Académique Documents
Professionnel Documents
Culture Documents
"
("That which you have inherited from your fathers,
acquire it so as to make it your own.")
-Goethe in Faust
Perturbation Methods
for Differential Equations
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.
987 6 543 2 1
Table of Contents
Preface xi
References 335
Answers to Selected Problems 343
Index 347
Permissions 353
Preface
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
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.
1.1. Introduction
I I
lim sin
,->0
(.!.), then one has three possibilities-
e
lim
' ....0
If (e)1 =1:, 0 < A < 00. (1.3.1)
00
e- ,e- ,e-
1 2 3
, •••
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,
4 Asymptotic Series and Expansions
f '( e)=-.!..-
e
2 e -1/£ ,
etc.
Thus,
I 2
j'(O) = lim -.!..-2 e--; = lim ~=O
£->0 e x->~ eX
etc.
where x == 1/e.
I
Therefore, the function e £ cannot be represented by a power series in e. In
I
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)
I 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
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)
Note that
m-I
f(e)- Lancpn(e)
am = ~~ cp:'(e)
which gives am uniquely.
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.
The error is
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
1.4 Asymptotic Series and Expansions 9
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.
- [(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 )]
10 Asymptotic Series and Expansions
is divergent for all e, no matter how small, but a few terms given good accuracy
for moderately small e.
U
" +U _ /32 -1/4
2
_0 .
U -
x
U - eix L an x- n , ao =1.
n=O
u-e
ix [
1+
(/32 -1/4) - (/32 -1/4)(/32 - 9/4) + ...]
2x 8x 2 •
Let
L gn (x) A- n/2
~
g[X,A] - Alf2
n=O
which leads to
etc.
Thus,
')
U ( X,I\. - x -1f4 e ±2/3 AI/I X'/I .
1.6. Exercises
=f ~ e-xdx.
~
f7 dt.
~ -I
f _e_
00 -Xl
2.1. Introduction
y(x;e) - L en Yn(x)
n=O
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).
zJ = 1-.JI=f}
Z2 = 1+.JI=f .
.JI=f =1 _! e _! e2 _ •••
2 8
we obtain
;:O~~:02:,0+ 1 =0 )
a~ + 2aOa2 - 2a 2 =0
etc.
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
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
-~ao + I: O}
ao -2a, -0 .
etc.
Thus,
etc.
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'
Then we obtain
from which
18 Regular Perturbation Methods
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
(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
We then obtain
((X o-1) + el/ 2 XI + .. ,][(xo-1) + el/2 (XI + (1) + ...]
=-e[xo+e1/2 XI + ...],
from which
20 Regular Perturbation Methods
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 ) +".
The roots of
cot(x _lC)
4
= __1_, xlarge.
8x
So, to first approximation, we have
cot(x - :}= 0,
from which
or
2.2 Algebraic Equations 21
x=(n+~) Tr+(j.
This leads to
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
t= °: =x
dx
0, -dt = V;0'
Non-dimensionalizing as follows -
t x
r=--
(Va/g) , y= (Va /g)'
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
2.3 Ordinary Differential Equations 23
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,
0.6 ...-----..-----.----""T""-----,
JJ 0.4
:;
"0 Numerical Solution
en 0.3 1----.fC--+ Yo +-_-":"':---:lo~
Yo +€y,
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
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
2.3 Ordinary Differential Equations 25
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 )+ ...
O(e): ul +u l = -u~ }
ul(O)=O, ul(O)=O'
etc.
where the dots overhead denote differentiation with respect to t.
26 Regular Perturbation Methods
Uo (t) =A cos t .
Substituting this, the 0(£0) problem becomes
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
2.3 Ordinary Differential Equations 27
Putting
u =-AcosO,
we obtain
t=± 1 j dO
~1 +eA 2 " ~1-m2 sin 2 0
where
2
m 2 = (e/2) A
- 1+eA 2
T=
2
~1+eA2
J
" dL}
~1-m2sin20
0
17
-;===4= J
,,/2
dO =K(m 2
)
- ~1+eA2 ~1-m2sin20
0
T=
4 ,J
/2 (1+~sin20+···
2 ) dO
~1+eA2 0 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
28 Regular Perturbation Methods
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.
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.
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 !
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:
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.
a [a 2 Ia ]
Using (2.5.1), equation (2.5.2) 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 -
(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
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)
£=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.)
34 Regular Perturbation Methods
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
5. Using a regular perturbation expansion, solve the initial-value problem
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).
6. Using a regular perturbation expansion, solve the initial-value problem
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
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
2.7. Appendix
~ = ~(x,y)}
(A.2)
T/=T/(x,y)
with the Jacobian
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)
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)
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)
(A.17)
where ~ and 17 are now complex conjugate functions. So, if one makes a
further transfonnation
(A.20)
(A.22)
for simplicity.
2.7 Appendix 39
: =; )
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
(3.1.3)
42 The Method of Strained Coordinates/Parameters
(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)
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)
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
(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)
(3.2.12)
which gives
at .
ul =-- sm roo t . (3.2.13)
2roo
44 The Method of Strained Coordinates/Parameters
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)
(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.
Thus, we have from (3.2.20),
Uo(s) =acosOJo s. (3.2.22)
Using (3.2.22), (3.2.24), and (3.2.25), we obtain from (3.2.18) and (3.2.19),
in agreement with the limiting form (3.2.17a) of the exact solution (3.2.16) for
small e.
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)
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)
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)
Case (i): n = O.
We now obtain from equation (3.2.46),
Uo = 1, say. (3.2.49)
U
.. = s:
-(1 -
1
- -
1
A cos 2t - - cos 4t . (3.2.53)
2 2 8 8
Removal of secular terms in equation (3.2.53) requires
1 1
O2 + - = 0 or O2 = - - . (3.2.54)
8 8
Using (3.2.51) and (3.2.54), (3.2.44) becomes
o=-i e +0(e 2 3
). (3.2.55)
o=1-.!.2 e - ~
32
e 2 + O(e 3
). (3.2.62a)
Case (iii): n =2 .
We now obtain from equation (3.2.46),
Uo =cos2t . (3.2.64)
Using (3.2.64), equation (3.2.47) becomes
..
u1 + 4u. =--1 - 5: cos2t - -1 cos4t.
VI (3.2.65)
2 2
Removal of secular terms in equation (3.2.65) requires
0, =0. (3.2.66)
o= 4 + 2-
48
e + o( e
2 3
). (3.2.70)
o 4
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)
3.2 Poincare-Lindstedt-Lighthill Method of Perturbed Eigenvalues 51
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)
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 + VI =(2a 1a - 8, b + %) sin t
-(2a l b + 81 a + ~) cost + nonsecular terms. (3.2.83)
(3.2.84)
4a~ - ( ±-
8,2 ) = 0 (3.2.85)
52 The Method of Strained CoordinatesIParameters
from which,
.!.+li
2
b =+- -1-- a,
I
li, :1:-1/2. (3.2.87)
--li
2 I
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.
(3.3.4)
2 2 2
0(e ):u; +n rr Un =-j(x)Un -An Un -An un}
2 1 I I I 2 0
L. .fi rr 2
(n 2 _m 2 ) anm sinmrr x
(3.3.8)
m=1
where
F"k =2 (3.3.10)
n 2 L (n 2 -r 2 ).fi bnrsinrnx
r=1
~ ~
(3.3.18)
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
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.
56 The Method of Strained Coordinates/Parameters
(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.
from which
c+2xo
x =---- (3.4.9)
, 2xo
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
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)
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)
,,
\
\
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).
which leads to
(3.5.3)
(3.5.4)
etc.
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)
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!
1 ]' 71+G(~).
U2(X,t)=-8[f'(~r (3.5.20)
u2(x,t) = = (3.5.22)
-± [f'(~O)]2
Equation (3.5.23) then becomes
~I = ~ f'(~o)' (3.5.26)
(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
]
etc.
From equation (3.6.4),
64 The Method of Strained Coordinates/Parameters
from which
or
(3.6.8)
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.
66 The Method of Strained Coordinates/Parameters
I·
I·
x
I·
~
\\
II
II
II
II
x=0 ----------_.
y
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
3.7 Applications to Solid Mechanics 67
where
2
)} = PL
0- EI .
(3.7.3)
Oe
(
3). d2
Y3 12 _ 121 1 (1 1) 3).,~
2 2
'-2-+IIoOY3--211001loIY2-1Io01lo1+211o2 Y I - - Y I -
(dyl)2
~ 2 ~
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
where
3.7 Applications to Solid Mechanics 69
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.
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
70 The Method of Strained Coordinates/Parameters
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
3.8 Applications to Fluid Dynamics 71
-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.
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.
72 The Method of Strained CoordinateslParameten
n=1
or
2
J( = I fS'(d4J') r +2I d 4J,2 ds
Ids ' "ds, 2c ds , I
d'Ato 1
or
3.8 Applications to Fluid Dynamics 73
(3.8.9)
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)
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)
y+l
(J(8)=--~
4c
8)
f '( t - - +O(e).
c
(3.8.14)
(3.8.15)
from which
8=!(S+x). (3.8.16)
2
Using (3.8.15), we have from equation (3.8.15),
(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 Applications to Fluid Dynamics 75
PI> pz
Figure 3.8. The interface accelerated normally
and subjected to a perturbation.
(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
where
3.8 Applications to Fluid Dynamics 77
where
e == a'(2n/}..').
The conditions at infinity are
y~(-I/oo: <l>jX ~O. (3.8.25)
y = 0 : a 1J r + lP jy = 0 (3.8.31)
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
1: en 71n(x, r)
~
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'
- A(') +71, -s (A('))
.0', 'l'lr
2 -
0'1 'l'2r +711 + kc 71 lxx -0 (3.8.45)
(3.8.46)
(3.8.50)
y~(-IYoo:t/J;~)~O (3.8.51)
_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)
(3.8.55)
(3.8.56)
etc.
One obtains for the O(e) problem, the linearized solution-
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)
from which
712 (x, r) = cosx ·(1- cosr) (3.8.66)
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)
3
K--(I-s) ] 1/2
a =+ __8"--__ (3.8.73)
2 - I +s
[
e
Linear 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.
(3.9.3)
3.9 Applications to Plasma Physics 83
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)
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.
84 The Method of Strained Coordinates/Parameters
N, = aQ cos~}.
w6 =1+e (3.9.10)
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
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
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-
(3.9.19)
(3.9.20)
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 : a 17 r -17x~x + ~y = 0 (3.9.27)
y~oo:lfI~O (3.9.31)
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
-l entPn (x,y, r)
~
(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)
'Illy + 11 lx =0 (3.9.41)
(3.9.49)
88 The Method of Strained Coordinates/Parameten
r = 0 : 71 = 0,
2 71 2 <= o. (3.9.53)
(3.9.57)
a llP3< +713 +k; 'lf3. = -a2(lP2< +lPlty71I)-a3lPl< -a l (lP2ty711 +lPlty712
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)
(3.9.65)
3.9 Applications to Plasma Physics 89
(3.9.66)
(3.9.72)
(3.9.73)
One may expect to construct unifonnly valid solutions only for wavenumbers
larger than kc '
lfI2y + 172> = ±
sin2x (I +cos2r) (3.9.77)
y~ -00: ~3 ~O (3.9.90)
(3.9.93)
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)
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.
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)
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
(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,
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
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.
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,
(L'v,h)=O, VhEH
or
(v, Lh)=O,
so
vERt·
A=G ~} (A 1.3)
(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)
(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.,
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)
0 T
I I
T t+T
= a (s) ds + a (s - T) ds
o T
I I
T t
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
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
Let us now generalize the above discussion to consider the case x E ~" ,
x= A(T) x (A2.11)
Then, <I> is called a fundamental matrix, and if <I>(to) =I, I being the unit
matrix, then <I> (t) is the principal fundamental matrix. We have
Further, if we introduce
W(t) =det<l>(t), (A2.15)
ftrA(S) ds
W(t) = W(t o ) e" (A2.16)
or
(A2.17)
or
(A2.18)a
Writing
(A2.18)b
So
W=WtrA, (A2.20)
from which
ftrA(sldS
W(t) = W(t o) e" (A2.16)
Also,
T
ftrA(s)ds
detB = eO (A2.22)
Let
'I'(t);: <I>(t + T). (A2.23)
Then
'P(t) =ei>(t + T) (A2.24)
and using equation (A2.13), we have
'P(t) = A(t+ T) <I>(t + T) = A(t)'I'(t). (A2.25)
Equation (A2.25) implies that 'I'(t) is also a fundamental matrix, and hence,
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)
B =<II(T). (A2.30)
(A2.32)
we have
x(t+T)=<II(t+T)· X o
we obtain
(A2.35)
If we introduce
(A2.36)
we have
ul(O)=O, u (0)=1 2
So,
<1>(0) =I.
Note,
<I>(t + T) = <I>(t)B
where, from (A2.22) and (A2.29),
U1(T)
B =<1>-1(0) <I>(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
where
(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
(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·
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
106 The Method of Strained Coordinates/Parameters
and
PI = P2 = O.
The general solution is
where
where
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
108 The Method of Strained Coordinates/Parameters
f(u;J.l)=O. (A3.2)
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)
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)
110 The Method of Strained Coordinates/Parameters
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;
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
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.
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)
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
4.2 Krylov-Bogoliubov Method of Averaging 115
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).
or
116 Method of Averaging
(4.2.12)
o
which gives a = a, say.
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
Pn [a (0)] =-
1
1r 0
Jh[a(O) cosu, a(O) sinu] sinu cos(nu) du
2H
Figure 4.2. Exact solution of equation (4.22) for x(O) = 0.5, x(O) =0 and
E =0.1 (From Nayfeh, 1981).
4
-I). (4.2.21)
f a (~a
a- -4
) =_.!. £(t+c)
8
-± loga+i log la
2
-41=-i £(t+c),
from which
118 Method of Averaging
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.
2. en un (a, 'II) + o( e
N
u = a cos'll + N 1
+ ) (4.3.1)
n=l
4.3 Krylov-Bogoliubov-Mitropolski Generalized Method of Averaging 119
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)
(4.3.3)
= -e (a cos VI + e U I + e U2 + ..
2 f
(4.3.5)
from which, we obtain to various orders in e,
etc.
Removal of secular terms in equation (4.3.6) requires
2
3a
AI =0, V'I = - . (4.3.8)
8(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
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)~
4.3 Krylov-Bogoliubov-Mitropolski Generalized Method of Averaging 121
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,
o(10 2) : -aa",
2
- 2 + u2 =
u2 [(2 '"2 + "',2) a - A, -dA) ] cos '"
da
2
+ (1- a cos
2
",)( AI cos'" - a "') sin", + ~~ )
+ a 2 u) sin 2", (4.3.17)
etc.
Removal of secular terms in equation (4.3.16) requires
()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
Removal of secular tenns in equation (4.3.42) requires
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
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),
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)
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)
(4.4.11)
~ (m a 2 ) + ~ [( 2k 3 - k) a
2
]= 0. (4.4.13)
(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)
4.5 Hamiltonian Perturbation Method 125
(4.4.17)
~ (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
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 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.
126 Method of Averaging
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'
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)
4.5 Hamiltonian Perturbation Method 127
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
ik8
HI (J,O) = fA k(J) e )
(4.5.16)
SI (J,O) = fisk (J) e ik8 '
Ir-I
(4.5.17)
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
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.
'2I (as
aqo )2 +'21 COo2 q 2 = E (4.5.24)
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
- -dS-os- m
0-
df
ffio)
. -I ( q -
2!
(4.5.27)
from which
~- f sm
q= . 0. (4.5.28)
Wo
aq 1W,oq-') ='IJ~
P = dSo =2 ( fw o - 1. 2!wo cosO. (4.5.29)
H o (I) =fw o }
1 2 . 4 (4.5.30)
HI (I, 0) =-"6 f sm 0
•
(4.5.32)
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)
4.5 Hamiltonian Perturbation Method 131
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
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.
L nk O)k
N
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).
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.
4.5 Hamiltonian Perturbation Method 133
- -
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
according to which
134 Method of Averaging
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
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
as - -)2
I, =;;-=1,-21
Cl'1
(4.5.47)
as -
I, =--= I,
• aCl'2 •
4.6 Applications to Fluid Dynamics 135
2j /
+-.!.- ( 2)3 2sin 3 (q,2 +2q,\)}+£ [w; (j\ +2j2) sin2q"
12 w2 2w\
II =constant (4.5.50)
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)
liquid
Figure 4.4. Deformed surface of fluid under gravity.
tPl + 2"1 (2
tPx + tPy2) + gy = 0 (4.6.5)
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.
4.6 Applications to Fluid Dynamics 137
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)
;£=_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
-a (a 2) +- - a 2) -_0 .
a (dW (4.6.15)
at ax dk
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)
where
(4.6.25)
(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)
4.7. Exercises
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
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
I
x,
derivatives and r
If a point function f(x,y) is defined throughout a region R and if the partial
Dy == E1](X). (AlA)
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.
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)
f
x,
I(y) = F(x,y,y') dx, (A 1.5)
X,
4.8 Appendix 1 143
we have
x,
= 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)
tSI = a~ by]Xlxxay
ay 1
1
+ [aF _ ~ (a~)] bydx.
1
ay
(A.19)
so,
or
(A1.1Ob)
144 Method of Averaging
111(X) g(x) dx = 0
1
I(y) = F(x,y,y',y") dx
X,
(A1.12)
(Al.B)
(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
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
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)
aF _.!!...(aF)+ d (aF)_
2
...+(_I)n d n (aF )=0. (AI.20)
2 n
~ dx~' dx ~" dx ~(n)
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:
146 Method of Averaging
Ifwe perturb the solution u about the extremal solution q,(x,y) according to
(A1.23)
(A1.24)
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
(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,
(Al.2?)
with u =f(x,y) on C.
4.9. Appendix 2
Hamilton-Jacobi Theory4
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 <I> 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
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
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.
(A2.12)
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
(A2.16)
_ ( _) ( ) 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.
152 Method of Averaging
--Y [r)
rat + H q.j'dqj'
--Y t = O'
,
j = 1, ... ,n. (A2.20)
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
so that
(A2.26)
(A2.27)
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
from which,
() =OJt + 8. (A2.32)
154 Method of Averaging
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
5.1. Introduction
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,
(5.2.6)
(5.2.7)
y(i) = 1p [ 1-e_!!..t]
m • (5.2.9)
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
where
-1 ± ..JI -4e 1
s ,
1.-
= 2e
'" -1, - - + 1.
e
The above exact solution may be approximated by
or
y = be l - x + (a - be) e- x/c + O(e) (5.3.3)
from which
lim (lim
c....o x....o
y) =a '* limo (limo y) =be
x.... c....
(5.3.5)
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)
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
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 -
160 The Method of Matched Asymptotic Expansions
(5.3.13)
etc.
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).
(5.3.15)
(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)
etc.
Thus, the inner expansion is given by
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
162 The Method of Matched Asymptotic Expansions
from which
Ao = a- be, AI = be, etc. (5.3.24)
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) _
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.)
5.3 The Inner and Outer Expansions 163
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-~
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
164 The Method of Matched Asymptotic Expansions
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) _
~ =x/t: (5.3.33)
(5.3.34)
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
5.3 The Inner and Outer Expansions 165
etc.
On solving these problems, we obtain
y~O)=~}
2x+1 . (5.3.41)
etc.
(5.3.42)
lim y~O)
x=>o
=lim y~i)
~=>~
(5.3.45)
ax _a
e( iiv at v)=a au
ax +b av
2
2 at 2 (5.4.1)
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
(from Kevorkian and Cole, 1996).
168 The Method of Matched Asymptotic Expansions
x
Figure 5.5. Domain of influence in a characteristic initial-value problem
(from Kevorkian and Cole, 1996).
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)
5.4 Hyperbolic Equations 169
p(x, t)
x
Figure 5.6. Domain of influence and time-like and space-like
subcharacteristics (from Kevorkian and Cole, 1996).
Assuming that v itself is continuous across r =ro one finds from equation
(5.4.7)
aK
-4e-=(b-a) K,
as
from which
with the associated inner limit process £0 => 0, x,i held fixed. Taking
(5.4.14)
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)
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)
so that
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,
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)
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)
(5.4.30)
(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
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
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
5.4 Hyperbolic Equations 175
a2
1--
K==~>O
b
which ensures that i =t is a positive time-like variable so that equation (5.4.35)
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)
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)
176 The Method of Matched Asymptotic Expansions
(5.5.1)
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)
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)
where
K"(;) == AII1]~~ - 2AI21]B~ + A22 •
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
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)
(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)
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.
180 The Method of Matched Asymptotic Expansions
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)
where,
u(i)
0,
=.!.2 u(i)'
0
- s' (I) u(i) + A (I) .
0
(5.6.12)
(5.6.13)
This leads to
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
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
182 The Method of Matched Asymptotic Expansions
be a little bit more complicated because the location of the layer is usually not
known until the expansions have been matched.
x=O:y=1 (5.7.2)
y'=z }
(5.7.4)
z' =~ y(z -1)
(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).
5.7 Interior Layers 183
z
3
c=-oo
-3 3 y
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
184 The Method of Matched Asymptotic Expansions
(5.7.11)
which gives
(i) _ A 1- BeA~
Yo - I +BeA~ , (5.7.13)
-1
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)
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
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.
186 The Method of Matched Asymptotic Expansions
h~ -ho =0 (5.8.6)
(5.8.9)
(5.8.10)
etc.
On solving these problems, we obtain
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 )
5.8 Latta's Method of Composite Expansions 187
or
x
y - b[l +e(l-x)] el - x +e-~ {(a - be)(1 +x)-ebe} +0(e 2 ) (5.8.13)
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)
On solving these problems, we obtain
188 The Method of Matched Asymptotic Expansions
O(lfe): g=X 2 +X
0(1): fr =~
o 2x+ 1 (5.8.22)
ho = a-3b
etc.
fo(b(r),r)=O (5.8.29)
ho(O, r) = 0 (5.8.31)
=hi [b(r), r] =0
hl(O, r) (5.8.37)
etc.
Solving the 0(1) problem, we obtain
(5.8.40)
Thus,
ho =ao(r) X.. (x,r). (5.8.42)
=L
~
hi Cs(r) Xs(x,r),
s=1
(5.8.45)
s=1
J(G~ X; +
b(r)
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'
0b- (;)
]
+O(e)
(5.8.50)
where
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)
Setting
gx = h or J
g = h dx , (5.9.4)
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)
192 The Method of Matched Asymptotic Expansions
(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.
x=O:y=a (5.9.14)
x=l:y=b. (5.9.15)
(5.9.16)
sinA(e -I)
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.
5.9 Turning-Point Problems 193
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
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.
194 The Method of Matched Asymptotic Expansions
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).
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)
(5.9.24)
Putting
(5.9.25)
(5.9.26)
from which
5.9 Turning-Point Problems 195
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
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)
~ ~
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)
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).
196 The Method of Matched Asymptotic Expansions
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
~ ~
(5.9.34)
I I,
Also for large ~ ~ < 0, we have from (5.9.27) and (5.9.28),
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)
Noting that
(5.9.40)
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
(5.9.45)
2 "
O(e)'Y
• Ii<
-~YI -
-
- g'
- y0" _L
g,2 y 0, (5.9.47)
etc.
We have from equation (5.9.46)
Yo (x,~) =Co(x) Ai(~). (5.9.48)
(5.9.53)
y-
~
Co e
.JJr I/(xt 4
. [x
Sill
I
eJI/(t)1 112
dt+n/4
] (5.9.54)
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.
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)
(
'I'
y
i.-
ax 'I' i.-
()y _1
xR
V'2) V'2\{1 = 0 (5 .1 0 .1 0)
E
(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)
(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)
or
d (i) (i) (i) (i) (i))_
(5.10.23)
dY 'IIIYYY + 'IIlx 'IIIYY - 'lilY 'IIlxY - 0
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).
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)
(5.10.32)
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),
5.10 Applications to Fluid Dynamics 205
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)
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)
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('
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 •
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
~ [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
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
5.12. Appendix 1
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)
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
!!.-
dt
[u(x(t),t)] =u, + dx
dt
ux =u, +uux =0. (A2.3)
where
(A2A)
Note: We have
(A2.(»
which lead to
Ur+UUx=O. (A2.9)
where
x = X o +cI.
Thus,
U= q,(x -C/) (A2.l2)
as expected!
O, x < 0,
t=o: u= { (A2.l4)
1, x> 0.
because
(x/t), + (x/t)(x/t)x = 0, (A2.16)
centered
expansion fan
___A _
t
o x
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
214 The Method of Matched Asymptotic Expansions
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.
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"---------~--------...,.-
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
If VI
s
L (u) dx dt =O. (A2.22)
If s
uL'('II) dxdt=O (A2.23)
where
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 ~
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'
from which
g, [u]+ g, [A (u)] = 0 (A2.29)
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= !'
6.1. Introduction
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
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.
Solving equation (6.2.7), we obtain
Xo ±
(t,1) = [Aa (1) eit + Aa (1) e- it ] (6.2.10)
a XI
2
--;-T + XI =-21 eit ( Aa - 21"~) + c.c. (6.2.11)
ot
6.2 Differential Equations with Constant Coefficients 221
Ao -2i~ =0 (6.2.12)
from which
1._
--ll
Ao = Ae 2 (6.2.13)
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)
Using (6.2.14) and (6.2.19), equation (6.2.16) gives the final solution,
t= °: x = 0, x = 1. (6.2.22)
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)
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
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
fix i +x\ _
--, - -21
.(
Ao +aA-o) eiT, +c.c. (6.2.33)
aT;j a1;
Removal of secular terms in equation (6.2.33) requires
Ao+aAo=O (6.2.34)
a1;
from which,
(6.2.35)
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!
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
(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)
2 aA =A-A 2 A. (6.2.52)
a~
Putting
A =.!. a(T.
p2 T.)e irp(7j.r,) (6.2.53)
2 ,
-B(T.I' T.)
U1 - 2 e
iTo + '8
1 1'A 3e 3iT, + c.c. (6.2.57)
226 Method of Multiple Scales
I = 0 : U = F (x), u, = O. (6.2.59)
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
from which,
(6.2.69)
6.2 Differential Equations with Constant Coefficients 227
where
Tm=emt. (6.2.71)
We then obtain from the initial-value problem (6.2.58) and (6.2.59),
0(1) : 4uo~. = 0 (6.2.72)
O(e): 4u,
~.
=2(-~+~)
a; dTJ U oI, +(-~+~)
a; dTJ U o (6.2.74)
etc.
We have from the initial-value problem (6.2.72) and (6.2.73),
Uo =fa (;,1;) + go (TI, 1;) (6.2.76)
(6.2.80)
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)
L e u (;,1;,Z;)
~
n
u(x,t;e)- n (6.2.87)
n=1
where
; == kx - cot, Tm == emt. (6.2.88)
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)
(6.2.97)
so that
(6.2.99)
Using (6.2.95), (6.2.99), and (6.2.100), the initial-value problem (6.2.93) and
(6.2.94) become
Tm = 0 : U •, = 0, u3To = O. (6.2.102)
Note that the linear problem associated with (6.2.106) and (6.2.107) gives
230 Metbod of Multiple Scales
/ = 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
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)
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 Differential Equations with Constant Coefficients 231
(6.2.121)
with
a(O,O) =1. (6.2.122)
Tm =0:
aa =0.
ar; (6.2.125)
~~~ +(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)
232 Method of Multiple Scales
where
o o
~ <I
E
Linear
cutoff
Nonlinear
cutoff
unstable stable
o k
where
(6.2.136)
and we assume,
k = k(X,), (0 = (0(1;).
234 Method of Multiple Scales
(6.2.139)
etc.
Solving equation (6.2.138), we have
qJo =A(X;,r;) e i8 +.:4(XI ,r;) e- i8 (6.2.140)
where
(6.2.141)
Using (6.2.140), equation (6.2.139) becomes
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
(6.2.142)
6.2 Differential Equations with Constant Coefficients 235
(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
Using (6.2.144)-(6.2.148), equation (6.2.143) becomes
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
we obtain from equation (6.2.149)
aa a(,
- + - wa 2) =0
2
(6.2.151)
ar; dx,
ap , ap 3a 2
-+w - = - . (6.2.152)
aT; dx l 8w
236 Method of Multiple Scales
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)
(6.2.164)
etc.
6.2 Differential Equations with Constant Coefficients 237
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
(6.2.169)
where
K=f'(O»O. (6.2.172)
Putting
W(x, 1;) =eK8 (X.r,) (6.2.173)
(6.2.175)
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
(6.2.178)
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
I
L="2m .' ( £+x )2 0"
[ x"+ "] -mg (£+x )(I-cosO) -"210:".
I , (6.2.185)
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)
where
(6.2.194)
6.2 Differential Equations with Constant Coefficients 241
, 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)
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
242 Method of Multiple Scales
x,~ + '""
r\2
x,- -_ - (2 l'r\A
U, 3B e
, + -2
2 -i<1/')
e iOl + NST
.. .+ C.C. (6.2.207)
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
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)
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 243
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) -
where a and q> are slowly varying functions of t and Yn{t) describe the higher
hannonics.
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
244 Method of Multiple Scales
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
(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
(6.3.13)
A, =0 (6.3.14)
-'I
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
6.3 Struble's Method 245
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)
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)
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 ,
246 Method of Multiple Scales
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).
where, t) =et is the slow time scale characterizing the slow variations
introduced by the weak excitation.
6.3 Struble's Method 247
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)
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
$(tl)=O't l +9(tl), (6.3.45)
A = Fo sin$. (6.3.47)
" 2
Putting further
a =Acos$, b =Asin$, (6.3.48)
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)
c>o
Figure 6.7. The solution curves in the ab-plane for the forced Duff'mg
oscillator in the case (1 > O.
6.3 Struble's Method 251
Figure 6.8. The solution curves in the ab-plane for the forced Duffmg
osciUator in the case 0' is sufficiently negative.
252 Method of Multiple Scales
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.)
where
x=et, e«1 (6.4.2)
and
~ ~
where
ddx(}j =',.(x-)
I\. i = 1, 2 ,
(6.4.6)
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)
Au =±i~qo(X) }
a b (6.4.11)
~=A,Bo=A
(6.4.12)
254 Metbod of Multiple Scales
fo
I
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
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
t
, =0 : =0, ax.
at =0
XI (6.4.25)
t =0 : X 2 =0, ~l =0 (6.4.27)
etc.
Solving equation (6.4.22), we have
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)
256 Method of Multiple Scales
(6.4.33)
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.
(6.4.38)
J
t
(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
258 Method of Multiple Scales
(6.4.43)
(6.4.45)
etc.
One obtains from equations (6.4.42) and (6.4.43)-
(6.4.47)
(6.4.48)
or
(6.4.49)
and
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
260 Metbod of Multiple Scales
2A2
-- 1A2
- COS (/1' +qJl ) COS (/2' +qJ2 )
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 )
(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)
6.4 Differential Equations with Slowly Varying Coefficients 261
:'I' , :'I'
o-q21 (OJ u"q21 2 (01 o-q21
-,-,-+-, -,-,-+ - ' - ,- , +q'l
:l'
=--,
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)
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)
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)
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 •
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:
O(e) . [g, a YI + a(~) dyl] g' =-2g' a yo _ g" dyo _ a(~) dyo - b(~)y
2 2
(6.5.8)
6.5 Generalized Multiple-Scale Method 263
(6.5.9)
etc.
Let us choose for convenience
f
x
a B~ + (a' - b) Bo =0 (6.5.19)
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.
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)
etc.
Solving equation (6.5.29), we have
6.6 Applications to Solid Mechanics 265
(6.5.31)
a:
Using (6.5.31), equation (6.5.30) becomes
a 2 .::l.,
1
+~ = -(A' +A)+(B' -B) e-1). (6.5.32)
B = /3e s . (6.5.36)
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
266 Method of Multiple Scales
* 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).
268 Method of Multiple Scales
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)
a
o x
(a)
(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
where e is a small quantity, which we shall identify below, and then equation
(6.6.3) gives, upon neglecting the nonlinear modal coupling,
270 Metbod of Multiple Scales
(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, ....
6.6 Applications to Solid Mechanics 271
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.
Using (6.6.8) and (6.6.9), equation (6.6.6) gives
(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)
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),
R
o
=1_ 2n2
~n
(i) p' 2
I
where
where
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),
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
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
276 Method of Multiple Scales
~, ;;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
_ 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 ~ ~
* if the temperature is nearly uniform in the flow field, then the quantity
;j2 ,
dxdx [(p-c:.p)8;j]
I )
* 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).
278 Method of Multiple Scales
ap +v.(pv)=o (6.7.16)
at
:i(pv)+v,(pvv)+Vp=iL [V2V+~ v(v.v)J. (6.7.17)
(6.7.19)
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
6.7 Applications to Fluid Dynamics 279
where
~ and "l, respectively, characterize the source field and the sound field.
Note that
V=V~+M~Vry. (6.7.24)
(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 ~-~
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
280 Metbod of Multiple Scales
(6.7.29)
where
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)
_ 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 Applications to Fluid Dynamics 281
(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
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
* 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
: = O. (6.7.36)
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)
284 Method of Multiple Scales
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.
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 Applications to Fluid Dynamics 285
(6.7.46)
(6.7.48)
(6.7.49)
(6.7.50)
(6.7.51)
;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)
286 Method of Multiple Scales
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
Looking for steady, progressing, and localized wave solutions of the form
tP(x,t) = tP(;), ;=x-Ut, (6.7.60)
from which
(6.7.64)
/ = f dq,
q,~I- aq, .
(6.7.65)
Putting
I(
a=- (6.7.66)
-3U'
(6.7.65) becomes
(6.7.69)
(6.7.70)
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
ah, + C
at 0
(1 + ~.!i.)
2h
ah, + r a h, = 0
ax
3
ax 3
(6.7.72)
o
where
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.
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)
Using (6.7.77) and (6.7.78), equation (6.7.75) gives
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 Applications to Fluid Dynamics 291
(6.7.84)
(6.7.86)
(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)
(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)
292 Method of Multiple Scales
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)
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)
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)
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 )
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
VII + VI xu =0 (6.7.109)
JAi(s) ds
x/(31)"'
VI = B(r) (6.7.110)
water
(6.7.114)
(6.7.115)
(6.7.116)
where 'IlqJ denotes the perturbation in the velocity potential of the liquid.
'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)
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)
298 Method of Multiple Scales
-.
2
I -+---+K
. aA , I a AI
ar 2 a(
IA
I
12 A - 0
I
(6.7.136)
Putting
(6.7.137)
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
6.7 Applications to Fluid Dynamics 299
.... .. -- ...
, ,,
,, ,,
'" '"
. o¢J (112 1 1 ) 2
I at +"21 02¢J
O~2 +/( ¢J - ¢Jo ¢J =0, (6.7.148)
PI +(paxt =0 (6.7.150)
1,1,1 ( )
a l +- a; +--, p; - - pxx
2 8p· 4p
- /( p - Po =O. (6.7.151)
(6.7.152)
6.7 Applications to Fluid Dynamics 301
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
L(;}O' (6.7.156)
302 Metbod of Multiple Scales
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
-2/CA" ~ da,(1
dr - A- II')} e iKx +c.c.
-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)
6.7 Applications to Fluid Dynamics 303
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.
where
e =(2tr)2 ~
A' p'g"
T' being the surface tension, p' being the density of water.
304 Method of Multiple Scales
Let us look for a travelling wave, and introduce a new independent variable -
~=x-ct (6.7.169)
, )-3/2
11 =cfP~ -"21 (fP~' + fP;, ) + k·11~~
, (
1+ 11~ (6.7.172)
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)
(6.7.181)
y ~ -00 : 1P2y ~ 0 (6.7.182)
1 2
Y = 0 : 1P3Y + 1P2 yy 111 + IPlyy112 +"2 IPlyyy111
(6.7.185)
y~ -00: 1P3y ~ O. (6.7.186)
Let
11 1 =Acos~. (6.7.187)
(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),
and then
306 Method of Multiple Scales
(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),
11~I) =Acos~ }
11~2) = Bcos 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)
Using (6.7.196) and (6.7.197), one obtains, from (6.7.179), (6.7.180), and
(6.7.182),
6.8 Applications to Plasma Physics 307
(6.7.199)
1
A +4coBc, -8kok l B =0,
2 A A
- - Co (6.7.200)
2
from which
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.
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
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
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
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)
(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
~ [IAI 2 ] + ~ [C(k)IAI
2
] =0, (6.8.19)
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 -
(i) conservation of mass -
an d
a;+ ax (nv)=v p (6.8.21)
(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 -
312 Method of Multiple Scales
_
~-e
1f2
[
f Ao(x')
X
dx' ]
-t ,
(6.8.24)
and take
(6.8.25b)
(6.8.29)
(6.8.30)
tPo == O. (6.8.31)
dn, 1 d ( ) (6.8.36)
-ag+ A d~ nov, +n,vo =0
o
(6.8.39)
This leads to
or
AO = 1 + vO • (6.8.40)
(6.8.43)
(6.8.44)
314 Method of Multiple Scales
Putting,
-v J-1-[.3..+{I-'o)]dTl
¢J. =g(1J) 1/11' g(1J)=e 0
2. -'0 '0 (6.8.46)
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)
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.
316 Method of Multiple Scales
6.9. Exercises
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.
and consider the cases (00 '" A and (00 '" A separately.
Miscellaneous Perturbation
Methods
etc.
Successive integrations of equation (7.1.5) along with use of the boundary
conditions (7.1.2) lead to
(7.1.7)
..
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).
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)
Now, assuming that the inverse operator L~I is defined, we can rewrite
equation (7.2.3) in the form
(7.2.5)
Upon operating on (7.2.8) by L o' and using equation (7.2.4), we find that
(7.2.12)
j
eA,,(r-r.), t > t',
G(t,t') = (7.2.13)
0, t < t'.
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)
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)
(7.2.20)
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~ ,
(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
326 Miscellaneous Perturbation Methods
exp
0
[( -1
1) ]
T] =(COST] SinT])
,
0 - sin T] cos T]
where
~
(7.2.30)
From equation (7.2.25), it follows that the mean electric field (E) is
governed by
(7.2.31)
(7.2.32)
7.2 A Perturbation Method for Linear Stochastic Differential Equations 327
)
(7.2.36)
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)
(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.
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
7.2 A Perturbation Method for Linear Stochastic Differential Equations 329
o (7.2.42)
(u(O)) == I
Putting
(u(t)) == e-/l l v(t) (7.2.43)
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
~
_~I[COS,/"'~
v(t) = e' - 4" t + ~11'
A/2 ~]
_ ~ Si.,/"' - 4" t. (7.2.48)
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).
7.2 A Perturbation Method for Linear Stochastic Differential Equations 331
(~ +}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)
4 This amounts to expressing the triple correlations in terms of pair correlations while
neglecting the residual fourth-order cumulants.
332 Miscellaneous Perturbation Methods
(7.2.56)
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
iP" = (xiPr }
iP(O) =1, iP(oo)=O
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
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,
Proc. Roy. Soc. (London) A211, 564.
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
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,
Proc. Roy. Soc. (London) A299, 6.
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
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
Chapter 3
2. x =eacos(mt+q»+O(e 3)
m=~;-m2 _e 2 :~2 ~;_m2 +...
3. u = e llt q{t) where
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
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
F I
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
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 5.17 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 6.16 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.
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.