearing modulus.

VII. Effect of Yield Surface Curvature

Predictions of plastic flow localization are very sensitive to the

formation of a vertex on subsequent yield surfaces (Rice, 1977). A
rounded vertex, or a relatively high local curvature of the yield surface
near the loading point, is often found in experiments, and kinematic
hardening can be used to approximately model such a rounded vertex.
Both for necking in biaxially stretched sheets (Tvergaard, 1978) and for
the onset of shear bands under plane strain conditions (Hutchinson and
Tvergaard, 1981), it has been found that predictions of kinematic
hardening are rather similar to those found for a solid that develops a
sharp vertex on the yield surface.
In order to account for the effect of increased yield surface curvature
during ductile fracture, Mear and Hutchinson (1985) have suggested a
kinematic hardening model for a porous ductile material. The model has
subsequently been extended by Tvergaard (1987a) to account for void
nucleation.

A. KINEMATIC
HARDENING
POROUS
MATERIAL

The model makes use of a family of isotropic/kinematic hardening

yield surfaces of the form @(d, a'', ~ , , f =
) 0, where a'' denotes the
center of the yield surface, and U, is the radius of the yield surface for the
matrix material. This radius is taken to be given by
OF = (1- b ) o , + bu, , (7.1)
 Material Failure by Void Growth 121

where uy and uM are the initial yield stress and the matrix flow stress,
respectively, and the parameter b is a constant in the range [0,1]. The
constitutive relations are formulated such that for b = 1 they reduce to
the isotropic hardening Gurson model, whereas a pure kinematic
hardening model appears for b = 0.
The approximate yield condition to be used here is of the form

where # I = & - &I 6e= (35ij5ji/2)l/2, and 5;’ = 6;i - Gii@’/

7 k 3. For f* =f
and q1 = 1, the expression (7.2) is that proposed by Mear and Hutchinson
(1985), which coincides with that of Gurson (1977a) if b = 1. The
parameter q1 is that also used in (3.1), and the function f*(f) is specified
by (3.2).
The plastic part of the macroscopic strain increment fi: and the
effective plastic strain increment iL for the matrix material are taken to
be related by
8’fi;= (1 - f ) ( l F & L . (7.3)
For f = O , (7.3) is an exact relationship for the classical kinematic
hardening solid, and for b = 1, the expression reduces to the equivalent
plastic work expression (3.11). Substituting the uniaxial true stress
natural strain curve (3.7) for the matrix material into (7.3) gives

Furthermore, the rate of growth of the void volume fraction is still taken
to be given by (3.4)-(3.6), with void nucleation specified by either (3.8)
or (3.9).
A fictitious Gurson yield surface aG = QG(& u M , f ) was used by
Tvergaard (1987a) to formulate the constitutive relations, where uMand f
are the current values and d& are a set of fictitious stress components
chosen such that

With this assumption, Q G = O is a direct consequence of =O. In most

cases the fictitious stresses d& will differ from the actual stresses ui’at
every point of the current yield surface (see Figure 19).
122 Viggo Tvergaard

FIG. 19. Schematic representation of the current yield surface @ = 0 for the kinematic
hardening model, and the fictitious yield surface = 0. The current stresses are d’, while
d& are fictitious stresses.

The expression for tj; in a point of the yield surface = 0 is chosen

identical to that given by the Gurson model at the point d& of the
fictitious surface (PG = 0. Thus, the plastic part of the macroscopic strain
increment is taken to be of the form (3.13). Here, the expressions for the
tensors rn; and mz and the hardening H are

(7 * 8)
Then, the instantaneous moduli to be used in the incremental stress-
strain relationship of the form (2.16) are still obtained from (3.17)-
(3.20).
The evolution equation for the yield surface centre during a plastic
increment is taken to be
$i=fi#i, fi 2 0 , (7.9)
which is a finite strain generalization of Ziegler’s (1959) hardening rule.
 Material Failure by Void Growth 123

The value of the parameter ,& is determined so that the consistency

condition, 6 = 0, is satisfied:

[
1
,& = (1 - b) - -
2 ."F (JF
2
-m:U
(JF OF
a@9
v k / -k2--Gk/U
af 3
Vkl

3a(l-f)+&--
EE,
E-E,l-f (@-+a-
U$ z)} mFlu .
vk/]
(7.10)

In cases where large rotations of the principal stress axes occur relative
to the material, the formulation of (7.9) in terms of the Jaumann rate
may give a poor representation of material behaviour, and other finite
strain generalizations using other corotational rates may be preferred
(Dafalias, 1983; Lee et al., 1983). However, in shear localization studies
the rotations of the principal stress axes prior to localization are quite
small, and for such studies, Mear and Hutchinson (1985) have found that
using the Dienes rate makes little difference.

B. LOCALIZATION
PREDI~IONS

A basic assumption in the development of the kinematic hardening

porous material model has been that in proportional stressing the
response should be identical to that for isotropic hardening. Clearly, the
instantaneous moduli remain identical in such circumstances, and there-
fore bifurcation predictions are not changed by using the kinematic
hardening model for a proportionally stressed solid. However, small
initial inhomogeneities give rise to non-proportional stressing, and then
the yield surface curvature has a strong effect on the imperfection-
sensitivity.
Mear and Hutchinson (1985) have used the M-K-type model (Figure 3)
with a larger initial porosity inside the band than outside the band to
illustrate the influence of kinematic hardening on plastic flow localization.
Similar studies have been carried out by Tvergaard (1987a) for cases
where there is no initial porosity, but where voids nucleate from
particles, which have a larger initial concentration inside the band, as
described by (4.4). Figure 20 shows results for a solid subject to uniaxial
plane strain tension with plastic strain controlled nucleation inside the
band and no nucleation outside the band f i = O . It is seen that
124 Viggo Tvergaard

0" 10" 20" 30" 40" 50" 0" 10" 20" 30" 40" WI 50"
(a) (b)
FIG.20. Shear band localization strain vs. initial band orientation for uniaxial plane
strain tension. No initial porosity, but strain controlled nucleation with YN= 0 and (a)
AfN = 0.01, (b) AfN = 0.001 (Tvergaard, 1987a).

localization occurs much earlier for kinematic hardening (b = 0) than for

isotropic hardening (b = l), particularly in the case of the very small
initial inhomogeneity AfN = 0.001. The same type of results were found
for solids under axisymmetric conditions.
The kinematic hardening material model makes it possible to study the
question discussed by various authors (for example, see Rice (1977)) that
even though failure occurs by shear localization with the final rupture
surface made up of a void sheet, it is not certain that localization was
initiated by porosity. In the case of no porosity, f = 0, the Gurson model
(b = 1) reduces to J,-flow theory, which is very resistant to localization,
whereas the localization predictions of the kinematic hardening model
(b = 0) are rather similar to those found for a solid that develops a vertex
on the yield surface (Hutchinson and Tvergaard, 1981).
Figure 21 shows localization predictions for a case in which the
inhomogeneity is represented by a slightly lower initial yield stress inside
the band than that outside, 0;= 0.99a;, while the concentration of void
nucleating particles is Constant, PN= 0.01 and AfN = 0, and nucleation
occurs at very large strains, specified by E~ = 0.9 and s = 0.1 in (3.8). For
the kinematic hardening solid, the first critical localization ( VI = 27")
occurs while there is essentially no porosity, so that here localization
occurs due to the increased yield surface curvature. When the computa-
tion is continued beyond the onset of localization, nucleation occurs
 Material Failure by Void Growth 125

1.2

&;
0.8

0.4

0
0" 10" 20" 30" 40' WI SO"

FIG. 21. Shear band localization vs. initial band orientation for uniaxial plane strain
tension. Inhomogeneity in the initial yield stress, u;= 0.99ue, while void nucleating
particles are uniformly distributed, fN = 0.01, AfN = 0, E,., = 0.9, s = 0.1, and there is no
initial porosity (Tvergaard, 1987a).

inside the shear band as the localized strains grow large, and final failure
occurs by the void sheet mechanism. Thus, for the kinematic hardening
solid, there is a competition between two effects on localization. If voids
nucleate late, vertex-type effects tend to dominate the onset of localiza-
tion, while porosity has a strong influence if the voids nucleate earlier;
but in both cases final failure occurs in a void sheet. On the other hand,
for the isotropic hardening material a significant amount of porosity has
to appear before the localization predicted in Figure 21.
The results in Figures 20 and 21 correspond to a homogeneously
stressed solid, for which localization can be analysed by the simple model
analysis (4.1)-(4.3). Shear band development in a solid subject to a
non-uniform state of deformation is a more complex problem, which
requires a full numerical analysis. Such numerical studies involving
localization in strongly non-uniform strain fields have been discussed
above, for example, see Figures 11 and 16. The strain field at the
stretched surface of a bent specimen represents a less extreme non-
uniformity, and this case has been used by Tvergaard (1987b) to study
the basic influence of nonhomogeneous deformation on the development
of localized shearing and shear failure.
An imperfection in the form of an initial surface waviness is assumed,
and periodic solutions are considered analogous with those illustrated in
Figure 14. The surface waviness has little effect initially but gives a
126 Viggo Tvergaard

1.0 1 1

0.8 -

0.6 -

0.4 .

0.2 - b=O
b I 0.5
_ _ _ . b_
.1 _ _ _

0
0 0.2 0.4 0.6 0.8 E0 0

FIG.22. Maximum principal logarithmic strains at two material points near the surface
of a bent specimen. The material has strain-controlled nucleation and no initial porosity
(Tvergaard, 1987b).

slightly nonuniform porosity distribution. Shear localization starts to

develop from the wave bottoms, leading finally to a shear crack that
grows into the material from the surface. For a kinematic hardening
material with no initial voids and strain-controlled nucleation specified by
fN = 0.04, E~ = 0.3, and s = 0.1 in (3.8), Figure 22 shows the correspond-
ing development of the maximum principal logarithmic strain at two
material points vs. an average surface strain measure cO, compared with
isotropic hardening predictions. As expected, localization occurs earlier
for the kinematic hardening model, but in all cases localization is
significantly delayed relative to the first loss of ellipticity near the surface.
The stability of a periodic pattern of shear cracks, such as that implied
by the results in Figure 22, has been investigated by a few computations.
Figure 23 illustrates the results of one of these computations. At the first
stage in Figure 23, two shear bands are developing nearly symmetrically
from the central wave-bottom on the surface, whereas at the two
following stages one of the bands has stopped growing while the other
band develops into void sheet fracture (painted black in Figure 23). This
 Material Failure by Void Growth 127

0.ozy

Y
f = 0,001 f~O001
A A

(bl
FIG.23. Contours of constant void volume fraction calculated for a surface region of
double length 21, in a bent specimen. The material has strain-controlled nucleation with no
initial porosity. (a) co= 0.323, (b) E, = 0.340, (c) c0 = 0.354 (Tvergaard, 1987b).

instability of the periodic growth pattern can explain a characteristic

spacing between the shear cracks that is observed on the surface of a bent
plate (for example, see Figure 10 in Hutchinson and Tvergaard, 1980).
This behaviour is analogous to the instability of a system of straight edge
cracks in a brittle solid found by Nemat-Nasser et al. (1980).
The kinematic/isotropic hardening model of a porous ductile solid has
also been used to study the interaction of two size-scales of voids, where
the fracture mechanism involves the formation of a void sheet between
two larger voids. Both for a planar array of cylindrical large-scale
inclusions (Tvergaard, 1987c) and for a full 3D array of spherical
inclusions (Tvergaard, 1987d), it is found that an increased yield surface
curvature gives rise to earlier final failure.

W I . Strain Rate Sensitive Material

The strain rate sensitivity of plastic yielding is often introduced in

material models by representing the inelastic part of the deformations in
terms of a nonlinear viscous behaviour. Such a rate sensitive version of
128 Viggo Tvergaard

the Gurson model has been used by Pan, Saje, and Needleman (1983)
and by Needleman and Tvergaard (1984~)to study the influence of the
strain rate on plastic flow localization, and it has been found that
localization is significantly delayed by the material rate sensitivity.
Subsequently, Tvergaard and Needleman (1986, 1987) have incorporated
a simple description of cleavage fracture in this material model and have
used the model to study the brittle-ductile transition in the Charpy
V-notch test.

A. VISCOPLASTIC
MATERIAL
MODEL

The strain rate sensitivity is included in the porous ductile material

model by taking the response of the matrix material to be elastic-
viscoplastic. Thus, the microscopic effective plastic strain rate iL is here
taken to be given by the power law relation

Fg$)-1
where rn is the strain rate hardening exponent and i.0 is a reference strain
rate. It is noted that in (8.1) and subsequently in this section (') denotes
the time derivative, since the material model is time-dependent.
The function g(&) in (8.1) represents the effective tensile flow stress
in the matrix material in a tensile test carried out at a strain rate such that
iP - i n . Thus, for a power hardening material with uniaxial stress strain

behaviour as that given by (3.21), the function g(.&) can be found from
the relationship

The concept of elastic unloading is not directly incorporated in the

present elastic-viscoplastic material model. Consequently, there is no
yield surface; but the function @(a", aM,f) = 0 given by (3.1) is used as a
plastic potential, so that the plastic part of the macroscopic strain rate is
given by the expression

The equivalent plastic work expression (3.11) is still assumed valid. Then,
 Material Failure by Void Growth 129

with iL specified by (8.1), the value of the parameter A is directly given

by

As in the time-independent material model, the total macroscopic

strain rate is taken to be the sum of the elastic and plastic parts,
rji, = rj: + 4.: Thus, using the elastic, incremental stress-strain relation-
ship, the constitutive relations for the elastic-viscoplastic material can be
written in the rate form (2.12).
For the viscoplastic material, the microscopic effective plastic strain
rate iL is not proportional to U M , but is given directly by (8.1).
Therefore, in the expression (3.4) for the rate of increase of the void
volume fraction, the contribution due to void nucleation is here taken to
be given by
@)nucleation = B(Uh4 + (d3-/3) + SiL 9 (8.5)
and plastic strain controlled nucleation is specified by

Since @ = 0 is used as the potential function, the consistency condition

6 = 0 must be satisfied. This equation together with (3.4), ( 3 4 , and
(8.5) determines the values off and U M .
In an incremental numerical solution for this elastic-viscoplastic porous
material, the stable step size can be significantly increased by using a
forward gradient method proposed by Peirce et al. (1984). At time t, the
microscopic effective strain rate iL to be used in (8.4) is expressed by a
+
linear interpolation between the rates at time t and t At, respectively,
i L = (1 + 8 ) Q ) + OiP(r+Ar)
M ? (8.7)
and a Taylor series expansion is employed to estimate the value i r + A r )
of the rate at time t + At. Then the constitutive relations based on (2.12),
(8.3), and (8.4) can be rewritten in the form (2.17), where analogous to
(3.17) the tangent moduli are given by
L$kI =yJk1- p,~i&kl
(8 * 8)
with the value p* given by
8 At
g 1 - QIe At
Q2 (8.9)
130 Viggo Tvergaard

and the initial stress rate type term given by

(8.10)

The functions Q l , Q 2 , and Q3 are rather lengthy expressions that will

not be repeated here. It is noted that 8 = 0 (no forward gradient method)
gives p* = 0, so that the tangent moduli (8.8) reduce to their elastic part.
A viscoplastic, kinematic hardening model of a porous ductile material
has been introduced by Becker and Needleman (1985). This model,
based on the time-independent model of Mear and Hutchinson (1985),
has been used to study the effect of material strain rate sensitivity in
plane strain and round tensile test specimens.
Failure by cleavage is included in the material model in a relatively
simple manner, by assuming that failure occurs if the maximum principal
tensile stress exceeds a critical value a,. Several investigations of
cleavage fracture in body-centered-cubic (b.c.c.) metals have shown that
such a constant critical stress is a realistic criterion for slip induced
cleavage failure in the low temperature range (Cottrell, 1958; Petch,
1958; Smith, 1968; Hahn, 1984). At somewhat higher temperatures, near
the transition to fibrous fracture, cleavage micro-cracks stretching over
one grain each are not large enough to be immediately unstable by the
Griffith criterion, and thus the final link-up of micro-cracks requires some
further plastic deformations. However, it has been assumed by Tvergaard
and Needleman (1986, 1987) that a constant critical value a, is a
sufficiently good criterion in the whole temperature range considered.
The effect of the elastic-viscoplastic material model is illustrated in
Figure 24 by a few results for a homogeneously strained material element
under plane strain tension (e3=0), subject to a fixed prescribed
logarithmic strain rate i l .The material considered has the rate hardening
exponent m = 0.01, the strain hardening exponent n = 10, and plastic
strain controlled nucleation (8.6). It is seen that the peak value of the
maximum principal tensile stress a1is much higher for a2/al= 0.6 (where
the stress triaxiality is high, ak/(3ae) = 2.2) than for u2= 0. Furthermore,
in both cases the peak value of o1is increased by about 15%, when the
strain rate is increased by a factor lo6 (note loo." = 1.148). Thus,
cleavage micro-cracks are most likely to nucleate in regions of high
triaxiality, and more likely at high strain rates. For example, if the critical
stress for cleavage is a, = 4.3a0, failure at the low strain rate occurs by
void coalescence, both for az/al=0.6 and for a 2 = 0 ; but at the high
 Material Failure by Void Growth 131

0.L 0.8 E,
(01

f , 1

0.1 -

I
I bl
FIG. 24. Effect of stress triaxiality and strain rate in plane strain tension. Material with
n = 10, m = 0.01, strain controlled nucleation and no initial porosity. (a) Maximum
principal stress. (b) Void volume fraction (Tvergaard and Needleman, 1986).

strain rate fracture occurs by cleavage in the high triaxiality case, while
the void volume fraction is still very low.

B. THE CHARPY
V-NOTCHTEST
The energy absorbed to fracture in the Charpy V-notch test is widely
used to evaluate the fracture resistance of materials. For b.b.c. metals,
fracture occurs by cleavage when the temperature is sufficiently low,
whereas at higher temperatures fracture occurs by ductile hole growth
and the absorbed energy is much larger. A similar transition can take
place at a fixed temperature, from cleavage fracture at high strain rates to
fibrous fracture at low strain rates.
132 Viggo Tvergaard

FIG.25. Contours of constant void volume fraction and maximum principal stress in a
Charpy V-notch test with U / B b , = lo3, at U / B = 0.125 (Tvergaard and Needleman, 1986).

The competition between the two failure mechanisms in the Charpy

V-notch test at a fixed temperature has been analysed numerically by
Tvergaard and Needleman (1986). For a material with m = 0.01, n = 10,
and the critical stress for cleavage a, = 3.35a0, while nucleation is plastic
strain-controlled, fN = 0.04, eN = 0.3, s = 0.1, calculations have been
carried out at three different rates of loading. Figure 25 shows distribu-
tions of the void volume fraction f and the maximum principal stress u,,,
in a case with U/BSo= lo3, where B is the specimen thickness and U is
the rate of displacement of the central point at which the hammer hits the
specimen. At the stage illustrated in Figure 25, large plastic strains at the
notch tip have given so much voidage that ductile failure is about to
initiate, but simultaneously the critical stress a, has been reached in the
material somewhat below the notch, and after this stage cleavage fracture
takes over completely. For U / B $ = lo5, cleavage occurs earlier and the
absorbed energy is significantly lower, while for U/BSo= 0.1 the ductile
failure mechanism is more dominant, in agreement with experimental
observations (see Rolfe and Barsom (1977)).
In a subsequent investigation (Tvergaard and Needleman, 1987), the
computational model has been used to directly analyse the temperature
dependence of the absorbed energy for a particular material: the
high-nitrogen mild steel investigated by Ritchie, Knott, and Rice (1973).
This tempeature dependence is generally a result of the variation of
material parameters with temperature, in particular the decreasing initial
yield stress for increasing temperature. Uniaxial tensile test results for the
 Material Failure by Void Growth 133

high-nitrogen mild steel show that in addition to the decreasing yield

stress, the degree of strain hardening increases with temperature, and it
turns out that also this temperature dependence has a strong effect on the
failure mode transition.
Curves of absorbed energies vs. temperature have been computed for
two different imposed velocities, U = 5 x m/sec and U = 5 m/sec,
corresponding to slow bending and impact loading, respectively. These
curves show a brittle/ductile transition in absorbed energy at T = -85°C
for the slow loading and at a somewhat higher temperature, T = -5O"C,
for impact loading. The transition in the fracture initiation mode from
cleavage initiation to ductile void growth initiation is predicted at
somewhat higher temperatures for both curves. In the range of tempera-
tures below T = 45"C, the general variation of the absorbed energies with
temperature and rate of deformation is found to be rather typical of
many steels. However, a sharp reduction in absorbed energy is found at
higher temperatures, which is untypical of most steels. This behaviour
results from an unusually strong increase in strain hardening, which is
characterized as "blue brittleness" for the particular material considered
here.
The computations discussed so far are quasi-static, neglecting the effect
of inertia, so that time dependence enters only through the viscoplastic
material model. A transient analysis that does not account for material
failure (Norris, 1979) indicates rather little effect of inertia at the loading
rates typically encountered in the Charpy impact test. To check this in
the presence of material failure Tvergaard and Needleman (1987) have
repeated a few of the quasi-static computations as transient analyses,
taking full account of material inertia. Figure 26 shows the dynamic
force-displacement curve for U = 5 m/sec at T = -55°C compared with
the response found by the quasistatic analysis. Plasticity damps the
bending mode oscillations rather quickly, and the absorbed energy found
by the transient analysis is only slightly higher than that found by
neglecting the effect of inertia. It might be expected that wave effects
have a significant influence on cleavage initiation, but in fact the
magnitude of the maximum principal stress oscillations are very small
once there is extensive plasticity.
Application of the elastic-viscoplastic material model is necessary in
studies of the brittle/ductile transition, since the difference between the
behaviour at slow loading and impact loading is entirely dependent on
material strain-rate sensitivity. It is noted that also the investigations of
 Viggo Tvergaard

10.0-
1

.o
Displacement (Meters) *1o - ~
FIG. 26. Computed force-displacement curves for Charpy test of a high-nitrogen mild
steel at U = 5 m/sec, T = -55°C. Full transient analysis compared with quasi-static result
(Tvergaard and Needleman, 1987).

crack growth by Needleman and Tvergaard (1987) and Becker et af.

(1987b) have been based on the viscoplastic version of the porous
material model, although only one rate of loading was considered.
Material strain-rate sensitivity does have an influence in these cases, due
to the non-uniform distribution of strain rates in the fields near the crack
tip; but here the predictions are not expected to differ much from those
of the time-independent material model.

M.Creep Failure by Grain Boundary Cavitation

In a polycrystalline metal subject to creep at elevated temperatures,

the nucleation and growth of voids to coalescence also plays a major role
in the failure process. In these circumstances, the voids occur primarily at
the grain boundaries and the growth mechanism differs significantly from
that at ductile fracture, discussed in the previous sections. Here, in
addition to dislocation creep of the grains, grain boundary diffusion plays
a significant role.
Experimental results show that high temperature cavitation occurs
 Material Failure by Void Growth 135

mainly on grain boundary facets normal to the maximum principal tensile

stress direction (Hull and Rimmer, 1959; Cocks and Ashby, 1982; Argon,
1982). Coalescence of these cavities leads to micro-cracks, and the final
intergranular creep fracture occurs as the micro-cracks link up. In cases
where diffusion gives the dominant contribution to the growth of cavities,
the rate of growth is often constrained by the rate of dislocation creep of
the surrounding material, as has been noted by Dyson (1976). Several
aspects of creep failure have been treated recently by Riedel (1986) and
Cocks and Leckie (1987).
A set of constitutive relations for creep with grain boundary cavitation
has been proposed by Tvergaard (1984b), as an extension of work by
Rice (1981) and Hutchinson (1983). For the rate of growth of a single
cavity both the contributions of diffusion and dislocation creep are
accounted for, and furthermore, the model incorporates the creep
constraint on the rate of cavity growth.

A. CREEPWITH GRAINBOUNDARY
CAVITYGROWTH

The constitutive relations are based on the assumption that microscopic

cavities nucleate and grow on a certain number of grain boundary facets,
while the grains deform by dislocation creep, represented as power law
creep (Tvergaard, 1984b). The model makes use of the idea of Rice
(1981) that the cavitating grain boundary facets can be represented as
penny-shaped cracks. Also an expression derived by Hutchinson (1983)
for the rate of creep of a material containing traction free micro-cracks is
applied.
The cavities are assumed to be uniformly distributed on each grain
boundary facet, with average spacing 26, as indicated in Figure 27. Then,
the average separation between two grains adjacent to a cavitating facet
is S = V/nb2, and the rate of growth of the average separation is

where V, V, and b are the cavity volume, the rate of growth of this
volume, and the rate of change of the cavity half-spacing, respectively
(see Figure 27b).
When the cavitating facets are modelled as penny-shaped cracks, the
average rate of separation 6 of the adjacent grains is also given as the
136 Viggo Tvergaard

I - - - - <
b

b-
\ I /
>---4
/ \
\
\ \
)---4 )--

FIG.27. (a) Spherical-caps shape of a single cavity. (b) Equally spaced cavities on a
grain boundary. (c) An isolated, cavitated grain boundary facet in a polycrystalline material.

average rate of opening of a crack. This average rate of opening has been
determined by He and Hutchinson (1981), and their expression, modified
to account for a non-zero normal tensile stress a, on the crack surfaces
(Tvergaard, 1984a), gives

s=p-
. S-a,
iF2R.
0,

Here, R is the current radius of the crack, and S = djfiifij represents the
value of the normal stress on the facet in the absence of cavitation, where
dj is the macroscopic Cauchy stress tensor and fit is the facet normal in
the current configuration. The macroscopic effective Mises stress cor-
responding to aij is a, (as in (3.1)), and the effective creep strain rate for
an uncracked material is taken to be i: = S,(a,/a,)”, where a, is a
reference stress quantity, n is the creep exponent, and ST is a
temperature-dependent reference strain rate.
 Material Failure by Void Growth 137

For this power law creeping material, the value of the parameter p is
given by the asymptotic expression (He and Hutchinson, 1981)
4 3 -112

B=;(l+--) ) (9.3)

which is highly accurate for all n as long as IS/a,l 5 2, while for higher
triaxialities the rate of opening of the micro-cracks is increasingly
underestimated by using (9.3) in (9.2).
The two expressions (9.1) and (9.2) for the rate of separation 6 must
be identical. This requirement determines the value of the normal stress
a,,on the facet and the cavity growth rate V , which is generally a function
of an.
Hutchinson (1983) has derived an expression for the macroscopic creep
strain rates in a material containing a certain density of penny-shaped
micro-cracks. The expression is obtained for traction-free micro-cracks,
partly based on the results of He and Hutchinson (1981) for a single
crack, and can also be modified to account for a non-zero traction a,,on
the crack surfaces (Tvergaard, 1984b). This modified expression for the
macroscopic creep strain-rates is

2n+la, n+l a, mij]}, (9.4)

where mij = f i i f i j , so that S = d m j jand p reflects the density of cavitating

facets. In terms of the crack radius R and the number of cracks per unit
volume A, Hutchinson (1983) found, using (9.3),

+ +
p = 4R3A(n 1)( 1 n ->3 -112
. (9.5)

The total strain-rate is taken to be the sum of the elastic part, the creep
part, and the thermal expansion part, fiij = 4: + +
fig fi;. Then, analo-
gous to (2.12) the macroscopic stress strain relationship can be written as

A forward gradient method has been used by Tvergaard (1984b) to

increase the stable step size in a numerical solution based on this
visco-plastic material model. The derivations follow along the lines
indicated in connection with (8.7) and result in constitutive relations of
the general form (2.17)) although the expressions differ from those given
by (8.8)-(8.10).
It is noted that equations (9.1) to (9.6) are independent of the
138 Viggo Tvergaard

particular mechanisms for cavity growth or nucleation to be considered.

These mechanisms will result in expressions for V and b to be substituted
in (9.1), and thus the material model is valid in cases where the cavities
grow in the spherical-caps shape as well as cases where crack-like cavity
shapes are favoured. The model has also been extended to account for
the effect of grain boundary sliding (Tvergaard, 1984b, 1985), and to
represent different nucleation or growth behaviour at different groups of
grain boundary facets; but these extensions will not be discussed here.
The particular void shape considered by Tvergaard (1984b) is the
spherical-caps shape (Figure 27a), which is the relevant shape in cases
where the diffusion along the void surface is sufficiently rapid, relative to
the diffusion along the grain boundary, to maintain this quasi-equilibrium
state. Detailed micromechanical studies of the rate of growth of such
cavities, by combined grain boundary diffusion and dislocation creep,
have been carried out numerically by Needleman and Rice (1980) and
Sham and Needleman (1983). At sufficiently low tensile stresses diffusive
cavity growth is completely dominant, while at somewhat higher stresses
dislocation creep also contributes noticeably to the growth rate.
Using an interpolation between the rigid grains model for purely
diffusional growth and approximate results (Budiansky et al., 1982) for
the growth of a spherical void in a power law creeping material, it has
been found (Sham and Needleman, 1983; Tvergaard, 1984a) that over a
wide range of stress states, the numerical results for the growth of a
single void are well approximated by the expression

where

Here, a is the cavity radius (Figure 27a), 26 is the average cavity spacing,
and 9 = &SBS2/kT is the grain boundary diffusion parameter, where
DsSB is the boundary diffusivity, SZ is the atomic volume, k is
Boltzmann’s constant, and T is the absolute temperature. Furthermore,
 Material Failure by Void Growth 139

a,,, a,,,, and a, are the average normal stress, mean stress, and Mises
stress, respectively, in the vicinity of the void, and the constants are
given by a,,= 3/(2n), /3" = (n - l)(n + 0.4319)/n2, and h ( q ) = [(1+
cos q)-' - 4 cos q]/sin q. A relevant value of the angleJ! , I (see Figure
27a) is q-7O0, and the sintering stress a, is often approximated by
a, = 0. When V is known, the rate of growth of the cavity radius is given
by a = V/[4na2h(q)].
Cavity coalescence on grain boundary facets requires a / b = 1 in cases
where creep constrained cavitation results in very small normal stresses
a,,. But, in other cases, where a,, is higher, failure of the remaining
ligaments between the cavities, by ductile tearing or by cleavage, may
occur at somewhat smaller values of a / b (Cocks and Ashby, 1982). Such
coalescence on grain boundary facets, leading to open micro-cracks, is
often used as a failure criterion; but is is noted that final fracture actually
occurs somewhat later, as these micro-cracks link up.
The parameter L in (9.7), defined by
- C 113
L=(gae/~e) (9.10)
was used by Needleman and Rice (1980) as a stress and temperature
dependent length scale. For a/L<0.1, the cavity growth rate is
completely dominated by diffusion, so that V2 is negligible relative to Vl ;
but dislocation creep growth plays an increasing role for higher values of
a/L, so that here V2 adds significantly to in (9.7).
Continuous nucleation of new cavities during the creep process is
observed in many metals (Argon, 1982; Dyson, 1983) and experiments
often indicate that the number of cavities is proportional to the effective
strain. An approximate way of incorporating such continuous nucleation
in the present material model has been discussed in some detail by
Tvergaard (1984b). With these approximations, the ratio b / b , to be
substituted into (9.1), is simply taken to be

(9.11)

where AI and AIl are the principal stretches on the plane of the facet and
N is the number of cavities per unit initial area of the facet. The value
dN/dee is taken to be constant in some cases, while in other cases this
value is taken to be stress dependent. It is assumed that no nucleation
occurs if N has reached a saturation value N,,, or if a, <0.
The temperature dependence of the material model is incorporated
140 Viggo Tvergaard

through the dependence of the effective creep strain rate i," and the grain
boundary diffusion parameter 9 on the absolute temperature T
(Needleman and Rice, 1980; Cocks and Ashby, 1982)
:i = io(2)fl$
exp[ Qv 1 - 1 (x F)], (9.12)

9 = 9 0 - To QB
-1- -1 .
exp
T [ W (To
-
T)]
(9.13)

Here, Bo , go,a,, and To are reference values, Qv and QB are activation

energies, and 3 is the gas constant.

B. CREEPRUPTUREIN STRUCTURAL
COMPONENTS

The material model has been used to analyse the creep failure of
thick-walled furnace tubes (Tvergaard, 1984b). The initial internal and
external radii are denoted by A. and B o , respectively, and the tube is
subjected to an internal pressure p and heated from the outside, so that
the absolute temperatures on the inside and outside are TA and TB with
the radial temperature distribution corresponding to steady heat
conduction.
Neglecting temperature variations along the tube, the solutions are
uniform in the axial direction, with a constant axial strain-rate deter-
mined so that the resultant axial force in the tube balances the axial
component of the internal pressure. Thus, the fields vary only in the
radial direction (the x2-direction), and a one-dimensional finite element
approximation is applied in (2.11).
The material parameters employed do not refer to a particular catalyst
tube steel, but are generally consistent with the orders of magnitude of
the parameters tabulated by Frost and Ashby (1982). The values used are
E = 10" N/m2, v=0.3, n = 5,
a, = 30 X lo6N/m2, Eo = h-', To= 1173 K,
QB
& = 6 x 104K, - = 3.5 X lo4K, go= m5/Nh,
94 94
and the thermal expansion coefficient is a! = 16.8 x K-'. The initial
dimensions of the tube are taken to be A. = 0.045 m and Bo = 0.055 m,
and the grain size is taken to be such that R , = 50 pm is representative of
 Material Failure by Void Growth 141

the initial radius of the grain boundary facets. Furthermore, in the case
illustrated in Figure 28, continuous nucleation is assumed with d N / d & ,=
104/AI and N m , = 4 0 0 / A I , where A I = n R : is the initial area of a
cavitating grain boundary facet. Very few cavities are assumed present
from the beginning, as specified by a J b 1= 0.005 and b I / R I= 1 , with the
density of cavitating facets p=O.5. A reference time is defined by
tR = a,/(Ei:), where ue is the Mises stress obtained from the average
axial and circumferential stresses, and :i is the corresponding effective
creep strain rate at temperature To.
At time t = 0, the tube is subjected to an internal pressure p =
3.5 x lo6 N/m2 and a temperature gradient specified by TA= To- 20 K,
Ts = + 20 K. Figure 28 shows that the thermal expansion gives rise to
significant stress gradients, with compressive hoop stresses S near the
outside of the tube wall, where the temperatures are highest. These
elastic stress peaks are relaxed by creep in the initial stages, until the

1
6,E

0.8

0.6

0.L

0.2
7
-7 4 0.61

0.L-

66.L 0.050
0
0.5 2 1 "0 0.5 2 1
x -A x -A
0 0
q& Bo- A.
(bl (C)
FIG. 28. Development of stresses and damage in a pressurized thick-walled tube with
non-uniform temperature. The material has continuous nucleation. (a) Macroscopic hoop
stress. (b) Normal stress on cavitating facets. (c) Cavity growth (Tvergaard, 1984b).
142 Viggo Tvergaard

steady state stress distribution is reached at t / t R= 5.75. Initially, a,/S = 1

(see Figure 28b), but the ratio a,/S decays rapidly towards zero, so that
creep constrained cavity growth takes place during most of the lifetime.
For tubes subject to uniform temperature, failure tends to occur first at
the inside, where the strains are larger; but here the higher temperature
at the outside results in nearly simultaneous failure at the inside and
outside (Figure 28c).
Creep crack growth has also been analysed by application of this
material model (Tvergaard, 1986). The particular configuration analysed
is a center cracked panel subject to plane strain conditions, with the
initial geometry specified by A o / B o = & where 2Ao is the initial crack
length and 2Bo is the initial specimen width. In contrast to the ductile
porous material description (Section 11), the present creep failure model
requires knowledge of the maximum principal stress directions, since the
orientation of the facets, on which cavitation is considered, has to be
chosen a priori. This is no problem in the thickwalled tube under internal
pressure, where the principal stress directions remain fixed, but during
crack growth the maximum principal stress directions rotate significantly.
The field of angles wI used (Tvergaard, 1986) to define the reference
normals of the cavitating facets has been computed as the maximum
principal stress directions in a steady-state creep solution with no damage
( p = 0). It has been checked in the computations that the angle between
the current normal of a cavitating facet and the current maximum
principal stress direction remains small in all material points where failure
is going to occur subsequently.
Figures 29 and 30 are results obtained in a computation where all
cavities are assumed to be present from the beginning, specified by
a I / b I= 0.1, b J R I = 0.1, and p = 0.2, while other material parameters are
n = 5 and Y = 0.3. The constant external load is specified by the initial
nominal stress on the crack ligament, uN= 0.0005E, and the value of the
grain boundary diffusion parameter is given by a I / L N= 0.01, where LN
appears by taking a, equal to aN in (9.10). In the figures, time t is
normalized by the failure time tP corresponding to uniaxial plane strain
tension at stress level a,.
Initially, at t = 0 when the load is applied, the elastic singular stress
field dominates near the crack tip; but already at t/tP = most of the
initial elastic stress peak has been relaxed by creep, and soon after this
stage the crack starts to grow. The creep damage in the material near the
crack tip, measured by alb, is shown in Figure 29 at two stages. To
 Material Failure by Void Growth 143

a / b = 0.9

1.0

A rlAo L ' rlA

0 0 01 0 02 0 0 01 0.02- 0

(0) (b)

FIG.29. Distribution of creep damage near the tip of a crack with initial length %lo.(a)
(b) t / t : = 0.0810 (Tvergaard, 1986).
t/tF = 0.0042,

bA/AO

0.2

0.1

0
0 0.1 0.2 t l t; 0.3
FIG. 30. Creep crack extension vs. time for crack with initial length 2Ao (Tvergaard,
1986).
144 Viggo Tvergaard

illustrate the growth of the macroscopic crack, elements in which failure

has occurred ( a l b = 1) are painted black, and it is seen that initially the
crack grows straight ahead ( 6 = 0"). When the crack extension AA is
about 0.014A0, the growth direction starts to deviate from the crack
plane, and all subsequent crack growth occurs along a plane inclined
about 40" to the initial crack plane.
The rate of crack growth is illustrated in Figure 30 by AA vs. t curves
computed for two different near-tip meshes, where AA denotes the
distance from the initial crack tip to the current crack tip in the reference
configuration. Crack growth starts nearly immediately when the load is
applied, with a relatively high growth rate that gradually decays as the
elastic stress peak is relaxed by creep. The growth rate increases again
when the growth direction starts to deviate from the crack plane, and
subsequently the growth keeps accelerating as the remaining ligament is
reduced by crack growth. Similar results for other sets of material
parameters are also presented by Tvergaard (1986).
As an extension of this anaiysis, Tvergaard (1988) has investigated the
effect of creep acceleration induced by degradation of the microstructure.
These crack growth analyses are carried out for both center-cracked and
double edge-cracked specimens, and it is found that creep acceleration
reduces the crack growth rate due to a lower stress level at the tip.
The same material model has been used recently by Li, Needleman,
and Shih (1988) to study creep crack growth under small scale yielding
conditions. In this investigation, the effect of material parameters and
load level on the dominant asymptotic stress and strain fields near the tip
of the growing crack has been studied.

X. Discussion

A characteristic feature of the material models discussed here is that

the damage parameter used is directly given by the material geometry on
the micro-scale, either as a void volume fraction or as an area fraction of
voids on grain boundary facets. This means that the current value of the
damage parameter can be directly determined by the use of microscopy.
Furthermore, this is the basis of the micromechanical studies that have
been used extensively to improve the understanding of material be-
haviour. By contrast, the approach often characterized as continuum
 Material Failure by Void Growth 145

damage mechanics makes use of damage parameters relating to macro-

scopic material behaviour, with no direct relation to the cause of damage
on the micro-level.
Although the void volume fraction can be measured, a given value of
this scalar damage parameter can represent many different distributions
and shapes of voids. Magnusen et al. (1987) have investigated the effect
of the void distribution by carrying out a series of tensile tests for
specimens with either random or regular arrays of equi-sized, cylindrical,
through-thickness holes. The volume fraction of holes is the same in all
cases, but different degrees of randomness are obtained by specifying
different values of a minimum accepted spacing between adjacent holes.
For each of steel, brass, and aluminium, it was found that the fracture
strain is smaller the more randomness is allowed. The regular square
array of holes that has often been used in micromechanical studies gives
the maximum fracture strain for the given volume fraction of holes. Thus,
the distribution of voids should certainly affect the material model, and
this needs to be investigated further. It may be expected, based on the
experiments, that the distribution will have a strong effect on the value of
the critical void volume fraction fc in (3.2), at which failure starts to
occur rapidly, while little influence is expected at smaller void volume
fractions.
Some materials contain long thin inclusions or rows of inclusions,
which may be essentially parallel, dependent on previous metal working
processes. For such materials, the influence of the voids on the stress
strain relationship is anisotropic, both in the nucleation and growth
phases, so that the isotropic material models discussed in the present
paper are not adequate. Some work has been done to find approximate
anisotropic yield conditions based on experimental results for various
patterns of perforations in sheets, and there has also been work
introducing a second order damage tensor Ji for the description of ductile
void growth. However, a full ductile porous material model that
accounts for anisotropies due to void distributions or shapes has not yet
been developed.
In spite of the problems raised above, it should be emphasized that in
many cases a reasonable description is obtained by the isotropic material
model, in which the void volume fraction is the only damage parameter.
This has been strongly supported by the detailed comparison of theoreti-
cal predictions and experimental results for sintered iron powder com-
pacts with known initial porosities, subjected to various non-uniform
146 Viggo Tvergaard

stress states (Becker et al., 1987a). The comparison with these experi-
ments supports the description of void growth and final failure incorpor-
ated in the material model. Nucleation models are not checked by these
experiments, but will have to be fitted with the actual material behaviour
for each particular application.
The size of the voids and the void spacing does not at all enter the
ductile material model, where only the void volume fraction is known,
and the same would be true if the void shapes and distribution patterns
were also known. The only necessary assumption is that the voids are
large enough relative to the dislocation cell size, so that the application of
continuum plasticity is reasonable. This is different for the material
subject to creep with grain boundary cavitation. Here, the grain size
provides a natural length scale, and the material model requires
knowledge of the number of cavities per grain boundary facet, so that in
fact both the size and spacing of the voids need to be known.
In the porous ductile material models, the same damage parameter,
the void volume fraction, applies to all possible stress states. This is more
complicated in the model for creep with grain boundary cavitation, since
here void growth occurs mainly on grain boundaries normal to the
maximum principal tensile stress. Thus, with only one damage parameter
(the area fraction of cavities on grain boundary facets), the maximum
principal stress direction must be known in each material point, and it
must be assumed that this field of directions remains approximately
constant throughout the material lifetime. In the more general case,
where the principal stress directions rotate during the lifetime, it is
necessary to use a number of damage parameters, where each parameter
has a direction associated with it (the facet normal). These damage
parameters represent area fractions of cavities on different groups of
grain boundary facets with different orientations of the normals.
Both the time-independent plasticity models in Sections 111 and VII
and the viscous material models in Sections VIII and IX have been
employed to analyse the gradually increasing damage and the mode of
final failure in several test specimens or structural components. The
predictions have been found to be at least qualitatively, and in some
cases also quantitatively, in good agreement with experimental observa-
tions. Thus, although there is a need for further developments to make
the descriptions more realistic in some respects, the material models in
their current form appear to be quite powerful.
 Material Failure by Void Growth 147

References

Anderson, H. (1977). Analysis of a model for void growth and coalescence ahead of a
moving crack tip. 1. Mech. Phys. Solids 25, 217-233.
Aoki, S., Kishimoto, K., Takeya, A., and Sakata, M. (1984). Effects of microvoids on
crack blunting and initiation in ductile materials. Int. 1. Fracture 24, 267-278.
Aravas, N., and McMeeking, R. M. (1985). Microvoid growth and failure in the ligament
between a hole and a blunt crack. Int. J . Fracture 29, 21-38.
Argon, A. S. (1982). Mechanisms and mechanics of fracture in creeping alloys. In “Recent
Advances in Creep and Fracture of Engineering Materials and Structures” (B. Wilshire
and D. R. J. Owen eds.), 1-52, Pinerage Press, Swansea, U.K.
Becker, R., and Needleman, A. (1986). Effect of yield surface curvature on necking and
failure in porous plastic solids. J. Appl. Mech. 108,491-499.
Becker, R., Needleman, A., Richmond, O., and Tvergaard, V. (1987a). Void growth and
failure in notched bars. Brown University, Division of Engineering.
Becker, R., Needleman, A., Suresh, S., Tvergaard, V., and Vasudevan, A. K. (1987b). An
analysis of ductile failure by grain boundary void growth. Brown University, Division
of Engineering.
Berg, C. A. (1970). Plastic dilatation and void interaction. In “Inelastic Behaviour of
Solids” (M. F. Kanninen et al. eds.), pp. 171-210. McGraw-Hill.
Bishop, J. F. W., and Hill, R. (1951). A theory of the plastic distortion of a polycrystalline
aggregate under combined stresses. Phil. Mag. 42, 414-427.
Brown, L. M., and Embury, J. D. (1973). The initiation and growth of voids at second
phase particles. I n “Proc. 3rd Int. Conf. on Strength of Metals and Alloys,” pp.
164-169. Inst. of Metals, London.
Budiansky, B. (1969). Remarks on theories of solid and structural mechanics. In “Problems
of Hydrodynamics and Continuum Mechanics” (M. A. Lavrent’ev et al. eds.). SIAM,
Philadelphia.
Budiansky, B., Hutchinson, J. W., and Slutsky, S. (1982). Void growth and collapse in
viscous solids. In “Mechanics of Solids, The Rodney Hill 60th Anniversary Volume”
(H. G. Hopkins and M. J. Sewell eds.), pp. 13-45. Pergamon Press.
Chu, C. C., and Needleman, A. (1980). Void nucleation effects in biaxially stretched
sheets. J. Eng. Materials Technol. 102, 249-256.
Cialone, H., and Asaro, R. J. (1980). Hydrogen assisted fracture of spheroidized plain
carbon steels. Met. Trans. U A , 1373-1387.
Cocks, A. C. F., and Ashby, M. F. (1982). On creep fracture by void growth. Prog.
Materiuls Science 27, 189-244.
Cocks, A. C. F., and Leckie, F. A. (1987). Creep constitutive equations for damaged
materials. A d v . Appl. Mech. 25, 239-294.
Cottrell, A. H. (1958). Theory of brittle fracture in steel and similar metals. Transactions
AIME 212, 192-203.
Cox, T. B., and Low, J. R. (1974). An investigation of the plastic fracture of AISI 4340 and
18 nickel-200 grade maraging steels. Mer. Trans. 5 , 1457-1470.
Dafalias, Y. F. (1983). Corotational rates for kinematic hardening at large plastic
deformations. 1. Appl. Mech. 50, 561-565.
Doraivelu, S. M., Gegel, H. L., Gunaselsera, J. S., Malas, J. C., Morgan, J. T., and
Thomas, J. F. (1984). A new yield function for compressible P/M materials. Inr. J .
Mech. Sci. 26, 527-535.
Dyson, B. F. (1976). Constraints on diffusional cavity growth rates. Metal Science 10,
349-353.
148 Viggo Tvergaard
Dyson, B. F. (1983). Continuous cavity nucleation and creep fracture. Scripta Metall. 17,
31-37.
Frost, H. J., and Ashby, M. F. (1982). “Deformation Mechanism Maps: The Plasticity and
Creep of Metals and Ceramics.” Pergamon Press, Oxford.
Goods, S. H., and Brown, L. M. (1979). The nucleation of cavities by plastic deformation.
Acta Metall. 27, 1-15.
Green, A. E., and Zerna, W. (1968). “Theoretical Elasticity.” Oxford University Press,
Oxford.
Guennouni, T., and Francois, D. (1987). Constitutive equations for rigid plastic or
viscoplastic materials containing voids. Fatigue Fract. Engng. Mater. Struct. 10,
399-4 18.
Gurson, A. L. (1977a). Continuum theory of ductile rupture by void nucleation and
growth-I. Yield criteria and flow rules for porous ductile media. 1. Engng. Materials
Technol. 99, 2T15.
Gurson, A. L. (1977b). Porous rigid-plastic materials containing rigid inclusions-Yield
function, plastic potential, and void nucleation. In “Proc. Int. Conf. Fracture” (D. M.
R, Taplin ed.), Vol. 2A, 357-364, Pergamon Press.
Hadamard, J. (1903). “Lecons sur la propagation des ondes et les &pations de
I’hydrodynamique.” Libraire Scientifique A., Hermann, Paris.
Hahn, G. T. (1984). The influence of microstructure on brittle fracture toughness. Met.
Trans. 15A, 947-959.
Hahn, G. T., and Rosenfield, A. R. (1975). Metallurgical factors affecting fracture
toughness of aluminum alloys. Met. Trans. 6A,653-668.
Hancock, J. W. (1982). Plasticity of porous metals. In “Yield, Flow and Fracture of
Polycrystals” (T. N. Baker ed.), pp. 69-79. Appl. Sci. Publishers.
Hancock, J. W., and Brown, D. K. (1983). On the role of strain and stress state in ductile
failure. J. Mech. Phys. Solids 31, 1-24.
Hancock, J. W., and Mackenzie, A. C. (1976). On the mechanisms of ductile failure in high
strength steels subjected to multiaxial stress-states. J . Mech. Phys. Solids 24, 147-169.
He, M. Y., and Hutchinson, J. W. (1981). The penny-shaped crack and the plane strain
crack in an infinite body of power-law material. J . Appl. Mech. 48, 830-840.
Hill, R. (1962). Acceleration waves in solids. J . Mech. Phys. Solids 10, 1-16.
Hull, D.,and Rimmer, D. E. (1959). The growth of grain-boundary voids under stress.
Phil. Mag. 4, 673-687.
Hutchinson, J. W. (1973). Finite strain analysis of elastic-plastic solids and structures. In
“Numerical Solution of Nonlinear Structural Problems” (R. F. Hartung ed.), p. 17,
ASME, New York.
Hutchinson, J. W. (1983). Constitutive behaviour and crack tip fields for materials
undergoing creep-constrained grain boundary cavitation. Acta Metall. 31, 1079-1088.
Hutchinson, J. W., and Tvergaard, V. (1980). Surface instabilities on statically strained
plastic solids. Int. J. Mech. Sci. 22,339-354.
Hutchinson, J. W., and Tvergaard, V. (1981). Shear band formation in plane strain. In?. J .
Solids Structures 17,451-470.
Hutchinson, J. W., and Tvergaard, V. (1987). Softening due to void nucleation in metals.
Harvard University, Division of Appl. Sci., Mech-99.
Koplik, J., and Needleman, A. (1987). Void growth and coalescence in porous plastic
solids. Brown University, Division of Engineering.
Lee, E. H., Mallett, R. L., and Wertheimer, T. B. (1983). Stress analysis for kinematic
hardening in finite-deformation plasticity. J . Appl. Mech. 50, 554-560.
Li, F. Z., Needleman, A., and Shih, C. F. (1988). Creep crack growth by grain boundary
cavitation: Crack tip fields and crack growth rates under transient conditions. Brown
University Report.
 Material Failure by Void Growth 149
Magnusen, P. E., Koss, D. A., and Dubensky, E. M. (1987). Ductile fracture and random
vs. regular hole/void arrays. Office of Naval Research, NW14-86-K-0381, Techn.
Rep. No. 5.
Marciniak, K., and Kuczynski, K. (1967). Limit strains in the process of stretch forming
sheet metals. Inr. J . Mech. Sci. 9, 609-620.
McClintock, F. A. (1968). A criterion for ductile fracture by growth of holes. J . Appl.
Mech. 35,363-371.
McMeeking, R. M. (1977). Finite deformation analysis of crack tip opening in elastic-plastic
materials and implications for fracture initiation. J. Mech. Phys. Solid 25, 357-381.
Mear, M. E., and Hutchinson, J. W. (1985). Influence of yield surface curvature on flow
localization in dilatant plasticity. Mech. Materials 4,395-407.
Moussy, F. (1985). Les differentes echelles du developpement de I’endommagement dans
del aciers. Influence sur la localization de la deformation a I’echelle microscopique. In
“Proc. Int. Symp. on Plastic Instabjlity; Considere Memorial” (J. Salencon ed.),
pp. 263-272. Presses Ponts et ChaussAes.
Needleman, A. (1972). Void growth in an elastic-plastic medium. J. Appl. Mech. 41,
964-970.
Needleman, A. (1987). A numerical study of void nucleation at carbides. Brown
University, Division of Engineering.
Needleman, A., and Rice, J. R. (1978). Limits to ductility set by plastic flow localization. I n
“Mechanics of Sheet Metal Forming” (D. P. Koistinen et ul. eds.), pp. 237-267.
Plenum Publishing.
Needleman, A., and Rice, J. R. (1980). Plastic creep flow effects in the diffusive cavitation
of grain boundaries. Acru Metull. 28, 1315-1332.
Needleman, A., and Tvergaard, V. (1984a). An analysis of ductile rupture in notched bars.
1.Mech. Phys. Solia3 32,461-490.
Needleman, A., and Tvergaard, V. (1984b). Finite element analysis of localization in
plasticity. In “Finite Elements, Special Problems in Solid Mechanics” (J. T. Oden
and G. F. Carey eds.), Vol. V, pp. 94-157. Prentice-Hall.
Needleman, A., and Tvergaard, V. (1984~).Limits to formability in rate-sensitive metal
sheets. In “Mechanical Behavior of Materials-IV” (J. Carlsson and N. G. Ohlson
eds.), pp. 51-65. Pergamon Press, Oxford.
Needleman, A., and Tvergaard, V. (1987). An analysis of ductile rupture modes at a crack
tip. J. Mech. Phys. Solids 35, 151-183.
Nemat-Nasser, S., and Taya, M. (1977). Model studies of ductile fracture. In “Continuum
Models of Discrete Systems,” pp. 387-405. University of Waterloo Press.
Nemat-Nasser, S., Sumi, Y.,and Keer, L. M. (1980). Unstable growth of tension cracks in
brittle solids: Stable and unstable bifurcations, snap-through, and imperfection
sensitivity. Inr. J . Solids Structures 16, 1017-1035.
Norris, D. M. (1979). Computer simulation of the Charpy V-notch toughness test.
Engineering Frucrure Mechanics 11,261-274.
Nutt, S. R., and Needleman, A. (1987). Void nucleation at fiber ends in AI-Sic composites.
Scriptu Metullurgicu 21,705-710.
Pan, J., Saje, M., and Needleman, A. (1983). Localization of deformation in rate sensitive
porous plastic solids. Inr. J. Fracture 21, 261-278.
Paris, P. C., Tada, H., Zahoor, A., and Emst, H. (1979). Initial experimental investigation
of tearing instability theory. In “Elastic-Plastic Fracture,” ASTM STP 668.
Peirce, D., Shih, C. F., and Needleman, A. (1984). A tangent modulus method for rate
dependent solids. Compur. Struct. 18, 875-887.
Petch, N. J. (1958). The ductile-brittle transition in the fracture of a-iron: I. Phil. Mug. 3,
1089- 1097.
Puttick, D. E. (1960). Ductile fracture in metals. Phil. Mug. 4,964-969.
150 Viggo Tvergaard
Rice, J. R. (1976). Elastic-plastic fracture mechanics. In “The Mechanics of Fracture” (F.
Erdogan ed.), pp. 23-53. ASME AMD 19.
Rice, J. R. (1977). The localization of plastic deformation. In “Theoretical and Appl.
Mech.” (W. T. Koiter ed.), pp. 207-220. North-Holland.
Rice, J. R. (1981). Constraints on the diffusive cavitation of isolated grain boundary facets
in creeping polycrystals. Acta Metall. 29, 675-681.
Rice, J. R., and Johnson, M. A. (1970). The role of large crack tip geometry changes in
plane strain fracture. In “Inelastic Behaviour of Solids” (M. F. Kanninen el al. eds.),
pp. 641-672. McGraw-Hill, New York.
Rice, J. R., and Tracey, D. M. (1969). On the ductile enlargement of voids in triaxial stress
fields. J . Mech. Phys. Solids 17, 201-217.
Riedel, H. (1986). “Fracture at High Temperatures.” Springer-Verlag, Berlin.
Ritchie, R. O., Knott, J. F., and Rice, J. R. (1973). On the relationship between critical
tensile stress and fracture toughness in mild steel. J. Mech. Phys. Solids 21, 395-410.
Rolfe, S . T., and Barsom, J. M. (1977). “Fracture and Fatigue Control in Structures-
Applications of Fracture Mechanics.” Prentice-Hall, Englewood Cliffs, New Jersey.
Rudnicki, J. W., and Rice, J. R. (1975). Conditions for the localization of deformation in
pressure-sensitive dilatant materials. J. Mech. Phys. Solids 23, 371-394.
Saje, M., Pan, J., and Needleman, A. (1982). Void nucleation effects on shear localization
in porous plastic solids. Inr. J. Fracture 19, 163-182.
Sham, T.-L., and Needleman, A. (1983). Effects of triaxial stressing on creep cavitation of
grain boundaries. Acta Metall. 31, 919-926.
Shima, S., and Oyane, M. (1976). Plasticity theory for porous metals. Int. J. Mech. Sci. 18,
285-291.
Smith, E. (1968). Cleavage fracture in mild steel. Znt. J . Fracture Mechanics 4, 131-145.
Speich, G. R., and Spitzig, W. A. (1982). Effect of volume fraction and shape of sulfide
inclusions on through-thickness ductility and impact energy of high-strength 4340 plate
steels. Met. Trans. W A , 2239-2258.
Stone, R. H., and Psioda, J. A. (1975). Discussion of Metallurgical factors affecting
fracture toughness of aluminum alloys. Met. Trans. 6A,668-670.
Tvergaard, V. (1978). Effect of kinematic hardening on localized necking in biaxially
stretched sheets. Int. J. Mech. Sci. 20, 651-658.
Tvergaard, V. (1981). Influence of voids on shear band instabilities under plane strain
conditions. Int. J. Fracture 17, 389-407.
Tvergaard, V. (1982a). On localization in ductile materials containing spherical voids. Int.
1. Fracture 18,237-252.
Tvergaard, V. (1982b). Ductile fracture by cavity nucleation between larger voids. J. Mech.
Phys. Solids 30, 265-286.
Tvergaard, V. (1982~).Material failure by void coalescence in localized shear bands. Int. J .
Solids Structures 18, 659-672.
Tvergaard, V. (1982d). Influence of void nucleation on ductile shear fracture at a free
surface. J . Mech. Phys. Solids 30,399-425.
Tvergaard, V. (1984a). On the creep constrained ditksive cavitation of grain boundary
facets. J. Mech. Phys. Solids 32, 373-393.
Tvergaard, V. (1984b). Constitutive relations for creep in polycrystals with grain boundary
cavitation. Acta Metall. 32, 1977-1990.
Tvergaard, V. (1985). Effect of grain boundary sliding on creep constrained diffusive
cavitation. J. Mech. Phys. Solids 33, 447-469.
Tvergaard, V. (1986). Analysis of creep crack growth by grain boundary cavitation. Int. J.
Fracture 31, 183-209.
 Material Failure by Void Growth 151
Tvergaard, V. (1987a). Effect of yield surface curvature and void nucleation on plastic flow
localization. J. Mech. Phys. Solids 35, 43-60.
Tvergaard, V. (1987b). Ductile shear fracture at the surface of a bent specimen. Mech.
Materials 6 , 53-69.
Tvergaard, V. (1987~).Effect of anisotropic hardening on ductile failure of a material
containing two size-scales of particles. Technical University of Denmark, DCAMM
Report No. 354.
Tvergaard, V. (1987d). 3D-analysis of localization failure in a ductile material containing
two size-scales of spherical particles. Technical University of Denmark, DCAMM
Report No. 365.
Tvergaard, V. (1988). Effect of microstructure degradation on creep crack growth. Dept. of
Solid Mech., Technical University of Denmark.
Tvergaard, V., and Needleman, A. (1984). Analysis of the cup-cone fracture in a round
tensile bar. Actu Metall. 32, 157-169.
Tvergaard, V., and Needleman, A. (1986). Effect of material rate sensitivity on failure
modes in the Charpy V-notch test. J. Mech. Phys. Solids 34, 213-241.
Tvergaard, V., and Needleman, A. (1987). Temperature and rate dependence of Charpy
V-notch energies for a high nitrogen steel. Brown University, Division of Engineering.
Tvergaard, V., Needleman, A., and Lo, K. K. (1981). Flow localization in the plane strain
tensile test. 1.Mech. Phys. Solids 29, 115-142.
Yamamoto, H. (1978). Conditions for shear localization in the ductile fracture of
void-containing materials. Int. J. Fracture 14, 347-365.
Ziegler, H. (1959). A modification of Prager’s hardening rule. Quart. Appl. Math. 17,
55-65.
This Page Intentionally Left Blank
 ADVANCES IN APPLIED MECHANICS, VOLUME 27

Complete Solutions to Some Mixed Boundary

Value Problems in Elasticity
V. I. FABRIKANT

Department of Mechanical Engineering

Concordia University
Montreal, Canada

I. Introduction. ... .... . . .. .. . .. . . .. . . .. . . _.... . .. . .. . . _.._.... ... .._.. . . 153

11. The0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
A. GeneralSolution ................................................... 155
B. Flat Crack under Arbitrary Normal Loading . . . . . . . . . . . . . . . . . . . . . . . . . . 158
C. Application 1: Point Force Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
D. Application 2: Concentrated Load outside a Circular Crack . . . . . . . . . . . . . 165
E. Contact Problem for a Smooth Punch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
F. Plane Crack under Arbitrary Shear Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
G. Application: Concentrated Tangential Loading of a Penny-Shaped Crack . 177
111. Illustrative Examples . . . . . . . . . . . . . . . . . . . . . . .
A. Example 1: Penny-Shaped Crack under Unif
B, Example 2: Flat Centrally Loaded Circular Punch . . . . . . . . . . . . . . . . . . . . . 187
C. Example 3: Inclined Punch on an Elastic Half-Space . . . . . . . . . . . . . . . . . . . 191
D. Example 4: Penny-Shaped Crack under Uniform Shear Loading . . . . . . . . . 192
IV. Discussion............................................................ 197
V. Conclusion . ..... . . .. ......... ... ...... . . . ....... . ... . .. 201
AppendixA .......................................................... 202
Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
Appendix C . . .. . .. . . . .. . . . .. .. . . . . . . . .. . . ... ... .. . . . . . .. ... . . . . . . . 206
AppendixD .......................................................... 211
AppendixE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
AppendixF ...............................................
Appendix G .. ..... ..... ... ... . _ . ... ... . _ ._ . .... . ... .... .... .. 218
Acknowledg . . . . . . . . . . , . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

I. Introduction
The great majority of solved punch and crack problems deal with the
stresses and displacements in the plane z = 0 only. There are just a few
153
Copyright 0 1990 Academic Ress. Inc.
All rights of reproduction in any form reserved.
ISBN 012002027-0
154 V. I. Fabrikant

complete solutions published (Sneddon, 1951; Elliott, 1949; Westmann,

1965) where explicit expressions are given for the field of displacements
and stresses for the most simple axisymmetric problems (a circular punch
and a penny-shaped crack). Some explicit expressions for the field of
displacements for an elliptic crack can be found in (Kassir and Sih, 1975).
We are unaware of any nonaxisymmetric punch or crack problem for a
transversely isotropic elastic half-space solved completely in elementary
functions. Knowledge of complete solutions is indispensable for con-
sideration of more complicated problems of crack interactions, influence
of external loads on punches and cracks, etc.
The integral transform method combined with coupled integral equa-
tions is the most widely used for the solution of mixed boundary value
problems. Some quite remarkable results were achieved by Uflyand
(1977), nevertheless, one has to admit that our inability to solve problems
directly is the main reason for a wide use of various integral transforma-
tions. Here we present some hints indicating that a direct method of
solving some problems in elasticity does exist and is elementary. Consider
the simplest case of a penny-shaped crack of radius a subjected to
axisymmetric pressure p ( p ) . The corresponding potential function f can
be found in Kassir and Sih (1975) and is as follows:

where

The user had to substitute the explicit expression for p in (1.2) and to
compute two consecutive integrals. The result is substituted in (l.l), and
an infinite integral with Bessel function is computed. It seems natural to
try to spare one integration by substituting (1.2) in (l-l),changing the
order of integration and computing the integral (Gradshteyn and Ryzhik,
1963, formula 6.752.1)
Lm sin stJo(ps)exp(-sz) ds =sin-’(’>.
S l2(t)
where the notation
12(t)= J { [ ( p + q2+ 22]1’2 + [ ( p - t)2+ 22]1’2}
 Solutions to Some Boundary Value Problems 155

was introduced. Now formulae (1.1) and (1.2) can be unified giving

Note that expression (1.5) contains no trace of the integral transform,

and thus gives us a hint that a direct and elementary solution is indeed
possible. For example, in the case of a uniform loading p = const., the
potential function (1.5) will take the form

The integral in (1.6) though looking formidable due to (1.4) can be

computed in elementary functions as shown in Appendix B (B.13),
namely,

Here l2 is defined by (1.4), and

ll(t)= t { [ p + t)2 + z y - [(p - t)2 + 2211’2). (1.8)
One can show that arbitary polynomial loading in (1.5) will lead to an
elementary expression for the potential function, and thus to an
elementary complete solution.
The considerations above give just a hint that a direct solution is
possible. This article describes how a general nonaxisymmetric solution
can be derived for various punch and crack problems. The solution is
elementary for the case of a circular line of division of boundary
conditions; no integral transforms or special function expansions are
involved. One may just wonder why this solution was not discovered at
least a century ago.

11. Theory

A. GENERAL
SOLUTION

Consider a transversely isotropic elastic body that is characterized

by five elastic constants Aik defining the following stress-strain
156 V . I . Fabrikant

relationships

au, au, dW
0,= A 13 -
ax
+ A 13 -
aY
+ A33 -,
az

TXY
dux
= A,(ay +,),au,
auy aw

The equilibrium equations are

Substitution of (2.1) in (2.2) yields

(2.3)
Now we introduce the complex tangential displacements u = u, + iu, and
ii = u, - iu,. This will allow us to reduce the number of equations in
(2.3) by one and to rewrite these equations in a more compact manner,
 Solutions to Some Boundary Value Problems 157

namely
1 d2U 1 dW
(A11 +AM) AU +A,,
dz
+-2 (A11 -AM) A2ii + (A13 +AM) A- dz = 0
(2.4)
a2w 1 d -
AM AW+ A 3 3 7 + - (A13 +AM)- (Au + Aii) =O.
az 2 az

Here the following differential operators were used

a2
A=-+- d2 a
A=-+i--,
a
ax2 dy2’ ax ay
and the overbar everywhere indicates the complex conjugate value. Note
also that A = M.One can verify that equations (2.4) can be satisfied by

u = A(fi + I$ + i&),
where all three functions Fksatisfy the equation (Elliott, 1948)

The values of mk and Yk are related by the following expression (Elliott

1948)

Y3 = (A44/&6)’5
Introducing the notation zk = z / y k , for k = 1, 2, 3, we may call the
function Fk= F(x, y, z k ) harmonic. Note the property mlmz = 1, which
seems to have escaped the attention of previous researchers, and which
will help us to simplify various expressions to follow. The other elastic
constants that will be used throughout the chapter are
G1= B + YlY2H, G2 = B - YlY2H,

Introduce the following parameters

u1= a, + uy, u2= a, - uy + Zit,,, tz= tzx+ityz. (2.10)
158 V. I. Fabrikant

This will simplify expressions (2. l ) , giving

We have now only four components of stress, instead of six, as it was in

(2.1). The substitution of (2.6) in (2.11) yields

(2.12)

= -AM + ml)F, + (1 + mz)F,],

A[(l
a
r, = A a A- [(1+ ml)F, + (1 + m2)&+ ie].
dZ

Here we used the fact that each Fk satisfies equation (2.7), and the
relation (for k = 1, 2) Ally: - A13mk = &(1+ mk), which is an immedi-
ate consequence of (2.8). Expressions (2.6) and (2.12) give a general
solution, expressed in terms of three harmonic functions &. It is very
attractive to express each function Fk through just one harmonic function

(2.13)
where zk = z / Y k , and ck is an as yet unknown complex constant. As we
shall see further, this is possible indeed.

B. FLATCRACKUNDER ARBITRARY
NORMAL
LOADING

Consider a transversely isotropic elastic space weakened by a flat crack

S in the plane t = 0, with arbitrary pressure p applied to the crack faces.
Due to symmetry, the problem can be formulated as follows: find the
solution to the set of differential equations (2.3) for a half-space z 20,
subject to the mixed boundary conditions on the plane z = 0
a, = -p(x, y) for (x, y ) E S , w = 0 for ( x , y ) @ S,
rz = 0 for --03 c ( x , y ) < m. (2.14)
 Solutions to Some Boundary Value Problems 159

These conditions can be satisfied by one harmonic function. Let us put,

according to (2.13),
F,(z) = c l F ( ~ l ) , 4.2) = c ~ F ( z ~ ) , F~(z)
= 0. (2.15)
Expressions of the type F,(z) and F ( z , ) , etc., everywhere in this article
should be understood as Fl(x, y, z) and F ( x , y, zl) respectively. The
substitution of (2.15) and (2.12) in the third condition (2.14) yields

(2.16)

We can present the function F through the potential of a simple layer,

namely,

(2.17)

where o stands for the crack face displacement w(x, y, 0), R(M,N) is
the distance between the points M ( p , 9, z) and N(r, 3, 0), and the
integration is taken over the crack domain S. Expression (2.17) satisfies
the second condition (2.14) identically, due to the well-known property
of the potential of a simple layer. Inside the crack the same property
gives

El
a2 ==O
= -2no = -2nw(x, y, 0). (2.18)

Now expressions (2.15)' (2.17), (2.18), and (2.6) give the second
equation for c1 and c2

(2.19)

The constants c1 and c2 are defined from (2.16) and (2.19) as

Y1 Y2
c1= - c2= - (2.20)
2n(m1 - 1) ' 2n(m2 - 1 ) .
The potential functions will be defined by

Y1 y2
F,(z)= - 2n(ml - 1)F(z1), F,(z)= - 2n(m2 - F(Z2). (2.21)

The substitution of (2.21) and (2.12) in the first condition (2.14) leads to
160 V. I. Fabrikant

the governing integral equation

(2.22)

where, as before, R(No, N ) stands for the distance between two points No
and N , and both No, N E S. The following identities were used
m,-1
mlm2=1, -- - 2nA44wY1 - Y2). (2.23)
m1 + 1
The case of a general crack under a uniform pressure was considered
by Fabrikant (1987a), where an approximate analytical solution was
presented. A similar method can be used in the case of arbitary
polynomial loading. An exact solution in elementary functions is possible
when the crack is circular. Let a be the radius of the crack. The
governing integral equation (2.21) can be rewritten in polar coordinates
as follows (see Fabrikant (1986a)):

(2.24)

Here the L-operator is defined by

(2.25)

and
(2.26)

The integral operator inverse to (2.24) was derived by Fabrikant (1986a),

and is

Another form of solution was derived by Fabrikant (1987b):

 Solutions to Some Boundary Value Problems 161
where

(2.29)

In this article, we do not stop at considering the situation in the plane

z =0: our purpose is obtaining a complete solution. We shall call
F ( p , @, z), as defined by (2.17), the main potential function since both
functions Fl and 6 become easily available, when F is found. The
substitution of (2.28) in (2.17) allows us to express the main potential
function as follows:

where the Green’s function K reads

(2.31)

Here R(-, stands for the distance between respective points:

a)

M ( p , 9, z), N(r, I), 0), and No(po,@o, 0). Although we can not compute
the integral in (2.31) in elementary functions, all its derivatives can be
expressed in elementary functions, due to the fundamental integral
established in Appendix C. Making use of (C.19), one can write

(2.32)

where

(2.33)

and the contraction 1, everywhere in this article stands for /‘(a), as it is

defined by (1.8). Note that h tends to q, as defined by (2.29), for 2-0
and p < a. Expressions (2.30) and (2.32) allow us to write
162 V. I. Fabrikant

The integral in (2.34), although looking difficult to compute even for

p = const, can be expressed in elementary functions for any polynomial
loading. This becomes evident, if we use the equivalent representations
through the L-operator (see Fabrikant (1986b) for details):

Here
g(x)=x
[l + r
p z- 2x 2 I1I2J
(2.36)

and the contraction 1’ everywhere in this article stands for &(a), as it is

defined by (1.4). Using the changes of variables x = Z2(t), t = g ( x ) ,
expression (2.35) can be rewritten as follows:

Since the function F vanishes at infinity, it can be defined from (2.37):

(2.38)
By using the property

ai2(t)- - [ g ( t )- p2I1/’arl(t)
--
at [p2--iI:(t)]ln az ’

which is a consequence of formulae (A.27) and (A.28), expression (2.38)

can be modified:

(2.39)

Expression (2.39) is convenient for exact evaluation of the potential

function F and proves that it can be expressed in elementary functions for
arbitrary polynomial loading. A simple change of variables gives another
 Solutions to Some Boundary Value Problems 163

formula, equivalent to (2.39),

dx

(2.40)
We can proceed now with the remaining derivatives of the Green's
function K, defined by (2.31). Differentiation of (2.31) yields
2~ a pei+ - re"Y
AK(p, 9, '; Po, $0) =
R3(M, N )
(a' - r2)'/'(a2 - P ; ) " ~ r dr d v
tan-'[ a R ( N , No) 1
R ( N , No) ' (2*41)
This integral is computed in Appendix E. By using ( E . l l ) , one can write
M P , 9 , z ; Po, 90)
2jd z
9 Ro
=-[-tan-(&) -
(a2 - p;y2 tan-'[
s
d
(1; - a2)1/'
I} '
(2.42)

where h is defined by (2.33) and

4 = pe-i+ - Poe-"tJ~ , J =
- PPoe-i(+-+o) )
la,
(2.43)
Ro = R ( M , No) = [P' + P; - 2 ~ COS(@
~ 0- $0) + ~'1''~.
The other derivatives, which will be needed for the complete solution,
are
a2
-~ ( p9,
, z ; po, $J~)
dZ2
= 2jd
[-tan-'
ii
(2.44)
d
-WP,
9 , ~PO,; 90)= 2x[
pe'+ - poei+Otan-'( &)
d2 Ri

+-R ; +h h 2 (-pe'@ + pe'@-poei'o)]

1g-I; R;
I (2.45)

A 2 K ( p , 9, 2 ; PO, 90)= 2jd tan-

- z(3Rz - 2') tan-'(

92R;
&)
(a2 - p;)'R(I$ - a2)"2poe'+0
+ q j 2 [ r f - PPoe-i(+-+~)]
164 V. I. Fabrikant

This concludes the general solution to the problem of a penny-shaped

crack subjected to arbitrary pressure. Formulae (2.32) and (2.39)-(2.46)
are the main new results of this section.

1: POINTFORCE
C. APPLICATION LOADING

Consider a penny-shaped crack opened by two equal concentrated

forces P applied in opposite directions at the point (pot @o, O*), po < a.
Formulae (2.6), (2.12), (2.21), (2.32), and (2.39)-(2.42) give a complete
solution for the field of displacements and stresses in elementary
functions, namely

(2.47)

(2.48)

(2.49)

(2.50)

(2.51)

(2.52)

where
tan-'[ s ]- ;
tan-'(&)), (2.53)
(1; - a 2 y 2
1
A(z) =-tan-' (2.54)
RO

(2.55)
 Solutions to Some Boundary Value Problems 165

(2.56)

(2.57)
+ +
Here Ro = [p' pi - 2pp0 cos($ - &) z2]'l2. It is believed that this
problem has not been solved before. The expression for a, (2.51)
simplifies when z = 0 and p > a, namely,

Defining the stress intensity factor

k l = lim [ ( p - a)'/2az],

the following result may be obtained from (2.58):

P (a' - p y
kl = (2.59)
n
2
(242)
1/2 2
a + p; - 2apo cos($ - $0).
One can write for arbitrarily distributed pressure,

1
k' = d ( 2 a ) ' " I, b
2n (a' - PiY2P(P0, $o)P0 dpo d$0
a2 + pi - 2ap, cos($ - $0) '
which corresponds to the well-known result (Cherepanov, 1974).

2: CONCENTRATED
D. APPLICATION LOADOUTSIDE A CIRCULAR
CRACK

Consider transversely isotroic space weakened by a penny-shaped

crack of radius a in the plane z = 0. Let a concentrated force P be applied
at arbitrary point ( p , $, z) in the Oz direction. The crack faces are
stress-free. The problem has not been considered before, even for an
166 V. I. Fabrikant

isotropic body. A complete solution is beyond the scope of this article

and will be published separately. Here we find the crack opening
displacement and the opening mode stress intensity factor k , only.
Consider the second system in equilibrium: two unit concentrated
forces Q applied normally to the crack faces in opposite directions at the
point (po, $o, O*). Denote the normal displacement in the space due to
the forces Q by we; wP is the crack opening displacement
w(po, &, O + ) - w ( p 0 , C # J ~ , 0-) due to force P. Application of the recipro-
cal theorem yields
QWP = PWQ,

which gives the crack opening displacement

with fi defined by (2.54). The stress intensity factor can be defined by

where
(2- 1:y2
f6G) = , +
r: = p2 a2 - 2pa COS($- $o) +.'2 (2.61)
r:
The stress intensity factor vanishes as z tends to zero for p 2 a.
In the case of an isotropic body, expression (2.60) transforms into

Here p is the shear modulus, and Y is Poisson's ratio. The corresponding

expression for the stress intensity factor will take the form

In the case of axial symmetry p = O , and formulae (2.62) and (2.63)

 Solutions to Some Boundary Value Problems 167

simplify as follows:

- p;)l”
+ 2(p;z2(a2
+ z’)(a2 + 2’) IJ
which is in agreement with the results reported by Collins (1962), who
considered the axisymmetric case only.

PROBLEMFOR
E. CONTACT A PUNCH
SMOOTH
A smooth punch is pressed against a transversely isotropic elastic
half-space z -> 0 by a normal force P. Let S denote the domain of contact.
The mixed boundary conditions on the plane z = 0 are

0, =o for (P, @)es, W = d P , @) for (P, @)ES,

rz = 0 for -m < ( x , y) < m. (2.64)
We may assume again that
F,(z) = c ~ F ( z ~ ) , &(z) =cZF(Z~), &(z) = 0. (2.65)
Substitution of (2.65) and (2.12) in the third condition (2.64) yields

C2Y1
c1= --*
mlY2
Now we need to define the main potential function F so that its second
z-derivative vanishes at z = 0 outside the domain of contact. Comparison
with (2.17) gives

F(p, @, z) = F ( M ) = 11
S
ln[R(M, N) + z ] a ( N )dS,. (2.67)

The simple layer potential property yields, inside the domain of contact,

(2.68)
168 V. 1. Fabrikant

which, combined with (2.12), gives the second equation for the constants
c1 and c2
2xAU[(1 + ml)cl + (1 + m2)c2]= 1. (2.69)

Equations (2.66) and (2.69) define the constants

c1=- HY 1
HY2
c2 = - (2.70)
ml - 1’ mz-l’

The simplifications made in (2.70) are due to the properties (2.23).

Finally, substitution of (2.65), (2.67), and (2.70) in the second equation
of (2.6) yields for z = 0 the well-known governing integral equation of
elastic contact problem for a smooth punch:

(2.71)

An approximate analytical solution of (2.71) for a flat punch of arbitrary

shape can be found in Fabrikant ( 1 9 8 6 ~1986d). The case of a curved
punch was considered in Fabrikant (1987d). An exact solution in closed
form is available for a circular punch of arbitrary profile (Fabrikant,
1986b). The integral equation (2.71) can be rewritten for a circular
domain of contact of radius a as follows:

Its exact closed form solution is (Fabrikant, 1986b)

(2.72)

The potential function F can be found in two stages. First of all, one has
from (2.67)

Substitution of (2.72) in (2.73) yields, after integration (see Appendix F

 Solutions to Some Boundary Value Problems 169

for details),

Here Ro = [P’ + pi - 2pp0 cos(@- @ ),, + z’]’“, and h is defined by

(2.33). The next integration of (2.74) with respect to z gives the
expression for the potential function directly through the prescribed
displacement under the punch, namely,
1 2n a
F@i $ 1 Z) = J G ~ H I, I,
@, Z ; PO, @ O ) W ( P O , @O)PO dpo d@o (2.75)
~ ( P P 9

where
%(pi @, Z ; P O , $0)
Ro
1
= - -tan-’
(&I + (a’ -p y
[ln[lz+ (1: - p2)1~1

f; = -
P er(+#Jo).
’
(2.76)
Po
The details of integation are presented in Appendix F. Now all the
Green’s functions, related to the field of displacements and stresses can
be obtained from (2.76) by differentiation in elementary functions:
ax
-=- z [?+
R tan-l(&)],
(2.77)
az Ri
(a’ - I:)’/’

(2.81)
170 V . I. Fabrikant

The following identities were used in the simplification of (2.77)-(2.81):

(I; - p$)(a’ - I:) + a2qq = a2(Ri + hZ),

(I; - p i ) ( q q + 2 2 ) - qq(a2 - p i ) = (1; - aZ)(R$+ h2).

Formulae (2.76)-(2.81) are the main new results of this section. They
give a complete solution for the considered contact problem when
combined with formulae (2.6), (2.12), (2.65), and (2.70). When the
prescribed displacement can be expressed as a sum of powers in x and y,
the complete solution is elementary. Some particular examples will be
considered in Section 111.

F. PLANECRACK
UNDER A~BITRARY
SHEAR
LOADING

Consider a transversely isotropic elastic space weakened by a flat crack

S in the plane z = 0, with arbitrary shear loading applied to the crack
faces antisymmetrically. The problem can be formulated as follows: find
the solution to the set of differential equations (2.3) for a half-space
z 2 0, subject to the mixed boundary conditions on the plane z = 0:

a, = 0 for --M < (x, y ) < -M.

It is no longer possible to present the solution in the form (2.15). A more
complicated representation is necessary, namely,
6 = Cl(AX1+ &I), F2 = ~ z ( A j 2+ Ax,), 5 = ~3(A23- &).
(2.83)
Here cl, c2, and c3 are the as yet unknown constants; xl, x2, and x3 are
the as yet unknown complex harmonic functions. Introducing the
notation z k = z / y k for k = 1,2,3, we assume also that
X d Z ) =X(Zlh X 2 ( Z ) = X(ZZ), X3(Z) =X e d . (2.84)
This assumption will allow us to reduce the problem to finding just one
harmonic function, which is much easier than searching for three. By
substituting (2.83) into the third equation of (2.12), we obtain the first
 Solutions to Some Boundary Value Problems 171

equation for the constants, namely,

c1 + m2c2= 0. (2.85)
The third condition in (2.82) is thus satisfied. Substitution of (2.83) in
(2.6) yields
u = c1(A2jr+ Ax1) + c2(A2j2+ Ax2) + ic3(A2j3- Ax3). (2.86)
The differential operators A and A are defined by (2.5). When z =0,
equation (2.86) transforms into
u = (cl + c2+ ic,) A 2 j + (cl + c2- ic3)AX. (2.87)
It is convenient to assume
c1 + c2 + ic, = 0. (2.88)
This assumption simplifies (2.87) giving
u = ( ~ 1 + ~2 - i ~ 3 )AX, (2.89)
and makes it possible to represent

x(M) = //
S
In[R(M,N ) + z ] u ( N ) dS,. (2.90)

The representation (2.90) satisfies identically the second condition (2.82),

and inside the crack the following equation becomes valid
1
c1 + c2 - ic3 = -.
2n
(2.91)

The solution of the set (2.85), (2.88), and (2.91) gives

1 1 1
c1= - c2= - ~ 3 = - . (2.92)
4n(m1- 1)’ 4n(m2- I)’ 4n
Substitution of (2.83) and (2.92) in the last expression (2.12) gives the
following expression for the tangential stress:

(2.93)
172 V. I . Fabrikant

Expression (2.93) simplifies, for z = 0,

Finally, satisfaction of the first condition (2.82) yields the governing

integro-differential equation
1
[G A ffR ( N , No) dSN
u( N )
t(No)= - 2n2(G?- GZ)
S

-I-G2 ff
S
d(N) d s ~ ] ,
R ( N , No)
(2.95)

where the elastic constants G1and G2 are defined by (2.9). Let the crack
boundary be described in polar coordinates by equation p = a ( @ ) , where
a ( @ ) is a single-valued continuous function. We can always choose the
coordinate axis orientation so that the first harmonic will vanish from the
Fourier expansion of a ( @ ) . An approximate analytical solution of (2.95)
for a general crack can be obtained by the method used in Fabrikant
(1987a). The method implies the following representation

ff A;o)
S R

Consider the case of a uniform shear loading. Let the tangential

displacements have the form

(2.97)

where uo is an as yet unknown complex constant. Substitution of (2.97) in

(2.96) yields, after integration and retaining the first two harmonics only,

ff
S R c A o )dSN
UO
16~d(G:- GZ)
P2 [1+ 3 cos 2(@- @o)]] d @ o .
(2.98)
 Solutions to Some Boundary Value Problems 173

By substituting (2.98) in (2.95) and performing the necessary

differentiation, we obtain the relationship between the shear loading and
the amplitude of the tangential displacements, namely,
1
t=
4n(G: - G;)(UoGiBi + ~ E o G z B z ) , (2.99)
where
(2.100)

Equation (2.99) can be solved:

(2.101)

The integrals in (2.100) can be computed easily for various crack shapes.
For example, a rectangular crack with sides 2a, and 2a2 is characterized
by the values

It is noteworthy that despite the fact that the integral representation

(2.96) is valid inside a circle pSmin{a(@)} only, and despite the
approximate nature of (2.98), the solution given by (2.99) and (2.100) is
exact for an ellipse (see Fabrikant (1987b) for details). We did not find in
the literature any reliable data related to a non-elliptical crack under
shear loading, therefore it is difficult to say how accurate the solution is
for various crack shapes. A complete investigation of a general crack
under shear loading is beyond the scope of this article and will be
published separately.
The integro-differential equation (2.95) can be solved exactly for a
penny-shaped crack. The closed form solution is (see Fabrikant (1987b)
for details)

(2.103)
174 V. I. Fubrikunt

where R and q are defined by (2.29), q is defined by (2.43), and

(2.104)

The potential functions can be found by substitution of (2.103) and

(2.90) in (2.83) and computation of the integrals involved. This looks at
first somewhat difficult, nevertheless, it will be shown here that all the
Green's functions can be expressed in elementary functions. Note the
following property

(2.105)
Introduce the following notation

(2.106)

Here the points N and No are characterized by the cylindrical coordinates

(r, 3,O)and (po, 40, 0) respectively. Note the following relationships of
symmetry:

Let R ( M , N) denote the distance between the points M ( p , $, z ) and

N(r, 3, 0). By using (2.105) one may write

(2.108)
 Solutions to Some Boundary Value Problems 175

Integration by parts in (2.108) leads to the important property

(2.109)

Two more properties can be obtained by applying A and A to both sides

of (2.109), namely,

(2.110)

(2.111)

The properties (2.109)-(2.111) will allow us to substitute the computa-

tion of various integrals involving E 3 , which look very formidable, by
computation of integrals involving expressions El and E 2 , some of which
have already been computed (2.41)-(2.46), and the remaining ones can
be computed relatively easy (see Appendix G).
Next, we introduce the notation
X = A2 -I-AX, Y = Ax -Ax. (2.112)
In order to obtain the complete solution, we shall need the following
expressions for various derivatives of X and Y: the tangential displace-
ments are defined by AX and AY; the normal displacements by axldz;
and the field of stresses may be computed through a2X/az2,A2X, A2Y,
A ( X / d z ) , and h ( d Y / d z ) . All the Green’s functions involved can be
expressed as various derivatives of two fundamental functions, namely,

K,(M, No) = 11
S
Ei(N, No) ln[R(M, N ) + 21 d&,
(2.113)

K z ( M , No) = 11
S
E2(N, No) lnfR(M, N ) + Z ] dSN.
176 V. I. Fabrikant

Rewrite formula (2.103) as

(2.114)
n S

By substituting (2.90) and (2.114) in (2.112) and using the properties

(2.109)-(2.111), we get the following results:

Kl+G'gz)TdS+A/( (Kl+C
51K 2 ) i d S ] ,
n S G1 S

Y=
n
[-A jj(KI - -
S
G2
GI
K 2 ) 5 d S + A jj(KI
-

S
-
G2
K2)tdS].

We shall only need the following derivatives of X and Y for the complete
solution:
H=- G2 -
n

+A2~~(Kl+~Kz)tdS], (2.11
S

AY=- G2 -
n S

+ A 2 j ( ( K 1 - G zGI
K2)tdS], (2.11
S

ax
-=-- G ~ - G ~ ~ G2 -
az n dz

+A (K1 + G2 K 2 ) t d S ] , (2.11
S Gl
dY - G l + G 2 d
az n az

+ A 1 5 ( K 1 - -G2
K2)tdS],
(2.11
S G1
 Solutions to Some Boundary Value Problems 177

a c1-c2a
-u=-- G2 -
az az S

+ A 2 ~ ~ ( K l +G2- K 2 ) f d S ] ,
(2.119)
S G1

G2 -
az

G2
+ A2 ( K , - -K 2 ) f d S ] , (2.120)
S
GI
a2x G, - c2a2 [ A I I ( K , + -G2 K- 2 ) r d S
az2- ~d az2 GI

+ A / / ( K I + GG1
'K2)idS], (2.121)
S

A2X = GI - G2
~
G2 -
Jd

+ A2 (K 1+ -
G2
G1
K 2 ) fdS], (2.122)
S

G2 -
3d
S

+ A 2 1 1 ( K I - -G2
K2)fdS].
(2.123)
S GI

All the Green's functions in (2.115)-(2.123) are computed in elementary

functions in Appendix G and in (2.42)-(2.46). The solution may be
considered complete. In order to avoid duplication, we do not write here
the explicit expressions for the fields of displacements and stresses since
the equivalent expressions for the relevant Green's functions are given in
the next section.

G. APPLICATION:
CONCENTRATED LOADING
TANGENTIAL OF A
PENNY-SHAPED
CRACK

Consider an infinite transversely isotropic solid weakened in the plane

z = 0 by a penny-shaped crack of radius a. Let two equal concentrated
178 V. 1. Fabrikant

forces T = T, + iTy be applied to the crack faces antisymmetrically at the

point No(po,#o, 0'). The previously obtained results give the complete
solution in elementary functions:

(2.125)

(2.126)

(2.127)

Here 3 indicates the real part, the elastic coefficients are defined by
(2.9), and the functions f with subindex less than 6 are defined by
formulae (2.53)-(2.57) , and the remaining functions are computed in
 Solutions to Some Boundary Value Problems 179

Appendix G, namely,
t 3(1; - a2)'I2tan-'[
s3 (1; -a
5

y
I} (2.130)

2 1/2 ( 5 -
fS(d= =1( a 2 - Po) { q [tan-'( (g 1
- 1)'/2
) - tan-'( (a2 - l Y ) ]
a( f - 1)'Q
4
(2.131)

1 a(1 - ?)'I2
- t( 1- t)3n tan-'( (l; - a 2 ) 1 / 2 ) } J (2.132)

(2.133)

x (tan-'[
1
( f - 1)'Q
] - tan-'[ (a2 - 1f)lQ
a(g - 1)'Q
I)]
180 V. I . Fabrikanl

5 2i pei@(3f; - a2f)
+ +
a ( f f- a2)1/2
fi&) =

3(1; - Itt) + f;t - 3p2)

+ ppoe'(@-@~)(21~
-
2 2i+ 2 2 10 2
a(1 - t)"2
(1 -3t)3'2 tan-'[ (+ a2)1'2 I} ' (2.137)

p e (a - po) (I2 - a2)1"(31; - ppoei(@-@O))

flS(Z) = (2.138)
- If)(/; - ppoe'(@-+0))2

fi6(z)= [R,4tan-'( &) + (a2 - pi)'"

1 Ri+z2

s ( 4 - 1)'/2
x (tan-'[ 1
( f - 1)'/2
1 - tan-1[(a2 - [:'li2])+
a(f - 1)'E
4
3 - ez:2} .

(2.139)
This problem was not considered before. The solution (2.124)-(2.139)
presents, in fact, the explicit expressions for the Green's functions, and
allows us to write a complete solution for the case of arbitrary tangential
loading in quadratures. The general results simplify significantly for
z = 0, namely,

pl[(qlg) - te2'@0] -
+G'
n [Rq
%tan-' 5
(R)
+
a2(1 - t)(l - f )
]T forpea, (2.140)

for p 5 a,

S
w=-
(p' - a y
s

(p2 - a2)10 e-i@o

- -- sin-'(:))] T) for p > a, (2.141)
?q UP0
 Solutions to Some Boundary Value Problems 181

u1= -z
2 %( (2nHAMy1y2- m2) [q~1 tan-'(:)
36 YlY2

+ 77

77pe-'@(3- f)
+-G2
Gl(u2 - p2)u2(1- f)2
IT} forp<a, (2.142)

al=O forp>u,

(2.143)

a, = 0,

(2.144)
Here R and 77 are defined by (2.29). The stress intensity factor of the
second and the third mode can be expressed through the decomposition
,+n) = ,z
,+ it,, which is related to t, by the relationship zz= dn)ei@.
Introducing the complex stress intensity factor
k2 + ik, = lim [ ( p - ~)'~~t,e-'@]], (2.145)
F a

one gets from (2.144)

k2 + ik3 =

(2.146)

In the general case of arbitrarily distributed loading, the stress intensity

factor takes the form

c2(3u - poei(@-@~) i6-

+-G1 le
~ ( -apoei(@-@o))2
@O)]pOdpo dG0, (2.147)

which is in agreement with the results of Fabrikant (1987b).

182 V. I . Fabrikant

111. Illustrative Examples

Here we present the complete solutions to some simple punch and

crack problems. The axisymmetric solutions were reported in the
literature, so that we shall be able to compare the results. The
non-axisymmetric ones seem to be new.

A. EXAMPLE CRACK
1: PENNY-SHAPED UNDER UNIFORM
PRESSURE

Let a penny-shaped crack of radius a be opened by the pressure

p = const. In this case one gets from (2.27)

w ( p , $) = 4 ~ p ( a '- (3.1)
The potential function F can be obtained by substitution of (3.1) in
(2.17). The integral can be computed in elementary functions (A.4):

( 2 d + 22' - p') sin-'

);( -
2a2-31:
(1; - (3.2)
a

The complete solution can be expressed through various derivatives of

the potential function, as prescribed by formulae (2.6) and (2.12). All the
derivatives are given in Appendix A. The solution is:
 Solutions to Some Boundary Value Problems 183

Here the following notation was introduced:

= f { [ ( a+P)’+2~]1”-[(a-P)2+2~]1’2},
12& = i{[(a+ p)’ + 2;]1’2+ [(a - p)’ + z:]l”}, (3-9 )
2
z&=-, k = 1, 2.
yk

The problem was first solved by Elliott (1949) by the integral transform
method. Our results are essentially in agreement with those of Elliott,
who expressed them in the form of integrals involving Bessel functions,
namely,

These integrals can be computed in elementary functions

S; = sin-’( E), S; =
a(a’ - 1
1; - 1:
y
, c’:=a(1;1;-- a’)”’
1:
-l y [ l ; + a2(p2 - 2a2 - 2 2 ) ]
c;= a(a’ (1; - 1 3 3
,

S! =
a(1; - u’))’n[a’(2a’ +
22’ - p’) - 1 3
,
(1; - 1 3 3

c:= z [ a - (a’ - l:)’n]

, s:=
a - (a’ -1 y
,
p(a2 - 1 y P
z [ a - (a’- 1 21 ) 1R] 2 a - (a’ -1 : y - al,(a2- 1 y
s: = , c: =
p ’ ( d - 1:)”’ P M -13 ’

’ ) ‘ ’ ’ ( l f+ 31: - 4a2)
c: = a’p(1; - ~(12” -1 3 3
,
a2p(a2- 1:)”’(31: + 1: - 4 ~ ’ )
s: =
(1; - 1 3 3 7

c;= 2a[(1$-P’a y - 23 - a(1; - a

1:--1:
y
’
184 V . 1. Fabrikant

sq =
2a[a - (a' - 1y2)]
- a(a2 - 1:)'"
P2 1;-1:
2a a - (az - Lq)'/2 - al,(a2 - I:)'"]
c;=-
P [ P 12(1$ - 13
- a ( a 2 - Lf)'"[l$ + a2(p2- 2a2 - 2 z 2 ) ]
(if - 1 3 3
There are some misprints (or errors) in the Elliott's paper. For example,
according to his formula (4.2.5), the tangential displacement u vanishes
on the plane z = 0, which can not be correct; there are missing terms and
obvious misprints in formula (4.2.6).
Though formulae (3.3)-(3.8) are valid for a penny-shaped crack
subjected to a uniform pressure only, they can be used to obtain some
general results that are valid for an arbitrary crack with a smooth
boundary subjected to arbitrary loading; namely, we can explore the
asymptotic behavior of displacements and stresses near the rim of a
general crack. Such expressions were derived first for the case of
transversely isotropic body by Kassir and Sih (1975) who considered a
more complicated case of an elliptical crack. Their derivation looks so
complicated that nobody so far has been able to verify its accuracy.
The case of a circular crack is much simpler, so that we can obtain
the same results by elementary means. Introduce the local system of
spherical coordinates ( r , 8, $), with the coordinate origin at the crack
rim. The following asymptotics are valid for the main parameters used in
(3.3)-(3.8):
p = a + r cos 8, z = r sin 8,
ilk a - irs;, 12k a +$rTi,
l$k - 2UrQk, (a' - i:k)'" (ar)1/2Sk,

Here the following notation was introduced:

Tk = ( Q k + cos 0)'l2, fork = 1, 2, 3. (3.11)

Introducing the opening mode stress intensity factor
p f i
kl=-, (3.12)
JT
 Solutions to Some Boundary Value Problems 185

the following asymptotic expressions can be derived by substitution of

(3.10) and (3.12) in (3.3)-(3.8):
u = u,, = - 2 n H k l f i r [fi+ 1- Y 2 T2 + 0(1), (3.13)
ml-1 m2-1
w =2 n H k l l h - [
mlSl + m2s2
m l - 1 m2-1
1- + O(r), (3.14)

Y: - (mk + 1 ) ~ T:k +0(1), (3.15)

(3.17)

tz= t,,,= - k1
~ ( Y I - Y Z )Q I
[3- S]+ O($),
Q2
(3.18)

These results were computed for @ = 0. This assumption allows us to

avoid a cumbersome axis transformation, without loss of generality. The
parameter u1 in this case is interpreted as the sum a,,+ a,, and
+
a, = a,, - a, 2it,,,.By taking the sum and the difference of (3.15) and
(3.16), one can get

Formulae (3.13)-(3.18) are essentially in agreement with the results of

Kassir and Sih (1975), except for some misprints; for example, one
should read fi and fi instead of n1 and n2 in the denominator of the
terms in curly brackets of formula (8.94~). In order to compare our
results with those of Kassir and Sih, one should keep in mind that their
definition of the stress intensity factor is lh times greater than ours, and
their notation nk corresponds to our y f . Kassir and Sih seem to have not
noticed the properties (2.23) and the relationship Sk = (sin B ) / ( y k T k ) ,
which in some cases can simplify their results significantly. For example,
they have an expression [A13mk -Al,yf]/[A,(mk l ) ] in formula+
(8.95a), without realizing that it is equal to -1 for k = 1, 2.
186 V. I. Fabrikant

Formulae (3.3)-(3.8) in the limiting case of yl+ y2+ y3+ 1, ml+

m2+ 1, H = ( 1 - v ) / 2 n p 9 and AM= A% = p9 give the complete solution
for an isotropic body. By using the L’HBpital rule, one obtains

(3.19)

(3.20)

[ a@; - a y 2
31
u1=-
2p
x [( 1 + 2 Y ) 1; - 1:
- sin-’(

+ az2[1;(1;+-a2(2a2+-2z2 - 3p2)]
a2)1a
1:)3(1;
(3.21)

(3.23)

= --
2 p z l l e i @ ( 1-~ - 5p2)+ 1 3
(3.24)
It i2(1; -1 3 3
This problem was first solved by Sneddon (1951) by the integral
transform method. He was seemingly unable to compute the potential
function (3.2), so that he resorted to differentiation under the integral
sign, with a subsequent computation of various integrals involving Bessel
functions. His final results are given as elementary functions of four
parameters, namely,

r=(I + (b)’) ’”, R~ = [(t)2+ (:)’ - 11 + 4( I)’,

e=tan-’ - ,
(3 $=tan-’(
p2 +2az
z 2 - a2
).
This choice of parameters is not the best possible. Here is one
illustration. The expression for S: in Sneddon’s notation takes the form
 Solutions to Some Boundary Value Problems 187

(Sneddon, 1951, p. 497)

with the limitation p # 0, and there is no indication of what the result

would be if p = 0. Our result is sin-'(a/12), with no limitations attached.
Introduction of Sneddon's parameters r and 8 seems to be unnecessary.
There exist relationships between his parameters R and C#J,and our lI and
1 2 , namely,
l2 - If f#l ( U Z - 1 y
R 2 = Z sin - =
a2 ' 2 (f: - 1 y *
These relationships may be used to compare the solutions, which are in
good agreement, except for some misprints: factor 5 is missing in formula
(139, p. 496), and the last term in formula (145, p. 499) should read -Sy,
rather than +Sg.
The asymptotic behavior of displacements and stresses near the crack
edge in an isotropic body can be found from either (3.13)-(3.18) or
(3.19)-(3.24). The result is

u = u, = -cos - 2(1 - Y) - cos2-

2cL
klfi 2e [ 2
+ 0(1), (3.25)

k lc f iL. 2e [
w = -sin - 2(1- Y)- cos2-
2
+ O(r), (3.26)

ol=-cos- 1+2v-sin-sin- +0 (1 ), (3.27)

fk1 i 2e [ 28 381
2

o2-fi
--cos-
k1 "[
2 1-2v-sin-sin-
2e 381
2
+o($), (3.28)

oz=-cos- 1+sin-sin- + 0 (1 ), (3.29)

$k1 2 e[ 2 301
2
3e
z, = tz,,= -sin e cos -+ ~(fi),
kl
(3.30)
2fi 2
which is in agreement with the results given in (Sih and Liebowitz, 1968).

B. EXAMPLE
2: FLATCENTRALLY
LOADED
CIRCULAR
PUNCH

Consider a transversely isotropic elastic half-space z 2 0, penetrated by a

rigid circular punch of radius a. The punch loading is statically equivalent
188 V. I . Fabrikant

to a centrally applied normal force P. Denote the punch settlement by

w = const. Solution of the integral equation (2.71) is given by (2.72), and
in this particular case takes the form

(3.31)

Integration of the last expression over the circle p 5 a relates the punch
settlement w with the total force P as
2wa
P=-.
3tH
The substitution of (3.31) in (2.67) leads to an integral which can be
computed by differentiation of the expression for the main potential
function (A.l)with respect to a. The result is

F
2w
=- z
JGH
[ sin-’( t)
- (a2- + a In[&+ (1; - p2)1’2]}. (3.32)

The appropriate differentiation of the potential function (3.32) gives the

complete solution:

(3.33)

This problem was first solved by Elliott (1949). His solution is

essentially in agreement with ours. The fact that he uses the notation p as
an elastic parameter and as a polar radius simultaneously, somewhat
complicates the comparison.
 Solutions to Some Boundary Value Problems 189

Introduce the stress intensity factor as

k l = lim [ ( a - p)'"aZ] for z = 0. (3.39)
F a

Substitution of (3.37) in (3.39) yields

0
k 1 = - n 2 H 6 -- --(-)
P a
2n 2
In
. (3.40)

The asymptotical behavior of the field of stresses and displacements near

the punch edge can be derived by substitution of (3.10) and (3.40) in
(3.33)-(3.38). The result is

u = u, = 2 n H k l f i r (3.41)

w = -2nHkl15r (ma, - 1 + -)mz T2

m2-1
+ 0(1), (3.42)

(3.45)

t, = tzn
= - (q+o(fi). (3.46)
~ ~ ( Y I - Y Q~ i I Q2

Comparison of (3.41)-(3.46) with (3.13)-(3.18) indicates that they

become identical if one substitutes formally 8 by n - 8. This is easy to
explain. The punch problem is mathematically equivalent to an external
crack problem. There exists a notion that the asymptotic behavior of
stresses and displacements near the edge of an arbitary flat crack with a
smooth boundary is completely defined by three stress intensity factors.
This means that the asymptotics of an internal crack and an external one
should be the same. The system of local axes was introduced in previous
paragraph so that the angle 8 was measured from the direction outside
the penny-shaped crack. In the case of the punch problem the axis On
should be inverted inside the circle, this will make the expressions
(3.41)-(3.46) identical to (3.13)-(3.18).
In the limiting case of an isotropic body formulae (3.33)-(3.38)
190 V. I. Fabrikant

transform into
wei@
U =
n(1- Y)
[- ( 1 - 2Y)
P
(3.47)

(3.48)

- 1 y 2 [ 1 : - a’(2a’ - p’ + 2z2)]
u]= -
(a’
1 - Y ) [ ( 1 + 2 v ) 1; - 1:
n (2wp
+ ( a 2 - i:y(i;- 1 3 3 1) (3.49)
2wpe2‘@ (a2 - 1 y 2[a - (a’ -
a, = n(1- Y ) (-(1-2.)[ 122 - 1 2 1 -
P’

(3.50)

a, =
n(1- Y)
+ z2[1: + a2(p2
(a’ - 1:)’n - 2a2 - 2 2 ) ]
- i y ( i ;- 1 3 3
1. (3.51)

2wp zpe’@(a2- 1f)”’(31’, + 1: - 4a’)

t = - (3.52)
n(1- Y ) (1; - 1 3 3
The equivalent expressions for the components in polar coordinates will
take the form:

up =
n(1- Y)
[-(1-2Y)
a - (a’ - 1
1 4 ; - 1:)
P
y
+ z l l ( f ;- 1, (3.53)

a - ( a 2 - 1:)’” (a’ - 1;)’”

up= 2wp ( ( 1 - 2 Y ) -
n ( 1 - Y) P2 1; - 1:

(3.54)

cT@= - 2wp ( ( 1 - 2 Y ) a - (a’ - 1 y

n(1- Y) P’
(a’ - r:)l”[a’ - ( 1 - 2v)1;]
+ lg(1: - 1:) IJ (3.55)

2wp zp(a’ - 1~)‘”(31; 1;”- 4 4 +

tpz= - n ( 1 - Y) (1; - 1 3 3 (3.56)

The integral transform solution can be found in (Sneddon, 1951). It is in

agreement with ours, except for the missing factor p in the last expression
 Solutions to Some Boundary Value Problems 191

of (56, p. 462). In the examples to follow we shall consider several

nonaxisymmetric problems that have not been solved before.

3: INCLINED
C. EXAMPLE PUNCH
ON AN ELASTIC
HALF-SPACE

The case of a flat circular punch pressed against a tansversely isotropic

elastic half-space by a noncentrally applied force P can be considered as a
superposition of two: that of a centrally loaded punch, which was
considered earlier, and a punch subjected to a tilting moment A,which
will be considered here. Let the displacements under the punch be
o = b,y - byx. (3.57)
Here b, and by are the tilting angles about the axes Ox and Oy
respectively. Introduce the complex tilting angle
b = b, + iby. (3.58)
Expression (3.57) can be rewritten now as
~ ( p +)
, = 3{6pe'@}. (3.59)
Here 3 indicates the imaginary part. The substitution of (3.59) in (2.72)
yields

Evaluation of the potential function (2.67) leads to the integral

(3.61)

Taking into consideration that

the integral in (3.61) can be computed by parts, with the result given by
(AS). The potential function takes the form

All the necessary derivatives are readily available from Appendix A, and
192 V . I . Fabrikant

we can write the complete solution:

(3.63)

(3.64)

(3.65)

(3.67)

(3.68)

To the best of our knowledge, this problem has not been solved before.

D. EXAMPLE
4: PENNY-SHAPED
CRACKUNDER UNIFORM
SHEARLOADING

Consider a penny-shaped crack of radius a in a transversely isotropic

elastic space, subjected to a uniform shear loading t, where t is a
complex constant. The solution of the integro-differential equation (2.95)
in this case is
U = 2(GT- ”) t ( a 2 - p2)1’2
for z = 0 and p 5 a. (3.69)
G1
Substitution of (3.69) in (2.90) leads to the integral
 Solutions to Some Boundary Value Problems 193

which has been computed in Appendix A, with all the necessary

derivatives. The complete solution will take the form:

(3.71)

(3.72)

(3.73)

(3.74)

(3.75)

Here
2 . 2 - (1:
) e"@
f i 7 ( ~=
+ 2u2)(a2- 1;)'" (3.76)
9
3P2

(3.77)

(3.78)

(3.79)
194 V . I . Fabrikant

h l ( z ) = -sin-’( I,> + a(l’21; - 1:

a -a2)’~

’
(3.80)

(3.81)

A complete solution to this problem for the case of isotropy can be found
in Westmann (1965); the case of transverse isotropy does not seem to
have been considered before.
We can derive again some results of general nature, namely, the
asymptotic behavior of the field of stresses and displacements in the
neighborhood of the edge of a flat crack with a smooth boundary. We
recall that at @ = 0 the decompositions u = u, + iu, and t, = tzx + ityr are
equal to dn) +
= u, iu, and t(”) = t,, + iz, respectively; u1 is understood
as a, + a,, and u2= a, - a, + 2it,,,. This will allow us to avoid a
cumbersome axis transformation. The complex stress intensity factor,
introduced in ( 2 . 1 4 9 , can be expressed through the prescribed shear
loading t as

k=k2+ik3=- (3.82)
Jd

and its inversion gives

(3.83)

Substitution of (3.10) and (3.83) in (3.70)-(3.81) yields

(3.85)

(3.86)

(3.87)
 Solutions to Some Boundary Value Problems 195

(3.88)

(3.89)

By taking the sum and the difference of (3.86) and (3.87), one gets

(3.90)

+ 2nHA,(” + &)}) (3.91)

Qi Qz

(3.92)

Our formulae (3.84)-(3.92) are in relatively good agreement with similar

results of Kassir and Sih (1975), except for formula (8.96b, p. 371) for u,,
which should correspond to the imaginary part of our (3.84). Formula by
Kassir and Sih (8.96b) seems to be in error because it implies that u,
depends on k2 y l , and y2 which is wrong: our results relate u, to k3 and
) )

y3 only, as it should be. There are several misprints in their formulae

(8.96a) and (8.96~).The remaining formulae are in agreement, though
the formulae by Kassir and Sih (1975) look more complicated than ours,
mainly because they did not notice the properties (2.23), which could
make some expressions much simpler.
In the case of isotropy, formulae (3.70)-(3.75) transform into

(3.93)
196 V. I. Fabrikant

u1=
+
2 ( ~ e ' + re-'+)
[
-2(1+ Y)
all(a2- 1:)"2
K(2 - Y) 1 2 ( G - 1:)

4 1 $+ I:]
+ zfl(lz- ~ ~i,(i;
) " ~ [ a ~-(5p2)
-1 3 3
(3.95)

2ei+ a11(a2 - 1 y
a,= - t
K(2 - Y) 12(G- 6 )

+ zl& - a2)1/2
[$?e2'+ -
~ ' ( 4 1- +
; 5p2) 1;'
(lq - 1:)' (t+ Ze"@)]], (3.96)
12(C - 1:)

( 3 = -
+
2 ( ~ e ' + re-'+) zll(lz - ~ ~ ) ) " ~ [ a - ( 4 1 $+ 1 3
~ 5p2)
9 (3.97)
K(2 - Y) i2(i;- 1 3 3
zz =
K(2 - Y )
z(a2 - l:)ln[lt + a2(2a2+ 22' - 3p2)]
+ (1; - 1 3 3
z(a'- 1:)"2[a2(61$- 21: + p2)- 51!] l:e2'@
(1; - 1 f ) Z 1
l f ( l i- 1:)
(3.98)
The asymptotic behavior near the crack boundary for the case of isotropy
will take the form

k2 30
u z = T s i n Ocos-,
2 r 2

,z, + iz, =F
cos(0/2)
[[
1 - sin

The last set of formulae is in agreement with the results presented in Sih
and Liebowitz (1968).
 Solutions to Some Boundary Value Problems 197

The case of nonuniform shear loading can be treated in a similar

manner. For example, in the case when the loading is proportional to the
term pei9, the main potential function will be proportional to the integral

It might be a good exercise for the reader to do the differentiation and

write a complete solution. As an example, here are the first two
derivatives

It is noteworthy that dZ/dz is proportional to the main potential function

for the case of linear normal loading of a penny-shaped crack.

IV. Discussion

Here we discuss some immediate and future applications of the method

presented in this paper. Comparison is made also with some results
reported in the literature.
The expressions for the energy release rate can be obtained from
(3.46), (3.47), (3.88), and (3.89) by the procedure similar to the one
employed by Kassir and Sih (1968). The result is
= 2n2Hk:, Y& = 2~r~Hy,y,k;, & = 2n2pk:,
which is in agreement with similar results of Kassir and Sih (1968), and
gives an additional reason for the introduction of the parameters H
and p.
Investigation of various interactions (loads with cracks and punches,
between cracks, etc.) becomes readily available due to the new theory.
For example, consider the interaction of a concentrated load Po, applied
198 V . I . Fabrikant

at an arbitrary point ( p , Cp, 2 ) in the z-direction, with a flat circular punch

of radius a. One can deduce from (3.34) that the normal displacement w
at the point ( p , 4, z), due to a unit force applied to the punch, is

Application of the reciprocal theorem immediately gives the punch

settlement w due to the load Po:

The tilting angle 6 of the punch can be obtained in a similar manner from
(3.64). The result is

Some well-known results can be simplified significantly by using the

method of computation of various integrals involving distances between
several points. For example, here is the set of governing integral
equations derived by Panasiuk et al. (1986, p. 83) for the problem of
interaction of N coplanar thin spheroidal inclusions:

Here fn is a known function, wn is the normal displacement at the

boundary of the n-th inclusion, S, is its median crossection, (xn , yn) E S, ,
and the integral operator r is defined by

Here ak is the radius of median crosssection of the k-th inclusion,

 Solutions to Some Boundary Value Problems 199

indicates the area outside s k . The double and quadruple integrals in

(4.2), which is a kernel of the integral equation (4.1), make any
numerical solution next to impossible. Panasiuk et al. (1986) have
managed to give an approximate solution for the case when the inclusions
are far apart, which is of little practical value, since there is almost no
interaction at such distances between the inclusions. Let us show that
(4.1) can be simplified so that its kernel be presented in elementary
functions. Making use of the integral (compare with (C.31))
(a’ - pg)’” 1 rdrdq
- q ) [p’ + r’
+ r’ - 2p0r COS(@~ - 2pr cos(@- q)]“’

one can change the order of integration in the second integral of (4.2),
perform the integration in s k , and the integral operator r simplifies in
polar coordinates significantly, namely,

which is much simpler than (4.2). It is reminded that R and 7 are defined
by (2.29). We can also compute r[(x, - xk)’ + ( y , - y k ) 2 ] - 3 ’ 2 in elemen-
tary functions. Indeed, expression (C.19) in the particular case of 2-0
gives

- 2n(a2 - p2)’”
- for r > a.
+
(r2- a2)’”[r2 p2 - 2rp cos(@- q ) ]
The results above simplify (4.1) so significantly that now it can be easily
solved by iteration numerically or analytically. The reader is referred for
more detail to Fabrikant (1987~).A full investigation of this and other
interaction problems is beyond the scope of this article and will be
published separately.
A complete solution to the problem of a semi-infinite plane crack in an
isotropic body was obtained by Uflyand (1965), who used a very
complicated Kontorovich-Lebedev integral transform. To the best of our
knowledge, the same problem for a transversely isotropic solid was not
considered as yet. All these results can now be obtained as a limiting case
200 V . I. Fabrikant

of the penny-shaped crack solutions, when the radius a-,m, and the
coordinate origin moves from the circle centre to its boundary. No
integral transform is necessary. All the Green's functions are elementary.
Some of the integrals, involving special functions, can now be
computed just by comparison of the solutions obtained by the integral
transform method with the similar solution, given by the present method.
For example, comparison of (2.39) with formulae (1.42) and (1.43) from
Kassir and Sih (1975) leads to

We have verified numerically that the last formula is correct for

noninteger n as well. An enormous amount of material on Bessel
functions exists in the literature, so it is difficult to claim that the last result
is new, but there is some chance that it is new, since our notation f I and l2
does not seem to have been used before. Here is another example of just
how useful this notation is. Suppose that we need to compute the integral

(4.3)

We could not find this integral in the tables, but we have found another
one (Gradshtein and Ryzhik, 1963, formula 6.752.2)
l' sin axJl(px)e-'" a
dx = - ( 1 - r), (4.4)
X P
where the parameter r is a positive root of the equation

The original integral (4.3) can be computed by integration of both sides

of (4.4) with respect to z. We need to get an explicit expression of r from
the fourth order algebraic equation ( 4 . 3 , substitute the result in (4.4),
and integrate the result with respect to z, which does not seem possible at
first. The introduction of the parameters f1 and l2 allows us to find the
positive root of (4.5) in a very simple form, namely, r = (a'- ff)In/a.
The z-integration can be performed by using (B.3), and the final result is
lm sin ax J$p")e-'"
dx=
(2a2 - ff)(Zg - a')''' - 2a22 p
--
2 sin-'(
a
i;). (4.6)
2@P
 Solutions to Some Boundary Value Problems 201

It is of interest to indicate several solutions to the governing integral

equation (2.71) in the case of a circular punch that are equivalent to
(2.72). One such solution was obtained in Fabrikant (1986b):

Here R and q are defined by (2.29), and A is a two-dimensional Laplace

operator. Taking into consideration that (T can be expressed through the
limiting case of (2.79) for z 0, the following expression can be obtained
--f

This result was first derived by Leonov (1953). Another type of solution
can be expressed through the second z-derivative of (2.75) for z = 0.
Taking into consideration that for a harmonic function the second
z-derivative is equivalent to -A, the following result is valid:

2n a
+ (a’ -ap2)3” I, I, 1 - tf 4 P 0 , 90)
(1 - t)(l - f) (a’-

The last formula corresponds to the solution by Mossakovskii et al.

(1985).

V. Conclusion

The new method is proven capable of solving various non-axisymmetric

problems as easy as axisymmetric ones. Some of the problems solved are
202 V . I . Fabrikant

still beyond the reach of the existing methods, and many more new
solutions are to come. The fact that the new method does not use any
integral transforms or special function expansions makes it available for
use by the average engineer. It is believed that some of the results
presented are of fundamental value not only to elasticity but to any
branch of engineering science where potential theory is being used.

Appendix A

Here the main potential function is given, together with selected partial
derivatives. Define the potential function by

r2n r a

where Ro = [p’ + pg - 2pp0cos($ - $-),I + z2I1I2. The integral can be

computed in elementary functions:

”[ (
Y = - z 2a2- p’
2
+ -32 z2
11
In[/, + (If - p 2 ) 1 ” ] } .(A.l)

The following derivatives may be computed:

Solutions to Some Boundary Value Problems 203

(A.lO)

(A.ll)

(A.12)

(A.13)

(A. 14)

(A.15)

(A.16)

(A.17)

(A.18)

(A.19)
204 V . I. Fabrikant

a AY = 2n[ -sin-’(
-
az
E) + a(lf - a y ] ,
1: - 1:
(A.20)

a’ (a’ - 1;)“’
-AY = 2na’pe’’ (A.21)
3.2’ lf(1f- 1:) ’
(a’ - 1;)”’
AAY = -2na’pe‘@ (A.22)
lf(1f- 1;) ’
A3\v = -2ne”@ 4[(17+ 2a2)(a’ - 1;)ln - 2a3] + all(a’ - 1;)“’
{ 3P3 1d1f - 1:) 1 3

(A.23)
d4Y
-- za[Z‘: + a’(2a’ + 22’ - 3p2)]
a z 4 - --2n * (A.24)
(if- 1:)3(1f - a2)1/2
a4y
-- ll(lf - a y
[a2(41: - 5 ~ ’ +
) l‘:], (A.25)
dp dz3 - -2n i2(i;- 1 3 3
a2 zi+ ~ ’ ( 6 1-f 21: + p2) - 51;
-A’Y=2npe az (A.26)
a2 i;(i;- 1:)3(1; - a2)1/2 *
The following identities were used during the derivation of (A. 1-A.26):
111’ = up, 1: + i f = a’ + p’ + 22,
(1; - p’)1/2(1f - u’)l/’ = 2 1 2 , (a’ - l;)”’(p’ - 1:)’” = 211, (A.27)
(a’ - 1;)’”(1f - a’)1/2 = za, (1; - p’)l/’(p’ - 1 y = zp.
all -- 4
-= a12 212

dz I f - 1:’ az -1f-1;’
a11
-=-- a12 - pl1 - p(a2 - 1:) dl’ - pl’ - a11 -
_ - p ( l f - a’)
dp ff-1; ll(lf-ly ap If-1: l’(lf-1;) *

(A.28)

Appendix B

Here we present some indefinite integrals of expressions containing ll

and 1’.

I (1; - dz = (a’ - 1 1 )
2 I/’ 1; - 2a2 + -ln[lz
-
2a
P2
2
+ (1; - p2)1R], (B.l)

I(1; - a’)l”l: dz = -a(a’ - 1,)

2 I/’ 1: + 2a2 + a’p’ In[/, + ( I f - p2)1R],
-
3
 Solutions to Some Boundary Value Problems 205

I (a’ - lf)’Dl:dz = -
1321: + 3p’) (1: - a y 2
8a

I t) sin-’( dz = z sin-’( ): - (a’ - 1:)’” + a In[l, + (1; - p’)’/’],

(B * 8)
I ):z sin-’( dz = -1 (2a’
4
+ 22’ + p’) sin-’(:) + (1; - a 2)‘I2 2a2 + 1:
4 a ’

I t)
z’ sin-’(
1
dz = 3 z 3sin-’(
a
I,) 1
+-
18
(a’ - 1:)’”(31; + 6p’ + 8a’ - 21:)

1
- - a(3p’
6
+ 2a’) 1n[1, + (1; - p’)’/’]. (B.lO)

The integration in (B.1)-(B.10) was performed by parts, with a con-

sequent change of variables: z = (a2 - l:)’R(p2 - l:)’D/ll or z =
(1; - a’)’/’(/; - p2)’R/12.
I p sin-’( E) y
dp = P’ sin-’( i)+ z(2a’ - 1:)
2(a’ - l y ’ ’
(B.ll)

I p’ sin-’( t) P3 sin-’( a
dp = -
3
G ) +Zp(242-l:)
6(a2 - 1:)”’
206 V. I . Fabrikant

a sin-’( %)da [( 2 2 + 22’

= - p’) sin-’
):(
+ Il(p’ - If)”’- 2z(a2 - If)’”
1. (B.13)

The integration in (B.ll) and (B.12) was performed by parts, with a

consequent change of variables: p = y[l + z 2 / ( a 2-y’)]’/’, which cor-
responds to a substitution 1’ = y. A similar remark is valid for formula
(B. 13).

Appendix C

Several integrals, used in this article, are computed here. Consider

three points in the system of cylindrical coordinates, namely, M ( p , @, z ) ,
Ma(po, &, zo), and N ( r , w, 0). The following notation is introduced:
I&) =i{[(p + t)’ + 2’11/’ - [ ( p - t)’ + z’]’/’}, (C. 1)
l’(t) = i { [ ( p + t)’ + 2 2 y + [ ( p - t)’ + ZZ]’”}, (C.2)
I&) + t)’ + z;y2 - [(pa - t)2 +
= ;{[(Po 2;]”2}, (C.3)
Iza(t)= ${[(Po+ t)’ + ,;]I/’ + [(pa- t)’ + z;]’”}. (C.4)
According to the earlier convention, llo stands as an abbreviation for
Il0(a), etc.; R ( . , .) denotes the distance between two points.
Consider the following integral:
1 h0
p’[ Z
R 3 ( M , N ) R ( N , Ma) tan-’[ R ( N , M,)]‘ dr ‘w,
(c‘5)

where
ho= (a’ - Ilo)’n(a’ - r’)lB/a. (C.6)
Make use of the integral representation (Fabrikant, 1986b):

where A(., .) is defined by (2.26). The substitution of (C.7) in (C.5)

 Solutions to Some Boundary Value Problems 207

yields, after changing the order of integration,

By substituting the integral representation (D.6)

z - z
R3(MJN) = [p’ + r’ - 2rp cos($ - 3 ) + z2I3/’

in (C.8), the following result can be obtained, after integration with

respect to 3 :

[Izo(x)- pi]’” (x’ - r2)’l2

Here the following property of the L-operators was used:

The well-known property of the Abel-type operators, namely,

(C. 12)

allows us to simplify (C.10) significantly:

It is noteworthy that the integrand in (C.13) is symmetric with respect to

the points M and Mo while it did not look so in the original expression
(C.5). The integrand in (C.13) is a perfect differential, so that the integral
can be computed as indefinite:

(C. 14)
208 V. I. Fubrikunt

where
R1= [p’ + pX - 2ppo COS($ - $0) + ( Z - ~ o ) ’ ] ’ ~ ’ ,
RZ = [p’ + p; - 2ppo COS($- $0) + ( Z + ~ o ) ~ ] ~ ’ ’ ,
(C.15)

Correctness of the integral in (C.14) can be verified by differentiation.

The algebra involved is not trivial. Here we present some intermediary
transformations:

X
1
[R: +@(x)
+R; + @(x) 1. (C.17)

Formula (C.14) allows to to compute the integral (C.5):

(C.18)
where the contractions 0’and 0’stand for @,(a) and @,(a) respectively.
Note an important particular case when z0 = 0. Formula (C.18) in this
case transforms into

(C.19)

and the integral vanishes when p o 2 u . Here the point No has the
cylindrical coordinates (po, $o, 0), and h is defined by (2.33).
 Solutions to Some Boundary Value Problems 209

The second fundamental integral to be considered is:

where ho is defined by (C.6). Make use of the integral representation for

the reciprocal distance

‘ dx [ I ; @ ) --x2I1/2A(”””, @ - w), (C.21)

(r2- x2)“’ I f ( x ) - l:(x) 12(x)r
which is a variation of the result established in Fabrikant (1986b). By
substituting (C.21) in (C.20), the following expression results, after
changing the order of integration:

(C.22)
We use the integral representation (Fabrikant, 1986b):

-
- L(r) d
r dr I, tdt [ l f o ( t ) -t2]’”
(t2- r2 )112 lz0(t)
2 - Ifo(t)

X A(-, $0- w) . (C.23)

The substitution of (C.23) in (C.22) yields, after integration with respect

to q:

(C.24)

We recall another well-known property of the Abel operators:

(C.25)
210 V. I. Fabrikant

Application of (C.25) to (C.24) yields

Note certain similarity between (C.26) and (C.13). The integrand in

(C.26) is a perfect differential and can be computed in elementary
functions:

(C.27)

where R1 and R2 are defined by (C.15) and

As before, correctness of the integration can be verified by

differentiation, using (C. 17) and the property

Finally, the integral (C.20) can be computed as follows:

(R($M,) )1 ig;:)
= Id H{-1 [-n - + 1 [n - tan-'(:)]]. (C.30)
zo R1 2
According to our convention, El and E2 denote =,(a) and =,(a)
respectively.
Consider a particular case, when zo=O and po>a. Due to the
relationship (see (A.27))
 Solutions to Some Boundary Value Problems 21 1

the integral (C.30) will take the form

1
(a’ - r2)l/’ p i + r2 - 2p0r cos( @o - q )
rdrdq
X
+
[p2 r2 - 2pr cos($ - q ) + 22]1/2

Here R,, = [p’ + pg - 2pp0 cos(@- Go) + ~ ~ 1 The ~ ’ integration

~ . in (C.30)
for z, = 0 and po < a yields n 2 / R o .The case of z = 0 can be considered in
a similar manner.

Appendix D

An integral representation for z / R 3 is presented here, for R = [p’ +

r2 - 2pr cos @ + ~ ~ 1 Consider
~ ’ ~ . the integral

By using the rule of differentiation under the integral sign

the integral (D.l) can be rewritten as

Introduce the new variable

(r2- x 2)1/2[12(x)
2
-
t= 9
X

2 ] 1/2[r212(x)
t ’ = -dt= - [lz(x)- x 2 -x21:(x)]
dx x2[1:(x)- 1:(x)](r2- x2)’/’
212 V . I . Fabrikant

Let us transform the expression for A:

I:(x) - p2r2/IT(x)

&) - p2r2/f:(x)
=-
&) + p2r2/I:(x)- 2pr cos + + +
p2 r2 z2 - I:(x) - & x ) + x 2 - r2
x21:(x) - r2&) [&) - I:(x)](r2- x2)'l2 t'
= - = - 03-41
x2(R2 t2) + [I~(x) - x 2 ] l n R 2 t2 ' +
The substitution of ( D . 3 ) and ( D . 4 ) in (D.2) allows us to continue the
transformations:

dx d (r2-x2)3nt'
= -'[ [
( r 2- x ~ ) " ~ , xt2(R2+ t2) 1
= -z(
(r2- x2)t' -
xt2(R2+ t2)
;1 jr
dt(x)
, t2(R2+ t2)
(r2-x2)tr 1 1 n z
= -z(
+ +
xt2(R2 t2) -
R2t -tan-'
R3 + (D.5)

Finally, (D.5) allows us to write the required representation

Appendix E

The following integral is computed here:

(a2- r 2)In(a2 - p$)'I2 r d r d q
,I' I,
2n
2n ope'@- rei*
R3(M, N )
tan-'[
a R W , N,) 1
R ( N , N,) ' (E'l)
The integral ( E . l ) can be computed indirectly by using (2.32), which
leads to an equivalent expression:
 Solutions to Some Boundary Value Problems 213

Let us make use of the following identities:

hpei@
M = --
If - 1: '

--
pe'4J - p0ei@O

Ri
tan-'(&) -~
h
-
[
R $ + h 2 l2-1'
riel
pei@- poei@"]
+

R;
(E.4)
The notation Ro in this appendix is used as a contraction for R ( M , No).
The substitution of (E.4) in (E.2) yields, after integration by parts,

=-
pe-W
1
- poe-'@O { tan-'( &) - [~0z d [tan-'( &)]
dz]

The following identities are now used:

dh - h(p2 - 1:)
--
dz Z ( l $ - 1:) '

The substitution of (E.7) in (E.5) allows us to proceed:

hdz
-I, Z2 + pel@- poei@o]
214 V. I. Fabrikant

(E.lO)

where I = (a2- ppoe-i(@-@n))l”. Finally, formulae (E.8) and (E.lO) allow

us to compute the original integral (E.1):
-
(a’ - r2)112(a2 r dr d q
& 6 z’;M,
- reiY
N)
tan-’[
a R W , No) 1 R W , NO)

(E.ll)

It is reminded that h is defined by (2.33), and R(M, No) = [p2 p; - +

2ppocos($ - $0) + ~ ~ 1 ’ ’The
~ . right hand side in (E.ll) simplifies in the
limiting case of p o - j p and $o+ $, namely,
 Solutions to Some Boundary Value Problems 215

Appendix F

Some integrals, related to elastic contact problems, are computed here.

The first integral to compute is (2.73)

where, according to (2.72),

1 d tdt
o(r, v )= - -L ( r )-
n2Hr dr I, (I

(t2- r

By substituting the integral representation for the reciprocal distance

(C.21) in ( F . l ) , we get, after changing the order of integration

We substitute (F.2) in (F.3) and use the property ( C . l l ) to obtain

The property (C.25) allows us to simplify (F.4) as follows

One can establish the following rule of changing the order of integration:

The rule (F.6), applied to ( F S ) , yields

216 V. I. Fabrikant

Now we need to compute the integral in curly brackets (F.7). Introduce a

new variable
[x' - I:(x)]'"(x2 - pi)1/2
h(x) = 9
Y

Formulae (F.8) may be considered as a particular case of (C.15) and

(C.16) for zo = 0 and x r po. By using the analogy with (D.4), one can
write

+
Here Ro = [p' pi - 2pp0 cos(4 - @o) + ~ ~ 1 Equations
~ ' ~ . (F.8) and (F.9)
justify the following identity:
z(x2 - pi)3/2h'(x)

xh2(x)[Ri+ h2(x)]
. (F.lO)

We can now compute the integral in (F.7).

d " [@x) -x 2 y 2
-I
dP0 PO 1;w - ax)
(1; - a2)1/2po " d
XdX

x d
=- a(a2- pol 2 - 1:)
2 112(12 + P O ] Po ( x 2 - pi)1/2 Z
X 9 - lu)]
(1; - a2)1'2po " du d z ( x 2 - p@3/2h'(x)
=-
2 1/2 2
a(a2 - Po) (12 - I:) + Po ] 2 112-
P" (x' - Po) dx xh2(x)[Ri + h2(x)1[
The last integral can be computed by parts:

i, .I
- (1: - a2)lnp, z(x2 - p i ) h ' ( x ) " h'(x)dx
2 112 2
a(a2 - Po) (12 - 1:)
+
+
xh2(x)[Ri h2(x)] +
p" h2(x)[Ri + h 2 W

-- z ( x 2 - p$)h'(x)
xhZ(x)[Ri+ h2(x)]

(F.ll)
 Solutions to Some Boundary Value Problems 217

Computation of the upper and lower limits in (F.ll) gives the final result
~(r, v)rdr d v

According to our convention, h stands as an abbreviation for h ( a ) .

Expression (F.12) validates (2.74).
The second integral to be computed is

(F. 13)

We proceed with integration by parts. The result is

1 dz d tan-'($). (F. 14)

We modify (E.7):
d Ro h(P2- 1:)
-tan-'($,
dz
=
[
R i + h2 z(1: - 1:)
~
+

i1
- z
hRo

(F.15)

By substituting (F.15) in (F.14), and taking into consideration (E.9) and

(A.28), we obtain
1
- -tan-'(
RO
&) + (a2
1
-p y 2

The integral in (F.16) is elementary:

+ (1: - P2)1121

'> 5 .+O). (F.17)

218 V. I . Fabrikant

Since the integration was indefinite, we might have lost a function of the
variables, other than z. This function can be found from the condition
that the result of the integration does not have a logarithmic singularity at
p = 0. The function eliminating such a singularity i s tan-'( - 1)"'. Now
the final result can be represented by

The last expression proves correctness of formula (2.76).

Appendix G

Some integrals related to the problem of a penny-shaped crack under

shear loading are presented here, without derivation. The first integral,
which can be computed directly, is

r dr d q

(a2- (l; - a2)'/2[f:(4 - t) - 3a2]

=Jt
a3 {$sin-'(:) -+ a(1 - t ) 2
+ 1 3z2
"1
-+ a2(3 - 2t) - - tan-'
( 1 - t)3" [l - t t
((1; -
- ')li2)}. (GJ)

Here t is defined by (2.104). Application of the operator A to both sides

of (G .1 ) yields

= -2~tpe'+(a2 - p
a3
y[3sin-'( E) + (1 - t)(/fa(lz--a2)'"
ppoe'(+-Q'O))
- 1 tan-'( a(1 - t)'" )].
t( 1- t)3/2 (/I
- a')'''
 Solutions to Some Boundary Value Problems 219

Another application of A to both sides of ( G . 2 ) gives

2% a (302 - rpoe'(V-@d)z(a2 - r2)1/2(02 - pg)1/2

I, I, (a2-rpoe
i(W-@a))2

t
a ~ 3 (

3(Z$ - a 2 y 2tan-'(
N~ ), rdrdq

S )]. (G.4)
s3 (1: - a2)1/2

Here h is defined by (2.33). Another differentiation of both sides of (G.4)

s2
with respect to z yields
- rpoei('-@o))(a2- r2)lI2(a2- pg)'"
- rPoei(V-@a) 12 aR3(M, N )
3z2
( l - R2(M, N )
) r dr d q

a(l2,- a2)1/2
= 21d
(1; - /:)(if - ppoei(@-@a))
+ lTt - 3p2) a(1 - t)l"
+ ppoei(@-@0)(21$
1; - PPoe'(@-@o) 1 3
- (1 - t)3D

(G.5)
Application of the operator A to both sides of (G.4) yields

3(pe" - reiV)rdr d q

hpei@(31f- a2t)
= 21d (G.6)
(19 - l:)(l; -
A different result appears if A is applied to a complex conjugate of
expression (G.4), namely,
(3a2 - r P o e - i ( ~ - @ o ) ) z(a2- r2)'l2(a2-
3(pe'@- reiV)rdr d q
(a' - rpoe-i(q-@"))2 &(M, N )
= 21dh - [ z2
-2 Po
[ ;u ~ )tan-'(
ei@o 15(g ~ ' ~
(12, -
5 ) -?15 ay2
25 pe'@(31$- a2t)
+
5
5y1: - a?)
+ (r; - a t ) 2- 2 1 +
(1: - f;)(f; - a 2 i ) 2
220 V . I. Fabrikant

Integration of both sides of ( G . l ) with respect to z gives

I
= n(a2- p;)'" 2 in[/, + (if - p2)ll2]-2

+ Z

Here f is defined by (2.76), and the bar, as usually, indicates the complex
conjugate value. A similar integration with respect to z of (G.3) yields

X {In[R(M, N ) + z ] } rdr d v

+-ei+ [ (a2;/:)la
P
I). (G.9)

It is reminded that q is defined by (2.43) and that

N )+z
A2ln[R(M, N ) + z ] = - ( p e i b - reiv)2R 3 ( M2R(M,
,N ) [ R ( M , N ) + 212 *
(G.lO)
 Solutions to Some Boundary Value Problem 221

And yet another application of the operator A to (G.9) gives

2n a (3a’ - rpoe’(V-+O) (a‘ - ,.‘)‘/’(a’ -
p31”A3{ln[R(M, N) + z ] } r d r d $ ~
I, I, -T
- rpoeitV-+~)2
(a’ 1
2n (a’ - p y [ 3 ( f i l ) ” ’ [tan-’( 1 )
- tan-l(( a’ - 1 y )I
4 ( f - 1)’“ a( f - 1)”2
e’id(a2 - l:)’Q 1f+P2 2p2(l; - a’)
-
a(/$- I:) [ +

lz - ppoe‘(+-+o) (1: - ppoei(+-+o))2 + l I

+-ei@[3 +2:
I+--
(a’ - l:)’O (
P 9 a
(G.ll)
Here
A3 In[R(M, N) + z] = (pe’@- re ;,,,) 8R2(M, N) + 9R(M, N)z + 32’
R3(M, N)[r(m, n) + 21’
(G.12)
Formula (2.42) can be used to obtain some additional results. Integration
of (2.42) with respect to z gives
- p$l” r dr d$J
(a’ - r2)1/2(a2
tan-’(
aR(N, No) R(N, NO)

= ?!
(I
[Rotan-’(&)Ro - (a’- pi)’”[ f tan-’( )
(If -S-a’)’/’

- ( 4 - l)’”(tan-’( ( f - 1)lI‘ ) tan-’( a(”( f -‘””))]].

- - 1)’Q (G.13)

The following indefinite integrals were used here

I tan-’( S-
(1’2 - a’)’“
) dz = z tan-’( -
S
(1: - a’)1/’
)+ S[ In[/’ + (lf - p’)”’]

+ (5; - 1)”’ tan-’

I “)
Ro tan-’( Ro dz = Rotan-’( $) - (a’ - p;)l/’
[-1n[1, + (lf -p2)’~l

+ ( f - I)”’ tan-’
(a’ - 1:)“’
+ (5‘ - 1)”’ tan-’ (G. 14)
222 V. I. Fabrikant

By applying the operator A to both sides of (G.13), one gets

(pei@-reiv)2[2R(M,N ) z ] + (a' - r2)1/2(a2- pi)1n r dr dq!J

R3(M, N ) [ R ( M , N ) + zI2
tan-'(
aR(N, NO) )
R ( N , NO)

x tan-(
tan-(

5
") +
RO
)+
(a2 - pi)"'[ z poei@O
( 7- a)
-tan-'((
(1: - a2)1/2
a2- - 12)'
1 "

a( I; - 1)'n
1) +$I - 7 1 .
~

ei@ha2
(G.15)

And yet another application of A to (G.15) yields

(pei@- rei"')3[8R2(M, N) + 9 R ( M , N ) z + 3z2]
R 5 ( M , N ) [ R ( M ,N ) + z]'
(a2- r2)ln(a2- p;)'" r dr dq!~
tan-'( a R ( N , NO) R ( N , NO)

Acknowledgment

The reported research was supported by a grant from the Natural

Sciences and Engineering Research Council of Canada.
 Solutions to Some Boundary Value Problems 223

References

Collins, W. D. (1962). Some axially symmetric stress distributions in elastic solids

containing penny-shaped cracks. 1. Cracks in an infinite solid and a thick plate. Proc.
Roy. SOC.London Ser. A 266, 359-386.
Cherepanov, G. P. (1974). “Mechanics of Brittle Fracture.” Nauka, Moscow (in Russian).
English translation: McGraw-Hill (1979).
Elliott, H. A. (1948). Three-dimensional stress distributions in hexagonal aeolotropic
crystals. Proc. Cambridge Philos. SOC.44, 522-533.
Elliott, H. A. (1949). Axial symmetric stress distributions in aeolotropic hexagonal crystals.
The problem of the plane and related problems. Proc. Cambridge Philos. SOC. 45,
621-630.
Fabrikant, V. I. (1986a). Inverse crack problem in elasticity. Acta Mech. 61, 29-36.
Fabrikant, V. I. (1986b). A new approach to some problems in potential theory. 2. Angew.
Math. Mech. 66, 363-368.
Fabrikant, V. I. (1986~).Flat punch of arbitrary shape on an elastic half-space. Internat. J.
Engineering Sci. 24, 1731-1740.
Fabrikant, V. I. (1986d). Inclined flat punch of arbitrary shape on an elastic half-space.
ASME J. of App. Mech. 53,798-806.
Fabrikant, V. I. (1987a). Flat crack of arbitrary shape in elastic body: Analytical approach.
Philos. Magazine A 56, 175-189.
Fabrikant, V. I. (1987b). Penny-shaped crack revisited: Closed form solutions. Philos.
Magazine A 56, 191-207.
Fabrikant, V. I. (1987~).Close interaction of coplanar circular cracks in an elastic medium.
Acta Mech. 67, 39-59.
Fabrikant, V. I. (1987d). Frictionless elastic contact problem for a curved rigid punch of
arbitrary shape. Acta Mech. 67, 1-25.
Gradshteyn, I. S., and Ryzhik, I. M. (1963). “Tables of Integrals, Series and Products.”
Moscow. English translation: Academic Press, New York (1965).
Kassir, M. K., and Sih, G. (1968). Three-dimensional stresses around elliptical cracks in
transversely isotropic solids. Engineering Fracture Mech. 1, 327-345.
Kassir, M. K., and Sih, G. (1975). “Three-Dimensional Crack Problems.” Noordhoff
International, Leyden.
Leonov, M. Ya. (1953). General problem of a circular punch pressed against an elastic
half-space. Prikl. Mat. Mekh. 17, 87-98 (in Russian).
Mossakovskii, V. I., Kachalovskaya, N. E., and Golikova, S. S. (1985). “Contact Problems
of the Mathematical Theory of Elasticity.” Naukova Dumka, Kiev (in Russian).
Panasiuk, V. V., Stadnik, M. M., and Silovaniuk, V. P. (1986). “Stress Concentration in
Three-Dimensional Bodies with Thin Inclusions.” Naukova Dumka, Kiev (in Russian).
Sih, G. C., and Liebowitz, H. (1968). Mathematical theories of brittle fracture. In
“Fracture”, Vol. 2 Academic Press, New York.
Sneddon, I. N. (1951). Fourier Transforms.” McGraw-Hill, New York.
Uflyand, Ya. S. (1965). Survey of articles on the application of integral transforms in the
theory of elasticity. North Carolina State University, Applied Mathematics Research
Group, File No. PSR-24/6.
Uflyand, Ya. S. (1977). “Method of Coupled Equations in Mathematical Physics Prob-
lems.” Academy of Sciences of the U.S.S.R., Nauka, Leningrad (in Russian).
Westmann, R. A. (1965). Asymmetric mixed boundary-value problem of the elastic
half-space. J . Appl. Mech. 32, 411-417.
This Page Intentionally Left Blank
 ADVANCES IN APPLIED MECHANICS, VOLUME 27

Convective and Radiative Heat Transfer in

Porous Media
CHANG-LIN TIEN?

Department of Mechanical Engineering

University of California,
Berkeley, California

and

KAMBIZ VAFAI

Department of Mechanical Engineering

Ohio State University
Columbus, Ohio

I. Introduction.. ... ... ... .. ... .. ... . .. ... .... ... .. . . ... .. ... ... . .. . , 226
11. Forced Convection in Porous Media . .. . ... . .. . ... ... . .. . .. ... . .. .... . . .. . 228
A. Flow over External Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
B. ConfinedFlows ........ ... ..... ............................. ..... 233
111. Natural Convection in Porous Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
A. Natural Convection over External Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . 236
B. Natural Convection in Confined Structures . . . . . . . . . . . . . . . . . . . . . . . 243
IV. Multiphase Transport in Porous Media . . . . . . . . . . . . . . .................. 252
A. Theoretical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . , . . . .. . . . . . 252
B. Application Areas ................................................... 255
V. Radiative Heat Transfer in Porous Beds . . . . . . . . . . . . ................... 260
A. Thermal Radiation Characteristics of Porous Beds . . . . . . . . . . . . . . . . . . . . . . 260
B. Modeling of Radiative Transfer in Porous Beds . . . . . . . . . . . . . . . . . . . . . . . . . 265
C. Multimode Heat Transfer in Porous Beds . . . . . . . . . . . . . . . . , . . . . . . . . . . . . . 270
Acknowledgments . . . . . . . . . . . . . . . ................................... 27 1
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 1

t Present affiliation: University of California, Irvine.

225
Copyright 0 1990 Academic Press, Inc.
All rights of reproduction in any form reserved.
ISBN 012UMU27-0
226 Chang-Lin Tien and Kambiz Vafai

I. Introduction

Convective and radiative heat transfer and multiphase transport

processes in porous media, both with or without phase change, have
gained extensive attention in recent years. This is due to the wide range
of applicability of these research areas in contemporary technology.
These applications include, but are not restricted to, areas such as
geothermal engineering, building thermal insulation, chemical catalytic
reactors, packed cryogenic microsphere insulations, petroleum reservoirs,
direct-contact heat exchangers, coal combustors, nuclear waste re-
positories, and heat pipe technology.
The reviews by Cheng (1978a) and Combarnous and Bories (1975)
provide some well-thought views on a number of different issues in heat
transfer in fluid-saturated porous media. The present work aims at
providing a comprehensive review that will cover most of the aspects of
interest in convective and radiative heat transfer and multiphase trans-
port in porous media.
Several applications related to porous media require a detailed analysis
of convective heat transfer in different geometrical shapes, orientations,
and configurations. Based on the specific application, the flow in the
porous medium may be internal or external. Most of the studies in
porous media carried out until now are based on the Darcy flow model,
which in turn is based on the assumption of creeping flow through an
infinitely extended uniform medium. However, it is now generally
recognized that the non-Darcian effects are quite important for certain
applications. Different models have been introduced for studying and
accounting for such non-Darcian effects as the inertial, boundary, and
variable porosity effects. The ultimate goal of studies in convective heat
transfer in porous media is to determine the dimensionless heat transfer
coefficient, the Nusselt number. A considerable amount of research has
been carried out to accomplish this, and empirical correlations for the
Nusselt number for a variety of configurations and boundary conditions
have been established, with certain limitations.
In keeping with the evolution of the non-Darcian model, it has become
increasingly important to consider and study the effects of multiphase
transport and phase change processes in porous media. Such an analysis
has become, in recent years, a more pronounced area of research. A
survey of the work performed in this area reveals several simplifying
assumptions that are commonly made in the modeling of multiphase
 Convective and Radiative Heat Transfer in Porous Media 227

transport phenomena. These include assumptions of constant transport

coefficients, absence of noncondensible gas effects, Darcian flow in the
porous matrix, and local thermodynamic equilibrium between the phases
of the porous system. It should be noted that such simplifying assump-
tions may be used for a limited number of phenomena, such as the drying
of different porous materials and moisture migration due to a tempera-
ture gradient in a porous material, without significant loss of accuracy.
However, for certain applications the assumption of Darcian flow and
local thermodynamic equilibrium will not be valid. At present there is a
lack of multiphase transport models for porous media that account for
the non-Darcian effects in momentum transfer, and also a lack of any
comprehensive analysis for multiphase transport in the absence of local
thermodynamic equilibrium.
Some modern technologies and industrial processes utilizing porous
and fluidized beds of solid particles operate at temperatures high enough
for thermal radiation to be a significant mechanism compared to
convective heat transfer (Flamant and Arnaud, 1984; Saxena et al.,
1978). Some examples are coal combustors, chemical reactors, and
nuclear fuel rods. Other types of packed beds in which thermal radiation
is important, even though the temperatures may not be high, are those
where other modes of heat transfer have been suppressed, such as packed
cryogenic microsphere insulations (Tien and Cunnington, 1973). For
these reasons, radiative heat transfer is considered to be a significant
mode of heat transfer for these types of porous systems.
The present work delivers a comprehensive and updated review of the
research performed in the aforementioned areas and serves as a conduit
for better understanding of the current research directions for convective
and radiative heat transfer as well as multiphase transport processes in
porous media. While providing a general and useful guide for future
research, this work primarily deals with the problems that have engineer-
ing applications. To make this work a convenient reference for research-
ers, an attempt has been made to, as much as possible, subdivide and
categorize the pertinent research areas and the available literature.
Furthermore, by concentrating on the infrastructure of each research
area and minimizing the peripheral information to the bare essentials, it
is hoped that this exposition will serve as an objective, accurate, and
readily accessible source of information for researchers in these areas. To
this end, a conscious effort has been made to avoid duplicating equation
after equation but instead concentrate on more physical and precise
228 Chang-Lin Tien and Kambiz Vafai

presentation with a practical point of view. However, every effort was

made to provide the reader with a precise path to the most pertinent
peripheral information related to a particular infrastructure. In what
follows, the aforementioned subjects are divided into four sections for
convenience. These are
11. Forced convection in porous media,
111. Natural convection in porous media,
IV. Multiphase transport in porous media,
V. Radiative heat transfer in porous and fluidized beds.
In Sections I1 and I11 the subject areas were well suited to be classified
according to their geometric configurations as shown later in Figures 1-3.

11. Forced Convection in Porous Media

A. FLOW OVER EXTERNAL

BOUNDARIES

Forced convection over external boundaries in the presence of a

porous medium (Figure 1)constitutes a very important subject area. This
is due to the very fundamental and generic nature of this type of problem
which makes it pertinent to a wide variety of applications.
Consider the flow over a semi-infinite flat plate first. Analysis of
convective heat transfer from this type of external boundary embedded in
a porous medium has many important applications, e.g., drying of
agricultural products, geothermal engineering, etc. Cheng (1977) and
Bejan (1984) documented the local Nusselt number for forced convection
over a constant temperature solid wall as
Nu, = 0.504 Pein, (2- 1)
where Nu,=hx/k, Pe,= U,X/(Y,h the heat transfer coefficient, k the
thermal conductivity, U, the free-stream velocity, and (Y is the thermal
diffusivity. They also documented the local Nusselt number for a solid
wall with constant heat flux as
Nu, = 0.886 Pri'*, (2.2)
where Nu, = qx/k(T, - T-), q is the supplied heat flux, and T, is the
temperature of the solid boundary. These two equations are valid if
Pe,>> 1, that is, when the longitudinal heat transfer is dominated by
 Convective and Radiative Heat Transfer in Porous Media 229

u,Tm -
- h
-c

(d2)
FIG. 1. Geometric configuration for forced convection heat transfer.
(a) Flat plate.
(b) Horizontal cylinder or sphere.
(c) Horizontal line source or point source.
(dl) Interface region between two different porous media.
(d2) Interface region between a fluid region and a porous medium.

convection. These results are based on Darcy’s law, which neglects the
effects of a solid boundary, inertia forces, and variable porosity on fluid
flow and heat transfer through porous media.
However, in many cases of practical interest, it becomes quite
important to consider the non-Darcian effects. In such cases, the fluid
velocity is high, the porous medium is bounded, and the porosity is
variable, thus rendering the Darcy flow model ineffective. Vafai and Tien
(1981) analyzed the effects of a solid boundary and inertial forces on
forced convection through constant porosity media, after establishing the
governing equations by local volume-averaging technique. The governing
equations established by Vafai and Tien (1981) and used in their analysis
230 Chang-Lin Tien and Kambiz Vafai

were
v . ( v ) =o, (2.3)
( P f / W V * V ) V ) = - V ( P ) ' + (Clf/S)V"V) - (Clf/K)W

-ptFS1'2y[(v) * ( v ) ] J , (2 4)
( ( v )* V ) ( T ) = a , V * ( T ) , (2.5)
where v represents the velocity vector, ( ) denotes the local volume
average of a quantity, pf the fluid density, S the porosity, K the
permeability, T the temperature, J a unit vector oriented along the
velocity vector, (P)' the intrinsic phase average of the pressure, p f the
fluid viscosity, y = (S/K)"' the porous medium shape parameter, and F
an empirical function which depends primarily on the microstructure of
the porous medium.
It was shown that for the flow field the boundary effect is confined
within a thin momentum boundary layer which often plays an insig-
nificant role in the overall flow consideration. The effect of the boundary
on heat transfer, however, was shown to be quite important and was
more pronounced for the thermal boundary layer with a thickness less
than or of the same order as that of the momentum boundary layer.
Recently, Kaviany (1987) obtained Karman-Pohlhausen solutions for
forced convection from a semi-infinite flat plate embedded in a porous
medium on the basis of the Brinkman-Forschheimer-extended equation.
These boundary and inertia effects were numerically and experimentally
investigated by Vafai and Tien (1982) for transient mass transfer in the
vicinity of an external impermeable boundary. It should be mentioned
that the concepts and the general results obtained from these studies are
equally valid for forced as well as natural convection over an external
impermeable boundary in the vicinity of a porous medium.
In many applications such as fixed-bed catalytic reactors, packed bed
heat exchangers, drying, chemical reaction engineering, and metal
processing, the constant-porosity assumption does not hold because of
the influence of an impermeable boundary. Therefore, there is a need to
focus on the variable-porosity effects on forced convection in the vicinity
of an impermeable boundary. Vafai (1984,1986) presented in investiga-
tion of the channeling effect and its influence on heat transfer and fluid
flow through variable-porosity media using the method of matched
asymptotic expansions. The variable porosity effects were shown to be
important for most heat transfer problems and the theoretical results
 Convective and Radiative Heat Transfer in Porous Media 231

were found to be in good agreement with the available experimental

data. Furthermore, Vafai, Alkire, and Tien (1985) performed an
experimental investigation on the effects of a solid impermeable bound-
ary and variable porosity on forced convection in porous media. Their
results confirm the qualitative and to a lesser degree the quantitative
agreement between the experimental and the theoretical results that
account for the variable porosity effects.
For flow over horizontal cylinders and spheres Cheng (1982) obtained
similarity solutions using the thermal boundary layer approximation for
forced convection around these types of geometries embedded in a
porous medium. The local Nusselt numbers for an isothermal cylinder
and an isothermal sphere are given respectively by
Nu, = 0.564 P e r m sin( 1 - cos Q)-”*, (2.6)

Nu, = 0,564 Pe? (2.7)

where Nu, =hroQ/k and Pe, = UwroQla with Q denoting the angle
measured from the stagnation point and ro denoting the radius of the
cylinder or the sphere. These results are based on Darcy’s law as the
governing momentum equation and are valid when Pe, >> 1.
For the cases of a horizontal line source and a point source placed in a
uniform flow Bejan (1984) presented the resultant temperature distribu-
tions. Also Cheng and Zheng (1985) have reported the inertia and
thermal dispersion effects on flow and temperature fields for a horizontal
line heat source in a porous medium.
Another configuration of engineering importance is the interface
region between a porous medium and another medium. In general, the
other medium could be a fluid, a solid or another porous medium. A
specific example of the interface region can be cited from petroleum
reservoirs wherein the oil flow encounters different layers of sand, rock,
shale, limestone, etc. Similar situations are encountered in many other
cases of practical interest such as geothermal operations, nuclear waste
repositories, water reservoirs, underground coal gasification, ground
water hydrology, iron blast furnaces, and solid matrix heat exchangers.
Vafai and Thiyagaraja (1987) analyzed a general class of problems
involving interface interactions on flow and heat transfer for three
different types of interface zones, the interface region between two
different porous media, the interface region between a fluid region and a
232 Chang-Lin Tien and Kambiz Vafai

porous medium where the fluid layer is sandwiched between an im-

permeable wall and a porous medium, and the interface region between
an impermeable medium and a porous medium. They obtained analytical
solutions for both the velocity and temperature distributions as well as
analytical expressions for the Nusselt number for all of the cases that
were investigated.
First for the interface region between an impermeable medium and a
porous medium, the Nusselt number was derived for the low and high
Prandtl number Pr fluids. For the low Prandtl number fluids, the Nusselt
number is

where r is the gamma function, f the nondimensional longitudinal

distance, and p is related to the temperature variation. For the high
Prandtl number fluids, the Nusselt number is

where 3, is a parameter that is explicitly related to the friction factor,

Reynolds number, porosity, Darcy number, and the Prandtl number,
whereas 52 is not dependent on the Prandtl number. The analysis shows
that for high Prandtl number fluids, the Nusselt number is proportional to
Pr113, and for low Prandtl number fluids, the Nusselt number is
proportional to Pr112. These Nusselt number expressions are extremely
accurate even for Prandtl numbers where the analytical temperature
distribution shows some deviations. The Nusselt number for the interface
region between two different porous media with different free stream
temperatures was also derived for the low and high Prandtl number fluids
based on the temperature distributions. For the low Prandtl number
fluids, the Nusselt number is

(2.10)

For the high Prandtl number fluids the Nusselt number for the interface
region between two different porous media with different free stream
temperatures is

NU^ =
+
Aotp:” Pr:12 [tp:12w/2 - A2Ao/(4tp1)]nf1/2
. (2.11)
+
[w Pr;l12 AO(n/f)’”](6, Dal)’l2
 Convective and Radiative Heat Transfer in Porous Media 233

For the interface region between a high Prandtl number fluid and a
porous medium the Nusselt number is found to be

In the above expressions V,, Ao, w , A2, A:l, Ac2, and Ac0 are given
explicitly by Vafai and Thiyagaraja (1987) in terms of the derived
interfacial velocity and temperature.
It should also be noted that their results for the interface between a
fluid region and a porous medium are found to be in good agreement
with the experimental hypothesis of Beavers and Joseph (1967).

FLOWS
B. CONFINED

For forced convection in confined structures two important and

interrelated cases are considered. First the convective heat transfer
process for porous channels is discussed. The importance of this
configuration is due to the fact that convective cooling can be augmented
by insertion of a high-conductivity porous material in a coolant passage.
Koh and Colony (1974) performed a numerical analysis of the cooling
effectiveness of a heat exchanger containing a conductive porous medium
while Koh and Stevens (1975) conducted an experimental investigation
for the same problem. It was shown for the case with fixed allowable wall
temperature that the heat flux at the channel wall can be increased by
over three times by using a porous material in the channel. For the case
with prescribed heat flux at the channel wall, a significant reduction in the
wall temperature was observed when a porous material was used in the
channel. This scheme has important practical applications, a typical
example being the cooling of rocket nozzles. Here, an augmentation of
regenerative cooling in the nozzle can be effectively achieved when the
convective cooling of the nozzle has reached its limit.
The Nusselt number for the fully developed regime in a porous
medium bounded by two parallel plates, based on the Darcy flow model
(slug flow) can be represented by (Rohsenow and Hartnett, 1973)

Nu = [i: for constant wall temperature,

for constant wall heat flux,
(2.13)

where Nu = q H / k ( T , - Tm), T, is the bulk mean temperature of the

fluid in the porous channel, and H the width of the channel.
234 Chang-Lin Tien and Kambiz Vafai

To account for the effect of a solid boundary, Kaviany (1985a)

performed a numerical study of laminar flow through a porous channel
bounded by isothermal plates based on the Brinkman-extended Darcy
model for constant porosity media. Recently Poulikakos and Renken
(1987) have investigated the effects of flow inertia, variable porosity, and
a solid boundary on fluid flow and heat transfer through porous media
bounded by constant-temperature parallel plates or a circular pipe. They
found that the general flow model predicts an overall enhancement in
heat transfer between the fluid-saturated porous medium and the walls,
compared to the predictions of the widely used Darcy flow model.
The convective heat transfer process through packed beds constitutes
the second case. Many industrial operations in the areas of chemical and
metallurgical engineering involve the passage of a fluid stream through a
packed bed of particulate solids to obtain extended solid/fluid interfacial
areas or good fluid mixing. Typical applications involving such systems
include catalytic and chromatographic reactions, packed absorption and
distillation towers, ion exchange columns, packed filters, pebble-type
heat exchangers, etc. The design of these systems is governed by the
pressure drop, fluid flow, and heat and mass transfer processes in the
packed bed arrangement. Considerable attention has been paid to these
aspects because of their direct influence on the optimiation and stability
of these systems. Since most packed and fluidized beds are characterized
by either high volume fractions of particles or large particles, or both,
many features are unique for analyzing such systems.
In the past, a one-dimensional model has been used to describe the
heat exchange process between the packed bed and its walls. This
continuum model assumes a uniform fluid temperature for a section
perpendicular to the flow direction. However, this condition of radial
uniformity is usually not satisfied, especially in chemical reactors with
important radial heat transfer effects. It has also been shown that the
mean temperature predicted by the one-dimensional model deviates
significantly from the actual radial mean temperature (De Wasch and
Froment, 1972). These considerations led to the use of a two-dimensional
model that accounts for the transverse heat transfer as well as the
longitudinal heat transfer by convection. In such a model, the heat
transfer in transverse direction can be characterized by the effective
thermal conductivity and the wall heat transfer coefficient. The latter
parameter was introduced to account for the higher thermal resistance
near the wall. These two parameters were estimated from the tempera-
 Convective and Radiative Heat Transfer in Porous Media 235

ture profile measurements performed in fluids flowing through packed

beds, heated or cooled from the wall (Lerou and Froment, 1977).
However, caution must be exercised while using many of the empirical
and semi-empirical correlations, especially while extrapolating into
regions where measurements are not available. The results of different
investigations show many discrepancies, the causes for which have been
attributed to the experimental techniques and the various definitions for
the parameters employed by different researchers (Balakrishnan and Pei,
1974). This type of analysis was also shown to be valid only after a certain
entrance length, thereby necessitating the introduction of a length effect
in these studies. The inclusion of the length effect in these studies does
not completely remedy this situation. This makes much of the data from
the above-mentioned investigations unsuitable for use in chemical reactor
design (Li and Finlayson, 1977).
Recently published research work emphasizes the importance of
porosity changes and the corresponding velocity changes near the
container wall and has shown, by comparison with experiments, that the
theoretical prediction of chemical reactor performance can be improved
by assuming a nonuniform velocity distribution within the packed bed
(Kalthoff and Vortmeyer, 1980). Vortmeyer and Schuster (1983) solved
the Brinkman-extended Darcy equation by a variational method and
explained the large deviations between calculated and measured profiles.
Vafai (1986) performed a boundary layer analysis for a variable-porosity
medium using the method of matched asymptotic expansions on the
Brinkman-Forschheimer-extended Darcy equation and the energy equa-
tion. The concept of the triple momentum boundary layer, to incorporate
the channeling effect in variable porosity media, was introduced in this
work. This type of analysis relies directly on the fundamental principles,
thus enabling a more thorough analysis of the problem. Cheng and Hsu
(1986) performed a numerical analysis for a fully developed forced
convective flow through a packed bed, confined by concentric walls
maintained at different temperatures. Their analysis was based on the
Brinkman model with variable permeability and the mixing-length theory
proposed recently by Cheng and Vortmeyer (1988) for transverse thermal
dispersion.
Hunt and Tien (1987) also investigated the non-Darcian convection in
cylindrical packed beds on the basis of Brinkman-Forschheimer-
extended Darcy equation. To account for the transverse thermal disper-
sion, they used the effective thermal conductivity varying with radial
236 Chang-Lin Tien and Kambiz Vafai

direction in the volume-averaged steady energy equation. Their results

agree well with experimental data for chemical catalytic reactors and are
independent of the extensive experimental relations needed in conven-
tional reactor models. However, most of the different approaches for
incorporating the thermal dispersion effects are essentially reduced to the
determination of the effective thermal conductivity by the use of more
adjustable parameters. A more fundamental approach for determination
of the thermal dispersion effects is needed for future research work.

111. Natural Convection in Porous Media

A. NATURALCONVECTION
OVER EXTERNAL
BOUNDARIES

Vertical flat plate boundaries are the most analyzed configuration for
natural convection over external boundaries (Figure 2). Cheng and
Minkowycz (1977) performed an analysis on natural convection from a
heated impermeable surface embedded in fluid-saturated porous media.
Their investigation was primarily motivated by their aim to simulate the
groundwater heating of an aquifer by a dike. They used a boundary layer
formulation based on Darcy’s law and the energy equation to obtain
similarity solutions. The local Nusselt number for a vertical isothermal
wall was found to be

Nu, = 0.444 Raia, (3.1)

where Nu, = h x / k and Ra, = Kgpx(Tw - T,)/(YY,g the gravitational
constant, Y the kinematic viscosity, and /3 is the volumetric thermal
expansion coefficient. This equation is valid in the boundary layer
regime, where Ra;’>> 1. The local Nusselt number for a vertical wall
with constant heat flux can be written as

Nu, = 0.772 Ra,*’”, (3.2)

where Ra,* = Kgpx’q/cuvk. This relation is valid in the boundary-layer
regime, Ra,*lI3>> 1. Cheng and Pop (1984) also considered transient free
convection about a heated vertical wall embedded in a porous medium.
To extend the range of applicability of boundary layer analysis to
relatively lower values of Rayleigh number, higher-order effects such as
entrainments from the edge of the boundary layer, axial heat conduction
and normal pressure gradients have to be included in the boundary layer
 Convective and Radiative Heat Transfer in Porous Media 237

X
t

(11) (12)
FIG.2. Geometric configuration for natural convection heat transfer (boundary layer).
(a) Vertical flat plate.
(b) Horizontal flat plate.
(c) Conjugate boundary layer.
(d) Vertical cylinder.
(el) Horizontal cylinder or sphere.
(e2) Horizontal cylinder buried beneath an impermeable horizontal surface.
(fl) Point source or horizontal line source in an infinite porous medium.
(f2) Point source or horizontal line source located on the lower boundary of a semi-infinite
porous medium.
238 Chang-Lin Tien and Kambiz Vafai

analysis. Cheng and Chang (1979) and Cheng and Hsu (1984) examined
the magnitude of these effects using the method of matched asymptotic
expansions for the problem of natural convection in a porous medium
near a vertical plate with a power law variation of wall temperature. It
was found that higher-order theory has a significant effect on the velocity
profiles but a relatively small effect on the temperature profiles.
However, their results have been found to have questionable validity.
Joshi and Gebhart (1984) have pointed out that those results are valid
only for a surface temperature variation specified a priori, such as an
isothermal surface condition, and that the problem with uniform heat flux
condition at the wall has to be treated separately. This is required
because the surface temperature variation is improved successively at
each order and is not imposed externally in the perturbation analysis. In
order to remedy the above problem, they developed consistent higher-
order approximations including second-order solutions using the method
of matched asymptotic expansions.
A number of recent studies have considered the various non-Darcian
effects for the problem of natural convection about a vertical flat plate
embedded in a porous medium; Plumb and Huenefeld (1981) and Bejan
and Poulikakos (1984) used Forschheimer’s equation to account for
inertial effects and obtained similarity solutions with the boundary layer
approximations. Based on Brinkman’s equation Evans and Plumb (1978)
investigated numerically the boundary effect on heat and fluid flow, and
Hsu and Cheng (1985) used the method of matched asymptotic expan-
sions for the same problem. Recently, Kim and Vafai (1989) obtained
analytical and numerical solutions for buoyancy-driven fluid flow and
heat transfer about a vertical flat plate embedded in a porous medium.
The governing equations were solved analytically using the method of
matched asymptotic expansions along with the modified Oseen method.
Their analysis, which included the boundary effects, was performed for
two general cases, the constant wall temperature and the constant wall
heat flux cases. For each case, two important categories were considered.
For the constant wall temperature case these categories were shown to be
physically related to either DaF2 << Rail’’ (DaF2<< RaE-1’3 for the
constant wall heat flux) or DaF2 >> Ra,’” (Da;’ >> Ra2-1’3 for the
constant wall heat flux). Hong, Tien, and Kaviany (1985) included both
boundary and inertia effects as well as the convective term in the
momentum equation. Also Kaviany (1985b) reported a Karman-
Pohlhausen solution to the same problem and Chen, Hung, and Cleaver
 Convective and Radiative Heat Transfer in Porous Media 239

(1987) studied numerically the non-Darcian effect of a vertical flat plate

embedded in a high-porosity medium on transient natural convection.
These investigations have shown that both inertia and boundary effects
decrease the heat transfer rate as the Rayleigh number is increased.
However, the influence of boundary effects on the Rayleigh number is
less pronounced than the influence of inertial effects. Another important
non-Darcian effect is the thermal dispersion effect. It has been found that
this effect needs to be considered when the inertia effects are prevalent.
Cheng (1981) and Plumb (1983) examined the thermal dispersion effects
in non-Darcian convective flows. Their analyses neglected the boundary
effect, which was subsequently included in the recent study by Hong and
Tien (1987). It is shown that the heat transfer rate is enhanced as a result
of the thermal dispersion effects.
Evans and Plumb (1978), Cheng, Ali, and Verma (1981), Cheng and
Ali (1981) and Kaviany and Mittal (1987) have performed experimental
investigations of natural convection around a vertical and an inclined
isothermal plate. Also Huenefeld and Plumb (1981) experimentally
investigated natural convection about a vertical plate with constant heat
flux.
The results reviewed in this section are also applicable to the case of an
inclined flat plate if the gravitational acceleration appearing in the
Rayleigh number is replaced by the gravitational acceleration component
that acts along the inclined surface. However, using the appropriate
component of the gravitational acceleration produces valid results only
up to a certain critical inclination angle. Little work has been done on
determining this critical inclination angle, where apparently the convec-
tion mode changes character from one type to another.
For natural convection over a horizontal flat plate, which has important
applications in heat transfer around a geothermal reservoir, Cheng and
Chang (1976) obtained a similarity solution leading to a local Nusselt
number given by
Nu, = 0.420 Rail3, (3.3)
where Nu, = hx/k and Ra, = Kg/3x(Tw - T,)/cuv. This equation is valid
only in the thermal boundary layer regime, Le., Rail3>>1. Also the local
Nusselt number for a horizontal wall with uniform heat flux was found
to be
Nu, = 0.859 Ra,*'14, (3.4)
where Ra,*= Kg/3x2q/kav. This relation holds also only in the thermal
240 Chang-Lin Tien and Kambiz Vafai

boundary layer regime, Ra,*”4>> 1. In a subsequent paper, Cheng

(1978b) obtained Karman-Pohlhausen type integral solutions for the
same problem. Also Pop and Cheng (1983) reported a Karman-
Pohlhausen solution for the growth of a thermal boundary layer in a
porous medium adjacent to a suddenly heated semi-infinite horizontal
surface.
To extend the applicability of the boundary layer theory to lower
Rayleigh number ranges, Chang and Cheng (1983) refined the boundary
layer analysis by including the higher-order effects such as the upward-
drift induced friction and fluid entrainment, using the method of matched
asymptotic expansions.
The upward-facing cold plate with finite length, another horizontal wall
configuration, was examined by Kimura, Bejan, and Pop (1985). They
have shown that the overall heat transfer rate between the porous
medium and the flat plate varies as

Nu- Rain, (3.5)

where Ra is the Darcy-modified Rayleigh number based on plate
half-length.
The analysis of the buoyancy induced flow occurring on both sides of a
vertical impermeable wall facing two fluid-saturated porous reservoirs
maintained at different temperatures could be useful for calculations in
the design of thermal insulations. This type of configuration falls under a
category which is usually referred to as “conjugate boundary layers.”
Bejan and Anderson (1981) analyzed such a configuration. They found
that a boundary layer exists on each side of the impermeable wall. The
overall Nusselt number correlation, correct to within 1%, is

Nu = 0.382(1+ 0.615 w)-0,875Ra’”

H7 (3.6)
where Nu = qO-HH/(Tm,H - Tm,L)k,w = (Wk/Hk,) Rag’ is the parameter
describing the size of the wall thermal resistance relative to the
convective resistance, k, is the thermal conductivity of the wall, and
descriptions for Tm,H,Tm,L,H , and W are given in Figure 2c.
Bejan and Anderson (1983) examined the natural convection flow
between a fluid-saturated porous medium and a fluid reservoir separated
by a vertical impermeable wall when the two fluids are maintained at
different temperatures. For this case, it was also shown that a boundary
layer exists on each side of the partition. The overall Nusselt number may
 Convective and Radiative Heat Transfer in Porous Media 241

be represented as
Nu = [(0.638)-’ + (0.888 B)-’]-’ Rag:f, (3.7)
where RaH,f= g(/3/~uv)~H’(T,,, - Tm,L),and B = k Rag2/kf Rag$.
For the case of a vertical cylinder, which has practical application in
predicting the convective flow generated in the ground water adjacent to
hot intrusives like magma, Minkowycz and Cheng (1976) obtained an
approximate solution using the local nonsimilarity method. Their solution
included different power law variations for the surface temperature of the
cylinder. In case of an isothermal cylinder, the local Nusselt number is
given by

Nu,=0.444Rain
[1+0.6 (3
- Ra;”’
1,

where ro is the radius of the cylinder and x refers to the curvilinear

coordinate along the surface of the cylinder. By comparing the above
result with the Nusselt number for the vertical flat plate, it becomes
apparent that for the same surface area, a vertical cylinder transfers more
heat than a vertical plate.
For horizontal cylinders and spheres the flow and temperature fields
are part of the information that can be used in the design of underground
electrical cables, thermal insulation of power plants, and steam and water
distribution lines. Merkin (1979) obtained similarity solutions for natural
convection boundary layers on axisymmetric and two-dimensional bodies
of arbitrary shape embedded in a saturated porous medium. Cheng
(1982), using a similarity transformation similar to Merkin’s, showed that
the local Nusselt numbers for boundary layer convection about a
horizontal isothermal cylinder or sphere embedded in an infinite porous
medium are respectively given by
Nu, = 0.444 R a f p n a sin( 1- cos @)-”’, (3.9)

Nu, = 0.444 R a r (3.10)

where Nu, = qwro/k(Tw- T,) and Ra, = Kg/3ro@(Tw- T,)/av with @

denoting the angle measured from the downward vertical (as shown in
Figure 2e) and ro denoting the radius of the cylinder or sphere. It should
be noted that Cheng neglected the buoyancy force normal to the heated
surface in order to obtain the above equations. Recently Fand, Steinber-
242 Chang-Lin Tien and Kambiz Vafai

ger, and Cheng (1986) performed an experimental investigation of heat

transfer by natural convection from a horizontal cylinder embedded in a
porous medium. They have shown that the overall range of Rayleigh
number can be divided into two subregions, called “low” and “high,” in
each of which the Nusselt number behaves differently. It is shown that
the low-Rayleigh number region corresponds to Darcy flow and the
high-Rayleigh number region to Forschheimer flow.
Natural convection around a cylinder buried beneath an impermeable
horizontal surface is another important configuration. Schrock, Fernan-
dez, and Kesavan (1970) and Fernandez and Schrock (1982) performed
experimental investigations of this geometry. Recently Farouk and
Shayer (1985) conducted a numerical study of natural convection from a
heated cylinder buried in a semi-infinite liquid-saturated porous medium.
It should be mentioned that this geometry is the same configuration
studied by Schrock et al.
Natural convection about a line source or a point source, which is the
limiting case for a horizontal cylinder or sphere, respectively, has an
important application to heat transfer problems in a nuclear-waste
repository. Bejan (1978) considered the problem of transient and
steady-state natural convection around a concentrated point heat source
in an infinite porous medium at low Rayleigh numbers and obtained a
perturbation solution for the transient case up to the first-order Rayleigh
number and for the steady-state case up to the third-order Rayleigh
number. For the same point source geometry, Hickox and Watts (1980)
obtained a numerical solution using a similarity transformation for the
steady-state natural convection. They also obtained a numerical solution
for a point source located on the lower, insulated boundary of a
semi-infinite region for a range of Rayleigh numbers from 0.1 to 100. In a
subsequent paper, Hickox (1981) has shown that in the limit of small
Rayleigh numbers, superposition of transient and steady convection
solutions for a concentrated heat source, a finite vertical line source and a
constant temperature sphere can be used for obtaining solutions for some
other geometries.
Prior to these works, Wooding (1963) had proposed a boundary layer
model for large-Rayleigh number natural convection about a point and
line source located on the lower boundary of a semi-infinite fluid-
saturated porous medium. Furthermore, Wooding was able to present an
analytical solution for the temperature and flow field around a horizontal
line source in a porous medium at high Rayleigh numbers. Bejan (1984)
 Convective and Radiative Heat Transfer in Porous Media 243

presented an analytical solution using a boundary layer analysis for the

flow and temperature field in the high-Rayleigh number regime for the
plume above a point source.

B. NATURAL
CONVECTION
IN CONFINED
STRUCTURES

Convective heat transfer in a rectangular porous cavity in which

vertical walls are maintained at two different temperatures is a fun-
damental problem (Figure 3). It has important applications to building
insulation and geothermal operations. The boundary layer regime that
develops at large Rayleigh numbers was studied by Weber (1975a) using
boundary layer approximations with the modified Oseen method. Darcy’s
equation was used as the governing momentum equation, and the Nusselt
number, defined as the ratio of the total heat transport to the heat
transferred by pure conduction, was found to be

Nu=0.577
(3
- Rag*. (3.11)

By replacing Weber’s impermeable horizontal boundary condition with

average zero energy flux conditions along the horizontal walls, Bejan
(1979) was able to get better agreement with the experimental results of
Klarsfeld (1963) as well as with the numerical calculations of Bankvall
(1974) and Bums, Chow, and Tien (1977). Bejan presented the Nusselt
number as

Nu=0.508
(3
- Rag2. (3.12)

Simpkins and Blythe (1980) arrived at similar results by using an integral

method.
Bories and Combarnous (1973) and Seki, Fukusako, and Inaba (1978)
performed experimental work to investigate the effects of the cavity
aspect ratio and the physical properties of the porous medium on the
overall heat transfer rates. Shiralkar, Haajizadeh, and Tien (1983)
performed a numerical study of high-Rayleigh number convection for the
same problem using a stable exponential differencing scheme. To account
for the non-Darcian effect on fluid flow and heat transfer, Tong and
Subramanian (1985) used the Brinkman-extended model based on
boundary layer approximations and the modified Oseen method. They
244 Chang-Lin Tien and Kambiz Vafai
 Convective and Radiative Heat Transfer in Porous Media

(1)
FIG. 3. Geometric configuration for natural convection heat transfer (confined flow).
(a) Rectangular porous cavity heated from the side.
(b) Rectangular cavity, partially filled with a porous medium, heated from the side.
(cl) Horizontal cylinder heated from the side.
(c2) Horizontal Annulus heated from the side.
(d) Vertical porous annulus heated from the side.
(e) Horizontal concentric cylinders or spheres with radial temperature gradient.
(f) Horizontal porous layers heated from below.
(g) Horizontal porous layers with localized heating from below.
(h) Inclined porous layer heated from below.
(i) Wedge-shaped porous layer heated from below.
(j) Open-ended cavity with a porous obstructing medium.
246 Chang-Lin Tien and Kambiz Vafai

reported significant contributions of the viscous diffusion term at high

Rayleigh and Darcy numbers. Poulikakos and Bejan (1985) used the
Forschheimer-extended Darcy model to obtain boundary layer solutions
and numerical results. Chan, Ivey, and Barry (1970), Lauriat and Prasad
(1987), and Tong and Orangi (1986) performed numerical analyses of a
rectangular cavity heated from the side for various aspect ratios, Darcy
numbers and Rayleigh numbers, using the Brinkman-extended Darcy
model.
In the limit of low aspect ratios the problem becomes one of natural
convection in shallow cavities with different end temperatures. Walker
and Homsy (1978) studied such a problem by using matched asymptotic
expansions and found the following expression for the Nusselt number
based on a Darcy model
1
(3.13)

The same result was also reported by Bejan and Tien (1978). Hickox and
Gartling (1981) also performed a numerical investigation of this problem
using the Galerkin form of the finite element method. To account for the
momentum boundary layer, recently Sen (1987) used the Brinkman-
extended Darcy model to obtain the boundary layer solutions for shallow
rectangular cavities. In order to investigate the effect of infiltration into a
shallow cavity Haajizadeh and Tien (1983) performed a numerical and
experimental study of natural convection in a shallow cavity with one
permeable vertical wall.
In a series of investigations, Blythe, Simpkins, and Daniels
(1982,1983) discussed the large Rayleigh number limit for a fixed aspect
ratio and presented some new scaling laws for the problem of natural
convection in a rectangular cavity. Assuming a constant buoyant term,
Philip (1982) found some exact solutions, using the separation of
variables techniques, in the small Rayleigh number limit in cavities of
rectangular as well as of other geometric cross-sections. Prasad and
Kulacki (1984a) studied numerically the effect of aspect ratio on fluid
flow and heat transfer for the rectangular cavity problem. It was shown
that for a tall cavity, an increase in the width decreases heat transfer
except when the flow exhibits boundary layer behavior on the vertical
walls.
Bejan (1983a) considered the case where the uniform heating and
cooling effects are prescribed along the vertical side walls. His solution
 Convective and Radiative Heat Transfer in Porous Media 247

was based on the Darcy flow model, the thermal boundary layer
approximation and the use of the modified Oseen method. For this case
the Nusselt number in the high Rayleigh number limit was found to be
1 H
Nu = -
2 L
(-) RaL2’, (3.14)

where RaL = Kg/3H2q/avk. To account for the various non-Darcian

effects on fluid flow and heat transfer, Poulikakos (1985) used the
Forschheimer-extended Darcy model to obtain the boundary layer
solution, and Vasseur and Robillard (1987) used the Brinkman-extended
Darcy model to study the boundary effects, which were shown to have a
significant influence on the flow field and heat transfer. Also Prasad and
Kulacki (1984b) studied numerically two-dimensional, steady, free con-
vection for a rectangular cavity with constant heat flux on one vertical
wall while the other vertical wall as isothermally cooled.
Furthermore, Bejan (1983b) studied numerically the effect of internal
flow obstructions, created by the addition of a variable size horizontal or
vertical plate to the corresponding vertical or horizontal boundary, on
heat transfer through a two-dimensional porous layer heated from the
side. Finally, in an attempt to observe the effect of non-uniform
permeability and thermal diffusivity on natural convection through a
porous layer heated from the side, Poulikakos and Bejan (1983a)
numerically investigated the flow field in a rectangular enclosure with
vertical (or horizontal) layers of various permeability porous media.
In building insulation, it is common to fill the enclosure entirely with a
porous material. To study the effect of partial filling of an enclosure with
a porous medium, Tong and Subramanian (1983) have numerically
analyzed natural convection in a rectangular enclosure that is vertically
divided into an open region and a region filled with a porous medium. By
assuming an impermeable interface, the open and the porous regions
were essentially decoupled. In essence their results could also be
obtained by a combination of available literature on open and porous
cavities by using an iteration procedure. An experimental investigation of
the problem by Tong, Faruque, Orangi, and Sathe (1986) led the authors
to conclude that the heat transfer could be minimized by filling the
enclosure partially with a porous medium rather than filling it completely.
A more practical case where there is no impermeable partition between
the fluid and the porous medium was considered both numerically and
experimentally by Beckermann, Ramadhyani, and Viskanta (1987).
248 Chang-Lin Tien and Kambiz Vafai

For a horizontal cylinder and annulus with ends at different tempera-

tures Bejan and Tien (1978) presented a parametric solution. It was
based on representation of the velocity components and the temperature
field by what amounts to a perturbation in the core region and integral
analyses in the end regions. Bejan and Tien (1979) also studied natural
convection in a porous medium bounded by two concentric cylinders with
different end temperatures.
Hickox and Gartling (1982) studied the case of a vertical porous
annulus heated from the side, in which the inner and outer walls are
isothermally heated and cooled, respectively, while the top and bottom
walls are insulated. They used a finite element method for obtaining heat
transfer results for Rayleigh numbers up to lo2. This type of geometry is
typically encountered in the storage of canisters containing high-level
nuclear waste material. However, in recent years the emphasis has been
placed in the horizontal emplacement of these canisters, thereby making
a modified version of the previously cited case more appropriate for this
type of application. Havstad and Burns (1982) used a finite difference
method, a perturbation technique, and an approximate analysis to
analyze the same problem. Prasad and Kulacki (1984c, 1985) also
considered the same problem for a range of Rayleigh numbers up to lo4
and for the height-to-gap width ratio of 1 I A I20 and 0.3 I A I0.9
respectively. The work, based on a finite difference formulation, con-
centrated on an investigation of the boundary layer and curvature effects
on temperature and flow-field structures. Furthermore, in their latter
work Prasad and Kulacki (1985), as well as Prasad et al. (1985),
performed experimental investigations of the same problem.
Studies of natural convective flow and heat transfer in porous
concentric cylinders and spheres with radial temperature gradients have
applications to buried pipelines, storage of solar energy in underground
containers, and insulating material for some types of nuclear reactors and
LNG facilities. Caltagirone (1976) examined free convection in porous
media when the cylindrical surfaces of an annulus are kept at different
temperatures. He obtained the steady, two-dimensional solutions numer-
ically and also observed the three-dimensional flows and associated
fluctuations experimentally. Facas and Farouk (1983) studied numerically
two-dimensional transient convection for the same problem. They were
able to show that the transient state finally does converge to a steady
state with no significant intermediate oscillatory behavior. Recently
Muralidhar and Kulacki (1986) numerically investigated the non-Darcian
 Convective and Radiative Heat Transfer in Porous Media 249

effects of inertia, boundary friction, and variable porosity for the same
problem.
Burns and Tien (1979) analyzed two-dimensional steady natural
convection in a porous medium bounded by concentric spheres and
horizontal cylinders. Their solutions were based on the finite difference
formulation and the use of regular perturbation method for the case
where the inner boundary was isothermal and the outer boundary was
allowed to interact with the surroundings. Masuoka, Ishizaka, and
Katsuhara (1980) also performed a numerical and experimental inves-
tigation of natural convection in porous media bounded by two con-
centric spheres.
For horizontal porous layers heated from below, there has been a
considerable amount of analytical, numerical, and experimental work
concerning the Benard convection. Lapwood (1948) examined the onset
of convection in an infinitely long horizontal layer heated from below by
using linearized stability analysis and found that the critical Rayleigh
number for the onset of convection is given by
(3.15)
where RaH = Kg/%f(TH- TC)/av. Later, Katto and Masuoka (1967)
performed an experimental investigation for the same problem.
Nusselt number dependence on Rayleigh numbers higher than the
critical value has been reviewed by Cheng (1978a), who compiled results
from nine sources for work that has been done over the past three
decades. These sources were Combarnous and Bories (1975), Schneider
(1963), Buretta and Berman (1976), Yen (1974), Elder (1967a), Kaneko,
Mohtadi, and Aziz (1974), Gupta and Joseph (1973), Strauss (1974), and
Combarnous and Bia (1971).
Robinson and O’Sullivan (1976), Somerton (1983), and Combarnous
and Boris (1975) tried to account for the wide spread in the data from the
sources that were reviewed by Cheng (1978a). Recently, Georgiadis and
Catton (1986) and Jonsson and Catton (1987) have shown through
numerical and experimental studies, respectively, that this scatter can be
attributed to the Prandtl number effect on Benard convection in porous
media heated from below. Georgiadis and Catton (1986) used the
Brinkman-Forschheimer-extended Darcy equation to account for the
inertial and boundary effects.
In a related configuration, Poulikakos (1986) studied numerically the
problem of buoyancy-driven flow instability in a horizontal composite
250 Chang-Lin Tien and Kambiz Vafai

(fluid-porous) system heated from below. The composite system consists

of a fluid layer overlaying a porous substrate and is bounded by two
isothermal horizontal walls and two adiabatic vertical walls. This con-
figuration has important applications to fibrous and granular thermal
insulation where the insulation occupies only part of the space separating
the hot boundary from the cold boundary. A Forschheimer-Brinkman-
extended Darcy equation was used to describe the flow inside the porous
bed. Also, the dependence of the flow and temperature field characteris-
tics on the pertinent dimensionless groups was discussed.
Natural convection through horizontal porous layers, where only a
portion of the bottom surface is heated while the rest of it is either
adiabatic or isothermally cooled, has some applications with respect to a
geologic repository for the storage of nuclear waste and also for some
geothermal operations. Elder (1967a) performed a numerical study of the
steady-state natural convection in a horizontal porous cavity with walls
that are isothermally cooled except for a centrally heated portion of the
bottom surface. Later, Elder (1967b) studied numerically and ex-
perimentally the transient natural convection for the same geometry and
boundary conditions.
Horne and O’Sullivan (1974) studied numerically and experimentally
the oscillatory flow behavior in a horizontal porous cavity with insulated
vertical walls and isothermally cooled top and bottom surfaces, except for
a centrally heated portion of the bottom surface. Also Horne and
O’Sullivan (1978) studied the effect of temperature dependent viscosity
and thermal expansion coeflicient on flow and temperature fields.
Recently Prasad and Kulacki (1986,1987) analyzed natural convection
in a horizontal porous cavity with insulated vertical walls, an isothermally
cooled top surface, and an insulated bottom surface with a centrally
heated portion. They used the finite difference formulation to charac-
terize the effects of the size and strength of the heat source as well as the
aspect ratio on heat transfer and fluid flow. They observed a plume-like
behavior above the heated region as the Rayleigh number was increased.
Bories and Combarnous (1973) have performed a theoretical and
experimental investigation of convective heat transfer in an inclined
porous layer heated isothermally from below. The three main structures
of free convection movements have been observed depending on the
values of the Rayleigh number, the inclination angle, and the longitudinal
extension of the layer. It was found that if Ra, < ~JT*/COS 8, where 8 is
the angle between the heated wall and the horizontal plate, the
 Convective and Radiative Heat Transfer in Porous Media 251

convective motion takes the form of a two-dimensional unicellular flow.

For the Rayleigh numbers higher than this critical value, a transition
from unicellular flow to stable three-dimensional flow is observed to
occur in such a way that the convective motion takes the form of either
polyhedral cells or longitudinal coils, depending on the value of 0 itself.
The investigations of Bories and Combarnous (1973) were based on a
finite-sized porous layer. For an infinite layer, Weber (1975b) analyzed
natural convection in a tilted porous layer using a perturbation technique.
Caltagirone and Bones (1985) obtained solutions and stability criteria for
natural convection in an inclined porous layer using a linear stability
theory and Galerkin’s spectral method.
Recently, Moya, Ramos, and Sen (1987) studied numerically two-
dimensional natural convective flow in a tilted rectangular porous cavity
where two opposing walls are kept at constant but different temperatures
while the other two walls are thermally insulated. Also, Vasseur, Satish,
and Robillard (1987) studied analytically and numerically natural convec-
tion in an inclined thin layer of porous medium when a constant heat flux
is applied on the two opposing walls while the other two walls are
insulated.
A somewhat related configuration to the horizontal layer is a wedge-
shaped porous layer. This type of geometry has direct use in estimating
the performance of the thermal insulation used in an attic-shaped
application. Bejan and Poulikakos (1982) studied natural convection in
an attic-shaped horizontal space heated from below by using an asympto-
tic analysis that was also used by Walker and Homsy (1978). Also
Poulikakos and Bejan (1983b) performed scaling analysis and numerical
simulation on transient and steady-state convection in the limit of high
Rayleigh numbers.
The final configuration considered in this section is natural convection
in open-ended porous cavities. The engineering applications of open
cavities are important. Examples include fire spread in rooms, solar
thermal central receiver systems, connections between reservoirs, nuclear
waste repositories, etc. This very important and fundamental geometry
can be modified to analyze a variety of complex geometries. It should be
mentioned that while that there has been many investigations on the
two-dimensional natural convection in porous enclosures, the related but
more complex problem of open cavities has received very little attention.
One of the reasons for this imbalance in the above-mentioned investiga-
tions can be traced to the difficulty in specifying the appropriate
252 Chang-Lin Tien and Kambiz Vafai

boundary conditions for an open cavity configuration due to its sig-

nificantly more complex geometry. Furthermore, in this configuration,
the physical attributes of both confined and external natural convection
are present, thus leading to a complex chain of interactions. Ettefagh and
Vafai (1988) have analyzed such a configuration and their results clearly
show the influence of the external corners in an open-ended cavity in
augmenting of the flow instabilities. They also discussed in detail the role
of the open boundaries, different temperature levels, and different aspect
ratios.

IV. Multipbase Transport in Porous Media

The subject of multiphase transport and phase change in porous media

has gained considerable attention during the past two decades. Among
the wide variety of the current applications of this subject the following
examples could be cited: drying of different porous materials (Berger and
Pei, 1973; Plumb et al., 1985), condensation in porous materials
(Ogniewicz and Tien, 1981; Vafai and Whitaker, 1986); heat pipe
applications (Udell, 1985); and geothermal applications (Faust and
Mercer, 1979; Cheng, 1978b).

A. THEORETTCAL
BASIS

The earliest studies of fluid transport including some aspects of heat

transport in porous materials were reported in the 1920s. Some early
examples are the works by Gardner and Widtsoe (1920), Lewis (1921),
Sherwood (1931), Newman (1931), and Richards (1931). In the pioneer-
ing works of chemical engineers in drying phenomena, it was assumed
that fluid motion in the porous media occurred by diffusion. Lewis
(1921), Newman (1931), and Sherwood (1929a,b, 1930) all used diffusion
equations for representing fluid transport. On the other hand, researchers
from other disciplines, such as hydrologists, ceramicists, and colloid
chemists, found out that diffusion equations were not adequate to explain
the liquid motion in unsaturated porous media and that capillary forces
(surface tension) were also important. Studies by Gardner and Widtsoe
(1920), Richards (1931), and Westman (1929) demonstrated this
influence.
 Convective and Radiative Heat Transfer in Porous Media 253

While the emphasis in the pioneering works on drying was given to

momentum transport, the influence of heat transport was ignored. The
first attempts to include the effect of heat transport were made by
Krischer (l938,1940a,b). This, afterwards, led to more serious con-
sideration of heat transport in related problems.
Relatively complete models of multiphase transport processes con-
sidering phase change and capillary action in porous media were first
established in the 1950s. Some of the related and more commonly used
works include: the studies reported by Philip and DeVries (1957) and
DeVries (1958) in which the multiphase transport processes in soil were
formulated, the model for drying of hygroscopic capillary porous solids
developed by Berger and Pei (1973) as an improved version of Krischer’s
model, the model of simultaneous transport processes in capillary-porous
media developed by Luikov (1964,1975), the studies by Eckert and
Pfender (1978) and Eckert and Faghri (1980) in which moisture migration
due to temperature gradient and transport processes in unsaturated
porous media are examined, and the drying model developed by
Whitaker (1977) using local volume-averaging technique.
One of the main problems encountered in the modeling of multiphase
tranport processes in porous media is the lack of information about the
numerous transport coefficients required for formulating the problem.
These coefficients may be functions of temperature as well as the
composition of the fluids flowing through the porous medium. In the case
of anisotropic media, some of these may even be tensors with nine
components. For simplifying the formulation of the multiphase transport
processes in porous media, a number of assumptions are commonly made.
The differences between various models of multiphase transport
processes in porous media usually arise from the different level of details
of the physical phenomena considered and the different simplifying
assumptions made according to the prevailing physical conditions.
Moreover, the use of different empirical correlations for transport
coefficients and different thermodynamic relations also gives rise to
differences in formulations.
Numerous studies dealing with specific problems related to multiphase
transport processes with change of phase in porous materials have been
reported. Some of the more widely used models and the pertinent
references on which they are based include the models by Whitaker
(1977), Luikov (1964, 1975), Philip and DeVries (1957), Berger and Pei
(1973), Eckert and Pfender (1978), and Eckert and Faghri (1980).
254 Chang-Lin Tien and Kambiz Vafai

Although some of these models have been developed for specific

problems, their different versions have been used quite extensively. The
different references cited for each of DeVries’ and Eckert’s models are
complementary to each other for the model concerned. In Luikov’s case,
there are a few differences in the two references cited. In what follows,
these models are analyzed and compared qualitatively, the assumptions
employed in each model are evaluated, and the advantages, disadvan-
tages, and general applicability of each model are discussed.
In Whitaker’s model the local volume-averaging technique is applied to
the governing balance equations resulting in volume-averaged equations.
With the help of constitutive equations such as Fourier’s law of heat
conduction and Darcy’s law for fluid flow, volume constraint relations,
and thermodynamic relations such as Kelvin equation and Clausius-
Clapeyron equation, a rigorous model of governing equations is formed.
Details of transport phenomena such as diffusion, dispersion, and
convective flow as well as capillary action in liquid phase are taken into
account.
Luikov, too, has established a very comprehensive system of governing
equations for multiphase transport processes in capillary-porous systems.
The primitive form of the equations for the liquid, vapor, and inert gas
phases is quite appealing. However, the simplifying assumptions used for
reducing the entire system of governing equations to three coupled
differential equations with temperature, pressure, and total moisture
content as system variables are not realistic.
The model developed by DeVries for simultaneous heat and mass
transport with phase change in soils is also widely used. It contains, to a
certain extent, information on the physics of the transport of liquid and
vapor in the porous medium. The fluid motion is represented by Darcy’s
law, and it is shown how the liquid flux term can be split into temperature
gradient, liquid moisture gradient, and gravity terms. This model is
suitable for applications with no noncondensibles or for applications in
which the effects of noncondensibles may be neglected. One drawback of
this model is that the correlation given for the liquid diffusivity is valid
only for the capillary condensation region. Analytical solutions for some
simplified steady-state cases have been presented in DeVries’ work.
The model presented by Berger and Pei (1973) for drying of hygro-
scopic porous materials is a simplified model in which the motion of both
liquid and vapor is represented by Fick‘s law. The vapor flow is taken to
be by diffusion, and the liquid motion is considered to be by capillary
 Convective and Radiative Heat Transfer in Porous Media 255

flow. The simplified nature of this model arises from the assumption of
constant transport coefficients, namely the vapor diffusivity and the liquid
conductivity. Further simplification is carried out by assuming that heat
transport occurs only by conduction through the solid matrix and by
transfer of latent heat due to phase change. This model is applicable to
cases with small temperature changes. For the different stages of drying,
two different coupling relations are used for the governing equations.
Eckert, in his works with Pfender (1978) and Faghri (1980), developed
a model for simultaneous heat and mass transfer in an unsaturated
porous medium. In his work with Faghri, no phase change is considered.
Rather, the motion of moisture due to temperature gradients is analyzed.
Vapor motion is represented by Fick’s law, and the problem is formu-
lated and solved for the one-dimensional case where the porous medium
is bounded by two impermeable walls. In this model, too, no nonconden-
sibles are taken into account. Some analytical solutions are obtained for
simplified cases with constant transport properties. Eckert, in his work
with Pfender, considers multiphase transport in porous media in the
presence of phase change. In this study, the liquid and vapor mass fluxes
are formulated by Darcy-like and diffusive terms respectively. Although
the energy transport equation is not developed, the heat flux per unit
area is defined and a description of the experimental work performed to
determine the thermal conductivity of the soil is given.
The models developed for simultaneous heat and mass transfer
processes in multiphase porous systems may have differences due to the
various assumptions made in developing each model. These assumptions
may change to a great extent depending on the physical conditions of the
particular problem considered. Moreover, there is lack of universal
methods for defining the transport coefficients and transport potentials.
This is a great drawback in establishing a generalizable model.

B. APPLICATION
AREAS

The application of multiphase transport in porous media cover a wide

variety of problems. The simplest problem that has been considered is
the effect of a temperature gradient on the transport processes in porous
media. Dybbs and Schweitzer (1973) formulated the problem of non-
isothermal flow in porous media using the volume-averaging technique
for low Reynolds number flows in which inertia terms are negligible.
256 Chang-Lin Tien and Kambiz Vafai

Dinulescu and Eckert (1980) studied the problem of moisture migration

due to temperature gradient in porous media. They performed a
one-dimensional analysis of this problem and came up with an analytical
solution. The first set of boundary conditions used were constant
temperature and impermeable wall boundaries at the two sides. Another
solution was obtained for constant heat flux boundary conditions. Huang
(1979) established a model of governing equations for the multiphase
transport processes in porous media. In Huang’s work the diffusive and
convective transport were described carefully and the effect of the
temperature gradient was emphasized. The governing equations for
drying processes in both funicular and pendular stages were developed.
In the funicular stage, the gaseous phases were assumed to be stationary,
while in the pendular stage, the liquid phase was taken to be immobile.
Huang’s model looks like a blend of Whitaker’s (1977) and Luikov’s
(1964) models. An account of the thermodynamic relations to be used in
conjunction with the transport equations was also presented in Huang’s
work. Huang et al. (1978) obtained a solution for the problem of
moisture migration in a concrete slab for natural drying conditions.
Baladi et al. (1981) studied the transport processes in soil around a buried
heat source. The phenomena of moisture migration and energy transport
were analyzed by a one-dimensional model. Numerical and experimental
works were carried out. In addition, closed-form solutions were obtained
by using approximate models in this study. In modeling the porous
system, two distinct wet and dry regions were assumed to exist.
Drying phenomena form an important part of the applications of
multiphase transport in porous media. One of the earlier works in this
area was carried out by Harmathy (1969) who investigated the drying
problem in a porous medium. An important assumption used in this work
was to consider all moisture transfer to take place in the gaseous phase
only. More recent studies have been reported by Lyczkowski and Chao
(1984). In their work, the authors give a comparison of the two-phase
drying model developed by Lyczkowski with the Stefan model for coal
drying. In the two-phase model discussed, noncondensibles are not taken
into account. Plumb et al. (1985) studied the problem of drying of
softwood. They formed a model for multiphase transport processes
taking into account both diffusion and capillary action in the liquid phase.
By the use of a mechanistic model they expressed the capillary porous
properties (permeability and capillary pressure) in terms of geometric
properties and measured permeability parameters. Numerical and ex-
 Convective and Radiative Heat Transfer in Porous Media 257

perimental studies were carried out for obtaining moisture profiles.

Another example is the study reported by Haber et al. (1984) on drying
in a semi-infinite porous system with emphasis on heat convection. A
one-dimensional model was developed in which two separate regions with
a moving interface were considered to form the system.
The problem of condensation in porous materials is one of the recent
problems studied. Ogniewicz and Tien (1981) reported an analytical work
on condensation in porous wall insulation. The model developed is
one-dimensional and mainly steady state, and it partially accounts for
convective and diffusive transport. The problem studied consists of a
porous slab subjected to two different environments on the two sides.
Vafai and Sarkar (1986,1987) reported a rigorous transient analysis of
moisture migration and condensation in porous and partially porous
enclosures. The analysis considered variation of properties as well as
noncondensible effects. The effect of Lewis and Prandtl numbers was
investigated by a model that utilized the volume-averaging technique. In
this problem, impermeable and adiabatic horizontal wall and permeable
vertical wall boundary conditions were employed. Properties such as
temperature, pressure, liquid fraction, and vapor density were solved for
separately, and the interface location was determined as part of the
analysis and not on a trial basis.
In another study, Vafai and Whitaker (1986) analyzed a different
problem using a two-dimensional transient model. In this study, the
multiphase transport processes with phase change in a porous slab were
also modeled by using the volume-averaging technique. Permeable wall
boundaries with constant temperature and liquid moisture content
boundary conditions were employed. The results of a numerical inves-
tigation solving for a large number of parameters were presented, and the
identification of significant transport mechanisms was performed. Based
on these findings, a simplified model for multiphase transport processes
in the presence of phase change in a porous medium was suggested.
Nilson and Romero (1980) investigated another condensation problem
in porous media. They formulated a one-dimensional problem of a
porous matrix with an inflow of saturated vapor at the boundary.
Hyperbolic/parabolic equations were obtained by Darcy formulation of
two-phase flow, and similarity solutions were obtained for the propaga-
tion of the condensation wave. In this study, correlations for relative
permeabilities of liquid and vapor as functions of liquid saturation have
been used. In a recent investigation by White and Tien (1987a), laminar
258 Chang-Lin Tien and Kambiz Vafai

film condensation in a porous medium was analyzed. In their analysis, the

no-slip condition was used for the velocity at the wall, and an exponential
model for the porosity and modified boundary layer equations was
employed for the region near the wall. Experimental studies on the same
topic were also reported by White and Tien (1987b).
Chuah and Carey (1985) investigated the transport processes in a
medium with variable porosity. In this study the porosity of the medium
is expressed in exponential form. Darcy’s law is used for liquid and vapor
transport. In this study, empirical correlations are used extensively for
some variables based on previous experimental data. Permeability is
expressed by Kozeny-Carman formula in terms of porosity. Relative
permeabilities of liquid and vapor are expressed in terms of liquid
saturation based on experimental data. Capillary pressure is also based
on Leverett’s correlation from experimental data, which expresses
capillary pressure in terms of porosity, permeability, and surface tension.
Besides local thermodynamic equilibrium, the assumption of equal
chemical potentials in the liquid and vapor phases is employed. This
yields an additional thermodynamic relation that relates temperature,
saturation pressure, and liquid and vapor pressures.
Another area of application of multiphase transport in porous media is
the heat pipe technology. A comprehensive review of the physical
phenomena and applications of heat pipes is given by Tien (1975,1985a)
and Tien and Rohani (1972). Udell (1985) reported analytical and
experimental studies for a one-dimensional problem of heat pipe effect in
porous media. In the model developed, gravity forces, capillarity, and
change of phase were considered. The problem studied was a steady state
one. Leverett correlation relating capillary pressure to permeability,
porosity, and surface tension was also used in this study. Relative
permeability correlations found by Flatt and Klikoff (1959) were also
employed. Numerical solution and experimental work were carried out
for model verification in Udell’s work. Udell and Fitch (1985) reported a
further study considering the effects of noncondensible gases. In this
work, the effect of noncondensible gases on Kelvin’s equation is
explained and utilized. Some additional empirical correlations are intro-
duced from previous studies for the thermal conductivity of a partially
saturated granular medium and for the effective molecular diffusivity.
Almost all the models developed for analyzing problems of multiphase
transport processes and phase change utilize the assumption of local
thermodynamic equilibrium between the phases in the porous system.
 Convective and Radiative Heat Transfer in Porous Media 259

While this is a very good assumption for many practical applications, it

might not be a realistic assumption for cases in which the fluid velocity is
high and/or where the system is subjected to an intense hat flux. Under
such situations, it may not be possible to have local thermal equilibrium
between the solid and fluid phases. One of the few examples of studies
that do not utilize the local thermal equilibrium assumption was carried
out by Wong and Dybbs (1978). In this work, the volume-averaged
energy equations for a single phase fluid flow are obtained for a saturated
porous medium. The heat transfer between the solid and fluid phases is
formulated by a convection term. An approximate means of accounting
for the radiative heat transfer between these phases by a convective-type
heat transfer coefficient is explained. A relationship is also established for
the effective thermal conductivity of a saturated porous medium in terms
of the effective thermal conductivities of the individual phases.
Limited experimental work has been performed for obtaining transport
coefficients for flow in porous materials. One investigation was reported
by Singh and Dybbs (1979). Besides forming a model for multiphase
transport processes, they have carried out experimental work for
determining the permeability of the porous matrix and the effective
thermal conductivity of the fluid saturated porous system. The per-
meability measurement was performed in an isothermal condition. The
effective thermal conductivity was measured, for the case where the
porous matrix was fully saturated with stagnant fluid, by the steady-state
method of comparison. The sintered metallic porous media used for
forming the solid matrix in this study were copper and nickel.
Other experimental works considering phase change in packed beds
have been reported by Tsai et al. (1984). Their study investigated the
dryout of an indutively heated packed bed made of metal particles.
Water was used as the working fluid. The variation of the dryout heat
flux with mass flux was determined. Similar work studying the effect of
pressure on the dryout of a packed bed of steel particles was carried out
by Catton and Jakobsson (1987).
For experimental investigations related to transport processes and
phase change in porous media, measurement of different variables such
as temperature, pressure, and relative humidity may be performed by
thermocouples and pressure and relative humidity measuring devices,
which are available in relatively small sizes. However, to our knowledge,
a very reliable commercial device for detection of the liquid moisture and
for measurement of the liquid moisture content in a porous medium does
260 Chang-Lin Tien and Kambiz Vafai

not seem to be available. This is crucial for experimental work aiming at

investigating condensation problems in porous media. The design and
construction of a device for in situ measurement of the liquid content in a
fibrous insulation material was reported by Motakef and El-Masri (1985).
The device utilizes a small electrical resistance probe.
The subject of multiphase transport processes and phase change in
porous media is now a very active research area. Besides the applications
considered to date, new frontiers are being explored for the use of these
results in technological applications. Among the recent proposed systems
for spacecraft applications related to this subject are the packed-bed type
heat exchangers utilizing phase change materials for forming the packed
bed and heat pipes with porous wicks. Both of these require a thorough
understanding of the fundamentals and analysis of the transport processes
in porous media. From the research carried out to date, it may be seen
that these physical phenomena are relatively well understood. But the
models established to date lack general applicability. This is partly
because of the lack of universally accepted methods of definition for
transport coefficients and transport potentials and partly because of the
different assumptions employed in forming the model in each particular
problem considered. Further theoretical work is required for obtaining a
generalizable model. Experimental work for determining empirical cor-
relations for the transport coefficients in different porous materials is also
an important necessity.

V. Radiative Heat Transfer in Porous Beds

A. THERMAL
RAD~AT~ON
CHARACTERISTICS
OF POROUS
BEDS

The porous beds in which thermal radiation can become important are
usually multiphase systems consisting of solid particulates and gases. The
role of thermal radiation in these porous beds is of major importance in
the design of fluidized beds, packed beds, catalytic reactors, and many
other advanced energy conversion systems, especially when the beds
operate at high temperatures. Radiation in these porous beds brings up
additional difficulties and challenging aspects in the analysis. These
difficulties arise in two major areas: the development of a radiation
transfer model, which prescribes the propagation of radiation within an
 Convective and Radiative Heat Transfer in Porous Media 261

absorbing, emitting, and scattering medium, and the corresponding

radiative properties for the model, and the development of a heat
conduction and/or convection model with properly determined thermo-
physical properties.
Previous studies of radiative heat transfer through porous beds have
employed a variety of analytical and experimental techniques. Vortmeyer
(1978) summarized some earlier radiation models that use unit cell
representations for analyzing packed beds. Such cell and layer models, in
conjunction with Monte Carlo methods, are also used by Chan and Tien
(1974a) to evaluate radiative characteristics of packed beds of fixed
porosity and regular structure, by Yang et al. (1983) for randomly packed
beds of uniform spheres, and by Kudo et al. (1985), who examined
different types of packing and variation in volume fraction. Also
Borodulya and Kovensky (1983) used the cell approach by evaluating
exact view factors in the unit cell by assuming the surfaces to be
isothermal and diffuse.
Thermal radiation within these beds usually is the result of emission by
the hot walls and the gas-particle mixture. This radiation undergoes
complex interactions with the bed, primarily due to absorption and
scattering processes. The three primary radiative properties that charac-
terize the interactions of radiation with the particulate bed are the
scattering coefficient, the extinction coefficient (i.e., sum of scattering
and absorption coefficients), and the scattering phase function. These
radiative coefficients are defined as the fractions of the corresponding
energy losses from the propagating wave per length of travel. The units
for the absorbing, emitting, and scattering coefficients are inverse of
length, whereas the phase function, which specifies the scattering
distribution, is dimensionless. The radiative coefficients are functions of
the optical constants of the bed materials and of the particle size, shape,
and packing, while the optical constants are functions of the wavelength.
The phase function is a strong function of shape and varies from
predominantly forward scattering for large particles to semi-diffuse for
small ones.
In homogeneous media such as gases, absorption and emission are the
major radiative mechanisms. If the medium contains inhomogeneities,
such as the particles in packed or fluidized beds, the additional
mechanism of scattering is introduced. These absorption and scattering
processes are governed by electromagnetic field equations and their
associated boundary conditions at all interfaces. The resulting analytical
262 Chang-Lin Tien and Karnbiz Vafai

problem is formidable and is usually solved by using simplifying

assumptions: idealized geometry of the scatterers, independent scattering
and absorption, and homogeneous distribution of particles.
Scattering and absorption characteristics of a particle are governed by
three factors: the particle shape, the particle size relative to the
wavelength of the incident radiation, and the optical properties of the
particle and the background medium (Tien, 1985b). Three characteristics
of a single particle are described by the solution of the electromagnetic
field equations. Physically, they can be explained by the processes of
reflection, refraction, and diffraction. When an electromagnetic wave
strikes the particle surface, a portion of it is reflected while the remainder
penetrates the particle. The beam within the particle may experience
some absorption and multiple internal reflections before it escapes out of
the particle in different directions, giving rise to scattering. This
scattering is the contribution by refraction. The diffraction scattering
process originates from the bending of the incident beams near the edge
of the particle. Consequently, even a completely absorbing particle
scatters radiation.
Scattering and absorption characteristics of many particles in close
packing or fluidization can be obtained from the single-particle charac-
teristics. The procedure depends on the scattering regime of the system
to which the system of particles belongs. Based on the size parameter and
the particle volume fraction fv the regime map shown in Figure 4
(Brewster and Tien, 1982a; Yamada et al., 1986; Tien and Drolen, 1987;
Drolen et al., 1987) is divided into two parts: one where independent
theory is an adequate representation and the other where interparticle
interactions must be accounted for. The independent theory is based on
the assumption that each particle in the assembly scatters and absorbs
radiation unaffected by the presence of other particles. Thus, the
extinction and scattering of energy by the system is expressed by a simple
algebraic addition of the energy extinguished and scattered by each
parimary particle. The cross-section for the system of N particles is the
sum of the cross-sections of these particles, and the individual particles
are assumed to scatter and absorb radiation independently of the others.
In essence, the independent theory assumes no electromagnetic inter-
action between the various particles in the system and that each particle
has the same incidence. Interference of the waves scattered from
different particles is also neglected. Thus, the resultant properties of the
system are simply an algebraic sum of the corresponding properties of the
 Convective and Radiative Heat Transfer in Porous Media 263
10 I I 1
FDGS b N D CLOUDS I PACKED AND& 1
I
I FLUIDIZED BEDS I
/ I
u 10
d
w
L z
IPULVER I ED
I COAL
I
I
SCATTER ING
I ,COMBUSTION
a IC I
a 1\ I
2
w
/----------------

N
wY l
J
0
k
a
I6 * SOOT IN FLAMES
AND SMOG LAYERS

PARTICLE VOLUME FRACTION, f,

FIG. 4. Independent and dependent scattering regime map.

individual particles that constitute the system , where each particle is

assumed to be alone in the imposed incident field.
Intuitively the assumptions associated with the independent theory
seem to be valid when the clearance between particles is significantly
larger than the particle diameter, as well as when the particles are
randomly distributed in space and time. Furthermore, the ratio of
interparticle clearance to the diameter of the particles is not of
significance for randomly distributed homogeneous particles, whereas the
ratio of clearance to the wavelength is. Results using the independent
theory are far simpler, and wherever justified, it is advantageous to use
them over the exact expressions that accounts for interparticle effects.
Departure from independent theory occurs in a densely packed system,
where the close spacing of the particles renders invalid the assumption
that each particle acts independent of the others. Dependent effects are
introduced into the radiative characteristics by two mechanisms. The first
is the near-field interparticle effect by which the net incidence on the
264 Chang-Lin Tien and Kambiz Vafai

particles as well as the internal fields is modified and consequently both

the extinction and scattering characteristics of the system are changed.
The second is due to coherent addition of scattered radiation in the
far-field that is manifested by a change in the scattering characteristics
only (Tien and Drolen, 1987; Kumar and Tien, 1987). In reality,
dependent effects are always present but may be neglected under certain
conditions. It is thus important to quantify demarcation criteria that
separate the regions where the independent assumption is a good
approximation from those where dependent effects cannot be ignored
(Tien and Drolen, 1987).
Experimental meaurements of radiant transmission through packed
and fluidized media have been reported by various researchers. Experi-
mentally, the radiative characteristics of packed beds are determined by
first determining the extinction characteristics through direct transmission
measurements. Next, the scattering cross-section and phase function are
obtained by measuring the angular distribution of scattered thermal
radiation. Finally, the absorption characteristics are inferred from the
difference between the extinction and scattering of the incident radiation.
The radiation source in the experiment is either a laser or glow-bar
(Drolen et al., 1987; Brewster and Tien, 1982a; Hottel et al., 1971).
The experimental results consist of transmittance and reflectance data
for a planar slab containing the scatterers from the particulate bed. Data
reduction to yield the desired radiative properties is achieved by adopting
a suitable radiative transfer model. For example, Brewster and Tien
(1982a) used the two flux model (Brewster and Tien, 1982b) for the data
inversion. Chen and Churchill (1963) used an open-ended tubular electric
furnace as a blackbody source. The test section comprised of packed
spheres supported on a screen and the source was placed underneath.
Using thermocouples as detectors, the characteristics of glass, aluminum
oxide, steel, and silicon carbide spheres, cylinders, and irregular grains
were found for various source temperatures. Two-flux models, with the
inclusion of an emission term, were used for the data inversion.
Whereas characteristics of packed beds can be measured in ex situ
situations, the thermal radiation characteristics of fluidized beds have to
be measured in sifu in order to preserve the operating conditions. This is
achieved by laser scattering techniques. Alavizadeh et al. (1984) have
designed an instrument for the measurement of the radiative component
of the heat transfer in a high-temperature gas fluidized bed. The design
uses a silicon window to transmit the radiative flux to a thermopile
detector located at the base of a cavity.
 Convective and Radiative Heat Transfer in Porous Media 265

B. MODELING
OF RADIATIVE
TRANSFER
IN POROUS
BEDS

Computation of the transport of thermal radiation in a particulate

system requires an accurate knowledge of the primary radiative prop-
erties. These properties are adopted primarily due to the following
considerations: (i) the theory of electromagnetic interaction with par-
ticles, (ii) direct experimental measurements, and (iii) inferred measure-
ments. For example, the absorption coefficient cannot be directly
measured from light scattering experiments. It is obtained indirectly by
measuring the scattering and extinction losses from the incident beam
and evaluating the difference between the corresponding extinction and
the scattering coefficients.
The propagation of radiation within an absorbing, emitting, and
scattering medium is governed by the equation of transfer (Kerker, 1961;
Siege1 and Howell, 1981; Ozisik, 1973):

ea * Vl,(r, en) = -(GA + U ~ A ) Z A ( ~en)

, + uaAkA(T(r))

where ZA is the monochromatic radiation intensity, T the medium local

temperature, r the position vector, eR the unit vector in the direction of
consideration, and 51 the solid angle centered around en. The coefficients
are denoted by 0 and the subscripts a and s refer to absorption and
scattering, respectively. The first term on the right hand side of the
equation of transfer represents the attenuation of intensity due to
absorption and scattering, the second term represents the gain due to
emission, and the last term is the gain due to the in-scattering into the
direction eR from all other directions. The intensity ZA is defined as the
energy per unit area per unit solid angle per unit wavelength and the
scattering phase function @(en,+ e,) is a specification of the radiation
intensity scattered from the direction eR, into the direction under
consideration, normalized by the isotropic scattered radiation intensity,
i.e., @(en.+ e,) = 1 for isotropic scattering. It should be noted that the
equation of transfer treats the medium as a continuum where each
volume element absorbs, emits, and scatters radiation. The exact
positions of the different particles in the volume are not considered; only
volume-averaged values of the radiative properties are used. Other
methods, which do not treat the medium as a continuum, but instead
take into account the position of particles and the boundaries between
266 Chang-Lin Tien and Kambiz Vafai

the solid and the gas phase, are termed the discrete models of radiative
transfer.
Neglecting the viscous dissipation, while including radiation, convec-
tion, and conduction modes of heat transfer (Siege1 and Howell, 1981;
Ozisik, 1973), will lead to the following energy equation

+ (V *
1
V ) T = V * kVT + V - qr,
-
where V qr is given by

-
V qr(r) = 4 n oaAZbA(T(r))
J* ,,J
dA - oaA &(r, en) dS2 dA.
JA
(5.3)

Here, T is the temperature, v the velocity vector, p the density, C,, the
specific heat, k the thermal conductivity, and t time. The energy equation
(5.2) is an integro-differential equation which in general does not lend
itself to straight-forward solutions even for simple cases. Furthermore,
due to the nature of the equation, numerical computations tend to be on
the heavy side.
Several methods have been used for modeling radiative transfer in
packed and fluidized beds. These approaches can be classified, according
to the modeling of the medium itself, into two major groups: discon-
tinuous or discrete models and continuous or pseudo-continuous models.
In discontinuous or discrete models, the medium is considered as a
regular assembly of units or cells of idealized geometry, resulting in a
simple algebraic formulation of the problem. In this case, volume-
averaged radiative properties of the medium, determined from the
radiative characteristics of particles, are used in the equation of transfer.
The second approach visualizes the medium as a random collection of
particles with some number density NIV, a reasonable approximation
when the characteristic dimension of the system is much larger than the
characteristic size of the particles in the system. In this case, the radiative
transfer modeling involves either the integro-differential equation of
radiative transfer or a simplified version of this equation, or Monte Carlo
and/or ray tracing methods.
Discontinuous models characteristically treat a particulate system as a
regular assembly of cells or units of idealized geometry such as parallel
flat plates, close-packed spheres, and cubic-packed spheres. In these
models, the scattering diagram of a unit or cell is first determined.
Standard resistance network or layer theory approaches are then applied
 Convective and Radiative Heat Transfer in Porous Media 267

to calculate radiation transfer through the system. Vortmeyer (1978)

reviewed many models that fall into this category for predicting radiant
conductivity. Chan and Tien (1974a), Kudo et al. (1985), and Borodulya
and Kovensky (1983) used different unit cells for calculating the optical
properties. In these works, Chan and Tien (1974a) examined a single
cubic cell that is representative of a cubic packed geometry. Using ray
tracing techniques and assuming specular surfaces, they calculated the
optical properties of a single layer based on the fraction of rays
transmitted, reflected and absorbed by a single cell. In the work of
Borodulya and Kovensky (1983), rather than using the ray tracing,
diffuse and isothermal surfaces were assumed and then the transmitted,
absorbed, and reflected fluxes were computed. In contrast to Borodulya
and Kovensky (1983), Kudo et al. (1985) used Monte Carlo techniques to
calculate the reflected, transmitted, and absorbed energy.
Predictions of radiative transfer in packed/fluidized beds based on
discontinuous or discrete models are expected to be very sensitive to the
packing geometry assumed. Nevertheless, the discontinuous approach is
the natural one to apply in order to predict radiative transfer in many
discrete systems such as nuclear fuel rods, solid-matrix beds, screens, and
packed beds of low tube-particle diameter ratio, such as the packed-bed
tubular reactors widely used in the chemical industry. Careful considera-
tion of the solid particle arrangement and recourse to a ray
tracing/Monte Carlo method or to other more complicated methods is
then required to accurately predict radiative transfer in these systems.
Furthermore, the modification of the ray tracing technique to account for
energy diffracted, especially for intermediate particle size parameters, is
then required. The application of the theory of geometric diffraction,
originally developed by Keller (1962) to radiative transfer problems
deserves special attention.
In the continuous model approach, the particulate medium is modeled
as a random assembly of particles with some number density. The
medium extinction characteristics can be calculated by a Monte Carlo
and/or ray tracing approach or by the solution of an integro-differential
transport equation such as Equation (5.1) or a simplified version of this
equation such as the two-flux model and the diffusion approximation.
The Monte Carlo and/or ray tracing models are best represented by
the work of Yang et al. (1983), who presented a novel approach for
modeling packed-bed radiative transfer. In their model, a randomly
packed bed of spheres is created mathematically, using a model that
268 Chang-Lin Tien and Kambiz Vafai

describes the slow sequential settling of individual rigid spheres of equal

diameter, where each sphere must be supported by at least three others,
and no sphere can overlap.
For this packing, rays are most likely to have their first interaction at
about one quarter of a sphere diameter. It was also found that the mean
penetration depth is about 0.66 sphere diameter, and that virtually all
rays have hit a sphere surface after traveling a distance of six diameters.
The above-mentioned probability information was then used to perform a
ray tracing analysis on thick beds. A ray was projected in a random
direction across a randomly chosen distance based upon the cumulative
distribution function. The process was then continued by randomly
choosing the position of a sphere around the endpoint of this ray and
assuming specular reflections at the sphere surface. Each of the models
that were discussed neglects the diffraction contribution by the particles.
This is a reasonable assumption for large a (a> 1 0 ) , where a is the size
parameter, since the diffracted energy is concentrated in the direction of
propagation. However, for intermediate and small a the diffracted
energy can no longer be considered as part of the propagating beam.
Thus, reflection at the particle surface is not the only contributor to the
scattered energy; diffraction must be considered as well.
The integro-differential approach involves the solution of an equation
of transfer as if the medium were a continuum. The extinction charac-
teristics of this continuum are based on the properties of the discrete
particles. Chen and Churchill (1963) used a two-flux approximation of the
equation of transfer to correlate their experimental data for transmission
of radiation through a packed bed of spheres.
The diffusion approximation method greatly simplifies the equation of
transfer in the case of optically thick media by assuming the radiative
transfer to be a diffusion process. Wang and Tien (1983) and Brewster
and Tien (1982b) have presented a diffusion based solution of the
radiative flux assuming a two-flux model for the radiative transfer in a
one-dimensional planar geometry.
The two-flux model was extended by Chandrasekhar (1960) by
approximating the radiation field in terms of a number N of discrete
streams. This reduced the problem to the solution of N coupled
differential equations. The direction of these streams, pi = cos mi,are
typically chosen so as to correspond to the zeros of the Legendre
polynomials PN(p). The integral in Equation (5.1) can then be approxi-
mated by Gaussian quadrature. The model just discussed is called the
 Convective and Radiative Heat Transfer in Porous Media 269

method of discrete ordinates and can be thought of as the exact solution

to the problem given a large enough N, say N 220. It is best applied to
problems involving anisotropic scattering and/or requiring directional
results. A simple algorithm for solving the general equation of transfer
using discrete ordinates is presented by Kumar et af. (1988). Other
methods for analyzing the radiative transfer equation have been reviewed
by Viskanta (1982, 1984) and Menguc and Viskanta (1983).
A few general comments are in order regarding the models that have
been proposed to predict the radiative transfer in particulate systems.
Models have been developed to predict primarily the transmission of
radiation and/or the effective radiant conductivity of the particulate
medium. Corresponding experiments have been performed to test these
two different approaches. Of these two major experimental methods, the
measurement of radiation transmittance through an evacuated and
isothermal bed of spheres is the fundamental and by far more reliable
method for evaluating theoretical models since these experiments com-
pletely isolate radiation from other modes of heat transfer. An excellent
review of many models developed to predict radiant conductivity can be
found in the literature (Vortmeyer, 1978). By far the most quoted
experimental data relating to radiant transport in packed beds are those
of Chen and Churchill (1963). Many of the authors previously referenced
have compared results of their respective models to these experimental
data. This experiment was the first to isolate the radiant mode of heat
transfer from the convective and conductive modes. This was achieved by
illuminating a bed of spheres (2-16 diameters deep) with a modulated,
high-temperature (700-1366 K) blackbody source and measuring the
transmitted energy via a spectrally independent detector.
Based on experimental comparisons it is concluded that the two-flux
model predicts consistently high values of the exchange factor, a quantity
which is used in the definition of the radiative conductivity. Particularly,
as the emittance is increased the experimental data indicate an increase in
exchange factor E, whereas the two-flux model predicts a decrease. In
the two-flux model, the individual spheres are assumed to be isothermal
and scattering independently. These may be poor assumptions for welded-
steel spheres. A model proposed by Vortmeyer (1978), in which he modi-
fied the plane-layer model to incorporate transmission of energy through
void spaces in the layer, appears to give the best predictions. A simple
model proposed by Schotte (1960), in which the exchange factor is approxi-
mated by the particle emissivity, also gives surprisingly good results.
270 Chang-Lin Tien and Kambiz Vafai

C. MULTIMODE
HEATTRANSFER
IN POROUS
BEDS

Combinations of radiation with the other modes of heat transfer were

first studied for cases where only radiation and conduction were present.
These types of multimode heat transfer processes in porous beds have
been the subject of many studies (Chan and Tien, 1974b; Bergquam and
Seban, 1971). Simultaneous radiative-convective transfer in packed and
fluidized beds has also been studied (Echigo et al., 1974; Tabanfar and
Modest, 1987). In fluidized bed systems, convection and radiation are the
important mechanisms of energy transfer as indicated by experimental
studies (Goshayeshi et al., 1986). Solutions have been obtained by
incorporating simplifying approximations such as an isotropically scatter-
ing gray medium (Yener and Ozisik, 1986), and a linearly anisotropic
scattering medium (Azad and Modest, 1981). A review of some of these
topics in heat transfer in fluidized beds was presented by Saxena et al.
(1978).
Equations (5.1)-(5.3) are all formulated in terms of volumetric-
averaged variables and are coupled through temperature. This set of
equations is a highly nonlinear system of differential and integro-
differential equations and it is a formidable task even to attempt a
numerical solution. The general equation of radiative transfer alone has
been the subject of many investigations, with several approximate
methods being proposed for its solution (Chandrasekhar, 1960; Siege1
and Howell, 1981; Viskanta, 1984; Menguc and Viskanta, 1983). An
iterative method or a variant of this method has been traditionally used
to solve the coupled energy and radiative transfer equation. In this
method, a temperature profile is assumed and used in Equation (5.3) to
calculate the divergence of the radiative heat flux, this being then
substituted in Equation (5.2) for recalculating temperature distributions.
The solution of these complex problems has been substantially facilitated
with the recent development of the Differential-Discrete-Ordinate
(DDO) method (Kumar et al., 1988) for solving the general equation of
radiative transfer. This simple but powerful method allows the simul-
taneous solution of Equations (5.1)-(5.3) in a direct and computationally
efficient way. The method uses a discrete ordinate technique to reduce
the integro-differential equation, Equation (5. l),to a system of ordinary
differential equations. The resulting set of coupled differential equations
are then solved by utilizing the existing software routines.
Heat transfer by simultaneous conduction and radiation between two
 Convective and Radiative Heat Transfer in Porous Media 271

reflecting surfaces with an intervening medium that absorbs, emits, and

scatters thermal radiation is a problem of considerable practical impor-
tance in stagnant packed beds, such as microsphere insulation and
nuclear fuel rods. There has been a significant number of investigations
concentrating on the conduction radiation modeling and interactions
(e.g., Viskanta, 1982; Yuen and Wong, 1980; Bergquam and Seban,
1971; Chan and Tien, 1974b; Vortmeyer, 1974; Vafai and Ettefagh,
1988).
The solution of combined convection and radiation in a particulate
medium is much more complicated than the solution of conduction
radiation problems because of the difficulties in the modeling of the fluid
in these systems. The understanding of convection heat transfer in
packed beds has developed substantially in recent years. Theoretical
models that include wall porosity variation effects and inertia effects
(non-Darcy effects) have yielded results in excellent agreement with
experiments (Hunt and Tien, 1987). The problem of combined convec-
tion and radiation in these systems has not been fully investigated.

Acknowledgments

The authors wish to thank Dr. S. J. Kim, Dr. M. Sozen, and Dr. H. C.
Tien for their careful reading of this manuscript.

References

Alavizadeh, N., Adams, R. L., Welty, J. R., and Goshayeshi, A. (1984). An instrument for
local radiative heat transfer measurement in a gas-fluidized bed at elevated tempera-
tures. ASME HTD 31, 1-8.
Azad, F. H., and Modest, M. F. (1981). Combined radiation and convection in absorbing,
emitting and anisotropically scattering gas-particulate tube flow. Int. J . Heat Mass
Transfer 24, 1681-1698.
Baladi, J. Y.,Ayers, D. L., and Schoenhals, R. J. (1981). Transient heat and mass transfer
in soils. Int. J. Heat Mass Transfer 24, 449-458.
Balakrishnan, A. R., and Pei, D. C. T. (1974). Heat transfer in tked bed. Industrial and
Engineering Chemical Process Design Development W, 441-446.
Bankvall, C. G. (1974). Natural convection in vertical permeable space. Warme und
Stofubertragung 7 , 22-30.
Beavers, G. S., and Joseph, D. D. (1967). Boundary conditions at a naturally permeable
wall. 1. Fluid Mech. 30, 197-207.
Beckermann, C., Ramadhyani, S., and Viskanta, R. (1987). Natural convection flow and
272 Chang-Lin Tien and Kambiz Vafai
heat transfer between a fluid layer and a porous layer inside a rectangular enclosure.
ASME J. Heat Transfer 109, 363-370.
Bejan, A. (1978). Natural Convection in an infinite porous medium with a concentrated
heat source. J. Fluid Mech. 89, 97-107.
Bejan, A. (1979). On the boundary layer regime in a vertical enclosure filled with a porous
medium. Lett. Heat Mass Transfer 6,93-102.
Bejan, A. (1983a). The boundary layer regime in a porous layer with uniform heat flux
from the side. Int. J. Heat Mass Transfer 26, 1339-1346.
Bejan, A. (1983b). Natural convection heat transfer in a porous layer with internal flow
obstructions. Int. J. Heat Mass Transfer 26, 815-822.
Bejan, A. (1984). “Convection Heat Transfer.” Wiley, New York.
Bejan, A., and Anderson, R. (1981). Heat transfer across a vertical impermeable partition
imbedded in a porous medium. Int. J. Heat Mass Transfer 24, 1237-1245.
Bejan, A., and Anderson, R. (1983). Natural convection at the interface between a vertical
porous layer and an open space. ASME J. Heat Transfer 105, 124-129.
Bejan, A,, and Poulikakos, D. (1982). Natural convection in an attic-shaped space filled
with porous material. ASME 1. Heat Transfer 104,241-247.
Bejan, A., and Poulikakos, D. (1984). The non-Darcy regime for vertical boundary layer
natural convection in a porous medium. Int. J. Heat Mass Transfer 27, 717-722.
Bejan, A., and Tien, C. L. (1978). Natural convection in a horizontal porous medium
subject to an end-to-end temperature difference. ASME 1. Heat Transfer 100, 191-198.
Bejan, A., and Tien, C. L. (1979). Natural convection in horizontal space bounded by two
concentric cylinders with different end temperatures. Int. J. Heat Mass Transfer 22,
919-927.
Berger, D., and Pei, D. C. T. (1973). Drying of hygroscopic capillary porous solids-A
theoretical approach. Int. J. Heat Mass Transfer 16,293-301.
Bergquam, J. B., and Seban, R. A. (1971). Heat transfer by conduction and radiation in
absorbing and scattering materials. ASME J. Heat Transfer 93,236-239.
Blythe, P. A., Daniels, P. G., and Simpkins, P. G. (1982). Thermally driven cavity flows in
porous media I. The vertical boundary layer structure near the corners. Proc. Roy.
SOC. London Ser. A 380, 119-136.
Blythe, P. A., Simpkins, P. G., and Daniels, P. G. (1983). Thermal convection in a cavity
filled with a porous medium: A classification of limiting behaviors. Int. J. Heat Mass
Transfer 26, 701-708.
Bories, S. A., and Combarnous, M. A. (1973). Natural convection in a sloping porous
layer. J. Fluid Mech. 57, 63-79.
Borodulya, V. A., and Kovensky, V. I. (1983). Radiative heat transfer between a fluidized
bed and a surface. Int. J. Heat and Mass Transfer 26, 277-287.
Brewster, M. Q., and Tien, C. L. (1982a). Radiative transfer in packed fluidized-beds:
Dependent versus independent scattering. ASME J. Heat Transfer 104,573-579.
Brewster, M.Q., and Tien, C. L. (1982b). Examination of the two-flux model for radiative
transfer in particulate systems. Int. J. Heat Mass Transfer 25, 1905-1907.
Buretta, R. J., and Berman, A. S. (1976). Convective heat transfer in a liquid saturated
porous layer. J. Appl. Mech. 43, 249-253.
Burns, P. J., and Tien, C. L. (1979). Natural convection in porous media bounded by
concentric spheres and horizontal cylinders. Int. J. Heat Mass Transfer 22, 929-939.
Burns, P. J., Chow, L. C., and Tien, C. L. (1977). Convection in a vertical slot filled with
porous insulation. Int. J. Heat Mass Transfer 20, 919-926.
Caltagirone, J. P. (1976). Thermoconvective instab es in a porous medium bounded by
two concentric horizontal cylinders. J. Fluid Mech. 76, 337-362.
 Convective and Radiative Heat Transfer in Porous Media 273
Caltagirone, J. P., and Bories, S. A. (1985). Solutions and stability criteria of natural
convective flow in an inclined layer. J . Fluid Mech. 155, 267-287.
Catton, I., and Jakobsson, J. 0. (1987). The effect of pressure on dryout of a saturated bed
of heat-generating particles. ASME J . Heat Transfer 109, 185-195.
Chan, C. K., and Tien, C. L. (1974a). Radiative transfer in packed spheres. ASME J . Heat
Transfer 96, 52-58.
Chan, C. K., and Tien, C. L. (1974b). Combined radiation and conduction in packed
spheres. In “Proceedings of the 5th International Heat Transfer Conference,” Vol. 1,
Tokyo, Japan. pp. 72-74.
Chan, B. K. C., hey, C. M., and Barry, J. M. (1970). Natural convection in enclosed
porous media with rectangular boundaries. ASME J . Heat Transfer 2, 21-27.
Chandrasekhar, S. (1960). “Radiative transfer.” Dover, New York.
Chang, 1. D., and Cheng, P. (1983). Matched asymptotic expansions for free convection
about an impermeable horizontal surface in a porous medium. Int. J . Heat Mass
Transfer 26, 163-173.
Chen, J. C., and Churchill, S. W. (1963). Radiant heat transfer in packed beds. AIChE
Journal 9, 35-41.
Chen, C. K., Hung, C. I., and Cleaver, J. W. (1987). NonDarcian effects on vertical plate
transient natural convection in porous media with high-porosities. In “Proceedings of
the ASME-JSME Thermal Engineering Joint Conference”, Vol. 2, pp. 313-318.
Honolulu, Hawaii.
Cheng, P. (1977). Combined free and forced convection flow about inclined surfaces in
porous media. Int. J. Heat Mass Transfer 20, 807-814.
Cheng, P. (1978a). Heat Transfer in geothermal systems. Adu. Heat Transfer 14, 1-105.
Cheng, P. (1978b). Convective heat transfer in porous layers by integral methods. Lett.
Heat Mass Transfer 5 , 243-252.
Cheng, P. (1981). Thermal dispersion effects in nonDarcian convective flows in a saturated
porous medium. Len. Heat Mass Transfer 8, 267-270.
Cheng, P. (1982). Mixed convection about a horizontal cylinder and a sphere in a saturated
porous medium. Int. J. Heat Mass Transfer 25, 1245-1247.
Cheng, P., and Ali, C. L. (1981). An experimental investigation of free convection about
an inclined surface in a porous medium. ASME Paper No. 81-HT-85.
Cheng, P., and Chang, I. D. (1976). Buoyancy induced flows in a saturated porous medium
adjacent to impermeable horizontal surfaces. Int. J . Heat Mass Transfer 19,1267-1272.
Cheng, P., and Chang, I. D. (1979). Convection in a porous medium as a singular
perturbation problem. Lett. Hear Mass Transfer 6 , 253-258.
Cheng, P., and Hsu, C. T. (1984). Higher-order approximations for Darcian free convective
flow about a semi-infinite vertical flat plate. ASME J . Heat Transfer 106, 143-151.
Cheng, P., and Hsu, C. T. (1986). Fully-developed, forced convective flow through an
annular packed-bed with wall effects. Int. 1. Heat Mass Transfer 29, 1843-1853.
Cheng, P., and Minkowycz, W. J . (1977). Free convection about a vertical plate embedded
in a saturated porous medium with application to heat transfer from a dike. J.
Geophys. Res. 82, 2040-2044.
Cheng, P., and Pop, I. (1984). Transient free convection about a vertical flat plate
embedded in a porous medium. Int. J . Engng. Sci. 22,253-264.
Cheng, P., and Vortmeyer, D. (1988). Transverse thermal dispersion and velocity profile in
a packed bed with forced convective flow. Chem. Engng. Sci. 43,2523-2532.
Cheng, P., and Zheng, T. M. (1985). Thermal dispersion effects in forced convection plume
above a horizontal line source of heat in a packed bed. Int. Comm. Heat Mass
Transfer.
274 Chang-Lin Tien and Kambiz Vafai
Cheng, P., Ah, C. L., and Verma, A. K. (1981). An experimental study of nonDarcian
effects in free convection in a saturated porous medium. Leu. Hear Mass Transfer 8 ,
261-265.
Chuah, Y. K., and Carey, V. P. (1985). Analysis of boiling heat transfer and two-phase
flow in porous media with nonuniform porosity. In?. J. Hear Muss Transfer 28,
147-154.
Combarnous, M. A., and Bia, P. (1971). Combined free and forced convection in porous
media. SOC.Per. Eng. J . 11, 399-405.
Combarnous, M. A., and Bones, S. A. (1975). Hydrothermal convection in saturated
porous media. A d v . Hydrosci. 10, 231-307.
DeVries, D. A. (1958). Simultaneous transfer of heat and moisture in porous media. Trans.
Amer. Geophys. Union 39, 909-916.
De Wasch, A. P., and Froment, G. F. (1972). Heat transfer in packed beds. Chem. Engng.
Sci. 27, 567-576.
Dinulescu, H. A,, and Eckert, E. R. G. (1980). Analysis of the one-dimensional moisture
migration caused by temperature gradients in a porous medium. In?. J. Hear Mass
Transfer 23, 1069-1078.
Drolen, B. L., Kumar, S., and Tien, C. L. (1987). Experiments on dependent scattering of
radiation. In “Proceedings of the AIAA 22nd Thermophysics Conference”, Honolulu,
Hawaii.
Dybbs, A., and Schweitzer, S. (1973). Conservation equations for nonisothermal flow in
porous media. 1. Hydrology 20, 171-180.
Echigo, R., Hasegawa, S., and Nakano, S. (1974). Simultaneous radiative and convective
heat transfer in a packed bed with high porosity. Trans. JSME 40,479.
Eckert, E. R. G., and Faghri, M. (1980). A general analysis of moisture migration caused
by temperature differences in an unsaturated porous medium. In?. J . Hear Mass
Transfer 23, 1613-1623.
Eckert, E. R. G., and Pfender, E. (1978). Heat and mass transfer in porous media with
phase change. In “Proceedings of the Sixth International Heat Transfer Conference,”
Vol. 6, pp. 1-12. Hemisphere Publishing Corp., Washington, D.C. Toronto, Canada.
Elder, J. W. (1967a). Steady free convection in a porous medium heated from below. J .
Fluid Mech. 27, 29-48.
Elder, J. W. (1967b). Transient convection in a porous medium. J. Fluid Mech. 27,
609-623.
Ettefagh, J., and Vafai, K. (1988). Natural convection in open-ended cavities with a porous
obstructing medium. In?. J. Hear Mass Transfer 31, 673-693.
Evans, G. H., and Plumb, 0. A. (1978). Natural convection from a vertical isothermal
surfaces embedded in a saturated porous medium. ASME Paper No. 78-HT-85.
Facas, G. N., and Farouk, B. (1983). Transient and steady-state natural convection in a
porous medium between two concentric cylinders. ASME J. Hear Transfer 105,
660-663.
Fand, R. M., Steinberger, T. E., and Cheng, P. (1986). Natural convection heat transfer
from a horizontal cylinder embedded in a porous medium. Int. J . Hear Mass Transfer
29, 119-133.
Farouk, B., and Shayer, H. (1985). Natural convection around a heated cylinder buried in a
saturated porous medium. I n “Natural Convection in Porous Media,” ASME
HTD-Vol. 56, pp. 181-189.
Faust, C. R., and Mercer, J. W. (1979). Geothermal reservoir simulation: 1. Mathematical
models for liquid- and vapor-dominated hydrothermal systems. Water Resources Res.
15, 23-30; Geothermal reservoir simulation: 2. Mathematical models for liquid- and
vapor-dominated hydrothermal systems. Water Resources Res. 15,31-46.
 Convective and Radiative Heat Transfer in Porous Media 275

Fernandez, R. T., and Schrock, V. E. (1982). Natural convection from cylinders buried in a
liquid-saturated porous medium. In “Proc. 7th Int. Heat Transfer Conference,”
Miinchen, Germany. Vol. 2, pp. 335-340.
Flamant, G., and Arnaud, G. (1984). Analysis and theoretical study of high-temperature
heat transfer between a wall and a fluidized bed. Int. J . Heat Mass Transfer 27,
1725- 1 735.
Flatt, I., and Klikoff, W. A. (1959). Effect of functional wettability on multiphase flow
through porous media. AIME Technical Note tt2043. AIME Transactions 216,256.
Gardner, W.,and Widtsoe, J. A. (1920). The movement of soil moisture. Soil Science 11,
215-232.
Georgiadis, J., and Catton, I. (1986). Prandtl number effect on Benard convection in
porous media. ASME J. Heat Transfer 108,284-290.
Goshayeshi, A., Welty, J. R., Adams, R. L., and Alavizadeh, N. (1986). Local heat
transfer coefficients for horizontal tube arrays in high-temperature large-particle
fluidized beds: An experimental study. ASME J . Heal Transfer 108,907-912.
Gupta, V. P., and Joseph, D. D. (1973). Bounds for heat transport in a porous layer. J.
Fluid Mech. 57,491-514.
Haajizadeh, M., and Tien, C. L. (1983). Natural convection in a rectangular porous cavity
with one permeable endwall. ASME J. Heat Transfer 105,803-808.
Haber, S., Shavit, A., and Dayan, J. (1984). The effect of heat convection on drying of
porous semi-infinite space. Int. J . Heat Mass Transfer 27,2347-2353.
Harmathy, T. Z. (1969). Simultaneous moisture and heat transfer in porous systems with
particular reference to drying. Ind. Eng. Chem. Fundamentals 8, 92-103.
Havstad, M. A., and Burns, P. J. (1982). Convective heat transfer in vertical cylindrical
annuli filled with a porous medium. Int. 1. Heat Mass Transfer 25, 1755-1766.
Hickox, C. E. (1981). Thermal convection at low Rayleigh number from concentrated
sources in porous media. ASME J . Heat Transfer 103,232-236.
Hickox, C. E., and Gartling, D. K. (1981). A numerical study of natural convection in a
horizontal porous layer subjected to an end-to-end temperature difference. ASME J .
Heat Transfer 103,797-802.
Hickox, C. E., and Gartling, D. K. (1982). A numerical study of natural convection in a
vertical annular porous layer. ASME Paper No. 82-HT-68.
Hickox, C. E., and Watts, H. A. (1980). Steady thermal convection from a concentrated
source in a porous medium. ASME J. Heat Transfer 102,248-253.
Hong, J. T., and Tien, C. L. (1987). Analysis of thermal dispersion effect on vertical-plate
natural convection in porous media. Int. J. Heat Mass Transfer 30, 143-150.
Hong, J . T., Tien, C. L., and Kaviany, M. (1985). NonDarcian effects on vertical-plate
natural convection in porous media with high porosities. Int. J. Heat Mass Transfer 28,
2149-2157.
Horne, R. N., and O’Sullivan, M. J. (1974). Oscillatory convection in a porous medium
heated from below. J . Fluid Mech. 66, 339-352.
Horne, R. N., and O’Sullivan, M. J. (1978). Convection in a porous medium heated from
below: The effect of temperature dependent viscosity and thermal expansion
coefficient. ASME J . Heat Transfer 100,448-452.
Hottel, H. C., Sarotim, A. F., Dalzell, W. H., and Vasalos, I. A. (1971). Optical
properties of coatings, effect of pigment concentration. AIAA Journal 9, 1895-1898.
Hsu, C. T., and Cheng, P. (1985). The Brinkman model for natural convection about a
semi-infinite vertical flat plate in a porous medium. Int. J. Heat Mass Transfer 28,
683-697.
Huang, C. L. D. (1979). Multi-phase moisture transfer in porous media subjected to
temperature gradients. Int. J. Heat Mass Transfer 22, 1295-1307.
276 Chang-Lin Tien and Kambiz Vafai
Huang, C. L. D., Siang, H. H., and Best, C. H. (1978). Heat and moisture transfer in
concrete slabs. In “Proc. 8th U.S. National Congress of Applied Mechanics,” LQS
Angeles, CA.
Huenefeld, J. S., and Plumb, 0. A. (1981). A study of nonDarcy natural convection
from a vertical heated surface in a saturated porous medium. ASME Paper No.
81-HT-45.
Hunt, M. L., and Tien, C. L. (1987). NonDarcian convection in cylindrical packed beds. In
“Proceedings of the ASME-JSME Thermal Engineers Joint Conference,” Vol. 2, pp.
433-438. Honolulu, Hawaii.
Jonsson, T., and Catton, I. (1987). Prandtl number dependence of natural convection in
porous media. A S M E J. Heat Transfer 109,371-377.
Joshi, Y., and Gebhart, B. (1984). Vertical natural convection flows in porous media:
Calculations of improved accuracy. Int. J . Heat Mass Transfer 27, 69-75.
Kalthoff, O.,and Vortmeyer, D. (1980). Ignition/extinction phenomena in a wall cooled
fixed bed reactor, Chem. Engng Sci. 35, 1637-1643.
Kaneko, T., Mohtadi, M. F., and Aziz, K. (1974). An experimental study of natural
convection in inclined porous media. Int. J . Hear Mass Transfer 17,485-496.
Katto, Y., and Masuoka, T. (1967). Criterion for the onset of convective flow in a fluid in a
porous medium. Int. J . Heat Mass Transfer 10, 297-309.
Kaviany, M. (1985a). Laminar flow through a porous channel bounded by isothermal
parallel plates. Int. J . Heat Mass Transfer 28, 851-858.
Kaviany, M. (1985b). An integral solution to vertical plate natural convection in porous
media. ASME Paper No. 85-WAIHT-1.
Kaviany, M. (1987). Boundary-layer treatment of forced convection heat transfer from a
semi-infinite flat plate embedded in porous media. A S M E J. Heat Transfer 109,
345-349.
Kaviany, M., and Mittal, M. (1987). Natural convection heat transfer from a vertical plate
to high permeability porous media; an experiment and an approximate solution. Int. J.
heat Mass Transfer 30,967-977.
Keller, J. B. (1962). Geometrical theory of diffraction. J . Optical Society of America 52,
116-130.
Kerker, M. (1961). “The Scattering of Light and Other Electromagnetic Radiation.”
Academic Press, New York.
Kim, S . J., and Vafai, K. (1989). Analysis of natural convection about a vertical plate
embedded in a porous medium. Int. J. Heat Mass Transfer 32, 665-677.
Kimura, S., Bejan, A., and Pop, I. (1985). Natural convection near a cold plate facing
upward in a porous medium. A S M E J . Hear Transfer 107,819-825.
Klarsfeld, S. (1963). Champs de temperature associes aux mouvements de convection
naturelle dons un milieu poreux limite. Rev. Gen. Thermique 108, 1403-1424.
Koh, J. C. Y.,and Colony, R. (1974). Analysis of cooling effectiveness for porous material
in a coolant passage. A S M E J . Heat Transfer %, 324-330.
Koh, J. C. Y., and Stevens, R. L. (1975). Enhancement of cooling effectiveness by porous
material in coolant passage. A S M E J . Heat Transfer W,309-311.
Krischer, 0. (1938). Fundamental law of moisture movement in drying by capillary flow
and vapor diffusion. VDIZ 82,373-378.
Krischer, 0.(1940a). The heat, moisture and vapor movement during drying porous
materials. VDZZ, Beih. 1,17-24.
Krischer, 0. (1940b). Heat and mass transfer in drying. VDI-Forschungsh. 415.
Kudo, K., Yang, W. J., Taniguchi, H., and Hayasaka, H. (1985). Radiative heat transfer in
packed spheres by Monte Carlo method. In “Proceedings of the U.S.-Japan Heat
Transfer Joint Seminar,” 5-4, pp. 1-12.
 Convective and Radiative Heat Transfer in Porous Media 277
Kumar, S., and Tien, C. L. (1987). Dependent scattering and absorption of radiation by
small particles. In “24th National Heat Transfer Conference,” Vol. 72, pp. 1-7.
Pittsburgh, Pennsylvania.
Kumar, S., Majumdar, A., and Tien, C. L. (1988). The differential-discrete-ordinate
method for solving the general equation of radiative transfer. In “25th National Heat
Transfer Conference,” pp. 24-27. Houston, Texas.
Lapwood, E. R. (1948). Convection of a fluid in a porous medium. Proc. Camb. Phil. SOC.
44,508-521.
Lauriat, G . , and Prasad, V. (1987). Natural convection in a vertical porous cavity: A
numerical study for Brinkman-extended Darcy formulation. ASME J . Heat Transfer
109,688-696.
Lerou, J. J., and Froment, G. F. (1977). Velocity, temperature and conversion profile in
fixed bed catalytic reactors. Chem. Engng. Sci. 32, 853-861.
Lewis, W. K. (1921). The rate of drying of solid materials. Ind. Eng. Chem. W, 427-432.
Li, C., and Finlayson, B. A. (1977). Heat transfer in packed beds-A reevaluation. Chem.
Engng. Sci. 32, 1055-1066.
Luikov, A. V. (1964). Heat and mass transfer in capillary-porous bodies. A d v . Heat
Transfer 1, 123-184.
Luikov, A. V. (1975). System of differential equations of heat and mass transfer in
capillary-porous bodies (review). Int. J . Heat Mass Transfer 18, 11-4.
Lynkowski, R. W., and Chao, Y.-T. (1984). Comparison of Stefan model with two-phase
model of coal drying. Int. J . Heat Mass Transfer 27, 1157-1169.
Masuoka, T., Ishizaka, K., and Katsuhara, T. (1980). Heat transfer by natural convection
in porous media between two concentric spheres. I n “Natural Convection in Enclos-
ures,” ASME HTD-Vol. 8, pp. 115-120.
Menguc, M. P., and Viskanta, R. (1983). Comparison of radiative transfer approximations
for highly forward scattering planar medium. J . Qunntitative Spectroscopy and
Radiative Transfer 29, 381-394.
Merkin, J. H. (1979). Free convection boundary layers on axi-symmetric and two-
dimensional bodies of arbitrary shape in a situated porous media. Int. J. Heat Mass
Transfer 22, 1461-1462.
Minkowycz, W. J., and Cheng, P. (1976). Free convection about a vertical cylinder
embedded in a porous medium. Int. J . Heat Mass Transfer 19, 805-813.
Motakef, S., and El-Masri, M. A. (1985). Liquid diffusion in fibrous insulation. ASME J .
Heat Transfer 107, 299-306.
Moya, S. L., Ramos, E., and Sen, M. (1987). Numerical study of natural convection in a
tilted rectangular porous material. Int. J . Heat Mass Transfer 30,741-756.
Muralidhar, K., and Kulacki, F. A. (1986). NonDarcian natural convection in a saturated
horizontal porous annulus. In “Natural Convection in Porous Media,” ASME
HTD-Vol. 56, pp. 23-32.
Newman, A. B. (1931). The drying of porous solids: Diffusion calculations. AIChE Journal
27, 310-333.
Nilson, R. H., and Romero, L. A. (1980). Self-similar condensing flows in porous media.
Int. J . Heat Mass Transfer 23, 1461-1470.
Ogniewicz, Y., and Tien, C. L. (1981). Analysis of condensation in porous insulation. Int.
1. Heat Mass Transfer 24, 421-429.
Ozisik, M. N. (1973). “Radiative transfer and interactions with conduction and convec-
tion.” Wiley, New York.
Philip, J. R. (1982). Free convection at small Rayleigh number in porous cavities of
rectangular, elliptical, triangular and other cross-sections. Int. 1. Heat Mass Transfer
25, 1503-1510.
278 Chang-Lin Tien and Kambiz Vafai
Philip, J. R., and DeVries, D. A. (1957). Moisture movement in porous materials under
temperature gradients. Trans. Amer. Geophs. Union 38, 222-232.
Plumb, 0. A. (1983). The effect of thermal dispersion on heat transfer in packed bed
boundary layers. In “Proc. ASME-JSME Thermal Engineering Conference,” Vol. 2,
pp. 17-22. Honolulu.
Plumb, 0. A., and Huenefeld, J. A. (1981). NonDarcy natural convection from heated
surfaces in saturated porous media. Int. J. Heat Mass Transfer 24, 765-768.
Plumb, 0. A., Spolek, G. A., and Olmstead, B. A. (1985). Heat and mass transfer in wood
during drying. Int. J. Heat Mass Transfer 28, 1669-1678.
Pop, I., and Cheng, P. (1983). The growth of thermal layer in a porous medium adjacent to
a suddenly heated semi-infinite horizontal surface. Int. J. Heat Mass Transfer 26,
1574- 1576.
Poulikakos, D. (1985). A departure from the Darcy model in boundary layer natural
convection in a vertical porous layer with uniform heat flux from the side. ASME J.
Heat Transfer 107, 716-720.
Poulikakos, D. (1986). Buoyancy-driven convection in a horizontal fluid layer extending
over a porous substrate. Phys. Fluids 29, 3949-3957.
Poulikakos, D., and Bejan, A. (1983a). Natural convection in vertically and horizontally
layered porous media heated from the side. Int. J. Heat Mass Transfer 26, 1805-1814.
Poulikakos, D., and Bejan, A. (1983b). Numerical study of transient high Rayleigh number
convection in an attic-shaped porous layer. ASME J. Heat Transfer 105, 476484.
Poulikakos, D., and Bejan, A. (1985). The departure from Darcy flow in natural
convection in a vertical porous layer. Phys. Fluids 28, 3477-3484.
Poulikakos, D., and Renken, K. (1987). Analysis of forced convection in a duct filled with
porous media. I n “Heat Transfer in Geophysical and Geothermal Systems,” ASME
HTD-Vol. 76, pp. 9-20.
Prasad, V., and Kulacki, F. A. (1984a). Convective heat transfer in a rectangular porous
cavity: Effect of aspect ratio on flow structure and heat transfer. ASME J. Heat
Transfer 106, 158-165.
Prasad, V., and Kulacki, F. A. (1984b). Natural convection in a rectangular porous cavity
with constant heat flux on one vertical wall. ASME J. Heat Transfer 106, 152-157.
Prasad, V., and Kulacki, F. A. (1984~).Natural convection in a vertical porous annulus.
Int. J . Heat Mass Transfer 27, 207-219.
Prasad, V., and Kulacki, F. A. (1985). Natural Convection in horizontal porous layers with
localized heating from below. In “Heat Transfer in Porous Media and Particulate
Flows,” ASME HTD--Vol. 46, 199-208.
Prasad, V., and Kulacki, F. A. (1986). Effects of the size of heat source on natural
convection in horizontal porous layers heated from below. I n “Proceedings 8th
International Heat Transfer Conference,” Vol. 5, pp. 2677-2682. San Francisco,
California.
Prasad, V., and Kulacki, F. A. (1987). Natural convection in horizontal porous layers with
localized heating from below. ASME J. Heat Transfer 109, 795-798.
Prasad, V . , Kulacki, F. A., and Keyhani, M. (1985). Natural convection in porous media.
J. Fluid Mech. 150, 89-119.
Richards, L. A. (1931). Capillary conduction of liquids through porous media. J. Appl.
Phys. 1, 318-333.
Robinson, J. L., and O’Sullivan, M. J. (1976). A boundary layer model of flow in a porous
medium at high Rayleigh number. J. Fluid Mech. 75, 459-467.
Rohsenow, W.M., and Hartnett, J. P. (1973). “Handbook of Heat Transfer.” McGraw-
Hill, New York.
 Convective and Radiative Heat Transfer in Porous Media 279
Saxena, S. C., Grewal, N. S., Gabor, J. D., Zabrodsky, S. S., and Galershtein, D. M.
(1978). Heat transfer between a gas fluidized bed and immersed tubes. Adu. Heat
Transfer 14, 150-248.
Schneider, K. J. (1963). Investigation of the influence of free thermal convection in heat
transfer through granular material. Paper 11-4. I n “Proc. 11th Int. Congr. Refrig.,” pp.
247-254.
Schotte, W. (1960). Thermal conductivity of packed beds. AZChE Journal 6, 63-67.
Schrock, V. E., Fernandez, R. T., and Kesavan, K. (1970). Heat transfer from cylinders
embedded in a liquid filled porous medium. I n “Proc. 4th Int. Heat Transfer
Conference,” Vol. 7, p. CT-3.6. Paris-Versailles.
Seki, N., Fukusako, S., and Inaba, H. (1978). Heat transfer in a confined rectangular cavity
packed with porous media. Int. J. Heat Mass Transfer 21, 985-989.
Sen, A. K. (1987). Natural convection in a shallow porous cavity-The Brinkman model.
Int. J . Heat Mass Transfer 30,855-868.
Sherwood, T. K. (1929a). The drying of solids. I. Znd. Eng. Chem. 21, 12-16.
Sherwood, T. K. (1929b). The drying of solids. 11. ind. Eng. Chem. 21,976-980.
Sherwood, T. K. (1930). The drying of solids. 111. Znd. Eng. Chem. 22, 132-136.
Sherwood, T. K. (1931). Application of the theoretical diffusion equations to the drying of
solids. AIChE Journal 27, 190-202.
Shiralkar, G. S., Haajizadeh, M., and Tien, C. L. (1983). Numerical study of high Rayleigh
number convection in a vertical porous enclosure. Numerical Heat Transfer 6,223-234.
Siegel, R., and Howell, J. R. (1981). “Thermal Radiation Heat Transfer,” Second Edition.
Hemisphere, New York.
Simpkins, D. G., and Blythe, P. A. (1980). Convection in a porous layer. Int. J . Heat Mass
Transfer 23, 881-887.
Singh, B. S., and Dybbs, A. (1979). Nonisothermal liquid flow and heat transfer in sintered
metallic porous media. Znt. J. Heat Mass Transfer 22, 1049-1057.
Somerton, C. W. (1983). The Prandtl number effect in porous layer convection. Appl. Sci.
Res. 40,333-344.
Straws, J. M. (1974). Large amplitude convection in porous media. J. Fluid Mech. 64,
51-63.
Tabanfar, S . , and Modest, M. F. (1987). Combined radiation and convection in absorbing,
emitting, nongray gas-particulate tube flow. ASME J . Heat Transfer 109, 478-484.
Tien, C. L. (1975). Fluid mechanics of heat pipes. Ann. Rev. Fluid Mech. 7, 167-185.
Tien, C. L. (1985a). Heat Pipes. I n “Handbook of Heat Transfer Applications,” Second
Edition, Chapter 5 . Editors: Rohsenow, W. M., Hartnett, J. P., and Ganic, E. N.
Hemisphere, New York.
Tien, C. L. (1985b). Radiation properties of particulates. I n “Handbook of Heat Transfer
Fundamentals,” Second Edition, pp. 14.83-14.91. Hemisphere, New York.
Tien, C. L., and Cunnington, G. R. (1973). Cryogenic insulation heat transfer. Adu. Heat
Transfer 9 , 349-417.
Tien, C. L., and Drolen, B. L. (1987). Thermal radiation in particulate media with
dependent and independent scattering. Ann. Rev. Num. Fluid Mech. Heat Transfer 1,
1-32.
Tien, C. L., and Rohani, A. R. (1972). Theory of two component heat pipes. J . Heat
Transfer 94, 479-484.
Tong, T. W., and Orangi, S. (1986). A numerical analysis for high modified Rayleigh
number natural convection in enclosures containing a porous medium. In Proceedings
8th International Heat Transfer Conference,” Vol. 5 , pp. 2647-2652. San Francisco,
California.
280 Chang-Lin Tien and Kambiz Vafai
Tong, T. W., and Subramanian, E. (1983). Natural convection in rectangular enclosures
partially filled with a porous medium. I n “Proceedings of the ASME-JSME Thermal
Engineering Joint Conference,” Vol. 1, pp. 331-338. Honolulu, Hawaii.
Tong, T. W., and Subramanian, E. (1985). A boundary-layer analysis for natural
convection in vertical porous enclosures-Use of the Brinkman-extended Darcy model.
Int. J . Heat Mass Transfer 28, 563-571.
Tong, F. W., Faruque, M. A., Orangi, S., and Sathe, S. B. (1986). Experimental results for
natural convection in vertical enclosures partially filled with a porous medium. I n
“Natural Convection in Porous Media,” ASME HTD-Vol. 56, pp. 85-94.
Tsai, F. P., Jakobsson, J., Catton, I., and Dhir, V. K. (1984). Dryout of an inductively
heated bed of steel particles with subcooled flow from beneath the bed. Nuclear
Technology 65, 10-15.
Udell, K. S. (1985). Heat transfer in porous media considering phase change and
capillarity-The heat pipe effect. Int. J . Heat Mass Transfer 28, 485-495.
Udell, K. S., and Fitch, J. S. (1985). Heat and mass transfer in capillary porous media
considering evaporation, condensation and noncondensible gas effects. I n “Heat
Transfer in Porous Media and Particulate Flows,” pp. 103-110.
Vafai, K. (1984). Convective flow and heat transfer in variable-porosity media. J . Fluid
Mech. 147, 233-259.
Vafai, K. (1986). Analysis of the channeling effect in variable porosity media. ASME J .
Energy Resources Technology 108, 131-139.
Vafai, K., and Ettefagh, J. (1988). Analysis of the radiative and conductive heat transfer
characteristics of a waste package canister. ASME J . Heat Transfer 110, 1011-1014.
Vafai, K., and Sarkar, S. (1986). Condensation effects in a fibrous insulation slab. ASME J .
Heat Transfer 108, 667-675.
Vafai, K., and Sarkar, S. (1987). Heat and mass transfer in partial enclosures. AIAA J.
Thermophysics Heat Transfer 1.3, 253-259.
Vafai, K., and Thiyagaraja, R. (1987). Analysis of flow and heat transfer at the interface
region of a porous medium. Int. J . Heat Mass Transfer 30, 1391-1405.
Vafai, K., and Tien, C. L. (1981). Boundary and inertia effects on flow and heat transfer in
porous media. Int. J. Heat Mass Transfer 24, 195-203.
Vafai, K., and Tien, C. L. (1982). Boundary and inertia effects on convective mass transfer
in porous media. Int. J. Heat Mass Transfer 25, 1183-1190.
Vafai, K., and Whitaker, S. (1986). Simultaneous heat and mass transfer accompanied by
phase change in porous insulation. ASME J . Heat Transfer 108, 132-140.
Vafai, K., Alkire, R. L., and Tien, C. L. (1985). An experimental investigation of heat
transfer in variable porosity media. ASME J . Heat Transfer 107, 642-647.
Vasseur, P., and Robillard, L. (1987). The Brinkman model for boundary layer regime in a
rectangular cavity with uniform heat flux from the side. Int. J . Heat Mass Transfer 30,
7 17-727.
Vasseur, P., Satish, M. G., and Robillard, L. (1987). Natural convection in a thin, inclined,
porous layer exposed to a constant heat flux. Int. J . Heat Mass Transfer 30, 537-549.
Viskanta, R. (1982). Radiation heat transfer: Interaction with conduction and convection
and approximate methods in radiation. I n “Proceedings of the Seventh International
Heat Transfer Conference,” Vol. 1, pp. 103-121. Munchen, Germany.
Viskanta, R. (1984). Radiative heat transfer. Progress Chemical Engineering (Fortschritte
der Verfahrenstechniks) Section A 22, 51-81.
Vortmeyer, D. (1974). Radiation interaction with conduction and convection. I n
“Proceedings of the 5th International Heat Transfer Conference,” Vol. 6, pp. 129-135.
Tokyo, Japan.
 Convective and Radiative Heat Transfer in Porous Media 281
Vortmeyer, D. (1978). Radiation in packed solids. In “Proceedings of the 6th International
Heat Transfer Conference,” Vol. 6, pp. 525-539. Toronto, Canada.
Vortmeyer, D., and Schuster, J. (1983). Evaluation of steady flow profiles in rectangular
and circular packed beds by a variational method. Chem. Eng. Sci. 38, 1691-1699.
Walker, K. L., and Homsy, G. M. (1978). Convection in a porous cavity. J. Fluid Mech.
87, 449-474.
Wang, K. Y., and Tien, C. L. (1983). Radiative heat transfer through opacified fibers and
powders. J. Quantitative Spectroscopy Radiative Transfer 30,213-223.
Weber, J. E. (1975a). The boundary layer regime for convection in a vertical porous layer.
Int. J. Heat Mass Transfer 18, 569-573.
Weber, J. E. (1975b). Thermal convection in a tilted porous layer. Int. J. Heat Mass
Transfer 18, 474-475.
Westman, A. E. R. (1929). The capillary suction of some ceramic materials. 1. Amer.
Ceram. SOC.l2,585-595.
Whitaker, S. (1977). Simultaneous heat, mass, and momentum transfer in porous media: A
theory of drying. A d v . Heal Transfer 13, 119-203.
White, S. M., and Tien, C. L. (1987a). Analysis of laminar condensation in a porous
medium. In “Proc. of the ASME-JSME Thermal Engineering Joint Conference,” Vol.
2, pp. 401-406. Honolulu, Hawaii.
White, S. M., and Tien, C. L. (1987b). Experimental investigation of film condensation in
porous structures. In “Proc. of the 6th International Heat Pipe Conference, Fun-
damental and Basic Research: Theoretical and Experimental Studies,” pp. 148-153.
Wong, K. F. and Dybbs, A. (1978). Energy equations for convection in saturated porous
media. ASME Paper No. 78-HT-54.
Wooding, R. A. (1963). Convection in a saturated porous medium at large Rayleigh
number or Peclet number. 1. Fluid Mech. 15,527-544.
Yamada, Y.,Cartigny, J. D., and Tien, C. L. (1986). Radiative transfer with dependent
scattering by particles: Part 2-experimental investigation. ASME J. Heat Transfer 108,
614-618.
Yang, Y. S., Howell, J. R., and Klein, D. E. (1983). Radiative heat transfer through a
randomly packed bed of spheres by the Monte Carlo method. ASME J. Heat Transfer
105, 325-332.
Yen, Y. C. (1974). Effects of density inversion on free convective heat transfer in porous
layer heated from below. Int. J. Heat Mass Transfer 17, 1349-1356.
Yener, Y.,and Ozisik, M. N. (1986). Simultaneous radiation and forced convection in
thermally developing turbulent flow through a parallel-plate channel. ASME J. Heat
Transfer 108, 985-988.
Yuen, W. W., and Wong, L. W. (1980). Heat transfer by conduction and radiation in a
one-dimensional absorbing, emitting, and anisotropically scattering medium. ASME J .
Heat Transfer 102, 303-307.
This Page Intentionally Left Blank
 ADVANCES IN APPLIED MECHANICS, VOLUME 27

Solitons Induced by Boundary Conditions

C. K. CHU
Department of Applied Physics
Columbia University
New York, New York

RU LING CHOU

Lamont Doherty Geological Observatory

Columbia University
New York, New York
and
NASA Goddard Institute for Space Studies
New York, New York

I. Introduction ............
11. Experiments . . . . . . . . . . . .
111. Numerical Results.. ............................
A. Soliton Formation. ...
B. Nonlinear Periodic Waves. ............. 295
IV. The0ry .................
A. Well-Posed Problem. .
B. Inverse Scattering.. ..
V. Conclusions ............
References ........................................................... 302

I. Introduction

Since the sighting of the solitary wave by Russell (1844) over a century
ago, the discovery of the remarkable properties of solitons by Zabusky
and Kruskal (1965) over two decades ago, and the invention of the
inverse scattering transform by Gardner, Greene, Kruskal, and Miura
283
Copyright 0 1wO Academic Press, Inc.
All rights of reproduction in any form reserved.
ISBN 0-12M12M7-0
284 C. K. Chu and Ru Ling Chou

(1967) shortly afterwards, the study of solitons has mushroomed into an

active field of research. Considerable interest has also been shown in the
generation of solitons.
In water, solitary waves can be generated by three different methods:
an initial profile evolving into one or many solitary waves, a moving ship
or equivalent pressure source on the surface of water, or boundary
excitation, such as a sluice opening or a wall pushing. In other media,
solitons can be produced by analogous situations.
In theoretical work, the first method corresponds to a pure initial value
problem (for some governing differential equation or system of equa-
tions), the second to a driving term or an inhomogeneous differential
equation (or equations), and the third to a mixed initial boundary value
problem. Indeed, the most exciting advances in soliton theory, such as
soliton interaction and collision and the inverse scattering transform,
have all come from pure initial value problems. Only recently have steps
been taken in the theoretical and numerical studies of the mixed initial
boundary value problems.
The differential equations used in the theoretical and numerical works
vary. The simplest and best studied is the Korteweg-deVries equation
(KdV equation), either in the ordinary form or in the regularized form
(the latter replaces the third x-derivative in the former by a mixed
xxt-derivative). For water, the Boussinesq systems (also in different
forms), and for nonlinear optics, the nonlinear Schrodinger equations
have been used extensively.
In this paper , we summarize briefly experiments and then describe
numerical results, mainly for water waves. We then describe some
theoretical results obtained to date for mixed initial boundary value
problems for the KdV equation and for the nonlinear Schrodinger
equation.
In the preparation of this review, the authors have benefited enor-
mously from valuable and insightful discussions with many colleagues,
particularly the following: Drs. M. Ablowitz, Y. Baransky, T. Fokas, M.
Tabor, and T. Y. Wu. We thank them deeply for their help.

II. Experiments

There have been numerous experiments done on solitons in water, as

well as in other media, with the express purpose of seeing the nonlinear
 Solitons Induced by Boundary Conditions 285

interaction that preserves wave identities, as predicted by numerical

experiments and by theory. The main results of these experiments need
not concern us in this paper, but the methods of generation of the
solitons are of significant interest. Indeed, in most experiments, the
solitons are boundary generated.
Maxworthy (1976) used a uniform bottom trough to study soliton
reflection and collision in water. In studying reflection, the method he
used to generate the solitons was to push a movable partition along the
length of the trough. The waves thus generated then reflect off a fixed
end wall. In studying collisions of solitons of different strengths, he used
partitions near both ends of the trough, and filled water behind the
partitions to different heights. The solitons are generated by withdrawing
the partitions, and solitons of different strengths then run toward each
other. Clearly, all the excitation processes involved here-pushing a wall
(or piston), reflecting waves from a fixed wall, and producing waves from
water at different heights-all correspond to boundary generation. The
last case can be argued as generation from initial conditions, but if either
the water level is maintained constant behind the partition after
withdrawal, or if the end wall is close to the partition, then the excitation
is, strictly speaking, still by boundary action.
Hammack and Segur (1974) also studied solitons in a trough by raising
or lowering a piston in the vertical direction, i.e., perpendicular to the
trough. This could approximate generation by initial conditions. Again,
however, the piston is small in comparison with the trough length, and it
is placed right next to a fixed wall. Thus, strictly speaking, the solitons
are affected by boundary action as well.
The ingenious laboratory experiments of J. Scott Russell (1844), which
followed his famous sighting of solitary waves in the Union Canal in
1834, were similar to the Hammack and Segur experiments, except that
instead of raising or lowering a piston, he dropped one end of a solid
weight into the water to produce positive waves (bumps), and lifted the
weight out of the water to produce negative waves (depressions). This
too is boundary generation of solitary waves.
Bona et al. (1981) studied nonlinear periodic waves in a trough. The
waves are generated by a paddle wheel at one end of the trough. The
generation is again obviously by boundary action.
These examples suffice to indicate the practical importance of studying
boundary generation of solitons, despite the theoretical convenience of
considering generation by initial conditions. The latter is rarely achiev-
286 C. K. Chu and R u Ling Chou

able experimentally, and the same situation holds in other media (e.g.,
solitons in ion-acoustic plasmas, nonlinear optics, etc.).

III. Numerical Results

A. SOLITON
FORMATION

The work of Chu, Xiang, and Baransky (1983) was one of the earliest
papers explicitly aimed at studying soliton formation in the quarterplane
solutions of the KdV equation, in its ordinary form
u,+ uu, + EU,, = 0. (3.1)
The initial condition and boundary conditions are the simplest possible:
u(x,0)= 0,
~ ( 0t), = U H ( t ) or U [ H ( t )- H ( t - to)],
where U is a constant and H ( t ) is the Heavyside function. The results
they found were striking and are summarized concisely below:
1. For a constant boundary excitation, u(0, t) = U H ( t ) , a continuous
stream of identical solitons are produced. As measured from the
computed results, the amplitude of each soliton is 2U, and the speed is
2U/3 in agreement with classical theory (Figure 1). Each soliton
“matures” in a certain formation time before the next soliton is “born”.
2. For a pulse excitation of length t o , u(0, t) = U [ H ( t )- H(t - to)], the
first solitons reach maturity, but the last one in general does not. Its
amplitude reaches a value determined by the cut-off time to (Figure 2).
3. Subsequntly, Camassa and Wu (1988) confirmed these findings
using more accurate numerical calculations. In addition, Chou (Chou,
1987 and Chou and Chu, 1988) showed in her doctoral thesis that solitons
calculated with the Boussinesq system
dh d
- + -(hu)= O
at ax
(3.3)
du au ah d3h
-+
dt
u-+-++p7=0
ax dx dX
Here h and u are the dimensionless height and velocity respectively, and
 Solitons Induced by Boundary Conditions 287

U 1
4.0

3.0

2.0

1 .o

-1.0
I
100 200 300 400 500 600
x(No. of Grid Points)
FIG.1. Solitons generated from KdV equation by the constant boundary condition
u(0, r ) = 2 and the zero initial condition u(x, 0) = 0 (Chu er al., 1983. Copyright 01983 by
John Wiley 13Sons, Inc. Reprinted by permission of John Wiley & Sons, Inc.).

U
4

0
0 10 20 30 40 50 60
x(No. of G r i d Points 1
FIG.2. Solitons produced from two pulses u(0, r ) of different durations ( 0 . U a s h e d
line and 0.6-solid line) showing maturing (Chu er al., 1983. Copyright 0 1983 by John
Wiley & Sons, Inc. Reprinted by permission of John Wiley & Sons, Inc.).
 C. K. Chu and Ru Ling Chou

gl& 0.0 d, 0.3

TIME
0.6 0.9

FIG.3. Trajectories in x-f space of soliton peaks, showing acceleration during maturing,
for a finite duration pulse (Camassa and Wu, 1988).

/3 is related to the E in the Korteweg-deVries equation: E = bp, for these

initial and boundary conditions are virtually identical to these solitons,
thus confirming that the two nonlinear equations indeed approximate
each other very well.
4. Solitons accelerate as they grow, until they reach the speed
appropriate for the mature solitons. Figure 3, taken from Camassa and
Wu (1988), shows the x-t curves of the soliton crests as they form,
accelerate, and reach steady speed.
5. Chou (Chou, 1987 and Chou and Chu, 1988) also considered more
general boundary conditions than square pulses. One aim of her work
was to see whether soliton “turbulence” would result if randomly
fluctuating boundary conditions were applied. This did not occur. In fact,
the random boundary data produced perfectly uniform solitons, with
amplitude equal to twice the expected value of the random values (Figure
 Solitons Induced by Boundary Conditions 289

1.0 -
.9 -
.B -
.7 -,.
.6

.5

.4

.3 -
.2 -
-
.1
I 1 d
0 -

FIG.4. Solitons generated from random boundary values of very short duration-1 time
step (Chou, 1987 and Chou and Chu, 1988).

4). However, more slowly varying boundary values did produce modu-
lated amplitudes in the solitons (Figure 5 ) . From this, and from slowly
varying sinusoidal boundary data, Chou concluded that the time variation
in the boundary data has a strong influence on the modulated soliton
amplitudes only if the characteristic time of the variation is comparable to
or greater than the soliton formation time; if it were much smaller, than
the solitons formed sense only averaged boundary values.
6. In Chu et al. (1983), they also considered the KdV-Burgers
equation, i.e.
u, + UU, - VU,, + EU, = 0. (3.4)
This equation has both dissipation and dispersion, but the interesting case
290 C. K. Chu and Ru Ling Chou

1.1

1.0

.9

.B

.I

.6

.5 -
.4 -
.3 -
.2 -
I
-
.l
J I
0 -

-.1'~~~~'~''~~''''~''''''''~~''''~''''~''''~~'''~'~''~'~r'~''~ I '

0 .5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0

FIG. 5. Solitons generated from random boundary values of longer duration-10 time
steps (Chou, 1987 and Chou and Chu, 1988).

is when the effect of dissipation is small. For the same initial boundary
value problem as described above the pure KdV equation (i.e., zero
initial data and a pure Heavyside function for the boundary data), the
resulting wave pattern now appears as in Figures 6a and 6b. The pattern
given in Figure 6b is a weak bore or collisionless shock, and is
significantly different from the solitons in Figure 1. While the former is a
continuously lengthening train of waves, the latter is a steady structure
that connects two uniform states, in this case, u = 0 and u = U . This
structure can be readily calculated from a set of ordinary differential
equations when we perform a Galilean transformation at the speed of the
wave in the usual way for obtaining shock wave structure. The length of
the train depends on the dissipation Y , and the speed is U / 2 , (as derived
 Solitons Induced by Boundary Conditions 29 1

U
(a) 3.0

2 .o

1.0

0 100 200 300 400 500 600

x (No. of G r i d Points)
U
(b) 3.0

2.0

1.0

0 100 200 300 400 500 600

x(No. of Grid Points)
FIG. 6. “Collisionless” shock from KdV-Burgers equation ( E = 0.022*, v = 2 ~ ) .(a)
Partially formed r = 2.4 (189 steps). (b) Fully formed t = 4.4 (34.9 steps). (Chu et al., 1983.
Copyright 0 1983 by John Wiley & Sons, Inc. Reprinted by permission of John Wiley &
Sons, Inc.).
292 C. K. Chu and Ru Ling Chou
from the Hugonot relation and as measured from the computed results),
in contrast to the speed of the leading soliton in Figure 1, which is 2U/3.
7. In Chou (Chou, 1987 and Chou and Chu, 1988), she extended the
class of initial conditions to u(x, 0) = uo, a constant, from the simple case
u(x, 0) = 0 of Figure 1. Rather surprisingly, waves of the form of Figure 7
now appear. The nonzero initial state uo> 0 has the effect of increasing
the soliton maturing time, so that now there are many solitons awaiting
maturation, and only the few near the front have attained full maturity.
In the case of Figure 1, we recall, each soliton matures almost fully
before the next one is born. This wave pattern of solitons growing is not
to be confused with the collisionless shocks of the previous paragraph, as
the shapes are completely different, and the former is an elongating wave
train while the latter is a steady structure.

U
2.6

2.4

2.2

2.0

1.8

1.6

1.4

1.2

1.0

.e
0 1 2 3 4 5 6 7 8 9 10 11 12

FIG.7. KdV solitons with u(x, 0) = uo= 1, u ( 0 , r ) = 1.75, showing much slower matur-
ing process (Chou, 1987 and Chou and Chu, 1988).
 Solitons Induced by Boundary Conditions 293

This situation can be seen more clearly if we make a transformation.

Thus. let
v =u - ug,
y = x - uot.
The original problem with u ( x , 0) = uo and u(0, t ) = U H ( t ) then trans-
forms into
v, + vvy + &Vyyy = 0
with
v(x, 0) = 0, v = U - uo on y = -uot.
Thus, the problem is equivalent to the original problem with u = U given
not on a fixed wall x = 0, but on a receding wall with velocity uo. The
later solitons will take a longer time to catch up with the earlier ones, and
thus the wave train will have the shape of Figure 7.
8. The equations used throughout these studies are the KdV equation
in its ordinary form (3.1), or the Boussinesq system (3.3). The regu-
larized form, where u, is replaced by u,~, was not used. The motivation
for not regularizing is that the authors wanted to see the explicit behavior
of the KdV equation in the quarterplane, and hoped to benefit from and
extend the vast theoretical literature on inverse scattering. Regularization
is necessary if one is particularly interested in the behavior of short
waves, which is not the aim of these authors. The numerical schemes
used were the same as given by Zabusky and Kruskal, i.e., leap frog for
the convective term and 5-point centered difference for the third
derivative terms; in Chu (1983), the leap-frog scheme used is the
standard second-order, while in Chou (1987,1988), both second-order
and fourth-order schemes were used. A matter of concern is that when
the wave lengths are short, “instability” would occiir. Figure 8, taken
from Chou et al. (1988), however, insures that this does not occur.
Shown in Figure 8 are the dispersion relations of the KdV equation, the
regularized KdV equation, and the numerical schemes using second- and
fourth-order leap-frog schemes for the convective operator. These
equations are first linearized with respect to a constant u = a # 0. We
remark that a is not the velocity at infinity, which would be zero, but the
local fluid velocity u ; the dispersion relation, as always, is local. The KdV
equation and the regularized KdV equation differ in the dispersive terms:
EU,,, and -(&/a)uzxtrespectively, and the dispersion relations are found
294 C. K. Chu and Ru Ling Chou

0 = phase
of r (degs)

0. 30 60. SO. 120. 150. 180.

E = k Ax (in degs)
FIG. 8. Dispersion relation for (linearized) KdV equation (Curve l), regularized KdV
equation (Curve 2), and leap-frog schemes (second orderxurve 3, fourth order-Curve
4). Abscissa is wave number k multiplied by grid size Ax. Ordinate is phase + of
amplification factor ( = o k At, the frequency multiplied by k and the time step). (Chou et
al., 1988.)

in the standard manner. For the numerical schemes, the dispersion

relations are found from the phase of the amplification factor, a complex
number, obtained from the usual von Neumann analysis. In contrast to
the exact KdV equation, which has negative o for k large, these
numerical approximations to the KdV equation have o always nonnega-
tive, thus having a regularizing effect on the equation. In any case, the
equation is not intended to be applied at small wave lengths, and this
regularizing effect protects against any short wave length instabilities.
 Solitons Induced by Boundary Conditions 295

B. NONLINEAR
PERIODIC
WAVES

Unknown to the authors of Chu et al. (1983), the KdV equation in the
quarterplane had been considered in great detail by Bona and coauthors.
A theory paper, Bona and Winther (1983), on the proper posing of the
problem will be discussed later, and an earlier paper by Bona, Pritchard,
and Scott (1981) solves the KdV equation numerically in the quarter-
plane. The authors were interested mainly in studying nonlinear periodic
waves and in obtaining the accurate dissipation and dispersion coefficients
to match experimental results, and not in soliton generation. Neverthe-
less, the problems are closely related, and their numerical scheme is
much more sophisticated, so that describing their work is in order.
The authors used the regularized KdV equation with dissipation
included, thus
rlt + (1 + ! r l ) r l X - P V X X - 8 L = 0. (3-5)
The regularizing, as mentioned earlier, kills off short-wave instabilities.
In contrast to the previously described works, here this regularization is
desirable, as the more sophisticated numerical scheme may preserve the
nature of the dispersion relation to large values of k. The initial value of
q ( x , 0) is taken as 0, while the boundary value ~ ( 0t), is taken to be a
given function h ( t ) .
The problems studied by Bona et al. (1981) are for initial conditions
and boundary conditions: ~ ( x 0), = 0, ~ ( 0t,) = small-amplitude sine
function of t or more general periodic function of t. If we transform
Equation (3.5) into (3.4), or into (3.1) with p set equal to zero, we have
E = 8, while the initial condition corresponds to u ( x , 0) = 1, and the
boundary condition corresponds to a periodic oscillation about u = 1.
The numerical scheme is based on an integral equation equivalent to
( 3 4 , which is also used for proving existence and uniqueness of the
solution:

rlt(x, t ) = h ’ ( W f i X +
lW, Y)(V+ irl*)(y, t ) dt
m

+ 6ptrl(t)e-GX - rl(x, t ) -~ ~ x y)v(y,

, 1) dt, (3.6)
0

where
~ ( xy)
, = 3[edfi(~+y) + sgn(x - y)e-filX-Yl I
296

0.04r
-
c
a
- 0
F

-0.04i ' ' I ' 0I

I0 I I ' I
'
' I
'
200 300
I I I I 400 I I 500'
I I 600
II

t/At
FIG.9. (a) Waves computed from KdV-Burgers equation at different instants. (b) Time
histories of computed waves compared with experiments at three different locations. (Bona
er al. 1981.)
 Solitons Induced by Boundary Conditions 297
0.04

h o

-0.04LI ' 1 1 1 1 I 1 1 ' I 1 I I I I I I 1 I I I I I I I I L I I I I 1

0 I00 200 300 400 5 00 6 00
t/At
FIG. 10. Typical boundary data used to generate the waves of Figure 9 (Bona ef ai.,
1981).

and

The numerical scheme uses a quadrature in x for the integrals on the

right hand side, and uses a fourth-order prediction-correction scheme in t
for the time integration. The method is considerably more accurate than
the scheme used in Chu et al. (1983). To test convergence, it was used on
a single soliton generated by boundary excitation, which compared
favorably with the exact solution.
The periodic waves were compared with experiment to determine the
correct parameter of p for water waves. Typical results are given in
Figure 9, where Figure 9a shows the shape of the excited waves at a fixed
time (for p = 0.014), and Figure 9b shows the time history of the wave
height at a fixed location (for p = 0.00168, and other coefficients all
altered), compared with experimental observations. A typical boundary
condition used for these studies is shown in Figure 10.
Based on these cited results, we can summarize boundary generated
waves to the KdV equation (3.1) and the KdV-Burgers equation (3.4)
(plus variants) in Table 1. Essentially, all the wave patterns that could be
generated by these two equations have been covered, and a reasonably
consistent picture, though still qualitative, has emerged from these works
seen together.

IV. Theory
There is a small amount of theoretical work for the KdV equation, and
somewhat more for the nonlinear Schrodinger equation, in the quar-
terplane. No such work has appeared in the literature on the Boussinesq
equation.
298 C.K. Chu and Ru Ling Chou

TABLE 1
WAVE PATERNS OF KdV AND KdV-BURGERS EQUATION

Initial state Boundary condition Wave patterns

No dissipation (v = 0):
u=o u = constant Continuous stream of
identical solitons, Figure 1
u = random Similar to above, Figure 4
u = periodic Stream of solitons,
and >O modulated, Chou (1987)
u=u,>o u = constant Continuous stream of
solitons maturing slowly,
Figure 7
u = periodic Figure 9a and Bona et al. (1981)*
and >O

Dissipation present (v > 0):

u=o u = constant Collisionless shock, Figure 4
u=u,>o u = periodic Figure 9a and Bona et al. (1981)'
and >O

* Recall these results are for the perturbed level q. The total height u = 1 + q is always
positive.

The theoretical results that have appeared in the literature so far can
be divided into two main groups. The first is the study of the existence
and uniqueness of the solution to the KdV equation in the quarterplane,
i.e., the well-posed nature of the problem. The second consists of the
attempts at extending the inverse scattering transform method from pure
initial value problems to those in the quarterplane.

A. WELL-POSED
PROBLEM

The problem of existence and uniqueness has been addressed in detail

by Bona and his coworkers (Bona and Winther, 1983). They showed that
prescribing initial conditions u(x, 0) and a single boundary condition
u(0,t) with suitable smoothness and compatibility conditions results in a
unique solution of the KdV equation in the quarterplane, with appropri-
ate smoothness. With minimal assumptions on smoothness, the existence
of a unique distribution solution is proved; with more stringent smooth-
ness assumptions, the unique existence of a classical solution is proved.
The method of proof uses the integral equation described earlier, from
 Solitons Induced by Boundary Conditions 299

which a priori estimates were obtained. The smoothness requirements

and the detailed proofs are technical, and we refer the interested readers
to their original papers.

B. INVERSE
SCATTERING

Extension of the inverse scattering method to the quarterplane is a

direction of research still in progress, and it is being studied by many
authors. There are two different paths of attack in this case. The first,
started by Kaup and Hansen (1986) for the nonlinear Schrodinger
equation and adopted by Camassa and Wu (1988) for the KdV equation,
is intended to model the numerical results, and necessitates some ad hoc
assumptions. The other, pursued by Fokas, Ablowitz, and colleagues
(Fokas, 1987 and Fokas and Ablowitz, 1988) uses Fourier sine transforms
to extend the initial data on the semi-infinite line to the entire line. The
procedure is thus self-consistent and rigorous, but up to the present,
results have only been obtained for the nonlinear Schrodinger equation,
and it is by no means clear how to extend the method for the KdV
equation.
We first summarize the inverse scattering transform method for pure
initial value problems. Consider the eigenvalue problem for the steady
state Schrodinger equation

qxx- u q = -Aq. (4.1)

The scattering problem is, given u ( x ) , to determine the discrete
eigenvalues A and the eigenfunctions q, assuming square integral
eigenfunctions entering from say, +w. With this assumption, the eigen-
values and eigenfunctions can be represented as

K = K1, Kz, * . . , Kn, q = "1, VZ, * * * 9 qnr

A= -K2,
m

f-m
q ' ( x ) dx = 1 and C(K) = lim
x-m
e"q(x),

For the continuous spectrum, A > 0,

300 C. K. Chu and Ru Ling Chou

For the continuous spectrum, U ( K ) and b ( ~ are ) the transmission and

reflection coefficients respectively; for the discrete eigenvalues I. = -K ~ ,
C ( K ) are the normalizing coefficients. The inverse scattering problem is to
give these coefficients and determine u ( x ) , using the well-known
Gelfand-Levitan or Marchenko integral equation. These facts and
procedures are standard in quantum mechanics.
The remarkable part of the inverse scattering transform method is the
theorem (Gardner et al., 1967) that if u is regarded as a function of x and
a parameter t in equation (4.1), and u satisfies the KdV equation (3.1)
with uu, replaced by -6uu,, with x and t now regarded as independent
variables, then the A’s are independent of t, the a’s are independent of t,
while the b’s and c’s evolve in t in a simple manner. This exciting result
permits the exact solution of the initial value problem of the KdV
equation using the following procedure: From the initial data u(x, 0), we
calculate by direct scattering the various values of A, a, b, and c at t = 0.
Then these values are known at all t. Hence by inverse scattering (now
considering t as a parameter), we get u(x, t ) for all t.
In Kaup and Hansen (1986) and in Camassa and Wu (1988), the
different authors attempted to extend this procedure to the quarterplane
problem. We reinterpret their procedures in a simple manner. Two
linearly independent eigenfunctions of the steady-state Schrodinger
equation (4.1) can be given as q ( x ) and &), which behave as
-
q ( x ) eiKx and $(x) - e-iKx as x-w.
Two other linearly independent eigenfunctions can be given as @ ( x ) and
$ ( x ) , which behave as
@(x) - eiKx and $(x) - e-jKX as x + - > .
From the definition of the scattering coefficients a and b, we have
a @ ( x )= v ( x ) - b$(x). (4.2)
Moreover, q satisfies the integral equation

In addition, $ satisfies the same equation with e-’= for the first term,
while @ and 4 satisfy similar equations except with the integral extended
from --oo to x .
The authors assume that the potential u(x; t ) , which satisfies the KdV
 Solitons Induced by Boundary Conditions 301

equation, to be u ( x ; t ) = O for all x < O and all t. Then, the integral

4
equations satisfied by 4 and all have integrals extended from 0 to x .
From (4.2) and (4.3), one easily deduces that q ( 0 , t) is a linear
combination of the scattering coefficients a and b. The t-evolution of a
and b thus depends on the evolution of q(0, t).
The time evolution of T) cannot be described by a simple equation for
3 alone, and is best seen in the form given by Zakharaov and Shabat, cf.
Ablowitz et al. (1974). Let v1 = 111, and let v 2 = i K q - q x . Then the
eigenvalue equation (4.1) can be written for the column vector u =
(q,v2) as a 2 x 2 system, and the time evolution equations can now be
written as a 2 x 2 system. A compact integral equation now replaces the
many equations corresponding to (4.3). From the time evolution equa-
tions, the evolution for q(0, t) can be determined, and hence the
evolution of the scattering coefficients are also found. Nevertheless, the
evolution equation for q ( 0 , t) requires the derivatives u,(O, t) and
u,,(O, t ) for the KdV equation, and the derivative u,(O, t) for the
nonlinear Schrodinger equation. These are not prescribable data for the
initial boundary value problems in question. The authors thus make the
ad hoc assumptions of setting them to zero, and obtained results that are
somewhat close to the numerical results. Hence, they describe the
procedure as a model and not yet a theory as such
Fokas (1987) used the Fourier sine transform (in x ) as a starting point,
and he did succeed in obtaining a theory for the nonlinear Schrodinger
equation. The sine transform, however, works for even derivatives in x ,
but not for odd derivatives. Hence the method cannot be extended
directly to the KdV equatior., and future work is still needed.

V. Conclusion

We have summarized scattered work on boundary generation of

solitons. Various authors doing numerical experiments have expressed
the hope that theory would develop, just as the inverse scattering
transform method blossomed following the early numerical experiments
of Zabusky and Kruskal on initial value problems. It is gratifying to see
that some beginnings in theoretical work have taken place, but con-
siderable work is needed before mixed initial boundary problems reach
the same degree of understanding as pure initial value problems with the
inverse scattering transform.
302 C. K. Chu and Ru Ling Chou

The works described here merely scratch the skin of a vast field of
interesting problems. First, theory should also develop for the forced
KdV equation, relevant for ship generated solitons, for example. For
linear equations, an inhomogeneous forcing function can always be
replaced by a homogeneous equation and inhomogeneous boundary data,
but for nonlinear equations, this is quite different. Second, there appears
to have been no work yet done on boundary generated solitons in more
than one dimension, even numerically. This seems to be a very fruitful
and exciting direction waiting to be pursued. The physical problems that
can be modelled are numerous, and undoubtedly, many new physical
phenomena will be discovered from such numerical experiments.

References

Ablowitz, M. J., Kaup, D. J., Newell, A. C., and Segur, H. (1974). The inverse scattering
transform-Fourier analysis for nonlinear problems. Srud. Appl. Math. 53, 249-315.
Bona, J. L., Pritchard, W. G., and Scott, L. R. (1981). An evaluation of a model equation
for water waves. Philos. Trans. Roy. SOC.London Ser. A 302, 457-510.
Bona, J. L., and Winther, R. (1983). The Korteweg-deVries equation posed in a
quarterplane. SIAM J. Math. Anal. 14, 1056-1106.
Camassa, R., and Wu, T. Y. (1988). The KdV model with boundary forcing. To appear.
Chou, R. L. (1987). Solitons induced by boundary conditions. Ph.D. dissertation, City
University of New York.
Chou, R. L., and Chu, C. K. (1988). Solitons induced by boundary conditions from
Boussinesq equation. Phys. Fluids., submitted.
Chu, C. K., Xang, L. W., and Baransky, Y. (1983). Solitary waves induced by boundary
action. Comm. Pure Appl. Marh. 36, 495-504.
Fokes, T. (1987). An initial-boundary value problem for the nonlinear Schrodinger
equation. Clarkson Univ. Rep. INS *81. (Accepted Phys. D.)
Fokas, T., and Ablowitz, M. J. (1988). Forced nonlinear evolution equations and the
inverse scattering transform. Clarkson Univ. Rep. INS ‘99.
Gardner, C. S., Greene, J. M., Kruskal, M. D., and Miura, R. M. (1967). A method for
solving the Korteweg-deVries equation. Phys. Rev. Lerr. 19, 1095-1097.
Hammack, J. L. and Segur, H. (1974). The Korteweg-deVries equation and water waves:
Part 2, Comparison with experiments. J . Fluid Mech. 65,289-314.
Kaup, D. J., and Hansen, P. J. (1986). The forced nonlinear Schrodinger equation. Phys.
D 18,77-84.
Maxworthy, T. (1976). Experiments on collisions between solitary waves. J . Fluid Mech.
76, 177-185.
Russell, J. S . (1844). Report on waves. “In Reports from the 14th Meeting of the British
Association for the Advancement of Science”, pp. 331-390. John Murray, London.
Zabusky, N. J., and Kruskal, M. D. (1965). Interaction of solitons in a collisionless plasma
and the recurrence of initial states. Phys. Rev. Leu. 15, 240-243.
 ADVANCES IN APPLIED MECHANICS, VOLUME 27

Physical Limnology
JoRG IMBERGER
and
JOHN C. PATTERSON
Centre for Water Research
The University of Western Australia
Nedlands, Western Australia

I. Introduction .......................................................... 303

11. Seasonal Behavior. ................ ................................. 306
111. SurfaceFlwes ........................................................ 32 1
IV. The Surface Layer .................................................... 334
........ 353
epening ................................................. 370
VII. Differential Heating and Cooling ................ 380
VIII. Outflow .............................................................. 391
IX. Inflow ............................................................... 405
X. Mixingbelowthe Surface Layer ........................................ 413
XI. Modeling ............................................................. 422
XII. Reservoir Destratification by Bubble Aerators ..................... 440
XIII. Summary ............................................................. 45 1
Acknowledgements.. .. ................. 455
References ........................................................... 455

1. Introduction

Our understanding of the hydrodynamics of lakes has developed

.rapidly over the last few years. With new, sophisticated instruments, our
powers of observation are greater, giving us deeper insight into the
physical principles governing the motion and mixing of water in a lake.
303
Copyright 0 1990 Academic Press. Inc.
All rights of reproduction in any form reserved.
ISBN 0-12-002027-0
304 Jorg Imberger and John C . Patterson

The hydrodynamics of lakes is intimately connected with the motion of

a stratified fluid affected by external forcing, the instability of these
motions, and the dynamics of turbulence in a stratified environment.
Most lakes become stratified as solar energy is input at the surface.
Through the expansion of the water, this thermal energy is converted to a
mechanical stability, which arrests motion in the lake. The stronger the
thermal insolation, the stronger the stratification and, in general, the
more quiescent the water body. Wind, stream inflow, and the withdrawal
of water all modify the stratification. However, as explored in this
review, there are mechanisms induced by the stratification that cause
active horizontal water transport and that become more intense as
stratification increases.
The dynamical balance within a lake may be likened to an engine; the
disturbing forces work against the potential energy gradient set up by
radiation. The internal dynamics of this engine are extremely inefficient.
A basin scale disturbance must begin large scale basin seiching, which
then degenerates into smaller scale internal waves and intrusions,
ultimately used by a host of instability mechanisms to produce tur-
bulence. This turbulence then induces a buoyancy flux upwards, raising
the overall center of gravity of the lake. Up to 90% of this mechanical
energy is lost to dissipation.
Lake hydrodynamics appears to have attracted research in two main
areas. First, there is a large body of work in the complicated patterns of
seiching in basins of arbitrary shapes, with and without rotation (Hutter,
1984, 1986; Stocker and Hutter 1986). The second focuses more on the
consequences that follow these internal wave motions, particularly stress,
small scale motions, and mixing processes (Fischer et al., 1979; Imberger
and Hamblin, 1982; Imboden et al., 1983; Imberger, 1985; Imberger,
1987).
Interest in small scale mean motions and turbulence as important
components of lake hydrodynamics is recent: reviews of lake hydrodyna-
mics by Mortimer (1974) and Csanady (1982) mention almost nothing
about these aspects, concentrating instead on internal and surface
seiching introduced by wind stresses.
The similarities between physical oceanography and physical limnology
are many, especially in the second category of work, and with this in
mind, the reader is referred to the following excellent reviews: Garrett
and Munk (1979), Gregg and Briscoe (1979), Caldwell (1983), Fritts
(1984), Holloway (1986), Gregg (1987), Hopfinger (1987), Thorpe
 Physical Limnology 305

(1987), and Muller and Garwood (1988), which discuss the motion of a
stratified fluid and mixing due to turbulence. There also have been two
recent major conferences dealing with the oceanic internal gravity waves
(Muller and Pujalet, 1984) and the dynamics of the oceanic surface mixed
layer (Muller and Henderson, 1987). Much of this material is relevant to
an understanding of the hydrodynamics of reservoirs and lakes. To avoid
duplication, this review concentrates on smaller scale motions and the
resulting mixing. A description of the action of rotation has therefore
been almost completely omitted since it concerns the dynamics of internal
waves, and it is adequately described in Hutter (1984,1986) and Stocker
and Hutter (1986). Circulation and mixing in ice-covered waters is an
extremely important subject for a large number of lakes, but this has also
been omitted because of the existence of the review by Carmack (1986).
Because of the rapid development of the subject, much of the
description in this review is speculative and concentrates on findings at
the Center for Water Research. Data from the Wellington, Canning, and
Harding Reservoirs in Western Australia are used as illustrations of
processes. These data sets have been used to set the stage for the review
and literature on the dynamics displayed by these data has been
referenced extensively.
The review begins with a discussion of seasonal behavior and the role
of the geographic location of a lake. Through a new, dimensionless
parameter the dynamics of a lake in the tropics at the time of peak
stratification can be compared with the dynamics of a lake in the
temperate zones at the beginning and end of the stratification cycle. This
section is followed by four sections on the dynamics of the surface layer
(formerly called the mixed layer) beginning with a discussion of the
surface fluxes, followed by the vertical dynamics of the surface layer.
Upwelling, differential deepening, and differential heating and cooling
are presented in separate sections as deviations from the vertical picture.
Our knowledge on outflow has increased considerably since the last
review by Imberger (1980) and this material is updated here. There are
also a number of significant developments in our understanding of the
inflow dynamics, particularly where the inflow separates from the river
channel.
Mixing below the surface layer is also discussed. Because of recent
reviews by Gregg (1987), Hopfinger (1987), and Thorpe (1987), in-
dividual mechanisms are not discussed, and the discussion concentrates
on the underlying dynamics contributing towards mixing in the hypolim-
306 Jorg Zmberger and John C. Patterson

nion and the metalimnion, on the classification of turbulent activity, and

on the parametrization of mixing internal to the hypolimnion and
boundary mixing. These sections set the stage for the review in Section
XI of modeling of lakes where stratification is the dominant force. Lastly,
because of eutrophication problems in lakes throughout the world and
the attention they are receiving, we have included a discussion of some
recent advances in the use of bubble aerators in reservoir destratification.
In this review, the words “lake” and “reservoir” are synonomous, with
“lake” defined as an impoundment that stores water from a catchment. A
reservoir is a man-made lake, which differs from a natural lake in some
ways. In a reservoir, the outflow can be controlled, and, because the
throughflow is more rapid, the water residence time is shorter. The
hydrodynamics of the two are not greatly different, unless the through-
flow is so strong that it affects stratification.

II. Seasonal Behavior

A lake or reservoir is a body of water contained in a river valley,

insulated by the bottom but with heat transfers at the surface. Short wave
solar radiation (of wavelength 300nm to 1OOOnm) and long wave
radiation (of wavelength greater than 1OOOnm) emitted by clouds and
atmospheric water vapor heat the water, while evaporation, sensible heat
transfer, and radiation from the surface most often cool the water. The
net heat input from these sources depends on the season and the
meteorological conditions at the time and site; the balance may change
dramatically not only from day to day but from hour to hour.
The net daily heat flux at the surface of the Canning Reservoir, 50 km
southeast of Perth, Western Australia, is shown in Figure 1. The
reservoir is at a latitude of 32.5”s with a Mediterranean climate of cool,
wet winters from June to August, and hot, dry summers from December
to March. The sensible and latent heat fluxes used in Figure 1 are
calculated from meteorological data gathered 2.0 m above the water
surface using bulk aerodynamic flux equations (see Section 111) with
coefficient values applicable to a neutral atmosphere (Tennessee Valley
Authority, 1972; Fischer et al., 1979).
The variance of the daily net heat flux was larger in spring and summer
(September to March) than in autumn and winter (April to August). This
was mainly due to variable cloud cover and higher water surface
 Physical Limnology 307
400

200

-
N

2
X

i *
+
z
w

m
YI

-200

-400 I I I 1 I I I I I I I I I I

J J A S O N D J F M A M J J A S

MONTH
1986 1987

FIG. 1. Net heat flux at the surface of the Canning Reservoir. The data were collected
with a meteorological station mounted on a specially designed buoy kept in position with a
taut mooring. The sensors were 2.0 m above the water level at all times. The smooth line
represents the averaged flux, and the tick marks represent the start of the month.

temperature, which led to large back radiation and strong evaporation on

cloudy, windy summer days. Also, the smoothed net heat flux was
negative only for a brief period in autumn (March to May). During
winter, the net flux was approximately zero. In spring and summer, the
smoothed net heat flux was positive with a mean value of approximately
80 W m-*.The total net heat exchange for the year was positive, but the
average lake temperature did not increase between June 1986 and June
1987.
This implies that the sum of the heat flux due to the river inflow, which
tends to be cold, and the loss of heat due to the outflow, which is
somewhat warmer, was comparable in magnitude to the fluxes through
the surface. Inflow and outflow patterns, which differentiate lakes from
reservoirs, must therefore have a significant influence on the heat budget
of a lake (see also Myrup et al. (1979) for a detailed heat budget of Lake
Tahoe).
308 Jorg Imberger and John C. Patterson

50.0

I
J
I
J A
I
S
I
O
I I
N
I
D
I
J
1
F
I
M
1
A
I
M
I
J
I
J A
I

MONM
1986 1987

FIG.2. The thermal characteristicsof the water in the Canning Reservoir corresponding
to the net heat flux shown in Figure 1. Drops were collected at fortnightly intervals
throughout the 15 months.

The Canning Reservoir thermal characteristics corresponding to the

period described in Figure 1, are shown in Figure 2. During winter, cold
weather mixed the lake over the whole depth, leading to an isothermal
water body of approximately 12°C. The inflow during the winter had
much the same temperature, as seen in Figure 2, and the temperature at
the bottom of the lake remained at 12°C. Just before the minimum
temperature was reached at the bottom of the lake, the surface layers
began to develop a weak thermal gradient in direct response to the
positive net heat flux (see Figure 1). This heating intensified until, at the
 Physical Limnology 309

end of December, the temperature at the surface had reached 24°C over
a depth of nearly 10 m. The stratification during this period appeared to
be unable to suppress the turbulence in the deep part of the lake between
20 and 30m depth, where the temperature rose from a low just above
11°C in September to 12°C in December. After December, the warmer
surface temperatures increased the back radiation, the evaporation, and
the sensible heat loss, and the surface temperatures rose only slightly,
reaching a maximum of 25°C in late February.
After this time, penetrative convection, introduced by a net surface
cooling, led to a characteristic decrease of the surface water temperature
of the top 15m, which continued from late February to June. Also
noticeable was the gradual deepening of the isotherms at the base of the
surface layer, induced by the erosion at the base by both natural
convection and wind-induced turbulence. Surface cooling has two effects.
First, the kinetic energy of the penetrative convection thermals erodes
the stable structure underlying the surface layer. Second, the cooling
decreases the stability (the density difference between the surface layer
and the underlying water), thus making the base more susceptible to
erosion by turbulence from within the surface layer. By June, the lake
again had become isothermal.
In limnology textbooks (Hutchinson, 1957; Wetzel, 1975; Henderson-
Sellers, 1984b), the upper warm layer is called the epilimnion, the strong
temperature gradient region below is called the metalimnion, and the
colder water in the lowest part of the lake the hypolimnion. These terms
are useful for description, but as lmberger (1985) pointed out, the
understanding of the properties of water behavior in these regions has
recently been revised.
As the following sections will show, the surface layer is not as well
mixed as is generally believed, nor is it in a state of uniform and constant
turbulence. The surface layer, or epilimnion, responds to diurnal changes
in surface fluxes. The metalimnion contains the major temperature
changes between the surface and bottom water. It is made up of many
temperature steps, and it is unrealistic to define the thermocline as that
depth at which the temperature gradient is maximum, as has been
suggested by previous authors (see Wetzel (1975) and Straskraba (1980)).
Typical profiles from the data set of Figure 2 are shown in Figure 3. The
temperature gradient is a strong function of the resolution of the
instrument. For the smoothed profile with a resolution roughly matching
the vertical extent of the interface (Figure 3b), the gradient reaches only
 w
b C
5;

L I I I I I I I I 1140
11 18 25 -15 15 -15 0 15 -15 0 15
TEMPERATURE StK dED FINESCALE MICROSTRUCTURE
(.C) dTMz ('C/rn) dT/dz ('C/rn) dT/& ( Q n )

FIG.3. Typical profile data obtained in the Canning Reservoir on 26 January 1987 at a central location. (a) Temperature obtained
with a standard conductivity, temperature, depth profiling instrument. (b) Smoothed temperature gradient. (c) Enhanced temperature
gradient with an effeective resolution of 0.02 m. (d) Temperature gradient obtained with a microstructure profiler which had a resolution
of 1 mm. This profile was taken a little earlier than that shown in (a), (b), and (c) and only extended to a depth of 20m.
 Physical Limnology 311

3°C per meter, whereas at the finest resolution obtained with a

temperature microstructure instrument (which resolved the profile to
1mm intervals) a maximum temperature gradient of 15°C per meter was
observed (Figure 3d). The coarse definitions find their origin in the early
measurements of Wedderburn (1906, 1911, 1912) where temperatures
were measured at every one or two meters.
The metalimnion is also poorly defined, for example, given the data in
Figure 3a, it is not clear whether the temperature step at 6 m depth
should be included. For this reason Imberger (1985) and Imberger (1987)
introduced the classification of a diurnal surface layer, a parent thermo-
cline, a metalimnion, and hypolimnion. The parent thermocline is the
temperature step caused by the most severe mixing or deepening event in
the immediate past. The metalimnion is the water body from the parent
thermocline to a depth where the smoothed temperature has reached a
certain percentage of the coldest temperature in the hypolimnion.
The parent thermocline is illustrated in Figure 4 with data collected
during a hot, cloud-free day at the Wellington Reservoir. At 0830 hours,
there is a clear surface layer, uniform in temperature from the surface to
approximately 7.0 m. This layer was mixing (Monismith and Imberger,
1988) by penetrative convection; the parent thermocline is the gradient
region found at 6.5 m. Following this, the weather became extremely hot
(temperature at 1200 hours was 42"C), there was no wind, and the water
surface was glassy all day. By 1500 hours (Figure 4), considerable surface
heating had formed a diurnal thermocline to a depth of approximately
1m, with a temperature rise in excess of 4°C. By 1700 hours a brisk wind
of 6ms-' caused the surface to mix (Figure 4), causing the diurnal
thermocline to deepen toward the parent thermocline. By definition,
once the diurnal thermocline has reached its maximum depth, it becomes
the new parent thermocline. In some cases, this may be deeper than the
previous parent thermocline (see Imberger (1987), Schertzer et al.
(1987), and Strub (1983)). Generally during spring, lakes stratify and the
location of the parent thermocline becomes shallower with time; in
autumn, during the lake's cooling phase, the parent thermocline lies
progressively deeper.
The temperature distribution of a lake depends on the cumulative
effect of the surface fluxes, the inflows, the outflows, and the rate of
vertical mixing. There has, however, been a great effort made (see
Straskraba (1980)) to provide empirical formulae to predict the thermal
characteristics of a lake according to its location. Three parameters are
312 Jorg Zmberger and John C . Patterson

TEMPERATURE (“C)

RG. 4. Evolution of a diurnal surface layer. Data taken in the Wellington Reservoir on
25 February 1985. Solid line: data collected at 0824 hours during calm conditions. Dashed
line: data collected at 1501 hours at time of maximum surface temperature gradient. Dotted
line: data collected at 1701 hours; cooling had commenced.

usually characterized: the surface temperature, the bottom temperature,

and the depth of the seasonal thermocline. Given that the solar incident
radiation is determined, at a particular time, by the latitude of the lake
fp, Straskraba (1980) suggested the following relationship for the surface
temperature:
8, = 28.1 - 0.34 fp’+ (0.54 - 0.045 fp’ + 0.0146 fpt2
- 1.97 x lo4 9’3)sin( z + y ) , (2.1)
where
fp’ = fp - 3.4,

[ 60
y = 240
for fp > 0 (northern hemisphere),
for fp < 0 (southern hemisphere),
and t is the Julian day.
 Physical Limnology 313

This formula was derived by curve fitting to the measured surface

temperatures of over 50 lakes, both in the southern and northern
hemispheres. Straskraba (1980) mentions that the expected temperatures
may be higher for shallower lakes and lower for deeper lakes than that
predicted by (2.1). Equation (2.1) was tested for two lakes in Western
Australia: Lake Argyle is very large, with a volume of 5.6 x lo9m3 and a
depth of 45m, located at a latitude of 16.5"s; Wellington Reservoir
(Patterson et al., 1984) has a volume of 186 X 106m2,a crest level depth
of 30m, and is located at 33.5"s. Data from these lakes are shown in
Figure 5. The fit for the Wellington Reservoir is reasonable, but for Lake
Argyle the departure both in the mean annual surface temperature and
the annual variation is severe, so such empirical generalizations must be
treated with caution. However, (2.1) illustrates the general trend with
latitude: tropical lakes are warmer and have smaller annual variability,
while lakes at higher latitudes are colder and have larger annual
variations.

LAKE
ARGYLE >

J F M A M J J A S O N D
MOMH
FIG.5. Mean surface water temperature for the Wellington Reservoir and Lake Argyle.
Solid line: data. Dashed line: empirical relationship (2.1) suggested by Straskraba (1980).
314 Jorg Zmberger and John C . Patterson

The temperature of water at the bottom of a lake is determined either

by the temperature of the coldest inflow (Hebbert et al., 1979) or the
temperature at the time of lake overturn (Robarts et al., 1982).
Straskraba (1980) assumes that the bottom temperature remains ap-
proximately constant, which is true for deep lakes such as Lake Tahoe
(Henderson-Sellers, 1984b), but is not true for lakes with active mixing.
Even deep lakes such as Lake Baldegg (maximum depth 70m) show
definite signs of warming during the summer months (Imboden et al.,
1983).
Ignoring the warming due to mixing, Straskraba (1980) suggested that
the bottom temperatures are given by
28.1 - 0.6 @ for 4 5 40",
e,'[4 for @ > 40".
This does not encompass lakes at extreme latitudes such as Lake Erie
(Schertzer et al., 1987), Lake Ontario (Boyce et al., 1983), or Lake
Babine (Farmer and Carmack, 1982) where temperatures fall below 4°C
in winter. The coldest temperatures given by (2.1) should match the
temperatures given by (2.2), since during the coldest period, most lakes
are isothermal. Again, agreement between (2.1) and (2.2) is only
marginal.
Such empirical generalizations ignore many factors that have a
profound influence on the temperature distribution and heat budget. As
mentioned above, inflows and outflows often contribute significantly to
the thermal budget, and this is illustrated in Figure 6 with data from
Patterson et al. (1984). In 1976, water was withdrawn from Wellington
Dam only from an offtake at mid-depth. In 1977, however, most of the
water was withdrawn from a scour valve at the base of the dam wall. The
mid-depth draw policy intensified the metalimnion, which in turn
increased the stability of the lake and thus reduced mixing throughout
(Fischer et al., 1979). By contrast, the bottom withdrawal reduced the
volume of the cold, salty bottom water, which reduced stability and
allowed surface fluxes to mix the lake much earlier in the year.
These variations cannot be incorporated into a simple empirical
formula, but require an estimate of stability of the lake water. Such
stability measures have long been advocated. Hutchinson (1957) intro-
duced the definition for stability:

s, = ( z - z,)A(z)p(z) dz, (2.3)

 Physical Limnology 315

40.001

1975 1976 1977 1978

FIG.6. Field measurements of the thermai characteristics of the Wellington Reservoir

from 1975-1978. The metalimnion present in 1976 was absent in 1977 due to the bottom
scour policy instituted in 1977 (Patterson et af., 1984).

where z is the vertical coordinate from the bottom of the lake, A(z) is the
area of the lake at height z , p(z) is the water density at a height z , z, is
the water depth, and zg is the height to the center of volume of the lake,
defined by

Straskraba (1980) suggests that the depth of the center of the meta-
limnion and the rate of turbulent mixing depend on the magnitude of S,.
He also observed that for a particular lake basin undergoing temperature
stratification given by (2.1) and (2.2), S, should be small near the
equator, rise to a maximum at middle latitudes, and again reach zero at
latitudes of approximately 70"; the actual stability being dependent on the
density of the water which is a highly nonlinear function of the
temperature (UNESCO, 1981). Idso (1973) reintroduced this concept of
stability, but multiplied S, by gravity g and divided (2.3) by A(z,),
removing the absolute size of the lake from the definition of the stability.
However, even with these modifications, S, is still dimensional. Further,
the concept of stability expressed by (2.3) is deficient since it only
accounts for the stabilizing influence of the stratification and not the
destabilizing influence of the disturbances of wind, inflows, and outflows.
The problem can be addressed by generalizing the surface layer
stability criterion first introduced by Sverdrup (1949, developed further
by Hurley Octavio et al. (1977), Spigel and Imberger (1980), Thompson
316 Jorg Imberger and John C. Patterson

and Imberger (1980), and formalized by Imberger and Hamblin (1982) as

the Wedderburn number

W = g’h2
- (2.5)
u2,L ’
where g’ is the modified acceleration due to gravity across the base of the
surface layer, h is the thickness of the surface layer, u , is the water
friction velocity due to wind stress, and L is the fetch length. Here
g’ = g Aplp,, where Ap is the density jump across the base of the
surface layer and po is the hypolimnion density. The friction velocity u, is
defined by (3.1) below. This number is dimensionless and represents, as
explained in Imberger (1985) and Imberger (1987), the ratio of the
baroclinic pressure force g’h2 at the point of upwelling and the surface
force u:L imposed by the wind stress; both forces being calculated per
unit width of the lake. This number will be elaborated upon in Section
IV, but a simple generalization is suggested to describe the behavior of
the lake as a whole.
Consider a general lake with an arbitrary stratification p ( z ) being acted
upon by a general wind field with a surface friction velocity u , ( x , y ) ,
where x and y are the horizontal coordinates embedded in the lake’s
surface. As this wind stress is imposed on the surface layer, there will be
a net force acting to overturn the density structure of the water column.
Taking moments about the center of volume located at zg, we obtain for
equilibrium

POUZ, U ( z m - zg) = (zg - zo)MgS, (2.6)

where zo is the center of gravity of the actual water mass with a density
stratification p ( z ) , M is the total mass of water, and /3 is the angle
subtended to the vertical by the center of mass at the center of volume.
This leads to the ratio we shall call the Lake number

(2.7)

Now noting that zo is defined by

 Physical Limnology 317

the Lake number may be rewritten as

Following the work of Spigel and Imberger (1980), the angle /3 may be
fixed so that LN is calculated at the point where upwelling commences,
that is when the metalimnion intersects the surface, so that

(2.10)

where zT is the height to the center of the metalimnion, and the fetch is
scaled with A'/*(z,,,), yielding the relationship

(2.11)

Assuming that the wind stress is constant over the surface, then LN
reduces to

(2.12)

where A. is the surface area of the lake A&). For large Lake numbers,
the stratification will be severe and dominate the forces introduced by the
surface wind stress. Under these circumstances, stratification is expected
to be horizontal, with little or no seiching and little turbulent mixing in
the metalimnion or the hypolimnion. Horizontality of the isotherms at
large LN is illustrated by the data set taken at Canning Reservoir on the
18 January 1987, which is depicted in Figure 7. A weak, warm lense of
water resident in the sheltered upstream part of the basin and a very
weak slope of "C m-l towards the dam wall, attributable to a first
mode seiche, are the only departures from this horizontality. As
discussed in Shay and Imberger (1988), the turbulence levels in the
epilimnion and the hypolimnion were extremely low. At the time, the
wind mean shear velocity (averaged over 24 hours) was 0.0033 m s-',
leading to a value of L , = 86. The full implication of LN with respect to
deep mixing will be discussed in Section X.
The interplay between the temperature distribution variation with
latitude and the changing water density due to the elevated temperatures
318 Jorg Zmberger and John C . Patterson

at lower latitudes may be captured effectively by taking the data

displayed in Figure 2, transposing this through the various latitudes via
(2.1) and (2.2), and then calculating L N from (2.12). This was done
by scaling the actual measured surface temperature by the ratio
8,(t, $2)/0s(t,32.5). The bottom temperatures were adjusted to vary
linearly so that at the equator the bottom temperature equalled the
surface temperature, and at 32.5"s the bottom temperature was equal to
that measured in the Canning Reservoir. Here, t is the time of a
particular profile and $2 is the second latitude. The scaled temperature
profiles were then used to compute L N from (2.12), yielding the
two-dimensional variation of LN shown in Figure 8. The diagram shows

TEMPERATURE ("C)
0.0
- 26.0'

10.0

--E
f 20.0
x

30.0

40.0 470 CA65 CA60 CA50 CA45 CAM CA35 CA30 CM5 CA20 CAiC
A j A a A t u
I I
1000.0 2000.0 3000.0 4000.0 5000.0 6000.0

CUMULATIVE DISTANCE

FIG. 7. (a) Longitudinal transect of isotherms in the Canning Reservoir taken on 18

January 1987. (b) Map of Canning Reservoir showing location of stations.
 Physical Limnology

9000

8000

7000

POISON GULLY
6000

- 5000
w
VI
+
a
z0
4
K
400C

8
J
a
0
s 300C

2000
CANNING RESERVOIR

1000

0000

0000 1000 2000 3000 4000 5000

LOCAL CO-ORDINATES (m)

FIG. 7. (conld.)
320 Jorg Zmberger and John C. Patterson

FIG.8. An example of the variation of the Lake number as a function of season and
latitude. The surface was constructed by computing the Lake number from data gathered in
the Canning Reservoir, but which was transposed to different latitudes using the empirical
relationship (2.1) suggested by Straskraba (1980). A constant wind speed of 8.0ms-' was
applied throughout.

that the maximum stability is obtained at mid-latitude, around the

summer period. Further, it illustrates that it is possible to trade latitude
and season when analyzing the dynamics of a lake. This would mean that
the Canning Reservoir in August, leaving aside the influence of the
earth's rotation, should be dynamically similar (in the sense of dynamic
similitude) to Lake Argyle in February. More generally, it may be said
that lakes in the tropics behave dynamically similar to lakes at mid-
latitudes at the beginning or end of the stratification cycle.
The seasonal pattern described above is often strongly influenced by
chemicals dissolved in the water, which also influence the density of the
water, thus contributing to the stability of the lake. The origin of these
dissolved salts are threefold (Wetzel, 1975). First, in coastal lakes, sea
water may enter the lake periodically. Second, ground or river salt water
intrusions may be present, and third, salts may be accumulated from the
decomposition of sediments and organic matter. If the dissolved salts
influence the density of water enough to prevent the normal winter
overturn, the lakes are called meromictic. Examples in the literature are
too numerous to list here, and it suffices to mention that they range from
very large lakes, such as the Dead Sea (Steinhorn and Assaf, 1980;
 Physical Limnology 321

Steinhorn and Gat, 1983); to small shallow lakes (MacIntyre and Melack,
1982; Bunn and Edward, 1984). The Wellington Reservoir is an example
(see Imberger, 1987) where a strong variability of the salinity of the
inflow led in some years to meromictic behavior.
In general, the dissolved salts influence the dynamics of the lake only
through the equation of state, and provided the Lake number and other
parameters allow for salinity variation, their dynamics are similar to
single species stratification. In the section on deep mixing, we discuss an
exception to this statement, and under certain circumstances, such as in
Lake Kivu (Newman, 1976), double diffusive instabilities may arise in
salt-stratified lakes due to different rates of molecular diffusion between
temperature and salt (see Turner (1985) for a complete review on
multi-components).
Many attempts exist to empirically correlate the dynamics of a lake to
variables such as the surface area (Ward, 1977) and geographical trends
such as altitude (Hutchinson, 1957; Wright, 1961; Loffler, 1968; Khom-
skis, 1969; Lerman and Stiller, 1969; Bella, 1970; Lerman, 1971;
Darbyshire and Colclough, 1972; Khomskis and Filatova, 1972; Blanton,
1973; Idso and Cole, 1973; Lewis, 1973; Sundaram and Rehm, 1973;
Tzur, 1973) and length of the lake (Yoshimura, 1936; Patalas, 1960 and
1961; Arai, 1964; Ventz, 1973). These correlations are useful only in that
they are special cases of the lake number LN formulation.

III. Surface Fluxes

The interaction between the atmosphere and a lake occurs at the lake’s
surface. A great deal of work has been done to determine the fluxes of
momentum, energy and mass across the air-water interface and many
numerical formulae have been developed. These methods are reviewed
by Tennessee Valley Authority (1972), Straskraba (1980), and
Henderson-Sellers (1986).
Short wave radiation (300 nm to 1000 nm) is usually measured directly.
Long wave radiation (greater than 1000nm) emitted from clouds and
atmospheric water vapor can be measured directly or calculated from
cloud cover, air temperature, and humidity (Tennessee Valley Authority,
1972). The reflection coefficient, or albedo, of the short wave and long
wave radiation varies from lake to lake and depends on the angle of the
sun, the color of the water, and the surface wave state. Back radiation
322 Jorg Zmberger and John C. Patterson

from the warm water surface may be calculated from the black body
radiation law, it also may be measured by pointing a radiation instrument
towards the water surface.
It was shown by Strub and Powell (1987), Marti and Imboden (1986),
Keijman (1974), Sadhuram et al., (1988), and many other authors that
simple aerodynamic bulk .,formulae, with constant transfer coefficients,
can be used to calculate the momentum, the sensible heat, and the latent
heat fluxes on a seasonal basis:

E -
-= q ' w ) = -u*q* = -C,U,(q, - q.), (3.3)
PLV
where z is the surface stress, p is the density of the air, u' and w' are the
horizontal and vertical fluctuation of velocity, the overbar is a time
average long enough to average out the short term energy but short
enough to allow for synoptic variability (Busch, 1977; Smith, 1980;
Geernaert et al., 1987), u , is the shear velocity in the air, CD is the
momentum or drag coefficient, U, is the air velocity at a certain height z
above the water surface, H is the sensible heat transfer, Cp is the specific
heat of water, 6" is the temperature fluctuation, 8, is the temperature
scale, CHis the heat transfer coefficient, often called the Stanton number
(Geernaert et al., 1987), 8, is' the water surface temperature, 8, is the
temperature of the air at the height z above the water, E is the latent heat
flux, Lv is the latent heat of vaporization, q' is the specific humidity, q* is
the specific humidity scale, Cw is the latent heat transfer coefficient or
Dalton number (Geernaert et al., 1987), q, is the specific humidity at a
height of z above the water surface, and q. is the specific humidity at
saturation pressure at the water surface temperature.
Equations (3.1), (3.2), and (3.3) are only valid for a stationary surface.
Since the water moves, U, should be replaced by U, - Us, where Us is the
water surface mean velocity (see Businger (1973) for a full discussion),
although for most wind speeds, this is a small correction.
The value of the transfer coefficients CH and Cw for such bulk
modeling has received a great deal of attention and, in the case of lakes,
may be estimated by carrying out a heat and water budget over seasonal
 Physical Limnology 323

time scales (for example, Strub and Powell (1987) for Castle Lake, Marti
and Imboden (1986) for Lake Sempach, Myrup el al. (1979) for Lake
Tahoe, and Taylor and Aquise (1984) for Lake Titicaca). These
references suggest a value of CH= Cw = 1.9 X (referenced to 10 m)
yields an adequate description of the total stored energy over a seasonal
time scale. This may be compared with a lower value of 1.45 x
suggested by Hicks (1972). Constant values are sufficient for seasonal
lake modeling since the thermal budget is self-regulating; an underestim-
ate of the heat loss will cause the water surface to heat, thus increasing
heat loss, while an overestimate of heat loss will cause the water surface
to cool and, by (3.2) and (3.3), this will lead to the heat loss being
decreased.
At smaller time scales (from hours to days), this negative feedback
mechanism no longer suffices because the thermal inertia of the water in
the surface layer is too great. The influence of the air column stability
and the water surface roughness must therefore be accounted for, and
these may introduce considerable variability to all three transfer
coefficients. Excellent reviews exist on this topic: Businger (1973), Dyer
(1974), Stewart (1974), Garratt (1977), Wu (1980), Donelan (1982), and
Blanc (1985). There are three main measurement techniques to deter-
mine the instantaneous surface fluxes at the water surface: eddy
correlation, profile, and dissipation. These have been used to calibrate
(3.1) to (3.3) thus giving values for CD, CH, and Cw. A description of
each technique is given by Dobson et al. (1980) and Blanc (1983). The
eddy correlation method is direct; the profile method uses mean velocity,
temperature, and humidity profile data; and the dissipation method uses
the spectral characteristics of the high wave number turbulence. Results
from the eddy correlation technique usually agree with each other within
10% (Blanc 1987); profile techniques usually agree with eddy correlation
methods to within 25% (Miyake et al., 1970; Wucknitz, 1976). Large and
Pond (1981) and Smith and Anderson (1984), in comparisons between
the dissipation method and the eddy correlation technique, found that
the two agreed to within 40%. Within similar schemes, there can be even
greater variability (see Lo and McBean (1978) for interprofile
comparisons).
The reliability of the bulk aerodynamic equations (3.1) to (3.3)
depends on the sensors being located in the internal boundary layer of
the lake. Air coming from land will meet a different roughness, humidity
and temperature over the lake and the internal boundary layer will be
324 Jorg Imberger and John C. Patterson

established with new equilibrium profiles (Taylor, 1970; Venkatram,

1977; Mulhearn, 1981; Claussen, 1987; Garratt, 1987). Peterson (1969)
recommends that the fetch be at least 100 times the height of the sensor,
but Hicks (1975) advocates a more cautious ratio of 1OOO. The recent
simulations of Garratt (1987) suggest that the growth of the internal
thermal internal boundary layer for stable conditions is given by
( 8 )- I y L ,
h = 0 . 0 1 4 x 1 ” Ugmhe
(3.4)

where h is the height of the internal boundary layer, x is the distance

downstream, U, is the free stream air velocity, g is the acceleration due
to gravity, A 8 is the potential temperature difference between the
continental air (constant throughout the approaching boundary layer and
equal to the land surface temperature) and the water surface tempera-
ture, and 8 is the absolute temperature of the air. For a typical
A 8 = 10°C and Um= 5ms-’, this would indicate a layer thickness at
lo00 m of 3.8 m, which lies between the recommendations of Hicks
(1975) and Peterson (1969). Equation (3.4) also agrees well with the
results presented by Mulhearn (1981) and Garratt and Ryan (1988) for
boundary layers close to the shore, even though (3.4) was derived for
considerably larger length scales. The momentum and water vapor
internal boundary layers may be expected to have similar thickness
(Garratt, personal communication).
The unstable case, where the water is warmer than the air, does not
appear to have been solved, but the case of cool ocean air overflowing a
warm land mass was recently reviewed by Venkatram (1986). The
roughness was fixed and not dependent on the shear stress, as in the case
of the model by Garratt (1987), but the internal boundary layer was still
observed to grow as x1I2.
For very small boundary layer thicknesses, the application of equi-
librium profiles severely constrains the position of the instruments, as they
must be placed in that area of the boundary layer described by the law of
the wall (Tennekes and Lumley, 1972) a region that extends only to about
10% of the layer thickness (Sorbjan, 1986). In the example above, where
the layer thickness was 3.8m, the instruments must be placed approxi-
mately 0.4m from the water’s surface. From Donelan (1980), the root
mean square surface roughness due to wave action may be estimated, for
a steady wind, to be 0.02m, which implies a value for Hi13 of O.lm ,
where Hi13 is the significant wave height. The region available for
 Physical Limnology 325

positioning of the sensors is thus limited to between 0.1 m and 0.4 m. To

prevent spray damage, the sensors would probably have to be placed
close to 0.4 m from the water surface.
If the meteorological sensors are embedded in this law-of-the-wall
region of the internal boundary layer, and the horizontal advection
effects within the boundary layer can be ignored, the normal turbulence
relationships can be used. Monin and Obukhov (1954) parameterized the
non-dimensional gradients of wind, potential energy, and specific humi-
dity in terms of the stability parameter z l L , where
-pu3,ev
L= (3 * 5 )
LV
+
8, = e ( l 0.61q) is the virtual temperature (“K), k is the von Karman
constant and L is the estimate of the Monin and Obukhov length.
The similarity relationships introduced by Monin and Obukhov (1954)
and Businger (1955), amongst others, are assumed to hold in the
law-of-the-wall region:

kz d e
(3.7)
e, dz
kz dq

Dyer (1974) summarized the various proposed forms for the functions
@ M , @ H , and GW. For a convecting boundary layer ( z / L< 0),

(3.10)

and for a stratified boundary layer with z l L > 0,

Hicks (1976) reanalyzed the Wangara data (Clarke et al., 1971) and
found (3.11) to hold for only weakly stable (0 < z / L < 0.5) boundary
326 Jorg Imberger and John C . Patterson

layers. He suggested that for z / L > 10,

z
&=0.8-, (3.12)
L
whereas Carson and Richards (1978) postulated that for the in-between
region (0.5 < z / L < lo),

@M [
= 8 - 4.25 ):( I'-+):( . (3.13)

Paulson (1970) substituted (3.9) to (3.13) into (3.6) to (3.8) and carried
out the integration to obtain

(3.14)

(3.15)

(3.16)

where zM , zH , and zw are roughness lengths, and defining

= [ - 16(;)] 114
, (3.17)
for the unstable case,
l+x
+ In( y - 2 tan-'(x)
VM= 2 In( 7)
1+x2
) + -,n2 (3.18)

and for the stable case,

z
-5(3 0 < - < 0.5,
L
(3.20)
0.5(i)-2-4.25(;)-1-71n(i)

- 0.852, 0.5 < < 10.0,

L
I (3.21)

I ln(;) - 0.76(:) - 12.093,

z
- > 10.0.
L
L (3.22)
 Physical Limnology 327

These relationships appear to have been tested over the range from
-15 < z / L < 15, but for the values larger than 15, the similarity functions
must be extended as suggested by Sorbjan (1986).
Equations (3.18) to (3.22) can now be used to relate the neutral to the
actual transfer coefficients, leaving only the surface roughness to be
estimated. The roughness lengths are related directly to the drag
coefficients, as can be seen by substituting (3.14) to (3.16) into (3.1) to
(3.4). Under neutral conditions,

(3.23)

(3.24)

(3.25)

For nonneutral conditions, the procedure yields

where a stands for D, H, or W. Geernaert and Katsaros (1986) point out

that the roughness, which is a function of u * , will be influenced by the
stability of the air and the boundary layer so that (3.23) becomes

(3.27)

Now eliminating z and U from (3.1), (3.27), and (3.14) yields

(3.28)

rather than (3.26). The correction (l/k)ln(zMN/zM) is small, seen by

noting that the roughness lengths may be given by the Charnock (1955)
relationship

(3.29)

(3.30)
328 Jorg Zmberger and John C. Patterson

so that (3.28) becomes

(3.31)

The correction at z / L = 1 is about 10%; a reduction for stable conditions

and an increase for unstable conditions. However, Keller et al. (1985)
also investigated the influence of stability on the surface roughness and
found an even greater correction was necessary.
The stability correction can now be implemented in two ways. First,
given the relationships (3.14) to (3.16), these may be substituted into the
definition of the bulk Richardson number (Panofsky, 1963; Deardorff,
1968; Rayner, 1981),
gz A 8 + 0.618,Aq (3.32)
Rig=-(8" U2
to yield a relationship between z / L and Rig that reduces to

(3.33)

where it was assumed that C H N = HWN = CHwN. Deardorff (1968),

however, did not make this assumption, and inverted the equilibrium
form of (3.32) to produce the following relationship (see Strub and
Powell 1987):

(3.34)

(3.35)

C W - e-24 Rie
7 (3.36)
CWN

for stable conditions, and

(3.37)

(3.38)
 Physical Limnology 329

for unstable conditions, where

a = 0.83C;:62, (3.39)
b = 0.25C,$'. (3.40)

These relationships were successfully used by Strub and Powell (1987) in

their heat budget calculations at Castle Lake.
The second approach was suggested by Hicks (1975) and involves an
iterative procedure. Neutral values of CDN,CHN and CWNare chosen and
neutral fluxes are calculated by (3.1) to (3.3). These values are
substituted into (3.5) to calculate the Monin and Obukov length L and
thus z / L . New values of the transfer coefficients, now partially corrected
for stability, may be obtained from (3.26). This procedure may be
repeated and the scheme is said to converge when the estimates for L
converge.
This scheme was implemented by Rayner (1981), and some typical
results from the data set presented in Figure 1 are shown in Figures 9a-i.
The air column, judging by the difference in the air and water
temperature shown in Figure 9a was unstable during the nights 29
January and 2 February particularly, but was stable for most of the day
period. The evaporative flux (Figure 9d) is a major contributor to the
heat budget, so the value of Cw is extremely important for an overall
heat budget calculation. The variation of the stability corrected transfer
coefficient for the anemometer height of 2.0m (Figures 9g and h) show
large variations throughout 24-hour periods. The algorithm was bounded
so that -15 < z / L < 15; larger values were not allowed. Dependence on
stability is now generally accepted to be given by (3.26), although direct
verification in terms of CD, CH, and Cw is scarce. For lakes, the reader is
referred to Strub and Powell (1987) for verification under stable
conditions, and to Graf et al. (1984) for both stable and unstable
boundary layers. In general, it appears that the stability corrections
implied by (3.26) describe a diurnal heat flux variation and are necessary
to obtain thermal budget closures in diurnal surface layer modeling
(Imberger 1985).
The remaining question concerns the choice of a neutral drag
coefficient, or as indicated by (3.23) to (3.25), the choice of the surface
roughness z,, zH, and zw. Charnock (1955) put forward, from purely
dimensional grounds, the relationship (3.29) and Stewart (1974) sug-
gested in his review, by fitting data available at the time, a value of (Y
330 b a
33

32

B 31

30

_-11
70
15 19 23 27 31
TEMPERATURE ib)

-Q
2.L

fw/m2)
-
110-31S.W. RADIATION (10-3) NET RADIATION LATENT HEAT FLUX SENSIBLE HEAT FLUX
(W/m2)
(10-3) TOTAL HEAT
FLUX (W/m2)

31 -

30 -

29 0.0 0.8 1.6 2.4 0.0 2.0 4.0 6.0 -15 15

(103)MOMENTUM EXCHANGE (103) HEATNASS TRANSFER STABILITY PARAMETER
COEFFICIENT COEFFICIENT
 Physical Limnology 331

between 0.0123 and 0.351. Garratt (1977) added the data from Kitaigor-
odski et al. (1973), Wieringa (1974), Kondo (1975), and Smith and Banke
(1975) to obtain a value of (Y equal to 0.0144; a value that also seemed to
fit inferred drag coefficients applicable to wind speeds up to 50ms-'.
More recently, Wu (1980) arrived at a value of 0.0185. Large and Pond
(1981) derived a value of CDN= 0.012 for wind speeds between 4 and
11m s-l, but increasing with wind speed beyond 11m s-'. Equating the
two formulations at a wind speed of 25 m s-' by using (3.27) leads to a
value of (Y = 0.0097, considerably smaller than suggested by most other
investigators. At low wind speeds (less than 5ms-'), the situation is
unclear, but in general, the results from the above investigators suggest a
constant value of 1 X for C,, (see Hicks (1972)). In general, the
inter-technique variability and data scatter (Blanc, 1985, 1987) outweigh
the above differences, but the most recent values of a; estimated by Wu
(1980), (0.0185); Geernaert et al. (1986), (0.0178); and Geernaert et al.
(1987), (0.0165) appear to suggest a mean value of (Y equal to 0.0175 for
a fully developed wave field.
The large variability in the surface roughness has recently received
considerable attention. Kitaigorodskii (1968) and Kitaigorodskii and
Zaslavskii (1974) already indicated that Charnock's (1955) relationship
(3.29) should be modified to include the wave age Co/u,:

(3.41)

where Co is the phase velocity of the dominant wave (wave number at

maximum energy) and large values of Co/u, represent decaying waves or
large swell and small values indicate a growing sea. This can be rewritten
in terms of the non-dimensional fetch g L / U 2 (Wu, 1985). Many authors
(see Donelan, 1982) have pointed out that for a young sea (short fetch or
small duration), the roughness is greater than under the equivalent
equilibrium conditions and so the neutral drag coefficients may be
expected to be larger. This explains why storm surge modelers (short

FIG. 9. Example of typical heat flux variations computed from data shown in Figure 1.
Day 29 is 29 January 1987. (a) Surface water temperature. (b) Air temperature. (c)
Short-wave incoming radiation. (d) Net total radiation. (e) Latent heat flux. (f) Sensible
heat flux. (g) Total heat flux at the water surface. (h) Drag coefficient C,. (i) Heat and
mass transfer coefficients CH= C, . (j) The bulk Richardson number.
332 Jorg Imberger and John C . Patterson

duration winds) generally require higher values for the neutral drag
coefficient.
Hsu (1974, 1986) and Geernaert et al. (1987) derived a relationship
between the wave age and wave steepness:

(3.42)

where is defined above and Lw is the wave length at the peak of the
energy spectrum. By using both statistics from the JONSWAP formula-
tion and from Chapter 3, Section 4, of the U.S. Army Corps of Engineers
(1977), the wave steepness for equilibrium seas becomes 0.023, yielding a
value of a slightly higher than that proposed by Wu (1980) (ais equal to
0.0185). Graf et al. (1984) investigated drag coefficients for shoaling
waves and large wave steepness and showed that water roughness using
(3.42) yielded good results. Geernaert et al. (1987) carried out a direct
correlation between the neutral drag coefficient C D N and the wave age:

indicating again an increased coefficient for small values of the wave age
(a developing wave surface). The relationships (3.41), (3.42), and (3.43)
also explain the reduction of drag coefficient in shallow water where
Co/u, tends to be larger (Hicks et al., 1974; Emmanuel, 1975; Geernaert
et al., 1987).
Donelan (1982) partitioned the water surface roughness into two parts:

2, = p[ E ( w )dw]lD, (3.44)
and

(3.45)

where wp is the frequency at the peak of the spectrum and /Iis a constant
equal to 0.0125. He then defined a neutral drag coefficient
 Physical Limnology 333

where c is the drag coefficient for an immobile plate with roughness

either z, or z, given by (3.27). Here UN is the neutral wind velocity at
10 m, C , = 0.83 g l w , , C , = 0.83 g / 2 0 , , 9 is the angle between the wind
and the waves at the spectral peak, R, = U N u / ~ u , is the root mean
square surface deviation, and Y is the kinematic viscosity of water. Once
again, the parameters C,, C,, u, z, and z, can all be estimated from an
understanding of the wave climate (Melville, 1977; Donelan, 1980;
Donelan and Peirson, 1987) and depend on properties such as wind
duration, fetch length, and depth of water. Thus three major new
procedures for estimating neutral drag coefficients over a lake water
surface exist: from wave steepness, equations (3.41) and (3.42); from the
age of the wave field, equation (3.43); and from a knowledge of the wave
frequency spectrum, equations (3.44), (3.45), and (3.46).
The dependence on water surface state is extremely important for the
study of lakes as the fetch is usually limited, the winds are often diurnal
and the water depth is highly variable. However, there does not appear
to be available a study comparing the accuracy of any of the above
techniques and the choice remains mainly one of personal preference.
The scatter of data in measurements of CHNand CwNso far exceeds
the trends with changing wind speeds or sea state. Rayner (1981)
suggested using the constant value of C H N = CWN= 1.35 X based on
the works of Hicks (1972 and 1975) and Pond et al. (1974). Friehe and
Schmitt (1976) analyzed data from nine different experimental programs
for various stabilities. They found a value of 0.86 X for CH under
stable conditions and 0.97 X for unstable conditions. However,
Friehe and Schmitt (1976) also pointed out that if the large flux data from
Smith and Banke (1975) are added, the unstable coefficients become
1.46 x the value of CW was 1.32 X independent of stability.
By contrast, Large and Pond (1981) found good agreement with
similarity formulations for CDNand CwNbut found that C H N was equal to
1.13 x lop3 for unstable conditions and 0.66 for lop3 for stable layers.
The data analyzed by Smith (1980), when corrected for stability (see his
Figure 13), indicated a constant value of CHN= 1 X with perhaps a
slightly larger value for the unstable profiles. Anderson and Smith (1981)
found a mean value of CWNof 1.3 X but again with a large amount
of scatter. So far, the dependence on the wave climate parameters has
not been clearly demonstrated (Geernaert et al., 1987).
In closing, it is important to mention that McBean and Paterson
(1975), Bean et al. (1975), and Parker and Imberger (1986) have shown
334 Jorg Zmberger and John C. Patterson

that the wind field over a lake can be extremely variable: this has
enormous consequences for the behavior of the surface layer, as will be
seen in Section VI.

IV. The Surface Layer

The diurnal surface layer is the water counterpart to the atmospheric

internal boundary layer. However, it should be remembered from the
outset that the characteristic time scale h / u , with which the internal
boundary layer responds is of order 30seconds, whereas the same time
scale for the water surface layer is in the vicinity of 900seconds for a
layer thickness of 5m. The energy spectrum of the wind record used to
calculate the fluxes depicted in Figure 9 is shown in Figure 10; it is seen
that there is a thousand fold decrease in wind energy from 10-6s-' to

FREQUENCY (cyclesh)

Ro. 10. Power spectral density of the wind speed during the period shown in Figure 9.
Seventeen days of data were used to compute the spectrum.
 Physical Limnology 335

10-3s-'. However, the energy at the higher frequencies is sufficient to

introduce considerable turbulent kinetic energy to the surface mixed
layer; since the period is comparable to the time scale of adjustment of
the water surface layer, it is unlikely that one would find, except perhaps
in the very near surface region, equilibrium profiles of the form given by
(3.14) to (3.16) (Dillon et al., 1981; Imberger, 1985; Muller and
Garwood, 1988). Even in the very surface region, it is unlikely that
equilibrium profiles exist because of the distortions introduced by wave
induced velocities. The only major exception would be in the case of a
strongly convecting surface layer, where the wind activity is small and the
time scale for the heat loss from the surface is set by the diurnal cycle or
by an even longer synoptic weather pattern. Such conditions have
recently been documented in lakes by Imberger (1985) and Brubaker
(1987) and in the ocean by Shay and Gregg (1984,1986).
Before proceeding with the discussion of the dynamics of the surface
layer and the associated vertical energy budget, let us consider some
actual examples of the evolution of the temperature field in a typical lake
situation. Recent microstructure measurements taken in the Wellington
Reservoir reveal the evolution of the diurnal surface layer. The data were
taken in February 1985 during an extremely hot period, when the
morning heating was strong and where, in the early morning, a southerly
wind with speeds up to 4 m sK1 prevailed which turned to a northwesterly
with a peak wind speed of 6 m s-l (at a height of 2 m) by 1300 hours. The
westerly remained active until 1900hours (see Monismith and Imberger
(1988) for a full description of meteorological conditions).
The initial profile from a region sheltered from the southerly wind is
shown in Figure lla-I. The surface (top 1 m ) had become strongly
stratified with a temperature differential of almost 1°C; the gradient
signal (Figure llb-I) shows little activity in the surface region except for
some isolated high gradient points (Imberger, 1988). The water below
this was turbulent, sustained by a weak gravitational underflow that had
been set up by the early morning differential cooling (Monismith and
Imberger, 1988). Ten minutes after the wind had moved north, causing a
gentle roughening of the surface, the temperature had changed to that
shown in the second profile. The temperature signal (Figure lla-11) and
the temperature gradient signal (Figure llb-11) show the formation of a
small mixed layer at the surface and also that the profile had become
active to a depth of 1.3m, even though the time since the surface had
become exposed to the wind was less than the time scale h l u , . The
336 Jorg Imberger and John C . Patterson

3
dT/dz (“C/m)

FIG. 11. (a) Examples of microstructure profiles collected at the entrance of Salmon
Brook, Wellington Reservoir on 26 February 1985. See Figure 26a for station locations.
Temperature:
I. 0928 SB20
11. 0939 SB20
111. 1002 c10
IV. 1021 c10
V. 1057 C10
VI. 1726 SBlO
(b) Microstructure temperature gradient signals corresponding to data in Figure 1la.
 Physical Limnology 337

turbulence was established at small scales and was not the result of a
large overturn as has been normally associated with turbulence in a
stratified fluid (Gregg, 1987). Imberger (1988) attributed this rapid
spread of the energy to the leakage by internal wave action from the base
of the small surface mixed layer. Twenty minutes later, and at a
somewhat more exposed site further away from the shore, the surface
layer had deepened to about 0.3 m and the temperature distribution
within the lake was more uniform (Figure lla-111). There is clear
evidence of active turbulence extending below the base of the surface
layer to almost 2 m (Figure llb-111). This deepening process continued
and the profile taken another twenty minutes later reveals a “uniform”
surface layer extending to nearly 0.7m. However, as described by de
Szoeke and Rhines (1976), Imberger (1985), and Spigel ef al. (1986), the
interface at the base of the surface layer, initially sharp and sustained by
the surface-introduced turbulence, is now smeared by billow activity
energized by the momentum of the surface layer, which is clearly
illustrated in Figures lla-IV and llb-IV. The leakage through the base of
the mixed layer remained and the turbulent activity extended to nearly
2 m.
During the measuring period, the lake was exposed to a very strong net
heating flux of around 750 W m-*. At approximately 1030 hours, the
wind intensified, reaching a peak of nearly 6 m s-l at about 1100 hours. A
well-defined diurnal thermocline formed at 1.2 m, reflected both in the
temperature (Figure 1la-V) and the temperature gradient signal (Figure
llb-V), but two features are noticeable. First, the heat flux introduced
sufficient buoyancy at the surface to prevent a completely mixed surface
layer from forming, but the temperature gradient signal shows strong
turbulence in this surface layer. Second, the temperature gradient signal
continued to show turbulence below the base of the surface layer,
extending to about 2.0 m.
The wind strength remained at about 4-6 m s-l for the remainder of
the day and only began to decrease at 1700 hours. At the time the last
profile was collected, the wind had decreased to about 3ms-’. As seen
from Figure lla-VI, the surface layer was not well mixed, even though
the turbulence obviously extended down to 2.3m (see Figure llb-VI).
The gradient signal shows an intermittent, patchy turbulence within the
surface layer, although there was a well-defined wind-stirred near-surface
region extending down to 0.06m. The water column below 2.2m
appeared to have become quiescent.
338 Jorg Imberger and John C. Patterson
a b
I I I 1 I

I I I 1 I I
L2:.5 ’ 22.0 22.5 23.0 -50 0 50

TEMPERATURE (“C) dT/dz (“C/rn)

FIG. 12. Temperature microstructure data from the Wellington Reservoir (13 March
1982) at the central basin after a very strong sea breeze had deepened and tilted the surface
layer. (a) Temperature. (b) Temperature gradient. (After Imberger (1985).)

Now consider an example (Figures 12a and 12b) from the diurnal
sequence documented in Imberger (1985). The profile was taken at a
time when there was a net heat loss of about 250Wm-’, a strong
wind-induced surface layer velocity of about 0.12 m s-’, and an active
wind stress. As seen in Figure 12, the surface layer was extremely
well-defined (variation less than O.O3”C),the base of the mixed layer was
diffuse due to billow activity (Imberger, 1985), and the water below the
base of the surface layer was relatively quiescent.
The third example, shown in Figure 13, also comes from the diurnal
data set presented in Imberger (1985) but was collected when natural
convection completely dominated the surface layer dynamics. The profile
is characterized by a well-mixed surface layer with obvious thermals
 Physical Limnology 339
a b

TEMPERATURE (“C) dT/dz (Qm)

FIG. 13. Temperature microstructure data from a penetrative convective period,

collected in the Wellington Reservoir on 14 March 1982. (a) Temperature. (b) Temperature
gradient. (After Imberger (1985).)

falling through the full extent of the layer and impinging on a sharp
interface.
In summary, these examples show that it is difficult to define the
surface layer in terms of the temperature profile. It is better to define the
layer as that depth of water directly energized by the surface fluxes and
the mean shear of the surface layer. The data also show that the
temperature, and thus the density, is rarely absolutely uniform in the
surface layer, especially when there is strong surface heating or when the
layer is retreating (wind stress reducing with time). Lastly, the surface
fluxes not only introduce turbulence into the surface layer but also, via
leakage through the base of the surface layer, into the deeper waters
making the above definition of the surface layer somewhat ambiguous.
Some of this surface-introduced energy energizes the water column as a
340 Jorg Irnberger and John C . Patterson

whole, but the exact fraction has not been documented in field
measurements.
The overall dynamics of the surface layer is determined by the
magnitude of the surface Wedderburn number (Imberger and Hamblin,
1982) given by (2.5). For W>>1, tilting of the isotherms due to the
applied wind stress will be small and horizontal variations are negligible.
This corresponds to strong stratification, light winds, and slow deepening
of the mixed layer dominated by surface-introduced turbulence reaching
the base of the mixed layer and eroding the interface. For W ((1,
deepening is dominated by internal shear production and occurs on a
time scale much shorter than horizontal convection in the surface layer.
This leads to a sharp interface downwind and a broad upwelling at the
upwind end of the lake (Monismith, 1986). In the initial classification (see
Spigel et al. (1986)), this regime was misinterpreted as having a
horizontal interface due to the envisaged rapid deepening. This question
is addressed in the next section. For intermediate values of the
-
Wedderburn number, W 1, upwelling and horizontal mixing become
important (Spigel and Imberger, 1980). This was confirmed by the model
of Imberger and Monismith (1986), the laboratory finding of Monismith
(1986) and by the field data of Imberger (1985) and Strub and Powell
(1986).
In the case where W > > l , the processes are essentially one-
dimensional, and as shown by Spigel (1980), the layer deepening can be
uncoupled from any internal seiching that may exist. There is a long
history of interest in such a one-dimensional surface layer. Munk and
Anderson (1948) discussed earlier observations and presented a theory
based on an eddy diffusion coefficient, dependent on stability, which
enabled the prediction of the surface layer behavior. The central task in
understanding the surface mixed layer is to quantify the rate at which the
kinetic energy of the turbulent velocity fluctuations is converted to
potential energy within and at the base of the surface layer as denser
water from beneath the layer is entrained and then mixed into the surface
layer. The turbulence available for this mixing may be generated at the
surface through pressure work, shear production, and wave breaking
(Kraus and Turner, 1967; Kraus, 1977; Cavaleri and Zecchetto, 1987;
Muller and Garwood, 1988) or in the surface layer and the base by shear
production (Pollard et al., 1973; Imberger, 1985). The transport and
redistribution of the turbulent kinetic energy within the surface layer is
dominated by large scale or secondary motions if present. The most
 Physical Limnology 341

popular concept is of a coherent, well-ordered Langmuir circulation

(Langmuir, 1938; Scott et al., 1969; Leibovich and Radhakrishnan, 1977;
Pollard 1977; Leibovich, 1977a, 1977b, and 1980; Leibovich and Paolucci,
1980; Leibovich, 1983). From the examples of surface layer development
discussed above, the presence of such organized cellular motions are
apparently not ubiquitous (see also Muller and Garwood (1988)) and
only few direct observations exist (Thorpe and Hall, 1982; Weller et al.,
1985; Smith et al., 1987).
The turbulent kinetic energy budget in the surface layer has so far
remained elusive. Because direct observations of the vertical turbulent
fluxes of momentum, mass, or energy are still lacking, one must rely
completely on indirect evidence from measurement of dissipation of
turbulent kinetic energy. Such measurements can be used to validate
integral models (Imberger, 1985), scaling laws (Kitaigorodskii et al.,
1983; Dillon et al., 1981; Shay and Gregg, 1986; Brubaker, 1987; Gregg,
1987) and estimates of the Reynolds stress by equating dissipation to
production (Dillon et al., 1981). Detailed comparisons of this kind have
led to better agreement in the lake cases than those in the deep ocean,
suggesting that better estimation of surface fluxes, the smaller variance of
the turbulent fluxes (smaller waves), and the weaker influence of
horizontal advection all contribute to this better match. Further, it
suggests that the problem lies in capturing the full variability with the
measurements rather than with turbulent kinetic energy budget assump-
tions. An example of the mismatch encountered in the ocean is given in
Figure 14 (from Muller and Garwood (1988)), where the observed
dissipation rates are considerably larger than the estimated surface input
of energy.
A greater understanding of mixed layer energetics has come from
one-dimensional models that simulate vertical processes but neglect all
horizontal variations. These models are in two categories. First, a
number of attempts have been made to obtain solutions to the fun-
damental equations of motions using some type of closure assumption.
Munk and Anderson (1948) may be viewed as a zeroth order model
(Andre and Lecarrere, 1985), whereas the models by Mellor and Durbin
(1975), Mellor and Yamada (1982), Findikakis and Street (1982a, 1982b,
1983), Edinger and Buchak (1983), Franke et al. (1987), Shih et al.
(1987), and Murota et al. (1988) are higher order schemes. Excellent
summaries of the applicability of such closure schemes to mixed layer
problems is given in the review by Rodi (1987). The range of length
342 Jorg Imberger and John C. Patterson

10-7 10-6 10-5 10-4 10-3

o . o o ~ - *, , , , , , , ,
~ , , , , , ,,.[ , , , , ,,,, , , , , , ,,,, ,q
, , , ,,

0.10

0.20
n
a
5
0.30
BURST 19

0.40

0.10

0.20

0.30

0.40

0.50

me-’
FIG.14. Two examples of dissipation as a function of depth taken in the ocean and
compared with the similarity scaling for turbulence produced by both wind stress and by
convection. (From Muller and Garwood (1987); data originally from Gregg, unpublished.)

scales characterizing the surface layer dynamics and the unsteadiness of

the turbulence makes the application of closure schemes difficult to
justify. For instance, it is now well known that in a penetrative
convection surface layer, the heat flux in the lower half of the layer is
against the mean gradient, a process that requires the application of a
high order closure model.
Large eddy simulation schemes fall into this same class of models,
but are even more computer intensive. The method has been applied
successfully to a homogeneous wind-driven reservoir by Ivetic et af.
 Physical Limnology 343

(1986), but once again, in the presence of stratification, the diversity of

scales and the non-stationarity of the turbulence has prevented the use of
the method.
Second, integral formulations pioneered by Niiler and Kraus (1977)
continue to be developed (Atkinson and Harleman, 1983; Spigel et al.,
1986; Heathershaw and Martin, 1987). As these authors point out, the
simplicity of physical insight afforded by the integral approach is
adequate justification for pursuing it.
Shear production of turbulence at the base of the surface layer was
recognized by Pollard et al. (1973) as an important source of energy for
the deepening process. Niiler (1975), Zeman and Tennekes (1977),
Sherman et al. (1978), and others have included shear production,
together with surface stirring as originally advocated by Kraus and
Turner (1967), in their mixed-layer models. Observations by Price et al.
(1978) verified the importance of shear production for oceanic mixed-
layer deepening, while the measurements of Thorpe (1978) in Loch Ness
provided evidence of billowing accompanying strong velocity shear at the
base of the mixed layer. Spigel and Imberger (1980) introduced a
classification for mixed-layer deepening that predicted the conditions
under which shear production would become important in small to
medium-sized lakes. Their mixed-layer-deepening algorithm was based
on the model of Sherman et al. (1978) but did include a separate
calculation to simulate billowing. However, Spigel and Imberger (1980)
did not explicitly account for the effects of billowing in the turbulent
energy budget, and their parametrization of the turbulent energy budget
assumed (in common with Sherman et al.) that mixed-layer turbulent
kinetic energy was a fixed proportion of external energy input by wind
and surface cooling. The assumption of fixed proportionality implies that
mixed-layer turbulence adjusts rapidly to changing external inputs as was
shown above; this may not be valid for diurnal simulations in which
meteorological forcing is varying rapidly. In addition, in deepening by
convective overturn, Denton and Wood’s (1981) experiments have shown
that the assumption of fixed proportionality is valid only under certain
boundary conditions.
The role of billowing is not so straightforward. Consider two parallel
streams separated by a sharp interface across which a velocity jump AU
and density jump A p (with the heavier layer on the bottom) occur.
Theoretical considerations (cf. Drazin and Reid (1981), Nishida and
Yoshida (1984), and Lawrence et al. (1987)) show that the interface is
344 Jorg Zmberger and John C . Patterson

unstable, and in practice, Kelvin-Helmholtz billows are always observed

along such an interface (Thorpe, 1969). The time for billows to form,
grow, and then collapse into small-scale turbulence is relatively short, of
order Tb = 20 A U / g f (Thorpe, 1973; Chu and Baddour, 1984), where
g’ = A p g / p , (po is a reference density). Values of T, in lakes are of the
order of minutes or less (Spigel, 1978). The ultimate mixing that occurs
after the breakdown of billows is of more relevance here than the details
of billow formation, and we use “billowing” broadly to include this final
mixing. The net result of billowing is that the sharp interface is replaced
by a shear layer of thickness 6 across which density and velocity vary
continuously. Simple energy arguments (Sherman et al., 1978) show that
6 is proportional to A U 2 / g f ;on the basis of numerical and laboratory
experiments, Sherman et al. suggest that

S = 0.3 AU2Jg’. (4.1)

Such a configuration is stable, and if nothing occurs to increase A U or
decrease the thickness 6, no further billowing will occur. If AU grows
(possibly due to an increasing wind stress) or 6 decreases (possibly due to
mixed-layer deepening-see Figure 1la-V) , then billowing will occur
until a new stable configuration is achieved.
There is thus a fundamental distinction between mixing caused by
billowing and mixing caused by thermocline erosion. Billowing broadens
an interface more or less symmetrically about the point of maximum
velocity gradient and thereby reduces the density gradient between
layers. Billowing by itself does not cause thermocline erosion or produce
any net mixed-layer deepening. Thermocline erosion is a one-way
process: an upper turbulent layer grows thicker at the expense of a lower
nonturbulent, or less turbulent, layer by entrainment of quiescent fluid
into the upper mixed layer. There is no tendency for density gradients to
be weakened during thermocline erosion and in some cases they may be
sharpened. Sharpening of density gradients is associated with convective
penetration or other stirring processes.
In lakes, billowing accompanies mixed-layer deepening (Thorpe, 1978;
Imberger, 1985) yet both processes drain kinetic energy from the mean
flow shear AU to produce turbulent kinetic energy. In this sense both
processes compete for a given supply of mean flow kinetic energy
proportional to $poAU2 per unit volume. The processes complement
each other in that a fraction of the turbulent kinetic energy produced in
both is used in working against gravity to increase the potential energy of
 Physical Limnology 345

the water column; i.e., both processes cause mixing. Billowing utilizes
mean flow kinetic energy that would otherwise be available to produce
mixed-layer turbulence for entrainment, but at the same time weakens
the density gradient at the base of the mixed layer. Billowing reduces the
energy required for further mixed-layer deepening, since the energy
required to entrain heavier fluid is roughly proportional to the strength of
the density gradient across the base of the mixed layer. There thus arises
a rather complex and unsteady interaction whereby mixed-layer deepen-
ing sharpens a gradient, making it unstable to shear so that billowing
occurs. Billowing weakens the gradient, which is then more easily eroded
by further deepening, which leads to further billowing.
One of the principal goals of the model by Rayner (1981), and later
Spigel et al. (1986), was to demonstrate that the interaction between
billowing and deepening can be successfully modeled by accounting
simultaneously for the energetics of surface stirring, shear production,
billowing, and variability in the store of turbulent kinetic energy.
In these simple one-dimensional integral models, it is assumed that the
surface layer may be approximated in the integral sense by three distinct
zones (Niiler and Kraus, 1977; Spigel et al., 1986): a comparatively thin
constant stress surface layer of thickness y (see Figure 15) where
turbulent kinetic energy is produced and then exported to the fluid
below; a uniform central layer in which part of the surface energy is used
to mix the fluid and is lost to dissipation; and a thin front of thickness

ATMOSPHERK: TRANSFERS

p s /fWlHL r/”L p
Z-H 3 V

?
F
-
rJ SURFACE
DRIFT
ZONE
WELL MMED LAYER
S, (FULLY TURBULENT)

HYPOLIMNION

Definition sketch of the surface layer model. (After Spigel et al. (1986))
346 Jorg Imberger and John C. Patterson

6 at the base of the surface layer. This transition marks the pronounced
density jump described above as the diurnal thermocline. Here, the
remainder of the turbulent kinetic energy generated at the surface plus
that generated internally by shear and less that which is locally dissipated
or radiated downwards by internal waves (Linden, 1975; E and Hopfin-
ger, 1986; Imberger, 1988) is used to entrain underlying fluid into the
central layer above. Using the symbols defined in Figure 15, the
governing differential equations (Niiler and Kraus, 1977; Spigel et al.,
1986) may be written as

dS
-=
a-
- -(s‘w’), (4.3)
at az
ap - - - ( ap ’ w-
_ ’)+--, a aQ
at dz cpaz (4.4)
au
-=
a-
- -(u’w’),
at 3.2 (4.5)

where E = u’* + v ’ +~w r 2 ,Q is the short-wave radiation, and the simplest

form of the momentum equation (4.5) is used. Niiler and Kraus (1977))
de Szoeke (1980), Ivey and Patterson (1984)) Janowitz (1986), and others
have included the influence of longitudinal pressure gradient and
Coriolos force. The rationale for using such a simple balance is discussed
in Spigel and Imberger (1980)) Kranenburg (1984)) Monismith (1985))
and Strub and Powell (1986). Monismith (1985) shows that (4.5) can be
used until the base of the mixed layer develops a baroclinic tilt due to the
second mode seiche; the time to the first deceleration is TJ4) where is
the period of the second mode seiche (the seiche associated with the
upper surface layer).
The boundary conditions at the free surface assume continuity of fluxes
across the interface:
-
Momentum: - -u’w ’(H) = u:, (4.7)
Heat: pOCpB’w’(H)
-
= QL+ HL+ H s , (4.8)
Salt: - = WS,,-
-s’w’(H) -
(4.9)
Mass: +
p ’ w ’ ( H )= p O [ - a e ’ w ’ ( H ) @ ’ w ’ ( H ) ] , (4.10)
 Physical Limnology 347

where W is the water loss at the surface, QL is the net long-wave

radiation, HL is the latent heat loss, Hs is the sensible heat loss, (Y is the
thermal coefficient of expansion, and B is the salt coefficient of volume.
Spigel et al. (1986) assumed that the turbulent heat and momentum
fluxes at z = 5 - 612 (the very bottom of the base of the surface layer)
are zero, so that
- - - -
elw' = s ' w ' = p'w' = u ' w t = 0. (4.11)

The leakage or transfer of turbulent kinetic energy through the base may
be written

(4.12)

As seen from Figures 11, 12, and 13, this is usually nonzero. Imberger
(1988) postulated that the mechanism for turbulent energy transport was
by internal wave propagation, away from the base of the surface layer
along rays. Linden (1975) estimated from laboratory experiments that up
to 50% of the turbulent kinetic energy budget arriving at the base of the
mixed layer seeps through the base. By constrast, E and Hopfinger
(1986) suggested that while internal waves were generated, energy
radiation generally did not affect the entrainment rate. Scaling suggests
that the leakage A L be parameterized by

AL = CLA3N3 (4.13)
where CL is a coefficient, A is the amplitude of the internal waves
generated, N is the buoyancy frequency below the base, and h is the
depth of the mixed layer. At present, there are no conclusive data that
allows the determination of the correctness of (4.13) or the magnitude of
CL,and for this reason, Spigel et al. (1986) assumed CL= 0. It is worth
remarking that since their model gave very good comparisons with
observed field data, the leakage term must be proportional to the flux of
turbulent kinetic energy arriving at the interface, which is seen by noting
-
that A E:n/N.
Spigel et al. (1986) followed Niiler and Kraus (1977) and integrated
(4.2) to (4.6) from the bottom to the surface of the lake using the
boundary conditions given by (4.7) to (4.12). Further, closure was
achieved by introducing a set of efficiencies for each of the source and
sink terms. Shear production in the base layer minus the dissipation was
348 Jorg Zmberger and John C . Patterson

approximated by
Uzdh 1 d U2dh 1 d
--+--(U?h)- cbh = Cs 2-+--(Uzh)], (4.14)
2 dt 12dt 2 dt 12dt
where Eb is the average dissipation in the base layer and Cs is the
efficiency of production of turbulent kinetic energy. Experimental evi-
dence that allows the evaluation of Cs is sparse. Sherman et al. (1978)
reanalyzed the experiments of Kato and Phillips (1969) and Kantha et al.
(1977) to arrive at a value of between 0.2 and 0.5. Imberger (1985)
presented field data giving a value of 0.24 and Spigel et al. (1986)
obtained best simulation results with a value of 0.2.
The turbulence introduced at the surface constant stress layer may be
parameterized (Kraus and Turner, 1967; Niiler and Kraus, 1977) as

(4.15)

where c is the velocity scale in the surface drift layer (see Figure 15), CN
is the efficiency of energy production at the water surface, and u* is the
water shear velocity.
The source of buoyancy flux due to surface heating may be written
(Deardorff, 1970; Zeman and Tennekes, 1977; Rayner, 1981) in the
form

w: = g ( h + !){
2
[
Ly Q ( H ) - Q ( 5 -
POCP
): ] + ( y e " ( H ) }
+ pwss- -
2*g
POCP
1"
5-m
Qdz. (4.16)

The dissipation in the mixed layer may be written (Spigel et al., 1986)

E,h = $ESn, (4.17)

where E, is the average dissipation in the mixed layer. This form was
introduced by Mahrt and Lenschow (1976); similar forms were suggested
by Garwood (1977) and Zeman and Tennekes (1977), as well as by the
results of Willis and Deardorff (1974). The alternative parameterization
used by Niiler and Kraus (1977), Sherman et al. (1978), and others
consists of introducing a set of efficiencies to reduce each energy source
by a certain fraction; an approach that does not allow for the storage of
 Physical Limnology 349

turbulent kinetic energy. Rayner (1981) proposed a necessary final

closure assumption which states that a fraction of the turbulent kinetic
energy iCFEF3 in the mixed layer together with the available shear
production (4.14) is utilized at the base to spin up the turbulence and to
energize the buoyancy flux (neglecting any leakage). This may be
expressed as
-E, +> U : dh
- + - - (IUd: 6 ) ]
F'
2
312
2 [ dt 6dt
dh 6d I d
Esdh
2 dt 2po
+
= -- - Aph -- -- (Aph)
[ dt 2dt
+ --
12dt
(ApS')]. (4.18)

Given these assumptions and carrying out the integration yields (Spigel et
al., 1986) the necessary set of equations for the unknowns U s , h, 0, , S, ,
6, and E,:
dEs
h-=-(C,+
dt
CE)E,'"+ W: + C&:, (4.19)

(4.21)

(4.22)

and (4.1) and (4.18).

The evaluation (Spigel et al., 1986) of the remaining three coefficients
CN= 1.33, CF=0.25, and CE = 1.15 follows directly from a careful
analysis of all available data (Kato and Phillips, 1969; Tennekes and
Lumley, 1972; Wu, 1973; Deardorff, 1974; Kamail et al., 1976; Mahrt
and Lenschow, 1976; Stull, 1976; Willis and Deardorff, 1974).
Recently, there has been a renewed interest in the entrainment
problem described by (4.1) and (4.18)-(4.22), and there is greater
awareness that there are two distinct sources of energy for entrainment at
the base of the surface layer as expressed by (4.18),
dh I d
'[
fCFEfi2 and IC U : - + - - ( U Z S )
'
dt 6dt 1
Atkinson (1988) reviewed the literature for shear driven experiments
(see also Christodoulou (1986)) and concluded that the data support the
350 Jorg Zmberger and John C . Patterson

empirical entrainment law

1 dh
--= a - f3 Ri P;lR
(4.23)
y + Ri
9
U dt
where a,f3, and y are constants, P, = U H / K , K is the coefficient of
species diffusion, and Ri = g'h/U2. Given that shear production domin-
ated the deepening process, (4.18) reduces to (assuming 6 to be small)

(4.24)

Comparison with (4.1) indicates that the depth of the mixed layer scales
the same as 6, which means that for a particular Us the depth is fixed and
cannot continue to deepen, contrary to (4.23). This was the basis of the
original Pollard et al. (1973) model, and may be explained by noting that
once a particular surface layer has a certain shear across its base, both h
and 6 will adjust until (4.1) and (4.24) are satisfied, after which time
there will be no further adjustment (Sherman et al., 1978; Lawrence et
al., 1987; Thorpe, 1987). The experiments of Narimousa et al. (1986) and
Narimousa and Fernando (1987) rely on a surface energy source u', that
will continually sharpen the interface and reduce 6, and the entrainment
law (4.23) therefore depends on the type of experiment and must be used
with caution.
This is best seen by substituting (4.22) into (4.24) and noting that g'h is
constant for no net heat flux, which then yields
1 dh Cy"
where Ri = g'hlu:. (4.25a)
u, dt -Ri'"'

Simple grid experiments contain no shear so that (4.18) reduces to the

relationship
1 dh - CF (4.25b)
E,ln dt 1 + g'hIE, '
Hannoun et al. (1988) have shown that the energy flux that arrives at the
interface is the same with and without the interface being present, and is
given by u3, where u is the root mean square velocity, which was also
observed to decay inversely proportional to the distance from the grid. In
grid experiments, we should therefore expect C F , the fraction of the
mean kinetic energy flux that arrives at the interface, to be a function of
both the depth of the surface layer and the turbulent intensities. Ignoring
 Physical Limnology 351

the spin-up term in (4.25) and defining ii as the mean (over the depth of
the mixed layer) root mean square velocity allows (4.25) to be rewritten
ldh
--== CF
(4.26)
ii dt Ri’
where

(4.27)

Nokes (1988) has done a thorough review of data from all available
grid experiments and concluded that the rate of non-dimensional
deepening ranges in proportionality from Ri-’ to Ri-’.75. In this context,
the results from Kranenburg (1984) and Murota and Michioku (1986b)
give convincing evidence that the non-dimensional deepening is propor-
tional to Ri-l. In all recent work, investigators have defined the
Richardson number Ri=g’l/u*, where 1 is the integral scale of the
turbulence at the position of the interface (but in the absence of the
interface) and u is the root mean square velocity of the turbulence
arriving at the position of the interface. Unfortunately, there are no data
available on the relationship between the depth of the mixed layer and
the integral scale for surface layers so we cannot take this comparison any
further. It would be constructive to carry out an integral analysis of the
grid configuration to determine the relationship between ii and u, and to
establish if (4.26) would lead to the relationship suggested by Nokes
(1988):
l d h - 0.15
u dt Ri-’.’ * (4.28)

This is now possible as all the necessary spatial variations can be derived
from Hopfinger and Toly (1976), Hannoun and List (1988), and Hannoun
et al. (1988). Last, it is important to reconcile, as discussed by Nokes
(1988), the difference in the measurement of the entrainment rate dhldt:
indirectly from a measure of the change of buoyancy in the surface layer
and directly from the propagation rate of the interface.
Murota and Michioku (1986b) also carried out an experiment where
they used grid stirring together with a convective energy source and
successfully compared their results with the model put forward by
Sherman et al. (1978) and embodied in (4.19), although their value for
C , was 2.9, not 1.33.
 b
0.0 w
VI
h,
1.0 -
2.0 -

7.0 1 ,/ /
21.50 21.75 22.00 22.25 22.50 22.75 23.00 23.25 23.50 23.75 24.00 21.50 21.75 22.00 22.25 22.50 22.75 23.00 23.25 23.50 23.75 24.00
TEMPERATURE ("C) TEMPERATURE ('C)

C d
0.0 0.0

1.o 1.o

2.0 2.0

x
E
3.0
-E 3.0

4.0 4.0
+ +
w
0 5.0 x 5.0

6.0 6.0

7.0 7.0

8.0
21.50 21.75 22.00 22.25 22.50 22.75 23.00 23.25 23.50 23.75
-
24.00
830
..
21.50 21.75 22.00 22.25 22.50 22.75 23.00 23.25 23.50 23.75 24.00
TEMPERATURE ('C) TEMPEAATURE (%)

FIG.16. Comparison of measured (solid line) and predicted (dashed line) temperature profiles. The profiles b, c, and
d have been corrected for effects of advection. (After Spigel el ul. (1986).)
 Physical Limnology 353

The model described by (4.11) and (4.18)-(4.22) was used by Spigel et

al. (1986) to simulate the diurnal cycle documented in Imberger (1985)
for a surface layer in the Wellington Reservoir in March 1982. The
24-hour period included strong heating in the early morning, severe wind
in the afternoon when the Wedderburn number decreased to below 0.02,
and intense penetrative convection from midnight to early morning.
Comparison from these simulations is shown in Figure 16. Given that
none of the coefficients C s , CN, CF, or CE were adjusted for the data
and also given that a period of heating, stirring, shear production, and
penetrative convection all contributed to the surface layer deepening, the
comparisons are remarkably good.
Recently it was pointed out (Dickey and Simpson, 1983; Stefan et al.,
1983) that the exact distribution of solar radiation q ( z ) is most important
to the dynamical response of the surface layer. From (4.16), the
departure from the linear absorption curve determines the mechanical
buoyancy flux, and it is this departure that contributes to the direct
influence on the dynamical response.
In summary, we have shown that the identification of the energy
sources in the surface layer has allowed, over the last few years, the
construction of successful simulation algorithms. The success of these
models, however, depends on an accurate description of the surface
fluxes, a realistic model for the momentum in the surface layer, and the
removal of changes due to horizontal advection. Last, the model allows
intercomparisons with the full range of laboratory experiments such as
natural convection, shear-generated turbulence, and grid-generated tur-
bulence, but as yet this comparison does not seem to have been carried
out. The leakage of turbulent kinetic energy through the base of the
surface layer remains to be parameterized.

V. Upwelling

When the surface wind stress increases, the isopycnals surface at the
upwind end and deepen at the downwind end (Wedderburn, 1912;
Keulegan and Brame, 1960; Blanton, 1973; Stefan and Ford, 1975; Spigel
and Imberger, 1980; Monismith, 1986; and many others). This phenome-
non is called upwelling.
A demonstration of upwelling is given by the longitudinal transects
taken in the Wellington Reservoir (Figure 17a) in August 1988. The
354 Jorg Imberger and John C . Patterson

isopycnals prior to the start of a wind event are shown in Figure 17d; the
wind event started at about 1000hours and lasted for about 6hours
(Figure 17b), with a direction of 320" (Figure 17c), the direction of the
transect. The isopycnals after 4 hours are shown in Figure 17e, indicating
a high degree of disturbance. The hypolimnion was also disturbed but
retained a degree of horizontality. The upwind surface layer showed
strong upwelling with a well-defined horizontal gradient supported by
cold water brought up from the metalimnion.
The idealized density profile assumed by Heaps and Ramsbottom
(1966) and later used by Spigel and Imberger (1980) in their lake
classification scheme consisted of only two layers. This two-layer ap-
proximation was used by Thompson and Imberger (1980) and Imberger
and Hamblin (1982) to formulate the Wedderburn number (2.5) as the
nondimensional parameter determining the response of the surface layer
to an impulsive wind stress.
The initial value problem for the two-layer situation was solved by
Spigel(l980). For the case with rotation, the reader is referred to Csanady
(1982), Heaps (1984), Kielman and Simons (1984), and Horn et al.
(1986); the influence of rotation will not be discussed here. From Spigel
(1980), in the first stages of development of the surface layer following
the startup of the wind stress, the surface layer moves according to
(4.22). This surface layer velocity was shown by Imberger (1985) to be
applicable until the interface tilt reaches the center of the lake and
induces a baroclinic pressure gradient, which retards the motion. The
period of the surface layer seiche was given by ( h , < H)

T,= (5.1)

where L is the length of the lake, Ap is the density difference across the

FIG.17. Upwelling as recorded along the central channel of the Wellington Reservoir in
response to a strong northwesterly wind. Data collected on 24 August 1988. (a) Location
map showing plane of projection and measuring stations. (b) Wind speed as recorded in the
central valley of the Wellington at 1.4m above the water surface. (c) Wind direction
recorded as above. (d) Isopycnals at 0930, before wind had commenced. (e) Isopycnals at
1330 extending from the Gervase River to the main basin. (f) Example of a temperature
microstructure record, collected during the strong upwelling event shown in Figure 17e. (g)
Corresponding temperature gradient. (h) Corresponding dissipation as computed from the
Batchelor spectra fitting technique (Bar graph) and the Wigner-Ville technique (sticks).
 23 000

22 000

21 000

20 000

-p 19000
GI
c
Y
LINE OF PROJECTDN
z l8OW AND WIND DIRECTION

:
8 17000
WELLINGTON RESERVOIR
2
s 16 000

15000

14 000

13 000

12 000
14 I 15000 16000 17000 18000 19000 20000 21000 22000 23000
LOCAL CO-ORDINATES Iml

-ur
-
E
0
6

L
fn
4
0
f 2
a
360

0
E
'f 270
k
:a:
0
g 180
5

90 u . .
I . ,. . . . .
2 '.O 237.5 238.0 238.5 239.0
DAY
356 Jorg Imberger and John C. Patterson

0.0
d DENSllY (kg/m3

2.0 --

4.0 --

6.0 --
E
E 8.0 --

12.0
’O’O 1
3000.0 4000.0 5000.0
PROJECTED DISTANCE (m)

0.0

2.0

4.0

6.0
--E
Eul 8.0
0

10.0
\
12.0 /*,*
‘+AO
o \ ,~
14.0
205
A
C60
A
C50
L
a2 /
A
’& A
2000 0 3000 0 4000 0 5000 0
PROJECTED DISTANCE (m)

FIG. 17. (conrd.)

Physical Lirnnology 357
2
358 Jorg Imberger and John C. Patterson

two-layer idealized density structure, h , is the depth of the surface layer,

h2 is the depth and p the density of the lower layer, and H is the depth of
the lake. Throughout it was assumed that the theory was applicable for
times up to T,/4 when the base of the surface layer begins to tilt. The
theory developed by Spigel and Imberger (1980), which provided the
background for the Wedderburn number concept, relied completely on
the two-layer model. However, it is apparent from the work of
Monismith (1986) that this model does not explain the upwelling that
already occurs at Wedderburn numbers considerably larger than one.
Monismith (1989, using the modal decomposition technique de-
veloped by Lighthill (1969) (see also Gill and Clarke (1974) and Csanady
(1972)), solved the n-layer case for wind stress varying arbitrarily in time.
He applied this general theory in three-layer form to the Wellington field
data reported by Imberger (1985). From the analysis, he showed that the
shear across the surface layer base AU,, was given by

which is the generalization of (4.22) to the case of an arbitrary number of

layers and where tl, is the displacement of the surface layer base
(between layer 1 and 2), A p 1 2is the density difference between layer 1
and 2, and h , is the thickness of layer 1 (the surface layer). The first term
on the right side of (5.2) is the pressure gradient induced by the slope of
the interface Cl2.
The importance of Monismith's (1985) analysis is that when hl and
A p l z are small (that is, W is small), mode 2 is excited, leading to a large
shear across the base of the surface layer with an associated strong tilt of
the interface f 1 2 ; the second interface 523 remains essentially horizontal
(Figure Ma). Conversely, where Ap,, is of the same order as Ap23 and
the top and bottom layers are of comparable size or larger than the
middle layer, the response is strongly influenced by the first mode and the
second mode with a comparable tilt in cl2 and f;23 and the introduction of
a large velocity in the bottom layer. By taking a range of surface layer
depths as measured in the Wellington field experiment, Monismith (1985)
obtained excellent comparisons between his theory and experimental
results for the thermocline tilt and induced shear; the three-layer
stratification example is a satisfactory model of the majority of lake
stratification patterns.
 Physical Limnology 359

A-
b

FIG. 18. Schematic of the modal response for a three layer

. system.
. (a) Second mode
response: W small, L , large. (b) First mode response: W small, L, small.

The results from this model have two profound implications:

(a) A mode 2 response (Figure 18a) concentrates the shear, and thus
induced turbulence, in the upper surface layer. The strong tilt induced at
the base of the surface layer introduces active upwelling at even quite
moderate Wedderburn numbers (Monismith, 1986). Further, the lower
layer remains essentially stationary with c23
remaining horizontal.
(b) A mode 1 response (Figure 18b) introduces motion throughout the
lake, with tilting isopycnals in both the metalimnion and the hypolim-
nion. This motion leads to active turbulence throughout the water
column.
It is important to categorize a lake’s response as either mode 1 or 2. It
is clear from the theory underlying the Wedderburn number W that a
mode 2 response is induced when W becomes small (less than 10); the
360 Jorg Imberger and John C. Patterson

tilting of the base of the surface layer becomes stronger with decreasing
W (Monismith, 1986). The three-layer model has been used successfully
to explain why upwelling is induced for moderately small Wedderburn
numbers. With increasing wind stress relative to the buoyancy influence,
W decreases and the amplitude of the mode 1 response increases, causing
the metalimnion (layer 2) to tilt, which induces upwelling not only from
the metalimnion but also from the hypolimnion. However, it is not the
Wedderburn number that determines the mode 1 response, but the Lake
number LN (2.12); LN is the mode 1 counterpart to W in mode 2. This
may be illustrated by examining in more detail the experiments of
Monismith (1986). The stratification used in these experiments is well
approximated by a three-layer stratification with a constant density in
layers 1 and 3 and a linear transition between the two. For a rectangular
basin of length L, width B and total density difference between the top
and bottom layers of Ap, the Lake number takes the form

where

and h , , hz , and h3 have the same definition as above.

The results from Monismith’s (1986) Experiment 8 (Figure 19a-d)
show that a mode 2 response was induced with the base of the
metalimnion remaining essentially horizontal, but with the surface of the
metalimnion developing distinct upwelling very similar to that in Figure
17c for the Wellington Reservoir field data. The value of W was 3.9 and
that of LN was 6.4, indicating that the base of the surface layer was rela-
tively susceptible to the mode 2 response, but that the overall stratifica-
tion contained a large amount of potential energy. By contrast, the values
for Experiment 22 were W = 0.4 and LN = 0.15; the value of W had
decreased by a factor of 10 but LN had decreased by as much as a factor
of 50, indicating that in this experiment both the base of the surface layer
and stratification were relatively weak and could not resist the overturn-
ing moment supplied at the surface by the belt drive (Figure 19e-h).
These experiments must thus be reinterpreted as both W and LN were
alIowed to vary. The low Lake number value in Experiment 22 explains
why the boundaries of the metalimnion remained almost parallel at even
 361

0.0-

5.0 -
I O ' O P

-
-b
I 10.0-

I
I-
n.
w
n

15.0-
I , I I I ~o,ol-~---ll,oo~

20.0- I I I I I
50 100 150 200 250 300 50 100 150 200 250 300
DISTANCE (cm) DISTANCE (cm)

0.0 0.0

5.0- 5.0-

-E
x

--5
F
a
w
0
i
w
O

15.0- 15.0-

20.0 I I I I I T 20.0 I I I I I

FIG. 19. Contours of density field measured by Monismith (1986) in a laboratory tank.
The stress was introduced at the bottom with a belt moving from left to right.
(a) Experiment 8: W = 3.9, L , = 6.4, t = 0.32 T, .
(b) -
. , Experiment 8: W = 3.9, L , = 6.4, I = 0.64 T, .
(c) Experiment 8: W = 3.9, L , = 6.4, = 0.96T,.
(d) Experiment 8: W = 3.9, L , = 6.4, = 8.0 T , .
(e) Experiment 22: W = 0.4, L , = 0.15, = 0.35 T, .
( f ) Experiment 22: W = 0.4, L , = 0.15, = 0.53 T, .
(9) Experiment 22: W = 0.4, L , = 0.15, = 0.70 TI.
(h) Experiment 22: W = 0.4, L , = 0.15, = 1.05 TI.
362 Jorg Imberger and John C. Patterson
6 f
0.0
1

50 100 150 200 250 50 100 150 200 250

DISTANCE (crn) DISTANCE (cm)

50 100 150 200 250 50 roo I50 200 250

DISTANCE (cm) DISTANCE (cm)

FIG.19. (conid.)

extreme deflections (Figure 19f). Indeed, the whole metalimnion sur-

faced, leading to a direct mixing of water from the bottom and surface
layers. Experiments in which LN is kept large while W is reduced have
not been carried out, but field data from Imberger (1985), in the initial
phases of the wind build-up, had a Wedderburn number of 0.02 and a
Lake number of 12. The isotherms (isopycnals) below the base showed
essentially no tilt throughout the wind period. The Lake number
decreased to a value of 1.4 at the time of peak wind speed, and the data
indicated some tilting of the isotherms and some mixing below the
 Physical Limnology 363

surface layer. By contrast, the Lake number for the data shown in Figure
17 was considerably smaller (0.24), explaining the greater mixing
observed at depth (see Section X ) .
W and LN are separate indicators, and for the case where W is small
but LN is large, only the surface layers respond to wind stress. Where W
and LN are small, the lake as a whole responds and we may expect that
vertical mixing will greatly increase throughout the lake.
The response in the case of continuous stratification (Monismith, 1986)
can now be interpreted in terms of these two nondimensional numbers.
For a linear stratification that reaches the surface, the mixed layer depth
is zero, as is the density jump across the base of the surface layer
(continuous stratification may be viewed as the limit where both
approach zero). This would mean that W is zero but LN is finite. The
expected response is a strong tilt of the surface isopycnals with active
upwelling, but at depth, the isopycnals remain horizontal. This case was
recently solved by Monismith (1987), using the same decomposition
technique mentioned above and assuming that the horizontal velocity
profile u(x, z, t) and the force field f ( x , z , t) could be decomposed as
follows:

n=l

where &(z) are solutions of the long wave form of the internal
eigenvalue problem (Gill, 1966) such that

and u,, fn, and 5, are the amplitude functions. Comparison with
experimental results from Monismith (1986) indicate that the fundamen-
tal mode was predicted well, but higher orders, which depend strongly on
the exact nature of the assumed stress distribution within the water
column were not as well predicted, although qualitatively, the agreement
was good.
So far the discussion has concentrated completely on the response of
times up to the quarter internal wave period for either mode 1 or 2.
Beyond this time, the internal displacement degenerates into seiches ( W
large, LN large) or damped motions (W and LN small). The damping of
the oscillations is caused by internal and boundary turbulence and we
364 Jorg Imberger and John C . Patterson

shall postpone the discussion of these processes to Section X, where

turbulence is reviewed.
For times longer than the damping time, it is possible to estimate the
deepening rate due to entrainment at the surface layer base for the case
where LN is large and W is small. Imberger and Monismith (1986) used
the results from Keulegan and Brame (1960), Kranenburg (1985), and
Monismith (1986) to formulate a model of a steady upwelling resulting
from a mode 2 response. Under such conditions, the experimental data
and the initial value problems discussed above lead to the following
conclusions:

(a) The interface tilts and opens at the upwind end over a period of
time approximately equal to the internal seiche period, (given
by (5.1)). The data suggest that the isopycnals are almost
horizontal at the bottom of the interface but slope considerably
upwards at the top of the interface, establishing a weak horizontal
density gradient in the mixed layer by a combination of upwelling
and horizontal variation in turbulent entrainment. The internal
circulation within the surface layer causes a general upwelling and
divergence of the flow at the upwind end, leading to a diffuse
interface. At the downwind end, there is downwelling, causing a
convergence at the interface, leading to an extremely sharp,
well-defined entraining interface.
(b) The velocity profile in the surface layer is such that the velocity at
the stress surface is approximately 20u, downwind (Kranenburg,
1985; Monismith, 1986) and about 224, downwind at the interface.
The flow at the interface was observed by Monismith (1986) to be
jet-like, concentrated a very short distance above the interface and
contained by a weak density gradient there. This differs from the
assumption made by Spigel and Imberger (1980) that the recircula-
tion flow in the mixed layer could be neglected for W > 1.
(c) The upwelling region is confined to the upwind end of the basin.
The density in the surface layer increases with time; for T > T,,
where T, is the characteristic longitudinal mixing time, the density
profile in the epilimnion varies linearly with distance along the
length of the surface layer.
(d) The net entrainment rule which best fits the experimental data
from Keulegan and Brame (1960), Kranenburg (1985), and Moni-
 Physical Limnology 365

smith (1986) may be written

dh 1
-= C l u e Ri-', (5.7)
dt
where C1 is a constant between 0.07 (Kranenburg, 1985; Moni-
smith, 1986) and 0.23 (Wu, 1973).
These conclusions permit the construction of a simple model of
upwelling and mixed layer deepening for reservoirs where W 1 and -
L N>> 1 (Imberger and Monismith, 1986, Figure 20).
For this case, as the stress is applied to the free surface, the interface
tilt is set up by the mode 2 response so that the upper isopycnals take on
an angle given by the dynamic balance developed by Wu (1973), and the
boundary layer thickness is proportional to the displacement of the upper
isopycnals

where 6 here is shown in Figure 20, gh = ( p , - p o ) g / p o , p1 - po is the

initial value of the density difference at the base of the surface layer, hl is
the initial surface layer depth, and 5 is the mean interfacial deflection.

STREAMLINESOF UPWELLINGFLOW
I

u UPWELLING REGION -
SHEARSTRESS
I

I\ \ \...
.....I.\ .................. .).
............................
......\I ....
.............
.I... ..-. ............
A,... ........
4 4

I ISOPYCNALS

I PZ
I
I I
x-0 XIL

FIG.20. Definition sketch for model of mixed-layer deepening due to upwelling. (After
Imberger and Monismith (1986).)
366 Jorg Imberger and John C. Patterson

The model assumes that the return flow of 2u, drags interfacial fluid with
it from the downwind end to the upwelling region. It is assumed that this
fluid, which can be visualized as being planed off the interface, enters the
upwelling region and then is distributed longitudinally by shear flow
dispersion. A simple mass balance would indicate that

where pa is the average density in the surface layer, gi = g(pl - pa)/po,

and h is the mixed layer depth at any time. It follows that

(5.10)

If 6 satisfies (5.8) then (5.10) produces a result consistent with (5.7).

Thus, the Ri-’ entrainment law can be arrived at without invoking
interfacial shear entrainment or entrainment energized by turbulence
imported from the surface; it is instead the consequence of upwelling.
The buoyancy flux (given by (5.9)) introduced by upwelling into the
main part of the surface layer at the upwind end is mixed downwind by
shear flow dispersion. Thus, the perturbation buoyancy field in the
surface layer g&(x, t ) = [p(x, t ) - po]g/posatisfies the diffusion equation
(Fischer ef al., 1979)
--
ag:,
- Ex- a2g:, (5.11)
at ax2 ’
where
E~ = Czhu, (5.12)
is the effective diffusion coefficient. The coefficient C2 depends on the
shape of the velocity profile and on the distribution and magnitude of the
vertical diffusivity E,. For the log profile, characteristic of two-
dimensional open channel flow, Elder (1959) derived a value of 6,
However, C2 will remain unspecified for the moment. The width I of the
end region is approximately hl and can be neglected (Cormack et af.,
1974).
A simple quantitative model of how upwelling leads to mixed-layer
deepening can be formulated by assuming that the mixed layer is initially
homogeneous and by modeling the effect of upwelling with an impul-
sively started, and later constant, buoyancy flux of strength B at x = 0.
Thus, the temporal and spatial variations of g& can be found by solving
 Physical Limnology 367
(5.11) subject to the following boundary and initial conditions.

ag:,
- B
(0, t ) = - H ( t ) , (5.13)
ax EX

ag:,
- (L, t) = 0, (5.14)
ax
g:,(x, 0) = 0. (5.15)
The required solution can be found in Carslaw and Jaeger (1978) as
m

gk(x, r ) = 2/3(t/&,)ln 2 [ierfc{(2n~+ L - ~ ) ( 2 ( ~ ~ t ) l ~ ) - ’ }

n=O

+ ierfc{(2nl+ x ) ( 2 ( ~ ~ t ) ’ ” ) - ’ } ] , (5.16)
where ierfc(x) = {exp(-x’)/dn} - x erfc(x) and erfc(x) is the com-
plementary error function. Equation (5.16) is plotted in Figure 21 in
terms of the quantity r = &(x, t ) ~ ~ ’ * / 3 - ’ t - ~for
’ ’ different values of

0
- ,
‘1L.10

L\\
0.2 a4 0.6 0.8 1

FIG.21. Curves of the solution (5.16) for I‘=g;(x, t ) ~ i ~ f i - ’ t - as

” ~functions of x at
different values of qL= L / ( E , ~ )(After
’ ~ . Imberger and Monismith (1986).)
368 Jorg Imberger and John C. Patterson

qL = L/(&,t)lR.When comparing experimental data with this solution, an

allowance must be made for the fact that horizontal density gradients are
created during the initial period of set-up by upwelling and by shear-
driven entrainment, i.e. gb(x, 0) =f(x). This can be done by adjusting
the time origin either through addition or subtraction of an offset.
Ideally, this offset should be much less than the characteristic timescale of
the dispersion process
L2
(5.17)
To==E,-
In addition, since /3 depends on h, it is not really constant throughout the
experiment. The solution will only be valid for times somewhat less than
hlRi
T,=-, (5.18)
ClU*

the time required for the mixed-layer depth to double.

From the solution plotted in Figure 21, it can be seen that the effect of
the upwelling flux is not felt at x = L until t -0.11 TO at which time
&(L) = O.O4gb(O). The total density difference in the mixed layer
reaches a maximum at t = 0.25 To, remains constant until t = TO and then
begins to drop as the mixed layer “heats up”. The solution for g,f,,(O, t)
can be shown to be
112 -112 112
gb(0, t ) =2Bt Ex Jc , (5.19)
for times less than 0.44 To. Finally, g,!,,(O, t ) will equal to 81, when

- 4Id
t = Tf - (“,W.($)
c: . (5.20)

If t > 0.44 TO, the factor appearing in (5.20) will not be n/4.
Figure 22 shows the data from three experiments reported in Keulegan
and Brame (1960), from two experiments reported in Kranenburg (1985),
and from Experiment 8 of Monismith (1986). In reducing these data,
Imberger and Monismith (1986) carried out the following operations:
(a) C,was first chosen to be 0.07 and, for all except Experiment 8, C2
was chosen to be 10.
(b) The resulting “raw” curves were plotted.
(c) So as to best fit the data to the theoretical curves, the dimension-
less time was offset by an amount A x and C1 was altered.
 Physical Limnology 369

r
-
'1L
10-

10:
102 10-1 1 I

4%)
FIG.22. Plot of values of (I'/qL)calculated from densities measured with correction for
the initial conditions and adjusted for a best-fit of data at two values of 5. The solid lines
are the theoretical solutions while the symbols represent:
Monismith (1986), W= 3. 9, At'=0.060, C,=O.O7, C,=6;
A Keulegan and Brame (1960), W = 3.0, AT' = 0.015, C , -0.11, C , = 10;
0 Keulegan and Brame (1960), W = 2.4, AT'= 0.025, C , = 0.11, C , = 10;
0 Keulegan and Brame (1960), W = 1.5, AT' = 0.030, C, = 0.05, C , = 10;
+ Kranenburg (1985), W = 3.7, AT' = -0.015, C , = 0.11, C , = 10;
x Kranenburg (1985), W = 2.3, AT'= +0.010, C , = 0.11, C , = 10.
(After Imberger and Monismith (1986).)

The last step was interactive and performed first for the data from the c
position, say el, c
closest to = 0. Once a satisfactory fit was obtained for
el, the data at a second position, c2
> 0.5, was reduced using the final
values of C1 and AT. From trial and error, it became apparent that for
each experiment analyzed, a best set of values of AT, C1, and C2 existed
that appeared to minimize the differences between theory and measure-
370 Jorg Imberger and John C. Patterson

ments for both values of t;. In all cases, reasonable values of all three
parameters were obtained: the time offsets were always less than 0.04 To,
and the entrainment coefficient C, was between 0.05 and 0.133, well
within the range reported in the literature and cited above. The
difference between values of C2 chosen for the belt-driven flow and the
wind-driven flow may reflect differences in the structure of those flows,
especially in the way the shear stress and surface velocity vary with x .
Recently Imberger and Spigel (1987) have carried out measurements in
Lake Rotognaio in New Zealand during conditions of strong heating,
weak winds, and very turbid water (active algal bloom reduced the light
extinction depth to 0.5 m). The general upwelling circulation remained,
but a strong stratification persisted within the whole surface layer,
inhibiting turbulence and eliminating the entrainment surface. The
surface buoyancy flux outpaced the mixing caused by wind-induced
circulation. This type of upwelling is characterized by W being very small
but L N very large, and the circulation appeared as that described by
Monismith (1986) for the continuously stratified case. The major
difference observed was that with the buoyancy flux present a steady state
circulation appeared to form quite rapidly.
The other important variation involves the case of a variable wind
field. If the wind changes direction (Dickey and Simpson, 1983) or speed,
gravitational adjustments lead to large intrusions that may degenerate
into solitons (Kao et al., 1978; Maxworthy, 1980) or undular bores
(Thorpe, 1974; Farmer, 1978; Smyth and Holloway, 1988).
In summary, the upwelling response of the epilimnion (diurnal
thermocline) is determined by the magnitude of W, and that of the
metalimnion by the magnitude of L N . Where W is small, but LN large, it
is possible to model the entrainment process at steady state as simple
recirculation, where the entrained fluid is swept along the interface into
the upwind upwelling region, then is mixed in the surface layer downwind
by shear dispersion. If the stratification is complicated and admits
numerous eigenfunctions of equal importance, then an equivalent num-
ber of nondimensional numbers become important.

VI. Differential Deepening

The wind field over a lake surface is rarely uniform. Mesoscale wind
variability (McBean and Paterson, 1975; Bean et al., 1975) causes
 Physical Limnology 371

variation over large lakes, and wind sheltering by the surrounding terrain
is the major factor affecting smaller lakes (Parker and Imberger, 1986).
Wind variability has two major effects. First, in areas with relatively high
wind speed, evaporation, and thus latent heat transfer, will be greater,
introducing a horizontal gradient in the surface heat flux. This phenome-
non, called differential cooling (Imberger, 1982), is discussed in Section
VII. Second, higher wind speeds on the exposed parts of the lake cause
both a more rapid local deepening of the surface layer and a greater
introduction of momentum. Both must ultimately be averaged across the
lake surface by gravitational adjustments of the isopycnals. The variabi-
lity in the surface layer deepening is called differential deepening
(Imberger and Parker, 1985).
Data from a field investigation near the headland of Salmon Brook in
the Wellington Reservoir illustrate these three phenomena extremely
well (Parker and Imberger, 1986). The data were collected along a
transect extending from a station in a sheltered embayment, station W4,
to the center of the basin exposed to the wind, Station C10 (Figure 23a).
The morning was warm and there was no wind. Shortly thereafter, a
southerly wind began to blow (wind at C10 was 2 m s-' and at W4 was
1m s-I); the surface layer was warm in the sheltered part of the transect
with some noticeable cooling out towards C10 (Figure 23b). This
difference was due to greater evaporative cooling in the open parts of the
lake. The wind intensified sharply at around 1540hours (rose to 4ms-'
at C10, but remained less than 0.6ms-' at W4) leading to noticeable
deepening at C10 (Figure 23c). The wind continued strongly until about
2000hours, when it decreased markedly to less than l m s - ' . The
horizontal density gradient associated with the isotherms is shown in
Figure 23d, implying a further deepening than that at 1805 hours. After
this time, the isotherms progressively relaxed (Figure 23e) until at
2300 hours (Figure 23f), the isotherms were once again virtually
horizontal.
Parker and Imberger (1986) investigated the evolution of the water
profiles by using the surface layer models described in the previous
section. They showed that the local deepening laws could be used to
describe the observations if local wind conditions were used and the
effects of horizontal advection were accounted for, although this ap-
peared to be a relatively minor correction. The results from Parker and
Imberger (1986) suggest that to obtain an average deepening law, careful
averaging must be done over the surface of the lake.
372 Jorg Imberger and John C. Patterson

15000
MAIN BASIN OF

1420

14kOO 15100 15600 16b00

X COORDINATE (m)

b: 1553 TEMPERATURE ("C)

N M K C10

1.0

FIG.23. An example of differential deepening recorded in the Wellington Reservoir in

March 1984. (a) Location map showing the isotherms. (b) Isotherms before the wind had
started at either the central or the sheltered station. (c) Data recorded at 1800 hours-the
wind speed in the central station had risen to 4 m s-1, but in the sheltered corner, the wind
was still less than 0.6111s-'. (d) Data recorded at 1948hourewind speed had dropped
everywhere to less than l.Oms-' immediately prior to taking this data set. (e) Data
recorded at 2137 hours-wind at both stations was less than 0.5 m s-', thermal structure was
relaxing and intruding out into the central part of the lake. (f) Data recorded at
2310 hours-the wind remained calm during the whole period, allowing the thermal
structure to adjust. The heat flux was beginning to go negative, causing slight cooling at the
surface. After Parker and Imberger (1986).
 Physical Limnology 373

c: 1805 TEMPERATURE (“C)

00

1.0-
-,..- *-.- .*.-

--E
2.0-
h
X I

N M WClO K C10

0.0
PROJECTED DISTANCE (rn)

d: 1948 TEMPERATURE (“C)

PROJECTED DISTANCE (rn)

4.0
100.0 200.0 300.0 400.0 500.0
PROJECTED DISTANCE

FIG.23. (conrd.)
374 Jorg Imberger and John C. Patterson

f: 2310 TEMPERATURE ("C)

4.0
100 0 200.0 300.0 4000 500.0
PROJECTEDDISTANCE (rn)

FIG.23. (conrd.)

The case just discussed was for conditions of moderate W and large
LN. In Figure 24a, we show a much larger transect extending along the
length of the Wellington Reservoir approximately six hours after a
westerly wind had begun. The speed of the wind was 2ms-' at
0600 hours, then rose steadily and peaked at 6 m s-l at 1430hours. The
direction remained quite steady at due west (see Figures 17b and 17c).
The dramatic two-dimensionality of the structure is observed in the
isopycnals in Figure 24a with severe deepening between Stations C10 and
C45 and again between Stations C80 and C90; both sections were more
exposed to the strong westerly. The deepening of the isotherms between
C10 and C45 was due to the setdown associated with upwelling at C10,
whereas the slope of the isopycnals between C40 and C60 was due to the
more sheltered nature of this part of the valley. Confirmation of the
general tilt west-east due to upwelling is given in Figure 24b, which
shows data from a west-east transect at Station C40. The water in both
exposed areas was warmer than its surroundings, which is distinct from
what is observed in Figures 23b to 23f. This indicates that advection was
dominating the deepening process (Imberger, 1985). Further, the average
Wedderburn number for the central basin surface flow at 0927 hours was
0.25, and LN was approximately 1, implying a much less stable lake as a
whole, with mixing expected throughout; the shear associated with
upwelling (4.20) was sufficient even at Station C40, close to the end of
 a DENSWY (kg / m3) b DENSITY (kg I m3)
0.0. 0.0.

5.0 5.0.

10.0- 10.0-

-
E E
--
15.0. I 15.0.
E I-
n
y1
n l !k
20.0- 20.0-

25.0- 25.0-

30.0.
:loo,

2000.0
CAW Iy
4000.0
CB", "I" I
6000.0 8000.0
Ac;\y"ycL

10000.0 12000.0
30.0-
0.0
A

200.0
A A

300.0
*
4dO.O
.I
500.0
A

6bO.O
A

7[
EAST WE
CUMULATIVE DISTANCE (m)

FIG.24. (a) Transect along the Wellington Valley (see Figure 17a). The wind was coming from the west (as shown in Figures 17b and
17c) and blowing across the lake between the 2000 and 5OOO m mark as well as between the 12000-16000m mark. The isopycnals show
extreme differential deepening, the wind strength at the time was 6ms-'. (b) Cross section at Station C40, from east to west, the
differential deepening, shown in Figure 24a was accompanied by strong upwelling. (c) Temperature. (d) Gradient temperature. (e)
Dissipation as recorded with microstructure instrument at Station C40 at 0947 hours. Dissipation as measured by both the Wigner-Vie
technique (sticks) and the Batchelor spectrum fitting technique (bar chart).
376 Jorg Imberger and John C. Patterson
F
0 1
OL
1 I I I I
0 (D N
l- z P
N
(u)Hld30
 Physical Limnology 377

the basin, to induce very active mixing throughout the water column (see
Figures 24c, 24d, and 24e).
De Szoeke (1980) addressed the problem of a variable wind stress over
the ocean. However, he aimed to determine the variability induced by
the Ekman pumping driven by the curl of the surface wind stress and not
by the differential rates of vertical entrainment. Maxworthy and Moni-
smith (1988) were motivated, in part, by the observations in the
Wellington (Parker and Imberger, 1986) and conducted a grid stirring
experiment where the grid extended over only part of a long, stratified
tank. Their results illustrated the sequence of events immediately after
the commencement of the grid oscillations. First, the turbulence intro-
duced by the grid mixed the underlying fluid and set up a downward
propagating entraining front immediately below the grid. Second, this
front continued to propagate downwards until either the turbulence
became weaker due to the greater distance from the grid (Hopfinger and
Toly, 1976), or the mixed fluid collapsed and intruded horizontally into
the neighboring quiescent ambient fluid. Third, as the intrusion propag-
ated out into the main tank, it reached a stage where the end of the tank
began to slow its progress, once again influencing the rate of descent of
the entrainment front immediately below the grid.
Using the various relationships for intrusion speed and entrainment
rates, Maxworthy and Monismith (1988) derived a relationship for the
speed of propagation of the entrainment front under the grid and
satisfactorily verified these relationships with the experiments. The grid
mechanism, while convenient, is a poor analogue for wind-induced
turbulence. Thus, even though the flow scenario will be the same, it is
more useful, as Maxworthy and Monismith (1988) did, to write down the
relationships for the case where the mixing is carried out by a wind stress.
Consider a strongly stratified surface layer over part of which a strong
wind stress is suddenly initiated with a shear velocity of u , . Under the
wind stress, the initial deepening will be given by (4.25b)

1 dh - C,
u , dt Ri* ’

which is the relationship relevant for a linearly stratified fluid and where
C , = 0.1 (Kranenburg, 1985). As shown by de Szoeke and Rhines (1976),
surface-induced turbulence and deepening dominates initially, until the
shear at the base of the mixed layer (5.2) builds up. For longer times,
378 Jorg Zmberger and John C. Patterson

shear production (4.25a) dominates and the entrainment rate is described

by:
1 dh U
-- = (2Ca)'n(-?)
u* dt Nh '
where C,= 0.2 to 0.3 (Spigel er al., 1986).
These deepening processes will continue (Maxworthy and Monismith,
1988) until the outflows/inflows become appreciable (see Figure 25) and
dilute the effort expended by the vertical mixing processes.
It is well established (Maxworthy, 1972; Imberger et al., 1976; Manins,
1976) that the intrusion can either be governed by an inertial buoyancy
balance or a viscous inertial buoyancy balance; in the reservoir, however,
the scales are such that invariably the inertial buoyancy force balance
prevails. This means that the intrusion thickness is given by (for a
two-dimensional intrusion)

and the speed by

c = 0.44(Nq)1'2,

where q is the two-dimensional volume flux and N the buoyancy

frequency. However, as observed by Maxworthy and Monismith (1988),
6 is not necessarily equal to the depth H of the entraining front under the

EXPOSED

- SHELTERED

~~

ENTRAINMENT
FRONT

STRATIFIED 4-

FIG. 25. Schematic of differential deepening.

 Physical Limnology 379

grid. The flow out of the turbulent area is governed by a gravitationally

driven shear flow, which is retarded by the vertical turbulent transport.
Such flows have received extensive treatment in estuarine situations;
detailed solutions are described in Scott and Imberger (1988). The flow
itself was recently modeled in the laboratory by Linden and Simpson
(1986), who, however, do not appear to have connected the dynamics of
their flow with the pioneering work of Taylor (1954) on shear flow
dispersion (see Fischer et al. (1979) for a general reference). It is
therefore better to re-analyze the Linden and Simpson (1986) experi-
ments in light of the estuarine work and the analysis presented by
Maxworthy and Monismith (1988).
Suppose the wind introduces a vertical mixing coefficient given by
(Fischer et al., 1979)
E, = 0.06 u,h. (6.5)
The flow under the grid then may be derived, if it is assumed that the grid
is much larger than the mixed region is deep ( h / L< I), from the long
box convection problem, where L is the length of the box.
A simple balance between baroclinic vorticity generation and vertical
diffusion leads to the scale of the velocity under the grid (Imberger, 1974)

where g’ is the reduced gravity across the turbulent front and Lf is the
length of the density variation at the outflow end; this length increases
with time as the intrusion increases to bring outer fluid into the turbulent
zone. This balance may be expected to be valid provided Lf > h2ule,
(Fischer et al., 1979).
The effective longitudinal dispersion coefficient is given by Fischer et
al. (1979) for such a velocity profile:
h2u2
K, = 5.3 x
€2

and a good approximation to Lfis given by

L‘ = (Kxt)1’2 (6.8)
for times which are short enough so that L > Lf.Combining (6.5) and
(6.8) yields the relationships for both the dispersion coefficient and the
380 Jorg Zmberger and John C . Patterson

peak velocity:
(6.9)
12 3 114
u=1.2(g@)
u*t , (6.10)

where it was assumed that the depth h adjusts slowly compared to the
time it takes for the flow to adjust to the changing Lf.
Unfortunately, the data presented in Linden and Simpson (1986) are
not sufficiently detailed to allow comparison. Maxworthy and Monismith
(1988) balanced the deepening given by (4.23) with the vertical advection
induced by the outflowing intrusional such that

(6.11)

However, if we consider the example shown in Fig. 23c, then K x =

4x m s-l, so that in the time deepening has taken place, Lfhas only
grown to 6 m , far too small to influence the process; it is therefore not
surprising that Parker and Imberger (1986) obtained good comparisons
between the one-dimensional model and the field data once they used
local wind speeds. The oversight in the paper by Maxworthy and
Monismith (1988) was that no distinction was made between the length
scale Lf and L.
In summary, the concept of differential deepening appears to be well
established and documented in the field. The laboratory experiments by
Maxworthy and Monismith (1988) have given an explanation for the
connection between the deepening under the exposed areas and the
intrusional flows set up by the buoyancy imbalance introduced by the
wind. However, it appears that for diurnal cycles, intrusional flow is of
secondary importance to the deepening. However, the intrusional flows,
which can reach velocities of up to 5cm per second (Parker and
Imberger, 1986), are of primary importance in redistributing water
horizontally once the wind event is over.

Vn. DEerential Heating and Cooling

Field data collected in Salmon Brook, a side arm of the Wellington
Reservoir (Monismith and Imberger, 1988), provides a good illustration
of differential heating and cooling. Salmon Brook (Figure 26a) runs
 Physical Limnology 381

north-south with a central uniform valley. The surrounding terrain is

hilly and the upper reaches of the side arm are sheltered from the wind.
Data were collected in late February 1985 during a hot (39"C), clear
day, where the peak net heat flux at the surface reached 960 W m-*. The
wind was less than 2 m s - ' all day except briefly between 1500 and
1830 hours, when a westerly wind with a speed of 5.5 m s-' sprang up.
The night had been relatively cool and still, with only a small northerly
breeze around midnight which reached 4 m s-'. This stopped completely
by 0700 hours.
An early morning transect along the central valley of Salmon Brook
from the upstream extreme to Station C10 (see Figure 26a) is shown in
Figure 26b. Cold water, which had formed around the perimeter of the
side arm during the night, fell under gravity from the surface to about
7 m . A transverse transect taken across SB20 (Figure 26c) shows the
symmetric plunging of the boundary water. Drogue tracking indicated a
drawing of central water into the side arm along the surface consistent
with the temperature observations that the night cooling had formed cold
boundary water, which downwelled around the perimeter of the side
arm. The inflow occurred over 50-70% of the depth of the underflow
(7 m), comparable to the laboratory results of Harashima and Watanabe
(1986).
In the early afternoon, after strong heating all day, the stratification
changed to that shown in Figures 26d and 26e, indicating strong boundary
heating around the perimeter with baroclinic pressure gradients of the
temperature field forcing the water away from the perimeter into the
main basin. However, Monismith and Imberger (1988) observed that
during this time the surface water was still slowly moving into the side
arm, holding the warm water near the perimeter.
By late afternoon, the surface water motion had reversed and a strong,
buoyant surface overflow (velocity approximately 0.10 m s-l) confined to
the top 1m was observed to be moving the water out into the main basin.
By 2200hours, the whole of the surface water in the top 2 m had been
replaced by this mechanism.
The westerly wind that sprang up at 1520 hours (see Figures 26f, 26g,
and 26h) immediately moved the thin surface buoyant plume onto the
east bank; the influence of the wind was progressively stronger towards
the mouth of the brook. It is interesting, in passing, to observe the
general shape of the upwelling pattern in Salmon Brook as one moves
382 Jorg Zmberger and John C . Patterson
a

1moo-

--
E
5 14600-

I -
>
14200-
SALMON

N
14b00 15dOO 1 seoo 16hOO
X COORDINATE (m)

TEMPERATURE (“C) TEMPERATURE (“C)

C
0.0-

?‘
2.0-

,
4.0-

-E-E 6.0-
0

8.0-

10.0- 10.0-

12.0- - 0 . 21
I 200.0 4000 600.0 800.0 1000.0 12 0 0.0 100.0 200.0 300.0 400.0 51 0

PROJECTED DISTANCE (m) PROJECTED DISTANCE (m)

FIG.26. An example of differential heating and cooling, illustrated with data collected
in Salmon Brook on 25 February 1985. (a) Bottom bathymetry of the site showing site and
transect positions. (b) Longitudinal transect taken early morning, starting at 0849 hours. (c)
Transect across Salmon Brook at Station SB20, early morning, commencing at 0909 hours.
(d) Longitudinal transect illustrating midday heating, taken at 1333hours. (e) Cross-
sectional transect at Station SB20, showing thermal heating near boundaries, data taken at
1340 hours. (f) Transect taken at SB25 at 1639 hours, illustrating the influence of a very
weak westerly wind. (g) Transect taken at SB20 at 1536 hours, the wind at this station was
somewhat stronger. (h) Transect taken at SBOO, commencing at 1639 hours. This section
was almost completely exposed to the wind, which now had reached a value of 4 m s-’. (i)
The net heat flux variation over the five days of the experiment as computed by the rate of
change of heat of the total water column at Station SB30.
 Physical Limnology 383
d TEMPERATURECC)

“-I 10.0

12.0
5830

2dO.O
--
4ob.O Sdo.0
A
----..._
- .-
8dO.O
L

PRQlECTED DISTANCE (m)

A
c10
;
lObO.0 12w.0

a TEMPERANRE (‘C) f TEMPERANRE (“C)

4.01 I 2.01
-..._ ,.-*.----.

10.0

s020w S020E S025W S025E

12.0
0.0 100.0 200.0 3W.O 400.0 500.0 0.0 20.0 40.0 60.0 80.0 100.0 120.0 140.0 1 1.0
PROJECTED DISTANCE (m) PROJECTEDDISTANCE ( m )

FIG.26. (cod.)

from the sheltered SB25 station to the more exposed SBOO (Figure 26h)
station, and to compare this with the discussion in Section VI. At SB25,
the Wedderburn number was around 5-10, as indicated by the very weak
surface upwelling and the almost horizontal isotherms immediately
below. At SB20, the whole diurnal thermocline was tilted, surfacing at
the west end of the side arm, indicating that the Wedderburn number was
around 1. Lastly, at SBOO, the Wedderburn number was calculated as 0.2
and the isotherms indicate a very similar pattern to that shown in Figure
19h. where the Wedderburn number was 0.4.
384 Jorg Imberger and John C. Patterson
TEMPERATURE (‘C)
g

SB2OW SBBOE
6.0
200.0 3W.O 400.0 500.0 600.0
PROJECTED DISTANCE (m)

TEMPERATURE (‘C) TEMPERATURE(‘C)

h i
0 0-
I

I
c

TOTALNETHEATFLUXAT I
I

-
iI
10-

2 0-
-1000 MUOBSERVED
HEAT INPUT AT WSi
-
, .
m DAY 65
I

,
7
DAY 68
DAY 67
DAVU
O A V ~

-E v-
cc

z30-
’=5.
a -
- - - - - M- - -
U OQSERMD
Lu * 8URFACECOOLING
4 0-

+I00
5 0- 0 0600 1200 I800 2400 0600

SBlOE
A
l.,/= 0- *O-

./--_*..-,..
,,
.-.o-

A
SBlOW
A
TIME (H)

60
I I I

FIG.26. (conrd.)

The consequences for the heat budget of this cyclic flow pattern are
illustrated in Figure 26i. The data show that the water column at Station
SB30 (the head of the side arm) responded as if twice the surface heat
load had been applied in both the heating and cooling phases. This is
further strong evidence that the inertia of the water leads to a large phase
lag between the thermal forcing and the flow response.
Flow patterns associated with the cooling phase are driven by the
greater temperature drop around the boundary; it is yet to be established
whether a variable heat flux also contributed to the excess temperature.
 Physical Limnology 385

On the heating phase, both the reduced depth and the reduced latent
heat loss (lower wind speed in the sheltered areas) around the perimeter
lead to the greater temperatures around the boundaries. Steady convec-
tion in a long, shallow cavity, an approximation of the side arm, has
received a great deal of attention for two-dimensional configurations
(Hart, 1972; Cormack et al., 1974; Imberger, 1974; Bejan and Tien,
1978; Bejan and Rossie, 1981; Bejan et al., 1981; Inaba et al., 1981;
Simpkins and Dudderar, 1981; Hart, 1983a, 1983b; Drummond and
Korpela, 1987) and in three-dimensions by Imberger (1976) and Scott
and Imberger (1988). These solutions rely on a viscous buoyancy balance
at first order and all have stable stratification that changes from one with
vertical isotherms at low Rayleigh numbers to one with horizontal
isotherms and boundary jets along the top and bottom boundaries at very
large Rayleigh numbers, such as those that operate in a side arm of a
reservoir. These solutions are of limited value, however, as they require
an imposed steady horizontal density gradient, with zero heat flux at the
upper and lower boundaries. Surface heat fluxes can be incorporated into
the solution technique (Cormack et al., 1975; Scott and Imberger, 1988)
but unless vertical velocities are allowed at first order, these heat fluxes
can only be incorporated at second order, making the solution techniques
less applicable.
Most importantly, in these diurnally cycled flows the inertia of the
water leads to a velocity field almost 90" out of phase with the thermal
forcing. In the data shown in Figure 26, the velocities did not reverse
until about 1500 hours, seven hours after the net heat flux had changed
from a loss to a gain (Monismith and Imberger, 1988). The explanation
of this observation lies in the following scaling arguments. A thermal
gradient is instantaneously imposed on a long cavity of fluid. This
gradient would induce a flow which would be governed by the balance

_
dP -
dz
- -gP,

leading to a time scale for spin-up;

T=-U (7.3)
Kgh'
386 Jorg Imberger and John C. Patterson

where u is a typical velocity of the final flow, K is the density gradient

and h is the depth of the flow. Inserting typical values derived from
Figure 26d leads to a time scale of two hours, which is reasonable given
the gradual build-up of the gradient.
The unsteady cavity problem has also been investigated by Patterson
and Imberger (1980) and Patterson (1984). The latter paper used internal
radiative heating to drive the flow rather than heating and cooling at the
vertical end boundaries. The critical parameter to emerge from the long
box problem is Gr A4, where

gcu ABh3
Gr = (7.4)
Y2 ’

A = -h (7.5)
L’

cu is the coefficient of thermal expansion, and At) is the temperature

differential. Once again, the applicability of these theories is marginal
due to the cyclic nature of the heat sources and the very different
boundary conditions. However, evaluation of the parameter P = Gr A4
leads to a value well in excess of 1 for a typical side arm, indicating that
the flow in side arms must be dominated strongly by inertia even if an
effective diffusivity Y of m2s-l is used, as would be indicated by the
turbulence measurements in these flows (Imberger, 1988).
The integral solutions advanced by Sturm (1981) and Jain (1982) seek
to explain the laboratory findings of Brocard and Harleman (1980). All
three deal with side arms with a strong surface heat loss, sustaining an
unstable upper water column and a longitudinal circulation driven by a
longitudinal temperature gradient. These investigations were motivated
by cooling pond configurations where there is a net surface heat loss at all
times and a steady state solution can be contemplated. Harashima and
Watanabe (1986) presented further laboratory verification that the
solutions found by Sturm (1981) and Jain (1982) meaningfully describe
these flows. However, since the surface temperature in a cooling pond is
generally above the normal equilibrium temperature for the prevailing
 Physical Limnology 387

meteorological fluxes, a longitudinal surface temperature gradient always

develops. By comparison (Figure 26b) no such gradient develops during
the cooling phase, which is better described by a source of negative
buoyancy distributed around the perimeter of the lake. By contrast, the
heating cycle may be modeled by a line source of buoyancy in
combination with a variable heat flux along the length of the side arm.
Imberger (1985) pointed to a potentially important convection problem
in lakes: differential absorption. Light and short wave radiation penetrate
the water column to considerable depth; the exact depth is dependent on
the concentration of particulate matter and the color of the water (Kirk,
1983). Where the water is transparent, short wave energy is distributed to
considerable depth; conversely, in areas of high turbidity, radiation will
penetrate to a much lesser depth and more heat will be absorbed near the
surface of the water column. This causes the surface temperature in the
turbid area to rise, and the water immediately beneath, shaded by the
turbidity, will become colder than in neighboring areas. Large tempera-
ture changes may therefore be expected to build up over successive days
of heating where the turbidity distribution is uneven.
Data collected recently in the upper reaches of the Canning Reservoir
are shown in Figures 27a and 27b, for the transect from C45 to C240 (see
Figure 27c). The isotherms show that a strong surface warm water lens
had formed where the water was turbid near Station CA100. Underneath
this lens, the water was considerably colder than the water at the same
depth at Station CA45. At the time of these measurements, a weak
variation in salinity existed at the upstream end of the reservoir, enough
to be used as a tracer, but at a low enough concentration not to
significantly affect the density field. The corresponding salinity structures
are shown in Figure 27b. An intrusion had formed at a depth of 3 m and
was propagating downstream and upstream. At the surface, the hot water
lens was spreading out, as evidenced by the spread of the isohalines,
although the speed was considerably slower than in the subsurface
intrusion. These flows are driven by the weak density gradients induced
by differential absorption. The flow may be termed a thermal siphon, and
will in many cases be a major contributor to a horizontal redistribution of
water in areas of patchy turbidity, an important factor in the kinetics of
phytoplankton communities.
Two theoretical models have been developed from the results shown in
Figures 27a and 27b. Patterson (1984) used the long cavity with a
longitudinally variable absorption as a model. He assumed a longitudinal
388 Jorg Imberger and John C. Patterson

dependence of the radiation flux

Q(x, 2 , t ) = 2Qoh(x - $), (7.6)

where h is the depth of the cavity, L is the length of the cavity, and Qo is
a constant, the value of which was used to fix the intensity of the
radiative heating.
The flow was assumed to be laminar so that the thermal energy

b SALINITY (psfi)

DISTANCE (m)

FIG.27. (a) Isotherms (“C) along a transect from Station CA45 (distance = 4377 m) in
the main Canning River tributary to Station CA24D (distance=9147m). Data was
collected between 1704 hours and 1813 hours, 15 December 1983. For station locations, see
Figure 7b. (From Imberger (1985).) (b) Isohalines (pss) for the same data set as in (a).
(From Imberger (1989.) (c) Map of Canning Reservoir showing measurement stations.
 Physical Limnology 389

POISON GULLY

CANNING RESERVOIR

CA180

CA230

oooo 1000 2000 3000 4Ooo 6000

LOCAL CO-OROINATES (m)
FIG.27. (contd.)

equation became
ae de ae
-+ u -+ W - = K V20 + Q(x, Z, t ) . (7.7)
at ax az
A scaling analysis revealed that the flow development, for the initial
value problem, depended critically on the magnitude of the parameter
P = Gr A4, (7.8)
390 Jorg Zmberger and John C . Patterson
where
gaQoh4L3
Gr = (7.9)
Y3

is the Grashof number for the problem and A is defined by (7.5). Eight
regimes of flow were defined. There are two extremes: P > Pr4 (where
Pr = Y / K , the Prandtl number) is the regime where inertia and convec-
tion dominate the final stages of flow. Second, P<Pr-'A4 gives a
thermal field described by a simple balance between conduction and
radiative heating.
Introducing typical values for the problem illustrated in Figures 27a
-
and 27b leads to the first regime (assuming Pr 1) as P is much greater
than one. For this regime Patterson (1984) observed a single gyre
centered at the transition between heating and cooling. The eddy was flat
and supported horizontal isotherms.
Trevisan and Bejan (1986) modeled the turbidity cloud as the perfect
reflector. This is the extreme in differential absorption and, once again, a
simple gyre was generated at the boundary between the absorbing and
reflecting areas although the gyre computed by these authors was not
dissimilar to that found by Patterson (1984). Neither model describes the
real situation; the absorption is differential in horizontal and vertical
directions and the patches are usually three-dimensional.
The original problem that motivated research into convective motions
was in cold lakes and concentrated on the thermal bar problems
(Rodgers, 1971; Spain et al., 1976; Mortimer, 1987). This phenomenon
occurs when a lake, initially at a temperature less than 4"C, begins to
warm. Similar to the observations described above, this warming occurs
more rapidly around the perimeter due to the shallow depth and the
warmer river inflows. In this sense, it is identical to what is believed to be
a good model for the data shown in Figures 26d and 26e. The extra
complication is that the equation of state is nonlinear at these low
temperatures, with a maximum density at 4°C.
Once the boundary fluid has a temperature greater than 4"C, it is
lighter and propagates along the surface. As it meets the colder (<4"C)
water offshore, mixing produces some 4°C water that sinks (Marmoush et
al. , 1984) and produces a complicated frontal mixing 'pattern, distinct
from the normal buoyant surface overflow which has a well-defined
frontal roller (Luketina and Imberger, 1987). Elliott and Elliott (1970)
and Ivey and Hamblin (1988) have carried out laboratory experiments in
 Physical Limnology 391

long cavities and Ivey and Hamblin (1988) in particular present graphic
data for the formation of a strong region of convergence, where the 4°C
water plunged in a plumelike sheet to the bottom of the tank. By analogy
with the long box problem, Ivey and Hamblin (1988) suggest the thermal
bar (the region of strong convergence) has a width of O(h Ra-"4), where
gtu( A q 2 h 3
Ra = 9 (7.10)
YK

and
p = po[ 1 - a(8 - 80)21 (7.11)
describes the equation of state in the range 0°C I8 5 8"C, Bo = 4°C.
Bennett (1971) included rotation to show that a long shore shear was
generated by the thermal bar density gradients. Further, a transverse
gyre motion supporting the convergence reinforced the thermal bar. This
analysis showed that the time scales for the establishment and propaga-
tion into the center of the lake is of the order of months, distinct from the
behavior in temperate lakes where, as shown above, it is a matter of
hours.
In summary, we have illustrated a new class of convective flows driven
by what has been called differential heating and cooling. The main
features of this flow are that it is three-dimensional and unsteady with the
inertia of the water introducing a large phase lag between the thermal
forcing and the resulting motions. The impact of these motions should
not be underestimated. In the smaller lakes discussed above, these
convective transports will communicate boundary water to the center of
the lake within hours and in large lakes, where rotation is important,
Ookubo et al. (1984) have shown that the boundary thermal inputs lead
to basin scale gyres, which again ensure a greatly enhanced horizontal
transport.

vm. outflow
The outflow from a stratified fluid is influenced by the buoyancy force,
which inhibits differential vertical motion, causing tilting of the isopyc-
nals. When an outlet is suddenly opened, the pressure wave instan-
taneously sets up a radial flow pattern towards the outlet. However, the
radial convergence near the sink quickly distorts the isopycnals, leading
392 Jorg Zmberger and John C. Patterson

to uneven vertical displacements. Buoyancy initiates a set of internal

waves that propagate out, adjusting the isopycnals back to a horizontal
neutral position, and the initial radial flow collapses into a jet-like
structure, flowing horizontally toward the outlet. Such a flow field is
called selective withdrawal, as most of the flow comes from the level of
the sink.
Selective withdrawal has wide application as a management technique
for the supply of water with desired water quality properties. The
technique was used intuitively to scour out poor quality water from the
bottom of the reservoir, long before any theory was developed. It was
not until the early 1960s that Raphael (1962), Koh (1966b), and
Wunderlich and Elder (1969) developed the foundations of the selective
withdrawal process.
The effect of selective withdrawal on the stratification of the reservoir
has received considerably less attention. The thermal structure shown in
Figure 6, measured in Wellington Reservoir in Western Australia, is an
example. Prior to 1976, the lake had a very strong thermocline from
November to June, which caused severe oxygen depletion in the
hypolimnion. In August 1976, the scour policy was first used (see
Imberger and Hebbert (1980)) to rid the reservoir of highly saline water
from the first inflows. The procedure weakened the density structure,
lowering the value of the Lake number so that the thermal structure was
completely changed for the rest of the year. This change influenced the
mixing, which affected the vertical flux of oxygen, and thus the reservoir’s
water quality. While the Wellington Reservoir is an extreme example
because of the influence of salinity on density structure, it displays the
principle well.
Selective withdrawal may thus be used to optimize the water quality in
the reservoir and so obviate the need for mechanical devices such as
pumps or bubblers. Research into selective withdrawal has been inten-
sive, and there is a large body of literature. This discussion is divided into
the following categories:
(1) line sink in a two-layered system,
(2) point sink in a two-layered flow,
(3) line sink in a linear stratification,
(4) point sink in a linear stratification,
(5) selective withdrawal in a rotating stratified fluid.
An empirical procedure for determining withdrawal characteristics from a
 Physical Limnology 393

nonlinear stratification is described in Fischer et al. (1979); few fun-

damental results exist.

1. Line Sink in a Two-Layered System

Consider first the simple case of an infinitely deep fluid of density

+
( p Ap) on top of which is a thin, finite layer of water with a density of
p. The case where there is no surface tension between the fluids, the
viscosity is small and a line sink is located a depth H below the interface
was first investigated by Craya (1949). The two layer case is mathemati-
cally identical to that of a single fluid with g' = ( A p / p ) g replacing g
(Imberger, 1980). Since the fluid is infinitely deep and the pressure is
constant along the free surface, the Bernoulli equation can be applied
along the free surface. Craya (1949) postulated a solution with a sharp
cusp extending from the surface to distance h above the sink. This is
distinct from the flow where a small stagnation mount forms over the sink
(Imberger, 1980). The flow at infinity is at rest and, assuming the flow in
the vicinity of the sink is radial and unaffected by the presence of the
cusp, the Bernoulli equation reads

where q is the two-dimensional volume flux at the sink. Craya (1949)

solved this equation for the Froude number F:

such that
F 2 = 8 n 2 ~ 2 (-
1 E),
where E is the ratio h l H . He postulated that the critical Froude number
(the Froude number at which the upper surface is drawn down into the
cusp) was given by
dF2
-- - 0.
dE
This leads to the solution E = $ and

l$ = 3.42. (8.5)
394 Jorg Imberger and John C . Patterson

Gariel (1949) performed experiments with a line sink at the bottom of

a flat bed tank. Modifying his results for the full q as above yields a
critical Froude number
F, = 3.0 (8.6)
Tuck and Vanden Broeck (1984) used conformal mapping and a series
solution technique to determine a cusp solution for the above semi-
infinite fluid domain problem. They obtained a value (for the full q)
F, = 3.55. (8.7)
Hocking (1985) treated the case of a sink above a sloping bottom with an
arbitrary angle. In the limit of a line sink embedded at the apex of the
sloping bottom and the slope going to zero, he found the critical value
F, = 2.88, (8 * 8)
which is smaller but of similar order to the Bernoulli solution (8.6).
Hocking (1988) gave experimental evidence that the critical value was
1.2, much smaller than that predicted by (8.5) or (8.7). Yih (1980),
however, speculated upon the existence of solutions with a cusp to the
problem with a linearized (constant velocity) free surface condition and
hence infinite Froude number. Such solutions are impossible in a fluid of
infinite depth, since the velocity at all points in the fluid must approach
zero at large distances from the sink. This conditipn would mean that the
velocity everywhere on the free surface was zero. If solutions exist in the
finite depth infinite Froude number problem, then solutions must also
exist for finite, but large, Froude numbers. This would mean that for this
geometry, there would be more than one solution with a cusp. Yih (1980)
suggested that the critical withdrawal layer would then be the lowest
value of the Froude number for which a cusp solution existed. Collings
(1985) and Hocking (1988) found solutions, the latter over a complete
range of sink to bottom depth ratios, into the range where the sink rises
above the level of the free surface at infinity (negative ratios).
Vanden Broeck and Keller (1987) found not only the infinite Froude
number solutions but, using a variation on the series method, solutions of
the full range of sink to bottom depth ratios between zero and one, for a
wide range of finite Froude numbers. They found that as the Froude
number was decreased from infinity to a given ratio, cusp-like solutions
were obtained at all values until some critical Froude number was
reached, or the Froude number reached 2. They speculated that where
 Physical Limnology 395

4
\
CUSPED SOLUTIONS
\.
1.
1 .
\.
A.

LL"2
CUSPED SOLUTIONS
WITH WAVES -
UNIQUE WAVELESS SPECULATIONONLY
SOLUTIONS

#tEXPERIMENT
HOCKING
(1 988)
0
-l:o -0.5 0.0 0.5 1 .o
(H - h)l H
FIG. 28. Solid line: theoretical results from Vanden Broeck and Keller (1987).
Dot-dash line: theoretical results with a sink (Hocking, 1988). The Froude number F, in
this diagram is based on the depth of water and not the sink depth.

solutions existed down to a Froude number of 2, solutions with waves on

the free surface could also be found for a Froude number less than 2,
although they did not calculate them. They also found the branch of
solutions on which no waves existed and which allowed solutions into the
range of Froude numbers much less than 2. This branch contains
solutions in which the fluid is of infinite depth. These results have been
summarized in the phase diagram in Figure 28.

2. Point Sink in a Two-Layered Fluid

Here, the Bernoulli approach for the semi-infinite domain becomes

(using the same notation)

1
H = h + - - Q2 (8.9)
(4nh2)' 2g'

where Q is the three-dimensional volume flux. Rearranging and finding

the maximum of this expression leads to the critical Froude number

F, = 5.09, (8.10)
396 Jorg Imberger and John C . Patterson
where

(8.11)

This result was first derived by Craya (1949). More recently Wood (1978)
generalized (8.10) to flows in a sector of a sphere and Bryant and Wood
(1976) treated the flow of multiple layers in a sector. Forbes and Hocking
(1988) derived mathematical solutions for the stagnation flow solutions
with no cusps up to a Froude number of 6.4. Experimental verification
was carried out by Lawrence and Imberger (1979) who found a value of
F, = 4.9, which was confirmed by Wood (1978). Harleman et al. (1959)
had found a value of 3.4, whereas the experiments of Jirka and Katavola
(1979) yielded a value of 3.15.
Blake and Ivey (1986a,b) also looked at the case where viscosity is the
controlling force and not inertia to derive the result
h c = 2 . 1 (VgQ
l ) 'I4
9
(8.12)

where h, is the critical draw-down depth. The transition between (8.12)

and (8.10) occurs when the parameter (Ivey and Blake, 1985)
S=(T)Q3g' =2+5.
(8.13)

In both the line and point sinks a diffuse interface will lead to lower
critical Froude numbers (Jirka and Katavola, 1979). This may be
explained by comparing the flow with the upwelling flow in a high
Wedderburn number regime; the three-layer solution of Monismith
(1986) showed a distinct divergence of the interface at the upwind end.
Such a mode 2 response is identical to the interface spreading near the
sink.
For discharges greater than the critical value given by (8.10) or (8.12),
both layers flow toward the sink, such that the interface makes a
well-defined angle at the sink and has a critical point at which the flow in
both layers is controlled. Wood (1978) derived an expression A = Q E / Q H
as a function of the Froude number

where QE is the flow from the epilimnion and QH is the flow from the
 Physical Limnology 397

\
\
““I
fly
50

40 -
KINEMATIC APPROACH

KINEMATIC APPROACH
I

30 -

20 -

0
1 01 0 2 °
05 1 2L
10 20 50 100 200 500 1000 2000 5000 10000

F3
FIG.29. The ratio 1 of epilimnion to hypolimnion discharge as a function of the Froude
number for flow towards a point sink. (After Lawrence and Imberger (1979).)

hypolimnion and g’ = (p2- p , ) g / p , . He also presented laboratory find-

ings that verify the concept of the critical point and the flow ratio once
the value of the Froude number was adjusted to compensate for the
non-radial flow at the critical section. Lawrence and Imberger (1979)
carried out further experimental verification of the above theory. Their
results and Woods’ are shown in Figure 29. Murota and Michioku
(1986a) generalized these results to a three-layer fluid.

3. Line Sink in a Linear Stratijication

The horizontal duct is the simplest configuration in which to consider

the initial value problem for a linearly stratified fluid. Initiation of the
sink discharge leads to an instantaneous potential flow solution

(8.14)

where 3 is the streamfunction nondimensionalized with respect to q, and

398 Jorg Zmberger and John C . Patterson

z and x are the vertical and horizontal coordinates nondimensionalized

with respect to H. However, as soon as the fluid particles begin to move,
they experience the buoyancy forces introduced by the stratification.
During these initial stages, inertia and viscous forces will be negligible
and the flow represents a balance between the unsteady vorticity term
and the baroclinic generation of vorticity as expressed by the streamfunc-
tion equation:
V21Vn + lVxx = 09 (8.15)
where 3 = V / q , z = z / H , x = x/H, t = tN, N is the buoyancy frequency
of the stratification, and V2 is the Laplacian operator.
Imberger (1970) used transform techniques to verify that close to the
sink, the solution to the duct problem was identical to the solution in the
unbounded fluid domain solved earlier by Koh (1966a). These authors
were able to show that for large time the horizontal velocity is

where = sin 8, 8 = arctan(z/x), r = (x’ + z ~ ) ” ~and , Jo is the Bessel

function of zeroth order. In the limit t + (x ~ and z fixed), this solution
reduces to
u(x, z , t ) - 4 2 ) ) (8.17)
which represents a singular jet flow at the level of the sink. The internal
waves responsible for changing the initial uniform flow in the duct into
the line jet must move at finite speed and hence steady state can never be
achieved in the model problem described by (8.15).
It was not until the work described in Kao et al. (1974) and Pao and
Kao (1974) that the role of the internal waves was fully understood.
These authors simplified (8.15) by omitting the x derivatives in the
Laplacian operator and showed that the double Fourier series describing
both the horizontal and vertical wave number space could be simplified to
a single sum of columnar disturbances.
1 2 “ 2
@= 2>
(4- + n = l -
nn
sin 2nnz H(nnx + t). (8.18)

McEwan and Baines (1974) independently defined these columnar

disturbances, but named them shear waves, in connection with a different
problem. They derived, by a modified method of steepest descent, the
behavior of the fluid in the vicinity of wave fronts. Their analysis showed
 Physical Limnology 399

that instead of remaining compact, as suggested by the step function in

(8.18), the shear waves are analogous to expansion waves in compressible
flow, and a wave travelling with a speed c,, = H N / n n has a frontal width
w,,= ~ ( ~ t ) ” ~ / n .
Physically, the initiation of the sink discharge thus leads to a potential
flow which, in turn, may be decomposed into a complete set of
progressive internal waves, containing the whole spectrum of vertical and
horizontal wave numbers. Once established this initial wave field
propagates away from the sink leaving a steady sink flow in its wake. As
Mahony and Pritchard (1977) point out the sole purpose of these waves is
to adjust the isopycnals so that they can return to a horizontal position at
steady state. McEwan and Baines (1974) proved that it is possible to sum
over the horizontal wave number space leaving an infinite series of shear
waves.
It may now be shown (Clarke, pesonal communication) that the long
wave limit of (8.15) is the Airy Equation (Cohen, 1979) and so shear
waves may be thought of as solutions to the Airy Equation. Clarke has
also recently shown that the inclusion of the nonlinear convective
acceleration term leads to the Korteweg-de Vries equation (Benney,
1966; Djordjevic and Redekopp, 1978) so that the nonlinear aspects of
shear waves can be thought of as analogous to an undular bore (Smyth
and Holloway, 1988).
The nonlinear aspects of a shear wave generation, distinct from the
propagating properties, were investigated experimentally by Monismith et
al. (1988). As the inertia of the flow increased, the observed shear wave
amplitude decreased relative to their theoretical values. When the
outflow was repeatedly pulsed, the strength of the average flow towards
the sink indicated a withdrawal layer thicker than that which would have
been found given either the average flow rate or the maximum
instantaneous flow rate.
Each shear wave has a vertical wave length A, = 2H/n and a wave
celerity C, = AnN/2n, so that the approach to the 6 function velocity
distribution given by (8.17) is brought about by the passage of successive
shear waves, each with a progressively smaller vertical wave length.
Steady state is achieved either by the dissipation of these waves or by
advection sweeping the waves back out through the sink.
First, consider the action of viscosity, which will dissipate the nth wave
in a time of O(Ai/v). However, in this time the wave has moved the
400 Jorg Imberger and John C. Patterson

distance A;N/2nv so that the non-dimensional thickness 6 / x becomes

-6 --_
An-
- O(Gr;1'6), (8.19)
x 2 x
where Gr, = N2x4/v2. Thus, a viscosity dominated layer has an x'13
upstream growth. This layer was first derived theoretically and verified
experimentally by Koh (1966a). The velocity profile is bell shaped with a
weak return flow immediately above and below the main layer.
Second, as a shear adds to its predecessor, the central velocity growth
is of order 2nxlNt. Advection will dominate once the induced velocity
O(v/An) is equal to the nth wave celerity O(AnN/2n). Successive waves
of order greater than n will no longer be able to propagate upstream.
This condition leads to a layer thickness
6
- = O(F;I2), (8.20)
X

where F, = q / 2 N x 2 and q is the total flow from both sides. Hence, in the
inertial case, the layer is of constant thickness converging only in the
potential flow region near the sink.
Combining (8.19) and (8.20) shows that the withdrawal layer changes
from one dominated by viscous forces to an inertially dominated layer at
a distance x , given by:

(8.21)

where RL = FL Grz3 and L is a reference length replacing x . The length x ,

was first recognized by Imberger (1972), and Silvester (1979) confirmed
its importance with a series of laboratory experiments.
The viscous flow given by (8.19) will no longer form the correct
upstream flow when the experimental tank is shorter than x , . Debler
(1959) described experiments in a short tank with an adjoining large
upstream reservoir. In such circumstances, pU2 is constant upstream, and
the motion is governed by the Euler equation together with the constraint
DplDt = 0. As demonstrated by Dubreil-Jacotin (1934), this leads to the
modified streamfunction equation
d p dH
VqJ' = gz
- 7 +7 , (8.22)
PodW d v
' (p/po)1'2qJand
where q ~ = A is the total upstream head. The first term
 Physical Limnology 40 1

on the right side of (8.22) represents the vorticity generated by the

buoyancy forces, while d H / d v ’ is the vorticity inherited from the
upstream boundary condition. Yih (1965) showed that if pU2 = constant
and the density varies linearly with z , then the right side of (8.22) is
constant and a solution may be found by separation of variables for
F = q / 2 N H 2> l/n. This solution consists of uniform flow upstream
necking down to a withdrawal layer near the sink. Above and below this
layer are weakly rotating eddies, which extend to infinity at F = l / n , thus
violating the upstream boundary conditions. This result was also derived
by Trustrum (1964). Kao (1970) constructed a numerical solution for
F < 1/n, along the lines of the method used by Huber (1960). Kao (1970)
showed that the layer thickness was given by (8.20), with a numerical
constant l h instead of the order sign. This value agrees with the
experimental results of Debler (1959).
In all experimental and prototype applications of the theory, some type
of rear wall exists that blocks the horizontal velocities. Imberger and
Fandry (1975) investigated the influence of such blocking by solving the
flow in a vertical duct. They demonstrated that internal waves were
indeed initiated but the vertical duct prevented their separation into
vertical shear waves. The internal wave group merely moved vertically
through a uniform flow, leaving behind the flow into the sink. This model
showed that the horizontal velocity varied linearly from x = 0 to x = L / 2 .
Imberger et al. (1976) solved the flow in a finite tank with a free
surface. The vertical constraints reintroduced shear waves that traversed
the tank, reflecting at the vertical boundaries. Their solution highlighted
a number of points:
The horizontal jet discharge varied linearly from the sink to the
end wall.
For the parameter R L < 1, the withdrawal layer thickness was
given by
6
- = 5.5 Gr,1’6 pr-1’6, (8.23)
L
with 64% of the layer lying above the sink center line and where Pr
is the Prandtl number.
For RL > 1, the flow was completely inertial and the withdrawal
layer thickness was given by
s
- = 4.OFZn. (8.24)
L
402 Jorg Imberger and John C . Patterson

(d) The withdrawal layer flow may be assumed to be steady through-

out the tank when it is steady at the end of the tank. This led to a
set-up time t = O(N-’Grg6) for RL < 1 and t = O(N-1FL1/2)for
R,> 1.
These conclusions, including the numerical factors in (8.23) and (8.24),
have been verified experimentally by Silvester (1979). Further, data given
by Bohan and Grace (1973) for the inertial flow regime may be
interpreted via the above formula, (8.24) yielding a numerical factor
of 3.4.
In the above, it was implicitly assumed that the Prandtl number was of
order unity. In many applications, especially where the water is stratified
with salt, this is not the case. Imberger et al. (1976) showed that the flow
structure was much more complicated for fluids with a large Prandtl or
Schmidt number. Basically, two complications arise. First, there are two
critical horizontal length scales x , , one where the viscous buoyancy
diffusive layer changes to a nondiffusive layer and one where the
influence of inertia becomes important. Second, the layer of thickness
order (Gr-’I6) is no longer stable but undergoes a secondary and tertiary
collapse. This leads to a hierarchy of layer types depending on the
relative magnitude of RL compared to the Prandtl (Schmidt) number Pr.
One of these layers has a thickness 6 / L = O(Gr-1’6R~s),experimentally
found by Mahony and Pritchard (1977), and one that is basically the
nondiffusive layer documented by Gelhar and Mascolo (1966) and
Walesh and Monkmeyer (1973). Silvester (1979) has experimentally
verified the existence of the flow hierarchy, but there remains a difficulty
in the regime Pr-’16 < R L < Pr-’13, where the potential energy of the flow
is zero over certain parts of the layer (see Imberger and Fischer (1970))
and where Mahony and Pritchard (1977) have observed hysterisis effects.

4. Point Sink in a Linear Stratification

The axisymmetric withdrawal layer was first investigated by Koh (1966a)

in the limit of very small discharge so that the flow was governed by a
viscous buoyancy balance. Lawrence and Imberger (1979) demonstrated
that this steady flow was established by the passage of a series of shear
waves originating from the sink at the initiation of the outflow. However,
in the axisymmetric case, these waves are cylindrical with amplitudes
depending inversely on the radius of propagation. These authors postu-
 Physical Limnology 403

lated, since the layer thickness was independent of the discharge, that the
thickness should be the same as in the two-dimensional case. This is
borne out by the data from Koh (1966a), who found
6
- = 5.8 Gr-116 pr-1/6 (8.25)
L
for the case where inertia was negligible. Spigel and Farrant (1984)
investigated the inertia-buoyancy limit and verified that the layer
thickness was given by the scale presented by Imberger (1980):
113
6=C@ 9 (8.26)

where C1= 1.58 (Bohan and Grace, 1973), 1.32 (Lawrence and Imber-
ger, 1979), 1.6 (Spigel and Farrant, 1984), and 1.32 (Ivey and Blake,
1985). Ivey and Blake (1985) also showed there was a layer thickness
6 = 4 . 2 (vQ
3)
(8.27)

that described selective withdrawal when inertia was small but species
convection was important.
The transition between these various layers is more complicated than
in the two-dimensional case. Ivey and Blake (1985) found a transition
parameter
(8.28)

but this parameter is now independent of distance from the sink so that
the inertia-free convective regime thickness (8.27) is also no longer
dependent on distance from the sink. A double criterion thus arises
depending on whether the convective layer can exist at all. These are
A: S > 3, S > (Gr Pr-2)1130 Inertial layer (8.26);
B: S > 3, S < (Gr P Y -~ )” ~ ’ Diffusive layer (8.25);
C: S < 3, S > (Gr Pr-2)1118 Convective layer (8.27);
D: S C 3, S < (Gr Pr-2)1118 Diffusive layer (8.25);
The difference here is that in Cases C and D, the inertial layer is thinner
than the convective layer and thus cannot exist. The first criterion on S is
a measure of the relative size of the inertial and convective layers, while
the second criterion on S differentiates the inertia and diffusive layers in
Cases A and B and the convective and diffusive layers in Cases C and D.
404 Jorg Imberger and John C . Patterson

5. Selective Withdrawal in a Rotating Stratijied Fluid

A homogeneous rotating fluid behaves analogously to a stratified

rotating fluid (Veronis, 1967), so it is not surprising that selective
withdrawal has its counterpart in a rotating fluid. Long (1956) obtained a
solution for steady flow of a rotating inviscid fluid into a point sink
located at the axis of the rotation to an infinitely tall cylinder. Using the
full nonlinear equations (Yih, 1965), he found necking near the sink with
an annular eddy around the axisymmetric layer. This eddy elongated to
infinity as the Rossby number

(8.29)

(where Q is the discharge, b the radius of the cylinder, and f the Coriolis
parameter) approached 0.261. Shih and Pao (1971) carried out experi-
ments and found that the withdrawal layer formed with a radius

(8.30)

whenever Ro < 0.261. Motivated by these experiments, Pao and Shih

(1973) constructed a slip line solution and found excellent agreement with
the prediction from (8.30). The viscous dominated solution was also
derived by Pao and Kao (1969) and they found in this case that the layer
thickness was given by
113
&=4.1(7) , (8.31)

the numerical fctor being obtained both theoretically and experimentally.

The transition between the two does not appear to have been studied,
but by analogy with a purely stratified case, the transition should depend
on the parameter

R --.Q (8.32)
-YL
For Rf > 1, (8.30) applies; for Rf < 1, (8.31) is relevant.
The above formulae predict a vertical withdrawal layer whereas in the
stratified case, the withdrawal layer is horizontal. Investigations are
underway (McDonald, personal communication) in an effort to reveal the
transition of the flow as the parameter f /N is varied from 0 to infinity.
 Physical Limnology 405

Monismith and Maxworthy (1988) performed experiments on the estab-

lishment of a selective withdrawal layer in a rectangular channel under
conditions of strong stratification and weak rotation ( f l N = 0.8- 0.25).
Their observations show that the initiation of the outflow resulted in a
Kelvin shear wave propagating cyclonically around the perimeter of the
tank, leaving in its wake an anti-cyclonic withdrawal layer. The thickness
of the withdrawal layer was, however, not influenced by the rotation and
was given by (8.26).

In summary, a great deal is known about selective withdrawal and most

situations can be predicted. There are three exceptions, however. First,
arbitrary stratification leads to premature withdrawal because of the
higher mode response, but no predictive procedure exists. Second, the
influence of sills or contractions has received only preliminary treatment
(Kao, 1976) and much work remains to be done. In particular, transitions
that occur through a contraction where the withdrawal layer changes
from a subcritical condition to supercritical and back to subcritical with
an associated hydraulic jump would appear to be extremely important to
mixing in lakes and the ocean. Third, the transition flow from a fully
rotating fluid to a fully stratified fluid has not been studied in detail.

Ix. Inflow

Reservoirs and lakes are supplied from rivers; as the river water meets
the relatively stagnant reservoir water, it usually encounters water of a
slightly different temperature, salinity or turbidity. These variables lead
to an inflow that is either lighter or heavier than the surface water of the
lake. When the inflowing water is very much heavier (Hebbert et al.,
1979; Ford et al., 1980) the river water enters the reservoir, and plunges
to the bottom of the lake. Alternatively, examples of where the inflow
water was not sufficiently heavy to take it all the way to the bottom are
given by Howard (1953), Wunderlich and Elder (1969), Ford (1978),
Ford and Johnson (1981), Fischer and Smith (1983), and Imberger
(1985). In all these cases, the inflowing water plunged beneath the lake
water at the entrance then flowed down the drowned river valley,
entraining lake water as it moved downstream. At some depth, the total
entrainment made the underflow neutrally buoyant relative to the
immediately adjacent water, and the underflow became an intrusion
406 Jorg Zmberger and John C . Patterson

FIG. 30. Schematic of inflow possibilities: overflow, plunge-point, intrusion, and

underflow.

(Figure 30). Last, there is the case where the inflowing water is lighter
than the resident surface water and the river overflows the lake as a
buoyant surface flow (Figure 30). The dynamics of such an overflow are
similar to the situation where a source of fresh water enters an estuary or
coastal environment (Chen, 1980; Ookubo and Muramota, 1981; Chao,
1988; Luketina and Imberger, 1987, 1988).
The surface buoyant jet occurs only rarely in a lake in tropical and
temperate regions but is normal in high latitude cold lakes, where such
inflows induce thermal bars (see Section VII). The dynamics of these
flows is a subject in itself and has been extensively treated by Chen
(1980), Sargent and Jirka (1982), Fischer and Smith (1983), O’Donnell
and Garvine (1983), Chu and Jirka (1985), Arita et al. (1986), and
Luketina and Imberger (1987, 1988), with an excellent review of the
frontal region being given by Simpson (1982). The reader is referred to
these articles for further information on the overflow case.
Consider the case where the density of the inflowing water is somewhat
heavier than the resident surface water in the lake and the river water
will plunge down the river bed until it reaches neutral depth. The
 Physical Limnology 407

propagation down the inclining plane of the leading head of the gravity
current has been investigated by Stacey and Bowen (1988) and Wilkinson
and Wood (1972). However, in most cases, the inflowing discharge is
sufficiently steady to allow the underflowing river to be analyzed as a
steady or quasi-steady gradually varying flow.
In this case, a balance is struck in the underflow such that bottom
friction, interfacial entrainment and inertia retard the flow, and the
downslope component of gravity accelerates it. This momentum balance
is influenced by the slope and flatness of the river bed and the rotation of
the earth. A discussion of the latter influence is omitted here.
As mentioned above, most authors have assumed the underflow
discharge to be steady, and if it is further assumed that the underflow is
confined laterally by the drowned river valley, a gradually varying flow
analysis yields satisfactory results (Ellison and Turner, 1959; Savage and
Brimberg, 1975; Hebbert et al., 1979; Akiyama and Stefan, 1984). The
main conclusion reached in these studies (see Fischer et al. (1979)) is that
the flow quickly adjusts, so that as it propagates downstream along a
particular bed slope, the Froude number of the underflow becomes a
constant equal to the normal flow Froude number F,, where the Froude
number is given by

where u is the velocity of the downflow, g' is the modified acceleration

due to gravity between the lake and underflowing water, and d is the
hydraulic depth of the underflow.
For the case where the density difference is derived from particulate
matter (turbidity current), the fallout or accumulation of particles must
be accounted for. However, provided the rate of change of density due to
deposition or erosion is slow compared to the time it takes to travel a
number of depths of the underflow, it is possible to retain the gradually
varying flow assumption, so that (9.1) is still valid even if only locally
(Akiyama and Stefan, 1985). Fischer et al. (1979) solved the vertically
integrated momentum and mass equation to derive an expression for the
normal Froude number:
408 Jorg Zmberger and John C. Patterson

where T is the top width of the underflow, E is the entrainment

coefficient, P is the perimeter of the underflow, h is the underflow depth,
h, is the depth of the underflow centroid below the interface, a is the area
of the underflow, 9 the valley slope, and CD is the channel bottom drag
coefficient.
The entrainment coefficient may be determined by applying a model
similar to the one used for the surface layer (see Section IV), but as
discussed in Fischer et al. (1979), the coefficients of the model may be
different due to the very different velocity profiles within the underflow-
ing current compared to the velocity distribution in the wind driven
surface layer. The entrainment at the interface is energized by the shear
there and by the turbulence imported from that generated at the river
bed. For mild slopes (9 = Hebbert et al. (1979) have shown that
shear production was negligible and that E had a value of 1.9 x
The location of the plunge point can now be found by neglecting the
local entrainment. This is done by substituting the entrance flow and
density difference into (9.1) and equating this to the normal Froude
number evaluated from (9.2). This will yield the first estimate of the
plunging depth h l , which may be improved by accounting for the
non-hydrostatic pressure distribution near the plunge point (Wilkinson,
1972; Hebbert et al., 1979; Jain, 1982; Akiyama and Stefan, 1984, 1985,
1987). This leads to a relationship

which implies an energy loss at the plunge point.

The three equations (9.1 to 9.3) can now be solved together for d and
H with only the assumption of no entrainment at the plunge point.
Comparison with field data is excellent (Hebbert et al., 1979).
When the slope becomes steeper (F, larger) or the inflowing momen-
tum is increased, the energy loss implied by (9.3) can no longer be
neglected and the associated entrainment at the plunge point will
invalidate the above analysis and the theory must be modified as
suggested by Britter and Simpson (1978) (see also Luketina and Imberger
(1987) and Jirka and Arita (1987)). The analysis for the inclusion for such
entrainment has been carried out by Akiyama and Stefan (1987), but no
data exist to fix the amount of entrainment except for the buoyant
overflow documented in Luketina and Imberger (1987, 1988). There are
some numerical simulations available (Farrell and Stefan, 1986; Buchak
 Physical Limnology 409

and Edinger 1984). However, these all use closure schemes so that the
estimates of entrainment would be overestimated (Franke et al., 1987).
For a divergent river valley with a relatively flat bottom, the flow may
separate from the side walls. Johnson et al. (1987a,b) observed six
different flow regimes depending on the entrance aspect ratio, the Froude
number, and the angle of divergence. These are schematically re-
produced in Figure 31. In general, the higher the Froude number, the
closer the flow is to a simple, entering jet. Once separated from the side
wall, the inflow behaves as a free jet and Johnson ef al. (1988) found that
the entrainment was much higher for such separated flows and could be
estimated from simple jet analysis (Fischer et al., 1979; Rodi, 1982).
The case of inflow into a lake with a flat bed morphology is similar to
the large divergence case and is characterized by a central momentum jet
followed by a radially spreading buoyant plume, originating from a
virtual origin. The virtual origin occurs at a distance from the inflow of
order M3/4B-112(Fischer et al., 1979; Luketina and Imberger, 1987),
where M is the entrance momentum flux and B is the corresponding
buoyancy flux. Luketina and Imberger (1987, 1988) presented an analysis
and a comparison with field data of a buoyant overflow, whereas
Hauenstein and Dracos (1984) presented a similar analysis and a
comparison with laboratory data for the case when the inflow is denser

SMALL DIFFUSER ANGLE LARGE

S T R O N P BUOYANCY WEAK

SURFACE III-IMMhA PLUNGELINE (REGION)

A1 A2 A3 B C E

STREAMLINES

FIG.31. Principal regimes of negatively buoyant flow into horizontal diverging channel.
(After Johnson er al. (1987a).)
410 Jorg Zmberger and John C. Patterson

than the lake water. Recently, Tsihrintzis and Alavian (1987) developed
a further integral model similar to that of Hauenstein and Dracos (1984)
for three-dimensional flow down an inclined plane.
Once the inflow reaches a depth of neutral buoyancy, the inflowing
water leaves the river valley and intrudes horizontally into a stratified
lake. Imberger (1985) presents field data from the Wellington Reservoir
that illustrate this lift-off process. The dynamics of the intrusion appear to
have only been thoroughly investigated for two-dimensional flow. In this
case, an intrusion entering a linearly stratified ambient water body will
have a length and width as summarized by Imberger et al. (1976) with the
propagation length being given by

i
0.44 LRli2t’, t’ 5 R, (9.4)
I = 0.57 LR2/3t’5i6, R -=t’ <PI?, (9.5)
CLR 3/4f’3i4, P? < t‘ < R-’, (9.6)
where the coefficients in (9.4) to (9.6) were determined from experimen-
tal data by Manins (1976) and Maxworthy (1980), but where the constant
C in (9.6) is still unknown. In these formulations, the parameters are
defined as
R = F Gr’I3, (9.7)

N2L4
Gr=- (9.9)
Y2 ’

(9.10)

and I is the length of the intrusion at a time t after initiation of the flow.
As already discussed for intrusions formed by differential deepening,
(9.4) to (9.6) are not valid when the intrusion approaches the rear wall
(Fischer et al., 1979).
Darden et al. (1975) showed that the accompanying velocity structure
due to such an intrusion consists of multiple reversing jets above and
below the entering intrusion, and Maxworthy (1980) demonstrated that
simple intrusions can degenerate into solitons if the volume flux to the
intrusion is terminated.
The second main case in a lake is where the downflow encounters a
 Physical Limnology 411

sudden change of density at a density interface. If the water below the

density interface is heavier than that of the downflowing water, an
intrusion will form along the interface. If the interface is very thin
compared to intrusion thickness, the intrusion behaves as two separate
gravity currents, the depth of the intrusion relative to the interface
adjusting so that the speeds match (Holyer and Huppert, 1980; Britter
and Simpson, 1981). Faust (1981) gives the speed of propagation V as

where d , is the depth of the upper layer, d2 is the depth of the lower
layer, p1 is the density of the upper layer, p2 is the density of the lower
layer, hl is the depth of the upper fluid above the intrusion, h2 is the
depth of the lower layer below the intrusion, and Ap = p2 - p,. Faust
(1981) generalized the same analysis and showed that a similar expression
could be derived for an intrusion into a three-layer fluid. However, in this
case, he found that internal waves generated at the interfaces could lead
to a slow-down of the intrusion.
The above intrusion dynamics have only considered laminar flow.
However, in lakes, the downflow is strongly turbulent (Imberger, 1985)
and it is important to assess the collapse of this turbulence as the
intrusion enters the lake. Lin and Pao (1979) and later Hopfinger (1987)
have reviewed this subject and concluded that the collapse takes place in
times of O(N-') and that the vertical motions decay most quickly,
leaving, in the absence of additional shear, two-dimensional turbulence
with a vorticity vector orientated predominantly vertically. The boundary
mixing experiments of Browand ef al. (1987) suggest that the decay is
extremely rapid and that we would not expect to find turbulence within
the intrusion at a horizontal distance from the lift-off point much greater
than the thickness of the intrusion. This has been confirmed by Roberts
and Matthews (1987) who injected a horizontally neutral buoyant
momentum source from a round nozzle into a linearly stratified fluid and
found that the mean shape of the jets spread horizontally at a distance of
M"4N-1",where M is the momentum flux of the jet and N is the
buoyancy frequency of the ambient water. Although these authors did
not measure the turbulent flux directly, they inferred that the largest
turbulent eddies became anisotropic at the same distance.
Now in the case of an underflow, the Froude number based on the
density difference between the underflow and the neighboring ambient
412 Jorg Imberger and John C . Patterson

water becomes infinity since g‘ tends to zero at the point of neutral

buoyancy. However, the inertia of the underflow would carry the fluid
through the lift-off into the intrusion, and a finite sized intrusion based on
an inertia buoyancy balance would form with a thickness (Manins, 1976)
112
6 = 4);( (9.12)

for the two-dimensional flow case. Assuming that the collapse starts at a
distance order (MN-2)”3, which is the equivalent result for the two-
dimensional flow, the turbulence would commence the decay at a
distance O(QN-l”) or one intrusional layer thickness into the lake.
Given these two pieces of evidence, one would expect not to find
turbulence in an intrusion. However, recently Imberger (1987) has
shown, by carrying out temperature microstructure measurements, that
the shear from the intrusions together with background shear from
internal wave activity can sustain a low level of turbulence. In a further
analysis of the same data, Imberger (1988) has shown that the turbulence
was at the point of decay, with

Fr, = (5) 112

= 15ll2= 3.87. (9.13)

Thus, even though the gradient Richardson number associated with

intrusion was of the order of 11, the turbulence remained viable,
although at the point of fossilization (Rohr et al., 1988).
In summary, we have seen that the inflow dynamics may be con-
veniently partitioned into three separate problems: the underflow, the
plunge point, and the intrusion dynamics. The inflow of the heavier water
mass into a drowned river valley appears to be adequately described by
existing gradually varying flow theories, although the entrainment over
steep slopes where shear production is the dominant source of energy for
entrainment has not been verified in the field. Equally, under these
conditions, we may expect severe entrainment at the plunge point, which
so far defies parameterization. The interaction of the inflow with a
resident internal wave field (Fischer and Smith, 1983) or with a surface
mixed layer deepening (Imberger , 1985) can introduce enhanced mixing
and must be treated on a case by case basis for each lake.
The most important problem remaining concerns the mixing associated
with inflows, both directly in the intrusion and also, as discussed in the
next section, in the lake as a whole.
 Physical Limnology 413

X. Mixing below the Surface Layer

Mixing below the surface layer is generally patchy and sporadic, with
individual events occupying no more than tens of cubic meters and
remaining at full intensity for no more than a few minutes (Imberger,
1985). The two central questions remain: What is the efficiency of energy
conversion at turbulent sites and what is the patch statistical distribution
(Imberger and Hamblin, 1982)?
No direct measurements of the local value of the vertical diffusion
coefficient K , exist, but many investigators have taken advantage of the
closed nature of the lake basin and obtained estimates of the vertical
diffusion coefficient averaged over the whole basin (Kullenburg et al.,
1974; Jassby and Powell, 1975; Imboden and Emerson, 1977; Imboden et
al., 1977; Robarts and Ward, 1978; Weiss et al., 1979; Lewis and Perkin,
1982; Imboden et al., 1983; Imboden and Joller, 1984; Sanderson et al.,
1986; Colman and Armstrong, 1987; Lake Biwa Research Institute, 1987;
Wuest, 1987; Wuest et a/., 1988). The data confirmed (Imberger and
Hamblin, 1982) that the lake-wide averaged vertical exchange coefficient
fluctuates between 2 x m2 s-l (near molecular) during periods of
very strong stratification and weak winds (Wuest, 1987) to values as high
as m2s-l during periods of very weak stratification in tropical lakes
(Lewis and Perkin, 1982). It has also been shown (Ward, 1977;
Straskraba, 1980; Wuest, 1987) that the vertical diffusivity depends on
the buoyancy frequency, the area of the lake and the depth of the lake.
Wuest (1987) attempted a correlation of the rate of mixing with the
Wedderburn number (2.5) and the buoyancy frequency, writing the
vertical diffusion coefficient as

(10.1)

He found most data sets required a value of cy in the range of

0.3 < LY < 0.9. This may be compared with oceanic variations, summar-
ized by Gargett and Holloway (1984), where 0.25<a<0.5. The
dependence on the Wedderburn number is shown in Figure 32, re-
produced from Wuest (1987), and indicates that if W ranges from
between 200 and 0, the vertical diffusivities range from around
10-6m2s-' to around 10-*m2s-'. In computing the value of the
Wedderburn number W, Wuest (1987) used the depth of the epilimnion,
but does not give details of the definition used. However, judging from
414 Jorg Zmberger and John C. Patterson

106 10-4 10.2

K, (rn21s)

FIG.32. A comparison of K , in Sempachersee for four different periods. The value of

K , was calculated using traces and are typical of the three regimes discussed in the text:
0.5 < W < 200, 0.1 < W < 0.5, and W < 0.1. (After Wuest (1987).)

his profile shapes, this would have been close to the depth of the seasonal
thermocline, which would make W close to LN. His W is really an
indicator of the stability of the seasonal thermocline and not the surface
layer.
We saw in Section I1 that in the majority of cases a lake is adequately
approximated by a three-layer stratification, so there are two numbers, W
and LN , which determine its response to a wind force. Small W and large
LN lead to a strong surface layer response, and small LN causes the main
thermocline to tilt with the associated stronger mixing in the hypolim-
nion. Three examples provide an illustration. In Figure 33, data from the
Canning Reservoir show a case where W =0.1 and LN=288. The
dissipation levels are close to zero everywhere, except in the top two
meters, where a very weakly stratified surface layer resided. During the
time of this profile, nearly 200 microstructure casts were taken over a
period of 10 days and only one profile showed the slightest sign of
turbulence in the hypolimnion. By contrast, Figure 17f depicts a
microstructure profile at W = 2.1 and LN = 0.24, whereas in Figure 24c
the conditions were such that W = 4.5 and LN = 0.44; in both cases the
whole profile had become turbulent with dissipation values in the
hypolimnion reaching m2 s - ~ . These data are presently being
analyzed more fully, but this preliminary analysis, together with the
evidence from Wuest (1987), strongly suggests that the response of the
 Physical Limnology 415

TEMPERATURE F) fldr ( a m ) DISSIPATON E (mas31

FIG.33. Microstructure taken in the Canning Reservoir during a period of large Lake
number L , = 288. (a) Temperature. (b) Temperature gradient. (c) Dissipation as calculated
from the Wigner-Ville technique (sticks) and the Batchelor spectra fitting technique (bar
chart). Note that there is no turbulence below the top 2 m.

seasonal thermocline, which determines the deep mixing in a lake, is in

general determined by the value of the Lake number LN.
The variability within the hypolimnion must depend on some empirical
relationship similar to that given by (10.1). Weinstock (1984) has
determined the following theoretical formulation:
&
K, = a (10.2)
N2+ k;u2 '
where E is the dissipation of turbulent kinetic energy, N is the buoyancy
frequency, ko is the wave number of the energy bearing eddies, u is the
root mean square velocity of the motion in the turbulent patch, and a is a
constant between 0.2 and 0.8 (Imberger and Hamblin, 1982). It is clear,
however, that the dependence on N is too strong in (10.2) and is not
supported by the empirical measurements of the average diffusion
coefficient (10.1). However, it would be interesting to investigate whether
the averaging process could decrease the dependency from N-' to N-'.
416 Jorg Imberger and John C . Patterson

The following overall scenario is appealing: as the wind, inflow, or

artificial mixing device (e.g., an aerator) start up, they induce seiching of
the thermocline (mode 1 response in a three-layer system). This seiching
provides a background shear within which internal waves generated by
differential deepening and by leakage through the base of the surface
layer increase the frequency and intensity of turbulent patches. This leads
to an increase in the buoyancy flux and turbulent kinetic energy
dissipation. This is a much simpler picture than in the ocean where the
background shear arises from a host of motions.
The Lake number can be extended easily to account for these three
major disturbances to read

(10.3)

where Qo is the air flow rate of the aerator, Qi is the river inflow rate at
the point where the intrusion forms, U is the inflow velocity of the
intrusion, and zi is the height of the intruding intrusion. All other terms
are defined in Sections I1 and XII. The additional terms are the
additional disturbing moments offered by the bubble flow pog2nQ$'3~g3A0
(Section XII) and the river momentum flux pOUQiacting at a moment of
(zg- zi) (Section XI). Whenever LN is small, the seasonal thermocline
may be expected to respond more actively, thus leading to higher
transport rates.
The origins of the mechanisms leading to internal mixing are numer-
ous, consisting of a range of instabilities, wave-wave, wave-shear,
Kelvin-Helmholtz billows, Holmboe instabilities, critical layer absorp-
tion, double diffusion, bores, and hydraulic jumps. These mechanisms
have recently been thoroughly reviewed and the reader is referred to
Hopfinger (1987) and Thorpe (1987). Mixing in the metalimnion is also
reviewed by Gregg (1987). For a review of double diffusion, the reader is
referred to Turner (1985).
The nature of the turbulence in a lake can be determined from the
turbulent Froude versus Reynolds number diagrams (Gibson, 1980, 1982,
1986, 1987a,b, 1988; Imberger and Boashash, 1986; Imberger, 1988). The
Froude number is defined

(10.4)
 Physical Limnology 417

and the Reynolds number is given by

(10.5)

where E is the dissipation of the turbulent kinetic energy, g' is the

acceleration due to gravity derived from the density anomaly within the
turbulence, and L, is the displacement scale (Imberger and Boashash,
1986). Turbulence close to the line
( T )= 3.87
Fr, = EL,
112

(10.6)

is at the point of fossilization (Itsweire et al., 1986; Imberger, 1988),

whereas active turbulence lies in the region Fr, > 1 and Re > 1. In Figure
34, we reproduce data derived from turbulence measurements from 10
different physical mechanisms operating in lakes. The mixed layer
turbulence is active; turbulence in the hypolimnion is at the point of
decay. It is necessary, and possible, to design a field experiment aimed at
correlating the location of the data points in the Froude versus Reynolds
number phase diagrams with the values of W and LN.Given that the
mixing efficiency (the flux Richardson number) can be correlated to the
location of the data in Figure 34, this would lead to an operational model
for the prediction of vertical transport in a lake (Imberger, 1988).
So far we have discussed only mixing internal to the hypolimnion.
There is evidence in the ocean (hey, 1987; Garrett and Gilbert, 1988;
Gregg and Sanford, 1988) that the energy input due to tidal motion
exceeds that appearing in the water column at depth, suggesting that
dissipation in the benthic boundary layer is a major sink for this energy.
Lueck and Osborn (1985) presented evidence that close to the bottom in
a submarine canyon, dissipation values increased considerably. h e y and
Boyce (1982) showed that a benthic boundary layer existed at the bottom
of Lake Erie.
In Figure 35, we show the development of turbulence in the water
column of the Harding Reservoir in response to a surface wind stress on
27 November 1988. Early in the morning the meteorological conditions
were warm and calm. At 1000 hours a wind with a speed ranging between
2 and 3ms-' commenced and continued until 1400hours, leading to a
Wedderburn number of 0.05 and a Lake number of 2.2. The sequence of
profiles (all extending from the bottom to the lake surface) shows the
establishment of turbulence both with the hypolimnion and in the benthic
418 Jorg Imberger and John C . Patterson
 Physical Limnology 419

boundary layer with dissipation values as high as 6 x m2s - ~in the

bottom 1 m of the profile (Figure 35f). This data set is presently being
analyzed but was included as it offers the first conclusive evidence of the
development of turbulence in the benthic boundary layer of a lake.
It is tempting to carry out a budget of the turbulent kinetic energy. For
instance, the data shown in Figure 33 indicate dissipation in the surface
layer (depth of 2 m) that almost exactly balances the energy input CLu:.
However, by comparison, integration of the total dissipation per m2 of
the lake’s surface for the profile shown in Figures 17 and 24 greatly
exceeds that introduced at the surface. Sufficient measurements are
required to account for the very strong variability (see Section VI) of
wind stress over the lake surface.
Boundary mixing can be sustained by two major mechanisms. First,
internal wave seiching will cause a velocity shear at the boundary and set
up a turbulent boundary layer over the rough lake bottom (see Imberger
and Hamblin (1982) for a review). Numerous experiments (Ivey and
Corcos, 1982; Thorpe, 1982; Phillips et al., 1986; Browand et al., 1987;
Ivey, 1987) have modeled this mechanism with a grid stirring the water
adjacent to a vertical or sloping wall in a stratified tank. These
experiments all produce a turbulent boundary layer close to the grid,
which then initiates intrusions that flow horizontally into the tank. The
mean motion induced by these intrusions is such that the density
structure in front of the intrusion changes in a way which mimics an
enhanced vertical diffusion adjustment. Ivey (1987) has summarized this
work and inferred an ocean averaged diapycnal eddy diffusivity of

FIG.34. The Froude versus Reynolds number plot containing the data from microstruc-
ture segments with well-defined spectral components protruding above a step gradient form.
All dissipations used in this plot were estimated by fitting a Batchelor spectrum to the
spectra computed for the whole segment.
0 Bubble plume thermal PT
0 Thermal falling in a weak ambient stratification
0 Thermal impinging on a sharp interface
A Penetrative convection
V Surface buoyant plume roller region
V Surface buoyant plume wake
x Well mixed surface region
+ Strong stratification, strong shear
Strong stratification, weak unknown shear
+ Strong stratification, no known shear
* Gravitational underflow
A Penetrative convection thermals impinging on an interface
0 Intrusion flows.
 0-

3-

-
r 6-
t
0 .

9- -19

26 26-30
TEMPERATURE (“C)
0
n/dz (“Ch)
3010”0
- 10’
& (m2/s3)
10.’ lo4
112

TEMPERATURE (“C)
0
dT/dz (“Urn)
--
3010’0
hw
lo4
Ws3)
10‘ loJ
’I2

h i (1406)
7r

I -3

I,
w
D
. -

6k
-
E
I

- 9

I
’ZLi4 ‘
-
I
- l L
26 28 -30 0 3d10’0’”’;06 ‘;‘Iz
TEMPERATURE (“C) dTidz ( T i m ) EuIx ( m Z / s 3 )

FIG.35. Examples of the development of turbulence in the Harding Reservoir at an

extremely low Lake number L , = 2.2. All profiles extend from 0.2 m above the rocky
bottom to the surface. (a) Temperature profile at 1127 hours. (b) Temperature gradient
profile at 1127 hours. (c) Dissipation. Sticks: Wigner-Ville estimate; bar chart: Batchelor
spectra fitting estimate. (d) As (a) at 1317. (e) As (b) at 1317. (f) As (c) at 1317. (9) As (a)
at 1406. (h) As (b) at 1406. (i) As (c) at 1406.
 Physical Limnology 421

3x m2s-l for the interior of the ocean. This result was derived from
the formula
Kl
K , =- (10.7)
BL ’
where K is the diffusion coefficient in the benthic boundary layer, 1 is the
turbulent benthic boundary layer thickness, 8 is the bottom slope, and L
is the basin scale.
The other mechanism is breaking internal waves (Cacchione and
Wunsch, 1974; Murota and Hirata, 1979; Murota et al., 1980; Thorpe,
1987; Garrett and Gilbert, 1988; Ivey and Nokes, 1989; Wallace and
Wilkinson, 1988). Internal waves shoal and break as they propagate over
a sloping bottom, especially if the bottom slope is at the same angle as
the rays of the internal wave field (a critical slope). Ivey and Nokes
(1989) found the turbulence in the benthic boundary layer resulting from
the breaking waves to be sustained by shear leading to Richardson
numbers as low as 0.01. For a critical slope, the steady state boundary
layer thickness resulting from the wave breaking was given by
1 = 5I;, (10.8)
where I; is the internal wave amplitude.
The mixing efficiency within this active boundary layer measured as the
flux-Richardson number (buoyancy flux divided by production) reached a
maximum of 20%. The diffusion coefficient K , can now be estimated
from (10.7) and the estimate of the diffusion coefficient in the benthic
boundary layer (Ivey and Nokes, 1989):
K = 0.09 cot2, ( 10.9)
where I; is the wave amplitude and w the wave frequency. Thus, given
(10.7)-(10.9), a recipe for estimations of the vertical diffusion coefficient
due to boundary mixing is now available. However, application still
requires a knowledge of the internal wave spectrum in a lake as a
function of L N, in order to determine what percentage of the bottom is at
critical slope. This is an urgent priority. Further, given the importance of
the first mode seiche, it is not clear that boundary mixing plays the same
important role in lakes as it appears to do in the ocean.
In summary, a great deal of progress has been made in the area of
hypolimnetic mixing. Recent measurements have conclusively shown the
turbulence level in the interior of the lake increased as LN decreases and
422 Jorg Zmberger and John C . Patterson

also that the effective basin average diffusion coefficient correlates to N- "
where a is approximately 1. The role of the benthic boundary layer and
the turbulence due to shear and internal wave breaking has been
quantified in laboratory experiments and these results are now ready to
be tested in field experiments with the data set from Harding Reservoir.
This should determine the functional relationship between the proportion
of transport in the interior compared with that in the benthic boundary
layer as a function of the Lake number LN .

XI. Modeling

The previous sections have given an overview of the seasonal behavior

of lakes and reservoirs, and discussed in detail the major processes that
determine their density structure. The complex interactions of these
processes which produce the overall thermal characteristics are best
studied with computer simulations. The need for these detailed descrip-
tions has been motivated by lake management requirements, which has,
in the last few decades, led to the development of a variety of modeling
techniques for stratified lakes and reservoirs.
Much of this development has occurred under the assumption of
one-dimensionality, where vertical motions are inhibited and transverse
and longitudinal variations are quickly evened out. Even with this great
simplification, it is difficult to model the interaction of a number of
complex processes to predict a density structure, and a number of models
of varying levels of complexity and success have been produced. There
has also been, to a lesser extent, some development of two- and
three-dimensional stratification models; the increasing complexity and
computational requirements have severely limited this development. The
question of circulation models is a separate issue and is not addressed in
this review.
The early one-dimensional models of stratification concerned the
solution of the one-dimensional heat transport equation. Perhaps the first
of these models was the Dake and Harleman (1969) formulation, in
which molecular diffusivity was assumed to be the vertical transport
mechanism that redistributed the heat input by the surface heat fluxes. In
a preliminary version of later developments, this model generated a
surface mixed layer by redistributing density instabilities produced by
surface cooling. The true diffusivity models followed: Orlob and Selna
 Physical Limnology 423

(1970), Huber et al. (1972), Sundaram and Rehm (1973), and Markofsky
and Harleman (1973) all developed models that simulated the vertical
transport of heat in the epilimnion by means of a diffusion-like process,
characterized by an eddy difhsivity E,. The definition of E, was the
major difference in these models. For example, Huber et al. (1972)
assumed a constant value of E, , calibrated from a data sequence. Orlob
and Selna (1970) used a constant value of E, in the epilimnion, with a
value exponentially decaying with depth in the hypolimnion, also
calibrated against data. These models were poor representatives of the
actual processes. Sundaram and Rehm (1973) incorporated some of the
physics into the representation by defining an E, that depended on the
stratification and the wind, through an overall Richardson number.
Similar formulations for E, have been suggested by, for example,
Thomas (1975) and Witten and Thomas (1976). Henderson-Sellers (1976)
has further developed these ideas for deriving E, for thermocline models
based on boundary layer theory and a mixing length turbulence closure
scheme.
Although the early models based on diffusivities were moderately
successful and, being relatively insensitive to the value of E,, were
simple to calibrate, Harleman (1982) showed that this was the result of
the dominance of other processes for the vertical transport of heat, and
the insensitivity may not occur in all cases. Further, in all cases,
calibration was required and transfer of the model from one lake or
reservoir to another required recalibration. Even within one lake,
conditions outside the range of calibration could not be modeled with
confidence. More importantly, the description of the turbulent transport
processes by a single bulk eddy transport coefficient meant that the
effects of the individual processes, such as turbulence generated by the
wind, convective mixing, internal waves, boundary mixing, and basin
scale seiching for example, were integrated into a single parameteriza-
tion. The coefficient was particular to both the lake and the combination
of processes acting at the time, and could have no real physical
interpretation. Because of its integrated, non-physical nature, there was
no independent means of verifying the parameter value or dependence.
Finally, the application of these diffusivities to other lake properties, for
example salinity or dissolved oxygen, was not necessarily correct. Fischer
et al. (1979) pointed out that the distribution of different properties may
arise from different combinations of physical processes, and the same
diffusivity may therefore not be appropriate.
424 Jorg Imberger and John C. Patterson

Recent developments for the incorporation of specific processes, for

example Henderson-Sellers (1984a, 1985), have minimized some of these
difficulties, but if an understanding of the lake behavior in terms of its
short time scale dynamics is required, models based entirely on diffusivity
formulations will be inadequate.
This conclusion led to a number of investigators taking a different
approach to modeling the epilimnion region of the reservoir or lake. The
early developments were based on the Kraus and Turner (1967) model of
thermocline formation in the ocean. As detailed in Section IV, this
provided a deepening rate based on a balance between the turbulent
kinetic energy (TKE) produced at the surface by the stirring of the
surface wind and the energy required at the thermocline to entrain water
from below the thermocline. The first lake models to incorporate a
parameterization of this process were those of Stefan and Ford (1975)
and Harleman and Hurley-Octavio (1977). Tucker and Green (1977)
utilized a version of the Kraus-Turner model for the epilimnion with a
diffusive model with a diffusivity that depended on wind speed and
stratification in the hypolimnion. Tucker and Green also included a
dependence on fetch length of the TKE made available for mixing.
The model development by Harleman and Hurley-Octavio (1977) was
expanded on by Hurley-Octavio et al. (1977) and Bloss and Harleman
(1979). The latter incorporated an effective efficiency of the transfer of
TKE from the surface to the base of the mixed layer, based on a
Richardson number formula. This incorporated the storage effects and
dissipation. This model formed the base for the standard MIT model.
These models all solved for the vertical heat transport on a fixed
discretization of the vertical direction; the lake was effectively split into a
number of horizontal layers of uniform property and fixed thickness.
Vertical transport was then achieved by transport across these boundaries
by mixing from the mixed layer model, diffusion from the eddy diffusion
model in the hypolimnion, or vertical advection as the result of inflow
and outflow. This latter process involved the calculation of vertical
velocities, and consequently, potential problems of numerical diffusion
arose.
A different approach was taken by Imberger et al. (1978) in the first
version of the model DYRESM. Here, the layers were Lagrangian in
character, of variable thickness, and able to move vertically. Thus, inflow
and outflow simply changed the thickness of the layers affected, with
those above moved vertically to accommodate the change in storage. The
TKE resulting from penetrative convection was formally incorporated in
 Physical Limnology 425

this version. Later versions of DYRESM also included the effects of

interfacial shear, parameterization of the TKE storage, an eddy
diffusivity for the hypolimnion based on the dissipation of the TKE and
the stratification, and the travel time of the inflows (Spigel and Imberger,
1980; Imberger and Patterson, 1981; Imberger, 1982; Spigel et al., 1986).
The clear advantage of these process-based models was the ability to
characterize the processes independently of the reservoir model. Thus,
although the mixed layer models, for example, contain a number of
parameters, these may be identified as efficiencies of the various
processes and their values derived from field or laboratory experiments.
This independence means that the resulting model may be transferred
from lake to lake without recalibration; a failure of the model to predict
observed data is not necessarily an indication of a faulty parameteriza-
tion, but rather an indication of the existence of processes not para-
meterized. Further, process-based models enable the user to determine
the detailed response of the lake to external forcing.
A third class of one-dimensional model also exists, based on a full
solution of the one-dimensional momentum and energy equations incor-
porating a turbulence closure scheme of a level higher than the eddy
diffusivities discussed above. A number of k--E models of this kind have
been developed (for example, Spalding and Svensson (1977), Svensson
(1978), and Sahlberg (1983)). The computational requirements of this
approach for long term simulations, however, are prohibitive.
The application of one-dimensional models is restricted to those lakes
for which the assumptions hold. Water Resources Engineers, Inc. (1969)
proposed an internal Froude number relating the average throughflow
velocity u from inflows and outflows and the stratification

U
Fr =- (11.1)
(g’H)”’ ’
where g’ was based on the density difference between inflow and surface
water, and H was the total depth. For Fr < l/n, they concluded that the
reservoir was sufficiently stratified for the one-dimensional assumption to
hold. Bloss and Harleman (1979) pointed out that the surface fluxes,
particularly the effects of the surface wind, were not represented in this
criterion and suggested that a limitation on wind speed U, be used in
addition. This was

(11.2)
426 Jorg Imberger and John C. Patterson

where d was the mixed layer depth, L the length of the lake, H the total
depth, and Ap and p o the density jump at the thermocline and the mean
hypolimnion density respectively. If this inequality was not satisfied, it
was assumed that the deviation of the thermocline from the horizontal
was sufficient to prevent the one-dimensional assumption from holding.
The numerical parameter was based on earlier empirical data relating
surface set up to wind speed.
This criterion was similar in character to that established by the
analysis of Spigel and Imberger (1980), based on rigorous dynamical
grounds. Here, as discussed in detail in Section IV, the stratification and
wind forcing were related through the Wedderburn number W
(Thompson and Imberger, 1980) in a series of criteria which categorized
the response of the lake to external wind forcing. These criteria, together
with modified inflow and outflow criteria and a parameter that charac-
terized the importance of the earth’s rotation, were discussed in
Patterson et al. (1984). These may be summarized as follows.
The deviation of the mixed layer was characterized by the value of W
(defined in Section IV). For W > 10, the deviation was negligible, and
the mixed layer deepening was characterized by the one-dimensional
effects of surface stirring. For 3 < W < 10, increasing evidence of shear
production was likely, however the one-dimensional model, with a shear
production algorithm was still valid. For W <3, the interface was
sufficiently close to the surface at the upwind end to be affected by
surface stirring, and the lake became mixed through an upwelling front
moving downwind. These criteria were based on the first mode wave
behavior in a rectangular basin containing a fluid with two densities.
Even with this simplification, the criteria proved a useful classification of
lakes (Patterson et al., 1984).
The inflow criterion was effectively the original Water Resources
Engineers form above (11.1). The approximate nature of the calculation
of Fr meant that there was little significance in the factor l/n, and Fr < 1
was taken to indicate one-dimensional behavior. Similarly, an outflow
Froude number was formed as

F, = Q (11.3)
H2(g’H)”2 ’
where Q was the discharge, which again reflected the relative importance
of inertial and gravitational adjustments. For F, < 1, the one-dimensional
structure was not disturbed. A criterion based on comparing the Rossby
 Physical Limnology 427

radius of deformation with the lake scale was used to characterize the
effects of the earth’s rotation.
The difficulty with the Wedderburn number and Froude number
calculations is that they are based on simple geometries and simple
two-layer density structures. To calculate the value of W appropriate to a
real lake involved solving the eigenvalue problem for the internal wave
field in the actual stratification and utilizing this to form an equivalent
two layer structure; depending on the stratification ambiguity could arise.
Likewise, the Froude number criteria were based on very simple
geometries, and took no account of, for example, the ratio of inflow to
storage.
The Lake number LN introduced above minimizes these difficulties of
application. The form discussed in Section I1 has been shown to be
equivalent, in the two layer rectangular case, to W, and its interpretation
is clear. For LN>> 1, the stratification is sufficiently strong to minimize
any disturbance of the metalimnion from the surface wind, and the para-
meterizations of Section IV are appropriate. Thus in an equivalent way to
W, the value of LN characterizes the assumption of the overall one-
dimensionality with respect to surface wind rather than just the surface
layer. Further, by redefining the maximum angle of deviation, the higher
mode responses may also be characterized.
The values of LN calculated for the Canning Reservoir, based on actual
values of u, (rather than the fixed value used in Section 11) and the field
profiles of temperature and salinity, are shown in Figure 36a. This shows
the same trend as Figure 8: LN values of order 50 for the summer period
when the stratification is strong and LN values of order 0-2 in winter
when the stratification is weak and the winds high. Clearly the one-
dimensional assumption is appropriate in summer, but there are periods
during winter when LN is close to 0, and the assumption may be invalid.
Examination of those times, however, reveals that the stratification is
extremely weak, and the error made by a vertical mixing model
compared to a model based on upwelling is small; both produce a
homogeneous reservoir over short time scales. On the other hand, if LN
were small in times of moderate to strong stratification, or W is small but
LN is large, upwelling will occur at the upwind end from the hypolimnion
in the former case and the surface layer in the latter. In both cases, the
one-dimensional assumption does not hold. In these cases, a two- or
three-dimensional circulation model may be more appropriate.
The form of LN for river inflow may also be easily determined.
428 Jorg Imberger and John C. Patterson

+
3w +

1W- ++
+ +
+
+ + t +
50- t
+ + +
+ +++ + + +
+ + + +

3 1.50-
L
9
3 0.00-

ol
0.30

0.00 I I-.
~ *++
I
+
+
+
t

I
+
+
+
t +
I
; k S OI N D J I F M I A I M J I J
MONTH

FIG.36. (a) The Lake number LNfor Canning Reservoir for those days on which field
data are available, based on the actual values of wind speed for the period 11 June 1986 to 7
September 1987. (b) The Lake number LN,[for Canning Reservoir, based on the inflows for
the same period as Figure 36a.
 Physical Limnology 429

Following Section X, the Lake number based on inflow can be defined by

(11.4)

where the length scale for the reservoir is taken as Ah”. Thus, large
values of LN,I correspond to small interface deviations.
The values of LN,, for the Canning Reservoir on those days on which
profile data are available are shown in Figure 36b. These are based on
the conservative assumption that zi = 0, that is, that the river underflows.
Clearly, LN.1 increases to relatively large values in the summer when the
stratification, and therefore, S, are large, and the inflows are small. LN,l
becomes small in periods of very weak stratification, and as in the
previous case, the weakness of the one-dimensional assumption causes
little damage.
Using LN and LN,, then gives an indication of the appropriateness of
the one-dimensional assumption for surface and inflow driven deforma-
tions of the density structure. The extension of LN to outflows and
deformations resulting from the earth’s rotations are not obvious and
have not been pursued here.
Given that the indicators for a lake or reservoir suggest that a
one-dimensional parameterization is appropriate and that a process-based
mixing model is preferred to a diffusivity model, the remaining question
is which of the available models should be chosen. As indicated above,
most of the development has been in the context of the MIT and
DYRESM models. In addition, the U.S. Army Corps of Engineers
model CE-QUAL-R1 is available. The temperature prediction com-
ponent of this model is, in its present form (U.S. Army Corps of
Engineers, 1982), very similar to DYRESM. In the following, the
structure and performance of DYRESM only will be discussed in detail.
As noted above, DYRESM is based on a Lagrangian representation of
the lake, with each horizontal layer of uniform property but variable
thickness and location. Thus, each layer expands or contracts as inflow
and outflow affect its volume, and those above move up or down. In this
way all vertical advection of mass is accounted for by layer movement,
and problems of numerical diffusion are not present. Further, conserva-
tion of mass is simpler to achieve without the necessity of computing
vertical velocities.
The model contains five basic process descriptions: surface fluxes of
heat, mass, and momentum; mixed layer dynamics; mixing below the
430 Jorg Imberger and John C . Patterson

surface layer; inflow; and outflow. Each of these is based on the

parameterizations given in the earlier sections and will be described only
briefly. The evolution of the model is described in the literature
(Imberger et al., 1978; Imberger and Patterson, 1981; Spigel and
Imberger, 1980; Imberger, 1982; Patterson et al., 1984; Hocking et al.,
1988). Various extensions to the model are given in Patterson et al.
(1985) (dissolved oxygen), Patterson and Hamblin (1988) (ice cover), and
Ivey and Patterson (1984) (bottom current induced mixing and effect of
rotation on surface layer dynamics).

1. Surface Heat, Mass, and Momentum Fluxes

The fluxes of heat, mass, and momentum at the surface are described
in detail in Section 111. Following the comments in that section with
respect to the self-regulation of the thermal budget, the bulk formulae
equations (3.1)-(3.3) are appropriate for modeling that reports on a daily
time scale, and frequently utilizes meteorological data from a single site,
averaged over a full day. The evaporative heat flux is calculated from
(3.3), using a bulk value of Cw of 2.6 X loF3, within the range of values
suggested. This corresponds to the formulation arising from the Lake
Hefner data (Tennessee Valley Authority, 1972) and is similar to the
form used by Orlob and Selna (1970). Sensible heat transfer (3.2) uses
CH= Cw and is related by the Bowen ratio (Henderson-Sellers, 1986).
The value of CD for momentum transfer (3.1) is wind speed dependent,
in the form (Donelan, 1982; Ivey and Patterson, 1984)
1.124 X lop3, u < 4 m s-', (11.5)
c,=( (0.96 + 0.041 U)x U >4 m s-'.
The remaining heat transfers are treated in the usual way. Incoming
short wave radiation is absorbed by the water column, after reflection at
the surface. The reflection is characterized by the albedo, which is
measured for the particular site. The albedo A, is a function of sun angle,
surface roughness, water color, etc., but is usually taken, for averaged
data, as a constant. Typically, A, is of order 0.04, corresponding to 4%
reflection. The absorption of short wave radiation is characterized by
Beer's Law
(11.6)
 Physical Limnology 43 1

where the sum is taken over a number of wavelength bands,

cPi==, (11.7)
i

and the qi are the attenuation coefficients for each band. Although two
and higher band relationships have been used, for example for the
absorption of short wave radiation in ice (Patterson and Hamblin, 1988),
in general a single band formulation is sufficient for use with daily data.
Incoming long wave radiation is either measured or predicted from
formulae such as the Swinbank equation (Swinbank, 1963), modified to
include the effect of cloud (Henderson-Sellers, 1986). Likewise, the long
wave emission from the surface may be measured or calculated from the
Stefan-Boltzmann Law
QL= m T 4 , (11.8)
where E is the surface emissivity, CT the Stefan-Boltzmann constant, and
T the surface temperature in "IS.The emissivity is fixed at 0.975.
These heat transfers are implemented on the layer structure of the
model, with all fluxes except short wave affecting the surface layer only.
The time step of the heating and surface mixing calculations is also set
here; the time is limited to that in which the surface layer temperature
changes by 3"C, up to a maximum of 12 hours. An additional limitation is
placed on the time step in terms of momentum transfer; the change in the
mean mixed layer velocity from the previous value is also limited.

2. Mixed Layer Dynamics

The mixed layer dynamics is modeled by an algorithm that represents a

simplified form of (4.18)-(4.22). Specifically, since the use of daily data
precludes rapid (within a diurnal time scale) changes, which prompted
the Spigel et al. (1986) formulation, the assumption that the turbulence in
the mixed layer adjusts rapidly to changing external inputs is made. This
is equivalent to putting (4.19) equal to zero. Equation (4.18) then
collapses onto the form used in DYRESM,

+--
Apgh
2p0 24p0 dh
+--I
gb2 d ( A p ) g A p b d b dh
12p0 dh dt
+ q3u:)+- u:+--+--
CK
=-(w:
2 "'2[
U f d d U,dd(U,)
6 dh 3 dh 1;'
dh
(11.9)
432 Jorg Imberger and John C. Patterson

where
q: = w: + q3u:, (11.10)
and the following equivalences apply:
77 = CN, (11.11)

(11.12)

(11.13)

CT (CF + cE)-u3. (11.14)

The values used are r,~= 1.23, C, = 0.2, CT= 0.51, and CK= 0.125. These
yield values of CF and CE similar to those given in Section IV. Equation
(4.22) is used to calculate the shear velocity.

3. Mixing below the Surface Layer

As discussed in Section X, mixing below the surface mixed layer is

characterized by patchy, sporadic, individual events that span regions of
only a few meters and last for only a few minutes. These events may arise
from, for example wave-wave interaction, double diffusion, boundary
mixing, or Kelvin-Helmholtz billows. It is not possible, in the context of
a model such as DYRESM, to model individually each of these processes
on the appropriate time and length scales. Rather, the turbulent
processes are modeled by an eddy diffusivity. This approach has been
commonly followed with a wide range of field determinations of K , (see
Section X).
DYRESM utilizes the formulation given by (10.2), as described by
Imberger (1982). Here, the energy inputs from the surface wind and the
inflow are assumed to be uniformly distributed over the surface layer and
the thermocline region. The wave characteristics are determined by
matching the K , with extreme values. Thus, for the case where wind
power input exceeds river inflow, K , from (10.2) is matched with
0.067(H - zT)u*, assuming that u u,. - This yields -
k,
12.4A,/[V‘(H-zo)], where V’ is the volume over which the energy is
dissipated, the lower boundary of which is at zo. Similarly, for the inflow
dominated case, ko-2x/do, where do is the depth of underflow, and
u-0.1 Qldgtan a’,where Q is the flowrate and a’ the stream angle.
 Physical Limnology 433

The parameter a in (10.2) is given the value 0.5 (Imberger and Hamblin,
1982).

4. Inflow

The inflow of the rivers into the reservoir is modeled in three separate
processes, as described in Section IX: determination of the plunge point,
underflow, and intrusion. Following Section IX, the plunge point
determination is parameterized by (9.1) and (9.3), the underflow by
(9.2), and the intrusions by (9.4), (9.9, and (9.7)-(9.11). The implemen-
tation of (9.4) and (9.5) is described in Imberger and Patterson (1981);
-
briefly the balance is changed from inertia-buoyancy at R 1 (R defined
by (9.7)). The cases of steep slopes, unconfined, or broad river valleys
are not modeled by DYRESM. The impact of a downflow on a sudden
density change is treated by the method of calculation of N2, as for the
withdrawal case below.
The downflow is modeled by tracking the day’s inflow down the slope
and locating its position and flowing depth. Thus, the underfiow is made
up of a number of parcels of inflows placed along the slope, moving down
on each day until their level of neutral buoyancy is reached. Insertion
following (9.4) or (9.5) generates the intrusion thickness from conserva-
tion of volume. The insertion lengths calculated from (9.4) and (9.5)
together with the downflow characteristics are essentially two-
dimensional parameterizations of the inflow; these are later utilized in a
quasi two-dimensional version (see below).

5. outfiow

In the context of the assumptions of both one- and two-dimensional

forms of DYRESM, the effects of the earth’s rotation are not considered
in the description of the process of withdrawal. Similarly, the behavior in
two-layer fluids is not explicitly modeled, with the behavior relevant to
sudden changes described in Section VIII being modeled in the use of
density gradients averaged across the interface. Although this procedure
is not strictly correct, the error is small. The averaging procedure is
described fully in Imberger and Patterson (1981). The outflow algorithm
is therefore based on (8.23) to (8.26), describing point and line sink
withdrawal from a continuously stratified fluid. Briefly, these equations
are implemented as follows. The inertial withdrawal layer thickness 6
434 Jorg Imberger and John C. Patterson

corresponding to the point source in a stratified fluid is calculated

according to (8.26), with a value of C1of 2, as discussed below. If 6 > W,
where W is the lake width, the layer has spread the width of the lake, and
for all but this immediate spreading region, the layer behaves as though it
is generated by a line sink across the full width of the lake (Lawrence and
Imberger, 1979).
For the point sink case, the inertial limit is described by (8.26), and the
viscous limit by (8.25). The transition flow is modeled, with the
distinction between the two cases being determined by the value of the
parameter S (8.28), following Cases A and B. The species convection
cases described by Ivey and Blake (1985) are not covered by the
algorithm.
If the line sink model is invoked by the condition referred to above,
the layer thickness is described by (8.23) and (8.24), depending on the
value of the parameter RL, as described in Section VIII. The value of C1
for the inertial point sink case is chosen so that the inertial layer thickness
of the point sink case will equal the inertial layer thickness arising from
the line sink, at the discharge at which the transition from point to line
sink occurs. Although slightly larger than the experimental determina-
tions of C1, the error is small and is accepted to ensure a smooth
transition from one type to the other.
The region from which the fluid is drawn is determined by the
algorithm developed by Hocking et al. (1988). Here, a velocity profile is
assumed, following Spigel and Farrant (1984), to be

t uo(1-); [1+ cos( n 3

1, z
0 s -.c1,
6
- 2,
112
(11.15)
z - 2,
0, -2 1,
61/2

which is a cosine profile about the level of the sink with a linear decay
away from the sink to the far end of the lake, where L is the lake length,
dIn the layer half thickness, uo the centerline velocity, and z, the height
of the sink.
Invoking conservation of volume enables computation of the vertical
velocities, and the streamlines were shown by Hocking et al. (1988) to
have the form
 Physical Limnology 435

where Y is a constant. Hocking et al. integrated along the streamlines

from the sink to obtain the envelope from which the withdrawal in a
given time step would be taken. Although closed form solutions were not
available, approximate forms for the curves for both large and small
withdrawal were found. The results were verified by comparison with the
Spigel and Farrant (1984) experimental results.
The calculation of these withdrawal envelopes gave an accurate
representation of the quantity of fluid taken from each of the model
layers. This allowed for a proper representation of the horizontal
transport in the withdrawal layer, noting that the envelope extended
above the calculated withdrawal layer near the sink; fluid from outside
the withdrawal layer translated vertically into the withdrawal layer before
moving horizontally into the sink.

These process descriptions are linked together in the framework of

Lagrangian layers introduced above. Thus, mixing in the mixed layer
algorithm takes place by the amalgamation of layers with redistribution
of the layer properties across the amalgamated layer according to
conservation of mass and energy. The thickness of the new layer will
equal the sum of the thicknesses of the individual layers; the lost
resolution is acceptable since there is no variability within the amalgam-
ated layer. Before a process affects that layer, for example heat fluxes at
the surface, the layer is checked against a maximum thickness criterion
and split if necessary. In other words, only the spatial resolution required
for each process is retained. In high gradient regions, the resolution is
correspondingly fine. The temporal resolution is set by the limitations on
heat and mass fluxes described above; again the resolution is fine only in
periods of relatively rapid change. The result is resolution of length scales
down to a few centimeters and time scales down to 15 minutes, but only
when necessary. This means a computationally efficient and accurate
code.
The quality of a simulation by any model depends largely on the
quality of the meteorological, inflow, and withdrawal data. As an
example of the application of DYRESM with relatively good data, a
simulation of the Canning Reservoir, described above, for the period 11
June 1986 to 7 September 1987, (450 days, covering a full seasonal cycle)
has been carried out. The result, presented as isotherm-depth histories, is
shown in Figure 37. This result should be directly compared with Figure
2. Clearly, the simulation has almost exactly followed the data, particu-
436 Jorg Imberger and John C . Patterson
TEMPERATURE ("C

50.0

40.0

- 30.0
N

20.0

10.0

I I I I I I I J I I I I I I
J A S O N D J F M A M J J A

DAY

FIG. 37. Isotherm-depth histories from the simulation of Canning Reservoir by

DYRESM for the period 11 June 1986 to 7 September 1987. This figure should be
compared with Figure 2, which shows the data for the corresponding period.

larly with respect to formation of the stratification through summer and

the subsequent mixed layer deepening in mid-May. There are two points
of mismatch. First, the shape of the 12°C isotherm in the deeper part of
the reservoir differs from the observed data. The error is minor,
however, as the temperature gradients in this region are extremely weak
and small deviations shift the isotherm large distances. Second, the
surface temperature early in the summer period is simulated as being
slightly higher than the data indicate. The error is within 1"C, however.
As a second example of the use of the model for longer term
simulations, an eight-year data set for the Wellington Reservoir was
assembled. Simulations of part of this period have been presented
 Physical Limnology 437

d 1914 1975 1978 1 gn 1978 1979 1980 1981 1982

TIME Pmmo

'7

0
1974 1975 1978 wn 1978 1979 1980 1981 1982
T I M PERKID

FIG. 38. The offtake salinities at Wellington Reservoir, measured (dashed line) and
predicted by DYRESM (dotted line), over the extended period 1974 to 1982: (a) mid-level
offtake and (b) scour offtake.

elsewhere (Imberger and Patterson, 1981); here, comparisons of the

predicted offtake properties from the two operating offtakes with the
measured values are shown (Figure 38). Clearly, there are some
deviations from the measured values, but over the long term the
agreement is remarkably good.
These applications show the strength of the one-dimensional version of
DYRESM,both for short term (one year) and longer term (eight years)
simulations. In both cases however, only the one-dimensional variations
in the stratifying species may be modeled, consistent with the assumption
of one-dimensionality. In many cases, however, the parameters of
interest may not influence the density, and even if the density field is
one-dimensional, these parameters may exhibit considerable horizontal
variability. Further, the motions in reservoirs and lakes are in general not
one-dimensional; for example, the wind accelerates the surface waters
and induces a return flow as well as tilting the surface, perhaps leading to
upwelling and internal and surface waves; the density gradients estab-
lished by these and other mechanisms such as unequal heat capture may
drive horizontal motions; inflows to a reservoir are usually of a different
density to the surface water and underflow for some distance before
438 Jorg Zmberger and John C. Patterson

being inserted horizontally. All of these two- and three-dimensional

effects are averaged out in a one-dimensional representation and the
details are lost.
Rodi (1987) reviewed the available models for a broad scope of lake
and reservoir problems. He concluded that in three dimensions the
difficulty of adequately resolving the important time and particularly
length scales limited the quantitative application of these models. In two
dimensions the situation is somewhat better, although models with
stratification are of limited application. These take a vertical surface
along the river valley of the lake, and simulate variations and motions in
the longitudinal and vertical directions. The best known of these models
is the LARM model of Edinger and Buchak (1983) (Gordon, 1981; Kim
et al. 1983) and its more recent version GLVHT (Buchak and Edinger,
1984). This model solves the laterally integrated equations of motion on a
finite difference grid, subject to a number of assumptions. Turbulence
closure was with a mixing length model with a Richardson number
correction. Again, however, resolution remains a major difficulty.
A different approach was taken by Jokela and Patterson (1985) with a
quasi two-dimensional version of DYRESM. The formulation considered
the development of horizontal structure only from inflow intrusions along
the previously defined DYRESM horizontal layers. To identify the
horizontal gradients, each horizontal layer was divided into a number of
Lagrangian parcels that moved horizontally within the layer, changing
length as the layer thickness changed, retaining the Lagrangian character
of the model. The total inflow for the day was distributed across a
number of horizontal layers in the usual way, with each layer’s
apportionment forming a new parcel for that layer. The length of the
parcel was determined from the calculation of the intrusion distance, and
all previous parcels in the layer were forced forward, contracting in
length as the layer thickness increased.
This preliminary model had no provision for adjustment of horizontal
gradients, and was applicable only for situations in which the one-
dimensional assumption held. However, the simulation of an inflow event
showed good agreement with the measured data.
An extended version of this model, with attention paid to other
processes besides inflow, has been completed by Hocking and Patterson
(1988). This model builds on the original Jokela and Patterson model,
including a gravitational adjustment. The vertical mixing processes from
the original one-dimensional model are retained, but are applied on a
parcel by parcel basis.
 Physical Limnology 439
a SALINITY (ppt)

--
E
N
30-

20-

1C-

1000 2000 3000 4000 5000 6000 7000

ACCUMULATED DISTANCE (m)

b SALINITY (ppt)

ACCUMULATED DISTANCE (m)

FIG. 39. (a) The isohalines resulting from salinity measurements at a sequence of
stations in Canning Reservoir on 27 October 1986. (b) The isohalines drawn from the
simulation of Canning Reservoir by two-dimensional DYRESM, after a short (30-day)
simulation.
440 Jorg tmberger and John C . Patterson

A simulation of an inflow event in the Canning Reservoir is compared

with the field data in Figure 39, following a 30-day simulation. The
intrusion is characterized by the isohalines shown, salinity being
effectively a tracer in the Canning Reservoir. The simulated intrusion is a
good representation of the field data in terms of the character of the
intrusion and the actual salinity values. The depth and progress of the
intrusion into the reservoir are accurately predicted, and the shape is
qualitatively the same. The thickness is slightly overpredicted. The tilting
of the 0.16ppt isohaline in the deeper part of the reservoir is also well
modeled.
The advantages of this model over those which solve some form of the
equations of motion on a fixed grid are obvious. The horizontal
resolution of the model is determined by the horizontal process acting
and is only fine where required. This means that the algorithm is
computationally economical and the dispersion problems associated with
fixed grids are dispensed with. Further, the tracking of tracers through
the reservoir is made particularly simple by the Lagrangian nature of the
representation.

XII. Reservoir Destratification by Bubble Aerators

The development of modeling techniques for lakes and reservoirs from

detailed process descriptions resulted from an increasing water resource
management requirement. For example, it may be desired to operate a
reservoir to optimize the quality of the withdrawn water, to minimize
evaporation, to prevent long-term build up of some dissolved substance
such as salt, or to prevent algae growth. To design operational strategies
to satisfy such diverse requirements, a process-based model of the kind
described above is essential. Only then, within the limitations of the
assumptions of the model, can the strategies be tested with the processes
properly represented.
As a case in point, reservoir destratification systems are frequently
installed to prevent the build up of stratification, or dismantle an existing
stratification in a reservoir, usually to improve water quality. In summer
the metalimnion is a barrier between the actively mixed epilimnion and
the hypolimnion. The hypolimnion may become deoxygenated, causing
water quality problems. Removing the stratification artificially or directly
aerating the hypolimnion have been practiced for several decades to
 Physical Limnology 441

combat this problem. The use of these management techniques in

reservoirs apparently dates from 1954 (Cooley and Harris, 1954),
although applications of bubble curtains as artificial breakwaters existed
in 1907 (see Bulson (1967) and Wilkinson (1979)). A number of reviews
have outlined the means available for destratification (Tolland, 1977;
Henderson-Sellers, 1982), including hydraulic pumping systems, mechan-
ical mixers, and air bubbler systems.
For reservoirs, the most common destratification device is an air
bubbler system. Compressed air is pumped to the bottom of the reservoir
and released; the buoyant plume rises to the surface, carrying with it
water from the hypolimnion, which is then ejected from the plume where
the density of the air-water mixture is approximately equal to the
ambient density. Once ejected, the fluid is locally heavy, and mixing
occurs, reducing the stratification. The release of the air occurs either as
one or more distinct plumes, or as a curtain, depending on the design.
The success of this procedure depends on the bubble plume being
buoyant enough to carry the hypolimnion water into the epilimnion
before discharge. The efficiency therefore depends on both the stratifica-
tion and the air discharge rate.
A typical bubble plume operation was recently documented in the
Harding Reservoir and is shown in Figure 40(a) and (b). In both
examples the wind was negligible, but during the acquisition of the data
shown in Figure 40(a) (see p. 448) the water surface layer was only very
weakly stratified. As evidenced by the shape of the isotherms the bubbler
was sufficiently active (PN = 0.3, see (12.1)) to completely penetrate the
whole water column and yet the stratification was sufficient to return the
cold surface water over the bubbler back to a depth of 6 m ; the falling
plume being nearly 30 m wide.
By contrast the data illustrated in Figure 40(b) was collected during a
windless hot afternoon when the stratification was much stronger at the
surface. Again the isotherms display the impact of the bubbler; the cold
water rising in the plume could not completely penetrate the surface
stratification (PN = 800 for the surface stratification) and the stronger
stratification led to a much thinner plume.
In spite of their relatively long period of use, the design of bubbler
systems falls far short of ideal, with much of the existing design based on
empirical rules determined for a particular water body. For example,
Davis (1980) gives a design procedure based on the stability of the water
body and the energy flux through the surface. Based on a calculation of
442 Jorg Imberger and John C . Patterson

the stability (2.3), the design procedure assumes an efficiency of 5% to

determine the energy input required by the destratifying device. Simi-
larly, determination of the length of diffuser pipe, the diffuser hole
diameters, and the diffuser separation are empirically based. The energy
requirement is based on the strongest stratification and no wind effects,
and is therefore overestimated. The calculations are sensitive to the value
of S,, and the resulting design depends strongly on the initial assumptions
made. Applications using these or similar procedures are described by
Brim and Beard (1980), Brown et al. (1982), and Brady et al. (1983). The
latter demonstrated the sensitivity of the Davis procedure to the assumed
initial profile. Burns (1977) describes some hydraulic model tests in-
tended to assist in the design procedure.
Even though the design procedures are inadequate for lake and
reservoir destratification systems, the bulk of research into bubble plumes
has been done in the context of a homogeneous environment (Kobus,
1968; Wilkinson, 1979; Milgram, 1983; Tacke el al., 1985; Sun and Faeth,
1986; Cheung and Epstein, 1987). These papers describe various aspects
of the plume characteristics, including determinations of the velocity
distribution, entrainment coefficient, and bubble slip velocity. In general
these are based on an integral model of the plume dynamics, with the
exception of the Sun and Faeth (1986) paper, which deals with the
turbulent properties of the plume.
Ditmars and Cederwall (1974), Milgram (1983), and Cheung and
Epstein (1987) presented integral models based on a homogeneous
ambient fluid. The conclusion drawn by these papers has been that, first,
it is not appropriate to treat the plume as a single phase flow unless the
aidow rate is very small, and second, that it is possible to treat the plume
as a simple plume with the buoyancy term modified by the presence of
the bubbles, provided that the slip velocity of the bubbles is
incorporated.
Stratified surrounding fluid was introduced in the models of McDougall
(1978) and Hussain and Narang (1984). The McDougall paper deals with
experiments in a strongly linearly stratified environment and shows that
the behavior of the plume is somewhat more complex than in the
unstratified case. Briefly, the plume rises, carrying relatively heavy water
and entraining ambient fluid, to the point where the mixture density is
approximately equal to the ambient, where the fluid detrains from the
plume. At this point, the detrained fluid has lost the additional buoyancy
of the bubbles and is locally heavy. It therefore plunges to a new level
 Physical Lirnnology 443

before spreading horizontally. The bubble plume continues to rise,

creating a new buoyant plume. In a strongly stratified fluid, this process
may occur a number of times, resulting in multiple intrusions, the spacing
of which depends on the stratification and the buoyancy flux. McDougall
proposes two integral models for the plume; the first based on the simple
plume model modified by the buoyancy flux input by the bubbles and the
bubble slip velocity; the second based on a double plume structure, with
an interior plume in which all the bubbles are contained, and surrounded
by an annular outer plume containing only liquid.
Hussain and Narang (1984) developed a similar double plume structure
model for application in weakly stratified fluids. The condition placed on
the stratification implicitly meant that multiple intrusions were not
possible.
In both stratified and unstratified experimental and model studies little
attention had been placed on the behavior under widely varying
stratification conditions and airflow rates. The paper by Asaeda and
Imberger (1988) attempted to resolve this question by conducting
experiments in both two layer and linearly stratified fluids of varying
degrees and varying airflow rates.
Asaeda and Imberger (1988) confirmed experimentally that the
efficiency of the destratification depended on both the stratification and
the airflow rate. For the linear stratification, they showed that the
behavior of a single plume, or, more accurately, the conversion of the
energy input by the plume into mixing, depended on the parameter

(12.1)

where N is the buoyancy frequency, H the depth of the aerator, and Qo

the airflow rate. In particular, they showed that the conversion efficiency
reached a maximum for a value of PN at approximately lo3. The efficiency
was calculated as

(12.2)

representing the ratio of the change in potential energy g AS, in the

stratification to the energy input by the plume in time At (Tolland, 1977).
The maximum efficiency observed was approximately 0.12. The maxi-
mum was also related to the character of the intrusions. For small values
of PN (relative to lo3), the plume travelled to the surface before
444 Jorg Zmberger and John C. Patterson

detraining, similar to the homogeneous experiments of Milgram (1983)

and others; for large values, multiple, unsteady intrusions formed. With
PN near lo3, the plume detrained at one or more locations, with steady
subsurface intrusions forming.
For a two-layer fluid, similar results were obtained. For a given
stratification, low airflow rates produced a plume which detrained at or
below the level of the interface, with the result that very little exchange
between upper and lower layers occurred. At high rates, the detrainment
occurred at the surface and plunged back to the interface. The critical
parameter here was

(12.3)

For PA < 30, the plume broke through the interface and mixed with the
upper layer. For PA > 30, the stratification was strong enough to prevent
the entrained fluid from penetrating the interface, and detrainment
occurred at or below the interface. In the first case, mixing occurred from
above, with a deepening of the interface; in the second, mixing was
weak, and characterized by a thickening of the interface.
These results were consistent with others; the McDougall (1978)
experimental results were of moderate PN, with steady subsurface
intrusions. The two layer results were similar to those obtained by
Graham (1980). While insufficient details are given by Graham to
calculate P A , it is clear that in the early part of the experiments, the
“diffusive” profiles are characteristic of PA > 30; the density jump at the
interface is sufficiently strong to force detrainment below the interface.
At later times, the stratification has been reduced to allow the effective
PA<30, and the plume penetrates fully, resulting in a sharper interface
as mixing from above deepens the upper layer. The experiments of
Kranenburg (1979) may also be put in the context of Pa. The series of
profiles shown in Kranenburg indicate, after the initial stages, a rapidly
deepening interface, consistent with a calculated value of PA of 0.1, much
less than the suggested value of 30 for penetration of the interface. This is
consistent with the qualitative description given by Kranenburg of an
intrusion plunging from the surface to the interface. The experiments of
Dortch and Holland (1980) were directed at comparisons with hydraulic
methods and insufficient details were given for a comparison to be made.
These conclusions have been incorporated into the reservoir model
described above. Although integral models of the plume itself exist, none
 Physical Limnology 445

have been coupled to a far field model, other than that of Kranenburg
(1979). Kranenburg developed a three component model; the plume, the
intrusion, and the far field two-layer stratification, in a cylindrical
geometry. Good comparisons with the experimental results were ob-
tained, but the model was not intended for reservoir applications.
The development of the coupled bubble plume and reservoir model is
fully described in Patterson and Imberger (1989); briefly, the single
plume model of McDougall (1978) was discretized using the layer
structure determined by the reservoir model, with the plume entraining
from each layer as it passes. As the plume rises, the effective buoyancy
anomaly decreases as entrainment lowers the plume density and the
ambient density decreases. At some level, the vertical velocity of the
entrained fluid becomes zero, the fluid is ejected and mixes with the
underlying reservoir water. It is assumed that the resulting horizontal
gravitational adjustment is sufficiently rapid for the one-dimensional
adjustment to remain in force. The bubble plume continues, starting a
new entrainment cycle.
The coupled model was verified against a one-year data series from
Myponga Reservoir in South Australia in which an aerator had been
operated over the summer at a depth of 14m below the surface
(Patterson and Imberger, 1989). The model in general reproduced the
resulting thermal structure, and it was concluded that it adequately
represented the coupling of the bubble plume dynamics with the
processes acting in the reservoir.
This enabled the use of the coupled model to determine various aspects
of plume behavior. Of particular interest was the maximum efficiency
airflow rate found in the Asaeda and Imberger (1988) experiments for a
simple geometry and simple stratifications. The question of the ap-
plicability of this to real geometries and stratifications is fundamental to
the optimum design of destratification systems.
One measure of performance of a destratification system is the
efficiency q, defined above. Asaeda and Imberger (1988) showed that the
efficiency depended on the parameters PN and Pa for the two specific
stratifications described. For the more general case the stability of the
system is characterized by S, and the airflow rate by Q,.Similar to the
cases for wind stirring and inflow, it is possible to form a version of the
Lake number appropriate to the bubble plume. Again, following Section
I1 and taking moments about the center of volume yields Stg/3, where /3 is
the angle of deviation of an isopycnal. If the disturbing force is taken as
446 Jorg Imberger and John C. Patterson

the bubble plume, characterized by a friction velocity w , , the moment of

the force scales with pow;H2, since the diameter of the plume scales with
H. The moment is therefore pow:H2L, where L is the length from the
plume to the center of volume. The work done by the plume is
characterized by poQoHg At (Dortch and Holland, 1980); this must be
equivalent to the work done by the disturbing force. Thus,
pow;H2w, At - poQoHg At (12.4)
or
113
w*-(%) . (12.5)

In the same way as in Section X,a generalized number may be defined:

(12.6)

where Bmax is the maximum deviation. For the two-layer case Bmax-
(2, - zT)/L;for the linearly stratified case & , - H / L . In general, if LN,B
is large, the deviation is small compared with Bmax and the density
structure will not be disturbed by the bubble plume; if LN,B is small, the
deviation is large. In the first case, the airflow rate is insufficient to
penetrate the stratification and cause mixing. The second case cor-
responds to an airflow rate well in excess of that required to penetrate the
stratification. If LN,B is large the efficiency will clearly be small since very
little mixing in a relatively strongly stratified environment will occur; if
LN,B is small, although mixing is occurring, the disturbance is more than
is required, and the efficiency is again low. The manifestation of the small
LN,B case is a plume detraining at the surface, which would also have
detrained at the surface with a much lower airflow rate. This qualitative
argument indicates that a maximum value of efficiency should occur at
intermediate values of LN,B and that, following the experiments of
Asaeda and Imberger (1988), the achievable efficiency should be of the
order of 0.12.
The parameters derived by Asaeda and Imberger (1988) are directly
related to L N , B . It is straightforward to show, using the definition of Bmax
above, that, for the linear stratification,

(12.7)
 Physical Limnology 447

where A. is the surface area of the experimental tank, and for the two
layer case,

(12.8)

-
For the linear stratification experiments, the value of PN lo3 gives
LN,B -8. For the two-layer experiments that yielded a transition of
character at PA-30, this puts transition at LN,B-5. The definition of
LN,B then unifies the two parameters defined by the experiments of
Asaeda and Imberger, and suggests that for other stratifications, LN,B -
5-8 will maximize efficiency of the bubble plume-induced mixing.
The numerical experiments with reservoir data by Patterson and
Imberger (1989) described above yielded single plume efficiency curves
which indicated a maximum efficiency of order 0.15, consistent with the
Asaeda and Imberger experiments. Patterson and Imberger introduced
the parameter P, where

(12.9)

and M is the total mass of the reservoir, derived from the ratio of energy
stored in the stratification to the energy input by the plume.
The simulations in Patterson and Imberger (1989) showed that a
maximum efficiency occurred at P - 1, as shown in Figure 41a for
Myponga Reservoir. Over the simulation period, a wide range of
stratification is present, and by varying Qo , a wide range of P values may
be obtained. Figure 41 is constructed from the results of several
simulations, with the efficiency plotted as a function of the P calculated
for a particular stratification and airflow without regard to the temporal
order of occurrence.
Myponga Reservoir is a small reservoir of storage 26.8 x lo6m3 and
surface area 2.8 x 106m2.The same calculation has been carried out for
the Harding Reservoir in the northwest of Western Australia. The
Harding Reservoir is substantially larger (storage 63.8 x lo6 m3 and
surface area 14.1 x lo6m2) and is exposed to quite different meteorologi-
cal forces, being located at latitude 20"s (compared with Myponga
Reservoir at 35"s). The calculation of P, however, yields a maximum
-
efficiency again at P 1, with a peak efficiency of 0.15 (Figure 41b). This
parameter therefore unifies the simulation data.
448 Jorg Imberger and John C . Patterson

I TEMPERATURE ("C)
0.0

2.0

--
E
4.0

E
0

6.0

8.0

10.0 I I I
21 I 40.0 60.0 80.0 100.0 120.0

PROJECTED DISTANCE (rn)

FIG.40. Transverse transect across the center of a 400 m long bubble diffuser (1 mm
diameter holes spaced at 4 m) located at a depth of 9 m. Data was obtained by yo-yoing a
temperature probe through the water column at intervals of approximately 8 meters from
the surface to within 1 m of the bottom. Gas flow rate was 40 I s-'. (a) Morning with weak
stratification (April 1989) and (b) Afternoon with strong stratification (April 1989).

The relationship between P and LN,B is given by

For the reservoir, the appropriate Pmax is (H - zT)/L, and the relationship
between L N , B and P is not direct, as both M and zT may change daily.
The calculation of L N , B would require new simulations; these have not
been performed.
The performance of a single plume is characterized by the value of P
and therefore indirectly by ; for optimum performance, a single
-
plume should be operated such that P 1. As the stratification changes,
the airflow rate should change. Further, the degree of destratification
obtained by a plume is given by A(gS,), which will generally be much less
 Physical Limnology 449

b TEMPERATURE (OC)

--
E
E
W
n
I

I
28.1
28 1

.BUBBLER
-280.
./
, 280
219-x

PROJECTED DISTANCE (m)

FIG.40. (conrd.)

than gS,. Either more plumes are required or the time over which the
destratification occurs must be extended. An estimate for the number of
-
plumes required is given by n gS,/A(gS,). Patterson and Imberger
(1989), however, show that the efficiency of multiple plumes is higher
than a single plume carrying the same total airflow. The optimum
operational strategy then is to operate as many separate plumes as
-
possible, each with P 1. This of couse involves altering the airflow rate
at perhaps daily intervals, depending on the stratification, a level of
control which may not be possible. In reality, both the number of plumes
and the airflow rate are usually fixed, or variable only in the simplest
sense, such as stopping and starting the aerator.
The effect of this nonvariable operational strategy was demonstrated
by Patterson and Imberger (1989); for Myponga Reservoir, the overall
efficiency achieved was only 0.025 for a 501s-' airflow, the result of
many times when the stratification was weak and the P value extremely
low. This result was strongly dependent on the number of plumes and
450 Jorg Imberger and John C . Patterson
a
20

- 1
I o9 1b-2 1b-1 102

FIG.41. (a) The efficiency of a single bubble plume as a function of the parameter P for
-
Myponga Reservoir, showing a clear maximum at P 1. (b) The efficiency for a single
bubble plume as a function of P for Harding Reservoir. The maximum is again at P 1.-
 Physical Limnology 45 1

was achieved with 10 separate plumes. A relatively simple control

strategy, in which the aerator was activated only on those days when the
temperature difference between aerator and surface exceeded 1"C,
doubled the total efficiency to 0.052. This was achieved with 100 separate
plumes. The interaction between the optimum number of plumes, the
total airflow rate, and the operational strategy is a complex one, though it
is clear that further substantial gains in efficiency could be made if control
over the airflow rate is introduced.

XnI. Summary
Advances in physical limnology in the last 10 years, both in the more
traditional area of internal wave seiching and in the newer area of
small-scale motions and mixing, has created a very large discipline. We
have highlighted this progress in this review, and now wish to draw
together those areas which we feel require more work.
(a) Seasonal behavior. Lakes in the tropics may be compared with
those in temperate regions if the stability and forcing are matched. This
may be done by evaluating two nondimensional numbers: the Wedder-
burn number W ,which describes the behavior of the surface layer, and
the Lake number L N , which describes the response of the entire lake.
This hypothesis was formed from a review of earlier work on the
intercomparison of lakes; a detailed intercomparison study of a large
range of lakes in different locations should now be conducted. On the
basis of this it may be found that further nondimensional numbers are
required to specify higher mode responses.
(b) Surface fluxes. A great deal of progress has been made, mainly by
oceanographers, in determining the exchange of momentum, heat, and
water vapor at the air-water interface. Provided the atmospheric internal
boundary layer is well established and the air is either not too stable or
unstable, there are adequate procedures for calculating surface fluxes.
However, few lakes studies have data of sufficient quality to enable the
use of the stability correction. This, together with the well-documented
fact that surface roughness has a strong influence, requires that meteoro-
logical stations be located in the internal boundary layer and that both
stability and roughness be incorporated through a model for the fetch
duration and radiation of waves. Last, the wind is spatially highly
variable over the surface of the lake. This leads to energy input at a
basin-scale length that causes deep hypolimnetic mixing and contributes
452 Jorg Imberger and John C . Patterson

to the strong, horizontal transport processes in the lake, a process so far

completely ignored and one requiring a measure of the wind variability
over the lake.
(c) The su$ace layer. The one-dimensional, vertical energy-based
mixed layer models appear to describe the surface layer well if local wind
and flux data are used. What is now required is a generalization of the
one-dimensional model incorporating the Wedderburn number W , which
governs the behavior of this layer, to account for horizontal advection of
momentum and mass. This should include gravitation-induced advection,
which becomes dominant at the cessation of a wind.
(d) Upwelling. When the Wedderburn number becomes low, the
mixed layer upwells at the upwind end. This upwelling occurs at all values
of the Wedderburn number; it is more severe at low values, but is still
present at values as high as 10 or 20. Recent models of upwelling have
indicated an entrainment process with the same deepening rate as the
rate usually attributed to surface-introduced turbulence. Once again,
these models need to be generalized to allow for two-dimensional effects
and the subsequent horizontal advection spawned at the cessation of
wind.
(e) Differential deepening. This new mechanism describes the uneven
deepening process brought about by the spatial variability of wind stress.
The high degree of variability of the wind over a lake causes extreme
deepening in exposed areas and almost no deepening in sheltered areas,
embayments, and side arms. This observation affects two areas. First, a
lake-averaged surface layer model must somehow account for this
variability, since at the .cessation of each wind period, gravitational
motions even out the isopycnals and the net amount of deepening will
depend on the exact distribution which, in turn, is determined by the
variability of the wind. Second, data from the Wellington Reservoir have
shown that the strong undulations introduced at the base of the mixed
layer by differential deepening lead to the introduction of energy from
the wind through the surface layer directly into the hypolimnion at a
basin scale. This is a much more effective way of getting energy into the
deep part of a lake than with a simple uniform upwelling mechanism
because not only are the first and second modes of the basin excited but
differential deepening energizes higher order internal wave modes. The
relationship between upwelling, basin seiching, and differential deepen-
ing as a function of Wedderburn and Lake numbers must be fully
investigated and the consequences for internal mixing assessed.
 Physical Limnology 453
(f) Diflei-enfial heating and cooling. This subject has received a great
deal of attention through the study of the thermal bar in cold lakes but,
because it is the strongest single process for horizontal advection in lakes,
it should receive a great deal more. An assessment of the influence of
frequency of cycling (diurnal) on embayments of various sizes must be
carried out to understand the influence of inertia of the water column.
Only then can we predict the horizontal transport in and out of sidearms.
This is important for both ecology and engineering management, since
recreation in reservoirs expose the lake to pollution. Differential absorp-
tion, the mechanism by which phytoplankton blooms are torn apart by
self-generated convective motions, is a new field. Much work needs to be
done in the interaction of convection and growth dynamics of various
species of plankton.
(g) OutJlow. Selective withdrawal has received a great deal of atten-
tion over the last 20 years and most fundamental flow configurations can
be predicted with reasonable accuracy. The outstanding problems center
around selective withdrawal in a variable stratification, and the influence
of rotation on the initial value problem and the final degeneration into
eddies. Last, the interaction of selective withdrawal and topographic
constraints (sills and contractions) is ready for analysis. This is especially
important since it has been shown that shear waves, which set up the
selective withdrawal layer, are solutions to the Korteweg de Vries
equation, allowing the full interaction to be explored while retaining
some of the nonlinearity of the fluid motion.
(h) lnjlows. An inflow plunges underneath the reservoir water, flows
down the drowned river valley until it reaches a point of neutral
buoyancy, then intrudes into the lake as a horizontal layer. The case
where the entrainment of the underflow is dominated by shear produc-
tion does not seem to have been investigated. Further, the entrainment
properties of the plunge point need urgent attention. The generation of
turbulence or the maintenance of turbulence in the intrusion layer is
important because intrusions can interact with the background internal
wave field to sustain a definite level of turbulent transport. Last,
lake-specific problems such as the interaction of the surface layer with an
inflowing stream and the interaction of the heaving (associated with
internal wave basin scale motions) and the inflow require generalization.
(i) Miring below the surface layer. The central problem here remains.
However, with the introduction of the Lake number, the way to proceed
is clearer. The distribution of turbulent patches and mixing efficiency
454 Jorg Zmberger and John C. Patterson

within each of these patches must be identified as the Wedderburn and

Lake numbers vary. Since, in a typically stratified lake, the volume of
fluid actively involved in turbulent motions at any time is at a maximum
of 1%, this is a most urgent and useful avenue to pursue. Once an
algorithm has been achieved that allows for a statistical description of the
patch distribution and identification of the mechanism underlying each
patch (and thus a description of the mixing efficiency), we will be able to
model deep mixing in a lake. A strong effort should be mounted to
investigate the formation of internal patches and the maintenance of the
turbulent benthic boundary layer.
(j) Modeling. Modeling has proceeded along two avenues: One-
dimensional parametric modeling and full eddy simulation or closure
scheme models. Neither approach is appropriate: two- and three-
dimensional motions are of central importance in a lake, and the scales of
motion range from the Kolmogorov scale to the basin scale. There does,
however, appear to be a spectral gap between the internal waves at the
basin scale and the motions within a turbulent patch. Further, as
mentioned above, the volume occupied by turbulent motions in a lake at
any one time is likely to be less than 1%. A two-dimensional model is
needed that accounts for the mean motions generally described by basin
seiching, superimposed on which is a model that determines the statistical
distribution of turbulence within the lake, in the benthic boundary layer
and surface layer. Last, a further model must be superimposed on these
two that adjusts the density field according to certain mixing efficiencies.
Such a closure scheme has a real chance of success in even quite large
lakes, with present computing power. Many variations are of course
possible, ranging from a simple parameterization of the patch distribution
and the patch efficiency to full turbulent simulations within each patch
once they have been identified.
(k) Reservoir destratijication by bubble aeration. It is unwise to try to
destratify a lake completely by introducing mechanical energy. It is much
more energy efficient to weaken the structure through the bubble flow
and allow the wind to complete the mixing. This requires an examination
of the bubble flow and the natural reservoir dynamics. This area of
research seems to be in its infancy, as does the equivalent research where
propellors are used rather than bubble plumes. Given that the field of
reservoir destratification is an enormous industry, large cost savings are
no doubt possible.
 Physical Limnology 455

Acknowledgements
Sections I to X were written by Jorg Imberger and Sections XI and XI1 were written by
John Patterson. We would like to thank Greg Ivey and Graeme Hocking for their
comments on the manuscript. Carolyn Oldham prepared the figures and performed the
computations, and Colleen Henry-Hall edited the manuscript and prepared it for
publication. This assistance is gratefully acknowledged. This work was supported by the
Centre for Environmental Fluid Dynamics, the Australian Research Council, the Australian
Water Research Advisory Council, and the West Australian Water Authority.

References
Akiyama, J., and Stefan, H. G. (1984). Plunging flow into a reservoir: Theory, J . Hydr.
Eng. 110,484-499.
Akiyama, J., and Stefan, H. G. (1985). Turbidity current with erosion and deposition. J .
Hydr. Eng. 111, 1473-1496.
Akiyama, J., and Stefan, H. G. (1987). Gravity currents in lakes, reservoirs and coastal
regions: Two-layer stratified flow analysis. Proj. Rept. No. 253, University of
Minnesota.
Anderson, R. J., and Smith, S. D. (1981). Evaporation coefficient for the sea surface from
eddy flux measurement. J . Geophys. Res. 86, 449-456.
Andre, J. E., and Lecarrere, P. (1985). Mean and turbulent structures of the oceanic
surface layer as determined from one-dimensional third-order simulations. J . Phys.
Oceanogr. 15, 121-132.
Arai, J. (1964). Some relations between the thermal property of lake and its fetch size.
Geogr. Rev. of Japan 37, 131-137.
Arita, M., Jirka, G. H., and Tamai, N. (1986). Classification and mixing of two-
dimensional buoyant surface discharges. 1. Hydr. Res. 24, 333-345.
Asaeda, T., and Imberger, J. (1988). Structures of bubble plumes in stratified environ-
ments. Environmental Dynamics Rept. ED-88-250, Centre for Water Research,
University of Western Australia.
Atkinson, J. F. (1988). Interfacial mixing in stratified flows. J. Hydr. Res. 26, 27-31.
Atkinson, J. F.,and Harleman, D. R. F. (1983). A wind-mixer layer model for solar ponds.
Solar Energy 31,243-259.
Bean, B. R., Emmanuel, C. B., Gilmer, R. O., and McGavin, R. E. (1975). The spatial
and temporal variations of the turbulent fluxes of heat, momentum and water vapor
over Lake Ontario. J. Phys. Oceanogr. 5 , 532-540.
Bejan, A., Al-Homoud, A. A., and Imberger, J. (1981). Experimental study of
high-Rayleigh number convection in a horizontal cavity with different end tempera-
tures. J. Fluid Mech. 109, 283-299.
Bejan, A., and Rossie, A. N. (1981). Natural convection in a horizontal duct connecting
two fluid reservoirs. J . Heat Transfer 103, 108-113.
Bejan, A., and Tien, C. L. (1978). Laminar natural convection heat transfer in a horizontal
cavity with different end temperatures. J . Heat Transfer 100,641-647.
Bella, D. A. (1970). Simulating the effect of sinking and vertical mixing on algal population
dynamics. J. Water Pollution Control Fed. 42(5), 140-152.
Bennett, J. R. (1971). Thermally driven lake currents during the spring and fall transition
periods. In “Proc. 14th C o d . Great Lakes Res.,” Ann Arbor p. 535. Intl. Assoc. for
Great Lakes Res.
456 Jorg Zmberger and John C . Patterson
Benney, D. J. (1966). Long nonlinear waves in fluid flows. J. Math. Phys. Sci. 45, 52-63.
Blake, S., and hey, G. N. (1986a). Density and viscosity gradients in zoned magma
chambers, and their influence on withdrawal dynamics. J . Volcanology and Geothermal
Res. 30,201-230.
Blake, S . , and hey, G. N. (1986). Magma-mixing and the dynamics of withdrawal from
stratified reservoirs. J . Volcanology and Geothermal Res. 27, 153-178.
Blanc, T. V. (1983). A practical approach to flux measurements of long duration in the
marine atmospheric surface layer. J. Clim. Appl. Met. 22, 1093-1110.
Blanc, T. V. (1985). Variation of bulk-derived surface flux, stability, and roughness results
due to the use of different transfer coefficient schemes. J . Phys. Oceanogr. 15,
650-669.
Blanc, T. V. (1987). Accuracy of bulk-method-determined flux, stability and sea surface
roughness. J . Geophys. Res. 92,3867-3876.
Blanton, J. 0.(1973). Vertical entrainment into the epilimnia of stratified lakes. Limnol.
Oceanogr. 18,697-704.
Bloss, S., and Harleman, D. R. F. (1979). Effect of wind-induced mixing on the seasonal
thermocline in lakes and reservoirs. Rept. No. 249, Massachusetts Institute of
Technology.
Bohan, J. P., and Grace, J. L. (1973). Selective withdrawal from man-made lakes. Tech.
Rept. No. H-73-4,Waterways Experiment Station, Vicksburg, Mississippi.
Boyce, F. M., Robertson, D. G., and hey, G. N. (1983). Summer thermal structure of
Lake Ontario off Toronto: Cooling the big city. Atmos. Ocean. 21,397-417.
Brady, J. A., Davis, J. M., and Douglas, E. W.(1983). First filling of Kielder Reservoir
and future operational considerations. J. Inst. Water Eng. Sci. 37, 295-312.
Brim, W. D., and Beard, J. D. (1980). Reservoir management by pneumatic induced
circulation. Paper No. 5-16.In “Proc. Symp. on Surface Water Impoundments,” (H.
G. Stefan, ed.), ASCE, pp. 877-885. Minneapolis, Minnesota.
Britter, R. E., and Simpson, J. E. (1978). Experiments on the dynamics of a gravity current
head. J. Fluid Mech. 88, 223-240.
Britter, R. E., and Simpson, J. E. (1981). A note on the structure of the head of an
intrusive gravity current. J . Fluid Mech. 112,459-466.
Brocard, D. N., and Harleman, D. R. F. (1980). Two-layer model for shallow horizontal
convective circulation. I. Fluid Mech. 100, 129-146.
Browand, F. K.,Guyomar, D., and Yoon, S. C.(1987). The behavior of a turbulent front
in a stratified fluid: Experiments with an oscillating grid. J . Geophys. Res. 92,
5329-5341.
Brown, I. K., Wooley, D. A., and Jory, A. G. (1982). Artificial destratification of Lake
Moms to improve water quality. In “Proc. Hydr. and Water Resources Symp.,”
Institution of Engineers, Australia, Melbourne, Victoria.
Brubaker, J. M. (1987). Similarity structure in the convective boundary layer of a lake.
Nature 330,742-745.
Bryant, P. J., and Wood, I. R. (1976). Selective withdrawal from a layered fluid. J. Fluid
Mech. 77,581-591.
Buchak, E. M., and Edinger, J. E. (1984). Generalized, longitudinal-vertical hydrodynam-
ics and transport: Development, programming and applications. Doc. No. 84-WR,
Waterways Experiment Station, Vicksburg, Mississippi.
Bulson, P. S. (1967).Theory and design of bubble breakwaters. Dock and Harbour Aurh.
48(560), 47-54.
Bunn, S . E., and Edward, D. H. D. (1984). Seasonal meromixis in three hypersaline lakes
on Rottnest Island, Western Australia. Aust. J . Mar. Freshw. Res. 35,261-265.
 Physical Limnology 457
Bums, F. L. (1977). Localized destratification of large reservoirs to control discharge
temperatures. Prog. Waf. Tech. 9, 53-63.
Busch, N. E. (1977). Fluxes in the surface boundary layer over the sea. In “Modeling and
Prediction of the Upper Layers of the Ocean” (E. B. Kraus ed.), pp. 72-91, Pergamon
Press, New York.
Businger, J. A. (1955). On the structure of the atmospheric surface layer. J . Meteor. U ,
553-561.
Businger, J. A. (1973). Turbulent transfer in the atmospheric surface layer. In “Workshop
on Micrometeorology” (D. A. Hauger ed.), pp. 67-100. Science Press, Ephrata,
Pennsylvania.
Cacchione, D., and Wunsch, C. (1974). Experimental study of internal waves over a slope.
J, Fluid Mech. 66, 223-239.
Caldwell, D. R. (1983). Oceanic turbulence: Big bangs or continuous creation? 1. Geophys.
Res. 88,7543-7550.
Carmack, E. C. (1986). Circulation and mixing in ice-covered waters. In “The Geophysics
of Sea Ice” (N. Untersteiner ed.), pp. 641-712. Plenum Press, New York.
Carslaw, H. S., and Jaeger, J. C. (1978). Conduction of Heat in Solids, Third Edition.
Oxford University Press.
Carson, D. J., and Richards, P. J. R. (1978). Modeling surface turbulent fluxes in stable
conditions. Boundary Layer Met. 14, 67-81.
Cavaleri, L., and Zecchetto, S. (1987). Reynolds stresses under wind waves. J. Geophys.
Res. 92,3894-3904.
Chao, S . Y. (1988). River-forced estuarine plumes. J. Phys. Oceanogr. 18, 72-88.
Charnock, H. (1955). Wind stress on a water surface. Quart. J . Roy. Meteor. SOC. 82,
639-640.
Chen, J. C. (1980). Studies on gravitational spreading currents, Rept. No. KH-R-40,
California Institute of Technology.
Cheung, F. B., and Epstein, M. (1987). Two-phase gas bubble-liquid boundary layer flow
along vertical and inclined surfaces. Nuclear Eng. and Design. 99, 93-100.
Christodoulou, G. G. (1986). Interfacial mixing in stratified flows. J . Hydr. Res. 24, 77-92.
Chu, V. H., and Baddour, R. E. (1984). Turbulent gravity-stratified shear flows. 1. Fluid
Mech. Us,353-378.
Chu, V. H., and Jirka, G. H. (1985). Surface buoyant jets and plumes. In “Encyclopedia of
Fluid Mechanics” (N. Cheremisinoff ed.), pp. 1053-1084. Gulf Publishing, Houston,
Texas.
Clarke, R. H., Dyer, A. J., Brook, R. R., Reid, D. G., and Troup, A. J. (1971). The
Wangara experiment: Boundary layer data. Tech. Paper No. 19, CSIRO Div. Met.
Phys., Melbourne, Victoria.
Claussen, M. (1987). The flow in a turbulent boundary layer upstream of a change in
surface roughness. Boundary Layer Met. 40,31-86.
Cohen, A. (1979). Existence and regularity for solutions of the Korteweg-de Vries
equation. Arch. Rat. Mech. Anal. 71, 143-175.
Callings, I. L. (1985). Two large Froude numbers cusped free-surface flows due to a
submerged line source or sink. Manuscript. Deakin University, Victoria.
Colman, J. A,, and Armstrong, D. E. (1987). Radon-222 determined vertical eddy
diffusivity in the benthic boundary layer of ice-covered lakes. Limnol. Oceanogr. 32,
577-591.
Cooley, P., and Harris, S. L. (1954). The prevention of stratification in reservoirs. J. Inst.
Water Eng. 8, 517-537.
Cormack, D. E., Leal, L. G., and Imberger, J. (1974). Natural convection in a shallow
458 Jorg Imberger and John C. Patterson
cavity with differentially heated end walls. Part 1: Asymptotic theory. J. Fluid Mech.
65,209-230.
Cormack, D. E., Stone, G. P., and Leal, L. G. (1975). The effect of upper surface
conditions on convection in a shallow cavity with differentially heated end walls. Zntl. J .
Heat Mass Transfer 18,635-648.
Craya, A. (1949). Recherches theoriques sur I’ecoulement de couches superposees de
fluides de densites differentes. La Houille Blanche Jan.-Feb., 44-55.
Csanady, G. T. (1972). Response of large stratified lakes to wind. J . Phys. Oceanogr. 2,
3-13.
Csanady, G . T. (1982). “Circulation in the Coastal Ocean.” D. Reidel Publishing, Boston.
Dake, J. M. K., and Harleman, D. R. F. (1969). Thermal stratification in lakes: Analytical
and laboratory studies. J. Water Resources Res. 5 , 484-495.
Darbyshire, J., and Colclough, M. (1972). Measurement of thermal conductivities in a
freshwater lake. Geofiica Pura i Applicata 93,151-158.
Darden, R. B., Imberger, J., and Fischer, H. (1975). Jet discharge into a stratified
reservoir. J. Hydr. Diu. ASCE 101, 1211-1220.
Davis, J. M. (1980). Destratification of reservoirs-A design approach for perforated-pipe
compressed-air systems. Water Services 84, 497-504.
de Szoeke, R. A. (1980). On the effects of horizontal variability of wind stress on the
dynamics of the ocean mixed layer. J . Phys. Oceanogr. 10,1439-1454.
de Szoeke, R. A., and Rhines, P. B. (1976). Asymptotic regimes in mixed-layer deepening.
J . Marine Res. 34, 111-116.
Deardorff, J. W. (1968). Dependence of air-sea transfer coefficients on bulk stability. J .
Geophys. Res. 73,2549-2557.
Deardorff, J. W. (1970). Convective velocity and temperature scales for the unstable
planetary boundary layer. J. Atinos. Sci. 27, 1211-1213.
Deardorff, J. W. (1974). Three-dimensional numerical study of the height and mean
structure of a heated planetary boundary layer. Boundary Layer Met. 7,81-106.
Debler, W. R. (1959). Stratified flow into a line sink. J . Eng. Mech. Diu. 85, 51-65.
Denton, R. A,, and Wood, I. R. (1981). Penetrative convection at low Peclet number. J .
Fluid Mech. 1l3, 1-21.
Dickey, T. D., and Simpson, J. J. (1983). The influence of optical water type on the diurnal
response of the upper ocean. Tellus 35B, 142-154.
Dillon, T. M., Richman, J. G., Hansen, C. G., and Pearson, M. D. (1981). Near surface
turbulence measurements in a lake. Nature 290, 390-392.
Ditmars, J. D., and Cedenvall, K. (1974). Analysis of air-bubble plumes. In “Proc. Coastal
Eng. Conf.,” Copenhagen. Ch. 128, pp. 2209-2226.
Djordjevic, V. D., and Redekopp, L. G. (1978). The fission and disintegration
- of internal
solitary waves moving over two-dimensional topography. J. Phys. Oceanogr. 8 ,
1016-1024.
Dobson, F. W., Hasse, L., and Davis, R. (1980). “Air-Sea Interaction Instruments and
Methods.” Plenum Press, New York.
Donelan, M. A. (1980). Similarity theory applied to the forecasting of wave heights,
periods and directions. In “Proc. Canadian Coastal Conf. ,” pp. 47-61. Burlington,
Ontario.
Donelan, M. A. (1982). The dependence of the aerodynamic drag coefficient on wave
parameters. In “First Intl. Conf. on Meteorology and Air/Sea Interaction of the
Coastal Zone” (preprint volume), pp. 381-387. The Hague, American Met. SOC.,
Boston.
Donelan, M. A., and Peirson, W. J. (1987). Radar scattering and equilibrium ranges in
 Physical Limnology 459

wind-generated waves with application to scatterometry. J . Geophys. Res. 92,

4971-5029.
Dortch, M. S., and Holland, J. P. (1980). Methods of total lake destratification. Paper No.
5-19. In “Proc. Symp. on Surface Water Impoundments,” ASCE, pp. 913-922.
Minneapolis, Minnesota.
Drazin, P. G., and Reid, W. H. (1981). “Hydrodynamic stability.” Cambridge University
Press, New York.
Drummond, J. E., and Korpela, S. A. (1987). Natural convection in a shallow cavity. 1.
Fluid Mech. 182, 543-564.
Dubreil-Jacotin, M. L. (1934). Sur les theorems dexistence relatifs a w ondes permanentes
periodiques a deux dimensions dan les liquids heterogenes. J. de Mathematiques 3,
217-291.
Dyer, A. J. (1974). A review of flux-profile relationships. Boundary Layer Met. 7 , 363-372.
E, X., and Hopfinger, E. J. (1986). On mixing across an interface in stably stratified fluid.
J. Fluid Mech. 166, 227-244.
Edinger, J. E., and Buchak, E. M. (1983). Developments in L A W : A longitudinal-
vertical, time-varying hydrodynamic reservoir model. Tech. Rept. E-83-1, Waterways
Experiment Station, Vicksburg, Mississippi.
Elder, J. W. (1959). The dispersion of marked fluid in turbulent shear flow. J . Fluid Mech.
5,544-560.
Elliott, G. H., and Elliott, J. A. (1970). Laboratory studies on the thermal bar. In “Proc.
13th Conf. Great Lakes Res.,” pp. 413-418. Intl. Assoc. Great Lakes Res. Ann Arbor.
Ellison, T. H., and Turner, J. S. (1959). Mixing of a dense fluid in a turbulent pipe flow.
Part 1: Overall description of the flow. J . Fluid Mech. 33, 514-528.
Emmanuel, C. B. (1975). Drag and bulk aerodynamic coefficients over shallow water.
Boundary Layer Met. 8, 465-474.
Farmer, D. M. (1978). Observations of long nonlinear waves in a lake. J. Phys. Oceanogr.
8, 63-73.
Farmer, D. M., and Cartnack, E. (1982). Wind mixing and restratification in a lake near the
temperature of maximum density. J . Phys. Oceanogr. 11, 1516-1533.
Farrell, G. J., and Stefan, H. G. (1986). Buoyancy induced plunging flow into reservoirs
and coastal regions. Proj. Rept. No. 241, St. Anthony Falls Hydraulic Lab., University
of Minnesota.
Faust, K. M. (1981). Modelldarstellung von Warmeinselstromungen durch Konvek-
tionsstrahlen. Paper No. SFB 80/ET/201, Universitat Karlsruhe.
Findikakis, A. N., and Street, R. L. (1982a). Finite element simulation of stratified
turbulent flows. J. Hydr. Div. ASCE 108, 904-920.
Findikakis, A. N., and Street, R. L. (1982b). Mathematical description of turbulent flows.
I. Hydr. Div. A X E 108, 887-903.
Findikakis, A. N., and Street, R. L. (1983). Approximate numerical model for stratified
flow. 1. Eng. Mech. 109, 950-969.
Fischer, H. B., and Smith, R. D. (1983). Observations of transport to surface waters from a
plunging inflow to Lake Mead. Limnol. Oceanogr. 28, 258-272.
Fischer, H. B., List, E. J., Koh, R. C. Y., Imberger, J., and Brooks, N. H. (1979).
“Mixing in Inland and Coastal Waters.” Academic Press, New York.
Forbes, L. K., and Hocking, G. C. (1988). Flow caused by a point sink in a fluid having a
free surface. J . Aust. Math. SOC. Ser. B (in press).
Ford, D. (1978). Unpublished data. Environ. Lab., Waterways Experiment Station,
Vicksburg, Mississippi.
Ford, D. E., and Johnson, M. C. (1981). Field observations of density currents in
460 Jorg Imberger and John C. Patterson
impoundments. Surface Water Impoundments, H. G. Stefan, ed. ASCE. Minneapolis,
Minnesota. pp. 1239-1248.
Ford, D. E., Johnson, M. C., and Monismith, S. G. (1980). Density inflows to Degray
Lake, Arkansas. In “Second Intl. Symp. on Stratified Flows,” pp. 977-987.
Trondheim.
Franke, R., Leschziner, M. A., and Rodi, W. (1987). Numerical simulation of wind-driven
turbulent flow in stratified water bodies. In “Proc. 3rd Intl. Symp. on Stratified Flows,”
(E. J. List, ed.) ASCE Pasadena, California.
Friehe, C. A., and Schmitt, K. F. (1976). Parameterization of air-sea interface fluxes of
sensible heat and moisture by the bulk aerodynamic formulas. J . Phys. Oceanogr. 6,
801-809.
Fritts, D. C. (1984). Gravity wave saturation in the middle atmosphere: A review of theory
and observations. Rev. Geophys. Space Phys. 22, 275-308.
Gargett, A. E., and Holloway, G. (1984). Dissipation and diffusion by internal waves
breaking. J . Marine Res. 42, 15-27.
Gariel, P. (1949). Recherches experimentales sur I’ecoulement de couches superposees
de fluides de densites differentes. La Houille Blanche Jan.-Feb., 44-55.
Garratt, J. R. (1977). Review of drag coefficients over oceans and continents. Monthly
Weather Review 105,915-929.
Garratt, J. R. (1987). The stably stratified internal boundary layer for steady and diurnally
varying offshore flow. Boundary Layer Met. 38,369-394.
Garratt, J. R., and Ryan, B. F. (1988). The structure of the stably stratified internal
boundary layer in offshore flow over the sea. Boundary Layer Met. 47, 17-40.
Garrett, C., and Gilbert, D. (1988). Estimates of vertical mixing by internal waves reflected
off a sloping bottom. In “Small-Scale Turbulence and Mixing in the Ocean: Proc. 19th
Intl. Leige Coll. on Ocean Hydrodyn.” (J. J. Nihoul and B. M. Jamart eds.), pp.
405-424. Elsevier, New York.
Garrett, C., and Munk, W. (1979). Internal waves in the ocean. Ann. Rev. Fluid Mech. 11,
339-369.
Garwood, R. W. (1977). An oceanic mixed layer model capable of simulating cyclic states.
J . Phys. Oceanogr. 7,455-468.
Geernaert, G. L., and Katsaros, K. B. (1986). Incorporation of stratification effects on the
oceanic roughness length in the derivation of the neutral drag coefficient. J. Phys.
Oceanogr. 16, 1580-1584.
Geernaert, G. L., Katsaros, K. B., and Richter, K. (1986). Variation of the drag coefficient
and its dependence on sea state. J. Geophys. Res. 91,7667-7679.
Geernaert, G. L., Larsen, S. E., and Hansen, F. (1987). Measurements of the wind stress,
heat flux and turbulence intensity during storm conditions over the North Sea. J .
Geophys. Res. 92, 13127-13139.
Gelhar, L. W., and Mascolo, D. M. (1966). Non-diffusive characteristics of slow viscous
stratified flow towards a line sink. Massachusetts Institute of Technology Hydrodynam-
ics Lab. Rept. No. 88, pp. 1-39.
Gibson, C. H. (1980). Fossil temperature, salinity and vorticity turbulence in the ocean. In
“Marine Turbulence” (J. C. J. Nihoul ed.), pp. 221-257. Elsevier, New York.
Gibson, C. H. (1982). On the scaling of vertical temperature gradient spectra. J . Geophys.
Res. 87, 8031-8038.
Gibson, C. H. (1986). Internal waves, fossil turbulence and composite ocean microstructure
spectra. J . Fluid Mech. 168,89-117.
Gibson, C. H. (1987a). Fossil turbulence and intermittency in sampling oceanic mixing
processes. J . Geophys. Res. 92, 5383-5404.
 Physical Limnology 46 1
Gibson, C. H. (1987b). Oceanic turbulence: Big bangs and continuous creation. J ,
Physiochem. Hydrodyn. 8, 1-22.
Gibson, C. H. (1988). Evidence and consequences of fossil turbulence in the ocean. In
“Small-Scale Turbulence and Mixing in the Ocean, Proc. 19th Intl. Leige Coll. on
Ocean Hydrodyn.” (J. J. Nihoul and B. M. Jamart, eds.), pp. 319-334. Elsevier, New
York.
Gill, A. E. (1966). The boundary-layer regime for convection in a rectangular cavity. J.
Fluid Mech. 26, 515-536.
Gill, A. E., and Clarke, A. J. (1974). Wind induced upwelling, coastal currents and sea
level changes. Deep-sea Res. 21, 325-345.
Gordon, J. A. (1981). LARM two-dimensional model: An evaluation. J. Enuiron. Eng.
Diu. 107, 877-886.
Graf, W. H., Merzi, N., and Pemnjaquet, C. (1984). Aerodynamic drag measured at a
nearshore platform on Lake Geneva. Arch. Met. Geophys. Bioklim. A33, 151-173.
Graham, D. S. (1980). Destratification characteristics and energy efficiency of air-plume
mixing of stratified fluids. Paper No. 5-20. In “Proc. Symp. on Surface Water
Impoundments,” ASCE pp. 923-932. Minneapolis, Minnesota.
Gregg, M. C. (1987). Diapycnal mixing in the thermocline: A review. J. Geophys. Res. 92,
5249-5286.
Gregg, M. C., and Briscoe, M. G. (1979). Internal waves, finestructure, microstructure and
mixing in the ocean. Rev. Geophys. Space Phys. 17, 1524-1548.
Gregg, M. C., and Sanford, T. B. (1988). The dependence of turbulent dissipation on
stratification in a diffusively stable thermocline. J . Geophys. Res. 93, 12381-12392.
Hannoun, I. A., and List, E. J. (1988). Turbulent mixing at a shear-free density interface.
1. Fluid Mech. 189, 211-234.
Hannoun, I. A., Fernando, H. J. S., and List, E. J. (1988). Turbulence structure near a
sharp density interface. J . Fluid Mech. 189, 189-209.
Harashima, A., and Watanabe, M. (1986). Laboratory experiments on the steady
gravitational circulation excited by cooling of the water surface. J. Geophys. Res. 91,
13056- 13064.
Harleman, D. R. F. (1982). Hydrothermal analysis of lakes and reservoirs. J . Hydr. D i n
ASCE 108,302-325.
Harleman, D. R. F., and Hurley-Octavio, K. A. (1977). Heat transport mechanisms in
lakes and reservoirs. In “Proc. 17th Cong. Intl. Assoc. Hydr. Res.,” Baden
Baden.
Harleman, D. R. F., Morgan, R. L., and Purple, R. A. (1959). Selective withdrawal from a
vertically stratified fluid. In “Proc. 8th Congr. Intl. Assoc. for Hydr. Res.,” Montreal,
pp . 10-C-1-10-C-16.
Hart, J. E. (1972). Stability of thin non-rotating Hadley circulations. J. Amos. Sci. 29,
687-697.
Hart, J. E. (1983a). Low Prandtl number convection between differentially heated end
walls. Intl. J. Heat Mass Transfer 26, 1069-1074.
Hart, J. E. (1983b). A note on the stability of low-Prandl number Hadley circulations. J.
Fluid Mech. l32,271-281.
Hauenstein, W., and Drams, T. H. (1984). Investigation of plunging density currents
generated by inflows in lakes. J . Hydr. Res. 22, 157-179.
Heaps, N. S. (1984). Vertical structure of current in homogeneous and stratified waters. In
“Hydrodynamics of Lakes” (K. Hutter ed.), pp. 153-207. Springer-Verlag. New York.
Heaps, N. S., and Ramsbottom, A. E. (1966). Wind effects on water in a narrow
two-layered lake. Phil. Trans. Roy. SOC.London Ser. A 259, 391-430.
462 Jorg Imberger and John C . Patterson
Heathershaw, A. D., and Martin, J. (1987). The theory and tuning of the Met 0.20 mixed
layer model. Admirality Research Establishment Accn. No. 75636.
Hebbert, R., Imberger, J., h h , I., and Patterson, J. (1979). Collie River underflow into
the Wellington Reservoir. J. Hydr. Div. ASCE 105,533-545.
Henderson-Sellers, B. (1976). Role of eddy diffusivity in thermocline formation. Proc.
ASCE 102,517-531.
Henderson-Sellers, B. (1982). Reservoir destratification techniques and models. In ASCE
“Proc. Applying Res. to Hydraulic Practice Conf. ,” pp. 697-706. Mississippi.
Henderson-Sellers, B. (1984a). Development and application of “U.S.E.D.”: A hydroclim-
ate lake stratification model. Ecol. Modelling 21,233-246.
Henderson-Sellers, B. (1984b). “Engineering Limnology.” Pitman, Marshfield,
Massachusetts.
Henderson-Sellers, B. (1985). New formulation of eddy diffusion thermocline models.
Appl, Math. Modelling 9, 441-446.
Henderson-Sellers, B. (1986). Calculating the surface energy balance for lake and reservoir
modeling: A review. Rev. Geophys. 24, 625-649.
Hicks, B. B. (1972). Some evaluations of drag and bulk transfer coefficients over water
bodies of different sizes. Boundary Layer Met. 3, 201-213.
Hicks, B. B. (1975). A procedure for the formulation of bulk transfer coefficients over
water. Boundary Layer Met. 8, 315-324.
Hicks, B. B. (1975). Wind profile relationships from the ‘Wangara’ experiment. Quart. J.
Roy. Met. SOC.102, 535-551.
Hicks, B. B., Drinklow, R. L., and Grauze, G. (1974). Drag and bulk transfer coefficients
associated with a shallow water surface. Boundary Layer Met. 6,287-297.
Hocking, G. C. (1985). Cusp-like free-surface flows due to a submerged source or sink in
the presence of a flat or sloping bottom. J. Aust. Math. SOC. Ser. B. 26, 470-486.
Hocking, G. C. (1988). Critical withdrawal from a two-layer fluid through a Line sink.
Environmental Dynamics Rept. No. ED-88-240, Centre for Water Research, Univer-
sity of Western Australia.
Hocking, G. C., and Patterson, J. C. (1988). Quasi two-dimensional modelling of
reservoirs. Environmental Dynamics Rept. ED-88-283, Centre for Water Research,
University of Western Australia.
Hocking, G. C., Sherman, B. S., and Patterson, J. C. (1988). Algorithm for selective
withdrawal from stratified reservoir. J. Hydr. Div. ASCE 114,707-719.
Holloway, G. (1986). Eddies, waves, circulation and mixing: Statistical geofluid mechanics.
Ann. Rev. Fluid Mech. 18,91-147.
Holyer, J. Y.,and Huppert, H. E. (1980). Gravity currents entering a two-layer fluid. J.
Fluid Mech. 100,739-767.
Hopfinger, E. J. (1987). Turbulence in stratified fluids: A review. J. Geophys. Res. 92,
5287-5303.
Hopfinger, E. J., and Toly, J. A. (1976). Spatially decaying turbulence and its relation to
mixing across density interfaces. J . Fluid Mech. 78, 155-175.
Horn, W.,Mortimer, C. H., and Schwab, D. J. (1986). Wind-induced internal seiches in
Lake Zurich observed and modeled. Limnol. Oceanogr. 31, 1232-1254.
Howard, C. S. (1953). Density currents in Lake Mead. In “Proc. Intl. Hydr. Convention,”
ASCE. pp. 355-368. Minnesota.
Hsu, S. A. (1974). A dynamic roughness equation and its application to wind stress
determination at the air-sea interface. J . Phys. Oceanogr. 4, 116-120.
Hsu, S. A. (1986). A mechanism for the increase of wind stress (drag) coefficient with wind
speed over water surfaces: A parametric model. J. Phys. Oceanogr. 16, 144-150.
 Physical Limnology 463
Huber, D. G. (1960). Irrotational motion of two fluid strata towards a Line sink. J. Eng.
Mech. Div.86, 71-86.
Huber, W. C., Harleman, D. R. F., and Ryan, P. J. (1972). Temperature prediction in
stratified reservoirs. J. Hydr.Div.ASCE 98, 645-666.
Hurley-Octavio, K. A., Jirka, G. H., and Harleman, D. R. F. (1977). Vertical heat
transport mechanisms in lakes and reservoirs. Tech. Rept. 227, Massachusetts Institute
of Technology.
Hussain, N. A., and Narang, B. S. (1984). Simplified analysis of air-bubble plumes in
moderately stratified environments. J. Heat Transfer 106,543-551.
Hutchinson, G. E. (1957). “A Treatise on Limnology. I. Geography, Physics and
Chemistry.” John Wiley & Sons, New York.
Hutter, K. (1984). “Hydrodynamics of Lakes.” Springer-Verlag, New York.
Hutter, K. (1986). Hydrodynamic modeling of lakes. In “Encyclopedia of Fluid Mechan-
ics,” pp. 897-998. Gulf Publishing, Houston, Texas.
Idso, S. B. (1973). On the concept of lake stability. Limnol. Oceunogr. 18,681-683.
Idso, S. B., and Cole, G. A. (1973). Studies on a Kentucky Knobs Lake: V. Some aspects
of the vertical transport of heat in the hypolimnion. J. Ecol. 61,413-420.
Imberger, J. (1970). Selective withdrawal from a stratified reservoir. Ph.D. thesis,
University of California, Berkeley.
Imberger, J. (1972). Two-dimensional sink flow of a stratified fluid contained in a duct. J.
Fluid Mech. 53, 329-349.
Imberger, J. (1974). Natural convection in a shallow cavity with differentially heated end
walls. Part 3: Experimental results. J. Fluid Mech. 65,247-260.
Imberger, J. (1976). Dynamics of a longitudinally stratified estuary. In “Proc. 15th Coastal
Eng. Conf.,” pp. 3108-3115. ASCE Hawaii.
Imberger, J. (1980). Selective withdrawal: A review. In “Proc. 2nd Intl. Symp. on Stratified
Flows,” Int. Ass. Hydraulic Research (IAHR) Trondheim.
Imberger, J. (1982). Reservoir dynamics modeling. In “Prediction in Water Quality” (E.
M. O’hughlin and P. Cullen eds.), pp. 223-248. Australian Academy of Science,
Canberra.
Imberger, J. (1985). The diurnal mixed layer. Limnol. Oceunogr. 30,737-770.
Imberger, J. (1987). Hydrodynamics of lakes. In “Proc. 12th Convention Australian Water
and Wastewater Assoc. ,” pp. 401-423. AWWA. Adelaide, South Australia.
Imberger, J. (1988). On the nature of turbulence in a stratified water body. Environmental
Dynamics Rept. No. ED-87-216, Centre for Water Research, University of Western
Australia.
Imberger, J., and Boashash, B. (1986). Application of the Wigner-Ville distribution to
temperature gradient microstructure: A new technique to study small-scale variations.
J. Phys. Oceunogr. 16, 1997-2012.
Imberger, J., and Fandry, C. (1975). Withdrawal of a stratified fluid from a vertical
two-dimensional duct. J. Fluid Mech. 70,321-332.
Imberger, J., and Fischer, H. B. (1970). Selective withdrawal from a stratified reservoir.
Water Pollution Control Res. Series 15040 EJZ 12/70, E.P.A. Water Quality
Office.
Imberger, J., and Hamblin, P. F. (1982). Dynamics of lakes, reservoirs and cooling ponds.
Ann. Rev. Fluid Mech. 14, 153-187.
Imberger, J., and Hebbert, R. H. B. (1980). Management of water quality in reservoirs.
Tech. Paper No. 49, Res. Proj. No. 74/70, Aust. Water Res. Council.
Imberger, J., and Monismith, S. G. (1986). A model for deepening due to upwelling when
W > 1. J. Fluid Mech. 171,432-439 (Appendix to Monismith (1986)).
464 Jorg Imberger and John C . Patterson
Imberger, J., and Parker, G. (1985). Mixed layer dynamics in a lake exposed to a spatially
variable wind field. Limnol. Oceanogr. 39,473-488.
Imberger, J., and Patterson, J. C. (1981). A dynamic reservoir simulation model-
DYRESM. In “Transport Models for Inland and Coastal Waters” (H. B. Fischer ed.),
pp. 310-361. Academic Press, New York.
Imberger, J., and Spigel, R. H. (1987). Circulation and mixing in Lake Rotongaio and Lake
Okaro under conditions of light to moderate winds: Preliminary results. N.Z. J . Marine
and Freshwater Res. 21,515-519.
Imberger, J., Thompson, R. 0. R. Y., and Fandry, C. (1976). Selective withdrawal from a
finite rectangular tank. J. Fluid Mech. 78,489-512.
Imberger, J., Patterson, J. C., Hebbert, R., and Loh, I. (1978). Dynamics of reservoirs of
medium size. J. Hydr. Diu. ASCE 104,725-743.
Imboden, D. M., and Emerson, S. (1977). Natural radon and phosphorus as limnologic
tracers: Horizontal and vertical eddy diffusion in Greifensee. Limnol. Oceanogr. 23,
77-90.
Imboden, D. M., and Joller, T. (1984). Turbulent mixing in the hypolimnion of
Baldeggersee (Switzerland) traced by natural radon-222. Limnol. Oceanogr. 29,
831-844.
Imboden, D. M., Weiss, R. F., Craig, H., Michel, R. L., and Goldman, C. R. (1977). Lake
Tahoe geochemical study. 1. Lake chemistry and tritium mixing study. Limnol.
Oceanogr. 22, 1039-1051.
Imboden, D. M., Lemmin, U., Joller, T., and Schurter, M. (1983). Mixing mechanisms in
lakes: Mechanisms and ecological relevance. Schweiz. Z. Hydrol. 45, 11-44.
Inaba, H., Seki, N., Fukusako, S., and Kanyama, K. (1981). Natural convective heat
transfer in a shallow rectangular cavity with different end temperatures. Numer. Heat
Transfer 4, 459-468.
Itsweire, E. C., Helland, K. N., and van Atta, C. W. (1986). The evolution of grid
generated turbulence in a stably stratified fluid. J. Fluid Mech. 162,299-338.
Ivetic, M., Iwasa, Y., and Inoue, K. (1986). Large eddy simulation of a shear driven flow in
a test reservoir. In “Proc. 2nd Intl. Conf. Hydrosoft 86,” (M. Radojkovic, C.
Maksimovic, and C. A. Brebbia, eds.) Southampton. pp. 209-216, Springer Verlag.
Ivey, G. N. (1987). The role of boundary mixing in the deep ocean. J . Geophys. Res. 92,
11873- 11878.
Ivey, G. N., and Blake, S. (1985). Axisymmetric withdrawal and inflow in a density-
stratified container. J. Fluid Mech. 161,115-137.
Ivey, G. N., and Boyce, F. M. (1982). Entrainment by bottom currents in Lake Erie.
Limnol. Oceanogr. 27, 1029-1038.
Ivey, G. N., and Corms, G. M.(1982). Boundary mixing in a stratified fluid. J . Fluid Mech.
121,1-26.
hey, G. N., and Hamblin, P. F. (1989). Convection near the temperature of maximum
density for high Rayleigh number, low aspect ratio, rectangular cavities. J . Heat
Transfer 111,100-105.
Ivey, G. N., and Nokes, R. I. (1989). Vertical mixing due to the breaking of initial waves
on sloping boundaries. J. Fluid Mech. 204,479-500.
Ivey, G. N., and Patterson, J. C. (1984). A model of the vertical mixing in Lake Erie in
summer. Limnol. Oceanogr. 29,553-563.
Jain, S. C. (1982). Buoyancy-driven circulation in free-surface channels. J. Fluid Mech. U2,
1-12.
Janowitz, G. S. (1986). A surface density and wind-driven model of the thermocline. J.
Geophys. Res. 91,5111-5118.
 Physical Limnology 465
Jassby, A., and Powell, T. (1975). Vertical patterns of eddy diffusion during stratification in
Castle Lake, California. Limnol. Oceanogr. 20, 530-543.
Jirka, G. H., and Arita, M. (1987). Density currents or density wedges: Boundary-layer
influence and control methods. 1. Fluid Mech. 177, 187-206.
Jirka, G. H., and Katavola, D. S. (1979). Supercritical withdrawal from two-layered fluid
systems. Part 2: Three-dimensional flow into round intake. J. Hydr. Res. 17,
53-62.
Johnson, T. R., Ellis, C. R., Farrell, G. J., and Stefan, H. G. (1987a). Negatively buoyant
flow in a diverging channel. I: Flow regimes. 1. Hydr. Eng. 1l3,716-730.
Johnson, T. R., Ellis, C. R., Farrell, G. J., and Stefan, H. G. (1987b). Negatively buoyant
flow in a diverging channel. 11: 3-D flow field descriptions. J. Hydr. Eng. lW,731-742.
Johnson, T. R., Ellis, C. R., and Stefan, H. G. (1988). Negatively buoyant flow in a
diverging channel: IV: Entrainment and dilution. J . Hydr. Eng. lW,437-456.
Jokela, J. B., and Patterson, J. C. (1985). Quasi-two-dimensional modelling of reservoir
inflow. In “Proc. 21st IAHR Intl. Congr.,” Vol. 2, pp. 317-322. Melbourne,
Institution of Engineers, Aust. Natl. Conference Publication No. 85/13.
Kamial, J. C., Wyngaard, J. C., Haugen, D. A., Cote, 0. R., and Izumi, Y. (1976).
Turbulence structure in the convective boundary layer. J. A m o s . Sci. 33,2152-2169.
Kantha, L. H., Phillips, 0. M., and &ad, R. S. (1977). On turbulent entrainment at a
stable density interface. J. Fluid Mech. 79, 753-768.
Kao, T. W. (1970). Free streamline theory for inviscid stratified flow into a line sink. Phys.
Fluids W, 558-564.
Kao, T. W. (1976). Selective withdrawal criteria of stratified fluids. J. Hydr. Diu. A X E
102, 717-729.
Kao, T . W., Pao. H. P., and Wei, S. N. (1974). Dynamics of establishment of selective
withdrawal of a stratified fluid from a line sink. Part 2: Experiment. 1. Fluid Mech. 65,
689-710.
Kao, T. W., Park, C., and Pao, H. P. (1978). Inflows, density currents and fronts. Phys.
Fluids 21, 1912-1922.
Kato, H., and Phillips, 0. M. (1969). On the penetration of a turbulent layer into stratified
fluid. J . Fluid Mech. 37,643-655.
Keijman, J. Q. (1974). The estimation of the energy balance of a lake from simple weather
data. Boundary Layer Met. 7,399-407.
Keller, W. C., Plant, W. J., and Weissman, D. E. (1985). The dependence of X band
microwave sea return on atmospheric stability and sea state. 1. Geophys. Res. 90,
1019-1029.
Keulegan, G. G., and Brame, V. (1960). Fourteenth progress report on model laws for
density currents mixing effect of wind induced waves. NBS Project No. 0603-11-06431,
U.S. Dept. of Commerce.
Khomskis, W. R. (1969). “Dynamics and Thermics of Small Lakes” Vilnyus: Mintis (in
Russian).
Khomskis, W. R., and Filatova, T. N. (1972). Principles of typology of stratified lakes in
relation to vertical exchange. Verhandlungen der Internationalen Vereinigung fur
Theoretische und Angewandte Limnologie 18, 528-536.
Kielmann, J., and Simons, T. J. (1984). Some aspects of baroclinic circulation models. In
“Hydrodynamics of Lakes” (K. Hutter ed.), pp. 235-285. Springer-Verlag, New York.
Kim, B. R., Higgins, J. M., and Bruggink, D. J. (1983). Reservoir circulation patterns and
water quality. J . Enuiron. Eng. 109, 1284-1294.
Kirk, J. T. 0. (1983). “Light and Photosynthesis in Aquatic Ecosystems.” Cambridge
University Press, Cambridge.
466 Jorg Imberger and John C. Patterson
Kitaigorodskii, S. A. (1968). On the calculation of the aerodynamic roughness of the sea
surface. Izu. A m o s . Oceanic Phys. 4, 498-502.
Kitaigorodskii, S. A., and Zaslavskii, M. M. (1974). A dynamical analysis of the drag
conditions at the sea surface. Boundary Layer Met. 6 , 53-61.
Kitaigorodskii, S. A., Kutznetsov, 0. A., and Panin, G. N. (1973). Coefficients of drag,
sensible heat and evaporation in the atmosphere over the surface of the sea. Izu.
A m o s . Oceanic Phys. 9, 1135-1141.
Kitaigorodskii, S. A., Donelan, M. A., Lumley, J. A., and Terray, E. A. (1983).
Wave-turbulence interactions in the upper ocean. Part 11: Statistical characteristics of
wave and turbulent components of the random velocity field in the marine surface
layer. J . Phys. Oceanogr. l3(11), 1988-1999.
Kobus, H. E. (1968). Analysis of the flow produced by air bubble systems. In “Proc. 11th
Coastal Eng. Conf.,” pp. 1016-1031. ASCE London.
Koh, R. C. Y. (1966a). Unsteady stratified flow into a sink. J. Hydr. Res. 4(2), 21-34.
Koh, R. C. Y. (1966b). Viscous stratified flow towards a sink. J. Fluid Mech. 24,
555-575.
Kondo, J. (1975). Air-sea bulk transfer coefficients in diabatic conditions. Boundary Layer
Met. 9, 91-112.
Kranenburg, C. (1979). Destratification of lakes using bubble columns. J. Hydr. Diu.
ASCE 105, 333-349.
Kranenburg, C. (1984). Wind-induced entrainment in a stably stratified fluid. J . Fluid
Mech. 145, 253-273.
Kranenburg, C. (1985). Mixed-layer deepening in lakes after wind set-up. J. Hydr. Diu.
ASCE. 111, 1279-1297.
Kraus, E. B. (1977). “Modeling and Prediction of the Upper Layers of the Ocean.”
Pergamon Press, New York.
Kraus, E. B., and Turner, J. S. (1967). A one-dimensional model of the seasonal
thermocline: 11. The general theory and its consequences. Tellus 14,98-106.
Kullenberg, G., Murthy, C. R., and Westerberg, H. (1974). Vertical mixing characteristics
in the thermocline and hypolimnion regions of Lake Ontario. (IFYGL). In “Proc. 17th
Conf. Great Lakes Res.,” Vol. 1, pp. 425-434. Intl. Assoc. Great Lake Res. Ann
Arbor.
Lake Biwa Research Institute (1987). Fundamental studies on the vertical transport of
materials in Lake Biwa. Rept. No. 86-A05.
Langmuir, I. (1938). Surface motion of water induced by wind. Science 87, 119-123.
Large, W. G., and Pond, S. (1981). Open ocean momentum flux measurements in
moderate to strong winds. J. Phys. Oceanogr. 11, 324-336.
Lawrence, G. A., and Imberger, J. (1979). Selective withdrawal through a point sink in a
continuously stratified fluid with a pycnocline. Environmental Dynamics Rept. No.
ED-79-002, Centre for Water Research, University of Western Australia.
Lawrence, G. A., Lasheras, J. C., and Browand, F. K. (1987). Shear instabilities in
stratified flow. Preprint from “3rd Intl. Symp. on Stratified Flows,” Vol. 1. Pasadena,
California.
Leibovich, S. (1977a). Convective instability of stably stratified water in the ocean. J. Fluid
Mech. 82, 561-585.
Leibovich, S. (1977b). On the evolution of the system of wind drift currents and Langmuir
circulations in the ocean. Part 1: Theory and the averaged current. 1. Fluid Mech. 79,
715-743.
Leibovich, S. (1980). On wave-current interaction theories of Langmuir circulation. J. Fluid
Mech. 99, 715-724.
 Physical Limnology 467
Leibovich, S. (1983). The form and dynamics of Langmuir circulations. Ann. Rev. Fluid
Mech. 15,391-427.
Leibovich, S . , and Paolucci, S. (1980). The Langmuir circulation instability as a mixing
mechanism in the upper ocean. J . Phys. Oceanogr. 10, 186-207.
Leibovich, S., and Radhakrishnan, K. (1977). On the evolution of the system of wind drift
currents and Langmuir circulations in the ocean. Part 2: Structure of the Langmuir
vortices. J. Fluid Mech. 80, 481-507.
Lerman, A. (1971). Time to chemical steady-state in lakes and ocean. A d v . in Chem. Ser.
106, 30-76.
Lerman, A., and Stiller, M. (1%9). Vertical eddy diffusion in Lake Tiberias.
Verhandlungen der Internationalen Vereinigung fur Theoretische und Angewandte
Limnologie 17,323-333.
Lewis, W. M. (1973). The thermal regime of Lake Lanao (Philipines) and its theoretical
implication for tropical lakes. Limnol. Oceanogr. 18, 200-217.
Lewis, E. L., and Perkin, R. G. (1982). Seasonal mixing processes in an Arctic fjord
system. J. Phys. Oceanogr. 12, 74-83.
Lighthill, M. J. (1969). Dynamic response of the Indian Ocean to the southwest monsoon.
Phil. Trans. R. SOC.London Ser. A . 26J, 45-92.
Lin, J. T., and Pao, Y. H. (1979). Wakes in stratified fluids. Ann. Rev. Fluid Mech. 11,
317-338.
Linden, P. F. (1975). The deepening of a mixed layer in a stratified fluid. J. Fluid Mech. 71,
385-405.
Linden, P. F., and Simpson, J. E. (1986). Gravity-driven flows in a turbulent fluid. J. Fluid
Mech. 172, 481-497.
Lo, A. K., and McBean, G. A. (1978). On the relative errors in methods of flux
calculations. J. Appl. Met. 17, 1704-1711.
Loftler, H. (1968). Tropical high-mountain lakes. Their distribution, ecology and zoo-
geographical importance. Colloquium Geographicum 9 , 57-76.
Long, R. R. (1956). Sources and sinks at the axis of a rotating liquid. Quart. J . Mech. and
Appl. Math. 9 , 385-393.
Lueck, R. G., and Osborn, T. R. (1985). Turbulence measurements in a submarine canyon.
Continental Shelf Res. 4, 681-698.
Luketina, D. A., and Imberger, J. (1987). Characteristics of a surface buoyant jet. J .
Geophys. Res. 92, 5435-5447.
Luketina, D. A,, and Imberger, J. (1988). Turbulence and entrainment in a buoyant
surface plume. J. Geophys. Res. (in press).
MacIntyre, S., and Melack, J. M. (1982). Meromixis in an equatorial African soda lake.
Limnol. Oceanogr. 27, 595-609.
Mahony, J. J., and Pritchard, W. G. (1977). Withdrawal from a reservoir of stratified fluid.
Rept. No. 79, Fluid Mech. Res. Institute, University of Essex.
Mahrt, L., and Lenschow, D. H. (1976). Growth dynamics of the convectively mixed layer.
J . A m o s . Sci. 33, 41-51.
Manins, P. C. (1976). Intrusion into a stratified fluid. J. Fluid Mech. 74, 547-560.
Markofsky, M., and Harleman, D. R. F. (1973). Prediction of water quality in stratified
reservoirs. J. Hydr. Div. ASCE 99, 729-745.
Marmoush, Y. R., Smith, A. A., and Hamblin, P. F. (1984). Pilot experiments on thermal
bar in lock exchange flow. J. Energy Eng. 110,215-227.
Marti, V. D. E., and Imboden, D. M. (1986). Thermische energieflusse an der
wasseroberflache: beispiel sempachersee. Schweiz. 2. Hydrol. 48, 196-229.
Maxworthy, T. (1972). Experimental and theoretical studies of horizontal jets in a
468 Jorg Zmberger and John C. Patterson
stratified fluid. In “IAHR Intl. Symp. on Stratified Flows,” Novosibirsk. IAHR. pp.
611-619.
Maxworthy, T. (1980). On the formation of nonlinear internal waves from the gravitational
collapse of mixed regions in two and three dimensions. J . Fluid Mech. 96, 47-72.
Maxworthy, T., and Monismith, S. G. (1988). Differential mixing in a stratified fluid. J.
Fluid Mech. 189,571-598.
McBean, G. A., and Paterson, R. D. (1975). Variations of the turbulent fluxes of
momentum, heat and moisture over Lake Ontario. J. Phys. Oceanogr. 5, 523-531.
McDougall, T. J. (1978). Bubble plumes in stratified environments. J. Fluid Mech. 85,
655-672.
McEwan, A. D., and Baines, P. G. (1974). Shear fronts and an experimental stratified
shear flow. J . Fluid Mech. 63, 257-272.
Mellor, G. L., and Durbin, P. A. (1975). The structure and dynamics of the ocean surface
mixed layer. J. Phys. Oceanogr. 5,718-728.
Mellor, G. L., and Yamada, T. (1982). Development of a turbulence closure model for
geophysical fluid problems. Rev. Geophys. Space Phys. 29,851-875.
Melville, W. K. (1977). Wind stress and roughness length over breaking waves. J. Phys.
Oceanogr. 7,702-710.
Milgram, J. H. (1983). Mean flow in round bubble plumes. J. Fluid Mech. l33,345-376.
Miyake, M., Donelan, M., McBean, G., Paulson, C., Badgley, F.,and Leavitt, E. (1970).
Comparison of turbulent fluxes over water determined by profile and eddy correlation
techniques. Quart. J . Roy. Met. SOC. 96, 132-137.
Monin, A. S., and Obukhov, A. M. (1954). Basic laws of turbulent mixing in the
atmosphere near the ground. Jr. Akad. Nauk SSSR Geofiz. Znst. 24(151), 163-187.
Monismith, S. G. (1985). Wind-forced motions in stratified lakes and their effect on
mixed-layer shear. Limnol. Oceanogr. 30,771-783.
Monismith, S . G. (1986). An experimental study of the upwelling response of stratified
reservoirs to surface shear stress. J. Fluid Mech. 171,407-439.
Monismith, S. G. (1987). The modal response of reservoirs to wind stress. J . Hydr. Diu.
ASCE 1l3,1290-1326.
Monismith, S. G., and Imberger, J. (1988). Horizontal convection in the sidearm of a small
reservoir. Environmental Dynamics Rept. ED-86-161, Centre for Water Research,
University of Western Australia.
Monismith, S . G., and Maxworthy, T. (1988). Selective withdrawal and spin-up of a
rotating, stratified fluid. J . Fluid Mech. 199,377-401.
Monismith, S. G., Imberger, J., and Billi, G. (1988). Shear-waves and unsteady selective
withdrawal. J . Hydr. Diu. ASCE 114,1134-1152.
Mortimer, C. H. (1974). Lake hydrodynamics. 1.V .L . u),124.
Mortimer, C. H. (1987). Fifty years of physical investigations and related limnological
studies on Lake Erie, 1928-1977. J . Great Lakes Res. l3,407-435.
Mulhearn, P. J. (1981). On the formation of a stably stratified internal boundary-layer by
advection of warm air over a cooler sea. Boundary Layer Met. 21,247-254.
Muller, P., and Garwood, R. W. (1988). Mixed layer dynamics: Progress and new
directions. EOS 69, 2-4, 10-12.
Muller, P., and Henderson, D. (1987). Dynamics of the oceanic surface mixed layer. In
“Proc. Aha Huliko’a Hawaiian Winter Workshop,” Hawaii Institute of Geophysics
Special Publication, Honolulu.
Muller, P., and Pujalet, R. (1984). Internal gravity waves and small-scale turbulence. In
“Proc. Aha Huliko’a Hawaiian Winter Workshop,” Hawaii Institute of Geophysics
Special Publication, Honolulu.
 Physical Limnology 469
Munk, W., and Anderson, E. R. (1948). Notes on a theory of the thermocline. J. Marine
Res. 7 , 276-295.
Murota, A., and Hirata, T. (1979). Breaking of internal gravity wave on slope. Tech. Rep&.
of Osaka University 29(1487), 255-264.
Murota, A., and Michioku, K. (1986a). Analysis on selective withdrawal from three-layered
stratified systems and its practical application. J. Hydrosci. Hydr. Eng. 4, 31-50.
Murota, A., and Michioku, K. (1986b). Characteristics of internal fluctuations in thermally
stratified fields induced by combination of mechanical and thermal stirring. Proc. JSCE
375(II-6), 171-179.
Murota, A., Hirata, T., and Michioku, K. (1980). Breaking of internal gravity wave on
slope: 11. Tech. Rep&. of Osaka University 30(1543), 255-261.
Murota, A., Michioku, K., and Kuchida, M. (1988). Numerical analysis on vertical mixing
phenomena induced by composite stirring. Proc. JSCE 11-9, 67-76.
Myrup, L. 0.. Powell, T. M., Godden, D. A., and Goldman, C. R. (1979). Climatological
estimate of the average monthly energy and water budgets of Lake Tahoe, California-
Nevada. Water Resources Res. 15, 1499-1508.
Narimousa, A., and Fernando, H. J. S. (1987). On the sheared-density interface of an
entraining stratified fluid. J . Fluid Mech. 174, 1-22.
Narimousa, S., Long, R. R., and Kitaigorodskii, S . A. (1986). Entrainment due to
turbulent shear flow at the interface of a stably stratified fluid. Tellus %A,
7687.
Newman, F. C. (1976). Temperature steps in Lake Kivu: A bottom heated saline lake. J.
Phys. Oceanogr. 6 , 157-163.
Niiler, P. P. (1975). Deepening of the wind-mixed layer. J . Marine Res. 33,405-422.
Niiler, P. P., and Kraus, E. B. (1977). One-dimensional models of the upper ocean. In
“Modeling and Prediction of the Upper Layers of the Ocean” (E.B. Kraus ed.), pp.
143-172, Pergamon Press, New York.
Nishida, S., and Yoshida, S . (1984). Stability of a two-layer shear flow. Theoretical and
App. Mech. 32, 35-45.
Nokes, R. I. (1988). On the entrainment rate across a density interface. 1. Fluid Mech. 188,
185-204.
O’Donnell, J., and Garvine, R. W.(1983). A time dependent, two-layer frontal model of
buoyant plume dynamics. Tellus 35A, 73-80.
Ookubo, K., and Muramoto, Y. (1981). Coriolis’ effects on density CurrentsExperiments
on the rotating exchange flow. Bulletin of the Disaster Prevention Research Institute
24B-2,339-365.
Ookubo, K., Muramoto, Y., Oonishi, Y., and Kumagai, M. (1984). Laboratory experi-
ments on thermally induced currents in Lake Biwa. Bulletin of the Dbaster Prevention
Research Institute 34/2(304), 19-54.
Orlob, G. T., and Selna, L. G. (1970). Temperature variations in deep reservoirs. 1. Hydr.
Div. ASCE 96.391-410.
Panofsky, H. A. (1983). Determination of stress from wind and temperature measure-
ments. Quart. J. Roy. Meteor. SOC.89, 85-94.
Pao, H. P., and Kao, T. W. (1969). Sources and sinks at the axis of a viscous rotating fluid.
Phys. Fluiah 12, 1536-1546.
Pao, H. P., and Kao, T. W. (1974). Dynamics of establishment of selective withdrawal of a
stratified fluid from a line sink. Part I: Theory. 1. Fluid Mech. 65, 657-688.
Pao, H. P., and Shih, T. H. (1973). Selective withdrawal and blocking wave in rotating
fluids. J . Fluid Mech. 57,459-480.
Parker, G. J., and Imberger, J. (1986). Differential mixed-layer deepening in lakes and
470 Jorg Imberger and John C . Patterson
reservoirs. In “Limnology in Australia” (P. de Deckker and W. D. Williams eds.), pp.
63-92. W. Junk Publishers, Melbourne, Victoria.
Patalas, K. (1960). Mieszanie wody jako czynnik okreslajacy intensywnose krazenia materii
w roznych morfologicznie jeziorach okolic Wegorzewa. Roczniki Nauk Rolniczych.
B77,223-242.
Patalas, K. (1961). Wind und morphologiebedingte wasserwegunstypen als bestimmender
Faktor fur die Intersitat des Stoffkreislaufes in nordpolnischen Seen. Verhandlungen
der Internationalen Vereinigung fur Theoretische und Angewandte Limnologie. 14,
59-64.
Patterson, J. C. (1984). Unsteady natural convection in a cavity with internal heating and
cooling. J. Fluid Mech. 140, 135-151.
Patterson, J. C., and Hamblin, P. F. (1988). Thermal simulation of a lake with winter ice
cover. Limnol. Oceanogr. 33, 323-338.
Patterson, J., and Imberger, J. (1980). Unsteady natural convection in a rectangular cavity.
J , Fluid Mech. 100,65-86.
Patterson, J. C., and Imberger, J. (1989). Simulation of bubble plume destratification
systems in reservoirs. Aquatic Sciences 51,3-18.
Patterson, J. C., Hamblin, P. F., and Imberger, J. (1984). Classification and dynamic
simulation of the vertical density structure of lakes. Limnol. Oceanogr. 29, 845-861.
Patterson, J. C., Allanson, B. R., and hey, G. N. (1985). A dissolved oxygen budget
model for Lake Erie in summer. Freshwater Biology 15,683-694.
Paulson, C. A. (1970). The mathematical representation of wind speed and temperature
profiles in the unstable atmospheric surface layer. J . Appl. Met. 9, 857-861.
Peterson, E. W. (1969). Modification of mean flow and turbulent energy by a change in
surface roughness under conditions of neutral stability. Quart. J . Roy. Met SOC.95,
561-575.
Phillips, 0 . M., Shyu, J. H., and Salmun, H. (1986). An experiment on boundary mixing:
Mean circulation and transport rates. J. Fluid Mech. 173,473-499.
Pollard, R. T. (1977). Observations and theories of Langmuir circulations and their role in
near suface mixing. In “A Voyage of Discovery: George Deacon 70th Anniversary
Volume” (M. Angel ed.), pp. 235-251. Pergamon Press, New York.
Pollard, R. T., Rhines, P. B., and Thompson, R. 0. R. Y.(1973). The deepening of the
wind mixed layer. Geophys. Fluid Dyn. 3, 381-404.
Pond, S.,Fissel, D. B., and Paulson, C. A. (1974). A note on bulk aerodynamic coefficients
for sensible heat and moisture fluxes. Boundary Layer Met. 6 , 333-339.
Price, J. F., Mooers, C. N. K., and Van Leer, J. C. (1978). Observations and simulation of
storm-induced mixed-layer deepening. J. Phys. Oceanogr. 8 , 582-599.
Raphael, J. M. (1962). Prediction of temperature in rivers and reservoirs. J . Power Diu. 88,
157- 182.
Rayner, K. N. (1981). Diurnal energetics of a reservoir surface layer. Environmental
Dynamics Rept. ED-80-005,Centre for Water Research, University of Western
Australia.
Robarts, R. D., and Ward, P. R. B. (1978). Vertical diffusion and nutrients transport in a
tropical lake (Lake McIlwaine, Rhodesia). Hydrobiologia 59, 213-221.
Robarts, R. D., Ashton, P. J., Thornton, J. A., Taussig, H. J., and Sephton, L. M. (1982).
Overturn in a hypertonic, warm, monomictic impoundment (Hartbeespoort Dam,
South Africa). Hydrobiologia 97, 209-224.
Roberts, P. J. W., and Matthews, P. R. (1987). Behavior of low buoyancy jets in a linearly
stratified fluid. 1. Hydr. Res. 25, 503-519.
Rodgers, G. IG. (1971). Field investigation of the thermal bar in Lake Ontario: Precision
 Physical Limnology 47 1
temperature measurements. In “Proc. 14th Conf. Great Lakes Res.,” pp. 618-624.
Intl. Assoc. Great Lakes Res. Ann Arbor.
Rodi, W. (1982). “HMT, Vol. 6: Turbulent buoyant jets and plumes.” Pergamon Press,
Sydney.
Rodi, W. (1987). Examples of calculation methods for flow and mixing in stratified fluids. J.
Geophys. Res. 92(C5), 5305-5328.
Rohr, J. J., Itsweire, E. C., Helland, K. N., and Van Atta, C. W.(1988). An investigation
of the growth of turbulences in a uniform-mean-shear flow. 1. Fluid Mech. 187,
1-33.
Sadhuram, Y., Vethamony, P., Suryanarayana, A., Swamy, G. N., and Sastry, J. S. (19b8).
Heat energy exchange over a large water body under stable atmospheric conditions.
Boundary Layer Met. 44, 171-180.
Sahlberg, J. (1983). A hydrodynamical model for calculating the vertical temperature
profile in lakes during cooling. Nordic Hydrology 239-254.
Sanderson, B., Perry, K., and Pederson, T. (1986). Vertical diffusion in meromictic Powell
Lake, British Columbia. J. Geophys. Res. 91, 7647-7655.
Sargent, R. E., and Jirka, G. H. (1982). A comparative study of density currents and
density wedges (with application to intermediate field dynamics of ocean thermal
energy conversion plants). Contract No. ET-78-S-02-4683, U.S.Dept. of Energy.
Savage, S. B., and Brimberg, J. (1975). Analysis of plunging phenomena in water
resources. J. Hydr. Res. W, 187-204.
Schertzer, W. M., Saylor, J. H., Boyce, F. M., Robertson, D. G., and Rosa, F. (1987).
Seasonal thermal cycle of Lake Erie. J. Great Lakes Res. 13,468-486.
Scott, C. F., and Imberger, J. (1988). Three-dimensional estuary circulation and class-
ification. In “Proc. Intl. Conf. Phys. of Shallow Estuaries and Bays, (R. Cheng, ed.)
Springer-Verlag. Pacific Grove, California. Lecture Notes on Coastal and Estuarine
Studies.
Scott, J. T., Myer, G. E., Stewart, R., and Walther, E. G. (1969). On the mechanism of
Langmuir circulations and their role in epilimnion mixing. Limnol. Oceanogr. 14,
493-503.
Shay, T. J., and Gregg, M. C. (1984). Turbulence in an oceanic convective mixed layer.
Nature 310, 282-285.
Shay, T. J., and Gregg, M. C. (1986). Convectively driven turbulent mixing in the upper
ocean. J. Phys. Oceanogr. 16, 1777-1798.
Shay, T. J., and Imberger, J. (1988). Mixing in a strongly stratified low-energy lake.
Environmental Dynamics Rept. ED-88-255, Centre for Water Research, University of
Western Australia.
Sherman, F. S., Imberger, J., and Corcos, G. M. (1978). Turbulence and mixing in stably
stratified waters. Ann. Rev. Fluid Mech. 10, 267-288.
Shih, H. H., and Pao, H. S. (1971). Selective withdrawal in rotating fluids. J. Fluid Mech.
49,509-527.
Shih, T. H., Lumley, J. L., and Janicka, J. (1987). Second-order modeling of a
variable-density mixing layer. J. Fluid Mech. 180, 93-116.
Silvester, R. (1979). An experimental study of end wall effects on selective withdrawal of
linearly stratified liquid from a reservoir. Ph.D. thesis, University of Western
Australia.
Simpkins, P. G., and Dudderar, J. D. (1981). Convection in rectangular cavities with
differentially heated end walls. J. Fluid Mech. 110, 433-456.
Simpson, J. E. (1982). Gravity currents in the laboratory, atmosphere, and ocean. Ann.
Rev. Fluid Mech. 14, 213-234.
472 Jorg Imberger and John C. Patterson
Smith, S. D. (1980). Wind stress and heat flux over the ocean in gale force winds. J. Phys.
Oceanogr. 5 , 709-726.
Smith, S. D., and Anderson, R.J. (1984). Spectra of humidity, temperature and wind over
the sea at Sable Island, Nova Scotia. J . Geophys. Res. 89,2029-2040.
Smith, S. D., and Banke, E. G. (1975). Variation of the sea surface drag coefficient with
wind speed. Quart. J. Roy. Met. SOC.101, 665-673.
Smith, J., Pinkel, R., and Weller, R. A. (1987). Velocity structure in the mixed layer
during MILDEX. J . Phys. Oceanogr. 17,425-439.
Smyth, N. F., and Holloway, P. E. (1988). Hydraulic jump and undular bore formation on
a shelf break. J . Phys. Oceanogr. 18,947-962.
Sorbjan, Z.(1986). On similarity in the atmospheric boundary layer. Boundary Layer Met.
34, 377-397.
Spain, J. D., Wernert, G. M., and Hubbard, D. W. (1976). The structure of the spring
thermal bar in Lake Superior. J. Great Lakes Res. 2, 296306.
Spalding, D. B., and Svensson, U. (1977). The development and erosion of the
thermocline. In “Proc. Seminar Intl. Centre for Heat and Mass Transfer,” Hemisphere
Publ. Corp. Washington D.C. Vol. 1, pp. 113-122.
Spigel, R. H. (1978). Wind mixing in lakes. Ph.D. thesis, University of California,
Berkeley.
Spigel, R. H. (1980). Coupling of internal wave motion with entrainment at the density
interface of a two-layer lake. J . Phys. Oceanogr. 10, 144-155.
Spigel, R. H., and Farrant, B. (1984). Selective withdrawal through a point sink and
pycnocline formation in a linearly stratified flow. 1. Hydr. Res. 22, 35-51.
Spigel, R. H.,and Imberger, J. (1980). The classification of mixed layer dynamics in lakes
of small to medium size. J . Phys. Oceanogr. 19, 1104-1121.
Spigel, R. H., Imberger, J., and Rayner, K. N. (1986). Modeling the diurnal mixed layer.
Limnol. Oceanogr. 31,533-556.
Stacey, M. W., and Bowen, A. J. (1988). The vertical structure of density and turbidity
currents: Theory and observations. J. Geophys. Res. 93,3528-3542.
Stefan, H.G.,and Ford, D. E. (1975). Temperature dynamics in dimictic lakes. J . Hydr.
Diu. A X E 101,97-114.
Stefan, H. G., Cardoni, J. J., Schiebe, F. R., and Cooper, C. M. (1983). Model of light
penetration in a turbid lake. Water Resources Res. 19, 109-120.
Steinhorn, I., and Assaf, G. (1980). The physical structure of the Dead Sea water column:
1975-1977. In “Hypersaline Brines and Evaporitic Environments” (A. Nissenbaum
ed.), pp. 145-153. Elsevier, Amsterdam.
Steinhorn, I., and Gat, J. R. (1983). The Dead Sea. Scienttjic American. 249(4), 102-109.
Stewart, R. W. (1974). The air-sea momentum exchange. Boundary Layer Met. 6 ,
151-167.
Stocker, T., and Hutter, K. (1986). One-dimensional models for topographic Rossby waves
in elongated basins on the f-plane. J . Fluid Mech. 170,435-459.
Straskraba, M. (1980). The effects of physical variables on freshwater production: Analyses
based on models. In “The Functioning of Freshwater Ecosystems” (E. D. Le Cren and
R. H. McConnell eds.), pp. 13-84. Cambridge University Press, England.
Strub, P. T. (1983).The response of a small lake to atmospheric forcing during fall cooling.
Ph.D. thesis, University of California.
Strub, P. T., and Powell, T. M. (1986). Wind-driven surface transport in stratified closed
basins: Direct versus residual circulations. J. Geophys. Res. 91, 8497-8508.
Strub, P. T., and Powell, T. M. (1987). The exchange coefficients for latent and sensible
heat flux over lakes: Dependence upon atmospheric stability. Boundary Layer Met. 40,
349-361.
 Physical Limnology 473
Stull, R. (1976). The energetics of entrainment across a density interface. J . Amos. Sci. 33,
1260- 1267.
Sturm, T. W. (1981). Laminar gravitational convection of heat in dead-end channels. J .
Fluid Mech. 110, 97-113.
Sun, T. Y., and Faeth, G. M. (1986). Structure of turbulent bubbly jets-1. Methods and
centerline properties. Inrl. J. Mukiphare Flow 12, 99-114.
Sundaram, T. R., and Rehm, R. G. (1973). The seasonal thermal structure of deep
temperate lakes. Tellus 25, 157-168.
Svensson, U. (1978). A mathematical model of the seasonal thermocline. Rept. No. 1002,
Dept. of Water Resources Eng., University of Lund, Sweden.
Sverdrup, H. U. (1945). “Oceanography for Meteorologists.” George Allen and Unwin,
London.
Swinbank, W. C. (1963). Long-wave radiation from clear skies. Quurr. J . Roy. Met. SOC.
89,339-348.
Tacke, K. H., Schubert, H. G., Weber, D. J., and Schwerdtfeger, K. (1985). Characteris-
tics of round vertical gas bubble jets. Metallurgical Trans. B. 16, 263-275.
Taylor, G. I. (1954). The dispersion of matter in turbulent flow through a pipe. Proc. Roy.
SOC. London Ser. A . 223,446-468; (1960). Sci. Pap. 2 , 466-488.
Taylor, P. A. (1970). A model of airflow above changes in surface heat flux, temperature
and roughness for neutral and unstable conditions. Boundary Layer Mer. 1,474-497.
Taylor, M., and Aquise, E. (1984). A climatological energy budget of Lake Titicaca
(Peru/Bolivia). I.V.L . 22, 1246-1251.
Tennekes, H., and Lumley, J. L. (1972). “A First Course in Turbulence,” MIT Press,
Cambridge, Massachusetts.
Tennessee Valley Authority (1972). Heat and Mass Transfer Between a Water Surface and
the Atmosphere. Lab. Rept. No. 14, Noms, Tennessee.
Thomas, J. H. (1975). A theory of steady wind-driven currents in shallow water with
variable eddy viscosity. 1. Phys. Oceanogr. 5 , 136-142.
Thompson, R. 0. R. Y., and Imberger, J. (1980). Response of a numerical model of a
stratified lake to wind stress. In “Proc. 2nd IAHR Symp. on Stratified Flows,
Trondheim.” (T. Carstens and T. McClimans eds.), pp. 562-570. IAHR.
Thorpe, S. A. (1969). Experiments on stability of stratified shear flow. Radio Sci. 4,
1327- 1331.
Thorpe, S . A. (1973). Turbulence in stably stratified fluids: A review of laboratory
experiments. Boundary Layer Mer. 5.95-119.
Thorpe, S . A. (1974). Near-resonant forcing in a shallow two-layer fluid: A model for the
internal surge in Loch Ness? J . Fluid Mech. 63, 509-527.
Thorpe, S. A. (1978). The near-surface ocean mixing layer in stable heating conditions. J .
Geophys. Res. 8 , 2875-2885.
Thorpe, S. A. (1982). On the layers produced by rapidly oscillating a vertical grid in a
uniformly stratified fluid. J. Fluid Mech. W, 391-409.
Thorpe, S . A. (1987). Transitional phenomena and the development of turbulence in
stratified fluids: A review. J. Geophys. Res. 92, 5231-5248.
Thorpe, S. A., and Hall, A. J. (1982). Observations of the thermal structure of Langmuir
circulation. J. Fluid Mech. 114, 237-250.
Tolland, H. G. (1977). Destratification/aeration in reservoirs. Tech. Rept. TR 50. Water
Research Centre, England.
Trevisan, 0. V., and Bejan, A. (1986). Convection driven by the non-uniform absorption
of thermal radiation at the free surface of a stagnant pool. Numerical Hear Transfer 10,
483-506.
Trusturn, K. (1964). Rotating and stratified flow. J . Fluid Mech. 19, 415-432.
474 Jorg Zmberger and John C . Patterson
Tsihrintzis, V. A., and Alavian, V. (1987). A model for predicting the properties of
three-dimensional negatively buoyant starting plumes on a sloping surface. In “Proc.
3rd Intl. Symp. on Stratified Flows,” California Institute of Technology, Pasadena.
IAHR.
Tuck, E. O., and Vanden Broeck, J. M. (1984). A cusp-like free-surface flow due to a
submerged source or sink. J. Aust. Math. SOC. Ser. B 25, 443-450.
Tucker, W. A,, and Green, A. W. (1977). A time-dependent model of the lake averaged,
vertical temperature distribution of lakes. Limnol. Oceanogr. 22, 687-699.
Turner, J. S. (1985). Multicomponent convection. Ann. Rev. Fluid Mech. 17, 11-44.
Tzur, Y. (1973). One-dimensional diffusion equations for the vertical transport in an
oscillating stratified lake of varying cross-section. Tellus 25, 266-271.
UNESCO (1981). Tenth report of the joint panel on oceanographic tables and standards.
Sidney, B.C., Canada.
U.S. Army Coastal Engineering Research Center (1977). “Shore Protection Manual,” Vol.
1, 3.35-3.37. U.S. Govt. Printing Office,Washington.
U.S. Army Corps of Engineers (1982). CE-QUAL-R1: A numerical one-dimensional
model of reservoir water quality. Instr. Rept. No. E-82-1, Waterways Experiment
Station.
Vanden Broeck, J. M., and Keller, J. B. (1987). Free surface flow due to a sink. J. Fluid
Mech. 175, 109-117.
Venkatram, A. (1977). A model of internal boundary-layer development. Boundary Layer
Met. 11,419-437.
Venkatram, A. (1986). An examination of methods to estimate the height of the coastal
internal boundary layer. Boundary Layer Met. 36, 149-156.
Ventz, D. (1973). Die Einzugsgebietsgrosse, ein Geofaktor fur den Trophiczustand
stehender Gewasser. Forrschnitte der Wusserchemie und ihrer Grenzgebiete 14, 105-
118.
Veronis, G. (1967). Analogous behavior of homogeneous, rotating fluids and stratified,
nonrotating fluids. Tellus 19,620-634.
Walesh, S. G., and Monkmeyer, P. L. (1973). Bottom withdrawal of viscous stratified fluid.
J. Hydr. Div. ASCE 99, 1401-1418.
Wallace, B. C., and Wilkinson, D. L. (1988). Run-up of internal waves on a gentle slope in
a two-layered system. J. Fluid Mech. 191,419-442.
Ward, P. R. B. (1977). Diffusion in lake hypolimnia. In “Proc. 17th Congr. Intl. Assoc.
Hydr. Res.,” Vol. 2, pp. 103-110. Baden-Baden. IAHR.
Water Resources Engineers, Inc. (1969). Mathematical models for prediction of thermal
energy changes in impoundments. U.S. E.P.A. Water Pollution Control Res. Series,
16130 EXT, Contract 14-22-422. U.S. Govt. Printing Office, Washington, D.C.
Wedderburn, E. M. (1906). The temperature of the freshwater lochs of Scotland with
special reference to Loch Ness. Trans. Roy. SOC. Edinburgh 45, 407-490.
Wedderburn, E. M. (1911). The temperature seiche: Observations in the Madusee. Trans.
Roy. SOC. Edinburgh 47,619-642.
Wedderburn, E. M. (1912). Temperature observations in Loch Earn with a further
contribution to the hydrodynamical theory of temperature seiches. Trans. Roy. SOC.
Edinburgh 48,629-695.
Weinstock, J. (1984). Effect of gravity waves on turbulence decay in stratified fluids. J.
Fluid Mech. 140,11-26.
Weiss, W., Lehn, H., Munnich, K. O., and Fischer, K. H. (1979). On the deep-water
turnover of Lake Constance. Arch. Hydrobiol. 86, 405-422.
Weller, R. A., Dean, J. P., Marra, J., Price, J. F., Francis, E. A., and Boardman, D. C.
 Physical Limnology 475

(1985). Three dimensional flow in the upper ocean. Science 227, 1552-1556.
Wetzel, R. G. (1975). “Limnology.” W. B. Saunders, Philadelphia, Pennsylvania.
Wieringa, J. (1974). Comparison of three methods for determining strong wind stress over
Lake Flevo. Boundary Layer Met. 7,3-19.
Wilkinson, D. L. (1972). Dynamics of contained oil slicks. J. Hydr. Diu. ASCE 98,
1013-1030.
Wilkinson, D. L. (1979). Two-dimensional bubble plumes. J. Hydr. Diu. ASCE 105,
139-154.
Wilkinson, D. L., and Wood, I. R. (1972). Some observations on the motion of the head of
a density current. J. Hydr. Res. 10, 305-324.
Willis, G. E., and Deardorf€, J. W. (1974). A laboratory model of the unstable planetary
boundary layer. J. Atmos. Sci. 31, 1297-1307.
Witten, A. J., and Thomas, J. H. (1976). Steady-wind driven currents in a large lake with
depth-dependent eddy viscosity. J. Phys. Oceanogr. 6, 85-92.
Wood, I. R. (1978). Selective withdrawal from two-layer fluid. 1. Hydr. Diu. ASCE 104,
1647- 1659.
Wright, J. C. (1961). The limnology of Canyon Ferry Reservoir. IV. The estimation of
primary production from physical limnological data. Limnol. Oceanogr. 6 , 330-337.
Wu, J. (1973). Wind-induced turbulent entrainment across a stable density interface. J.
Fluid Mech. 61, 275-287.
Wu, J. (1980). Wind-stress coefficients over sea surface near neutral conditions-A revisit. J.
Phys. Oceanogr. 10,727-740.
Wu,J. (1985). Parameterization of wind-stress coefficientsover water surfaces. J. Geophys.
Res. 90,9069-9072.
Wucknitz, J. (1976). Determination of turbulent fluxes of momentum and sensible heat
from fluctuation measurements and the structure of wind field over waves above the
tropical Atlantic during ATEX. “Meteor” Forschungsergab 11, 25-50.
Wuest, A. (1987). Ursprung und grosse von mischungsprozessen im hypolimnion natur-
licher seen. Rept. No. 8350, Eidgenossischen Technischen Hoschschule, Zurich.
Wuest, A., Imboden, D. M., and Schurter, M. (1988). Origin and size of hypolimnic mixing
in Urnersee, the southern basin of Vienvaldstattersee (Lake Lucerne). Schweiz. Z .
Hydrol. 50.
Wunderlich, W. D., and Elder, R. A. (1969). Selective withdrawal from density stratified
reservoirs. Rept. No. 21, Water Resources Res. Advance.
Yih, C. S. (1965). “Dynamics of non-homogeneous fluids.” Macmillan, New York.
Yih, C. S. (1980). “Stratified Flows.’’ Academic Press, New York.
Yoshimura, S. (1936). A contribution to the knowledge of deep water temperature of
Japanese lakes. Japanese J. Astro. Geophys. 13, 61-120.
Zeman, O., and Tennekes, H. (1977). Parameterization of the turbulent energy budget at
the top of the daytime atmospheric boundary layer. J. Amos. Sci. 34, 111-123.
This Page Intentionally Left Blank
 Author Index
Numbers in italics refer to the pages on which the complete references are listed.

A Baines, I! G., 398-399,468

Ablowitz, M. J., 4, 8, 76, 79,299, 301,302 Baladi, J. Y.,256, 271
Adams, R. L., 264,270,271,275 Balakrishnan, A. R., 235,271
Akiyarna, J., 407408,455 Banejee, I! E, 19, 79
Alavian, V,410,473-474 Banke, E. G., 331,333,472
Alavizadeh, N., 264,270,271,275 Bankvall, C. G., 243,271
Al-Homoud, A. A,, 385,455 Baransky, Y.,286-287,291,293,295,297,302
Ali, C. L., 239, 273-274 Barry, J. M., 248,273
Alkire, R. L., 231,280 Barsom, J. M., 132,150
Allanson, B. R., 430, 470 Bean, B. R., 333,370,455
Andersen, C. M.,30, 76 Bead, J. D.,442,456
Anderson, E. R.,340-341.469 Beavers, G. S., 233,271
Anderson, R.,240,272 Becker, R., 8445.92, 108, 118-119, 130, 134,
Anderson, R . h , 323-333,455,472 146, 147
Andersson, H., 91, 107, 147 Beckemann, C., 247,271-272
Andre, J. E., 341,455 Bejan, A., 228,231,238,240,242-243,
Aoki, S., 116, 147 246-248,251,272,276,278,385,390,
Aquise, E., 323,473 455,473
Arai, J., 321, 455 Bella, D. A., 321, 455
Aravas, N., 116, 147 Benjamin, T. B., 16, 76
Argon, A. S., 135, 139, 147 Bennett, J. R., 391,455
Arita, M., 406,408,455,465 Benney, D. J., 399,456
Armstmng, D. E., 413,457 Berg, C. A,, 94,147
Amaud, G., 227,275 Bergarnasco, L., 8-9, 11-12,74,80-81
Asaeda, T., 443,44546,455 Berger, D., 252-254.272
Asaro, R. J., 109,147 Bergquarn, J. B., 270-271,272
Ashby, M. E, 135, I39-140,148 Beman, A. S., 249,272
Ashton, I! J. 314,470 Best, C. H.,256,276
Assaf, G., 320,472 Bia, E, 249,274
Atkinson, J. E, 343,349,455 Billi, G., 399, 468
Ayers, D. L., 256,271 Bishop, J. E W , 84.94,147
Azad, E H.,270,271 Blake, S.. 296,403,434,456,464
Azad, R. S., 348,465 Blanc, T. V, 323,331,456
Aziz, K., 249, 276 Blanton, J. O., 321,353,456
Bless, S., 424-425,456
B Blythe, I! A., 243,246,272,279
Boardman, D.C., 341,474
Baddour, R. E., 344,457 Boashash, B., 416-417,463
Badgley, E, 323,468 Bohan, J. E, 402-403,456
Baer, E, 66, 76 Bona, J. L., 16, 76,285,295-298,302

477
 418 Author Index

Bones, S. A., 226,243,249-251,272-274 Cavaleri, L., 340,457

Borodulya, V A., 261,267,272 Cederwall, K., 442,458
Bowen, A. J., 407,472 Chan, B. K. C., 248,273
Boyce, E M., 311,314,417,456,464, 471 Chan, C. K., 261,267,270-271,273
Boyd, J. E, 10-11, 19-20,22-23,25-27,30-32, Chan, T.E , 81
34,36,38,41-42,44,50-52,54-55,57, Chandrasekhar, S., 268,270,273
59-61,63-65,67,69,71-72,74, 7679 Chang, H.-C., 26, 78
Brady, J. A ., 442,456 Chang, I. D., 238,240,273
Brame, V;, 364,368,465 Chang, E, 239,273
Brewster, M. Q., 262,264,268,272 Chao, S. Y., 406, 457
Briggs, W L., 46, 78 Chao, Y.-T., 256,277
Brim, W D., 442,456 Chamock, H., 327,329,331,457
Brimberg, J., 407, 471 Chen, B., 29,36, 78
Briscoe, M. G . , 304,461 Chen, C. K., 238,273
Britter, R. E., 408,411,456 Chen, J. C., 264,268-269,273,406,457
Bmard, D. N.,386,456 Cheng, I?, 226,228,231,235-236,238-241,
Brook, R. R., 325,457 249,273-275.277-278
Brooks, N. H., 459 Cherepanov,G.I?, 165,223
Browand, E K., 411,419,456,466 Cheung, E B., 442,457
Brown, D. K., 89, 111,147 Chou, R. L.,286,288-290,292-294.302
Brown, I. K., 442,456 Chow, L. C., 243,272
Brown, L. M., 91,93, 108,147 Christodoulou, G. G.,349,457
Brubaker, J. M., 335,341,456 Chu, C. C., 93,147
Bruggink, D. J., 438,465 Chu, C. K., 286-295,297,302
Bryant, I? J., 396,456 Chu, M. T.,39, 78
Buchak, E. M., 341,409,438,456,459 Chu, V; H., 344,406,457
Budiansky, B., 86, 138,147 Chuah, Y.K., 258,274
Bulson, E S.,441, 456 Churchill, S. W,264,268-269,273
Bunn, S. E., 320,456 Cialone, H.,109, 147
Buretta, R. J., 249,272 Clarke, A. J., 358, 461
Bums, E L., 442,457 Clarke, R. H., 325,457
Bums, I? J., 243,248-249,272,275 Claussen, M., 324, 457
Busch, N. E., 322,457 Cleaver, J. W., 238,273
Businger,J. A., 322-323,325,457 Cocks, A. C. E, 135, 139-140.147
Cohen, A., 399,457
C Colclough, M., 321, 458
Cole, G. A., 321, 463
Cacchione, D., 421,457 Collings, I. L., 394, 457
Caldwell, D.R., 304,457 Collins, W. D.,167,223
Caltagirone,J. E, 248,251,272-273 Colman, J. A., 413,457
Camassa, R., 286,288,300,302 Colony, R., 233,276
Canuto, C., 41,44, 78 Combarnous, M. A., 226,243,249,250-251,
Cardoni, J. J., 353,472 272,274
Carey, V; P,258,274 Cooley, E, 441,457
Carmack, E., 459 Cooper, C. M., 353,472
Carmack, E. C., 305,314,457 COEOS,G. M., 343-344,348,350-351,419,
Carslaw, H.S., 367,457 464,471
Carson, D. J., 326,457 Cormack, D.E., 366,385,457-458
Cartigny,J. D., 262,281 Cote, 0. R. 349,465
Cation, I., 249,259,273, 275-276.280 Cottrell, A. H., 130,147
 Author Index 479

Cox, T.B.,110,147 Duddelar, J. D., 385, 471

Craig, H., 413,464 Durbin, E A., 341,471
Craya, A., 393,396,458 Dybbs, A., 255,259,274,279.281
Csanady, G. T., 304,354,358,458 Dyer, A. J., 323,325,452 459
Cunnington, G. R.,227,279 Dyson, B. E, 135, 139,142 148
Cuong, F! G., 28, 79
E
D
E, X.,346-347,459
Dafalias, Y.E, 123,147 Echigo, K.,270,274
Dake, J. M. K., 422,458 Eckert, E. G., 256,274
Dalzell, W H., 264,275 Eckert, E. R. G., 253-255,274
Daniels, I? G., 246,272 Edinger, J. E., 341,409,438,456,459
Darbyshire, J., 321,458 Edward, D. H. D., 320,456
Darden, R.B.,410,458 Eilbeck, J. C., 78
Dashen, R. E,61, 78 Elder, J. W , 249-250,274,366,459
Date, E., 11, 78 Elder, R. A., 392,405,475
Davis, J. M., 441-442,456,458 Elliott, G. H., 390,459
Davis, R.,323,458 Elliott, H. A., 154, 157, 188,223
Dayan, J., 257, 275 Elliott, J. A., 390,459
Dean, J. E, 341,474 Ellis, C. R., 409, 465
Deardorff, J. W , 328,348,349,458,475 Ellison, T.H., 407, 459
Debler, W R.,400401,458 El-Masri, M. A,, 260,277
Deem, G. S., 29, 78 Embury, J. D., 91,108,147
Denton, R.A., 343,458 Emerson, S., 413,464
deszoeke, R.A,, 337,346,377-378,458 Emmanuel, C. B.,332-333,370,455,459
DeVries, D. A., 253,274,278 Epstein, M., 442,457
De Wasch, A. E,234,274 Ernst, H., 117,149
Dhir, V K., 259,280 Ettefagh, J., 252,271,274,280
Dickey, T. D., 353,370,458 Evans, G. H., 238-239,274
Dillon, T.M., 335,341,458 Eydeland, A., 28,4447, 78
Dinulescu, H. A., 256,274
Ditmars, J. D., 442,458
Djordjevic, V D., 399,458 F
Dobson, E W , 323,458 Fabrikant, V I., 160, 162, 168, 172-173, 181,
Dodd, R.K., 78 199,201,206,209,223
Donelan, M. A., 323-324,331-333,341,430, Facas, G. N., 248,274
458459,466,468 Faeth, G. M., 442,473
Doraivelu, S. M., 92, 147 Faghri, M., 253-255,274
Dortch, M. S., 444,446,459 Fand, R. M., 241,274
Douglas, E. W , 442,456 Fandry, C., 378,401402,410,463-464
Dracos, T. H., 409410,461 Fanner, D. M., 314,370,459
Drazin, F! G., 343,459 Farouk, B.,242,248,274
Drinklow, R. L., 332,462 Farrant, B.,472
Dritschel, D. C., 29, 78 Farrell, G. J., 408409,459,465
Drolen, B.L., 262,264,274, 279 Faruque, M. A., 247,280
Drummond, J. E., 385,459 Faust, C. R., 252,274
Dubensky, E. M., 145,149 Faust, K. M., 411,459
Dubreil-Jacotin, M. L., 400,459 Ferguson, W E., 4, 11, 78
Dubmin, B. A,, 11, 78 Fernandez, R.T., 242,275,279
480 Author Index

Fernando, H. J. S., 350-351,461 Gibson, C. H., 416, 46U-461

Feshbach, H., 21,23,80 Gilbert, D., 417,421,460
Filatova, T.N., 321, 465 Gill, A. E., 358,363,461
Findikakis, A. N., 341,459 Gilmer, R. O., 333,370,455
Finkel, A., 17, 80 Godden, D. A., 469
Finlayson, B. A., 235,277 Goldman, C. R., 413,464,469
Fischer, H. B., 304,306,314,366,379,393. Golikova, S. S., 223
402,405-410,412,423,458,459,463 Goods, S. H., 93,148
Fischer, K. H., 413,474 Gordon, J. A., 438,461
Fissel, D. B., 333,470 Goshayeshi, A., 264,270,271,275
Fitch, J. S., 258,280 Grace, J. L., 402-403,456
Flamant, G., 227,275 Gradshteyn, I. S., 154,200,223
Flaschka, H., 4, 11, 78 Graf, W H., 329,332,461
Flatt, I., 258,275 Graham, D. S., 444,461
Flierl, G. R.,26,45,54,71-72, 78-79 Grammaticos, B., 52,56,69-70,73,80
Fokas, T.,299, 301,302 Grauze, G., 332,462
Forbes, L. K., 396,459 Greatbatch, R.J., 61, 78
Ford, D. E., 353,405,424,459-460,472 Green, A. W , 424
Forest, M. G., 4.7, 78 Green, A. E., 86,148
Francis, E. A., 341, 474 Greene, J. M., 283,300,302
Francois, D., 92,148 Gregg, M. C., 304-305,335,337,341,416417,
Franke, R.,341,409,460 461, 471
Friehe, C. A,, 333,460 Grewal, N. S., 227,270,279
Fritts, D. C., 304,460 Guennouni, T.,92, 148
Froment, G.E, 234-235,274,277 Gunaselsera,J. S., 92, 147
Frost, H. J., 140,148 Gupta, V F!, 249,275
Fukusako, S., 243,279,385,464 Gurson, A. L., 84,89-90,92-94, 102, 121,148
Guyomar, D., 411,419,456
G
Gabor, J. D., 227,270,279 H
Galershtein, D. M., 227,270,279 Haajizadeh, M., 243,246,275,279
Gardner, C. S., 283,300,302 Haber, S., 257,275
Gardner, W , 252,275 Hadamard, J., 96,148
Gargett, A. E., 413,460 Hahn, G. T., 109, 130,148
Gariel, F!, 394,460 Hall, A. J., 341, 473
Ganatt, J. R.,323-324,331,460 Hamblin, I! E,304,313-314,316,340,354.
Garrett, C., 304,417,421,460 390-391,413,415,419,426,430431,
Gartling, D. K., 246,248,275 433,463-464,462 470
Garvine, R. W , 406,469 Hammack, J. L., 17, 74, 78,285,302
Ganvood, R. W , 305,335,340-341,348, Hancock, J. W , 89,92, 109, 111,148
460,468 Hannoun, I. A,, 350-351,461
Gat, J. R.,321, 472 Hansen, C. G., 335,341,458
Gebhart, B., 238,276 Hansen, E ,460
Geer, J. E, 30, 76 Hansen, I!J., 299-300,302
Geernaert, G.L., 322,327,331-333,460 Harashima, A., 461
Gegel, H. L., 92, I47 Hamshima, I. A., 381,386
Gelhar, L. W., 402,460 Harleman, D. R.E, 315,343,386,396,
Georgiadis, J., 249, 275 422-425,455456,458,461,463,467
Gibbon, J. D., 78 Harmathy, T.Z., 256,275
 Author Index 481

Hams, S. L., 441, 457 Huenefeld, J. A,, 238,278

Hart, J. E., 385,461 Huenefeld, J. S., 239,276
Hartnett, J. I?, 233,278 Hull, D., 135, 148
Hasegawa, S., 270,274 Hung, C. I., 238,273
Hasse, L., 323,458 Hunt, M. L., 235,271,276
Hasslacher, B., 61, 78 Hunter, J. K., 69,73, 79
Hauenstein, W,409410,461 Huppert, H. E., 411,462
Haugen, D. A., 349,465 Hurley-Octavio, K. A., 315,424, 461
Haupt, S. E., 7, 13-16,30, 77-79 Hussain, N.A., 4 4 2 4 3 , 4 6 3
Havstad, M. A,, 248,275 Hussaini, M. Y., 41,44, 78
Hayasaka, H., 261,267,276 Hutchinson, G. E., 309, 314,321,463
He, M. Y., 136-137.148 Hutchinson, J. W ,88, 101, 103-106, 120-121,
Heaps, N.S., 354,461 123-124, 127, 130, 135-138,147-149
Heathershaw, A. IL, 343,462 Hutter, K., 304-305,463,472
Hebbert, R.,314,392,405,407408,424,430, Hyman, J. M., 4,7-8, 79
462,464
Hebbert, R. H. B., 463
I
Helland, K. N.,417, 464, 471
Henderson, IL, 305,468 Idso, S. B., 315,321,463
Henderson-Sellers, B., 309,314,321,423-424, Imberger, J., 304-305,309,311,313-314,
430431,441,462 316-317,321,329,333,335,337-338,
Hickox, C. E., 242,246,248,275 340-341,343-351,353-354,358,362,
Hicks, B. B., 323-325,329,331-333,462 364-366,368,370-371,377-381,
Higgins, J. M., 438, 465 385-388,390,392-393.396403,
Hill, R.,84,94,96,98,147-I48 405-413,415417,419,424426,
Hirata, T., 421,469 430434,437,443,44547,449,455,
Hirota, R.,10-11, 79 457-459,462-464,46&473
Hocking, G. C., 394,396,430,434,438, Imboden, D.M., 304,314,322-323,413,464,
459,462 462 475
Holland, J. I?, 444,446,459 Inaba, H., 243,279,385,464
Holloway, G., 304,413,460,462 Ingersoll, A. I?, 28, 79
Holloway, I? E. 370,399,472 Inoue, K., 342,464
Holton, J. R.,60,79 Ishizaka, K., 249,277
Holyer, J. Y., 411, 462 Ito, M., 11, 79
Hornsy, G. M., 246,251,281 Its, A. R.,10, 79
Hong, J. T,238-239.275 Itsweire, E. C., 412,417,464,471
Hopfinger, E. J., 304-305.346-347,351,377, Ivetic, M., 342, 464
411,416,459,462 hey, C. M., 248,273
Horn, W ,354,462 Ivey, G. N., 314,346,390-391,396,403,417,
Home, R. N., 250,275 419,421,430,434,456,464,470
Horton, W ,71, 79 Iwasa, Y.,349, 464
Hottel, H. C., 264,275 Izumi, Y., 349,465
Howard, C. S., 405,462
Howell, J. R., 261,265-267,270,279,281
Hsu, C. T.,235,238,273,275
J
Hsu, S. A., 322,462 Jaeger, J. C., 367,457
Huang, C. L. D., 256,275-276 Jain, S.C., 386,408,464
Hubbani, D. W., 390,472 Jakobsson, J. O., 259,273
Huber, D. G . , 401, 463 Jakobsson, J., 259,280
Huber, W. C., 423,463 Janicka, J., 341, 471
 482 Author Index

Janowitz, G. S., 346,464 Kindle, J. C., 61, 79

Jassby, A., 413,465 Kirk, J. T. 0.. 387,465
Jaworski, M., 4, 11, 79,82 Kishimoto, K., 116, 147
Jirka, G. H., 315,396,406,408,424,455,SZ Kitaigorodskii, S. A., 331,341,350,466,469
463,465,471 Klarsfeld, S.,243,276
Johnson, M. A., 116,150 Klein, D.E., 261,267,281
Johnson, M. C., 405,459,460 Klikoff, W A., 258,275
Johnson, R. S., 9, 79 Knott, J. E, 132,150
Johnson, T.R., 409,465 Kobus, H.E., 442,466
Jokela, J. B., 438, 465 Koh, J. C. Y.,233,276
Joller, T., 304, 314,413,464 Koh, R. C. Y.,392,398,400,402-403,
Jonsson, T., 249,276 459,466
JOT, A. G., 442,456 Kondo, J., 331,466
Joseph, D. D., 233,249,271,275 Koplik, J., 107-108, 148
Joshi, Y.,238,276 Korpel, A., 19, 79
Korpela, S. A., 385, 459
Koss, D.A., 145,149
K
Kovensky, V: I., 261,267,272
Kachalovskaya, N.E., 223 Kranenburg, C., 346,351,364-365’368, 378,
Kadomtsev, B. B., 16, 79 444-445,466
Kaimal, J. C., 465 K ~ ~ uE.s ,B., 340,343,345-348,424,466,469
Kalthoff, O., 235,276 Krischer, O., 253,276
Kamail, J. C., 349, Krishnamurthy, V:, 66-69, 79
Kaneko, T., 249,276 Kmskal, D. M., 283,293,300,302
Kantha, L. H., 348,465 Kmskal, M. Jl,9,52,56,62-64,69, 71,
Kanyama, K., 385,464 73,804
Kao, T. R, 370,398,401,404-405,465,469 Kubota, T.,81
Kassir, M. K., 154, 184-185, 195, 197,200,223 Kuchida, M., 469
Katavola, D. S., 396, 465 Kudo, K., 261,267,276
Kato, H.,348, 349,465 Kulacki, E A., 246-248,250,277-278
Katsaros, K. B., 327, 331, 460 Kullengberg, G., 413,466
Katsuhara, T., 249,277 Kumagai, M., 391,469
Kano, Y.,249,276 Kumar, S., 262,264,269-270,274,277
Kaup, D.J., 299-301,302 Kuscynski, K., 97,149
Kaviany, M.,230,234,238-239,275-276 Kutznetsov, 0. A,, 331,466
Keer, L. M., 127,149
Keijman, J. Q., 322,465
Keller, H. B., 36, 79 L
Keller, J. B., 267,276, 394, 474 Lake Biwa Research Institute, 413,466
Keller, W C., 328,465 Landau, L., 74
Kerker, M., 265,276 Langmuir, I., 314,466
Kesavan, K., 242,279 Lapwood, E. R., 249,277
Keulegan, G. G., 364,368,463 Large, R G . , 323,331,333,466
Keyhani, M., 248,278 Larsen, S. E., 460
Khalatnikov,J. M., 56, 80 Larson, J., 8-9, 11,81
Khomskis, R R., 321,465 Lasheras, J. C., 466
Kielmann, J., 354,465 Launat, G., 246,277
Kim, B. R., 438,465 Lawrence, G. A., 343,350,396-397,402-403,
Kim, S. J., 238, 276 434,466
Kimura, S., 240,276 Lax, I?, 8, 79
 Author Index 483

Leal, L. G., 366,385,457-458 Manins, F! C., 378,410,412,467

Leavitt, E., 323, 468 Marciniak, K., 97, 149
Lecarrere, E, 341,455 Markofsky, M., 423,467
Leckie, E A., 135, 147 Marmoush, Y. R., 390,467
Lee, E. H.,123,148 Marra, J., 341, 474
Lehn, H., 413,474 Marti, V D.E., 322-323,467
Leibovich, S., 341, 466-467 Martin, J., 343,462
Lemmin, U., 304,314,413,461 Mascolo, D. M., 402, 460
Lenschow, D. H., 348-349,467 Masuoka, T., 249,276-277
Leonov, M. Y.,201,223 Matsuno, Y., 10-11,74, 79-80
Lerman, A., 321,467 Matthews, F! R., 411,470
Lerou, J. J., 235,277 Matveev, V B., 10, 78-79
Leschziner, M. A,, 341,409,460 Maxwotthy, T., 285,302,370,377-380,405,
Lewis, E. L., 413,467 410,467-468
Lewis, W K., 252,277 McBean, G. A., 323-333,370,467-468
Lewis, W M., 321,467 McClintock, E A., 84, 89,149
Li, C., 235,277 McDougall, T.J., 442,444445,468
Li, E Z., 144,148 McEwan, A. D., 398-399,468
Liebowitz, H., 187, 196,223 McGavin, R. E., 333,370,455
Lighthill, M. J., 358,467 McLaughlin, D. W ,4,7, 11, 78
Lin, J. T., 411, 467 McMeeIung, R. M.,116, 118,142 149
Lindemann, G. J., 72, 79 McWilliarns, J. C., 24,68,71-72, 79
Linden, F! E , 346-347,379-380,467 Mear, M. E., 120-121, 123, 130,149
List, E. J., 350-351, 459, 461 Meiron, D. E., 31, 79
Lo, A. K., 323,467 Meiss, J. D., 71, 79
LO,K. K., 114-115,150 Melack, J. M., 321, 467
Loffler, H., 321,467 Mellor, G. L., 341,468
Loh, I., 314,405,407408,424,430,462,464 Melville, W.K., 333, 468
Long, R. R., 350,404,467 469 Menguc, M. F!, 269-270,277
L~IEIIZ, E. N., 24.66-69, 79 Mercer, J. W ,252,274
Low, J. R.,110, 147 Merkin, I. H.,241,277
Lueck, R. G., 417,467 Meni, N., 329,332,461
Luikov, A. V,253,256,277 Meyer, R. E., 56, 79
Luketina, D. A., 390,406,408409,467 Michel, R. L., 413,464
Lumley, J. A., 341,466 Michioku, K., 341,351,397,421, 469
Lumley, J. L., 324,341,349,471,473 Mied, R. F,! 72, 79
Lyczkowski, R. W ,256,277 Milgram, J. H., 442,444,468
Miloh, T., 19, 80
M Minkowycz, W J., 236,241,273,277
Mittal, M.,239,276
Ma, Y-C., 4, 78-79 Miura, R. M., 283,300,302
MacIntyre, S., 321,467 Miyake, M., 323,468
MacKenzie, A. C., 89, 109, 111, 148 Modest, M. E, 270,271,279
Magnusen, I? E., 145,149 Mohtadi, M. E, 249,276
Mahoney, J. J., 16, 76 Monin, A. S., 325,468
Mahony, J. J., 399,402,467 Monismith, S. G., 311,335,340,346,353,
Mahrt, L., 348-349,467 358-360,363-365,368,370,381,385,
Majumdar, A., 269-270,277 388,396,399,405,460,463,468
Malas, J. C., 92,147 Monkmeyer, I? L., 402,474
Mallett, R. L., 123,148 Mooers, C. N. K., 343,470
484 Author Index

Morgan, A. I?, 27,80 Ogniewicz, Y.,252, 257,277

Morgan, J. T.,92,147 Olmstead, B. A , , 252,256,278
Morgan, R. L., 396,461 Ookubo, K., 391,406,469
Moms, H.C., 78 Oonishi, Y., 391, 469
Morse, I? M., 21,23,80 Orangi, S., 246-247,279-280
Mortimer, C. H., 304,354,390,462,468 Orlob, G. T,422-423,430,469
Mossakovskii, V I., 201,223 Orszag, S. A , , 40-41,43-44,80
Motakef, S., 260,277 Osborn, T. R., 417,467
Moussy, I?, 109.149 Osbome, A. R., 8-9, 11-12,74,80-81
Moya, S. L., 251,277 Overman, E. A., 29,81
Mulheam, I? J., 324,468 Oyane, M., 84,91,96,150
Muller. I?, 305, 335, 340, 341,468 Ozisik, M. N., 265-266,270,272 281
M u d , W H., 24,80
M u d , W.,304,340-341,460,469
Munnich, K. O., 413, 474 P
Muralidhar, K.,248,277 Pan, J., 100, 102-103, 111, 115, 128,149-150
Muramoto, Y., 391,406,469 Panasiuk, V V, 198,223
Murota, A., 341,351,391,421,469 Panin, G. N., 331,466
Murthy, C. R., 413,466 Panofsky, H.A., 328,469
Myer, G. E., 341,471 Pao, H.I?, 370,398,404,465,469
Myrup, L. O., 307,323,469 Pao, H.S., 404, 471
Pao, Y.H., 411,467
Paolucci, S., 341, 467
N
Paris, I? C., 117,149
Nakmura, A., 10-11,74,80 Park, C., 370,465
Nakano, S., 270,274 Parker, G . J., 333,371,377,380,464 469-470
Narang, B.S., 44243,463 Patalas, K.,321,470
Narimousa, A., 350,469 Paterson, R. D, 333,370,468
Narimousa, S., 350,469 Patterson, J. C., 313-314,346,386-387,390,
Needleman, A., 84-85,91-94,97, 100, 405,407-408,424-426,430-431,433,
102-103, 106-108, 111-112, 114-119, 437-438,445,447,449,462,
128-134, 138-140, 144, 146,147-151 464-465,470
Nernat-Nasser, S., 102, 127,149 Paulson, C. A., 323,326,470
Neveu, A , , 61, 78 Paulson, C., 468
Newell, A. C., 301,302 Pearson, M. D , 335,341,458
Newman, A. B., 252,277 Pederson, T., 413,471
Newrnan, E C., 321,469 Pedlosky, J., 60,80
Niiler, P P, 343,345-348,469 Pei, D.C. T., 235,252-254,271-2Z?
Nilson, R. H.,257,277 Peirce, D, 129,149
Nishida, S., 343,469 Peirson, N! 1,333,458-459
Nokes, R. I., 351,421,464,469 Peitgen, H.-O., 39,80
Norris, D.M., 133,149 Peregrine, D. H., 16,24,80-81
Novikov, S. E, 78 Perkin, R. G., 413,467
Nutt, S. R.,107,149 Pemnjaquet, C., 329,332,461
Perry, K.,413,471
Petch, N. J., 130,149
0
Peterson, E. W,324,470
O’Donnell, J., 406, 469 Petviashvili, V I., 16,44,47-48,72, 79-80
O’Sullivan, M. J., 249-250,275,278 Pfender, E., 253-255,274
Obukhov, A. M., 325,468 Philip, J. R., 246,253,277-278
 Author Index 485

Phillips, 0. M., 348-349,419,465,470 Richman, J. G., 335, 341,458

Pierini, S., 17,80 Richmond, O., 84-85.92, 108, 134, 146,147
Pierrehumbert, R. T., 29,80 Richter, K., 331,460
Pinkel, R., 341, 472 Richter, I!H., 39,80
Plant, W. J., 328,465 Riedel, H.,135,150
Plumb, 0. A., 238-239,252,256,274,276,278 Rimmer, D. E., 135,148
Pokrovskii, V L., 56,80 Ritchie, R. O., 132, 150
Pollard, R. T., 340-341,343,350,470 Robarts, R.D., 314,413,470
Pomeau, Y.,52,56,69-70,73,80 Roberts, I!J. W., 411,470
Pond, S , 323,331,333,466,470 Robertson, D G., 311,314, 456, 471
Pop, I., 236,240,273,276,278 Robillard, L., 247,251,280
Poulikakos, D., 234,238,246-247,249,251, Robinson, J. L., 249,278
272,278 Rodgers, G. K., 390,470,471
Powell, T.M., 307,322-323,328-329,340, Mi,W , 341,409,438,460,471
346,413,469,472 Rohani, A. R., 258,279
Powell, T,465 Rohr, J. J., 412, 471
Prasad, V , 246-248,250.277-278 Rohsenow, W. M., 233,278
Price, J. E , 341,343,470,474 Rolfe, S. T., 132,150
Pritchard, W G., 285,295-296,298,302,399, Romero, L. A., 257,277
402,467 Rosa, E , 311, 314,471
Psioda, J. A., 110,150 Rosenfield, A. R., 109,148
Pujalet, R., 305,468 Rossie, A. N., 385, 455
Purple, R. A,, 396,461 Rudnicki, J. W , 96,150
Puttick, D. E., 93,149 Russell, J. S., 283, 285,302
Ryan, B. E , 324,460
Ryan, E J., 423,463
Q Ryzhik, I. M., 154,200,223
Quartentni, A., 41.44, 78
S
R
Sadhuram, Y., 322,471
Radhakrishnan, K., 341,467 Saffman, E G., 29,31,36, 78-79
Ramadhyani, S., 247,271-272 Sahlberg, J., 425,471
Ramani, A., 52,56,69-70,73,80 Saje, M., 100, 102-103, 111, 115, 128,149-150
Ramos, E., 251,277 Sakata, M., 116,147
Ramsbottom, A. E., 354,461 Salmun, H., 419,470
Raphael, J. M., 392,470 Sanderson, B., 413,471
Rayner, K. N., 328-329,333,337,340,343. Sanford, T. B., 417,461
345-349,353,425,431,470,472 Sargent, R. E., 406,471
Redekopp, L. G., 399,458 Sarkaz S.. 257,280
Rehm, R. G., 321,423,473 Sarotim, A, E 264,275
Reid, D. G., 325,457 Sastry, J. S., 322,471
Reid, W. H., 343,459 Sathe, S. B., 247,280
Renken, K., 234,278 Satish, M. G., 251, 280
Rhines, I! B., 337,340,343,350,378,458,470 Savage, S. B., 407, 471
Rice, J. R., 84,89,93-94,9698, 102-103, 111, Saxena, S. C., 227,270,2?9
116,120, 124, 132,135, 138-140, Saylor, J. H., 311, 314,471
149-150 Schertzer, W. M., 311, 314,471
Richards, L. A., 252,278 Scheurle, J., 69, 73, 79
Richards, E J. R., 326,457 Schiebe, E R., 353,472
486 Author Index

Schmitt, K. E, 333,460 Smith, A. A., 390,467

Schneider, K. J., 249,279 Smith, E., 130,150
Schoenhals, R. J., 256,271 Smith, J., 472
Schotte, W , 269,279 Smith, S. D., 322-323,331,333,455,472
Schrock, V E., 242,275,279 Smyth, N. E , 370,399,472
Schubert, H. G., 442,473 Sneddon, I. N., 154, 186-187, 190,223
Schurter, M.,304,314,413,464,475 Somerton, C. W , 249,279
Schuster, J., 235,281 Sorbjan, Z., 324,327,4Z?
Schwab, D. J., 354,462 Sozen, M.,253,279
Schwa-, L. W., 30-31,80 Spain, J. D., 390, 472
Schweitzer,S., 255,274 Spalding, D.B., 425, 472
Schwerdtfeger,K., 442,473 Speich, G. R.,115, 150
Scott, C. E, 379,385,471 Spigel, R. H., 315, 317, 337,340,343-349,
Scott,J. T.,341,471 353-354,358,364,370,403,425426,
Scott, L. R., 27,81,285,295-296,298,302 430431,434,464,472
Seban, R. A., 270-271,272 Spitzig, W. A , , 115,150
Segur, H., 8, 17,52,56,62-64,69,71,73, 76, Spolek, G. A., 252,256,278
SO,285,301,302 Stacey, M. W., 407,472
Seki, N., 243,279,385,464 Stadnik, M. M.,223
Selna, L. G., 422-423,430,469 Stefan, H.G., 353,407-409,424,455,
Sen, A. K., 246,279 459,465,472
Sen, M., 251,277 Steinberger, T.E., 241,274
Sephton, L. M., 314,470 Steinhorn, I., 320-321.472
Seydel, R.,36-37,81 Stenger, E , 4142,81
Sham,T.-L., 138,150 Stem, M. E., 71, 78
Shavit, A., 257, 275 Stevens, R. L., 233,276
Shay, T. J., 317, 335, 341, 471 Stewart, R., 323,329,341,471
Shayer, H.,242,274 Stewart, R.W., 472
Sherman, B. S., 462 Stiassnie, M.,24,81
Sherman, E S., 343-344,348,350-351,471 Stiller, M., 321, 466
Shewood, T. K.,252,279 Stocker, T., 304,305,472
Shih, C. E, 129, 144,148-149 Stone, G. E, 385,458
Shih, H. H., 404,47l Stone, R. H., 110, I50
Shih, T. H., 341,404,469,471 Straskraba, M., 309,311-315,321,413,472
Shima, S., 84,91,96, 150 Straws, J. M., 249,279
Shiralkar, G. S., 243,279 Street, R. L., 341,459
Shyu, J. H., 419,470 Streett, C. L., 41,48,81
Siang, H. H., 256,276 Strub, P. T,311,322-323,328-340,
Siegel, R., 265-266,270,279 346,472
Sih, G., 154, 184-185, 187, 195-197,200,223 Stull, R., 349,473
Silovaniuk, V E, 223 Sturm, T.W , 386,473
Silvester, R., 400,402,471 Subramanian, E., 243,247,280
Simons, T. J., 354,465 Sumi, Y.,127,149
Simpkins, D. G., 243,279 Sun, T. Y.,442, 473
Simpkins, E G., 246,272,385,471 Sundaram, T. R., 321,423,473
Simpson, J. E., 379-380,406,408,411,456. Suresh, S., 118-119,147
462 471 Suryanarayana, A., 322,471
Simpson, J. J., 353,370,458 Svensson, U., 425,472-473
Singh, B. S., 259,279 Sverdrup, H. U., 315,473
Slutsky, S., 138,147 Swamy, G. N., 322,471
 Author Index 487

Swenson, M., 72,81 lhng, K.-K., 81

Swinbank, W C., 431,473 'hrkington, B. A,, 28,44-47, 78
'hrner, J. S., 321,340,343,348,407,416,424,
459,466,474
T
Tvergaard, V , 84-85,90-92,95,98, 100-121,
Tabanfar, S., 270,279 123-128, 130-144, 146,147-151
Tacke, K. H., 442,473 Tzur, Y.,321, 474
Tada, H., 117,149
Takeya, A., 116, 147
Tamai, N., 406,455 U
Tanaka, S., 11, 78 Udell, K. S., 252,258,280
Taniguchi, H., 261,267,276 Uflyand, Y.S., 154, 199,223
Tanveer, S., 29,81 UNESCO, 315,474
Taussig, H. J., 314,470 U.S. Army Coastal Eng. Research Ctr., 429,474
Taya. M., 102,149 U.S. Army Corps of Engineers, 474
Taylor, G. I., 379,473
Taylor, M., 323,473
Taylor, I? A., 324,473 V
Tennekes, H., 324,343,348-349,473,475 Vafai, K., 229-231,233,235,238,252-253,
lhnessee Valley Authority, 306,321,430,473 257,271,274,276,279-280
'knay, E. A,, 341,466 van Ana, C. W., 417,464, 471
Thiyagaraja, R., 231,233,280 Vanden-Broeck, J. M., 29,81,394,474
Thomas, J. E,92,147 Van Dyke, M., 30,81
Thomas, J. H., 423,473,475 Van Leer, J. C., 470
Thompson, R. 0. R. Y.,354,378,401-402, Vasalos, I. A., 264,275
410,426,464,470,473 Vasseur, I?, 247,251,280
Thornton, J. A., 314, 470 Vasudevan, A. K., 118-119,147
T h o p , S. A., 304-305,341,343-344,350, Venkatram, A,, 324,474
370,416,421,473 Ventz, D., 321, 474
Tien, C. L., 227,229,230-231,235,238-239, Verkley, W. T. M., 28,46,48,81
243,246,248-249,252,257-258, Verma, A. K., 239,274
261-262,264,267-271,2?2-272 Veronis, G., 404, 474
279-281,385,455 Vethamony, I?, 322,471
Toda, M., 19.81 Viskanta, R., 247,269-271,271-272,272 280
'Iblland, H. G., 441,443,473 Vortmeyer, D., 235,261,267,269,271,273,
'Ibly, J. A,, 351, 377,462 2 76.280-281
'Ibng, T.W., 243,246-247,279-280
Tracey, D.M., 84,89, 116,150
%CY, E. R., 8-9, 11, 81
W
Trevisan, 0. V ,390,473 Walesh, S. G., 402, 474
Tribbia, J. J., 66, 76 Walker, K. L., 246,251,281
Troup, A. J., 325,457 Wallace, B. C., 421, 474
'Rustrum, K., 401, 473 Walther, E. G., 341,471
Tsai, E I?? 259,280 Wang, K. Y.,268,281
Tsihrintzis, V A., 410,473,474 Ward, I? R. B., 321,413,470,474
Tsuchiya, Y., 24,81 Watanabe, M., 381,386,461
Tsvelodub, 0. Y.,44,80 Water Resources Engineers Inc.,
Tuck, E. O., 394,474 425-426.474
Tucker, W A., 424 Watson, L. T., 27.81
'hlin, M. I?, 19, 80 Watts, H. A,, 242, 275
488 Author Index

Weber, D. J., 442, 473 Wuest, A., 413414,475

Weber, J. E., 243,251,281 Wunderlich, W.D., 392,405,475
Wedderburn, E. M.,311,353,474 Wunsch, C., 421,457
Wei, S. N., 398,465 Wyngaard, J. C., 349,465
Weinstock, J., 415,474
Weiss, R. G., 464
X
Weiss, R. R., 413
Weiss, W., 413,474 Xiang, L. W., 286-287,291,293,295,
Weissman, D. E.,328,465 297,302
Weller, R. A., 341, 472.474
Welty, J. R., 264,270,271,275
Y
Wernert, G. M.,390,472
Wertheimer, T. B., 123,148 Yamada, T., 341,468
Westerberg, H.,413,466 Yamada, Y., 262,281
Westman, A. E. R., 252,281 Yamamoto, H., 98, 100, I50
Westmann, R. A., 154,194,223 Yang, W. J., 261,267,276
Wetzel, R. G., 309,320,474 Yang, Y. S., 261, 267,281
Whitaker, S., 252-253,256-257,280-281 Yasuda, T., 24,8I
White, S. bl.,258,281 Yen, Y.C., 249,281
Whitehead, Jr., J. A., 71, 78 Yener, Y., 270,281
Whitham, G. B., 19,23,81 Yih, C. S., 394,401,404,475
Widtsoe, J. A., 252,275 Yoon, S. C., 411,419,456
Wiennga, J., 331, 474 Yoshida, S., 343,469
Wilkinson, D. L.,407408,421,44142, Yoshimura, S . , 321,475
474-475 Yuen, H. C., 31, 79
Williams, G. I?, 51, 61, 81 Yuen, W W , 271,281
Willis, G. E., 348-349, 475
Wilson, R. J., 51, 61, 81
Z
Wimp, J., 23,81
Winther, R., 295,298,302 Zabmdsky, S. S., 227,270,279
Witten, A. J., 423,475 Zabusky, N. J., 9,29, 78,81,283,293,302
Wong, K. E,259,281 Zagrcdzinski, I., 4, 11, 79,81-82
Wong, L. W ,271,281 Zahoor, A., 117,149
Wood, I. R., 343,396,407,456, 458,475 Zaitsev, A. A., 19, 82
Wooding, R. A., 242,281 Zang, T. A,, 41,44,48, 78,81
Wooley, D. A.. 442,456 Zaslavskii, M.M., 331,466
Wright, J. C., 321,475 Zecchetto, S., 340,457
Wu, H. M.,29,81 Zeman, O., 343,348,475
WU,J., 323,331-332,349,365,475 Zema, W ,86, I48
Wu, T.Y., 286,288,300,302 Zheng, T. M., 231,273
Wucknitz, J., 323,475 Ziegler, H., 122,150
 Subject Index

A linear stratification, 443

Abel operator, 207,209 relation to Lake number, 446-447
Absorbed energy, 133 two-layer stratification, 444
Absorption coefficient, 261, 265 Building thermal insulation, 226,243,247
Aerator Buoyancy frequency, 398
diffuser, 442 Burst of nucleation, 102
number of ports, 449
Axisymmetric, pressure, 154 C
Axisymmetnc, problem, 167 Calibration, model, 423,425
Catalytic reactors, 226,230,236,240
B Cavitation, 134
Cell models, 102
Bessel function, integral of, 154, 183, 186,200 Center of volume, 315
Boundary Channeling effect, 230,235
conditions, 155, 158, 167, 170 Chemical reaction engineering,230
value problems, 154 Chemical reactorfs), 227,234-235
Boundary effect(s), 226,229-231,234, Circular crack, 160, 165, 182-187, 192-197
238-239,247,249 Circular disk, 154
Boundary mixing, 421 Circular punch, 168, 187, 191, 198,201
Breaking waves, 421 Cleavage fracture, 131
Brinkman, 230,234-235.238, Cnoidal wave, relationshipwith solitary wave, 17
243,247,249-250 Coal combustors, 226-227
Brittle-ductile transition, 133 Collisionless shocks, 291, 295
Bubble plume Complex
bubbler system, 441 conjugate, 157, 170,219
data, field, 441 constant, 158, 172, 192
design history, 441 function, 170
efficiency, 441,443,44647 stress, 157
dependence on number of plumes, 449 stress intensity factor, 181, 194
peak, 447 tangential displacements, 156
relationshipto r: 447,450 tilting angle, 191
Lake number for, 446 Composite system, 250
mixing due to, 441 Concentrated load
model, 442-444 inside acrack, 164-165, 177-182
coupling with fan field, 445 outside a crack, 165-167
double plume, 442-443 outside a punch, 198
simulation results, 449-451 Conjugate boundary layer, 240
single plume, 442 Contact problems
plume parameter P for a circular punch, 168-169, 187-192
general stratification, 447 for a smooth punch, 167
489
490 Subject Index

Continuous/pseudo-continuousmodel, 266 Diffusivity, eddy, 423-424

Crackgrowth, 117-118, 142 Direct contact heat exchangers,226
Crack problems Discontinuous/discretemodel, 266
asymptotic behavior of stresses and Dislocation creep, 135- 136
displacements near crack rim, 184-185, Displacement
187, 194-196 complex tangential, 156
circular crack under shear loading, 192- 197 normal, 157
concentrated load outside a circular outside a penny-shaped crack,
Crack, 165-167 182, 186, 193, 195
flat crack under arbitrary normal under a circular punch, 188, 190, 192
loading, 158-164 Distillation towers, 234
general crack under uniform shear, 172-173 ' Diurnal variablility, surface energy budget, 337
penny-shaped crack under uniform Drawdown, 393
pressure, 182-187 Drying, 227-228,230,252-253,255-256
penny-shaped crack under uniform shear Ductile fracture, 91, 111
loading, 192-197 DYRESM 2D model, 438
plane crack under arbitrary shear DYRESM model, see Model, DYRESM
loading, 170- 177
point force loading of a penny-shaped crack,
164-165, 177-182 E
Creep constrained cavitation, 135, 142 Eddy diffusivity, 423-424
Creep failure, 134, 141 Efficiency
Criteria for assumption of one-dimensionality, bubble system, 441,443,446-447
426,429 dependence on number of plumes, 449
Cup-cone, 115 Peak,447
r:
relationshipto 447,450
D Elastic constants, 155, 157, 166
Element vanish technique, 113, 143
DarcyIDarcian flow, 226-227,229,231-236, End wall, influence of, 400
242-243,247,249-250, Entrainment
254-255,257-258 surface energy budget, 350-351
Decohesion, 106 upwelling, 365-366
Deformation, energy of, 197 Epilimnion, 309
Destratification,440 Equilibriumequations, 156
Differential Extinctioncoefficient, 261, 265
absorption, 388-391
influence of source strength, 388
cooling F
example, 381-385 Fluidized bed(s), 227,234,260-261,
time scale, 385 264,266-267,270
deepening, 370-380 Force
example, 371-377 complex tangential, 177
intrusion formation, 378 normal, 164-165
model, 377-380 resultant, 167
heating, 380-391 unit, 166
convectivemotions, 386,387 Forschheimer, 230,235,238,242-243,
example, 381-385 247,249-250
phase lag, 385 Forward gradient method, 129, 137
Diffusion, 138 Fourier series, 172
Diffusive models, one-dimensional, 422-423 Fracture, see Crack problems
 Subject Index 49 1

Fracture toughness, 120 underflow, 407

Froude number Instrument resolution effect, 311
inflow, 425,427 Integral equations
outflow, 426-427 approximate solution for general
turbulent, 416 domain, 172-173
Function for a flat crack under normal loading, 160
Green’s, 161, 169, 174-175, 180 for a flat crack under shear loading, 172
potential, 154, 159, 167, 182, 188, 191 for a smooth punch problem, 168
Funicular stage, 256 Integral representation
for the reciprocal of the distance between
two points, 209
G
involving inverse cube of the distance, 211-212
Geothermal engineering, 226,228 of more general type, 206-211
GLVHT model, 438 Interface, 231-233,257,261
Grain boundary facet, 135 Inverse scattering, 299-301
Grain boundary voids, 118, 136 Ion exchange columns, 234
Green’s function, see Function, Green’s Iron blast furnaces, 231
Ground water heating, 236 Isotropic medium, 186, 190, 195-196
Ground water hydrology, 231
K
H
Kadomtsev-Petviashvili equation polycnoidal
Heat budget, 307 waves, 16
influence of latitude on, 312 Karman-Pohlhausen solution(s),230,238,240
Heat flux, influence of surface layer, 337 KdV-Burgers equation, 289,291,295-297
Heat pipe technology, 226,258 Kernel of integd equation, 199
Hypolimnion, 309 Kinematic hardening, 120
mixing in, 413-422 Korteweg-deVries (KdV) equation, 16,284,
dependence on N,413 286,288,290,292-298,300-302
influence of Wand I.,, 414 cnoidal waves, 17
modeling procedure of, 432-433 imbricate series, 22
polycnoidal waves, 4
preconditioned Newton flow solution
1
for the soliton, 42
Imbricate Fourier series Kozeny-Cannan formula, 258
definition, 20
table of solitonknoidal wave applications, 19
table of solitonknoidal wave examples, 22
L
Inclusion problem, 198- 199 L-operator, 160
Inertia effect(s), 226,229-231,234, Lagrangian formulation of field equations, 86
238-239,249,271 Lake number, 317,360
Inflow, 405-413 for bubbler systems, 446
diverging channel, 410 for Canning Reservoir, 427
end wall influence, 410 for inflows, 429
Froude number, 425,427 interpretation for modeling, 427,429
Lake number for, 429 variation with latitude, 320
modeling procedure, 433-435 Langrangian
multiple velocity concentration, 410 layer mixing, 435
separated flow, 409 layer thickness increase and decrease, 435
turbulence, 412 model layers, 424,435
492 Subject Index

Laplace operator, 201 DYRESM 2 4 438

LARM model, 438 GLVHT, 438
Latitude LARM, 438
influence on heat budget, 312 results, 438-439
variation of Lake number with, 320 Modeling procedure
Layer inflow, 433
increase and decrease, 435 interpretation of Lake number for, 427,429
mixed, dynamics modeling mixed layer dynamics, 431-432
procedure, 431-432 mixing in the hypolimnion, 432-433
mixing, 435 outflow, 433-435
Leverett’s Correlation, 258 surface fluxes, 430-431
Loading Modons, 31.45,71,73
axisymmetric, 154 Moisture migration, 227,253,256
point force, 164-167, 177- 182 Moment, tilting, 191
shear, 170-177 Momentum boundary layer, 230
uniform, 182-187, 192-197 Monte Carlo method, 266-267
Local thermodynamicequilibrium, 227,258 Multiphasetransport, 226-227,253-258,260
Local volume-averaging technique,
229,253-255,257
Localization, 96, 112, 114, 123 N
Noncondensible(gas) effects,
227,254,257-258
M
Non-Dmian effect(s), 226-227,229,238-239,
Mechanismsfor turbulence, 416 243,247,249,271
Metal processing, 230 Nonlinear periodic waves, 285,295-298
Metalimnion, 309 Nonlocal solitaly waves
Mixing applications, 60
boundary, 421 capillary-gravitywater waves
due to bubble system, 441 (PRG equation), 69
efficiency,417 numerical weather prediction
horizontal, in upwelling, 366-369 (slow manifold), 65
Model plasma modons, 71
bubble plume, bubbler system, 442-444 quasi-geostmphic monopole vortices, 71
coupling with fan field, 445 Rossby waves, 60,71-73
double plume, 442-443 @breather (Higgs boson), 61
simulation results, 449-450 definition, 49
single plume, 442 exponentialsmallness, 54
calibration,423,425 far field analysis, 52
differential deepening, 377-380 nanopteroidalwave, definition, 50
DYRESM, 424,429,435 nanopteron, definition, 50
integral, of surface energy budget, 343-350 numerical methods, 57
layers, Lagrangian, 424,435 cnoidal matching, 58
MIT, 429 radiation basis functions, 57
one-dimensional perturbation theoy, 55
criteria for assumption of, 426-429 radiatively decaying soliton, 49
diffusive models, 422-423 Non-normality, 95.97
process-based models, 424-425 Nuclear fuel rods, 227,267,271
turbulence closure, 425 Nuclear waste repositories, 226,231,242,251
results, 435-437 Number
two-dimensional Froude
 Subject Index 493

inflow, 425,427 Packed cryogenic microsphere insulations,

outflow, 426-427 226-227
Lake, 317, 360 Packed filters, 234
for bubbler systems, 446 Parent thermocline, 311
for Canning Reservoir, 427 Pebble-type heat exchangers,234
for inflows, 429 Pendular stage, 256
interpretation for modeling, 427,429 Penny-shaped crack, 135
variation with latitude, 320 Permeability,230,256,258-259
Reynolds, turbulent, 417 Petroleum reservoirs, 226,231
Richardson, 328 Phase change process, 226
Wedderburn, 316,340,358,426-427 Plastic potential function, 128
Numerical boundary value methods for direct Poisson, coefficient, 166
computation of solitons Polycnoidal waves
iteration initialization strategies definition, 4
continuation, 26,36 Hill’s spectrum method, 8
in an artificial parameter, 27,38 Phase Variable Boundary Value
low order finite difference or Problem, 12- 16
spectral method, 26 relationship with multiple solitons, 4
perturbation theory, 25 Stokes series for, 13
residual inhomogeneity (“homotopy ”) theta function method, 10
method, 27 variational principle, 8
iteration schemes Porosity, 229-230,232,234-235,239,258,271
artificial diffusion, 28 Porous ductile solids, 89, 120
Eydeland-nrkington method, 44 Powder compacted metals, 85,91
Petviashvili scheme, 47 PRG equation, 53
artificial time methods, 39 Process-based models, one-dimensional,
Newton-Kantorovich, 28.39 422423
nonlinear Richardson’s iteration, 29, 39 Pseudospectral method
preconditioned Newton flow, 40 algorithmdescription, 33
choice of basis set, 30
0 rational Chebyshev functions, 33
One-dimensionalcriteria for assumption sinc basis, 42
of, 426-429 staggered grid preconditioning,43
One-dimensionaldiffusive models, 422423 symmetry, exploitationof, 31.35
One-dimensionalprocess-based tensor product basis, 35
models, 424-425 Punch, see also Contact problems
circular, 187-191
One-dimensionalturbulence closure, 425
outflow inclined, 191-192
Froude number, 426427 settlement, 188
linear stratification smooth, 165-167
line sink, 397-402
point sink, 4 0 2 4 5 R
point sink, 395-397
two layer, line sink, 395-397 Radiation, 431
withdrawal envelopes, 435 absorption remission, 431
surface flux, 321
Rate sensitivity, 128
P
Ray tracing method, 266-267
Packed bed(s), 227,230,234-235,259-261, Regenerativecooling, 233
264,266-268,271 Resonant triad, imbricate series for, 22
 494 Subject Index

Reynolds number, turbulent, 417 Surface fluxes, 321-334

Richardson number, 328 bulk aerodynamiccoefficients, 430
Rossby radius, 426 bulk formulae, 322
Rotation, 404 influence of stability, 325-328
modeling procedure of, 430-431
S
Monin-Obukhov length, 325
radiation, 321,431
Scattering coefficient, 261, 265 absorption remission, 431
Seasonal behavior, 306-321 Richarson number, 328
Shearbands, 96, 114, 124 similarity pmfile, 325
Shear waves, 398-399 surface roughness, 331-333
Size-scalesof particles, 109, 112, 116 transfer coefficients,327-334
Softening, 102 wave steepness, 332
Solar thermal central receiver, 251 Surface layer, 334-353
Solid matrix heat exchangers,231 examples of, 335-340
Soliton experiments, 284-286 heat flux influence, 337
Soliton formation, 286-294 modeling procedure, mixed layer dynamics,
Solitons, 283-302 431-432
Spherical-capsshape,138
Stability, 314
influenceon surface fluxes, 325-328 T
Stokes series, 13,30,58 Raring modulus, 117, 119
Stratification Rmperature dependence of materials,
general, 447 133, 137, 140
in relation to plume parameter P Rmperature minimum, 391
linear, 443 Rnsile test specimen, 115
line sink, 393-395 Thermal bar, 390-391
point sink, 395-397 Thermal boundary layer, 230-231,239
two-layer, 444 Thermal characteristics, 308-309
line sink, 393-395 Thermal dispersion (effects), 231, 235-236, 239
point sink, 395-397 ?taction, see Stress
influenceon bubble plume, 442-443 ?tansient analysis, 133
stress Transversely isotropic material, 155-156
complex, 157 Triaxiality, 130, 137
components of in cylindrical polar 'Ikiple momentum boundary layer, 235
coordinates, 190 Tropical lakes, 313
intensity factor, 165-167, 181, 184, 189, 194 'Ihrbulence
Stress-strain relations, 156 closure, one-dimensionalmodels, 422-423
Surface and energy budget mechanisms for, 416
billowing, 343 -0-dimensional models
closure, 349 DYRESM,438
dissipation, 348 GLVHT, 438
diurnal variability, 337 LARM, 438
energy leakage, 347 results, 438-439
entrainment, 350-351
integral model, 343-350
penetrativeconvection, 348 U
shear production, 348 Underground coal gasification, 231
surface stirring, 348 Uniform pressure, 182
Surface cracks, 114, 127 Uniform shear tractions, 192
 Subject Index 495

Upwelling Viscoplasticity, 128

associated shear, 358 Viscous-inertiatransition, 400
entrainment, 365-366 Void coalescence, 91, 108, 139
example of, 354-358 Void growth, 93, 107
interface tilting, 354 Void nucleation, 93, 102, 106, 139
modal response, 359,363 Void sheet failure, 101, 110, 114, 126

W
V
Wedderburn number, 316,340,358,426-427
Variable porosity effects, 226,229-231,234 Whiskers, 107
Vertex effects, 101, 106, 120, 125 Withdrawal
Vertical diffusion coefficient, 413415,421 envelopes, 435
Virtual works, 87 layer thickness, 401-403
This Page Intentionally Left Blank