Linear and Nonlinear Waves Textbook)

A Course on Nonlinear Waves
Nonlinear Topics in the Mathematical Sciences

An International Book Series dealing with Past, Current and Future Advances and
Developments in the Mathematics of Nonlinear Science
Editor:
MELVYN S. BERGER
University of Massachusetts at Amherst, U. S.A.
VOLUME 3
A Course on
Nonlinear Waves
by
Samuel S. Shen
Department of Mathematics,
University of Alberta,
Edmonton, Alberta, Canada
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

Library of Congress Cataloging-in-Publication Data
Shen. Samuel S.
A course on nonl inear waves ! Samuel S. Shen.
p. ce. -- (Nonl inear topics in the mathematical sciences ; v.
3)
Includes index.
ISBN 978-94-010-4932-0 ISBN 978-94-011-2102-6 (eBook)
DOI 10.1007/978-94-011-2102-6
1. Non 1 inear waves. 1. Title. II. Series.
QA927.S48 1994
531' .1133--dc20 93-19296
TSBN 978-94-010-4932-0
Printed on acid-free paper
AU Rights Reserved
1993 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 1993
Softcover reprint of the hardcover 1st edition 1993
No part of the material protected by this copyright notice may be reproduced or

utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner.
dedicated to my teacher and friend,
Professor Mei-Chang Shen
Contents
Preface xi
Chapter 1
Asymptotic Expansion 1
1.1 Concepts of Asymptoticity . 2
1.1.1 Is a divergent series useful? 2
1.1.2 The symbols "", 0 and 0 . 5
1.1.3 Asymptotic sequences 6
1.2 Method of Multiple Scales . . . . 10
1.2.1 Introduction . . . . . . . 10
1.2.2 Description of the method of multiple scales. 12
1.2.3 The van der Pol oscillator . . . . . 15
1.2.4 A forced oscillator and its stability . . . . . . 18
Chapter 2
Hyperbolic Waves 25
2.1 Conservation Laws . . . . . . . . . . . . . . . . . 26
2.1.1 The traffic flow problem . . . . . . . . . . 26
2.1.2 Conservation laws in a continuum media. 28
2.1.3 Jump boundary conditions 33
2.2 Characteristic Method . . . . . . . . . . 35
2.2.1 Linear initial value problem 35
2.2.2 Nonlinearity and wave breaking. 41
2.2.3 Shocks . . . . . . . 45
2.2.4 Entropy condition 46
2.2.5 Shock structure . . 49
Chapter 3
Water Waves 53
3.1 Governing Equations for Water Waves 53
3.1.1 Euler equations .. 53
3.1.2 Potential flow . . . . . . 54
3.2 Shallow Water Equations '" 58
3.2.1 Shallow water equations 58
3.2.2 Wave breaking on a beach . 62
viii Contents
3.3 Dispersive Water Waves . . . . . . . . . . . . . . . . 64

3.3.1 Dispersive waves . . . . . . . . . . . . . . . . 64
3.3.2 Boussinesq equations and the KdV equation. 66
3.3.3 Solutions to Korteweg-de Vries equations ... 70
Chapter 4
Scattering and Inverse Scattering 75
4.1 Scattering Method . . . . . . . 77
4.1.1 String-spring scattering 77
4.1.2 Linear Schrodinger equation. 78
4.2 Inverse Scattering for the KdV . 82
4.2.1 Inverse scattering method 82
4.2.2 KdV solitons . . . . . . . 85
4.2.3 KdV solitons with a wake 99
4.3 Lax Pair and KdV Hierarchy* .. 102
4.4 Biicklund Transform* . . . . . . . 105
4.5 Derivation of Inverse Scattering Method* 115
4.6 Soliton Fission . . . . . . . . . . . . . . . 119
Chapter 5
Burgers Equation 123
5.1 Viscous Fluid on an Inclined Plate 124
5.2 Cole-Hopf Transformation . . . . . 131
5.3 Stability of the Burgers Shock Wave 140
5.4 Interfacial Boundary Conditions of Two Viscous Fluids* 142
Chapter 6
Forced KdV Equation 147
6.1 An Ideal Flow Over a Small Bump 148
6.2 Supercritical Solitary Waves . . . . 153
6.2.1 Locally forced supercritical waves. 154
6.2.2 Non-locally forced supercritical waves 156
6.3 Subcritical Cnoidal Waves and Hydraulic Fall 161
6.3.1 Locally forced sub critical flows .. 162
6.3.2 Non-locally forced subcritical flows .. 165
6.4 Transcritical Periodic Soliton Radiation . . . 167
6.4.1 Approximate solutions and mass-momentum-energy-work
relationship . . . . . . . . . . . . . . . . . . . . . . 168
6.4.2 Spectral method for finding locally forced solitons 174
6.4.3 Program of the spectral scheme in Mathematica 178
6.5 Stability of Solitary Waves . . . . . . . . . . . . . . . . 183
Contents ix
Chapter 7
Sine-Gordon and Nonlinear Schrodinger 189
7.1 Physical Models of the Sine-Gordon Equation 190
7.1.1 Coupled torsional pendulums . . . . . 191
7.1.2 One-dimensional crystal dislocation . 192
7.1.3 Magnetic flux in a long one-dimensional Josephson junction 193
7.2 Solutions of the Sine-Gordon Equation 196
7.3 Optical Self-focusing . . . . . . . . . . . . 206
7.3.1 Pulse broadening due to dispersion 206
7.3.2 Optical self-focusing . . . . 209
7.4 A Simple Solution of the NLS . . . 212
7.5 Arctan Type of Solutions of the sG 213
Chapter 8
Selected Examples of Flow Instabilities 219
8.1 Concept of Stability . . . . . . . . . 220
8.2 Kelvin-Helmholtz: Gravitational Instability 225
8.3 Benard Problem: Thermal Instability . 230
8.4 Taylor Problem: Centrifugal Instability 237
8.5 Benjamin-Feir: Side-Band Instability 242
Chapter 9
Wave Interactions and X-Ray Crystallography 247
9.1 Wave Interactions . . . . . . . . 248
9.1.1 Introduction . . . . . . . . . . . . . . 248
9.1.2 Forced harmonic motion. . . . . . . . 249
9.1.3 Resonance conditions for nonlinear systems 250
9.1.4 Four-wave interactions. . . . . . . . . . . . 255
9.1.5 Nonlinear wave interactions in other systems 258
9.2 Phase Problem in X-ray Crystallography. . . . . 258
9.2.1 Bragg's law of X-ray diffraction . . . . . 258
9.2.2 Fourier representation of electron density 259
9.2.3 Coordinates in crystal cells 263
9.2.4 The phase problem. . . 265
9.2.5 Structure Invariants . . . . 267
9.2.6 Neighborhood principle . . 270
9.2.7 Probability distributions of structure invariants. 270
Appendix A
KdV Solitons via Inverse Scattering 277
Appendix B
KdV Solitons via Backlund Transform 283
B.1 Backlund Transform Program . . . . 283
x Contents
B.2 Two Solitons 289

B.3 Three Solitons 289
B.4 Four Solitons 290
B.5 Five Solitons 291
B.6 Six Solitons .. 293
B.7 Seven Solitons. 297
Appendix C
Derivation of the Stationary KdV 309
Bibliography 317
Index 323
Preface
The aim of this book is to give a self-contained introduction to the mathe-

matical analysis and physical explanations of some basic nonlinear wave phe-
nomena. This volume grew out of lecture notes for graduate courf;!es which I
gave at the University of Alberta, the University of Saskatchewan, and Texas
A&M University. As an introduction it is not intended to be exhaustive iQ
its choice of material, but rather to convey to interested readers a basic; yet
practical, methodology as well as some of the more important results obtained
since the 1950's. Although the primary purpose of this volume is to serve as a
textbook, it should be useful to anyone who wishes to understand or conduct
research into nonlinear waves.
Here, for the first time, materials on X-ray crystallography and the forced
Korteweg-de Vries equation are incorporated naturally into a textbook on non-
linear waves. Another characteristic feature of the book is the inclusion of four
symbolic calculation programs written in MATHEMATICA. They emphasize
outcomes rather than numerical methods and provide certain symbolic and nu-
merical results related to solitons. Requiring only one or two commands to run,
these programs have user-friendly interfaces. For example, to get the explicit
expression of the 2-soliton of the Korteweg-de Vries equation, one only needs
to type in soliton[2] when using the program solipac.m.
The structure of graduate programs in applied mathematics is changing. It
xi
xii
was common in the 1960's and 70's that a graduate student in applied math-
ematics would spend two years taking courses on linear mathematics in order
to pass a qualifying exam, only to be assigned a nonlinear topic for the the-
sis. Therefore graduate students did not have an opportunity to be exposed
to nonlinear equations until they got their Ph.D. thesis topics. The discovery
of solitons and strange attractors in the 1960's has revealed the overwhelming
importance of nonlinearity in nature and forced more and more schools to offer
nonlinear wave courses to beginning graduate students. There are many books
with titles like "Nonlinear Waves", "Solitons", "Nonlinear Wave Propagation" ,
etc. For introductory courses, many of these books are either too specialized or
too technical. G. B. Whitham's highly regarded book "Linear and Nonlinear
Waves" (1974) is perhaps the best choice as a text. Nevertheless, there appears
to be room for a book which complements Whitham's by including numerous
examples and by reporting on some recent discoveries about nonlinear waves.
In the process of writing this book, I have tried to introduce physical concepts
intuitively and in a direct manner. Many of the results are represented by
diagrams, and the physical and mathematical significance of these results is
carefully explained. As such, this book should be useful text at the beginning
graduate or advanced undergraduate level.
My own approach to understanding has been almost exclusively through
examples and pictures. Upon hearing a law or theorem, I would try to come
up with an example to which it could apply. Only then would I remember
the law and use it for my own purposes. This is particularly how I came
to understand nonlinear theories. Some people might argue that there exist
no nonlinear THEORIES but only nonlinear EXAMPLES. I would think that
a graduate student in applied mathematics would take the same approach.
He/she would comprehend nonlinear theories much better at the beginning
through reading or manipulating simple examples and through examining well
designed modern computer graphics. In this spirit, the book offers numerous
carefully selected examples, most of which are furnished with detailed solutions
and plots. Chapter 4 best shows the use of examples to illustrate a theory.
There are enough exercise problems in the book for a one-semester course.
Additional reading materials listed at the end of each chapter are for research
use and for help with the exercise problems. Beginning students need not to
spend too much time on these additional reading materials.
Since I have designed this book as a text for a one-semester course, the
book only elucidates some basic nonlinear problems. Many important and
fashionable topics (such as Yang-Mills equations and instantons, Lorenz equa-
tions and strange attractors, etc.) are not covered. To read this book one
needs background knowledge in general physics, undergraduate linear algebra
and differential equations. The sections whose titles have asterisks require
more mathematical background, and may be deferred. The book includes nine
chapters. Chapter 1 is on asymptotics and nonlinear ODEs. This chapter in-
tends to leave the readers with the impression that to mathematically handle
a nonlinear problem in nature, one should first apply asymptotic analysis to it.
Conservation laws are discussed in Chapter 2, with emphasis on the comparison
xiii
between the consequences of linear conservation laws with those of nonlinear

conservation laws. Chapter 3 treats water waves. Chapter 4 delineates the
scattering method and the inverse scattering method. The procedure of us-
ing the inverse scattering method to find 1-, 2- and 3-soliton solutions of the
standard Korteweg-deVries equation is presented in full length. Lax pairs and
Backlund transforms are briefly discussed in section 4.3. Chapter 5 studies
Burgers' equation. Chapter 6 discusses the forced Korteweg-deVries equation.
It emphasizes the steady state bifurcations and the unsteady state periodic
soliton generation. Most of the results are from research published after 1982.
Chapter 7 studies the sine-Gordon equation and the nonlinear Schrodinger
equation. Derivations of the sine-Gordon equation from various perspectives
are recapitulated. Several soliton type solutions for both sine-Gordon equa-
tion and nonlinear SchrOdinger equation are found via simple means. Chapter
8 presents some typical examples of wave instability and Chapter 9 studies
wave resonance. In the first section of Chapter 9, the basic theory of water
wave resonance is presented. In section 9.2, the Karle-Hauptman contributions
to the phase problem of X-ray crystallography are described in detail. This
chapter exposes the analogy between the water wave resonance theory and the
methodology established by Hauptman for solving the phase problem in X-ray
crystallography. Readers may be interested in knowing that Hauptman, who
got his Ph.D. in Mathematics, won the chemistry Nobel prize in 1985 together
with Karle for the work described in section 9.2.
I started to write this text for a nonlinear wave course in 1988. I offered a
special topics class on nonlinear wave propagation at Texas A&M University
in the spring of 1989. The first draft of this book constitutes the notes I used
in that class. I subsequently revised the notes, and used the revised version to
teach a nonlinear wave course already in the curriculum at the University of
Saskatchewan in 1990. The third major revision was done in 1991. Chapters
4 and 8 were significantly enhanced. The computer program soli pac. m was
then written and tested by G. E. Sarty. The explicit expressions of 1-, 2-,
... , and 7 - solitons were obtained. I used the third revision as lecture notes
in a nonlinear wave class at the University of Alberta in the winter term of
1992. At the University of Alberta, the nonlinear wave course had also been
included in the graduate curriculum. The final revision of the book was done
in 1992. Chapters 6 and 9 were almost totally rewritten because at that time
new results on the forced Korteweg-de Vries equation were available and the
phase problem in determining a crystal structure became more attractive than
ever.
I have to thank many people who helped me in writing this book. Pro-
fessor M. C. Shen provided me with many invaluable materials and comments
about the manuscript. J. A. Brooke, L. Gong, H. A. Hauptman, R. Karsten,
N. Lemire, R. E. Meyer, T. B. Moodie, A. Nip, L. Quinlan and R. Read have
proofread various parts of the manuscript, while S. X. Shen read the entire
manuscript. The students who took my nonlinear wave classes were most en-
thusiastic. They pointed out many errors in the notes and even contributed
solutions to some of the exercise problems. For example, L. Gong contributed
xiv
the explicit expression of the 3-soliton by the inverse scattering method, and
G. E. Sarty contributed three MATHEMATICA programs and wrote the ap-
pendices. Cheryl Piche typed part of the manuscript. I would like to thank
the Department of Mathematics and Statistics, University of Saskatchewan and
the Department of Mathematics, University of Alberta for their warm support
of this book. I would also like to express my appreciation to several staff from
Kluwer for their cooperation in the production process.
Last but not least, I would like to extend my most sincere gratitude to
Professor M. S. Berger for his constant encouragement and precise advice.
Without his invaluable guidance, it would not be possible for this book to have
come to fruition.
Sam Shen
Edmonton, Canada
March 1993
Chapter 1
Asymptotic Expansion
Asymptotic expansion is a method that approximately evaluates an integral

or solves an initial and/or boundary value problem for ordinary or partial
differential equations. Among these integrals or boundary value problems, there
must be a small parameter. This small parameter characterizes the scales of the
practical problem whose mathematical model is the integral or boundary value
problem. An asymptotic expansion is a formal integer power or fractional
power series expansion of the small parameter. The coefficient of the each
power yields a simpler integral or equation, which sometimes can be solved
analytically. This greatly reduces the complexity of the original problem. When
many terms are taken, the expansion gets so messy that one can never carry out
the expansion to many orders. Here, "many" usually means more than three
or four. Fortunately, the formal series expansion as an asymptotic expansion is
usually a divergent series, yet its partial sum of few terms usually represents a
good approximation to the original quantity we would like to find. Here, "few"
usually means two or three.
1
2 Chapter 1. Asymptotic Expansion
The asymptotic expansion technique is widely used in physics, chemistry

and engineering. Many physical laws are the first order asymptotic expansion
of the original complex problem, such as the soliton theory in water waves
governed by the Korteweg -de Vries equation and the optic soliton theory gov-
erned by the cubic nonlinear Schrodinger equation. This chapter introduces
some basic concepts regarding the asymptotic expansion and some simple ex-
amples. In other chapters of this book, this asymptotic expansion technique is
often applied.
1.1 Concepts of Asymptoticity
1.1.1 Is a divergent series useful?
It is clear that the integral
G(x) = 1
00
o
-xt
-Ie dt
+t
(1.1.1)
converges for any positive x. Let us evaluate this integral by the following
method. Disregarding convergence and divergence, we expand l/(l+t) formally
as
1
--=l-t+t 2
-t 3 + .... (1.1.2)
l+t
Substituting (1.1.2) into (1.1.1) and integrating the resulting equation term by
term like
(1.1.3)
we have
S(x)
(1.1.4)
This series is divergent for any real x. Hence the series (1.1.4) does not
represent the function (1.1.1). What is wrong? The reason is simple. The
series on the right hand side of (1.1.2) diverges when t > 1, and hence does not
represent the function 1/(1 + t). So we see that the reason we have divergence
is because of the non uniformity of the expansion (1.1.2).
The above fact is not surprising at all. However, the reader may be surprised
by the following fact. Using numerical integration on a machine (I used the
software called Mathematica on a SiliconGraphics computer), one can easily
obtain that
G(lO) = 0.091563 ...

1.1. Concepts of Asymptoticity 3
II Z 1.0 5.0 10.0 20.0 30.0 II

II G(z) 0.596347 0.170422 0.091563 0.047719 0.032290 II
82(Z) 0.0 0.160000 0.090000 0.047500 0.032222
8 4 (z) -4.0 0.166400 0.091400 0.047713 0.032289
8 5 (z) 20.0 0.174080 0.091640 0.047720 0.032290
8 6 (z) -100.0 0.166400 0.091520 0.047718 0.032290
81O(Z) -326980.0 0.146199 0.091546 0.047719 0.032290
8 13 (Z) 4.4x lOS 0.449415 0.091590 0.047719 0.032290
8 120 (Z) _5.5xlO H1 t; -7.1 X lOll:.! -5.1 X 1O(t; -3.6x10"lU -2.5x lOll!
II Z 40.0 50.0 60.0 100.0 300.0 II

II G(Z) 0.024404 0.019615 0.016398 0.009902 0.003322 II
82(Z) 0.024375 0.019600 0.016389 0.009900 0.003322
8 4 (z) 0.024404 0.019615 0.016398 0.009902 0.003322
8 6 (z) 0.024404 0.019615 0.016398 0.009902 0.003322
81O(Z) 0.024404 0.019615 0.016398 0.009902 0.003322
8 13 (Z) 0.024404 0.019615 0.016398 0.009902 0.003322
8 120(Z) -2362.20 0.019615 0.016398 0.009902 0.003322
8 300(z) -6.8 xlO l t;1S -1.8x10lU:.! -3.0xlO(1S -7.7xlO 11 0.003322
Table 1.1: The values of the function G(x) and their asymptotic approxima-
tions.
Let us compute the partial sum 8 n (10) of the series (1.1.4) up to 4 terms
8 4 (10) = 0.100000 - 0.010000 + 0.002000 - 0.000600 = 0.091400. (1.1.5)
We see that 8 4 (10) is a surprisingly good approximation of G(10). The relative
error IS
- 8 (10) I 100~
IG(10)G(10) 4
x <
1~
70 70.
This amazing phenomenon implies that even though a series is divergent,

its partial sum sometimes may still approximate the original function. This is
not a lucky coincidence. The reader may now like to look at Table 1.1 and look
for values of G(10), G(20), G(30) and G(40). You can see more evidence that
the partial sums 8 n (z) of the divergent series (1.1.4) approximately represent
the corresponding values of G(z) as long as z is large enough and n is properly
chosen.
To understand this fortunate success, we consider the difference, Rn (z),
between G(z) and the nth partial sum 8 n (z) of the series (1.1.4)
Rn(z) = G(z) - 8 n (z)
where the nth partial sum Sn (x) is
Sn (x) = .! - ~2 + ~3 - ... + (-1 t- d n - I)! . (1.1.6)

X x x xn
From
we have
(1.1.7)
It is clear that
Rn(X):S 1
0
00 n
t e
-xt n!
dt = x n+1' x> O. (1.1.8)
From this expression we can see that if n is small and x is large, then Rn(x)
is small. Hence Sn(x) is a good approximation of G(x). To get some feeling for
this idea, we refer again to some numerical results in Table 1.1. Since Sn (x)
is an asymptotic approximation to G(x) under the condition that x -7 00, it
is not surprising that when x = 1, Sn(l) deviates far away from G(l). On
the other hand, when x = 5, it can already be considered to be an "infinity"
and Sn(5) approximately represents G(5) for some n. The bold faced numbers
in the table represent the most accurate asymptotic approximations. Never-
theless, the best asymptotic approximation is usually not defined as the most
accurate one. Instead, the best asymptotic approximation must have the fol-
lowing two characteristics: good accuracy and few terms. In our case of G(x),
the best approximations are not those bold faced numbers. Instead, they are
S2(X) as x -7 00. It is obvious that the best asymptotic approximation defined
this way is not unique since "few" terms may mean one, two or three terms.
The reason to define the best asymptotic approximation this way is for practi-
cal applications. For practical complex physical systems, after determining the
scales ofthe problem, one can do asymptotic expansion like that of (1.1.4), but
it is normally not possible to carry out the expansion to more than four terms.
Therefore, people always look for asymptotic approximations which have only
one term, two terms, three or four terms. These are the best asymptotic ap-
proximations and have the so called "parsimonious" property of the asymptotic
expansion. A simple explanation of the "parsimonious" asymptotics was given
by Ludwig (1983). For a nontrivial example, the reader may see an interesting
paper by Keller and Rubinow (1960). They found a very good approximation
to some PDE eigenvalue problems. For an application to a complex physical
system, the reader may see a paper by Shen (1992). He derived some simple
physics laws from complex two-layer flows.
1.1.2 The symbols "', 0 and 0

In this section we will compare the sizes of two functions when their common
independent variable approaches a certain point (infinity is included). Let f(x)
and g(x) be two real valued functions defined in (-00,00). If If(x)jg(x)1 is
bounded in a neighborhood of x = a (a may be equal to oo), then one of the
following three cases must hold.
(i) limx-ta[J(x)jg(x)] = 1. In this case f(x) and g(x) are of almost the same
size in a neighborhood of a. We write
f(x) '" g(xj (x -+ a).
In the situation of no possible ambiguity, we simply write
We say that f(x) is an asymptotic approximation to g(x) as x is near a

(or g(x) is an asymptotic approximation to f(x) as x is near a).
(ii) limx-ta[J(x)jg(x)] = O. Now f(x) is qualitatively smaller than g(x) in a
neighborhood of a. We write
f(x) = o{g(x)} (x -+ a) ,
or, simply, f = o(g). We say that f( x) is of order smaller than g( x) when
x IS near a.
(iii) If it is neither case (i) nor case (ii), and if If(x)jg(x)1 is bounded in a
neighborhood of x = a, we write
f(x) = O{g(x)} (x -+ a) ,
or, simply, f = O(g). We say that f(x) is of order not exceeding g(x)
when x is near a.
Usually if none of the above three cases occurs, then we exchange the roles
of f(x) and g(x). Most likely, case (ii) will apply to g(x)j f(x). For example, if
f(x) = x and g(x) = 1- cos x, the limit liffix-to f(x)jg(x) does not exist. But
limx-tog(x)jf(x) = o. So g(x) = o{f(x)} as x -+ 0, i.e. 1- cos x = o{x} as
x -+ O.
Examples:
a) x 2 =0{x} (x-+O);
b) x'" sin x (x -+ 0);
c) ~ '" X~l (x -+ 00);

d) eX = O{sinhx} (x -+ 00);
e) 1 - 2
X2 ' " cos X (x -+ 0);
f) x 2 = o{tan x} (x -+ 0);
g) x1 - ~
1! + X3
2!
- ... + (- 1)n-1 (n-1)!
xn '" Jo l+t dt
roo e-'" (x -+ 00),
roo ,,+1
h) If I(x) '" XV (x -+ 00), then Jx I(t)dt '" -~+1 when v < -1 .
Remark 1: The notations o(g) and O(g) are often used to denote the classes
of functions 1 with the properties (ii) and (iii) respectively. Particularly, they
are used to denote unspecified functions which have these properties. This is
generic and has been adopted in the study of asymptotic solutions to differential
equations.
Remark 2: Selected operation rules of 0 and o.
a) 0(0(1)) = 0(1);
b) 0(0(1)) = 0(0(1));
c) O(l)O(g) = O(lg);
d) O(g)o(g) = o(l)o(g) = o(lg) if 9 = 0(1);

e) 0(1) + O(g) = 0(1) + 0(1) = 0(1) if 9 = 0(1);
f) 0(1) + o(g) = 0(1) if I'" g.
Exercises:
1. Show that if i = O( 1/Ji), i = 1,2, ... , n, and ai are constants, then
2. Show that if i = O( 1/Jd, i = 1,2, ... , n, then TIi=l i = 0 (TIi=l1/Ji)

3. Verify the example h).
1.1.3 Asymptotic sequences

The first concept in this subsection is the asymptotic sequence. We denote the
sequence of functions 1 (x), 2 (x), ... by {n (x n,
or {n}. The sequence of
functions {n} is said to be an asymptotic sequence as x -+ a if n+1 o(n). =
In practical applications, the functions under consideration depend not only
on x but also on a parameter f. Let{ 1/Jn(x; En be such a sequence, where
x ERe R (the real line) and R is a prescribed set in R. If 1/Jn+1 (x = a; E) =
o{ 1/Jn (x = a, E)} as E-+ 6, then we say that {1/Jn (x; E) } is an asymptotic sequence
at x = a. If 1/Jn+1(X; E) = oNn(x; En as E -+ 6 is valid for every x E R , then
we say that {tPn(Xj En is a uniformly asymptotic sequence in the domain R.

An asymptotic sequence which is not uniform in R is said to be singular in R.
Example 1: The following sequences are all asymptotic sequences.

a) {xn} (x--+O),
b) {x- n } (x --+ 00),
c) {eXx->',,} (x --+ 00), An real positive, and An+! > An for each n,
d) {E~=l(-I)k-dk~l)!}, (x --+ 00).
Example 2: {Ene- lxl } (E --+ 0) is a uniformly asymptotic sequence in R.
We say that E:=l !k(X) is an asymptotic expansion of !(x) as x --+ a if

n
I!(x) - :L: !k(x)l = o(Jn(x)) (x --+ a) .
k=l
Exercise: Show that
~(_I)n-l (n - I)!
L..J xn
n=l
is an asymptotic expansion of
1
00
o
e-xt
--dt
1 +t
as x --+ 00.
An asymptotic expansion does not need to be a convergent series. For in-
stance, the asymptotic expansion in the above exercise is divergent for any
value of x. In fact, almost all asymptotic expansions in practical applications
are divergent series. We take only the first two or three terms in our approxi-
mation. Sometimes, we take only one term. From Table 1.1, one can see that
taking more terms does not always provide a higher degree of accuracy. Most
often, taking more terms provides less accuracy since the asymptotic series are
usually divergent and the partial sums of many terms are either very large or
have oscillatory signs. The optimal number of terms to be taken varies from
problem to problem.
The second concept in this subsection we want to address is the asymptotic
expansion. We say that E:=l !n(Xj E) is an asymptotic expansion of !(x, E) if
n
1!(Xj E) - :L: !k(Xj E)I = o{fn(Xj En (E --+ 0) .
k=l
If the above expression holds for any x in R, then we say that E:=l !n (x; E) is
a uniformly asymptotic expansion of !(x, E) in R. If the asymptotic expansion
is not uniform in R, then it is said to be singular in R. These concepts are
ym J-==-""H-~
(1-a}(1-1/e)+O( E)
o~~-*--------------------------~-------x
1
Figure 1.1: Asymptotic solution of the boundary layer problem (1.1.9) -

(1.1.11).
particularly useful in finding asymptotic solutions for differential equations.

The boundary layer theory for viscous fluid flow is a typical example.
Example: Consider a boundary layer equation
f.yI' + y' = a, 0 < a < 1, 0 < f. 1, 0 < x < 1, (1.1.9)
y(O) = 0, (1.1.10)
y(l) = 1. (1.1.11)
The boundary layer is the narrow region near x = 0 (see Fig. 1.1).
Before carrying out the formal mathematical analysis on the above problem,
let us qualitatively examine the solution behavior. It is common knowledge that
in a well defined physical problem all the additive quantities in an equation have
the same order in terms of size. In order to make (1.1.9) meaningful, y" must
be of order aC 1 (which is very large!) at least in a subinterval of [0, 1]. Hence
the curvature is very large in this subinterval. The solution has to increase from
y = 0 at x = 0 to Y = 1 at x = 1. Since there is no non-uniform disturbance
in the interval [0,1], the largest curvature can only appear near the boundary
where the boundary constraint acts as the force that bends the solution. The
large curvature characterizes a sharp turn of the solution curve y = y( x). This
sharp turn separates the narrow boundary layer from the outside region.
The above argument shows the existence of the boundary layer. Later, we
will argue that the boundary layer can only occur at the left boundary.
The boundary value problem (1.1.9) - (1.1.11) is a linear problem and has
an exact solution
(1.1.12)
We desire an asymptotic solution which can be compared with the exact solu-
tion (1.1.12). The significance of finding an asymptotic solution is to demon-
strate the method that can be applied to find asymptotic solutions for some
nonlinear problems whose analytic solutions can not be found.
Since ( is small, the term ('1/' should not play the dominant role in the entire
region (0,1). Let us first assume ( = O. Then
Y = az + b
=
is the general solution of the differential equation '1/ a. But y az+b can not=
satisfy both boundary conditions. Now we are in a dilemma. Which boundary
condition should make the solution satisfy? This is equivalent to asking where
the boundary layer is. Let us try to satisfy the y(O) =
0 condition. Then
y = az is the solution. Hence y(l) =
a < 1. So y(z) needs to increase very
fast and is concave up near z = 1. The increasing rate must be '1/ > a (faster
=
than d(az)jdz a) near z = =
1. This implies that ('1/' a - '1/ < O. But this
contradicts the fact that y(z) is concave up near z = 1. Therefore, the large
curvature can only appear at the left boundary and hence the boundary layer
=
can only be at the left boundary. Thus, we choose b 1 - a so that the right
boundary condition y(l) = 1 is satisfied. We call
Yo = az + 1- a (1.1.13)
an outer solution (the reader will see why we use the word "outer" upon finish-
ing reading this example). Actually, Yo represents the exact solution outside of
the boundary layer.
To satisfy y(O) = 0, we make a transform
e= ~.
(
The geometrical or physical meaning of such a transform is the application of

the scaling technique near z = O. We know that the physical length (measured
by z) is small near z =
O. So we enlarge this length scale by dividing it by a
small number (. Thus, this transform is like placing a magnifying glass on the
boundary layer. Then (1.1.9) and (1.1.10) become
d2 y dy
+ - = a, y(O) = O. (1.1.14)
de
-2
de
Omitting a, (1.1.14) has a solution
Yi = C (1 - e-e) . (1.1.15)
We call Yi an inner solution since Yi represents the exact solution inside the
boundary layer. Because the boundary layer is thin (of order (), we used the
magnified scale e=z j ( instead of the original physical scale z.
Near the right boundary of the boundary layer, there should be a constant
which bridges the outer solution (the solution outside the boundary layer) to
the inner solution. This bridging constant is denoted by Ym. So we have
lim Yo
",-+0
= Ym = E-+oo
lim Yi . (1.1.16)
This gives Ym = 1 - a and c = 1 - a. The above limit on the left hand side
stands for the value of the outer solution value at the right boundary of the
boundary layer, and the limit on the right hand side for the inner solution value
at the same boundary.
Finally we have an asymptotic solution
Yc = Yo + Yi - Ym = ax + (1- a)(l - e-~). (1.1.17)
The difference between Yc and the exact solution Y given by (1.1.12) is
Iy - Ycl = O(f). (1.1.18)
By definition, Yc is indeed an asymptotic approximate solution to the problem

(1.1.9) - (1.1.11). This method of constructing an approximate solution is
called the matched asymptotic expansion method.
Numerous examples show that when a sharp turn exists in a solution curve
of a physical quantity, the matched asymptotic expansion method is one of the
most efficient tool for finding an approximate solution. Sometimes, even nu-
merical methods may fail to treat the sharp turn while the matched asymptotic
expansion method works.
1.2 Method of Multiple Scales

1.2.1 Introduction
Consider the superposition of two harmonics:
Then
If IWi - w21 is small and both IwtI and IW21 are large, then
sinti - W2 t + (Pi - <p2)

2 2
oscillates much more slowly than
Namely,
modulates the faster oscillation
sin(Wi +W2 t + <Pi + <P2).

2 2
1.2. The Method of Multiple Scales 11
xl
x2
xl + x2
Figure 1.2: Pat phenomenon: superposition of two fast oscillations of similar

frequencies. Two time scales: faster time scale 2/(Wl + W2) ~ I/Wl ~ l/w2,
and slower time scale 2/lwl - w21.
This modulation is called the pat phenomenon in physics (see Fig. 1.2). Ap-
parently in this physical problem, there are two time scales involved. One is
the regular time t and the other is the slow time T = ft with
Physically speaking, with the slow time scale T, we can only observe the pattern
of the pat phenomenon. With the regular time scale t, we can only observe
the fast oscillation. Therefore, to seek a physically meaningful solution for a
problem, we need to identify how many time scales the problem has. Within
a specified time scale, an originally nonlinear problem may degenerate into a
linear one which can be solved more easily. This is the central theme of the
method of multiple scales.
Introducing multiple scales is natural in our common practices. For in-
stance, to describe the temperature change with respect to time, we may use
"hour" as the time scale to describe the temperature variation over a day or
several days. Whereas, when we describe the temperature variation over a year
or several years, "season" is used as the time scale. "Season" is a slower time
scale than "hour". Some weather events occur in the seasonal scale, such as
the seasonal cycle of the global climate, and others in the scales of days or even
hours and so on. For example, the city of Calgary in Canada is strongly influ-
enced by the weather event called "Chinook" which is due to the strong Pacific
atmospheric current and the Rocky Mountains blocking. The "Chinook" can
cause a temperature fluctuation of 20C within an hour. A Calgarian may say,
"It was snowing an hour ago. You see, now it is 20C. What a joke!". So, the
"Chinook" is a weather event whose time scale is hours or minutes. If you only
look at the monthly temperature, you will not be able to find the "Chinook"
weather event at all. On the other hand, some weather events have to be de-
scribed by seasons or years. For example, the EI Nino event is in this time
scale. The EI Nino is an event of the Pacific Ocean warming and happens once
every two to six years. It strongly influences the precipitation distribution in
the continent of Latin America, South Asia and part of North America. Pre-
diction of the EI Nino event is very valuable to those people who live in those
areas and is a very important research area of meteorology.
Another example is the topographical description of land surface. To de-
scribe the topography of the Saskatchewan River around the University Bridge
in Saskatoon, Saskatchewan, we perhaps use 20 meters as the length scale. On
the other hand, when we describe the topography of the whole Province of
Saskatchewan, we may use 50 km as the length scale. Here, 50 km is a larger
length scale than 20 m.
1.2.2 Description of the method of multiple scales

We will describe the method of multiple scales through an example. Consider
a damped linear oscillator
x+ x = -2f.i:, 0 < f 1, t > O. (1.2.1)

The left hand side represents a perfect oscillator and the right hand side a
damping force that resists the oscillatory motion. The perfect oscillator x +
x has its own frequency (a time scale), and the damping 2f.i: characterizes
another motion behavior (another time scale). Since the damping force is
scaled down by f, the motion resistance is presumably small. Consequently, the
time scale characterizing this motion is large. Let us now check this analysis
mathematically.
The general solution of this equation is
x = ae-tcos(~ t + 8). (1.2.2)

Here ae-t is the slowly decaying amplitude ofthe oscillation cos(~ t+8)
(see Fig.1.3). From equation (1.2.2) and Fig. 1.3, we can see that ft and
v'f="f2 t ~ t are the two time scales, t and ft, for the damped linear oscillator.
Next we will show how to use the method of two scales to find an asymptotic
solution to (1.2.1). Let
t, (1.2.3)
1.2. The Method of Multiple Scales 13
-1
-2
-3 0 50 100 150 200

Figure 1.3: Two time scales: faster oscillation time scale 1 and slower decay
time scale 1/e.
T2 = t, (1.2.4)
Z z(T1, T 2, e). (1.2.5)
Then
:i: ZT1 + eZT2'

Z = ZT1TI + 2ezT1T2 + e2zT2T2 .
Equation (1.2.1) becomes
ZT1T1 + Z + 2e(zT1T2 + ZT1) + e2(zT2T2 + 2ZT2) = O. (1.2.6)
Assume (1.2.6) has an asymptotic expansion of the form
Z = zo(T1 , T2) + eZ1(T1, T2) + e2z2(T1, T2) + .... (1.2.7)
Substituting (1.2.7) into (1.2.6) and assembling the terms accordipg to like
powers of e, we obtain the following equations
Order eO: ZOT1T1 + Zo = 0, (1.2.8)

Order e1 : Z1 T1TI + Z1 = -2(ZOT1T2 + ZOT1 ) (1.2.9)
Equation (1.2.8) has general solution: .
Zo = A(T2) cos(T1 + (T2)) (1.2.10)
Inserting (1.2.10) into (1.2.9), we have
Z1 T1TI + Z1 = 2 [ A' (T2) sin(T1 + (T2)) + A(t2)' (T2) COS(T1 + (T2))

+A(T2) sin(T1 + (T2))] . (1.2.11)
The right hand side is all harmonics offrequency 1, which is equal to the natural
frequency of the operator of the left hand side. Hence the general solution of
this equation will grow linearly in time T1 with no upper bound. This is the
phenomenon of resonance in physics. Those terms on the right hand side which
make the general solution grow indefinitely in a time scale are called the secular
terms. Apparently the solution growing indefinitely in time T1 is unphysical.
Hence the physically meaningful perturbation is the one whose secular terms
always vanish. Thus,
A' (T2) sin(T1 + tP(T2)) + A(T2)tP' (T2) COS(T1 + tP(T2))

+A(T2) sin(T1 + tP(T2)) = O. (1.2.12)
This further implies

(1.2.13)
and
(1.2.14)
Hence
A(T2) = ae- T2 , tP = constant.
From (1.2.8) and (1.2.10), we have formally obtained an asymptotic solution
to (1.2.1)
Xa = ae- T2 COS(T1 + tP) + O(c) ,
or
Xa = ae- lt cos(t + tP) + O(c). (1.2.15)
This solution agrees well with the exact solution (1.2.2). Indeed, one can easily
check that Ix - xal = O(c).
In summary, using the method of multiple scales to find an asymptotic
solution to a problem, we first need to identify how many temporal and spatial
scales there are in the problem. For instance, the problem (1.2.1) has two time
scales. One is the oscillation scale and the other is the amplitude decaying
scale. Secondly, we introduce the new scales T1 = =
T1 (t; c), T2 T2(t; c), T3 =
T3(t; c), etc. and assume the unknown u an asymptotic expansion form: u =
Lf=D cAkuk(T1, T2," .), where 0 ~ A1 < A2 < A3 < .... Upon assembling all
the terms according to like powers of c, one obtains a group of equations, which
usually are all linear even though the original problem is nonlinear. In many
cases, the nonlinearity is the manifestation of an interaction of some physical
quantity at different spatial and time scales. In a singled out scale, the same
physical quantity may manifest itself linearly. This phenomenon is sometimes
referred to as the method of nonlinear separation of variables. Thirdly, we solve
these linear equations with vanishing secular terms. Finally, changing the new
scales back to the original physical scale t and c, we can obtain an asymptotic
solution.
1.2. Method of Multiple Scales 15
1.2.3 The van der Pol oscillator

As an application of the method of multiple scales, we consider the van der Pol
oscillator
=
ii + u ( 1 - u 2 )u . (1.2.16)
There are three time scales in this problem. The first is t, which is for the
natural oscillation of the operator on the left hand side of (1.2.16) whose fre-
quency is one. The second is ft, which is for the amplitude decaying due to the
damping when lui> 1. The third is 2t, which is for the slow magnification
when lui < 1.
Here, it is not obvious why the third time scale should be introduced. The
reader may have already realized this point. Of course, if we just include the
scales t and ft and obtain a solution, then the time scale 2t will not be shown
in the solution and the asymptotic solution with only scales t and t will not
show the slow magnification phenomenon when lui < 1.
The reader may wonder how one can find out all the "hidden" time scales
before one actually has known the solution. The beauty of the multiple scales
method is that one does not have to know all the scales in advance. Fewer
scales included may give less accurate solutions. Sometimes, it may give you
the solution of another phenomenon associated with the physics which you
are not considering. That means you have chosen the wrong scaling. But,
you would know that the answer is not right for the phenomenon under your
investigation from the physics itself. Then, you would try another scaling or
finer scaling. For example, in the case of van der Pol oscillator (1.2.16), one can
use two scales: t and ft. Correspondingly, the solution obtained according to
the two scales will reflect the physics within these two scales. To improve the
solution, one can include finer scales like 2t or 3t. However, one must always
include the largest scales, such as t and ft for the van der Pol oscillator, in each
computation since macrophysical quantities normally manifest most noticeably
at the largest scale. Let
To t, (1.2.17)
Tl ft, (1.2.18)
T2 2 t, (1.2.19)
u = u(To, Tl , T2; )
uo(To, Tl. T2 ) + ul(To, Tl , T2 ) + 0(2). (1.2.20)
Substituting (1.2.17)-(1.2.20) into (1.2.16) and equating the coefficients of like

powers of , we have
UOToTo + Uo 0, (1.2.21)
Ul To To + Ul = - 2U OTo Tl + 2( 1 - u~ )UOTo , (1.2.22)
- 2U lToT1 - UOT1 Tl - 2UOToT2
+( 1 - u~ )Ul To + (1 - u~ )UOT1 - 2UOUl To. (1.2.23)
The general solution of (1.2.21) is
(1.2.24)
where c.c. stands for complex conjugate of the remaining part on the right hand
side of the equation. Substituting Uo into (1.2.22) results in
U1ToTo+Ul=-Z.( 2AT,- A + A 2A-) eiTo -ZA

3e3iTo + c.c. , (1.2.25)
where A denotes the complex conjugate of A. To avoid the indefinite growth

. of Ul, in To, the secular term -i( 2AT, - A + A2 A )eiTo should be set equal to
Zero, 1.e.
2AT, - A + A2 A = o. (1.2.26)
The complex function A can be written in the exponential form
1 ....
A = -ae'''' , (1.2.27)
2
where a and <p are real functions of Tl and T 2 Separating real and imaginary
parts of (1.2.26), we have
<PT, 0, (1.2.28)
1 1 2
aT, -(1-
2 -a
4 )a . (1.2.29)
Hence
<P <p(T2) , (1.2.30)

4
a2 (1.2.31 )
1 + c(T2)e- T, .
At this point, we have found the first order asymptotic solution
4
= cost + O(f) (1.2.32)
1 + (4/a5 - 1 )e- a
U
which satisfies the initial condition u(O) = ao, u(O) = O(f). So far the smaller
scale f 2 t has not been reflected in the solution (1.2.32).
Next we proceed to find the second order asymptotic solution. The equation
(1.2.22) has the general solution .
(1.2.33)
To determine B, we need to use the solvability condition of the third order

problem. Substituting (1.2.24) and (1.2.33) into (1.2.23), we have
(1.2.34)
where n.s.t. stands for nonsecular term and h.o.t. for higher order terms, and
Set the secular term Q equal to zero and write B in the exponential form
Then we have
(1.2.36)
(1.2.37)
From these equations we can derive
(1.2.38)
Thus
(1.2.39)
Since b should be bounded in the time scale T l , equation (1.2.39) implies
or,
1
<P = -16T2 + <Po. (1.2.40)
Finally, we have obtained the second order asymptotic solution
(1.2.41)
Here
4
(1.2.42)
a = 1 + (4/ a~ - 1 )e-et .
Equation (1.2.41) can be written in a more concise form
(1.2.43)
where
1 2 1 7 2
(J = 16{ t+Sdoga- 64 m +(Jo, (1.2.44)
00 = -t/Jo - {bo = constant. (1.2.45)
Now all the three time scales t, d and (2t are reflected in the solution (1.2.43).
1.2.4 A forced oscillator and its stability

Consider the following nonlinear oscillator
(1.2.46)
where
(1.2.47)
We include the two largest time scales: t and d. Let
To t, Tl =d,
x = x(To, T 1; () = x(O)(To, TI) + (x(l)(To TI) + O({2).
Then
From (1.2.46), we can derive equations for x(O) and x(l).
Order {a:
(1.2.48)
Order (l:
x(l) +w 2x(l) = _ux(O) _ 2x(0) _ (x(0)2 x(O) + h cos( at _ R). (1.2.49)

ToTo ,.. To ToTo To fJ
Equation (1.2.48) has the general solution

x(O) = A(TI)eiwTo + C.c. (1.2.50)
where A is a complex valued function of Tl and c.c. stands for the complex
conjugate.
From equation (1.2.50) and at = wTo + OTl + O({2), equation (1.2.49)
becomes
xlTo + w2x(1) = [-lliwA - 2iwA' - iwAIAI2 + ~ei(nTl-p) ]e iwTo

+n.s.t. + h.o.t. + C.c. (1.2.51)
where h.o.t. stands for higher order terms (i.e., the terms which are of order
O(c) or smaller) and n.B.t. for nonsecular terms whose angular frequency for
To is not w). Setting the secular term equal to zero, we have a differential
equation for A:
(1.2.52)
We write A into exponential form
and separate real and imaginary parts of (1.2.52) to obtain
P
I 1 3 + -p
+ -p Jl.
= . (A
-hsm uTl -
R
/J -
A.)
'I'
2 2 4w '
ptf>' = - 4~ cos( OTl - (J - tf> )
The above equations can be transformed to an autonomous system by let-

ting
(1.2.53)
Then the resulting autonomous system is
I 1 3 Jl h.
p +"2 P +"2 P = 4w sm"),, (1.2.54)
h
P"Y' - Op = 4w cos")'. (1.2.55)
If the oscillator is stationary in T l , then pi = ")" = O. Consequently, the

oscillation is according to the fast scale Tl and there is no modulation effect in
slower time scales. Hence for steady state oscillation we have
1 3 Jl h.
-p +-p= -sm")' (1.2.56)
2 2 4w '
h
Op = 4w cos")'. (1.2.57)
These two equations lead to
(1.2.58)
This equation shows how the oscillation amplitude p depends on the am-
plitude h of the forcing oscillation. p is a multiple valued function of h. This
implies that for a given forcing there may exist more than one steady state
oscillatory response, each of which is of a different amplitude. The curve p VB.
h is called the operating curve by meteorologists and physicists, or bifurcation
diagram by mathematicians, and is determined by (1.2.58). When Jl = -9,
0=1 and w = 1, the curve is shown in Fig. 1.4.
o
,.;
m'"N
"0
:>
...>
.>0
~t"i
E
o
CD'"
>..:
o
:J
o
LO
l.":
o
Q~~~--~----r---~--~--~----r---~--~---'
0.0 3.0 6.0 9.0 12.0 15.0 18.0 21.0 :M.O 27.0 30.0
h
Figure 1.4: Operating curve P vs. h : I-' = -9, n = 1 and w = 1.

The oscillation for a determined P is
x = 2pcos( wt + fnt - f3 -,) + O(f). (1.2.59)
When h = 18, there exist three sinusoidal responses according to Fig. 1.4.
The corresponding values of pare 1.13, 2.42 and 3.3 (see Figs. 1.4 and 1.5 ).
Among many of those steady oscillatory states, some are unphysical. They
can never be reached in practice. Hence, they can never be observed in nature.
This is because they are unstable. We will see that p = 2.42 in Fig. 1.4
corresponds to an unstable solution. Next we will prove a remarkable theorem
on the stability of the steady state oscillation.
Theorem 1.1 Ifdpfdh ~ 0 and2 p2:::; -1-', then the steady oscillation {1.2.59}
is stable. Otherwise, the steady state oscillation is unstable.
This theorem implies that if the bifurcation diagram has a negative slope at
a point, then the solution corresponding to this point is necessarily unstable.
A stability theorem described by the slope of the bifurcation diagram is usually
called the slope stability theorem. Many bifurcation phenomena in nature yield
slope stability theorems similar to the one presented here.
Proof of theorem:
Let Po and 'Yo be a steady state solution of (1.2.54)-(1.2.55). Then p = Po
and, = 'a
satisfy (1.2.56)-(1.2.57). Suppose that small perturbations tSp and
tS, are imposed to Po and 'a
respectively. Namely in (1.2.54)-(1.2.55), let
p = Po + tSp, (1.2.60)
x
6
4
2
-2
-4
-6
x
6
4
2
-2
oot
-4
-6
Figure 1.5: Three solutions when h = 18 {see equations (1.2.58) - (1.2.59)).
'Y = 'Yo + ~'Y . (1.2.61)
Substituting (1.2.60)-{1.2.61) into (1.2.54)-{1.2.55) and linearizing the resulting

equations around (po, 'Yo), it follows that
(1.2.62)
where
M - ( -3pU2 -1l/ 2 (hcos'Yo )/( 4w) ) (1.2.63)
- 0/po ( -h sin 'Yo )/( 4wpo )
and (sin 'Yo, cos 'Yo) in the matrix M is determined by (1.2.56)-(1.2.57).
Let ~p ex e>'t, ~'Y ex e>.t. Then (1.2.62) leads to an eigenvalue problem. The
characteristic polynomial is
det ( M - ,xI) =0 ,
where I is an identity matrix. This equation in turn can be written as
(1.2.64)
Consider the quadratic formula for a quadratic equation z2 + pz + q = 0:
-p Vp2 - 4q
z = 2 .
If both p and q are real numbers, then Re(z) < 0 if and only if p > 0 and
q > O. Thus for the two roots of (1.2.64) to have negative real parts, the
necessary and sufficient condition is that
2p~ < -1', (1.2.65)
and
(1.2.66)
From (1.2.58), we have
d (~~) (( 2~ ) 2) = 4 [ 0 2+ ~ (I' + 3p~ )( I' + p~) ] (1.2.67)
Since dpo/ dh and d( 2: )2/

dp~ have the same sign, a solution is stable if and
only if 2p~ < -I' and dpo/dh > O. This proves the theorem.
Therefore in Fig.1.4, only the lowest branch corresponds to stable steady
oscillations. The reason for this result is quite simple. When we look at our
original problem (1.2.46) , we find that as the oscillator reaches a threshold
(the criterion 2p~ < -1') that is higher than the certain value, the system gets
amplified. This amplification gets stronger as the time increases due to the
positive feedback of the system and finally leads the system to go unbounded
and hence unstable. On the other hand, the resistance force can always drag
the perturbation back to the stationary oscillation. In this case, the solution
is stable.
Exercises:
1. For x+wz = fI'X+fZ 2 X+fhcos(ot-p), I' > 0,0 =W+fO+0(f 2 ),h =
0(1) and f 1, do the similar computations as we did for (1.2.46). Show that
there is only one steady state oscillation for a given h. Prove that this unique
solution is stable.
2. In equation (1.2.58), if we fix 0, I' and h, then p is a multiple valued
function of W in some interval. Graph this multiple valued function for some
given values of 0, I' and h (see Fig. 1.4). Do we have the following stability
theorem?
Theorem: A solution is stable if and only if dp/dw > o.
Prove it if the theorem is true (see Reference [7]). Then try to explain the
physical meaning of your result. Can you do the same thing for the 1', p relation
or 0, p relation?
Additional Reading Materials
[1] F. W. J. Olver (1974), Asymptotics and Special Functions, Academic Press,

New York.
[2] A. ErdeIyi (1956), Asymptotic Expansions, Dover, New York, Chapter 1.
[3] D. Ludwig (1983), Parsimonious asymptotics, SIAM J. Appl. Math. 43,

664 - 672.
[4] J. B. Keller and S. I. Rubinow (1960), Asymptotic solution of eigenvalue

problems, Ann. Phys. 9, 24 - 75.
[5] S. S. Shen (1992), Forced solitary waves and hydraulic falls in two-layer
flows, J. Fluid Mech. 234, 583 - 612.
[6] G. H. Nayfeh (1973), Perturbation Methods, John Wiley, New York, Chap-
ters 1 - 4.
[7] A. H. Nayfeh and D. T. Mook (1979), Nonlinear Oscillations, Wiley, New

York, Chapter 4.
Chapter 2
Hyperbolic Waves
When describing the motion of any matter, it is always required that the mass
is conserved. This mass conservation law, whether conjugating with other
conservation laws or not, puts a constraint on the material motion so that a
material property of the matter, such as the density, moves along a real curve
in the space (x, y, z, t). This curve is usually called the characteristic of the
conservation law. The term "hyperbolic" is equivalent to the characteristic
curves being real. In this chapter, we will describe some basic examples of the
hyperbolic conservation laws and show how to use the characteristics method
to solve initial value problems for hyperbolic conservation laws. Here, we em-
phasize finding solutions to certain simple problems and explaining the physical
meaning of these solutions. Mathematical rigor and extensive studies may be
found in more specialized books (such as that by Smoller (1983)) on hyperbolic
conservation laws.
25
26 Chapter 2. Hyperbolic Waves
Car flow
p(X,t)
"",~.
x
The street
Figure 2.1: Trafic flow on a street.
2.1 Conservation Laws
2.1.1 The traffic flow problem
We consider the traffic flow on a single lane street. The x-axis is aligned along
the street. The density, p(x, t), denotes the number of cars on a unit length
of the street. The flux, q, denotes the number of cars passing through a fixed
point x in a unit time. The conservation of the number of the cars in the street
from Xl to X2 (see Fig.2.1) is
(net increase of cars)f(unit time) =

(incoming cars - outgoing cars)f(unit time).
The mathematical expression is
d
-d
t
l
x,
x
p(x, t)dx = q(x!) - q(X2). (2.1.1)
When IX2 - xd ~ 0, equation (2.1.1) becomes
Pt + qx = O. (2.1.2)
Equation (2.1.2) is a conservation law. To solve (2.1.2) one needs an additional

condition defining a relation between p and q and this additional condition
is called the state equation. To get such a state equation, let us analyze the
relationship between the cars' cruising velocity (s), the flux (q) and the density
of the cars (p). The following equation holds:
q = sp. (2.1.3)
It is clear that when the density is equal to zero, the flux must be zero. On the
other hand, when the density is very large, the cars can not move that fast. The
2.1. Conservation Laws 27
Pmax P
Figure 2.2: Curves of the state equations.
extreme case is that the cars are bumper to bumper. Now the density reaches
its maximum value (Pmax) and the cars cruising speed reaches its minimum
(8 = 0). Somewhere between minimum density (p =
0) and the maximum
density (p = Pmax), there is a value for the density (Pm) such that the flux
reaches its maximum (qmax). Therefore, the curve that describes the state
equation must pass through points (0,0), (Pm, qmax) and (Pmax, 0). There are
many curves that can connect these three points (see Fig. 2.2). The simplest
case is that the three points are connected by two straight lines:
if p$.Pm
(2.1.4)
if P>Pm
Let
a = qmax/Pm ~ 0, fJ =Pmax/Pm ~ 1,
and
I' = a/(fJ - 1) ~ O.
Then, the corresponding differential conservation law is
Pt +apx = 0, if p$.Pm, (2.1.5)

Pt -I'Px = 0, if P>Pm. (2.1.6)
Before analyzing this conservation law, let us look at the simplest hyperbolic
PDE
Ut + CUx = 0, -OO<X,t<OO, (2.1.7)
with an initial condition
p(X, t = 0) = I(x), -00 < x < 00. (2.1.8)

The solution to the IVP is

u=f(x-ct). (2.1.9)
This is a traveling wave solution that characterizes the uniform motion of the
initial profile f(x) toward the right at the speed c.
Now let us go back to the conservation law (2.1.5)-(2.1.6). When the density
is small (p < Pm), all the cars tend to accumulate and the density wave travels
forward (i.e., toward the right on the x-axis) at the speed c = 0:. This speed
is, however, not the cars' cruising speed. This phenomenon occurs when a
red traffic light is turned on to block a long train of sparsely distributed cars
which cruise at high speeds. On the other hand, when the density is very large
(p > Pm), the cars tend to distance from each other and the density wave
travels backward (i.e., toward the left on the x- axis). This happens when a
red traffic light turns to green allowing the many stopped cars (p = Pma:c) to
advance. A driver of the 6th or 7th car behind the red light can clearly see the
density wave traveling toward him/her. This density wave travels at a speed
c= ,.
There can be other state equations that connect the three points (0,0),
(Pm,qma:c) and (Pma:c, 0) on the (p,q) plane. The following two are useful in
the actual traffic flow modeling:
Quadratic State Equation:
qma:c ( ) (2.1.10)
q= ( ) P Pma:c - P ;
Pm Pma:c - Pm
Logarithmic State Equation:
q= q(ax) pIn (pm ax) . (2.1.11)

Pm In P;~'" P
According to Whitham (pp. 69-70 of Ref. [3]), the second state equation fits
some traffic flows quite well.
2.1.2 Conservation laws in a continuum media

Let 0 be a closed bounded domain in R3. A mass with density p(x, t) (unit:
mass volume-I) moves through the boundary 80. The flux vector field is
denoted by q(x, t) (unit: mass area -1 . time-I). The mass conservation in
o is
(net mass increase in O)/(unit time) =
(net mass flowing in through (0)/( unit time). (2.1.12)
The mathematical expression of this conservation is
~ f P dO = - f q. ds.
vt Jn Jan
The divergence theorem implies that
l (~: q) + \1 . dO = O. (2.1.13)
When the volume 101-+ 0, the above equation implies that
(2.1.14)
In this derivation, we used two assumptions: (i) 0 is a fixed domain; (ii) Pt+'\l.q
is a continuous function of x in o. These two assumptions do not hold in some
situations. Assumption (i) is broken when we use Lagrangian description for
a fluid flow. Assumption (ii) is broken when there is a shock across 0 since
the quantity Pt + \7 . q is no longer continuous across the shock. Nonetheless,
the physical fact of mass conservation still exists. Thus, in the case that the
above two assumptions are broken, the corresponding modified conservation
laws should be found. The rest of this section is devoted to this topic.
Next, we give a brief description of the Eulerian and the Lagrangian coor-
dinates in fluid flows and resolve the modification problem for the conservation
laws in the Lagrangian coordinates. The shock situation will be resolved in
section 2.2.1.
A description of fluid motion based on identifying the individual elements
of fluid is called Lagrangian. We start by picking up an element whose position
is initially identified by a. Its motion is described by x = x(a, t). Vector
x is the position of the element a at time t. To determine the fluid motion
in Lagrangian coordinates is to find the function x = x(a, t). In contrast to
tracing the motion of an element using the Lagrangian description, another way
to describe the fluid motion is to measure the status of the fluid at every fixed
point in the fluid domain at every moment. In fluid dynamics laboratories, most
of the observations are made this way since the instruments usually need to be
fixed at certain positions. A description of fluid motion by means of velocity
field v, density p, pressure p, and temperature T, considered as functions of
the observation point x and the observation time t, is called Eulerian. Such
a representation of fluid motion is used with only rare exceptions by fluid
dynamicists. In the Lagrangian description, the material domain V is moving
with respect to the observer. Equation (2.1.14) must take another form in the
Lagrangian space-time coordinates (a, t). This is the transport theorem.
Theorem 2.1 (Transport Theorem) Let \It be a fluid domain and f be a

state variable. Then
(2.1.15)
where f1t = :t + v . '\l is called the material differentiation operator, and v is

the velocity field.
Proof: Let Va be the initial position of the fluid domain \It (see Fig.2.3) and
J be the Jacobian of the domain transformation between Va and \It.
Instantaneous material domain
Initial material domain
Figure 2.3: Lagrangian description of the fluid motion.
Then
~ l ldv
v.
~ 1 JI dV
1~
Vo
= (fJ)dV
1
Vo
DI J dV +
= ivo Dt Vo
I DJ dV
Dt
1~ dV +1 IV v dV.
v. v.
In the last step we used
DJ
Dt = JV v. (2.1.16)
The proof is finished.
Exercise: Show that equation (2.1.16) holds.

The conservation of mass is a special case of equation (2.1.15) when 1= P,
the density. Since the material does not change in Vi with respect to its own
motion, we have
o=~
Dt
1
v.
Pdv=l
v.
(DP +PV.V)dV.
Dt
(2.1.17)
normal
traction
Figure 2.4: Surface traction and body force.
As the volume Ivtl --+ 0,

Dp
Dt + P V7 . v = o. (2.1.18)
This can be written as a conservation law
ap
at + V7 . (pv) = 0, (2.1.19)
or,
ap
at + V7 . q = 0, q = pv (flux). (2.1.20)
This equation agrees with equation (2.1.14) and is referred to as the continuity
equation in continuum mechanics.
The velocity field v has n components in the space R n (n = 1,2,3). The
conservation law (2.1.20) has only one equation, but itr has four unknowns:
v = (u, v, w) and p. To solve the conservation law, one needs some additional
conditions, unless the density p is known and the problem is one dimensional.
One of these additional conditions is the conservation of momentum, i.e., New-
ton's Second Law of Motion:
(net increase of momentum in vt)/( unit time)
= (total external force applied on the boundary avt)
+ (total external body force applied on vt).
The mathematical expression is (See Fig. 2.4)
~ { p v dV = { p f dV + { t ds. (2.1.21)
J~ J~ J&~
Here f is called the body force with unit [force mass- 1], and t is called the
surface traction with unit [Jorce . area- 1 ]. For example, in a purely gravita-
tional field of earth, the body force f = 9 = 9.8[newton.]kg- 1] = 9.8[m sec]-2.
The standard atmospheric pressure acting on an open surface of water is 10.1
[newton cm- 2] . On the boundary alit, t = II n where II is called the stress
tensor and n is the unit outer normal of alit.
By the transport theorem (2.1.15), the divergence theorem and the conser-
vation of mass, the above expression can be written as
f
lv,
(p Dv
Dt
_ 'V . II - Pf) dV = o. (2.1.22)
Letting IlItl-+ 0, we then have

Dv
P Dt = 'V. II + pf. (2.1.23)
The stress tensor II is related to v and this relation is called the constitution
relation. The constitution relation is usually determined by the properties (not
the motions) of the fluids. This constitution relation can be very different for
water and for polymer fluid.
For a Newtonian viscous fluid,
II= (-p+oX'Vv)I+2pD (2.1.24)
where D = (1/2) [( Vi,j + vj,i)l~,j=l is a 3 x 3 matrix and called the tensor of the
deformation rate, p the viscosity, oX the factor of volume compression, p the
pressure, and I the identity matrix. Using this expression, equation (2.1.23)
becomes
p (~; + (v . 'V)v) = -'Vp + (oX + p)'V ('V. v) +p 'V 2 v + pf. (2.1.25)
Equation (2.1.25) together with equation (2.1.19) are called the Navier-Stokes
equations. When oX = p = 0, the Navier-Stokes equations become the Euler
equations. In water wave studies, we usually take water to be incompressible
(oX =0) and inviscid (p =
0). It is very difficult to solve either Navier-Stokes
equations or Euler equations except for some special cases. These equations
have been challenging people for more than a century, and they continue to do
so. People have devoted much more attention to find approximate solutions to
the boundary value problems (BVPs) or the initial boundary value problems
(IBVPs) for these equations with certain assumptions of the physics, such as the
long wave assumption, and Boussinesq assumption, etc. With some of these
assumptions, one can apply the asymptotic procedure (discussed in Chapter
1) that reduces the original Navier-Stokes equations or Euler equations to a
simpler form. Solutions of these reduced equations are much more transparent
and can occasionally be found by analytical methods. Examples illustrating
this asymptotic reduction procedure will be presented in Chapters 3, 5 and 6.
Exercise: Derive equation (2.1.25) from equations (2.1.23) - (2.1.24).
Front L
Figure 2.5: Jump discontinuity on a front.
2.1.3 Jump boundary conditions

Let I; be the moving surface, called the front, across which a state variable 
is discontinuous (see Fig. 2.5). The velocity of the front at x E I; is u(x, t).
The velocity of the fluid element at x is v(x, t). The general form of the
conservation law for is
 dO + f
J'E(t)nO(t)
r
[ (v - u) . n ds
1
+f Q . n ds = f X dO. (2.1.26)
Jao(t) JO(t)
Here 0 is shown in Fig. 2.5. Quantity X is the external field, Q is the flux
of through a~, and [H = (x E I; + 0, t) - (x E I; - 0, t), the difference
of the state variable on two sides of the front. Applying equation (2.1.26) to
0 1 , O2 and 0, we canobtain
f [(v - u) . n + Q . n] 2 ds = 0. (2.1.27)
J'Eno 1
Letting II; n 01 -t 0, we have
[(v-u).n+Q.n]: =0. (2.1.28)
This is the conservation law on a discontinuity surface, such as a shock front.

Examples:
(i). Mass: CI> = p, Q = O. We have
[p(v - u) . n]: = O. (2.1.29)
(ii). Momentum: CI> = pv, Q = n (the stress tensor). We have

[pv(v - u) . n + n . n]i = O. (2.1.30)
(iii). Energy: CI> = U) pv + pE,

2 Q = v . n, where E is the internal energy.
We have
(2.1.31)
(iv). Rankine-Hugoniot condition: for the conservation law Ct + Ix (c) = 0, we

have: CI> = C , U = dx/dt = s = shock speed, n = 1 (one dimensional),
Q = I(c). The velocity of the material particle v s and is hence
regarded as zero. Thus
s [c]~ = [t(c)]~. (2.1.32)
It is this condition which determines the shock speed s in the gas dynam-
ICS.
(v). Free surface of water: Let z = I(x, y, t) be the free surface of a water
region. Hence
Ow = {(x, y, z) I z < I(x, y, t)} nO (2.1.33)
I
is water, and OA = {(x, y, z) z > I(x, y, t)} no is not water (air,
maybe). The air density and pressure are assumed to be zero on the free
surface. The conservation here is for mass. The state variable CI> is hence
the density p. There is no mass crossing the free surface (Q = 0). Then
by (2.1.28),
Pwater(v . n- U n) = Pair (v .n - U n) ~ O. (2.1.34)
Thus,
v n= u'n. (2.1.35)
For the free surface z = I(x, y, t), the unit outer normal vector is n where
n = (-Ix, -Iy, l)/Jr; + I~ + 1.

The velocity of the front (the free surface) is u = (0,0, It). The velocity
of the fluid particle at x and t is v = (u, v, w) (the velocity field). Con-
sequently, by equation (2.1.35) we have the kinematic condition for the
water waves on the free surface:
01 +u 01 +v 01 -w=O (2.1.36)
ot ox oy
where (u, v, w) = v is the velocity field of water.
2.2. Characteristic Method 35
2.2 Characteristic Method

In this section we will discuss how to use the characteristic method to solve the
hyperbolic conservation laws. Conservation laws are named for those differen-
tial equations which can be written into the form
Ut + f",(u) = O. (2.2.1)
Hence the continuity equation (2.1.19) is a conservation law, u and fERn, Z

and t E R.
We say that the conservation law is hyperbolic if all the eigenvalues of
(afdauj)~ j=1 are real for every u. For example, the following equations
are all hyperbolic conservation laws:
(i). Pt - 2p", =0 , eigenvalue = -2;
(ii). Pt + p2 p", = 0, eigenvalue = p2;
(iii). Pt + c p", = 0, Ct + pc c'" = 0, eigenvalues = (1 + c 11 - cl)/2.

A conservation law is said to be strictly hyperbolic if all eigenvalues of
(afdauj)~ j=1 are real and distinct for every u.
2.2.1 Linear initial value problem

Consider the initial value problem
Ut + a u'" = 0, -00 < Z < 00, t > 0, a is constant, (2.2.2)

u(z, t = 0) = 9(Z) E C 1(_00, 00). (2.2.3)
One can guess that

u = 9(Z - at) (2.2.4)
is a solution to the problem (2.2.2) - (2.2.3) . By uniqueness, this is the only
solution.
Proof of the uniqueness theorem: Suppose there were at least two solutions
for the IVP. Pick any two solutions Ul and U2. Let v = U2 - Ul. Then v(z,t)
satisfies the following IVP:
Vt + av", = 0 and v(z, t = 0) = O.

Multiplying the above equations by v, one gets
and v2(z,t = 0) = O.
This IVP is equivalent to
dv 2 dx
-=0,
dt
dt = a, and v2 (z(t), t = 0) = O.
Characteristic
1; = x - at
u(x.t=O) = g(x)
= g(x-at)
t = t
x
----------~---------------x
1;
a) Traveling wave solution b) Characteristic plane
Figure 2.6: The traveling wave solution for the simplest hyperbolic conservation
law (2.2.2).
This ODE initial value problem has only the null solution. Thus, Ul = U2 and
the uniqueness proof is finished.
So far, life is too easy! Now, what does equation (2.2.4) say? It implies that
the solution at time t is a copy of the initial data, and this copy moves along
the x-axis with velocity a (see Fig. 2.6). Thus the solution (2.2.4) is called the
traveling wave solution. The trace of the initial point ~ on the x-axis is given by
x - at =~. The line x - at = ~ on the (x, t) plane is called the characteristic.
By (2.2.4), along a characteristic, the solution U is a constant. Introducing
moving coordinates, one can also derive the solution (2.2.4) instead of guessing.
It is common knowledge that one can never integrate a PDE directly by the
fundamental theorem of calculus. To find a solution to a PDE by derivation
inevitably means that one needs to convert the PDE problem into an ODE
problem. One does this by tracing the solution along a certain line which is
called the characteristic. Let
r = t, ~ = x - at. (2.2.5)
For a fixed ~, x - at = ~ defines a line on the (x, t) plane. These types of

lines are called the characteristics. Along the characteristics, equations (2.2.2)
- (2.2.3) become
U1'(~'r) = 0, -00 < ~ < 00, r> 0, (2.2.6)

u(x(~, r), r = 0) = g(~). (2.2.7)
Hence u is a function of only ~. By the initial condition, we find the solution
u = u(~) = g(~) = g(x - at). (2.2.8)
So, the solution has been found. Such a method (that projects the PDE
onto its characteristic and converts the PDE problem into an ODE problem)
of finding solutions to hyperbolic conservation laws, is called the characteristic
method. Let us work out more examples by the characteristic method.
Example 1:
ut+au~+bu=/(x,t), -oo<x<oo, t>O, (2.2.9)

u(x, t = 0) = uo(x) (2.2.10)
where both a and b are positive constants, and b f:. 1.

Let
r =t, e =x - at (characteristic) . (2.2.11)
Along the characteristics, equations (2.2.9) -(2.2.10) become
u r + bu = f(e + ar, r), (2.2.12)

u(e, r = 0) = uo(e). (2.2.13)
For a fixed e,
the problem (2.2.12) - (2.2.13) is an initial value problem of
an ODE. We say that the PDE problem becomes an ODE problem along a
characteristic. The solution of the ODE problem is
(2.2.14)
In terms of the original variables x and t, equation (2.2.14) becomes
u(x, t) = uo(x - at)e- bt + lot 1 (x - a(t - (7), (7)) e-b(t-u) du. (2.2.15)
If I(x, t) is a gradually diminishing square wave
(2.2.16)
where X[O,l] is the usual characteristic function defined by
when x E [0,1],
X[O,l](X) = { ~: otherwise.
The upper half plane R x R+ is divided into six sub-domains (see Fig. 2.7).
Then, the solution expressed in terms of elementary functions can be written
as:
u(x, t) = uo(x - at)e- bt + Uregion (2.2.17)

x
Figure 2.7: Regions for solution (2.2.17).
where
0,
in region I: < 0,
z
ae(.,-al)/a ( z:(b-l)/a _ 1)
b-l e ,
in region II: z - at < 0 and 0 < z < 1,
....!Le(z:-at)/a (e(b-l)z:/a _ e(b-l)(z:-at)/a)
b-l '
e-bz:/ a in region III: 0 < x - at < 1 and 0 < x < 1,
Uregion = -a -
....!Le(z:-at)/ a(e(b-l)/ a _ e(b-l)(z:-at)/ a
b-l '
in region IV: 0 < x - at < 1 and z > 1,
0,
in region V: x - at> 1,
....!Le(z:-at)/a (e(b-l)/a - 1)
b-l ,
in region VI: x - at < 0 and x> 1.
Example 2:
Ut + XUz: = 0, -00 < x < 00, t > 0, (2.2.18)

u(x, t = 0) = uo(z). (2.2.19)
Since the coefficient of Uz: is not a constant, the characteristics are no longer
straight lines. The nonconstant coefficients of the unknowns or their derivatives
are due to the nonuniformity of the medium in which the wave is propagating.
The diffraction index of the medium varies from point to point and hence the
medium bends the wave traveling direction. We have already seen from the
Characteristics
~---L-----L----~----------------------x
Figure 2.8: Curved characteristics for conservation laws m a non-uniform

medium.
previous examples that the waves travel along characteristics. The bending of
the wave traveling direction is equivalent to the bending of the characteristics.
The characteristics are defined by the following ODE problem:
dx
dt = x, x(t = 0) = e (2.2.20)
See Fig. 2.8 for the characteristic curves.

e
Here = xe- t are the characteristics. Thus we introduce the following new
variables
T = t, e
= xe- t . (2.2.21)
Problem (2.2.18) and (2.2.19) becomes
Ur = 0, u(x(e, T = 0), T = 0) = uo(e). (2.2.22)
Hence u is only a function of e. This function is found to be

(2.2.23)
If
1
Uo (x) = -1--2 '
+x
then
1
u (x t) = ----=-__=_ (2.2.24)
, 1 + x e-
2 2t
This evolution process is shown in Fig. 2.9.

Next we extend the characteristic method to strictly hyperbolic systems
with constant coefficients. Consider an IVP of n unknowns
Ut +A Ux = F(x, t), u(x, t = 0) = uo(x) (2.2.25)

1
0.75
U(x,th.5
0.25
o
t
Figure 2.9: Evolution of an initial data of Example 2 in a non-uniform medium

according to (2.2.24).
where A is an n x n constant matrix and all its eigenvalues are real and distinct;
u, Uo and FERn; and x and t E R. Then there is a nonsingular matrix P
such that
(2.2.26)
Under the change of variable v = Pu, we have
Vt + A V z = PF(x, t) == F(x, t). (2.2.27)
Thus (2.2.25) is decoupled and the decoupled equations (2.2.27) can be solved
individually for each equation by the previous characteristic method.
Example: Consider
(2.2.28)
u(x, t = 0) = uo(x), v(x, t = 0) = vo(x). (2.2.29)
The matrix A = (i ;) has eigenvalues al = 1, a2 = 3. The correspond-

ing eigenvectors are rl = (1, -1)/-12 and r2 = {1, 1)/-12 respectively. The
similarity transform matrix P is
P=-12 1(1 -1)

1 1 .
Making a transform
equations (2.2.28) - (2.2.29) become
Ut + U., = 0, Vt + 3v., = 0, (2.2.30)

u(x, t = 0) = uo(x), v(x, t = 0) = vo(x). (2.2.31)
Hence
U(x, t) = uo(x - t), v(x, t) = vo(x - 3t). (2.2.32)
Finally,
1 ( uo(x - t) + vo(x - 3t) )

= v'2 -uo(x - t) + vo(x - 3t) .
Thus, the solution of the IVP (2.2.28) - (2.2.29) is
(~ )
_ ~ ( uo(x - t) - vo(x - t) + uo(x - 3t) + vo(x - 3t) ) (2233)
- 2 -uo(x - t) + vo(x - t) + uo(x - 3t) + vo(x - 3t) ..
The reader can easily verify that (2.2.33) indeed solves the IVP (2.2.28) -
(2.2.29).
Exercise: Solve the IVP
1
Ut + u., + v., = 0, Vt + u., - 2v., = 0, (2.2.34)
u(x, 0) = uo(x), v(x, 0) = vo(x). (2.2.35)
2.2.2 Nonlinearity and wave breaking

Consider the conservation law
Pt + q.,(p) = 0, -00 < x < 00, t > O. (2.2.36)
It can also be written as

Pt + c(p)p., = O. (2.2.37)
The quantity c(p) is therefore the characteristic speed. Let us impose an initial
condition on (2.2.36) by
p(x, t = 0) = po(x), -00 < x < 00. (2.2.38)

P(X,t), t > 0
t __________________________ ~ __________ __________ x

~
Figure 2.10: A wave with a smooth initial data breaks at t = tB .
We intend to solve problem (2.2.37) - (2.2.38) by the characteristic method.

Characteristics are defined by
dx
C: dt = c(p), x(t = 0) ={. (2.2.39)
Along a characteristic C,
d: = 0, p(x({, t = 0), 0) = po({). (2.2.40)
Hence p = Po({) equals a constant along C. From (2.2.39), the characteristic

Cis
x = {+ c (po({)) t. (2.2.41 )
Thus far, it seems that the problem (2.2.37) - (2.2.38) has been solved
without much effort. Actually, this is not the case. LIFE IS NEVER EASY.
Let us consider a concrete example:
Ut + U U", = 0, u(x, t = 0) = ae-'" 2

, a> o. (2.2.42)
By equation (2.2.41), the characteristic solution is given by
(2.2.43)
From this, 8p/8t and 8p/8x can be evaluated as follows
8p 2a 2{e- 2e
(2.2.44)
8t 1- 2a{e-et'
8p 2a{e-e 2
(2.2.45)
8x 1- 2a{e-et
We can see that at t = tB = ee /(2a{), the quantities 8p/8t and 8p/8x

become infinity. So, the initial data's profile at the point { breaks at the time
Figure 2.11: Breaking time tB and overlapped characteristics.
tB (see Fig. 2.10). This happens because starting on the initial profile the
point with a higher value of P propagates faster than that with a lower value
(see equation (2.2.39)). When t > tB, the density P becomes triple valued
at some points. This is totally unphysical. Thus, the characteristic solution
(2.2.39) - (2.2.40) is no longer valid after the breaking time tB' Therefore,
expression (2.2.41) does not completely solve problem (2.2.37) - (2.2.38) . It is
only a solution valid for t < tB. What is the solution of (2.2.37) - (2.2.38) after
the breaking time? This question is the basis of the shock-fitting problem. The
reader may now be aware that LIFE IS VERY TOUGH.
Next we compute the breaking time of expression (2.2.41) from the geomet-
rical point of view. From Fig. 2.11, the breaking time is the earliest intersection
time of two neighborhood characteristics. Therefore
x= e+ c(po(e))t (2.2.46)
and
x= e+ oe + c (po(e + oe)) t (2.2.47)
hold simultaneously. Letting oe ~ 0, we have
(2.2.48)
When t > tB, there exists a region in which at any point there will be at
least two characteristics passing through. This region may be considered as a
fold of the (x, t) plane made of three sheets, with different values of P on each
sheet. The boundary of the region is an envelope of characteristics (see Fig.
2.11).
Exam.ple 1:
Pl > 0, X ~ 0
Pt+PPx=O, p(x, t=O)= { P2>0, x<O (2.2.49)
P (X, t)
t =0
I--_P_l_t_= t > L__ p=-I_
------------------------~------------------------- X
Figure 2.12: The solution for PDE (2.2.49) in Example 1 when PI < P2'
Case (i): PI < P2.

Then tB = O. The characteristic solution is only valid outside of the region
x - P2t < 0 and x - PIt> 0 (see Fig. 2.12).
Case (ii): PI > P2
A solution is (see Fig 2.13)
if f > PI,
if P2 < f < PI, (2.2.50)
if f < P2
The solution in the fan P2 < f < PI is called the centered simple wave, and is
also called the rarefaction. Along a ray x = at (P2 < a < pI), the solution P
is constant and equal to a. Hence the fan is a transient region which bridges
the lower and the higher density regions in the direction of wave propagation.
This is in contrast to the overlap of the characteristics in Fig. 2.12.
The reader may have been wondering why the breaking happens. It is be-
cause of the nonlinearity. The fact that nonlinearity causes breaking has been
experimentally observed and theoretically verified for many nonlinear wave phe-
nomena. In contrast to the nonlinearity term causing breaking, the dispersive
terms cause waves to disperse. The balance of the nonlinearity and the disper-
sion may produce a wave of permanent form. Solitons are a typical example of
waves of permanent form and will be studied in detail in Chapter 4.
P (X ,t)
PI
t = 0
x
Figure 2.13: The solution for PDE (2.2.49) in Example 1 when Pl > pa.
2.2.3 Shocks
From the previous subsection we have seen that the characteristic solutions
cease to be valid after the breaking time tB. The mathematically multiple
valued density in a region is unphysical. The only way to establish a solution
after the breaking time tB is to allow discontinuities of p. This discontinuity
is physically observable and called a shock (or shock wave). The shock travels
with respect to its neighborhood media. The traveling shock waves have been
observed in many situations, such as jet booms, shocks induced by cannon
shells, and ocean tides. The book entiled "An Album of Fluid Motion" by
Milton Van Dyke shows many examples of shocks. This is an excellent book
for students to get some feeling about fluid motion.
Consider
Ut +uuo: = 0, (2.2.51)
with continuous data
< 0,
x>
1, x
'(', t = 0) = .0(') = { 1- x, :5 x :5 1, (2.2.52)
0, 1.
The characteristic solution of (2.2.51) - (2.2.52) is
(2.2.53)
In terms of the original variables z and t, we have
r-'
if z < t < 1,
I-t if t::;z::;l,
u= 0 if z> 1 and t < 1, (2.2.54)
1 if z <!! and t > 1,
0 if z< $
and t> 1.
The peak of the initial data is at z = O. Most likely, the earliest breaking
occurs at this peak point. So the breaking time can be computed from (see
(2.2.48))
(2.2.55)
Hence the characteristic solution (2.2.54) is continuous up to t = 1. When

t > tB = 1, a region of overlapped characteristics appears. Here we will fit
a discontinuity line (called shock path) in this region. The Rankine-Hugoniot
condition determines the position of a shock at a given time. Hence, it de-
termines the shock path in the (z, t) plane as well. The Rankine-Hugoniot
=
condition is s[u) [J(u)). Here f(u) =
(1/2)u 2 Hence
1
s[u] = 2" [u 2 ] (2.2.56)
or
(2.2.57)
where Ul and U r are values of u at the left and right sides of the shock path,
respectively. In our situation, Ut = 1, U r = O. Thus, s = 1/2 and the shock
path is
t = 2z-1. (2.2.58)
The complete solution u is shown in Fig. 2.14.
2.2.4 Entropy condition

Even with the Rankine-Hugoniot condition, some characteristic solutions can
still be unphysical. This is because they do not satisfy an additional condition:
the entropy condition. This condition says that in every physical process the
entropy of the system is nondecreasing. This is a fundamental assumption in
thermodynamics.
Consider
< 0,
Ut + UU x = 0, u(x, to) = { ~: x
x> O. (2.2.59)
This problem has at least two solutions:
0, z < 0,
Ul = { 0,
1,
and U2 = { f, 0< x < t, (2.2.60)
1, z > t.
t
t=2x-1
Shock path
u =1
u =1
u=0
u =1
x = t
1-x
U= ___
u=o
1- t
0 1
x
Figure 2.14: Evolution of continuous initial data (2.2.52) according to the

nonlinear conservation law (2.2.51).
The first solution '1,11 is obtained by fitting the fan with a shock (called the
rarefaction shock) and a discontinuous solution that satisfies the Rankine-
Hugoniot condition. The second solution '1,12 is obtained by fitting the fan
with a centered simple wave (a continuous solution!). Of course, the physically
meaningful solution can only be one of the two. So, which one?
Before we answer this question, let us look at another example:
-I x< 0,
Ut + UU x = 0, u(x, t = 0) = { 1,' x> o. (2.2.61)
Then
-I,
-a
ua(x, t) = { ' (2.2.62)
a,
1,
are solutions for every a ~ 1 (see Fig. 2.15). All the jumps satisfy the Rankine-
Hugoniot condition. So we have infinitely many solutions.
Here comes the same question. What value of a should we choose in order
to make the solution unique and physically meaningful?
From (2.2.40) - (2.2.41), we can easily show that for Ut+ fx(u) = 0, u(x, t =
0) = uo(x), we have
(2.2.63)
2x = t 2x = (a -1)t
U =-a
u = -1 U =1
--------------------~&--------------------------x
o
Figure 2.15: Nonuniqueness of solutions to a nonlinear hyperbolic problem.
If u~(e) > 0 and f"(uo) > 0, then

1
(2.2.64)
u'" < /" (uo) t .
Let E = suPuo (J"(uo)' which is independent of x and t. Then
u(x + h, t) - u(x, t) E
(2.2.65)
h $ t'
for any h > 0, t > O. This condition is called the entropy condition. Across the
shock, the nondimensionalized quantity h (that measures the shock thickness)
is much smaller than the jump of the nondimensionalized quantity u (that
measures the strength of the shock). Hence,
Eh
u(x + 0, t) - u(x - 0, t) $ -t- R:j O. (2.2.66)
Therefore, condition (2.2.65) implies that if we fix t > 0 and let x scan from
-00 to +00, then u can only jump down at the discontinuity point of u(x, t).
Such a restriction on the solutions of hyperbolic conservation laws is called the
entropy condition.
Apparently the rarefaction shock Ul of (2.2.60) does not satisfy the entropy
condition (2.2.65) since it jumps up when it crosses the shock: xlt = 1/2. Also
it is clear that none of the solutions U of (2.2.62) satisfy the entropy condition.
Q
Hence they are all unphysical. Actually the fan region is a rarefaction. It can
only admit simple waves but not shocks. In contrast, the solution represented
by Fig. 2.14 satisfies the entropy condition. Hence that shock is physical.
In summary, to have a shock which is physical, both the Rankine-Hugoniot
condition and the entropy condition must be satisfied. The Rankine-Hugoniot
condition determines the shock positions (hence, the shock paths), while the
entropy condition checks which shock is ,physical. In many situations, it is this
entropy condition that makes the solution to an IVP for a hyperbolic PDE
unique. In practical applications, it is clear whether a shock or a rarefaction
occurs. For example, a supersonic piston of a compressible gas can generate a
shock before the piston and a rarefaction after the piston. Everybody can see
that the shock and the rarefaction can not be reversed. This is why many engi-
neers never mention the entropy condition when they construct shock solutions,
for they KNOW what makes sense!
Exercise: If Ut + f:c(u) = 0, J"(u) > 0, U E Rand s is the shock speed,
then
(2.2.67)
Prove this claim.
2.2.5 Shock structure

Students may still find that fitting a shock in the region of overlapped char-
acteristics is mathematically unnatural, even though the physical meaning of
shock solutions is clear. In this section, we will present a more plausible and
natural approach to construct such discontinuous solutions.
We add a viscosity I/p:c:c term to the hyperbolic conservation law
(2.2.68)
So we have a new equation
Pt + c(p)P:c = 1/ P:c:c, c(p) = q'(p). (2.2.69)
It is well known that viscosity can smear out some mathematical singular-
ities. In our current situation, the singularity is the nonsmoothness. This is
because equation (2.2.69) has a structure like the heat equation and the heat
transfer in a continuous medium is gradual and smooth. Hence we expect to
obtain a smooth solution of (2.2.69) even with discontinuous initial data. On
the other hand, the data can evolve into a shock according (2.2.68). Since
equation (2.2.69) becomes equation (2.2.68) when the viscosity vanishes, we
naturally expect that the smooth solution of (2.2.69) will approach a shock
wave as 1/ -t O. The smooth solution of (2.2.69) is called the viscosity solution.
Let us look for a traveling wave solution of (2.2.69) with the initial data
p(x,t = 0) = { PI, if x> 0, (2.2.70)

P2, if x< O.
Assume that
e= x - st, (2.2.71)
Then (2.2.69) becomes
[c(p) - s]p' = I/P". (2.2.72)
P (x t)
t
P2 P2
l.
P1 P1
t = 0 ~ t> 0
I
f x
Shock thickness
Figure 2.16: Viscosity and shock structure.
The first integration of this equation gives

q(p) - sp+ G = vp' (2.2.73)
where G is an arbitrary constant. We expect our solution to be the same as
the shock when Ixi is large, i.e.,
lim p(e) = { P1, (2.2.74)

(-doo P2
See Fig. 2.16 for the profile of p. Hence
P'(oo) = O. (2.2.75)
Thus the limits of (2.2.73) as eapproaches -00 and +00 respectively give
(2.2.76)
i.e.
(2.2.77)
This is exactly the Rankine-Hugoniot condition. The above shows again that
the Rankine-Hugoniot condition gives nothing but the velocity of the traveling
wave of permanent form (see Fig. 2.16).
Next we will examine the thickness IX2 - xli of the transition region from
P2 to Pl. Here Xl and X2 are chosen to fit a measure scale of the shock, and
IX2 - xli is called the thickness of the shock. For instance, one may choose
Ip(6) - p(6)1 = 0.81p2 - P11
or
Ip (et) - P (6)1 = 0.951p2 - pd .
Then the thickness of the shock can be found by integrating (2.2.73) from p (6)
to p (6) as follows
d :1:2 - :1:1 r({t) dp
(2.2.78)
;= v = J (6)
p q(p) - q (pt) - pP2-P;1 (q (P2) - q (pt))'
Here we have used 6 - 6 = Xl - X2. Thus, the thickness of the shock is

proportional to the viscosity II. The transition region becomes shorter and
shorter as the viscosity becomes smaller and smaller so that finally a shock
wave is obtained.
This is a way, which is totally different from the characteristic method, to
approach a shock solution. However, we have seen that to approach a shock
wave in this way we need to solve (2.2.69) . This is not easy. Nevertheless, for
the special case, c(p) = p, equation (2.2.69) can be solved analytically. This
special equation is called the Burgers equation. The method to solve it is called
the Cole-Hopf transformation. We will study this in Chapter 5.
[1] J. Smoller (1983), Shock Waves and Reaction-Diffusion Equations, Springer-

Verlag, New York, Chapters 15 - 18.
[2] R. E. Meyer (1982), Introduction to Mathematical Fluid Dynamics, Dover,
New York, Chapters 1 - 3, and Chapter 6.
[3] G. B. Whitham (1974), Linear and Nonlinear Waves, John Wiley, New
York, Part I.
[4] J. Marsden and T. J. R. Hughs (1983), Mathematical Foundation of Elas-
ticity, Prentice-Hall, New York, Chapter 2.
[5] W. F. Lucas (1978), Models in Applied Mathematics, Springer-Verlag, New
York, Vol. 1.
[6] M. Van Dyke (1982), An Album of Fluid Motion, The Parabolic Press,
Stanford, California.
Chapter 3
Water Waves
Many of the general ideas about dispersive waves originated in the problems of
water waves. This is a fascinating subject because the phenomena are familiar
and the mathematical problems are various.
- - - G. B. Whitham
3.1 Governing Equations for Water Waves
3.1.1 Euler equations
In this chapter, water is assumed to be an inviscid, incompressible fluid. Body

force is the earth's gravity. From section 2.1.2, the conservation of mass and
53
54 Chapter 3. Water Waves
momentum yields
Ux + Vy + W z = 0, (3.1.1)
1
Ut + UUx + vUy + WU z = -- Px, (3.1.2)
P
1
Vt + UVx + VVy + WVz = -- Py, (3.1.3)
P
1
Wt + UWx + VWy + WW z = -- pz
p
- g. (3.1.4)
These equations are valid in the fluid domain and are called the Euler equations.
=
Here (u, v, w) u is the velocity field, p is the density, P denotes pressure, and
9 is the gravitational acceleration (see Fig. 3.1).
Boundary conditions vary from problem to problem. For surface waves in
an open ocean (without shorelines), there are two boundaries. One is on the
ocean bottom, which is assumed to be rigid. The other is on the free surface of
the water, which is to be determined. On the free surface two conditions need
to be specified. The first one is the dynamical condition, reflecting the external
actions on the free surface. The other one is the geometrical condition, which
states that the water particles on the free surface should always stay on the
free surface. At the bottom, since there is no other fluid interacting with the
water, the only condition which applies is the geometrical one showing that it
is impossible for water to penetrate the rigid bottom.
Let z = 1/(:C, y, t) and z =-h(:c, y) be the free surface and the bottom
topography respectively. Then, on the free surface z = 1/(:C, y, t), we have
P = p(:c, y, t) (dynamical) , (3.1.5)

1/t + U1/x + v1/y = W (geometrical). (3.1.6)
And on the bottom z = -h(:c, y), we have
uh x + vhy + W = 0 (geometrical). (3.1.7)
In (3.1.5), p is a given surface pressure disturbance and the geometrical bound-

ary condition (3.1.6) was derived earlier in section 2.2. Therefore, the water
wave problem in an open ocean is a free surface problem defined by (3.1.1) -
(3.1.7). Solving such a wave motion problem is a challenging task.
3.1.2 Potential flow

The potential flow is a flow whose velocity field is irrotational, i.e.,
w=V'xu=o
where w is called the vorticity. From (3.1.1) - (3.1.4), we can derive an equation
(3.1.8)
3.1. Governing Equations for Water Waves 55
o~ __~__~_________
~g z = -h(x.y)
x
Bottom topography
Figure 3.1: Surface water waves in an open ocean
This equation is called the vorticity equation.

Exercise: Derive (3.1.8) from (3.1.1) - (3.1.4).
Equation (3.1.8) has a solution w = O. Hence, W = 0 is a solution to the
Euler equations. Thus, irrotationality is a mathematically realistic assumption.
Thus, it is claimed that an irrotational flow does exist in the Euler equation
sense. Specifically, a uniform flow in a channel is irrotational. Of course, we
always bear in mind that mathematical models (such as the Euler equations)
and laboratory observed phenomena (such as the water waves) are only mutual
approximations under certain conditions. For instance, any flow in the real
world has a viscosity which is not zero and a density which is not uniform. The
viscosity and the nonuniform density (the later is called the stratification of a
jluid) , always cause rotation of fluid elements. So, abosulte irrotational flows
can never be found in nature. But in many cases, the vorticity is so weak that
the flow can be considered to be approximately irrotational.
If 'V x u = 0, then there exists a potential <jJ such that
u='V<jJ.
In this situation, the equations (3.1.1) and (3.1.8) may be considered as the
governing equations since they are dependent on (3.1.1) and (3.1.2) - (3.1.4).
Equation (3.1.8) is satisfied if w = O. Hence the only equation which needs to
be satisfied is (3.1.1). This is the case when <jJ is a harmonic function, i.e.
(3.1.9)
Using
'V(u u) = 2(u . 'V)u + 2u x ('V xu),
equation
1
Ut + (u V)u = --Vp- gk
P
(i.e. equations (3.1.2) - (3.1.4)) can be written as
(3.1.10)
Let
(3.1.11)
This H is called the head of the flow. If w = =

V x u 0, there exists a scalar
potential function ~ such that u = V~ . Consequently, equation (3.1.10) gives
Hence,
(3.1.12)
where C(t) is the arbitrary constant from spatial integration and is a function
oft. Equation (3.1.12) is called the Bernoulli equation, which is very useful in
hydrodynamics.
Now we are ready to formulate the boundary conditions for the Laplace
equation (3.1.9) for an open ocean. On the free surface z = 71(x, y, t), we have
~t + JP dp
p
+ !IV~12
2
+ 971 = C(t) (dynamical condition) ,(3.1.13)
71t + ~:c71:c + ~y71y = ~z (kinematic condition). (3.1.14)
On the bottom z = -h(x, y), we have
~:ch:c + ~yhy + ~z = 0 (kinematic condition). (3.1.15)
The function C(t) in (3.1.13) can be determined by a known condition at a

point on the free surface, such as the point at infinity.
Example 1: If h(x, y) = -1, and p = 1 , then ~ = Va y, 71 = 0, C(t) = Va2 /2
is a solution of the problem (3.1.9), and (3.1.13) - (3.1.15). Physically, this
solution represents a uniform flow with velocity Va along the y-axis.
Example 2: There is right cylindrical water container which holds water of
depth H. At the bottom of the container there is a small hole through which
the water flows out. The area of the small hole is S which is much smaller than
the base area A of the right cylindrical container. We may use the Bernoulli
theorem to compute the total time T needed for all the water to drain out (see
Fig. 3.2).
By the Bernoulli theorem, we can first evaluate the flow speed at the drain-
ing hole. Since the depth decrease is very slow, the flow may be considered
3.1. Governing Equations for Water Waves 57
A Air pressure Pa
u
Air pressure s
li
Figure 3.2: Application of the Bernoulli theorem to the draining water problem.
approximately stationary. We also regard the water density as a constant.

Then the head on the free surface of the water is equal to the head at the hole:
1 2 1 2
-Pap + -u
2
+ gH = Pa
-
p
+ -v
2
.
In the above, we have chosen the bottom of the container as the zero point
for the vertical coordinate. On both the free surface and the hole, the water
pressure should be equal to the air pressure Pa. The water flowing speed at the
hole is v and the free surface is moving down at a speed equal to u. Both u and
v are unknowns. We need an additional equation to solve the above problem.
This is the conservation of flux:
uA = vS.
Solving the above coupled equations, we obtain
S
v= and u=-
A
Since S ,< A, we have an approximation

v ~ V2gH.
The total time needed to drain the water is denoted by T. The conservation
of mass leads to
-Adh = Svdt.
Integrating the above with respect to h from H to zero and with respect to t
from zero to T, we can obtain that
T=J 2 H meg
(3.1.16)
where me = (S2jA2)(1- S2jA2)-1 is considered as an effective gravitational

coefficient. If we consider meg as the effective gravitational acceleration, then
equation (3.1.16) gives the time needed for a free falling particle to reach the
ground from the height H in this effective gravitational field. This is a familiar
result from general physics or elementary calculus.
3.2 Shallow Water Equations

3.2.1 Shallow water equations
Consider the situation that the wave length A (as the horizontal length scale
L) is much larger than the water depth H. Hence
(3.2.1)
Let us consider two-dimensional problems. The x-axis is along the horizontal

direction, and the y-axis is along the vertical direction opposite to the gravi-
tational force. Assume that the free surface is not subject to external forces.
Then from section 3.1., we have the following mathematical setup of the prob-
lem:
U;. + v~. = 0, (3.2.2)

u t* + U **
u x + v **
u y = - -1 Px.,
* (3.2.3)
P
v t* +U* v*x + v **
vy = - -p1*
py - g, (3.2.4)
(irrotational); (3.2.5)
with boundary conditions
P* = 0, 1ft
'fJt. + U *'fJx. =
1ft
v 1ft on y lit = 'fJ 1ft ( *
x, t*) ; (3.2.6)
u*h;. +v* =0 on y* = -h*(x*). (3.2.7)
Here p is a constant, and "* " signifies that the quantities are dimensional. To
nondimensionalize the above equations, we introduce the following dimension-
less quantities:
H t*
(x, y) ( H x* Y*)
L H' H ' t =L JHjg'
u* L v* ) p*
(U, v) ( (3.2.8)
..JiH' H..JiH ' p = pgH'
In the above non-dimensionalization process, the length scale and the time
scale are the most crucial. The condition of the horizontal length scale being
not much greater than the vertical length scale implies that the fluid motion
3.2. Shallow Water Equations 59
is quite violent. The vertical acceleration is important. In order to observe

the relatively fast event, the time scale has to be short. The relative size of
the vertical velocity with respect to the horizontal velocity is determined by
the conservation of mass when the length scales are prescribed. The mass
conservation equation suggests a relation
U V
+=0
where U and V are scales of the horizontal and vertical velocities respectively.
Therefore, we have the relative size
V H
U = L
We carry out this process of nondimensionalizing the physical quantities in
the above way because we would like to bring every dimensionless quantity to
the size 0(1). The nondimensionalized form of (3.2.2) - (3.2.7) is
U:.: + Vy = 0, (3.2.9)
Ut + UU:.: + VUy = -P:.:, (3.2.10)
f (Vt + UV:.: + vVy ) = -Py - 1, (3.2.11)
(3.2.12)
on the free surface y = 1](x, t) :

P = 0, 1]t + U1]:.: - V = 0; (3.2.13)
and on the bottom y = -h(x) :
uh~ + v = o. (3.2.14)
Assume that the problem (3.2.9) - (3.2.14) has an asymptotic solution of

the form
(3.2.15)
Substituting (3.2.15) into (3.2.9) - (3.2.14) and assembling the terms of like
powers of f gives
Order fO :
+ VOy = 0,
UO~ (3.2.16)
-UOt + UOUo~ + vouoy = -Po~, (3.2.17)
POy = -1, (3.2.18)
uO y = 0; (3.2.19)
on y = 1]0: Po = 0, + uo1]o~ = Vo;

1]Ot (3.2.20)
and on y = -h(x) : uoh~ + Vo = o. (3.2.21)
In the above, the boundary conditions on the free surface are approximated by
a Taylor expansion of (3.2.11) about y = 1]0.
From equations (3.2.18) and (3.2.20),
Po = -y + 1]0 (3.2.22)
From equation (3.2.19), we find that uo is independent of y and hence a function
of only z and t:
Uo = I(z, t). (3.2.23)
This equation clearly shows that the zeroth order horizontal velocity does not
have a vertical structure. Intuitively, this is generally false when the moving
water is very deep. Hence, equation (3.2.23) is a striking manifestation of the
shallow water flow.
Integrating (3.2.16) with respect to y and using the bottom boundary con-
dition (3.2.21), we have
Vo = - Iz y - (fh)z. (3.2.24)
Then equations (3.2.17) and (3.2.20) become
It + I Iz + 1]oz = 0,
1]Ot + 11]oz + Iz1]o + (fh)z = O.
These two equations can be written as
1]Ot + [/(1]0 + h) L= 0, (3.2.25)

It + I Iz + 1]oz = O. (3.2.26)
Equations (3.2.25) - (3.2.26) are called the shallow water equations. These are
the equations that the zeroth order asymptotic approximate solution (3.2.15)
should satisfy. Equations (3.2.25)-(3.2.26) are in dimensionless variables. The
dimensional form of these two equations (3.2.29)-(3.2.30) is going to be shown
later.
If the asymptotic assumption (3.2.15) holds, it is clear that when h= con-
stant (flat bottom), then
I/(z, t) - ~ [ : u(z, y, t)dyl = O(f) (3.2.27)
where u(z, y, t) is from the exact solution of (3.2.9) - (3.2.14). Hence, I(z, t) is
approximately the average horizontal velocity.
However, the mathematical rigor of the asymptotic assumption (3.2.15)
requires a proof.
Exercise: Suppose that h = constant and u(z, y, t) satisfies the problem
(3.2.9) - (3.2.14). Show that if 1]0 and I solve the problem (3.2.25) -(3.2.26),
then equation (3.2.27) holds.
A simpler way to derive the shallow water equations is from the physical
point of view. The physics assumption for shallow water flow is that there is
no vertical acceleration. Since, in an inviscid fluid, the driving force for the
acceleration in the vertical direction consists of only the pressure gradient and
the gravity, the vanishing vertical acceleration of the fluid implies that the
pressure gradient must be balanced by the gravity. Thus, the pressure is equal
to the hydrostatic pressure. Namely,
p* = pg(r/* - y*). (3.2.28)
Under the shallow water assumption, the continuity equation can be re-
placed by the conservation of mass flux
11;. + [u*(h* + l1*)t. = o. (3.2.29)
In the Euler equations, there are two equations for the conservation of mo-
mentum. One is for the horizontal direction and the other one for the vertical
direction. Since under the shallow water assumption the vertical acceleration
is zero, only the conservation of the horizontal momentum plays a role. The
conservation of the horizontal momentum is
u;. + u*u;. = -p; ..

By equation (3.2.28), we have
u;. + u* u;. + 911;. = o. (3.2.30)
Equations (3.2.29) and (3.2.30) are the dimensional form of the shallow water
equations (3.2.25) and (3.2.26). The nondimensionalization rule is, of course,
still according to (3.2.8).
From the above analysis, we have seen that the physical argument is sim-
ple and straightforward. In contrast, the mathematical analysis is tedious but
shows explicitly the size of the terms thrown out in the approximation process.
It seems that the physical analysis and the mathematical analysis are compli-
mentary to each other and we should hesitate to suggest that one analysis be
superior to the other.
If h is a constant, then equations (3.2.25) - (3.2.26) can be written into a
conservation law
(3.2.31 )
The eigenvalues of the coefficient matrix are
'\1,2 = f V110 + h. (3.2.32)
Hence, equation (3.2.31) is a strictly hyperbolic system. The wave propagation

speeds relative to the mean flow f are the differences between the characteristic
speeds and the mean flow velocity. Namely,
ut = '\1,2 - f = V110 + h (3.2.33)

Wave moving direction
~ Front
Figure 3.3: Wave breaking on a beach of uniform slope
and uf are called the the propagation speeds of shallow water waves. The
"+" and "-" signs indicate that a disturbance in a shallow water propagates
in both directions (upstream and downstream). This shallow water wave speed
can be observed very easily by slightly disturbing a shallow water pond. The
small wave propagates in the rest water at the shallow water speed .J9 H.
Shallow water equations are particularly interesting when h is not a con-
stant. Offshore engineers use them to study water motions on beaches, such as
the well known running-up problem. In addition, the shallow water equations
are of rich mathematical structure and applicable to many fields of sciences and
engineering. Actually, the gas dynamics equations have the same mathematical
structure as the shallow water equations (3.2.25) - (3.2.26). Therefore, it is not
surprising for shallow water equations to have discontinuous solutions, which
are known as the undular bores. As we discussed in Chapter 2, discontinuous
solutions of gas dynamics equations exist and are called the shocks.
3.2.2 Wave breaking on a beach

Consider a surface wave of smooth profile moving toward the shoreline of a
uniformly sloping beach (see Fig. 3.3). Our common sense suggests that this
profile would soon or later become non-smooth and break on the beach.
The shallow water equations
71t+ [u(71 + h)]z = 0, (3.2.34)

Ut + U U z + 71z = O. (3.2.35)
are used as the governing equations. The x-axis is chosen to be the undisturbed
= =
free surface and x i is the fixed shoreline. The wave front x s(t) is defined
by u(x, t) =
0, 71(x, t) =
0 when x > s(t). The following assumptions are
taken:
(a) U and 1] are continuous functions;
(b) The first and the second partial derivatives of U and 71 suffer at most jump
discontinuities;
(c) u(x, t) = 0, 7](x, t) = 0 if x> s(t).

Assumption (a) is reasonable since we are considering the wave motion be-
fore the breaking of the surface. An undular bore (a discontinuity) can form
only after the breaking. Assumption (c) is consistent with the statement that
"x = i is a fixed shoreline." This is more or less true since waves break long be-
fore they reach the initial position of the shoreline. Let f- = limx_u(t)_O f(x)
denote the value of the quantity f right behind the wave front. Then, assump-
tions (a) and (c) yield (du/dt)- = (d7]/dt)- = 0, i.e.,
(3.2.36)
where
ds
c=- (3.2.37)
dt
is the velocity of the front. Just behind the wave front (i.e., s - x = 0+),
equations (3.2.34) - (3.2.35) become
{7]t}- = -h (u x )-, {ut}- = - (7]x)- . (3.2.38)
If we assume that (7]x) - =P 0, then (3.2.36) and (3.2.38) yield
c = ~; = v'h. (3.2.39)
This is exactly the propagation speed of shallow water waves as we wanted to

find.
We are mainly interested in the breaking, i.e., where and when (7]x)- be-
comes infinity. Let
a = a{x) = (7]x)- . (3.2.40)
From (3.2.34) - (3.2.35) and the above derived results, we can have
c2 ( Uxx )- - ( 2hx
Utt )- + -
2
a + -3a = O. (3.2.41)
c c
Exercise: Derive equation (3.2.41).
By
ddx (ux)- = (u xx )- + (uxt} - ::'
:x (ut}- = (Utx)- + (Utt)- ::'

we have
c2 d~ (u x)- - c d~ (ut}- = c2 (u xx )- - (Utt)- . (3.2.42)
From (3.2.38) - (3.2.42), one can derive
da 3h x 3 2
dx + 4h a + 2h a = O. (3.2.43)
Exercise: Derive equation (3.2.43).

To integrate equation (3.2.43), let a = lib. Then
db _ 3h x b _ ~ = 0 (3.2.44)
dx 4h 2h .
This first order ODE can be solved by variation of constants. Finally,
b(x) = [h(O) + I(x)] (~~~~) 3/2 (3.2.45)
where
(3.2.46)
e
Whether a wave will break depends on whether b(e) = 0 for some less than
or equal to i. There are two cases: a depressive wave (a(O) = (7]x)-(O) > 0) and
an elevation wave (a(O) = (7]x)-(O) < 0). Elevation waves always break. But,
e
depression waves will break if and only if I(i) < 00 and = i. If la(O)II(i) > i,
e
then < i, and the wave breaks before the shoreline. If la(O)II(i) ~ 0, then
e = i, and the wave breaks at the shoreline.
For more details on wave breaking on beaches, see Refs. [3] and [4] listed
as additional reading materials at the end of this chapter.
3.3 Dispersive Water Waves

3.3.1 Dispersive waves
If waves of different wave lengths propagate at different speeds, then we say that
the waves are dispersive. Whether propagated waves can be called dispersive
waves depends on not only the wave maker but also the media in which the
waves propagate. We can send dispersive waves through most materials in
nature. Water is a dispersive medium. The following simple experiment can be
done by everybody to demonstrate the dispersion behavior of the water waves
of small amplitude.
Stand near a pond and get two stones in your hands, one of which is much
larger than the other. Throw the smaller one into the pond. The stone gen-
erates water ripples, whose wave lengths are relatively short and which propa-
gate at relatively low speeds. Then throw the larger piece into the pond. The
larger stone generates water ripples whose wave lengths are relatively longer
and which propagate at relatively higher speeds. In summary, the results of
this experiment imply that the water waves with longer wave lengths propagate
faster than those with shorter wave lengths.
To see the dispersive property of waves mathematically, let us look at the
following three examples.
Example 1. Consider the Klein-Gordon equation
tPtt - tPxx + tP = o. (3.3.1)

3.3. Dispersive Water Waves 65
Assume this Klein-Gordon equation has a wave solution of the form
= acos(kx -wt) (3.3.2)
where w is called the frequency, k is called the wave number and A = 11k is
called the wave length. Substituting (3.3.2) into (3.3.1) we have
( - w2 + k 2 + 1) cos(kx - wt) = O.
Hence, expression (3.3.2) is a nontrivial solution of equation (3.3.1) if and only

if
(3.3.3)
This equation is called a dispersion relation (or eikonal equation in optics,
or Hamilton-Jacobi equation in the Hamiltonian mechanics). The observable
velocity of a wave of frequency wand wave length A is WA (or wi k) and is called
the phase velocity. Hence, the phase velocity of the wave is
(3.3.4)
Therefore waves of different wave lengths propagate at different speeds. The

larger the wave length, the higher the propagation speed. Furthermore, the
dispersive wave (3.3.2) with a fixed wave length propagates in two directions
since the phase velocity can be positive or negative according to (3.3.4).
Example 2. Consider the linear Korteweg-de Vries equation
(3.3.5)
Let
= a cos(kx - wt). (3.3.6)
Inserting (3.3.6) in (3.3.5), we can obtain a dispersion relation
3 1 3
W = -k - -k . (3.3.7)
2 6
Then the phase velocity is
(3.3.8)
Thus the wave (3.3.6) is dispersive and the wave with a fixed wave length
propagates only in one direction since there is only one given sign for a given
k.
Example 3. Consider the cubic nonlinear Schrodinger equation
(3.3.9)
where i = V-I is the imaginary unit and u(x, t) is a complex valued function.
~-....~, 11" Free surfac___e-___
9
H Domain of water
+
Flat bottom
Figure 3.4: Two-imensional shallow water waves in an open ocean of flat bot-
tom.
Let us look for a solution of the form
u = a exp[i(kx + wt)], a = constant (may be complex valued). (3.3.10)

Substituting (3.3.10) into equation (3.3.9), we have the following dispersion
relation
(3.3.11)
This dispersion relation depends on the amplitude of the solution, and so the
phase velocity also depends on the solution amplitude. The phase velocity is
~ = lal 2 _ k (3.3.12)
k k .
From this expression, we can see that a wave of very long wave length has a
phase velocity which is proportional to the amplitude of the solution. But a
wave of very short wave length has a phase velocity which is inversely propo-
tional to the amplitude of the solution. This property should not be due to
the wave maker. Instead, this property has something to do with the nature of
the medium in which the wave propagates. We will discuss the cubic nonlinear
Schrodinger equation and its solution properties further in Chapter 7.
3.3.2 Boussinesq equations and the KdV equation

Consider tlie long wave motion of water with flat bottom (see Fig. 3.4). Let L
be the horizontal length scale which characterizes the typical wave length. Let
H be the vertical scale which is the upstream mean depth. Assume the motion
is irrotational. Then we have the following mathematical problem:
.6. * * = 0, 0 < y* < H + 1/*, (3.3.13)
on the free surface y* = H + "1* ,

1/;. + ;. "I;. - ;. = 0, (3.3.14)
1/* + ; + '12 (V'* *) 2 = constant, (3.3.15)

and on the bottom

;. = o. (3.3.16)
We introduce the following dimensionless variables
x* y* ....(iH *
x = L' y = H' t = -y-t ,
TJ* A. -1 *
TJ= H' '1'= Vg HL '
and
= (~) 2 1 (long wave assumption).
The time scale L j V9 H is the time needed for a linear shallow water wave
to travel a distance L. This horizontal length scale may be regarded as a
typical wave length for periodic waves. The horizontal velocity is u* ;. '" =
LVgHjL = ....(iH. The vertical velocity scale is v* = ;. '"
L....(iHjH =
1/2 V9 H. From the above two statements it seems that the vertical velocity is
in the order 1/2 Vg H and the horizontal velocity is in the order VgH. Hence
the vertical velocity is greater than the horizontal velocity. This is against our
physical intuition and is an inappropriate interpretation of the two statements
above. As we discussed in section 3.2 the relative size of the horizontal velocity
with respect to the vertical velocity, i.e. the relationship V* jU* = H j L, should
still hold. Hence, we expect that the leading order term of is independent
of y in order to make the nondimensionalization consistent with the physics.
Consequently, the vertical velocity is v* = ;. '" 3/2....(iH which is consistent
with the relation V* jU* = H j L = 1/2. The nondimensionalized problem is
xx + yy = 0 , 0<y < 1 + TJ; (3.3.17)
on y = 1 + TJ,
(3.3.18)
(3.3.19)
on y = 0,
y = O. (3.3.20)
Let
(3.3.21)
n=O
Substituting (3.3.21) into (3.3.17) and using (3.3.20), we have
hn-1=0, n=I,2,3,,
2
12 = -2/0xx, 14 = 24/0xxxx,,
(_I)"n {)2n
hn = (2n)! {)x2n 10, ....
Let I = 10 and take only quantities of order lower than 0(f2) in (3.3.18) -
(3.3.19). Then
(3.3.22)
(3.3.23)
In the above, let U = lc and differentiate (3.3.23) with respect to x. This leads
to
= 0,
f
TJt + [(1 + fTJ)U]z - 6"uzzz (3.3.24)
f
Ut + f U U z + TJz - 2"uzzt = O. (3.3.25)
These equations are called the Boussinesq equations. Linearizing the above
equations and dropping high order derivatives, we obtain
TJt+ U z = 0, (3.3.26)
Ut + TJz = O. (3.3.27)
Hence
TJtt - TJzz = o. (3.3.28)
This equation admits waves traveling in two directions, TJ(x, t) = TJR(X - t) +
TJdx+t). Furthermore, equations (3.3.26) - (3.3.27) can be satisfied with U = 71.
We may take this result as a hint and assume
(3.3.29)
in (3.3.24) - (3.3.25). Substituting (3.3.29) into (3.3.24) - (3.3.25), we have
6" Azzz = 0,
f f
TJt + TJz + f Az + 2fTJTJz - 6"TJzzz -
f f
TJt + TJz + fAt + fTJTJz - 2"TJzzt - 2"Azzt = O.
If we consider those waves traveling only to the right, then TJt + TJz = 0 is a
constraint on the above two equations. We apply this constraint only in such
a way that 71 TJt =
-71 TJz and fTJzzt =
-fTJzzz' The higher derivatives of the
higher order terms ~Azzt and (f/6)A zzz are omitted. With these adjustments,
to make the two equations consistent, A may take
2
A = _'!L + TJzz
43'
In turn, 71 yields
3f f
TJt + TJz + 2"71 TJz + 6"TJzzz = O. (3.3.30)
Let us take the following transformation of coordinates
t -+ f-1t, x -+ x - (1 - f)t. (3.3.31)

Physically, this transformation puts the observer on a moving reference frame

which moves at a speed equal to (1 - f)..{iH to the right. At the same time,
the time is enlarged by c 1 . It implies that the physical phenomenon observed
by this moving observer is going to be in a state of very fast motion. Equation
(3.3.30) in the new coordinates becomes
(3.3.32)
This is the so called I<orteweg-de Vries equation (KdV in short) in terms of

dimensionless variables. When it is expressed by dimensional quantities, the
KdV becomes
(3.3.33)
From Example 2 of section 3.3.1., we know that the linear KdV equation
of (3.3.30) admits waves traveling only in one direction. So does the nonlinear
KdV. The accuracy of the dimensional KdV (3.3.33) to serve as a model equa-
tion is discussed in an excellent paper by Hammack and Segur (1974). Their
conclusion is that
the I<orteweg-de Vries equation appears to provide an accurate model
for determining the evolution from various set of initial data of grav-
ity waves of moderate amplitude propagating in one direction in a
non-dissipative or slightly dissipative fluid of uniform depth.
There is another derivation of the Korteweg-de Vries equation shown in
Appendix C, which is more systematical and possesses clear physical meanings.
Before we go to the next subsection, let us examine some conservation
quantities. We do our inspection on a little faster moving reference frame than
that defined by (3.3.31) for the KdV (3.3.32). This new frame is arranged in
the following way:
t --t -t, x --t x - 2,Xt. (3.3.34)
Then, the KdV (3.3.32) in this new reference frame becomes
3 1
"It + 'xTJx - 2TJTJx - "6TJxxx = O. (3.3.35)
Recall that in Chapter 2 we called the PDE
Pt +qx = 0, -00 < x,t < 00, q(oo) =0 (3.3.36)
a conservation law because the mass J~oo p(x, t) dx is a constant (i.e., conserved
in time). It has been proved that the KdV (3.3.35) or (3.3.32) has infinitely
many conservation laws. Namely, there are infinitely many quantities associ-
ated with (3.3.35) or (3.3.32) which are conserved during the motion of the
fluid. The reader may see Whitham's book (p. 600) for more details. Not all
the conserved quantities have physical meaning. But, the first few quantities
have clear physical meanings such as mass, momentum, and energy.
It is straightforward to derive the following from the KdV (3.3.35):
(3.3.37)
(3.3.38)
Since
77(00) = 77x(OO) = 77xx(OO) = 77xxx(OO) = 77xxxx(oo) = 0,
i:
we have
= = const
I:
m 77(X, t) dx (conservation of mass), (3.3.39)
M = 772(x, t) dx = const
(conservation of momentum). (3.3.40)
Exercise: Derive the conservation law like (3.3.37) and (3.3.38) for the
mechanical energy for the KdV (3.3.35) equation.
3.3.3 Solutions to Korteweg-de Vries equations

Consider the KdV equation
Ut + AUx + 2auu x + pUxxx = 0, -00 < x, t < 00 (3.3.41)
where A, a and p are real constants. We look for its traveling solutions. Let
e=x-ct, u=u(e). (3.3.42)
Then
(A - c)u' + 2auu' + PU"' = O. (3.3.43)
The first integral of (3.3.43) gives
(A - c)u + au 2 + PU" + C = 0 (3.3.44)
where C is the integration constant. Multiplying both sides of (3.3.44) by u'

and integrating it, we can integrate one order further to obtain:
(3.3.45)
where D is the integration constant. So
3 P ,)2 _ 3 3 2 3C 3D _
--(u --U--(A-C)U --u--=P(u). (3.3.46)
2a 2a a a
If f310: > 0, then solutions of (3.3.46) fall into one of the following three cases.
=
Case (i). If P(u) 0 has three distinct real roots, then (3.3.46) has a cnoidal
wave solution. In this case, we write (3.3.46) in the form
~~
20:
(u,)2 = (r1 - u)(u - r2)(u - r3) , (3.3.47)
Let v = u - r2, (= ((20:)/(9f3))1/2 e, then it can be written as

a1 (dd~ )2 = v (81 - v) (v - 81 + 82) , 0 < 81 < 82 (3.3.48)
where
(3.3.49)
The solution of the differential equation (3.3.48) can be expressed by a
Jacobi elliptic function
2[3i;
V=81 cn V4' (3.3.50)
This is a periodic function of ,. Its wave length is

4
A= ~K(m) (3.3.51)
y382
where m = /'i < 1 is called the modulus of the Jacobian elliptic function,
and K (m) is the complete elliptic integral of the first kind
(3.3.52)
Example: In equation (3.3.46), suppose that
0: = -9, f3 = -2, A- C = 6, C = -6, D = O.

Then, equation (3.3.46) becomes
a1 (u') 2
=u(2-u)(u-2+3). (3.3.53)
So
(3.3.54)
Its wave length is
A= ~K ( fa) ~ 2.7. (3.3.55)
The graph of the function (3.3.54) is shown in Fig. 3.5.

v
1.
Figure 3.5: Cnoidal waves
Case (ii). If P(u) = 0 has a real double root, ro, and ro is smaller than the
third real root r, then (3.3.46) has a solitary wave solution. What we mean by
a solitary wave solution of (3.3.46) is that
lim u(x)
1:r:I~oo
=0 (or denoted by u(oo) = 0).
In this situation, rl = r, r2 = r3 = ro, 81 = r - ro > 0 and 82 = r - ro = 81,
and equation (3.3.47) becomes
2
1 dv ) 2
'3 (
d( = V (81 - v) . (3.3.56)
This equation can be directly integrated by using techniques in elementary

calculus:
(3.3.57)
(see Fig. 3.6).

Example: In (3.3.46), assume
1 3
f3 = --,
6
Q' = --, A -
4
C < 0, C = D = O.
Then
-1 (U ') 2 = -u3 - 2 ( A - C) u2 (3.3.58)
3
and
u = 2 (c - A) sech 2 J~(C - A)(. (3.3.59)
This example comes from a fluid mechanical model of near critical flow of water
in a two-dimensional channel.
o
Figure 3.6: Solitary wave solution of the KdV.
P(u)
Solitary
/ wave Unbounded
solution
wave
region
u
Figure 3.7: Regions of unbounded, cnoidal and solitary waves.

Case (iii). If P( u) = 0 has complex roots (There must be two complex roots.
Why?), or if P(u) = 0 has a double real root which is larger than the third real
root, then (3.3.46) does not have bounded solutions.
The above classification is shown in Fig. 3.7.

[1.] J. J. Stoker (1957), Water Waves: the Mathematical Theory with appli-
cations, Interscience, New York.
[2.] G. B. Whitham (1974), Linear and Nonlinear Waves, John Wiley, New
York, Chapter 13.
[3.] M. E. Gurtin (1975), On the breaking of water waves on a sloping beach
of arbitrary shape, Q. Appl. Math. 33, 187-189.
[4.] M. C. Shen and R. E. Meyer (1963), Climb of a bore on a beach, J. Fluid
Mech. 16, 113-126.
[5.] J. L. Hammack and H. Segur (1974), The Korteweg-de Vries equation
and water waves. Part 2. Comparison with experiments, J. Fluid Mech.
65, 289-314.
Chapter 4
Scattering and Inverse

Scattering
For a given potential, the scattering method has been commonly used to find
the wave functions in quantum mechanics. An inverse process of this scattering
is to find the potential from known scattering data. Such a process is called
the inverse scattering method. If the potential satisfies a nonlinear evolution
equation (the differential equation Ut = E[u], where E is a nonlinear time
independent operator), sometimes there exists a linear operator whose potential
is u(x, t) such that the spectrum of the linear operator is independent of time
t. Hence the inverse scattering method can generate solutions to the nonlinear
evolution equation by solving linear problems. This remarkable method that
solves nonlinear evolution equations was invented by Kruskal, Greene, Gardner
and Miura (1967), and it was first applied to find soliton solutions of an initial
value problem for the Korteweg-de Vries equation. Later it was applied to
75
76 Chapter 4. Scattering and Inverse Scattering
other nonlinear evolution equations, such as the cubic-nonlinear Schrodinger

equation, the sine-Gordon equation, the Ginzburg-Landau equation, and the
Yang-Mills equations, etc. Many historical papers on this aspect can be found
in the book edited by Rebbi and Soli ani (1984) which is a collection of reprints.
The KdV solitons for long water waves conspicuously manifest the balance
of the weak nonlinearity and dispersion. The weak nonlinearity results from
the fact that the wave amplitude is not infinitesimal. In theory, the amplitude
is small (of order t). But, this t can be moderately large (as large as 0.46
in certain cases according to Hammack and Segur (1974) [12]). Therefore, the
waves prescribed by the KdV equation can actually have finite amplitudes. The
dispersion is a property of the water medium in two or three spatial dimensions.
The strength of the dispersion varies with the time and space scales. Under
the scales required for the derivation of the KdV, the strength of the dispersion
and that of the nonlinearity are balanced. Hence, a single hump is possibly
stable and this stable single hump is a soliton. This is an unusual phenomenon
and was first observed by John Scott Russell in 1834.
The further significance of the KdV solitons can be greatly appreciated
when one learns about the Fermi-Pasta-Ulam's problem (1955) concerning the
motion of a chain of N identical particles each of which interacts with its
nearest neighbors (see Dodd et al. (1982) [3, p. 5], Newell (1985) [13, p. 3], or
Drazin and Johnson (1989) [10, p.14]). It is expected that the energy initially
in the lowest mode would eventually be evenly distributed among all the N
modes. However, FPU's numerical results showed that the energy was not
evenly distributed among all the high modes. Instead, more than 90% of the
energy swings back and forth among only a few of the lowest modes. So, energy
can be recollected! Kruskal and Zabusky studied the FPU problem from the
point of view of a continuous system. Instead of using N ODEs as the governing
equations, they obtained a PDE whose first order asymptotic solution yields
the Korteweg-de Vries equation (KdV). By solving the KdV, they confirmed
FPU's discovery and coined the word "soliton". The results were announced in
1965 [14]. It was this famous paper that stimulated an overwhelming number
of studies of the nonlinear evolution equations ( Ut = Eu ) in the 1960's and
1970's.
In this chapter, we first study the spring-string scattering to get some feel-
ings about the scattering process. Then we proceed to the scattering method
applied to the Schrodinger equation in quantum mechanics. The major part of
this chapter deals with the inverse scattering method, but we restrict ourselves
to its application to the Korteweg-de Vries equation. The Lax pair and the
KdV hierarchy are going to be discussed in section 4.3. Baclund transform can
also be used to find soliton solutions of the KdV and is discussed in section
4.4. Section 4.5 gives a concise derivation of the inverse scattering method for
the KdV. Soliton fission in a non-uniform medium is studied at the end of this
chapter.
We emphasize the description of the method of finding solutions to the
Korteweg-de Vries equation and the interpretation of the motion and interac-
tion of solitons. So, sections 4.3, 4.4 and 4.5 can be omitted in the first year
4.1. Scattering Method 77
y
String
a)
Figure 4.1: (a) Mass-string-spring scattering system; (b) forces acting on the
mass m.
graduate course on nonlinear waves.
4.1 Scattering Method

4.1.1 String-spring scattering
Consider a mass-string-spring system shown in Fig. 4.1a. The mass point is at
one end of the spring and is attached to the string at x = o.
=
A harmonic wave Yin ei(k:l>+wt) , kw < 0 is sent along the string from x =
-00. After being scattered by the mass-spring, part of the wave is transmitted
to x = =
+00, denoted by Ytr T(t)ei(b+wt). The other part is reflected back
to x = -00, denoted by Yre = R(t)ei(-b+wt). Here T and R are called the
transmission coefficient and the reflection coefficient respectively.
Let Y> and Y< denote the displacement of the string for x 2: 0 and x < 0
respectively. Then
Y< = Yin + Yre = (eik:l> + Re-ib)eiwt, (4.1.1)

Y> -- Ytr -- Tei(k:l>+wt) . (4.1.2)
The reflection coefficient R and the transmission coefficient T can be deter-

mined by the dynamical condition at the attachment point x = O. The string
displacement is continuous at x = 0 and equal to that of the mass point. How-
ever, the slope of the string mass suffers ajump discontinuity (see Fig. 4.1(b)).
From Fig. 4.1(b), the linearized equation of motion of the mass point is
m d2y = -Ky+T$ (a y>

dt 2 ax I - ay<
:1>=0ax I ):1>=0
(4.1.3)
where m is the mass of the mass point, J< the spring constant, and T. the
tension of the string. Please notice that this equation is valid only for small
amplitudes of y since the nonlinearity is ignored.
Assume that (4.1.3) has a solution of the form
y = Ye iwt . (4.1.4)
Then y = y< Ix=o = y> Ix=o implies
Y = 1+ R= T. (4.1.5)
By (4.1.1) - (4.1.2) and (4.1.4)' equation (4.1.3) becomes
- mYw 2 = -J<Y + T.[ikT - ik(l - R)]. (4.1.6)
This equation together with (4.1.5) implies

-2ikT.
T (4.1.7)
J< - mw 2 - 2ikT. '
mw 2 - J<
R = (4.1.8)
J< - mw 2 - 2ikT.
Therefore, the motion of the whole mass-string-spring system is determined.
The above procedure is similar to that presented in Lamb's book (1980) [2].
We now summarize this procedure in determining the motion of a mass-string-
spring system. The entire motion is caused by an incident wave. We then
assume that part of this wave is transmitted and the other part is reflected
by the mass-spring. The mathematical expressions of this motion are given by
(4.1.1) - (4.1.2). These waves are determined by R, T, k and w which are called
the scattering data. The spring constant J< is called the scattering potential.
With given scattering potential J<, the dynamical condition at the mass point
determines the scattering data, hence the entire motion of the system.
The above procedure, using the superposition of the transmission and re-
flection waves to determine the motion of a system caused by an incident wave,
is called the scattering method. This method is particularly useful in quan-
tum mechanics. It can also be used to solve acoustic problems. However, it
applies to only linear systems since the method generically depends on the
superposition principle, which is invalid for nonlinear systems.
4.1.2 Linear Schrodinger equation.

About 350 years ago, Decartes established the coordinate system which con-
nected geometry and algebra and made it possible to describe a physical motion
in terms of mathematical formulas. Since then, it has become possible to be
a scientist without being a philosopher and a believer in God. Decartes also
invented the concept of mass point, an abstract mass that does not have di-
mensions. The position of a mass point in principle can be exactly determined.
Newton, Laplace, Lagrange, Hamilton, and Poisson, etc., all used these con-
cepts and developed the classical mechanics.
4.1. Scattering MethDd 79
Based upDn the abDve, .one can imagine that the existence of matter should
depend neither on .observers nor on the observation method. Thus, the motion
of matter is deterministic in principle and is the unique solution of an initial-
boundary value problem of a differential-integral equation. Indeed, the classical
mechanics in this way successfully explained almost all mechanical motions of
macro-mass-elements. Even the orbits of satellites have been computed from
the classical mechanics. Nevertheless, the classical mechanics cannot explain
many phenomena of microparticles' motion. At the beginning of this century,
Planck, Einstein, Heisenberg and de Broglie found the duality of microparti-
cles. The most well known duality is that of light and is demonstrated in the
following. In a photoelectric process, the light behaves like particles which kick
out some electrons on the surface of a metal and hence produce an electric volt-
age on the metal. These light particles, called the photons, have momentum
but zero mass. On the other hand, when a parallel light beam pass through
a grate, the phenomenon of interference can be observed on a screen behind
the grate. Hence, the light behaves like waves. This property of behaving like
particles in one case and like waves in the other is called the duality of light. In
classical mechanics, particles are particles and waves are waves. Nothing can
be both particles and waves. Existence is independent of observations. Thus,
the duality cannot be admitted in the classical mechanics. This dilemma forced
the birth of a new mechanics: quantum mechanics.
In quantum mechanics, the position of a mass point (called a particle), al-
though without dimensions, cannot be exactly determined in principle. This is
Heisenberg's uncertainty principle. So we no longer talk about the position of a
particle. Instead, we talk about the probability of a particle in a neighborhood
of a certain point in a domain. The corresponding probability distribution
function is the modulus of a complex valued function '1/;, called the wave func-
tion. Therefore the motion of a particle is determined by its wave function.
The profile of a wave function is a wave and this wave travels as the particle
moves.
Let us consider the one dimensional case. The wave function 'I/; of a particle
is a function of x and t. The probability of the particle in the interval [a, b] is
given by
1:
Hence
I'I/; (x , t)12 dx = 1.
In classical mechanics, we have observable quantities x (position), p (mo-

mentum) and H (energy). In quantum mechanics, they are replaced byopera-
tors i:, P (momentum operator) and E (energy operator). The actions of these
operators on the wave function 'I/; are as follows:
x 'I/; , (4.1.9)
= -iii 81/1 (4.1.10)

8x'
iii ! 1/1. (4.1.11)
We define the Hamiltonian operator ii 1/1 as follows:
:m 1/1 + V(i)1/1
A2
H1/1 =
li 2 821/1
- 2m 8x2 + V(x)1/1 (Hamiltonian operator).
Here Ii = 6.626 X 10- 34 [joule sec] (or 10.5 x 1O- 27 [erg . sec]) is the Planck's
constant, i = R the imaginary unit, m the mass of the particle, and V(x)
the potential of the field in which the particle moves.
In classical mechanics, the total energy E is equal to the kinetic energy
p2j(2m) plus the potential energy V, i.e. E = p2j(2m) + V. In contrast to
Newton's second law in classical mechanics, analogous to E = p2 j(2m) + V,
the appropriate physical law that determines the time evolution of a quantum
mechanical system is
E1/1 = ii 1/1. (4.1.12)
This is the famous Schrodinger equation. From (4.1.10) - (4.1.11), equation
(4.1.12) can also be written as
(4.1.13)
This is a complex partial differential equation for the wave function 1/1.
Once again we emphasize that Schrodinger equation is a fundamental law in
quantum mechanics like F = ma in the classical mechanics. Hence (4.1.12) or
(4.1.13) is not a derived equation. Nonetheless, equation (4.1.12) has an alge-
braic analogy of energy conservation in classical mechanics: H =
Ij(2m)p2 +
= =
V E constant energy.
A lot of times, one needs to look for solutions of equation (4.1.13) in the
harmonic oscillation form in time t
(4.1.14)
Since the unit of Ii is [joule][sec], the unit of w must be [joule]. Thus, w rep-
resents the energy of the particle. When the potential of a system is given,
the boundary conditions are settled. Substituting (4.1.14) into the Schrodinger
equation (4.1.13), one obtains a self-adjoint eigenvalue problem. The eigen-
function is wand the eigenvalue is w. This problem has infinitely many modes.
The eigenvalue for a mode is called the energy level of the mode. For an at-
tractive potential, the self-adjoint eigenvalue problem has only discrete eigen-
values. Between each two eigenvalues there is an energy gap, i.e., energy levels
are quantized.
4.1. Scattering Method 81
Exercise: Let
V(x) = {O, ~f Ixl < i' (4.1.15)
00, If Ixl ~ 2".
Solve equation (4.1.13) with the potential given by (4.1.15). What can you
conclude when a is large (macro-mass-elements) and when a is small (micropar-
ticles)?
Next we show how to use our scattering method to solve (4.1.13). Send a
wave 1/Jin from x -00. =
1/Jin = ei(lc:r:+wt), kw < O. (4.1.16)
For real k and w, the above wave does not represent a particle's motion since
the integral of the modulus of the above wave is divergent. Hence, from the
quantum mechanical view point the above wave does not have a physical mean-
ing. However, the mathematical method used to determine such a solution is
useful.
The transmission and reflection waves are respectively
= Tei(lc:r:+wt) , (4.1.17)
Rei(-Ic:r:+wt). (4.1.18)
Hence if 1/J is a solution of (4.1.13) due to the disturbance 1/Jin, then the solution
takes the form
ei(lc:r:+wt) + R ei ( -Ic:r:+wt), x --+ -00,
1/J(x, t) '" { T ei (Ic:r:+w t) , x --+ +00. (4.1.19)
To determine Rand T, we need to use (4.1.13). Let us consider a specific case:

V(x) = Q&(x), where &(x) is the Dirac delta function and Q is a constant.
Now, the solution has the following form
ei(lc:r:+wt) + R ei ( -Ic:r:+wt), x :5 0,
1/J(x,t) ={ T ei(lc:r:+wt), x ~ O. (4.1.20)
By the continuity of 1/J at x = 0, we have
T= l+R. (4.1.21)
Integrating (4.1.13) over an interval (-e, e) yields
o
in-
ot
1_
'l/Jdx = --(-I
1,,2 o1/J
2m ox --I
o'I/J
ox )+ Q'I/J(O).
:r:= :r:=-
(4.1.22)
Using (4.1.20) and letting e --+ 0, we have
ikh2
2m (1 - R - T) + QT = O. (4.1.23)
Using (4.1.21), we finally obtain
T (4.1.24)
ikh2 - Qm'
Qm
R (4.1.25)
ikfl.2 - Qm
Thus, the solution (4.1.20) is completely determined.
4.2 Inverse Scattering for the KdV

4.2.1 Inverse scattering method
In the previous section, using the transmission and the reflection waves we
found the wave function (called scattering data) from a given potential function
V(x). In contrast, the goal of the inverse scattering method is to recover the
potential V(x) from the given scattering data ,p.
Consider the initial value problem for a Korteweg-de Vries equation (KdV):
Ut - 6uu x + U xxx = 0, -00 < x < 00, t > 0 (4.2.1)

u(x, t = 0) = uo(x). (4.2.2)
The equation (4.2.1) is called the standard KdV equation. There is a condition
on the initial value
and
1: (1 + Ixl)luo(x)ldx < 00.
This is the sufficient condition for the existence and uniqueness of the solution
to the initial value problem for the KdV. From fluid mechanics, the shallow
water KdV equation is
3 1
Ut + ,\ U x - '2 u U x - '6 U xxx = O. (4.2.3)
Let
U -+ ~ (,\ - 62 / 3 u)
X -+ _6- 1/ 3 x
t -+ t.
Then this shallow water KdV is transformed into the standard KdV (4.2.1).
Let
(4.2.4)
4.2. Inverse Scattering for the KdV 83
where A is independent of z. If we take u to be a known function and v to

be the unknown function, then (4.2.4) is a Riccati equation for v. A Riccati
equation can be linearized by a standard transform
I()IIJ
v-- (4.2.5)
I()
Equations (4.2.4) - (4.2.5) imply that
I()IIJIIJ - (u - A)I() = O. (4.2.6)
This is an eigenvalue problem for a linear Schrodinger equation as described
in section 4.1.2, with a potential u, the eigenvalue (or energy level) A and the
eigenfunction (or called wave function) I(). Since the function u is a solution of
the initial value problem (4.2.1) - (4.2.2), it must be a function of both z and
t. For this reason, one would easily conclude that A depends on t. Surprisingly,
this is not the case and A is independent of time t. This is stated in the following
theorem:
Theorem 4.1 If u(z, t) is a solution to (4.2.1)-(4.2.2), then
A = Ao = constant (4.2.7)
where Ao is an eigenvalue of
(4.2.8)
The proof of this theorem is quite easy and is described in section 4.3.
Therefore the eigenvalues in (4.2.6) are completely determined by (4.2.8). The-
orem 4.1 will be used to derive the inverse scattering method in section 4.5.
In contrast to the scattering problem that finds the scattering data, the
inverse scattering problem finds the potential u(z, t) from known scattering
data I()(z, t). Gel'fand and Levitan (1955) [8] established a procedure to find
u(z, t) by solving a linear integral equation. They showed that the function
u(z, t) in (4.2.6) can be written as
{)
u(z,t) = -2{)zIz,z;t) (4.2.9)
where I< is the solution of the following linear integral equation
Iz, y; t) + B(z + y, t) + 100

B(y + z, t)Iz, z; t)dz = O. (4.2.10)
The equation (4.2.10) is called the Gel'fand-Levitan equation. The integration

kernel B is given by
B(e,t) = L C;'(t)exp(-kme) + 211'1 1_

N
m~l
00
00
R(k,t)exp(ike)dk. (4.2.11)
In this expression, the quantities km, Cm(t) and R(k, t) are defined as fol-
lows:
(a) km's are given by

k;'=-Am, m=1,2,,N (4.2.12)
and {>.m}~=l are distinct eigenvalues of (4.2.8). The corresponding
eigenpairs are {Am, t,Om}~=l with I~oo t,O;'(x)dx = 1;
(b) Cm(t)'s are determined by
(4.2.13)
(c) A wave exp(ikx) is sent from x = +00 and hence
t,O "" exp(-ikx) + R(k,t)exp(ikx) asx -+ 00, (4.2.14)

t,O "" T( k, t) exp( -ikx) as x -+ -00. (4.2.15)
The relationship between the reflection coefficient R and the transmission co-
efficient T is
(4.2.16)
To determine T and R, we have the following theorem.
Theorem 4.2
Cm(t) = Cm(O) exp(4k~t),

{ R(k, t) = R(k, 0) exp(8ik3 t), (4.2.17)
T(k, t) = T(k, 0)
where Cm(O) = limx-too t,Om(x, 0) exp(kmx), and R(k,O) and T(k,O) are ob-
tained from the initial data for the KdV equation u(x, t = 0) = uo(x) using
(42.8).
To end this section, we summarize the inverse scattering method for the
KdV presented above. We want to solve the initial value problem for a KdV
equation
Ut - 6uux + U xxx 0, = -00 < x < 00, t > 0,

u(x, t = 0) = uo(x)
4
where E I~oo Id~nn uo(x)j2dx < 00 and I~oo(1 + IxDluo(x)ldx < 00. First of
n=l
all, we need to solve the eigenvalue problem
t,OOxx - (uo(x) - A)tpo = 0
to obtain km , Cm(O) and R(k, 0). In some situations, T(k, 0) (and hence R(k, 0))
is determined a priori from physical considerations. For an arbitrary potential
Uo, it is not that easy to find the reflection coefficient R. We will see this in
the next subsection. From Theorem 4.2, Cm(t) and R(k, t) are determined.
Hence the integration kernel B(e, t) in the Gel'fand-Levitan equation (4.2.10)

is determined as
N
B(e, t) L: C!(O) exp(8k~t - kme)
1
m=l
+-
1
00
R(k, 0) exp[i(8k 3 t + ke)]dk. (4.2.18)
211" -00
Next we solve the linear Gel'fand-Levitan integral equation (4.2.10) to obtain

K (z, Yj t). Usually, people seek solutions whose variables z, Y are separated.
Namely, let
M
K(z,y,t) = L:Gn(z,t)Hn(y,t),
n=l
or
M
K(z, y, t) = L: In (t)Jn(z)Kn(y).
n=l
Finally the solution of the initial value problem for the KdV is given by
o
u(z, t) = -2 oz K(z, Zj t).
The power of the inverse scattering method is that it makes it possible to get
solutions of a nonlinear problem by solving two linear problems (see Fig. 4.2).
One of the two linear problems is the eigenvalue problem for a time indepen-
dent Schrodinger equation. This problem can sometimes be solved analytically
by using special functions which are associated with the second order ODE.
The other one is the linear Gel'fand-Levitan integral equation. This equation
can sometimes be solved analytically by using the method of separation of vari-
ables. Therefore we sometimes can use the inverse scattering method to find
analytic solutions of some specific nonlinear initial value problems. This type
of systematic method is a remarkable discovery in mathematics and nonlinear
wave research!
4.2.2 KdV solitons
For some initial conditions, the problem (4.2.1) - (4.2.2) has solutions which
consist of only solitons. Solitons are the waves dying out at infinity and having
profiles which unaltered after collisions with other solitons. In this subsection,
we present I-soliton, 2-soliton, 3-soliton and N-soliton results by following the
four steps:
Gel'fand-Lavitan Equation:
K(x, Yi.t) + B(x + y, t) + 1 B(y + z,

00
t) K(x, Zi t) dz = 0
B(E, t)
seperation of varibles to get
/
K(x,Yi t)
Solve
a:a: + (-u.o(x) + A) = 0 \
\
using special functions. Find:
k! = -Am, 4>m(X) , R(k,O),
Cm(O), Cm(t), R(k, t).
u(x,t) = -2!K(x,xjt)
/ \
uo(x) u(x, t)
Ut - 6uux + Ua:xx =0
Figure 4.2: Solution diagram for the inverse scattering method.
(a) Solve
xx - (uo(x) - >') = 0, (oo) = 0
and find the eigenpairs {Am, m(X)}~=l' The eigenfunctions are normal-
ized in such a way that J~oo lml 2 dx = 1, m = 1,2"", N. Then, the
total number of solitons is N.
(b) Compute:
km = V->'m'
Cm(O) = x-+oo
lim m(x) exp(kmx),
R(k,~) = 0,
N
B(e, t) = L C!(O) exp(8k!t - kme)
m=l
(c) Solve the Gel'fand-Levitan equation:
K(x, y, t) + B(x + y, t) + 1 00
B(y + z, t)K(x, z, t)dz = 0
by the method of separation of variables.

(d) Find the solution by differentiation:
a
u(x, t) = -2 ax K(x, x, t)
and interpret the physical meaning of the solution.

In the rest of this subsection, we are going to find 1-, 2-, 3- and n-soliton
solutions of the standard KdV equation using the above four steps.
(i) I-soliton example.
By direct integration, the KdV derived in Chapter 3 gives a traveling soliton
solution. This solution has one hump (or dent) and travels at a uniform speed.
It is a I-soliton solution. This solution can be recovered by using the inverse
scattering method to solve an initial value problem for a KdV equation.
Consider the initial value problem
Ut - 6uu x + U xxx = 0, -00 < x < 00, t > 0, (4.2.19)

u(x, t = 0) = uo(x) = -2sech 2 x. (4.2.20)
By (4.2.8), the related eigenvalue problem
<POxx + (2 sech 2 x + >')<po = 0 (4.2.21)
can be solved exactly (see (ii) 2-soliton example below). The solution is
kl = 1, C(O) = v'2. (4.2.22)

The reflection coefficient R(k,O) = O. This can be proved from the scattering
theory by using the hypergeometric functions.
With kl = = =
1, C(O) v'2 and R(k, 0) 0, the Gel'fand-Levitan equation is
K(x, y, t) = 2 exp(8t - x - y)
+2exp(8t-y) 1 00
K(x,zjt)exp(-z)dz. (4.2.23)
Let K(x, Yj t) = L(x, t) exp( -y) (separation of variables). Then
L(xjt) + 2exp(8t - x) + 2 exp(8t)L(x,t) 1 00

exp(-2z)dz = o.
Hence
L -2exp(x)
(Xj t) = 1 + exp(2x _ 8t) ,
and
-2exp(-x - y)
K (x, Yj t) = ( )"
1 +exp 2x - 8t
Therefore, the solution of the initial value problem (4.2.19) - (4.2.20) is
u(x,t)
o
-2 ox K(x, Xj t)
8 exp(2x - 8t) 2
[1 + exp(2x _ 8t)]2 = -2sech (x - 4t). (4.2.24)
This result agrees with the single solitary wave solution found in the last chap-
ter. This wave has a single dent and travels at a uniform speed equal to 4. Of
course it has a permanent form (can be called a permanent shape). This wave
solution is called a I-soliton.
(ii). 2-soliton eXaIIlple.
We will show that the following initial value problem for the standard KdV
equation
Ut - 6uu x + U xxx = 0, -00 < x, t < 00, (4.2.25)

u(x, t = 0) = uo(x) = -6 sech 2 x (4.2.26)
has a 2-soliton solution. That is, a solution consists of two waves of permanent
form which die out at infinity and travel at two different speeds. The inverse
scattering method is employed to find this 2-soliton solution.
Step 1. Solve the linear eigenvalue problem.
-n> solve the eigenvalue problem
t/Jxx + (6 sech 2 x + >..)t/J = 0, -00 < x, 00, (oo) = 0, (4.2.27)

e
let = tanh x. This transformation maps (-00,00) for x to [-1,1] for e.
Then
.,p", = .,pe{1 - e),

.,p",,,, .,pee{1 - e)2 - 2e(1 - e).,pe.
Equation (4.2.27) becomes
(4.2.28)
with .,p(e) = O. Comparing (4.2.28) with the generalized Legendre equation
d [ (1 -
de e )d.,p]
de + [l(l + 1) -
2
1m
_ e
2
2
]
t{; = 0,
we have
l(l + 1) = 6, l ~ 0,
_m 2 = A, 0< Iml ~ i.
Hence, l = 2 and
(4.2.29)
Then the eigenfunctions .,p are given by the associated Legendre polynomials
pr(e). It is useful to recall the following. When land m are positive integers
or zero, one has the Rodriques formula for Legendre polynomials
where Pl(Z) are Legendre polynomials:
Po(Z) = 1,
P1 (z) = z,
1
P2 (z)= 2(3z 2 - 1),
1
P3(Z) = 2(5z3 - 3z),
1
P4(Z) = S(35z4 - 30z 2 + 3),
1
P5(Z) = S(63z 5 - 70z 3 + 15z),
1 dn [2 n] 2n - 1 n- 1
Pn(z) = -nn. zn (z - 1) = - -n z Pn- 1 (z) -
2'-d - n- Pn- 2 (z).
Therefore,
tP1 pi (e)
(e - 1)~ !p2(e)
de
i3(1- e)~e
3i sech x tanh x.
tP2 pi (e)
d
(e - 1) de2P2(e)
2
3(e - 1)
-3 sech 2x.
Since
1: ItP1(XWdx = 91: tanh 2x sech 2xdx = 6,
and
1: ItP2(X)1 2dx
the normalized eigenfunctions are
= 91: sech 4 xdx = 12,
(4.2.30)
(4.2.31)
Step 2. Compute km and the coeficients Cm(O), Rand T.

~w we can compute C 1 (0) and C 2 (0) using formula (4.2.13) as t = O.
C1(0) lim 1F1(X) exp(k1x)

",-+eXl
lim
"'-too
I! tanh x sech x exp( x)
V"2
v'6,
lim 1F2(X) exp(k2x)
"'-too
lim 3
1!2 sech 2 x exp(2x)
"'-too V"2
2V3.
From (4.2.17), we have
C1(t) C1(0) exp( 4krt)
v'6 exp( 4t),
C 2 (t) C 2 (0) exp(4k~t)
2..J3 exp(32t).
By using the hypergeometric functions, one can prove that R(k, 0) = O. By

(4.2.18), the integration kernel of the Gel'fand-Levitan equation is
2
B(e, t) L C;, (t) exp( -kme)
m=l
6exp(8t)exp(-e) + 12exp(64t)exp(-2e). (4.2.32)
Step 3. Solve the Gel'fand-Levitan equation by the method of separation of variables.

-so we have the Gel'fand-Levitan equation as follows:
+ 6 exp(8t) exp( -(x + y)) + 12 exp(64t) exp( -2(x + y))
1
K(x, y; t)
+ 00
[6 exp(8t) exp( -(y + z)) + 12 exp(64t) exp( -2(y + z))]
K(x,z;t)dz=O. (4.2.33)
The solution of separation of variables is assumed to be
K(x, y; t) = L1 (x, t) exp( -y) + L 2(x, t) exp( -2y). (4.2.34)
Substituting (4.2.34) into the Gel'fand-Levitan equation (4.2.33) and compar-

ing the coefficients of exp( -y) and exp( -2y), we have
(1 + 3eSt-2X)L1 + 2e St - 3x L2 = _6e St - x , (4.2.35)

4e64t-3x L1 + (1 + 3e64t-4x)L2 = _12e64t-2x. (4.2.36)
The solution for the linear algebraic equations (4.2.35) - (4.2.36) is
(4.2.37)
where
1 + 3e 8t - 2x 2e8t - 3x )
(
det 4e64t - 3x 1 + 3e64t - 4x
1 + 3e 64t - 4x + 3e 8t - 2x + e72t - 6x,
_6e8t - x 2e8t - 3x )
(
D.1 det _12e64t - 2x 1 + 3e64t - 4x
_6{e8t - x _ e72t - 5x),
1 + 3e 8t - 2x _6e8t - x )
(
D.2 det 4e64t - 3x _12e64t - 2x
_12{e64t - 2x + e72t - 4x).

From (4.2.34) and (4.2.37),
K(x, y; t) L1e- Y + L 2 e- 2y
~ [ _ 6(eSt - x _ e72t - 5x)e- y _
12(e64t - 2x + e72t - 4x)e-2y].
Hence
K(x, x;t) = -1 [eSt - 2x + e72t - 6x + 2e 64t - 4x] . (4.2.3S)
Step 4. Find the solution u(x, t) and discuss its properties.
8
u(x,t) -2 8xK(x,x;t)
~~ {[(_2e 8t - 2x _ 6e72t-6x _ Se64t-4x) .
.(1 + 3e64t-4x + 3e 8t - 2x + e72t-6x)]
_[(_12e64t-4x _ 6e 8t - 2x _ 6e72t-6x) .
.(e8t-2x + e72t-6x + 2e64t-6X)]).
This expression can be simplified into the following form:
( ) _ 12 3 + 4 cosh(2x - St) + cosh (4x - 64t)

(4.2.39)
u x, t - - [3 cosh(x _ 2St) + cosh(3x - 36t))2 .
This is a solution to the problem (4.2.25) - (4.2.26). Since uo(x) = -6 sech 2 x

given by (4.2.26) satisfies
'L,!=o f~oo I(dn /dxn)uo(x) 12 dx < 00,

and
f~oo(1 + Ixl)luo(x)ldx < 00,
the solution to the problem (4.2.25) - (4.2.26) is unique (Lax (196S) [5]). Thus,
(4.2.39) is the only solution to the problem (4.2.25) - (4.2.26).
What remains to be shown is that (4.2.39) consists of two solitons. To
observe these two solitons, we need to move along with the waves. So we
introduce the following two moving coordinates
t = t, em = x - 4k;'t, m = 1,2. (4.2.40)
We expect that
(4.2.41)
where ~! are constants. Indeed, this is the case. We show the details as follows.
Case (a). For the smaller soliton, kl = 1'~1 = Z - 4t.
In this case,
= -12 3 + 4 cosh(2{t) + cosh(4~1 - 48t)
[3 COSh(~l - 24t) + cosh(3~1 - 24t)]2
le- 4e1
~ -12-::--.,..L.2--,.-".,......,..".
(~e-el + ~e-3el)2
= -2 sech2({t - ~t> as t -t 00, (4.2.42)
where e-ei = v'3. Similarly,

lim u(z, t) = -2 sech2(~1 - ~l)
t-+-oo
(4.2.43)
where e-ei" = v'3.

Case (b). For the larger soliton, k2 = 2,6 = z -16t.
In this case,
= -12 3 + 4 cosh(26 + 24t) + cosh (46)
u(z,t) = u(~,t)
[3 cosh(6 - 12t) + cosh(36 + 12t)]2
41e 2e2
~ -12 ':'ft'"----:---.2--.--:::-:--:-::"
[~e-e2 + ~e3e2)2
= -8 sech2[2(~2 - ~t)]
=-8sech2[2(z-16t-~t)] as t-too, (4.2.44)
where e- 2et = v'3.

Similarly,
lim u(z, t) = -8 sech2[2(z - 16t - ~2)] (4.2.45)
t-+-oo
where e- 2G = v'3.
The physical meaning of (4.2.42) and (4.2.44) is very clear. If we ride with
a reference frame of velocity 4, then we eventually see only the smaller soliton
-2 sech 2(z - 4t - ~t>. On the other hand, if we ride with a reference frame of
velocity 16, then we eventually see only the large soliton -8 sech 2[2(z - 16t -
~t)]. Therefore, the solution (4.2.39) consists of two solitons. When t < 0, the
larger soliton is behind the smaller one. The larger the soliton is, the faster it
travels. Hence, the larger soliton catches up with the smaller one at t = 0 and
eventually passes the smaller soliton as time t > 0 approaches infinity. See Fig.
4.3 for this soliton passing process.
(iii). 3-soliton example.
We will show that the following initial value problem for the K-dV equation
Ut - 6uu:1: + U:1::1::1: = 0, -00 < z < 00, t > 0, (4.2.46)
u(z, t = 0) = uo(z) = -12 sech 2z (4.2.47)
Figure 4.3: Collision of two KdV solitons: the evolution of the initial pro-
file (4.2.26) in both positive t direction and negative t direction according to
(4.2.39).
has a 3-soliton solution. Namely, a solution consists of three waves of permanent

form which die out at infinity and propagate at three different velocities. The
inverse scattering method is employed to find such a 3-soliton solution.
Consider the eigenvalue problem
<pxx + (12sech 2 x + -\)')1/1 = O. (4.2.49)
Let e= tanh x, then the above equation becomes

de + ( 12 + 1 _>.)
ded [(1 - e )d1/l]
2
e 1/1 = o. (4.2.50)
Comparing the above equation with the generalized Legendre equation
-d [(1- e)-
2 d1/l]
de + [l(l + 1) -
2
-m-2 ] 1/1 = 0
de 1- e
we have
1(1 + 1) = 12, I ~ 0,
_m 2 = A, 0 < Iml ~ o.
Hence, 1= 3, and
m1 = 1, A1 = -1; (4.2.51)
m2 = 2, A2 = -4; (4.2.52)
m3 = 3, A3 = -9. (4.2.53)
Then the eigenfunctions t/Jn are given by the associated Legendre polynomials
pr(f.):
3
t/J1 = P31 (tanh x) = 2"sechx(4 2
- 5sech x),
t/J2 = P;(tanhx) = 15 tanhx sech 2x,
t/J3 = P;(tanhx) = 15 sech 3x.
Since
The normalized eigenfunctions are

-t/J1 v'3
= Tsechx (4 - 5sech x),
2
(4.2.54)
-t/J2 = -2-tanhx
Vl5 2
sech x, (4.2.55)
-t/J3 = -4-sech
Vl5 3 x. (4.2.56 )
Now we can compute Ck(O) (k = 1,2,3) and then find Ck(t) (k = 1,2,3) :
C1(0) = lim t/J1(x)e X = 2V3,
x~oo
and
C1 (t) = 2v'3e 4t , (4.2.57)
C2 (t) = 2Vl5e 32t , (4.2.58)
C3 (t) = 2Vl5e 10Bt (4.2.59)
The potential is reflectionless and thus R(k,O) = O. The integration kernel

of the Gel'fand-Levitan equation is
B(e, t) = 12e Bt -e + 6e64t-2e + 60e216t-3e (4.2.60)
and hence the Gel'fand-Levitan equation is
+ 12e Bt -e + 6e64t-2e + 60e216t-3e +

1
-K(x, y, t)
00
(12eBt-y+z + 6e 64t - 2(y+z) + 60e216t-3(Y+Z)
xK(x, z, t)dz = O. (4.2.61)
To solve the above equation, let

3
K(x,y,t) = - I: C;'(t)Lm(x,t)e- kmY (4.2.62)
m=l
Substituting the above expressions into the G-L equation, we can get the fol-
lowing algebraic equations:
Putting Cm(t) (m = 1,2,3) into the above equation and solving them for
Lm(x, t) (m = 1,2,3), we have
where ~1' ~2, ~3 and ~ are respectively given by
~ = e6x + 10e 216t + 15e64t+2x + 6e2BOt-4x + 6e Bt +4x

+15e224t-2x + 10e72t+2x + e2BBt-6x.
Upon simplification, the desired solution is given by
(4.2.63)
Figure 4.4: Collision of three KdV solitons: the evolution of the initial profile
(4.2.47) according to (4.2.63).
where Ut{x, t) and U2(X, t) are expressed respectively as
u1 (x, t) = 3024 + 24 cosh(10x - 2BOt) + 240 cosh(Bx - 224t)

+360 cosh(6x - 72t) + 720 cosh(6x - 216t)
+960cosh(4x - 20Bt) + 1920cosh(4x - 64t)
+600 cosh(2x - 152t) + 1200 cosh(2x - Bt)
+3240cosh(2x - 56),
U2 (x, t) cosh(6x - 144t) + 6cosh(4x - 136t)
+15cosh(2x - BOt) + lOcosh(72t).
The collision process of the three solitons found above is shown in Fig. 4.4.
(iv). N-soliton example.
If (4.2.B) has N distinct eigenvalues and the reflection coefficient R is zero,
then the initial value problem for the KdV equation
Ut - 6uu z + U zzz = 0, -00 < x, t < 00, (4.2.64)

u(x, t = 0) = uo(x) = -N(N + 1)sech 2 x (4.2.65)
has aN-soliton solution. In the following, we will show how to find these N
solitons.
The corresponding Gel'fand-Levitan equation is
N
K(x, Y; t) + L: C,;(t) exp[-km(x + y)] +
t
m=1
m=1
C,;(t)exp(-kmY) 1
It:
00
exp(-kmz)K(x,z;t)dz = O. (4.2.66)
To solve this equation, let

N
K(x, y;t) =- L: Cm(t)tPm(x) exp( -kmy). (4.2.67)
m=1
Substituting this into (4.2.66) and requiring the coefficients of exp(kmy) to
vanish, we obtain
L: CmCn exp[~(k: ~ kn)x]tPn

N
tPm(x) +
n=1 m n
Cmexp(-kmx), m= 1,2,,N.
This equation can be written in a matrix form
MtP=E (4.2.68)
with
M [cmCn exp[-(km
k
+ kn)x] ~ ]
k + mn
N
,
m+ n m,n=1
tP [tPm]~=1'
E = [Cmexp(-kmX)]~=1.
Solving (4.2.68), one can derive that
8
K(x, x; t) = 8x logdet(M). (4.2.69)
Hence
82
u(x, t) = -2 8x 2 logdet(M). (4.2.70)
The following theorem can be proved (see Miura (1976) [4]).
Theorem 4.3 The function u(x, t) given by (4.2.70) consists of N solitons.

Asymptotically, on the frame moving with the mth soliton ((x, t) -t (em, t))
one has
(4.2.71)
where em = x - 4k! t and emo is a constant. The mass of the m-th soliton is
4m, the height is 2k!, and the speed is 4k! (m 1,2, ... , N). =
4.2.3 KdV solitons with a wake

From the previous examples, we have seen that if the initial profile uo(x) yields
N discrete negative eigenvalues (-k~, -k~, ... , -kJ.,) and has no continuous
spectrum, then the initial value problem for the standard KdV has an N-soliton
solution. By the uniqueness of the solution to the IVP
Ut - 6uu z + U zzz = 0, -00 < x, t < 00,

u(x, t = 0) = uo(x), -00 < x < 00,
the potential uo(x) in the eigenvalue problem for the Schrodinger equation is
reflectionless, i.e., R(k, t) =
O. We will see that the condition R(k, t) - 0 is
responsible for the oscillatory part of a solution. This non-solitary oscillation is
called a wake in fluid mechanics. As an example, let us take the delta function
as the initial condition. This potential yields not only the discrete negative
spectrum, but also the positive continuous spectrum.
Consider the IVP:
Ut - 6uu z + U zzz = 0, -00 < x,t < 00, (4.2.72)

u(x, t = 0) = Pc5(x), -00 < x < 00. (4.2.73)
The corresponding eigenvalue problem in the inverse scattering method is
4>" - (Pc5(x) - >')4> = 0, -00 < x < 00. (4.2.74)
The discrete spectrum is the eigenvalue >'1 = _P 2 /4. The corresponding eigen-
function is 4>(x) = JP/2 exp(Plxl/2) when P < O. When t --t 00, this eigen-
pair corresponds to a soliton:
Ul (x, t) = -2k~ sech 2 [k 1 (x - 4k~t)]

where kl = .../->'1 = -P/2 > O.
However, the eigenvalue problem also has a continuous spectrum. Let us
look at the following exercise problem:
Exercise: Find the scattering data for the Schrodinger equation when the
potential is Qc5(x).
Solution: The scattering problem is
(4.2.75)
where tjJ is a bounded function. Let
tjJ(x, t) = exp(iEt/h)4>(x).
Then, Schrodinger equation becomes
4>" - (Pc5(x) - >')4> = 0 (4.2.76)

where
2m
P= 2~Q and A = -,;:E. (4.2.77)
h
In the following, we will see that the scattering data (A and ) critically depends
on the sign of P.
Let us send a wave from x = 00:
in(X) = I exp(ikx).
In order to satisfy (4.2.76), we set k =.../X. To make (oo) finite, we must have
Im(k) ~ O. The general scattering pattern is
I eik:t: + Re-ik:t: x> 0,

(x) ={ T eik:c , ' x< O.
(4.2.78)
The continuity condition at x = 0 implies that
I+R=T.
The jump condition at x = 0 is
This implies that

ik[I - R- T] = PT.
We can solve for Rand T and obtain
PI
R(k) = - P _ 2ik' (4.2.79)
T(k) = 2ikI (4.2.80)

P- 2ik
Now, let us discuss the value of k. There are two cases:
Case (i) k is real: In this case, the modulus of the wave function does not
represent the probability distribution function of a particle and has no physical
constraint. Therefore, k can take on any value. This spectrum is A = k 2 > 0
and is continuous. The coefficients of reflection and transmission are given by
equations (4.2.79) and (4.2.80) respectively.
Case (i) k is purely imaginary: In this case, k = .../X represents the energy
level of the system and the eigenfuction must be normalized as J~oo 11 2 dx = 1.
This requires R = 0 in order to satisfy (oo) = O. Therefore, T = I and
eik:c x> 0,
(x) = I { e -ik~ , x < O.
From the jump condition for :c at x = 0
ik(I + I) = PI,
we have
k = ~i, P< O. (4.2.81)
Hence the spectrum is A = k 2 = _p 2 /4 < O.

In summary, when P < 0, there are a discrete spectrum and a continuous
spectrum. For the discrete spectrum, the reflection coefficient is zero. The
energy level is
E = li 2 p2 /(8m)
and the wave function is
1/J(x, t) = Pexp [ilip 2 /(8m)t - Plxl/2] .
When P > 0, there is a only continuous spectrum. For the continuous spec-
trum, the reflection coefficient is given by equation (4.2.79). If the magnitude
of the incident wave is I = 1, then the reflection coefficient is
P
R(k,O) = 2ik _ P (4.2.82)
Using (4.2.17), we obtain
R(k, t) = 'k P P exp(8ik 3 t). (4.2.83)

2z -
The kernel of the Gel'fand-Levitan equation is
1
B(e, t) = C 1 (t) exp( -k1e) + 271' 1 00
-00 2ikP_ P exp[ki(8k 2 t + e)] dk. (4.2.84)
The Gel'fand-Levitan equation is

+ C 1 (t) exp[-kdx + V)]
1
Ix, Y; t)
00
1
+ 271' -00 2ikP_ P exp[ki(8k 2t + x + v)] dk
+1 00
{C1(t)exp[-k1(y+z)]
+ 1
00
1
271' -00 2ikP_ P exp[ki(8k 2t + Y + z)] dk}
Ix,z;t)dz = o. (4.2.85)
This is a linear equation. The following superposition principle can be applied
to the solution of the Gel'fand-Levitan equation:
Ix,y;t) = I<.(x,y;t) +I<w(x,y;t)
where I<. is the soliton part and I<w is the wake part. Then, the soliton part
satisfies the following integral equation:
I<. (x, y; t) + C 1 (t) exp[-kl (x + V)]
1 00
C 1 (t) exp[-k 1 (y + z)] I<. (x, z; t) dz = O. (4.2.86)
The soliton is
(4.2.87)
where the phase shift is Zo = Pin 2. The wake part Kw satisfies another
integral equation:
Kw(z,y;t) + Pexp [~(-2P2t+Z+y)] +
100
Pexp [~ (-2p 2t + z + V)] Kw(z, z;t)dz = O. (4.2.88)
In the above, we have used the residue theorem for the contour integral with
respect to the complex variable k. Finally, we can solve this equation and use
(4.2.89)
to obtain the wake part of the solution of the IVP. However, we cannot find
an analytic solution of the integral equation (4.2.88) and can only obtain an
approximate solution.
The total mass for the wake part is P < 0 and that for the soliton part is
-2P> O. The soliton and the wake together comprise the solution of the IVP.
This solution is shown in Fig. 4.5.
4.3 Lax Pair and KdV Hierarchy*

The most fundamental property that makes the inverse scattering method ap-
plicable to the KdV is that the eigenvalues of the linear Schrodinger equation
associated with the KdV are time independent. In this section, we will discuss
how to use the Lax pair to generate a class of PDEs whose associated eigen-
value problems have time independent eigenvalues. The KdV is a member in
this class.
The partial differential equation of the following type
Ut = N (u) (N is an operator independent of t) (4.3.1)

is called an evolution equation. The standard KdV equation
Ut - 6uu., + u.,.,., = 0
is an evolution equation with
N = 6u8., +8:.
Sometimes, the evolution equation Ut = N(u) can be written in the form
L t = [M,L] (4.3.2)
4.3. Lax Pair and KdV Hierarchy* 103
P~(x)
o x
U(x, t)
o x
-p
Figure 4.5: Solution of the IVP for the standard KdV with the initial data as
uo(x) = Po(x), P < O. The solution consists of a soliton plus a wake.
where both M and L are linear operators independent of t, and
[M,L] = ML - LM (4.3.3)
is called the commutator of M and L. The operator pair M and L is called
the Lax pair and the evolution equation (4.3.2) is called the Lax equation. For
example, for the simple PDE Ut - u., = 0, we have L = -0; + U and M = 0.,.
So, L t = Ut and
[M,L] 0.,(-0; +u) - (-0; +u)O.,

O.,U (note that uo., = 0).
Hence, Lt = [M, L] is equivalent to Ut u.,. =

Another example comes from the standard KdV equation
Ut - 6uu., + u.,.,., = o.
Let
L = -0; + U and M = -40: + 6uo., + 3u.,.
Then, the operator equation L t = [M, L] is equivalent to the standard KdV
equation.
The following theorem is very important and crucial for establishing the
inverse scattering method (ISM thereafter).
Theorem 4.4 If Lt + [L, M] = 0 and L is a self-adjoint operator, then the

spectrum A from
L(x) = A(x), -00 < x, t < 00 (4.3.4)
is independent oft.
Proof:
Differentiate L = A with respect to time t. One obtains that
At = Lt + Lt - At
(M L - LM) + (L - A)t
= MA - LM + (L - A)t
(L - A)(t - M).
Multiplying the above equation by * (the complex conjugate of ) and inte-
grating the resulting equation with respect to x from -00 to 00, one has
At (, } (, (L - A)(t - M)}
(L - A), t - M}
= (O,t- M )
o
where (, ) = f~oo ~ dx is an inner product, and ~ is the complex conjugate
of . Since is an eigenfunction, (, ) is positive. Hence, we must have
At = O. (4.3.5)
Therefore, A(t) = Ao is a constant.
In the last part of this section, we ahve briefly described the KdV hierarchy.
Recall that for the Lax equation
Lt + [L,M] = 0,
=
if L -o~ + u and M =
-4o~ +6uo.: + 3u.:, then the Lax equation is equivalent
to the standard KdV eqaution
As we might have noticed in the proof above, one has the freedom to choose
M. Other choices of M do not destroy the property At = o. The question is
how to choose M such that the resulting evolution equation has a physical
meaning. One choice is
+L
n
Mn = _ao~n+l (Umo~m-l + o~m-l) + A (4.3.6)
m=l
where a = const, Um = Um(x, t) and A = A(t). The equations
Lt + [L, Mn] = 0, n = 1,2,3, ... (4.3.7)
4.4. Bii.cklund Transform 105
constitute a class of PDEs which can all be solved by ISM. When n = 1, a =

4, Ul = 3u and A = 0, the corresponding equation is the KdV equation Ut -
3uu3: + U3:3:3: = O. Therefore, the class of Lax equations defined by (4.3.7) is
called the KdV hierarchy.
Example 1: n =0
Now, M = -a83: + A. Let a = -1 and A = 0, then the Lax equation is
equivalent to the PDE Ut - U3: = o.
Example 2: n = 2
Now, M =
-a8! + U28! + 8!U2 + U183: + 83:Ul + A. Let a -4, U2 = =
3u, Ul = 3u and A = 0, then the corresponding evolution equation is
Ut - U3:3:3: - U3:3:3:3:3: + 3uu3: + 3uu3:3:3: = o.

For an arbitrary n,
and
E [8~m+1Um + a8~m+lu] - E Um8~m-lu = o.

n n
[L, M] =-
m=l m=l
Hence the corresponding evolution equations in the KdV hierarchy are
n n
Ut - E 8~m+1Um + a8~m+lu - E Um8~m-lu = 0, n = 0,1,2,. (4.3.8)
m-l m=l
4.4 Backlund Transform*

Besides the inverse scattering method (ISM), Backlund transform (BT) is an-
other systematic and analytic way to find solitons. A. V. Backlund discovered
his transform around 1876. He himself and others found a few minor applica-
tions of the transform within three decades after its discovery. Nonetheless, the
transform has no generally accepted definition. Perhaps due to its "ambigu-
ity" and "too" large freedom in use, the transform was more or less forgotten
between 1910 and 1960. In the late 1960's, the transform started to interest
people again as it was successfully used to find solitons (in many nonlinear
systems) which were then and still are of importance to theoretical physicists
and mathematicians.
In this section, we will first briefly explain what Backlund transform is. A
simple example that applies BT to Liouville equation will be shown. Finally
we present the method due to Wahlquist and Estabrook (1973) [15] to find
solitons by solving the initial value problems for the standard KdV equation.
This method is in the category of Backlund transform and enables one to find
multiple-soliton solutions of the standard KdV without quadrature.
To solve a partial differential equation, denoted by P(u) = 0, for the un-

known u(x, y), one may find a conjugate partial differential equation, denoted
by Q(v), for the conjugate unknown v(x, y). Here the conjugate unknown v is
introduced via a Backlund transform
v~ = F1(u,v,D OI u;x,y), (4.4.1)

Vy = F2 (u,v,D OI u;x,y) (4.4.2)
where DOI U denotes the set of partial derivatives of all orders. In practice, one
only needs to use these derivatives up to at most one order higher than the
order of the partial differential equation P(u) = O. Therefore, we have the
following equivalence relation
Original PD E <' '> Backlund Transform

P(u}=O ~==::>7
Conjugate PDE v~ = F!(u, v, DOIu; x, y)
Q(v}=O Vy = F2(U, v, DOIu; x, y)
A simple example of these equivalent classes is the correspondence between

the harmonic conjugates and the Cauchy-Riemann condition for analytic func-
tions of a complex variable.
Example: Cauchy-Riemann equations act as a Backlund transform for the

Laplace equation.
Laplace Equation Backlund Transform is

~u(x,y}=O -< Cauchy-Riemann Equations
~------~~------~
Harmonic Conjugate Vx =-uy
~v=O Vy =Ux
In the above table, the harmonic function v is the harmonic conjugate of u.
Hence, the conjugation defined in the BT is consistent with that for harmonic
functions.
It appears that the conjugate PDE Q(v) = 0 can sometimes be solved, or
some of its special solutions can be easily found. Once the conjugate function
v(x, y) is found, one may solve the BT to get the function u(x, y) which, in turn,
automatically solves the original PDE P(u) = O. This procedure is the same
as finding a harmonic function by integrating the Cauchy-Riemann equations
from a given harmonic conjugate, and is graphically shown below:
P(u} =0 u(x, y) is a solution u(x,y) is found

is solved = vx=FJ
vy =F2
Step 1. Solve Special solution v(x, y)

Q(v} =0 ==} Step 2. Solve the BT
4.4. Backlund Transform 107
Sometimes, the conjugate equation Q( v) = 0 is the same as the original

equation P( u) = 0, such as the previous example for the Laplace equation.
This seems disappointing. Nonetheless, the BT and the conjugate equation
may still be useful. An easy (or even a trivial) solution of Q(v) = 0 may still
lead to an interesting nontrivial solution of the original PDE P( u) = 0 via
solving the BT. A typical example is the BT for sine-Gordon equation.
Example: The equation u xy = sin u is called the sine-Gordon equation.
Let its BT be
Vx = Ux + 2'xsin[(v + u)j2],
Vy = -Uy + (2j'x)sin[(v - u)j2].
Then, the function v(x, y) yields
v xy = SIn v.
This is the conjugate equation which is the same as the original sine-Gordon
equation. Despite this, the BT can still lead to soliton solutions u(x, y) of the
original sine-Gordon equation u xy = sin u (see the book by Drazin and Johnson
(1989) [10]).
As a more concrete example, we next show how to use the BT to find the
general solution to Liouville equation:
(4.4.3)
We introduce the following BT:
VX = -U x + V2e(u-v)/2, (4.4.4)
Vy = U y - V2e(u+v)/2. (4.4.5)
Then the operation (ajay)(4.4.4) + (ajax)(4.4.5) leads to the conjugate
equation for v:
v xy = O. (4.4.6)
The general solution of this conjugate equation is
v(x, y) = f(x) + g(y). (4.4.7)

Substituting (4.4.7) into (4.4.4) and multiplying (4.4.4) by eJ(x) , one can obtain
eJ(x) [u + f - g]x = V2e(u+J-g)/2.
The integral of this equation is
_V2e-(U+J-g)/2 = J e-J(x) dx.
Similarly, from (4.4.5) one can get
_V2e-(u+J-g)/2 = J eg(y) dy.

From the sum of the above two equations, we solve for u(x, y):
u(x, y) = 2ln [- f e-f("') d:! f eg(y) 1

dy + g(y) - f(x). (4.4.8)
This is the general solution of the Liouville equation (4.4.3). Choosing f(x) =
-In x and g(y) = In y, we can obtain a special solution
8xy ] (4.4.9)
u(x, y) = In [ (2 _ x 2 _ y2)2 .
In the following, we explain how to use the BT that is due to Wahlquist and
Estabrook (1973) [15] to establish a nonlinear superposition principle for the
standard KdV and consequently to find the KdV solitons. In 1973, Wahlquist
and Estabrook published an ingenious paper that delineates a BT to find multi-
ple KdV solitons without quadrature. Let us consider the initial value problem
PI : Ut - 6uu", + u"''''''' = 0, -00 < x < 00, (4.4.10)

u(x, t = 0) = -N(N + l)sech2x, -00 < x < 00. (4.4.11)
Let u = W",. Then, (8/8x)(4.4.10) leads to
Wt - 3w; + w"''''''' = 0, -00 < x < 00. (4.4.12)

We take the standard KdV (4.4.10) and equation (4.4.12) as the original equa-
tions P(u) = O. They are denoted by P1(u) =
Ut - 6uu", - u"''''''' 0 and =
P2(u) =Ut - 3u; + u"''''''' =
0, respectively. Let (u*, w*) be the conjugate
variables of (u, w) and introduce the following BT:
w; = -w", + 2A + (1/2)(w - w*)2, (4.4.13)
w; = -Wt - (w - w*)(w - w*)",,,,
+2[u 2 + uu* + (u*)2]. (4.4.14)
To derive the conjugate equations, we calculate (8 2/8x 2)(4.4.13) + (4.4.14) and

obtain
w; - 3(W;)2 + w;"'''' = o. (4.4.15)
Calculating (8/8t)(4.4.14), one can obtain
u; - 6u*u; + u;"'''' = O. (4.4.16)
=
Hence, Qdu*) PI(U*) = O.
Next, with special solutions of (4.4.15)-(4.4.16)' we generate solutions of
(4.4.10) via solving the BT (4.4.13) - (4.4.14).
=
I-soliton as a special solution:
Take the trivial solution of (4.4.16): u*(x, t) O. Then, w* = u; = O. Let
A < O. The BT (4.4.13) - ( 4.4.14) becomes
A < 0, (4.4.17)
Wt = -w - w"'''' = 2u 2 (4.4.18)
4.4. Bii.cklund Transform 109
Let A = _k2 and k > O. Then, equation (4.4.17) can be directly integrated to
yield
W == Wl (:c, t) = -2k tanh[b - c(t)], if Iwi < 2k. (4.4.19)
Here, the function c(t) is arbitrary and resulted from the integration with
respect to :c. Notice that U = W",. Equation (4.4.18) yields
c'(t) = 4Ak.
Hence,
c(t) = -4k3 t - b o
where :Co is the integration constant. Therefore,
Wl(:C, t) = -2k tanh[k(:c - 4k 2t - :co)], Iwi > 2k, (4.4.20)
and
Wl(:C, t) = -2kcoth[k(:c - 4k 2t - :co)], Iwi < 2k. (4.4.21)
Thus, the solution of Pl(u) =0 is (using = w"')
U
Ul (:c, t) = -2k2sech 2[k(:c - 4k 2t - :co)], Ul ~ -2k 2. (4.4.22)

To satisfy the initial condition (4.4.11), one must have
k= 1, :Co = O.
Hence,
(4.4.23)
This is the I-soliton solution of the initial value problem (4.4.10) - (4.4.11)
when N = 1.
With this procedure, we have found a solution of the KdV without a direct
quadrature of the equation. Instead, we can begin with a simple solution of
the conjugate equation. Then, relatively easy integration of the BT yields a
solution of the KdV. This is a very charming method! More elegantly, this
method can lead us to find the N-soliton solution of (4.4.10) - (4.4.11) through
a purely algebraic recursive procedure. Let us first work on the two-soliton
solution before we get to the general N-soliton solution.
2-soliton as a special solution:
Let Wo, Wl and W2 be solutions of P2(w) = O. Introduce the following BT:
(Wl + wo)", = -2Al + ~(Wl - wO)2, (4.4.24)

1
(W2 + Wo)", = -2A2 + 2'(W2 - wo) ,
2
(4.4.25)
New solutions -Wl2 and -W2l of P2(W) = 0 are constructed from the pairs
(WI, A2) and (W2' Al) respectively via the following BT
(Wl2 + wI)", = -2A2 + ~(W12 - wI)2, (4.4.26)
(W2l + W2)", = -2Al + ~(W2l - W2)2. (4.4.27)

Bianchi's theorem of permutation implies that W21 = W12. Then, compute

[(4.4.24) - (4.4.26)] - [(4.4.25) - (4.4.28)]. This leads to
4(Al - A2)
W12 = Wo - . (4.4.28)
Wl-W2
This is called the nonlinear superposition principle. It states that a new so-
lution of the PDE P2(U) = 0 may be a nonlinear superposition ( namely a
pure algebraic operation) of some previously known solutions without using
quadrature.
As a simple example of a 2-soliton solution, we take
Wo == 0,
Wl = -2 tanh(x - 4t),
W2 = -4coth(2x - 32t),
There are special cases of equations (4.4.20) and (4.4.21). The reason why we
take W2 ex: coth is to avoid the case Wl - W2 = 0 in (4.4.28). With this choice
of Wo, Wl, W2, Al and A2, the quantity W12 is obtained from (4.4.28) as:
W12 = 6[2 coth(2x - 32t) - tanh(x - 4t)]-1. (4.4.29)

Then
--W12
a
ax
-12 3 + 4 cosh(2x - 8t) + cosh (4x - 64t)
(4.4.30)
[3 cosh(x - 28t) + cosh(3x - 36t)]2
This is the 2-soliton solution of (4.4.10) -(4.4.11), which were found earlier in
section 4.2.2 via the inverse scattering method.
The function W12(X, t) given by (4.4.29) is plotted in Fig. 4.6. The graph
consists of ladder steps. Thus, the function W12 is called the 2-soliton ladder.
Each ladder step preserves its shape after its collision with the other steps. The
total step jump is 12 in the case of (4.4.29). This is the total mass. Hence we
may say that
W12(X, t) = - {:Coo U12(e, t) de
is the mass of solitons distributed in the semi-infinite interval (-00, x].
From the above solution procedure, it appears that the nonlinear superpo-
sition principle is the most important result since it is relatively easy to find
Wo, Wl and W2. It would certainly be interesting to generate a new solution
W123 ...n (denoted by W(n) from the previously known solutions wo, Wb, W n .
If U123 ...n (denoted by U(n) is an n-soliton solution (U(n) = -(a/ax)W(n), then
W(n) is called the n-soliton ladder. Wahlquist and Estabrook were the first to
provide such a general superposition principle. We present their results below.
Let Wo == 0, Wl, W2,, Wn be solutions of P2 (w) = 0 where (Wl' W2,, w n )
may be generated by Wo == 0 by using the BT like equation (4.4.20) or (4.4.21).
4.4. Backlund Transform 111
-U(x,t=-l) W(x,t=-l)
10 10
7.5
8 5
6 2.5
0 x
4 -2.5
2 -5
0_ 30
-20 -10
A0 10 20 30
-7.5
-10_ 30
-20 -10 0 10 20 30
-U(X,t=-O.l) w(x,t=-O.l)
10r----------r--------~ 10
7.5
8 5
6 2.5
0 x
4
2
-2.5
-5
-7.5
J
0_ 30 -20 -10 o 10 20 30 -1<>-'30 -20 -10 0 10 20 30
-U(X,t=O) w(x,t=O)
10r----------,----------~ 10
7.5
8 5
6 2.5
0 x
4 -2.5
2 -5
1 -7.5
-10_ 30
~30 -20 -10 0 10 20 30 -20 -10 0 10 20 30
-u(x,t=O.l) w(x,t=O.l)
10r----------r--------~ 10
7.5
8 5
6
4
2.5
o IfV x
-2.5
2 -5
-7.5
J
0_30 -20 -10 0 10 20 30 -10_ 30 -20 -10 0 10 20 30
-u(x,t=l) w(x,t=l)
10 10r----------r--------~
7.5
8 5
6 2.5
Or----------+-----+----~x
4 -2.5
2 -5 -------1-'
0_ 30
1\ -7.5
-1~~3-0---2-0----1-0--~0---1-0---2-0--3~0
-20 -10 0 10 20 30
Figure 4.6: Collision of two KdV solitons and their ladders.

The function Wj is associated with the constant >"j < 0 (j 1,2, n). = -kJ =
Let W(n) be a new solution superposed by the known solutions WI, W2,"', W n .
Here, we may understand the subscript (n) as the permutation (>"1, >"2, ... , >"n).
Then, by induction, we can derive that
W(n) = W(n-2) + 4(>"n-1 - >"n)

, n >2 (
4.4.31)
W(n-1)' - W(n-1)
where (n-1)' denotes the permutation (>"1,>"2, .. ,>"n-3,>"n-2,>"n).
Examples:
(I) Compute W123.
W123 -- W(3)
4(>"2 - >"3)
W(1) + ---'---~
W(2)' - W(2)
We denote W(1) = W1, W(2) = W12 and W(2)' = W12 and W(2)' = W13. Hence,
(4.4.32)
(II) Compute W1234.
W1234 -- W(4)
4(>"3 - >"4)
W(2) + ---'-~-...::..!.
W124 - W123
where W123 is given above and
W124 == W(3)'
4(>"2 - >"4)
W(1) + W14 - W12
.
(III) Express W123 in terms of WI, W2, and W3.
4(>"1 - >"2)
W12=WO+ , Wo==O,
W2 - W1
4(>"1 - >"3)
W13=WO+ , Wo==O.
W3- W1
Using equation (4.4.32), we obtain

>"1W1(W2 - W3) + >"2W2(W3 - w!) + >"3W3(W1 - W2)
W123 = W(3) = - >"1(W2 _ W3) + >"2(W3 - w!) + >"3(W1 _ W2) . (4.4.33)
This is the 3-soliton superposition principle and W(3) is the 3-soliton ladder.
When
>"1 = -1, W1 = -2 tanh(x - 4t),
>"2 = -4, W2 = -4coth(2x - 32t),
>"3 = -9, W3 = -6 tanh(3x - 108t).
4.4. Biicklund Transform 113
The 3-s01iton ladder given by (4.4.33) is shown in Fig. 4.7. The corresponding
3-s01iton is
(4.4.34)
and is shown in Fig. 4.4.

Generally, to get an N-soliton solution for the KdV by this procedure, one
needs first to derive a nonlinear superposition principle for the N-soliton ladders
(e.g. (4.4.33) for the 3-s01iton). Then choose AL = _2, W2k = -2(2k) coth[2kx-
4(2k)3t] and W2k+1 = -2(2k+1)tanh[(2k+l)x-4(2k+1)3t). Here, ~ 1 and
2k ::; N or 2k + 1 ::; N. For instance, for the 4-s01iton solution, we choose
A1 = -1, W1 = -2 tanh(x - 4t),

A2 = -4, W2 = -4coth(2x - 32t),
A3 = -9, W3 = -6 tanh(3x - 108t),
A4 = -16, W4 = -8coth(4x - 256t).
Then W(N) given by the nonlinear superposition principle is the N-soliton lad-
der. Therefore, the N-soliton is U(N) = -(8/8x)W(N).
Finally, we list some facts about the N-soliton for the standard KdV:
Ut - 6uu x + U xxx = 0, -00 < x < 00,
u(x, t = 0) = -N(N + l)sech 2x, -00 < x < 00.
As t --* oo, we have an asymptotic expression for u(x, t):

N
u(x, t) ex: I) -22sech2[(x - 42t - XL)]}
l=l
N
ESl(X,t). (4.4.35)
l=l
Here, Sl is the -th soliton
where Xl is the phase shift which is NOT arbitrary. The -th soliton ( =
1,2,3, ... , N) has the following properties:
Mass of the -th soliton: Me = -4.

Amplitude of the -th soliton: Al = _22.
Speed of the -th soliton: Ce = 42.
Total mass of the N-soliton: M = L~l Me = -2N(N + 1).

Recalling the BT procedure presented above, we notice that it is basically a
recursive routine. Hence, modern computers are the best tool to deal with this
-U(X,t=-0.2} W(X,t=-0.2}
20 15
17.5 10
15
12.5 5
10 0
fl i~
7.5 -5
5
2.5 -10
0 J ~
-15
-10 -5 0 5 10 -10 -5 o 5 10
-u(x,t=-0.1} w(x, t=-O.1)
20 15
17.5 10 V
15
12.5
10
5
o / x
7.5
5
2.5
0 A"
-5
-10 J
-10 -5 0 5 10 -15 -10 -5 o 5 10
-u (x, t=O) W(X,t=O)
20 15
17.5
15
12.5
10
5 /
10 0 x
7.5
5
2.5
-5
-10 )
0 -15
-10 -5 5 10 -10 -5 0 5 10
-u(x,t=O.I) w(x,t=O.l)
20 15
17.5 10
J
15
12.5 5
10 0 x
7.5
5
2.5 /' A
-5
-10 ./
J
0 -15
-10 -5 0 5 10 -10 -5 o 5 10
-u(x,t=0.2) W(x,t=0.2)
20 15
17.5 10
15
12.5 5
10 0 x
J\
7.5 -5
5
2.5 -10
0 ./ -15
-10 -5 0 5 10 -10 -5 0 5 10
Figure 4.7: Collision of three KdV solitons and their ladders.

4.5. Derivation of Inverse Scattering Method 115
recursive procedure. Appendix B at the end of this book provides a detailed

description on how to implement this BT using the symbolic computer language
called Mathematica. A symbolic package written in Mathematica has been
included in Appendix B. The reader can easily generate an n-soliton, say n =
3, simply by typing in soliton[3] on a computer after loading the package
solipac.m given in Appendix B.
Both the inverse scattering method (ISM) and the BT produce soliton so-
lutions. You may wonder what the connection between ISM and BT is. This
question was answered by H. H. Chen (1974) [16]. He derived a BT from ISM
for a KdV equation.
4.5 Derivation of Inverse Scattering Method*

This subsection gives a concise derivation of the inverse scattering method
(ISM). Using the ISM, a nonlinear IVP for a KdV on the entire line is reduced
to solving two linear problems: an eigenvalue problem for Schrodinger equation
and a linear integral equation called the GL equation. The key is to derive
the Gel'fand-Levitan (GL) equation. Our presentation here follows the same
approach as in Gel'fand and Levitan's original paper (1955) [8].
Consider the IVP for the standard KdV on the entire real line:
Ut - 6uu", + u"''''''' = 0, u(x, t = 0) = uo(x). (4.5.1)

The Miura transform between functions u and v is
v", + v2 = U - A (Riccati equation). (4.5.2)

The general Riccati equation is y' = f(x)y2+ g (x)y+h(x). The Miura transform
(4.5.2) is a Riccati equation for v for a given function u. This Riccati equation
can be linearized by a nonlinear transform:
4>",
v= . (4.5.3)
Substituting (4.5.3) into (4.5.2), we have
4>",,,, - (A - u)4> = 0, x E R, (4.5.4)

4>(oo) = 0 (for discrete spectrum). (4.5.5)
This is a scattering problem whose potential is u(x, t) and whose eigenvalue is
A. If the eigenvalue is negative (the discrete spectrum), the stationary state
corresponding to negative eigenvalue of the Schroinger equation is called the
bound state. If the eigenvalue is non-negative, the spectrum is continuous and
the state is unbounded. There is no physical meaning for the wave function
in an unbounded state. Our goal is to find u(x, t) from the scattering data 4>.
This is why the procedure is called the inverse scattering method. So, we need
to derive one more equation which conjugates with (4.5.4). This is obtained
by incorporating (4.5.4) and (4.5.1).
116 Chapter 4. 8cattering and Inverse Scattering
Step 1: If u(x, t) < 0 solves (4.5.1)' and if E!=l f~oo lu~n)12 dx < 00, then
At = 0 for any bound state (i.e. A < 0).
From (4.5.4),
(4.5.6)
In the case of the bound state, the scattering data vanishes at infinity, i.e.
=
<p(oo) 0 when A < O. The material wave is locally confined. In the case of
an unbounded state, the scattering data does not vanish at infinity. Instead,
it admits a harmonic wave at infinity, such as an incident wave <p <X exp[iv'Xx]
as x -+ 00 when A ~ o.
Substituting (4.5.6) into (4.5.1), one can eventually obtain
(4.5.7)
where
(4.5.8)
For discrete spectrum (eigenvalue) Am, the bound state has u(oo) = 0, <Pm (oo)
o.
I:
Then, the integration of (4.5.7) over (-00,00) results in
At <p~(x) dx = o. (4.5.9)
Since <Pm is an eigenfunction, f~oo <p~ dx # O. Hence
At = O. (4.5.10)
Step 2: Derivation of the equation pair for u and <P for a bound state.
If uo(x) < 0 and satisfies the integral condition, then A = AO = constant is
determined by uo(x). The spectrum of (4.5.4) - (4.5.5) is discrete when A < 0
and continuous when A ~ O. This claim is now standard and its proof can be
found in GL's paper.
From equation (4.5.10), we find that the first integral of (4.5.7) becomes
<pRx - R<px = a function of time only. (4.5.11)

This purely time dependent function vanishes at x = oo. Hence,
(4.5.12)
This further implies that

R
= C(t). (4.5.13)
This integration constant C(t) can be determined by the scattering data at 00

(u(oo) = 0 and <p(x) ,...., exp[-RxD. Since
(R/<P)(x = oo,t) = _4(_A)3/2
4.5. Derivation of Inverse Scattering Method* 117
by straightforward computation, we have
From equation (4.5.8) and the above, we obtain

t + :c:c:c - 3( u + >'):c + 4( _>.)3/2 = O. (4.5.14)
Exercise: Derive (4.5.14) for>. ;::: 0 ( unbounded state). Notice at x = 00,

,..., exp[-iJXx] + b(JX, t) exp(iJXx) and b(JX, t) is the reflection coefficient.
Therefore, we have two equations for and u. They are equations (4.5.4)
and (4.5.14). These two equations can be considered as the Fourier transform
of the following two equations with respect to y:
1/J:c:c - 1/Jyy - u1/J = 0, (4.5.15)

1/Jt + 1/J:c:c:c - 3u1/J:c - 31/J:cyy - 41/Jyyy = 0 (4.5.16)
i:
where x, y, t E Rand is the Fourier transform of 1/J(x, y, t) with respect to y
defined by
= Fy[1/J] = 1/J(x, y, t) exp(i~y) dy.
The equation pair (4.5.15) - (4.5.16) can be written in a more compact form:
M 1/J == 1/J:c:c -1/Jyy - u1/J = 0, (4.5.17)

N 1/J === 1/Jt + 83 1/J - 3u81/J = 0 (4.5.18)
where
Step 3: Equivalence of the solutions to the equation pair and that of the
IVP for the KdV (i.e. the inverse scattering).
It can be calculated by straightforward means that
(4.5.19)
Using equations (4.5.15) - (4.5.16), we find that the above equation yields
Ut - 6uu:c + U:c:c:c = O. (4.5.20)
Therefore, for any function pair (1/J, u) that solves equations (4.5.15) - (4.5.16),
u is a solution of the standard KdV (4.5.20) with u(oo, t) = 0 at any time t.
Step 4: Direct scattering problems and search for B.
Next we consider the scattering problem (4.5.14). Let 1/J = 6(x + y) be an
=
incident wave from x 00 and B(x - y, t) be the reflected wave. Then
1/J00 = 6(x + y) + B(x - y, t). (4.5.21)

It is found that if I< (x, e; t) yields

I<ee - I<:c:c + u(x, t)I< = 0, e > x, (4.5.22)
a u(x, t)
(4.5.23)
axIx,x;t) = --2-'
Ix = oo,e;t) = I<e(x = oo,e;t) = 0, (4.5.24)
then
(4.5.25)
solves equation (4.5.14) for any x, y and t. But by 1/;(x, y; t) = 0 when x+y < 0,
we have
(4.5.26)
Since 1/;00 =
o(x + y) + B(x - y, t), from equations (4.5.25) and (4.5.26) we
obtain that
B(x - y, t) + Ix, -y; t) + 1 00

Ix, e; t)B(e - y, t) de = o. (4.5.27)
Let y -t -y and e-t z. We then have the GL equation used in section 4.2:
B(x + y) + Ix, y; t) + 1 00
Ix, z; t)B(z + y, t) dz = o. (4.5.28)
Solving this GL equation, one can obtain Ix,y;t). Using equation (4.5.23),
one finally obtains the solution to the IVP for the standard KdV equation:
a
u(x,t) = -2 ax Ix,x;t). (4.5.29)
What remains is to find B(x - y, t). Since u(oo, t) 0, substituting

1/;(x, y, t) = o(x + y)
+ B(x - y, t) as x -t 00into the equation pair (4.5.17) -
(4.5.18), the function B satisfies
B:c:c - Byy = 0, (4.5.30)

B t + a3 B = 0 (4.5.31)
when x :f; y. For t > 0, Bo(x, y) == B(x, y, t = 0) is obtained from the scattering
problem (4.5.15), i.e.
Bo:c:c - Boyy - uo(x)Bo = O. (4.5.32)
The solution to this scattering problem is
Bo(x, y, t = 0) = L:
N
C;(O) exp[-km(x -
m=l
y)] + -1
211"
100
-00
R(k) exp[ik(x - y)] dk
(4.5.33)
4.6. Soliton Fission 119
where km = V-Am, Cm(O) = limx-+oo tPOm(x) exp(kmx), and tPOm(x) is the

eigenfunction of
tPOmxx + (k;" - uo(x))tPOm = 0, m = 1,2,3, "', N. (4.5.34)

This step is not trivial. The details of the derivation can be found in GL's
paper.
With the initial condition B(x, y, t = 0) given by (4.5.33), equation (4.5.31)
can be solved for t > 0:
N
L
1:
B(x, y, t) 'Ym exp[-km(x - y) + 8k!t]
m=1
+ 2~ R(k) exp[ik(x - y) + 8ik 3 t] dk. (4.5.35)
This step is not trivial either and can also be found in GL's paper. Since
the variables x and y always appear in the form x - y, the kernel B(x, y, t) is
more directly denoted by B(x - y, t) as in the GL equation.
4.6 Soliton Fission

In this section, we study soliton solutions of KdV equations with variable co-
efficients. Let us consider the problem of a KdV equation
1 1/2
u 8 + '32 d- 7/4 uu{ + 6d ueee = 0, 0 < e< 00, s:f. 0, (4.6.1)
u(e, s = 0) = uo(e) = A sech 2 (,ae) (4.6.2)
where A and (3 are constants, d is a piecewise constant function of s defined by
d = d(s) ={ 1, ~f s < 0, (4.6.3)

do < 1, If s > O.
This problem models a soliton surfing from one uniform section of a two di-
mensional channel to another uniform section with a small transition region
at s = 0 (see Johnson (1973) [7]). We want to see how many solitons will be
generated from a given initial soliton. That is, we want to find aN-soliton
solution of (4.6.1) for s > 0 with the initial single soliton condition (4.6.2).
We do this by using the inverse scattering method. By changing variables,
(4.6.1) can be transformed to the standard KdV equation for s > O. Then the
associated eigenvalue problem (4.2.8) is
tPee + [3: d~9/4uo(e) + A] tP = O. (4.6.4)
Let u = tanh e, then (4.6.4) becomes

~ [ _ 2 dtP] [~-9/4 A{32 + {32 (1 A
du (1 u) du + 2 d
]_
_ ( 2 ) tP - O. (4.6.5)
This is the generalized Legendre equation. The discrete eigenvalues are given
by
Here m and n are integers. Each of these discrete eigenvalues corresponds to a

soliton as s -t +00, while a continuous spectrum of (4.6.5) corresponds to an
oscillatory wave train.
Thus,
d-;9/4j32 A = ~n(n + 1), n = integer (4.6.6)
is the necessary and sufficient condition for teh existence of an n-soliton solution
of (4.6.1) - (4.6.2) on the channel section of depth do. This process of one soliton
disintegrating into several solitons is called the soliton fission.
After the fission, the solitons have their amplitudes
2Am 2
(4.6.7)
Am = n(n+ 1)'
and the propagation speeds
Vm = 2Am, (m= 1,2,,n). (4.6.8)
The initial condition (4.6.2) should agree with the one-soliton solution of the
equation (4.6.1) for s < O. Hence
(4.6.9)
Thus,
(4.6.10)
According to (4.6.6),
_ [n(n +
do - 2
1)] -4/9
' n = 1,2,3,. (4.6.11)
This is the final form of the fission condition. Fission occurs when n > l.
This requires do < 1 by the fission condition (4.6.11). Physically, this fission
condition implies that a single soliton disintegrates into several solitons only
when it surfs from a deeper section of a two-dimensional channel to a shallower
section. However, this conclusion is not applicable to a channel of elliptical
cross section or when a current is present.
4.6. Soliton Fission 121
[1] C. S. Gardner, J. M. Greene, M. D. Kruskal and R. M. Miura (1974),

Korteweg-de Vries equation and generalizations. VI. Methods for exact
solution, Comm. Pure Appl. Math. 27, 97-133.
[2] G. L. Lamb (1980), Elements of Soliton Theory, John Wiley, New York.
[3] R. K. Dodd, J. C. Eilbeck, J. D. Gibbon and H. C. Morris (1982), Solitons
and Nonlinear Wave Equations, Academic Press, New York.
[4] R. M. Miura (1976), The Korteweg-de Vries equation: A survey of results,
SIAM Review 18, 412-459.
[5] P. D. Lax (1968), Integrals of nonlinear equations of evolution and solitary
waves, Comm. Pure Appl. Math.21, 467-490.
[6] L. Landau and E. Lifschitz (1958), Quantum Mechanics, Nonrelativistic
Theory, Pergamon Press, New York.
[7] R. S. Johnson (1973) , On the development of a solitary wave moving over
an uneven bottom, Proc. Camb. Phil. Soc.73, 183-203.
[8]1. M. Gel'fand and B. M. Levitan (1955), On the determination of a dif-
ferential equation from its spectral function, Amer. Math. Soc. Transl.
Ser. 2, 253-304.
[9] C. Rebbi and G. Soliani (1984), Solitons and Particles, World Scientific
Publishing Co., Singapore.
[10] P. G. Drazin and R. S. Johnson (1989), Solitons: An Introduction, Cam-
bridge University Press.
[11] P. G. Drazin (1983), Solitons, Cambridge University Press.
[12] J. L. Hammack and H. Segur (1974), The Korteweg-de Vries equation
and water waves, Part 2. Comparison with experiments, J. Fluid Mech.
65, 289-314.
[13] A. C. Newell (1985), Solitons in Mathematics and Physics, SIAM, Philadel-
phia, Pennsylvania.
[14] N. J. Zabusky and M. D. Kruskal (1965), Interaction of "solitons" in a
collision less plasma and the recurrence of the initial states, Phys. Rev.
Lett., 15, 240-243.
[15] H. D. Wahlquist and F. B. Estabrook (1973) , Backlund transformation
from solitons of the Korteweg-de Vries equation, Phys. Rev. Lett. 31,
1386-1390.
[16] H. H. Chen (1974), General derivation of Backlund transformations from
inverse scattering problems, Phys. Rev. Lett. 33,925-928.
[17] M. J. Ablowitz and H. Segur (1981), Solitons and Inverse Scattering

Transform, SIAM, Philadelphia, Pennsylvania.
[18] T. Kuusela, J. Hietarinta, K. Kokko and R. Laiho (1987) , Soliton ex-
periments in a nonlinear electrical transmission line, Eur. J. Phys. 8,
27-33.
Chapter 5
Burgers Equation
In this chapter, we study the initial value problem of Burgers equation on

the entire real line. The unknown of the Burgers equation models the first
order elevation of the free surface of viscous fluid flow down an inclined plate.
The Cole-Hopf transform can convert the nonlinear Burgers equation into a
linear heat equation. Hence the initial value problem for the Burgers equation
can be solved analytically. Analytic solutions to the Burgers equation for two
different initial conditions are found. These solutions are the Burgers shock
waves and the triangular waves respectively. The Burgers shock waves have
a jump discontinuity when the viscosity approaches zero and they are stable.
This stability claim will be proved in section 5.3. But the triangular wave is
not stable and is only a transient state.
123
124 Chapter 5. Burgers Equation
Viscous fluid x
Figure 5.1: Configuration of a viscous fluid flow on an inclined plate
5.1 Viscous Fluid on an Inclined Plate
Consider a viscous fluid of constant density flowing on an inclined plate (see

Fig. 5.1). The governing equations are the Navier-Stokes equations (see section
2.1.2):
U;. + v;. = 0, {5.1.1}

*
Ut. + U * u z* + v * u y* = 1 *
--Pz. . ()+ II~
+ gSln A* u * , {5.1.2}
p
v t* +U * v z* +v * v y* = --Py.
1 * -gcos () +II~
A* V* , (5.1.3)
P
with the boundary conditions as follows (see the last section of this chapter):
on the free surface y* = '1*(x*,t*},
(5.1.4)
(5.1.5)
(5.1.6)
and on the plate y* =- H ,

u* = v* = 0 (no slip condition). (5.1.7)
Here 11 = III p is called the kinematic viscosity, Il is called the viscosity, and
H is the upstream uniform depth. The difference between viscous fluid flows
and inviscid fluid flows is that the viscous flows have: (a) viscous deceleration
II~ *u* in the entire flow field; (b) dynamical surface condition that includes
the viscous traction; and (c) no slip condition. When the viscosity is small, the
viscous deceleration does not plays an important role in the fluid motion, yet
boundary conditions, particularly the no slip condition on a rigid boundary,
5.1. Viscous Fluid on an Inclined Plate 125
greatly affect the fluid motion behavior. The most distinguished feature of
the fluid motion is that there is a large velocity gradient at the boundary.
This gradient in turn generates strong vorticity that induces turbulence. The
complicated flow phenomena occurring at a thin layer offluid near the boundary
have to be understood when designing a turbine and other fast moving objects.
The special theory that studies this viscous influence on fast fluid flows is called
the boundary layer theory.
To nondimensionalize the above equations, we introduce the following di-
mensionless variables
(u, v) = (u*, L/Hv*)/Viii, (x,y) = (H/Lx*,y*)/H,

p../illH ( Transport effects )
R = J.t
Reynolds number '" D'ff' ffi
I USlve e ects
'
t (H2/L2)t*/VH/g, p=p*/pgH, TJ=TJ*/H.

Let
O<f=H/Ll
be the small number in our asymptotic analysis. In the above, L is the length
scale along the x-axis, which may be a typical wave length. Hence f 1 is
the long wave assumption. This is equivalent to assuming that the vertical
acceleration can be ignored. Violent fluid motions, such as the liquid spilling,
are directly induced by vertical acceleration. If the vertical acceleration of the
fluid is absent, the motion of the fluid is smooth and calm: a laminar flow. The
aforementioned Reynolds number is the most important factor that determines
whether a flow is smooth (laminar) or turbulent. Since the transport effect
excites the chaotic motion of fluid particles and the diffusion effect suppresses
the excited chaotic state to a calm state, the large Reynolds number implies
ultimately turbulent motion and small ReynolCls number, on the other hand,
implies a laminar motion of fluids. The property of a fluid and the domain
in which this fluid moves determine the value of the Reynolds number R. For
instance, in an atmospheric flow, R is very large and the flow is usually turbu-
lent. For a high speed water jet in an open channel, the flow becomes turbulent
when the Reynolds number R is large (say, R > 2000). For the same water
flow in the same channel, if the flow speed is low, then the flow is laminar and
the Reynolds number R is small (say, R < 1000).
The dimension ofthe kinematic viscosity v is [length]2[time]-I. The Reynolds
number can be written as
R= ../illH .
v
To get some feeling about the viscous properties of fluids, we list the values
of some common fluids under the condition of 1 [atm] (= 1013 [millibars]
or 1.013 [kg][cm]-2 ) pressure and 20 C temperature: v = 4.22 X 10- 7 for
gasoline, 1.01 x 10- 6 for water, 1.5 x 10- 5 for air, 3.25 x 10- 4 for SAE30 oil
and 1.18 x 10- 3 for glycerin, where v has the dimension [meter]2[second]-l.
As an example of computing the value of the Reynolds number, we suppose
an SAE30 oil flow of depth H = 0.2[m] at a velocity U near VgH. Then the
Reynolds number is
= ",,9.8 x 0.2 x 0.2 = 861
R 3.25 x 10-4 .
Since the Reynolds number is less than 1000, this flow is most likely laminar.
It is clear that every fluid found in nature has viscosity. In certain cases,
the viscosity can be ignored, such as the case of shallow water waves discussed
in Chapter 3. But for the same water, if it flows in a thin pipe at a relatively
high speed, the viscosity of water must be considered.
In terms of dimensionless variables, equations (5.1.1) - (5.1.7) become:
Navier-Stokes equations
Uo: + Vy = 0, (5.1.8)
f2Ut + f( uU x + vU y ) = -fPo: + sin 0 + R- 1 (f2uo:o: + U yy ), (5.1.9)
f3 Vt + f2(uvo: + vV y ) = -Py - cos 0 + R- 1 (f3 vxx + fV yy ), (5.1.10)
on the free surface y = .,,(t, Xj f)
f"'t + u"'o: - v = 0, (5.1.11)

f(Rp - 2wo:)."o: + Uy + f 2 Vx= 0, (5.1.12)
Rp - 2fVy + f(f 2Vx + Uy)"'x = 0, (5.1.13)
and on the flat plate y = -1

U = v = O. (5.1.14)
The simplest motion described by the above equations is the fluid flow
whose free surface is flat (i.e., .,,(x, t) = 0). In this case, the pressure satisfies
the hydrostatic law (p = -y cos 0), the y direction velocity vanishes (v = 0),
and the x direction velocity satisfies a parabolic distribution
Uo = (RsinO/2)(1- y2).
The gravitational force overcomes the viscosity resistance and maintains this
stationary motion. If 0 = 0, the gravitational force is perpendicular to the resis-
tance force and hence there is no push on the fluid to overcome the resistance.
Consequently, the fluid is at rest.
Recall that for an inviscid fluid flow in a horizontal two-dimensional channel,
a trivial solution has a flat free surface and hydrostatic pressure but an indef-
inite uniform flow velocity, because there is no need of a force that overcomes
the viscous resistance force. Hence, an inviscid channel flow can have infinitely
many trivial solutions when, of course, the molecular friction between the fluid
and the boundaries has been ignored. But the viscosity imposes a constraint
on the fluid motion so that for a trivial solution, although the free surface can
still be flat and the pressure can be hydrostatic, the longitudinal velocity must
satisfy the unique parabolic distribution law. Therefore, the trivial solution of
viscous fluid flow is unique. It can be proved that this unique trivial solution
Viscous fluid x
Figure 5.2: The velocity field of the trivial flow of a viscous fluid down an
inclined plate.
is stable and so can be observed in a laboratory. See Fig. 5.2 for the flow field.
More interesting solutions are those perturbed from this trivial solution.
One type of non-trivial solutions is described by the Burgers equation. The
Burgers equation is an equation satisfied by the first order free surface elevation
quantity 'TJ1 in an asymptotic expansion. The asymptotic expansion is assumed
to have the following form:
(u,V,P,'TJ) = (UO(y),O,Po(y), 0)
+ c(U1(X, y, i), V1(X, y, i),P1(X, y, I), 'TJ1(X, i))
+ c 2(U2(X, y, i), V2(X, y, i),P2(X, y, i), 'TJ2(X, I))
+ 0(c3 ). (5.1.15)
Substituting (5.1.15) into (5.1.8)-(5.1.14), and assembling the terms according

to the like powers of c, we have
Order cO : (The trivial solution)
R-1uoyy + sin B = 0, (5.1.16)

POy + cos B = 0, (5.1.17)
= = on y = 0, (5.1.18)

UO y 0, Po
Uo = on y =-1. (5.1.19)
Order c 1 :
U1x + V1y = 0, (5.1.20)

U1yy 0,= (5.1.21)
Ply = 0, (5.1.22)
on y = 0,
UO'TJ1x - VI = 0, (5.1.23)
Ul y + UOyy'1l = 0, (5.1.24)
POy'1l + PI = 0, (5.1.25)
on Y = 1,
VI = 0. (5.1.26)
Order f2 :
U2., + V2y = 0, (5.1.27)

=
UOUt.. + VI Uoy -PI., + R- 1U 2yy, (5.1.28)
-P2y + R-1Vl YY = 0, (5.1.29)
on Y = 0,
'1lt + UI'1l., + U0'12., = V2 + Vl y '1lo (5.1.30)
U2y + Ulyy'1l + UOyy'12 = 0, (5.1.31)
RP2 - 2Vly = 0, (5.1.32)
on Y = -1,
(5.1.33)
Those terms of order f3 or higher are omitted. From the equations of order
fO, we have
Uo = ~R(l - y2) sin () (parabolic velocity distribution), (5.1.34)

Po = -ycos() (hydrostatic pressure). (5.1.35)
This is the trivial solution. From the equations of order fl, we have
Ul R(l + Y)'1l sin(), (5.1.36)
VI = -~R(1+y)2'1h:sin(), (5.1.37)
PI = '11 cos (). (5.1.38)
These quantities allow all the equations (5.1.20) - (5.1.26) to be satisfied except
equation (5.1.23).
The solutions of the equations of order fO and order fl, given by (5.1.34) -
(5.1.38). Equations (5.1.28), (5.1.30) and (5.1.33), yield
-1 < y < 0, (5.1.39)

R'12 sin () on y = 0, (5.1.40)
o on y =-1. (5.1.41)
Straightforward integration of these equations gives
U2 = R{ [~(y2 - 1) cos ()
+R2(~: -~ + 254)]'11.,+ (1+Y)'12sin() }. (5.1.42)

Integrating (5.1.27) with respect to y from -1 to 0 and making use of (5.1.39),

(5.1.30), (5.1.33) and (5.1.42), we obtain
71lt + 2Rsin () 711711:1: + ~ (-5 cos () + 2R2 sin2 ()) 71111:11: = O. (5.1.43)
This is the partial differential equation which determines the first order ap-
proximation to the elevation of the free surface. By changing variables, this
equation can be written in the form
(5.1.44)
This is the so called Burgers equation, which can be solved analytically by

the Cole-Hopf transformation. In this book, we also call (5.1.43) a Burgers
equation.
When the Burgers equation serves as the governing equation for the first
order free surface elevation 711, the upstream velocity profile is
u = [R(I- y*2 I H2) sin (}12]J9Ii +

[(HI L)R(1 + y* I H)71il H sinO]J9Ii + O((HI L)2).
The upstream velocity must be around the profile
and its perturbation is related to the solution 71i of the Burgers equation and
in the order of HI L. This is similar to the case of the Korteweg-de Vries
equation (KdV) discussed in Chapter 3. In a KdV model, the perturbation
of the upstream velocity about the critical speed of shallow water wave is in
the order of (HI L)2 and also related to the solution of the corresponding KdV
equation.
Like the KdV equation, the Burgers equation also describes many interest-
ing wave phenomena, such as the Burgers shock waves, triangular waves and
N waves. We will discuss only shock waves and triangular waves in this book.
Similar to the heat equation, the problem becomes ill-posed if the coefficient
of 7]111:11: in (5.1.43) is positive. To avoid this, we have the following constraint
R R
<e
= (52cot' (())}t' (5.1.45)
sm
Here, Re is called the critical Reynolds number. The coefficient of 7]111:11: in
(5.1.43) vanishes when R = Re. In fact, R < Re is the stability criterion for
the Burgers equation. We will show this in section 5.3.
Fig. 5.3 shows the relationship between the critical Reynolds number Re
and the inclined angle () when the angle () is in the range of 1 to 15 degrees.
We see that the value of Re quickly falls below 40 when () is larger than 3
degrees. Recall the example of the computation of Reynolds number earlier in
this chapter. We can find out that for a relatively large depth and for most
Rc
100
80
60
40
20
theta
o 2 4 6 8 10 12 14
Figure 5.3: The relationship between the critical Reynolds number Rc and the
inclined angle ().
common fluids that are familiar to us, the corresponding Reynolds numbers
are much larger than 40. Hence only very viscous fluid flow or a very thin fluid
layer on an inclined plate is stable.
Let us look at a few concrete numbers to get some feeling about when a
viscous fluid flow on an inclined plate is laminar and stable. Suppose () =
l[degree]. Then Rc = 91. For the very viscous fluid glycerin (II = 1.18 X
10- 3 [m]2[sJ-l ), the relation @H/II = Rc = 91 results in H = 105.6 [mm].
Hence, for a glycerin flow on a plate inclined at one degree to be stable, the
fluid depth must be less than 105.6 [mm]. For the same inclined plate, the
maximum depths of stable laminar SAE30 oil flow (II = 3.25 X 10- 4 [m]2[s]-l
) and water flow (II = 1.01 X 10- 6 [m]2[s]-1 ) are 44.7 [mm] and 1.0 [mm]
respectively.
The smaller the Reynolds number, the calmer the fluid motion. So the
Burgers' equation describes a class of calm motions of viscous fluids down an
inclined plate with moderate Reynolds number.
For a fluid flow to be stable on an inclined plate, the fluid must have high
viscosity. For example, when () = 3 and 10, the corresponding values of Rc are
30 and 9 respectively. At these Reynolds numbers, for a fixed finite upstream
depth and the upstream velocity, the fluid must have high viscosity. Another
way to make the fluid flow stable on an inclined plate is to reduce the thickness
of the fluid layer. In this case, the transport effect in the fluid motion is greatly
reduced. Accordingly, the Reynolds number reduces dramatically and hence is
below the critical value R e .
In section 5.2, we will use the Cole-Hopftransform to solve a general IVP for
the Burgers equation. A very interesting solution which is related to the shock
5.2. Cole-Hopf Transformation 131
wave associated with a hyperbolic conservation law is a traveling wave that

has a smooth jump. This solution is called the Burgers shock wave. In section
5.3, we prove that such a Burgers shock wave is stable. Another interesting
solution is the evolution of a very narrow single hump of the free surface. This
hump grows to a triangular shape, and it finally dissipates and approaches the
trivial solution. Thus, the triangular wave is not stable.
5.2 Cole-Hopf Transformation

Consider the initial value problem for the Burgers equation
Ut + UU", = jJU",,,,, -00 < x < 00, (5.2.1)

u(x, t = 0) = uo(x). (5.2.2)
There is a remarkable nonlinear transform that can reduce the nonlinear prob-
lem (5.2.1) - (5.2.2) to a linear problem. This transform is known as the
Cole-Hopf transformation (Cole (1951) (51, Hopf (1950) U1, and also see A.
R. Forsyth (1906), Theory of Partial Differential Equations, Part IV - Partial
Differential Equations, vol. VI, p.l0l). Even though Burgers' work on the
derivation of the Burgers equation did not appear until 1948, this equation
was already solved much earlier. The evidence can be found in Forsyth's book
referred to above.
Formally the Cole-Hopf transform is similar to the nonlinear transform for
Riccati equation which is an ODE with a second order nonlinearity:
y' = a(x)y2 + b(x)y + c(x), a(x) # O.

A nonlinear transform
w' 1
y=---
w a(x)
can cancel the second order nonlinearity so that the unknown function w sat-
isfies a second order linear ODE
w" - [a'(x)ja(x) + b(x)]w' + a(x)c(x)w = O.

As an example, consider
y' = e'" y2 - Y + e'" .
Let
y = -(w'jw)e-"'.
Then w(x) satisfies
w" + w = O.
This ODE has a general solution
132 Cha.pter 5. Burgers Equa.tion
So the general solution for the original ODE is

C 2 sinx - C 1 cos X -z
y= e
C 1 sin x + C 2 cos x
We can divide the numerator and the denominator in the above expression
by a non-vanishing C 1 or C2 so that there is only one free parameter as the
arbitrary integration constant.
Now, we come back to the Burgers equation (5.2.1) and rewrite it as
Ut + (~u2) z = JJUzz
This equation is defined on the (x, t)-plane. There are two directions: spatial
direction x and evolution direction t. When we look at the evolution direction t
only, this is a first order differential equation with a second order nonlinearity.
Perhaps one would naturally think that the nonlinear transform for the Riccati
equation may linearize this nonlinear PDE. As a matter of fact, the transfor-
mation does exactly that. This transformation is now known as the Cole-Hopf
transformation defined by
U = -2JJ IfJz . (5.2.3)
IfJ
Substituting (5.2.3) into (5.2.1) we have
(IfJt - JJlfJzz )lfJz = (IfJt - JJlfJzz)z 1fJ (5.2.4)

Exercise: Derive this equation from (5.2.1) and (5.2.3).
Therefore, if lfJ(x, t) solves
- 00 < x < 00, (5.2.5)

then u(x, t) computed from the Cole-Hopftransform (5.2.3) solves the Burgers
equation (5.2.1).
The initial value problem for the heat equation (5.2.5) can be solved ana-
= =
lytically. Let lfJ(x, t 0) lfJo(x) be the initial condition. Taking the Fourier
transform with respect to x for both equation (5.2.5) and the initial condition
lfJo(x), we obtain
and
<p(t = 0) = <Po
where <p = J~oo lfJ(x, t) exp(iwt) dx is the Fourier transform of lfJ(x, t) with
respect to x and is a function of wand t. The solution for this first order ODE
problem is
<p = <poexp(w 2 JJt).
The inverse Fourier transform recovers lfJ(x, t):
where * denotes the convolution product and F-l the inverse Fourier trans-
form. This equation has a quite nice algebraic structure. It reveals clearly how
the initial data CPo evolves according to the evolution factor F-l[exp(w 2 Jtt)].
Because of this interesting algebraic structure, an algebraic method using semi-
groups has been developed to study parabolic PDEs.
Equation (5.2.3) can be written as
U
a
= -2Jt ax logcp. (5.2.6)
JU~~
Hence
cp(x, t) = exp ( - t) dX)' (5.2.7)
Since a constant multiplier of cp does not affect u by (5.2.6), we may write

(5.2.7) as
(5.2.8)
The initial condition (5.2.2) should be transformed to an initial condition

for (5.2.5) by using (5.2.8).
cp(x, t = 0) == cpo(x) =exp ( -1'" u~~) dU). (5.2.9)
Note that
F-l[exp(w 2 Jtt)] = v'4~Jtt exp (- :;t) .
So, the linear initial value problem (5.2.5) and (5.2.9) has an analytic solution
which can be found by using the technique of Fourier transform:
1
cp(x, t) = 2v'rrJtt
1-00
00
CPo (e) exp (-

(x-e)2
4Jtt ) de (5.2.10)
From (5.2.3), we in turn have an analytic solution for the nonlinear problem
(5.2.1) - (5.2.2)
(5.2.11)
When Jt approaches zero (vanishing viscosity), u",(x, t) may approach infinity

at some point for a certain initial condition. This is like a solution to the hyper-
bolic conservation law Ut + UU'" = O. For more details about the conservation
law, see Whitham (1974) [8], p.98-101 and Chapter 2 of this book.
Exercise: Find the necessary derivatives of u(x, t) to verify that (5.2.11)
solves the problem (5.2.1) - (5.2.2).
Hint:
as t -t 0+
and
as t-tO+.
Next we present two examples: the Burgers shock wave and the triangular
wave. These examples will show how the Cole-Hopf transform works.
Example 1: Burgers shock wave
Consider the initial value problem of the Burgers equation (5.1.43) with the
initial condition
(5.2.12)
We assume that R < Rc and let

x
z= . (5.2.13)
2Rsin ()
Then (5.1.43) becomes the standard Burgers equation like (5.2.1):
"I1t + "I1"11z = m"llzz (5.2.14)

where
1
m = 60R(5cot 8 csc 8 - 2R2) > O. (5.2.15)
From (5.2.9),
<po(z) exp (-1o Z

f(u)/2m dU)
{ exP(-lz/2m) if z<O, (5.2.16)
if z ~ O.
Using the expression (5.2.11), we obtain
1
(5.2.17)
"I1(X, t) = 1 + h(z, t) exp(Z;~e)
where
h(z, t) =
erfc(-z/v'4fflt) . (5.2.18)
erfc((z - t)/V4mt)
In the above, erfc is the complementary error function defined by
erfc(s) = 100
exp(-x 2 ) dx.
As t -t 00, h(z, t) -t 1/2. Hence,

1
7]1 (x, t) '" 1 1 (Z-t/2) as t -t 00. (5.2.19)
+ 2 exp 2m
Figure 5.4: Burgers shock wave (see equation (5.2.19)).
This is a smooth "shock" called the Burgers shock wave. When m = 0.4, the
evolution of the initial data is shown in Fig. 5.4.
Exercise: Derive (5.2.17) by using formula (5.2.11). Hint: Verify the
following relations:
<po(e) = { ~xp(-e/(2m)) if e> 0,

if e~ OJ
and
For a fixed to, we can find the following limit values:
lim 7]l{X, to) = 0 and lim 7]l{X, to) = 1. (5.2.20)

"'-+00 "'-+-00
This implies that the initial profile of the free surface at infinity is not affected
by the wave propagation in a finite time. This is plausible since the wave front,
which is initially at x = 0, propagates at a finite speed. In practical applica-
tions, the heat equation describes the strength of the molecular diffusion while
the Burgers equation describes some other phenomena, different from the heat
transfer process, such as the free surface wave propagation in viscous fluid flow
on an inclined plate. For a heat transfer process, the temperature of a con-
ductor eventually becomes uniform through out the entire conductor, yet the
Burgers equation can represent the maintenance of a permanent difference of
two uniform states within one fluid domain (i.e. Burgers shock wave (5.2.19)).
The Burgers shock wave is a traveling wave solution and can be found by
a simple method: the traveling wave approach. Consider the traveling wave
solution to the Burgers equation
(5.2.21)
with the boundary conditions
7]l{-oo,t) = 1 and 7]l(OO,t) =0. (5.2.22)
Let Z =z - st and 7]l(Z,t) = ({Z). Then (5.2.21) becomes
- s(' + ((' = m(". (5.2.23)
The first integral of this equation is
(( ( - 2s) = m(' + C
where C is the integration constant. When Z = 00, we have ( = (' = 0 and

hence C = O. When Z = -00, we have ( = 1, (' = 0 and hence s = 1/2. Thus
we get
((( - 1) = 2m('. (5.2.24)
Further integration of this equation gives
1-( z
In-(- = -2m'
Solving for (, we obtain
1
7]l{Z, t) =
1 + exp
(Z-t I 2)
2m
(Burgers shock wave). (5.2.25)
This is a traveling smooth "shock" wave which is similar to solution (5.2.19)

found by the Cole-Hopf transform when t -t 00. The only slight difference is in
the transition shape because there is no 1/2 factor in denominator of (5.2.19).
This difference is caused by the different initial conditions for the two solutions.
The jump in these solutions becomes sharper when m gets smaller. We will
show in the next section that these Burgers shock waves are stable.
Example 2: Triangular wave

Another interesting phenomenon that can be described by the Burgers equa-
tion is the triangular wave. Initially, let the free surface be a single narrow
hump which can be approximated by a Dirac delta function due to the long
wave assumption. Under the action of the gravitational force, the peak of the
hump will fall and push the base of the hump downstream and hence drag a
tail upstream. So the free surface in the hump region becomes a triangular
shape. This shape can be found analytically by solving an IVP for the Burgers
equation.
The initial condition is
771(Z, t = 0) = Ac5(z). (5.2.26)

To compute tpo(e), we regard the lower limit of the integration (5.2.9) to be 1
instead of 0 since this lower limit is arbitrary anyway. Hence
if e> 0, (5.2.27)
if e::; 0,
Then, from formula (5.2.11) the triangular wave is
~ exp(-(2)
771 (z, t) = V-;;: erfc() + 2/(exp(-y) _ 1) (5.2.28)
where
z
and (= ..J4mt (5.2.29)
The evolution of this triangular wave for m = 0.04 and A = 1.0 is shown in Fig.
5.5. The instantaneous shape of the same triangular wave at t = 2 is shown in
Fig. 5.6. In this figure, a clear triangular shape is displayed.
Recall that the Burgers shock wave is a traveling wave solution for a large
time. Traveling wave allows a reduction procedure that reduces a PDE to an
ODE. It is interesting to try another common reduction procedure: similarity
solution.
To find a similarity solution for a PDE, one should begin with dimension
analysis by examining the equation itself or another dimensionalized equation
of the same mathematical structure. These equations to be examined must
be expressed in terms of dimensional quantities. For the Burgers equation,
the most handy equation is the one-dimensional Navier-Stokes equation (5.1.2)
when () = 0:
u;. + u* u;. = IIU; ., .
Figure 5.5: Evolution of the triangular wave (see equation (5.2.28)).
0.8
-3 -2 -1 3
Figure 5.6: The triangular wave at t = 2 (see equation (5.2.28)).

Let the initial condition be
u*(x*,t* =0) =AJ(x*).
Then, v and A have the same dimension: [lengthj2[timej-1. From the dimen-
sional quantities in the above two equations, one can construct a few most basic
dimensionless quantities:
A
and
v
There must be a definite relationship among these three dimensionless quanti-
ties. Since A/v is a constant, it must serve only as a parameter. Hence there
is the following relationship:
where F represents a functional relation. When taking square root, this relation
is reduced to
where f is a new unknown function which is going to satisfy an ODE. In this

way, we have set up the frame of similarity solution.
To find similarity solutions of Burgers equation TJ1t + "11 "liz = mTJ1zz, let
"11 = f!f f(()'

z
(= ..J4mt
(5.2.30)
From this assumption, we can also derive the triangular wave (5.2.28).
Substituting (5.2.30) into the Burgers equation (5.2.1), one obtains that
f(() + (f'(() - f(()!'(() + ~!"(() = 0, (5.2.31 )

f(oo) = O. (5.2.32)
The boundary condition comes from the fact that the wave propagates at a
finite speed and hence the infinity is not disturbed. The above boundary value
problem has a first integral
!'(() - f2(() + 2(f(() = O. (5.2.33)
This is a Riccati equation which can be solved analytically by a nonlinear

substitution:
Wi
f(() = --. (5.2.34)
w
From (5.2.33), one can derive an equation for w:
wI!
-+2(=0.
Wi
Integrations of this equation result in

w' = C 1 exp(-(2), (5.2.35)
w = -C1 erfc() + C2 (5.2.36)
Hence,
f() = exp(-(2) (5.2.37)
erfc() - C
where C = C2 /C1 is the arbitrary integration constant for the Burgers equation
and is determined by the initial condition.
The initial condition '71(Z, t = 0) = A6(z) implies that
lim
t-+O+
/1-1
'71(Z, t) dz = A.
i:
Integration by substitution yields
A = 2m f() d(
= _2m/
00 l
w d(
-00 w
2mln j(C2 - 2Cd/C21.
From this equation, we find that
2
C = 1 _ exp(-y) . (5.2.38)
So the constant C in (5.2.37) is determined. Using equation (5.2.30), we have

found the triangular wave (5.2.28) as a similarity solution.
The triangular wave is not stable and gradually decays to the trivial solution
of a flat free surface (see Fig. 5.5).
5.3 Stability of the Burgers Shock Wave

The Burgers equation
Ut + UUzo = I'Uzozo , I' >0 (5.3.1)
has a steady state solution of the form
uo(z) = -H tanh ( ~:), H > o. (5.3.2)
Because lim uo(z) = TH and the transition interval is short when I' is small,
zo-+oo
the solution (5.3.2) looks like a shock wave. Actually, this is the same Burgers
shock wave as that found in the last section when H = 1/2 in (5.3.2) and '71 is
shifted to '71 - 1/2 in (5.2.25). We want to examine the linearized stability of
the steady state solution (5.3.2). The following theorem holds.
5.3. Stability of the Burgers Shock Wave 141
Theorem 5.1 The Buryers shock wave as either asymptotically stable or neu-
trally stable with respect to the infinitesimal disturbance.
Proof:
Following the conventional linear stability theory, we superimpose an in-
finitesimal disturbance v(z, t) upon the steady state solution uo(z). Let
u = uo(z) + v(z,t), IIv(,t)lIoo lIuolioo. (5.3.3)
Substituting (5.3.3) ~nto (5.3.1) and linearizing it with respect to v around uo,
we obtain the following equation for v:
Vt + UoV., + UO.,v = /JV.,.,. (5.3.4)
Let
v(z, t) oc l(z)e 17t , u = constant.
Then I satisfies
/J/.,., - uo/., - (UO., + u)1 = o. (5.3.6)
Since the disturbance is small and does not affect the boundary condition at
infinity in a finite time interval, we have
I(z = oo) = O. (5.3.7)
The boundary value problem (5.3.6) - (5.3.7) is an eigenvalue problem with

eigenvalue u and eigenfunction I. There are three cases for u.
Case (i): Re(u) > O. Then v(z, t) grows indefinitely in time t. Thus the
solution uo(z) is unstable.
Case (ii): Re(u) < O. Then v(z, t) decays to zero exponentially in time t.
Thus the solution uo(z) is asymptotically stable in the Lyapunov sense.
Case (iii): Re(u) = O. Then v(z, t) is bounded. Thus the solution uo(z) is
neutrally stable.
Next we will show that case (i) does not occur. Let
y = (Hz)j(2/J), a = (4/Ju)jH2. (5.3.8)
Then equations (5.3.6) and (5.3.7) become
f" + 2(tanh y)f' + (2 sech 2 y - a)1 == L(f) = 0, (5.3.9)

f'(oo) == 0 (5.3.10)
where f' = (dl)j(dy). Since the differ~ntial operator is even in y"both od<,l and
even functions of y can separately be solutions of (5.3.9). An even function is
independent from an odd function. Thus the following two special solutions of
(5.3.9) - (5.3.10) are linearly independent from each other
{ lodd = sechy . [k sinh(ky) - tanhy cosh(ky)] , (5.3.11)

leven = sechy . [k cosh(ky) - tanhy sinh(ky)] ,
where k = ..;r+Q. The first one is odd and the second one is even. Since (5.3.9)
is a second order linear equation, its general solution is a linear combination of
two linearly independent special solutions.
So, when either k f. 1 or k f. 0 (i.e., either 0: f. 0 or 0: f. -1), the neces-
sary and sufficient condition for eigenfunctions (5.3.11) to satisfy the boundary
condition (5.3.10) is that Re(k) < 1. This implies that Re(o:) < 0, and further
implies Re(u) < O. Therefore the Burgers shock wave uo(x) given by (5.3.2) is
asymptotically stable. Thus, this is case (ii).
If k = 1, then
fodd = tanh y + y sech 2 y and feven = sech 2 y. (5.3.12)
These are two independent solutions of (5.3.9). However, fodd does not satisfy
the boundary condition (5.3.10). So the eigenspace is spanned only by feven. If
k = = =
1, then 0: O. The eigenfunction is feven (k=l) sech 2 y and the eigenvalue
u is zero. Thus uo(x) is neutrally stable. This is case (iii).
If k = 0, then
feven = sech y (1 - y tanh y) and fodd = - sechy tanh y. (5.3.13)
When k = 0, then 0: = -1. The eigenfunctions are given by the above formulae
and the eigenvalue u is real and negative. Thus uo(x) is asymptotically stable.
This is again case (ii).
In summary, for the Burgers shock wave we either have the asymptotically
stable case (ii) or the neutrally stable case (iii) and we can never have the
unstable case (i). The proof is thus complete.
This proof is extracted from a paper by Jeffery and Kakutani (1970) [7].
5.4 Interfacial Boundary Conditions of Two

Viscous Fluids*
This section serves to explain the boundary conditions of the free surface pre-
sented in section 5.1. We will see that when viscosity is present, the mathe-
matical expression for the dynamical condition on a free surface is not obvious.
Fig. 5.7 illustrates a smooth interface between two viscous fluids. We would
like to formulate the dynamic boundary conditions on this interface.
The following notations are employed
uW: the stress tensor for fluid (1) (i,j = 1,2,3),

u~): the stress tensor for fluid (2) (i,j = 1,2,3),
ni: the unit normal vector on the interface pointing upward (i = 1,2,3),
x; = ~l + ~2: the total curvature of the interface E,

'Y: the tension coefficient
5.4. Interfacial Boundary Conditions of Two Viscous Fluids 143
Fluid (1)
Fluid (2)
Figure 5.7: The configuration of the smooth interface of two viscous fluids.
T = 'Y (il + i2): the surface tension.

In the above expression, Rl and R2 are the radii of the curvature of the
interface E in any two orthogonal planes that contains the normal vector n =
(nl, n2, n3). Take E to be an infinitesimal element of the interface E. The
balance of the forces acting on the element E leads to the equilibrium equation
(5.4.1)
Here the usual summation convention is used.

Decomposing (5.4.1) into the normal and the tangential components, we
have
Normal ,. .....(l)n.n.
Vij 'J -
.....(2)n.n. _ ..., (
Vij J - I
1+ R21) '
Rl (5.4.2)
Tangential: uU)niTj - U}J>niTj =0 (5.4.3)
where (Tl,T2,T3) is a unit tangent vector.

For the Newtonian fluid (see Chapter 2), the stress tensor is related to the
velocity field ui(i = 1,2,3) in the following way
(5.4.4)
We use two examples to illustrate how to use the above formulas (5.4.2) and
(5.4.3).
ExaIllple 1: If the fluid (1) is air and the fluid (2) is oil, then one often takes
u~J> = 0 and hence the interface is approximately viewed as a free surface. The
dynamic boundary condition is then
_ ".(2)n -
Vij 3 -
"V
I
(2- + 2-)
R1 R2
n'
3'
(5.4.5)
EXaIllple 2: If the fluid flow is a two dimensional motion, then one can derive
(5.1.5)-(5.1.6).
The profile of the free surface is denoted by y - 7J(x, t) = O. Then the unit
outer normal vector is
n = (-7Jz, l)/Vl + 7J~.
The unit tangent vector is
t = (1, 7Jz)/Vl + 7J~.

If the surface tension is vanishing and the upper fluid is air, then the dy-
namic condition on the free surface is that the surface traction is equal to zero
(see equation (5.4.5)). Namely,
2
L: [-pi5 ij + J.l(Ui,j + Uj,;)] ni = O. (5.4.6)
i=1
Projecting the above vector into the normal direction n and the tangential
direction t respectively, we obtain the following
2
L: [-pi5ij + J.l(Ui,j + Uj,i)] ninj = 0, (5.4.7)

i,j=l
2
L: [-pi5ij + J.l(Ui,j + Uj,i)] nitj = O. (5.4.8)
i,j=1
Let x = index 1, y =index 2,1.11 U and 1.12 = =

v. Then the above two
equations can be simplified to (5.1.5) - (5.1.6) respectively.
[1.] M. C. Shen and S. M. Sun (1987), Critical viscous surface waves over an
incline, Wave Motion 9, 323-332 ..
[2.] C. S. Yih (1963) , Stability of liquid flow down an inclined plate, Phys.
Fluids 6, 321-334.
[3.] T. B. Benjamin (1957), Wave formation in a laminar flow down an inclined
plate, J. Fluid Mech. 2,554-574.
5.4. Interfacial Boundary Conditions of Two Viscous Fluids 145
[4.] E. Hopf (1950), The partial differential equation Ut + UU., = Jl.U.,." Com-
mun. Pure Appl. Math. 3,201-230.
[5.] J. D. Cole (1951), On a quasilinear parabolic equation occurring in aero-
dynamics, Q. J. Math. 9,225-236.
[6.] Shih- I Pai (1956), Viscous Flow Theory, I - Laminar Flow, D. van
Nostrand Co. Inc., New York.
[7.] A. Jeffery and T. Kakutani (1970), Stability of the Burgers shock wave
and the Korteweg-de Vries soliton, Indiana Univ. Math. J. 20,463-468.
[8.] G. B. Whitham (1974), Linear and Nonlinear Waves, John Wiley, New
York, Chapter 4.
Chapter 6
Forced KdV Equation
In this chapter we study the forced Korteweg-de Vries equation (fKdV):
Ut + AU x + 2auu x + f3u xxx = J'(x), -00 < x < 00

where A, a < 0 and f3 < 0 are constants, and f(x) is a given function (called
the forcing) which is differentiable and has a compact support (i.e. it is nonzero
only in a closed bounded set). This equation is an asymptotically reduced result
from Euler equations of fluid motion and corresponding boundary conditions.
The unknown function u(x, t) represents the first order elevation of the free
surface of the fluid. The forcing function f(x) is due to the bottom topography
of a fluid domain (such as a bump on the bottom of a two dimensional channel),
or due to an external pressure on the free surface (such as the wind stress on
the surface of an ocean). Solutions of this fKdV are characterized according to
the value of A. We will show that there exist two values of A (AL < 0, AG > 0)
such that
147
148 Chapter 6. Forced KdV Equation
External pressure
Y Free surface
~c'~~~I~H~~~,~~~~~~~~~~x
" t,. (x') bump
Figure 6.1: Configuration of the flow: a uniform flow disturbed by a bump
and/or a surface pressure.
(a) when A ~ AC, the fKdV admits at least two stationary solitary wave
solutions and A = AC is the turning point of the bifurcation curve;
(b) when A $ AL, the fKdV admits only one downstream cnoidal wave solution
and A = AL is the cut-off point at which the cnoidal wave becomes a
hydraulic fall;
(c) when AL < A < AC, the fKdV admits no steady state solutions and solitons
are periodically generated at the site of forcing and radiated upstream.
These results are only about ten years old and valuable not only in practical
applications (such as stream blocking by mountains or ocean bottom topogra-
phies) but also in developing analytic theory for forced nonlinear evolution
equations which do not have group symmetries.
In this chapter, we will derive the fKdV and show its solution properties.
The derivation of the fKdV based on potential theory is given in section 6.1.
We study supercritical (A ~ AC) solitary wave solutions in section 6.2. and
subcritical (A $ AL) downstream cnoidal wave solutions in section 6.3. Tran-
scritical (AL < A < AC) soliton radiations are investigated in section 6.4. An
efficient numerical scheme based upon the spectral method is discribed and a
user-friendly Mathematica program for this scheme is attached for students'
use. The stability of the multiple solitary wave solutions is discussed in section
6.5.
Since the results presented in this chapter will not be used in the subse-
quent chapters, the reader may skip this chapter if he or she is not particularly
interested in forced systems.
6.1 An Ideal Flow Over a Small Bump
We consider the two dimensional motion of an inviscid fluid of constant density

induced by a moving distributed pressure on the free surface, or by a moving
6.1. An Ideal Flow Over a Small Bump 149
bump on the bottom of a channel. Both the moving pressure and the moving
bump are assumed to have compact support. The fluid flow configuration is
shown in Fig. 6.1. In practical applications, p* may be the wind stress on the
surface of ocean water, u* a mountain blocking for atmospheric current, or as
a bottom topography blocking for an ocean current.
Let H be the upstream depth and L be the typical wave length. We want to
find the behavior of long waves of the fluid motion. This long wave assumption
is specified by defining a small parameter c
0<f=(HjL)21. (6.1.1)
Assume that the distributed pressure on the free surface and the bump on the
bottom move at the same speed c* toward the upstream. A reference frame
(x*, y*) is chosen to be fixed with the bump (see Fig. 6.1). If the flow is
irrotational, the governing equation is the Laplace equation for the potential
cJ>* of the flow field .
.6,*cJ>* = 0, u*(x*) < y* < H + 1]*(x*,t*) (6.1.2)

where .6,* = {)2jfJ2x* + ()2j{)2y* is the two-dimensional Laplacian.
There are two boundary conditions on the free surface. One is the geometric
condition. It states that any fluid particle on the free surface always stays on
the free surface. This can be expressed by
l2.-(y* - H - 1]*) = 0 on y* = H + 1]* (6.1.3)

Dt*
where Dj Dt* is the material derivative operator. Another is the dynamic
condition. It is expressed by the Bernoulli equation
{)cJ>* 1 -* 1
()t* +"2(V'*cJ>*)2+ g(H+1]*)+: = "2(c*)2+ gH on y* = H+1]*. (6.1.4)
Here p is the density of the fluid, and c* is the speed of the upstream flow.
There is only one boundary condition on the bottom: the geometric condi-
tion. It states that the fluid can not penetrate into the rigid boundary. This
condition is expressed by
- D (*
Dt*
y -u *) =0 on y* = u*. (6.1.5)
Equations (6.1.2) - (6.1.5) have all been expressed in terms of dimensional

quantities, which are signified by the superscript "*". To nondimensionalize
the equations, we introduce the following dimensionless quantities:
x x* j L, y = y* j H,
t f3/2Ift* (assumption of long time scale or smooth motion),
1] 1]* j H, cJ> = cJ>* j(LV9H) , p = p* j(pgH),

P f-2p* j(pgH) (small surface stress assumption),
u f - 2 U* j H (small bottom obstruction assumption).
The "long time" above means that the time scale is longer than L / J 9 H: the
time needed for a shallow water wave to travel a distance L ( the horizontal
length scale). So the time scale of the motion considered here is longer than
that in the shallow water wave (see Chapter 3) by a multiple of c 1 . Hence our
time scale here is: t* "" c1(L/ViJl).
By the above dimensionless quantities, equations (6.1.2) - (6.1.5) can be
nondimensionalized. The nondimensionalization results in
(6.1.6)
with boundary conditions on the free surface y = 1 + 77(X, t),
f77t + (c + ~x)77x = f-l~y, (6.1.7)

1
ftpt + 2(~~ + f-l~n + c~x + 77 + f2p = 0, (6.1.8)
and with a boundary condition on the bottom y = f 20'(X),
(c + ~x)O'x = f-3~y. (6.1.9)
In the above, let

tp=~-cx.
Since c x is the potential of the uniform upstream flow, tp(x, y, t) is the potential
of the perturbed wave motion. If 77(X, t) is small, we approximate the boundary
conditions (6.1.7) - (6.1.8) on y = 1 + 77(X, t) by their Taylor expansions about
=
y 1. It follows that
on y = 1, (6.1.10)
on y = 1. (6.1.11)
Similarly, the Taylor expansion of (6.1.9) about y = results in
(c + tpx + f2tp xy O')O'x = f-3tpy + f-1tpyyO'. (6.1.12)
The speed of the upstream flow is assumed to be near critical. Namely,
(6.1.13)
Here Co is called the critical speed of the upstream uniform flow, which is going
to be determined. So A is a measurement of the perturbation of the upstream
uniform flow velocity c from its critical value co. If A> 0, then c> Co and the
flow is called supercritical. If A < 0, then c < Co and the flow is subcritical.
The unknown functions are tp =
tp(x, y, t; f) and 77 =
77(X, t; f). If we know
these two functions, then we know everything about the flow and the wave.
However, it is impossible to analytically find these functions tp and 77 as solutions
ofthe corresponding BVP exactly. The alternatives are numerical solutions and
6.1. An Ideal Flow Over a Small Bump 151
asymptotic approximations. In this book, we choose to find asymptotically

approximate solutions.
To find an asymptotic approximation of these two functions, we assume
that i.p and TJ have the following asymptotic expansions
i.p = (i.p1(X,y,t) + (2i.p2(X,y,t) + (3i.p3(X,y,t) +O({4), (6.1.14)

TJ = (TJ1(X, t) + (2TJ2(X, t) + O({3). (6.1.15)
Substituting (6.1.12) - (6.1.15) into (6.1.6), (6.1.10) - (6.1.11), and rearranging

the terms according to the powers of (, we obtain a sequence of boundary value
problems. The problems of the first three lowest orders are as follows:
Lowest order:
i.plyy = 0, 0< y < 1, (6.1.16)

1 2
2<P1y + COi.p1:r: + TJ1 = 0 on y= 1, (6.1.17)
i.p1y =0 on y= 1, (6.1.18)
i.p1y = 0 on y= O. (6.1.19)
First order:
i.p2yy = -i.p1:r:x, 0 < y < 1, (6.1.20)

1 2
i.plt+ 2i.p1X + i.p1yi.p2y + i.p1yi.plyyTJ1 + Coi.p2x
+Ai.p1:r: + COi.p1:r:yTJ1 + TJ2 + P = 0 on y = 1, (6.1.21)
CoTJ1:r: = i.p2y + i.p1yyTJ1 on Y = 1, (6.1.22)
i.p2y =0 on y = O. (6.1.23)
Second order:
i.p3yy = -i.p2xx, 0 < y < 1, (6.1.24)

+ <P1ty + CoTJ2x + ATJ1:r: + i.p1xTJ1:r:
7]lt
+ i.p1yyTJ2 + TJ1i.p2yy on Y =
= i.p3y 1, (6.1.25)
COUx = i.p3y + i.p1 yy U on Y = O. (6.1.26)
Integrating the lowest order problem (6.1.16) - (6.1.19), we obtain that i.p1 is a
function of only x and t and independent of y:
i.p1 = i.pl(X,t).
Also from the direct integration of (6.1.16) - (6.1.19), we get
<P1:r: = -7]1 (x, t)/co. (6.1.27)
By the first order problem (6.1.20) - (6.1.23), we can get
i.p2 = -~i.plxxy2 + F(x, t) (6.1.28)

where F(:c, t) is an arbitrary function. By (6.1.27),
1
CP2 = -2 T/1:cY
2
+ F(:c, t ) . (6.1.29)
Co
Using (6.1.21), we can obtain
1 2 1
F:c = -CPlt
--
Co 2co 2co
T/2
Co
P
-T/1 - -T/1:c:c + AT/1 - - - - .
Co
(6.1.30)
Integration of equations (6.1.29) and (6.1.30) yields
1 2 CPlt T/r T/1:c:c, T/2 P

CP2:c = -T/1:c:cY - - - - - - - + AT/1 - - - - . (6.1.31)
2~ ~ 2~ 2~ ~ ~
It follows from (6.1.29) and (6.1.22) that
(c~ - 1)T/1:c = o. (6.1.32)
For a nontrivial solution T/1:c, we have T/1:c t 0 and hence
c~ = 1. (6.1.33)
So the critical speed of the upstream flow is one. From the flow configuration,
the flow at the upstream :c =
-00 is along the positive x-axis direction, and
hence Co = 1.
Integrating (6.1.24) with respect to y from y = 0 to Y = 1 and using the
boundary conditions (6.1.25) - (6.1.26), we obtain
T/lt + COT/2:c + AT/1:c + CP1:cT/1:c - T/1CP2yy - Co(J':c

_ [T/1:C:C:C _ CPlt:c _ (T/i):c _ T/1:c:c:c + AT/1:c _ T/2:c _ P:c]. (6.1.34)
6co Co 2co 2co Co Co
From (6.1.27), (6.1.31), (6.1.33) and (6.1.34), it follows
311
1Jlt + AT/1:c - 2T/1T/1:c - 6T/1:c:c:c = 2(p:c + (J':c). (6.1.35)
Here we have taken Co = 1. Equation (6.1.35) is called the forced Korteweg-de

Vries equation (f K dV). Apparently, the forcing term on the right hand side
of the equation is due to the external pressure exerted on the free surface and
the bump on the bottom. Therefore, for a given surface pressure p, a given
bump (J', and a given upstream near critical flow speed c = 1 + fA, we can find
an asymptotically approximate shape of the free surface T/ = fT/1 (:c , t) + 0(10 2)
by solving the partial differential equation (6.1.35). In the next four sections,
we will discuss the solution behavior of the fKdV equation (6.1.35). It turns
out that there are four types of solutions:
(i) Supercritical stationary solitary waves: They occur only when A is positive
and sufficiently large;
6.2. Supercritical Solitary Waves 153
(ii) Subcritical stationary cnoidal wave: It occurs only when A is negative

and sufficiently small;
(iii) Stationary hydraulic fall: It occurs only when A is equal to a special
value AL < 0;
(iv) Unsteady periodic soliton radiation: This solution appears when IAI is
small. Such a solution is called the transcritical solution.
6.2 Supercritical Solitary Waves
In this section, we study solitary wave solutions of the time independent (i.e.
stationary) forced Korteweg-de Vries equation (6.1.35). Solitary wave refers
to any surface wave profile that dies out at infinity. If v(x) is a solitary wave
solution of (6.1.35), then it satisfies
AV' + 20:vv' + f3v"' = !'(x), -00 < x < 00, A> 0, (6.2.1)
v(x = oo) = v'(x = oo) = v"(x = oo) = 0 (6.2.2)
where v' = dv/dx.

Of course, for the case of (6.1.35), 0: = -3/4, and f3 = -1/6. However,
to keep the generality of (6.2.1), we do not specify values for 0: and f3 in this
section. Instead we just assume that 0: and f3 are negative constants. Such
a general theory has applications in studying disturbed flows which are near
critical in channels of arbitrary cross section, such as a triangular cross section,
an elliptic cross section, etc. The values of 0: and f3 are uniquely determined
by the geometry of the cross section.
Integrating (6.2.1) with respect to the independent variable from -00 to x,
we have
AV + o:v 2 + f3v" = f(x), -00 < x < 00, A> 0, (6.2.3)

v'(x = oo) = v(x = oo) = O. (6.2.4)
We specify the parameters as follows: 0: < 0, f3 < 0 are fixed, and f(x) is
nonzero only in a finite interval [x _ , x +l, and A is posi ti ve since the flow is
supercritical.
Recall that (1 + tA)VgH is the upstream near critical flow speed. Physical
intuition tells us that for a given 0:, f3 and f(x), the shape of the free surface is
totally controlled by the upstream flow velocity (i.e. A). Therefore we would
like to study the solution behavior of (6.2.3) - (6.2.4) for different values of
A. First of all, we want to know the range of the parameter A in which the
problem (6.2.3) - (6.2.4) has solutions. The existing research results indicate
that there exists a number AC > 0 such that (6.2.3) - (6.2.4) has: (i) at
least two solutions when A > AC, (ii) one solution when A = AC and (iii)
no solutions when A < Ac. The proof of this existence theorem is beyond the
scope of this book. Interested readers are referred to Shen's paper (1992). Here
we present only analytic solutions when the forcing is a Dirac delta function
(locally forced) and numerical solutions when the forcing possesses a compact
support (non-locally forced).
6.2.1 Locally forced supercritical waves

In the scaling process described in section 6.1, the height 110'11 of the bump is
divided by H and the length of the bump base, B, should be divided by the
horizontal length scale L. If the length of the bump base is very short (i.e.
B is small), then B / L is very small since L is very large for the long wave
assumption. After the scaling, we approximately regard the base length of the
bump as zero [in the dimensionless coordinates]. But the area under the bump
after scaling is not zero. The Dirac delta function possesses such a property.
We thus employ it to approximate the short base bump in the dimensionless
coordinates. The forcing that has a very short base length is called the local
forcing. Results from such an approximation have proved reasonably accurate
for semi-circular bumps. So we consider the following fKdV
P
..\v + O'v 2 + f3v" = '2 8(x), ..\ > 0 . (6.2.5)
We look for solitary wave solutions of (6.2.5). Hence v satisfies the boundary
conditions:
v(oo) = v'(oo) = 0 . (6.2.6)
By direct integration, the solutions of (6.2.5) - (6.2.6) can be expressed as
follows
3..\
v(x) = - 20' sech
20
V4jj (x - L+), x> 0, (6.2.7)
3..\
v(x) = - 20' sech
20
V4jj (x - L_), x<o (6.2.8)
where L+ and L_ are constants. The continuity condition of the free surface
at x = 0
v(O+) = v(O-) == v(O) (6.2.9)
implies that
(6.2.10)
Because of 6(x), v'(x) must have a jump discontinuity at x = O. Namely,
p
v'(O+) - v'(O-) = 2f3 . (6.2.11)
From (6.2.7) and (6.2.8), this condition can be written as
-fj "(0) [-tanh (fJ L+) (fJ L_)]- ~.

+ tanh (6.2.12)
This equation holds for a nonzero P only if L+ ::f. L_, and from (6.2.10), we
get
L+ = -L_ = Lo .
Hence (6.2.12) can be written as
(6.2.13)
where
f = tanh (J~: LO) , (6.2.14)
Po: 1
e = -~ ";-f3>' . (6.2.15)
Since P - 1 = I(J + 1)(1 - 1), it is clear that:

(i) when lei < 2/(3..;3), equation (6.2.13) has three distinct real roots and
only two of them are in (-1, 1);
(ii) when lei > 2/(3..;3), equation (6.2.13) has only one real root whose
absolute value is greater than one;
(iii) when lei = 2/(3..;3), equation (6.2.13) has a double real root whose
absolute value is less than one and the third real root is not in (-1, 1).
Therefore
Po: 1 2
e = -~ ";-(3)' = 3..)3
determines the critical value of >.:
_ (30:2 P2) t (6.2.16)

>'c - -16f3
It follows that
Lo = Vf4ii
=x arctanh(J) (6.2.17)
has: (i) two values if >. > >'c ; (ii) one value if >. = >'c ; and (iii) no solution
if 0 $ >. < >'c . As soon as one finds Lo, the solution (6.2.7) - (6.2.8) is
determined.
3>' { sech 2
v(x) = - - I~
/"j[
(x - Lo) , x ~ 0,
(6.2.18)
20: sech V ~ (x + Lo) ,
2 x $ O.
For a given L o, (6.2.18) defines a cusped solitary wave (see Figs. 6.2(c) and
6.2(d) ). The cusp is concave up (down) if Lo > 0 0 respectively). From
(6.2.13) - (6.2.15) and (6.2.17), we have:
P < 0 =? c < 0 =? f > 0 =? Lo > 0 =? cusp is concave up,

P > 0 =? c > 0 =? f < 0 =? Lo < 0 =? cusp is concave down, Hence
sign(P) = - sign(Lo) .
Selected solutions of (6.2.5) - (6.2.6), determined by (6.2.18), are shown in
Figs. 6.2(c), 6.2(d), for a triangular channel: a = -5./2/8,(3 = -13./2/192.
Correspondingly, the saddle node bifurcation diagrams are shown in Fig. 6.2( a).
The relationship between Lo and A is also shown in Fig. 6.2(b). The turning
points of the bifurcation diagrams are computed from equation (6.2.16).
When P > 0, IIvll oo =-
~~ sech 2 ( N
Lo) < - ~~, which is the amplitude
of the free solitary wave. The bifurcation diagram IIvll oo versus A is given by
= -"41fJ. { [3 ( )+ 1}
-1
P -3 1 Pa 411"/3
IIvll oo cos arccos - 6v-Ji>.3 {211"/3} . (6.2.19)
The (IIvll oo , A) curve has two branches. The upper branch and the lower
branch correspond to 411" /3 and 211" /3 respectively in the above formula. The
two branches are joined at AC, at which 117]11100 = -AC/a. As P ~ 0+, L01 ~
0- and L02 ~ -00. Hence II v 1100 approaches -3A/2a and zero respectively.
When P < 0, the amplitude IIvll oo equals -3A/2a all the time. The cusped
solitary waves have two peaks in each single solution. As P ~ 0-, the two
peaks of a cusped solitary wave merge and the cusp disappears gradually. The
limit is the usual solitary wave in the case of no forcing. At the same time, the
two peaks of the other cusped solitary wave move further apart to upstream
and downstream respectively. The limit is the usual null solution in the case
of no forcing since the peaks have moved to negative and positive infinity.
From sign(P) = -sign(Lo) and from equation (6.2.18) which determines
the free surface profile, we see that if P < 0 (> 0), then the cusps of the
solitary waves are concave up (down respectively). Namely, a surface suction
(P < 0) corresponds to a dent ofthe free surface and a surface pressure (P > 0)
corresponds to a crest of the free surface. This is consistent with the result
obtained by J. W. Miles (1986), but not consistent with our intuition, and so
it is a paradox.
6.2.2 Non-locally forced supercritical waves

The non-local forcings are a class of forcings whose bases arenot very short and
can not be approximated by a Dirac delta function. The support of the forcing
is denoted by supp(f) = [x_, x+l. In general, a numerical method has to be
used to solve the BVP for the fKdV. A computer can only solve a problem
in a bounded domain. Hence, we first need to convert the BVP on (-00,00)
into an initial value problem (IVP) with the initial point at x_. According to
Fig. 6.3, we solve (6.2.3) - (6.2.4) from -00 to x_ analytically. With the initial
condition at x_ determined by the analytic solution in (-00, x_), one can solve
the IVP up to x+ numerically by IVP solvers such as those in Mathematica,
(a)
(b) le-S
p."
.IS
P.1.
.
P ....1
-I
U
,..
I.'
(c) ...
I:t
...
1.1
1.1 P.1.o
a_lAG
L'
",,1.4
(d)
...
. 1
.
..
'..
Figure 6.2: Supercritical flows in a triangular channel.
f(x)
------------~--~------~--------------~x
x_ x+
Figure 6.3: A non-local forcing function f(x) with a compact support.
IMSL and other software libraries. Finally the solution in (x+, 00) can also be
found analytically by matching with the numerical solution at x+.
The analytic solution of (6.2.3) - (6.2.4) in (-00, x-l is
3..\ sech 2
v(x) = - 2a (
V/-i
2jf(x - Lo) ), (6.2.20)
Here Lo is a phase shift. Because of the presence of the bump f(x), Lo can not
be an arbitrary constant. Instead, we will see that Lo can only take certain
discrete values. From (6.2.20), we have different solutions of the BVP (6.2.3) -
(6.2.4) for different values of Lo.
To help search for the values of Lo, we introduce a new quantity B>.(Lo):
1 "'+
B>.(Lo) = "'_ f(x)v'(x)dx,
or
(6.2.21)
Multiplying (6.2.3) by v'(x) and integrating the product from x_ to a point

x > x+ with respect to the independent variable, we have
"2(3 (v' (x)) 2 + [..\"2 + 3a v (x) ] v 2 (x) = B>.(Lo), (6.2.22)
For this equation to have a half solitary wave solution (i.e., x > x+ and v( +00) =
0), the polynomial of v
must have a double real zero which is smaller than the third real zero (see the
end of Chapter 3). This is the case only when B>. = 0 (see Fig. 6.4). Therefore
we have
v(+oo) = 0 (6.2.23)
6.2. SupercriticaJ Solitary Waves 159
Figure 6.4: The curve of the cubic polynomial Q(v).
When B>.. = 0, the function v(x) satisfies 0 < v(x) ~ -3.x/(2a) for x> o.
When a = -3/4,/3 = -1/6, and
/(x) = { ~sin(1rx), Ixl ~ 1,

Ixl> 1,
the B>..(Lo) function is shown in Fig. 6.5. The .x value at the turning point
is .xc = 1.391133. When .x = 3.3, B>..(Lo) has two zeros: L01 =
-1.1791390,
L02 = -0.5662725. When .x = 1.1, B>..(Lo) has no zero.
Hence, B>..(Lo) = 0 is the condition used to determine Lo. In this way, Lo is
an implicit function of .x, which may be multiple valued. To find B>.., we solve
the following initial value problem up to x+
!: ,ech' [~(z_ - L+
.xv + av 2 + /3v" = 0, x> x_, (6.2.24)
v{z-) = - (6.2.25)
v'(z_) = -fi- [~(z_ -LO)].

v(z_) tanh (6.2.26)
Given a trial value of L o, the initial conditions (6.2.25) - (6.2.26) are deter-
mined. This IVP has a unique solution since the differential equation (6.2.24)
satisfies the Lipschitz condition. One can solve the problem by an IVP solver
on a computer. The numerical results presented here were obtained by using
an initial value solver called NDSolve[ ] in Mathematica. When we solve
the problem up to the point x = x+, the value of B>..(Lo) can be computed
by (6.2.21). Hence, for a fixed .x, a given trial value of Lo generates a unique
value of B>..(Lo) by solving (6.2.24) - (6.2.26). We can plot the function B>.. vs.
"
.;
..
0
i
... ,.
~
.~
~
0
~ -3.0
-J.o -1.0 0.0 1.0 2.0
LO
Figure 6.5: The graph of B>. (Lo) vs. Lo for a fixed A = 1.1,1.391133,3.3.
Lo for a fixed A (see Fig. 6.5). Our numerical results show that there exists a
positive number AC such that: (i) if A > AC, B>.(Lo) has at least two distinct
=
zeros, (ii) if A AC, B>.(Lo) has a double zero, and (iii) if A < AC, B>.(Lo) has
no zero.
If there exists an Lo such that B>.(Lo) = 0, then (6.2.22) has a solution
corresponding to this Lo that satisfies v( +00) = v' (+00) = 0. Therefore, the
original boundary value problem (6.2.3) - (6.2.4) is solved. The number of
solutions is equal to the number of zeros of B>.(Lo).
We summarize our solution procedure as follows:
(i) Pick a value of L o, use NDSolve[ ] in Mathematica to solve (6.2.24) -
(6.2.26) up to x+, and compute B>.(Lo) by (6.2.21).
(ii) Repeat step (i) for different values of Lo and plot the function B against
L o, and find zeros of B>.(Lo).
(iii) With the correct phase shift Lo determined by B>.(Lo) = 0, solve (6.2.24)
- (6.2.26) up to a satisfactory point. This solution matched with the
solution (6.2.20) for x < x_ is a complete solution of the BVP problem
(6.2.3) - (6.2.4).
When 0' = -3/4, (3 = 1/6, A = 3.3, and the non-local forcing is defined by
Ixl ~ 1,
Ixl> 1,
the corresponding two solutions are shown in Fig. 6.6. Solution I corresponds
=
to L01 -1.1791390, and the other one corresponds to L02 -0.5662725. =
6.3. Subcritical Cnoidal Waves and Hydraulic Fall 161
v
L01=-O.5662725
7
6 L02=-1.179139
x
-3 -2 -1 o 1 2
Figure 6.6: Multiple (two) solitary wave solutions of the stationary forced
Korteweg-de Vries equation.
In the above we have shown how to find numerical solutions of the BVP
problem (6.2.3) - (6.2.4) for a fixed A. From the numerical examples, we have
noticed that the maxima of solutions depend on A. As A decreases, the maxi-
mum of one solution (the upper solution) decreases, but the maximum of the
other solution (the lower solution) increases. Since A measures the upstream
flow speed, from our intuition the amplitude of the wave (i.e. maximum of
the solution) should be proportional to A. This intuition suggests the upper
solution is physical and the lower one is unphysical. Actually, it has been con-
jectured that the lower solution is unstable and the higher solution is asymp-
totically stable. So it is interesting to know the global dependence of the wave
amplitude on A. This is called the bifurcation diagram. This diagram can be
obtained by solving (6.2.3) - (6.2.4) repeatedly for different values of ). using
the procedure prescribed above. So the maxima of solutions against). can be
plotted. This curve is the bifurcation diagram. The one corresponding to the
forcing in Fig. 6.6 is shown in Fig. 6.7.
6.3 Subcritical Cnoidal Waves and Hydraulic

Fall
In this section, we study the time independent, subcritical solutions of the
forced Korteweg-de Vries equation (6.1.35). If v(x) is such a solution then v
v
max
7
lambda
2L---~--------~----~~~~------~~~-----
1.5 2 2.5 3
Figure 6.7: Bifurcation diagram of a supercritical flow over a non-local forcing.
satisfies
AV' + 2avv' + f3v'" = !,(x),-00 < x < 00, A < 0, (6.3.1)
v(-oo) = V' (-00) = v"(-oo) = 0 (6.3.2)
where v' = dv/dx.
Integrating (6.3.1) with respect to the independent variable from -00 to x,
we have
= f(x),
AV + av 2 + f3v" A < 0, (6.3.3)
v(-oo) = v'(-oo) = O. (6.3.4)
We specify the parameters as follows: a < 0, f3 < 0 are constants, and
f(x) is nonzero only in a finite interval [x_,x+], and A is negative (because
of subcritical flows). Like the discussion for the supercritical flows in the last
section, we classify the cases according to the base length of the forcing: local
forcing for a short bump and non-local forcing for a long bump. In the case
of local forcing, solutions can be found analytically. Otherwise, in general,
solutions have to be found numerically.
6.3.1 Locally forced sub critical flows

The forcing term for a local forcing is expressed in terms of (P/2)8'(x) in the
fKdV.
P
AV + av 2 + f3v" = "2 8(x), A < 0 , (6.3.5)
v(-oo) = v' (-00) = 0 . (6.3.6)
When x < 0, the solution vanishes identically.

v(x) =0 when x ~ 0.
Therefore solving the problem (6.3.5) - (6.3.6) is equivalent to integrating
the following initial value problem
AV + av 2 + f3v" = 0, x > 0, (6.3.7)
v(O+) = 0, (6.3.8)
v'(O+) = P/(2f3) . (6.3.9)
The first integral of the above gives

3f3
- (v')2 = _v 3
2a
_ -
3A
2a
v2 + -
3
8af3
p2 = Q(v) , x> 0, (6.3.10)
v(O+) = 0 . (6.3.11)
Here, A can be chosen to make Q(v) have three distinct real zeros, a double real
zero, or only one real zero. Correspondingly, the problem (6.3.10) - (6.3.11)
has a cnoidal wave solution, a wave free solution (Le. hydraulic fall), and an
unbounded solution respectively. If we choose
(6.3.12)
then Q(v) has three distinct real zeros, a double real zero, and only one real
zero when A < AL, A = AL and A > AL respectively. This claim can be easily
verified by either plotting the Q(v) curve or factorizing Q(v).
Thus, the IVP (6.3.10) - (6.3.11) has: (i) a cnoidal wave solution when
A < AL; (ii) a hydraulic fall solution when A = AL, and (iii) no bounded
solutions when A > AL.
From the condition (6.3.12), one can see that for the given geometry of a
channel (aand f3 are determined), AL depends only on P. Comparing (6.2.16)
and (6.3.12), we see that IALI > Ac.
When A < AL, the cnoidal wave solution of (6.3.10) - (6.3.11) can be ex-
pressed in terms of a Jacobi elliptic function
v(x) = ~ [cos (0 +~) - ~ +(coso - cos (0 +~))
x cn 2 ( 6~ (coso - cos (0 + 2;)) (x - xo)) 1(6.3.13)

when x > O. The phase shift Xo is in [0, T] and is determined by v(O+) = 0, l.e.
= (cos 0- cos (0 + *) )
xcn 2 ( :~ (coso-cos (0+ 2;)) x o) (6.3.14)
Here T is the period of the cnoidal wave (6.3.13) and is given by
(6.3.15)
The parameters 0 and k 2 are
o= ~ arc cos [-1 + ~ (a:r] ~ i ' (6.3.16)
cosO - cos(O + ~)
n-
2
k = 2 <1. (6.3.17)
cosO - cos(O +
And K(k 2 ) is the complete elliptic integral of the first kind defined by
(6.3.18)
The closer>. is to >'L, the larger the period T of the cnoidal wave is. When
>. t >'L, this period approaches infinity and the cnoidal wave solution becomes
a wave free solution. This is the hydraulic fall. This conclusion can be easily
derived from (6.3.12), and (6.3.15) - (6.3.16). As a matter offact, when>. = >'L,
=
we have 0 0, k 2 = 1. Since K(l) =
00, the period T =
00 by (6.3.15). The
solution (6.3.13) becomes
>'L [ -1 + 2
v(x) = -;- 3 sech 2 ( Vr;;
4jj (x - xo) )] , x>o (6.3.19)
where Xo is determined by v(O+) = 0, I.e.
Xo = Vf4ii
4 arc sech Vf23 . (6.3.20)
The downstream depth is HD = (1 - f >'L/a)H < H. So the free surface falls

to a lower level from the upstream higher level. Next we show that in a square
channel the downstream flow is supercritical for such a hydraulic fall.
Let UD and HD be the downstream velocity and depth. The conservation
of mass flux yields
1.0
.8
.6
.'1
.2
'1, 0
-.2
-. 'I
-.6
-.8
-1.0
-1.2
-I. 'I
-1.6
-1.8
-2.0
-1 0
x
Figure 6.8: Profiles of three typical stationary subcritical surface waves: sinu-
soidal wave (I), cnoidal wave(lI) and hydraulic fall (III).
Then the downstream Froude number FD is
FD UD/JgHD
1 - f. >"L + O( f.2) > 1 (supercritical)
Another limit is the case when>.. -+ -00. Then () ~ 1r/3, k 2 ~ 0, and
= =
K(k 2 ) ~ K(O) 1r/2. Hence the period is T 21r/V-6>" . The cnoidal waves
become approximately sinusoidal waves whose amplitudes approach zero as
>.. -+ -00. The subcritical cnoidal waves, hydraulic falls and sinusoidal waves
are shown in Fig.6.8 for a triangular channel: a = -50/8, f3 = -130/192, P =
96/13 and>" = -1.455 (for the cnoidal wave), -1.451 (for the hydraulic fall)
and -4.0 (for the sinusoidal wave).
6.3.2 Non-locally forced subcritical flows

We consider solutions of (6.3.3) - (6.3.4) in three intervals: (-00, x-J, [x_, x+l
and [x+, 00). In (-oo,x-J, (6.3.3) - (6.3.4) has only trivial solution:
v(x) = 0, if x ~ x_. (6.3.21)
Then we solve an initial value problem from x_ to x+:
>..v + av 2 + f3v" = f(x), (6.3.22)
V(~L) = v'(x_) = O. (6.3.23)

This IVP satisfies the Lipschitz condition and hence has a unique solution.
With the existence of a solution of (6.3.3) - (6.3.4) from -00 to x_ and
from x_ to x+, we now extend the solution to [x+, 00). Multiplying (6.3.3) by
v'(x) and integrating the resulting equation from x+ to x > x+, we have
3{3 , 2 3 3A ( )
2a (v) = -v - 2a v + D == Q v , (6.3.24)
where
(6.3.25)
For the given geometry of a channel, D is a function of A only. The value of
A can be chosen to make Q(v) (defined by (6.3.24)) have three distinct zeros,
a double zero, or only one real zero. Correspondingly, (6.3.24) has a cnoidal
wave solution, a wave free solution (hydraulic fall), or an unbounded solution
respectively. It is easy to show that if
(6.3.26)
then Q(v) has a double root which is smaller than the third root. The solution
of (6.3.26) is denoted by AL and is negative. Numerous numerical experiments
show that D(A) is positive and bounded as A ~ -00. Hence the D(A) curve and
the A3 j(2a 3 ) curve have an intersection. Consequently, equation (6.3.26) has a
solution AL and the solution of equation (6.3.24) is a cnoidal wave, a hydraulic
fall or an unbounded function when A < AL, A = AL, and A > AL respectively.
Since the unbounded solution is unphysical, we consider only those solutions
when A ::; AL.
When A ::; AL, equation (6.3.24) has a solution
v(x) = ~ [cos (0 + ) - ~ + (cosO - cos (0 + ~7r))

xcn 2 ( 6~ ( cos - cos (0 + 2;)) (x - xo)) ] (6.3.27)
for x > x+. The phase shift Xo is in [0, T] and is determined by
v{ x +) = ~ [ cos (0 + ) - ~ + ( cos 0- cos (0 + 4;))

xcn 2 ( 6~ ( cos - cos (0 + ~)) (x+ - xo)) ]. (6.3.28)
T is the period of the cnoidal wave of (6.3.27) and is given by
(6.3.29)
6.4. Transcritical Periodic Soliton Radiation 167
The parameter (J and k 2 are
(J '13 arccos (-1 + 40: 3 D/ A3 ), (6.3.30)

cos (J - cos((J + 4;)
-------.:-~2~ < 1. (6.3.31)
cos (J - cos((J + ;) -
And K(k 2 ) is again the complete elliptic integral.

The hydraulic fall solution is the limit of (6.3.27) as A tAL. In this case,
(6.3.26) holds. By (6.3.30) - (6.3.31), we have (J = 0, k 2 = 1. Since K(1) = 00,
the period T = 00 by (6.3.29). Equation (6.3.27) becomes
(6.3.32)
and Zo is determined by
(6.3.33)
As z -t 00, v(z) -t -AL/O: < O. Therefore, there occurs a cascade of fluid in a

channel flow.
Another case is when>. -t -00. Then v(z) of (6.3.27) becomes a sinusoidal
wave as predicted by linear theory.
If A ~ AL, then the problem (6.3.3) - (6.3.4) has a bounded solution. The
numerical solution can be easily found by an initial value solver NDSolve[ ]
in Mathematica. One can set the initial point at z = z _ and initial value as
=
v(z_) v'(z_) O. =
Now the question is what the value of AL is for a given geometry of a channel.
We recall that AL is determined by (6.3.26). To find AL, one can use a "do
loop" to solve (6.3.22) - (6.3.23) for each A. In this way a curve D(A) (A < 0)
defined by (6.3.25) is obtained. Then the intersection of the D(A) curve and
the A3 /(20: 3 ) curve gives the value of AL. Therefore, (i) when A < AL, the
downstream free surface consists of a cnoidal wave; (ii) when A = AL, the
downstream free surface is wave free (hydraulic fall); and (iii) when A > AL,
the fluid flow can not reach an equilibrium (i.e., no stationary state exists for
AL < A ~ 0).
6.4 Transcritical Periodic Soliton Radiation
In sections 6.2 and 6.3, we found that if IAI is sufficiently small (more specif-
ically, AL < A < AC), then the forced Korteweg-de Vries equation does not
have a stationary solution. The fluid flow is intrinsically unsteady even though
the forcing is stationary. The flows in the regime AL < A < Ac are called the
tmnscritical flows.
y
n
ds
-y-----.
~(t)
1 + (A
1
x
Figure 6.9: Illustration of the transcritical water wave problem. The bump is
denoted by a Dirac delta function P.:5(x).
In this section, we consider those unsteady state, transcritical solutions of

the forced Korteweg-de Vries equation
"It + ATJx + 2aTJTJx + f3TJxxx = !' (x) (6.4.1)

where A is in the transcritical range (AL' Ac). Then the phenomenon of the
periodical soliton generation at the site of forcing can be found. A bump
moving at constant transcritical velocity c = (1 + (A)Vill in a uniform layer
of water initially at rest is equivalent to a bump fixed in a uniform stream of
velocity c. Please see Fig. 6.9 for an illustration.
The Caltech group led by Professor T. Y. Wu first observed the transcritical
periodic soliton generation in a water flume (1982). Set a bump on the bottom
of a flume. This bump can slide freely along the bottom. Instead of having a
uniform upstream flow, one can move the bump upstream at a uniform speed.
In a two dimensional channel, if one moves the bump at a speed near Vill, then
one can observe the following phenomenon. Solitons are periodically generated
in front of the bump and surge ahead at a speed faster than the speed of the
bump. Immediately behind the bump, there is a uniform depression zone.
Behind this zone, there is a zone of wake, which propagates downstream. The
number of the upstream radiated solitons, the length of the depression zone
and the length of the wake zone are all increasing at a constant rate. Fig. 6.9
illustrates such an interesting phenomenon.
6.4.1 Approximate solutions and mass-momentum-energy-

work relationship
In this subsection, we discuss the approximate solutions of the fKdV (6.4.1)
when the forcing f(x) can be expressed in terms of the Dirac delta function.
This is the case of local forcing, i.e. the support of the forcing is very short in
comparison with the typical wave length.
Recall that the bump is assumed small and described by 0'* (z*) (2 H 0'( z), =
where the small number (is defined by ( = (HjL)2 1. The quantity L is
the horizontal length scale. Hence z = zL. The free surface is assumed to
be .,]* = (H"I1(Z, t) + 0((2). When c = (gH)1/2(1 + (A), the function "Idz, t)
satisfies a forced Korteweg-de Vries equation (fKdV):
3 1 P
"lit + A"I1:I: - '2"11"11:1: - 6"113:3:3: = '2&3:(z), -00 < z < 00. (6.4.2)
Here P = C 3 / 2Sj H2 is the dimensionless area of the bump, S is the actual

area of then bump, and &(z) is the Dirac delta function. The solution to an
initial value problem for the fKdV gives an approximation to the free surface
profile by "I. ~ (H"I1(Z, t). Meantime, the approximate velocity and pressure
fields are
(u, v*) ~ (-("11. (3/2 "11:1: y)(g H) 1/2 ,

p* = pg[-(1 - y* j H) + ("11].
A schematic solution of equation (6.4.2) is shown in Fig. 6.9. Many examples
have shown that the above fKdV is a very good model for the flows under
investigation when the base length of the bump is in a comparable scale with
the bump height (i.e. in the case of local forcing), even when ( takes a relative
large value, say 0.5.
We are concerned exclusively with the initial condition "1* (z, t* = 0) = O.
i:
The mass conservation property of the wave motion gives the following identity
P"I*(z*, t) dz* = 0 (6.4.3)
for any t ~ O. Here, the dimension for the density P is [massHarea]-l.

The horizontal momentum Mh is
The above yields
(6.4.4)
The negative sign "-" implies that the horizontal momentum is oriented toward
upstream.
Similarly, it can be shown that the vertical momentum MtJ is of order
(5 pH2(gH)1/2, which is negligible in relation to Mh.
The total mechanical energy E is equal to the sum of the kinetic energy Ek
and the potential energy Ep. Here we take the potential with respect to half
1:
the depth of the rest water. Hence we have
E = d:e*l~+11 dy* [~((u*(:e*,y*,t*))2+(v*(:e*,y*,t*))2)

H +h* ]
+pg(y* - 2 )
= 1 00
-00 d(f- 1/ 2H:e)
lH(1+1I1)
~Hh d(Hy) {~[(f(-771(:e,y,t))(gH)1/2)2
+ (fl/2(gH)1/2f7713>y)2] + pgH (y - ~(1 + f 2h)) }

+ pgH30(f 7/ 2).
The above yields
E = _(gH)1/2 Mh + pg2H3 f5/2 1 00

-00
1
(77~ + 3771~ - 771h] d:e
+ pgH 30(f7/2). (6.4.5)

Next, we evaluate the energy for upstream solitons (E,), downstream de-
pression (Ed) and the downstream wake (Ew). The k-th upstream soliton
solution of the tKdV (6.4.2) may be expressed in the following form
77~k)(:e, t) = 2(A + s)sech2{[(3/2)(A + s)F/2(:e + st - :ek)} (6.4.6)
where s is the upstream advancing speed of the solitons, a, = 2(A + s) is the
amplitude of the soliton, and :ek is the phase shift. For each soliton 77~k), one
has
m~k) = 1: 77~k)(:e, t) d:e = 4(a,/3)1/2
1:
(mass of one soliton),
(77~k)2 d:e = 8 (~(A + s)) 3/2 = 8 (~ ) 3/2 ,

1: [(77~k)3 + ~77~:)2] d:e = 332 (A + s)5/2 = 4~ a~/2 .
Let N,(t) be the total number of mature solitons upstream at a large time t.
Then the mechanical energy of the upstream solitons is
a, ) 3/2
E, = pgH3 f 3/2N,(t) 8 ( "3
2V2
+pgH3f5/2N,(t)-3-a~/2. (6.4.7)
The first term is -Mh,@ and
Mh, _pgH 2.;gJif3/ 2N,(t) 1: (77~k)2 d:e

_pgH 2.;gJif3/2 N, (t) 8 (~ f/2 . (6.4.8)
is the total horizontal momentum of the all upstream solitons. The momentum
of one soliton is
M h
(k) _
- 8
(a.3 )3/2 . (6.4.9)
To find the mechanical energy of the downstream depression, we evaluate
{:Cd 3 1 2 3
Jo (771 + 3111:c ) dx = -hdxd (6.4.10)
where hd ~ 0 is the depth of the depression and Xd ~ 0 is the length of the

depression zone. The depression depth hd may be determined by the mass
postulate due to Wu (1987) which supposes that the soliton mass comes solely
from the depression. Numerical tests have shown that this mass postulate is
a very good approximation of the actual mass distribution when the upstream
velocity is in a subinterval of the transcritical range. The average height of
the upstream is h., that is, the average of 111 (x, t) with respect x over a period
d., the distance between the two peaks of any two adjacent solitons. When
time is large, we regard h. as an upstream uniform state which falls to the
downstream uniform state h d . Both of these uniform states extend to infinity
as time t --+ 00 and form a stationary state v(x) which yields the following
boundary value problem:
3 1 P
AV:c - '2vv:c - 6v:c:c:c = "2 c5:c(x), -00 < x < 00, (6.4.11)
v(-oo) = h. and v(oo) = -hd. (6.4.12)
=
Let v(x) (x)+h. and find the first two integrals. A downstream (at x = 00)
bounded and wave free solution can happen only when
and (6.4.13)
Exercise: Derive formulas (6.4.13) from equations (6.4.11)-(6.4.12). Hint:

Let v(x) = (x) + h. and see equations (33)-(44) in a paper by Shen (1991)
[14].
The quantity Xd may be determined by the same mass approximation that
leads to
N.m~k) Xdhd = (6.4.14)
= =
where m~k) 4[(2/3)(A+s)F/2 4(a./3)1/2 is the mass of an upstream soliton.
Equations (6.4.5) and (6.4.10), together with the above equation, result in
(6.4.15)
Here the first term is equal to -MhdViiH and
Mhd = - f3/2 pH 2 Verr (a.) 1/2

gH N. 4hd 3" (6.4.16)
is the momentum of the depression zone. The negative sign for this quantity
means that the impulse exerted to the flow by the bump is toward to the
upstream direction.
Now, it is appropriate for us to estimate the soliton amplitude a. and the
soliton generation period T. The following two first integrals are obtained by
doing J~~ (6.4.2) dx and J~~ 171 (x, t)(6.4.2) dx
(k) 3
m;. = -'x1J1 (0, t) + 41Jr (0, t), (6.4.17)
M(k)
;: = -'x1Jr(O, t) + 1J~(0, t)
1 ) 1 2
+31J1 (0, t)1J1xx(0-, t + 61J1X (0-, t) (6.4.18)
where M~~) is the horizontal momentum of an upstream soliton. After adopting

the following approximation,
lim -
T-+ooT
liT 0
1J1(0,t) dt = h., (6.4.19)
1 fT 1Jr(O, t) dt = h;,
lim -T (6.4.20)
T-+oo io
lim .!..
iofT 1J~(0, t) dt = h;, (6.4.21)
. liT
T-+oo T
hm T- 1J1 (0, t)1J1xx(0-, t) dt = 0, (6.4.22)

T-+oo 0
1
lim
T-+oo
-T
iofT 1Jix(O-, t) dt = 0, (6.4.23)
the long time average of the above two first integrals becomes
(6.4.24)
(6.4.25)
The operation (6.4.25)/(6.4.24) results in
as = 2(hd + ~'x)(hd
hd
+ l,X)
. (6.4.26)
To find T., perform the operation J~:::' (6.4.2) dx where XD is any point in the
uniform depression zone. This integral yields
T. = 16 [ 2(hd + l,X) ] 1/2

(6.4.27)
3 3h~(hd + ~,x)
It seems not easy to find the wake energy directly. So we evaluate the total
work done to the water by the bump. Then the wake energy Ew can be found
as W - E& - Ed.
To evaluate the total drag on the bump, we perform the integration
l X1
-00 (6.4.2) x 111 (x, t) dx
for any fixed Xl in the depression zone. This operation yields
M (k) D*
h& 'h2 h3 w (6.4.28)
-r.
+ A d + d = ( 3/2 pg H2
i:
where the drag D:'v is defined as
D~ = p*(x*,y* = 0';. (x*),t*)O';. (x*) dx*.

From equations (6.4.28), (6.4.26), (6.4.27) and (6.4.9), we can obtain
D:;' = ~pgH2(3/2P2. (6.4.29)
It is quite startling that the long time average of the total drag is independent
of the flow velocity. This information is very valuable to floating body designs.
This conclusion is supported by Fig. 6 in the paper by Lee et. al (1989) when
the upstream Froude number is between 0.6 and 1.1. In their figure, the wave
resistance decreases only about 10% when the upstream Froude number F
changes from 0.6 to 1.1.
The total work done by the bump up to the time N&T.(H/g)1/2 when N.
solitons are mature is
This yields
3
W = _N&pgH 3(3/2 p 2T.(1 + (A). (6.4.30)
2
Then, we can obtain the following energy distribution results
E. = N.P9Hl/2~a;3/2(4V3H + 3V2a;), (6.4.31)
E d -N 3 h*d a&*1/2(2H_h*)
- &pg Hl/22V3 d, (6.4.32)
W = ~N pgH- 2(-3/2S2 T* c* (6.4.33)

2 &
where T&* = C 3/ 2T& (H/g)1/2 is the dimensional soliton generation period,

= =
h'd (H hd the dimensional depth of the depression, a; (H a& the dimensional
amplitude of the solitons, and S = (3/2 H2 P.
Now comes the question when the estimation formulas (6.4.26), (6.4.27),
(6.4.31), (6.4.32) and (6.4.33) are valid. The transcritical range has been found
to be (AL,AC) with
and

Because hd + A = 0, we have T3 --t 00 as A --t AL + by (6.4.27). Conse-
quently, it takes infinitely long time for a soliton to mature and to be radiated
upstream. Therefore, the mass postulate becomes invalid and formulas (6.4.26)
and (6.4.27) become inaccurate when A is near AL. Numerical results confirmed
this conclusion. For a different and more complicated reason, the estimation
formulas are become invalid when A = Ac. Therefore, the formulas for the
approximate solutions are valid only in a subinterval of (AL' AC).
Further discussions about the approximate solutions can be found in papers
by Grimshaw and Smyth (1986), Wu (1987) and Lee et al. (1989).
6.4.2 Spectral method for finding locally forced solitons

It appears that there is no method available that solves the IVP for the fKdV
analytically. Here we extend the Chan-Kerkhoven scheme for the unforced KdV
equation to solve the following forced KdV equation:
(6.4.34)
where a < 0, f3 <

and P are constants. This equation is integrated in
time by the leap-frog finite difference scheme in the spectrum space. The
infinite interval is replaced by -L < z < L with L sufficiently large such that
the periodicity condition u( -L, t) = u(L, t) is true in the sense of accurate
approximation. In order to simplify the Fourier transform we introduce X =
Hz + L) so that the solution is 211'-periodic. We also write s = f ' v(X, t) =
u(x, t) and w = sv 2 Thus the equation (6.4.34) is now
(6.4.35)
For the numerical solution of equation (6.4.35) we discretize the interval [0,211']
by N + 1 equidistant points Xo, Xl, ... , XN-l, XN so that !:lX = 211'/N and
denote the approximation of v(Xj, t) by V(Xj, t). We always take N to be a
power of 2 and let M = N /2. The discrete Fourier transform of V(Xj, t) for
j = 0,1,2, ... , N - 1 is denoted by V(p, t):
. N-l
V(p, t) = ~ ~ V(Xj, t)e-(271"jp/N)i, p = -M, -M + 1, ... , M -1 (6.4.36)
V IV 3=0
=
where i vCT. We then take the Fourier transform of equation (6.4.35) with
respect to X and use the following approximation:
t"T ( ),..., V(p, t + .6.t) - V(p, t - .6.t)

Yt p, t ,.., 2.6.t '
3V" ( ) 3 V(p, t + .6.t) + V(p, t - .6.t)
P p,t ~ p' 2 .
Given V(Xj, t) and V(Xj, t - .6.t) for j = 0,1, ... , N - 1, we can calculate
V(p, t + .6.t) from the formula
(6.4.37)
where W(p, t) is the Fourier transform of W(Xj, t) = 3s[V(Xj, t)j2, j =

0,1,2,, M. The last term in the above equation is the Fourier transform
of the forcing term. From V(p, t + .6.t) we can calculate V(Xj, t + .6.t) by the
inverse Fourier transformation formula
V(Xj, t + .6.t) = Jw L: M-1
p=-M
V(p, t + .6.t)e(21fjp /N)i, j = 0,1,2, ... , N - 1.
(6.4.38)
In the rest of this section, we show how to implement the above numerical
scheme using Mathematica.
In order to implement the above algorithm using Mathematica and take
advantage of its command for the discrete Fourier transform, we change the
indices j = 0,1, ... , N - 1 and p = -M, -M + 1, ... , M - 1 to k = j + 1 and
=
q p+ M + 1 respectively. Thus we have U(q, t) for q 1,2, ... , N -1 in place =
of V(p, t) where
~ L: U(k, t)e- 21fi (k-1)(q-N/2-1)/N,

N
U(q, t) = q = 1,2, ... , N. (6.4.39)
V IV k=l
Similarly, we have U(k, t) in place of V(Xj, t) such that

N-1
U(k, t) = .~ L: U(q, t)e 21fi (k-1)(q-N/2-1)/N, k = 1,2, ... , N. (6.4.40)
V IV q=l
The Mathematica code is within the form of a package forcedKdV.m which can
be found in the next subsection. In the package, u, urn, up denote U(., t), U(., t-
.6.t), U(., t + .6.t) respectively. Similarly ut, urnt, upt represent their Fourier
transforms. The parameters of the equation, N,L,a,/3,p,>", and b..t are repre-
sented by n, I, alpha, beta, p, lambda, and deltat. The package forcedKdV . m
contains the following definitions:
1. 'fourier[g], performs the Fourier transform of a list g; (see equation (6.4.39)).
2. 'invfourier[gt], performs the inverse fourier transform of a list gt; (see

equation (6.4.40)).
3. 'step' calculates one time step; (see equation (6.4.37)).
4. 'tsteps[m]' calculates 'm' time steps. The parameters 'm' and 'timesave'
are integers supplied by the user. Intermediate results are overwritten in
a file called "intres" at intervals of 'timesave' steps. Once 'm' is chosen,
the value of 'timesave' should be chosen so that it divides 'm'.
5. 'msteps' calculates blocks of 'tsteps[m]' and saves the solution u in "res"

files, such as "res01", "resI6", at fixed intervals of 'm' time steps. The
number of "res" files is equal to 'lastres' minus 'firstres' plus one.
6. 'plot' displays the solution u graphically from any "res" file that is loaded.
To implement the package, we proceed with the following steps:
(i) Load the package <forcedKdV.m).
(ii) Load the values of the following parameters:
the parameters appearing in the equation (n, I, alpha, beta, p,

lambda, and deltat).
'm' is chosen so that the results are saved in "res" files after every
time interval of 'm' times b..t.
'first res , determines the name of the first "res" file created in the
current session. For example, if 'firstres' = 1, then the first saved
result is in "resOl". If 'firstres' = 15, then the first saved result is in
"resI5" .
'lastres' determines the name of the last "res" file created in this
seSSIOn.
'timesave' is chosen so that the intermediate results are overwritten
in the "intres" file after every time interval of 'timesave' times b..t.
For example, if 'timesave' = 5, then the results are overwritten in
the "intres" file after every time interval of 5b..t. The time at which
the results are overwritten appears on the output device.
These initial values can be declared by either loading them directly or

putting them into a file and then loading the file.
15
J
-1#
I
12
9
(Y'
500 x (grid points)

100 200 300 400
Figure 6.10: A transcritical upstream radiated soliton solution of the fKdV

equation.
(iii) To start the calculations from time t=O:

issue the command 'forcedKdV'. This calculates 'm' x 'lastres' +1
time steps. At the intervals set by 'timesave', sufficient intermediate
results are saved in the "intres" file to permit the calculation of
additional time steps.
(iv) To obtain results for additional time steps:
load the program if it is not already loaded.
supply new values for 'firstres' and 'lastres'. The value of 'firstres'
should be one more than the number of the final res file from the
previous run. The value for 'lastres' should be the final res file now
desired from the supplementary run.
issue the 'forcedKdV' command which loads all the necessary pa-
rameters and values from the "intres" file, performs further time
steps and creates additional "res" files. If the parameter 'firstres' is
greater than 1, the same command 'forcedKdV' automatically does
the supplementary run for additional time steps.
(v) To plot the results from any res file:
load the res file.
issue the 'plot' command.
Once the "res" files are created, either the numerical values of u can be
viewed directly III the "res" files or u can be plotted by issuing the 'plot'
command.
As a numerical example, we use the following values of parameters: n=512j

1=80j alpha=-3/4j beta=-1/6j p=lj lambda=O.Oj deltat=O.lj m=lOj firstres=lj
lastres=15j timesave=10. Fig. 6.10 shows the evolution of the solution u(x, t)
from the initial zero profile.
As a consequence the numerical results are saved in files "res01", "res02" , ...
"res15" at times t=l, t=2, ... , t=15 respectively. The "intres" file after running
the package contains all the parameters and the required results at t=15. This
final "intres" file can be used to start a supplementary run in order to get
results for t greater than 15. For example, if 'lastres' = 15 in the previous
session, then in the supplementary run 'firstres' must be 16. If one wants to
run up to the time equal to 80m.6.t, then the 'lastres' in the supplementary
run should be 80.
6.4.3 Program of the spectral scheme in Mathematica
(******************************************************)
(* *)
(* Program name: forcedKdV.m *)
(* *)
(* This program solves forced KdV equation of the form*)
(* *)
(*u + lambda u + 2 alpha u u + beta u =(1/2) sigma *)
(* t x xxx x *)
(* using the spectral method. *)
(* The forcing function sigma(x) = P DiracDelta(x). *)
(* The initial condition is u(x,t=O) =0. *)
(* *)
(* The student version, 1992 *)
(* by R. P. Manohar, L. Quinlan *)
(* University of Saskatchewan *)
(* and *)
(* S. S. Shen *)
(* University of Alberta *)
(* Canada *)
(* *)
(******************************************************)
BeginPackage["forcedKdV''']
forcedKdV::usage = "forcedKdV solves the equation
using m*lastres time steps"
plot: : usage = "plots the solution u at a fixed
time tot
(*The following parameters require inital values

for the program to run:
- the parameters appearing in the equation

(n, 1, alpha, beta, p, lambda, and del tat)
- 'm' controls the number of time steps done
before an "res" file is created containing
the results.
- 'firstres' determines the number of the first
"res" file created (for the initial run, set
firstres = 1).
- 'lastres' controls the total number of "res"
files created.
- 'timesave' determines how often the
intermediate results are saved in the
"intres" file. The time at which the results
are saved is printed on the output device. *)
l=lambda=alpha=beta=p=n=m=firstres=lastres\
=deltat=timesave=tottime=time=umt=ut=reu=s=O;
Begin[" 'Private''']
(*Procedures Section:*)
(* 1): The command "fourier[g_List]" gives the

fourier transform of a given list, g. The
command "invfourier[gt_List]" gives the
inverse fourier transform of a list, gt.
Modifications are introduced to take
advantage of the fast fourier transform in
Mathematica. The letter t denotes a
transformed list. *)
fourier[g_List]:=
Block[{k,n,a},
n = Length[g];
a = Table[(-1)~(k-1)g[[k]],{k,1,n}];
InverseFourier[a]
]; (*end of fourier Block*)
invfourier[gt_List]:=
Block[{k,n,a},
n = Length[gt];
a = Fourier[gt];
Table [( -1) ~ (k-1)a[[k]] ,{k,1,n}]
]; (*end of invfourier Block*)
(* 2): The command

"step [umt_List,ut_List,wt_List]"
calculates the list upt from the given
lists umt,ut and wt; that is, one time step
is calculated using the formula for ut at
t+de1tat. The letter p denotes the values
at t+de1tat and m the values at t-de1tat
for u and w.*)
step[umt_List,ut_List,wt_List]:=
B1ock[{j,q,n,a,tl,t2,t3},
n = Length[ut];
Do[tl = q - n/2. - 1;
t2 = N[l/(l. - beta * de1tat *
s~3 * I * tl~3)];
t3 = N[(-l)~(q+l) * P * de1tat *
I * s * tl* N[Sqrt[n]]/(2 1)];
a[q] = t2 * 1 + beta * de1tat * I
* s~3 * tl~3) * umt[[q]]
- 2. lambda de1tat s I tl ut[[q]]
- 2. alpha deltat I tl wt[[q]]
+ t3),
{q,l,n}]; (*end of Do loop*)
Table [a [j] ,{j , 1 ,n}]
]; (*end of step B1ock*)
(* 3): The command "tstep[nsteps]" calculates

'nsteps' time steps; the value of nsteps is
supplied by the initial parameter m. "tsteps"
requires u, umt, ut and uses "step" repeatedly.
It also saves the intermediate results in a
file called intres after a certain number of
time steps specified by the parameter 'timesave'.
'time' measures the number of time steps in m.*)
tsteps[nsteps_]:=
B1ock[{j},
Do[w = s u~2; wt = fourier[w];
upt = step[umt,ut,wt];
up = invfourier[upt];
umt = ut; ut = upt; u = up;
(*updating 1ists*)
time = time + 1;
(*updating time variab1es*)
tottime = del tat + tottime;
If [Mod[time-l,timesave]==O,
(*then*)
Print[tottime]; "!rm intres"; save,

(*else*)
Continue],
{j,1,nsteps}];(*end of Do loop*)
]; (*end of tsteps Block*)
(* 4): Initially, the values of the list umt=O

at t=O. The list ut is calculated at deltat/10
and then u is obtained. With this small
deltat, ie. deltat/10, nine additional steps
are calculated to remove any disturbance due
to initial approximation of u at deltat.*)
init:=
Block[{},
time=tottime=O.;
umt=Table[O,{q,1,n}]; Clear[q]; time=1;
s=N[Pi/l] ;
deltat=deltat/10.;
ut=Table[O.6(-1)~(q+1)*p*s*deltat
*N[Sqrt[n]]*I*(q-n/2-1)/(2. 1),
{q,1,n}]; Clear[q];
u=invfourier[ut];
tottime=tottime + deltat;
save;
tsteps [9];
deltat=10.*deltat;
umt=Table[O,{q,1,n}];
time=1 ;
]; (*end of init Block*)
(* 6): 'msteps' calculates blocks of 'tsteps[m]'

and saves the solution u in "res" files at
fixed intervals of 'm' time steps. The number
of "res" files depends upon the parameters
'firstres' and 'lastres', supplied by the user.*)
msteps:=
Block [{i},
Do [tsteps Em] ;
end = ToString[i];
If [i > 9,
(*then*) file = StringJoin["res",end],
(*else*) file = StringJoin["resO",end]];
reu = Re[u];
Save[file,reu,tottime,lambda,deltat],
{i,firstres,lastres}];
(*end of Do loop*)
]; (*end of msteps Block*)
(* 6): The "forcedKdV" block contains all the

commands needed to run the program once
the initial values are specified in the "data"
file, and the package, forcedKdV, has been
loaded using "forcedKdV". *)
forcedKdV:=
Block[{},
If [firstres==O,
(*then*)
Return[Print["If first run,
data file is not loaded;
if sup run, value for
firstres is missing."]]
] ;
If [firstres==l,
(*then*) init,
(*else*) intres;
u = invfourier[ut];
] ;
msteps;
]; (*end of forcedKdV Block*)
(* 7): Bookkeeping commands:

The command "save" is designed to save the
intermediate results and is used in "tstep" above.
Similarly the command "plot" is
used to plot the real part of u which is saved
in files named res##.*)
save:=Save["intres",l,lambda,n,m,alpha,beta,p,
deltat,s, timesave,tottime,
time,umt,ut];
plot:=
Block[{templ,temp2,tl,t2,t3,t4},
templ = Min[reu];
temp2 = Max[reu];
tl = N[tottime - deltat,4];
t2 = ToString[tl];
t3 = ToString[lambda];
6.5. Stability of Solitary Waves 183
t4 = StringJoin["time=",t2,
"lambda=",t3];
ListPlot[reu,PlotJoined->True,
PlotRange->{temp1,temp2},
PlotLabel->t4]
]; (*end of plot Block*)
(*----------end of Procedures Section ----------*)
End[ ]
EndPackage [ ]
6.5 Stability of Solitary Waves

Stationary solitary wave solutions of a forced KdV equation were described in
section 6.2. One may recall that there are multiple solitary wave solutions. It
was pointed out that in the case of two solitary waves the upper one is stable
and the lower one is unstable. In this section we will demonstrate this claim
numerically.
The problem of solitary wave stability may be loosely described in the
following way for a locally forced KdV. Consider numerical solution of the
IVP for the fKdV:
p
Ut + AU~ + 2auu~ + f3u~~~ = 2d~(Z), -00 < Z < 00, (6.5.1)
u(z, t = 0) = uo(z) (6.5.2)
where uo(z) is a stationary solitary wave solution of the fKdV. Solve this IVP
numerically up to a certain length of time to . If the profile of u( z, to) is almost
same as that of uo(z), then we say that the solitary wave uo(z) is stable. If
the profile of uo(z, to) is dramatically different from that of uo(z), then we say
that the solitary wave uo(z) is unstable.
In the above statement, we need to clarify two points. One point is where
the small perturbation comes into the problem since the stability and instabil-
ity are relative to small perturbations. Because every numerical scheme is not
100% accurate, this numerical error (sometimes called the numerical noise) is
the small perturbation. The other point is why the change has to be "dra-
matical". Indeed, this may not be true universally. But for most systems, the
instability is a manifestation of either an exponential growth (for most non-
linear dispersive systems) or an algebraic growth (for second order nonlinear
hyperbolic systems). Despite that there are some neutrally stable solutions to
certain systems, the numerical computations can not catch the evolution of
1.4
1.2
0.8
o.
o.
-20 -10 10 20
X.
Figure 6.11: Two solitary wave solutions of the stationary locally forced KdV
equation.
these solutions since the regime in which the neutrally stable solutions exist is
so narrow that it is smaller than the tolerance of numerical errors. Therefore,
numerical demonstration of instability either follows an exponential growth or
an algebraic growth and both growths are dramatic.
We summarize the way to find uo{x) in section 6.2.1. A solitary wave uo{x)
is a solution of the following BVP:
P
>.uox + 2auuox + {3uoxxx = "28x{x), -00 < x < 00, (6.5.3)
uo{oo) = uox{oo) = uoxx{oo) = o. (6.5.4)
A solution can be expressed analytically by
uo{x) == --sech
3>'
2a
2{J>.
- {
4{3
(x - xo),
(x + xo),
x> 0,
x< o.
(6.5.5)
The phase shift Xo is determined by
Xo = )-4{3
T arctanh{b) (6.5.6)
where b E (-1, 1) is a solution of the following cubic algebraic equation
3 Pa 1
f - f + 6>: -)-(3). = O. (6.5.7)
time=5. lambda=0.6
1.4
1.2
O.
o.
o.
-20 -10 10 20
"%
Figure 6.12: Numerical demonstration of the stability of an upper solitary wave
solution of the stationary locally forced KdV equation.
time=5. lambda=0.6
1.4
1.2
0.8
0.6
0.4
-20 - 0 ox.
Figure 6.13: Numerical demonstration of the instability of a lower solitary wave
solution of the stationary locally forced KdV equation.
As A > AG = (3n 2p2/(-16,8))1/3, equation (6.5.7) has two solutions in

(-1,1). Hence, there are two solitary wave solutions. When P > 0, the two
solutions are ordered, i.e. one is above the other. If P = 1.0, n = -3/4,,8 =
-1/6, and A = 0.6, then the upper (lower) solitary wave corresponds to Xo =
-0.294817 ( Xo = -0.72709). These two solutions are shown in Fig. 6.11.
One can easily modify the numerical scheme described in section 6.4.2 to
solve the IVP (6.5.1) -(6.5.2). Fig. 6.12 demonstrates that the upper solitary
wave (xo = -0.294817) is stable since it shows the profile u(x,5) (i.e. to = 5)
which is almost the same as the initial profile. The small difference is due to
the numerical noise.
Fig. 6.13 demonstrates the instability of the lower solitary wave (xo =
-0.72709). Fig. 6.13(a) shows the initial profile uo(x). Fig. 6.13(b) shows
the profile u(x, 5) (i.e. to = 5) which is dramatically different from the initial
profile.
An extensive study of the stability of forced solitary waves was conducted
by Camassa and Wu (1991). Interested readers may read their paper.
[1] T. Yao-Tsu Wu (1987), Generation of upstream advancing solitons by

moving disturbances, J. Fluid Mech. 184,75-100.
[2] T. R. Akylas (1984), On the excitation of long nonlinear water waves by a
moving pressure distribution, J. Fluid Mech. 141,455-466.
[3] P. G. Baines (1987), Upstream blocking and airflow over mountains, Ann.
Rev. Fluid Mech. 19, 75-97.
[4] R. Grimshaw (1987), Resonant forcing of barotropic coast ally trapped
waves, J. Phys. Oceanogr. 17, 53-65.
[5] S. S. Shen (1989), Disturbed critical surface waves in a channel of arbitrary
cross section, J. Appl. Math. Phys. (ZAMP) 40,216-229.
[6] J. W. Miles (1986), Stationary, transcritical channel flow, J. Fluid Mech.
162, 489-499.
flows, J. Fluid Mech. 234,583-612.
York.
[9] H. Lamb (1945), Hydrodynamics, 6th ed., Dover Publications, New York.
[10] S. S. Shen and M. C. Shen (1990), A new equilibrium of subcritical flow
over an obstruction in a channel of arbitrary cross section, Euro. J. Mech.
B/Fluids 9,59-74.
[11] S. J. Lee, G. T. Yates and T. Y. Wu (1989), Experiments and analyses

of upstream-advancing solitary waves generated by moving disturbances,
J. Fluid Mech. 199,569-593.
[12] R. H. J. Grimshaw and N. Smyth (1986), Resonant flow of a stratified
fluid over topography, J. Fluid Mech. 169,429-464.
[13] R. Camassa and T. Y. Wu (1991), Stability offorced solitary waves, Phil.
Trans. R. Roy. Lond. A 337,429-466.
[14] S. S. Shen (1991), Locally forced critical surface waves in channels of
arbitrary cross section, J. Appl. Math. Phys. (ZAMP) 42, 122-138.
Chapter 7
Sine-Gordon and
Nonlinear Schrodinger
The sine-Gordon equation is a very important partial differential equation not

only in the modern theory of condensed matter physics but also in many other
fields of sciences, such as chemical reaction kinetics and high energy physics.
It models far more physical phenomena than the Korteweg-de Vries equation
(KdV). It is known from Chapter 4 that an initial value problem (IVP) for the
KdV may yield a soliton solution which can be found analytically using the
inverse scattering method. The sine-Gordon equation also possesses soliton
solutions and its IVP can also be solved by the inverse scattering method.
The (cubic) nonlinear Schrodinger equation (NLS) can be derived from
the modulation of deep water surface waves or from Maxwell equations in a
medium whose diffraction index is weakly but not infinitesimally dependent
on the electric field (i.e., n = no(w) + n2(w)IEI2). It has also been found
189
190 Chapter 7. Sine-Gordon and Nonlinear SchrOdinger
that the nonlinear Schrodinger equation possesses soliton solutions and that
its initial value problem can also be solved by the inverse scattering method.
The nonlinear Schrodinger solitons constitute the foundation of the theory of
nonlinear transmission lines in modern optic communication technologies.
Since the beginning of 1970s, many interesting and important results have
been discovered about the sine-Gordon equation (sG). It is impossible to de-
scribe all of these results in this chapter. Hence, in this chapter I will discuss
only a few of the topics on these two equations which in my personal opin-
ion are among the most elementary and fundamental. These topics are listed
below:
(i) Physical phenomena characterized by the sine-Gordon equation,
(ii) Some special solutions of the sine-Gordon equation,
(iii) Pulse broadening due to dispersion and beam focusing due to nonlinearity
in an optic fiber (derivation of the nonlinear SchrOdinger equation),
(iv) Some traveling wave solutions of the nonlinear Schrodinger equation.
The inverse scattering method for the sine-Gordon equation and nonlinear
Schrodinger equation is a little more sophisticated than that for the KdV and
hence is not included in this book. Interested readers may learn about the
inverse scattering method to solve the sG and the NLS from the book by
Drazin and Johnson (1989) and that by Ablowitz and Clarkson (1991).
7.1 Physical Models of the Sine-Gordon Equa-

tion
The sine-Gordon equation
ipxx - iptt = sin ip (7.1.1)
was first derived when studying the geometry of the surface whose Gaussian
curvature equal to -1. Later it has been used to model numerous physical
phenomena, which range from the motion of coupled pendulums to unitary
field theory. The book by Rebbi and Soliani (1984) has collected various types
of physical models described by the sG and other soliton equations. The book
by Dodd et al. (1982) is another good source to find physical models for
nonlinear evolution equations such as the sG and NLS. We list a few examples
of applications as follows:
(i) Mechanical model: continuously coupled pendulums acting under gravity

while fixed on an elastic bar, where ip measures the displacement angle
of the pendulums.
7.1. Physical Models of the Sine-Gordon Equation 191
(ii) One-dimensional solid crystal dislocation under a cos t.p like potential,
where t.p is the measurement of the crystal displacement.
(iii) Magnetic flux in a thin Josephson junction, where t.p is the phase difference
of electric wave functions in the two superconductors being joined.
(iv) Plane wave propagation in a ferromagnetic or antiferromagnetic field,
where t.p is the angle between the direction of magnetization and that of
the external magnetic field.
(v) Propagation of ultra short pulses (lasers, for instance) in a two state media,
where E = 8t.p / at describes the electric field modulation (envelope).
(vi) Unitary theory of elementary particles where the Lagrangian of the field
is
(vii) Waves in superfluid 3He , where a double sine-Gordon equation is proved

to be a proper mathematical model.
In this section we describe only the first three examples. A reader interested
in knowing details of other models should read the relevant additional reading
materials listed at the end of this chapter.
7.1.1 Coupled torsional pendulums

A chain of pendulums is evenly mounted on an elastic bar which can store
torsional elastic energy: ET = (1/2)KE (t.pj+i_t.pj)2. Here, t.pj is the angular
displacement of the pendulum j and K is the elastic constant of the elastic bar.
Each pendulum has mass m and rotational inertial moment I (see Fig. 7.1).
In the gravitational field, the equations of motion of the chain of the pen-
dulums are
d2t.p .
I dt 2J = -mg 1sin t.pj + K (t.pj+i + t.pj-i - 2t.pj) , j = 1,2,, N (7.1.2)
where N is the total number of pendulums.

Let a be the distance between two adjacent pendulums. As a -t 0, the
continuity limit of (7.1.2) becomes
It.ptt - Ka 2t.p:c:c + mglsint.p = 0, (7.1.3)
or
2
t.ptt - Cot.p:c:c + Wo2 sm

t.p = 0 (7.1.4)
where c~ Ka 2/I and w~ = mgl/I. Since the dimension of I is [ML2]
and the dimension of K is [M L 2 T- 2 ], the dimensions of Co and Wo can be
derived as [LT- i ] and [T- i ] respectively. Therefore, the wave propagation
P=F=~~~~~~~~====~~---x
Elastic bar
y
z
Figure 7.1: Coupled pendulums in a gravitational field
speed is Co and the pendulum oscillation frequency is woo The consequence

that the wave propagation speed is proportional to the elastic coefficient K and
inversely proportional to the inertial moment is intuitively acceptable. From
the expression of Wo = (m 9 l / I) 1/2, it seems that the oscillation frequency is
proportional to the pendulum arm length I. This is a false interpretation of
the expression and this interpretation contradicts the physical intuition. If we
notice that I is proportional to 12, then it is clear that the frequency Wo is
inversely proportional to I as one expects.
7.1.2 One-dimensional crystal dislocation

Consider two chains of crystal vertices. The lower chain is fixed and the upper
chain is about to be dislocated (see Fig. 7.2). The total interaction energy is
(7.1.5)
where ej is the displacement of the particle j and K is the elastic constant
of the crystal. Hence the system is favorable to ej+1 = ej (i.e., the uniform
dislocation state). The forces acting on the jth particle includes K (ej+1 - ej),
K (ej-1 - ej) and -(d/de)( -A cos (211'e/a)). The equation of motion for the
jth particle on the upper chain is
(7.1.6)
where m is the mass of each particle on the upper chain. Let <p = 211'e/ a. As
0, the continuum limit of (7.1.6) becomes
a --t
2
<Ptt - co<Prcrc + Wo2 sm

<P = 0 (7.1.7)
V!/ I
,, , ,,
IIil II i J
: : :
IIV
: : :
: :
.!
II II II II
if if V II 1/ !
j-2 j-l j+l
Restoring Potential: -cos(2Pi xl
Figure 7.2: One-dimensional crystal dislocation
where w~ = (27r/a)2(A/m) and c~ = Ka 2/m. Dimensions of K and A are

[MT- 2] and [M L 2T- 2] respectively. Dimensions of Wo and Co can be derived.
The result is [T-l] for Wo and [TL-l] for Co. Therefore, Co represents the
propagation speed of the dislocation wave and Wo the oscillation frequency of
a dislocated particle about its equilibrium.
7.1.3 Magnetic flux in a long one-dimensional Josephson

junction
Two pieces of thin superconductors are joined together by a very thin insulator
as shown in Fig. 7.3. Such a device is called the Josephson junction. Josephson
(1964) announced that there was a tunneling current through the insulator even
in the absence of a voltage between the two superconductors and this tunneling
current altered its direction rapidly.
Here, we consider the so called a.c. Josephson effect: a voltage V is applied
across the junction and an alternating current flows through the junction. The
direction of this Josephson current alternates at a microwave frequency (2e V In)
where n = 6.626 X 1O-34[joule. sec] is the Planck's constant and e is the charge
of an electron. By l[electronvolt] = 1.60218925 x 1O-19[joule], if the applied
voltage is l[volt], the alternating frequency is about 0.5[P H z] ( l[petahertz] =
1015[hertz]). This frequency is near ultraviolet radiation frequency and hence
is extremely high. Its wave length is about 0.6[JLmJ = 6 x 1O-7[mJ. Hence,
the Josephson junction is a good micrometer microwave source if a voltage
as high as l[volt] can be applied to the device. Based upon this principle, a
type of device called SQUIDS (Superconducting QUantum Interference Device)
has been extensively studied since the 1970s. SQUIDS has many industrial
The thin insulator
x
Applied magnetic field
,,,.,,,,,,,,,,.,~. H
Figure 7.3: Josephson junction of two superconductors
applications when certain microwave sources are needed, such as high frequency
digital operations in logic circuits.
Dimensions of the long one-dimensional Josephson junction are as follows:
the thickness of the thin junction layer is about d = 25A ( lA = lAngstrom =
lO-lOmeter), the thickness of the superconductors is in the order of several
thousands Angstroms, the width of the junction W is in the order of microme-
ters (10 4 A), and the longitudinal dimension may vary from several micrometers
to several decimeters. Hence, this conventional Josephson junction is a very
small device and has a very thin rectangular box shape. Since SQUIDS has an
even smaller volume than semiconductor devices, it is amenable for a variety
of industrial applications. Although modern SQUIDS have various types of
shapes, we consider only this conventional shape for our theoretical interest in
the fundamental theory.
The thin insulator barrier permits the quantum tunneling of superconduct-
ing electron from one piece of superconductor layer to the other. The electric
wave functions in the two superconductors are
(7.1.8)
Here P2 and P2 are the electron charge densities on superconductors 1 and 2

respectively. The Josephson tunneling current I per unit length in the width
dimension is related to the phase difference
(7.1.9)
The relation is
1= 10 sin cp (7.1.10)
where 10 is the maximum dc (direct current) Josephson current which depends
on the materials of the barrier, the junction geometry and the environmental
temperature. As mentioned above, the current oscillation (described by the
L L L
x x+dx
Figure 7.4: The L-C circuits in a Josephson junction.
changing rate of the phase difference) is related to the applied voltage. This
relation is
(7.1.11)
where rpo = h/2e = 2 x 1O-15[voltHsec] is the reciprocal of frequency per unit

voltage.
We regard the Josephson junction as a collection of infinitely many L-C
circuits as shown in Fig. 7.4. When applying Kirchoff's law of the conservation
of electric current to an infinitesimal section from x to x + dx, one has
i[x + dx, t] - i[x, t] + (Cdx) a;; + (10 sin rp)dx = O.
where i is the electrical current in the width dimension and C is the elec-
tric capacitance per unit length in the width dimension. Dividing the above
expression by dx and letting dx -+ 0, one has
ai
-
av .
ax = - Cat- - 10sInrp. (7.1.12)
Another useful relationship between i and V is the so called the gauge

invariant property:
av _ _Lai
ax - at (7.1.13)
where L is the inductance per unit length in the width dimension.
Combining expressions (7.1.12), (7.1.13) and (7.1.11), one can obtain the
sine-Gordon equation:
(7.1.14)
The length scale and the time scale can be normalized by the Josephson pen-
etration length >'J = (rpo/(21rLlo))1/2 and the Josephson plasma frequency
WJ = (27rlo/(CrpOW/2 respectively:
x
x--- >,J' T = t WJ.
The value of )"J varies in a large range. A typical value may be in the order of
lOO[Jlm), thousands times ofthe thickness of the superconducting layer. With
these explicit length and time scales, the above sine-Gordon equation can be
normalized to the standard one as we wanted to derive:
(7.1.15)
7.2 Solutions of the Sine-Gordon Equation
In the last section, we have seen that the sine-Gordon equation may model
various physical phenomena. To describe evolution of the physical systems and
the corresponding physical properties, we need to solve the sG. In this sec-
tion, we will discuss a special solution technique due to G. L. Lamb, Jr. This
technique is straightforward and can be easily mastered by students, yet the
method reveals all the important properties, such as kink collisions and oscil-
latory breathers, of the solutions of various IVPs for the sG. A general way to
find the analytic solution of an IVP for the sG is the inverse scattering method.
It requires to solve a 2 x 2 matrix eigenvalue problem. In this matrix, the di-
agonal elements are differential operators {) / {)x. The whole inverse scattering
process is a little too sophisticated to be discussed here. Interested readers are
referred to the book by Drazin and Johnson (1989) and that by Ablowitz and
Clarkson (1991).
Consider the sine-Gordon equation
CPxx - CPtt = sin cpo (7.2.1)
We want to look for solutions to this equation of the following form
U(x))
cP = 4 arctan ( v(t) . (7.2.2)
This method is due to G. L. Lamb, Jr. according to Barone et al. (1971). Sub-
stituting (7.2.2) into (7.2.1), one can derive the following (see the last section
of this chapter):
1 ( u" ) , 1 ( v" ) ,
UU' -;- (x) =- vv' -; (t). (7.2.3)
Equation (7.2.3) holds for all x and t only if both sides of (7.2.3) are equal to
a same constant. Let this constant be -4k 2 Then
U")' (x) = -4k uu',

(-;- 2 (7.2.4)
and
v" )' (t) = 4k 2 vv'.
( -; (7.2.5)
7.2. Solutions of the Sine-Gordon Equation 197
Integrations of (7.2.4) - (7.2.5) give (for the derivation details see the ap-
pendix of this chapter)
(U')2 = _k 2u 4 + m 2u 2 + n 2, (7.2.6)
(v')2 = k 2v4 + (m 2 _ l)v 2 _ n 2. (7.2.7)
Here m 2 and n 2 are integration constants.

Next we investigate several special cases for different values of the integra-
tion constants k, m and n.
Case I Single kink: =
k 0, m 2 > 1, n O. =
In this situation, equations (7.2.6) - (7.2.7) yield
(7.2.8)
where Cl and C2 are integration constants. Hence
<p(x,t) = 4 arctan [I'exp ( n)] (7.2.9)
where I' = Ct/C2, and {i = Jm 2 - l/m is the speed of the traveling wave.
Equation (7.2.11) represents a class of traveling wave solutions of the sine-
Gordon equation (7.2.1). The wave traveling speed is determined by the initial
condition (<po(x) = 4 arc tan ( I'exp(x/vl- (i2)) where I' is determined by
the height (denoted by 1") of the initial profile at the middle point of the
transition zone that connects the <p = -211" uniform state to the <p = 211" uni-
form state. As a matter of fact, this traveling wave solution can be directly
found from (7.2.1) by letting <p = <p(x - (it) and using the boundary conditions
<p(oo) = 211".
Let , = x - {it. Then the sG becomes
(1 - {i2)<p(( = sin <po
Multiplying this equation by <PC and integrating the resulting equation with
respect to the independent variable from -00 to " one obtains
1 - {i2 2
-2-<P( = cos<p-1.
This expression can be written as
Vl-2 {i2 <Pc = SlD"2.

. <p
The following operation
VI - (i21'" _d,-u...,.....,.. =
2 ..,' sin( u/2)
i(
0
ds
yields
i.p ( "('
4 = V~
Intan-
1- {32 4
+lntan-
After solving for i.p in the above, one obtains the single kink solution (7.2.9).
For every real m (Iml > 1), the solution (7.2.9) is a stable traveling wave
that joins the -211" state at -00 and the 211" state at 00 together by a smooth
transition state. That is why the solution is called the single kink. The stability
can be seen most easily from the pendulums model by one's physical intuition.
Satisfying the sine-Gordon equation, the solution (7.2.9) arranges the lowest
potential energy for the pendulum system with the constraint that all the
pendulums near both ends (z = oo) of the elastic bar hang in a vertical
position and the pendulum chain has two complete 360 0 twists. This solution
is a moving smooth state transition and the transition zone consists of the
twists. Any disturbance would increase the internal (potential) energy of the
system. The solution is insensitive to small perturbations and is hence stable.
When talking about solitons, people usually refer to the bell shape solitons
which die out at infinity and keep their original shape after collision. Here, we
take away the requirement of the "bell shape" for the definition of solitons. The
soliton property of a wave is more generally defined the property of the wave
resuming its original shape after its collision with other solitons. Under this
general definition of solitons, one can actually show that the above single kink
traveling wave has the soliton property. As a matter of fact, this kink can be
viewed as a soliton step as discussed in Chapter 4. Its derivative with respect
to z is an ordinary bell shape soliton that dies out at infinity. The derivative
of this kink is
exp(~)
i.px(z, t) = 4"( v1-/P . (7.2.10)
VI - {32 1 + "(2 exp ( 2~)
V -f32 1
If {3 = 0.8 and "( = 1, then
i.p(z, t) = 4 arctan [ exp ((z - 0.8t) /0.6)]
and
20 exp [(z - 0.8t)/0.6]
i.px(z,t) = - [ ]
3 1 + exp (z - 0.8t)/0.3
are shown in Figs. 7.5 (a) and (b) respectively.

Case II State splitting: k = 0, m> 1, n # 0.
In this situation, equations (7.2.6) - (7.2.7) yield that
u = (n/m) sinh(mz + cd, (7.2.11)

v = (n/Vm2 - 1) cosh ( V~~-2---1 t + C2). (7.2.12)
7.2. Solutions to the Sine-Gordon Equation 199
(a)
phi
(b)
x
Figure 7.5: (a) Single kink solution of the sG that joins the -211" state at -00
to the 211" state at 00; (b) x derivative of the kink.
Hence
vm2 - 1 sinh(mx + cd ].
IfJ ( x, t ) = 4 arctan [ (7.2.13)
m cosh(Vm2 -1 t + C2)
Obviously, these solutions are independent of n. They can model the splitting
process from a one-step state (from -271' to 271') to a two-step state ( from -271'
to 0 and then 0 to 271') as time t increases. This process is shown in Fig. 7.6
(a). The transition rate is proportional to the value of m. This transition wave
is an even function of t and an odd function of x. Its derivative with respect
to x is
(7.2.14)
This expression shows the collision process of two state-splitting sG solitons

(see Fig. 7.6 (b) when m = 2).
=
When m 2, Cl C2 0, we get = =
lfJ(x, t) = 4 arctan ( va'
V3 sinh(2X))
2 cosh ( 3t)
(7.2.15)
and
1fJz;
( x,t ) -16
-
/;;3
Vi)
cosh(2x)cosh(V3t)
2 . J7i 2' (7.2.16)
4cosh (v3t) + 3 sinh (2x)
The IfJ function is plotted in Fig. 7.6 (a) and its derivative function with respect
to x is plotted in Fig. 7.6 (b).
Case III: k =P 0, n = 0, and m falls into one of the following three
sub-cases.
Case III (a). Collision of the sG solitons m2 > 1.
Then
IfJ (x, t ) = 4 arctan [ vmm2 - 1

sinh(vm2-1t+C2)]
()
cosh mx + Cl
. (7.2.17)
This solution is related to the solution of case (II). Exchanging the numerator
and denominator of the argument of the arctan function, we can convert one
solution to the other. Notice that
arctan x + arctan (~) = i.

Hence, the explicit relation between the solution in the above two cases is
solution of case II + solution of case III (a) = i.

When Cl C2= = 0, this solution can demonstrate a collision of a soliton
kink with an antisoliton kink.
(a)
-10
5
phi
o 10 20
x
(b)
t
-10 15
o 10
x
Figure 7.6: (a) State-splitting solution ofthe sine-Gordon equation; (b) collision
process of two state-splitting sine-Gordon solitons.
When m = 2, take
2 sinh( y'3 t) 1
<p(x, t) = 4 arctan [ y'3 cosh (2x) (7.2.18)
This function is plotted in Fig. 7.7 (a). It shows the collision of a kink
soliton with an antikink soliton.
As we have seen in case (II), the derivative of a soliton kink (or called a
soliton ladder) is the usual bell shape soliton that dies out at infinity. When
Cl = C2 = 0, the derivative function <PII: can be evaluated from (7.2.17):
<PII: =4 mvm 2 1 sinh(mx) sinh(vm2 - 1 t)

2 -
(7.2.19)
(m 2 -1)cosh2(mx) +m 2 sinh2(vm2 -1 t)
Similar to case (II), this expression is the evolution equation which shows the
collision of a soliton with an antisoliton. For m =
2, let u =
<p1I:/4. The
function u(x, t) is plotted in Fig. 7.7 (b) and the plot shows the collision process
of a sine-Gordon bell shape soliton with a sine-Gordon bell shape antisoliton
(u(x, t) :::; 0).
Case III (b). Asymptotically stationary solitons m = l.
Then
<p(x, t) = -4 arctan[(t + C2) sech(x + cd]. (7.2.20)
This solution is not a traveling wave. Instead, it is asymptotically stationary
as t -+ 00. The asymptotic state consists of a soliton kink and an antisoliton
kink. The derivative of <p(x, t) with respect to x describes the collision process
of two asymptotically stationary bell shape solitons. The derivative is
<PII: = - 4 (t +1c2)sech(x + cd tanh(x + cd

2
+ (t + c2)2sech (x + cd
()
7.2.21
When Cl = C2 = 0, the functions <p(x, t) and <P1I:(x, t) are plotted in Figs.

7.8 (a) and (b) respectively.
III (c). Sine-Gordon breather m 2 < l.
Then
m sin(vl- m 2 t + c2 )]
<p(x,t) = 4arctan [ h( ). (7.2.22)
vI - m 2 cos mx + Cl
This solution is called the breather solution to the sine-Gordon equation, since
its shape looks like the breather pressure curve of an animal lung. For m =
= =
0.8, Cl C2 0, the solution <P expressed as a function of x and t is shown in
Fig. 7.9 (a).
The derivative of <P with respect to x describes a repeated collision process
of a bell shape soliton with a bell shape antisoliton. The derivative is
4m 2 VI - m 2 sinh(mx + cd sin(VI - m 2 t + C2)
<PII: (
x,)
t = =f 2 2 (7.2.23)
(1- m 2 ) cosh (mx + cd + m 2 sin (v1- m 2 t + C2)
When m = 0.8, Cl = C2 = 0, this derivative function is shown in Fig. 7.9 (b).
7.2. Solutions of the Sine-Gordon Equation 203
(a)
-10
5
phi
-20 -10 o 10 20
x
(b) t
-10 15
-20 -10 o 10
x
Figure 7.7: (a) Collision of a sine-Gordon kink with an antikinkj (b) collision
of a sine-Gordon bell shape soliton with a sine-Gordon bell shape antisoliton.
204 Chapter 7. Sine-Gordon and Nonlinear Schrodinger
(a)
-100
5
phi
x
Figure 7.8: (a) Collision of asymptotically stationary soliton kinks; (b) collision
of asymptotically stationary bell shape soliton.
(a)
-20
5
phi
(b)
-10
1
Figure 7.9: (a) Breather solution of the sine-Gordon equation; (b) repeated
collision process of sine-Gordon solitons.
7.3 Optical Self-focusing

This section consists of two subsections: Section 7.3.1 discusses pulse broad-
ening due to dispersion property of the material of a transmission line, and
section 7.3.2 discusses the balance of nonlinearity and dispersion. This balance
demonstrates that optical focusing, i.e. the elimination of the pulse broadening
effects, can be achieved in a nonlinear transmission line.
7.3.1 Pulse broadening due to dispersion

Most media in nature can transmit dispersive waves. Because of the dispersion,
the sound signals in telecommunication are broadened and hence distorted in
the process of signal transmission from one place to the other. This inevitably
requires that wave transmission stations be relatively close to each other so as
to "pick up"the true signals and "pass"them to the next station. Therefore,
it is highly desirable to find those materials in which the sound signals can be
transmitted from one place to another conform ally. The nonlinear optic fiber is
one of these kinds of materials. In these nonlinear optic fibers, the light beam,
i.e. the carrier electromagnetic waves of frequency in the order of [GHz] =
10 9 [H z] which is high compared with the human audio frequency 20[H z] '"
20[ J{ Hz], can be automatically focused and the dispersion is cancelled out by
the nonlinearity. This diffraction index relates to the electric field in the form
of the second order nonlinearity. The signals become solitons which propagate
conform ally.
In this subsection, we discuss the pulse broadening distortion due to the
dispersion effects and see how serious this dispersion problem is in telecommu-
nications. In next subsection (7.3.2) we will study the optieal focusing.
Consider a Gaussian pulse
Here T is called the width of the pulse, Wo is the frequency of the carrier
wave, and f{t) is called the modulated wave (or called a signal). Usually, Wo
is quite large (i.e. high frequency). In telecommunications, the width T is of
order 1 - 10,000 ps (I ps = 10- 12 seconds), and the frequency Wo is of order
10 Hertz. As we know, a human being's aural frequency range is 20 - 20000
9
Hertz. So, the frequency of the carrier wave is at least about 10 5 times higher
than that of signals. From the scattering analysis in Chapter 4, we learned
that the wave of higher frequency has better transmission capability. So the
carrier wave must have very high frequency to reduce the energy loss due to
the reflection! In other words, the carrier wave must oscillate much faster than
the sound wave in telecommunication devices and hence is modulated by the
sound wave. Because of the large difference of the frequencies, relative to the
number of events of the carrier wave passing, the event of the signal passing
is rare (at a ratio of lover 100,000). Thus, the signal may be regarded as
a traveling solitary wave when measured by the "clock" of the carrier wave.
For T = 0.001 and Wo = 40000, the signal f{t) is plotted in Fig. 7.10. This
7.3. Optical Self-focusing 207
f(t)
1.5
t
-0.01 -0.005 0.005 0.01
-1.5
Figure 7.10: A signal of width T = 0.001 with carrier wave frequencey Wo =
40000.
Wo = 40000 is unrealistically low in telecommunication technologies. Here we

use it only to illustrate the modulation idea since it is impossible to visualize
an oscillation of l[GH z] by our eyes.
Let F(w) be the Fourier transform of the signal /(t). In the transmission
process, the Fourier transform F(w) does not change in a linear transmission
line. Let 9(t) be the output signal which corresponds to the input signal /(t).
The phase difference between /(t) and 9(t) is kL, where k = k(w) is the wave
number and L is the distance between the input point and the output point.
So the output signal 9(t) is
9(t) = -1
211"
100
-00
F(w)e-(wt-kL)dw
. (7.3.1)
Here k and ware related by the dispersion relation k = k(w). The Taylor
expansion of k(w) about Wo is
k = k(wo) + -dw
dk IWo (w - wo) + -2!1 -dwd k IWo (w - wo)
2
2
2 + ... (7.3.2)
Substituting this expression into the 9(t) formula above, we have
9(t) = 1 00
-00
dk
F(w) exp { - i[wt - k(wo)L - dw Lo (w - wa)L-
d2 k I (w -
.!..2! dw WO)2 L - ... J}dw
2 Wo
exp{i [k(wo)- :~Iwowo] L} x

x l: F(w)exp { -;w [t - (dw~dk) LJ }

x{exp d2k l
[!..2 dw (W-W O)2 L + ... ]}dW.
2 Wo
From the expression
we see that (dw / dk) I

Wo
is the velocity at which the signal energy is transmit-
ted. This velocity is called the group velocity. Notice that the inverse Fourier
transform of exp( -(wm)2) is exp( -(t/m)2) / -..Ii. Hence
(7.3.3)
Therefore the width of the output signal is
Tout =
L d2k)21
T2+ ( - - (7.3.4)
2 dw 2 Wo
From this expression, one sees that (d 2w/ dk 2) I

Wo
measures the signal broaden-
ing acceleration due to the dispersion.
In engineering applications, one often uses the full width at a half maximum
height of the signal as the measurement scale of the signal width. We denote the
width at the half maximum of the input signal by r and that ofthe output signal
by r'. According to Mollenauer and Stolen (1982) [5], there is an empirical
formula for the broadening effect:
2
r, = r 1
+ (1.47LD>..2)
r2 (7.3.5)
where D is the dispersion parameter ofthe wave propagation medium, L is still

the distance between the input point and the output point, and>" is the wave
length of the carrier wave. The unit of LD is [ps/nm] (l[ps] = 1O-12[second] ,
l[nm] = 1O-9[meter], and IJl m = 1O-6[m]). The following example may give
you some feeling how a signal is distorted in a dispersive linear transmission line.
Suppose that LD ofthe line is LD = 450[ps/nm]. The width ofthe input signal
= =
is r 10 [ps]. The length of the carrier wave is >.. 1.5[Jlm] (about 200[GH z]).
The distance between the input point and the output point is L30[km]. Then
the width of the output signal is T' = 150 [ps]. So T'IT = 15 I! Such a
large distortion is intolerable in telecommunication. It is thus desirable to keep
the width of a signal unchanged in the process of signal transmission. The
self-focusing property in an optic fiber transmission line serves this purpose.
7.3.2 Optical self-focusing

When the diffraction index depends on the strength of the electric field, it is
possible for an optical fiber wave guide to have the function of self-focusing.
This function reduces the pulse broadening effect described in the last sec-
tion. Particularly, when the diffraction index relates the electric field in the
second order nonlinearity, the signal wave (i.e. the modulated wave) satisfies
the cubic nonlinear Schrodinger equation (NLS). The initial value problem for
this equation can be analytically found and it has soliton solutions. Hence
the dispersion is balanced by the nonlinearity and the wave broadening effect
is eliminated. Materials of the second order nonlinearity are considered most
often in theoretical research and industrial applications.
Consider the electromagnetic wave propagation in a polarizable medium.
We use Maxwell equations as our governing equations:
=
\7 x E 0 (or \7. D 471'p) = (Coulomb's law), (7.3.6)
\7 . B = 0 (no magnetic monopoles), (7.3.7)
8B
at + \7 x E = 0 (Farady's law), (7.3.8)
-8E
8t
+ -1 8P
-
fO 8t
+ 2
Co \7 x B =0 (Ampere's law) (7.3.9)
where E and B are electric and magnetic fields respectively, P is the polariza-
tion vector, fO is the dielectric constant, and Co is the speed oflight in vacuum.
Let
E = E(x, y, z, t)i (7.3.10)

(along the fiber's longitudinal direction),
B = B 1 (y,z,t)x+B 2 (x,z,t)y (7.3.11)
(in the fiber's transverse direction),
and
P = P(x, y, z, t)i. (7.3.12)
From (7.3.6) - (7.3.9) we can derive a governing equation for E, which is
E tt + -1 Ptt = Co2 \7 2 E (7.3.13)

fO
where \7 2 is the three dimensional Laplacian.

Assume
(7.3.14)
This is equivalent to assuming that the diffraction index is related to the electric
field by second order nonlinearity:
n = no + n2E2.
Then we have
(7.3.15)
=
where (3 l/c~ + al/{o, and'Y a2/{o, =
Consider the slow amplitude variation of the plane wave exp[i(kz -wt) +8].
We introduce the slow space and time scales as follows
x = {X, Y = {y, Z = tz, T = {t. (7.3.16)
It is assumed that E is of the following asymptotic expansion
E = (E(l)(X, Y, Z, T; z, t)
+ (2 E(2) (X, Y, Z, T; X, y, z, t)
+ (3 E(3) (X, Y, Z, T; X, y, z, t) + O({4). (7.3.17)
Substituting (7.3.17) into (7.3.15) and separating equations of successive orders,

we obtain that
O({): ( V2 - (3~)E(l)
8t 2 = 0' (73
..
18)
( V2 - (3~) E(2)
8t 2
( 82 82
-2 8x8X + 8y8Y
82( )
+2 - - - (3-- E(l)
. 2
(73 19)
8z8Z 8t8T' ..
( V2 - (3~)E(3)
8t 2
82 lj2 82 82 2
-2(8x8X + 8y8Y + 8z8Z -(38t8T)E()
+ 'Y 8t8 2 (E(1)3

2
82 8 82 82 1
- (8X2 + 8y2 + 8Z2 - (3 8T2) E( ). (7.3.20)
With the assumption (7.3.17), the linear equation (7.3.18) can have a plane
wave solution of the form
E (1) = -E(X, Y, Z, T) exp(zO)

.
+ C.C. (7.3.21)
where c.c stands for the complex conjugate, and
0= kz -wt +8 (phase of the fast plane wave), (7.3.22)

w2 = k 2 /{3, dw / dk = {3- t (group velocity). (7.3.23)
Here, E is the modulation (signal) of the faster wave (i.e. carrier wave). It is
our desire to find out which equation the modulated wave E satisfies and what
properties it has since E is the desired transmission signal.
Equation (7.3.19) can be written as
2 - -
(V2 - {J !2 )E(2) = -2{JiW(~; + {J-! ~;) exp(iO) + C.c. (7.3.24)
The right hand side of (7.3.24) is secular and hence it should vanish. The
general solution of the PDE
IS
E = E(e,X, Y), (7.3.25)
Then the 0(3) problem becomes
(V2 _{J !22 ) E(3)
a2E a2E a2E aE

= - [ax2 + ay2 + ae2 + 2ik ae] exp(iO)
+ 'Yw2[9F exp(3iO) - 3IEI2Eexp(iO)] + C.c.
Vanishing of the secular terms yields
2-
a E a E
2- 2-
a E 21-12- ' aE
-
aX2 + ay2 + ae2 + 3'Yw E E + 2tk ae = O. (7.3.26)
This is the nonlinear Schrodinger equation (NLS) we wanted to derive. It

describes the wave modulation (7.3.21) in the reference frame moving at the
group velocity along the z-axis.
The NLS equation can have soliton solutions. Each soliton represents a
filament diverging from the z-axis at a small angle. It is this filamental tendency
of the initial profile of the electric field that makes the self-focusing possible.
The focusing requires 'Y > O. We will see that this requirement is necessary by
studying solutions of the NLS equation in the next section.
An optic fiber can be very thin. The diameter of the core is in the order of
10 [pm]. The optical fibers are much thinner than copper lines. Yet, this optic
fiber can transmit signals of much higher intensity in the order of mega-watts
per square center, about 1 watt per line.
The nonlinear Schroodinger equation (7.3.26) can describe many physical
and chemical processes besides the self-focusing mechanism in an optical fiber
guide. For example, it can be used to describe the modulated motion of long
surface waves in deep water.
7.4 A Simple Solution of the NLS

Consider the following special type of nonlinear Schrodinger equation equation
(NLS)
iUt + u.,., + vluI 2 u = 0, -00 < x < +00, t > O. (7.4.1)
Since the nonlinearity is of order three, sometimes this equation is called the
cubic nonlinear Schrodinger equation. This is a complex equation of the un-
known U with independent variables x and t. The quantity v> 0 is related to
=
the material nonlinearity n2: n no + n21EI2 (the diffraction index).
The IVP for the above NLS can be analytically solved by an inverse scat-
tering method which requires a 2 X 2 matrix differential operator eigenvalue
problem and is not described here. Instead, we look for the solution of the
form
U = exp(i(kx - wt))v(e), e=x-ct (7.4.2)
which will yield a single-soliton solution. Substituting this into (7.4.1), we
obtain
V" + i(2k - c)v' + (w - k 2 )v + vlv 2 1v = O. (7.4.3)
This is a special case of (7.3.26) when E is independent of X and Y.
By choosing
k = cj2, w = c2 j4- a, a> 0, (7.4.4)
the v' term is eliminated and it results in
v" - av + vv 3 = O. (7.4.5)
The first integral of this equation is
(7.4.6)
where A is the integration constant. The general solution of (7.4.6) can be

expressed by elliptic functions.
A special case is the single-soliton solution: v(oo) = O. This results in the
integration constant A = O. We have
v= V~
-;- sechy'a(x - ct). (7.4.7)
Therefore a soliton modulated wave solution to the NLS equation (7.4.1) is
u(x, t) = f exp(i(kx - wt)) sechy'a(x - ct). (7.4.8)
When a = 0.5, v = 2.0, k = 5, w = 24.5, and c = 10, the function u(x, t) is

plotted in Fig. 7.11 with t = 0, 2 and 4 respectively.
More recent research in this area is on forced nonlinear Schrodinger equa-
tion. The forcing is due to the longitudinal inhomogeneity of the fiber or an
external disturbance due to an interference signal. Because of the forcing, the
7.5. Arctan Type of Solutions of the sG 213
usual traveling wave solution does not exist any more. Mathematically, one
says that the group symmetry is broken in this situation. The developed in-
verse scattering method and Backlund transformation are no longer applicable.
The research on finding the fiber materials, interference signals and solution
method for the forced NLS will have great impact on future communication
industries.
7.5 Arctan Type of Solutions of the sG

In section 7.2, we investigated various solutions of the type
~(x,t) = 4 arctan (:~:1) (7.5.1)
to the sine-Gordon equation
~xx - ~tt = sin~. (7.5.2)
Of course, one can also assume
~(x,t) = 4arctan[u(x)v(t)]
to get the same result.
From (7.5.1), one can derive
4vu' U"(U 2 + v 2) - 2U(U')2

~x = - u2 + v2' ~xx = 4v (u2 + v2)2 , (7.5.3)
4uv' v"(U 2 + v 2 ) - 2v(v')2

~t = - u2 + v2 ' ~xx = -4u (u2 + v2)2 (7.5.4)
Using equation (7.5.1) and
sin(40) 2 sin(20) cos(20)

4sinOcosO (cos 2 0 - sin 2 0)
sec u (+.
4tanO~ sec u -1)
4 tanO{l - tan 2 0)
{I + tan 2 0)2
one can derive
. [ (U)] v 2_u 2
sm 4 arctan ~ = 4uv {u2 + v 2)2. (7.5.5)
By equations (7.5.3), (7.5.4) and (7.5.5), the sG (7.5.2) can be written as
(7.5.6)
V(x,t=Ol
o t-----~fHi
-1 -20 o 20 40 60
V(x, t=2}
Or------------+------~~
-1 -20 o 20 40 60
v(x,t=4}
Or------------+--------------------~ >
-1 -20 0 20 40 60
Figure 7.11: The single-soliton solution of the nonlinear Schrodinger equation
with a = 0.5, v = 2, k = 5, w = 24.5, c = 10
Taking derivative of (7.5.6) with respect to :z:, we have
2uu' VII) + (u 2 + v 2) (u-;-lI )' -

UII + -;
(-;- 4u'u" = -2uu'. (7.5.7)
Taking derivative of (7.5.6) with respect to t, we have
ull + -;
2vv' ( -;- VII) + (u 2 + v 2) (VII)'
-; - 4v'v" = 2vv'. (7.5.8)
The operation (7.5.7)/((u 2 + v 2)uu') + (7.5.8)/((u 2 + v 2)vv') yields
-1 ( ~ ~ (t) = O.
")' (:z:) + - 1 (")' (7.5.9)
uu' u vv' V
This is the same as equation (7.2.3) in section 7.2. Hence there is a constant,
say -4k2, such that
-1 ( ~
")' (:z:) = -4k2,
uu' u
_1
vv'
(VII)' (t) = 4k 2.
V
Although k could be complex valued to keep the separation of the variables in

a general form, here we consider only real valued k. The first integral of the
above two equations results in
We rewrite the above two equations in the following form:
u" = _2k 2 u 3 + CIU, (7.5.10)

v" = 2k 2v 3 + d 1 v. (7.5.11)
The usual energy integral of (7.5.10), i.e. J(7.5.10) u'(:z:) d:z:, yields
(7.5.12)
Similarly, the energy integral of (7.5.11) yields
(7.5.13)
Since we once took derivatives of equation (7.5.6), we might have some extra
integration constants in equations (7.5.12) and (7.5.13). In other words, there
may exist some relationships among Cl, C2, d 1 and d 2. To find these relation-
ships, we substitute (7.5.10), (7.5.11), (7.5.12) and (7.5.13) back into equation
(7.5.6). It follows that
(d 1 - Cl + l)u 2 + (Cl - d1 - l)v 2 + C2 + d 2 = o.

Thus, we obtain the following relationships:
d2 = -C2 (7.5.14)
Let Cl = m 2 and C2 = n 2 Then equations (7.5.12) and (7.5.13) become

(U')2 = _k 2u 4 + m 2u 2 + n 2, (7.5.15)
(v')2 = k 2v 4 + (m 2 _ l)v 2 _ n 2. (7.5.16)
These are the same as equations (7.2.6) and (7.2.7).

We need to point out that in general m and n are complex valued in order
to keep Cl and C2 either positive or negative. Yet, we consider only the case
Cl ~ 0 and C2 ~ O. Hence, m and n are real valued constants.
[1] G. Rebbi and G. Soliani (1984), Solitons and Particles, World Scientific
Publishing, Singapore.
[2] R. K. Dodd, J. C. Eilbeck, J. O. Gibbon and H. C. Morris (1982), Solitons
and Nonlinear Wave Equations, Chapter 8, Academic Press, New York.
[3] A. Bishop and T. Schneider (1978), Solitons and Condensed Matter Physics,
Springer-Verlag, New York.
[4] A. Barone, F. Esposito, C. J. Magee and A. C. Scott (1971), Theory and
applications of the sine-Gordon equation, Rivista Del Nuovo Cimento 1,
227-267.
[5] L. F. Mollenauer and R.H. Stolen (1982), Solitons in optical fibers, Fiberop-
tic Technology, April, 1982, 193-198.
[6] G. L. Lamb, Jr. (1980), Elements of Soliton Theory, John Wiley, New
York, Chapter 5.
York, Chapter 17.
[8] H. Hasimoto and H. Ono (1972), Nonlinear modulation of gravity waves,
J. Phys. Soc. Japan 33, 805-811.
[9] D. J. Kaup and P. J. Hansen (1986), The forced nonlinear Schrodinger
equation, Physica D 18, 77-84.
[10] S.S. Shen (1990), Blocking of solitary pulses in a nonlinear fiber, Wave
Motion 12, 551-557.
[11] A. Barone (1974), Josephson Effect: Achievements and Trends, World
Scientific, Singapore.
[12] D. Saint-James, E.J. Thomas and G. Sarma (1969), Type II Supercon-
ductivity, Pergamon Press, Toronto.
[13] P.G. Drazin and R.S. Johnson (1989), Solitons: an Introduction, Cam-
bridge University Press, New York.
[14] J.C. Gallop (1991), SQUIDS, the Josephson Effects and Superconducting
Electronics, Adam Hilger, New York.
[15] R. D. Parmentier (1978), Fluxions in long Josephson junction, in Solitons
in Action (ed. K. Lonngren and A. Scott), pp. 173 - 199.
Chapter 8
Selected Examples of Flow

Instabilities
Although certain fluid flow configurations perfectly satisfy all the necessary
conservation laws (the conservation of mass and of momentum) and boundary
conditions, they cannot be seen in nature and cannot be used as engineering
designs. The reason is that these flows exist only when the boundary conditions,
external forces, internal structure of the fluids and initial conditions are in
accordance with the mathematical formulation perfectly. Yet, it is our common
sense that none of these conditions can be perfect in nature or in engineering
practice. But, on the other hand, these flow configurations can be so sensitive
to the imperfection of these conditions that the flows spontaneously change
their configurations and become different types of flows. Such a sharp and
quick response of a fluid flow system to a small external disturbance is known
as the flow instability. At the end of Chapter 6, we have seen that the lower
219
220 Chapter 8. Selected Examples of Flow Instabilities
solitary wave on a bump is unstable. Any small external disturbance, such as

the small numerical noise, can result in a sharp and quick change of the solitary
wave configuration.
In contrast to these unstable flow configurations, there are some flow config-
urations which are robust to a small, sometimes even a quite large, disturbance.
We say that these flows are stable. The upper solitary wave over a bump pre-
sented at the end of Chapter 6 is an example of the stable flows.
The fluid flow instability is a very complicated subject and a very impor-
tant area of scientific research. In this book, we plan only to discuss some basic
concepts of instability and present a few examples of stable and unstable flows.
There are some specialized books that treat the subject of hydrodynamic sta-
bility exclusively. Interested readers are referred to the book by Drazin and
Reid (1981) and those by Lin (1955) and Chandrasekhar (1981). The most
important pieces of classical literature are cited in one of these two books.
8.1 Concept of Stability

It is known to every body that to erect an egg on the brim of a coffee mug
is impossible. Scientifically, we say that the position of the egg relative to
the coffee mug is unstable. Yet, in principle the egg can have its equilibrium
under the action of the gravitational force and the reaction force from the brim
of the mug. This equilibrium, nevertheless, is very sensitive to the external
disturbances which can cause the immediate fall of the egg. Moreover, almost
all the unsuccessful erections of an egg on the brim of a coffee mug are due to
the fact that people, possibly except a few magicians, cannot put the center of
gravity of the egg exactly on the brim. Because the egg's position relative to
the coffee mug is unstable, this tiny deviation of the position of the center of
gravity results in the immediate fall of the egg.
In contrast to the example above, if the position of a rigid body relative
to its surroundings is stable, then a small, or even a fairly large, deviation
of the center of gravity from its equilibrium will not cause a collapse of the
configuration. For example, the normal positions of our study desks are stable.
Let us consider the stability of a wedge as a rigid body as shown in Fig.
8.1. Among the five positions of the wedge, position I is unstable and all
the other four positions are stable. Positions II and III are stable relative to
small disturbances, but they can change to position IV or V when an external
disturbance is large enough. We say that the positions II and III are nonlinearly
unstable. The nonlinear stability theory deals with relatively large disturbances
while the linear stability theory deals with, in principle, infinitesimally small
disturbances. Despite that the disturbance is infinitesimally small, we can
not ignore it. It is very often that this small disturbance causes the sensitive
unstable system to collapse. The prediction problem of the linear stability of
mechanical systems is complicated enough, needless to mention the nonlinear
stability. Fortunately, in most cases, linearly stable systems are stable enough
to be used in engineering designs.
B.1. Concept of Stability 221
D,--_ _ _---.C
E
I
f
Af--'------( f
f
f
f
f
f
fA
}-:--------
D c
Position I Position II
~------~~E
B A
Position III Position IV
~ _ _ _ _ _ _--=:::M E
Position V
Figure 8.1: Stable and unstable positions of a rigid wedge.

There is a Chinese proverb that says: "Water flows down and people move
up". Unstable rigid body positions or flow configurations imply relatively high
potential energy of the systems. These unstable systems would tend to gain
their status of lower potential energy. It is quite clear from Fig. 8.1 that the
position I of the wedge has the highest potential energy and position V has
the lowest potential energy. The unstable position I does not, however, have
to switch to the most stable position V. It can switch to any of the four stable
positions after its fall.
In this aspect, fluid flows are the same. A linearly unstable flow configura-
tion may switch to a stable state which may be nonlinearly unstable. Further
relatively large disturbances introduced to the system may result in further
switching of the flow configuration to one that has even lower potential en-
ergy or smaller length and time scales. However, the mechanism of the state-
switching of fluid flows is generally much more complicated than that of rigid
body motion. In fluid flows, it usually involves the disintegration of the length
scales and the time scales: a large scale disintegrates into a series of smaller
scales. The potential energy associated with the large scale motion or the input
of the external energy through the disturbance is transformed to the kinetic
energy, in one part, of the smaller scale motion and is consumed by viscosity
in another part. The breaking of water wave on a beach, the transition of a
laminar jet to a turbulent jet, the dissipation of a smoke ring and numerously
many other examples all show the scale disintegration and energy dissipation.
Fig. 8.2 shows the instability and the state transition of Couette flows. This
Couette flow is a viscous fluid flow between two concentric rotating cylinders.
Let me first describe the state bifurcation for the case of fixing the outer cylinder
and rotating the inner cylinder. The difference in the diameters of the cylinders
are small compared with the diameters. The space between the two cylinders is
sealed and filled with the viscous fluid. The trivial, but sensible, configuration
of the fluid flow is that the fluid is dragged by the rotating cylinder without
any longitudinal and latitudinal patterns. However, this is not always the case.
When the rpm of the inner cylinder increases and reaches a certain critical
value, a longitudinal pattern appears (see Fig. 8.2 (a)). And a further increase
of the rpm of the inner cylinder to a new threshold results in another more
complicated pattern. Finally, the flow becomes turbulent when the rpm is large
enough (see Fig. 8.2 (d)). From the successive increases of the rpm and the
switching of one state to another, we see a common feature similar to that found
in other flow instability phenomena, i.e. the length scale and the time scale in
the fluid motion become smaller and smaller after each successive transition of
state. Mathematically speaking, each of this state transition is a bifurcation.
The first time I observed this type of state transition was in a biochemistry
laboratory, it happened in mixture bottles which are devices commonly found
in chemistry and other bioscience laboratories. The viscous fluid in a glass
bottle was in motion caused by the rotation of a magnetized stirring bar at
the bottom of the bottle. The rotation of the stirring bar was driven by a
rotating magnetic field generated by another device (called stirerfhotplate)
on which the mixture bottle sits. The rotational speed of the bar could be
8.1. Concept of Stability 223
controlled. I gradually increased the rotational speed and observed a series of

state transitions (bifurcations).
The laminar flow configuration is usually man made and distinguishes itself
from its surroundings. The nature of the fluid motion tends to homogenize
the distinguished features by disintegrating the large scales associated with the
laminar motion of the fluid. In a developed turbulent flow, there are scales
ranging from as small as the mean free path of the fluid molecules of the
fluid and as large as the scale of the entire flow configuration. This is in a
sharp contrast to the solitary wave case where there are only two length scales
involved: a long wave length scale and a depth scale. It is because there are
infinitely many length scales in a turbulent flow that turbulent flows are a
difficult subject to study. It is improper to study turbulent flows by using the
methods for laminar flows.
Viscosity plays a crucial role in the transition from a laminar flow (say,
Reynolds number is less than 1,000) to a turbulent flow (say, Reynolds number
is larger than 2,000). Viscosity is the most subtle effect in the stability and
instability exchange process. It is well known that a viscous fluid flow is more
stable. But in many initial states of the instability, it is exactly the viscosity
that leads to the development of vortices. Without viscosity, a laminar ideal
fluid flow cannot be transmitted to a turbulent flow. Fortunately, when one is
only interested in the criterion of the linear instability, there is no need to fully
understand such a subtle effect of viscosity in the post stability motion.
In every fluid motion, there is resistance: viscosity or inertia. These are the
stabilization effects. For an instability to occur, there must be some destabiliza-
tion effects. The destabilizing forces may be the gravitational buoyancy force
(heavier fluids on lighter fluids), thermal buoyancy force (fluids being heated
below), centrifugal force (fluids subject to a net angular momentum), and a
few others. In this book, we will exclusively analyze the simplest examples of
the instabilities driven by the gravitational buoyancy force, thermal buoyancy
force and centrifugal force.
Although there are many methods in the linear stability theory, the under-
lying principles are the same. One basically assumes that after an infinitesi-
mal disturbance the motion of the system under consideration obeys linearized
laws derived from the original nonlinear equations and the sensitive response
of the system to the external disturbance obeys the exponential laws of growth
(unstable) or decay (stable). Consequently, the linear stability problem be-
comes basically a linear eigenvalue problem. The corresponding eigenvectors
(or eigenfunctions) usually constitute a basis of a Hilbert space. Each eigen-
function represents a normal mode. Hence, this eigenvalue approach is called
the normal mode method in stability theory.
224 Chapter B. Selected Examples of Flow Instabilities
Figure 8.2: Instability and state transition of the rotating Couette flow. Panels
a, b, c, and d correspond to increasing rpm of the inner cylinder. [From Coles
(1965), by permission of Cambridge University Press]
B.2. Kelvin-Helmholtz: Gravitational Instability 225
Fluid 2
Interface x
------... p
----.~
1
Fluid 1
Figure 8.3: Flow configuration of two fluids that demonstrates the Kelvin-
Helmholtz instability.
8.2 Kelvin-Helmholtz: Gravitational Instabil-

ity
Few unstable flows can be analyzed by simple analysis that can be taught in a
course for the first year graduate students. The Kelvin-Helmholtz instability,
the Benard problem and the Taylor's Couette flow instability are among the
few.
The Kelvin-Helmholtz instability describes the instability of the flow con-
figuration of one ideal fluid on the other with a horizontal flat interface (see
Fig. 8.3). Both fluids are infinitely deep and flow only in the horizontal di-
rection. The densities and the velocities of the two fluids are Pl, P2 and Ul, U2
respectively. The trivial solution of the Euler equations for the flow system is
u = Ul ,2, V = 0, P = Po - Pl,2 y, P = Pl,2 when y> 0 0). (8.2.1)
To check whether the flow configuration shown in Fig. 8.3 is stable or

not, we apply a small disturbance to the system. Of course, the response of the
system to the small disturbance must show on the interface. So the flat interface
now becomes a wavy one as shown in Fig. 8.4. The stability question is hence
whether the wave will fade and go back to the flat interface as it was (stable) or
it will grow indefinitely (unstable). We are only concerned with linear stability
(sometimes called linearized stability in mathematics) here. The result from the
linear stability analysis will yield the criterion of the instability. Immediately
after the appearance of the instability, a vortex sheet is developed. Further
development of the interface can result in various types of chaotic motion of
the interface zone. This post instability motion of the fluids is completely
nonlinear and thus the linearized stability analysis will not be able to describe
the evolution of the interface wave after the occurrence of the instability.
From physical intuition, it is obvious that when Ul = U2 = 0, the configu-
ration of the fluid system is stable (unstable) if P2 < Pl (Pl < P2, respectively).
Figure 8.4: The wavy response of the interface to a small external disturbance.
But, when the shear velocities UI and U2 do not vanish, the situation is not
obvious any more. These velocities sometimes suppress the instability and at
other times excites the instability.
According to Drazin and Reid (1981, p.14), Helmholtz noticed the instabil-
ity ofthe two superimposed fluid flows in 1868 and Kelvin solved this instability
problem in 1871. To analyze this instability, let us use the potential theory.
The goal is to use the linearized stability analysis to find out when the flow
system is stable and when it is unstable.
The potential of the trivial flow (also called the basic flow in the language
of bifurcation) is
<p(x, y, t) = UI ,2 x when y < 0 (y> 0).
After a small disturbance is applied to the system, the potential and the pres-
sure become
<PI = UIX + <p~,
<P2= U2x + <p~,
PI = Po - 9 PI Y + p~ ,
P2 =Po - gP2Y + p~.
The quantities with the superscript ", " are the flow field in response to the
external disturbance. The boundary conditions at y oo are <p~ = 0 and =
<pi = 0 since the small disturbance does not affect the uniform motion at
infinity in finite time.
The normal mode method requires we assume that the response to the
disturbance is in the form !(x, y) exp( -iwt). Here the frequency w is complex
valued: w = Wr + iWi.
Definition: If Wi > 0 0), the flow is unstable (stable).
The unique solution of the Laplace equation for <pi 2 with the boundary
conditions <p' (x, oo) = 0 is '
<P~,2= C I ,2 exp [ky + i(kx - wt)) + c.c. for y < 0 (y> 0), (8.2.2)
TJ = aexp [i(ka: - wt)] + c.c. (8.2.3)
B.2. Kelvin-Helmholtz: Gravitational Instability 227
or
CP~,2 = Cl,2 exp(ikz) exp( -iwt) + c.c.
where z = z - iy, k is called the wave number and real valued, and c.c. stands
for complex conjugate. The quotient w / k is called the phase velocity and is
denoted by c: the speed of the interface wave. Hence c Cr + iCi is complex=
valued and Ci < 0 (> 0) implies stability (instability).
Another assumption of the linearized stability analysis is that the response
of the system to the small disturbance satisfies linearized laws of the original
nonlinear equations. In the potential theory, the nonlinearity comes into the
system only through the free surface or interface boundary conditions. Here,
we employ the linearized interfacial boundary conditions
81] U 81] _ 8cp~
8t + 18z - 8y'
81] U 81] _ 8cp~
8t + 28z - 8y ,
, , 8 2 1]
P2 - Pi = T 8z 2 '
P~ 8cp~ 8cp~
- = - - - - Ul-- - 91],
Pi 8t 8z
P~ 8cp~ 8cp~
- = - - - - U2-- -91],
P2 8t 8z
where T is the surface tension. The first two equations are the kinematic
conditions: the two fluids do not penetrate into each other. The third one
is a dynamical condition from the force balance. The last two equations are
linearized Bernoulli equations which can also be considered as dynamical condi-
tions. One can eliminate P~ and P~ from the three dynamical conditions above.
Hence the last three equations become one.
The useful equations and the solution forms are summarized as below:
81] U 81] _ 8cp~
8t + 18z - 8y'
81] U 81] _ 8cp~
8t + 28z - 8y'
Pi [ at
8CP~ 8cp~ ]
+ Ul 8z + 91] - P2
[8cp~
at + U2 8cp~ ] _ 8 1]
2
8z + 91] - T 8z2 '
cp~ = C l exp[ky + ik(z - ct)] for y < 0,
cp~ = C2 exp[-ky + ik(z - ct)] for y> 0,
1] = a exp[ik(z - ct)].
Substitution of the last three expressions into the first three equations re-
sults in an eigenvalue problem of a 3 x 3 matrix:
Mb=O (8.2.4)
where
(8.2.5)
and
(8.2.6)
In this eigenvalue problem, b is the eigenvector and c is the eigenvalue. The

eigenvalue is determined by putting det(M) = O. This yields
c = Pi Ul + paUa (8.2.7)
Pi +Pa
where
(8.2.8)
When Ul = Ua = 0, we have c = co. Hence Co is the phase velocity of the

interfacial wave in the absence of currents.
Recall the definition of linear instability: Im(c) > 0 f--t unstable. As long
as there is one unstable mode, the whole system is unstable. Therefore, the
flow is linearly unstable if and only if
a
( Ul - Ua ) a
P1Pa +
Pi Pa
> Co . (8.2.9)
Otherwise, the flow is stable.

In the following, we discuss several special cases of this Kelvin-Helmholtz
instability.
Case I: Pa > Pi
In this case, the upper fluid is heavier than the bottom one. It is our com-
mon sense that the flow must be unstable. Indeed, this proves to be true when
examining the relationship between c~ and k according to equation (8.2.8). The
curve of the function is shown in Fig. 8.5. When k ::; k., we have c~ ::; O. So
the instability condition (8.2.9) is automatically satisfied and this instability
is caused by the modes in the disturbance whose wave numbers are smaller
or equal than k., i.e. low frequency oscillations. But if the disturbance has
a very large wave number (k is large, or the wave length A is small), whether
the flow is stable or unstable depends on the densities and the difference of the
flow velocities. The c~, k relationship and the instability criterion (8.2.9) imply
something one may not have understood from intuition, i.e. longer waves are
more unstable than shorter waves when the upper fluid is heavier. This is the
direct manifestation of the phenomenon that the collapse of a heavier fluid on
a lighter fluid is sudden and in a large scale.
8.2. Kelvin-Helmholtz: Gravitational Instability 229
-- ~
----
~
~~ ~ k
V'
/
k*
-1 I
-1
I
-2 ,
-2 ~.
Figure 8.5: The relationship between c~ and k when the upper fluid is heavier.
A random disturbance contains sinusoidal components of every wave num-

ber. Of course, it includes the sinusoidal components of large and small wave
numbers. Thus, when considering stability, we usually include disturbances of
all wave lengths (i.e. a white noise in space). A fluid flow is stable only when it
is stable under the disturbances of waves at every wave length. Hence, in this
sense, the configuration of a heavier fluid on a lighter fluid is always unstable.
Case II: Pl > P2

In this case, the upper fluid is lighter than the bottom one. According to
our common sense, this flow configuration should be stable. But, this is not
always true when the difference of the flow velocities does not vanish. The
shear due to the velocity difference can induce the instability of certain mode
(i.e. the wave of certain wave length).
According to equation (8.2.8), the relationship between c~ and k is shown
in Fig. 8.6. In this figure,
The most unstable mode is neither the very long wave nor the very short wave,
but the wave with wave number k = kcr The instability is caused by the
shear of the current and the surface tension. The interaction of the large shear
velocity with the surface tension yields a sheet of vortices. As the shear flow
continues to transfer kinetic energy to the vortices, the vortex sheet develops
further and finally induces a zone of turbulence between the two fluids. Fig. 8.7
shows this type of instability exited at the most unstable mode (k = kcr ). The
upper stream of water is moving to the right faster than the lower one, which
1~+--------+---------+---------r--------~~-------+-
1~4--------+--------~--------~--~~~~---------+-
1~~-------+---------+------~~~------~---------+-
k
10
Figure 8.6: The relationship between c6 and k when the upper fluid is lighter.
contains dye that fluoresces under illumination by a vertical sheet of laser light.
The faster stream is perturbed sinusoidally at the most unstable wave number
k = kcr in the first panel, and at half that wave number k = (1/2)kcr in the
second panel.
The very long wave whose wave length A = k- 1 is near infinity is stable
when the upper fluid is lighter. This agrees with our common sense.
The very short wave whose wave length A = k- 1 is near zero is also stable.
Such a short wave disturbance interacts with the surface tension and creates a
very thin and stable vortex sheet.
Case III: U1 = U2
In this case, there is no shear between these two fluids. The flow is unstable
if and only if the upper fluid is heavier than the bottom one (i.e. P2 > pd.
Hence, the shear enhances the instability of the flow system.
8.3 Benard Problem: Thermal Instability

Benard (1900) observed the instability of a viscous fluid driven by a thermal
gradient. He experimented on a very thin layer of oil of depth about 1 mil-
limeter on a plate. The top surface of the fluid was free. When the plate was
uniformly heated and reached a certain temperature, he observed a pattern
that consisted of regular hexagonal cells. This is a new state. The transition
from the original uniform, trivial basic state to this new state is due to the
development of convection. Rayleigh (1916) employed the linear stability anal-
ysis and found the criterion for this thermal instability to happen: the ratio
of the thermal buoyancy force to the viscous stablization force needed to be
B.3. Benard Problem: Thermal Instability 231
Figure 8.7: The Kelvin-Helmholtz instability at the most unstable mode when
the upper fluid is heavier. [From Van Dyke (1982), by permission of The
Parabolic Press]
sufficiently large. This ratio is commonly referred to as the Rayleigh number:

Ra.
When the fluid layer is thicker, one may see other patterns rather than the
hexagonal cells as the first post instability state. One can easily experiment
this with his usual flat bottom cooking pan. Pour a thin layer of oil in the
pan and put the pan on a hot stove. One can observe the development of the
cells. If the temperature gradient is maintained at a constant value (this is very
difficult to do with a household cooking stove), then the cells do not change.
Since it is almost impossible to maintain such a constant temperature gradient,
what one actually sees is a series of states: larger cells turn into smaller ones
due to the instability.
Although our experiment with a cooking pan on a household stove yields
different stability patterns, we still call this type of stability problem the Benard
problem and the corresponding instability the Rayleigh instability. Actually, in
modern days, every such type of thermally driven instability problem of a fluid
layer is generally referred to as the Benard problem.
We plan to analyze a Benard problem for a layer of fluid sandwiched between
two horizontal plates of distance d apart. The bottom plate has a higher
temperature and the temperature difference is e. See Fig. 8.8 for the fluid
configuration and the coordinate system for the analysis below.
The surface tension plays a very important role in Rayleigh instability when
there is a free surface. One may still think that this thermal instability is
actually the Kelvin-Helmholtz instability since the heated fluid at the bottom
becomes lighter. Indeed, this is part of the reason. Particularly, when the
fluid layer is thick. The thermal gradient induces the density gradient which
in turn induces a convective buoyancy force. But, when the fluid layer is very
d ,.
X or y
Figure 8.8: The fluid configuration of the Benard problem between two plates.
[From Kundu (1990), by permission of Academic Press]
thin, such as the original case considered by Benard, the Kelvin-Helmholtz

instability is not the driving mechanism. Instead, the temperature gradient
induces inhomogeneities of the surface tension. The inhomogeneous surface
tension then forces the fluid to convect and the cells to be formed.
Let us make a dimensional analysis and find out the control parameter:
the Rayleigh number Ra. The relevant physical quantities are: the vertical
temperature gradient r [I<L-l], the thermal expansion coefficient a [I<-l],
the depth of the fluid layer d [L], the kinematic viscosity v [L2T- 1 ], and the
gravitational acceleration g [LT-2]. Here, what is inside [ ] is the dimension of
the quantity ahead of [ ]. The buoyancy force per unit mass in an infinitesimal
fluid element due to the thermal expansion is
[Buoyancy force]j[unit mass] '" gaT'
where T' '" rd = e is the perturbed temperature from the basic uniform state.
This buoyancy force due to the thermal expansion is the driving force of the
instability. What battles with this destabilizing force is the stabilizing force
which is the viscous force. Numerous experiments have shown that the viscous
force acting on a fluid or a solid particle is proportional to the particle's speed
when its velocity is small but proportional to the square of the speed when
the velocity is large. In our case, the convection velocity of the fluid is small.
Hence, the viscous force per unit mass is
[Viscous force]j[unit mass] '" d- 2 vw
where w is the convection velocity of a fluid element at the perturbed state

and d- 2 is used to balance the dimension. The thermal diffusivity (not the
molecular diffusivity) is mainly the eddy diffusivity caused by the convection:
'" '" wd [L2T-l]. The instability criterion is based upon the ratio of the
destabilizing force (the Buoyancy force) to the stabilizing force (the viscous
force). This ratio is
Buoyancy force gaT' gard4

Viscous force '" vd- 2 w '" --;;;;:-
B.3. Benard Problem: Thermal Instability 233
This is the Rayleigh number Ra:
Ra = gafd4 .
(8.3.1)
11K.
For the fluid configuration shown in Fig. 8.8, the critical Rayleigh number
is Ra cr = 1708. When Ra > 1708, the fluid becomes unstable and quickly
transforms to the convective state. Next, let us work out the mathematical
analysis of the linearized stability theory and determine this critical Rayleigh
number.
The gravitational force per unit mass is neglected and the vertical buoyancy
force is written as
fb = g[1 - a(T* - To)]
where To is the surrounding temperature. The governing equations are
p* = Po[1 - a(T* - To)], (8.3.2)

u~ .= 0 (8.3.3)
J,J '
u*t, + u~u~
J I,J
= -~p*.
0
Po ,J - g[1 - a(T* - To)]Jo" 3 + vD.u~ , (8.3.4)
~; + Uj~j = K.D.T*. (8.3.5)
Here, Xl = =
X, X2 Y and X3 =
Z are the spatial coordinates, Ul U, U2 =
v =
and U3 = ware the velocity components, Ji3 is the Kronecker delta (equals zero
when if:. 3 and one when i = 3), the repeated indices mean the summation for
j from 1 to 3, and D. is the three dimensional Laplacian.
The basic state has zero velocity. The temperature (1') and the pressure
(P) satisfy the following equations:
1 -
--P,j - g[1- a(T - TO)]J;3 0, (8.3.6)
Po
o. (8.3.7)
= =
Applying the boundary conditions: 1'(z -d/2) To and 1'(d/2) To - G, =
we can solve equation (8.3.7) to obtain a linear temperature distribution:
(8.3.8)
where G/d is the temperature gradient f.

Let the perturbed state be: Ui(X, y, z, t), T(x, y, z, t) and p(x, y, z, t). Set
= =
ut Ui, T* 1'(z) + T and p* =
P(z) + p and make use of equations (8.3.6)
- (8.3.7). After neglecting the nonlinear terms, we can derive from equations
(8.3.4) and (8.3.5) that
1
Ui,t = --P,; + gaTJi3 + vD.u;, (8.3.9)
Po
T,t - fw = K.G. (8.3.10)
Equation (8.3.10) has only two unknowns: T and w. We would like to eliminate
p, U and v from the three equations (8.3.9) (when i = 1,2,3) and the continuity
equation (8.3.3) to get an equation with only T and w as unknowns. The
reason for this manipulation is that since T and ware the main causes of the
instability, we would like to work with them. The third equation in (8.3.9)
(when i = 3) does not include u and v. We start from this equation and try
to get rid of p. Taking the Laplacian for the i = 3 component for both sides of
equation (8.3.9), we get
(8.3.11)
To make use of the continuity equation ui,i = 0, we take the divergence of the
vector (8.3.9):
1
--D.p + goT z = o.
Po '
Taking derivative of this equation with respect to z, we get
1
--D.(p z)
po'
+ goT,zz = o.
With this, equation (8.3.11) becomes
D.w,t = goD.HT + vD. 2 w (8.3.12)
where D.H is the horizontal two-dimensional Laplacian.

Now we have two equations (8.3.10) and (8.3.12) for two unknowns T and
w. We nondimensionalize the independent variables by taking
t ~ (d/w)t, (x, y, z) ~ (x d, y d, z d)
and convert the dimension of w to the dimension of T:

If,
w ~ fd2W
where til has the temperature dimension [J{j. Now, a comparison of the size of
variables becomes possible since equations (8.3.10) and (8.3.12) are linear and
homogeneous. We rewrite these two equations in terms of the newly defined
variables as
(8t - D.)T = til, (8.3.13)

(Pr- 18t - D.)D.tiI = Ra D.HT (8.3.14)
where Pr = v / If, is called the Prandtl number.

Now, the normal mode can be introduced as
til = W(z) exp(ikx + ily + ut),

T = S(z) exp(ikx + ily + ut).
8.3. Benard Problem: Thermal Instability 235
Substituting these two expressions into equations (8.3.13) and (8.3.14), we get
[0" - (D2 - K2)] S(z) = W(z), (8.3.15)

[;r - (D2 - K2)] (D2 - K2)W(z) = -Ra K 2S(z) (8.3.16)
where D = d/dz and K = ../k 2 + [2. The boundary conditions are

1
W=DW=S=O when z= "2. (8.3.17)
Equations (8.3.15), (8.3.16) and (8.3.17) constitute an eigenvalue problem. It

can be proved that the operator of this eigenvalue problem is self-adjoint and
hence its eigenvalues are all real valued. The fluid configuration is linearly
unstable when 0" > o. Next, let us find out when 0" is positive by solving the
eigenvalue problem (8.3.15), (8.3.16) and (8.3.17).
It is clear from (8.3.15), (8.3.16) and (8.3.17) that 0" is a function of Ra,K
and Pr. It appears that it is not easy to determine this function in general.
Yet, to determine this function is not crucial. The important information for
= =
us is when 0" 0 (marginally stable) for Ra Racr So the fluid configuration
is stable (unstable) when Ra < Racr (Ra > Racr , respectively). Therefore,
we set 0" = o. Then, equations (8.3.15) and (8.3.16) reduce to
(D2 - K2)S(z) = -W(z),

(D2 _ K2)2W(z) = RaK 2S(z).
Applying (D2 - K2) to the second equation, we can eliminate S from the above
two equations to obtain a sixth order equation for W
(8.3.18)
with the corresponding boundary conditions
(8.3.19)
For a given wave number K, we may view the above problem (8.3.18) - (8.3.19)
as a new eigenvalue problem with Ra being the eigenvalue and W(z) being
the eigenfunction. This self-adjoint eigenvalue problem has only a discrete
spectrum: Ral(K), Ra2(K),, 00. Then, Racr = minRal(K) is the smallest
eigenvalue that corresponds to an even mode: a mode without any node point.
The second mode is odd and has exactly one node point. See Fig. 8.9 for the
illustration of these two modes.
Now, let us find this even mode. Setting W = exp(qz) and substituting it
into equation (8.3.18), we get the characteristic equation for (8.3.18)
Gravest even mode

8 Gravest odd mode
W(z}
Figure 8.9: The first two modes of the convective flows after the appearance
of the instability: (a) the first mode (even), (b) the second mode (odd). [From
Kundu (1990), by permission of Academic Press]
The three roots for q2 are
q2 = _f{2 [(~~r/3 -1], (8.3.20)
q2 = f{2 [1 + ~ (~~ ) 1/3 (1 iV3)]. (8.3.21)
Then the six roots are:
iqo, q, and q*
where
1/3 ]1/2
qo = f{ [ (:~) -1
and q and q* are the complex conjugates of the square root of (8.3.21). If we
consider the smallest eigenvalue Ra1 for an even mode, the general solution of
(8.3.18) can be written as
W = A cos(qoz) + B cosh (qz) + C cosh(q* z) + c.c.

where c.c. stands for complex conjugate. Substituting this into the boundary
conditions (8.3.19), one gets a 3 x 3 matrix eigenvalue problem whose eigen-
vector is (A, B, C) (complex valued). This coefficient matrix Mis:
cosh(~)
qsinh(V
(q2 _ f{2)2 cosh(q/2)
To find the eigenvalue Ra1, we set
det(M) = o. (8.3.22)
8.4. Taylor Problem: Centrifugal Instability 237
Ra
unstable
----~--~
1708
stable
Kcr=3.12
K
Figure 8.10: The function Ra1(K).
For every given K, one can solve equation (8.3.22) numerically to get Ra1.
Thus, the numerical solutions determine the functional dependence of Ra1 on
K: Ra1(K) which is shown in Fig. 8.10. The minimum of this function is
approximately Racr = 1708 when K = Kcr = 3.12. This result is similar to
the Kelvin-Helmholtz instability when the top fluid is lighter: there is a most
unstable mode kcr Since a usual nonmonochromatic perturbation is composed
of harmonics of all wave numbers, the fluid configuration is hence unstable
when the Rayleigh number is larger than 1708. This result agrees with the
laboratory experiments very well and is considered one of the major successes
of the linear stability theory.
8.4 Taylor Problem: Centrifugal Instability

The Taylor problem is the instability problem of a viscous fluid flow between
two concentric rotating cylinders as shown in Figs. 8.2 and 8.11. The radii of
the inner and outer cylinders are R1 and R2 respectively, and the corresponding
angular velocities are 0 1 and O2 ,
The driving force of the instability is the centrifugal force. When the fluid
elements near the inner cylinder have a larger angular momentum than the fluid
elements near the outer cylinder, the former is subject to a larger centrifugal
force and this force tends to pull the inner fluid outward. This difference in
the angular momenta is caused by the difference between the angular velocities
of the cylinders. If there were no viscosity, then the fluid flow would become
unstable as soon as such a momentum difference appears. This inviscid in-
stability criterion was obtained by Rayleigh (1888) and is very similar to the
c: T
o II
0 1
>.cr
~
......y '-.Y
0, 01
Figure 8.11: The Couette flow between two rotating cylinders. [From Kundu
(1990), by permission of Academic Press]
Kelvin-Helmholtz instability criterion: the top fluid is heavier than the bottom
one in the absence of shear currents, where the driving force of the instability
is the gravitational buoyancy force.
When the fluid is viscous in the Couette flow, the centrifugal force that tends
to cause the instability is opposed by the viscous resistance force (the stablizing
force). This is very similar to the Benard problem, where the thermal buoyancy
force has to battle with the viscous resistance to trigger an instability. The
ratio of the thermal buoyancy force to the viscous resistance force (measured
by the Rayleigh number Ra) has to be sufficiently large for an instability to
occur. Similarly, for the Couette flow problem, we may require the ratio of the
resultant centrifugal force on fluid elements to the viscous resistance force to be
large enough for an instability to happen. And indeed, this is true. The relevant
mathematical analysis and laboratory experiments were first conducted by G.
I. Taylor (1923). The aforementioned ratio is called the Taylor number Ta:
Centrifugal force per unit mass
Ta
Viscous force per unit mass
n (R - R ) n,R~-n2R~
1 2 1 R 2 -Rr
where 11 is the kinematic viscosity.

Let us now do the mathematical analysis for a special case of this problem
when the gap between the two cylinders is very small:
The governing equations for this problem are the Navier-Stokes equations
in cylindrical coordinates:
where
D a a a
Dt = at + r ar + z az '
U U
and A is the Laplacian in the cylindrical coordinates.

The basic flow is an axisymmetric laminar flow: velocity = (0, V(r), 0) and
pressure = P(r). One can easily find that
B
V=Ar+- (8.4.1)
r
where
and (8.4.2)
The pressure P(r) ofthe basic state will not appear in the perturbed equations,
so we do not need to compute it.
We denote the perturbed field also by (u r , U(I, uz ) and p without any ambi-
guity:
Ur -+ U r ,
U(I -+ V(r) + U(I,
p -+ P(r) + p.
Again, we assume that the perturbed flow is axisymmetric. Experiments have

shown that this assumption is a very good approximation for the first unstable
state (or, as it is called mathematically the first bifurcation state). This flow is
shown in the first panel of Fig. 8.2. We substitute the above expressions into
the Navier-Stokes equations and omit the nonlinear terms. The result is
OUr Ur OUz 0
or + -;:- + oz = ,
OUr _ 2V Ue = _! op + V (Llu r _ Ur ) ,
ot r p or r2
oue + (dV + V) Ur = v
ot dr r
(LlUe _ ue)
r2
,
oU z lop
- = ---+vLlu z .
ot poz
Now the normal mode assumption comes into the play:
(u r , Ue, uz,p) = (u, v, tV,p) exp(ikz + wt) + c.c. (8.4.3)
Substituting this expression into the above linearized perturbed equations and
eliminatingp and tV from the first, the second and the fourth equations, together
with the third equation we get two equations for u and v:
w)
v ( DD. - k 2 - -;; (DD. - k 2"
k2 )u = 2-;:-v,
V" (8.4.4)
v (DD. - k 2 - ~) v = D. V u (8.4.5)
= =
where D d/dr, D. d/dr+ l/r and Ll-1/r2 DD. + = 0;.
Next we apply the small gap assumption and nondimensionalize certain
variables.
1= kd (dimensionless wave number),

(T = wd2 /v
(dimensionless frequency),
( = (r - Rd/d E [0,1] (dimensionless radial coordinate),
1
D. '" D =
'd(d/d() (d/d( is still denoted by D below),
V '" Od1 - (1 - m)(]

r
where d =R2 - Rl and m =

O2 /0 1 . When ( =
0,1 for the inner and outer
=
boundaries, the tangential velocities are V 01R 1 and V 02R2 respectively. =
The velocity in between satisfies a parabolic distribution with respect to r.
Substituting these approximations into equations (8.4.4) and (8.4.5), we get
(8.4.6)
(8.4.7)
Let
Equations (8.4.6) and (8.4.7) can be written in a nicer form:
(D2 _12 - U)(D2 _/2)U = (1 + O'()V, (8.4.8)

(D2 _/2 _ U)V = -Ta 12 U (8.4.9)
where 0' = m - 1,
Ta = _4Afhcr (8.4.10)
1I 2
is the Taylor number and A = (!hR~ - fhRn/(m - Rn

is the same as that
given by (8.4.2). The boundary conditions for the above equations are
U = DU = v= 0 when (= 0, 1. (8.4.11)
Equations (8.4.8), (8.4.9) and (8.4.11) constitute an eigenvalue problem with

u(Ta, I) being the eigenvalue. This problem is self-adjoint and hence it only
has real discrete spectrum. The linear stability criterion requires u < 0 for
stable flows.
For marginal stability, similar to what we did for the Benard problem in
the previous section, we set u = o. From (8.4.8), (8.4.9) and (8.4.11), we get a
new eigenvalue problem of the sixth-order equation
(D2 _/ 2)3v = _/2 Ta (1 + O'()v, (8.4.12)

v = (D2 _/ 2)v = D(D2 -12)v = 0 for (= 0,1. (8.4.13)
Here, Ta may be viewed as the eigenvalue which is a function of the wave

number I: Tal(I), Ta2(/),, 00. The minimum of the function Tal(l) is the
critical Taylor number Ta cr . The corresponding wave number Icr marks the
most unstable mode. When Ta > Ta cr (Ta < Ta cr ), the Taylor Couette flow
is unstable (stable). The eigenvalue problem here looks similar to that of the
Benard problem (8.3.18), but here we have a variable coefficient ( on the right
hand side of the equation (8.4.12). This makes the problem much more difficult.
It appears that there are no analytic solutions to this eigenvalue problem.
Taylor did numerical calculations and experiments for this problem, the results
of which agreed quite well.
Before 1920s, there were no sophisticated measuring instruments such as
today's laser devices. Accurate measurements of fluid flows were impossible.
Yet, G. I. Taylor chose to study this Couette flow that requires no direct mea-
surements of fluid flow field. The only control data needed is the rotational
speed of cylinder. Even at that time it was not that difficult to have a motor
with a fairly accurate rpm. Thus, it has been said in the literature that Taylor
provided the first evidence showing the validity of the N avier-Stokes equations.
Chandrasekhar (1961) considered many different types of approximations
and produced agreeable results with experiments too. We only list a small
portion of the data obtained from one of his approximations:
(m,lcr,Ta cr ) = (1,3.12,1708); (0,3.12,3390.3); (-1,4.0,18677)

Figure 8.12: Stokes wave in deep water.
where m = O2 /0 1 is the ratio of the rotational speed of the outer cylinder O2

to that of the inner cylinder 0 1 as defined earlier. One may have noticed that
1708 is also the value we obtained for the critical Rayleigh number for a fluid
flow between two parallel plates. The mathematical reason for this agreement
is the similarity between the two eigenvalue problems.
With the known values of the critical Taylor number, one may select the
viscosity of the fluid, the rotation speeds of the two cylinders and the radii of
the two cylinders in such a way that the fluid flow is either stable or unstable.
These types of experiment and theory have very important applications in the
machine lubrication designs for rotating parts.
8.5 Benjamin-Feir: Side-Band Instability

What we want to show in this section is that Stokes waves are unstable. Stokes
waves are the periodic finite-amplitude irrotational waves in deep water. The
possibility of the existence of this kind of wave was first pointed out by Stokes
(1847), about the same time when John Scott Russell (1844) reported his ob-
servation of the beautiful solitary wave in a canal. Stokes showed that the
surface elevation of the irrotational waves in deep water was given by
1
1/ = acos(kz-wt)+2"ka 2 cos2(kx-wt)
3 2 3
+ Sk a cos3(kx -wt) + ... (8.5.1)
He also found that the phase velocity depends on the wave amplitude (not the
case for small amplitude linear waves):
(8.5.2)
Expression (8.5.1) is the Fourier cosine expansion of the free surface profile.
As pointed out in H. Lamb's book (p.420), a question as to the convergence of
the cosine series was raised by Burnside (1916) and he even doubted whether
this kind of periodic wave would exist. This led Rayleigh to undertake an
extensive investigation (1917). Finally, the existence theorem for Stokes waves
was rigorously established by Levi Civita (1925).
8.5. Benjamin-Feir: Side-Band Instability 243
The summation of the cosine series in (8.5.1) yields a periodic wave of

permanent form traveling at the phase velocity determined by (8.5.2). The
profile of 77(X, t) obtained from the summation has a flattened trough and a
peaked crest (Fig. 8.12). The maximum possible amplitude is ama~ = 0.07/k,
at which point the crest becomes a sharp 120 0 angle. The maximum wave
height with a 120 0 angle was found by both Stokes (1847) and Michell (1893)
using different methods.
In nature, we hardly see any wave of permanent form with a large ampli-
tude. A naive person may conclude that this type of periodic wave is unstable.
Ironically, perhaps due to Russell's discovery of the stable solitary wave, peo-
ple at that time never thought that Stokes waves were unstable. If there were
no Russell's discovery, probably people would never have thought that a large
amplitude wave of permanent form was possibly stable. Scientists, like Russell,
all knew that it was quite difficult to produce a solitary wave even in a labo-
ratory. Engineers who tested ship models in long water tanks made effort to
generate regular wave trains: waves of permanent form. For decades, engineers
had been troubled by the gradual steeping and final breaking of the initially
regular waves. People had no doubt about the instability of the waves but
instead they blamed the insophistication of the laboratory apparatus and the
imperfection of their experimental conditions. Despite this interesting history,
these finite amplitude Stokes waves are in fact unstable.
The historical discovery of the instability of Stokes waves was due to Ben-
jamin and Feir (1967). The instability is due to the fact that the amplitude
of the Stokes wave (8.5.1) is very vulnerable to even a very small disturbance
near its phase (b - wt). The instability criterion is that infinitesimal distur-
bances of the phase near kx - wt will undergo unbounded magnification which
eventually lead to an instability of the initial regular wave train if
0< &~ v'2 ka (8.5.3)
for a small positive number &defined below.

Although the detailed analysis of the instability is necessarily complicated,
the essential mechanism is rather simple and was ingeniously explained by
Benjamin and Feir (1967). In the Stokes wave train (8.5.1), let us denote the
phase by
(= kx - wt.
The disturbance, that sucks energy away from this basic wave train to amplify
its own amplitude indefinitely, consists of a pair of progressive wave trains
whose phases are slightly different from (:
(1 = (+e - ")'1 (Upper side band), (8.5.4)

(2 = (- e- 1'2 (Lower side band) (8.5.5)
e
where = K.X - &t is only a small fraction of ( (K. is a small fraction of k and &
is a small fraction of w), 'Y1 and 1'2 may be functions of t or constants. The &
appeared in (8.5.3) is the same as the &here. The phases (1 and (2 are referred
Figure 8.13: Benjamin-Feir instability: (a). view near to wave maker (basic
regular wave); (b) view at 200 ft. further down from the wave maker (irregular
wave). The basic wave length is 7.2 ft. [From Benjamin (1967), by permission
of The Royal Society of London]
8.5. Benjamin-Feir: Side-Band Instability 245
to as the side band phases of the basic phase (, and the corresponding frequen-
cies are called the side band frequencies. So, the Benjamin-Feir instability is
sometimes called the side band instability.
The amplitudes of the side band wave trains are assumed to be f1 and
f2 which are much smaller than the amplitude of the basic wave a (0 < fda,
f2/ a 1). This is the common assumption of small disturbance when studying
the stability. The nonlinear interactions among these three wave trains exist
because of the nonlinear boundary conditions. After the interaction, there will
be wave components with phases
2( - (1 = (2 + ("n + 12), (8.5.6)

2( - (2 = (1 + (11 + 12). (8.5.7)
The corresponding amplitudes are a 2 f1 and a 2 f2' respectively. It is well known

that two waves of the same frequency resonate with each other and the ampli-
tude of the new wave grows indefinitely in time. Thus, if it happens that
() = 11 + 12 --t constant (8.5.8)
as the nonlinear processes develop in time, each mode of the f1 wave train
will generate a cosine component that resonates with a cosine component of
the f2 wave train and vice versa. Therefore, if () # 0,11", the two wave trains
amplify each other and their amplitudes grow indefinitely. Benjamin and Feir
showed that the amplitudes actually grow exponentially in time. So the basic
wave becomes unstable and its energy is sucked away by the side band waves.
Although the frequencies are only slightly apart (almost undetectable) from
that of the basic wave, the amplitudes of the side band wave grow quickly and
eventually the waves break. See Fig. 8.13 for the experimental evidence of the
Benjamin-Feir's side band instability. The basic wave train is produced by a
wave maker. As it travels to the other end of the tank, the side band waves get
amplified, then become irregular and finally break. The existence of the side
band waves is inevitable since any regular wave produced in laboratory or in
nature are subject to some kind of slow modulation. As we know from Chapter
1, slow modulation is due to the superposition of two harmonics of similar
frequencies. Thus, when a wave maker generates the basic finite amplitude
periodic wave, it also generates a number of unnoticeably small amplitude side
band waves at the same time.
The mathematical analysis of the Benjamin-Feir instability is in the cate-
gory of nonlinear stability analysis. The method is completely different from
what we learned in the last three sections. The entire analysis is quite com-
plicated and omitted in this book. Interested readers may read their original
paper (Benjamin and Feir (1967) [9]) which was very well written and compre-
hensible even for first year graduate students.
Later developments on the theory of deep water waves have simplified the
original analysis by deriving a cubic nonlinear Schrodinger equation. Further,
people have analytically and numerically found that the irregular waves induced
by several side band small waves can reorganize themselves into a nicely modu-
lated wave. This is similar to the recurrence phenomenon of Fermi-Pasta-Ulam
(1955) .
[1] P. G. Drazin and W. H. Reid (1981), Hydrodynamic stability, Cambridge

University Press, New York, Chapters 1, 2 and 3.
[2] P. K. Kundu (1990), Fluid Mechanics, Academic Press, New York, Chapter
11.
Stanford, California.
[4] S. Chandrasekhar (1981), Hydrodynamic and Hydromagnetic Stability,
Dover Publications, New York.
[5] C. C. Lin (1955), The Theory of hydrodynamic stability, Cambridge Uni-
versity Press, New York.
[6] C. S. Yih (1988), Fluid Mechanics, West River Press, Ann Arbor, Michigan,
Chapter 9.
[7] E. Infeld and G. Rowlands (1990), Nonlinear Waves, Solitons and Chaos,
Cambridge University Press, New York, Chapter 1.
[8] H. Lamb (1945), Hydrodynamics, Dover Publications, New York, Chapter
9.
[9] T. B. Benjamin and J. F. Feir (1967), The disintegration of wave trains on
deep water, Part 1. Theory,}. Fluid Mech. 27,417-430.
[10] D. Coles (1965), Transition in circular Couette flow, J. Fluid Mech. 21,
385-425.
[11] T. B. Benjamin (1967), Instability of periodic wavetrains in nonlinear
dispersive systems, Proc. Roy. Soc. Lond. A 229, 59-75.
Chapter 9
Wave Interactions and

X-Ray Crystallography
The Fourier representation of plane waves in fluid motions may be expressed

in the form
u(r, t) = Lan exp[i21l"n(kn . r - wnt)] + c.c.
n
The coefficients an, the wave numbers k n , and the wave frequencies Wn are
determined by the conservation laws of mass and momentum as well as bound-
ary and initial conditions. Each term in the Fourier representation is called a
mode. In a linear system, the modes do not interact with each other. In a non-
linear system, these modes do interact with each other and these interactions
generate new modes. So, strictly speaking, the above Fourier representation of
a plane wave for a nonlinear system is valid only for a specified moment of time
or a specified short time interval. The new modes are generated only when
247
248 Chapter 9. Wave Interactions and X-Ray Crystallography
resonance conditions
kl k2 k3 = 0 for three waves,

Wl W2 W3 = 0,
or
kl k2 k3 k4 = 0 for four waves,

Wl W2 W3 W4 = 0
are satisfied. The third mode is generated by the first two modes in a three-
wave interaction process and similarly the fourth mode is generated by the first
three modes in a four-wave interaction process.
In X-ray crystallography, the electron density has a Fourier representation:
p(r) = L: IEHI exp(i21l'1PH) exp(i211'H . r)

H
where H is a reciprocal lattice vector and 4>H is the phase of the Fourier
coefficient EH' In the direct method to find a crystal structure, one needs to
assign a value to 4>H for each H. The expectation value of the third phase 4>L
(as a random varaible) can be calculated when the first two phases (4)H' 4>K)
are known and the invariant condition
H+K+L=O
is satisfied. The expectation value of the fourth phase 4>M can be calculated
when the first three phases (4)H' 4>K' 4>L) are known and the invariant condition
H+K+L+M=O
is satisfied.
This is the major part of the mathematical similarity between the wave
interaction problem and the phase problem in the X-ray crystallography. Both
problems are very fashionable scientific research subjects in the modern days.
In this chapter, we will explain the basic ideas behind the pertinent subjects
and try to relate the methodologies used in these two seemingly uncorrelated
research areas.
9.1 Wave Interactions

9.1.1 Introduction
Strictly speaking, purely monochromatic waves do not exist in nature. Waves
with different wave numbers or frequencies exist in the same physical, chemical
or engineering system although there may be one or more primary waves with
specified wave numbers and frequencies that initially dominate the wave mo-
tion. Although those waves with different wave numbers or frequencies from
9.1. Wave Interactions 249
that of the primary waves have very small amplitudes, they may playa very
important role in changing the motion behavior of the primary waves and con-
sequently cause the instability of the system. This is the direct result of the
wave-wave interaction. The Benjamin-Feir side band instability discussed in
section 8.5 is one of the numerous examples of this type.
The consequences of the wave-wave interaction in certain cases are quite
clear, yet the mathematical analysis is usually very complicated and the precise
mechanism responsible for the energy transfer among the participated modes
is far from trivial. Students generally feel that wave-wave interaction is one of
the most difficult subjects in a nonlinear wave course. We plan not to have
the ambition to make the students understand all the mechanisms and mathe-
matical analysis involved in the wave-wave interaction for all systems. Instead,
we would like to clearly elucidate a few wave interaction mechanisms through
some specific examples. Here, we emphasize the understanding of the outcome
of the wave interactions. This is the place where readers might benefit a lot by
changing their philosophy from understanding the outcome from mathemati-
cal solutions to finding the mathematical solutions from understanding what
is the expected outcome. Perhaps, when we were graduate students, most of
us had the unproductive experience of being so attracted by the mathematical
formality that we forgot to understand the basic mechanisms and fundamental
ideas involved.
We will start with a simple example of resonance in a linear forced harmonic
motion. The resonance condition in this example leads the students to foresee
the conditions of resonance in other linear and nonlinear systems. With little
effort, the students will be able to come up with resonance conditions for three-
and four-surface wave interactions in deep water.
These resonance conditions, in mathematical formality, are the same as
the invariant conditions in the phase problem in X-ray crystallography. The
resonance condition guarantees the energy transfer- from the primary modes
to the newer modes and are hence a type of conservation laws. The invariant
condition guarantees the invariance of certain linear combinations of phases
and is hence also a conservation law.
9.1.2 Forced harmonic motion

Consider a one dimensional harmonic oscillator subjected to an infinitesimal
periodic forcing. This simple example reveals a very neat resonance condition
that can be extended and applied to many more complex systems of wave-wave
interactions. The mathematical equations that describe the forced system are
Ii + w 2x = f cos[f2t], i:(t = 0) = i: o, x(t = 0) = Xo (9.1.1)
where w is the natural frequency of the system, f2 the frequency of the periodic
forcing, f the infinitesimal amplitude, and i:o and Xo the initial conditions. The
response of the system to the forcing can be found by the solution to the ODE
(9.1.1):
[XO coswt + ~ sinwt] + w2~n2 cos nt, when n2 :f. w2 ,

x(t) = { (9.1.2)
[xocoswt + ~sinwt] + 2~tsinnt, when n2 = w2
The first part of the above solution is the response to the initial condition
and has little interest to us. The second part of the solution which is pro-
portional to f is the response of the system to the infinitesimal forcing. The
consequence of this forcing is dramatically different in the two cases: n2 = w2
(resonant case) or n2 :f. w2 (nonresonant case). When the forcing is not reso-
nant with the system, the response of the system to the forcing is periodic and
bounded (or infinitesimal when n2 is far away from w 2 ). But when the forcing
resonates with the system (n 2 = w2 ), the response of the system increases
linearly with time ( ftj(2n)). Eventually, the response becomes unbounded.
The oscillation at the resonant frequency absorbs away all the energy of the
forcing. Hence, the resonant forcing causes the instability of the original oscil-
lation system which is supposedly set in motion by an initial condition [there
is no dissipation here].
We can rewrite the resonance condition n2 = w 2 in another form
nw = o.
The resonant condition in this form can be easily extended to a system that
has several natural frequencies with the same order of amplitude. For a linear
system of many degrees of freedom, each mode of a fixed frequency does not
interact with any other mode to produce new modes. Hence the resonance
occurs if and only if the frequency of the external forcing is exactly equal to
one of the fundamental frequencies, i.e.
for only one j. This j can be any number of 1,2,, N, and N is the degree
of freedom of the system.
In a linear system, a mode can only resonate with an external forcing but
not with other modes of the system. In contrast to this, in a nonlinear system,
a mode can sometimes resonate with other modes of the same system and new
modes are generated from the nonlinear interactions.
9.1.3 Resonance conditions for nonlinear systems

For capillary nonlinear surface gravity waves in water, the nonlinearity comes
into the system through the free surface boundary conditions. The nonlinearity
is of the second order in the Bernoulli equation on the free surface. Sometimes,
we need to expand the boundary conditions on the wavy free surface in Taylor
series about the flat water surface. Depending on the truncation order, the
resulting system may have a nonlinearity of the third order or higher (e.g.
cubic nonlinear Schrodinger equation for deep water surface waves).
In general, when the wave amplitude is small (ofthe order e), a wave system
may be described by an equation of the form
.c(u) = eN(u) (9.1.3)

where .c represents a linear operator, N a nonlinear operator, e a small positive
number and u the unknown physical quantity in question. The linear part of
the equation has a traveling wave solution u/ =
exp[i(k . x - wt)] where the
frequency wand the wave number k satisfy a dispersion relation:
W(w,k) = O. (9.1.4)
This dispersion relation is obtained from
.c (exp[i(k . x - wt)]) = W(w, k) exp[i(k . x - wt)] = O.

For instance, the Boussinesq equation for long shallow water waves
Utt - (1 + fUx)U xx + -YUxxxx = 0

is an example of the form (9.1.3). For this equation, we have
.c(u) = Utt - Uxx + -YUxxxx ,

N(u) = UxUxx .
Then, the dispersion relation is
and
W(w,k) =w 2 _k2+-yk4.
When we substitute u/ = exp[i(k . x - wt)] into equation (9.1.3), the right
hand side of the equation may be considered as the forcing of a linear system.
If the frequency on the right hand side is the same as the natural frequency of
the linear operator .c, then a resonance occurs.
Suppose N has a second order nonlinearity. Let Wl and W2 be the frequencies
oftwo primary waves which are solutions of the linear equation .c(u) = 0:
Ul = al exp[i(k 1 . x - wlt)],
U2 = a2 exp[i(k2 . x - w2t)].
The corresponding wave numbers satisfy the dispersion relations W (Wi , ~) =
o (i = 1,2).Let u be a linear combination of Ul and U2. Then .c(u) is also a
linear combination of .c(u!) and .c(U2). But, N(u) includes terms of the type
exp i[(k 1 k 2 ) x - (Wl W2)t].

The corresponding amplitude is ala2 = 0(e 2) if al = O(e) and a2 = O(e). So
the term
(9.1.5)
is a small amplitude forcing of C. The response of the system to this forcing is

in the order of a3 = O( f3) unless resonance occurs. The resonance occurs only
=
when the frequency of the third mode U3 a3 exp[i(k3 . x - W3t)] of the linear
operator equals the frequency of the interaction product (9.1.5). In other words,
ki k2 and WI W2 satisfy the dispersion relation W(WI W2, ki k 2) = O.
Thus, the resonance condition is
W; = (WI W2)2 ,
Ik312 = Iki k212,
or
WI W2 W3 = 0, (9.1.6)
ki k2 k3 = O. (9.1.7)
The third mode U3 is now amplified by the interaction product and withdraws
energy away from the two primary modes. The amplitude of the third mode
is zero at time t= 0 and grows initially in time linearly. Consequently, as the
energy is transferred from the original two primary modes to the third, the
amplitudes al and a2 decrease. Therefore, the amplitudes at, a2 and a3 are
functions of time. One can obtain three coupled nonlinear ODEs that govern
the evolution of these amplitudes. These equations are called the interaction
equations and usually can be written in the following form:
ial = C I a;a;,
ia2 = C2aiar,
ia3 = C3 aia;
where i = V-I is the imaginary unit, " * "is for complex conjugation, and
C I , C 2 and C3 are constants independent of t and the amplitudes aI, a2 and
a3. To derive the interaction equations, let
3
U = L:>j(t) exp[i(kj . x - Wjt)];
j=1
substitute this u into the equation (9.1.3) and set the secular terms to zero
using the resonance condition. Although the steps are straightforward, it is
still very tedious to derive the interaction equations even for a very simple
system, such as the Boussinesq equation. We hence avoid the derivations in
this book.
The energy transfer mechanism among the three modes can be applied to
a mode interacting with itself (2wt) , an old mode interacting with a newly
generated mode (W3 +wt) to yield the fourth new mode, and so on. Because of
these nonlinear interactions and the generations of new modes , periodic water
waves are hardly stable as the result of the disintegration of the length and
time scales.
~ ,, ,
,
Beach
:
I
,
I
--.---
-,------r----'---
k, I I
-.-- -.-----.--- -.-----,---

I I I
-
I I I
I I I
Beach Ita 2k1 I
I
--.,----------
I
I
I
I I I
I
--+----
I
I I
I I
'-_1- ____ --!----
J
- f-""-----
I I I
I I I
- -~-----~-----~---
I Pliiiism I
Figure 9.1: The top view of a wave interaction water tank. [From Phillips
(1974), by permission of Cornell University Press]
101. -_ _ _.....
o 100 200
t
h
t
Ii J.I t
Y'i
2Jj7fz
Figure 9.2: Phillips' experimental results of water wave interactions. [From
Phillips (1967), by permission of the Royal Society of London]
Figure 9.3: The experimental results due to McGoldrick et. al (1966) on water
wave interactions. [From Phillips (1967), by permission of the Royal Society of
London]]
The aforementioned three waves can be observed in a water tank. One

may design a water tank similar to that shown in Fig. 9.1. Primary waves
are generated by plungers on the sides AB (plunger 1) and Be (plunger 2 ).
The circular frequency of the plunger 1 is It = wd21r and that of the plunger
2 is h = W2/21r. The wave numbers of the primary waves generated by the
= =
plungers are kl kd and k2 -k2i respectively. Here, i and j are unit vectors
in x and y directions respectively. One can adjust the ratio of the frequencies
so that W(Wl W2, kl k 2) = 0 holds (resonant) or not (nonresonant). In
the case of resonance, the new mode k3 will be observed. The observational
data due to Phillips (1967) is shown in Fig. 9.2. The horizontal axis is the
frequency of the waves and the vertical axis is a measure of the energy given
by the square of the Fourier coefficients of the resulted wave. The peaks at It
and h are apparently due to the primary waves generated by the two plungers.
Other peaks correspond to the third new mode whose frequency fa are It + h,
21t - h and others.
Fig. 9.3 shows the experimental results due to McGoldrick et. al (1966).
The two panels demonstrate the difference in the system's responses to the
resonant and non-resonant primary waves. The top panel is at resonance while
the bottom one is off resonance. The largest new mode due to the nonlinear
resonance is at 211 - h. This corresponds to the wave number 2kd - k2i and
agrees with the results due to Phillips presented above.
9.1.4 Four-wave interactions

Sometimes, it is impossible for a quadratic nonlinearity to generate a new res-
onant mode from two primary modes, i.e. the three-wave interaction described
in the above section is impossible. The reason is that the new modes generated
by the sum or the difference of two primary waves do not satisfy the dispersion
relation determined by the linear operator C. But, the third order nonlinearity
may generate new modes from the three primary waves. The phases of the new
modes are determined similarly by the triple difference or the triple sum of the
phases of the three primary waves. So, four-wave interactions are possible in
certain systems. One may generalize this philosophy to five-wave interactions
due to the fourth order nonlinearity in the nonlinear operator N when the three
and four wave interaction become impossible at the second and the third order
nonlinearities. It seems that people have not found any laboratory evidence of
five or more wave interactions. Possibly, the nonlinear effect in five wave reso-
nance is in the order of ala2a3a4a5 = 0((5) which is too small to be observable
in a relatively short time interval. Yet the energy transfer at the fifth or an
even higher order of nonlinearity, in principle, can still cause an instability in
a longer time.
As an example, let us show that no second-order resonant interactions are
possible in pure gravity waves in deep water. The lowest order resonance
occurs at the four-wave interaction caused by the third order nonlinearity in
the nonlinear operator N and the outcome of the four-wave interaction is in
the order of 0((4). People have already found numerical and experimental
evidence of this type of four-wave interactions. In section 8.5, we have already
learned that the Stokes wave in deep water are unstable because of the side
band interaction and energy transfer (Benjamin-Feir instability).
The configuration of the deep water gravity waves in three dimensions is
shown in Fig. 9.4. By pure gravity surface waves, we mean that the capillary
force is neglected and the potential theory can be applied. The governing
equations of the wave motion are as follows:
in the fluid domain
d!.p =0 when - 00 < Z < ((x, y, t); (9.1.8)
on the free surface z = ((x, y, t) (expanded in Taylor series about z = 0)

1 2
(t - !.pz - (!.pzz - 2'( !.pzzz
+!.px(x + !.pxz((x + !.py(y + !.pyz((y = 0, (9.1.9)

(2
g( + !.pt + (!.pzt + 2!.pzzt
z=l;(x, y, t) Air
~
y
Water
z=-co
x
Figure 9.4: The three-dimensional configuration of surface waves in deep water.
(9.1.10)
and at z =-00
ep = epx = epy = epz = O. (9.1.11)
The nonlinearity in the system appears only in the surface conditions. For
the leading order term, one may eliminate ( from equations (9.1.9) and (9.1.10)
to obtain an equation for ep. In symbolic terms, we have
(9.1.12)
The linear part on the left hand side determines the linear operator C and the
dispersion relation
(9.1.13)
The quadratic term on the right hand side for two waves Ul and U2 generates
a mode whose phase is
The frequency Wl W2 and the wave number kl k2 can not satisfy the dis-
persion relation (9.1.13) since
{Wl W2)2 = glkl k21, w~ = glkll, and w~ = glk21
hold simultaneously only when kl = 0 or k2 = O. Thus, there is no third mode

that can resonate with the mode generated by the second order nonlinearity
from the two primary waves.
The interaction of three waves in the third order generates new modes whose
phases are in the form
Since
k1 . k2 k2 . k3 k3 . k1 = 0
is possible, it may happen that
If this occurs, then the new mode will be at the resonant state and this new
mode has frequency W4 and wave number k 4 . The resonance condition is ap-
parently
Ik412 = Ik1 k2 k312,

IW412 = IW1 W2 w312,
1.e.
W1 W2 W3 W4 = 0, (9.1.14)
k1 k2 k3 k4 = o. (9.1.15)
To find the interaction equations, we suppose
4
'P = f L Al(t) exp [ikl . r + iWlt + IkLlz]
j=l
4
+f 2 L Alm(t)exp[i(kl+km)r+i(wl+Wm)t+lkl+kmlz]
l,m=l
4
+f3 L ALmn(t)exp [i(kL +km +kn) r+i(wL +wm +wn)t
L,m=l
where r = (x, y) and kL, k m and k n are two-dimensional wave numbers. This
expression satisfies the Laplace equation and the boundary condition at z =
-00 automatically. The surface boundary conditions can solely determine the
interaction equations. Substituting the above 'P into the surface boundary
conditions (9.1.9) and (9.1.10) and making the secular terms equal to zero
using the resonance conditions, one finally obtains the interaction equations
for A1 (t), A2(t), A3(t) and A4(t) (the leading order terms in the expansion
above). The interaction equations are in the form
4
A1 = iA1 LC1jlAjl2 + d1A;A3A4exp(ia1t),
j=l
where clj,d1 and 0:'1 are constants independent oft and Aj (j = 1,2,3,4).
The complete derivation of the above interaction equations is very complicated
and thus omitted here. Interested readers may find the details in the book by
Craik (1985) and the references therein.
9.1.5 Nonlinear wave interactions III other systems

It seems that the theory of nonlinear wave interactions was first established
for studying surface wave interactions in water. It explained and demonstrated
many interesting phenomena: wave instabilities, wave reorganizations, and oth-
ers. The idea of the wave-wave interaction is not restricted to water waves and
has already been used in other fields. The applications of the theory in study-
ing the successive state transitions in the Taylor-Couette flows, the Rayleigh -
Benard convections and the onset of turbulence near the critical layer in shear
flows have proven to be successful. All these have been known to many applied
mathematicians, physicists and engineers. But a less known success of the use
of the wave resonance conditions is found in X-ray crystallography. A crucial
problem in X-ray crystallography is to solve the phase problem. Linear com-
binations of three or four phases in the electron density function are structure
invariants if certain invariant conditions are satisfied. These invariant condi-
tions are, in mathematical formality, exactly the same as the wave resonance
conditions. We devote the next section to studying the phase problem in X-ray
crystallography.
9.2 Phase Problem in X-ray Crystallography

9.2.1 Bragg's law of X-ray diffraction
Centuries ago, people noticed that certain materials in nature always main-
tain a hexagonal or a tetragonal shape no matter how fine the materials were
ground. These materials, such as salt, are crystals. The group symmetries in
crystal structures were studied as early as the last century. One can find some
crystallography books dated back to 1903 or earlier. However, it is a much
more complicated task to detect the structure of a given crystal. This task has
been challenging people and will continue to do so.
The first set of experiments on scattering of X-rays by a crystal were not
done until 1912, nearly two decades after the discovery of X-rays in 1895. The
experiments were only concerned with exploring the periodicity in a crystal.
Shortly afterwards, on the spectrometer built by W. H. Bragg, spectra were
measured for some alkali halides. The Braggs, father and son, solved these
simple structures requiring only the concepts that various sets of atoms were
scattering either in phase (211", bright spot) or out of phase (11", dark area) with
9.2. PhaBe Problem in X-ray Crystallography 259
each other for each of the diffracted beams. This is the famous Bragg's law for
the in phase scattering
2d sin () = 211"n'\ (9.2.1)
where d is the perpendicular distance between two parallel planes passing
through certain crystal lattice points, () the angle between the incident rays
and the plane of diffraction, and ,\ the wave length of the X-ray. Both d and ,\
are in the order of angstroms ( l[angstrom] = lO-lO[meter]). See Fig. 9.5 for
details.
The incident X-rays may be regarded as a traveling wave
lin = a; exp[i211"(ko . r - wot)].
So the incident direction is ko with Ikol = 1/,\, and wo'\ being the speed of the
X-ray. The constant ai is complex valued and its magnitude is the strength of
the incident X-rays.
Suppose that the diffraction from the parallel planes is uniform. Then, when
reaching the screen, the diffracted rays have the same strength but different
phases. Each diffracted ray may be expressed in the form
d = ad exp[i211"(k . r - wot - tP)]

where k is the direction of the diffracted ray with Ikl = 1/,\, and tP being the
phase difference. The constant ad is again complex valued and its magnitude
is the strength of the diffracted wave. The phase difference between two nearby
diffracted rays is apparently 2d sin 0/,\, which can be written in the form
2dsinO/,\ = N . (k - k o)
where N is a normal vector of the diffraction plane and its magnitude is INI = d.
The superposition of the diffracted waves from these parallel planes is
2: adn exp[i2n1l"(k . r - wot)].

n
As shown in Fig. 9.5 (d), the two near by diffracted rays can maximally
enhance each other (positive interference) if the phase difference between them
takes the value 211"n for any integer n. This is Bragg's law (9.2.1). Because
of this interference, bright lines and dark bands appear periodically on the
screen (or a film) in the case of diffraction by a two-dimensional grating (see
Fig. 9.6 (b) for the strengths of the waves on the film). In the case of a
three-dimensional crystal, the intersections of the bright lines result in periodic
bright spots on the screen (or the film). These spots are called the diffraction
pattern or Laue spots (Fig. 9.6 (a)).
9.2.2 Fourier representation of electron density

In a crystal, most electrons are around nuclei. If we know the distribution of
the electrons (i.e. the electron density function), then we know the relative
Film
Two diffracted waves

0 t
-1
-2
0 0.1 0.2 0.3 0.4
Larger wave:
in phase superposition
0 t
-1
Smaller wave:
out phase superposition
-2
0 0.1 0.2 0.3 0.4
Figure 9.5: Bragg's diffraction by paraHel planes in a crystal.

9.2. Phase Problem in X-ray Crystallography 261
(a)
(b)
Figure 9.6: (a) Laue spots on a film [From McPherson (1982), by permission
of John Wiley]; (b) Wave interference on a film.
positions of the atoms that constitute the crystal. So, the structure of the
crystal is known.
Let p(r) be the density function of electrons. This function must take on
very large values in the neighborhood of an atom and be almost zero in other
places. Hence, in the vicinity of an atom, this density function looks like a
three-dimensional Dirac delta function. Since a crystal consists of a periodic
arrangement of atoms, the density function p(r) must also be periodic in space.
In 1915, Bragg suggested to express this periodic function in terms of Fourier
series
00
1
=
p(r) IIVII L: F(h, k, f) exp[-21l'i(hz + ky + fz)].
h,k,l=-oo
The coefficient F(h, k, f) can be computed in a three dimensional period (called

a cell and denoted by V), and II VII is the volume of V). The Fourier coefficients
are
F(h, k, f) = [p(r) exp[21l'i(hz + ky + fz)] dzdydz.
The strength of the diffracted wave depends on the IF(h, k,f)1 and these
quantities, in turn, satisfy the Parseval identity
~ IF(h, k, fW = [p2(r) dzdydz.

h,k,t_-oo
From the diffraction pattern recorded on a film, an intensity II(h, k, f)1 can be
measured from each spot. One can then determine the strength ofthe diffracted
wave, i.e. IF(h,k,f)1 = v'II(h,k,f)l. But,
F(h,k,f) = IF(h,k,f)lexp[icfo]
has a phase cfo to be determined and the diffraction pattern records no phase
information. Hence, the phase can not be measured by Bragg's diffraction
experiment and so far there is no way one can experimentally measure this
phase. An alternative is to make many measurements of the intensity and use
the magnitudes IF(h, k, f)l to derive the phase values.
In summary, one can only measure the magnitudes of the coefficients in the
Fourier series of the electron density function and the phases of the coefficients
have to be calculated. Determining the phases from the magnitudes is called
the phase problem in X-ray crystallography. As a matter of fact, the phase
problem is still a major difficulty in using X-ray crystallography to determine
the crystal structure despite the large amount of work having been done in this
area. In the 1950s and 1970s, J. Karle and H. Hauptman made tremendous
contributions to the problem by introducing the so called direct method. In
1985, these two pioneers were suitably honored by the award of the Nobel prize
for Chemistry.
Figure 9.7: Unit cell in a crystal.
9.2.3 Coordinates in crystal cells

To proceed further in the phase problem, we need to understand some basics
about the geometry and symmetry of a crystal and define coordinates.
It is well known that among all polygons only identical triangles, rectangles
(parallelograms) and hexagons can fill up a plane. Similar to this, there are
only finitely many identical polyhedrons that can fill up three dimensional
space. The symmetry of a crystal may be classified by space groups. All types
of symmetries and groups of real crystal lattices have been described in the
International Tables for X-my Crystallogmphy. However, in understanding the
phase problem and the direct method, familiarity with the space groups and
the symmetries associated with real crystal lattices is not required.
Under whatever circumstances, there are unit cells which are parallelepipeds
spanned by three linearly independent vectors: a, h, c. The lengths of these
three vectors are in the order of angstroms. See Fig. 9.7 for a unit cell in a
crystal. A completed crystal consists of the translation of the unit cells along
the a, hand c directions. Therefore, we choose a, h, c as the basis of our
crystal space and adopt their lengths as the unit lengths in each direction. A
point inside a cell is represented by a vector r = xa + yh + zc or (x, y, z) for
o ~ x, y, z ~ 1 (see Fig. 9.7).
From Bragg's law described in subsection 9.2.1, we have seen that in the X-
ray diffraction process the positions of the set of parallel diffracting planes play
a very important role. But we do not need to determine the position of each
specific plane, rather we only need to determine: (i) the normal of the plane
set, and (ii) the perpendicular distance between any two neighboring planes.
In crystallography, we determine the positions of this set of planes by so called
Miller indices (h, k, l) where h, k and l are integers. The Miller indices are
determined by conditions: (i) ha + kh + lc is a normal vector to the plane;
and (ii) h = the number of intersections of the plane set with the wctor a
in the unit cell and k and l are similarly defined. If the plane set intersects
with -a, then h takes a negative value. Thus, a set of given Miller indices
(h, k, l) determines a light grating: a set of parallel diffracting planes. The
lattice points in a crystal can form many light gratings since each three lattice
points, in principle, can form a diffraction plane. The Miller indices for a plane
(a) (b)
Figure 9.8: (a) Parallel plane sets; (b) reciprocal lattice points. [From McPher-
son (1982), by permission of John Wiley]
set are not unique: (I, I, I) and (1,1,1) represent the same set of planes. Here
I stands for -1. The larger the Miller indices, the closer the planes in the
gratings. See Fig. 9.8 (a) for some examples in determining Miller indices.
Orthogonal to this (a, b, c) frame is the reciprocal frame (a*, b*, c*):
a* = b xc/V, b* = c x a/V, c* = a x b/V (9.2.2)
where V = a (b x c). The space spanned by (a*, b*, c*) is called the reciprocal
space. Then, the vector ha* + kb* +fc* corresponds to a point in the reciprocal
space. Hence, each parallel plane set in the crystal space corresponds to two
points in the reciprocal space and these two points are symmetric with respect
to the origin. Considering all possible plane sets in a crystal, we get a set
of points in the reciprocal space. These points constitute a new lattice and
this new lattice is called the reciprocal lattice. See Figs. 9.8 (a) and (b) for
the correspondence between the parallel plane sets and the reciprocal lattice
points.
The purpose to construct a reciprocal lattice is not to find another real
lattice, rather it is for computational convenience. It is relatively difficult to
count all the possible diffraction plane sets. But it is very easy to count the
lattice points in the reciprocal lattice. We will see in the next subsection that
the vectors in the reciprocal space correspond to the wave numbers. Hence,
the reciprocal space may be considered as the Fourier space (also called the
spectrum space). When using the Fourier representation of electron density,
we can convert the relevant computations into the operations in the Fourier
space, i.e. on the reciprocal lattice.
9.2.4 The phase problem

The intensities of a sufficient number of X-ray diffraction maxima determine
a crystal structure. The available intensities usually exceed the number of
parameters needed to describe the structure. From these intensities, a set of
numbers !PH I can be derived, one for each intensity. To determine the phase
of the structure factor FH = !PHI exp(i<PH), one has to determine the phase
<PH which can not be measured from the diffraction experiment. Since we have
more than enough magnitude data, in principle it is possible to determine the
phase by using the magnitude data. The problem to use the magnitude data
!PH I to find the phase is the phase problem in crystallography.
The electron density is expressed in terms of Fourier series
p(r) =~ L FH exp( -21riH . r) (9.2.3)

H
where V represents a unit cell and its volume as well, H = ha* + kb* + I.e* is
a reciprocal lattice vector and the summation of H is over the entire reciprocal
lattice. The Fourier coefficients are called the structure factors and can be
written as
FH = [p(r) exp(21riH . r) dV = !PHI exp(i<PH)' (9.2.4)
The integration above is over the unit cell V.

To solve the phase problem, we write the above Fourier pair in a discrete
form. Suppose that there are N discrete, non-vibrating, point atoms in the
cell. Then the Fourier pair becomes:
if
(9.2.5)
if
N
EH = L(Zj/u)exp(21riH .rj) == IEHlexp(i<PH) (9.2.6)
j=l
where Zj is the atomic number of the jth atom inside the cell,
N
U = (L Zj2)1/2
j=l
is proportional to the total number of electrons in a unit cell, and rj is the po-
sition ofthe jth atom. Equation (9.2.5) shows that the positions of the maxima
of the Fourier series (9.2.5) are at the desired atomic position rj; but in order
to calculate (9.2.5) it is necessary to know the complex Fourier coefficients EH'
that is to say, both the magnitudes IEHI, derived from experiment, and the
phase <PH, lost in the diffraction experiment. Apparently, for the above dis-
cretization of the Fourier transform pair to be valid approximately, a necessary
and sufficient condition is that electron density function is very close to a three
dimensional Dirac delta function. The resolution of the atom positions is in
the order of one angstrom. This is true for small molecules of several dozen
or even a few hundred atoms. But, for large protein molecules it is sometimes
questionable whether the discretization (9.2.5) and (9.2.6) is still valid since
these protein molecules contain a large number of atoms and the resolution of
diffraction maxima is severely limited.
The chemical components are supposed to be known (i.e., Zl,, ZN are
known). The unknowns are the positions of the atoms (rl,, rN). So we have
3N unknowns which should be determined by the equations
N
IEHI = ~)Zj/(T) exp(21riH . rj) . (9.2.7)
j=l
The magnitudes IEHI have already been determined by the diffraction experi-
ment. We usually have far more than 3N lattice points in a reciprocal lattice
since we perform the X-ray diffraction experiments on a crystal in various di-
rections of incident X-rays by rotating the crystal. So the system of equations
(9.2.7) includes more than 3N equations for only 3N unknowns. Hence, the
system is redundant.
If all the measurements were absolutely accurate, the equations in the family
(9.2.7) would not contradict one another. Therefore, the phase problem is, in
principle, solvable when formulated in terms of non-vibrating, point atoms.
However, the measurements can not be absolutely accurate. Then there are
two technical difficulties in solving (9.2.7) to find rjs. Firstly, the equations
might contradict each other due to measurement errors. Secondly, it is not
practical, even with available supercomputers, to solve the 3N highly nonlinear
equations for a relatively large N, say N = 500. A modified scheme is to find
an approximate solution to the phase problem. This scheme is to find rjs that
minimize the weighted sum of squares
(9.2.8)
where the sum is taken over all points in the reciprocal lattice, and WH are a
suitably chosen set of weights and can be regarded as a Lagrange multipliers.
This scheme overcomes the first difficulty pointed out above. Although the
global minimum of (9.2.8) may not be zero, it, in principle, corresponds to
an approximate solution. But, in practice, it is still not possible to solve this
minimization problem directly as formulated in (9.2.8). What would be a way
to get out of this dilemma?
Recalling the definition of the phase problem, we would like to determine the
phase <PH. Why do we need this quantity if what we want to know is the crystal
structure rjs? It is expected that the calculation of phases would help us with
finding the desired atomic positions rjs as suggested by equation (9.2.5). Thus,
it is pointless to find rjS and then to compute 4>H by using (9.2.6). We have
just walked back to where we started. Mathematically, it is both good and bad
to have a computation loop. It is bad that one can not find an exact solution.
But, it is good that one can break the loop by assigning trivial values to certain
unknowns and start an iterative process to get an approximate solution. This
iteration scheme and an approximate solution are exactly what we need. Next,
we will develop an iterative scheme to find an approximate solution.
The best place to break this loop seems to be at the step (9.2.6) where
we assume a value for 4>H' At first glimpse, it seems that 4>H can take any
value between -11' and 11'. But because of the minimization formulation (9.2.8),
one may naturally think that for a given point H in the reciprocal lattice
(corresponding to a parallel plane set in the crystal) the structure factor EH
must have a preferred direction like that in the steepest descent method in
mathematical optimization. In other words, 4>H has a probability distribution
that peaks at the preferred direction. Fortunately, there is a way to find such
a probability distribution function (pdf). So according to this pdf we assign a
value to 4>H for each reciprocal vector H. With this 4>H, one can compute the
left hand side of (9.2.5):
(9.2.9)
This sum is an approximation of a periodic three dimensional Dirac delta func-

tion. An example of a two dimensional sum as an approximation to a Dirac
delta function is shown in Fig. 9.9:
6
D{x, y) = L exp[211'i{mx + ny)].
m,n=-6
The peak points of the expression (9.2.9) correspond to the first iterative rjs.
Then one can use these trial rjs to compute 4>H using (9.2.6). These 4>Hs,
in turn, enable one to compute (9.2.9). This yields the second trial atomic
positions rjs. The iteration process continues in this manner until the resulted
sequence of positions converges.
The techniques which employ the probabilistic distribution of phases to
determine crystal structures are known as the direct method, since the phases
4>H are determined directly from the observed magnitudes JEH J [rather than
from a presumed known structure via (9.2.6)]. Thus, in the direct method the
first thing is to find a pdf for 4>H and to assign 4>H the expected value at the
starting step of the iteration.
9.2.5 Structure Invariants

In this subsection, we prepare ourselves with preliminaries to make a good guess
for 4>H to start the first iteration. The problem is how to choose a coordinate
origin so that the phases or combinations of several phases are only dependent
on the structure of the crystal and independent of the choice of the origin.
80
60
Figure 9.9: An approximation to the two dimensional periodic Dirac delta

function by a finite sum.
The vector rj depends on both the relative atom position and the choice
of origin of the frame (a, b, c). Hence the individual phases 4>H' a function of
the atomic position vectors rj via equation (9.2.6), also depend on the choice
of origin (as well, of course, as on the crystal structure). Although the relative
positions of atoms and the orientations of the cell edges are fixed for a given
crystal, the origin of the frame (a, b, c) can still vary. If we shift the origin
from 0 to 0' and the position of 0' relative to 0 is ro, then the phase for the
same crystal is changed from 4>H to 4>' H:
This creates another dilemma: If 4>H is the phase of the structure factor
E H , it should be independent of the choice of the origin as its strength IEHI
is . A frame dependent structure factor cannot be used to determine a crystal
structure.
Although a single phase is not an invariant when an origin moves arbitrarily,
the sum of a few phases, each for a different point in the reciprocal lattice, is
an invariant. The most important sums are the sums of three and four phases.
Because
these sums are origin independent when a nice condition is satisfied: the sum
of the three or four reciprocal lattice vectors vanishes, i.e.
H+K+L=O.
This is similar to the resonance condition in three- or four-wave interactions

discussed in the last section. For water waves, one can see the energy trans-
fer from a larger scale motion to a smaller scale motion clearly only when the
resonance condition is satisfied. It is the same here. Only when the invariant
condition is satisfied, the pdf of the phase becomes distinguishable and hence
can be calculated. Such an interesting connection between the wave-wave in-
teraction theory and the X-ray crystallography may inject new ideas to both
research fields.
We will demonstrate that the sums of three and four phases yield enough
information to determine the structure of a crystal.
The triplet
3 = H + K + L (9.2.10)
is an invariant when
H+K+L=O. (9.2.11)
The quartet
4 = H + K + L + M (9.2.12)
is an invariant when
H+K+L+M=O. (9.2.13)
In the above, 3 and 4 are called the structure invariants and equations
(9.2.11) and (9.2.13) are called the structure invariant conditions.
We know that a single phase is not an invariant. If a sum of two phases
is an invariant, then K + L = 0 implies that K = -L. So the two phases
correspond to the same set of parallel scattering planes. Hence, K and L
become the same phase (but with opposite signs).
One can generalize the above sums to linear combinations of five, six or more
phases. But, it seems that this generalization only fulfills one's curiosity and
has no practical value in detecting a crystal structure. A reason for this is that
the sums of three and four phases have already provided enough information
to determine a crystal structure.
When we talk about an invariant, we mean that the origin can be moved
arbitrarily. A quantity is an invariant only if it does not change with any
arbitrary shift of the origin. However, real crystals have certain symmetries,
such as the symmetry about the center, symmetries about a two-fold screw
axis and many others. These symmetries can be classified by space groups. It
is rather easy for a mathematics or physics student to learn this classification
theory and understand all the symmetries. One can find the relevant materials
in crystallography books or solid state physics books.
Because of the presence of these symmetries, the choice of origin is quite
restricted. For instance, in the case of the center symmetry, it is natural to
choose the center as the origin. The shift of the origin can only be from one
center to another. Even a single phase is invariant under this shift. When
considering the symmetry, only certain choices of origins are permissible. Those
linear combinations of phases which are invariant with respect to the shift
among the permissible origins are called the structure seminvariants.
The pdf of a structure seminvariant can also be calculated and hence is

likewise very useful in the direct methods.
9.2.6 Neighborhood principle

When IEH I is fixed, EH can vary on a circle when taking different phases.
If T is an invariant, the amplitudes of its related structure factor constitute
a neighborhood of the invariant. For example, the first neighborhood of the
triplet invariant tPs consists of
The first neighborhood of the quartet invariant tP4 consists of
The second neighborhood of the same quartet consists of the four amplitudes
above plus three other additional magnitudes
IEH+KI, IEK+LI, IEL+HI

Since L + M = - (H + K), L + M and H + K represent the same set of parallel
diffraction planes. Thus, IEL+M I is not included in the second neighborhood.
Hauptman established the neighborhood principle: the value of any structure
invariant T is primarily determined, in favorable cases, by the values of one or
more small sets of magnitudes lEI (the neighborhoods of T) and is relatively
insensitive to the values of the great bulk of the remaining magnitudes lEI.
With this principle, we can compute the values of triplets and quartets from
a few magnitudes IEHI, IEKI, IELI, IEMI. The results calculated from these
very few magnitudes are robust.
9.2.7 Probability distributions of structure invariants

The probability distribution was introduced into X-ray crystallography by A.
J. C. Wilson (1949). A few years later, the concept of the joint distribution of
two or more structure factors was introduced by H. Hauptman and J. Karle.
They demonstrated the importance of the probabilistic approach to the phase
problem. In this subsection, we will show how to get a good guess of the phases
which are necessary to start an iteration process in the direct method.
Let us consider the triplet invariant tPs. Suppose there are N atoms in
a unit cell. We denote the reciprocal lattice by W. Regard H, K, L as three
random variables each of which is uniformly distributed in the reciprocal lattice
W. Consider only those structure factors whose magnitudes are equal to one
of the three values Ri, R2, Rs > 0 and satisfy the invariant condition, i.e.
and
H+K+L= o.
Then the structure invariant <Pa = <PH + <PK + <PL is a function of the primitive
random variables H, K, L and therefore is itself a random variable. Its domain
is defined as (-11",11"]. Its conditional probability distribution in this domain is
(9.2.14)
which is apparently not uniform. Here,
Io(A) = E
00 1
(k!)2
(A)2k
2"
is the zeroth order modified Bessel function and
2ua
A= a/2 Rl R2Ra
0'2
with
N
Un = ~Zj, n = 1,2,3""
j=l
The distribution (9.2.14) is bell shaped and the shape is determined by the
value of A. The larger the value of A, the sharper the distribution curve
P1Ia(<pa) and the less the deviation of the invariant <Pa away from zero. Since
Un is proportional to N, the value of A is proportional to l/VN. Hence, the
estimation <Pa ~ 0 is less valid when N is very large (i.e. the molecules contain
a large number of atoms). When A = 10.000,2.316,0.731, the distributions are
plotted in Fig. 9.10. From the figure, we can see that when A is large, <Pa has
a large chance to be around zero:
<Pa ~ o. (9.2.15)
Here, " ~ " is a probability equality sign, i.e. the right hand side equals the
expected value of the left hand side. Large magnitudes imply a larger A, a
smaller variance in the distribution, and a more reliable estimate of <Pa, zero in
this case.
Our goal is to find <PH for each H to start the first iteration. How can
we make use of the triplet invariants: <Pa = <PH + <PK + <pL? If two phases
are known, then the third phase is equal to the negative of the sum of the
other two phases only in the probability sense according to (9.2.15). Is there
any better estimate for the third phase? A decisive step was taken by Karle
and Hauptman (1956). They derived what is now called the tangent formula
that can determine the new phase when some pairs of phases are known. This
determination is in the sense of equality not the probability equality, hence it
is a more precise estimation than
The tangent formula can be expressed in the following way. If ifoK and ifoL
are known (or preassigned), then the tangent formula
tan[ifo ] = _ L:K IEKIIELI sin[ifoK + ifoLl (9.2.16)

H L:K IEKIIELI cos[ifoK + ifoL]
determines the value of the third phase ifoH' In the above, L = H - K. The
summation is supposed to be carried out over the entire reciprocal lattice.
Because of the neighborhood principle, we can actually choose a small number
of points in the reciprocal lattice to compute this sum. The more points we
take, the more reliable is the estimate. Therefore, the tangent formula is usually
used to refine and extend a basis set of phases, presumed to be known. One can
use the result from the tangent formula as an induction and hence ifomKnL can
be found also by the tangent formula for any integers m and n. Therefore, the
phases corresponding to the points on the entire cross section in the reciprocal
lattice spanned by K and L can be computed by repeated use of this tangent
formula. This procedure may be used as an iterative step when computing the
phases.
How to find values of the phases for the lattice points which are not on the
plane spanned by K and L? At this point it should be mentioned that, since
the phases ifoH are not uniquely determined by the crystal structure but depend
also on the choice of origin, the method of fixing the origin in a direct methods
procedure is to specify arbitrarily the values of three phases ifoH 1 ' ifoH 2 , ifoH 2 '
where HI, H2 and H3 are three suitably chosen non-coplanar reciprocal lattice
vectors. With proper choices of HI, H2 and H 3, these vectors span the whole
of reciprocal space, not merely a plane in the reciprocal space. In this way, one
uses the triplets and the tangent formulas in different planes. Complementary
to this is to investigate another invariant in three dimensions: the quartet
invariant.
Now, let us consider the distribution of the quartet ifo4 in its domain (-7r, 7r].
In the first neighborhood, we choose those structure factors whose magnitudes
are equal to one of the four positive numbers R I , R 2 , R 3 , R4 and whose recip-
rocal vectors satisfy the invariant condition, i.e.
and
H+K+L+M=O.
Then the structure invariant ifo4 = ifoH + ifoK + ifoL + ifoM is a function of the
primitive random variables H, K, L, M and therefore is itself a random variable.
Its conditional probability distribution over the domain (-7r, 7r] is
(9.2.17)
where 10 is still the zeroth order modified Bessel function and

20"4
B= - 2 R I R 2 R 3 R4.
0"2
-150 150
Degrees
Figure 9.10: The probability distribution of the triplet invariant.
Hence the distribution of 4>4 is also bell shaped and has the same form as
that of 4>3 except that A is replaced by B. The most probable value for 4>4
is also zero: 4>4 ~ O. This estimate is more reliable when B is larger. But,
unfortunately this estimate is not so good as that for 4>3. The reason is that
B = O(l/N) < A = O(l/VN) for a large N. In summary, the distribution of
4>4 is bell shaped and the peak of the bell is at 4>4 = O. The estimate 4>4 ~ 0 is
not so reliable as 4>3 ~ O.
To obtain a more reliable estimate for the quartet invariant, we make use of
the second neighborhood. We impose more constraints on the structure factors
by assuming as known the seven constant magnitudes:
IEHI = Rl, IEKI = R2, IELI = R3, IEMI = R4,

IEH+KI = R12, IEK+LI = R23, IEL+HI = R31
The additional constraint makes the estimate of 4>4 more reliable and reveals a
possibility of favorable values other than zero.
The distribution formula is due to Hauptman (1975) and written as
Pll7 ~ p(4)4I R l, R2, R3, R4, R12, R23, R31)
exp (-2B' cos 4>4) 10 (:;~ R12X12 )

(9.2.18)
tt- '7 I 2
,,
R, -2.918 N_29
I
"
,
,
I;!= 1 2 1
~ -3 4;;
R, -2.863 8-,273
R, -2.27614>1 _96"
,
I
I \
\
\
!!-3! 4
H+K-6 i i'
R. =1.733 fru.
RI2 .'.631 Ifmod4!o-'OS-
I
I
\
\ i<+i-4 2 ~ R,,O.223
,,
I \ i+Y*4SS R11 -1.540
,,
I
I
I
/,/
I
I
\
\
,,
,, ,,
,,
.- /
---------- -..,,-------- cp
-180 -160 -140 -120 -100 -80 -60 -40 -20 20 40 60 80 100 120 140 160 180
o"gr""s
Figure 9.11: The pdf of the quartet invariant. [From Hauptman (1985), by
permission of John Wiley]
where
1 (2
B , = "3 20'3 - 0'20'4 ) R1R2R3R4, (9.2.19)
0'2
X 12 = (RiR~ + R~R~ + 2R1R2R3R4 cos <p4)1/2, (9.2.20)
X 23 = (R~R~ + R~Ri + 2R1R2R3R4 COS <P4)1/2, (9.2.21)
X 31 = (R~Ri + R~R~ + 2R1R2R3R4 cos <p4)1/2 (9.2.22)
where L is a normalization constant that forces the area under Pll7 to be unity.
Fig. 9.11 shows the probability distribution functions computed from the above
formulas P l l 4 and P l 17 ' The peak of the distribution of the quartet invariant
<P4 occurs at 105, which is relatively close to the true value 96 for this known
structure. In X-ray crystallography, a difference fo 20 is quite acceptable and
this 9 difference is considered to be a very good approximation.
The shape of P l l 7 is very sensitive to the values of R 12 , R23 and R 31 . If
R 12 , R23 and R31 are all large, then there is only one peak in P l l 7 which is at
zero. When the values of R 12 , R 23 , R31 decrease, there appear two peaks which
are symmetric with respect to zero. Finally, when R 12, R 23 and R31 are all
close to zero, then the pdf reduces to
P l l7 ~ ~ exp( -2B' cos <p4)'

So the peaks are now pushed to the end points and the most probable value
for <P4 is now 'Tr or -'Tr.
In any case, when three phases <PH, <PK and <PL are known, the fourth phase
<PM can be estimated from <PM ~ <P4 - <PH - <PK - <PL where <P4 takes the most
probable value. Hence, one can use this formula and the appropriate recipe for
origin fixing to estimate all the phases for all the points other than those on
the spanned plane described for the triplet invariants.
For a more reliable estimate of <PM' there is a modified tangent formula
that can be used. The formula is quite complicated and thus omitted here.
[1] O. M. Phillips (1974), Wave interactions, pp. 188 - 211, in "Nonlinear

Waves" ed. by S. Leibovich and A. R. Seebass, Cornell University Press,
Ithaca, New York.
[2] A. D. D. Craik (1985), Wave Interactions and Fluid Flows, Cambridge
University Press, New York.
[3] O. M. Phillips (1967), Theoretical and experimental studies of gravity wave
interactions, Proc. Roy. Soc. A 299, 104 - 119.
4] L. F. McGoldrick (1965), Resonant interactions among capillary-gravity
waves, J. Fluid Mech. 21,305-331.
[5] H. Hauptman (1985), Phase problems of X-ray crystallography, in "Ency-
clopedia of Statistical Sciences", Vol. 6, ed. by Kotz and Johnson, John
Wiley & Sons, New York, pp. 702-709.
[6] H. Hauptman (1983), The phase problem of x-ray crystallography, Proc.
Indian Acad. Sci. (Chem. Sci.) 92, 291-321.
[7] H. Hauptman (1975), A joint probability distribution of seven structure
factors, Acta Cryst. A 31, 671-679.
[8] H. Hauptman (1975), A new method in the probabilistic theory of the
structure invariants, Acta Cryst. A 31, 680-687.
[9] M. M. Woolfson (1987), Direct methods - from birth to maturity, Acta
Cryst. A 43, 593-612.
[10] G. Bricogne (1984), Maximum entropy and the foundations of direct
methods, Acta Cryst. A 40, 410-445.
[11] A. McPherson (1982), Preparation and Analysis of Protein Crystals, John
Wiley & Sons, New York, Chapters 5, 7, and 10.
[12] M. Sachs (1963), Solid State Theory, McGraw-Hill, New York, Chapters
2 and 4.
Appendix A
KdV Solitons via Inverse

Scattering
by G. E. Sarty
The inverse scattering method for finding soliton solutions of the KdV equa-
tion as presented in Section 4.2 can be automated with the symbolic manipula-
tion program Mathematica (and probably by other such programs as well). At
the end of this appendix is a Mathematica routine that will generate n-soliton
solutions for the KdV equation
Ut - 6UUx + Uxxx = 0
subject to the initial condition of
U(x, t = 0) = -n(n + 1)sech2(x).

To use the program (assuming you have purchased Mathematica), type the
code as given into a file called "invskat.m". Then start up Mathematica and
type invskat .m. This loads the soliton generation code. Then to generate,
say, a 2-soliton solution type soli ton [2] ; or soli ton [3] for a 3-soliton solu-
tion, etc. The solution is then returned as the function u[x_. tJ. The code
277
278 Appendix A. KdV Solitons via Inverse Scattering
has been tested for up to 6 solitons and no attempt has been made to simplify
the answer to any extentj it is given in terms of exponential functions. The
program of Appendix B simplifies answers in terms of hyperbolic trigonometric
functions which gives a neater looking final result.
The logic of the program follows that given in Chapter 4 and comments
in the code explain the steps followed. We give a brief elaboration here of
the steps. First we must use the result that the eigenfunctions of the linear
Schrodinger equation at t =0 for the initial data under our consideration are
e
the associated Legendre functions in the variable = tanh(z). Symbolically,
we write the associated Legendre functions as p;:'(e) where n is the number
of solitons sought and 1 ~ m ~ n. (In the code, LegendreP[n.m.Tanh[x]] is
P;:'(tanh(z)).) It is known that
1 1 [pm(e)]2 ~ =
-1 n 1- e2
.!. (n + m)!
m (n - m)!
and this is used to normalize the eigenfunction <Pm(z) = P;:'(tanh(z)) to

<Pm(z) = (~~-:)t <pm(z) (which is Phi [m.x] in the code).
Next, Cm (0) is calculated as Co [m] in the program. Since Mathematica has
trouble doing infinite limits for complicated expressions involving hyperbolic
trigonometric functions, some substitutions are done that mimic the calculation
as it would be done manually. With Cm(O) in hand, Cm(t), written as Cm[m. t]
in the code, can be easily calculated by a formula in Theorem 4.2.
Now the kernel function B(e, t), written as B[y_. tJ, can be determined
using the fact that R(k, t) = O. The program is now ready to solve the Gelfand-
Levitan equation for K(z, Yj t) using the method of separation of variables.
That is, we assume that
n
K(z,Yjt) = L Lm(z,t)e- k ",11
m=l
where K(z,Yjt) is K[x_.y_.tJ and Lm(z,t) is L[m] at this point in the pro-
gram. Note that km = m in our case. The expression for K(z, Yj t) is sub-
stituted into the Gelfand-Levitan equation ( GL in the code). Terms involving
e- k ",11 for 1 ~ m ~ n are separated out. That being done, the coefficient
functions Lm(z,t) and hence K(z,Yjt) can be solved for algebraically.
Finally, the solution is calculated as
{)
U(z, t) = -2 {)z K(z, Zj t).
The symbolic program written in Mathematica is given below:

279
..
( ...................................................................... )
(.
(. IIVSKAT." -- A "athematica soliton generator. This
.)..
)
. .
(.. package contains subroutines that generate n-soliton solutions )
(.. ~or the KdV equation: )
(. .)
. .
(. U - 6 UU + U = 0 )
(. t x xxx )
(. .)
(. subject to the initial condition )
. .
(. 2 )
(. U(x,t=O) = -n(n+1) sech [x]. .)
(. .)
(. The inverse scattering method and the Gel~and-Levitan )
(. equation is used. .)
..........................................................................
(. Programmed by: )
(. G. Sarty, July 1992 )
(. .)
( )
invskat: : usage = "The package invskat.m returns n soli ton solutions to the
standard KdV equation subject to the initial condition
U(x,t=O) = -n(n+1) sech[x]2 using the inverse scattering method. For more
in~ormation, type '?soliton."
soliton: :usage = "soliton[n] will return an n soliton solution to the

standard KdV equation where n is an integer."
sOliton[n_Integer] := Block[{m,Phi,Co,Cm,B,K,L,GL,y,eqn,k},
I~[ n <= 0, Return["Use a positive integer."] ];
(. define normalized eigen~unctions: .)
Do[
Phi[m,x] = (-1)m Sqrt[ m(n-m)!/(n+m)! ] LegendreP[n,m,Tanh[x]];
Phi[m,x] = Factor[Phi[m,x]];
Phi[m,x] = Phi[m,x] II. {(1 - u_) (1 + u_) -> (1 - u2),
(-1 + u_) (1 + u_) -> - (1 - u2),
(1 - u_)k_ (1 + u_)k_ -> (1 - u2)k,
(-1 + u_)k_ (1 + u_)k_ -> - (1 - u2)k};
Phi[m,x] Phi[m,x] II. {(1 - Tanh[x_]2) -> Sech[x]2,
(-1 + Tanh[x_]2) -> -Sech[x]2,
(1 - Tanh[x_]2)(k_/2) -> Sech[x]k,
(-1 + Tanh[x_]2)(k_/2) -> -Sech[x]k,
(1 - Tanh[x_]2)k_ -> Sech[x](2 k),
(-1 + Tanh[x_]2)k_ -> -Sech[x](2 k)};
Phi[m,x] = PowerExpand[Phi[m,x]],
{m,1,n,1}] ;
(. de~ine Cm(O) .)
Do[
Co[m] Phi[m,x] E(m x);
Co[m] Co[m] II. Sech[x] -> 2/(Ex + E(-x;
Co[m] Co[m] II. Tanh[x] -> 1;
Co[m] = ExpandAII[Co[m]];
280 Appendix A. KdV Solitons via Inverse Scattering
top = lumerator[Co[m]]/E-(m x);

bot ExpandAll[Denominator[Co[.]]/E-(m x)];
Co[m] = top/bot;
Co[m] Limit[ Co[m] , x -> Ininity],
{m,l,n,H] ;
Do[
Cm[m,t] = Co[m] E-(4 m-3 t),
{m,l,n,H] ;
(. deine B(y,t) .)
Do[
B[y_,t_] := Sum[Cm[m,t]-2 E-(-m y), {m,l,n}],
{m,l,n,H] ;
(. using separation of variables, define l(x,y;t) .)
(. substitute into the Gelfand-Levitan equation .)

(. first solving the integral of B[y+z,t] l[x,z,t] from x to ininity, .)
(. vrt z: .)
inint = ExpandA11[B[y+z,t] l[x,z,t]] II. E-(b_ + c_ z) -> -E-(b + c x) / c;

GL = l[x,y,t] + B[x+y,t] + infint;
(. pullout E-(-km y) terms: .)
GL = ExpandA11[GL];
GL = ExpandAll[E-n+l) y) GL];
Do[
GL = GL II. {E-(h_ + m y) -> y[m] E-h, E-(m y) -> y[m]},
{m,l,n,H] ;
Do[
eqn[m] = GL;
Do[
I[k == m,eqn[m]=eqn[m] II. y[k] -> l,eqn[m]=eqn[m] II. y[k] -> 0],
{k,l,n,H] ,
{m,l,n ,H];
(. solve for L[m] .)
system = {eqn[l] == O};

vars = {L[1]};
Do[
system = Append [system , eqn[m] == 0];
vars = Append[vars, L[m]] ,
{m,2,n,l}] ;
soln = Solve[system,vars];
(. use L[m] to reconstruct l[x,x,t] .)
Do[
l[m] = Together[Expand[Part[Part[Part[soln,l],m],2]]],
{m,l,n,H];
281
k[x_,t_] := Together[Expand[Sum[l[mJ E-(-m x), {m,1,n}]]];
(. finally, calculate the solution, u, of the initial value problem .)
Return[u[x_,t_] = Together[Expand[-2 D[k[x,t] ,x]]]]

]
Appendix B
KdV Solitons via Backlund

Transform
by G. E. Sarty
B.1 Backlund Transform Program

Biicklund transform and the nonlinear superposition principle were described
in section 4.4. In this subsection, we describe how to use a symbolic computer
code to implement the nonlinear superposition principle automatically to find
multiple soliton solutions of the initial value problems for the KdV equation.
The code is written in Mathematica and given in the later part of this subsec-
tion. Specifically, the code has been designed to calculate soliton solutions to
= =
the KdV equation using the initial condition U(z, t 0) -n(n+ 1) sech 2 (z).
The results of 2-,3-, 4-, 5-, 6-, and 7-soltion solutions computed from this code
are presented in sections B.2, B.3, B.4, B.5, B.6 and B.7 respectively.
The computer program essentially has two parts. One part is a set of
routines that enables specific hyperbolic trigonometric function simplifications
to be done. This set of routines is called "hyppac.m" . The other part calculates
283
284 Appendix B. KdV Solitons via Backlund Transform
the soliton solutions. These two parts have been put into a package called
"solipac.m". It can be used by copying the code into a file called "solipac.m"
and entering solipac.m at the Mathematica prompt. Typing soliton[N]
will cause the program to generate an N -soliton solution. One may like to test
the code by typing in soliton[l] or soliton[2] to generate 1- and 2-soliton
solutions. These calculations take only a few seconds on any workstation.
If you desire only to calculate an expression for the soliton ladder, type
ladder [N] . After the ladder calculation is completed, if one want to see the cor-
responding soltion, simply type finish. For instance, the command ladder [2]
generates a 2-soliton ladder. Then the command finish will generate a 2-
soliton solution. The program has been tested up to 7 solitons (the 7-soliton
solution took about a week's worth of CPU time on a SUN Sparcstation 1+
(1991)).
The hyperbolic function routines are pretty much self-explanatory with the
locally defined hyperbolic functions being designated with small first letters
(e.g. sinh, cosh) to avoid confusion with Mathematica's definitions (Sinh, Cosh,
etc.). The soliton calculation itself proceeds by calculation of the the func-
tions Wen) via the nonlinear superposition principle derived using the Backlund
transform in section 4.4. We use notations: (i) (n) == 12 [n - l]k ( a per-
mutation of 1,2,3,, n - 1 and k) where k is a number greater than n - 1;
(ii) (n - 1)' == 12 .. [n - 2]k (a permutation of 1, 2, 3, ... , n - 2 and k and still
for k> [n - 1]). We have the nonlinear superposition principle as:
4(An - i - An)
Wen) = W(n-2) - W(n-i) - W(n-i)'
, (B.1.1)
where Am == _m 2 Note that equation (B.1.1) is valid for each value of k

such that [n - 1] > k ~ N where N is the number of solitons sought. These
equations show the relationship between the various Wen). This relationship
is shown schematically by the connecting arrows in Figure B.1 which is an
example for N = 5. The tricky part of the program is to reindex the Wen)
for any N universal way so that a one-dimensional array w[n] can be used to
define the functions. In our program, we number the Wen) in the following way:
Wi = w[l], W2 = w[2], W3 = w[3], W4 = w[4], W5 = w[5],

W12 = w[6], W13 = w[7], W14 = w[8], W15 = w[9],
W123 = w[10], W124 = w[ll], W125 = w[12],
W1234 = w[13], W1235 = w[14],
W12345 = w[15]
for the case of the 5-soliton solution.

The functions in the first column of Figure B.1 are defined by
Wn(X, t) = -2k tanh(kx - 4k 3t), when k is odd,
wn(X, t) = -2k coth(kx - 4k 3t), when k is even.

B.l. Backlund Transform Program 285
W;:----
N345
"
" "w:. / " "w

1'25 3
",, ,
, ",
'w
4
"w
5
Figure B.l: Diagram of the nonlinear superposition principle for 5 solitons.
The N-soliton solution is then found by differentiating W12 ... N (or w[(N-2 N)/2]+)
with respect to x.
That is enough about the ideas used in developing the computer code. The
whole package of code solipac.m is given below:
286 Appendix B. KdV Solitons via Biicklund Transform
(
(.
(.
.
........................................................................ )
SOLIPAC." -- A "athematica soliton generator. This
.
.)
)
(. * package contains subroutines that generate n-soliton solutions * *)
(* * or the IdV equation: * *)
(* * *)
(* * U - 6 UU + U 0 * *)
(* * t x xxx * *)
(* * * *)
(* * subject to the initial condition * *)
(* * 2 * .)
(* * U(x,t&O) -n(n+1) sech [x]. * *)
(* * * *)
(* * The package HYPPAC." is included (and described below) * *)
(* * along with the routines soliton, ladder and inish. * *)
(* * * *)
(* Programmed by: G. Sarty, Jan. 26, 1993 * *)
(* * * *)
(. ** ***.****.****.***** ********.**********.************************ *)
solipac::usage = "Solipac.m contains routines or calculating n - soliton
solutions or the IdV equation. For more inormation type ?hyppac,
?soliton, ?ladder and ?inish. The commands soliton[n], ladder[n] and
inish use the routines in hyppac.m to simpliy expressions containing the
hyperbolic trig unctions cosh and sinh."
(* ********************************************************************** *)
(* * * .)
(* * HYPPAC." -- A collection o hyperbolic trig unction * .)
(* * substitution routines or "athematica. The routines are * *)
(. * called: * *)
(* * 1) hypdif * *)
(* * 2) hypsubs * *)
(* * 3) hypac * *)
(* * 4) hypconv * *)
(* * 5) hypmath * *)
(* * 6) hypunmath. * *)
(* * * *)
(* ********************************************************************** .)
hyppac: : usage .. "The ile hyppac. m is included in the ile solipac. m and
contains routines or manipulating hyperbolic trig unctions. The routines
included are hypdi, hypsubs, hypac, hypconv, hypmath and hypunmath.
(Type ?hypdi, ?hypsubs, ?hypac, ?hypconv, ?hypmath or ?hypunmath or more
ino on these routines.) To use hyppac, use small letters to denote the trig
unctions. That is, denote the hyperbolic cosine by cosh instead o Cosh
and denote the hyperbolic sine by sinh instead o Sinh. This will prevent
"athematica rom converting everything into exponential notation. The
routines hypmath and hypunmath will convert between the two types."
(. DIFFEREITIATIOI OF HYPERBOLIC TRIG FUICTIOIS *)
hypdi: : usage = "The routine hypdi lets "athematica know what the
derivatives o the hyperbolic trig unctions sinh and cosh are. The
deinitions were made when the ile hyppac.m was read in."
sinh'[x_] := cosh[x]
cosh'[x_] := sinh[x]
(* HYPERBOLIC TRIG SUBSTITUTIOI RULES *)
hypsubs: :usage .. "The routine hypsubs, applied thus:
expr II. hypsubs , does hyperbolic trig unction product to sum
conversions in the expression expr. This command was written primarily or
use by the hypconv command."
B.l. Backlund Transform Program 287
hypsubs = { sinh[u_] cosh[v_] -> (1/2) sinh[u + v] + (1/2) sinh[u - v],

cosh[u_] sinh[v_] -> (1/2) sinh[u + v] - (1/2) sinh[u - v],
cosh[u_] cosh[v_] -> (1/2) cosh[u + v] + (1/2) cosh[u - v],
sinh[u_] sinh[v_] -> (1/2) cosh[u + v] - (1/2) cosh[u - v] ,
sinh[u_]-(n_Integer) -> sinh[u]-(n-2) (1/2) (cosh[2 u] - 1),
cosh[u_]-(n_Integer) -> cosh[u]-(n-2) (1/2) (cosh[2 u] + 1),
sinh[-u_ v_.] -> -sinh[u v],
cosh[-u_ v_.] -> cosh[u v],
sinh[u_Real v_.] -> If[ u < 0, -sinh[-u v], sinh[u v] ],
cosh[u_Real v_.] -> If[ u < 0, cosh[-u v], cosh[u v] ],
sinh[u_Integer v_.] -> If[ u < 0, -sinh[-u v], sinh[u v] ],
cosh[u_Integer v_.] -> If[ u < 0, cosh[-u v], cosh[u v] ] }
(. FACTORIIG THE ARGUMEIT .)
hypfac::usage = "The routine hypfac, applied as expr //. hypfac ,
to an expression expr factors out common factors in the arguments of
hyperbolic trig functions (in particular, -1 may be factored out and thus
made visible to the hypsubs routine). This command vas vritten primarily
for use by the hypconv command. "
hypfac = { sinh[u_] :> sinh [Factor [u]],
cosh[u_] :> cosh[Factor[u]] }
(. HYPERBOLIC TRIG COIVERSIOI ROUTIIE .)
hypconv: : usage = "The routine hypconv, implemented on an expression expr
thus: hypconv[expr] , uses the routine hypsubs to COMPLETELY
convert all products of hyperbolic trig functions in expr into sums of
hyperbolic trig functions. It uses the commands hypfac and hypsubs in a
loop."
hypconv[f_] := Block[ {nevf,oldf,j}, j = 1; oldf = f;
While [ j == 1,
nevf oldf II. hypsubs;
nevf = Expand[nevf];
nevf = nevf //. hypfac;
If[ SameQ[nevf,oldf], j = 0, j 1];
oldf = nevf]; Return[nevf] ]
(. COlVERT TO MATHEMATICA FORM .)
hypmath: :usage = "The routine hypmath, implemented on an expression expr
thus: hypmath[expr] , converts cosh to Cosh and sinh to Sinh."
hypmath[f_] := f //. {cosh[u_] -> Cosh[u], sinh[u_] -> Sinh[u]}
(. COlVERT TO HYPPAC ROUTIIE FORM .)
hypunmath: : usage = "The routine hypunmath, implemented on an expression expr
thus: hypunmath[expr] , converts Cosh to cosh and Sinh to sinh."
hypunmath[f_] := f //. {Cosh[u_] -> cosh[u], Sinh[u_] -> sinh[u]} ;
(. EID OF HYPPAC . . )
(. COMPLETE SOLITOI LADDER AID MULTIPLE SOLITOI CALCULATIOI .)
soli ton: : usage = " soli ton [n] , vhere n is an integer, viII give
an n-soliton solution for the KdV equation Ut - 6 U Ux + Uxxx = 0 subject to
the initial condition U(x,t=O) = -n(n+1) sech-2[x]. The solution is
returned as U and the associated soliton ladder is returned as W. lormally,
U is printed out upon completion of this command. To execute this same
procedure in tvo steps, use the ladder and finish commands. For more
information type ?ladder and ?finish."
soliton[n_Integer] := Block[{1ambda,v,s},
If[ n <= 0, Return["Use a positive integer."] ];
Array [lambda ,n] ;
Array [v, (n-2 + n) /2] ;
Array[s,n];
Do[ lambda[k] -(k-2), {k,n} ];
Do[ v[k] -2 k sinh[k x - 4 k-3 t] / cosh[k x - 4 k-3 t], {k,1,n,2}];
Do[ v[k] = -2 k cosh[k x - 4 k-3 t] / sinh[k x - 4 k-3 t], {k,2,n,2}];
s [1] = 1;
Do[ s[l] = s[l-l] + n - (1-2), {1,2,n,1}];
j = 2;
Do[ v[s[j]+i] = -(4 (lambda[j-1]-lambda[j+i]/(v[s[j-1]]-v[s[j-1]+i+1]),
{i,O,(n-j),l} ];
Do[ Do[ v[s[l]+i] = v[s[1-2]]-4 (lambda[l-l]-lambda[l+i]/(v[s[l-l]]
v[s[l-l]+i+1], {i,O,(n-l),l}], {1,3,n,l}];
W = Together[v[(n-2 + n)/2]] ;
top lumerator[W];
top = hypconv[top];
top hypmath[top];
top Simplify[top];
bot Denominator[W];
bot hypconv[bot];
bot hypmath[bot];
bot Simplify[bot];
W = Expandlumerator[Together[top/bot]];
U = D[W,x];
first = Take[U,l];
second = U - first;
first = first Denominator[W]~2;
second = second Denominator[W]~2;
first = Expand[first];
second = Expand[second];
topu = first + second;
botu = Denominator[W]~2;
topu = hypconv[topu];
topu = hypmath[topu];
topu = Simplify[topu];
U = Expandlumerator[topu/botu] ]
(. SOLITOI LADDER OILY CALCULATIOI .)
ladder: : usage = " , vhere n is an integer, viII give

ladder[n]
an n-soliton ladder for the KdV equation Ut - 6 U Ux + Uxxx = subject to
the initial condition U(x,t=O) = -n(n+1) sech~2[x]. The solution is
returned as W. lormally, W is printed out upon completion of this command."
ladder[n_Integer] := Block[{lambda,v,s},
If[ n <= 0, Return["Use a positive integer."] ];
Array[lambda,n];
Array[v,(n~2 + n)/2];
Array[s,n];
Do[ lambda[k] -(k~2), {k,n} ];
Do[ v[k] = -2 k sinh[k x - 4 k~3 t] / cosh[k x - 4 k~3 t], {k,1,n,2}];
Do[ v[k] = -2 k cosh[k x - 4 k~3 t] / sinh[k x - 4 k~3 t], {k,2,n,2}];
s [1] = 1;
Do[ s[l] = s[l-l] + n - (1-2), {1,2,n,1} ];
j = 2;
Do[ v[s[j]+i] = -(4 (lambda[j-l]-lambda[j+i]/(v[s[j-1]] - v[s[j-l]+i+l]),
{i,O,(n-j),l} ];
Do[ Do[ v[s[l]+i] = v[s[1-2]]-4 (lambda[l-l]-lambda[l+i]/(v[s[l-l]] -
v[s[1-1]+i+1], {i,O,(n-l),l}], {l,3,n,l}];
W = Together[v[(n~2 + n)/2]];
top lumerator[W];
top hypconv[top];
top hypmath[top];
top Simplify[top];
bot Denominator[W];
bot hypconv[bot];
bot hypmath[bot];
B.2. Two Solitons 289
bot =
Simplify[bot];
=
W Expandlumerator[Together[top/bot]] ]
(. CALCULATE MULTIPLE SOLITOIS FROM THE LADDER .)
finish: :usage = ,, finish . Typing this command after a
ladder[n] command has been executed viII return the multiple soliton
solution, U, associated to the previously generated ladder, W. lormally, U
is printed out upon completion of this command."
finish := ( U = D[W,x];
first = Take[U,l];
second = U - first;
first =
first Denominator[W]-2;
second = second Denominator[W]-2;
first = Expand[first];
second =
Expand[second];
topu =
first + second;
botu = Denominator[W]-2;
topu =
hypconv[topu];
topu = hypmath[topu];
topu = Simplify[topu];
U = Expandlumerator[topu/botu]
The results of 1-, 2-, ... , and 7-soliton solutions are presented in the follow-
ing sections
B.2 Two Solitons

The analytical expression for the two soliton ladder is:
6sinh[3(12t - x)] + 6sinh[28t - x]
W12 = cosh[3(12t - x)] + 3 cosh [28t - x]
The analytical expression for the two soliton solution is:
-36 - 48 cosh [2 (4t - x)] - 12 cosh[4(16t - x)]
=
[ cosh[3(12t - x)] + 3 cosh [28t - x] F
U12
B.3 Three Solitons

The analytical expression for the three soliton ladder is:
12 sinh[6(24t - x)] + 48 sinh[4(34t - x)] + 60 sinh[2( 40t - x)]
W123 = 10 cosh[72t] + cosh[6(24t - x)] + 6 cosh[4(34t - x)] + 15 cosh[2( 40t - x)]
The analytical expression for the three soliton solution is:
U123 = [-3024 -1200cosh[2(4t - x)]- 360cosh[6(12t - x)]-
1920cosh[4(16t - x)]- 3240cosh[2(28t - x)]- 240cosh[8(28t - x)]-
24cosh[10(28t - x)]- 720cosh[6(36t - x)]- 960cosh[4(52t - x)]-
600 cosh[2(76t - x)] ]
[ 10 cosh [72t] + cosh[6(24t - x)] + 6 cosh[4(34t - x )]+

15cosh[2(40t - x)] F
290 Appendix B. KdV Solitons via. Backlund Transform
B.4 Four Solitons

The analytical expression for the four soliton ladder is:
W1234 = 4 [ 5 sinh[10( 40t - x)] + 140 sinh[4( 46t - x)] + 40 sinh[8( 49t - x )]+
135sinh[6(56t - x)] + 100sinh[4(82t - x)] + 175sinh[2(88t - x)]-
35sinh[2(56t + x)] ]
[ 126cosh[120t] + cosh[10(40t - x)] + 70cosh[4(46t - x)]+

10 cosh[8( 49t - x)] + 45 cosh[6(56t - x)] + 50 cosh[4(82t - x )]+
175cosh[2(88t - x)] + 35cosh[2(56t + x)] ]
The analytical expression for the four soliton solution is:
U1234 = [-650400 - 224000 cosh[144t] - 98000 cosh[288t]-

2520 cosh[2(292t - 7x)] - 63000 cosh[2(284t - 5x )]-
9000 cosh [2 (332t - 5x)] - 686880cosh[2(4t - x)]-
19600 cosh[6(12t - x)] - 224640 cosh[4(16t - x )]-
2240cosh[12(24t - x)] - 176400cosh[2(28t - x)]-
50400 cosh [8 (28t - x)] - 25200cosh[10(28t - x)]-
80640 cosh [8 (34t - x)] - 399840 cosh [6 (36t - x )]-
504000cosh[4(40t - x)]- 40cosh[18(44t - x)]-
720cosh[16(46t - x)]- 22400cosh[12(48t - x)]-
201600cosh[4(52t - x)] - 25200cosh[10(52t - x)]-
3600cosh[14(52t - x)] - 49000cosh[6(60t - x)]-
8000cosh[12(60t - x)] - 211680cosh[8(64t - x)]-
630000 cosh[2(76t - x)] - 141120 cosh[4(76t - x )]-
204120 cosh[6(76t - x)] - 70000 cosh[6(84t - x )]-
201600 cosh[4(112t - x)] - 176400 cosh[2(148t - x)]-
63000cosh[2(220t - x)] -17640cosh[2(116t + x)] ]
[ 126cosh[120t] + cosh[1O(40t - x)] + 70cosh[4{46t - x)]+

10 cosh [8(49t - x)] + 45cosh[6(56t - x)] + 50cosh[4{82t - x)]+
175 cosh[2(88t - x)] + 35 cosh[2(56t + x)] ]2
B.5. Five Solitons 291
B.5 Five Solitons

The analytical expression for the five soliton ladder is:
W12345 = [390sinh[892t - 13x] + 4410sinh[388t - 7x]+

14700sinh[676t -7x] + 30sinh[15(60t - x)]+
1701Osinh[5(76t - x)] + 5040sinh[9(76t - x)]+
2310 sinh[11 (76t - x)] + 3150sinh[9(92t - x)]+
22050sinh[3(108t - x)] + 11760sinh[5(124t - x)]+
3600sinh[172t - x] + 2940sinh[3(204t - x)]+
4410 sinh[316t - x] - 1260 sinh[5(20t + x )]-
5040 sinh[3 (36t + x)] - 4860 sinh[164t + x] ]
[ 15cosh[892t - 13x] + 315cosh[388t - 7x] + 1050cosh[676t - 7x]+

cosh[15(60t - x)] + 1701 cosh [5 (76t - x)] + 280 cosh[9(76t - x )]+
105 cosh[11 (76t - x)] + 175cosh[9(92t - x)] + 3675cosh[3(108t - x)]+
1176cosh[5(124t - x)] + 1800cosh[172t - x] + 490cosh[3(204t - x)]+
2205cosh[316t - x] + 126cosh[5(20t + x)] + 840cosh[3(36t + x)]+
2430 cosh[164t + x] ]
The analytical expression for the five soliton solution is:
U12345 = [-640926000 - 31752000 cosh[144t]- 200037600 cosh[240t]-

129654000 cosh[288t] - 17496000 cosh[336t]-
111132000 cosh[432t] - 42865200 cosh[480t]-
29635200 cosh[720t] - 1680 cosh[2(868t - 13x )]-
20160cosh[2(644t - 11x)] -134400cosh[2(788t -11x)]-
42000cosh[2(860t - 11x)] - 1128960cosh[2(292t - 7x)]-
24696000cosh[2(580t - 7x)] - 9525600cosh[2(604t -7x)]-
5268480cosh[2(652t -7x)]- 6585600cosh[2(724t -7x)]-
28224000 cosh[2(284t - 5x)] - 54432000 cosh[2(332t - 5x )]-
18522000 cosh[2(356t - 5x)] - 567000 cosh[4(392t - 5x )]-
73500 cosh [4 (416t - 5x )]- 32256000 cosh[2( 428t - 5x )]-
24696000cosh[2(572t - 5x)]- 8232000 cosh [2 (644t - 5x)]-
1632960cosh[6(212t - 3x)]- 333090240cosh[2(4t - x)]-
67939200cosh[6(12t - x)]- 385188720cosh[4(16t - x)]-
317520cosh[12(24t - x)] - 611452800cosh[2(28t - x)]-
40000800cosh[8(28t - x)]- 6585600cosh[10(28t - x)]-
11430720cosh[8(34t - x)] -157288320cosh[6(36t - x)]-
71442000cosh[4(40t - x)]- 12600cosh[20(40t - x)]-
241920 cosh[18 (44t - x)]- 61740000cosh[8(46t - x)]-
952560 cosh[16( 46t - x )]- 16669800 cosh[12(48t - x )]-
1260000 cosh[16(49t - x)] - 160003200 cosh[4(52t - x )]-
77565600 cosh[10(52t - x)]- 11707200 cosh[14(52t - x)]-
25515000cosh[12(56t - x)]- 296352000cosh[6(60t - x)]-
10584000cosh[12(60t - x)]- 269090640cosh[8(64t - x)]-
2143260cosh[12(64t - x)]- 340200cosh[20(64t - x)]-
60cosh[28(64t - x)] - 10080cosh[24(66t - x)]-

705600cosh[16(67t - x)] - 1058400cosh[18(68t - x)]-
20412000 cosh[8(70t - x)] - 12600 cosh[24(72t - x )]-
164640000cosh[2(76t - x)] - 186701760cosh[4(76t - x)]-
91445760cosh[6(76t - x)] - 12130560cosh[14(76t - x)]-
12594960 cosh[16(76t - x)] - 235200 cosh[20(76t - x )]-
31500000 cosh[8(82t - x)] - 215460000 cosh [6 (84t - x )]-
57697920 cosh[12(84t - x)] - 2963520 cosh[18(84t - x )]-
1176000 cosh[16(85t - x)] - 385875000 cosh[4(88t - x )]-
50009400 cosh[8(88t - x)] - 14288400 cosh[12(88t - x )]-
4445280 cosh[16(9lt - x)] - 97977600 cosh[6(92t - x )]-
1470000cosh[16(94t - x)] - 46305000 cosh[12 (96t - x)]-
163900800cosh[10(100t - x)] - 136080000cosh[8(106t - x)]-
9878400 cosh[12(108t - x)] - 150028200 cosh[4(112t - x )]-
60011280cosh[6(116t - x)] - 34574400cosh[8(118t - x)]-
3087000cosh[12(120t - x)] - 52920000cosh[4(124t - x)]-
129820320cosh[8(124t - x)] - 67737600cosh[6(132t - x)]-
148803480cosh[4(136t - x)] -163296000cosh[6(140t - x)]-
573652800 cosh[2(148t - x)] - 2304960 cosh[8(154t - x )]-
66044160 cosh[6(156t - x)] - 32413500 cosh[4(160t - x )]-
201519360cosh[4(196t - x)] - 28224000cosh[2(220t - x)]-
8643600 cosh [4 (232t - x)] - 240045120cosh[2(244t - x)]-
63221760 cosh[2(364t - x)] - 38102400 cosh[2(388t - x)]-
423360 cosh[8(26t + x)] - 4898880 cosh[6( 44t + x )]-
15435000cosh[4(56t + x)] - 16329600cosh[4(68t + x)]-
10001880 cosh[4(104t + x)] - 106686720 cosh[2(116t + x )]-
24192000 cosh[2(140t + x)] - 59270400 cosh[2(212t + x )]-
3951360 cosh[2(356t + x)] ]
[ 15cosh[892t - 13x] + 315cosh[388t - 7x] + 1050cosh[676t - 7x]+

cosh[15(60t - x)] + 1701 cosh[5(76t - x)] + 280 cosh[9(76t - x)]+
105 cosh [11 (76t - x)] + 175cosh[9(92t - x)] + 3675cosh[3(108t - x)]+
1176cosh[5(124t - x)] + 1800cosh[172t - x] + 490cosh[3(204t - x)]+
2205 cosh[316t - x] + 126 cosh[5(20t + x)] + 840 cosh[3(36t + x )]+
2430 cosh[164t + x] F
B.6. Six Solitons 293
B.6 Six Solitons

The analytical solution for the six soliton ladder is:
W123456 = [798 sinh[1756t - 19x] + 40950 sinh[1252t - 13x]+

114660 sinh[1540t - 13x] + 30492 sinh[764t - llx]+
261954 sinh[1244t - llx] + 155232 sinh[1484t - llx]+
432180sinh[1180t -7x] + 25200sinh[3{516t - 5x)]+
727650sinh[548t - 5x] + 14700sinh[3{564t - 5x)]+
554400 sinh[692t - 5x] + 510300 sinh[1028t - 5x]+
69300sinh[7{4t - x)] + 8316sinh[9{4t - x)]+
161700 sinh[3{84t - x)] + 232848 sinh[9{84t - x )]+
42 sinh[21 {84t - x)] + 748440sinh[7{100t - x)]+
7140sinh[17{100t - x)] + 661500sinh[9{132t - x)]+
378000sinh[7{148t - x)] + 74088sinh[9{164t - x)]+
1018710sinh[3{180t - x)] + 181104sinh[244t - x]+
151200 sinh[3{324t -x)] + 224532 sinh[484t - x]+
10584sinh[964t - x] - 173250sinh[3{12t + x)]-
258720 sinh[3{60t + x)] - 231000 sinh[188t + x]-
58212 sinh[476t + x] - 242550 sinh[28t + 5x))
[ 21 cosh[1756t - 19x] + 1575 cosh[1252t - 13x]+

441Ocosh[1540t -13x] + 1386cosh[764t - llx]+
11907 cosh[1244t - llx] + 7056 cosh[1484t - llx]+
30870cosh[1180t - 7x] + 840cosh[3{516t - 5x)]+
72765 cosh[548t - 5x] + 490 cosh[3{564t - 5x)]+
55440cosh[692t - 5x] + 51030cosh[1028t - 5x]+
4950 cosh [7{4t - x)] + 462cosh[9{4t - x)]+
26950 cosh [3 {84t - x)] + 12936 cosh[9{84t - x )]+
cosh[21{84t - x)] + 53460cosh[7{100t - x)]+
210 cosh[17{100t - x)] + 36750 cosh[9{132t - x )]+
27000 cosh[7{148t - x)] + 4116 cosh[9{164t - x)]+
169785cosh[3{180t - x)] + 90552cosh[244t - x]+
25200 cosh[3{324t - x)] + 112266 cosh[484t - x]+
5292 cosh [964t - x] + 28875cosh[3{12t + x)]+
43120cosh[3{60t + x)] + 115500cosh[188t + x]+
29106 cosh[476t + x] + 24255 cosh [28t + 5x] ]
294 Appendix B. KdV Solitons via Backlund Thansform
The six soliton solution is:
U123456 = [-1990617855408 - 611966880000cosh[144t]-

81327286656 cosh[240t]- 259891443000 cosh[288t]-
565820640000 cosh[336t]- 391727952000 cosh[432t]-
705144542256 cosh[480t]- 352983015000 cosh[576t]-
103733784000 cosh[672t]- 458859582720 cosh[720t]-
13070456784 cosh[960t]- 52390800000 cosh[1008t]-
123773265000 cosh[1056t]- 77014476000 cosh[1152t]-
3360cosh[2(1732t -19x)]- 100800cosh[2(1508t -17x)]-
564480 cosh[2(1652t - 17x)]- 164640 cosh[2(1724t - 17x)]-
12418560 cosh[2(868t - 13x)] - 14968800 cosh[2(892t - 13x )]-
37255680cosh[2(1156t -13x)]- 360581760cosh[2(1228t -13x)]-
254741760cosh[2(1396t -13x)]- 493920000cosh[2(1444t -13x)]-
186701760cosh[2(1468t -13x)]- 94832640cosh[2(1516t -13x)]-
110638080cosh[2(1588t -13x)]- 23284800cosh[2(644t - 11x)]-
576723840 cosh[2(788t - 11x )]- 310464000 cosh[2(860t - 11x )]-
507091200 cosh[2(1004t - 11x )]- 5119329600 cosh[2(1124t - 11x )]-
1825528320 cosh[2(1148t - 11x )]- 3353011200 cosh[2(1196t - 11x )]-
926100000 cosh[2(1220t - 11x )]- 1451520000 cosh[2(1292t - 11x )]-
968083200cosh[2(1436t -l1x)]- 290424960cosh[2(1508t -l1x)]-
179292960cosh[2(4t - 7x)]- 5524989120cosh[2(292t -7x)]-
6601240800 cosh[2(388t - 7x)]- 2390572800 cosh[2(508t - 7x)]-
28523880000cosh[2(580t - 7x)]- 36966948480cosh[2(604t - 7x)]-
27165600cosh[4(628t - 7x)]- 69065821440cosh[2(652t - 7x)]-
73347120000 cosh[2(676t - 7x)]- 39734217600 cosh[2(724t - 7x)]-
176329440cosh[4(736t -7x)]- 41074387200cosh[2(748t -7x)]-
14817600cosh[4(772t - 7x)]- 53343360cosh[4(796t -7x)]-
17287200cosh[4(808t -7x)]-197743835520cosh[2(892t -7x)]-
50709120000cosh[2(940t - 7x)]- 159016556160cosh[2(1012t - 7x)]-
7302113280 cosh[2(1084t - 7 x )]- 50409475200 cosh[2(1108t - 7x )]-
11379916800cosh[2(1228t - 7x)]- 6721263360cosh[2(1252t - 7x)]-
138124728000cosh[2(284t - 5x)]- 37966320000cosh[2(332t - 5x)]-
136914624000 cosh[2(356t - 5x)]- 1571724000 cosh[4(392t - 5x)]-
2376990000cosh[4(416t - 5x)]- 37255680000cosh[2(428t - 5x)]-
46103904000cosh[2(476t - 5x)]- 4400088000cosh[4(488t - 5x)]-
10584000 cosh[6( 492t - 5x )]- 6112260000 cosh [4 (524t - 5x )]-
28523880000cosh[2(572t - 5x)]- 1530900000cosh[4(572t - 5x)]-
2716560000cosh[4(596t - 5x)]- 3500658000cosh[4(608t - 5x)]-
84135744000cosh[2(644t - 5x)]- 8573040000cosh[4(644t - 5x)]-
2074464000 cosh[4(668t - 5x)] - 441712656000 cosh[2(716t - 5x)]-
114095520000cosh[2(788t - 5x)]- 403558848000cosh[2(836t - 5x)]-
10886400000cosh[2(1004t - 5x)]- 24893568000cosh[2(1076t - 5x)]-
1886068800cosh[6(212t - 3x)]- 1185528960cosh[6(244t - 3x)]-
12070840320 cosh[6(364t - 3x)]- 10287648000 cosh[6(380t - 3x)]-
5881105440cosh[6(404t - 3x)]- 14402707200cosh[6(428t - 3x)]-
B.6. Six Solitons 295
1411200cosh[16(203t - 2x)] - 1905176582400cosh[2(4t - x)]-

9147600cosh[16(4t - x)] - 86063160960cosh[6(12t - x)]-
760656711360 cosh[4(16t - x)] - 6119668800 cosh[12(24t - x )]-
774568448640cosh[2(28t - x)] - 156168260640cosh[8(28t - x)]-
70624689408 cosh[10(28t - x)] - 220308076800 cosh[8(34t - x )]-
1145413301760cosh[6(36t - x)] - 1376925480000cosh[4(40t - x)]-
5122656cosh[20(40t - x)] - 168739200cosh[18(44t - x)]-
25101014400cosh[8(46t - x)] - 1703585520cosh[16(46t - x)]-
52257653280cosh[12(48t - x)] - 512265600cosh[16(49t - x)]-
624673042560 cosh[4(52t - x)] - 84544653312 cosh[10(52t - x )]-
11410398720cosh[14(52t - x)] -10373378400cosh[12(56t - x)]-
206705520000cosh[6(60t - x)] - 18928728000cosh[12(60t - x)]-
66528 cosh [30 (60t - x)] - 536220594000cosh[8(64t - x)]-
69313028400 cosh[12(64t - x)] - 555043104 cosh[20(64t - x )]-
1940400 cosh[28 (64t - x)] - 27941760cosh[24(66t - x)]-
11670785280 cosh[16 (67t - x)] - 4107438720cosh[18(68t - x)]-
56582064000 cosh[8(70t - x)] - 221288760 cosh[24 (72t - x )]-
1765617235200cosh[2(76t - x)] - 333902761920cosh[4(76t - x)]-
447524118720 cosh[6(76t - x)] - 192492181728 cosh[10(76t - x )]-
102769793280 cosh[14(76t - x)] - 27938267280 cosh[16(76t - x )]-
5215795200cosh[18(76t - x)] - 3940823040cosh[20(76t - x)]-
733471200 cosh[22(76t - x)] - 138600 cosh [32(79t - x )]-
12806640000cosh[8(82t - x)] - 3630682440cosh[16(82t - x)]-
234846259200cosh[6(84t - x)] - 185533286400cosh[12(84t - x)]-
20925273600cosh[18(84t - x)] - 17463600cosh[24(84t - x)]-
3725568cosh[30(84t - x)] - 38031840000cosh[16(85t - x)]-
156881340000cosh[4(88t - x)] - 878295088440cosh[8(88t - x)]-
236333401920cosh[12(88t - x)] - 20956320 cosh[28 (88t - x)]-
84cosh[40(88t - x)] -17854784640cosh[16(9lt - x)]-
420431679360cosh[6(92t - x)] - 2037420000cosh[18(92t - x)]-
30240 cosh[36 (92t - x)] - 28426536600cosh[16(94t - x)]-
2381400 cosh[32 (94t - x)] - 75547533600 cosh[12(96t - x )]-
770144760cosh[24(96t - x)] - 35280cosh[36(96t - x)]-
15688134000 cosh[8(100t - x)] - 904089263616 cosh[1O(100t - x )]-
11209968000 cosh[18(100t - x)] - 1232231616 cosh[20(100t - x )]-
1122660000cosh[24(100t - x)] - 10584000cosh[28(100t - x)]-
16003008cosh[30(100t - x)] - 456382080cosh[24(102t - x)]-
3333960 cosh[32(103t - x)] - 31120135200 cosh[12(104t - x )]-
75014100cosh[24(104t - x)] - 222017241600cosh[8(106t - x)]-
411600cosh[32(106t - x)] - 898502220000cosh[6(108t - x)]-
252558583200 cosh[12(108t - x)] - 24222401280 cosh[18(108t - x )]-
24893568 cosh[30(108t - x)] - 470318879520 cosh[4(112t - x )]-
88978594320 cosh[16(112t - x)] - 24863975136 cosh[20(112t - x )]-
1730937600 cosh[24(114t - x)] - 443603381760 cosh[6(116t - x )]-
11855289600 cosh[12(116t - x)] - 41074387200 cosh[18(116t - x )]-
420078960cosh[24(116t - x)] - 913242274560cosh[8(118t - x)]-
15717240000 cosh[16(118t - x)] - 79866864000 cosh[12(120t - x )]-

1296540000 cosh[24(120t - x)] - 34496689920 cosh[16(12lt - x )]-
880310323200 cosh[4(124t - x)] - 360972612000 cosh[8(124t - x )]-
335988487296 cosh[10(124t - x)] - 107820266400 cosh[12(124t - x )]-
109919779200cosh[14(124t - x)] - 27871905600cosh[18(124t - x)]-
6883833600cosh[22(124t - x)] - 248935680cosh[24(126t - x)]-
36966948480cosh[16(127t - x)] - 74875185000cosh[16(130t - x)]-
430694288640 cosh[6(132t - x)] - 15717240000 cosh[12(132t - x )]-
72606240 cosh[24(132t - x)] - 412483246560 cosh[4(136t - x )]-
49417622100cosh[8(136t - x)] - 27725211360cosh[16(136t - x)]-
7633434816 cosh[20(136t - x)] - 7938000000 cosh[16(139t - x )]-
188606880000cosh[6(140t - x)] - 6096384000cosh[18(140t - x)]-
21874111560cosh[16(142t - x)] - 592215624000cosh[12(144t - x)]-
559109537280 cosh[2(148t - x)] - 395176320000 cosh[6(148t - x)]-
216062652960 cosh[8(148t - x)] - 4537890000 cosh[16(148t - x )]-
5547709440cosh[18(148t - x)] - 116169984cosh[20(148t - x)]-
149845449600 cosh[8(154t - x)] - 720170922240 cosh[6(156t - x )]-
13691462400cosh[12(156t - x)] - 25391439360cosh[16(157t - x)]-
1048252590000cosh[4(160t - x)] -173282571000cosh[8(160t - x)]-
79214889600 cosh[12(164t - x)] - 1016487360 cosh[16(166t - x )]-
50316124320 cosh[12(168t - x)] - 215550720000 cosh[2(172t - x )]-
130704567840cosh[6(172t - x)] - 424462500000cosh[8(172t - x)]-
215442767232cosh[1O(172t - x)] - 5511240000cosh[12{172t - x)]-
201264497280 cosh[8(178t - x)] - 33339600000 cosh[12(180t - x )]-
471247761600 cosh[4(184t - x)] - 12602368800 cosh[12(184t - x )]-
158403168000 cosh[8(190t - x)] - 570007247040 cosh[4(196t - x )]-
199168865280 cosh[6(196t - x)] - 69313028400 cosh[8(196t - x )]-
59147117568cosh[10(196t - x)] - 371482372800cosh[6(204t - x)]-
7468070400 cosh[12(204t - x)] - 355914377280 cosh[8(208t - x)]-
147867793920 cosh[6(212t - x)] - 187928328000 cosh[2(220t - x )]-
461039040000cosh[4(220t - x)] - 456382080000cosh[6(228t - x)]-
142967119680cosh[4(232t - x)] - 1694947161600cosh[2(244t - x)]-
11980029600cosh[8(244t - x)] - 1394039808cosh[1O(244t - x)]-
10287648000 cosh[8(250t - x)] - 218001611520 cosh[6(252t - x )]-
152488662480cosh[4(256t - x)] -77962500000cosh[4(268t - x)]-
5881105440 cosh[8(268t - x)] - 87886149120 cosh[6(276t - x )]-
134469720000cosh[2(292t - x)] - 58090919040cosh[4(292t - x)]-
621599378400 cosh[4(304t - x)] - 323460799200 cosh[2(316t - x )]-
4320812160cosh[6(332t - x)] - 133111440000cosh[4(340t - x)]-
905964998400cosh[2(364t - x)] - 22632825600cosh[4(364t - x)]-
77617552320 cosh[4(376t - x)] - 147867793920 cosh[2(388t - x)]-
152996659200cosh[2(436t - x)] - 533433600cosh[4(484t - x)]-
79068195360 cosh[2(508t - x)] - 188606880000 cosh[2(532t - x )]-
93139200000cosh[2(580t - x)] - 250847150400cosh[2(604t - x)]-
18255283200 cosh[2(724t - x)] - 960498000 cosh[12x]-
2801452500cosh[8(8t + x)] - 717171840cosh[12(12t + x)]-
B.7. Seven Solitons 297
20582100000 cosh[8(20t + x)] - 1056198528 cosh[10(20t + x )]-

22058467200cosh[8(26t + x)] - 122245200000cosh[6(36t + x)]-
36212520960cosh[6(44t + x)) - 860606208 cosh[10(44t + x)]-
32955753600 cosh[4(56t + x)] - 10373378400 cosh[8(56t + x )]-
365858428320 cosh[4(68t + x)] - 35565868800 cosh[6(76t + x )]-
22590912960 cosh[6(84t + x)] - 19921440000 cosh[4(92t + x )]-
27725211360cosh[4(104t + x)] - 74413987200cosh[2(116t + x)]-
244490400 cosh[8(116t + x)] - 192022760160 cosh[4(128t + x )]-
262497312000 cosh[2(140t + x)] - 3353011200 cosh[6(156t + x )]-
392841187200 cosh[2(164t + x)] - 10040405760 cosh[4(164t + x )]-
316808849280 cosh[2(212t + x)) - 12474000000 cosh[4(236t + x))-
9241737120 cosh[4(248t + x)] - 103733784000 cosh[2(260t + x )]-
263560651200cosh[2(284t + x)] - 123295011840cosh[2(332t + x)]-
4563820800 cosh[2(356t + x)] - 81130896000 cosh[2(500t + x )]-
7302113280 cosh[2(572t + x)] - 1232231616 cosh[1440t]-
2286900000 cosh[2( 4t + 5x)) - 6830208000 cosh[2(76t + 5x)] ]
[ 21 cosh[1756t - 19x] + 1575 cosh[1252t - 13x]+

4410cosh[1540t - 13x] + 1386cosh[764t - 1b]+
11907 cosh[1244t - 11x] + 7056 cosh[1484t - 11x]+
30870cosh[1180t - 7x] + 840cosh[3(516t - 5x)]+
72765 cosh [548t - 5x] + 490cosh[3(564t - 5x)]+
55440 cosh[692t - 5x] + 51030 cosh[1028t - 5x]+
4950 cosh[7( 4t - x)] + 462 cosh[9(4t - x )]+
26950 cosh [3 (84t - x)] + 12936 cosh [9 (84t - x)]+
cosh[21(84t - x)] + 53460 cosh[7(100t - x )]+
210cosh[17(100t - x)] + 36750cosh[9(132t - x))+
27000cosh[7(148t - x)] + 4116cosh[9(164t - x)]+
169785cosh[3(180t - x)] + 90552cosh[244t - x]+
25200 cosh[3(324t - x)) + 112266 cosh[484t - x]+
5292cosh[964t - x] + 28875cosh[3(12t + x)]
+43120 cosh[3(60t + x)] + 115500 cosh[188t + x]+
29106 cosh [476t + x] + 24255cosh[28t + 5x] J2
B.7 Seven Solitons

The analytical expression for the seven soliton ladder is:
W1234567 = [1456sinh[2(1564t -13x)] + 92400sinh[2(1460t - 11x)]+

51744sinh[2(1532t - 11x)] + 7606368sinh[2(1276t - 7x))+
4004000sinh[2(164t - 5x)] + 28028000sinh[2(596t - 5x)]+
231000sinh[4(656t - 5x)] + 18018000sinh[2(668t - 5x)]+
588000sinh[4(728t - 5x)) + 11858000sinh[2(812t - 5x)]+
67914000sinh[2(956t - 5x)] + 8316000sinh[2(1172t - 5x)]+
299376 sinh[6(356t - 3x)] + 2095632 sinh[6(436t - 3x )]+
1143072sinh[6(476t - 3x)] + 9225216sinh[8(22t - x)]+
21525504sinh[4(28t - z)] + 27243216sinh[6(28t - z)]+

48048 sinh[14(28t - z)] + 648648 sinh[12(32t - z )]+
5605600sinh[8(40t - z)] + 4708704sinh[2(52t - z)]+
4953312 sinh[6(68t - z)] + 192192 sinh[16(88t - z )]+
20966400 sinh[4(100t - z)] + 2402400sinh[14(100t - z)]+
22198176 sinh[8(112t - z)] + 13621608 sinh[12(112t - z)]+
56 sinh[28 (112t - z)] + 18144sinh[24(128t - z)]+
3311616 sinh[16(133t - z)] + 85621536 sinh[6(148t - z )]+
80080000 sinh[8(148t - z)] + 16166304 sinh[14(148t - z)]+
26195400 sinh[12(160t - z)] + 7761600 sinh[16(160t - z )]+
28658448sinh[2(172t - z)] + 19160064sinh[12(172t - z)]+
6930000sinh[14(172t - z)] + 790272sinh[16(178t - z)]+
63567504sinh[6(188t - z)] + 14968800sinh[12(200t - z)]+
42499072sinh[8(202t - z)] + 116540424sinh[4(208t - z)]+
48498912sinh[8(232t - z)] + 54885600sinh[6(260t - z)]+
16816800sinh[4(280t - z)] + 1862784sinh[8(292t - z)]+
12573792 sinh[6(308t - z)] + 26426400 sinh[2(340t - z )]+
12705000 sinh[4(352t - z)] + 18213888 sinh[4(388t - z )]+
33297264sinh[2(412t - z)] + 7114800sinh[2(700t - z)]-
8845200sinh[6(20t + z)] - 33434856sinh[4(32t + z)]-
38220000 sinh[2(92t + z)] - 3171168 sinh[4(152t + z )]-
11099088 sinh[2(308t + z)] - 243936 sinh[2(668t + z)] ]
[ 6019650 cosh[192t] + 5054400 cosh[336t] + 8494200 cosh[672t] +

490050 cosh[1344t] + 28 cosh[2(1564t - 13z )]+
2100cosh[2(1460t - lIz)] + 1176cosh[2(1532t -l1z)]+
271656cosh[2(1276t -7z)] + 200200cosh[2(164t - 5z)]+
1401400 cosh[2(596t - 5z)] + 5775 cosh[4(656t - 5z)]+
900900 cosh [2 (668t - 5z)] + 14700cosh[4(728t - 5z)]+
592900cosh[2(812t - 5z)] + 3395700cosh[2(956t - 5z)]+
415800 cosh[2(1172t - 5z)] + 8316 cosh[6(356t - 3z )]+
58212cosh[6(436t - 3z)] + 31752 cosh [6 (476t - 3z)]+
576576 cosh[8(22t - z)] + 2690688 cosh[4(28t - z )]+
2270268 cosh [6 (28t - z)] + 1716cosh[14(28t - z)]+
27027 cosh[12(32t - z)] + 350350 cosh [8 (40t - z )]+
1177176 cosh [2 (52t - z)] + 412776 cosh[6(68t - z )]+
6006cosh[16(88t - z)] + 2620800cosh[4(100t - z)]+
85800 cosh[14(100t - z)] + 1387386 cosh[8(112t - z )]+
567567 cosh[12(112t - z)] + cosh[28(112t - z )]+
378 cosh[24(128t - z)] + 103488 cosh[16(133t - z )]+
7135128 cosh[6(148t - z)] + 5005000 cosh[8(148t - z )]+
577368cosh[14(148t - z)] + 1091475cosh[12(160t - z)]+
242550 cosh[16(160t - z)] + 7164612 cosh[2(172t - z )]+
798336cosh[12(172t - z)] + 247500cosh[14(172t - z)]+
24696 cosh[16(178t - z)] + 5297292 cosh[6(188t - z )]+
623700cosh[12(200t - z)] + 2656192cosh[8(202t - z)]+
14567553cosh[4(208t - x)] + 3031182cosh[8(232t - x)]+

4573800 cosh[6(260t - x)] + 2102100 cosh[4(280t - x )]+
116424 cosh[8(292t - x)] + 1047816 cosh[6(308t - x )]+
6606600 cosh[2(340t - x)] + 1588125 cosh[4(352t - x )]+
2276736 cosh[4(388t - x)] + 8324316 cosh[2(412t - x )]+
1778700 cosh [2 (700t - x)] + 737100 cosh [6 (20t + x )]+
4179357 cosh[4(32t + x)] + 9555000 cosh[2(92t + x )]+
396396 cosh[4(152t + x)] + 2774772 cosh[2(308t + x)]+
60984cosh[2(668t + x)] ]
The expression for the seven soliton solution is:
U1234S61 = [-27553846057384192 - 2222678091110400 cosh[144t]-

8495891967467520 cosh[240t] - 6538746445862400 cosh[288t]-
1156175262950400 cosh[336t] - 10989038170880000 cosh[432t]-
2090436576205824 cosh[480t] - 1840147482700800 cosh[528t]-
3546915573801600 cosh[576t] - 8622020177971200 cosh[672t]-
12392672682120192 cosh[720t] - 1571137568000000 cosh[864t]-
8108388570522240 cosh[960t] - 3702318344524800 cosh[1008t]-
407798251660800 cosh[1056t] - 1065086243020800 cosh[1152t]-
1124534572761600 cosh[1248t] - 886004184665600 cosh[1296t]-
1495260256089600 cosh[1392t] - 1628197592845824 cosh[1440t]-
713646940406400 cosh[1536t] - 543855312000000 cosh[1584t]-
1904612024899584 cosh[1680t] - 53328898022400 cosh[1728t]-
111217704998400 cosh[1968t] - 179485571865600 cosh[2016t]-
83431755703296 cosh[2160t] - 1735555852800 cosh[2736t]-
75600 cosh[2(3028t - 25x)] - 831600 cosh[2(2636t - 23x )]-
11642400 cosh[2(2876t - 23x)] - 6350400 cosh[2(2996t - 23x )]-
14817600 cosh[2(3020t - 23x)] - 1778112 cosh[2(3068t - 23x )]-
129729600 cosh[2(1732t - 19x)] - 148324176 cosh[2(1756t - 19x )]-
908107200 cosh[2(2164t - 19x)] - 6904209312 cosh[2(2236t - 19x )]-
384199200 cosh[2(2380t - 19x)] - 15402895200 cosh[2(2524t - 19x )]-
21824510400 cosh[2(2572t - 19x)] - 8762535936 cosh[2(2596t - 19x )]-
1344697200 cosh[2(2620t - 19x)] - 28560470400 cosh[2(2740t - 19x )]-
6845731200 cosh[2(2764t - 19x)] - 3734035200 cosh[2(2884t - 19x )]-
1045529856 cosh[2(2956t - 19x)] - 713512800 cosh[2(1508t - 17x )]-
12913284384 cosh[2(1652t - 17x)] - 6356750400 cosh[2(1724t -17x)]-
399567168 cosh[2(1772t - 17x)] - 35276072832 cosh[2(2012t - 17x )]-
229785117408 cosh[2(2132t - 17x)] - 90810720000 cosh[2(2156t - 17x )]-
133491758400 cosh[2(2204t - 17x)] - 174243484800 cosh[2(2348t - 17x )]-
48193947648 cosh[2(2372t - 17x)] - 229209750000 cosh[2(2420t - 17x )]-
895103104896 cosh[2(2492t - 17x)] - 93884313600 cosh[2(2564t - 17x )]-
112954564800 cosh[2(2588t - 17x)] - 261954000000 cosh[2(2660t - 17x )]-
61611580800 cosh[2(2708t - 17x)] - 223914087936 cosh[2(2732t - 17x )]-
3136589568 cosh[2(2852t - 17x)] - 185513328 cosh[2(388t - 13x )]-
59940197856 cosh[2(868t - 13x)] - 18551332800 cosh[2(892t - 13x )]-
764172208800 cosh[2(1156t - 13x)] - 1930518285696 cosh[2(1228t - 13x )]-
834323490000 cosh[2(1252t - 13x)] - 389577988800 cosh[2(1372t - 13x )]-
11026603224000 cosh[2(1396t - 13x)] - 3496212720000 cosh[2(1444t - 13x )]-
3775200863616 cosh[2(1468t - 13x)] - 5411713507200 cosh[2(1516t - 13x )]-
300 Appendix B. KdV Solitons via. Bii.cklund Transform
6541096161600 cosh[2(1540t - 13x)] - 6048 cosh[4(1552t - 13x )]-

2531003739264 cosh[2(1588t - 13x)] - 13586989852800 cosh[2(1660t - 13x )]-
2621560189248 cosh[2(1708t - 13x)] - 4724610508800 cosh[2(1732t - 13x )]-
16152502766400 cosh[2(1756t - 13x)] - 38830500878400 cosh[2(1876t - 13x )]-
50067668851200 cosh[2(1900t - 13x)] - 15556748118912 cosh[2(1948t - 13x )]-
5041461902400 cosh[2(1996t - 13x)] - 46481536886400 cosh[2(2020t - 13x)]-
1843734638592 cosh[2(2068t -13x)] -11956069540800 cosh[2(2092t - 13x)]-
15462224870400 cosh[2(2116t - 13x)] - 2161120500000 cosh[2(2164t - 13x )]-
1734982115328 cosh[2(2308t - 13x)] - 9624609086400 cosh[2(2356t - 13x )]-
6037934918400 cosh[2(2380t - 13x)] - 617463000000 cosh[2(2404t - 13x)]-
3630682440000 cosh[2(2452t -13x)] - 1355454777600 cosh[2(2476t -13x)]-
739338969600 cosh[2(2596t - 13x)] - 35618558976 cosh[2(284t - 11x )]-
43286443200 cosh[2(356t - 11x)] - 643727855232 cosh[2(644t -11x)]-
646100110656 cosh[2(764t - 11x)] - 4456344816000 cosh[2(788t -11x)]-
1498504946400 cosh[2(860t - 11x)] - 247913265600 cosh[2(908t - 11x )]-
24442044314688 cosh[2(1004t - 11x)] - 30667008468096 cosh[2(1124t - 11x )]-
864864 cosh[4(1136t - 11x)] - 46015153984800 cosh[2(1148t - 11x )]-
14444353123200 cosh[2(1196t - 11x)] - 35756721000000 cosh[2(1220t - 11x )]-
47684591018064 cosh[2(1244t - 11x)] - 14906579688000 cosh[2(1268t - 11x )]-
25734028320000 cosh[2(1292t - 11x)] - 29804544 cosh[4(1316t - 11x )]-
2045284441200 cosh[2(1340t - 11x)] - 69854400 cosh[4(1424t - 11x )]-
6852576931200 cosh[2(1436t - 11x)] - 104315904 cosh[4(1436t - 11x )]-
97140626751168 cosh[2(1484t - 11x)] - 123227725420800 cosh[2(1508t - 11x )]-
12236744520000 cosh[2(1556t - 11x)] - 24723218520000 cosh[2(1652t - 11x)]-
5753767219200 cosh[2(1700t - 11x)] - 368415513097632 cosh[2(1724t - 11x )]-
5177084220000 cosh[2(1772t - 11x)] - 88190182080000 cosh[2(1796t - 11x )]-
328945638155904 cosh[2(1844t - 11x)] - 137650889376000 cosh[2(1868t - 11x )]-
14825286630000 cosh[2(1916t - 11x)] - 63003869603136 cosh[2(1964t - 11x )]-
47851661088000 cosh[2(1988t - 11x)] - 46593757980000 cosh[2(2012t - 11x )]-
110937519000000 cosh[2(2060t - 11x)] - 51953075564544 cosh[2(2084t - 11x )]-
39204784080000 cosh[2(2132t - 11x)] - 84565256041728 cosh[2(2204t - 11x )]-
2587686393600 cosh[2(2348t -11x)] - 2074675680000 cosh[2(2372t -11x)]-
1138582013184 cosh[2(2444t - 11x)] - 6922501185600 cosh[2( 4t - 7x )]-
59003357212800 cosh[2(100t - 7x)] - 5235928169472 cosh[2(172t - 7x )]-
19392326553600 cosh[2(220t - 7x)] - 6363107150400 cosh[2(244t - 7x )]-
55214108078976 cosh[2(292t - 7x)] - 8181137764800 cosh[2(388t - 7x)]-
16921669564800 cosh[2(508t -7x)] - 788566622659200 cosh[2(580t - 7x)]-
343364762664000 cosh[2(604t -7x)] - 89503045632 cosh[4(628t - 7x)]-
881619665639328 cosh[2(652t - 7x)] - 90901530720000 cosh[2(676t - 7x )]-
54540918432 cosh[4(688t - 7x)] - 661400971862400 cosh[2(724t -7x)]-
2154745756800 cosh[4(736t -7x)] - 176943325759200 cosh[2(748t - 7x)]-
2440992153600 cosh[4(772t - 7x)] - 260308928880000 cosh[2(796t - 7x )]-
813664051200 cosh[4(796t - 7x)] -1454763518208 cosh[4(808t -7x)]-
1077372878400 cosh[4(832t -7x)] - 406639733900400 cosh[2(844t - 7x)]-
289766400 cosh[6(876t - 7x)] - 1153750197600 cosh[4(880t - 7x )]-
1455822468067968 cosh[2(892t - 7x)] - 2444204431468800 cosh[2(940t - 7x )]-
315555609600 cosh[4(940t - 7x)] - 1551699072 cosh[6(948t - 7x )]-
123480000 cosh[6(972t - 7x)] - 4405228027200 cosh[4(976t - 7x )]-
5638691874816 cosh[4(988t - 7x)] - 432081216 cosh[6(988t - 7x )]-
138297600 cosh[6(996t - 7x)] - 2106689124443904 cosh[2(1012t - 7x )]-
B. 7. Seven Solitons 301
9310194432000 cosh[4(1012t -7x)]- 10594584000000 cosh[4(1024t - 7x)]-

5289520465152 cosh[4(1048t - 7x )]- 4417778534400 cosh[4(1060t - 7x )]-
1094699164759200 cosh[2(1084t - 7x)]- 1858472260444800 cosh[2(1108t -7x)]-
1485034405991136 cosh[2(1132t - 7x)]-19334952806400 cosh[4(1132t - 7x)]-
852864012000000 cosh[2(1156t - 7x)]- 4819394764800 cosh[4(1156t - 7x)]-
320513711918400 cosh[2(1180t - 7x)]-14311490054400 cosh[4(1192t -7x)]-
586419759525600 cosh[2(1228t - 7x)]- 630902587392 cosh[4(1228t - 7x)]-
4066364332800 cosh[4(1240t - 7x)]- 287525390390784 cosh[2(1252t - 7x)]-
1093890798000000 cosh[2(1300t - 7x)]- 844958822400 cosh[4(1300t - 7x)]-
492892646400 cosh[4(1312t - 7x)]- 1008129759436800 cosh[2(1444t - 7x)]-
128456449273152 cosh[2(1492t - 7x)]- 395625567330000 cosh[2(1516t - 7x)]-
533308702449600 cosh[2(1588t - 7x)]- 985948768861056 cosh[2(1612t - 7x)]-
388281316500000 cosh[2(1660t - 7x)]- 495724666719744 cosh[2(1732t - 7x)]-
12704483994048 cosh[2(1852t - 7x)]- 110937519000000 cosh[2(1900t - 7x)]-
67095011491200 cosh[2(1948t -7x)]- 975927439872 cosh[2(2092t - 7x)]-
241026786000000 cosh[2(68t - 5x)]- 80951270400 cosh[4(68t - 5x)]-
151502551200 cosh[4(176t - 5x )]- 14847250017600 cosh[2(212t - 5x )]-
1380352701974400 cosh[2(284t - 5x)]- 2268124008796800 cosh[2(332t - 5x)]-
660840681362400 cosh[2(356t - 5x )]- 17643451795200 cosh[4(392t - 5x )]-
4327213363200 cosh[2( 404t - 5x )]- 4857046286400 cosh[4( 416t - 5x )]-
1029964568371200 cosh[2( 428t - 5x )]- 4533271142400 cosh[4( 452t - 5x )]-
2061259200 cosh[6( 468t - 5x )]- 932243478566400 cosh[2( 476t - 5x )]-
82176955660800 cosh[4( 488t - 5x )]- 214013654400 cosh[6( 492t - 5x )]-
237318681600 cosh[6(516t - 5x )]- 20138185267200 cosh[4(524t - 5x )]-
1780813429995600 cosh[2(548t - 5x )]- 80754273600 cosh[6(564t - 5x )]-
345945600 cosh[8(566t - 5x )]- 788566622659200 cosh[2(572t - 5x )]-
62531742873600 cosh[4(572t - 5x )]- 71034163200 cosh[6(588t - 5x )]-
8950304563200 cosh[4(596t - 5x)]- 315085047076800 cosh[4(608t - 5x)]-
115413160262400 cosh[4( 632t - 5x)] - 558835200 cosh[8( 632t - 5x )]-
1390768579200 cosh[6(636t - 5x)]- 920912547693600 cosh[2(644t - 5x)]-
80100999014400 cosh[4(644t - 5x)]- 172422809395200 cosh[4(668t - 5x)]-
326761419600 cosh[6(676t - 5x)]- 4120116000000 cosh[6(684t - 5x)]-
1877353928049600 cosh[2( 692t - 5x )]- 3113510400 cosh[8( 692t - 5x )]-
61535160086400 cosh[4(704t - 5x )]- 5867769600 cosh[8(704t - 5x )]-
2857318556400 cosh[6(708t - 5x )]- 2364884899977600 cosh[2(716t - 5x )]-
1066867200 cosh[8(722t - 5x )]- 6525096916800 cosh[6(756t - 5x )]-
3781810618185600 cosh[2(764t - 5x )]- 1120324867200 cosh[6(772t - 5x )]-
134358588000000 cosh[4(776t - 5x )]- 5829432527318400 cosh[2(788t - 5x )]-
2495269022400 cosh[6(796t - 5x )]- 4991679000000 cosh[6(804t - 5x )]-
146544065740800 cosh[4(812t - 5x )]- 244940382086400 cosh[4(824t - 5x )]-
480249000000 cosh[6(828t - 5x )]- 3345607551686400 cosh[2(836t - 5x )]-
1307045678400 cosh[6(836t - 5x )]- 508803837141600 cosh[4(848t - 5x )]-
263560651200 cosh[6(852t - 5x )]- 1396529164800 cosh[6(876t - 5x )]-
214807309516800 cosh[4(884t - 5x )]- 99891792000000 cosh[4(896t - 5x )]-
613328943427200 cosh[2(908t - 5x )]- 169009008537600 cosh[4(908t - 5x )]-
2380754376000000 cosh[2(932t - 5x )]- 52935349975200 cosh[4(944t - 5x )]-
136525305470400 cosh[4(992t - 5x)]- 2756456140838400 cosh[2(1004t - 5x)]-
53013342412800 cosh[4(1004t - 5x )]- 875839426862400 cosh[2(1028t - 5x )]-
159040027238400 cosh[4(1028t - 5x )]- 4088185101000000 cosh[2(1052t - 5x )]-
71940531801600 cosh[4(1064t - 5x )]- 569475841334400 cosh[2(1076t - 5x )]-
3303876523147200 cosh[2(1124t - 5x)] - 1644025772188800 cosh[2(1148t - 5x )]-

1161818380800 cosh[4(1184t - 5x)] - 533032591291200 cosh[2(1196t - 5x )]-
2009885514436800 cosh[2(1268t - 5x)] - 4375563192000000 cosh[2(1292t - 5x )]-
75731041108800 cosh[2(1484t - 5x)] - 921952867651200 cosh[2(1508t - 5x )]-
41653340467200 cosh[2(1556t - 5x)] - 380205065116800 cosh[2(1628t - 5x )]-
20376279000000 cosh[2(1844t - 5x)] - 14910002553600 cosh[2(1868t - 5x )]-
1434355322400 cosh[6(44t - 3x)] - 2208907196496 cosh[6(92t - 3x)]-
52141956273792 cosh[6(212t ...:. 3x)] - 45773273145600 cosh[6(244t - 3x )]-
16868018591424 cosh[6(292t - 3x)] - 247591795651200 cosh[6(364t - 3x )]-
30893806944 cosh[12(368t - 3x)] - 235344607898400 cosh[6(380t - 3x )]-
32438497291200 cosh[6(388t - 3x)] - 227069481038400 cosh[6(404t - 3x)]-
430359242288976 cosh[6(412t - 3x)] - 138221899200 cosh[12(416t - 3x)]-
101949562915200 cosh[6( 428t - 3x)] - 86910050304 cosh[12( 428t - 3x )]-
1275914226787200 cosh[6(484t - 3x)] - 292999673584800 cosh[6(508t - 3x)]-
396196624107648 cosh[6(532t - 3x)] - 320412912019200 cosh[6(548t - 3x)]-
279921170390400 cosh[6(580t - 3x)] - 262902902169600 cosh[6(604t - 3x )]-
82343877739200 cosh[6(628t - 3x)] - 30114332430336 cosh[6(652t - 3x )]-
10082923804800 cosh[6(700t - 3x)] - 118756205568 cosh[16(203t - 2x )]-
17548876800 cosh[16(22lt - 2x)] - 159826867200 cosh[16(25lt - 2x )]-
423783360000 cosh[16(257t - 2x)] - 76822276608 cosh[16(263t - 2x )]-
152562009600 cosh[16(275t - 2x)] - 1278942787584 cosh[16(293t - 2x )]-
1337851468800 cosh[16(311t - 2x)] - 253018225152 cosh[16(323t - 2x )]-
251475840000 cosh[16(329t - 2x)] - 15951524583399552 cosh[2( 4t - x )]-
75906083661408 cosh[10(4t - x)] - 61374015166464 cosh[12(4t - x)]-
11641026196800 cosh[14( 4t - x)] - 1506939033600 cosh[16( 4t - x )]-
71788901184 cosh[18( 4t - x)] - 2706757871596800 cosh[6(12t - x )]-
122426304000000 cosh[12(12t - x)] - 4722157440000 cosh[16(13t - x)]-
158694639P0127488 cosh[4(16t - x)] - 134975000160000 cosh[8(16t - x)]-
46855607198400 cosh[12(16t - x)] - 7229150824896 cosh[16(16t - x )]-
47048958899712 cosh[12(24t - x)] - 24360820844371200 cosh[2(28t - x)]-
1700485353907200 cosh[8(28t - x)] - 406119170338560 cosh[10(28t - x)]-
923444121600 cosh[18(28t - x)] - 498659825664 cosh[20(28t - x )]-
5496691200 cosh[24(30t - x)] - 12217165728768 cosh[16(3lt - x)]-
842924192850432 cosh[8(34t - x)] - 7675481467506048 cosh[6(36t - x )]-
30165841305600 cosh[12(36t - x)] - 280560280000 cosh[18(36t - x)]-
5001025704998400 cosh[4( 40t - x)] - 518140820736 cosh [20( 40t - x )]-
10080551150208 cosh[18( 44t - x)] - 2538890021606400 cosh[8( 46t - x )]-
41584549450752 cosh[16( 46t - x)] - 758932057699200 cosh[12( 48t - x )]-
51814082073600 cosh[16( 49t - x)] - 6801941415628800 cosh[4(52t - x )]-
3253700893722048 cosh[10(52t - x)] - 493262525448000 cosh[14(52t - x)]-
1049235161990400 cosh[12(56t - x)] - 12348675159004800 cosh[6(60t - x)]-
462050549452800 cosh[12(60t - x)] - 82450368 cosh[30(60t - x)]-
11745902525220864 cosh[8(64t - x)] - 141631469711424 cosh[12(64t - x)]-
11120484380544 cosh[20(64t - x)] - 3964935744 cosh[28(64t - x )]-
313661365248 cosh[24(66t - x)] - 37404596275200 cosh[16(67t - x)]-
38151640296000 cosh[18(68t - x)] - 635164264627200 cosh[8(70t - x)]-
610295347200 cosh[24(72t - x)] - 8931262685587200 cosh[2(76t - x )]-
8181669932384256 cosh[4(76t - x)] - 4472342754397056 cosh[6(76t - x)]-
238561977221568 cosh[10(76t - x)] - 500067319175424 cosh[14(76t - x)]-
421634418847488 cosh[16(76t - x)] - 6464108851200 cosh[18(76t - x)]-
23284561563648 cosh[20(76t - x)] - 909015307200 cosh[22(76t - x)]-

456648192 cosh[32(79t - x)] - 1295352051840000 cosh[8(82t - x )]-
55380009484800 cosh[16(82t - x)] - 9282339715756800 cosh[6(84t - x )]-
2164981130188800 cosh[12(84t - x)] - 164404944265408 cosh[18(84t - x)]-
1487143257600 cosh[24(84t - x)] - 16049281248 cosh[30(84t - x)]-
336336 cosh[42(84t - x)] - 77712740582400 cosh[16(85t - x)]-
15868062635040000 cosh[4(88t - x)] - 2422262233036800 cosh[8(88t - x )]-
757443074572800 cosh[12(88t - x)] - 173145772800 cosh[28(88t - x)]-
13837824 cosh[40(88t - x)] - 636287642597376 cosh[16(9lt - x)]-
3248675370864000 cosh[6(92t - x)] - 2525042520000 cosh[18(92t - x)]-
461260800 cosh[36(92t - x)] - 47318078126592 cosh[8(94t - x )]-
85673348006400 cosh[16(94t - x)] - 22250277504 cosh[32(94t - x)]-
1513621485129600 cosh[12(96t - x)] - 11184151410432 cosh[24(96t - x)]-
3175460288 cosh[36(96t - x)] - 89451613742400 cosh[6(100t - x )]-
1303947496051200 cosh[8(100t - x)] - 6562938297928320 cosh[10(100t - x)]-
961238736057600 cosh[14(100t - x)] - 18510857164800 cosh[16(100t - x)]-
102771371102400 cosh[18(100t - x)] - 104932828256256 cosh[20(100t - x)]-
3698850355200 cosh[24(100t - x)] - 303005102400 cosh[26(100t - x)]-
879708211200 cosh[28(100t - x)] - 113277292128 cosh[30(100t - x)]-
14832417600 cosh[34(100t - x)] - 3770730163200 cosh[24(102t - x)]-
50854003200 cosh[32(103t - x)] - 2618841534236928 cosh[12(104t - x)]-
12357522777600 cosh[24(104t - x)] - 4448193752217600 cosh[8(106t - x )]-
494300911104 cosh[24(106t - x)] - 67805337600 cosh[32(106t - x)]-
1113543751320000 cosh[6(108t - x)] - 1775022616012800 cosh[12(108t - x)]-
805305653152000 cosh[18(108t - x)] - 3847683840000 cosh[24(108t - x )]-
917764079232 cosh[30(108t - x)] - 33633600 cosh[42(108t - x )]-
197440148505600 cosh[16(109t - x)] - 6830388519292800 cosh[4(112t - x )]-
898012680364800 cosh[16(112t - x)] - 258664201443648 cosh[20(112t - x )]-
1109908800 cosh[36(112t - x)] - 290594304 cosh[40(112t - x )]-
27468783014400 cosh[24(114t - x)] - 2141123807614176 cosh[6(116t - x)]-
1952992987545600 cosh[12(116t - x)] - 192497587682400 cosh[18(116t - x)]-
34915618902528 cosh[24(116t - x)] - 239183407008 cosh[30(116t - x )]-
112 cosh[54(116t - x)] - 6004413002649600 cosh[8(118t - x )]-
1338428931840000 cosh[16(118t - x)] - 685235381107200 cosh[12(120t - x )]-
12114039974400 cosh[24(120t - x)] - 24991366400 cosh[36(120t - x)]-
369600 cosh[48(120t - x)] - 644267332380672 cosh[16(12lt - x)]-
4188857283072000 cosh[4(124t - x)] - 150216121036800 cosh[6(124t - x )]-
4343519961639936 cosh[8(124t - x)] - 9628480327834368 cosh[10(124t - x)]-
355237588113408 cosh[12(124t - x)] - 3026131633324800 cosh[14(124t - x)]-
452541072791808 cosh[18(124t - x)] - 30505427656704 cosh[20(124t - x )]-
181510631108352 cosh[22(124t - x)] - 2136026640384 cosh[28(124t - x )]-
352418242176 cosh[32(124t - x)] - 12915302400 cosh[36(124t - x )]-
226328256 cosh[42(124t - x)] - 84672 cosh[50(124t - x )]-
49337596019712 cosh[24(126t - x)] - 1881600 cosh[48(126t - x )]-
313260659712000 cosh[16(127t - x)] - 1144615547275200 cosh[12(128t - x )]-
526848 cosh[48(129t - x)] - 501365466201600 cosh[16(130t - x )]-
5192377344 cosh[40(130t - x)] - 9869105897481600 cosh[6(132t - x )]-
1758176259840000 cosh[12(132t - x)] - 482298817000000 cosh[18(132t - x )]-
1107487180800 cosh[24(132t - x)] - 3196137776832 cosh[30(132t - x )]-
14874675200 cosh[36(132t - x)] - 97020000 cosh[42(132t - x )]-
4630347489132288 cosh[4(136t - x )]- 8957763150336000 cosh[8(136t - x )]-

912587038114752 cosh[16(136t - x )]- 115683241250304 cosh[20(136t - x )]-
29675173312512 cosh[24(136t - x )]- 16328978265600 cosh[28(136t - x )]-
149361408 cosh[44(136t - x )]- 47883239040000 cosh[24(138t - x )]-
873700834406400 cosh[16(139t - x )]- 5214195627379200 cosh[6(140t - x )]-
826188665702400 cosh[12(140t - x)] - 165924550209600 cosh[18(140t - x )]-
1775545168320 cosh[30(140t - x)] - 48680755200 cosh[36(140t - x )]-
333653114995200 cosh[16(142t - x)] - 12793131187200 cosh[24(142t - x )]-
616548240000 cosh[32(142t - x)] - 2190633984 cosh[40(142t - x )]-
9186418297728672 cosh[12(144t - x)] - 43073340710400 cosh[24(144t - x )]-
22411620000 cosh[36(144t - x )]- 515540933107200 cosh[16(145t - x )]-
59810410243584 cosh[24(146t - x)] - 24391994032929600 cosh[2(148t - x)]-
504841559040000 cosh[4(148t - x)] - 2797255581120000 cosh[6(148t - x)]-
3295675666483200 cosh[8(148t - x)] - 530879469120000 cosh[14(148t - x)]-
747553847040000 cosh[16(148t - x)] - 240134914656000 cosh[18(148t - x)]-
805349068812288 cosh[20(148t - x)] - 250951237992000 cosh[22(148t - x)]-
51455250000 cosh[34(148t - x)] - 33264000000 cosh[36(148t - x )]-
20537193600 cosh[38(148t - x)] - 1194891264 cosh[40(148t - x)]-
11099088000000 cosh[24(150t - x)] - 595672948975104 cosh[16(15lt - x)]-
2053719360000 cosh[32(15lt - x)] - 151444339200 cosh[36(152t - x )]-
8434538836770816 cosh[8(154t - x)] - 420526282176000 cosh[16(154t - x )]-
10954287590400 cosh[24(154t - x)] - 586642839552 cosh[32(154t - x)]-
5970939347493120 cosh[6(156t - x)] - 1758463625318400 cosh[12(156t - x )]-
725159398163200 cosh[18(156t - x)] - 254721153426432 cosh[24(156t - x)]-
2428566970368 cosh[30(156t - x)] - 20445917184 cosh[36(156t - x )]-
414330679526400 cosh[16(157t - x)] - 461039040000 cosh[32(157t - x )]-
2141957412302400 cosh[4(160t - x)] - 2396444046796800 cosh[8(160t - x )]-
282745738207296 cosh[20(160t - x)] - 10882058140800 cosh[28(160t - x)]-
5808499200 cosh[36(160t - x)] - 7831516492800 cosh[24(162t - x )]-
717724922923008 cosh[12(164t - x)] - 577398790096128 cosh[18(164t - x)]-
1796593696128 cosh[30(164t - x)] - 84487176603648 cosh[16(166t - x)]-
76920826397184 cosh[24(166t - x)] - 586776960000 cosh[32(166t - x )]-
3735371428588800 cosh[12(168t - x)] - 20517013440000 cosh[24(168t - x )]-
219689293824 cosh[32(169t - x)] - 267139192320000 cosh[2(172t - x )]-
5046503364302400 cosh[6(172t - x)] - 5254913664000000 cosh[8(172t - x)]-
6429553414764480 cosh[10(172t - x)] - 907899632640000 cosh[12(172t - x)]-
1143953537126400 cosh[14(172t - x)] - 2451962261548800 cosh[16(172t - x)]-
135963828000000 cosh[18(172t - x)] - 248102442491904 cosh[20(172t - x )]-
78172462368000 cosh[26(172t - x)] - 85515570201600 cosh[24(174t - x )]-
984562561382400 cosh[16(175t - x )]- 946122837660000 cosh[12(176t - x )]-
1662892001971200 cosh[8(178t - x)] - 14106505301288400 cosh[6(180t - x )]-
988500114201600 cosh[12(180t - x)] - 690986069989600 cosh[18(180t - x)]-
20917512000000 cosh[24(180t - x)] - 53670532608 cosh[30(180t - x )]-
2181455406434304 cosh[16(18lt - x)] - 7353614629670400 cosh[4(184t - x)]-
1214497338582528 cosh[8(184t - x)] - 1060522604773632 cosh[12(184t - x)]-
355900254893568 cosh[20(184t - x)] - 71914186713600 cosh[24(184t - x)]-
37917927032832 cosh[24(186t - x)] - 4759411629772800 cosh[12(188t - x)]-
791472635553600 cosh[18(188t - x)] - 5942763873446400 cosh[8(190t - x)]-
654158158617600 cosh[16(190t - x)] - 530453163240000 cosh[12(192t - x)]-
235459224000000 cosh[16(193t - x)] - 18044209654732800 cosh[4(196t - x)]-
5391238529188512 cosh[6(196t - x)] - 7410917562048000 cosh[8(196t - x)]-

2713122404280288 cosh[1O(196t - x)] - 6831800479760256 cosh[14(196t - x)]-
391770827078400 cosh[18(196t - x )]- 77436854820864 cosh[20(196t - x )]-
16943184720000 cosh[22(196t - x)] - 9581833046016 cosh[24(196t - x )]-
1646568000000 cosh[24(198t - x)] - 198530590809600 cosh[16(199t - x )]-
581055764889600 cosh[12(200t - x)] - 1777434590131200 cosh[16(202t - x)]-
4034067230793600 cosh[6(204t - x)] - 1476943753678848 cosh[12(204t - x)]-
98132630948352 cosh[18(204t - x)] - 3614546073600 cosh[24(204t - x )]-
3592050721459200 cosh[8(208t - x)] - 110937519000000 cosh[16(208t - x )]-
325634455770624 cosh[16(211t - x )]- 3032999496727200 cosh[6(212t - x )]-
39306093750000 cosh[18(212t - x )]- 950320962759168 cosh[8(214t - x )]-
4208054410560000 cosh[12(216t - x)] - 184013254656 cosh[24(216t - x )]-
730175189990400 cosh[16(217t - x)]- 3303269697974400 cosh[2(220t - x)]-
4307653724774400 cosh[4(220t - x)] - 2295924752721600 cosh[6(220t - x)]-
1563641491059648 cosh[10(220t - x)] - 824107284000000 cosh[14(220t - x )]-
140967296870400 cosh[16(220t - x)] - 300854099145600 cosh[18(220t - x )]-
27842332041216 cosh[20(220t - x )]- 1794365268119424 cosh[12(224t - x )]-
4320386598528000 cosh[8(226t - x)] - 116326419922944 cosh[16(226t - x )]-
6748314452880000 cosh[6(228t - x)] - 4359044020531200 cosh[12(228t - x)]-
61848899481600 cosh[18(228t - x)] - 458206304371200 cosh[4(232t - x )]-
2720862144000000 cosh[8(232t - x )]- 56929100659200 cosh[16(235t - x )]-
34151723020800 cosh[18(236t - x)] - 1048624075699200 cosh[8(238t - x )]-
126785736000000 cosh[16(238t - x )]- 11515760532593856 cosh[2(244t - x )]-
1897652091535200 cosh[6(244t - x )]- 2606082555428352 cosh[8(244t - x )]-
1130356339047936 cosh[10(244t - x)] - 110912161852800 cosh[14(244t - x)]-
60857153280000 cosh[16(244t - x)]- 161880027724800 cosh[16(247t - x)]-
156715857043200 cosh[12(248t - x )]- 925964219980800 cosh[8(250t - x )]-
8643047938324800 cosh[6(252t - x)]- 486606785280000 cosh[12(252t - x)]-
17329677589224000 cosh[4(256t - x)] - 2335571804966400 cosh[8(256t - x )]-
762887885760000 cosh[12(256t - x)] - 2181859353273600 cosh[6(260t - x)]-
193636396800 cosh[18(260t - x )]- 4672211544000000 cosh[8(262t - x )]-
10843638220800 cosh[16(262t - x )]- 576842316078336 cosh[12(264t - x )]-
345027653006400 cosh[2(268t - x )]- 6925450371840000 cosh[4(268t - x )]-
89706461644800 cosh[8(268t - x )]- 2349516450048768 cosh[10(268t - x )]-
47544651000000 cosh[14(268t - x)] - 154044669240000 cosh[12(272t - x )]-
2125029880157184 cosh[8(274t - x)]- 1114493329757808 cosh[6(276t - x)]-
386555621760000 cosh[12(276t - x )]- 1108239497164800 cosh[8(280t - x )]-
175769368879104 cosh[12(284t - x )]- 2085793892582400 cosh[8(286t - x )]-
7831516492800 cosh[12(288t - x )]- 1853346697020000 cosh[2(292t - x )]-
2718655011072000 cosh[4(292t - x)]- 2787430509864000 cosh[6(292t - x)]-
936425614924800 cosh[8(298t - x )]- 7534519696704000 cosh[6(300t - x )]-
5836657428201600 cosh[4(304t - x)] - 98168384160000 cosh[12(312t - x )]-
400875750475200 cosh[2(316t - x)] - 1645840454049792 cosh[6(316t - x)]-
4913823064550400 cosh[8(316t - x )]- 69778811950848 cosh[1O(316t - x )]-
353575170748800 cosh[6(324t - x )]- 8482134786048 cosh[12(324t - x )]-
6718449497760000 cosh[4(328t - x )]- 166826557497600 cosh[6(332t - x )]-
354442522097664 cosh[8(334t - x)] - 144851259571200 cosh[2(340t - x )]-
438564923596800 cosh[4(340t - x )]- 6165631520848800 cosh[6(340t - x )]-
19084803268608 cosh[10(340t - x )]- 83936853000000 cosh[6(348t - x )]-
25216214229581760 cosh[2(364t - x)] - 2956506448896000 cosh[4(364t - x )]-
381317845708800 cosh[6(364t - x)] - 384820065907200 cosh[8(370t - x)]-

3373899977448000 cosh[6(372t - x)] - 5804890554887424 cosh[4(376t - x)]-
126582878822400 cosh[8(376t - x)] - 3293299379462400 cosh[2(388t - x)]-
151618159300608 cosh[6(396t - x)] - 305918613000000 cosh[4( 400t - x)]-
95067567302400 cosh[8( 400t - x)] - 59640010214400 cosh[8( 406t - x )]-
1275914226787200 cosh[6(412t - x)] - 893282450860800 cosh[6(420t - x)]-
1401142209667200 cosh[2( 436t - x)] - 9477350553775104 cosh[4( 436t - x )]-
571381050240000 cosh[4(448t - x)] - 7302858393600 cosh[8(460t - x)]-
11299191750000 cosh[6(468t - x)] - 368241102028800 cosh[4(472t - x)]-
4239062826014016 cosh[2(484t - x)] - 87875717529600 cosh[4(484t - x)]-
80690064840000 cosh[6(484t - x)] - 97007053651200 cosh[6(492t - x)]-
1824248406236256 cosh[4( 496t - x)] - 16263969163041600 cosh[2(508t - x)]
1281520880640000 cosh[4(508t - x)] - 331783452000000 cosh[4(520t - x )]-
812494863180000 cosh[2(532t - x)] - 36970720617600 cosh[6(532t - x)]-
1173259064577600 cosh[4(544t - x)] - 709273278185472 cosh[4(556t - x)]-
2247025904323200 cosh[2(580t - x)] - 1502683414936704 cosh[2(604t - x )]-
710000121600 cosh[6(612t - x)] - 545363851608576 cosh[4(616t - x )]-
2213927440124400 cosh[2(628t - x)] - 1580764905720000 cosh[2(652t - x )]-
224471126880000 cosh[2(676t - x)] - 12452170500000 cosh[4(688t - x)]-
4954936218248016 cosh[2(724t - x)] - 35702863257600 cosh[4(724t - x)]-
6645601138176 cosh[4(736t - x)] - 565667239737600 cosh[2(748t - x)]-
1080329730480000 cosh[2(796t - x)] - 4089600700416 cosh[4(796t - x)]-
3284939753327232 cosh[2(844t - x)] - 2622379301942400 cosh[2(868t - x)]-
9419178472704 cosh[2(964t - x)] - 875839426862400 cosh[2(988t - x)]-
223311668580000 cosh[2(1012t - x)] - 120869068320000 cosh[2(1036t - x)]-
591849277987968 cosh[2(1084t - x)] - 63008109964800 cosh[2(1228t - x )]-
6973215480000 cosh[2(1372t - x)] - 4998400856064 cosh[2(1444t - x )]-
86451906340800 cosh[12x] - 921387359692800 cosh[8(2t + x )]-
766289987438976 cosh[6(4t + x)] - 461500079040000 cosh[8(8t + x)]-
486101585970000 cosh[6(12t + x)] - 10939261132800 cosh[12(12t + x )]-
68021553600 cosh[18(12t + x)] - 1371314520576 cosh[16(14t + x)]-
412527673958400 cosh[4(20t + x)] - 500457359001600 cosh[8(20t + x)]-
1308982042368 cosh[10(20t + x)] - 5713810502400 cosh[14(20t + x)]-
73735364102400 cosh[12(24t + x)] - 732082129422336 cosh[8(26t + x )]-
826188665702400 cosh[6(28t + x)] - 2834413319337600 cosh[6(36t + x)]-
7313677443072 cosh[12(36t + x)] - 112687848000000 cosh[8(38t + x )]-
56689500067200 cosh[2(44t + x)] -1040997718785024 cosh[4(44t + x)]-
174785616948096 cosh[6(44t + x)] - 380920700160000 cosh[8(44t + x)]-
119482370463456 cosh[10(44t + x)] - 319470049080000 cosh[6(52t + x)]-
3315151046592000 cosh[4(56t + x)] - 298505443332096 cosh[8(56t + x )]-
15069390336 cosh[16(59t + x)] - 856823325758400 cosh[6(60t + x)]-
558320911948800 cosh[8(62t + x)] - 1996996309862400 cosh[4(68t + x)]-
329642913600 cosh[14(68t + x)] - 385874538940800 cosh[6(76t + x)]-
402532261840800 cosh[4(80t + x)] - 1907219714400 cosh[12(80t + x )]-
2490573014721792 cosh[6(84t + x)] - 1562751590400 cosh[12(84t + x )]-
3281778339840000 cosh[4(92t + x)] - 50904857203200 cosh[8(92t + x )]-
311230489667328 cosh[4(104t + x)] - 479960082201600 cosh[6(108t + x)]-
5804960407424640 cosh[2(116t + x)] - 675972544665600 cosh[4(116t + x)]-
805527410688 cosh[8(116t + x)] - 2531657576448 cosh[10(116t + x )]-
92774102252832 cosh[6(124t + x)] - 4056375921998400 cosh[4(128t + x )]-
21976194240000 cosh[8(128t + x)] - 327190543680000 cosh[6(132t + x )]-

51277786560000 cosh[8(134t + x)] - 2777210426944800 cosh[2(140t + x )]-
731367744307200 cosh[4(140t + x)] - 39190461133824 cosh[8(146t + x )]-
14444353123200 cosh[6(156t + x)] - 486861178003200 cosh[2(164t + x )]-
3253525211768832 cosh[4(164t + x)] - 124633509000000 cosh[6(180t + x )]-
1438441804800 cosh[8(182t + x)] - 4716877347000000 cosh[2(188t + x)]-
80103228004800 cosh[6(196t + x)] - 1350883319385600 cosh[4(200t + x)]-
199154946454464 cosh[6(204t + x)] - 4377355143638400 cosh[2(212t + x)]-
1077964967321424 cosh[2(236t + x)] - 641271056025600 cosh[4(236t + x)]-
25716257094528 cosh[6(244t + x)] -76357285804800 cosh[4(248t + x)]-
2483906162774400 cosh[2(260t + x)] - 80121541500000 cosh[4(272t + x)]-
3125509285298400 cosh[2(284t + x)] - 789970928947200 cosh[4(308t + x )]-
53873070451200 cosh[4(320t + x)] - 193390509312 cosh[6(324t + x )]-
8644100901836400 cosh[2(332t + x)] - 189030432374784 cosh[4(344t + x)]-
1738836535822848 cosh[2(356t + x)] - 65539004731200 cosh[4(368t + x )]-
140967296870400 cosh[4(380t + x)] - 1442910297628800 cosh[2(404t + x)]-
649296648000000 cosh[2( 428t + x)] - 3857812216562304 cosh[2( 476t + x )]-
+
4777681708800 cosh[4( 488t x)] - 434366614281600 cosh[2(500t x )]-+
149777752924800 cosh[2(572t + x)] - 839632436042400 cosh[2(620t + x)]-
533142472262400 cosh[2(644t + x)] - 1033760177049600 cosh[2(692t + x)]-
118790120344896 cosh[2(716t + x)] - 719130999787200 cosh[2(764t + x)]-
288389772582912 cosh[2(836t + x)] - 10878216148800 cosh[2(980t + x )]-
35729866972800 cosh[2(1004t + x)] - 119540836800 cosh[2(1340t + x )]-
397207256846400 cosh[2( 4t + 5x)] - 197868158888400 cosh[2(28t + 5x )]-
156250741046400 cosh[2(76t + 5x)] - 12322416178800 cosh[2(124t + 5x )]-
69994178654400 cosh[2(148t + 5x)] - 170053884000000 cosh[2(172t + 5x )]-
2337467932800 cosh[2(364t + 5x)] - 20885233168800 cosh[2(508t + 5x )]-
7499376284400 cosh[2(116t + 7x)] - 1632171340800 cosh[32(133t - x)] ]
[ 6019650 cosh[192t] + 5054400 cosh[336t] + 8494200 cosh [672t]+

490050 cosh[1344t] + 28 cosh[2(1564t - 13x )]+
2100 cosh[2(1460t - 11x)] + 1176 cosh[2(1532t - 11x )]+
271656 cosh[2(1276t - 7x)] + 200200 cosh[2(164t - 5x )]+
1401400 cosh[2(596t - 5x)] + 5775 cosh[4(656t - 5x )]+
900900 cosh[2(668t - 5x)] + 14700 cosh[4(728t - 5x )]+
592900 cosh[2(812t - 5x)] + 3395700 cosh[2(956t - 5x )]+
415800 cosh[2(1172t - 5x)] + 8316 cosh[6(356t - 3x)]+
58212 cosh[6( 436t - 3x)] + 31752 cosh[6( 476t - 3x )]+
576576 cosh[8(22t - x)] + 2690688 cosh[4(28t - x)]+
2270268 cosh[6(28t - x)] + 1716 cosh[14(28t - x )]+
27027 cosh[12(32t - x)] + 350350 cosh[8( 40t - x )]+
1177176 cosh[2(52t - x)] + 412776 cosh[6(68t - x )]+
6006 cosh[16(88t - x)] + 2620800 cosh[4(100t - x )]+
85800 cosh[14(100t - x)] + 1387386 cosh[8(112t - x )]+
567567 cosh[12(112t - x)] + cosh[28(112t - x )]+
378 cosh[24(128t - x)] + 103488 cosh[16(133t - x )]+
7135128 cosh[6(148t - x)] + 5005000 cosh[8(148t - x )]+
577368 cosh[14(148t - x)] + 1091475 cosh[12{160t - x )]+
242550 cosh[16(160t - x)] + 7164612 cosh[2(172t - x )]+
798336 cosh[12(172t - x)] + 247500 cosh[14(172t - x )]+
308 Appendix B. KdV Solitons via. Backlund Transform
24696 cosh[16(178t - x)] + 5297292 cosh[6(188t - x )]+

623700 cosh[12(200t - x)] + 2656192 cosh[8(202t - x )]+
14567553 cosh[4(208t - x)] + 3031182 cosh[8(232t - x )]+
4573800 cosh[6(260t - x)] + 2102100 cosh[4(280t - x )]+
116424 cosh[8(292t - x)] + 1047816 cosh[6(308t - x )]+
6606600 cosh[2(340t - x)] + 1588125 cosh[4(352t - x )]+
2276736 cosh[4(388t - x)] + 8324316 cosh[2( 412t - x )]+
1778700 cosh[2(700t - x)] + 737100 cosh[6(20t + x )]+
4179357 cosh [4(32t + x)] + 9555000 cosh[2(92t + x )]+
396396 cosh[4(152t + x)] + 2774772 cosh[2(308t + x)J+
60984cosh[2(668t + x)] ]2
Appendix C
Derivation of the
Stationary KdV
by G. E. Sarty
Here we derive the stationary Korteweg - deVries equation
A/'1 - ~2/1/'
1 - ~/'"
61 = 0
(C.O.I)
using an asymptotic analysis similar to the one used in Chapter 6. A Math-

ematica program is used this time to help us do some relatively complicated
substitutions and simplifications during the course of a manual derivation. In
theory, a Mathematica program could be written to do all of the substitu-
tions and transformations required in the derivation but since it only needs
to be done once, there is no real saving of effort by doing this. In fact, it
would take more effort because the program would need to be debugged and
some experimentation would be required to get Mathematica to simplify cer-
tain mathematical expressions in the appropriate form. However, writing such
a program might be a useful exercise in learning how to apply Mathematica to
the types of problems we've encountered in these appendices.
309
310 Appendix C. Derivation of the Stationary KdV
We begin with the long wave assumption by assuming that the number
[= (H/L)2 is positive and much smaller than 1. Here, as in Chapter 6, His
the height of the fluid stream and L is the typical wavelength. The coordinate
system (x*, y*) is as defined by Figure 6.1. Since we are interested in the time-
independent and forcing free scenario, O'*(x*) and p*(x*), shown in Figure 6.1,
are both assumed to be zero. The fluid velocities in the x* and y* directions
are given by u and v* respectively.
In the metric system, for example, the gravitational constant 9 has units of
meters/second 2 and H has units of meters. The quantity ..fill consequently
has units of meters/second which are units of velocity. So we can use the
quantity ..fill to nondimensionalize the velocities u* and v*. This is done by
introducing the nondimensionalized velocities u and v as:
u*
u=-- u* = vIfiii u (C.O.2)
..fill
and
v = v*_
[-1/2 _ v* = [1/2 vIfiii v. (C.O.3)
..fill
This choice of non-dimensionalization is chosen because (as will be seen) it
leads to
a1/l a1/l
u = uc ay and v = -u c ax
where U c is the upstream fluid velocity and 1/1 is the normalized stream function
which will be defined below. If we had non-dimensionalized u and v* as
u
u = .JgH u = vlfiiiu
and
v= [1/2 _v*_
.JgH
then we are be led to
a1/l a1/l
u=u c -
ay and v = -[u c ax
instead. Except for that difference, we would otherwise end up with the same
results (i.e. with equations (C.O.4), (C.0.5) and (C.0.6) below).
Our immediate goal is to define a change of variables (x, y) I-t (,1/1), where
( = x and 1/1 = 1/I(x, y), from the non-dimensionalized position variables x
and y to variables defined by the normalized stream function 1/1. The inverse
transformation is denoted in symbols by (,1/1) I-t (x, y) where x = ( and
=
y f(, 1/1). This will transform the fluid domain in the x, y plane into a strip
no = R x [0, 1] in the (,1/1 plane. Then we will use the Euler equations and the
boundary conditions to derive:
311
fIJI" - 2f/(/.pI(.p + (1 + f/[)I.p.p = 0 in no, }

u~(1 + f/l) + [2(1 - 1) - u~JfJ = 0 on 'I/J = 1, (C.O.4)
1=0 on 'I/J = o.
After deriving (C.O.4), let
I = 'I/J + f'I/J II (() + f2 h ((, 'I/J) + ...

and
Uc = c+ fA + ....
Then we can get
(C.0.5)
and
(C.0.6)
The stationary KdV equation (equation (C.O.l)) then follows from equation
(C.0.6) by differentiating it with respect to (.
We now define the change of variables (x, y) H- ((, 'IjJ). Recall that the
partial derivative of the potential function <1>* with respect to x*, <1>;., is the
velocity u* and that v* = ~.. The stream function \):1* is defined as the
harmonic conjugate of the potential function <1>*. That is, <1>* and \):1* are the
real and imaginary parts of an analytic function. So we have, as the Cauchy-
Riemann equations:
(C.0.7)
and
(C.0.8)
Since the units of u* are the same as those of -JiiH (velocity), the units of
\):1* / H are also the same as those of -JiiH. In other words, the units of \):1* are
the same as those of H-JiiH and so we nondimensionalize the stream function
to \):I:
\):1*
\):1= H-JiiH (C.0.9)
The nondimensional coordinates are

y*
y= - :} y* = H y (C.0.I0)
H
and
fl/2
x= --x * (C.O.11)
H
We do not deal with time since we are considering a stationary process.
Now we normalize \):I. Let 17*(X*) describe the free surface. The downstream
condition at x = -00 is then given by 17* ( -00) = H with fluid velocity
u~ == u*(-oo,y*) = \):IZ.(-oo,y*) for all 0::; y* ::; H (C.0.12)

(note that the lower limit for y* follows from the assumption of a flat bottom,
i.e. u*(x*) == 0.) Also, on the top and bottom surfaces, "1)* (x , H) = C1 and
"I). (x ,0) = C2 respectively where C1 and C2 are constants.
Integrating equation (C.0.12) with respect to y* gives:
"I). (-00, y*) = u:y* +K (C.0.13)

where K is an arbitrary constant. Using equation (C.0.13):
at y* =H "1)(-00, H) = u:H + K = C1
at =0 "1)*(-00,0) = K = C2.
So choose K = 0 and deduce that
0<
_ < uc H
"I). _ (C.O.14)
between the top and bottom surfaces. Dividing equation (C.O.14) by H.../ill
and using equation (C.O.9) gives
(C.O.15)
With equation (C.O.2), u; = VgH Uc, then inequality (C.O.15) implies
(C.O.16)
Defining
(C.O.17)
and dividing equation (C.O.16) by Uc yield
05: "p 5: 1,
so "p is the normalized "I).
Using equation (C.O.17) in equation (C.O.9) gives
"1)*
"p = ---== (C.O.1S)
ucHVgH
Now, from equations (C.O.2), (C.O.7), (C.O.1S) and (C.O.10), we have
=> (C.O.19)
From equations (C.O.3), (C.O.S), (C.O.1S) and (c.o.n) we have
1 1
V = _(.-1/2 J::lY "I);.
ygH
= _(.-1/2 J::lYucH
ygu
Viii"pzx z = _(.-1/2 uc H-"pz
(.1/2
H
313
(C.0.20)
Now we can look at the transformation (x, y) I-t ( , w) where ( = x and
w = tP(x, y). Note that we have replaced the notation tP = tP(x, y) with the
more precise w =
tP(x, y). This will avoid confusion when we need to clearly
separate the domain of the inverse transformation from the functions that define
the transformation. The inverse transformation is given by (, w) I-t (x, y)
where x = ( and y = f(', w). So w = tP(x, y) = tP(, f(, w)). Put
<p(, w) = tP(, f(, w)) - w = 0

and differentiate <p implicitly:
with respect to ,: <P, = tPz: + tPyf, =0 (C.0.21)

=
twice w.r.t. (: <P" tPz:z: + tPz:yf, + (tPyz: + tPyyfdh + tPyf" =0
or since tP is harmonic, tPz:z: + 2tPz:yf, + tPyyfl + tPyf" = 0 (C.0.22)
with respect to w: tPyfw - 1 = 0 (C.0.23)
twice with respect to w: tPyyf~ + tPyfww = 0 (C.O.24)
with respect to , then w: tPz:yfw + tPyyfwf, + tPyf,w = 0 (C.0.25)
with respect to w then ,: (tPyz: + tPyyfdfw + tPyfw, O. = (C.0.26)
We will assume that f have second order continuous derivatives so that equa-
tions (C.0.25) and (C.0.26) are the same. Since c)* is harmonic, .6.*c)* =
c);oz:o + c)Zoyo =
0 in the fluid domain. V*C)* =
0 is Euler's equation for
incompressible, irrotational fluid flow. Now
\);-3:- = (-v* )z:o (_f.l/2,fiH v)z:o

(f.l/2,fiH uctPz:)z:o f.l/2JgH uctPz:z:xz:o
1/2 tP
= f. 1/2,fiH
9 uc-U z:z: = f.U cAtPz:z:
where we have used equations (C.O.B), (C.0.3), (C.O.20) and (C.O.ll). Also,
WZoyo (u*)yo = (,fiH u)yo = (,fiH uctPy)yo

= JgH uctPyyyyo = ucAtPyy
where we have used equations (C.0.7), (C.0.2), (C.0.19) and (C.0.10). Hence
wc.j"ftPz:z: + uc~tPyy = 0,
or
, or
(C.0.27)
Substituting equation (C.0.27) into equation (C.0.22) yields
1
- -tPyy
f
+ 2tPxyf( + tPyyf(2 + tPyf" = o. (C.0.2S)
Equation (C.0.24) implies

tP - - tPyy f~ (C.0.29)
y- fww
and equation (C.0.29) in equation (C.0.25) gives
tPxy = tPyy (fj~~w - f( ) . (C.0.30)
Substituting equation (C.0.29) and (C.0.30) into equation (C.0.2S), we get
1.1.
--o/yy + 2.1.o/yy (fwf(w ., ).,
-.,-- - J( J( + .1..,2 tPyyf~., = 0 .
o/yyJ( - --.,-J"
f JWW Jww
This can be simplified to
ff~f" - 2ffwf(wf( + (1 + ff{)fww = o. (C.0.31)
This is the exact differential equation that f((, w) must satisfy.

Now we deal with the boundary conditions. On the bottom w(= tP) = 0
and y = O. Since y = f((, w) we have f((,O) = 0, i.e. the lower boundary
condition for equation (C.0.31) is
= 0 on w = O.
f (C.0.32)
On the top, w = 1 and y* = 7]*(x*) + H or using equation (C.O.lO), Hy =

7]* + H ~ y = ~ + 1. But f((, 1) = y so
*
f((, 1) = ~ +1 !? 7]* = H(J - 1) on w = 1. (C.0.33)
Also on the top, Bernoulli's equation must be used
(C.0.34)
Now, II 'V'*<)* 112= (<);.)2 + (<);.)2 = (W;.)2 + (W;.)2 by equations (C.0.7) and
(C.O.S). Differentiate equation (C.0.1S)
with respect to y*: w;. = ucH../iiitPyyy = uc../iii tPy

where we have used equations (C.0.10) and (C.O.11). So
(W;.)2 = f.U~gH tP; (C.0.35)

315
and
(W;.)2 = u~gH t/J~. (C.O.36)
Substituting equations (C.O.33), (C.O.35) and (C.O.36) into equation (C.O.34)
gives
1 1
2U~gH(ft/J; + t/J~) + g(H + H(I - 1)) = 2gHu~ + gH
? U~(ft/J; + t/J~) + 2(1 - 1) = u~
? u~ (f (-J~ ) + I~) + 2(1 -

2 1) - u~ = 0
? u~(f/t + 1) + 1~2(1 - 1) - I~u~ = 0
? u~(1 + fin + [2(1 - 1) - u~]!~ = 0 on w = 1 (C.O.37)
where we have used equations (C.O.21) and (C.O.23). This is the upper bound-
ary condition for equation (C.O.31). Together, equations (C.O.31), (C.O.32) and
(C.O.37) define the boundary value problem (C.O.4).
Now, as promised, we let
(C.O.38)
and
(C.O.39)
Then we substitute equations (C.O.38) and (C.O.39) into the expressions of
(C.O.4). For this, the following Mathematica code is useful.
(* ******************************************************************* *)
(* * * *)
(* * skdv.m -- IIathematica code 1:or use in deriving * *)
(* * the stationary KdV equation. lote that e is used for * *)
(* * epsilon, z for zeta, 1 1:or lambda and u2 for u squared. * *)
(* * c * *)
(* * * *)
(* * Programmed by: * *)
(* * G. Sarty June 1992 * *)
(* * * *)
(* ******************************************************************* *)
f[z_,v_] := v + e v 1:1[z] + e-2 1:2[z,v] + 0[e]-3
u2 = c-2 + 2 e c 1 + e-2 1-2 0[e]-3
pde := e D[f[z,v] ,v]-2 D[D[f[z,v] ,z] ,z] -
2 e D[f[z,v] ,z] D[f,v] D[D[f[z,v] ,z] ,v] +
(1 + e D[1: [z ,v] ,z]-2) D[D[f[z,v] ,v] ,v] == 0
top := ExpandAll[
u2 (1 + e D[1: [z ,v] ,z]-2) + ( 2(f[z,v] - 1) - u2) D[1:[z.v] .v]-2 == 0]
Using the code we find that (C.O.4) becomes

[w(ft},d() + (h)ww(, W)]f2 + 0(f3) = 0 in no
[ - 2 + 2w] + [6wft () - 4ft () - 2c 2ft ()]f+
[- 4c,Xft() - 2ff() - c 2(ft}2() + 6wff()+
on w = 1 (C.0.40)
212(, w) - 4(h)w(' w) - 2c2(h)w(, w)
+4w(h)w(,w)]f 2 + 0(f3) = 0
on w = 0
The fO terms from the top boundary condition give
- 2 + 2w = 0 on w = 1
which shows only that we have not made any mistakes here. The f1 terms from
the top boundary condition give
which implies that c2 = 1 if ft # O. Using c = 1 and w = 1 in the f2 terms of

that same top boundary condition we get
212(,1) - 2(h)w(' 1) + 3ff() - 4,Xft() = O. (C.0.41)

Finally, the f2 terms of the partial differential equation itself gives:
wf~'() + (h)ww('w) = 0 for (,w) E nO. (C.0.42)

Integrate equation (C.0.42) with respect to w to get
(h)w(' w) = -~f~'()w2 + K 1 ( ) (C.0.43)
where K1 is an arbitrary function of (. Integrate equation (C.0.43) with respect

to w again to get
(C.0.44)
where K2 is another arbitrary function of (. The f2 terms of the bottom

boundary condition give 12(,0) = 0 so K 2 ( ) = 0 when this information is
put into equation (C.0.44). Thus, equations (C.0.43) and (C.0.44) become
12(,1) = -!ff'() + K 1 (), }

(C.0.45)
(h)w(' 1) = -!ff'() + K1()'
Substituting (C.0.45) into (C.0.41) gives:
-~ff'() + 2K1 () + ff'() - 2K1 () + 3ff() - 4,Xft () = 0
and this simplifies to equation (C.0.6). As was previously mentioned, equation
(C.0.6) can be differentiated once to yield the stationary Korteweg-deVries
equation (C.O.l) and our derivation is complete.
Bibliography
[1] M. J. Ablowitz and H. Segur (1981), Solitons and Inverse Scat-

tering Transform, SIAM, Philadelphia, Pennsylvania. [4] (This [4]
means that this entry is cited in Chapter 4. The same rule applies to
other entries in this bibliography.)
[2] T. R. Akylas (1984), On the excitation oflong nonlinear water waves by

a moving pressure distribution, J. Fluid Mech. 141,455-466. [6]
[3] P. G. Baines (1987), Upstream blocking and airflow over mountains, Ann.
Rev. Fluid. Mech. 19, 75-97. [6]
[4] A. Barone (1974), Josephson Effect: Achievements and Trends, World

Scientific, Singapore. [7]
[5] A. Barone, F. Esposito, C. J. Magee and A. C. Scott (1971), Theory and

applications of the sine-Gordon equation, Rivista Del Nuovo Cimento 1,
227-267. [7]
[6] T. B. Benjamin (1957), Wave formation in a laminar flow down an in-

clined plate, J. Fluid Mech. 2, 554-574. [5]
[7] T. B. Benjamin (1967), Instability of periodic wavetrains in nonlinear

dispersive systems, Proc. Roy. Soc. Lond. A 229, 59-75. [8]
[8] T. B. Benjamin and J. F. Feir (1967), The disintegration of wave trains

on deep water, Part 1. Theory, J. Fluid Mech. 27,417 - 430. [8]
[9] A. Bishop and T. Schneider (1978), Solitons and Condensed Matter

Physics, Springer-Verlag, New York. [7]
[10] G. Bricogne (1984), Maximum entropy and the foundations of direct

methods, Acta Cryst. A 40, 410-445. [9]
[11] R. Camassa and T. Y. Wu (1991), Stability offorced solitary waves, Phil.

Trans. R. Roy. Lond. A 337,429-466. [6]
[12] S. Chandrasekhar (1981), Hydrodynamic and Hydromagnetic Stability,

Dover Publications, New York. [8]
317
318 Bibliography
[13] H. H. Chen (1974), General derivation of Biicklund transformations from

inverse scattering problems, Phys. Rev. Lett. 33, 925-928. [4]
[14] J. D. Cole (1951), On a quasilinear parabolic equation occurring in aero-
dynamics, Q. J. Math. 9, 225-236. [5]
[15] D. Coles (1965), Transition in circular Couette flow, J. Fluid Mech. 21,
385-425. [8]
[16] A. D. D. Craik (1985), Wave Interactions and Fluid Flows, Cambridge
University Press, New York. [9]
[17] R. K. Dodd, J. C. Eilbeck, J. D. Gibbon and H. C. Morris (1982), Solitons
and Nonlinear Wave Equations, Academic Press, New York. [4, 7]
[18] P. G. Drazin (1983), Solitons, Cambridge University Press, London. [4]
[19] P. G. Drazin and R. S. Johnson (1989), Solitons: An Introduction, Cam-
bridge University Press, New York. [4, 7]
[20] P. G. Drazin and W. H. Reid (1981), Hydrodynamic stability, Cambridge
University Press, New York, Chapters 1, 2 and 3. [8]
[21] A. Erdelyi (1956), Asymptotic Expansions, Dover, New York, Chapter 1.
[1]
[22] J. C. Gallop (1991), SQUIDS, the Josephson Effects and Superconducting
Electronics, Adam Hilger, New York. [7]
[23] C. S. Gardner, J. M. Greene, M. D. Kruskal and R. M. Miura (1974),
Korteweg-de Vries equation and generalizations. VI. Methods for exact
solution, Comm. Pure Appl. Math. 27, 97-133. [4]
[24] I. M. Gel'fand and B. M. Levitan (1955), On the determination of a
differential equation from its spectral function, Amer. Math. Soc. Transl.
Ser. 2, 253-304. [4]
[25] R. Grimshaw (1987), Resonant forcing of barotropic coast ally trapped
waves, J. Phys. Oceanagr. 17,53-65.
[26] R. H. J. Grimshaw and N. Smyth (1986), Resonant flow of a stratified
fluid over topography, J. Fluid Mech. 169,429-464. [6]
[27] M. E. Gurtin (1975), On the breaking of water waves on a sloping beach
of arbitrary shape, Q. Appl. Math. 33, 187-189. [3]
[28] J. 1. Hammack and H. Segur (1974), The Korteweg-de Vries equation
and water waves. Part 2. Comparison with experiments, J. Fluid Mech.
65, 289-314. [3, 4]
[29] H. Hasimoto and H. Ono (1972), Nonlinear modulation of gravity waves,
J. Phys. Soc. Japan 33, 805-811. [7]
Bibliography 319
[30] H. Hauptman (1975), A joint probability distribution of seven structure

factors, Acta Cryst. A 31, 671-679. [9]
[31] H. Hauptman (1975), A new method in the probabilistic theory of the

structure invariants, Acta Cryst. A 31, 680-687. [9]
[32] H. Hauptman (1983), The phase problem of x-ray crystallography, Proc.

Indian Acad. Sci. (Chern. Sci.) 92, 291-321. [9]
[33] H. Hauptman (1985), Phase problems of X-ray crystallography, in "En-

cyclopedia of Statistcal Sciences", Vol. 6, ed. by Kotz and Johnson, John
Wiley & Sons, New York, pp. 702-709. [9]
[34] E. Hopf (1950), The partial differential equation Ut + UU", = jlU",,,,, Com-
mun. Pure Appl. Math. 3, 201-230. [5]
[35] E. Infeld and G. Rowlands (1990), Nonlinear Waves, Solitons and Chaos,
Cambridge University Press, New York, Chapter 1. [8]
[36] A. Jeffery and T. Kakutani (1970), Stability of the Burgers shock wave
and the Korteweg-de Vries soliton, Indiana Univ. Math. J. 20, 463-468.
[5]
[37] R. S. Johnson (1973) , On the development of a solitary wave moving
over an uneven bottom, Proc. Camb. Phil. Soc.73, 183-203. [4]
[38] D. J. Kaup and P. J. Hansen (1986), The forced nonlinear Schrodinger

equation, Physica D 18, 77-84. [7]
[39] J. B. Keller and S. I. Rubinow (1960), Asymptotic solution of eigenvalue

problems, Ann. Phys. 9, 24-75. [1]
[40] P. K. Kundu (1990), Fluid Mechanics, Academic Press, New York, Chap-
ter 11. [8]
[41] T. Kuusela, J. Hietarinta, K. Kokko and R. Laiho (1987) , Soliton exper-

iments in a nonlinear electrical transmission line, Eur. J. Phys. 8, 27-33.
[4]
[42] G. L. Lamb (1980), Elements of Soliton Theory, John Wiley, New York.
[4, 7]
[43] H. Lamb (1945), Hydrodynamics, 6th ed., Dover Publications, New York.
[6, 8]
[44] L. Landau and E. Lifschitz (1958), Quantum Mechanics, Nonrelativistic
Theory, Pergamon Press, New York. [4]
[45] P. D. Lax (1968), Integrals of nonlinear equations of evolution and solitary

waves, Comm. Pure Appl. Math. 21, 467-490. [4]
320 Bibliography
[46] S. J. Lee, G. T. Yates and T. Y. Wu (1989), Experiments and analyses

of upstream-advancing solitary waves generated by moving disturbances,
J. Fluid Mech. 199, 569-593. [6]
[47] C. C. Lin (1955), The Theory of hydrodynamic stability, Cambridge Uni-
versity Press, New York. [8]
[48] W. F. Lucas (1978), Models in Applied Mathematics, Springer-Verlag,
New York, Vol. 1. [2]
[49] D. Ludwig (1983), Parsimonious asymptotics, SIAM J. Appl. Math. 43,
664 - 672. [1]
[50] J. Marsden and T. J. R. Hughs (1983), Mathematical Foundation of

Elasticity, Prentice-Hall, New York, Chapter 2. [2]
[51] A. McPherson (1982), Preparation and Analysis of Protein Crystals, John
Wiley & Sons, New York Chapters 5, 7, and 10. [9]
[52] L. F. McGoldrick (1965), Resonant interactions among capillary-gravity
waves, J. Fluid Mech. 21, 305-331. [9]
[53] R. E. Meyer (1982), Introduction to Mathematical Fluid Dynamics,
Dover, New York, Chapters 1 - 3, and Chapter 6. [2]
[54] J. W. Miles (1986), Stationary, transcritical channel flow, J. Fluid Mech.
162, 489-499. [6]
[55] R. M. Miura (1976), The Korteweg-de Vries equation: A survey of results,
SIAM Review 18, 412-459. [4]
[56] L. F. Mollenauer and R. H. Stolen (1982), Solitons in optical fibers,
Fiberoptic Technology, April, 1982, 193-198. [7]
[57] G. H. Nayfeh (1973), Perturbation Methods, John Wiley, New York,
Chapters 1 - 4. [1]
[58] A. H. Nayfeh and D. T. Mook (1979), Nonlinear Oscillations, John Wiley,
New York, Chapter 4. [1]
[59] A. C. Newell (1985), Solitons in Mathematics and Physics, SIAM,
Philadelphia, Pennsylvania. [4]
[60] F. W. J. Olver (1974), Asymptotics and Special Functions, Academic
Press, New York. [1]
[61] Shih- I Pai (1956), Viscous Flow Theory, I - Laminar Flow, D. van Nos-
trand Co. Inc., New York. [5]
[62] R. D. Parmentier (1978), Fluxions in long Josephson junction, in "Soli-
tons in Action" ed. by K. Lonngren and A. Scott, Academic Press, New
York, pp. 173 - 199. [7]
Bibliography 321
[63] O. M. Phillips (1967), Theoretical and experimental studies of gravity

wave interactions, Proc. Roy. Soc. A 299, 104 - 119. [9]
[64] O. M. Phillips (1974), Wave interactions, in "Nonlinear Waves" ed. by

S. Leibovich and A. R. Seebass, Cornell University Press, Ithaca, New
York, pp. 188-211. [9]
[65] C. Rebbi and G. Soliani (1984), Solitons and Particles, World Scientific
Publishing Co., Singapore. [4, 7]
[66] M. Sachs (1963), Solid State Theory, McGraw-Hill, New York, Chapters
2 and 4. [9]
[67] D. Saint-James, E. J. Thomas and G. Sarma (1969), Type II Supercon-

ductivity, Pergamon Press, Toronto. [7]
[68] M. C. Shen and R. E. Meyer (1963), Climb of a bore on a beach, J. Fluid

Mech. 16, 113-126. [3]
[69] M. C. Shen and S. M. Sun (1987), Critical viscous surface waves over an
incline, Wave Motion 9, 323-332. [5]
[70] S. S. Shen (1989), Disturbed critical surface waves in a channel of arbi-

trary cross section, J. Appl. Math. Phys. (ZAMP) 40, 216-229. [6]
[71] S. S. Shen (1990), Blocking of solitary pulses in a nonlinear fiber, Wave

Motion 12, 551-557. [7]
[72] [14] S. S. Shen (1991), Locally forced critical surface waves in channels
of arbitrary cross section, J. Appl. Math. Phys. (ZAMP) 42, 122-138. [6]
flows, J. Fluid Mech. 234,583 - 612. [1,6]
[74] S. S. Shen and M. C. Shen (1990), A new equilibrium of subcritical flow

over an obstruction in a channel of arbitrary cross section, Euro. J. Mech.
B/Fluids 9,59-74. [6]
[75] J. Smoller (1983), Shock Waves and Reaction-Diffusion Equations,

Springer-Verlag, New York, Chapters 15-18. [2]
[76] J. J. Stoker (1957), Water Waves: the Mathematical Theory with appli-
cations, Interscience, New York. [3]
Stanford, California. [2, 8]
[78] H. D. Wahlquist and F. B. Estabrook (1973) , Backlund transformation

from solitons of the Korteweg-de Vries equation, Phys. Rev. Lett. 31,
1386-1390. [4]
322 Bibliography
York, Part I. [2, 3, 5, 6, 7]
[80] M. M. Woolfson (1987), Direct methods - from birth to maturity, Acta
Cryst. A 43, 593-612. [9]
[81] T. Yao-Tsu Wu (1987), Generation of upstream advancing solitons by
moving disturbances, J. Fluid Mech. 184, 75-100. [6]
[82] C. S. Yih (1963) , Stability of liquid flow down an inclined plate, Phys.
Fluids 6, 321-334. [5]
[83] C. S. Yih (1988), Fluid Mechanics, West River Press, Ann Arbor, Michi-
gan, Chapter 9. [8]
[84] N. J. Zabusky and M. D. Kruskal (1965), Interaction of "solitons" in a
collisionless plasma and the recurrence of the initial states, Phys. Rev.
Lett., 15, 240-243. [4]
Index
Ablowitz, 122, 317 Bragg, 262

Akylas, 186, 317 Bragg's law, 259
Ampere's law, 209 breaking time, 43
antisoliton, 200 breather solution, 202
asymptotic approximation, 5 Bricogne, 275, 317
best, 4 bump, 147
parsimonious, 4 buoyancy force, 231
asymptotic expansion, 1,7,127,210 Burgers equation, 129
matched, 10 Burgers shock wave, 131, 135, 141
singular, 7
uniform, 7 Camassa, 186
asymptotic sequence, 6 carrier wave, 206
singular, 7 Cauchy-Riemann equations, 106
uniform, 7 cell, 262
asymptotic solution, 10, 14 centrifugal force, 237
atomic number, 265 Chan-Kerkhoven scheme, 174
Chandrasekhar, 220, 241, 246, 317
Backlund transform, 105 characteristic method, 35, 39
Benard problem, 231 characteristics, 36, 39
Baines, 186, 317 Chen, 115, 121,318
Barone, 216, 217, 317 cnoidal wave, 164
basic state, 233 Cole, 145, 318
beach, 62 Cole-Hopf transformation, 131
bell shape, 202 Coles, 246, 318
Benjamin, 144, 243,246,317 collision process, 200
Benjamin-Feir instability, 255 commutator, 103
Bernoulli equation, 56, 149 complete elliptic integral, 167
Bessel function, 271 complex conjugate, 227
bifurcation, 239 conjugate equation, 107
bifurcation diagram, 19, 161 conservation
Bishop, 216, 317 mass, 30
boundary conditions, 54 momentum, 31
dynamical, 54 conservation law, 25, 35
geometrical, 54 hyperbolic, 35
kinematic, 56 constitution relation, 32
boundary layer, 8, 125 continuity equation, 31
Boussinesq equation, 68, 251 Couette flow, 222
323
324 Index
Coulomb's law, 209 subcritical, 162

coupled pendulums, 190 supercritical, 153
Craik, 258, 275, 318 transcritical, 168
crystal, 258 Forsyth, 131
unit cells, 263 four-wave interactions, 255
crystal dislocation, 191 Fourier representation, 248
crystallography, 258 Fourier pair, 265
Fourier transform, 132, 174
damping force, 12 free surface, 34
decaying, 14
deformation rate, 32 Gallop, 217, 318
depression, 168 Gardner, 75, 121, 318
diffraction index, 209 Gaussian pulse, 206
diffraction pattern, 259 Gel'fand, 83, 115, 121,318
diffraction plane, 259 Gibbon, 121, 216, 318
dimension analysis, 137 Ginzburg-Landau, 76
direct method, 262 gratings, 263
dispersion relation, 65, 251 Greene, 75, 121, 318
dispersive waves, 64 Grimshaw, 186,318
Dodd, 76,121,216,318 group velocity, 208, 210
Drazin, 76, 121, 217, 220,226, 246, Gurtin, 74, 318
318
Hamilton-Jacobi, 65
eikonal equation, 65 Hamiltonian operator, 80
Eilbeck, 121, 216, 318 Hammack, 74, 121, 318
electromagnetic waves, 206 Hansen, 216, 319
electron density, 248 harmonic function, 55
energy transfer, 252 harmonics, 14
entropy, 46 Hasimoto, 216, 318
entropy condition, 48 Hauptman, 262, 270, 275, 319
Erdelyi, 23, 318 heat equation, 132
Esposito, 216, 317 Hietarinta, 122, 319
Estabrook, 105, 110, 121, 321 Hopf, 145, 319
Euler equations, 32, 54 Hughs, 51, 320
Eulerian, 29 hydraulic fall, 164, 167
evolution equation, 102 hydrostatic, 61
external pressure, 147 hyperbolic, 25
hyppac.m, 283
Farady's law, 209
Feir, 243, 246, 317 inductance, 195
Fermi-Pasta-Ulam,76 Infeld, 246, 319
fission condition, 120 inner solution, 9
flow instability, 219 intensity, 262
flux, 26 intensity data, 265
forced Korteweg-de Vries, 147 interaction equations, 252, 257
locally, 154 interface, 142, 143
non-locally, 154 inverse scattering, 82
Index 325
Jacobian, 29 Lipschitz condition, 166

Jeffery, 142, 145, 319 Lucas, 51, 320
Johnson, 76, 121,217,318,319 Ludwig, 23, 320
Josephson junction, 191, 194
Josephson penetration length, 195 Magee, 216, 317
Josephson plasma frequency, 195 magnetic monopoles, 209
marginal stability, 241
Kakutani, 142, 145, 319 Marsden, 51,320
Karle, 262, 270 mass conservation, 28
Kaup, 216, 319 Mathematica, 2, 115, 175,277
KdV hierarchy, 102, 105 Maxwell equations, 209
Keller, 4 McGoldrick, 254, 275, 320
Kelvin-Helmholtz instability, 225 McPherson, 275,320
kinematic condition, 34 Meyer, 51, 74,320, 321
kinematic viscosity, 124 Michell, 243
kink, 198 microwave frequency, 193
Kirchoff's law, 195 Miles, 156
Klein-Gordon, 65 Miller indices, 263
Kokko, 122, 319 minimization formulation, 267
Korteweg-de Vries, 65, 82 Miura, 75, 98, 121, 318, 320
equation, 69 modes, 249
Kruskal, 75, 76, 121, 318, 322 newer, 249
Kundu, 236, 246,319 primary, 249
Kuusela, 122, 319 modulated wave, 211
Mollenauer, 208,216,320
L-C circuits, 195 Mook, 23, 320
Lagrangian coordinates, 29 Morris, 121, 216, 318
multiple scales, 11
Laiho, 122, 319
Lamb, 78, 121, 186, 216, 246, 319 natural oscillation, 15
laminar flow, 223 Navier-Stokes equations, 32, 124, 239
Landau, 121, 319 Nayfeh, 23, 320
Laplace equation, 56, 106, 226 neighborhood principle, 270
lattice, 259 Newell, 76, 121, 320
reciprocal, 264 Newtonian fluid, 32, 143
Laue spots, 259 no slip condition, 124
Lax, 121, 319 nonlinear Schrodinger, 76, 189, 211
Lax equation, 103 forced, 212
Lax pair, 102 nonlinear Schrodinger, 65
Lee, 187, 320 normal mode, 223
Legendre equation, 89 numerical noise, 183, 220
Legendre polynomials, 89
Levitan, 83, 115, 121, 318 Olver, 23, 320
Lifschitz, 121, 319 one-step state, 200
Lin, 246, 320 Ono, 216, 318
linearized stability, 225 operating curve, 19
Liouville equation, 107 oscillator, 12
326 Index
harmonic, 249 Rubinow,4

nonlinear, 18 Russell,76
stationary, 19
outer solution, 9 Sachs, 275, 321
Saint-James, 217, 321
Pai, 145, 320 Sarma, 217, 321
paradox, 156 scales, 14
Parmentier, 217, 320 hidden, 15
partial sum, 3 spatial, 14
pat phenomenon, 11 temporal, 14
phase, 210 scattering, 81
phase difference, 194 scattering data, 78
phase problem, 262, 265 scattering method, 78
direct method, 267 scattering potential, 78
neighborhood principle, 270 Schneider, 216, 317
probabilistic distribution, 267 Schrodinger equation, 80
phase velocity, 65, 66 Scott, 216, 317
Phillips, 254, 275, 321 secular terms, 14, 252
Planck's constant, 80 Segur, 74, 121, 122, 317, 318
plane waves, 210, 247 self-focusing, 209
polarization vector, 209 seminvariant, 270
potential flow, 54 separation of variables, 14, 91
potential theory, 255 nonlinear, 14
Prandtl number, 234 shallow water equations, 60
probability distribution, 270 shallow water waves, 62
pulse broadening, 206 Shen, 4, 74, 144, 186,217,321
quantum mechanics, 79 shock, 45
quartet invariant, 273 rarefaction, 47
strength, 48
Rankine-Hugoniot condition, 34, 46, thickness, 51
49 shock path, 46
rarefaction, 48 shoreline, 63
Rayleigh, 230, 237 side band frequencies, 245
Rayleigh instability, 231 side band instability, 245
Rayleigh number, 232 signal broadening, 208
critical, 233 Silicon Graphics, 2
Rebbi, 76, 121, 216,321 similarity solution, 137
reflection coefficient, 77 simple wave, 47
Reid, 220, 226, 246, 318 sine-Gordon, 76, 107
resonance, 14 sine-Gordon equation, 189
resonance condition, 248, 250 sinusoidal, 20
Reynolds number, 125 Smoller, 51, 321
critical, 129 Smyth, 187,318
Riccati equation, 83, 132 Soliani, 76, 121, 216, 321
Rodriques formula, 89 solipac.m, 284
Rowlands, 246, 319 solitary wave, 153
Index 327
stability, 183 . traveling wave, 28, 36

soliton, 76 triangular wave, 131, 137, 140
N-soliton, 85 triplet invariant, 270
I-soliton, 87 tunneling current, 193, 194
2-soliton, 88 turbulent flow, 223
3-soliton, 93 two-step state, 200
fission, 119
KdV, 85 ultraviolet radiation frequency, 193
ladder, 110 uncertainty principle, 79
N-soliton, 97
van der Pol, 15
radiated, 177
Van Dyke, 45, 51, 246, 321
spectral scheme, 178
viscosity, 32, 124
spectrum, 75
vortex sheet, 229
discrete negative, 99
vorticity, 54
positive continuous, 99
vorticity equation, 55
SQUIDS, 193
stability, 20 Wahlquist, 105, 110, 121, 321
standard KdV, 82, 88, 102, 103 wake, 99, 168
state equation, 26 wave function, 79
Stoker, 74, 321 wave number, 227
Stokes waves, 242 white noise, 229
Stolen, 208, 216, 320 Whitham, 51, 74, 133, 145, 186,216,
stratification, 55 322
stress tensor, 142, 143 Woolfson, 275, 322
structure factors, 265 Wu, 168, 186, 322
structure invariants, 269
conditions, 269 X-ray crystallography, 262
Sun, 144,321
SUN Sparcstation, 284 Yang-Mills, 76
superconductor layer, 194 Yates, 187,320
superposition principle, 110 Yih, 144, 246, 322
nonlinear, 108
surface tension, 143, 231 Zabusky, 76,121,322
tangent formula, 271

Taylor expansion, 60
Taylor number, 238
critical, 242
Taylor problem, 237
tension coefficient, 142
thermal instability, 231
Thomas, 217, 321
traffic flow, 26
transcritical flows, 167
transmission coefficient, 77
transport theorem, 29
Nonlinear Topics in the Mathematical Sciences
An International Book Series dealing with Past, Current and Future Advances and
Developments in the Mathematics of Nonlinear Science
1. M.S. Berger: Mathematical Structures of Nonlinear Science. An Introduction.

1990 ISBN 0-7923-0728-3
2. P.G. Bakker: Bifurcations in Flow Patterns. Some Applications of the
Quantitative Theory of Differential Equations in Fluid Dynamics. 1991
ISBN 0-7923-1428-X
3. S.S. Shen: A Course on Nonlinear Waves. 1993 ISBN 0-7923-2292-4
KLUWER ACADEMIC PUBLISHERS - DORDRECHT I BOSTON I LONDON

Linear and Nonlinear Waves Textbook)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Linear and Nonlinear Waves Textbook)

Uploaded by

Copyright:

Available Formats

A Course on Nonlinear Waves

Nonlinear Topics in the Mathematical Sciences

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

Printed on acid-free paper

No part of the material protected by this copyright notice may be reproduced or

3.3 Dispersive Water Waves . . . . . . . . . . . . . . . . 64

B.2 Two Solitons 289

The aim of this book is to give a self-contained introduction to the mathe-

between the consequences of linear conservation laws with those of nonlinear

Asymptotic expansion is a method that approximately evaluates an integral

The asymptotic expansion technique is widely used in physics, chemistry

1.1 Concepts of Asymptoticity

1.1.1 Is a divergent series useful?

It is clear that the integral

G(lO) = 0.091563 ...

II Z 1.0 5.0 10.0 20.0 30.0 II

II Z 40.0 50.0 60.0 100.0 300.0 II

This amazing phenomenon implies that even though a series is divergent,

where the nth partial sum Sn (x) is

Sn (x) = .! - ~2 + ~3 - ... + (-1 t- d n - I)! . (1.1.6)

1.1.2 The symbols "', 0 and 0

f(x) '" g(xj (x -+ a).

In the situation of no possible ambiguity, we simply write

We say that f(x) is an asymptotic approximation to g(x) as x is near a

c) ~ '" X~l (x -+ 00);

d) O(g)o(g) = o(l)o(g) = o(lg) if 9 = 0(1);

1. Show that if i = O( 1/Ji), i = 1,2, ... , n, and ai are constants, then

2. Show that if i = O( 1/Jd, i = 1,2, ... , n, then TIi=l i = 0 (TIi=l1/Ji)

1.1.3 Asymptotic sequences

we say that {tPn(Xj En is a uniformly asymptotic sequence in the domain R.

Example 1: The following sequences are all asymptotic sequences.

d) {E~=l(-I)k-dk~l)!}, (x --+ 00).

Example 2: {Ene- lxl } (E --+ 0) is a uniformly asymptotic sequence in R.

We say that E:=l !k(X) is an asymptotic expansion of !(x) as x --+ a if

Figure 1.1: Asymptotic solution of the boundary layer problem (1.1.9) -

particularly useful in finding asymptotic solutions for differential equations.

The geometrical or physical meaning of such a transform is the application of

Yc = Yo + Yi - Ym = ax + (1- a)(l - e-~). (1.1.17)

The difference between Yc and the exact solution Y given by (1.1.12) is

Iy - Ycl = O(f). (1.1.18)

By definition, Yc is indeed an asymptotic approximate solution to the problem

1.2 Method of Multiple Scales

sinti - W2 t + (Pi - <p2)

modulates the faster oscillation

sin(Wi +W2 t + <Pi + <P2).

Figure 1.2: Pat phenomenon: superposition of two fast oscillations of similar

1.2.2 Description of the method of multiple scales

x+ x = -2f.i:, 0 < f 1, t > O. (1.2.1)

x = ae-tcos(~ t + 8). (1.2.2)

-3 0 50 100 150 200

:i: ZT1 + eZT2'

ZT1T1 + Z + 2e(zT1T2 + ZT1) + e2(zT2T2 + 2ZT2) = O. (1.2.6)

Assume (1.2.6) has an asymptotic expansion of the form

Z = zo(T1 , T2) + eZ1(T1, T2) + e2z2(T1, T2) + .... (1.2.7)

Order eO: ZOT1T1 + Zo = 0, (1.2.8)

Equation (1.2.8) has general solution: .

Zo = A(T2) cos(T1 + (T2)) (1.2.10)

Inserting (1.2.10) into (1.2.9), we have

Z1 T1TI + Z1 = 2 [ A' (T2) sin(T1 + (T2)) + A(t2)' (T2) COS(T1 + (T2))

A' (T2) sin(T1 + tP(T2)) + A(T2)tP' (T2) COS(T1 + tP(T2))

This further implies

1.2.3 The van der Pol oscillator

Substituting (1.2.17)-(1.2.20) into (1.2.16) and equating the coefficients of like