2003 Book PerturbationMethodsForDifferen PDF

"Was du ererbt von deinen Viitern hast, erwirb es, urn es zu besitzen.
"
("That which you have inherited from your fathers,
acquire it so as to make it your own.")
-Goethe in Faust
In memory ofmy father, Krishnarao Shivamoggi

Bhimsen K. Shivamoggi
Perturbation Methods
for Differential Equations
Springer Science+Business Media, LLC

Bhimsen K. Shivamoggi
Department of Mathematics
University of Central Florida
Orlando, FL 32816-1364
U.S.A.
Library of Congress Cataloging-in-Publication Data
Shivamoggi, Bhimsen K.
Perturbation methods for differential equations I Bhimsen K. Shivamoggi.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-4612-6588-7 ISBN 978-1-4612-0047-5 (eBook)
DOI 10.1007/978-1-4612-0047-5
1. Perturbation (Mathematics) 2. Differential equations-Numerical solutions. 3.
Differential equations, Partial-Numerica! solutions. 1. Title.
QA871 .S44 2002

515'.35-dc 21 2002066657
CIP
AMS SubjectC1assifications: 34E05, 34EI0. 34E13. 34E15, 34E20, 35B20, 35B25
Printed on acid-free paper.

©2003 Springer Science+Business Media New York
OriginalIy published by Birkhăuser Boston in 2003
Softcover reprint ofthe hardcover Ist edition 2003
AlI rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher, Springer Science+Business Media, LLC,
except for brief excerpts in connection with reviews or
scholarly analysis. Use in connection with any form of information storage and retrieval, electronic
adaptation, computer software, or by similar or dissimilar methodology now known or hereafter
developed is forbidden.
The use of general descriptive names, trade names, trademarks, etc., in this publication, even if the
former are not especially identified, is not to be taken as a sign that such names, as understood by the
Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone.
ISBN 978-1-4612-6588-7 SPIN 10764753
Typeset by the author.
987 6 543 2 1
Table of Contents
Preface xi
Chapter 1 Asymptotic Series and Expansions 1

1.1 Introduction 1
1.2 Taylor Series Expansions 2
1.3 Gauge Functions 2
1.4 Asymptotic Series and Expansions .4
1.5 Asymptotic Solutions of Differential Equations 10
1.6 Exercises 11
Chapter 2 Regular Perturbation Methods 13

2.1 Introduction 13
2.2 Algebraic Equations 14
2.3 Ordinary Differential Equations 22
2.4 Partial Differential Equations 29
2.5 Applications to Fluid Dynamics: Decay of a Line Vortex 31
2.6 Exercises 33
2.7 Appendix. Review of Partial Differential Equations 35
AI. Transformation to Canonical Forms 35
vi Contents
Chapter 3 The Method of Strained CoordinateslParameters •.••••..•..•.........•..41

3.1 Introduction 41
3.2 Poincare-Lindstedt-Lighthill Method of Perturbed Eigenvalues .42
3.3 Eigenfunction Expansion Method 52
3.4 Lighthill's Method of Shifting Singularities 55
3.5 Pritulo's Method of Renormalization 59
3.6 Wave Propagation in an Inhomogeneous Medium 63
3.7 Applications to Solid Mechanics: Nonlinear Buckling of
Elastic Columns 65
3.8 Applications to Fluid Dynamics 71
3.8.1 Nonlinear Hyperbolic Waves in a Gas 71
3.8.2 Rayleigh-Taylor Instability of Superposed Fluids 75
3.9 Applications to Plasma Physics 82
3.9.1 Nonlinear Waves in an Electron-Plasma 82
3.9.2 Rayleigh-Taylor Instability of a Plasma in a Magnetic Field 84
3.10 Limitations of the Method of Strained Parameters 91
3.11 Exercises 93
3.12 Appendix 1. Fredholm's Alternative Theorem 95
3.13 Appendix 2. Floquet Theory 97
3.14 Appendix 3. Bifurcation Theory 107
Chapter 4 Method of Averaging 113

4.1 Introduction 113
4.2 Krylov-Bogoliubov Method of Averaging 114
4.3 Krylov-Bogoliubov-Mitropolski Generalized Method of Averaging 118
4.4 Whitham's Averaged Lagrangian Method 122
4.5 Hamiltonian Perturbation Method 125
4.5.1 Systems with Constant Parameters 125
4.5.2 Systems with Slowly-Varying Parameters 130
4.6 Applications to Fluid Dynamics: Nonlinear Evolution of
Contents vii
Modulated Gravity Wave Packet on the Surface of a Fluid 135

4.7 Exercises 139
4.8 Appendix 1. Review of Calculus of Variations 141
Al.l Functionals with Second-Order Derivatives 141
A1.2 Functionals with Higher-Order Derivatives 144
A1.3 Functionals with Several Independent Variables 145
4.9 Appendix 2. Hamilton-Jacobi Theory 148
A2.1 Hamilton's Equations 148
A2.2 Canonical Transformations 150
A2.3 Hamilton-Jacobi Equation 152
A2.4 Action-Angle Variables 153
Chapter 5 The Method of Matched Asymptotic Expansions 155

5.2 Physical Motivation 155
5.3 The Inner and Outer Expansions 158
5.4 Hyperbolic Equations 166
5.5 Elliptic Equations 176
5.6 Parabolic Equations 179
5.7 Interior Layers 181
5.8 Latta's Method of Composite Expansions 185
5.9 Turning Point Problems 190
5.9.1 JWKB Approximation 190
5.9.2 Solution Near the Turning Point.. 194
5.9.3 Langer's Method 197
5.10 Applications to Fluid Dynamics: Boundary-Layer Flow Past
a Flat Plate 199
5.10.1 The Outer Expansion 199
5.10.2 The Inner Expansion 201
5.10.3 Flow Due to Displacement Thickness 206
viii Contents
5.11 Exercises 207

5.12 Appendix 1. Initial-Value Problem for Partial Differential Equations 209
5.13 Appendix 2. Review of Nonlinear Hyperbolic Equations 211
Chapter 6 Method of Multiple Scales 219

6.2 Differential Equations with Constant Coefficients 219
6.3 Struble's Method 243
6.4 Differential Equations with Slowly Varying Coefficients 252
6.5 Generalized Multiple-Scale Method 262
6.6 Applications to Solid Mechanics: Dynamic Buckling of a
Thin Elastic Plate 265
6.7.1 The Problem of Aerodynamically Generated Sound 273
6.7.2 Mathematical Formalization of Lighthill's
Theory of Aerodynamically Generated Sound 276
6.7.3 Nonlinear Shallow Water Waves 281
6.7.3(a) Governing Equations 282
6.7.3(b) Korteweg-deVries Equation 284
6.7.3(c) Solitary Waves 286
6.7.3(d) Stokes Waves 288
6.7.3(e) Perturbed Solitary Wave Propagation 290
6.7.3(t) Wave Modulation and Nonlinear Schr6dinger
Equation 295
6.7.3(g) Long-Time Evolution of the Modulation 299
6.7.3(h) Second-Harmonic Resonance 303
6.8.1 Nonlinear Longitudinal Waves in a Hot Electron-Plasma 307
6.8.2 Ion-Acoustic Solitary Waves in an Inhomogeneous Plasma 310
6.9 Exercises 316
Contents ix
Chapter 7 Miscellaneous Perturbation Methods 319

7.1 A Quantum-Field-Theoretic Perturbative Procedure 319
7.1.1 Blasius Equation 320
7.2 A Perturbation Method for Linear Stochastic Differential Equations 322
7.2.1 Application to Wave Propagation in a Random Medium 325
7.2.2 Renormalization Procedure 327
7.2.2(a) Exact Solution 327
7.2.2(b) Perturbative Solution 328
7.2.2(c) Renormalized Solution 329
7.3 Exercises 333
References 335
Answers to Selected Problems 343
Index 347
Permissions 353
Preface
In nonlinear problems, essentially new phenomena occur which have no

place in the corresponding linear problems. Therefore, in the study of nonlinear
problems the major purpose is not so much to introduce methods that improve
the accuracy of linear methods, but to focus attention on those features of the
nonlinearities that result in distinctively new phenomena. Among the latter are -
* existence of solutions of periodic problems for all frequencies rather than
only a set of characteristic values,
* dependence of amplitude on frequency,
* removal of resonance infinities,
* appearance ofjump phenomena,
* onset of chaotic motions.
On the other hand, mathematical problems associated with nonlinearities are
so complex that a comprehensive theory of nonlinear phenomena is out of the
question.' Consequently, one practical approach is to settle for something less
than complete generality. Thus, one gives up the study of global behavior of
solutions of a nonlinear problem and seeks nonlinear solutions in the
neighborhood of (or as perturbations about) a known linear solution. This is the
basic idea behind a perturbative solution of a nonlinear problem.
1 In an alternative approach via numerical solution of nonlinear problems, computers
have contributed enormously by providing detailed information. But, this approach lacks
organization and does not afford complete insight into the underlying complex physical
behavior. Besides, for problems of singular-perturbation type (see below), determination
of the accuracy of a numerical solution is extremely difficult without first obtaining some
idea about the salient features of the solution afforded by the singular perturbation
approach.
xii Preface
The method of perturbation expansion is carried out with respect to a small

parameter £, which may be a non-dimensionalized amplitude of a typical
perturbation. The coefficients in these expansions are obtained as solutions of a
sequence of linear problems. The lowest-order tenns are governed by problems
that typically result from a linearization of the original problem and are known.
The higher-order quantities are produced as solutions of linear inhomogeneous
differential equations where the inhomogeneities involve only the previously
detennined lower-order quantities.
One has a regular perturbation problem if a straightforward perturbation
expansion is unifonnly valid. However, if a straightforward expansion is not
unifonnly valid, and if this difficulty is compounded in higher approximations,
then one has a singular perturbation problem. Over the course of more than a
century, starting with problems of dynamical astronomy, a host of singular
perturbation procedures have been developed to resolve these non-unifonnities.
A number of excellent accounts of these procedures exist in the literature
emphasizing several different points of view.
This book has grown out of my lecture notes for an interdisciplinary graduate
level course on perturbation methods which I have taught for the past several
years to a mix of students in applied mathematics, physics, and engineering. I
have endeavored to describe the standard procedures (which together cover a
wide area of applicability) along with the underlying basic concepts in a unified
way. To facilitate an interdisciplinary readership, I have adopted a
straightforward intuitive approach and paid more attention to the procedures and
the underlying ideas than to mathematical rigor. Thus, I have side-stepped the
theoretical aspects such as error estimates and soft-pedaled the fonnal proofs.
The book starts with an introduction to the idea of asymptotic
approximations to the evaluation of transcendental functions and the solution of
differential equations and a discussion of the regular perturbation approach. It
then embarks on a systematic and unified discussion of the resolution of the
non-unifonnities produced, as a rule, by the regular perturbation method. We
discuss several singular perturbation methods, primarily
* the method of strained parameters,
* the method of averaging,
* the method of matched asymptotic expansions,
* the method of multiple scales,
* the quantum-field-theoretic renonnalization method.
Preface xiii
These procedures are illustrated via simple but useful worked examples and
are applied to ordinary and partial differential equations arising in various
problems of solid mechanics, fluid dynamics, and plasma physics. I have tried to
provide brief background material for several mathematical as well as
application topics to make the discussion reasonably self-contained.
Several exercises are provided at the end of each chapter along with answers
to selected problems (given at the end of the book) to consolidate the ideas and
procedures and to occasionally indicate developments from the research
literature.
The selection of the application topics reflects, to some extent, my own
research interests. In the same vein, the list of references given at the end of the
book contains primarily books and papers I have used in developing the
graduate lecture material (and this book) which may also provide the reader with
some useful leads into the more advanced aspects of the subject. Perturbation
methods has a vast literature and the list of references given in this book cannot
pretend to be fully representative.
Acknowledgements
In the preparation of this manuscript (and the graduate lecture material), I
have drawn enormously from the ideas, procedures, and treatments developed
by several eminent authors - J.D. Cole, K.O. Friedrichs, M.H. Holmes, J.B.
Keller, 1. Kevorkian, P.A. Lagerstrom, A.H. Nayfeh, R.A. Struble, M.D. Van
Dyke, and others. I have acknowledged the various sources, in this connection,
in the References, given at the end of this book and wish to apologize to any
others I may have inadvertently missed. I wish to thank my students and
colleagues who rendered valuable assistance as the course material was
developed. I wish to express my thanks to my colleagues, Professors Larry
Andrews, John Cannon, David Rollins, and Mike Taylor for several valuable
discussions on many aspects of the subject. Most of the manuscript was read by
Professors John Cannon and David Rollins who made valuable suggestions. I
wish to thank also the referees for their valuable suggestions.
My sincere thanks are due to Jackie Callahan for the exemplary patience,
dedication, and skill with which she did the whole typing job. My thanks are due
to Ronee Trantham for her excellent job in doing the figures. My thanks are also
due to Birkhauser for their splendid cooperation with this whole project and
working with them was really a delight. Last, but by no means least, my
immense thanks are due to my wife, Jayashree, and my daughters, Vasudha and
xiv Preface
Rohini for their enormous understanding and cooperation with this endeavor (as
they have done with my previous book projects). My father Krishnarao could
never attend college but was always immensely enthusiastic about higher
learning and knowledge. This book is dedicated to his memory.
Orlando, Florida Bhimsen K. Shivamoggi

Chapter 1
Asymptotic Series and

Expansions
1.1. Introduction
The method of perturbations can be used to develop approximate solutions to

differential equations, which have nonlinearities or variable coefficients so that
an exact solution cannot be constructed. Questions of convergence of the
perturbation expansion in terms of a suitable small parameter are abandoned if
such an expansion can provide a useful representation of the exact solution.
However, the perturbation expansions are usually not uniformly valid and
various techniques have been developed to render them uniformly valid.
A perturbative solution of a nonlinear problem becomes viable if it is "close"
to another problem we already know how to solve. We study the solution of the
simpler problem and then try to express the solution of the more difficult
problem in terms of the simpler one modified by a small correction. This
approach is sequential, in the sense that once an improved approximation has
been found, the process may be repeated to obtain a still better approximation.
The latter step involves developing a formal perturbation expansion with respect
to a small parameter e. The accuracy of this approximation gets better for
smaller values of e. For example, e may be the nondimensionalized amplitude
of a typical perturbation. The coefficients in these expansions are obtained as
solutions of a sequence of linear problems. The quantities of relevance are
essentially developed in a Taylor series, which is taken to be uniformly
convergent in e in any compact part of the respective domain of analyticity. The
justification to expect the solution to exhibit such an analytic dependence on the
parameter e follows if the parameter e enters the differential equations and the
2 Asymptotic Series and Expansions
boundary conditions in an analytic way. The lowest-order terms in these series

are given by quite simplified forms of the original problem, so that often these
terms are well known. The higher-order terms are then produced as solutions of
linear inhomogeneous differential equations where the inhomogeneities involve
only the previously determined linear-order quantities.
1.2. Taylor Series Expansions
Suppose a function f(x) is infinitely differentiable at x = x o' then we may

express it in a power series of (x - x o) as
(1.2.1)
where,
_ 1 .c<n) ( ) (1.2.2)
an = - } xo '
n!
Example 1: Taylor series expansions of some standard functions are -

x3 x5
sin x = x - - + - - ...
3! 5!
x~ x4
cosx=I--+-- ...
2! 4!
,
x X x-
e =1+-+-+ ...
l! 2!
x~ x3
£n(l+x) = x - - + - - ....
2 3
1.3. Gauge Functions
Consider the limit of a function f (£) as £ ~ O. This limit may depend on

whether £ ~ 0+ or £ ~ 0- .
1.3 Gauge Functions 3
I I
Example 2: lim e '

£-+0+
=0, lim e '
£-+0-
=00.
Suppose e> O. Suppose f(e) has no essential singularity at e = 0 (e.g.,
lim sin
,->0
(.!.), then one has three possibilities-
e
lim
' ....0
If (e)1 =1:, 0 < A < 00. (1.3.1)
00
In order to determine the rate at which f(e) tends to 0 or 00, we compare

these rates with the rates at which known functions, called gauge functions, tend
to 0 or 00.
The simplest set of gauge functions are
2
1,e,e ,.,. }
e- ,e- ,e-
1 2 3
, •••
which satisfy, for e near 0,

(1.3.2)
Example 3: We have
from which
2 4
. sine
hm--=hm e e
. ( 1--+--,.. )
=1
' ....0 e ' ....0 3! 4!
2
r 1- cos e r (1 e 1
=,~ 2! -4'!+,.. = 2!
)
,~ e2
Thus,
sine ~ 0 at the same rate as e~0 }

2
l-cose~O at the same rate as e ~O .
On the other hand, in order to detennine the rate at which

e ~ 0, we try to expand e- I/£ in a Taylor series for small e.
Letting f(e) = e- I/£, we have
f '( e)=-.!..-
e
2 e -1/£ ,
etc.
Thus,
I 2
j'(O) = lim -.!..-2 e--; = lim ~=O
£->0 e x->~ eX
1"(0) = lim (-.!..-4 - 3

£->0 e e
2-) e-;
2 3
=lim 4
x - x =0
x-+oo eX
etc.
where x == 1/e.
I
Therefore, the function e £ cannot be represented by a power series in e. In
I
fact, e £ ~ 0 faster than any power of e, because

I
£ n
•1m
" -e n = ."1m --;-=
x 0.
£->0 e x->~ e
1.4. Asymptotic Series and Expansions
Definition: Let f (e) and g (e) be defined in some neighborhood of e = 0 "
We write
f(e) =o(g(e)) as e~0 (1.4.1)
if
1.4 Asymptotic Series and Expansions 5
lim If(e)1
£->0 g(e)
=0
and we write
f(e) =O(g(e)) as e -7 0 (1.4.2)
if there exists a positive bounded constant M such that
If(e)1 ~ M ~(e)1 or l;i;~1 is bounded

for all e in some neighborhood of e = o.
I I
Example 4: e--; = 0 ( ea ), \;j a. Thus, e f is transcendentally small with respect

to ea.
Example 5: e 2ene = o(e) as e -7 0+.

By L'Hospital's rule
· -
I1m - = I'1m -ene
e2ene - = I'1m
£->0' e £->0' (lie) £->0'
Thus, e 2ene = o(e) as e -7 0+.
Example 6: sine = O(e) as e -7 0+.

By the Mean Value Theorem of calculus, there exists a number c between 0
and e such that
sine -sinO
----=cosc.
e-O
Thus,
Isin el = Ie cos cl ~ lei
since Icos cl ~ 1. Therefore,
sine = O(e) as e-70+.

The above definition may be extended to functions of e and another variable

lying in an interval I.
Definition: Let f(t,e) and g(t,e) be defined for all tEl and all e in a
neighborhood of e = o. We write
f(t,e) =o(g(t,e)) as e -? 0 uniformly on I (1.4.3)
if
limlf(t,e)l=o "d tEl.

£-->0 g(t,e) ,
If there exists a positive boundedfunction M(t) on I such that

\I(t,e)1 ~ M(t) Ig(t,e)1
for all tEl and for all e in some neighborhood ofzero, then we write
f(t,e) =O(g(t,e)) as e -? 0, uniformly on I. (1.4.4)
A measure of decreasing orders of magnitude is provided by an asymptotic

sequence of functions.
Definition: A sequence {lpn (e)}~=1 is an asymptotic sequence if
A sum of terms in an asymptotic sequence is an asymptotic expansion.

Various linear operations and multiplications can be performed with asymptotic
expansions.
If a function possesses an asymptotic approximation in terms of an
asymptotic sequence, namely, an asymptotic expansion, then the latter is unique
for the given asymptotic sequence. Besides, with an asymptotic expansion, the
error of approximation can be understood as well as controlled.
Example 7: Lan lpn (e) is an asymptotic expansion of f(e) as e -? 0 if

n=1
N
f(e)-Lancpn(e)=O(CPN(e)) or O(CPN+I(e)) as e~O.
n:O
Note that
m-I
f(e)- Lancpn(e)
am = ~~ cp:'(e)
which gives am uniquely.
In a physical problem, the coefficients in an asymptotic expansion will

depend on space or time variables other than e. The series is said to be
uniformly valid (in space and time) if the error is small uniformly in these
variables.
Definition: An asymptotic expansion

~
f(x,e) - LOm (e) am (x) as e ~ 0 (1.4.5)

m:O
is uniformly valid if
N
f(x,e)- LOm(e) am(x)=RN(x,e) (1.4.6)
m:O
where RN(x,e) =O[ON+I (e)] or O[ON (e)] uniformly, for all x; otherwise it is
nonuniform.
Example 8: sin(x+e)=sinx+e cosx- e~ sinx+ ... is a uniformly valid

2!
expansion as e ~ O.
Example 9: -J x + e =.,J; (1 + e/x)1/2

e e~ +....
=.,J; ( 1+---, )
2x 8x'
This is a nonuniform expansion because this converges only if Ie/xl $1.
Example 10: y(t,e) = e- tt , t > 0, e« 1.

The first three terms in the Taylor expansion in powers of e provide an
approximation
Yapprox ( t,e ) =1-et+'2

1 e -" r .
The error is
E(t,e) = e- Ct -1 + e t _.!. e 2 t 2 = _..!.. e 3 t 3 +....

2 3!
For a fixed t, the error can be made as small as desired by choosing e small
enough. However, for a fixed e, no matter how small, the approximation breaks
down, as t becomes large. Thus, this approximation is not uniformly valid on
1=[0,00). In other words, we cannot write E(t,e)=o(e 3 ), as e~O,
uniformly on [0,00).
Example 11: Consider

dy 1
dx +y=~, x»!.
This has the solution given by
J-;e dz
x Z
x
y=e-
- x
ez
f ......"z· dz
1 -x
=-+e
x
x
eZ
f
1 1 -x
=-+-, +2e -dz
x x· Z3
O! l! 2! (n-l)!
=-+-, +3+···+---+
n
n! e
-x J -;;:;:r
x
e
Z
dz.
x x· x x Z
Note that this series diverges for all x, as n ~ 00. However, for x < 0, the
residue after n terms is
I~I = n! e- X
1Z~:I dZI
< I II I
- n. x n +1 e
-x IX
_ e
Z d _
Z -lxn+II'
n!
Thus, the error committed in truncating the series after n terms is in magnitude
less than the first neglected term. Observe that, as Ixl ~ 00, with n fixed,
IRnl ~ 0; so, for a fixed large n, the first n terms in the series can represent y
with an error which can be made arbitrarily small by taking x sufficiently large,
the divergence of the series notwithstanding.
An infinite asymptotic series may either converge for some range of e or

diverge for all e. Mathematical convergence depends on the behavior of terms
of indefinitely high order. However, in developing an approximation in practice,
one can calculate only the first few terms and hope that they rapidly approach
the true solution. This requirement may sometimes be better met with a
divergent series than with a convergent series which requires many terms to give
an accurate approximation. The error incurred by the nth sum is usually smallest
when e-{l/n).
Example 12: The expansion of the Bessel function

~ (_1)k e2k
J (e) =
o 2 2k (k!)2 t;
has an infinite radius of convergence, but many terms are needed for accurate
results unless e is small. On the other hand, the asymptotic expansion
- [(1 -
(eI) - f!e
2 2 2 2 2 2
J - 1-.3- e +----;--
1 .3 .5 .7 e + .
2 4 )
o 1C 2! 8 4! 82 4 ••
xcos (~- :)
+(_1_
2
1!8
e 1
2 2 2
1_'_3_,.....5_
I __
3
3! 8
e3 + ...) sin (-e1 - ~4 )]
is divergent for all e, no matter how small, but a few terms given good accuracy
for moderately small e.
1.5. Asymptotic Solutions of Differential Equations
Frequently, even though the solution of a given differential equation is not a

simple combination of elementary functions, an asymptotic representation of
that solution can be constructed in terms of elementary functions.
Example 13: Consider the differential equation -
U
" +U _ /32 -1/4
2
_0 .
U -
x
This equation results by putting y = U$) in Bessel's equation
x2 y" + X y' + (x 2 - /32) Y =O.

For x » I, this equation gives
U" + U '= 0,
from which
We may therefore seek the following solution-
U - eix L an x- n , ao =1.
n=O
Following the usual procedure for a power-series solution, we obtain
u-e
ix [
1+
(/32 -1/4) - (/32 -1/4)(/32 - 9/4) + ...]
2x 8x 2 •
Note: If /3 = ± 1/2, we have
as is to be anticipated readily from the given equation.

1.5 Asymptotic Solutions of Differential Equations 11

U"-AXU=O, A»1.
Look for a solution of the form

jgdt
u=e
Substituting this in the given equation, we obtain
g' + g2 = AX.
Let
L gn (x) A- n/2
~
g[X,A] - Alf2
n=O
which leads to
[(g~ - x) + A- I/ 2 (2g og, + g~)+ ± (g; + 2g0 g2 + gn+ ... J=O.

Equating coefficients of the various powers of II A to zero, we obtain
etc.
Thus,
')
U ( X,I\. - x -1f4 e ±2/3 AI/I X'/I .
This solution breaks down near x = O. The domain of nonuniformity

corresponds to g' -l -
AX or (on using g - X1f2 ) X Alf3 - 0(1).
1.6. Exercises
1. Show that .Je =o(e- 3/ 2 ) as e ~ 0+ .

1- cose
2. Show that lne =o(e- p ) as e ~ 0+ for alI p> o.

3. Show that e,ant = 0(1) as e ~ o.
=f ~ e-xdx.
~
4. Find an asymptotic expansion for large w for f(w)

o w+x
f7 dt.
~ -I
5. Find an asymptotic expansion for large x for F(x) =

x
f _e_
00 -Xl
6. Find an asymptotic expansion for large x for En (x) = dt.n

tI
7. Find an asymptotic solution for large x for the equation dy =e-2xy .

dx
Chapter 2
Regular Perturbation Methods
2.1. Introduction
The regular perturbation method works only for exceptionally special

problems, and fails in general.
The procedure consists of -
(i) substituting the power series
y(x;e) - L en Yn(x)
n=O
into the differential equation and the boundary/initial conditions,

(ii) expanding all quantities in a power series in e,
(iii) collecting terms with same powers of e and equating them to zero,
(iv) solving this hierarchy of boundary/initial value problems sequentially,
i.e., if YO'YI' ... 'Yk-1 are known, then Yk is detennined by an equation of
the fonn
where ;;£ is the linearized operator of the reduced problem evaluated at

the known solution Yo; the possibility that the above problem has a
solution locally whenever ;;£ is invertible is indicated by the Implicit
Function Theorem.
Theorem 1: Consider an implicit equation

f(x,e)=O.
14 Regular Perturbation Methods
If there exists x = X o such that
and if Ix (x o' 0) is an invertible linear map, then there exists a unique solution
ofthe given equation in the neighborhood of e =0 given by
x=g(e).
2.2. Algebraic Equations
Many of the essential concepts of perturbation methods can be discussed in

the simpler context of algebraic equations before differential equations are
considered.
Example 1: Consider the algebraic equation (Simmonds and Mann, 1986)

Z2 - 2z + e = 0, e« l.
This equation has two roots given by
zJ = 1-.JI=f}
Z2 = 1+.JI=f .
Expanding .JI=f as a Taylor series
.JI=f =1 _! e _! e2 _ •••
2 8
we obtain
Let us now obtain these results using a regular perturbation expansion,

z(e) ~ ao +ea +e 2az +... +a NeN +O(e N+
J
J
).
On substituting this into the above equation, we obtain

[a o +ea l +e 2a2+... +eNaN+o(eN+')r

- 2 [a o +ea, +e 2a2+ ... + eNaN +O(e N+1 )]+e =0
from which
[a~ +e·2aoa, +e 2(a~ +2aOa2)+o(e 3 )]
2
- 2 [a o + e al + e a2+ O( e )] + e =O.
3
Rearranging this, we obtain

(a~ -2ao )+e(2aoa\-2a\ +1)+e2(a~ +2aOa2 -2a2 )+O(e 3 )=0.
Equating coefficients of like powers of e to zero, we obtain an infinite system

of equations
;:O~~:02:,0+ 1 =0 )
a~ + 2aOa2 - 2a 2 =0
etc.
which is to be solved recursively.

Thus,
ao =0 or 2.
For ao = 0, we obtain
1 1
al =2' a2 =8' etc.
which gives the first root

1 1 2
ZI - - e+- e +....
2 8
For ao =2, we obtain
1 1
a l = -2' a2 = -8' etc.
which gives the second root

1 1,
Z - 2 - - e - - e- - .. '.
2 2 8
Example 2: Consider the algebraic equation (Simmonds and Mann, 1986)

e Z2 - 2z + 1 =0, e« 1.
Note that, in the limit e -7 0, the degree of the equation drops from 2 to 1. So,
this problem cannot be treated adequately by a regular perturbation expansion.
This is a singular-perturbation problem.
This equation has two roots given by
Expanding ~ as a Taylor series
~ =1-1- e-! e2 _ •••

2 8
II}
we obtain
z =-+- e+ .. ·
z~. =~-~_!
e 2
e+ .. · .
8
Let us now obtain these results using a regular perturbation expansion,
z(e)- ao +ea l +e 2 a2 +0(e 3 )
(this presupposes that e Z2 is small compared with -2z + 1 In the given
equation). On substituting this into the above equation, we obtain
e[ao+ ea, +e 2 a2 + 0(e 3 )f - 2[ao+eat +e 2a2 + 0(e 3 )]+ 1 =0

from which,
e[a~ +e ·2aoa, +0(e 2 )]-2[ao+eat +e 2a2 +0(e 3 )]+ 1 =O.

Rearranging this, we obtain
(-2a o+ 1)+e(a~ - 2a t )+ 0(e 2 ) =O.
Equating coefficients of like powers of e to zero, we obtain
-~ao + I: O}
ao -2a, -0 .
etc.
Thus,
etc.
These give only the first root

I I
z, ='2+8e+ ....
Now, in order to find the second root Z2' note that we may write
ez 2- 2z+ 1= e(z-z,)(z -Z2)
=ez 2 -e(z, +z2)z+ez,z2
from which
e Z,Z2 = I or Z2 - eI
so that, for the root Z2' e Z2 is not small compared with - 2z + I, in the limit
e ~ 0, so the previous regular perturbation expansion does not set up the right
dominant balance for Z2' and is, therefore, not suited for recovering Z2'
This suggests that, in order to obtain Z2' we should set
ez(e) =w(e), with w(O)i= 0,

so that the given equation becomes
2
w -2w+e= O.
Observe that the singular problem posed by the second root has been
rendered regular by a change of variable (z to w).
The roots of this equation can now be found using a regular perturbation
expansion
Then we obtain
from which
bg -2bo = 0 }
2bObi - 2bl + I = 0 .
etc.
We then obtain
bo = 2, bl = -1/2 , etc.
Note that the other root for bo' viz., bo = 0, has to be discarded since w (0) *" O.
Thus, we have
from which, we get for the second root,

2 I
z, =---+O(e).
. e 2
Example 3: Consider the algebraic equation (Nayfeh, 1981)

(x-I)(x-r)=-ex.
When e =0 , this equation becomes
(xo-I)(xo-r)=O
from which,
Xo =I or Xo =r.
Look for a regular perturbation expansion,
x- Xo + e XI + e 2x 2 + ....
Substituting into the given equation, and expanding we obtain
(x o -I)(xo -r)+e[(2xo -I-r) XI +xo]
+e 2 [(2x o -1- r) x 2 +x~ +x l ]+ .. · =O.
Equating the coefficients of each power of e to zero, we obtain
(X o -1)(xo - r) = 0
(2x o - 1- r) XI + X o = 0
2
(2x o -1-r) X 2 +X 1 +X I =0
etc.
Thus,
X o =1 or X o =r
1 r
XI = - - - or - -
l-r l-r
r r
X, = - - - - or - - -
" (1- r)3 (1 _ r)3
etc.
Therefore,
~
x - 1- 1 < - (I ~',: )' + .. ).
er e"r
x-r+--+---+· ..
l-r (l-r)3
Observe that these expansions break down as r ~ 1, and the singularity gets
worse at higher orders. The region of non-uniformity can be determined by
examining the conditions under which the successive terms in these expansions
are of the same order. This happens when
l-r=0(eI/2 ).
In order to obtain a uniform expansion in this region, introduce a detuning
parameter (1 defined by
and look for a solution of the form
We then obtain
((X o-1) + el/ 2 XI + .. ,][(xo-1) + el/2 (XI + (1) + ...]
=-e[xo+e1/2 XI + ...],
from which
Thus,
Xo =1 or 1 }
XI =± (_(J±~(J2 -4), etc..
Therefore,
X = 1-± el/
2
(J+~(J2 -4 )+".}
X =1- ±el/ 2
(J - ~ (J2 - 4 ) +".
which are regular at (J =0 or r =1!
Example 4: Considerthe roots of Bessel's function J o(x) for large x.
Note the asymptotic expression
Jo(X)-~:x [cos(x-~)+L sin(x-:)l x large.
The roots of
are then given by
cot(x _lC)
4
= __1_, xlarge.
8x
So, to first approximation, we have
cot(x - :}= 0,
from which
or
x "" ( n + ~ ) Tr, n large.
To improve on this approximation, let us put
x=(n+~) Tr+(j.
This leads to
cot [(n +.!..) Tr +(j] = _ 1 +...

2 2Tr(4n+3)+8(j
or
cot ( n + ~) Tr' cot (j - 1 _ 1 1

--:---- ---.,....---+ ...
cot ( n + ~) Tr + cot (j - 2Tr(4n+3) 1+ 4
Tr(4n+3)
(j
or
from which
(j "" 1
2Tr(4n+3)
Therefore,
Root No. 1 2 3 4 5
Perturbation 2.40308 5.52004 8.65372 11.79153 14.93092
Tabulated 2.40482 5.52008 8.65373 11.79153 14.93092
Table 2.1. Comparison of Approximate and Tabulated Roots

of Bessel's Function of order zero (from Nayfeh, 1981).
2.3. Ordinary Differential Equations
We will now consider applications of regular perturbation methods to ordinary

differential equations.
Example 5: Consider the motion of an object projected radially upward from

the surface of the earth (Holmes, 1995). Letting x(e) denote the height of the
object, measured from the surface, we have from Newton's Second Law, the
following equation of motion -
d 2x gR
2
-2- t>O
dt - - (x + R)2 ,
where R is the radius of the earth and g is the gravitational constant.

Let the initial conditions be
t= °: =x
dx
0, -dt = V;0'
Non-dimensionalizing as follows -
t x
r=--
(Va/g) , y= (Va /g)'
2
The above initial-value problem becomes

2
d y =- (l+ey)2'
dr2 1 r>O } ,
r = °: = y 0, dy
dr
=I
where
V;2
e= _0_« I.
Rg
Look for a solution of the form-
2
y- yo(r)+ey, (r)+0(e ).
Substituting into the initial-value problem and equating the coefficients of like
powers of e, we obtain
Y~ =I}
0(1): y;=-I
r =0: Yo =0,
O(e) : y;'= 2yo }
r =0 : YI =0, Y; =0
etc.
Here, primes denote differentiation with respect to r.
Solving these problems successively, we obtain
Yo =-"2I r 2
+r
I I
='3 r -12 r
3 4
YI
etc.
Thus,
A comparison between the asymptotic expansion and the exact numerical

solution is shown in Figure I - the two-term asymptotic solution is quite close to
the exact numerical solution! Observe that YI contributes to increase the flight
time of the object. This is due to the fact that the decrease of the gravitational
force with height allows the object to stay up longer.
0.6 ...-----..-----.----""T""-----,
JJ 0.4
:;
"0 Numerical Solution
en 0.3 1----.fC--+ Yo +-_-":"':---:lo~
Yo +€y,
o 0.5 1.0 1.5 2.0

1: -axis
Figure 2.1. Comparison between the asymptotic solutions and the
numerical solution in Example S. In these calculations e =10-1 • There is
little difference between the numerical solution and the two-term expansion
(From Holmes, 1998).
24 Regular Perturbation Metbods
dy
dt
=_y + e /, t> o} .
y(O) =1
Look for a solution of the fonn
Substituting into the above equations and equating coefficients of like powers of
e, we obtain
0(1) : Yo = -Yo}
yo(O)=1
O(e): YI =-YI +y~}

YI(O)=O
0(e 2): Y2 =-Y2 +2YOY1)

Y2(O)=O
etc.
Solving these problems successively, we obtain
-I
Yo =e
YI = e- I - e- 21
-I 2 -21 -31
Y2 =e - e +e
etc.
Thus,
y(t)- e- I +e(e- I _e- 21 )+e 2 (e- I -2e- 21 +e- 31 )+ ...
In order to find the exact solution of this problem, note that the given
equation is a Bernoulli equation; so put
I
y=-.
w
We then obtain
--w=-e
dw }
dt .
w(O) = I
This initial-value problem has the solution
In tenns ofy, this is

1 e- t
Yexact = e+(I-e)e = I+e(e- t -1)'
t
from which
e= 1 : Yexact =1 ,
which shows the fact that the unperturbed problem (corresponding to e =0)
describes motion which dies away, while the motion of the perturbed problem
does not die away! (In fact, when e =1, we have y(t);: 1.) Thus, the "small
perturbation" e alters the character of motion. The expansion of the exact
solution in powers of e gives
y(t)=e- t +e(e- t _e- 2t )+e 2 (e- t _2e- 2t +e- 31 )+ ...
which agrees with the perturbation solution.

Observe also that the convergence of the series in the approximate solution is
uniform, as e ~ 0, for 0 $ t < 00 because, for a fixed t > 0, the error
approaches zero as e ~ O.
Example 7: Consider the initial-value problem for Duffing's equation -

2
u + U + e u = 0, t > 0}
3
d2
-
dt .
u(O) = A, u(O) = 0
Look for a solution of the fonn
u(t;e) - Uo(t)+eu, (t)+e 2u2 (t)+ .. ·.
We then obtain, to various orders in e,
0(1) : Uo + U o = 0 }
U o (0) = A, Uo (0) = 0
O(e): ul +u l = -u~ }
ul(O)=O, ul(O)=O'
etc.
where the dots overhead denote differentiation with respect to t.
Solving the 0(1) problem, we obtain
Uo (t) =A cos t .
Substituting this, the 0(£0) problem becomes
u, + u, = -A~ 00" / =- ~ (3 oost + co,3/ l}

u,(O)=O, u,(O)=O
from which,
A3 3A 3
U, (t) =- (cos3t -cost)-- t sint.
32 8
Thus,
u(t)-Acost+£OA 3 [J...-
32
(cos3t-cost)-~ tsint]+ ....
8
Observe that the second term has a secular (secular is the Latin word used for
century - this adjective comes from the astronomical context) term which causes
the 0(£0) term to dominate the 0(1) term as t becomes larger than £0-'. Note
that it is not possible to alleviate this difficulty by including higher-order terms
because the latter only worsen this difficulty.
In order to see the source of this difficulty, let us look at the exact solution of
this problem.
The first integml of the given equation is given by
1.0 I 0 I 4
- u· + - u· + - £0 u = h
224
where h is a constant of integration. On using the given initial conditions, we
obtain
- I A2 +-
h -- £0 A4 .
2 4
Integmting the first integml further, we obtain
J~
u
=+ du
t _ ,
A 2h -u
2
-"2I lOu 4
which may be rewritten as
Putting
u =-AcosO,
we obtain
which may be rewritten as
t=± 1 j dO
~1 +eA 2 " ~1-m2 sin 2 0
where
2
m 2 = (e/2) A
- 1+eA 2
The period of the oscillation is then given by
T=
2
~1+eA2
J
" dL}
~1-m2sin20
0
17
-;===4= J
,,/2
dO =K(m 2
)
- ~1+eA2 ~1-m2sin20
0
where K (m 2 ) is the complete elliptic integral of the first kind.
For small e, T may be expanded in powers of e,
T=
4 ,J
/2 (1+~sin20+···
2 ) dO
~1+eA2 0 2
=~1+:A2 (~ +m:n +...)

=2n (1-% eA +.. ) 2
Thus, the source of the difficulty with the regular perturbation solution is that
the correction terms of O(e), o(e2 ), etc. do not correct for the difference
between the exact period of oscillation T and the approximate period 2tr of the
leading-order term Acost. The cumulative effect of this error over several
oscillations destroys the uniform validity of the regular perturbation solution.
Example 8: Consider the boundary-value problem -

e y" + y' + y = 0, 0 x I}
$ $
y(O)=a, y(l)=b '
where primes denote differentiation with respect to x.

Look for a regular perturbation expansion of the form
y(x,e) - Yo (x)+eYI (x)+···.
We then obtain at various powers of e:
y~ + Yo = 0 }
YI, + YI = -Yo" .
etc.
Note that the order of the differential equations for yo'y..... is reduced.
Consequently, these differential equations cannot take on both of the boundary
conditions, and one of these boundary conditions, viz., y(O) =a, must be
dropped l . Thus, we have the following boundary conditions -
YO(l)=b}
YI(l)=O .
etc.
We therefore have the following solutions-
Yo = be'-x }
l x
YI =b(l-x) e - •
etc.
Thus,
, We will discuss in Chapter 5 why the boundary condition at x = 0 is the one that need!
to be dropped for this problem.
2.4 Partial Differential Equations 29
Observe that this solution does not satisfy the other boundary condition
y( 0) =a, and so this solution is valid everywhere except near the point x =O.
However, for e« 1, this solution is close to the exact solution everywhere
except in a small interval at x =0, where the exact solution changes rapidly in
order to retrieve the boundary condition there which is about to be lost.
The source of the difficulty with the regular perturbation solution is that the
correction terms of O( e), o(e2 ), ... do not recover the boundary condition
y( 0) = a that is missed by the lowest order solution Yo' The successive terms
actually worsen the behavior near x =a !
2.4. Partial Differential Equations
We will now consider applications of regular perturbation methods to partial

differential equations (see Appendix for a review of partial differential
equations).
Example 9: Consider the nonlinear hyperbolic wave-propagation problem -
Uxx =UxU" , x ~ 0, 1 ~ 0
u/l - )
u(O,/)=ef(/), f(/)=O for 1::;0,.
u(x,O)=O
Let us expand the solution as follows -
u-eu, +e 2u2 + ....
We then obtain from the given equation to various orders in e:
O(e) : u\tt - u1xx =0)

U1 (0,/) = f(/)
u1 (x,O) =0
o( e2) : U211 - U2xx =U)x UItt}

U2 (O,t)=0
U2 (X,0)=0
etc.
Solving the O(e) problem, we obtain
U) (x,t) = f(t - x).

Let us introduce the following characteristics of the (linearized) problem-
~=t - x, 1J = t + x.
The O( e 2
) problem then becomes
iYu 2 = _.!. f'(~) f"(~)

a~drJ 4 '
from which we have
Using the condition u2 (O,t) =0, we obtain
G (t) ='8I [I' (t)r' t .

Thus,
Finally,
Observe the nonuniformity in the above solution. Physically, the latter is due to
the erroneous assumption of constant speed of propagation. The cumulative
effects of variation in the speed of propagation produce errors in the far field.
Thus, two common sources of non-uniformities in asymptotic expansions

are-
• infinite domains which allow long-term effects of small perturbations to
accumulate,
• singularities in governing equations, which lead to localized regions of rapid

changes.
2.5. Applications to Fluid Dynamics: Decay of a Line

Vortex
Consider the decay of a vortex filament in an incompressible fluid

(Shivamoggi, 1998). Let us use the cylindrical coordinates and take the z-axis
along the axis of the filament. The vorticity component along the z-direction
given by
r = duo +~ (2.5.1)
~
rar
where U o is the azimuthal component of the fluid velocity, evolves according to
asdt = v (a2s
ar +.!. as)
ar '
2 r (2.5.2)
v being the kinematic viscosity of the fluid.
a [a 2 Ia ]
Using (2.5.1), equation (2.5.2) becomes
dt (ruo)=v ar (ruo)--; ar (ruo)'

2
(2.5.3)
Non-dimensionalize the various quantities using a reference length L and a
a (ruo)=e [aar2(ruo)-;I ara (rue) ]

reference time r, so that equation (2.5.3) becomes
at 2
(2.5.4)
where
vr
e:=-.
L
The boundary conditions are
r =0 : U o =0 }
for t> O. (2.5.5)
r==)oo : Uo ==) 0
Let us assume e« I and seek a solution of equation (2.5.4) in the form of a
straightforward expansion -
UOapprox - Len Uo. (r,t). (2.5.6)

n=O
Substituting in equation (2.5.4), and equating the coefficients of equal powers
of e, we obtain
(2.5.7)
(2.5.8)
etc.
From the nature of the equations (2.5.7) and (2.5.8), it is clear that their
solutions cannot satisfy both of the boundary conditions (2.5.5), and one of them
must be dropped.
Solving equations (2.5.7) and (2.5.8), and requiring the satisfaction of the
boundary conditions u8 ~ 0 as r ~ 00 , one finds
u8approx - A~r) +e [A"(r)- A';r)] t+.... (2.5.9)
According to (2.5.9), u8approx ~ 00 at r ~ 0 so that the error in (2.5.6) is not

uniform over [0,00), and the expansion (2.5.5) breaks down at the axis. It is of
interest to note that the higher-order terms in the expansion (2.5.5) introduce
successively higher-order singularities at the axis. This simply implies that the
problem (2.5.3) is of singular-perturbation type.
In order to understand further the nature of the nonuniformity, note the exact
solution of equation (2.5.3),
u8exact = ~
2nr
[1- exp(-~)]
4et
(2.5.l0a)
which is in agreement with the first term in (2.5.9) and satisfies the boundary
condition u8 ( 00) ~ 0 (ro can be shown to be the circulation of the vortex
filament at time t = 0). In order to understand what happens at the boundary
r =0, write (2.5. lOa) in the form
u8exact -
_2nr ~ exp (r
~ - 2nr - 4et
2
)
. (2.5.l0b)
The second term in (2.5.l0b) is not negligible even as e ~ 0 since we are

interested in the region r ~ O. Besides, in this form, the order of the error is
uniform in [0,00). The behavior of u8exact is shown in Figure 2.2 together with
the first term of u8approx' Note that near r =0, the motion is a rigid-body
rotation. It can be seen that for small e, u8exact agrees with u8approx except in the
2.6 Exercises 33
£=0
o r
Figure 2.2. Velocity distribution in a vortex fIlament.
small region near the axis where it changes rapidly in order to satisfy the
boundary condition there, which is about to be lost. Physically, this means that
for small times there is a narrow vortex core near the axis, and the rest of the
flow field is irrotational.
For r« FW, we have, from (2.5.l0a),
ro
ueexaci == - - r
81rvt
which corresponds to an almost rigid rotation with angular velocity ro/81rvt.
The intensity of the vortex thus decreases with time as the "core" spreads
radially outward.
2.6. Exercises
1. Using suitable perturbation expansions, find the two roots of the equation
(Hinch,199l)
(I-e) x 2 -2x+l =0.
(Note: The perturbation expansion here is in non-integral powers of e.)
2. Using suitable perturbation expansions, find the three roots of the equation
D:
3
+ x 2 -1 = O.
3. Using the quadratic-root fonnula, find the two roots of the equation
(x -1)(x - r) =-D:.
Expand these roots of powers of e and compare them with the roots obtained
by using regular perturbation expansions for the cases
(i) r*l,
(ii) r == 1.
4. Using a regular perturbation expansion, solve the initial-value problem
(Nayfeh,1973)
dy
dt
+y =t:l, t> o}.
t=O:y=l
Y
d +1
dt
=ty}.
t=O:y=e
Compare the solution with an expansion in powers of e of the exact solution
of this problem. Discuss the change in the nature of the solution produced by
the perturbation (e * 0).
2
-t+(I-ee- at )y=O, a>O
ddt } .
t =0 : y =0, y' =1
Compare this solution in the limit a ~ 0 with an expansion in powers of e
of the exact solution.
7. Using a regular perturbation method to solve the initial-value problem
d
dr
2
----?'- + =f(t) ty } .
t =0 : y =a, y' =b
Compare this solution with an expansion in powers of e of the exact

solution.
2.7 Appendix 3S
8. The asymptotic expansion of J,.(x) for large x is
J,. (x) - I! 1
[cos( x - v; - :) - 4 v: - 1 sin ( x - v; - :)
x
Show that the roots of J,.(x) =0 are given by
3 v) 4v -1
2
x= ( n+4"+2" rc- 2rc(4n+3+2v) + ....
2.7. Appendix
Review of Partial Differential Equations
At. Transformation to Canonical Forms

Consider a second-order, linear partial differential equation (PDE)
AUxx +BUxy +CUyy +DUx +EUy +FU= G (A.l)
where A, B, C, ... are functions of x and y.

Consider a change of independent variables, given by
~ = ~(x,y)}
(A.2)
T/=T/(x,y)
with the Jacobian
J=I~x ~yl~o (A.3)

T/ x T/y
in the region under consideration; here ~(x,y) and T/(x,y) are assumed to be
twice continuously differentiable.
Note the relations -
U x = U~ ~x + U~ 11x
uy = u~ ~y +u~ 11y
un: =u~~ ~; + 2u~~ ~x 11x + u~~ 11; + u~ ~n: + u~ 11n: (A.4)
u.y = u~~ ~x ~y + u~~ (~x 11y + ~y 11x) + u~~ 11x11y + u~ ~.y + u~ 11.y
uyy =u~~ ~: + 2u~~ ~y 11y + u~~ 11: + u~ ~yy + u~ 11yy
Using the relations (AA), equation (A.l) becomes
A* u~~ + B* u~~ + c* u~~ + D* u~ + E* u~ + F* = G* (A.5)
where,
A* = A~; +B~x~y +C~:
B* = 2A~x 11x +B(~x 11y +~y 11x)+2C~y 11y
• 2 2
C = A 11x + B11x11y + C11y
D* =A~n: + B~.y + C~yy + D~x + E ~y (A.6)
E' = A 11n: + B11.y + C11 yy + D11x + E11y

F' =F
G* =G
Noting that
(A.7)
we see that the nature of equation (A. I ) is invariant under the transformation
(A.2), provided J:;; O.
For the purpose of classification, write equation (A.I) in the form
(A.8)
which transforms to
A* u~~ + B' u~~ + C* u~~ =H*. (A.9)
Suppose now that ~ and 11 are such that
A' = A~; + B ~x~y + C~; = 0 }. (A.IO)

C* = A 11; + B11 x11y + C11; = 0
Along the curves ~ = canst and 11 = canst, we have, respectively,

2.7 Appendix 37
d~ : ~xdx + ~ydy ~ 0 }
I
(A.lla)
dT7 - T7xdx + T7 ydy - 0
or
dy = _
dx ~y
~x
(A.llb)
dy T7x·
-=--
dx T7 y
Thus, we have, along the curves ~ =cons! and T7 =cons! ,
(d (dY)
A dx )2 -B dx +C=O,
y
(A.l2)
from which
2
dy = B+.JB -4AC)
dx 2A
(A.B)
dy B-.JB 2 -4AC .
-=
dx 2A
If the solutions of these equations are
1/'1 (x,y) = c\ }
(A.14)
1/'2 (x,y) = c 2
which represent the characteristic curves, then
~ = 1/'1 (x,y)}
(A.l5)
T7 = 1/'2 (x,y)
are the relevant transformations.

2
If B - 4AC > 0, then we have two real/distinct families of characteristics,
and equation (A.9) becomes
(A.16)
If one makes a further transformation
(A.17)
then equation (A.16) becomes

(A.l8)
38 Regular Perturbation Metbods
which is the canonical fonn ofthe hyperbolic PDEs.

If, on the other hand, B 2 - 4AC < 0, then the characteristics are not real;
nonetheless, equation (A.9) becomes again
B* U~ry = H*, B* ~ 0 (A.l9)
where ~ and 17 are now complex conjugate functions. So, if one makes a
further transfonnation
(A.20)
where a and f3 are real functions, then equation (A.l6) becomes

(A.2l)
which is the canonical fonn ofthe elliptic PDE's.

Finally, if B 2 -4AC=O, then two real families of characteristics (A.14)
degenerate into one family.
Suppose that we choose ~ such that
(A.22)
We then have, from (A.6),

B* = 2A~x17x + B (~x17y +~y17x)+ 2C~y17y
(A.23)
=2 (..fA ~x +-fC ~y)(..fA 17x +-fC 17y) =0
for any 17 such that J ~ O.
Equation (A.9) then becomes
C* Uryry = H* , C* ~ 0 (A.24)
which is the canonical fonn of the parabolic PDEs.

One may choose
17=Y (A.25)
for simplicity.
2.7 Appendix 39
Example A.I: / Uxx - x 2 uyy =O.

Here,
So, this equation is hyperbolic everywhere except on the lines x =0 or y =O.

The characteristics are given by
: =; )
dy _ x
----
dx y
or
/ _x 2 =c, and y2 +x 2 =c2 •
Therefore, consider a canonical transfonnation generated by
~=/ _x
2
, 11=/ +x 2 •
The given PDE then becomes
U'n ,., = 2~·

(,11- 1]"') u,, - (}- 11"') Un'
2~· .,
Chapter 3
The Method of Strained

Coordinates/Parameters
3.1. Introduction
This technique is devised to prevent the appearance of secular tenns in

perturbative solutions of equations such as
u+OJ~u=ef{u,u), e«1. (3.1.1)
The secular tenns in this context represent non-unifonnities associated with
infinite domains.
This technique takes note of the fact that the nonlinearities alter the frequency
of the system. This is accounted for by introducing a new independent variable
T =OJI and expanding u and OJ in powers of f as (Poincare, 1892)
2
u -_uo(T)+eu) (T )+f u2 (T)+"'j.
(3.1.2)
OJ - OJo + e OJ) + e 2 OJ 2 + ...
The parameters OJ;, i ~ 1 are chosen so as to prevent the appearance of secular

tenns.
In a generalization of this method (Lighthill, 1949), one expands not only the
dependent variable u but also the independent variable I in powers of f and in
tenns of a new independent variable s as
(3.1.3)
42 The Method of Strained Coordinates/Parameters
The latter expansion can be viewed as a near-identity transfonnation from t to s.

The functions c;m are detennined such that the expansion for u is unifonnly
valid.
3.2. Poincare-Lindstedt-LighthiU Method of

Perturbed Eigenvalues
Suppose we have to find the eigenvalues of the linear operator A perturbed by

a small operator, i.e., (A + e B), e« 1. We look for solutions of
(A+eB)x=Ax (3.2.1)
and suppose that they can be expressed as power series (Poincare, 1892,
Lindstedt, 1882, Lighthill, 1949)-
Lex Le
~ ~
j j
x- j ' ..1= Aj . (3.2.2)
j=O j=O
Substituting these expressions into the eigenvalue equation and collecting tenns
of like powers of e, we obtain
0(1): (A-A o I) X o =0 (3.2.3)
O(e):(A-AoI)XI =A1xo-Bxo (3.2.4)

etc.
Note that the operator L =A - ..1 I, which occurs in every equation, is not
0
invertible, so, according to the Implicit Function Theorem, we cannot solve the
equations for X I 'X 2 ' .•.. However, Fredholm's Alternative Theorem (see
Appendix 1) comes to the rescue and implies that solutions can be found if and
only if the right-hand sides are orthogonal to the null space of the adjoint
operator L· =A· - I o I, i.e., if the null space of L~ is spanned by Yo, then we
have from equation (3.2.4)
AI (xo,yo)-(Bxo'yo) = O. (3.2.5)
Since (xo,Yo)::F- 0, equation (3.2.5) uniquely detennines AI:
A _ (Bxo'yo) (3.2.6)
I - (xo,Yo)
3.2 Poincare-Lindstedt-Lighthill Method of Perturbed Eigenvalues 43
For this value of AI' the solution XI exists and can be made unique by the
further requirement (xl,XO) =O. Similarly, all of the xk's, k> 1, can be solved
in succession.
Example 1: Consider
2
d2
- u + roo2 u =eu }
dt (3.2.7)
du .
t =0 : u =a, -;j( =0
Construct a regular perturbation expansion of the form

u(t;e) - Uo (t) + eU I (t) + .... (3.2.8)
We then obtain from (3.2.7) to various orders of e:

2
o(1) . -ddt2u- + roo2 Uo -_
O
0 }
(3.2.9)
du
t =0 : Uo =a, _ 0 =0
dt
2
o( e) . -,-
d u, + roo2 u, =Uo
dr
}
du (3.2.10)
t=O: u, =0, _I =0
dt
etc.
Thus, we have from (3.2.9),
U o =a cos roo t . (3.2.11)
Using (3.2.11), (3.2.10) then becomes
(3.2.12)
which gives
at .
ul =-- sm roo t . (3.2.13)
2roo
Using (3.2.11) and (3.2.13), (3.2.8) becomes

at .
u (t; e ) ~ a cos COo t + e - - sm COo t + .. '. (3.2.14)
2COo
Observe the presence of a secular term in (3.2.14); the region of nonuniformity
for large t is
t=O(l/e) as e~O. (3.2.15)
Note that the exact solution of(3.2.7) is, however, given by
u = acos (COo~l-e/co~ t) (3.2.16)
which represents a purely periodic oscillation. Therefore, the unbounded

behavior indicated by the regular expansion (3.2.8) is spurious and needs to be
rectified. First, note that the origin of the secular term in (3.2.14) can be seen by
expanding the exact solution (3.2.16) in powers of t:
acos (COo~l-e/co~ t)=acos [COo (1- 2~~ + ...) t] (3.2.17a)
and further expanding the cosine function on the right,
acos (COo~I-e/CO~ t)= a cos COo t . cos (E-+ ...)+asincoot. sin(E-+ ...)
2 COo 2 COo
at .
= a cos COo t + e - - sm COo t + .. "
2COo
(3.2.17b)
in agreement with (3.2.14). Further, the effect of the perturbation is to modify
the frequency from the unperturbed value COo to the perturbed value
COo ~1- e/ co~ . Thus, one introduces a new variable
(3.2.18)
and constructs a new expansion

u (t; e) ~ Uo(s) + e u, (s) + .... (3.2.19)
We then obtain from (3.2.7) to various orders in e:

2
o (1) . -dds2u-o + OJo2 uo -_ 0 )
(3.2.20)
s =0 : uo = a, _du0 = 0
ds
2 2
o ( e) .. d U1
-2-+OJO
ds
2
ds
u, --2OJ,
_ d uO
-2-+UO )
du du (3.2.21)
s=O: u =0 - ' =-OJ _ 0
, 'ds I ds
etc.
Uo(s) =acosOJo s. (3.2.22)
Using (3.2.22), (3.2.21) then becomes

2
d u,
-2- + OJo2 u, =(1+ 2OJ, OJo2) acosOJos. (3.2.23)
ds
One may now choose OJ) so as to remove the secular term on the right-hand
side, i.e.,
2 1
1 + 2OJ, OJ o = 0 or OJ, = - --2 . (3.2.24)
2OJo
Using (3.2.24), we then have, from equation (3.2.23),
u,(s)=O. (3.2.25)
Using (3.2.22), (3.2.24), and (3.2.25), we obtain from (3.2.18) and (3.2.19),
u=acosOJos=acos [OJ o (l- 2:~ +".) t] (3.2.26)
in agreement with the limiting form (3.2.17a) of the exact solution (3.2.16) for
small e.
Example 2: The equation of motion for a simple pendulum is
(; + 1f. sin 0 = 0 . (3.2.27)

£
For small 0, put
(J=(-~/6) u(i), i =~ t (3.2.28)
and drop the hats. We then obtain, from equation (3.2.27), on expanding in
powers of £ , Duffing's equation -
d 2u
- , +u+£u 3 =0. (3.2.29)
dr
Introduce a new independent variable, s, given by the expansion
s=t (1+£(0, +£2(02 +...) (3.2.30)
and consider an expansion
L £n un (s).
~
u- (3.2.31)
n=O
(3.2.32)
(3.2.33)
(3.2.34)
etc.
Solving equation (3.2.32), we obtain
U o = a cos (s + Ip). (3.2.35)
Substituting (3.2.35), equation (3.2.33) becomes
d2~, + U, = _.!.. a 3 cos3(s + Ip) _ (~ a 2 - 2(0,) acos(s + Ip). (3.2.36)

ds- 4 4
Removal of secular tenns in equation (3.2.36) requires
3
4
2
- a - 2(0
'
=0 or (0
,
=-83 a 2 (3.2.37)
We then obtain the following solution for equation (3.2.36)-

a3
u, = - cos3(s+Ip). (3.2.38)
32
Using (3.2.35) and (3.2.38), equation (3.2.34) becomes
2
d u, +U, = (21
-,_. - a4 +2co, ) acos ( s+m ) + nonsecular terms. (3.2.39)
ds· . 128 . '1'
Removal of secular terms in equation (3.2.39) requires

21 4
- a + 2co 2 = 0 or co 2 =- -21 a
4
(3.2.40)
128 256
Using (3.2.35), (3.2.37), (3.2.38), and (3.2.40), we obtain from (3.2.30) and
(3.2.31),
3
a
U- acos(cot +<p)+ e - cos3(cot +<p)+ o( e 2 ) (3.2.41)
32
with
3
co=I+-ea 2 21 e 2 a 4 +0(e 3 ) •
-- (3.2.42)
8 256
Observe that the first-order regular perturbation solution gives an error of O(e)
for t =0(1), but a larger error of 0(1) for t =O( e- I ). However, the first-order
singular perturbation solution gives an error of only O(e) for t = O( e- I ) and
does not lead to an error of 0(1) until t = O( e- 2 ).
Example 3: Consider the Mathieu equation

ii + (0 + e cos 2t) U =O. (3.2.43)
Let us determine transition curves in the (0, e) plane, which separate stable
and unstable solutions and correspond to periodic solutions (see Appendix 2).
Let us consider expansions
o= n 2
+ e 01 + e 2 O2 +... (3.2.44)
u-uO +eu1 +e 2 u2 +··· (3.2.45)

where n is an integer including zero. We then obtain from equation (3.2.43) to
various orders in e,
0(1): iio +n 2 Uo = 0 (3.2.46)
2
O(e): iii +n u1 = -(OJ +cos2t) Uo (3.2.47)
0(e 2 ): ii 2 +n 2 u2 = -(01 +cos2t) U1 -02 Uo (3.2.48)

etc.
We now have the following different cases to consider.
Case (i): n = O.
We now obtain from equation (3.2.46),
Uo = 1, say. (3.2.49)

U, = -0, - cos2t. (3.2.50)
Uniform validity of the solution of equation (3.2.50) requires

0, =0. (3.2.51)
We then have from equation (3.2.50),
u, =-41 cos 2t + A . (3.2.52)
Using (3.2.49), (3.2.51), and (3.2.52), equation (3.2.48) becomes
U
.. = s:
-(1 -
1
- -
1
A cos 2t - - cos 4t . (3.2.53)
2 2 8 8
1 1
O2 + - = 0 or O2 = - - . (3.2.54)
8 8
o=-i e +0(e 2 3
). (3.2.55)
Case (ii): n =1.

Uo =cost. (3.2.56)
.. +u 1 =- (s:(11 +'2 cost-'2 cos3t.

U,
1) 1 (3.2.57)

1 1
0, + - = 0 or 0, = - - . (3.2.58)
2 2

1
u, =- cos3t. (3.2.59)
16
u..2 +u2 =- (1 5:) cost+-

-+v2 1 cos3t--
1 cos5t. (3.2.60)
32 32 32
1 1
0, + -
• 32
=0 or 0,
•
=- -.
32
(3.2.61)
o=1-.!.2 e - ~
32
e 2 + O(e 3
). (3.2.62a)
Had we instead taken for the solution of equation (3.2.46),

Uo =sint, (3.2.63)
we would have obtained in place of(3.2.62a),

5: 1 1 2 (3)
v=1+-e--e+Oe. (3.2.62b)
2 32
Case (iii): n =2 .
Uo =cos2t . (3.2.64)
..
u1 + 4u. =--1 - 5: cos2t - -1 cos4t.
VI (3.2.65)
2 2
0, =0. (3.2.66)

1 I
u = - - + - cos 4t . (3.2.67)
, 8 24
u2 + 4u2 =-(02 - 2.-)

48
cos2t - ~ cos6t.
48
(3.2.68)

5
0, - -
. 48
=0 or 0,
.
=-485 . (3.2.69)
o= 4 + 2-
48
e + o( e
2 3
). (3.2.70)
Had we instead taken for the solution of equation (3.2.46),

Uo =sin2t (3.2.71)
we would have obtained in place of(3.2.70),
O=4--.!..- e 2 +0(e 3 ). (3.2.72)

48
The above results are schematically shown in Figure 3.1.
o 4
Figure 3.1. Parameter diagram for Mathieu equation:

Shaded regions unstable.
The method of strained parameters gives the transition curves and the
periodic solutions along them, as discussed in the foregoing. In order to
determine the solutions in the neighborhood of the transition curves, we use
Whittaker's method (Whittaker, 1914). For this purpose, we cast the solution
first in the normal form
u(t) = v(t) eat. (3.2.73)
Equation (3.2.43) then becomes
v+ 20' v+ (0 + 0'2 + e cos 2t) v = O. (3.2.74)
Near the transition curves, a is small, so one may look for a solution of the
form-
v(t;e) - Vo (t) + eVI (t) + e 2 v2 (t)+··· (3.2.75)
8 =80 + e 81 + £2 82 + ... (3.2.76)
a=£a l +e 2 a 2 +.... (3.2.77)
Substituting these expansions into equation (3.2.74), we obtain

0(1) : Vo +80 Vo =0 (3.2.78)
O(e) : VI +80 VI =-2a l Vo - 81 Vo - Vo cos2t (3.2.79)

etc.
We have from equation (3.2.78)-
Vo =acos~ t+bsin~ t. (3.2.80)
Now, according to the Floquet theory (see Appendix 2), v(t) is periodic, with
period 1r (21r), if
80 =n 2 , n even (odd) integer. (3.2.81)

VI +n 2 VI =-2a l (-ansinnt+bncosnt)-8] (acosnt+bsinnt)
- (acosn t + bsinnt) cos2t (3.2.82)
which may be rewritten, for n =I, as
VI + VI =(2a 1a - 8, b + %) sin t
-(2a l b + 81 a + ~) cost + nonsecular terms. (3.2.83)
(3.2.84)
The solvability condition for equations (3.2.84) gives
4a~ - ( ±-
8,2 ) = 0 (3.2.85)
52 The Method of Strained CoordinatesIParameters
from which,
(J, =±~ ~±_liI2. (3.2.86)
Using (3.2.86), equations (3.2.84) give
.!.+li
2
b =+- -1-- a,
I
li, :1:-1/2. (3.2.87)
--li
2 I
Using (3.2.73), (3.2.80), (3.2.81), (3.2.86), and (3.2.87), we obtain
U-
.!.
al e2
'~.!.-o'
4 I
I [
cost - i---
.!. + lil sint ]
--li
2 I
,
+a e- 2I'V'4o,1 .!. + li sint ] +...
fI;, [ cost+ _2- (3.2.88)
2 I '
--li
2 I
liI :I:-.!.
2
which shows that the solution is bounded when li~ >.!. and unbounded when
4
lil2 <.!.. Therefore, the transition from stability to instability is described by
4
li~ =.!., and the transition curves emanating from li =I are given, from (3.2.76)
4
and (3.2.81), by
I
li=I±-f+'" (3.2.89)
2
in agreement with the result (3.2.62) given by the method of strained
parameters.
3.3. Eigenfunction Expansion Method
Consider a perturbed linear eigenvalue problem

3.3 Eigenfunction Expansion Method 53
U"+[A+£j(X)]U=O, j(X)=j(-X), u(O)=u(I)=O. (3.3.1)
Corresponding to £ =0, we have, from (3.3.1),

Uo =.fi sin n rr x, Ao_= n rr
2 2
}.
(3.3.2)
n -1,2,3
For £ :F- 0, let us put (Schrodinger, 1926)

'2 sinnrr x+eunl +e 2 un::! +...
Un - vi. (3.3.3)
(3.3.4)
Using (3.3.3) and (3.3.4), equation (3.3.1) gives to various orders in £,

2 2
0(£) : u;' + n rr un = - j(x) Un - An Un }
I I 0 I 0
(3.3.5)
Un (O)=U n (1)=0
I I
2 2 2
0(e ):u; +n rr Un =-j(x)Un -An Un -An un}
2 1 I I I 2 0
Un, (O)=un,(I)=O . (3.3.6)

etc.
Let us expand un I in terms of the eigenfunctions (3.3.2) of the 0(1) problem,

as follows-
Un, = L. anm.fi sinmrrx.

m=1
(3.3.7)
Then equation (3.3.5) gives
L. .fi rr 2
(n 2 _m 2 ) anm sinmrr x
(3.3.8)
m=1
=-.fi j(x)sinnrrx-.fi An , sinnrrx.
Multiply equation (3.3.8) by .fi sin k rr x, and integrate from 0 to 1; we then

obtain
(3.3.9)
where
Jo j(x) sinn rr x sinkrr x dx =0n'

I
F"k =2 (3.3.10)
If k =n, equation (3.3.9) gives

54 The Method of Strained CoordinatesiParameten
J f(x) sin nnx

I
2
An, =-F,.n =-2 dx. (3.3.11)
o
If, on the other hand, k;:;:. n, equation (3.3.9) gives
F,.k
(3.3.12)
Using (3.3.12), equation (3.3.7) gives

F
un =- L '( ,nk ,).fi sinknx+ann.fi sinnnx. (3.3.13)
, k"n n- n- - k-
In order to determine ann' we need to proceed to equation (3.3.6). Let us

expand un, as follows-
~
un, =L bnr.fi sinrnx. (3.3.14)

r=1
Equation (3.3.6) then gives

~
n 2 L (n 2 -r 2 ).fi bnrsinrnx
r=1
~ ~
= - Lank.fi f (x) sin k n x - Lank An, .fi sin k n x

i=1 bl
- An, .fi sin nn x. (3.3.15)
Multiply equation (3.3.15) by .fi sins n x, and integrate from 0 to 1; we then

obtain
2
n (n
2
- S2) bns =- L ani 0s - ans An, - An, ~ns . (3.3.16)
i=1
If s =n, equation (3.3.16) gives

~ ~ F,.2k
An,=-L ani 0n=L 2( 2_k2)' (3.3.17)
k"n bn n n
(3.3.18)
Let us normalize un as follows -

f (Uno+EUn,+E2Un,)
J 2
dx=l, (3.3.19)
o
from which we have
J
fo Un Un dx = 0
o ,
f (2uno un,+u;,)dx=0
I (3.3.20)
o
etc.
Using (3.3.7) and (3.3.14), equation (3.3.20) gives
(3.3.21)
Using (3.3.2), (3.3.7), (3.3.17), and (3.3.21), we have from equation (3.3.3),
r;:;
un --y2 sinnnx-e £.J
~
2( F"t
2 2) r;:;
-y2 sinknx
bn n n - k
~ {[~
+ 10 2 £.J£.J F"s ~s
--,--:---.....,....,.-:---~
2 2
F"n F"t 2 ] -yr;:;
2 sin k n x
s"n n (n _s2)(n -e) n 4 (n 2 -e)
4
t"n
_.!. F,,2t 2.fi sin n n x} + 0(10 3 ), (3.3.22a)

2 n
4
(n 2 -e)
A.=n2n2-eF"n+e2 L 2(~2t 2)+0(10

t"n n n -k
3
). (3.3.22b)
3.4. LighthiU's Method of Shifting Singularities
Lighthill's (1949) technique is based on the premise that the source of some
non-uniformities may be traced to the fact that the regular perturbative solution
may have the right form, but not quite at the right place. Lighthill (1949) then
suggested that this non-uniformity may be remedied by straining the
independent variable.
Example 4: Consider (Van Dyke, 1975)
(x+ey) dy +Y=I)
dx • (3.4.1)
y(l) =2
Observe that whereas the singularity for the full equation (3.4.1) is along the
line x = -e y, the limit e ~ 0 shifts it to x =0 (see Figure 3.2). The method of
strained parameters (Lighthill, 1949) makes it move toward the true position
even as e ~ 0, (so long as e:t:- 0).
A straightforward perturbation-expansion gives the following solution -
2
l+x (l-x)(I+3x) 2 (l-x )(1+3x)
y---e 3 +e 5 +... (3.4.2)
x 2x 2x
which breaks down as x ~ O. Further, the singularity at x =0 is compounded
at higher orders.
Figure 3.2. Shifting of singularities.
Let us strain the independent variable x as follows-

x = X o + e XI (x o) + .... (3.4.3)
Equation (3.4.1) then gives
[x o +e(x, + y)] dy . (l-e dx l )+y= 1, (3.4.4)

dx o dx o
from which we obtain to various orders in e-

0(1) : X o dy + y =1 (3.4.5)
dx o
O(e): (x, +y) dy -xo dy dx, =0

dx o dx o dx o (3.4.6)
etc.
c+xo
y=--, (3.4.7)
Xo
c being an arbitrary constant.

dx, x, _ c+xo
(3.4.8)
dx o - X o -~'
from which
c+2xo
x =---- (3.4.9)
, 2xo

c+2x
x=-x-e---o (3.4.10)
o 2X o '
from which
c+2x
Xo =-x+e - - . (3.4.11)
2x
Using (3.4.11), we have, from equation (3.4.7),
c+2x
c+x+e ~ 2x 2 +2(c+e) x+ec
y=- =- (3.4.12)
c+2x
x+e-- 2x 2 +2ex+ec
2x
Imposing the initial condition
y(I)=2, (3.4.13)
58 The Method of Strained Coordlnates/Parameten
equation (3.4.12) gives
3e
c=I+-+0 (')
e- . (3.4.14)
2
Observe that the singular perturbative solution (3.4.12) no longer shows a
singularity on the line x =0 !
On expanding in powers of e, the singular perturbative solution (3.4.12) can
be written as
y - 1+-1 + e
x
(3X 2
-lx-I) + (')
2x
3 0 e- , x:;t 0 (3.4.15)
in agreement with the regular perturbation solution (3.4.2)!

Next, in order to find the exact solution for this problem, put
z=x+ey (3.4.16)
so that equation (3.4.1) becomes

dz
z -=e+x. (3.4.17)
dx
The solution of equation (3.4.17) is
z =[2e x + x 2 + c] 1/2, (3.4.18)
c being an arbitrary constant. Thus,
x
(3.4.19a)
e
Imposing again the initial condition (3.4.13), equation (3.4.18) gives
2
c =4 +-. (3.4.20)
e
Using (3.4.20), equation (3.4.19a) becomes
2
y= [ (~)
2
+2 C: x
)+4
]1/ x
e
(3.4.19b)
On expanding this in powers of e, this can be written as

y - 1+-1 + e (3X -2X-l) + 0 ()
2
e
3
2
(3.4.21)
x 2x
in agreement with the regular perturbative solution (3.4.2)!

The exact solution (3.4.19b) as well as the regular perturbative solution

(3.4.2) are schematically shown in Figure 3.3.
,,
\
\
o \ Linearized
I \
i \
\
,
X=-EY~" i ~
Exact
......,;---_.
" 1 ,I
, 0
, 0
'I'0
----T------------·------~x
o
I
Figure 3.3. Integral curves for equation (3.4.1) (from Van Dyke, 1975).
3.5. Pritulo's Method of Renormalization
Sometimes a perturbation expansion can be rendered unifonnly valid by

introducing the straining of the independent variable directly in the solution
rather than in the equation (Pritulo, 1962).
Example 5: Consider the initial-value problem for Duffing's equation -

d 2u
- , +u+eu 3 =0, u(O)=a, u(O)=O. (3.5.1)
dr
Look for a straightforward expansion -
u(t;e) - Uo(t) + eU I (t) + ... (3.5.2)
60 The Method of Strained CoordinateslParameten
which leads to
(3.5.3)
(3.5.4)
etc.
We obtain from (3.5.3) and (3.5.4),

Uo = acost (3.5.5)
3a 3 a3
u1 =- - t sin t + - (cos 3t - cos t)
8 32 (3.5.6)
etc.
3
3a3 a ]
u(t;e)-acost+e [ -8 tsint+ 32 (cos3t-cost) +.... (3.5.7)
Pritulo (1962) suggested that one may introduce the straining of the
independent variable
t =s (1 + e (0, +...) (3.5.8)
directly into the solution (3.5.7). Expanding and collecting the coefficients of
equal powers of e, we obtain
1 3
u(t;e) - acoss + e [ -a((0, + %a
2
) ssins + 3 2 a (cos3s - coss)] + .... (3.5.9)
Removal of secular terms in equation (3.5.9) then requires
(0, =--3 a 2 . (3.5.10)

8
Therefore, a uniformly valid expansion is
1
u(t;e)-acoss+- ea 3 (cos3s-coss)+···, (3.5.11)
32
where
t=s (I-~ £a 2
+ .. J (3.5.12)
in agreement with the result (3.2.41) and (3.2.42) obtained by the method of
strained parameters!
Example 6: Consider the nonlinear hyperbolic wave-propagation problem -

a 2u a 2u au a 2u
at 2 - ax 2 = ax at 2
u(o,t)=£f(t), f(t)=O for t~O . (3.5.13)
u(x,O)=O
Let us expand the solution as follows -

u - £ u1+ £2 u2 + .... (3.5.14)
We then obtain from (3.5.13) to various orders in £:
0(£) : a2~1 _ a2u~ =

ar ax·
o} (3.5.15)
u1(O,t) = f(t) ,
2
0(£2) : a2~2 _ a u; = a u1 a2~1}
ar ax· ax ar
(3.5.16)
u2 (O,t) = 0
etc.
Solving (3.5.15), we obtain
u,(x,t)=f(t-x). (3.5.17)
Let us introduce the following characteristics of the linearized problem

(3.5.15)-
~ =t - x, 1] =t + x . (3.5.18)

2
a u =_.!.4 f'(~) f"(~) '
a~dry
2
(3.5.19)
from which we have

1 ]' 71+G(~).
U2(X,t)=-8[f'(~r (3.5.20)
Using the condition U2 (O,t) =0, we obtain

G(t)=81 [f'(t)]"t.
'
(3.5.21)
i [f'(~W (~-71) -± [f'(~W x.

u2(x,t) = = (3.5.22)
u(x,t)-ef(~)-e2 {± [f'(~W x}+ .... (3.5.23)
Observe the nonuniformity in (3.5.23). Physically, the latter is due to the

erroneous assumption of constant speed of propagation in the far field. The
cumulative effects of variation in the speed of propagation produce errors in the
far field. In order to remove this nonuniformity, let us strain one of the
linearized characteristics as follows (Lin, 1954 and Fox, 1955)-
(3.5.24)
-± [f'(~O)]2
u(x,t) - e f(~o) +e 2{~I f'(~o) x} +.... (3.5.25)
The nonuniform term in (3.5.25) can be eliminated by choosing
~I = ~ f'(~o)' (3.5.26)
Using (3.5.26), equation (3.5.25) then becomes
u(x,t)-ef([t-x J- e x f'[t-x J)+ .... (3.5.27)

4
Thus, the uniformly valid first-order solution for the nonlinear problem is simply
the solution for the corresponding linear problem with the respective
characteristic replaced by the characteristic calculated by including the first-
order nonlinearities in the problem (Whitham, 1952). The variation in the speed
of propagation is thus described by "bending" the linearized characteristics.
3.6 Wave Propagation in an Inhomogeneous Medium 63
3.6. Wave Propagation in an Inhomogeneous Medium
The method of strained parameters is used in the following to treat the

general problem of wave propagation in an inhomogeneous medium through the
model example (Shivamoggi, 1978c)-
ul/-{I+ea(x,t)ruxx=O, e«I} (3.6.1)

u(x,O)=f(x), u,(x,O)=g(x)
where the inhomogeneities in the medium are modeled by the parameter a(x,t).
The main idea is to obtain a uniformly valid solution by straining the
characteristic corresponding to the wave propagation in a homogeneous medium
- in analogy to that in Lin's method (1954) for a nonlinear hyperbolic system
(Section 3.5).
Thus, introduce the characteristic parameters
~,11 =x +kt (3.6.2)
and seek solutions of the form
(3.6.3)
where the functions 1C (~, 11) are determined by imposing the condition that
(3.6.3) be uniformly valid for large distances.
Putting a(x,t) = a(~, 11), one obtains from equation (3.6.1),
0(1) : :;~ = 0, (3.6.4)
d U 1 _ a(~, 11) [
2
]
o (e) .. d~ dry - - - 2 - u044 + 2u047J + U07J7J
1C(g,11) [
+ - 2 - u044 - 2u047J + U07J7J
]
1 dIC (g, 11) 1 d1C( g, 11)

+'2 d~ u04 +'2 dry U07J ' (3.6.5)
etc.
From equation (3.6.4),
U o(;, 11) = Ap(;) + Bq(11), (3.6.6)
where A and B are arbitrary constants determinable by the given initial

conditions. For the sake of determining 1((;,11), one may consider only the
right-running waves so that Uo =Uo (;); then the removal of the secular terms in
equation (3.6.5) requires that
01((;,11) +1((;,11) uo~~ -0(;,11) uo~~ =0, (3.6.7)

d; uo~ uo~
from which
or
(3.6.8)
Thus, the uniformly valid first-order solution for wave propagation in an

inhomogeneous medium is simply the solution corresponding to the wave
propagation in a homogeneous medium with the respective characteristic
replaced by the characteristic calculated by including the first-order
inhomogeneities in the medium. In other words, the solution corresponding to a
homogeneous medium, when the first-order inhomogeneities are included, may
still have the right form, but not quite at the right place.
For the case 0(;,11) == 1, equation (3.6.8) gives
I( =1, (3.6.9)
and if g(x) = - f'(x), equation (3.6.6) becomes
1 2+£ f [x -(1 +e) t ]
U o (;,11) = - -
2 l+e
(3.6.10)
+.!. _e_ f [x + (1 + e) t]
2 1+e
which is the exact solution to (3.6.1) with a(x,t) = I!
3.7 Applications to Solid Mechanics 65
3.7. Applications to Solid Mechanics: Nonlinear

Buckling of Elastic Columns
The buckling problem involves the investigation of a potentially unstable

equilibrium between the external loading and the internal resistance of the
column. A sudden breakdown of the internal resistance constitutes a
characteristic feature of the buckling phenomenon, regardless of whether at the
instant offailure the elastic limit is exceeded or not.
Under a compressive load P, a possible state for the column is that of pure
compression; but experience shows that for sufficiently large values of P,
transverse deflections can occur. The classical linear theory for infinitesimal
deflections due to Euler (see Timoshenko, 1959) establishes the conditions
under which a column under the action of a concentrated axial load can take up
equilibrium configurations other than the one corresponding to a uniform
compression. This phenomenon is discussed here as a bifurcation scenario.
A characteristic feature of a bifurcation scenario is that the load is a
monotonically increasing function of the deformation. I It is to be noted that
although the critical load of bifurcation problems is a characteristic value
corresponding to the linear problem, the study of the supercritical behavior in a
bifurcation scenario requires the consideration of the nonlinear problem.
Besides, the linear theory leads to lateral deflections whose magnitude remains
indeterminate. The bifurcation problem associated with the buckling of elastic
columns is treated here using the method of strained parameters.
Consider a long, slender, inextensible column simply supported at both ends,
and subjected to a centrally applied compressive load P at its upper end (see
Figure 3.4). The column is assumed to be perfectly straight and of uniform
cross-section, while the material is homogeneous and it behaves elastically. The
shape of the deflected central line of the column is described by y(x), which
satisfies the following differential equation and the boundary conditions:
I The buckling of a column can also occur via a snap-through scenario which is
characterized by the occurrence of a jump from one stable state to another stable state, by
passing unstable stables between the two.
I·
I·
x
I·
~
\\
II
II
II
II
x=0 ----------_.
y
Figure 3.4. Buckled column.
2 2
d y2 + PL y [1+(dy )2]3/2 =0]
dx EI dx ,
y(O) = y(l) =0 (3.7.1)
where all distances have been nondimensionalized using L, E being the modulus
of elasticity, and I the moment of inertia of the cross section.
If (dy/dx)« 1, (3.7.1) may be written as
2
d y 12oY _ 3A.~ (dY )2}
dx 2 +llo --T y dx , (3.7.2)
y(O) =y(l) =0
where
2
)} = PL
0- EI .
Look for solutions of the fonn
(3.7.3)
One then obtains from (3.7.2),

2
O(e): dd YI> +).,20 YI = 0 )
(3.7.4)
YI~~)=YI(I)=O '
2
o( e2).. -d2Y2- + 1100
12 Y2 = 12 1
-21100 IIoIYI
)
ds , (3.7.5)
Y2(O) = Y2 (1) =0
Oe
(
3). d2
Y3 12 _ 121 1 (1 1) 3).,~
2 2
'-2-+IIoOY3--211001loIY2-1Io01lo1+211o2 Y I - - Y I -
(dyl)2
~ 2 ~
Y3(O) = Y3(1) = 0 . (3.7.6)

etc.
From (3.7.4), one obtains the characteristic values and the characteristic
solutions corresponding to the linear problem
YI(s)=A1sinnJfs }
(3.7.7)
=n n ; n =1,2, ...
2 2 2 '
).,0
The plot lYl vs. ).,~ is sketched in Figure 3.5 - a picture, which on physical
grounds is not acceptable. Therefore, in order to have a reasonable description of
the buckling phenomenon, one proceeds to include the nonlinear effects.
Using (3.7.7), it is obvious that the removal of the secular tenns in (3.7.5)
requires
(3.7.8)
Equation (3.7.5) then yields
(3.7.9)
68 The Method of Strained CoordinatesiParameten
Figure 3.5. Buckling amplitude vs. load: Linear theory.
Using (3.7.7)-(3.7.9), equation (3.7.6) becomes

2 4 3 4 3
----2:l.
d + A2 .y = _n2 A, A 3A A ) sin nns - _o_1
- _o_1 3A A sin 3nns}
° °-
(
2
ds 3 I 8 8 , (3.7.10)
Y3(0)=Y3(1)=0
so that the removal of the secular terms in (3.7.10) requires
1
/\', =--3 /\'1"0 A - =--
3 ",
n-n-Aj.
I (3.7.11)
- 16 16
3n 2 n 2 A 3
Y3(S)=- I sin3nns. (3.7.12)
64
Thus, on using (3.7.7)-(3.7.9) and (3.7.11), (3.7.3) becomes
Y (x;e) - e AI sin Ax + 0(e 2 ), (3.7.13)
where
or
(3.7.14)
so that the column buckles into a shape given by (3.7.13) with an amplitude AI'
Note the bifurcation phenomenon, (see Appendix 3 for a review of bifurcation
theory) which corresponds to the branching of a nonlinear solution from the
solution corresponding to the undeflected configuration at A," = A,~ = n"n", as
shown graphically in Figure 3.6.
Figure 3.6. Buckling amplitude vs. load: Nonlinear theory.
This simple example also affords a somewhat exact fonnulation which may
be used to check the accuracy of the results obtained in the foregoing using the
method of strained parameters. The coordinates sand () are introduced to
describe the deformed configuration (see Figure 3.7). The potential energy of
the deformed column is given by
u=-
2 0 ds
f-
El L(dO)2 ds+P J
{L cosOds-L } .
0
(3.7.15)
x
Figure 3.7. Buckled column geometry.
Extremization of U yields
d 20
El - , +PsinO=
ds·
° (3.7.16a)
x=O,L: El(dO/ds}=O,
Writing (3.7.16a) as
-d [El
- (dO)2
- -PcosO] =0,
ds 2 ds
one obtains
-El (dO)2
- =P cosO+c (3.7.16b)
2 ds
where c is an arbitrary constant. Let the boundary conditions be
dO
x=O,L: -=0, O=±a, (3.7.17)
ds
so that equation (3.7.16b) becomes
-dO =[2P
- (cos 0 - cos a)
]1 /2 (3.7.18)
ds £1
Also noting that
sinO = dyjds, (3.7.19)
one obtains from equation (3.7.16a),
£1 dO
y=-p ds'
and using equation (3.7.18), this becomes
y =-~2£ljP (cosO - cosa)J/2. (3.7.20)
We have from equation (3.7.20),
Ymax _ 2 sin (aj2)
(3.7.21)
T- 11: ~P/(11:2£I/L2)'
a trend which agrees with that given by (3.7.14) using the method of strained
parameters.
3.8. Applications to Fluid Dynamics
3.8.1 Nonlinear Hyperbolic Waves in a Gas

Uniformly valid simple-wave solutions are developed here for nonlinear
hyperbolic equations using the method of strained parameters (Shivamoggi,
1977a and 1978a).
The potential function ep (the fluid velocity v =Vep) of the flow due to the
propagation of nonlinear acoustic plane waves in an inviscid, irrotational, and
initially quiescent medium is governed by
;iep 202ep 0 (oep)2 of/J o2f/J 1 (Of/J)2 02 f/J

ot2- C ox2=-ai ox +(I-r)aiox2-"2(I+r) ox ox2' (3.8.1)
where c is the speed of sound in the undisturbed medium and r the ratio of
specific heats of the gas. We consider only simple right-running waves subjected
to the boundary condition
f/J(O,t) = Ej(t), (3.8.2)
where E is a small nondimensional parameter.
Since we are looking for travelling waves, introduce characteristic

coordinates -
): _kx
~,7J = t +-, (3.8.3)
c
and seek solutions of the boundary-value problems (3.8.1) and (3.8.2) of the
form
n=1
k(~;e) = I +eJ((~)+O(e)2, (3.8.4)
so that one obtains from equation (3.8.1),

a'
O(e):~=O (3.8.5)
as, aS 2
O(e 2):4 a24J2 =[2aJ(+2J(d24J,/dS~ _r:1 d2~,]d4J', (3.8.6)
as, aS 2 as, d4J, Ids, c ds, ds,
etc.,
where
_x
S'.2 = t+-.
C
Using (3.8.2), one obtains from equation (3.8.5),
4J1 (s,) =f (sJ (3.8.7)
The removal of the secular terms in equation (3.8.6) requires

2
2 aJ( +2J( d24JI/ds~ r+ 1 d 4J, =0 (3.8.8)
as, d4J, Ids, c2 dS,2
from which,
or
2
J( = I fS'(d4J') r +2I d 4J,2 ds
Ids ' "ds, 2c ds , I
d'Ato 1
or
(3.8.9)
Using (3.8.4) and (3.8.9), we have from (3.8.3),
s, =t-~=~+e
C
Y+/ x
4c
(#1).
ds,
(3.8.10)
Writing <p, (s,) as a Taylor series about s, = ~, and using equation (3.8.7),
one readily obtains
(3.8.11)
The flow velocity is then given by
<Px --~ f'(~) [1+e :;/ Xf"(~)]+0(e3), (3.8.12)
which contains a secular term. In order to eliminate it and render the expression
uniformly valid, we use Pritulo's method, and put
(3.8.13)
so that the removal of the secular term in (3.8.12) requires
y+l
(J(8)=--~
4c
8)
f '( t - - +O(e).
c
(3.8.14)
Using equation (3.8.14), we have from (3.8.10) and (3.8.13),
(3.8.15)
from which
8=!(S+x). (3.8.16)
2
Using (3.8.15), we have from equation (3.8.15),
S=±(S+x)[I+e :;/ f'(t-~)]+0(e3)

or
s=x [1+£ ~:/ I'(t-~)]+O(£T (3.8.17)
Using (3.8.13) and (3.8.17), we have from equation (3.8.12)
¢>x --~ f' {t-~ [1+£ ~:/ I' (t-~)]}+O(£3) (3.8.18)
which is uniformly valid.

Equation (3.8.18) implies that the uniformly valid first-order solution for the
nonlinear problem is simply the solution for the corresponding linear problem,
with the respective characteristic replaced by the characteristic calculated by
including the first-order nonlinearities in the problem, as we saw previously in
Example 6.
For compressive waves, the solution (3.8.11) contains regions of multivalues,
which have to be resolved by the introduction of shocks. Since the potential
function is continuous across shocks, their locations are given by
¢> (ga ' X ) ~ ¢> (gb , x)) , (3.8.19)

t(ga ' x) - t(gb ' x)
where ga and gb denote the characteristics ahead and behind the shocks. One
then has
f( ga) - f( gb) = £ ( ~:31 ) [1'2 (ga) - 1'2 (gb)]).

x
gb -ga =£ (~:/) x [I'(ga)- I'(gb)] (3.8.20a)
Eliminating £x, one obtains
(3.8.20b)
which is essentially the equal area rule (Landau, 1945; Whitham, 1952).
Incidentally, equation (3.8.20b) shows that the slope of the line ¢>, = const is
equal to the arithmetic mean of the slope of the characteristics in the
unperturbed flow and the slope of the characteristic at the point in question. For
the case of a shock running into the unperturbed flow, this is one form of the
bisector rule as pointed out earlier by Van Dyke (1952).
3.8.2 Rayleigh-Taylor Instability of Superposed Fluids

The instability of the interface between two fluids having different densities
and accelerated towards each other is called the Rayleigh-Taylor instability
(Chandrasekhar, 1961 and Shivamoggi, 1998). A static state, in which an
incompressible fluid of variable density subjected to a vertical acceleration is
arranged in horizontal strata and the pressure and the density p are functions of
the vertical coordinate z only, is obviously a kinematically realizable one.
However, whether this state is also dynamically realizable is related to the issue
of stability with respect to small disturbances.
If viscosity is neglected, the flow can be assumed to be irrotational and the
velocity potential of each fluid satisfies the Laplace equation. The difficulty in
solving this type of problem, however, arises from the nonlinear boundary
conditions at an unknown interface. On the other hand, if one assumes the
disturbance-amplitude to be infinitesimal, the boundary conditions can be
linearized and a solution to the linear problem can be readily obtained. For the
linear problem of two different inviscid and incompressible fluids in contact and
subjected to an acceleration directed from the heavier fluid towards the lighter
one, it turns out, for the case of zero surface tension, that any slight disturbance
in the plane interface grows exponentially with time. The effect of the surface
tension in the linear problem is to produce a critical wave number kc ' so that the
interface is unstable or stable according as the wave number is less or greater
than kc .
(a) The Linear Problem

The two fluids are taken to be inviscid and incompressible, and if the motion
of the whole system is supposed to start from rest, it may be assumed to be
irrotational. The applied acceleration g' is directed from heavier to lighter fluid
(see Figure 3.8). Initially, the interface is taken to be disturbed according to a
simple sinusoidal standing wave with an amplitude a' and a wavelength ,1,'.
If y =1J denotes the disturbed shape of the interface, then, one has with the
velocity potentials, j' such that the fluid velocity v is given by v = - V,
2
Y§1J: V <1>j =0; j=I,2. (3.8.21)
PI> pz
Figure 3.8. The interface accelerated normally
and subjected to a perturbation.
Further, one has the following boundary conditions at the interface:

(i) Kinematic Condition: This expresses the fact that a fluid particle initially
on the interface must stay on the interface during the course of perturbation.
This requires
y= 11 : 11, -11 x et>jx +et>jy =0; j =1,2. (3.8.22)
(ii) Dynamic Condition: This refers to the force balance across the interface.
In order to derive the dynamic condition, note that in the presence of a surface
tension T', at the interface, for equilibrium, the normal stresses acting on the
two sides of an interfacial element dS' must differ by an amount given by
where p' is the normal pressure exerted by the fluid; ilk is the outward normal
to dS', and R; is the principal radius of curvature of dS'. The convention
regarding signs is that R; is to be considered positive if the corresponding center
of curvature lies on side 1 of dS'. For the case under consideration, note that
1 , ( ,,)-3/2
R' = 11x'x' 1 + 11x ; •
I
The dynamic condition at the interface becomes
where
and the initial conditions are

(=0: 1J=ecosx, 1J1 =0, (3.8.24)
where
e == a'(2n/}..').
The conditions at infinity are
y~(-I/oo: <l>jX ~O. (3.8.25)
All quantities in equations (3.8.21)-(3.8.25) have been nondimensionalized with

respect to a reference length }..'/2n, a time (}..'/2ng't, where primes denote
dimensional quantities.
Make a change of variable
r=a!, (3.8.26)
so that the boundary-value problem (3.8.21)-(3.8.23) becomes

y;; 1J : V'2<1>j = 0; j = 1,2 (3.8.27)
y=1J: a1J r -1]x jX +jy =0 (3.8.28)
y= 1J: a<l>'r -± (V',)2 +1J-S [a<l>2r -± (V'S +1J]
= -k 21Jxx ( 1 + 1Jx2)-3/2 • (3.8.29)
Upon linearizing the system (3.8.27)-(3.8.29), (3.8.24) and (3.8.25) we

obtain
y;; 0: V'2lfrj = 0; j = 1,2 (3.8.30)
y = 0 : a 1J r + lP jy = 0 (3.8.31)
y=O: alP'r+1J-s (alP2r +1J)+k21Jxx =0 (3.8.32)
y~(-IYoo: lPjy ~ 0 (3.8.33)
r =0 : 1J =cos x, 1Jr =O. (3.8.34)
From (3.8.30), (3.8.31), (3.8.33), and (3.8.34), one obtains

1J(x, r) =cosx· cosr (3.8.35)
tPj (x,y, r) =-(-ly cosx· sin r· e-(-I)' Y. (3.8.36)
Using (3.8.35) and (3.8.36), equation (3.8.32) gives the dispersion relation
, k 2 + 5-1
(r = (3.8.37)
5+1
Note from (3.8.37) that-
(i) in the absence of the surface tension, the interface is stable or unstable
according as p~ S p~ ;
(ii) in the presence of the surface tension, the interface is stable or unstable
according as k'~k: where
g' (p~ - p;)

k'c = (3.8.38)
T'
(b) The Nonlinear Problem

When the interface is unstable according to the linear theory, the evolution of
the interface quickly gets out of the linear regime because the predicted growth
is exponential. It turns out that the nonlinear problem of instability of a
vertically accelerated horizontal two-dimensional interface between two
different fluids is one of singular perturbation type. We therefore use the method
of strained parameters to develop a uniformly valid solution for the above
problem for wave numbers near the linear cut-off value (Nayfeh, 1969 and
Shivamoggi, 1979).
A few remarks about the issue of transferring the boundary conditions to
known boundaries in a perturbation scheme are in order. Often a boundary
condition is imposed at a boundary, which is unknown without the solution and
varies slightly with the perturbation parameter e. A knowledge of the way in
which the solution varies in the vicinity of the basic configuration of the
boundary corresponding to e =0 is necessary in order to effect the transfer of a
boundary condition. If the solution is analytic in spatial coordinates, this transfer
is simply accomplished by an expansion in a Taylor's series about the values at
the basic configuration. If this is not possible, e will appear both implicitly and
explicitly in the perturbation expansion so that the resulting operations will not
be consistent.
Seek solutions to the boundary-value problem (3.8.27)-(3.8.29), (3.8.24) and
(3.8.25), of the form, for wavenumbers near the linear cut-off value kc '
-1: en t/J;n) (x,y, r)

~
cf>j (x,y, r;e) (3.8.39)

n=l
1: en 71n(x, r)
~
71 (x, r;e) = (3.8.40)

n=1
1: en-' O'n(k)
~
O'(k;e) = (3.8.41)
n=1
(3.8.42)
In (3.8.42), one could include an O(e) tenn on the right-hand side, but it turns
out to be zero anyway.
One obtains upon substitution of (3.8.39)-(3.8.42) into (3.8.27)-(3.8.29),
(3.8.24) and (3.8.25):
O(e): y§ 0 : t/J;~x +t/J;;y =0 (3.8.43)
y = 0 : 0', 71 lr + t/J;'J = 0 (3.8.44)
Y -0'
- A(') +71, -s (A('))
.0', 'l'lr
2 -
0'1 'l'2r +711 + kc 71 lxx -0 (3.8.45)
(3.8.46)
r=O: 71, =cosx, 71'r =0 (3.8.47)
o( e2).. y::>",: : 0 .. 'l'jxx

111(2) 111(2) - 0
+'I'jyy- (3.8.48)
Y -- 0 .. 0', 712r + 'l'jy

A(2) - III{I)
-71,x 'l'jx III{I)
- 'I'jyy 71, - 0' 2 71, t (3.8.49)
(3.8.50)
y~(-IYoo:t/J;~)~O (3.8.51)
r = 0 : 712 = 0, 712r = 0 (3.8.52)
0(e 3 ): y§ 0 : t/J;~)x +t/J;~~ = 0 (3.8.53)
_0. 1 III{I)
Y - . 0', 713r + 'l'jy - -'I'jyy 712 - 'l'jyy 71, - 2" 'l'jyyy 71, + 'l'jx 71 lx
111(3) - III{I) 1II(2) 2 1II(2)
+ t/J;~)y 71171,x + t/J;~) 71 2x - 0'2 71 2r - 0'3 71 lr (3.8.54)

(3.8.55)
(3.8.56)
1" = 0 : 113 = 0, 113< = o. (3.8.57)
etc.
One obtains for the O(e) problem, the linearized solution-
111(x, 1") =cosx· COS1" (3.8.58)
<t>Y) (x,y, 1") = -(-If cosx· sin 1"' e-(-IY y (3.8.59)

,
(Jj = k; s+1
+s-I
, (3.8.60)
the linear cut-off wave number kc being given by
(J~ =0 or k; + s - 1 =O. (3.8.38)

One may expect to construct uniformly valid solutions only for wavenumbers
larger than k c ' But it turns out, upon a consideration of the nonlinear problem in
the folIowing, that this is possible only for wavenumbers greater than kc by a
definite amount, namely, £2 K.
For wavenumbers near kc ' usmg (3.8.58) and (3.8.59), the system
(3.8.48)-(3.8.52) becomes
.::: 0 .. 'f'jxx
Y:::>
111(2) 111(2)
+ 'f'jyy -
-
0 (3.8.61)
Y =0 : <t>g) = (J2 cosx· sin 1" (3.8.62)
Y =0 : (I - s) 112 + k; 112xx =0 (3.8.63)
Y~ (-If c<:>: <t»~) ~ 0 (3.8.64)
1" = 0 : 112 = 0, 112 < = 0 (3.8.65)

from which
712 (x, r) = cosx ·(1- cosr) (3.8.66)
lPy)(x,y,r)=-(-ly a 2 cosx·sinr·e-(-IYY, (3.8.67)
where a 2is now non-zero, but is arbitrary in the 0(£2) problem.

Next, using (3.8.58), (3.8.59), (3.8.66), and (3.8.67), the system
(3.8.53)-(3.8.57) becomes
,.::: 0 'Yjxx+Yjyy-
. 11>(3) 11>(3) - 0 (3.8.68)
y::>
y=o: IPY) =- ~2 cos2x·sin2r-(a2 - (3 ) cosx·sinr (3.8.69)
y =0 : (1- s) 713 + k; 713xx =[-(I + s)a; + K - %(1- s)] cos x· cos r (3.8.70)
y~(-IYoo:IP;~~O (3.8.71)
r =0 : 713 =0, 713r =O. (3.8.72)
The removal of the secular terms in (3.8.70) requires

3
(I + s) a; - K + - (I - s) = 0
8
or
3
K--(I-s) ] 1/2
a =+ __8"--__ (3.8.73)
2 - I +s
[
so that one has
K < ~ (1- s) : instability

8
K =~ (1- s) : neutral stability

8
K > ~ (1- s) : stability

8
and corresponding to neutral stability, one has
e =k; +~8 (I-s) £2 +0(£3) (3.8.74)
which is graphically represented in Figure 3.9.

e
Linear cut - off
Nonlinear cut - off
Unstable Stable
o kc k
Figure 3.9. Linear- and nonlinear-cutoffs for Rayleigh-Taylor instability.
It is to be noted that the interfacial waves grow even at k =kc despite the
cutoff predicted by the linear theory. On the other hand, the onset of instability
even below the linear stability threshold when the disturbance has finite
amplitude implies that this instability is a subcritical instability.
3.9. Applications to Plasma Physics
3.9.1 Nonlinear Waves in an Electron-Plasma

Consider a one-dimensional wave motion in a warm isothermal electron
plasma, with the ions forming an immobile neutralizing background (Infeld and
Rowlands,1979; Shivamoggi 1988a). The equations governing such a wave
motion are:
(i) conservation of mass:
an a
- + - (nv)=O (3.9.1)
at ax
(ii) conservation of momentum:
av av I an aiP
-+v-=---+- (3.9.2)
at ax n ax ax
(iii) Gauss' law:
(3.9.3)
where n is the electron number density normalized by the mean number density
no, v is the electron fluid velocity non-dimensionalized by the electron thermal
speed Vr, =~Kr,,/me' ¢J is the electric potential normalized by T,,/e, the time t
is normalized by W;I and x by the Debye length A. D = Vr./w p ' T" is the
temperature of the electron plasma and K is the Boltzmann constant.
If we look for a stationary wave traveling in the x-direction, the various
physical quantities will depend on x and t only through the combination
~ = lex - Q)(. Then on putting
n=l+N (3.9.4)
we may derive from equations (3.9.1)-(3.9.3):
W2 e] (iN [ k
2 2
3w ](aN)
[ -(1-+-N~)3 --l+-N- -a~-2 + (1+N)2 - (1+N)4 -a~ +N=O. (3.9.5)
Let us now introduce a small parameter e« 1, which may characterize a

typical wave amplitude, and seek solutions to equation (3.9.5) of the form
N(~;e) - ; e"N" (~)),

_ (3.9.6)
w(k;e) = L enwn(k)
n=O
o (e) .(2 2) a2 N,
. Wo - k a~2 + N, -_ 0 (3.9.7)
2 2
O(e2).(w2-e) a N2 +N =(3W2-K2)[N a N, +(aN,)2]
• 0 a~2 2 0 'a~2 a~
(3.9.9)
etc.
We obtain from equation (3.9.7), the linear result:
N, = aQ cos~}.
w6 =1+e (3.9.10)
Using (3.9.10), the removal of secular tenns on the right-hand side of

equation (3.9.8) requires
w, == 0 (3.9.11)
and then the solution to equation (3.9.8) is given by

,
N2 = a; (3+2k 2
) cos2~. (3.9.12)
Using (3.9.10)-(3.9.12), the condition for the removal of secular tenns on the
right-hand side of the equation (3.9.9) gives
W2
_e (8k
- ~
2
+9) 2
aQ • (3.9.13)
24"1'I+e
Thus, the nonlinear dispersion relation is given by
w2 =(1 +e)+e 2 :~ (8e +9) a~ +O(el (3.9.14)
3.9.2 Rayleigh-Taylor Instability of a Plasma in a Magnetic Field

Consider wave motions at the surface of an incompressible, infinitely-
conducting plasma of infinite depth supported by a magnetic field H', (Figure
3.10).
If the linear problem can be used as a guide it (see Chandrasekhar, 1961 and
Shivamoggi, 1986) shows that -
(i) the magnetic field affects the development of Rayleigh-Taylor
instability only if the disturbances propagate along the magnetic field;
(ii) the plasma motion is irrotational.
Though the validity of these results for the nonlinear problem is not established,
we shall, in the following, for the sake of simplicity, consider only wave
motions parallel to the magnetic field and take the plasma motion to be
irrotational (Shivamoggi, 1982a).
3.9 Applications to Plasma Physics 8S
Vacuum
H'
\ • ~~x,t,£)
x "'=7 ~
Plasma
Figure 3.10. The plasma surface accelerated normally
and subjected to a perturbation.
The applied acceleration g' is directed away from the plasma. Initially, the
surface of the plasma is taken to be disturbed according to a simple sinusoidal
standing wave with an amplitude a' and a wavelength A'/2n.
Nondimensionalize the various physical quantities with respect to a reference
length A'/2n, a time (A'/2ng'( and a magnetic field H'. Here the primes
denote the dimensional quantities. Introduce the velocity potential tfr and the
magnetic potential 'II for the perturbations according to
v = -'\1tfr, h =-'\1'11. (3.9.15)
If Y = 1J (x, t; £), where £ = a' (2n/ A'), denotes the disturbed shape of the
plasma surface, one has the following boundary-value problem.
(i) Laplace equation for the velocity potential-
y < 1J : tfr;; + <Pyy = O. (3.9.16)
(ii) Laplace equation for the magnetic-field potential-
Y > 1J : '11:0 + 'II yy = 0 . (3.9.17)
(iii) Kinematic condition at the plasma surface-
Y =1J : 1J, -1J.r<P\ + <P =O. y (3.9.18)
(iv) Frozen-in field constraint at the plasma surface-
(3.9.19)
(v) Dynamic condition at the plasma surface-

(3.9.20)
(vi) Infinity conditions-

(3.9.21)
(3.9.22)
(vii) Initial condition-

/=0: 17=t: COSX, 171 =0. (3.9.23)
Here
2 _
k ---
H,2 (21r)2
p'g' ).,' ,
p' being the mass density of the plasma. Note that (3.9.19) expresses the fact
that the plasma surface remains a magnetic-field line even in a perturbed state. 2
Make a change of variable
r =a/ (3.9.24)

Y < 17 : ~xx + ~yy = 0 (3.9.25)
Y > 17 : lfI n + lfIyy =0 (3.9.26)
Y = 17 : a 17 r -17x~x + ~y = 0 (3.9.27)
-17x + 17 xlfl x -lfl y = 0 (3.9.28)
a~r -± (~; +~:)+ 2

17 = k (-lfIx +± +±
lfI; lfI: ) (3.9.29)
Y~-oo: ~~O (3.9.30)
y~oo:lfI~O (3.9.31)
r=O: 17=t:cosx, 17 r =0. (3.9.32)
For wavenumbers k near the linear cut-off value kc ' seek solutions to
(3.9.25}-(3.9.32) of the form (Shivamoggi, 1982a)
2 The variational basis of the governing equations (3.9.16) and (3.9.17) and the boundary
conditions (3.9.18)-(3.9.22) for this problem was given by Shivamoggi (1983).

-l entPn (x,y, r)
~
tP(x,y, r;e) (3.9.33)
'II(x,y,r;e)-l en'lin (x,y,r) (3.9.34)

n=1
(3.9.35)
n=1
(j(k;e) = l
~
en-I(jn(k) (3.9.36)
n=1
(3.9.37)
In (3.9.37) one could include an O(e) term on the right-hand side, but it turns
out to be zero anyway.
One obtains upon substitution of (3.9.33)-(3.9.37) into (3.9.25)-(3.9.32), and
equating coefficients of equal powers of e, a hierarchy of problems of various
orders in e:
O(e): y< 0: tPlxx +tPlyy = 0 (3.9.38)
y > 0 : 'IIlxx + 'IIlyy = 0 (3.9.39)
Y = 0 : (j1111t + tPly = 0 (3.9.40)
'Illy + 11 lx =0 (3.9.41)
(jltPl t + 111 + k; 'IIlx =0 (3.9.42)
y~-oo: tPl ~O (3.9.43)
y ~ +00 : 'III ~ 0 (3.9.44)
1'=0: 111 =cosx, 11lt =0. (3.9.45)

(3.9.46)
y > 0 : '112xx + '112yy = 0 (3.9.47)
Y =0 : (j1112t + tP2y =11lx tP,x - tPlyy111 - (j211 1t (3.9.48)
(3.9.49)
88 The Method of Strained Coordinates/Parameten
a llP2< +Th +k;, 'lf2. =21 (lPi.' +lPI~V, ) -a2lPI< -a,lP,ty711

+k~ [-'lfI.y711 +~ ('If~. +'lfn] (3.9.50)
y~-oo: lP2 ~O (3.9.51)
y~+oo: 'lf2 ~O (3.9.52)
r = 0 : 71 = 0,
2 71 2 <= o. (3.9.53)
o( £3) :y < 0 : lP3xx + lP3yy = 0 (3.9.54)
y> 0 : 'lf3xx + 'lf3yy =0 (3.9.55)

1 ,
y = 0 : a l713< + lP3y = -lPlyy712 - lP2yy711 - 2 lP,yyy71i + lP2.71I.
+lPl.712x +lP,xy71 1711. - a 2 71 2 <-a371 1< (3.9.56)
(3.9.57)
a llP3< +713 +k; 'lf3. = -a2(lP2< +lPlty71I)-a3lPl< -a l (lP2ty711 +lPlty712
+~ lP,ryy71; )+(lPI.lP2x +lP,.lP"y711 +lPl ylP2y

, [ 1,
+lP,ylP,yy71 1) + k; -'lfl xy712 - 'If 2x.l,711 - 2 'If"yy71i + 'lflx'If2.
+'lflx'lf"y711 + 'If Iy 'If2y + 'If, y'lf,yy71I] - J( 'lfl x (3.9.58)
y ~ -00 : lP3 ~ 0 (3.9.59)
y~+oo: 'lf3 ~O (3.9.60)
r = 0 : 713 = 0, 713< =0 (3.9.61)
etc.
For k'" kc where a 1 = 0, the system (3.9.38)-(3.9.45) becomes
y < 0 : lPlxx + lPlyy = 0 (3.9.62)
y> 0 : 'lflxx + 'lfl yy = 0 (3.9.63)
y =0: lPly =0 (3.9.64)
(3.9.65)
(3.9.66)
y~-oo: tPl ~O (3.9.67)
y ~ +00 : lfIl ~ 0 (3.9.68)
r=O: 171 =COSX, 17lr =0, (3.9.69)
from which one obtains

171 =COSX· cosr (3.9.70)
lfIl = -e- Y sinx· cosr (3.9.71)
(3.9.72)
(3.9.73)
One may expect to construct unifonnly valid solutions only for wavenumbers
larger than kc '
Using (3.9.70)-(3.9.73), the system (3.9.46)-(3.9.53) becomes

y < 0 : tP2xx =0
+ tP2yy (3.9.74)
y> 0: lfI2xx + lfI2yy =0 (3.9.75)
y = 0 : tP2Y =(j2 cosx· sin r (3.9.76)
lfI2y + 172> = ±
sin2x (I +cos2r) (3.9.77)
172 +lfI2x =-"41 cos2x·(I+cos2r) (3.9.78)
y~ -00: tP2 ~ 0 (3.9.79)

y~+OO:lfI2~0 (3.9.80)
(3.9.81)
from which one obtains

1
172 = - - cos2x (I + cos2r) (3.9.82)
4
lfI2 == 0 (3.9.83)
tP2 = +(j2 eY cosx· sin r (3.9.84)
where (j 2is nonzero but is arbitrary in the 0 (£2) problem.
Next, using (3.9.70)-(3.9.73) and (3.9.82)-(3.9.84), the system (3.9.54)-
(3.9.61) becomes
y < 0 : ~3xx + ~3YY =0 (3.9.85)
y > 0: Vl3xx + Vl3yy = 0 (3.9.86)
y = 0 : ~3Y = <1'3 COSX· sin r - <1'2 cos2xsin2r (3.9.87)
Vl3y + 713x = -23.smx· cos·, X· cos 3

r+
_.!.. sin x . COS'Z" cos2x· (1 + cos2r) +

4
+.!.. cosx· coS'Z" sin 2x· (1+ cos 2r) (3.9.88)

2
713 + Vl3x = ( -<1'; + I( - ~) cos x . cos r + Higher Harntonics (3.9.89)
y~ -00: ~3 ~O (3.9.90)
y~+oo: Vl3 ~O (3.9.91)
r =0 : 713 =0, 713r =O. (3.9.92)
The removal of the secular ternts in (3.9.89) requires

2 3
<1'2 -1(+-=0
8
or
(3.9.93)
so that one has
I( < -3:.msta b'I'

t tty
8
I( =~ : neutral stability
8
I( > -3 : stab'l'
t tty
8
and corresponding to neutral stability, one has
e =1+'83 e 2
+0(e
3
) (3.9.94)
which is graphically represented in Figure 3.11.

3.10 Limitations of the Method of Strained Parameters 91
Linear
cut -off
Unstable Stable
k
Figure 3.11. Linear and nonlinear cut-offs for
Rayleigh-Taylor instability of a plasma.
It thus appears that the waves on the plasma surface grow even at k =kc = I,
despite the cut-off predicted by the linear theory.
In the present case, we have a fluid, which is subject to a magnetic field, and
there is no surface tension. The opposite case wherein the fluid is subject to a
surface tension and there is no magnetic field was considered in section 3.8.2,
where we found for the cut-off wavenumber the same expression as (3.9.94)!
This seems to indicate that the magnetic field and the surface tension have
dynamically equivalent effects on the present problem even in the nonlinear
case.
3.10. Limitations of the Method of Strained

Parameters
Generally, the method of strained parameters succeeds when the singularity

predicted by the regular perturbative solution actually exists, though at a slightly
different location (Van Dyke, 1975). However, this method, which is effective
in determining periodic solutions, is incapable of determining transient
responses of dissipative systems because any expansion procedure that is based

on approximating e-et by a finite number of terms in a series expansion is
doomed to fail for large t!
Example 7: Consider the initial-value problem

2
d 2u +u=_£du ]
dt dt
(3.10.1)
du = 0 .
t= O : u= I,
dt
Let us construct a solution to the initial-value problem (3.10.1) of the form
u(t;e)-uo (t)+£u,(t)+0(£2). (3.10.2)
Then, we obtain from (3.10.1), to various orders in £:
0(1): d'·~o + u = 0 ]
dr 0
(3.10.3)
t =0 : o =1,
U duo =0
dt
2
O(e): d u, +u = _ duo }
dt 2 1 dt
(3.10.4)
t =0 : u, =0, du, =0
dt
etc.
Solving (3.10.3) and (3.10.4), we obtain
Uo = cost (3.10.5)
u, =.!. (-t cos t + sin t) (3.10.6)

2
etc.
u - cost + e.!. (-t cost + sint) + O( £2). (3.10.7)

2
In order to remove the secular term in (3.10.7), we use Pritulo's method and
renormalize according to
(3.1 0.8)
3.11 Exercises 93
so that (3.10.7) becomes

. 1 1.
u - cos s - e f. sm s - - e s cos s + - e SID s + .... (3.10.9)
I 2 2
In order to remove the secular terms in (3.10.7), one may elect to choose
1
f.. =-- scots (3.10.10)
2
which is, however, unacceptable because f.. becomes unbounded for
s = n,2n,3n, .... Thus, the method of strained parameters cannot handle
dissipative systems.
3.11. Exercises
1. The equation of motion for a pendulum rotating about its vertical axis with
angular velocity r is
?+(~I _r
2
d 2
coso) sine =O.
dr
Use the method of strained parameters to calculate the amplitude-dependent
frequency.
2. A point mass moves freely along a parabola x
2
=2pz (p> 0) rotating about
its axis with angular velocity 0.>. The equation of motion of the mass is
(1+;:)X+~22 +(;-o.>2)X=0.
Solve it using the method of strained parameters.
3. For the Mathieu equation
ii+(o+ecos2t) u=O,
use Whittaker's method to determine the solutions in the neighborhood of the
transition curve given by
4. Solve using the eigenfunction expansion method

5. Obtain the two-tenn straightforward expansion of the solution of (Nayfeh,

1973)
(x+ty) dy +y=O,
dx
y(l) = 1.
Use Lighthill's method to remove the singularity at x =O.
6. Obtain the two-tenn regular perturbation expansion of the solution of
(x+ty) dy +2y=2,
dx
y(l) = 2.
Use the method of strained parameters to obtain the unifonnly valid one-tenn
expansion
where,
x=s-e(l+ 3~2 )+o(e 2),

C=I+~e+O(e2).
3
Show that it does not have the problem that the regular perturbation
expansion has, as x=>O.
7. Solve using the method of strained parameters

(l+eu) Ux +u y =0,
t = 0 : U = e t/> (x ).
8. Solve using the method of strained parameters
t =0: U = j(x)+ g(x), ut =g'(x)- j'(x).

9. Consider the weakly nonlinear wave equation
3.12 Appendix 1 95
For e =0, this equation has the solution
U o =aosin(O+l/'o), O=kx-~I+e t
for arbitrary constants ao'l/'o and k. Construct a periodic solution, for
0< e« I, of the form
u(x,t;e) - a o sin(e + l/'o)+ e u, (e) + ...
where
e=kx-~I+k" (O(e)t,
(0 (e) =1+ e(Ol + e"(O" + "',
3a"
(0\ = 8 (I +k"
0 ')' etc.,
3
a
u 1 =- 3; sin3(e+l/'0), etc.
3.12. Appendix 1
Fredholm's Alternative Theorem

Let L be a bounded linear operator, and suppose we wish to solve the
equation -
Lf =g, X E [a, b]. (AU)
We want to know if there is a solution, and if so, how many solutions are
possible.
Theorem (Uniqueness). The solution of equation (AI. I) (if it exists) is unique if
and only if the only solution of L fo = 0 is fo == O.
Proof: Suppose L fo =0 for some fo :;It O. If L f =g, then J; =f + a fo also

solves L J; = g for any a so that the solution is not unique. Conversely, if
solutions of L f =g are not unique, then there are solutions J; and h with
f =J; - h :;It 0 satisfying L J; - L h =o.
Theorem (Existence): Equation (AI. I) has a solution if and only if
f g(x) v(x) dx = °
b
(g, v) = (Al.2)
a
for every v (x) satisfying L' v =0, where L' is the adjoint ofL.
Proof: Suppose RL denotes the range of Land N L denotes the null space of L
while Rt denotes the orthogonal complement of R L • The proof is done if we
show that Rt =N L,. Suppose vERt. This implies that
(v,Lf)=o, VfEH,
or
(L'v,f) =0,
from which
L·v=O.
Therefore,
Conversely, suppose v E Nt. Then,
(L'v,h)=O, VhEH
or
(v, Lh)=O,
so
vERt·
Example At: Consider
A=G ~} (A 1.3)
N(A) is spanned by the vector-
(AlA)
3.13 Appendix 2 97
so that solutions of Ax =b, if they exist, are not unique. Since N A" is spanned
by the vector -
(Al.5)
solutions of Ax =b exist only if b is of the fOlm
(A 1.6)
This is no surprise since the second row of A is three times the first row.
3.13. Appendix 2
Floquet Theory
Consider non-autonomous linear differential equations with periodic
coefficients, i.e.,
x =a(t)x, a(t + T) =a (t),} ,

(A2.l)
x(O) = X o
where x E~.
Though a(t) is periodic, x(t) may not be periodic.
We have, from equation (A2.l),

fa(sldS
x(t)=xoe o =qJ(t) x o' (A2.2)
where
fa(S)dS
qJ(t):e o (A2.3)
is a fundamental solution of the problem.
Theorem: Let p be a Floquet multiplier for the system -
x'=a(t)x, a(t+T)=a(t), ":it (A2.l)

and let J1 be the corresponding Floquet exponent so that p =e/l T. Then, there
exists a solution x(t) such that
x (t + T) = P x (t), 'ift . (A2.4)
Further,
x(t) =elll v(t), 'ift (A2.5)
where v(t + T) =v(t).
Proof: Noting that
Io a(s) ds = Ia(s) ds + Ia(s) ds

~T T ~T
0 T
I I
T t+T
= a (s) ds + a (s - T) ds
o T
I I
T t
= a(s) ds + a(s) ds,

° °
we have
1+1"
Jo(s)ds
cp(t+T)=e °
fa(S)dS Ja(S)dS
=e O
• eO (A2.6)
= cp(T) cp(t).
Thus,
x(t + T) = px(t) (A2.7)
where p is the Floquet multiplier
p =cp(T). (A2.8)
If P =e/l T, J1 being the Floquet exponent, we have
1 2kni 1 IT 2kni
J1 = - log cp( T) + - =- a( s) ds +-- (A2.9)
T T ToT
so that J1 is determined only up to multiples of 2ni/T.
Further, if we put
3.13 Appendix 2 99
v(t) = x(t) e- 1l1 (A2.5)
we have, on using (A2.7) and (A2.8),

v(t + T) = x(t + T) e- 1l1 • e- IlT
= px(t) e- 1l1 'e- IlT
= x(t) e- 1l1 = v(t) (A2.10)
so that v(t) is periodic and bounded.

The Floquet exponents enable one to determine the stability of the origin
(which is a stationary point). If Re (.u) < 0, solutions tend to zero, while if
Re (.u) > 0, solutions become unbounded as t=}oo.
Example A2: Consider

x=(o+cost) x.
Then, we have
a(t)=o+cost, a(t+2n)=a(t).
Noting
Ja(s) ds
2/r
= 2nO,
o
we have that the Hoquet exponent is
Ja(s) ds = O.
1 2/r
.u = -
2n 0
Therefore, the origin is stable if 0 < 0 and unstable if 0 > O.
Let us now generalize the above discussion to consider the case x E ~" ,
x=A(t)· x, A(t+T)=A(t). (A2.1l)
Definition: Let xl(t),···,xn(t) be n linearly independent solutions of

x= A(T) x (A2.11)
on t e 9' , and put

(t) =[x'(t), ... ,x" (t)] (A2.12)
where (t) is an n x n matrix solution of
<1>= A <1>. (A2.13)
Then, is called a fundamental matrix, and if (to) =I, I being the unit
matrix, then (t) is the principal fundamental matrix. We have
x (t) = (t) . X o. (A2.14)
Further, if we introduce
W(t) =det<l>(t), (A2.15)
W(t) is called the Wronskian.
Lemma: The Wronskian W(t) is given by
ftrA(S) ds
W(t) = W(t o ) e" (A2.16)
Proof: Let us write
or
(A2.17)
Using (A2.15), we then have

W(t)= W(t o) det[I +(t-to ) A(to)+o(t-to)]
or
(A2.18)a
Writing
(A2.18)b
we have, from (A2.18)a,

(A2.19)
3.13 Appendix 2 101
So
W=WtrA, (A2.20)
from which
ftrA(sldS
W(t) = W(t o) e" (A2.16)
Theorem: Let (t) be afundamental matrix for

x=A(t)x, A(t+T)=A(t), '<;ft. (A2.]])
Then, (t + T) is also a fundamental matrix, and there exists a non-singular

constant matrix B such that
(t+T)=(t) B, '<;ft. (A2.2])
Also,
T
ftrA(s)ds
detB = eO (A2.22)
Proof: Since (t) is a fundamental matrix, we have

ei> = A(t) <1>. (A2.13)
Let
'I'(t);: (t + T). (A2.23)
Then
'P(t) =ei>(t + T) (A2.24)
and using equation (A2.13), we have
'P(t) = A(t+ T) (t + T) = A(t)'I'(t). (A2.25)
Equation (A2.25) implies that 'I'(t) is also a fundamental matrix, and hence,
'I' (t) ;: (t + T) = (t) B (A2.26)
for some constant non-singular matrix B.

We have, from (A2.l5),
ftrA(sldS I+T ]
W(t+T)= W(O) eO ~trA(s) ds . (A2.27)
[
Ifwe write
W (t + T) = W (t) det B, (A2.28)
we then have from (A2.27),

,..,T T
ftrA(s)ds ftrA(slds
detB=e' =e' (A2.22)
Equation (A2.2l) implies that
B =<II-'(O)<II(T). (A2.29)
If <II(t) is the principal fundamental matrix, so <11(0) =I, then we have
B =<II(T). (A2.30)
Definition: Let the eigenvalues of B be p,,"',Pn' which are the Floquet

multipliers. If Pi =ell, T; i =1"", n, then f.ii are the Floquet exponents.
Note that
1 2kni
II. =-Iogp +--' k=12 .... (A2.3I)
,.." T ' T' , ,
So, Pi are only determined up to multiples of 2nijT.

Let e), .. ,en be the eigenvectors ofB. Writing
(A2.32)
we have
x(t+T)=<II(t+T)· X o
= <II(t)· <II(T)· (~aiei) (A2.33)
= <II(t) . ( ~ ellJ aiei ).

Thus, if
(A2.34)
3.13 Appendix 2 103
we obtain
(A2.35)
If we introduce
(A2.36)
we have from (A2.35),

vk(t + T) =e-IJ.(t+T) x k(t + T)
=e-lJ,(t+T) $(t) . ak elJ•T ek (A2.37)
= e- IJ , I x (t) = v (t)
k k
so that vk (t), 1 $ k $ n, is periodic and bounded.
Furthennore, noting that

x(t)= L xk(t)=L elJ.1vk(t), (A2.38)
k k
we have that, if Re (,uk) $ 0, V k, the origin is asymptotically stable, while if

Re (,uk ) ~ 0 for any k, 1 $ k $ n, the origin is unstable.
Example A3: Consider Hill's equation

u+a(t)u=O,
where
a(t+T)=a(t), Vt.
Putting
we have
which may be rewritten as,

1
x=A(t)x, A(t)=[ 0 0].
-aCt)
The fundamental matrix $(t) is given by
where u 1 and u 2 are linearly independent solutions such that

2
u ' (O)=I,(0)=0}.U
ul(O)=O, u (0)=1 2
So,
<1>(0) =I.
Note,
(t + T) = (t)B
where, from (A2.22) and (A2.29),
U1(T)
B =<1>-1(0) (T) =
[ u'(T)
and
T
JtrA(sjds
detB=eo =1.
The Floquet multipliers P, which are the eigenvalues of B, are then given by
p 2 _ 2 t/J P + 1 = 0,
where
Thus,
PI.2 =t/J ±~t/J2 -1
with
PIP2 =1, PI + P2 =2t/J.
T
The Floquet exponents J.l1.2' where PI.2 = el'I.2 , are therefore given by
There are now several cases to consider.

(i) t/J> 1: Here, PI.2 are both real and positive, and PI > 1 > P2 > O. Thus, J.l1
is real and positive while J.l2 is real and negative. The general solution is
3.13 Appendix 2 lOS
where
There are no periodic solutions, and in general, lui ~ 00, as t ~ 00.
(ii) tfr < -1: Here, PI.2 are both real and negative, and P2 < -1 < P, < O.
Thus,
J.l, =y:-Y,
in J.12 =-y:+Y
in . 1
and cosh rr = -tfr
The general solution is
where
iffl
±-
K,.2(t)=e T vl.2(t)
and
Kdt + 2T) =KI.2(t), "1 t
Again, there are no periodic solutions, and in general, lui ~ as t ~

00, 00 .
(iii) -1 < tfr < 1: Here, PI.2 are both complex, with Ip,.21 =1. We have
PIO =e±i<JT, 0 < e; T < n
and
J.lI.2 = ± ie;, cos aT = tfr·

u = c, Re[e i<Jlv(t)]+c 21m [ei<Jlv(t)]
where
v(t + T) = v(t), "1 t.
The solutions are bounded and oscillatory.
(iv) tfr = 1: This case is the boundary between cases (i) and (iii). Here, we
have
and
PI = P2 = O.
where
The solutions are purely oscillatory.

(v) l/J =-1: This case is the boundary between cases (ii) and (iii). Here, we
have
PI =pz =-1,
and
where
The solutions are again purely oscilIatory.

Thus, the boundary between bounded and unbounded solutions is
characterized by periodic solutions with period Tor 2T.
As a special case of Hill's equation, consider a parametric resonance system
ii + {co 2 +cb(t)} u=O
where
b (t + T) =b (t).
When c = 0, this equation describes a simple harmonic oscillation of period
1;, =. 2,,/ co.
We have,
3.14 Appendix 3 107
so, the stability boundaries are now curves in the OJ 2 ,e-plane.
Further, we have, for the case e = 0,
u' = cosOJt, u2 =~ sinOJt.

OJ
So,
t/J = cosOJT,
and hence, -1::;; t/J ::;; 1, for e =o. The stability boundaries are given by
t/J=I~OJT=2k1r or T=k1;"
t/J=-I~OJT=(2k+I)1r or 2T=(2k+l)To,
where k =0,1,2, .... The analyticity of the function t/J(OJ2, e) then implies that
*
the above stability boundaries can be extended into the e 0, but small e-
regime by expanding the solutions as a power series in e. The unbounded
behavior of the solutions, when T or 2T is an integer multiple of 1;" is called
parametric resonance.
3.14. Appendix 3
Bifurcation Theory
Bifurcation or branching occurs in a system when the state of the system
depends on some parameter f.l, and as that parameter varies the state branches to
another state at some critical value f.l c of the parameter f.l with usually a
concomitant change of stability. Stability considerations determine the solution
sought out by the system when there is a multiplicity of solutions. The
bifurcation theory is concerned with how the multiplicity of solutions varies
with the parameter and the stability properties of the bifurcating solutions.
Bifurcations are classified according to how the stability of an equilibrium
solution changes. There are two ways in which this can occur. An eigenvalue of
the system linearized about this solution can pass through zero, or a pair of non-
zero eigenvalues may cross the imaginary axis. The first case corresponds to a
saddle-node or tangent bifurcation and describes the birth or collapse of two
equilibria (like a stable node coalescing with a saddle and annihilating it). This
corresponds to a manifold associated with a given fixed point intersecting itself.

On the other hand, when manifolds associated with different fixed points
intersect, an exchange of stability occurs - this corresponds to a transcritical
bifurcation or a pitchfork bifurcation. The second case corresponds to the so-
called Hopf bifurcation and describes the birth of a family of periodic orbits
(limit cycle) following the change in stability of a focus.
Consider the first-order autonomous differential equation describing flow in a
one-dimensional phase space R,
du
-=f(u;J.l), 1>0 (A3.l)
dl
where J.l is a real parameter, and f is a given analytic function of u and J.l with
continuous partial derivatives of all order with respect to u and J.l.
The equilibrium solution u =U o of equation (A3.I) is found from
f(u;J.l)=O. (A3.2)
If ~ = 0 in an open neighborhood of J.l = J.l c' then corresponding to one value

of J.l, several equilibrium solutions
o may exist. This is guaranteed by the
U
3
Implicit Function Theorem which allows the solutions of equation (A3.2) to
become non-unique whenever C!f/ = 0 . au
The graph of equation (A3.2) is called the branching or bifurcation diagram.
The intersecting branches are the bifurcating solutions and the points of
intersection, which correspond to change of stability, are called bifurcation
points.
The stability of equilibrium solutions changes as J.l varies. Often an
equilibrium solution uo(J.l) will be stable for J.l < J.lc and unstable for J.l ~ J.l c'
Thus, as J.l is increased slowly, the equilibrium solution uo(J.l) becomes
unstable at J.lc' the system may therefore branch to another stable solution. The
bifurcation theory seeks to explore how the stability of various equilibria
3 THEOREM: Suppose that f(u;/l) : R x R ~ R is a C l function satisfying
f(uo;/l,)=O and ~ (uo;/l,);tO.

Then, there exists a unique solution of the implicit equation f(u;/l) = 0 given by
U =g(/l) in some open subset Wof /lc'
3.14 Appendix 3 109
changes as J.l is varied near J.lc' This issue, of course, depends in an essential
way on the nonlinear nature of the problem.
In order to detennine stability of the equilibrium solution Uo(J.l), consider a
small perturbation u(t) about uo(J.l) , so that
u(t) = uo(J.l) + u(t). (A3.3)
Equation (A3.l) then becomes

du= f( Uo +u;J.l
- '). (A3.4)
dt
Linearizingfin U, equation (A3.4) becomes
du r ( ) , (A3.5)
dt '" J u uo;J.l u,
from which we have
u(t) =cef.(uo:I')'. (A3.6)
Thus, if /'. (u o;J.l) < 0,
o is stable and vice versa. The bifurcation point J.lc is
U
here defined by fu(uo;J.l)=O along with f(uo;J.l)=O. Now, by the Implicit

Function Theorem, f(uo;J.l) =0 implies J.l =J.l(uo) whenever fl' (uo;J.l) ~ O.
We have, on differentiating f(uo;J.l) =0 with respect to uo'
dJ.l
fu +/p - = O. (A3.7)
o duo
Equation (A3.7) shows that dJ.l
duo
=0 at a bifurcation point (where fu
0
=0), if
fl' ~ 0 there.
In higher dimensions, one has, in place of equation (A3.1),
du
-=f(u;J.l). (A3.8)
dt
Let uo(J.l) be an equilibrium point of equation (A3.8), so that
f(uo(J.l);J.l) =O. (A3.9)
The stability of uo(J.l) is then determined by the eigenvalues
A) (J.l), ~ (J.l), "', An (J.l) of the Jacobian matrix
A(J.l) = df(uo(J.l);J.l)
au . (A3.l0)
If all the eigenvalues have a negative real part, then "0 (J1) is stable. On the other
hand, if one or more eigenvalues have a positive real part, then "0(J1) is
unstable. Further, if the eigenvalues depend on the parameterJ1, this stability
may change as the parameter J1 varies. In fact, the value of J1 =J1 c' say, for
which
ReAo;{J1c) =0 for some i }
ReAAJ1c) < 0 for all j ::F- i
define the bifurcation points. It is apparent that a bifurcation point can arise in
two ways:
(i) AI (J1) is real-valued, AI (J1c) = 0, and Re A;(J1J < 0 for i = 2, ... , n;
(ii) AI (J1) and A2 (J1) form a complex conjugate pair, so that

A) (J1) =A2 (J1) =a (J1) + i f3 (J1), a (J1c) =0, f3 (J1c ) ::F- 0, and Re Ai (J1c ) < 0
for i =3, ... ,n.
Note that for a one-dimensional system, only Case (i) can occur. Case (i) is
called the saddle-node bifurcation. Case (ii) is called the Hopf bifurcation.
Figure 3.12 shows the schematic diagram of the variation of the eigenvalues as
J1 varies through J1 c in cases (i) and (ii) for a two-dimensional system.
Case (i) corresponds to the transition of the critical point "0(J1) from a stable
node into a saddle point. Case (ii) corresponds to the transition of the critical
point "0
(J1) from a stable focus into an unstable focus and appearance of a
periodic solution (see Shivamoggi, 1997, for futher details).
3.14 Appendix 3 111
ImA. ImA.
A.\ (/l)
ReA. ReA.
A. 2 (/l) A.1(/l)
A. 2 (/l)
(a) (b)
Figure 3.12. Schematic diagrams for (a) saddle-node bifurcation
and (b) Hopf bifurcation.
Chapter 4
Method of Averaging
4.1. Introduction
Consider a weakly nonlinear oscillation problem

x+ e h (x, x) + x =0, e« 1. (4.1.1)
The effect of the perturbation is to cause the parameters of the unperturbed

system to vary with time. If the perturbation is small, the variation of the
parameters within one period of the unperturbed motion will also be small.
Thus, the perturbation theory is based on the premise that during small intervals
of time the perturbed system moves along a route, which has the same functional
form as the unperturbed system but with parameters changing with time. The
latter variation may be periodic or secular. In the latter case, the change in a
parameter is cumulative so that no matter how small this change may be over a
small interval of time, eventually the perturbed route will differ greatly from the
unperturbed route. The periodic variations in the parameters, on the other hand,
do not materialize over long intervals of time because they can be eliminated by
averaging the perturbation effects over the unperturbed period!
Assume that this system has at least one periodic solution, so that its phase
diagram consists of either a limit cycle or a center. Such an equation is in a
sense "close" to the linear equation
x+x=O (4.1.2)
and the paths will be close to being circular. One may use this fact in
constructing approximate solutions. In order to determine periodic solutions, in
particular, we assume the existence of a nearly circular closed path in the phase
diagram. This method calculates the average value of the amplitude of a slowly-
varying oscillation over a cycle of the path. It can calculate approximate periods
114 Method of Averaging
of limit cycles and the shape of the spiral paths around limit cycles, or it can
calculate amplitude-frequency relations in the case of a center.
4.2. Krylov-Bogoliubov Method of Averaging
Introduce the phase-plane coordinates (x,y =x), then equation (4.1.1) can be
written as the following system of first-order differential equations:
x=y }. (4.2.1)
y= -eh-x
Let (a(t), O(t)) be the polar coordinates of a representative point on a phase
curve of the system. Thus,
x = aCOSO}
(4.2.2)
y = asinO '
from which
(4.2.3)
and
0= tan-I (y/x). (4.2.4)
We then have from (4.2.3) the following result-

ao = xX +.y.y = -eyh(x,y)
or
0= -eh(acosO, asinO) sinO. (4.2.5)
Further, we have from (4.2.4) the following result-
. e
0= -1- - h(acosO, a sin 0) cosO. (4.2.6)
a
Putting
f3(t):O(t)+t, (4.2.7)
we obtain from (4.2.6),
. e
13=-- h(acosO, asinO) cosO. (4.2.8)
a
Thus, both f3(t) and a(t) are slowly-varying with t. Figure 4.1
schematically shows these variables for the Duffing equation, for which
h(x,x) =x 3 •
~~(t)
n 1\ a(t
I u(t)
u V u V V U U V V V V
Figure 4.1. The variation of a, p, and u with t for a(O) = 1.5, P(O) = 0.0,
and E = 0.05, for the Duff'mg equation with h (x, x) = x 3
(From Nayfeh, 1981).
The perturbed period is given by (Krylov and Bogoliubov, 1947)

21< dO 21<
r=-f-· "'2n-~ fh(acosO, asinO) coso dO. (4.2.9)
o 0 a 0
The perturbed frequency is then given by

'I<
CO = 2n '" 1 +_t:_ "fh(acosO, asinO) cosO dO. (4.2.10)

r 2na 0
We have on the other hand, using (4.2.5) and (4.2.6),
da = ~ '" t:h(acosO, asinO) sinO. (4.2.11)

dO 0
If the solution is periodic with period nearly equal to 2n, we obtain
21<
f da dO =0
o dO
or
Jh(acosO, asin 0) sinO dO =°

2H
(4.2.12)
o
which gives a = a, say.
Example 1: Consider the nonlinear oscillator described by the Duffing equation

x+x+£x 3 "'0. (4.2.13)
Equation (4.2.10) now gives for the perturbed frequency
J(a cos 0) cosO dO

2H
OJ '" 1 + _£- 3 3
(4.2.14)
21ra 0
from which
3£ ,
OJ '" 1+- a- (4.2.15)
8 '
in agreement with the result in Example 2, Chapter 3.
Let us expand the nonlinear term in a Fourier series as follows-
h [a(O) cosO, a(O) sinO] sinO
=Po [a(0)] +
2
i
n=\
{PH [a(O)] cos nO + qn [a (0)] sin nO}, 0 < 0 < 21r (4.2.16)
where
Po [a (0)] =-
1
1r
Jh[a(O) cosu, a(O) sinu] sinu du
2H
f h[a(O)
>H
qn [a (0)] =.!- cosu, a(O) sinu] sinu sin (nu) du . (4.2.17)

1r 0
Pn [a (0)] =-
1
1r 0
Jh[a(O) cosu, a(O) sinu] sinu cos(nu) du
2H
Using (4.2.16) in (4.2.5) and integrating over a cycle, we obtain
(a) =_.!. £ Po (a). (4.2.18)

2
Similarly, we obtain
21<
(0)=_1 __£_ fh(acosu, asinu) cosu duo (4.2.19)
2na 0
Example 2: Consider the van der Pol oscillator described by

x+ £ (x 2
-I) x+ x = 0, £ > O. (4.2.20)
This system is unstable for small-amplitude perturbations, with a linear

growth rate £/2. The nonlinear dissipative term £ x 2 X eventually limits this
growth, and the amplitude of the oscillation saturates (as shown by the
numerical solution sketched in Figure 4.2).
Figure 4.2. Exact solution of equation (4.22) for x(O) = 0.5, x(O) =0 and
E =0.1 (From Nayfeh, 1981).
We have from (4.2.17) and (4.2.18), for this case,
0== (0) = _.!.- a

2n
f( a
0
2
cos 2 u -I) sin 2 u du =-~ ea(a
2
2
4
-I). (4.2.21)
Separating the variables and integrating, we obtain from equation (4.2.21)
f a (~a
a- -4
) =_.!. £(t+c)
8
where c is a constant of integration. This leads to
-± loga+i log la
2
-41=-i £(t+c),
from which
a(t) =[
1-
( 2)
4
1- - e
-£/
]1/2 ,
(4.2.22)
2
ao
=
where ao a(O).
Equation (4.2.22) indicates trajectories spiraling inward or outward
(according to ao > 2 or ao < 2) into a limit cycle, which is a circle given by
a = 2, in the phase plane (see Figure 4.3). This result shows the relaxation
nature of the oscillations of the van der Pol oscillator.
Figure 4.3. Orbits approaching the limit cycle for the van der Pol equation.
4.3. Krylov-Bogoliubov-Mitropolski Generalized

Method of Averaging
In this method, one assumes an asymptotic expansion of the form
2. en un (a, 'II) + o( e
N
u = a cos'll + N 1
+ ) (4.3.1)
n=l
where each un is a periodic function of VI with a period 21r, and a and VI are
taken to vary with time according to (Bogoliubov and Mitropolski, 1961)
da =fen An (a)+ O(e N+I ) )

dt n=1
(4.3.2)
d N
:: =roo + ~en Vln(a)+O(e N 1
+ )
while the derivatives are transformed according to
(4.3.3)
Example 3: Consider the Duffing oscillator described by

.• 2 3
U+ roo U= -e U . (4.3.4)
Using (4.25)-(4.27), we obtain from equation (4.3.4),
[( eA I +e2 A2 + ... )2 ~+(eA

aa2 I
+e 2A2 + ...)(eA'+e
I
2A' +...)
2
i-
aa
+2(eA I +e 2A2 + ... )(ro0 +ell/ 1/ + ...) ~
+e 21't'2
't'I dVl aa
+(roo +e lff
'I' 1
+e 2Iff
'I' 2
+ ... )2 ~+(eA
dVl2 I
+e 2A2 + ...)(elff'
'I' I
+e 2IIr'
'I' 2
+ ... )~]
dVl
2
x (acosVl + eUI + e U2 + ...) + ro~ (a cos VI + eUI + e2U2 + .. -)
= -e (a cos VI + e U I + e U2 + ..
2 f
(4.3.5)
from which, we obtain to various orders in e,
o (e) : roo2 (ddVl 2 + U\ =2roo VII a cos VI + 2roo AI SID. VI -

2
UI (4.3.6)
)
a3 cos 3 VI
etc.
2
3a
AI =0, V'I = - . (4.3.8)
8(i)o

3
a
ul = - - , cos3V'. (4.3.9)
32 (i)o
(dU2
(i)o2 :IT
2
+ U2 ) _
-
(
2(i)o V'2+ -15a
4
- 2 ) a cos V' + 2(i)o A 2 sm
.
V'
vV' 128 (i)o
(4.3.10)
as
+ - - 2 (21cos3V'-3cos 5V').
128 (i)o

4
15a
A2 = 0, V'2 = - 256(i)~ . (4.3.11)
Equation (4.3.10) then yields,

as
u2 = 4 (21 cos 3V' - cos 5V') . (4.3.12)
1024 (i)o
Using (4.3.8), (4.3.9), (4.3.11), and (4.3.12), we have from (4.3.1) and (4.3.2)
ea 3 e 2 as 3
u-acosV'+--, cos3V'- 4 (21cos3V'-cos5V')+O(e )
32 (i)o 1024 (i)o
da
- =
dt
° or a = a o = constant . (4.3.13)
2 2 4
dV'
-=(i) +3ea
- - - 15e a + O( e3)
dt 0 8 (i)o 256 (i)~
Example 4: Consider the van der Pol oscillator described by

u+u=e(l-u 2 ) u. (4.3.14)
Using (4.3.1)-(4.3.3), we obtain from equation (4.3.14),
[( lOA ) +102 A2 + ... )2 ~+(eA , +10 A2 + ...)(eA)' +e A'2 + ... )!...

2 2
da2 da
X a 1+10"') +10 "'2 + ...

2
+2 ( lOA) +10 2 A2 + ... 1+10"'1 +10 2 "'2 + ... --:1+
)
aau",
( 2 )2
a 2 + ( lOA) +10 2 A2 + ...X10"', +10 2 "'2 + ... - a ]( acos",+eu) +102

2
x-- u2 + )
...
a", a",
I I )
2
+ ( a cos'" + 10 u) + 10 u2 + ...) =10 [1 - ( a cos'" + 10 u) + 10 2 u2 + .. Y]
2
x[(eA) +10 A2 + ...) ~ +(1+10"'1 +10 2"'2 + ...) :",]
(4.3.15)
from which, we obtain to various orders in e,
0(10): a2~) +u) =2",) acos",+2A) sin",-a (1-..!. a 2 ) sin",+..!. a l sin3",

a"'j 4 4
(4.3.16)
o(10 2) : -aa",
2
- 2 + u2 =
u2 [(2 '"2 + "',2) a - A, -dA) ] cos '"
da
+[2(A2 +A) ",))+aA I d~l] sin",

2 2
-2 a u) -2A a u)
"') a",2 ) aa a",
2
+ (1- a cos
2
",)( AI cos'" - a "') sin", + ~~ )
+ a 2 u) sin 2", (4.3.17)
etc.
"') =0, A) ="21 a ( 1-"41 a-') . (4.3.18)


3
a .
U, =- 32 sm3",. (4.3.19)
()2
()"'-
U,2 +U,
-
=[2a IlF , -A
y - I
dA I
da
+(1-~4 a 2) A +~]
128 I
COS"F
y
(4.3.20)
3 2
a (a +8) 5a 6
+ 2 A, sin", + cos 3", + - cos 5",.
- 128 128
A, =0, IIF, =~ (dA -1+~ a2)_~. I

(4.3.21)
- Y • 2a da 4 256
3 2
5a 5
a (a +8)
u,
-
=---
3072
cos 5", -
1024
cos 3", . (4.3.22)
Using (4.3.18), (4.3.19), (4.3.21), and (4.3.22), we have from (4.3.1) and
(4.3.2),
(4.3.23a,b,c)
da=e!!"(I_"!"a 2 )
dt 2 4
d", =1 + e2 [~( d AI _ 1+ ~ a2 ) _ ~].

dt 2a da 4 256
Observe that equation (4.3.23b) is identical to equation (4.2.21) that was

obtained by using the method of averaging.
4.4. Whitham's Average Lagrangian Method
The averaged Lagrangian method (Whitham, 1965, 1967) is based on

variational principles. The main appeal of the variational principle is its power
of synthesis. The whole dynamics of the problem is expressed in tenns of a
4.4 Whitham's Average Lagrangian Method 123
single function through which the governing equations as well as all the
necessary boundary conditions follow by taking the functional derivatives.
In order to illustrate this method, consider the following nonlinear dispersive
wave propagation problem (Bretherton, 1964),
tPtI +tPxxxx +tP.o. +tP = EtP 3 , E« 1. (4.4.1)

The linear problem associated with equation (4.4.1) has the following
solution
cp=acosO,O=kX-OJt}
(4.4.2)
OJ2 = k 4 _ k 2 + 1 .
It may be readily verified that equation (4.4.1) has a variational character set
forth by the following Lagrangian -
L =.!.2 'I' _.!.2 'I'xx +.!.2 'I'x _.!.2 'I' +.!.4 E 'I'
2
11I/
2
11I . 11I
2
11I
2
11I
4
(4.4.3)
because extremization of L, namely,

8 If Ldxdt = 0 (4.4.4)
leads to (see Appendix 1 for a review of calculus of variations),
~ (aL )+~ (~)_.!.- (~)_ aL =0 (4.4.5)

at atP, ax atPx ax 2
atPxx atP
which, on using (4.4.3), in tum gives equation (4.4.1).
In the presence of a weak nonlinearity, one may argue that the form of the
solution (4.4.2) is maintained, but the amplitude a will not be a constant, and 0
will not be linear in x and t. The wave number k and frequency OJ may now be
generalized by defining them as
k=ao OJ=_ao. (4.4.6)
ax ' at
The parameters a, k, and OJ are now slowly-varying functions of x and t. We
have from (4.4.6) the following compatibility condition-
ak + aOJ =0. (4.4.7)
at ax
The variations of the parameters a, k, and OJ are governed by the variational
equations for the averaged Lagrangian. Thus, putting tP =a cos 0, using (4.4.6),
and averaging the Lagrangian (4.4.3) over a period, we obtain
5£=_1
2rr
f Ld8
2"
0
(4.4.8)
=.!..(m2_k4+e_l)a2+3e a4 +0(e 2).
4 32
The variational equations for 5£ are obtained by extremizing 5£, first with
respect to a,
d5£ = 0 (4.4.9)
da
which gives on using (4.4.8)
m2 = k4 _k 2 +1+2. ea 2+0(e 2). (4.4.10)

4
Next, extremizing 5£ with respect to 8, we have
(4.4.11)
which, on using (4.4.6), becomes
_i. (d5£) +J.... (d5£) _ d5£ =O. (4.4.12)

at dm ax dk d8
~ (m a 2 ) + ~ [( 2k 3 - k) a
2
]= 0. (4.4.13)
We have from (4.4.10)
(4.4.14)
d (ma-» + ax
at d ( m dm
dk a- 0) =O. (4.4.15)
from which
da+ -
m -
2
d (dm
-a 2)] +a 2(dm - -_ 0 .
- +dm-dm) (4.4.16)
[ at ax dk at dk ax
We have from equation (4.4.7)
(4.4.17)
Using equation (4.4.17), equation (4.4.16) becomes
~ (a 2
)+ ~ (~~ a )=o. 2
(4.4.18)
Equation (4.4.18) implies that the quantity a 2 (which may be energy in some
sense) propagates with the group velocity dm.
dk
4.5. Hamiltonian Perturbation Method
4.5.1 Systems with Constant Parameters

This method applies to systems with near-integrable Hamiltonians of the
form
H(I,() =Ho(I) + e HI (I,() + e 2 H2 (I,() + ... (4.5.1a)
where I is the action and () is the angle (see Appendix 2 for the action-angle
formulation). The system associated with Ho (I) is soluble (or integrable), and
its solution is considered known. Canonical perturbation theory is based on the
assumption that there is a region of phase space in which the phase curves of H
and H o may be continuously deformed into each other. (In other words, during
small intervals of time, the perturbed system moves in an orbit of the same
functional form as the unperturbed system but with parameters changing with
time.) The basic idea of the present method ' is to find a new set of action-angle
variables (J,qJ) for the perturbed system H(I,() such that the new Hamiltonian
K (J) found by means of a canonical transformation is a function only of J.
For the unperturbed system, Hamilton's equations are
i=- aHo =0 (4.5.2)

a()
. dH
() = _ 0 = m (I) (4.5.3)
dI 0
I While the present procedure uses canonical variables, it may be mentioned that other
perturbation theories, such as the Lie transfonn theory (Cary, 1981) are also available for
this purpose.
and for the perturbed system, Hamilton's equations are
I· - - -_- edH)
- - dH --e2 - + O( e3)
dH2 (4.5.4)
d8 d8 d8
. dH
8=-=(00 () dH)
I +e -+e 2-dH+2
0 (3)
e . (4.5.5)
dI dI dI
The object is to find the generating function S(8,1) which produces the
transformation H(I,8) ~ K(J) through the relations
I_dS _dS (4.5.6)
- d8' <p- dJ'
where <p is the new angle variable.
Let us expand S in a power series in e as follows:

S =So + e S) + e 2 S2 + ... (4.5.7)
where So is the identity transformation, So =J8 so that, from equations (4.5.6),

(4.5.7) gives a near-identity transformation-
I = dS =J + e dS) + e 2 dS2 +.... (4 5 8)

d8 d8 d8 ..
Let us further expand K (J) in another power series in e,
K (J) = K o (J) + eK, (J) + e2 K 2 (J) + .... (4.5.1 b)
Using the expansions (4.5.7), (4.5.8), and (4.5.1b), the Hamilton-Jacobi

equation
H(~~, 8) =K(J), (4.5.9)
on expanding each term in these expansions in Taylor series, becomes,
H (J) + e [dHo dS) +

o dJ ( 8 )
n]
+ e2
2
[.!-
d Ho (dS))2
2 dJ2 d8
dH dS, dn dS ] 3 , 3
+ dJo dO + aJ) d~ +H2 +0(e )=Ko (J)+eK, (J)+e-K 2 (J)+0(e ).
(4.5.10)
Equating coefficients in equation (4.5.10), of equal powers of e, we obtain
(4.5.11)
O( e) .. aHo ao + HI(J' 0) =K I (J)

()J as, (4.5.12)
2
0(e2) : ! a Ho (aS I )2 + aHo aS2 + aHI aSI + H (J 0) =K (J) (4.5 13)
2 ()J2 ao ()J ao ()J ao 2' 2 •
etc.
Recalling that the motion is periodic in 0, we average equation (4.5.12) over
0; we then obtain
(4.5.14)
where
f HI (J,O) dO.
_ 1 2/[
H, (J,O) = -
2rc 0

a
ao SI (O,J) = - W
I -
(J) HI (J,O) (4.5.15)
o
where
HI (J,O) =HI (J,O)- HI (J,O)}

W (J):: d H o •
o dJ
Writing SI (O,J) and HI (J,O) as Fourier series in 0,
ik8
HI (J,O) = fA k(J) e )
(4.5.16)
SI (J,O) = fisk (J) e ik8 '
Ir-I
we obtain from equation (4.5.15)
(4.5.17)
Using (4.5.17), (4.5.16) gives
SI(J,O)=i iAk(J) eik8 . (4.5.18)

k=1 kw o (J)
Using (4.5.18) in (4.5.6), the new angle is given by

as
qJ=-=O+e -+
l
...
aS (4.5.19)
()j ()j
while, from (4.5.8), the new action is given by
J=I-e - +
laS
... (4.5.20)
ao
The perturbed frequency from (4.5.16) is given by
co(J) =COo (J)+e

aK
al + .... (4.5.21)
Note that if COo is small, the effect of the perturbation can be quite large.
Since the nth-order correction has terms proportional to 1/co~ , this difficulty
gets compounded at the higher-order corrections. Usually, COo is small near a
separatrix dividing phase-space into regions containing different types of
motion. This problem becomes much more serious, as we will see later, for
systems of n degrees of freedom (n ~ 2) because, there are n fundamental
frequencies. These frequencies and all their integer linear combinations occur in
the denominator of the expressions for the various orders of corrections, so that
even if all the frequencies are large, a particular linear combination of them may
be small. Thus, the convergence of perturbation expansions for systems of more
than one degree of freedom is a very tricky problem.
Example 5: Consider the motion of a simple pendulum executing oscillations

about the downward vertical. For this problem, the unperturbed Hamiltonian is
Ho(P,q)=~ (p 2+ cog q2) (4.5.22)
while the perturbation is
HI (p,q) =-co~ (cosq-I +~ l)- (4.5.23)
The Hamilton-Jacobi equation for the linear problem given by H o (p, q) is
'2I (as
aqo )2 +'21 COo2 q 2 = E (4.5.24)
where S (q, I) is the generating function such that ~ = p.

The action I is given by
f = - 1.£
j 2 ( E - -1 W o
"q-) dq = -E (4.5.25)
2n c 2 Wo
where C is the closed path connecting the turning points ± .fiE. Thus, the
Wo
generating function So (q,I) with f, the action, as the new "momentum" is given
from equation (4.5.24) by
So (q,I) = J 2(fWo-~w~l) dq.

qo
(4.5.26)
The new coordinate, which is merely the angle 0, is then given by
- -dS-os- m
0-
df
ffio)
. -I ( q -
2!
(4.5.27)
from which
~- f sm
q= . 0. (4.5.28)
Wo
Next, we have, from (4.5.26),
aq 1W,oq-') ='IJ~
P = dSo =2 ( fw o - 1. 2!wo cosO. (4.5.29)
Using (4.5.28) and (4.5.29), we have from (4.5.22) and (4.5.23),
H o (I) =fw o }
1 2 . 4 (4.5.30)
HI (I, 0) =-"6 f sm 0
•
On averaging over 0, we have from (4.5.14),

- ( J,O ) =- -1f-
J =HI
K I () , "" - - 1J-
,. (4.5.31)
16 16
Using (4.5.10), (4.5.28), and (4.5.31), (4.5.1b) gives the new Hamiltonian to
0(£),
(4.5.32)
and hence the new frequency,
w(J) = ~ =Wo - £ i+O(£2). (4.5.33)

4.5.2 Systems With Slowly Varying Parameters

In classical mechanics, whenever a system has a periodic motion, the action
integral f pdq taken over a period is a constant of motion. This is nearly so
even when slow variations (compared with the period) occur and the motion
then being not quite periodic; the approximate constant is then called an
adiabatic invariant. More generalIy, a Hamiltonian system with constant
parameters has certain constants of motion; when these parameters vary slowly
these constants of motion break up but new approximate constants of motion,
calIed adiabatic invariants, appear.
Consider a Hamiltonian system with a parameter a, with the Hamiltonian
given by H =H(p,q,a). By means of a generating function F(q,O,a),
f
( F (q, 0, a) is related to the usual generator S = pdq by F = S - JO) one may
go to an action/angle variable representation (see Appendix 2). The new
Hamiltonian is given by
dF dF.
K(J,O,a) = H(J,a)+at' = H(J,a)+ da a. (4.5.34)
Hamilton's equations in the new frame then give
j=- ~~ =-a :0 (:} (4.5.35)
If a is slowly-varying, J will be smalI, so in order to find its cumulative value

over long times, one averages j over a period .. (treating a to be a constant in
this interval),
a
J"" = - ~ I.j dO d (dF)
da dt + 0(00a, a'2) . (4.5.36)
f
From S = pdq and J = 2~ f pdq, it is obvious that when (J increases by 21r,
S wilI increase by 2;r J. Noting that JO will also increase by 2;r J during this
period, we see that F = S - J(J is a periodic function of (J and thus can be
represented by a Fourier series:
~k8
F= .L...Ak (J,a) el
• (4.5.37)
k
a fL'k
J"" = - -
.. k
I -dAk eik8dt + 0(00a a'2) = 0(00a a'2)
da ' ,. (4.5.38)
Thus, J is zero to within 0(a,a 2 ); J is an adiabatic invariant. An adiabatic

invariant is a dynamical quantity which is approximately constant under slow or
adiabatic changes in the Hamiltonian.
One may characterize slow variations by introducing t =e t, e« I, thus
a=a(t); then a=ea'(t).
Equation (4.5.38) then implies that J is zero within 0(e 2).2
Example 6: Consider a simple pendulum with constant string length. Then, both
energy E and frequency (0 are constants. If the string length is now allowed to
vary slowly, neither E nor (0 will be constants; however, the ratio of E and (0,
namely, J = E, will be an adiabatic invariant, i.e., the change in J will be of a
(0
higher order than the change in the parameter a.
A procedure to construct adiabatic invariants of nearly periodic systems with

many degrees offreedom was also given by Gardner (1959) and Kruskal (1962).
If the original system is Hamiltonian, then one can define the usual action
integral
taken around a closed curve. J remains constant in the course of time if the curve
evolves in accordance with the equations of motion. This result is then used to
reduce the given system to a system of one less degree of freedom by means of a
canonical transformation to new coordinates, one of which is the constant J, in
which case its conjugate coordinate, viz., the angle variable t/J describing the
phase around the closed curves, is ignorable. If the reduced system can itself be
put in Hamiltonian form, and if its solutions are nearly periodic, the whole
procedure can be reapplied.
2 In studying adiabatic invariants using perturbation theory, it must be noted that a

quantity might appear to be constant to all orders in some perturbation parameter E'
without being only constant. For example, the change in the adiabatic invariant may be
described by a term like e-I/£ which does not have a power series expansion in E about
E = 0 (see Example 3 in Chapter I).
As an illustration of the method of canonical transformations on the

Hamiltonian of systems with slowly varying parameters, let us consider the
motion in a system of nonlinearly-coupled oscillators with slowly varying
individual frequencies.
For a system of nonlinearly coupled oscillators there is, in general, no
effective energy sharing without an internal resonance in the system (see also
Example 9, Chapter 6). The energy sharing is usually brought about by resonant
interactions among the natural modes of the system. Mathematically, an internal
resonance occurs if the uncoupled frequencies {O)k} of the various oscillators
L nk O)k
N
are such that =: 0 for a set of integers {n k }. Here, N is the number of

k=1
oscillators in the system. However, if one allows the various system parameters,
such as the uncoupled frequencies of the various oscillators, to vary slowly in
time (the restriction to slow variation of the system parameters is crucial for the
workability of an analytic perturbation theory), then an internal resonance can
prevail only temporarily. One issue of interest in these problems is how the
presence of nonlinear coupling affects the adiabatic invariance of the individual
actions of the two oscillators. It turns out that for times t sufficiently far from the
instant, say T, corresponding to the advent of an internal resonance, the two
oscillators individually exhibit adiabatic action invariants. The latter have
different values in the pre- and post-resonant regions. During the period when an
internal resonance prevails, a certain linear combination of the various actions
will be an adiabatic invariant (Kevorkian, 1980, Shivamoggi, 1987).
Example 7: Consider a system of two nonlinearly-coupled oscillators with

slowly varying individual frequencies O)k (O)k'S change very little during the
individual periods 2n/O)k) having the Hamiltonian
H =2.I (PI2 + P22) + 2.1 (0)12 ql2 + 0)22 q22) + e (ql2 q2 - '31 q23) (4.5.39)
where the small parameter e characterizes the weak coupling between the two
oscillators. Let the parameters 0)1 (i) and 0)2 (i), where t =e t, evolve in such
a way that the system passes through a state of internal resonance. We now
apply successive canonical transformations on the Hamiltonian to eliminate the
resonant variables (when an internal resonance prevails) from the Hamiltonian.
Using a canonical transfonnation generated by a function
fo ~2IlJ)1 -OJ~q~ dql +f ~2I2OJ2 -OJ;q; dq2

~ ~
S(q"q 2J"I2,t) = (4.5.40)

0
we have the following action-angle fonnulation
- -
q, -f!II
OJI
.
sm<p" q2 - -fE -
I2
OJ 2
sm<p2
.}
(4.5.41)
PI = ~2II OJ I cos <PI' P2 = ~212 OJ2 cos <P2
where II and 12 are the actions and <PI and <P2 are the corresponding angles of
the two oscillators.
In the action-angle fonnulation, the new Hamiltonian H is then given by
OJ' sin2<P2 ]
OJ' J sin2<P1 +_2_
+e _1_1
[ 2OJ, 2OJ2
(4.5.42)
where primes denote differentiation with respect to the argument.
Because of the nonlinear coupling between the two oscillators and the
variations of the OJ/s (represented by the O(e) tenns on the right in equation
(4.5.42» the actions I k of the two oscillators will undergo slow variations.
However, since the nonlinear coupling is weak and the variations in the OJk's are
slow, for times t sufficiently far from the instant T corresponding to the advent
of an internal resonance, the changes in I k become effective only when they
accumulate over long times. Thus, averaging over <Pk' we obtain
(iI) = I,OJ I + 12 OJ 2 (4.5.43)
according to which
I, = constant, 12 = constant. (4.5.44)
Thus, for times t sufficiently far from the instant T corresponding to the advent
of an internal resonance, the individual actions of the two oscillators are
adiabatic invariants.
In order to treat the situation for times t near the instant T when an internal
resonance prevails, i.e., m2 (i) = 2m, (i)
(which can be recognized from the
0(£) tenns in if), let us eliminate the resonant variables from if by a
canonical transfonnation to a frame of reference that rotates with the resonant
frequency. The new coordinates then describe the slow variation of the variables
about their values at resonance. We then average over the rapidly rotating phase.
The new canonical transfonnation is generated by the function
S(i')2 ,CI'"CI'z) = (CI'z -2C1't) i2 +CI', it (4.5.45)
where ik are the new "momenta". The new coordinates are then given by
_ as
CI', = --- = CI',
all
(4.5.46)
_ as
Cl'2 = ai =Cl'2 - 2C1',
2
On the other hand, the old "momenta" are given by
as - -)2
I, =;;-=1,-21
Cl'1
(4.5.47)
as -
I, =--= I,
• aCl'2 •
2j /
+-.!.- ( 2)3 2sin 3 (q,2 +2q,\)}+£ [w; (j\ +2j2) sin2q"
12 w2 2w\
+ w; j2 Sin2(q,2 +2q,\)]. (4.5.48)

2w2
Note that near the internal resonance, w2 = 2w 1, q,2 is the slow variable. On
averaging over the fast variable q,\, we obtain
~ +I~2(W 2-2w\)-£ [1-21

(H~) =I\w\ 1 2 (4.5.49)
2w\
Since, according to (4.5.49), q,\ is now the ignorable coordinate, we have
II =constant (4.5.50)
or, from (4.5.47),

II + 212 = constant. (4.5.51)
Thus, during the period an internal resonance prevails, a certain linear

combination of the actions I k is an adiabatic invariant, which of course signifies
an exchange of energy between the two oscillators that occurs during an internal
resonance.
4.6. Applications to Fluid Dynamics: Nonlinear

Evolution of a Modulated Gravity Wavepacket
on the Surface of a Fluid
Let us consider the nonlinear evolution of a wavepacket (for which the

energy of the motion is concentrated in a narrow wavenumber band)
propagating on deep water using the averaged Lagrangian method (Lighthill,
1967, Whitham, 1965, 1967, and Yuen and Lake, 1975).
Consider an initially quiescent semi-infinite fluid confined in the region
y < 0, and subjected to a gravitational field g = -giy' (see Figure 4.4). Let
Y = T'/(x,t) denote the disturbed shape of the free surface (whose mean level is
given by y =0). The motion of this fluid system is described by the following
Lagrangian (Luke, 1967)-
L= L
I)(x./
l[ 1 ]
tPt + 2" (tP; + tP:) + gy dy (4.6.1)
where tP(x,y,t) is the velocity potential describing the flow, so the fluid
velocity v is given by v = V tP .
The variational principle for the wavemotions on the free surface of a fluid is
postulated to be
o JJLdxdt =0
t, x,
(4.6.2)
subject to the restrictions that 0tP, 077 = 0 at x = XI' Xl and t =t), t 2 .
liquid
Figure 4.4. Deformed surface of fluid under gravity.
Following the usual procedure of calculus of variations, we obtain from

(4.6.1) and (4.6.2):
y < 77(x,t) : tPxx + tPyy = 0 (4.6.3)
y= 77(x,t): -77/ +tPy -77xtPx = 0 (4.6.4)
tPl + 2"1 (2
tPx + tPy2) + gy = 0 (4.6.5)
y~-oo : tPy ~ O. (4.6.6)
Equation (4.6.4) describes the kinematic condition on the velocity field at the
free surface. Equation (4.6.5) describes the dynamic condition of force balance
at the free surface.
Let us now consider a finite-amplitude stationary gravity wave of frequency

0.>0 and wavenumber ko propagating in the x-direction, and superpose on it a
slowly-varying weak modulation, and study the evolution of such a modulation.
If, following Whitham (1967), we assume that the wave can still be taken to be
sinusoidal locally, i.e.,
1J = acosO (4.6.7)
but with amplitude and phase varying slowly in x and t, i.e.,
a =a(x,t) }
, (4.6.8)
0= O(x,t) = kox - o.>ot + cp(x,t)
then we may introduce a generalized frequency co and wavenumber k:
= -Or = 0.>0 - CPt}
0.>
. (4.6.9)
k=Ox =ko +CPx
We have from (4.6.9), a compatibility condition:
dt + do.>
dk ax =0 . (
4.6.10)
Using (4.6.3), (4.6.4), and (4.6.6), we obtain

1J = acosO+ka 2cos20 (4.6.11)
2
iP=[o.>a sinO+ o.>~x (l-ky)cosO+.!. at COSO] eky + o.>a sin20e 2ky . (4.6.12)
k k k 2
In order to obtain the equations describing the long-time evolution of the
wave, we use (4.6.11) and (4.6.12) in (4.6.1), and calculate the averaged
Lagrangian ;£:
;£=_1
2n
10
LdO= (_ 0.>2
4k
+~) a2 + a;
4 4k
+ a all
2k
0.> ax at 3 0.>2 a aX). 3 0.> a a" 0.>2 a; 1 , ka 4
+--+- +- - - - + - -3+ -
2 4k 8 e 4 e 8k 8
0.>- •
(4.6.13)
Variation of ;£ with respect to 0 gives
~ (d;£)_~ (d;£)+ d;£ =0 (4.6.14)

dt do.> ax dk dO
which, on using (4.6.10), gives the well-known equation-
-a (a 2) +- - a 2) -_0 .
a (dW (4.6.15)
at ax dk
Next, variation of ;t with respect to a gives

a;t =0 (4.6.16)
aa
which, on using (4.6.15), gives
W=W (1 +-
o 2
1 k2
a 2 - -1- -
2wo
2
d w
-O
de -
a '
an) (4.6.17)
where
Wo = ~gko . (4.6.18)
For weak modulations, using (4.6.17), we may write

2
dwo (k-k ) +-
W=W +_. d w O [( k-k )2 -a-
1 -_. xx ] +_.
aw (a-' -a-') . (4.6.19)
°dko °2dkg 0 a aa~ 0
Using (4.6.9), we obtain from (4.6.19) and (4.6.15),

o [2 2(2 2) _
2
dw o lfJx +-
lfJ, +-- 1-d -W2 1
lfJx - - +- woko a -aD - 0
axx ]
(4.6.20)
dko 2 dko a 2
2
(a 2)' ddw o ( a 2) +-,-

+--
k
d W o ( lfJxa 2) =0.
dk
(4.6.21)
o x x
o
Ifwe put
(4.6.22)
then we may combine equations (4.6.20) and (4.6.21) into the SchrOdinger
equation:
2
i (ax + dwo ax )+.!.. d w2Oa2X2 =.!.. w k 2(I 1_I 1 2 2
). (4.6.23)
at dk ax
o
2 dk ax 0
2 X Xo 0 0
In order to investigate the stability of the modulation governed by equation

(4.6.23), we put
(4.6.24)
where
(4.6.25)
Then, equations (4.6.20) and (4.6.21) give

4.7 Exercises 139
(4.6.26)
(4.6.27)
Let us put
(4.6.28)
and assume
in _ j(I(~-Ot)
PP't'l e (4.6.29)
and linearize in PI and (fJ1' We then obtain from equations (4.6.26) and (4.6.27):
(4.6.30)
(4.6.31)
from which we obtain for n:

n =± W o I(
2 )] 1/2
8eo ~
2 2 ( 2
[ 8e 0
_ k2 1 1
0 Xo (4.6.32)
Equation (4.6.32) shows that the modulation is unstable (n 2 < 0) if
II(I <.J8 kg IXol·
4.7. Exercises
1. Use the Krylov-Bogoliubov method of averaging to construct an approximate

solution of
u+(8+ecos2t)u=0, e«1.
2. Use the Krylov-Bogoliubov-Mitropolski method to find the limit cycle of
u+e(u 2 -1) u+u-eu 3 = 0, e« 1.
3. Solve the forced Duffing equation-

2
d u
- , + W o' 3 1.
•u = eu + eF:o cos II.l
dr
140 Method of Averagiog
using the Krylov-Bogoliubov-Mitropolski method for W o ~ A. and W o "" A.

separately.
4. Use the Krylov-Bogoliubov-Mitropolski method to construct an approximate
solution for a system of two nonlinearly-coupled oscillators
x.. + W\2X = -e· 2xy,
y+w;y=e(i _x
2
), e«1.
Show that this solution breaks down for an internal resonance w 2 =2w\ .
Construct a solution that deals with this case.
5. Show that the Boussinesq equation
tPlt - tPxx - tPxxxx - (tP 2 ) xx

=0
has the variational characterization
8 If Ldxdt =0
where the Lagrangian is
1 1 1 1
L =1. l/Iltl/lxx - 1. l/Ixx2 + 1. l/Ixxx2 +"3 l/I xx'
3
tP =l/Ixx'
Find the averaged Lagrangian ;£ and the variational equations.
6. Find the first-order corrections to the motion of the linear oscillator with the
Hamiltonian
perturbed by
7. A particle of mass m slides under the influence of gravity on a smooth rigid

wire in the shape z =.!. a
2 2
x , where the z-axis is vertically upwards. Find
2
the Hamiltonian for the system and show that an approximation to this
Hamiltonian, correct to O(p2 x 2), is
P 2 mg a4
H=-+- a 2x 2 - - X
2
p2
2m 2 2m
Find an approximation to the motion which is correct to 0 (p2 x 2).
4.8 Appendix 1 141
4.8. Appendix 1
Review of Calculus of Variations
Al.I Functionals with Second-Order Derivatives

Calculus of variations is a generalization of the theory of extremals of point
functions. The problem is to find a function y =q,(x) from a suitable class r of
admissible functions, which are single-valued with continuous first and second
derivatives on the interval of interest, such that the functional
I
x,
I(y) = F(x,y,y') dx (AU)

X,
is an extremal. A necessary condition for the functional I to have an extremal is

the vanishing of its first variation.
In this context, it is useful to briefly review first the extrema of point
functions.
derivatives and r
If a point function f(x,y) is defined throughout a region R and if the partial
a; exist and are continuous in R, then necessary
conditions that f possess an extremal at some point (xo,Yo) in R are that

C!flax =C!f Idy =0 at the point (xo,Yo) or that the differential dfsatisfies
df= af dx+ af dy=O (Al.2)

ax dy
at the point (xo,Yo)' Let us now generalize the concept of differential (Al.2) so
that we can treat functionals in a similar manner.
Ifwe write
y =q,(X)+E1](X) (Al.3)
the change E1](X) is called the variation ofy,
Dy == E1](X). (AlA)
Here, Dy plays a role similar to that of the differential df in the discussion of

point functions. However, from a geometrical viewpoint, the differential df of a
point function f(x) of one variable is a first-order approximation to the change
in that function along a particular curve. On the other hand, the variation oy
represents a first-order approximation to the change in y from curve to curve.
More generally, we have the following definition for a variation offunctional.
Definition: The first variation ofthe functional I is
oI=e lim
£--+0
3-
de
I(t/J+E1J)
whenever this limit exists for all 1] in the class r.
In order to calculate the first variation of the functional I, one compares the
values of I that correspond to neighboring curves y(x) anchored on the same
end points at x = XI and x 2 (see Figure 4.5).
y=~(x)
Figure 4.5. Variations about the optimal curve.
Applying the variational operator 0 to the functional -
f
x,
I(y) = F(x,y,y') dx, (A 1.5)
X,
4.8 Appendix 1 143
we have
x,
tSI(y) = JtSF(x,y,y') dx (Al.6)

X,
under suitable continuity requirements on F.

Note,
tSF =e lim..:!.... F(x,t/> + E7J, t/>' + ETf')

£-->0 dE
= e [7J(x) aF aF].
ay +7J'(x) ay'
Setting
E7J =tsy and E7J' =tsy' ,
we have
aF aF,
tSF=- tsy+- tSy . (Al.7)
ay ay'
Therefore,
(Al.8)
Noting the commutativity of the two operators ts and d/dx,

,
by' == ETf' = (E7J) = ( by)
and integrating by parts, we have
tSI = a~ by]Xlxxay
ay 1
1
+ [aF _ ~ (a~)] bydx.
1
ay
(A.19)
We have the forced boundary conditions

x=x 1 : y=y,}
(A 1.1 Oa)
x = x 2 : Y = Y2
so,
or
(A1.1Ob)
Lemma: If g(x) is a continuous function in Xl $ X $ x 2 • and if
111(X) g(x) dx = 0
for arbitrary 11 in class r for which 11 (Xl) = 0 and 11 (X 2 ) = 0, then g( x) == 0

in XI $ X $ x2 ·
Using the above lemma, an extremal of I(y), which corresponds to 81=0,
with 81 given by (A. 19), is given by the Euler-Lagrange equation -
dF _.!!...- (dF) =O.

dy dx dy'
(A 1.11)
Al.2 Functionals with Higher-Order Derivatives

Consider extremals of functionals of the form
1
I(y) = F(x,y,y',y") dx
X,
(A1.12)
where F satisfies general continuity and differentiability requirements in each of

its arguments x,y,y',y". We seek extremals of (Al.12) subject to the forced
boundary conditions -
(Al.B)
A necessary condition for 1 to have an extremal is the vanishing of its first

variation. In accordance with the variational notation introduced above, we have
(A 1.14)
, "
Replacing 8y' by (8y) and 8y" by (8y) , and then integrating by parts we
obtain for an extremal of I,
(A1.15)
4.8 Appendix 1 145
from which, invoking the Lemma, we deduce the Euler-Lagrange equation
aF _.!!...(aF)+ d (aF)=o. 2
2
(Al.16)
~ dx~' dx ~"
Because of the forced boundary conditions (A 1.13), we have <5y =0 and
Dy'=O at X=X 1,X2 ' and no natural boundary conditions are required. On the
other hand, if no forced boundary conditions are specified, then in order to
satisfy (AI.I5) we must prescribe the natural boundary conditions
[:- ~(:,)]I X=X I

=0 (Al.17)
[:- ~ (:)]1 X=);2

=0, aFI
~"
=0. (Al.18)
One may generalize the above results to functionals of the more general form
1
l(y) = F(x,y,y', ... ,y(n)) dx
X,
(Al.19)
where n is a fixed integer satisfying n > 2. For extremals of such functionals, it

can be shown that the corresponding Euler-Lagrange equation has the form
aF _.!!...(aF)+ d (aF)_
2
...+(_I)n d n (aF )=0. (AI.20)
2 n
~ dx~' dx ~" dx ~(n)
A1.3 Functionals with Several Independent Variables

Let us now consider a functional that is a function of a point function u of
two independent variables x and y. We assume that u is continuous with
continuous first and second partial derivatives in some bounded, simply-
connected region R with smooth boundary curve C and assumes prescribed
boundary values on C but is otherwise arbitrary. The functional that we shall
study is then defined by
l(u)= If F(x,y,u,uX'u
R
y) dxdy (A 1.2 1)
where ux ' uy and F are continuous and F has continuous first and second partial
derivatives in each of its arguments.
In considering functionals of the type (A1.2I), it proves to be useful to note
Green's Theorem in the Plane:
Theorem: If R is a bounded, simply-connected region bounded by a simple

closed curve3 C in the x,y-plane, and if p(x,y) and Q(x,y) together with their
first partial derivatives are continuous in R, then
Jf( a;-:) tb:dy=f[p(x,y) tb:+Q(x,y) dy].
Ifwe perturb the solution u about the extremal solution q,(x,y) according to
u = q,(x,y) + £17(x,y) = q,(x,y)+l5u (A 1.22)
then we have for the first variation of I,
(A1.23)
(A1.24)
then we can rewrite (A1.22) as
81= JJ[aF
R au _~(aF)_~(aF)]l5utb:dY
ax au ()y au
x y
(AI.25)
+ JJ[~ax (aF
R au l5u) + ~
x
(aF l5u)] tb: dy.
()y au y
By applying Green's Theorem in the Plane to the second integral in (A1.25),

we obtain
(A 1.26)
-' A simple closed curve is a curve consisting of a finite number of arcs with continuously
turning tangents, joined at the end points, but not crossing over itself.
4.8 Appendix 1 147
Here,
For stationary values of I(u), we have ~I = 0, which upon invoking a two-

dimensional form of the lemma, leads to the two-dimensional Euler-Lagrange
equation
(Al.2?)
along with either the forced boundary condition -

u = f(x,y) on C (Al.28)
in which case 8u = 0 on C or the natural boundary condition -

aF _ aF dy = 0 on C (Al.29)
au au
y x dx
in order that the line integral in (A 1.26) vanishes.
Example 1 Consider the functional

I(u) = ff(u~ +un dxdy
R
with u =f(x,y) on C.
We have for this functional -

aF aF
-=0, -=2u x ' -=2uy .
aF
au au x au y
The Euler-Lagrange equation (Al.27) associated with the stationary condition of

the above functional then reduces to the Laplace equation
The solution of this equation in a particular region R of the xy-plane, subject to

the forced boundary condition u = f(x,y) on C, is the Dirichlet problem.
4.9. Appendix 2
Hamilton-Jacobi Theory4
A2.1 Hamilton's Equations

Consider a system of n degrees of freedom which is described by a
Lagrangian L 5 which is a function of the generalized coordinates
qj' (j =1,2, ... ,n), the generalized velocities qj (j =1,2, ... ,n) and the time t-
L = L(qj,qj,t). (A2.l)
Then, the actual motion of the system from a position qj'l at time tl to a
position qfl at time t2 can be determined from Hamilton's principle of least
action which requires that the integral of the Lagrangian function takes the
minimum possible value between the initial time t\ and the final time t 2 for the
actual path. More precisely, the actual path in configuration space between two
configurations qitl) and qit2) at times (I and t2 , respectively, is that which
makes the time integral of the Lagrangian function stationary with respect to
variations &i j of the path which vanish at the end points, i.e.,
JL(qj,qj,t) dt = o.
I.
&I> = 8 (A2.2)
The functional is called Hamilton's principal function (or the action integral)
for the path q(t).
Using the techniques of variational calculus, (A2.2) can be expressed as
&I>=t
j=1
J[aL()qj _.!!...(a~)]&ijdt=O'
I dt aqj
j=1,2, ... ,n. (A2.3)
I
Because of the arbitrariness of &ij between t =t1 and t =t2 , equation (A2.3)
leads to Lagrange's equations
4 See Shivamoggi (1997) for a more detailed discussion.

5 A Lagrangian of a system is typically the difference between the kinetic energy and the
potential energy of the system.
4.9 Appendix 2 149
aL _!!... ( aL ) =0, j =1,2, ... , n. (A2A)

dqj dt dqj
Ifwe introduce generalized momenta according to
aL
Pj = dqj' j = 1,2, ... ,n, (A2.5)
Lagrange's equations (A2A) become

. aL
Pj = dqj , j = 1,2, ... ,n. (A2.6)
Let us now use qj'Pj,t as the independent variables (qj and Pj are called
the conjugate variables) rather than qj'qj' t. Noting that
<5L = L aL 8qj + L a~ 8i;j + aL &

j dqj j dqj at
= LPj 8qj + LPj 8i;j + ~ & (A2.7)

J J
=<5 (LPAj)+ Lh 8qj -

J J
Lqj 8pj +
J
~ &,
we have
which describes a Legendre transformation from L to the Hamiltonian H,6

H=H(qj,pj,t)= LPjqj-L. (A2.9)
j
We have from (A2.9)
8H =L aH 8qj + L aH 8pj + aH & (A2.l0)

j dqj j dpj at
which, on comparing with equation (A2.8), leads to Hamilton's equations-
6 H is typically the energy of the system.

dH
qj = dpj
. dH
p.=-- (A2.1l)
J dqj
dH dL
-=--
at at j=1.2 .... n
Observe that if a particular coordinate is cyclic, i.e., it does not appear explicitly
in H, the corresponding momentum is a constant of the motion. Similarly, if L
does not depend on t explicitly, H is a constant of motion and we have
conservation of energy.
A2.2 Canonical Transformations

Canonical transformations can be of practical use in simplifying the
integration of Hamilton's equations.
Let L(qj,qj,t) and I(q>t,t) be two Lagrangian functions involving the
same number of degrees of freedom. Let qi and qi be related so that a path in
the qi -space for which the time integral of L is stationary corresponds to a path
in qi-space for which the time integral of I is stationary. Then these two
Lagrangians provide two different descriptions of the same system. This implies
that
(A2.12)
because L is unique to within an additive total derivative of a scalar function VI. 7

Thus, we have from (A2.12),
oI (([j,qj,t) =0L(qj,qj,t)_!!.-
dt
OVl(qj'([j,t)
or
7The total time derivative cannot contribute to the variation of the time integral in
Hamilton's principle (A2.2).
4.9 Appendix 2 151
~ aI ~ _ . aI
£..
I
::J;:;.
vtJ.,
Iiij; + £.. P; Iiij; +
I
at &
~ aL ~ . aL
=£.., ..:In. Oq; + £..pJiq; +ii &
Vlf, J
_!!... (L a",
dt ; dq;
Oq; + L a", Iiij; + a",
; iJii; dt
&). (A2.13)
Choosing
Pi = a",
dq; , P-, =_a:,." 1=,
. 1 2,... ,n,
VIf, (A2.14)
we find, from (A2.13) that the Lagrange equations remain form invariant under
a group of canonical transformations generated by the function '" (qj' qj' t) .8
Using (A2.l2) and (A2.l4), the new Hamiltonian is then given by

- ~ . -
H = £"Pi qi -L = £..P/l; -L+;;
~ a",
I ,
(A2.l5)
=H(q;.p;.t) + ~ ",(q;.q;.t).
'" is called the generating function, and we have from (A2.l4) and (A2.15),
(A2.16)
One may find other types of generating functions by making Legendre

transformations. Thus, letting
'" (qj,qj,t) =tit (qj'Pj,t) - LPj qj (A2.l7)
j
we find that tit (qj' Pj' t) generates a canonical transformation according to

atit(qj,pj,t) _ atit(qj,pj,t)
Pj = dqj ,qj = dpj . (A2.l8)
Using (A2.15) and (A2.18), we have for the new Hamiltonian
_ ( _) ( ) ati/ (q j' Pr t)
H qj'Pj,t =H qj'Pj,t + dt . (A2.19)
8 Note that the coordinates and momenta do not remain necessarily distinct under such
transformations.
A2.3 Hamilton-Jacobi Equation

Let us choose a canonical transformation so that li =constant. Then, we may
take li = 0, which trivially integrates the canonical equations to give
p; =constant and qj =constant.
Noting (A2.19), the Hamilton-Jacobi equation for the generating function tjI
is then given by
--Y [r)
rat + H q.j'dqj'
--Y t = O'
,
j = 1, ... ,n. (A2.20)
If H is not an explicit function of t, then

H(q;,pj) = constant = ai' say. (A2.21)
We now seek a canonical transformation generated by tjI = S( q;, a;) such that
all the new momenta pj are constants, say aj" The new Hamiltonian li will
then be equal to H or a l and will be cyclic in all the new coordinates qj' The
new equations of motion are
q = ali = ali = {I, i = 1

I (Jp; aa 0, i:f:. 1 j
(A2.22)
. ali
P = - dq; = 0; i = 1, ... , n
The generating function S( q;, a j ) produces the following transformation:

as_ as as
aa j
Pj = dq; , q; = (Jpj = (A2.23)
and satisfies the following equation:
H(q;, ::)=a t • (A2.24)
a l is called an isolating integral (Whittaker, 1964) since it isolates one degrees

offreedom from the other (n -1) degrees offreedom.
In order to see the physical significance of S, note that

dS~aS. ~.
di= L I
.:lr..
CAfE
q; = LPjq;
I
(A2.25)
4.9 Appendix 2 153
so that
(A2.26)
which is simply the action integral!
A2.4 Action-Angle Variables

In cases where the motion is periodic, one may be interested in some average
characteristics of the motion rather than the details of the motion. Toward this
objective, one modifies the Hamilton-Jacobi approach slightly so that the
integration constants a j are chosen to define a set of n constants called action
variables J;.
Consider, for the sake of illustration, a system with one degree of freedom
described by a Hamiltonian H =H(q,p). Apply a canonical transformation
generated by S(q,p) so that
(A2.27)
Hamilton's equations in the new coordinates are then

~ ali ~ aH
p=- aq' q =a;;' (A2.28)
Let p=constant=J, say. Then 'if =() is a cyclic coordinate, and the Hamilton-
Jacobi equation (A2.20) for the generating function S becomes
- -
H=H(J)=H(q,p)=H q'iJq ( as) (A2.29)
with
()= asdJ' (A2.30)
On the other hand, from equations (A2.28), we have

. dli
() =""d:i = OJ, say, (A2.31)
from which,
() =OJt + 8. (A2.32)
Note that if J has the dimension of action, 0 is dimensionless and is called

the angle variable.
Let us suppose now that q and p are periodic functions of t, and let us take J
to be the action evaluated over one period
J=_l 1. pdq. (A2.33)

2n '1
Note that during one period of the motion, 0 increases by an amount
110 = 1. dO =
'1
1.~(aS)dq = ~
'1aq ()J
1. as dq = ~ 1. pdq = 2n
aJ'1aq ()J'1
(A2.34)
where we have use (A2.27), (A2.30), and (A2.33). (A2.34) shows that CiJ is the
frequency of the periodic motion. Since the integral t pdq represents the area
enclosed by an orbit of energy H = E in the phase plane, each orbit is uniquely
labelled by J, which is constant along every orbit. Each point on an orbit is
labelled by a single-valued function of O. Thus, the action-angle formulation
enables us to calculate the frequencies of the periodic motions directly without
finding the variations of the coordinates with time.
On using (A2.27), (A2.33) shows that
J = 2~ t(~) dq. (A2.35)
So, the change in S during one period is given by
I1S == tdS = 21t1 (A2.36)
The relation (see (A2.l7))

S(q,p) =S(q,O)+JO (A2.37)
then shows that S (q, 0) is a periodic function 0 with period 2n.

Chapter 5
The Method of Matched

Asymptotic Expansions
5.1. Introduction
In cases where a small parameter multiplies the highest derivative in a

differential equation, there occurs a sharp change in the dependent variable in a
certain region of the domain of the independent variable. In constructing a
solution to the differential equation through uniformly-valid expansions, one
characterizes the sharp changes by a magnified scale that is different from the
scale characterizing the behavior of the dependent variable outside the
"boundary-layer" regions. In other words, one represents the solution by two
different asymptotic expansions using the independent variables x and x/e say.
Since they are different asymptotic representations of the same function, they
should be related to each other in a rational manner in an overlapping region
where both are valid (Friedrichs, 1955); this leads to the asymptotic matching
principle (the latter makes the two representations completely determinate).
5.2. Physical Motivation
In order to physically motivate this method, consider the response of a linear

spring-mass-damping system initially at rest to an impulse 10 , Denoting the
mass by m, the damping constant by p, and the spring constant by k, we have
the initial-value problem (Cole, 1968):
156 The Method of Matched Asymptotic Expansions
my+ fJY+ ky =f oO(t)}.

(5.2.1)
y(O-)=O, y(O-)=O
For very small values of m, the initial-value problem (5.2.1) may be
approximated by
(3y<0) + ky(Ol =f o 0 (t)}.
(5.2.2)
y(O)(O-) = 0
Since the order of the differential equation in (5.2.2) has decreased from 2 to 1,
one of the two initial conditions in (5.2.1) must be dropped. This implies the
existence of a boundary layer at t =0, and the above truncated equation is valid
only for t away from t = O. Noting, from (5.2.2), that
y(O)( 0+) = ~ ,
we have
I _E.
y(O) =...Q. e f3 (5.2.3)
{3
However, in a short interval near t = 0, the displacement is produced infinitely
rapidly from 0 to ~. In order to describe the motion during this initial interval,
we note that inertia is dominant at t =0 (impulse-momentum balance) and that

due to the large initial velocity, damping becomes effective immediately, but the
spring does not, since the deflection is small. Thus, for t=:O, the initial-value
problem (5.2.1) may be approximated by
my(i) + (3y(i) = foO(t)}
. (5.2.4)
y(i )(0- ) =y(i)(O-)=O
Integrating with respect to t from 0- to t, we have from (5.2.4),

mil + {3y(i l = f o (5.2.5)
from which we obtain
(5.2.6)
where c is an arbitrary constant.

Noting from equation (5.2.5) that
5.2 Physical Motivation 157
(5.2.7)
we have, on using (5.2.6),

1
c =_-iL (5.2.8)
f3'
Using (5.2.8), (5.2.6) becomes
y(i) = 1p [ 1-e_!!..t]
m • (5.2.9)
Equation (5.2.3) and (5.2.9) appear to imply that

y(i)( 00) =y(o)( 0+). (5.2.10)
Equation (5.2.10) embodies the general idea behind the asymptotic matching
principle.
The solutions (5.2.3) and (5.2.9) are sketched in Figure 5.1 along with the
exact solution of (5.2.1). Observe that the exact solution looks like y(l) near
t = 0 and like yt0) as t becomes large.
o t
Figure 5.1. Inner, outer, and exact solutions.

5.3. The Inner and Outer Expansions
Let us now illustrate the method of matched asymptotic expansions through

an example.
Example 1: Consider the boundary-value problem:

t)''' + y'
+ y = 0, 0$ x $I}.
(5.3.1)
y(o) = a, y(l) = b
The exact solution of the boundary-value problem (5.3.1) is
(e) _ (ae s, - b) eS'x + (b - ae s, ) eS'x
(5.3.2)
y - eS ' _e s,
where
-1 ± ..JI -4e 1
s ,
1.-
= 2e
'" -1, - - + 1.
e
The above exact solution may be approximated by
y'" (_;_1) [-be- X

+(b-ae-
I
) /-;]
or
y = be l - x + (a - be) e- x/c + O(e) (5.3.3)
which shows a uniformly valid form.

Note that
lim
c....o
y =bel-X}
(5.3.4)
lim y=a
.\"-+0
from which
lim (lim
c....o x....o
y) =a '* limo (limo y) =be
x.... c....
(5.3.5)
so that the two limits are non-commutable.

Note that the expansion (5.3.3) cannot be obtained by keeping either x or x/e
fixed. In the former case, one obtains
y(O) = be l - x +O(e) (5.3.6)
which is not valid in the boundary layer near x = 0 since

y(O)(O) = be:t a. (5.3.7)
In the latter case, one obtains, on the other hand,

y(i) =be+(a-be) e- x/ t +O(e) (5.3.8)
which is not valid, as x ~ 1, since

y(i) (1) =be :t b . (5.3.9)
This suggests that we represent the solution by using two different asymptotic
expansions using the variables x and x/e. The occurrence of the term e- x/ t
makes this problem singular because there is no expansion of e- x/ t , which is
valid in a neighborhood of e =0 .
Let us seek, for the boundary-value problem (5.3.1), an outer expansion
(valid away from x =0) ofthe form-
N-I
y(O)(x;e) - Ieny~o)(x)+o(eN). (5.3.10)
n=O
We then obtain, from equation (5.3.1), to various orders in e,
0(1) : y~(O)' + y~O) =0 }
O(e) : y~O) + y~O) = _y;(O)

" (5.3.11)
etc.
Note that in the outer expansion the order of equation (5.3.1) is reduced.
Therefore, the outer equations (5.3.11) cannot take on both of the boundary
conditions in (5.3.1), and one of these boundary conditions, viz., y(O) =a, must
be dropped. This means that y(O) is valid everywhere, except in the region
x =O(e). Thus, we have the following boundary conditions on y(O) -
(5.3.12)
Using these conditions, we obtain, for equations (5.3.11), the following
solutions -
(5.3.13)
etc.
Thus, the outer expansion is given by

y(O) =b[l+e(l-x)] el - x +0(e 2 ). (5.3.14)
For small e, y(O) is close to y(e) everywhere except in a small interval at x =0,
where y(e) changes rapidly in order to retrieve the boundary condition there
which is about to be lost (see Figure 5.2).
be
b
I
(c) I
Y I
I
I
________ IL_
a
I
I
o x
Figure 5.2. Inner, outer, and composite expansions for equation (5.3.1).
To determine an inner expansion which is valid in the boundary layer in

x =O(e), we introduce a new independent variable ~ = x/e. This allows the
width of the boundary layer region to become independent of e as e=>O and
enables us to retain the highest derivative in the given equation when e ~ O.
This is essential to representing the rapid variation of y in the boundary layer.
We then have
(5.3.15)
Let us seek an inner expansion of the following form-

N-I
y(i)(~;e)_ Ie n y~i)(~)+o(eN). (5.3.16)
n=O
We then obtain, from equation (5.3.15), to various orders in e-

2 (i) (i)
0(1): d Yo + dyo =0
d~2 d~
(5.3.17)
etc.
Noting that y(i) is valid only in the region x =O(e), we have the following
boundary conditions on y(i) -
y~i)(O) =a}
yll)(O) = 0 . (5.3.18)
etc.
Using these conditions, we have, for equations (5.3.17), the following solutions -
0(1): y~i) =a-Ao(I-e-~) (5.3.19)
O(e): y~i) =A,(I-e-~)-[a-Ao(I+e-~)] ~ (5.3.20
etc.
Thus, the inner expansion is given by
y(i) = a- Ao (1-e-~)+e {AI (1- e-~)- [a - Ao (1 +e-~)]~} + 0(e 2 ). (5.3.21)
In order to relate y(i) to y(O) in an overlapping domain of validity, let us use

the following asymptotic matching principle (Shivamoggi, 1978b)'-
I There are other types of asymptotic matching principles available in the literature. One
such type is due to Kaplun (1957) which involves intermediate limits and the other type
"The n - termfannal Laurent series

expansion of the outer expansion about = "The n - term formal outer
the inner boundary written in terms limit of the inner expansion."
of the inner variable."
Thus, near x = 0, let us write
y(O)(x) =y(O) (0) + xylOl' (0) + 0(x 2 )
=YbO) (0) + e [~ YbO/ (0) + YiO)(O)] + 0(e 2 ) (5.3.22)
=YbO) (0) + e [~Y~O)' (0)+ YiO)(o)J + 0(e 2 ).

We then have according to the above matching principle
be+e[be-be~]+0(e2)=(a-Ao)+e [AI -(a-Ao)~]+0(e2), (5.3.23)
from which
Ao = a- be, AI = be, etc. (5.3.24)
Using (5.3.24), the inner expansion (5.3.21) becomes
y(i) = be +(a - be) e-; + £ {be(l- e-;) - [be- (a - be) e-;] ~} + 0(£2), (5.3.25)
y(O) is valid everywhere except in a small interval of 0(£) near the origin
while y(i) is valid only in a small interval of O(e) near the origin, Although
y(O) and y(i) have overlapping domains, one needs to switch from one expansion
to the other if a numerical solution is desired over the whole interval. However,
the switching location is not known precisely. This difficulty can be
circumvented by combining both expansions into a single composite expansion
y(c) _
y(c) = y(O) + y(i) _ y(o), (or /J.), (5.3.26)
where y(O), represents the inner limit of the outer expansion and /), represents
the outer limit of the inner expansion,
Note,
is due to Van Dyke (1975). (See Fraenkel, 1969 and Eckaus, 1979 for a critical
assessment of the various asymptotic matching principles.)
and
y(C),= yeo), + i)' _/0),. = y(O), + y(i) _ yeo), = y(i)
so that yCd reproduces y(O)in the outer domain while it reproduces y(i) in the
inner domain. Therefore, y(d is valid everywhere.
For the present example, on using (5.3.14) and (5.3.25), we have for the
composite expansion -
y(c) =b [1+e(l-x)] e1- x +0(e2)+be+(a-be)e-~
+e {be (I - e-~ )- [be - (a - be) e-~] ~}
+0(e 2)- [be + e(be - be~)] + O( e 2 ) (5.3.27)

or
x
y(c) =b[l+e(l-x)] e l - x +[(a-be)(l+x)-ebe] e--; +0(e 2).
Here, note that
e-~ = {o(e n ) as e => 0, \::in, if x = 0(1),

(5.3.28)
0(1) as e => 0, if x = O(e).
Let us now consider the determination of the location of boundary layers in a

singular-perturbation problem.
Example 2: Consider the boundary-value problem

tY" + a(x) y' + b(x) y = O}. (5.3.29)
y(O)=a, y(I)=f3
In the limit e ~ 0, equation (5.3.29) gives
a(x) y' + b ( x) y =0 (5.3.30)
whose solution cannot satisfy both boundary conditions, and one of them must
be dropped as a consequence. The boundary condition that must be dropped
depends on the sign of a(x) in the interval (0, I). In general, if the outer limit of
the inner solution diverges and does not exist, the boundary layer does not arise
at the assumed location. Thus, the location of the boundary layer is determined
so as to have the resulting inner expansion possess a proper outer limit. If
a(x) > 0, y(O) =a must be dropped, and an inner expansion near x =0 must
be developed and matched with the outer solution. If a(x) < 0, y(l) ={3 must be
dropped, and an inner expansion near x =1 must be obtained and matched with
the outer expansion. If a(x) changes sign in (0,1), y may change from
oscillatory to monotonic across the zeros of a(x) (called the turning points - see
Section 5.9).
Let us consider here the case a(x) > O. Then, we have for the outer solution
y~o) _
a(x) y~o)' + b(x) y~o) =o} (5.3.31)

y~o) (I) = {3
from which we have
y~o) ={3 exp[-J a(t)

x
b(t) dt]. (5.3.32)
I
In order to obtain the inner solution, introduce a new independent variable
~ =x/t: (5.3.33)
so we have, in the limit x ~ 0,
(5.3.34)

y~i) =a - B+ be-a(og (5.3.35)
where B is an arbitrary constant.

Matching asymptotically y~o) with y~i) , we obtain
B =a - {3 exp [i a(t)
o
b(t) dt]. (5.3.36)
Using (5.3.32), (5.3.35), and (5.3.36), the composite expansion is then given
by
it) = y(O) + y(i) _ y(O),

X b(t) ] {
=f3exp [ -{ a(t) dt + a-f3exp
[I[a(~)
b()
dt
]} a(O)x
e--'-+O(e).
(5.3.37)

e y" + (2x + I) y' + 2y =O}. (5.3.38)
y(O) =a, y(l) =f3
Look for an outer expansion of the form -
~
y (0)= ~>n y~O) (x). (5.3.39)

n=O
We then obtain from the boundary-value problem (5.3.38)
(2x + I) y~O)' + 2y~0) = o}

y~O) (I) =f3 (5.3.40)
etc.
On solving these problems, we obtain
y~O)=~}
2x+1 . (5.3.41)
etc.
Look for an inner expansion of the form -
(5.3.42)
We then obtain from the boundary-value problem (5.3.38)
y~i)" + y~i)' =OJ

y~i) (0) =a . (5.3.43)
etc.
y~i) =a - A (1- e -~)}

(5.3.44)
etc.
where A is an arbitrary constant.

Asymptotic matching between y(O) and y(i) then gives
lim y~O)
x=>o
=lim y~i)
~=>~
(5.3.45)
from which, we have on using (5.3.41) and (5.3.44),

3{3=a-A
or
A=a-3{3. (5.3.46)
Using (5.3.46), (5.3.44) becomes

y~i) =a-(a-3{3)(I-e-~). (5.3.47)
The composite expansion is then given by

y(c) =y(O) + y(i) _ y(i)o
(5.3.48)
=~+(a-3{J)e-~ +O(e).
2x+l
5.4. Hyperbolic Equations
Consider the linear hyperbolic partial differential equation (Cole, 1968)-
ax _a
e( iiv at v)=a au
ax +b av
2
2 at 2 (5.4.1)
which has real characteristics (see Figure 5.3)

r = t - x, s = t + x. (5.4.2)
The characteristics serve to define the region of influence, propagating into
the future, of a disturbance at a point Q (see Figure 5.3). The manner of
specification of boundary conditions on an arc for a fully-posed boundary-value
problem depends on the nature of the arc with respect to the characteristic
directions of propagation. One boundary condition is specified on the time-like
arc (see Figure 5.4) corresponding to one characteristic leading into the adjacent
region in which the solution is defined. Two boundary conditions are given on
the space-like arc (see Figure 5.4) corresponding to the two characteristics
leading into the adjacent domain. When the boundary curves are along the
characteristic curves, only one condition can be prescribed, and the
characteristic relations must hold. The characteristic-initial-value problem
describes one condition each on AB and on A C to define the solution in ABCD
(see Figure 5.5).
t
~
Q
o x
Figure 5.3. Characteristics and region of influence
(from Kevorkian and Cole, 1996).
rx
time-like
arc
s
space-like
" " / arc
" ~~
x x
Figure 5.4. Time-like arc and spac~like arc
x
Figure 5.5. Domain of influence in a characteristic initial-value problem
Consider the initial-value problem corresponding to equation (5.4.1) in

-00 < X < 00 with
t =0 : V =F (x), v, =G (x). (5.4.3)
According to the general theory of characteristics, the solutions at a point
P(x,t) (see Figure 5.6) can depend only on that part of the initial data which
can send a signal to P. This is part (XI < X < x 2 ) of the initial line contained
between the backward running characteristics through P.
Now, the outer problem associated with equation (5.4.1) corresponds to
taking the limit e=>O and is given by
av(O) adO)
a --+b - - =0 (5.4.4)
ax at
from which,
v(O)(x,t)=f(X-~ t). (5.4.5)
In the limit e=>O the solution v(x,t) depends only on the data connected to P
along a subcharacteristic of equation (5.4.1) given by
bx - at =const. (5.4.6)
p(x, t)
x
Figure 5.6. Domain of influence and time-like and space-like
subcharacteristics (from Kevorkian and Cole, 1996).
Now the subcharacteristic, reaching P, originated at point A between x"x 2 if

Ibfal> 1 i.e., if it is time-like. Then the limit E=>O preserves the domain of
influence. However if Ibfal < 1 the subcharacteristic reaching P is space-like and
lies outside the domain of influence, originating at B (see Figure 5.6). In this
case, the limit E=>O increases the domain of influence and implies that the
outer solution cannot be obtained by taking the outer limit of the solution of the
full problem for equation (5.4.1), contrary to what should be the case!
It turns out that even the issue of stability of the solution u(x,t) (here
stability/instability refers to the exponential decay/growth along the
characteristic of the discontinuities in the derivatives of the solution across
characteristic) is related to whether the subcharacterstics are time-like or space-
like. In order to see that, note that in terms of the characteristic coordinates
(5.4.2), equation (5.4.1) becomes
a2 u au
-4E-=(b-a) -+(b+a) - .
au (5.4.7)
aras ar as
Consider the propagation of a jump in av/dr along r = ro = const. Let,
K=[av] =(av) _(av) . (5.4.8)

dr r=ro dr r; dr r·
Assuming that v itself is continuous across r =ro one finds from equation
(5.4.7)
aK
-4e-=(b-a) K,
as
from which
Now, a jump across a characteristic propagates to infinity along that

characteristic. Using (5.4.8) and (5.4.9), this implies
(b - a) > 0 ~ stability }
. (5.4.10)
(b - a) < 0 ~ instability
Similarly, a consideration ofajump in av/as across a characteristic s =So gives
(b + a) > 0 ~ stability }
(5.4.11)
(b + a) < 0 ~ instability .
From (5.4.10) and (5.4.11) one obtains,

Ib/al> 1 for stability. (5.4.12)
We restrict further discussion to the stable case.

Now, note that the solution v(OJ(x,t) given in (5.4.5) can only satisfy one
initial condition, so that one may expect the existence of a boundary layer on the
line t =O.
Assume an initially-valid expansion
v(i)(x,i;e) - vg (x,i)+ /31 (e) vii) (x,i) +...
J
(5.4.13)
where, i is the time scale relevant for t == 0 ,

- t
t=--
o(e)
and
with the associated inner limit process £0 => 0, x,i held fixed. Taking
(5.4.14)
the initial conditions (5.4.3) give
i =0 : v~i) =F(x), di) =0 for n>0

"
_ av(i) au(i)
t =O·_o_=O
. di ' al = G(x), (5.4.15)
av(i)
-"-
di
=0 for n >1
Choosing 0(£0) =£0, and substituting (5.4.13) and (5.4.14), equation (5.4.1) gives
2 (i) (i)
0(1) .. adiV2o + b avo
di
=0 (5.4.16)
a2v(i) av(i) av(i)

0( <»' --'-+b
<0di 2
• di ax
_l_=_a _0_ (5.4.17)
etc.
Notice that the boundary-layer equations (5.4.16) and (5.4.17) are ordinary-
differential equations, which is a feature of the boundary layers not occurring on
a subcharacteristic. This is true for any hyperbolic-initial value problem, since a
space-like arc can never be a subcharacteristic.
Using the initial conditions (5.4.15), equations (5.4.16) and (5.4.17) give
v~i)(x,i)=F(x) (5.4.18)
vli)(x,i)=[ G(x)+~ F'(x)][I-e-i]-~iF'(x) (5.4.19)
so that
v(i)(x,i;£O) = F(x)+£O[ {G(x)+~ F'(X)}{I- e- i } - ~ iF'(x)] + .... (5.4.20)
Note that (5.4.20) possesses terms that persist in the limit t=>oo, as well as
terms that decay in time which are typical of a boundary layer.
Next, construct an outer expansion, with the associated outer limit process,
£0 => 0, x andt held fixed,
v(O)(x,t, e) - v~O) (x, t) + £0 vlO) (x,t) + ... (5.4.21)

so that equation (5.4.1) gives

:l (0) :l (0)
0(1): a a~ + b a~ =0 (5.4.22)
av(O)
O(e)' a _ adO) (a 2 v(0)
1 - + b _ 1 _ = __
a 2 v(0»)
0 0_ (5.4.23)
. ax at ax 2 at 2
etc.
One obtains from equation (5.4.22)
v~O) = f(~), ~ = x-E. t. (5.4.24)

b
a
:l
u~
(0)
+b
:l
a~
(0) (
= 1- :2
2)
f"(~) (5.4.25)
from which
(5.4.26)
so that
v(O)(x,t;e)=f(~)+e [ J;(~)+ a2 2 - a 2
b b +a
b 2 2
( b)
x+- t f"(~) + ....
a
] (5.4.27)
The asymptotic matching between v(i) and v(O) requires

V(O)(x , O', e) + e i v(O)
J(x' O', e) + ... = v(i) (x , 00', e) • (5.4.28)
Using (5.4.20) and (5.4.27), (5.4.28) gives

f(x) =F(x),
b2 2 (5.4.29)
J;(X)+~ 2 -a 2 x!,,(x)=G(x)+E.F'(x)
b b +a b
(5.4.30)
Consider, next, a radiation problem in which boundary conditions are

prescribed on a time-like arc and propagate into the quiescent medium in x > 0,
(Figure 5.7). When the boundary condition is prescribed for instance, at x =0,
one has to distinguish two cases depending on whether the subcharacterstics run
into or out of the boundary x = o. RecaIl, from (5.4.6), that the subcharacterstics
are given by
a
~ = x - - t = const. (5.4.6)
b
Subcharacteristics
t
Characteristic X =t
u=o
a> 0 X
Characteristic X =t
u=o
a < 0 X
Figure 5.7. Radiation problem with boundary conditions prescribed on a

finite portion of the boundary (from Kevorkian and Cole, 1996).
Note, that the characteristics are incoming or outgoing according as a:§ o.

Let the boundary condition be
x = 0: V = F(t), t> O. (5.4.31)
Outgoing Characteristics: Assume an outer solution,

dO)(x,t;e) - v~O) (x,t)+ev;O) (x,t)+··· (5.4.32)
where
b
v~O) =f(n '=t--x. (5.4.5)
a
Substituting (5.4.5) into the boundary condition (5.4.31), we obtain
b
t<-x,
0, a
°-
(O) -
V (5.4.33)
b
t >-x.
a
This solution obviously has a discontinuity on the particular subcharacteristic

through the origin. However, such a discontinuity is not permitted in the solution
to equation (5.4.1) with e:;t: O. Thus, in order to obtain a uniformly-valid
subcharacteristic ,=
solution, a suitable boundary layer must be introduced on the particular
0 which supports the discontinuity in the outer solution
v(O) . Assume an inner expansion
v(i)(x,t;e)-v~i)(x,i)+f.l(e) v;i)(x,i)+ ...
where,
_. x-!!..t}
b
x=-- (5.4.34)
8(e)
t =t
and
8==>0 as e==>O,
with an associated inner limit process e ==> 0, xand t held fixed. Choosing
8=.Jf
and substituting (5.4.34), equation (5.4.1) gives in the limit e ==> 0,
a v(i) adi)
2
K ax ° - at°
2 -
(5.4.35)
where
a2
1--
K==~>O
b
which ensures that i =t is a positive time-like variable so that equation (5.4.35)
outer expansion dO) on the subcharacteristic ,=

is a diffusion equation that describes the spreading of the discontinuity in the
O. Matching v(i) to v(O)
asymptotically, as before, one obtains
v~i)(x,i)= F~+) erfc [2k). (5.4.36)
Incoming Characteristics: Assume an outer expansion

v(O) (x,t;e) - v~O) (x,t) + e v~O) (x, t) +.... (5.4.37)
Since the disturbances now propagate along the subcharacteristics from the
quiescent region to the boundary, one has
dO) == O. (5.4.38)
Then the discontinuity in dO) occurs at the boundary x =0 so that one has a
boundary layer at x = O. Since the line x = 0 is not a subcharacteristic, the
boundary layer equations should now be ordinary differential equations. Assume
an inner expansion
(5.4.39)
where
x =~
<5 (e) ,
t =t and VI ~ 0 as e ~ 0 (5.4.40)
with an associated inner limit process, e ~ 0, xand i held fixed. If one

chooses <5 =e, substituting (5.4.39) into equation (5.4.1), we obtain
a 2 v(t) av(l)
= a_o_
ax
_ _0_
2 ax ..
(5441)
On using the boundary condition (5.4.31), one obtains from equation (5.4.41)
v~i) (x,i) =F(i) exp( aX), a < O. (5.4.42)
5.5. Elliptic Equations
(5.5.1)
where a's, a, and b are constants, and e« I, and

a~2 -a'la22 < o.
In order to determine a solution u(x,y;e) to equation (5.5.1) uniquely, it is
sufficient to prescribe one boundary condition on u or its normal derivative, or a
combination on a closed boundary.
Consider an interior boundary-value problem with u =U B (x B ) prescribed on
a closed boundary curve, (see Figure 5.8).
Figure 5.8. The subcharacteristics.
The curves
~ =bx - ay = const. (5.5.2)
are the characteristics of equation (5.5.1), in the limit e => 0, and are the
subcharacteristics of equation (5.5.1).
Introducing another independent variable
1J=ax+by, (5.5.3)
S.S Elliptic Equations 177
transfonning the independent variables x,y to C;, 7], equation (5.5.1) becomes
(5.5.4)
where
2
A = a ll b 2 -
2a l2 ab + a n a
II 2
a +b2
allab + a l2 (b 2 - a 2)- a 22 ab
A 12 = a2 +b 2 '
2 2
A _ a ll a + 2a l2 ab + a 22 b
22 - a2 + b2
Now, in equation (5.5.4), let
lim u (c;, 7]; £) => u(O) (C;) (5.5.5)
£=>0
~.ry fixed
where the boundary condition on one side of the domain (see Figure 5.9) is
sufficient to detennine u(O)(c;) uniquely in the whole domain, but u(°l(c;) does
not, in general, satisfy the boundary condition on the other side of the domain,
so that one may expect a boundary layer to arise there. In order to study the
latter region, we introduce a new independent variable
• 7]-7]B(c;)
(5.5.6)
7] = <5(£)
u =u(x)
Figure 5.9. Production of a boundary layer on "="11 (~)

with an associated limit process e ~ 0, 1]' and; held fixed. The retention of
the highest-order derivatives in equation (5.5.4) then requires 8(e) = e. Seeking
an asymptotic solution of the form
U' (;,1]';e) - u~ (;,1])+O(e), (5.5.7)
equation (5.5.4) gives, in the limit e ~ 0,

K"(;) UOry'ry' = U~ry' (5.5.8)
where
K"(;) == AII1]~~ - 2AI21]B~ + A22 •
The elliptic nature of equation (5.5.4) ensures that K"(;) > 0.

We obtain from equation (5.5.8),
u~ (;,1]') = A(;)+B(;) exp ( K"(~)). (5.5.9)
Equation (5.5.9) shows that the boundary layer occurs on the upper boundary.
Matching (5.5.9) asymptotically to the interior solution u~O)(;), we obtain
where the subscript U refers to values on 1] = 1] B ( ;).
This solution breaks down for the case when the boundary is a
;s.
subcharacteristic, say ; = We then introduce a new independent variable
(5.5.10)
and assume an asymptotic expansion

u· (f, 1]; e) =u~ (f, 1])+ O(e) (5.5.11)
with an associated limit process e ~ 0, ;* and 1] held fixed. Then equation

(5.5.4) gives
(5.5.12)
Since All> 0,1] is a time-like coordinate, which means that one requires the
prescription
5.6 Parabolic Equations 179
(5.5.13)
Further, the requirement of matching of u~ with the interior solution u~O) gives
(5.5.14)
Thus, the boundary layers arising on the subcharacteristics are characterized by

a diffusion-like behavior, which we also saw in Section 5.4.
If us(lJ) = const., then we have
u~(f,lJ)= Us +[u~O)(S)-us] erf ( 2~). (5.5.15)
5.6. Parabolic Equations
Consider a nonlinear diffusion equation (also called the Burgers equation

(Burgers, 1948»-
Ut +uux =EUxx" -00 < x < 00, t > 0 (5.6.1)
U(x,O)=ep(x). (5.6.2)
It is assumed here that ep(x) is smooth and bounded except for a jump
discontinuity at x =O. Moreover, ep'(x);;:: 0 for x * 0, and ep(O-) > ep(O+).
Look for an outer solution ofthe form -
(5.6.3)
Substituting (5.6.3) into the initial-value problem (5.6.1) and (5.6.2), we obtain
u~~) + u~o) u~~) = 0 (5.6.4)
U~O)(x,O) = ep(x). (5.6.5)

The characteristics of this equation are given by (see Appendix 2)
x=xo+ep(xo)t. (5.6.6)
Therefore, the solution of the above equation is given by

u~O) (x,t) =ep(x o)' (5.6.7)
where Xo is given implicitly in terms of x and t via (5.6.6). This solution is valid
as long as the characteristics do not intersect.
Let us assume that there is a single (smooth) curve x = s(t) - the shock (see
Figure 5.10), where the characteristics intersect, and hence, the outer solution
becomes discontinuous across such a curve. In order to develop a proper
description in the transition layer around this curve, introduce a new coordinate
~=x-s(t) (5.6.8)
e
shock
X = s(t)
Figure 5.10. The shock curve.
and assume an inner expansion

u(I)(~,t) = u~i)(~,t) + e U~i)(~,/) + .... (5.6.9)
Substituting (5.6.9) into equation (5.6.1), we obtain to 0(1),

-S'(/)U(i) +u(i) u(i) = u(i) (5.6.10)
0, 0 0, 0"
~~ too : u~i) =u~O)± (5.6.11)
where,
We obtain from equation (5.6.10), on one integration,
u(i)
0,
=.!.2 u(i)'
0
- s' (I) u(i) + A (I) .
0
(5.6.12)
The boundary conditions (5.6.11) then give

(5.6.13)
This leads to
s' (t) = ~ HOl' + U~OL ), (5.6.14)
which detennines the position of the shock.

Using these expressions, another integration of equation (5.6.12) gives
1 ((01_ (01,) ~
. u(oj, + u(OL B(t) e -2"0 -"0
° - °
u('J(J:~, t) - ° -~(u,\O)--u,\O),)~
(5615)
..
1+ B(t) e 2
where B(t) is an arbitrary function in the zeroth order problem. The solution
(5.6.15) is sketched in Figure 5.11.
o x
Figure 5.11. The shock structure.
5.7. Interior Layers
The rapid variations in the solution that are typical of a boundary layer do not
have to occur only at the boundary. When this happens, the problems tum out to
be a little bit more complicated because the location of the layer is usually not
known until the expansions have been matched.
Example 4: Consider the nonlinear differential equation (Lagerstrom, 1988)-

ey" - yy' + Y =0, 0 <x <1 (5.7.1)
x=O:y=1 (5.7.2)
x=1 :y=-l. (5.7.3)

Equation (5.7.1) was advanced by Lagerstrom (1988) to model shock layers in
gas dynamics.
Equation (5.7.1) may be rewritten as
y'=z }
(5.7.4)
z' =~ y(z -1)
which admits an integral -

,
y- - e z + e en 11- zl =C, (5.7.5)
2
where C is an arbitrary constant.
This shows that z =1 or y =x + C is an exact solution represented by a
straight line in the (x,y)-plane, which no solution curve can cross. (see Figure
5.12).
For the two outer regions on either side of the interior layer (to be located
below), look for an outer expansion of the form
y(O)(x;e) - ybO)(x)+ey~O)(x)+ .. ·. (5.7.6)
Equation (5.7.1) then leads to
(5.7.7)
which gives
YbO) =x+a (5.7.8)
where a is an arbitrary constant to be determined by imposing the boundary
conditions (5.7.2) and (5.7.3).
z
3
c=-oo
-3 3 y
Figure 5.12. First integral: (Y; - EZ + E en 11- zl = c) curves
(from Lagerstrom, 1988).
If there is an interior layer at x =x o, then we have

(0) _ {x + 1, 0::; x< oX
(5.7.9)
Yo - x - 2, Xo < x ::; I.
In order to determine the solution that is valid in the interior layer, introduce
; = x-xo (5.7.10)
e
and look for an inner expansion of the form
(5.7.11)
Equation (5.7.1) then leads to

" ,
y~i) _ y~i)y~i) =0 (5.7.12)
which gives
(i) _ A 1- BeA~
Yo - I +BeA~ , (5.7.13)
A and B being arbitrary constants.

The boundary conditions to be satisfied by the inner solution (5.7.13) are
provided by the asymptotic matching of the outer and inner solutions
y~i)(±oo) = y~O)(x~). (5.7.14)
Using (5,7.9) and (5.7.13), this leads to

A = xo + 1 }
(5.7.15)
-A = x o -2
1
x =-
o 2'
A=~. (5.7.16)
2
In order to determine the other constant B, one reasonable assumption is that
y(i) is anti-symmetric about ~ =0 (see Figure 5,13). This leads to
-1
Figure 5.13. Interior layer transition.

B=l. (5.7.17)
The composite expansion for the region 0 < x < 1 is then given by using
(5.7.9), (5.7.13), (5.7.16), and (5.7.17),
3
(5.7.18)
which is shown in Figure 5.14.
2
I
- ---, - - Numerical -
Solution
~ - Composite
,...- Expansion -
c
.S!
:; 0
-----
"0
Vl
-1
-
-2
o 0.2 0.4 0.6 0.8 1.0
x - axis
Figure 5.14. Comparison between the composite expansion given in (5.7.18)

and the numerical solution of (5.7.1). In the calculations E= 10- 2
(from Holmes, 1995).
5.8. Latta's Method of Composite Expansions
Rather than detennining outer and inner expansions, matching them, and then
fonning a composite expansion, as in the above, Latta (1951) suggested that,
one may instead start with a solution which has the fonn of the composite
expansion.

e y" + y' + Y = 0, 0 :$; x :$; 1 (5.8.1)
y(O)=a, y(I)=b. (5.8.2)
Let us look for a solution of the form

00 x 00
y - Len fn(x) + e-; Lenhn(x). (5.8.3)

n=O n=O
Substituting (5.8.3) into equation (5.8.1), we obtain
e ~ enf:' +~ enf:+ ~ enfn +e-~(e~ enh;'- ~ enh; + ~ enhn)=O'

(5.8.4)
This gives at various order in e,
0(1) : fJ + fo =0 (5.8.5)
h~ -ho =0 (5.8.6)
J;(I)=b, fo(O)+ho(O)=a (5.8.7)
O(e): ft'+ J; =-fJ' (5.8.8)
(5.8.9)
(5.8.10)
etc.
0(1) : fo = be'-x x}
(5.8.11)
ho =(a-be)e
O(e): J; =b(l-x)e'-X }
(5.8.12)
hi =[-be + (a - be) x] eX
etc.
Substituting (5.8.1 1) and (5.8.12), into (5.8.3), we obtain
y=[be'-x +eb(l-x) e'-x]+e-x/t{(a_be)e X+
+e[-be +(a - be)x] eX} + 0(e 2 )
or
x
y - b[l +e(l-x)] el - x +e-~ {(a - be)(1 +x)-ebe} +0(e 2 ) (5.8.13)
in agreement with the results in Example 1.
Example 6: Consider the boundary-value problem (Nayfeh, 1973)

ey"+(2x+1)y'+2y=0,0:S;x:S;1 (5.8.14)
y(O)=a, y(l)=b. (5.8.15)

- g~) -
y- L en!,,(x)+e--' L enhn(x) (5.8.16)
n=O n=O
where g(x) is another function to be determined. Substituting (5.8.16), equation
(5.8.14) gives
-
e L enl:'+ (2x + 1) L
-I: +-
2 Len!"
n=O n=O n=O
+ e[(- g" + g'22)

e e n=O
i
enh n - 2g'
e n=O
i enh~ +i enh~']
n=O
(5.8.17)
+(2x + 1)[ - ~ ~ enhn+~ en h~ ] + 2 ~ enhn =O.

This gives at various orders in e,
O(I/e) : hog'[g' -(2x + I)] =0 (5.8.18)
0(1): (2x+1)f;+2fo =0 (5.8.19)
(-2g' +2x + 1) h~ +(2 - g") ho =0 (5.8.20)
etc.
Substituting (5.8.16), the boundary conditions (5.8.15) give
g(O)=O, fo(I)=b, fo(O)+ho(O)=a. (5.8.21)
O(lfe): g=X 2 +X
0(1): fr =~
o 2x+ 1 (5.8.22)
ho = a-3b
etc.
Substituting (5.8.22) into (5.8.16), we obtain

2
x +x
3b ----
y---+(a-3b) e e +O(e). (5.8.23)
2x+1
Example 7: Consider now a boundary-value problem for a diffusion equation

(Keller, 1968)-
eut=u xx ' O~x~b(r), r=et (5.8.24)
u(O, r) = tp(r), u[b(r), r] = O}.

(5.8.25)
u(x,O) ='If (x)
_ g(t) _
u= L e'f,,(x,r) + e--e L e'h.(x,r). (5.8.26)

.=0 .=0
Substituting (5.8.26) into the boundary-value problem (5.8.24) and (5.8.25), we

obtain to various orders in e:
0(1) : fr0" = ° (5.8.27)
10(0) = tp(r) (5.8.28)
fo(b(r),r)=O (5.8.29)
ho" + g'h o =0 (5.8.30)
ho(O, r) = 0 (5.8.31)
ho(b ( r), r) = ° (5.8.32)
fo(x, 0) + ho(x, 0) = 'If (x ) (5.8.33)
O(e) : J;" = 10, (5.8.34)
;;(0, r) = ;; [b (r), r] = ° (5.8.35)

hi" + g'h, = ho, (5.8.36)
=hi [b(r), r] =0
hl(O, r) (5.8.37)
J;(x,O)+ hl(x,O) =0 (5.8.38)
etc.
Solving the 0(1) problem, we obtain
fo =<t>(r{l- b~r)l (5.8.39)
Since the boundary conditions (5.8.31) and (5.8.32) on ho are homogeneous,

equation (5.8.30) has a nontrivial solution if and only if g' is one of the
eigenvalues
(5.8.40)
The corresponding eigenfunctions are

2
2 ] 1/ . IOrX
(5.8.41)
X.' =[ b(r) sm b(r)'
Thus,
ho =ao(r) X.. (x,r). (5.8.42)
Using (5.8.39)-(5.8.42), the O(e) problem becomes

(5.8.43)
hi (0, r) =hi [b(r), r] =O. (5.8.44)
Expanding hi in terms of the eigenfunctions X5'
=L
~
hi Cs(r) Xs(x,r),
s=1
L (g; - g;) Cs Xs = a~ X.. +ao X.."

~
(5.8.45)
s=1
Multiplying equation (5.8.45) by X..' and integrating over (O,b(r)) we have

J(G~ X; +
b(r)
Go X.-, X.-) dx = 0 (5.8.46)

o
which is just the solvability condition for equation (5.8.43).
From the normalization condition,
JX; dx = 1,
b(r)
(5.8.47)
o
we have
d
-dr JX.- dx =2 JX.- X.- dx =O.
b(r) 2 b(r)
(5.8.48)
0 0'
Using (5.8.48), we have from the solvability condition (5.8.46)

G~ = 0 or Go = constant. (5.8.49)
Using (5.8.39)-(5.8.42), and (5.8.49), we have finally
_]+ La [-] sin

d;
u- tp(r) [1 _b(r)
X -
J<~I
2
b(r)
-
b(r)
exp
J<
1/'
IC1tX
[
, ,
- leTr-
e
J-,-
r
0b- (;)
]
+O(e)
(5.8.50)
where
=[ b~O) ] [ljI(X)-tp(O) 1- b~O)

1/2 b(O) { []}
GJ< sin :;) dx.
5.9. Turning-Point Problems
5.9.1 JWKB Approximation

In this procedure, one assumes an exponential dependence for the fast
variation while the slow variation is described via a polynomial correction. With
this assumption, the first-term approximation of the solution is determined by
solving two first-order differential equations. One of these equations, called the
eikonal equation, is nonlinear and determines the fast variation in the solution.
The other, called the transport equation, is linear and determines the slow
variation. It turns out that the higher-order terms in the JWKB (Jeffreys, 1924,
5.9 Turning-Point Problems 191
Wentzel, 1926, Kramers, 1926, Brillouin, 1926) expansion are linear and can be
determined in principle conveniently.
Consider an equation of the form
d2
~+A? q(x)y= 0 (5.9.1)
dx-
where A is a constant.
If A» 1, then between two successive zeros of y(x), q(x) is nearly
constant. This suggests that we look for a solution of the form
y(x,A) =e"g(d). (5.9.2)

1 ,
A gxx + g; + q = o. (5.9.3)
Setting
gx = h or J
g = h dx , (5.9.4)
equation (5.9.3) becomes

1 2
A hx +h +q = O. (5.9.5)
Expanding h in powers of 1/,1"

1
h(x,A) - ho(x) + I hI (x) +"', (5.9.6)
equation (5.9.5) gives to various orders in ±:

0(1): h; +q =0, eikonal equation (5.9.7)
0(1/,1,): 2hoh, +h~ =0 (5.9.8)
etc.
We have, on solving equations (5.9.7) and (5.9.8),
0(1): ho(x)=±j~q(x) (5.9.9)
1 '
0(1/,1,): hI (x) = - - [enq(x)] (5.9.10)
4
etc.
Consequently, we have from (5.9.2), (5.9.4), (5.9.6), (5.9.9), and (5.9.10)
(5.9.11)
Thus,
(5.9.12a)
(5.9.12b)
This is called the JWKB approximation and is valid as long as x is away from
the zeros of q (x). The latter is called a turning point because y is oscillatory on
one side ofa zero of q(x) while it is monotonic on the other side.
Example 8: Consider the boundary-value problem (Holmes, 1995),

y" + A? e y
2x
=0, 0<x <1 (5.9.13)
x=O:y=a (5.9.14)
x=l:y=b. (5.9.15)
The solution (5.9.12) then becomes
y-e -~2 [be sinA(e -1)-asinA(e -e)] •

If2 X x
(5.9.16)
sinA(e -I)
The exact solution for this problem is

y=C,Jo(Ae x )+C2 Yo (Ae x ) (5.9.17)
where,
1
C, == ~ [bYo(A)-aYo(k)] ,
1
C2 == ~ [aJo(k)-bJo(A)] ,
~ == Jo(k) Yo(A)- Yo(k) Jo(A) .
Figure 5.15 shows that (5.9.16) and (5.9.17) are in good agreement.
1.0 ~---T"""---""T"""---"""'T"-----,.----..,
0.5 I-~~-t----t-+---l\-+---+f-l\-----t
c
,2
:; 0 t - - - + + - - - -......'----~+_--+__+_-+_-~
"0
Vl
-0.5 t----~--_+_+----~r___J-_+--~_+_I
-1.0 L- ..I...- ....... -'- --L --I
o 0.2 0.4 0.6 0.8 1.0

x - axis
Figure 5.15. Comparison between the exact solution (5.9.17)

and the JWKB approximation given in (5.9.16). In the
calculations, E = 10- 1 , a =1, and b = O. (The two curves are
so close they are essentially indistinguishable from each other.)
(From Holmes, 1995).
Example 9: Consider the eigenvalue problem (Holmes, 1995),

y" -+- ;} e 2x y = 0, 0 < x < I (5.9.18)
x =0 and I : y =O. (5.9.19)
From the solution (5.9.16), the eigenvalues are given by

sinA(e-l) == 0
or
A(e-l)==nJr
or
A == nJr . (5.9.20)
e-l
For the exact solution, on the other hand, the eigenvalues are given by
J o(Ae) Yo (A) - Yo (Ae) J o(A) =o. (5.9.21)
The comparison between these two sets of numbers (see Figure 5.16) shows that
the JWKB approximation does well even for the first few eigenvalues.
10· 1
~
"Q.
~.....
- ~&
'G--o.... G-
-v-.o-e..
~ -Q-o-o-~ "0
5 10 15 20
n
Figure 5.. 16. Relative error, in absolute value, between the JWKB
approximation (5.9.20) of the eigenvalues and the values obtained by
solving (5.9.18) numericaUy. Here E =10- 1 (from Holmes, 1995).
5.9.2 Solution Near the Turning Point

Let us write, for x == J1.,
q(x) =(x - J1.) f(x), f(x) > O. (5.9.22)
The JWKB solutions (5.9.12) are the outer expansions valid for
x > J1. and x < J1. which break down for x == J1.. In order to determine the inner
expansion valid near x =0, we introduce
~ =(x - J1.) )}/3 (5.9.23)
so equation (5.9.1) becomes
(5.9.24)
Putting
(5.9.25)
(5.9.26)
from which
ylil = a3 Ai(z)+ b3 Bi(z) 2 (5.9.27)
where Ai(z) and Bi(z) are Airy functions, which have the following
asymptotic values for large z:
A I.( Z) == 2/ii
1 -1/4-'
z e
Ai( -z) == In \zl-I/4 sin(~ + 1f/4)

(5.9.28)
.() 1
BIZ==/iiZ -1/4'
e
Bi( -z) == In Izl-1/4 cos(~ + 1f/4)
where
~ = ~ Izl3/2·
3
In order to match the inner solution (5.9.27) with the outer solution (5.9.l2a)
for ~ > 0, we note first that
)., f ~q('f) d'f =)., f ~'f - J.l ~ f( 'f) d'f ="32 ~ f(J.l) ~3/2 + o(
x x
).,-2/3) (5.9.29)
~ ~
so that (5.9.12a) becomes, for x == J.l and x> J.l,
y(O) _ ~[al cos(~ ~f(J.l) ~3/2)+ b sin(~ ~f(J.l) ~3/2)]+ ..., ~ > O.
l
4 ~f(J.l) 3 3
(5.9.30)
_r[ (2"3 )/2)

2 The solution of equation (5.9.26) is given by
y-v z C,IIf) z +C2 I_If) (2"3 z)/2)] .

Putting
Ai(z) a
==~-E [I_If) ZV2 )- 1 (~ZV2)1/) 1
Bi(z) == ~-E[ 1_ 1
/) (~ZV2 ) + IIf) (~Z)/2 ) J
the above solution becomes
y==C1 Ai(z)+C2 Bi(z).
Further, we note that for large ~ > 0, we have from (5.9.27) and (5.9.28),
.Jii [(2
. ~-1{4rI{12
l}- °3 sin 3~f(J1)~3/2+n/4)+b3cos (23~f(J1)~3/2+n/4 )] +...,
~>O.
(5.9.31)
Matching y(O} with l}, we obtain
).-1{6/16 [ . n n]
°1 = .Jii °3 SIn "4 + b3cOS"4 '
).-1{6 fl{6 [ 1! . n]
bl = .Jii °3 cos "4 - b3S1n"4 . (5.9.32)
Next, in order to match the inner solution (5.9.27) with the outer solution
(5.9.l2b) for ~ < 0, note
~ ~
). J~-q(-r) d-r=). J~(J1--r)f(-r) d-r

x x
= ~ ~ f(J1) (_~)3/2 + o( K 2/3) (5.9.33)

3
so that (5.9.l2b) becomes for x'" J1, x < J1,
y(O} _ ~_~I~:g {02 eXP[~~f(J1)(-~t ]+b2exp [ _~~f(J1)(_~)3/2 ]}+...,

~<o.
(5.9.34)
I I,
Also for large ~ ~ < 0, we have from (5.9.27) and (5.9.28),
y(i} _ (-~t;r1{12 {± 03 exp [ -~~f(J1) (_~t2]

(5.9.35)
+b3exP[~~f(J1)(_~)3/2]}+ ... , ~<o.
Matching y(O) with y(i}, we obtain
).-1{6 1'/6
°2 = .Jii b3 , (5.9.36)
Thus, an approximate solution to an equation with a turning point at x = J1 is

given by three separate expansions (Rayleigh, 1912)-
(i) one near X"" J.l;

(ii) one for x < J.l;
(iii) one for x > J.l.
5.9.3 Langer's Method

One may seek to describe the behavior of the solution of a turning-point
problem of the form
e 2 y" - f(x) y= 0 (5.9.37)
over the given entire domain in terms of the Airy functions. For this purpose, we
introduce another independent variable (Langer, 1931, 1935, Wasow, 1965),
~=e-2/3g(x) with g(O)=O, (5.9.38)
and look for a solution of the form

y=y(x,~;e). (5.9.39)
Noting that
(5.9.40)
gg,2 y~~+e2/3 g (2'

g yx~+g"y~) +e 4/3 J:
gyxx-~ f()
x y= 0 . (5.9.41)
If, for convenience, we set

gg,2 =f, (5.9.42)
y~~+e
2/3 ( 2 g")
g,Yx~+g'2Y~ +e
4/3 (1g,2Yxx )-':»'-0.
J:.. _ (5.9.43)
%f If(t)11/2 dt
2/3
g(x) = sgn(x) (5.9.44)

[ ]
Expanding the solution as

(5.9.45)
we obtain from equation (5.9.43), to various orders in e,

0(1) : Yo" - ~Yo =0 (5.9.46)
2 "
O(e)'Y
• Ii<
-~YI -
-
- g'
- y0" _L
g,2 y 0, (5.9.47)
etc.
Yo (x,~) =Co(x) Ai(~). (5.9.48)
YJ.. -~YI g" Co ] Al0' (~).

=- [ --;2 Co, +-;}"" (5.9.49)
" g g
Removal of the secular term in equation (5.9.49) requires

2 "
-C'+Lc
,0 ,2 0 =0 (5.9.50)
g g
from which
1/4
C X - Co - C g (5.9.51)
o( )- [g'(x)t - 0 ( 1) '
where Co is an arbitrary constant.
Using (5.9.51), (5.9.48) becomes

1/4
y(x)-Co [ ~~:~ ] Ai[e-213g(x)]+0(e213). (5.9.52)
Note that as x ~ +00, (5.9.52) gives
(5.9.53)
while as x ~ -00, (5.9.52) gives
y-
~
Co e
.JJr I/(xt 4
. [x
Sill
I
eJI/(t)1 112
dt+n/4
] (5.9.54)
in agreement with the WKBJ approximation discussed in the previous section.

5.10. Applications to Fluid Dynamics: Boundary

Layer Flow Past a Flat Plate
For flows past streamlined bodies at large Reynolds numbers (i.e., in fluids of
small viscosity), Prandtl (1904) proposed that one need recognize the effects of
viscosity only in a thin layer - boundary layer adjacent to the body and the rest
of the flow may be considered inviscid. As a first approximation, the inviscid-
flow equations are solved with appropriate boundary conditions, ignoring the
presence of the boundary layer. However, in general, the inviscid flow will not
satisfy the condition of the no-slip of the fluid at the body, and it is necessary to
introduce a boundary layer between the inviscid flow and the body to adjust the
inviscid solution toward satisfying this condition on the body.
Lagerstrom and Cole (1955) pointed out that Prandtl's boundary-layer theory
can be embedded in a systematic scheme of successive approximations via the
method of matched asymptotic expansions. However, this solution, which is
represented by an infinite series in powers of R~1/2 (R E being the Reynolds
number, RE =UL/v, U being a reference velocity, say, the velocity of the fluid
in the free stream far away from the body, L being a characteristic length of the
body, and V being the kinematic viscosity of the fluid) turns out to be
asymptotic.
We consider here the problem of two-dimensional viscous incompressible
flow past a flat plate. It turns out that the external inviscid flow is associated
with an outer limit process, and the boundary layer with an inner-limit process.
The order of the differential equations is lowered in the outer limit, and the
boundary condition of no-slip of the flow at the plate is lost so that the problem
under consideration is one of singular-perturbation type.
5.10.1 The Outer Expansion

Let x measure the distance along the plate from the leading edge and y the
distance normal to the plate.
The equations governing this flow are -
(i) conservation of mass -
V·v=o, (5.10.1)
(ii) conservation of momentum -
av +(v.V') v=_-!.. V'p+VV'2 V , (5.10.2)

at p
where v is the fluid velocity,p is the fluid pressure, and p is the fluid density.
Eliminating the pressure p, equations (5.10.1) and (5.10.2) lead to
a(i) + (v. V') (i) = VV'2(i) (5.10.3)

at
where (i) is the vorticity
(i):V'xv·f.
z (5.10.4)
Introducing the stream function l/I according to
v = V''P x fz ' (5.10.5)
equations (5.1 0.3) and (5.1 0.4) become
i. V' 2'P + (cl'¥ i. _cl'¥ i.) V' 2'P =V V' 4'P (5.10.6)
at ()y ax ax ()y
and
(5.10.7)
The boundary conditions on this flow are
y =0 :v=0 (5.10.8)
upstream: v =Uix ' (5.10.9)
Equation (5.10.8) describes the no-slip condition at the plate.

Assuming the flow to be steady, the boundary-value problem in
nondimensionalized flow variables is given by
(
'I'
y
i.-
ax 'I' i.-
()y _1
xR
V'2) V'2\{1 = 0 (5 .1 0 .1 0)
E
y =0 : 'I' =0, 'Py =0, 0 < x < 1 or oo}. (5.10.11)

upstream : 'I' - y
If the plate is semi-infinite there is no natural length in the problem; however,

the apparent difficulty may be circumvented by choosing some reference length
L.
Let us seek a straightforward asymptotic expansion, as RE ~ 00 , of the form
S.10 Applications to Fluid Dynamics 201
(5.10.12)
with the associated outer limit process RE ~ 00, and x,y fixed. Then, equation
(5.10.10) gives
11/(0)
"f' Iy
~
ax _II/(O)~)
Ix "f'
V 2 11/(0) =0
~ I "f'
(5.10.13)
(
from which
V2V1~0) =_(i)~0) (VI~O)). (5.10.14)
If the oncoming stream is irrotational, equation (5.10.14) gives

V2V1~0) =O. (5.10.15)
The boundary-conditions (5.10.11) give

y =0 : VI~O) =0 }
(5.10.16)
upstream : VI~O) - y .
Note that the no-slip condition

y =0 : VI~~) =0
has been dropped since, in the outer limit, the order of the differential equation
(5.10.10) has dropped.
From (5.10.15) and (5.10.16), we obtain
VI~O)(x,y)=y (5.10.17)
so that, in the limit R E ~ 00, a flat plate causes no disturbance.
S.10.2 The Inner Expansion

Because of the loss of the no-slip condition, the basic inviscid solution is not
valid close to the flat plate. Therefore, assume an inner expansion valid within
the boundary layer
VI(i)(x,y;RE) - O~i)(RE) Vlri)(x, Y) + of)(RE) VI~;)(x, Y) +... (5.10.18)
with the associated inner limit process RE ~ 00, x, Y fixed, where

- y (5.10.19)
Y - 0(i)( R ).
1 E
Then, equation (5.10.10) gives

(5.10.20)
so that the retention of the highest derivative (the tenn on the right-hand side in
(5.10.20» requires
(5.10.21)

Y=.[R; y (5.10.22)
and equation (5.10.20) becomes

;i (i)
( dy2 - 'III Y axd + 'IIlx dYd) 'IIIYY - 0
(i) (i)_
or
d (i) (i) (i) (i) (i))_
(5.10.23)
dY 'IIIYYY + 'IIlx 'IIIYY - 'lilY 'IIlxY - 0
from which we have

11I(i)
Y'IYYY
+ 11I(i)
Y'lx Y' I YY Y'IY Y' IxY - f(x)
11I(i) _1IF(i) II' (i) - (5.10.24)
where f(x) is proportional to the pressure gradient impressed on the plate by

the inviscid flow. Note that this implies that the pressure is almost constant
across the boundary layer.
The asymptotic matching between the outer and the inner solutions gives
'111i) (x, 00) =Y'III; (x, 0) +...
from which
(5.10.25)
which simply implies that the tangential velocity of the boundary-layer flow
approaches at the outer edge of the boundary layer, Y ~ 00, the inviscid flow-
velocity 'IIt~) (x,O).
Using the outer inviscid solution, equation (5.10.24) becomes

(i) (i) (i) (i) (i) _ (0) (0)
'IIIYYY + 'IIlx 'IIIYY - 'lilY 'IIlxY - -'Illy (x,O) 'IIlxy (x,O). (5.10.26)
Notice that the boundary-layer equation is parabolic with x acting as a time-

like variable - although the original Navier-Stokes equations were elliptic. This
is in agreement with the result in Section 5.4 that the boundary layers arising on
the subcharacteristics are characterized by a diffusion-like behavior. This means
that the upstream influence is lost so that the first-order boundary-layer solution
on a flat plate is not affected by the trailing edge (if the plate is finite) and the
wake beyond.
For a flat plate, equation (5.10.26) becomes
II/(i)
"I"ITY
+ "I'Ix"l'IYY
II/(i) I,,(i) _ II/(i) I,,(i) -
"I"Y"I"xY-
0
, (5.10.27)
the boundary conditions being

i
Y =0 : lJIi ) =0, V'i1 =0 foe 0 <x <I 0' ~}. (5.10.28)
Y ~ 00 : lJI~) =1
Now, the problem given by (5.10.27), (5.10.28) is invariant under the
transformation
lJIi i ) ~ c lJI~i), X ~ c" x, Y ~ c Y
so that it is possible to look for solutions in which lJI~i), x and y occur in certain
combinations - self-similar solutions. This also reduces equation (5.10.27) to an
ordinary differential equation. Thus, putting
(i) Y
lJI, (x,Y)=-v2x J;(TJ),
~
TJ=- (5.10.29)
-fiX
equation (5.10.27) gives the Blasius equation-
J;"'+ J; J;"=0 (5.10.30)
while the boundary conditions (5.10.28) give

TJ = 0:J; = 0, J;' = O} . (5.10.31)
TJ ~ 00 : J;' ~ 1
Here the primes denote differentiation with respect to TJ. Note that the flow
velocity components are given by
u = J;'(TJ), v = (TJ J;'- J;)
which implies that the velocity profile in the boundary layer is the same for all x
except for the change of scale.
In order to find the asymptotic behavior of the solution of (5.10.30) and
(5.10.31), note first that equation (5.10.30) admits a solution of the form
(5.10.32)
This implies that equation (5.10.30) has the scaling group

];=a-IJ;, Ti=a11· (5.10.33)
We may, therefore, introduce the following canonical coordinates-

, dl'
S =J; 11, 1 =11- _!I_II. (5.10.34)
d11
The transformation from (s,/) to (1,11) is given differentially by
ds d11
-=- (5.10.35)
1+s 11
The transformation rules of the various derivatives are
d J; 2
2 1 dl
-=--I+--(/+S)
3
d11 2
11 11 3 ds '
3 2
d J; 6 5 dl I d 1
-=-I--(/+S)-+---(/+s) 2
+ 1- - (/+S) ( -+1
(dl) dl ) . (5.10.36)
2
d11 3 11
4
11 4 4
ds 11 ds 11 4 ds ds
In terms of the new coordinates (S,/), the boundary-value problem (5.10.30)
and (5.10.31) becomes
2
d ,1 +(/+S) ( -+1
(/+S) 2 - dl ) --5(/+s)-+6/+s
dl dl [ -2/+(/+s)-
dt] =0
ds· ds ds ds ds
(5.10.37)
s=O:(=O }.
(5.1 0.38)
s~oo: t~oo
Near s = 0, equation (5.10.37) shows that

1= As (5.10.39)
with
A. (A. + 1)2 - 5A.(A. + 1) + 6A. = 0 (5.1O.40a)
or
A. =1,2. (5.1O.40b)
A. =1 turns out to be the spurious root. For A. =2, we obtain from equation
(5.10.35),
ds d11
-=- (5.10.41)
3s 11
from which we have
(5.10.42)
Hence, we have from (5.10.34),
11 ~ 0 : 1; == 11 2 • (5.10.43)
Next, near s ~ 00, equation (5.1 0.37) shows that

t =As (5.10.44)
with
-2A+A.(A.+l)=O (5.10.45a)
or
,1,=1. (5.10.45b)
We then obtain from equation (5.10.35),
ds d11
-==- (5.10.46)
2s 11
from which we have
(5.10.47)
Hence, we have from (5.10.34),
11 ~ 00 : f.. == 11 - PI (5.10.48)
as expected! Here, PI is an arbitrary constant.
Figure 5.17 shows the numerical solution of equation (5.10.30) due to

Schlichting (1972) compared with the experimental data for several values of
Reynolds numbers. The agreement between the two shows the validity of the
various approximations and assumptions made in the boundary layer theory.
Besides, the velocity profile is seen to preserve its shape as one moves
downstream despite the fact that the boundary layer thickness is changing.
1.0 +
f'
.08
-U
U
0.6
Ux
0.4 v
'9.5xI0·
.3.0xIO'
+ I.Ix 10'
0.2
11 y( Ulvx )1('
Figure 5.17. Theoretical Blasius profile (Schlichting, 1972) and

experimental confirmation (Dhawan, 1952), plotted by Tritton, 1988.
5.10.3 Flow Due to Displacement Thickness

Using (5.10.12), (5.10.17), (5.10.18), (5.10.21), (5.10.27), (5.10.43), and
(5.10.48), the asymptotic matching gives
O+o(O)(R
2 • )11/(0) 2 ' O)+"'+~+'"
. , . . (x ..[R;
1
=..[R; (i)
VII (x,oo)+ ...
= k [~iJ(k)]+"" as Y~oo
= ./R;- ~2~E f31~+'''' as Y~oo. (5.10.49)
Thus,
5.11 Exercises 207
8~°l(RE) = k } (5.10.50)
y = 0 : ",;0) = -/3, -E
so that ",(0) =0 at y =(1/.JR;) /3, -E, which implies that the presence of a
boundary layer endows a certain thickness to the plate, which then displaces the
outer inviscid flow like a solid parabola of nose radius /3,2/ RE •
For a semi-infinite flat plate, we then have

V'2 ",~O) = 0 (5.10.51)
y =0 : ",~O) =0, x <0
",~O) =-/3, -th, x> 0 (5.10.52)
upstream : ",~O) =o(y)

which corresponds to linearized flow for a body given by y =/3, (~2xj Re ), and
we have
",~O) =-/3, Re [~2(X+iY)]. (5.10.53)
Thus, even though the flow outside the wake and the boundary layer is
essentially irrotational, it is not accurately described by the solution for potential
flow past the given body. One has to take into account the apparent change of
shape of the body caused by the displacement-thickness effect of the boundary
layer.
Note that the foregoing theory is not valid within a distance of order vjU
from the leading edge of the plate where the thickness of the boundary layer is
comparable with the distance from the leading edge.
5.11. Exercises
1. Solve by the method of inner and outer expansions

e y" + xy' - xy =0 }
y(O)=a, y(I)=b .
2. Solve by using the method of inner and outer expansions

£ y" + (I + m) y' + ay =0, 0 < x < I, a > -I}.

yeO) = 0, y(l) = I
3. Solve by using the method of inner and outer expansions
£(y" + ~ y') - y =0).

y(O)=o, y'(I)=O
4. Solve by using the method of inner and outer expansions (Friedrichs, 1942)
£y" + y' = a }
y(O)=O, y(I)=I'
5. Solve using the method of matched asymptotic expansions-
tY" + Y.Y' - xy =0,
y(O)=I, y(I)=-l.
6. Solve by using both the method of inner/outer expansions and the method
of composite expansion
£y" + (2x + I) y' + 2y =O}.
yeO) = a, y(l) = b
7. Solve by using both method of inner/outer expansion and the method of
composite expansion
£y" -(2x + I) y' + 2y =O}.
y(O)=a, y(I)=b
8. Solve by using Langer's method
y" + (02 (sin x) y =0, (0)> I}.
yeO) = y(l) = 0
9. Consider the eigenvalue problem (Holmes, 1995)
~ [p(x):]-[r(x)-A?q(X)]y=o, .1.»1,
yeO) =0, y(l) =0,
where p(x), q(x), and rex) are smooth, positive functions. By putting
(Liouville, 1837),
5.12 Appendix 1 209
show that the above equation becomes

p(x) w" (x) +[ A?q (x) - f(x)] w = °
where
,
f(x) == r(x)+.!.
2
~p(x) [ '\jp(x)
~] .
Use a first-term WKBJ approximation to show that, for large A,
A - An =mf [5o~M
rqw dx]-I
10. Use the JWKB method to find an approximate solution of (Holmes, 1995)
ty"+2y'+2y=0,
y(O) = 0, y(l) = 1.
Compare the result with
* the composite expansion obtained using the method of matched

asymptotic expansions,
* the exact solution.
5.12. Appendix 1
Initial-Value Problem for Partial Differential Equations

The fact that each equation in characteristic form involves a particular linear
combination of the derivatives can be used to gain insight into the structure of
solutions of the equations, such as the correct number of boundary conditions
and the domain of dependence, by considering a construction of the solution at
successive small time increments. In order to see this, let us consider a
hyperbolic partial differential equation in the characteristic form
dx
d:k+!t(X,t,'I')=O on -=ck(x,t,'I'). (ALl)
dt
Consider the initial value problem in x > 0, t > 0, with data prescribed on the
x-axis (which is transverse to the characteristics, i.e., nowhere tangent to them)
at t =0. 3 If P and Qk are two neighboring points on the kth characteristic, then
one obtains from equation (AU)
Vtk (P)-Vtk (Qk)+ it (Qk) [t(P)-t(Qk)]= OJ. (Al.2)

x(P) - X(Qk) =ck(Qk )[t(P) - t(Qk)]
Further, the values at P will depend only on the data between I; and ~ on the
x-axis where PI; and P~ are the two characteristics through P (see Figure
5.18). In other words, I;~ is the domain of dependence of P. Thus, for the fully
initial problem, with Vtk given on t =0, - 00 < x < 00, the solution can be
constructed in t > 0 and it is unique.
Figure 5.18. Characteristics through a point.
Therefore, it is as if the characteristics carry information from the boundaries

into the region concerned. Physically, the characteristics correspond to paths of
waves propagating with the velocities Ck.
3 If data are prescribed on the characteristic, the differential equation does not detennine
the solution at any point not on the characteristic.
5.13 Appendix 2 211
5.13. Appendix 2
Review of Nonlinear Hyperbolic Equations

Consider
u, +uux =0, u(x,O)=q,(x). (A2.l)
The characteristic curves of this equation are given by

dx
C: - =u(x,t). (A2.2)
dt
Observe that u(x,t) is constant on C, because
!!.-
dt
[u(x(t),t)] =u, + dx
dt
ux =u, +uux =0. (A2.3)
Further, each characteristic is a straight line (see Figure 5.19).
Figure 5.19. Characteristic.
The characteristic through (x o, 0) and (x, t) is given by

x-x dx
C: _ _ 0 =- =u(x,t) =u(xo'O) =q,
t -0 dt
or
C : x = X o + q, (x o) t, (A2A)
which gives Xo implicitly as a function of x and t.
Thus, the solution u(x,t) is given in the parametric form-

u(x,/) = q,(xo(X,/)), (A2.5)
where
(A2A)
Note: We have
(A2.(»
where we have from (A2A),
1 = xo, + I q,' (x o) xo, }

(A2.7)
o = xo, + q, (x o) + I q,' (x o) xo, .
Thus, (A2.6) becomes

q,'(x o)
(A2.8)
U
x
= 1 + I q,' (x o)'
which lead to
Ur+UUx=O. (A2.9)
Example A.I: For the linear case

Ur +cux = 0, (A2.IO)
the solution becomes

(A2.lI)
where
x = X o +cI.
Thus,
U= q,(x -C/) (A2.l2)
as expected!
Example A.2: Consider the initial-value problem

Ur + UU x = 0, I > 0 (A2.13)
5.13 Appendix 2 213
O, x < 0,
t=o: u= { (A2.l4)
1, x> 0.
Observe that u(x,O) is an increasing function of x. A continuous solution is an

expansion wave -
o, x S; 0,
u(x,t)= x/t, OS;xS;t, (A2.l5)

{
1, x ~ t,
because
(x/t), + (x/t)(x/t)x = 0, (A2.16)
(see Figure 5.20).
centered
expansion fan
___A _
t
o x
Figure 5.20. Centered expansion fan.
If two characteristics intersect - see Figure 5.21 (this happens, for instance,
when u(x,O) is a decreasing function of x which leads to larger-amplitude
waves traveling faster than the smaller-amplitude ones), the solution u(x,t)
ceases to be single-valued at the point of intersection. This situation has to be
resolved by inserting a jump discontinuity called a shock which is not along a
characteristic. (By contrast, in the linear problem, discontinuities for t > 0 arise
from discontinuities in the initial conditions and occur along characteristics.)
The integral fonn of the partial differential equation dictates that certain
compatibility conditions have to be satisfied across such a discontinuity.
Figure 5.21. Intersection of characteristics.
Example A.3: Consider

u, +uux = 0 (A2.l7)
t = 0 : u =sin x . (A2.l8)
The solution is therefore
u = sinxo (x,t) (A2.l9)
where
x =X o + t . sin x o.
Thus,
u = sin(x - ut) (A2.20)
which shows a steepening of the initial wavefonn because for this problem the
"crest" moves faster than the "trough". The solution eventually becomes triple-
valued which is to be remedied by inserting a "shock" as shown in Figure 5.22.
5.13 Appendix 2 us
.
,,
I
27t
u= 01'---------,+----------::::-
,
7t "
,
I
.
u= 0 I'--------~ro=_::_--------....,.;.;.
u= 0 I"---------~--------...,.-
u= 0 I'--------~ ......- -------r_
OL------------,.;:::",-...:::;;-----.... x
Figure 5.22. Successive solutions of ", + ""x =0, "(x, 0) =sinx at times
t =O,.!., 1, 2 showing many-valuedness when t > 1.
2
In order to discuss discontinuous "shock" solutions, let us write equation

(A2.1) in the conservation form-
L(u) == u, +{A(u)L =0 (A2.21)
where x,teS and A(u)=u 2 J2.

For 8 c 8, let VI be an arbitrarily smooth test function which vanishes

outside 8. Then, we have
If VI
s
L (u) dx dt =O. (A2.22)
Integration by parts leads to
If s
uL'('II) dxdt=O (A2.23)
where
If u is smooth, then equation (A2.21) and (A2.23) are equivalent. However, if

u is not smooth, (A2.23) may remain valid even when (A2.21) does not.
Definition: u is said to be a weak solution of L(u) =0 in 8 if u is piecewise

continous and the condition
JJ uL' ('II) dxdt = 0 (A2.24)

Ii
holds for all test functions VI which vanish outside S.

Let C: g(x,t) =0 be a smooth curve in S where u is discontinuous. Let C
,. ... " "I't"
divide 8 into 8\ and 82 , so 8 = 8 1 U S2 (see Figure 5.23). u is assumed to be
" " ,.. "
smooth in 8 1 and 8 2 so that (A2.22) holds in 81 and 8 2 , Thus,
If u L' (VI) dx dt if
= If
u L' ('II) dx dt + u L' ('II) dx dt =O. (A2.25)
s ~ ~
Note that,
JJ uL' (VI) dxdt =JVI {u+g, + A(u+) gx} ds- JJ VI L(u) dx dt (A2.26)
~ c ~
JJ u L' ('II) dx dt =- JVI {u-g, + A(u-) gx} ds - JJ VI L(u) dx dt (A2.27)

~ c ~
where u± are the values taken by u on C as the limits are taken from the regions
.5. .5
and 2 , respectively, and we have noted that the outward normal n for 1 is .5
the inward normal for S2'
Using (A2.26) and (A2.27), equation (A2.25) gives

5.13 Appendix 2 217
fc lfI {g, [u] + g, [A (u)]} ds = 0 (A2.28)
from which
g, [u]+ g, [A (u)] = 0 (A2.29)
where [q] denotes the jump in q across C.
Figure 5.23. The shock curve.
If U =-g,/gx denotes the speed of propagation of the "shock" C, we have
U[u] = [A(u)]. (A2.30)
Noting that A(u) = u2 /2, this yields

U=u++u- (A2.31)
2
So, the speed of the shock is the average of the values of u ahead and behind the
shock.
Example A.4: Consider the initial-value problem -
u, +uux = 0, t > ° (A2.32)

t, x < 0,
t=O: u= { (A2.33)
0, x>O.
Observe that u(x,O) is a decreasing function ofx. So, the solution u(x,t) is
discontinuous. The discontinuity is resolved by inserting a "shock" given by
x = Vt (see Figure 5.24), where
(A2.34)
shock
x= !'
Figure 5.24. Intersection of characteristics.

Chapter 6
Method of Multiple Scales
6.1. Introduction
In the method of matched asymptotic expansions (Chapter 5), the solution is

constructed in different regions that are then patched together to form a
composite expansion. The method of multiple scales ' , on the other hand, starts
with a generalized version of a composite expansion. This involves separate
coordinates for each region, which are considered to be independent of one
another. Consequently, the given equation is transformed into a partial
differential equation even if it was an ordinary differential equation to begin
with. On the other hand, the method of multiple scales may also be viewed as a
generalization of the method of strained parameters in that the relevant scales
are given implicitly rather than explicitly in terms of the original variables
(Kevorkian, 1966).
6.2. Differential Equations with Constant Coefficients
In order to illustrate the method of multiple scales, consider the equation

x+ x =ex, e« 1 (6.2.1)
which has the exact solution

x(t,e) =acos(.JI=e t + a). (6.2.2)
I This name is a bit awkward because the method of matched asymptotic expansions also
uses multiple scales though each scale is effectively confined to a certain region in the
latter method.
220 Metbod of Multiple Scales
Exp~nding (6.2.2) in powers of e, we have

x(t,e) - acos(t + a) + -!.. eatsin(t + a) +
2
+i e2a[tsin(t+a)- t; cos(t+a)]+ ... (6.2.3)
which is a nonuniform approximation. A better approximation is obtained by

noting that
2
(1-e)'/2 t=(l-± e-i e2) t+o(e t) (6.2.4)
so that (6.2.2) may be approximated by
x(t,e) - acos[(1-±e} + a] +ie2atsin[(1-±e}+a]

+o(e 4 t 2) (6.2.5)
which shows that the problem in question is characterized by two time scales t
and et =1, say. Thus, x(t;e) =x(t,1;e). We may now look for a solution of the
form
x(t,e) - Xo (t,1)+ex, (t,1)+ .... (6.2.6)
(6.2.7)
(6.2.8)
(6.2.9)
etc.
Xo ±
(t,1) = [Aa (1) eit + Aa (1) e- it ] (6.2.10)
where the overhead bar denotes the complex conjugate.

a XI
2
--;-T + XI =-21 eit ( Aa - 21"~) + c.c. (6.2.11)
ot
where c.c. denotes complex conjugate terms.

Removal of the secular terms in equation (6.2.11) requires
Ao -2i~ =0 (6.2.12)
from which
1._
--ll
Ao = Ae 2 (6.2.13)
Using (6.2.13), (6.2.10) becomes
Xo =ACOS(t-± 1+ a)- (6.2.14)
Equation (6.2.11), then, has the solution
XI = ±[A) (1) e il
+ A) (1) e- il ]. (6.2.15)
a x, + X,
2
-,--
ar
1 (A
- =-2 )- A) + -4 A e
2' , 1 I
+t)
- e it + c.c. (6.2.16)

1 -.!. it
AI - 2i A; + - A e 2 = 0 (6.2.17)
4
from which
1 _.!."
A=--iAle
) 8
2
(6.2.18)
Using (6.2.18), (6.2.15) becomes
XI 8" At~ sin ( t - '21-t + a ) .

= 1 (6.2.19)
Using (6.2.14) and (6.2.19), equation (6.2.16) gives the final solution,
X - ACOS(t -±l + a)+ie Ai sin(t - ±i +a)+o(e 3

) (6.2.20)
which is same as (6.2.5)!
Example 1: Consider the initial-value problem (Nayfeh, 1973)

x+x=-2ex, t>O (6.2.21)
222 Method of Multiple Scales
t= °: x = 0, x = 1. (6.2.22)
If we take the solution of this initial-value-problem to be a regular

perturbation expansion of the form -
(6.2.23)
we then obtain
x(t) - sint - etsint + .... (6.2.24)
On the other hand, the exact solution of the initial-value problem (6.2.21) and
(6.2.22) is
(6.2.25)
which may be approximated by
x(t)-= bl-e-
e- sin(t-"!"e t).
fl
2
2
(6.2.26)
Figure 6.1 shows the comparison between the regular perturbation expansion
(6.2.24) and the exact solution (6.2.25). The two curves are reasonably close to
one another for small t, but differ from one another significantly for large t.
2
- Exact Solution 'I II
'I I
- - yo(t)+eyl(t)
I I
§ I I
.-§ 0
I
-0 I
rJ:J
I
-1
I I I I
-2
" "
U U
II
I
0 25 50 75
t -axis
Figure 6.1. Comparison between the regular perturbation expansion in

(6.2.24) and the exact solution given in (6.2.25) when E =2 X 10-1 (from
Holmes, 1995).
Let us now use the method of multiple scales to construct a more accurate
solution than the one given by the regular perturbation expansion (6.2.24). Let
us look for a solution of the form
L emxm(7~,1;,J;, ...)+o(eM)
M-I
x(t;e)- (6.2.27)
m=O
where
(6.2.28)
(6.2.29)
(6.2.30)
(6.2.31)
etc.
We have for equation (6.2.29), the solution
o=Ao(1;,1;) e ; +:40 (1;,1;) e- ;.

i7 i7
X (6.2.32)
fix i +x\ _
--, - -21
.(
Ao +aA-o) eiT, +c.c. (6.2.33)
aT;j a1;
Ao+aAo=O (6.2.34)
a1;
from which,
(6.2.35)
The solution of equation (6.2.33) is then given by

XI = AI (1;,1;) eiT" + AI (1;,1;) e-iTo. (6.2.36)

2
a x-2 + x = - (2'A
- 1 + 2'1 -aAI - a e -T. + 2'1 -aao e-T.)
'eOiT. +c.c. (6.2.37)
aT;} a1; a1;
I
2 I 0
-oA. + A -_ -i ( -a + 2'l -oao ) e -r. . (6.2.38)

oJ; I 2 ° 01;,
Removal of secular terms from equation (6.2.38) (to prevent XI from
becoming O(x o)), in tum, requires
-ao + 2i oao =0 (6.2.39)

en:2
from which
_!-r,
ao =ao e 2 '. (6.2.40)
Using equation (6.2.39), equation (6.2.38) yields

Al =al (1;,) e-r.. (6.2.41)
Using (6.2.32), (6.2.35), (6.2.36), (6.2.40), and (6.2.41), (6.2.27) becomes
x - e-' {a, :(,-~) + ii, / ("~)} + O(e), (6.2.42)
or
x- p -tt (I e t)+O(e)
sin t-- 2
(6.2.43)
I-e- 2
in agreement with the approximate version (6.2.26) of the exact solution!
Example 2: Consider the van der Pol oscillator given by

d u
2
-+u=e
2
(2) du,
l-u - (6.2.44)
dt dt
t =0: u =ao' au
at =0 . (6.2.45)
2. e u (To,Z;,1;,)+O(e
2
n 3
u- n ). (6.2.46)
n=O
We then obtain, from equation (6.2.44), to various orders in e:

zzs
(6.2.47)
(6.2.48)
(6.2.49)
etc.
Solving equation (6.2.47), we have
Uo =A(~,Z;) eiT, +A(~,Z;) e-iT,. (6.2.50)
Using (6.2.50), equation (6.2.48), becomes

2
a u
ar;l +U t =-i (aA
2 a~ -A+A A e' -iA e
2 - ) or, 3 3°r,
0 J 0 +C.c. (6.2.51)
2 aA =A-A 2 A. (6.2.52)
a~
Putting
A =.!. a(T.
p2 T.)e irp(7j.r,) (6.2.53)
2 ,
we obtain from equation (6.2.52),
~ =0, ~ =(1- a:) a. (6.2.54)
On using the initial condition (6.2.45), we obtain from (6.2.54),

<p=<p(Z;) (6.2.55)
4
2
= .
I+C~ -I) e-£I
a (6.2.56)
Using (6.2.52), we obtain from equation (6.2.51),
-B(T.I' T.)
U1 - 2 e
iTo + '8
1 1'A 3e 3iT, + c.c. (6.2.57)
We now consider the application of the method of multiple scales to partial

differential equations.
Example 3: Consider the initial-value problem (Holmes, 1995),

U xx =u/l +eup -oo<x<oo, 1>0 (6.2.58)
I = 0 : U = F (x), u, = O. (6.2.59)
Let us take the solution of (6.2.58) and (6.2.59) to be a regular perturbation

expansion of the form -
U- Uo (X,/) + eU I(X,/) +.... (6.2.60)
We then obtain from the initial-value problem (6.2.58) and (6.2.59),

0(1): uo" -uo" =0, -oo<x<oo, 1>0 (6.2.61)
1=0: Uo =F(x), uo, =0 (6.2.62)
O(e): u1" -u1" =uo,' -oo<x<oo, 1>0 (6.2.63)
I = 0 : u1 = 0, u1, = 0 (6.2.64)
etc.
We have from the initial-value problem (6.2.61) and (6.2.62),
Uo =± [F(g)+F(1J)] (6.2.65)
where
g,1J == x +: I. (6.2.66)
Using (6.2.65), the initial-value problem (6.2.63) and (6.2.64) becomes
4u,. =.!.. [-F'(g) + F'(1J)] (6.2.67)

,. 2
1=0: u1 =0, -u1, +u 1 =0 . (6.2.68)
from which,
(6.2.69)
The appearance of secular terms in (6.2.69) implies that the regular

perturbation expansion (6.2.60) breaks down for this problem. In order to
eliminate the secular terms, let us now use the method of multiple scales and
look for a solution of the form -
u - Uo (;, TI, 1;) + e u, (;, TI, 1;) +... (6.2.70)
where
Tm=emt. (6.2.71)
0(1) : 4uo~. = 0 (6.2.72)
t=O: Uo =F(x), -uo +uo =0 ~ . (6.2.73)
O(e): 4u,
~.
=2(-~+~)
a; dTJ U oI, +(-~+~)
a; dTJ U o (6.2.74)
t=O: u, =0, -u, +u, =-uo (6.2.75)

( " TI
etc.
We have from the initial-value problem (6.2.72) and (6.2.73),
Uo =fa (;,1;) + go (TI, 1;) (6.2.76)
10 (x,1;)+ go (x,1;)= F(x). (6.2.77)
4u, = - : (210 + f)+~ (2go +go). (6.2.78)

~. a~ 11 dTJ "
210 I, +10 = 0 (6.2.79)
(6.2.80)
from which, on using (6.2.77), we obtain
fo(;,1;)=~ F(;) e-T,/2 (6.2.81)
go (17,1;) =.! F(TI) e-T,/2. (6.2.82)

2
U(x,t)-..!.. [F(x-t)+F(x+t)] e- ct/ 2 +O(e). (6.2.83)

2
Example 4: Consider a nonlinear dispersive wave propagation described by

_ 3
Uti -uxx +u-u (6.2.84)
t = 0: U= ecoskx, Ut = ecosinkx. (6.2.85)
Note that the linear problem associated with (6.2.84) and (6.2.85) gives
U= ecos(kx -cot), co 2 = k 2 + 1. (6.2.86)
In order to determine the contributions from the nonlinear terms in equation

(6.2.84), let us consider the solution to be an expansion of the form-
L e u (;,1;,Z;)
~
n
u(x,t;e)- n (6.2.87)
n=1
where
; == kx - cot, Tm == emt. (6.2.88)

O(e) : (co 2 -e) U,~, +u, = 0 (6.2.89)
Tm= 0: u, = coskx, u1 = co sinkx

To
(6.2.90)
Tm = °:u,
- = 0, u, + u, =
-TO 11
° (6.2.92)
etc.
Solving the initial-value problem (6.2.89) and (6.2.90), we obtain
u,=cos[;+tPj(1;,J;)], co 2 =e+1. (6.2.95)
Using (6.2.95), the initial-value problem (6.2.91) and (6.2.92) becomes

(m 2 -e) U2~~ +u2 =2mtP'T, Sin(;+tPl)' (6.2.96)
(6.2.97)

tP, =0
T,
(6.2.98)
so that
(6.2.99)
and equation (6.2.96) then has the solution

(6.2.100)
Using (6.2.95), (6.2.99), and (6.2.100), the initial-value problem (6.2.93) and
(6.2.94) become
(0/ -e) u3". +u3 =-2(W tP,T, -~)

8
cOS(~+tPl) (6.2.101)
Tm = 0 : U •, = 0, u3To = O. (6.2.102)

3
wtP, - -8 = 0
T,
(6.2.103)

3
tP, = - T,. (6.2.104)
8w -
Using (6.2.95), (6.2.100), and (6.2.104), we obtain
U- ecos[ fa -( w- ~:) t]+ O(e 2

) (6.2.105)
showing the amplitude-dependent frequency shift produced by the nonlinear

terms.

(Nayfeh,1973)
(6.2.106)
t=O: u=ecosfa, u, =0. (6.2.107)
Note that the linear problem associated with (6.2.106) and (6.2.107) gives
U =ecos a/cos lex, a2 =k 2 -1. (6.2.108)

Let us detennine here a solution valid for wavenumbers near the cut-off value
= =
k k c 1 (when the linear frequency a vanishes). For this purpose, we
introduce
(6.2.109)
so that the initial-value problem (6.2.106) and (6.2.107) becomes

Uti -
2
k u~~ - U =U 3
(6.2.110)
/ = 0 : U = e cos~, u, = O. (6.2.111)
Let us take the solution of the initial-value problem (6.2.110) and (6.2.111) to
be an expansion of the fonn
u(~,/;e) - L- enun (~,1'o,I;,Z;) (6.2.112)

n=\
k= 1+e 2 K (6.2.113)
where
(6.2.114)
We then obtain, from the initial-value problem (6.2.110) and (6.2.111), to
various orders in e:
O(e):
aa1'/u - aa~2'u - u, =0
2 2
(6.2.115)
T = 0 : u\ =
m
au\o
cos~, a1' =0 (6.2.116)
2): a2u2_ a2u2-u =-2~

0(e a1'o2 a~2 2 a1'o aT; (6.2.117)
T = 0: u, = 0,
m •
au2 + au\ = 0
a1'o aT;
(6.2.118)
2 2 2 2 2
0(e 3): a u3_ a u3-u = u 3+ 2K a u\ _ a u, _2 a u2 _ 2~ (6.2.119)
a1'o2 a~2 3 \ a~2 aT;2 a1'oT; a1'o aT;
T = 0: u = 0
m 3'
au + au
3
aT: au,
aT. + aT.
2 =0 (62120)
..
012
etc.
Solving the initial-value problem (6.2.115) and (6.2.116), we obtain
(6.2.121)
with
a(O,O) =1. (6.2.122)
Using (6.2.121), the initial-value problem (6.2.117), and (6.2.118) becomes

a u, a
2 2
U,
aT;f - a~2· - U2=0 (6.2.123)
Tm =0: u, =0, ~2 + aa COS;=O. (6.2.124)

. uTa ar;
Removal of secular terms in (6.2.124) requires
Tm =0:
aa =0.
ar; (6.2.125)
We have from equation (6.2.123),

u 2 =b(r;,1;) cos;. (6.2.126)
Using (6.2.126), (6.2.124) gives

b(O,O) = O. (6.2.127)

2u3 a2u3
aaTa2 ( 3 3 a 2a ) 1 3
- a~2 -u3= '4 a -2Ka- ar;2 J: J:
cos~+'4 a cos 3~. (6.2.128)
~~~ +(2K-~ a 2
) a=O. (6.2.129)
Using the initial conditions, (6.2.122) and (6.2.125), we obtain, from equation
(6.2.129),
(6.2.130)
where
fJ = 16K -1. (6.2.131)

3
Since 0 is real, the right-hand side in (6.2.130) must be positive, hence 0 2 must
be outside the interval (l,fJ) or (fJ,I), depending on whether fJ>1 or fJ<1.
Since 0(0) =I, 0 increases without bound if fJ I (see Figure 6.2). The case fJ =I corresponds to neutral stability,
which, from (6.2.113) and (6.2.131), is therefore described by
3
k = I +- e 2 +0(e 3 ). (6.2.132)
8
Equation (6.2.132), graphically shown in Figure 6.3 (a situation similar to the
one in Section 3.8.2), indicates that the waves in question grow even at
k = k3 = I, despite the cut-off predicted by the linear theory. On the other hand,
the onset of instability even below the linear stability threshold when the
disturbance has finite amplitude implies that this instability is a subcriticol
instability.
o o
~ '>';'>'>0';4

o
~ >I
Figure 6.2. Unbounded (IJ < 1) and bounded (IJ> 1) solutions
(shaded region is where solution exists).
E
Linear
cutoff
Nonlinear
cutoff
unstable stable
o k
Figure 6.3. Linear and nonlinear cut-offs.

(Bretherton, 1964),
(6.2.133)
The linear problem associated with equation (6.2.133) has the solution
lp =acos6, 6 =kx - OX}
2 4 2 • (6.2.134)
(0 =k -k +1
For the nonlinear problem (e:t; 0), let us look for a solution of the form
(Nayfeh and Hassan, 1971)
lp -lpo (6, X" 1;) + e lp, (6,X, , 1;) +... (6.2.135)
where
(6.2.136)
and we assume,
k = k(X,), (0 = (0(1;).
Noting the following rules of differentiation,

a a a
-=-m-+e-
at ae ar;
a a a
-=k-+e-
dx ae aXI
2 a2 a2 am a , a2
-ata2=m-, -ae 2- 2em - -
aear;
- e- - + e--
ar; ae ar;2
a 2 , a 2 a2 ak a , a2
dx 2 =k- ae2+ 2ek aeax) + e ax) ae + e- aX.2
il 4 a
4 a4 +6ek- - ak-a+3
-=k 4 -+4ek
4
3 0
... (6.2.137)
dx ae ae 3 aXI aX; ae 3
(6.2.139)
etc.
qJo =A(X;,r;) e i8 +.:4(XI ,r;) e- i8 (6.2.140)
where
(6.2.141)
2 4
m2 + k2) --+
a qJ,2 k4 __
a qJ,4 +r/\ _- [2'1m -aA+ 2I'k(2k 2- 1) -aA+ 1
. am- A
(
ae ae "t'l aT.I dx 1 aT.I
+i(6k 2-1) ~ A+3A 2.:4] ei8 +A 3e 3i8

dx,
+c.c.
(6.2.142)
Removal of secular terms in equation (6.2.142) leads to the modulation

equation-
2iw aA+ 2ik (2e-l) :~+i~w A+i(6k 2-1) ak A+3A 2 A=0.(6.2.143)

aT; VAl vr; dx l
Now, from (6.2.141), we have
(J)(J)' =k(2e -1),
and
(J)(J)
" -+w
dk ,2 ak _ (6k2
-- -1)-ak. (6.2.144)
dx, dx l dx l
Further, we have from the compatibility condition
ak + aw =0 (6.2.145)
ar; dx l
the following result -
ak +w' ak = O. (6.2.146)
ar; dx l
Using (6.2.146), we have
aw , ak ,2 ak
-=w -=-w - (6.2.147)
aT; ar; dx l
so that we have, from (6.2.144),
aw + (6k2 -1) ak =
(J)(J)" ak . (6.2.148)
aT; dx l dx l
aA+ 2W ' -+w
2- aA "ak
- A- _ 3i
- A2A- . (6.2.149)
ar; dx l dx l w
Putting
- a ;/3
A --e (6.2.150)
2
aa a(,
- + - wa 2) =0
2
(6.2.151)
ar; dx,
ap , ap 3a 2
-+w - = - . (6.2.152)
aT; dx l 8w
Equation (6.2.151) implies that a quantity proportional to a 2 (which may be

energy in some sense) propagates with the group velocity OJ', as we saw in
Section 4.4.
Example 7: Consider the nonlinear-diffusion problem (Berman, 1978 and

Holmes, 1995)-
u,=euxx+f(u), -OO<X<OO, t>O (6.2.153)
t=O: u=g(x), -oo<x<oo. (6.2.154)
Let 0 ~ u ~ I, and we impose
f(O)=O,I'(O»O}
(6.2.155)
f(I)=O, 1'(1)<0
and
O~g(x)~1 }
(6.2.156)
x~~ g(x) =1, !~ g(x) =0 .
Good choices for f(u) and g(x) satisfying (6.2.155) and (6.2.156) are
f(u)=u(l-u) (6.2.157)
1
g(x)=--" , A>O. (6.2.158)
1+ e x
Let us look for a solution of the initial-value problem (6.2.153) and (6.2.154)
of the form-
u - Uo (x, To, 1;) + eU I (x, To, 1;)+ ... (6.2.159)
where
(6.2.160)

0(1) : UOTo =f(u o) (6.2.161)
Tm =0: Uo =g(x) (6.2.162)
o(e) : u 1TO = I' (uo) u, + Uo - Uo

.u 1j
(6.2.163)
(6.2.164)
etc.
We obtain from the initial-value problem (6.2.161) and (6.2.162),
J-(v)
"0
f
1/2
dv
=To +O(x,1;) (6.2.165)
with
J-.
g(x) dv
O(X,O)= (6.2.166)
f(v)
1/2
Equations (6.2.165) and (6.2.166) may be fonnally expressed as

uo=Uo(s), s=To+O(x,1;). (6.2.167)
Using (6.2.167), and noting the chain rules-

a a a a
aT. = uor,
I
au'
0
ax = uo. au0
(6.2.168)
and hence, the relations -
(6.2.169)
equation (6.2.163) becomes-

u lro - f'(u o) U I = f(u o) [On -Or, + f'(U o) 0;]. (6.2.170)

On -Or, +K 0; =0 (6.2.171)
where
K=f'(O»O. (6.2.172)
Putting
W(x, 1;) =eK8 (X.r,) (6.2.173)
equation (6.2.171) leads to

Wn = Wr,' - 00 < x < 00, 1; > 0 (6.2.174)
from which, we have

(6.2.175)
where we have from (6.2.166) and (6.2.173),

,f.f) dV
K J [(V)
h(x) = W(x,O) = e v' (6.2.176)
For the choice of f(u) given by (6.2.157), we obtain, from (6.2.166) and
(6.2.167),
1
Uo(s)=--_ . (6.2.177)
1+e s
Noting, from (6.2.177) that
(6.2.178)
we obtain from (6.2.166),
8(x,0) = UOI (g) = en -L. (6.2.179)

l-g
For the choice of g(u) given by (6.2.158), (6.2.179) leads to
8(x,0) = -A,x. (6.2.180)
W(x, 1;) = .k Ie-).K(x+2JT,~)-fdg

or
W(x, 1;) =e-AKX+A'K'r,. (6.2.181)
Comparing (6.2.181) with (6.2.173), we have

8(x,1;) =-A, x + A,2 K 1;. (6.2.182)
Using (6.2.182), we have from (6.2.167),

Uo (x, To, 1;) =Uo(To - A, x + A,2 K 1;) (6.2.183)
which represents a travelling wave with velocity

2
V _1+EA, K
A, . (6.2.184)
Observe that the speed of the wave depends on the shape of the initial profile -
the steeper the initial profile (the larger A. is), the slower it moves (see Figure
6.4).
"
1.0
'\ '\
\ \ \,
0.8
S 0.6
.~ \ \
'0
0.4
1\ \ 1\
VJ
" '"
0.2
o ~
-10 -5 o 5 10 15 20
x -axis
Figure 6.4. Comparison of the asymptotic traveling wave solution (6.2.183)

and the numerical solution (dashed curves), of equation (6.2.153) at
t =0, t =5, and t =10. In these calculations, E =10- 2 and A =1, (from
Holmes, 1995).
Example 8: Consider oscillations of a spring swinging in a vertical plane (see

Figure 6.5).
The Lagrangian for this system is given by (Kane and Kahn, 1968)
I
L="2m .' ( £+x )2 0"
[ x"+ "] -mg (£+x )(I-cosO) -"210:".
I , (6.2.185)
The equations of motion are then
x+!.. x+g(l-cosO)-(e+x) Ii =0 (6.2.186)

m
8+~ sinO+-2- x8=0. (6.2.187)
£+x £+x
I
I
r- e x
I mg
Figure 6.5. Spring-pendulum system.
Putting
x _
- = ex O=e6 (6.2.188)
f '
and dropping the tildes, expanding the sine and the cosine functions, equations
(6.2.186) and (6.2.187) become
.. no 2
x+:lo" X = e
(0' 2 -'2I (2) (6.2.189)
ij + 0 = e(x 0 - 2x 0) (6.2.190)
where
2 _ k
WI =-, (6.2.191)
m
Let us look for a solution of equations (6.2.189) and (6.2.190) ofthe fonn -
x(t;e) - Xl (t,tl)+ex z(t,t l)+ .. · (6.2.192)
O(t;e) - 01 (t,tl)+eO z(t,t l )+·.. (6.2.193)
where
(6.2.194)
We then obtain from equations (6.2.189) and (6.2.190),

0(1): XI" +Q2 x, =0 (6.2.195)
6,,, +6, =0 (6.2.196)
, 1"
o()
£ : x,
-"
+Q-x,
•
=-2 XI
'I
--
2
6,- +61-
'
(6.2.197)
(6.2.198)
etc.
We have from equations (6.2.195) and (6.2.196),
x, =A(t,) eiQt +:4(t,) e- iOI (6.2.199)
6, =B(t l) eit + B(t,) e- it • (6.2.200)
Using (6.2.199) and (6.2.200), equations (6.2.197) and (6.2.198) become

n'
x, +;lo"-X, =-2·nA
!;lo" Ie
iQl 3 B-e-
-- "il 1 BB- +C.C.
+- (6.2.201)
-" • I 2 2
62 + 62 = -2 i B I e it + (1 + 20.) A Bei(o+')t + (1- 20.) A Bei(O-')t + C.c. (6.2.202)
" I
The removal of secular terms in equations (6.2.201) and (6.2.202), requires

(6.2.203)
We then have from equations (6.2.201) and (6.2.202),

2
1 - 3 B 2il
X =-- BB-- - -e +C.C. (6.2.204)
2 20. 2 2
2 0. -4
-
62 - (1+20.) AB i(O+I)1 (1-20.) AB- i(O-')1 (6.2.205)
- 0.(0.+2) e - 0.(0.-2) e +C.C.
The solution (6.2.204) and (6.2.205) breaks down when an internal resonance
sets in, i.e., when 0. = 2. This is because the solution (6.2.204) and (6.2.205)
misses new secular terms that arise on the right-hand sides of equations
(6.2.201) and (6.2.202) when 0. =2. Remedying for this omission leads to
inclusion of variations on the amplitudes of the oscillations in x and 6 which
become non-negligible during an internal resonance when a considerable energy
exchange occurs between the x and 6 motion components. In order to treat this
resonance case, let us put
0.-2=£0'. (6.2.206)
Equations (6.2.201) and (6.2.202) then become
x,~ + '""
r\2
x,- -_ - (2 l'r\A
U, 3B e
, + -2
2 -i<1/')
e iOl + NST
.. .+ C.C. (6.2.207)
=-(2iB/ -(1-20) ABe <1l e il

j
(}2" +(}2 l
) + N.s.T.+C.c. (6.2.208)
where N.s. T. denotes non-secular tenns.

The removal of secular tenns in equations (6.2.207) and (6.2.208) requires
2iOA,
I
=-~2 B 2e- <1/, i
(6.2.209)
2 i B, I
=(1- 20) A Be i
<1/,. (6.2.210)
Putting
- - ib
B- - e;/J (6.2.211)
2 '
equations (6.2.209) and (6.2.210) yield
3 2
aI, = - b cosr (6.2.212)
80
b/ =--43 abcosr (6.2.213)
3 - b 2 smr
aI, = - - •
(6.2.214)
8a0 2
3 .
f3I, =--
4 asmr (6.2.215)
where
r=Oa-2!3+(0-2)t. (6.2.216)
Equations (6.2.212) and (6.2.213) give

(6.2.217)
from which
2a 2 + b 2 =constant. (6.2.218)
Equation (6.2.218) may be rewritten as
2
OJ~ a + OJ; b
2
=constant, (6.2.219)
which describes the redistribution of energy between the two natural modes with
frequencies OJ) and OJ 2 of the system during an internal resonance.
6.3. Struble's Method
This method deals with weakly nonlinear wave systems of the fonn
dy +OJoy=e!(d
-2
2
2Y)
y,-,t, e«1 (6.3.1)
dt dt
and consists in looking for a solution of the fonn (Struble, 1962) -
y =a cos (OJot - q» + L en Yn(t) (6.3.2)

n=1
where a and q> are slowly varying functions of t and Yn{t) describe the higher
hannonics.
Example 9: Consider a system of two nonlinearly coupled oscillators

(Shivamoggi and Vanna, 1988)-
d 2x
-2 +OJ~ X = -e2xy (6.3.3)
dt
2
d y, "
-
dr
, +OJ;y=e(y-x").
.
(6.3.4)
This is the constant parameter counterpart of the system considered in Example

6 in Chapter 4.
Let us look for a solution of the fonn -
x(t;e) - AI(t l ) cos( q>1 (t l )) + e u1 (t,t l )+ O( e2 ) (6.3.5)
y(t;e)- A2 (t l ) COS(q>2 (tl))+ev i (t,t l )+0(e 2 ) (6.3.6)
where t l = et is the slow time scale characterizing the slow variations

introduced by the weak coupling among the oscillators, and
(6.3.7)
Note that the solution (6.3.5)-(6.3.7) expresses the fact that the solutions of
equations (6.3.3) and (6.3.4), for e« 1, are very nearly equal to the set of
hannonics represented by the first tenns in the expansions (6.3.5) and (6.3.6),
which they would identically be if e = O. The perturbations induced by the
tenns of O{e) in equations (6.3.3) and (6.3.4) may then be expected to show up
as slow variations in the Ak's and the (Jk'S, and as higher hannonics in the
solution through the Uk'S and Vk's.
Substituting the above expansions (6.3.5) and (6.3.6) in equations (6.3.3) and
(6.3.4), we obtain
(2em,A I (J",) COS(j'1 + (-2em,A", ) sin(j', +e(u,,, +m~u,)+ ...

(6.3.8)
=-e (2A A2COS(j'1 COS(j'2) + ...
I
(2em2A2(J2" ) cos (j'2 + (-2em 2A2" ) sin (j'2 + e(v," + m;v,) +...
(6.3.9)
=e (A; cos2 (j'2 - A~ cos 2 (j',) +....
By equating the coefficients of sin (j'" cos (j'1' sin (j'z, cos (j'z, and the rest to
zero separately, we obtain from equations (6.3.8) and (6.3.9), to O(e):
(JI
'I
=0 (6.3.10)
AI =0 (6.3.11)
'I
u'" +m~ul =-2A,Azcos(j', cos(j'z (6.3.12)
(6.3.13)
A, =0 (6.3.14)
-'I
VI" + m22V, =A22 cos 2 (j'2 - A2,cos 2(j',. (6.3.15)
where the At's and (Jt's are constants to O(e). Thus, in general, the two
oscillators move as if they were effectively uncoupled, and there is no
appreciable energy sharing between them. Note, however, that the solutions
(6.3.16) and (6.3.17) break down when mz =2m" because of small divisors,
which of course, corresponds to an internal resonance in the system. This
difficulty arises because the straightforward perturbation solution given by
(6.3.5)-(6.3.7) does not properly recognize all the potential secular tenns on the
right-hand sides in equations (6.3.3) and (6.3.4), and, in particular, fails to
account for the ability of such tenns to cause variations in the amplitudes of the
oscillators. The latter becomes nonnegligible during an internal resonance when
a considerable energy sharing occurs in the system. Thus, a proper treatment of
the case when an internal resonance prevails would be to sort out all the
potential secular tenns on the right-hand sides in equations (6.3.12) and (6.3.15).
Let us write the O(e) tenns on the right in equations (6.3.12) and (6.3.15), as
follows:
-2A I A2 COSq>1 COSq>2 = -A IA2 cos(q>2 - 2q>,) COSq>1 - A,A2cos(q>2 - 2q>.) cos3q>1
+ AI A2sin (q>2 - 2q>,) sin q>1 + AI A2sin (q>2 - 2q>,) sin 3q>1
2 2
-AI2 cos 2 q>. A A~ cos (q>2 - 2)
=-2-2 l
q>. COSq>2 -2 . (q>2 - 2)'
AI sm q>1 smq>2'
(6.3.1S)
Using these expressions, we obtain now
2eoAl\ = -A.A2 cos(q>2 - 2q») (6.3.19)
-2eo I A",=AIA2sin(q>2 -2q>,) (6.3.20)
U1" + eo~UI =-AI A2 cos (q>2 - 2q>1) cos 3q>, + AI A2sin (q>2 - 2q>,) sin 3q>1 (6.3.21)
2
2eo,A,(J,
• - -',
=__
A
2
1 cos(m,-2m,)
't'. 't'
(6.3.22)
2
-2eo 2A2" =- A2I sin(q>2 - 2q> , ) (6.3.23)
, 1 ( ' - AI·') + -1 A;, cos 2m , .

" + eo;v
. l = -2A;
VI (6.3.24)
· 2 - 't'.
Solving equations (6.3.21) and (6.3.24), we obtain
ul = A
SI A22 [COS(q>2 -2q>1) cos3q>, -sin(q>2 -2q>,)Sin3q>1] (6.3.25)
eo,
(6.3.26)
Observe that the solutions (6.3.25) and (6.3.26) no longer exhibit any small
divisors.
Further, we obtain from the modulation equations (6.3.20) and (6.3.23) for
amplitudes AI and A 2 ,
from which,
1 , ,
- o>,A," + O>,A;
2 . "
=constant. (6.3.28)
Noting that for the present case 0>2 =20» , (6.3.28) leads to
2
0» 2A ) + 0>22A 22 = constant, (6.3.29)
which indicates the exchange of energy between the two oscillators during an
internal resonance (see Figure 6.6).
Figure 6.6. Variation of the individual actions with time during the first-
order internal resonance (due to Ford and Waters 1963).
Example 10: Consider forced oscillations of the Duffing system-

x+ x =&3 + ePa cos At. (6.3.30)
Let us look for a solution of the form-

x(t,e) - A(t) )cos(t - 9(t))) + &) (t,t l ) + ... (6.3.31)
where, t) =et is the slow time scale characterizing the slow variations
introduced by the weak excitation.
On substituting (6.3.31), equation (6.3.30) leads to

(2eAO,,) cos(t-0)+(-2eAJ sin(t-O)
(6.3.32)
+ e(x lu + XI) = eA 3 cos 3(t - 0) + eFo cosAJ + ....
By equating the coefficients of sin(t - 0) and cos(t - 0) and the rest to zero
separately, we obtain, from equation (6.3.32), to O(e):
0,
,
=i8 A 2 (6.3.33)
A/, =0 (6.3.34)
A3
XI +x I = - cos3(t-0)+~cOSAt. (6.3.35)
u 4
If A. ~ 1, equations (6.3.33)-(6.3.35) yield-
3 ,
0= 00 +- A-t l (6.3.36)
8
A = constant (6.3.37)
XI =-(~)
32
cos3(t-0)+(A)
1- A.
COSAJ (6.3.38)
which shows that to O(e), the amplitude of the near harmonic is not affected by
the excitation.
However, if A. -= 1, the solution (6.3.36)-(6.3.38) breaks down. In order to
treat this case, we need to allow the excitation to affect the amplitude of the near
harmonic especially during a resonance. For this purpose, we write
e~cosAt=e~cos[(A.-l) t+O] cos(t-O)
(6.3.39)
- e ~ sin [(A. -1) t + 0] sin(t - 0).
Putting
A. -1 =ea (6.3.40)
and using (6.3.39), we have from equation (6.3.32)
0/, =iA2 + ;~ cos[at +O(t l l )]

(6.3.41)
A" =~ sin [at} +O(t l )] (6.3.42)

A3
XI +X I = - cos3(t-9). (6.3.43)
" 4
Equation (6.3.43) yields
A3
XI = - - cos3(t-9) (6.3.44)
32
which is well behaved at A =1!
Putting
$(tl)=O't l +9(tl), (6.3.45)
equations (6.3.41) and (6.3.42) become
$ =~A2 +0'+ Fo cos$ (6.3.46)

"8 2A
A = Fo sin$. (6.3.47)
" 2
Putting further
a =Acos$, b =Asin$, (6.3.48)
we obtain from equations (6.3.46) and (6.3.47),
at, =-[~(a2+b2)+O']b (6.3.49)
b" = ~ +[~ (a 2+b 2)+O'] a. (6.3.50)
The equilibrium solutions of equations (6.3.49) and (6.3.50) correspond to

b=O (6.3.51)
2 2
~ +[~ (a +b )+O'] a = O. (6.3.52)
Let us write equation (6.3.52), in the form-

(a+,u)(a 2-,ua+v)=O (6.3.53)
where
8 4
v-,u
2
="30', ,uv ="3 Fo > O. (6.3.54)
Since (v- ,u2) should change sign depending on whether O'§ 0, we require
v > O. Further, since ,uv > 0, this implies ,u > O.
The roots of equation (6.3.53) are
6.3 Struble's Metbod 249
(6.3.55)
Thus, we have
4v> Ji! : one real root a =-Ji,
4v < Ji! : three real roots a =-Ji,
(6.3.56)
Therefore, if (J > 0, there is one equilibrium point, and if a is sufficiently

negative, there will be three equilibrium points.
On the other hand, we have for the equations (6.3.49) and (6.3.50), the
following integral -
Faa + a( a 2 + b 2 ) + ~ (a 2 + b 2 )2 = constant = c (6.3.57)

16
_
which shows that all solutions are bounded, since t h e term 3 (a! +b!)!
16
completely dominates all others for large a and b. The integral curves, therefore,
form a family of closed paths centered about the equilibria (see Figures 6.7 and
6.8).
Thus, if A. = 1, for the nonlinear problem, there exist bounded, periodic
responses entrained at the impressed frequency so that the resonance infinities
have been removed by the nonlinearity. This is because the frequency of the
natural oscillation varies with amplitude due to the nonlinearity, so that the
natural oscillation does not remain in step with the excitation. The amplitude-
frequency response diagram (see Figure 6.9) shows that, for a given impressed
frequency, there can be more than one steady-state response. The initial
conditions determine which of the possible responses actually develops. Observe
the possibility of jump phenomena exhibited by the response. Note that the
stability of a given branch is determined by the criterion that IAI increases as Fa
increases and vice versa. Observe the possibility of jump phenomena exhibited
by the response. As A. increases from below 1, IAI increases along AB. At B, IAI
jumps abruptly to the value corresponding to E, afterwards decreasing along EC
as A. increases further.
c>o
Figure 6.7. The solution curves in the ab-plane for the forced Duff'mg
oscillator in the case (1 > O.
Figure 6.8. The solution curves in the ab-plane for the forced Duffmg
osciUator in the case 0' is sufficiently negative.
IAI
-8 -6 -4 -2 o 2 4 6 8 10
2(1- A)
a=--
e
Figure 6.9. The amplitude-frequency response for the forced Duffing
oscillator. (Adapted from Struble, 1962.)
6.4. Differential Equations with Slowly Varying

Coefficients
Consider an equation with slowly varying coefficients,
~{ + p(x;e) : +q(x;e) y =0 (6.4.1)
where
x=et, e«1 (6.4.2)
and
~ ~
p= LEnpn(X), q= LEnqn(X), (6.4.3)

n=O n=O
Let us look for a solution of the fonn

~ ~
y~ LenAn(x) eel +LenBn(x) ee, (6.4.4)

n=O n=O
where
ddx(}j =',.(x-)
I\. i = 1, 2 ,
Ai being the roots of

(6.4.5)
(6.4.6)

(2A\ + Po) ~ +(A; +A1PI +ql) ~ =0 (6.4.7)
(2A2 + Po) B~ +(A; +A2PI +ql) Bo =0 (6.4.8)
from which we have
i = 1,2. (6.4.9)
For the case P = 0, qn = 0 for n ~ 1, i.e., for (Liouville, 1837 and Green,
1837)
d1
dx~ +qli) y =O. (6.4.10)
We therefore have from (6.4.5) and (6.4.9),
Au =±i~qo(X) }
a b (6.4.11)
~=A,Bo=A
where a and b are constants.

Thus,
(6.4.12)
This is the JWKB approximation considered in Section 5.9.
Example 11: Consider a harmonic oscillator with a slowly-varying frequency

(Shivamoggi and Muilenburg, (1991» for which we have
2
d,
- x +Clr'(-)
t x=O (6.4.13)
dr
dx
t = 0 : x = 0, - = a (6.4.14)
dt
where
t =et, e« 1. (6.4.15)
To treat this problem we shall give a systematic procedure which involves a

combination of the method of strained parameters and the method of multiple
scales. This is necessary since the instantaneous frequency of the basic gyration
is indeed not co(i), but rather some average value of co(i) over the basic
gyration (Kuzmak, 1959 and Cole, 1968).
Let us introduce a new scale,
fo
I
i = I(e~) d~. (6.4.16)
Noting the following rules of differentiation,
iJt ta e ai
a =1 (-)
ai+ a }
(6.4.17)
a2 2 a2 a2 a 2 a2
iji2=/ aF +2if aiai + if' ai +e ai 2
the initial-value problem (6.4.13) and (6.4.14) becomes
if' (i) dx co 2 (i)

d 2x 2 a2 x 2 1 (J2 X
~+-(-) -~
dt 1 i dt 1 i
2 +-(-)
2 x+e--~ -~ +e - -
1 atat
- 2 =0
12 ai
(6.4.18)
x(O) =0, 1(0) dx~O) =a. (6.4.19)

dt
x(i;e) - X o(iJ)+&, (iJ)+e 2x 2 (iJ)+". (6.4.20)
(6.4.21)
We then obtain from the initial-value problem (6.4.18) and (6.4.19) to various
orders in e:
0(1):
ilx
a/ +xo =0 (6.4.22)
, =0 : X =0, 0) .---at
t
axo =a (6.4.23)
o
.a2x 0)'(1) axo _ 2 ilx o 2J;(1)

(6.4.24)
ai2 0)2 (I) at - 0)(1) at ai + 0)(1)
1
O(e). +x l - - Xo
t
, =0 : =0, ax.
at =0
XI (6.4.25)
2 • a2x 2 _ O)'(i) ax 2 a2x

o(e ). ai 2 +x2- - 0)2 (I) at - 0)(1) at ai
l I
_[ J;'(I) _ 2J;(1)0)'(1)] ax o + 2J;(1) a2x o

0)2 (I) 0)3 (I) at 0)2 (I) at ai
1 a2x o 2J; (i) [J;2 (i) 2;; (i)]
- 0)2(1) af2 + O)(i) XI - 0)2 (I) - 0)(1) X
o (6.4.26)
t =0 : X 2 =0, ~l =0 (6.4.27)
etc.
Xo (t,i) =Ao (i)sint. (6.4.28)
Using (6.4.28) in equation (6.4.24), the removal of secular terms in the latter
then requires
J; (I) =0 (6.4.29)
Ao(I)O)'(t) 2~(1)
(6.4.30a)
0)2 (I) + 0) (I) =0
or
~ (I) 0)(1) = constant == I (6.4.30b)
and we may then take

(6.4.31)
Using (6.4.28)-(6.4.31) in equation (6.4.26), the removal of secular terms in

the latter then requires
~'(i) - 2w (1) J; (1) Ao (1) =0, (6.4.32)
from which, on using (6.4.30), we have
(6.4.33)
We thus have the following solution:

x(t,e) =Ao (1) sin! +0(e 2 ). (6.4.34)
Differentiating (6.4.34), with respect to t, we obtain

1 xw' t(_)]2 2
_
2
(6.4.35a)
1= w(i) [p+e 2w(i) +x w(t )+0(e )
where
dx
p=-. (6.4.36)
dt
Equation (6.4.35a) may be rewritten as
(6.4.35b)
2 2
which shows that the well-known adiabatic invariant f.l = p2 + w x is the
W
value of I (which is constant) only to the lowest order in e! It is a pity that f.l
cannot look better.
Example 12: Consider a system of two nonlinearly coupled oscillators with

slowly varying parameters, i.e., the linear frequencies wk are slowly varying
functions of time (Kevorkian, 1980, Shivamoggi, 1987)
ql. + w~ ql = -e· 2qlq2 (6.4.37)
(6.4.38)
Here, e is a small parameter characterizing the weakness of the couplings

between the two oscillators.
The parameters (01 and (02 may then evolve in such a way that the system
passes through a state of internal resonance. Let (01 =(01 (il ) and (02 =(02 (il)
where i l =et, and e is the small coupling constant that also characterizes the
slow variations in (01 and (02' Let us introduce two new fast time scales
J
t
il.2 = (01.2 (il) dt (6.4.39)

o
and look for solutions of the form
q, (t;.e) - q,o (i~,~,~,i~)+eql\ (i~,~,i~,i~)+0(e2}}

(6.4.40)
q2 (t,e) - q20 (t p t2,t, ,t2) + eq21 (t p t2,t, ,t2)+ o( e )
where i2 =e 2t is another slow time scale introduced in the system.
Use (6.4.40), and note the following transformation rules-

d a a a ,a
-= (0\ 7 + (0, 7 + e 7 + e- 7 '
dt at l - at 2 uti at 2
2 2 2 2 2
d ,a ,a (a ( )
dt 2 =(OJ ail2 + (0; ai22 + (01(02 ail ai2 + ai2ail
2 2 2
a2 ( ) (a ( )
(
+e(Ol ail ai, + ai, ail + e(02 ai2ai, + ai, ai2
a
+e(O\. 7+e(02. -a. +0 (e-') . (6.4.41)
at,
'I at 2 'I
(In (6.4.41) the mixed differentiation operators are written separately because in
some cases they do not yield the same result when the order is reversed. Since
the present problem turns out not to be one of these cases we will not distinguish
the orders of the mixed derivatives in the following.) One obtains from
equations (6.4.37) and (6.4.38)-
:I' , :I' :I'
0(1) .. ~., +,(0; ~., + 2 (02 ..:0lJL
•• + qlo -- 0 (6.4.42)
at l- (OJ at; (0\ at\ at 2
(6.4.43)
(6.4.45)
etc.
One obtains from equations (6.4.42) and (6.4.43)-
qlO =A) (~J~) cos [~ +(/1) (t~J:)]}.

q20 = A 2 (t pt2 ) cos h + (/12 (tptJ]
(6.4.46)
(6.4.47)
(6.4.48)
The removal of the secularity-inducing terms in equation (6.4.47) requires-

2A,_ A,co,_
__
'I +--z-" =0
COl CO,
or
(6.4.49)
and
fP'il = 0 or fPl = fPl (/2 ) • (6.4.50)
qll = AlA; z cos [(i1+iz)+(fPl +fPJ]

(CO, + CO2 ) - COl
+( A,~; 2 cos [(il -iz)+(fPJ -fPz)]. (6.4.51)

COl - CO 2 - COl
One has to add the homogeneous solution in (6.4.51), but this can be included in
qlO by redefining AI·
One obtains from equation (6.4.48) similarly
2Ao Aocoo
_"_;1 +~=O
co 2 CO;
or
(6.4.52)
and
fPz il
=0 or fPz = (1
fP2 2) (6.4.53)
_ Ai - A~ Ai (' ) A,z /2
q2' - 2 ---z cos2 t2+fPz + z (6.4.54)
2co z 6co z (2co,) - CO;
Equations (6.4.49) and (6.4.52) show that for times t sufficiently far from T
which correspond to the advent of an internal resonance co 2 (et) =2(0, (et), the
two oscillators individually exhibit adiabatic action invariants. However, the
latter have different values in the pre- and post-resonant regimes.
For t == T, which corresponds to (Oz (I,) == 2(0, (II)' the results (6.4.49)-
(6.4.54) break down. In order to treat this internal resonance, let us write
2A2
-- 1A2
- COS (/1' +qJl ) COS (/2' +qJ2 )
COl
=- AI~2 COS [(/2 - 2q + (qJ2 - 2</'1)] COS (II + qJl)

COl
- AI~2 COS [(/2 -2/1)+(qJ2 -2</'1)] cos 3(/1+qJl)

COl
A I A2 sm
+ -2- /2 - 2/'1) + ( </'2 - 2</'1 )] sm
• [(' . (/1
' + qJ, )
COl
A I A2 sm
+-2- . [(' /2-2/1' ) + ( qJ2- 2</'1 ) ]sm3/
. ( ' +</'1 )
1
COl
2
-AI2 cos 2 (/1' +qJl ) --2-2
_ A,2 A. cos [('/2 -2/') (
1 + lfJ 2 -2qJ,
)] cos ('/2 +qJ2 )
- ~2 sin [(/2 -2/1)+(</'2 -2qJl)] sin (/2 +qJJ
(6.4.55)
Using (6.4.55), equations (6.4.47) and (6.4.48) give
I
2A " AI COl" AlA,. [(' ') ( )]
-+-,-=--,-"
CO CO" CO"
SID /, -2/ + qJ, -2qJ
"I " I
(6.4.56)
I I I
-2A I </'1. -_ -
A I,A-
2
cos [('/, - 2/'1) + ( </', - 2qJI )] (6.4.57)
COl "COj " -
(6.4.58)
(6.4.59)
(6.4.60)
:'I' , :'I'
o-q21 (OJ u"q21 2 (01 o-q21
-,-,-+-, -,-,-+ - ' - ,- , +q'l
:l'
at2" (02 at I" (02 atl at2 •
=--,
1 (A;' -
2(02 -
AI-') + --,
1 A;, cos2 ('t, + q>, ).
2(02· .-
(6.4.61)
It might appear that, in equations (6.4.56), (6.4.57), (6.4.59), and (6.4.60), there
is some inconsistency because the A's and q>' s depend on II and 12 whereas the
right-hand sides involve I) and 12 , This can be resolved by noting that the
function (12 - 211) is nearly constant because its derivative ((02 - 2(01) is very
small near an internal resonance, and we assume that (12 - 211) is constant
enough to remove the apparent inconsistency.
On solving equations (6.4.58) and (6.4.61), one obtains
qll = AlA;
8(01
[COS{(12 -211 )+(q>2 -2q>1)} COS3(11 +q>I)
-sin {(12 -21,)+(q>2 -2q>1)} sin3 (II +q>I)),
q21 =
Ai _A2 12 ---2
Ai cos2 ('t 2 +q>2 ).
2(02 6(02 (6.4.62)
Observe that (6.4.62) no longer exhibits any small divisors. Also, one obtains
from equations (6.4.56) and (6.4.59)
(A 12(01)'/. + 2 (Ai (02)''1 =0
or
AI2 (01 + A22 (02 =C t2 .
(- ) (6.4.63)
Ifone introduces the action J k =A; (Ok' then (6.4.63) gives
J I + 2J2 = C(12 ). (6.4.64)
Equation (6.4.64) exhibits the fact that the energies of the two oscillators are
redistributed as the system evolves through the internal resonance.
If the system parameters (01 and (02 are constant, equation (6.4.64) becomes
(on noting that (02 '" 2(01)
JI(OI +J2(02 = C(12) (6.4.65)
as deduced previously in Example 9 in Chapter 6.

6.5. Generalized Multiple-Scale Method
We illustrate a generalized method of multiple scales (Nayfeh, 1964) by the

following example.
Example 13: Consider the following boundary-value problem:

d2 d
e-++a(x).2:.+b(x)y=O, O<x<1 (6.5.1)
dx dx
y(O) =a, y(l) = P (6.5.2)
where a(x) is continuously differentiable and positive in (0, I) .
Let us introduce new independent variables
~=x, 1J=g(x) with g(O)=O, (6.5.3)

e
and note the following rules of differentiation -
d d1J a a
-=--+-
dx dx iJ1] a~ ,
2
d = (d1J
- - )2 - + - a2 2
d 1J- + 2d1J- - - + -a a2 a2 (6.5.4)
dx 2 dx iJ1]2 dx 2 iJ1] dx iJ1] a~ a~ 2 •
Looking for a solution of the form
L enYn(~,1J)+o(eN)
N-I
y- (6.5.5)
n=O
we then obtain, from the boundary-value problem (6.5.1) and (6.5.2) to various
orders in e:
0(1) : [g, ~~O + a(~) ~] g' = ° (6.5.6)
Yo(O,O)=a, Yo(l,g~I))=p, (6.5.7)
O(e) . [g, a YI + a(~) dyl] g' =-2g' a yo _ g" dyo _ a(~) dyo - b(~)y
2 2
. iJ1]2 a1J a~ iJ1] iJ1] a~ 0
(6.5.8)
6.5 Generalized Multiple-Scale Method 263
(6.5.9)
etc.
Let us choose for convenience
f
x
g{x) = a{t) dt. (6.5.10)

o
The boundary-value problem (6.5.6) and (6.5.7) then becomes
iiy"o + dyo =0 (6.5.11)

drt dry
Yo (O,O) = a, Yo{1,oo)=/3. (6.5.12)
Equation (6.5.11) yields

(6.5.13)
Imposing the boundary conditions (6.5.12), we obtain

Ao{O) + Bo{O) = a (6.5.14)
Ao{I) = /3. (6.5.15)

Using (6.5.13)-(6.5.15), the boundary-value problem (6.5.8), and (6.5.9)
becomes
( d2~1 + dyl) =[B~ + (a' ~b) Bo] e-ry _(~ + b~)

d1}- d1} a a- a a-
(6.5.16)
YI{O,O)=O, YI{I,oo)=O. (6.5.17)

a~+bAo=O (6.5.18)
a B~ + (a' - b) Bo =0 (6.5.19)
from which we have
A, (~) = p exp[! :~:~ dl] (6.5.20)
B.(~)= a~) exp[I:~:~ dl] (6.5.21)
where C is an arbitrary constant.

Using these results, the boundary conditions (6.5.14) then yield-
f3. exp [i b(t)

oa(t)
dt]+~:::
a(O)
a. (6.5.22)
Using (6.5.22), (6.5.21) becomes
I a(l)
I b(l) d
I I' b(l) dl
J:.)::: a a(O) - f3a(O) e'
Bo (~
.;;(1)
(6.5.23)
a(x) e.

e y" + y' + Y ::: 0 (6.5.24)
y(O)::: a, y(I)::: b (6.5.25)
which was solved in Section 5.3 by using the method of matched asymptotic
expansions.
Let us introduce new independent variables
x
~::: x, ry:::- (6.5.26)
e
and note the following rules of differentiation -
d
-:::-+--
a 1 a
dx a~ e dry'
d2 a2 2 a2 1 a2
- :2 : : - + - - - +2- - (6.5.27)
dx a~2 e a~ ary e dry2 .
Looking for a solution of the form
y- Yo (~,ry)+ey, (~,ry)+ ... (6.5.28)
we then obtain, from equation (6.5.24), to various orders in e:

2
0(1) . a yo + dyo
. dry2 dry
::: 0 (6.5.29)
O(e) . a YI + dy, ::: -2 a yo _ dyo -y

2 2
(6.5.30)
. dry 2 dry a; dry a; 0
etc.
(6.5.31)
a:
a 2 .::l.,
1
+~ = -(A' +A)+(B' -B) e-1). (6.5.32)
The removal of secular terms in equation (6.5.32) requires

A' +A = 0 (6.5.33)
B'-B=O (6.5.34)
from which we have
(6.5.35)
B = /3e s . (6.5.36)

x
x--
y=ae- x +/3e f +0(£). (6.5.37)
Imposing the boundary conditions (6.5.25), we finally obtain

y= be '- x +(a-be) e X
-
Xlf
+0(£) (6.5.38)
in agreement with the results in Example I in Section 5.3.
6.6. Applications to Solid Mechanics: Dynamic

Buckling of a Thin Elastic Plate
The classical study of buckling of elastic systems (see Section 3.7) is based
on the assumption that the external loads are applied statically. This means that
their magnitude increases with sufficient slowness that any inertial forces may
be neglected. Such an approach is not valid when one adds a periodic load to the
stationary one, and the former has a rapidly varying magnitude. Then the forced
vibrations of the elastic body take the place of the static deflections. The study
of the evolution of the response of an elastic body under time-dependent loading
is the subject of dynamic stability.
The classical linear theory (see Bolotin, 1964), that considers infinitesimal
deflections and establishes the conditions under which a thin plate under the
action of a uniaxial dynamic compressive load can take up equilibrium

configurations other than the one corresponding to a uniform compression,
reveals that the forced vibrations of the elastic plate show a parametric
resonance, so that periodic forces acting in the middle plane of the plate can
excite intense transverse vibrations for certain relationships between the forcing
frequency and the natural frequency of transverse vibrations of the plate. A
distinguishing feature of the parametric resonance (see Section 3.2) is that
whereas the ordinary resonance occurs when the frequency of the exciting force
equals the natural frequency of the system, in the parametric resonance the
exciting frequencies may be multiples of the natural frequency. There is also a
possibility of the occurrence of excited vibrations for the frequencies lower than
the principal resonance frequency. On the other hand, the forced vibrations are
bounded if the frequency and amplitude of the applied load are properly chosen.
In particular, if the applied load P(t) consists of a stationary part and a periodic
part,
P(t) = Po + ~ cosOt,
and if Po were a compressive load greater than the static critical load ~r for the
onset of buckling, the plate in the undeflected configuration is unstable. But, if
~ and 0 are chosen properly, the small motions in the neighborhood of the
undeflected configuration can be stabilized (this is possible even if the time

average of P(t) over a cycle can be much larger than the static critical load for
the onset of buckling).
Here we consider the dynamic buckling of a rectangular thin elastic plate
under a uniaxial harmonically varying load (with a compressive stationary part).
The linear theory is known not to lead to a valid description of the dynamic
post-buckling behavior of the plate because it predicts vibration-amplitudes that
increase unboundedly with time. It is found that experimentally observed
amplitudes do not increase indefinitely with time under these circumstances, but
level off to a stationary value signifying the advent of a supercritical state of
equilibrium. In order to determine whether the linearly unstable vibrations
become stationary, and the magnitude of the stationary amplitudes, one has to
consider the nonlinear problem (Shivamoggi, I977b).
Nonlinear theory of a vibrating plate results by abandoning one or more
assumptions made in developing the linear theory. Two types of nonlinearities
arise:
* nonlinear inertia,
* nonlinear elasticity.
Nonlinear inertia constitutes additional inertia forces which arise during
coplanar displacements U,v; the latter are coupled nonlinearly with the transverse
deflection of the plate, w. Nonlinear elasticity arises in considering the dynamic
response of thin plates loaded beyond their buckling limits so that one needs to
retain in the expressions for strain components, the squares and products of the
deflections and their derivatives. Further, the linear theory involves an
assumption that the middle plane of the plate is free from stress, so that there are
no resultant forces, called the membrane forces, acting in the middle plane of the
plate. However, the latter arise if
(i) there are external loads acting in the middle plane of the plate;
(ii) the plate is bent into a nondevelopable surface that results in strains in the
middle plane of the plate.
Under these circumstances, it is necessary to take into consideration the effect of
bending vibrations of the plate of the stresses acting in the middle plane of the
plate. Thus, in the case of a plate, the elastic nonlinearity is not only geometrical
but also physical in origin. The membrane forces produce an additional resultant
transverse load,
where the membrane forces Nx ' N xy ' Ny are measured per unit length.
We consider only the inertia effect due to the lateral motion of the plate.
Then w will satisfy a nonlinear partial differential equation,
2 2 2 2
V'4 W + ph a
D
w= -!.. (N a w_2N a w + N a w)
iJt2 D x ax 2 Xyaxay y ay2' (6.6.1)
where,
p denotes the mass density, E being the modulus of elasticity, and V the
Poisson's ratio. Equation (6.6.1) in the static case reduces to von Karman's
celebrated nonlinear differential equation for the bending of thin plates
corresponding to large deflections (von Karman, 1910).
Consider the dynamic post-buckling response of a rectangular thin elastic

plate of constant thickness h, simply supported along the edges and subjected to
a compressive distributed load P(t} per unit length at the edges x =O,a. We
confine our attention to the middle plane (taken to be the x,y plane) so that we
may ignore the displacement components U,v. When a uniaxial applied periodic
load
P(t} =Po + r; cosOt
per unit length at edges x =O,a, where Po is stationary, and r; is of small
magnitude, acts in the middle plane of the plate (see Figure 6.10), upon using
(see Timoshenko and Woinowsky-Krieger, 1959 for the derivation of strain
terms),
N.x =1_Eh
--
v 2 -2 -ax
[1 (aw)2 +-21 (aw)2]
ay -(p, +P. cosOt)
V -
0 I ,
N
xy
= Eh
l+v
(aw aw)
axay'
N
y
=~
1- v
[.!.2 (aw)2
2
ay
+.!.2 v (aw)2]
ax '
(6.6.2)
where J..l == h/2, with the boundary conditions
x=O,a: w=O, ;Yw/ax 2 =O}.

a2w/ ay2 =°
(6.6.4)
y = 0, b : w =0,
In deriving equation (6.6.3), we have, in effect, assumed that the forces P(t}
applied at the edges of the plate are transmitted throughout the plate to a
sufficiently close approximation without any change in their magnitude along
the x-direction.
a
o x
(a)
p(t) ~---+-------~ p(t) x
(b)
Figure 6.10. Geometry of the loaded plate: (a) plan, (b) section.
Consider cases wherein the amplitude 1; of the periodic load is much smaller
than the static critical load P"r for the onset of buckling. Seek solutions of the
form
w(x,y,t;e)=e 2,2, fmn(t;e) sin(m~/a) sin(nny/b) (6.6.5)

m=l n=1
where e is a small quantity, which we shall identify below, and then equation
(6.6.3) gives, upon neglecting the nonlinear modal coupling,
(6.6.6)
where
and (Omn is the natural frequency of transverse vibration of the plate. It may be
noted that it is the particular choice of the boundary conditions as in (6.6.4) that
made possible separating the variables as in (6.6.5).
Let the initial conditions be
r = 0; fmn = Amn , dfmn / dr = O. (6.6.7)
One thus has a nonlinear Mathieu's differential equation (6.6.6) so that the
nonlinear problem of the dynamic buckling of a thin elastic plate under a
periodic load P(t} reduces to the investigation of the parametric stability (where
the time-dependent load P(t) appears as a parameter) of the solution of the
nonlinear Mathieu's differential equation. In particular, the issue of stability
requires that all solutions fmn (r) of this equation, for m, n =1,2, ... , and
o$ r < 00, remain bounded when arbitrary initial conditions are imposed.
The general theory of linear differential equations with periodic coefficients
(viz., Floquet theory; see Section 3.13 which becomes useful here since we are
constructing nonlinear solutions in the neighborhood of the linear solution)
shows that the a,e-plane is divided into regions of stability and instability
separated by transition curves. Further, along the transition curves, periodic
solutions of period 21C or 41C exist (as well as linearly increasing solutions
except at the so-called critical points). In the unstable regions the growth takes
place exponentially. The transition curves intersect the e =0 line at the critical
points:
a c =n 2 /4; n=0,1,2, ....
In order to find an approximation to the transition curves for e« 1, one looks

for a (e) such that periodic solutions of period 2n or 4n result. Consider the
cases for which the stationary part Po of the applied load P{t) is slightly greater
than the static critical load ~r for the onset of buckling so that, corresponding to
the neighborhood of the critical point a c = 0, one writes
(6.6.8)
and seeks solutions of the form

(6.6.9)
where
of =er.
Here, r characterizes the time scale of simple harmonic oscillations at the
critical points, and of characterizes the time scale of the effective spring forces
e cos rand e 2cf;'n' that are expected to produce a cumulative effect of long
times on the solution.
0(1): iifm.,no = 0 (6.6.10)

ar"
a2 1'
O{e)'. ~=-2
a2 1'
ar ~+(a -cosr) {'
arai
2 Jmno I
(6.6.11)
(6.6.12)
etc.
In order to have bounded solutions, one has, from equation (6.6.10),
fmno{r, of) = Bo(of). (6.6.13)
Using (6.6.13), in order to have bounded solutions for equation (6.6.11), one
requires
a 1 =0. (6.6.14)

fmnl{ r, of) = B1(of) + Bo(of) cosr. (6.6.15)
Using (6.6.13), (6.6.15), equation (6.6.12) becomes
dh~·2 =2 d sinr-[B, (i)+Bo(i) cosr] cosr

2
d Bo
-~+a2Bo
(_)
r -c Bo [(_)]3
r . (6.6.16)
dr
d
~
2
+( Bo -a 2 +- I) Bo + cBo - 0 .
3 _
(6.6.17)
dr 2
Using equation (6.6.7), one obtains from equation (6.6.17),
( dB
di
o )2 =_!:.
2
(A m.2 _B02)(c_B02) (6.6.18)
where
__ 2 [-a 2 +(1/2)] 2
c- -Am.'
-c
Different cases arise depending on the signs and magnitudes of c and c. If
c<O, since Bo(O) = Am.' B; cannot exceed A;'. if c>A;'., and cannot be
smaller than A;'. if c< A;'.; otherwise, dBo/ di will be imaginary. Thus, if
c> A;'., Bo is bounded and oscillates between Am. and -Am.' and if
c< A;'., Bo is unbounded. The special case c= A;'. separates the stable
oscillations from unstable oscillations, and hence, the transition curve (there is
only one in this case) corresponds to
I ,
a 2 =Z+c A;;'.. (6.6.19)
On the other hand, if c > 0, B; is bounded and oscillates between Am. and c if
c> 0, and oscillates between 0 and Am. if C < O. In all cases, the solution of
equation (6.6.18a) is given by Jacobian elliptic functions.
Using (6.6.19), one obtains from (6.6.8),
a =_e 2 (~+CA;'. )+o(e3 ), (6.6.20)
which determines the amplitudes of a periodic vibration corresponding to a

given load (see Figure 6.11). Note the branching of a nonlinear solution with
respect to the undeflected configuration at
Figure 6.11. Plate vibration amplitude vs.load (from Shivamoggi, 1977b).
R
o
=1_ 2n2
~n
(i) p' 2
I
which signifies that the load is a monotonically increasing function of the

deformation. It is also important to note the backward shift of the branching
point caused by the periodic load r:
which signifies that the state of equilibrium
corresponding to Po '" ~r can always be destabilized by a periodic load
irrespective of its frequency. However, once the branching occurs, the amplitude
A mn of the vibration depends only on the stationary part Po of the applied load
P{t).
6.7. Applications to Fluid Dynamics
6.7.1 The problem of aerodynamicaUy generated sound

As we saw in Section 3.8.1, the problem of sound-wave propagation is one of
singular perturbation type. Here, we consider sound radiation from low Mach-
number flows. The method of multiple scales provided an elegant approach to
this problem by linking the mathematical requirement of uniform validity of the

solution directly to the physical possibility of a sound radiation in the far field
(Shivamoggi, 1981a). (Alternatively, one may use the method of matched
asymptotic expansions, wherein a wavefield is matched to an incompressible
source flow field (Sears, 1969).)
For linearized, unsteady, compressible potential flow, we have
_1_
2a2ep' + 2M~ a2ep' + M 2a2ep' =V',2ep'
c: at'2 c: ax' at' ~ ax,2 (6.7.1)
where the fluid velocity is given by

v' == V"ep' ,
c: being the ambient velocity of sound, M: being the ambient Mach number
(M: == U:/c:, U: being the ambient flow speed in the x-direction). The primes
denote dimensional quantities.
Assuming harmonic time-dependence and putting
'AI x,y,z,t
.... ' ( ', , ')
=I/J , ( x,y,z
, ,
e
') -jm't'
, (6.7.2)

, "
110'
'l'y"y" + I/Jz', z' + 132110' 2 'M 2 (j)
'l'x'x' + I ~"2
110'
'l'x'
(j) -
+"""";2 110'
'I' =0 , (6.7.3)
c~ c~
where
Looking for solutions of the form

M x'(J)'
i
I/J'(x',y',z') ='II'(X',y',z') e ;;'fj , (6.7.4)
where

, , , 2, 0 (6.7.5)
'II x'x' + 'IIy'y' + 'IIz'z' + X 'II = ,
where
(j)'
X'-
- c: 13'
Non-dimensionalizing according to
X' _ y' z'
x=/(", y- b" z=- (6.7.6)
b' ,
m'b'
e=--«I
, f3
C~
'
b' being a length scale characterizing the source field, and seeking solutions of
the form
",(x,y,z, e) ="'0 (x,i,y,ji,z,z) + e 2"'1 (x,i,y,ji,z,z) + o( e4 ), (6.7.7)
where
i =t)', ji =t)', i =ez,
we obtain from equation (6.7.5),
0(1): '"Oxx + "'Off + "'0== = 0 (6.7.8)
"'Ixx + "'Iff + "'1== = -( '"Oii + '"Oyy + "'0:: + "'0)

2
o( e ) :
(6.7.9)
etc.
According to equation (6.7.9), "'I would behave asymptotically as
"'I - _1_
4nr
0 f"'o (x,i,y,ji,z,i) dV(x,y,z),
2
(6.7.10)
v
where V is the volume of the source field, and

2 2 2
2 2 1/2 ~ a a a )
r= ( X +/+Z ) , 0
2 _ (
= ai 2 + aji2 + ai 2 +1,
so that "'I will at least be of the same order as "'0 in the far field. This means
that (6.7.7) will not provide a uniformly valid approximation to the properties of
the flow. In order to preclude this, one requires
[]2",0 = 0 (6.7.11)
or, if
\II
TO ="'oC-ii , i =m't', (6.7.12)
equation (6.7.11) implies
(6.7.13)
where
~, ;;z a2 a2
V "a;;2
=-+ a;2- +ai-2'
Equation (6.7.13) simply implies the existence of a sound radiation in the far
field.
One then has the following program: Vlo can be represented in terms of
transformed incompressible singularities. In order to make Vlo uniformly valid,
one replaces the latter by the corresponding acoustic singularities.
Thus, for a source in motion, one obtains
A..' -
'V _ - I
r'
f'(' t - -r'- +M~X')
--
c: /32 c: /32 '
(6.7.14)
where
r' =[(x' +V't,)2 + rf (y,2 +Z'2)]1/2,

V' being the velocity of the source.
6.7.2 Mathematical formalization of Lighthill's theory of aerodynamicaUy

generated sound
According to Lighthill's (1952) theory, which was proposed to serve as a
model for treating the sound radiation field by relatively small regions of
distributed sources embedded in an infinite homogeneous fluid at rest with
sound speed c~ (and Curle's (1955) extension to incorporate the effect of the
presence of solid boundaries on the sound field), one has
_ 1
---2 - -
a 2
JT;j[Y,t-(~-~/c~)] dV(Y )
p-p~
4Jrc~ ax/Jxj t> ~- yl
+_1_ 3..- Je/~j[y,t-~-~/c~)] dS(y) (6.7.15)
4Jrc~
2
ax; s ~ ~
where p is the mass density of the fluid,

T;j =pv;vj + P;j - c;, Pbjj ,
V; is the fluid velocity, p is the pressure, J1 is the coefficient of viscosity which

is taken to be constant, €; is the direction of cosines of the outward normal to
the solid surface S from the fluid, and the solid surface S is supposed to be either
fixed or moving in its own plane. The physical situation under consideration is a
flow occupying a volume V in the presence of solid boundaries S, and the
subscript 00 denotes the uniform conditions outside v.
Lighthill adopted a force-oscillation view for this aerodynamically generated
sound on the grounds that the latter is so weak that it does not react significantly
upon the flow producing it. Thus, sound is conjectured to be generated in the
quiescent atmosphere in the same manner as that produced in a classical
stationary medium by a continuous distribution of acoustic dipoles of
instantaneous strength I/>;} per unit area over the solid surface S. Further
assumptions in the Lighthill theory are that:
* the influence of viscosity is negligible on the ground that the term

accounting for viscous stresses contains an additional spatial derivative
and this leads to a source of essentially higher order and lower efficiency
than those due to the momentum fluxes;
* the existence of some fictitious sources is then postulated in order to

simulate the phenomena of convection, refraction, and dissipation of the
sound in a real medium;
* if the temperature is nearly uniform in the flow field, then the quantity
;j2 ,
dxdx [(p-c:.p)8;j]
I )
is small, so that the flow-noise generation is assumed to be insensitive to

deviations of the pressure from its adiabatic approximation;
* the speed of sound is constant and the sources of sound are allowed to
move, whereas the medium into which they radiate is assumed to be
stationary.
Given these limitations, it therefore appears that Lighthill's theory is satisfactory
for either low-convection Mach number or low frequencies with wavelengths
much larger than the linear dimension of the region of the distributed sources.
One may then expect to formalize the approximations otherwise made
intuitively in Lighthill's theory by developing a small Mach-number expansion
of the Navier-Stokes equations (Shivamoggi, 1977c).
At low speeds, a valid approach to study the compressibility effects is

through a formal perturbation of the basic equations for incompressible flow,
using the ambient Mach number M~ as the perturbation parameter, as seen in
Section 6.7.1.
The Navier-Stokes equations for a fluid are
ap +v.(pv)=o (6.7.16)
at
:i(pv)+v,(pvv)+Vp=iL [V2V+~ v(v.v)J. (6.7.17)
One has for an ideal gas, the equation of state

p=pRT, (6.7.18)
where 11. is the gas constant.

Consider an isothermal situation for analytical convenience and
nondimensionalize as follows:
v
v=~,
U
(6.7.19)
where the adiabatic exponent r (i.e., p - pY) is taken to be unity accordingly.

However, it may be readily verified that the results given in the following are
valid for any value of r , namely, an adiabatic case.
One then obtains
ap + v.(p v) = 0 (6.7.20)
at
apv
-+V· (pvv )- V [2 v+-I V{V·v) ] =--,
I Vp. (6.7.21)
at 3 M:"
It is a simple matter to derive from equations (6.7.20) and (6.7.21),
v P- M:'
2
[-{
at
1
3
]
a - VV : (pvv) + v2 (V· v) + - v2 (V· v) =0 .
2
(6.7.22)
Consider the cases for which M~ « I, and seek the solutions of the form
where
~ and "l, respectively, characterize the source field and the sound field.
Note that
V=V~+M~Vry. (6.7.24)
Using (6.7.23) in equation (6.7.22), and noting (6.7.24), one obtains

O(M~): V~PI=-V~V~ :vovo (6.7.25)
(6.7.26)
etc.
The general solution of equation (6.7.25) is
_ 1
PI (t "l, t ) - -
J V,V, : vovo(t'''l,t)
I yl
()
dV t
4n v ~-~
- 4~ !PI(t'''l,t) V'(I~~~) .tts(t)

+ 4~ !I~~~ V,Pl (t'''l,t) .dS(t), (6.7.27)
where the solid surface S has been assumed to be stationary and impervious.
Using the divergence theorem and noting that the fluid velocity nonnal to the
surface S is zero, one finds
(6.7.28)
Noting that
(6.7.29)
it is seen that the first term in equation (6.7.25) introduces a monopole

contribution to P2' whereas PI has only a quadruple contribution. To elaborate,
note that in the far field, PI behaves asymptotically as
where
while P2 would behave asymptotically as
so that P2 would dominate PI in the far field. This means that a perturbation
expansion in terms of a small Mach number will not provide a uniformly valid
approximation to the properties of the flow. In order to preclude this, one
requires
(6.7.30)
which physically involves admission of the possibility of sound radiation in the

far field. This further determines the functional form of the solution (6.7.28)
with regard to 1J, and one has
_ 1
PI(~,l1,t)--V~V~.
. f vovoI(t,t -1111)
I dV(t)
4n v ~- t
+_1 V f
PI (t,t- 1lll) .dS(t) (6.7.31)
4n ~ s I~- tl '
or
(6.7.32)
Corresponding to the far field and cases of weak sound-radiation, if one
interprets this as the first tenn in a Taylor series expansion in the retarded time
,
i -IXI/c~ the subsequent tenns in the expansion can be incorporated by writing
this as
p_p~ _ 1 ~ ~ . fVOVO{t,(U2/V~) [i -(IX-tl/c~)]} ()

-~---~2 V}lx'
p~ 4Jl'c~ v
I:.
I""-t
I dV t
1 ~ f(p-p~)/p~{qU2/V~) [1 -(IX-tl/c~)]}
+ 4Jl'c~ V; IX-tl ·dS(t).
(6.7.33)
Incidentally, (6.7.33) shows that:
* the so-called "Reynolds stresses" are the principal generators of
aerodynamic sound;
* upon a comparison with Curle's (1955) result, that the quantity
axaax
2
[( ~ ~'~) 0i j]'
p-c;, P
I J
is small under nearly isothennal situations;

both in confonnity with Lighthill's remarkable conjectures made intuitively.
6.7.3 Nonlinear Shallow Water Waves

There are two types of wave motions on a water surface. Shallow-water
waves arise when the wavelength of the waves is much greater than the depth of
water. Surface waves correspond to disturbances that do not extend far below
the surface. The wavelength is much less than the depth of water.
The features that make an analysis of surface wave motions difficult are
* the presence of nonlinearities;
* the free surface being unknown a priori, besides being variant with time.
In order to make progress with the theory of surface wave motions it is, in
general, necessary to simplify the model by making special hypotheses of one
kind or another which suggest themselves on the basis of general physical
circumstances contemplated in a given class of problems. Thus, two
approximate theories result when
* the amplitude of the surface waves is considered small, or
* the depth of the water is considered small with respect to wavelength.

The first hypothesis leads to a linear theory and to boundary-value problems of
nearly classical type; while the second hypotheses leads to a nonlinear theory for
initial-value problems, which, in the lowest order is of the type corresponding to
sound wave propagation in compressible fluids.
6.7.3(a) Governing Equations

The equilibrium configuration of a liquid in a container of finite size is one of
rest with a plane surface. One may produce a wavemotion on the surface which
is due to the action of gravity that acts in the direction of restoring the
undisturbed state of rest. If the wavemotion is assumed to have started from rest
relative to the undisturbed state of flow, which is itself assumed irrotational,
then, the wavemotions will be irrotational. The wave propagation is taken to
occur along the x-direction, and the gravity acts opposite to the z-direction. We
will assume that the fluid flow is two-dimensional and there is no dependence
on the transverse horizontal coordinate.
The velocity potential ct> defined according to
v=Y'ct>, (6.7.34)
v =(u, v, w) being the fluid velocity, satisfies

a ct> a ct>
2
--+--=0
2
(6.7.35)
ax az
2 2 •
Figure 6.12. Surface waves on water of mean depth ho•
At a rigid stationary boundary, we have the fluid impenetrability condition -
: = O. (6.7.36)
If z =h(x,t) denotes the displacement of the free surface, since a fluid

particle on that surface will remain there, we have the following kinematic
condition expressing the fact that the free surface remains a material surface:
D
z=h: -(z-h)=O (6.7.37)
Dt
or
0tI>
z=h·. -=-+V·Vh
ah
az at . (6.7.38)
The dynamic condition at the free surface is,
z=h: P=T(~J (6.7.39)
where R1 is the radius of curvature of the free surface, counted positive when
the center of curvature is above the surface,
/
R)
2
-.!.- = a ~
ax
[1 + (ahax )2]-3 2 (6.7.40)
T is the surface tension, and the pressure p is given by the Bernoulli integral -
2
p ~ u 2 +w
-=---gz- (6.7.41)
P at 2
p being the density of water.
6.7.3(b) Korteweg-deVries Equation

Nonlinear effects in a train of surface waves lead to many essential observed
phenomena. In deep water, the wave profile becomes distorted, with the crests
slightly sharper than the troughs and the phase speed increasing slightly with
increasing wave slope. In shallow water, the nonlinear effects are stronger and
so more easily observed; consequently, the linearized model here entails much
more stringent conditions on the wave amplitude than that in deep water.
Consider small (but finite)-amplitude long waves of amplitude a and
wavelength ), in shallow water of depth h so that
a h
a:; - « I and f3:; - « l. (6.7.42)
h ),
Two limiting cases arise for such waves depending on whether the parameter
a =~
S :; f32 a),2.IS muc h greater than one (h
t e Stokes approximation
. . )or.IS near
one (the Boussinesq approximation). The first case leads to nonlinear periodic
waves while the second case affords a balance between nonlinearity and
dispersion and leads to solitary waves.
Consider a homogeneous, incompressible fluid whose undisturbed depth
above a rigid horizontal boundary is ho. Let the disturbed free surface be at a
height h(x,t) (see Figure 6.12).
One, then, has the following boundary-value problem for the velocity
potential <1>:
(6.7.43)
(6.7.44)
(6.7.45)
(6.7.46)
Consider propagation of waves moving only to the right, and so introduce

If2
~=-fi (x-cot), r=e 3/2t, Vt=e rf>}
(6.7.47)
co=~gho' e«1
(6.7.48)
(6.7.49)
(6.7.50)
(6.7.51)
Look for solutions of the form -

2
h - ho + eh, + e h2 +...}
(6.7.52)
Vt-eVt,+e 2Vt2+'" '
then equation (6.7.48) leads to
;/
O(e): ;:' =0 (6.7.53)
;r-Vtn + a>-Vtn-I =0
0( "'") .
<:, • iJz2 af , n > 1. (6.7.54)
One obtains, from equations (6.7.53) and (6.7.54), on using the boundary
condition (6.7.49),
etc. (6.7.55)
The boundary conditions (6.7.50) and (6.7.51) are imposed at a boundary

which is unknown without the solution and varies slightly with the perturbation
parameter £. If the solution is analytic in spatial coordinates, the transfer of the
boundary conditions to the basic configuration of the boundary corresponding to
£ = 0 is effected by expanding the solution in a Taylor series about the values at
the basic configuration, as in Section 3.8.2. Thus, (6.7.50) and (6.7.51) give,
z=h . 0(£2). dVt2 =-c dh,

o· . ()z °d;'
2
o( 3). dVt3 +h d Vt2 = dh l _ dh2 + dVtl dh l
£ . {)z I ()z2 ar Co d; d; d;'
etc. (6.7.56)
and
z=ho .• 0(,,2).
". -C
0
dVt,
d; +ghI =0,
0(£3) . .1l-
d ~+.!.
d .1l +gh =0
(d)2
'ar c°d; 2d; 2'
(6.7.57)
etc.
From (6.7.55)-(6.7.57), one derives (Jeffrey and Kakutani, 1972)
3
dh, + 3co h dh l =_coh~ d hi (6.7.58)
ar 2ho I d; 6 d;3
which is the Korteweg-deVries equation.

Thus, in the shallow-water approximation, the elliptic Laplace equation
(6.7.43) is replaced by the nonlinear hyperbolic equation (6.7.58).
6.7.3(c) Solitary Waves

Write the Korteweg-deVries equation (6.7.58) in the form
tPI + KtPtPx + tPxxx =O. (6.7.59)
Looking for steady, progressing, and localized wave solutions of the form
tP(x,t) = tP(;), ;=x-Ut, (6.7.60)

q,~(I(q,-u)+q,~~~ =0,
I~ I~ 00 : q"q,~,q,~~ ~ o. (6.7.61)
Upon integrating, equation (6.7.61) gives
q,~~ = Uq,-~2 q,2 (6.7.62)
and again, equation (6.7.62) gives

.!. q,2 = U q,2 _ ~ q,3 (6.7.63)
2 ~ 2 6
from which
(6.7.64)
The integral in (6.7.64) is of the form
/ = f dq,
q,~I- aq, .
(6.7.65)
Putting
I(
a=- (6.7.66)
-3U'
(6.7.65) becomes
/=-2 f~=ln 1-",. (6.7.67)

1_",2 1+",
Using (6.7.67), (6.7.64) leads to
1-", =e..{i)~ (6.7.68)
1+", '
from which
(6.7.69)
Noting, from (6.7.66), that
(6.7.70)
one obtains, on using (6.7.69),

iP= 1 4e.JU~ 2
1
=- 2 .JU;
3U 2
sech - - = - sech -(x-Ut)
[.JU ] (6.7.71)
a(l+e.JU~) a 2 1( 2
which represents a unidirectional solitary wave. Equation (6.7.71) shows that

* this solitary wave moves with a velocity proportional to the square root of
its amplitude (this is a consequence of the nonlinearity of the wave);
* the width of the solitary wave is inversely proportional to the square root
of its amplitude (this is a consequence of the spreading of the wave due to
dispersion); this result, of course, confirms the condition a - If
discussed before (see (6.7.42»!
Solitary waves are localized waves propagating without change of shape and
velocitl. The essential quality of a solitary wave is the balance between
nonlinearity, which tends to steepen the wavefront in consequence of the
increase of the wavespeed with amplitude, and dispersion, which tends to spread
the wavefront. Solitary waves are found to preserve their identity and to be
stable in processes of mutual collisions. Solitary waves are strictly nonlinear
phenomena with no counterparts in linear theory (see Drazin and Johnson, 1989,
for further details).
6.7.3(d) Stokes Waves

Irrotational, steady, progressive waves were considered first by Stokes and
are called Stokes waves. The Korteweg-deVries equation (6.7.58) can be written
as
ah, + C
at 0
(1 + ~.!i.)
2h
ah, + r a h, = 0
ax
3
ax 3
(6.7.72)
o
where
2 Solitary waves were first observed by 1. Scott Russell on the Edinburgh-

Glasgow canal in 1834. Russell also perfonned laboratory experiments on
solitary waves and empirically deduced that the speed u of the solitary wave is
given by
Let us find the next approximation to the linear periodic wavetrain using the
method of strained parameters. Thus let
~Ol -
2 3
£h!') (0)+ e h!2) (0) + e hp) (0) +. "j (6.7.73)
m = mo(k) + em, (k) + e2m2(k) + ...
where
0== kx - mt.
Using (6.7.73), equation (6.7.72) leads to the following hierarchy of

equations -
I ,,,
O(e): (m-cok) h(') -yeh(') =0 (6.7.74)
O(e 2): (m-cok) h(2)' _yk 3h(2(' =% cokh!l) h(l/ -m1h(l)' (6.7.75)
etc.
Equation (6.7.74) gives the linear result-
h!l) = cosO, mo(k)=cok-ye. (6.7.77)
Using (6.7.77), the removal of secular terms on the right-hand side in
equation (6.7.75) requires
(6.7.78)
h(2) =~ cos20. (6.7.79)

, 8ye
Using (6.7.77)-(6.7.79), the removal of secular terms on the right-hand side
in equation (6.7.76) requires
3c~
m =-- (6.7.80)
2 32yk'
Using (6.7.77)-(6.7.80), (6.7.73) becomes

290
Equation (6.7.81) exhibits two essential effects of nonlinearities -

* periodic solutions of the fonn ei(kx-IDt) may exist, but they are no longer
sinusoidal;
* the amplitude appears in the dispersion relation.
6.7.3(e) Perturbed Solitary Wave Propagation

Consider a perturbed Korteweg-deVries equation describing nonlinear wave
propagation in a slowly-varying medium or a channel of slowly-varying cross
section (Smyth, 1984)
(6.7.82)
where e is a small parameter. The above perturbation adds energy to the wave
in question.
We will analyze the behavior of the perturbed solitary wave solution of
equation (6.4.66) using the principle of energy conservation. The perturbed
solitary wave turns out not to conserve "mass". So, following Johnson (1973)
and Grimshaw (1979), we introduce a "tail" behind the solitary wave to restore
the satisfaction of "mass" conservation. Further, following Knickerbocker and
Newell (1980) and Smyth (1984), the "tail" is divided into two distinct "near"
and "far" tail regions.
Let us assume that the solution of equation (6.7.82) comprises mainly a
solitary wave with slowly-varying parameters given by the expansion-
U - U o (;, r) + £u 1 (;, r) + ... (6.7.83)
where
(
;=x-Jm(r)dt,
o
r being the slow-time scale r = et. This ansatz is valid if the solitary-wave
width is small compared with the scale length of the variations in the medium.
Substituting (6.7.83) into equation (6.7.82), we obtain to 0(1),
(6.7.84)
Equation (6.7.84) has the solitary-wave solution

(0 2 ,fO)
Uo =- sech -~. (6.7.85)
2 2
In order to determine (00 ( r), consider the energy-conservation law for
equation (6.7.82)-
(6.7.86)
Substituting (6.7.85) into (6.7.86), we obtain

, 4
(0=-(0 (6.7.87)
3
from which
(6.7.88)
where (00 is the initial value of (0.
The solitary-wave solution (6.7.85) does not satisfy the "mass" conservation
law-
(6.7.89)
because, substituting (6.7.85), the left-hand side of equation (6.7.89) gives

i e,fO) while the right-hand side gives 2e,fO) .
3
So, one needs to introduce a "tail" behind the solitary wave to remedy this
"mass" defect. Further, it turns out that the expansion (6.7.83) is not uniformly
valid as x ~ - 0 0 , so one needs to introduce a new expansion to describe the
"tail" region. In order to see this, first note that, substituting (6.7.83) into
equation (6.7.82), we obtain to O(e),
(6.7.90)
Let v be a solution of the adjoint equation associated with the left-hand side
of equation (6.7.90), i.e.,
(6.7.91)
Multiplying equation (6.7.90) by v, equation (6.7.91) by u, and adding, we

obtain
-m(vul)+(VUI~~)~ -(V~UI~)~ +(V~~UI)~ = (uo-uor ) v. (6.7.92)
Integrating (6.7.92) with respect to ~,over (-00,00), we obtain
J(uo- UOr ) vd~ = [-m(vu,) + VUI~~ - V~UI~ + V~~UI r (6.7.93)
Now, the solitary wave in question will have no effect on the region ahead of
it so that we have the boundary condition -
(6.7.94)
Further,
~ ~ -00 : u, bounded. (6.7.95)
Equation (6.7.93) has the following bounded solutions-

V= U o (6.7.96)
v =1. (6.7.97)
Using (6.7.94) and (6.7.95), and making the choice (6.7.96), equation
(6.7.93) gives
J(uo -
~
Uo r) Uo d~ =0 (6.7.98)
which is simply the energy-conservation law (6.7.86)!

Next, making the choice (6.7.97), and using (6.7.94), equation (6.7.93) gives
1
u,(-OO)'" C' (6.7.99)
6-ym
Comparison of (6.7.99) with (6.7.85) shows that the expansion (6.7.83) is not
uniformly valid as ~ ~ -00, so one needs to introduce a new expansion to
describe the "tail" region.
As we saw above, the nonuniformity in u, as ~ ~ -00, is O(e) so that the
region behind the solitary wave may be expected to be of O(e) in height and a
slowly-varying function of x and t. Therefore, in the near-tail region, consider an
expansion of the form -
(6.7.100)
where X is the slow space scale X =ex .

Substituting (6.7.100) into equation (6.7.82), we obtain to O(e),
(6.7.101)
from which
VI =f(X) e'. (6.7.102)
The function f(X) is to be determined by matching the near-tail solution
(6.7.100) with the solitary-wave solution (6.7.85). This requires knowledge of
the solitary-wave position X =X s (-r), which is obtained by considering the
solitary-wave speed
dX
_s =co(r) (6.7.103)
dr
from which, on assuming
r=O:Xs=O, (6.7.104)
we obtain,
(6.7.105)
Matching the near-tail solution (6.7.100) with the solitary-wave solution

(6.7.83), and using (6.7.85), (6.7.99), and (6.7.102), we obtain
f(X) e' = 6",co

~,
and using (6.7.88) and (6.7.105), this leads to
2
f(X) = 5/4 • (6.7.106)
3J{i; (1 + 4X)
3coo
The near-tail solution (6.7.100) remedies the "mass" defect associated with
the solitary-wave solution (6.7.85). In order to see this, consider the "mass"
defect
d
tim = dt L L
~
u dx - e
~
u dx
d d x,
_f o dx + e-dr _f VI dX
~
=e -dr U
x,
f o dx - e f VI dX + O(e
~
2
-e U )
and using (6.7.85), (6.7.102), (6.7.105), and (6.7.106), this becomes

tim=0(e 2 ) (6.7.107)
so that the near-tail plus the solitary-wave solution conserve "mass" to O(e).
Next, observe that the near-tail expansion does not remain bounded for large
times and is therefore not valid in this regime, so, one needs to introduce another
new expansion to describe the far-tail region trailing the near-tail region (see
Figure 6.13).
soliton
Figure 6.13. Schematic diagram of the perturbed

Korteweg-deVries solitary wave.
In the far-tail region, consider an expansion of the form

u - e VI (x,t, r) + e 2V 2 (x,t, r)+···. (6.7.108)
Substituting (6.7.108) into equation (6.7.82), we obtain to O(e),
VII + VI xu =0 (6.7.109)
which has the similarity solution -

JAi(s) ds
x/(31)"'
VI = B(r) (6.7.110)
where Ai(x) is the Airy function and B(r) is a constant of integration to be

determined by matching the far-tail solution (6.7.108) with the near-tail solution
(6.7.100) at x =O. Using (6.7.102) and (6.7.106), this leads to
B(r) = ~e'. (6.7.111)

3,,0)0
Figure 6.13 gives a schematic diagram of the perturbed Kortweg-deVries
solitary wave, which shows a decaying oscillatory far-tail.
6.7.3(f) Wave Modulation and Nonlinear SchrOdinger Equation

Let us superimpose a slowly-varying weak modulation on a stationary
weakly nonlinear wave, and study the evolution of such a modulation. To make
the formulation simple, let us assume that, consequent to the superimposition of
the modulation, the wave is still periodic, but with the amplitude and phase
slowly-varying in x and t.
Consider an initially quiescent water subjected to a gravitational field g'
(Figure 6.14). Here the prime denotes dimensional quantities. Let us suppose
that at time t' = 0, a progressive wave is established such that the elevation of
the free surface is raised to y' =1]' where
2rra'
t' = 0 : 1]' = a' (et') e 2trix'/ A' + C.C., e=7«1, (6.7.112)
water
Figure 6.14. Deformed surface of water subjected to gravity.

Equation (6.7.112) represents a sinusoidal fonn with slowly-varying amplitude.

If we nondimensionalize all the physical quantities using a reference length
?L'/21C, and a time (?L'/21Cg'), the wave motions for subsequent times at the
disturbed free surface described by y =17 (x,t;e) are governed by the following
boundary-value problem:
Y< 17: qJxx +qJyy = 0 (6.7.113)
(6.7.114)
(6.7.115)
(6.7.116)
where 'IlqJ denotes the perturbation in the velocity potential of the liquid.
Since the disturbance is assumed to be a progressive wave, let us introduce

the following independent variables -
(6.7.117)
where C is the group velocity of the primary progressive wave, and look for
solutions of the fonn
<P(X,y't;~~ t..£.<p. (S,y", fll. (6.7.118)

17(x,t,e) - L e 17. (~,s, r)
n=l
This leads to a hierarchy of problems of various orders in e:

O(e) : y < 0 : eqJI~~ +qJ1.w = 0 (6.7.119)
y =0 : qJ\y + CO'I1I~ =0 (6.7.120)
-ClXpI~ + 171 = 0 (6.7.121)
y ---? -00 : qJly ---? 0 (6.7.122)
O( e2 ) : y < 0 : k2qJ2~~ + qJ2.w =-2kqJl~( (6.7.123)

2
y = 0 : qJ2y + CO'I12~ = k qJl~ 171~ - C17I ( - qJ 1yy 171 (6.7.124)
-ClXp2~ + 172 =-'2I (k·qJ~~

" + qJ\:r, ) + CqJl( + ClXp,g,171 (6.7.125)
y ~ -00 : ({)2y ~ 0 (6.7.126)
0(£3) : y < 0 : k2({)3~~ + ({)3yy =-2k({)2~' - ({),,, (6.7.127)

2
Y = 0 : ({)3y + 0J1}3~ = k ({)2~ 'fJ1~ + e ({),~ 'fJ2~ +
+ e({)I~Y'fJ,'fJ,~ + k({),~'fJ" + k({),,'fJ,~ - C'fJ2, + 'fJ lt
- ({)2yy'fJ l - 2"1 ({)lyyy'fJl2 - ({)lyy'fJ2

(6.7.128)
-oxp3~ + 'fJ3 =-( k2({)1~({)2~ + e({)'~({)'f,/lt

+({)ly({)2y + ((),y({)lyy'fJl) + C({)2' - k({).~({)" - ({)It
+ C({)I,y'fJ, + oxpl~y'fJ2 + oxp2~y'fJl + 2"1 OXPI~yy'fJ.2 ,

(6.7.129)
y ~ -00 : ({)3y ~ O. (6.7.130)
From (6.7.119) and (6.7.122), we obtain the linear results
'fJ,=A,~s,r)e;~+c.c. }
(6.7.131)
IOJ A (s ) ;~+ky
({). --T
_
t ,r e +c.c.
and
:'==~;=J (6.7.132)
Using (6.7.131) and (6.7.132), (6.7.123)-(6.7.126) give

2
'fJ2 =A, e2i~ + C.c. }
(6.7.133)
({)2 = kAI, ( 1 ); +
20J - ~ e ~ ky + c.c.
.
Using (6.7.131)-(6.7.133), we obtain from (6.7.127), (6.7.128), and (6.7.130):
m = iOJ A
.,..3 2k I" ( / _ r)k ei~+ky
1[A't 2wki AI" . (5'2- 6M k
+"k + +lwk 2 2 )
2
IA, 1 AI ] e i~+ky
+ higher harmonics + C.c. (6.7.134)
Using (6.7.131)-(6.7.134), elimination of the secular terms in (6.7.129) gives

the nonlinear Schrodinger equation (Hasimoto and Ono, 1972)-
2
. --
IA 2k- IAI 12 AI =0.
I A ,,- (6.7.135)
lt I
2
8w w
Let us now construct a localized stationary solution of equation (6.7.135).
Let us first write equation (6.7.135) in the form
-.
2
I -+---+K
. aA , I a AI
ar 2 a(
IA
I
12 A - 0
I
(6.7.136)
Putting
(6.7.137)
.!.2 v" +~2 (2y- u) v' +(s_L)

2
v+1C I\f v =O. (6.7.138)
Here primes denote differentiation with respect to the argument.

Letting
u2 a
s=--- (6.7.139)
2 2'
v" - av + 21(ll3 =O. (6.7.140)
Upon integrating once, equation (6.7.139) gives
(6.7.141)
from which
Jv~adv ,
-I(ll-
=(~-Ur). (6.7.142)
Putting
(6.7.143)
(6.7.142) leads to
-I
.J(i
J-dw-w·,
1-
=(~-Ur), (6.7.144)
from which
.!- In 1- W = j(i (~-Ur). (6.7.145)

2 l+w
If 1( > 0, a> 0, one obtains, from (6.7.145),
v= ~ sechj(i (~- Ur) (6.7.146)
which represents an envelope soliton (Figure 6.15) that propagates unchanged in

shape and with constant velocity. The latter result arises because the nonlinearity
and dispersion exactly balance each other - a result which turns out to be unique
to one-dimensional solutions.
.... .. -- ...
, ,,
,, ,,
'" '"
Figure 6.15. An envelope soliton.
6.7.3(g) Long-Time Evolution of the Modulation

It is found that the long-time evolution of the wave modulation exhibits a
non-ergodic behavior. The numerical solution of the nonlinear Schrodinger
equation (6.7.136) with periodic boundary conditions and with a modulationally
unstable initial condition shows that (Lake et aI., 1977) a state of maximum
modulation is reached by the unstable wave system. After that, the solution
demodulates and eventually returns to an unmodulated state. This process is
repeated in time. Thus, the end state is neither random (no thermalization) nor
steady, but consists of a time-periodic spreading and regrouping of wave energy
initially confined to carrier wavenumber and its linearly unstable harmonics and
sidebands (Figure 6.16). This is due to the fact that the actively participating
modes in this long-time evolution are few and clearly identifiable (with those
which are modulationally unstable,) and these linearly unstable modes

nonlinearly evolve into a superperiodic state.
Figure 6.16. Recurring modulation and demodulation of the wave envelope

(from Lake et aI., 1977).
It turns out (see below) that a finite-amplitude unifonn wavetrain is unstable

to infinitesimal modulations with sufficiently long wavelengths, while it is
stable to modulations with short wavelengths so that a threshold for instability
exists. The long-time behavior of the linearly unstable modulation near this
threshold for instability shows that (Janssen, 1981) the nonlinear effects stabilize
the linearly unstable modulations and produce a periodic motion. For this
purpose, let us put
- -;,'1%1'1 111
A ,-e 'I' (6.7.147)
and first write equation (6.7.136) in the fonn
. o¢J (112 1 1 ) 2
I at +"21 02¢J
O~2 +/( ¢J - ¢Jo ¢J =0, (6.7.148)
where ¢Jo corresponds to the basic wave.
In order to investigate the modulational instability of the wavetrain whose

evolution is governed by equation (6.7.148), one puts
(6.7.149)
so that equation (6.7.148) gives
PI +(paxt =0 (6.7.150)
1,1,1 ( )
a l +- a; +--, p; - - pxx
2 8p· 4p
- /( p - Po =O. (6.7.151)
In order to perfonn the linear stability analysis, one next puts
(6.7.152)
where K is the wavenumber and n is the frequency of the modulation.

Assuming that IpII« IPol,
and keeping only the terms linear in and 0"1' we PI
obtain, from equations (6.7.150) and (6.7.151),
n 2 =.!..K 2 (K 2 -4K'Po)' (6.7.153)

4
Thus, if K' > 0, n 2 is negative for IKI < ~4K'Po .
We will now consider the nonlinear development of the initially linearly
unstable modulation. For this purpose, we consider the initial-value problem for
modulations with wavenumbers near the threshold for instability given by
K
2
=4K'Po'
We look for a solution of the following form (Shivamoggi, 1990):
2
p(x,t) - Po + ePI (x, r) + e P2 (x, r) + ...}
O"(x,t) - eO"\(x, r) + e20"2 (x, r) + ... (6.7.154)
K 2
=4PoK'+ e3 X + ...
where e is a small parameter that characterizes the departure of K 2 from the
linear stability threshold value 4poK', and r =el is a slow time scale
characterizing slow time evolutions near the stability threshold. We have
introduced an explicit detuning parameter X in (6.7.154).
Substituting (6.7.154) into equations (6.7.150) and (6.7.151), we obtain the

following systems of equations to various orders in e:
O( e·) : L (;:J = S. (p"P, .. ·· ,P•.,; G p '" ,G•• ,), n = 1,2,. ... (6.7.155)
where
0
L == 1 a
[ --+Kip
2
4 dx 2 0
and the function Sn depends on the solutions up to O( en-I).

We obtain, from equation (6.7.155), to O(e):
L(;}O' (6.7.156)
the solution for which corresponds to the neutrally-stable case of the linear
problem:
iKx
PI = A(r) e +c.c.}
0'1 = a(r) (6.7.157)
K2 =4po/(
where c.c. means complex conjugate.
Using (6.7.157), we obtain, from equation (6.7.155), to 0(e 2 ):
(6.7.158)
from which,
2
p, =I -da
- - 1A 12 +A- e 2iKx +c.c.}
- /( dr 2po
I dA iKx . (6.7.159)
0',=-,--- e +c.c.
- 4po/( dr
Using (6.7.157) and (6.7.159), we obtain from equation (6.7.155), to 0(e 3 ):

I da dlAI 2
2
- - - -2 + - -
/( dr dr
2
I d A I/(,
{- - , +-xA+-IA/"A + NST. (6.7.160)
4po/( dr 4 2po
-2/CA" ~ da,(1
dr - A- II')} e iKx +c.c.
Removal of the secular terms in the first member of equation (6.7.160)

requires
-I da 2
- - 1A 1 = const. (6.7.161)
/( dr
Let us take the constant above to be zero. Removal of the secular terms in the
second member of equation (6.7.160), then, requires
2
ddrA2 + ( PoKX + 2/(-'I A"I') A =O. (6.7.162)
If we impose the following initial conditions,

• dA
r =0 : A =A, - =0 (6.7.163)
dr
and take A to be real, we obtain, from equation (6.7.162),
(~y = (A A )(A 13)

1(2
2
_
2 2
_ (6.7.164)
where
p=_PoX _A2.
I(
Equation (6.7.164) shows that A is bounded and oscillates between A and Iii
if 13 > 0 and oscilIates between 0 and A if 13 < O. This demonstrates the
nonlinear saturation of the linearly unstable modulation (X < 0) near the linear-
instability threshold.
6.7.3(h) Second-Harmonic Resonance

Suppose a typical surface disturbance is characterized by a sinusoidal
travel1ing wave with amplitude a' and wavelength A', then, let us
nondimensionalize all physical quantities in the following with respect to a
reference length (A'/2tr), a time (A'/2trg,)1/2 where g' denotes the acceleration
due gravity, the primes here denote the dimensional quantities. The potential
function of the motion of the water is taken to be g,I/2 (A' f/2 tP. If y =17 /2tr
denotes the disturbed shape of the surface (whose mean level is given by y =0),
one has the following boundary-value problem:
y < 17 : tPxx + tPyy = 0 (6.7.165)
y = 17 : tPy = 171 + 17xtPx (6.7.166)
tP, +"21 (tP;' +I/J;, ) + 17 = k-17

, ( , )-3/2
xx 1+17; (6.7.167)
y~ -00 : tPy ~ 0, (6.7.168)
where
e =(2tr)2 ~
A' p'g"
T' being the surface tension, p' being the density of water.
Let us look for a travelling wave, and introduce a new independent variable -
~=x-ct (6.7.169)
y < 11 : fP~~ + fPyy =0 (6.7.170)
y= 11 : fPy =(fP~ -c) 11~ (6.7.171)
, )-3/2
11 =cfP~ -"21 (fP~' + fP;, ) + k·11~~
, (
1+ 11~ (6.7.172)
y ~ -<x> : fPy ~ O. (6.7.173)
Seek solutions to (6.7.170)-(6.7.173) of the form-
fP(~,y;e) - L enfPn (~,y)

n=1
11(~;e)- L en11n(~)
n=1
(6.7.174)
L encn(k)
~
c(k;e) -
n=O
L enkn
~
k(e)=
n=O
where
2n
e=a ._« 1. I
A'
Using (6.7.174), one obtains from the boundary-value problem (6.7.170)-
(6.7.173), the following hierarchy of boundary-value problems:
O(e):y<O:fPl~~+fPIYY=O (6.7.175)
y = 0 : fPly = -co11l~ (6.7.176)
11. = ClfPl~ + kg 111~~ (6.7.177)
Y ~ -<x> : fPly ~ 0 (6.7.178)

2
0(e ) : y < 0 : fP2~~ +fP2yy = 0 (6.7.179)
Y = 0 : fP2y + fP,yy111 = -C0112~ + (fPl~ - C1) 111~ (6.7.180)

112 = Co (1P2; + IPly;11,) - ~ (IP~; + 1P1 V) + C11P1; + kg 112;; + 2kokl 11I;;

2
(6.7.181)
y ~ -00 : 1P2y ~ 0 (6.7.182)
o(e 3) : y < 0 : 1P3;; + 1P3yy =0 (6.7.183)
1 2
Y = 0 : 1P3Y + 1P2 yy 111 + IPlyy112 +"2 IPlyyy111
=-C0 113; + (IPI; - CI ) 112; + (1P2; + 1P,;y111 - C2 ) 111; (6.7.184)
113 =Co (1P3; + 1P2;y111 + IP I;y112 + ~ 1P,;yy11; )

+ ci (1P2; + IPI;y111 ) + C21P1; -IPI; (1P 2; + 1P1;y111 ) -IPIY (1P2Y + lP,yy11,)
+ kg 113;; + 2kok, 112;; + 2kok2 111;; - %kg 11 ;;11;;

1
(6.7.185)
y~ -00: 1P3y ~ O. (6.7.186)
Let
11 1 =Acos~. (6.7.187)
Then, from (6.7.175)-(6.7.178), one obtains

IPI (~,y) =AcoeY sin~ (6.7.188)
(6.7.189)
Next, letting
112 = Bcos2~ (6.7.190)
and using (6.7.187)-(6.7.189), one then obtains, from (6.7.179), (6.7.180), and
(6.7.182),
1P2 (~,y) =Co ( B- ~2 e sin 2~ + c i Ai! sin ~ .

2y (6.7.191)
)
Using (6.7.187)-(6.7.191), one finds that the removal of secular terms in

(6.7.181) requires
(6.7.192)
and then
(6.7.193)
The case ko = ±.J1ji, where the above solution breaks down, corresponds to
the second-harmonic resonance (Wilton, 1915 and McGoldrick, 1970), which
we shall treat shortly.
Using (6.7.187)-(6.7.193), one obtains, from (6.7.183), (6.7.184), and
(6.7.186),
(/), (~,y) = A[-c oA2 (

-
c~/4 +~)
1- 2ko8
2 +C 2 ] e sin~
Y
+ higher harmonics. (6.7.194)
Using (6.7.187)-(6.7.194), one finds that the removal of secular terms in

(6.7.185) requires
C = Co ( c~ /2 +.!. _~ kg2 ) A2 k2 = 0 (6.7.195)

2 2 1- 2ko2 2 8 c0 '
which is again not valid for k o = ±.J1ji.
In order to treat the case of second-harmonic resonance corresponding to

k = ±.J1ji, first note that for this case, the fundamental component
11~I) =Acos~ }
(/)?) = AcoeY sin~
and its second harmonic
11~2) = Bcos 2~ }
(/)~2) = Bcoe 2Y sin 2~
have the same linear wave velocity co' so that the two can interact resonantly
with each other. In order to treat this nonlinear resonant interaction, put
111 = A cos~ + Bcos2~ (6.7.196)
(/)1 = Co (Ae Y sin~ + Be 2Y sin2~). (6.7.197)
Using (6.7.196) and (6.7.197), one obtains, from (6.7.179), (6.7.180), and
(6.7.182),
+ (-A 2c O + 2EC,) ~ e 2y sin2c; + higher hannonics. (6.7.198)
Using (6.7.196)-(6.7.198), one finds that the removal of secular tenns in

(6.7.181) requires
-co B + 2coc, - 2kok, =0
, A
(6.7.199)
1
A +4coBc, -8kok l B =0,
2 A A
- - Co (6.7.200)
2
from which
E= kok, + ~-c2e (6.7.201)

2 - 4 0 I
Co
C
,
= k, (3ko2c+ c~ / k~ ) -+~
c
2
A2
-4- c0'
k
2 2
(6.7.202)
o
which show that purely phase-modulated waves are possible for wavenumbers
near the second-hannonic resonant values.
6.8. Applications to Plasma Physics
6.8.1 Nonlinear Longitudinal Waves in a Hot Electron-Plasma

Consider nonlinear longitudinal waves propagating in a hot electron plasma
(Shivamoggi, 1982b). One has the usual governing equations:
an a
-+-(nv)= 0 (6.8.1)
at ax '
av +v av +_l_iJp =_..!- E (6.8.2)
at ax men ax me '
(iii) conservation of energy -
iJp iJp av
-+v -+3p-=0 (6.8.3)
at ax ax '
(iv) Ampere's law-

aE
iii =41l'env, (6.8.4)
where n is the electron number density, v the mass velocity, P the pressure, and
E the electric field. In equation (6.8.3), note that the adiabatic exponent =3, r
since we have considered a one-dimensional motion.
Seek solutions to equations (6.8.1 )-(6.8.4), representing slowly varying
wavetrains of the form
n- no +enl (S, r,O)+e 2 n2 (s,r,O)+0(e 3 )
v - ev. (S, r,O)+e 2v2 (S, r,O)+0(e 3 )
(6.8.5)
2
P - Po +epi (S, r,O)+ e P2 (S, r,O) + 0(e )
3
E - eEl (S, r,O)+ e2 E2 (S, r,O)+ 0(e 3 ),

where
s=ex, r =eI, e« 1,
0= k(S, r)x - w(S, r) t.

Using (6.8.5), one obtains from (6.8.1)-(6.8.4), to O(e),
anI +n avl
at 0 ax =0 (6.8.6)
av l +_I_()P1 +~ E =0 (6.8.7)
at me no ax
me I
()PI +3p avl =0 (6.8.8)

at °ax
aE
-at-4n:enov,
I
=0. (6.8.9)
Upon solving equations (6.8.6)-(6.8.9), one obtains

E, = A(S, r) ei9 + A' (S, r) e- i9
iW
vj = 4-n:en [A(s,r) ei9 -A'(s,r) e- i9 ]
o
(6.8.10)
nl = ~: [A(s,r) ei9 -A'(s,r) e- i9 ]
PI =- ik:~me [A(s,r) ei9 -A'(s,r) e- i9 ]
where the dispersion relation is given by

m2 = mpe2 + k 2 V;2re' (6.8.11)
mpe is the plasma frequency, and Vre is the electron thermal speed-
, == 4trn e-'jme'
m;.. Vie,
=3po/meno'
o
Using (6.8.5), one obtains from (6.8.1)-(6.8.4), to o(e 2 ),
dn 2 + n av2 =-~ (n v ) _ n av l
(6.8.12)
ar °ax ax °d;
dnl _
dr II
av2 +_I_dp2 +.!!.... E =-v avl +~ dpl _ avl __I_dpl (6.8.13)

ar meno ax m 2 ax men~ ax dr meno d;
I
dp2 + 3p av2 =-v dpl _ 3p avl _ dpl _ 3p avl (6.8.14)

ar °ax lax lax ar °d;
dE - 4n:en v 4n:en dEl
----at 2 _
ar . (6.8.15)
o 2 - l VI -
One deduces, from equations (6.8.12)-(6.8.15),

2 2
(ard Vre axd2 +mpe )(dE d Vre axd )(dE
2 2
2 -
2 2
----at2 )_(
- -a;r+ dr -4n:en lv
2
2
I
1
)
d (..:l...
4n:e_
+_ v _vP_1 +3p _av..:l... avI )
I +_VY_I +3p _
me ax J ax I ax ar 0d;
(6.8.16)

+ [2OJ 2 aA 2,.,I.V? aA V? A- ak] -i9
ar + Te ag + OJ Te ag e
(U\.
(6.8.17)
In order to render the assumed form of the solution (6.8.5) uniformly valid,
one has to remove the secular terms in equation (6.8.17). This requires
2
OJ 2 aA
ar + 2OJ kV?Te aA ag =
ag + OJ V?Te Aak o}
_ . (6.8.18)
aA 2 kV? aA V;2 A- ak =0
2 OJ 2
ar + OJ Te ag + OJ Te ag
Using (6.8.11), (6.8.18) becomes
~ [IAI 2 ] + ~ [C(k)IAI
2
] =0, (6.8.19)
where C(k) is the group velocity
C(k) = dOJ = kV;' . (6.8.20)

dk OJ
2
Equation (6.8.19) implies that a quantity proportional to IAI (which may be
energy in some sense) propagates with the group velocity C(k), as we saw in
Example 6.
6.8.2 Ion-Acoustic Solitary Waves in an Inhomogeneous Plasma

Nonlinear waves get modified in an essential way in an inhomogeneous
plasma. For instance, the amplitude, width, and propagation velocity of an ion-
acoustic solitary wave will change as the latter moves in an inhomogeneous
plasma (Nishikawa and Kaw, 1975, Shivamoggi, 1981b, 1988b, Kuehl and
Imen, 1985).
Theoretical treatments describing the properties of an ion-acoustic solitary
wave in an inhomogeneous plasma have been based on the Korteweg-deVries
equation, modified either by variable coefficients or by additional small terms.
These theories give, for example, the spatial dependence of the slowly-varying
soliton-amplitude, width, and speed in an inhomogeneous plasma. The soliton
behavior is determined by carrying out a perturbation expansion based on the
assumption that the soliton width is small compared with the scale length of the
plasma inhomogeneity. Under this condition the soliton retains its identity, and
its amplitude, width and speed are slowly-varying functions of position, hence,
providing the raison d'etre for looking for a JWKB type solution for this
problem.
Consider an ion-acoustic solitary wave propagating in an inhomogeneous
plasma which is otherwise in a time-independent state. Assuming that the ions
are cold and that the electrons follow a Boltzmann distribution, we have for the
ions the following governing equations -
an d
a;+ ax (nv)=v p (6.8.21)

dv dv dt/J v
-+v-=---v - (6.8.22)
at ax ax In'
(iii) Gauss law-
(6.8.23)
where n is the number density of the ions, v is their velocity, t/J is the
electrostatic potential, and VI is the ionization rate per unit volume. n is
normalized by a reference value of the unperturbed number density, v is
normalized by the ion-acoustic speed Cs ' t/J is normalized by kT,,/e, and r' is
normalized by the ion-plasma frequency ro p,' The subscript 0 refers to the
undisturbed state.
Let us introduce two new independent variables -
_
~-e
1f2
[
f Ao(x')
X
dx' ]
-t ,
(6.8.24)
where e is a small parameter characterizing the typical amplitude of a wave,

and seek solutions of the form-
2
n - no +en, +e n2+ }
tP-tPo +etP, + e2 tP2 + . (6.8.25a)
V - Vo + e VI + e V2 + ...
2
and take
(6.8.25b)
Since no and Ao (to be determined below) are to be independent of t, we

have
and~o =0, dAo

d~
=0
.
(6.8.26)
Using (6.8.24)-(6.8.26), equations (6.8.21)-(6.8.23) give to 0(1):
dvo =0 dtPo =0 (6.8.27)

d~ , d~ .
Equations (6.8.26) and (6.8.27) imply that the unperturbed quantities depend
only on the slow-space variable Tf. These are governed by -
d
dTf (novo) =V (6.8.28)
(6.8.29)
(6.8.30)
We have from equation (6.8.30)-
tPo == O. (6.8.31)
Using (6.8.31), equations (6.8.28) and (6.8.29) give-

avo
n -=-v (6.8.32)
°dTf
dn
V
o _ 0 =2v. (6.8.33)
dTf
We have from equations (6.8.32) and (6.8.33)
no = -.!- (V1] + A)2 (6.8.34)

B
B
v = -,------ (6.8.35)
o (v1]+A)"
A and B are constants to be determined using prescribed boundary conditions.
On using (6.8.25), to 0(£), equations (6.8.21)-(6.8.23) give-
dn, 1 d ( ) (6.8.36)
-ag+ A d~ nov, +n,vo =0
o
_dv, +~ dv l +~ dq,1 = 0 (6.8.37)

d~ Ao d~ Ao d~
(6.8.38)
from which, using equations (6.8.26) and (6.8.27), we may derive-
(6.8.39)
This leads to
or
AO = 1 + vO • (6.8.40)
Using (6.8.40), equation (6.8.36) then gives-

(6.8.41)
Next using (6.8.25), to 0(£2), equations (6.8.21 )-(6.8.23) give-
dn, 1 d d
- d~ + A d~ (nOv2+n2vO+nlv,)+ dry (nov,+n,vo)=O (6.8.42)
o
_ dv2 + _1 (v dv2 + V dv') + V dv, + V mo + _1 dq,2 + dq,1

l
= -VA (1- V o)
d~ Ao 0 d~ 1 d~ 0 dry 'd1] Ao d~ dry '1', no
(6.8.43)
(6.8.44)
Using (6.8.26)-(6.8.33), (6.8.38), (6.8.40), and (6.8.41), we may derive from

equations (6.8.42)-(6.8.44):
a¢J1 + ~ ¢JI a¢J, + (_1_ ano ) ¢JI + _~_ a ~. = o.

3
(6.8.45)
dry Ito as 2.A..ono a1J non o as
Putting,
-v J-1-[.3..+{I-'o)]dTl
¢J. =g(1J) 1/11' g(1J)=e 0
2. -'0 '0 (6.8.46)
equation (6.8.45) gives-

al/l.+
- g
, (6.8.47)
dry (1 +vo)"
In order to determine the properties of the solitary wave governed by
equation (6.8.47), consider the equation-
a a a3
~1+a(1J)1/I1 ~'+I3(1J) a~I=O. (6.8.48)
To lowest order in e, equation (6.8.48) has the solution-

2
1/11 '" asech 1p (6.8.49)
where
lp=bO, a= 3ro b=
ak' v4i~)
3i1 (6.8.50)
ao ao
k(1J) = as' ro(1J) = - dry
From
ak aro
-=--=0 (6.8.51)
dry as '
we have
k = constant. (6.8.52)
Equation (6.8.48) has an integral invariant to lowest order in e given by-
(6.8.53)

~(a:)=o
or
(6.8.54)
where the subscript 0 refers to some reference values. Using (6.8.54), (6.8.50)
gives-
(6.8.55)
Returning to equation (6.8.47), the amplitude of the solitary wave is then

given by-
3
g/A~ )1/ (6.8.56)
Vtl_. - ( Ij2 A~ no
Thus, from (6.8.38) and (6.8.46), we have

/
n,_ -nog ( 2/lol '0 no g )1 3
which on using (6.8.34), (6.8.35), (6.8.40), and (6.8.46), becomes

rl~(I_ rlJ
1)
-- J2V;;;; 'In, d ry
"0
n _n ( ~ + -JB)2/3e 3 2B(,+aJ
I~ 0 "1/"0
or
(6.8.57)
On the other hand, the speed c of the solitary wave is given approximately
from (6.8.40) to be
(6.8.58)
Equations (6.8.57) and (6.8.58) show that the amplitude of the solitary wave
increases and the speed decreases as the solitary wave propagates into regions of
increasing density.
6.9. Exercises
1. Solve by using the method of multiple scales,
y" + 2ey' + y = °.
2. Solve by using the method of multiple scales,
y" + e xy' + y = 0, x > 0,
y(O) = 0, y'(O) = 1.
3. Solve using the method of multiple scales,

Y"+.(Oeiy
, 3 14-
= ey + J1 cos IU
and consider the cases (00 '" A and (00 '" A separately.
4. Solve the following initial-value problem using the method of multiple

scales-
Uti - Un + U = U3 ,
t= °:u = ecoskx, u, = esinkx.
5. Use Struble's method to solve Mathieu's equation
ii + ((02 - e cos t) U = °
1 1
for cases (0'" - , (0 '" - , (0 '" 1 separately.
2 2
6. Solve by using the generalized multiple-scale method,
ey" + y' = a,
y(O) =0, y(l) = 1.

ey" +(1 +x) y' +x 2y = 0,
y(O) =0, y(l) = 1.

ey" + xy' +x 2y =0,
y(O) = 0, y(l) = 1.
9. Solve using the generalized method of multiple scales-

ty" - (2x + I) y' + 2y =0,
y(O) =a, y(l) = p.
6.9 Exercises 317
10. Solve using the generalized method of multiple scales-

ey"+(I+ax)y'+C9J=O, a>-I,
y(O) = 0, y(I) = 1.
11. Investigate the evolution of the solitary-wave solution of the perturbed
regularized long-wave equation (Shivamoggi and Rollins, 2002) -
Introduce a "tail", of which the near-tail portion remedies the "mass"

defect and the far-tail portion exhibits a plateau structure.
Chapter 7
Miscellaneous Perturbation
Methods
In addition to the standard perturbation methods discussed in Chapters 2-6, a

variety of other perturbation methods have been developed for several
applications. We will consider here only a couple of these methods since they
have proved to be very useful in some applications.
7.1. A Quantum-Field-Theoretic Perturbative

Procedure
Bender et al. (1989) gave a quantum-field-theoretic (QFT) perturbation

procedure to solve nonlinear differential equations. In this procedure, one
identifies a parameter 8 as a measure of the nonlinearity in the differential
equation in question. So, when 8 =0, the equation becomes linear and can be
solved analytically. As 8 increases smoothly from zero, the nonlinear
development gradually turns on. One assumes that the effect of changing 8
from 8 =0 to 8 ~ 0 is not associated with any sudden nonanalytic effects. The
parameter 8 is now treated as a small parameter and the solution is expanded as
a formal perturbation series in powers of 8, even though the original equation
actually corresponds to 8 = 1.
The validity of this procedure has been demonstrated by applying it to a
number of differential equations of mathematical physics (Bender et aI., 1989;
Shivamoggi and Rollins, 1997 and 1999). The rapidity of convergence of this
procedure has been found to be quite good because the 8 -perturbation series has
a finite (nonzero) radius of convergence.
320 Miscellaneous Perturbation Methods
7.1.1 Blasius Equation

Blasius equation (see equation (5.1.30» is a third-order nonlinear differential
equation that describes the velocity profile of the fluid in the boundary layer
formed by a fluid flowing along a flat plate -
1''' + f" / =0 (7.1.1)
along with the boundary conditions -

1J = 0 : / = 0, I' = 0 (7.1.2a)
1J-7 oo :/==1J. (7.1.2b)
Here, the primes denote differentiation with respect to 1J.

Equation (7.1.1) is now replaced by an equation that contains a parameter 0
/'" + /" /b = o. (7.1.3)

Note that equation (7.1.1) is recovered when 0 =1, and 0 =0 corresponds to
the linear zero-order approximation. By identifying 0 as the perturbation
parameter, the solution/is then expanded in a power series in 0 -
(7.1.4)
This then leads to a set of linear equations for In:

0(1) : 1;"+ 1;'= 0 (7.1.5)
o(0) : 1;"'+1;"= - /;'. in fa (7.1.6)
etc.
Successive integrations of equation (7.1.5) along with use of the boundary
conditions (7.1.2) lead to
(7.1.7)
Equation (7.1.7) is compared with the exact numerical solution of equation

(7.1.1) in Figure 7.1. The agreement is seen to be very good.
..
Next, we obtain from equation (7.1.6),
f
1;"(0)= d1J. e-ry.ln(1J- 1+ e-ry). (7.1.8)
o
7.1 A Quantum-Field-Theoretic Perturbative Procedure 321
4 "T"'""---------------'FT'1
0+-..,.::~-+--+_+---l-1----+-+_+_......-+--+-+--t
o 0.5 1.5 2 2.5 3
11
Figure 7.1. Comparison of zeroth order approximate solution and
numerical solution (bold) for the Blasius equation (from Shivamoggi and
Rollins, 1999).
Numerical integration (Bender et aI., 1989) gives

'/;"(0) =-2 ·1332745. (7.1.9)
Using (7.1.7) and (7.1.9), we have from (7.1.4)-

!"(O) -1-2 ·13327458 +0(8 2 ). (7.1.10)
In order to compare (7.1.10) with the exact numerical result (Schlichting,

1972) -
f" (0) =0 . 46960 ... (7.1.11)
following Bender et al (1989), one converts (7.1.10) to a (0,1) Pade

approximation and evaluates the latter at 8 =1 -
1
1 + 2 ·13327458
I =0 . 31915 (7 1 12)
..
0=1
which differs from (7.1.11) by 32%.
7.2. A Perturbation Method for Linear Stochastic

Differential Equations
Wave propagation in a random medium' is usually described by a stochastic

differential equation with the characteristics of the medium represented by the
stochastic coefficients (Bourret, 1962, Keller, 1964, van Kampen, 1976, and
Shivamoggi et aI., 2000). A random medium is a family of media, each labeled
by one value of a parameter a which ranges over a space A in which a
probability density p(a) determines the probability of a given value of a, and
therefore, represents the source of the waves which, in some cases, may be
random. The objective of the theory of stochastic equations is the determination
of the probability distribution of various statistical properties of the solution u,
such as its expectation or mean value, its variance, and its higher-order
moments.
Consider a linear operator M = M (a) in some space that depends on a
parameter a, which ranges over a set A in which a probability density p(a)
determines the probability of the given parameter a. In this case M(a) is a
I Propagation of waves in a random medium includes a number of applications, such as
propagation of starlight through the turbulent atmosphere, propagation of radio waves

through the ionosphere, and sound wave propagation in the ocean (Tatarskii, 1967,
Chernov, 1969 and Andrews and Phillips, 1998). Although variations in the refractive
index from its mean value in a turbulent medium are very small (on the order of 10-]),
the wave typically propagates through a large number of refractive-index
inhomogeneities and hence, the cumulative effect can be very significant.
stochastic operator. If g is a given element of the space and u is an unknown

element, we may consider the linear stochastic equation for u given by
M(a)u=g, (7.2.1 )
which is a family of equations depending on the parameter a. We assume

equation (7.2.1) has a unique solution u(a) for each a. Moreover, since the
probability density p(a) is known, it determines the probability density of the
solution u(a) of the stochastic equation. We seek to find an equation for the
mean value (u), defined by
J
(u) = u(a)p(a)da. (7.2.2)
Let us assume that M(a) depends on a small parameter e, and that for
e = 0, M(a) reduces to a sure linear operator Lo. Upon expanding M(a;e) in
powers of e, we may rewrite equation (7.2.1) in the form
(Lo + e LI (a)] u (a; e) =g . (7.2.3)
The stochastic linear operator L I (a) represents a stochastic perturbation of the

sure linear operator L o. We shall denote by Uo a particular solution of the sure
equation obtained by setting e =0 in equation (7.2.3), leading to
(7.2.4)
Now, assuming that the inverse operator L~I is defined, we can rewrite
equation (7.2.3) in the form
(7.2.5)
We may solve (7.2.5) for u iteratively as follows:

u - U o - e L~I (Llu o]+ e2 L~I {(L,L~I LI ) uo} +.... (7.2.6)
On taking the expectation value, equation (7.2.6) gives

(u) - Uo - e L~I [(L I )uo] + e2 L~' {(L,L~I LI ) uo} +"', (7.2.7)
Upon operating on (7.2.8) by L o' and using equation (7.2.4), we find that
{L o + e(L J ) + e 2 [(L I ) L~I (L I ) - (LIL~ILI)]}(u) + 0(e 3 ) =g. (7.2.9)

Let G(t,t') be Green's function associated with the unperturbed operator Lo

which is defined by
J
L~' f{t) = G(t,t')f(t')dt'. (7.2.10)
Hence, G(t,t') satisfies
LoG(t,t') =t5 (t - t') }

(7.2.11)
G(t,t') = 0, for t < t' .
That solution of (7.2.11) which corresponds to outgoing waves at large distances
should be chosen.
In order to give an explicit fonn of equation (7.2.9), let us write Lo and L I as
(7.2.12)
Let us take Ao to be independent of t and (AI) =O. The Green's function

G(t,t') associated with the unperturbed operator Lo is then given by
j
eA,,(r-r.), t > t',
G(t,t') = (7.2.13)
0, t < t'.
(~ - Ao) (u{t)} =e 2 f(AI (t)eA,,(r-r') Al (t'))(u(t')) dt' +

o
g, (7.2.14)
which is valid if et « 1.
Noting that
(u{t)) - eA,,(I-r') (u(t')) + o(e 2 ), (7.2.15)
(~ - A") (U(I)} =[e' [(A, (I)e" A, (I - r)} e-" dr] (U(I)} +g. (7.2.16)
If 'fe is the autocorrelation time of AI{t), i.e.,
(AI {t)eA"t AI (t - 'f)) =0, if 'f > 'fe' (7.2.17)
and if t > 'fe' we may write equation (7.2.16) as

I
[; - A"] (u(t)} ~ [e' (A, (t)e" A, (t - r)} e-" dr] (u(t)} + g, (7.2.18)
which is valid if e'fc «I. Combining this restriction with the previous one,
viz., et« I, we have for the region of validity of equation (7.2.18), the interval
(7.2.19)
7.2.1 Application to Wave Propagation in a Random Medium

Consider a linear wave propagating in a turbulent atmosphere, which has
small random variations in the refractive index. The electric field of such a
monochromatic wave satisfies
(7.2.20)
where k o is the wavenumber of the wave and the refractive index of the medium
n is given by
n2 =I+e,u(z). (7.2.21)
,u (z) describes the small random variations.
Let us introduce the quantities
j; k F= dE (7.2.22)
~ = oz, d~ ,
so that equation (7.2.20) can be written as the system of first-order equations
(7.2.23)
Identifying equation (7.2.23) with equations (7.2.3) and (7.2.12), and applying
equation (7.2.18), we obtain
(7.2.24)
Noting that
exp
0
[( -1
1) ]
T] =(COST] SinT])
,
0 - sin T] cos T]
equation (7.2.24) can be expressed as
d [(E)] [( 0 1) £2 (0 0)] [(E)] (7.2.25)

d~ (F) = -1 0 +2 c\ -C
2
(F)'
where
~
C1 == J(JJ(~/ko) JJ [(~-T])jko]) sin2T] dT]

o
(7.2.26)
J
~
C2 == (JJ (~/ko) JJ [( ~ - T])jko]) (1- cos2T]) dT]

o
If the wave propagation in a random medium is assumed to be stationary,

Gaussian and a Markov process, then by Doob's Theorem (Doob, 1942) it is an
Uhlenbeck-Ornstein process, so that
(7.2.27)
then (7.2.26) gives

C 2e
= _ _0_ C = _4k
2
_0_ (7.2.28)
I 1+4e' 2 1 +4e .
o o
If, on the other hand, we take the commonly used Gaussian form
(JJ(~/ko) JJ [(~_T])jko])=e-~2/2kg, (7.2.29)
then (7.2.26) gives
(7.2.30)
From equation (7.2.25), it follows that the mean electric field (E) is
governed by
(7.2.31)
If the mean electric field (E) is assumed to vary in space, according to
(7.2.32)

e2) +1-
.e
(
2 2 2k2
0 3
k =ko 1- - c2 +O(e ). (7.2.33)
2 C1
2
Taking the square root of(7.2.33) and substituting into (7.2.32), we obtain
2 2
(E}="2I ~exp { e
iko[1-"4 C 1] e c 2koz } +c.c.,
z-"4 (7.2.34)
where c.c. denotes the complex conjugate of the preceding term.

Equation (7.2.34) shows the attenuation of the coherent wave due to random
inhomogeneities in the medium. (This damping is due to phase interference
among the individual solutions for different a, like the Landau damping of
plasma oscillations (Landau, 1946), and has nothing to do with actual
dissipation; thus, in this process, the medium does not gain energy.)
7.2.2 Renormalization Procedure

Application of perturbative procedures to random differential equations can
sometimes run into qualitative problems. In order to see this, consider the
following problem (Kraichnan, 1961 and Leslie, 1973):
~:to+::i:(tl] Ultl=Oltl } (7.2.35)
where J1 is a determinate constant, and bet) is a real, centered stationary

Gaussian random function of t having the auto-correlation function
(b(t') b(t")) = (J2 e-J.(I'-I
H
)
(7.2.36)
befitting an Uhlenbeck-Omstein process, but otherwise unspecified.
7.2.2(a) Exact Solution

The initial-value problem (7.2.35) has the exact solution
j
-pH b(t')dl'
u(t)=e 0 (7.2.37)
Jb (t') dt', being a linear functional of the centered Gaussian

I
Noting that
o
random function b(t), is a centered Gaussian random variable, we have
J "
-1l1-:; jf(b(/')b(/")dl'dt"
(u(t)) = e - 00 .2 (7.2.38)
Using (7.2.36), (7.2.38) becomes
(7.2.39)
For long-range correlations (i.e., small A.) when the fluctuations are almost
time-independent, (7.2.39) yields
I '2
IlI
(u(t)) '" e- - 2 (J'/ (7.2.40)
On the other hand, for short-range correlations (i.e., large A.) when the
fluctuations differ very little from pure white noise, (7.2.40) yields
(7.2.41)
In this limit, the effect of randomness on the mean amplitude is only to cause a
frequency shift.
7.2.2(b) Perturbative Solution

If the fluctuations b(t) are taken to be very small, one may apply the
perturbative procedure, namely, the result (7.2.9) to the initial-value problem
(7.2.35), which leads to (Shivamoggi et a\., 1999)
2 This is based on the result (see van Kampen, 1992)

[ (x)+~ ((x'})+~ ((X'})+H]
(eX) =e 2. :l.
where,
((X")) E (x- (x>r).
3 In canying out this integration, use has been made of the result
]Jp(t'-t)dtdt'=2 J(I-":")p(r)dr.
00 0 T
( ~ +J1)(u(t)) + J(b(t)b(t')) e-/l(I-I')(U(t')) dt'==O(t)}.

l
o (7.2.42)
(u(O)) == I
Putting
(u(t)) == e-/l l v(t) (7.2.43)
and using (7.2.36), the initial-value problem (7.2.42) becomes
dt
'J e
-dv + 0'"
I
0
-'«1-1') v (') 5:()}
t dt ' == u t
• (7.2.44)
v(O) == 0
Upon Laplace transforming according to
J
~
V(p):= e- P1 v(t) dt (7.2.45)

o
the initial-value problem (7.2.44) yields
p V(p) + 0'2 V(p) == I, (7.2.46)

p+A
from which
V()== p+A (7.2.47)
P p(P+A)+0'2 .
Upon inverting the Laplace transform, (7.2.47) leads to
_~I[COS,/"'~
v(t) = e' - 4" t + ~11'
A/2 ~]
_ ~ Si.,/"' - 4" t. (7.2.48)
Observe that, when A == 0, the perturbative solution (7.2.48) oscillates

indefinitely, as t ~ 00, though the exact solution (7.2.40) vanishes in this limit
(see Figure 7.2)!
7.2.2(c) Renormalized Solution

In an effort to remedy the qualitative difficulty mentioned above, Kraichnan
(1961) proposed a renormalization of the perturbative procedure. (Kraichnan
called this procedure the direct interaction approximation.) This involves
replacing EQl in the bracket in equation (7.2.9) by u{t); we then obtain in place
of the initial-value problem (7.2.42),
1. (J 2n,} =1/2
2. (J 2/')...2 = 1/3
3. (J2/')...2 =.26
0.8
0.6
0.4
0.2
o 1.5 2
Figure 7.2. Comparison of the exact solution (solid line) with the
perturbative solution (7.2.48) (dashed line) (from Shivamoggi
et. ai, 1999).
(~ +}l) (u (t)) + Jot (b(t) b(t') u(t - t'))(u(t')) dt' =<5 (t)}.
(7.2.49)
(u(O)) =1
Following Kraichnan (1961), we make the quasi-normality assumption4
(b(t) b(t') u(t-t')) =(b(t)b(t'))(u(t -t')) (7.2.50)
and use (7.2.36) and (7.2.43); we then obtain from the initial-value problem
(7.2.49),
dv +0'2
dt
J0
e-A.(t-t') v(t -t') v(t') dt' =<5 (t)}
• (7.2.51)
v(O) =0
Upon Laplace transforming with respect to t, the initial-value problem
(7.2.51) yields the functional equation-
p V(p) +0'2 V(p + A) V(p) =1, (7.2.52)
from which
1
V(p) = , ( )" (7.2.53)
p+O'" V p+A
Equation (7.2.53), on iteration leads to the continued-fraction representation:
1
V(p) =----0'---2- - - (7.2.54)
p+ 2
(p+ ,1)+ 0'
(p+2A)+ ...
Successive truncations of this continued fraction yield the following
approximants -
V( )=.!. ' (
p+A
p p P p+/I,~) +0'", ,
(p + A)(p + 2A) + 0'2
(7.2.55)
On inverting the Laplace transform, (7.2.55) leads to the following

approximants:
4 This amounts to expressing the triple correlations in terms of pair correlations while
neglecting the residual fourth-order cumulants.
(7.2.56)
Comparison of (7.2.56) with the perturbative solution (7.2.48) shows that-

* the first approximant corresponds to the total neglect of the random
aspects in equation (7.2.35),
* the second approximant corresponds to the incorporation of the random
aspects in equation (7.2.35) in aperturbative way,
* the third (and higher) approximant corresponds to the incorporation of the
random aspects in equation (7.2.35) in a non-perturbative (renormalized)
way.
Further comparison of (7.2.56) with the exact solution (7.2.39) shows that
(see Figure 7.3)-
* the third approximant is quite close to the exact solution,
* the renonnalized solution becomes more accurate as the ratio (1/ A.
becomes smalIer.
7.3 Exercises 333
I. a'/..' =1/3
2. a'/ ..' =1/9
3. a '/..' = 1/27
0.8
0.6
I
I
I
I
0.4 I
I
I
I
I
I
I
I
I
0.2 I
I'I
I
\
\
\
\
,,
0
0 2 4 6 10
Figure 7.3. Comparison of the exact solution (solid line) with the
third approximant (dashed line) in equation (7.2.56)
(from Shivamoggi et. ai, 1999).
7.3. Exercises
1. Use the QFT perturbation procedure to solve the Thomas-Fermi equation

(Bender et aI, 1989)-
2. Use the QFT perturbation procedure to solve the Kadomtsev equation

(Shivamoggi and Rollins, (1997»-
iP" = (xiPr }
iP(O) =1, iP(oo)=O
f'" + ff" - agg" = 0

g" + PUg' - f'g) = 0
17 =0 : f =0, f' =0, g =0
I
3. Use the QFT perturbation procedure to solve the Greenspan-Carrier
equations (Shivamoggi and Rollins, (1999»-
17 ~ 00 : f "" 17, g "" 17

Here, a and Pare positive parameters.
4. Use the perturbation method of Section 7.2 to solve (Shivamoggi et aI.,
(2000» -
d2~
dz-
+k; [l+e,u(z)] E+e(aE 2 )+e 2 (bE 3 )=O
where ,u(z) describes the small random variations in the refractive index,
and a,b are constant parameters.
References
Andrews, L.c. and Phillips, R.L. (1998), Laser Beam Propagation through
Random Media, SPIE Press.
Bender, C.M., Milton, K.A., Pinsky, S.S., and Simmons, L.M. (1989), A new
perturbative approach to nonlinear problems, J Math. Phys. 30, 1447.
Berman, V.S. (1978), On the asymptotic solution of a nonstationary problem on
the propagation of a chemical reaction front, Soviet Math. Dokl. 19, 1076.
Bogoliubov, N.N. and Mitropolski, Y.A. (1961), Asymptotic Methods in the
Theory ofNonlinear Oscillations, Gordon and Breach.
Bolotin,V.V. (1964), Dynamic Stability ofElastic Systems, Holden Day.
Bourret, R.C. (1962), Stochastically perturbed fields with applications to wave
propagation in random media, Nuovo Cimento 26, 1.
Bretherton,F.P. (1964), Resonant interactions between waves: The case of
discrete oscillations, J Fluid Mech. 20,457.
Brillouin, L. (1926), Remarques sur la mechanique ondulatoire, J Phys. Radium
7,353.
Burgers, J.M. (1948), A mathematical model illustrating the theory of
turbulence, Adv. Appl. Mech. 1, 171.
Cary, J.R. (1981), Lie transforms and their use in Hamiltonian perturbation
theory.
Chandrasekhar, S. (1961), Hydrodynamic and Hydromagnetic Stability,
Clarendon Press.
Chemov, L. (1969), Wave Propagation in a Random Medium, Dover.
Cole, J.D. (1968), Perturbation Methods in Applied Mathematics, Ginn-
Blaisdell.
Curle, N. (1955), The influence of solid boundaries on aerodynamic sound,
Proc. Roy. Soc. (London) A231, 505.
Doob, J.L. (1942), The Brownian movement and stochastic equations, Ann.
Math. 43,351.
Drazin, P.G. and Johnson, R.S. (1989), Solitons, Cambridge University Press.
336 References
Eckaus, W. (1979), Asymptotic Analysis of Singular Perturbations, North-

Holland.
Erdelyi, A. (1956), Asymptotic Expansions, Dover.
Ford, J. and Waters, J. (1963), Computer studies of energy sharing and
ergodicity for nonlinear oscillator systems, J Math. Phys. 4, 1293.
Fox, P.A. (1955), Perturbation theory of wave propagation based on the method
of characteristics, J Math. And Phys. 34, 133.
Fraenkel, L.E. (1969), On the method of matched asymptotic expansions, Parts
I, II, and III, Proc. Cambridge Phil. Soc. 65,209,233, and 263.
Friedrichs, K.O. (1955), Asymptotic phenomena in mathematical physics, Bull.
Am. Math. Soc. 61,484.
Friedrichs, K.O. (1942), The mathematical structure of the boundary-layer
problem, in Fluid Mechanics, Ed. von Mises, R. and Friedrichs, K.O.,
Springer-Verlag.
Gardner, C.S. (1959), Adiabatic invariants of periodic classical systems, Phys.
Rev. 115, 791.
Green, G. (1837), On the motion of waves in a variable canal ofsmall depth and
width, Trans. Cambridge Phil. Soc. 6,457.
Grimshaw, R.HJ. (1979), Slowly-varying solitary waves, I. Korteweg-deVries
equation, Proc. Roy. Soc. (London) A368, 359.
Grimshaw, R.H.J. (1990), Nonlinear Ordinary Differential Equations,
Blackwell.
Hasimoto, H. and Ono, H. (1972), Nonlinear modulation of gravity waves, J.
Phys. Soc. Japan 33, 805.
Henon, M. and Heiles, C. (1964), The applicability of the third integral of
motion: Some numerical experiments, Astron. J 69, 73.
Hinch, EJ. (1991), Perturbation Methods, Cambridge University Press.
Holmes, M.H. (1995), Introduction to Perturbation Methods, Springer-Verlag.
Infeld, E. and Rowlands, G. (1979), On the stability of electron-plasma waves,
J Phys. A12, 2255.
Janssen, P.A.E. (1981), Modulational instability and the Fermi-Pasta-Ulam
recurrence, Phys. Fluids 24, 23.
Jeffrey, A. and Kakutani, T. (1972), Weak nonlinear dispersive waves: A
discussion centered around the Korteweg-deVries equation, SIAM Rev. 14,
582.
Jeffreys, H. (1924), On certain approximate solutions of linear differential
equations of the second order, Proc. London Math. Soc. 23, 428.
References 337
Johnson, R.S. (1973), On an asymptotic solution of the Korteweg-deVries

equation with slowly-varying coefficients, J. Fluid Mech. 60,813.
Kane, T.R. and Kahn, M.E. (1968), On a class of two-degree-of-freedom
oscillations, J. Appl. Mech. 35,547.
Kaplun, S. (1957), Low Reynolds number flow past a circular cylinder, J. Math.
Mech. 6,595.
Keller, lB. (1964), Wave propagation in random medium, Proc. Symp. Appl.
Math. 16,84.
Keller, J.B. (1968), Perturbation Theory, Lecture Notes, Department of
Mathematics, Michigan State University.
Kevorkian, l (1966), The two variable expansion procedure for the approximate
solution of certain nonlinear differential equations, in Space Mathematics,
Part 3, Ed. Rosser, lB., American Mathematical Society.
Kevorkian, l (1980), Resonance in a weakly-nonlinear system with slowly-
varying parameters, Stud. Appl. Math. 62,23.
Kevorkian, l and Cole, J.D. (1996), Multiple Scale and Singular Perturbation
Methods, Springer-Verlag.
Knickerbocker, C.J. and Newell, A.C. (1980), Shelves and the Korteweg-
deVries equation, J. Fluid Mech. 98,803.
Kraichnan, R.H. (1961), Dynamics of nonlinear systems,J. Math. Phys. 2, 124.
Kramers, H.A. (1926), Wellenmechanik und halbzahlige quantisierung, Z. Phys.
38,828.
Kruskal, M.D. (1962), Asymptotic theory of Hamiltonian and other systems
with all solutions nearly periodic, J. Math. Phys. 3, 806.
Krylov, N. and Bogoliubov, N.N. (1947), Introduction to Nonlinear Mechanics,
Princeton University Press.
Kuehl, RH. and Imen, K. (1985), Finite-amplitude ion-acoustic solitons in
weakly-inhomogeneous plasmas, Phys. Fluids 28, 2375.
Kuzmak, G.E. (1959), Asymptotic solutions of nonlinear second-order
differential equations with variable coefficients, J. Appl. Math. Mech. 23,
730.
Lagerstrom, P.A. (1988), Matched Asymptotic Expansions: Ideas and
Techniques, Springer-Verlag.
Lagerstrom, P.A. and Cole, J.D. (1955), Examples illustrating expansion
procedures for the Navier-Stokes equations, J. Rat. Mech. Anal. 4,817.
338 References
Lake, B.M., Yuen, H.e., Rungaldier, H., and Ferguson, W.E. (1977), Nonlinear
deep-water waves: Theory and experiment, Part 2. Evolution of a
continuous wavetrain, J Fluid Mech. 83,49.
Landau, L.D. (1945), On shock waves at large distances from the place of their
origin, Soviet J Phys. 9, 496.
Landau, L.D. (1946), On the vibrations of the electronic plasma, J Phys.
(USSR) 10,25.
Langer, R.E. (1931), On the asymptotic solution of ordinary differential
equations with an application to the Bessel functions of large order, Trans.
Am. Math. Soc. 33,23.
Langer, R.E. (1935), On the asymptotic solutions of ordinary differential
equations with reference to Stokes' phenomenon about a singular point,
Trans. Math. Soc. 37,397.
Latta, G.E. (1951), Singular Perturbation Problems, Ph.D. Thesis, California
Institute of Technology.
Leslie, D.C. (1973), Developments in the Theory of Turbulence, Clarendon
Press.
Lighthill, MJ. (1949), A technique for rendering approximate solutions to
physical problems uniformly valid, Phil. Mag. 40, 1179.
Lighthill, M.J. (1952), On sound generated aerodynamically, I. General theory,
Lighthill, MJ. (1967), Some special cases treated by the Whitham theory, Proc.
Roy. Soc. (London) A299, 28.
Lin, C.e. (1954), On a perturbation theory based on the method of
characteristics, J Math. And Phys. 33, 117.
Lindstedt, A. (1882), Uber die integration einer rur st6rungsthorie wichtigen
differtialgleichung, Astron. Nachr. 103,211.
Liouville, J. (1837), Sur Ie developpement des fonctions on partices de fonctions
en series, J Math. Pures Appl. 2, 16.
Luke, J.e. (1967), A variational principle for a fluid with a free surface, J Fluid
Mech. 27,395.
McGoldrick, L.F. (1970), On Wilton's ripples: A special case of resonant
interactions,J Fluid Mech. 42, 193.
Nayfeh, A.H. (1964), A Generalized Methodfor Treating Singular Perturbation
Methods, Ph.D. Thesis, Stanford University.
Nayfeh, A.H. (1969), On the nonlinear Lamb-Taylor instability, J Fluid Mech.
38,619.
References 339
Nayfeh, A.H. and Hassan, S.D. (1971), The method of multiple scales and
nonlinear dispersive waves, J. Fluid Mech. 48,463.
Nayfeh, A.H. (1973), Perturbation Methods, Wiley.
Nayfeh, A.H. (1981), Introduction to Perturbation Techniques, Wiley.
Nishikawa, K. and Kaw, P.K. (1975), Propagation of solitary ion-acoustic waves
in inhomogeneous plasmas, Phys. Lett. A50,455.
O'Malley, R.E. (1974), Introduction to Singular Perturbations, Academic Press.
Poincare, H. (1967), New Methods in Celestial Mechanics, (1892), Translation
NASA-TIF 450.
Prandtl, L. (1904), Uber Fliissigkeitsbewegung bei sehr kleiner Reibung, in
Verhandlungen des III. Internationlen Mathematiker-Kongresses,
Heidelberg.
Pritulo, M.F. (1962), On the determination of uniformly accurate solutions of
differential equations by the method of perturbation of coordinates, J. Appl.
Math. Mech. 26,661.
Rayleigh, l.W.S. (1912), On the propagation of waves through a stratified
medium with special reference to the question of reflection, Proc. Roy. Soc.
(London) A86, 208.
Schlichting, H. (1972), Boundary Layer Theory, McGraw-Hill.
Schrodinger, E. (1926), Quantisierung ais Eigenwertproblem, Ann. Phys. 80,
437.
Sears, W.R. (1969), Aerodynamics, noise, and the sonic boom, AIAA J. 7, 577.
Shivamoggi, B.K. (1977a), Nonlinear hyperbolic waves, J. Sound and Vibr. 54,
603.
Shivamoggi, B.K. (1977b), Dynamic buckling of a thin elastic plate: Nonlinear
theory, J. Sound and Vibr. 54, 75.
Shivamoggi, B.K. (1977c), Uniformly-valid mach number expansion of the
Navier-Stokes equations and mathematical formalization of Lighthill's
theory of aerodynamically generated sound, J. Sound and Vibr. 51, 303.
Shivamoggi, B.K. (1978a), Nonlinear hyperbolic waves, J. Sound and Vibr. 55,
594.
Shivamoggi, B.K. (1978b), Method of matched asymptotic expansions:
Asymptotic matching principle for higher approximations, Z. Angew. Math.
Mech. 58,354.
Shivamoggi, B.K. (1978c), Wave propagation in an inhomogeneous medium, J.
Acoust. Soc. Am. 63, 1926.
340 References
Shivamoggi, B.K. (1979), Nonlinear theory of Rayleigh-Taylor instability of

superposed fluids, Acta Mech. 31, 30 I.
Shivamoggi, B.K. (198Ia), Method of multiple scales and the problem of
aerodynamically generated sound, Arch. Mech. 33,603.
Shivamoggi, B.K. (198Ib), Effect of finite-ion temperature on ion-acoustic
solitary waves in an inhomogeneous plasma, Can. J. Phys. 59,719.
Shivamoggi, B.K. (1982a), Rayleigh-Taylor instability of a plasma in a
magnetic field: Nonlinear theory, J. Plasma Phys. 27, 129 (erratum in 30,
511, (1983».
Shivamoggi, B.K. (1982b), Weakly-nonlinar dispersive waves: Group velocity,
J. Plasma Phys. 27,507.
Shivamoggi, B.K. (1983), A variational principle for gravity wavetrains in
magnetohydrodynamics, Q. Appl. Math. 41,31.
Shivamoggi, B.K. (1986), Theory of Hydromagnetic Stability, Gordon and
Breach.
Shivamoggi, B.K. (1987), Internal resonances in weakly-nonlinearly coupled
oscillators with slowly-varying parameters, Z. Angew. Math. Mech. 69,23.
Shivamoggi, B.K. (1988a), Introduction to Nonlinear Fluid-Plasma Waves,
Kluwer.
Shivamoggi, B.K. (1988b), Ion-acoustic solitary waves in an inhomogeneous
plasma, J. Plasma Phys. 40,579.
Shivamoggi, B.K. and Varma, R.K. (1988), Internal resonances in nonlinearly-
coupled oscillators, Acta Mech. 72, Ill.
Shivamoggi, B.K. (1990), Nonlinear development of modulated gravity
wavetrain in deep water, J. Phys. A: Math. And Gen. 23,4289.
Shivamoggi, B.K. and Muilenburg, L. (1991), On Lewis' exact invariant for a
linear harmonic oscillator with time-dependent frequency, Phys. Lett. A
154,24.
Shivamoggi, B.K. (1997), Nonlinear Dynamics and Chaotic Phenomena,
Kluwer.
Shivamoggi, B.K. and Rollins, D.K. (1997), An analytic perturbative solution
for the Kadomtsev equation for a heavy atom in a very strong magnetic
field, J. Phys. A: Math. and Gen. 30, 3681.
Shivamoggi, B.K. (1998), Theoretical Fluid Dynamics, II Ed., Wiley.
Shivamoggi, B.K. and Rollins, D.K. (1999), Magnetohydrodynamic boundary
layer on a flat plate: Further analytic results, J. Math. Phys. 40,3372.
References 341
Shivamoggi, B.K., Taylor, M.D., and Kida, S. (1999), On some mathematical

aspects of the direct interaction approximation in turbulence theory, J.
Math. Anal. Appl. 229,639.
Shivamoggi, B.K., Andrews, L.e., and Phillips, R.L. (2000), Nonlinear wave
propagation in a turbulent medium, Physica A 275, 86.
Shivamoggi, B.K. and Rollins, D.K. (2002), Evolution of solitary-wave solution
of the perturbed regularized long-wave equation, Chaos, Solitons, and
Fractals, 13, 1129.
Simmonds, lG. and Mann, lE. (1986), A First Look at Perturbation Theory,
Krieger.
Smyth, N.F. (1984), Soliton on a Beach and Related Problems, Ph.D. Thesis,
California Institute of Technology.
Strauss, W.A. (1992), Partial Differential Equations, Wiley.
Struble, R.A. (1962), Nonlinear Differential Equations, McGraw Hill.
Tatarskii, V. (1967), Wave Propagation in a Turbulent Medium, Dover.
Timoshenko, S.P. and Woinowsky-Krieger, S. (1959), Theory of Plates and
Shells, McGraw-Hill.
Timoshenko, S.P. (1959), Elastic Stability, McGraw Hill.
Tritton, D.l (1988), Physical Fluid Dynamics, Clarendon Press.
Tsien, H.S. (1956), The Poincare-Lighthill-Kuo method, Adv. Appl. Mech. 4,
281.
Van Dyke, M.D. (1952), A study of second-order supersonic flow theory, NACA
Report 1081.
Van Dyke, M.D. (1975), Perturbation Methods in Fluid Mechanics, Parabolic
Press.
van Kampen, N.G. (1976), Stochastic differential equations, Phys. Rep. 24, 171.
van Kampen, N.G. (1992), Stochastic Processes in Physics and Chemistry,
Elsevier.
von Karman, Th. (1910), Festigkeitsprobleme im Maschinenbau, Enzyklopadie
der Math. Wiss. 4, 349.
Wasow, W. (1965), Asymptotic Expansions for Ordinary Differential Equations,
Wiley.
Wentzel, G. (1926), Eine verallgemeinerung der quantenbedingungen fur die
zwecke der wellenmechanik, Z. Phys. 38, 518.
Whitham, G.B. (1952), The flow pattern of a supersonic projectile, Comm. Pure
and Appl. Math. 5,301.
342 References
Whitham, G.B. (1965), A general approach to linear and nonlinear waves using
a Lagrangian, J Fluid Mech. 22, 273.
Whitham, G.B. (1967), Variational methods and applications to water waves,
Whittaker, E.T. (1914), On the general solution of Mathieu's equation,
Edinburgh Math. Soc. Proc. 32,75.
Whittaker, E.T. (1964), Analytical Dynamics of Particles and Rigid Bodies,
Cambridge University Press.
Wilton, J.R. (1915), On ripples, Phil. Mag. 29,688.
Yuen, H.C. and Lake, B.M. (1975), Nonlinear deep-water waves: Theory and
experiment, Phys. Fluids 18, 956.
Answers to Selected Problems
Chapter 1
4.
l' 2' (-I)" ,
f(w)-I-'-:'+~+ ... +~+(n+I}!
w w w
~ (-1)"+1
0 (W+X)
J :2 e-xdx
-x -x -x -x
5. En(x} - ~-n~+n(n+I};- + ... +(_1}r-1 n(n+ 1} ... (n+r-2}~
x x x x
00 -Xl
+(-lrn(n+I} .. ·(n+r-l) Jr(n+r)X~ dt

I X
7. y-a- e-
2ax
1 ) e-4ar +...
--'!'(x+_
2a 4a- 4a
Chapter 2
5e 2
e
2. XI =1--+-+",
Z 8
e 5e 2
x =-1----+· ..
2 2 8
x =
3
_.!.e + e - 2e 3 + ...
5. y = -t + e(1- t; ) + e 2 (t _ t~ ) + ...
-al I 2 (e-al - I) ]
6. y=sint+e
[
~
4+a-
sint+ ( ') cost +...
a 4+a-
344 Answers to Selected Problems
7. y=a+bX+]{X_S)!{S)dS+e(_.!..ax 2 _.!..bX 3_]{X-t)3 !(t)dt)

o 2 6 0 6
Chapter 3
2. x =eacos(mt+q»+O(e 3)
m=~;-m2 _e 2 :~2 ~;_m2 +...
3. u = e llt q{t) where
q =Ao +e( AI +~ Ao cos 2t) +...

J.l""±i~8+te2
4.
{;:;2 .
q>=v'- smn1tt-
e .fi
"
.3
sm n1tt+···
16n"tr"
, "
/l,=n"tr" --3 e+···
2
5. y= ;+e( x~x~ 1 )+...

Application of Lighthill's method leads to
2x+ex
Y"" 2x 2 +e·
8. u =! {x -ct[1 +1 f'{x -ct)]} +g{x +ct [I +1 g'{x +ct)]} +...
Chapter 4
2. u=acosVt+e(~32 cOS3Vt-~
32
Sin3 Vt )+..
where
Answers to Selected Problems 345
2
da a( a )
di=ei 1-
4
2
dV! a
-=I-e-
dt 8
3J
6. m=l+e-+···
4
Chapter 5
~ f
x/.Jt _.::.
1. y= bex-1_be- 1 +a+ - (be-I -a) e 2 dr
1r 0
2. _ _ 1+ a e-~ + e
- _l+_a
Y-l+m ( )
[L1 a + ~(I_+_a_)~a..,...}
(l+a)(I+ax) (l+ax)3
1 x2 a --
+{ -a(l+a)-+---a(l+a)
2
}
e £X] +...
2 e 1+a
2
x 1 5/2 1
-+ - 5( I)' O:S;x:s; "'2
2 1+ e-4£ x- .f1 -v 4
5. Y~
x2 3 5/2 1
---+
z 2 -45( x -I)'
-
"'2
-V4
<x:S;l
1+ e £ .f1
_3 1- x
7. Y = a (2x + 1) + (b - 3a) e £ +...
9( b - 3a) x'+x-2
y
c
=a (2x + 1) + (2x+l) 2 e £ +...
Chapter 6
where
346 Answers to Selected Problems
{
~ laf a, (00 '# A
da 2(00
dt, = - 3i Ia12 a -Jli

-, (00 "'I\,
,
2(00 4(00
9(b - 3a) _ 2-x'-x

9. y =a(2x+l)+ , e f +...
c (2x+lt
1+a -Hx+~')
10. y=---(l+a) e
l+m
a(l+a) a a 2 (2+a) _Hx+ a ;2)]
+e - e +...
[ (1+m)3 (l+a)(l+m) l+a
Index
A Burgers equation, 179
Adiabatic invariants, 130, 131 c

Aerodynamically generated sound,
273 Calculus of variations, 141
Algebraic equations, 14 Canonical transformations, 125,
Asymptotic matching principle, 155, 131,132,133,134,150,152
161 Centered expansion fan, 213
Asymptotic sequence, 6 Characteristics, 30, 61, 63, 72, 166,
Asymptotic series (expansions), 1,4, 209,211
6 Composite expansion method, 185
Bessel functions, 9, 20
uniform validity, 7, 28 D
Asymptotic solutions, 9
Averaged Lagrangian method, 122, Differential equations with slowly-
137 varying coefficients, 252
Averaging method, 113 Duffing's equation, 25,46,59,115,
116,119,139,246
B Dynamic buckling of a thin elastic
plate, 265
Bessel functions, 9, 20
Bernoulli equation, 24 E
Bifurcations, 65
Blasius equation, 203, 320 Eigenfunction expansion, 52
Boundary layer flow past a flat Eigenvalue perturbations, 42
plate, 199 Elastic column buckling, 65
Blasius equation, 203 Electron-plasma nonlinear waves,
displacement thickness, 206 82,307
Boundary layers, 155, 156, 160,
164, 177
348 Index
F I
Floquet theory, 51, 97, 270 Initial-value problem for partial

Fluid dynamics applications, differential equations, 209
aerodynamically generated Inner-outer expansions, 158
sound,273 Interior layers, 181
boundary layer flow past a flat Ion-acoustic solitary waves in an
plate, 199 inhomogeneous plasma, 310
decay of a line vortex, 31
modulated gravity waves, 135 J
nonlinear shallow water waves,
281 JWKB approximation, 190
nonlinear waves in a gas, 71
Rayleigh-Taylor instability, 75 K
Fredholm alternative theorem, 42,
95 Korteweg-deVries equation, 284
Frequency shift, 41, 47,84,114, Krylov-Bogoliubov method, 114
116,128,229,289 Krylov-Bogoliubov-Mitropolski
method,118
G
L
Gauge functions, 2
Langer's method, 197
H Lighthill's method, 55
Limit cycle, 118
Hamilton-Jacobi theory, 148
action-angle variables, 151 M
canonical transformations, 150
Hamilton-Jacobi equation, 152 Matched-asymptotic expansions,
Hamilton's equations, 148 155
Hamiltonian perturbation method, asymptotic matching principle,
125 155,161
systems with constant boundary layers, 155, 156, 160,
parameters, 125 164,177
systems with variable composite expansion method,
parameters, 130 185
elliptic equations, 176
Index 349
hyperbolic equations, 166 P

inner-outer expansions, 158
interior layers, 181 Partial differential equations, 29, 35
JWKB approximation, 190 elliptic equations, 38, 176
Langer's method, 197 hyperbolic equations, 38, 166
turning-point problems, 190 initial-value problem, 209
Mathieu equation, 47, 270 nonlinear hyperbolic equations,
Modulated gravity waves in a fluid, 211
135 parabolic equations, 38, 179
Multiple-scale method, 219 Perturbation expansion, 1
differential equations with Perturbation methods, 1
slowly-varying coefficients, 252 algebraic equations, 14
generalized multiple-scale averaging method, 113
method,262 averaged Lagrangian method,
Struble's method, 243 122, 137
eigenfunction expansion, 52
N eigenvalue perturbations, 42
Hamiltonian perturbation
Nonlinear diffusion problem, 236 method,125
Nonlinear hyperbolic equations, 211 Krylov-Bogoliubov method,
Nonlinear SchrOdinger equation, 295 114
Nonlinear shallow water waves, 281 Krylov-Bogoliubov-Mitropolski
Korteweg-deVries equation, method,118
284 Lighthill's method, 55
nonlinear SchrOdinget equation, matched asymptotic expansions,
295 155
perturbed solitary waves, 290 multiple-scale method, 219
second-harmonic resonance, nonuniformities, 19,30
303 ordinary differential equations,
solitary waves, 286 21
Stokes waves, 288 partial differential equations, 29
wave modulation, 295 Poincare-Lindstedt-Lighthill
Nonlinearly coupled oscillators, 132, method,42
239,243,256 Pritulo renormalization method,
Nonuniformities, 19,30 59, 73
quantum-field-theoretic
perturbation method, 319
350 Index
regular perturbation methods, Rayleigh-Taylor instability of

13 superposed fluids, 75
secular terms, 26 Regular perturbation methods, 13
singular perturbation algebraic equations, 14
expansions, 17, 19,31 nonlinear hyperbolic wave
strained coordinates/parameters, propagation, 29
41,63,72,78,83,86 ordinary differential equations,
stochastic differential equations 21
perturbation method, 322 partial differential equation, 29
Perturbed solitary wave propagation, projectile motion, 22
290
Plasma physics applications s
electron-plasma nonlinear
waves, 82 Secular terms, 26
hot electron-plasma nonlinear Shock waves, 74,180,220
waves, 307 Simple pendulum, 45,128,131
ion-acoustic solitary waves in Singular perturbation methods, 17,
an inhomogeneous plasma, 310 19,31
Rayleigh-Taylor instability, 84 averaging method, 113
Poincare-Lindstedt-Lighthill averaged-Lagrangian method,
method,42 122, 137
Pritulo renormalization method, 59, eigenfunction expansion, 52
73 eigenvalue perturbation, 42
Projectile motion, 21 Hamiltonian perturbation
method,125
Q Krylov-Bogoliubov method,
114
Quantum-field-theoretic Krylov-Bogoliubov-
perturbation method, 319 Mitropolski, method, 118
Blasius equation, 320 Lighthill method, 55
matched asymptotic expansions,
R 155
Mathieu equation, 47, 270
Radiation problem, 172 multiple-scale method, 219
Rayleigh-Taylor instability of a nonlinear wave propagation, 59,
plasma, 84 61,71,123
Index 351
Poincare-Lindstedt-Lighthill v
method,42
Pritulo renormalization method, van der Pol oscillator, 117, 121,224
59, 73 Vortex line decay, 31
simple pendulum, 45
strained coordinates/parameters, w
41,63,72,78,83,86
wave propagation in Wave propagation
inhomogenous medium, 63 inhomogeneous medium, 63
Solid mechanics applications modulation, 293
elastic column buckling, 65 nonlinear, 61, 71,123,228,229,
dynamic buckling of an elastic 233
plate, 265 random medium, 322, 325
Solitary waves, 286 Whittaker's method, 50
Stochastic differential equations
perturbation method, 322
renormalization method, 327
Stokes waves, 288
Strained coordinates/parameters, 41,
63,72,78,83,86
Struble's method, 243
Subcharacteristics, 168, 176,203
Subcritical instability, 82, 232
Supercritical equilibria, 65
Taylor series, 1, 2
Turning-point problems, 190
JWKB approximaiton, 190
Langer's method, 197
u
Uniform validity, 7, 27
Permissions
Table 2.1 reprinted from Nayfeh, A.H., Introduction to Perturbation

Techniques, 1981, with permission ofJohn Wiley & Sons, Inc.
Figures 2.1,5.14-5.16,6.1, and 6.4 reprinted from Holmes, M.H., Introduction

to Perturbation Methods, 1995, with permission of Springer-Verlag.
Figures 4.1 and 4.2 reprinted from Nayfeh, A.H., Introduction to Perturbation
Techniques, 1981, with permission ofJohn Wiley & Sons, Inc.
Figure 5.3-5.7 and 5.9 reprinted from Kevorkian, 1. and Cole, J.D., Multiple
Scale and Singular Perturbation Methods, 1996, with permission of Springer-
Verlag.
Figure 5.12 reprinted from Lagerstrom, P.A., Matched Asymptotic Expansions:

Ideas and Techniques, 1988, with permission of Springer-Verlag.
Figure 5.17 reprinted from Tritton, DJ., Physical Fluid Dynamics, 1988, with
permission of Oxford University Press.
Figure 6.6 reprinted from Ford, 1. and Waters, 1., Computer Studies of Energy
Sharing and Ergodicity for Nonlinear Oscillator Systems, Journal of
Mathematical Physics, 4,1293, 1963, with permission of the American Institute
of Physics.
Figure 6.9 adapted from Struble, R.A., Partial Differential Equations, 1962,
with permission of McGraw-Hili Education.
Figure 6.11 reprinted from Shivamoggi, B.K., Dynamic Buckling of a Thin

Elastic Plate: Nonlinear Theory, Journal of Sound and Vibration, 54, 75, 1977,
with permission of Elsevier.
354 Permissions
Figure 6.16 reprinted from Lake, B.M., Yuen, H.C., Rungaldier, H., and
Ferguson, W.E., Nonlinear Deep-Water Waves: Theory and Experiment. Part 2.
Evaluation of a Continuous Wavetrain, Journal of fluid Mechanics, 83, 49,
1977, with permission of Cambridge University Press.
Figure 7.1 reprinted from Shivamoggi, B.K. and Rollins, D.K.,

Magnetohydrodynamic Boundary Layer on a Flat Plate: Further Analytic
Results, Journal ofMathematical Physics, 40, 3372, 1999, with permission from
the American Institute of Physics.
Figures 7.2 and 7.3 reprinted from Shivamoggi, B.K., Taylor, M.D., and Kida,
S., On Some Mathematical Aspects of the Direct Interaction Approximation in
Turbulence Theory, Journal of Mathematical Analysis and Applications, 229,
639,1999, with permission of Academic Press.

2003 Book PerturbationMethodsForDifferen PDF

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

2003 Book PerturbationMethodsForDifferen PDF

Transféré par

Droits d'auteur :

Formats disponibles

"Was du ererbt von deinen Viitern hast, erwirb es, urn es zu besitzen.

In memory ofmy father, Krishnarao Shivamoggi

Springer Science+Business Media, LLC

Library of Congress Cataloging-in-Publication Data

QA871 .S44 2002

AMS SubjectC1assifications: 34E05, 34EI0. 34E13. 34E15, 34E20, 35B20, 35B25

Printed on acid-free paper.

ISBN 978-1-4612-6588-7 SPIN 10764753

Typeset by the author.

Chapter 1 Asymptotic Series and Expansions 1

Chapter 2 Regular Perturbation Methods 13

Chapter 3 The Method of Strained CoordinateslParameters •.••••..•..•.........•..41

Chapter 4 Method of Averaging 113

Modulated Gravity Wave Packet on the Surface of a Fluid 135

Chapter 5 The Method of Matched Asymptotic Expansions 155

5.11 Exercises 207

Chapter 6 Method of Multiple Scales 219

Chapter 7 Miscellaneous Perturbation Methods 319

In nonlinear problems, essentially new phenomena occur which have no

1 In an alternative approach via numerical solution of nonlinear problems, computers

The method of perturbation expansion is carried out with respect to a small

Orlando, Florida Bhimsen K. Shivamoggi

Asymptotic Series and

The method of perturbations can be used to develop approximate solutions to

boundary conditions in an analytic way. The lowest-order terms in these series

1.2. Taylor Series Expansions

Suppose a function f(x) is infinitely differentiable at x = x o' then we may

Example 1: Taylor series expansions of some standard functions are -

1.3. Gauge Functions

Consider the limit of a function f (£) as £ ~ O. This limit may depend on

Example 2: lim e '

Suppose e> O. Suppose f(e) has no essential singularity at e = 0 (e.g.,

In order to determine the rate at which f(e) tends to 0 or 00, we compare

which satisfy, for e near 0,

sine ~ 0 at the same rate as e~0 }

On the other hand, in order to detennine the rate at which

1"(0) = lim (-.!..-4 - 3

fact, e £ ~ 0 faster than any power of e, because

1.4. Asymptotic Series and Expansions

Definition: Let f (e) and g (e) be defined in some neighborhood of e = 0 "

if there exists a positive bounded constant M such that

If(e)1 ~ M ~(e)1 or l;i;~1 is bounded

Example 4: e--; = 0 ( ea ), \;j a. Thus, e f is transcendentally small with respect

Example 5: e 2ene = o(e) as e -7 0+.

Thus, e 2ene = o(e) as e -7 0+.

Example 6: sine = O(e) as e -7 0+.

since Icos cl ~ 1. Therefore,

sine = O(e) as e-70+.

The above definition may be extended to functions of e and another variable

limlf(t,e)l=o "d tEl.

If there exists a positive boundedfunction M(t) on I such that

A measure of decreasing orders of magnitude is provided by an asymptotic

Definition: A sequence {lpn (e)}~=1 is an asymptotic sequence if

A sum of terms in an asymptotic sequence is an asymptotic expansion.

Example 7: Lan lpn (e) is an asymptotic expansion of f(e) as e -? 0 if

In a physical problem, the coefficients in an asymptotic expansion will

Definition: An asymptotic expansion

f(x,e) - LOm (e) am (x) as e ~ 0 (1.4.5)

Example 8: sin(x+e)=sinx+e cosx- e~ sinx+ ... is a uniformly valid

Example 9: -J x + e =.,J; (1 + e/x)1/2

Example 10: y(t,e) = e- tt , t > 0, e« 1.

Yapprox ( t,e ) =1-et+'2

E(t,e) = e- Ct -1 + e t _.!. e 2 t 2 = _..!.. e 3 t 3 +....