Perturbation Methods (Wiley Classics Library)

Perturbation Methods Perturbation Methods ALI HASAN NAYFEH WILEYVCH Wiley-VCH Verlag GmbH & Co. KGaA All books pu...

Author: Ali Hasan Nayfeh

226 downloads 2145 Views 14MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Perturbation Methods

Perturbation Methods

ALI HASAN NAYFEH

WILEYVCH

Wiley-VCH Verlag GmbH & Co. KGaA

All books publiskd by Wiley-VCH are carefully produced. Nev&les, authors, editors, and publisher do not warrant the information mtainedinthesebooks,inchldmg thisbook, to be h e ofamrs.

Readas~advisedtokeepinmindthatstatements,data,illUStlati~, procedural details or other items may inadvertently be inaccurate. Library of Congress Card No.: Applied for

Blitlsb Ilbrary C a ~ - i n - ~ b l i c a t i oIhtn: n

A cataloguereun-d for thisbook is available from the British Library

Bibliographic informatioa published by Die Deutscbe Bibliothek Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at
All rights reserved (includingthose of translation into Otha languages). No part ofthis book may be rqaoducedin any form- nor transmittedor translated into machine language without written pamissiOnfimm the publishers. Registered names, trademarks,etc.used in this book, even whtn not specifically marked as such,are not to be considered unprotectedby law.

Printed in the Federal Republic of Germany Printed on acid-fkee paper Printing Straw GmbH, MGrlenbach Bookbinding J. Schaffer wlchbindeaei GmbH, Grtinstadt ISBN-13: 978-0-471-3991 7-9

ISBN-10:0 4 71-399 17-5

Many of the problems faced today by physicists, engineers, and applied mathematicians involve difficulties, such as nonlinear governing equations, variable coefficients, and nonlinear boundary conditions at complex known or unknown boundaries, which preclude their solutions exactly. To solve these problems we are forced to resort to a form of approximation, a numerical solution, or a combination of both. Foremost among the approximation techniques is the systematic method of perturbations (asymptotic expansions) in terms of a small or a large parameter or coordinate. This book is concerned only with these perturbation techniques. According to these perturbation techniques, the solution of the full problem is represented by the first few terms of a perturbation expansion, usually the first two terms. Although these perturbation expansions may be divergent, they can be more useful for a qualitative as well as a quantitative representation of the solution than expansions that are uniformly and absolutely convergent. It is the rule rather than the exception that the straightforward (pedestrian) expansions in powers of a parameter have limited regions of validity and break down in certain regions called regions of nonuniformity. To render these expansions uniformly valid, investigators working in different branches of physics, engineering, and applied mathematics have developed a number of techniques. Some of these techniques are radically different, while others are different interpretations of the same basic idea. The purpose of this book is to present in a unified way an account of some of these techniques, pointing out their similarities, differences, and advantages, as well as their limitations. The different techniques are described using examples which start with model simple ordinary equations that can be solved exactly and progress toward complex partial differential equations. The examples are drawn from different branches of physics and engineering. For each example a short description of the physical problem is first presented. The different techniques are described as formal procedures without any attempt at justifying them rigorously. In fact, there are no rigorous V

vi

PREFACE

mathematical justifications available yet for the expansions obtained for some of the complex examples treated in this book. At the end of each chapter, a number of exercises have been included, which progress in complexity and provide further references. The reader need not understand the physical bases of the examples used to describe the techniques, but it is assumed that he has a knowledge of basic calculus as well as the elementary properties of ordinary and partial differential equations. Chapter 1 presents the notations, definitions, and manipulations of asymptotic expansions. The sources of nonuniformity in perturbation expansions are classified and discussed in Chapter 2. Chapter 3 deals with the method of strained coordinates where uniformity is achieved by expanding the dependent as well as the independent variables in terms of new independent parameters. Chapter 4 describes the methods of matched and composite asymptotic expansions; the first method expresses the solution in terms of several expansions valid in different regions but related by matching procedures, while the second method expresses the solution in terms of a single expansion valid everywhere. In Chapter 5 the idea of fast and slow variables is used in conjunction with the variation of parameters method to study the slow variations of the amplitudes and phases of weakly nonlinear waves and oscillations. The methods of Chapter 3, 4, and 5 are generalized in Chapter 6 into one of three versions of the method of multiple scales. Chapter 7 treats available methods for obtaining asymptotic solutions of linear ordinary and partial differential equations. My first technical debt is to Dr. W. S. Saric and to my brothers Dr. Adnan Nayfeh and Mr. Munir Nayfeh for their comments and encouragement throughout the writing of this book. I am indebted to several colleagues for helpful comments and criticism, including in particular Drs. D. T. Mook, D. P. Telionis, A. A. Kamel, and B. H. Stephan and Messers 0. R. Asfar and M. S. Tsai. This book would not have been written without the patience and encouragement of my wife, and the insistence of my parents Hasan and Khadrah, in spite of their illiteracy, that I acquire a higher education. Therefore I dedicate this book to my parents and wife. ALI HASANNAYFEH Blacksburg. Virginia May I972

1. Introduction

1

Parameter Perturbations, 1 1.1.1. A n Algebraic Equation, 2 I . I .2. The uan der Pol Oscillator, 3 1.2. Coordinate Perturbations, 4 I .2. I . The Bessel Equation of Zeroth Order, 5 I .2.2. A Simple Example, 6 1.3. Order Symbols and Gauge Functions, 7 1.4. Asymptotic Expansions and Sequences, 9 I .4.I . Asymptotic Series, 9 I .4.2. Asymptotic Expansions, 12 1.4.3. Uniqueness of Asymptotic Expansions, 14 1.5. Convergent versus Asymptotic Series, 15 1.6. Nonuniform Expansions, 16 1.7. Elementary Operations on Asymptotic Expansions, 18 Exercises, 19 1.1.

2.

Straightforward Expansions and Sources of Nonuniformity

2.1.

2.2.

23

Infinite Domains, 24 2. I . I. The Du$ing Equation, 24 2.1.2. A Model f o r Weak Nonlinear Instability, 25 2.1.3. Supersonic Flow Past a Thin Airfoil, 26 2.1.4. Small Reynolds Number FIow Past a Sphere, 28 A Small Parameter Multiplying the Highest Derivative, 31 2.2.1. A Second-Order Example, 31 2.2.2. High Reynolds Number Flow Past a Body, 33 2.2.3. Relaxation Oscillations, 34 2.2.4. Unsymmetrical Bending of Prestressed Annular Plates, 35

vii

viii

CONTENTS

Type Change of a Partial Differential Equation, 37 2.3.1. A Simple Example, 38 2.3.2. Long Waves on Liquids Flowing down Incline Planes, 38 2.4. The Presence of Singularities, 42 2.4.1. Shqt in Singularity, 42 2.4.2. The Earth-Moon-Spaceship Problem, 43 2.4.3. Thermoelastic Surface Waves,45 2.4.4. Turning Point Problems, 48 2.5. The Role of Coordinate Systems, 49 Exercises, 52

2.3.

3. The Method of Strained Coordinates

3.1. The Method of Strained Parameters, 58 3.1.1. The Lindstedt-Poincard Method, 58 3.1.2. Transition Curvesfor the Mathieu quation, 60 3.1.3. Characteristic Exponents for the Mathieu Equation (Whittaker’s Method), 62 3.I .4. The Stability of the Triangular Points in the Elliptic Restricted Problem of Three Bodies, 64 3.1.5. Characteristic Exponents for the Triangular Points in the Elliptic Restricted Problem of Three Bodies, 66 3.I .6. A Simple Linear Eigenvalue Problem, 68 3.1.7. A Quasi-Linear Eigenvalue Problem, 7 1 3.I .8., The Quasi-Linear Klein-Gordon Equation, 76 3.2. Lighthill’s Technique, 77 3.2.1. A First-Order Direrential Equation, 79 3.2.2. The One-Dimensional Earth- Moon-SpaceshipProblem, 82 3.2.3. A Solid Cylinder Expanding Uniformlyin Still Air, 83 3.2.4. Supersonic Flow Past a Thin Airfoil, 86 3.2.5. Expansions by Using Exact Characteristics-Nonlinear Elastic Waves, 89 3.3. Temple’s Technique, 94 3.4. Renormalization Technique, 95 3.4.1. The Dufing Equation, 95 3.4.2. A Model for Weak Nonlinear Instability, 96 3.4.3. Supersonic Flow Past a Thin Airfoil, 97 3.4.4. Shijit in Singularity, 98 3.5. Limitations of the Method of Strained Coordinates, 98 3.5.1. A Modelfor Weak Nonlinear Instability, 99

56

CONTENTS

3.5.2.

A Small Parameter Multiplying the Highest Derivative,

ix

100 3.5.3. The Earth- Moon-Spaceship Problem, 102 Exercises, 103

4. The Methods of Matched and Composite Asymptotic Expansions 4. I .

110

The Method of Matched Asymptotic Expansions, 1 1 1 4. I . I . Introduction-Prandtl‘s Technique, 111 4.1.2. Higher Approximations and Rejined Matching Procedures, 114 4.1.3. A Second-Order Equation with Variable Coeficients, 122 4.1.4. Reynolds’ Equation f o r a Slider Bearing, 125 4.1.5. Unsymmetrical Bending of Prestressed Annular Plates,

128

4.1.6. Thermoelastic Surface Waves, 133 4.1.7. The Earth-Moon-Spaceship Problem, 131 4.1.8. Small Reynolds Number Flow Past a Sphere, 139 4.2. The Method of Composite Expansions, 144 4.2. 1. A Second-Order Equation with Constant Coeficients, 4.2.2.

145

A Second-Order Equation with Variable Coeficients,

148

4.2.3. An Initial Value Problem f o r the Heat Equation, 150 4.2.4. Limitations of the Method of Composite Expansions,

153 Exercises. 154

5. Variation of Parameters and Methods of Averaging 5.1. Variation of Parameters, 159 5.1 . I . Time-Dependent Solutions of the Schrodinger Equation, 160 5.I .2. A Nonlinear Stability Example, 162 5.2. The Method of Averaging, 164 5.2.1. Van der Pol’s Technique, 164 5.2.2. The Krylov-Bogoliubov Technique, 165 5.2.3. The Generalized Method of Averaging, 168 5.3. Struble’s Technique, 171 5.4. The Krylov-Bogoliubov-Mitropolski Technique, 174 5.4. I . The Dufiing Equation, I75 5.4.2. The van der Pol Oscillator, 116 5.4.3. The Klein-Gordon Equation, 178

159

The Method of Averaging by Using Canonical Variables, 179 5.5.1. The Dufing Equation, 182 5.5.2. The Mathieu Equation, 183 5.5.3. A Swinging Spring, 185 5.6. Von Zeipel’s Procedure, 189 5.6.I . The Dufing Equation, 192 5.6.2. The Mathieu Equation, 194 5.7. Averaging by Using the Lie Series and Transforms, 200 5.7.I. The Lie Series and Transforms, 201 5.7.2. Generalized Algorithms, 202 5.7.3. SimpliJied General Algorithms, 206 5.7.4. A Procedure Outline, 208 5.7.5. Algorithmsfor Canonical Systems, 2 12 5.8. Averaging by Using Lagrangians, 216 5.8.1. A Model f o r Dispersive Waves, 21 7 5.8.2. A Model f o r Wave- Wave Interaction, 219 5.8.3. The Nonlinear Klein-Gordon Equation, 221 Exercises, 223 5.5.

6. The Method of Multiple Scales

228

6.1. Description of the Method, 228 6. I . I . Many- Variable Version (The Derivatiue-Expansion Procedure), 236 6.1.2. The Two- Variable Expansion Procedure, 240 6.1.3. Generalized Method-Nonlinear Scales, 241 6.2. Applications of the Derivative-Expansion Method, 243 6.2.1. The Dufing Equation, 243 6.2.2. The van der Pol Oscillator, 245 6.2.3. Forced Oscillations of the van der Pol Equation, 248 6.2.4. Parametric Resonances-The Mathieu Equation, 253 6.2.5. The van der Pol Oscillator with Delayed Amplitude Limiting, 257 6.2.6. The Stability of the Triangular Points in the Elliptic Restricted Problem of Three Bodies, 259 6.2.7. A Swtnging Spring, 262 6.2.8. A Model for Weak Nonlinear Instability, 264 6.2.9. A Model for Wave- Wave Interaction, 266 6.2.10. Limitations of the Derivative-Expansion Method, 269 6.3. The Two-Variable Expansion Procedure, 270 6.3.1. The Du$’ng Equation, 270 6.3.2. The van der Pol Oscillator, 272

xi

CONTENTS

6.3.3. The Stability of the Triangular Points in the Elliptic Restricted Problem of Three Bodies, 275 6.3.4. Limitations of This Technique, 275 6.4. Generalized Method, 276 6.4.1. A Second-Order Eqtration with Variable Coeflcients, 276 6.4.2. A General Second-Order Equation with Variable Coeficients, 280 6.4.3. A Linear Oscillator with a Slowly Varying Restoring Force, 282 6.4.4. An Example with a Turning Point, 284 6.4.5. The Dufing Equation with Slowly Varying Coeficients, 286 6.4.6. Reentry Dynamics, 291 6.4.7. The Earth-Moon-Spaceship Problem, 295 6.4.8. A Model for Dispersive Waves, 298 6.4.9. The Nonlinear Klein-Gordon Equation, 301 6.4.10. Advantages and Limitations of the Generalized Method, 302 Exercises, 303

7. Asymptotic Solutions of Linear Equations 7. I .

308

Second-Order Differential Equations, 309 7. 1. 1. Expansions Near an Irregular Singularity, 309 7.1.2. An Expansion of the Zeroth-Order Bessel Function for Large Argument, 3 12 7.1.3. Liouville’s Problem, 314 7.1.4. Higher Approximations for Equations Containing a Large Parameter, 3 15 7. I .5. A Small Parameter Multiplying the Highest Derivative, 317 7.1.6. Homogeneous Problems with Slowly Varying Coeflcients, 3 18 7.I . 7. Reentry Missile Dynamics, 320 7.1.8. Inhomogeneous Problems with Slowly Varying Coeficients, 32 1 7.1.9. Successive Liouville-Green ( WKB) Approximations, 324 7.2. Systems o f First-Order Ordinary Equations, 325

xii

CONTENTS

Expansions Near an Irregular Singular Point, 326 Asymptotic Partitioning of System of Equations, 327 Subnormal Solutions, 331 Systems Containing a Parameter, 332 Homogeneous Systems with Slowly Varying Coeficients, 333 7.3. Turning Point Problems, 335 7.3.1. The Method of Matched Asymptotic Expansions, 336 7.3.2. The Lmger Transformation, 339 7.3.3. Problems with Two Turning Points, 342 7.3.4. Higher-Order Turning Point Problems, 345 7.3.5. Higher Approximations, 346 7.3.6. An Inhomogeneous Problem with a Simple Turning Point-First Approximation, 352 7.3.7. An Inhomogeneous Problem with a Simple Turning Point-Higher Approximations, 354 7.3.8. An Inhomogeneous Problem with a Second-Order Turning Point, 356 7.3.9. Turning Point Problems about Singularities, 358 7.3.10. Turning Point Problems of Higher Order, 360 7.4. Wave Equations, 360 7.4.1. The Born or Neumann Expansion and The Feynman Diagrams, 361 7.4.2. Renormalization Techniques, 367 7.4.3. Rytov’s Method, 373 7.4.4. A Geometrical Optics Approximation, 374 7.4.5. A Uniform Expansion at a Caustic, 377 7.4.6. The Method of Smoothing, 380 Exercises, 382 7.2.1. 7.2.2. 7.2.3. 7.2.4. 7.2.5.

References and Author Index

387

Subject Index

417

Perturbation Methods ALI HASAN NAYFEH Copyright Q 2004 WILEY-VCH Valag GmbH & Co. KGaA

CHAPTER 1

Introduction Most of the physical problems facing engineers, physicists, and applied mathematicians today exhibit certain essential features which preclude exact analytical solutions. Some of these features are nonlinearities, variable coefficients, complex boundary shapes, and nonlinear boundary conditions at known or, in some cases, unknown boundaries. Even if the exact solution of a problem can be found explicitly, it may be useless for mathematical and physical interpretation or numerical evaluation. Examples of such problems are Bessel functions of large argument and large-order and doubly periodic functions. Thus, in order to obtain information about solutions of equations, we are forced to resort to approximations, numerical solutions, or combinations of both. Foremost among the approximation methods are perturbation (asymptotic) methods which are the subject of this book. According to these techniques, the solution is represented by the first few terms of an asymptotic expansion, usually not more than two terms. The expansions may be carried out in terms of a parameter (small or large) which appears naturally in the equations, or which may be artificially introduced for convenience. Such expansions are called parameter perturbations. Alternatively, the expansions may be carried out in terms of a coordinate (either small or large); these are called coordinate perturbations. Examples of parameter and coordinate expansions and their essential characteristics are presented in Sections 1.1 and 1.2. To formalize the concepts of limits and error estimates, definitions of order symbols and other notations are introduced in Section 1.3. Section 1.4 contains definitions of an asymptotic expansion, an asymptotic sequence, and a power series, while Section 1.5 presents a comparison of convergent and asymptotic series. Uniform and nonuniform asymptotic expansions are then defined in Section 1.6. A short summary of operations with asymptotic expansions is given in Section 1.7. 1.1. Parameter Perturbations

Many physical problems involving the function u(x, e) can be represented mathematically by the differential equation L(u, x, E) = 0 and the boundary

1

2

INTRODUCTION

condition B(u, E) = 0, where z is a scalar or vector independent variable and E is a parameter. In general, this problem cannot be solved exactly. However, if there exists an E = e0 (e can be scaled so that E, = 0) for which the above problem can be solved exactly or more readily, one seeks to find the solution for small E in, say, powers of E; that is u(z; E) = uo(z)

+ EUl(Z) +

E2Uz(z)

+- -

(1 .l. 1)

where u, is independent of E and uo(z) is the solution of the problem for E = 0. One then substitutes this expansion into L(u, 2, 6) = 0 and B(u, E) = 0, expands for small E, and collects coefficients of each power of E. Since these equations must hold for all values of E, each coefficient of c must vanish independently because sequences of E are linearly independent. These usually are simpler equations governing u,, which can be solved successively. This is demonstrated in the next two examples. 1.1.1. AN ALGEBRAIC EQUATION Let us consider first the solution of the algebraic equation 24

for small E. If

E

= 0, u = 1. For u =

1

=1 E

+

(1. I .2)

EU3

small, but different from zero, we let

+ EU1 +

E2U2

+

E3U3

+-

(1. I .3)

* *

and (1.1.2) becomes

+ e2uZ+ c3u3+ . -

6u1

*

+

= ~ ( 1 cul

+ c2u2 + e3u3 + -

Expanding for small E, we rewrite (1.1.4) as EU1

+ c2uz +

E3U3

+

* * *

+

- 1)

ez(u2

+

= ~ [ 1 3 ~ ~ 3 E12 ( U 2

Collecting coefficients of like powers of E(U1

*

- 3241)

- 3242 - 3U12)

+

Since this equation is an identity in E , each coefficient of pendently. Thus 24,-l=O

- 3u, = 0 - 3u12 = 0

u2

Us

- 324,

.]

(1.1.5)

=0

(1.1.6)

*

we have

E,

e3(U3

+ U12) +

(1.1.4)

-)3

a

E

*

vanishes inde(1.1.7)

(1.1.8) ( 1.1.9)

The solution of (1.1.7) is u1

=1

(1.1.10)

1.1.

PARAMETER PERTURBATIONS

3

Then the solution of (1.1.8) is u2 = 324, = 3

(1.1.1 1)

and the solution of (1.1.9) is u3

Therefore (1.1.3) becomes u =1

= 3u2

+ 3u12 = 12

(1.1.12)

+ E + 3e2 + 12€3 + -

( I . 1.13) where the ellipsis dots stand for all terms with powers of E" for which n 2 4. Thus (1.1.13) is an approximation to the solution of (1.1.2), which is equal to 1 when E EE 0. *

1.1.2. THE VAN DER POL OSCILLATOR As a second example, we consider van der Pol's (1922) equation

-+ u

for small

t.

If

c

d' u du = E(1 - 2)dt2 dt = 0 this equation reduces to

with the general solution

(1.1.14)

d2u -+u=o dt2 u = acos (t

(1. l .15)

+ p)

(1.1.16)

where a and q are constants. To determine an improved approximation to the solution of (1.1.14), we seek a perturbation expansion of the form u(t; c) = uo(t) +

EUI(1)

+

E2U2(t)

+

*

(1.1.17)

*

where the ellipsis dots stand for terms proportional to powers of than 2. Substituting this expansion into (1.1.14), we have

=41

- (uo + EU1

+ EZU2 + . .

-+€! dt !!! + I E 2 ddt llZ

dt

E

greater

]

(1.1.18)

.. .

(1.1.19)

+ . ..

Expanding for small E , we obtain

= E(1

""1 +

du, - U l ) duo + 2[(1 - uo ) - 2u0u1dt dt 2

4

INTRODUCTION

Since u, is independent of E and (1.1.19) is valid for all small values of E, the coefficients of like powers of E must be the same on both sides of this equation. Equating the coefficients of like powers of E on both sides of (1.1.19), we have Coefficient of

EO

d2uo

(1.1.20)

-+uo=o

Coefficient of

dt2

E

d2ul

+ u1 = (1 - UO ) -duo dt 2

-

dt2

Coefficient of d2u2 -

dt2

+ u2 = (1 -

uo2)

( 1.1.21)

du, - 2u0u, duo dt

dt

(1.1.22)

Note that (1.1.20) is the same as (1.1.15) and its general solution is given by (1.1.16); that is uo = a cos ( t + 9) (1.1.23) Substituting for uo into (1.1.21) gives d2u,

-

dt2

+ u1 = -[I

- a2 cos2( t

+ 9)la sin ( t + p)

Using the trigonometric identity cos2 ( t

+ p) sin ( t + 9) = sin ( t + 9)+4 sin 3(t + p)

we can rewrite this equation as

+ p) + &a3sin 3(t + p)

(1.1.24)

+ p) - &a3 sin 3(t + q ~ )

(1.1.25)

d2u1 a3 - 4a +ul=sin ( t dt2

4

Its particular solution is ul=--

a3 4a t cos ( t 8

With uo and u1 known the right-hand side of (1.1.22) is known, and one can solve it for u2 in a similar fashion. The usefulness of such an expansion is the subject of this book. 1.2. Coordinate Perturbations

If the physical problem is represented mathematically by a differential equation L(u, x) = 0 subject to the boundary conditions B(u) = 0, where x

1.2.

COORDINATE PERTURBATIONS

5

is a scalar, and if u(x) takes a known form uoas x xo (xois scaled to 0 or co), one attempts to determine the deviation of u from uo for x near xo in terms of powers of z if zo = 0, or x-l if xo = 00. This technique is demonstrated by the next two examples. --f

1.2.1. THE BESSEL EQUATION OF ZEROTH ORDER We consider the solutions of x -d2Y + -dy+ dx2 dx

zy= 0

(1.2.1)

This equation has a regular singular point at x = 0,which suggests that a power series solution for y can be obtained using the method of Frobenius (e.g., Ince, 1926, Section 16.1). Thus we let m

y = 2 umxp+m

(1.2.2)

m=O

where the number p and the coefficients a, must be determined so that (1.2.2) is a solution of (1.2.1). Substituting (1.2.2) into (1.2.1) gives

2 ( p + m)(p + m W

m=O

-

+ 2 (p + M)a,2"+m--l+ 2 amZP+m+l= 0 m

m=O

or

m=O

2 ( p + m)2umxp+m-1+ 2 a,xp+m+l m

W

m=O

-0

=0

(1.2.3)

which can be written as p2uozp-'

+ ( p + 1)2alx' + 2 (p + m)2umzP+m-'+ 2

Replacing m by m it as p2aoxp-'

m

m

m=2

m=O

=O

+ 2 in the first summation of this equation, we can rewrite

+ (p + 1)2ulzp+ 2 [(p + m + 2)'~,+~+ am]zp+"+' = 0 a0

m=O

(1.2.4)

Since (1.2.4) is an identity in x , the coefficient of each power of x must vanish independently; that is p%, = 0 (1.2.5) (1.2.6) (p 1)"l = 0 (1.2.7) m = 0, 1 , 2 , . . (p m 2)2am+2 a, = 0,

+ +

+

+

.

6

INTRODUCTION

The first equation demands that p = 0 if a, # 0; then (1.2.6) gives a, = 0 and (1.2.7) gives (1.2.8) a,,

m = 1 , 2 , 3 , .. . ,

= 0,

(1.2.9) The solution thus obtained if a, = 1 is a Bessel function of zeroth order, and it is often denoted by J,. Thus

+...

(1.2.10)

Since the ratio of the nth term to the (n - 1)th term is -x2/(2n)$ and tends to zero as n + co irrespective of the value and sign of x, the series (1.2.10) for J, converges uniformly and absolutely for all values of x. An expansion valid for large values of x is obtained in Section 7.1.2 and compared with the above expansion in Section 1.5. 1.2.2. A SIMPLE EXAMPLE As a second example, we consider the solution of

dY -+y=dx

1 X

(1.2.11)

for large x. For large x we seek a solution in the form m

y = 2 amx-m m=l

(1.2.12)

Substituting this expansion into (1.2.1 1) yields

2 -mamx-m--l + 2 a,xm

00

m=l

m=2

Replacing m by m

+ (a1 - 1 ) P = 0

(1.2.13)

+ 1 in the second summation series, we can rewrite this

1.3.

ORDER SYMBOLS AND GAUGE FUNCTIONS

7

equation as (a, - 1 ) d

+ 2 (a,,, a,

=0

- ma,)x-m-'

m=l

(1.2.14)

Since this equation is an identity in x , the coefficient of each x - ~must vanish independently; that is a, = 1 ,

Hence

u2 = 1,

for m 2 1

= ma,

a,

u3 = 2 ! ,

(1.2.15)

a , = (n - l)!

a4 = 3 ! ,

and (1.2.12) becomes

( n - l)! +-+...

y = - +1 - + -l !+ - +2! . . . 3! x x2 2 3 2 4

-

Xn

( 1.2.16)

Since the ratio of the nth to the (n - 1)th term is ( n - 1)x-1 and it tends to infinity as n a3 irrespective of the value of x , the series (1.2.16) diverges for all values of x . In spite of its divergence, this series is shown in Section 1.4 to be useful for numerical calculations, and it is called an asymptotic series. 1.3. Order Symbols and Gauge Functions

Suppose we are interested in a function of the single real parameter E , denoted byf'(E). In carrying out our approximations, we are interested in the limit off(€) as E tends to zero, denoted by E -+ 0. This limit might depend on whether E tends to zero from below, denoted by E 0, or from above, denoted by E 10. If the limit off(€) exists (i.e., it does not have an essential singularity at E = 0 such as sin E - , ) , then there are three possibilities

-1

f(€1

--+

f(e)--A

f (€1

0

as E - + O , O < A

< co

(1.3.1)

00

In the first and last cases, the rate at which f ( c ) -+ 0 and f ( ~4 ) 00 is expressed by comparing f ( ~ with ) known functions called gauge functions. The simplest and most useful of these are . . . , E - - n , . . . ,E - 2 ,

€-I,

1,

E , €2,.

. . ,E n , . . .

In some cases these must be supplemented by log E-,, log (log E-'),

e'

-1

, e-'

-1

, and so on

Other gauge functions are sin E , cos E , tan

Q,

sinh c , cosh

E,

tanh

E,

and so on

8 as

INTRODUCTION

The behavior of a functionf(E) is compared with a gauge function g ( e ) E -+ 0, employing either of the Landau symbols: 0 or 0.

The Symbol 0 We write

f ( ~= ) 0[g(~)Jas

E

-0

(1.3.2)

if there exists a positive number A independent of E and an co > 0 such that

If(4l s A lg(4

for all 1.1

I €0

(1.3.3)

This condition can be replaced by

For example, as E

-

(1.3.4) 0

sin E = O(E),

sin c2 = O(c2)

sin 7~ = O ( E ) ,

sin 2~ - 2~ = 0(c3)

cos E = 0(1),

1 - cos E = O(E2)

Jo(E)

= 0(1),

Jo(E)

- 1 = O(c2)

sinh E = O ( E ) ,

cosh E = O(1)

tanh E = O(E),

tan E = O(E)

coth E = O(E-'),

cot E = O(6-I)

Iff is a function of another variable x in addition to c , and g ( x , G) is a gauge function, we also write f(z,

c)

= O [ g ( x ,E ) ]

as

(1.3.5)

E -+0

if there exists a positive number A independent of c and an

If@.,

€11I A I&,

E~

€11 for all I 4 I co

> 0 such that (1.3.6)

If A and c0 are independent of x, the relationship (1.3.5) is said to hold uniformly. For example sin (x

+ c ) = O(1) = O[sin (x)]

uniformly as

E -+

while e"'

- 1 = O(c) nonuniformly as

JFE -:/, = ~ ( r )nonuniformiy as

E

-+

E -+

0

o

0

1.4. ASYMPTOTIC EXPANSIONS AND SEQUENCEs 9 The Symbol o

We write

f ( ~=) o [ ~ ( E ) ] as

E

if for every positive number 6, independent of

(1.3.7)

-0 E,

there exists an

c0 such

that

If<.>l I 6 Ig(4 for I 4 I €0

(1.3.8)

I A€)I

(1.3.9)

This condition can be replaced by

lim .&) Thus as

E -+

t-0

0

=0

sin E = o(l),

sin E~ = O(E)

cos E = o(E-1’2),

.TO(€) = o(E-1)

coth

cot E = o [ E - ( ” + ~ ) / ~ ]

for positive n

exp (-€-I)

for all n

1

E

= ~(e-~”),

- cos 3~ = o ( E ) ,

= o(E”)

Iff = f ( x , E ) and g = g(x, E), then (1.3.7) is said to hold uniformly if 6 and e0 are independent of x. For example while

sin (z

+

E)

= o ( E - ~ / ~uniformly ) as

e--Et- 1 = o(rl”)

JGT - Jx 1.4.

E -+

nonuniformly as

= o ( ~ ~ ‘ ~nonuniformly ) as

0

o o

E -+ E

-+

Asymptotic Expansions and Sequences

1.4.1. ASYMPTOTIC SERIES We found in Section 1.2.2 that a particular solution of 1

is

(1.4.1)

dx Y = - f1 - + l-!+ - +2 !. . .3! x x2 x3 x4

+-(n - l)! + . . . X”

(1.4.2)

which diverges for all values of x. To investigate whether this series is of any value for computing a particular solution of our equation, we determine the remainder if we truncate the series after n terms. To do this we note that

10

INTRODUCTION

a particular integral of our differential equation is given by y = ee2

["x-'ez

J-m

ds

(1.4.3)

which converges for negative x. Integrating (1.4.3) by parts, we find that y =1

+ e-z~-~x-zezd x = -1 + -1 + 2e-2[mx-3ez x

X

dx

x2

1 - -1 + + -2 + 3! eP2

x 1

22

23

l! x2

2! 9

+-(n - I)!

3!

=-+-+-+,+... x

2"

4- n! e-zs_',x-"-lez

dx (1.4.4)

Therefore if we truncate the series after n terms, the remainder is R , = n! e-zrmx-n-l ez dx

(1.4.5)

which is a function of n and x. For the series to converge, lim R, must be n-m

zero. This is not true in our example; in fact, R, + 00 as n + 03 so that the series diverges for all 2 in agreement with what we found in Section 1.2.2 using the ratio test. Therefore, if the series (1.4.2) is to be useful, n must be fixed. For negative x (1.4.6)

Thus the error committed in truncating the series after n terms is numerically less than the first neglected term, namely, the (n 1)th term. Moreover, as 1x1 -+ a)with n fixed, R, + 0. Therefore, although the series (1.4.2) diverges, for a fixed n the first n terms in the series can represent y with an error which can be made arbitrarily small by taking 121 sufficiently large. Such a series is called an asymptotic series of the Poincare' type (Poincart, 1892) and is denoted by

+

(1.4.7) on

-In general, given a series 2 (am/xnl),where a,,, is independent of x, we say m=O

1.4.

ASYMPTOTIC EXPANSIONS AND SEQUENCES

11

that the series is an asymptotic series and write

(1.4.8) if and only if n

-

(1.4.9)

The condition (1.4.9)can be rewritten as

-

n-1

(1.4.10)

As another example of an asymptotic series, we consider, after Euler

(1 754), the evaluation of the integral

(1.4.11) for large positive w . Since

- _0 Q

and

+x -1

(-1)mXm

m=O

if

om

(1.4.12)

x < o

(1.4.13)

Jo

f (0)=m=O 2:

(-l)"m!

(1.4.14)

om

Since the ratio of the mth to the ( m - 1)th term, -mu-', tends to infinity as m + co, the series (1.4.14)diverges for all values of o. To investigate whether (1.4.14)is an asymptotic series, we estimate the remainder if the series is 'truncated after the nth term. To d o this we note that

-+ -2 w +x m=O wrn

0 --

Hence

n-l

(-1)"s"

(-

(1.4.15)

1)nX"

on-yw

+

2)

( 1.4.16)

where lRnl =

m wn-l

0

xne-x

-dx

w + x

I <- in

n!

r x n e - . dx = Wn

(1.4.17)

12

INTRODUCTION

Hence the error committed by truncating the series after the first n terms is numerically less than the first neglected term, and (1.4.18)

Therefore the series (1.4.14) is an asymptotic series, and we write

f (0)

-I:

(-l)mm!

m=o

(1.4.19)

wm

1.4.2. ASYMPTOTIC EXPANSIONS One does not need to use a power series to represent a function. Instead, one can use a general sequence of functions an(€)as long as dn(e)

= O[S,-,(E)]

as

c

---f

(1.4.20)

0

Such a sequence is called an asymptotic sequence. Examples of such asymptotic sequences are E", en'3, (log E ) - ~ , (sin z ) ~ ,(cot E ) - ~ (1.4.21) In terms of asymptotic sequences, we can define asymptotic expansions. W

Thus, given 2 umdm(c) where a, is independent of m=o

B

and

am(€)is an asymp-

totic sequence, we say that this expansion is an asymptotic expansion and write y - 2 umdm(~) as a - 0 (1.4.22) m=O if and only if n-1 Y = 2 amdm(~) o[d,(e)] as E 0 (1.4.23) W

m=O

-

+

Clearly, a n asymptotic series is a special case of an asymptotic expansion. As an example of an asymptotic expansion that is not an asymptotic power series, we return to the integral (1.4.11). Following van der Corput (1962), we represent f ( w ) in terms of the factorial asymptotic sequence [(w I ) ( o + 2) . (w + n)]-l as w -+co. To d o this we note that

--

+

- -1- - 1 w

+x

0

- - 1_ w

1 -0

5

w(O

+

5)

- 1)

+ 1) + w(w + l)(0 + - 1) - l)(x - 2) + 40 + l)(w o ( w + 1) + 2) 0 ( w + l)(0 + 2)(w + X

5(2

w(w

5)

X

x(5

5(5

5)

(1.4.24)

1.4.

ASYMPTOTIC EXPANSIONS AND SEQUENCES

13

In general 0 (-l),z(z - 1). . (z + 1 - m) -2 0 + z m=O ( 0 + l)(o + 2) . . ( 0 + m) (- I)n+lz(z- 1) . . . (z - n) *

*

+ ( w + 1 x 0 + 2) . .

*

(0

+ n)(w +

5)

(1.4.25)

This equation can be proved by induction as follows. If (1.4.25) is valid for n, we show that it is valid for (n 1). To do this we note that

+

w -co

(- l)"z(z

+ z- 2

- 1) . . . (x + 1 - rn)

+ l)(w + 2 ) .

. (O + m) (-1)"+'z(z - 1) . . (2 - n) ( 0 + l)(w + 2) ' . ( w + n + 1) - (-1)"+1z(z - 1) . (x - n) ( 0 + 1 x 0 + 2 ) . . ( w + n + 1) (- l)n+'z(z - 1) . . . (z - n) (w + 1 x 0 2) . ( w + n)(w + 2)

m=O

(W

+

*

*

*

* *

*

+

+

* *

By combining the last two terms and extending the summation to n can rewrite this expression as 0 -0

n+l(-l)m2(z-

+z-zo

(w

+ (-

l ) . . . ( z + 1 - rn)

+ l)(w + 2). . . (w + m)

l)n+2

z(2 - 1) *

(w

+ 1, we

+ l)(w +

. (z - n 2) . . ( 0 n *

- 1)

+ + l)(o+ z)

(1.4.26)

Thus if (1.4.25) is true for n , (1.4.26) shows that it is true for n + 1. Since (1.4.25) is true for n = 0, 1 , and 2 according to (1.4.24), it is true for n = 3, 4, 5, . . . . Therefore it is true for all n. Multiplying (1.4.25) by exp (-2) and integrating from z = 0 to z = a, we have n

f ( w > = 2 arnarn(w) m=o

where a, = r z ( z

- 1) . . . (z

dm(w) = (-l)"[(w

+ Urn)

(1.4.27)

- m + 1)eP d x

(1.4.28)

+ l)(w + 2) . . . (w + rn)]-'

(1.4.29)

14 INTRODUCTION Since w is a positive large number

(1.4.31)

= Ian1 Idn(w)l

Thus the error committed by keeping the first n terms is numerically less than the nth term, hence (1.4.32)

Since dm(w) is an asymptotic sequence as w -+ co m

f ( w ) - 2 am6,(o) -0

as w 4 co

(1.4.33)

1.4.3. UNIQUENESS OF ASYMPTOTIC EXPANSIONS We have shown in the previous two sections that (1.4.34)

and f(w)

1

m-0

(-l)mJmx(x 0

(w

- 1) . . . (z + 1 - m)e-'

+ l)(w + 2).

. (w

dx

+ m)

as w - + m

(1.4.35)

Thus the asymptotic representation off(w) as w co is not unique. In fact, f(w) can be represented by an infinite number of asymptotic expansions --f

because there exists an infinite number of asymptotic sequences that can be

used in the representation. However, given an asymptotic sequence 6,(w),

the representation off(o) in terms of this sequence is unique. In this case

-

f(o)

W

1a,d,(o)

m=O

as

w

-+a0

(1.4.36)

where the amare uniquely given by

(1.4.37) an = lim 0 '

co

m=O

d,(w)

1,5.

CONVERGENT VERSUS ASYMPTOTIC SERIES

15

1.5. Convergent versus Asymptotic Series We found in Section 1.2.1 that one of the solutions of Bessel's equation x

is given by the series z2

Jo(x) = 1 - 22

-+

d2Y dx2

dY dx

+ xy = 0

X6 +- 22. 4 2 . 62 + . . . + ( - 1 y 42 24

2 2 .

(1.5.1)

X2n

22.42.

. . (2n)2

+

'

(1.5.2) which is uniformly and absolutely convergent for ail values of x. Another representation of Jo can be obtained if we note that the change of variable y = x-l/Zy, (1.5.3) transforms (1.5.1) into

(1.5.4) As x 4 co, this equation tends to (1.5.5)

with the solutions

y1 = &a"

(1.5.6)

y1 = &y2

(I. 5.7)

This suggests the transformation which gives (1.5.8)

This equation can be satisfied formally by yz = 1 - -

1 . I

8s

-

1 . 32 g 2 . 2 ! . x2

1.32'52

+

+

83.3!. x3

1 . 3 2 . 5 2 - 72 g 4 . 4! . x4

+ ...

(1.5.9)

By combining this series with that obtained by changing i into --i, obtain the following two independent solutions x-l/2

y'l'

y(2)

N

(u cos x

x-l'z(u sin x

+

t)

sin x)

- D cos x)

we

(1.5.10)

16

INTRODUCTION

where u= U =

To determine the connection between Jo(x) and these two independent solutions, we use the integral representation rr

7rJO(x)=J cos (zcos 8) d8

(J.S.12)

0

and obtain (see Section 7.1.2)

The ratio test shows that yz. u, and u , and hence the right-hand side of (1.5.13), are divergent for all values of x. However, for large x the leading terms in u and B decrease rapidly with increasing rank so that (1.5.13) is an asymptotic expansion for large x. For small x the fist few terms of (1.5.2) give fairly accurate results. In fact, the first 9 terms give a value for J0(2)correct to 11 significant figures. However, as x increases, the number of terms needed to yield the same accuracy increases rapidly. At x = 4, eight terms are needed to give an accuracy of three significant figures, whereas the first term of the asymptotic expansion (15.13) yields the same accuracy. As x increases further, an accurate result is obtained with far less labor by using the asymptotic divergent series (1 513). 1.6. Nonuniform Expansions

In parameter perturbations the quantities to be expanded can be functions of one or more variables besides the perturbation parameter. If we develop the asymptotic expansion of a function f ( x ; P), where x is a scalar or vector variable independent of E, in terms of the asymptotic sequence d,,,(c), we have f(z;P)

-2 m

m=O

a,,,(z)d,(~)

as

E +

0

(1.6.1)

where the coefficients a,,, are functions of x only. This expansion is said to be uniformly valid if (1.6.2a) R,(x;

E)

= O[d,4~)]

uniformly for all z of interest

(1.6.2b)

I .6.

NONUNIFORM EXPANSIONS

17

Otherwise the expansion is said to be nonuniform4 valid (it is often called a singular perturbation expansion). For the uniformity conditions (1.6.2) to hold, a,(x)d,(~) must be small compared to the preceding term for each m. Since am(€) = O [ ~ , - ~ ( E ) ] as e -+ 0, we require that am(x) be no more singular than a,-,(%) for all values of x of interest if the expansion is to be uniform. In other words, each term must be a small correction to the preceding term irrespective of the value of 2. A uniformly valid expansion is sin (x

+

E)

= sin x cos E

= sin x

+

E

+ cos x sin

6!

2

E2

cos z - - sin x - - cos 5 2! 3! t7

-- -sin x - - cos x E6

E

7!

+ ...

as

+ E4- sin 5 + E5- cos 4!

€-to

5!

(1.6.3)

Note that the coefficients of all powers of E are bounded for all values of hence a,(z) is no more singular than a,-,(x), and the expansion is uniformly valid as a consequence. For a nonuniformly valid expansion, let us expandf(x; E) = J F c for small 6 ; that is

z,

(1.6.4)

Each term of this expansion except the first is singular at z = 0 and is more singular than the preceding one, hence the expansion is not uniformly valid. It. breaks down near x = 0. An estimate of the size of the region of nonuniformity can sometimes be obtained by assuming two successive terms to be of the same order; that is

5 = 0(1) 22

gives x = o(E)

(1.6.5)

+

We can see this if we look at [l (~/x)]l'~ whose Taylor series expansion converges only if I E / z ~ is less than unity. For a second nonuniformly valid expansion, we consider the expansion of exp ( - - ~ f ) for small E. It possesses the following uniformly convergent series for all t (1.6.6)

18 INTRODUCTION It is clear that exp ( - c f ) can be approximately represented by a finite number of terms only if the combination ct is small. Since c is small, this means that t = O(1). When t is as large as 0(4), rt is not small and the truncated series breaks down. For example, if t = the first two terms give -1 for exp (-2). If the above series is truncated after a finite number of terms, it is found that the truncated series is satisfactory up to a certain value of t after which exp ( - d ) and the truncated series differ from each other by a quantity which exceeds the prescribed limit of accuracy. Adding more terms to the truncated series increases the value o f t to a new value t' for which this truncated series is satisfactory. However, for t > t' the difference between exp (- d)and the new truncated series again exceeds the prescribed accuracy. Thus all terms of the series are needed to give a satisfactory expansion for exp (- E t ) for all t. It is the rule rather than the exception that asymptotic expansions in terms of a parameter are not uniformly valid and break down in regions called regions of nonuniformity, which are sometimes referred to as boundary layers. Friedrichs (1955) discussed in a survey article the occurrence of these nonuniformities in different branches of mathematical physics. Most perturbation techniques were developed to render these nonuniform expansions uniformly valid. The sources of nonuniformity are discussed in Chapter 2 and the techniques of rendering nonuniform expansions uniform are taken up in the remaining chapters. 1.7. Elementary Operations on Asymptotic Expansions

To determine approximate solutions to differential and integral-differentia1 equations, we assume expansions, substitute them into equations, and perform on them elementary operations such as addition, subtraction, exponentiation, integration, differentiation, and multiplication. These operations are carried out without justification although some of these expansions are divergent. Conditions under which these operations are justified are discussed by van der Corput (1956), ErdClyi (1956), and de Bruijn (1958). Addition and subtraction are justified in general. For example, if

srf(z; a

6)

dx

-2

a , ( ~ ) r u , ( z )dx as a

E

-+

0

(1.7.3)

EXERCISES

Iff&;

c)

and q,,(e) are integrable functions of

E,

19

then

Multiplication is not justified in general because in the formal product of ~ a , ( z ) p , , ( ~and ) z b , , ( z ) y , , ( ~all ) products q,,(c)y,(e) occur, and it is generally not possible to arrange them so as to obtain an asymptotic sequence. In other words, multiplication is justified if the result is an asymptotic expansion. This is the case for all asymptotic sequences qn such that qnym either form an asymptotic sequence or possess asymptotic expansions. An important class of such sequences is the collection of powers of e. Thus if

then

(1.7.6)

where

n

(1.7.7)

Exponentiation is not justified in general. When it is not justified, it leads to nonuniformities. For example = &(1+

-l -e

1E2 - -+ .-

.)

as c + O (1.7.8) 8x2 is not justified when e/x = O(1) because its right-hand side is nonuniform in the region 2 = O(c). Similarly J

T

e

22

--1 - 1 - + 1+EX EX

€822

- €323 + .

*

-

as

E --+

0

(1.7.9)

is not justified when ex = O(1) because its right-hand side is nonuniform for large x. It is not justified in general to differentiate asymptotic expansions with respect to a variable such as x or with respect to the perturbation parameter e. As in exponentiation, when it is not justified, differentiation leads to nonuniformities. Exercises

1.1. Determine the order of the following expressions as c

+0 :

20

INTRODUCTION

1.2. Arrange the following in descending order for small

c:

In (In €-I), 1 , €112 In C-1, c In E-1, e-1lE, In c-1, €312, E, €2 In e-1 1.3. Expand each of the following expressions for small c and keep three terms: €2, 2'2,

(a) d1 - +2t - &4t (b) (1 + E C O S ~ ) - ~

+

+

1 +2E

- E2 m

(c) (1 EW1 E2W2)-2 (d) sin (s + ew1s + c 2 0 g )

(f) In

q

+

1.4. Let p = p0 e p l + e2p2 in h = (3/2)[1 - d l - 3p(1 - p ) ] , expand for small e, and keep three terms. 1.5. Find a second-order expansion for the solution of

x =1

+

€22,

E

<< 1

and compare it with the exact solution for c = 0 . 1 and c = 0.001. 1.6. Show that the asymptotic expansion for large x of

is 1

1 -

4!

2!

x2

3!

5!

)

-+ - - ... cosx 24 26

-

Does the series converge? Find an upper bound for the remainder and show that the remainder approaches zero faster than the last term in the expansion as x to. 1.7. If x = s - (c/3s2) - (3c2/10s">, find a two-term expansion for s when x = 0. 1.8. If x = s 4 2 - (2/3)~'/~)+ ( 4 2 / 5 ) ~ ~ $ / ~show , that the solution of dx/ds = 0 is

+

= +2

-

MP3

and then find x corresponding to this value. 1.9. If Y,(S)

and show that

s =1

+ EYIW +

€2Y&)

+

~

~

4

+ . .. = A

1 =s

+ cxl(s) + c2x2(s) + . . .

- € Xl ( l )

- €2[X2(1) - x;(l)xl(l)]

and then find ~ ~ ( 1 y,(l), ). and ~ ~ ( 1 ) .

)

+ ...

EXERCISES 1-10. Consider the equation Y'

+y =

y(0) = 1

cy2,

(a) Determine a three-term expansion for small (b) Show that the exact solution is y = epz[l

21

+ E(e-*

e.

- 1)I-l

(c) Expand this exact solution for small c and compare the result with (a). (d) Is this expansion valid for all x? 1.11. Determine a coordinate expansion for

Y" of the form

6

+A)Y

=o

1.12. Determine second-order (three-term) expansions for

(a) ii (b) ii

+ u = €2,

+u =

c

<< 1

-
with u(0) = a and u(0) = 0. Are these expansions uniformly valid? 1.13. Find a first-order (two-term)expansion valid for small E for the solution of

dx

s-

ds

=z

+ y'

dY

s- = -(2 ds

+ z)y ~ ( 1 )= 1

y(1) = e-l,

1.14. Use the asymptotic expansion (1.5.13) to show that the large zeros Jn(z)are

the solutions of cot

(t - *Tr)

= -

1 85

33 512t3

- + -+ . .

'

and then show that

t

= tTr(4n

1.15. Show that

1 + 3) + 2~r(4n + 3) + . . . with n an integer.

uii

+ + tu = r 2 U

is satisfied by the expansions (Levinson, 1969) =

as t

--

u =t 00,

t3

-6

+clt

+c2

-9tint

+of:)

+ (b, sin r + b2 cos t)t-1'2 + O(tp1)

where ci and bi are constants.

t of

22 INTRODUCTION 1.16. Show that the equation of Bellman (1955) (U)Z

=u

+u

is satisfied by the expansions (Levinson, 1969) ae-t

- qze-2:

u = h t 4 + &-t3

as t

+ 03,

+ 0(~-39 + c,t3 + c 2 r 2 + O(r2In2r )

In t

where a and ci are constants.

Perturbation Methods ALI HASAN NAYFEH Copyright Q 2004 WILEY-VCH Valag GmbH & Co. KGaA

Straightforward Expansions and Sources of No n u n ifo r m ity

As pointed out in Section 1.6, it is the rule rather than the exception that expansions of the PoincarC type (straightforward expansions), such as

f(?

-2 m

E)

m=O

4n(E)fm(4

where am(€) is an asymptotic sequence in terms of the parameter E, are nonuniformly valid and break down in regions called regions of nonuniformity. Some of the sources of nonuniformities are: infinite domain, small parameter multiplying the highest derivative, type change of a partial differential equation, and presence of singularities. In the infinite domain case, the nonuniformity manifests itself in the presence of so-called secular terms such as xn cos z and zn sin z,which make fm(x)lfm-l(z)unbounded as x approaches infinity. In the case of the small parameter multiplying the highest derivative, the perturbation expansion cannot satisfy all the boundary and initial conditions, and the expansion thus is not valid in boundary and initial layers. Since the boundary and initial conditions required to form a well-posed problem depend on the type of the partial differential equation under consideration, nonuniformities might arise if the type of the perturbation equations is different from the type of the original equation. In the fourth class, singularities that are not part of the exact solution'appear at some point in the expansion, generally becoming more pronounced in succeeding terms. To illustrate how nonuniform expansions arise and how they are recognized, we give several examples for each source of nonuniformity. These examples also serve to describe the mechanics of carrying out parameter perturbations. Moreover, most of these examples are taken up again in the remaining chapters where they are rendered uniformly valid. To conclude this chapter, we discuss the role of coordinates (dependent as well as independent) in making the expansions uniform or nonuniform and the role of perturbation

23

24

STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY

methods in choosing the coordinate systems that render the expansions uniformly valid. 2.1. Infinite Domains 2.1 .l. THE DUFFING EQUATION Consider the oscillations of a mass connected to a nonlinear spring described by Dufling’s equation

ii

+ u + eu3 = 0,

u(0) = a,

zi(0) = 0

(2.1.1)

where E is a small positive number. This problem admits the integral (2.1.2)

Equation (2.1.2) shows that u is bounded for all times when is positive. Let us seek an approximate solution in the form of a Poincart-type asymptotic expansion a,

u = 2 ErnUm(f)

(2.1.3)

m=O

Substituting into (2.1 .l), expanding, and equating coefficients of equal powers of E lead to the following problems for uo and u1

+ uo = 0, ii, + u1 = -uO3, ii,

Iio(0)= 0

uo(0) = a , u,(O) = 0,

zil(0) = 0

(2.1.4) (2.1.5)

The solution for uo that satisfies the initial conditions is uo = a cos t

(2.1.6)

Substituting for u,, in (2.1.5) and using the trigonometric identity cos 3t = 4 C O S ~t - 3 cos t lead to iil+ul=

--a

cos 3t

+ 3 cos t 4

(2.1.7)

The solution of (2.1.7) that satisfies the initial conditions in (2.1.5) is u, =

Thus u = a cos t

3a3 8

- -t sin t

+ €a”[-

i t sin t

-a3 +(cos 3t - cos t ) 32

(2.1.8)

+ ~ ( C O3tS- cos r ) ] + O(e2)

(2.1.9)

The two-term expansion above cannot approximate the solution as t + co because the term t sin t makes uJu0 -+ 00 as t -+ co; t sin t is called a secular

2.1.

INFINITE DOMAINS

25

term. It tends to infinity as t + 00, whereas u should be bounded for all t as discussed above. The variable t does not need to be infinite for (2.1.9) to break down; if t = O ( E - ~the ) , second term is of the same order as the first, contrary to our assumption that EU, is a small correction to u,, when we derived (2. I .9). When more terms in the series are calculated, secular terms of the form ~ " ( C O St , sin t ) are obtained. Although the resulting series is convergent, it is slowly convergent, and one cannot represent the solution for all t using a finite number of terms of this series. The appearance of secular terms is characteristic of nonlinear oscillation problems, hence one should not expect to obtain a uniformly valid straightforward expansion in these cases. 2.1.2. A MODEL FOR WEAK NONLINEAR INSTABILITY As a model for the weak nonlinear instability of a standing wave, we consider the problem (2.1 .lo) U t t - u,, - u = u3 u(5,O) = E cos kx,

u,(z,0) = 0

(2.1.11)

The initial conditions suggest an expansion of the form u = €241

+ E2U2 + E3U3 + -

(2. I .I 2)

* *

Substituting this expansion into (2.1.10) and (2.1.1 1) and equating coefficients of like powers of c, we obtain

Order

E

(2.1 .1 3)

Order UZtt

U&,

- uzzx - u2 = 0

Order e3 Ugtt U,(Z,

(2.1.14)

0) = uzt(s,0) = 0 -

UWX

- ug

0) = Us&,

= u13

(2.1.15)

0) = 0

The solution of the first-order problem (2.1.13) is u1

= cos a,t cos k z ,

a12 = k2

-1

(2.1.16)

Thus the wave is stable or unstable depending on whether k is greater or less than unity. The special case k = 1 separates stable from unstable waves.

26

STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY

The solution of the second-order problem (2.1.14) is uz = 0. Substituting for u1 from (2.1.16) into (2.1.15) and solving the resulting problem, we have us = -[12u1t sin ult

128~:

1 +3 [3(COS U1t -

+ cos ult - cos 3ult] cos kx +

C O S ~ ~ ) k2(cos361t

- C O S ~cos ~ ) 3] k ~(2.1.17)

where p2 = 9k2 - 1. Therefore tt

= E cos ult cos kx

+ e 3 [32U1 L t sin alt cos kx + terms bounded as t

m]

(2.1.18)

t =

O(E@) or

-+

Here again the straightforward expansion is invalid when larger because of the presence of the secular term t sin a,?.

2.1.3. SUPERSONIC FLOW PAST A THIN AIRFOIL

As a third example showing how an infinite domain is responsible for a nonuniformity in an expansion of the Poincar6 type, let us consider a uniform inviscid supersonic flow past the thin symmetric airfoil shown in Figure 2-1.

'SYMMETRI c Figure 2-1

AIRFOIL

+

If we assume that the velocity vector q = Ugrad (x +), the equation governing the potential function for steady two-dimensional, irrotational, isentropic motion is

where B2 = M 2 - 1 and M is the free-stream Mach number. The normal velocity vanishes on the surface; that is, the flow must be tangent to the

2.1.

INFINITE DOMAINS

27

surface. Hence

where I is the airfoil chord. The boundary condition upstream is

+hY) = 0

(2.1.21)

Van Dyke (1952) obtained a second-order solution to (2.1.19) through (2.1.21) by iteration for small but finite c. Let us obtain an expansion of the PoincarC type using c as the perturbation parameter. Let

6 = €61 +

E”2

+

(2.1.22)

* * *

Since E is small, we can greatly simplify the problem by transferring the boundary condition (2.1.20) from y = ET(z)to y = 0 by using the following Taylor series expansion

0)

d(x, ET) = +(.,

Thus we rewrite (2.1.20) as

+ .T+,(.,D) + &2T2#,,(x, 0) + * *

Substituting (2.1.22) into (2.1.19), (2.1.21), and (2.1.23), expanding for small E , and equating coefficients of equal powers of E , we have

Order

E

61,<.,

61,, - B261x, = 0

0) = T’(4,

$J~(Z,y)

Order 62YY

c2

- B242xr= M 2 “ y

(2.1.25)

(upstream)

q51,,T at y = 0 and 0

(2.1.26)

5x 5I

(upstream)

The general solution of (2.1.24) is where

I1

+ 1)6126Iz2 + (Y - 1)6lZ+lWu + 26lYArn1

42Y= q51zT’&(z, y) = 0

=0

05 2

(2.1.24)

61 =f(@

(2.1.28) (2.1.29)

.

+ g(v)

(=x-By,

(2.1.27)

(2.1.30) T=X+BY

The upstream condition (2.1.26) demands that g

= 0,

while the condition

28

STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY

(2.1.25)demands that f = -T ( t ) / B .Therefore

91 = -T(E)/B

Substituting for

into (2.1.27) gives

(2.1.31)

+ 1)f'f"

(2.1.32) Transforming the left-hand side of (2.1.32) into the 5' and 7 coordinates, we have M4(y 1) (2.1.33) f'f: at%=4B2 whose solution is #em

- BZ+aza!= M 4 ( Y

3

+

(2.1.34) The function h'(E) can be determined from (2.1.28) to be

Since u (the axial component of velocity) is given by U(l

+ 9.J

1+

---+

M4 yT'T" - TT" 0(e3) (2.1.36) 2 B3 For y = 0(1), the third term in (2.1.36) is bounded and thus is a small correction to the second term which is in turn a small correction to the first term as E -+ 0. However, as y increases to O(&) and larger, the third term becomes of the same order as the second, and then of the same order as the l)M4E3yT'T" which first because of the presence of the term (1/2)(y makes uz/ulunbounded as y -+ 00. Although there are no circular functions in this problem, this term can be considered a secular term.

+

2.1.4. SMALL REYNOLDS NUMBER FLOW PAST A SPHERE The fourth example given here to show the difficulty arising from an infinite domain is the small Reynolds number, incompressible, uniform flow past a sphere. In the spherical coordinate system shown in Figure 2-2, the full Navier-Stokes equations give the following dimensionless equation for the stream function y ( r , 0 ) (u, = y&a sin 8, ue = -y,/r sin 0) for axisymmetric flow

2.1.

INFINITE DOMAINS

29

Figure 2-2

where the Reynolds number R = Ualv

(Y

is kinematic viscosity) and (2.1.38)

The boundary conditions at the sphere's surface call for the vanishing of the velocity; that is, in dimensionless form, y(1,e) = %%(I, 0) = 0

(2.1.39)

The condition that the flow is uniform upstream gives

y ( r , 0) -+

+r2 sin2

0 as r

---f

00

(2.1.40)

Equations (2.1.37) through (2.1.40) form a well-posed problem for the stream function y . Let us seek a formal expansion of the PoincarC type valid for small R m

Substituting (2.1.41) into (2.1.37) through (2.1.40), expanding for small R, and equating coefficients of equal powers of R lead to

Order Ro

94y0

=0

(2.1.42)

yo(1,e) = yor(l, 6 ) = 0 yo(', 0) tr2 sin2 8 as r + 00 --f

(2.1.43) (2.1.44)

Order R 1 a a g4y1= - voe - - yorr2 sin e ar ae

(

+ 2 cot e yor- 2 (2.1.46) (2.1.47)

30

STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY

Equation (2.1.44) suggests that yo has the form yo = f(r) sin2 8

(2.1.48)

Substituting this assumed form into (2.1.42) leads to 4f"

8f'

8f

f i V - I - P + - r3 ---

-0

(2.1.49)

+ c1r-l

(2.1 S O )

r4

whose general solution is

f = c4r' + c2r2+ clr

The boundary condition (2.1.44) demands that c4 = 0, and c2 = 1/2, while the boundary conditions (2.1.43) demand that c1 = -3/4 and c - ~= 1/4. Therefore yo =

A4

(2r2

- 3r + -

rl)

sin2 o

(2.1.51)

This solution was obtained by Stokes (185 1). Substituting for yo from (2.1.51) into (2.1.45) gives .9*1yl

=

')

- 9- ( 2 - - + - sin2 e C O S e 4 re

r3

r5

(2.1.52)

Equation (2.1.52) and the boundary conditions (2.1.46) and (2.1.47) suggest that a particular solution for y1 has the form y1 = g(r) sin20 cos 0

(2.1.53)

g(1) = g'(1) = 0

(2.1.55)

With this form for yl,g should satisfy the following equation and boundary conditions 12g" gi* - - + - = 24g' (2.1.54) r2 r3 g(r) = o(r2) as r -+ The general solution of (2.1.54) is g = b-,r-'

00

3 9 3 1 + b, + b,r3 + b,r5 - --r2 + -r + - 16 32 32 r

(2.1.56)

(2.1.57)

The boundary condition (2.1.56) demands that b, = b, = 0. However, even with this choice for b3 and b,, g does not behave properly as r 4 co because of the presence of the term -(3/16)r2 in the particular solution for g. It is clear that no values for b, and b-, can be found that correct this shortcoming. Moreover, no other complementary solution of (2.1.52) can

2.2.

PARAMETER MULTIPLYING HIGHEST DERIVATIVE

31

be found which makes y1 behave properly as r + co.The boundary conditions (2.1.55) demand that bo = 6-, = -3132, hence 32

2r2 - 3r

+ 1- r

(2.1.58)

r2

Here again difficulty with the straightforward expansion arose because of the infinite domain. The two-term expansion 32

2r2 - 3r

x sin2 8 cos 8

+ 1- r

+ O(R2)

r2

as R + 0 (2.1.59)

satisfies the surface boundary conditions, but it does not satisfy the boundary condition at infinity. Thus, this expansion breaks down for large r and this breakdown is called Whitehead's paradox because Whitehead (1 889) was the first to obtain this solution, although by iteration, and the first to point out its nonuniformity.

2.2. A Small Parameter Multiplying the Highest Derivative 2.2.1. A SECOND-ORDER EXAMPLE In order to show the difficulty that arises when a small parameter multiplies the highest derivative, we consider the following example after Latta (1964)

EY"+y'+Y=o, Y(0) = a ,

o j s j

Y(1) = b

1

(2.2.1) (2.2.2)

where e is a small positive number. First we seek a straightforward expansion of the form a3

y = 2Pyn(s), n=O

E

<< 1

(2.2.3)

Substituting into (2.2. I) and equating coefficients of equal powers of E lead to (2.2.4) Y; Yo = 0

+

+

(2.2.5) Y:, Y, = - Y L It can be seen that at any level of approximation n, ynV1is known, hence yn for any n is given by a first-order differential equation. Consequently, the solutions of (2.2.4) and (2.2.5) cannot satisfy both of the boundary conditions (2.2.2), and one of them must be dropped. It is shown in Section 4.1.2 that the boundary condition at the origin must be dropped. Solving

32

STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY

for the first two terms and imposing the boundary condition y(1) = b give

+ Ebe'-'(l

y = be1-'

+

- z) + O(E*)

(2.2.6)

- 6)eslX+ ( b - ae"')es*"

(2.2.7)

E ) , which is in general different from the a in At the origin y = be(1 (2.2.2). Hence the error in (2.2.3) is not uniform over [0, I], and the expansion breaks down at the origin. To understand further the nature of the nonuniformity, let us look at the exact solution of (2.2.1) and (2.2.2), namely

y=

(ae"

e"2 - esl

where s1.2

It can be shown that

=

- I f J l - 4€

(2.2.8)

2E

lim y(x,

E)

(2.2.9)

=

E+O

z fixed

is in agreement with the first term in (2.2.6) and satisfies the boundary condition y(1) = b. To understand what happens a t the boundary x = 0, let us estimate y to order E from the exact solution and denote it by @.As €+O s1

= -1

+ O(E),

Therefore @

= be'-x

+ (a

s2

=

- -1 + 1 + O(E)

(2.2.10)

€

- be)"'"'''fx

+ O(E)

+

(2.2.11)

In the above estimation the term proportional to exp [-(z/c) x] was not neglected because an estimate is being made not only when E 0, but when x 0 also. From the manner in which (2.2.1 1) was constructed, the order of error is uniform on [0, 13. The behavior of g is shown schematically in Figure 2-3 together with the first term of (2.2.6) denoted by $. It can be seen that, for small E , # agrees with tj except in a small region near the origin where it changes quickly in order to satisfy the boundary condition there. Thus y(z; E) is continuous for E > 0 but discontinuous for E = 0. In fact

-

--f

lirn lim y(x;

E)

=a

(2.2.12)

lirn lirn y(z;

E)

= be

(2.2.13)

E-rO I-0

2-0

r-0

which shows the nonuniform convergence of the exact solution y(x; [the first term in (2.2.6)].

E)

to

2.2.

PARAMETER MULTIPLYING HIGHEST DERIVATIVE

33

I

I

I

I

I

I I

I

0.0

0.5

1.0

Figure 2-3 IN THIS SPECIAL CASE

(I

= 0 AND

+X

b = 1.

2.2.2. HIGH REYNOLDS NUMBER FLOW PAST A BODY Consider a two-dimensional viscous incompressible uniform flow past the body shown in Figure 2-1. The complete steady Navier-Stokes equations give for the stream function y ( u = yr,u = - y z with u and 11 the velocity components along the z and y coordinates) (2.2.14)

where the Reynolds number R = U L / v with v the kinematic viscosity of the fluid. Equation (2.2.14) needs to be supplemented by boundary conditions. At the body surface, y = F ( z ) , both components of velocity vanish; thus F’yJz, F(z)l

+ yzcz, F(z)l = 0 $4.9

Qz)l

=0

(2.2.15) (2.2.16)

The second condition represents the vanishing of the velocity tangent t o the body (the so-called no-slip condition), while the first condition represents the vanishing of the normal velocity component. The third boundary

34

STRAIGHTFORWARD EXPANSIONS AND NONUNlFORMlTY

condition is

(2.2.17)

Seeking a straightforward solution of the form m

we obtain the following equation for the first term (inviscid flow) VZY0 (Yaw a - Yoza) aY

=0

(2.2.19)

which is a third-order rather than a fourth-order differential equation. Thus yo cannot satisfy all the boundary conditions (2.2.15) through (2.2.17), and one of them must be dropped. Since an inviscid flow can slide on the body, the boundary condition (2.2.16) must be dropped. Therefore the resulting solution for yo, although it approximates the exact solution very well as R -+ 03 away from the body, breaks down near the body. Regardless of how small the viscosity is (how large R is), the tangential velocity must vanish a t the surface. Therefore for large R the exact solution is close to yo except in a thin layer near the body where it undergoes a quick change in order to retrieve the no-slip condition. This thin layer is Prandtl’s boundary

layer.

2.2.3. RELAXATION OSCILLATIONS The problem considered next is that of finding the periodic solutions of equations of the form €U“ = f ( u ’ , u ) (2.2.20) for small E whenf(u’, u ) = 0 has no periodic solutions. Van der Pol (1927) was the first to treat a problem of this kind in connection with explaining the jerky oscillations (relaxation oscillations) of an electronic circuit governed by the following equation which is named after him

+ u = a(u’ - i d 3 )

u”

If we let u

= u’, x = u/a,and

E

=

dv = c-

(2.2.21)

a-2, (2.2.21) becomes u3 u---x

3

(2.2.22) dx u If E = 0,x = o - (u3/3), which is shown in Figure 2-4. We assume that E is very small, but different from zero, and consider a solution curve that starts at P. Since P is away from F, duldx is approximately - co down to P, where

2.2.

PARAMETER MULTIPLYING HIGHEST DERIVATIVE

t'

35

Figure 2-4

it reaches I'. At this point, dvldx = 0. Since dvldx is approximately fa, away from I', the solution curve tends to follow r, staying below it, until it reaches the vicinity of P,. At this point the solution curve turns almost vertically upward until it intersects I' at P,. Since dv/dx M fa, away from I',the solution curve tends to follow I? from P3 clockwise, staying above r, until it reaches the vicinity of P,,where it turns almost vertically downward to intersect r at P,,Then it tends to follow the path from P, to P2.Therefore the limit of the periodic solution as E -+ 0 consists of the two segments P,P, and P,P, of I?, and the two vertical lines P,P, and P2P3. Thus the limit solution as E -+ 0 satisfiesf(u', u ) = 0 except at certain points where v = u' has jump discontinuities.

2.2.4. UNSYMMETRICAL BENDING OF PRESTRESSED ANNULAR PLATES The last example in this class is the unsymmetrical bending of a prestressed annular plate introduced by Alzheimer and Davis (1968). The plate

36

STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY

X

Y

A.

TOP VIEW UNDEFORMED PLATE

B.

DEFORMED CONFIGURATION

2

Figure 2-5

is clamped at the outer edge and contains a central rigid inclusion as shown in Figure 2-5. A moment is applied to the rigid inclusion which rotates about a diameter, out of the plane of the plate. For a thin annular plate with in-plane forces and no surface loading, Timoshenko and WoinowskyKrieger (1959) derived the following equation for the lateral displacement w v 4 w = --[n,S 112 a20

+ n e (1;aw s +

$2)+

a

-)]

2n,-(- l a o ar r ar

(2.2.23)

where n,, n,, and n, are the in-plane forces per unit length. In (2.2.23) the radial distance is made dimensionless with respect to the outer radius r , , and the lateral displacement is made dimensionless with respect to the plate thickness h. The flexural rigidity D = Eh3/12(1 - v2), where E is Young's modulus and Y is Poisson's ratio. We assume that the in-plane forces are applied in the form of an initial uniform radial prestressing, and that they are large enough to remain essentially constant during any subsequent lateral motion (i.e., n,. = = n = constant, n,, = 0). Thus (2.2.23) reduces to where

2 V 4 W

- V'W

=0

(2.2.24)

2.3.

TYPE CHANGE OF EQUATION

37

Figure 2-5 yields the following boundary conditions

am

- = a c o s 0 at r = b

o = bacos 0,

ar

(2.2.25) (2.2.26)

where a is assumed to be small so that sin a m a, and b = r2/rl with r2 the radius of the rigid inclusion. The boundary conditions (2.2.25) and (2.2.26) suggest a solution of the form o = u(r) cos 8 (2.2.27) Hence

t2(;i;l d2

+ -1 -d - 7) 1 (2 d2u + -1 du - 7) u - (2 d2u + -1du - %) r dr

As

E+

r dr

r dr

du

= 0 (2.2.28)

r2

u(b) = ba,

-(b) dr

=a

(2.2.29)

u ( 1 ) = 0,

- (1) = 0

(2.2.30)

du

dr

0, (2.2.28) reduces to 1 du -d2u + -dr2 r dr

u

r2

=0

(2.2.31)

which is of the second order, and its solution cannot satisfy the four boundary conditions (2.2.29) and (2.2.30). Hence two boundary conditions have to be dropped. If we seek a straightforward expansion of the form m

u = Ze"u,(r) n=O

(2.2.32)

we find that each u, satisfies (2.2.31). Consequently, (2.2.32) is not valid for all r in [ b , 11. A uniformly valid expansion is obtained in Section 4.1.5 by using the method of matched asymptotic expansions. 2.3. Type Change of a Partial Differential Equation

Since the boundary and initial conditions required to make a well-posed problem depend on the type of the partial differential equation, difficulties might arise if the original equation changes, say, from elliptic to parabolic or hyperbolic as the small parameter vanishes. This class can be considered a

38

STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY

subclass of that discussed in Sections 2.2.1 through 2.2.4. In the following discussion we describe two examples and the difficulties encountered in the expansion of one of them. 2.3.1. A SIMPLE EXAMPLE Let us consider the following Dirichlet problem for the function &x, y,

c)

For E > 0 the above problem is a well-posed problem and admits a unique solution. However, if E = 0, (2.3.1) reduces to

4z* - Al = 0

(2.3.6)

which is parabolic (diffusion equation). The solution of (2.3.6) cannot in general satisfy all the boundary conditions (2.3.2) through (2.3.5) and one of them must be dropped. As discussed in Section 4.1.2, (2.3.5) must be dropped, and thus the resulting solution is not valid near y = 1. For small E the solution of the reduced equation is expected to be close to the exact solution except in a narrow region near y = 1 where the latter changes rapidly so that it satisfies the boundary condition which is about to be lost. It should be noted that the singular nature of the problem depends not only on the change in type of equation but also on the given region in which the solution is obtained. Although the solution of (2.3.1) in the region 0 5 5, y I 1 does not tend to that of (2.3.6) uniformly, the solution of (2.3.1) tends uniformly to the solution of (2.3.6) in the upper half-plane. Next we discuss an example in which the change in type of equation does not lead to nonuniformities.

LONG WAVES O N LlQUlDS FLOWlNG DOWN INCLINED PLANES In this section we consider the characteristics of waves on the surface of a liquid film flowing down an inclined plane (Figure 2-6). This is a rather involved example, and it is discussed here because it illustrates a general technique for long nonlinear dispersive waves. The flow is governed by the 2.3.2.

2.3.

TYPE CHANGE OF EQUATION

39

Figure 2-6

Navier-Stokes equations

-ali+ - =ae o a? ag aa + a-aa + u"afi 1 a$ = -- + gsine + ~ Q ' l i a?

ag a5 afi a/; - + u ,ao + 8 -ae = - -1 _

ai

(2.3.7)

(2.3.8)

+

- g c o s e v ~ 2 ~(2.3.9) ai a., ag P a?l where li and fi are the components of the velocity in the x and y directions, 8 is liquid pressure, t is time, and p and Y are the liquid's density and kinematic viscosity, respectively. At the liquidlsolid interface, both components of velocity vanish; that is

a=fi=o

at $ = O

(2.3.10)

If the liquid surface is flat (i.e., no waves), the following laminar steady solution exists g sin 0 P=0 (2h$ - g2),

o=-

P

2v

= 0,

(2.3.1 1)

- pg cos 8(g - h,)

In this solution we used the boundary condition ati/a$ = 0 at g = h, (i.e., no shear). We next consider fluctuations in this steady-flow configuration. We introduce dimensionless quantities according to a = ho/l,

= OIUL,

h = h/h,, u

y = $/h,,

iu = t/U,,

x = $11 u = fi/aU,

(2.3.12)

40 STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY where U, = gh,2 sin 13/2v, 1is a characteristic length of the waves, and a is a dimensionless quantity measuring the shallowness of the liquid. Substituting (2.3.12) into (2.3.7) through (2.3.9) and using (2.3.11), we obtain the following equations for the dimensionless fluctuations u, Ut

+ u, = 0

(2.3.13)

+ (u + U)U, + u(u, + U ’ ) = Ut

4- (u

R

+ uR1

+ a2u,,)

-(tlvv

1 2 + V)U, + uu, = - p + - (uvv + a2u,,) Ra2 ’ aR

(2.3.14) (2.3.15)

where R = ULh$v is a Reynolds number for the liquid film, U = 2y - y2, and primes denote differentiation with respect to y. Equation (2.3.13) can be solved by introducing a stream function ~ ( zY,, t ) such that ( u , 4 = (Yv9 -YJ Then (2.3.14) and (2.3.15) can be combined into Yw

+ w + Y J Y ~ , ,- (u” + Y ~ J Y , -I 2 a 2 ~ ~ ~ + a’R[~,,t + ( U + YJY,,, - v,v,,I - a 4 ~ x z z z (2.3.16)

= aRrywt

Equation (2.3.16) must be supplemented by boundary conditions. At the solid/liquid interface, (2.3.10) gives Yv(% 0 9 0 = Y@,0,t ) = 0

(2.3.17)

At the free surface the nornial velocity of the liquid is equal to the velocity of the interface; that is

h,

+ ( U + yv)hZ+ yZ = 0

a t y = h(s)

(2.3.18)

Moreover, the condition that the tangential stress vanishes at this free surface gives

(U’

+ yw - a2y,,)(l

- a2hZ2)- 4a2yzvh,= 0

at y = h

(2.3.19)

Finally, the continuity of the normal stress across this free surface demands that - p -t (h

- 1)cot 0 -

Th,, cosec 8 (1

+ u2h,8)”’

-2a-

1 - u2h:

- 2uYzu 1

+ u8h,-

+ uzh:

Q2Yzs)hz

(YV”

1

=C

at y = h (2.3.20)

~ ~

2.3. TYPE

CHANGE OF EQUATION

41

where from (2.3.14)

P x = ka-lYgYY - &R"f',t 4- (u -I- Y,)Yzy - (u' and

+ Y,,)W,I + 8 u ~ , , ~(2.3.21)

T=-U

Pd2

with CT the liquid surface tension. To determine an equation describing the dimensionless elevation of the disturbed surface, we follow Benney (1 966a) by first finding a perturbation solution for (2.3.161, (2.3.17), and (2.3.19) through (2.3.21) in powers of u and substituting the expansion for y into (2.3.18). Thus we let Y = yo

+ ay, +

P = r*-'p-1

+ po + up1

(2.3.22)

in the former equations and equate coefficients of like powers of u to obtain Order a" YOVYVV

=0

yo = yoy= 0 at

yoyy= 2(y Order

y =0

(2.3.24)

at y = h

(2.3.25)

ywVy= 0 at y = h dc

Ylvvvv

where

- 1)

(2.3.23)

= R"P'ovvt

+ ( u + y o c / ) ~ o ~ -v v (u" +

~ovyy)yo~I

(2.3.27)

y1 = yly= 0 at y = 0

(2.3.28)

ylVV = 0 at y = h

(2.3.29)

po = (h - 1) cot 8 - Tcosec 8 h,,

PO== &"~vvu

(2.3.26)

- &R"Povr

at y =

/I

(2.3.30)

+ ( u + Wou)Ylo~Y - (u' + W o v v ) ~ o ~ l (2.3.31)

The solution of (2.3.23) through (2.3.26) is Then (2.3.27) becomes

Yo = (h - l)Y2

Y l Y V V U = 2R(h, whose solution subject to (2.3.28) is

(2.3.32)

+ 2hh,Y)

Substituting for y1 into (2.3.29) through (2.3.31) and solving the resulting

42

STRAIGHTFORWARD EXPANSIONS AND NONUNlFORMITY

equations for A and B, we obtain

B = &(ha cot 8 - T cosec e h,,) A

=

-h(h, cot 8 - T cosec 0 h,,,)

Since

a

- EY(Z, Y, f)lu=hl ax

- &Rh2(h,+ Qh'h,)

= (Yuh,

(2.3.34)

+ Yz)lv=n

we can rewrite (2.3.18) as

h,

a + (2h - h2)h, + ax [y(x,y,

f)ly=hl

=0

+ O(a) into this equation, we have h, + 2h2h2 = 0 Then y+, h , t ) = -$h3(h, cot 8 - T cosec 8 h,,,) + h ~ h ~ Letting y = yo + uyl + O(a2)in (2.3.35), we obtain h, + 2h2h2 + u [-Qh3(c0t 8 - +Rh3)h, + Q T cosec 8 h3h,, - 2h2(cot 8 - 3 h 3 ) h , 2 + 2Tcosec 8 h2h2,,,] + 0(u2) = 0

(2.3.35 )

Substituting y = yo

(2.3.36) h ,

(2.3.37)

Let us summarize what we have accomplished in this section. We started with the elliptic equation (2.3.16) which was replaced by (2.3.23) and (2.3.27) which are clearly not elliptic. From these perturbation equations we arrived at (2.3.37) which is clearly hyperbolic. Thus nonuniformities due to the type change of equation did not arise because of the unboundedness of the domain. 2.4.

The Presence of Singularities

In this class the expansions have singularities within the region of interest, which are not part of the exact solution. Moreover, in the higher-order terms, the singularities are not only preserved, but they become even more pronounced. 2.4.1. SHIFT IN SINGULARITY As a first example in this class, we consider the problem discussed by Lighthill (1949a)

dY dx

(x + q )-

+ ( 2 + x)y

=0

with y(1)

= e-'

(2.4.1)

2.4.

THE PRESENCE OF SINGULARITIES

43

This equation is singular along the h e x = -EY; the boundary condition makes the exact solution ~(z) positive for x 2 0, hence y(x) is regular in O<X
To determine a straightforward expansion, we let Y = Yo(4

+

EYl(4

+.

*

.

(2.4.2)

Substituting (2.4.2) into (2.4.1), expanding, and equating coefficients of and E to zero lead to dY0 xdx 5

dY1 dx

+ (2 + z)yo = 0,

+ (2 + 4 Y l = -Yo

-1

yo(l) = e dY0 dx

Yl(U = 0

-,

c0

(2.4.3) (2.4.4)

The solution of the zeroth-order problem is (2.4.5)

yo = x-2e-x

Substituting for yo into (2.4.4) and solving the resultant equation, we have y1 = z-2e-"~e-tt-3(l

+ 2t-l) dt

(2.4.6)

As z -P 0, yo = O(X-~),while y1 = O(X-~).Thus although the exact solution is regular at x = 0, the zeroth-order solution is singular at x = 0, and the

singularity grows stronger. 2.4.2. THE EARTH-MOON-SPACESHIP PROBLEM We consider next the motion of a spaceship of mass m which is moving in the gravitational field of two fixed mass-centers. The mass M eof the earth is much larger than the mass Mm of the moon. With respect to the rectangular Cartesian coordinate system shown in Figure 2-7, the equations of motion in dimensionless form are (2.4.7) d2Y

-=

dt2

where

rez = 2

Y -(1 - p ) - 3 - p - 3 r&?

+ y2,

r,,,' = (z

+

Y rm

-

(2.4.8)

+ y2

(2.4.9)

1~ = M m I ( M m Me) Distances and time were made dimensionless using, respectively, the distance

44 STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY

Y

Figure 2-7

d between the mass centers, and

I*"

+

L M ,d3 M e ) where G is the universal gravitational constant. This problem has been studied by Lagerstrom and Kevorkian (1963b) for the initial conditions

s=o,

dY-- -pc -

y=o,

ds

at i = O

(2.4.10)

h=-p, p f l (2.4.11) where h is the total energy of the spaceship. We interchange the roles of x and t and assume the following straightforward expansions for small p (2.4.12) =to(4 ptdx) * * * y = py1+ * * (2.4.13) Substituting (2.4.12) and (2.4.13) into (2.4.7) through (2.4.9) and equating coefficients of equal powers of p lead to

+

+

(2.4.14) t; tgt; --+3,=-+tA3 to

Y; ti2

1 52

g,3 y; + Y1 =0 to

22

1

(1 - z).)"

(2.4.15) (2.4.16)

2.4. THE PRESENCE OF SINGULARITIES 45 The solutions of these equations subject to the initial conditions (2.4.10) and (2.4.1 1) are 1 $ to = 9 sin-’ pJ. P

x In

1

1 P

Jx(1

- pzx)

+ (1 - 2pZ)x + 2[(1 - p y 1 -

p2x)x]”2

1-x y1=

-cx

(2.4.17)

(2.4.18) (2.4.19)

Thus the above expansion breaks down as x + 1 because t , has a logarithmic singularity. We find that the higher approximations are increasingly more singular near x = 1. In fact t , = 0[(1 - %)-I]

as x+ 1

(2.4.20)

2.4.3. THERMOELASTIC SURFACE WAVES We consider next the effect of heat conduction on waves propagating along the free surface of an isotropic elastic half-space. Maxwell’s modified heat conduction law is used rather than that of Fourier in order to take into account the short time required to establish a steady-state heat conduction when a temperature gradient is suddenly produced in the solid. Thus the heat flux h is assumed to be given by

ah

70-

at

+ h = --kgradO

(2.4.21)

where O denotes the change in the absolute basic temperature Oo, k is the coefficient of thermal conductivity, and T~ is the thermal relaxation time. Equation (2.4.21) implies that thermal signals have finite rather than infinite speed of propagation. Our discussion here follows that of Nayfeh and Nemat-Nasser (1971). Since the material is assumed to be isotropic, we consider two-dimensional motions in the 2-31plane, and we denote the corresponding displacements by u and v , respectively. The x axis is taken to coincide with the free surface and the y axis normal to it and directed toward the body. If B2 = (A 2p)/,u with A and ,u the Lam6 coefficients of elasticity, b = [2 + (3A/p)]aOOwith a the coefficient of linear thermal expansion, and g = a(3A 2,u)/pc, with p the material density and c, the specific heat at constant volume, then the

+

+

46

STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY

coupled elastic and heat conduction equations can be written as

aZv

aZv

+--b-

ax2

P 2 = P - + ( aY2 P2-1)=

ae ay

(2.4.23)

In the above equations time was made dimensionless by using l / w * and lengths were made dimensionless by using v,/w* where

1 + 2p vp = -

(2.4.25)

P

At the free surface the normal and tangential stresses as well as the temperature gradient vanish; that is, a t y = 0

-au+ - = av o

(2.4.26)

ay ax aU av (b2 - 2)- + 6'- - be = 0 ax ay -ae= o aY

(2.4.27) (2.4.28)

The solution of (2.4.22) through (2.4.28) is assumed to be of the form where the real part of u is positive. The wave speed is then given by c = w/Real(q) and the attenuation constant by s = Imaginary(q). Substituting (2.4.29) into (2.4.22) through (2.4.25) and setting the determinant of the resulting linear system of equations for a, to zero lead to the following three solutions for u in terms of q and w u12

= 42

- pwo"

u2 + u32 = 2q2 - 7w2 - w2(1 + T E ) az2u32

where

E

=0

4 7

+ iw(1 +

(2.4.30) E)

- w2q2 - iw3 + iwq*(l + E) + 4 4 - qZwzT(1 + E)

(2.4.31) (2.4.32)

= bg/b2.For each uk there exists an eigenvector which is given by

4

b9

1

ulk for

k = 2 and 3

(2.4.33)

2.4.

THE PRESENCE OF SINGULARITIES

47

and 1, i

,0)ol,

atq

k =1

for

(2.4.34)

Substituting the eigenfunctions corresponding to the eigenvectors (2.4.33) and (2.4.34) into (2.4.26) through (2.4.28) and setting the determinant of the set of equations for alk to zero lead to, as o + 00 (2.4.35)

where A' = (1

- c2)(1 - c2T ) - C'TE,

G(c') =

(1

- 4!?"2)4 1 - pzcc"

(2.4.3 6)

The classic Rayleigh wave speed can be obtained from (2.4.35) and (2.4.36) by letting E = 0; if 6' = 3 (corresponding to a Poisson's ratio of 1/4), c2 = 0.2817. For small E we may attempt to determine an expansion for cz whose first term is the Rayleigh solution. Thus we let c2 = CR2(1

+

EC1

+ c2c2 + .

*

.)

(2.4.37)

Substituting (2.4.37) into (2.4.35) and (2.4.36), expanding, and equating coefficients of equal powers of E on both sides, we obtain

where F = 2 - CR'(1

+ + 2J(1 T)

CR')(~

- TCR')

(2.4.39)

Equation (2.4.38) shows that the expansion (2.4.37) breaks down as T increases toward the value cg' where c1 becomes unbounded. Higher approximation can be shown to be more singular than the second terms at T = ck2. If T > ck2 the expansion (2.4.37) must be modified; otherwise A 2 is negative, hence from (2.4.32) a real part of one of the a terms becomes negative. The modified expansion is of the form c2

- - (1 7

+ €C1 + c2c1 + . . *)

(2.4.40)

The functions c1 and c2 can be determined by substituting (2.4.40) into (2.4.35) and equating coefficients of E and c2 on both sides. The result is c l = --

7

7 -

1'

c2=--

(1

T3 -T)3

-' M2 T - 1

(2.4.41)

48 STRAIGHTFORWARD EXPANSIONS

AND NONUNIFORMITY

where M is a root of

p-'(T-')+ IM2-

22

7

-

7

1-7

=0

(2.4.42)

The coefficient of M 2 vanishes at r = ci;", hence the expansion (2.4.40) is singular at this point. It can also be shown that this singularity is compounded in the higher approximation. Thus the first of the above expansions is valid for r < cz' and the second is valid for r > c;' both break down near r = cz'. An expansion valid near this singularity is determined in Section 4.1.6 by using the method of matched asymptotic expansions. 2.4.4. TURNING POINT PROBLEMS As a last example, we consider the asymptotic expansions of the solutions

of

y"

+

- z')y

A2(1

(2.4.43)

=0

for large A. For Izi < 1, the solutions of this equation are oscillatory, while for Is1 > 1 they behave as exponentials. This suggests an expansion of the form = eA4(s;A) (2.4.44) where d = d o ( 4 A-1+1(4 * . (2.4.45)

+

+

Substituting this expansion into (2.4.43) and equating coefficients of like powers of A, we obtain + 2;

2+;4;

= -(1

+ 4;

The solutions of these equations are

- 22)

(2.4.46)

=0

(2.4.47)

f i Y d 5 d . r f

IsdrP_l d7

9, = -4 In 4;

for 1x1 < 1

for 1.1

+ a constant

>1

(2.4.48)

(2.4.49)

Hence

for 1x1 < 1 (2.4.50)

2.5.

' T H E ROLE OF COORDINATE SYSTEMS

49

and = i.2

- 1)1/4

[

a2 exp

(AJ'JZ

d7)

+ b, exp (-A/'JFZ

dT)

]

for 1x1 > 1 (2.4.51) where ai and bi are constants. The expansions (2.4.50) and (2.4.51) are called the Liouville-Green or WKB approximation (Section 7.1.3). These expansions are singular at z = k l , hence they are not uniformly valid. The points z = & l are called turning points. The nonuniformity in this example arose because of the representation of the solutions in terms of elementary functions (namely, exponential and circular functions). The expansions show that the behavior of the solution changes across 1x1 = 1 from oscillatory to exponential, hence we need to represent the solutions by functions that exhibit this qualitative behavior. In this case the appropriate functions are the Airy functions (Section 7.3.1). 2.5. The Role of Coordinate Systems

In obtaining a parameter perturbation for a quantity such as u ( z ; E), we first choose an independent variable which need not be the physical independent variable z but a function 5 of z and the small parameter E. Then we assume

2 dm(~)um[5(z; 4,

u =

E)]

m=O

as

E -+ 0

(2.5.1)

where is an asymptotic sequence. We substitute this expansion into the governing equations, expand for small E keeping now 5 fixed, and then equate the coefficient of each 6, to zero. Thus

(2.5.2)

It is clear that, for a given sequence d,, u, depends on the choice of ((s;6). , Some of these choices 5 lead to nonuniform expansions, whereas others lead to uniform expansions. For example, choosing 5 = f in (2.1.1), we obtain the nonuniform expansion u ( t ; E) = a cost

+

EU~[-$

sin t

+ &(cos 3t - cos t ) ] + 0(c2)

(2.5.3)

50 STRAIGHTFORWARD EXPANSIONS

AND NONUNIFORMITY

for Duffing's equation (Section 2.1.1). Had we chosen we would have obtained u ( t ; €) = 0

cos 5

+ €a3 +os 32

35 - cos

5 = [l + (3/8)cu2]t,

g+ O(E2)

(2.5.4)

+

which is uniformly valid. Coordinates such as 5 = [l (3/8)€u2]t, which lead to uniform expansions, are called optimal coordinates (Kaplun, 1954). As a second example, we consider the model for the weak nonlinear instability problem of Section 2.1.2. Equation (2.1.18) gives the following expansion u = E cos u,t cos kx

+

1

+

~ t sin [ a,t cos ~ kx terms bounded as t -+ co (2.5.5) 320, This nonuniform expansion corresponds to the choice 5 = t. Had we chosen 5 = o,[l - (9/32012)]t,we would have obtained E

u = E cos 5 cos k x

+ O(e3)

(2.5.6)

which is uniformly valid for all 1. Hence this latter 5 is an optimal coordinate. As a third example, we consider supersonic flow past a thin airfoil (Section 2.1.3). Equation (2.1.36) gives the following expansion for the axial velocity component

1+

_--

M4 yT'T" - TT" O(e3) (2.5.7) 2 B3 where T = T ( x - B y ) ; y = eT(x) (airfoil shape). This nonuniform expansion was obtained holding x and y fixed. Had we chosen y + 1 ~ 4 t = y and x - B y = ( - c - ET'(5) (2.5.8) 2 B2 to be fixed, we would have obtained

1+

O ( E ~ )(2.5.9) - = I - € - "(') + E 2 [ $ ( 1 - M4(y '))T" - TT" U B 4B2 which is uniformly valid. Hence (2.5.8) are optimal coordinates. It should be mentioned that a coordinate may be optimal for O(E) but not optimal for O(e2). For example u

5 = (1

+ &a2)t

(2.5.10)

2.5.

THE ROLE OF COORDINATE SYSTEMS

51

is an optimal coordinate for O ( 0 for the Duffing equation (2.1.1), whereas it is not optimal for any higher order. However,

1; = (1

+

-* M ) t

(2.5.1 1)

$€a2

is optimal for O(c2) for (2.1.1). Since most straightforward perturbation expansions (obtained by holding the physical coordinates fixed) are nonuniform, perturbation techniques have been devised to render these expansions uniformly valid. In the method of strained coordinates (Chapter 3), some of these expansions are rendered uniformly valid by determination of the optimal coordinates as near-identity transformations. In some of the problems considered, such as (2.2.1) and (2.2.2), a uniformly valid expansion was given by (2.2.1 1) as y = b exp (1 - x)

+ ( a - be) exp (x - 1) - + O(r)

(2.5.12)

+ O(e)

(2.5.13)

This expansion could not be obtained by keeping x fixed or by keeping x / e fixed. Had we kept x fixed, we would have obtained y = b exp (1 - x)

= a # be in general. However, which is not valid near x = 0 because ~(0) had we kept x/c fixed, we would have obtained

y = be

+ ( a - be) exp (-

3

(2.5-14)

which is not uniformly valid because y(1) = b # be. Thus the solution is represented by two different expansions using the coordinates (scales) x and x / e . Since they are different asymptotic representations of the same function, they can be related to each other by what is called the matching principle (Chapter 4). This suggests that uniformly valid expansions can be obtained by determining different expansions using different scales, relating these expansions by matching, and then combining the expansions. This is the method of matched asymptotic expansions described in Chapter 4. Rather than obtaining, say, two expansions using two different scales to represent the asymptotic solutions of (2.2.1), we keep both x and x / e or some functions of them fixed in carrying out the expansions. This means that we increase the number of independent variables to two and transform our original ordinary differential equation into a partial differential equation. This is the method of multiple scales described in Chapter 6. In oscillation problems such as ii

+ u = Ef(u, li)

52

STRAIGHTFORWARD EXPANSIONS AND NONUNIFORMITY

the unperturbed solution (i.e.,

E

= 0) is

u=acos+,

(2.5.15)

+=t+e

where Q and 0 are constants. If E # 0 the solution can still be expressed in the above form provided that a and 0 are taken to be time-dependent. The method of variation of parameters (Section 5.1.1) can be used to find the following equations for (I and

+

da - --E

dt

sin + f [ a cos

+, --a

d4 € - = 1 - - cos + f [ a cos dt

U

(2.5.16)

sin +]

+, -a

(2.5.17)

sin +]

To determine a uniformly valid expansion of the solution of these equations, we can introduce near-identity transformations for both dependent variables a and rather than a near-identity transformation for the independent variable as in the method of strained coordinates. This is the method of averaging described in Chapter 5.

+

Exercises 2.1. Determine a three-term expansion for the solution near unity of (5

- 1)(Z - T) +

= 0,

€

<< 1

Is it valid for all values of T ? 2.2. Calculate three terms in the asymptotic expansion of q’ + z y = -1,

y(0) = 1

What is its region of nonuniformity? 2.3. The problem of isoenergetic cylindrical shock waves can be reduced to (Levey, 1959) 2 dg =g(l awg dw

where fi and of

a

when

a

a

- w2) - w(1 - Bw2)

are constants. Determine a second-order expansion for g in terms

<< 1 and discuss its uniformity.

2.4. Determine a first-order (two-term) expansion for small

x

+ z = €(k - +9), z(0) = a, (z

+

EYb’

+ Y = 0,

YU)

for

k(0) = 0

Is this expansion uniformly valid? 2.5. Consider the problem

z

=1

EXERCISES

53

(a) Determine a second-order (three-term) expansion for this problem assuming that E << 1. (b) What is its region of nonuniformity? (c) Show that the exact solution to this problem is

: J:

y=--+

-+;+1

(d) Expand this exact solution for small E and compare the result with (a). Can you conclude anything about the source of nonuniformity ? 2.6. Find a first-order (two-term) expansion for small (z

+ .y)y'

- +y

=1

+ 22,

What is its region of nonuniformity? 2.7. Determine a first-order expansion for small (.-

+ ~ y ) y ' + xy

E

of

y(1) = 1 E

of

y(1) = e-l

= be-",

What is its region of nonuniformity? 2.8. Determine a two-term expansion for the particular solution of EU"

+ (1 - x2)u

=f(z)

What conditions must be imposed o n f f o r this expansion t o be uniform? 2.9. Determine an expansion for large L for the solution of

+ y' + P2(1 - x2)y = 0 = exp [ACI(x) + &(x) + . . .]. Where does zy"

in the form y down? 2.10. Determine a two-term expansion of ii

+ wO2u+ k cos w t =

this expansion break

EU'

Is this expansion valid for all values of w ? 2.11. Determine a second-order (three-term) expansion for small solution of ii + (S + E cos 2t)u = 0

u(0) = a ,

U(0) = 0

For what values of 6 is this expansion nonuniform? 2.12. Show that a first-order expansion for small p of

+ -1, P- x is -&InWhat is its region of nonuniformity?

t(0) = 0

+ O[/L2(1 -.)-'I

E

for the

54 STRAIGHTFORWARD EXPANSIONS 2.13. For what values of a

AND NONUNIFORMITY

> 0 is the expansion

u=acost--f(a3-44a)rcost 8

uniformly valid? Is there an a

+$U3COS3t]+.~.

> 0 for which the expansion

€a3 u=acost+-((fcos3t-3rsinr)+.~. 8 is uniformly valid. 2.14. Let t = (1 EU)S in the second expansion in Exercise 2.13 and expand the result to O(P) keeping s fixed. Can you choose u to make this expansion uniformly valid? 2.15. Introduce the new variable

+

+ ii + u + €2= 0 5

in

= (1

fca2)t

Determine a first-order expansion for the resulting problem. Is it uniformly valid? Can you conclude anything about the role of the independent variables in rendering the expansions uniform? 2.16. Consider the problem

p(0) = p ( O = 1 where h = h ( z ) is a known function. Determine an expansion for large A. Discuss the nonuniformity of this expansion. Calculate two terms if p(1) = 1 is dropped. 2.17. Consider the problem

9 = a y s i n (6 + p) dE2 d2f = cos ( E + j3) dP

which arises in the bending of circular cylindrical tubes (Reissner and Weinitschke, 1963). Determine an expansion to O(a2) for small a and discuss its uniformity. 2.18. The laminar flow through a channel with uniformly porous walls of different permeabilities can be reduced to (Proudman, 1960; Terrill and Shrestha, 1965) f“

-+ R( ff” -f ’ 2 )

f (0) = 1 , f(1) = 1

-

=c

f’(0)= 0 a, f’(1) = 0

EXERCISES

Show that for small a f = 1 +

aA[2(ePR"

+ Rx - 1) - R(l - ePR1. + O(E2)

c = 2 a R 2 A ( F R - 1) and determine A . 2.19. Determine a first-order expansion for

- u,,

utt U(Z,

+ O(a2)

+ u = €U3,

0) = a cos kx,

E

<< 1

ut(x,0 ) = 0

and discuss its uniformity. 2.20. Determine a first-order straightforward expansion for small Utt

4%0) = f (2)

-

+ g(4,

CZU,,

=

E

of

EUU,

ut(x. 0 ) = clg'(z) -f'(x)I

wheref(x) and g(z) are bounded functions of x. Discuss its uniformity.

55

Perturbation Methods ALI HASAN NAYFEH Copyright Q 2004 WILEY-VCH Valag GmbH & Co. KGaA

CHAPTER 3

The Method of Strained Coordinates

In this chapter we describe techniques of rendering the approximate solutions to some of the differential equations discussed in the previous chapters uniformly valid by introducing near-identity transformations of the independent variables. This technique goes back to the nineteenth century when astronomers, such as Lindstedt (1882), Bohlin (1889), and GyldCn (1893), devised techniques to avoid the appearance of secular terms in perturbation solutions of equations such as

ii

+ wo2u = E f ( U , a),

E

<< 1

The fundamental idea in Lindstedt's technique is based on the observation that the nonlinearities alter the frequency of the system from the linear one w,, to w ( E ) . To account for this change in frequency, he introduced a new variable 7 = o t and expanded w and u in powers of E as U = Uo(T) 0

+

= oo

+

EUl(7)

€01

+

+

E2W2

E2U2(T)

+.

+.

' '

* .

Then he chose the parameters m i , i 2 1, to prevent the appearance of secular terms. PoincarC (1892) proved that the expansions obtained by Lindstedt's technique are asymptotic. Various forms of this idea have been utilized to obtain approximate solutions to problems in physics and engineering. The idea is to find a parameter in the problem (such as frequency, wave number, wave speed, eigenvalue, or energy level) that is altered by the perturbations and then expand both the dependent variables as well as this parameter in, say, powers of the strength of these perturbations. The perturbations in the parameter are then chosen to render the expansion uniformly valid. Thus we call this technique the method of strained parameters. This idea is the basis of the Rayleigh-Schrodinger method of obtaining approximate stationary solutions to the Schrodinger equation, in which one expands not only the wave function but also the energy level (Schrodinger,

56

THE METHOD OF STRAINED COORDINATES 57

1926). It is also the basis of Stoker’s method of treating finite-amplitude water waves in which the stream function and the wave speed are expanded in powers of the steepness ratio of the waves (Stoker, 1957). If we interpret this parameter expansion as a near-identity transformation, then Lighthill’s technique of rendering approximate solutions uniformly valid is a generalization of this technique. According to Lighthill (1949a, 1961), if we encounter a nonuniformity in expanding a function such as u(zl, x2,.. . ,zn;E) in powers of E, we expand not only the dependent variable u but also the independent variable exhibiting the nonuniformity, say zl,in powers of E in terms of a new independent variable as N-1

u =

2 Emu&,

m=O

21

=s

22, 23,

N

+2

m=l

ErnSm(S,

. . . , xn) + O(r”)

22, 23,.

. . ,2 3

+

O(EN+l)

The last expansion can be viewed as a near-identity transformation from z1 to s. The functions Em are called straining functions, and they are determined such that the expansion for u is uniformly valid. In other words, U , / U , - ~ < CO for all values of z1of interest, or equivalently higher approximations are no more singular than the first. Note that if t,,,= 0,s with 0, constant, Lighthill’s technique becomes the Lindstedt-Poincart technique. Since Lighthill’s transformation strains a coordinate rather than a parameter, his technique is called the method of strained coordinates. For hyperbolic differential equations, Lighthill’s technique is equivalent to expanding the dependent and independent variables in terms of some, or all, of the exact characteristics of the equations (Whitham, 1952, 1953; Lin, 1954; Fox, 1955). Rather than introducing the transformation into the differential equations and then carrying out the expansion in terms of the new variable, Pritulo (1962) suggested introducing the transformation into the nonuniform straightforward expansion. The transformation can then be found directly therefrom by solving algebraic rather than differential equations. This is another form of the method of renormalization (Section 7.4.2) introduced by Rayleigh in his analysis of scattering. Rayleigh carried out an expansion u = uo eul for scattering from a thin slab and then recast this expansion in the form u = uo exp (eul/uo) to make it valid for many slabs. The method of strained parameters is described by applying it to several physical examples in the following section. In Section 3.2, Lighthill’s technique is applied first to ordinary differential equations and then to partial differential equations. This is follo\yed by a description of Temple’s method of linearization. The method of renormalization is then taken up in Section 3.4,

+

58

THE METHOD OF STRAINED COORDINATES

while limitations of the method of strained coordinates are discussed in Section 3.5. 3.1. The Method of Strained Parameters

As discussed above, this method is based on the presence of a parameter in the problem which is altered by the perturbations. As shown in the following sections, this parameter may be the frequency of a weakly nonlinear system, the energy level in a quantum mechanical problem, the characteristic exponent in the normal solution of a linear problem with periodic coefficients, the wave number or frequency in plasma oscillations, and the wave speed or frequency in finite-amplitude surface waves. 3.1.1. THE LINDSTEDT-POINCARE METHOD The simple examples of Sections 1.1.2 and 2.1.1 show that truncated straightforward expansions in powers of E of equations of the form u

+

Oo2U

= Ef(u, li)

(3.1.1)

are valid only for short intervals of time because of the presence of secular terms. The essence of the Lindstedt-PoincarC technique is to prevent the appearance of these secular terms by introducing a new variable in (3.1.1) to obtain (1

+

€01

+

E2W2

t =

+ ..

-)-2

s(l

+

d2u

ds2

€01

+

€202

+ . - .)

(3.1.2)

+ w:u

Letting

03

u = 2:EnU,(S) n=O

(3.1.4)

in (3.1.3) nd equatin coefficients of like powers of c, we obtain equations to determine u, in succession. The solutions for u, contain secular terms unless the w , have specific values. This technique was applied to a wide range of physical and mathematical problems. For example, Keller (1968) adapted this technique to boundary value problems for systems of ordinary differential equations. A combination of this technique and the Ritz-Galerkin procedure is frequently used in the dynamic response of elastic bodies (e.g., Han, 1965; Bauer, 1968; Sweet, 1971). As an example, we consider Duffing's equation

@+u dt2

+

EU3

=0

(3.1.5)

3.1.

THE METHOD OF STRAINED PARAMETERS

59

Under the transformation (3.1.2) it becomes d2u ds2

+ (1 + Em1 +

E2W2

+ . . .)“u + €U3) = 0

(3.1.6)

We substitute (3.1.4) into (3.1.6) and equate coefficients of like powers of to obtain d2uo -+u,=o

(3.1.7)

ds2

- + u1 = d2ul ds2

E

- 2w1uo

-Ud

(3.1.8) (3.1.9)

The general solution of (3.1.7) is uo = a cos (s

+ 4)

(3.1.10)

where a and I$ are constants of integration. With (3.1.10), (3.1.8) becomes

d2ul

- + u1 = - & ~ ’ c o s3(s ds2

+ 4) - ($a2 + 2wJa

+ 4)

COS(S

(3.1.11)

If a straightforward perturbation expansion is used, w , = 0, and (3.1.1 1) reduces to (2.1.7) whose particular solution contains a secular term. In order to avoid this secular term, w1 is chosen to eliminate the coefficient of cos (s 4) on the right-hand side of (3.1.1 1). This condition determines

+

w1 to be

q=

Then the solution of (3.1.11) becomes u1 = &-a3

(3.1.12a)

-3 8a2

cos 3(s

+ 4)

(3.1.12b)

Substituting for uo, ulr and w1 into (3.1.9), we obtain

+ 4) + NST

d2u -I- ~2 = (i%sa 5’ 4 - 2w2)a COS(S

~

ds2

(3.1.13)

where NST stands for terms that do not produce secular terms. Secular terms are eliminated if a>* = &+2.4 (3.1.14) Therefore u = a cos (wt

+ 4) + 32 u3 cos 3(wf + 4) + O(c2)

(3.1.15)

60

THE METHOD OF STRAINED COORDINATES

where a and 8 are constants of integration, and 0

- $a24

= (1 =1

+ &$&z

+

+ $a26 - L 2 566-d

2

+

* *

-)-I

0(E3)

(3.1.16)

3.1.2. TRANSITION CURVES FOR THE MATHIEU EQUATION As a second example, we determine the transition curves that separate stable from unstable solutions of the Mathieu equation

ii

+ (6 + c cos 2t)u = 0

(3.1.17)

which has been studied extensively. This equation is a special case of Hill's equation which is a linear differential equation with periodic coefficients. Equations similar to this appear in many problems in applied mathematics such as stability of a transverse column subjected to a periodic longitudinal load, stability of periodic solutions of nonlinear conservative systems, electromagnetic wave propagation in a medium with periodic structure, the lunar motion, and the excitation of certain electrical systems. The qualitative nature of the solutions of (3.1.17) can be described by using Floquet theory (see, e.g., Coddington and Levinson, 1955, Chapter 3). This equation has normal solutions of the form (3. I. 18) where 4 is a periodic function o f t having a period 7r or 27r, and y may be real or complex depending on the values of the parameters 6 and E . Floquet theory shows that the transition curves, in the d-e plane, separating stable from unstable solutions, correspond to periodic solutions of (3.1.17). Some of these curves are determined below by - both 6 and II as functions - expanding of E . Thus we let 6 = n2 cd, + e2d2 + * (3.1 .19) u = eY'+(t)

+ +

u ( t ) = 240

EUI

+

E2U2

+

- *

..

(3.1.20)

where n is an integer including zero, and u,/uo is bounded for all m in order that (3.1.20) be a uniformly valid asymptotic expansion. Substituting (3.1.19) and (3.1.20) into (3.1.17), expanding, and equating coefficients of equal powers of E , we obtain

+ ii, + ii2 + ii,

n2uo

=0

(3.1.21)

n2u1

=

(3.1.22)

n2u2

= -(6,

-(al + cos 2t)u0

+ cos 2t)u, - 6,u0

(3.1.23)

The solution of the zeroth-order equation is cos nt sin nt

n=0,1,2,

...

(3.1.24)

THE METHOD OF STRAINED PARAMETERS

3.1.

61

Next, we determine the higher approximations for the cases n = 0, 1, and 2. The Case of n = 0 In this case uo = 1 , and (3.1.22) becomes

- cos 2t

ii, = -6,

(3. I .25)

In order for (3.1.20) to be a uniformly valid asymptotic expansion, 6, must vanish; and 24,

+c

= Q cos 2t

(3. I .26)

where c is a constant. With uo and u1 known (3.1.23) becomes

ii, = -6, - t - c cos 2t - Q cos 4t

(3.1.27)

In order for uz/uoto be bounded, 6, must be equal to - l / S , hence 6

-Q

=

€2

+

(3. I .28)

0(€3)

The Case of n = 1 I n this case uo = cos t or sin t. Taking uo = cos t, we find that (3.1.22) becomes ii, u, = -(& 4)cos I - g cos 3t (3.1.29)

+

+

In order that ul/uobe bounded, 6, must be equal to -1/2, and then u, = -r'&cos 3t

(3.1.30)

Equation (3.1.23) then becomes ii,

+

= -(&

+ &)COSt +

cos 3t -

cos 5t

(3.1.31)

The condition that u,/uo be bounded demands that 6, = - 1/32, hence

-~

6=I -

€

+

2 o(€3)

(3.1.32)

Had we used uo = sin t, we would have obtained the transition curve 6=I

+ ic -

~

€

+

20 ( € 3 )

(3.1.33)

The Case of n = 2 In this case uo = cos 2t or sin 2t, and (3.1.22) becomes, in the former case, ii,

+ 4u,

=

-+

- 6, cos 2t - g cos 4t

(3.1.34)

Since ul/uo must be bounded, 6, must vanish, hence u, =

-Q

+ & cos 4r

(3.1.35)

62

THE METHOD OF STRAINED COORDINATES

Substituting for u, and ul into (3.1.23) yields ii2

+ 4242 = -(a2 - A)cos 2t - 2s cos 6t

(3.1.36)

Because u2/uomust be bounded, we require that 6, = 5/48, hence

6 =4

+

+

&€2

Taking u, = sin 22 leads to

(3.1.37)

O(E3)

+O(4

6 = 4 - &€2

(3.1.38)

3.1.3. CHARACTERISTIC EXPONENTS FOR THE MATHIEU EQUATION (WHITTAKERS METHOD) Floquet theory (see, e.g., Coddington and Levinson 1955, Chapter 3) shows that (3.1.17) has solutions of the form indicated in (3.1.18), where is periodic (having a period 57 or 257) and y is a real or a complex constant. Substituting (3.1.18) into (3.1.17), we transform the latter into

+

6 + 2y4 + (6 + y2 +

f

cos 2t)+ = 0

(3.1 .39)

Since the transition curves correspond to y = 0, Whittaker (1914) obtained an approximation to near the transition curves by assuming the following expansions for 4, 6, and y

+

4 = +o 6 = 6,

+

.+l

+

+ €dl +

y = ‘y1

+

e2yz

+ -

e2+2 €26,

+ *

+

*

*

*

(3.1.40) (3.1.41) (3.1.42)

Next we carry out the solution for the case 6, = 1. Substituting (3.1.40) through (3.1.42) into (3.1.39) and equating like powers of z , we obtain (3.1.43) $0 40 =0

+

61 + 41 = -2y14, - (6, + cos 2t)bo 4 2 + 42 = -2~141 - 2y2& - (?I2 + &)+o - (6, + cos 2t)+1

(3.1.44) (3.1.45)

The general solution of (3.1.43) is +o = a cos 2

+ b sin t

(3.1.46)

where a and b are constants. Equation (3.1.44) then becomes $1

+ 4, = ~ y , a+ (+- W]

sin t - [2y1b

+ (4 + 6,)al cos 2

- $a cos 3t - &bsin 32 (3.1.47)

3.1.

THE METHOD OF STRAINED PARAMETERS

63

Since I#J is periodic, the terms that produce secular terms must vanish; that is

+ (t - 6,)b = 0

2y1a

(3.1.48)

(S + &)a + 2y1b = 0

(3.1.49)

For a nontrivial solution for a and 6, the determinant of the coefficients of u and b in (3.1.48) and (3.1.49) must be zero. Hence (3.1.50)

Then

(3.1.51)

The solution of (3.1.47) subject to the conditions (3.1.48) and (3.1.49) becomes (3.1.52) I#J1 = &a cos 3t &b sin 3t

+

With the above results (3.1.45) becomes

BZ + I#Jz = [2yza - (6,

+ y12+ & P I

+ +

sin - [(d, y12 &)a 2 ~ &cos ] t NST

+

+

(3.1.53)

Terms that produce secular terms will be eliminated if

+ y12 + &)b = 0 (6, + + &)u + 2yzb = 0 2y2a - (6,

(3.1.54) (3.1.55)

Y12

Since b is related to a by (3.1.51), (3.1.54) and (3.1.55) are satisfied simultaneously if and only if y z = 0 and

6 , = -ylz-&

(3.1.56)

Therefore, to first approximation 14

= ae

(cos t

+

cos 3t)

+-612Yl- t (sin t +

6=1

If we let then

+ €a1 + iE2(d:

&c

sin 3t)

- $1 + 0 ( € 3 )

1+

6, = 4 cos 20

y1 = t sin 2a,

O(c2)

(3.1.57) (3.1.58) (3.1.59)

6 , = a c o s 40 - 2)

b - sin2a = --cot0 a cos20- 1

(3.1.60)

64

THE METHOD OF STRAINED COORDINATES

Hence (3.1.57) and (3.1.58) become = a"ee(l/4)(sin

6 =1

lokt

+ .Q

[sin (t

cos 2a

- a) + &c

+ +z(cos

sin (3t - 4 1 4a - 2)

+

+ O(r2)

0(€3)

(3.1.61) (3.1.62)

where a" is a constant.

3.1.4. THE STABILITY OF THE TRIANGULAR POINTS IN THE ELLIPTIC RESTRICTED PROBLEM OF THREE BODIES We consider next a fourth-order system with periodic coefficients involving the stability of the triangular points in the elliptic restricted problem of three bodies. The problem is governed mathematically by 2 ' '

- 2y' -

y"

+ zx' -

1

h2" ecosf

+

=o

(3.1.63)

+hly

=o

(3.1.64)

- y)]

(3.1.65)

1 ecosf where primes denote differentiation with respect to f,and h1.z = $[I f J1 - 3p(l

Equations (3.1.63) through (3.1.65) describe the linearized motion of a particle near the triangular points in the restricted problem of three bodies. Here e is the eccentricity of the orbit of the two primaries and p is the ratio of the smaller primary to the sum of the two primaries. If e = 0, it is known that the motion is stable if 0 p < ji m 0.03852 and unstable for p 2 p. Therefore F is the intersection of a transition curve (from stable to unstable motion) with the y axis in the e-y plane. Also, it is known from Floquet theory that periodic solutions with periods 277 and 47r correspond to transition curves. In the interval 0 y < p, the period 2a corresponds to y = 0, while 47r corresponds to po = (1 - 2&/3)/2; y = 0 corresponds to a transition curve that coincides with the e axis. In the following discussion we present an analysis following Nayfeh and Kame1 (1970a) for the determination of the transition curves that intersect the y axis at yo. We assume that

<

<

m

(3.1.66) Y =

m

Z: e"Yn(f)

n=O

(3.1.67) (3.1.68)

3.1.

THE METHOD OF STRAINED PARAMETERS

65

Substituting (3.1.68) into (3.1.65) and expanding for small e , we have m

m

hz = 2 bn(po,

. . . ,pn)en

n=O

where 00, 60

= [+(I

+ k ) , +(l - k ) ] ,

(3.1.70) -

(3.1.71)

k = J" 1 2

-

b, = -al = 3J&p1

(3.1.72)

Substituting (3.1.66) through (3.1.70) into (3.1.63) and (3.1.64) and equating coefficients of like powers of c, we obtain z::

- 2y;

=

2

(- 1)krbsc o s t j

(3.1.73)

y;

+ 22;

=

2

(-l)tyrus

(3.1.74)

t=o.s=o.r=o n=s+r+t

t=O. s=O.r=O n=r+s+t

COStf

The zeroth-order equations admit the following periodic solutions of period 4rr xo = cos 7, yo = -a sin 7 (3.1.75)

xo = sin 7,

where

f

r = - ,

2

a = bo

yo = a cos 7

+ 4 = (ao + a)-'

(3.1.76) -

= f(7 - J33)

(3.1.77)

There are two transition curves intersecting the ,u axis at ,u = pocorresponding to the above two independent solutions. If we take (3.1.75) we find that the first-order problem becomes

- 60x1 = ( b , - ibo) COS 7 - Sbo COS 37 + 2 4 - aoyl = -a(al + $ao)sin T + &aaosin 37

Xp

y;'

- 2yi

(3.1.78) (3.1.79)

The terms proportional to cos T and sin T lead to secular terms in the particular solutions for x1 and yl. To determine the condition necessary for the removal of these secular terms, we assume a particular solution of the form xD = 0,

yp = c sin r

(3.1.80)

Substituting (3.1.80) into (3.1.78) and (3.1.79) and equating the coefficients of cos 7 and sin T on both sides, we obtain

66

THE METHOD OF STRAINED COORDINATES

Elimination of

c

leads to

Since b, = --a,

b, - &bo= -a2(al

+ $ao)

(3.1.82)

bo - UOU' * -0.1250 2(1 - a') p1 w -0.05641

(3.1.83)

b, = Hence from (3.1.72)

(3.1.84)

Therefore the transition curve to first order is given by

+ O(ez)

(3.1.85)

+ 0.05641e + O(e2)

(3.1.86)

p = 0.02859 - 0.05641e

Had we used the solution given by (3.1.76) for the zeroth-order problem, we would have obtained the second branch p = 0.02859

The above analysis could be continued to higher orders in a straightforward manner and has been carried out to fourth order by Nayfeh and Kame1 (1970a). 3.1.5. CHARACTERISTIC EXPONENTS FOR THE TRIANGULAR POINTS IN THE ELLIPTIC RESTRICTED PROBLEM OF THREE BODIES It is known from Floquet theory that (3.1.63) and (3.1.64) have normal solutions of the form (3.1.87) 2, Y = eY"4(f ), Y(f)l where 4 and y are periodic, having a period of either 27r or 4v, and y is a real or complex number. Substitution of (3.1.87) into (3.1.63) and (3.1.64) transforms the latter into

6" + 2Y+' - 2Y' + Y24- 2YP -

4'+ 2YY' + 24' + Y'Y + 274 -

=0

(3.1.88)

+ ecosf = 0

(3.1.89)

hZ4 ecOS

+

'"

The transition curves correspond to y = 0, hence near these transition curves y is small. Thus to obtain expansions for 4 and y valid near the transition curves intersecting the p axis at polwe let (3.1.90)

(3.1.91) P = po

+ epl +

(3.1.92) (3.1.93)

3.1.

THE METHOD OF STRAINED PARAMETERS

67

Substituting (3.1.90) through (3.1.93) into (3.1.88), (3.1.89), and (3.1.65) and equating like powers of e, we obtain

Order e0 (3.1.94) (3.1.95)

Order e

+ 2y,r~,+ - bo$o Y; + 2$; - a o ~ = 1 - 2 ~ 1 ~; 2 ~ 1 4 0+ aiyo &' - 2

~; bo4i =

-2y1$;

b140

~

0

C O S ~ (3.1.96) C O ~ S 0~ (3.1.97)

The general solution of (3.1.94) and (3.1.95) is $o = A cos T

+ B sin

T

(3.1.98)

yo=aBcos7-aAsin7

(3.1.99)

This solution determines the right-hand sides of (3.1.96) and (3.1.97). Thus

4; - 2y; - bo$l

= PI, cos T

+ Q,, sin

T

- &boACOS 37 - 4boB Sin 37

y; where

+ 24; - aoyl = P,, cos + Qlz sin - &aoaBcos 37 + &aoaAsin 37 T

P,, = yl(a - 2)A

+a

Qlz = yl(a - 2)B

-a

T

(3.1.101)

3

a -- B

( 1

(a , + -3A

Since 4 and y are periodic, the secular terms in the solution of must vanish. To eliminate these secular terms, we let

dl = 0,

(3.1.100)

y, = c, cos T

+ c2 sin

T

dl and y1 (3.1.102)

68

THE METHOD OF STRAINED COORDINATES

Substituting (3.1.102) into (3.1.100) and (3.1.101) and equating the coefficients of cos T and sin T on both sides, we obtain

Elimination of c1 and cz from (3.1.103) results in Pi1

= aQ127

Qll

(3.1.104)

= --PIZ

Substitutingthe above expressions for P and Q into (3.1.104) and rearranging, we obtain [bl -

2 + bo

yl(l - 4a

(

ct2 u1

31

+-

+ a2)A +

A - yl(l - 4a

+ a2)B = 0

(3.1.105)

= 0 (3.1.106)

For a nontrivial solution the determinant of the coefficients of A and B in (3.1.105) and (3.1.106) must vanish. This condition leads to

Then (3.1.108)

Therefore, to first approximation 2, y

= eeY1'[cos(if- a), -a sin (if- a)]

+ O(e)

(3.1.109)

The transition curves (3.1.85) and (3.1.86) correspond to y1 = 0, while (3.1.75) and (3.1.76) can be obtained from (3.1.109) by letting y1 = 0 and a = 0 or 4 2 . The present analysis can be continued to higher order in a straightforward manner even though the algebra is involved. The expansion has been carried out to second order by Nayfeh (1970a). 3.1.6. A SIMPLE LINEAR EIGENVALUE PROBLEM Let us now consider the problem of determining the eigenvalue L and the eigenfunction u where 24''

+ [A + . f ( q u

= 0,

f ( 4= f ( - - 4

u(0) = u(l) = 0

(3.1.1 10) (3.1.111)

3.1.

THE METHOD OF STRAINED PARAMETERS

and E is a small quantity. If given, respectively, by

E

69

= 0, the eigenfunctions and eigenvalues are

n = 1,2,3,. ..

u, = J i s i n n n x ,

(3.1.112)

A, = n2n2

(3.1.1 13)

The above eigenfunctions are orthonormal; that is (3.1.114)

s,’.n(x)um(z) dz = d m n where,,a,

the Kronecker delta function, is defined as follows amn

For a small but nonzero

A,, by letting

u, =

E,

=

[

0

m#n

1

m=n

we obtain an approximate solution to u, and

JZ sin nnx + E u n l +

A,

= n2n2

E2unZ

+ ..

*

+ cAnl + e2An2+ . . .

(3.1.115)

(3.1.116)

Substituting (3.1.115) and (3.1.116)into (3.1.110) and (3.1.111) andequating coefficients of like powers of E, we obtain

45

where the zeroth-order problem is satisfied identically, and u,,, = sin nnx. We assume that unl can be expressed as a linear combination of the zeroth-order eigenfunctions u,,,; that is m

u , , ~= 2 an,,,$ m=l

(3.1.1 19)

sin mnx

This solution satisfies the boundary conditions on unl. Substituting (3.1.1 19) into (3.1.117), we obtain

2 Jz n2(n2 - rn2)an, m

m=l

Multiplying (3.1.120) by

sin mmx = -,/?f(x)

sin nnx

- JZ

A,, sin nnx (3.1.120)

& sin k m , integrating from 0 to 1, and using the

70 THE METHOD OF STRAINED COORDINATES orthonormality property (3.1.1 14) of the eigenfunctions (3.1.1 12), we obtain m2(n2

where

- k2)a,

= -Fnk - An16,k

I’

(3.1.121)

F , = 2 f (x)sin nnx sin k m d x

(3.1.122)

If k = n, the left-hand side of (3.1.121) vanishes, hence

A,, = -Fnn

= -2

However, if k # n

s,’

f (2) sin2n n x dx

(3.1.123)

F, (3.1.124) r2(n2 - k2) Condition (3.1.123) is equivalent to the removal of secular terms. The function unl is then ank=

-

$z sin knx

+ an$

sin n i m (3.1.125)

Note that unnis still undetermined. It is determined in the final solution by normalization of u,. Proceeding to second order, we assume that

2 bn& Q)

un2 =

r=l

sin rnx

(3.1.126)

Substituting for un2,unl, and un0into (3.1.118), we obtain

7r2z(n2 - r2)>Jzb,, sin r m = - 2an& m

03

r=1

k=1

w ‘

- za,l,$

f(x) sin k m

sin k n x

k=1

- An2JZ

sin n m (3.1.127)

Multiplying (3.1.127) by & sin m x , integrating from 0 to 1, and using the orthonormality property (3.1.114), we obtain r 2 ( n 2 - S2)bns =

m

-k=1 z U n k F k J - U n s ~ n l- ln26,,

(3.1.128)

If s = n, the left-hand side of (3.1.128) vanishes, while the right-hand side gives (3.1 1.29)

THE METHOD OF STRAINED PARAMETERS

3.1.

71

If s # n, (3.1.128) gives

2

-

FnkFks

-

annFns

FnnFna (3.1.130) n2(n2 - s2) 7r4(n2 - s2)' Here again b,, is still undetermined; it will be determined from the normalization of un. To normalize u, we require that

bns

=

k+n

n4(n2 - k2)(n2 - .?)

(3.1.131)

Since u,, is normalized, we obtain fl

J,

d~ = 0

(3.1.1 32)

~1(2un0un2 u;d d x = 0

(3.1.1 33)

unounl

+

Condition (3.1.132) gives an, = 0, while condition (3.1.133) gives bnn

=

2 azk

-4

Therefore, to second order u, =

42 sin n n x - E 2

k#n

(3.1.134)

k+n

f l sin kmx

Fn,

n2(n2 - k2) FnsFks

n4(n2 - s2)(n2 - k2)

- -1

-

F n n F n k 2 2]$

m4(n2 - k )

sin knx

'" JZ sin nnz + O(c3)

2 r4(n2 - k2)'

2. = n2n2- E

F ~ ,

+ 2 E2

k#n

- kz) + o ( ~ ~ )

'ik

mZ((n2

(3.1.135) (3.1.136)

The expansion method described in this section is called the RayleighSchrodinger method; it was developed by Schrodinger (1926) to treat stationary solutions of the Schrodinger equation. For more references and more complete treatment, we refer the reader to the book edited by Wilcox (1966) and to the article by Hirschfelder (1969). 3.1.7.

A QUASI-LINEAR EIGENVALUE PROBLEM Let us next consider the eigenvalue problem

H+

+ A#

=4

4)

(3.1.137)

72

THE METHOD OF STRAINED COORDINATES

subject to the linear homogeneous boundary condition (3.1.13 8)

B(+) = 0

where H is a linear operator and F is a nonlinear operator of 9. We seek an approximate solution for small c by letting (3.1.139) (3.1.140)

Substituting (3.1.139) and (3.1.140) into (3.1.137) and (3.1.138) and equating coefficients of like powers of E, we obtain

w o+

A040

H+I -I- ~ O $ I=

= 0,

-~I+o

a+,) =0

+ F(+o),

B(41) = 0

(3.1.141) (3.1.142)

We must distinguish between two cases depending on whether the eigenvalues of (3.1.141) are distinct or not. The first case is called the nondegenerate case, while the second case is called the degenerate case because more than one eigenfunction corresponds to a repeated eigenvalue. Both of these cases are treated in order below. The Nondegenerate Case. Let us assume that (3.1.141) is solvable and that its solution yields the eigenfunctions u, corresponding to the eigenvalues pn, n = 1 , 2 , . . . . We assume further that pm# p,, if m # n,and that the eigenfunctions {un} form an orthonormal set so that J;nkmdx

= dm,

(3.1.143)

where x is the vector representing the coordinates, k is the complex conjugate of u , and the integration is over the domain D of interest. To solve (3.1.142) we expand in terms of the orthonormal set {u-} as in the previous section; that is (3.1.144)

Thus +1 satisfies B(+,) = 0 because B(u,) = 0 for each m. Letting do= u, and A, = p,, and substituting (3.1.144) into (3.1.142), we obtain

Multiplying (3.1.145) by 4,integrating over D, and using the orthonormality condition, we obtain (1.n

- ~ s ) a=, - 4 S m

+

Fns

(3.1.146)

3.1.

THE METHOD OF STRAINED PARAMETERS

73

where (3.1.147)

If n = s, (3.1.146) gives (3.1.148)

Ifn#s Fns

a, = ___ Pn - Ps

Thus

(3.1.149)

The coefficient an,, can be determined to be zero if we assume that 4 = $o O ( E ~is )normalized as in the previous section. Therefore, to first order

+

4=un+.z: 1 = p,,

Fnm

___ 11,

m+n P n

- Pm

+ O(E2)

+ eFnn + O(e2)

+

(3.1.1 51) (3.1.152)

As an example, let us consider the problem (3.1.1 53) 4(0) = $(I) = 0 Here D is the interval [0, 13, and u, = &sin

Since F(+) =

-43

nrx,

p, = n 2 r r 2

(3.1 .1 54) (3.1.155)

and ii, = u m sin3 nrrx sin m r x dx =[(sin = @rn.3n

3nrrx - 3 sin n r x ) sin mrrx d x

- @nm

(3.1.156)

Therefore (3.1.151) and (3.1.152) become

J2 4 = $ sin nrrx - sin 3nrrx + 0 ( E 2 ) 16n2r2 E

1 = n2r2- Q€ + O ( 2 )

(3.1.157) (3.1.158)

74

THE METHOD OF STRAINED COORDINATES

The Degenerate Case. In this case let p,+& = p, for k = 0, 1 , 2 , . . . ,M . Then M 40 = bkun+k (3.1.1 59)

We substitute for +1 and take 1, = p, to obtain

4,

k=O

from (3.1.144) and (3.1.159) into (3.1.142) and

m

M

m=l

k=O

z‘(Pn - Pu,)amuu, =

b/Pn+k

+F 2

L M O

bkun+k

We multiply (3.1.160) by zi, and integrate over D to obtain M

@n

- Pu,)as = -IlZbkds,,+k

1

+ SAbo, bl, . . . ,b,)

(3.1.160)

(3.1.161)

where (3.1.162)

+ k for k = 0, 1 , 2 , . . . ,M, (3.1.161) gives k = 0, 1 , 2 , . . . , A4 (3.1.163) F,+k(bo, b , . . . ,b,) - Iibk = 0 , These constitute M + 1 homogeneous algebraic equations for the M + 1 unknown b , terms and the eigenvalue I,. If s # n + k , k = 0,1, . . . , M , If s = n

a, =

SAbo, bi,. Pn

-

9

b ~ )

-K

(3.1.164)

As an example, let us consider the problem -d4+ ++n

dx4

-+I+=€+-

d4

dx2

dx

2d2+

+(O) = d”(0) = $(l) = d”(1) = 0 In this case the solution of the linearized problem is u, =

JZ

sin nnx,

pn = n2(5

- n2)T4

(3.1.165) (3.1.166) (3.1.167)

Thus ,ul = p2 = 4m4,and we have degeneracy. We assume that, corresponding to the eigenvalue p1 Then

+o = boJZ sin nx

= n[-bobl sin TX

+ blJZ sin 2nx

(3.1.168)

+ bo2sin 2nx + 3bob, sin 3nx + 2b; sin 4nzJ (3.1.169)

3.1.

THE METHOD OF STRAINED PARAMETERS 75

Hence

p8= J I [ F ( + ~ ) sin sax dx =

+Jz~[-bobldlr + bO2d, + 3bobldb + 2b12d,,]

With Ss known (3.1.163) becomes

- Ilbo = O &JZab,2 - I,bl = 0

(3.1.171)

-aJ%rb&,

While (3.1.164) gives

3b0b1 a3 = 4oJz a3' Since b, f 0, (3.1.171) yields

(3.1.170)

(3.1.172)

b12 a, = 90$ a '

(3.1.173)

(3.1.174)

Substituting for 6 , into (3.1.172) and solving for A,, we obtain

A,

(3.1.175)

J2

Hence

b, = fibo

Therefore

+ = b&

-1 inbo

=

sin ax f iboJT sin 2nx 3 f -ib: sin 3mx 40a3

+

E

(3.1.176)

[a,$

-b; 90a3

I = 47r4 q= T rib0

Jz

sin T X

+ a,$

1+

sin 4 a x

+ O(2)

sin 2 a x

O(E*) (3.1.177)

(3.1.178)

The constants a, and a2can be related to b, by normalizing +o. The solutions corresponding to p,, n > 2 are

+ = @sin A

= n2(5

n m

+ r 15n(n' - l)a3sin 2 n a z + O(ra)

- n2)a4 + O ( 2 )

(3.1.179) (3.1.180)

76 THE

METHOD OF STRAINED COORDINATES

3.1.8. THE QUASI-LINEAR KLEIN-GORDON EQUATION We consider in this section the problem of the determination of periodic finite-amplitude traveling waves for the equation

ult - a%,

+ y2u = /3u3

(3.1.181)

If we neglect the nonlinear term @*, we obtain the linear traveling harmonic wave u = u cos (kz

- of),

o2= azkz

+ ye

(3.1.182)

The phase speed for this wave is o l k which is independent of the amplitude u. In the nonlinear problem the phase speed is in general a function of the amplitude. To determine the dependence of the phase speed c on the amplitude, we assume that 6 = z - ct

u = u(6),

(3.1.183)

so that (3.1.181) becomes (cz

- a2)u" + y2u = @u3

(3.1.184)

where primes denote differentiation with respect to 6. We assume the amplitude to be small and expand both u and c as

(3.1.185)

Had we included the terms ac, and as#,, we would have found that c, = 0 and u2satisfies the same eqmtion as u,, hence u2 is not included. Substituting (3.1.185) into (3.1.184) and equating coefficients of equal powers of u, we obtain

- a')#; (cgB - a">; (ct

+ y2u1 = o + y2u3 = -2coc2u~ + Su,"

(3.1.186) (3.1.187)

We take the solution of (3.1.186) to be u1 = cos k0,

coz = a2

+ yak-z

(3.1.188)

3.2.

LIGHTHILL’S TECHNIQUE

77

so that (3.1.185) coincides with (3.1.183) to O(a).Then (3.1.187) becomes

+ y2u3 = (2c0c2k2 + $/I) cos k0 + $/Icos 3k0

(cO2- u’)u~,

(3.1.189)

Secular terms are eliminated if c2 = -3/I/8q,ka. Then

’

us=---

32y2

cos3kB

Therefore a3i3 cos 3kO u = a cos k0 - 32y2

c = vIu 2

+ y2k-’

[

1-

+

*

]

343 + ... 8(a2k2 y2)

+

(3.1.190)

The technique used in this section was formalized by Stoker (1957) for the treatment of surface waves in liquids. This technique was used to treat the interaction of capillary and gravity waves in deep and finite-depth water by Pierson and Fife (1961) and Barakat and Houston (1968), respectively. It was also used by Maslowe and Kelly (1970) to treat waves in KelvinHelmholtz flow. Instead of expanding the phase speed, we could have expanded the wavenumber to determine the wave number shift or the frequency to determine the frequency shift. Variations of this technique were applied to a variety of problems. For example, Malkus and Veronis (1958) treated the Binard convection problem. Pedlowsky (1967) determined the response of a bounded ocean to a surface wind oscillating near one of the Rossby wave frequencies. Keller and Ting (1966) and Millman and Keller (1969) obtained periodic solutions for various systems governed by nonlinear partial differential equations, while Keller and Millman (1969) treated nonlinear electromagnetic and acoustic wave propagation. Rajappa (1970) investigated nonlinear Rayleigh-Taylor instability. 3.2. Lighthill’s Technique The essence of Lighthill’s technique is t o expand not only the dependent variable u(z,, zl, . . , xn; c) in powers of the small parameter E, but also to expand one of the independent variables, say zl,in powers of E . Lighthill (1949a, 1961) introduced a new independent variable and then expanded both u and x1 in powers of E with coefficients depending on s. To a first approximation he assumed that x1 and s are identical. Thus Lighthill assumed

.

78

THE METHOD OF STRAINED COORDINATES

the following expansions for u and xI m

u = 2 (Emum(& 22,5 3 , . . . , xn) m=O

m

XI

= s 4-

Z : E r n t r n ( S , x.2, x3,. m=l

. . , xn)

(3.2.1) (3.2.2)

It is clear that the straightforward expansion (Poincark type) consists of (3.2.1) alone with s replaced by zl.Since this straightforward expansion is not uniformly valid, Lighthill introduced (3.2.2) and chose Em (called straining functions) so as to make both of the above expansions uniformly valid; that is, he chose 5, so that the resulting approximation is uniformly valid. In some cases this is accomplished by requiring that and Urn-1

Em

be bounded

(3.2.3)

tm-1

In other words, higher approximations shall be no more singular than the Jirst. Comparing (3.2.1) and (3.2.2) with (3.1.2)and (3.1.4), we see that Lighthill’s technique is an extension of the method of strained parameters. This technique was modified by Kuo (1953, 1956) to apply to viscous flows. For this reason Tsien (1956) called it the PLK method, that is, the Poincarb Lighthill-Kuo method. This method was applied to a variety of problems, especially wave propagation in nondispersive media. Lighthill (1949b) treated conical shock waves in steady supersonic flow. Whitham (1952) determined the pattern of shock waves on an axisymmetric projectile in steady supersonic flow and treated the propagation of spherical shocks in stars (1953). Legras (1951, 1953) and Lee and Sheppard (1966) applied it to steady supersonic flow past a thin airfoil, while Rao (1956) applied it to sonic booms. Holt and Schwartz (1963), Sakurai (1965), Holt (1967), and Akinsete and Lee (1969) investigated nonsimilar effects in the collapsing of an empty spherical cavity, while Jahsman (1968) treated the collapse of a gas-filled spherical cavity. Sirignano and Crocco (1964) analyzed combustion instability which is driven by chemical kinetics. Savage and Hasegawa (1967) studied the attenuation of pulses in metals, while Sakurai (1968) discussed the effect of plasma impedance in the problem of inverse pinch. Einaudi (1969, 1970) applied it to the propagation of acoustic gravity waves. Lewak (1969) and Zawadzki and Lewak (1971) solved Vlasov’s equation. Espedal (1971) used a combination of this technique and the method of matched asymptotic expansions to determine the effect of ion-ion collision on an ion-acoustic plasma pulse. Asano and Taniuti (1969, 1970) and Asano (1970) extended this technique to nondispersive wave propagation in slightly inhomogeneous media.

3.2.

LIGHTHILL'S TECHNIQUE

79

Melnik (1965) applied it to the entropy layer in the vicinity of a conical symmetry plane. Mclntyre (1966) treated optimal control with discontinuous forcing functions. Ross (1970) applied it to diffusion-coupled biochemical reaction kinetics. Barua (1954) analyzed secondary flows due to rotation in an unheated tube, Morton (1959) treated laminar convection in a heated tube, and Morris (1965) investigated the case of laminar convection in a vertical tube. Chang, Akins, and Bankoff (1966) analyzed the free convection of a liquid metal from a uniformly heated plate. Crane (1959) rendered an asymptotic expansion for boundary layers uniformly valid. Goldburg and Cheng (1961) discussed the anomaly arising from the application of this technique and parabolic coordinates to the trailing edge boundary layer. Ockendon (1966) investigated the separation points in the Newtonian theory of hypersonic flow. Since Lighthill's technique is a generalization of the method of strained parameters, the first technique gives results identical to those obtained using the latter technique whenever it is applicable. Therefore the examples discussed next are problems that cannot be treated by the method of strained parameters. 3.2.1. A FlRST-ORDER DIFFERENTIAL EQUATION The first example treated by Lighthill is the first-order differential equation (3.2.4) where y(z) and Y(.c) are regular functions for all 5 of interest. Wasow (1955) determined the necessary conditions for the convergence of Lighthill's expansion for this problem; the proof had an error which was corrected by Lighthill. Usher (1971) investigated the necessary conditions for the applicability of this technique to equations of the form y'

=fk, Y> + .g(%

Y)

+

*

..

Comstock ( 1 968) showed that Lighthill's technique may lead to an erroneous expansion (see Exercise 3.28) for ( 2 " + 6y)y' + nx"-ly = rnz'"-l , Y(1) = a > 1

while Burnside ( 1 970) investigated the uniformity of the expansion obtained by straining z = x n rather than 2. It is clear that the region of nonuniformity is in the neighborhood of x = 0. For E = 0, (3.2.4) has the solution y = [exp

jz-

[J' y ( e x p J ' v

d7)

+ c]

(3.2.5)

80

THE METHOD OF STRAINED COORDINATES

Let q(0) = qo, then expf

4O dt = xaoR(x)

(3.2.6)

t

where R(x) stands for a function regular at x = 0. Since r(x) is regular at

x=o

y = R(x)

+ O(x-*o)

as

2 -+

(3.2.7)

0

except when qo is a negative integer. In the latter case y = R(z)

+ O(x-90 In x)

as x -+ 0

(3.2.8)

Equations (3.2.7) and (3.2.8) show that the zeroth-order solution of (3.2.4) is bounded or unbounded as x -+ 0 depending on whether qo < 0 or qo 2 0. In order to show the details of the method, we apply it to a specific example for which qo = 2. In this case we consider the following problem treated by Lighthill (1949a) and Tsien (1956)

+ ey)- dY + (2 + x)y = 0, dx

(z

y(1) = Ae-l

(3.2.9)

where A is a constant. Following Lighthill, we assume that (3.2.10) x =s

+ 2 ErnXm(S) m

(3.2.11)

m=1

Then

(3.2.12) In order to apply the boundary condition, we need to de-Ermine the value of s, denote it by S, corresponding to x = 1 ; that is, we must solve m

s = 1 - 2 Pxm(S)

(3.2.13)

m=l

We expand 2 in powers of

E

according to

s = 1 + ES1

+ €2s2+ - -

(3.2.14)

*

Substituting (3.2.14) into (3.2.13), expanding, and equating coefficients of equal powers of E lead to

s = 1 - q ( 1 ) - €2[X2(1) - xl(l)x;(l)]

+

*

.

*

(3.2.15)

3.2.

LIGHTHILL'S TECHNIQUE

81

The boundary condition can now be written as

+

Ae-l = ~ ~ ( 1 4y,(l) )

or

- yX(l)z,(l)]

+ ...

(3.2.16)

yo(I) = Ae-l

(3.2.17)

Ydl) = Y6(1)21(1)

(3.2.18)

Substituting (3.2.10) through (3.2.12) into (3.2.9), expanding, and equating the coefficients of qo and E to zero yield sy;,

+ (2 + S)Yo = 0 + ( 2 + S>Yi = -(2 + s)Yoz; - (yo + Y;)Z~ - yoy;

S Y ~

(3.2.19) (3.2.20)

The solution for yo is

yo = A e - Y 2

(3.2.21)

With this solution (3.2.20) becomes

If x1 = 0, (3.2.22) reduces to the equation for the first-order term in the straightforward expansion, where y1 is more singular at z = 0 than yo. In fact, yo = 0(r2), while y1 = 0 ( c 5as ) x -+ 0. To render the above expansion uniformly valid, x1 can be chosen so that y1 is no more singular than yo by eliminating the right-hand side of (3.2.22). However, Lighthill found that a uniformly valid expansion can be obtained by choosing z1to eliminate the worst singularity. Thus he put (3.2.23)

or A 3s2

xl=--

Then (3.2.22) becomes

Hence

3s3

(3.2.24)

+ Ae-"($ +

s4

-

s,'

e-6 t

5)

- + - dE 4

(3.2.25)

1

(3.2.26)

The straining function x2 can be found from the elimination of the worst singularity in yz to be 3A2 x2=-(3.2.27) 1oS4

82

THE METHOD OF STRAINED COORDINATES

Therefore

where x = s - - - -eA 3s2

3e2A2 los*

(3.2.29)

The roughest approximation that is uniformly valid near the origin is y = A~-'s-~

where s is the root of

(3.2.30) (3.2.31)

which is approximately x when x 2 0 and E << 1 . This expansion is assumed to start from a positive value of x, and it is required to continue downward and through x = 0. For a physical problem this continuation stops if there exists a real branch point of s as a function of x before the origin. The branch point is given by &/is = 0, or equivalently by x EY = 0, which is a singularity of the original equation (3.2.9). In this case the branch point is given by s w (--2Ae/3)llS which is positive if and only if A < 0. Therefore the above expansion would be valid up to the origin if A > 0, and stops at x w (3/2)(-2Ae/3)ll8 if A < 0. If A = 1 then x = 0 corresponds to

+

S =

Hence at x = 0

Y=

yr3+ (7/a- y y 3 +

(;)""+

10 3 10

U(r)

(3.2.32)

O(1)

(3.2.33)

E

3.2.2. THE ONE-DIMENSIONAL EARTH-MOON-SPACESHIP PROBLEM The one-dimensional earth-moon-spaceship problem was studied by Nayfeh (1965a) and can be reduced to (see Section 2.4.2)

+-1 -P- 2 We assume that

t(0) = 0

+ Ptl(4 + 0(P2) =s + + 0(P2)

t = tds) -2

,

Pl(S)

(3.2.34) (3.2.3 5 ) (3.2.36)

3.2.

LIGHTHILL'S TECHNIQUE

83

Substituting (3.2.35) and (3.2.36) into (3.2.34) and equating coefficients of equal powers of p , we have 202 = s,

The solution of (3.2.37) is

JZ

to =

t,(O) = 0

(3.2.37)

p

(3.2.39)

If z1= 0, I, = 0 t h (1 - z)] as z -+1. The singularity in t , can be removed if the right-hand side of (3.2.38) is eliminated; that is (3.2.40) (3.2.41) Therefore, to first approximation where s is the root of

J2

t = gs3/2

+~

( p )

(3.2.42)

3.2.3. A SOLID CYLINDER EXPANDING UNIFORMLY I N STILL AIR Let us next solve the problem of a cylindrical shock wave produced by a cylindrical solid body expanding uniformly from zero in inviscid nonconducting still air. This problem was also studied by Lighthill. The radial expansion velocity is assumed to be ca, where a, is the speed of sound in still air and is a small quantity. The shock propagates with a uniform radial velocity Ma, where M is the shock Mach number. The flow between the cylinder and the shock is adiabatic and isentropic, hence it can be represented by a potential function q ( r , t ) (the radial velocity q = qJ given (3.2.44) where u is the local speed of sound and is related to a, by Bernoulli's equation; namely (3.2.45)

84 THE METHOD OF STRAINED COORDINATES

where y is the ratio of the gas specific heats. The gas is assumed perfect with constant specific heats. Three boundary conditions must be satisfied by pl. The velocity of the air at the cylinder's surface is equal to the velocity of its expansion, that is (3.2.46) The second condition is the continuity of pl across the shock. Since pl = 0 in still air pl(Ma,t) = 0 (3.2.47) The third condition is the Rankine-Hugonoit relation between the shock velocity and the velocity of the air behind it; that is

2a,(M2 - 1)

- (Maor) = apl

(3.2.48) ar M(Y 1) Since there is no fundamental length in this problem, all flow quantities are functions only of r/uot.Thus we let ql

+

= a02tf(x),

Then the problem becomes

r x =a0t

(3.2.49)

subject to the boundary conditions df dx

=€

(3.2.51)

f ( M )= 0

(3.2.52)

-(€)

(3.2.53) Since there are three boundary conditions imposed on a second-order differential equation, a relationship must exist between M and E. Since E is small, f is small, hence the zeroth-order term in the straightforward expansion is the solution of the linearized form of (3.2.50); that is (1 - 5 2 ) -d y dx2

1df =0 + -xdx

(3.3

3.2.

LIGHTHILL'S TECHNIQUE

85

Using the above boundary conditions, we find that -.

f = e2[,/l

2 2

- 1 dx,

(3.2.55)

M =1

This approximate solution has no physical meaning for x > 1, and the shock Mach number must be greater than 1 for there to be any propagation. In order to determine a valid solution beyond z = 1 , and in order to determine by how much M exceeds 1 , we find it convenient to transform the second-order equation (3.2.50) into a system of two first-order equations by letting -df= g

(3.2.56)

dx

Then we assume the following expansions

+ €"fi+ . . g = E 2 g o + e4g, + x = s + 2x1(s) + e4x2(s)+ . . . M = 1 + C ~ M+, r4M, + . . f = €yo

(3.2.57)

*

(3.2.58)

*

(3.2.59) (3.2.60)

*

The zeroth-order term is given by (3.2.54) if x is replaced by s; that is

(3.2.61)

The first-order problem is (1 - s2)g;

+

S

- (1 - sZ)g;x; + [-2sx, S

As s --t 1, go -+ J 2 ( 1 - s);fO

+ ( y + 1)sgo - ( y - 1)folg; goxl = 0

- 4f; = g ,

+go4

2 (1

(3.2.62) (3.2.63)

- s ) ~ ' ~hence , (3.2.62) becomes

-

-xl + y + 1 + O ( , / G ) as s 1 (3.2.64) J2(1 - S) Thus gl will have a singularity at s = 1 unless x1 = 0, and gl(l) = y 1 as a consequence. g, =

+

86

THE METHOD OF STRAINED COORDINATES

In order to determine M,, we use the boundary conditions at the shock. If S corresponds to the position of the shock x = M, then to order e2 d= 1

+ €2Ml+

(3.2.65)

- * .

and the boundary condition (3.2.53) gives 2g0(1

+ €2M,+ - - *) +

* * *

+-

4

= -c2M1 Y + l

Substituting for go from (3.2.61) and equating the coefficients of sides, we obtain MI = 0. Hence S= 1

+ e4[[M,

- s,(l)]

+

* *

.

Then (3.2.52) and (3.2.57) give

(3.2.66)

* *

c2

on both (3.2.67) (3.2.68)

h(1) = 0 The second-order equation gives for g2

In order to remove the singularity from g,, we set 5 2

Hence d = 1 + e4[[M2 - ( y condition (3.2.53) gives J
+

whose solution is M , = 3(y

= ii(y

+ 1),/2] - 2M,

+ 11,

(3.2.70)

and, to fourth order, the boundary

4Mz + ( y + 1) = Y+l

(3.2.71)

+ 1)2/8.Consequently

M =1

+ 3(Y + 1yE4+

0(€6)

(3.2.72)

Pandey (1968) treated the case of a solid cylinder expanding uniformly in still water rather than air. 3.2.4. SUPERSONIC FLOW PAST A THIN AIRFOIL The fourth example in the application of Lighthill's method is the determination of a uniformly valid expansion for supersonic flow past a thin airfoil (Legras, 1951, 1953) discussed in Section 2.1.3. To carry out the expansion, it is more convenient to transform the original second-order differential equation into a system of two first-order equations by letting 11

= 422,

v = 4"

3.2.

LIGHTHILL'S TECHNIQUE

87

in (2.1.19) through (2.1.21) and obtaining u,

- B2u, = M 2 [ ( y+ l)uu, u

+ ( y - l)uo, + 2uu, + cubic terms]

+ ETU, + . .

+ u + €Tug+ . .

1

(3.2.74)

u, = 0, *

(3.2.73)

= c ~ ' ( z ) at

y =0

(3.2.75)

*

u(z, y) = u(z, y) = 0

(upstream)

(3.2.76)

The straightforward expansion of this problem was found in Section 2.1.3 to break down as y + 03. Since u and D vanish upstream, a uniformly valid expansion can be obtained by straining the linearized outgoing characteristic (i.e., z - By = a constant). Thus we let 24

=

%(t,

0 = EUl(5,

where z

- BY= E

+ 7) +

7)

'"2(t,

E2U2(5,

+ 7) + . . 7)

'

(3.2.77)

*

(3.2.78)

*

+ ~ G i ( 5 7), + e2Gz(5, 7) +

*

*

(3.2.79) (3.2.80)

Y = 7

The straining functions Gi can be determined by imposing the condition that (3.2.77) and (3.2.78) be uniformly valid for large distances; that is, uz/ul and 02/ul are bounded. This condition has been shown to be equivalent to requiring 5 to be the outgoing characteristic of the nonlinear equations (Lighthill, 1949a; Whitham, 1952, 1953; Lin, 1954; Fox, 1955). Since the characteristics of (2.1.19) are given by [I - M2(7' - l)+=

+

*

+ [2M24, + *) dx dy - [B2 + M 2 ( y + l)$, + *](dy)' = 0

*](d~)'

*

* *

(3.2.81)

the equation for the outgoing characteristic is (3.2.82)

where c =B

M2 +( [ B 2 ( y - 1) + ( y + 1)lu - 2BU) + . . . 2B

(3.2.83)

Equation (3.2.82) can be rewritten as (3.2.84)

88

THE METHOD OF STRAINED COORDINATES

The problem is thus reduced to expanding the dependent variables u and D as well as the independent variable x in terms of Z, the variable 7 = y 7 and the outgoing characteristic 5. Thus (3.2.79) is equivalent to where

= x0(t7 7)

xo = E

+

+ Bq

r) + E2x2(E7

Exl(t,

7)

+.

(3.2.85)

* *

and xi= G , for i 2 1

(3.2.86)

To fix the parametrization we need to place an initial condition on 2, This condition is taken to be x(E, 0) = t (3.2.87) which is equivalent to choosing Gi to vanish at y = 0. To transform from the independent variables x and y to that

a --

at

and 7,we note

a

x5z

a + y,- a = c - a + a ax ay ax ay

a

- = 5,-

a7

on account of (3.2.80) and (3.2.84). Hence (3.2.88) Substituting (3.2.77), (3.2.78), and (3.2.85) into (3.2.73) through (3.2.76), (3.2.83), (3.2.84), and (3.2.87), using (3.2.88), and equating coefficients of equal powers of E, we obtain (3.2.89) (3.2.90) (3.2.91)

XdE,

0) = 0

(3.2.93)

The solution of (3.2.89) through (3.2.91) that vanishes upstream is u1 = T ' ( t ) ,

UI

= -B- 'T' (E )

(3.2.94)

which coincides with the linearized solution. Then (3.2.92) becomes Zlv

= --gM4(y

+ l)B-ZT'(E)

(3.2.95)

3.2. whose solution subject to (3.2.93) is Z,

= -+M4(y

LIGHTHILL’S TECHNIQUE

+ l)B-’yT’(t)

89

(3.2.96)

Therefore a first-order uniformly valid expansion is given by the first terms in (3.2.77) and (3.2.78) where 5

- By = E

- +M*(y

+ 1)Bp2yT’([) + O ( 3 )

(3.2.97)

on account of (3.2.85), (3.2.86), and (3.2.96). This solution shows that the uniformly valid first-order expansion for a hyperbolic system of equations is simply the linearized solution with the linearized characteristic replaced by the characteristic calculated by including the first-order nonlinear terms. Higher-order approximations can be obtained in a straightforward manner. A second-order expansion was obtained by Lee and Sheppard (1966). For general problems the velocity potential 4 does not vanish upstream. In this case a uniformly valid expansion can be obtained by expanding the dependent as well as both of the independent variables 5 and y in terms of 6 , and both characteristics E and y of the nonlinear equations. Thus we augment (3.2.83) and (3.2.84) by equations describing the ingoing characteristic y and introduce another expansion for y similar to (3.2.85). We next illustrate such a procedure for a more general system of hyperbolic equations. 3.2.5. EXPANSIONS BY USlNG EXACT CHARACTERISTICSNONLINEAR ELASTIC WAVES In the hyperbolic differential equation discussed above, a uniformly valid expansion was obtained by straining one of the characteristics of the linearized equation. The resulting strained variable was a better approximation to the exact characteristic. Lin (1954) and Fox (1955) generalized Lighthill’s technique for problems of hyperbolic differential equations in two independent variables by adopting characteristic parameters as the independent variables, a procedure that amounts to straining two families of characteristics. Thus they were able to treat general waves in fluid flow in which the ingoing and outgoing waves interact. This method was applied by Verhagen and Van Wijngaarden (1965) to the hydraulic jump problem. Guiraud (1965), Oswatitsch (1969, and Zierep and Heynatz (1965) applied it to gas dynamics waves of finite amplitude. Gretler (1968) devised an indirect method to calculate the plane flow past an airfoil, while Van Wijngaarden (1968) analyzed the oscillation near resonance in open pipes. Thermally driven nonlinear one-dimensional oscillations were treated by Chu and Ying (1963) and Rehm (1968) and by Gundersen (1967) for a conducting fluid. Chu (1963) and Mortell (1971) studied self-sustained oscillations in a pipe. Lick (1969) analyzed propagation of waves in isentro’pic and chemically reacting compressible fluids, while

90

THE METHOD OF STRAINED COORDINATES

Lesser (1970) investigated wave propagation in inhomogeneous media. Parker and Varley (1968) treated the nonlinear interaction of stretching and deflection waves in elastic membranes and strings, while Mortell and Varley (1971) discussed the nonlinear free vibration of an elastic panel. Richmond and Morrison (1968) applied this technique to an axisymmetric plasticity problem, while Davison (1968) obtained a second-order expansion for nonlinear elastic waves in an isotropic medium by using two characteristics as independent variables. Now, we explain the technique by determining a firstorder expansion for nonlinear elastic waves in anisotropic materials. If u and v represent the displacements in the x and y directions, then (3.2.98) (3.2.99)

where p is the material density. We let P = u, and Q = v, and assume the stresses c and 7 to be polynomials in P and Q such that (3.2.100)

-

as P and Q 0. Here, iland p are the Lame moduli of the linear theory of elasticity. Thus - = C,'P

+ $alP2 + azPQ + ta,Q2 + . .

7

+ +blP2 + bzPQ + $b3Q2+ . .

(T

P

- = C>Q

where

P

(3.2.101)

*

(3.2.102)

(3.2.103)

with c, and c, the principal and shear speeds of propagation. By letting

R

= ut and

(3.2.104)

S = vt

and substituting (3.2.101) and (3.2.102) into (3.2.98) and (3.2.99), we have

+

R,- cP2Pz= UP, fIQ, where

+ uZQ, = blP + bZQ,

*

*

+ aQ, + B = a2P + a3Q 6 = bzP + b3Q

St - c,2QZ = ~ p ,

a = alp

+

* *

(3.2.105) (3.2.106) (3.2.107) (3.2.108)

3.2.

LIGHTHILL’S TECHNIQUE

91

Since P = u, and Q = uz, (3.2.104) gives Pt=

4,

Qt

=

sz

(3.2.109)

Note that we have replaced the system of the two second-order differential equations (3.2.98) and (3.2.99) by a system of four first-order equations, which is more convenient for the application of the method of strained coordinates. To complete the problem formulation, we need to specify the initial conditions. We consider the case investigated by Davison (1968) in which the material is initially unstressed, at rest, and occupying the half-space x 2 0 when a disturbance is introduced at x = 0; that is Q(0, t ) = q ( t ) for t 2 0

P(0, I ) = e + ( t ) ,

P ( z , 0 ) = Q(z, 0) = R ( x , 0) = S ( x , 0 ) = 0 for x 2 0

(3.2.110)

with

+

+ ( t ) = y ( r ) =I 0 for

r

0

(3.2.1 11)

where and y are known functions and is a small but finite dimensionless quantity. The condition (3.2.111) is equivalent to the vanishing of P and Q along the ingoing characteristics. To obtain a uniformly valid expansion for the above problem, we expand the dependent as well as the independent variables in terms of E and the outgoing characteristic parameters E and 7. Thus we let

+ 7) + - . Q= 7) + e2Qe,(6, 7) + R = € R i ( E ,7) + c2&(E, 7) + s = ESl(E, 7) + c2Sz(E, 7) + . fJ = €Pl(E, 7)

C2Pz(E,

€C?i(E,

z = 50(E,

r) +

t = t o ( f , 7)

+

EXl(5, Efl(E,

(3.2.112)

*

* * *

(3.2.113)

* * *

(3.2.114)

* *

(3.2.115)

+ e2x2(E, 7) + . . . 7) + ~ ~ t , (7)f ,+ - . 7)

+

(3.2.116) (3.2.117)

To first-order quantities in P and Q , the characteristic wave speeds of

(3.2.105), (3.2.106), and (3.2.109) are given by

(3.2.118)

Thus to O ( P , Q ) the outgoing characteristics E and 7 are given by xrl = c,t,

(3.2.119)

x 5 = CZtC

(3.2.120)

92

THE METHOD OF STRAINED COORDINATES

where c1 and cz are the positive speeds in (3.2.118). To fix the parametrization we need to place initial conditions on x and 1. These conditions are taken to be (3.2.121) 4 5 , 5) = 0 and t ( 5 , 5) = 5 In terms of these new independent variables, the initial conditions (3.2.1 10)

To transform from the independent variables x and t to 5 and q , we note that a --- - a + at- _ a = at

a6

ax

atax

-(- a + $)

a = at

-(t$

aq aq at a -ax

Hence

-(-ata + .$)

at

atat

1

c2

- c1 t

a

(3.2.123)

/---

a)

(3.2.124)

Substituting (3.2.1 12) through (3.2.1 17) into (3.2.105), (3.2.106),(3.2.109), and (3.2.118) through (3.2.122), using (3.2.124), and equating coefficients of equal powers of E , we obtain Order

EO

Order

E

zo, - c,to, = 0

(3.2.125)

3.2.

LIGHTHILL’S TECHNIQUE

93

The solution of (3.2.125)through (3.2.127)is

(3.2.136) which is simply the linearized characteristics

Substituting for to from (3.2.136)into (3.2.128) through (3.2.131) and solving the resulting equations subject to (3.2.135),we obtain Pi(6, 7) = $ ( E l , &(Et

7) =

7) = ~ ( 7 )

-cA(E),

Sl(t7 7) = --#(TI)

(3.2.137)

which are, except for an E factor, simply the solution of the linearized problem. With this solution (3.2.132)and (3.2.133) become

- CVt1), = rla,b(E) + rlaZy(7) (3.2.138) (zl - ‘st1){ = ‘2b24(6) + r2b3W(7) where (rl,l?,) = -(1/2)(c, - C ~ ) - ~ ( C ~ / C , , -cp/cs). The solution of (3.2.138) (21

subject to (3.2.134)is 21 %1

- Cph - ‘stl

= rlal(77

1 5

=

rZb2

Jg

- WO) d<

+ rla2s*Y(0 4

+

5

r2b3(E

(3.2.139)

- 7)w(q)

Therefore a first-order uniformly valid expansion is

where 5‘ and 7 are given by

As in the problem of supersonic flow past a thin airfoil treated in the previous section, the first-order uniformly valid expansion is simply the linearized solution with the linearized characteristics replaced by the characteristics calculated by including the first-order nonlinear terms.

94

THE METHOD OF STRAINED COORDINATES

The solution can be continued in a straightforward manner to higher orders. For the isotropic case the solution was carried to second order by Davison (1968) and Nair and Nemat-Nasser (1971) for homogeneous and inhomogeneous materials, respectively. 3.3. Temple's Technique In order to determine a uniformly valid expansion for the problem du = F(x,u, c) dx

with u(xo) = uo

(3.3.1)

Temple (1958) introduced a new independent variable s, as Lighthill did, and assumed that u = u(s, c) and x = z(s, e) (3.3.2) Whereas Lighthill assumed that (3.3.3) u = u&) EU1(S) €2U&)

+

x =s

+ €XI(S) +

+

+--+ -

€%&)

*

*

(3.3.4)

and chose zi so that the above two expansions are uniformly valid, Temple replaced the original equation (3.3.1) by two new equivalent equations

du dx - = U ( z , u , s, c), (3.3.5) = X ( x , u , s, €) ds ds such that U and X are regular in c. Then he determined a straightforward perturbation expansion for u and x. Thus Temple's technique determines zi in a systematic manner. A similar approach has been used by Whitham, Lighthill, Fox, Lin, and Davison and discussed earlier in the case of hyperbolic equations where the uniformization was achieved by expanding in one or more characteristic parameters. As an example, let us consider

+

dY -

+ +

~ ( 1 )= e-' (3.3.6) (2 z):>y= 0, dx This example has been discussed by Temple, and it is a special case of the problem given by (3.2.9). Temple replaces the above equation by (z

EY)

+

dx + €y, s dY - = -(2 x)y (3.3.7) ds ds These equations are analytic in E and possess the following expansions s -= z

y = s-%-~[ 1 -

~l+(t) + dg

O(2)

(3.3.8)

3.4.

RENORMALIZATION TECHNIQUE

95

where (3.3.10) ASS-O

x =s-g y = s-2

Hence at z

=

0 y=

+

~ - 2

+

- +-4

(:T+

o(9-4)

o(E2s-6)

(3.3.1 1) (3.3.12) (3.3.13)

0(c-1/3)

in agreement with (3.2.33) obtained by Lighthill's technique. 3.4. Renormalization Technique

Pritulo (1962) showed that in order to determine a uniformly valid perturbation expansion for a given problem we need not introduce the transformation (3.2.2) into the differential equations and determine 5,. Instead, we carry out the straightforward expansion in terms of the original variables and then introduce the transformation (3.2.2) into this straightforward expansion. In order to render this expansion uniformly valid, we impose the condition of Lighthill that the singularities do not grow stronger as the order of approximation increases. Thus we obtain algebraic equations for the determination of 6, rather than differential equations, thereby simplifying the whole procedure. However, Pritulo assumed that the coefficients of the series (3.2.1), with the exception of perhaps uo, satisfy linear equations, and thus he asserted that under this condition his method becomes effective. This technique was rediscovered by Usher (1 968). This technique is closely related to the method described in Section 7.4.2, which was originated in the work of Rayleigh on scattering. After determining the scattering from a thin slab, Rayleigh recast it into an exponential to render it valid for scattering from many slabs. We apply this technique to several examples treated earlier in this book and render them uniformly valid. 3.4.1. THE DUFFING EQUATION

This problem was introduced in Section 2.1.1 where we determined the straightforward perturbation expansion u = a cos t

+ <$[-it

sin t

+ .I,(cos

3t

- cos t)] + O ( E ~ )

(3.4.1)

A uniformly valid expansion was obtained in Section 3.1.1 by using the Lindstedt-Poincare procedure.

96 THE

METHOD OF STRAINED COORDINATES

To determine a uniformly valid expansion from (3.4.1), we introduce the transformation (3.1.2) into the above series. Expanding and collecting coefficients of equal powers of E, we have u = Q cos s

- ~ [ u ( w+~%a2)ssin s - .I,a3(cos 3s - cos s)]

+ O(e2)

(3.4.2)

The secular terms in (3.4.2) are eliminated if 0 1

=

Therefore a uniformly valid expansion is where

u = Q cos s

+ &€d(COS

t = s(1

(3.4.3)

-%a2

3s

- cos s) + O(E2)

- *€d)+ O(E2)

(3.4.4) (3.4.5)

which is in agreement with (3.1.15) and (3.1.16) obtained by using the Lindstedt-Poincar6 procedure. 3.4.2. A MODEL FOR WEAK NONLINEAR INSTABILITY As a second example, we render uniformly valid the straightforward expansion u = E cos ult cos kx

9 + e3(320, t sin ult cos kx + terms bounded as t -+03 )

(3.4.6)

obtained in Section 2.1.2 for the model problem (2.1.10) and (2.1.11). We let t = s(l €em2 * .) (3.4.7)

+

+-

in (3.4.6) and expand for small E to obtain u = E cos uls cos

kx

+ c3[(-&

sin (ils cos kx

- G,uJ,),

+ terms bounded as s -+

Secular terms will be eliminated if w 2 = 9/32a12. Therefore a uniformly valid expansion is u = E cos at cos kx O(e3) (3.4.8) where

,[,-

+

9e2 (3.4.9) 32(k2 - 1) If k > 1, c is real and (3.4.8) is valid for times as large as O(E-~)and it represents oscillatory standing waves with amplitude-dependent frequencies.

a=v/k2-

3.4.

RENORMALIZATION TECHNIQUE

97

However, if k < 1, 0 is imaginary and (3.4.8) represents growing waves. Since after a short time cosh 35t, where d is real, dominates cosh d t , (3.4.8) is valid only for short times. Equation (3.4.9) shows that (T co as k + 1, and the second term on the right-hand side is of the same order as the first term when k - 1 = O ( E ~Therefore ). although the above expansion is valid for a wide range of k , it breaks down when k - 1 = O(e2). We show in Section 3.5.1 that the application of the method of strained parameters to determine an expansion valid near k = 1 leads to erroneous results. An expansion valid near k = 1 is obtained by using the method of multiple scales in Section 6.2.8. ---f

3.4.3. SUPERSONIC FLOW PAST A THIN AIRFOIL As a third example for the application of the renormalization technique, we render uniformly valid the straightforward expansion for the velocity component in the axial direction obtained in Section 2.1.3 for flow past a thin airfoil. From (2.1.36) the straightforward expansion is

In order to render this expansion uniformly valid, we let

6 = s + €61(S, Y ) + O ( 4

(3.4.1 1)

in (3.4.10), expand, and collect coefficients of equal powers of

U -I-U

E

to obtain

+ 2[+( 1 - M‘(4YBht 1)) T q S ) - T(s)T”(s)

ET‘(s) B

Y + l M4yTf(s))?]

+ O(e3)

(3.4.12)

This expansion can be rendered uniformly valid for all y by choosing E1(S,

y) =

- y+M l4 y T ’ ( s ) 2B2

(3.4.13)

Therefore a uniformly valid expansion is given by U = 1-E TYs) -

U

B

+

0(E2)

(3.4.14)

98

THE METHOD OF STRAINED COORDINATES

where $=s-€-

+

2B2

M4yT'(s)

+ O(2)

(3.4.15)

which is in full agreement with (3.2.94) and (3.2.97) obtained by using Lighthill's technique. 3.4.4. SHIFT IN SINGULARITY As a fourth example, we consider the problem given in (2.4.1). The straightforward expansion was obtained in Section 2.4.1 to be (3.4.16)

To render this expansion uniformly valid, we let x =s

+ €X1(S) + -

(3.4.17)

* *

in (3.4.16) and collect coefficients of equal powers of

E

to obtain (3.4.18)

The above expansion can be rendered uniformly valid by choosing XI

=

-gs-2

(3.4.19)

to remove the worst singularity. Therefore a uniformly valid expansion is given by y = sPe-" O(E) (3.4.20) where 2 = s - i.s-2 WE2) (3.4.21)

+

+

which is in full agreement with (3.2.30) and (3.2.31) obtained by Lighthill's technique. 3.5. Limitations of the Method of Strained Coordinates

In the preceding sections it was evident that the method of strained coordinates is a powerful technique for determining uniformly valid expansions for widely varying physical problems. However, in spite of the successful treatment of hyperbolic differential equations for waves traveling in one or two directions, the method may not yield uniformly valid expansions for elliptic differential equations. Although Lighthill (1 951) obtained a uniformly valid second-order expansion for incompressible flow past a rounded thin airfoil, Fox (1953) found higher-order expansions are not uniformly valid.

3.5.

LIMITATIONS OF STRAINED COORDINATES

99

She also found that even a second-order uniformly valid expansion could not be obtained for the case of compressible flow past a thin airfoil. Thus Lighthill (1961) advised in a later article that his method be used only for hyperbolic differential equations. In spite of this, Vaglio-Laurin (1962) successfully applied this technique in conjunction with the method of integral relations to the blunt body problem (a mixed boundary value problem). Moreover, Emanuel (1966) and Kuiken (1970) successively applied this technique to parabolic problems involving, respectively, unsteady, diffusing, reacting tubular flow, and flow down an inclined surface originated by strong fluid injection. It should be mentioned that Hoogstraten (1967) modified this technique to treat subsonic thin airfoil problems by introducing a function approximating uniformly a mapping of the physical plane onto a plane in which the airfoil is represented by its chord. It was suggested by Tsien (1956) that the failure of the method of strained coordinates in treating the thin airfoil problem is the expansion of a function near an irregular point. Fortunately, one can see that the singularities are transferred from the dependent variables to the straining functions and thus realizes the nonuniformity of the resulting expansion. Yuen (1 968) expanded a function near an irregular singular point to obtain an expansion valid near the cutoff wave number in the nonlinear stability of a cylindrical liquid jet. However, the resulting expansion did not have any singularity although it fails at the cutoff wave number as shown by Nayfeh (1970~).We show the difficulties encountered by Yuen by using a model problem for weak nonlinear instability of a standing wave (Section 3.5.1). The method of strained coordinates has been shown by Levey (1959) to fail for the class of singular perturbations in which the small parameter multiplies the highest derivative (Section 3.5.2). He showed that this method yields erroneous results for the problem of cylindrical shock waves. However, it can be shown that straining the dependent rather than the independent variable leads to a uniformly valid expansion (Exercise 3.33). Although this technique yields uniformly valid expansions for the periodic orbits of weakly nonlinear oscillations, it was shown by Nayfeh (1966) that it does not yield any information about any solution besides the limit cycle and limit points. In general, if the amplitude is varying, then the method of strained coordinates does not apply. We show the difficulties of the method of strained coordinates with the following examples. 3.5.1. A MODEL FOR WEAK NONLINEAR INSTABILITY As discussed in Section 3.4.2, the expansion (3.4.9) for u breaks down when k - 1 = 0 ( c 2 ) . To apply the method of strained parameters to this

100

THE METHOD OF STRAINED COORDINATES

expansion, we let

k =a

+ e2k,

(3.5.1)

in (3.4.9), expand for small E , collect coefficients of powers of u = J.B-[l-

c

- ak2] aa - I

c,

and obtain

+ q23)

(3.5.2)

In order that the coefficient of c2 be no more singular than the first term as a 1, we choose k, to be 9/32. Then (3.5.2) becomes --j

+

which is bounded as a -+ a = 1 , or from (3.5.1)

]

92 + .. (3.5.3) 32(a 1) I. Neutral stability corresponds to cr = 0 ; that is, k =1

+ &c2 +

(3.5.4)

To show the invalidity of the expansion (3.5.3), it is enough to show the invalidity of the neutral stability condition (3.5.4). The neutral stability configuration is by definition independent of t , hence it is governed by the equation u,, u = -u3 (3.5.5) Letting

+

u = E cos k x

+ 2 A , cos nkx, m

A , = O(c2)

n=2

in (3.5.5) and equating the coefficient of cos kx to zero, we obtain

Hence (3.5.6)

which is different from (3.5.4). Therefore the expansions (3.5.3) and (3.5.4) are incorrect. A correct expansion valid near k = 1 is obtained in Section 6.2.8 by using the method of multiple scales. A SMALL PARAMETER MULTIPLYING THE HIGHEST DERIVATIVE Levey (1959) showed that the application of the method of strained coordinates to the problem of cylindrical shock waves (introduced in Exercise 2.3) leads to incorrect results. The thickness of the shock was found by Wu (1956) to be independent of its strength, contrary to the result obtained by Levey using topology analysis. Rather than showing the invalidity of the expansion by discussing the problem of cylindrical shock waves, we follow 3.5.2.

3.5.

LIMITATIONS OF STRAINED COORDINATES

101

Levey and discuss a simpler problem that has the same features and has an exact solution for comparison. The problem is (3.5.7)

where E is a small positive number. Its exact solution which passes through the point (xo,yo) is

Y = Yoe(so'-z')/zc

-

I

dt

e-zvfe*2/ze

(3.5.8)

E

The straightforward perturbation expansion of (3.5.7) can be obtained by letting 00

Y = 2 E"Y,(Z)

(3.5.9)

n=O

Substituting (3.5.9) into (3.5.7), equating coefficients of equal powers of and solving the resulting equations, we obtain y =

-x-1

- €5-3 - . . . - 1.3.5

* * *

(2n -

~)E"Z-~"-'

* *

.

E,

(3.5.10)

It can be verified that (3.5.10) is the asymptotic expansion for large x of the exact solution (3.5.8). It can be seen that the above expansion breaks down near x = 0 because the first term is singular and higher-order terms are more and more singular. When x = O ( E ~ ' ~all ) , terms in the above expansion are of O(E-~'~). Thus the above expansion is never an adequate expansion in the region x = O(E~/~). To apply the method of strained coordinates to this problem, we let z =s

+

EX&)

+

E%,(s)

+ . ..

(3.5.1 1)

in (3.5.10), expand for small E, and collect coefficients of equal powers of E . The straining functions x, are then chosen so that the higher-order terms are no more singular than the first. In this problem this amounts to eliminating all the terms except the first. The result is y=

where

- -1 S

(3.5.12) (3.5.13)

102 THE METHOD

OF STRAINED COORDINATES

It follows from the last relations that a,

> 2n-yn - l)!,

n

>2

(3.5.15)

Therefore the expansion (3.5.13) is divergent; in fact, it is “more” divergent than (3.5.10), and the expansion breaks down as z approaches O ( E ~ ’ ~All ). that has been achieved is the exchange of an invalid expansion of one variable for an invalid expansion of the other. The reason for the breakdown of the expansion is the dropping of the highest derivative whose effect is small for large z but becomes significant as z approaches the region O ( P ) . 3.5.3. THE EARTH-MOON-SPACESHIP PROBLEM Next we show that the application of the method of strained coordinates to the two-dimensional earth-moon-space problem (introduced in Section 2.4.2) leads to an invalid expansion. To render the expansion (2.4.17) and (2.4.1 8) uniformly valid using the method of strained coordinates, we substitute (3.5.16) x =s pl(s) .*

+

+ -

into this expansion and obtain

J2

1 t = 3 sin-’ pJs P

1 -7 Js(l P

- p2s)

(3.5.17) The straining function x1 is chosen so that the term between the brackets is no more singular than the first term as s 1 ; that is XI =

-

1

2(1 - $1

ln(1 - s)

(3.5.18)

The straining function could have been chosen to eliminate the term in the brackets. The resulting first-order expansion is then

y = -#us

+ O(F2)

(3.5.20)

EXERCISES

103

where

(3 S . 2 1) Comparing this expansion with the exact solution, Nayfeh (1965a) showed that this expansion gives a divergent trajectory near the moon although the expansion obtained in the one-dimensional case agrees fairly well with the exact solution. In the one-dimensional case there is a singularity at x = 1 p/(l - p z ) O(p2) which is outside the range of interest 0 5 x 5 1. In the straightforward expansion the singularity is shifted to 5 = 1, a n d the straining of z moves the singularity from 5 = 1 toward its right position. However, in the two-dimensional case there is a sharp change in the direction of the spacecraft in the neighborhood of the moon, and the straining which is O ( p ) cannot cope with such a sharp change.

+

+

Exercises 3.1. Consider the problem

ii

+ u = €2,

u(0) = a,

ri(0) = 0

(a) Determine a second-order (three-term) straightforward expansion and discuss its uniformity. (b) Render this expansion uniformly valid using the method of renormalization. (c) Determine a first-order (two-term) uniformly valid expansion using the method of strained parameters and compare the result with (b). 3.2. (a) Show that the motion of a point mass that moves freely along a parabola z2 = 2pz rotating about its axis with an angular velocity w is given by

(b) Determine a two-term straightforward expansion for small amplitude and discuss its uniformity. (c) Render this expansion uniformly valid using the method of renormalization. (d) Determine a one-term uniformly valid expansion using the method of strained parameters and compare the result with (c). 3.3. Determine a two-term uniformly valid expansion for small amplitudes for the solution of

which describes the oscillations of a pendulum.

104

THE METHOD OF STRAINED COORDINATES

3.4. Determine a second-order uniformly valid expansion for the periodic solution of ii u = r(1 - u2)U

+

Note that the amplitude is not arbitrary. 3.5. Determine a first-order uniformly valid expansion for the periodic solution of ii u = €(1 - z)U

+

Ti

where

is a constant. 3.6. Consider the equation

+z =

u2

7

ii

+ wO2u= €2+ k cos w t

Determine first-order uniformly valid expansions for the periodic solutions when (a) wo M 2w [Hint: let wo = 2w PU and u = uo cul . . . where uo = a cos (2wt B) ( 1 / 3 ) k ~ cos - ~ w t , then determine a and 3/ from the equation for u,], and (b) wo M w / 2 (a is arbitrary in this case).

+

+ +

3.7. Consider the equation ii

+

+

+ wo2u = cu3 + k cos w t

Determine first-order uniformly valid expansions for the periodic solutions when (a) wo M 3 0 and (b) wo M 4 3 . 3.8. Determine second-order expansions for the odd solutions corresponding to the transition curves of ii (6 € cos 2t)u = 0 when 6 is near 1 and 4. 3.9. Consider the equation

+ +

ii

+

1

+ 6u

€COS2t

=o

(a) Determine second-order expansions for the transition curves near

6 = 0, 1, and 4 (Shen, 1959). (b) Use Whittaker’s technique to determine second-order expansions for

u near these curves. 3.10. Consider the equation

ii

+

6 1

- ccos2t - ECOS2t u = o

(a) Determine second-orderexpansions for the first three transition curves (Rand and Tseng, 1969) (Le., near 6 = 0, 1, and 4). (b) Use Whittaker’s technique to determine u near these curves. 3.11. Consider the equation ii

+ (6 + rcos3t)u = o

EXERCISES

105

Determine second-order expansions for the first three transition curves using both the method of strained parameters and Whittaker’s technique. 3.12. Determine a periodic solution to O ( E )of the problem (Mulholland, 1971) ...

u

+ ii + U + u

= (1 - u2

-2

- ii2)(ii

+ U)

3.13. Determine first-order expansions for ii

+ lu = EU3

subject to (a) u(0) = u ( n ) = 0, and (b) u ( t ) = u(t 3.14. Determine a first-order expansion for ii

+ 277).

+ lu = r(sin 2t + uz)u u ( r ) = u(t + 2a)

3.15. Determine first-order expansions for

+ ilu = €tu U O ) = U ( f + 277) (b) ii + Au = .(a cos 2t + u ( t ) = n ( t + 277) (a) ii

sin 2r)u

3.16. Determine a first-order expansion for small amplitudes of the periodic solution of k t - x + g ( i - c0s e) - ( I z)~2 = o m

+

+

e + -l + x

sine

+

xe

=o

which describe the oscillation of a swinging spring of length I and constant k when oI2= k / m m 4oz2 = 4gll. 3.17. The free vibrations of a simply supported beam on an elastic foundation are given by u,,,, yu EYU3 Ut* = 0

+ +

u(0, t ) =

U(T,

+

t ) = u,,(O, t ) = u Z z ( a ,t ) = 0

u(x, 0) = u sin 2,

ut(z, 0 ) = 0

where >’, a , and t are constants. Determine an expansion to O(E)of the frequency of oscillation (Han, 1965). 3.18. Carry out the expansions of Sections 3.1.4 and 3.1.5 to second order. 3.19. Consider uniform traveling wave solutions of Utt

- uzz

+ u = €U3

106 THE METHOD OF STRAINED COORDINATES of the form u = a exp i(k.l: - w t ) and wave number shifts. 3.20. Consider the problem

+ higher harmonics and determine the frequency

+ u,, + u,x2r = €2

Utt

U ( X ,0) = a cos kx,

u~(z, 0) = 0

(a) Determine a first-order straightforward expansion. (b) Render this expansion valid using the method of renormalization. (c) Determine an expansion valid for t = O(e-l) using the method of strained parameters. (d) Show that the frequency is invalid near k = 1. (e) Apply the method of renormalization to this frequency to remove the singularity. (f) Show that the resulting expansion is erroneous. 3.21. Consider the eigenvalue problem +xx

+ +YV + A+

= ex2+

+ ( x , 0 ) = C(x, r) = C(0, Y) =

+(X,

Y)

=0

Determine first-order expansions when 1 is near 2 and 5 . 3.22. Consider the problem

v2+ + A+

+

=

E

f ( x , Y,

z)+

with vanishing on the surfaces of a cube of length X. Determine first-order expansions when I = 3 and 6 if (a)f = z2and (b)f = z2y. 3.23. The free transverse vibrations of a simply supported beam are given by ElwrZZz

-

T=

+ =0 ’( w , ) ~d.z

Tw,,

ES

w ( 0 , I ) = w(1, t ) = W,,(O, w(x,O)

=

TX

asin-

I

,

PW~(

I) =

w,,(l,

t) =0

wt(x,O) = 0

where E, I, p. S, a, and lare constant. Determine a first-order expansion for small amplitude (Evensen, 1968). 3.24. Consider the problem (.1:

+ €Y)Y’ + y = 0 ,

y(1) = 1

(a) Determine a second-order straightforward expansion. What is its region of nonuniformity ? (b) Render this expansion uniformly valid using the method of renorrnalization.

EXERCISES

107

(c) Determine a first-order expansion (two terms in y and three terms in z) using Lighthill's technique and compare it with (b). (d) Determine the exact solution by interchanging the roles of dependent and independent variables and compare it with (b) and (c). 3.25. Show that a uniformly valid expansion of

is where z = t - ~ ( Inb t + b + 1) 3.26. Consider the problem

+ O(c2).

(z + .y)y' - *y = I + 2 2 , y(1) = 1 (a) Determine a second-order straightforward expansion and discuss its uniformity. (b) Render this expansion uniformly valid using the method of renormalization. (c) Obtain a first-order expansion using Lighthill's technique and compare the result with (b). 3.27. Use the method of renormalization to render the expansion of Exercise 2.12 uniformly valid. 3.28. Consider the problem (sn

+ q ) y ' + nxn-ly Y(l) = a

- mxm--1

=O

>1

(a) Show that its exact solution is given by zny

+

&!.2 I

= xm

+ (a +

*€a2

- 1)

(b) Show that application of Lighthill's technique gives y "Yo

= (5"

+ a - ?)I

(c) Show that (Comstock, 1968) theapproximate solution is erroneous near (x = 0) except for special values of m and n. (d) Introduce a new variable z = xn in the original problem and then strain the variable z to determine an approximate solution to y. Determine the conditions under which this new expansion is valid near the origin (Burnside, 1970), hence determine the role of changing the independent variable on rendering the approximate solution uniformly valid. 3.29. Consider the problem (1

+

EU)

au

ax

au += 0, ay

u(z,0) = E+(Z)

108 THE METHOD OF

STRAINED COORDINATES

(a) Determine a first-order straightforward expansion for E << 1 and discuss its uniformity. (b) Render this expansion uniformly valid using the method of renormalization. (c) Determine a first-order expansion using Lighthill’s technique and compare the result with (b). 3.30. Consider the problem Utt

- c2ux, = u,u,,

+ ( t ) = 0 for t u(x, 0) = 0 for z 2 0

u(0, t ) = ~ + ( t ) ,

0

(a) Determine a first-order straightforward expansion and render it “uniformly valid” using the method of renormalization. (b) Determine a first-order expansion using Lighthill’s technique and compare it with (a). Show that renormalizing u rather than u, leads to an erroneous result. 3.31. Consider the problem (Lighthill, 1949a)

au -ax + - X +n Y 40,Y)

u(x, 0) = u(x, 0) = 0 =

€+(y)Y-n,

0

where +(O) = 0. Show that a first-order uniformly valid expansion is u =

E+(r1)(2

+ TI-” + O ( 2 )

3.32. Consider the problem Ut*

- c2u,, = €U,U,,

4%0 ) =f(4+ g ( x ) ,

ut(x, 0 ) = c(g’(4

-fW)

where f (x)and g ( x ) are bounded functions of x. (a) Determine a first-order straightforward expansion. Can you render it uniformly valid using the method of renormalizaion? (b) Obtain a first-order expansion using the method of strained coordinates. 3.33. Consider the problem EY’

for

+ y = 1,

y(1) = 1

(a) Show that Lighthill’s technique fails to give a uniformly valid expansion. (b) Show that straining y rather than x gives a uniformly valid expansion. (c) Investigate whether straining y‘ would yield a uniformly valid expansion

+ +y =0

€yl’ y’

Y(0) = a ,

Y(1) = B

EXERCISES

3.34. Consider the problem

+u =

ii

€f(U,

109

ti)

(a) Show that the method of strained coordinates (MSC) leads to it

where

+ O(r), + = s + c t = s + et,(s) + . . .

= asin

4

Io2?[u sin 4, a cos +] cos 4 d4

ut;=a=-77

2at;

=

p

= - j o z T / [ usin 7r

4, a cos 41 sin 4 d4

+

= 0 and (b) Show that is a constant so that t , = (1/2)Bu-'s a constant. (c) Hence show that the MSC yields only the limit cycles or limit points for this problem (Nayfeh, 1966).

Perturbation Methods ALI HASAN NAYFEH Copyright Q 2004 WILEY-VCH Verlag GmbH & Co. KGaA

CHAPTER 5

Variation of Parameters and Methods of Averuging

5.1.

Variation of Parameters

This technique was originally developed to solve inhomogeneous linear equations when the general solutions to the corresponding homogeneous equations are known. As an example, consider the general linear secondorder inhomogeneous equation

Y”

+ P(4Y‘ + 4(4Y = m)

(5.1 .l)

Let yl(s) and y,(s) be two linearly independent solutions of the corresponding homogeneous equation. Then we assume that the solution of (5.1.1) is Y = Al(4Yd4

+ Az(z)Y2(4

(5.1.2)

where the functions A, and A, need to be determined. Note that in the homogeneous problem A, and A, are constants, while in the inhomogeneous case they are allowed to vary, hence the name “variation of parameters.” Differentiating (5.1.2) with respect to z gives

Y’ = AlYi

+ &Y;l + AlYl + A h 2

(5.1.3)

Since we have three unknown functions A l , A,, and y, while we have only the two equations (5.1.1) and (5.1.2), we are free to impose one more condition on A,, A z , and y. Let us demand that Aiyi

Then (5.1.3) becomes

+ A;Y~= 0

(5.1.4)

+

(5.1.5)

Y’ = AIY; AZY; Differentiating (5.1.5) with respect to z gives

y” = AlYl’

+ A,yl + A:?!; + A h ;

(5.1.6)

159

160

VARIATION OF PARAMETERS AND AVERAGING

Substituting for y, y’, and y” into (5.1.1) and using the fact that y1 and y2 are solutions of the corresponding homogeneous equation, we obtain A;yi

+ A;&

=R

(5.1.7)

Solving (5.1.4) and (5.1.7) for A; and A; yields A; = A; =

--R(4!/2(4

(5.1.8)

W(Z) R(4Yd4

W(4 where W(x)is called the Wronskian and is given by W(X> = Y,(X)YXX) The general solution of (5.1.1) is then

Y = ClYl(4

+

-YX4Y44

c2v2(4

+ Y,(4

(5.1.9)

(5.1.10) (5.1.1 1)

where c1 and c2 are constants, and the particular solution y9 is given by

This technique has been extended for the determination of solutions of problems in which the inhomogeneity is a function of both the dependent and independent variables. The dependence on the dependent variable may be nonlinear. We next discuss two examples; the first is linear, while the second is nonlinear. 5.1.1. TIME-DEPENDENT SOLUTIONS OF THE SCHRODINGER EQUATION Let us consider the Schrodinger equation (5.1.13)

subject to homogeneous boundary conditions, where H , and HI are timeindependent and time-dependent linear operators. We assume that (5.1.14)

subject to the same homogeneous boundary conditions, has the solution 00 2a y = ~ L Z ~ U , , ( S ) ~ - ’ ~ ~ ~ , w , = -E , n=l h

(5.1.1 5)

5.1.

VARIATION OF PARAMETERS

161

where a, is a constant, and u, and En are, respectively, the eigenfunctions and the corresponding eigenvalues of (5.1.16)

H,u = EM

with the same homogeneous conditions. The eigenfunctions u, are assumed to be orthonormal over the domain D. Following Dirac (1926), we assume that the solutions of the perturbed problem are still given by (5.1.15) but with time-varying a,. Substituting (5.1.15) into (5.1.13) gives

m

=

-n=l 2 H1[anun(z)e-'""']

(5.1.17)

The first term on the left-hand side of this equation vanishes according to (5.1.16). Then (5.1.17) becomes

Multiplying (5.1.18) by ii,(z), integrating over the domain D, and using the orthonormality of u,, we obtain

dam - = - - 2rri ze dt h n=l

where

iornt

(5.1.19)

Hlmn

Hlmn = /Dz7m(z)Ifl[ ~ ~ u , ( z ) e - " "d~z]

If Hl does not contain any derivative with respect to becomes dam dt

-=

where

- -27ri I'ane

2rr = - ( E m - En), h

Q,,

h

iomnt

(5.1.20) t,

then (5.1.19)

-

Hlm,

(5.1.21)

Rim, = ~ i i , ( z ) H 1 [ u n ( z ) dl x

(5.1.22)

n=l

Equation (5.1.21) is equivalent to the full problem given by (5.1.13). If H, is a small perturbation, then we can expand a, as a, = a,

+ a, +

Q,2

+ .-

(5.1.23)

+

where a,, is a constant and equal to a,,(t = 0), and am,

<< am(,-,).

Then the

162

VARIATION OF PARAMETERS AND AVERAGING

first approximation to a, is given by (5.1.24)

If in addition an, = d n k , then (5.1.24) becomes dam1 dt

2rri eiomktfllmk h

(5.1.25)

As an example, let

H l y = y f ( z )sin wt

Then where

H1,

= fmk sin wt = -iifM(eiot

(5.1.26)

- e-'ot)

(5.1.27)

f & =Suli,(z)f(z)ud4 d z Substituting in (5.1.25) and solving for a,,, we obtain

5.1.2. A NONLINEAR STABILITY EXAMPLE

The method of variation of parameters in conjunction with an expansion in terms of eigenfunctions has been developed and applied extensively to nonlinear stability problems by Stuart (1958, 1960a, b, 1961), Watson (1960), Eckhaus (1965), and Reynolds and Potter (1967). This technique has been unified and presented in a coherent manner by Eckhaus (1965). To describe this technique let us consider the following example after Eckhaus (1965) (5.1.29)

where L is a linear operator and F(+) is a nonlinear operator. Let us assume that L depends on one space variable, say x, with 0 5 x 1, and let 4 satisfy the linear homogeneous boundary conditions

B,(+) = 0 at z = 0 and B,(+) = 0 at x = 1

(5.1.30)

It is obvious that the linearized problem

a+ = 0 L(+) - at

(5.1.3 1)

5.1.

VARIATION OF PARAMETERS

163

subject to the boundary conditions (5.1.30), admits a solution of the form with

4 = u(x)e-At

(5.1.32)

+ Au = 0

(5.1.33)

L(u)

Bl(u) = 0 at x = 0 and B,(u) = 0 at x = 1

(5.1.34)

We assume that the eigenvalue problem (5.1.33) and (5.1.34) possesses a denumerable set of eigenvalues A, (real or complex), corresponding to the eigenfunctions u,, for which we can solve. The eigenvalues are assumed to be different from each other and ordered such that Real (A,) > Real (A,J. Let us assume that L is a self-adjoint operator so that the eigenfunctions u, are orthogonal, and we assume that they have been normalized so that

Then the general solution of the linearized problem can be written as m

(5.1.36)

where a, are constants which can be determined from the initial conditions. For the nonlinear problem we assume that the solution is still expressible in the form (5.1.36) but with time-varying a,, and we write it as 03

d = 1An(t)un(z)

(5.1.37)

n=l

where A,

= a,

Since L(u,) =

exp (-Ant).

-A&,,

Substituting (5.1.37) into (5.1.29) leads to

(5.1.38) can be rewritten as

Multiplying (5.1.39) by Cm(x),integrating from z = 0 to x = 1 , and using the orthonormality condition (5.1.39, we obtain

form= 1,2, . . . .

164

VARIATION OF PARAMETERS AND AVERAGING

If L is not self-adjoint, the eigenfunctions u, are not mutually orthogonal. However, we can define the adjoint operator M by P A Y , ) - YZWYl) =

d

[P(Yl?Y d l

(5.1.41)

where p1 and y2 are two functions of x and P is a bilinear form. With this definition the adjoint problem is given by Mii

+ Iii = 0

(5. I .42)

and the boundary conditions are chosen such that P(u, ii) vanishes at both x = 0 and x = 1. Under these conditions u, and 12, are orthogonal, and they can be normalized so that (5.1.43) The analysis in this case would be the same as in the self-adjoint case if we replace Em(%)by Gm(x). In particular, (5.1.40) becomes

5.2.

The Method of Averaging

5.2.1. VAN DER POL'S TECHNIQUE Van der Pol (1926) devised a technique, which is described in this section, to investigate the periodic solutions of the equation d2u -++w,2u dt2

du =€(1 -uU2)-+~klcosill dt

(5.2.1)

which is named for him. In (5.2.1) E is assumed to be small, and I (the frequency of the excitation) is assumed to differ from wo (the natural frequency) by a small quantity which is of the order of E. Under these assumptions the solution of (5.2.1) is assumed to have the form u ( t ) = a l ( t ) cos ;It

+ az(t)sin ilt

(5.2.2)

where a l ( t ) and a,(t) are assumed to be slowly varying functions of time; that is, dai/dt = O(E)and d2a,/dt2= O(e2). Differentiating (5.2.2) twice gives

+ 2azl cos I t + 5, cos At + u, sin At

ii = - P a l cos I t - 12a, sin I t - 2all sin At

(5.2.3)

5.2.

THE METHOD OF AVERAGING

165

where the overdots denote differentiation with respect to t. Substituting (5.2.2) and (5.2.3) into (5.2.1), neglecting terms of order higher than t, keeping in mind that ai = O(E) while Bi = O(e2), and equating the coefficients of cos At and sin I t on both sides, we obtain 24

+ 2a2 - ta,(l - p) = 0 1 0:

i2 - 0," 2a2 - -a, 1

where

a2

p=-=

- ta2(l - p) = tic a: + a :

4

(5.2.4)

(5.2.5)

(5.2.6)

4

To analyze the periodic solutions of (5.2.1), we note that they correspond to the stationary solutions of (5.2.4) and (5.2.5); that is, they correspond to the solutions of (5.2.7) 20azO- alo(l - po) = 0 -20ulO - azO(l- Po) = k

(5.2.8)

where u is the detuning factor, and it is given by (T=-

1- 0

0

(5.2.9)

€

Terms of O(e2) in (5.2.4) and (5.2.5) have been neglected. By adding the squares of (5.2.7) and (5.2.8) and using (5.2.6), we obtain the frequency response equation k' (5.2.10) po[4ge -k (1 - Po)'] = 4

5.2.2. THE KRYLOV-BOGOLIUBOV TECHNIQUE We discuss this technique in connection with the general weakly nonlinear second-order equation d2u + w;u dt2

When

E

=

(5.2.11)

= ff

0 the solution of (5.2.11) can be written as u = acos (mot

+ 0)

(5.2.12)

where a and 6 are constants. To determine an approximate solution to (5.2.11) for E small but different from zero, Krylov and Bogoliubov (1947) assumed

166

VARIATION OF PARAMETERS AND AVERAGING

that the solution is still given by (5.2.12) but with time-varying a and 6 , and subject to the condition du = -amo

dt

# = mot + 8

sin #,

(5.2.13)

Thus this technique is similar to van der Pol’s technique which was discussed in the previous section. The only difference is in the form of the first term. Differentiating (5.2.12) with respect to t gives du dt

-=

-am,

sin

da de +cos # - a - sin 4 dt dt

Hence

# - a dB -sin 4 = 0

da

-cos dt

(5.2.14)

df

on account of (5.2.13). Differentiating (5.2.13) with respect to t gives d2u dt2

- = -awo2 cos #

da dt

- oo- sin

4 - am, de - cos # dt

Substituting this expression into (5.2.1 1) and using (5.2.12), we obtain W,

da dt

- sin

de + + am, cos # = dt

-Ef[a

cos #, -am, sin #]

(5.2.15)

Solving (5.2.14) and (5.2.15) for daldt and deldt yields

-da- _- dt

sin + j [ a cos 4, -am, sin 41

de = - Ecos + j [ a cos 4, dt

(5.2.16)

WOO

-am,

sin 41

(5.2.17)

am0

Thus the original second-order differential equation (5.2.1 1) for u has been replaced by the two first-order differential equations (5.2.16) and (5.2.17) for the amplitude a and the phase 8. To solve (5.2.16) and (5.2.17), we note that the right-hand sides of these equations are periodic with respect to the variable 4, hence daldt = O(E) and deldt = O(E).Thus a and 8 are slowly varying functions of time because E is small; hence they change very little during the time T = 2 7 i - l ~(the ~ period of the terms on the right-hand sides of these equations). Averaging (5.2.16) and (5.2.17) over the interval [ t , t TI, during which a and 8 can

+

THE METHOD OF AVERAGING

5.2.

167

be taken to be constants on the right-hand side of these equations, we obtain (5.2.18) (5.2.19) where

Note thatf, and gl are simply two coefficients of the Fourier series expansion off As an example, let us consider Duffing’s equation (2.1.1) in which f ( u , li) =

Hence

gl(4 =

fi(4= 0,

(5.2.22)

-u3

-%a3

(5.2.23)

Consequently, a = a constant from (5.2.18), and U2

(5.2.24)

e=QE-t+e, WO

from (5.2.19). Therefore, to first approximation (5.2.25) As a second example, we consider the van der Pol oscillator in which f(u,

li)

In this case f1

= -Ww,U(l

= (1

du - 2)-

dt

- fa2),

g, = 0

(5.2.26) (5.2.27)

Hence 0 = 0 , = a constant from (5.2.19), whereas da dt

- = -(1 EU

2

- *a2)

(5.2.28)

168

VARIATION OF PARAWTES AND AVERAGING

Integrating (5.2.28)yields

(5.2.29)

The same basic idea of this technique underlies the method of Le Verrier (1856).

5.2.3. THE GENERALIZED METHOD OF AVERAGING In this technique we consider (5.2.12) and (5.2.13) as a transformation from u and duldt into u and 4 so that

-da- _-dt

sin + f [ a cos 4, -amo sin 41 0 0

f 9- mo - cos + f [ a cos +, -amo

am0

dt

(5.2.30) sin +]

where the variable t$ is called a rapidly rotating phase. Rather than integrate these equations as in the previous section, we define a near-identity transformation (see Bogoliubov and Mitropolski, 1961,p. 412)

+ 4)+ €%z(H, 4)+ (5.2.31) + = 4+ Et$r(H, 4)+ f2+,(H, 6) + . from (a, 4) to (Z,r$), which is 27r periodic in 4such that the transform of the u,= i

€Ua,(G,

* * *

*

system (5.2.30) has the form

dH- € A , ( i ) + E2Az(rf)

dt

d4 = 0 dt

0

+ ...

+ €B,(H) + E2Bz(H)+ . . .

(5.2.32)

with Ai and Bi independent of In this procedure a and 4 need not be restricted to scalar functions (Mettler 1959;Sethna 1963;Morrison, 1966b). Higher-order effects were obtained by Volosov (1961,1962), Musen (1965) and Zabreiko and Ledovskaja (1966). The inverse transformation of (5.2.31) was proposed by Kruskal (1962) and an order-by-order algorithm for this

4.

5.2. THE

METHOD OF AVERAGING

169

latter procedure was obtained by Stern (1970b). Stern (1971b) used this technique to analyze slowly varying perturbed systems. Substituting (5.2.31) and (5.2.32) into (5.2.30), expanding, and equating coefficients of like powers of E , we obtain equations of the form

(5.2.33)

where the right-hand sides are known functions of the lower-order terms in (5.2.31) and (5.2.32). In general, F, and Gn contain short-period terms (denoted by superscript s) and long-period terms (denoted by superscript I). We choose A n and B, to be equal to the long-period terms; that is A,=FL,

Then

(5.2.34)

B n = G,'

(5.2.35) which can be solved successively for a,, and 4,. As an example, we consider the van der Pol oscillator in which J ( u , zi) = (1 - UZ)li,

wo = 1

In this case (5.2.30) become

da - -- & [ 4 4 dt

- u') - 4~ cos 2 4 + a3 cos 441

d4 = 1 -I- &[2(2 dt

- a')

sin 2 4

- a'

(5.2.36) sin 4+]

Substituting (5.2.31) and (5.2.32) into (5.2.36) and equating coefficients of like powers of 6 , we have Order

c

170 Order

VARIATION OF PARAMETERS AND AVERAGING E*

- &iiul(2sin 2 4 + sin 44) + 4+1[(2 - 5') cos 2 6 - C2 cos 461

(5.2.39)

Equating Al and B, to the long-period terms on the right-hand sides of (5.2.37), we have A1 = QG(4

- ti'),

Bl

=0

(5.2.40)

Consequently, (5.2.37) become

841 = ~ ( 2 whose solution is

a4

a,

=

+1

= -9(2

(5.2.41)

2)sin 2 6 - gci2 sin 4 4

+ 252 sin 44 - 2)cos 2 4 +

-$ti sin 24

cos 44

(5.2.42)

With (5.2.40) and (5.2.42), (5.2.38) and (5.2.39) become

8%

-

+ A , = short-period terms

34 842 + B2 -- -& -

a4

+ &az - &$a" + short-period terms

(5.2.43)

Equating A , knd B, to the long-period terms on the right-hand sides of (5.2.43),we obtain

5.3.

STRUBLE’S TECHNIQUE

171

Therefore, to second order u = a cos

where a =

r#

(5.2.45)

i - &5[sin 2 4 - &iz sin 461 + O(r2)

4 = 4 - &[(2 - 5 2 ) cos 2 6 - & 2 C O S 4r#J + O ( 2 ) d‘ = $ 4 4 - 52)

+

0(~3)

dt

d$ - = 1 - &€2 11 - 3 dt

(5.2.46)

2 + &2]

+ O(2)

(5.2.47)

This solution is in full agreement with that obtained in Section 5.7.4 by using Kamel’s algorithm. For canonical systems the transformation (5.2.31) and (5.2.32) can be effected in a more elegant manner by using von Zeipel’s procedure (Section 5.6) or Lie series and transforms (Section 5.7.5). The latter is a simple and efficient algorithm which is builtfrom the recursive application of a few elementary operations, thereby making it well suited for implementation on

a computer. For noncanonical systems an efficient recursive algorithm was formulated by using Lie transforms (Section 5.7). 5.3.

Struble’s Technique

Struble (1962) developed a technique for treating weakly nonlinear oscillatory systems such as those governed by ii

+

002u

(5.3.1)

= Q(u, zi, t )

He expressed the asymptotic solution of this equation for small E in the form u = a cos

- e) +n=l zEnun(t) N

+ o(P+~)

(5.3.2)

where a and 0 are slowly varying functions of time. If we set each u, = 0, (5.3.2) reduces to the form of the solution used by Krylov and Bogoliubov to obtain a first approximation to u (see Section 5.2.2). Rather than carry out the details for a general function f,we use the specific function corresponding to the Duffing equation. Thus we consider (5.3.3) ii Wo%J = -€US

+

172

VARIATION OF PARAMETERS AND AVERAGING

Substituting (5.3.2) into (5.3.3) gives

[

2a0,-

d2a + -- a dt dt2

de

dt

- 8) - 3e2U,a2 cos2

- e) + . . .

(5.3.4) If we consider terms through O(E)and equate the coefficientsof cos (mot - 0) and sin (mot - 8) on both sides, we obtain the following so-called variational equations =

-E43

cos3

d8 2ac0,dt

+ d2a - - a(;)d e dt2

= -tea3

da d28 dad0 +ay + 2-=0

-20,

dt

dt

dt dt

(5.3.5) (5.3.6)

This leaves the following so-called perturbational equation

-+ wozUl = -i f fcOs 3 3(m,t - e) d2u, dt2

To first order in

E,

(5.3.7)

(5.3.5) and (5.3.6) reduce to (5.3.8)

Hence a=ao,

8=--

Eatt 8WO

+ eo

(5.3.9)

where a, and 8, are constants. Then the solution of (5.3.7) to first order is obtained by considering 8 and a constants. The result is 1 a3 cos 3(Oot - 0) 3200 Hence the first-order solution becomes u1 = 2-

u = a COS

- e) +

where a and 8 are given by (5.3.9).

1 3201,

Ea3

cos 3(m,t - 8)

(5.3.10)

(5.3.1 1)

5.3.

STRUBLE'S TECHNIQUE

173

With u1 known the term -33E2U,a2cos2 (mot - e) =--

e~a5[COS (mot

128~2

- e) + 2 cos 3(Oot - e)

+ cos qw,t - e)l (5.3.12)

+

Moreover, we have to calculate the terms of O(E) in (d2ul/dt2) mo2u1; that is, we must include the term

9 do -a 3 -cos 3(o,t

- 0)

(5.3.13)

160, dt Now considering terms through O ( 2 ) leads to the variational equations 2~~0,+ - -- a dt dt2 do

d2a

=

-&a3

- -e2u5 1280;

d28 dad0 - + a -+ 2 - - = 0 dt dtZ dt dt and the perturbational equation -2w,

-+ .,,"u2 d2u2

dt2

da

=

3 -u5[2cos 3(Wot 12802

- e) + C

(5.3.14)

(5.3.15)

O qO,t ~

- e)]

9 do a 3 - cos ~ ( O O? e) (5.3.16) 16w0e dt The solution of (5.3.14) and (5.3.15) can be obtained by iteration starting with (5.3.9) to be

--

a=u,,

e=--

8WO

razt

+E2a;t + 8, + O(r3) 25602

(5.3.17)

where a, and 8, are constants. Substituting for dO/dt from (5.3.9) into (5.3.16) and solving the resulting equation, we obtain within an error of O ( E )the following

Therefore the solution to second order is

c2a6 +-10240,4 cos 5(wt - 0,) + 0(e3)

(5.3.19)

174 VARIATION OF PARAMETERS AND AVERAGING

where (5.3.20)

To carry out the solution to third order, we need to calculate the terms of O ( 8 ) in dzu,/dtZ,and the terms of O ( E )in d2u,/dt, then write the variational and perturbational equations. This constitutes a major limitation of this technique. The second limitation is the iteration solution of the variational equations. Systematic ways of handling such problems are the LindstedtPoincarC technique (Section 3.1. l), the Krylov-Bogoliubov-Mitropolski technique (Section 5.4), Lie series and transforms (Section 5.7), and the method of multiple scales discussed in Ch. 6.

5.4. The Krylov-Bogoliubov-Mitropolski Technique

In the course of their refinement of the first approximation for (5.2.11), which was discussed in Section 5.2.2, Krylov and Bogoliubov (1947)developed a technique for determining the solution to any approximation. This technique has been amplified and justified by Bogoliubov and Mitropolski (1961) and extended to nonstationary vibrations by Mitropolski (1965). They assumed an asymptotic expansion of the form u = a cos y

+ I e n u n ( a , y ) + O(eN+') N

(5.4.1)

n=l

where each u, is a periodic function of y with a period 27r, and a and y are assumed to vary with time according to

2 e"An(a) + O(eN+')

(5.4.2)

- = wo + 2 enyn(a)+ ~ ( c ' " + ' )

(5.4.3)

da dt

-=

dY dt

n=l

N

n=l

where the functions u,, A,, and w, are chosen such that (5.4.1) through (5.4.3)satisfy the differential equation (5.2.1 1). In order to uniquely determine A,, and y,, we require that no u, contains cos y. The derivatives are transformed according to

(5.4.4)

5.4.

THE KRYLOV-BOGOLIUBOV-MITROPOLSKI

TECHNIQUE

175

We next illustrate this technique by its application t o the Duffing, van der Pol, and Klein-Gordon equations. 5 4.1. THE DUFFING EQUATION We consider the nonlinear oscillator ii

+ wo2u =

(5.4.8)

-a43

which was treated previously in Sections 3.1.1, 5.2.2, and 5.3. Substituting (5.4.1) through (5.4.7) into (5.4.8) and equating coefficients of like powers of E through e2, we obtain 2

oo

a2u1 + oo2u1= 2w,yla

a2/’2

+ 2w0A1 sin y - a3 c0s3 y

cos y

- 3u,a 2 cos 2 y

(5.4.9)

a%, - 2w0A1a2ul - 2w0y1 7

aw

aa 8~

(5.4.10)

In order that u1 be periodic, the terms that produce secular terms on the right-hand side of (5.4.9) must vanish. Since c0s3 y = (3 cos y cos 3y)/4, this condition gives 3a2 A, = 0, yx= (5.4.11) 800 Then the solution of (5.4.9) is

+

u1

a3 320;

-cos 3 y

=

(5.4.12)

Substituting this first-order solution into (5.4.10) gives w:

-++ a2u

wo2u2=

15a4p ) a cos

aY

a5 +-1280,

y

+ 2w0A2sin y

(21 cos 3y

- 3 cos 5 y )

(5.4.13)

Elimination of secular terms yields A, = 0,

y2 =

’15a4

-2560:

(5.4.14)

176 VARIATION

OF PARAdETERS AND AVERAGING

Then the solution of (5.4.13) is uz =

-a5 (21 cos 3y - cos 5y) 10240,

(5.4.15)

Therefore, to second order, u is given by

where

+

dt

= 0 or

* = m o + - - 3ea2

dt

80,

[

y = o 0 l + - 3Ea2 -

802

(5.4.17)

a = a, = a constant

15e2a4 2560;

+0

152a4

256od1

~ ~ ) +

+

0(e3)

(5.4.18)

where yo is a constant. This solution agrees with (5.3.19) and (5.3.20) obtained using Struble's method. 5.4.2. THE VAN DER POL OSCILLATOR We discuss next the nonlinear oscillator ii

+ u = E(l - UZ)ti

(5.4.19)

treated in Sections 5.2.2 and 5.2.3. Substituting (5.4.1) through (5.4.7) into (5.4.19) and equating coefficientsof equal powers of E through 8,we obtain

aZu, - + u1 = 2yla cos y

ap2

+ 2A, sin y (5.4.20)

A, cos y - a y , sin y

+

+ u2u1sin 2y

(5.4.21)

5.4.

THE KRYLOV-BOGOLIUBOV-MITROPOLSKI

TECHNIQUE

177

Elimination of secular terms from the right-hand side of (5.4.20) gives

y1 = 0,

Hence

u1 =

Al = $a(l - id)

(5.4.22)

a3 -sin 3y

(5.4.23)

32

With this solution (5.4.21) becomes aZu -

aw2

dA 2ay2 - A , > da

+ (1 -

+ 2A2 sin y + a

3

128 5a5 +cos 5y 128

2 +

128

8, cos 3y

(5.4.24)

For there to be no secular terms in u2

Hence 5a5

Me=--

3072

cos 5 y -

(5.4.26)

1024

Therefore the solution to second order is given by

€2

u = a cos y - - sin 3y

32

2U3

-where

1024

da EU - = -(I dt 2

[$a2cos 5 y

- *a2),

+ (a2 + 8) cos 3y] + O(e3) 4

a2 =

1

(5

+

(5.4.27)

(5.4.28)

-l ) F

+

dY 2 A1 dA1 (5.4.29) - 1 $a2) -dt= [2a(do 256 where a, is a constant. Using (5.4.22) and (5.4.28), we can write (5.4.29) as

+'

2 - - ( EI dY-- 1 - Hence

16

dt

2

8a

E

y=t--t--lnu+-a 16 8 where y o is a constant.

A]

da dt

-1 4a 21-

7r 64

2

+yo

(5.4.30)

178

VARIATION OF PARAMETERS AND AVERAGING

5.4.3. THE KLEIN-GORDON EQUATION As a third example, we consider nonlinear waves governed by Utt

- c2u,,

+ A2u = € j - ( U , ut, us)

(5.4.3 1)

after Montgomery and Tidman (1964). If E = 0, (5.4.31) admits solutions of the form u = a cos ( k -~mot 4) (5.4.32)

+

where a and 4 are constants, and ko and oosatisfy the dispersion relationship

+ A2

wo2 = c2k02

(5.4.3 3)

For small but finite E , we seek an expansion of the form

+ €#,(a,y ) + - -

u = a cos y

(5.4.34)

*

where a is a slowly varying function of both time and position according to

- = cA,(a) + c2A2(a)+ . . .

(5.4.35)

+ e2&(a) + . . .

(5.4.36)

aa at

i3U - &,(a)

ax

and y is a new phase variable which coincides with the phase of (5.4.32) for r = O

aw - = -wo + at 'a -- k,

ax

Ec,(u)

+ c2C,(a) + . . .

(5.4.37)

+ E D , ( ~+) e2D,(a) + . . .

(5.4.38)

In this case also, no u, contains the fundamental cos y. Substituting (5.4.34) through (5.4.38) into (5.4.31), using (5.4.33), and equating the coefficients of E on both sides, we obtain 12

(z:2

+

1 ul) = -2(w0A1

+ c2koB,)sin y

- 2a(w0C1 + C2koDl) cos y

+ f [ a cos y, amo sin y , -ako

sin py]

(5.4.39)

We now Fourier analyze f in terms of y as f [ a cos y, am, sin y , -ako sin y ]

+ 2 [f,(a) sin n y + gJa) cos n y ] W

= go(a)

Tl=l

(5.4.40)

5.5.

AVERAGING USING CANONICAL VARIABLES

179

Eliminating secular terms, we have 2w0A, Sa(woC1

+ 2c%,B1 =fl(u)

+ c2koD,) = g1(a)

(5.4.41) (5.4.42)

Then the solution of (5.4.39) is

and D1from (5.4.35) through (5.4.38) into (5.4.41) Substituting for A,, B,, C1, and (5.4.42), we obtain

- + wo-aa aa

I

ax

at

!

at

h(a)

= E-

2w,

&(a) ’ + w o, ag - = f

ax

2aw0

(5.4.44) (5.4.45)

where 0; = dwo/dko,the group velocity, and /? =

Iffl = 0

a

B = E(Z

I+J

- k0x + mot

=

h,(s

- wit)

+ wit)* 4aw0wh + hz(s - w i t )

(5.4.46) (5.4.47) (5.4.48)

where h, and hz are determined from the initial or boundary conditions. Equations (5.4.44) and (5.4.45) can be solved easily if a and fl are functions of either time or position only. 5.5.

The Method of Averaging by Using Canonical Variables

Let us consider a conservative dynamic system governed by the following Lagrange equations (5.5.1)

where q = ( q l , q 2 ,. . . , q N } is the generalized coordinate vector, t is the independent variable, overdots denote differentiation with respect to I , L(Q, q, t ) = T - V is the Lagrangian, and T and V are the kinetic and potential energies. Let us define the generalized momentum vector p = {Pl?PZ, . P N ) by aL (5.5.2) Pi = aqa ’

’ 7

180 VARIATION

OF PARAMETERS AND AVERAGING

and the Hamiltonian H by H = pT4 - L

(5.5.3)

where pT denotes the transpose of p (if p is a column vector, pT is a row vector). Considering H as a function of p, q, and t only, we can write aH dH = 2- dqi i=l 6’4, Also, from (5.5.3)

aH + 2-aH dp, + - d t api at

(5.5.4)

i=l

The second and third terms on the right-hand side of (5.5.5) cancel according to (5.5.2). Moreover, since pi = aL/aq, from (5.5.1) and (5.5.2), we can write (5.5.5) as (5.5.6)

Comparing (5.5.4) and (5.5.6), we obtain the following canonical equations of Hamilton . = aH q. (5.5.7) ’ api aH (5.5.8) pi=-aqi aL aH - -- - (5.5.9) at at These equations replace Lagrange’s equations. Under a transformation from q and p to Q(q, p, t) and P ( q , p. f ) , (5.5.7) and (5.5.8) are transformed into

Q, t )

(5.5.10)

Pi = gi(P, Q, t ) If there exists a function K ( P , Q, t) such that

(5.5.1 1)

Qi

f i = -a K

=fi(P,

and gi =

api then (5.5.10) and (5.5.11) assume the form

K - a-

(5.5.12)

aQi

(5.5.13)

5.5.

AVERAGING USING CANONICAL VARIABLES

181

and Q and P are called canonical variables, and the transformation from q and p to Q and P is said to be a canonical transformation with respect to the function K . Canonical transformations can be generated using a so-called generating function S(P, q, t ) according to (e.g., Goldstein, 1965, Chapter 8; Meirovitch, 1970, Chapter 9)

p. =as,

as (5.5.14) ap, Once these equations are solved for q = q(P, Q, t ) and p = p(P, Q, t ) , K is related to H by '

K(P,

Q, t )

Qi=-

841

= fWP,

Q, t ) , W, Q, 0, t l

+ as

(5.5.15)

If a canonical transformation can be found such that K = 0, then P is a constant vector according to the second relationship in (5.5.13), hence S is a function of q and t only. Since p i = from the first relationship in (5.5.14), S must satisfy the following so-called Hamilton-Jacobi equation

If S is a complete solution of (5.5.16), then (5.5.14) furnishes a complete set of integrals of the equations (5.5.17) Complete solutions of (5.5.16) are not available for a general H . However, if H = H , fi where is small compared to Ho and a complete solution S,,(P,, . . . , P N ,q l , . . . ,.qN,t ) is available for

+

H,(flSP, . . . , (5.5.18) a41 then an approximate solution of (5.5.17) can be obtained by using the method of averaging in conjunction with the method of variation of parameters. Thus we use a generating function

s = S,(P,, . . . , PN, 41, . . . ,qhr, t )

(5.5.19)

where P is time-varying rather than being a constant. Hence P and Q are given by aK aK p . = -(5.5.20) ' aQi ap, 9

ea=-

182

VARIATION OF PARAMETERS AND AVERAGING

where

K = H o + A +as0 -- -A

(5.5.21)

at

as a result of (5.5.18). If the solution qo(P,Q, t) and po(P, Q , t) of (5.5.17) with H = Ho has a period T with respect to t , then an approximate solution to (5.5.17) is still given by qo and po but P and Q are given by (5.5.20) with K replaced by its average over T; that is, by

In (5.5.22), P and Q are held constant. We next illustrate this technique by its application to three specific examples. 5.5.1. THE DUFFING EQUATION We consider again the equation q

+

oo2q

+ €93 = 0

(5.5.23)

The Hamiltonian Corresponding to this equation is

+ wo8q2)+ &q4

H = $(p2

(5.5.24)

The Hamilton-Jacobi equation corresponding to the case e = 0 is (5.5.25)

This equation can be solved by separation of variables; that is, by letting

s = wl) +dt)

(5.5.26)

Equation (5.5.25) separates into or a = - u t

3=-u

and

(27+

wo2q2

= 2a

Hence

or

s

s = -ut +

s

(5.5.27)

1-I

J2a

- ~ 2 -a wzq2 dq

(5.5.28)

- w:q2

(5.5.29)

dq

with u the new momentum. Consequently, the new coordinate 2

2 -112

(2a-w0q)

1

d q = -t+-arcsin0 0

Ooq

Jz;;

is given by (5.5.30)

5.5.

AVERAGING USING CANONICAL VARIABLES

183

so that

J%.

q = -sin

wo(t

+ 8)

(5.5.31)

0 0

We could have written this solution for (5.5.23) by inspection if E = 0. However, the canonical variables CL and ,6 were obtained naturally by solving the Hamilton-Jacobi equation (5.5.25). Since E? = (1/4)q4 = (ca2/wO4) sin4 wo(t p), the variational equations (5.5.20) become aaf@ i ’ a = -8. = -a f i (5.5.32a)

+

aa

Now

so that (5.5.32b)

3ca and ,6 = 4t + Po 4w0 where Po is a constant. Therefore, to first approximation ct

= a constant

(5.5.33)

(5.5.34) in agreement with the expansions obtained in Sections 5.4.1 and 5.3 by using the Krylov-Bogoliubov-Mitropolski and Struble techniques if we identify &/w, by uo. 5.5.2. THE MATHIEU EQUATION As a second example, we consider for positive then

w.

If we let

q

+

( 0 2

+

E

cos 299 = 0

9=P

p

=

- (WZ + E cos 2t)q

(5.5.35) (5.5.36) (5.5.37)

from (5.5.35). These equations can be written in the form (5.5.38)

184

VARIATION OF PARAMETERS AND AVERAGING

where

H = )(p2

+ w2q2) + &42 cos 2t

As in Section 5.5.1, the solution of (5.5.38) for

4%

q = -cos w(t W

E

(5.5.39)

= 0 is

+ g)

Hence & = (1/2)q2 cos 2t = (ca/w2) cos 2t cos2 w ( t equations (5.5.20)become

(5.5.40)

+ g)and thevariationaI (5.5.41)

Since

0 if

(I?)=

i

wisawayfrom 1

-cos 2[(w - 1)t + wp]

L:2

if w - 1 = O(E)

(5 5 4 2 )

In the former case M and g are constants to first approximation. In the second case we inttoduce the new canonical variables a* and p* using the generating function so that

s* = cc*[(w - 1)t + wj?]

(5.5.43) (5.5.44)

p * = - -as* - (w - 1)t a%*

+0p

(5.5.45)

Consequently, a* and ,8* are canonical variables with respect to the Hamiltonian K = ( ~ ) + - =as* - c o s€a* 2/3*+(wl)a* (5.5.46) at 4w Hence (5.5.47) (5.5.48)

5.5.

AVERAGING USING CANONICAL VARIABLES

185

Elimination of t from (5.5.47) and (5.5.48) gives c

- d(cos 2p*) 40

du* -= a* ~

w

- 1+

Hence In a* = -ln

[

w

-1

-f_

40

cos 2p*

1

+ 420 cos 28* + a constant

(5.5.49)

Thus the motion is unstable (a* unbounded) if

That is, to first approximation o < l + f e

or w > l - $ f s

u) = 1 f ) c

or

The curves w2 = 1 f

&

(5.5.50) (5.5.51)

separate the stable from the unstable regions in the w - E plane. These curves are in agreement with those obtained in Section 3.1.2 by using the Lindstedt-PoincarC method and in Section 3.1.3 by using Whittaker's method.

5.5.3. A SWINGING SPRING Following Kane and Kahn (1968), we consider the nonlinear oscillations of a spring swinging in a vertical plane as shown in Figure 5-1. This problem

Figure 5-1

was introduced by Gorelik and Witt (1933) to illustrate internal resonances. The kinetic and potential energies of the mass m are

+ (I + v = tkz2 + mg(1 + z)(l - cos 0)

.T = t r n [ i 2

.)"2]

(5.5.52) (5.5.53)

186

VARIATION OF PARAMETERS AND AVERAGING

where x is the stretch in the spring beyond its equilibrium. Therefore L =T

-V

= $rn[i2

+ ( 1 + x)'%'] - mg(l + x)(1 - cos 0) - $kx2 (5.5.54)

Since

i3L p,=,,-

H =ips

- mi,

aL pe = - = rn(1

ae

+ x)%

(5.5.55)

+ ep, - L

For small x and 8 and x = O(e), H can be expanded into

3x2p; XPe2 ----2:rng164 + -+ 0 ( 0 5 ) (5.5.57)

rn 13 2m14 If we keep the quadratic terms in H, then a complete solution of the corresponding Hamilton-Jacobi equation can be obtained as follows. The Hamilton-Jacobi equation in this case is

(5.5.58) where S = S(x, 6 , t ) . T o solve this equation we let

(5.5.59)

hence

(5.5.60) (5.5.61) Therefore p z = as - = Jm(2u,

ax

- kx2)

(5.5.62) (5.5.63)

S

= -(ul

+ a2)t+

s

dm(2a,

s

- kx') dx +

Jrn~(2u, mg18') $0

(5.5.64)

5.5.

AVERAGING USING CANONICAL VARIABLES

187

Consequently

p1 -- - = -

m dx = m(2a1 - kx')

-2

+ &arcsin

B2=G=-r+/ as

m12 de = --t JmP(2a, - m g W )

Hence

x =

(5.5.65)

Z J ;

+ Jiarcsin

Odz

J"k sin Bl

e=J%

(5.5.66)

2%

(5.5.67)

sin B,

(5.5.68)

p , = J2ma1 cos B,

(5.5.69)

pe = 1J2mu, cos B,

(5.5.70)

-

where

,-

I-

To first approximation the variational equations correspond to

Ei

2

"P,2

= arngxe - -

rn 13

.-

-

azJo(l {sin B, lJ2k

+ % sin [(wl

+ 3 sin (Bl + 2Bz)

- 2wdt

+ olB1- 2w28,1)

Thus is fast varying unless w , - 2w2 = E , where the latter case the slowly varying part of A is

E

(5.5.71)

is a small quantity. In

To eliminate the explicit dependence of (fl)on t , we make a further canonical transformation from u1 and B1 to al* and PI* according to S*(al*, pl,

t) =

€al* -I 2w,

+3 al*B1 2w,

(5.5.73)

188 VARIATION OF PARAMETERS AND AVERAGING Thus

(5.5.74)

Therefore K = a constant because aK/at = 0. The variational equations become al* =

-

aK -- 202cu,Ju1* cos y

(5.5.76)

881* & , = - - - a K - -202cl%,Ju1* cos y

(5.5.77)

aB2

pl*

-

E = aK = -

aEl*

ZK

202

- ~ C U , E ~ - "sin' y -

@z = - = - c J a , *

aE2

where

(5.5.78)

sin y

(5.5.79)

(5.5.80)

Equations similar to (5.5.76) through (5.5.80) were obtained by Mettler (1959) and Sethna (1965) using the method of averaging. Adding (5.5.76) and (5.5.77) and integrating, we obtain a,*

+ u2 = E = a constant

(5.5.8 1)

Hence the motion is completely bounded. Elimination of y from (5.5.75) and (5.5.77) gives

(2J

= C'E;(E

where

- Ut) -

p2i2a2) - K]'

= C 2 [ F 2 ( ~2 ) G'(E~)]

F = fa2JE - tcZ,

G =~

~

(5.5.82)

2

~

2

-E K 2]

(5.5.83) )

The functions F(ctZ)and G(E,) are shown schematically in Figure 5-2. For

5.6.

VON ZEIPEL’S PROCEDURE

189

F. G

Figure 5-2

real motion, F2must be greater than or equal to G2.The points where G meets F correspond to the vanishing of both u, and al*.A curve such as GIwhich meets both branches of F or one branch of F at two different points corresponds to a periodic motion for the amplitudes and the phases, hence it corresponds to an aperiodic motion. The solution for the amplitudes and the phases can be written in terms of Jacobi elliptic functions. However, the points where G2 touches the branches of F represent periodic motions where the nonlinearity adjusts the frequencies o1and o2to produce perfect resonance. 5.6.

Von Zeipel’s Procedme

To determine a first approximation to Hamiltonian systems, the method of variation of parameters in conjunction with the method of averaging was used in Section 5.5. To determine higher approximations, von Zeipel (1916) devised a technique which is described and applied in this section to the first two examples discussed in the previous section. The essence of this technique is to expand the generating function S in powers of a small parameter E as

190 m

VARIATION OF PARAMETERS AND AVERAGING

2 ens,,,

and then determine S,, recursively by solving a chain of partial

n=O

differential equations. Let the system under consideration be described by the Hamiltonian m

WP, q, t ) = 2enHn(p, q, t ) , n=O

E

<< 1

(5.6.1)

with q the generalized coordinate vector and p the conjugate momenta vector. Let So = S,(P, q, t) be a complete solution of the Hamilton-Jacobi equation Ho - , q , t

12 +-=O

(5.6.2)

and let p = p(P, Q, t) and q = q(P, Q, t) be the solutions of

(5.6.3) If we assume P and Q to be time varying rather than being constants and use S = So(P,q, t) as a generating function to transform from the canonical system p and q to the canonical system P and Q, we transform the Hamiltonian from H into

m

= ~ E ~ H ~ Q, [ ~0, ( q(P, P , Q, r), t l n=1 m

=

I:enRn(P,Q, 0

(5.6.4)

n=l

Hence P and Q are governed by the variational equations (5.6.5) (5.6.6)

To determine an approximate solution to (5.6.5) through (5.6.6) to any order, we introduce a near-identity transformation fro& the canonical system P and Q to the canonical system P* and Q* using the generating function

2 Pi*Qi + 2 enSn(P*,Q, t ) N

m

i =1

n=l

S =

(5.6.7)

191

VON ZEIPEL'S PROCEDURE

5.6. so that m

+ 2 en aQ, (P*, Q, t )

Pi = Pi*

(5.6.8)

n=l

and

fi is transformed into Lo

K

= 2 en K,(

m

[

P*, Q, t ) = 2 ~ " 8(P* ,

n= 1

n=l

+ 2 ern "-) m=l

aQ

, Q, t ]

+2 n=l

E

n s n

at

(5.6.9) To determine K, we expand the right-hand side of (5.6.9) for small E and then equate the coefficients of equal powers of E on both sides to obtain K , = A,(P*, Q, t )

+ as -at'

(5.6.10)

K , = fi,(P*, Q, t )

fi as,aR, as, + 2- (P*, Q, t ) + dPi at

(5.6.11)

i=iaQi

K, = F,

as, +-

(5.6.12)

at

where F,t = F,,(P*,Q, t ) is a known function of HI, a,,.. . , fin and S1, S,, . . . , Sn-l. The functions S, are still undetermined and they can be chosen in any manner desired. Since F, contains, in general, a short-period term Fmsand a long-period term FnL,we choose

K,=

F>

and

at

-FF,"

(5.6.13)

Thus K,, contains long-period terms only, while S, contains short-period terms only. The functions S, can be obtained by solving successively the chain of partial differential equations in (5.6.13). The basic idea underlying this technique is the same as that underlying the generalized method of averaging of Section 5.2.3. Stern (1971~)showed that, for Hamiltonian systems, Kruskal's technique is equivalent to von Zeipel's technique. In both techniques we introduce near-identity transformations from the old dependent variables, which contain long- as well as short-period terms, into new dependent variables which contain longperiod terms. The basic difference between the two techniques is that the transformation in the von Zeipel method must be canonical, while the transformation in the generalized method of averaging need not be canonical and the system need not be described by a Hamiltonian. Morrison (1966b) showed that up to second order the von Zeipel procedure is a particular case of the generalized method of averaging. Giacaglia (1964) carried

192

VARIATION OF PARAMETERS AND AVERAGING

out the expansion to any order, while Barrar (1970) investigated the convergence of the von Zeipel procedure. Musen (1965) showed that the equations governing the expansion take a concise form if written in terms of Faa de Bruno operators (1857). We next determine second-order expansions for the first two problems discussed in the previous section. 5.6.1. THE DUFFING EQUATION As a first example, we consider the Duffing equation (5.5.23) which corresponds to the Hamiltonain (5.5.24). The solution of the problem corresponding to H, is given by (5.5.31) as found in Section 5.5.1. Hence

+ b)

(5.6.14)

B. = - a f i

(5.6.15)

2

A = 4E- M sin‘

wo(r

0 0

aA ag’

&=--

aM

To determine an approximate solution to (5.6.15), we introduce a nearidentity transformation from u and ,8 to u* and B* using the generating function (5.6.16) (5.6.17)

and the Hamiltonian

is transformed into

m

K = 2enKn(a*,@, t ) n=l

- -‘,sin4wO(r + b) 0 0

[ as,

a*

+ as + a@

as, +€--+8--+... at at Equating coefficients of equal powers of K1

=

Me2 0 0

[-:

E

E

2

as

l2

+ . .. ~

(5.6.18)

on both sides, we have

as - 4 cos 2w0(t + ,!I)+ Q cos 4w0(t + @)I + -at‘

(5.6.19) (5.6.20)

Equating Kl to the long-period term on the right-hand side of (5.6.19), we obtain 3M*2 K, =(5.6.21) 8wO4

5.6.

VON ZEIPEL'S PROCEDURE

193

Hence (5.6.19) becomes

as,

-

at

a*2 +7 [-4 oo

cos 2w0(t

+ ,8) +

cos 4w0(t

+ @)I = 0

(5.6.22)

+ p)]

(5.6.23)

The solution of (5.6.22) is

+

a*2

S1 = 5[sin 2w0(t ,8) - $ sin 4w,(t 40, With this value of S , , (5.6.20) becomes a*3 K , = *- [cos2wo(t 0 0

+ @)- /cos4wo(t + /?)I sin4 wo(t + 8) + asat

(5.6.24)

Equating K , to the long-period term on the right-hand side of this equation, we have 17a*3 (5.6.25) K z = - z Hence, to second order (5.6.26)

and u* -

aK

=0

or a* = aconstant

(5.6.27)

ag*

or (5.6.28)

where gois a constant. Having determined S , , we obtain a = a*

+ as, + . . . E -

= a* -4-

p*

ag a

*'

E d2WO

[cos 2w0(t

=g

+ as, aa* + . . .

=

+

+ @) - & cos 4w0(t + #?)I + O(E') (5.6.29)

E -

a* E 5-

200

[sin 2w0(t

+ p) - isin 4w0(t + p)] + O(e2)

(5.6.30)

194

VARIATION OF PARAMETERS AND AVERAGING

Solving (5.6.29) and (5.6.30) for u and @ in terms of a* and u = u*

+

€U*2

EU*

2w0

gives

[COS 2%(t

+ p*) -

[sin 2w0(l

+ p*) - & sin 4w0(t + b*)] + O(E')

2w0

B = p* - 5-

p*

COS 400(t

+ p*)] 4- O(E') (5.6.31) (5.6.32)

To compare this expansion with those obtained by other methods, we substitute for u and /?from (5.6.31) and (5.6.32) into (5.5.31), expand for small E , keeping a* and p* fixed, and obtain q =

J2a* -(I 0 0

-

8.5)

sin(wt

a* + pol - J2a* sin 3(wt + Po) + O ( 2 ) 16w: E -

(5.6.33)

where w = wo(l

If we let

+ I*)

(5.6.34)

then J2u* --

-a

&€

2-

a3

+ o(€2)

(5.6.35)

w0

(00

Hence

+

where (5.6.37)

This expansion is in agreement with those obtained in Section 5.3 by using Struble's method and in Section 5.4.1 by using the Krylov-BogoliubovMitropolski method. 5.6.2. THE MATHIEU EQUATION Next we determine a second-order expansion for Mathieu's equation (5.5.35) which corresponds to the Hamiltonian (5.5.39). The solution corresponding to Ifo can be written as (see Section 5.5.1)

JZ

q = -cos o ( t w

+ p),

p = -J20(

sin w(t

+ p)

(5.6.38a)

5.6. Hence a and

Transforming from a and (5.6.16), we obtain =

rK,

195

are canonical variables with respect to

E7 =

K

VON ZEIPEL'S PROCEDURE

= w2

/? to

CO2

o(2

a* and

+ g) cos 2t g* using

(5.6.38b)

the generating function

+ e2K2 + . . .

(5.6.39) since a is given by (5.6.17). Equating coefficients of equal powers of E in (5.6.39) yields K,

a* + - {cos 2t + * cos 2[(w + 1)t + w g ] at 2w2 + B cos 2[(0 - 1)t + w p ] )

as

= l-

(5.6.40) (5.6.41)

Two cases arise depending on whether w is near 1 (resonance) or w is away from 1. We consider both cases next starting with the latter case.

The Case of w Away from I . In this case all the terms on the right-hand side of (5.6.40) are fast varying. Hence Kl = 0, and

1

S , = - - sin 2t u* 4cu2

+ sin 2[(w + 1)t + w p ] + sin 2[(w2(w--1)t1)+ w g ] 2(w

I

+ 1)

(5.6.42)

Substituting for S , into (5.6.41), we obtain K,

as,

a*

at

4w3

= - - - cos'

w(t

+ ,8)

+ cos 4Kw2(w++1)t1)+ wg1 + cos 4[(w2(w--1)t1)+ Og] + w

cos 4w(t

+ g) + cos 4t + 2 cos 2w(t + p) w2 -

1

1

(5.6.43)

196 VARIATION OF If

w

PARAMETERS AND AVERAGING

is also away from 2

K,=-- 1

a*

16 UP(^

- 1)

(5.6.44)

because the rest of the terms on the right-hand side of (5.6.43) are fast varying and must be set equal to -as@ In this case (5.6.45) Hence

b * = - a- K - - -1

16 ~

aa*

Consequently

/3*=--

a* = a constant

(5.6.46)

- 1) + o(4

(5.6.47)

€2 ~

(

j

o

~

1 €9 + /3Q 16 o ~ ( o 1) ~

-

(5.6.48)

and q is given by (6.2.95) if we make the following replacement of variables u + q , q,-+w , a +o

--+

w/30 and w -+ w[l

-

--+J2u*/w

- ~ ~ / 1 6 w ~-( wl ) ]~

(5.6.49)

Thus q is bounded and the motion in this case is stable. If w is near 2, however, cos2[(w - 2)t wp] is slowly varying, and it has to be included in K,; otherwise S, will contain secular terms or a small divisor depending on whether w is exactly equal to 2 or not. Equating K2 to the long-period terms on the right-hand side of (5.6.43), we obtain

+

K2=---

w cos 2[(w - 2)t + wp] “ * ( + ] 16w3 -1 w-1 0’

(5.6.50)

To an error of O(E), /?can be replaced by /3* in (5.6.50). To analyze the motion in this case, we remove the explicit dependence of K on t by transusing ’ the generating function forming from a* and @*to a‘ and ,!I Thus

S’ = a‘[(w - 2)t

+

0/3*]

(5.6.5 1)

(5.6.52)

as- (w - 2)t

P I = % -

+

0/3*

(5.6.53)

5.6. VON ZEIPEL’S PROCEDURE 197 and

K‘ = K

+ (w - 2)a’ = (w - 2)a’ -

E’

Hence a.,= - - a- K’

aP‘

-

-€

2

“[* 160~

0 -

1

+

‘“-1

w2 - 1

a’ sin 28’ 8w2(w - 1)

(5.6.54)

(5.6.55)

(5.6.56)

The solution of (5.6.55) and (5.6.56) can be obtained as in Section 5.5.2 to be €2

160(0’ - 1)

- 16w2(w2 - 1) cos 28‘1

+ a constant

(5.6.57)

Therefore, for instability

2

E2

16w(w2 - 1)

16w2(w - 1)

or

5t2 w<2+-+0(€3),W>2-192

t2

192

+ o(4

(5.6.58)

Consequently, the transition curves that separate stable from unstable motions in the w2 - E plane and emanating from w = 2 are given by w2 = 4

52 + 48 - + O(t3)

and w2 = 4

E2 - 48 + o ( € ~ ) (5.6.59)

in agreement with those obtained in Section 3.1.2.

+

The Case of w Near I. In this case cos2[(w - 1)t w p ] is slowly varying, and therefore it should remain in K , ; otherwise S, is singular at w = 1 as evident from (5.6.42). Equating Kl to the long-period terms in (5.6.40), we have K , = -cos 2 [ ( 0 - 1)t U*

4w2

Hence

+ wp]

a* - -- - - {COS 2t + 4 cos 2[(w + 1)t + w P ] } as,

at

2w2

(5.6.60)

(5.6.61)

198

VARIATION OF PARAMETERS AND AVERAGING

The solution of (5.6.61) is sin 22

0 + 1)t + w,B] + sin 2 [ (2(w + 1) )

(5.6.62)

Substituting for S1into (5.6.41) gives K z = -as2 -

U*

+ +

+

+ 8)cos 2t

cos 2[(w l ) t wP3 cos2 w(t 403(w 1) Equating K2 to the longperiod terms in (5.6.63), we have at

Kz =

-

Therefore, to second order K = - cos 2[(0 4w2 €a*

U*

32w3(o

+ 1)

- 1)t + COB]-

€'a* 32w3(o

Moreover

- a*

-

(5.6.64)

+ 1) + O(P)

+ 1) cos 2[(0 + 1)t + wp] + O(e2)

€U*

440

(5.6.63)

(5.6.65)

(5.6.66)

p* = #? + E -as, + . . . aa*

=B

E

- ,(sin

+ sin 2I(w + 1)t + wB]) + O(@)

+

(5.6.67) 40 2(w 1) Solving (5.6.66) and (5.6.67) for u and B in terms of a* and ,8* gives

2t

Substituting for ,Iinto ? (5.6.65) gives K = -cos 2[(w 4m2 €U*

- 1)t + wP*] -

6%

*

32w3(w

+ 1)

5.6.

VON ZEIPEL’S PROCJZXJRE

199

The presence of the last term on the right-hand side of (5.6.70) exhibits a shortcoming of the von Zeipel procedure in its present form, in which the fast and slowly varying terms in (5.6.40) were determined using mixed variables (new momenta but old coordinates). Had we expressed (5.6.39) in terms of the new variables a* and /3*, this last term would have been absorbed into S,, which might contribute to the slowly varying part of K3. In fact, such a representation in mixed variables was recognized by Breakwell (see Schechter, 1968) to lead to invalid results for the motion of a particle near the triangular points in the restricted problem of three bodies (Breakwell and Pringle, 1966). Using this suggestion, Schechter obtained a valid expansion by expressing the Hamiltonian in terms of the new variables before averaging to determine the slowly varying part (long period). Musen (1 965) developed algorithms for effecting the transformation, to any order, of variables and arbitrary functions from the old to the new variables and vice versa. Lacina (1969a, b) and Stern (1970a, 1971a) obtained expressions for general near-identity canonical transformations from the old variables to the new variables. With these transformations, they modified the HamiltonJacobi equation. The resulting perturbation schemes may be related to other perturbation schemes which use canonical variables by the proper choice of certain expressions entering into these transformations. A technique of determining integrals of motion of a system governed by a Hamiltonian was developed by Whittaker (1916, 1937), Cherry (1927), Contopoulos (1963), McNamara and Whiteman (1967), and Coffey (1969). The technique is based on the fact that the equation

is satisfied by any integral of the canonical equations of motion

aH . Q=--,p=

aP

aH -aP

An effective and powerful technique of effecting the transformation of variables and arbitrary functions to new variables has been developed by Hori (1966, 1967) using the Lie series, and Deprit (1969) and Kame1 (1969, 1970) using the Lie transforms. This technique is described in Section 5.7. Had we expressed (5.6.39) in terms of the new variables, we would have found that EU*

K = -cos Z[(W - l)t 4w2

+ o#?*]- 3203(w + 1) + O(P) E2a*

(5.6.71)

We remove the explicit dependence of K on t by changing from u* and ,6*

200

VARIATION OF

PARAMETERS AND AVERAGING

to u1 and /?' by using the generating function

+ w/3*]

S' = d [ ( w - 1)t

Hence

(5.6.72)

a s = wu' a* = -

(5.6.73)

a@*

as - /3'=z(w

and

K' = K

as' = -COS +at 40 Eu~

28'

-

+ w/3*

1)t

e2a'

32w2(w

Therefore

+ 1) +

(5.6.74)

(W

- 1 ) ~ ' (5.6.75) (5.6.76)

As in Section 5.5.2, the solution of (5.6.76) and (5.6.77) is

In u' = -In[&

-

cos 28'

2 32w2(w

1

+ 1) + w - 1 + a constant

Hence the transition curves are given by €2

Consequently

or

4w w = 1 f *€ w2 = 1 f 4.

32w2(w

+ 1)

+ O(2) -# +

-

(5.6.78)

&€2

O(E3)

(5.6.79) (5.6.80)

These curves are in agreement with those obtained in Section 3.1.2 by using the Lindstedt-Poincark technique, and in Section 3.1.3 by using the Whittaker technique. 5.7. Averaging by Using the Lie Series and Transforms In analyzing the oscillations of a weakly nonlinear system, the method of variation of parameters is usually used to transform the equations governing these oscillations into the standard form * k = f(x; €) = 2

m=O

em ,f,(X)

m.

(5.7.1)

5.7.

AVERAGING BY USING LIE SERIES A N D TRANSFORMS

where f,(x) =

201

amf 1

a E m c=o

Here x and f a r e vectors with N components. The vector x may represent, for example, the amplitudes and the phases of the system, or the orbital parameters of the unperturbed two-body problem. If we denote the components of the vector f, by fmn, then a component xk of the vector x is said to be a rapidly rotating phase iff,, f 0. To analyze this standard system, we found it to be useful (see Section 5.2.3) to introduce a near-identity transformation

x = X(y;

€)

=y

+ .X,(y) + cZX,(y) + .

* *

(5.7.2a)

from x to y such that the system (5.7.1) is transformed into En

(5.7.2b) n! where g, contains long-period terms only. In Section 5.2.3, X, and g, were determined by substituting (5.7.2) into (5.7.1) and separating the short- and the long-period terms assuming that X, contains short-period terms only. Y = g ( y ; 4 = 2 --gg,(y) n=o

5.7.1. THE LIE SERIES AND TRANSFORMS In this section we define the transformation (5.7.2a) as the solution of the N differential equations dx

- = W(x; E ) , de

X(E

= 0) = y

(5.7.3)

The vector W is called the generating vector. It seems at first that we are turning in circles because we are proposing to simplify the original system of differential equations by solving a system of N differential equations. This is not the case, because we are interested in the solution of (5.7.1) for large 1 , whereas we need the solution of (5.7.3) for small E ; which is a significant simplification. Equation (5.7.3) generates the so-called Lie transforms (Kamel, 1970), which are invertible because they are close to the identity. If W does not depend on C , (5.7.3) generates the so-called Lie series. For a canonical system, Hori (1966, 1967) and Deprit (1969) took (5.7.4a)

202

VARIATION OF PARAMETERS AND AVERAGING

with q the system’s coordinates,, p the conjugate momenta, and t the time, and defined

[9

w = -s,

,

s = S(q,p, t ;

€)

(5.7.4b)

with S the generating function. Hori (1966) constructed a nonrecursive algorithm using the Lie series to m

determine the transform K = 2: (P/n!)K,(Q, P,t ) of a Hamiltonian m

n=o

H = 2 ( P / n ! ) H , ( q ,p, t). Deprit (1969) constructed another algorithm to ,=O

generate K recursively using the Lie transforms. Kamel (1969b), Campbell and Jefferys (1970), and Mersman (1970) showed the equivalence of Hori’s and Deprit’s theories, while Kamel (1969a) simplified Deprit’s algorithm. Hori (1970) showed that, to second order, Lie transforms are equivalent to von Zeipel’s technique. Shniad (1970) proved that the von Zeipel transformation is equivalent to the Deprit transformation, while Mersman (1971) established the equivalence of the Hori, Deprit and von Zeipel transformations. It should be mentioned that a perturbation theory based on the Lie series and transforms has several advantages over von Zeipel’s procedure. The generating function is not a mixed function of the old and new variables, the theory is canonically invariant, and it is possible to give a direct expansion of any function of the old variables in terms of the new variables. Kamel (1970) introduced the transformation (5.7.3) and constructed an algorithm to transform the standard system (5.7.1) into (5.7.2b). Moreover, he constructed algorithms (1) to transform any vector function from the old to the new variables and (2) to determine (5.7.2a) and its inverse. Henrard (1970) and Kamel (1971) investigated more deeply the mathematical and operational significance of these algorithms. We next construct these generalized algorithms and then specialize them to the canonical case in Section 5.7.5. 5.7.2. GENERALIZED ALGORITHMS Let the solution of (5.7.3) be x = X(y; e ) and let its inverse be y = Y(x; c) so that dx = X, dy and dy = Y,dx (5.7.5) where and

xY -- ax. * a Y;

(Jacobian matrix)

5.7.

AVERAGING BY USING LIE SFBIES AND TRANSFORMS 203

From (5.7.5) we have

dx = XYYxdx

so that

XyYx = Z

Hence

(identity matrix) (inverse of X,)

Y, = (XY)-l

(5.7.6)

The second relationship in (5.7.5) gives = Yxi

jl

which can be rewritten using (5.7.1) as

i = g(y; ). of

= Yxf

IXSXW)

(5.7.7)

We are interested in developing the right-hand side of (5.7.7) in powers E so that (5.7.8)

From (5.7.7)

=

[az

(Yxf)

Now

dx

a

+

a

(YXO]x-Xw€)

(5.7.9)

0"

(5.7.10) -(Yxf) = Yx- + -f ae af ax ae Since y = Y(x; E ) is the inverse of x = X(y; e) [the solution of dx/de =W(x; e) subject to x(c = 0) = y]

ae

ay + Yx-dx Q =0=so that

aE

dc

ay

-=

a€

de

=

ay + YXW ae

-YxW

Hence (5.7.10) can be rewritten as

a

- (Yxf)

a€

= Yx

ar a - - - (Y,w)f a€ ax

(5.7.11)

Using this expression and noting that dXldE = W , we rewrite (5.7.9) as d

9 = - t(Yxf)x=xcY;€)l dc dc

204

VARIATION OF PARAMETERS AND AVERAGING

which can be simplified to (5.7.12) where (5.7.13)

By recurrence, (5.7.12) gives d"g = mxDnflx=x(Y;€)

Since X(E = 0) = y

(5.7.14)

den

Y&=, = I

(identity matrix)

so that (5.7.15) To determine D"f, we assume that W can be expanded as

(5.7.16)

so that the transformation (5.7.3) can be generated, successively, to any m

order. I f f = 2 (rn/n!)fn, (5.7.13) becomes n-0

+

Letting n = k 1 in the first term, and n = k - m in the second term, we can rewrite this expression as

or

2

€k

Df=zk . +zk! OD

k=o

Yfk+l

m

k=O

k

%k-m

m=o z c m k [ z w m + l -

Df = k=O

where

2 - ff) k!

ax (5.7.17) (5.7.18a) (5.7.18b)

cmk =

k! (k - m)! m !

5.7.

AVERAGING BY USING LIE SERIES AND TRANSFORMS

By recurrence, (5.7.17) and (5.7.18) with fk = f:)

205

give (5.7.19)

where k

f y = f&l’

+ 2 CmkLm+lf:--,l’

(5.7.20)

m-0

Hence (5.7.21) The recurrence relationship (5.7.20) can be best visualized using the Lie triangle introduced by Deprit (1969); it is somewhat reminiscent of the Pascal triangle and shown in Figure 5-3.

Figure 5-3. Lie triangle. (ftn) = &(”), f, = fn(0))

(5.7.22)

In carrying out perturbation solutions, we often need to express a vector 2

F(x; E) = 2 - F,(x), n=O n !

F,(x) = !??!

acn

1

c=o

(5.7.23a)

206

VARIATION OF PARAMETERS AND AVERAGING

in terms of the new variable y in the form

= 2 - F'n'(y), n=o n! a,

En

F'n'(y) = d"F dan

1

X=Y.

(5.7.23b) C=O

where (5.7.23~) Using (5.7.16) and (5.7.23a), we express this [as we expressed (5.7.17)] in the form (5.7.24) where

By recurrence (5.7.24) and (5.7.25) give (5.7.27) where m-0

with FLo' = Fk. Hence

* En G(y; 6 ) = 2 - Ftn), n=O n !

F(n) = Fc)[

(5.7.29)

X-Y

Equations (5.7.27) through (5.7.29) have the same form as (5.7.19) through (5.7.21) except for the different operator 9, so that (5.7.28) can also be visualized using the Lie triangle. 5.7.3. SIMPLIFIED GENERAL ALGORITHMS To simpiify an algorithm such as (5.7.20), Kame1 (1969, 1970) wrote it first as

5.7.

AVERAGING BY USING LIE SERIES AND TRANspoRMs

207

He then eliminated successively the functions on the right-hand side to obtain eventually f r ) as a linear functional of ftn+,), f(n+*-l),. . . ,f(n).Thus we assume that

fp)=

f(n+K)

- 2k CjkGjf(n+k-i)

(5.7.3 1)

j=l

where Gj is a linear operator which is a functional of 4,JC.+~, . . . ,L1. Substituting (5.7.31) into (5.7.30) yields the following recursion relationship j-1

G, = Li For example

-m=l 2 C~21LmGi-m,

15j

In

G, = L, G, = L, - L1L1 G , = L3 - Ll(L.2 - L1L1) - 2L&1

(5.7.32)

(5.7.33)

For n = 0 and n = 1, (5.7.31) gives fCk)= f,

+ zcjvj,k-j k

i=l

k

f p = f'k+k+l'- IcjYj,k-j+l

(5.7.34) (5.7.35)

j=1

where

5-1

This is the simplified algorithm of Kamel in which fj., = 0 if fti) = 0 because the second index i, in the recursion relationship, is fixed. A more convenient form of this algorithm was obtained by Kamel by writing (5.7.35) as (5.7 * 37) However, from (5.7.18a)

j=l

so that (5.7.37) can be rewritten as

Since dx/dc = W from (5.7.3) (5.7.39)

208

VARIATION OF PARAMETERS AND AVERAGING

where

I

x(n+l)(y) = -

dnW den

Hence from (5.7.16)and (5.7.34)

for n 2

E=O.X=Y

k-1

= wk -k zC;-'Xj,p+j,

X"'

k 2

(5.7.40)

j=1

where = gjX(i'

XjSi

- z.C;21=gmXj--m.a j-1

(5.7.41)

m-1

To determine the inverse transform y =x

* +z - y'"'(x) n! En

(5.7.42)

n=l

we eliminate x - y from (5.7.39)and (5.7.42)to obtain

However

2 - y'"'(x) O0

En

n=l

n!

u=

u =2

n=l

=

En -2 x'"'(y) O0

n=l

n!

En co E n un(x) = 2 r u'"'(y)

n!

n-1

n.

(5.7.43a)

(5.7.43b)

(5.7.43c)

(5.7.44) where

is defined by (5.7.41).

5.7.4. PROCEDURE OUTLINE Consider a system of differential equations written in the standard form n=o

n.

(5.7.45)

The essence of the algorithms of the previous section is to introduce a transformation from x to y so that (5.7.45)becomes

i = 2: 7 gn(Y) n. O0

En

(5.7.46)

n=O

where g, does not contain short-period terms. To do this we construct a

5.7.

AVERAGING BY USING LIE SERIES A N D TRANSFORMS

209

near-identity transformation

x =y

+2 x'"'(y) n! En

n=l

(5.7.47)

Under such a transformation a vector F(x; E) = 2 - Fn(x) n=O n ! En

becomes

2

F(x; E ) = 2 - Ftn'(y) n=o n!

(5.7.48)

(5.7.49)

The algorithms described in the previous section to generate these transformations can be implemented on a computer because they can be effected by the recursive application of elementary operations. We describe below the procedure to second order. It is initiated by putting (5.7.50)

Then we begin the first-order expansion by writing the linear partial differential equation (5.7.51) gl(Y) = fdY) + LlfO We choose g, to equal the long-period terms in f,, solve the resulting equation for W1, and compute x") = w,

Fl,o= 6P1F'" F'" = F1

+ Fl,o

(5.7.52)

To prepare for the second-order expansion, we compute g1.1 = L1g1 Then we set up the differential equation

+

+

+

(5.7.53) g2 = f, Llfl g1.1 b f o and choose g, to be equal to the long-period part of the right-hand side. This completes the expansion to second order. We illustrate this procedure by applying it to van der Pol's equation q

whose solution for

E

=0

+ q = E(1 - q")4

(5.7.54)

can be written as

q=acos+,

+ = t + p

(5.7.5 5)

210

VARIATION OF PARAMETERS AND AVERAGING

By the method of variation of parameters, (5.7.54) can be replaced by (see Section 5.2.3) ci = *€[U(l - tu2) - U C , ta3C,] (5.7.56) where

+ = 1 + &[(I

C , = cos n+

+

(5.7.57)

- &zz)S, - ta2S4]

and S, = sin n+

LI

Equations (5.7.56) and (5.7.57) have the same form as (5.7.1) with

x= fo=

[

&z(l - @)

f1=

C]

(5.7.58)

(5.7.59)

- *uc,

+

+(1 - ifu2)Sz- &l2S4

We transform now from x = we obtain

>1

[i] [-I].

f, = 0 for n

to y =

go=

1

Qu3C4

(5.7.60) (5.7.61)

From (5.7.50) and (5.7.59)

c3

(5.7.62)

From (5.7.18b) and (5.7.59), we find that awn

L,fo = - -

(5.7.63)

awl g, = fl - -

(5.7.64)

ad*

Hence (5.7.51) becomes

a+*

Choosing W, to remove the short-period terms in fi gives &2*(1

g,=[

Solving the resultant equation gives

w1=

[

--)u*St*

- &*a) 0

+

-t(l - *U*2)C2*

where S,* = sin n+* and C,* = cos n+*.

1

&U*3S4*

+

(5.7.65)

I

&2*2c4*

(5.7.66)

AVERAGING BY USING LIE SERIES A N D TRANSFORMS

5.7.

21 1

With (5.7.61) and (5.7.63), (5.7.53) becomes (5.7.67)

Choosing Wz to remove the short-period terms, we obtain gz = (Llfl)

+ (L1g1)

(5.7.68)

Since g, consists of long-period terms only, and W1consists of short-period terms only, (Llgl) = 0. Hence (5.7.69)

Therefore

in agreement with the expansion obtained in Section 5.2.3 using the generalized method of averaging. To compare the expansion obtained in this section with that obtained in Section 5.4.2 by using the Krylov-Bogoliubov-Mitropolski technique, we need to express (5.7.55) in terms of the new variables. Here En

27 n=O n.

4 =

with From (5.7.50)

Fo = acos(6,

(5.7.71)

Fn = 0 for n 2 1

P o )= a* cos $*

(5.7.72) (5.7.73)

Then (5.7.52) and (5.7.26) give

Hence

Fl,o= 2Z1(a* cos $*) = [cos +*, --a* = -$a*(l - $ u * ~ sin ) #* -

q = a* cos d* - $ea*[(l

sin +*]W, sin 3+*

- &a*z)sin $* + +a*2sin 34*] + O(e2)

(5.7.74)

which can be rewritten as q = a* cos p

where dy=l-rz dt

*3 - ra sin 3 y

32

(I8 - - 8 + ?*) 256 a*2

+ O(2) + O(P)

(5.7.75a)

(5.7.75b)

212 VARIATION OF

PARAMETERS AND AVERAGING

in agreement with the expansion obtained in Section 5.4.2 using the KrylovBogoliubov-Mitropolski technique. 5.7.5. ALGORITHMS FOR CANONICAL SYSTEMS Hori (1966, 1967) and Deprit (1969) used the Lie series and transforms, respectively, to transform a Hamiltonian

into a new Hamiltonian K(P, Q, t ; c) = 2 - U P , Q, 2) n=o n! En

If we let

(5.7.77)

(5.7.78)

X =

(5.7.79) then g can be generated from a Hamiltonian K as g=

[

-Kq

(5.7.80)

In this case the algorithm of Section 5.7.3 reduces to the scalar form (Kame), 1969a) Ko = Ho(Q, P, 2 )

where

af as, --af as, L;f = -aQ ap apaa

(5.7.82a) (5.7.82 b)

and

5.7.

AVERAGING BY USING LIE SERIES AND TRANSFORMS

213

Under the above transformation the old variables are given in terms of the new variables as (5.7.83) where (5.7.84a) and (5.7.84b)

3-1

p3.. 1. = L'.3"'

- Z: C21LAPj-m,i m=l

To third order, the above algorithms become

K O = Ho

(5.7.85) (5.7.86)

K2 = H, K3 = H3

+ L;H1 + L;Ki

-9s2

+ L;H2 + 2LLH1 + 2LiK2 + L;Kl

(5.7.87)

9 t

- Liz&

-9s3 (5.7.88)

9 t

214 VARIATION OF PARAMETERS AND AVERAGING We next illustrate this procedure by its application to the swinging spring represented by the Hamiltonian (5.5.57). Using the solution (5.5.67) through (5.5.70), we transform this Hamiltonian into H = H,

where

+ 3H2+

(5.7.91)

*..

I-

H, = %J3 [sin2B, - 2 cos2B,] sin B1 l k

(5.7.92)

1 a: 12a,a, . 2 H , = --s i n 4 ~ , -sin B, cos2B, 3 mgl k12

(5.7.93)

+

+

and Bi = w i ( t pi). We transform a, p, and H into a*, @*,and K = KO Kl &K2 . . . using the algorithm defined by (5.7.85) through (5.7.87). Since H , = 0, KO = 0 by (5.7.85), and 9 S n / 9 r = as,/& by (5.7.82b). With (5.7.92), (5.7.86) becomes

+ +

-

K 1 -- -

",/$ 1

+

(sin B1*

+ + sin (Bl* + 2B2*)

sin [(ol- 20,)t

+ wlpl* - 2w2p2*])-

3

+

(5.7.94)

All the terms in K, have short periods unless w1 w 20,. In the latter case sin [(ol- 2022 olpl* - 20,/?,*] has a long period (slowly varying). Choosing S, to eliminate the short-period terms results in

+

-*

+

1

3 cos(Bl* 2B2*) (5.7.96) awl 202) If we choose Szto eliminate the short-period terms in (5.7.87), we obtain

+

+

K , = (H,) (GHd The averaged values in K , are given by

+ (&Kl)

(5.7.97)

(5.7.98)

*=

3al*a2* k12

a2 +-w-(H,)= - -

8mgl

a$ 2k12

30(1*Ue*

+

k2"

(5.7.99)

AVERAGING BY USING LIE SERIES AND TRANSFORMS

5.7. (LiH,) =

215

&H asl\ + /aH, as,\ ''\ag, a d \Z ZJ -

= -[I+

4(W1

+ 20.4

]

*= - 9a1*a2*

a2

4kZ2

o2

+ 2w2

4k12 o1

(5.7.100)

In (5.7.99) and (5.7.100) use has been made of the fact that o1M 2w2 (i.e., kl M 4mg from the definitions of o1and 02). Hence K2 -

and

33af 32k12

+-39al*a2* 16k12

(5.7.101)

K = Ko+K1+hK2+..* -I

2

39a1*a2* ++ ... 64k12 32kl'

33a: --

(5.7.102)

To remove the explicit dependence of K on t , we transform from a * and u' and B' using the generating function

#?*to

S' = a;

[(w, o12w2)t+ A*] + adSz*

(5.7.103)

so that (5.7.104) (5.7.105) (5.7.106) (5.7.107)

and

asi

K'=K+-=-at

3a;Ja; - sin (wl/li - 2w2,9i) 21J2k a 33a; 3944

+64k12, + 32kl

--

o1- 202 , 0 1

a1 (5.7.108)

216

VARIATION OF PARAMETERS AND AVERAGING

(5.7.109) U'2 --

-

1J2k

-sin y

pz=-.I

where

3JU; - sin y 21,/2k

cos y

394 ++ 32k1'

w1-

334 394 -+ -, 32k12

y = 0lt% - 2w,B;,

(5.7.110)

32kl

2w,

0 1

(5.7.11 1) (5.7.112) (5.7.113)

5.8. Averaging by Using Lagrangians

Instead of using canonical variables, which require the use of Hamiltonians, Sturrock (1958, 1962) developed a technique that does not require canonical variables. It consists of averaging the Lagrangian and then writing down the corresponding Euler-Lagrange equations. Whitham (1965a, 1967a, b, 1970) developed a similar technique for waves in which the frequency and wave number-as well as the amplitude are slowly varying functions of space and time. Bisshop (1 969) supplied a more rigorous justification of this technique. Although this technique is not as elegant as those using canonical variables, it has the advantage of being directly applicable to partial as well as ordinary differential equations. Kawakami (1970) and Kawakami and Yagishita (1971) used canonical variables in conjunction with Hamiltonians to solve the nonlinear Vlasov equation. This technique was applied to a variety of problems of wave propagation in fluids and plasmas. Lighthill (1965, 1967) applied Whitham's theory to moderate waves in deep water where pseudofrequencies are absent, while Karpman and Krushkal' (1 969) used Whitham's theory to study the decay of a plane wave into separate wave packets. Howe (1967) studied open-channel steady flow past a solid surface of finite wave group shape. Bretherton (1968) treated linear wave propagation in slowly varying wave guides, while Bretherton and Garrett (1968) investigated slowly varying waves in inhomogeneous media. Garrett (1968), Drazin (1969), and Rarity (1969) analyzed nonlinear internal gravity waves; the effects of shear and slight atmospheric stratification were determined by Garrett and Drazin, respectively. Simmons (1969) studied the interaction of capillary and gravity waves, while Grimshaw

5.8.

AVERAGING BY USING LAGRANGIANS

217

(1970) discussed solitary waves in water of variable depth. Crapper (1970) investigated the generation of capillary waves by gravity waves. Dougherty (1970), Galloway and Crawford (1970), and Galloway and Kim (1971) treated nonlinear waves in plasmas. Dewar (1970) investigated the interaction between hydromagnetic waves and an inhomogeneous medium, while Tang and Sivasubramanian (1 971) studied the nonlinear instability of modulated waves in a magnetoplasma. Lowell (1970) analyzed wave propagation in lattices with an anharmonic potential. We describe this technique and its application to three examples. 5.8.1. A MODEL FOR DISPERSIVE WAVES As a first example, we analyze slowly varying wave train solutions for Bretherton’s (1 964) model equation

ht

+

4222,

+ 422+ 4 =

€43

(5.8.1)

with the nonlinear term e d 3 rather than €4” If we neglect the nonlinear term ed3, (5.8.1) admits the uniform traveling wave solution

+ = a cos 0,

where

(I)

0 = k x - cot

(5.8.2)

and k satisfy the dispersion relationship 0)’

= k4 - k2

+1

(5.8.3)

To determine a slowly varying wave train solution by using the variational approach, we first write the Lagrangian corresponding to (5.8.1); that is L =

b4t2 - 44:2

+ 4402 - BV +

(5.8.4)

It can be easily verified that (5.8.1) is the Euler-Lagrange equation corresponding to this Lagrangian. We assume an expansion of the form

4 = a c o s e + E p A , c o s n+e cqr2) 4,

n=2

where

k = 0=,

=

-et

(5.8.5) (5.8.6)

and a , o,k , and A i are slowly varying functions of x and t . If 0 is twice continuously differentiable, (5.8.6) gives the compatability relationship kt

+

W,

=0

(5.8.7)

Since secular terms appear first at O(€) in the straightforward expansion,

218

VARIATION OF PARAMETERS AND AVERAGING

we assume that a,, a,, w,, w , , k,, and k , are O(E).Thus

4,

+ a, cos 8 + cw 2 nAn sin n8 + O ( 2 ) sin 8 + a, cos 8 - E k 2 nAn sin n8 + O(e2) 03

= aw sin 8

+, = -ak

n=2

m

n=2

$ ,, = -ak2 cos 8 - 2a,k sin 8

..

- c k 2 z n2Ancos n8 + 0 ( e 2 ) m

n=2

Substituting these expressions into (5.8.4), we find that the resulting Lagrangian depends implicitly on z and r through 8 , a, o,k , and A j . Its variation with z and t through 8 is much faster than that through the other parameters, because when 6 varies over [0,211], the period of the 8 terms in L, the other parameters hardly change. As in the other versions of the method of averaging, we average L over 8 from 0 to 2rr, keeping a, o,k, and A t constant. To do this we first average each term in L separately. Thus

$ ,: = 4a2k4

-

Hence

$4

= $a4

+ O(c2)

+ O(c)

-

9 = L = f ( 0 2- k4 + k2 - l)a2 + 2 a4 + O(2)

(5.8.8) 32 The averaged Lagrangian 2 is an explicit function of a and an implicit function of 8 through w and k. Now we write the Euler-Lagrange equations with respect to the variables a and 8. The Euler-Lagrange equation with respect to a is a9’EpIaa = 0; it gives the dispersion relationship w2 = k4 - k2

+ 1 + $€a2+ O(r2)

(5.8.9)

Note that this dispersion relationship can be obtained by using the principle of harmonic balance; that is, we let = a cos 8 in (5.8.1) and equate the coefficients of cos 8 on both sides. Since w = -8, and k = 8,, the EulerLagrange equation with respect to 8 is

+

or

”(”)ae, at

+

d(?z)= i&E= 0 ax ae, ae

(5.8.10) (5.8.11)

5.8.

AVERAGING BY USING LAGRANGIANS

219

However,

Hence

aw

= $ma2 + O(E'),

a

- (OM') at

ak

= -$(2k3

a [(2k3 - k)a'] +ax

- k)a' + O(e2) =0

(5.8.12)

To simplify (5.8.12), we differentiate (5.8.9) with respect to k to obtain WW'

= 2k3 - k

+ O(e2)

where a)' = dwjdk is the group velocity. Hence we rewrite (5.8.12) as

or

(5.8.13) Since w = ~ ( k ) , = o'kt, and the second term in (5.8.13) vanishes according to (5.8.7). Hence (5.8.13) simplifies into (5.8.14)

Moreover, since o = w(k), (5.8.7) can be rewritten as

ak + 0' ak =0 at

ax

(5.8.15)

Therefore the spatial and temporal variations of the amplitude a, frequency a), and wave number k are given by (5.8.9), (5.8.14), and (5.8.15). 5.8.2. A MODEL FOR WAVE-WAVE INTERACTION Had we carried out the expansion of the previous section to second order, we would have obtained (b = a cos 0

€a3 cos 30 + 0 ( e 2 ) + 32(9k4 - 1)

(5.8.16)

Although valid for a wide range of values Df k , this expansion breaks down near k2 = 1/3. This case is referred to as the third-harmonic resonance case in which both cos 0 and cos 38 satisfy the same dispersion relationship; that is, both the fundamental and its third harmonic have the same phase speed wlk.

220

VARIATION OF PARAMETERS AND AVERAGING

To determine an expansion valid near k2 = 113 for 5.8.1), we assume the expansion to be

4 = a, COS 6, where

+ a3 cos + 2 A,

8, = k,x - o,t

with

a,

63

E

n+1.3

+ pn,

cos (6, f vn)

on2= kn4 - kn2

+ O(E’)

(5.8.17)

+1

(5.8.18)

k , w 3 k , and w 3 m 3 0 1

(5.8.1 9)

Note that the first-order term contains the fundamental cos 8 , and its third harmonic 8,. Since we are interested in the case k12 M 113, we take w i and ki to be constants, whereas we take p i, v i , ai, and A j to be slowly varying functions of x and t. We next substitute this expansion into (5.8.4) and average the resulting Lagrangian over ei, keeping pi, vi, ui, and Ad constants. In performing the averaging we note that, although Oi are fast varying, 6 3 - 38, is slowly varying. Thus

#zz = 1 2 (k:

-

-

2 k1.3

+ 4ki3piZ)a: + O(E’)

With these expressions 9becomes

where 6

= 63

- 38,

34 + 32 - (a: + 4al2a; + a:) + u13u3cos 6 8

(5.8.20)

+ p3 - 38,

(5.8.21)

= (k,

- 3kl)x - (w3 - 3w,)t

In arriving at (5.8.20) we used the dispersion relationship (5.8.18) and the definition of the group velocity o:= (2ki3 - ki)/wj. With wi and ki constants, us and pi are governed by the Euler-Lagrange equations

a- =9o aai

(5.8.22)

5.8.

AVERAGING BY USING LAGRANGIANS

221

and (5.8.23) Substituting for 2 from (5.8.20) into these equations and using (5.8.21), we obtain

B3t

+

~ 4 8 3 ,=

a01

-

at

8w3

[6ai2

+ 3a3' + a13ag1cos 61

3€ + w118% - = - -a,-a-, sin 6 9

ax

801

aa3 + w3, aa, = E 3 a , sin 6

at

8w3

ax

(5.8.25) (5.8.26) (5.8.27)

These equations are in agreement with those obtained in Section 6.2.9 by using the method of multiple scales. 5.8.3. THE NONLINEAR KLEIN-GORDON EQUATION As a last example, we consider after Whitham (1965a) the nonlinear Klein-Gordon equation (5.8.28) utt - u,, V'(u) = 0 where V ( u ) is any nonlinear potential function which yields oscillatory solutions. Scott (1970) treated the special case V(u) = -cos u , which represents the propagation of a magnetic flux on a long Josephson tunnel junction. The Lagrangian corresponding to this equation is

+

L = $Ut2 - $11;

- V(u)

(5.8.29)

For a uniform wave train solution u ( @ , (5.8.28) becomes where

(0'

-

+ V ' ( U )= 0

k2)U,,

and

w = -Ot

k = 0,

(5.8.30) (5.8.31)

A first integral of (5.8.30) is

4(0' - k 2 ) U ;

+ V ( U )= E

(5.8.32)

Integrating this equation gives w2 -

k2

du

(5.8.33)

222 VARIATION OF PARAMETERS AND AVERAGING We assume that u is periodic with a period that can be normalized to unity so that

(5.8.34) To determine approximate equations for slowly varying E, o,and k, we substitute u(O) for u in (5.8.29) and average the resulting Lagrangian over the interval [0, 11. Thus L = $(w2 - k2)U; - V(U)= (0’ - k2)u; - E

as a result of (5.8.32), hence L[u(B)]dO = (w2 - k2)J

1

1482

d8 -

(5.8.35)

EP (5.8.36)

The variation of 9 with respect to E gives the dispersion relation (5.8.34). Since w = -Ot and k = O,, the Euler-Lagrange equation corresponding to the variable 8 is

(5.8.37) Substituting for 2’ into (5.8.37), we obtain -a( w W ) at

where

+a (kW) = 0 ax

(5.8.38) (5.8.39)

The problem formulation is completed by augmenting (5.8.34) and (5.8.38) by the cornpatability relation ak - + - =aoo at

ax

(5.8.40)

EXERCISES

223

Exercises 5.1. Use the Stuart-Watson-Eckhaus solutions to (a) ii

technique to determine approximate

+ Iu = €12

u(0) = u(n) = 0 (b)

~

t

+

-t ~~,, u =E U

u ( x , 0) = a cos 2 ,

~

Ut(Z,

0) = 0

5.2. Use Struble’s technique to determine uniform second-order expansions for (Struble, 1962)

+ u = r ( l - u2)li (b) ii + ( h + cos 2r)u = 0 (a) ii

E

5.3. Use the Krylov-Bogoliubov technique to determine approximate solutions for (a) ii + W o % = - E l i (li( (b) ii

(c) ii

+ wo2u = r ( l - u2)u + cu3 + (8 + c cos 2t)u = 0

5.4. Consider Mathieu’s equation

ii

+ (6 + rcos2r)u = 0

Determine uniform second-order expansions using (a) the Krylov-BogoliubovMitropolski technique, (b) the generalized method of averaging, and (c) the Lie transforms. 5.5. Consider the equation

+ fao2q =

tq3

+ K cOS

or

(a) Show that it corresponds to the Hamiltonian

H

=

&(pz + wo2y2) - t c @ - K y cos wf

(b:) Determine a first-order expansion when K = O(1) and w is away from 3 0 0 , 00, (c) K = 0 ( 1 ) and w is near 3w0. (d) K = 0(1) and w is near 4 3 . (e) K = O ( E )and w is near w0. 5.6. Consider the equation

7 + w02q = ~ ( 1- q2)q + K cos wf

Use the Krylov-Bogoliubov method to determine first-order expansions for the cases enumerated in Exercise 5.5.

224 VARIATION

OF PARAMETERS AND AVERAGING

5.7. Use the generalized method of averaging, the Krylov-Bogoliubov-Mitre polski method, and Kamel's method to determine second-order expansions for ii

+ u = E(l

- u2)U

+

€113

Compare the results of the three methods. Which of these techniques would you recommend for such problems? 5.8. Consider the problem

+ 2pli + v2u = - € f ( u , li)

ii

AS E -0

u = uoe-pt cos (mot

(a) For

by letting u

c

+ (bo),

wo =

.\/v2

- pz

# 0 determine a uniform expansion following Mendelson (1970)

= u(a, v),

where

u=u,+ru,+". du - -- - p u eA,(a) dt

+

+ ...

9 = wo + '&(a) + . . . dt (b) Determine an expansion using the Krylov-Bogoliubov-Mitropolski technique. (c) Which of these expansions is more accurate? 5.9. Consider the problem 12

+ 6~ + Eblurn + EbzUn--l

- cb.&-' cos At

=0

where 6, E , bi, and L are constants and m is an odd while n is an even integer with m > n. (a) For small E find a solution of the form u = a(t)cos 0 ,

0 = wt

- C(t),

o =

A/n

and use the method of averaging to determine equations for a and (b (Tso and Caughey, 1965). (b) Determine a Hamiltonian corresponding to the above equation and then determine a first-order expansion using canonical variables for 6 near 02. (c) Compare the results of the two techniques. 5.10. The problem of a spherical pendulum (i.e., a particle moving under the action of gravity on the surface of a smooth fixed sphere) is represented by the Hamiltonian (Johansen and Kane, 1969)

where qi andpi are the coordinates and momenta of the particle, m is its mass, g is the gravitational acceleration, and I is the radius of the sphere.

EXERCISES

22 5

(a) Determine a first-order solution for small amplitudes using the method of averaging with canonical variables. (b) Determine a second-order expansion using Lie transforms. (c) Determine a second-order expansion using the generalized method of averaging. (d) Compare the three resulting expansions. 5.11. Consider the problem of a swinging spring with damping j.

+ f+ + ik

--3'

m

+g(i

- cos e) - ( I

+ +I2 = o

2 xe = o e + s,e + g sine + I+x l + x :

If w12 = k / m and w2' = g / l , determine second-order uniform expansions using the generalized method of averaging, and the Lie transforms for the cases (a) w1 w 20, and (b) (0, * 3w,. 5.12. Consider (3.1.63) through (3.1.65). (a) Show that the corresponding Hamiltonian is H =

+ p,') + 7 2 ~ 1- 7 1 ~ + 2 &Q12 + 72')

- +(h1722

+ hzy12)(1 + e cosf)-'

(b) Use the method of averaging with canonical variables to determine a first-order expansion near the transition curves starting at (p,,, e) where p,,= (1 - 2\'2/3)/2. (c) Use the Lie transforms to determine a second-order expansion near these transition curves.

5.13. Consider the motion of a particle described by the Hamiltonian

+

+

+

(6 + &)7? 413 + 29i7z2 H = +(pi2 + pz2) 4(42p1 - qlpz) + where 6 is a constant parameter. (a) Show that the circular frequencies of the linearized problem are 1 and 2 when 6 = 1 , (b) use the method of averaging with canonical variables to determine a first-order expansion for small amplitudes when 6 * 1 , (c) determine a first-order expansion using the method of averaging, and (d) which of these techniques would you recommend for such problems? 5.14. Consider the problem (Sethna, 1965) i. !J

+ = e(3b12' + 2bZzy + b&) - €612 + K ~ C OAtS + w 2 y = e(b,x2 + 26,xy + 3b4y2) - eS,$ + K , sin I t W

~

~

T

If internal resonance means w1 w 202 or w, = 2w1, resonant excitation means I, m w1 or w 2 , soft excitation is denoted by K i = EkE with ki = O (l), and hard excitation is denoted by K i = 0(1), use the method of averaging to determine firstorder expansions for the following cases:

(a) Hard nonresonant excitation in the absence of internal resonance. (b) hard nonresonant excitation in the presence of internal resonance. (c) soft resonant excitation in the absence of internal resonance. (d) soft resonant excitation in the presence of internal resonance.

226

VARIATION OF PARAMETERS AND AVERAGING

5.15. Traveling waves in a cold plasma are governed by

a 2 +(pu) = 0 at ax au

au

at

ax

-+u-+E=O

+

Let p = 1 O ( E ) ,u = O ( E ) ,and E = O ( E ) . Use the method of averaging to determine the temporal as well as the spatial variation of the amplitude and phase of a monochromatic traveling wave. 5.16. Consider the problem $ t t - 4zx 4- 4 = 0 (a) Show that e = kx - wt 4 = a cos e,

+

is a solution of this equation if w2 = k2 1. (b) Show that the above equation can be written in the conservation forms (Whitham, 1965b)

a

a

at M4," + 4x2 + 491 + ax ( - + 2 4 t )

=0

(c) Let 4 = a cos 8 with a = a ( x , t ) , k = ,e, and w = --Bt in the conservation equations. Assume that a, w , and k are slowly varying functions of x and t. Hence average these equations over @ = 0 to 2 ~ keeping , a, w , and k constants, and obtain

ak + o r ak at

aa2

5.17. Consider the equation

at

Utt

=0

ax

a +(Wh2) ax

- c2ux,

=0

+ u = €U3

(a) Write down the Lagrangian corresponding to this equation, (b) determine a first-order expansion for traveling waves with constant wave number and frequency but both spatially and temporally varying amplitude and phase. 5.18. The problem of nonlinear transverse oscillations in a homogeneous free-free beam with a nonlinear moment-curvature relationship is described by the Lagrangian

EXERCISES

227

where p , ,8, and t are constants. Determine a first-order expansion for traveling waves with slowly varying amplitudes and phases using (a) the variational approach, and (b) writing down the governing equation and then using the method of averaging. 5.19. Consider Bretherton's (1964) model equation 4tt

+ 4,xx + h x +

+

=

P

(a) Show that the linear problem has the dispersion relationship W'

= k 4 - k2 + 1

(b) Determine the wave number corresponding to nth-harmonic resonance. (c) Use the method of averaging to determine a first-order expansion near

the second-harmonic resonance condition (let the amplitudes and phases be functions of z and t). (d) Write down the corresponding Lagrangian, and then use the variational approach to determine a first-order expansion near the second-harmonic resonance condition.

Perturbation Methods ALI HASAN NAYFEH Copyright Q 2004 WILEY-VCH Valag GmbH & Co. KGaA

CHAPTER 6

The Method of Multiple Scales

6.1. Description of the Method There are three variants of the method of multiple scales. We describe them by discussing the linear damped oscillator x+x=

-2d

(6.1.1)

We chose this example because its exact solution is available for comparison with the approximate solution obtained, and because we will be able to display the different variants of the method more clearly without involving ourselves in algebra. To start, let us determine a straightforward asymptotic expansion for small 4. Thus we assume that x = xo + EX1

+ GX, + . . .

(6.1.2)

Substituting (6.1.2) into (6.1 .I) and equating coefficients of equal powers of E to zero lead to (6.1.3) xo xo = 0

+ XI + x1 = -2x o xz + 2.2 = -2x1

The general solution of (6.1.3) is 20

= a cos ( t

+ 4)

(6.1.4) (6.1.5)

(6.1.6)

where a and (b are arbitrary constants. Substituting for xo into (6.1.4) and solving the resulting equation, we obtain 21

= -at cos ( t

+ 4)

(6.1.7)

Substituting for x1 in (6.1.5) and solving for x,, we obtain xZ = +at2cos ( t

+ 4) + Qat sin ( t + 9) 228

(6.1.8)

6.1.

DESCRIPTION OF THE METHOD

229

Therefore 2

= a cos ( 2

+ 4)- Ear cos ( r + 4) + +r2a[t2cos (t + 4) + t sin (t + I#)] + O ( 2 )

(6.1.9)

It is obvious that (6.1.9) is a poor approximation to 5 when t is as large as r-l, because then the second (a1) and the third (c2z2) terms are not small compared to G, and rzl, respectively (xl and z2 contain secular terms), as was assumed when we carried out the above expansion. Thus the straightforward expansion is not valid when t increases to O(c-l), and the source of the difficulty is the infinite domain as discussed in Section 2.1. The failure of the above straightfonvard expansion can be seen by investigating the exact solution of (6.1.1), which is given by 2

= ae-"cos

141- e2 t + 41

(6.1.10)

Equation (6.1.9) can be obtained by expanding (6.1.10) for small E with t kept fixed. Thus the exponent and cosine factors are represented by

+ gc2t2 + . . c o s ( J 1 - r 2 t + 4)= cos(t + 4) + +2tsin exp (-d) = 1 - r t

(6.1.11) (t

+ 4) i...

(6.1 .12)

It is clear that exp (-d) can be approximated by a finite number of terms only if the combination d is small. Since E is small, this means that t = O(1). When t is as large as r-l, et is not small and the truncated expansion breaks down. The above truncated series is satisfactory up to a certain value of t after which exp (- rf) and the truncated series differ from each other by a quantity that exceeds the prescribed limit of accuracy. Adding more terms to the truncated series increases the value of r to a new value 1' for which this truncated series is satisfactory. However, for t > t', the difference between exp (- r t ) and the new truncated series again exceeds the prescribed limit of accuracy. All terms of the series are needed to give a satisfactory expansion for exp (- d) for all 1. Thus to determine an expansion valid for times as large as cl, the combination rt should be considered a single variable TI = O(1). Then any truncated expansion for exp (-d) valid for times as large as r--l is of the form exp (- e t ) = exp (- T I )

(6.1.13)

Similarly, the truncated expansion (6.1.12) is not satisfactory when t is as large --as O(E-~). To obtain a truncated asymptotic expansion for cos [Jl - c2 t 41 valid for t = O ( c 2 ) , r2t should be considered a single

+

230

THE METHOD OF MULTIPLE SCALES

variable T2 = O(1). With this condition

cos [JI - e2 t

+ 41 = cos [t - iT2 + 4 - $e4t + ‘1 = cos (t - i T z + 4) + &*t sin (t - $T2 + $) + . * .

*

.

(6.1.14) Expansion (6.1.14) is valid when t = O(zc2) because the correction term (second term) is O(e2) or less for all times up to O(6-3. However, this expansion breaks down when t = O(E-~)because the second term ceases to be small compared to the first. To obtain an expansion valid for times as large as O(c4), another variable, T4= e4t = O(1), should be introduced. The above discussion suggests that z(t; e) depends explicitly on t , d , e 2 t , . . . , as well as E itself. This can also be seen from the exact solution. Thus, in order to determine a truncated expansion valid for all f up to O<E-”‘>,where M is a positive integer, we must determine the dependence of z on the M 1 different time scales To, T I ,. . . , T M , where

+

T , = emt

(6.1.1S)

The time scale Tl is slower than To,while the time scale T, is slower than Tl. In general, T,, is slower than Tn-l. Thus we assume that s(t; E ) = Z(T0, TI,

=

. . . , TIM;E )

2 emsm(To,Ti, . . . , T M )+ O(eTfi1)

M-1

m=O

(6.1.16)

The error in (6.1.16) is stated O(eTM)to remind the reader that this expansion is valid for times up to O ( E - ~ ) .Beyond these times, we must use other time scales to keep the expansion uniformly valid. Equations (6.1.15) and (6.1.16) show that the problem has been transformed from an ordinary differential equation to a partial differential equation. If the original problem is a partial differential equation, then the introduction of different time scales increases the number of independent variables. By using the chain rule, the time derivative is transformed according to (6.1.17) Equations (6.1.15) through (6.1.17) formulate one version of the method of multiple scales ; namely, the many-variable version. This technique has been developed by Sturrock (1957, 1963), Frieman (1963), Nayfeh (196% d , 1968), and Sandri (1965, 1967). Equations (6.1.16) and (6.1.17) show that a uniformly valid expansion is obtained by expanding the derivatives as well as the dependent variables in powers of the small parameter. Hence Sturrock and Nayfeh called this technique the derivative-expansion method.

6.1.

DESCRIPTION OF THE METHOD

231

Substitution of (6.1.16) and (6.1.17) into (6.1.1) and equating coefficients of like powers of E , we obtain equations for determining xo,xl, . . . ,xM. The solutions of these equations contain arbitrary functions of the time scales T I ,T,, . . . , Ti?,. In order to determine these functions, additional conditions need to be imposed. If (6.1.16) is to be valid for times as large as E-M, P x , should be a small correction to which in turn should be a small correction to E ~ - ~ X , - ~ .Thus we require that

2,< oo for all T,,T,, . . . , T~ 2,-1

This condition does not mean that each x, is bounded. In fact, each x, may be unbounded. However, this condition requires, as in Lighthill's technique (Section 3.2), that higher approximations be no more singular than the first term. This condition is equivalent to the elimination of secular terms. The second version of the method of multiple scales was introduced by Cole and Kevorkian (1963) and applied by Kevorkian (1966a) and Cole (1968) to several examples. Morrison (1966a) showed that this procedure is equivalent to the method of averaging to second order, while Perko (1969) established their equivalence to nth order. Kevorkian (1966b) showed the equivalence of this procedure and von Zeipel's method to first order. If we investigate the exact solution (6.1.10), we find that t appears in either of the combinations E? or 41 - 3f.Hence to determine an expansion valid for large times, one introduces the two time scales f =

E?

and

7 = 41

- G t = (1 - +cz - 4c4 +

. . a )

t

(6.1.18)

Therefore Cole and Kevorkian (1963) assumed that x(t; 4 = q f , 17; 4

2 E m X m O ; 7)+ 0 ( c M )

M-1

=

where

(6.1.19)

m=O

t =Et,

q = (1

with constant w,. In this case transformed according to

+

+

E ~ C O ~ c3w3

+ - - - +c

~ c o ~ (6.1.20) ) ~

is slower than 7,and the time derivative is

(6.1.21) These two versions can be generalized considerably. Thus the manyvariable version can be generalized (Nayfeh, 1967b) by using an asymptotic

232

THE METHOD OF MULTIPLE SCALES

sequence d , ( ~ ) rather than powers of

E.

Thus

T,, = B n ( E ) f

(6.1.22)

n = 2 dn(E) a dt

qr=o

aTn

(6.1.23)

Equations (6.1.22) and (6.1.23) can be generalized further by letting Tn

= 4I(E)ga[rU,,(4fl

(6.1.24) (6.1.25)

where ,u,,(E) is another asymptotic sequence. Thus (6.1.24) allows for linear as well as nonlinear time scales. Similarly the two-variable expansion procedure can also be generalized. Thus (6.1.20) and (6.1.21) can be generalized to

(6.1.27) This general form was developed by several investigators including Kuzmak (1959), Cochran (1962), Mahony (1962), and Nayfeh (1964, 1965b). Klimas, Ramnath, and Sandri (1970) investigated the role of gauge transformations for uniformization of asymptotic expansions. The method of multiple scales is so popular that it is being rediscovered just about every 6 months. It has been applied to a wide variety of problems in physics, engineering, and applied mathematics. Cole and Kevorkian (1963), Nayfeh (1965c, 1967b, 1968), Kevorkian (1966a), Davis and Alfriend (1967), Schwertassek (1969), and Musa (1967)Rasmussen (1970), and Reiss (1971) analyzed weakly linear and nonlinear vibrations governed by second- or third-order ordinary differential equations. Kuzmak (1959) studied nonlinear oscillations in second-order differential equations with slowly varying coefficients. Cochran (1962), Nayfeh (1964, 1965b), and Fowkes (19681) used the generalized version to analyze turning point problems for second-order linear differential equations. Cochran (1962), Nayfeh (1964, 1965b), and Ramnath and Sandri (1969) used the generalized method to study linear equations with variable coefficients, while Cheng and Wu (1970) analyzed the effect of the scales on the problem of an aging spring. Noerdlinger and Petrosian (1971) discussed a linear inhomogeneous equation with slowly varying coefficients which describes the

6.1.

DESCRIPTION OF THE METHOD

233

effect of cosmological expansion on self-gravitating ensembles of particles, while Kevorkian (1971) investigated the problem of passage through resonance for a one-dimensional oscillator with a slowly varying frequency. Cochran (1962), O’Malley (1968a, b), and Sear1 (1971) applied the generalized method to boundary value problems for certain nonlinear second-order differential equations, while Cochran (1 962) and Ackerberg and O’Malley (1970) applied this method to second-order equations that exhibit turning points and boundary layers. Tam (1968) used the generalized version to solve the Orr-Sommerfeld equation. In orbital mechanics, Nayfeh (1965a) used the generalized version to analyze the earth-moon-spaceship problem. Ting and Brofman (1964) and Nayfeh (1966) analyzed the problem of takeoff of a satellite from a circular orbit with a small thrust, Shi and Eckstein (1966) investigated takeoff from an elliptical orbit with a small thrust, Kevorkian (1966a) and Brofman (1967) studied the motion of a satellite subjected to a small thrust or drag, and Eckstein and Shi (1967) analyzed the motion of a satellite with variable mass and low thrust. Eckstein, Shi, and Kevorkian (1966a) determined the motion of a satellite around the primary in the restricted problem of three bodies, while Alfriend and Rand (1969) determined the stability of the triangular points in the elliptic restricted problem of three bodies. Eckstein, Shi, and Kevorkian (1966~)evaluated higher-order terms in the motion of a satellite using the energy integral and evaluated the effects of eccentricity and inclination (1966b). Shi and Eckstein (1968) analyzed the motion of an artificial satellite having a period commensurable with the rotation period of the primary. Alfriend (1970) and Nayfeh (1971b) studied the two-to-one resonances, while Nayfeh and Kame1 (1970b) and Alfriend (1971b) studied the three-to-one resonances near the equilateral libration points. Alfriend (1971a) analyzed two-to-one resonances in two-degree-of-freedom Hamiltonian systems. In flight mechanics, Ashley (1967) discussed the role of different time scales in flight mechanics, while Nayfeh and Saric (1971b) analyzed nonlinear resonances in the motion of a missile with slight asymmetries. Nayfeh (1969a) used the generalized version to study the motion of a rolling missile with variable roll rate and dynamic pressure but linear aerodynamics, while Nayfeh and Saric (1972a) studied the motion of a missile with nonlinear aerodynamics and variable roll rate and dynamic pressure. Ramnath (1970b) studied the transition dynamics of VTOL aircraft. In solid mechanics, Amazigo, Budiansky and Carrier (1970) analyzed the nonlinear buckling of imperfect columns, while Reiss and Matkowsky (1971) investigated the nonlinear dynamic buckling of a compressed elastic column. Mortell (1968) analyzed the problem of a traveling wave on a cylindrical shell and the propagation of waves on a spherical shell (1969). Kelly (1965)

234

THE METHOD OF MULTIPLE SCALES

and Morino (1969) studied nonlinear panel flutter, while Sprigs, Messiter, and Anderson (1969) discussed membrane flutter. In partial differential equations, Cochran (1962), Nayfeh (1965b), and Comstock (1971) treated elliptic equations. Fowkes (1968, Part 11) obtained uniformly valid expansions for caustic problems. Neubert (1970) obtained solutions for the Helmholtz equation for turbulent water. Wingate and Davis (1970) discussed the propagation of waves in an inhomogeneous rod. Keller and Kogelman (1970) treated a nonlinear initial boundary value problem for a partial differential equation. Luke (1966) studied the Klein-Gordon equation and general variational equations of second order, while Emery (1970) treated the case of several dependent variables and several rapidly rotating phases. Ablowitz and Benney (1970) investigated the evolution of multiphase modes for the Klein-Gordon equation. Nayfeh and Hassan (1971) and Nayfeh and Saric (1972b) discussed nonlinear dispersive waves on the interface of two fluids and in a hot electron plasma. Parker (1969) analyzed the effects of relaxation and diffusive damping on dispersive waves. In wave interactions, Benney and Saffman (1966), Benney (1967), Davidson (1967), Benney and Newell (1967), Hoult (1968), Newell (1968), and Benney and Newell (1969) investigated the nonlinear interaction of random waves in a dispersive medium. Davidson (1969) studied the time evolution of wave correlations in a uniformly turbulent ensemble of weakly nonlinear dispersive systems. In water waves, Carrier (1966) analyzed gravity waves in water of variable depth, while Hoogstraten (1968) and Freeman and Johnson (1970) studied shallow water waves in shear flows. Jacobs (1967) solved the tidal equations. Murray (1968) treated free surface oscillations in a tank resulting from drainage. Chu and Mei (1970) studied slowly varying Stokes’ waves. McGoldrick (1970) and Nayfeh (1 970b) treated the second-harmonic resonance case, while Nayfeh (1970d, 1971a) investigated the third-harmonic resonance case in the interaction of capillary and gravity waves. In atmospheric science, Newell (1969) treated the resonant interaction of Rossby wave packets, while Stone (1969) analyzed the problem of baroclinic waves. Shabbar (1971) discussed the side-band resonance mechanism in the atmosphere supporting Rossby waves, while Lindzen (1971) studied the propagation of equatorial Tanai and Kelvin waves through shear. In plasma physics, Ball (1964), Taussig (1969), and Tam (1969, 1970) analyzed the propagation of nonlinear waves in a cold plasma, while Nayfeh (1965d) and Das (1971) investigated nonlinear oscillations in a hot electron plasma. Davidson (1968) treated nonlinear oscillations in a Vlasov-Maxwell plasma. Peyret (1966) analyzed plasma waves in an accelerator, while Butler and Gribben (1968) discussed nonlinear waves in a nonuniform plasma.

6.1.

DESCRIPTION OF THE METHOD

235

Maroli and Pozzoli (1969) studied the penetration of high-frequency electromagnetic waves into a slightly ionized plasma. Abraham-Shrauner (1970a, b) investigated the suppression of runaway of electrons in a Lorentz gas. Chen and Lewak (1970), Chen (1971), and Prasad (1971) studied parametric excitation in a plasma, while Lewak (1971) discussed the interaction of electrostatic waves in a plasma. Dobrowolny and Rogister (1971) and Rogister (1971) analyzed the propagation of hydromagnetic waves in a highbeta plasma. In hydrodynamic and plasma stability, Frieman and Rutherford (1964) developed a kinetic theory for weakly unstable plasmas, while Albright (1970) analyzed the stabilization of transverse plasma instability. Kelly (1967) investigated the stability of an inviscid shear layer. Benney and Roskes (1969) analyzed the instability of gravity waves. Kiang (1969) and Nayfeh (1969b) studied Rayleigh-Taylor instability, while Newel1 and Whitehead (1969) analyzed postcritical Rayleigh-Binard convection. Nayfeh (1 970c) investigated the nonlinear stability of a liquid jet. Nayfeh and Saric (1971a) studied nonlinear Kelvin-Helmholtz instability, while Puri (197 1) analyzed the effects of viscosity and membrane on the oscillation of two superposed fluids. Stewartson and Stuart (1971) treated the nonlinear stability of plane Poiseuille flow. Mitchell (1971) applied this technique to combustion instability. In fluid mechanics, Germain (1967) and Lick (1970) reviewed recent developments in aerodynamics and nonlinear wave propagation in fluids including the methods of matched asymptotic expansions, strained coordinates, and multiple scales. Benney (1965) analyzed the flow field produced by finite-amplitude oscillation of a disk about a steady state of rotation, while Barcilon (1970) studied the linear viscous theory of steady rotating fluid flows. Rubbert and Landahl (1967) discussed the transonic airfoil problem. Peyret (1970) treated the problem of steady flow of a conducting perfectly compressible fluid in a channel. Chong and Sirovich (1971) studied the problem of steady supersonic dissipative gas dynamics. Cheng, Kirsch, and Lee (1971) analyzed the behavior of a strong shock wave initiated by a point explosion and driven continuously outward by an inner contact surface. In general physics, Caughey and Payne (1967) used a combination of the method of multiple scales and the method of matched asymptotic expansions to solve the Fokker-Planck equation arising from the response of selfexcitated oscillators to random excitations. Brau (1967) obtained a stochastic theory for the dissociation and recombination of diatomic molecules. Ramnath (1 970a) obtained an approximation for the Thomas-Fermi model in atomic physics and treated a class of nonlinear differential equations arising in astrophysics (1971). Meyer (1971) investigated Rayleigh scattering

236

THE METHOD OF MULTIPLE SCALES

of a laser beam from a massive relativistic two-level atom, while Nienhuis (1970) studied Brownian motion with a rotational degree of freedom. In statistical mechanics, Maroli (1966) solved Boltznann's equation to obtain a kinetic theory of high-frequency resonance gas discharge breakdown, while Caldirola, De Barbieri, and Maroli (1966) solved Boltzmann's equation for the electronic distribution fiinction. De Barbieri and Maroli (1967) solved the Liouville equation to analyze the dynamics of weakly ionized gases, while Goldberg and Sandri (1967) and Ramanathan and Sandri (1969) derived sets of hierarchical equations. In the remainder of this section, we describe the three versions of the method of multiple scales and their application to the simple linear damped oscillator given by (6.1.1). In the following sections we apply these techniques to different problems in mathematical physics. MANY-VARIABLE VERSION (THE DERIVATIVEEXPANSION PROCEDURE) Substituting (6.1.16) and (6. I. 17) into (6.1.1) and equating coefficients of like powers of E, we obtain the following equations for xo, x l , and x2 6.1 . I .

azx0 +xo=o

(6.1.28)

aT,"

azxl + XI

= -2-

aT,"

azX2 8x1 - 2 + x2 = -2 -

aT,"

aTo

aZX, ~

aToaTl

ax0

aTo

- 2-

aaXo

azx0 azx0 -2 ___

aTlz

(6.1.29)

aTo aTl aToaTz

8x0

- 2 - (6.1.30)

aT

The general solution of (6.1.28) is

-+

xo = Ao(Tl, T2)eiT0 AO(Tl,T2)e-iT0

(6.1.31)

where A, is the complex conjugate of A,. This solution is simply equivalent to (6.1.6) where a and #J are taken to be functions of the slow time scales T I and T , rather than being constants. Substituting for xo from (6.1.3 1) into (6.1.29), we obtain

The general solution of (6. I .32a) is

+ xl(Tl,T2)e-iT0

x1 = A,(T,, T2)eiT0

6.1. DESCRIPTION

OF THE METHOD

237

Comparing (6.1.32b) with (6.1.31) shows that EX, is a small correction to x only when €To = cf is small. In order to obtain an expansion valid for times as large as O(E-~),the secular terms, Toexp (&iT,,), in (6.1.32b) must vanish; that is (6.1.33) or A = a,( Tz)e-T1

(6.1.34)

Then (6.1.32b) becomes

+

x1 = A,(T,, Tz)etTo Al(T,, T2)e-'To

(6.1.35)

- + x2 = -Q(Tl, T2)eiTo- Q(Tl, Tz)e-'To

(6.1.36)

Using xo and x1 in (6.1.30), we obtain aZx,

where

a T,'

Q(T,, T2)= 2iA1

aA + 2i -,-

+

-aoe-T1 2i -e- T I (6.1.37) aT1 The terms on the right-hand side of (6.1.36) produce secular terms because the particular solution is x2 = $iQ(Tl, Tz)ToeaT0- ,+i&(T,,T2)Toe-'To (6.1.38) These secular terms make e2x2 the same order as €5, when t is as large as O(c-l). In order to eliminate these secular terms, Q must vanish; that is

(6.1.39) In general, one does not need to solve for x2 in order to arrive at (6.1.39). One needs only to inspect (6.1.36) and eliminate terms that produce secular terms. The general solution of (6.1.39) is

[

A, = al(Tz)

1

Tl + ti (--ao + 2i a TZ

Substituting for A, into (6.1.35), we obtain --a,

1

eaT1

+ 2i-aT* T, e-T1eiTD+ CC

(6.1.40)

(6.1.41)

where CC stands for the complex conjugate of the preceding expression. However xo = [aoe'To + go,e-iT~]e-T1 (6.1.42)

238

THE METHOD OF MULTIPLE SCALES

Therefore, as TI+ co, although z, and 5, + O , E Z , becomes O(z,) as t increases to O(ce). Thus the expansion z, €5, breaks down for t as large ) the coefficients of TI in the brackets in (6.1.41) vanish; as O ( E - ~unless that is, unless

+

(6.1.43)

or

- a 00e-iT,/2

(6.1.44)

a,-

where a,, is a constant. Then (6.1.40) becomes A, = al(T2)e-T1

(6.1.45)

Therefore

The function a,(T,) can be determined by carrying out the expansion to third order al(T2) = alle-"~'2

(6.1.47)

where a,, is a constant. If we assume that the initial conditions are such that z(0) = a cos 4 and i ( 0 ) = --(sin $J1 - + E cos 4) and replace T, by €9,we obtain z = ae-Ftcos i t - +e2t 4) R (6.1.48)

+ +

where R is the remainder. From (6.1.10) and (6.1.48), we find that

For linear equations such as (6.1.1), we may introduce the different time scales without expanding z.Thus using (6.1.17) in (6.1.1), we obtain a2

a2

[%+2.--

aT, aTl

+

-+2-) az 4aT;

a2

aT, aT,

a +-

+

.)z

(6.1.50)

6.1.

DESCRIPTION OF THE METHOD

Equating the coefficients of like powers of

E

239

to zero yields (6.1.51) (6.1.52) (6.1.53)

The general solution of (6.1.51) is

+ x(Tl, T2)ePiTO

x = A(T,, Tz)e'TO

(6.1.54)

Substituting into (6.1.52), we obtain (6.1.55)

Since (6.1.55) is valid for all To, the coefficients of exp (iTo)and exp (-iTo) must vanish; that is aA -+A=O (6.1.56) aTI or A = a(Tz)e-T1 (6.1.57) Substituting (6.1.54) into (6.1.53) yields (6.1.58)

Thus A aA + 2-JaTl + 2i =0 aT2

a2A aT12

(6.1.59)

Substituting for A from (6.1.57) into (6.1.59) gives

aa

2i--u=O

Hence

(6.1.60)

a T,

a = aoe-'T*'2

(6.1.61)

where a, is a constant. Therefore (6.1.54) becomes =a

e-Tiei(T.-T,12)

+

cc

(6.1.62)

240

THE METHOD OF MULTIPLE SCALES

Expressing (6.1.62) in terms of t yields x = ae-Etcos( t - fr2t

+ 4)

(6.1.63)

where a, = (1/2)a exp (+I. This result is in full agreement with (6.1.48). 6.1.2. THE TWO-VARIABLE EXPANSION PROCEDURE Changing the independent variable from r to 6 and q as defined by (6.1.21), we transform (6.1.1) into

- -241 We assume that 5

= xo(5, q)

+2wz+**.)-

ax - 2 2 - ax

a7

at

+ 4 5 , q ) + e2x*(t,7)+ - -

(6.1.64)

(6.1.65)

Substituting (6.1.65) into (6.1.64) and equating coefficients of equal powers of E on both sides, we obtain ---o+x,=o a2

(6.1.66)

atx

atxz

- + XZ + 2 arlz

azxl ,+x,+22=-2-

8x0

817

%

azx0 w , 7 arl

a t aq

zx ' = -2- ax, - 2-ax, + 2-ataq + a-at2 aq at agXl

(6.1.67) (6.1.68)

The general solution of (6.1.66) is xo = A,(E)e'q

+ A,(t)e-'q

(6.1.69)

With this solution, (6.1.67) becomes (6.1.70)

Eliminating the terms that produce secular terms in (6.1.70) gives

Hence

dAo -+A,=O dt

(6.1.71)

6.1.

DESCRIPTION OF T H E METHOD

241

The solution of (6.1.71) is

A , = a,e-e (6.1.73) where a, is a constant. Substituting the above solutions for xo and z1into (6.1.68) gives

-_a Z x ,

aq

+ x2 = [-2i(%

+ A') + ( 2 0 , + l)a,ed

1

es

+ CC

(6.1.74)

Eliminating the terms that produce secular terms in (6.1.74) yields

dA1 + whose solution is

= -+i(2w2

AE

+ l)a,e-c

(6.1.75)

+

A , = alept - f i ( 2 0 , l)a,Ee+ (6.1.76) Substituting for A , into (6.1.72) and comparing the result with (6.1.69) show that x l / z 0 is unbounded as 6 + 00 unless

-$

(6.1.77)

02=

Therefore in terms of t (6.1.65) becomes

+ +

x = ae-" cos ( t - f 2 t 4) O ( 2 ) (6. I . 78) where Q, €al = (1/2)a exp ti+). This expression is in full agreement with that obtained using the many-variable version (derivative-expansion method).

+

6.1.3. GENERALIZED METHOD-NONLINEAR SCALES We first introduce a new variable T = ct to transform (6.1.1) into

c ( ;d'x ;,+2$)+z=o

(6.1.79)

In order to determine a uniformly valid expansion, we let E = 7 ,

r / = - g-1(7) e

+

&(T)

+

€g1(T)

+

' ' '

,

gi(0) = 0 (6.1.80)

where g iis determined in the course of analysis. The derivatives with respect to T are then transformed according to (6.1.81)

+ [+g

+ g&) + cgi(E) +

* *

-1'5

(6.1.82)

242 THE METHOD OF MULTIPLE SCALES We assume that x possesses a uniformly valid expansion of the form x = xo(E, q)

+ q ( E , 7) +

E2X2(E, q)

+

* *

.

(6.1.83)

Substituting (6.1.81) through (6.1.83) into (6.1.79) and equating coefficients of like powers of E, we obtain (6.1.84)

(6.1.85)

The general solution of (6.1.84) is

(6.1.86)

ie +

Substituting for xo into (6.1.85) gives r2

a2x,

+ zl = -[(- % + g-1

2i)A,

g-1

The terms on the right-hand side of (6.1.87) produce secular terms. In order to eliminate secular terms

(- + +i9 + 2i g-1 g-1

7Ao = 0 (6.1.88)

Since (6.1.88) must be valid for all q , and A, # 0 for a nontrivial solution, we require that g,: = 0 or g-, = CE since q(0) = 0 (6.1.89) where c is an arbitrary constant which can be taken to be unity without loss of generality. Then (6.1.88) becomes A;

whose solution is

+ (1 + ig;)Ao

=0

A 0 -- aoe-S-im(QJ

(6.1.90) (6.1.91)

where a, is a constant. With A, and g-, known 2,

= aoe-rei(r/~)

+ (ige-re-i(r/c)

(6.1.92)

Equation (6.1.92) shows that gocancels out, hence the solution is independent

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

243

of the value of go. Therefore we might as well set it equal to zero without loss of generality. Hence A, becomes A, = a,e&

(6.1.93)

As a result of (6.1.88), the solution for x, is

+

(6.1.94) z, = Al(t)eiq A,(t)e-a'J With g-, = and go = 0, the equation for x 2 can be determined by substituting (6.1.81) through (6.1.83) into (6.1.79) and equating the coefficient of cz to zero. Thus

azxz +s,+2-

a2x, az, ,aZx, azx, ax + 2-+2 g 1 2 +7+2 ata7 arl arl a t at O-

=0

(6.1.95)

Substituting for zo and z, into (6.1.95) yields

- + r2 = -[2 i(A;

ar2

+ A,) - (2g; 4- l)~,e-*]e*~+ CC

(6.1.96)

Eliminating the terms that produce secular terms in (6.1.96), we obtain A;

+ A , = -4i(2g; + l)a,e-c

(6.1.97)

+ t)e-[

(6.1.98)

The solution of (6.1.97) is A, = ale-( - +iao(2g,

where a, is a constant. Equation (6.1.98) shows that xl/xo is unbounded as t 4 00 unless g1= --*t (6.1.99) In terms of t = T / C , the expansion becomes

+

L

= ue-'tcos ( t

- +c2t + +) + O(2)

(6.1.100)

where a, al= (lj2)a exp ($). This expansion is again in agreement with those obtained using the derivative-expansion and the two-variable expansion versions of the method of multiple scales. 6.2.

Applications of the Derivative-Expansion Method

6.2.1. THE DUFFING EQUATION The second example to which we apply the derivative-expansion method is the Duffing equation

- + w,"u d2u

dt2

+

Ell3

=0

(6.2.1)

244

THE METHOD OF MULTIPLE SCALES

We assume that

2

a

Then

d (6.2.3) -=Do+~Dl+~2D2+..., D =dt aT, Substituting (6.2.2) and (6.2.3) into (6.2.1) and equating coefficients of each power of E to zero, we have

+ oo2uo= 0 Do%, + oo2u1= -2DoDluo - uO3

Do%,

(6.2.4) (6.2.5)

+

DO2u2 wO2u2= -2DoD1~1 - 2DoDZuO - Dl2U0 - 3Uo2u1 (6.2.6) The solution of (6.2.4) is

+

uo = A(Tl,T2)eiWoTa A(Tl,T2)e-aWoTa Equation (6.2.5) then becomes

+

+

(6.2.7)

+

D,2ul o 2 u 1 = -[2iw,D1A 3A2A-]eiWoTo - A I e 3 i W o T o CC (6.2.8) In order that u,/uo be bounded for all To, terms that produce secular terms must be eliminated. Hence

+

2 i o , ~ , A 3A2K = O

(6.2.9)

and the solution for u1 becomes (6.2.10)

To solve (6.2.9), we let A = (1/2)uei6 with real a and $, separate real and imaginary parts, and obtain (6.2.11)

Hence a = u(T2),

4 = -a 2 T + $o(T2) 3

8WO Substituting for uo and u1 into (6.2.6) yields

(6.2.12)

- Q(T,, TZ)eiWaTo + CC (6.2.13)

where Q

= 2i0oD1B

+ 3A2B + 6 A A B+ 2ioODzA - 8w:

(6.2.14)

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

245

Secular terms are eliminated if B=O

and

(6.2.15)

15A3A2 2iwoD2A = -

(6.2.16)

8W,2

With Q = 0, the solution of u2, disregarding the homogeneous solution, is (6.2.17) Letting A = (l/2)aei+ in (6.2.16) and separating real and imaginary parts, we obtain

w

-aa= o , aT2

15

-ma-=-

aT2

256~0,

(6.2.18)

(I4

Equations (6.2.12) and (6.2.18) lead to a = a constant, hence (6.2.19a) where

x

is a constant. Therefore

(6.2.19b) Substituting for uo, ul, and u, into (6.2.2), keeping in mind that A = (1/2)aexp (i4) and expressing the result in terms of t , we obtain u =a

cos ( O l t

+ x) +

2-

32w0 Ea3

(1 -

320,~

2a +-10240:

where w =

W"

3a2 +80,

E

cos 5(wt

15a4 - -2

2560,~

+ x)

cos 3(wt

€=)

+ x) + O(c3)

+ O(E3)

In the last two terms of (6.2.20a), w0 is replaced by

w

(6.2.20a)

(6.2.20b)

with an error O(e3).

6.2.2. T H E VAN DER POL OSCILLATOR As a second example, we consider the van der Pol oscillator

-+ u d2u dt2

= E(l -

du

2)-

dt

(6.2.21)

246

THE METHOD OF MULTIPLE SCALES

Substituting (6.2.2) and (6.2.3) into (6.2.21) and equating the coefficients of like powers of e, we obtain

+ u, = 0 Do%, + = -2D,D1~0 + (1 - U;)D,U, Do%, + = -2DoD1~1 - Dl2uo - 2DoDZuo + (1 - U;)D,U,

Do2uo

~1

UZ

+ (1 - U ~ ) D l U O- 2u,u,D,u,

(6.2.22) (6.2.23) (6.2.24)

The solution of (6.2.22) is

+

uo = A(Tl, T2)eiTa J(Tl, Tz)e-iTa

Substituting for D:u,

uo into

(6.2.25)

(6.2.23) gives

+ u1 = -i(2D1A

-A

+ A2x)eaTo- iA3e3iT0+ CC

(6.2.26)

To eliminate terms that produce secular terms, we require the vanishing of the coefficients of exp ( f i T , ) ; that is 2D1A = A - A Z K

(6.2.27)

Then the solution of (6.2.26) is

+ iiA3e3'To+ CC

u1 = B(Tl, Tz)eiTo In order to solve (6.2.27), we let

A = M T l , T2) exp i4V1, T,)

(6.2.28) (6.2.29)

Separating real and imaginary parts in (6.2.27), we obtain (6.2.30)

Hence (6.2.31)

If we are interested in the first approximation to u , then we consider B, and c as constants. Mofeover, if u(0) = a, and du(O)/dt = 0, then where

u = a cos t

+ O(E)

4,

(6.2.32) (6.2.33)

which is in agreement with the expansion obtained in Section 5.4.2 using the Krylov-Bogoliubov-Mitropolski technique.

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

247

To determine the second approximation, we need to determine the functions B, 4, and c. Thus we substitute for u, and u1 into (6.2.24) and obtain where

Do%,

+ u, = Q(Tl, Tz)eiTo+ &(Tl, T2)eCiT0+ NST

+ i(l - 2AX)B - iA2B- 2iD,A

Q = -2iD,B

(6.2.34a)

- D12A

+ (1 - 2AK)DIA - A2D1K+ &A3K2 (6.2.34b)

Secular terms will be eliminated if Q = 0. To solve (6.2.34b) with Q = 0, we let B = (1/2)ib exp i$ with b real and $ defined in (6.2.29). We substitute for A and B into (6.2.34b) with Q = 0, separate real and imaginary parts, and obtain

aa a T2

- = 0,

ab 2 -- ( I - &')b a TI

=

or

a = a(Tl)

-,

d4 d2a - 2 ~- + - (1 dT, dTl

(6.2.35a) 2

)-da

dTl

--

128

a5

(6.2.35b)

With the help of (6.2.30), (6.2.35b) can be expressed in the form

(:I

d -

- - -+2: : (

6 :)

dT

+ (&a

-

k)

du

(6.2.36b)

Integrating, we obtain h = --a($

+ ;ITl + %%a3- &u In a + ab,(T,)

(6.2.36~)

In order that u,/uo be bounded for all TI, the coefficient of Tl in the above expression for h must vanish. This condition gives

4 = --AT2 where tl

+

(6.2.37)

40

is a constant. The expansion of u to second approximation is then

= a cos [(l

- -r'G.2)t + $,I

- e { ( ( ; t a 3-

In a

+ ab,)

+ &u3

sin [(l

sin 3[(1 - &;).'t

- i%c2)f + $,I

+ do])+ O(r2)

(6.2.38)

where a is defined by (6.2.33) and b, is considered a constant to within the order of error indicated. To an error of O(rz), this expression can be written

248

THE METHOD OF MULTIPLE SCALES

as

u = a cos (t

where

- 6) -

6 = &E2t

sin 3(t - 6)

+ O(8)

+ 8. In a - +a2 + 6 ,

(6.2.39a) (6.2.39b)

and 6, = -4, - ~ b = , a constant. This last form of the solution is in full agreement with that obtained in Section 5.4.2 using the Krylov-BogoliubovMitropolski method. 6.2.3. FORCED OSCILLATIONS OF THE VAN DER POL

EQUATION We consider next the response of the van der Pol oscillator, discussed in the previous section, to a periodic external force; that is, the oscillations of the equation du d2U w t u = ~ ( -1 u2)K cos I t (6.2.40) dt2

+

dt

+

where K and 1 are real constants. Four cases arise depending on whether the excitation (external force) is "soft" [i.e., K = O ( E ) ]or "hard" [i.e., K = O(l)], and whether the excitation is resonant [i.e., I - w, = O ( E ) ]or nonresonant [i.e., 1 - o, = O(l)]. Soft Nonresonant Excitation. In this case K = ck, where k = 0(1),and we express cos I t in the form cos IT,. To determine a first approximation to u , we let (6.2.41) = u,(T,, TI) %(TO, Tl) O(E2)

+

+

with To = t and TI = ct. Substituting (6.2.41) into (6.2.40) and equating the coefficients of eo and ,E on both sides, we obtain D,Zu, Do2u,

+

wo2uo =

0

+ wo2u1 = -2D,Dlu, + (1 - u,~)D,u,+ k cos IT ,

(6.2.42) (6.2.43)

The solution of (6.2.42) is u, =

A(T,)e'"OTO

+ A(T,)e-i"oTo

(6.2,44)

Substituting for u, into (6.2.43) gives Dtu,

+ w2u1 = iw0(-2A' + A - A A-) e + 3ke'"' - iwnA3e3i00T0+ CC 2

iwoTo

(6.2.45)

For there to be no secular terms, we require that 2A' = A - A 2 K

(6.2.46)

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

6.2.

249

where the prime denotes differentiation with respect to T I .Then the solution of (6.2.45) is u1 = B(Tl)eimOTO+

A

!

eiATo

2 coo - I 2

+iA3 e3iooTo

+ CC

80,

(6.2.47)

Letting A = (l/2)a exp i# in (6.2.46), separating real and imaginary parts, and solving the resulting equations, we find that # is a constant, while a is given by (6.2.33). Therefore, to first approximation u = a cos mot

+ O(E)

(6.2.48)

where a is given by (6.2.33). Equations (6.2.33) and (6.2.48) show that, to first approximation, neither the phase nor the amplitude is affected by the presence of a soft nonresonant excitation. Moreover, the natural response (i.e., the case with k = 0) dominates the forced response, as expected, since the forcing function is soft. However, as the forcing frequency I approaches the natural frequency coo, the forced response becomes more significant and approaches infinity as can be seen from (6.2.47), and the above expansion is no longer valid. Hard Nonresonant Excitation. (6.2.43) are modified to

+ Do2u1 +

Do2u0

In this case K = O(1) and (6.2.42) and (6.2.49)

= K cos IT,

W ~ U , ~

0

= ~-2DoD,uO ~ 1

+ (I

- UO~)D~UO

(6.2.50)

The solution of (6.2.49) is

+

uo = A(Tl)eiooTo A(Tl)e-i"oTo

Substituting for

uo into

+ w: ~

- I2

cos IT,

(6.2.51)

(6.2.50) gives

+

D ~ u , w:ul = ioo[-2A'

where = 1 - K2/2(wO2require that

+ A7 - A2A]eimoTo + CC + NST

(6.2.52)

. In order to eliminate the secular terms, we

2A' = A17 - A2K

(6.2.53)

To solve (6.2.53), we let A = (1/2)aexp i+, separate real and imaginary

250 THE METHOD OF MULTIPLE SCALES parts, and obtain

4 = a constant and (6.2.54)

The solution of (6.2.54) can be obtained by separation of variables to be In u2 - In (q - &z2) = qT,

If u(0) = a, u is given by

+ a constant

+ [K/(wO2- A2)3 and du(O)/dt = 0, the first approximation to u = a cos w,t

where 2

a =

+

1

+ w,2 - A2 cos At + O(E) ~

411,

(2-

(6.2.55)

(6.2.56)

The steady-state solution (i.e., t co) depends on whether q is positive or negative [i.e., K 2is less than or greater than 2(oO2For negative q, exp ( - q f ) + co as t co, hence a + 0 as f co, and the steady-state solution is ---f

--f

us =

---f

~

m:

- A2

cos I t

+ O(e)

(6.2.57)

However, for positive q , exp ( - c q f ) + O as t + 00, and a+ 2 4 % as 1 a.Consequently, the steady-state solution is ---f

U, =

-

2Jq cos mot

+ -cos I t + O(E) - A2 6J02

(6.2.5 8)

Therefore, if 11, is negative, the natural response fades away and the steadystate solution consists of the forced response only. However, if q is positive, the steady-state solution is a combination of the natural and forced responses, with the amplitude of the natural response modified by the presence of the hard excitation. Soft Resonant Excitation. In this case K = ck with k = O(1), and A CT = O(1). In order to determine a valid asymptotic expansion in this case, we express the excitation in terms of To and TI according to w, = o c with the detuning

K C O SA t = ck cos (mot

+ oct) = ck cos (w,T, + aT,)

(6.2.59)

With this expression for the excitation, the equations for u, and u1 of (6.2.41)

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

are Di2u1

+

+ wo2uo= 0 + (1 - uo2)D,u0 + k cos Do%,

0 0 ~ ~= 1

-2DODluo

The general solution of (6.2.60) is

(wOTO

+

uo = A(T,)eiUoTO K(Tl)e-*wOTO Hence (6.2.61) becomes Dtu,

+ w;u1

= [io0(-2A'

+ oT,)

25 1

(6.2.60) (6.2.61) (6.2.62)

+ A - A2K)+ gkeiuT1]eimoTo - io0A3e3iw0T0 + CC

(6.2.63) The terms proportional to exp ( &iwoTo)in (6.2.63) produce secular terms with respect to the time scale To because the terms in the brackets are functions of TI only. In order that u,/uo be bounded for all To 2A' = A - A2K - 1 ike'"T1 (6.2.64) 200 To solve (6.2.64), we let A = (1/2)a exp i4, separate real and imaginary parts, and obtain (6.2.65) (6.2.66) To eliminate the explicit time dependence of the right-hand sides of (6.2.65) and (6.2.66), we let (6.2.67) Hence (6.2.65) and (6.2.66) become da

- = ga(1

dT,

k . - $2)+ sin y

200

(6.2.68)

k cos y (6.2.69) d TI 20,a Periodic solutions of the externally excited oscillator (6.2.40) correspond to the stationary solutions of (6.2.68) and (6.2.69); that is, da/dTl = d y / d T l = 0 , or k . (6.2.70) $41 - 42) -sin Q = O

d v-- - a + -

+

o+-

k 2w06

200

cos

4=0

(6.2.71)

252 THE =HOD

OF MULTIPLE SCALES

where the tilde refers to the stationary solution. Elimination of Q from these equations leads to the following frequency response equation p(1

k2

- P )+ ~ 4a2p = 7 = F2, 400

a"2 p =-

4

(6.2.72)

+

For a given excitation amplitude rk and frequency 1 = wo e a , (6.2.72) furnishes p , hence the amplitude of harmonic oscillations. To first approximation the harmonic oscillation is given by

+ 6)+ O ( € )

u = a"C0S (w,t

while the frequency of oscillation is

(6.2.73)

(6.2.74) Therefore, as 1 approaches wo, the natural response is entrained by the forced response. The result is a synchronization of the output at the excitation frequency. The stability of these harmonic oscillations can be obtained by letting a = a"+ Aa,

y

=4

+ Ay

(6.2.75)

Developing the right-hand sides of (6.2.68) and (6.2.69) in powers of Aa and Ay and keeping only linear terms, we have = &(l- $r?")Aa

k +cos @ A y

(6.2.76) dT1 200 k d(Aw) - - - k (6.2.77) cos @ ha sin $7 A y dT1 2w0h 2w0a If we let Aa a exp mT, and A y a exp mT,,then rn must satisfy the equation

-

m2 - a m

where =1

- 2p,

+A =0

A = t(l - 4p

(6.2.78)

+ 3p2) +

(6.2.79)

where use has been made of (6.2.70) and (6.2.71). The discriminant of (6.2.78) is D = p2 - 40' (6.2.80) The loci iR = A = D = 0 are called separatrices and shown in Figure 6-1. The locus A = 0 is an ellipse whose center is p = 213, 0 = 0, while the locus D = 0 is the two straight lines p = f2a. The interior points of the ellipse correspond to saddle points, hence the corresponding harmonic oscillations are unstable. The points exterior to the ellipse are nodes if

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

253

P

t

10

0.5

0

0 5

0-

I .o

Figure 6-1

D 2 0 and foci if D < 0. The harmonic oscillations corresponding to these points are stable or unstable according to whether p is greater or less than 1/2. Hard Resonant Excitation. The analysis for this case can be obtained as a special case of the previous case, with k = K ~ Ewhere , the amplitude of excitation K = O(1). Hence k is very large because c is small. Thus, for B near zero, (6.2.72) shows that there exists only one amplitude p for harmonic oscillation and it is stable. As k increases without bound the amplitude increases without bound also.

6.2.4. PARAMETRIC RESONANCE-MATHIEU EQUATION Let us return to the Mathieu equation discussed in Section 3.1.2, namely ii

+ (6 +

f

(6.2.81)

cos 2t)u = 0

According to the Floquet theory of linear differential equations with periodic coefficients, the 6-r plane is divided into regions of stability and instability which are separated by transition curves along which u is periodic with a period of either 7r or 2 ~ In. Section 3.1.2, we determined approximations to the transition curves using the Lindstedt-Poincare method. In this section we find not only the transition curves but also the solutions, hence the degree of stability or instability, as we found in Section 3.1.3 using Whittaker's method. To this end we let 6

and assume that u = uo(To, Ti,

Tz)

= coo2

with positive wo

+ fui(To, Ti, + 7'2)

f2u2(T0,

(6.2.82) Ti, Tz)

+

*

. (6.2.83)

254

THE METHOD OF MULTIPLE SCALES

Different cases have to be distinguished depending on whether far away from an integer n.

near or

coo is

Solution f o r wo Far Awayfrom an Integer. We express cos 2t in terms of the time scale To as cos 2T0. Substituting (6.2.83) into (6.2.81) and equating the coefficients of zo, z, and 8 to zero, we obtain

Do2u2

+

Do2u1 ~

0

+

Do%, 0 0 ~ ~= 1

+ wo2uo= 0

(6.2.84)

-~DOD,U, - U O cos 2To

(6.2.85)

= ~-2DoD1~1 ~ 2 - (01'

+ 2DODz)uo -

The solution of (6.2.84) is u,, = A(T', T2)eiooTo 2(Tl, T&-

+

~1

cos 2To (6.2.86)

i o o To

(6.2.87)

Substituting for uo into (6.2.85) yields

D;ul

+ oopul= -2io

-

+ cc

- 4Aei(oo-2)To

0D1A ~ ~ W O T ~ OA ~ ~ ( w o + ~ ) T o

(6.2.88) Since wo is far away from 1, secular terms will be eliminated if D,A = 0 or A = A(T,). Then the solution of u1 is

Substituting for uo and u1 into (6.2.86) yields

D;u2

+ w:u2

= -2

1qo: - 1)

-

+ 1)

+--l6(oO- 1)

~~i(go+r)TO

l6(oO

+ cc

A~~(QIo-~)To

(6.2.90) Since wo is far away from 1 or 2, secular terms will be eliminated if i

D2A = -

A (6.2.91) 1600(w: - 1) If we let A = (1/2)a exp iq5 and separate real and imaginary parts, we obtain

-da =o, Therefore

dT2

a = a constant,

1 16w0(w2- 1)

(6.2.92)

4 = - l 6 ~ 0 ( 0 0-~ 1) T2 + $0

(6.2.93)

d 4_

dT2

and

-

1

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

255

where $o is a constant. With (6.2.91) the solution of (6.2.90) is u2

=

Aei(oo+ltTo

1

W w O

+ l)(wo + 2) + 128(00 - 1l)(oo- 2) Aei(oo-4)To+ CC

(6.2.94)

Summarizing, to O(e2), the solution for u is u = a cos (cot

+ +o)

where (6.2.96) We emphasize again that this expansion is valid only when coo is away from 1 and 2. As wo -+ 1 or 2, u -+ 03. An expansion valid near oo= 1 is obtained next. Solution for wo Near 1. In this case we let

6=1

+

Ed1

+ €26, + -

(6.2.97)

* *

with 6, and 6, = O(1). Equation (6.2.97) modifies (6.2.84) through (6.2.86) into (6.2.98) Do2uo uo = 0

+

+ = -2DoDl~o - 61~0- cos 2To Do%, + u2 = -2DoD1ul - (D12 + 2D0D2)u0 Do2u,

(6.2.99)

UO

UI

The solution of (6.2.98) is

6 1 ~ 1-

+ K(T,, ~

uo = A(T,, Tz)eiTO

62u0 - ~1 cos 2To (6.2.100)

~ ) e - ~ (6.2.101) ~ ~

Substituting for uo into (6.2.99) gives D ~ u+ , u1 = ( - 2 i D , A - 6,A - &A)eiTo- gAe3*To+ CC (6.2.102) Secular terms with respect to the time scale To will be eliminated if DIA = fi(6,A

+ $2)

(6.2.103)

256

THE METHOD OF MULTIPLE SCALES

Then the solution of (6.2.102) is =116 (

u,

1

~ ~ 3 i T +o ~ - ~ - 3 i T o

To solve (6.2.103), we assume that A = A,

+ iA,

(6.2.104) (6.2.105)

where A, and A ; are real, and separate real and imaginary parts to obtain (6.2.106) (6.2.107) The solution of these equations is

A, = ul(T2)eYITI

+ uz(T2)e-Y1T1

(6.2.108)

A, = 2y1 [ U ~ ( T , ) ~-~uz(7'.z)e-y1T'] '~' t - 61

where

(6.2.109)

Y12 = H i - d l 7

(6.2.110) Here a, and u2 are real valued functions of the time scale T,; however, to first approximation, u, and a, are constant. Equations (6.2.108) through (6.2.1 10) show that A grows exponentially with TI(i.e., with t) if y1 is real or I6,l 5 1/2, and A oscillates with Tl if y1 is imagi&ry or IS,l 2 1/2 (in this case the solution is written in terms of cos ylTl and sin ylT; to keep A, and Ai real). Hence the boundaries (transition curves) that separate the stable from the unstable domains emanating from 6 = 1, E = 0 are given to a first approximation by 6, = f 1/2 or

+ O(3)

(6.2.111)

+ DlzA + (6, + &)AleiTo + CC + NST

(6.2.112)

6 = 1 f 4.

To determine a second approximation to u and the transition curves, we substitute for u, and u1 into (6.2.100) and obtain

+

D ~ u , u2 = -[2iD2A

The condition that must be satisfied for there to be no secular terms is

+

2iD2A

+ D12A + (6, + &)A

=0

(6.2.1 13)

Since A = A, iA,, (6.2.113) gives the following equations for A, and A i upon separation of real and imaginary parts 'A 2 2 +aA;=O (6.2.114) aT2 -2 *-a A . aA, = 0 (6.2.115) aT2

+

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

where

~ 1 +2

257

+

(6.2.116) a = 6, Replacing A, and A i by their expressions from (6.2.108) and (6.2.109) and equating the coefficients of exp (fylTl) to zero because they are functions of T,, we obtain da1 dT2 da2

2-+-

2Yl

3 - 61 2Yl

dT, - g - 6,

aa, = 0, Ma2

= 0,

These equations lead to da, _ --da,- - 0 dT2

- -4y1 d a l + aa, = 0 (6.2.117) 4 - 61 dT, 4y1 da2 + aa, = 0 (6.2.118) -4 - 61dTz

or a, = a constant and a2 = a constant (6.2.119)

dT,

(6.2.120) Therefore, to second approximation, the solution is given by (3.1.57) through (3.1.62) which were obtained by Whittaker's method. 6.2.5. THE VAN DER POL OSCILLATOR WITH DELAYED AMPLITUDE LIMITING The next example is a third-order problem as opposed to the previous examples which are of the second order. It is given in dimensionless quantities by de (6.2.121) 7-dZ

+z

= u2

(6.2.122) dr where v is voltage, t is time, e is excitation, coo is natural frequency, T is delay time, Z is output of the low-pass filter, and p is a measure of the servo-loop gain. This oscillator was first studied by Golay (1964) and then by Scott (1966) and Nayfeh (1967b, 1968). We consider here free oscillations (i.e., e = 0) and refer the reader to Nayfeh (1968) for the forced oscillation case. To determine a first approximation to the above equations, we assume that = uo(To, Ti) poi(Tm Ti) * . . (6.2.123)

+

+

(6.2.124) (6.2.125)

258

THE METHOD OF MULTIPLE SCALES

Substituting (6.2.123) through (6.2.125) into (6.2.121) and (6.2.122) and equating coefficients of like powers of p , we obtain

+ T D J ~+ Zo = uo2

(6.2.126)

D O 2 ~ , w ~ =~0 u ~ Do2u,

(6.2.127)

+ wo2u1 = 20,[( 1 - Z,)U, - D p , ] +

TDJ1

2 1

= -TDlzo

(6.2.128)

+ 2UoU1

(6.2.129)

The solution of (6.2.126) is

+

(6.2.130)

+ Z o = A K + A2e2iwoTo+ CC

(6.2.131)

uo = A(Tl)eiwoTo ,4(Tl)e-ieoTo

Substituting for u, into (6.2.127) gives .rD,Zo Its solution is Z , = B(T,)e-To"

Ape-2imoTo

+ 2 A R + 1 + 2ioO7 + 1 - 2iw07 ~2~2i~~DoTo

(6.2.132)

With u, and Z , known (6.2.128) becomes

- 6i00 where

+ 2iw07 + CC

A3e3ia0To

1

(6.2.133)

A2K (6.2.134) - DIA 1 2ioO7 Secular terms will be eliminated if Q = 0. Then the solution for u1 is Q =A-2A2K-

+

+-

+

u1 = 2AB7 1 - iOOT e[ioo--(llr)]To 3i A3e3ie0T0 CC (6.2.135) 1 - 2iwO7 40, 1 2iwor To solve the equation Q = 0, we let A = (1/2)aei* with real a and separate real and imaginary parts in (6.2.134), and obtain

+

4,

(6.2.136)

d 4- -&u$a2

(6.2.137)

d TI

where a, =

3 1

+ 8w,2r2 + 4W02T2'

a.=

- 1 +2WoT 40,272

(6.2.13 8)

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

259

The solutions of (6.2.136) and (6.2.137) are (6.2.139)

where a, is the initial amplitude and 4, is a constant. To determine B, we substitute for u,, Z,, and u1 into (6.2.129); that is

In order that Zl/Zo be bounded for all To, the coefficient of exp ( - T , / T ) must vanish; hence (4.2.142)

where we made use of A = (1/2)a exp i4. Substituting for u2 from (6.2.139) into (6.2.142) and solving the resulting equation, we obtain (6.2.143)

where

Therefore, to first approximation u = a cos (mot

z

=

+

a'

2J1

+ 4) + 001)

cos [ h o t

+ 4W,2T2

+ 2 4 - tan-'

(6.2.144)

+ * a z + O(p)

~WOT]

(6.2.145)

with a, 4,and B given, respectively, by (6.2.139), (6.2.140), and (6.2.143). 6.2.6. THE STABILITY OF THE TRIANGULAR POINTS I N THE ELLIPTIC RESTRICTED PROBLEM OF THREE BODIES The next two examples are fourth-order problems-one is linear and the second is nonlinear. Consider first the stability of the triangular points in the restricted problem of three bodies treated in Sections 3.1.4 and 3.1.5 using the Lindstedt-Poincart and Whittaker techniques. This problem was

260

THE METHOD OF MULTIPLE SCALES

treated first using the method of multiple scales by Alfriend and Rand (1969). The problem reduces mathematically to the stability of solutions of (3.1.63) through (3.1.65). In this section we use the method of multiple scales to determine the transition curves intersecting the p axis at po = (1 - 2d5/3)/2 and determine the behavior of x and y near these transition curves. We set cosf = cos Toand assume that 2

+

= xo(To,Tl) exl(To,Tl)

+- -

+ eYl(To9 Tl) + - p = po + epl + - - -

Y = YO(T0, Tl) where

To =f and

Hence

(6.2.146)

*

(6.2.147)

*

(6.2.148)

Tl = ef

d - Do+eDl+-..,

D,=-

(6.2.149)

a

(6.2.150) aT, Substituting (6.2.146) through (6.2.150) into (3.1.63) through (3.1.65) and equating coefficients of eO and e to zero, we obtain

df

Order e0

Order e

- 2Doy1-

Do2x1

+

Do2xo - 2Doyo - bgo = 0

(6.2.15 1)

Dozy,

(6.2.152)

+ 2D0x0 - a d o = 0

box1 = -220081~0

Do2Y1 2Dg1- a091 = -2DoD1yo

+ 2Dlyo + b1xo - bso cos To

(6.2.153)

- 20120 - b1yo - U

(6.2.154)

~ ~ C O ToS

where ui and b, are given by (3.1.71) and (3.1.72). The solution of (6.2.151) and (6.2.152) is

+

where

xo = A(Tl) cos &To B(T,) sin &To

(6.2.155)

yo = uB(Tl) cos *To- uA(T,) sin *To

(6.2.156)

-

+ &)-l = b, + & = 4(7 - 433)

a = (ao

(6.2.157)

The zeroth-order solution determines the right-hand sides of (6.2.153) and (6.2.154). Thus they become

+ Qlsin *To+ NST Dozyl + 2Dg1 - aoyl = P, cos &To+ Q2sin &To+ NST DO2q- 2D0yl - bgl = P, cos gTo

(6.2.158) (6.2.1 59)

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

6.2. where

261

P, = (20: - 1)B’ + (b, - $b,)A

(6.2.160)

Pz = (a - 2)A’ - a(b,

(6.2.161)

Qi

(6.2.162)

+ &z,)B = - ( 2 ~- 1)A’ + (b, + 4bo)B Qz = (a - 2)B‘ + a(b, - Bao)A

(6.2.163)

To determine a first approximation, we need not solve for x, and y1 but only insure that xl/zoand yl/yoare bounded for all To. This is the reason why we spelled out the terms that produce secular terms. To eliminate the secular terms, we can find the particular secular solution and then determine the condition for its vanishing. The resulting particular solution is of the form x = 0, (6.2.164) y = R1cos &To S, sin 4To or (6.2.165) x = R, cos &To S2sin *To y = 0,

+

+

Thus we can determine the conditions for the elimination of the secular terms by assuming a particular solution of the form (6.2.164) or (6.2.165). The results of using either form are the same. Substituting (6.2.164) into (6.2.158) and (6.2.159) and equating the coefficients of cos (T0/2) and sin (T0/2) on both sides, we obtain

Ri = Q i ,

s, = -P1

(6.2.166)

R, = -aP,,

S , = -aQ2

(6.2.167)

Elimination of R, and S, from (6.2.166) and (6.2.167) leads to the required conditions; that is PI = aQz, Ql = --UP, (6.2.168) Substituting for P,,P,,Q,, and Q, from (6.2.160) through (6.2.163) into (6.2.168) leads to the following two equations for A and B (1 - 4a (1

+ a2)A’ + [(l - a2)bl + $(bo- aauo)]B= 0

- 4a + a2)B’ -. [(l - a2)b, - $(b, - a2u,)]A = 0

(6.2.169) (6.2.170)

To solve these equations, we let A = a exp y,T,

B = b exp y,T,

(6.2.17 1)

and obtain

+ a2)yla+ [(l - a2)bl + +(bo- a2ao)]b= 0 - [(l - a2)b, - i(bo - a2uo)]u+ (1 - 4a + aX2)byl =0 (1 - 4a

(6.2.172) (6.2.173)

262

THE METHOD OF MULTIPLE SCALES

These equations are the same as (3.1.105) and (3.1.106) obtained using Whittaker's method. Hence y l and b/a are given by (3.1.107) and (3.1.108), while x and y are given by (3.1.109). This expansion was continued to second order by Alfriend and Rand (1969). 6.2.7. A SWINGING SPRING We consider next the nonlinear swinging spring discussed in Sections 5.5.3 and 5.7.5 and described by the Lagrangian (5.5.54). The equations of motion corresponding to (5.5.54) are 2

k ++ g(i - C O e)~ - ( I + x)P = o m

e+-

sine+l + X

2

I+%

ke = 0

(6.2.174) (6.2.175)

We seek an asymptotic solution of these equations for small but finite z and 8 of the form

+ +e ( t ) = .el(r0, rl)+ w T 0 , + . - .

x ( r ) = E X ~ ( Tl) T ~ , E ~ ~ Tl) ~ ( T .~. ,

(6.2.176) (6.2.177)

where T, = E n t and E is of the order of the amplitudes of oscillation. Substituting (6.2.176) and (6.2.177) into (6.2.174) and (6.2.175) and equating coefficients of like powers of E, we obtain Order

e

D:x,

+ w;x1

= 0,

+ wz28, = 0,

Do201 Order

k m

o12 =-

(6.2.178)

oz2=

(6.2.179)

1

€2

D : ~ , + w:x2 = -20,01x1

o;e,

- ige12+ i(ooe,)2

(6.2.180)

w2 2 + wz2e2= - 2 ~ , 0 ~ e +, 2 x,81 - - ( ~ o z l ) ( ~ , e l (6.2.181) ) 1 1

The solution of the first-order equations is

+ ,4(Tl)e-imiTn = B ( T l ) e i m ~+ T ~B(T,)e-'+'o

z1= A(Tl)eimiTn

(6.2.182)

8,

(6.2.183)

6.2.

APPLICATIONS OF THE DERNATIVE-EXPANSION METHOD

263

Then (6.2.180) and (6.2.181) become

+ +gBB + CC

Do2x,+ q 2 x 2= -2iqD1Aeim1To - +g Bze2i*,To

(6.2.184)

I f A and B are constants, the particular solutions of (6.2.184) and (6.2.185)

are

8, =

-

wz(w2 h ( W 1

-

0z(w2

4 ( 0 1

+ 201)

+ 2%)

ABei(Ol+O2)T0

- 2w1) ABei(-ol--02)To + cc - 2%)

(6.2.187)

which tend to co as w1 20,. Consequently, the expansions (6.2.176) and (6.2.177) break down when o1R!20,. To obtain an expansion valid when w1 2wz,we let ---f

w,

- 2w, = fU,

U

(6.2.188)

= O(1)

and let A and B be functions of TI rather than being constants. Moreover, using (6.2.188) we express exp (2iozTo)and exp [i(wl - 03T0]in (6.2.184) and (6.2.185) as exp (2iw2T0)= exp (iwlTo- iaT,) exp [i(wl - w2)To]= exp (iwzTo iaTl) to obtain

+

Dtx2

+ wI2x2= -(2iw,D,A + +gB2eiuT1)eiOlT0 + CC + N S T

028,

+ wz28, = - [2iwzDlB- wz(w*

- 2w1)

I

1

~ j ~ i u T eimo,To , +

Eliminating secular terms, we have

(6.2.189)

cc + NST (6.2.190)

2iw1D,A = -$gB2 exp (-iaT,) 2iwzDlB = wz(w2 - 20,) A B exp (iaT,) I

(6.2.191)

264 THE METHOD OF

MULTIPLE SCALES

Letting A = -(1/2)ia1 exp (iolB1) and B = -(1/2)ia2 exp (io&, with real a, and pi,and separating real and imaginary parts, we obtain (6.2.192) Liz=--

3EWz

41

(6.2.193)

a1a2cos y

(6.2.194)

A = - &3P a, sin y where

(6.2.195)

Y = 4 3 1 - 2wzBz + (01 - 2w2P

(6.2.196) If we let aI2 = co,a,*/o,k and a22= 2a2/rngl, (6.2.192) through (6.2.196) go over into (5.5.76) through (5.5.80) which were obtained using the method of averaging in conjunction with canonical variables. 6.2.8. A MODEL FOR WEAK NONLINEAR INSTABILITY We next consider the model problem us

- u,, - u = u3

16.2.197) cos kx, u ~ ( z0) , =0 for weak nonlinear instability of standing waves, which was discussed in Sections 2.1.2, 3.4.2, and 3.5.1. Away from k = 1, a uniformly valid solution for standing waves is (Section 3.4.2) U(X, 0) = E

where

u = c cos ot cos kx

+ O(E~)

9Ez

I+

32(k2 - 1)

(6.2.198)

...

It is clear that this expansion breaks down when k - 1 = O ( 2 ) . To determine an expansion valid near k = 1, we introduce the new variable 5' = kx in (6.2.197) to obtain Utt

- k 2 U c 5 - u = u3

u ( l , 0) = E cos E,

Moreover, we let k =1

+ C2k,

ut(t, 0) = 0

with k z = 0(1)

(6.2.199)

(6.2.200)

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

265

and assume that

~ ( 51 ,; E )

( 5To, , T l , T,;E )

=~

=

+ e2u2+ e3u3 + . . .

(6.2.201)

E U ~

where T , = E n t . Substituting (6.2.200) and (6.2.201) into (6.2.199) and equating coefficients of equal powers of B , we obtain Order

E

a2ul - - azUl aT:

u1 = cos 5,

Order

B,

u2

(6.2.202)

8% -0

at

a TO

azU3 at2

u3

at2

= 0,

Order e3 aT:

=0

aZu2 - - a z U z - u2 = -2aT:

aZu3

u1

at2

aZu,

aToaTl

8% - - au1 at TO T l

a

T,=O

a

T, = 0

azUl - - - 2-

= u13 + 2k2-

at2

= 0,

aT:

(6.2.203)

a2u2

aToaTl

- 2-

a% -- - -aul - -au2 at T,, = 0 a TO

aT,

aZu

a~~

aTo (6.2.204)

aTl

The solution of the first-order problem is u1 = u(T1,T2) cos

E,

u(0,O) = 1

(6.2.205)

Then (6.2.203) becomes

The solution of (6.2.206) will contain a term proportional to To making u2/ul unbounded as To-,co unless aa/aTl = 0 at TI = Tz= 0. Then U, =

b(T,, T2)cos E ,

b(0,O) = 0

(6.2.207)

With the first- and second-order solutions known, the equation for ug

266

THE METHOD OF MULTIPLE SCALES

becomes

aZu3 ar,2

azU3

2a3 - 2k2a - - cos 5

aq2

ap

+ &a3cos 3[

(6.2.208)

Secular terms will be eliminated if

a% + (2k2 - $a2)a = 0 aT,2

(6.2.209)

The'initial conditions for a were obtained above as

aa and - = 0 at T,=O (6.2.210) a TI To determine b(T,, T,) and the dependence of a on T,, we need to carry out the expansion to higher order. If we limit the analysis to O ( 4 , we can regard a as a function of TI within an error of O(E2f). A first integral for (6.2.209) and (6.2.210) is a = l

Since a(T,) is real, the right-hand side of (6.2.211) must be positive, hence p. Since a(0) = 1, a2 increases without bound if j3 < 1 and oscillates between 0 and 1 if j3 > 1. Therefore /? = 1 or k, = 3/8 separates the stable from the unstable regions. Hence the condition of neutral stability is

a2 must be outside the interval whose ends are 1 and

k =1

+ $e2

(6.2.212)

in agreement with (3.5.6). The solution for a is a Jacobian elliptic function.

6.2.9. A MODEL FOR WAVE-WAVE INTERACTION We again consider Bretherton's (1964) model equation

+

++

A t 42222 + 4Zz =C V (6.2.213) which was treated in Sections 5.8.1 and 5.8.2 using the variational approach. The linear problem admits the uniform traveling wave solution

4 = a cos ( k z - wt

+ j3)

(6.2.214)

where a, k, w , and j3 are constants and w and k satisfy the dispersion relationship w2 = k4 - k2 1 (6.2.215)

+

Harmonic resonance may occur whenever ( w , k) and (nw,nk) for some integer n 2 2 satisfy (6.2.215). This occurs at all k2 = 1/n for n 2 2. At

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

267

these wave numbers the fundamental and its nth harmonic have the same phase speed. Since the nonlinearity is cubic in our equation, the fundamental corresponding to k2 = 1/3 interacts to O(E)only with its third harmonic (k2 = 3). If the nonlinearity is ~ r j 5for ~ some integer m , then the fundamental ( k 2 = l/m) interacts to O(E)with its mth harmonic (k2= m).If we consider interactions to orders higher than E, harmonic resonances other than the third can occur even for a cubic nonlinearity. To determine a first-order expansion valid near k2 = 1/3, we let where

4 = +o(To, Ti,xo,xi) + ~+i(To,Ti, xo,x i ) +

* * *

(6.2.21 6)

X,,= cnx

T, = c n t ,

Substituting this expansion into (6.2.213) and equating coefficients of like powers of c , we obtain

a v o+ a'+, + a v o + 4, = 0

L(4,) =2aT, L(+J

ax: ax, a", - 4 8'4, - 2 - a240 = 40"- 2a ~ a, ~ , ax: ax, ax, ax, 2-

(6.2.217) (6.2.218)

The solution of (6.2.217) is taken to be

+, = A,(Tl, Xl)eiel + A,(Tl, Xl)eies + CC

where

(6.2.219)

8, = k,X, - wnTo = k:

- k:

w 3 M 3w1,

k,

0 , '

+,

+1

M

(6.2.220)

3k1

Note that is assumed to contain the two interacting harmonics. Had we assumed it to contain exp (&), we would have found it t o be invalid (see Sections 5.8.2 and 6.4.8). Substituting for 4, into (6.2.218), we obtain

+ 3(2A1X1+ A3X3)A,eie3+ A13e3ie1+ A,3e3ies +

3,4 ZA ei(ea+zel)

+

3A 1A 3 zei(ze3+e1) + 3A-1A 3 zei(ze3-81)

1

3

+

3 ~ 2~ 1

where w , = dw,/dk: group velocity.

ei(B~--2h) 3

+ cc

(6.2.221)

268

THE METHOD OF MULTIPLE SCALES

Because of the interaction between the two modes, terms that produce secular terms other than the usual exp (id,) occur in (6.2.221). To recognize these terms we consider the perfect resonance case in which d3 = 30, so that exp (id,) and exp (3i0,) produce secular terms. We immediately see that exp ( 3 i Q and exp [i(d, - 20,)] produce secular terms. For the near-resonance case, we indicate their secular behavior by expressing them in terms of exp (id,) and exp (id3). To do this we observe that

e3 - 30, = r = (k, - 3k,)x0 - (W3 - 301)~0 Although X o and To are 0(1), I' becomes slowly varying as k3 w3 -+ 30,, hence we express this slow variation by rewriting I' as

---f

3kl and

(6.2.222) With this function we express exp (3i0,) and exp [i(0, - 20,)] as exp (3i0,) = exp [i(0, - r)],

exp [i(0,

- 20,)l

= exp [i(0,

+ r)].

Eliminating the terms that produce secular terms on the right-hand side of (6.2.221), we have

(6.2.224) Letting A, = (1/2)u, exp (i/?,) with real a, and ?/, in (6.2.223) and (6.2.224) and separating real and imaginary parts, we have (6.2.225)

at% aT1

3 + ol-ax, = -801 (a,

2

+ 2a3 + alu3cos 6) 2

aa3 = -a,3sinS 1 + wjaT, ax, 803

(6.2.227)

& 3

at% + o3fa/?, 1 2 - = -( 6 ~ 1+ 3a: where

a

ax,

gW3

6 =r

+ p3 - 38,

(6.2.226)

+ a 1 3 u ~cos ' S)

(6.2.228) (6.2.229)

6.2.

APPLICATIONS OF THE DERIVATIVE-EXPANSION METHOD

269

Equations (6.2.225) through (6.2.228) are in full agreement with (5.8.24) through (5.8.27) obtained using the variational approach. 6.2.10. LIMITATIONS OF THE DERIVATIVEEXPANSION METHOD This method applies to wave-type problems only. It does not apply to unstable cases except when the instability is weak, such as the nonlinear stability problem discussed in Section 6.2.8. For k > 1, u is bounded and the expansion (6.2.198) is valid for times as large as e-2 if k is away from 1. This expansion is valid only for small times for k < 1 and away from 1. Near k = 1 , the instability is weak and a valid solution is given by (6.2.211) for times as large as e-l. In the case of hyperbolic equations, this method applies to dispersive waves only when the initial conditions can be represented by the superposition of a finite number of sinusoidal functions. For a linearly nondispersive wave problem such as (6.2.230) u ( 5 , 0) =f(z),

au -( 5 , O ) at

=0

(6.2.231)

this method does not provide a solution even iff(z) is a sinusoidal function such as cos 5. To see this we let u = uo(5, To, Ti)

where

+

To = t ,

E U ~ ( Z ,To, Z'1)

+

* * *

(6.2.232)

Tl = et

Substituting (6.2.232) into (6.2.230) and (6.2.231) and equating like powers of E, we obtain Order

EO

(6.2.233) uo(z, 0,O)= cos

Order

5,

E

au0 aT0

(5,

0,O) = 0

(6.2.234) (6.2.235)

u(5, 0,O) = 0,

au, (x,0,O)=

aT0

- -(x,0,O) aT1

(6.2.236)

270 THE METHOD OF MULTIPLE SCALES

The solution of (6.2.233) and (6.2.234) is uo = ,4(Tl)ei(z-To)+ A-( Tl)e-i(Z-TO)

where

(6.2.237)

A(0) = 3 Substituting this zeroth-order solution into (6.2.235) gives

+

2AA + ,42e2”’2-TO’

+ pe-2i(z-To)

(6.2.238) The right-hand side of this equation contains terms that produce secular To)]in addition to terms. They are the terms proportional to exp [f2i(z exp [fi(z - To)].In order that ul/uobe bounded for all To,all these terms must be eliminated. However, there is no way in which this can be done. In the previous examples such terms were proportional to exp [fi(z - To)], hence A was chosen to eliminate them. In this case, if a nontrivial solution is desired, A can be chosen in such a way as to eliminate the terms proportional to exp [fi(z - To)].The resulting expansion contains secular terms, hence it is not valid for large times. Expansions valid for large times and general initial conditions for nondispersive waves were obtained in Sections 3.2.4 and 3.2.5 using the method of strained coordinates.

-

6.3. The Two-Variable Expansion Procedure

6.3.1. THE DUFFING EQUATION Let us consider again the equation

+ €U3 = 0

+4

+

d2u dt2

We assume that where

- + w,”u u = uo(5, 17)

5 = €f,

5 , 17)

17 = (1

+

E2U2(E,

+

(6.3.1)

17)

+

* * *

+ .**)t

(6.3.2)

(6.3.3) Substituting (6.3.2) and (6.3.3) into (6.3.1) and equating like powers of E, we obtain t2W2

€304

(6.3.4) (6.3.5)

6.3.

THE TWO-VARIABLE EXPANSION PROCEDURE

The general solution of (6.3.4) is ~0

= A d E ) cos woq

Then (6.3.5) becomes azUl

-y a7

+ w;u1

+ B,(E) sin w,q

271

(6.3.7)

+ :(A: + A,B,2)] cos woq + [ 2 w , ~ ;- %(B: + A ~ B , )sin ] woq

= -[2w,BA

-

-MOB:) cos 30,q Secular terms will be eliminated if

+ $(B:

- ~4,213,)sin 30,q

+ $(A: + A , B t ) = 0 2w,A; - $(B: + A:Bo) = 0 2woBA

(6.3.8)

(6.3.9) (6.3.10)

Adding B, times (6.3.9) to A , times (6.3.10) gives

+

+

d -(A,' B:) = 0 or A; B: = u 2 = a constant dE Using (6.3.11), we can express (6.3.9) and (6.3.10) in the form

B; + w I A O

where

Ah - O J ~ B=, 0

= 0,

(6.3.11)

(6.3.12)

3 8WO

w1 = - a'.

Hence

A , = a cos (wlE 4-+),

B, = --a sin (wlE

+ $)

(6.3.13)

where 4 is a constant. With the secular terms eliminated, the solution of u1 becomes

-- I

(B: - ~ A Z B , )sin 3w0q (6.3.14)

320, Substituting for A , and Bo into u, and ul, we get u,, =

cos e

u1 = A,(@ cos 8

where

(6.3.15) u3 + B,(E) sin 0 + 320: -

0 = woq

cOS

+ olt + 4

313

(6.3.16) (6.3.17)

272

THE METHOD OF MULTIPLE SCALES

Replacing uo and u1 in (6.3.6) by their expressions from (6.3.15) through (6.3.17), we obtain

aZu2+ o;u2 -

=

+ 3a')

cos 8 - (-2w0A; -2w0w1B1

128~:

-

(2w0B1' +

pu2A,

- 200w1Al - awl2 - 2uo2wo2

+ $a2Bl)sin 8 + NST

(6.3.18)

Secular terms are eliminated if A, = B, = 0 and

or 02=--

15

a4

(6.3.19)

2560;

Therefore, to second approximation

u = u cos (of where

d dt

= - (w0q

or

€a3 + +) + 32w0 cos 3(wt + (6) + O(E') 2-

d + colt) = [(ao + dt

+E

~

+ .w. . ) t ] ~

(6.3.20)

~

(6.3.21)

This expansion is in full agreement with those obtained in Section 3.1.1 by using the Lindstedt-PoincarC methods, in Section 5.4.1 by using methods of averaging, and in Section 6.2.1 by using the derivative-expansion method. 6.3.2. THE VAN DER POL OSCILLATOR The second example we consider is the van der Pol oscillator

d2u du - + 11 = E ( l - u2)(6.3.22) dt2 dt discussed in Sections 5.4.2, 5.7.4, and 6.2.2. We assume that u possesses the following uniformly valid expansion (Cole and Kevorkian, 1963 ;Kevorkian, 1966a) u = Uo(5,V)

+ % ( 5 , 7) + c2u,(5, rl) +

'

..

(6.3.23)

where 5 and '7 are defined in (6.3.3). Substituting (6.3.3) and (6.3.23) into

~

6.3.

THE TWO-VARIABLE EXPANSION PROCEDURE

(6.3.22), and equating like powers of

E,

a2u,

we obtain

-+u,=o

(6.3.24)

ar12

- + u1= - 2 2

+ (1 - u i ) -au,

aZu, + u 2 = -2

aZu, - aZu, - - 20,-

a2u

ar2

at aq

a7

star

aZu

273

(6.3.25)

all

at2

ar12

+ (1 - u:)(- aU, + 2 au ,- 2u,u,- au, all

The general solution of (6.3.24) is uo = A o ( t ) cos q

at

arl

+ B,(5) sin q

(6.3.26)

(6.3.27)

Hence (6.3.25) becomes

aZu + u 1 = [-2BA

f (I - A,2

+

4

+ [ 2 4 , - (1 - A'

7 B . I cos q B ' ) ~ , ] sinq

+ $(A03 - 3A0B,2)sin 377 + #(~,3- 3A02B,-,) cos 317

(6.2.28)

Elimination of secular terms necessitates satisfying the following conditions (6.3.29)

4

(6.3.30) Subtracting B, times (6.3.29) from A, times (6.3.30) gives p'

where

p is

- p(l

- gp) = 0

(6.3.31)

the square of the amplitude of the zeroth-order solution; that is

+ Bo2

p = u2 = A,'

(6.3.3 2)

By separation of variables we integrate (6.3.31) to obtain

(5 4

a' =

1

+

- 1)c'

(6.3.3 3)

274 THE METHOD OF MULTIPLE

SCALES

where a, is the initial amplitude. Expressing A, and B, in terms of the phase 4 and the amplitude a, we obtain A,, = a cos

+,

B, = --a sin 4

(6.3.34)

Substituting into either (6.3.29) o r (6.3.30) and using (6.3.31), we find that

+' = 0

+ = $,

or

=a

Hence u, can be expressed as u, = a c o s

+

(q

(6.3.3 5)

constant

(6.3.36)

$0)

With (6.3.29) and (6.3.30) satisfied, the solution of (6.3.28) is

u1 = ME) cos ( q

+ do) + B1(t) sin ( q + 4,) - a3 sin 3(q + 4,) 32

(6.3.37)

Substituting for u, and u1 into (6.3.26) gives aZU + uz =

[-2B;

1

+ (1 - &a2)B,- a" + 2wza + (1 - &?)a' + us 128

(6.3.38)

To eliminate secular terms we require that 2B;

- (1 - &a2)Bl= 2w,a - a" + (1

a5 - $a2)ar+ 128

2A; - (1 - &?)A, = 0

(6.3.39) (6.3.40)

Using (6.3.31) and (6.3.32), we express the above two equations as

A'-

('a"' 1

(6.3.42)

--lA,=O

The solutions of these equations are B, = a(oz

+ &)l - b,a + &a In a - &-a3

(6.3.43) (6.3.44)

A, = ~ , a ~ e - ~

where a, and bl are constants. Since as t unbounded as E 03 unless

-+

-+

oz= -2. 16

co, 6

-+

co and a

--f

2, ul/uo is (6.3.45)

6.3.

THE TWO-VARIABLE EXPANSION PROCEDURE

275

Therefore, to second approximation u = (a

+ ~ a ~ a ~ e cos - ' ~[(I ) - &e2)t + +,I - J ( ~ a 3- &a In a + bla) sin [(I - +z))t + +,I \ a3 +sin 3[(1 32

where

a=

h 2 ) t

+

+0]

I+

2

,

O(E') (6.3.46) (6.3.47)

This expansion is in full agreement with (6.2.38) obtained using the derivative-expansion method if we identify a, with (5, 4 , 3 a l .

+

6.3.3. THE STABILITY OF THE TRIANGULAR POINTS IN THE ELLIPTIC RESTRICTED PROBLEM OF THREE BODIES Let us consider again the parametric resonance problem treated in Section 6.2.6 using the derivative-expansion method. The problem is described mathematically by (3.1.63) through (3.1.65). To determine a uniformly valid expansion near the transition curves using the two-variable expansion procedure, we need to use different time scales from those given by (6.3.3). The appropriate time scales are

E

= (e

+ wze2+

. .)t,

(6.3.48)

7 =r We assume that z and y possess expansions of the form 5

= zo(5,q)

*

+ ez,(E, q ) + e252(E,11) + - . .

Y = YO(5, 7) f ey1(5,7)

+ e2y2(E, 77) +

*

.

*

(6.3.49) (6.3.50)

The algebraic details of the solution are not presented here. The details for the first-order solution are the same as in Section 6.2.6, with E = Tl and 7 = To. The reader is referred to Alfriend and Rand (1969) for details of the second-order solution. Their results are in full agreement with those obtained in Section 3.1.5 using Whittaker's method. 6.3.4. LIMlTATlONS OF THIS TECHNIQUE The above examples demonstrate that by choosing the two variables appropriately the results of the two-variable expansion procedure agree with those of the derivative-expansion method. In some cases more than two variables are needed to obtain uniformly valid expansions such as in the case

276 THE

METHOD OF MULTIPLE SCALES

of satellite motion around the smaller primary in the restricted problem of three bodies (Eckstein, Shi, and Kevorkian, 1966a) and the motion of an artificial satellite having a period commensurable with the rotational period of its primary (Shi and Eckstein, 1968). In the case of hyperbolic equations, this technique as well as the derivativeexpansion method applies only to dispersive wave problems and it does not provide solutions for nondispersive wave problems such as that discussed in Section 6.2.10. 6.4

Generalized Method

6.4.1. A SECOND-ORDER EQUATION WITH VARIABLE COEFFICIENTS Let us consider the following special second-order problem (Nayfeh, 1964, 1965b) d2Y dy (6.4.1) lz(22 -I-1)2y = 0 dx2 dx

+

+

Y(0) = YO) = B (6.4.2) discussed in Sections 4.1.3 and 4.2.2 using the methods of matched and composite asymptotic expansions. As discussed in Section 4.1.3, the straightforward expansion possesses a nonuniformity at x = 0. The size of the region of nonuniformity is x = O(F). To treat this problem using the method of matched asymptotic expansions, an inner expansion valid when x = O(E) was introduced using the inner variable r] = x/e. This inner expansion was matched to the outer expansion and then a composite expansion was formed to give a uniformly valid expansion. To obtain a uniformly valid expansion using the generalized version of the method of multiple scales, we introduce the scales E = X

(6.4.3)

where g, is determined from the analysis. We require that go(0) = g,(O) = 0 so that go(z)--+ z as x -+ 0, hence 17 approaches the inner variable X/Q. Then the derivatives with respect to x are transformed according to (6.4.5)

6.4.

GENERALIZED METHOD

277

These variables transform (6.4.1) into

.)

a2y

+

+ g; + €g; + . .

€(&

:(

+2E - + g g ; + € g ; + . .

4- (2E

+ I)[ (& + gi + E

+ 2y = 0

(6.4.7)

where primes denote differentiation with respect to the argument. Note that we expressed the z variables appearing in (6.4.1) in terms of 5; namely, we expressed 22 1 as 2E 1. Moreover, we expressed gn and its derivatives in terms of 5. Now we assume that there exists a uniformly valid asymptotic representation of the solution of (6.4.7) in the form

+

+

(6.4.8)

where

ynY7l-1

(6.4.9)

for all f =: 5 and = ~ ( zE); where z is in the domain of interest. This last condition is the mathematical expression of the fact that the expansion (6.4.8) is regular in the whole domain of interest. Substituting (6.4.8) into (6.4.7) and equating like powers of e , we obtain the following equations for yo, yl, and yz (6.4.10)

a2y0 + (26 + 1) -2 a Y + 2y1 = 0 ++ 2g;-a5 a7 + 2g1' aEa7 at= a6 a2y0

(6.4.12)

278

THE METHOD OF MULTIPLE SCALES

Since g; f 0 because g,(x)

--f

z as x + 0, the solution of (6.4.10) is

+

yo = ~ ~ ( 6 ) Bo(E)e-y(*)q

where

?(El

2E

y %)g:

87

= -[(2E

+1

(6.4.14)

=-

Then (6.4.1 1) becomes

$!( +

(6.4.13)

g;

+ 1)A; + 2A0] (6.4.15)

The solution of this equation is

- (2 - ygg + g;g;y2)B, - Bog~y']~)e-yq(6.4.16) In order that yl/yo be bounded for all 7, the coefficients of 7, qeCYq,and qze-yq must vanish; that is (6.4.17) (2E + 1)Ah 2A0 = 0

+

+

Boy' = 0

2&(B0y)' - (2E 1)B; - (2 - 7s: The general solution of (6.4.17) is

+ &y2)BO - B&y'

(6.4.18) =0

(6.4.19)

a0 A, = (6.4.20) 25 1 where a, is a constant. If y, hence yo, is to satisfy two boundary conditions, Bo f 0, thus (6.4.18) yields y' = 0 (6.4.21)

+

Therefore y is a constant, which is taken to be unity without loss of generality. Then (6.4.14) gives (6.4.22) go = E2 E

+

since g,(O) = 0 to reflect the fact that the nonuniformity is at 5 (6.4.19) becomes B; - giB, = 0

= 0.

Equation

(6.4.23)

6.4.

GENERALIZED METHOD

279

whose solution is

B, = boeEl(!) (6.4.24) where b, is another constant of integration. Therefore, to first order

Since g,(E) disappears from the expansion irrespective of its value, we can set g, = 0 without loss of generality. The above analysis shows that y in (6.4.13) must be a constant which is taken to be unity without loss of generality. If y is not a constant, then since it multiplies 7 the derivatives with respect to E always create terms proportional to powers of 7 which make yl/yo unbounded as 7 co. Therefore, whenever such a situation arises, y is set equal to unity from the start. Moreover, we did not need to solve (6.4.15) to determine the conditions for yl/yo to be bounded for all 7. We could have investigated (6.4.15) and required the vanishing of all terms leading to particular solutions that make yl/younbounded. Such terms include all those that are proportional to the solutions of the homogeneous equation. Since e-ys and 1 are solutions of the homogeneous equation, we require the satisfaction of the conditions (6.4.17) and (6.4.19). To determine the second approximation, we let g, = 0, y = 1 , substitute for yo, yl, and go into (6.4.12), and obtain

-

r3+

& = -[(25

%)

+ 1)Ai + 2A1 + A:] + g;[Bi - Bogd]e-v

(6.4.26)

In order that yz/yobe bounded for all (2E

+ 1)A; + 2A, + A,” = 0

B; - Bog; = 0 Using (6.4.20) in solving (6.4.27), we obtain

(6.4.27) (6.4.28)

(6.4.29) where a, is a constant of integration. The condition (6.4.28) is satisfied if Thus

Bi = 0 and g;.= 0

B, = b, and g , = a constant where b, is a constant of integration and gz = 0 since g,(O) = 0.

(6.4.30) (6.4.31)

280 THE METHOD OF

MULTIPLE SCALES

Therefore, to second approximation, y is given by

+ ble-[(Z2+S)/€]

]+

O(e2) (6.4.32)

Imposing the boundary conditions y(0) = a and y(1) = ,5, we obtain = 38, bo = cc - 38, a, = -28/3, and b, = -16/3/3. Hence (6.4.32) becomes

If we expand e-(Z2/E) for small .‘/I., (6.4.33) agrees with (4.2.50) obtained using the method of composite expansions. Thus the method of multiple scales gives a single uniformly valid expansion in contrast with the method of matched asymptotic expansions which gives two expansions that must be matched. 6.4.2. A GENERAL SECOND-ORDER EQUATION WITH VARIABLE COEFFICIENTS As a second example, we consider (Cochran, 1962; Nayfeh, 1964, 1965b)

+ a(.)!/‘ + b(x)y

q”

(6.4.34)

= c(2)

y(0) = 0: and y(1) = ,5

(6.4.35)

where a(.) > 0 in [0, I]. The case in which ‘ ( 2 ) vanishes in the interior of [0, 13 is called a turning point problem. Turning point problems are discussed briefly in Section 6.4.4 and in detail in Sections 7.3.1 through 7.3.9. If c = 0, this example will be the same as that treated in Section 4.1.3 using the method of matched asymptotic expansions. Since a(.) > 0, the nonuniformity is at 1: = 0. In Section 4.1.3, we introduced an inner variable X / E to determine an expansion valid in the region x = O(E), which was matched to an outer expansion. To determine a uniformly valid first approximation using the method of multiple scales, we assume that there exists an asymptotic representation for y of the form

Y = Yo(& rl)

where

E = 5,

+ EYd6, r ) + -

1 with 17 = d. E

g(x)-+ x

*

.

as s + O

(6.4.36) (6.4.37)

6.4.

281 Substituting (6.4.36) and (6.4.37) into (6.4.34) and equating the coefficients of co and to zero, we obtain GENERALIZED METHOD

(6.4.38)

(6.4.39)

where we have expressed a(z), b(z), c(z), and g(z) in terms of E . Since g $ 0, the general solution of (6.4.38) is where

yo = A(5) y=-

+ B(S)e+l)V

45)

(6.4.40) (6.4.41)

g’

As discussed in the previous section, y must be a constant; otherwise the derivatives with respect to 6 would produce terms proportional to y ‘ q in (6.4.39), hence make ylIyo unbounded as q + 00. For a uniformly valid expansion, we require that y = 1 without loss of generality. Hence g =[u(t)dt

since g(z)+ z as 2-0

(6.4.42)

Substituting for yo into (6.4.39) gives (a72

+ 2 gf2= - [ a ~ ’ + b~ - C] + [g’B’ + (g” - l ~ ) B ] e - (6.4.43) ~

In order that yl/y, be bounded for all q , we require that aA’

g’B’

+ bA = c

+ (g” - b)B = 0

(6.4.44) (6.4.45)

The solutions of these equations are

(6.4.47)

where a, and 6, are constants of integration.

282

THE METHOD OF MULTIPLE SCALES

To first approximation, y is given by

bo

e S ~ [ b ( t ) / a ( t ) l d t e - ~ - l S o Z a ( t ) d+ t

4%)

O(r)

(6.4.48)

The limits of the integrals in (6.4.46)and (6.4.47)were chosen so that a, and b, could be expressed in a simple manner in terms of the boundary conditions (6.4.35).Thus a, = 8 and Expansion (6.4.48)is a composite expansion which agrees with the inner and outer expansions obtained in Section 4.1.3 in the inner and outer regions, respectively. If we specialize (6.4.48)for the case a(.) = 1

+ 22,

b(x) = 2,

c(.)

=0

(6.4.50)

discussed in the previous section, we obtain

(6.4.51)

where we have used a, = 8 and (6.4.49).This expansion is in full agreement with the first term in the expansion obtained in the previous section.

6.4.3. A LINEAR OSCILLATOR WITH A SLOWLY

VARYING RESTORING FORCE The two cases discussed above can be treated by using either the method of multiple scales or the method of matched asymptotic expansions. Let us consider next an example that cannot be treated by the latter method; namely y” + b(€s)y = 0 (6.4.52) where b ( a ) # 0 and E is a small parameter. To obtain an expansion uniformly valid for large x, we assume that there exists an asymptotic representation for y of the form Y = >[(OY q) EYl(6, q) * * * (6.4.53) where ,$=Ex, q = - +g ( t ) ... (6.4.54)

+

+

E

This form of q was chosen in order that the frequency of oscillation w = dq/dx = g‘(6) = O(1). Substituting (6.4.53) and (6.4.54) into (6.4.52)and

6.4. GENERALIZED METHOD 283

equating the coefficients of

eo

and e to zero, we obtain

The general solution of (6.4.55) is (6.4.57)

where

(6.4.58) As argued in the previous two sections, we set y = 1 to obtain an expansion in which yl/yois bounded for all 7. Hence g

=J6m dt

Substituting for yo into (6.4.56), and remembering that y

a2y1 + y1 g" (av' )

= -i(g"A

(6.4.59) =

1 , we obtain

+ 2g'A')e'" + i(g"B + 2g'B')e-'q

(6.4.60)

In order that yl/yo be bounded for all 17, we require the vanishing of the coefficients of exp (&ill) on the right-hand side of (6.4.60); that is

+ 2g'A' = 0 g"B + 2g'B' = 0 g"A

(6.4.61) (6.4.62)

The solutions of these equations are A = = 6, 0 B=-- 6 0 J g' J g' where E0 and 6, are constants of integration. If b(Ez) > 0, y is given by

where a, and b, are constants. If b(ez) < 0, y is given by

(6.4.63)

(6.4.64)

(6.4.65)

284 THE METHOD OF MULTIPLE SCALES

Expansions (6.4.64) and (6.4.65) are called the WKB approximation to the solution of (6.4.52) (see Section 7.1.3). These expansions are clearly not valid near a point where b(cx) vanishes. In fact, they tend to infinity as x approaches a zero of b(~x).The zeros of ~ ( E z ) are called turning points and are discussed in detail in Sections 7.3.1 through 7.3.9. An example of a turning point problem is treated in the next section using the method of multiple scales. If we change variables in (6.4.52) from x to 6, we find that (6.4.66)

which is a problem containing a large parameter 1.Thus the above-obtained approximation is applicable to this problem as well. 6.4.4. AN EXAMPLE WITH A TURNING POINT Let us consider the problem y"

+ P ( 1 - x)j-(x)y = 0

(6.4.67)

where 1 is a large positive number andf(x) is regular and positive. The WKB approximation tends to infinity as x -+ 1 as can be seen from (6.4.64) and (6.4.65) if we let b ( d ) = (1 - x ) f ( x ) and E = P. To determine an expansion valid everywhere using the method of multiple scales, we first determine the size of the nonuniformity. Thus we let 5 = (1 - x)A" with v > 0 in (6.4.67) and obtain (6.4.68)

As A -+ co, the following different limits exist depending on the value of

y=o

d2Y + f ( l ) t y dt2

if v < g }

= 0 if

Y

(6.4.69)

v =Q

The last limit is the appropriate one because its solution has an exponential behavior for 5 < 0 (i.e., x > 1) and an oscillatory behavior for 5 > 0 (i.e., x < 1). Thus it can be used to connect (6.4.64) and (6.4.65) as the turning point is crossed. Therefore we assume that there exists an asymptotic representation o f the

6.4.

GENERALIZED METHOD

285

solution of (6.4.67) of the form (Cochran, 1962; Nayfeh, 1964, 1965b; Fowkes, 1968, Part I)

Y

where

= yo((,

f = 2,

with

7)

+ A - 2 / 3 ~ 1 ( E , 7) + . . . + ...

(6.4.7 1)

h(5) > 0

(6.4.72)

7 =Pg(5)

g(5) = (1 - 2 ) / ? ( 5 ) ,

(6.4.70)

The functions of the independent variable z that appear in (6.4.67) are expressed in terms of 6, except 1 - 5 is replaced by ~ i l - ~ / ~ / hbecause (z) it reflects the nonuniformity. Therefore (6.4.67) becomes

Substituting (6.4.70) into (6.4.73) and equating the coefficients of 1213 to zero, we obtain

A41a

and

(6.4.74)

The general solution of (6.4.74) is

+

yo = A ( E ) ~ ~ / ~ ~ ~ / ~ [ Y (~E()~~) ~~ /1~i 2I ~ J - 1 / 3 [ ~ ( s )(6.4.76) ~3~2~

where Jill3 are Bessel’s functions of order f 1/3 and

(6.4.77)

In order that yl/yo be bounded for all q , y = 1 as discussed in Sections 6.4.1 and 6.4.2. Therefore g’(5)WE) =

-3rfcnl”2

(6.4.78)

where the negative sign in (6.4.77) was taken so that h ( s ) > 0. Multiplying both sides of (6.4.78) by (1 - E ) l I 2 , we obtain Since g(1) = 0

g”2g’ = -$[(1

- E )f ( E ) ] 1 / 2

g3l2= -S,’[(I

- t ) f ( t ) ] 1 / 2d t

(6.4.79)

286

THE METHOD OF MULTIPLE SCALES

With yo known and y = 1, (6.4.75) becomes

a [(2g‘A’ = -a7

+ g”A)Tf2J1/3(lj13/2)+ (2g’B‘ + g”B)$/2J-1/3(~3’2)](6.4.80)

In order that yl/yo be bounded for all 7,the right-hand side of (6.4.80) must vanish; that is

I

+ g“A = 0 2g’B’ + g”B = 0

2g‘A‘ Hence

(6.4.81)

(6.4.82)

where a and b are constants of integration. Thus, to first approximation

where a, and b, are constants. As x --t 1

where Z, and 6, are constants. Since J,(t) = tV

+ o(tv)

as t -+ 0

we conclude that (6.4.83b) is bounded as x --t 1. 6.4.5. THE DUFFING EQUATION WITH SLOWLY VARYING COEFFICIENTS Let us next consider the equation d2u dt2

-

where

+ a(E)u + B(E)u3 = 0 6 = Ef,

E

<< 1

(6.4.84)

6.4.

GENERALIZED METHOD

287

The asymptotic solutions of this equation were studied by Kuzmak (1959) using the method of multiple scales. If a and p are constants, the solution of (6.4.84) can be expressed in terms of Jacobi's elliptic functions; that is, in terms of u = A sn(Kt, v), A cn(Kr, v), A dn(Kt, v)

(6.4.85 )

Here Y is the modulus and K(v)is the complete elliptic integral. The differential equations satisfied by these functions are (6.4.86a) (6.4.86b) (6.4.86~) where

r

= Kt. Differentiating both sides of (6.4.86) yields

+ (1 + v2)sn - 2v2sn3= O

(6.4.87a)

+ (1 - 2v2)cn + 2v2cn3= O

(6.4.87 b)

d2sn d7'

d2cn dr2 dr2

+ (v'

- 2) dn

+ 2dn3 = 0

(6.4.87~)

Since these elliptic functions are tabulated for the case 0 < v < 1, we express the solution in terms of one of these tabulated functions. If cc and are slowly varying functions rather than constants, we expect the solution to depend on the slow time scale 5 = ct as well as on the fast time scale t. Moreover, to first approximation, the solution can be expressed as in (6.4.85) but with A = A ( c ) , K = K(Q,and v = v(E). Thus in the case of slowly varying coefficients, we let (6.4.88)

This form of solution differs from that of Kuzmak in that he assumed q = g'(5)t. Substituting (6.4.88) into (6.4.84) and equating coefficients of

288

THE METHOD OF MULTIPLE SCALES

like powers of

E,

we obtain

We write the solution of (6.4.89) in terms of one of the elliptic functions in (6.4.85), say sn; that is uo = A(E)sn[q, V ( 8 l (6.4.91) Hence uo/A should satisfy (6.4.87a) with q = 7 ; that is (6.4.92)

In order that (6.4.89) and (6.4.92) be identical [1

+ vz(t)Ig'Yt) = 48

(6.4.93)

(6.4.94) These are two relationships among A(E), v(E), andg(5). A third relationship is determined from the condition that uJu0 be bounded for all q in order that (6.4.88) be a uniformly valid asymptotic expansion. Differentiating (6.4.89) with respect to q leads to the homogeneous part of (6.4.90). Hence i?uo/aq is a solution of the homogeneous part of (6.4.90). In order that ul/uobe bounded for all q, the inhomogeneous part in (6.4.90) must be orthogonal to the solution of the homogeneous part; that is 2v2(E)gf2(E)= -P(E)A'(E)

(6.4.95) where sn(q,, v) = 0 and T is the period of sn(q, v ) with respect to 11. This condition is a generalization of the elimination of terms that produce secular terms. Equation (6.4.95) can be rewritten in the form

or

(6.4.96) Since uo = A sn(n, v), ql can be taken to be zero and T = 4K where K is the following complete elliptic integral of the second kind (6.4.97)

6.4. GENERALIZED METHOD

289

Substituting for uo from (6.4.91) into (6.4.96), we have where c is a constant, and (6.4.99)

with

5 = sn(q, Y).

Using (6.4.86a), we express L as J(l

or L=

(1

- <"(1 - Y y 2 ) d5

+ 2 ) E ( v ) - (1 - VZ)K(V)

(6.4.100) 3 a*' where E ( Y )is the following complete elliptic integral of the first kind

=lo1,,/-

1 - v2x2d x (6.4.101) 1 - x2 The conditions (6.4.93), (6.4.94), and (6.4.98) constitute three relationships for the determination of A([), v(E), and g'(5). Solving for g' from (6.4.93) gives E(Y)

(6.4.102)

Eliminating g' from (6.4.93) and (6.4.94) and solving for A , we obtain (6.4.103)

Squaring (6.4.98) and substituting for g' and A from (6.4.102) and (6.4.103), we obtain (6.4.104)

With the aid of these last-mentioned relationships, we can compute v(6) from (6.4.104), then g' and A from (6.4.102) and (6.4.103). The graph for the solution of (6.4.104) was given by Kuzmak and is shown in Figure 6-2. Three different cases arise depending on the signs of a(5) and @ ( f ) :

> 0 , @ ( E ) < 0. In this case p > 0 and (6.4.94) shows that > 0. Hence the curve that determines y lies in the first quadrant. The solution for y exists if 0 < p < 219. At 5, such that p(5,) = 219, the (1) a(5) y = v2(5)

--".P , - -2 --4

- -6 - -8 -40

--I4 --I2

-46

--I8

290

6.4. GENERALIZED METHOD 291 asymptotic solution ceases to be oscillatory. If p > 2/9, or if a(5) < 0 and < 0, (6.4.89) has no periodic solutions. (2) a(5) > 0, B(5) > 0. In this case p > 0 and (6.4.94) shows that y < 0. Hence the curve that determines y lies in the fourth quadrant. The solution for y exists for 0 < p < 03. (3) a(6) < 0, B(E) > 0. In this case p < 0 and (6.4.94) shows that y < 0. Hence the curve that determines y lies in the third quadrant. The solution for y exists if -a < p < -419. Since the elliptic functions and integrals are usually tabulated for real Y such that 0 < Y < 1, alternative asymptotic solutions in terms of cn(q, Y ) and dn(ljr, Y ) are preferred in cases (2) and (3).

@(E)

6.4.6. REENTRY DYNAMICS The motion of a reentry rolling body with variable spin under the influence of nonlinear aerodynamic forces and slight center of gravity offset and aerodynamic asymmetries is governed by (Nayfeh and Saric, 1972a)

+ ic2x1E + ixz 151' 6

+=P p = e2v0

+ E Y ~ G+

E ~ Y ~ P ,

(6.4.105) (6.4.106)

G = Imaginary {Ee-a+}

(6.4.107)

+

where 6 = fl icc, [ E l is the sine of the total angle of attack, p is the roll rate, EKis the amplitude of the excitation due to aerodynamic asymmetry, and E is a small but finite quantity of the order of the sine of the initial total angle of attack. Here wo, K, y , pi,xi, and vi are slowly varying functions of time, and I and I, are constants. In the absence of damping and nonlinear terms (i.e., y = pi = xi = 0), the solution of (6.4.105) for constant p, K, and wo is

5 =~

~+ Azeimort ~

+

i

(01

~

EK

- P)(P

t

-

e'(H+90)

(6.4.108)

02)

where A, and A, are complex constants and w1.2

=pzz f

21

J(+F

+ 002

(6.4.109)

The frequencies w1and w, are called the nutation and precession frequencies.

292

THE METHOD OF MULTIPLE SCALES

For statically stable bodies (ie., coo2> 0) and positivep, w1 is positive while negative. Two cases arise depending on whetherp is near w1 or not. The first case is called roll resonance, and the forced response tends to infinity as p + wl.Before p approaches wl, the damping as well as the nonlinear aerodynamic forces significantly modify the response. The roll resonance case is discussed in this section for the case K = C2k,and we refer the reader to Nayfeh and Saric (1972a) for the nonresonant case. To determine an approximate solution to (6.4.105) through (6.4.107) when p M o1using the generalized version of the method of multiple scales, we make use of the fact that actual fight test data and six-degree-of-freedom numerical calculations show that there are at least four time scales: a slow time scale T2 = cat characterizing K, coo, y , vi, xi, and pi;and three fast scales characterizing the nutation, precession, and forced components of the angle of attack. Thus we assume expansions of the form o2is

where wl(Tz) is the nutation frequency and w2(T2) is the precession frequency. In terms of these variables, the time derivatives are transformed according to da a a -o1-+w2-++-+2-

dt

%I

aqz

a+

a aT2

(6.4.113)

(6.4.1 14)

Substituting (6.4.1 10) through (6.4.1 14) into (6.4.105) and (6.4.107) and

6.4. equating coefficients of like powers of

6,

GENERALIZED METHOD

293

we have

LO,) = 0

(6.4.1 15)

+ i(x1 f xz 1S1I2)t1

(6.4.116)

+ v2po+ v1 Imaginary (tlePi') (6.4.1 17)

where

+ w:

a

- i-(

(6.4.118)

I The solution of (6.4.115) is

t1= A1(T2)e"1 + A2(~2)eiq2

(6.4.1 19)

Then (6.4.1 17) becomes

+ d a l sin (ql - 4 + 8,) + a2 sin (qz - 4 + 8,)l

(6.4.120)

where A,, = a, exp (ie,,) with real a,, and 8,. Since po w wl, v1 - 4 is a slowly varying function of time, and we consider it a function of T2.Now the solution of (6.4.120) contains terms that tend to infinity as ql, q2, or 4 -+ co (i.e., t + co), thereby invalidating our expansion for long times unless dP0

- = yo d T2

+ v2p0+ vial sin (ql - 4 + 0,)

(6.4.121)

Then pz becomes P2

a2v = ___ cos(q2 Po - -2

+ 6, - 4)

(6.4.122)

294 THE METHOD OF

MULTIPLE SCALES

With El known (6.4.1 16) becomes

+ Qzeivz + (iq,uZ + ixz + y)A~&ea(2''-q~a) + (iwZp2+ ix2 + y ) ~ l A ~ e i ( 2 v 2 - - 4 1 )

L(E3)= Qleivl where

(6.4.123)

Q1 = - i ( q - w z ) dAi - - iw;A, dT2

+

+ xl)+ [-2iy + 2x2 + (wl + dpZIal2 + xz + %Pz)a,2)A2

i((W2pl

+ (-iY

(6.4.125)

Secular terms in (6.4.123) will be eliminated if Q , = Qz= 0. Letting A , = a , exp (ion) with real a, and 0 , in (6.4.124) and (6.4.125) with Ql = Q , = 0 and separating real and imaginary parts, we obtain

(6.4.127)

(6.4.129) where

r = +-ql

-08,

+

(6.4.130)

(6.4.131)

(6.4.132)

6.4.

GENERALIZED METHOD

295

Combining (6.4.128) and (6.4.130) and introducing the detuning parameter a defined by po = w1

we have

+

6%

6.4.7. THE EARTH-MOON-SPACESHIP PROBLEM The next example is the one-dimensional earth-moon-spaceship problem discussed in Sections 2.4.2, 3.2.2, and 4.1.7 and given by

, t(0) = 0 (6.4.134) 1-x The straightforward expansion for r , for small p , in terms of x is singular at x = 1, and the region of nonuniformity is 1 x = O ( p ) . Thus to determine an expansion valid for all z using the method of multiple scales, we introduce the two variables (Nayfeh, 1964, 1965a)

-

7"-

E=x,

- (-

With these variables (6.4.134) becomes J2(- at

+IG)

at

at

1-x P

-5p

+

T2 7

(6.4.135)

where all functions of x are expressed in terms of E except the term 1 - x, the source of the nonuniformity, which is expressed in terms of q as pq. Now we assume that t possesses the following uniformly valid expansion t .= t n ( f , 11)

+ ~ f i (11)t +, P ' G ( ~ ,7) + .

* *

(6.4.136)

Substituting (6.4.136) into (6.4.135) and equating coefficients of like powers of p , we obtain (6.4.137) (6.4.138) (6.4.139)

The general solution of (6.4.137) is (6.4.140)

296

THE METHOD OF MULTIPLE SCALES

where A is determined from the condition that f l / t obe bounded for all 7. The solution of (6.4.138) becomes

-J2

+J

tl = -A'(t)-q

j - PI2sinh-'&

m

As q-+ co, (6.4.141) becomes -fitl

=? /,[

- A'(6)Iq

+ E ( t ) (6.4.141)

i

+ it,/? - t3"In 2 - + E ( t ) + O(q-') (6.4.142)

Thus t , contains two terms which make t , singular as q -+co; a term proportional to q and a term proportional to In (11). The first term can be eliminated if ~ ' ( 8=) & or A = $5312 a (6.4.143)

+

where Q is an arbitrary constant. As for the second term, In (9) is slowly varying with z and p although q is fast varying. Thus it should be expressed in terms of E ; that is

Then tl is singular as 8

-

1, and it will be bounded as 5

B ( t ) = $E3'/" In (1 - t )

+ C(6)

-

1 if (6.4.144)

The function C(5) is determined by requiring that t,/t, be bounded as q + 00. Substituting the above solutions into (6.4.139), we obtain -,

As q

-+

-I

co, (6.4.145) becomes

+---l3I2 21-5

C'(t) + O(q-')

as 3-f

00

(6.4.146)

6.4. GENERALIZED METHOD 297

Here again In

J.1 is expressed in terms of 5 . Consequently, (6.4.146) becomes as q - + w

In order that t J t , be bounded as 1;1+

00

or

where c is a constant of integration. Expressing to and t , in terms of x and using the initial condition t(x = 0) = 0, we obtain a = c = 0. Hence

$1

+ $x3/' + JZ - 4 In -k + O(pz) 1 - JZ

(6.4.148)

We next discuss an alternative method (Nayfeh, 1965a) of determining A(E) and B(6). Since

is assumed to be uniformly valid for all x, it must reduce to the straightforward expansion (see Exercise 2.12)

JZ t = gx3'/"+ p

3x3I2

+ JZ

")

- 4 In - + O(p2) (6.4.150) 1 - Jx

away from x = 1 . We can use this condition to determine the functions A(E) and B(6) rather than use the condition that te/t,-l be bounded for all 5 and

298

THE METHOD OF MULTIPLE SCALES

q. Expressing (6.4.149) in terms of x , and expanding for small p , we obtain

1+

+4

~ In ~4(1-’ ~ - B(x)

O(p2) (6.4.151)

11% In order that the first terms in (6.4.150) and (6.4.151) be the same A ( z ) = 8x3‘2

(6.4.1 52)

Then the second terms are the same if

Substituting these expressions for A and B into (6.4.149) and expressing the result in terms of x , we obtain (6.4.148) exactly. 6.4.8. A MODEL FOR DISPERSIVE WAVES We again consider Bretherton’s (1964) model equation

4t*+

L

2

+

$22

+ 4 = 43

(6.4.154)

The linearized equation admits the uniform wave train solution

4 = a cos 8

8 = kx

- wt,

w2 = k4 - k2

(6.4.155)

+1

To determine a wave train slowly varying with position and time, we follow Nayfeh and Hassan (1971) by assuming that

4 = +d8, Xl, TI) + .41(8,

where

8 = e-’((X1, Tl),

Xl, Tl)

X , = EX,

+

* * *

(6.4.156)

Tl = ~f (6.4.157)

In terms of the new variables 8, X I , and T,, the time and space derivatives become a 2 -- 0 2 - a2 - 2EW

at2

ad2

a2 aw a+ -E -

aeaT,

aT,ae

€2-

a2

aT:

ak a - + 2-za 2 aeax, + ax,ae ax, a4 - = k4 - + 4€k3___ + 6 e k 2 -a-k a3 + . . . a84 a83 ax, ax, a83 a2 a2 -- k2-+2ck-

ax2

ao2

a4

a4

ax4

a2

E -

6.4.

GENERALlZED METHOD

299

Substituting (6.4.156) into (6.4.154) and equating coefficients of like powers of c, we have

~ ( 4 =~ (w' ) 3

+ k2)

*'+ ae2

ae4

(E-g)$!?a34 ax, ae3 a2+0

a2+0

L(+l)=+O +2a--2k--4k3ae aTl

+

+ +' = 0

k4 ?h0

ae ax,

(6.4.158) a93 ax,

6k2--oak

(6.4.159)

The solution of (6.4.158) is taken to be

LO'

Substituting for where aA Q =2 i~ a TI

+ A(X,, T,)e-" = k4 - k' + 1

(6.4.160)

+o = A(X,, Q e "

where

into (6.4.159), we have

L(+,)= Q ( X , , T,)e"

+ A3e3" + CC

(6.4.161)

aA ak + 2ik(2k2 - 1)+ i am -A + i(6k2 - 1) A + 3A26

ax,

aq

8x1

(6.4.162) The condition that must be satisfied for there to be no secular terms is Q = 0. To simplify this condition, we note that LOW'

= 2k3 - k

(6.4.163)

where w' = dw/dk, the group velocity. Differentiating (6.4.163) with respect to XI, we have (6.4.164)

If 5 is twice continuously differentiable, w and k satisfy the compatabllity relationship am ak +-=O or

a ~ , ax,

(6.4.165)

Hence

300

THE METHOD OF MULTIPLE SCALES

so that (6.4.164) can be rewritten as (6.4.166)

With (6.4.163) and (6.4.166) the condition Q = 0 can be simplified to aA

2-

aT1

i-2w'-

aA

8x1

+ w " -akA

3x1

3i

= -A2A 0

(6.4.167)

Letting A = (112)~exp (ip) in (6.4.167) and separating real and imaginary parts, we have

a (w'a') -+ aTl ax,

=0

(6.4.168) (6.4.169)

The solution obtained in this section using the method of multiple scales is a different representation of the solution obtained in Section 5.8.1 by averaging the Lagrangian. In fact, the equations governing the amplitude and the wave number have exactly the same form. However, in Section 5.8.1 there is no phase, but the dispersion relationship (5.8.9) is amplitudedependent; in this section the dispersion relationship is independent of amplitude but the solution describes the phase variation. To show the equivalence of these representations, we expand 8 of Section 5.8.1 in the form

e = eo - E~

so that

0

= wo

+ E- at

(6.4.170)

Substituting (6.4.170) into (5.8.9) and equating coefficients of like powers of E, we obtain W : = k t - kO2 1

a,!7

+

-+w,-=at

ax

The last equation is the same as (6.4.169)

3a2 8 ~ 0

6.4. GENERALIZED METHOD 301 6.4.9. THE NONLINEAR KLEIN-GORDON EQUATION The last example considered in this chapter is the equation

+

utf - u,, V’(u) = 0 (6.4.171) which was treated in Section 5.8.3 using Whitham’s method of averaging the Lagrangian. Our analysis follows that of Luke (1966). We assume that u possesses a uniformly valid expansion of the form

u(z, t ) = uo(0, Xl, Tl)

+ .u,(e, x,,Tl) i.

(6.5.172)

+ V’(u,)

(6.4.173)

* *

where 0, XI,and T , are defined in (6.4.157). Substituting (6.4.172) into (6.4.171), using the expressions for the derivatives from the previous section, and equating coefficients of like powers of E , we obtain aZu

(w2 -

ae2

aZu + V ” ( U ~ ) U , ae

=0

(wZ- k2)<

= 2k-

a2u,

ae ax,

+

2w

am auo a2uo + ak auo + a e m , ax, ae aT, ae

(6.4.174)

Equation (6.4.173) can be integrated once to give (6.4.175)

whose solution is

where E ( X l , T,) and v(X1, T I )are unknown functions to be determined by examining (6.4.174). Inverting (6.4.176), we find that U O ( ~XI, , TI)

=f(e

+ v , E , w 2- k2)

(6.4.177)

We assume thatfis periodic with a constant period which can be normalized to unity so that dt (6.4.178) ( 2 [ E - V(5)])”2 = This provides one relationship among o,k, and E which is a dispersion relationship. Now the particular solution of (6.4.174) contains terms that make uI/uo unbounded as 0 + 00 unless the right-hand side of (6.4.174) is orthogonal to the solution of the adjoint homogeneous equation. This condition is

302

THE METHOD OF MULTIPLE SCALES

sometimes referred to as the solvability condition, and it is a generalization of the condition of elimination of secular terms that has been used extensively in this book. Since (6.4.174) is self-adjoint, the solvability condition demands that its right-hand side be orthogonal to the solution of the homogeneous equation which can easily be shown to be u1 = au,/ae. Thus the solvability condition requires that

which can be rewritten as

Changing the variable of integration from 0 to u, and substituting for au,/aO from (6.4.175), we can rewrite this condition as

This provides a second relationship among w , k, and E. The third relationship is the compatability equation (6.4.165). The results of this section are in agreement with those obtained in Section 5.8.3 using the variational approach.

6.4.10. ADVANTAGES AND LLMLTATIONS OF THE GENERALIZED METHOD This method can certainly be applied to all problems that can be treated by either the derivative-expansion method or the two-variable expansion procedure. Moreover, it can also be applied to cases in which both of these two methods fail, such as problems requiring nonlinear scales (e.g., an oscillator with slowly varying coefficients) or problems having sharp changes (e.g., the earth-moon-spaceship problem). However, the algebra is more involved, and the derivative-expansionmethod and the two-variable expansion procedure are preferable for nonlinear oscillation problems with constant coefficients. The method of multiple scales can be used to obtain uniformly valid expansions for problems that can be treated using the method of strained coordinates. Moreover, the method of multiple scales can be applied to cases in which the method of strained coordinates cannot be applied, such as

EXERCISES

303

problems involving damping or sharp changes. In cases in which the method of strained coordinates applies, it may have an advantage because of the implicitness of the solution. For nondispersive hyperbolic equations expansions in terms of exact characteristics are desirable. However, the method of multiple scales can be viewed as a generalization of the method of strained coordinates if the scales are given implicitly rather than explicitly in terms of the original variables. The examples considered in this chapter demonstrate that the method of multiple scales can be applied to problems that can be treated by the method of matched asymptotic expansions, such as the earth-moon-spaceship problem, as well as to problems that cannot be treated by the latter method, such as nonlinear oscillations. The method of multiple scales yields a single uniformly valid expansion, in contrast with the method of matched asymptotic expansions which yields two expansions that must be matched. Although an ordinary differential equation is transformed into a partial differential equation by the method of multiple scales, the first approximation is not more difficult to solve than the first inner equation. However, the equations for determination of the different scales may be difficult to solve (Mahony, 1962). Moreover, this method has not been applied yet to partial differential equations in which the first term in the expansion is nonlinear, such as viscous flow past a body, or elliptic partial differential equations with inhomogeneous boundary perturbations such as flow past a thin airfoil. The method of multiple scales is applicable to problems that can be treated by the method of averaging, the method of Krylov-Bogoliubov-Mitropolski, and the Lie transforms as well as to cases that cannot be treated by these methods. If the system is represented by a Hamiltonian, the Lie transforms have an advantage because the higher approximations can be obtained recursively. However, the method of multiple scales can be applied in conjunction with the Lie transforms directly on the Hamiltonian. Exercises 6.1. Determine a first-order uniform expansion for

ii

+ wozu = €f(u, li)

then specializeyour results to the cases:f=- ri + bu3, bu3 6.2. Determine second-order uniform expansions for

+ +

ii (6 E cos 2r)u = 0 for 6 near 0 and 4. 6.3. Determine second-order uniform expansions for ii

+ (6 +

C O S ~I ) U

=0

+ (1 - u2)u, and - 1li)U.

304 THE METHOD OF MULTIPLE SCALES 6.4. Determine ht-order uniform expansions for

ii

+ (6 +

E

cos 2t)u = E f ( U , li)

and specialize your results to f = Pu3, - (lit u, and (1 6.5. Consider the problem ii

+

+ (1

~ 0 % =E[U~

- u2)li.

- u2)li] + KCOSw t

Determine first-order uniform expansions for (a) K = O(1) and w away from wo, 3w0, and w0/3 (b) K = 0(1) and w = 3w0 (c) K = 0(1) and w = w0/3 (d) K = O ( E )and w fis wo. 6.6. Consider the problem

ii

d + wo2u = 2~ [(l dt Ti

+z

Z)U]

+ 2K cos wt

= u2

Determine first-order uniform expansions for the cases enumerated in Exercise 6.5. 6.7. The problem of takeoff of a satellite from a circular orbit with a small thrust can be reduced to U”

€V2 + u - 02 = - u3 (suf + cu)

uf=--

cm3 u3

40) = 1,

~ ‘ ( 0= ) 0, and 140) = 1

where primes denote differentiationwith respect to 8, and E, s, and c are constants. For small c show that (Nayfeh, 1966) v =f

+ 3~cf-3Inf+ O ( 2 )

I

+ cfd2(6Inf - 1) + o(2) wheref = (1 - 4 ~ ~ 0 )Is ~ this ’ ~ . expansion valid for all O ? 6.8. Consider the problem defined by (6.4.105) and (6.4.106) with constant w,,, p , Y . P i and x i . (a) Determine a first-order uniform expansion for the case K = O(1) and p away from wl. (b) Show that this expansion is not valid whenp fis 0 or 20, - w2 and determine first-order uniform expansions for these cases (Nayfeh and Saric,

EXERCISES

305

1971b). (c) Determine a first-order uniform expansion for K = O ( 2 )and p w w1 using the methods of multiple scales (Nayfeh and Saric, 1971b) and the method of averaging (Clare, 1971); then compare both results. 6.9. Use the method of multiple scales (MMS)to determine second-order uniform expansions for the problems

+

(a) CY“ F y’ y =0 (b) ry” F y’ = 22 (c) ry” f (22 + 1)y‘ = 1 subject to the boundary conditions

Y(0) = a ,

Y(1)

(a) ry“ - a(2)y’ b(z)y = 0, (b) EY” F y’ + y2 = 0

a(r)

=

B

6.10. Determine first-order uniform expansions for the problems

+

(C)

cy”

>0

yy’ - y = 0

(22 + 1)y’ + y2 = 0 (d) EY” (e> ry” 7 y’ yn = 0 ; n is a positive integer

+

subject to the boundary conditions

Y(1)

Y(0) = a,

=B

6.11. Use the MMS to determine a first-order uniform expansion for the problem ry“

+ a(2)y’ = 1

d o ) = a,

Y(1) = B

if a(z) has a simple zero at p in [O, I]. 6.12. Use the MMS to determine a first-order uniform expansion for y”

+ P(1 - s)y(z)y

=0

where n is a positive integer,f(z) > 0 and 1 >> 1. 6.13. Determine first-order uniform expansions for (6.4.84) of the form

(a)

u = A ( S ) c n [ q , 4 8 1 and

(b) u

= A(E)dntrl, 4~31.

6.14. Consider the problem

ii where

4

+ oo2(Et)u = €us + Kcos 4

= ~ ( r f )Determine .

first-order uniform expansions for the cases

(a) K = O(1) and w away from (b) K = O(1) and o m 30, (c) K = O(1) ahd w w 4 3 (d) K = O(e) and o w w,

oo,3w,,

and w0/3

306

THE METHOD OF MULTIPLE SCALES

6.15. Consider the problem defined by (6.4.105) through (6.4.107) with variable coefficients. Determine a first-order uniform expansion when K = 0(1) and p is away from wl. Show that this expansion is not valid whenp M 0 or 2 9 - w,, and determine uniform expansions for these cases (Nayfeh and Saric, 1972a). 6.16. Solve Exercise 5.14 using the MMS. 6.17. Determine a first-order uniform expansion for small amplitudes for

+ 2x + 3x2 + 2y2 = 0 y + x + 26y + 4xy = 0

x -y when 6 w 1. 6.18. Consider the problem

Determine a first-order uniform expansion when

6.19. Consider the equation

Use the MMS to determine first-order expansions for traveling waves if (a) the amplitude and phase vary slowly with position and time and (b) the wave number, frequency, amplitude, and phase vary slowly with position and time. 6.20. Nonlinear transverse oscillations in a homogeneous free-free beam with a nonlinear moment-curvature relationship are given by

where c and E are constants. Determine first-order uniform expansions for small E using the MMS for the cases enumerated in Exercise 6.19. 6.21. Use the MMS to determine second-order uniform traveling wave solutions for the problem defined in Exercise 5.15 and the cases of Exercise 6.19. 6.22. Consider again Bretherton's model equation

but with a general nonlinear function$ Use the MMS to determine first-order uniformly valid expansions for small c for the case of nth-harmonic resonance. Specialize your results for (a) second-harmonic resonance and (b) third-harmonic resonance.

Perturbation Methods ALI HASAN NAYFEH Copyright Q 2004 WILEY-VCH Valag GmbH & Co. KGaA

CHAPTER 7

Asymptotic Solutions of Linear Equations

In this chapter we describe new techniques and use some of the techniques described in the previous chapters to obtain asymptotic solutions of linear ordinary and partial differential equations. Interest here is in differential equations with variable coefficients. The approach is to utilize a small or large parameter and develop parameter perturbations or to utilize a small or large coordinate and develop coordinate perturbations. If the point at infinity is an ordinary point or a regular singular point of an ordinary system of differential equations, convergent solutions can be obtained in inverse powers of the coordinate, except in some special cases of the regular singular problem when one of the solutions may involve the logarithm of the coordinate. Our interest in this chapter is in the irregular singular case, when the solution must be represented by an asymptotic expansion. In the parameter perturbation case, the parameter can be small or large; the first case includes the case of slowly varying coefficients. The expansions in these cases are obtained using the Liouville-Green (WKB) transformation and its generalizations. The resulting expansions are valid everywhere except at certain points called turning or transition points. Expansions valid everywhere including the turning points are obtained using Langer’s transformation and its generalizations. The discussion of partial differential equations is limited to the case of the reduced wave equation with variable index of refraction. The expansion is first developed for the case in which the index of refraction deviates slightly from a constant, using the Born-Neumann procedure, and the solution is then represented by Feynman diagrams. The resulting expansion is valid only for short distances, and its region of uniformity is extended using the renormalization technique. Then the geometrical optics approach and the method of smoothing are described. Since these problems are linear, a vast amount of literature exists on their asymptotic solutions as well as their mathematical justification. We describe

308

7.1.

SECOND-ORDER DIFFERENTIAL EQUATIONS

309

in this chapter the techniques of developing formal asymptotic expansions to the solution of the equations without any mathematical justification. Moreover, a selective number of articles is cited. For further references and mathematical rigor, the reader is referred to ErdClyi (1956), Jeffreys (1962), Cesari (1971), Bellman (1964), Wilcox (1964), Wasow (1965), Feshchenko, Shkil', and Nikolenko (1967), Wasow (1968), and Frisch (1968). Second-order ordinary differential equations are treated first in Section 7.1, while systems of first-order ordinary differential equations are discussed in Section 7.2. Turning point problems are then taken up in Section 7.3, while the reduced wave equation is treated in Section 7.4. 7.1. Second-Order Differential Equations In this section we are concerned with the asymptotic development of the solutions of the equation (7.1.1) where E is a parameter which can be either small or large. We assume that p and q do not simultaneously vanish in the interval of interest. We discuss first the asymptotic solutions of this equation near an irregular singular point. Then we describe techniques for determining the asymptotic solutions of (7.1.1) when it contains a large parameter. Next we consider singular perturbation problems in which a small parameter multiplies the highest derivative. Finally, we describe techniques for obtaining asymptotic expansions when p , q , and r are slowly varying functions of x. 7.1.1. EXPANSIONS NEAR AN IRREGULAR SINGULARITY Let us investigate the asymptotic development of the solutions of (7.1.2) as x + co, if the point at infinity is an irregular singular point. Before doing so let us define a regular singular and an irregular singular point. Assume thatp(z) and q(s) can be developed in ascending powers of (2 - so),zo < co, as P(X) = Po(" - xJa[l P ~ ( Z- 4 . . .I, Po # 0 (7.1.3) q ( z ) = qo(x - zo)'[1 qi(5 - 2 0 ) . . .I, qo # 0 The point Z o is called an ordinarypoint if a 2 0 and fi 2 0; otherwise it is called a singular point. A singular point is called a regular singular point if a 2 - 1 and ,6 2 -2; otherwise it is called an irregular singular point.

+ +

+ +

3 10

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

The above definitions indicate that the nature of a finite point such as zo can be decided almost at a glance. The nature of the point at infinity can be determined by transforming it first into the origin. Thus we let x = z-l in (7.1.2) and obtain (7.1.4) The point at infinity is an ordinary point of the original equation if the origin is an ordinary point of the transformed equation; that is

which correspond to p(2) = 22-'

q(x) = o

+O ( d )

as

(~-~)

x+03

(7.1.5)

in the original equation. In order that infinity be a regular singular point of (7.1.2), the origin must be a regular singular point of the transformed equation; that is as

z+O

which correspond to p ( z ) = 0(z-1) q ( 2 ) = o(z-2)

as

x-+w

(7.1.6)

Thus if p ( x ) and q(x) can be developed in descending power series of x as

+

+

= p02' . . . , q(z) = qoxs . . . , po and qo # 0 then the point at infinity is an ordinary point if /? -4, and either u = - 1 withpo = 2 or a -2. In this case (7.1.2) has two convergent solutions in - 1 and /? -2, powers of x-l. If these conditions are not satisfied, and a the point at infinity is a regular singular point. In this case (7.1.2) has two of the form (Frobenius, 1875) convergent solutions in powers of z1

<

<

<

y = zy1

+

u15-1

+ a22-2 +

*

. .)

(7.1.7)

7.1.

SECOND-ORDER DIFFERENTIAL EQUATIONS

where a satisfies what is called the indicia1 equation a2

+ (po - 1)a + q o = 0

if

a = -1

and

p

=

3 11

-2

except in the special cases in which the roots of this equation are equal or differ by an integer, when one solution may involve log 2. If one or both of the inequalities &>-I,

B>-2

is satisfied, the point at infinity is an irregular singular point. This is the case of concern to us in this section. In this case (7.1.2) can be satisfied by solutions of the form y(x) = eh(z)sau(x) (7.1.8a)

co, which need not converge, and A(s) is a polywhere u(z) = O(1) as x nomial in x"ln. Letting Ax" be the leading term in A(%),substituting the above solution into (7.1.2), and extracting the dominant part of each term, we obtain pv2x2v-2 poAYxv+"-l 402fl = 0 (7.1.8b) Therefore v = u + l or 2 v = p + 2 (7.1.8~) ---f

+

+

whichever furnishes the greater value of Y. If Y is an integer, the above solution is called a normal solution (Thom6, 1883), and it has the form y = exp ( L z V

+ 2v-lzv-1+ . . . + il,z)z"(l + u,z-' + . . .)

(7.1.8d)

If v is not an integer, the above solution is called a subnormalsolution, and A is a polynomial in x1I2while u is an ascending series in x-lI2. So far, we have assumed that p and q are represented by power series of x so that a and /? are integers. If a and 8, are not integers, Y is a rational function which can be expressed in its lowest terms as v = n/k. Then the subnormal solutions have the general form

+ . . . + A2x2' + illx')xu(l + ulx-' +

+ . . .)

y = exp (Anznr

with

T

=

k-'

(7.1.8e)

To determine a normal or subnormal solution, we substitute the corresponding form of the solution into (7.1.2), equate the coefficients of like powers of 5 , and obtain equations which, in turn, can be solved for A,, 0, and a, in succession. Let us consider as an example the special case 00

m

3 12

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

wherep, and qn are independent of x . In this case, Y = 1 according to (7.1.8~) and (7.1.2) and (7.1.9) have a formal asymptotic solution of the form m

2

y = eAxxa c , , x - ~

(7.1.10a)

n=O

according to (7.1.8d), where 1 is a root of

1'

+ Po1 + 40 = 0

(7.1.10b) on account of (7.1.8b). By substituting (7.1.10a) into (7.1.2), using (7.1.9), and equating coefficients of like powers of x, we find that (7.1.10~) and we arrive a t a recurrence relation for c,. Jacobi (1849) developed normal solutions for Bessel functions of large argument and first order, while Stokes (1857) developed them for Airy's equation. Horn (1903) gave justification for the asymptotic solutions in the form of products of exponentials and descending series in x . 7.1.2. AN EXPANSlON OF THE ZEROTH-ORDER BESSEL FUNCTION FOR LARGE ARGUMENT The zeroth-order Bessel function is given by

"Y+'dy+y=o dx2

Here

(7.1.11)

xdx

p 1 = 1 and p m = 0 for rn # 1 qo = 1 and q m = 0 for rn # 0 Hence (7.1.10) indicates that = ei2x-1/2

m

as x +

m=O

00

(7.1.12)

Substituting this expansion into (7.1.11) and equating like powers of x , we obtain the following recurrence relationship c,+~ = - i

Hence if we take co = 1 = eixx-l/2

[

1--

+ !d2

(rn 2(m

+ 1)

1 1. 1 . 32 2 4" 2 . 2!. x 2 4 . 22 1 . 32 . 5 2 . 72 -k 44 . 24. 4! . x4

+

(7.1.13)

c?n

1 . 32. 43. Z3. 3 ! . x3

+ . . .]

as x + co (7.1.14)

7.1.

SECOND-ORDER DIFFERENTIAL EQUATIONS

3 13

Since the ratio of two successive terms

the right-hand side of (7.1.14) diverges for all values of x. However, for large x, it is an asymptotic expansion because the leading terms diminish very rapidly as m increases. Another linear independent expansion iJ can be obtained by replacing i by --i in (7.1.14); that is

17

=

1

e-’xs-l’2

(I

+

4.22

1 . 3’ - 1 . 3 2 . s2 I. 4’. 2’. 2!. x2 43.23. 3 ! . x3 1 . 32. 5 2 . 7 2 . -k 4 4 . 2 4 . 4 ! . x4

.

f -

+.

(7.1.1.5)

Real solutions can be obtained by linearly combining (7.1.14) and (7.1.15) according to

+

Y1

ye

Y 5= -

x-1/2

( u cos x

Y--d

s-112

( u sin x - u cos x)

2

N

__.

=

2i

where u(x) = 1

-

1 . 3*

4’. 2’. 2 ! . x2

+ u sin s)

1 . 32 . 5 2 . 7 2 4 4 . 2 4 . 4!.x4

o(x) = -~ 1 - 1 . 3 2 . s2 4 . 2 ~ 43.23.3!.x3

+...

+..

(7.1 . I 6)

( 7 . I . 17a)

(7.1.17b)

Therefore the asymptotic form of the Bessel function Jo is given by

+

Jo-Ayl By,asz+m where A and B are constants. From (7.1.16) through (7.1.18) lim X ~ ’ ~ J ~ = ( XA) cos x 5-t

m

+ B sin x

Iim x’”J~(x) = - A sin x 5-CO

+ B cos s

(7.1.1 8) (7.1.19) (7.1.20)

Hence A = lim x ” ~ [ J , ( ~cos ) x

- J ~ ( s )sin X I

(7.1.21)

B = lim Z ” ~ [ J ~sin ( ~ x)

+ J;(z) cos z]

(7.1.22)

x-m

5-.m

3 14

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

However, .To has the integral representation (e.g., Ince, 1926, Section 8.22) (7.1.23) Substituting this expression into (7.1.21), we obtain

1-03

lr

z+m

X

[cos x cos (x cos 0 )

+ sin x cos 8 sin (x cos 0)J de

If we let J%sin 812 = 4 in the first integral, and second integral, we obtain A = v/i[rcos lr

Similarly, we find that B = l/&. we obtain J 0 -J 2 - [u cos (x - t TX

4'd+ +lOmcos'c

d g cos 012 = 0: in the

da] = 7 1

JT

Therefore combining (7.1.16) and (7.1.18),

~+ )v sin (x - tlr)] as x

--f

co (7.1.24)

7.1.3. LIOUVILLE'S PROBLEM Liouville (1837) and Green (1 837) simultaneously considered the behavior of the solutions of

d2Y

-+ [

ax2

m X >

+ qn(4lY

=0

(7.1.25)

for large A , where q1 is a positive and twice continuously differentiable function and q2(z)is continuous in the interval [a,b] of interest. Using the transformation z = &x), u = W(X)Y(X> (7.1.26) we change (7.1.25) into d'v

+ dz'

1 ,, 2 4 ' ~ ' du p($ - ,)d2+ $[l"q,(z) + q2(x) - y

Choosing

4 and ly such that

7.1.

SECOND-ORDER DIFFERENTIAL EQUATIONS

3 15

or (7.1.29) we reduce (7.1.27) to (7.1.30) where (7.1.3 1) Since 91is twice continuously differentiable and q2 is continuous in [a, b ] ,

6 is small compared to L2. Hence, to first approximation, o is the solution of (7.1.30) with 6 = 0; that is

+

u = a cos ilz b sin Az where a and b are constants. Therefore, to first approximation

(7.1.32)

(7.1.33) If ql(z)is negative rather than positive, (7.1.33) is replaced by

Y=

Q

+ b exp [--i/,/--41(2) iGm

exp [ I / , / -

dz]

dx

(7.1.34)

These expansions are in full agreement with those obtained in Section 6.4.3 using the method of multiple scales. It should be mentioned that these expansions break down at or near the zeros of ql(z). These zeros are called turning or transition points. Turning point problems are discussed in Section 7.3. The transformation (7.1.26) and (7.1.29) is called the Liouville-Green transformation by mathematicians, while the solutions (7.1.33) and (7.1.34) are called the WKB approximation by physicists after Wentzel (1926), Kramers (1926), and Briliouin (1926). However, an approximate solution of this kind was obtained by Carlini (1817) for Bessel functions when the order and argument are both large. 7.1.4. HIGHER APPROXIMATIONS FOR EQUATIONS CONTAINING A LARGE PARAMETER We consider the asymptotic development of the solutions of

d2y

dx2

+ q(x, A)y = 0

(7.1.35)

316

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

for large A when q(z, A) =

A2kz A-nqn(z) a.

n=O

as

A -+ co

(7.1.36)

with qo # 0 in the interval of interest and k a positive integer. The asymptotic solution of this problem can be sought in either of the following two formal expansions (7.1.37) (7.1.38)

The justification of these expansions was given,by Horn (1899). Substituting either of these formal expansions into (7.1.35) and (7.1.36) and equating coefficients of like powers of A, we obtain equations for the successive determination of a, and g,. In the case k = 1 and the formal expansion (7.1.37), these equations are &Iz

2daL

and

+ 40 = 0

(7.1.39)

=0

(7.1.40)

+ + g%o (41

The solution of (7.1.39) is (7.1.42)

Then the solution of (7.1.40) is

which can be rewritten as (7.1.43)

where c is a constant. Hence, to first approximation

Y=

c1 cos p(z)

+ c2 sin p(z) [l + O(A")l

r40(411/4

(7.1.44)

where cl and c, are constants and

(7.1.45)

7.1.

SECOND-ORDER

DIFFERENTIAL EQUATIONS

3 17

Higher approximations can be obtained by solving successively for a, from (7.1.41). Had we used the second formal expansion (7.1.38) instead of (7.1.37) for the same case k = 1, we would have obtained the following equations for the determination of g, (7.1.46) gAZ q o = 0

+

+ q , + 2 gig',-, + gkp1 = 0 m-1

2g;gk

s= 1

for

rn 2 1

(7.1.47)

The solution of(7.1.46) is given by (7.1.42), and f o r m = 1 (7.1.47) becomes

2g;g;

whose solution is

+ 41 + gs = 0 (7.1.48)

Substituting for go and gl into (7.1.38), we obtain exactly (7.1.44) and (7.1.45). 7.1.5. A SMALL PARAMETER MULTIPLYING THE HIGHEST DERIVATIVE In this section we consider the equation (7.1.49) for E + 0. Asymptotic expansions of the solutions of this equation have been obtained in Section 4.1.3 by using the method of matched asymptotic expansions, in Section 4.2.2 by using the method of composite expansions, and in Section 6.4.2 by using the method of multiple scales. Using the transformation (Goldstein, 1969) (7.1 S O ) we transform (7.1.49) into the normal form dx2

2E

E

(7.1.51)

For r = 0, (7.1.51) has the same form as (7.1.35) and (7.1.36), with k = 1, 1 = 1/26, qo = --p2 and q1 = 2q - p'. Hence the complementary solution of (7.1.51) is (7.1.52)

3 I8

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

where =

' 2E

j'p(x)[l Zo

+

E

2q1

M

7 d x = - + 4 In p P 2E

An approximate particular integral to (7.1.49) can by obtained by putting E = 0; that is y+ = 1 /'g d x with E = e x p j 2 d x (7.1.54) E 20 P 20 P Hence to first approximation (7.1.55) Higher approximations can be obtained by assuming an expansion of the form m

where

y = 2 cnA,(x)e-M'E n=O

+ 2 engn(x) m

n=O

(7.1.56)

A o = - c2E -, B - 5 + i / 21 0 "p Ed rx (7.1.57) P OE This form is the same as that assumed when we employed the method of composite expansions in Section 4.2.2 where we determined M, as well as A, and B,, by substituting this expansion into the original equation and equating the coefficients of en and E" exp ( - M / E ) to zero. For the case p = p ( x , E ) and q = q(z,E ) , Wasow (1965, Chapter 7) assumed an asymptotic expansion of the form

(7.1.58)

where it, and A, are the roots of

+

+

€2' p ( x , .)A q ( x , €1 = 0 as E 4 0 (7.1.59) Thus if p and q are independent of E, A, = -p/E and A, = 0, and (7.1.58) takes the form (7.1.56). 7.1.6. HOMOGENEOUS PROBLEMS WITH SLOWLY VARYING COEFFICIENTS In this section we consider the equation

4 2 + p(Ex; .) 59 + ' ( € 5 ; .)y dx2

dx

=0

(7.1.60)

7.1.

where

E

SECOND-ORDER DIFFERENTIAL EQUATIONS

3 19

is a small parameter, and OD

m

n=O

n=O

p = z:."p,(t),q = z c n q n ( t ) with

f = ex

(7.1.61)

An asymptotic expansion of the general solution of (7.1.60) has the form (see Feshchenko, Shkil', and Nikolenko, 1967, for history and references) y

where

m

cnA,(E)eel

= n=O

+ 2 EnB,(E)eez 00

(7.1.62)

n=O

(7.1.63)

and A, and A, are the roots of

+

+

(7.1.64) 1, p o ( W qo(E) = 0 We have assumed that A, and 1, are different in the interval of interest. In (7.1.62), Oi and E are assumed to be independent. This is equivalent to the method of multiple scales described in the previous chapter. The derivatives are transformed according to -=

d dx

a + A, a A, 8% ae,

d2

1,2

a + c

at a2 - -2 + 2211, -+ A Z 2a2T dx2 a01 ae, ae, 8% a2 + 24+ €1;-a + 4;- a + a2 +2 4ae, at ae, at 801 ae, at a2

€

a2

2

7

where 2; = dA,/d[. Let A and B denote the coefficients of exp (0,) and exp (0,). Substituting (7.1.62) into (7.1.60) and equating the coefficients of exp (ei) to zero, we obtain (A,' + 2,p + q)A 4 2 2 , p)A' EA;A €'A" = 0 (7.1.65) (A,' + A,p + q)B 421, p)B' E ~ L B E'B" = 0 (7.1.66) Letting again

+ +

+ +

m

A = ~ E " A ,and n=O

+ +

+ +

m

B = zcnBn

(7.1.67)

n=O

in (7.1.65) and (7.1.66) and equating coefficients of like powers of E, we obtain equations to determine successively A, and B,. The first terms A , and Bo are given by (21,

(21,

+ P o ) 4 + (1;

+@l

+ 41)Ao = 0

+ Po)% + (1; + A2Pl + 4 1 P O = 0

(7.1.68) (7.1.69)

320

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

Their solutions are A,, B, oc exp

-

21,

Inthecaseofp=Oandq,=Ofotn> 11,

and

(7.1.70)

+ Po

1

1' = &qqo(E)11/2

A - - ,

a

b

B---

O - J X

O

- $1

where a and b are constants. Therefore, to first approximation J

J

Y=

(7.1.71)

[40(t)1''4

This is the Liouville-Green or WKB approximation to the solution of

fi + qo(Ex)y = 0 dx'

(7.1.72)

7.1.7. REENTRY MISSILE DYNAMlCS For a symmetric missile the complex angle of attack is given by (Nayfeh, 1969a) i p ( l + y ) + - +u- - c D EQ ] d u

u

In this equation u, Q, D, and M are slowly varying functions of time, while p and y are constants. Hence it has the same form as (7.1.60), with

q,, = y(p2

- p'),

q1 = i p

- D -M +-

(7.1.74)

Here primes denote differentiation with respect to the slow time E = ~ t , while overdots denote differentiation with respect to the fast time t. Substituting for po and qo into (7.1.64) gives

A2

+ i d 1 + y)A + y(pC2- P')

=0

(7.1.75)

7.1. Hence i= -&i(l

Then (7.1.70) gives

SECOND-ORDER DIFFERENTIAL EQUATIONS

+ y ) p f iw,

w =

A - &,-A+AA O

where

,

- JWU

(Q/u - D ) d l ,

AA

Jf(1- y)'p2 + y p e 2

B - = e O

='s

- JOU

-A-AA

321

(7.1.76) (7.1.77)

( i-t r+ - + Q D)(I-y)+ZM

4

dl

w

(7. I .78)

Therefore, to first approximation fi=--- I {a exp [-A AA - i i ( 1

+

,fi

+ b exp [-A

+ y)pz + i - AA - gi(1 + y ) p x - i J w d z ] } (7.1.79)

Analysis of missile dynamics was performed by Fowler et al. (1920), Fowler and Lock (1921), Green and Weaver (1961), Murphy (1963), and Coakley (1 968) among others. 7.1.8. 1NHOMOGENEOUS'PROBLEMS WITH SLOWLY VARYING COEFFICIENTS We consider in this section the asymptotic development of the general solution of (7.1.80) where

3 = w(5)

with

dx

and m

p = 2E~P&), n=O

E

= EX

(7.1.81) m

m

q =

2 enqn(E),

n=O

r =

2

n=O

Enrn(E)

(7.1.82)

Two cases arise depending on the values of iw and the roots 1, and A, of

+

+

(7.1.83) A2 pol 40 = 0 If iw = A, or 1, at one or more points in the interval of interest, we have a resonant case; otherwise we have a nonresonant case. We treat the latter case first. The Nonresonant Case. In this case we assume that

322

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

where (7.1.85) The equations for A and B are the same as (7. I .65) and (7.1.66). To determine C , we let y = C e x p (i$) in (7.1.80), equate the coefficients of exp (i4) on both sides, and obtain (-0'

+ imp + q)C + 42iw + p)C' + iaw'C + &"= r

(7.1.86)

Letting rn

(7.1.87)

in (7.1.86) and equating coefficients of like powers of E, we obtain equations to determine c, successively. The first one is given by co =

r0

-0'

+ impo + q,,

-

10

(io

- A,)(io

- A,)

(7.1.88)

The solution for A and B is the same as in Section 7.1.6. Therefore, to first approximation

21,

+ Po

24

+ (io - Al)(iw - A,) r0

ei6

+ Po (7.1.89)

where a and b are constants. The Resonant Case. The expansion (7.1.89) breaks down whenever iw is equal to either 1, or A, at one or more points because the last term is unbounded at such points. We assume that iw is equal to L1 at one or more points while iw # 1, in the interval of interest. Asymptotic expansions valid in this case have been developed by Fowler et al. (1920) and Fowler and Lock (1921). The particular solution in this case has the form

where

y = ~ ( x:)ei+ ,

(7.1.90)

7.1.

SECOND-ORDER DIFFERENTIAL EQUATIONS

323

Substituting (7.1.90) and (7.1.91) into (7.1.80) and equating the coefficients of each of 7 exp i+ and exp i+ on both sides, we obtain

+ cG' = 0 (G + p + i o ) H + EH' = r G2 + p G + q

Letting

(7.1.92) (7.1.93)

m

4,

G =n=O Z:"Gn(6) and

H =n=O ~E~H,(E)

(7.1.94)

in (7.1.92) and (7.1.93) and equating coefficients of like powers of E , we obtain equations for the successive determination of the G, and H,,. The first two terms are given by G: poco qo = 0 (7.1.95)

+ + 2GoG1 + PoGi + PiGo + + GA = 0 41

+ Po + i 4 H O = (Go + + io)Hi + (GI + pi)Ho + HA = (Go

10

(7.1.96) (7.1.97)

PO rl (7.1.98) Equation (7.1.95) shows that Go = A, or 1,. We take Go = Al because 1, is assumed to be equal to io at one or more points. Equation (7.1.96) gives

G1 = - PIGO

+ GA + 41 + Po

2Go

(7.1.99)

The solutions of (7.1.97) and (7.1.98) are

The general solution of (7.1.81) is then y = A(E, €)eel

+ B(6, E)ee* + ~ ( xE)e'+ ,

(7.1.101)

where A and B are determined as in Section 7.1.6. The complementary solution of (7.1.80) can be obtained by using the following technique which is an alternative to that used in Section 7.1.6. We assume that ye = ('(2, c )

d5 = F(5, E ) < with dx

(7.1.102)

Substituting (7.1.102) into the homogeneous part of (7.1.80), we obtain Letting

F2

+ p F + 4 + EF' = 0 m

F = ZcnFn(5) n=O

(7.1.103) (7.1.104)

324

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

in (7.1.103) and equating coefficients of like powers of c , we obtain equations to determine F,,. The first two are F,2+P$~+40=0 (7. I. 105)

+

+

+ +

2FoF1 PoFi ~ i F o 41 F; = 0 (7.1.106) Equation (7.1.105) has two roots, A, and A,, as (7.1.64). Then (7.1.106) gives

(7.1.107) Hence

Integrating (7.1.108), we find that yc is the same as that obtained in Section 7.1.6. This method of expansion is the-same as that represented by (7.1.38). 7.1.9.

SUCCESSIVE LIOUVILLE-GREEN (WKB) APPROXIMATIONS To obtain higher approximations to the solution of (7.1.109) for small E, Imai (1948) proposed the use of successive Liouville-Green (WKB) transformations. Thus we introduce the transformation

ak, = EX) ak,

y1 = [ k ( c ~ ) ] ” ~ y ( ~ )

(7.1.1 10)

thereby transforming (7.1.109) into

iedX1 + k:y,

where

(7.1.111)

=0

k12 = 1 - - -

(7.1.112)

Since k varies slowly with x, k12 !w 1, and an approximate solution to (7.1.1 11) is y1 = a cos x1 b sin x1 (7.1.113)

+

with constant a and 6. Hence a first approximation to the solution of (7.1.109) is

s

a cos k dx Y=

+ b sin

k1/2

s

k dx (7.1.114)

7.2.

SYSTEMS OF FIRST-ORDER ORDINARY EQUATIONS

325

To determine a second approximation to y, we note that (7.1.11 1) has the same form as (7.1.109). Thus an improved solution to (7.1.111) can be obtained by introducing the transformation k dx

1 d2k

(7.1.115) (7.1.1 16)

L

Then (7.1.111) is transformed into d2y, dx;

where

+

(7.1.1 17)

kZ2y2= 0

(7.1.118)

Since dkl/dxl = k-l ( d k , / h ) = O(r), the last two terms in (7.1.118) are small compared to 1, hence a first approximation to y2 is YZ

= a cos x2

+ b sin x2

(7.1.119)

Therefore a second approximation to y is given by

Y=

k112

(a cos x2

+ 6 sin x 2 )

(7.1.120)

where (7.1.121)

Higher approximations can be obtained in the same manner by introducing the new transformations dxn+l = kn dxn,

Yn+l

-

= Jkn

Yn

(7.1.122)

7.2. Systems of First-Order Ordinary Equations

In this section we also consider first the asymptotic solutions of equations near an irregular singular point which is assumed to be 03. Then we discuss equations with either a small or a large parameter. Finally, we describe asymptotic expansions for equations with slowly varying coefficients.

326

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

7.2.1. EXPANSIONS NEAR AN IRREGULAR SINGULAR POINT We consider the behavior of the system of n linear equations

dy = x*A(x)y dx

as x

+

cu, where q is an integer A(z) =

(7.2.1)

2 - 1, and the matrix

m

2A,X-~

m=O

as x + cu

(7.2.2)

If q = - 1, x = co is a regular singular point of (7.2.1), while if q > - 1, x = m is an irregular singular point of the system. The behavior of the solution near an irregular singularity depends on whether all the eigenvalues of A. are distinct or not. We discuss in this section the case of distinct eigenvalues. If (7.2.1) is a scalar equation, it can be solved explicitly and its solution has the form y ( x ) = U(x)z“e*(z’ (7.2.3) where G is generally a complex constant, Q(z) is a polynomial of z having the form (0 if q = - 1 Q ( x ) = Q+I (7.2.4) 12 m=l Qmxm and m

U(X) = 2 umx-m m=O

(7.2.5)

In the case of systems of equations, the asymptotic solution still has the form (7.2.3) but G, Q,, and Urnare constant matrices. This expansion was called a “normal” solution by Thomi (1 883). To calculate the asymptotic expansions for the solutions of (7.2.1) and (7.2.2), we seek formal solutions of the form where

G

y = u(x)x‘eA(z)

is a constant,

(7.2.6) (7.2.7)

with I - , = 0 for m

2 0, and U(X)

=

m

2U,X-~

m=O

as x + co

(7.2.8)

Here A is a scalar quantity, while y and u are column vectors. Substituting (7.2.6) through (7.2.8) into (7.2.1) and (7.2.2) and equating the coefficients

7.2.

SYSTEMS OF FIRST-ORDER ORDINARY EQUATIONS

327

of equal powers of x, we obtain equations to determine successively A,, a, and u,. The first equation gives (A, - Aa+lZ)uo= 0,

I: identity matrix

(7.2.9)

For a nontrivial solution the determinant in (7.2.9) must vanish. This condition gives the following nth-order algebraic equation 1-40

- &+I[l = 0

(7.2.10)

If the eigenvalues of A , are distinct, (7.2.10) gives n distinct values for A,, which correspond to n linear independent solutions of the form (7.2.6). 7.2.2. ASYMPTOTIC PARTITIONING OF SYSTEMS OF EQUATIONS Sibuya (1958) developed the following scheme for simplifying the system of equations (7.2.1) by reducing them to some special differential equations whose solutions can be found more readily than those of the original system. To accomplish this we let (7.2.1 1) Y = W)V(X) where P is an n x n nonsingular matrix to be determined and v is a column vector. Hence (7.2.1) is transformed into

-

dv = zQB(z)v dx

where

A(x)P(z) -

(7.2.12) 5-9

-

or

dx

(7.2.13) The essence of the technique is to choose the matrix P(x) such that the matrix B(s) has a canonical Jordan form. To d o this we let m

B = z B m x - m as x-+

a,

m=O m

(7.2.14)

P = z P m x - m as x + a , m=O

where Bm represents a Jordan canonical matrix. Substituting (7.2.14) into (7.2.13) and equating coefficients of like powers of 5 , we obtain (7.2.15) AoPo - P,B, = 0 m-1

328

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

for rn 2 1, with Pm-a-l= 0 if m then Po can be chosen such that

- q - 1 < 0. If A, has distinct eigenvalues, (7.2.17)

B, = P,'AoP,,

is diagonal. We multiply (7.2.16) from the left with P;' to obtain

- W,Bo

BOW,

where

= Bm

+F,

(7.2.18)

w, = K I P r n

F , = --P,'A,Po

+ PF1 2 (P,Bm, m-1 S=l

(7.2.19)

- A,,PJ

- (m - 4 - l)Wm-a-i (7.2.20)

If we denote the components of Fmby FC, we choose B, such that B: = -F:,

B g = 0 for i # j

(7.2.21)

Since B, has distinct eigenvalues, (7.2.18) with (7.2.21) can be solved to determine W,, hence P,, from (7.2.19). In the case of multiple eigenvalues, we can partition the system of equations into simpler systems using the same scheme. We assume that A . has multiple eigenvalues and that there exists a matrix Po such that B, = P,'A,P,

(

= 'B

0

BF

)

(7.2.22)

where B:' has the eigenvalues iZi ( i = 1,2, . . . ,r) while BF has the eigenvalues 1, ( j = r 1, r 2, . . . , n) such that iZi Aj. Let us partition W,,, and F, according to

+

+

+

w: w:/

(7.2.23)

F, =

where Wil and Fkl are r x r matrices while W;' and Fk2 are (n - r) x (n - r) matrices. We choose

wt = W E = 0

and B:

=

-FE,

B g = -pg

(7.2.24)

Then (7.2.18) becomes

(7.2.25) These equations can be solved uniquely for Wi' and WE because BF and Bia do not have common eigenvalues.

7.2.

SYSTEMS OF FIRST-ORDER ORDINARY EQUATIONS

329

As an example, we consider Bessel's equation 2 ' -

Letting

+

d2Y + x-dY dx2 dx

2

(5

Y = u1,

- n 2)y

=o

(7.2.26)

dY = u2 dx

(7.2.27)

we transform (7.2.26) into

du

- = A(s)u dx

Hence q = 0, A,,, = 0 for m

> 2 , and

Since the eigenvalues of A, are f i ,

Po=

(

1 i

7

-i

(I 3 1

and P 2 =

2

so that

Bo = PilAoPo = Hence from (7.2.20)

(7.2.30)

1

(

0 -i

)

(7.2.31)

(7.2.32) Therefore from (7.2.21) and (7.2.25) we obtain

B1= With W,and B, known

--( ), 1 1 0 2 0 1

w&(

-1

0

)

1 0

(7.2.33)

(7.2.34)

330

ASYMPTOTIC SOLUTlONS OF LINEAR EQUATIONS

from (7.2.20). Hence

Substituting the expressions for B,, B,, and B2into (7.2.12),we obtain

Hence

v2

b

=Jexp

[-ix

- i-

1

(7.2.36)

+ 4n2

where a and b are constants. Since P = Po + 1PI X

+ 0(x--2)= Po(z +: - w, ) +O(X3

(7.2.37) Therefore 1--

+

%(

+ -i

i)

--

exp [-ix

- i-]

82

+ O ( K ~ ” ) (7.2.38)

7.2.

SYSTEMS OF FIRST-ORDER ORDINARY EQUATIONS

33 1

Note that we did not use P, in (7.2.37) and (7.2.38) because the error in v is O ( Z - ~ / ~To ) . compare these results with those obtained in Section 7.1.2, we expand exp [ f i ( l 4n2)/8s] in powers of x-l. Since

+

-ix-i-

+824n2) + 0(~-5’2)

in agreement with (7.1.14) when n = 0. Higher approximations can be obtained by calculating the higher values of B, and W , in a straightforward, though tedious, way. The technique employed in Section 7.1.2 is very much easier to implement than the technique described in this section. 7.2.3. SUBNORMAL SOLUTIONS If A, of (7.2.1) has multiple eigenvalues, we cannot decouple all these equations by choosing B to be a diagonal matrix. Instead we partition this system of equations to obtain simpler systems of the form m

dv’ = z ~ B ~ ( z ) ~ ’ , B’ = 2 Bmiz - ~ dX

(7.2.40)

m=O

where the eigenvalues li of B l are different from l j of B ; for i # j . Thus corresponding to each single eigenvalue A,, B” is a scalar, and (7.2.40) can be solved. If q = 0 (7.2.41) where a is a constant and c, are known in terms of Brmfor r 2 2. Hence a normal solution exists corresponding to this eigenvalue of the original system (7.2.1). For a multiple eigenvalue 1, of multiplicity m,,B* has a rank of m,.It turns out that this reduced m, system of equations, hence the original system, may not have a normal solution corresponding to this eigenvalue. However, it may have what is called a subnormal solution of the form (7.2.3) through (7.2.5), but Q and U are expanded in powers of xIir with integer r. As an example, the equation dx

dx

(7.2.42)

332

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

has the general solution = aed/z(xa/4 - x-514) + be-d;(23/4 + ~ - 5 1 4 )

(7.2.43)

which consists of two subnormal solutions. Equation (7.2.42) is equivalent to the system

-dx

(7.2.44)

In this case

has the eigenvalue , I= 0 with a multiplicity of two, which precludes the existence of normal solutions. 7.2.4. SYSTEMS CONTAINING A PARAMETER Let us consider the system of n linear equations = A(x, e)y (7.2.45) dx where E is a small positive number, h is an integer, and A(%, E) is an n x n matrix which possesses the asymptotic expansion eh*

m

A(s, e) = 2 emA,(x) as m=O

0

E -+

(7.2.46)

If h is zero or negative, y possesses asymptotic expansions of the form m

If h > 0 the asymptotic expansions of the solutions of (7.2.45) depend on whether the eigenvalues of Ao(x) are distinct in the whole interval of interest or not. A point at which A,(x) has multiple eigenvalues is called a turning or transition point; turning point problems are discussed in Section 7.3. If the eigenvalues of A,(%) are distinct, the asymptotic representations of n linearly independent solutions of (7.2.45) have the form Y = u(x, E) exp

[S

1

A(x, E) ds

(7.2.48)

7.2. SYSTEMS

OF FIRST-ORDER ORDINARY EQUATIONS

where

I:€-'A&) h

A(%, u(x,

=

(7.2.49)

.>

€)

333

7-1

m

= 2 E'rd(5)

(7.2.50)

7EO

Substituting (7.2.48) through (7.2.50) into (7.2.45) and (7.2.46) and equating coefficients of like powers of .s, we obtain equations that determine successively il,and u,. There are n linear independent solutions of the form (7.2.48) through (7.2.50) corresponding to the n eigenvalues of A o ( z ) ; that is, the solutions of (7.2.51) P O ( 4 - AO(4Zl = 0 7.2.5. HOMOGENEOUS SYSTEMS WITH SLOWLY VARYING COEFFICIENTS In this section we consider the asymptotic solutions of

where

dY = A(5, .)y, dx m

A(6, c) = 2 E"A,(E) m=O

5 =EZ as

E

-+

(7.2.52)

0

(7.2.53)

In this problem x is a fast variable while 6 is a slow variable. As in Section 7.2.2, we assume the existence of a nonsingular matrix Po(5)such that

where BY has the eigenvalues Ai (i = I , 2, . . . ,r ) and Bizhas the eigenvalues A j ( j = r + 1 , r + 2, . . . , n) such that Ai # Aj. In this case we can reduce the original coupled system of equations (7.2.52) into two decoupled systems of orders r and n - r . To do this we let (7.2.55)

where

dx

= B(&

E)V

(7.2.56)

334

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

or (7.2.57) We seek asymptotic representations of P and B of the form

where B , represents a block diagonal matrix. Substituting (7.2.58) into (7.2.57) and equating coefficients of like powers of E, we obtain

m-1

F,

=

2 (P,B,,

8-1

- A,,P,)

- AmPo +dP,l

d5

(7.2.60)

As in Section 7.2.2, we choose B, and Po in accordance with (7.2.54), multiply the second of equations (7.2.59) from the left by Pi', and use (7.2.19) to obtain BOW, - W,B, = B, F, (7.2.61) where W, = PG'P,, F , = PG1Frn (7.2.62)

(zzi),

+

To solve (7.2.61) we partition F, and W , according to F, =

(

w, = 'w w') w: w2,2

(7.2.63)

where F: and WE are r x r matrices. If we choose W 11, = W E = 0 and B:=

''

-Fm,

B:2= - F Z

(7.2.64)

(7.2.61) becomes BA'WE - WzB:' = F',2 BFWZ - WZB: = F:

(7.2.65)

(7.2.66) These equations can be solved uniquely for W;' and Wkl since B: and BF do not have common eigenvalues. If A , has distinct eigenvalues, one can use the above scheme to reduce the original system to an uncoupled system of n equations having the form (7.2.56) with B a diagonal matrix. The details are the same as in Section 7.2.2. An easier technique can be used to determine the asymptotic solutions of

7.3.

TURNING POINT PROBLEMS

335

the system (7.2.52) if A , has distinct eigenvalues. The asymptotic representation has the form (7.2.67) (7.2.68)

df, = A(E)

(7.2.69)

dx

where A ( f ) is an eigenvalue of A,(<). There are n linearly independent solutions of the form (7.2.67) corresponding to the n eigenvalues of A,. Substituting (7.2.67) through (7.2.69) into (7.2.52) and (7.2.53) gives equations that determine u, successively.

7.3. Turning Point Problems We found in Section 7.1.3 that the Liouville-Green or WKB approximation to the solutions of (7.3.1) for large 1 is

Y‘

al cos

[AIJq,(dx] z) + b, sin [IsJq,(z)dx]

(7.3.2)

for positive ql(x) and

Y=

a2 exp

[.3JJ--q,(z)dz

1v + [ I b2 exp

-qdx)

-A ,/--ql(x) dx] (7.3.3)

for negative ql(x). As remarked in Section 7.1.3, these approximations are valid as long as x is away from the zeros of ql(x). Equations (7.3.2) and (7.3.3) show that y is oscillatory on one side of a zero of ql(x) while it is exponential on the other side, hence such a zero is called a transition point. It is also called a turningpoint because in classic mechanics it is the point at which the kinetic energy of an incident particle is equal to its potential energy and the particle therefore reverses direction. The point x = p is called a turning point or a transition point of order a where a is the order of the zero of ql(x) a t x = p . If q2(x) is singular a t a turning point, the turning point is called a singular turning point; otherwise it is regular. In this section we

336

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

describe techniques for determining the asymptotic solutions of turning point problems starting with second-order equations such as (7.3.1). 7.3.1. THE METHOD OF MATCHED ASYMPTOTIC EXPANSIONS Let us assume that qz(z)is a regular function while

ql(z) = (z - p ) f ( s ) with positivef(z)

(7.3.4)

Hence the approximate solution to (7.3.1) is given by (7.3.2) for z > p and by (7.3.3) for z < p . These expansions are called outer expansions and break down near z = p. To determine the region of nonuniformity of these expansions, we let 5 = (z - p)Av with positive v in (7.3.1) and obtain

5% dF2

+ {1""'6fflu

+ 51--']+ L - Z ' q , ~ + 6A-VI)y

=0

(7.3.5)

As 1 + 00 the third term in (7.3.5) tends to zero for all positive v ; however, the resulting equation depends on the value of v. As 1 + co (7.3.5) tends to y=O d2Y

-=0 dF2

if v < j if v > %

(7.3.6)

It is obvious that the first two limits are not acceptable because their solutions do not match the outer expansions (7.3.2) and (7.3.3). Therefore the acceptable limit is the distinguished limit with Y = 2/3 yielding the third equation in (7.3.6). If we let =

-N%)

(7.3.7)

the first-order inner solution is governed by (7.3.8) Its general solution is

y = u,Ai(z)

+ b,Bi(z)

(7.3.9)

where Ai(z) and W ( z ) are the Airy functions of the first and second kind, respectively. Let us now digress to give some properties of the Airy functions which we will need in the ensuing discussion. These functions have the following

7.3.

TURNING POINT PROBLEMS

337

integral representations see (e.g., ErdClyi, 1956, Section 4.6) Ai(z) = T

Bi(z) = T

S" S" O

O

+

cos (it3 z t ) d t [exp (-it3

(7.3.10)

+ z t ) + sin ( i t 3 + zt)] d t

(7.3.11)

These functions can also be related to Bessel functions of order 1/3 according to Ai(z) = iJz[I-i/dc)

- 11/dc)l =

(7.3.12)

77

(7.3.13) (7.3.14) (7.3.15)

where 5 = (2/3)z3l2. For large positive asymptotic expansions

z

these functions have the following (7.3.16)

1 Ai(--z) = -z - ~ sin / ~

,I;

1

Bi(-z) = T2-II4 cos

Jn

i5 + 3

(7.3.1 7)

-

(7.3.18)

3

5+-

(7.3.19)

To match the inner solution (7.3.9) with the outer solution (7.3.2), we express the latter in terms of 6 = (z - p ) P and determine its limit as A- 03. In this case z > ,u and

+ O(A-2'3) Hence

+...

(7.3.20)

338

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

Expressing (7.3.9) in terms of

Y = a3Ai(-

Its expansion for large

E , we obtain

6qf(p)) + b3Bi(- E qf (p))

6,obtained

by using (7.3.17) and (7.3.19), is

+ b, cos (845) + ')] + .4 E3I2

*

(7.3.2 1)

Since the matching principle demands that (7.3.20) and (7.3.21) be equal, we obtain al = --_[a,sinA---116f116

J"

"I

"4 + b3cos-4

7.3.2) becomes

(7.3.23)

where

To match (7.3.9) with the outer solution (7.3.3), we note that

Hence (7.3.3) becomes

(7.3.24) Since 5 is negative in this case, z = is positive, and the asymptotic behavior of the inner solution (7.3.9) for large z, obtained from (7.3.16) and (7.3.18), is

Evm

+ b3 exp (gJfe)<-E)3'2)]

+ . ..

(7.3.25)

7.3.

TURNING POINT PROBLEMS

339

Equating (7.3.24) and (7.3.25), we obtain (a23 b2)

= (639

34)

(7.3.26)

Hence the outer solution (7.3.3) for negative ql(x) is

(7.3.27) Therefore an approximate solution to (7.3.1) with a turning point at x = p is given by the three separate expansions: (7.3.9) near the turning point, (7.3.23) for z > p , and (7.3.27) for x < p. The matching provided the connection between the constants a, and b,, and u2 and b2. This connection was first given by Rayleigh (1912) in his investigation of the total reflection of sound waves from a transition layer, and he presented an explicit solution for the exponentially decreasing solution only. Gans (1915) gave connection formulas for both solutions, and Jeffreys (1924) rediscovered them in an application to Mathieu’s function. Wentzel (1926), Kramers (1926), and Brillouin (1926) rediscovered them about the same time in their investigations of the Schrodinger equation. Thus in the physics literature this method is usually named with some permutation of the letters W , K , and B, and recently the letter J has been added to these letters for the contribution of Jeffreys. Zwaan (1929) established the connection formulas by integration along a path in the complex plane that avoids the turning point. This technique was developed further by Kemble (1935). A disadvantage of this technique is the fact that the solution is given by three different expansions. A single expansion uniformly valid for all x was obtained in Section 6.4.4 by using the method of multiple scales. In the next section we discuss a powerful technique of treating turning point problems originated by Langer (1931, 1934) and developed by Langer and several researchers as indicated in Sections 7.3.2 through 7.3.10. 7.3.2. THE LANGER TRANSFORMATION The gist of Langer’s approach is that approximately identical equations have approximately identical solutions. He realized that any attempt to express the asymptotic expansions of the solutions of turning point problems in terms of elementary functions must fail in regions containing the turning points. A uniformly valid expansion for all z must be expressed in terms of nonelementary functions which have the same qualitative features as the solutions of the equation. The decisive step in Langer’s approach is to introduce a transformation

340

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

of the dependent and independent variables of the form = 4(4, 7J = Y(.)Y(X) which transforms

dx2

into

(7.3.28)

+ [A2q1(x) + 42(2)lY = 0

The middle term vanishes if y =

leaving

(7.3.29)

JZ

(7.3.3 1) (7.3.32)

If we choose

41 = 1 or

4’z

4=

s-

(7.3.33)

Jq1(7) d7

we recover the Liouville-Green transformation. The resulting solution is -7.1.3, and it is expressed in terms of the circular functions as in Section singular at the turning points (zeros of ql). Since y = and 4’ = y = and the transformation (7.3.28) is singular at the zeros of q,(z). To obtain a uniformly valid expansion for a problem with a turning point at z = p such that q 1 ( 4 = (x - p)fW with f(4 > 0 (7.3.34)

44’

+&

&,

we follow Langer (1931, 1934) by choosing (7.3.35)

so that (7.3.32) becomes

d2u + A2zv = dv dz2

where

(7.3.36) (7.3.37)

The solution of (7.3.35) is

g43‘z

=Ix,/(. -

p ) f ( 7 ) d7 for

x 2p

c

g( - 4)3’2=r J ( p - 7 ) f ’ ( 7 )d7 X

(7.3.3 8) for

x _< p

-

7.3.

-

-q z ,

TURNING POINT PROBLEMS

341

As z - p , (b v / f ( p ) ( x - p ) and y hence S = O(1) if q2 is continuous. Moreover, the transformation (7.3.28) is regular everywhere including the turning point z = p. Since 6 = O(1) and 1 is large, v is given approximately by what Langer calls the related equation (7.3.39)

whose solution is

+

u = c , A ~ ( - ~ ~ / ~c z2 )~ i ( - P 3 4

(7.3.40)

where c1 and c2 are constants of integration. Hence, to first approximation

+

1

y = T { C ~ A ~ [ - ~ ' / ~ + ( Z~ )~]S i [ - I * / ~ ~ j 5 ( z ) ](7.3.41) }

J +'(4

where (b is defined in (7.3.38). This single expansion is uniformly valid for all x including the neighborhood of the turning point x = p. Using the asymptotic expansions (7.3.16) through (7.3.19) for large argument of the Airy functions Ai and Bi, we obtain 1F6 . c1 sin 2 J q l ( r ) d r Y = J%i?'/q __ i(x) 4

(

[6

+

"1

for x

>p

(7.3.42)

for z

(7.3.43)

and

in agreement with the results of the previous section. Olver (1954) generalized Langer's transformation to the form

(7.3.44)

where the independent variable z is still any undefined function of x. With

342

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

this transformation (7.3.29) becomes (7.3.45) where (7.3.46) If we can choose

((2)

such that 6 = 0(1), then the related equation d2v dz2

+ A2(’2v= 0

(7.3.47)

has solutions which are asymptotically equivalent to the solutions of (7.3.29) for large A. In order for 6 to be 0(1),x must be regular and not vanish in the interval of interest. Consequently, 5 must be chosen so that 5‘2 has zeros and singularities of the same order and at the same places as those of ql(x) so that and ql(x) are positive or negative together. Moreover, it is desirable to choose so that the related equation (7.3.47) is solvable in terms of known functions. This general transformation was rediscovered by Moriguchi (1 959). If ql(z) = (z - p)f(z) with f ( z ) > 0, = fz (7.3.48) [I2

which yields Langer’s transformation. 7.3.3. PROBLEMS WITH TWO TURNING POINTS We consider the case in which

q1W = (x - p l ) ( p 2 - x > f ( z ) , p2 > p1 and f(4 > 0 (7.3.49) so that (7.3.29) has two simple turning points at z = p1 and p2. Such problems with two turning points arise, for example, in the solutions of the Schrodinger equation (e.g., Jeffreys, 1962; Pike, 1964) for tunneling or classic oscillator problems and in the determination of heat transfer in a duct (e.g., Jakob, 1949, pp. 451-480). Applying the results of the previous section to the turning point x = pl, we obtain

where

-

7.3.

TURNING POINT PROBLEMS

343

However, as z p2, $; = O[(x Therefore (7.3.50) breaks down in the neighborhood of x = p z , and it is valid for p2 - x > 6, where 6, is a positive small quantity. Applying the results of the previous section to the turning point z = p2, we obtain

+

1

y = ={a2Ai[-~2/3$2(x)] b2Bi[-12/3~2(x)]} for

J$W

x

- p, > 6, (7.3.52)

where 6, is a small positive number, and

3$Y2

K-+P

=s”’&

=J [T

- PI>(P2 - T)f(T) dT for

- pl)(T - ~ M T I

2

for x

< pz > p2

Since both (7.3.50) and (7.3.52) are valid in the interval ,ul

(7.3.53)

+ 6, < z <

p2 - a,, we can connect these expansions by matching. Expanding (7.3.50)

for large argument and for x

> pl,using (7.3.17) and (7.3.19), we obtain

+ b, cos ($A$:’” + :)] Similarly, expanding (7.3.52) for large argument for x

(7.3.54)

< ,u2,we obtain

+ b, cos (31$:/’ +

$1

(7.3.55)

Equating (7.3.54) and (7.3.55) gives a, sin ($1+:”

+): + b, cos ($A&”

(

= a 2 sin #A$;’2

If we let then

+ ): + - + b2 cos (

3

+ 4)

(7.3.56)

344

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

hence from (7.3.56)

a , = 6, sin A - a2 cos A b, = a2 sin h 6, cos h

+

(7.3.58)

-

NOWif y is a bounded function of z, as is the case in solutions of the Schrodinger equation, 6 , and 6 , = 0 because Bi(z)+ n-1p2z-1/4 exp [(2/3)z3//"] as z co. Hence from the second equation of (7.3.58) sin A = 0 or

A

with n an integer

= nr

(7.3.59)

Therefore from (7.3.57)

Rather than representing the solution by two expansions, Miller and Good (1953), Kazarinoff (1958), and Langer (1959b) suggested expressing the solution by a single uniformly valid expansion using parabolic cylinder functions. Using the transformation (7.3.44), we choose Cf2 so that it has two simple zeros. We take those to be z = f I with z = - 1 corresponding to z = p l , and we let (Pike, 1964) ['* = 4a2(1 - z z )

(7.3.61)

We choose a so that z = 1 corresponds to z = p2. Thus from (7.3.44) we obtain (7.3.62)

-1

where the branches of the square roots should be chosen so that z is a regular function of z and the regions where ql(z)> 0 and ql(z)< 0 transform into z2 < 1 and z2 > 1, respectively. Choosing z = 1 to correspond to z = p 2 , we obtain the following equation for a

I

Hence

(7.3.63) With (7.3.61) the related equation is d21; -

dz2

+ 4a2R'(I

- Z')D

=0

( 7.3.64)

7.3.

TURNING POlNT PROBLEMS

whose solutions are given by I--

u = Wv(2t'uilz),

Y

+ 3 = uil

345

(7.3.65)

where W , is Weber's function of order v. If y is bounded at infinity, u must be bounded, hence 1' = n where n is an integer. Therefore

-

A=

+ a>

pa3 4 n

(7.3.66)

dT

in agreement with (7.3.60). Problems with two turning points have also been analyzed by Olver (1959), and Moriguchi (1959). Several turning point problems were treated by Evgrafov and Fedoryuk (1966), Hsieh and Sibuya (1966), Sibuya (1967), and Lynn and Keller (1970) among others. 7.3.4. HIGHER-ORDER T U R N I N G POINT PROBLEMS In this section we let q , = (z

.- p)"f(x),

f(z)> 0 and

a is a positive real number (7.3.67)

To determine a single uniformly valid expansion, we let [ ' z = z' so that has the same number of zeros as q l . Hence

['2

(7.3.68) where the branches of the square root are chosen so that the regions where ql(x) > 0 and ql(x) < 0 correspond to z' > 0 and z' < 0, respectively. This transformation leads to the related equation (Langer, 1931) (7.3.69) The solution of (7.3.69) is

where v = (2

+ x)

Hence, to first approximation

346

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

McKelvey ( 1 955) expressed the asymptotic solutions of a second-order turning point problem (i.e., c/. = 2 ) in terms of Whittaker's functions. A second-order turning point problem arises in the diffraction by elliptic cylinders whose eccentricities are almost unity (Goodrich and Kazarinoff, 1963), and in the solution of the Schrodinger equation (Voss, 1933). The first analysis of a second-order turning point problem was given by Goldstein (1931) using the method of matched asymptotic expansions as described in Section 7.3.1. 7.3.5. H 1G H ER APPROXI MAT 10N S So far in our presentation, only the first term in the asymptotic expansion has been obtained. There are four different approaches for the determination of the higher-order terms.

Lunger's Approach. The gist of this approach is always to relate the solution of the equation to be solved to that of some simpler but structurally similar problem that can be solved explicitly in terms of transcendental functions (Langer, 1949). The drawback to this approach is that it is unsuitable for numerical calculations because the coefficients of the asymptotic expansions are functions of the independent variable as well as the perturbation parameter. Moreover, the expansions are established using several transformations. Equivalent expansions can be obtained in an easier way by using Olver's approach as indicated later. CI1erry's Approach. In 1949 and 1950, Cherry developed a technique for obtaining the higher-order terms of a simple turning point problem, which has been transformed using the Langer transformation (Section 7.3.2) into d2V

cIz2

where

+ [-A% + Ag(2, A ) ] u = 0

(7.3.72)

(7.3.73)

In Cherry's analysis all g , with even n are missing. We assume a formal expansion of the form u

=

A(z;

A)5,[P3+(2;

A)],

i

=

I, 2

where 5, and 5, are the Airy functions of the first and second kind, respectively. Since

TURNING POINT PROBLEMS

7.3.

347

and

(7.3.72) becomes

Equating the coefficients of

5, and dtild4 to zero, we obtain (7.3.76)

A” + 1g + =0

(7.3.77)

_- 1

(7.3.78)

- 2)

A“$l$”

A

From (7.3.76) A-= hence (7.3.77) becomes

- 2)

+ A+“

=0

2A‘4‘

J

+I

+Ig +

(7.3.79)

To solve this equation we let (b = 2

+ A P $ , ( Z ) + I-”&)

+.

*

.

(7.3.80)

and equate coefficients of like powers of 1 to obtain equations for the successive determination of 4,. The first two equations are

24;

+ + go = 0 244 + = -g, $1

$2

whose solutions are

(7.3.81)

- z+;z - 241+;

The lower limit in these expressions was chosen so that at z = 0. Therefore v is given by the formal expansion

6,

and

$2

are regular

348

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

This formal expansion is a uniform approximation to the solution of the original problem except in small neighborhoods of the zeros of v , where Cherry used the following modified formula u =

Ai(A2/32)(l

+ A-2al + A%, + . . .) + @(A2/3z)(A-2b, + A-4b2 + . . .) dz (7.3.84)

If

w, 4= z: m

n=O

A-2ng2n(Z)

an and b, are determined from (7.3.83) by expanding = z and using the relation d2Aildz2= A2zAi.

+

4‘ and Ai(Az134)about

Oluer’s Approach. Olver (1954) proposed to determine a complete asymptotic expansion by assuming that

+

u = A ( Z ;A ) < ~ ( A ~ ’ ~ z )B ( Z ;A)<;(A~’~z)

(7.3.85) This form is the same as the final form of the expansion obtained by Langer (1949) and the modified formula (7.3.84) of Cherry. This expansion can also be regarded as an application of the method of composite expansions described in Section 4.2. Since 5; = A2zci, U’ = A’Ci ( A B’)[: B<; = (A’ A2zB)<, ( A B’)
+ + + + + +

+ A2B+ A2~B’)
U“ = (A”

Consequently, (7.3.72) becomes

(A” + A2B + 2A2zB’ + AgA)<, + (2.4’ + B” Equating the coefficients of A”

<$

+ AgB)(I = 0

(7.3.86)

51 to zero, we obtain

and

2A’ + B” + AgB = 0 + A2B + 2A2zB’ + AgA = 0

(7.3.87)

These equations are satisfied by formal expansions of the form A =

2 AXnAn(Z) rn

n=O

B

m

=

2 A-nBn(z)

n=l

(7.3.88)

7.3. where

TURNING POINT PROBLEMS

+ g,B, = 0 2zBI + B, + goAo = 0 2A;

349

(7.3.89)

n

2zBA+,

+ B,+l + goA, =

2

= B,,

-m=lgm&,

The solution of (7.3.89) is

n 2 1

A , = c o s h go(? l 2 ~ dT 7

(7.3.91 )

hence the solution of (7.3.90) is

where a n = + J : [ % Z ( ~ ) A O ( ~) pn(T)B1(7)1 dT bn =

i[

[JT

-

an(T)~l(T)

In the case of

BdT)AO(T)l J;

(7.3.93)

dT

Ag = L\ il-Zngn(2) io

n-0

(7.3.87) is satisfied by formal expansions of the form A = A,

+ IA-2nAn(~), 4)

n=l

A, = 1

(7.3.94)

m

B = 2 A-2nBn(z) n=l

Substituting this expansion into (7.3.87) and equating coefficients of like powers of A, we obtain 2Ak = - B i - gOB, 2zBh

+ B , = -A;-l

n-1

- 2: g,Bn-,

= an

m=l

n-1

- goAn-l - 2 gmAn-m-l m=l

(7.3.95) = p,

350

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

Their solutions are (7.3.96)

f z

An = ?J an(T)dT

Successive Langer Transformations. To determine higher approximations to the solutions of

d2Y + A2q(x)y = 0 dx2

(7.3.97)

where q(z) vanishes in the interval of interest, Imai (1948, 1950) proposed repeated application of the Langer transformation. This technique has been applied and extended considerably by Moriguchi (1959). For a simple turning point at x = p, we first introduce the Langer transformation

in (7.3.97) to obtain

- + [A22 d2v

dz2

where 6=x

-114

-

d2(x1I4)

-

-x

(7.3.99)

=0

S(Z)]U

-314

d (x ) 2

-114

dz2 dx2 Since 6 = O(1) and 1 is large, v is given approximately by

- + 1%= 0 d2u dz2

that is

(7.3.100)

v =&(A~/~z)

(7.3.101)

where C1 and 5, are the Airy functions of the first and second kind. To improve (7.3.101) we rewrite (7.3.99) in the same form as the original equation (7.3.97) by changing the independent variable from z to x1 according to

Zl= 2

- p1

(7.3.102)

where ,ulis the root of A2z - 6(z) = 0; that is 1%1

- S(p1) = 0

(7.3.103)

Then (7.3.99) can be written as

- + 12[alsl + R1(XJV d2v dx12

=0

(7.3.104)

7.3.

TURNING POINT PROBLEMS

351

where

(7.3.105) This equation has the same form as (7.3.97), hence an approximate solution can be obtained using the transformation g

x1

2

c,/mdT,

41

=-

Then (7.3.104) becomes

where

=

41 = 4 x 1

?

21

d'v, dz12

v =

+

+ MXl)

- Sl(Z1)]Ul

=0

(7.3.106)

(7.3.107)

(7.3.108)

A first approximation to u1 is the solution of d2vl dz,

-

Hence

+ A2ZlU1

=0

(7.3.109)

(7.3.110) Now transforming back to x and y yields an improved approximation to the solution of the original equation. From (7.3.106) u1

= 5a(P3z1)

two independent solutions of (7.3.97) are approximately given by 5i[122/3u:/3(~(x) - pl)]

where z(z) is defined in (7.3.98).

(7.3.112)

352

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

Higher approximation can be obtained by means of repeated application of the above procedure. 7.3.6. AN INHOMOGENEOUS PROBLEM WITH A SIMPLE TURNING POINT-FIRST APPROXIMATION In this section we determine a first approximation to a particular solution of the equation

d2Y + [A2q,(x) + q,(x)]y = A2G(x) dx2

(7.3.113)

for large 1 when ql(z) has a simple zero at x = p. If we divide (7.3.113) by A2 and let 1 00, we obtain --f

(7.3.114) as an approximate particular solution. This solution is singular at x = p unless G(z) has a simple or a higher-order zero at x = p. To determine a first approximation to a particular solution in the case of G(p) # 0, we first employ the transformation z =

$(XI,

g.32 '

to transform (7.3.114) into

=I

y=z

z -

U

JqdT)

- + (2% - b)u d2u dz2

= A"@)

(7.3.115)

(7.3.116)

where 6 is defined by (7.3.37), and g(z) =

(6' [x(z)l

G[x(z)l

(7.3.1 17)

Since 6 = O(1) and A is large, a first approximation to (7.3.116) is

fi + a2zu = ).("A dz2

(7.3.118)

To determine a particular solution, we write g(z) as the sum of two terms according to g(4 = g(0) k(4 - g(0)I (7.3.1 19)

+

and determine particular solutions corresponding to each term. A particular solution corresponding to the second term is given approximately by (7.3.120)

7.3.

TURNING POINT PROBLEMS

353

uniformly for all z if g(z) is differentiable at z = 0. To find a particular solution corresponding to the first term, we let E = A2/3z so that (7.3.118) becomes d2v tv = ~ 2 / 3 ~ ( 0 ) (7.3.121) with a particular solution

dE2

+

u2

where d2T

-

dE2

=

+ 6T = 1 ,

P~~(o)T(E)

(7.3.122)

1 T(E) = - as 161-

E

co

(7.3.123)

Thus T(E) may be represented in terms of Lommel functions (e.g., Watson, 1944, pp. 345-351) according to

m

l)(-t)" n!

= 3-2/313n/3p(" n=O

5 n=O

(- 1)"(3n)!

3"n!

E-'~~',

>

151 <

277 larg 61 < 3

(7.3.124)

where So,,,, denotes a specific Lommel function. Therefore a particular solution to (7.3.1 18) is approximately given by (7.3.125) Transforming back to x and y, we obtain

f ( z )=

qdX)

(x - P ) ~

Inhomogeneous problems with turning points arise in the analysis of boundary layer stability (Holstein, 1950) a n d thin elastic toroidal shells and bending of curved tubes (e.g., Clark, 1964). The technique presented above was developed by Holstein (1950), Clark (1958, 1963) and Tumarkin (1959). Steele (1965) obtained a single particular solution for an inhomogeneous second-order equation in terms of general Lommel functions SPev.

354

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

7.3.7. AN INHOMOGENEOUS PROBLEM WITH A SIMPLE TURNING POINT-HIGHER APPROXIMATIONS To determine a complete asymptotic representation of d2Y dx2

+ [A2f(x) + Ax(",

A)]?/

= A2G(z, 1)

(7.3.127)

wheref(z) has a simple zero at x = ,u and m

~(z, A) = 2 A-"x,(x),

G(x, A) =

n=O

m

2P G n ( x )

n=O

as

A -+

01)

we first employ the transformation z = +($),

?iZ3/'

to transform (7.3.127) into d2v dz2

-

where

4 ( w ,

=[m) dT,

+ [A2z+ A&,

A)

y = o/d@

A)]v = A2g(z, A)

= P4(2)x(z,A)

1

m

q(z,

n=O

(7.3.130)

P(z) = -

JG)

We assume that

A) = 2 A-"g,(z),

(7.3.129)

+ A-1P3(x)p"(2)

g ( z ( 4 , A ) = P3(z)G(z, A),

g(z,

(7.3.128)

m

A) = 2 A-"q,(z)

as 1 -+

n=O

01)

(7.3.131)

and restrict our attention to (7.3.129). We assume a complete asymptotic expansion of a particular solution of (7.3.129) and (7.3.131) to have the form v = C(z, A)

+ A 2 / 3 A ( ~A)z)T(F) , + A1l3B(z,A)?+(&

5

= A2/32

(7.3.132)

where T(E) is defined in (7.3.124) as the solution of T"

Since

+ ET = 1,

1

T = - as

F

Ill-+co

(7.3.133)

+ A213A'T + A1l3(AA+ B')T' + ABT = C' + AB + A213(A' - h B ) T + A1'3(AA+ B')T' - = C" + 2AB' + A2A + A213(A"- A2Az - AB - 2hB')T dz2 + ~ l / y 2 1 -~ 'A Z ~+B WIT' do- C' dz

d20

7.3.

TURNING POINT PROBLEMS

355

(7.3.129) becomes

[(A%

+ Iq)C + C" + 2IB' + A2A - A2g]+ A213[A"- IB - 2kB' + %A]T + A1I3[2AA'+ B" + IqBlT' = 0 (7.3.134')

In order for (7.3.134) to be an identity, each of the coefficients in square brackets must vanish; that is

(2%

+ Iq)C + A2A - A2g + 2AB' + C" = 0 2 k B ' + I B - A" - I q A = 0 2AA' + AqB + B" = 0

(7.3.135) (7.3.136) (7.3.137)

To solve (7.3.135) through (7.3.137), we assume formal expansions of the form A = 2 l-nAn(z), n=O

B =

I:A-nBn(~),

m

C =

'n=O

n=O

P C , , (7.3.138)

and equate coefficients of like powers of A to obtain

zc,

= gn

- An - 2Bb-1 - Ci-2

n-1

- 2 qn-&lCk k-0

(7.3.139) (7.3.140) (7.3.141)

where all coefficients with negative subscripts are defined to be zero. Letting n

we combine (7.3.140) and (7.3.141) into

-

2 Ji u; - iqoun = iMzAn-l - J z M,Bn-l

(7.3.143)

whose general solution is

where a,t and /?, are arbitrary constants and (7.3.145)

356

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

Since coefficients with negative subscripts are zero, (7.3.144) reduces for n=Oto (7.3.146) uo = (a, ip,)eie Hence A, = a, cos 8 - Po sin 8 (7.3.147) JZB, = sin 8 po cos 8

+

+

Since O(0) = 0, Do = 0 in order that B, be bounded as z -+0. Now (7.3.139) reduces for n = 0 to (7.3.148) zC, = go - A ,

In order that C, be bounded as z

-

0 (7.3.149)

AO(0) = go@) = a 0

Therefore the first approximation to u is

which reduces to (7.3.125) when qo = 0 (i.e., 8 = 0). In general, for B, to be regular at z = 0, we require that for C , to be regular at z = 0, we require that

Pn = 0, while

n = 0, 1, 2 , . . . (7.3.151)

Hence

- +.Io b ,

(7.3.150)

COS [ q Z )

B,(z) = a, sin 8(z)

+4

- 8(~)]kf,B,,-,dr cos [O(z)

(7.3.152)

- 8 ( ~ ) ] M + 4 , -dT~

- 4 l S i n [ 8 ( Z ) - 8 ( ~ ) ] h f $ ? ,d-r~

(7.3.153)

This general solution was obtained by Tumarkin (1959) and justified by Clark (1963).

A N INHOMOGENEOUS PROBLEM WITH A SECONDORDER TURNING POINT In their investigation of toroidal membranes under internal pressure, Sanders and Liepins (1963) encountered an inhomogeneous problem with a 7.3.8.

7.3.

TURNING POINT PROBLEMS

357

turning point of second order of the form

d2Y d x2 + A2q(r)y = A2G(z)

(7.3.154)

where q(z) has a zero of order two at x = p. To determine a first approximation to a particular solution of this equation, we let z =

$(x),

*Z2

in (7.3.154) and obtain

=I’d& P

d2v + (A22 dz2

d7,

y =U

J.

- S)u = il2g(z)

(7.3.155)

(7.3.156)

where 6 and g are defined by (7.3.37) and (7.3.117). Since 6 = O(1) and il is large, a particular solution to (7.3.156) is given approximately by

- + A22% d2u dz2

= A2g(z)

(7.3.157)

To find an approximate particular solution to (7.3.157), we express g(z) as the sum of three terms according to g(z) = g(0)

+ g ’ W + [g(4- g(0) - g’(0)zl

(7.3.158)

and determine particular solutions corresponding to these three terms. A particular solution corresponding to the last term is approximately given by (7.3.159) uniformly for all z if g”(0) exists. To find particular solutions corresponding to the other two terms in g ( z ) , we let 5‘ = jll/% thereby transforming (7.3.157) into d2v 6% = Ag(0) P g ’ ( 0 ) 5 (7.3.160) dE2 Sanders and Liepins (1963) defined the two functions T , ( [ ) and T 2 ( [ )by

+

+

In terms of these functions, a particular solution of (7.3.160) may be written as UZ = Ag(0)TdE) jl1’2g’(O)TdE) (7.3.162)

+

358

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

Therefore a particular solution of (7.3.157) is approximately given by v=

Az)-

2

- g’(o)z + Ag(O)T,(E) + n””’(o)T,(~) (7.3.163)

Transforming back to x and y, we can obtain a uniformly valid first approximation to the original equation. 7.3.9. TURNING POINT PROBLEMS ABOUT SINGULARITIES We consider the asymptotic expansions of the solutions of (7.3.164) for large 1 under the conditions

The problems considered in the previous sections (ro = 0 and a 2 0) are special cases of the present problem where ro # 0 and a may be negative. As z ---t p , (7.3.164) tends to ]y=o

(7.3.166)

Therefore we choose (7.3.167) as the related equation with solutions expressed in the form 0

= Z1’2$Yv(yZ@)

(7.3.168) J1 - 4r0 2 2+a where the cylindrical functions W v ( f )satisfy the differential equation

&=A,

p=- a + 2 ,

1 d%‘, -d2V, + + dt2 t dt

v=

(7.3.169)

The Bessel, Neumann, and Hankel functions ( J v ( t ) ,Y v ( t ) ,H v l ( t ) ,and H V 2 ( f ) ) are special cylindrical functions. Investigation of turning point problems about singularities was started by Langer (1935). The other principal contributors to the investigation of this problem were Cashwell (1951), Olver (1954), Swanson (1956), Kazarinoff and McKelvey (1956), Erdklyi (1960) , and Wasow (1965).

7.3. TURNING POINT PROBLEMS 359

To determine the asymptotic expansions of the solutions of (7.3.164), we introduce the transformation 2

= $(.I,

= Y().Y().,

y(4 =

@

(7.3.170)

thereby transforming (7.3.164) into d2v

(7.3. 7 1 )

In order that this equation be approximately identical to (7.3.167), we require that (7.3.17 2 ) V V = 4(4 so that (7.3.173)

Hence (7.3.171) becomes (7.3.174)

where (7.3.175)

(7.3.176) (7.3.177)

Hence a first approximation to u is given by (7.3.167) whose solution is given by (7.3.168). Therefore y is given approximately by

Higher approximations to (7.3.174) can be obtained by using Olver's approach (Section 7.3.5) by assuming that where 6, and

6,

+

= A(z, W J z ; 4 B(z, m z ; 4 are the independent solutions of (7.3.167).

(7.3.179)

360

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

7.3.10. TURNING POINT PROBLEMS OF HIGHER ORDER Most of the interest in turning point problems for differential equations of order higher than two arose from the hydrodynamic stability of parallel flows. The linear stability problem for a parallel flow can be reduced to the solution of the so-called Orr-Sommerfeld equation (e.g., Lin, 1955)

giV

- 2a2f + a45b = iaR{[U(y)- c ] ( q - a”) - U”(y)d}

(7.3.180)

for the disturbance amplitude $(y). In this equation, U(y) is the velocity profile of the undisturbed flow which is a known function. The parameters a and R are positive constants representing, respectively, the disturbance wave number and a flow Reynolds number. The parameter c is a complex constant whose real part c, determines the wave speed, while its imaginary part ci determines the damping or growth rate of the disturbance. This equation is supplemented by four homogeneous boundary conditions to form an eigenvalue problem for the eigenfunction 4 and the eigenvalues c, and ci if u and R are known. The system is unstable if ci > 0, stable if ci < 0, and neutrally stable if ci = 0. For large uR, two independent solutions of this problem can be obtained in the form (7.3.181) d = M Y ) (Ca-l+,(y) * *

+

+ -

while the remaining two solutions can be obtained in the form

-

4 = efd=R: [(v- c)-”~

+ (aR)-’/”fi(y) + . . .I

(7.3.182)

where 5 = J:o J i ( U - c) dy. The above solution breaks down near the zeros of U - c, which are turning points of (7.3.180). First-order uniformly valid asymptotic solutior,s were obtained for (7.3.180) by Tollmien (1947) and Wasow (1953), while complete uniformly valid expansions were obtained by Langer (1957, 1959a), Rabenstein (1959), and Lin and Rabenstein (1960, 1969). K. Tam (1968) obtained uniformly valid expansions for (7.3.180) using the method of multiple scales. Turning point problems for equations of the nth order were treated by Sibuya (1963a, b). 7.4.

Wave Equations

In this section we describe some of the available techniques for determining approximate solutions of linear wave problems and their related elliptic problems. In our description of these techniques, we use the wave problem c2(r)V2u - utt - ot(r)u = cz(r)g(r)eawf (7.4.1)

7.4.

WAVE EQUATIONS

361

If we let we obtain

0%

u = u(r)eiwt

(7.4.2)

+ k2n2(r)u= g(r)

(7.4.3)

where k and n are the wave number and refractive index given by (7.4.4)

with co a reference speed. In this problem we assume that g is deterministic, while n can be a random function so that the results may be applied to wave propagation in a random medium. For a constant n the homogeneous problem admits a plane wave solution of the form = Aeink.r (7.4.5) with constant A , while the inhomogeneous problem has the integral (7.4.6)

where 5. is a variable vector ranging over the scattering volume V . However, if n is not a constant, we seek asymptotic expansions to the solutions of (7.4.3). The choice of an asymptotic technique to obtain an approximate solution depends on the value of k and the spatial variation of n. If n deviates slightly from a constant, we can use the so-called Born expansion developed by physicists and the Neumann expansion developed by mathematicians, renormalization techniques, or the Rytov technique. If k is large or n is a slowly varying function of position, we can use the geometrical optics technique. Although these techniques were developed for deterministic problems, they can also be used for stochastic problems. For the latter problems we also describe the so-called smoothing technique which is the counterpart of the method of averaging discussed in Chapter 5. For an account of the applications of these techniques and for more references, we refer the reader to the books by Chernov (1960), Tatarski (1961), Babich (1970, 1971), and the survey article by Frisch (1968). 7.4.1. THE BORN OR NEUMANN EXPANSION AND THE FEYNMAN DIAGRAMS This technique is applicable when n deviates slightly from a constant. In this case we assume that k and n have been normalized so that the constant part of n is unity, thereby allowing n2 to be written as n2

=1

+ q(r)

(7.4.7)

362

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

x

x

with E small and = O(1). For the statistical problem we assume that is a centered random function of r so that its mean, denoted by (x), is equal to

zero. To obtain a Born expansion (Born, 1926), we let u =

2 W

(7.4.8)

Emum

m=O

Substituting this expansion into (7.4.3), using (7.4.7), and equating coefficients of like powers of E , we obtain W O )

=g

L(u,) = -k2~u,-1 where the operator L is defined by

L = V2

(7.4.9)

for m

2

1

(7.4.10)

+ k2

(7.4.11)

Equations (7.4.9) and (7.4.10) can be solved successively. Thus for a given m,the right-hand side of (7.4.10) is known from the previous solution. Hence its solution is given by

u,(r) = -~'SFX(13u,-,(r~)G0(r; rm)dr,,

m

21

(7.4.12)

where r, is a variable vector ranging over the scattering volume V , and Go is the free-space Green's function

n= -

If g

3

eikb-Ei

47r Ir - !3 0, (7.4.9) admits the following plane wave solution uo = Aeik.1

(7.4.13)

(7.4.14)

where A is a constant. Then from (7.4.12) (7.4.15) u2 = A k ' ~ v ~ ( r ~ ~ ( r , ) ~ t k . r lrz)Go(r2; G o ( r ; rJ

= a(-l)"k""Jvx(rJx(rd

*

*

dr, dr,

(7.4.16)

x@,)

---

r,-J Go(ra;r,) dr, dr, * . * dr, (7.4.17) In this expansion, a, is called the first Born approximation, while emurnis called the mth Born approximation. &k.r

lGo(r; r,)Go(r,;

7.4.

WAVE EQUATIONS

363

If x is a centered random function of r, the mean of u can be obtained by averaging (7.4.8). The result is ( u ) = Aeik"

+ A.'k4Sy(~(rl)~(rz))e"IG.(r;r,)G,(r,;

+ ~ ( - E ) - k z m / V(x(rl)x(rz) . eik.rl

G,(r; rm)Go(rm;rm-,)

'

r,) dr, dr,

+ ...

x(rrn)>

1

..

Go(rz;rl) dr, dr, . . dr,

+. .

(7.4.18)

-

The averaged quantities (x(rl)x(r2) * x(r,)) depend on the configuration of the points rl, r,, . . . ,rm because for most random media there exists a correlation length I (ie., the values of x at points of separation larger than 1 are uncorrelated). To express the dependence on the correlation length, we expand these averaged quantities in the following cluster (cumulant) expansions (x(r,)x(rz)> = Wl, r2) (x(r,)x(rz)x(rs)>= Nr1, r2. r3) (x(rl)x(rz)x(r3)x(r4>)= W1,rz)R(r3,r4) -k R(rl, r4)R(r2?

(X(rl)x(r2) . *

'

X(rm)>

=

2

kl+...+k,=m

+ W1, r3)R(rz,r4) r3)

R(gl, .

x R(ql, . . .

* *

7

+ R(rl, 9

(7.4.19)

r2, r3r r4)

gkl)R(
a

'

.

3

Ckr) ' '

.

qk,)

where ki 2 2. Thus the summation in the last equation is extended over all possible partitions of the set r,, r2, . . . ,r, into clusters of at least two points. If x is a centered Gaussian random function, all correlation functions vanish except the two-point correlation functions. Using (7.4.19) in (7.4.18), we obtain an expression for (u) which exhibits the dependence on the k-point correlation functions. If g f 0 the particular solution of (7.4.9) is

$64 ASYMPTOTIC

SOLUTIONS OF LINEAR EQUATIONS

or in operator form U,

(7.4.22)

= (-k2MX),Mg

where the operator M is defined in (7.4.20). Therefore u(r) = M g

+ 2 cm(-k2MX)"Mg m

m=l

(7.4.23)

This series is also called the Neumann series by mathematicians. It can also be obtained by converting (7.4.3) into the integral equation u = M g - ck2MXU and solving it by iteration. From (7.4.23) we can define Green's function G for (7.4.3) as

G =

2 (G,,9)mGo

m=o

with 9, -+-ck2X(r)+(r)

(7.4.24)

(7.4.26)

This series was represented by Frisch (1968) by what he called "bare" diagrams using the following conventions: Go is represented by a solid line and 9 is represented by a dot. Then G is represented by the diagram series G=

-+-+

r

ro r

rl

ro

r

r1

ro

+...

(7.4.27)

This series has a multiple scattering physical interpretation. The mth term corresponds to a wave which propagates freely from ro to r,, is scattered at rl by the inhomogeneities,propagates freely to r,, is scattered at r,, and so on. Frish (1968) represented the double Green's function (the tensor product of G and its complex conjugate) by G0

G = G(r; ro)G(E; b)

7.4,

WAVE EQUATIONS

in the following “bare” double-diagram series

t

l o t

r

r1

€0

ro

+ t

€1

€

€1

365

€0

...

( 7.4.28)

€0

where each double diagram is the tensor product of the operator corresponding to the upper line, with the complex conjugate of the operator corresponding to the lower line. If x is a centered random process, ( G ) can be represented by the following “dressed” diagram series (Frisch, 1968).

+...

(7.4.29)

In this diagram series the following conventions were used: (1) Points belonging to a given cluster were connected by a dotted line. (2) To every “bare” diagram involving k factors of we associated as many “dressed” diagrams as there are different partitions of rl, r2, . . . , rk into clusters of at least two points. (3) To calculate a “dressed” diagram, the solid lines are replaced by G o , the cluster of dotted lines ending at rl, r2, . . . , rs by the factors

x,

(--k2)’R(rl, r2,. . . , r8),

366

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

and the integration is performed over all intermediate points. Thus

(7.4.30)

Similarly, we can express the covariance (G @ G) in the following “dressed” double-diagram series r-.

where, for example r

r 1;

+

. - *

(7.4.32)

ro

I

I

I I

(7.4.33)

For a Gaussian random function, only two-point correlation functions are nonvanishing, hence diagrams such as 3 and 7 in (7.4.29) and 5 in (7.4.32) vanish. Representation of formal perturbation series by diagrams was first used by Feynman (1948). These diagrams are called Feynman diagrams and are

7.4.

WAVE EQUATIONS

367

widely used in statistical thermodynamics (see, for example, Prigogine, 1962), the many-body problem (e.g., Van Hove, Hugenholtz, and Howland, 1961), and quantum electrodynamics (e.g., Balescu, 1963). The first use of Feynman diagrams for solving linear stochastic equations was made by Kraichnan (1961). Feynman diagrams were introduced for the solution of wave propagation in random media by Bourret (1962a, b), Furutsu (1963), and Tatarski (1964) for the case of a Gaussian process, and by Frisch (1965, 1968) for the general case. The expansions obtained in this section were shown to be divergent by Frisch (1968). Moreover, he also showed that these expansions contain secular terms which restrict the validity of these asymptotic expansions to small arguments. Since (X(rl)X(rJ * . X(rz,)) is the sum of 1.3.5 * . (2m - 1) two-point correlation functions, this number increases rapidly with increasing m and this is another reason why (7.4.29) and (7.4.32) are divergent. Shkarofsky (1971) modified Born's expansion for backscattering from turbulent plasmas to obtain saturation and cross-polarization. A technique for making such expansions more uniformly valid was developed by Rayleigh (1917) to make a similar expansion which he obtained for first scattering from a thin slab valid for many slabs. This technique is called renormalization, and it has been developed further and extended by several researchers, as discussed in the next section. 7.4.2. RENORMALIZATION TECHNIQUES To illustrate the nature of the nonuniformity that might arise in the Born (Neumann) expansion and explore methods of remedying it, we discuss the simple example u" k2(r)u = 0, ~ ( 0= ) 1 (7.4.34)

+

where k is constant and given by N

(7.4.35)

k =ze"kn n=O

Outgoing waves are given exactly by

N

(7.4.36)

However, the Born expansion can be obtained from (7.4.34) to be

+ e3(ik,x - k,k,z2 - +ik13z3) + .

*

*]

(7.4.37)

368

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

It is clear that this expansion is valid only for short distances, and it breaks down when klx = O ( d ) or larger. The origin of the nonuniformity can be seen by comparing this Born expansion with the exact solution (7.4.36). The Born expansion can be obtained from the exact solution by N

expanding exp (ix 2: 8""') N

n=l

in Taylor series in terms of ix

z:Pk,, then exN

n=l

panding (2Enk,Jnzin powers of E and collecting coefficients of like powers of n=l

Although the Taylor series expansion of exp+ in terms of 4 converges a finite number of terms uniformly and absolutely for all values of cannot be used to represent exp 9 with a given accuracy for all values of Consequently, any uniformly valid expansion of expd for all 4 must be expd itself. Thus to determine an expansion from (7.4.37) valid for z as large as O(E-~),we must sum the sequence E.

+,

1 + iek,x

+ - . . + (kk1x)*/m! -I-

+.

* *

Its sum is exp (iklz), and (7.4.37) becomes u = exp [i(ko = exp [i(k,

+ rkl)z]%=om. 2 $( i xn=z 2 Pk.) m

N

m

+ akl)x](l + iE2k,z + ir3k,z - 4e4k,2z2 +

a)

(7.4.38)

This expansion breaks down when k p = U ( c * ) . To increase the range of validity of this expansion to values of x = U ( E - ~ )we , must sum the sequence m

2 (ia2k2x)m/m!. An effective technique of summing these sequences without

m=O

knowing their explicit functional dependence is the method of multiple scales of Chapter 6. Another technique for effecting the summation of the secular terms is the renormalization technique. The renormalization technique was originally developed by Rayleigh (1917) to generalize his first scattering from a thin slab to scattering from many slabs. He obtained an expansion of the form

for first scattering from one slab. To obtain a solution valid for many slabs, he recast this expansion into an exponential; that is

+ ep)~l

u = exp [i(ko

(7.4.40)

7.4. In this manner he effectively summed the sequence

WAVE EQUATIONS

2

I

369

(icpx)"/m! of secular

m=l

terms. The process of summing expansions to make them "more" uniformly valid is called renormalization. This technique was rediscovered by Pritulo (1962) as described in Section 3.4. To make a more uniformly valid expansion from a two-term Born expansion u = u, E U ~ we , recast it into the following exponential

+

(7.4.41) (7.4.42)

+

where A = A , exp [ e Real (ul/uo)] and S = So c Imaginary (ul/u,,). This renormalization technique has been extended to obtain the kinetic equations for weakly nonlinear systems (see, for example, Van Hove, 1955, 1957; PrigoginC, 1962; Balescu, 1963, and Al'tshul' and Karpman, 1966). According to this technique, sequences of secular terms are separated and summed with or without the use of Feynman diagrams. The summation of the principal sequences of secular terms leads to quasi-linear equations. The renormalization technique has also been widely used in the study of wave propagation in random media (e.g., Tatarski, 1961, Chapter 6; Keller, 1962; Karal and Keller, 1964). Thus to determine a more uniformly valid expansion for (G) of (7.4.29), we recast it into an exponential (G) = G,ev

(7.4.43)

Hence the iirst renormalization gives y=-

where G, =

G2

GO

G0(r.> r2)G 0(r2 ,. r1)G,,(rl; rn)R(rl; r2) dr, dr,

(7.4.44)

(7.4.45)

Diagrammatic summation techniques have been used to determine renormalization equations to any order by Bourret (1962a, b), Furutsu (I 963), Tatarski (1 964), and Frisch (1965). Tatarski (1 964) arranged the diagrams for (G) and (G @ G) in such a manner that he could recognize that they were the Neumann expansions of two integral equations with two kernels having an infinite nhmber of terms. For a centered Gaussian index of

370

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

refraction, the diagrammatic expansion of (G) is

12

13

1A

To obtain an integral equation for (G) with a kernel consisting of an infinite series, we look at the topology of the diagrams in (7.4.46).Let us make the following definitions.

(1) A diagram without terminals is a diagram stripped of its external solid lines, such as diagram 2 without terminals f i and diagram 4 without terminals (2) A diagram without terminals is connected if it cannot be cut into two or more diagrams without cutting any dotted lines. Diagrams 4 , 5, 10, 11, and 13 through 20 are connected, while diagrams 3, 6, 7, 8, 9, and 12 are unconnected. Unconnected diagrams can be factored; for example,

m.

I'

--'\,

I

1

,--. \

following five diagrams

can be written as the product of the

7.4. (3) The mass operator Q denoted by diagrams contributing to (G); that is

0

WAVE EQUATIONS

371

is the sum of all possible connected

(7.4.47)

.

All unconnected diagrams composed of two connected diagrams occur in the sum of diagrams -, while all unconnected diagrams composed 0 . of three connected diagrams occur in the sum of diagrams From this we note that ( G ) , which is denoted by -, is governed by the following Dyson equation (Tatarski, 1964)

--

+

-=-

or in analytic form

(7.4.48)

I

+

(G(r; ro)) = G o @ ;ro>

-

G&; rl)Q(rl; r2)(G(rz; ro)) dr,

(7.4.49)

A similar equation has been used extensively in quantum electrodynamics, quantum field theory, and the many-body problem, and it was first introduced by Dyson (1949). If x is homogeneous, the mass operator Q is invariant under translations, hence it is a convolution operator whose Fourier transform Q(k) is a multiplication operator of ordinary functions. Hence, taking the Fourier transform of (7.4.49), we obtain

because Go(%) =

(7.4.51)

k2 - 'K

Solving (7.4.50)for ( G ( x ) ) , we obtain

Thus if Q is known, (G) can be obtained by inverting ( G ( x ) ) . However, the exact expression for Q is as difficult to find as ( G ) .Thus one approximates Q. The simplest approximation based on the Dyson equation retains only the first term in the mass operator; that is -I-

+

r-.

I'

'\

(7.4.53)

372

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

or in analytic form W r ; ro)) = Go@;ro)

+ c Z k 4Gs (r.,r

; rz)Nrl ;d(WZ ; ro)) d r drz ~

(7.4.54) This equation is called the first renormalization equation, and it was introduced in diagrammatic form by Bourret (1962a, b). It should be mentioned that (7.4.49) and (7.4.54) cannot be solved by iteration, because this would lead to secular terms. An approximate solution to (7.4.54) is given by Varvatsis and S a n e r (1971) where

(G(r; ro)> = Go(r; ro)e'"z'Go

(7.4.55)

Gz = k4 Go(r; rl)Go(rl; r2)Go(rz;ro)R(rl; rz) drl dr,

(7.4.56)

s

This solution is valid for homogeneous as well as inhomogeneous random media. It is the same solution that would be obtained if we had recast (G) = Go 8 G Zinto an exponential. Using diagrammatic techniques, Tatarski (1964) and Frisch (1968) obtained the following Bethe-Salpeter equation for a centered Gaussian and general refractive index

+

=

-+

where the intensity operator

M

I>a

(7.4.57) consists of all connected diagrams

without terminals in the expansion of (G @

c> ; that is

'L-/

(7.4.58) This equation was first introduced by Salpeter and Bethe (1951) for relativistic bound-state problems.

7.4.

WAVE EQUATIONS

373

7.4.3. RYTOV'S METHOD To obtain an approximate solution of

+

+

V2u k2[1 q(r)]u = 0 (7.4.59) we follow Rytov (1937) by assuming that (see also Tatarski, 1961, pp. 121-128; Chernov 1960, pp. 58-67) u = ev

(7.4.60)

thereby transforming (7.4.59) into Now we assume that y possesses the following asymptotic expansion m

(7.4.62) Substituting (7.4.62) into (7.4.61) and equating coefficients of like powers of E, we obtain (7.4.63) V2vo Vyo . Vyo = -k2 V2y1

V2y, -k

+ + 2 V y 1 . VyO = -k2X

2Vy;

m=O

(7.4.64)

n 22

VynPm= 0,

(7.4.65)

These equations can be solved successively. The resulting expansion m

corresponds exactly to the Born expansion u =

2 P u n of (7.4.59). In fact,

n-0

the Rytov expansion can be obtained from the Born series by recasting the latter into an exponential. Setting \

(7.4.66) expanding the exponential for small z, and equating coefficients of like powers of E , we obtain for the first terms u, = evO,

u1 = eWOyl, u3

= eYy3

u2 = evo(yz

+ YlY2 + 3Y13)

+ ty?) (7.4.67)

Thus the Rytov expansion is a renormalized Born expansion, hence it should be more uniformly valid than the Born expansion. However, this conclusion is controversial (Hufnagel and Stanley, 1964; deWolf 1965, 1967; Brown, 1966, 1967; Fried 1967; Heidbreder, 1967; Taylor, 1967; Strohbehn, 1968; Sancer and Varvatsis, 1969, 1970).

374

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

THE GEOMETRICAL OPTICS APPROXIMATION The object is to obtain an asymptotic solution for

7.4.4.

+

(7.4.68) V2u k2n2(r)u = O for large wave number k (ix., small wavelength 1 = 2a/k). For large k, (7.4.68) has an asymptotic expansion of the form (Keller, 1958) =pS(r)

m

2 (Wrn4nW

(7.4.69)

m=O

Substituting (7.4.69) into (7.4.68) and equating coefficients of like powers of k, we obtain VS * VS = n2(r) (eiconal equation) (7.4.70) 2 vs * vu, 2 VS Vu,

is

+ uov2s = 0 + umV% = -V2u,1

(transport equations) for m 2 1

(7.4.71) (7.4.72)

Equation (7.4.70) can be solved using the method of characteristics; that

* do

=21vs

(7.4.73)

where o is a parameter and 1 is a proportionality function. Elimination of VS from the first and third equations gives

->

dr da 21da -(-

=2 ~ n ~ n

(7.4.74)

Then the solution of the second equation in (7.4.73) is (7.4.75)

where r(o) is the solution of (7.4.74) subject to the initial conditions r(aO)= r,, dr(ao)/da= io. Choosing 21 = n-l and u to be the arc length along the rays, we rewrite (7.4.74) and (7.4.75) as

-[n(r(o)) d do

1-'

= Vn

da

s = so + r n [ r ( ~ ) ]d7 Jso

(7.4.76) (7.4.77)

7.4. Equation (7.4.71) becomes

WAVE EQUATIONS

375

along the rays. Its solution is (7.4.78) Similarly, the solution of (7.4.72) is

where c is a constant to be determined from the initial conditions. The expansion obtained in this section is not valid at a caustic (i.e., an envelope of rays), shadow boundaries, foci of the rays, and source points. In such regions neighboring rays intersect and the cross-sectional area of a tube of rays becomes zero. Since the energy is conserved in a tube of rays, the amplitude of the field must be infinite in these regions. The unboundedness of the field at a caustic is shown below, and an expansion valid everywhere including a caustic is obtained in the next section. To show the breakdown of the expansion of this section at or near a caustic, we specialize it to the case n(r) = 1. In this case the rays are straight lines according to (7.4.76), and S = So - oo o according to (7.4.77). Now we express the solution of (7.4.71) and (7.4.72) in terms of a coordinate system with respect to the caustic which we assume to be smooth and convex. Figure 7-1 shows a point P outside the caustic and two rays passing through it. If we assign a direction to the caustic, this induces a direction for each ray which must be tangent to it at some point. Thus each point P outside the caustic lies on two rays-one has left the caustic and the other is approaching the caustic.

+

Figure 7-1

376

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

In two dimensions we let s measure the arc length along the caustic and u measure the length from the point of tangency to the point P. Thus the position of P is determined by s1 and ul or s2 and 02.In terms of these coordinates, the position vector r of P can be expressed as

r = ro(s)

+ ue,

(7.4.80)

where r = ro(s) is the equation of the caustic, and el = dro/ds is a unit vector tangent to the caustic. Differentiating (7.4.80), we have

+ el d o + -o e2 ds

dr = el ds because de,/ds = p-le, where the variables from u and s to

P

is the curvature of the caustic. Changing

p-I

5 = S

(7.4.81)

q=s+u

we obtain

71-6 dr =

Hence

dv +

e2 d t

(7.4.82)

P

af + -P af

Vf = -el all 11Since from (7.4.81)

17 -

tate2

(7.4.83)

"1

+---v - t a ta v[- t a E

ta7

(7.4.84)

we rewrite (7.4.83) and (7.4.84) as

v f =-afe l au

+

(-asaf - ;-e2 -

+ oo,then S(s, a) = s + a

If we take the special case S(s, uo) = s

(7.4.86) which is double-valued corresponding to either the ray that has left or the one that is approaching the caustic. Using (7.4.85) and (7.4.86) in (7.4.78)

7.4.

WAVE EQUATIONS

377

and (7.4.79), we obtain (7.4.87)

-

which are unbounded as G 0 (i.e., a caustic). A modified expansion valid everywhere including the neighborhood of the caustic is obtained in the next section. 7.4.5. A UNIFORM EXPANSION AT A CAUSTlC To determine an expansion valid at the caustic, we first must determine the size of the region of nonuniformity and the form of the solution in this region. To do this we let u = y(r, k)eiks"' (7.4.89) in (7.4.68) with n(r) = 1 and obtain k Z y [l - (VS)']

+ ik(2VS . V y + y V 2 S ) + V 2 y = 0

(7.4.90)

Assuming that the k 2 term is the leading term, we obtain

VS . V S = 1 ik(2VS. V y

(eiconal equation)

+ yV2S) + V 2 y = 0

We take the solution of (7.4.91) as S = s

+

0.Hence

(7.4.91) (7.4.92)

(7.4.92) becomes

To analyze (7.4.93) near the caustic, we introduce the stretching transformation T = k"a, 1> 0 and obtain

The parameter 1, is determined by requiring that the highest power of k within the braces (terms neglected in the leading term of the straightforward expansion of the previbus section) be equal to the power of k in the first

378

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

term (is., the leading term in the ray expansion). That is, we let 1+A=41 Then as k

---f

or

A=&

(7.4.95)

co, (7.4.94) tends to

(7.4.96) Letting 3

2

p = e-fr / 3 p V(Z), we obtain

2 7

z =P

(7.4.97)

FG

(7.4.98) whose solutions are the Airy functions A i ( - z ) and Bi(-z) (see Section 7.3.1). Had we analyzed the behavior of the solution near a shadow boundary, we would have found that it must be represented by Weber or parabolic cylinder functions. Now we can either match this inner expansion with the outer expansion obtained in the previous section and with another outer expansion inside the caustic (Buchal and Keller, 1960), or form a single uniformly valid expansion imitating the expansions for turning point problems (see Section 7.3). Following Kravtsov (1964a, b), Ludwig (1966), and Zauderer (1970b), we assume an asymptotic expansion of the form u = e"fl(r'(g(r, k)V[k2/3+(r)]

+ rk1/3 1 h(r, k)v'[k2'3+(r)]]

(7.4.99)

where 8 and 4 are determined from the analysis and V(z)is given by (7.4.98). Substituting (7.4.99) into (7.4.68), using the fact that V" zV' V = 0, and equating the coefficients of V and V' to zero, we obtain

+

+

+ +(V+)2 - 13 - 2k2+hV8 . V+ + ik[2V8. Vg + gV28 + 2 4 V + . Vh + +hV2+ + h(V+)2] + v2g = o

-k2g [(V8)2

-k2h

[(vq2+ + ( v + y

+ ik[2V+

*

Vg

- 11 - kZgve.

v+

+ gVz+ + 2V8 . Vh + hV28] + V2h = 0

(7.4.100) (7.4.101)

The coefficients of k2 in (7.4.100) and (7.4.101) vanish if (7.4.102)

7.4. WAVE EQUATIONS 379

To solve the resulting equations, we let

g(r, k ) = go@) h(r, k ) = h,(r)

+ k-’gl(r)

+..

+ k-’h1(r) + . . .

(7.4.103)

and equate coefficients of like powers of kl. The first-order equations are

2V8 . Vg,

+ goV28 + 2 4 V 4 2 V 4 . Vg,

*

Vh,

+ go”’+

+ +h,V2+ + II,(V$)~ = 0 + 2Vf3. Vh, + h,V28 = 0

(7.4.104)

Equations (7.4.102) are equivalent to the eiconal equation (7.4.70), while (7.4.104)are equivalent to the transport equation (7.4.71).Similar expansions were obtained by Fowkes (1968, Part 11) by using the method of multiple scales. The system of equations (7.4.102)is a nonlinear system of equations which is elliptic where 4 < 0 (shadow region), hyperbolic where 4 > 0 (illuminated region), and parabolic where 4 = 0 (caustic curve or surface). For a convex analytic caustic, the systems (7.4.102) and (7.4.104) can be solved by expanding 8 and 4 in power series in terms of a coordinate system with respect to the caustic. To see the connection between the system (7.4.102) and the eiconal equation (7.4.70), we multiply the second equation in (7.4.102) by f2& (illuminated region where 4 > 0 is considered) and add the result to the first of these equations to obtain

(ve f J$V+)2

or

=

1

(VS)2 = 1

where

(7.4.105)

s* = 6 f &$3/2

(7.4.106)

Similarly, we multiply the second equation in (7.4.104) by &dT, add the result to the first of these equations, and use VO * V 4 = 0 to obtain where

2T-s+ vy* ’

+ [V2S* 7 $+-”2(v+)”y* Y* = go f

43 ho

=0

(7.4.107) (7.4.108)

Equation (7.4.107) differs from the transport equation (7.4.71) only by the term &(1/2)4-1/2(V4)2 which makes the coefficient of y* bounded near the caustic. Replacing V and V’ by their asymptotic expansions for large argument, we recover the geometrical optics expansion of the previous section (4 > 0, illuminated region). Replacing V and V’ by their asymptotic expansions for

380

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

large argument for the case q4 < 0, we obtain an expansion in the shadow region which can be interpreted in terms of complex rays, phases, and transport coefficients. Uniform asymptotic expansions for wave propagation and diffraction problems were reviewed by Ludwig (1970b) and Babich (1970, 1971). The role of coordinate systems in rendering the expansions uniformly valid was investigated by Zauderer (1970a). A single uniformly valid expansion for a point source problem was obtained by Babich (1965), while three matched expansions for this problem were presented by Avila and Keller (1963). Problems of diffraction by convex objectswere treated by Zauderer (1964b), Buslaev (1964), Grimshaw (1966), Ludwig (1967), and Lewis, Bleistein and Ludwig (1967) among others. Diffraction by a transparent object was analyzed by Rulf (1968). Uniform expansions for the problem of diffraction near the concave side of an object (whispering galley modes) were obtained by Kravtsov (1964b), Matkowsky (1966), and Ludwig (1970a) among others. Characteristic transition regions in which two caustics are near each other are analogous to second-order turning points. Their uniform expansions involve Weber or parabolic cylinder functions. Some of these problems were treated by Kravtsov (1965), Babich and Kravtsova (1967), Weinstein (1969), and Zauderer (1970a, b). Diffraction by a thin screen (Fresnel diffraction) was investigated by Wolfe (1967), Kersten (1967), and Ahluwalia, Lewis, and Boersma (1968). Multiple transition regions arise from the tangential intersection of two or more caustics and shadow boundaries, such as near terminated caustics (Levey and Felsen, 1967), cusps on caustics (Ludwig, 1966), and points of diffraction for smooth objects (Ludwig, 1967). Problems involving transition regions were treated by Zauderer (1964a, 1970a), Fock (1965), Rulf (1967), and Bleistein (1967) among others. 7.4.6. THE METHOD OF SMOOTHING

To apply this technique to (7.4.3) with n2(r) a centered random function,

we first convert it to the integral equation

u = M g - Ek2MXU where the operator M is defined by

(7.4.109)

(7.4.110) with Gothe free-space Green’s function. Frisch (1968) pointed out that many other linear equations of mathematical physics, such as the Liouville equation

7.4.

WAVE EQUATIONS

381

for an ensemble of classic interacting particles, the Hopf equation for turbulence, and the Fokker-Planck equation, can be put in the above integral equation form. In general, we are not interested in the function u but in its projection Pu on a subspace of the original space over which u is defined. For example, in the problem of wave propagation in a random medium, Pu = ( u ) , while in the case of N interacting particles, Pu is the N-particle velocity distribution function obtained by integrating u over all position coordinates. Let Pu = u,,

and

u, = ( I

- P)u

(7.4.1 11)

For wave propagation in a random media, u, is the coherent part of the field (mean part) while u, is the incoherent part (fluctuating part). For a deterministic g and a centered random x Pg = g ,

PM = M P ,

PXP = 0

(7.4.1 12)

Applying the operator P from the left on (7.4.109), we obtain U,

Applying the operator I U,

= -ek2M(I

=

M g - ck2MPX(Pu - ck2MPXUz

= Mg

+ u,)

(7.4.1 13)

- P from the left on (7.4.109) also, we obtain

- P ) ~ ( u+ , u,) = -d?M(Z

- P)xuC - ~ k ' M ( 1- P ) p ,

(7.4.114)

We solve (7.4.1 14) by formal iteration for u, in terms of u, to obtain tli =

m

2 [--k'M(I

n-1

Substituting for

U,

- P)x]''u,

(7.4.115)

into (7.4.1 13) gives

Since u, = Pucr this equation can be written as where

u, = Mg

+ MQu,

Q = - z € k 2 P X [ - c k 2 M ( Z - P)x]"P n= 1

(7.4.1 16) (7.4.1 17)

For random operators (7.4.116) is just the Dyson equation (7.4.49) with Q the mass operator. This technique was called the method of smoothing by Frisch (1968) because it is the counterpart of the method of averaging (Chapter 5) in which the dependent variables have short- and long-period parts. This

382

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

technique was first introduced by Primas (1961), Ernst and Primas (1963), and Tatarski and Gertsenshtein (1963) for random equations, and by Zwanzig (1964) for the Liouville equation. The first-order smoothing is given by (7.4.116), with n = 1 in (7.4.117); that is Q = E W P ~ M (-I P)xP = c2k4PxMxP Hence u, = Mg e2k4MP~MxPu, (7.4.118)

+

which coincides with the Bourett equation (7.4.54) for linear random media, and with the Landau equation for the Liouville problem (see, for example, PrigoginC 1962). For linear random media this first approximation was also obtained by Keller (1962) and Kubo (1963). Exercises 7.1. Determine asymptotic expansions for large x for the solutions of

(c) x2y” + xy’ + 2.(

- n2)y

(d) Y“

1 + (I +; y):+

( e ) y”

+ xy‘ + y = 0

=0

=o

7.2. Determine asymptotic expansions for large x for the solutions of

+

(b) y” f (gx 1)y = 0 (c) y“ xy’ yx3y = 0 (d) y” + ~-‘/~y‘ + X-~Y= 0

+

+

7.3. Consider the equation

y” - Ps-2y = 0

(a) Show that the exact solution is

y = azm1 + b P z where m1,2 = 4 f %‘A2 + $. (b) Determine the first WKB approximation. (c) Compare this approximation with the exact solution (Jeffreys, 1962).

EXERCISES

383

7.4. Determine a first approximation to the eigenvalue problem

+ t Y ( z ) u = 0,

u”

f(z)

u(0) = u(1)

for large 1.

>0

=0

where Determine an equation for 4 and then determine the 7.6. Determine a uniform asymptotic expansion for ii

+

W2(Ct)U

+n

(Brull and Soler, 1966).

=f(ct)

where E is a small parameter and f is a bounded nonperiodic function of t. 7.7. Consider the problem ii

4

+ w 2 ( c t ) U = k cos 4

where = t ( d ) .Determine uniform asymptotic expansions for the cases (a) I is not near w , and (b) 1 = w for some t = to > 0 (Kevorkian, 1971). 7.8. Consider the problem ii

+ wz(cr)Jri- w,(ct)wz(ct)u

- i[W1(Et)

=f ( E t )

where overdots denote differentiation with respect to t and c is a small parameter. Determine uniform asymptotic approximations to the solution of this equation if (a)f(d) = 0, (b)f(rt) is a bounded nonperiodic function oft, (c)f(ct) = k exp (ie) with 4 = A ( c t ) # w , , and (d) f ( r t ) has the same form as in (c) but t = w1 at some I = to.

7.9. Determine a uniform asymptotic expansion for the general solution of 2y”

+ 4 2 2 + 1)y’ + 2xy

= 0,

0

7.10. Bessel functions J,(nz), Y,(nz), H ~ l * z ) ( n are x ) solutions of the differential equation x2y“ + xy’ + nZ(x2 - 1)y = 0 Determine a uniform asymptotic expansion for the general solution of this equation for large n. 7.11. Determine uniform asymptotic expansions for small E for the solutions of y”

where

CL

+

is a constant.

x(1

+

(1

2)

- ca

+ x)2

?4 = O*

-1 < x

<

384

ASYMPTOTIC SOLUTIONS OF LINEAR EQUATIONS

7.12. Consider the Graetz problem for heat transfer in a duct u"

+ P(1

- x2)f(x)u

u(1) = u(-1)

=0

=0

where d >> 1 andf(x) =f( -z) > 0. Determine a first approximation to 1 and the eigenfunctions using (a) the method of matched asymptotic expansions (Sellers, Tribus, and Klein 1956), and (b) the method of multiple scales or Langer's transformation (Nayfeh, 1965b). 7.13. Given that f(x) > 0 and d >> 1, determine first approximations to the following eigenvalue problems (a) y"

+ d2(1 - x)f(z)y

=0

y(0) = 0 and y(m)

(b) xy"

+ +

<

00

y' d2xf(z)y = 0 y(1) = 0 and y(0) < w

+

(c) y" 12(1 - z)"f(x)c)y = 0, y(0) = 0 and y(m) < m

n is a positive integer

(d) zy" y(1)

n is a positive integer

+ y' + A2xnf(z)y = 0, =0

and y(0)

+ A2(x - l)"(2

<

w

- z)"f(x)y = 0, m and n are positive integers for all x. 7.14. Consider the Graetz problem for heat transfer in a pipe (e) Y'' y

<

run

+ u' + Pr(1 - r2)f(r)u

u(1) = 0,

u(0)

=0

<

where d >> 1 andf(r) > 0. (a) Determine an expansion valid away from r = 0 and r = 1 ; (b) determine expansions valid near r = 0 and r = 1;(c) match these three expansions, hence determine A, and form a uniformly valid composite expansion (Sellers, Tribus, and Klein, 1956). 7.15. Consider Exercise 7.14 again. (a) Determine an expansion valid away from r = 0 using Langer's transformation; (b) determine an expansion valid away from r = 1 using Olver's transformation; (c) match these two expansions to determine 2, (d) form a composite uniformly valid expansion; and (e) compare these results with those of Exercise 7.14. 7.16. Consider the problem y" - 12(z - 1)(2 - z)y = 0 Determine an approximate solution to y if 1 >> 1, and y 7.17. Consider the problem u"

+ d2(1 - z)"(x

- p)rnf(z)u

u(1) = u(p)

=0

< co for all x.

=0

EXERCISES

385

where 1 >> I , f ( . r ) > 0, n and rn are positive integers, and < 1 . Determine expansions valid away from z = and z = 1 using Olver’s transformation, match them to determine 1, and then form a composite expansion, 7.18. Consider the problem ru”

+ pu’ + W(l - r)mf(r)u

= 0,

u(1) = 0 and u(0)

<

m

where ), >> 1, f ( r ) > 0, 11 is a real number, and n and rn are positive integers. Determine expansions valid away from r = 0 and 1 using Olver’s transformation, match them to determine 1, and then form a composite expansion (Nayfeh, 1967a). 7.19. Given that f(z) > 0 and 1>> 1 , determine first approximations to the eigenvalue problems (a) s y ”

+ y’ + Pz(1 - z)f(z>y = 0 + +

(b) r y ” y’ A2zn(l- z)*f(z) = 0, rn and n are positive integers subject to the condition that y < rn for z 2 0. 7.20. How would you go about determining a particular solution for U”

+ ,4223~= n2g(z)

when 1 >> 1 if (a) g(0) # 0, (b) g(0) = 0 but g‘(0) # 0, (c) g(0) = g’(0)= 0 but g”(0) # 0, and (d) g(0) = g’(0) = g”(0) = O?

5. A molecular study on the depth-dependent oxidation and condensation gradients of aged dammar and mastic varnish films

Abstract

The depth-dependent oxidation and condensation gradients of extremely aged, ~ 55 µm thick dammar and mastic films are studied on the molecular level. Direct Temperature-resolved MS (DTMS), Matrix-Assisted Laser Desorption/Ionisation Time-Of-Flight MS (MALDI-TOF-MS) and High Performance Size Exclusion Chromatography (HP-SEC) were employed to analyse the films after 248 nm laserinduced removal of 3.5, 7, 11.5, 15, 20, 25 µm and 3, 6, 10, 12, 16, 20, 25 µm from the aged dammar and mastic films respectively. Electron ionisation (16 eV) DTMS total ion currents indicated that polarity and condensation decrease depth-wise, since lower temperature is required to volatilise the incorporated polar compounds and to induce pyrolysis of the high MW condensed fraction as the thickness of both films reduces. The relative abundance of oxidised dammarane, oleanane and ursane type triterpenoids gradually decreases with depth. Multivariant Factor Discriminant Analysis (DA) quantified the oxidative gradient and showed that a depth of 15 µm from surface of the aged films is the threshold between highly and poorly deteriorated

Chapter 5

A molecular study on the depth-dependent gradients

material. MALDI-TOF-MS showed that UV-induced oxidation resulting in A-ring openings at position C-2 of the oleanane / ursane type molecules stops at the 15 µm below the surface. The upper layers completely absorb radiation with λ < 350 nm. HP-SEC determined that the high molecular weight fraction becomes less prominent as a function of depth. All the data presented establish the depth-dependent compositional gradients and determine that deep layers in the bulk of aged dammar and mastic films remain unaffected from autoxidative degradation processes. The fact that non-UV-induced oxidation and unaffected material were detected in the bulk of the laser ablated varnish films indicates that excimer laser ablation at 248 nm is a non-oxidative process.

5.1 Introduction Ageing of triterpenoid resins, such as dammar and mastic, under light, is in principle an oxidative process, leading, as a first step, to the conversion of triterpenoid hydrocarbon type carbon skeletons (Figure 2.2.1) to carbon skeletons with oxygen containing functional groups (Figure 2.2.2) (De la Rie 1988a, Van der Doelen 1999). Today, it is understood that these changes are only the beginning of an oxidative process leading to radical polymerisation (crosslinking or condensation), oxidative modifications in the side chain or in the functional groups on the tetra- or the pentacyclic ring structures, shortening of the side chain, especially in triterpenoid compounds with the dammarane skeleton (Van der Doelen, et al. 1998a, Van der Doelen, et al. 1998b), and probably to eventual defunctionalisation, bond breaking and disintegration of the triterpenoid carbon skeleton (Boon and van der Doelen 1999).

250

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

As already described above (Section 2.2), the extent of oxidative degradation is significantly affected by the presence or absence of UV wavelengths in the incident radiation. The main difference is that UV wavelengths lead to an opening of the Aring at position C-2 of dammarane and oleanane/ursane type molecules (Figure 2.2.2c) (Van der Doelen, et al. 2000), while in the absence of UV wavelengths most of the oxidative products are associated with oleanane and ursane type triterpenoids oxidised at positions C-11 and C-28 (Figure 2.2.2b) (Van der Doelen, et al. 1998a). UV wavelengths have been employed for various studies on the ageing of the triterpenoid varnishes (De la Rie 1988b, Zumbühl, et al. 1998, Van der Doelen, et al. 2000, Scalarone, et al. 2003). It has been determined, herein, that for the study of extremely aged varnishes, UV wavelengths are essential during the ageing process (Section 2.3). Even freshly harvested triterpenoid resins, having been irradiated by sunlight during exudation from the bark of the resin trees (Koller, et al. 1997), contain dammarenolic acid and 20,24-epoxy-25-hydroxy-dammaran-3-one (Mills and Werner 1955, Poehland, et al. 1987, De la Rie 1988a, Van der Doelen, et al. 1998a, Van der Doelen, et al. 1998b, Scalarone, et al. 2003), both having oxidised A-rings. The presence of these highly degraded molecules in ‘fresh’ resins is explicable by the free radical chain reactions which were initiated in the resin teardrops during the harvesting period (Dietemann 2003). A DTMS study of 55 µm thick dammar and mastic films, aged under intense, accelerated deteriorative conditions1, showed that both films contain a mixture of a few intact triterpenoid molecules, some oxidised triterpenoids with A-ring openings, evidently produced upon UV irradiation, and 1

500 h xenon arc exposure (λ > 295 nm) at 60°C, 45 days near an open window and over a month storage in the dark

251

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

some oxidised oleanane/ursane type molecules generated upon non-UV irradiation (Section 2.3). It is should be stressed that UV-induced and non-UV-induced oxidised triterpenoids were found to be almost in equal relative amounts in the investigated films (Table 2.3.5). In Section 4.3.2 it was demonstrated that the distribution of carbonyl groups is not random in the thickness of the films but influenced by the optical absorption lengths at the various wavelengths contained in the incident light. In particular, upon ageing dammar and mastic films absorb completely wavelengths with λ < 350 nm at the uppermost 10-15 µm surface layers, wavelengths with λ > 350 nm penetrate deeper and wavelengths λ > 400 nm fully penetrate the 55 µm films tested (Figure 4.3.2.6). A theoretical model based on earlier GC/MS findings on aged dammar and mastic indicated that, owing to the strong light intensity and the abundant presence of oxygen in the surface, most of the oxidative products are generated in the surface, while the reduced amounts of oxygen in the bulk result in termination reactions of free radicals producing non-oxidative crosslinks (Boon and van der Doelen 1999). In other words, it is suggested that the oxidative deterioration of aged dammar and mastic films reduces with depth. The most apparent clues towards this conclusion are that: –

aged dammar films thicker than 10 µm were found to be less deteriorated than thinner dammar films aged under the same conditions (De la Rie 1988b).

–

solubility tests of laser ablated, naturally aged varnishes on paintings using a KrF excimer laser and a mixture of cleaning solvents showed that polarity decreases as a function of depth (Theodorakopoulos and Zafiropulos 2003, Section 1.4).

252

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

– the amount of free radicals in fresh mastic resin teardrops, harvested in the same period, increases as the teardrops decrease in size, indicating that mastic blocks the propagation of light in the bulk (Dietemann 2003). These findings are supported by theoretical models on the ageing of various organic materials. Thomson suggested the presence of unreacted substance across depth of light aged organic films and based on Beer’s Law postulated that the abundance of this unreacted material should increase with increasing distance from the surface (Thomson 1965, 1979). Other models have shown that the consumption of oxygen across depth is dependent upon a reducing rate of diffusion, which leads to ‘oxygen starvation’ in the bulk (Thomson 1978, Cunliffe and Davis 1982, Fukushima 1983). The integration of data resulting from reducing light propagation into films and from the reduction of oxygen availability with depth produced models showing that the overall degradation of organic substrates is reduced as a function of depth (Schoolenberg and Vink 1991). A comprehensive review about depth-dependent degradation in various materials due to light propagation and oxygen diffusion has been already provided (Feller 1994a). Heat-bodied oil coatings, such as a thick copal oil varnish film tested, are the exception of this rule, because of the prepolymerisation process (Mantell, et al. 1949, Mills and White 1994, Carlyle 2001), which leads to high degrees of saturation both in terms of absorption (Sections 2.4, 2.5 and 4.3.2) and in terms of bonding of the incorporated molecular structures (Chapter 6). The depth-dependent gradients in oxidation and crosslinking across the thickness of aged dammar and mastic films have been determined experimentally by the changes 253

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

in the interaction of these films at consecutive depths with a KrF excimer laser (Zafiropulos,

et

al.

2000,

Theodorakopoulos

and

Zafiropulos

2003,

Theodorakopoulos, et al. 2005, Chapters 1 and 3). In particular, it was shown that (a) the ablation rate per pulse and the ablation yield per incident laser photon (Section 3.3.2), (b) the carbon dimers emitted in the ablation plume monitored by LIBS (Section 3.3.4), (c) the ablation step obtained using a common laser cleaning method (Section 3.3.5) and (d) the laser light transmitted towards the underlying substrate (Section 4.3.1.2), gradually change across the reducing thickness of the ablated films. UV/VIS and ATR-FTIR measurements established the decreasing gradients in absorption, oxidation, and crosslinking as a function of depth (Section 4.3.2), and the results were in a good agreement with earlier findings (Zafiropulos, et al. 2000). Moreover, evidence was provided about the non-oxidative contribution of the 248 nm pulses of the KrF excimer laser to the remaining films, as has been also determined elsewhere (Castillejo, et al. 2002). The latter finding shows that the determined gradients are not a result of the interaction of the resin films with the UV laser photons but a genuine characteristic of aged dammar and mastic films. This chapter aims at the establishment of the compositional gradients of aged dammar and mastic films on the molecular level. The evaluation of the dammar and mastic films prior to and after ageing (Section 2.3) enables a comparison with the degradation state of the films across their depth profiles. The following work is based on

Direct

Temperature-resolved

MS

(DTMS),

Matrix-Assisted

Laser

Desorption/Ionisation Time-Of-Flight MS (MALDI-TOF MS) and High Performance – Size Exclusion Chromatography (HP-SEC). Analytical data from earlier studies in

254

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

particular of aged and unaged natural resins, based on HPLC-MS, GC/MS and DTMS, are employed for an efficient interpretation of the complex DTMS summation mass spectra (De la Rie 1988a, Papageorgiou, et al. 1997, Van der Doelen, et al. 1998a, Scalarone, et al. 2003). Quantification of the DTMS data is enabled with Multivariant Factor Discriminant Analysis. MALDI-TOF-MS is employed to provide complementary molecular information on the ablated surfaces and interpretation of the MS is enabled by comparison with related analytical data (Zumbühl, et al. 1998, Dietemann 2003). Finally, SEC is employed to determine the molecular weight modifications across depth to support findings based on excimer laser ablation (Chapter 3) and ATR-FTIR (Section 4.3.2). Interpretation of the SEC traces was based on previous molecular weight studies of aged dammar films (De la Rie 1988b, Van der Doelen and Boon 2000).

5.2 Experimental 5.2.1 Direct temperature-resolved mass spectrometry The laser-ablated zones were examined with DTMS in order to investigate the potential molecular variation across the depth profile. From each zone a thin film of approximately 4 mm2 was mechanically subtracted, then homogenized and brought in suspension with a few drops of ethanol. A volume of 2-3 µl of the mixture was applied to a Pt/Rh (9:1) filament (100µm diameter) of a direct insertion probe, and dried in vacuo (using the Purevap) by evaporation of the ethanol. After insertion of the probe in the ionisation chamber a gradual temperature increase of the filament was set at a rate of 1A/min to a final temperature of approximately 800ºC, while the MS was monitoring the evolved compounds in electron ionisation (EI) or ammonia

255

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

chemical ionisation (NH3/CI) modes. The compounds were ionised at 16 eV energy in EI mode and 250 eV in NH3/CI mode, and analysed in a JEOL SX-102 double focusing mass spectrometer (B/E) over a mass range from 20-1000 Dalton at a cycle time of 1 s. The acceleration voltage was 8 kV. The total ion currents and summation mass spectra were examined. 5.2.2 Factor discriminant analysis (DA) EI-DTMS summation mass spectra were numerically analysed by factor discriminant analysis coupled with the FOMpyroMAP multivariate analysis program, that is a modified version of the ARHTUR package from Infometrix Inc. (Seattle, USA; 1978 release) and with FOM developed Matlab® (The Mathworks Inc., Natick, MA, USA) toolbox ChemomeTricks. The discrimination was based on a double stage component analysis (PCA) that is described elsewhere (Hoogerbrugge, et al. 1983). Mass spectra of triplicate (at least) measurements for each sample were inserted in the aforementioned software to enable discrimination. To minimise variance in the data due to variance in the mass spectrometer readings, the DT mass spectra of the same films were measured on the same day prior to DA analysis. 5.2.3 Matrix-assisted laser desorption/ionisation Time-Of-Flight mass spectrometry (MALDI-TOF-MS) Sampling was enabled with ethanol wetted minute TLC plates coated with cellulose that were simply brought in contact with the surface of the aged, unaged and laser ablated films. The samples were coated with a thin layer of 2,5-dihydroxybenzoic acid (DHB, Aldrich, Steinheim, Germany). A sample of neat DHB coating was also prepared for reference. The TLC plates were attached on the stainless steel MALDI

256

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

probe, which was then placed on the x-y-z translator of the TOF-MS system (BruckerFranzen Analytik, Bremen, Germany) for the appropriate orientation into the ionisation chamber (10-7 mbar) of the system. The ion source consists of positively or negatively charged metal electrode that is the MALDI probe and a grounded accelerating grid at a ~ 2 cm distance. The accelerating potential was 19.9 kV in the reflex mode. As soon as the vacuum lock was fastened the probe was in position to face the laser beam at an angle of 60º with respect to the surface normal. The nitrogen discharge (N2) laser (337 nm) used had a 4 ns pulse duration and a repetition rate of 12 Hz (Photon Technology PL2300). Manipulation of the sample was enabled by monitoring the ion source via a CCD camera. Processing was carried out with Bruker software (DataAnalysis for TOF 1.6.g). 5.2.4 High Performance Size Exclusion Chromatograph (HP-SEC) The study of the molecular weight changes, across the depth profiles of the light aged, laser ablated resin films was carried out with High Performance Size Exclusion Chromatography (HP-SEC). Unaged, aged and laser ablated samples were dissolved in THF (~10mg/µl) and centrifuged. An amount of 20 µl was analysed on a Shimadzu HP-SEC system, consisting of a SCL-10AD vp control panel, an LP-10AD vp pump, a DGU-14A degasser, a SIL-10AD vp autoinjector, a CTO-10AS column oven and a FRC-10A fraction collector (Shimadzu Benelux, ‘s-Hertogenbosch, The Netherlands). Two different detectors connected in series were used for the signal detection. These were a Shimadzu SPD-10A vp UV/VIS detector operated at 240 nm and a Shimadzu RID-10A refractive index detector in combination with a Shimadzu Class vp 5.03 software. The temperature in the system was 40ºC and the flow rate was 1ml/min.

257

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

Finally, calibration was carried out with polystyrene standards (Polymer Laboratories) with an average mass ranging from 580 to 370,000 Daltons.

5.3 Results and Discussion 5.3.1 DTMS of the laser-ablated depth-profiles of the aged dammar and mastic films Tables 5.3.1.1 and 5.3.1.2 present the lists of identified compounds in the aged and laser ablated dammar and mastic films with DTMS. It should be noted that because of the sampling and the subsequent dissolution of the aged films in ethanol for the technical requirements of sampling for DTMS, the final mass spectra provide average readings with respect to the abundance of these compounds across depth. Both the aged ~ 55 µm dammar and mastic films (Appendix 2.7) are basically a mixture of triterpenoids with oxidised A-rings at position C-2 (5, 7, 152, 17-20, 22 and 23), oxidised oleanane/ursane type triterpenoids at positions C-11 and C-28 (24-28) and a series of unreacted triterpenoids (1-16). According to the optical absorption lengths at various wavelengths (Figure 4.3.2.6), this mixture indicates that the aged dammar and mastic films can be divided in three different parts in their thicknesses; namely a surface section that absorbed the UV wavelengths of the incident radiation, a deeper part that absorbed only the visible radiation and an underlying part in the bulk at which the transmitted visible light was of low intensity, which combined with the absence of oxygen did not lead to the destruction of the triterpenoid molecules.

2

Compounds 5, 7 and 15 are typical oxidised dammarane type molecules with an A-ring opening at C2 (Mills and Werner 1955, Poehland, et al. 1987, De la Rie 1988a) and are present in triterpenoid resin films even prior to ageing (Van der Doelen, et al, 1998a, Dietemann 2003).

258

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

Table 5.3.1.1 List of Compounds found in aged and laser ablated dammar resin films

Label Compound name

Mw

EI mass/charges*

NH3/CI mass/charges

1

Nor-α-amyrone (3-oxo-28-nor-urs-12-ene)

410

204, 410

428

2

Nor-β-amyrone (3-oxo-28-nor-olean-12-ene)

410

204, 410

428

3

Dammaradienone (3-oxo-dammara-20(21),24-diene)

424

109, 205, 424

442

4

Dammaradienol (3β-hydroxy-dammara-20,24-diene)

426

109, 189,207,408, 426

444

5

Dammarenolic acid (20-hydroxy-3,4-seco-4(28),24-dammaradien-3oic acid)

458

109, 440

476,458

6

Hydroxydammarenone (20-hydroxy-24-dammaren-3-one)

442

109, 315, 355, 424

442

7

20,24-epoxy-25-hydroxy-dammaran-3-one

458

143, 399

143

8

Oleanonic acid (3-oxo-olean-12-en-28-oic acid)

454

189, 203,248,409, 454

472

9

Ursonic acid (3-oxo-12-ursen-28-oic acid)

454

189, 203,248,409, 454

472

11

Oleanonic aldehyde

438

203, 232, 409, 438

456

12

Ursonic aldehyde

438

203, 232, 409, 438

456

15

20,24-epoxy-25-hydroxy-3,4-seco4(28)dammaren-3-oic acid

474

143

143

17

3,4-seco-28-nor-urs-12-en-3-oic acid

428

204, 428

446

18

3,4-Seco-28-nor-olean-12-en-3-oic acid

428

204, 428

446

19

3,4-seco-28-nor-urs-12-en-3,28-dioic acid

472

203, 248 472

490

20

3,4-seco-28-nor-olean-12-en-3,28-dioic acid

472

203, 248 472

490

21

Dihydro-dammarenolic acid (20-hydroxy-3,4-seco-24-dammaren-3-oic acid)

460

387, 456

478

3-oxo-25,26,27-trisnordammarano-24,20-lactone

414

95, 99, 205, 315, 414

432

440

234, 275, 422, 440

458

468

217, 257, 276, 317, 482

486

440

234, 275, 422, 440

458

468

257, 276, 317, 482

486

24 25 26 27 28

17-hydroxy-11-oxo-nor-β-amyrone (3,11-dioxo-17-hydroxy-28-norolean-12-ene) 11-oxo-oleanonic acid (3,11-dioxo-olean-12-en-28-oic acid) 17-hydroxy-11-oxo-nor-α-amyrone (3,11-dioxo-17-hydroxy-28-norurs-12-ene) 11-oxo-ursonic acid (3,11-dioxo-urs-12-en-28-oic acid)

*

m/z in bold correspond to the highest relative intensity of the series according to previous findings (Van der Doelen, et al. 1998a)

259

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

Table 5.3.1.2 List of Compounds found in aged and laser ablated mastic resin

Label Compound name

Mw

EI mass/charges

NH3/CI mass/charges

2

Nor-β-amyrone (3-oxo-28-nor-olean-12-ene)

410

204, 410

428

3

Dammaradienone (3-oxo-dammara-20(21),24-diene)

424

109, 205, 424

442

5

Dammarenolic acid (20-hydroxy-3,4-seco-4(28),24-dammaradien-3oic acid)

458

109, 440

476,458

6

Hydroxydammarenone (20-hydroxy-24-dammaren-3-one)

442

109, 315, 355, 424

442

7

20,24-epoxy-25-hydroxy-dammaran-3-one

458

143, 399

143

8

Oleanonic acid (3-oxo-olean-12-en-28-oic acid)

454

189, 203,248,409, 454

472

10

Moronic acid (3-oxo-olean-18-en-28-oic acid)

454

189, 203,248,409, 454

472

11

Oleanonic aldehyde

438

203, 232, 409, 438

456

13

(Iso)masticadienonic acid (3-oxo-13α,14β,17βH,20αH-lanosta-8,24-dien26-oic acid or 3-oxo-13α,14β,17βH,20αHlanosta-7,24-dien-26-oic acid)

454

439, 454

472

14

3-O-Acetyl-3epi(iso)masticadienolic acid (3α-Αcetoxy-13α,14β,17βH,20αH-lanosta-8,24dien-26-oic acid or 3α-Acetoxy-13α,14β,17βH,20αH-lanosta-7,24dien-26-oic acid)

498

438, 498

516

16

28-nor-olean-18-en-3-one

410

163, 191, 410

428

18

3,4-seco-28-nor-olean-12-en-3-oic acid

428

204, 428

446

20

3,4-seco-28-nor-olean-12-en-3,28-dioic acid

472

203, 248 472

490

22

3,4-Seco-28-nor-olean-18-en-3-oic acid

428

204, 428

446

23

3,4-seco-28-nor-olean-18-en-3,28-dioic acid

472

203, 248 472

490

24

3-oxo-25,26,27-trisnordammarano-24,20-lactone

414

95, 99, 205, 315, 414

432

440

234, 275, 422, 440

458

468

217, 257, 276, 317, 482

486

25 26

17-hydroxy-11-oxo-nor-β-amyrone (3,11-dioxo-17-hydroxy-28-norolean-12-ene) 11-oxo-oleanonic acid (3,11-dioxo-olean-12-en-28-oic acid)

*

m/z in bold correspond to the highest relative intensity of the series according to previous findings (Van der Doelen, et al. 1998a)

260

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

100 Aged Dammar: surface 50

100 Aged Mastic: 50 Surface

0

0 100

- 3 µm

50

% Relative Intensity

% Relative Intensity

100 - 3.5 µm 50 0 100 - 15 µm 50 0 100 - 20 µm 50

0 100

0 100

- 25 µm

50

0

0

100

100

50

- 16 µm

50

50

Unaged Dammar

0

Unaged Mastic

0 20

40

60

80

20

Scan No.

40

60

80

Scan No.

Figure 5.3.1.1 EI-DTMS TIC’s of laser-ablated depth steps of (a) aged dammar and (b) aged mastic

Figure 5.3.1.1 presents the electron ionisation (16 eV) total ion currents (EI-DTMSTIC’s) of the aged films prior to ablation and that of the laser ablated depth-steps after removal of 3.5, 15 and 20 µm from the aged dammar film and 3, 16 and 25 µm from the aged mastic film. The EI-DTMS-TIC’s of the unaged films are also presented for comparison reasons. The major shifts towards lower scan numbers across the successive depth-steps and especially of the volatile fraction that is desorbed via evaporation (Boon 1992) demonstrate that lower temperatures are required for sufficient desorption with depth. Consequently, the more material is removed from the surface of both aged varnish films the more volatile the remaining films become. The apparent shifts are so intense that after removal of 20 µm from the aged dammar and 25 µm from the aged mastic films the corresponding volatile fractions are evaporated with ~ 80% less heating compared to the temperatures that induced evaporation in the unaged and aged films prior to ablation. As discussed in Section 2.3, such shifts are 261

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

related to changes in the polarity (Figure 2.3.1) (Van der Doelen 1999, Scalarone, et al. 2003). Moreover, the lower peak that emerges at high scan numbers represents the high molecular weight, crosslinked and condensed fraction of the analysed samples that desorbs in the ionisation chamber via pyrolysis at high degrees of temperature (i.e. T > 450°C) (Boon 1992). In both cases, pyrolysis occurs at lower temperatures, while the corresponding peaks become weaker with depth indicating that the high MW fraction becomes less polar and is reduced as a function of depth. It should be noted that the pyrolysis peak gives only an estimation of the high MW fraction, because at high temperature windows (> ~500°C) thermally stable material is formed corresponding to breakage of more or stronger bonds (Boon 1992) (see also Section 2.3). Despite this complexity the pyrolysis event is used often as a good indicator of the high MW fraction of aged dammar and mastic resins (Van der Doelen 1999, Scalarone, et al. 2003). According to the EI-DTMS TIC’s then, light ageing of dammar and mastic films leads to lower degrees of oxidation, polarity and condensation or crosslinking as a function of depth. With regards to polarity this finding supports preliminary results across the laser ablated depth profile of hundred-year-old natural resin varnishes, according to which solubility increases as a function of depth (Section 1.4, Theodorakopoulos and Zafiropulos 2003). In this particular example, the polarity of the remaining varnish across the laser ablated depth-steps was reducing almost logarithmically in the deeper parts of the tested varnishes (Figure 1.4.1). A similar phenomenon is observed here. Figure 5.3.1.2 shows the % volatilisation shifts against the logarithm of the laser262

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

ablated depth-steps (lower xx’ axis) and the remaining film thickness (upper xx’ axis) (Table 5.3.1.3). The scan number of the EI-DTMS-TIC volatilisation peaks of the films prior to laser ablation was set as the maximum (100%). The corresponding plots quantify the shifts in the temperature windows that are responsible for the evaporation of the volatile fraction. An estimation of the degree of crosslinking versus depth is also provided based on the apex and broadness of the pyrolysis peak of the EI-DTMSTIC’s. However, the corresponding SEC traces are more accurate with respect to the molecular weight modifications across depth (Section 5.3.4). Given that both intensity and broadness of the pyrolysis peaks enclose significant information about molecular weight, the areas of the pyrolysis peaks were measured. The plots of the pyrolysed high MW condensed fractions as a function of depth are presented in Figure 5.3.1.3. It is observed that the crosslinking gradient performs two apparent trends with depth in both films tested. The first trend is terminated at 15 µm from surface and from that point inwards new trends commence, which tend to reach the degree of pyrolysis determined prior to ageing. The outlined jump in the plots after removal of 15 µm from surface coincides with the change of the ablation step using a KrF excimer laser Table 5.3.1.3 Data extracted from EI-DTMS TIC’s Laser ablated depth% Volatilization % Pyrolysis yield steps. Measured (Normalized with scan (Normalized with %area depth steps from surface (µm) no of aged surface) of aged surface)

Surface 1 2 3 4 5 6 7 Unaged

dammar 0 3.5 7 11.5 15 20 25 – 0

mastic 0 3 6 10 12 16 20 25 0

dammar 100 80 77 73 71 66 62 – 54

263

Mastic 100 98 95 94 92 89 88 84 77

dammar 100 86.44 81.76 80.42 78.96 50.68 46.79 – 41.80

mastic 100 85.13 82.59 79.64 78.79 78.35 59.61 58.52 53.24

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

Residual Resin Films (µm) 55

45

50

52

35

25

aged surface both films

100

% Volatilization (with reference to EI-DTMS TIC's)

53

95 90 85 80

unaged mastic

75 70 65

Dammar Mastic

60

unaged dammar

55

0 0.01

2

3

4

5

6 7 8 9 10

20

30

Depth-profile (µm)

Figure 5.3.1.2 Gradient in volatilisation across depth. 55

45

40

35

30

aged surface (both films)

100

% Pyrolysis yield (with reference to EI-DTMS TIC's)

50

25

20

Dammar Mastic

90

80

70

60

unaged mastic 50

unaged dammar 40 0

5

10

15

20

25

30

35

Depth-profile (µm)

Figure 5.3.1.3 Gradient in pyrolysis yield relative intensity across depth.

(Figure 3.3.5.3), as well as with the minimisation of the carbon dimer emission as recorded with LIBS (Figure 3.3.4.2). From a photomechanical point of view, excimer laser ablation of organic materials is dependent on the degree of polymerisation and crosslinking (Srinivasan and Braren 1989, Zafiropulos 2002). Thus, all the results

264

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

presented in the present work demonstrate that below the 15 µm surface layers neither of the aged dammar and mastic films is heavily crosslinked. Figures 5.3.1.4 and 5.3.1.5 present the ammonia chemical ionisation (NH3/CI) DTMS summation mass spectra of the aged dammar and mastic films prior to laser ablation (i), after removal of ~12 and 25 µm from surface (ii, iii) and prior to ageing (iv). The electron ionisation (EI) DTMS summation mass spectra of the aged films prior to ablation, after removal of increasingly thicker surface layers ranging from 3 to 25 µm, and prior to ageing are presented in Figures 5.3.1.6 (dammar) and 5.3.1.7 (mastic). In NH3/CI mode (250 eV) ammonia adduct molecular ions [M+NH4]+, [M+H]+ and [M+NH4-H2O]+ cations are obtained resulting in molecular weight information, while in EI mode (16 eV) minor fragmentation is obtained resulting in structural and more detailed information. Interpretation of the mass spectra was based on earlier DTMS studies (Van der Doelen, et al. 1998a, Van der Doelen 1999, Van der Doelen and Boon 2000, Scalarone, et al. 2003), as well as on GC/MS and HPLC-MS (Van der Doelen, et al. 1998a, Van der Doelen, et al. 1998b). The most characteristic molecular ions and ion fragments of the identified compounds are presented in Tables 5.3.1.1 and 5.3.1.2. Details of the incorporated compounds (Figure 2.2.1) and the ion fragments thereof are given in Section 2.3. It is demonstrated that the thicker the layer removed from the highly deteriorated surfaces, the more the molecular configuration of the remaining films tends to resemble the composition of the resins prior to ageing. The depth-dependent shift towards a less oxidative state is evident even after removal of only 3.5 and 3 µm from the aged dammar and mastic films respectively, although the molecular changes are 265

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

376

334

20

(i)

460 476

224 254

428

(Aged Surface) 40

446

490

(-11.5 microns) 224 254

334

376

460 476

(ii)

490 0

(-25 microns) 224 254

376

472

334

428 460 426 476

(iii)

458

50

442

rel. abundance %

50

426 428 442

0

486

0

224

334

254

376

458

50

460 442 428 426 472

(Unaged)

(iv) 476

0 200

300

400

500

mass/charge

Figure 5.3.1.4 NH3/CI DTMS summation mass spectra of the aged dammar film (total thickness: ~ 55 µm) prior to ablation (i), after removal of 11.5 µm (ii) and 25 µm (iii) from surface, and the NH3/CI DTMS of the same film prior to ageing (iv) (i)

428

(Aged Surface)

50

408

458

100

442

472 476 486 490

0

(ii)

442

50

408

458 472

428

(-12 microns)

476 486 490

0

100

428

(-25 microns)

50

408

(iii) 458 472

rel. abundance (%)

100

444

486

0

428

(Unaged)

408

(iv)

458 472

50

444

486

0 200

300

400

500

mass/charges

Figure 5.3.1.5 NH3/CI DTMS summation mass spectra of the accelerated aged mastic (total thickness: ~ 55 µm) prior to ablation (i), after removal of 12 µm (ii) and 25 µm (iii) from surface, and the NH3/CI DTMS of the same film prior to ageing (iv)

266

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

more obvious with increasing depth. The unreacted compounds observed are the dammarane type triterpenoids (3, 4, 5, 6 and 7) with EI marker peaks at m/z 109, 143, 424 and NH3/CI marker peaks at m/z 442, 446, 478 and the oleanane/ ursane triterpenoids (1, 2, 8, 9, 10, 11 and 12) with EI marker peaks at m/z 189, 203, 204, 248, 409, 410 and NH3/CI marker peaks at m/z 428, 456, 472. These marker peaks are evidently increasing with depth. Although both types of triterpenoids are contained in both dammar and mastic, molecules with the dammarane skeleton are more abundant than those with the oleanane / ursane skeleton in dammar resin, while the opposite occurs in mastic resin. The recovery of the ion fragments corresponding to these marker compounds is more impressive in dammar, because of the abrupt reduction of the triterpenoid compounds that is observed in dammar resin upon ageing (Section 2.3.1, Van der Doelen, et al. 1998a, Van der Doelen 1999, Scalarone, et al. 2003). Oxidised compounds such as the oleanane / ursane type molecules with the A-ring opening at position C-2 (17/18, 19/20 for dammar and 18/22, 20/23 for mastic) with EI marker peaks at m/z 428, 472 and NH3/CI marker peaks at m/z 446, 490, which are in trace amounts in the aged films prior to ablation seem to disappear after removal of the uppermost 15 µm in both cases. Since these particular compounds are only formed upon UV irradiation during accelerated ageing (Van der Doelen, et al. 2000), it is indicated that the UV wavelengths of the incident radiation do not penetrate deeper than a 15 µm surface zone of the films. This finding is in a very good agreement with the optical absorption lengths, lα, determined for the aged dammar and mastic films tested (Figure 4.3.2.6), showing that both films absorb completely the incident

267

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

radiation with λ < 350 nm at depths shorter than 15 µm from surface. This observation is confirmed with MALDI-TOF-MS (Section 5.3.3). 143

80

399

428

454 456 470 472

408

359 355 371

482

300

350

400

450

248

424 426

95

234

257 275

315

359 355 371

414

408 410

189

40

399

109

387

60

500

(- 3.5 microns) (ii)

204

80

20

250

440 454 456 470 472

200

428

150

143

99

482

0 100

150

200

350

400

450

500

(- 7 microns) (iii) 248

109

189

99

95

235

257 275

315

359 371 355

426 424 440 408 454 456 470 472

204

60

414

80

20

300

143

100

40

250

399

0 100

150

200

100 80

20

300

350

400

450

500

(- 11.5 microns) (iv) 204 248

60 40

250

143

109

424 426

189

95

235

257

315

275

355

359 371

408 399

440 454 456 470 472

% rel. abundance

315

0

100

% rel. abundance

234 257 275

414

95

424 426 440

387

109

40 20

248

410

204

60

100

% rel. abundance

(Aged Surface) (i)

99

% rel. abundance

100

0 100

150

200

250

300

350

400

450

500

mass/charge

Figure 5.3.1.6 EI-DTMS summation mass spectra of the accelerated aged dammar film (total thickness: ~ 55 µm) prior to ablation (i), and after removal of 3.5 µm (ii), 7 µm (iii) and 11.5 µm (iv) from surface.

268

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

143

% rel. abundance

100 80

424 426

60 40

408

189 257

234

20

315

275

410 440

355359 371 399

454

150

200

300

350

400

450

60

424 248

109

189

20

235 257 275

315

355

440

408 359 371 399

454

0 150

200

250

300

350

143

100

400

470

450

500

(- 25 microns) (vii) 204

80 60

500

(- 20 microns) (vi) 204

80

40

250

143

100

424 248

109

40

189

20

235

257 275

315

359 355 371

410 408

440 454 470

0 100

150

200

250

300

350

400

450

500

424

100

% rel. abundance

470

0

100

% rel. abundance

248

109

100

% rel. abundance

(- 15 microns) (v) 204

(Unaged) (viii)

204

80 60 40

109

143

189

248 315

20

355

408

440 454

0 100

150

200

250

300

350

400

450

500

mass/charge

Figure 5.3.1.6 (continued) EI-DTMS summation mass spectra of the accelerated aged dammar film after removal of 15 µm (v), 20 µm (vi) and 25µm (vii) from surface and the corresponding EI-DTMS of the unaged film (viii) (total thickness: ~ 55 µm).

269

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

95 109

20

399 395

250

300

350

468 470 472 484 498

400

450 500 (- 3 microns) (ii) 408

143

426

95 109

424

248

218 235

395 275 315 327

20

440 454

422

191203

399

163

60 40

200

468

355

484498

0 150

40

250

143

300

350

400 408

248

80 60

200

163

218 235

395 275

20

500

(- 6 microns) (iii) 424

191 203

95 109

450

440 454

399

100

468

315 327 355

484 498

0 150

200

250

100 80

143 95

60

163

109

203 191

300

350

400

450 500 408 (- 10 microns) (iv)

248

424 440

218 235

40

315 327

273

20

395 355

399

100

454 468 484

498

0 100

150

200

250

350

400 408

100

143

80 163

203 191

95 109

218 235

20

450 (- 12 microns)

500 (v)

424

248

60 40

300

395

315 327

275

399

% rel. abundance

150

80

100

% rel. abundance

315 327 355

0

100

% rel. abundance

218

275

100

% rel. abundance

410 424 426 440 454 422

40

203 205 191

428

163

408

234

% rel. abundance

80 60

(Aged Surface) (i)

248

143

100

440 454 468

355

484 498

0 100

150

200

250

300

350

400

450

500

mass/charge Figure 5.3.1.7 EI-DTMS summation mass spectra of the accelerated aged mastic film (total thickness: ~ 55 µm) prior to ablation (i), and after removal of 3 µm (ii), 6 µm (iii), 10 µm (iv) and 12 µm (v) from surface.

270

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

408 (- 16 microns) (vi) 143

60

163

424

191 203

95 109

440454

218 235

315 327 355

275

20

399

80

40

248

484 498

0 100

150

200

250

300

350

400 408

60 40

163

191

203

95109

218

235

20

315

440 454 468

327

355

395

484 498

0 100

150

200

250

100

% rel. abundance

450 500 (- 20 microns) (vii)

424

248

399

143

80

400

450

408

191

440

163

95 109

500

(- 25 microns) (viii)

424

203

143

60

20

350

248

80

40

300

218 235

399

% rel. abundance

100

275

315327

355

454 468

395

484

0 100

150

200

250

203

100

300

350

400

60

20

191

143

40 109

163

450

248

424440 454 218 235 315

150

200

(ix)

408

327

355

395

468 484 498

0 100

498 500

(unaged)

80

95

% rel. abundance

468

395

399

% rel. abundance

100

250

300

350

400

450

500

mass/charge Figure 5.3.1.7 (continued) EI-DTMS summation mass spectra of the accelerated aged mastic film after removal of 16 µm (vi), 20 µm (vii), 25 µm (viii) from surface and the corresponding EI-DTMS of the unaged film (ix) (total thickness: ~ 55 µm).

The DTMS summation mass spectra also detect traces of some oleanane / ursane type molecules oxidised at positions C-11 and C-28, which are listed in Tables 5.3.1.1 and 5.3.1.2 as compounds 25-28 for dammar and 25 and 26 for mastic. The dammarane type compound 24, which is present in both films, has the same type of side chain

271

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

oxidation (Figure 2.2.2b, Van der Doelen, et al. 1998a). Marker peaks of these particular oxidation products (24-28) are monitored at m/z 234, 317, 414, 482 in the EI mode and at m/z 432, 458, 486 in the NH3/CI mode. These particular oxidation products have been detected by Van der Doelen and co-workers (1998a, 1999) in aged varnishes sampled from a large number of paintings kept in several galleries and museums. These products are reproducible in dammar and mastic resins upon light ageing excluding UV wavelengths or using an elaborate ageing process involving dissolution of the resins in a mixture of solvents with photosensitisers followed by irradiation in a fluorescent tube device (Van der Doelen and Boon 2000). Judged from the relative intensity of the corresponding m/z marker peaks, the abundance of these compounds is found to be equivalent to that of the UV-induced oxidation products with the oxidised A-rings in both the aged dammar and mastic 55 µm films prior to ablation (Tables 2.3.5 and 2.3.6). The marker peaks of the non-UV induced oxidation products 24-28 are in particular notable in the summation MS of the remaining films after removal of the 15 µm highly deteriorated surface layers. Some (minor) traces are also observed in the films even prior to ageing, indicating that the films prior to ageing were somewhat oxidised, as shown in Section 2.3. According to the presented data, the molecular changes are apparent by monitoring the gradual recovery of the abundances of the ion fragments corresponding to the unreacted, non-oxidised dammarane (3, 4, 5, 6 and 7) and oleanane / ursane type triterpenoids (1, 2, 8, 9, 10, 11 and 12) with depth. The molecular changes are less pronounced in the uppermost 15 µm, which absorb highly the UV wavelengths of the incident radiation. Strong molecular changes between surface and deeper parts of the

272

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

films are observed after the removal of the highly deteriorated surface layers, especially in the case of dammar, where after removal of 20 µm from surface the characteristic ion fragments of the unreacted compounds produce relative intensities comparable to those of the same film prior to ageing. After removal of 25 µm from the surface of the aged mastic film the ion peak m/z 143, which is the characteristic ion fragment of the ocotillone type molecules, such as compound 7, decreases, indicating that the dammarane type molecules and their oxidation products influence the composition of mastic less than that of dammar. This is a genuine characteristic of mastic resin since the oleanane / ursane molecules are more abundant than the dammarane ones (Van der Doelen, et al. 1998a, Van der Doelen 1999, Scalarone, et al. 2003). The demonstrated molecular changes across the depth-profiles of the aged films determine that with increasing depth the composition of the films is gradually shifted towards a less deteriorated state similar to that prior to ageing. These findings are in a very good agreement with the results of the excimer laser ablation (Chapter 3), the laser induced transmission, the UV/VIS and the ATR-FTIR measurements (Chapter 4). 5.3.2 Multivariant Factor Discriminant Analysis (DA) of the DTMS data across the depth-profiles of the aged dammar and mastic films. The observation of the depth-dependent compositional gradients in the evidently complex DTMS summation mass spectra of the aged dammar and mastic films basically allows a qualitative demonstration of the molecular changes across depth. Multivariant factor discriminant analysis (DA) (Hoogerbrugge, et al. 1983, Windig, et al. 1983) was employed to quantify the modifications observed via the multivariate

273

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

mass data. Figure 5.3.2.1 shows the first discriminant function (DF1) mass spectrum for the EI-DTMS summation MS series of the remaining films after laser ablation at 159 194

248 232

189

(+)

354 375 385 387 413 431 438 456 472

dammar

315 342

224

399

440

(-)

355 205 109

143

410

203

(a) 426

204 424

(+) 472

387 413 428 431

mastic 354

0.2 0.0

109 143

-0.4

82

468

438

315 355 235

399

(-)

189 218 191 440

-0.2

454

-0.6 -0.8

(b)

439 203 424

-1.0

248 408

Figure 5.3.2.1 Projections of the EI-DTMS data classified with MFDA for the aged, laser ablated and unaged films of dammar (a) and mastic (b).

274

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

successive depth-steps of both dammar (a) and mastic (b). The DTMS mass spectra of the unaged films are also included, so that the two opposite compositional cases, i.e. aged-unaged, would mark the discrimination boundaries of the inserted analytical data. In short, judging from the molecular ions of the identified compounds that are listed in Tables 5.3.1.1 and 5.3.1.2, DA analysis shows that across the thickness of dammar, oleanane / ursane molecules 1 and 2, i.e. [M]+· = 410, and the dammarane molecules 3, 4 and 6, i.e. [M]+· = 424, 426, 442 respectively, peak intensely at the negative side of DF1. The characteristic and commonly most abundant ion fragments of the oleanane and ursane molecules 8, 9, 11 and 12, such as m/z 189, 203 and 409, peak at the negative DF1 side. The molecular ions of the latter compounds, i.e. [M]+· = 454 for 8, 9 and [M]+· = 438 for 11, 12 and the characteristic ion fragment at m/z 248, peak at the positive side of DF1, possibly due to overlapping with ion fragments of oxidation products of these molecules. The molecular ions of the UV-induced oxygenated products with the characteristic A-ring openings, i.e., 17/18, ([M]+· = 428), 19/20 ([M]+· = 472) and 21 ([M]+· = 460), peak weakly at the positive DF1 side. Similarly, m/z 234, 317, 414 and 482 representing the non-UV-induced oxidation products (24, 25, 26, 27 and 28) peak in the negative side of DF1. The molecular ions of oxygenated compounds 5, 7 ([M]+· = 458) and 15 ([M]+· = 474) with oxidised A-rings at C-2, which are also detected in the composition of unaged dammar (Mills and Werner 1955, Poehland, et al. 1987, De la Rie 1988a), peak weakly at the positive DF1 side. The ionic fragments of their dammarane skeletons, i.e. m/z 109 and 440 (5),

275

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

m/z 399 (7) and the ionic fragment m/z 143 derived from the ocotillone side chain (7, 15) peak at the negative DF1 side. The first discriminant function mass spectrum of the corresponding EI-DTMS mass spectra of mastic is somewhat different compared to the DA results of dammar. In agreement with the DA analysis of dammar, oleanane molecule 1, i.e. [M]+· = 410, and the dammarane type molecules 3 and 6, i.e. [M]+· = 424, 442 respectively, peak intensely at the negative side of DF1. In contrast to dammar, the molecular ions of the oleanane type molecules 8, 10 ([M]+· = 454) and 11 ([M]+· = 438), peak at the negative DF1 side, confirming that these molecules are abundant throughout the depth of mastic. The same molecular ion with that of 8 and 10 corresponds to (iso)masticadienonic acid (13), the abundant presence of which is monitored at the intense marker peak m/z 439 on the negative DF1 side. Compound 14, which has an ionic fragment at m/z 438 that is also the molecular ion of 11, does not seem to play any significant role in the final discriminant analysis, since its molecular ion at m/z 498 coincides with the DF1 axis. The same occurs with m/z 163 that is characteristic of compound 16, although m/z 191, also corresponding to 16, peaks clearly at the negative DF1 side. In addition, the molecular ion of compounds 5 and 7 ([M]+· = 458) peaks at the positive DF1 side as in the case of dammar. The ionic fragments of these compounds, characteristic of their dammarane skeletons, i.e. m/z 109 and 440 for 5 and m/z 143, 399 for 7, peak at the negative side, although their relative intensity is lower than that in dammar. Finally, the molecular ions of the UV-induced oxidised triterpenoids 18/22, 20/23 (m/z 428, 472) and the marker peaks of the non-UV-induced oxidation

276

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

0 .6 0.6

*1 *1

0.5 0 .5

5

*1 *1

5

Discrimination Function 2

0 .4 0.4 0 .3 0.3

4

4

0 .2 0.2

*9 *9

0 .1 0.1 *9 *9

00

77 88 4 4

77 77

8 6 4 2 0

2

4

6

8

-0 . 6

Discrimination Function 2

0.3

(a)

33

0 10

-0 . 4

-0 . 2

YAxisTitle

-0.6 -0.4 -0.2

0.4

33

33

66 66

10

-0 . 4 -0.4

*1 *1

22

88

-0 . 2 -0.2

0.5

44

2 2

55 6688

-0 . 1 -0.1

-0 . 3 -0.3

22

*9 55 *9

0

0.2

0.4

0.6

0.8

1

1 .2

0 0.2 0.4 0.6 0.8 1 Discrimination Function 1

1 .4

1.2

1.4

7 7 XAxisTitle

7

0.2

5

0.1

*10 *10 9

0 -0.1 *10 *10

-0.2

9 9

7 8 8 8 8

6 5 5 6 6 6

2 24 4 3 4 43 2 2 5 33 1 1 1 1

-0.3

(b)

9

-0.4 -0.2

-0.1

0 0.1 0.2 0.3 0.4 0.5 Discrimination Function 1

0.6

0.7

Figure 5.3.2.2 Score maps of DF2 versus DF1 of dammar (a) and mastic (b) depth steps. The ascending numbers represent the compositions of the successive laser ablated depth steps. Data points number (1) correspond to the aged surfaces, while numbers (9) for dammar and (10) for mastic to the unaged materials. The consecutive shifts across the DF1 from the very deteriorated surface layers towards the less deteriorated bulk, which resembles the composition prior to ageing, are evident.

277

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

products 24, 25, 26 (m/z 414, 234, 276 and 482) peak weakly at the positive DF1 side, since these compounds are found in traces across the depth of the aged mastic film. Figure 5.3.2.2 presents the plots of the MS data points as coordinates in the score map of the first discriminant function (DF1) versus the second discriminant function (DF2). The geometric distance between the data points is an accepted measure of the demonstrated compositional differences (Hoogerbrugge, et al. 1983, Windig, et al. 1983). Therefore this distance is utilised here to illustrate the chemical changes across the depth-profiles of the tested films. The gradual shifts of the data points across DF1 represent the compositional gradient that is observed with the EI-DTMS mass spectra. The data points from the lower to the higher numbers correspond to the consecutive depth-steps from the surface towards the bulk, while the highest numbers represent the MS data of the resins prior to ageing. Figure 5.3.2.3 demonstrates the data points along DF1 as a function of depth for the ~ 55 µm, aged dammar and mastic films. Both cases reveal a remarkable compositional gradient, which decreases from the surface. These results quantify the molecular modifications observed with DTMS and verify previous findings on the excimer laser ablation (Theodorakopoulos and Zafiropulos 2003, Theodorakopoulos, et al. 2005, Chapter 3) and the determined gradients in the optical properties characteristics of both films across depth (Zafiropulos, et al. 2000, Section 4.3.2). Although the delineated trends of the aged dammar and mastic films are not similar, there is a very significant common characteristic. As shown in Figure 5.3.2.2, DA analysis placed the origin of DF1 coordinate at a depth corresponding to 15 µm from the original aged surfaces. This is not irrelevant with the change of the ablative interaction upon 248 278

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

DTMS discrimination function 1

1.2

Depth steps

1.0 0.8 Thresholds of highly aged and less deteriorated material

0.6

(a)

0.4 0.2 0.0 fresh dammar: independent signal from film thickness

-0.2 -0.4 0

5

10

15

20

25

30

35

40

45

Depth from surface (µm)

-0.2

Thresholds of highly aged and less deteriorated material

DTMS discrimination function 1

-0.1 0.0 0.1 0.2

(b)

0.3 0.4 fresh mastic: independent signal from film thickness

0.5

Depth steps 0.6 0

5

10

15

20

25

30

35

40

45

Depth from surface (µm) Figure 5.3.2.3 MS projections of DF 1 as a function of depth in the ~ 55 µm thick, aged dammar (a) and mastic (b) films.

nm pulses (Sections 3.3.2, 3.3.3, 3.3.5). At this particular depth the ablation step changes upon KrF excimer laser pulses in both films (Figure 3.3.5.3). Since photochemical ablation is basically affected by the degree of oxidation and

279

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

crosslinking of the laser irradiated material (Srinivasan 1994), these findings indicate that the 15 µm depth from surface of both the aged dammar and mastic films is the threshold between highly and poorly deteriorated material. This finding supports earlier findings by De la Rie (1988b), who determined that 10 µm dammar films deteriorate more than thicker films aged under the same conditions. It should be also noted that the data points corresponding to the deepest laser-ablated depth-steps (2025 µm from surface) emerge near the data points of the unaged films. Given that the total film thicknesses were of the order of 55 µm, it should be expected that below a depth of 25 µm from the original surface the composition of the films would eventually become identical to that prior to ageing. 5.3.3 Matrix assisted laser desorption/ionisation (MALDI) time-of-flight (TOF) MS Some quantification of the molecular changes – resulting in the compositional depthdependent gradients – of the aged dammar and mastic films was obtained with MFDA based on the DTMS data. However, the changes monitored provided mean compositional information because the samples required for the DTMS analysis contained the remaining films. MALDI-TOF-MS, using 2,5-dihydroxybenzoic acid (DHB) as the matrix, was carried out to shed light on the composition of the ablated surfaces of the successive depth-steps across the thicknesses of the aged dammar and mastic films (Figure 5.3.3.1). Unfortunately, the low signal-to-noise ratio of the mass spectra obtained did not permit analysis of mastic and all of the depth steps of

280

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

(i)

dammar. The problem of low signal-tonoise ratios of the MALDI/MS was also

f

M

M

d

M

ab

M

c

M

encountered in earlier studies of aged

h

?

(ii)

e? g M

?

M? M

triterpenoid resin films and therefore similar measurements were carried out

M

i

ab

MM

j M

c ? ? M

M

f g

?

(iii) M

with graphite instead of the matrix

M

(GALDI/MS), which results in spectra f M

a M 460

M

M

470

480

c

? M 490

? M

(iv)

g

M

500

mass/charge

Figure 5.3.3.1 MALDI-TOF-MS of the DHB matrix (i), the aged dammar surface (ii), the surface of the film after laser ablation of the uppermost 15 µm (iii) and the surface of the dammar film before ageing (iv). Despite the relatively low peak-to-noise ratio some molecular changes between the aged surface and the ablated surface are discriminated. For the assignments refer to the text.

without chemical noise (from DHB) (Zumbühl, et al. 1998, Dietemann 2003). However, the novel TOF technique with the TLC plates (see Section 5.2.3), employed for the present analysis, is different MALDI/MS

from

the

method.

established It

has

been

observed that with the present MALDI-TOF-MS technique the interference of DHB is minimal, while all the m/z peaks generated correspond to the material on the TLC plates. Hence, the low signal-to-noise ratio is most likely ascribed to the minimal subtraction of resin from the TLC plate. This phenomenon has been observed to be typical of the highly aged resin films (AMOLF, internal report). In the present work crucial conclusions were drawn from the mass spectra at the surface of the aged dammar film prior to ablation, the surface of the remaining film after removal of 15 µm and the surface of the unaged dammar resin. The identification of the mass peaks is tentative and proposed by comparison with data from the

281

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

literature (Zumbühl, et al. 1998, Dietemann 2003). A representative mass / charge range observed in the MALDI-TOF mass spectra is within m/z 450 to 510, at which the sodiated ions [M+Na]+ of unreacted and oxidised triterpenoids are discriminated from the corresponding ion peaks of DHB (Zumbühl, et al. 1998). However, even within this range there are several m/z peaks attributed to DHB (marked with M in Figure 5.3.3.1), which can potentially interfere with the sodiated molecular ions of characteristic triterpenoid compounds (Zumbühl, et al. 1998). A relatively intense peak of DHB is at m/z 451 (not shown), where the sodiated molecular ions of UVinduced oxidation products 17/18 (428 Da, Tables 5.3.1.1, 5.3.1.2) with oxidised Arings are detected (Dietemann 2003). Other overlapping DHB mass / charges are at m/z 461 prohibiting detection of oleanonic and ursonic aldehydes 11/12 (438 Da), and m/z 477 that coincided with [M+Na]+ of oleanonic and ursonic acids, 8/9 (454 Da) (Dietemann 2003). The highest DHB peak within the examined range was at m/z 480 near a marker peak of dammarenolic acid, 5 (458 Da) at m/z 481 due to the [M+Na]+ (Dietemann 2003). In the unaged dammar film surface (Figure 5.3.3.1iv) peaks a, c, f and g at m/z 465, 481, 497 and 497.7 are discriminated respectively. Peak a at m/z 465 is assigned to the sodiated molecular ion of hydroxydammarenone (6) (442 + 23 = 465), c at m/z 481 to [M+Na]+ of dammarenolic acid, 5 (458 Da), and f at m/z 497 to [M+Na]+ of compound 15 (474 Da) (Dietemann 2003). Peak g at m/z 497.7 is attributed to the molecular ion of an oxidised dammarenolic acid 5, [M+Na+O]+, i.e 458+23+16 = 497 (Zumbühl, et al. 1998). Some changes compared to the MALDI MS of the unaged dammar film are monitored in the mass spectrum of the aged film surface (Figure 5.3.3.1ii). It is shown that the

282

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

sodiated molecular ion peak of hydroxydammarenone, 6 (m/z 465, a) is less intense than in the unaged film. This is in good correspondence with the DTMS data showing that there is an abundance reduction of dammarane type triterpenoids upon ageing (Van der Doelen, et al. 1998a). Dammarenolic acid, 5, is recognised by marker peaks, m/z 481, c, which is attributed to [M+Na]+, and m/z 497.7, g, which is attributed to [M+Na+O]+, with the latter peak being more intense compared to the unaged dammar film. The high intensity at m/z 497 (f) indicates that the dammarane type degradation compound 15, which is generated during the biosynthesis of triterpenoid resins upon sunlight irradiation during exudation from the tree (Koller, et al. 1997, Dietemann 2003), is also abundant in the aged film. In the surface of the aged dammar film there are some peaks, such as m/z 467 (b), 483 (d), 495 (e) and 499 (h), which are not detected prior to ageing. These peaks correspond to the sodiated molecular ions of dammarenediol with a molecular weight of 444 Da (Mills and Werner 1955, Mills 1956, Poehland, et al. 1987, De la Rie 1988a), which was not reported in the present DTMS study, (b: 444 + 23 = 467), dihydro-dammarenolic acid 21, (460 Da, d: 460 + 23 = 483), the UV-induced oleanane / ursane seco-products with oxidised A-rings 19/20, (472 Da, e: 472 + 23 = 495) and 20,24-epoxy-25-hydroxy-3,4-seco-dammaran3-oic acid that is a UV-induced dammarane type seco-product with a molecular weight of 476 Da (Van der Doelen, et al. 2000, Scalarone, et al. 2003). The sodiated molecular ion of this particular compound peaks at m/z 499 (476 + 23), h. The MALDI MS of the film after the removal of the 15 µm surface layers is shown in Figure 5.3.3.1iii. It is observed that the marker peaks of the UV-induced secoproducts, at m/z 483, 495, 499 (d, e, h) are not present at this particular depth. Instead,

283

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

peaks at m/z 463 and m/z 491 (i and j respectively) emerge, which are not detected in the mass spectrum of the aged or on the surface of the unaged dammar. These peaks represent the sodiated molecular ions of the oxidation compounds 25/27 (440 Da) and 26/28 (468 Da) respectively (Dietemann 2003). These are typical compounds formed upon oxidation of oleanane/ursane triterpenoids in the absence of UV wavelengths and have been reported in aged dammar varnishes on paintings (Van der Doelen, et al. 1998a, Van der Doelen 1999). This finding is in line with the decreasing number of carbonyl groups according to the ATR-FTIR measurements (Section 4.3.3), as well as with the decreasing optical densities determined with the UV/VIS measurements (Section 4.3.2). Given that the optical absorption lengths, lα , of both the aged dammar and mastic films at λ > 350 nm are longer than 15 µm from the surface (Figure 4.3.2.6), it is suggested herein that A-ring oxidation occurs under irradiation with λ < 350 nm. This suggestion agrees with findings by Van der Doelen and co-workers (1999, 2000), who determined the A-ring oxidation upon irradiation with λ > 315 nm and the non-UV-induced oxidation upon irradiation with λ > 380 nm. Regardless of the wavelengths employed, it is demonstrated here that the bulk of aged dammar and mastic films is protected from UV irradiation anyway. 5.3.4 A study on the molecular weight across depth of aged dammar and mastic films with High Performance - Size Exclusion Chromatography (HP-SEC) Although a decreasing gradient in the condensed, high molecular weight fraction of the aged dammar and mastic films was indicated with the weakening of the EI-DTMS TIC’s pyrolysis peak with depth (Figure 5.3.1.3), size exclusion chromatography was carried out to determine depth-wise molecular weight changes. Interpretation was

284

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

carried out using earlier SEC studies of triterpenoid varnishes (De la Rie 1988b, Van der Doelen 1999). De la Rie (1988b) used Gel Permeation Chromatography (GPC), which is a synonymous method to SEC, and determined an average increase in the molecular weight of accelerated aged dammar resins up to 1500 Da. Zumbühl and coworkers (1998) also reached the same conclusion measuring the mass/charges of light aged dammar films on the basis of GALDI/MS. Van der Doelen and co-workers (1999, 2000), used SEC with UV (240 nm) and VIS (400 nm) detectors to detect compatibility of naturally aged varnishes and accelerated aged resins. In the present work, SEC traces at 240 and 400 nm of the unaged, aged and laser ablated dammar and mastic films were studied to detect MW changes across depth. The same quantity of material was used from all the films tested and dissolved in THF (~10mg/µl), while about 20 µl of the prepared solutions were injected in the column. It was therefore expected that the final chromatograms would provide information about the molecular weight distribution across depth as a function of the absolute absorbance at 240 and 400 nm. However, this was not possible because of the depth-dependent gradients of the optical properties of the aged films (Figures 4.3.2.1b, 4.3.2.2b), which affected the intensities of the resulting SEC traces accordingly. Therefore, all the SEC traces were normalised at a % relative absorption. Figure 5.3.4.1 presents the SEC traces at 240 and 400 nm of the dammar film prior to and after accelerated ageing. The molecular weight increase upon ageing is evident with both detectors. Absorption at both wavelengths is separated in a small fraction of sesquiterpenoids (~ 200 Da), the group of triterpenoid molecules (400/500 Da), which being the most abundant compounds absorb strongly, a small fraction of dimerised

285

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

and oxidised triterpenoid molecules (900/1000 Da) and traces of the condensed high MW fraction at 10 kDa, that are mainly formed upon the radical polymerisation of polycadinene (Van Aarssen, et al. 1990). The highest absorption at 240 nm of the aged dammar film is obtained at 900/1000 Da, while the triterpenoid molecules (400/500 Da) absorb slightly less than their dimers, forming a double peak within the range 400 to 1000 Da. This double peak broadens towards the higher MW regime, up to 15 kDa. Absorption of the ~ 200 Da group is critically reduced upon ageing indicating that both low MW components and triterpenoid molecules participate in the formation of longer molecular chains during degradation or that the low MW fraction escaped the film (Boon and van der Doelen 1999). The SEC trace at 400 nm of the aged film shows that the group of molecules absorbing intensely in the blue have a molecular weight in the order of 400 Da. This MW points to triterpenoid molecules, while the second intense peak at 800/900 Da shows that oxidised and dimerised triterpenoids absorb strongly in the blue region. SEC traces of aged varnishes from paintings at 400 nm detected a MW increase absorbing strongly at molecular weights greater than 1000 Da, while artificially aged dammar resins in solution absorbed intensely at ~ 2000 Da with shoulders at higher molecular weights (Van der Doelen 1999, Van der Doelen and Boon 2000). This mass increase was associated with yellowing. In the present work, the strongest absorption in the blue was obtained at 400 Da.

286

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

100

% Relative absorption

80

aged at 240 nm unaged at 240 nm aged at 400 nm unaged at 400 nm

60

40

20

0 10000

1000

100

Molecular Weight (Da) Figure 5.3.4.1 SEC traces of the aged and unaged dammar films at 240 nm and 400 nm.

Figure 5.3.4.2 demonstrates the SEC traces at 240 nm of consecutive depth-steps down to 15 µm from the surface of the aged dammar film and the SEC trace of the same dammar film prior to ageing. It is observed that the molecular weight across the highly deteriorated surface layers of the aged dammar film does not change significantly, which is in agreement with the intense carbon dimer emission upon 248 nm laser ablation (Figure 3.3.4.2) as well as with the constant ablation step in the 15 µm surface layers (Figure 3.3.5.3). The results agree also with the DTMS data of the 15 µm surface layers of the aged dammar film, showing that there is little change in the abundance of the mass/charges corresponding to the triterpenoid molecules, (400/500 Da). A small shoulder peak, increasing at about 200 Da on the SEC traces at 240 nm (Figure 5.3.4.2) of the ablated depth-steps, indicates the presence of some low MW products with depth, which eventually escape in the atmosphere (Boon and van der Doelen 1999). Figure 5.3.4.3 shows the SEC traces of the same depth-steps of the

287

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

aged dammar film at 400 nm and the SEC trace of the film prior to ageing. Two facts are apparent in this chromatogram. First, ageing results in a slight increase of the MW 100

3-6

% relative intensity

80

1 60

40

2 20

0 100000

10000

1000

100

Molecular Weight (Da) Figure 5.3.4.2 HP-SEC plots of dammar at 240 nm before ageing (1), after ageing (2) and those of the laser ablated depth steps between 3.5 and 15 µm (3-6). 100

2 80

% relative intensity

3-6 60

1 40

20

0 100000

10000

1000

Molecular Weight (Da) Figure 5.3.4.3 HP-SEC plots of dammar at 400 nm. For comparison the traces of the unaged (1), aged (2) and the ablated depths from 3.5 to 15 µm (3-6) are demonstrated.

288

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

of the triterpenoid compounds (400/500 Da), which is related to oxidation. Second, there is an abrupt decrease in the absorption from the crosslinked fraction of the dimerised triterpenoids (900 Da). These products participate in the formation of yellow compounds absorbing strongly in the blue and are responsible for the colouration effect in the aged dammar (Van der Doelen and Boon 2000). The decrease in the absorption at 400 nm between the aged surface and the layers below, especially at 900 Da, indicates a decreasing degree of yellowing with depth, which has been determined in laser ablated aged varnishes on paintings (Theodorakopoulos and Zafiropulos 2003, Section 1.4). 100

% Relative absorption

80

aged at 240 nm unaged at 240 nm aged at 400 nm unaged at 400 nm

60

40

20

0 100000

10000

1000

Molecular Weight (Da) Figure 5.3.4.4 HP-SEC traces of the aged and unaged mastic film at 240 and 400 nm

Figure 5.3.4.4 presents the SEC traces of mastic prior to and after ageing at 240 nm and 400 nm. At 240 nm the unaged film absorbs strongly at 400 Da owing to the triterpenoid molecules. There is a shoulder peak at ~ 1000 Da caused by oxidised triterpenoid dimers and an apparent hump at ~ 10 kDa that represents a condensed fraction of mastic. At 400 nm the unaged film absorbs strongly at ~ 300 Da owing to

289

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

unsaturated triterpenoid ketones and quinones (Formo 1979) that are responsible for yellowing (Figure 2.2.4). The fraction of 1000 Da absorbs less intensely at 400 nm than at 240 nm, while there is also a small fraction at 10 kDa that absorbs the radiation equally at both wavelengths. Upon ageing absorption due to triterpenoids molecules (~400 Da) is strong in both wavelengths, but a shift towards higher MW indicates oxidation. The increased absorption at 1000 Da (240 nm) and ~ 900 Da (400 nm), indicates the increasing concentration in oxidised triterpenoid dimers (De la Rie 1988b). The broadening of the peaks suggests the existence of triterpenoid trimers, tetramers and longer crosslinked chains that appear as shoulders at 10 kDa. Absorption at 400 nm is relatively strong from about 3000 Da to almost 80 kDa compared to the SEC trace at 240 nm. According to Van der Doelen and Boon (2000) these long molecules absorb strongly in the blue and are responsible for the yellowing effect of the triterpenoid varnishes. Indeed, unlike the aged dammar film, the aged mastic film obtained a stronger yellow hue after several weeks in the dark following an extreme light ageing procedure, which initially bleached the yellow chromophores (Chapter 2). Therefore, the depth-dependent molecular weight decrease of the aged mastic film is better illustrated with the SEC chromatograms at 400 nm (Figure 5.3.4.5). The SEC traces of the unaged mastic film and the aged film prior to ablation are shown for reasons of comparison. It is observed that there is an evident gradient in the absorption from the high MW fraction across depth. In particular, the abundances of the 900/1000 Da components and the condensed fraction in the high MW regime up to 80 kDa are gradually decreasing with depth. As in the case of DTMS, sampling for SEC included the remaining varnish film (total thickness ~ 55 µm) resulting in average molecular weight determination of the remaining films. Therefore, the 290

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

absolute molecular weight as a function of depth will most likely display a sharper gradient than that demonstrated in Figure 5.3.4.5.

Relative absorption at 400 nm (%)

100

aged surface - 16 µm

80

- 3 µm 60

40

- 27 µm 20

unaged 0 100000

10000

1000

100

Molecular Weight (Da) Figure 5.3.4.5 HP-SEC traces at 400 nm of the aged and unaged and laser ablated mastic film at successive etchings at 3, 6, 9, 12, 16, 20, 25 and 27 µm from the surface. The apparent MW reduction across the depth profile is evident.

5.4 Rationalisation of triterpenoid varnish ageing Recently an ageing model of triterpenoid varnishes has been proposed (Boon and van der Doelen 1999). The model was based on experimental observations of molecular studies carried out with GCMS and LCMS analysis of fresh and aged triterpenoid varnishes and orientational studies on sterols, which have comparable carbon skeletal ring structures as the triterpenoid molecules. In view of the current mass spectrometric and chromatographic analytical data based on DTMS, MALDI-MS, HP-SEC, UV/VIS and ATR-FTIR (Chapter 4) on successive depths of the excimer laser ablated triterpenoid films, an updated schematic model on the degradation of these varnishes is proposed. The current findings are integrated into the previous model, aiming at a better understanding of the nature of aged triterpenoid varnishes. In line with the 291

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

previous model (Boon and van der Doelen 1999), we assume the existence of a considerable degree of orientation of the triterpenoid molecules forming flat sheet-like structures on a molecular level before the degradation process begins. A first phase in triterpenoid varnish degradation corresponds to photoinitiation and the early oxidation stages. As soon as the varnish is exposed to light, the triterpenoid molecules lying on the surface varnish layers absorb the short wavelengths of the incident ambient radiation and hence the longer wavelengths (λ > 350 nm) are transmitted to the triterpenoids in the bulk. In line with Beer’s Law, absorption of light from the resin compounds reduces logarithmically across the depth of the film. The depth-dependent qualitative and quantitative reduction of light leads to decreasing degrees of excitation. Thus, the free radicals formed across depth are quenching as the light propagates deeper into the bulk. Given the findings in the present work it is understood that the free radicals in the surface layers are more abundant than in the bulk because more intense bond-breakage occurs in the surface. Certainly, the presence of oxygen and the mobility of the radicals into the triterpenoid films play a significant role in the degradation process and the delineated quenching with depth. The excited molecules and the free radicals generate new covalent bonds between each other forming so new high molecular weight compounds. Because of the gradient in light propagation (Chapter 4), the extent of this condensation process decreases accordingly with depth. At the same time, the oxygen diffusion from the air-surface interface into the bulk is also decreasing (Thomson 1965, 1979), thereby leading to high degrees of functionalisation on the surface layers and oxygen depletion in the bulk. Because of 292

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

the absorption of the short wavelengths (λ < 350 nm) from the surface layers only, different oxidation products are formed in the surface and the deeper layers, which is supported by the existence of A-ring oxidation at C-2 in the surface layers and oleanane / ursane type molecules oxidised at C-11 and C-12 in the bulk (Section 5.3.3). The degree of oxygen consumption across depth is important for the propagation reactions (Section 2.2), which originally are initiated in an decreasing rate across the depth profile of the film (top to bottom) (Schoolenberg and Vink 1991). Hence, depending on the degree and the type of oxidation, water molecules are attracted by the oxidised triterpenoids across the film thickness. This attraction leads to hydrogen bonding between the polar groups and the water molecules creating weak hydrogen bonded ‘crosslinks’ (Boon and van der Doelen 1999), the abundance of which decreases with depth. The sheet-like orientations of the triterpenoid compounds are disturbed, especially at the surface, as soon as these structural modifications take place from the first stages of degradation. Both types of crosslinking reduce the mobility of the triterpenoid compounds especially on the surface, thereby leading to surface stiffness, the degree of which is accordingly reducing with depth. At the same time an equivalent gradient in polarity across thickness is also formed. At a second phase, the varnish proceeds to further degradation that is led by further oxygen consumption, bond breaking and polymerisation processes. The already modified triterpenoid molecules at the uppermost layers of the film are further degraded forming smaller molecular species, which no more have the characteristic triterpenoid carbon ring skeleton, as determined by analytical data based on GC and HPLC (Van der Doelen, et al. 1998b, Boon and van der Doelen 1999). The high

293

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

degree of functionalisation of the triterpenoids at the surface results in even stronger absorption of the short wavelengths by the outer layers, which prevents penetration of UV wavelengths into the bulk. The overall propagation of light into the film is shortened further and Beer’s Law is no longer valid for all the film thickness because of the abrupt optical changes across depth. This is now well understood, because of the decreasing trends of the absorption coefficients progressing into the depths of aged mastic and dammar films (Section 4.3.2) (Zafiropulos, et al. 2000). According to the previous theoretical model on triterpenoid ageing, small degradation products, such as functionalised carbon compounds (aldehydes, ketones and acids) yielded from bond breaking and ring disintegration, escape to the atmosphere (Boon and van der Doelen 1999). At the same time bond-breakage, excitations and oxygen consumption is continued gradually in the underlying layers although all trends would be decreasing with time (Thomson 1965, 1979). Hence, further attraction of water molecules is consistent with the degree and the depth of oxygen consumption across the depth-profile. Crosslinked fractions have reduced molecular weight with depth. Crosslinks at the surface may well contain more polar that apolar groups, changing gradually the plasticity of the varnish at the exposed surface. A final phase corresponds to further degradation of highly deteriorated triterpenoid varnish. Elimination reactions in the high molecular weight condensed fractions may lead to formation of aromatic centres, further influencing the light absorbing properties of the surface layers. On the surface, residual polar triterpenoid and polar parts of high molecular weight species are linked with water molecules. The high concentration of the three dimensional covalent-bonded polymer network may lead to

294

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

a certain degree of insolubility of the surface layers. Further exposure to radiation leads to depolymerisation of the surface by continuous bond-breakage resulting from excitation of the absorbing functional groups attached to the condensed surface layers. On the other hand, the bulk is preserved in a somewhat ‘unaged’ state, since all the deteriorative actions are carried out in the uppermost layers. Both the extent of polarity and crosslinking in the deep layers are comparable to that of the starting film, protecting the substrate from the deteriorative consequences of the ambient conditions.

5.5 Conclusions In conclusion, deterioration of dammar and mastic varnish films formed upon ageing is critically dependent on their thicknesses. Based on the results shown above it is determined that thick dammar and mastic films protect photosensitive substrates from deteriorative UV light wavelengths and from exposure to oxygen, preserving them in the best possible condition for a considerable lifetime. The transition from the highly degraded surface to the less reacted or even unreacted bulk layers, is a matter of a depth-dependent gradient both in terms of oxidation, as shown herein with DTMS, and in terms of the photochemically induced high MW condensed material, detected by HP-SEC. The presented results support all the findings demonstrated above on the change of the interaction of the aged resin films with the UV laser photons (Figures 3.3.2.2, 3.3.2.4, 3.3.3.1), the decreasing carbon dimer emission upon consecutive laser pulses (Figures 3.3.4.2a, b), the reduction of the ablation step (Figure 3.3.5.3), the change of the laser-induced transmission rate (Figures 4.3.1.2.2a, b, 4.3.1.2.4), the decreasing optical densities (Figures 4.3.2.1b, 4.3.2.2b, 4.3.2.4b) and the decreasing

295

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

number of carbonyls (Figures 4.3.3.2i, ii, 4.3.3.3a, b, 4.3.3.4) as a function of depth, on the molecular level. Oxidation not only reduces gradually as a function of depth, but the transition from highly oxidative and polar states in the surface towards nonoxidative states in the bulk is, in addition, qualitative. Based on the MALDI-TOF-MS results, it is determined herein that UV-induced oxygenated triterpenoid compounds, such as oleanane / ursane type molecules with oxidised A-rings (Figure 2.2.2c), are formed only in the 15 µm surface layers despite the fact that both the 55 µm films tested were irradiated with light of λ > 295 nm for about 500 hours (Section 2.7.2.2). At less than 15 µm below surface only oleanane / ursane type molecules with side chain oxidation at positions C-11 and C-28 (Figure 2.2.2b); these are detected as typical compounds of oxidised varnishes found in paintings (Van der Doelen, et al 1998a). This finding is in line with the optical absorption lengths of light with λ < 350 nm that is completely absorbed in the 15 µm surface layers (Figure 4.3.2.6), which indicates that wavelengths shorter than 350 nm are responsible for the A-ring oxidation. Hence, wavelengths longer than 350 nm penetrate in the deeper layers of the film, as also shown in Figure 4.3.2.6. Similar results have been reported by (Feller 1994b), who noted that the penetration depth of 350 nm radiation is ten times that of 320 nm into aromatic polyester films. Although oxidation is considered as the main problem of natural resin varnishes (Section 2.2), the decreasing gradient in the high MW condensed fraction that is formed during photochemical degradation should not be underestimated. Provided that high MW crosslinks are insoluble (Stolow 1985, De la Rie 1988b), the suggestion that upon chemical treatments only the volatile compounds are removed and the condensed material remains on the treated surface (Boon and van der Doelen 1999) is fully supported here. As discussed in Section 296

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

2.2.2, this potential influences the appearance of the substrate, especially if this is paint, because of the overall rise of the MW and the high viscosity (De la Rie 1987, Berns and De la Rie 2002, 2003) formed in the new varnish, which eventually will be integrated with the high MW condensed fraction of the older varnish that remains on the surface. This possibility and the fact that no oxidative contribution of the 248 nm laser pulses to the remaining films was detected is a good argument for the excimer laser ablation of aged varnish films.

297

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

5.6 References Berns, R. S. and De la Rie, E. R. 'The relative importance of surface roughness and refractive index in the effects of varnishes on the appearance of painitngs'. In Preprints of the 13th triennial meeting of the ICOM Committee for Conservation, Ed. V. R., Vol. I, James & James (Science Publishers) Ltd, Rio de Janeiro, (2002), 211216. Berns, R. S. and De la Rie, E. R., 'The effect of the refractive index of a varnish on the appearance of oil paintings', Studies in Conservation 48 (2003) 251-262. Boon, J. J., 'Analytical pyrolysis mass spectrometry: new vistas opened by temperature-resolved in-source PYMS', International Journal of Mass Spectrometry and Ion Processes 118/119 (1992) 755-787. Boon, J. J. and van der Doelen, G. A. 'Advances in the current understanding of aged dammar and mastic triterpenoid varnishes on the molecular level'. In Firnis: Material - Aesthetik - Geschichte, International Kolloquium, Braunschweig, 15-17 Juni 1998, Ed. A. Harmssen, Hertog-Anton-Ulrich-Museum, Braunschweig, (1999), 92-104. Carlyle, L. A. The artist's assistant: Oil painting Instruction manuals and handbooks in Britain 1800-1900: with reference to selected eighteenth century sources, Archetype Publications, London, 2001. Castillejo, M., Martin, M., Oujja, M., Silva, D., Torres, R., Manousaki, A., Zafiropulos, V., van den Brink, O. F., Heeren, R. M. A., Teule, R., Silva, A., and Gouveia, H., 'Analytical study of the chemcial and physical changes induced by KrF laser cleaning of tempera paints', Analytical Chemistry 74 (2002) 4662-4671. Cunliffe, A. V. and Davis, A., 'Photo-oxidation of thick polymer samples. Part II: The influence of oxygen diffusion on the natural and artificial weathering of polyolefins', Polymer Degradation and Stability Vol. 4 (1982) 17 - 37. De la Rie, E. R., 'The influence of varnish on the appearance of paintings', Studies in Conservation 32 (1987) 1-13. De la Rie, E. R., 'Stable Varnishes for Old Master Paintings', PhD Thesis University of Amsterdam, (1988a). De la Rie, E. R., 'Photochemical and thermal degradation of films of dammar resin', Studies in Conservation 33 (1988b) 53-70.

298

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

Dietemann, P., 'Towards more stable natural resin varnihses for painintgs', PhD Thesis Swiss Federal Institute of Technology, Zurich, (2003). Feller, R. L. 'Depth of Penetration of Light into Coatings; and Influence of Sample Thickness and Oxygen Diffusion'. In Accelerated Aging: Photochemical and Thermal Aspects, Ed. D. Berland, The Getty Conservation Institute, USA, (1994a), 56-61 and 135-137. Feller, R. L. 'Depth of Penetration of Light into Coatings'. In Accelerated Aging: Photochemical and Thermal Aspects, Ed. D. Berland, The Getty Conservation Institute, USA, (1994b), 56 - 61. Formo, M. W. 'Paints, varnishes and related products: Discolouration'. In Baile's Industrial Oil and Fat Products, Ed. D. Swern, Vol. 1, John Wiley & Sons, New York, (1979), 722-724. Fukushima, T., 'Deterioration Processes of Polymeric Materials and their Dependence on Depth from Surfaces', Durability of Building Materials Vol. 1 (1983) 327 - 343. Hoogerbrugge, R., Willig, S. J., and Kistemaker, P. G., 'Discriminant analysis by double stage principal component analysis', Analytical Chemistry 55 (1983) 17101712. Koller, J., Baumer, U., Grosser, D., and Schmid, E. 'Mastic'. In Baroque and Rococo Lascquers, Ed. J. Koller, Vol. 81, Arbeitshefte des Bayerischen Landesamtes fuer Denkmalpflege, Karl M. Lipp Verlag, Muenchen, (1997), 347-358. Mantell, C. L., Kopf, C. W., Curtis, J. L., and Rogers, E. M. 'Oil Varnishes'. In The technology of natural resins, John Wiley & Sons, Inc., (1949), 265-319. Mills, J. S. and Werner, A. E. A., 'The chemistry of dammar resin', Journal of Chemical Society (1955) 3132-3140. Mills, J. S., 'The constitution of the natural, tetracyclic triterpenes of dammar resin', Journal of Chemical Society (1956) 2196-2202. Mills, J. S. and White, R. The Organic Chemistry of Museum Objects, 2nd edition, Butterworth-Heinemann, Oxford, 1994. Papageorgiou, V. P., Bakola-Christianopoulou, M. N., Apazidou, K. K., and Psarros, E. E., 'Gas chromatographic-mass spectrometric analysis of the acidic triterpenic fraction of mastic gum', Journal of Chromatography A 769 (1997) 263-273.

299

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

Poehland, B. L., Carte, B. K., Francis, T. A., Hyland, L. J., Allaudeen, H. S., and Troupe, N., 'In vitro antiviral activity of dammar resin triterpenoids', Journal of Natural Products 50 (1987) 706-713. Scalarone, D., van der Horst, J., Boon, J. J., and Chiantore, O., 'Direct-temperature mass spectrometric detection of volatile terpenoids and natural terpenoid polymers in fresh and artifficially aged resins', Journal of Mass Spectrometry 38 (2003) 607-617. Schoolenberg, G. E. and Vink, P., 'Ultra-violet degradation of polypropylene: 1. Degradation Profile and thickness of the embrittled surface layer', POLYMER Vol. 32 No 3 (1991) 432 - 437. Srinivasan, R. and Braren, B., 'Ultraviolet laser ablation of organic polymers', Chemical Reviews 89 (1989) 1303-1316. Srinivasan, R. 'Interaction of laser radiation with organic polymers'. In Laser Ablation: Principles and Applications, Ed. J. C. Miller, Vol. 28, Springer Series of Material Science, Springer, Berlin, Heidelberg, (1994), 107. Stolow, N. 'Part II: Solvent Action'. In On picture varnishes and their solvents., Ed. E. H. Jones, Revised edition 1971. Cleveland, Ohio: Case Western Reserve University. Revised and enlarged edition 1985. Washington DC: National Gallery of Art., (1985). Theodorakopoulos, C. and Zafiropulos, V., 'Uncovering of scalar oxidation within naturally aged varnish layers.' Journal of Cultural Heritage (Suppl. 1) 4 (2003) 216s222s. Theodorakopoulos, C., Zafiropulos, V., Fotakis, C., Boon, J. J., van der Horst, J., Dickmann, K., and Knapp, D., 'A study on the oxidative gradient of aged traditional triterpenoid resins using 'optimum' photoablation parameters', In Lacona V Proceedings, Osnabrück, Germany, September 15-18, 2003, Eds. K. Dickmann, C. Fotakis, and J. F. Asmus, Springer Proceedings in Physics, Vol. 100, SpringerVerlang, Berlin Heidelberg, (2005), 255-262. Thomson, G. 'Topics in the conservation chemistry of surface.' In Application of Science in Examination of Works of Art, Museum of Fine Arts, Boston, (1965), 78 85. Thomson, G., 'Oxygen Starvation in Paint and Other Films', National Gallery Technical Bulletin Vol. 2 (1978) 66 -70. Thomson, G., 'Penetration of Radiation into Paint Films', National Gallery Technical Bulletin Vol. 3 (1979) 25 -33.

300

C. Theodorakopoulos

Chapter 5

A molecular study on the depth-dependent gradients

Van Aarssen, B. G. K., Cox, H. C., Hoogendoorn, P., and De Leeuw, J. W., 'A cadinene biopolymer present in fossil and extant dammar resins as a source for cadinanes and bicadinanes in crude oils from Southern Asia', Geochimica et Cosmochimica Acta 54 (1990) 3021-3031. Van der Doelen, G. A., van der Berg, K. J., and Boon, J. J., 'Comparative chromatographic and mass spectrometric studies of triterpenoid varnishes: fresh material and aged samples from paintings', Studies in Conservation 43 (1998a) 249264. Van der Doelen, G. A., Van der Berg, K. J., Boon, J. J., Shibayama, N., De la Rie, E. R., and Genuit, W. J. L., 'Analysis of fresh triterpenoid resins and aged triterpenoid varnishes by high-performance liquid chromatography-atmospheric pressure chemical ionisation (tandem) mass spectrometry', Journal of Chromatography A 809 (1998b) 21-37. Van der Doelen, G. A., 'Molecular studies of fresh and aged triterpenoid varnishes', PhD Thesis University of Amsterdam, (1999). Van der Doelen, G. A., Van den Berg, K. J., and Boon, J. J., 'A comparison of weatherometer aged dammar varnish and aged varnishes from paintings'. In Art Chimie: La Couleur: Actes du Congres, Ed. J.-P. Mohen, CNRS Editions, Paris, (2000), 146-149. Van der Doelen, G. A. and Boon, J. J., 'Artificial ageing of varnish triterpenoids in solution', Journal of Photochemistry and Photobiology A: Chemistry 134 (2000) 4557. Windig, W., Haverkamp, J., and Kistemaker, P. G., 'Interpretation of sets of pyrolysis mass spectra by discriminant analysis and graphical rotation', Analytical Chemistry 55 (1983) 81-88. Zafiropulos, V., Manousaki, A., Kaminari, A., and Boyatzis, S., 'Laser Ablation of aged resin layers: A means of uncovering the scalar degree of aging', ROMOPTO: Sixth Conference on Optics, Vlad V. I. (Ed.), SPIE Vol. 4430 (SPIE The International Society for Optical Engineering, Washington, (2001) 181-185. (2000). Zafiropulos, V. 'Laser ablation in cleaning of artworks'. In Optical Physics, Applied Physics and Material Science: Laser Cleaning, Ed. B. S. Luk'yanchuk, World Scientific, Singapure, New Jersey, London, Hong Kong, (2002), 343-392. Zumbühl, S., Knochenmuss, R., Wülfert, S., Dubois, F., Dale, M. J., and Zenobi, R., 'A graphite-assisted laser desorption/ionisation study of light-induced ageing in triterpene dammar and mastic varnishes.' Analytical Chemistry 70 (1998) 707-715. 301

C. Theodorakopoulos

Perturbation Methods ALI HASAN NAYFEH Copyright Q 2004 WILEY-VCH Valag GmbH & Co. KGaA

References and Author Index Ablowitz, M. J., and D. J. Benney, (1970). The evolution of multi-phase modes for nonlinear dispersive waves. Stud. Appl. Math., 49,225-238. [234] Abraham-Shrauner, B. (197Oa). Suppression of runaway of electrons in a Lorent2 plasma. I. Harmonically time varying electric field. J. phsmrr Ahys., 4,387402. [235] Abraham-Shrauner, B. (1970b). Suppression of runaway of electrons in a Lorentz plasma. 11. Crossed electric and magnetic fields. J. Phrmrr B y . 4,441450. [235] Ackerberg, R. C., and R. E. O’Malley, Jr. (1970). Boundary layer problems exhibiting resonance. Stud. AppL Math., 49,277-295. [233] Ahluwalia, D. S., R. M. Lewis, and J. Boersma, (1968). Unifonn asymptotic theory of diffraction by a plane screen. S U M J. AppL k t h . , 16,783-807 [3801 Akins, R. G. See Chang, Akins, and Bankoff. Akinsete, V. A., and J. H. S. Lee, (1969). Nonsimilar effects in the collapsing of an empty spherical cavity in water. Phys Fluids, 12,428434 /78] Albright, N. W. (1970). Quasihear stabilization of the transveme instability. Phys. Fhrido, 13 1021-1030.[235] Alfriend, K. T.(1970). The stability of the triangular Lpgrangian points for commensurability of order two. Celesriol Mech., 1,351-359. I2331 Uriend, K. T. (1971a). Stability and motion in two-degreeaf-freedom Hamiltonian systems for two-twne commensurability.CeZestkfMech, 3,247-265. (2331 Alfriend,K.T.(1971b). Stability of and motion about L, at three-to-one commensurability. CelestiaZMech.,4,60-77. [233] M e n d , K. T. See Davis and Alfriend. Alfriend, K. T., and R. H. Rand, (1969). Stability of the trhnguh points in the elliptic restricted problem of three bodies. A U A J., 7,1024-1028. [233,260,262,276] Al’tshul’, L. M., and V. I. Karpman, (1966). Theory of nodinear oscillations in a collisionless plasmaSovietPhys. JEW (English trand.), 22, 361-369. [369] Alzheimer, W. E., and R. T. Davis, (1968). Unsymmetrical bending of prestressed annuh Plates. J. Eng. Mech. Div. Proc. ASCE, 4, 905-917. (35, 128) Amazigo, J. C., B. Budiansky, and G. F.Carrier (1970). Asymptotic analyses of the buckling of imperfect columns on nonlinear elastic foundations. Znt. J. Solids Structures. 6, 1341-1356. [233] Anderson, W. J. See Spriggs, Messiter, and Anderson. Asano, N. (1970). Reductive perturbation method for nonlinear wave propagation in inhomogeneous media. IIl. J. Phyd Soc. Jupun,29,220-224. [781 Asano, N. and T. Taniuti, (1969). Reductive perturbation method for nonlineax wave propagation in inhomogeneous media. I. J. Phys. Soc. J a m , 27,1059-1062.

387

388

REFERENCES AND AUTHOR INDEX

N., and T. Taniuti (1970). Reductive perturbation method for nonlinear wave propagation m inhomogeneous media Il. J. Phys. Soc. Japun, 29,209-214.[78] Ashky, H. (1967). Multipie pcaliag in flight vehicle dynamic analysis - a preliminary look. AIAA Paper No. 67-560.(2331 Avila, G. S. S., and J. B. Keller, (1963). The high frequency asymptotic field of a point source in an inhomogeneous medium. Cbmm Aue AppL Ahth.. 16,363-381. [ 3801 Babich, V. M. (1965). A point source in an inhomogeneow medium. 2. V y d . Mar. f Mut. Fyz., 5,949-951;EpglishtrcmsL,usSRCbmput~rkloth.Mth.Ph~8., 5,247-251.[380] Babich, V. M. (1970). MarhemtiuaI ptoblcms in Wave Ropugation Thawy, Part I. Plenum, New York. [ 361,3801 Babich, V. M. (1971).hfathemaffalh b k m s in Wave hpugation Theury. Part II. Pknum, New York. [361.380] Babich, V. M., and T. S. Kravtsova, (1967). Propagation of wave film type oscillations of quantizedthfclmcrs.P W . 31,204-210. [380] Balcscu,R. (1963). Smt&tidMech&sof CkgdAvrider Why, New York. [367.369) BaU, R. H. (1964). Nonlinea~theory of waves in a cold homogeneous plasma without magnetic &ld. Stauford Microwave Laboratory Rept. No. 1200. [234] Bankoff, S. G. See Chang, Akius, and Bankoff. Barakat, R., and A. Houston, (1968). Nonlinear periodic capiliarymvity waves on a fluid of fdte depth.J. Ceophyg Rer., 73 65456554.[771 Barcilon. V. (1970). Some inertial modifications of the h e a r viecous theory of steady rotating fluid flowaPh~aFhddr. 13,537-544.[235] Barnu, R. B. (1970). Converlpnce of the von Zeipel procedure. CehtiulMe&, 2,494-504. (1921 Barua, S . N. (1954). sbcondary flow in a rotating stm&#t pipe. Roc. Roy. Soc. (London), A227,133-139.[79] Bauer, H. F. (1968). Nonlineat response of elastic plates to pulse excitations. J. AppL Mech., 35,47-52.[S8] Bellman, R. (1955). Remuch problems. Buu Am Math Soc., 61,192.[22] Bellman, R. (1964). PaMkrtfon Techniques in Mathemrrtia, Physics, and Engineering. Holt, New York. [3091 Benney, D. J. (1965). The flow induced by a disk oscillating about a state of steady rotation. QUd. J. M.APPL M t h . , 18,333-345.[235] Benney, D.J. (1%6). Long waves on liquid h . J . Moth. amfPhyx 45,150-155.[41] Benney, D. J.(1%7). The asymptotic bellavior of nodinear dispersive waves. J. Marh. und P&&, 46,115-132.[234] Benney, D. J. See Ablowitz and €bnney. Bcnney, D. J., and A. C. Newcll, (1967). Sequential time closures for interacting random waves. J. b¶iath. and my&, 4,363-393.(2341 Bcnney, D. J., and A. C. Nenen, (1969). Random wave closures. Stud. Appl. Moth.,48, 29-53.[234] m y , D. I., and G. J. R o b s , (1969). Wave instabilities. Sfud. AppL Muth.,48, 377-385. [2351 Benney, D. J. and P. G. Wman, (1%6). Nonlinear mteractions of random waves in a dhpcrsb medium.Awz Roy. Soc. (London). -9,301-320. [234] Bethe, H. A. See Salpeter and &the.

-0,

REFERENCES AND AUTHOR INDEX

389

Bisshopp, F.(1969). A modified stationary principle for nonlinear waves. J. Diff. Equations, 5,592405. [216] Bleistein, N. (1967). Uniform asymptotic expansions of integrals with many nearby stationary points and algebraic singulatities. J. Math. Mech., 17,533359.[380] Bleistein, N. See Lewis, Bleistein, and Ludwig. Boersma, J. See Ahluwalia, Lewis, and Boersma. Bogoliubov, N. N. See Krylov and Bogoliubov. Bogoliubov, N. N., and Y. A. Mitropolski, (1961). Asymptotic Methods in the Theory of Nonlineor Osci&@on& Gordon and Breach, New York [ 168,174J Bohlin, K. P. (1889). Ube~eine neue Annaherungsmethode der Storungstheorie, Akad Handl. BUmng.. 14,(Afdl, Stokholm). [56] Born, M. (1926).Quantum mechanics of impact processes. 2.Phyr, 38,803-827.(3621 Bourret, R. C. (1962a). Propagation of randomly perturbed Filds. &n. J. Phys., 40, 782-790.[367,369,372] Bourret, R. C. (1962b) Stochastically perturbed fwlds, with applications to wave propagation in random media. Nuovo Cimenro, 26, 1-31.[ J 367,369,3721 Brau, C. A. (1967). On the stochastic theory of dissociation and recombination of diatomic molecules in inert dilutens. J. Chem Phyr, 47,1153-1163.[235] Breakwell, 1. V., and L. M. Perko, (1966). Matched asymptotic expansions, patched conics, and the computation of interplanetary trajectories. In &ogress in Asfromutics and A m m u t i a , Vol. 17, Method in AJhrodymmics and Celestid Mechanics (R. L. Duncombe and V. G. Szebehely, Eds.), Academic, New York, pp. 159-182.[ 1381 Breakwell, J. V., and R. Pringle, Jr.(1966). Resonances affecting motion near the earth-moon equilateral libration points. In Rogress in Astronmrticsand Aeronautics, Vol. 17,Methods in A d y ~ m i c s a n CelesrrblMechanics d (R. L. Duncombe and V. G. Szebehely, Eds.), Academic, New York,pp. 55-74.[ 1991 Bretherton, F. P. (1962). Slow viscous motion round a cylinder in a simple shear. J. a i d Mech., 12,591613.[110] Bretherton, F. P. (1964). Resonant interactions between waves. The case of discrete oscillations. J. FluidMech., 20,457479. [217,227,267,298] Bretherton, F. P. (1968). Propagation in slowly varying waveguides hoe. Roy. Soc. (London), A 302,555-576.1216) Bretherton, F.P., and C. J. R. Garrett (1968).Wavetrains in inhomogeneous moving media. Roc. Roy. Soc fLondon).A302,529-554. [216J BriUouin, L. (1926).Remarques sur la mecanique ondulatoire. J. Phyr Rudium, 7,353-368. [ 114,315,3391 Brofman, S. See Ting and Brofman. Brofman, W. (1967). Approximate analytical solution for satellite orbits subjected to small thrust or drag.AIAA J., 5,1121-1128.[233] Bromberg, E. (1956). Nonlinear bending of a circular plate under normal pressure. Comm Pure A p p l Mkrh.. 9,633659.[ 144,1461 Brown, W. P., Jr. (1966). Validity of the Rytov approximation in optical propagation calculations.J. Opt. Soc. A n , 56,1045-1052.[373] Brown, W.P., Jr.(1967). Validity of the Rytov approximation. J. Opt. Soc. A m 57,15391543. [ 373J

390

REFERENCES AND AUTHOR INDEX

B d M. A., and A. I. S o h , (1966). A new perturbation techniquefor differential equations with ~ p p t p m c t e r s m . t.Appf. M t h . , 24,143-151. [383] B u W R. N., and J. B. Keller, (1960). Boundary layer problems in diffraction t h m u . Comm Rue A& Math., 13,85-144. I3781 Budiansky, B. See Amazigo, Budianslry, and C d e r . Bumside, R. R. (1970). A note on Lighthill’smethod of strained coordinates. SIAMJ. Appf. Math. 18,318-321. [79,107] Buslaev, N. S. (1964). On formulas of short wave asymptotics in the problem of diffraction by convex bodies. Ak d. N d S S R M y Mat. Inst. Steklov, 73,14-117. [ 3801 Butler, D. S., and R. J. Gribben (1968). Relativistic formulation for non-linear waves in a nowniform plasma. J PhslM P h p . 2,257-281.1234) Caldirola, P., 0. De Barbier&and C. Maroli, (1966). Electromagnetic wave propagation in a d y ionized plasma. I. Nuovo cimento, B42,266-289.[2361 Campbell, J. A., and W. H. Jefferys, (1970). Equivalence of the perturbation theories of Hori and Deprit. Cele#&l Mech. 2,467473. [202] Carlini, F. (1817). Ricerche Julia convergenza della serie che serve alla soluzione del problem di Keplero. Memoria di Francesco Carlini, Milano; also kcobi’s Ges Werke, 30,197-254. I3151 7,189-245;alSo A m Na&i&tm, Carrier, G. F. (1953). Boundary layer problems in applied mechanics. Advan Appl. Mech., 3,1-19. [119] Carrier, G. F. (1954). Boundary layer problems in applied mathematics. C o r n R u e Appl. Math., 7,ll-17 (1141 Carrier, G . F. (1966). Gravity waves on water of variable depth. J. Fluid Mech., 24, 641-659. [234] Carrier, G. F. (1970). Singular pertubation theory and geophysics. SUMRev.. 12,175-193. r1101 Carrier, G. F. See Amazigo, Budiansky, and Carrier. Cashwell, E. D. (1951). The asymptotic solutions of an ordinsuy differential equation in which the coefficient of the parameter is sbguk.Pa&J. Moth.. 1,337-353. [3581 Caughey, T. K. See Tso and Caughey. Caughey, T.K., and H. J. Payne, (1967). On the response of a class of self-excited oscillators to stochastic excitation. Int J. Non-Lhear Mech.. 2,125-151.1235 J Cesari, L. (1971). Avmptotic Be?wvior and Stabiliv Roblems in Ordinmy Differential E q w t h s Springer-Vmlag,New York. [309] Chang, K. S., R. G. Akins, and S. G. Bankoff, (1966). Free convection of a liquid metal from a uniformly heated vertical plate. Ind. Ens. Chem Fundurn,5,26-37.1791 Chen, C. S. (1971). Parametric excitation in an inhomogeneous plasma. J. Pbsma Phys., 5, 107-113. [235] Chen, C. S., and G. J. Lewak, (1970). Parametric excitation of a plasma. J. pbsntrr Phys, 4, 357-369. [235] Cheng, H.and T. T. Wu (1970). An aging spring. Stud. Appf.Math., 49,183-185. [232] Cheng, H. K., J. W. Kirsch, and R. S. Lee, (1971). On the reattachment of a shock layer produced by an instantaneous energy release. J. Fluid Me&. 48, 241-263, [235] cheng, s. 1. See Goldburg and cheng. Chernov, L. A. (1960). Wave Proprgcrtion in a Random Medium Dover, New York. [361, 3731

REFERENCES AND AUTHOR INDEX

39 1

Cherry, T. M. (1927). On the transformation of Hamiltonian systems of linear differential equations with constant or periodic coefficients. Roc. Lond. Muth. SOC., 26, 211-230. [199] Cherry, T. M. (1949). Uniform asymptotic expansions. J. London Math. SOC.,24, 121-130. 346 1 Cherry, T. M. (1950). Uniform asymptotic formulae for functions with transition points. Trans. Am. Math. Soc., 68,224-257. [346] Chong, T. H., and L. Sirovich, (1971). Non-linear effects in steady supersonic dissipative gas dynamics. Part 1-Two-dimensional flow. J. FluidMech., 50,161-176. [235] Chu, B. T. (1963). Analysis of a self-sustained thermally driven nonlinear vibration. Phys. Fluids, 6 1638-1644. [89] Chu, B. T., and S. J. Ying, (1963). Thermally driven nonlinear oscillations in a pipe with travelling shock waves. Phys Fluid 6,1625-1637. [89] Chu, V. H., and C. C. Mei (1970). On slowly-varying Stokes waves. J. Fluid Mech., 41, 873-887. [234] Chudov, L. A. (1966). Some shortcomings of classical boundary layer theory. NASA ( t w s l . ) ‘IT F-360, IT 65-501 38. [ 1451 Clare, T. A. (1971). Resonance instability for f m e d configurations having nonlinear aerodynamic properties. J. Spucecraft Rockets, 8,278-283. [ 3051 Clark, R. A. (1958). Asymptotic solutions of toroidal shell problems. Quurt. Appl. Math., 16,47-60. [ 353 J Clark, R. A. (1963). Asymptotic solutions of a nonhomogeneous differential equation with a turning point. Arch. Rut. Mech. Anal., 12,34-51. [353,356] Clark, R. A. (1964). Asymptotic solutions of elastic shell problems. inAsymptotic Solutions of Differentid Equations and Their Applimtions (C. H. Wilcox, Ed.), Wiley, New York, pp. 185-209. [353] Coakley, T. J. (1968). Dynamic stability of symmetric spinning missiles. J. Spucecraft Rockets, 5 , 1231-1232. [321] Cochran, J. A. (1962). Problems in singular perturbation theory. Ph. D. Thesis, Stanford Universiy. [232, 233, 234, 280, 2851 Coddington, E. A., and N. Levinson, (1955). Theory of Ordinary Differentid Equations. McCraw-Hill, New York. [60,62] Coffey, T . P., Jr. (1969). Invariants to all orders in classical perturbation theory. J. Math. Phys., 10,426438. (1991 Cole, J. D. (1968). Perturbution Methods in Applied Mathematics. Blaisdell, Waltham, Mass. Cole, J. D. See Lagerstrom and Cole. Cole, J. D., and J. Kevorkian, (1963). Uniformly valid asymptotic approximations for certain nonlinear differentjal equations. Nonlinear Differentid Equations und Nonlinear Mechanics (J. P. LaSalle and S. Lefschetz, Eds.), Academic, New York, pp. 113-120. [231,232,273] Comstock, C. (1968). On Lighthill’s method of strained coordinates. SZAMJ. Appl. Muth., 16,596602. [ 79,1071 Comstock, C. (1971). Singular perturbations of elliptic equations. I. SZAM J. Appl. Muth., 20,491-502. [ 2341 Contopoulos, G . (1963). On the existence of a third integral of motion. Asrron. J.. 68, 1-14. (1991

392 REFERENCES AND AUTHOR INDEX Crane, L. J. (1959). A note on Stewartson’s paper “On asymptotic expansions in the theory of boundary layers.”J. m t h . undfiys.. 38,172-174. (791 Clapper, G. D. (1970). Nonlinear capillary waves generated by steep gtaVity waves. J. nuid Me&, 40,149.159. [217] Crawford, F. W. See Galloway and Crawford. Crocco, L. See Skignano and C r o c c ~ . Das, K. P. (1971). Interaction among three extraordinary waves in a hot electron plasma. Phys. Fluids, 14,124-128. (2341 Davidson, R. C. (1967). The evolution of wave correlationsin uniformly turbulent, weakly nonlinear systems. J. Phmm Phys.. 1,341-359. I2341 Davidsbn, R. C. (1968). Nonlinear osdllationsin a Vlasov-Maxwellplasma. Phys. Fluids, 11, 194-204. (2341 Davidson, R. C. (1969). General weak turbulence theory of resonant fourwave proce8~e8. P h y F ~ l ~ i d 12,149-161. ~, I2341 Davis, R. T. See Atzheimer and Davis;Wink~teand Davis. Davis, R. T., and K. T. Alfriend, (1967). Solutions to van der Pol’s equation using a perturbation method. Inr. J. Non-LinecuMech., 2,153-162. (2321 Davison, L. (1968). Perturbation theory of nonlinear elastic wave propagation. Inr. J. solids S m t u r e s , 4,301-322. [90,91,941 De B a r W , 0.See Cddirola, De Barbieri, and Maroli De Barbieri, O., and C. Maroli, (1967). On the dynamics of weakly ionized gases. Ann. Phya, 42,315-333. (2361 De Bruijn, N. G. (1958). Aqynrptotc N t h o d s in A w @ & North-Holland, Amsterdam; Interscience, New York. ( 181 Deprit, A. (1969). Canonical transformations depending on a small parameter. Celestkrl Mech., 1,12-30. ( 199,201,202,205,212] D e w , R. L. (1970). Interaction between hydromagnetic waves and a timedependent, inhomogeneous medium. pnVr FtWk.. 13,2710-2720. (217 J DeWolf, D. A. (1965). Wave propagation through quasi-optical irregularities. J. o p t . SOC. Am, 55,812-817. [373] DeWolf, D. A. (1967). Validity of Rytov’s approximation. J. Opt. Soc. A m . 57,1057-1058. (3731 DiP~ima,R. C. (1%9). Higher order approximations in the asymptotic solution of the Reynolds equation for slider bearings at high bearing numbers. J. LuhTication Techndogy. 91,45-51.[125] Ditac, P. A. M. (1926). On the theory of quantum mechanics. Roc. Roy. Soc. (London), A112.661677. ( 1611 Dobrowolny, M., and A. Rogister, (1971). Non-linear theory of hydromagnetic waves in a high p plasma. J. Psomrrr Phys., 6,401412. (2351 Dorodnicyn, A. A. (1947). Asymptotic solution of van der Pol’s equation. J Appl. &th. Me&, 11,313-328;Am m r h . Soc. Zhnd,88,1953. (1141 Dougherty J. P. (1970). Lagmngm . methods in plasma dynamics. I. General theory of the method of the aversged Lagmngm.J. PbmraPhy%.4,761-785. (2171 Drazin, P. G. (1969). Nonlinear internal gravity waves in a slightly stratitied atmosphere. J. FtuiiiMedr, 36,433446. (2161

REFERENCES AND AUTHOR INDEX

393

Dyson, F. I. (1949). The radiation theories of Tomonaga, Schwinger, and Feynman. P h y ~ Rev., 75,486-502. (3711 Eckhaus, W. (1965). Studies in Nonlinem Stability Theory. Sphger-Vuhg, New York. 11621 Eckstein, M. C. See Shi and Eckstein. Eckstein, M. C. and Y. Y. Shi (1967). Low-thrust elliptic sphl najecta*tsof a arrtenite of wriable ~ S S AIAA . J., 5,1491-1494.12331 Eckstein, M. C., Y. Y. Shi, and I. Kcvmkian, (1966a). A uniformly valid asymptotic representation of satellite motion around the smaller primary in the lestricted three-body problem. In Rmgress in Asbonmrricr a d Aavnautics. Vol. 17, Methods in Astraiynamfcs and Celcsthl Meelvrnicr (R.L. Lhmcombe and V. G. Szabehely, Eds.), Academic, New York, pp. 183-198. (233.276) Eckstein, M. C., Y. Y. Shi, and J. Kevorkian, (1966b). Satellite motion for arbitrary eccentricity and inclination around the smaller primary in the restricted three-body problem. Amon J.. 71,248-263. [233] Eckstein, M. C.,Y. Y. Shi, and J. Kevorkian (1966~).Use of the energy integral to evaluate higher-order terms in the time history of sateIUte motion. Ashon. J., 71, 301-305. 12331 Einaudi, F. (1969). Singular perturbation analysis of acoustiqpavity waves. mys. -8, 12,752-756. [78] Einaudi, F. (1970). Shock formation in acoustic gravity waves. J. Geophyr Re&, 75, 193-200. [78] Emanuel, G. (1966). Unsteady, diffusing, reacting tubular flow with application to the flow in a glow discharge tube. Aerospace Corporation Rept. No. TR669 (6240-201-9. (991 Emery, V. J. (1970). Eikonal approximation for nodinear equations. J. Math. Phyr, 11, 1893-1900. [234] Erddyi, A. (1956).Asymptotic Expansion& D o m , New York (18,309,3371 Erddyi, A. (1960). Asymptotic solutions of differential equations with transition points or singularities. J. W t h . M y s , 1,16-26.13581 Erdclyi, A. (1961). An expansion procedure for singular perturbations. Atti A d ScL Torim, Cl. Sci Fk Mat. Nat., 95,651672. (113,1211 Emst, R. R., and H. Primas (1963). Nuclear magnetic resonance with stochastic high-frequency fmlds.Helv. Phyr Ackr., 36,583600. ( 382) Espedal, M. S. (1971). The effects of ion-ion collision on an ionacoustic plasma pulse. J. P h s m Phyr, 5,343-355. ( 781 Euler, L. (1754). Novi CornmenWi'iAmd. Scf. Pe?Popolitanae, 5, 205-237; Opcrcr Om?&, Ser. I. 14,585617. I l l ] Evensen, D. A. (1968). Nonlinarr vibrations of beams with various boundary conditions. AL4A J., 6,370-372. ( 1061 Evgzafov, M. A., and M. B. Fedoryuk, (1966). Asymptotic behavior of solutions of the equation w"(z) - p(z'. A)w(z) = 0 as A + in the complex a-plane. U@.&t.Nauk, 21,3-50. [ 3451 Faa De Bruno, F. (1857). Note sur une nouvek formule de c a l d differentiel. Quart. J. Pure Appl. Math., 1,359-360. [ 1921

-

394 REFERJ3NCES AND AUTHOR INDEX Fedory~k,M. B. See Evgrafov and Fedoryuk. Felsen, L.B. See Levey and Felsen. Feshcheniro, S. F., N. I. S W , and L. D. N h h k o , (1967). Asymptoric Methods in the Tireoty of Linear Dtffacnthr Equationx. American Elseer, New York. [309, 319) Feynman, R. P. (1948). Space-time approach to nomlativistic qUantilm mechanics. Rev. bhdmplr)rr, 20,367-387. [%I Fife, P. Sea Pierson and Fife. F&, V. A. (1965). Electronwgneetic Diffnrctn and Ropagution Roblems. P e w o n , New York. [3801 Fowkes, N. D. (1968). A singulpr perturbation method. Parts I and 11. Quart. AppL Math, 26,5749,71435. [232,234,379,285] Fowler, R. H., and C. N. H. Lock, (1921). Approximate solutions of linw differential equations. Roc. London mth. Soc., 20,127-147. [321,322] Fowler, R. H., E. C.Callop, C. N. H.Lock,and H. W. Richmond (1920). The aerodynamics Roy. Soc. London, 221,295-387. [ 321,3221 of a 8pinning ShelL tlha. Fox, P. A. (1953). On the use of coordinate perturbations in the solution of physical problems. Project DIC 6915 Technical Rept. No. 1, Massachusetts Institute of T ~ hn o l o g yCambridge. , [98] Fox, P. A. (1955). Perturbation theory of wave propagation based on the method of characteristics.J. Math. andPhyr, 34,133-151. [57,87,89] Fraenkel, L. E. (1969). On the method of matched asymptotic expansions, Parts 1-111. Roc Chrnbr&&e€?hilSot, 65,209-231,233-251,263-284. [ 1191 Freeman, N. C., and R. S. Johnson, (1970). Shallow water waves on shear flows. J. Fluid Me&, 42,401409. [2341 Fried, D. L. (1967). Test of the Rytov approximation. J. Opt. Soc. Am., 57, 268-269. [3731 Friedrichs, K. 0. (1942). Theory of viscous fluids. Fluid Dynamics. Brown University Press. Providence, R. I., Chapter 4. [1191 Friedrichs, K. 0. (1955). Asymptotic phenomena in mathematical physics. BulL A m Math SOC.61,485-504. [18] Frieman, E. A. (1963). On a new method in the theory of irreversible processes. J. Math. Phys., 4,410-418. [230] Frieman, E. A., and P. Rutherford, (1964). Kinetic theory of a weakly unstable plssma. A m my&, 28,134-177. (2351 Frisch, U. (1965). Wuue Promation in Random Medm. Institut d'Astrophysique, Paris. r367.3691 Frisch, U. (1968). Wave propagation in random media. h b a b i l i v Methods in Applied Mathem& Vol. 1 (A. T. Bharucha-Reid, Ed.), Academic, New York, pp. 75-198. (309,361,364,365,367,372,380,381]

Frobenius, G. (1875). Veber die mguGren Integrale der linearen Differentialgleichungen.J. Reine Angew. Math., 80,317-333. [310] Furutsu, K. (1963). On the statistical theory of electromagaetic waves in a fluctuating medium. J. Re&NatL Bur. S t a M s , Sec. D,67,303-323. [ 367,3691 Gallop, E. G. See Fowler, Qhp, Lock, and Richmond. Galloway, J. J., and F. W. Crawford, (1970). Lagrangian derivation of wave-wave coupling coefficients. Proc. 4th European Conference on Controlled Fusion and Plasma Physics, Rome (CHEN), 161. [217]

’

REFERENCES AND AUTHOR INDEX

395

Galloway, J. I. and H. Kim, (1971). Lagrangian approach to non-linear wave interactions in a warm plasma. J. Phsma Phys., 6 53-72.[217] Gans, R. (1915). Propagation of light through an inhomogeneous media. Ann. Phys., 47,

709-736.[ 114,3391

Garrett, C . J. R. (1968). On the interaction between internal Bavity waves and a shear flow. J. FluidMech. 34, 711-720.I2161 Garrett, C. J. R. See Bretherton and Garrett. Germain, P. (1967). Recent evolution in problems and methods in aerodynamics. J. Roy. Aeron. Soc., 71,673491.[110,235] Gertsenshtein, M. E. See Tatarski and Gertsenshtein. Giacaglia, G. E. 0.(1964).Notes on von Zeipel’s method. NASA X-54764-161.[ 1911 Golay, M. J. E. (1964). Normalized equations of the regenerative oscillator-noise, phase-locking, and pulling. ROC. IEEE, 52,1311-1330.[258] Goldberg, P., and G. Sandri, (1967). Power series of kinetic theory. Parts I and 11. Phys. Rev., L154,188-209.[236] Goldburg, A., and S. I. Cheng, (1961). The anomaly in the application of Poincak-Lighthill-Kuo and parabolic coordinates to the trailing edge boundary layer. J. Math.

Mech., 10,529-535.[79]

Goldstein H. (1965).C&ssiml Mechunics Addison-Wesley, Reading, Mass. [ 1811 Goldstein. S. (1929).The steady flow of viscous fluid past a f j e d spherical obstacle at small [ 1411 Reynolds numbers. Roc. Roy. Soc. (London). A-123,225-235. Goldstein, S. (1931). A note on certain approximate solutions of linear differential equations of the second order. Roc. London Math. SOC.,33,246-252.(3461 Goldstein, S. (1969). Applications of singular perturbations and boundary-layer theory to simple ordinary differential equations. In Conmktions to Mechunics (D. Abir, Ed.), Pergamon, Oxford, pp. 41-67. [317] Good, R. H.See Miller and Good. Goodrich, R. F., and N. D. Kazarinoff, (1963). Diffraction by thin elliptic cylinders. Michigan Mark. J., 10,105-127.[346] GoreWr, G., and A. Witt, (1933). Swing of an elastic pendulum as an example of two parametically bound linear vibration systems. J. Tech. P k p (USSR,) 3, 244-307. ~351 Green, G. (1837). On the motion of waves in a variable canal of small depth and width. Trens, Cambridge Phil. Soc., 6,457462. (3141 Green, G. S., and A. K. Weaver, (1961). The estimation of the three-dimensional gyrations of a ballistic missile descending through the atmosphere. Royal Airuaft Establishment Tech. Nore G . W. 596 (London). I3211 Gretler, W. (1968). Em indirekte Methode zur Berechnung der ebenen Unterschallstromung. 1.Miamque, 7,83-96.[ 89) Gribben, R. J. See Butler and Gribben. Grimshaw, R. (1966). High-frequency scattering by finte convex regions. Comm Rue A p p l Math., 19, 167-198.(3801 Grlmshaw, R. (1970). The solitary wave in water of variable depth. J. Fluid Mech. 42,

639-656.(2161

Guiraud, J. P. (1965). Acoustique geomktrique, bruit balistique des avions supersoniqueset focalisation. J. Mebanique, 4,215-267.(891

396

REFERENCES AND AUTHOR INDEX

Gundersen, R. M. (1967). Self-sustained thermally driven nonlinear oscations in magnctohydrodynamic flow. Znt. J. Eng. Sci,5,205-211.[89] Gly&n, H. (1893). N o u v e k rscerches SUT les &s empby& dans les thiories des p l a n h . Actrr Mirth. 9,l-168.(561 b, L. S. (1965). On the free vibration of a beam on a nonlinear elastic foundation. J. AppL Mech., 32,445447. (58,1051 Hanks, T . C. (1971). Model relating heat-flow value near, and vertical velocities of mass transport beneath, ocean rises. J. Ceophys Reg. 76,537-544.[ 1561 Hasegawa, H. S. See Savage and Hasegawa. Hassan, S. D. See Nayfeh and Hassan. Heidbredcr, G. R. (1967). Multiple scattering and the method of Rytov. J. Opt. Soc. A n , 57,1477-1479.13731 Henrard, J. (1970). On a perturbation theory using Lie transforms. Cekstk?l Mech., 3, 107-120.[202] Heynatz, J. T.See Zierep and Heynatz. Hirschfelder, J. 0.(1969). Formal Rayleigh-SchrikIhger perturbation theory for both desnerate and nondegenerate energy states. Znt. J. Quantum C?&?m.3,731-748. t711 Holstein, V. H. (1950). Uber die aussere and h e r e Reibungsschicht bei Storungen hminarer Stromungen. ZAMM, 30,2549. [3531 Holt, M. (1967). The collapse of an imploding spherical cavity. Rev. Roumaine, S d Tech. Ser. Mec AppL. 12,407415. [78] Holt, M.,and N. J. Schwartz, (1963).Cavitation bubble collapse in water with finite density behind the interface. Phya Fluidr. 6,521-525.[78] Hoogstraten, H. W. (1967). Uniformly valid approximations in two-dimensional subsonic thia aitfoil theory. L Eng. Muth., 1,5145. [99] Hoogstraten, H. W. (1968). Dispersion of non-linear shallow water waves. J. Eng. Muth., 2, 249-273.(2341 Hori, G. I. (1966). Theory of general perturbations with unspecified canonical variables. PUbL Astron. Soc Japan. 18,287-296,[199,201,202,212] Hori, G. I. (1967). Nonlinear coupling of two harmonic oscillations. PubL Astron. Soc. Japn, 19,229-241.[199,201,212] Hori, G. I. (1970). Comparison of two perturbation theories based on canonical transformations. PubL Astron. Soc. Japan, 22,191-198.[ 2021 Horn, J. (1899). Ueber eiw h e a m DifYerentialgleichung zweiter ordnung mit einezn wiuhulichen Pa?ameter.Mzth. Ann. 52,271-292.(3161 Horn, J. (1903). Untersuchung der Intergrale einer linearen Differentialgleichung in der Umgebung ciner Unbestimmtheitsstelle vexmittelst succeSSiver Annahemngen. Arch. Mirth. Phy~, 4,213-230.[312] Hoult, D. P. (1968). Euler-Lagrangerelationship for random dispersive waves. Phys.Fluid%, 11,2082-2086.[ 2343 Houston, A. See Barakat and Houston. Howe, M. S. (1967). Nonlinear theory of openchannel steady flow past a solid surface of f~te-wave-groupshape. J. FkktMech. 30,497-512.[216] Howland, L. See Van Hove, Hugenholtz, and Howland.

REFERENCES AND AUTHOR INDEX

397

Hsieh, P. F., and Y. Sibuya, (1966). On the asymptotic integration of second order linear ordinary differential equations with polynomial coefficients. J. Mkth. Anal AppL, 16,84-103. (3451 Hufnagel, R. E., and N.R. Stanley (1964). Modulation transfer function associated with image transmission through turbulent media. J. Opt. Soc. A m . 54,5261. [373] Hugenholtz, N. See Van Hove, Hugenholtz, and Howland. Imai, I. (1948). On a refmement of the W. K. B. method.Phys. Rev.. 74, 113. [ 324, 3501 Imai, I. (1950). Asymptotic solutions of ordinary linear differential equations of the second order.Phys Rev.. 80, 1112. [350] Ince, E. L. (1926). Ordimry DifferentialEquations Longmans, Green, London. [ 5 , 3141 Jacobi, C. G. J. (1849). Versuch einer Berechnung der grossen Ungleichheit des Saturns nach einer strengen Entwickelung. Astron. Nu&. 28,65-94. [ 312) Jacobs, S. J. (1967). An asymptotic solution of the tidal equations. J. Fluid Mech., 30, 417438. [234] Jahsman, W. E. (1968). Collapse of a gas-fiUed spherical cavity. J. Appl. Mech., 95,579-587. [781 Jakob, M. (1949). H a t Tnma;er, Vol. 1. Wiley, New York. I3421 Jefferys, W. H. See Campbell and Jefferys. Jeffreys, H. (1924a). On certain approximate solutions of linear differential equations of the second order. Roc. London Muth. Soc., 23,428436. [ 114,3391 Jeffreys, H. (1924b). On certain solutions of Mathieu’s equation. Roc. LondonMufh. Soc. 23,437476. [114,339] Jeffreys, H. (1962). Asymptotic Approximations. Oxford University Press, Oxford. [ 309, 342,3821 Johansen, K. F., and T. R. h e , (1969). A simple description of the motion of a spherical pendulum. J. Appl. Mech., 36,408-41 1. (2241 Johnson, R. S. See Freeman and Johnson. Kain, M.E. See Kane and Kahn. Kamel, A. A., (1969a). Expansion formulae in canonical transformations depending on a small parameter. Celesziul Me&, 1,190-199. [199,202,206,212] Kamel, A. A. (1969b). Perturbation theory based on Lie transforms and its application to the stability of motion near sun-perturbed earth-moon triangular libration points. SUDAAR Rept. No. 391, Stanford University. [ 199,202,2061 Kamel, A. A. (1970). Perturbation method in the theory of nonlinear oscillations. Celestial Mech., 3,90-106. [ 199,201,202,206) Kamel, A. A. (1971). Lie transforms and the Hamiltonization of non-Hamiltonian systems. CelestialMech. ,4,397405. [ 2021 Kamel, A. A. See Nayfeh and Kamel. h e , T.R.See Johansen and Kane. Kane, T. R., and M. E. Kahn, (1968). On a class of two-degree-of-freedom oscillations. J. Appl. Mech., 35,547-552. [ 1851 Kaplun, S. (1954). The role of coordinate systems in boundary-layer theory. Z. Angew. Math. phys, 5, 111-135; also Chapter 1 of Kaplun (1967). [SO, 1141 Kaplun, S . (1957). Low Reynolds number flow past a circular cylinder. J. Mkth. Mech., 6 , 595603; also Chapter 3 of Kaplun (1967). [ 1141

398

REFERENCES AND AUTHOR INDEX

Kaplun, S. (1967). FIuid Medronics and Singuhr Perturbations (a collection of papers by S. Kaplun, edited by P. A. Lagerstrom, L. N. Howard, and C. S. Liu), Academic, New York. [114,1191 Kaplun, S., and P. A. Lagerstrom, (1957). Asymptotic expansions of Navier-Stokes solutions for small Reynolds numbem. J. Murh. Me& 6, 585-593; also Chapter 2 of Kaplun

(1967). [114] Karal, F. C., and J. B. Keller, (1964). Elastic, electromagnetic, and other waves in a random medium. J. mrh.Rhys. 5,537-547. [369] Karpman,V. I. See Al'tshul' and Karpman. Karpman, V. I., and E. M. Km&kal' (1969). Modulated waves in nonlinear dispersive media. soviet Phys. 28,277-281. [216] Kawakami, I. (1970). Perturbation approach to nonlinear vlasov equation. J. Phys. SOC. Japn. 28,505-514. [216] Kawakami, I., and T.Yagishita, (1971). Pertubation approach to nonlinear Vlasov equation. 11. Nonlinear plasma oscillation of fmite amplitude. J. P?iys Soc. Japan. 30,244-253. [2161 Kazarinoff, N. D., (1958). Asymptotic theory of second order differential equations with two simple turning points.Arch. Rat. Me&. Anal. 2,129-150. [ 3441 Kazarinoff, N. D. See Goodrichand -off. Kazarinoff, N. D., and R. W. Mc Kelvey, (1956). Asymptotic solution of differential equations in a domain containing a regular singular point. Con J. Math., 8, 97-104. [3581 Keller, J. B. (1958). A geometrical theory of diffraction, calculus of variations and its applications. Aoc. Symp. AppL Mkth.. 8,27-52. [ 3741 Keller, J. B. (1962). Wave propagation in random media. Proc. Symp. Appl. Marh., 13, 227-246. [ 369,3821 Keller, J. B. (1968). Perturbation Theory. Lecture notes, Mathematics Department, Michigan State University.[58,150] Keller, J. B. See Avila and Keller, Buchal and Keller; Karal and Keller; Lynn and Keller; Millman and Keller. Keller, J. B., and S. Kogelman, (1970). Asymptotic solutions of initial value problems for nonlinea partial differential equations. SIAMJ. AppL Marh., 18,748-758. [234] Keller, J. B., and M.H.MiUmau, (1969). Perturbation theory of nonlinear electromagnetic wave propagation.P?iys Rev.. 181,1730-1747.1771 Keller, J. B., and L. Ting, (1966). Periodic vibrations of systems governed by nonlinear partial differential equations. Comm Rue. AppL Murh. 19,371420. [77] Kelly, R. E. (1965). Stability of a panel in incompressible, unsteady flow. AZAA J., 3, 1113-1118. [233] Kelly, R. E. (1967). On the stability of an inviscid shear layer which is periodic in space and time. J. Fhrid Mech., 27,657689. [2351 Kelly, R. E. See Maslowe and Kelly. Kemble, E. C. (1935). A contribution to the theory of the B.W.K. method. Phys, Rev.. 48, 549-561. [339] Kersten, P. H. M. (1967). Diffraction of an electromagnetic wave by a plane screen. Ph. D. Thesis. Technbche Hochschull, Eindhoven. [380]

rnN%.ENCES AND AUTHOR INDEX

399

Kevorkian, J. (1966a). The two variable.expansion procedure for the approximate solution of certain nonlineax differential equations. Spurn Marhenrrrt&x Part 3. (J.B. Rosser, Ed.) American Mathematical Society, Providence, R.I.,pp. 206-275. (231,232,233, 273) Kevorkih, J. (1966b). von Zeipel method and the twwariable expansion procedure. Adtron J., 71, 878-885. [231] Kevorkian, J. (1971). Passage through resonance for a omxbmmional oscillstox with alowly varying frequency. SCQMJ.AppL hizth.. 20,364-373. [233,383] Kevorkian, J. See Cole and Kevorkian; Eckstein, Shi, and Kevorkian; Lagcrstrom and Kevorkian. Kiang, R. L. (1969). Nonlinear theory of inviscid Taylor instability near the cutoff wave number. Phys. Fluids, 12,1333-1339. [235] Kim, H. See Galloway and Kim. Kirchhoff, G. (1877). Zur Theorie des Cbndenazttas, Berlin, Akad., Monatsber. pp. 144-162. [114] Kirsch, J. W.See Cheng. Kirsch, and La. Klein, J. S. See Sellers, Tribus, and Klein. Klimas, A., R. V., Ramnath, and G.Sandri, (1970). On the compatibility problem for the uniformization of asymptotic expansions. 1. Math. AnaL Appl., 32,482-504. 12321 Kogelman, S. See Keller and Kogelman. Kraichnan, R. H. (1961). Dynamics of nonlinear stochastic systems. J. Mzth. Phys, 2, 124-148. [ 367) Kramers, €I A. . (1926). WeUenmechanik und ha&* Quantisierung. 2. Phys., 39, 828-840.[114,315,339) Kravtsov, Y. A. (1964a). A modification of the geometrical optics method. Rudbjizika, 7, 664-673 (in Russian). [378] Kravtsov, Y. A. (1964b). Asymptotic solutions of Maxwell's equations near a caustic. Rudiofuika, 7,1049-1056 (in Russian>. [378,380] Kravtsov, Y. A. (1965). Modification of the method of geometrical optics for a wave penetrating a caustic. &&&ah, 8,,659667. I3801 Kravtsova, T. S. See Babich and Kravtsova. Krushkal', E. M. See Karpman and Kndkal'. Kruskal, M. (1962). Asymptotic theory of Hamiltoninn and OW systems with all solutions nearly periodic. J. Math. My&, 3,806-828. /I681 W l w , N., and N. N. Bogoliubov (1947). Introduction to Nonlinear Mechanic& Princeton University Press, Princeton, N. J. [165,174] Kubo, R. (1963). Stochastic Liouville equation. J. Mufh. Phy&,4,174-183. [ 3821 Kuiken, H. K. (1970). Inviscid film flow over an inclined surface originated by strong fluid injection. J. Fluid Me&, 42,337-347. [99] Kuo, Y. H. (1953). On the flow of an incompmssible viscous fluid past a flat plate at moderate Reynolds numbers. J. Math and My%,3 5 83-101. [ 78) Kuo, Y. H. (1956). Viscous flow along a flat plate moving at high supersonic speeds. J. A m n S d , 23,125-136. I781 Kuzmak, G. E. (1959). Asymptotic solutions of nonlimar ascond order differential equations with variable coefficients.J. AppL Math. Me&, 23, 730.744. [232,2871

400 REFERF,NCESAND AUTHOR INDEX Lcina, J. (19693. New CBllOnicBL perturbation method for complete set of integrals of

motion. Cu&. J. tlhyr. B19,130-133. (1991 Lacine,J. (1969b). New COllOniCPl perturbation method for complete set of integrals of motion. Ann. Ahys, 51,381-391. [ 1991 Lagerstrom, P. A. See Kaplun and Lagerstrom. Lqerstrom, P. A., and J. D. Cole, (1955). Examples illustrating expansion procedures for the Nav&&tokw equations. J. Rat. Meah. A n d , 4,817-882. [ 1401 Lagerstrom, P. A., and J. Kevorkian, (1963a). Earth-to-moon trqjectories in the restricted -body problem.J. M&nt4ue, 2,189-218. [138] approximation to the two Lagerstrom, P. A., and J. Kevorkian, (1963b). Mat&ed&c fixed force-cater problem.Altron. L68,84-92. (44.1381 Landahl, M. T.See Rubbert and Laadahl. Langer, R. E. (1931). On the asymptotic solutions of differential equations, with an application to the Besael functions of large complex order. Zhzm A m Murh. Soc., 33,2344. [339,340,3451 Langer, R. E. (1934). The asymptotic solutions of certain finear ordinary differential equations of the second order. Zhm A m Mfh.Soc, 36,90406. [339,340] Langer, R. E. (1935). On the asymptotic solutions of ordinary differential equations with reference to the Stokes phenomenon about a singulpi point. Pam A m Mth. Soc., 37,397416. [ 3581 Langer, R. E. (1949). The asymptotic solutions of ordinary bear differentialequations of the second order, with special reference to a turning point. Zhzm A m Mark Soc., 67,461490. [ 346,3481 Langer, R. E. (1957). On the asymptotic solutions of a class of ordinary differential equations of the fourth order, with specint reference to an equation of hydrodynamics. Ih.urs.Am. Wh. Sot. 84,144491. [360] Langer, R. E. (1959a). Formal solutions and a related equation for a class of fourth order differential equations of a hydrodynamic type. Zhzm A m Murk Soc., 92,371410. [ 3601 Langer, R. E. (1959b). The asymptotic solutions of a linear differential equation of the second order with two turning points. Zhzm A m Mfh.Soc.. 90,113-142. (3441 Laplace, P. S. (1805). On the figure of a large drop of mercury, and the depression of mercury in a glass tube of a great diameter. In CClrrHl Mechunics (transl. by Nathaniel Bowditch, Boston, 18391, Chelsea, New York,1966. [ 1141 Latta, G. E. (1951). Singular perturbation problems. Ph.D.Thesis, California Institute of Technology. [114,1451 Latta, G. E. (1964). Admced Ordimwy DiffwentizlEquutiora Lecture notes, Stanford University. I3171 LedovsEsja,L. B. See Zabreiko and LedovaLaja. Lee, D. H.,and L. hi. Sheppard, (1966). An approximate second-order wing theory. AIAA L, 4,1828-1830. [78,89] Lee, J. H. S. See Akinsete and Lee. Lee, R. S. Sa Cheng, Kirsch,and Lee. Legns, J. (1951). Application dz la Athode de Lighthill ‘a un ikoulement plan supersonique. C o w .R e d , 233,1005-1008. [78,86]

REFERENCES AND AUTHOR INDEX 401 Legras, J. (1953). Nouvelles applications de la mithode de LighthiU APdtude des ondes de choc. O.N.E.R.A. PubL No.66. [78,86] Lesser, M. B. (1970). Uniformly valid perturbation series for wave propagation in an inhomogeneous medium. J. Aooust. Soc. A m . 47,1297-1302. [W] Le Verrier, U. J. J. (1856). Sur la determination des longitudes terrestres. Paris, C0m;pt. Rend., 43,249-257. [ 168) Levey, H. C. (1959). The thickuess of cylindrical shocks and the PLK method. Qurut. AppL Math., 17,77-93. [52,99, 1001 Levey, L., and L. B. Fefsen, (1967). On transition functions Occurring in the theory of diffraaion in inhomogeneous media. J. Inst. M t . AppL, 3,76-97. [3801 Levinson, N. (1969). Asymptotic behavior of solutions of nonlinear differential equations. Shld. AppL Math.. 48,285-297. [21,22] Levinson, N.See Coddington and Levinson. Lewak, G. J. (1969). More uniform perturbation theory of the vlapov equation. J. Phsma PhyS., 3,243-253. [78] Lewak, G. J. (1971). Interaction of electrostatic waves in collisionless plasmas. J. Phsma Phys, 5,5143. I2351 Lewak, G. J. See Chen and Lewak;Zawadzki and Lewak. Lewis, R.M. See Ahluwalia, Lewis, and Boersma. Lewis, R. M.,N. Bleistein. and D. Ludwig, (1967). Uniform asymptotic theory of creeping waves. Comm €We Appl. Math., 20,295-328. [380] Lick, W. (1969). Two-variable expansions and singular perhubation problems. SZAMJ. AppL Mth., 17,815-825. [89] Lick, W. (1970). Nonlinear wave propagation in fluids. Annual Review of Fluid Medmnics, Vol. 2 (M.van Dyke, W. G. Vinciuti, and J. V. Wehausen, Eds.), Annual Reviews, Palo Alto, Calif., pp. 113-136. [235] Liepins, A. A. See Sanders and Liepins. Lighthill, M. J. (1949a). A technique for rendering approximate solutions to physical problems uniformly valid. Phil Mag.,40,1179-1201. [42,57,77,80,87,108] Lighthill,M. J. (1949b). The shock strength in supersonic “conical fields.” I412Mag., 40, 1202-1223. [78] Lighthill, M. J. (1951). A new approach to thin airfoil theory. A e o n Quart., 3, 193-210. [981 Lighthill, M. J. (1961). A technique for rendering approximate solutions to physical problems uniformly valid. Z. F7ugWisk. 9,267-275. [57,77,991 Lighthill, M. J. (1965). Contributions to the theory of waves in non-linear dispersive systems. J: Inst. Math. AppL. 1,269-306. [216] Wthill, M. J. (1967). Some special cases treated by the Whitham theory. h c . Roy. Soc., (London). A m , 28-53. [216] Lin, C. C. (1954). On a perturbation theory based on the method of characteristics.J. Math andPhys, 33,117-134. [57,87,89] Lin, C. C. (1955). The Theory of Hydroctmamic StubW. Cambridge University Press, Cambridge. [ 3601 Lin, C. C., and A. L. Rabenstein (1960). On the asymptotic solutions of a claw of ordinary differential equations of the fourth order. Zhzm A m Mth. Soc., 94,2457. I3601

402

REFERENCES AND AUTHOR INDEX

Lin, C. C., and A. L. Rabenstein (1969). On the asymptotic theory of a class of ordinary differentid equations of the fourth order. 11. Existence of s o l u t i o ~which ~ ~ are approximated by the fcKmal solutions. snd.ApPr Mkth.,48,311-340. [360] Lindstedt, A. (1882). Ueber die Integration eiuer fur die strorungs~rkwichtigen DifferenWgbkhung. A-n, Nuah., 103, h i . 211-220. [56] Lindzen, R. S. (1971). Equatorial planetary waves in shear: Part 1. J. Atmos. S C L 28, 609622. [2341 Liouville, J. (1837). Second m6moixe sur le d&eloppment des fonctions en shies dont divers tenncs mt 'bsujettis a satisfaire 'a une &me kuationdiff&ntielle du second ordm contenant un paramktre variable. J. Mkth. h e A&, 2,16-35.[314] Lock,C. N. H. See Fowler and Lock;Fowler, Gallop, Lock,and Richmond. Lowll, S. C.(1970). Wave propagation in monatomic lattices with anharmonic potential. hx.Roy. Soc (London). A318,93-106. [217] Ludwig, D. (1966). Uniform asymptotic expausions at a caustic. &nun Rue AppL Muth., 19,215-250. [378,380] Ludwig, D. (1967). Uniform asymptotic expansion of the fwld scattered by a convex object at high frequencies. C o r n Pure AppL W M . 20,103-138. [380] Ludwig, D. (197Oa). Diffraction by a circular cavity. J. W t h . P h p . 11,1617-1630.[380) Ludwig, D. (1970b). Uniform asymptotic expansions for wave propagation and diffraction problems. SIAMRCV., 12,325-331. [380] Ludwig, D. See Lewis, Bleistein and Ludwig. Luke, J. C. (1966). A perturbation method for nonlinear dispersive wave problems. Boc. Roy. Soc. (London). A292,403412. [234,301) Lynn,R. Y. S., and J. B. Keller (1970). Uniform asymptotic solutionsof second order linear ordinary differentid equations with tumiug points. Comm Rue AppL W f h . , 23, 379408.[345] Lyusternik, L. A. See V&k and Lyusternik. Mc Goldrick, L. F. (1970). On Wilton's ripples: A special case of resonant interactions. J. FhddMech., 42,193-200. [2341 Mc Intyre, J. E. (1966). Neighboring.optimal terminal control with discontinuous forcing functions.ALQA J., 4,141-148. [79] Mc Kelvey, R. W. (1955). The solution of second order linear ordinary differentialequations about a turning point of order two. Zhzm. A m Muth. Soc., 79,103-123. (3461 Mc Kelvey, R. W. See Kazariuoff and Mc Kelvey. Mc Namara, B., and K. J. Whiteman (1967). Invariants of nearly periodic Hamiltonian systems. J. Mth. Phy~.,8,2029-2038. [ 1991 Mahony, J. J. (1962). An expansion method for singular perturbation problems. J. Aushvrlian Wf. Soc,2,440463. [232,303] Malleus, W. V. R.;and G. Veronis (1958). Finite amplitude cellular convection. J. Fluid Meek. 4,225-260. [77] Maroli, C. (1966). Kinetic theory of high-frequency resonance gas discharge breakdown. Nuovo CYmmto, B41,20&224.[236] Maroli, C. See Calttirola. De Barbieri, and Maroli; De Barbjeri and Maroli. Maroli, C., and R. Pozzoli (1969). Penetration of high-frequency electromagnetic WVH into a slightly ionized plpsma Nuow Cirnento.B61,277-289. [235]

REFERENCES AND AUTHOR INDEX 403 Maslowe, S. A., and R. E. Kelly (1970). Finite amplitude oscillations in a Kelvin-Helmholtz flow. Int. J. Non-Lineor Mech., 5,427435.(771 Matkowsky, B. J. (1966). Asymptotic solution of partial differential equations in thin domains. Ph.D.Thesis, New York University. (3801 Matkowsky, B. J., See Re& and Matkowsky. Maxwell, J. C. (1866). On the viscosity or internal friction of air and other gases. Phil. l h n s Roy. Soc. London, 156,249-268.11141 Mei, C. C. See Chu and Mei. Meirovitch, L. (1970).Methods of Amlytiarl Dymmks Maraw-Hill, New York. [ 1811 Melnik, R. E. (1965). Newtonian entropy layer in the vicinity of a conical symmetry plane. A M J., 3,520-522.[79] Mendelson, K. S. (1970). Perturbation theory for damped nonlinear oscillations. J. Muth. Phy~.,11,3413-3415.(2241 Mersman, W. A. (1970). A new algorithm for the Lie transformation. CelestiuI Mech., 3, 81-89.[202] Merman, W. A. (1971). Explicit recursive algoxithms for the construction of equivalent canonical transformations. Celestial Mech..,3,384-389.[202] Messiter, A. F. See Spriggs, Messiter, and Anderson. Mettlex, E. (1959). StabilitZtsfragen bei freien SchwingUngen mechanischer Systeme. IngenieW-Archiv., 28,213-228.[ 168,1881 Meyer, J. W. (1971). Rayleigh scattering of a kser beam from a massive relativistic two-level atom. Phys. Rev.,A3,1431-1443.(2351 Miller, S. C., and R. H. Good (1953). A WKB-type approximation to the SchrSdinger equation. Phys Rev., 91,174-179.[ 3441 Millman, M. H. See Keller and Millman. Millman, M. H., and J. B. Keller (1969). Perturbation theory of nonlinear boundary-value problems. J. Math. phys, 10,342-361.[77] Mitchell, C. E. (1971). Analysis of a combustion instability problem using the technique of multiple scales. AIAA J., 9,532-533.[235] Mitropolski, Y. A. (1965). Problems of the A&yymptotic Theory of Non-stutionury Vibmtions. Daniel Davey, New York. [ 1741 Mitropolski,Y. A, See Bogoliubov and Mitropolski. Montgomery, D., and D. A. Tidman (1964). Secular and nonsecular behavior for the cold plasma equations. Phys Fluas, 7,242-249.[ 1781 Moriguchi, H. (1959). An improvement of the WKB method in the presence of turning points and the asymptotic solutions of a class of Hill equations. J. Phys Soc. k p a n , 14,1771-1796.[342,345,350]

Morino, L. (1969). A perturbation method for treating nonlinear panel flutter problems. MAA J., 7,405411. [234] Moms, W. D. (1965). Laminar convection in a heated vertical tube rotating about a parallel axis. J. Fhtid Mech., 21,453464.(791 Morrison, H. L. See Richmond and Morrison. Morrison, I. A. (1966a). Comparison of the modifKd method of averaging and the two variable expansion procedure. SIAMRev., 8,6645. 12311

404 REFERENCES AND AUTHOR INDEX Mordson, J. A. (1966b). Gcosralized method of averand the von zeipel method. In &ogres in Amornutics and A m m t i e r . VoL 17. Mcthodr in Asbrodynrunicr ond ccluthlM .(R. L. Duacombs and V. G. Szebehdy, Eda), Acadedc, New York, pp. 117-138. [168,1911 M o w M. P. (1%8). Traveling load on a cylincticrl sbsll. J. Aaowt. Soc. Am. 44, 1664-1670. [233] Mortell, M. P. (1969). Waves on a spherical shell. J. Aoourt. Soc Am. 45,144-149. [233] Morten, M. P. (1971). Resonant therm&acmstic osciDotionn Int. J. Eng. S d , 9,175-192. I891 Mortell, M. P., and E. Varley (1971). Finite amplitude waves in boundedmedia: Noalinear free vibrations of an elastic panel. Roc. Roy. Soc ( L o d n ) . A318,169-1%. I901 Morton, B. R. (1959). Laminar convection in uniformly heated horizontal pipes at low Rayleigh numbem Quort. L Me& Ap& k t h . 12,410426. [79] MulhoJland, R. J. (1971). Nonlinear oscillations of a third-order dafsrcntial equation. Int. J. Non-Lhw Me&, 6,279-294. [ 1051 Murphy, C. H. (1963). Free flight motion of symmetric micsiles. JMlistic Reqarch Laboratorbs Rept. No. 1216, Abodeem hoviug Grounds, Md. (3211 Munay, J. D. (1968). On the effect of drahuge on free surface osdllptionr. A& Sci Re&. 19,234-249. [2341 Muss, S. A. (1967). Integral constraints in wealdy nonlinear periodic system.SIAM J. AppL k t h . , IS, 1324-1327. I2321 Musen, P. (1965). On the high order effects in the methods of Krylov-Bogoliubov and Pom&. J. AItron S d , 12,129-134. [168,192,199] Nair, S., and S. Nemat-Nmer (1971). On Wte amplitude waves in heterogeneous elastic soIids.Int. J. Eng. S d , 9,1087-1105. [94] Nayfeh, A d ~ n ,and S. Nemat-Nasser (1971). Thennoelastic waves in solids with thermal relaxation.Ado Mechunk, 12,5369. [45,136] Nayfeh, A. H. (1964). A gewralized method for treating shgular perturbation problems. Ph. D.Tbgis, Santfurd Udvadty. [232,276,280,285,295] Nayfeh, A. H. (1965a). A comparison of three perturbation methods for the earth-moonspaceship probbm. AL4A J.. 3,1682-1687. [82,103,233,295,297] Nayfeh, A. H. (1965b). An expansion method for treating sjngular perhubation problems.J. k t h . My&.6,1946-1951. [232,234,276,280,285,384] Nayfeh, A. H. (1965~).A paturbation method for treating nonlinear oscillation problems. J. kloth. ondMy&.44,368-314. [230,232] Nayfeh, A. H. (19659. Nonlinear oscillations in a hot electron plasma. Phyo Fhjdp. 8, 1896-1898. (230,2341 Nayfeh, A. H. (1%6). Taboff from a circular orbit by a anall thrust. In &ogres in Arbolurcticr and Acnmcrutia. Vot 17, Metho& in Asmdynsmia ond &leatiul Mechanics (R. L. Duncornbe and V. G. !kebehely, Eds.), Academic, New York, pp. 139-152 199.109,233,3051 Nayfeh, A- H. (1967a). hplptotk solutions of an ejgenvalue problem -0 turning -poislts--heet transfer in a tube. J. ddirtrh ond fiya, 16,349-354. (3851 Nayfeh, A. H. (1967b). The van &r Pol 0aCill.tor with delayed amplitude limiting. ptoe. IhXE, 55,111-112.[231.232,258]

REFERENCES AND AUTHOR INDEX

405

Nayfeh, A. H. (1968). Forced osdllations of the van der Pol oscillator with delayed amplitude limiting.IEEE Tranr GWdt Th-, 15,192-200. [230,232,258] Nayfeh, A. H. (1969a). A multiple time scaling aaalysis of rwntry vehicle roll dynamics. AIM J.. 7,2155-2157. (233,3201 Nayfeh, A. H. (1969b). On the nonliDear Lamb-Taylor instability. J. Fhrfd Mech., 38, 619631. [235] Nayfeh, A. H. (197Oa). Characteristic exponents for the triangular points in the elliptic restricted problem of three bodies. AlAA J.. 8,1916-1917. [68] Nayfeh, A. H.(1970b). Finite amplitude surface waves in a liquid layer. J. Flicid Mech., 40, 671684. (2341 Nayfeh, A. H. (1970~).Nonlinear stability of a liquid jet. H y a &ids, 13, 841-847. [99, 2351 Nayfeh, A. H. (197Od). Triple- and quintupledimpled wave profrles in deep water. Phya WS 13,, 545-550. [234] Nayfeh, 4. H.(1971a). Third-harmonic resonance in the interaction of capillary and gravity waves. J. FhridMech.. 48,385-395. [234] e ~ the equilateral libration points. AIAA Nayfeh, A. H. (1971b). Two-toone r e s o ~ n c near J., 9,23-27. [233] Nayfeh, A. H.,and S. D. Hassan (1971). The methd of multiple scales and nonlinear dkpmive waves. J. Fluid Mech., 48,463475. [234,298] Nayfeh, A. H., and A. A. Kamel (1970a). Stability of the triangukr points in the elliptic restricted problem of three bodies. A U A J., 8,221-223. [64,66] Nayfeh, A. H. and A. A. Kamel (1970b). Threetoone resonances near the equilateral libration points. AIAA J., 8,2245-2251. [233] Nayfeh, A. H., and W. S. S d c , (1971a). Nonlinear Kehrin-Helmholtz instability. 1. Fluid Mech., 46,209-231. [235] Nayfeh, A. H., and W. S. Saric (1971b). Nonlinear resonances in the motion of rolling re-entry bodies. AIM Paper No. 7147. [233,305] Nayfeh, A. H., and W. S. Saric (1972a). An analysis of asymmetric r o w bodies with nonlinear aerodynamics.ALAA J., 10,1004-1011. [233,291,292,306] Nayfeh, A. H.,and W. S. W c (1972b). Nonlinear waves in a Kelvin-Helmholtz flow. J. FhtidMech., 55,311-327. [234] Nemat-Nasser, S. See Adnan Nayfeh and Nemat-Nasser; Nair and Nemat-Nasser. Neubert, J. A. (1970). Asymptotic solution of the stochastic Helmholtz equation for turbulent water. J. A m s t . Soc. Am, 48,1203-1211. [334] Newell, A. C. (1968). The closure problem in a system of random gravity waves. Rev. C e o p h y ~6,l-31. , (2341 Newell, A. C. (1969). Rossby wave packet interacti0ns.J. Fluid Mech., 35,255-271. (234) Newell, A. C. See Bemey and Newell. Newell, A. C., and J. A. Whitehead (1969). Finite bandwidth, fUrite amplitude convection. J. W M e c h . , 38,279-303. (235) Nienhuis, G. (1970). On the microscopic theory of Brownian motion with a rotational degree of freedom. PhyJla. 49,2648. (2361 Nikolenko. L. D. See Feshchenko, Shkil', and Nikolenko. Nocrdlinger, P. D., and V. Petrosian (1971). The effect of cosmological expansion on self-gravitatingensembles of particles. Astrophya J.. 168,l-9. [232]

406 REPERENCES AND AUTHOR INDEX Ockendon, J. R. (1966). The separation of Newtonian shack layers. J. Fhrid Me&., 26, 563-572. [79] Olvsr, F. W. J. (1954). The asymptotic solution of linur differential equptions of the second order for large valuea of a ppramcter and t b m P b t i c a W o n of-1 w w of lslgC order. piML IZMI. Roy. See. LondDn Sa. A, 247,307-368. [341, 348,3581 Ohnr, F. W. J. (1959). Uniform asymptotic e~pansiOnafor Weber pprabolic cylinder functirms of b g a orders. J. Reg NutL Bw.stcmchrdr, 63B, 132-169. [ 3451 for cssfpin nontineareecoadorder 0’M.lley. R. E., Jr. (19680). A boundary value differential equations with a mall parameter. Arch. Rat. Me& A d , 29, 66-74. (2331 Omalley, R. E., Jr. (1968b). Topics in singular perturbations. Ad~tan.Marh., 2, 365470. [ 110,2331 O’Malley, R. E., Jr. (1971). Boundary layer methods for nonlincpr initial value problems. SUMRev., 13,425-434. [1451 Omalley, R.E., Jr. See Ackerberg and O’Malley. OSCW, C. W. (1910). &x die S t o w F d and uber eiw verwandte Aufgabe in der Hydrrdynamk Ark mr. Artron. 0 s . 6, . No. 29. [ 1401 Oswatitsch, V. K. (1965). Ausbreitungsprobleme.ZAMM. 45,485498. [89] Pandey, B. C. (1968). Study of cylindrical piston problem in water using PLK method. ZAMP, 19,962-963. [86] Parker, D. F. (1969). Nonlinearity, relaxation and diffusion in acoustics and ultrasonics.J. FhridMeali. 39,793-815. (2341 Parker, D. F., and E. Vatley (1968). The interaction of fdte amplitude deileciion and stretchiug waves m drstic memlmmes and strings. Q w f . J. Mech.AppL Wth., 21, 329-352. [go] Payne, H.J. See Caughey and Payne. Pearson, J. R. A. See Roudman and Pearson. Pcdtowsky, J.(1%7).Fluctua~winds and the ocean circulation. Z ‘ e h , 19,25@257. [77] Perko, L. M. (1969). Highsr order averaging and related methods for perturbed periodic and quasi-perbdic systems. SIAMJ. Appl Math.. 17,698-724. I2311 Perko, L.M. See Brsakwcll and Perko. Pertosian, V. See N0er-r and Petrosian. peyret, R. (1966). b e n t quasi unidimensionnel dans un ~ t e dewplasma ‘a ondes L M.m*, 5,471415. (2341 Peyret, R. (1970). Etude de 1’6coulement d’un fluide umducteur dans un canal par la mGthode dcs belles multiples.J. M k n i q u e , 9,61-97. [2351 Pierson, W. J., and P. Fife (1961). Some nonliaeer properties of long-aested periodic waves with lengthsnear 2.44 centimeters. J. Caopnva Reg. 66,163-179. [771 pilrc, E. R. (1964). On the rektedupation method of asymptotic approximation.Quwt.J. Mech.AppL Math., 17,105-124.369-379. [242,344] Pohc& H. (1892). New McthodsofCckrtiplMCchmJa.VOLI-III (English transl.), NASA TTF450,1%7. [ 10,561 Potter, M. C. See Reynolds and Potter. Pozzoli, R. See Mudi and Pozzoli.

REFERENCES AND AUTHOR INDEX

407

Pmndtl, L. (1905). h e r Flhsigkeitsbewegung bei sehr klemer Reibung. Procedugs Third Internat. Math. Koagr., Haidelberg, pp. 484491. Motion of fluids with very little viscosity. Tech.Memo N.A.C.A. (English ttansl.), No. 452,1928. [ 113) Rasad, R. (1971). Effect of ion motion on parametric oscillations of a cold plasma in a magnetic fkld. J. Phmm Plryr, 5,291-302. (2351 F5gogin6, I. (1962). NonequUilrrfurnStCrtMcal Meehonicg Wiley, New York. (367, 369, 3821 Primas, H. (1961). ijber quantenmeehdshe Systeme mit ehem stochastischen Hamiltonoperator. Helv. my&Ado. 34.36-57. [ 3821 Primas, H.See Ernst and Primas. Pringle, R. Jr. See Breakwell and Pringle. Pritulo, M. F. (1962). On the determination of uniformly accurate solutions of differential equations by the method of perturbation of coordinates. J. AppL Mbth. Mech.. 26, 661-667. [57,95,369] Proudman, I. (1960).An example of steady laminuflow at large Reynolds number. J. * i d Mech.,9,593-602. [54,156] Proudman, I., and J. R. A. Pearson (1957). Expansions at Smpu Reynolds numbers for the flow past a sphere and a ciwular cylinder. J. FfuidMech., 2, 237-262. 1114, 141, 1441 Puri, K. K. (1971). Effect of viscosity and membrane on the oscillations of superposed f l ~ & J. AppL Plry~.,42,995-1000. (2351 Rabenstein, A. L. (1959). The determiuation of the inverse matrix for a basic reference equation for the theory of hydrodynamic stability. Arch. Rot. Me& Anal, 2, 355-366. (3601 Rabensteiq, A. L. See Lin and Rabenstein. Rajappa, N. R. (1970). Nontineat theory of Taylor instability of superposed fluids. J. my& Soc. Jupon, 28,219-224. (771 Ramamthan, G. V., and C . Sandri (1969). Model for the derivation of kinetic theory. J. Murk my&.10,1763-1773. (2361 Ramnath, R. V. (19700). A new analytical approximation for the Thomas-Femi model in atomic physics. J. &rh. AML AppL, 31,285-296. (2351 Ramnath, R. V. (1970b). Transition dynamics of VTOL aircraft. AIAA J., 8, 1214-1221. (2331 Ramauth, R. V. (1971). On a class of nonlinear differential equations of astrophysics. J. Moth. AWL AppL. 35,2747. [235] Ramnath, R. V.See Klimap, Ramnath, and Sandxi. Ramnath, R. V., and G. Sandri (1969). A g e m d .m d multiple scales approach to a class of linear differentialequations.J. &th. AML AppL. 28,339-364. (2321 Rand, R.H.See Alfnend and Rand. Rand, R. H.. and S. F. Tmng (1969). On the stability of a differential equation with appli&tion to the vibrations of a particle m the p k . J. AppL Mech.,36,311-313. 11041 Rao, P. S. (1956). Supersonic bangs. Aeron. Quorr., 7,135-155. 1781 Rarity, B. S. H. (1969). A theory of the propagation of internal gravity waves of finite amplitude. J. W M e c h . , 39,497-509. [216]

408

REFERENCES AND AUTHOR INDEX

Rasmussen, M. L. (1970). Uniformly valid approximations for non-lineiu oscillations with small damping. In?. J. NOR-LinemMe&.. 5,687-696. [2321 Rayleigh, Lord (1912). On the propagation of waves through a stratified medium, with special reference to the question of reflection. BOCRoy. Soc. ( L o h n ) , A86, 208-226.[114,339] Rayleigh, Lord, (1917). On the reflection of light from aregularly stratifkd medium. Roc. Roy. Soc ( L o d n ) , A93,565-577. [95,367,368] Rehm, R. G. (1968). Radiative energy addition behind a shock wave. Phys. Fluids, 11, 1872-1883.1891 Reiss, E. L. (1971). On multivariable asymptotic expansions. SlAM Rev., 13, 189-196. I2321 Reiss, E. L., and B. J. Matkowsky (1971). Nonlhear dynamic buckling of a compressed elnstic column. Qwrt. Appl Mth., 29,245-260. [233] Reissner, E., and H. J. Weiuitschke (1963). Finite pure bending of circular cylindrical tubes. pumt. A@ Wtk,20,305-319. (541 Reynolds, W. C., and M. C. Potter (1967). Finite-amplitude instability of parallel shear flows. J. FluidMech.. 27,465-492.11621 Richmond, H. W.See Fowler, M o p , Lock, and Richmond. Richmond, O., and H.L. Morrison (1968). Application of a perturbation technique based on the method of characteristics to axisymmetric plasticity. 1 AppZ. Mech., 35, 117-122. [90] Rogister, A. (1971). Parallel propagation of nonlinear low-frequency waves in high-8 plasma. M ~ sW . S 14,2733-2739. (2351 Rogister, A. See Dobrowolny and Rogister. Roskes, G. J. See Benuey and Roskes. Ross, L. W.(1970). Perturbation analysis of diffusion-coupled biochemical reaction kinetics. SLAMJ. Appl. Mth., 19,323-329. (791 Rubbed, P. E., and M. T. Landahl(1967). Solution of the transonic airfoil problem through parametric differentiation. ALAA J., 5,470479.1235 J Rulf, B. (1967). Relation between creeping waves and lateral waves on a curved interface. L Math. M p . , 8,1785-1793. [380] Rulf, B. (1968). Uniform asymptotic theory of m ra c tion at an interface. Comm Pure AppL Mth., 21,67-76. [ 3801 Rutherford, P. See Frieman and Rutherford. Rytov, S. M. (1937). Mr act i o n of light by ultrasonic wave. Zzv. Akad. Nuak SSSR Ser. Fiz. No. 2,223-259 (in Russian). [ 373) Saffman, P. G. See Benney and Wman. Sakurai, A. (1965). Blast wave theory.InBasicDevebpments in Fluid Dymmiks, Vol. I (M. Holt, Ed.), Academic, New York, pp. 309-375. [78] Sakurai, T. (1968). Effect of the plasma impedance on the time variation of the inverse pinch.J. B y & Soc. Japn. 25,1671-1679.1781 Salpeter, E. E., and H. A. Bethe (1951). A relativistic equation for bound-state problems. Phys. Rev., 84,1232-1242. (3721 Saner, M. I. See Varvatsis and Sancer.

REFERENCES AND AUTHOR INDEX

409

Saucer, M. I., and A. D. Varvatsis (1969). An investigation of the renormalization and Rytov methods as applied to propagation in a turbulent medium. Northrop Corporate Laboratories Rept. No. 69-281. [ 3731 Sancer, M. I., and A. D. VarvataiS (1970). A comparison of the Born and Rytov methods. Roc. IEEE, 58,140-141. [373] Sanders, J. L., Jr., and A. A. Liepius (1963). Toroidal membrane under internal pressure. AIAAJ., 1,2105-2110. [356,357] Sandri, G. (1965). A new method of expansion in mathematical physics. Nuovo amento, B36,67-93. I2301 Sandri, G. (1967). Uniformization of asymptotic expansions. In Nonlinaar W & l Differentrbl Equations: A Symposium on Methods of Solutions gV. F. Ameq Ed.), Academic, New York, pp. 259-277. [230] Sandri, G. See Coldberg and Saudri; Klimas, Ramnath, and Sandri; Ramanathan and Sandri; Ramnath and Sandri. Saric, W.S. See Nayfeh and Saric. Savage, J. C., and H. S. Hasegawa (1967). Evidence for a linear attenuation mechanism. Geophysics, 32,1003-1014. (781 Schechter, H. B. (1968). The effect of three-dimensional nonlinear resonances on the motion of a particle near the earth-moon equilateral libration points. Semnd Compibtwn of Papers on Tkajectoty Am&sis and Guidcrnce Theory. NASA PM-67-21, 229-344. [ 1991 SchrSdinger, E. (1926). QuantisierUng als Eigenwertproblem. Ann. Phys., 80,437490. [56, 711 Schwartz, N.J. See Holt and Schwartz. Schwertassek, V. R. (1969). Grenzen von Mitnahmeberkhen. ZAMW, 49,409421. [232] Scott, A. C. (1970). Propagation of magnetic flwc on a long Josephson tunnel junction. Nuow Cimento, B69,241-261. [221] Scott, P. R. (1966). Equations of the oscillator with delayed amplitude limiting. Roc. IEEE, 54,898-899. [ 2581 Searl, J. W. (1971). Expansions for singular perturbationsJ. Inst. Muth. Appl. 8,131-138. [2331 Sellers, J. R., M. Tribus, and J. S. Klein (1956). Heat transfer to laminar flow in a round tube or flat conduit-The Graetz problem extended. Dun& A W E , 78, 441448. (3841 Sethna, P. R. (1963). TransKnts in certain autonomous multiple-degree-of-freedom nonlinear vibrating systems. J. Appl Mech.,30,4450. [ 168) Sethna, P. R. (1965). Vibrations of dynamicat systems with quadratic nonlinearities. J. AppL Me&.., 32,576-582. [ 188,2251 Shabbar, M. (1971). Side-band resonance mechanism in the atmosphere supporting Rossby waves. J. Atmoa Sci. 28,345-349. (2341 Shen, C. N. (1959). Stability of forced oscillations with nonlinear second-order terms. J. Appl Mech.,26,499-502. [lo41 Sheppard, L. M. See Lee and Sheppard. Shi, Y. Y. See Eckstein and Shi;Eckstein, Shi,and Kevorkian.

410 RE-

AND AUTHOR INDEX

S N Y. Y., and M. C. Eclrrtsin (1966). Ascent or dsllcsnt from satellite orbit by low thrust.

AL4AA 4,2203-2209. [233] Shi, Y. Y., and M. C. Eckstsjn (1968). Application of dDeulpr pertuxbath methods to ptobbma A m I;, 73,275-289. [233,276] Shlcarofdqr, I. P. (1971). Moduisd Born back scattehg from turbulent p h m u ~ ~ : Attonuation kdhg to saturation and aosepddatiua 1Iplrdyo Sct, 6,819-831. 13671 Shkil’, N. I. Ses Ferhchenlro, shidl’, and Nikohko. Shniad, H. (1970). The equtvllcPcc of von Zeipel mappings and Lis transforms. Celestial M&, 2,114120. (2021 Sbrestba, G. M.See T d and shrerth.. Sihya, Y. (1958). Sur r h c t i o n anslytisue d’un sy& d’&uations diff6rentielles ordinaims li&ires amtenant un paramhe. J. FBC scicnc~.Uni~.T o m , 7, 527-540. [3271 Sibuya, Y. (1963~).Asymptotic solutions of a liaeu ordinruy ditkmtid equation of nth order about a 3impb turning In Z n t a m S~ y m p ~ dDdffcrmthr ~~ Equa&ns mi N o d k a r Mcahrrrricr (J. P. La SaUm mnd S. Lafsch~tz.Eds.), Academic, New York, pp. 485488. [360) Sibuya, Y. (1963b). Simplification of a liwar ordinary differentid equation of the nth order at a tumbg point. Ar& f i t . Me& A d , 13,206221. [3601 Sibuya, Y. (1967). Subdomhutsolulionsof the differenthl equation y” - A’ Oe -a,) Or U ~ ) . . . ~ - U ~ ) Y =AO~.k d h t l r119,235-271.[345] . Sibuya, Y.See Hdch and Sbuya. Simmons, W. F. (1969). A variational method for weak maonant wave interactions. Roc Roy. Soc (I;onaon)),AJo9,551-575. [216] Sirignano, W. A., pad L. Crocco (1964). A shack wave model of unstable racket combustors. AIAA 1.. 2,1285-1296. [78) Sirovich, L. See Chcmg and Sirovich. SiVaSU‘ A. See Tang and Sivasubramanian. Soh, A. I. See Brull and SOlQ. Spriggs. J. H. A. F. Mcssitcr, and W. J. Anderson (1969). Membrane flutter paradox-An explnaation by -bation methods. ALQA J., 7,1704-1709. I234J Stanley, N. R. !ke Hufnagd and Stanley. Stceh, C. R. (1965). On the asymptotic solution of nonhomogeneousordinary differential equations with a parameter. Qutvt.A& Mth., 23,193-201. [353] Stetn, D. P. (197Qr). Dirtct tranrformationa 1. Wth. Aby~., 11, 2776-2781. [1WI Stern, D. P. (1970b). K ” s perturbation method. J. Mth f i y k . 11,2771-2775. [ 169) Stem, D. P. (1971a). A new formulption of CMoIljCpl perturbation theory. CelerthlMeJI., 3,241-246. [199] St-, D. P. (1971b). cla+dcp1 a d i a h t i ~theow. J. AWh. phyr, 12, 2231-2242. [169] Stem, D. P. (1971~).The canonization of nicc vari.bbk J. Math. my&, 12, 2226-2231. [911 Stewatson, K., and J. T. Stuart,(1971). A non-lineat instability theory for a wave system in plane Pohsuillc flow.1FMd Me&. 48,529-545. (2351

mt.

REFERENCES AND AUTHOR INDEX

4 11

Stoker, J. J. (1957) Water W m s . Wiley, New York. 157,771 Stokes, G. G. (1851). On the effect of the internal friction of fluids on the motion of pendulums. %ns. GmibMge Phil. Soc,9,&106. [ 301 Stokes, G. G. (1857). On the discontinuity of arbitrary constants which appear in divergent developments. CIlrnbrWgePhiL lhms.. 10,106128; all.thpem,4,77-109. 13121 Stone, P. H. (1969). The meridional structure of baroclinic waves. J.Atmos. Sci., 26, 376-389. [ 234) Strohbehn, J. W. (1968). Comments on Rytov’s method. 1. Opt. Soc.A m , 58, 139-140. (3731 Struble, R. A. (1962). Nonlineur Dflferent&zlEquutionr McGraw-Hill, New York. [171, 223) Stuart, J. T. (1958). On the nonlinear mechanics of hydrodynamic stability. J. FhridMech., 4,l-21. (1621 Stuart, J. T. (19606). Nonlinear effects in hydrodynamic stability. Roc. Xth Int. Cong. Appl. Mech. Stresa, Italy. [ 1621 Stuart, J. T. (1960b). On the nonlinear mechanics of wave disturbances in stable’and unstable pardel flows. Part 1. The basic behaviour in plane Poiseuille flow. 1. h i d Mech., 9,353-370. [ 1621 Stuart, J. T. (1961). On three-dimensional nonlinear effects in the stability of parallel flows. Adwn. A w n . &A. 3,121-142. Pergamon, Oxford. (1621 Stuart, J. T. See Stewartson and Stuart. Sturrock, P. A. (1957). Nonlinear effects in electron plasmas. Roc.Roy. Soc. (London), A242,277-299. [230] Sturrock, P. A. (1958). A variational principle and an energy theorem for small amplitude disturbances of electron beams and of electron-ion plasmas. Ann. Phys.. 4,306-324. [2161 Sturrock, P. A. (1962). PIoJm HydromgnetiC, Stanford University Press, Stanford, California. (2161 Sturrock, P. A. (1963). Nonlinear theory of electromagnetic waves in plasmas. Stanford University Microwave Laboratory Rept. No. 1004. [ 2301 Swanson, C. A. (1956). Differential equations with singular points. Tech. Rept. 16, Contract Nonr.-220(11), hpartment of Mathematics, California Institute of Technology. [3581 Sweet, J. (1971). Impulse of a ring with nonlinear matsial behavior. AIAA J., 9,332-334. [581 Tam, C. K. W. (1969). Amplitude dispersion and nonlinear instability of whistlers. phys. Fluids, 12,1028-1035. [234] Tam, C. K. W. (1970). Nonlinear dispersion of cold plasma waves. J. P h m . MYS.,4, 109-125.12341 Tam, K. K. (1968). On the asymptotic solution of the Orr-Sommerfeld equation by the method of multiple scales. L FhridMech., 34,145-158. I233.3601 Tang, T. W.,and A. Sivasubramanian, (1971). Nonlinear instability of modulated waves in a magnetoplasma. Phys. Fluids, 14,444-446. I 2171 Taniuti, T. See Asano and Taniuti. Tatarski, V. 1. (1961). Wuve Ropagntion in 4 Zbbukwt Medium McGrawHill, New York. [361,369,373]

4 12 REFERENCES AND AUTHOR INDEX Tatarslci, V. I. (1964). Propagation of electromagnetic waves in a medium with strong -mtant fluctuatiom Sovict ~ Y SJETP . @I@& -1, 19, 946-953.

[367,369,371,3721

Tamski, V. I., and M. E. Gertrenshteiu, (1963). Propagation of waves m a medium with strong ti^^ Of the refr+ctive index. Sodt WS. Jm &&), 179

458469.[382]

Taussig, R. T. (1969). Macroscopic quasilinear theory of high-frequency radiation in a cold pkrmra P h y ~ Fhdds. . 12,914-922.[234] Taylor, L. S. (1967). On Rytw’s method. Radio M,2,437441. I3731 Tempk, G. (1958). Lineariution and delhuization. Proceedings of the International Congrea~of Mathsmoticc, Edinburgh, pp. 233-247.[941 Te& R. M., and G. M. shrsrth, (1965). Lamiuar flow throughacbannelwith uniformly p o r ~ u walls s of different permeability.AppL S d Re&,A15,440468. [54,156] Thorn-, J. (1883). iiber Intsgrale zwaiter Gattung. 1. Reine A-. Miah., 95, 241-250.

-

[311.326]

Tidman, D. A. See Montgomery and Tidman. Timoshenko, S., and S. Womowsky-Wger, (1959). Theory of Phres und She&, 2nd ed., McGraw-Hill,New York. [361 Tin& L. See Kdlcr and Ting. Ting, L., and S. Brofman, (1964). On take-off from circular orbit by mail thrust. ZAhXf,

44,417428. [233] Tollmien, W. (1947). Asymptotische Integration der StonmgsdiBerentidgkichung ebener S t r 6 m v bei hohan Reynoldschen zahlea. ZAMM, 2S/27, 33-50, 70.83. [3W] T n h q M. Ste !Mers, Tnbus, and Klein. Tsen& S. F. See Rand and Tseng. Tsien, H.S. (1956). The Pomcar&Lighthill-Kuo method. Adwn. Appl. Me&, 4,281-349. [78,80,991 Tso, W. K., and T. K. Caughey, (1965). Parametric excitation of a nonlinear system. J. AppL Me&, 32,899-902.[2241 Tumptlrin, S. A. (1959). Asymptotic solution of a linear nonhomogeneous second order differential equation with a transition point and its application to the computations of toroedrl rbellr and Propdler blpdes.AppL A W L Me&. (H &%t.Me&, A W E mvrdj 23,1549-1565. [353,356] Usher, P. D. (1968). Coordinate stretching and interface location. 11. A new PL expansion. J. COnpntterPhy~.,329-39.(951 Usher, P. D. (1971). Necegtary conditions for applicability of Pd~~&Lighthiuperhubtion theory. Qauwt, A& firh., 28,463-471. [79] V a d b b w h , R. (1962). On the PLK method and the supersonic blunt-body problem. J. A m a S d . 29,185-206. [99] Van der Corput, J. G. (1956). Asymptotic developments I. Fundamental theorems of asymptotic+ J. AmL bhth., 4,341418. [18] Van der Corput, J. G. (1962).Asymptotic Expmdons Lecture notes, Stanford University. I121 Van der Pol, B. (1922). On a type of osciuation hysteresisin a simple triode generator.Phil 43,177-193. [3)

REFERENCES AND AUTHOR INDEX 4 13 Van der Pol, 8. (1926). On oscillation hysteresis in a simple triode generator. Phil Mug..43, 70@719.[164] Van der Pol, B. (1927). h e r Relaxations sch-. 1 . b . h h t l Tekgr. Te&pk. 28, 178-184.134) Van Dyke, M. D. (1952). A study of second-order supersonic flow theory.N.A.C.A. Rept. No. 1081. [271 Van Dyke, M. (1964). &mubution Methods in F?a& Mechmica Academic, New York. ~ll0,114,119,130] Van Hove, L.(1955). Quantum-mechanicalperturbations giving rise to a statistical transport equation. Hysicn, 21,517-540.[369] Van Hove, L. (1957). The approach to equilibhun m quantum statistics. P h y b , 23, 441-480. [3691 Van Hove, L., N. Hugenholtz, and L. Howland, (196l).Quuntum Theory ofMunyporticle System Benjamin, New York. (3671 Van Wijngaarden, L.(1968). On the oscillations near and at resonance m open pipes. J. Eng. &rh., 2,225-240.1891 Van Wijngaarden, L.See Verhagen and Van Wijngaarden. Varley. E.See Mortell and Varky; Parker and Varley. Vawatsis, A. D. See Sancer and Varvatsis. Varvatsis, A. D., and M. 1. Sancer, (1971). On the renormalization method in random wave propagation. Rudio Sci, 6,87-97. [ 3723 Vasil'eva, A. B. (1959). On repeated differentiation with respect to the panmeter of solutions of systems of ordinary differential equations with a small parameter in the derivative. fit. Sb., 48,311-334 (in Russian). [ 114,121J Vasil'eva, A. B. (1963). Asymptotic behavior of solutions of certain problems for ordinary nonlinear differential equations with a small parameter multiplying the highest dexivatives. Usp. Mar. Nuuk, 18, 15-86 (in RueSian); Rurdon f i t h . SrpveYs. 18, (1963) 13-81. [114] Verhagen, J. H. G.,and L. Van Wijngaarden, (1965). Non l i n a osciIlatioIlg of fluid in a container. 1.W M e c h . , 22,737-751. 1891 Veronis, C . See Malleus and Veronis. Visik, M. I., and L. A. Lyusternilq (1957). Regular degeneration and boundary layer for linear differential equations with small paraiierer. Usp. Mt. Nu&, 12,3-122 (in Russian); Am. Math. Soc. Transl., Serv.2,20,239-364,1962. [ 114,1441 Volosov, V. M.(1961). Higher approximations in averaging. Sovfer Moth. Do& 2,221-224. 11681 Volosov, V. M. (1962). Averaging in systems of ordinary differential equations. RugZlrn Murh. surveys, 7, 1-126. [ 1681 Von Zeipel, H. (1916). Movements of minor phuets. Ark. Mzr. Aslron. Stockholm, 11, NO. 1,l-58, NO. 7 , 1 4 2 . [189] Voss, W. (1933). Bsdingunsen F% das Aufbten dcs Ramsauereffektes. 2. Phyr, 83, 581-618. [ 3461 Wasow, W. A. (1953). Asymptotic solution of the differential equation of hydrodynamic stability in a domain containing a transition poiut. A m Mth., 58,222-252.13601 Wasow, W. A. (1955). On the convergence of an approximation method of M. J. Lighthill. J. Rat. Meek Anal. 4,751-767. [79]

m&,

4 14 REFERENCES AND AUTHOR INDEX Wmw, W. A. (1965). Asymptotic Expansions for Ordintuy Differential Eqwtions Wiley, New York. [110,309,318,358] W w w , W. A. (1968). Connecti~nproblems for asymptotic series. BulL Am. Math. Soc.. 74, 831-853. (3091 New York. Watson, G. N. (1944). A Trertise on the Theory of Besse2 Fwtctiollg Ma&, [3531 Watson, J. (1960). On the nonlinear mechanics of wave diptwbances in stable and unstable p d e l flows. Part 2. The development of a solution for plane PoiseuiUe flow and for plane Couette flow. J. FIuid Me&, 9,371-389. ( 1621 Weaver, A. K. See Green and Weaver. Weinitchke, H.J. See Reispner and Weinitschke. Weinstein, L. A. (1969). Open Resonutors and Open Waveguides Golem Press, Boulder, Colorado. ( 3801 Wentzel, G. (1926). Eine Verallgemeinerung der Quantenbedingung fur die Zwecke der Wellenmedwnik. Z. my&,38,518-529.[114,315,339] Weyl, H. (1942). On the differential equations of the simplest boundary-layer prohlems. Ann. Math., 43,381407. 11141 Whitehead, A. N. (1889). Second approximations to Viscous fluid motion. Quart. J. Math., 23,143-152. 1311 Whitehead, J. A. See Newell and Whitehead. Whiteman, K. J. See Mc Namara and Whiteman. Whitham, G. B. (1952). The flow pattern of a supersonic projectile. C o r n Pure Appl. Math., 5,301-348. [57,78,87] Whitham, G. B. (1953). The propagation of weak spherical shocks in stars. Comm Pure AppL Math., 6,397414. [57,78,87] Whitham, G. B. (1965a). A general approach to linear and nonlinear waves using a Lagrangian. 1. FluidMech.. 22,273-283. (216,2211 Whitham. G. B. (1965b). Nonlinear dispersive waves. mu.Roy. Soc. (London), A283, 238261. [226] Whitham, G. B. (1967a). Nonlinear dispersion of water waves. J. mUid Mech., 27,399412. 12161 Whitham,G. B. (1967b). Variational methods and applications to water waves. Roc. Roy. Soc (London),A299,6-25. I2161 Whitham, G . B. (1970). Twetiming, variational principles and waves. J. a i d Mech.. 44, 373-395. 12161 Whittaker, E . T. (1914). On the general solution of Mathieu’s equation. Edinburgh Math. SOC b., 32,75-80.[62] Whittaker, E. T. (1916). On the adelphic integral of the differential equations of dynamics. Proc. ROY. SOC. Edinb~rgh.37,95-116. 11991 Whittaker, E. T. (1937). Anulytial Dynamics of Particles and Rigid Bodies, 4th ed., Cambridge Univemity Press, Cambridge. ( 1991 Wilcox, C . H. (1964). Asymptotic Solutions of Dvfwentkzl Equations and Their A p p b t i o m Wiley, New York. 13091 Wilcox, C. H. (1966). Perhubrtion T h e m and its Appliaations in Qwnium Mechanics. Why, New York. [711

REFERENCES AND AUTHOR INDEX

41 5

Whmte, R. T.,and R. T. Davis, (1970). Perturbation solution of a hyperbolic equation governing longituainal wave propagation in cextain nonuniform bars. J. Acoust. Soc. Am., 47,1334-1337. I2341 Witt, A. See Gore& and Witt. Woinowsky-Krieger,S. See Timoshenko and Woinowsky-Krieger. Wolfe, P. (1967). A new approach to edge diffraction. SIAM J. Appl. Math., 15, 1434-1469. [ 3801 Wu, T. T. See Cheng and Wu. Wu, Y. T. (1956). Two-dimensional sink flow of a Viscous, heat-conducting, compressible fluid; cylindricalshock waves. Q w t . Appl. Math., 13,393418. [loo] Yagishita, T.See Kawakami and Yagishita. Ying, S.J. See Chu and Ying. Yuen, M. C. (1968). Nonlinear capillary instability of a liquid jet. J. Fluid Mech., 33, 151-163. [99] Zabreiko, P. P., and L. B. Ledovskaja, (1966). Higher approximations of the BogolinbovKrylov averaging method. Dokl. A h d . Nuuk. SSSR,171,1453-1456. [ 1681 Zauderer, E. (1964a). Wave propagation around a convex cylinder. J. Muth. Mech., 13, 171-186. [380) Zauderer, E. (1964b). Wave propagation around a smooth object. J. Murh. Mech., 13, 187-199. [380] Zauderer, E. (1970a). Boundary layer and uniform asymptotic expansions for diffraction problems. S W J .Appl. hkth., 19,575600. I3801 Zauderer, E. (1970b). Uniform asymptotic solutions of the reduced wave equation. J. Moth. Anal Appl., 30,157-171. [378,380] Zawadzki, E. M.,and G. J. Lewak, (1971). Penetration to second order of an electrostatic field into a warm plasma. J. P k s m Phys.. 5,73-87. I781 Zierep, V. J., and J. T. Heynak, (1965). Ein analytisches Verfahren zur Berechnung der nichtlinearen Wellenausbreitung. Z B ,45,3746. (891 Zwaan, A. (1929). lntensitaten im Ca-Funknspektrum. Ph. D. Thesis, Utrecht. (3391 Zwanzig, R. (1964). On the intensity of three generalized master equations. Physicu, 30, 1109-1123. [382]

Perturbation Methods ALI HASAN NAYFEH Copyright Q 2004 WILEY-VCH Valag GmbH & Co. KGaA

Subject Index

Acoustic, 77, 78 Aerodynamic, 110 Airfoil theory, 98,99,235, 303;see Supersonic airfoil theory Airy’s equation, 312,336 Airy’s functions, 49,336,378 Algebraic equation, 2,57,74,95,327 Algorithm, 168,171,199,209 for canonical systems, 212-214 generalized, 202-206 simplified, 206-208 Anomaly, 79 Aperiodic motion, 189 Astrophysics, 235 Asymptotic expansion, 23,78 of Airy’s functions, 337 of Bessel’s function, 16 defiition of, 12 divergent, 16 elementary operation on, 18-19 uniform, 17-19 uniqueness of, 14 see UlSO Asymptotic series Asymptotic matching principle; see Matching Asymptotic partitioning, 327-331 Asymptotic sequence, 12,14,16,18,19 factorial, 12 fractional powers in, 136, 137 logarithms in, 137,144 Asymptotic series, 10-12 defiition of, 1 1 versus convergent series, 15-16 see Asymptotic expansion Attenuation, 46,78 Averaging, method of, 159-227 generalized method of, 168-171,191,

211,223-225,231

Krylov-Bogolinbov, method of, 165-168

Krylov-Bogolinbov-Mopolski,method Of, 174-179,183,194,211,212, 223,224,246,248 Struble’s method of, 171-174,176,183, 194,223 using, canonical variables, 179-189 Lagrangian, 216-222 Lie series and transforms,200-246 von Zeipel transformation, 189-200 van der Pol’smethod of, 164-165 see ulso Smoothing Beam, niration of, 105,106,155,226,306 Bearing, slider, 54,125-128 Bellman, equation of, 22 Benard problem, 77 Bending of, shells and tubes, 54,353;see nlso Unsymmetricalbending of plates Benney’s technique, 3842 Bernoulli’s equation, 83 Besselfunctions, 1,5,6,15,312,315,383 asymptotic expansion of, 16,312-314,

329-331

integral representation of, 16,314 zeros of, 21 Bethe-Salpeter equation, 372 Blunt body problem, 99 Boltzmann’s equation, 236 Born approximation, 362 Born expansion, 308,361-367,373 renormalization of, 367-372 Boundary conditions, loss of, 31, 34,37,

38,54,111,114,122

transfer of, 27 Boundarylayer,18,23,79,111,112,147,

233

location of,114-116,122 Prandtl, 34

417

418

SUBJECT INDEX

problem with two, s, 128-133 stability of, 353 Branch point, 82 Bretherton's equation, exercises involving, 227,306 treated, by variational approach, 217-221 by method of multiple scales, 266-269, 298-300 wavewave interaction for, 219-221,266269 Brownian motion, 236 Buckling, 233 Canonical, averaging, variables, 179-189 equations, 180,199 Jordan, form, 327 mixed, variables, 199 system, 190,201 transformation, 181,187,191 variables, 181.183,184,195,202,216, 224 Caustic, 234,375-380 Change, 111,113 of characteristics, 89 see also Sharp change; Type change of Characteristic, parameters, 89,94 wave speeds,91 Characteristic exponent, 58,62,66 Characteristics, expansion in terms of, 57, 86-94,303 method of, 374 Circuit, electronic, see van der Pol oscillator Cluster expansions, 363 Compatability relationship, 217,222 Composite expansion, 114,144,384,385 construction of, 121 for earth-moon-spaceshipproblem, 139 for equation with variable coefficients, 125 for simple example, 121-122 for slider bearing, 128 for thermoelastic waves, 136 for unsymmetrical bending of plates, 133 Composite expansions, applied to turning point problems, 348-350 method of, 144-154,317,318 Composite solution, 113 Conservation form ofequations, 226 Coordinate, optimal, 50-51 parabolic, 79

perturbations, 1,4-7,21, 309-314,379 role of, systems, 23,49-52,54, 380 see olso Strained coordinates Correlation, 363 Cosmologicalexpansion, 233 Cycle, limit, 35 Cylinder, elliptic, 346 a solid, expanding, 83-86 Cy-lindrical,functions, 358,378,380 jet, 99 Degenerate, 72, 74 Derivative-expansion procedure, 302 applications of, 243-269 description of, 230,236-240 limitations of, 269-270 see ulso Multiple scales, method of Detuning, 250 Diagram, 366,367 bare, 364 connected, 370,371,372 double, 365 dressed, 365 Diffraction, 346,380 Diffusion equation, 38 Dirichlet problem, 38 Discontinuity, see Singularity Dispersion relationship, 178,217,218, 219, 220,222,227,266,301 Dispersive waves, 78,234, 303 long nonlinear, 3842 see ulso Bretherton's equation; KleinGordon equation; Thermoehstic waves; Wave-wave interaction Divisor, small, 196 Domain, effect of, on nonuniformity of expansions, 38,42 infinite, 24-31 Duffing equation, 5 0 , s 1 exercises involving, 54,104,105,223, 224,304 with slowly vary coefficients, 286 straightforward expansion for, 24-25 treated, by averaging using canonical variables, 182-183 by derivativeexpansion procedure, 243245 by generalized method of multiple scales, 286-291 by Krylov-Bogoliubov method, 167

SUBJECT INDEX

4 19

by Krylov-Bogoliubov-Mitropo~~ method, 175-176 by Lindstedt-Poincard method, 58-60 by renormalization, 95-96 by Struble's method, 171-174 by two-variable expansion procedure, 271-273 by von Zeipel's procedure, 192-194 Dyson equation, 371, 381

treated by method of matched asymp totic expansions, 139-144 Flutter, 234 Foci, 253 Fokker-Planck equation, 235, 381 Fourier, 45,167,178 Frequency, 56,58,96, 165,252 Fresnel diffraction, 380 Frobenius, method of, 5,310

Earth-moon-spaceship problem, 233,302, 303 exercises involving, 53, 107 illustrating limitations of method of strained coordinates, 102-103 straightforward expansion for, 4 3 4 5 treated, by Lighthill's technique, 82-83 by method of composite expansions, 153-154 by method of matched asymptotic expansions, 137-139 by method of multiple scales, 295-298 Eccentricity, 64, 233, 346 Edge layer, 111 Eiconal equation, 374, 377, 379 Eigenvalue, 56 Eigenvalue problem, linear, 68-71 quasilinear, 71-76 Elastic, 46, 58,90, 353 waves, 89-94 Elliptic equation, 37,42,98, 189,234, 303, 360 Energy level, 56, 58 Entropy layer, 79 Euler-Lagrangeequation, 216, 217, 218, 220,222

Gauge, function, 7,8 transformation, 232 Gaussian, 363, 367,369,372 Generalized expansion, see Composite expansion Generalized version of method of multiple scales, applications of, 276-302 description of, 232,241-243 limitations of, 302-303 Generalized method of averaging, see Averaging, method of Generalized vector, 179 Generating function, 181,184, 189,190, 192,195,196,200,202,215 Generating vector, 201 Geometrical optics, 308,361, 374-377, 379 Geophysics, 110 Graetz problem, 384 Green's function, 362, 364, 380 double, 364 Group velocity, 179,219,220,267,299

Faa de Bruno operators, 192 Feynman diagrams, 308,361-372 night mechanics, 233 Floquet theory, 60,62,64,66 Flow, down an inclined plane, 38-42,99 hypersonic, 79 jet, 99 past a body, 33-34, 113 through a channel, 54,56,216 see also Flow past a sphere; Supersonic airfoil theory Flow past a sphere, exercise involving, 158 straightforward expansion for, 28-31

Hamilton-Jacobi equation, 181, 182, 183, 186,190,199 Hamiltonian, 223,224,225 definition of, 180 for D u f f i equation, 182 for Mathieu equation, 184 for swingii spring,186 transformation of, using von Zeipel procedure, 189-200 using Lie transforms, 202, 212-216 Harmonic balance, 218 Harmonic resonance, 219-221, 234, 262269 Harmonic wave, 76 Heat, 45,79,84,156,342,384 problem for, equation, 150-152 Helmholtz equation, 234 Hill's equation, 60

420

SUBJECTINDEX

Hopf equation, 381 Hydraulicjump, 89 Hyperbolic equation, 37,42,57,99,379. See also Dispersivewaves; Elastic waves; and Supersonic airfoil theory

Jacobi elliptic functions, 189 Jerky oscillations, 34 Jordan, form, 327 matrix, 327 Josephson tunnel, 221

Inclined plane, see Flow, down an inclined Plane Indicial equation, 311 Induction, 13 Infiite domain, as a source,of nonuniformity, 23-31,229 of uniformity, 38,42 Inhomogeneous, 361 problems with turningpoints, 352-359 Initial, boundary value problem for heat equation, 150-152 layer, 23 Inner and our expansions, see Matched asymptotic expansions, method of Inner expansion, 110,112,114,119,145,

Kamel's algorithm. 171 Kamel's method, 224

146,148

for bending of plates, 129-132 d e f i t i o n , 117-118 for earth-moon-spaceship problem, 139 for equation with variable coefficients, 122-124

for simple example, 117-118 for slider beating, 127 for thermoelastic waves, 134-135 for turning point problems, 336 Inner limit, 112,118,119,129,131,139 Inner region, 113,122,146,153 Inner solution, 112,113 Inner variables, 119,120,121,144,149, 153,154

choice of, 114-116, 122-124, 126-127, 134-135,137-138

generalized, 145 Instability, 78,269. See also Model for nonlinear w b i l i t y ; Stability Integral, 10.11,12 differential equation, 18 equation, 370,380 of motion, 24,199 relations, 99 Intermediate limit, 119 Intermediate matching, see Overlapping Jacobian matrix, 202

Kelvin-Helmholtz,77,235

Kernel, 370 Kleincordon equation, 175,234 exercises involving, 105,226 treated, by averaging using Lagrmgh, 221-222

by Krylov-Bogoliubov-Mitropohki method, 178-179 by method of multiple scales, 301-302 by method of strained parameters, 7677

Krylov-Bogoliubov technique, 165-168,223 Krylov-Bgoliubov-Mitropolskitechnique, 174-179,183,194,211,212,223, 224,246,248,303

Kruskal's technique, 168,191 Lagrange equations, 179,180 Lagrangian, 179

averaging using, 216-222,301 for Bretherton's equations, 2 17 for KleinGordon equation, 221 for swinging spring, 186 Lamd coefficients of elasticity, 45,90

Laminar, 39,54

Landau, equation, 382 symbols, 8,9 Langer's transformation, 308,346,384 for first-order turning point problems, 339-341

for generalization of, 341-342 successive, 350 Latta's technique, 144-154 Layer, 23,79,111; see also Boundary layer Libration points, see Stability of elliptic triangular points Lie series and transforms, 171,199,200216,223,225,303

Lie triangle, 205,206 Lighthill's technique, 57,77-95 exercises involving, 107,108 limitations of, 79,98-100,107,109

SUBJECT INDEX Limit, 7 cycle, point, or solution, 35,99, 109 distinguished, 336 Oseen’s, 140 Stokes’, 140 see also Inner limit; Intermediate limit; and Outer limit Limitations of, method, of composite expansions, 153-154 of matched asymptotic expansions, 144-, 145,155,156,303,339 of multiple scales, 269-270, 275-276, 302-303 of strained coordinates, 79,98-103, 107,109,110,302,303 Struble’s method, 174 von Zeipel’s procedure, 199 Lindstedt-Poincard technique, 56,57,5860,95,96,185,200 Linear damped oscillator, treated by multiple scales, 228-243 Linearization, method of, 57,94-95 LiouviUe, equation, 236, 380, 382 problem, 314 LiouviUeGreen approximations, 49, 320, 335 higher, 315-317 successive, 324-325 LiouviUeGreen transformation, 308, 315, 340 Logarithms, 7,12,45,83,137,144,308, 311 Lommel functions, 353 Long period part, 169,191,209,214 Lunar motion, 60; see also Earth-moonspaceship problem Mach number, 26,83,85 Matched asymptotic expansions, method of, 37,48,51,78,110-144,148,153, 154,235,317,346,378,384 applied to turning point problems, 336339 limitations of, 144,155,156,303 Matching, 51,110,114,115,116

asymptotic, principle, see van Dyke’s principle as guide to forms of expansions, 141 intermediate, 119 Kaplun’s, principle, 119

421

Randtl, procedure, 112,118 refined, 118-119 Matching of inner and outer expansions, bending of plates, 130-132 caustic problem, 378 earth-moon-spaceship problem, 139 equation with variable coefficients, 124 now past a sphere, 141-144 simple example, 120-12 1 slider bearing, 127 term by term, 130 thermoelastic waves, 135-136 turning point problems, 337-339,343344 Mathien equation, 339 exercises involving, 53,104, 223, 303, 304 treated, by averaging using canonical variables, 183-185 by Lindstedt-Poincard technique, 6 0 6 2 by method of multiple scales, 253-257 by von Zeipei’s procedure, 194-200 by Whittaker’s method, 62-64 Maxwell’s heat conduction law, 45 Membrane, 356 Method, of strained coordinates, see Strained coordinates, method of of strained parameters, see Strained parameters, method of of multiple scales, see Multiple scales, method of Missile dynamics, 233 exercises involving, 304, 305, 306 linear, 320-321 nonlinear,291-295 Model for nonlinear instability, 50 exercises involving, 55 illustrating limitations of method of strained coordinates, 99-100 straightforward expansion for, 25-26 treated, by method of multiple scales, 264-266 by method of renormalization, 96-97 Momenta, 179,182,190,199,224 Moon,see Earth-moon-spaceship problem; Lunar motion Multiple scales, method of, 51,97, 100, 153,221,228-307,315, 317,339, 368,384 applied, to caustic, 379

422

SUBJECT INDEX

to equations with slowly varying coefficients, 318-320 to Orr-Sommerfeld equation, 360 Navier-Stokesequations, 28,33, 39 Neumann, expansion, 361; see also Born expansion series, 364 Newtonian theory, 79 Nondispersive waves, 78,269-270,303. See also Dispersive waves; Elastic; Shock waves; and Supersonic airfoil theory Nonuniformity, in airfoil theory, 28,50 in asymptotic expansions, 16-18 bibending of plates, 37 in Born’s expansion, 367 dependence of, on coordinates, 49-51 on size of domain, 24-31,38,42 in Duffii’s equation, 24-25,49-50 in earth-moon-spaceshipproblem, 45 in equation with constant coefficients, 32 in flow past a sphere, 31 in geometrical optics approximation, 376377

in interior of domain, 137 in jet instability, 99 in linear oscillator, 229 in model for nonlinear instability, 26,50 in relaxation oscillations, 35 in slider bearing, 126 in shift in singularity, 43,79 in thermoelastic waves, 4 7 4 8 in turning point problems, 49,284,315, 336

variable exhibiting, 57,285 see a h Boundary layer; Region of nonuniformity; Sources of nonuniformity; and Type change of Node, 252 Normal solution, 58,60,66,311,326,331, 332

No-dip condition, 33,34 Olver’s transformation, 341,384,385 Operator, 206 adjoint, 164 Faa de Bruno, 192 intensity, 372 mass, 371,381 self-adjoint. 163,164

Optics, see Geometrical optics Optimal, control, 79 coordinate, 50,51 Orbit, 64.99 Orbital mechanics, 233 Order symbols, 8,9 Orr-Sommerfeld equation, 233,360 Oscillations, 25,34,89,226,232,303 Oseen’s equation, 141 Oseen’s expansion, 140-144 Oseen’s limit process, 140,141 Oseen’s variable, 143 Outerexpansion, 110,112.114,119,146 for bending of plates, 128-129 deftnition of,117 for earth-moon spaceship problem, 45, 138

for equation with variable coefficients, 124

for simple example, 117 for slider bearing, 126 for thermoelastic waves, 4 7 4 8 Outer limit, 112,117,119,128,138,144 Outer region, 113,122 Outer solution, 111,113,115 Outer variable, 119, 121,144,145 Overlapping, 113,119 Parabolic coordinates, 79 Parabolic cylinder function, 344,378 Parabolic equation, 37,38,99,379 Paradox, 31 Parameter, 89 large, 315; see also Turning point problems perturbation, 14,16,308 see a h Small parameter multiplying highest derivative; Strained parameters, method of Parametric resonance, see Mathieu equation; Stability of elliptic ttiangularpoints Parameters, 201; see also Variation of parameters Parametrization, 88,92 Partitioning,see Asymptotic partitioning Pascal triangle, 205 Pendulum, 103,224; see Swinging spring Period, 169,191,192 Periodic, 58,60 motion, 189

SUBJECT INDEX orbit, 99 solutions, 34, 77 see Mathieu equation; Stability of elliptic triangular points Perturbations, coordinate, 4-7, 308-314 parameter, 14, 16, 308 Phase, 178, 380 rapidly rotating, 168, 201, 234 speed, 76,71 Plasma, 58, 78,216,217,226,234, 235, 367 Plasticity, 90 Plates, see Unsymmetrical bending of plates Poincard-Lighthill-Kuomethod, see Lighthill‘s technique Poisson ratio, 36,47 Potential function, 26,83,89 Prandtl’s boundary layer, 113 Prandtl’s technique, 111, 114 Pritulo’s technique, see Renormalization, method of Quantum, 58, 371; see also Schrodinger equation Random, 361,362,363,365,367, 369, 372,380,381,382 Rankine-Hugonoit relation, 84 Ratio test, 6, 7, 10, 11, 16 Rayleigh-Schrodingermethod, 56, 71 applied to eigenvalue problem, 68-71 Rayleigh-Taylorinstability, 77, 235 Rayleigh wave speed, 47, 134 Reaction kinetics, 79 Reentry dynamics, see Missile dynamics Regionof nonuniformity, 17,18, 19,23, 79 for earth-moon-spaceshipproblem, 137138 for equation with constant coefficients, 116 for flow past a sphere, 140 near a caustic, 377-378 for thermoelastic waves, 134-135 for turning point problems, 284, 336 Related equation, 341 Relaxation oscillations, 34-35 Renormalization, method of, 57,95-98, 308,361, 367-372 exercises involving, 103,106,107,108

423

Resonance, 89, 233,248 external, 225 internal, 185, 225 in linear systems, 321-324 near, 195 perfect, 189 parametric, 235 passage through, 233 see also Harmonic resonance Reynold’s equation, see Slider bearing Reynolds number, 40,360 high, 33,130 small, 28,29,139 Rigidity, flexural, 36 RitzGalerkin procedure, 58 Rossby wave, 77,234 Rytov’s method, 361,373 Saddle point, 252 Satellite, 233,276 Scales, 51, 110,230,231, 232,233, 242, 303; see also Multiple scales, method of Scattering, 57,95,235, 361, 364, 367, 368 Schrodinger equation, 56,71,160, 339, 342,344,346 Secular terms, 23-26 elimination of, 6548,261, 288, 302 Self-sustained oscillations, 89; see also van der Pol oscillator Series, see Asymptotic series; Lie series and transforms Shadow boundary, 378 Sharp change, 103,110,303 Shell, 233 Shift, 42,77,79,106 in singularity, 103 exercises involving, 52,53, 106, 107 straightforward expansion for, 42-43 treated, by Lighthill’s technique, 79-82 by renormalization, 98 by Temple’s technique, 94-95 Shock waves, 52,78,83, 84,85,99, 100, 235 Short period part, 169,191,214 Singular,17,57,78,81,231 Singular perturbation, 17,99,340 dependence on region size, 38,42 see also Nonuniformity Singular point, definition of, 309,310

424

SUBJECT INDEX

expansion near an irregular, 309-312,326327 regular, 5,308 Singularity, 85,86 as a jump discontinuity, 32,35 branch point, 82 essential, 7 growing, 23,42,43,45,95 logarithmic, 45,83,102,137 tuning point with, 358-359 worst, 81,98 see also Nonuniformity Skin layer, 111 Slider bearing, 54,125-128 Small parameter multiplying highest derivative, 23,31-37,99, 317-318 in limitations of method of strained coordinates, 99,100-102 Smoothing, method of, 308,361,380-382 Solvability condition, 152,288,302; see Secular terms Sonic boom, 78 Sources of nonuniformity, 23-55, 140 Spaceship, see Earth-moon-spaceshipproblem Speed, 56,57,58,76,77,83,90,91 Sphere, see Flow past a sphere Spherical, 28,78,224 Spring, 232. See also D u f f i i equation; swinging spring Stability, 60,77,78,99,100,162-164,235, 252,353,360. See also Mathieu equation; Model for nonlinear stability; Stability of elliptic triangular points Stability of elliptic triangular points, treated, by method of multiple scales, 259262,275 by method of strained parameters, 6466 by Whittaker’s technique, 6 6 6 8 Statistical mechanics, 236 Stochastic, 361,367 Stokes’ expansion, 140 Stokes’ limit process, 140 Stokes’ variable, 143 Stokes’ solution, 30 Strained coordinates, method of, 51.52, 56-109,110.114.156,235,302, . . 303

Strained parameters, method of, 56,58-77, 78,79,97,99,100,103,105,106 Strainitg, of characteristics, 87,89 dependent variable, 99,108 function, 57,78,79,81,87,99,101,102, 103,107 Stratification, 216 Stream function, 28,29,33,40,57,113, 140 Stretching transformation, for bending of plates, 128 for caustic, 377-378 for dependent and independent variables, 134 for earth-moon-spaceship problem, 137138 for equation with variable coefficients, 122,123 for heat equation, 151 for turning point problems, 284,336 Struble’s method, 171-174,176,183,194, 223 Stuart-Watson-Eckhaustechnique, 162-164 Subnormal solution, 311,331-332 Supersonic airfoil theory, 50,78,93 straightforward expansion for, 26-28 treated, by Lighthill’s technique, 86-89 by renormalization, 97-98 swinging spring,exercises involving, 105, 225 treated, by averaging Hamiltonian, 185189 by Lie series and transforms, 214-216 by method of multiple scales, 262-264 Temple’s technique, 57,94-95 Thermoelastic waves, straightforward expansion for, 45-48 treated by method of matched asymptotic expansions, 133-137 Thomas-Fermi model, 235 Three body problem, 199,233,276. See also Earth-moon-spaceship problem; Stability of elliptic triangular points Triangle, 205,206 Triangular points, 233; see also Stability of elliptic triangularpoints Transfer of boundary condition, 27 Transform, see Lie series and transforms Transformation, canonical, 181,190,

SUBJECT INDEX 191,199 contracting, 140 Deprit, 202 Hori, 202 Lie, 201 LiouvfileGreen, 49,315,340 near identity, 51,57,168,190,201 stretching, 111,113,114,116,117,377 von Zeipel, 171,191,199,202 Transition, 380; see Turning point problems Transition curves, exercises involving, 104, 105,225,303,304 for libration points, 64-68, 259-262.275 for Mathieu’s equation, 60-64, 183-185, 194-200,253-257 Transport equation, 374, 379 Tunneling effect, 342 Turning point problems, 114,308,309, 335-360 definition of, 49,122,284,335 exercises involving, 53, 305, 383, 384, 385 near caustic, 377-380 treated by method of multiple scales, 232, 233,284-286 Two-body problem, 201 Two-variable expansion procedure, 243, 302 applications of, 270-275 description of, 231, 240-241 limitations of, 275-276 see also Multiple scales, method of Type change of, 23,3742

Uniformity, see Nonuniformity Uniformization, 94, 107, 232 Unsymmetrical bending of plates, exercises involving, 155 straightforward expansion for, 35-37 treated by method of multiple scales, 128‘133 van der Pol, method of, 164-165,166 van der Pol oscillator, exercises involving, 52,54,104,223,224,303,304 straightforward expansion for, 3 4 , 34-35

425

treated by generalized method of averaging, 169-171 by Krylov-Bogoliubov technique, 167168 by Krylov-Bogoliubov-Mitro~l~i technique, 176-177 by Lie series and transforms, 209-212 by method of multiple scales, 245-253, 272-275 van der Pol oscillator with delayed amplitude liiting, 104,257-259,304 van Dyke’s matching principle, 114, 119 mechanics of, 120-122 Variables, see Canonical variables; Inner variables; and Outer variables Variation of parameters, 52,159-222 Variational, approach, 216-222, 227 equations, 172,183, 184, 190 Vibrations, 114,232. See also Oscillations; Waves Vlasov’s equation, 78, 216 von Zeipel’s procedure, 189-200, 202,231 Water waves, 56,57,58,77,216,217,234 Wave equations, 360-382 Wave number, 58,77,99,360 Waves, 58,60,77, 78,89,90,94, 216, 217, 233,234,235. See also Bretherton’s equation; Dispersive waves; Elastic waves; Flow down an inclined plane; KleinGordon equation; Non dispersive waves; F’lasma; Shock waves; Supersonic airfoil theory; and Thermoelastic waves Weber function, 345, 378,380 Whitham’s method, 216-222 Whittaker’s functions, 346 Whittaker’s method, 6244,6668, 104, 105,185,200 WKBJ approximation, 49,308, 315, 320, 335,339,382 successive, 324-325 Wronskian, 160 Young’s modulus, 36

Differential Geometry (Wiley Classics Library)

Read more

Statistical Optics (Wiley Classics Library)

Read more

Perturbation Methods

Read more

Perturbation Methods

Read more

Perturbation methods

Read more

Perturbation methods

Read more

Perturbation Methods

Read more

Perturbation methods

Read more

Perturbation methods

Read more

Statistical Design for Research (Wiley Classics Library)

Read more

Introduction to perturbation methods

Read more

Perturbation methods with Mathematica

Read more

Supermoney (Wiley Investment Classics)

Read more

Supermoney (Wiley Investment Classics)

Read more

Introduction to Perturbation Methods 2e

Read more

Perturbation Methods in Fluid Mechanics

Read more

Singular Perturbation Methods in Control

Read more

Bayesian Inference in Statistical Analysis (Wiley Classics Library)

Read more

Water Waves: The Mathematical Theory with Applications (Wiley Classics Library)

Read more

Quantum Statistical Properties of Radiation (Wiley Classics Library)

Read more

Water Waves: The Mathematical Theory with Applications (Wiley Classics Library)

Read more

Water Waves: The Mathematical Theory with Applications (Wiley Classics Library)

Read more

Linear Operators, Part I: General Theory (Wiley Classics Library)

Read more

Foundations of Differential Geometry (Wiley Classics Library) (Volume 1)

Read more

Foundations of Differential Geometry (Wiley Classics Library) (Volume 2)

Read more

We (Modern Library Classics)

Read more

Otello (Opera Classics Library)

Read more

Salome (Opera Classics Library Series) (Opera Classics Library Series)

Read more

Madama Butterfly (Opera Classics Library Series) (Opera Classics Library)

Read more

Aida (Opera Classics Library Series) (Opera Classics Library)

Read more

Recommend Documents

Differential Geometry (Wiley Classics Library)

Statistical Optics (Wiley Classics Library)

Statistical Optics JOSEPH W. GOODMAN Wiley Classics Library Edition Published 2000 A Wiley-Interscience Publication ...

Perturbation Methods

Perturbation Methods

Perturbation methods

Perturbation methods

Perturbation Methods

Perturbation Methods ALI HASAN NAYFEH Wiley-VCH Verlag GmbH & Co. KGaA This Page Intentionally Left Blank Perturb...

Perturbation methods

Perturbation methods

Statistical Design for Research (Wiley Classics Library)

Statistical Design for Research This Page Intentionally Left Blank Statistical Design for Research This Page Inte...