Professional Documents
Culture Documents
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
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
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
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
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
Bibliography 317
Index 323
Preface
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
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
1
2 Chapter 1. Asymptotic Expansion
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
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.
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. Concepts of Asymptoticity 5
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);
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);
Exercises:
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
8 Chapter 1. Asymptotic Expansion
ym J-==-""H-~
(1-a}(1-1/e)+O( E)
o~~-*--------------------------~-------x
1
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= ~.
(
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
Then
If IWi - w21 is small and both IwtI and IW21 are large, then
Namely,
xl
x2
xl + x2
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
12 Chapter 1. Asymptotic Expansion
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.
t, (1.2.3)
1.2. The Method of Multiple Scales 13
-1
-2
T2 = t, (1.2.4)
Z z(T1, T 2, e). (1.2.5)
Then
Substituting (1.2.7) into (1.2.6) and assembling the terms accordipg to like
powers of e, we obtain the following equations
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,
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
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)
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)
16 Chapter 1. Asymptotic Expansion
(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
<PT, 0, (1.2.28)
1 1 2
aT, -(1-
2 -a
4 )a . (1.2.29)
Hence
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)
(1.2.34)
1.2. Method of Multiple Scales 17
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)
(1.2.38)
Thus
(1.2.39)
or,
1
<P = -16T2 + <Po. (1.2.40)
(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
18 Chapter 1. Asymptotic Expansion
(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).
To t, Tl =d,
x = x(To, T 1; () = x(O)(To, TI) + (x(l)(To TI) + O({2).
Then
Order {a:
(1.2.48)
Order (l:
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
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)
P
I 1 3 + -p
+ -p Jl.
= . (A
-hsm uTl -
R
/J -
A.)
'I'
2 2 4w '
ptf>' = - 4~ cos( OTl - (J - tf> )
(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.
20 Chapter 1. Asymptotic Expansion
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
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)
1.2. Method of Multiple Scales 21
x
6
4
2
-2
-4
-6
x
6
4
2
-2
oot
-4
-6
(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)
22 Chapter 1. Asymptotic Expansion
-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
and
(1.2.66)
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?
1.2. Method of Multiple Scales 23
[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.
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.
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
d
-d
t
l
x,
x
p(x, t)dx = q(x!) - q(X2). (2.1.1)
Pt + qx = O. (2.1.2)
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
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
~ f P dO = - f q. ds.
vt Jn Jan
The divergence theorem implies that
l (~: q) + \1 . dO = O. (2.1.13)
2.1. Conservation Laws 29
(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.
(2.1.15)
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.
30 Chapter 2. Hyperbolic Waves
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)
o=~
Dt
1
v.
Pdv=l
v.
(DP +PV.V)dV.
Dt
(2.1.17)
2.1. Conservation Laws 31
normal
traction
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&~
32 Chapter 2. Hyperbolic Waves
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)
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,
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
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).
2.1. Conservation Laws 33
Front L
<I> dO + f
J'E(t)nO(t)
r
[<I> (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 <I> through a~, and [<I>H = <I>(x E I; + 0, t) - <I>(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 [<I>(v - u) . n + Q . n] 2 ds = 0. (2.1.27)
J'Eno 1
(2.1.31)
(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
Ut + f",(u) = O. (2.2.1)
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.
36 Chapter 2. Hyperbolic Waves
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)
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.
2.2. Characteristic Method 37
Example 1:
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)
u(x, t) = uo(x - at)e- bt + lot 1 (x - a(t - (7), (7)) e-b(t-u) du. (2.2.15)
(2.2.16)
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:
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:
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
2.2. Characteristic Method 39
Characteristics
~---L-----L----~----------------------x
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)
If
1
Uo (x) = -1--2 '
+x
then
1
u (x t) = ----=-__=_ (2.2.24)
, 1 + x e-
2 2t
1
0.75
U(x,th.5
0.25
o
t
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)
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)
Making a transform
Hence
U(x, t) = uo(x - t), v(x, t) = vo(x - 3t). (2.2.32)
Finally,
(~ )
_ ~ ( 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)
P(X,t), t > 0
Along a characteristic C,
(2.2.43)
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
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)
44 Chapter 2. Hyperbolic Waves
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'
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.
2.2. Characteristic Method 45
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)
< 0,
x>
1, x
'(', t = 0) = .0(') = { 1- x, :5 x :5 1, (2.2.52)
0, 1.
(2.2.53)
46 Chapter 2. Hyperbolic Waves
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)
0, z < 0,
Ul = { 0,
1,
and U2 = { f, 0< x < t, (2.2.60)
1, z > t.
2.2. Characteristic Method 47
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
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)
48 Chapter 2. Hyperbolic Waves
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.
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
2.2. Characteristic Method 49
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.
(2.2.68)
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)
t
P2 P2
l.
P1 P1
t = 0 ~ t> 0
I
f x
Shock thickness
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
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
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
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),
56 Chapter 3. Water Waves
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)
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)
A Air pressure Pa
u
Air pressure s
li
Figure 3.2: Application of the Bernoulli theorem to the draining water problem.
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
T=J 2 H meg
(3.1.16)
58 Chapter 3. Water Waves
(3.2.1)
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
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)
uh~ + v = o. (3.2.14)
(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)
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
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
3.2. Shallow Water Equations 61
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,
Under the shallow water assumption, the continuity equation can be re-
placed by the conservation of mass flux
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
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 )
~ Front
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.
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;
3.2. Shallow Water Equations 63
(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)
c2 ( Uxx )- - ( 2hx
Utt )- + -
2
a + -3a = O. (3.2.41)
c c
Exercise: Derive equation (3.2.41).
By
db _ 3h x b _ ~ = 0 (3.2.44)
dx 4h 2h .
This first order ODE can be solved by variation of constants. Finally,
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.
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.
(3.3.4)
(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.
66 Chapter 3. Water Waves
9
H Domain of water
+
Flat bottom
Figure 3.4: Two-imensional shallow water waves in an open ocean of flat bot-
tom.
~ = 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.
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, ....
68 Chapter 3. Water Waves
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)
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)
(3.3.32)
(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)
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.
70 Chapter 3. Water Waves
(3.3.37)
(3.3.38)
Since
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.
where A, a and p are real constants. We look for its traveling solutions. Let
Then
(A - c)u' + 2auu' + PU"' = O. (3.3.43)
The first integral of (3.3.43) gives
(3.3.45)
3 P ,)2 _ 3 3 2 3C 3D _
--(u --U--(A-C)U --u--=P(u). (3.3.46)
2a 2a a a
3.3. Dispersive Water Waves 71
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)
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)
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)
a1 (u') 2
=u(2-u)(u-2+3). (3.3.53)
So
(3.3.54)
1.
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)
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.
3.3. Dispersive Water Waves 73
o
Figure 3.6: Solitary wave solution of the KdV.
P(u)
Solitary
/ wave Unbounded
solution
wave
region
u
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.
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
String
a)
Figure 4.1: (a) Mass-string-spring scattering system; (b) forces acting on the
mass m.
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)
Y = 1+ R= T. (4.1.5)
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.
x 'I/; , (4.1.9)
80 Chapter 4. Scattering and Inverse Scattering
: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)
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)
T= l+R. (4.1.21)
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)
ikh2
2m (1 - R - T) + QT = O. (4.1.23)
82 Chapter 4. Scattering and Inverse Scattering
T (4.1.24)
ikh2 - Qm'
Qm
R (4.1.25)
ikfl.2 - Qm
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
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:
84 Chapter 4. Scattering and Inverse Scattering
(4.2.13)
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
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
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.
4.2. Inverse Scattering for the KdV 85
1
m=l
+-
1
00
R(k, 0) exp[i(8k 3 t + ke)]dk. (4.2.18)
211" -00
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!
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:
86 Chapter 4. Scattering and Inverse Scattering
Gel'fand-Lavitan Equation:
B(E, t)
/
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.
4.2. Inverse Scattering for the KdV 87
(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
K(x, y, t) + B(x + y, t) + 1 00
B(y + z, t)K(x, z, t)dz = 0
can be solved exactly (see (ii) 2-soliton example below). The solution is
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)
and
-2exp(-x - y)
K (x, Yj t) = ( )"
1 +exp 2x - 8t
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
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
e
let = tanh x. This transformation maps (-00,00) for x to [-1,1] for e.
Then
(4.2.28)
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
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).
90 Chapter 4. Scattering and Inverse Scattering
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)
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).
4.2. Inverse Scattering for the KdV 91
+ 00
[6 exp(8t) exp( -(y + z)) + 12 exp(64t) exp( -2(y + z))]
K(x,z;t)dz=O. (4.2.33)
(4.2.37)
where
1 + 3e 8t - 2x 2e8t - 3x )
(
det 4e64t - 3x 1 + 3e64t - 4x
_6e8t - x 2e8t - 3x )
(
D.1 det _12e64t - 2x 1 + 3e64t - 4x
1 + 3e 8t - 2x _6e8t - x )
(
D.2 det 4e64t - 3x _12e64t - 2x
K(x, y; t) L1e- Y + L 2 e- 2y
~ [ _ 6(eSt - x _ e72t - 5x)e- y _
Hence
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:
We expect that
(4.2.41)
4.2. Inverse Scattering for the KdV 93
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- 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)
94 Chapter 4. Scattering and Inverse Scattering
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).
-d [(1- e)-
2 d1/l]
de + [l(l + 1) -
2
-m-2 ] 1/1 = 0
de 1- e
4.2. Inverse Scattering for the KdV 95
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
-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)
96 Chapter 4. Scattering and Inverse Scattering
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
(4.2.63)
4.2. Inverse Scattering for the KdV 97
Figure 4.4: Collision of three KdV solitons: the evolution of the initial profile
(4.2.47) according to (4.2.63).
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
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)
Hence
82
u(x, t) = -2 8x 2 logdet(M). (4.2.70)
The following theorem can be proved (see Miura (1976) [4]).
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:
(4.2.75)
tjJ(x, t) = exp(iEt/h)4>(x).
Then, Schrodinger equation becomes
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+R=T.
ik(I + I) = PI,
4.2. Inverse Scattering for the KdV 101
we have
k = ~i, P< O. (4.2.81)
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)
+ 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)
102 Chapter 4. Scattering and Inverse Scattering
The soliton is
(4.2.87)
where the phase shift is Zo = Pin 2. The wake part Kw satisfies another
integral equation:
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.
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.
[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
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).
104 Chapter 4. Scattering and Inverse Scattering
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
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
and
[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
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
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
From the sum of the above two equations, we solve for u(x, y):
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
=
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
New solutions -Wl2 and -W2l of P2(W) = 0 are constructed from the pairs
(WI, A2) and (W2' Al) respectively via the following BT
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
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
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
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)
W1234 -- W(4)
4(>"3 - >"4)
W(2) + ---'-~-...::..!.
W124 - W123
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
The 3-s01iton ladder given by (4.4.33) is shown in Fig. 4.7. The corresponding
3-s01iton is
(4.4.34)
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:
where Xl is the phase shift which is NOT arbitrary. The -th soliton ( =
1,2,3, ... , N) has the following properties:
-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
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)
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
(4.5.12)
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:
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
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
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
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)
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(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
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.
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
This is the generalized Legendre equation. The discrete eigenvalues are given
by
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)'
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
Burgers Equation
123
124 Chapter 5. Burgers Equation
Viscous fluid x
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)
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
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)
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
5.1. Viscous Fluid on an Inclined Plate 127
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)
Order c 1 :
on y = 0,
UO'TJ1x - VI = 0, (5.1.23)
128 Chapter 5. Burgers Equation
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 :
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
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
U2 = R{ [~(y2 - 1) cos ()
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)
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
130 Chapter 5. Burgers Equation
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
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:
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
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):
5.2. Cole-Hopf Transformation 133
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)
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
From (5.2.3), we in turn have an analytic solution for the nonlinear problem
(5.2.1) - (5.2.2)
(5.2.11)
as t -t 0+
134 Chapter 5. Burgers Equation
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)
erfc(s) = 100
exp(-x 2 ) dx.
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:
and
136 Chapter 5. Burgers Equation
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)
(( ( - 2s) = m(' + C
1-( z
In-(- = -2m'
1
7]l{Z, t) =
1 + exp
(Z-t I 2)
2m
(Burgers shock wave). (5.2.25)
5.2. Cole-Hopf Transformation 137
~ 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; ., .
138 Chapter 5. Burgers Equation
0.8
-3 -2 -1 3
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
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
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
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)
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
Substituting (5.3.3) ~nto (5.3.1) and linearizing it with respect to v around uo,
we obtain the following equation for v:
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
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
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
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
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].
Fluid (1)
Fluid (2)
Figure 5.7: The configuration of the smooth interface of two viscous fluids.
(5.4.1)
Normal ,. .....(l)n.n.
Vij 'J -
.....(2)n.n. _ ..., (
Vij J - I
1+ R21) '
Rl (5.4.2)
(5.4.4)
We use two examples to illustrate how to use the above formulas (5.4.2) and
(5.4.3).
144 Chapter 5. Burgers Equation
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
Projecting the above vector into the normal direction n and the tangential
direction t respectively, we obtain the following
2
[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
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.
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 .
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)
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)
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)
(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
First order:
Second order:
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).
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)
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
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)
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
154 Chapter 6. Forced KdV Equation
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).
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)
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
Po: 1
e = -~ ";-f3>' . (6.2.15)
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:
156 Chapter 6. Forced KdV Equation
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.
(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.
158 Chapter 6. Forced KdV Equation
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)
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
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
!: ,ech' [~(z_ - L+
.xv + av 2 + /3v" = 0, x> x_, (6.2.24)
v{z-) = - (6.2.25)
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.
160 Chapter 6. Forced KdV Equation
"
.;
..
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.
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.
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
= (cos 0- cos (0 + *) )
xcn 2 ( :~ (coso-cos (0+ 2;)) x o) (6.3.14)
(6.3.15)
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)
Xo = Vf4ii
4 arc sech Vf23 . (6.3.20)
.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).
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).
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
(6.3.29)
6.4. Transcritical Periodic Soliton Radiation 167
(6.3.32)
and Zo is determined by
(6.3.33)
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.
168 Chapter 6. Forced KdV Equation
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).
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)
i:
The mass conservation property of the wave motion gives the following identity
(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.
170 Chapter 6. Forced KdV Equation
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
= 1 00
-00 d(f- 1/ 2H:e)
lH(1+1I1)
~Hh d(Hy) {~[(f(-771(:e,y,t))(gH)1/2)2
1:
(mass of one soliton),
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)
=
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)
(6.4.15)
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)
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
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)
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
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
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
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 d -N 3 h*d a&*1/2(2H_h*)
- &pg Hl/22V3 d, (6.4.32)
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).
(6.4.34)
(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
6.4. Transcritical Periodic Soliton Radiation 175
=
where i vCT. We then take the Fourier transform of equation (6.4.35) with
respect to X and use the following approximation:
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)
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
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
176 Chapter 6. Forced KdV Equation
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:
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'.
6. 'plot' displays the solution u graphically from any "res" file that is loaded.
15
J
-1#
I
12
9
(Y'
(******************************************************)
(* *)
(* 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
l=lambda=alpha=beta=p=n=m=firstres=lastres\
=deltat=timesave=tottime=time=umt=ut=reu=s=O;
Begin[" 'Private''']
(*Procedures Section:*)
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*)
"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*)
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*)
6.4. Transcritical Periodic Soliton Radiation 181
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*)
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],
182 Chapter 6. Forced KdV Equation
{i,firstres,lastres}];
(*end of Do loop*)
]; (*end of msteps Block*)
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*)
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[ ]
EndPackage [ ]
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
184 Chapter 6. Forced KdV Equation
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)
Xo = )-4{3
T arctanh{b) (6.5.6)
3 Pa 1
f - f + 6>: -)-(3). = O. (6.5.7)
6.5. Stability of Solitary Waves 185
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.
186 Chapter 6. Forced KdV Equation
Sine-Gordon and
Nonlinear Schrodinger
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:
(iii) Pulse broadening due to dispersion and beam focusing due to nonlinearity
in an optic fiber (derivation 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).
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:
(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
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
192 Chapter 7. Sine-Gordon and Nonlinear SchrOdinger
P=F=~~~~~~~~====~~---x
Elastic bar
y
z
(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)
7.1. Physical Models of the Sine-Gordon Equation 193
V!/ I
,, , ,,
IIil II i J
: : :
IIV
: : :
: :
.!
II II II II
if if V II 1/ !
j-2 j-l j+l
x
Applied magnetic field
,,,.,,,,,,,,,,.,~. H
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)
(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
7.1. Physical Models of the Sine-Gordon Equation 195
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 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)
(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.
196 Chapter 7. Sine-Gordon and Nonlinear SchrOdinger
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)
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
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
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)
(7.2.8)
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
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.
VI - (i21'" _d,-u...,.....,.. =
2 ..,' sin( u/2)
i(
0
ds
198 Chapter 7. Sine-Gordon and Nonlinear SchrOdinger
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
and
20 exp [(z - 0.8t)/0.6]
i.px(z,t) = - [ ]
3 1 + exp (z - 0.8t)/0.3
(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.
200 Chapter 7. Sine-Gordon and Nonlinear SchrOdinger
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)
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
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
(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.
202 Chapter 7. Sine-Gordon and Nonlinear SchrOdinger
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):
(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.
7.2. Solutions to the Sine-Gordon Equation 205
(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.
206 Chapter 7. Sine-Gordon and Nonlinear SchrOdinger
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.
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)
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
(7.3.3)
Tout =
L d2k)21
T2+ ( - - (7.3.4)
2 dw 2 Wo
2
r, = r 1
+ (1.47LD>..2)
r2 (7.3.5)
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 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
and
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
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)
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()
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
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
v" - av + vv 3 = O. (7.4.5)
(7.4.6)
v= V~
-;- sechy'a(x - ct). (7.4.7)
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.
~(x,t) = 4arctan[u(x)v(t)]
to get the same result.
From (7.5.1), one can derive
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)
(7.5.6)
214 Chapter 7. Sine-Gordon and Nonlinear Schrodinger
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
7.5. Arctan Type of Solutions of the sG 215
ull + -;
2vv' ( -;- VII) + (u 2 + v 2) (VII)'
-; - 4v'v" = 2vv'. (7.5.8)
-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
The usual energy integral of (7.5.10), i.e. J(7.5.10) u'(:z:) d:z:, yields
(7.5.12)
(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
216 Chapter 7. Sine-Gordon and Nonlinear SchrOdinger
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
d2 = -C2 (7.5.14)
[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.
[7] G. B. Whitham (1974), Linear and Nonlinear Waves, John Wiley, New
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.
7.5. Arctan Type of Solutions of the sG 217
[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
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
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
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
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.
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)
228 Chapter 8. Selected Examples of Flow Instabilities
where
(8.2.5)
and
(8.2.6)
c = Pi Ul + paUa (8.2.7)
Pi +Pa
where
(8.2.8)
-- ~
----
~
~~ ~ k
V'
/
k*
-1 I
-1
I
-2 ,
-2 ~.
Figure 8.5: The relationship between c~ and k when the upper fluid is heavier.
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
230 Chapter 8. Selected Examples of Flow Instabilities
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.
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]
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]
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
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
u*t, + u~u~
J I,J
= -~p*.
0
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)
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
t ~ (d/w)t, (x, y, z) ~ (x d, y d, z d)
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
Substituting these two expressions into equations (8.3.13) and (8.3.14), we get
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)
(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)
236 Chapter 8. Selected Examples of Flow Instabilities
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]
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
cosh(~)
qsinh(V
(q2 _ f{2)2 cosh(q/2)
det(M) = o. (8.3.22)
8.4. Taylor Problem: Centrifugal Instability 237
Ra
unstable
----~--~
1708
stable
Kcr=3.12
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.
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
8.4. Taylor Problem: Centrifugal Instability 239
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
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.
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:
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.
(8.4.6)
(8.4.7)
Let
8.4. Taylor Problem: Centrifugal Instability 241
where 0' = m - 1,
Ta = _4Afhcr (8.4.10)
1I 2
U = DU = v= 0 when (= 0, 1. (8.4.11)
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
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
244 Chapter 8. Selected Examples of Flow Instabilities
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
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
246 Chapter 8. Selected Examples of Flow Instabilities
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) .
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
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.
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.
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
250 Chapter 9. Wave Interactions and X-Ray Crystallography
(9.1.1):
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.
In general, when the wave amplitude is small (ofthe order e), a wave system
may be described by an equation of the form
W(w,k) = O. (9.1.4)
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
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.
9.1. Wave Interactions 253
~ ,, ,
,
Beach
:
I
,
I
--.---
-,------r----'---
k, 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]
254 Chapter 9. Wave Interactions and X-Ray Crystallography
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]]
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.
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
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
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
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
258 Chapter 9. Wave Interactions and X-Ray Crystallography
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.
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
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
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
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)).
Film
Larger wave:
in phase superposition
0 t
-1
Smaller wave:
out phase superposition
-2
0 0.1 0.2 0.3 0.4
(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.
262 Chapter 9. Wave Interactions and X-Ray Crystallography
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 strength of the diffracted wave depends on the IF(h, k,f)1 and these
quantities, in turn, satisfy the Parseval identity
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.
9.2. Phase Problem in X-ray Crystallography 263
(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*):
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. Phase Problem in X-ray Crystallography 265
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
266 Chapter 9. Wave Interactions and X-Ray Crystallography
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,
9.2. Phase Problem in X-ray Crystallography 267
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)
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.
80
60
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.
9.2. Phase Problem in X-ray Crystallography 269
The second neighborhood of the same quartet consists of the four amplitudes
above plus three other additional magnitudes
and
H+K+L= o.
9.2. Phase Problem in X-ray Crystallography 271
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)
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
272 Chapter 9. Wave Interactions and X-Ray Crystallography
The tangent formula can be expressed in the following way. If ifoK and ifoL
are known (or preassigned), then the tangent formula
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)
-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:
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
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.
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
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)!
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).
..
( ...................................................................... )
(.
(. 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[n_Integer] := Block[{m,Phi,Co,Cm,B,K,L,GL,y,eqn,k},
I~[ n <= 0, Return["Use a positive integer."] ];
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
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] ;
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];
Do[
l[m] = Together[Expand[Part[Part[Part[soln,l],m],2]]],
{m,l,n,H];
281
by G. E. Sarty
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)
W12345 = w[15]
W;:----
N345
"
"w
5
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
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
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)] ]
Derivation of the
Stationary KdV
by G. E. Sarty
A/'1 - ~2/1/'
1 - ~/'"
61 = 0
(C.O.I)
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
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* = <I>~.. 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)
(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:
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)
(C.O.16)
Defining
(C.O.17)
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
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
wc.j"ftPz:z: + uc~tPyy = 0,
or
, or
(C.0.27)
314 Appendix C. Derivation of the Stationary KdV
1
- -tPyy
f
+ 2tPxyf( + tPyyf(2 + tPyf" = o. (C.0.2S)
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
= 0 on w = O.
f (C.0.32)
(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)
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~
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]
316 Appendix C. Derivation of the Stationary KdV
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
(C.0.44)
[3] P. G. Baines (1987), Upstream blocking and airflow over mountains, Ann.
Rev. Fluid. Mech. 19, 75-97. [6]
317
318 Bibliography
[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]
[40] P. K. Kundu (1990), Fluid Mechanics, Academic Press, New York, Chap-
ter 11. [8]
[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]
[69] M. C. Shen and S. M. Sun (1987), Critical viscous surface waves over an
incline, Wave Motion 9, 323-332. [5]
[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]
[73] S. S. Shen (1992), Forced solitary waves and hydraulic falls in two-layer
flows, J. Fluid Mech. 234,583 - 612. [1,6]
[76] J. J. Stoker (1957), Water Waves: the Mathematical Theory with appli-
cations, Interscience, New York. [3]
[77] M. Van Dyke (1982), An Album of Fluid Motion, The Parabolic Press,
Stanford, California. [2, 8]
[79] G. B. Whitham (1974), Linear and Nonlinear Waves, John Wiley, New
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
323
324 Index