A V9r9'APPLIED FIRST COURSE IN

I /I
f T ar

CIF
TINS

1:
MICHAEL K.. KEANE
A Very Applied First Course
in
Partial Differential Equations

MICHAEL K. KEANE
Department of Mathematical Sciences
United States Air Force Academy

hail

Prentice Hall
Upper Saddle River, New Jersey 07458
Library of Congress Cataloging-in-Publication Data
Keane, Michael, K.
A very applied first course in partial differential equations / Michael K. Keane
p. cm.
ISBN: 0-13-030417-4
1. Differential equations, Partial. I. Title

QA377.K38 2002
515'.353-dc21 2001040032

Acquisition Editor: George Lobell

Editor-in-Chief: Sally Yagan
Vice President/Director of Production and Manufacturing: David W. Riccardi
Executive Managing Editor: Kathleen Schiaparelli
Senior Managing Editor: Linda Mihatov Behrens
Production Editor: Bob Walters
Manufacturing Buyer: Alan Fischer
Manufacturing Manager: Trudy Pisciotti
Marketing Manager: Angela Battle
Marketing Assistant: Vince Jansen
Director of Marketing: John Tweeddale
Editorial Assistant: Melanie VanBenthuysen
Art Director: Jayne Conte
Cover Design: Bruce Kenselaar
Cover Illustration: Michael K. Keane

©2002 Prentice-Hall, Inc.

Upper Saddle River, New Jersey 07458

All right reserved. No part of this book may be

reproduced, in any form or by any means,
without permission in writing from the publisher.

Printed in the United States of America

10 9 8 7 6 5 4 3 2 1

ISBN: 0-13-030417-4

Pearson Education LTD., London

Pearson Education Australia PTY, Limited, Sydney
Pearson Education Singapore, Pte. Ltd.
Pearson Education North Asia Ltd., Hong Kong
Pearson Education Canada, Ltd., Toronto
Pearson Educaciun de Mexico S.A. de C. V.
Pearson Education-Japan, Tokyo
Pearson Education Malaysia, Pte. Ltd.
 To my loving wife, Jean,
and
in memory of my friend, Sonya
Contents

Preface xvii

1 Introduction 1

2 The One-Dimensional Heat Equation 5

2.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 DERIVATION OF HEAT CONDUCTION
IN A ONE-DIMENSIONAL ROD . . . . . . . . . . . . . . . . . . . 5
2.2.1 Derivation of the Mathematical Model . . . . . . . . . . . . . 7
2.2.2 Initial Temperature . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 BOUNDARY CONDITIONS
FOR A ONE-DIMENSIONAL ROD . . . . . . . . . . . . . . . . . . 17
2.3.1 Boundary Conditions of the First Kind . . . . . . . . . . . . 17
2.3.2 Boundary Conditions of the Second Kind . . . . . . . . . . . 18
2.3.3 Boundary Conditions of the Third Kind . . . . . . . . . . . . 19
2.4 THE MAXIMUM PRINCIPLE AND UNIQUENESS . . . . . . . . . 23
2.5 STEADY-STATE TEMPERATURE DISTRIBUTION . . . . . . . . 30

3 The One-Dimensional Wave Equation 41

3.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 DERIVATION OF THE ONE-DIMENSIONAL
WAVE EQUATION . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3 BOUNDARY CONDITIONS . . . . . . . . . . . . . . . . . . . . . . 49
3.3.1 Boundary Conditions of the First Kind . . . . . . . . . . . . 50
3.3.2 Boundary Conditions of the Second Kind . . . . . . . . . . . 50
3.3.3 Boundary Conditions of the Third Kind . . . . . . . . . . . . 50
3.4 CONSERVATION OF ENERGY
FOR A VIBRATING STRING . . . . . . . . . . . . . . . . . . . . . 54
3.5 METHOD OF CHARACTERISTICS . . . . . . . . . . . . . . . . . 57
3.6 D'ALEMBERT'S SOLUTION TO THE
ONE-DIMENSIONAL WAVE EQUATION . . . . . . . . . . . . . . 64
vi Contents

4 The Essentials of Fourier Series 73

4.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2 ELEMENTS OF LINEAR ALGEBRA . . . . . . . . . . . . . . . . . 74
4.2.1 Vector Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2.2 Linear Dependence, Linear Independence, and Basis . . . . . 76
4.2.3 Orthogonality and Inner Product . . . . . . . . . . . . . . . . 78
4.2.4 Eigenvalues and Eigenvectors . . . . . . . . . . . . . . . . . . 83
4.2.5 Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.3 A NEW SPACE: THE FUNCTION SPACE
OF PIECEWISE SMOOTH FUNCTIONS . . . . . . . . . . . . . . . 90
4.3.1 Inner Product, Orthogonality, and Basis
in a Function Space . . . . . . . . . . . . . . . . . . . . . . . 92
4.3.2 Definition of Trigonometric Fourier series . . . . . . . . . . . 100
4.3.3 Fourier series Representation of Piecewise
Smooth Functions . . . . . . . . . . . . . . . . . . . . . . . . 103
4.4 EVEN AND ODD FUNCTIONS AND
FOURIER SERIES .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 119

5 Separation of Variables:
The Homogeneous Problem 131
5.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.2 OPERATORS: LINEAR AND HOMOGENEOUS
EQUATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.2.1 Linear Operators . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.2.2 Linear Equations . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.3 SEPARATION OF VARIABLES: HEAT EQUATION . . . . . . . . 139
5.3.1 Spatial Problem Solution . . . . . . . . . . . . . . . . . . . . 141
5.3.2 Time Problem Solution . . . . . . . . . . . . . . . . . . . . . 144
5.3.3 The Complete Solution . . . . . . . . . . . . . . . . . . . . . 145
5.4 SEPARATION OF VARIABLES: WAVE EQUATION . . . . . . . . 153
5.4.1 Spatial Problem Solution . . . . . . . . . . . . . . . . . . . . 155
5.4.2 Time Problem Solution . . . . . . . . . . . . . . . . . . . . . 156
5.4.3 The Complete Solution . . . . . . . . . . . . . . . . . . . . . 156
5.5 THE MULTIDIMENSIONAL SPATIAL PROBLEM . . . . . . . . . 164
5.5.1 Spatial Problem for X (x) . . . . . . . . . . . . . . . . . . . . 167
5.5.2 Spatial Problem for Y(y) . . . . . . . . . . . . . . . . . . . . 168
5.5.3 Time Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.5.4 The Complete Solution . . . . . . . . . . . . . . . . . . . . . 168
5.6 LAPLACE'S EQUATION . . . . . . . . . . . . . . . . . . . . . . . . 180
5.6.1 An Electrostatics Derivation of Laplace's Equation . . . . . . 180
5.6.2 Uniqueness of Solution . . . . . . . . . . . . . . . . . . . . . . 181
5.6.3 Laplace's Equation in Cartesian Coordinate System . . . . . 182
 Contents vii

6 The Calculus of Fourier Series 191

6.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.2 FOURIER SERIES REPRESENTATION OF A
FUNCTION: FOURIER SERIES AS A FUNCTION . . . . . . . . . 191
6.3 DIFFERENTIATION OF FOURIER SERIES . . . . . . . . . . . . . 195
6.4 INTEGRATION OF FOURIER SERIES . . . . . . . . . . . . . . . 206
6.5 FOURIER SERIES AND THE GIBBS PHENOMENON . . . . . . . 213

7 Separation of Variables: The Nonhomogeneous Problem 223

7.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
7.2 NONHOMOGENEOUS PDES
WITH HOMOGENEOUS BCS . . . . . . . . . . . . . . . . . . . . . 224
7.3 HOMOGENEOUS PDE WITH
NONHOMOGENEOUS BCS . . . . . . . . . . . . . . . . . . . . . . 236
7.3.1 Homogeneous PDE Nonhomogeneous Constant BCs . . . 236
7.3.2 Homogeneous PDE Nonhomogeneous Variable BCs . . . . . 242
7.4 NONHOMOGENEOUS PDE AND BCs . . . . . . . . . . . . . . . . 258
7.5 SUMMARY . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . 277

8 The Sturm-Liouville Eigenvalue Problem 279

8.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
8.2 DEFINITION OF THE STURM-LIOUVILLE
EIGENVALUE PROBLEM . . . . . . . . . . . . . . . . . . . . . . . 281
8.3 RAYLEIGH QUOTIENT . . . . . . . . . . . . . . . . . . . . . . . . 293
8.4 THE GENERAL PDE EXAMPLE . . . . . . . . . . . . . . . . . . . 298
8.5 PROBLEMS INVOLVING HOMOGENEOUS BCS OF THE THIRD
KIND ...................................302
9 Solution of Linear Homogeneous Variable-Coefficient ODE 317
9.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
9.2 SOME FACTS ABOUT THE GENERAL
SECOND-ORDER ODE . . . . . . . . . . . . . . . . . . . . . . . . . 318
9.3 EULER'S EQUATION . . . . . . . . . . . . . . . . . . . . . . . . . . 320
9.4 BRIEF REVIEW OF POWER SERIES . . . . . . . . . . . . . . . . 325
9.5 THE POWER SERIES SOLUTION METHOD . . . . . . . . . . . . 327
9.6 LEGENDER'S EQUATION AND
LEGENDRE POLYNOMIALS . . . . . . . . . . . . . . . . . . . . . 334
9.7 METHOD OF FROBENIUS AND BESSEL'S EQUATION . . . . . 341

10 Classical PDE Problems 351

10.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
10.2 LAPLACE'S EQUATION . . . . . . . . . . . . . . . . . .
. . . . . . 351
10.2.1 Laplace's equation in the Polar Coordinate System . . . . . . 352
viii Contents

10.2.2 Laplace's equation in the Spherical

Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . 356
10.3 TRANSVERSE VIBRATIONS OF A THIN BEAM . . . . . . . . . 364
10.3.1 Derivation of the Beam Equation . . . . . . . . . . . . . . . . 364
10.3.2 Transverse Vibrations of a Simply
Supported Thin Beam . . . . . . . . . . . . . . . . . . . . . . 367
10.4 HEAT CONDUCTION IN A CIRCULAR PLATE . . . . . . . . . . 373
10.5 SCHRODINGER'S EQUATION . . . . . . . . . . . . . . . . . . . . 380
10.6 THE TELEGRAPHER'S EQUATION . . . . . . . . . . . . . . . . . 389
10.6.1 Development of the Telegrapher's Equation . . . . . . . . . . 389
10.6.2 Application of the Telegrapher's Equation
to a Neuron . . . . . . . .
. .. . .
. .. . . . . . . . . . . . 391
10.7 INTERESTING PROBLEMS IN DIFFUSION . . . . . . . . . . . . 398

11 Fourier Integrals
and Transform Methods 403
11.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
11.2 THE FOURIER INTEGRAL . . . . . . . . . . . . . . . . . . . . . . 404
11.2.1 Development of the Fourier Integral . . . . . . . . . . . . . . 405
11.2.2 The Fourier Sine and Cosine Integrals . . . . . . . . . . . . . 414
11.3 THE LAPLACE TRANSFORM . . . . . . . . . . . . . . . . . . . . 421
11.3.1 Laplace transform Solution Method of ODEs ..........422
11.3.2 The Error Function ........................424
11.3.3 Laplace Transform Solution Method of PDEs ..........425
11.4 THE FOURIER TRANSFORM . . . . . . . . . . . . . . . . . . . . . 439
11.4.1 Fourier Cosine and Sine Transforms . . . . . . . . . . . . . . 442
11.4.2 Fourier Transform Theorems . . . . . . . . . . . . . . . . . . 444
11.5 FOURIER TRANSFORM SOLUTION METHOD
OF PDES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450

Appendices 462

A Summary of the Spatial Problem 463

B Proofs of Related Theorems 465

B.1 THEOREMS FROM CHAPTER 2 . . . . . . . . . . . . . . . . . . . 465
B.1.1 Leibniz's Formula . . . . . . . . . . . . . . . . . . . . . . . . 465
B.1.2 Maximum-Minimum Theorem . . . . . . . . . . . . . . . . . 466
B.2 THEOREMS FROM CHAPTER 4 ....................469
B.2.1 Eigenvectors of Distinct Eigenvalues
Are Linearly Independent . . . . . . . . . . . . . . . . . . . . 469
B.2.2 Eigenvectors of Distinct Eigenvalues of an n by n
Matrix Form a Basis for WL 470
 Contents ix

B.3 THEOREM FROM CHAPTER 5 . . . . . . . . . . . . . . . . . . . . 470

C Basics from Ordinary Differential Equations 473

C.1 SOME SOLUTION METHODS FOR FIRST-ORDER ODES . . . . 473
C.1.1 First-Order ODE Where k(t) Is a Constant . . . . . . . . . . 473
C.1.2 First-Order ODE Where k(t) Is a Function . . . . . . . . . . 475
C.2 SOME SOLUTION METHODS OF
SECOND-ORDER ODES . . . . . . . . . . . . . . . . . . . . . . . . 479
C.2.1 Second-Order Linear Homogeneous ODES . . . . . . . . . . . 479
C.2.2 Second-Order Linear Nonhomogeneous ODES . . . . . . . . . 485

D Mathematical Notation 489

E Summary of Thermal Diffusivity of Common Materials 491

F Tables of Fourier and Laplace Transforms 493

F.1 TABLES OF FOURIER, FOURIER COSINE,
AND FOURIER SINE TRANSFORMS .................494
F.2 TABLE OF LAPLACE TRANSFORMS . . . . . . . . . . . . . . . . 498

Bibliography 501

Index 504
List of Figures
2.1 One-dimensional rod of length L. 6
2.2 One-dimensional rod of length L . . . . . . . . . . . . . . . . . . . . . 10
2.3 Arbitrary slice [a, b] of one-dimensional rod. . . . . . . . . . . . . . . 10
2.4 Temperature distribution of slice [a, b] . . . . . . . . . . . . . . . . . . 12
2.5 Heat flow from end x = L .. . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Heat flow at end of rod x = 0 .. . . . . . . . . . . . . . . . . . . . . . 19
2.7 The domain of u(x, t) in the rectangle 0 < x < L and 0 < t < T. 24
2.8 Steady-state temperature distribution with u(0) = a and u(L) = b. 32
2.9 Graph of 5 cos x + 20. . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.1 Finite string segment. . . . . . . . .

. . . . . . . . . . . . . . . . . . 42
3.2 String attached to elastic ends with their ends fixed. . . . . . . . . . 51
3.3 String attached to elastic ends with their ends displaced.. . . . . . . 51
az az
3.4 The solution sketch for - 3 =0. . . . . . . . . . . . . . . . . . 60
at ax
3.5 Solution sketch for u(x, t) = sin(x - 0.25t)e-o.5t
61
3.6 Graph of d'Alembert's solution for t = 0 .. . . . . . . . . . . . . . . . 67
a
3.7 Graph of d'Alembert's solution for 0 <t < . . . . . . . . . . . . . . 68
c
a
3.8 Graph of d'Alembert's solution for t = - . 68
c
a
3.9 Graph of d'Alembert's solution for t> - . . . . . . . . . . . . . . . . 69
c

4.1 f (x) _ 1] on the interval [-2,2]...................90

4.2 A function, 1(x), with several jump discontinuities.. . . . . . 90 . . . .

4.3 1(x) = tan x on the interval [0, ?C] .. . . . . . . . . . . . . . . . . . . . 91

4.4 1(x) _ x on the interval [-2, 2]......................91
4.5 The graph of 1(x) _ -x and g(x) = 2 on the interval [-L, L]. . . . . 94
4.6 1(x) = x3 + 7x2 - 4x - 10 on the interval [-2, 2].............104
4.7 The Fourier series representation of 1(x) = x3 + 7x2 - 4x - 10 for
the partial sum Sl (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.8 The Fourier series representation of 1(x) = x3 + 7x2 - 4x - 10 for
the partial sum S5 (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

xi
xii List of Figures

4.9 The Fourier series representation of 1(x) = x3 + 7x2 - 4x - 10 for

the partial sum S12 (x) .. . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.10 The Fourier series representation of 1(x) = x3 + 7x2 - 4x - 10 for
the partial sum S25(x) .. . . . . . . . . . . . . . . .
. .. . 106
. . . . . .

4.11 h(x) = x2 + 2x - ion the interval [-5,5]................ 106

4.12 Fourier Series representation of h(x) = x2 + 2x - 1 for the partial
sum Si (x). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.13 Fourier Series representation of h(x) = x2 + 2x - 1 for the partial
sum S5(x). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.14 Fourier Series representation of h(x) = x2 + 2x - 1 for the partial
sum S25(x). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.15 Fourier Series representation of h(x) = x2 + 2x - 1 for the partial
sum S50(x). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.16 Fourier Series representation of h(x) = x2 + 2x - 1 for the partial
sum S75(x). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.17 Fourier Series representation of h(x) = x2 + 2x - 1 for the partial
sum S1oo (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.18 h(x) = 32 + >IIn=1
100 100(-1)n
n2r2
L
COS
nirx
5
- n 20(-1)n sin n5x
graphed on
the interval [-15,15] .. . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.19 The graph of the function g(x) on the interval -1 < x < 1.......109
4.20 g(x) = 2 + n=1 [['r1
sin (nirx)] on the interval [-1,1]..... 111

4.21 g(x) = 2 + [[1_1) sin (nirx)] on the interval

n
>n°_1 [-1,11..... 111

4.22 g(x) = 2 + n5 i [[1_(_1r1 sin (nirx)] on the interval [-1,1]..... 111

n
4.23 g(x) = 2 + >I=i 1-( 1 sin (nirx)] on the interval [-1,1]..... 111
4.24 g(x) = 2 + >jnooi [1) 1-( n
sin (nirx)] on the interval [-1, 1]. . . . . 111

4.25 g(x) _ 23 +1000 [1-(-1)n]

n=1 n sin (nirx) on the interval [-1,1]..... 111
4.26 g(x) = 2 + nooo 1-(

4.27 cos x on the interval [-?C, ?C] .. .

i n
sin (nirx)] on the interval [-3,3]..... 112
. . . . . . . . . . . . . . . . . . . . . 113
4.28 cos x on the interval [-371, 3ir]. . . . . . . . . . . . . . . . . . . . . . 113
4.29 An even function on the interval [-L, L] .. . . . . . . . . . . . . . . . 120
4.30 An odd function on the interval [-L, L] . . . . . . . . . . . . . . . . . 120
4.31 1(x) = x on the interval [0, 2] .. . . . . . . . . . . . . . . . . . . . . . 121
4.32 7(x) = x on the interval [-2, 2]. . . . . . . . . . . . . . . . . . . . . 122
4.33 Fourier cosine series representation of 7(x) = x for the partial sum
Si (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.34 Fourier cosine series representation of 7(x) = x for the partial sum
Sio (x) .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.35 Fourier cosine series representation of 7(x) = x for the partial sum
S25 (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
 List of Figures xiii

4.36 Fourier cosine series representation of 7(x) = x for the partial sum
S5o (x) .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.37 Fourier cosine series representation of 7(x) = x) on the interval [-8, 8].124
4.38 f (x) = x on the interval [-2, 2] .. . . . . . . . . . . . . . . . . 124 . . . .

4.39 Fourier sine series representation of 1(x) = x for the partial sum Si(x). 125
4.40 Fourier sine series representation of 1(x) = x for the partial sum
Sio (x) .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.41 Fourier sine series representation of f (x) = x for the partial sum
S25 (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.42 Fourier sine series representation of f (x) = x for the partial sum
S5o (x) .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.43 Fourier sine series representation of f (x) = x on the interval [-8,8]. 126

5.1 y = cosh x. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

5.2 y = sink x .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.3 y = sin x. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.4 The graph of u(x, t) for 0 < t< 2. . . . . . . . . . . . . . . . . . . . 148
5.5 The graph of u (x, t) for 0 < t < 5 .. . . . . . . . . . . . . . . . . . . . 159
5.6 The graph of u(x, y, t) for time t = 0 .. . . . . . . . . . . . . . . . . . 175
5.7 The graph of u (x, y, t) for time t = 0.2 .. . . . . . . . . . . . . . . . . 175
5.8 The graph of u(x, y, t) for time t = 0.4 .. . . . . . . . . . . . . . . . . 175
5.9 The graph of u (x, y, t) for time t = 0.8 .. . . . . . . . . . . . . . . . . 175
5.10 Heat conduction in a rectangular plate of length L and height H. . . 183
5.11 Assumption of u (x, y) = ul (x, y) + u2 (x, y) + u3 (x, y) + u4 (x, y) . . . 184
5.12 Heat conduction in a rectangular plate, u3 (x, y), of length L and
height H. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . 185 . .

6.1 The graph of the Fourier series representation off (x) . . . . . . . . . 193
6.2 The graph of the Fourier series representation of g(x). . . . . . . . . 193
6.3 The graph of 1(x) = x2. . . . . . . . . . . . . . . . . . . . . . . . . . 213
6.4 The graph of 1(x) = x2 and the partial sum Sl (x). . . . . . . . . . . 214
6.5 The graph of 1(x) = x2 and the partial sum S2 (x). . . . . . . . . . . 214
6.6 The graph of 1(x) = x2 and the partial sum S5 (x) . . . . . . . . . . . 214
6.7 The graph of 1(x) = x2 and the partial sum Slo(x). . . . . . . . . . 214
6.8 The graph of 1(x) = x2 and the partial sum S15 (x) . . . . . . . . . . 215
6.9 The graph of 1(x) = x2 and the partial sum Sioo (x). . . . . . . . . . 215
6.10 The odd periodic extension of 1(x) = 5 on [-L, L]. . . . . . . . . . . 216
6.11 The graph of the odd periodic extension of 1(x) = 5 and the first
partial sum S1 x . 216
6.12 The graph of the odd periodic extension of 1(x) = 5 and the first
partial sum S3 (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
xiv List of Figures

6.13 The graph of the odd periodic extension of 1(x) = 5 and the first
partial sum S7 (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
6.14 The graph of the odd periodic extension of 1(x) = 5 and the first
partial sum S15 (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
6.15 The graph of the odd periodic extension of 1(x) = 5 and the first
partial sum S99 (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
6.16 The graph of the odd periodic extension of 1(x) = 5 and the first
partial sum S2oi x . 217
6.17 The graph of 1(x) = x2 + x - 1. . . . . . . . . . . . . . . . 219 . . . . .

6.18 The Fourier sine series representation of 1(x) = x2 + x - 1. . . . . . 219

6.19 u(x,t) for 0<t<0.5. . . . . . . . . . . . . . . . . . . . . . . . . . . 219

7.1 u(x, t) for 0 < t < 20 .. . . . . . . . . . . . . . . . . . . . . . . . . . . 232

7.2 u(x,t) for0<t<0.05........................... 241
7.3 The graph of the solution for u(x, t) for 0 < t < 2. . . . . . . . . . . 251
7.4 The graph of u(x, t) for 0 < t < 2. . . . . . . . . . . . . . . . . . . . 268

2 Ho
8.1 o fl(z) = tonhzand f2(z) =
Graph of L 30 4
z H0-HL)
..............305
(

z (HL
8.2 Graph of fl(z) = tan z and f2(z) = Ho )
z2 HHL + L 0

8.3 Graph of f i (z) = tan z and 12(z) _

-z . . . . . . . . . . . . . . . . . 309

9.1 Graph of Airy's function. . . . . . . . . . . . . . . . . . . . . . . . . 331

9.2 The plot of the first five Legendre polynomials, Pn (x) . . . . . . . . . 337
9.3 The plot of the first five Legendre functions, Qn(x). . . . . . . . . . 338
9.4 Bessel's function of the first kind for n = 0, 1, 2, 3.. . . . . . . . . . . 347
9.5 Bessel's function of the second kind for n = 0, 1, 2, 3. . . . . . . . . . 348

10.1 Cylindrical coordinate system. . . . . . . . . . . . . . . . . . . . . . 352

10.2 Spherical coordinate system .. . . . . . . . . . . . . . . . . . . . . . . 352
10.3 Laplace's equation in a circular plate. . . . . . . . . . . . . . . . . . 353
10.4 The sphere of radius, r. . . . . . . . . . . . . . . . . . . . . . . . . . 357
10.5 An elastic beam. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
10.6 A small section (x, x + Lxx) of an elastic beam. . . . . . . . . . . . . 366
10.7 Simply supported beam . . . . . . . . . . . . . . . . . . . . . . . . . . 367
10.8 The cantilever beam: rigidly fixed on the left end and free on the
right end. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
10.9 Beam rigidly fixed at both ends. . . . . . . . . . . . . . . . . . . . . 371
10.lOPolar coordinate system.. . . . . . . . . . . . . . . . . . . . . . . . . 373
10.11A fixed point in three-dimensional space . . . . . . . . . . . . . . . . 388

11.1 Graph of x3 - 3x2 - 2x + 3 on [-2,2].................. 404

 List of Figures xv

11.2 Graph of Fourier series representation of x3 - 3x2 - 2x + 3 on [-2, 2]

where n = 150................................405
11.3 Graph of Fourier series function on [-oo, oo] where n = 150.. . . . . 405
11.4 Graph of 1(x)................................408

11.5 Graph of the Fourier integral of 1(x) ...................409

11.6 Graph of g (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
11.7 Graph of the Fourier integral of g (x) .. . . . . . . . . . . . . . . . . . 411
11.8 Graph of h(x). . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . 411
11.9 Graph of The Fourier integral of h(x). . . . . . . . . . . . . . . . . . 412
11.loGraph of 1(x) = 2e-x, where 0 < x. . . . . . . . . . . . . . . . . . . 416
11.11Graph of The Fourier cosine integral of 1(x) = 2e-x, where 0 < x. . 416
11.12Graph of The Fourier sine integral of 1(x) = 2e-x, where 0 < x. . . 417
11.13The Graph of erf (x) and erfc (x).................... 425
11.14The graph of f (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
11.15The Fourier transform, F(8) of f (x) . . . . . . . . . . . . . . . . . . . 440
11.16Graph of f (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
11.17The Fourier cosine transform, F(8) of f (x). . . . . . . . . . . . . . . 443
11.18The Fourier sine transform, F(8) of f (x). . . . . . . . . . . . . . . . 443
11.19Temperature distribution, u(x, t), for several values of t. . . . . . . . 455

B.1 u(x, t) in the rectangle 0 < x< L and 0 < t < T . . . . . . . . . . . . 467

C.1 Plot of ce 2t + 2 for c = -2, -1, 1, 2 .. . . . . . . . . . . . . . . . . . 475

-t
C.2 Plot of -e 2 + 2 .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
3
C.3 The graph of u(t) =
2
-et2

- -2 for 0 < t < 0.5...............477

1
Preface
This text is designed for a one-semester course in partial differential equations for
the undergraduate student of engineering, physics, applied mathematics, social sci-
ence, biology, and other sciences, for example, economics. The text covers the
method of separation of variables, Fourier series, classical problems of physics and
engineering, Sturm-Liouville eigenvalue problems, power series solutions of variable
coefficient ordinary differential equations, and transform methods. Wherever pos-
sible, mathematical topics are motivated by physical laws or problems. As such,
mathematical modeling of physical data and applications are stressed. Throughout
the text, completely worked examples/counterexamples are used to develop math-
ematical concepts. This reduces the potential for the student to see mathematics
as a set of magical steps, and it allows the student time to develop his/her own
methodology for solving problems based on comprehension of the mathematical
process. When a purely mathematical topic is developed, such as Fourier series, the
approach taken is constructive in methodology by building on material the student
has encountered in other courses. This provides the student a framework of con-
nections allowing easy comprehension of the material, and it assists the instructor
in developing the student's insight into higher mathematics.
Mathematical texts can be very intimidating to many students. Therefore, this
text is designed to be truly readable and "student friendly." Whenever the text or
parts of the text have been used in class, student end-of-course critiques indicated
that the readability and usability of the text is in the 99th percentile.
The text is motivated by applications, which help the student in his/her studies
in other areas of engineering and science. Many topics are introduced by using
a physical model as opposed to a purely theoretical approach. For example, the
section on the method of characteristics for first-order partial differential equations
with constant coefficients is introduced by the physical example of a surfer catching
a wave. Another example is the uniqueness of solution for the one-dimensional
wave equation, which is developed by first considering conservation of energy for
a vibrating string, a concept that most students should understand from either
their first physics or calculus courses. Theoretical topics, such as Fourier series, are
introduced by first discussing real vector spaces and the fact that different basis
can be developed for n-dimensional space by considering an n x n matrix with n
distinct eigenvalues and their corresponding eigenvectors.
The prerequisites for a student in a course using this text are the calculus se-

xvii
xviii

quence and elementary ordinary differential equations. An introduction to linear

algebra would be helpful, but not necessary.
I have included a review of ordinary differential equations in the appendices. I
have found this extremely valuable for many students. Also, for a more theoretical
approach an appendix with proofs of selected theorems is provided.

Course Outline
A possible outline for a one semester course is the following:
Chapters 1 through 8, which is the core material. This provides for the develop-
ment of the three classes of linear second-order partial differential equations, ellip-
tic, parabolic and hyperbolic and the three types of boundary conditions, Dirichlet,
Neumann, and Robin. Additionally, Chapters 1 through 8 gives a thorough discus-
sion of the separation of variables technique, coverage of the relevant theorems of
Fourier series and an introduction to the Sturm-Liouville boundary value problem.
Once Chapters 1 through 8 are covered, there are several options. For a complete
development of classical solution methods of second-order linear partial differen-
tial equations, I would suggest including Chapter 11, which develops the Fourier
and Laplace transforms. For a wider set of applications, I would suggest including
Chapters 9 and 10. It is also possible to chose selected topics from Chapters 9, 10,
and 11 for a broad discussion of applications and technique.
Although the text is not directly tied to a mathematical software package, such
as Mathematica, many of the exercises require the student to find partial sums of
Fourier series. Also, students are required in the exercises to graph both the Fourier
series representation of a function and a three-dimensional view of the solution of
a partial differential equation for various partial sums. Thus, students should be
familiar with some type of mathematical software package.
You may contact the author directly for Mathematica files and other comple-
mentary material related to the text by email at
[email protected]

Acknowledgments
In writing this text, I have drawn from my classroom experience, and I have
been influenced by many sources. I benefited from discussions with colleagues and
students. I thank everyone who has made comments and suggestions. Their interest
in improving the book encouraged me.
It is with great pleasure that I thank George Lobell, my Prentice Hall Editor, for
his support, valuable suggestions, and encouragement. In addition, I would like to
thank Bob Walters, my Production Editor for his assistance and advice and Adam
Lewenberg for his assistance and work on the graphics.
I want to take this opportunity to express my gratitude to the following math-
ematicians who reviewed earlier versions of this text and offered many valuable
 XiX

suggestions and insights, which certainly improved the content, readability, and ac-
curacy of the text:
Dr. Daniel C. Biles, Western Kentucky University;
Dr. Christine M. Guenther, Pacific University;
Dr. Jose Barrionuevo, University of South Alabama;
Dr. Mark Kon, Boston University;
Dr. William Margulies, California State University;
Dr. Mikhail Shvartsman, University of Saint Thomas;
Dr. Johnny Henderson, Auburn University.

I would like to express my thanks to the following individuals who gave valuable
support, advice, and suggestions during the development of the text:
Dr. Keith Bergeron, United States Air Force Academy;
Dr. Patricia Egleston, United States Air Force Academy;
Dr. Ali Haghighat, Nuclear Engineering Program University Park;
Colonel and Dr. Daniel W Litwhiler, United States Air Force Academy;
Dr. Glenn E. Sjoden United States Air Force Academy;
Dr. Dawn L. Stewart United State Air Force Academy.

Finally, I want to thank Dr. Michael Round, United States Air Force Academy,
for his time in reviewing several drafts of this text for grammatical accuracy.
I am particularly grateful to my wife, Jean, and to a friend of mine, Sonya, for
their support during this long project. It is with great pleasure that I dedicate this
book to them.

Michael K. Keane
Chapter 1

Introduction
The theory of ordinary differential equations (ODEs) was well established in the
early part of the 18th Century. However, the theory of partial differential equations
(PDEs) was still in its infancy and only studied by a few pioneers. Then, in 1747,
the famous mathematician Jean Le Rond D'Alembert, while studying the problem
of vibrating strings, developed the following form of the wave equation:

a2u(x, t) a2u(x, t)
ate - axe
He also published, in the Memoirs of the Berlin Academy, the solution to Equation
(1.1), i.e.,

u(x,t) = f(x+t)+g(x-t),
where f and g are arbitrary functions.
The field was advanced further by another famous mathematician, Leonhard
Euler. He developed the solution

u(x, t) = f (x + ct) + g(x - ct),

which is the solution to a more general wave equation
a2u(x, t)
= c 2 a2u(x,
ax2
t)
ate
Also in 1752, while studying hydrodynamics, Euler developed the three-dimensional
equation
52u(x, g, z)
ax
a2u(x, g, z)
ay
a2u(x, g, z)
5z2
-- 0 (1.3)

However, it wasn't until 1782 that P. S. Laplace formalized Equation (1.3) while
working in celestial mechanics. Equation (1.3) is now known as Laplace's equation.

1
2 Chapter 1: Introduction

Next, the mathematical theory of heat, which is characterized by the equation

Du(x, t)
at
-k D2u(x,
ax2
t)

was studied extensively by J. B. Fourier. In 1822, his paper on the solutions of the
heat equation, Theorie analytique de la chaleur set forth his idea that any function
y = 1(x) could be represented by a trigonometric series, now known as a Fourier
series. The Fourier series is one of the main topics of this text because it provides
a solution of second-order PDEs that are linear and homogeneous with linear and
homogeneous boundary conditions.
Careful inspection of Equations (1.2, 1.3, and 1.4) reveals something that they
have in common: they are all second-order PDEs.
Although the study of second order PDEs is two and a half centuries old, it is
far from obsolete. In fact, you can develop a second order PDE for almost anything
that occurs in nature or is constructed by man. They have been used traditionally
in physics, celestial mechanics, and meteorology. However, second order PDEs are
being developed and applied to problems in economics, mathematical physiology,
nuclear transport theory, aerospace industry, geophysics (particularly in the areas of
lava flow and plate tectonics), car design, electrical engineering, forestry, industrial
and community pollution, oceanography, and a host of other areas. With these
diverse fields researching second-order PDEs, one might get the impression that
there is no common thread to all the different fields. However, this impression is
quite wrong, second-order PDEs can be broken down into three major classes, known
as elliptic, hyperbolic, and parabolic. These names may seem strange; however, they
are rooted in the study of quadratic equations.
A form of the quadratic equation is axe + 2bx + c = 0, which may be solved by
using the quadratic formula

r= -b f a
b2 -ac

In Equation (1.5), we call b2 - ac the discriminant. If b2 - ac < 0, then the equation

is called elliptic. If b2 - ac > 0, then the equation is called hyperbolic. If b2 - ac = 0,
then the equation is called parabolic.
The quadratic form of an equation applies to a linear PDE in two variables.
Consider the general linear second-order PDE in two variables:

52u(x,y) +2b5u(x,Y) 52u(x,y)

axe 8x8y 8y2

______ +e Du(x,y) +fu(x,y)=g. (1.6)

y
Mathematicians call
52u(x,y) +2b5u(x,Y) 52u(x,y)
(1.7)
8x2 8x8y 8ya
 Chapter 1: Introduction 3

the principal part of Equation (1.6). From the principal part of the equation, you
can determine if a linear second-order PDE is elliptic, hyperbolic, or parabolic. For
instance, Laplace's equation, Equation (1.3), has a discriminant equal to -1. Thus,
it is representative of the elliptic class of PDEs. We first encounter this class of
PDE in Chapter 5. The wave equation, Equation (1.1), has a discriminant equal
to 1. Hence, it is representative of the hyperbolic class of PDEs. We encounter
this class of PDE in Chapter 3. Finally, the heat equation, Equation (1.4), has
a discriminant equal to 0. Therefore, it is representative of the parabolic class of
PDEs, which is introduced in Chapter 2.
Laplace's equation and the heat and wave equations are the primary represen-
tatives of their particular classes. However, not all linear second order PDEs are
readily classified. For example, Tricomi's equation,

a2u(x, y) a2u(x, y)
axe + a2
y
is elliptic for y> 0, parabolic for y = 0, and hyperbolic for y < 0. Thus, one must
be very careful when classifying PDEs.
As we study second-order PDEs, it is very important to remember the three
classes, since everything that we learn about the representative of each class of PDE
applies to the entire class. Now, we move on to the study of PDEs by investigating
the heat equation.

EXERCISES 1

1.1. For each of the following linear second order PDEs, identify in what regions
of the two-dimensional plane the equation is elliptic, hyperbolic, or parabolic.

(a) 2 a2u(x, y) + 4 a2u(x, Y) + 4 a2u(x, y) _ u(x, y) = 0.

8x2 8y2 8y8x

(b)
a2u(x,y) + 2 a2u(x,y) + a2u(x,y) + u ( x' ) 0.
axe y y
a2u(x , y> a2u(x , y) a2u(x , y) au(x , y> _
(c) Sin(xy)
ax 2 - 6 ayax + ay 2 + a y
- °'
82u(x,y) 82u(x,y) 82u(x,y) 8u(x,y) au(x,y)
(d) axe -cos(x)
ay8x + 8y2 + 5y 8x
+5u(x,y)= 0.
Chapter 2

The One-Dimensional Heat

Equation
2.1 INTRODUCTION
For over one million years, fire and its product, heat, have contributed significantly
to the rise of man. Thus, naturally, the study of heat transfer has become very
important.
At first, the physical properties of heat could only be guessed at through the
senses (e.g., the object is too hot or too cold). Only after Sir Isaac Newton and
Gottfried Leibniz1 developed calculus in the seventeenth century could we math-
ematically model the physical properties of heat transfer. Today we know three
types of heat transfer: conduction, convection, and radiation.
Although the mathematical formulation of all three types of heat transfer is
important, this chapter concentrates on mathematical formulas for heat conduction
in a one-dimensional rod. It begins with the derivation of the heat equation, then
branches out to discuss various boundary conditions and their physical meanings.
Next, we investigate uniqueness of solution for the heat equation. Finally, the
steady-state (equilibrium) temperature distribution solution for heat conduction in
a one-dimensional rod is discussed.

2.2 DERIVATION OF HEAT CONDUCTION

IN A ONE-DIMENSIONAL ROD
Why is heat transfer in a one-dimensional rod called heat conduction? To an-
swer this question, we define what we mean by heat transfer, investigate how it
occurs in a one-dimensional rod, and examine the three different types: convection,
radiation, and conduction.
'Sir Isaac Newton and Gottfried W. Leibniz were contemporary seventeenth century mathe-
maticians who independently developed calculus.

5
6 Chapter 2: One-Dimensional Heat Equation

Let's start by defining heat transfer. Simply put, heat transfer is the way heat is
distributed throughout a material (sometimes steady-state, sometimes time depen-
dent) when either an internal or external heat source is applied. Next, we discuss
how it occurs in a one-dimensional rod.
Physically, every one-dimensional rod has a cross-sectional area, A. Thus, the
one-dimensional rod is actually three-dimensional. See Figure (2.1). A rod is con-
sidered one-dimensional if the heat transfer occurs in a one directional manner,
down the length of the rod. This means the heat transfer across the cross-sectional
area is uniform. Thus, the heat moves in one direction through the length of the
rod. For example, consider a lit match and two different rods. The first rod has the
cross-sectional area equivalent to the cross-sectional area of one human hair; the
second rod has the cross-sectional area of a large steel I-beam. When the end of
the first rod is placed near the flame of the lit match, the end is uniformly heated,
and the heat moves uniformly down the length of the rod. When the end of the
second rod is placed near the flame of the lit match the flame doesn't even cover
the entire cross-sectional area of the rod. Therefore, the cross-sectional area of the
second rod is not uniformly heated, and the flow of heat through the rod is not in
one direction. Now, we'll investigate the three types of heat transfer, starting with
convection.
Convection of heat is the transfer of heat by the actual motion of the material
being heated. For example, if you hold your hand above the surface of boiling
water your hand feels the heat. Some of this heat is from the movement of the air
molecules, initially against the surface of the boiling water. As the air heats above
the boiling water, the air molecules rise, taking heat with them. As these molecules
rise, other cooler air molecules replace them. The rising molecules can even be seen
as steam. If you think of a rod as a thin piece of copper wire, do the molecules of
the wire move away from the wire as it is heated? No. Thus, heat transfer in a
one-dimensional rod is not convection. Having eliminated convection, we proceed
and consider radiation as a possible reason for heat transfer in a rod.

Figure 2.1: One-dimensional rod of length L.

 Section 2.2: Derivation of Heat Conduction 7

Radiation refers to the continual emission of energy from the surface of all bodies.
For example, hold your hand near the side of a radiator. (Note: do not put your
hand near the top of the radiator, because then heat convection also occurs). You
feel the heat. This is due to heat coming out of the radiator in a wave motion, in
the infrared electromagnetic spectrum similar to electromagnetic waves. In a one-
dimensional rod, heat transfer doesn't move in a wave-like motion down the length
of the rod. Therefore, just as in convection, it seems radiation of heat doesn't
account for heat transfer in the rod.
The last type of heat transfer is conduction. Conduction comes from collisions
of neighboring molecules, transferring heat by kinetic energy. Here, the molecules
move slightly, but do not actually move out of position like they do in convection;
nor do they move like an electromagnetic wave as in radiation. Now their movement
is more like that of tennis balls in a cylindrical container that is a little wider than a
tennis ball, and long enough to hold several tennis balls. Imagine the tennis balls are
molecules. When heat is applied, they can move slightly back and forth, bouncing
against each other and against the side of the can. But they remain essentially in
the same position. This is what happens when heat is applied to one end of a solid
rod. The molecules get excited, but they do not move away from the heat, as in
convection, nor do they move in a wave similar to an electromagnetic wave, as in
radiation; instead, they collide with one another and retain their relative position.
Using these brief explanations of heat transfer, we can see that heat transfer in
a one-dimensional rod is best described by conduction. Why is this so important?
Once we understand that heat transfer in a one-dimensional rod is conduction, we
may use three very important basic properties of conduction, called laws, which are
attributed to Joseph Fourier. These heat conduction laws are as follows:

Law 1. Heat flow is from points of higher temperature to points of lower tempera-
ture.

Law 2. Conduction can only take place in a body when different parts of the body,
including the ends, are at different temperatures.

Law 3. Heat flow changes depending on the material. Different materials, even at
the same temperature have different heat flows.

These three laws form the basis for the mathematical model of heat conduction
in a one-dimensional rod that we'll now develop.

2.2.1 Derivation of the Mathematical Model

One equation governing heat conduction in a one-dimensional nonuniform rod with-
out lateral insulation may be written in words as: rate of change with respect to
time of thermal energy in the rod equals the rate of flow of thermal energy due to
conduction, minus thermal energy loss via convection, plus thermal energy produced
8 Chapter 2: One-Dimensional Heat Equation

within the rod. This word equation is translated into mathematics as

c(x)p(x)
Du(x,t) =
(K0(X)
5u(xt))
at ax ax
-/3(x) [u(x, t) - v(x, t)} + Q(x, t). (2.1)

This is quite an intimidating equation, and its solutions aren't easy to find. Even-
tually, you will be able to solve Equation (2.1) for particular cases. However, we
will simplify Equation (2.1) to develop a careful approach to its solution. First,
we must identify the terms, then make reasonable assumptions about them. This
allows us to derive a new equation. Mathematicians use this method all the time.
Thus, viewing Equation (2.1), we see that there are three double-variable functions
u(x, t), Q(x, t), and v(x, t) and four single variable functions c(x), p(x), Ko(x), and
,3(x).
The first double-variable function, dependent on distance and time, is u (x, t) :
a continuous function that describes the temperature distribution in the rod. This
function depends on distance because the temperature, u(x, t), usually is not the
same at every point x along the length of the rod. It depends on time because
the temperature, u(x, t), is expected to change with time due to some external or
internal source.
The second double-variable function, dependent on distance and time, is Q (x, t) :
a function representing internal heat source, called "source" for short. It describes
the heat energy generated inside the rod. This is an interesting function because
not all materials that can be used to make a rod generate heat; some materials
will only generate heat after an external heat source is applied. For example, a rod
made of steel does not generate heat even when an external heat source is applied,
but a rod composed of a radioactive material generates its own heat, regardless
of external source. Generally, the source function is either known or determined
experimentally.
The last double-variable function is v(x, t), the temperature of the surrounding
medium. It depends on position and time, since the rod could connect different
temperature areas that may change over time. This function plays a big part in
Equation (2.1) if there is no lateral insulation.
Moving into the single-variable functions, we have c(x), the specific heat of the
rod. Specific heat is the heat energy required to raise the temperature of one unit
of material mass one degree. Specific heat is a spatial function, since the material
composition of the rod depends on position. If the rod is composed of only one
material, or the mixture of materials is uniform, the specific heat, c(x), becomes
the constant, c.
The next single-variable function is p(x). It is the mass density of the rod. It
also depends on position in the rod because the rod could be composed of several
materials, each having a different mass density. Mass density is usually measured
as mass per unit volume. As in specific heat, if the rod is composed of one material,
or the mixture of materials is uniform, the mass density, p(x), becomes a constant,
p
 Section 2.2: Derivation of Heat Conduction 9

The third single-variable function is Ko (x), the thermal conductivity of the

material. It is usually determined experimentally. Each different material, or mix-
ture of materials, has a different thermal conductivity. For instance, the thermal
conductivity of copper at 273 Kelvin is 390 watts per meter per degree Kelvin;
whereas, the thermal conductivity of asbestos at 273 Kelvin is 0.15 watts per meter
per degree Kelvin. Thus, thermal conductivity depends on the material composi-
tion of the rod, and it may also depend on the temperature. However, if the range
of temperature is not allowed to have large swings (the case we will consider), then
thermal conductivity will primarily depend on the material at each position, x, like
specific heat, c(x), and mass density, p(x). Hence, like specific heat and mass den-
sity, if the rod is composed of one material or has a uniform mixture of various
materials, the thermal conductivity, Ko(x), becomes a constant Ko.
The last single-variable function in Equation (2.1) is ,3(x). It appears in the heat-
conduction equation multiplied by [u(x, t) - v(x, t)], which is a convection term.
Hence, ,fi(x) is named the convection proportionality function. In many cases, ,fi(x)
is actually a constant. However, in our equation we treat it as a function.
In the table below the terms of Equation (2.1) are listed with a short description:

Function Description
u(x, t) temperature distribution of the rod

Q(x, t) internal heat source

v(x, t) temperature of surrounding medium

c(x) specific heat of material

p(x) mass density of material

Ko(x) thermal conductivity of material

,fi(x) convection proportionality function

Now, let's make some intelligent assumptions to simplify Equation (2.1).

First, let's make the assumption that the rod is made of a uniform material,
then c(x), p(x), and Ko (x) all become constants c, p, and Ko. When we impose a
second assumption, the rod has perfect lateral insulation, which means convection
can't take place. This assumption forces ,fi(x) to equal 0 in Equation (2.1). Using
these two assumptions, we change Equation (2.1) to

cp
8u(x, t) =KO 8zu(x, t) + Q(x t).
(2.2)
at axe
Equation (2.2) is the mathematical model for heat conduction in a one-dimensional
uniform rod with perfect lateral insulation.
10 Chapter 2: One-Dimensional Heat Equation

Now, we are going to derive Equation (2.2). Consider the physical model of a
one-dimensional uniform rod with perfect lateral insulation. Apply the conservation
of thermal energy law, which basically states that the rate of change with respect
to time of thermal energy in the rod must equal the rate of flow of thermal energy
across the boundaries, plus the thermal energy produced within the rod. However,
when conservation of thermal energy is applied, the problem must be approached
from a calculus point of view. In other words, consider only a thin slice first and
then expand it mathematically to the full rod.
Shown in Figure (2.2) is a one-dimensional rod oriented in the positive x direc-
tion. That is, the x-axis runs down the center of the rod and the rod has length L,
sothat 0<x<L.

Figure 2.2: One-dimensional rod of length L.

Figure 2.3: Arbitrary slice [a, b] of one-dimensional rod.

 Section 2.2: Derivation of Heat Conduction 11

Now, applying the conservation of thermal energy to any arbitrary slice [a, b] of
our rod (see Figure (2.3)) we can claim that

Rate of change with respect to time of total heat inside the slice
= Rate of flow of heat across the boundaries of the slice
+ Total heat generated inside the slice.
By using the previously defined functions, variables, axioms, and some calculus,
the foregoing equation can be expressed mathematically. (Note: In the following
equations, A is the cross-sectional area of the rod).
First,
b

Total heat inside the slice of rod [a, b] = c p A u(x, t) dx.

a
Next, using differential calculus on this equation, we may express the left side of
Equation (2.3) as

d [fbA(t)d]
We will keep the constants c, p, and A inside the integral for the moment. This
makes the calculations easier.
The second term in Equation (2.3), rate of flow of heat across the boundaries of
the slice [a, b], is somewhat more complicated. Since there is a rate, we know from
differential calculus that a derivative is somehow involved. This derivative must
relate temperature distribution at the boundaries of the slice to heat flow at these
same boundaries. The correct solution lies in understanding heat flow as described
by the three heat Conduction laws:
Law one says heat flow is from points of higher temperature to points of lower
temperature. This law describes how heat flows.
Law two says conduction can only take place in a body when different parts of
the body, including the ends, are at different temperatures. Therefore, Law two
describes when heat flow takes place.
Law three says heat flow changes depending on the material. Different materials,
even at the same temperature, have different heat flows. This law indicates heat
flow depends on material. Hence, we must introduce a function describing the
thermal conductivity of the material. At present our rod is uniform, so the function
describing thermal conductivity becomes a constant.
Now, suppose the temperature in the rod is hotter to the left of the slice [a, b]
than it is to the right of the slice. See Figure (2.4).
By Law 1, this means that heat flow is in the positive x direction. The equation
for heat flow across a cross-sectional area A is

heat flow = _K o
u(x, t) A
(2.5)
ax
12 Chapter 2: One-Dimensional Heat Equation

a Heat Flow b

Figure 2.4: Temperature distribution of slice [a, b] .

The constant Ko is the thermal conductivity of the material. The partial derivative,
au(x, t)
, is negative since it indicates the slope of the derivative in the x direction;
ax
therefore, the negative sign in front of Ko must be introduced to show that heat
flow is to the right, the positive x direction. The relationship between the three
heat conduction laws to heat flow, expressed in Equation (2.5), is Fourier's2 law
of heat conduction.
Using Equation (2.5), the slice [a, b] gains thermal energy at the boundary a and
loses thermal energy at the boundary b. Therefore, the second term in Equation
(2.3), rate of flow of heat across the boundaries of the slice [a, b], is mathematically
modeled as

_xoa [au(a, t) _
8x
au(b, t)(2.s)
8x

Again using calculus, we can model the third and final term in Equation (2.3),
total heat generated inside the slice [a, b] per unit time t as
fb

AQ(x, t) dx. (2.7)

Therefore, utilizing Equations (2.4, 2.6, and 2.7), we can rewrite Equation (2.3) as

t F 6
cpAu(x,t)dx = -KoA raal st) _ aat)l
Lx
J

fb
+ AQ(x, t) dx. (2.8)
Ja
2J. B. Fourier (1768-1830), a very influential mathematician, first postulated the heat conduc-
tion laws, and developed what is known as Fourier's law of heat conduction.
 Section 2.2: Derivation of Heat Conduction 13

The constant cross-sectional area, A, occurs in every term of Equation (2.8).

Thus, it is easily canceled, yielding

b p uxt
(s ) dx =- x0 a(x,t)
_aax,t)]

Ia
b

+f a
Q(x,t)dx.

Equation (2.9) can also be simplified by using calculus on

t d
Ia
6
c p u (xt) dx

and

rau(a,t) _ au(b,t)l
-K0
L 8x 8x J

Applying the fundamental theorem of integral calculus to the second term of

Equation (2.9), namely,
rau(a, t) _ au(b, t) 1
-K0
L 8x 8x J

we get

au(a, t) au(b, t) b a2u(x, t)

-K0 K° dx,
ax ax a axe
a2u(x, t)
provided that we assume that is continuous. This assumption is really
axe
a2u(x, t)
not out of line as the function axe is the rate of change of the heat flow. We
expect it to be continuous, since any kind of discontinuity implies a sharp change
in heat flow, violating the three heat conduction laws.
For the first term of Equation (2.9), a new calculus formula is required. It is
called Leibniz's3 formula. The theorem is stated as follows and a proof may be
found in Appendix B.
a f (x, t)
Theorem 1. Suppose f (x, t) and the partial derivative are continuous in
at
some region of the xt-plane where a < x < b and c < t < d, then

[fbf(x,t) t)
dx.
dt dx Ia b of at,

3 G. W. Leibniz (1646-1716), a contemporary of Sir Isaac Newton, developed the modern no-
tation for calculus in 1676.
14 Chapter 2: One-Dimensional Heat Equation

The first term of Equation (2.9) satisfies Leibniz's formula because u(x, t) is
au (x, t)
continuous and is expected to be continuous. Therefore,
at
dF
dt if
La
b

cp u(x, t)dx =
]
f a
6

cp
au(x,t)
dx.

Using the two previous simplifications, we can rewrite equation (2.9) as

/'6 p 8u(x, t)
c dx = b Ko a2 a( 2' t) dx + J b Q(x, t) dx. (2.10)
J
f
For equation (2.10) to be true, we must have

b au(x, t) a2u(x, t)
cp - K° - Q(x, t) dx = 0. (2.11)
a at axe
Since the choice of a and b was arbitrary, Equation (2.11) must be true for all choices
of a and b within the length of the rod. Using proof by contradiction, we can show
that Equation (2.11) is true only if the integrand is zero. Thus, we can omit the
integration and consider only the integrand that is,
au(x, t)
at
- K0 a2u(x,
axe
t)
- Q(x, t) = 0

or

au(x, t) a2u(x, t)
p at =K0
K° axe
+Q(x,t). (2.12)

Equation (2.12) is identical to Equation (2.2). Thus, using the conservation

of thermal energy law and calculus, a one-dimensional uniform rod, with perfect
lateral insulation, has been correctly modeled mathematically. In Equation (2.12),
the source heat energy Q(x, t) is usually given. Therefore, the only unknown is
u(x, t). If Q(x, t) = 0, then Equation (2.12) becomes
au(x, t) a2u(x, t)
(2.13)
at = ' axe '

where k = K° and is called the thermal diffusivitY. Appendix E contains a table of

cp
the thermal diffusivity of common materials, which will be used in this text.

2.2.2 Initial Temperature

Whenever an experiment is conducted, the initial state for that experiment is de-
termined beforehand. In the case of heat conduction in a one-dimensional rod, we
 Section 2.2: Derivation of Heat Conduction 15

have the initial temperature distribution of the rod, known as the initial condition
(IC) of the rod. Since we are talking about temperature distribution, the IC is a
function of x. It is the condition that must exist before the experiment starts (at
the start of time for the experiment). Therefore, at t = 0, we have u(x, 0) = f (x),
0 < x < L. Remember, even if the IC is a constant c throughout the rod, the IC
may still be thought of as a function of x.
The following equation, Equation (2.14), describes heat conduction in a one-
dimensional rod with the constraints of perfect thermal lateral insulation, constant
mass density, constant thermal conductivity, constant specific heat, no heat source,
and an IC:
au(x, t) _ t)
at axe
(2.14)

u(x, 0) = 1(x), 0 < x < L.

Also, remember that Equation (2.14) is the primary representative of the parabolic
class of linear second-order PDEs.

EXERCISES 2.2

2.2.1. Suppose you are given a thin slice of a uniform one-dimensional rod with
perfect lateral insulation. State the word equation for conservation of thermal
energy.
2.2.2. Given specific heat, c(x); mass density, p(x); thermal conductivity, Ko (x);
and temperature, u(x, t), state Fourier's law of heat conduction.
2.2.3. Briefly explain the basic idea of Fourier's law of heat conduction.
2.2.4. Suppose we have heat conduction in a rod with perfect lateral insulation, no
internal heat sources, and specific heat, mass density, and thermal conduc-
tivity as functions of x, that is, c(x), p(x), and Ko(x). Starting with the
conservation of thermal energy law, derive a new form of the heat conduction
equation.
2.2.5. Suppose we have heat conduction in a uniform rod with an internal source of
heat energy, but there is no lateral insulation. Thus, the heat flows freely in
and out across the lateral boundary at a rate proportional to the difference
between the temperature, u(x, t), in the rod and the surrounding medium,
/3(x, t). Starting with the conservation of thermal energy law, derive a new
form of the heat conduction equation. Hint: This problem requires a new
formulation of the conservation of thermal energy law.
2.2.6. Suppose you have a perfect laterally insulated uniform rod, but instead of a
heat source it has a heat sink (a sink is where heat is absorbed). Starting with
the conservation of thermal energy law, derive the heat conduction equation.
 16 Chapter 2: One-Dimensional Heat Equation

2.2.7. Suppose you have a perfect laterally-insulated uniform rod, but the cross-
sectional area is a function of x (that is, A(x)). Starting with the conservation
of thermal energy law, derive a new form of the heat conduction equation.

2.2.8. Show that u (x, t) = e - t sin x is a solution of

au(x,t) a2u(x,t)
at axe

subject to

u(x, 0) = sin x.

2.2.9. Show that u(x, t) = e-3t cos 2x is a solution of

au(x, t) 3 a2u(x, t)
at 4 ax 2

subject to

u(x,O) = cos2x.
5e-3x)
2.2.10. Show that u(x t) = 1 + e2t(4e3x + is a solution of

au(x, t) 2 a2u(x, t)
at -9 ax 2

subject to

4e6x + 5
u(x, 0) = 1 +
e3x

As time goes to infinity, what do you expect to happen to the one-dimensional

rod?

2.2.11. Solve a2u( _- xy.

y

Exercises (2.2.12-2.2.14) involve first-order ordinary differential equations (ODEs).

A review of this material may be found in Appendix C.

2.2.12. Solve u'(t) = atu(t) where a E R. Graph several members of the family of
solution curves.

2.2.13. Solve u' (t) = atu(t) + sin t where a E R. Graph several members of the family
of solution curves.
 Section 2.3: Boundary Conditions 17

2.2.14. Solve the following first-order ODEs

(1) g'(t) + 6g(t) = t, g(0) = 1.

(2) y' = 18y + xy, y(1) = 8.

(3) h'(x) - h(x) = sinx, h(0) = 7r.

(4) 3g'(z) = cosy - 9g(z), g U = 0.

2

2.3 BOUNDARY CONDITIONS

FOR A ONE-DIMENSIONAL ROD
In Section 2.2, we derived Equation (2.12) and Equation (2.13), commonly referred
to as the heat equations, for a one-dimensional rod. Also, we discussed the initial
temperature distribution in the rod, then modeled it as an initial condition. How-
ever, when you model physical phenomena, it is good practice to model as much
as possible; to satisfy this demand, we must discuss the ends of the rod. The ends
of the rod are usually called the boundaries; when their state is modeled, they
are called boundary conditions (BCs). In this section, we cover the mathematical
models of various types of BCs.

2.3.1 Boundary Conditions of the First Kind

A name for BCs of the first kind is Dirichlet4 conditions. When the boundaries
of a one-dimensional rod take on the temperature of surrounding mediums, the
boundaries are said to have specified temperatures. For example, if the rod's
boundary at x = 0 is held in a hot water bath where the temperature changes with
time, then that boundary may be mathematically modeled as

u(0, t) = gl (t). (2.15)

Also, suppose that the rod's boundary at x = L is held in a different bath of hot
water where the temperature changes in time, but differently than at x = 0. Then
that boundary may be mathematically modeled as

u(L,t) = g2(t).
(Note: gl (t) and g2 (t) describe the temperature of the different baths as they change
with time. Some texts use ubl (t) and ub2 (t) to describe the temperature in different
baths).
Sometimes, the temperatures of baths do not depend on time and are there-
fore constants, Tl and T2. If Tl and T2 equal zero, then we have u(0, t) = 0 and
4 Peter Gustav Lejune Dirichlet (1805-1859) was a Prussian born mathematician who was highly
influenced by Fourier in the early 19th century
18 Chapter 2: One-Dimensional Heat Equation

u(L, t) = 0. In this case, the boundaries being described are in contact with zero-
degree baths, and they are known as homogeneous boundary conditions of the first
kind. Homogeneous boundary conditions are required to develop the separation of
variables solution technique. This technique is one of the major topics covered
in this book.

2.3.2 Boundary Conditions of the Second Kind

BCs of the second kind are also known as Neumann5 conditions, and they de-
scribe rate of heat flow across the boundaries. Here, ends of the rod are covered
by insulation material. For example, consider the rod's boundary at x = L where
there is insulation material between the end of the rod and the surrounding medium.
Now, suppose heat is flowing outward from the end of the rod at x = L. That is,

L Heat Flow L+S

Figure 2.5: Heat flow from end x = L.

heat flow is from left to right in the positive x direction. This means the rod is
hotter than the surrounding medium. See Figure (2.5).
From Fourier's law of heat conduction, the equation that models this example
is
au(L, t) _
-KoL
f ) (t), (2.16)

where fi(t) is given. (Note: Ko(L) is the thermal conductivity of the insulation.
Also note, Equation (2.16) describes the value of the derivative at the point L. Thus,
we cannot simply integrate to make it a specified temperature. A very special case
of Equation (2.16) occurs when /(t) = 0. Then Equation (2.16) can be written as
au(L, t) _ 0
ax '

meaning we have the case of perfect insulation. (This may seem impossible, but it
is a standard case to model mathematically).
5 F ranz Neumann, (1798-1895) was a mathematical physicist who worked in Konigsberg.
 Section 2.3: Boundary Conditions 19

2.3.3 Boundary Conditions of the Third Kind

A name given to BCs of the third kind is Robin's6 conditions. These conditions
are based on Newton's law of cooling, which states heat loss by convection from
one body to another is proportional to the temperature difference between the two
bodies. How can heat be lost by convection when we are discussing conduction?
Consider the left boundary of a rod of length L. That is the boundary at x = 0.
Suppose the rod is hotter than the surrounding medium -a liquid, which is being
stirred rapidly so that it maintains a constant temperature throughout, which is
changing with time t. Let the temperature of the medium be gl (t). However, the
liquid adjacent to the end of the rod is slightly hotter than the liquid just a little
farther away. Hence, it heats up slightly, moving away from the end of the rod in a
convective manner. Once away from the end of the rod, the slightly hotter liquid is
rapidly mixed in with the rest of the medium. The mathematical model for this is
au(o, t)
-Ko(0) = -h[u(0,t) -gi(t)J. (2.17)
ax
Here h is a constant of proportionality and is assumed to be positive. It is called
the coefficient of convective heat transfer. The minus signs in Equation (2.17)
describe the physical situation of the rod being hotter than the liquid medium.
Thus, heat is leaving the rod on the left side, which is in the negative direction. See
Figure (2.6).

O-3 Heat Flow 0

Figure 2.6: Heat flow at end of rod x = 0.

Robin's conditions can be thought of as the generalization of Dirichlet and Neu-

mann boundary conditions. For example, if h -+ 0 in Equation (2.17), then Equa-
tion (2.17) mathematically models Neumann's Condition of perfect insulation. If
h -* oo in Equation (2.17), then Equation (2.17) mathematically models Dirichlet's
Condition of a time-dependent specified temperature.
This completes our discussion of the three types of BCs. Using what we learned
6 Victor G. Robin (1855-1897) was a French mathematician
20 Chapter 2: One-Dimensional Heat Equation

in Sections (2.2.1 and 2.3), we may now mathematically model a physical situation
involving heat conduction. This is done in the following example.
EXAMPLE 2.1. Mathematically model heat conduction in a one-dimensional
uniform rod of length L with no internal heat source, thermal diffusivity of k, perfect
lateral insulation, and initial condition as a function of x. Also, the left boundary,
at x = 0, is in direct contact with a zero-degree bath, and the left boundary, at
x = L, is perfectly insulated.

Solution: The equation is

au(x, t) _ t)
at axe
subject to the IC
u(x,0) = f(x)

and the BCs

u(0, t) = o,

8u(L,t) _ o
8x

Notice, both the initial condition and the boundary conditions must be given for
a complete physical description. When both are given, we can generate an accurate
mathematical model of the physical description.
Knowing how to mathematically model a physical situation involving heat con-
duction is important. However, just as important is the capability of giving a
physical description from a mathematical model. This is done in the next example.
EXAMPLE 2.2. Given that
au(x, t) _ a2u(x, t)
cQ(x, t),
at - ax2
subject to the BCs
8u(0, t)
- 6 watts,
8x
(L)
8u(L,tt)
=0
8x
ax
and IC

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

write a short paragraph developing a possible physical model.

 Section 2.3: Boundary Conditions 21

Solution: A possible physical model for the equations in this example is heat
conduction in a perfect laterally insulated uniform one-dimensional rod of length L,
with thermal diffusivity of k. Also, the source term is a sink, and it is being modi-
fied by a proportionality constant, a. This means heat energy is being withdrawn
from the rod. The left boundary, at x = 0, has a rate of heat flow of a constant
six watts. This may be due to imperfect insulation. The right boundary, at x = L,
is perfectly insulated. Finally, the initial condition is a function of x. Please note,
the initial condition does not have to match the boundary conditions at x = 0 and
x = L.

EXERCISES 2.3

2.3.1. Write a short paragraph (five sentences or less) that describes the physical
problem modeled by the equations
au(x, t) a2u(x, t)

subject to IC

u(x, 0) = -x,
and the BCs
8u(0, t)
_ 0 and u(L, t) = a(t).
8x

2.3.2. Write a short paragraph (five sentences or less) that describes the physical
problem modeled by the equations
au(x, t) _ 1 a2u(x, t) _t
at 4 ax2 - 15[ u(x, t -
) (8x - 2e + 2 )],

subject to IC

u(x, 0) = 8x,

and the BCs

u(0, t) = 0 and u(7r, t) = 167r cos .

3

2.3.3. Suppose a nonuniform metal rod of length it, with perfect lateral insulation,
has an initial temperature distribution of sin x. Initially, one end of the rod
is fixed at a temperature of 0°C, while the rest of the rod is placed in liquid
nitrogen. What would be a possible mathematical model that describes this
problem? Explain your answer.
 22 Chapter 2: One-Dimensional Heat Equation

2.3.4. Consider a one-dimensional rod of length L. Assume that heat energy is flow-
ing into the rod at x = 0 proportional to the temperature difference between
the end temperature of the rod and the known external temperature. Develop
the mathematical model for this condition. Briefly justify your answer.
2.3.5. Consider a one-dimensional rod of length L. Assume heat energy is flowing
into the rod at x = L proportional to the temperature difference between the
end temperature of the rod and the known external temperature. Develop the
mathematical model for this condition. Briefly justify your answer.
2.3.6. Given Equation (2.17), show that if h - * 0, then we get Neumann's condition
of perfect insulation. If h - * oo, then we get Dirichlet's condition.
2.3.7. Consider a one-dimensional rod of length H. Assume the rod has no lateral
insulation, is nonuniform, and an internal heat source doesn't exist. Also,
assume at the boundary x = 0, the rod is held at a constant temperature of
15°C, and at the boundary x = H, the rod is imperfectly insulated, allowing a
heat energy flow of a constant -8 watts. Develop the mathematical model that
includes a possible initial temperature-distribution equation. Briefly explain
your choice of initial temperature-distribution equation.
2.3.8. Show that u(x, t) = e-2t sin x is a solution of
au(x, t) a2u(x, t)
at = 2 axe '
subject to the BCs
u(0, t) = 0 and u(27r, t) = 0,

and IC

u(x, 0) = sin x.

2.3.9. Show that u(x, t) = e-5t cos 3x is a solution of

au(x, t) _ 5 a2u(x, t)
at 9 axe '
subject to the BCs
au (0, t) _ au (3 , t
ax - 0 and ax
and IC

u(x,0) = cos3x.

a2u(x, y) _
2.3.10. Solve
axay xy.
 Section 2.4: The Maximum Principle 23

Exercises (2.3.11 through 2.3.15) involve first-order ODES. A review of this material
may be found in Appendix C.

2.3.11. Carbon 14 obeys the law of radioactive decay. Determine k if the half-life of
carbon 14 is 5568 years. Next, suppose the initial amount of carbon 14 is 9
gm. Determine the amount of carbon 14 left after 256 years.,

2.3.12. Suppose a mothball loses volume by sublimation at a rate proportional to its

surface area. Write an ODE to describe this phenomenon, making sure to
define each term in your equation. Now, suppose the mothball's initial radius
is 4 cm. If it takes 30 days for the radius to decrease 2 cm, how long will it
take for the radius to decrease to 1 mm?

2.3.13. Suppose in a population of yeast cells, growing exponentially, the initial pop-
ulation of cells is 1200. Fifteen minutes later it is 1700. Find the growth rate
for the population.

2.3.14. A cup of coffee is initially at boiling point, 100°C. The temperature of the
room is 20°C. Find the temperature of the coffee as a function of time. (Hint:
use Newton's law of cooling.)

2.3.15. In a furnace, the temperature of the inner wall of an area 2m2 is 450°C. The
temperature of the outer wall is 80° C. There is a 0.5m of brick insulation
(thermal conductivity of brick is 0.38) between the walls. How much heat
escapes in three minutes? (Hint: Assume steady state has been reached
across the walls and remember Fourier's law of heat conduction:
du(x) A
heat flow = -K
° dx )

2.4 THE MAXIMUM PRINCIPLE AND UNIQUENESS

So far, we can develop the heat equation for a one-dimensional rod, and we realize
that we must have ICs and BCs to fully state the problem. However, we have no idea
if we have enough initial data to have a solution for any time, t, or if the solution
is unique, that is, is the problem well-posed (see Definition (2)). In this section,
we address these issues and answer them. Since this text is primarily application
oriented, we do not go into great depth.
Definition 2. (Well-posed) A problem involving a partial differential equation is
said to be well-posed if there exists a unique solution, and the solution depends
continuously on the data of the problem.

Suppose we are given

au(x, t) _ ka2u(x, t)
(2.1s)
at axe
24 Chapter 2: One-Dimensional Heat Equation

subject to Dirichlet BCs

u(0,t) = g(t)
(2.19)
,u(L(L,t) = h(t
n(t )

and to IC
u(x, 0) = 1(x). (2.20)

We want to determine if a unique solution exists. Our intuition tells us that there
is a unique solution. However, proving it is a slightly different matter. We start by
stating a maximum-minimum theorem for diffusion of heat in a one-dimensional
rod. Then we state and prove an immediate result of the maximum-minimum
theorem, which shows that the problem given in Equations (2.18, 2.19, and 2.20)
has an unique solution. Finally, we state another result of the maximum-minimum
theorem, that of continuous dependence of the solution on the initial data. Thus,
the problem is well-posed.

0 L

Figure 2.7: The domain of u(x, t) in the rectangle 0 < x < L and 0 < t < T.

Figure (2.7) is a two-dimensional diagram of heat flow in a one-dimensional

uniform rod with perfect insulation and Dirichlet BCs. The vertical sides of the
rectangle 0 < x < L and 0 < t < T refer to the BCs of the one-dimensional rod,
which may be changing with respect to some function of time, t, throughout the
experiment. The lower base of the rectangle indicates the IC of the one-dimensional
rod at the start of the experiment.
Theorem 3. (Maximum-Minimum theorem) Let T be an element of the real num-
bers such that T> 0. Suppose the function u(x, t) is continuous in a closed rectan-
gle, R, given by
0<x<L and0<t<T,
as shown in Figure (2.7), and satisfies the heat equation given in Equation (2.18)
in the interior of the rectangle. Then, u(x, t) attains its maximum or minimum on
 Section 2.4: The Maximum Principle 25

the base of the rectangle t = 0 or on the vertical sides of the rectangle x = 0 or

x=L.
A proof of the maximum-minimum theorem may be found in Appendix B.
However, a common sense approach may give us some understanding of the theorem.
Fourier's heat conduction axioms tell us that heat flows from high to low. In
the absence of a source or sink in the rod, the heat must flow from any hot spots
to cooler spots within the rod or out of the ends. If the rod is being burned at one
end and is in a freezer at the other end, the heat will flow from the burning end
to the end in the freezer. However, the end that is being burned will still always
be hotter than any other point of the rod and the end in the freezer will always be
cooler than any other point on the rod.
We now proceed to Corollary (4), which tells us that if a solution exists, then it
is unique.
Corollary 4. (Uniqueness) There is at most one solution to the problem given in
Equations (2.18, 2.19, and 2.20).
Proof. Suppose u1 (x, t) and u2 (x, t) are both solutions to the problem given in
Equations (2.18, 2.19, and 2.20). Let w(x, t) = u1(x, t) - u2 (x, t). Then, w(x, t)
satisfies Equation (2.18) with w(x, 0) = 0, and w(0, t) = w(L, t) = 0. Letting T be
any number greater than zero, we have w(x, t) = 0 throughout the entire rectangle
given in the maximum-minimum theorem. Thus, since T is arbitrary, we have
w(x, t) = u1(x, t) - u2 (x, t) = 0, which implies u1(x, t) = u2 (x, t). 0
Using a method called the energy integral method, it can be shown that
au(x, t) _ a2u(x, t)
at - k axe
subject to Neumann BCs

au(L,t)
(L
=ht
ax ()'
and to IC

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

has a unique solution. However, that method is beyond the scope of this text.
Another result of the maximum-minimum theorem is continuous dependence of
the solution on the data (both the boundary conditions and the initial conditions).
Corollary 5. (Continuous dependence of the solution on the initial data) The so-
lution of the problem given in Equations (2.18, 2.19, and 2.20) depends continu-
ously on the initial and boundary conditions in the following way: let u1 (x, t) and
26 Chapter 2: One-Dimensional Heat Equation

u2 (x, t) be solutions to the problem with initial data gl (t), hl (t), and Ii (x) and g2 (t),
h2 (t), and f 2 (x), respectively. Let T and s be any positive real numbers. If

Om I- .f2(x)I C E+
om I91 (t) - 9a (t) I e, and

max I hi(t) - ha(t)) s,

0<t<T
then
max
0<x<L, 0<t<T
I E.
Therefore, the maximum-minimum theorem and its two immediate corollaries
provide for the well-posedness of the problem given in Equations (2.18, 2.19, and
2.20).
Equations (2.18, 2.19, and 2.20) describe the heat equation, sometimes called
the diffusion equation, in a one-dimensional rod. From the preceeding discussion we
know that the heat equation is well-posed. At this time, it is convenient to state the
heat equation for multiple spatial dimensions for the Cartesian coordinate and in
different coordinate systems. The heat equation in two and three spatial dimensions
for Cartesian coordinates are given by Equations (2.21 and 2.22) respectively.
Du(x, y, t) (D2u(x,y,t) D2u(x, y, t)
at - axe + a2
y
(2.21)

Du(x, y, z, t) (52u(xYzt) a2u(x, y, z, t) a2u(x, y, z, t)

at - Dx2 + D y2 + uz2
(2.22)

The heat equation in polar, cylindrical, and spherical coordinate systems are given
by Equations (2.23, 2.24, and 2.25) respectively.
au (r,O,t) _ k1 1 8 rr8u(r, B, t) 1 D2u(r, e, t)
ae2 (2.23)
at r 8r L 8r +r2

au(r,e,z,t) _ 1 a rau(r,e,z,t) 1 a2u(r,e,z,t)

at r ur ar + r2 ae2

a2u(r, e, z, t)
(2.24)
uz2

au(r, e, , t) _ -k(----- 1 a2u

e
at r2 Dr ar + r2 Sine Sln De2

1 a2u
(2.25)
r2 sin2 0
 Section 2.4: The Maximum Principle 27

The right side of Equations (2.21, 2.22, 2.23, 2.24, and 2.25) is called the Laplacian,
and it is given in many equations as V2u where the arguments of u are understood
from the nature of the problem.
We will state and use Equations (2.21, 2.22, 2.23, 2.24, and 2.25) in later chapters
in the text. However, you should know how to derive each of the equations from
the Cartesian counterpart. For instance, the heat equation in polar coordinates,
Equation (2.23), may be derived from the Cartesian two spatial dimension heat
equation, Equation (2.21). A demonstration of the derivation is given in Example
(2.3

EXAMPLE 2.3. Derive the polar form,

au(r, 8, t)
at
_
k -r1ara [r0t)]
au(r,
ar +
1 a2u(r, 8, t)
r2 a82
,

of the heat equation from the two spatial dimension Cartesian form,

au(x, y, t) (D2u(x,y,t) a2u(x, y, t)

at = ax2 + ay2

Solution: Remembering that x = r cos 8 and y = r sin 8 we have r = x2 + y

and 8 = tan-1 (v). Therefore,
x

ar _ x _ a2r _ y2 _ sine 8
ax - cos 8, which implies ax2
/x2 + y2 (x2 + y2) 3 r

Similarly,

Dr a2 r cost 6
ay
= sin 8, which implies
a y2 = r
Also,

a8 _ sin 8 D20 2 cos 8 sin 8

Therefore, ax2 =
ax r r2

and

a8 cos 8 D20 2 cos 8 sin 8

Thus,
ay = r .
ay 2 =- r2

We are now ready to find the polar coordinate form of the heat equation. Using
the chain rule, we have

au Du Dr Du DO
ax = Dr ax + ae ax'
 28 Chapter 2: One-Dimensional Heat Equation

which means
D2 u _ 3 Du Dr Du DO (D2u Dr D2 u DO Dr Du D2 r
Dx2 Dx Dr ax + 093x Dr2 d x+ DBr d x d x+ Dr Dx2
D2 u DO D2 u Dr DO Du D20
+ Dal dx + are dx dx + de Dx2
D2 u 2 D2 u 2 sin 8 cos 8 D2 u sing 8 Du sing 8 Du 2 cos 8 sin 8
cos 0- r
Dr2 d r9 + 382 r2 + dr r + 38 r2 .

Similarly,
32 u D2 u 2 D2 u 2 sin 8 cos 8 D2 u cost 8 Du cost 8 Du 2 cos 8 sin 8
32 sin 0+
Dr2 are r + 382 r2 + dr r De r2
Thus,
D2 u D2 u D2 u D2 u 2 sin 8 cos 8 D2 u sing 8 Du sing 8
cos2 0-
3x2 + 5 2 5r2 are r + 382 r2 + 3r r
Du 2 cos 8 sin 8 D2 u 2 D2 u 2 sin 8 cos 8
+ r2 +D-2 sin e +
38 are r

+- 32 u cost 8
502 r2
+ Du cost 8 - Du 2 cos 8 sin 8 _ D2 u
dr r de r2 d r2
.+' _
1 Du
rdr
.+'
1 D2 u
--
r2 382

_ 13 du 1 32u
r
r Dr Sr + r2 392
.

Hence, we have
Du(r, 8, t) _ 1D Du,r(rdr8, t) 1 D2u(r, 8, t)
St \r 3r + r2 382 .

In the exercises, you must derive the cylindrical and spherical form of the Lapla-
cian.

EXERCISES 2.4

2.4.1. Prove Corollary (5).

2.4.2. Show that A cos w0t + B sin w0t can be written in the form r sin (wot + 0).
Determine r and 8 in terms of A and B. (Hint: Use the trigonometric angle-
sum relations.)
2.4.3. The Laplacian in the Cartesian coordinate system is defined as

72vi(r yi
'
- - - -L
D2u
Dx2
D2u
Dy2
D2u
Dz2 .
 Section 2.4: The Maximum Principle 29

(1) If x = r cos 8, y = r sin 8, and z = z, show that the Laplacian can be

written in the cylindrical coordinate system as
1D au 1 a2 u a2 u
Vu(r,
2
e,z) -_ r + r2 a82 + az2 .
r Dr Dr

(2) If x = r sin 8 cos , y = r sin 8 sin , and z = r cos 8, show that the
Laplacian can be written in the spherical coordinate system as
1a au a2u
V2u('r 8 ,q5) =
1
r2 sin 8
r2 ar ar + r2 sin 8 a82

1
a2 u
+
r2 sin2 a

2.4.4. (1) Given the cylindrical coordinates

p2 = x2 + y2 and b=tan- 1(Y) show that

(a) x = cos b.

(b) py = sin .
8¢ _ -sin q5
8x p
dJ 8¢ _ cosh.
ay p
(2) Given r = p cos i + p sin b7 + zk. Show that the square of the element
of arc length is
(ds)2 = drdr = (dp)2 + (dz)2.

2.4.5. Given the cylindrical coordinates

x = p cos b,

y = p sin b, and

z=z;
p2 = x2 + y2 and q5 = tan-1(a
), prove that the cylindrical coordinate system
x
is orthogonal.
 30 Chapter 2: One-Dimensional Heat Equation

2.4.6. (1) Given the spherical coordinates

x = p sin 8 cos b,

y = p sin 8 sin q5, and

z = p cos 8;

p2 = x2 -+
y2 + z2 and b = tan-1
(v), show that
x
ap
(a) ax = sin 8
a
(b) p = sin 8 sin b.
Dy
ap
(c) = cos 8.
az
sin b
(d) Dx p sin 8

(e)-=
Dy
Dqcos b
psin 8
(2) Given r = p sin 0 cos p sin 0 sin bj + p cos 8k, show that the square of
the element of arc length is

(ds)2 = drdr = (dp)2 + p2(dB)2 + p2 sin2

2.4.7. Given the parabolic cylindrical coordinates

y = aQ, and

z=z;
show that the square of the element of arc length is

(ds)2 = (a2 +132)(da)2 + (a2 + 32)(d/3)2 + (dz)2.

2.5 STEADY-STATE TEMPERATURE DISTRIBUTION

From the previous two sections we know how to set up the mathematical model for
heat conduction in a uniform rod with perfect lateral insulation, no internal sources,
specified temperatures at both ends, and length L. This mathematical model is
au(x, t) _ ka2u(x, t) (2.26)
at axe
 Section 2.5: Steady-State 31

subject to the BCs

u(0, t) = g(t)
(2.27)
u(L, t t)== h(tt )
and to IC
u(x, 0) = f (x). (2.28)

Now, suppose g(t) and h(t) are the fixed constants a and b, respectively. Then,
the solution to this model will simplify to a steady-state temperature distribution,
independent of time t. This solution is known as the equilibrium solution.
To understand what we mean by steady-state, think of a uniform rod with perfect
lateral insulation and no internal sources. Now, apply a constant temperature
forever to both ends. Since neither end is insulated, eventually the temperature in
the rod will adjust to the temperature distribution specified by the heat sources at
both ends.
Definition 6. A steady-state temperature distribution is a temperature distribu-
tion that does not depend on time.
t)
From Definition (6), u(x, t) from Equation (2.26) becomes u(x). Thus,
au(x,

at
Du(t) 32
= 0 and k
8(xz t) k d dx ax) - 0 or
d2u(x)
= 0, (2.29)
dx2

with BCs

u(0) = a and u(L) = b. (2.30)

Since the IC is concerned with the temperature distribution at time t = 0 and

steady-state is time independent, the IC is generally ignored.
Equation (2.29) is a rather simple second-order ordinary differential equation.
By integrating it twice, we arrive at the general solution
u(x) = Clx + C2. (2.31)

Applying the BC u(0) = a in Equation (2.31), we get C2 = a. Applying the second

a
b
-
boundary condition, u(L) = b, we get Ci = T . Thus our specific solution is

u(x) =
b-a (2.32)

In this problem, the steady-state temperature distribution is a straight line with

slope
b
-a
L and u(x) intercept of a. Figure (2.8) graphically illustrates u(x).
32 Chapter 2: One-Dimensional Heat Equation

0 L

Figure 2.8: Steady-state temperature distribution with u(0) = a and u(L) = b.

What would happen if we changed Equation (2.26) so that it had a time-

independent source term, Q(x), and we still wanted to solve for the steady-state
solution? Equation (2.26) becomes

Du(x, t) __ D2u(x, t)
Dt +
Q(x).

Dx2

Would the solution, Equation (2.32), change? How would this effect Figure (2.8)?
These questions are important to answer for a fuller understanding of this steady-
state problem; they'll be asked again in the exercises at the end of the section.
Next, suppose we want to find the steady-state temperature distribution in a
uniform one-dimensional rod with no internal source, perfect lateral insulation, and
perfect insulation on the boundaries. We again start with Equation (2.29)
d2u(x)
dx2 = 0.
However, the B Cs change to
du(0) _ du(L)
dx dx

Again, the general solution to the ODE is

u(x) = Clx+C2.
Applying both the first and second boundary condition yields
du(0)=O=C.
dx 1

This implies u(x) = C2. It appears that steady-state temperature in a uniform rod,
with perfect insulation everywhere, is an arbitrary constant. But is it arbitrary? No.
 Section 2.5: Steady-State 33

In this case, it turns out that the initial condition of the rod, u(x, 0) = 1(x), plays
an important part. The energy associated with the initial temperature distribution
cannot escape because of the perfect insulation. Since temperature distribution in
the rod must follow the three heat conduction axioms, the temperature distribution
levels out to a constant. But what constant? This question can only be answered
by going back to Equation (2.8), which is the mathematical representation of con-
servation of thermal energy. It states that for any thin slice of rod [a, b],

d
1b rau(a, t) _ au(b, t) 1
cpAu(x, t)dx = -K°A
dt L 8x 8x J

This equation is valid for the entire rod, not just the thin slice [a, b]; therefore, it
can be written as

t [1L cpAu(x, t)dx = -KoA La xt auax t J

L
+ AQ(x, t)dx. (2.33)
0

Because there is no internal source, Q (x, t) = 0. Also, we have a uniform rod, which
means c p, and Ko are constant and can be written as k = K° Thus, Equation .
cp
(2.33) may be written as

d 8u(0, t) 8u(L, t)
u(xt)dx] = - (2.34)
LLL 8x 8x

We can further reduce Equation (2.34) by remembering that both ends are perfectly
au(0, t) = 0 and au(L, t) = 0.
insulated, which means Thus Equation (2.34)
' ax ax
becomes
L

f u(x, t)dx = constant.

This tells us that the total initial heat energy inside the rod must equal the total final
heat energy inside the rod. We did not say that the initial temperature distribution
is the same as the final temperature distribution. Since u(x, 0) = 1(x), the total
initial heat energy is

L u(x, 0)dx = L f (x)dx.

J0 J0
34 Chapter 2: One-Dimensional Heat Equation

Also, since u(x) = C2, the total steady-state heat energy in the rod (the total final
heat energy) is
L

f C2 dx = C2 L.

Setting total initial heat energy equal to total final heat energy and solving for C2,
we arrive at
fL
C2 = L f (x) dx,
J0
which is the average of the initial temperature distribution.
As a final example, consider the following problem:
EXAMPLE 2.4. Consider a one-dimensional uniform rod with perfect lateral in-
sulation, internal heat source of 25 cos x, thermal diffusivity of 5, and initial temper-
ature distribution of 25 cos x. At the boundary x = 0, the rod is in direct contact
with a bath held at the constant temperature of 25°C, and at the boundary x = it,
the rod has perfect insulation. Determine and graph the steady-state temperature
distribution.
Solution: First, describe the above physical problem as a mathematical model.
We have
Du(x, t) a2u(x, t)
at - 5
axe
+ 25 cos x,

subject to the IC

u(x, 0) = 25 cos x

and BCs
u(O,t) = 25

8x

Second, state and solve the steady-state problem. The steady-state problem is
d2u(x)
5 + 25 cos x = 0,
dx2
subject to BCs

u(0) = 25

du('ir)
l = 0.
dx
 Section 2.5: Steady-State 35

The steady-state problem has the general solution

u(x) = 5 cos x + Clx + C2.
du(7r)
Applying the BC u(0) = 25, implies C2 = 20. Applying the BC = 0, implies
dx
Cl = 0. Thus, the specific steady-state solution is
u(x) = 5cosx+ 20.
Figure (2.9) is the graph of the steady-state temperature distribution.

u(x)

25

15

x
0
it

Figure 2.9: Graph of 5 cos x + 20.

EXERCISES 2.5

2.5.1. Given aone-dimensional uniform rod with perfect lateral insulation, determine
and graph the steady-state heat distribution given the following boundary
conditions, source, and thermal diffusivity:
(1) u(0) = 0, u(L) = a, Q(x) = 0, k = 2.

(2) u(0) = b, u(L) = 0, Q(x) = x, k = 1.

(3) u'(0) = 0, u(2ir) = a, Q(x) = 0, k = 1.

(4) u(0) = 0, u'(L) = 0, Q(x) = x2, k = 5.

(5) u(0) = u'(0), u(7r) = 15, Q(x) = 0, k = 1.

(6) u(0) = 2, u(L) = u'(L), Q(x) _ -x, k = 2.

 36 Chapter 2: One-Dimensional Heat Equation

2.5.2. Consider the mathematical model

Du(x, t) __ D2u(x, t)
k + au(x, t); a > 0, k> 0,
3x2
subject to the BCs

u(O,t) = 1 and -1
and IC
u(x, 0) = cos x - x.
(1) Write a short paragraph (five sentences or less) describing a possible
physical model.
(2) Determine the steady-state temperature distribution.
2.5.3. Determine the steady-state solution for a uniform rod with perfect lateral in-
sulation if the boundary at x = 0 is kept at a constant temperature of -10°C,
the boundary at x = 100 cm is kept at a constant temperature of 15°C, and
there exists a time-independent heat source, which is a linear function based
on the position in the rod.
2.5.4. Find the steady-state solution for a uniform one-dimensional rod with no inter-
nal source and perfect lateral insulation that satisfies the radiation condition
Du(0t)
- u(0, t) = 0
Dx
at the end x = 0 and is kept at a constant temperature T2 at the end x = L.
2.5.5. Find the steady-state solution for a uniform one-dimensional rod with an
internal source of Q(x, t) = x, perfect lateral insulation, that is kept at a
constant temperature Tl at the end x = 0, and satisfies the radiation condition
Du(L, t)
Dx
-u(L,t)=0
at theendx=L.
2.5.6. Consider the mathematical model
Du(x, t) D2u(x, t) tee_x
(cos ?Cx) =e + t2
Dt 3x2 1'
subject to the BCs

u(o,t)= o and a i

and IC
u(x,0) = x2.
 Section 2.5: Steady-State 37

(1) Write a short paragraph (five sentences or less) describing a possible

physical model.
(2) Determine the steady-state temperature distribution.

2.5.7. Consider
au(x, t) _ a2u(x, t)
8t 8x2
+x-i3,
subject to the BCs

as°'t)=0andaax't) =o
and IC

u(x, 0) = cos
x
(1) Find the equilibrium temperature distribution.
(2) For what values of 3 does the equilibrium temperature distribution exist?
Explain physically.

2.5.8. Suppose you were given

Du(x, t) __ a2u(x, t)
9 +9x,
Dt Dx2

subject to the BCs

u(0, t) = 0 and u(ir, t) = 0

and IC

u(x, 0) = e_x sin 3x.

Write a technical report, including explanations and mathematical details,

which contains at the least the following:

(1) a physical interpretation of the previous problem and

(2) for very large time:
i. the solution,
ii. the heat energy generated per unit time inside the rod,
iii. the heat energy flowing out of the rod per unit time at each end,
and
iv. the relationship between parts (b) and (c).
 38 Chapter 2: One-Dimensional Heat Equation

2.5.9. Suppose you were given

Du(x, t) a2u(x, t) 7r2u(x, t)
at = axe + 4

subject to the BCs

u(0, t) = 0 and u(6, t) = 0
and IC

u(x, 0) = it (6x - x2).

Write a technical report, including explanations and mathematical details,
which contains at the least the following:
(1) a physical interpretation of the above problem and
(2) for very large time:
i. the solution,
ii. the heat energy generated per unit time inside the rod,
iii. the heat energy flowing out of the rod per unit time at each end,
and
iv. the relationship between parts (b) and (c).
2.5.10. Suppose you were given
au(x, t) a2u(x, t)
+ e,
8t 8x2
subject to the BCs
u(0, t) = 0 and u(7r, t) = 0

and IC

u(x, 0) = 7rsinx.

Write a technical report, including explanations and mathematical details,

which contains,at the least the following:
(1) a physical interpretation of the above problem and
(2) for very large time:
i. the solution,
ii. the heat energy generated per unit time inside the rod,
iii. the heat energy flowing out of the rod per unit time at each end,
and
iv. the relationship between parts (b) and (c).
 Section 2.5: Steady-State 39

2.5.11. Consider a uniform one-dimensional rod with no internal source and perfect
lateral insulation. Assume that u(x, 0) = 1(x), and suppose the following
boundary conditions are given:
Du(s t) __ Du(-7r,t)
u(7r, t) = u(-7r, t) and
Dx Dx
(1) State a possible physical explanation. Briefly justify your answer.
(2) Does a steady-state solution exist? If so, what is the steady state so-
lution? Explain your answer. If a steady-state solution does not exist,
explain why.
2.5.12. Given the Cauchy-Riemann equations
Du - Dv _0
Dx ay
Du Dv °
Dy+Dx= '
show that both u and v satisfy Laplace's equation
D2z D2z
Dx2 + Dy 2 = °'
which may be considered as the multi-dimensional steady-state equation.
Laplace's equation is discussed in Chapter 5.
2.5.13. Fick's law of diffusion for chemical species in mathematical physiology is given
by

Q = -DVc,
where Q is the flow of a chemical across a membrane, D is the diffusion
coefficient and is dependent on the solute and the fluid in which the chemical is
dissolved, and c is the amount of chemical. Fick's law can be used to derive an
analogue of Ohm's law for a membrane of thickness, L, with different chemical
concentrations on each side of the membrane. If the medium is isotropic
(diffusion occurs the same regardless of the direction of the measurement),
then we get
Dc D2 c
at = D Dx2 '
subject to
c(0, t) = Cl, the chemical concentration on the left of the membrane,
and
c(L, t) = Cr, the chemical concentration on the right of the membrane.
Find the steady-state solution.?
7James Keener and James Sneyd, Mathematical Physiology, ©1998 by Springer-Verlag, New
York, pp. 36-38. Reprinted by permission.
Chapter 3

The One-Dimensional Wave

Equation
3.1 INTRODUCTION
The study of wave equations covers a wide range of physical problems. For instance,
the wave equation that governs vibration of a microphone diaphragm is

/a atD2u at\
+b
axe + x ax - 2 - °'
where u(x, t) is the displacement of a diaphragm in a capacitor microphone. Other
physical examples are the electromagnetic waves in a transmission line, wave motion
in an ocean, and vibrations in a beam. In later chapters, we will develop equations
and solutions for some of these physical examples. However in this chapter, we
consider the most basic wave equation: vibration of a one-dimensional string.
We derive the wave equation for vibration in a one-dimensional string in Sec-
tion (3.2). Boundary conditions (BCs) for the one-dimensional wave equation are
discussed in Section (3.3). Section (3.4) covers conservation of energy for the wave
equation and uniqueness of solution when using Neumann boundary conditions. We
conclude this chapter with the method of characteristics for first-order PDEs and
d'Alembert'sl solution for the one-dimensional wave equation.

3.2 DERIVATION OF THE ONE-DIMENSIONAL

WAVE EQUATION
When you think of vibrations in a one-dimensional string, you should ask yourself,
"Is the string vertical or horizontal, tightly stretched or loose?" There may be a
difference in the derivation of the equations. In fact, there are no vibrations in
'Jean Le Rond d'Alembert (1717-1783) was the first to develop a form of the wave equation

41
42 Chapter 3: One-Dimensional Wave Equation

a string that is not tightly stretched. Gravity could play a completely different
role in a string that is vertical versus a string that is horizontal. In this section,
we restrict ourselves to a tightly stretched horizontal string. Also, we make the
following assumptions:

Vibrations of the string are small. This means that as the string vibrates, we
have a very small change in the slope of the string from the "at rest" position,
which is horizontal.
Only vertical vibrations will be considered horizontal vibrations can be ne-
glected because of the small slope.
Vertical displacement then depends on the position on the string, along with
time, and can be modeled as y = u(x, t). Thus, the slope is represented by
dy _ au
tan[B(x, t)].
dx ax =
The string is perfectly flexible. It offers no resistance to bending.
Newton's law of motion, F = ma (which we will write as ma = F), is applied
to a small section of the string (x to x + Ox).
Consider Figure (3.1), which is out of proportion for labeling purposes:

'L(x + &, t)

8(x+&, t)

x+&

Figure 3.1: Finite string segment.

Here T, a tangential force, represents the tension in the string, and B is the angle
of the displacement of the string from the horizontal at the ends x and x + Ox.
It is unknown whether the string is uniform. Therefore, we represent the mass of
the string as a function. However, we only expect the mass of the string to change
with position, x, not with time, t. Thus, we chose p(x) to model mass, m, per unit
length and assume it is a known quantity.
In addition to mass, we must consider other forces on the string. One example
is the restoring force. This is the force that tries to return the string to its at-
rest position. This force models the added tension on the string caused by vertical
 Section 3.2: Derivation of the One Dimensional Wave Equation 43

displacement. Another example is the resistance force to the string's velocity.

The medium (air, water, oil, etc.) that the string vibrates in causes this force,
which attempts to slow any motion.
Using our assumptions, plus the discussion on forces and mass, we can now give
a fairly good mathematical model of the one-dimensional wave equation. Using
Newton's law of motion we arrive at
2
(x)p(x) = T(x + Ox, t) sin [9(x + x, t)] - (x, t) sin [9(x, t)]

-(Ox)a - (x)u + (x)Q(x,t). (3.1)

D2 u
Mass times acceleration is represented by the term (/2x)p(x) . The right side
Dt2
of Equation (3.1) is the sum of forces. We have two terms for tension: one for the
left end at x and one for the right end at x + Ox. Since only vertical vibrations are
considered, the terms r(x + Ox, t) sin [9(x + Ox, t)] and r(x, t) sin [9(x, t)] are the
vertical components of the tensile force. Assuming that a > 0, the term (L2x)c-
models the resistance force. (Ox)/3u with j3> 0 models the restoring force. Other
possible external forces, such as gravity, are modeled as (x)Q(x, t).
Dividing Equation (3.1) by Ox and taking the limit as Ox - 0, we get
D2u Du
p(x)
Dt2
D
= x ((x, t) sin [9(x, t)]) - a
a a
t - u + Q(x, t). (3.2)

In our assumptions, the slope was

dy _ 8u
dx ax -tan [B(x, t)] .
When the angle B is small (B ti 0), which is another of our assumptions, we have
dy au [B(x, t)]
=tan [9(x, t)] =sin sin [9(x, t)]
dx - Dx cos[9(x,t)]
or

dy
dx
- au
8x
sin [e(x, t)].

Replacing sin [8(x, t)] by in Equation (3.2) yields

a Du
D2 u
p(x) Dt2 = Dx T(x, t)
Dx
- a Du
Dt
- /% + Q(x, t). (3.4)

Equation (3.4) is the mathematical model for small vibrations in a small piece of
perfectly flexible string that is horizontal and tightly stretched.
If we make the assumption that the string is perfectly elastic (a valid assumption
for most strings), then the tensile force r(x, t) may be approximated by the constant
44 Chapter 3: One-Dimensional Wave Equation

T, which is the initial tension on the unperturbed string. Also, if we assume the
string is made of a uniform material, then the mass density p(x) becomes the
constant p and Equation (3.4) becomes
82u 82u 8u Nqu
p 8t2 - T 8x2 - a 8t - + Q(x, t).

Another assumption we can make is that the term Q (x, t) only models gravity.
If we know the tension force is high compared to the force of gravity, we may neglect
gravity. Doing so yields
012 012
u u au
pate = T cat - ,3u'
axe -
which is a form of the well-known Telegrapher's equation.
As its name implies, the Telegrapher's equation models electromagnetic wave
transmission in a wire. Also, it is an important equation in analyzing the time-
dependent Boltzmann's equation in the theory of neutron transport. We discuss
solutions to the Telegrapher's equation in a later chapter.
To simplify Equation (3.5) we may, for the moment, neglect the forces of friction
and restoration. Mathematically, this means we assume that a and j3 are zero. Thus,
the mathematical model for vibrations in a one-dimensional, uniform, perfectly
flexible, highly stretched string is
012 012

p=u TT axeu
or in its more usual form
012
u 2 Du
u
ate
T= c ax2'

where c2 = T . Because T is tension and p is mass per unit length, c has the
p
dimension of length/time, known as velocity. Actually, c is the specific velocity of
wave propagation along the string.
The one-dimensional wave equation applies to many different physical systems.
For example,
012 012
u u
at2 01x2

is the mathematical model of longitudinal or torsional vibrations in a rod. Here,

the constant k is a physical parameter known as the Young's modulus.
Another example is electrical current along a wire. Kirchoff's2 laws give us
ai C, av
+Gv = 0
ax + at
2 Gustav Kirchoff (1824-1887) was a physicist who obtained significant results in the study of
PDEs
 Section 3.2: Derivation of the One Dimensional Wave Equation 45

and
Dv L Di
Ri = 0
ax + at +
where the variables have the following meaning:

x is the location along the wire.

t is time.

i(x, t) is the current along the wire.

v(x, t) is the potential along the wire.
C is the capacitance.
G is the leakage conductance.

R is the resistance.
L is the self-inductance.

Using Equations (3.8 and 3.9), we can derive a form of the wave equation by
differentiating Equation (3.8) with respect to x and differentiating Equation (3.9)
with respect to t. This yields
D2 i D2v Dv
(3.10)
axe + cDxat + GDx = °
and
D2v D2 i Di
+ L at2 + R =0. (3.11)
Dtax Dt

Multiplying Equation (3.11) by C, then subtracting it from Equation (3.10), yields

a22 (DDx2vDt a2v av a2i

c ax - cL at2 - cR at = a (3. 2
atax +
D2v D2v
In Equation (3.12) the terms in the parenthesis, will equal zero if, in
axat Dtax ,
Dv Dv D2 v
the function v(x, t), the first derivatives, and , and the mixed partials,
Dt Dx axat
D2v
and are all continuous. This condition is expected in an electrical system.
Dtax ,
Therefore, Equation (3.12) becomes

Dv
cR Di =0.
D2 i D2 i
+G - CL 5t2 - (3.13)
axe Dx Dt
46 Chapter 3: One-Dimensional Wave Equation

av
Now, using Equation (3.9) in the form = -L a2 - Ri, we can rewrite Equation
Dx Dt
(3.13) as

D2i Di D2i Di
axe - GL Dt - cL Dt2 - cRDt -GRi = o
or in the more familiar wave form,
D2 i D2 i Di
CL = - (GL + CR) - GRi. (3.14)
a t2 a x2 at
In a similar fashion, we can derive a wave equation for the potential:
D2CLv D2v Dv
at 2
=
ax
2 - (GL + CR) t - GRv.
a
(3.15)

If, in Equations (3.14 and 3.15), G = R = 0, then we arrive at the equations

D2 i 1 D2 i
ate = CL Dx2
and
D2v 1 D2v
Dt2 - CL Dx2 '
which are easily recognized as the one-dimensional wave equation. Please remember
that the wave equation is the primary representative of the hyperbolic class of linear
second-order PDEs.
Other examples of physical systems that use the wave equation as a mathe-
matical model include sound waves, water waves, probability waves of quantum
mechanics, and vibrations in solids. However, to get a more complete picture of the
wave equation, we must know what conditions were present when we started the
experiment. These are called initial conditions; we discuss them next.
When we derived the heat equation we also discussed an initial condition (IC).
This IC described the heat distribution in the rod at the start of time. In the wave
equation, Equation (3.7), we have a second partial derivative with respect to time.
Thus, we must have two initial conditions (ICs).
One IC of the wave equation describes the starting position of the string. The
starting position of the string is the location of the string when a stopwatch starts
(t = 0). Therefore, the starting position depends only on x, and we model it as
u(x, 0) = f (x).
For a second IC, we will consider a plucked violin string. Suppose after the
violin string is plucked, we allow it to vibrate for a few seconds before we start a
stopwatch. We then determine the starting position. We also notice the string is
moving. This movement is the instantaneous velocity. It must be mathematically
modeled for a more accurate picture of the experiment. We know velocity is always
Du
the first derivative with respect to time of our function. Thus we have . Initial
Dt
Du(x'0)
velocity implies t = 0 in the function, resulting in = g (x) .
at
 Section 3.2: Derivation of the One Dimensional Wave Equation 47

Thus, our two ICs are

u(x,0) = f(x)

8u(x, 0)
l 8t - 9(x)

EXERCISES 3.2

3.2.1. Given the one-dimensional wave equation

82u 2 82u 8u
c
ate - 9x2 - a at - au + Q(x, t),
explain in two sentences or less the physical meaning of the terms
au

/3u, and

Q(x, t).

3.2.2. Show that for all positive integers m, each of the following functions satisfies
82u
ate
=c 2 82u
ax2
(1) ul(x, t) = sin[mirx] sin[mirct].
(2) u2(x, t) = sin[mirx] cos[mirct].

(3) u3 (x, t) = cos [(m + 2 I irx] sin [(m + 2 I irctJ.

ai + C av
Gv = 0
ax at +

av - L a2
+ Ri = 0.
ax at

CL t2
2 2
x2 - (GL + CR) t - GRv.
 48 Chapter 3: One-Dimensional Wave Equation

3.2.4. Show that

1
u(x, t) = 2
cosh2(x - 4t)

is a solution of the Korteweg-deVries equation

au au a3 u
at
+6u-
Dx + ax3
.

3.2.5. A linear approximation of one-dimensional isentropic flow of an ideal gas (a

gas in which the only stress across any element of area is normal to it) is given
by

where u = u(x, t) is the velocity of the gas and p = p(x, t) is the density of
the gas. Show that u and p satisfy the wave equation.

The following exercises involve second-order ODEs. A review of this material

may be found in Appendix C.

3.2.6. Solve the following second-order homogeneous ODES:

(1) u"(t) + 6u'(t) + 3u(t) = 0, u(0) = 1, u'(0) = 0.5.
(2) y" + 6y = 0.
(3) s"(t) - 6s(t) = 0, s(1) = 0, s'(0) = 1.
(4) g" (x) + 4g' (x) = 0.
(5) h" (t) - 3h' (t) + 2h(t) = 0, h(0) = 4, h'(3) = 1.
3.2.7. Consider a weight of 5 pounds attached to a steel spring that has a natural
length of 1 foot. The mass stretches the spring 0.25 foot. Suppose the system
is started in motion by stretching the spring an additional 0.1 foot in the
downward direction, then released. Determine and solve the resulting equation
of motion neglecting air resistance.
3.2.8. Solve the following second-order nonhomogeneous ODES.
(1) y"(x) + 3y'(x) - 4y(x) = 2e-4x.
(2) z"(y) + 2z'(y) - 3z(y) = 3e-3y.

(3) x"(t) + 2x'(t) + 4x(t) = 3 sin t.

 Section 3.3: Boundary Conditions 49

(4) u"(t) + au'(t) + wou(t) = coswt, a2 - 4w2 <0.

(5) ddt -2dt -y=tet+4.

3.2.9. Determine the solution of the differential equation

mu"(t) + cu'(t) + ku(t) = Fcoswt

satisfying the following ICs. Assume c2 - 4km < 0.

(1) u(0) = c, u'(O) =0.

(2) u(0) = 0, u'(O) = c.
(3) u(0) = c, u'(O) =

3.2.10. Solve

u"(t) + u(t) = coswt, w 1

subject to ICs

u(0)=0
u'(O) = 0.

Show that the solution may be written as

((1+w)t) 2sin ((1_w)t\

u(t) = 2sin
1-w2
2
)
Using your favorite mathematical software, graph the solution for at least
three different values of w.

3.3 BOUNDARY CONDITIONS

In the previous section, we derived Equation (3.6) and Equation (3.7), which are
commonly referred to as the equations that govern wave motion in a one-dimensional
au (x 0)
string. Also, initial position, u(x, 0) = f (x), and initial velocity, = g(x),
the ICs, were discussed. However, when we model physical phenomena, we must
try to model as many constraints as possible. We would fall short of this goal if we
did not discuss the ends of the string, called boundaries, as we did in Chapter 1.
In this section, we discuss the mathematical model of various boundary conditions
for a one-dimensional string.
50 Chapter 3: One-Dimensional Wave Equation

3.3.1 Boundary Conditions of the First Kind

For a one-dimensional string, Dirichlet conditions describe where boundaries are
attached. A fixed attachment would be considered a fixed boundary condition. For
example, consider a one-dimensional string fixed at some constant displacement
S, then u (O, t) = S and u (L, t) = S. A special case of constant displacement is
u(0, t) = 0 and u(L, t) = 0. These describe boundaries fixed with zero displacement
from the horizontal axis, homogeneous BCs.
Alternatively, we can use Dirichlet conditions to describe how boundaries of the
one-dimensional string are controlled or specified. That is, a function describes the
physical displacement of the end of the string over time from the horizontal axis.
For example, if the boundary x = 0 of the string moves with time in an up and
down motion only, then that boundary may be mathematically modeled as
u(O,t) = gl(t). (3.16)

3.3.2 Boundary Conditions of the Second Kind

Neumann conditions describe the tensile force on a one-dimensional string at the
boundaries. Note: A mathematical model of tensile force (from the derivation of
DU(X, t)
the wave equation) is T For example, suppose at the boundary x = L, the
tensile force applied changes with time. The equation governing this is
au(L, t) _
T - g2(t). (3.17)
ax
The tensile force does not have to change with time. It could be a constant. In
this case, the mathematical model for the boundary at x = L is
8u(L,t) _ S
8x
A very special case of constant tensile force is when S = 0, and we have at the
boundary x = L, as in
8u(L,t) _ o
8x
This particular boundary condition means that at the boundary x = L the string
is attached to a frictionless sleeve which moves vertically. This may seem impossi-
ble, but it is a standard mathematically modeled case. A more interesting case is
boundary conditions of the third kind.

3.3.3 Boundary Conditions of the Third Kind

Robin's conditions for a one-dimensional string describe some type of an elastic
attachment at both ends of the string. For example, consider Figure (3.2). Here,
the ends of the string are attached to a spring. The spring has its other end fixed.
 Section 3.3: Boundary Conditions 51

Figure 3.2: String attached to elastic ends with their ends fixed.

The spring constant is assumed to be positive. Also, we will assume the spring
constant is the same at both ends, and we will denote it as k.
The mathematical model for this physical condition is
au(0t)
T = ku(0 t
ax )

for the end x = 0. For the end, x = L, we have

5u(L, t)
ax ( )

Note: The spring constant may not be the same at both ends. This condition would
indicate different springs attached to either end of the string.
Another example of Robin's conditions is shown in Figure (3.3). Here, the end

Figure 3.3: String attached to elastic ends with their ends displaced.

x = 0 is attached to a spring where the end of the spring can move in a vertical
 52 Chapter 3: One-Dimensional Wave Equation

direction. The displacement of the end of the string is described by u (O, t). The
displacement of the left end of the spring can be described as d(t). The tension on
au(0, t)
the end of the spring at its attachment to the string is described by T By
ax
setting the vertical tension of the spring equal to the difference of the displacements,
we arrive at the mathematical model
Du(O, t)
______ = k[u(O,t) - d(t)]. (3.18)

Assuming the spring obeys Hooke's3 law, k is the spring constant. The signs in
Equation (3.18) are identical to Newton's law of cooling, which was developed in
Chapter 1. The end x = L has a similar mathematical model, and it is left as an
exercise.

EXERCISES 3.3

3.3.1. Explain in your own words what happens to the Robin's condition
5u(O,t)
= h[u(O,t) - 'yi(t)]

when

(1) h -f oc.
(2) h -f 0.
3.3.2. State the mathematical model for the given information:
(a) Small vertical vibrations in a uniform tightly-stretched string.
(b) Fixed left end at 4.
(c) Free right end.
(d) String length of 37r.
(e) Tension of 4.
(f) Mass density of 3.
(g) Initial displacement of x2 + 2x + 4.
(h) Initial velocity of 0.
3.3.3. State the mathematical model for the given information:
(a) Small vertical vibrations in a nonuniform tightly-stretched string.
(b) Free left end.
3R. Hooke (1638-1703) was professor of geometry at Gresham College and secretary of the
Royal Society.
 Section 3.3: Boundary Conditions 53

(c) Fixed right end at -2.

(d) String length of .
2
(e) Tension of 4.
(f) Mass density of 2x + 1.
(g) Initial displacement of 0.
(h) Initial velocity of -2 sin x.
3.3.4. Consider a one-dimensional tightly stretched string in the horizontal position.
Suppose the mass density of the string is constant; the left end, at x = 0,
is fixed at a height of 3.5 off the horizontal axes; the right end, at x = 7r, is
allowed to move freely; the initial position of the string is given by 1(x) =
x2 - 27rx + 3.5, and the initial velocity is 0. State the mathematical model.
3.3.5. Given the following equation, describe the physical situation:
a2u a au -x2
x e -t
3
(x+5) ate = (2x + 1)
ax ax +
subject to BCs:
Su(0,t)
=t2+2t+1
u 4,
(4,t t) =0
and ICs:
u(x, 0) = cos x + sin x - 1

au(x, 0)
_ In (x + 1) - 2.60759.
at
3.3.6. Given the following equation, describe the physical situation:

ate = a
a2u au
(x + 1) 2
ax
(3x+5)-)
ax
+ sin xt,

subject to BCs:
u(0, t) = o

8u(L, t)
_ 3 cost
8x

and ICS:
u(x,0) =ln(x+1)
Su(x,0)
at
-0
 54 Chapter 3: One-Dimensional Wave Equation

3.3.7. Develop the mathematical model for Robin's conditions at the boundary x =
L described in Figure (3.3).

3.3.8. A uniform string with fixed ends has an initial displacement of 2x for 0 < x <
r and 37r - x for 7r < x < 37r. It is known that the initial velocity is 0 and
the string is vibrating in a medium that resists the vibrations. (The medium
produces a resistance proportional to the velocity.) Suppose the resistance
constant of proportionality is 0.01. State the mathematical model.

3.3.9. A uniform string with a fixed end at 0 and free end at 27r has an initial
displacement of -x for 0 < x < and 3x - 67r for x < 27r. It is
- 2 2 - -
known that the initial velocity is 0 and the string is vibrating in a medium
that resists the vibrations. (The medium produces a resistance proportional
to the velocity.) Suppose the resistance constant of proportionality is 0.03.
State the mathematical model.

3.3.10. A uniform string with free ends has an initial displacement of

,0<x<1
u(x,0)= x-,1x<3
-2x+8,3<x<4.
It is known that the initial velocity is 0 and the string is vibrating in a medium
that resists the vibrations. (The medium produces a resistance proportional
to the velocity.) Suppose the resistance constant of proportionality is 0.13.
State the mathematical model.

3.4 CONSERVATION OF ENERGY

FOR A VIBRATING STRING
The principle of conservation of energy states that if no energy is lost due
to friction or other possible forces, then in a mechanical system the sum of the
instantaneous kinetic and potential energy is equal to a constant. From physics, we
know kinetic energy
gY of a point with mass m is E k 12 mv2 where v is the velocit Y
,

of the mass point at any time t. Also, potential energy is given by E = lkx2 ,
where k is a force constant of proportionality and x is the coordinate of the body.
Consider the equations for a uniform tightly stretched vibrating string with free
ends and initial conditions,

52 u 252u
c (3.19)
ate = ax2 '
 Section 3.4: Conservation of Energy for a Vibrating String 55

subject to BCs

(3.20)

and ICs

(3.21)
8u(x, 0)
8t -
Here, the kinetic energy is

Ek=-2 L fau(x, t)J12 dx,

and the potential energy is

EP = 2 fL fa"(x't)Jl2dx.
(3.23)

Equations (3.22 and 3.23) are different than the single-point mass equations because,
for a string, we must sum over the entire length. Thus, we have

E=Ek+E
= 2 L (5u(x,t)2d x + 2 J L (5u(x,t)2dx. (3.24)
J0 0
1

We need the sum of the instantaneous kinetic and potential energy, which means
taking the derivative with respect to time, t. Hence, the equation

dE d [11L (5u(xt))2 2 L 8u(x, t) 2

dx + dx
dt dt 2 at 2 ( 8x
)
becomes

dE_ d [1 t)1 2
+ c2
ty12
dx (3.25)
dt 2 ,f L l
Using Leibniz's formula on Equation (3.25) yields
2
dE La au(x, t)2 2a 1' 't)
dt - 2, at at
+
at ax ) dx
56 Chapter 3: One-Dimensional Wave Equation

which reduces to

dE L a2u(x, t) (au(xt)\ 2 a2u(x, t) au(xt))d x (3.26)

8t2 8t + axat ( ax

dE _ L 2 a2u(x, t) I au(x, t) t) (au(xt) dx (3.27)

dt o C
ax2 at + axat
2 a2u(x,
ax

The integrand of Equation (3.27) may be recognized as the derivative of a product,

which is
a au(x, t) au(x, t) a2u(x, t) (au(x,t)\ + a2u(x, t) (au(x, t)
ax at a- axe at axat ax
Therefore, Equation (3.27) becomes

dE 2 La au(x, t)au(x, t)
dt CL 8x [ at ax ] dx

2 [au(xt) au(x, t)l L

at ax o

which means the sum of the instantaneous kinetic and potential energy is equal to
a constant.
We may now proceed and prove that if Equations (3.19, 3.20, and 3.21) have a
solution, then the solution is unique.

Proof. Let ul (x, t) and u2 (x, t) be solutions to the given equations. Then, v(x, t) _
av(0, t) _ aul (0, t) au2 (0, t) av(L, t) _
ul (x, t) - u2 (x, t) . Thus, = 0 and
ax ax ax ax
_
Dui(L,t) au2(L, t) _ av(x, 0) _
ax ax -
0. Also, v(x, 0) = ul (x, 0) - u2 (x, 0) = 0 and
at
aul (x, o) _au2 (x, o) _
at at -
0. Thus, the sum of the instantaneous kinetic and potential
energies of v(x, t) must equal a constant. We form the energy equation
L
1 c2o L (av(xt)\2
dx dx. (3.28)
E- 2o at
(av(x,t)\2 + 2 ax
Taking the derivative with respect to t yields

dE
2 dt f oL
t)12
J
dx + 2 L (av(xt)
J o
a 2 dx
'
(3.29)
 Section 3.5: Method of Characteristics 57

which becomes

dE _ rav(x, t)av(x, t) 1 L
dt L at ax 0
This means that E(t) = 0, which in turn implies that the integrands of Equation
Dv (x, t) (x, t)
(3.28) are zero. Thus and must be identicallyY zero which implies
cat Dx
v(x, t) - 0. Thus, ul (x, t) = u2 (x, t). D

EXERCISES 3.4

3.4.1. Given a uniform tightly stretched vibrating string with a fixed end at x = 0,
a free end at x = L, and initial conditions, determine what happens to the
total energy E.
3.4.2. at the end x = L, and the springs other end fixed, and initial conditions,
determine what happens to the total energy E.
Note: The spring constant is assumed to be a positive.
3.4.3. Given the equations for a uniform tightly stretched vibrating string with ho-
mogeneous fixed ends and initial conditions,
a2 u 232u
=c axe' (3.30)
ate
subject to BCs:

u(0, t) = o
(3.31)
{
u(L, t) = 0
and ICs:
u(x,0)=f(x)
(3.32)
Du(x,0)
a:
at 9(x).

Show that if this problem has a solution, then the solution is unique.

3.5 FIRST-ORDER PDES:

METHOD OF CHARACTERISTICS
In this section, we use the method of characteristics to analyze first-order constant
coefficient PDEs. A characteristic of a first-order PDE is a curve where the PDE
58 Chapter 3: One-Dimensional Wave Equation

becomes an ODE. A common example of this is a surfer catching a wave. The

surfer moves along the wave with a velocity equal to the wave's. The wave may
move faster or slower depending on any number of variables. But, from the surfer's
point of view, the wave does not change.
We begin our investigation with the simple first-order PDE
Dz - c Dz _ 0 (3.33)
Dt 8x
with IC

z(x,0) = f(x), (3.34)

where -oc <x < oc and 0 <t < oc. Unlike the second-order wave equation, which
has two initial conditions, a first-order PDE will only have one initial condition.
This initial condition describes the initial position at time t = 0. Remember, initial
conditions depend on the number of partial derivatives with respect to time that
are in the equation. A first-order PDE only has one partial derivative with respect
to time. Also, there are no boundary conditions because the variable, x, varies from
negative infinity to positive infinity.
One way to solve this problem is to consider the rate of change of z(x(t), t) as
measured by a moving observer, x = x(t). The chain rule implies

tz(x(t),t) - 5zdx 5z
ax dt + at'
(3.35)

Comparing these quantities with equation (3.33), we see that if the observer moves
dx
along at velocity -c then = -c and z(x(t)' t) = 0. That is, z is a constant.
dt dt
Dz
The term in Equation (3.35) represents the change in z with respect to time
Dt
Dz dx
at the fixed position x. The first term in Equation (3.35), represents the
Dx dt '
change that the observer sees as the observer moves into different regions of the
domain of z.
Using Equation (3.35), we have reduced the PDE in Equation (3.33) to two
dx dz
first-order ordinary differential equations: = -c and = 0. If we consider
dt dt
that
dx _
dt
-c '
we see that the solution is

x(t) _ -ct + a, (3.36)

where a represents the initial point on the characteristic curve when t = 0. Using
the IC (3.34), we have z(x(t), t) = z(a, 0) = f (a).
 Section 3.5: Method of Characteristics 59

Solving for a in Equation (3.36) yields a = x(t) + ct, or simply a = x + ct.

Therefore,

z (x, t) = f (x + ct). (3.37)

Equation (3.37) is known as the general solution of Equation (3.33).

EXAMPLE 3.1. Consider
az - 35z0
at ax
with initial condition

z(x, 0) = sin x,

where - oc <x < oc and 0 < t < oc. Letting

z(x, t) = z(x(t), t).
We find the derivative of z with respect to t, which is
dz _ az dx az
dt ax dt + at
Thus,
dx _ -3
dt '

which implies

x(t)= -3t+a.
Also,

dz _ 0
dt '

which implies

z(x(t),t) = c.
Thus,

z(x(0), 0) = sin(x(0)).

Therefore, using Equation (3.37), we find that the general solution is

z (x, t) = sin(x + 3t).

60 Chapter 3: One-Dimensional Wave Equation

z(x, t)

it

Dz Dz
Figure 3.4: The solution sketch for -3 =0
Dt Dx

Figure (3.4) shows a sketch of the solution for values 0 < t < 1 and -?C <x <7r.
Two of the parallel characteristics x = a - 3t are clearly evident in the solution
sketch. They are represented by a solid line for a = and a dashed line for a = it.
2

We will now consider the more general first-order constant coefficient PDE
Du Du
a u=0 (3.38)

with IC

u(x, 0) = 1(x), (3.39)

where-oc<x<oo,0<t<oo, and8>0.
As introduced in the beginning of this section, one way to solve this problem
is to consider the rate of change of u(x(t), t) as measured by a moving observer,
x = x(t). Applying the chain rule yields
du _ Du dx Du
(3.40)
dt Dx dt + Dt
dx
If we assume that = a the left side of Equation (3.40) can be substituted into
dt
Equation (3.38). Then, we have
du du
+/3u=0 or = -/3u,
dt
 Section 3.5: Method of Characteristics 61

which is a first-order ODE with solution

u(x, t) = c(x)et.
Applying the IC (3.39) yields a general solution for u(x, t) of
(x)e-at.
u(x, t) = f
Note: The solution of the first-order PDE is the solution of a first-order ODE, where
the arbitrary constant of the ODE is replaced by an arbitrary function.
dx
To determine a more specific solution for u(x t) based on our assumption
dt
c, we must solve for x and place that solution in the function 1(x). The solution
dx
to = c is x = cxt + a where a is the constant representing the initial position
dt
of the characteristic curve when t = 0. Therefore, when x = a and t = 0, we have
u(x, t) = u(a, 0). Thus, the complete solution of Equation (3.38) with IC (3.39) is
u(x, t) = f (x - (3.41)

Because of the term e-13t, we know as time goes to infinity, u(x, t) tends to zero. This
seems quite reasonable if we assume Equation (3.38) is a first-order wave equation.
In this case, the term 3u is a damping function that tends to flatten out the wave
as time passes.

5
1

u(x, t)

Figure 3.5: Solution sketch for u(x, t) = sin(x - 0.25t)e-o.5t

EXAMPLE 3.2. Consider

at + o.z5a + 0.5u = 0 (3.42)

 62 Chapter 3: One-Dimensional Wave Equation

with IC
u(x, 0) = sin x, (3.43)
where -oo <x < oo and 0 < t < oo. Since c = 0.25 and ,3 = 0.5, we know from
Equation (3.41),
u(x, t) = f(x - cEt)e_t,
that the specific solution is
u(x, t) = sin(x - 0.25t)e-o.5t.

Figure (3.5) shows a sketch for -2ir < x < 2ir and 0 < t < 5. Two parallel
characteristics are shown on the graph. However, remember these characteristics
take into consideration the exponential e-o.5t

EXERCISES 3.5

3.5.1. Solve and graph the following using the method of characteristics.
az az
(1) 5 = 0 with IC z x 0= 2x.
az az
(2) -3 = 0 with IC z x 0 = x2 - 2x 1.

aw aw
(3) ax +6w = 0 with IC w(x, 0) = cos x.
at
au au
2u = 0 with IC u(x, 0) = sinx.
(4) at + 4ax -
3.5.2. Consider the simple concentration problem
au au
= o, 0 < t < oo, - oo < x < oo,
at + ax
with IC
u(x, 0) = cosx.
Solve and graph the solution using your choice of mathematical software. Does
the solution satisfy the PDE and the IC?
3.5.3. Solve
av av0<t<oo
-+x-=0, -oo<x<oo
at ax
with IC
v(x, 0) = 0.5x2.
Graph the solution using your choice of mathematical software. Does the
solution satisfy the PDE and the IC?
 Section 3.5: Method of Characteristics 63

3.5.4. Solve
au au
at
+ax+tu=0, 0<t<oo, -oo<x<oo,
with IC

u(x, 0) = x2.

Graph the solution using your choice of mathematical software. Does the
solution satisfy the PDE and the IC?
3.5.5. Solve the surface wave problem
Du +a Du + au + u=0 0<t<oo -oo<x<oo -oo< <oo
at ax a "
with IC

u(x,y,0) =sin(x+y),
where c, ,3, and 'y are constants.
3.5.6. Solve the surface wave problem
au
at
au
+cEax
au
+/3-
ay
+u = 0 0< t <00, -00 <x <00, -00 <yy <00,
o0 ,

with initial condition

u(x,y,0) =sin(x+y),
where c = 2, ,3 = -1, and 'y = 1.
3.5.7. Given the first-order semilinear PDE
au(x,y) au(x,y)
a (x,y, u(x, u)) + b(x ,y, u(x ,y)) ax = c (x,y, u(x,y)),
ax
show that the method of characteristics yields
dx dy du(x, y)
a(x, y, u(x, y)) b(x, y, u(x, y)) c(x, y, u(x, y))

When an IC is given, Equation (3.5) is sometimes called the first-order Cauchy4

problem.
a2 u 1 a2 u au
3.5.8. Given cat + /3u = 0, which is another form of the Telegraph
axe c2 at2 +
4Augustin-Louis Cauchy (1789-1857) a very famous French mathematician who derived the
name of determinant
64 Chapter 3: One-Dimensional Wave Equation

au au
Equation, let v = u w = and z = . Show that v w, and z must
ax ' at >

satisfy the following system of three equations,

av
z = 0,
at
aw az
0
at - ax
and
az 2 aw
+ cxz + 3v = 0.
at - c ax

3.6 D'ALEMBERT'S SOLUTION TO THE

ONE-DIMENSIONAL WAVE EQUATION
In the last section, we considered the method of characteristics as a solution to
first-order PDEs. In this section, we show the method of characteristics may be
applied to solving the one-dimensional wave equation. Here, the chain rule has an
important role. This was first published by Jean Le Rond d'Alembert in 1747. We'll
start with the one-dimensional wave equation with no BCs, That is, the string is
considered infinite:
a2u 2 a2 u
=C (3.44)
ate axe
with ICs
u(x, 0) = f(x)
-oo <x < oo. (3.45)
8u(x, 0) }

Remembering our calculus, if u(x, t) is a real-valued function (which it is) such

that
au au and a2u or a2 u (called the mixed partials) are continuous at every
ax , at , atax axat
a2u
point in the domain, then at every point in the domain the derivatives or
atax
a2u a2u a2u
exist and = To use this theorem, we must assume that the
.
axat atax axat
mixed partials are continuous at every point in the domain. Thus, we can rewrite
Equation (3.44) as
a2u a2u a2u 2 a2u
=0.
ate + ca-; -
C (3.46)
axat - ax2
This can be "factored" in the following way:
a a au au
=0.
at - ax at + ax
 Section 3.6: d'Alembert's Solution to the One-Dimensional Wave Equation 65

Du
Letting z = c ields the first-order PDE
yields
Dt + Dx

az az
at-cax=°,

which we know from the previous section has the general solution
8u 8u
z(x, t) = at + sax = Q(x + ct), (3.47)

where Q is some arbitrary function. Note: In Equation (3.37), z(x, t) = f (x + ct),

where the function 1(x) was the initial condition. In Equation (3.47), we do not
know the function explicitly. Therefore, we assign it the arbitrary function Q.
Similarly, if we "factored" Equation (3.46), that is, we found that

(a + sax) (at -sax) =0

au - c au
and let v = we would have another first-order PDE
at Dx

av av
at
+c ax =0,
which has the solution
au au
v(x, t) = at -sax =Pox - Ct) (3.48)

where P is some arbitrary function.

Adding Equation (3.47) to Equation (3.48), we obtain
Du 1
=_ [Q(x+ct)+P(x-ct)].
at 2

aF(x - ct) 1 DG(x + ct) 1

If we let -c
Dt = 2 P(x - ct) and c at = 2 Q(x + ct), then use some
calculus, we obtain

u(x, t) = G(x + Ct) + F(x - ct) (3.49)

as the general solution of Equation (3.44). It is interesting to note that Equation

(3.49) expresses the solution of the one-dimensional wave equation as the sum of
two moving waves, moving in opposite directions with velocity c. This makes sense
if you remember that we are dealing with an infinite string. You pluck the string,
and the wave moves off in both directions.
Applying ICs (3.45), we obtain a specific solution. The first IC says that

u(x, 0) = G(x) + F(x) = f (x). (3.50)

66 Chapter 3: One-Dimensional Wave Equation

To apply the second IC, we use the chain rule. Consider the term G(x + ct) in
Equation (3.49). It is actually a composition of the functions G(X) and X (x, t) _
aG _ dG aX ax aG _ aG _ dG
x + ct. Thus, The fact that = c implies c c
at dX at ' at at aX dX '
G
since G is a function of X only. Therefore, cddX = cG' (X ) = cG' (x+ ct). Similarly,
aF _
at - -cF' (x - ct)
Using the method of the previous paragraph and applying the second IC, we
obtain
8u(x, 0)
= c [G'(x) - F'(x)] = g(x). (3.51)

To find a solution for G(x), we take the derivative of Equation (3.50) with respect
to x, multiply by c, and add it to Equation (3.51), which yields

2
[+i'x)]
c
( )

By integrating, we see that

G(x) = Zf(x) + 2c LX 9(s)ds + kl. (3.52)

A similar calculation yields for F(x)

x
()
F (x) 12f(x) - 1
2c 0
9( s)ds +k2. (3.53)

In Equations (3.52 and 3.53), k1 and k2 are constants of integration. However, since
G(x) + F(x) = 1(x), we must have k1 + k2 = 0. Therefore, the general solution of
Equation (3.44) is the sum of Equations (3.52 and 3.53), each shifted a distance of
ct; u(x, t) can be written in the form
f9(S)dsJl
u(x, t) = 2 [f(x + ct) + f (x - ct)] + g(s)ds -
2c [,I o +ct

u(x, t) = 2 [f(x + ct) + f (x - ct)] +

1p/'a+ct g(s)ds. (3.54)

This is known as d'Alembert's solution to the one-dimensional wave equation. Phys-

ically, we combine two sets of characteristic curves (sometimes called two families
of characteristic curves) to form the solution of the one-dimensional wave equation.
EXAMPLE 3.3. Consider
a2u(x, t) 2 a2u(x, t)
ate
=c ax2 ' -oo< x < oo,
 Section 3.6: d'Alembert's Solution to the One-Dimensional Wave Equation 67

subject to
h, if x <a
u(x, 0) _
0, if x > a
8u(x,0) _ p
at
Determine the solution using d'Alembert's method and graph the solution for sev-
eral values of time t.

Solution: The general form of d'Alembert's solution is

1 1
x+ct
u(x, [f(x ct) f (x g(s)ds.
t) = 2 + + - ct)] + 2c -ct
Since g(x) = 0, we have

u(x,t) = {f(x+ct)+f(x-ct)]

and know that F(x) = G(x) = 2 f (x). This means

-a a

Figure 3.6: Graph of d'Alembert's solution for t = 0.

h
if x <a
F(x)=G(x)=
I 0, if x > a.
In Figure (3.6), we have t = 0 and u(x, t) = 2 [f(x) + 1(x)] = f (x) = h. Since
x <a, we have the endpoints of the graph at -a and a. Remember, the full pulse
h
is made up of two smaller pulses. Each smaller pulse has a height of 2 and as soon
68 Chapter 3: One-Dimensional Wave Equation

-a-ct a-ct a+ct

-a + ct

a
Figure 3.7: Graph of d'Alembert's solution for 0 < t < .
c

h/2

H H
-2a 2a

a
Figure 3.8: Graph of d'Alembert's solution for t = .
c

as t 0 the pulses will start to separate, as shown in Figure (3.7).

a
In Figure (3.7), 0 < t < , we have half of the pulse going to the left and
c
h
the other half of the pulse going to the right, both with a height of 2 . However,
a
since t < , we still have the height of h where the pulses overlap. We determine
c
the endpoints for each change of the pulses by again considering x < a. Here, x
actually is x+ct and x-ct. For x+ct, we have -a < x+ct < a or -a-ct < x < a-ct.
For x - ct, we have -a < x - ct <a or -a + ct <x <a + ct. Also, note the pulse
going to the left has endpoints of -a - ct and a - ct, and the pulse going to the
right has endpoints of -a + ct and a + ct.
a h
In Figure (3.8), t = both pulses are now at the height of 2 , and the pulses no
c'
longer overlap. Using the formula x <a where x = x + ct = x + a, we obtain the
 Section 3.6: d'Alembert's Solution to the One-Dimensional Wave Equation 69

endpoints for the left pulse, which are -2a and 0. The right endpoints are obtained
similarly by letting x = x - ct = x - a. Thus, the right endpoints become 0 and 2a.
a
Finally, for Figure (3.9), t> , and we have two separate pulses, each of height
c

h/2

-a - ct a-ct -a +ct a + ct

a
Figure 3.9: Graph of d'Alembert's solution for t> .
c

h
2. One pulse is traveling to the left and the other pulse is traveling to the right.
The pulse traveling to the left has endpoints of -a - ct and a - ct. These endpoints
are determined by letting x = x + ct while remembering x < a. Thus, we have
x + ct <a or -a < x + ct <a, which becomes -a - ct <x <a - ct. Similarly,
using x = x - ct, the right endpoints are found to be -a + ct and a + ct.

EXERCISES 3.6

3.6.1. What is the solution to the initial-value problem

a2 u a2 u
=2 0 < t < oo - 00 <X < 00,
ate axe
subject to

u(x, 0) _

au(x, 0) _ 0?
at
Sketch the solution for several values of time, t.
 70 Chapter 3: One-Dimensional Wave Equation

3.6.2. What is the solution to the initial-value problem

a2u a2u
=3 ; 0 < t < oo; - oo < x < oo,
O`lt2 axe
subject to

u(x, 0) =0
and

au(x, 0)
at 00, , 1?

Sketch the solution for several values of time, t.

3.6.3. What is the solution to the initial-value problem
a2u _ a2u
ate axe, 0<t<oo; -oo<x<oo,
subject to

u(x, 0) _
0,
0, x>1
and

8u(x, 0)
0?
at

Sketch the solution for several values of time, t.

3.6.4. What is the solution to the initial-value problem
a2u a2u
ate =4ax2; 0<t<oo; -oo<x<oo,
subject to

u(x, 0) =0
and

8u(x, 0)
at

Sketch the solution for several values of time, t.

 Section 3.6: d'Alembert's Solution to the One-Dimensional Wave Equation 71

3.6.5. What is the solution to the initial-value problem

a2 u a2 u
9 ax2 ; 0 < t < oo; -oo < x < oo,

subject to
- sin x, x < 7r
u(x, 0) _
0, >r
and
au(x, 0)
0?
at

Sketch the solution for several values of time, t.

3.6.6. What is the solution to the initial-value problem
a2u a2u
' 0 < t < oo; - oo < x < oo,
axe
subject to
- cos x,
u(x,0) _
0, > it
and

8u(x, 0)
0?
8t

Sketch the solution for several values of time, t.

3.6.7. What is the solution to the initial-value problem
a2 u 152u
ate - 4ax2'
0<t<oo; -oo<x<oo,
subject to

u(x, 0) = 0
and

8u(x, 0)
sin x, x < r
at 0, > it?
Sketch the solution for several values of time, t.
 72 Chapter 3: One-Dimensional Wave Equation

3.6.8. What is the solution to the initial-value problem

a2u _ 1 a2u
ate
9ax2'
0<t<oo; -oo<x<oo,
subject to

u(x, 0) = 0

and

cosx, <
8u(x, 0)
at o xx> 2-.
Sketch the solution for several values of time, t.
3.6.9. Solve for propagation of electrical vibrations in an infinite conductor for the
condition

GL = CR,
where G, L, C, and R are the leakage conductance, self-inductance, capac-
itance, and resistance, respectively, per unit length of the conductor. At
the start of the experiment, the current is i (x, 0) = 1(x) and the voltage is
v(x, 0) = g(x).
3.6.10. Given
a2u 25uu
=c axe' (3.55)
ate
subject to
u(x,0) = f(x)

8u(x, 0) } -00<x<00, (3.56)

consider the coordinate transformations = x - ct and i = x + ct. Using the

chain rule on v( , i ), develop d'Alembert' solution for Equations (3.55 and
a2u a2u
3.56). Hint: Let u(x, t) = v( , i) and determine ate and axe ni terms of
v(, i).
Chapter 4

The Essentials of Fourier

Series
4.1 INTRODUCTION
In previous chapters, we discussed the derivation of the heat and wave equations.
Also, we discussed the solutions to the steady-state problem for the heat equation,
and d'Alembert's method for the wave equation in an infinite string. Although these
solution methods are important in their own right, they do not tell us the entire
story of the partial differential equations governing heat conduction in a rod, or
motion of a tightly stretched finite string. In fact, the steady-state method tells us
nothing about heat conduction in a rod at time t = 5 seconds, or, for that matter,
at any finite time t. Although there are ways of developing d'Alembert's method for
a finite string, they aren't always the best or the preferred solutions, nor is there an
equivalent d'Alembert's method for the vibrations in a drumhead. However, other
methods can solve wave motion in a string and temperature distribution in a rod,
as well as multispatial dimensions; these methods are based on Fourier series.
Traditionally, Fourier series is introduced as the closed form solution to a second-
order PDE with boundary conditions. Although this method is completely correct,
and was originally introduced by Fourier, it sometimes leaves you with the feeling
that something is missing. The effect is that students must rely on brute memory
to get through the material. In this chapter, we introduce Fourier series by making
use of an earlier concept, which you've already used or have been introduced to,
linear algebra.
Linear algebra is central to a fuller understanding of mathematics. It lays the
foundation for the higher mathematics used in physics, engineering, and applied
mathematics. The framework of linear algebra plays a crucial role in the devel-
opment of quantum mechanics, solving systems of ODEs in engineering, and the
advanced mathematical tools used by applied mathematicians.
Using linear algebra as an introduction to Fourier series may seem to you like
the blind leading the blind. Many engineering and physics students have never been

73
74 Chapter 4: The Essentials of Fourier Series

formally introduced to linear algebra. But, all students who have taken the calculus
sequence, ODEs, and some type of engineering mathematics course have worked
with linear algebra through most of their college career. Therefore, this chapter
builds on this material to develop the ideas from linear algebra that are needed
to understand the connection with Fourier series in the section titled "Elements of
Linear Algebra." The chapter then develops the mechanics of Fourier series. Finally,
even and odd functions and their Fourier series representation are discussed.
The linear algebra portion of this chapter should not be construed as a replace-
ment for a linear algebra course. We cover only those theorems and ideas relevant
to developing Fourier series. Also, we do not present the proofs nor much of the
mechanical manipulations that are necessary for a better understanding of linear
algebra. If you enjoy the material covered in the linear algebra portion, I would
suggest you take a linear algebra course specifically designed for applications. Also,
the notation of this chapter may be new to you. If this is the case, I suggest you
read Appendix D on mathematical notation.

4.2 ELEMENTS OF LINEAR ALGEBRA

First, linear algebra needs to be defined. This is not an easy task because, quite
literally, it means the algebra of lines or the algebraic study of lines. Therefore, we
need to think of a more informative approach. I'll start with an explanation of the
term, "linear," in linear algebra. Then, I'll explain how the word "algebra" fits in.
You have a geometric and algebraic understanding of linear. For example, the
equation y = mx + b geometrically describes a line in the xy-plane called IIS2. In
algebra the same equation y = mx + b is called a linear equation. However, in linear
algebraic vocabulary this equation is known as an affine equation. An affine
equation is basically a shift of a linear equation. In linear algebra, a linear equation
must go through the origin: This means that b = 0 in the equation y = mx + b.
This concept of "linear" in linear algebra forces a mathematical assumption. That
assumption is that a line going through the origin contains all the information about
all the lines parallel to it in the plane, except the y intercept, which is nothing more
than the shift of the line. In IR3, this concept is a little more complicated, but the
idea is the same.
Now that we understand what "linear" means in linear algebra, the concept of
"algebra" in linear algebra must also be explained. In high school, you learned
about algebra on the real numbers. This is basically the study of multiplication
and addition on general real numbers, using equations to explain simple physical
phenomena. In the equations, scalars (commonly called constants) came from the
real numbers. This really does not change much in linear algebra. You still work
with equations. Also, scalars are still real numbers. However, the general numbers
in the equations have changed to vectors. Thus, 1lS2 is a space in linear algebra.
Linear algebra, as in all subjects, has evolved past its humble beginnings of
Jft2 and has been applied to other spaces. To discuss these more evolved spaces,
development of a more general space structure was needed. The general space
structure was called a vector space. Don't let the words "vector space" scare you.
 Section 4.2: Elements of Linear Algebra 75

The definition is just a mathematical formalization describing something that occurs

in nature. In fact, you have been living in a vector space all your life. For example,
if you consider where you are standing as the origin, then an axis running through
your head into the ground could be the z-axis and the xy-plane would be the ground
you are standing on. This is actually a rough copy of IIJ3, which is a vector space.
Understanding the definition of a vector space is fundamental to the study of
linear algebra. We define vector space in the next subsection.

4.2.1 Vector Space

Definition 7. A vector space V is a non-empty set of elements x and y (called
vectors), such that the algebraic operations, vector addition and scalar multiplica-
tion, hold.

This is a very concise definition, but what does it mean? We will start with
vector addition.
Vector addition includes several ideas. First, when adding two vectors in a vector
space V, the sum is also in V. This concept is known as closure of the space. Second,
vector addition is commutative: If x and y are vectors in V, then x + y = y + x.
Third, vector addition is associative. This means that if x, y, and z are vectors in
V, then x + (y + z) = (x + y) + z. Last, there exists a unique vector 0 in V such
that if x is any vector in V, then x + 0 = x. Also, there exists the vector -x in V,
such that x + (-x) = 0. Next, we consider scalar multiplication.
Scalar multiplication also includes several ideas. First, a scalar is nothing more
than a number, and sometimes you will read or hear the phrase "a real vector
space." This means that the scalars are all real numbers. If you are reading a math
text and you see the phrase "complex vector space," then the scalars come from
the complex plane and have the form a + bi. In this text, we will work exclusively
with real vector spaces. Also, a vector space V is closed under scalar multiplication.
Second, scalar multiplication distributes over vector addition. This means if x and
y are vectors in V, and a is any real number, then a(x + y) = ax + ay. Third, scalar
multiplication is associative. That is, if x is any vector in V and a and b are real
numbers, then a(bx) = (ab)x. Finally, there exists an identity element for scalar
multiplication. Since we are working exclusively with real vector spaces, you can
clearly see that 1 is our identity element.
In many texts, Definition (7) is written in the following form:
Definition 8. A real vector space V is a non-empty set of elements x and y (called
vectors), such that:

1. For any pair of vectors x and y in V, there exists a unique vector in V that
is the sum of x and y.

2. Vector addition is commutative. For any pair of vectors x and y in V, then

x+ y = y+ x.
76 Chapter 4: The Essentials of Fourier Series

8. Vector addition is associative. For any vectors x, y, and z in V, then x+ (y+ z) =

(x + y) + z.

4. There exists a unique vector 0 in V, such that for all vectors x in V, x + 0 = x.

5. For every vector x in V, there exists the vector - x in V, such that x+ (-x) = 0.

6. For all vectors x in V and all real numbers, a, the unique vector ax is in V.

7. Scalar multiplication distributes over vector addition. For any pair of vectors
x and yin V and real number a, we have a(x + y) = ay + ax.

8. Scalar multiplication is associative. For all vectors x in V and the real num-
bers a and b, a(bx) = (ab)x.

9. The scalar 1 is the identity element for scalar multiplication.

You may use either Definition (7) or Definition (8). Examples of real vector
spaces are IIS2 (the xy-plane) and II83 (three-dimensional space).
The next subsection increases our knowledge of a vector space V by introducing
terms you already know from previous courses. These terms are linear dependence,
linear independence, and basis.

4.2.2 Linear Dependence, Linear Independence, and Basis

Central to the concepts of linear dependence and linear independence is the concept
of linear combination of vectors in a vector space. A linear combination of vectors
is the term for the addition of two or more vectors in a vector space. In general, we
use the following definition:
Definition 9. Let V be a vector space. Suppose the vectors ul, u2, u3, ..., un are
elements of V and al, a2, a3, ..., an are scalars. Then, a linear combination of
the vectors u1, u2, u3, ..., un is
alul + a2u2 + a3u3 +... + anon.
Note: A linear combination of vectors in a vector space yields a vector. Remem-
ber, the zero vector, written 0, is a valid vector.
EXAMPLE 4.1. Consider the following vectors in standard form in II.
1 -2 5
u= -2 ,v= 0 ,andw= -6
3 1 8

A possible linear combination is

1 -2 5
-2 +2 0 5 -6
3 1 8
 Section 4.2: Elements of Linear Algebra ??

Also, at first glance, the three vectors in the above example seem to be unrelated.
But, a careful inspection shows 3u - v = w or
-2
0
1
1

This means the vector w can be written as a linear combination of the other two
vectors. Thus, the vectors u, v, and w are linearly dependent. The formal definition
of linear dependence in a general vector space follows:
Definition 10. Let V be a vector space. Suppose the vectors u1, u2, u3, ..., un
are elements of V. If al, a2, a3, ..., an are scalars and a linear combination of the
vectors u1, u2, u3, ..., un ,
a1 u1 + a2 u2 + a3 u3 + ... + an un = 0,
(where ai, 1 < i < n) are not all equal to zero, then the vectors u1, u2, u3, ... , un
are linearly dependent.
To explain this definition, consider the three vectors u, v, and w given in Ex-
ample (4.1). Since w = 3u - v, we have w - 3u + v = 0. Here a1 = 1, a2 = -3,
and a3 = 1. Now that we know what linearly dependent vectors are, we need to
define linearly independent vectors.
Definition 11. Let V be a vector space. Suppose the vectors v1, v2, v3, ..., vn are
elements of V and a1, a2, a3, ..., an scalars. Then, the linear combination of the
vectors v1, V2, V3, ..., vn is given by
a1v1+a2v2+a3v3+...+anvn = 0.
The vectors v1, v2, v3, ..., vn are said to be linearly independent only when ai =
0 f or all i such that 1< i< n.
To demonstrate this definition clearly, consider the following example.
EXAMPLE 4.2. Let
1 -1 0
v1= 1 ,v2= 1 ,and v3= 1
-1
-1 --11 1

We will show that these vectors are linearly independent. Suppose we form the
linear combination of the vectors v1, v2, and v3 as
1 - 1 0 0
al 1 +a2 1 + a3 1 0
-1 -1 1
1 0

where we assume at least one of the scalars a1, a2, or a3 does not equal 0. From the
first line across all three vectors, we find that a1 = a2 because a1 - a2 = 0. Also,
78 Chapter 4: The Essentials of Fourier Series

from the second line across all three vectors, a3 = -2a1 since a1 + a2 + a3 = 0.
Using substitution we have 2a1 + a3 = 0. However, from the third line across all
three vectors, -a1 - a2 + a3 = 0, implying a3 = 2a1. Thus, 2a1 = a3 = -2a1. This
in turn implies a contradiction to our assumption, since 2a1 = a3 = -2a1 implies
al = 0, which means a2 = 0 and a3 = 0.
It is easy to see that the vectors

L
] and [o]
in R2 and
1 0 0
0 1 and 0
0 0 1

in 113 are linearly independent. These vectors are referred to as the standard basis
vectors for the vector spaces IR2 and IR3, respectively. Considering the format of the
standard basis in R2 and J3, we could guess that the standard basis in Rn would
have n vectors, each vector with n components, one in the nth position and all other
positions with a zero. This would be a good guess and quite correct. In fact, the
usual way to state the standard basis in 1Rn is
1 0 0
0 0 0
0 1 0
0 0

0
L0i 0 r
0 r r
1

where it is understood that there are n vectors with n components in each vector.
There are other bases for IR2, 113, and IRn. These will be discussed in a later
subsection. Next, we'll discuss orthogonality and inner product. These topics may
sound new to you, but they really are not.

4.2.3 Orthogonality and Inner Product

1182 and liS3 are the vector spaces most familiar to you. The following is a good
geometric definition of orthogonality: If two vectors intersect and the intersection
forms a right angle, then the two vectors are orthogonal to each other. This means
the two vectors are perpendicular to each other. A more mathematical definition is
as follows:
Definition 12. Two vectors are orthogonal if and only if their dot product is zero.
Since we know how to perform the dot product, Definition (12) gives us a way
to determine if two random vectors are orthogonal. This removes the uncertainty
 Section 4.2: Elements of Linear Algebra 79

of guessing or trying to geometrically determine orthogonality, which may require

the use of a straight edge and protractor.
You have determined in many previous courses, by using the dot product, that
the standard bases in JR2 and JR3 is an orthogonal set, and you can see that the
standard basis in JR ' is also an orthogonal set. Also, it is standard practice in
calculus courses to show two or more random vectors are orthogonal. Therefore, I
will not run through an example.
The question that should come to your mind is "Why use the term orthogonal
if it just means perpendicular in a geometric sense?" In JR2 and IR3, it is easy
to understand when two vectors are perpendicular, thus orthogonal. In JR ' for
n > 3, it is impossible to see, but you can somewhat imagine perpendicular lines by
relating them to JR2 or JR3. Thus, you have an idea of the concept of orthogonality
here, also. However, a vector space was defined in general terms and real vector
spaces JR2, JR3, and JR are not the only vector spaces, they are just the vector
spaces you are most familiar with. For instance, another vector space is the vector
space composed of all 2 x 2 matrices with real-number components. It is called
the matrix space of 2 x 2 matrices, and it is denoted M2,2. In the matrix space,
M2,2, perpendicular matrices have no meaning. Thus, "orthogonal" has become
the word most preferred. Also, the term "orthogonal" has become tied to another
term, "inner product."
The standard inner product of two vectors in JR2, J3, or Rn is our old friend the
dot product. However, the mathematical notation has changed. The dot product
between two vectors u and v was always denoted as u. v. Now the inner product
between the same two vectors u and v is denoted as (u, v) , and when we discuss
the standard inner product on JR3, or Jn, the notation is (u, v) = u. v.
Since the dot product is now an inner product, the definition for orthogonality
can be restated:
Definition 13. Two vectors are orthogonal if and only if their inner product is
zero.

We use the term inner product because the dot product only works on the vector
spaces Jn, for n e N. In other vector spaces, the dot product would make no sense.
For instance, the matrix space M2,2 cannot use the dot product as its inner product.
This naturally leads us to the general definition of an inner product.
Definition 14. Let V be any real vector space. An inner product on V is a func-
tion that assigns a real number to each pair of vectors u and v of V, written
(u, v) = a e III, satisfying:

1. (u, u) > 0 for u 0, and (u, u) = 0 if and only if u = 0.

2. (u, v) = (v, u) for any u and v in V.

3. (u + v, w) = (u, w) + (v, w) for any u, v, and win V.

4. (bu, v) = b (u, v) for u and v in V and b a scalar in III.

80 Chapter 4: The Essentials of Fourier Series

It is left as an exercise for you to show that the inner product (dot product) in
the vector spaces IIS2 and ii is, in fact, an inner product. Also, it should be noted,
any vector space where an inner product has been defined is known as an inner
product space.
An example of an inner product space, that is not one of the W' 's is the inner
product space composed of all 2 x 2 matrices with real-number components. If A and
B are elements of M2,2, then the inner product is defined as (A, B) = tr(BT A).
In the next example, we prove the space of all 2 x 2 matrices with real-number
components is an inner product space.

EXAMPLE 4.3. We must show that the space of all 2 x 2 matrices with real-
number components, M2,2, is an inner product space. We are using the inner
product, (A, B) = tr(BT A). We start with the vector A, where

a b
A c d

First, show that

(A,A) > O for A 0 and A E M2,2.

Thus, we have

(A, A) = tr(AT A) = tr (f a c1 ra
\lb dJ L
a2 + c2 ab + cd
= tr ba + do b2 + d2

= a2 + c2 + b2 + d2 > 0 for a, b, c, and d 0.

If

A- 00
0 0 '

then

(A,A) = O for A E M2,2.

Also, if
 Section 4.2: Elements of Linear Algebra 81

and we have

(A,A) = tr(AT A) = tr\L a

I I

a2 + c2 ab + cd 11
= tr
ba + do b2 + d2

= a2 + c2 + b2 + d2 = 0

then a, b ,c, and d = 0. Second, we show that

(A,B) = (B,A) for any A and B in M2,2.
Let

A=La d andB=Lg e
Then,

(A,B)=tr(BTA)=tr(F e g1 [a b 1Lf

h] Lc d])
=tr rI ea + gc eb + gd
fa+hc fb+hd ])
L

= ea+gc+ fb+hd,

and

c1 re f 11
(B,A)=tr(ATB)=tr (f
\lb
a
dJ[9 h]J
=tr ae+cg of +ch
be+dg bf +dh ])

=ae+cg+bf +dh.
Therefore,
(A,B) _ (B,A) for any A and B in M2,2.
Next, we show that
(A+B,C) = (A,C)+(B,C) for any A, B, and C in M2,2.
82 Chapter 4: The Essentials of Fourier Series

Let

A=I a dJ,B=I g hJ,andC=Lo pJ.

Lc L

Then,

(A+B,C) = tr (CT (A + B)) = tr I [ m

o a+e b+ f
n p c+g d+h )

=tr(f m(b+f)+o(d+h) 11
n(a + e) + p(c + g) n(b +1) + p(d + h) J J

= m (a + e) + o(c + g) + n(b + f)+p(d+h),

and

(A, C) + (B, C) = tr(CT A) + tr(CT B)

Lra
tr pJ
pJ Lg hJ/
ma+oc mb+od r me+og mf +oh
tr
na+pc nb+pd ) +tr I( ne+pg of +ph I

= ma+oc+nb+pd+me+og+nf +ph

= m(a + e) + o(c + g) + n(b + f) + p(d + h).

Therefore,

(A+B,C) = (A,C)+(B,C) for any A, B, and C in M2,2.

Finally, we show that

(3A,B) = ,3 (A, B) for any A and B in M2,2 and /3 in III.

Let

A=La dJ andB=[e
9
h]'
 Section 4.2: Elements of Linear Algebra 83

Then,

(f e g 1 ,<3a 3b 11
(,<3A, B) = tr (BTA) = tr
\Lf hi L dj)
e/3a + g/3c e(33 + g/3d
= tr
f /3a + h/3c f /3b + h/3d )

= era+g/3c+ f3b+h3d = /3 (ea + gc + fb+hd),

and

e g 1 [a b n"
h][c d])
=,6tr I L fa+hc fb+hd ])

= /3(ea+gc+fb+hd).
Thus,

(3A, B) = ,3 (A, B) for any A and B in M2,2 and /3 in III.

Therefore, we have shown that M2,2 is an inner product space with inner product
defined by (A, B) = tr(BT A).

Later in this chapter, a completely different inner product in a completely differ-

ent space will be introduced. This new inner product is crucial to our understanding
of how Fourier series works with PDEs. For now though, we move onto topics you
should be familiar with, eigenvalues and eigenvectors.

4.2.4 Eigenvalues and Eigenvectors

In this subsection, the one definition is stated for the general case n x n. However,
all discussions and examples will be with respect to the vector space This vector
space is quite familiar, and it facilitates the comprehension of the ideas underlying
the material of eigenvalues and eigenvectors.
The first time you encountered eigenvalues and eigenvectors may have been in an
ODE course. Eigenvalues and eigenvectors played an important part in the solution
of systems of first-order linear ODES. However, the emphasis was on the solution to
the system, not on eigenvalues and eigenvectors. Here, we are primarily concerned
with eigenvalues and eigenvectors themselves. Let's start with a definition.
84 Chapter 4: The Essentials of Fourier Series

Definition 15. Let A be an n x n matrix. Then a real number \ is called an

eigenvalue of the matrix A, if and only if for some nonzero n x 1 vector x, the
equation
Ax= ;
is true. If the vector x exists, then the vector x is called the eigenvector corresponding
to the eigenvalue ).
The definition does not indicate how to find the eigenvalue and the associated
eigenvector. However, we determine the eigenvalues by solving the equation
A - all = 0.
Remember the symbol means to find the determinant of the matrix.
EXAMPLE 4.4. Consider the matrix

Determine the eigenvalues and associated eigenvectors for the matrix A.

First, form the matrix A - Al. This matrix is

A - aI = C5 3 1 Ol(5-\
)_ 3 1
3 5 0 1 3 5-,\
Next, find the determinant of A - ,\I, and set A - aI = 0. That is,
5-a 3
0= 35-a = (5-a )(5-a -9=a2-10a+25-9
)

=a2-10\+16.
The equation ,\2 - 10\ + 16 = 0 is called the characteristic polynomial, and it
can be solved for roots by factoring. Once factored, the roots are = 2 and \ = 8.
These are the eigenvalues. Generally, they are labeled as = 2 and '\2 = 8.
To determine the associated eigenvector for each of the eigenvalues, form the
equation (A - \I) x = 0 for each eigenvalue. Thus, for the eigenvector correspond-
ing to the first eigenvalue, we solve

0= (A-)1I)x= 5 3- 1 5-\1
3 x1
x2

This equation can be solved using the Gauss-Jordan reduction method and the
eigenvector x is given by
(xi(
x
We now have the corresponding eigenvector to the eigenvalue )1 = 2. In a similar
fashion, the corresponding eigenvector x to the eigenvalue )2 = 8 is
 Section 4.2: Elements of Linear Algebra 85

It is interesting to note that the eigenvectors

-1) and I 1

are orthogonal to each other.

The previous example shows that eigenvalues and eigenvectors can be deter-
mined in a systematic way. Also, it is important to note an n x n matrix generally
has n distinct eigenvalues, each of which has a corresponding eigenvector. In the
next subsection, we tie all the pieces of linear algebra that are analogous to the
development of Fourier series together.

4.2.5 Significance
In the previous subsections, vector spaces, linear combinations, linear independence,
inner product, orthogonality, and eigenvalues and eigenvectors were discussed, but
they were not related. Here, hopefully, we will fit the pieces of the puzzle together.
All of the examples will be centered on the real vector spaces, and R3, since you
are most familiar with them.
Everything that occurs in this section occurs in a vector space. This is an impor-
tant concept. It means mathematics and reality are tied together in some fashion
and that fashion is the development of mathematical objects that act the same as
three space,
In a vector space, a basis is extremely important and necessary. A basis is
composed of linearly independent vectors. Another important point in the real
vector spaces W, n e N, is that the n in the superscript position tells us how many
vectors are in the basis. For instance, IIhas three vectors in the standard basis,
and they are

Also, the n in the superscript position identifies the vector space to be finite di-
mensional. This does not mean the space does not go on to infinity, it just means
there are only a finite number of copies of the real axis being used to make up
the space. Are there infinite dimensional spaces? The answer is yes, and in the
next section you will be introduced to one, the function space of piecewise smooth
functions. But, before we discuss infinite dimensional space, let's tie together some
of the tools we need from the real vector spaces.
Basis vectors are linearly independent. This means any vector in the space can
be expressed as a linear combination of the basis vectors.
EXAMPLE 4.5. Express the vector
-3
18
7
86 Chapter 4: The Essentials of Fourier Series

as a linear combination of the standard basis vectors. The solution is

-3 1 0 0
18 =-3 0 +18 1 +7 0
7 0 0 1

Although Example (4.5) is easy to complete and, in fact, can be completed al-
most unconsciously, it really should bring up two very important questions. First, is
the standard basis the only basis? Second, if another basis exists, how do we deter-
mine the multiplicative constants of the basis to form a linear combination as in the
standard basis? We will answer each of these questions because an understanding
of the underlying concepts is necessary.
The standard basis is not the only basis for R3, or for that matter any vector
space. The standard basis is the easiest one to work with because in most vector
spaces the standard basis is known to be orthonormal, meaning the vectors that
make up the basis have unit length and are orthogonal to each other.
Another orthonormal basis for IIis
1 1
r -1 1

1 L/J
If you want to express

as a linear combination of the orthonormal basis in Equation (4.1), you must deter-
mine the constant multipliers. It does not seem to be an easy task. However, the
inner product is the tool that makes it happen. If you compute the inner product
of the vector with each vector in the basis in Equation (4.1), then the answers are
the respective constants.
EXAMPLE 4.6. Express the vector

as a linear combination of the orthonormal basis

r -1 1
and

LVJ LJ
 Section 4.2: Elements of Linear Algebra 87

Solution: First, compute the inner products of the vector and the orthonormal
basis. That is, find

r-1-1
2
1
LJ
and
r -1 1

10

L/J
Note: Here the inner product yields the contribution of the vector in the direction
of each unit vector; this is because the basis vectors are orthonormal. Next, form
the linear combination, which is
r-1-1 r -1 1

22 32 10
0
1 L/J
1

LJ
Example (4.6) shows us how to find the constants of the linear combination
when we have an orthonormal basis.
Since we know other bases exist for the real vector spaces, how do we find them
and how do we make them orthonormal? The answer lies in two theorems stated
here, with proofs in Appendix B and a process, and it involves topics discussed in
the previous subsections. The topics are eigenvalues and eigenvectors.
Theorem 16. Let , '\2, /\3, ... ) be distinct eigenvalues of an n x n matrix.
Then, the corresponding eigenvectors x1, x2, x3,. . xn form a linearly independent
.

set of vectors. That is, c1 x1 + c2 x2 + c3 x3 +. . + c7xn = 0 if and only if ci = 0 for

.

1 <i <n.
Theorem 17. If an n x n matrix A has n distinct eigenvalues, then the correspond-
ing eigenvectors form a basis for W.
 88 Chapter 4: The Essentials of Fourier Series

Remember, eigen means characteristic. Thus, eigenvalues are the characteristic

values of an n x n matrix, and the corresponding eigenvectors are the characteristic
vectors of the n x n matrix. Knowing that characteristic means distinguishing
trait or feature tells us that the eigenvalues and corresponding eigenvectors are a
distinguishing feature of the n x n matrix and the corresponding n-space. What
could be a more distinguishing feature of n-space than the basis. Using a basis in n-
space allows one to describe every vector in that space by using a linear combination
of the basis and Theorem (17) tells us that eigenvectors are a basis.
Both theorems are necessary. Theorem (16) tells us the eigenvectors are linearly
independent. This is required for any basis. Theorem (17) informs us the eigen-
vectors form a basis for the real vector space, which has the same superscript as
the size of the matrix. But how do we know if the eigenvectors are orthonormal?
Generally, the eigenvectors are not orthogonal and usually do not have unit length.
This is why a process is needed: The Gram-Schmidt orthonormalization process,
which makes the eigenvectors orthonormal. However, the process is rather lengthy
to explain and not needed for this course. What is important is that eigenvalues
and eigenvectors have a role in determining a basis in vector spaces. This idea will
show up again in the next section.

EXERCISES 4.2

4.2.1. Which of the following sets of vectors in IIare linearly dependent, and which
are linearly independent:
0 1 3
(1) 2 2 6 ,

LaJ 3 3 6 J
( 1 -1 0
(2) 2 2 1 ,or
( 3 -5 7
J
1

2
3

4.2.2. Find the characteristic polynomial, eigenvalues, and corresponding eigenvec-

tors for the following matrices:

(1)
2 ii
[
2 -2 3
(2) 0 3 -2 and
0 -1
-1 2
2

2 23
(3) 1 2 1

2 -2 1
 Section 4.2: Elements of Linear Algebra 89

4.2.3. Given the matrix

102
021,
-1 0 6

find the eigenvalues and eigenvectors. Then, perform the following process:

(1) Name the eigenvectors a1, a2, and a3 (order doesn't matter).
(2) Let vector /1 = a1.
(3) Find the vector /32 by using the formula

/32-a2 - (a2, /31) /31,

where (a, b) is the standard inner product for

(4) Find the vector /33 by using the formula

(a3,,3i) (a3,132)
/33 = a3 - (/3/3)/31 - (/32,/32)/32

(s) The vectors i1, /32, and /33 are an orthogonal basis for Make them
an orthonormal basis.
(6) This process is known as the Gram-Schmidt process. Prove you have
produced an orthonormal basis for R3, which can be used to describe
every vector in II.

4.2.4. Show that the dot product in the vector spaces IIg2 and IIis an inner product.

4.2.5. Show that the space of all 2 x 3 matrices with real-number components, M2,3,
is an inner product space. Note: Use the inner product (A, B) = tr (BTA),
where A and B are elements of M2,3.

4.2.6. Show that the space of all 3 x 2 matrices with real number components, M3,2,
is an inner product space. Note: Use the inner product (A, B) = tr (BTA),
where A and B are elements of M3, 2 .

4.2.7. If w1 and w2 are positive real numbers, show that the definition

(xl,x2) (yl,y2) = wlxlyl + W2x2y2

yields a different inner product on 2 (Hint: Show all four properties of an

inner product hold.)
90 Chapter 4: The Essentials of Fourier Series

4.3 A NEW SPACE: THE FUNCTION SPACE

OF PIECEWISE SMOOTH FUNCTIONS
The function space of piecewise smooth (PWS) functions is an inner product space
where the elements are PWS functions. Therefore, a new inner product must be
defined, because the inner product on PWS functions can't be the dot product.
Also, unlike the vector spaces R, orthogonality in a function space may not easily
be recognized geometrically. But, we are getting ahead of ourselves. Inner product
and orthogonality are discussed in the next subsection. First, we must define the
elements of the function space of PWS functions.
For our purposes, the function space of PWS functions is composed of PWS
functions, each of which exists on the closed finite interval [a, b]. Also, a PWS
function seems to imply the function can be in pieces. This implication is true to a
certain extent. To understand piecewise smooth functions, we first need to define
the term jump discontinuity.
Definition 18. A function f (x) has a jump discontinuity at the point x = xo, if
the limit of f (xo) from the left, written limo 1(x), exists and the limit of f (xo)
from the right, written lim+ f (x), exists, but limo f (x) limo f (x).
It is important to note that the word "exist" in the definition of jump disconti-
nuity means the limit must be finite. Below, in Figure (4.1) and Figure (4.2), are
two graphical examples of functions with at least one jump discontinuity. Figure
(4.1) is f (x) _ [x + 1 ] , the step function on the interval [-2, 2]. Whereas, Figure
(4.2) is a function with random jump discontinuities on the interval [-L, L]. Also,

1.5

0.5
a b
x
-2 -1 1 2
- 0.5

Figure 4.1: f (x) _ 1] on the Figure 4.2: A function, f (x), with

interval [-2, 2]. several jump discontinuities.

a good counterexample of the definition of jump discontinuity is the function tan x

on the interval [0, ?C], which is graphed in Figure (4.3). It is a good counterexample,
because at x = , tan x has a discontinuity, but it doesn't fit our definition of a
2
jump discontinuity.
With a good understanding of jump discontinuity, we can define a piecewise
smooth function.
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 91

0 x

Figure 4.3: 1(x) = tan x on the interval [0,71].

Definition 19. A function f (x) on the interval [a, b] is piecewise smooth if the
interval [a, b] can be broken up into a finite number of subintervals such that the
df
function, 1(x), and its derivative (x) are continuous on each of the subintervals.
dx
This definition of a piecewise smooth function given implies that the function
1(x) may not be continuous on the interval [a, b], but the discontinuities are only
jump discontinuities. Thus, the function f (x) is bounded on the interval [a, b]. Both
Figures (4.1 and 4.2) are good examples of piecewise smooth functions. Another
example is f (x) = ( x, graphed in Figure (4.4) on the interval [-2,2]. Note: The

Figure 4.4: f (x) _ x on the interval [-2, 2].

function f (x) = x is continuous on the interval [-2,2], but at x = 0 the derivative

does not exist. Thus, we break the function into the intervals [-2, 0) and (0, 2]
where the function and the derivative are both continuous, and we recognize it as
a PWS function.
Summarizing the properties of the function space of piecewise smooth functions
we have the following:

The function space exists on the closed interval [a, b].

The elements are functions.
92 Chapter 4: The Essentials of Fourier Series

The functions and their first derivatives are continuous except at possibly a
finite number of jump discontinuities.

Understanding the setting for the function space of piecewise smooth functions
is only part of the story. The other pieces of the story are discovering reasonable
definitions for inner product, orthogonality, and basis.

4.3.1 Inner Product, Orthogonality, and Basis

in a Function Space
In the vector space IRn, n e N, the dot product is the most familiar inner product.
However, in the function space of PWS functions on the interval [a, b] the dot
product would make no sense. Therefore, a new inner product must be determined.
To aid us, the definition for an inner product is restated below for a general vector
space.

Definition 20. Let V be any real vector space. An inner product on V is a func-
tion that assigns to each pair of vectors u and v of V a real number, written
(u, v) = c e Ilk, satisfying the following:

1. (u, u) > 0 for u 0, and (u, u) = 0 if and only if u = 0.

. (u, v) _ (v, u) for any u and v in V.

3. (u + v, w) _ (u, w) + (v, w) for any u, v, and w in V.

4. (bu, v) = b (u, v) for u and v in V and b a scalar in Ilk.

Remember, the words "real vector space" mean the associated scalar field to the
vector space is the real numbers.
Let's look at the different operations we can perform with functions. First, there
is addition. In general, when two functions are added together, we get another func-
tion not a real number. Next, we could consider multiplying two functions together.
But, multiplication of two functions usually yields another function. Moving into
the realm of calculus, differentiation could be considered. However, differentiation is
really defined at a point and does not apply to functions that are piecewise smooth
because of the jump discontinuity. Therefore, differentiation would not be a good
candidate for the inner product. Finally, we can consider integration. A definite
integral can be applied to a function on the entire interval and produces an answer
that is a real number. Also, a function that has jump discontinuities, as defined
previously, can be integrated, and we can easily integrate the product of two func-
tions. Thus, integration is a good candidate for the inner product. Therefore, we
define a possible new inner product as
e
(u, v) = f u(x)v(x) dx.
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 93

Note: u and v are no longer boldfaced since they are functions. Thus, not vectors
in the traditional sense.
Now, it must be determined if this definition meets all the criteria of the defin-
ition of an inner product.
First, we must ask, "is

(u,u)=f v.(x)u(r)
e dx = 16 [u(T)]2dz > 0
J

on the interval [a, b]?" We know, from calculus, the square of a function is always
nonnegative. The integral of a nonnegative function is also nonnegative, since the
integral is always nonnegative when the curve of the entire function is above the
x-axis. Therefore, (u, u) > 0 for all functions u(x) in the function space of PWS
functions. Also, if u = 0, then
f6 fb
(u,u) = (0,0) = Odx=J Odx=O.
Ja
If
b b

(u, u) = u(x)u(x) dx = [u(x)] dx = 0,

Ia Ia

then wehave u=0.

Second, we must show (u, v) = (v, u) for any u and v in the vector space. Using
the general theorems of integration, we have
6 /'6
(v., v) _
f u(x)v(x) da = I v(x)v.(x) dx = (v, u)
J

for any u and v in the function space of PWS functions.

Third, we must show (u + v, w) = (u, w) + (v, w) for any u, v, and w in the
function space of PWS functions. Again, using the general theorems of integration,
we have

(u+v,w) /eu(a)w(x)+v(x)w(x)dz
= f [v.(x)+v(x)]w(x)dx= J
6

u(a)w(a) dx + v(x)w(x) dx = (w, w) + (v, w)

Jo J
for any u, v, and w in the function space of PWS functions.
Finally, we must show (cu, v) = c (u, v) for u and v in the function space of PWS
functions and c a scalar in JR. The theorems on general integration provide us with
the solution

(cu, v) = 6 cu(x)v(x) dx = c f b u(x)v(x) dx = c (u, v)

Ja a
94 Chapter 4: The Essentials of Fourier Series

for any u and v in the function space of PWS functions and c a scalar in IIt.
We have shown the definition
b

( u, v) = u(x)v(x) dx
f
is an inner product on the function space of PWS functions. In fact, it is the
standard inner product for the function space of PWS functions, which is analogous
to the dot product in the vector spaces IIYn, for any n e N.
It is convenient to change the notation sightly as in
b

(f,g) = f (x)g(x) dx.

a
This change actually makes the inner product easier to recognize. Also, in the text's
study of PDEs, the interval of greatest importance to us is the interval [-L, L], in-
stead of the general interval [a, b]. Therefore, we change our inner product notation
to
L
(f,g) = f (x)g(x) dx
J-L
in the function space of PWS functions.
Having an inner product for the function space of PWS functions implies a
concept of orthogonality. By the definition of orthogonality, two functions are or-
thogonal if the inner product is zero. This means, if f (x) is orthogonal to the
function g(x), then
rL
(f,g) = I f(x)g(x) dx = 0.
L

x
-L L

Figure 4.5: The graph of 1(x) = -x and g(x) = 2 on the interval [-L, L].

EXAMPLE 4.7. Let f (x) _ -x and g(x) = 2. Then,

f(_x)(2) dx = -2x dx = -x2 I LL = -(L2 - (-L)2) = 0.

L J L
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 95

Figure (4.5) shows the graph of 1(x) and g (x) .

In Example (4.7), the functions f (x) _ -x and g(x) = 2 are shown to be
orthogonal in the function space of PWS functions. We now must find the analogous
orthogonal basis in the function space of PWS functions.
In the vector space IR3, the natural basis, axes, can be defined by the three
vectors
1 0
0 1

0 0

which are perpendicular, orthogonal, to each other. An analogous definition of the

basis of the vector space IIS3 would involve elements from the function space of PWS
functions, and these elements must be functions defined on the interval [-L, L].
The functions that are analogous to the standard basis in Rn, n e N, are or-
thogonal to each other on the interval [-L, L]. However, this is where the analogies
stop. The standard basis in W, n e N for a given n (say n = 15) means there are
15 vectors in the basis. Thus, we know we have a finite dimensional vector space.
The standard basis in the function space of PWS functions is infinite. Thus, our
function space is infinitely dimensional. The questions to be answered are "What
are the functions?" and "How do we get them?"
To answer the first question, the functions a where a is a constant and the
trigonometric functions, sine and cosine form the standard basis for the function
space of PWS functions. However, the argument of the functions is not a simple x.
The argument is n e N. Thus, the standard basis is a, sin and cos
L L L
nEN.
The second question is answered by considering the general ODE

o"(x) = -A(x),
subject to the BCs

dcp(L) dcp(-L
dx dx

on the interval [-L, L]. We must solve Equation (4.2) for A < 0, A = 0, and A > 0.
For A <0, we let -s = A, s > 0. Then, Equation (4.2) becomes

p"(x) = sp(x).
From your previous courses you should recognize this ODE and know that the
solution is dependent on A. First, for A <0, we have the solution

(x) = cl cosh (vx) + c2 sinh (/x). (4.4)

96 Chapter 4: The Essentials of Fourier Series

Using the first BC from Equation (4.3) yields

cp(L) = cl cosh (/L) + c2 sinh (/L)

= cp(-L) = cl cosh (-'L) + c2 sinh (-/L). (4.5)

Since the hyperbolic cosine function is an even function and the hyperbolic sine
function is an odd function, we rewrite Equation (4.5) as

cl cosh (/L) + c2 sinh (/L) = cl cosh (/L) - c2 sinh (/L) ,

which becomes

2c2 sinh (/L) = 0. (4.6)

Equation (4.6) implies c2 = 0. Therefore, Equation (4.4) reduces to

(x)=cicosh(s/x). (4.7)

Using the second BC from Equation (4.3) on Equation (4.7) yields

d(L) = /c1 sinh (/L) = dcp(-L)

= V/c1 sinh (-/L),
dx dx

which can be rewritten as

2/c1 sinh (/L) = 0,

implying c1 = 0. Thus, for A <0, we have only the trivial solution.
Next, we assume A = 0. Then, Equation (4.2) becomes

o"(x) = 0.
Integrating twice yields

cp(x) = dix + d2.

Applying the first BC from Equation (4.3) yields

cp(L) = d1L + d2 = cp(-L) = -d1L + d2. (4.9)

Equation (4.9) can be rewritten as

2d1 L = 0,

which implies d1 = 0. Thus, Equation (4.8) becomes

(x) = d2. (4.10)

 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 97

The second BC from Equation (4.3) indicates that we must find the derivative of
cp(x) = d2. Since d2 is a constant, the second BC is trivially true. Thus, for A = 0
we have (x) equal to the constant d2.
Finally, we assume that A > 0. The solution, in this case, is
(x) = a cos Vx + b sin Vx. (4.11)

Applying the first BC from Equation (4.3) yields

(L) = a cos (v5L) + b sin (v5L)

= cp(-L) = a cos (_vL) + bsin (-vL). (4.12)

Since the cosine function is an even function and the sine function is an odd function,
we rewrite Equation (4.12) as

a cos (v'L) + b sin (vL) = a cos (vL) - b sin (vL),

which becomes
2b sin (/5L) =0. (4.13)

Applying the second BC from Equation (4.3) yields

d(L) _ (b cos (vL) - a sin (vL))
dx

d( -L) L) _ (bcos (_L) - a sin (_L)). (4.14)

dx
Again, since the cosine function is an even function and the sine function is an odd
function, we rewrite Equation (4.14) as

b cos (vL) - a sin (vL)) = v b cos (vL) + a sin (vL)),

which, after some algebraic manipulations, becomes
2a sin /5L =0. (4.15)

Equations (4.13 and 4.15) are very similar. Dividing out the 2a and 2b yields

sin (v5L) = 0
2
for both equations. Solving for A, we find A = (-i--) n = 1,2,3,... Remember,
the sine function equals zero when the argument of the sine function is an integer
multiple of 71. Thus, when A > 0, the solution for Equation (4.2) is
n7rx
cp=
(x) a cos + b sin
L
. (4.16)
L
98 Chapter 4: The Essentials of Fourier Series

We have all the values of A, called eigenvalues, which allow for a nontrivial solu-
tion to Equation (4.2) subject to the BCs given in Equation (4.3). These eigenvalues
are analogous to the eigenvalues in Theorem (16) and Theorem (17). Also, Equa-
tions (4.10 and 4.16) make up the general solution of Equation (4.2). The solution
is made up of three different types of functions. The three different functions are a
constant, cos and sin called eigenfunctions. They are analogous to the
L L
corresponding eigenvectors of Theorems (16 and 17). The eigenfunctions are often
referred to as orthogonal sequences. The orthogonal sequences are cos and
L
n7rx n7rx n7rx
sin . Remember, when n = 0, sin = 0 and cos = 1. Thus, the con-
stant may be included in the cos sequence. However, in your previous course
L
work, the orthogonal sequences were referred to as the fundamental solution set.
We know the functions of a fundamental solution set are linearly independent, and
linear independence is a basic property of any basis, and since they are also known
as orthogonal sequences, we know that they are orthogonal. However, we will not
take for granted that they are orthogonal. This will be proved.
As the previous paragraph suggests, solving the ODE, Equation (4.2), is analo-
gous to solving an n x n matrix for the distinct eigenvalues and the corresponding
eigenvectors. Remember, eigenvalues are found when you solve the matrix equation
Ax = Ax or A - AI x = 0. Also, when eigenvalues for an n x n matrix are found,
the eigenvectors may be determined. By Theorem (17), Eigenvectors form a basis
for the n-dimensional space that is represented by the n x n matrix. The space
we are considering is a function space. Therefore, instead of eigenvectors, we have
eigenfunctions, for the function space formed on the interval [-L, L]. It remains for
us to show they are orthogonal.
To show that the eigenfunctions found are orthogonal on the interval [-L, L], we
must show several things. You must remember there is one constant and an infinite
number of both sine and cosine eigenfunctions, all of which must be orthogonal to
each other. We start by showing that
L
L cosmirx
sinn7rx L dx = 0, n, m E N.
J-L
First, the trigonometry identities
si n ( a + b) = s i n a cos b + cos a s i n b
and
sin (a - b) = sin a cos b - cos a sin b
must be used. We have
n7rx m7rx
s in
(-i- + L si n L cos m7rx
n7rx
L + cosn7rx
L sin .
L ( 4 . 17)

and
nirx nirx
si n L - mirx
L = si n L - cos nirx
L cosmirx L sin mirx
L
,
. ( 4 . 18 )
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 99

Adding Equation (4.17) to Equation (4.18) yields

n7rx m7rx n7rx m7rx n7rx m7rx
sin (-L_ + _---) + sin = 2 sin cos
L
(-_-L-

L L -z
Dividing both sides by two and integrating from -L to L, we find
mirx nirx nirx
1

2 J-L
L
{sin
n'irx
L
+
L
+ sin
L
- mirx
L
dx =
J-L
L
sin
L
mirx
cos L dx.

Integrating yields

2-L(LL)
1
+ sin
nirx mirx
L . nirx L- mirx
L dx

2
fLsin/ (n
L )xdx
L L

11/ -L 1 f n+m 7rl / (n--L m) n-m 71 L

(n + m)) cos
c
2L L Jx om S L x -L

\ ((n + m)ir)L+(n_m)4CO5 /(n - m) Ll

1 ( -L cos
L
-L L J
L

(n+m)n / -L (n-m)n
z
L )(_L)+(n_m)4CO5( L )(_L)]

=0.
Remember, the cosine of the angle equals the cosine of minus the angle.
The rest of the orthogonality integrals are
L nirx mirx
sin-----sin------- dx =0, n, m E N, n m;
-L L L

L
cosnirx
L cosmirx
L dx = 0, n, m E N, n
I L
m;

L n7rx
d2 cos L dx = 0, d2 a constant, n e N, n 0;
-L
and
L
f d2 sin n7rx
L
dx = 0, d2 a constant, n e N, n 0.
L
They are left as exercises.
100 Chapter 4: The Essentials of Fourier Series

We have shown that the eigenfunctions d2, a constant, sin and cos
L L
are orthogonal eigenfunctions. In fact, they form the standard basis of the function
space of PWS functions. Therefore, we are now ready to define a linear combination
of our basis. Note: A linear combination of our basis must represent every function
in the function space of PWS functions.

4.3.2 Definition of Trigonometric Fourier series

The eigenfunctions d2, sin
-- , and cos
--
, n e N, were identified as the standard
L L
basis for the function space of PWS functions on the interval [-L, L]. This can only
be true if every function in the function space of PWS functions can be represented
in a linear combination of sine and cosine functions. Remember, a linear combina-
tion is the sum of the product of the eigenvectors, in this case eigenfunctions, and
a scalar, a constant. Also, we know that there are an infinite number of eigenfunc-
tions in our proposed standard basis. Therefore, a linear combination must include
an infinite sum. Thus, if our standard basis was identified correctly, we have, for
any function 1(x), in the function space of PWS functions, the definition
00
n7rx n7rx
f (x) a0 + ::i: [an cos
n=1
(-z--)
+ bn sin
(--)]
L
(4.19)

where a0, an, and bn are the scalar multipliers of our eigenfunctions and means
approximately. Note: A scalar times a constant is just another constant. Thus, a0
is the product of an arbitrary scalar and the constant d2.
The right side of Equation (4.19) is known as a trigonometric Fourier series. It
is called trigonometric, since sine and cosine functions are used in the sum. Also,
Equation (4.19) is sometimes called an eigenfunction expansion, where sine and
cosine functions are the eigenfunctions. There are other types of Fourier series.
They also form a standard basis but for different intervals/coordinate systems.
However, at this time, there is no need to go into detail about them. They will be
discussed in Chapter 8, when generalized Fourier series are introduced, and used
extensively in Chapter 10. For purposes of this text, the trigonometric Fourier
series above will be called a Fourier series for short. When we introduce generalized
Fourier series, appropriate names will be applied.
Equation (4.19) identifies the linear combination of our proposed standard ba-
sis. The constants a0, an, and bn, are known as Fourier coefficients, and we must
determine them.
For each different function in the space of PWS function, the scalars a0, an,
and bn are found in the same way the scalar multipliers were found in Example
(4.6). In Example (4.6), the scalar multipliers were found by using the standard
inner product for IR, called the dot product. The vector to be represented was
used in the dot product with each eigenvector, yielding the scalar contribution in
the direction of each eigenvector. When the eigenvectors and corresponding scalars
were put in a linear combination, they formed another representation of the vector in
question. Thus, it would seem natural to find the scalars a0, an, and bn in Equation
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 101

(4.19) for each function in the space of PWS functions by using the standard inner
product for the space of PWS functions, integrating the product of the function to
be represented and the eigenfunctions from [-L, L]. The obvious solution is not as
easy as the question. How do we do it?
First, for convenience, the ti in Equation (4.19) will be changed to =. Thus,
Equation (4.19) becomes
00

1(x) = ao + [an cos (nLx) -I- bn sin UL)]

\ (4.20)
n=1

The inner product for the function space requires the integration over the in-
terval [-L, L] of the product of two functions. This means we have to multiply
Equation (4.20) by some function, and the object of the multiplication and subse-
quent integration is the determination of a Fourier Coefficient. This implies using
the orthogonality of the sine and cosine functions to our advantage. Thus, to find
an, for any n e N, we would multiply Equation (4.20) by cos
L
for some (f)
particular m e N, m 0. This yields
mirx 1
f (x) cos ( L x) = ao cos ( L
l

00

+ cos [an cos (nLx) -I- bn sin (nLx) ] (4.21)

n=1
\ Lx / .

After some small algebraic manipulation on Equation (4.21), we get

mirx 1
1(x) cos ( L x) = ao cos ( L
l

00
-I- cos Lan cos (nLx) -I- bn sin (nLx) ] . (4.22)
n=1
\L x1
We integrate Equation (4.22) from -L to L. Thus, we have

L
f (x) cos ( L x) dx = a0 f L cos ( L x) dx
L
+
f L n-1
cos ( L x) [ate, cos (nLx) + sin (nLx) ] dx. (4.23)

The first integral,

'Lmirx
LLcL) dx,
102 Chapter 4: The Essentials of Fourier Series

in Equation (4.23) is in its final form unless we know the function, 1(x). The second
integral,

L mix
cos ( L ) dx,
Jf L
ao

in Equation (4.23) is known to equal zero by orthogonality. The last integral,

(mix 1 rcos (nLx) + bn sin (nLx) ] dx,

[an
L

in Equation (4.23) presents a completely different problem. We must ask ourselves,

"When does the integral of an infinite sum equal the infinite sum of the integrals?"
This question is extremely important, and it will be discussed in more detail in
Chapter 6. For the moment, we will accept as fact the integral can be taken inside
the infinite sum, and we know that the infinite sum of sums is equal to the sum of
the infinite sums. This yields

an L cos (nLx) cos ( L x) dx

n=1 J L
0o L
+ bn sin (nLx) cos ( L x) dx. (4.24)
n=1 J I'
The second integral in Equation (4.24),
(fl1X\)
bn J
L sin cos ( L x) dx,

is known to equal zero by orthogonality. The first integral in Equation (4.24),

an
f cos (nZ)(\)
xcos L xdx,
will, by orthogonality, equal zero whenever m n. However, when m = n, we have
(nirx) 2 /
dx = an LL 2 -I- 2 cos 12 L x ) dx = an L.
an LL cos
Thus, the entire infinite sum in Equation (4.24) becomes one term, and that term
is an L . Therefore,

L nix
f (x) cos L dx = anL.
fL
Solving for the Fourier Coefficient an yields
1 L nirx
an = zf f (x) cos L dx.
L
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 103

In a similar fashion, we find the Fourier coefficients

1 L
a0 2L , fL(x) dx

and

bn = f (x) sin dx.

L Il LL \nLxl
In conclusion, we have the definition of the trigonometric Fourier series repre-
sentation of a function in the space of PWS functions,
00

1(x) = ao -I- [an cos (nLx) -I- bn sin nx

c-z-)]' (4.25)

and the integral equations for determining the Fourier coefficients, which are
L
(4.26)
J
dx,

L f(x)

/'L
an = L f (x) cos (nLx) dx, (4.27)
J L

and
L
(nirx)
bn = L
I L
1(x) sin dx. (4.28)

Thus, for any function, f (x), in the function space of PWS functions on [-L, L],
we can define a Fourier series and determine the Fourier Coefficients. It still must
be determined if a Fourier series actually converges to the function, f (x), in the
function space of PWS functions. This important point is discussed in the next
subsection.

4.3.3 Fourier series Representation of P iecewise

Smooth Functions
This subsection starts with several examples of functions and their Fourier series rep-
resentations. The examples are followed by the general theorem governing Fourier
series convergence, called the Fourier's convergence theorem. We conclude this sub-
section with several theorems developing absolute and uniform convergence of a
Fourier series to the function it represents. The definitions of absolute and uniform
convergence will be provided at that time.
The first example is a Fourier series representation of a well-behaved polynomial,

f(x)=x3-4-7x2-4x-1O.
104 Chapter 4: The Essentials of Fourier Series

Figure 4.6: 1(x) = x3 + 7x2 - 4x - 10 on the interval [-2, 2].

EXAMPLE 4.8. Consider the function 1(x) = x3 + 7x2 - 4x - 10 on the interval

[-2, 2] shown in Figure (4.6). The Fourier series representation of 1(x) is
00
x3 + 7x2 - 4x - 10 = ao + [an cos (nLx) + bn sinnx
(-z-)i'

f2
a° 2L f
,
L

L
f fix) dx = 4
J 2
x3 + 7x2 - 4x - 10 dx = 3
and

11
L
f(x
f () cos os-
nirx
dx

p2
1
(x3 -I- 7x2 - 4x - 10) cos (nLx) dx
2
/2
1 (224 cos (nom) 224 sin (nom) 72 sin (nor)
2 n
n

since (_l)n and sin(ner) = 0 for all n. Also

fL
bn = L 1(x) sin (nLx) dx
J L

= 2I ' (x3 -I- 7x2 - 4x - 10) sin (nLx) dx

 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 105

1 (l92cos(nir) 1 sin (nir) 64 sin (nir)

2 n3ir3

since cos(nir) _ (-1)n and sin(ner) = 0 for all n. Thus, the Fourier series represen-
tation of the function 1(x) is
00
-2
x3 + 7x2 - 4x - 10 =
3
+
n=1 n
112(-1)n
2 2
cos
nirx
L
+
96(-1)n
n3 3 sinnirx
L()j

Figures (4.7,4.8,4.9 and 4.10) are four graphs of the Fourier series of the function
1(x) for the partial sums Si(x), S5 (x), S12 (x), and S25 (x), respectively. Note: The
Fourier series converges to the entire function, 1(x), rather quickly.

_ -2 96(-1)n
Si(x) 3 +
1

n=1 n
112(-1)n
22 cos
nirx
(----)
L + n3 3
sin
nirx
(-h--)

5
(fl1X) (fl7IX)J
cos + 96(313)n sin .

L
n=1

x
x

Figure 4.7: The Fourier series repre- Figure 4.8: The Fourier series repre-
sentation of 1(x) = x3 + 7x2 - 4x -10 sentation of f (x) = x3 + 7x2 - 4x - 10
for the partial sum 81(x). for the partial sum Ss (x) .

-2
12
r112(-1)
cos
/nlrxl + 96( )n sin
Siz(x) -
3
n=1
L n2 l J

-2 25 r112 -1)n
n2 nixl 96(-1) (fl7IX
Szs (x) = 3 + cos (-i l + sin .

n=1
I

L
-z )j
106 Chapter 4: The Essentials of Fourier Series

Figure 4.9: The Fourier series repre- Figure 4.10: The Fourier series rep-
sentation of 1(x) = x3 + 7x2 - 4x -10 resentation of 1(x) = x3 -1-7x2 -4x- 10
for the partial sum 512(x). for the partial sum S25(x).

In the next two examples, the functions chosen demonstrate the action of the
Fourier series at a jump discontinuity. An explanation is provided, then formalized,
in Fourier's convergence theorem.

EXAMPLE 4.9. Consider the function h(x) = x2 -I- 2x - 1 on the interval [-5, 5]
shown in Figure (4.11) . The Fourier series representation of h (x) is

Figure 4.11: h(x) = x2 + 2x - 1 on the interval [-5, 5].

00
nix nix
x2 + 2x - 1 = a0 -E-
n=1
[an cos
5n-
-I- bn sin
---j
5

where

L f5
a° 2L h(x) dx = 10
J 5
x2 + 2x - 1 dx = 3 ,
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 107

an = L J L h(x) cos
(fl7X)
dx = 5 5(x2 + 2x - 1) cos
(nx) dx
J
100 100 48 100(-1 n
= n sin(ner)
and
(fl7IX) nrx
sin dx = 5 J (x2 -I- 2x - 1) sin (-) dx
5

20 20 20(-1)n
= n27t2 sin(ner) - n cos(nir) _ -
nor
n7r
Thus,

x2+2x-1= 2+ 100(-1)nirx
00

n2 cos - - 20(-1)n
n sin
nirxl
5 5
n=1

Figures (4.12, 4.13, 4.14, 4.15, 4.16, and 4.17) are six graphs of the Fourier series
representation of h(x) for the partial sums Si (x), S5 (x), S25 (x), S5o (x), S75 (x), and
S1oo (x), respectively. Note: The partial sums converge to the function h(x) much
more slowly than the partial sums represented by Figures (4.7,4.8,4.9 and 4.10) of
Example (4.8). The reason for the slower convergence of the partial sums are the x-
values, x = -5 and x = 5. At these x-values h(-5) h(5). Thus, the Fourier series
representation of the function does not seem to converge to the values of h(-5) or
h(5). This example demonstrates the action of the Fourier series representation of
a function with a jump discontinuity at the endpoints.

x
x

Figure 4.12: Fourier Series represen- Figure 4.13: Fourier Series represen-
tation of h(x) = x2 -I- 2x - 1 for the tation of h(x) = x2 + 2x - 1 for the
partial sum 81(x). partial sum S5 (x).

1.
22 r100(-1)n nirx 20(-1)
n
Sl (x) = 3 + L 5J
L
108 Chapter 4: The Essentials of Fourier Series

22 . r100(-1)n
SS(x) - 3 + I,
n=1 L
nirx
n
20(-1)n
5
nirx

25
Szs(x) = 3 +
22 r100(-1)n
L
nix n
20(-1)n nirxl
5J

n=1
X100(- 1)1
cos -
nirx
5
20(-1)n
nor
nirxl
sin-I

30

25

20

15

10

x x
-4 2 4

Figure 4.14: Fourier Series represen- Figure 4.15: Fourier Series represen-
tation of h(x) = x2 + 2x - 1 for the tation of h(x) = x2 + 2x - 1 for the
partial sum S25 (x) . partial sum S5o (x) .

35

30
25

20
15

10

x
4 2 4

Figure 4.16: Fourier Series represen- Figure 4.17: Fourier Series represen-
tation of h(x) = x2 + 2x - 1 for the tation of h(x) = x2 + 2x - 1 for the
partial sum S75 (x) . partial sum Sioo (x)

22 + 7s r100(-1)"` nix 20(-1)n nirx

S7s(x) - cos sin 5 I .
3
n=1
L 5 nir
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 109

Sioo(x) = 3 +
22
100
r100(-1)n
L
nix 20(-1)
n 5
nirxl
J

In Figure (4.18), the graph of the Fourier series representation of f (x) over the
interval [-15,15] is shown. We see the Fourier series duplicates itself three times
on this new interval. This is known as the periodic extension of the Fourier series.
Also, we see that the Fourier series representation of 1(x) does converge at f(-5)
and f(5). Although, what it actually converges to is not yet known.

x
15 -10 -5 5 0 15

` 100
Figure 4.18: h (x) = 22
+ Li n=1
100(-1)n
cos - 20(-1)n
n sin 9-] graphed on
the interval [-15,15] .

2.5

1.5

0.5
x
-0.5 0.5

Figure 4.19: The graph of the function g(x) on the interval -1 < x < 1.

Next, we will consider a function that has a jump discontinuity in the interior
of the interval.
EXAMPLE 4.10. Consider the function
1, -1<x<0
g(x) =
2,0<x<1,
shown in Figure (4.19).
110 Chapter 4: The Essentials of Fourier Series

The Fourier series representation of g(x) is

00
nirx nix
g(x) = ao -I- ::i: [an cos -I- bn sin ,
1 1
n=1

where
/'L 1

ao = 2L
L
j(x) dx = 2 f 9(x) dx = 2
1

an L g(x) cos (nLx) dx = 1 J 1 g(x) cos (nix) dx

J-L

3
= sin(nir) = 0,
nir
and

bn L
f Lg(x)sin
(nirx)
dx = 1 J 1 g(x) sin (nlx) dx

1 1 [i_(_i)n]
_ - cos(nir) _
nir nor nor

Thus,

(-1)]n sin (nix)

n

The Fourier series representation of the function g(x) are graphed below for the par-
tial sums Si (x), S1o (x), S25 (x), S5o (x), S1oo (x), and S1ooo (x), respectively. Also,
note that as n increases from 1 to 1000, the graphs of the Fourier series represen-
tation of the function g(x) become increasingly more accurate. Again, please note
that at the x values of -1 and 1, the Fourier series representation of g(x) does not
seem to converge to the points g(- 1) and g(1). Also, at x = 0 the Fourier series
representation of g(x) seems to equal both 1 and 2. In Figure (4.26), the Fourier
series representation of g(x) is graphed over the interval [-3, 3]. Please note that
the Fourier series representation of g(x) repeats itself. Again, this is known as the
periodic extension of the Fourier series.

In Example (4.9), the function, f (x) = x2 + 2x - 1, on the interval [-5, 5] was

a continuous function on that interval. However, f(-5) f(5). If we consider
the graph of the Fourier series representation of f (x) in Figure (4.18), we can see
that at the point (-5, f(-5)) and the point (5, 1(5)), the Fourier series converges
to something.
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 111

2.5

1.

0.5

x x
-1 -0.5 0.5 -0.5 0.5 1

Figure 4.20: g (x) _ 3

Figure 4.21: g(x) _ 3 +
2 2
1- (-1) n n10=1
L iifl1 nor
sin (nirx) on the nor
sin (nirx)] on the

interval [-1, 1]. interval [-1, 1].

1 1

0.5 0.5

x x
-1 -0.5 0.5 1 -1 -0.5 0.5 1

Figure 4.22: g(x) _ 3 + Figure 4.23: g (x) _ 3 +

2 2
251[1-(-1)] sin (nirx)1 on 501-(-1)n

>n=1 n the >In=1 nor Sln (n7rx)j On the

interval [-1, 1]. interval [-1, 1].

3 3

2.5 2.5

2 2

1.5

0.5

x x
-1 -0.5 0.5 -1 -0.5 0.5

Figure 4.24: g(x) = 2+ Figure 4.25: g(x) = 2+

[1-(-1)n] -n=1
1000 [1-(-1)n]
'-` 100
Lin-1 L
n
sin (n7rx)] on the n sin (n7rx)] on the
interval [-1, 1]. interval [-1, 1].
112 Chapter 4: The Essentials of Fourier Series

2.5

1.5

0.5

x
-3 -2 -1 1 2

Figure 4.26: g(x) = 2 + La-i

n
sin (nirx)] on the interval [-3, 3].

In Example (4.10), the function,

1, -1 <x<0
20x,
2, , 0 <<x < <1,
1

on the interval [-1,1 ] has a jump discontinuity at x = 0 and g(- 1) g(1). Again,
if we consider the graph of the Fourier series representation of g(x) in Figure (4.26),
we see that at the points (-1, g(-1)), (0, g(0)), and (1, g(1)), the Fourier series does
converge to something.
The convergent point of the Fourier series representation of a function 1(x) on
the interval [-L, L] with jump discontinuities within the interval is the average of
the left and right limits of the function at the jump discontinuity. For instance, this
means in Example (4.10) where
1, -1<x<0
9(x)
2,0<x<1,
at x = 0, the left limit of the g(0) = 1 and the right limit of g(0) = 2. Therefore,
the Fourier series representation of g (x) at x = 0 converges to the value
g(0) +2 g(0) =1.5.

The Fourier series representation of a function 1(x) on the interval [-L, L],
where f (- L) f (L) , converges to the average of the left and right limits of the
endpoints on the periodic extension of the function. What is the periodic extension
of a function? The easiest way to answer this question is to think of the cosine
function on the interval [-7r, 7r] graphed in Figure (4.27). A periodic extension of
the cos x on the interval [-7r, 7r] is the cos x graphed on the interval [-37r, 37r] shown
in Figure (4.28).
Figure (4.28) repeats the original curve shown in Figure (4.27), once on the
left and once on the right. This is an example of a periodic extension. Ba-
sically, the curve is repeated on the left of the original curve as many times as
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 113

Figure 4.27: cos x on the interval [-7r, 7r].

x
-2. 2.5 5 7.
0.5

Figure 4.28: cos x on the interval [-37r, 37r].

needed to reach negative infinity, and on the right of the original curve as many
times as needed to reach positive infinity. Therefore, in Example (4.9), where
f (-5) f(5), the Fourier series representation of 1(x) converges at x = -5 to
f(-5) + f(-5) _ f(5) + f(-5) _ 34+ 14 = 24 at x = 5 the Fourier series
2 _ 2 _ 2
converges to f (5 ) + f(5) = f (5 ) + f(-5) = 34 + 14 = 24.
2 2 2
Theorem (21) is Fourier's convergence theorem, which was proved by Peter Gus-
tav Lejeune Dirichletl in 1828 in the journal Journal fur die refine and angewandte
Mat hematik.

Theorem 21. (Fourier's Convergence Theorem) If f (x) is an element of the func-

tion space of piecewise smooth functions on the interval [-L, L], then the Fourier
series representation of f (x) converges to

1. the function f (x), wherever f (x) is continuous, and to

2. the average of the left and right limits of the function f (x) at any jump dis-
continuity. That is, the Fourier series converges at a jump discontinuity x = xo to

'P. G. L. Dirichlet (1805-1859) was a very famous mathematician from Prussia.

114 Chapter 4: The Essentials of Fourier Series

f (xo) + f (x)
2

3. if the endpoints, f (-L) and 1(L) are not equal, then the Fourier Series ton-
verges at -L and at L, to f (L
) + f (-L)
2

The proof of Theorem (21) may be found in many different texts on Fourier
series. In particular, Fourier series, by Georgi P. Tolstov, provides a complete
derivation of this theorem and many of the other theorems governing convergence
of Fourier series.
Since the Fourier series of a piecewise smooth function is guaranteed to converge
to the function whenever the function is continuous, or to the average of the left-
and right-hand limits wherever the function has a jump discontinuity, we do not
have to graph the function or the periodic extension of the function before writing
the Fourier series down and determining the Fourier coefficients. This concept is
very important. We now develop the notion of uniform convergence of a Fourier
series to a function.
Understanding how a Fourier series converges requires some knowledge of series
convergence. Hence, we start with two definitions; absolute and uniform conver-
gence. If you have not had a real analysis course, this definition and several other
concepts may be difficult to understand. However, it gives you an idea of conver-
gence and how it must be shown. So let's get started.
Definition 22. An infinite series is said to be absolutely convergent if the series
formed from it by replacing each term by its absolute value is convergent.
Definition 23. Let (fn) be a sequence of functions defined on a subset S E R.
Then, (fn) converges uniformly on S to a function F defined on S, if for each
> 0 there exists a number N such that

I- f(x)I <E
for allxES and alln>N.
This definition gives us a precise way of determining uniform convergence. The
following theorem, called the Weierstrass M-test,2 provides a way to determine if a
sequence of functions converges uniformly. The proof of this theorem may be found
in Analysis with an Introduction to Proof by Steven R. Lay as well as many other
texts on real analysis.
Theorem 24. (Weierstrass M-test) Suppose that (fn) is a sequence of functions
defined on a subset S E R, and (Ma) is a sequence of nonnegative numbers such
that

fn(x)I <M
2Karl Weierstrass (1815-1897) was one of the greatest mathematics teacher of the mid-
nineteenth century. His greatest contributions to mathematics were in the field of power series
representation of a function.
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 115

00 00

for all x E S and all n E N. If Mn converges, then f n converges uniformly

n=1 n=1
on S.
It should be stressed that uniform convergence is much stronger than both point-
wise or mean-square convergence. Although both pointwise and mean-square con-
vergence are important concepts, pointwise convergence is not powerful enough for
our purposes and mean-square convergence is beyond the scope of this text.
We now consider a simple and important fact. It is stated in a theorem and
followed by a short proof.
Theorem 25. Given a trigonometric Fourier series
00
n7rx n7rx
a0 + [an cos + bn sin , (4.29)
n=1
(-IL-) (-IL-)]
which is not assumed to be the Fourier series of any function. If the series
00

converges, the Fourier series, Equation (4.29), converges absolutely and uniformly.

Proof. Since
n7rx n7rx n7rx n7rx
an cos L + bn sin L < an cos L + bn sin L

by triangle inequality and

nzx x I I I I+I I'
an coS I + I bn sin - --) an bn
L L
because the sine and cosine functions are bounded by ±1. We have, by the Weier-
strass M-test, that the Fourier series, Equation (4.29), converges uniformly. E
The previous theorem and the Fourier convergence theorem, Theorem (21) imply
Theorem (26).
Theorem 26. The Fourier series representation of a continuous, piecewise smooth
function f (x) on [-L, L] converges to f (x) absolutely and uniformly.
Theorem (26) is a very important theorem, and it implies that the Fourier series
representation of a continuous piecewise smooth function f (x) on [L, L] is also
continuous. This fact, with the necessary conditions, is stated in the next theorem.
Theorem 27. If f (x) is a piecewise smooth continuous function on the interval
[-L, L] and f (-L) = f (L), then the Fourier series representation of the function
f (x) is continuous on [-L, U.
 116 Chapter 4: The Essentials of Fourier Series

Similar theorems are stated for the Fourier sine and cosine series in the next
section.

EXERCISES 4.3

4.3.1. Show that

nirx mirx
J(L)(L)
L

-L
sin dx = 0, n E N, n m

by direct computation.
4.3.2. Show that
nirx
J(L)
L

-L
cos dx = 0, n E N

by direct computation.
4.3.3. Show that
nirx
J(L)
L
sin dx = 0, n E N
-L

by direct computation.
4.3.4. Show that
nirx mirx
J(L)(L)
L

-L
cos dx = 0, n E N, n m

by direct computation.
4.3.5. Plot the functions 1(x) = x and g (x) = 1 on the interval [-L, L]. Are the
functions 1(x) and g(x) orthogonal? Show by direct computation
pL
f(x)g(x) dx = 0.
J_L

4.3.6. Plot the functions

u(x) = 2 3x2 - 1) and v(x) = 2 5x3 - 3x)

on the interval [-1, 1]. Are the functions u(x) and v(x) orthogonal? Show by
direct computation that
1

u(x)v(x) dx = 0.
11
 Section 4.3: A New Space: Function Space of Piecewise Smooth Functions 117

Also, show by direct computation that

1
2
I [u(x)]2 dx =
-1 5

and
1
2
I [v(x)J2 dx =

4.3.7. Plot the functions

u(x) = 1 35x4 - 30x2 +3) and v(x) = 1 63x5
8 8
- 70x3 + 15x
on the interval [-1, 1]. Are the functions u(x) and v(x) orthogonal? Show by
direct computation that
1

u(x)v(x) dx = 0.

Also, show by direct computation that

1
2
I [u(x)]2 dx =

and

2
[v(x)] dx=-.
z

11

4.3.8. Some functions are orthogonal with a common weight function. For example,
the Hermite polynomials are orthogonal with weight function ex In Chapter
.

8, we discuss weight functions. However, for the purpose of this exercise, we

give you the general formula and ask you to show that the first four Hermite
polynomials are orthogonal to each other. The general formula is
[00
e-xZ Hn (x)Hm (x) dx.
- 00

The first four Hermite polynomials are

Ho (x) = 1,

Hl (x) = 2x,

H2(x) = 4x2 - 2, and

H3(x) = 8x3 - 12x.

 118 Chapter 4: The Essentials of Fourier Series

4.3.9. Show, by direct computation, that the Fourier coefficients a0 and bn have the
formulas
L
1
ao =
2L -L

and

1 L n7rx
bn = L
f L
1(x) sin L dx.

4.3.10. Determine the Fourier series for the following functions on the given bounds:
(1) 5for-2<x<2.
(2) 2x2 - 3x for -7r < x < 7r.
(3) 4 cos 3x for - <x < .
2 2

(4)
1 + cos 2x + sin 5x for
3 3 - x <-
4.3.11. Determine the Fourier series representation for the function 1(x) = 2x - 1
on the interval [-5,5]. Using your favorite mathematical software, plot the
function 1(x) and the Fourier series representation of 1(x) for n = 1 to 5,
n = 1 to 10, n = 1 to 50, and n = 1 to 200. In your own words, explain how
the Fourier series representation of 1(x) converges at the points -5, 0, and 5.
4.3.12. Determine the Fourier series representation for the function g (x) = 3 sin x
on the interval [-7r, 7r]. State in your own words any conclusions you may
determine about the Fourier series representation for the function g(x).
4.3.13. Let

1
'
- 2 <- x <
37r
2

-1, 27r < x < 2 .

Determine the Fourier series representation for the function h(x). Using
your favorite mathematical software, plot the function h(x) on the inter-
- 37r 37r
val 2 , 2 and the Fourier series representation of h(x) on the interval
[-37r, 37r] for n = 1 to 5, n = 1 to 10, n = 1 to 50, and n = 1 to 200. In your
own words, explain how the Fourier series representation of h(x) converges at
3-r
the points - and 0.
2 ' 2

4.3.14. Do you need to solve for the Fourier coefficients before graphing the Fourier
series representation of a function?
 Section 4.4: Even and Odd Functions and Fourier series 119

4.3.15. Suppose 1(x) are piecewise smooth functions on [-L, L]. Prove Bessel's in-
equality:
0o L
a2 + a2 + b2 < 1 f[f(x)]2dx.
n= 1
Bessel's inequality implies the sum of the squares of the Fourier coefficients
of any square integrable function always converge.
4.3.16. Show that the Fourier series
00
n7rx n7rx
a0 + an cos + bn sin
n=1
L L

can be written in the form

00
n7rx
A0 + An cos + 9n ,
L
n=1

where An = an + b. Also, find an expression for Bn in terms of an and b.

4.3.17. Prove Riemann-Lebesgue's lemma: If 1(x) is a piecewise smooth function of
period 27r, then
1
lim
n-400
1(x) sin [(n + 2 x dx = 0.
-7t

Hint: You must use Bessel's inequality.

4.3.18. Prove Parseval's equality: If 1(x) is a piecewise smooth function on [-L, L],
then

L [f (x)]2 dx = 2a+ (an +b)

j' n=1

where an and bn are the Fourier coefficients of 1(x).

4.4 EVEN AND ODD FUNCTIONS AND

FOURIER SERIES
The question to consider next is "How do Fourier series help us solve PDEs, since
the heat and wave equations are developed on the interval [0, U?" This question is
answered in the following discussions.
In Chapters 2 and 3, the equation for heat conduction in a one-dimensional rod
and the wave equation for transverse vibrations in a string were developed. The
complete equations depended on developing sets of boundary conditions. All of the
boundary conditions were stated on the interval [0, L]. Also, at the beginning of
120 Chapter 4: The Essentials of Fourier Series

this chapter, Fourier series were described as useful for solving PDEs like the heat
and wave equations. However, Fourier series were developed on the interval [-L, L].
Therefore, there must be some kind of mathematical translation. This translation
is based on even and odd functions.
The definition for an even function is as follows:
Definition 28. The function 1(x) is an even function on the interval [-L, L] if

f (-x) = f (x)
for all x E [-L, L].

x
-L

Figure 4.29: An even function on the interval [-L, L].

The following definition is for an odd function:

Definition 29. The function f (x) is an odd function in the interval [-L, L] if
f (-x) = -f(x)
for all x E [-L, L].

Figure 4.30: An odd function on the interval [-L, L].

Definitions (28 and 29) explain how to show a function as even or odd. However,
another way of describing a function as even or odd is graphically. The graph of an
 Section 4.4: Even and Odd Functions and Fourier series 121

even function on the interval [-L, L] is symmetric about the y axis. Figure (4.29)
demonstrates this feature.
The graph of an odd function on the interval [-L, L] is not symmetric about
any axis. However, if you perform a 180° rotation of the first quadrant of the graph
of a function, and it is identical to the third quadrant of the graph of the function,
then the function is odd. Figure (4.30) demonstrates this feature.Please note, the
graph of the function in the third quadrant is identical to a 180° rotation of the
graph in the first quadrant.
When two even or two odd functions are multiplied together, the result is an
even function. When an even and an odd function are multiplied together, the
result is an odd function. Also, integration benefits from even and odd functions.
If we integrate an even function, 1(x), from -L to L, then we can simplify the
integral by multiplying it by 2 and integrating from 0 to L, that is,

L f (x) dx = 2 L f (x) dx, f (x) even.

L 0
J J
However, the integral of an odd function, g(x), from -L to L is 0.
In the function space of PWS functions on the interval [-L, L], most functions
are neither even nor odd. Therefore, knowing what even and odd functions are does
not appear to help us. Also, as we said before, the heat and wave equations are
usually given on the interval [0, L]. However, a mathematical technique of extending
a function has been developed. Thus, if a function exists on the interval [0, L], then
we can extend it to the interval [-L, L] by making an even or odd extension. This
means we can choose which extension is convenient for our situation, then expand
it in a Fourier series.

Figure 4.31: 1(x) = x on the interval [0,2].

EXAMPLE 4.11. Show how the function 1(x) = x on the interval [0,2] can be
expanded in a Fourier series.
Solution: There are two choices and both are correct. First, Figure (4.31) shows
the graph of f (x) = x on the interval [0,2]. Next, the function 1(x) is extended
as an even function, which we will call 1(x) _ Ix on the interval [-2, 2]. Figure
(4.32) graphically illustrates the even periodic extension of 1(x) _ x on the interval
122 Chapter 4: The Essentials of Fourier Series

[-2, 2]. The Fourier series representation of 7(x) = x) on the interval [-2, 2] is then
developed. The Fourier series representation of the function 1(x) = x I is

Figure 4.32: 7(x) = xi on the interval [-2, 2].

00
Lnirx nrx
J(x) = x) = a0 + an cos + bn sin
n=1
L

00
nirx nirx
= a0 + an cos + bn sin
2 2
n=1
where
2
ao = 12L-fL (x) dx = -1 xi dx.
L 4 -2
However, 7(x) = x l on the interval [-2, 2] is an even function. Therefore,
2 2
a0=
1

4
fIxl dx= fx dx= 1.
2
1

Also,

an = L
J LL f (x) cos
nLx dx = 2J2 x cos n2x dx.
2
nrx lx) cos nrx on the interval [-2, 2] is an even function; therefore,
7(x) cos -i=
2
2 2
1 nrx 4 ((-1)n -1 )
an = 2 lxi cos 2 dx = x cos 2 dx =
-2 0 (nor)

nrx nix
Since 1(x) sin = lx sin on the interval [-2, 2] is an odd function, the
2 2
integral is zero. Thus,
00 4((_1)n_1)
n=1 (nit)
2 cos 2xn
 Section 4.4: Even and Odd Functions and Fourier series 123

Figures (4.33,_4.34, 4.35, and 4.36) are four graphs of the Fourier cosine series of
the function 1(x) _ x for the partial sums Si (x), Sio (x), S25 (x), and S50 (x),
respectively. Figure (4.37) is the Fourier cosine series representation of 1(x) _ x
for the partial sum S50 (x) on the interval [-8, 8].

Si (x) = 1 + 1 4((-1)-1) cos n7rx

2
(nr) 2
n=1
and
10
4((-1)n - 1) nirx
Sio (x) = 1 + 2 cos
2
n=1 (nit)

x x
-2

Figure 4.33: Fourier cosine series Figure 4.34: Fourier cosine series
representation of 1(x) _ x for the representation of 1(x) _ x for the
partial sum Si (x). partial sum Sio (x) .

2 2

1.5 1.5

0.5 0.5

+- x x
-2 -1 2 -2 -1

Figure 4.35: Fourier cosine series Figure 4.36: Fourier cosine series
representation of 1(x) _ x for the representation of 1(x) _ x for the
partial sum S25 (x) . partial sum S50 (x) .

S25 (x) = 1 +
25
4((-1)-1) coS nix
2

n=1 (nit) 2
124 Chapter 4: The Essentials of Fourier Series

and

S50 (x) = 1 +
50
4((-1)-1) 2 COS
n7rx
(nor) 2
n=1

The interval [-8, 8] was chosen because it shows that the convergence is exact on

Figure 4.37: Fourier cosine series representation of 7(x) _ x on the interval

-8,8].

the interval [-2,2].

Finally, the function 1(x) is extended as an odd function, which we will call 1(x) = x
on the interval [-2, 2]. Figure (4.38) graphically illustrates 1(x) = x on the interval
[-2, 21, and the Fourier series representation of 1(x) = x on the interval [-2, 2] is
then developed. The Fourier series representation of the function 1(x) = x is

x
-2 -1 1

-2

Figure 4.38: f (x) = x on the interval [-2,2].

00
n7rx
f (x) = x = ao + :i: an cos + bn sin
n=1
L L

00
n7rx
= ao + [an cos + bn sin
2 2
n=1
 Section 4.4: Even and Odd Functions and Fourier series 125

Here, f (x) = x on the interval [-2, 2] is an odd function. Thus,

a0
1

2L -L
L
1(x) dx =
1

4
fx 2

2
dx = 0,

and

1 L n7rx 1 2
an = L 1(x) cos L dx = 2 x cos 2 dx = 0.
-L -2
Whereas,
L 2
1 nrx 1 2 nrx nrx
bn = L
f L
f (x) sin L dx = 2
-2
x sin
2
dx =
0
x sin
2
dx

nor

Thus,

°° -4 (-1)n n7rx
f (x) = x = > sin ,
nor 2
n=1

Figures (4.39, 4.40, 4.41, and 4.42) are four graphs of the Fourier sine series of

Figure 4.39: Fourier sine series rep- Figure 4.40: Fourier sine series rep-
resentation of f (x) = x for the partial resentation of f (x) = x for the partial
sum Si(x). sum Sio (x) .

the function f (x) = x for the partial sums Si (x), Sio (x), S25 (x), and S50 (x),
respectively. Finally, Figure (4.43) is the Fourier series representation of f (x) = x
for the partial sum S50 (x) on the interval [-8, 8].

Si (x) _
1
-4 (-1)n
sin
nrx
and S10 (x) _
10 -4(-l)' sin nrx
n7r 2 n=1 n7r 2
n-1
126 Chapter 4: The Essentials of Fourier Series

2 2

1 1

x
1

-2 -2

Figure 4.41: Fourier sine series rep- Figure 4.42: Fourier sine series rep-
resentation of 1(x) = x for the partial resentation of 1(x) = x for the partial
sum S25(x). sum S50(x).

x
-7.5 -2. 5 2.5

Figure 4.43: Fourier sine series representation of f (x) = x on the interval [-8,8].

25
_4 (-1)n 50 -4 (-1)n nirx
S25 (x) _ sin 2and S50 (x) = sin
n7r n=1 nor 2
n-1

The interval [-8, 8] was chosen because it shows the convergence is not exact
on the interval [-2, 2]. At the endpoints, x = -2 and x = 2, the Fourier series
representation of the function, 1(x) = x, converges to 0.

Note: In Figure (4.43), as x approaches -2 or 2 and -6 or 6 the Fourier series

representation of the function, 1(x) = x, seems to oscillate. This is known as the
Gibbs phenomenon. The Gibbs phenomenon is discussed in greater detail at the
end of Chapter 6.
Example (4.11) shows both an even and an odd extension of the function
1(x) = x. The even extension is called the Fourier cosine series representation
of the function, and it is given by
00
n7rx
1(x) = ao + an cos
L
n=1
 Section 4.4: Even and Odd Functions and Fourier series 127

where
11L
ao = L f (x) dx

2 L n' rx
an = L 1(x) cos L dx.
0

The odd extension is called the Fourier sine series representation of the function,
and it is given by
00
n7rx
1(x) _ bn sin
n=1
L

where

bn =
2 JL f (x) dx.
L 0 L

However, which of the two different extensions is the right one to use? This is
generally answered by the format of the problem. In most cases, either the Fourier
sine series representation will be the correct one to use or the Fourier cosine series
representation will be the correct one to use, but not both.
We conclude this section with two theorems on when the Fourier sine and cosine
series representation of a function are continuous. As in Theorem (27), they follow
directly from Theorem (26) stated at the end of the last section. These theorems are
important and will be continually used throughout the rest of the text particularly
in Chapter 6 and 8.
Theorem 30. If 1(x) is a piecewise smooth continuous function on the interval
[0, L], then the Fourier cosine series representation of the function 1(x) is contin-
uous.

Theorem 31. If 1(x) is a piecewise smooth continuous function on the interval

[0, L] and f(0) = 1(L) = 0, then the Fourier sine series representation of the
function f (x) is continuous.

EXERCISES 4.4

4.4.1. Determine the Fourier cosine series for the following functions on the given
interval:
(1) 7for0<x<2ir.
(2) sin x for 0 < x < 0.5.
 128 Chapter 4: The Essentials of Fourier Series

(3) x3 - 2x2 + 1 for 0 < x < ir.

(4) cos x for 0 < x < 3.1.
4.4.2. Determine the Fourier sine series for the following functions on the given
interval:
7for0<x<2ir.
sin x for 0 < x <0.5.
x3-2x2+1 for0<x<7r.
cosx for 0 < x < 3.1.
4.4.3. Given
p"(x) = -Ap(x),
subject to
cp(0) = 0 and p(L) = 0,
find all eigenvalues and eigenfunctions.
4.4.4. Given
p"(x) =
subject to

p' (O) = 0 and p' (L) = 0,

find all eigenvalues and eigenfunctions.
4.4.5. Given

subject to

p' (O) = 0 and p(L) = 0,

find all eigenvalues and eigenfunctions.
4.4.6. Given
p"(x) = -Ap(x),
subject to

cp(0) = 0 and cp'(L) = 0,

find all eigenvalues and eigenfunctions.

 Section 4.4: Even and Odd Functions and Fourier series 129

4.4.7. Find the Fourier cosine series representation for the function 1(x) = x2 on the
interval [0, 1]. Using your favorite mathematical software, plot the function
1(x), the even extension of 1(x), and the Fourier cosine series representation
of the function 1(x). Determine the necessary n, so that the Fourier cosine
series representation of the function 1(x) is accurate.
4.4.8. Find the Fourier sine series representation for the function 1(x) = x2 on the
interval [0, 1]. Using your favorite mathematical software, plot the function
1(x), the odd extension of 1(x), and the Fourier sine series representation of
the function 1(x). Determine the necessary n, so that the Fourier sine series
representation of the function 1(x) is accurate. Compare your results with
those of the previous problem.
4.4.9. Is the Fourier cosine series representation of the function g(x) = x3 on the
interval [0, 3], or the Fourier sine series representation of the function g(x) _
x3 on the interval [0, 3], equivalent to the Fourier series representation of the
function h(x) = x3 on the interval [-3, 3]?
4.4.10. Given the fact that 1(x) is a continuous function on the interval [-L, L],
(1) state the conditions when the Fourier series representation of the function
1(x) is equal to the function 1(x) for all x in the interval,
(2) state the conditions when the Fourier cosine series representation of the
function 1(x) on the interval [0, L] is equal to the function 1(x) for all x
in the interval [0, L], and
(3) state the conditions when the Fourier sine series representation of the
function 1(x) on the interval [0, L] is equal to the function 1(x) for all x
in the interval [0,L].
Chapter 5

Separation of Variables:
The Homogeneous Problem
5.1 INTRODUCTION

So far, we have discussed both the steady-state temperature solution for the dis-
tribution of heat in a rod and d'Alembert's solution for the one-dimensional wave
equation. In this chapter, we introduce a third method for solving PDEs: separation
of variables.
Separation of variables is an important technique to master when studying solu-
tion methods of PDEs, because this technique leads to an infinite series solution, the
Fourier Series solution. However, separation of variables does have its drawbacks.
First, unlike the steady-state temperature solution, which makes no requirement on
the homogeneity of the PDE or boundary conditions (BCs), separation of variables
requires both the PDE and the BCs be homogeneous. Second, in general the spa-
tial variable must have finite boundaries. If the spatial variable has semi-infinite
or infinite boundaries, separation usually does not work. We tackle the problem of
semi-infinite and infinite boundaries in Chapter 11. Finally, separation of variables
requires the PDE and the BCs to be linear. Therefore, this chapter starts with
a brief discussion of linear and homogeneous equations and BCs. To discuss this
properly, we introduce the notion of an operator (don't get nervous, you've already
seen lots of them). Operators are very important to mathematics and are used
extensively by physicists and engineers.
Once we complete our discussion on operators and linear equations, we proceed
with separation of variables. First, we cover separation of variables and the heat
equation, followed immediately by separation of variables and the wave equation.
The similarities between the two sections is quite evident and should be no surprise.
Then, we apply separation of variables to multidimensional spatial problems. We
complete the chapter with another application of separation of variables, Laplace's
equation in Cartesian coordinates.

131
132 Chapter 5: Separation of Variables: The Homogeneous Problem

5.2 OPERATORS: LINEAR AND HOMOGENEOUS

EQUATIONS
You have been working with operators all of your academic life. In first grade, when
you learned about adding two numbers together, you began your study of operators.
The operator in that case was addition, and addition on the real numbers could be
considered a mapping written as + : III x III -* III. Please refer to Appendix D.
Note: III is a vector space. In calculus, you were introduced to two very important
d_ b
operators: They are the differential operator and the integral operator _ dx.
dx a
In both of the calculus operators, the underline, _, holds the place for the function
on which the operation occurs for instance,
d (x2
+2x-1 )
or

/' b

(x + 2x -
a

In the study of ODEs, you learned that a differential operator can look like
d2- d_
a b,
dx2 + dx +
which is the sum of other operators. Again, III is the vector space that is involved
with the two calculus operators and the differential operator.
Judging from these three operators, it would seem reasonable to assume a partial
differential operator could be
a_ a2_
at - ax2
In fact, Equation (5.1) is a partial differential operator known as the heat operator.
Actually, an operator is like a function. However, it is more versatile. An opera-
tor is a mapping of vector spaces or function spaces. This is important to remember
because the objects that operators operate on are, in this course, functions.
Now that we have looked at several different examples of operators, let's discuss
a very special type of operator: the linear operator.

5.2.1 Linear Operators

Definition 32. Let L be an operator. Then, if L has a vector space as it's domain
and the range of L is a vector space over the same field, in our case III, and Lsatisfies

L [au(x, t) + bv(x, t)] = aL [u(x, t)] + bL [v(x, t)]

where a and b are arbitrary constants, then L is called a linear operator.

 Section 5.2: Operators: Linear and Homogeneous Equations 133

Note: Vector space in the previous definition may be replaced by function space.
b
For example, we can show that L = Ja _ dx is a linear operator. This means
that for two different functions 1(x) and g(x) and two constants c and d, we have
b

L [cf(x) + dg(x)] _ [cf(x) + dg(x)] dx.

a

By the property of definite integrals, we know the integral of the sum of two func-
tions is equal to the sum of the integrals of each function. Thus, we have
b b b
f [cf(x) + dg(x)] dx = c f (x)dx + dg(x)dx.
a a a

Applying the property of definite integrals again, we find that the integral of the
product of a constant and a function is equal to the product of the constant and
the integral of the function, resulting in

b c f (x)dx + J b dg(x)dx = c 6 f (x)dx + d 6 g(x)dx.

Ja a Ja Ja
This becomes

cJ fb (x)dx + d b g(x)dx = cL [1(x)] + dL [g(x)].

a Ja
Therefore,

L [cf(x, t) + dg(x, t)] = cL [f(x, t)] + dL [g(x, t)],

b
proving that the integral operator L = Ja _ dx is a linear operator.
The heat operator, Equation (5.1), is also a linear operator. The following is a
proof that the heat operator is linear. It is slightly more complicated than proving
the integral operator is linear.
EXAMPLE 5.1. Consider the operator
a_ a2_
L= at -kax2 - g(x,t)
where the underline (_) indicates where a function would be placed.
To show L is a linear operator, we must show that
0(ciu1 + c2u2) _ 02(ciu1 + c2u2) _g(x,
k t)(ciul + c22G2)
C7t ox2

02(ui)
=c1 -cig(x,t)(ui)+
Ox2
134 Chapter 5: Separation of Variables: The Homogeneous Problem

2
-c2g(x,t)(u2).
a2
Let's start with the left side of this equation and work to the right side. We have
the equation

a(c,u, + c2u2) _ ka2(clu, + c2U2)

_ 9(x, t)(ciui + c2u2)
at axe

a(ciui + c2U2) _ ka2(clu, + c2u2)

_ c19(x t)ul - c29(x, t)u2
at axe
by distribution of multiplication over addition and commutative properties of mul-
tiplication. By the linearity of the differential operator, we obtain

a(clul + C2U2) _ ka2(clu1 + Cau2)

_ c19(x t)ul - c29(x t)u2
at axe

l a(u1)
at
a(u2) _
at
lka2(ui) _ 2ka2(u2) _ c,g(x, t)ul- c2g(x,
ax2 ax2
t)u2.

This equation can be rearranged by the commutative property of addition to pro-

duce

Cl
a("') +CZ a(u2) - a2(u') -cZk a2(u2)
- cig(x, t)ui - c2s(x, t)u2
at at ax2 ax2

2
a(ui)
C1
ax2
-cig(x,t)(ui)+
at

2
a2(u2)
C2 aat ) 2) - cZg(x, t)(u2),

which is the desired result. Thus,

L=
at
-k-x2 - 2

9(x, t)-

is a linear operator.
Understanding a new concept is often easier if a counterexample is given. Con-
sider the following example, also based on the heat operator, a nonlinear heat
operator.
EXAMPLE 5.2. Consider the operator

- () 2

Oax2 + ax
3

- at
 Section 5.2: Operators: Linear and Homogeneous Equations 135

For the operator J, the parentheses, (), indicate a holding place for the function
being operated on. To prove that J is a nonlinear operator, we show that
a(C1u1 + C2u2)
- (c1u1 + c2 U2) a2(C1u1 + C2u2) a3(C1u1 + C2u2)
at ax2 ax3

3(u1)
____
at
- 11 2(u1)
Dx2 ax3

a(u2) a2(u2) + C2 a3(u2)

+C2 at - C2u2
ax2 ax3
Notice the not equal to sign in the previous equation. As in the last example, we
start with the left side and work through the problem. By the distribution property
of multiplication over addition, we find that
a(C1u1 + C2u2) a2(C1u1 + C2u2) a3(C1u1 + C2u2)
(C1U1
at +C2u2) ax2 + ax3

a(Clul +CZU2) _ C2u2)

ac axe

a2(C1u1 + C2u2) a3(C1u1 + C2u2)

(C2U2)
ax2 + ax3
Applying the linearity of the differential operator once, twice, or three times where
needed, and the commutative property, we obtain
a(C1u1 + C2u2) _ (C1U1) a2(C1u1 + C2u2)
at ax2

-(r-ii- .l a2(C1u1 + c2U2) -+-

a3(CLUi + C2U2

a(u1) a2(u1) a(u2) a2(u2)

= Cl - Cl(ClUl) + C2 - C2(ClUl)
at 3x2 at 3x2

+Cla3(ui)
ax2 ax2 ax3 ax3
No matter what algebraic method we apply, we can not rearrange this to be
Cl a(u1) _ Clul a2(u1) + a3(u1)
Cl
at ax2 ax3

a(u2)
+CZ
at ax2 ax3
Thus, J is a nonlinear operator.
Now, having the skill to identify a linear operator only comes in handy if you
understand its placement in a linear equation.
136 Chapter 5: Separation of Variables: The Homogeneous Problem

5.2.2 Linear Equations

A linear equation contains a linear operator, L, operating on some function,
u(x), which is equal to another function, f. The function f (x) can be constant or
a function of x. Written mathematically, this translates to

L [u(x)] = f (x) or L [u] = f. (5.2)

An example is the following heat equation with a source term Q (x, t) :

au(x,t) - Q(x,t).

Here, the linear operator is a partial differential operator giving us a linear PDE.
Another example is the wave equation with a damping function, Q(x, t),

2u(x,t) a2u(x,t)
ate
-c axe
=Q(x,t).

The general linear second-order PDE in two spatial variables is

) y) + d(x, y) y)
a(x,y)a ux'y)
z + b(x, y) a ayax + c(x, y) a uay ,

Ou(x, y)
-- a(.r.. ail + f(x, y)u(x, y) = g(x, y).
ay

A nonlinear PDE is one where one or more of the functions a(x, y), b(x, y),
c(x, y), d(x, y), e(x, y), or f (x, y) in Equation (5.3) is a function of u(x, y). For
example, consider the equation

2, J) + b(x, y) a2u(x, y + (x, J) + d(x, J) a"(x, y)

u(x, y) a2"(
a2u(2,

y)
8x 8y8x 8y 8x

Ou(x, y)
+ e(x, ii) + f(x, y)u(x, y) = g(x, y).
ay
It is a nonlinear equation because a(x, y) in Equation (5.3) has been changed to
u(x, y). For another example, consider

02u(x,
u(x, t) t) - CZ a2u(x, t)
= Q(x, t).
ate axe
02u(x, t)
This equation is nonlinear, since we have the product u(x, t)
t2 Note: The
study of nonlinear PDEs is beyond the scope of this text. However, you still need
.

to be able to identify a PDE as linear or nonlinear.

 Section 5.2: Operators: Linear and Homogeneous Equations 137

1f f = 0 in Equation (5.2), then Equation (5.2) is called a linear homogeneous

equation. The heat equation without a source term,
au(x,t) a2u(x,t)
___ - x (.)
54
is an example of a linear homogeneous partial differential equation. It should be
clear that u(x, t) - 0 is a solution for the linear homogeneous heat equation. This
solution is called the trivial solution, a simple test to determine if a linear equation
is homogeneous. If u(x, t) - 0 is not a solution, then the equation is called nonho-
mogeneous. A solution technique for nonhomogeneous linear PDEs is discussed
in Chapter 7.
Notice that in Equation (5.4) we have a second-order partial derivative with
respect to the variable x. Recalling that a second-order linear homogeneous ODE
has two unique solutions, it would not be surprising to find out that the linear
homogeneous heat equation also has two or more unique solutions, in addition to
the trivial solution. When you studied ODEs, you learned that if there are two
solutions, then a linear combination of those solutions is also a solution. This is a
property of linear operators and is known as the principle of superposition.
Definition 33. Principle of superposition: If u1 and u2 are solutions to a linear
homogeneous equation, then an arbitrary linear combination of them, clue + c2u2,
is also a solution of the same linear homogeneous equation.
The principle of superposition is used extensively when solving PDEs, which will
be pointed out later in this chapter. Also, in the previous definition, it is important
to note that the word "homogeneous" is used. If we have a nonhomogeneous linear
equation, then the principle of superposition does not apply. However, just as you
learned in ODEs, there are methods to solve some nonhomogeneous linear PDEs
that involve the principle of superposition. This subject will be covered in Chapter
7.

Boundary conditions can also be defined as linear and homogeneous. The special
au(0, t) _ 0
BCs studied in Chapters 2 and 3 where u(0' t) = 0 and u(L' t) = 0 or
ax
au(L, t)
and = 0 are obviously linear and homogeneous. It is more difficult to
ax
au (0, t) au (L, t )
determine whether Robin's conditions (where = - hu(0 t) and
ax ax
hu(L, t), h is a positive constant) are linear and homogeneous. To determine if BCs
are homogeneous, set u(x, t) - 0 in the BCs. If you get 0 = 0, then the BCs are
homogeneous. To determine if BCs are linear, consider whether there is a product
of the function u(x, t) with the operator on the function u(x, t). For instance,
au(o,t) __ a2u(0,t)
ax -u(0' t) axe
is a nonlinear BC, whereas
au(o, t) _ _hu(o, t),
8x
 138 Chapter 5: Separation of Variables: The Homogeneous Problem

when h is a positive constant, is linear. In fact, h can be a function of x and t,

h = h(x, t), and the BC will still be linear. In the next section, we will apply what
we have learned about linearity.

EXERCISES 5.2

5.2.1. Show that

a a2
L=
ax
- ax2
is a linear operator.
5.2.2. Consider the equation
a2u a2 u
c (x, t)u + /3(x, t).

ate axe +
(a) Identify the operator.
(b) Show that the operator is linear.
(c) If ,3(x, t) = 0, what can you say about the equation?
5.2.3. Let
8u 8u
L(u)=u--k--.
8x

au
Note: Simply stating that there is a product between u and is not enough.
at
5.2.4. Show that
iaral ia2
L r 8r Lr ar + r2 ae2
is a linear operator.
5.2.5. Consider the equation
a2u 1 a / aul 1 a2u a2u
8t2 r ar r 8r + r2 ae2 + 8z2 .
(a) Identify the operator.
(b) Show that the operator is linear.
5.2.6. Consider the equation
au _ 1 a 2 au 1 a2u 1 a2 u
sin 0 ae2 + r2 .
at r2 ar r ar + r2 sin o
sin2

0 acb2
 Section 5.3: Separation of Variables: Heat Equation 139

(a) Identify the operator.

(b) Show that the operator is linear.
5.2.7. Suppose that ul and u2 are solutions to the linear homogeneous equation
L(u) = 0. Prove that an arbitrary linear combination of ul and u2 is a
solution of L(u). Note: Simply stating the principle of superposition is not
enough.

5.3 SEPARATION OF VARIABLES: THE HEAT

EQUATION IN AONE-DIMENSIONAL ROD
Consider the heat equation for the uniform one-dimensional rod with perfect lateral
insulation and no source term
au a2u
at = k axe
subject to BCs
u(0, t) =o
u(L(L, t)
t) =0
and IC
u(x, 0) = f (x). (5.7)

We want to have a solution for Equation (5.5) subject to the BCs, Equation (5.6),
for any time t and 0 < x < L. This is called solving the initial value problem.
To do this, we first assure ourselves that Equations (5.5 and 5.6) are linear and
homogeneous, using the tests described in the previous section. If Equations (5.5
and 5.6) are linear and homogeneous, we can apply the separation of variables
technique, and we assume the function u(x, t) is a product of two functions, one of
time G(t) and one of space p(x). In other words
u(x, t) =
Then we find the appropriate derivatives. We have
au
= G' (t)'p(x)

and

G(t)cp " (x)

8x
Substituting into Equation (5.5), we have
G'(t)p(x) = (5.8)
140 Chapter 5: Separation of Variables: The Homogeneous Problem

Separating Equation (5.8) so that one side of the equation is in terms of t and the
other side is in terms of x is called separation of variables. After doing so, we have
G'(t) _ p"(x)
5.9
kG(t) p(x)
This may be very difficult to satisfy, since we are saying that a function of time is
equal to a function of space. Or, more precisely, the left side of Equation (5.9) is a
function of time, t, and does not vary with the variable x. However, it is equal to a
function of x, which does not vary with time, t. Thus, both sides of Equation (5.9)
must be equal to the same constant. The constant is called the separation constant,
and we will denote the constant as -A. Thus, we have the equation
G'(t) - o"(x) - (5.10)
kG(t) p(x)
The choice of -A as the separation constant is for convenience. We could have just
as easily have chosen A.
(tt)
Having separated Equation (5.5) into time _ _A) and space
G
G/
" (x) _ -A equations, we now separate the BCs, Equation (5.6). Using our
\o(x)
assumption for u(x, t) = G(t)p(x), we find that
u(0, t) = G(t)p(0) = 0. (5.11)

If G(t) = 0 for all time t, then the solution to Equation (5.11) is trivial. That is,
u(x, t) = 0, which is not a very interesting problem. But since a BC usually refers
to a spatial variable, it makes sense to assume that
'p(0) = 0.

Similarly

u(L, t) = G(t)cp(L) = 0
implies that
p(L) =0.
Thus, our BCs, Equation (5.6), have become

(,p(0) =0
J (5.12)
(L) = 0.

Using Equation (5.10), the equation in time is

G'(t) _ -AkG(t), (5.13)
 Section 5.3: Separation of Variables: Heat Equation 141

and the boundary value problem for the spatial variable is

p"(x) = -Ap(x), (5.14)

subject to the BCs, Equations (5.12). Thus, we have completely separated Equation
(5.5) subject to the BCs, Equation (5.6). Note: The initial condition for the time
problem will be applied at a later stage of the solution process. Therefore, it does
not need to be separated.
We will solve the spatial equation first, though, in general, either way will suffice.
The second, more time-consuming, method would be to solve the time equation first,
but then you'd have to fill in the variable A after solving the spatial problem. Also,
it is important that you develop a standard methodology that always works when
solving PDEs using separation of variables. This helps when we start solving much
more complicated problems using this method.

5.3.1 Spatial Problem Solution

Many texts assume A> 0, since this is the only solution that relates to a physical
situation for Equation (5.14) subject to the BCs, Equation (5.12). In this text,
we do not make this assumption. I feel it is necessary for a student of physics,
engineering, or mathematics to be aware of all solutions and to be able to work
out all solutions for a given situation. Developing this awareness is not an easy
task, but it is a necessary one. Thus, we will not make an assumption about A but
proceed in an orderly fashion to completely resolve Equation (5.14) subject to the
BCs, Equation (5.12).

Case 1: A <0.
In this case it is convenient to assume that A _ -s, where s > 0. Thus, Equation
(5.14) becomes
gyp" (x)
= sp(x),
subject to
p(0) =0
(L) = 0.

We now have a second-order constant coefficient ODE where the coefficient is s > 0.
The solution to this problem is
o(x) = c1e"x + c2e-`fix.

Using the noncomplex form of Euler's equations,

eax - e-ax
sinh ax =
2
142 Chapter 5: Separation of Variables: The Homogeneous Problem

and
eax + e-ax
cosh ax = 2

it is convenient to rewrite the solution as

co(x) = c3 cosh \x + c4 sinh \x. (5.15)

Following are the graphs of cosh x, in Figure (5.1), and sinh x, in Figure (5.2).

Figure 5.1: y = cosh x.

Figure 5.2: y = sinh x.

The graphs point out the advantage of using the functions cosh x and sinh x.
Figure (5.1) indicates that cosh(0) = 1, while Figure (5.2) indicates that only the
sinh(0) = 0. This helps us when we apply the BCs, Equation (5.12), to Equation
(5.15). Applying o(0) = 0 yields
p(0) = 0 = c3 cosh(0) + c4 sinh(0).
 Section 5.3: Separation of Variables: Heat Equation 143

Since sinh (0) = 0 and cosh (0) = 1, we must have c3 = 0. Applying o (L) = 0 yields
o (L) = 0 = c4 sinh L;

since the sinh (x) is 0 only when x = 0, we must have c4 = 0. Therefore, when A <0
there is only the trivial solution to Equation (5.14) subject to the BCs, Equation
(5.12).

Case 2:A=0.
Here Equation (5.14) becomes
o" (x) = 0.

This ODE is solved by integrating twice, which yields

co(x) = d1x + d2.

Applying the first BC, co(0) = 0, we find that

co(0) =0=d2.
Applying the second BC, p(L) = 0, we get
p(L) =0= d1 L,
which indicates that d1 = 0, since L 0. Therefore, when A = 0 there is only the
trivial solution to Equation (5.14) subject to the BCs, Equation (5.12).

Case 3: A> 0.
Equation (5.14) is a second-order ODE with constant coefficients and a solution of
co(x) = hoe-i'x

Using Euler's equations,

eiax - e-iax
sin ax =
2i

and
eiax + e-iax
cos ax = 2

this solution can be rewritten as

co(x) = h5 cos Vx + h6 sin Vx.
Applying the first BC, o(0) = 0, we get
co(0) = 0 = h5 cos 0 + h6 sin 0.
144 Chapter 5: Separation of Variables: The Homogeneous Problem

Since sin 0 = 0 and the cos 0 = 1, we have co(0) = 0 = h5. Applying the second BC,
o (L) = 0, we have
o (L) = 0 = h6 sin V L .
We could always claim that h6 = 0, but this would only give us the unusable trivial
solution again, co(x) = 0. Hence, we must look for other possible solutions. The real
clue comes from considering the function y = sin x shown in Figure (5.3). We see

x
0

Figure 5.3: y = sin x.

that there are numerous places that sin /XL = 0. For instance /XL = 71, L =
271, \/XL = 37r,... , L = n7r. Solving for A in general yields
nor 2
n= L
,n=1,2,3,...
We use the notation ) to indicate an infinite collection of solutions known as
eigenvalues. Corresponding to each eigenvalue is an eigenfunctibn cpn (x) =
hn sin nx = hn sin (--)
n'rxNote that we have changed the coefficient from h6 to
L
hn to indicate that each eigenfunction may have a different coefficient.
Thus, the complete solution to the spatial problem is given by the collection
2
n = CL)
,n=1,2,3,...
nirx
co (x) = hn sin
(az)
Note: This is not a general solution to Equation (5.14) subject to the BCs, Equation
(5.12). A general solution would be a linear combination of all possible solutions.

5.3.2 Time Problem Solution

From the spatial problem, we found that we could only solve the problem for selected
nor 2
values of A denoted by n = , n = 1,2,3,... Thus, Equation (5.13) becomes
L
 Section 5.3: Separation of Variables: Heat Equation 145

the family

G(t) = -AnkGn(t) = - ()2kG(t).

L
From your ODE course, you know that we have solutions
cne-1n,t
Gn(t) = , n = 1,2,3,...

5.3.3 The Complete Solution

We now have the solutions for both the time ODE and the spatial ODE. From our
assumed form of

u(x, t) = G(t)cp(x),

we find an infinite collection of solutions

u(x, t) = sin n = 1, 2, 3, .. .
L
Since cn and hn are arbitrary constants being multiplied together, we can replace
them by B. Therefore,

u(x, t) = Bne_ \nt sin , n = 1, 2, 3, .. .

L
The question now remains as to which n gives the solution we really want. We know
that for n= 1, we have

u(x, t) = B1e- gal t sin ?fx

L

When n=2, we have

t 27rx
u(x, t) = Bee_
sin L ,

which is different from the case n = 1. When n = m, we have

(rn7rx)
u(x, t) = Bme-k\ftt
sin
L
which is different from the previous two cases. In reality, we have an infinite number
of independent solutions for u (x, t). By the principle of superposition, we know that
any sum of independent solutions is also a solution. Therefore, we have
7rx 27rx
u(x, t) = B1e-salt
sin + Bee-Ia2t
sin
L L

Bme- kAm t sin + ... ,

L
146 Chapter 5: Separation of Variables: The Homogeneous Problem

which becomes, using summation notation,

00
Bne-pant flux
u(x, t) = sin (5.16)
L
n=1

From Chapter 4, we recognize this equation as similar to a Fourier sine series.

Having the general solution, we can apply the IC (Equation (5.7)), u(x, 0) =
1(x), to arrive at the specific solution. We have
00
n7rx
u(x, 0) = f (x) = Bn sin (jj_).
n=1

Recall that, in Chapter 4, we learned that any piecewise smooth function can be
represented by a Fourier series for 0 < x < L. Therefore, we can solve for the
coefficients Bn by using the orthogonality of the sine function. They are
L
B=
n
fp f (x) sin (nLx) dx
= f(x)sin-_) dx.
L
fsin2()dx L JoL nLxC

This completes the solution of Equation (5.5) subject to the BCs, Equation
(5.6), and IC Equation (5.7). It describes temperature distribution in the one-
dimensional rod for all time t. In fact, if we take the limit as time goes to infinity
in Equation (5.16), we get the steady-state solution for Equation (5.5). We state
this mathematically as
roo
_ 'n t nurx
lim u(x, t) lim Bne sin =0.
t-+00 t-+00 L
n=1

A fully worked example of the separation of variables method for the heat equa-
tion is provided in Example (5.3).
EXAMPLE 5.3. Find the time-dependent solution for u(x, t) when
au _ a2 u
at -
1 .14 axe (5.17)

subject to

u(o, t)= o
8u(1, t)
8x
=0
and

u(x, 0) = -x2 + 2x. (5.19)

 Section 5.3: Separation of Variables: Heat Equation 147

The PDE, Equation (5.17), and BCs, Equation (5.18), are linear and homogeneous.
Therefore, separation of variables technique is valid. Letting u(x, t) = G(t)(p(x),
yields the time equation
G'(t) = -1.14AG(t), (5.20)

and the spatial equation

(p"(x) = -A(p(x), (5.21)

subject to the BCs

(5.22)

When solving the spatial equation, we must determine the valid values of A. Thus,
we first assume A <0. In this case, let A = -s and solve
p "(x) = scp(x),

subject to the BCs, Equation (5.22). This yields

co(x) = c1 cosh /x + c2 sinh \x. (5.23)

Applying the first BC, cp(0) = 0, to Equation (5.23) yields c1 = 0. Applying

the second BC, p' (l) = 0, to Equation (5.23) yields c2 = 0. Thus, there are no
eigenvalues for A < 0. Next, we consider A = 0. Here, Equation (5.21) becomes
(p" (x) = 0,

which, after integrating twice, has as a solution

(o(x) = c3x + C4. (5.24)

Applying the first BC in Equation (5.22) to Equation (5.24) yields c4 = 0. Applying

the second BC, p' (l) = 0, to Equation (5.24) yields c3 = 0. Therefore, there are
no eigenvalues for A = 0. Finally, we assume A > 0. In this case, the solution to
Equation (5.21) is

co(x) = c5 cos \/5x + c6 sin \/5x. (5.25)

The first BC, cp(0) = 0, applied to Equation (5.25) yields c5 = 0. The second BC,
p' (l) = 0, indicates either c6 = 0, in which case we only have the trivial solution,
or the eigenvalues

(2n-1)ir 2
An= 2 ,n=1,2,3, .. (5.26)
L

with corresponding eigenfunctions

(2n-1)irx
cpn (x) = cn sin (5.27)
2
148 Chapter 5: Separation of Variables: The Homogeneous Problem

Substituting the eigenvalues from Equation (5.26) into Equation (5.20), we deter-
mine the solution for the time equation, which is
Gn (t)
= dne-1.14Ant

Thus, the solution for u(x, t) is expressed as

00
(2n - 1)?fx
u(x, t) = Gn (x) _ bne sin
2
n=1
We determine the constant bn by using the IC, Equation (5.19), and applying or-
thogonality. We have
00
(2n-1)7rx
u(x, 0) = -x2 + 2x = bn sin
2
n=1

Therefore, remembering L = 1, we find that

Figure 5.4: The graph of u(x, t) for 0 < t < 2.

2 1 (2n - 1) rx 32
bn = - (-x 2+ 2x) sin dx =
1 0 2 r
(2n -1)33
Replacing the values for bn in the Fourier sine series yields the specific solution for
u (x, t) , which is

°° 32 (2n-1)7rx
u(x, t) _ I e-1'14ant sin
n-1 (2n - 2
 Section 5.3: Separation of Variables: Heat Equation 149

A graph of the solution is shown in Figure (5.4) for the partial sum of n = 25 with
0 < t <2. Also, the initial temperature distribution is outlined on the graph by a
boldface line. It is interesting to note from Figure (5.4) how fast e-1'14ant reduces
the initial temperature distribution to the approximate steady-state solution.

EXERCISES 5.3

5.3.1. Separate the following PDEs into appropriate ODES:

au(x,t) = kax,t)
(1) +u(x,t).
at ax
8u (x, t) _ k82u( t) _ m 8u(x, t)
2,

(2) + u(x, t).

at ax ax
t) _ ax [k(x)tht)l
(3) + u(x, t).
J

5.3.2. Consider the PDE

au(x, t) _ a2u(x, t)
at 4 8x2
For each set of BCs and ICs, solve the initial value problem; using your favorite
mathematical software, graph the solution for 0 < t < 5. Clearly indicate the
initial temperature distribution on your graph.
u(o, t) = o
(1) Bcs: and IC: u(x, 0) = x2 - 27rx.
{ 0

(2) Bcs: and IC: u(x, 0) = sin x.

{ u(271, t) = 0
8u(0, t)
=0
8x
(3) BCs: and IC: u(x, 0) = cos x.

0
(4) Bcs: and IC: u(x 0)= x3 - - x2
2
=0
5.3.3. Consider the PDE
au(x, t) a2u(x, t)
at axe
 150 Chapter 5: Separation of Variables: The Homogeneous Problem

(a) Solve the initial value problem subject to

BCs and IC: u(x, 0) = 7 sin 3x - 2 cos 2x.

au(-7r, t) au(7r, t)
ax ax

(b) In five sentences or less give a physical description of this problem.

5.3.4. Consider the following information:

(a) heat conduction in a one-dimensional rod of length 2 m

(b) perfect thermal insulation on the lateral sides,

(c) right end is held at a constant temperature of 0 degrees,
(d) left end has perfect thermal insulation,
(e) thermal diffusivity of 1.7 cm2/sec (thermal diffusivity of silver),
(f) no internal heat source.
(1) Set up the mathematical model.
(2) Give a series solution of your mathematical model.
(3) Check to see that this solution satisfies the equation, the boundary con-
ditions, and the initial conditions.
(4) Using your favorite mathematical software, graph the solution. Then
write a short paragraph discussing the surface at the following times:
t = 1 sec, t = 5 secs, and t = 10 secs.
5.3.5. Let a metallic rod 20 cm long be heated to an initial temperature which is
77rx °
modeled by x sin (-20 --)) C. Suppose that at t = 0 the ends of the rod are
plunged into an ice bath of 0°C, and thereafter maintain this temperature.
Also, suppose no heat is allowed to escape from the lateral surface of the rod.
Note: You must model and explain your choice of boundary conditions for
this experiment.
(1) Solve the initial value problem for a rod made of copper, cast iron, and
asbestos. The thermal diffusivity of these materials may be found in
Appendix E.
(2) Using your favorite mathematical software graph the solution and the
initial condition, then determine the approximate solution, for the fol-
lowing three cases:
(a) Use only the first term in the series for u(x, t) to find the approximate
temperature at x = 7 cm when t = 12 secs for each material.
(b) Use the first three terms in the series for u(x, t) to find the approxi-
mate temperature at x = 7 cm when t = 12 secs for each material.
Compare this answer with the previous answer.
 Section 5.3: Separation of Variables: Heat Equation 151

(c) Use the first 25 terms in the series for u(x, t) to find the approxi-
mate temperature at x = 7 cm when t = 12 secs for each material.
Compare this answer with the previous two answers.
(3) Use the first five terms of the series to determine the amount of time,
t, it takes for each material to reach a temperature of 0°C at the center
of the rod. Compare the amount of time it takes for each material, and
write a short essay on the reasons why the time is different or the same.
5.3.6. Consider a uniform one-dimensional rod of length L, without an internal heat
source, which is not laterally insulated. (Heat can flow in and out across
the lateral boundary.) By experimentation, you discover that heat is flowing
across the lateral boundary at a rate proportional to the difference between
the temperature u(x, t) and the surrounding medium that is kept at 0°C.
(1) Given that the convection constant of proportionality is greater than 0
and the ends of the rod are not insulated and held at 0°C, set up the
mathematical model.
(2) Suppose the initial temperature of the rod is x when 0 -< x < L
2 and

x - L when L < x < L. Give a series solution of your mathematical

model.
2- -
(3) Check to see if this solution satisfies the equation, the BCs, and the ICs.
5.3.7. Consider a uniform one-dimensional rod of length L that has no internal
heat source and is not laterally insulated. (Heat can flow in and out across
the lateral boundary.) By experimentation, you discover that heat is flowing
across the lateral boundary at a rate proportional to the difference between the
temperature u(x, t) and the surrounding medium that is kept at uo degrees.
(1) Given the convection constant of proportionality is greater than 0 and
the ends of the rod are not insulated and held at uo degrees, set up the
mathematical model.
(2) Solve the steady-state temperature distribution.
(3) Suppose after steady-state is reached, the rod is suddenly plunged into
a 0°C bath. State the mathematical model governing this situation and
give a series solution of your mathematical model.
(4) Check to see if this solution satisfies the equation, the BCs, and the ICs.
5.3.8. Consider a uniform one-dimensional rod of length L, that has no internal
heat source and is not laterally insulated. (Heat can flow in and out across
the lateral boundary). By experimentation, you discover that heat is flowing
across the lateral boundary at a rate proportional to the difference between the
temperature u(x, t) and the surrounding medium that is kept at uo degrees.
(1) Given the convection constant of proportionality is greater than 0 and the
ends of the rod are perfectly insulated, set up the mathematical model.
 152 Chapter 5: Separation of Variables: The Homogeneous Problem

(2) Solve the steady-state temperature distribution.

(3) Suppose after steady-state is reached, the rod is suddenly plunged into a
0°C bath. State a the mathematical model governing this situation and
give a series solution of your mathematical model.
(4) Check to see if this solution satisfies the equation, the BCs, and the ICs.
5.3.9. Consider a thin metallic rod made of cast iron (thermal diffusivity of 0.12)
of 3 m length. Suppose the rod does not have lateral insulation and the
temperature of the medium in which the rod is held is maintained at 0°C.
Given that the boundaries are also held at 0°C and the initial temperature
distribution is 1 x3 + 2 x2 - 3x> and there is no source term complete the
following:

(1) Set up the mathematical model.

(2) Solve the initial value problem for your model and graph the solution for
time, t = 3. Also, plot the initial temperature distribution on the graph.
(3) Solve the steady-state problem for your model.
(4) For what time t does the initial value problem approximate the steady-
state solution?
5.3.10. Consider a thin metallic ring made of copper (thermal diffusivity of 1.14).
Suppose the ring has lateral insulation. Determine the solution to the initial
value problem if the initial temperature distribution is x2 and there is no
source term.
5.3.11. Consider a thin metallic ring made of aluminum (thermal diffusivity of 0.86).
Suppose the ring does not have lateral insulation and the temperature of the
medium the ring is held in is maintained at 0°C. Determine the solution to
the initial value problem if the initial temperature distribution is an arbitrary
function of x and there is no source term.
5.3.12. Consider
au(x, t) a2a(2,t)
= 0.245 +x2-8x,
subject to
a-g
e_xsin sin 8
BCs:
{ and IC: u(x, 0) = x- 8 x.

(1) Give a physical interpretation of this problem.

(2) For very large time find the following:
(a) the solution.
(b) the heat energy generated per unit time inside the entire rod,
 Section 5.4: Separation of Variables: Wave Equation 153

(c) the heat energy flowing out of the rod per unit time at each end,
and
(d) the relationship between parts (b) and (c).
(3) Suppose that after Part (5.3.3) has completed, the heat source is turned
off.
(a) Find the equations that describe the mathematical model at this
time.
(b) Using the equations that you just found, solve the heat flow problem
for any time t.
(c) Using your favorite mathematical software, graph the solution for
0<t<5.
5.3.13. Diffusion through a Membrane: Fick's Law can be used to derive an analogue
of Ohm's Law for a membrane of thickness, L, with different chemical concen-
trations on each side of the membrane. If the medium is isotropic (diffusion
occurs the same regardless of the direction of the measurement), then we get
ac a2 c

at - D axe'

subject to
c(0, t) = C1, the chemical concentration on the left of the membrane
and
c(L, t) = Cr, the chemical concentration on the right of the membrane.
(1) Find the time-dependent solution.
(2) Show that as we let t -- oc, we get the same steady-state solution as in
Chapter 2 Section 2.5.1

5.4 SEPARATION OF VARIABLES: THE WAVE

EQUATION IN AONE-DIMENSIONAL STRING
Consider the wave equation for a perfectly elastic vibrating string with no external
forces
a2u 2
a2u
T= c (5.28)
ate
ax2'

subject to the BCs

u(0, t) =o
(5.29)
t)
u(L(L,t) = = 00 ,
1 Adapted from James Keener and James Sneyd, Mathematical Physiology, ©1998 by Springer-
Verlag, New York, pp. 36-38. Reprinted by permission.
154 Chapter 5: Separation of Variables: The Homogeneous Problem

and with ICs

u(x,0) = f(x)
(5.30)
8u(x, 0)
at - 9(x)

As in the previous section, we want a solution for Equation (5.28) subject to the
BCs, Equation (5.29), for any time t and 0 < x < L. To obtain the solution, we
must assure ourselves that Equations (5.28 and 5.29) are linear and homogeneous
using the techniques developed in Section 2. If Equations (5.28 and 5.29) are linear
and homogeneous, we again assume the solution function u(x, t) can be written as
the product of two functions, one of time G(t) and one of space p(x). In other
words,

u(x,t) = G(t)(x).
We then find the appropriate second partial derivatives,

a2u
= U" (t)cp(x)
ate

and

a2 u
.
ax 2
Now substituting into Equation (5.28), we obtain

G"(t)5p(x) = (5.31)

Equation (5.31) is now ready for separation of variables, where one side of the
equation is in t and the other side of the equation is in x. Again, we'll set the
equation equal to a separation constant, as we did in the previous section, resulting
in
G"(t) - cp"(x) - (5.32)
c2G(t) p(x)

Using Equation (5.32), the equation in time is

G"(t) _ -Ac2G(t), (5.33)

and the equation in space is

p"(x) = -A(x). (5.34)

As in the previous section, we have separated our PDE into two ODEs, one
ODE in time and one ODE in space. Now let's separate the BCs, Equation(5.29).
 Section 5.4: Separation of Variables: Wave Equation 155

Using our assumption for u(x, t) = we find that

u(0, t) =

G(t) = 0 for any time t, then our solution is trivial. Also, a BC usually refers to
a spatial variable; therefore, it makes sense to assume that

(0) = 0.

Similarly,

u(L, t) = G(t)cp(L) = 0,

implying that

cp(L) =0.

Thus, our BCs, Equation (5.29), have become

0
(5.35)
(L) = 0.

We have now completely separated Equation (5.28) and BCs, Equation (5.29),
into the boundary value problem

(5.36)

and a time problem

G"(t) = -Ac2G(t).
Note: The initial conditions will be applied after a general solution to the problem
is found. Thus, they do not need to be separated.

5.4.1 Spatial Problem Solution

The spatial problem, Equation (5.36), is identical to that of the previous section
and has solution
2

CL)
n = 1,2,3,...
n'rx
Pn (x) = hn sin L
156 Chapter 5: Separation of Variables: The Homogeneous Problem

5.4.2 Time Problem Solution

From the spatial problem, we found that an =
(),
nor
L
2
n = 1,2,3,... This changes
Equation (5.33) into
G(t) (rllr)2 c2G(t).
L
You have seen this problem before, in your ODE course, so we know this has the
solution

n t) = an cos n = 1,2,3,...
L L

5.4.3 The Complete Solution

As in the heat equation example, we have the solution to the time ODE and to the
spatial ODE. From our initial assumption

we find that for n = 1, 2, 3, ... ,

(nlrct)] nlrx
u(x, t) = [an cos L -+- bn sin L hn sin L

Since an, bn, and hn are arbitrary constants and hnan and hnbn are the products,
we will let An = hnan and Bn = hn b. Thus, we have for n = 1, 2, 3, ... ,
nlrx
u(x, t) = LAn cos
L
+ Bn sinL ()] sin L

Again the question remains, which n gives the solution we really want? We know
that for n = 1, we have
rrct\
u(x, t) = I Al cos I Lt I + Bl sin sin (--)
L \ / \ /J
When n = 2, we obtain
/2irct\ 2irct
r
u(x, t) = I AZ cos I L I + BZ sin
L \
L sin2irx
L
which is different from the case n = 1. When n = m, we have

u(x, t) = [Am cos

mrrct
L + Bm sin
()] mlrct
L sin
mlrx
L

which is different from the previous two cases. Again, we have an infinite number of
independent solutions for u(x, t). By the principle of superposition, we know that
 Section 5.4: Separation of Variables: Wave Equation 157

any sum of independent solutions is also a solution. Therefore, using summation

notation, we have
00
r nlt x
u(x, t) = An cos
n7r ct
L
+ Bn sinn7rL()]
ct
sin L
n=1

which we can identify as similar to a Fourier sine series.

To get a specific solution that satisfies our problem, we now use the ICs. To
satisfy the first IC in Equation (5.30), which is the initial displacement of the string,
we set the following:
00
n7rx
u(x, 0) = f (x) = >An5fl (__).
n=1
L

Thus 1(x) is set equal to a Fourier sine series. Now to find An, we use the orthog-
onality of the Sine function, resulting in

=
ff(x)sin()dx =f f (x)
L
L
L
sin n7tx
L dx.
fo sin2 ()dx 0

To satisfy the second IC in Equation (5.30), which is the initial velocity of the
string, we set

au(x, 0)
= >BTh (

Just like f (x), g(x) is set equal to a Fourier sine series. Now, to find Bn, we again
use the orthogonality of the sine function, resulting in
1L
Bn
CL/cL 0
9(x) sin (nLx) dx.

Solving for Bn, we find that

B= nor2 g(x) sin l

;;J/l
LL

This completes the solution of Equation (5.28) subject to the BCs, Equation
(5.29), with ICs, Equation (5.30). We provide the following example for the wave
equation.

EXAMPLE 5.4. Consider

a2 u a2 u
=4 (5.37)
Dt2 axe
158 Chapter 5: Separation of Variables: The Homogeneous Problem

subject to
8u(0, t) _
8x (5.38)
u(1, t) = 0
and

u(x, 0) = x3 + 2x2 - 3
(5.39)
8u(x, 0) _ x-
2 1.
at -
Find a time-dependent solution for u(x, t). Since the PDE in Equation (5.37) and
BCs, Equation (5.38), are linear and homogeneous, separation of variables technique
can be applied. Assuming u(x, t) = we have the spatial equation

d'(x) = -A(x), (5.40)

subject to

,d(0) = 0
(5.41)

and a time equation,

G"(t) _ -4AG(t). (5.42)

Working with the spatial equation first, it can be determined that for A < 0 and
A = 0 there is only the trivial solution. For A> 0, we have

cp(x) = Cl cos v'5x + c2 sin /5x. (5.43)

Applying the first BC in Equation (5.41), we find c2 = 0. Applying the second BC

yields eigenvalues,
2
(2n - 1)ir
n= 2 ,n=1,2,3, ...
L

with corresponding eigenfunctions,

(2n - 1)irx
cps (x) = cncos
2

Substituting an into Equation (5.42) yields

G( t) = -4A Gn (t),
 Section 5.4: Separation of Variables: Wave Equation 159

which has as a solution

Gn(t) = an cos [(2n - Dirt] + bn sin [(2n - 1)irt].

Thus, the general solution for u(x, t) is

00

u(x, t) = {Ancos [(2n - 1)art] + B nsin [(2n - 1)art]llI cos (2n -

2
n=1

where An = ancn and Bn = bncn. Applying the IC, u(x, 0) = x3 + 2x2 - 3, yields
00
(2n - 1) rx
u x 0 = x3 2x2 -3= A cos
2
n=1

Using orthogonality, we determine An, which is

1
(x3 (2n -
An = 2 2x2 - 3) cos dx
2

-2 (96 + 80ir(-1)1(1 - 2n))

(2n - 1)44
au(x, 0) _ x2 - 1, yields
Repeating the process with the second IC
at
5

Figure 5.5: The graph of u(x, t) for 0 < t < 5.

 160 Chapter 5: Separation of Variables: The Homogeneous Problem

2 (2n - 1)irx
B= (2n - 1)
1
f (x2 - 1) cos 2 dx

(5.45)
(2n - 1)44 '
Therefore, the specific solution for u(x, t) is

°O (2n -
u(x, t) _ {An cos [(2n - Bn sin [(2n - cos 2
n=1

where An and Bn are given by Equations (5.44 and 5.45) respectively. Figure (5.5)
is the graph of u(x, t) for the partial sum of n = 15 with 0 < t < 5. Please look
carefully at Figure (5.5). The boundary at x = 1 is fixed at 0. Whereas, the
boundary at x = 0 is freely oscillating. This is exactly what the BCs indicated.
Also, the initial displacement of the string is indicated by a boldface line.
The one-dimensional problems are useful for developing Fourier Series Solutions
to a PDE. However, more interesting problems arise in the multidimensional spatial
variable PDEs, which are discussed in the next section.

EXERCISES 5.4

5.4.1. Determine if the following PDEs are separable. If so, separate the PDEs into
appropriate ODEs. If not, explain why.
C72u(x, t) a2u(x t)
(1) =C 2 v +2l(x t) , .
ate axe
a2u(x, t) 2 t
mau'x, t>
m ax +u(x , t)
(2)
at2 - c ax 2 - .

(3) c(x)p(x)

5.4.2. Consider the PDE

tax
a2u(x, t) _ a
ae- [T'1)] -
au(x, t)
ax at + u(x, t).

a2u(x, t) _ a2u(x, t)
ate - 9 axe
For each set of BCs and ICs, solve the initial value problem. Using your
favorite mathematical software, graph the solution for 0 < t < 2. Clearly
indicate the initial displacement on your graph.

u(0, t) = o u(x, 0) = rx -
(1) BCs: and ICs:
8u(x, 0) _ 0
u(rr, t) = 0
at
 Section 5.4: Separation of Variables: Wave Equation

u(0, t) = 0 u(x, 0) = sin x

(2) Bcs: ?r t and ICs:
au au x 0)
= 0.
I ax
au(o, t)
=0 ( u(x, o) = o
dx
(s) Bcs: and ICs:
8u(x, 0)
8u(ir,t)
-0 = cos x
8x
at
8u(O,t) _ 0 u(x,0) = cosx + 1
(4) BCs: 8x and IC:
clu(x,U)
t) = 0 = 3x3 - 3irx2.
at
5.4.3. Consider the PDE
a2u(x, t) _ a2u(x, t) au(x, t) au(x, t)
ate - axe - at + ax
(1) In five sentences or less, give a physical description of this problem.
(2) Solve the initial value problem subject to

u(O,t) = 0 ( u(x,0) = cos x

BCs: and ICs: 2

8u(x, 0) _ 2
uI2,tI=0 8t

5.4.4. A uniform string with mass density 0.03 lbs/ft, and tension 300 lbs, is fixed
at the left end and has a freely moving right end. The string has length 20 ft
and is initially at rest, with linear displacement from 0 to 1.
(1) Model this problem mathematically. Note: You must model and explain
your choice of boundary conditions.
(2) Solve the mathematical model.
(3) Write the solution with all known quantities substituted into it.
(4) Check to see that this solution satisfies the equation, the boundary con-
ditions, and the initial conditions.
(5) Using your favorite mathematical software, graph the solution for 0 <
t <5.
5.4.5. Consider the following information:
(a) a perfectly flexible string of length 2ir ft,
(b) tension of 50 lbs/ft,
(c) mass density of 0.02 lbs/ft,
 162 Chapter 5: Separation of Variables: The Homogeneous Problem

(d) vibrating motion that is entirely vertical,

(e) fixed left end at 0,
(f) free right end,
(g) initial displacement of x for 0 <x < it and of 2ir - x for it <x < 2ir,
(h) initial velocity of 0, and
(i) no gravity effects.

(1) Set up the mathematical model.

(2) Solve your mathematical model for any time t.
(3) Check to see that this solution satisfies the equation, the BCs, and the
ICs.
(4) Using your favorite mathematical software, graph the solution. Then
write a short paragraph discussing the surface at the following times:
t = 1 sec, t = 5 secs, and t = 10 secs.

5.4.6. Consider a slightly damped vibrating string with a restoring force that satisfies
a2 u _ 252u au
ate c axe - a - /3u, where a and 3 are constants.
at
(1) Explain why a > 0.
(2) Explain the action of the restoring force.
(3) Find a series solution subject to

u(x,0) = f(x)
u(o> t)= o
BCs: J and ICs:
8u(x, 0)
u(L, t) = 0
l at
Explain your choice of the magnitude of a and 3 as you solve this prob-
lem.

5.4.7. A uniform string with fixed ends is excited by the impact of a rigid plane
hammer, which gives it the following initial distribution of velocities:

8u(x, 0)
at - 225 , 7r < x < 27r,

0, 2ir -< x < 3ir.

Find the vibrations of the string, if the initial displacement was zero.
 Section 5.4: Separation of Variables: Wave Equation 163

5.4.8. A uniform string with fixed ends is excited by the impact of a rigid sharp
hammer, which gives it the following initial distribution of velocities:
0' 0 < x < 7r
au x 0
(
at
) = 225 cos(x - r )' it-2 <x < 3ir2
2

-3ir<x
0'2 - <2r.
-
Find the vibrations of the string, if the initial displacement was zero.
5.4.9. Consider longitudinal vibrations of a uniform flexible rod with free ends. De-
termine the following:
(1) The mathematical model if the initial displacement and velocity are ar-
bitrary functions of x in the longitudinal direction.
(2) Solve the initial value problem.
Note: Consider the possibility of uniform linear motion of the rod for
the entire problem.
5.4.10. A uniform string with fixed ends has an initial displacement of 2x for 0 <
x < it and 3ir - x for 7r < x < 3ir. The initial velocity is zero and the string
is vibrating in a medium that resists the vibrations. (The medium produces
a resistance proportional to the velocity.) Suppose the resistance constant of
proportionality is 0.01. Find the solution to the initial value problem.
5.4.11. A uniform string with a fixed end at 0 and free end at 2ir has an initial
3zr 3zr
displacement of -x for 0 < x < and 3x - 67r for < x < 27r. It is known
- 2 2 - -
that the initial velocity is zero and the string is vibrating in a medium that
resists the vibrations. (The medium produces a resistance proportional to the
velocity.) Suppose the resistance constant of proportionality is 0.03; find the
solution to the initial value problem.
5.4.12. A uniform string with free ends has an initial displacement of
-x,0-<x<-1

u(x,0)= x-,1x<3
-2x+8,3<x<4.
The initial velocity is zero, and the string is vibrating in a medium that
resists the vibrations. (The medium produces a resistance proportional to the
velocity.) Suppose the resistance constant of proportionality is 0.13; find the
solution to the initial value problem.
5.4.13. This problem develops the mathematical model for longitudinal vibrations of
a gas in a tube. Consider an ideal gas performing small longitudinal vibrations
164 Chapter 5: Separation of Variables: The Homogeneous Problem

when it is enclosed in a cylindrical tube. We know that plane cross sections,

consisting of particles of the gas, are not deformed, and that all the gas parti-
cles move parallel to the axis of the cylinder. Determine the boundary value
problem for the density p, the pressure p, the velocity potential of the gas
particles, the velocity v, and the displacement u(x, t) if the ends of the tube
are closed by a rigid impermeable surface.

5.5 THE MULTIDIMENSIONAL SPATIAL PROBLEM

We have covered separation of variables for both heat conduction in a one-dimensional
rod and transverse vibrations of a one-dimensional string. These are the simplest
examples of PDEs. This section is devoted to building on this knowledge of PDEs
by considering the multidimensional problem.
An example of a multidimensional problem is the two-dimensional heat problem
in a uniform rectangular plate, with no internal source and perfect lateral insula-
tion. What does perfect lateral insulation on a rectangular plate mean, physically?
Consider a sheet of paper the rectangular plate. The perfect lateral insulation would
cover the front and back of the sheet of paper.
The equation for the problem just described is
au Du 02u\
(5.46)
at - ax2 + a 2
subject to any necessary BCs and ICs. It is quite easy to see that the heat equation
in a uniform three-dimensional parallelepiped (rectangular box), with no internal
source and perfect lateral insulation is
au a2u a2u a2u
(5.47)
at = ax2 + a 2 + az2
Again, this equation would be subject to the appropriate boundary and initial
conditions. What would the equation be for a four, or even a five, dimensional
parallelepiped? Or, for that matter, a circle, a sphere, or an ellipsoid? These
equations become quite messy to write, so mathematicians and physicists developed
the operator V2, called Del squared, from the operator V, called Del. In calculus
III, you learned Del is the vector,

8x
V=
a
L ay J
for a two-dimensional system. Thus, V2 = V V or
rai
ax
a
ax a2 a2
V2=V.V= =ax2 a2 ,
5115
ayJ LayJ
 Section 5.5: The Multidimensional Spatial Problem 165

which is the dot product of V with V.

The operator V2 handles the dimensions of Equations (5.46 and 5.47). For
example, we can rewrite Equation (5.46) as
au
= kV2u. (5.48)
at
Here,

a2 a2
V2 = + ay2
ax2

Equation (5.48) is the compact form of Equation (5.46).

For another example, Equation (5.47) has the same compact form as Equation
(5.46), which is Equation (5.48) with the added dimension. The expanded V2
operator becomes
a2 52 a2
V2 =
C axe + aye + az2 /
In this section, the V2 operator is not used much since we deal with uniform
two- and three-dimensional rectangular-shaped plates for both the heat and the
wave equations. However, in later chapters, this operator becomes very important,
and you need to get accustomed to it.
Consider the problem of heat conduction in a uniform rectangular plate with
no internal source, perfect lateral insulation, a prescribed temperature of 0° on the
sides x = 0 and x = L, perfect insulation on the sides y = 0 and y = H, and
an initial temperature distribution that is a function of x and y. This problem is
mathematically modeled as
au 82u 82u
= kV2u = k (5.49)
at C ax2 + ay2
subject to the BCs
u(0, y, t) =0
u(L, y, t) =0
au(x, 0, t) _ 0 (5.50)
ay
au (x, H, t)

and IC

u(x, y, 0) = f (x, y). (5.51)

166 Chapter 5: Separation of Variables: The Homogeneous Problem

Equation (5.49) and BCs, Equation (5.50), are linear and homogeneous. Hence,
the separation of variables technique can be applied as a solution method for this
problem. As in the one-dimensional problem, we assume that

u(x, y, t) = G(t)cp(x, y),

so that Equation (5.49) becomes

1 dG ia2 a2
kG dt cp (c9x2 +

where -A is the separation constant. Also, the BCs are separated and become

p(0, y) =0
p(L, y) =0
8cp(x, 0) (5.52)
0
8y

8cp(x, H)
0 .

8y

Thus, we have a time problem,

dG
= - AkG
dt '

and a spatial problem,

(5.53)

subject to the BCs, Equation (5.52).

The spatial problem is still a PDE, where the PDE and the BCs are linear and
homogeneous. Therefore, we can apply the separation of variables technique. If we
assume that p(x, y) = X (x)Y(y), then we can rewrite Equation (5.53) as

d2X d2Y
Y + X d 2 = - AX Y. (5.54)
dx2 y

Separating Equation (5.54) yields

l d2X _ l d2Y _
-T,
X dx2 - Y dy2

where -T is the new separation constant. Again, we must separate the boundary
 Section 5.5: The Multidimensional Spatial Problem 167

conditions, and they become

=o
X(L) =o
dY(0)
=1)

H)
Ii dy
0.

We now have two spatial problems. One in terms of X (x),

d2X
dx2 = -TX, (5.55)

subject to the BCs

(5.56)
X(L)
(L) = = o0;
the other in terms of Y(y),
d2Y
(5.57)
y
subject to the BCs

0
dy
(5.58)
dY(H
dY(H)
0.
d dy

5.5.1 Spatial Problem for X (x)

We recognize that Equation (5.55), subject to the BCs, Equation (5.56), has eigen-
values
nor 2
Tn =
L
,n=1,2,3,... (5.59)

and eigenfunctions
X (x) = bn sin n= 1,2,3,...
L
Thus, the complete solution for the spatial problem for X (x) is
2
Tn =
\L)
,n=1,2,3,...
nlrx
Xn (x) = bn sin
(-z)
168 Chapter 5: Separation of Variables: The Homogeneous Problem

5.5.2 Spatial Problem for Y(y)

Letting _ (A - T), Equation (5.57) becomes

d2Y
dy2 = -Y,
subject to the BCs, Equation (5.58); from previous work, we recognize that the
eigenvalues are
mir 2
m=
(-)
H
,m=0,1,2,3,... (5.62)

and the eigenfunctions

may,
Y(y) = am cos m=0,1,2,3,... (5.63)
H
Therefore, the complete solution for the spatial problem for Y(y) is

m=0,1,2,3,...
Ym,(2,/) = p,m, cos H
C may I

Since our original separation constant is in terms of A, we must determine A

from Equations (5.59 and 5.62). Thus, the eigenvalues are
nor 2+ ml r\2
Anm =Tn - m =
L H '
n= 1'2'3'...' m=0'1'2'3'...
Note that Anm has subscripts of n and m. The reason is that Anm is made up of
Tn and m

5.5.3 Time Problem

We have
dG __
Anm kG.
dt
From previous work, we know the solution is

Gnm(t) = cnme-Anmkt. (5.64)

5.5.4 The Complete Solution

We know that cpnm (x, y) = Xn (x)Ym (y);
nirx
Xn (x) = bn sin L , n = 1, 2, 3,... ;
 Section 5.5: The Multidimensional Spatial Problem 169

and
miry
Ym(y) = am cos , m=0,1,2,3,...
H
Therefore, we have for n = 1,2,3,... and m = 0,1,2,3,...,
nirx miry
(p(x, y) = Cnm sin L cos H , (5.65)

where
Cnm = bn am .

Also u(x, y, t) = Gnm(t)cpnm(x, y) Thus, using Equations (5.64 and 5.65) we get
for n = 1,2,3,... and m = 0,1,2,3,...,
nirx miry
u(x, y, t) = (5.67)

Anm = Cnm Cnm , (5.68)

which is an infinite collection of solutions.

In previous sections of this chapter, we used the principle of superposition to
give a simple summation of all the solutions that would give the correct answer.
Equation (5.67) has a slightly different problem. That is, there are two indices, n
and m. Let's consider different possible solutions for u(x, y, t). We have
rx mgr
u( x, y, t) = Alme-Ai,nkt sin cos y, m = 0, 1, 2, 3, ... ,
L H
which, using the reasoning of previous sections of this chapter, naturally leads to
00
irx may
u(x, y, t) = sin Alme-Almt cos . (5.69)
L m=o H
We could also have
2?rx Hmay
u(x,y,t) = A2me-)\2m kt sin L cos m=0,1,2,3,...
which leads to
00
2irx m ay
u (x,y, t) = sin A2m e-A2mkt cos (5.70)
L
m=o
H
Using the principle of superposition on both, we have
00
ir y
u(x, y, t) = sin Alme-)\1mlct
cos
L
m=o
H

00
2irx may
- sin A2me -A 1mkt cos .
L
m=o
H
170 Chapter 5: Separation of Variables: The Homogeneous Problem

Expanding this process on the cases n = 1,2,3,... gives

00
irx Alme-almkt may
u(x, y, t) = sin cos
L
m=o
H
00
may
+ sin A 2m cos
L
m=o
H
00

+ ... + sin
njrx Anme-Anmkt cos may -+... (5.71)
L
m=o
H

But since equation (5.71) is an infinite sum of an infinite sum, it could be easily
written as
00 00
n?rx may
u(x, y, t) = sin cos
n=1
L m=0 H

or in the more usual form of

00 00
miry n?Cx
u(x, y, t) = Anme -anm1t cos sin
n=1 m=0
H L

00
nirx
= Ano e -An0kt sin
L
n=1

00 00
miry nirx
+ Anme coS -fl- sin , (5.72)
L
n=1 m=1

which is the complete general solution to our problem.

To determine the specific solution, we apply the IC, u(x, y, 0) = f (x, y), to
Equation (5.72). In doing so, we get
00 00 00
nirx + may nirx
u(x, y, 0) = f (x, y) = Ano sin Anm cos sin
n=1
L
n=1 m=1
H L

This leaves us the task of determining the equations for the coefficients Anm and
Ano
Let's first determine the equation for the coefficient Anm. In other words, we
consider the case when n = 1, 2,3,. . . and m = 1, 2, 3, ..., which is

ux ,y )=f( ,y)=(5.73)
0 x
00

n=1 m=1
00

A cos
may
H
sin
L

Note: To do this we are actually applying the orthogonality of the sine and cosine
functions.
 Section 5.5: The Multidimensional Spatial Problem 171

In Equation (5.73), the inner sum actually equals a function in y for each value
of n. That is,
00
miry
B(y)= = >Anmcos--.
H
(5.74)
m=1

Thus, Equation (5.73) may be rewritten as

00
nirx
f (x, y) = >Bn(y)sin__.
n=1
L

Since f (x, y) is now set equal to a Fourier sine series, we can use the method that
we learned in Chapter 3 to solve for Bn (y). The solution is
nirx
B(y)=-L2 L
f (x, y) sin
L
dx.
0

Thus, we have Bn (y) equal to a Fourier cosine series. Next, solving for Anm yields
H
2 miry
Anm = H Bn (y) cos dy,
0 H
and on replacing Bn (y) with its integration equation, we get
2 H2 L nirx miry
Anm f (x, y) sin L dx cos dy
H 0 L 0 H

4 HL miry nirx
f (x, y) cos sin L dxdy. (5.75)
HL 0 0 H
Likewise we arrive at
H L
_ 2 nirx
f (x, y) sin L dxdy. (5.76)
Ano LH J 0 0

This completes the answer to the problem at the beginning of this section, giving
us
00
nhrx
u(x, y, t) = Anoe sin
n=1
L

00 00
miry nirx
+ Anm e - fin" cos sin
n=1 m=1
H L

where
2
HL nirx
Ano = f (x, y) sin L dxdy
LH 0 0
172 Chapter 5: Separation of Variables: The Homogeneous Problem

and
H
4 L miry nirx
Anm = HL f (x, y) cos sin L dxdy.
0 0 H
This completes the solution of Equation (5.49), subject to the BCs, Equation (5.50),
with IC, Equation (5.51). See below for a worked example for a multi-dimensional
problem.

EXAMPLE 5.5. Find the time-dependent solution for

a2 u a2 u a2 u
9 axe + a y2 (s.77)
ate

subject to
u(0, y, t) = 0

u(2, y, t) = 0
(5.7s)
u (x, 0, t) = 0

u(x, 3, t) = 0,

with
0<x<1 0<y<1.5
0<x<1 1.5<y<3
u(x, y, 0) = f (x, y) (5.79)
1<x<2 0<yG1.5
1<x<2 1.5<y<3,
and
au(x, y, 0) _ 0
at
The PDE, Equation (5.77), and BCs, Equation (5.78), are linear and homogeneous.
Therefore, separation of variables technique is valid. Letting u(x, y, t) = G(t)Sp(x, y)
and substituting into Equation (5.77) and Equation (5.78) yields the time equation

G"(t) = -9AG(t),
and the spatial equation

y) y)
axe + aye ) _ -ap(x, y), (5.82)
 Section 5.5: The Multidimensional Spatial Problem 173

subject to the BCs

(p(O,y) =0
cp(2, y) =0
(5.83)
p(x, 0) =0
o(x, 3) = 0.

The spatial equation, Equation (5.81), subject to the BCs in Equation (5.83), is a
linear and homogeneous system. Hence, separation of variables technique is again
valid. Letting Sp(x, y) = X (x)Y(y), and substituting into Equation (5.81) and
Equation (5.83), yields the spatial equation
X"(x) _ -TX (x), (5.84)
subject to the BCs
X(0) = o
J (5.85)
X(2) = o.

We know from Section 5.3, Equation (5.84), subject to the BCs, Equation (5.85),
has a nontrivial solution only when 'r> 0. For 'r> 0, the eigenvalues are
n?C
'rn= (---)
2
2
,n=1,2,3,... (5.86)

with corresponding eigenfunctions

Xn (x) = an sin 2 n = 1, 2, 3, .. . (5.s7)

The separation of variables technique also yields the spatial equation

Y"(y) _ -(A - T)Y(y)> (5.88)

subject to the BCs

(5.89)
Y(3) =0 .

Letting _ (A - z) in Equation (5.88) yields

I'"(y)
(y).
_ -I' (5.90)

Again, we know from Section 5.3, Equation (5.90), subject to the BCs, Equation
(5.89), has a nontrivial solution only when > 0. For > 0, the eigenvalues are
m?t 2
gym= 3 ,m=1,2,3,..., (5.91)
174 Chapter 5: Separation of Variables: The Homogeneous Problem

with corresponding eigenfunctions

may
Ym (y) = bm sin , m=1, 2, 3,... (5.92)
3
Using Equations (5.86 and 5.91) and the relationship _ (A - r), yields
Amn_ nor 2 mgr 2

2 +(-)
3 ,
(5.93)

for n = 1, 2, 3, ... and m = 1, 2, 3, ... Substituting Amri into Equation (5.81)

gives us the solution to the time equation, which is
Gmn (t) = Cmn cos (3/t) + hmn sin (3/t). (5.94)

We now can put together the total solution for u(x, y, t). Remember Sp(x, y) _
X (x)Y(y), so we have
u(x, y, t) = Spmn (x, y)Gmn (t) = Xn (x)Ym (y)Gmn (t) _
00

n=1 m=1
00

[Amn cos (3/t) + Bmn sin (3\/t)] sin miry

3
sin
nirx
2
(5.95)

Applying the IC, u(x, y, 0) = f (x, y), yields

u(x,y,0) = f (x,y) =
00 00

n=1 m=1
A mn sin
miry
3
sin
-
nirx
2
(5.96)

Using orthogonality, we determine Amn, which is

3 2
2 miry nirx
Amn = 3 f (x, y) sin sin 2 dx dy
JJ
0 0
3

96 mir nit
sin sin (5.97)
m2 n2 ir4 3 2
au(x, y, 0) = 0
Repeating the process with the second IC , , Y fields
at
00 00
au(x, y, 0) nirx
at
=0 = [3/Bmn] Sin miry
3
Sin
2
n=1 m=1
which implies Bmn = 0. Therefore, the complete solution is
00 00
miry nirx
u (x, y, t) _ Amn cos 3 Amn t sin sin
3 2
n=1 m=1
where Amn is given in Equation (5.97). Since we can't present four-dimensional
graphs, we graph the solution for the partial sums n = 25 and m = 25 and several
different times, t. Figure (5.6) depicts the surface u(x, y, t) at time t = 0. Here,
the four faces of the figure indicate the initial conditions. Figure (5.7) shows the
surface of u(x, y, t) at time t = 0.2. The third graph, Figure (5.8), portrays the
surface u(x, y, t) at time t = 0.4. Finally, Figure (5.9) depicts u(x, y, t) for time
t=0.8.
 Section 5.5: The Multidimensional Spatial Problem 175

Figure 5.6: The graph of u (x, y, t) for Figure 5.7: The graph of u (x, y, t) for
time t = 0. time t = 0.2.

2 2

Figure 5.8: The graph of u(x, y, t) for Figure 5.9: The graph of u(x, y, t) for
time t = 0.4. time t = 0.8.

EXERCISES 5.5

5.5.1. For the following PDE, separate the PDE into its respective ODES. Also,
completely separate the BCs,
a2 u a2 u au a2 u
C7t2
2
= C ax2 b
t +aa
y
2,

subject to BCs
 176 Chapter 5: Separation of Variables: The Homogeneous Problem

5.5.2. For the following PDE, separate the PDE into its respective ODES.

a2u 2 a2u a2u a2u au au

ate - c C axe + aye + az2 - ax + ay

5.5.3. Consider a thin rectangular plate of length L = 27r meters and width W = 47r
meters with perfect lateral insulation. Find the temperature distribution in
the plate given the following:

(1) The rectangular plate is made of silver (he thermal diffusivity of silver
may be found in Appendix E) and is subject to the BCs

u(0, y, t) = 0

au(L, y, t) _ 0
ax

with the IC

cos(xy) - 1; 0<x<L
2
0<y<W
- 2

W
y cos(x - L);
2 <x<L
L
- - 0y
<- <
u(x, y, 0) _
xcos(y - W);
-0x<2 -- 2
y<W
2

W
cos((x - L)(y - W));
2 <x<L
L
- - 2<--<W.
y

(2) The rectangular plate is made of granite (the thermal diffusivity of gran-
ite may be found in Appendix E) and is subject to the BCs

u(L, y, t) = 0

u(x, 0, t) = 0

au(x, W, t)
,
ay
 Section 5.5: The Multidimensional Spatial Problem 177

with the IC
W
0<x< L3 0<
_y < 4

_W . 0<x< L W < <W

- 3 4
_y_
-L); 3 <x<L 0<yG 4

-L)(y-W); 3 <x<L 4 <yGW.

(3) The rectangular plate is made of brick (the thermal diffusivity of brick
may be found in Appendix E) and is subject to the BCs

with IC
u(x,y,0)=cos(3x) 0<x<L 0<y<W.
5.5.4. Consider a thin rectangular plate of length L = m and width W = 7r m,
2
which offers no resistance to bending. Find the time-dependent solution given
the following conditions:
(1) The plat is subject to the BCs
u(0, y, t) = 0

u(L,y,t) = 0
u(x, 0, t) = 0

u(x, W, t) = 0,

with the ICs

u(x, y, 0) = 0,
178 Chapter 5: Separation of Variables: The Homogeneous Problem

and

xy; 0<x<L2 0<-y <W2

x(y - W); 0<x<L2 W<

2 y<W
--
au(x, y, 0)
at
(L - x)y; L<x<L
2 - - 0<y<W
- 2

W
(L-x)(y-W); L2 <x<L
- - 2< --
y<W.

(2) The plate is subject to the BCs

r au(0, y, t)
ax
au(L, y, t) _ 0
ax
au(x, 0, t) _ 0
ay

u (x, W, t) = 0,

with the ICs

x2y 0<x<L2 0<-y <W2

W < <W
u(x, y, 0) =
2 -y-
W
2 - 2Lx)y;
2 <x<L
L
- - 0<y<
- 2

-2Lx)(W-y); 2 <x<L W <y<W,

and

au(x, y, 0) _ 0
at
 Section 5.5: The Multidimensional Spatial Problem 179

(3) the plate is subject to the BCs

with the ICs

y sin x; 0<x<L
- - 0<-y <W2
u(x, y, 0)
W
(y-W)sinx; 0<x<L
- - 2 <-y-
<W
and

x(cosy+l); 0<x< 2 0<y<W

au(x, y, 0)
at
(x-L)(cosy+1); 2 <x<L 0<y<W.
5.5.5. Consider a situation with the following characteristics:
vertical vibrations in a rectangular membrane,
membrane length of 37r m,
membrane width of it m,
initial velocity of x sin y,
initial displacement of 0,
tension of 4 kg/m,
mass density of 0.3 kgs,
fixed on two sides meeting at the point (0, 0), and
frictionless moving sides meeting at the point (37r, 7r).

Set up the mathematical model.

Solve your mathematical model for any time t.
Using your favorite mathematical software, graph the solution. Then
write a short paragraph discussing the surface at the following times:
t = 1 sec, t = 5 secs, and t = 10 secs.
 180 Chapter 5: Separation of Variables: The Homogeneous Problem

5.5.6. A granite brick of length x = L, width y = W, and height z = H is initially

at the uniform temperature of f (x, y, z) degrees C. Suppose that at time t = 0
the sides z = 0 and z = H are cooled to 0°C while the other sides are perfectly
insulated, and thereafter maintained in that fashion. Find the temperature
distribution in the brick at any time t. Assume L = 27rft, W = 7r ft, H = ft,
2
and f (x, y, z) = 1- cos z - sin z. Note: The thermal diffusivity of granite may
be found in Appendix E. Graph the solution for several different times t.

5.5.7. Consider a thin vibrating rectangular uniform membrane of length m and

2
width m. Suppose the sides x = 0 and y = 0 are fixed and that the other
2
two sides are free. Given an initial velocity of zero and initial displacement
of f (x, y) _ (sin x) (sin y), determine the time-dependent solution and graph
the solution for several different times t.

5.5.8. Consider a perfect laterally insulated thin sheet of glass Pyrex with length
37r ft and width 27r ft. Suppose the sides x = 0 and x = 37r ft are perfectly
insulated and the other two sides are held at zero degrees. Find the solu-
tion if the initial temperature distribution in the sheet is given as f (x, y) _
(cos x) (sine y). Note: The thermal diffusivity of glass Pyrex may be found in
Appendix E.

5.6 LAPLACE'S EQUATION

Laplace's equation is an extremely important equation in mathematical physics.
It naturally arises in electrostatics, steady-state temperature field, magnetostatics,
and potential flow of an incompressible liquid. In this section, we derive Laplace's
equation from electrostatics. Then we determine the solution for Laplace's equation
in the Cartesian coordinate system.

5.6.1 An Electrostatics Derivation of Laplace's Equation

Coulomb's Law the force of attraction or repulsion between two point charges is
directly proportional to the product of the charges and inversely proportional to
the square of the distance between them is the starting point for the study of
electrostatics. It directly leads to two important differential equations describing
an electric field, E. The first is the differential form of Gauss's law, which is

(5.98)

where p is the external charge density. Mathematically, Equation (5.98) is the

divergence of the electric field. The second equation is the curl of the electric field
as a function of position. It is

VxE=0. (5.99)
 Section 5.6: Laplace's equation 181

From Equation (5.99) we know that E is the gradient of a scalar function, known
as the scalar potential, I. Thus,
E=-V. (5.100)

Substituting Equation (5.100) into Equation (5.98) yields

which may be written as

v2 = -p. (5.101)

Equation (5.101) is known as Poisson's2 equation. In regions of space where

there is no charge density, p, the scalar potential satisfies Laplace's equation,
v2 = 0. (5.102)

Since a scalar potential is used to develop Poisson's and Laplace's equations,

they are sometimes referred to as the potential equations. (Note: In mathematics,
we usually use the scalar function u instead of the potential scalar notation of ).
Thus, Equation (5.102) becomes

V2u = 0. (5.103)

Note: Poisson's and Laplace's equations are time-independent second-order PDEs

in two or more spatial dimensions.
Equation (5.103) has the following form in Cartesian coordinates,
a2u a2u a2u
(5.104)

5.6.2 Uniqueness of Solution

We now consider the uniqueness of the solution for Laplace's equation. First, we
need to define a harmonic function and state a different form of the maximum
principle in R2, which you first encountered in Chapter 2. For those of you who
have already had a complex analysis course, the definition of a harmonic function
is quite familiar.
Definition 34. A harmonic function is a function which solves Laplace's equa-
tion.
Theorem 35. (Maximum Principle) Let SZ be a bounded set in R2. Let u (x, y) be a
harmonic function in SZ, while u(x, y) must be continuous in the union of SZ and the
boundary of SZ, denoted D. Then, the maximum and minimum values of u(x, y)
are attained on the DSZ, unless u(x, y) is identically equal to a constant.
2Simeon-Denis Poisson (1781-1840) was a French mathematician who once stated that life is
good for to things: mathematics and teaching it.
182 Chapter 5: Separation of Variables: The Homogeneous Problem

(Note: The maximum principle is easily extended to the space Rn). Proof of
Theorem 35 may be found in Appendix B. However, a common sense approach
gives us the idea behind the maximum principle.
If we consider u(x, y) in Laplace's equation as the steady-state temperature
distribution in a plate, then u(x, y) can't be greater at one point in the plate than
all the other points of the plate because Fourier's heat conduction axioms say that
heat diffuses from high to low. Thus, if one point in the plate is hotter than the
rest, then the heat must flow away from that point to the surrounding points, thus,
reducing the temperature at the hot point. However, this would mean that the
temperature would change with time, a contradiction to the steady-state nature of
Laplace's equation. The same reasoning applies to a minimum point.
Having stated the maximum principle, we can state and show uniqueness of
Laplace's Equation in I12 for the Dirichlet problem. The Dirichlet problem is
2
a2 u a2 u
u =axe -}- 2= 0 in SZ,
ay
subject to
u(x, y) = f (x, y) on aSZ,

where 1 is as given in the previous theorem. The strategy of the proof is to assume
there exist two different solutions. Then, show that the solutions must be equal to
each other.
Proof Suppose there exists two solutions to the Dirichlet problem stated earlier,
u1(x, y) and u2 (x, y). Then let v(x, y) = u1(x, y) - u2 (x, y). Thus, v(x, y) is also
a harmonic function in 1 and zero on 9, because u1 (x, y) = f (x, y) = u2 (x, y)
on Dft Also, v(x, y) is continuous on 1 n aSZ since both u1(x, y) and u2(x, y)
are continuous and the sum of continuous functions is continuous. Hence, by the
maximum principle, v(x, y) must attain its maximum and minimum values on 1K.
Thus, v(x, y) = 0. Therefore, v(x, y) = u1(x, y) - u2 (x, y) = 0, which implies
u1(x, y) = u2 (x, y).

Now that we have shown that the solution to Laplace's equation is unique, we
solve Laplace's equation in Cartesian coordinate system.

5.6.3 Laplace's Equation in Cartesian Coordinate System

Consider the problem of steady-state temperature distribution in a rectangular plate
with perfect lateral insulation, no source, and length L and height H. Also, suppose
the boundary conditions are nonhomogeneous Dirichlet conditions. Figure (5.10)
shows the indicated conditions. The mathematical formulation for this problem is
the following:

_ y) a2u(x, y)
V2u(x,
(x, y) _ u(x, y) - a2u(x,
axe
+ a2
y
__
0 (5.105)
 Section 5.6: Laplace's equation 183

u (x, H) =f4(x)

u(O, y) -.f (Y) u (L, y) =f2(y)

u(x, 0) =f3(x)

Figure 5.10: Heat conduction in a rectangular plate of length L and height H.

subject to the BCs

(5.106)

H)= = f4(x ).
u(x, H)

The only solution technique you currently have to solve a PDE is based on sep-
aration of variables, which requires a linear homogeneous PDE and BCs. The BCs,
Equation (5.106), are linear, but not homogeneous. Thus, separation of variables
does not appear to help us. However, suppose we have four separate solutions to
Equation (5.105). Then, by the principle of superposition, the sum of the four
solutions would also be a solution. Thus, consider

u (x, y) = u1(x, y) + u2 (x, y) + u3 (x, y) + u4 (x, y) . (5.107)

We really need to know two things about the four solutions. First, what do the
four solutions look like? Second, we know the solutions satisfy Equation (5.105),
but how do they satisfy the BCs, Equation (5.106)? The answer to both of these
questions is tied together. We let each of the solutions, u( x, y), i = 1,... , 4,
have one nonhomogeneous boundary condition from Equation (5.106) and three
homogeneous boundary conditions. Therefore, we have the following solutions,
184 Chapter 5: Separation of Variables: The Homogeneous Problem

which are graphically illustrated in Figure (5.11):

V2u1(x, y) = 0 V2u2 (x, y) = 0 V2u3 (x, y) = 0 V2u4(x, y) = 0

subject to subject to subject to subject to

u1(0, y) = f '(Y) u2 (0, y) = 0 u3 (0, y) = 0 u4 (0, y) = 0

u1(L, y) = 0 u2 (L, y) = 12(Y) u3 (L, y) = 0 u4 (L, y) = 0

u1(x, 0) = 0 u2 (x, 0) = 0 u3 (x, 0) = f3(x) u4 (x, 0) = 0

u1(x, H) = 0 u2 (x, H) = 0 u3 (x, H) = 0 u4 (x, H) = f4(x)

u = f(x) uI = 0 u2 = 0

u =f(y) u,=0

+
u = f(y)

u = f(y)
uI = 0 u2 =f2( )

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

ua =.fa(x)

U, = 0

+ +

tt, = 0 u0 ua = 0

U, =f(x) U, = 0

Figure 5.11: Assumption of u(x, y) = u( x, y) + u2 (x, y) + u3 (x, y) + u4 (x, y).

From our original supposition, we know the sum of the us is a solution to

Equation (5.105), and by our assumption about each solution, we know, for instance,
at y = H that
ul (x, H) = u2 (x, H) = u3 (x, H) =0
 Section 5.6: Laplace's equation 185

and

u4 = f4(x).

Thus, applying the principle of superposition at y = H yields

u(x, H) = u1(x, H) + u2 (x, H) + u3 (x, H) + u4 (x, H) = f4 (x),

the nonhomogeneous condition we require for u (x, y). Similarly, we can show
u(x, 0) = f3(x), u(L, y) = 12(y), and u(0, y) = fl(y). Therefore, the original
problem may be broken down into four separate problems, each with three homo-
geneous BCs and one nonhomogeneous BC. Can we use the separation of variables
technique on the four separate problems? If we consider each of the four separate
problems as a boundary value problem, then we cannot. However, if we consider
the variable with the nonhomogeneous condition as a "time-like" variable, then we
can use separation of variables.
Since the method of solution is nearly the same for any of the four, the solution
for u3 (x, y) is given here. u2 (x, y), and u4 (x, Y). Figure (5.12) depicts the rectan-
gular plate u3 (x, y) It demonstrates how a simple shift of a coordinate aids us in
.

the solution. The coordinate shift is possible because the solution of the resulting
ODE is invariant under a translation. That is, the translation does not change the
solution. The remaining three are left as an exercise.

u3(x, H) = 0

u3(0, y) = 0 u3(L, y) = 0

u3(x, 0) =f3(x)

Figure 5.12: Heat conduction in a rectangular plate, u3 (x, y), of length L and
height H.

For u3 (x, y), we have

V2u3 (x, y) = 0 (5.108)

186 Chapter 5: Separation of Variables: The Homogeneous Problem

subject to

u3(0, y) = 0

u3(L, y) = 0
(5.109)
u3 (x, 0) = f3 (x)

u3(x, H) =0.
We assume at x = 0 and x = L that we have boundary conditions and at y = 0
and y = H that we have "time-like" conditions. Letting u3 (x, y) = 3(y)G3(x) and
separating in the usual manner we arrive at the two ODEs

GPs x) _ -A3(x) (5.110)

subject to
fX3(0)=0

Ps(L) _

and

G3(y) _ AG(y) (5.111)

subject to
G(0)=f(x)
G3(H)=0.
The spatial ODE, Equation (5.110), subject to its BCs, has the solution
(fl7 2

L)
n= 1,2,3,... (5.112)
n7rx
SP3n (x) =sin L

The second ODE, Equation (5.111) that is now stated as

(fl71) 2
Gin (y) = A Gin (y) = Gin (y)> (5.113)

subject to

Gsn(0) = f3(x)

G3(H)
Gin (H) _ 0,
 Section 5.6: Laplace's equation 187

has the solution

Gin (y) = C3n cosh

nay
+ din sinh i
ny . (5.114)
L
Equation (5.114) really does not help us when we want to use the condition G3n (H) _
0. However, a simple shift in the y coordinate may help us. Once we shift the y
coordinate, we must make sure it solves Equation (5.113). We replace y with y - H
in Equation (5.114). Thus, Equation (5.114) becomes
H) nr(y H)
G 3n (y) = c3n
cosh L + d3n sinh L . (5.115)

As practice, you should show that Equation (5.115) solves Equation (5.113).
We may now apply the condition G3n (H) = 0, which yields

Gsn (H) _ 0 _ c.
Therefore, we have

n7r(y
Gsn (y) =din sinh - H) . (5.116)
L

Combining the solutions for 3n (y) and G371 (y), and using the principle of su-
perposition, we arrive at
00
nir(y H) nirx
u3(x, y) = d 3n sinh sin
n=1
L L

Using the "time-like" BCs, u3 (x, 0) = 13(x), we may determine the last unknown,
d3n , by using the orthogonality of the Fourier sine series. Thus, we have

00
nom( H) fx n3(
u x 0) = f3( x) = d3 sinh
L
sin , (5.117)
n=1

which implies

-2 L
f 3 (x) sin L dx. (5.118)
L sinh L o

The solutions for u1 (x, y), u2(x, y), and u4(x, y) are determined similarly, and
are left as an exercise. Note: For u1 (x, y), a coordinate shift must be determined.
You should prove to yourself that the translated solution of the ODE does not
change the differential equation. For u2 (x, y) and u4 (x, y), no shift is required.
Once all four solutions are found we use the principle of superposition to combine
the four solutions, which yields the solution for u (x, y). Also, please remember
that Laplace's equation is the primary representative of the elliptic class of linear
second-order PDEs.
 188 Chapter 5: Separation of Variables: The Homogeneous Problem

In Chapter 10, we will again meet Laplace's equation. Only then we will work
in polar, cylindrical, and spherical coordinate systems.

EXERCISES 5.6

5.6.1. Show that u(x, y) = xy and u(x, y) = x2 + y2 are harmonic in I12.

5.6.2. Find the complete solution of
52u(x, y) _
V2u(x,
(x, y) - D2u(x,
5x2
y)
+ ay2
0,

subject to

u(x,H) = f4(x).
Remember that u3 x, y) was solved in the text.
5.6.3. Find the complete solution of
52u(x, y) _ 0,
V2u(x,
(x, y) - D2u(x,
axe
y)
+ ay2
subject to
8u(0, y)
- .fi (y)
8x

au(x, H) _
f4(x).
ay
5.6.4. Consider Laplace's equation inside a rectangle. Suppose .fi(y) = f3(X)
f4(x) = 0 and 12(y) = y + 1. Determine the solution.
5.6.5. Given Laplace's equation in Cartesian coordinates

y) _ _
D2u(x, y) D2u(x, y)
2
v u(x, - oyy

find the solution for each of the following BCs:

 Section 5.6: Laplace's equation 189

(,y),
1 u0 =0 au(2'y) = ux 0 =0 andux 3 =x.
(2) u0 = 2 au(, y) =
u(x, 0)
- 0, and u(x, 27r) = sin x.
ax ' a
aua(0x, y) au(x, 0)
=0 ' u ('y) =cos y'
a
0, and u(x,1) = 0.
(3) 2

au(0, y) au(x, 0) au(x, 7r)

4 =0u1 =sin 0, and = x2.
() ax ' (' y) y' ay ay

(5) u(0, =0 u(2, - au(2,y) =0 u x 0 =0 andux 3 =0.

au(0, y) au(x, r) _ 0
(6) u(0,y) _ = 0, u(2 ,y) = cosy, u(x,0) = 0, and
ax ay
au (x, 0)
(7) u(0,y) = 0 u(,y) = sin ux0_ = 0 andux 3= 0.
y

(8) u(0, =0 = 0 u x 0 = 0 andux 1_ _u(x, 1) _ 0

y
5.6.6. If u1(x, y) is a solution of Laplace's equation, prove that the
partial derivative of u1 (x, y) with respect to one or more of the rectangular
au1 a2u1 a2u1
coordinates (for example, and also a solution.
ax , axe , a yax) are
5.6.7. Formulate and solve the general Laplace's equation in a
three-dimensional parallelepiped with BCs of the first kind (Dirichlet condi-
tions).
5.6.8. Formulate and solve the general Laplace's equation in a
three-dimensional parallelepiped with BCs of the second kind (Neumann con-
ditions).
5.6.9. Formulate and solve the general Laplace's equation in a
three-dimensional parallelepiped Dirichlet BCs on the boundaries x = 0 and
x = L and Neumann BCs on the boundaries y = 0 and y = H.
5.6.10. Formulate and solve the general Laplace's equation in a
three-dimensional parallelepiped Dirichlet BCs on the boundaries x = 0 and
y = 0 and Neumann BCs on the boundaries x = L and y = H.
5.6.11. Formulate and solve the general Laplace's equation in a
three-dimensional parallelepiped Dirichlet BCs on the boundaries x = L and
y = 0 and Neumann BCs on the boundaries x = 0 and y = H.
5.6.12. Consider a region bounded by conducting plates x = 0, y = 0, and y = H.
Suppose the plate x = 0 is charged to a potential V, the plates y = 0 and
y = H are earthed, and there are no charges inside the region, find the
electrostatic field inside the region.
 190 Chapter 5: Separation of Variables: The Homogeneous Problem

5.6.13. PROJECT:3 This project models the fluid in the cochlea surrounding the
basilar membrane, which is part of the inner human ear. It is assumed that the
fluid in the cochlea is incompressible and inviscid (not thick). If the equations
are nondimensionalize we arrive at
D2u(x, y) 52u(x, y) =0
axe + D y2

subject to
8u(0, y)
=1
ax

u(1, y) = 0

2u(x, 0)
' I
u(x, H) = 0.
where I is called the impedance in the damped harmonic oscillator.
(1) Find the general solution.
(2) Show that after truncating the series at N terms, multiplying by cos(mirx),
and integrating from 0 to 1, we obtain the system of linear equations
N
An nm fm,
n=0

where

f
1
cos(nirx) cos(mirx)
nm = 2 cosh(nirL) dx - 1 nor sink( nii H )nm
S
I 2

where Sip = 1 if i = j and 0 otherwise and

fm=LSmp-
f 1 x(2 - x) cos(mirx)
I
dx.

3Adapted from James Keener and James Sneyd, Mathematical Physiology, ©1998 by Springer-
Verlag, New York, pp. 707-711. Reprinted by permission.
Chapter 6

The Calculus of Fourier

Series
6.1 INTRODUCTION
Chapter 4 introduced you to Fourier series. You learned that a Fourier Series is a
linear combination of orthonormal functions. This linear combination can represent
any function in the function space of piecewise smooth functions.
In Chapter 5, you solved PDEs by separation of variables. The separation of
variables technique led to a Fourier series solution for the PDE. Also, the Fourier
series had a function of time, t, as part of the solution. The natural question to ask
is, how do we know the Fourier series solution of the PDE is the actual solution to
the PDE? This is a very important question, and it must be answered. There are
really two questions here: (1) how and when can a Fourier series be differentiated
and integrated?; (2) If a Fourier series can be differentiated and integrated, what is
the result?
Before we attempt differentiating or integrating a Fourier series, we must deter-
mine if a Fourier series is a function. We'll do that in the next section.

6.2 FOURIER SERIES REPRESENTATION OF A

FUNCTION: FOURIER SERIES AS A FUNCTION
What is a function? Here's one definition:
Definition 36. A function f from a set A, called the domain, to a set B, called
the range, is a rule of correspondence that assigns to each x in a certain subset D
of A, a uniquely determined element 1(x) of B.
How does this definition apply to a Fourier series? To answer this question, we
must identify everything in the definition of a function with a particular object in
our study of Fourier series.

191
192 Chapter 6: The Calculus of Fourier series

First, the sets A and B in the definition are the set of real numbers, R, for
Fourier series. Note: Both sets A and B can be the same. Second, the subset
D of A in the definition refers to the interval [-L, L], a subset of Ilk for Fourier
series. Finally, the very nature of Fourier series is to assign to each element in the
interval [-L, L] a unique element in the range. Thus, a Fourier series is a function.
Although knowing a Fourier series is a function is important, the really important
concept is knowing if the Fourier series is a continuous function.
From Chapter 4, we know the Fourier series representation of a function on the
interval [-L, L] converges to the function at all points the function is continuous
and to the average value at any point of discontinuity. Consider the following two
examples.

EXAMPLE 6.1. Consider

-7r -7r

f(x)

r
7r-x, 2 <x <?r.
The function 1(x) is a piecewise smooth continuous function. Its Fourier series
representation is

1(x)
- 16 + n=1 (urn \(-1)n - cos 2) cos nx

00,
-- ((6 sin
nor
- r -1 ))
n
1 sin nx.
n= 1
2 ---)
2n r
The solution is graphed in Figure (6.1) on the interval [- 37r, 3ir]. It shows the
Fourier series representation of 1(x) converges to - at x = -3ir and x = 37r.
4
Since Fourier series are periodic with period 2L (where L is the length), we know
the Fourier series representation of 1(x) at x = -?r or it also converges to .
4
Also, the Fourier series representation of 1(x) at x = -?r or it converges to ,
4
but the function equals
-it
and 0 respectively. This means = f
-it f
2 4 2 '
or the average value of the endpoints of the function. Thus, a jump discontinuity
exists in the Fourier series representation for 1(x). Therefore, the Fourier series
representation of 1(x) is not continuous and can not be differentiated term-by-
term.
Look closely at Figure (6.1). As the Fourier series Representation of 1(x) approaches
it, from the left or from the right, the graph seems to oscillate. Then the Fourier
 Section 6.2: Fourier series as a Function 193

series representation of 1(x) converges to - The oscillation is known as the Gibbs

.
4
phenomenon. The Gibbs phenomenon will be discussed at greater length in the last
section of this chapter.

1.5

Figure 6.1: The graph of the Fourier series representation off (x) .

EXAMPLE 6.2. Given

2, 2

g(x)= -x,---x<
-2,2<x<7r.
37r 7r

The function g(x) is a piecewise smooth continuous function. Its Fourier series

Figure 6.2: The graph of the Fourier series representation of g(x).

194 Chapter 6: The Calculus of Fourier series

representation is

37r 2 n7r
g (x) = + ((_i) n - cos cos nx
16 7rn 2 2
n=1

+
00
-4 sin
n7r
sin nx.
7rn 2 2
n=1

The solution is graphed in Figure (6.2) on the interval [-37r, 37r], which shows the
Fourier series representation of g(x) converges to at x = -37r and x = 37r. As
2
in the last example, since the Fourier series is periodic with period 2L, we know
that the Fourier series converges to at x = -it or it. Thus, the Fourier series
2
representation for g(x) is continuous at the endpoints. And since there are no other
discontinuities in the function, the Fourier series representation of g(x) is continuous
on the interval [-7r, it].
Knowing when a particular Fourier series is a continuous function, without being
forced to graph the Fourier series, is very important. At the end of Chapter 4, three
theorems were given to clarify when a particular Fourier series is continuous. These
definitions are restated below:
Theorem 37. If f (x) is a piecewise smooth continuous function on the interval
[-L, L] and f (-L) = 1(L), then the Fourier series of the function 1(x) is contin-
uous on [-L,L].
Theorem 38. If f (x) is a piecewise smooth continuous function on the interval
[0, L], then the Fourier cosine series representation of the function f (x) is contin-
uous on [0, L].
Theorem 39. If f (x) is a piecewise smooth continuous function on the interval
[0, L] and f(0) = f (L) = 0, then the Fourier sine series representation of the
function f (x) is continuous on [0, L].
Notice in all three theorems the function f (x) is continuous and piecewise smooth
on the interval. Both characteristics are required. A function may be continuous
on the interval [-L, L] but not be piecewise smooth. A good example is the tan x
on the interval [- , ]. At - and the tan x goes to - oc and oc, respectively.
2 2 2 2
Likewise, if a function is piecewise smooth and not continuous, the Fourier series
representation will have jump discontinuities. Thus, the Fourier series as a function
will not be continuous. Also note, for two of the theorems, specific values for
the function at the end points are given. The endpoint values are very important
because they insure the Fourier series representation of the function in Theorems
(37 and 39) converges to the value at the endpoints of the function.
Why is continuity of a Fourier series so important? The answer is quite straight-
forward. Since you cannot differentiate a function on an interval if the function has
 Section 6.3: Differentiation of Fourier series 195

a jump discontinuity, you cannot even consider differentiation of a Fourier series

on an interval unless the Fourier series is continuous. This is the first step toward
differentiation of a Fourier series; actual differentiation of a Fourier series is the
second step, which we go to now.

6.3 DIFFERENTIATION OF FOURIER SERIES

When solving a linear second-order ODE, you were taught to plug the solution back
into the ODE to see if you actually got the correct solution. For example, consider
y"(x) - 3y'(x) + 2y(x) = 0,
where the solution is
y (x) = cl e2x + c2ex .

Using Equation (6.2), you found

y' (x) = 2c1 e2x + c2ex

and

y" (x) = 4c1 e2x + c2ex . (6.4)

Replacing y' (x) and y" (x) in Equation (6.1) yields

4cle2x + c2e-x
-3 (2cie2x + c2ex) + 2 (cle2x + c2ex) = 0. (6.5)

The obvious conclusion is that you found the correct solution.

Does a similar procedure prove that the solution to a linear second-order PDE
is the correct solution? Consider the PDE:
au a2u
at - axe
subject to
u(-L, t) = u(L, t)
au(-L, c) au(L, t)
at - at
and

u(x,O) = 1(x).
The solution is
n ix
00

u(x, t) = ao + [ane_(O2t cos n?Cx + bne-(L )2t sin

, (6.6)
n=1
L
196 Chapter 6: The Calculus of Fourier series

where the three Fourier coefficients are given by

_ 1 L
ao f (x)
2L -L dx,

1 L n7rx
an = L J 1(x) cos dx,
-L L
and
1 L
bn L (x) sin L dx.
- Lf
Note: The Fourier coefficients do not depend on the variable x.
Following the same procedure, which exists for second-order linear ODEs, and
noting the Fourier series solution in Equation (6.6) is continuous in the spatial
variable by Definition (37) and is assumed to be continuous for the time variable,
we must find
00
au = a [ane_(t)2t cos nrx + bne - (L) t sin nirx
2
a0 +
at at L n=1
L L

a a nirx
+bne - (L) sin
2t 2t n7rx
[ao]+ [ane (L) cos
at at Ln=1
L L

au a
a0 +
°°
[ane_()2t cos nrx + bne - ( L) 2 t sin nrx
ax ax n=1
L L

a a °° 2t nrx 2t
[ao] + [ane_ (L) cos +bne _ (L) sin ,
ax ax n-1 L L

and
a2 u a a °O nrx +bne n7rxi
-_L
= a0 + [ane_()2tcos - ( L ) 2 t sin
ax 2 ax ax n=1
L L

a
[aol
ax 125;;

aJa °° nrx
ax ax [anetcos
_ (L)
L
- bne _ (L)
2 2t
sin
-i-i] }.
(6.9)
n=1

What we need to perform on the left side of Equations (6.7, 6.8, and 6.9) is quite
straightforward. However, the right sides of Equations (6.7, 6.8, and 6.9) present
 Section 6.3: Differentiation of Fourier series 197

interesting questions. How do you differentiate an infinite series, and what do you
get for a solution? Also, an interesting follow-on question is if we can differentiate a
Fourier series, which is an infinite series, can we differentiate all infinite series by the
same method? This is not a course in general infinite series, but the answer to the
last question depends on whether the infinite series is continuous and convergent, as
well as if the derivative of the infinite series is uniformly convergent. For Definitions
(37, 38, and 39), the Fourier series representation of the function is continuous,
converging uniformly to the function. This simple fact makes working with Fourier
series much simpler. It means differentiating a Fourier series is possible.
Since differentiation of a Fourier series is possible under certain conditions, what
is the end result of differentiation of a Fourier series, and how do we perform the
differentiation? We perform the differentiation term-by-term, which must be shown.
The end result is hopefully another Fourier series, which is also the Fourier series
representation of the derivative of the original function.
To prove these statements, we consider a general Fourier series representation
of a function, u(x, t), where all the Fourier coefficients depend on the variable t:
00

u(x, t) = ao(t) + [an(t) cos nLx + bn(t) sin nLx] . (6.10)

n=1

Also, we assume that the function u(x, t) is continuous for both time and space
variables and u(-L, t) = u (L, t). These conditions insure the Fourier series rep-
resentation of u(x, t) is a continuous function on the interval [-L, L]. Next, we
assume that u(x, t) is continuously differentiable with respect to both variables. Fi-
nally, we hope that the derivative with respect to either the spatial variable, x, or
the time variable, t, of a Fourier series is another Fourier series, which converges to
the derivative of the original function, u (x, t). Then, we assume that the derivative
with respect to either variable of u(x, t) is a piecewise smooth function. We find the
derivative with respect to time first, then we state the resulting theorem. Next we
find the derivative with respect to the spatial variable, and we state the resulting
theorem.
We have

u(x,t) = ao(t)+ [an(t) cos bn (t) sin zH (6.11)

n=1

where u(x, t) is continuously differentiable with respect to t, and the Fourier series
representation of u(x, t) is continuous. Also, we know that the derivative of u(x, t),
au(x, t) au(x, t)
is p iecewise smooth and thus may
Y be written as a Fourier series
at at
which is
au(x, t) °O n7rx n7rx
= Ao (t) + [An(t) cos + Bn (t) sin (6.12)
at
n=1
L L

Note: The coefficients of Equation (6.12) are different from the coefficients in Equa-
tion (6.11).
198 Chapter 6: The Calculus of Fourier series

It suffices to show term-by-term differentiation of a Fourier series with respect to

the parameter t is possible if we can show that the Fourier coefficients of Equation
(6.12) are derived from the Fourier coefficients of Equation (6.11).
The equations for the Fourier coefficients corresponding to Equation (6.11) are
L
a°(t) 2L JL u(x, t) dx, (6.13)

L
an(t) = 1L-u(x,t) cos n7rx
L dx, (6.14)
L

and
L
1 n7rx
bn (t) = L-u(x,t) sin L dx. (6.15)
L

The equations for the Fourier coefficients corresponding to Equation (6.12) are

1 L 8u(x, t)
Aoff)
t = ZL L at dx (6.16)

_ 1 L au (x, t) n7rx
An cos dx, (6.17)
(t) L -L at L

and
1 L au(x, t) n7rx
Bn (t) _ L sin dx. (6.18)
-L at L

The form of the integrals in Equations (6.16, 6.17, and 6.18) should remind you of
Leibniz's theorem in Chapter 2, which we restate here.
af
Theorem 40. Suppose f (x, t) and the partial derivative (x' t) are continuous in
at
some region of the xt-plane where a < x < b, then

td J f r6(x, t) dx b of (at x, t)
dx.
Ia

au(x t)
Thus, if we further assume that at' is continuous in some region of the
xt-plane where -L < x < L, then Equations (6.16, 6.17, and 6.18) become
L
A0(t)
dt
[fu(xt)
2L
dx (6.19)
L
 Section 6.3: Differentiation of Fourier series 199

L
,nLx dx (6.20)
`4n(t) dt L ,

and
rL
Bn(t) _ t L u(x, t) sin nLx dx (6.21)
J L

Using Equations (6.13, 6.14, 6.15), and substituting equivalent terms, we find that
Equations (6.19, 6.20, and 6.21), respectively, become

`4°(t)
dt
[u(x,t)
,f
2L
L
dx = dt
[ao(t)] = ao(t),
L

f u(x, nLx dx
L
A(t) _ [Lt)cos - at [an(t)] - an(t),

and

L
Bn (t)
dt L fL u(x, t) sin nLx dx = dt
[b(t)] = b( t).

Thus, the Fourier coefficients of Equation (6.12) are derived from the Fourier coef-
ficients of Equation (6.11). Therefore, term-by-term differentiation with respect to
a parameter t is valid, and we have proved the following theorem:
au(x t)
Theorem 41. If u(x, t) and are continuous functions, then the Fourier
series representation of u(x, t) on the interval [-L, L],
00

u(x, t) = ao(t) + [an(t) cos nLx + bn(t) sin nLx] ,

n=1

may be differentiated term-by-term with respect to the parameter t, and the result is
the Fourier series representation of the derivative of u(x, t) with respect to t,
00
au(x, t) n7rx , n7rx l
at
= a( t) + [a(t) cos L
+ bn (t) sin
L
n=1

Next, we show that term-by-term differentiation of a Fourier series with respect

to the variable x is valid. Again, we have
00

u(x, t) = ao(t) -}- [an(t) cos nLx + b(t) sin nLx] , (6.22)
n=1
200 Chapter 6: The Calculus of Fourier series

where u(x, t) is continuously differentiable with respect to x, u(-L, t) = u(L, t).

Therefore, the Fourier series representation of u(x, t) is continuous. Also, we assume
au(x, t) au(x, t)
the derivative of u(x, t), is Piecewise smooth and thus may be
ax ax
written as a Fourier series, which is

au (x, t) °O n7rx n7rx

= ao (t) + [(t)
an cos + ,Qn (t) sin (6.23)
ax
n=1
L L

Again, it suffices to show that the Fourier coefficients of Equation (6.23) may be
derived from the Fourier coefficients of Equation (6.22). The equations for the
Fourier coefficients corresponding to Equation (6.22) are
L
ao(t) = ZL IL u(x, t) dx, (6.24)

fL
an (t) = LL u(x, t) cos nLx dx, (6.25)
L

and
L
1 nrx
bn(t) = u(x, t) sin L dx. (6.26)
L -L

The equations for the Fourier coefficients corresponding to Equation (6.23) are
L
1 Du(x,t)
(6.27)
2L f L 8x '

_ 1 L au(x, t) nrx
an (t) cos dx, (6.28)
L -L ax L

and

1 au(x, t) nrx
sin dx. (6.29)
L ILL ax L

Integrating Equation (6.27) yields

L L

L
a ax't)

dx = 2Lu(x, t)
-L

- 1
[u(L,t) - 'u(-L,t)] = 0,

since u(L,t) = u(-L, t).

 Section 6.3: Differentiation of Fourier series 201

The Fourier coefficient ao (t) = 0 is not unexpected. Remember, a function of t

is considered a constant when finding the derivative with respect to the variable x.
For Equation (6.28),
_ 1 L au(x, t) mix
an (t) cos dx,
L -L ax L
au(x, t)
we employ the method of integration by parts with dv = dx and w =
flit- ax
cos L. This yields

an(t) = L (u(x,t)cos__Z_) nxL + L J

L /'L
Lu(x, t) sin nLx dx

L
1

L
[u(Lt)cosmi - u(-L, t) cos flit + 1J u(x, t) sin mix
flit
L
dx
-L

1 nit L mix
L
[(u(Lt) - u(-L, t)) cos flit +
L- L
u(x, t) sin
L
dx

1 n7r L mix
Lu(x, t) sin dx (6.30)
L L - L

Rewriting Equation (6.30) yields

an (t) =
flit [fLu(x,t)sin
1
dx . (6.31)
L L

Using Equation (6.26) and substituting equivalent terms, Equation (6.31) becomes
L
_ flit 1 mix flit
an (t) L
L -L
u(x, t) sin L dx = L bn (t).

Thus, an (t), in Equation (6.23), depends on the coefficient bn (t), in Equation (6.22).
d flit- = flit flit-
Again, this is not unexpected since sin cos
dx L L L
Similarly, we find the Fourier coefficient /n (t) depends on the coefficient an (t) .
Thus, we have proved term-by-term differentiation of a Fourier series representation
of a continuous function with respect to the variable x. We state our results in the
following theorem:
Theorem 42. If u(x, t) is a continuous function on the interval [-L, L] with u(-L, t)
au(axx' t)
= u(L, t), and is a piecewise smooth function, then the Fourier series
representation of u(x, t) is continuous and it can be differentiated term-by-term.
 202 Chapter 6: The Calculus of Fourier series

There are two more theorems stated subsequently, with proofs left as exercises.
These theorems develop term-by-term differentiation of Fourier Cosine and Fourier
sine series representation of a function u(x, t). As you may have guessed, knowing
these theorems is very important in the chapters that follow.
au(x t) .

Theorem 43. If u(x, t) is a continuous function on the interval [0, L], and Zs
ax'
a piecewise smooth function, then the Fourier cosine series representation of u(x, t)
is continuous and it can be differentiated term-by-term.
Theorem 44. If u(x, t) is a continuous function on the interval [0, L] with u(0, t) _
au(axx' t)
u(L, t) = 0, and is a piecewise smooth function, then the Fourier sine series
representation of u(x, t) is continuous and it can be differentiated term-by-term.
We have covered all cases where Fourier series may be differentiated term-by-
term. If the function u(x, t) in Theorems (42, 43, and 44) is a function of one
variable, that is, f (x), the theorems still hold. Next, we turn our attention to
term-by-term integration of a Fourier series.

EXERCISES 6.3

6.3.1. Show that the Fourier coefficient Bn (t) in Equation (6.29) depends on the
coefficient an (t) in Equation (6.25).
6.3.2. Prove Theorem 43.
6.3.3. Prove Theorem 44.
6.3.4. Given
a2 u 2 a2 u
5;=C
axe
subject to
u(0, t) = 0 and u(L, t) = 0
and

u(x, o) _ .f(x) and auat' ° = g(x)>

the solution for u(x, t) is
00
cn7rt cn7rt n7rx
u(x, t) = an cos L + bn sin L sin L .

n-1

Prove, using term-by-term differentiation, that the solution satisfies the PDE
and BCs.
 Section 6.3: Differentiation of Fourier series 203

6.3.5. Given
au a2 u
at axe
subject to
au(0, t) au(L't)
=0 and -0
ax ax
and
u(x,0) = f(x),

the solution for u(x, t) is

00
n?fx
u(x, t) = ao + ane-(L )2t cos
n=1
L

Prove, using term-by-term differentiation, that the solution satisfies the PDE
and BCs. State reasons for all differentiations.
6.3.6. Given

271 - x, 7 < x < 27,

find the Fourier sine series representation of 1(x). Then, determine the Fourier
cosine series representation of f' (x) using term-by-term differentiation.
6.3.7. Consider the Fourier sine series
00
n7rx
1(x) bn sin L,

n=1
where 0 < x < L and 1(x) is a continuous function, ti means approximately,
and f' (x) is piecewise smooth. Suppose 1(0) = a 0 and f (L) = 0, and
show that
n7rx
1
[-a]+ Ln((_l)_]c05
n71 2 n 1

n=1

6.3.8. Consider
00
nix
cosh x = ao + an cos L, 0 < x < L.
n=1

Differentiating both sides of the equation with respect to x yields

00
-nom nix
sink x = an sin L, 0 <x <L.
n=1
L
 204 Chapter 6: The Calculus of Fourier series

Differentiating again yields

cosh x =
00

- n 2
an cos
nix
, 0 < x < L.
L L
n=1

Since the Fourier cosine series representation of a function on an interval is

unique, we must have

°° nor 2 n?fx
a0 + an cos = - (-i-) an cos
L n=1
L
n=1

This implies that

a0=0andan=0.
This conclusion is clearly false. Determine the error in the logic.
6.3.9. Consider
00
nix
sinh x = a0 -F- an cos , 0 < x < L.
L
n=1

Differentiating both sides of the equation with respect to x yields

00

cosh x = - ?- an sin nix , 0 < x < L.

L L
n=1

Differentiating again yields

00
n- nix
- (1)
2
sinh x = an cos
L
n=1

Since the Fourier cosine series representation of a function on an interval is

unique, we must have

°° n?rx °° n?r 2 n?rx

a0 + an cos = - (-E-) an cos
L L
n=1 n=1

This implies that

a0=0andan=0.
This conclusion is clearly false. Determine the error in the logic.
6.3.10. Given
au a2u
ax2
 Section 6.3: Differentiation of Fourier series 205

subject to
au (L, t)
()
u 0 t= 0 and
ax
=0
and

u(x,0) = f(x)

the solution for u(x, t) is

(2n-1)71x
u(x, t) = bne-an of sin ,

n=1
2L

2
[(2n_1)71
n= 2L ,n=1,2,3,...

Prove, using term-by-term differentiation, that the solution satisfies the PDE
and BCs. Briefly discuss any assumptions you make.

a2u
=16 a2u
axe
ate

u(0, t) = 0 and u(L, t) = 0

u(x, 0) = f (x) and aunt' = g(x),

the solution for u(x, t) is

4n71t mix
an cos L + bn sin 4n71t
L
sin
L
n=1

2
An = n= 1,2,3,...
L
Prove, using term-by-term differentiation, that the solution satisfies the PDE
and BCs. Briefly discuss any assumptions you make.
6.3.12. Given
au a2 u
at ax2 '
206 Chapter 6: The Calculus of Fourier series

subject to

and
u (,),)
-L t = u L t and au(-L, t) _
a
au(L, t)
ax

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

the solution for u(x, t) is

u(x, t) = ao +
00

n=1
e- an cos
n?rx
L
+ bn sin
n7rx
L
_),
nor 2
An=
[-IL-I
,n=1,2,3,...
Prove, using term-by-term differentiation, that the solution satisfies the PDE
and BCs. Briefly discuss any assumptions you make.

6.4 INTEGRATION OF FOURIER SERIES

When expressing a piecewise smooth function, 1(x), on the interval [-L, L], we
use the orthogonality of the sine and cosine functions to determine the Fourier
coefficients. This requires multiplying the equation
00
nzrx
1(x) = ao + [an cos bn sin
L L
n=1
Lmix mix
by either the sin or cos and integrating the equation from -L to L.
L
Thus, for instance, when finding bn we have
L
mrx L minx
1(x) sin L dx = ao sin L dx
J-L J-L
nhrx nhrx mrx
+ [an cos L + bn sin L sin L dx.

mrx
Since sin L is an odd function,
L
mrx
ao sin L dx = 0.
-L
Therefore, we have remaining
L
mrx L°° n7rx n7rx mrx
I L
1(x) sin
L
dx =
-L n=1
[an cos
L
+ bn sin
L
sin
L
dx. (6.32)
 Section 6.4: Integration of Fourier series 207

In Equation (6.32), once we know the function 1(x), the left integral is feasible.
However, to perform the integration of the infinite series in Chapter 5, you were
asked to believe the integral of the infinite sum is equal to the infinite sum of the
integrals, and the result was a formula for the term b. This would tend to mean
term-by-term integration of a Fourier series representation of any piecewise smooth
functions on the interval [-L, L] is possible. We confirm this suspicion with a
theorem.
Theorem 45. 111(x) is a piecewise smooth function on the interval [-L, L], then
the Fourier series representation of f (x) can always be integrated term-by-term.
The result is an infinite series, not necessarily a Fourier series, which converges to
the integral of the function f (x) for -L < x < L. Note: It does not matter if the
function f (x) has jump discontinuities.

The proof of Theorem (45) can be found in the book titled Infinite Series, by
Earl D. Rainville, published in 1967.
Two examples are given showing term-by-term integration of Fourier series.
These examples demonstrate the unique features of Theorem (45). Remember, you
are guaranteed an infinite series that converges to the integral of the function the
Fourier series represents.

EXAMPLE 6.3. Consider the function f (x) = x on the interval [-2, 2]. Since
f (x) is an odd function, the Fourier series representation of f (x) = x is the same as
the Fourier sine series representation of f (x) = x on the interval [0, 2]. Therefore,

°O -4(-1)n n7rx
x= sin (6.33)
nor 2
n=1

x2
Integrating Equation (6.33) will yield an infinite series for the function 2 . That
is,

_4(_1)72 fl717
T dT = / sin dT =
0 nor 2
0 n=i
00
-4(-1)n
nor 0
x
sin-dr.
n71T
2
(6.34)
n=1

Notice the integrals in Equation (6.34) are from 0 to x. We want the integrals this
way because we want a function after integration. Performing the integration yields
x
x2
_
°° -4(-1)n -2 coS n71T =
2 n=1
n71 [n71 2 Jo

°° 4(-1)72 2 mix
cos -1 (6.35)
n-1 nor nor \ 2
208 Chapter 6: The Calculus of Fourier series

Rewriting Equation (6.35) yields

x2
2
_ 00.

L (n)
8(-1)n
2 (cos 2
nix
- 1)
n=1
00
8(-1)n
(mu)2'
nix
2
2
- 8(-1)n
(mu)
a

00
8(-1)n n7rx 8(-1) n
_ 2
COS
2
n=1 (n7r) n=1 (n71)
or

x2 =
-
16(1)n
cos
n7rx
2
00

.- (nit)2 n-1
(1Z7C12

00
16(-1)n , 16(-1)n mix
cos 2 (6.36)
n= n=1 (n71) 2

We know the infinite series in Equation (6.36) converges to x2. But, as expected,
the infinite series in Equation (6.36) is not a Fourier series. Sometimes, we really
need a Fourier series solution, and with a little algebraic manipulation we can
transform the infinite series, like that in Equation (6.36), into a Fourier series.
EXAMPLE 6.4. Starting with the infinite series in Equation (6.36),
00 o0
2 16(-1)n 16(-1)n n7rx
X=- 2 -F- 2 cos
2
(6.37)
n=1 (mu) n=1 (n71)

We notice the second term on the right has a cosine function. The cosine function
indicates the possibility of transforming it into a Fourier cosine series. Thus, we set
up the standard Fourier cosine series for g(x) = x2,
00
n7rx
g(x) = x2 = ao + an cos
2
n=1

16( -1)n
Comparing similar terms, we find the term - 2
, in Equation (6.37), must
n=1 (nit)
16(-21) n
be the ao term in Equation (6.38), and the term , in Equation (6.37), must
(mu)
be the an term in Equation (6.38). From Equation (6.38), we determine an and ao.
We have
mix
2
an = L J g(x) cos
L
L
dx =
2
x2 cosn7rx
L dx
0 0

16(-1)n
(6.39)
 Section 6.4: Integration of Fourier series 209

and
f2
a0 = 2J x2dx=3. (6.40)
0

Thus, Equation (6.38) becomes

2 4 °O 16(-1) n nirx
x=3+ (n7r)
2 (6.41)
n=1

The right side of Equation (6.41) is almost what we have in Equation (6.37). The
16(-1) n
an term in Equation (6.41) is identical to the term 2
in Equation (6.37).
(n71)
To complete the process, we must show

4 16(-1) n
3
n=1

Consider Equation (6.41). Evaluated at x = 0,

00
4 °O
16(-1) n n71(0) _ 4 16(-1) n
g(0) = 02 = 3 + cos
(nr) 2'
(n7r)2

2 3
n=1 n=1

which implies
00
4 16(-1) n
'
(6.42)
3
n=1
n2

Therefore, we rewrite the infinite series in Equation (6.37) as

16(-1)n n7rx
-3
x2=+ 2 cos
2
(6.43)
n=1 (n71)

which is the Fourier cosine series of the function f (x) = x2.

Suppose we would like to integrate Equation (6.43) and express the solution as
the Fourier sine series representation of x3. This is done in the next example.
EXAMPLE 6.5. Consider
4 °° 16(-1) n n7rx
x2 = -+ 2 cos
2
(6.44)
3 (n71)
n=1

on the interval [0,2]. Integrate Equation (6.44) and express the solution as a Fourier
sine series representation of x3.
210 Chapter 6: The Calculus of Fourier series

Solution: Integrate Equation (6.44) from 0 to x. That is,

x
4 x
x 00 16(-1)n n71Y
T2 dT = - dT + 2 cos dT
0 3 o o n=1 (n7r) 2

x o0 x
4 16(-1)n n71Y
dr + 2 cos dT.
(n71) o 2
3 o n=1

Performing the indicated integrations yields

3X
4 °° 16(-1) n 2 n717
sin
0 n_1 (n71)2 n7r 2 0

This becomes

x3 4 °° 16(-1) n 2 n7rx
_-x+ 2 sin
3 3 n=1 (nr) [mu 2

-4 °O 32(-1)n mix
-x+ 3
n=1 (n7r)
3
sin
2
(6.45)

However, Equation (6.45) does not look like the Fourier sine series of 1(x) = x3.
Therefore, first we multiply Equation (6.45) by 3, which yields

°O 96(-1)n , n7rx
x3 =4x+ sin (6.46)
n=1 (n71)

Second, the right side of Equation (6.46) must be completely expressed as a Fourier
sine series on the interval [0, 2]. Therefore, we must express the 4x term in Equation
(6.46) as a Fourier sine series on the interval [0, 2]. We have
°O
n7rx
4x = bn sin ,
2
n=1

where
2
mix -16(-1 n
b= 4x sin 2 dx =
n7r
0

Thus,

4x=
°° (-)
161
n
sin
n7rx
(6.47)
n=1
mr 2
 Section 6.4: Integration of Fourier series 211

Combining Equations (6.46 and 6.47) yields

°O -16 (_i)12 n7rx 96(-1)n n7rx

x3 = sin sin
n_1
n7r 2 n=1 (n) 3
2

96(-1)n 16(1)2I nix

- n=(n)3
- n sin -
2

This is the correct Fourier sine series for f (x) = x3.

As Examples (6.3 and 6.5) show, term-by-term integration of a Fourier series

produces an infinite series. The infinite series, with a little manipulation, may be
written in the more useful Fourier series form.
Determining the value of an infinite series is one use of integrating Fourier series.
The following example clearly shows this aspect.
EXAMPLE 6.6. Consider the Fourier sine series of the function 1(x) = 1 on the
interval 0 < x < L,
00
n7rx
1= bn sin (6.48)
L
n=1

where

b=
n L
2 L
sinn7rx
L dx = 2 n
(1 - (-1)).
0 n7r

Integrating both sides of Equation (6.48),

x
n7rs
fl71(1 - (-1)n) f sin L ds,
f n=1

yields
X

x=(1-(-1))
°° n) I- L cos nlrs
2
n7r
n_ 1
[n7r L j0

-2 n x - 1]
(1 - (-1)) [cos_Z_
n_1 (n7f)2

00 00
_
2L
2 (1 - (-1)n) - (n7r)2 (1 - nx(6.49)
(-1))cos--.
n=1 n n=1
 212 Chapter 6: The Calculus of Fourier series

Equation (6.49) has the form of the Fourier cosine series for the function f (x) = x
2L_(1
where ao = - (-1)n). Finding ao in the usual way, we have
n=1 (nor)

ao=
1

L
L
xdx= L2 .

0
00
2L
Therefore the infinite series 2 (1 - ( -1)n) converges to L written as
n=1 (nit)
2

00
L
(nit)2 (1 - (-1)n)
2
n=1
for any positive number L.
Another use of term-by-term integration of Fourier series is determining the
Fourier coefficients, which is left as an exercise. This concludes our discussion
about integrating Fourier series. We move on to the Gibbs phenomenon.

EXERCISES 6.4

6.4.1. Term-by-term integration has other advantages. Using term-by-term integra-

tion, determine the Fourier coefficients bn by integrating both sides of the
following equation twice:
00
nx
sinh x = >bnsin , 0 < x < 271.
2
n=1

6.4.2. Using term-by-term integration, determine the Fourier coefficients ao and an

by integrating both sides of the following equation twice:

e- = ao + an cos nirx, 0 < x < 1.

n=1

6.4.3. Using term-by-term integration, determine the Fourier coefficients bn by inte-

grating both sides of the following equation twice:
00
fl7IX
cosh x = bn sin --, 0 < x < 2.
2
n=1

6.4.4. Show that

IT2
°° (-1)n
12 n2
n=1

Hint: Find the Fourier sine series for the function f (x) = x, then integrate the
Fourier sine series to get the Fourier cosine series of f (x) = x2 where L = it.
 Section 6.5: Fourier series and the Gibbs phenomenon 213

6.4.5. Show that

1 6 ((2 - n27r2)(-1)n - 2J
4
n=1 ren)

Hint: Find the Fourier sine series for the function 1(x) = x2, then integrate
the Fourier sine series to get the Fourier cosine series of f (x) = x3 where
L=7C.

6.5 FOURIER SERIES AND THE GIBBS PHENOMENON

In our study of PDEs, the stated problems and examples always had matching
BCs and ICs. That is, the ICs satisfied the BCs of the problem. The question
should arise, what happens if the ICs do not satisfy the BCs of the problem? Also,
in Chapter 4 and again in Chapter 6, we had examples of Fourier series that did
not converge to the function at the endpoints. In fact, the Fourier series seemed
to oscillate rapidly around any jump discontinuity the function had. For example,
consider the Figures (4.18 and 4.43) in Chapter 4 and Figure (6.1) in Chapter 6. We
indicated that the oscillations were known as the Gibbs phenomenon, and we stated
that the Fourier series would converge to the average of the jump discontinuity of
the function. In this section, we examine the Gibbs phenomenon.
An examination of the Gibbs phenomenon requires a more in-depth explanation
of the convergence of Fourier series. For example, consider the graph function
1(x) = x2 given in Figure (6.3) on the interval [-L, L]. We know the Fourier series

Figure 6.3: The graph of 1(x) = x2.

representation of 1(x) is

x2 _ L+ 2 °O 4L2(1)72
cos (6.50)
3 n_1 n7r 2 \ L
214 Chapter 6: The Calculus of Fourier series

Now, consider the following sequence of partial sums of the Fourier series given in
Equation (6.50) with their corresponding Figures(6.4-6.9):

Sl
()x _ LZ+
3
1 4L2(-1)n
(n)2 cos
rn7rxl
\LI
L2 4L2
3(n)2 cos \ L /
n=1

S2(x) _ L2 + r. 4L2(-1)n
cos
3 L (n7r)2 \L
= 3-
2
4L2 rx 4L2
cos
( 27rx
(
ncos ( L )+ ( )2

Figure 6.4: The graph of f (x) = x2 Figure 6.5: The graph of f (x) = x2
and the partial sum Sl (x). and the partial sum S2 (x).

L2

-L

Figure 6.6: The graph of f (x) = x2 Figure 6.7: The graph of 1(x) = x2
and the partial sum S5(x). and the partial sum Slo(x).

S L2 + 5 4L2(-1)n nirx
3
n=1
(n cosL().
LZ 10
4L2(-1)n nirxl
Sio(x) = + cos
n=1
(n
 Section 6.5: Fourier series and the Gibbs phenomenon 215

2f_ \n
2
Sis (x) = -a-- + :i:
15
4 n x,
cos (----)
n=1

Sioo(x) =
LZ ioo
+ :i:
4L2(-1)n
cos
nixn=1
(n7r)2 (-z--)

- L2

-L L -L L

Figure 6.8: The graph of 1(x) = x2 Figure 6.9: The graph of f (x) = x2
and the partial sum S15(x). and the partial sum Sloo(x)

It is easy to see that the sequence of partial sums of the Fourier series represen-
tation of the function x2 converge uniformly to the function x2. In fact, by partial
sum Sioo (x) the graph of the partial sum and the graph of the function are almost
indistinguishable, which means convergence is relatively fast. Note, each partial
sum, Sn, n = 1,2,3,..., is a trigonometric polynomial. This fact is very important
and the following theorem brings it to light.
Theorem 46. (Weierstrass approximation theorem) If the function 1(x) is contin-
uous and has period 2L, then it can be uniformly approximated by trigonometric
polynomials.

The proof of the Weierstrass 1 approximation theorem may be found in The

Elements of Real Analysis by Robert G. Bartle.
Remember, trigonometric polynomials are functions of x, and when we discuss
uniform approximation by trigonometric polynomials, we are actually considering a
sequence of trigonometric polynomial functions uniformly converging to the function
1(x) when f (x) is a continuous function.
What happens when the function f (x) is not continuous on the period 2L, but
has a finite number of jump discontinuities? This question is answered in part by
the following example.
EXAMPLE 6.7. Consider the Fourier sine series representation of the function
f(x)=5
1 Karl Weierstrass, (1815-1897) was the leading mathematical analysts in the second half of the
nineteenth century in Berlin
216 Chapter 6: The Calculus of Fourier series

on the interval 0 < x < L. Please note the period of this function is still 2L.
Remember, you must do the odd periodic extension shown in Figure (6.10) to
generate the Fourier sine series. The Fourier sine series is

-L L

Figure 6.10: The odd periodic extension of 1(x) = 5 on [-L, L].

00
10 [1 - (_i)12] nirx
f(x) = 5 =
n=1
n sin
4

Because of the term [1 - (-1)n], we recognize that for even natural numbers, n =
2n, S2n (x) = 0. Therefore, the following sequence of partial sums and graphs-
Figures (6.11-6.16)-are for odd n.

-L

-5

Figure 6.11: The graph of the odd Figure 6.12: The graph of the odd
periodic extension of f (x) = 5 and periodic extension of 1(x) = 5 and
the first partial sum Si(x). the first partial sum S3 (x) .

1 10[1 - (_i)12] n7rx 20 7rx

S1(x) _ sin = sin .
n?r 4 ?r 4
n=1

3
10 [1 - (_i)12] 20
S3 (x) _
n?r
sin
,

4
=
?r
sin
7rx
4
+ 20
37r
371x
-sin----.
4
n=1
 Section 6.5: Fourier series and the Gibbs phenomenon 217

Figure 6.13: The graph of the odd Figure 6.14: The graph of the odd
periodic extension of 1(x) = 5 and periodic extension of f (x) = 5 and
the first partial sum S7 (x) . the first partial sum S15 (x) .

S7(x) _ . 10 [1 - (-1)n] nix

L,
n=1
n sin
4

10 [1 - (-1)n] nix
S15(x) _ sin ----
4
n=1

5 5

-L -L

-5

Figure 6.15: The graph of the odd Figure 6.16: The graph of the odd
periodic extension of 1(x) = 5 and periodic extension of f (x) = 5 and
the first partial sum S99(x). the first partial sum S201 (x).

10 [1 - (_i)12]
sin
nix
S99(x) _
n=1
nit 4

2:i: 10 [1 _(_i)12]
S2o1(x) =
n sinnix
4
-
n=
218 Chapter 6: The Calculus of Fourier series

By the partial sum, S201 (x), we see that the Fourier sine series representation
of the function f (x) = 5 is converging to 5 at points away from the jump discon-
tinuities at x = 0, x = ±L. At the points x = 0, x = ±L, we notice an overshoot
of the Fourier series. This overshoot is known as the Gibbs2 phenomenon. As
n --f oc, we could expect the overshoots to disappear. However, in reality they
don't. It can be shown that the Fourier series converges to the average value at any
jump discontinuity. It should be noted that in higher mathematics, we can show
that the Fourier series actually has weak convergence to the endpoints of a jump
discontinuity. This weak convergence is from the left and from the right at a jump
discontinuity. However, the concept of weak convergence is beyond the scope of this
course.
Does the Gibbs phenomenon occur when we solve a PDE? Consider the following
example.
EXAMPLE 6.8. Suppose we are given the heat problem
on 02 u
O`lt O`lx2 '

subject to

and

u(x, 0) = x2 + x - 1.
We see that the IC has the value of -1 at x = 0 and 5 at x = 2. From our previous
work in Chapter 5, we know that the general solution is
00

u(x, t) = brie-(2 )2t sin

2
n=1
Applying the IC yields
00
n?f x
u (x, 0) = x2 + x - 1 = bri sin
2
n=1

Thus, we find that bri is given by

bri =
2
L f L , nirx 2
f (x) sin L dx = 2
0
L
(x2 + x - 1) sin 2 dx

f 2
(x2 + x - 1) sin 2 dx =
2 (1 + 7 (-1)"d)
n71

2Josiah Gibbs (1839-1903), an American physicist, brought the overshooting phenomenon to the
'

attention of the scientific community in a paper. However, it was first noticed by the mathematician
Henry Wilbraham and was rediscovered by the British during World War II while developing radar.
 Section 6.5: Fourier series and the Gibbs phenomenon 219

Therefore, the Fourier sine series representation of x2 + x - 1 is

16((-1)-1) 2(1+7(-1)) sinnix

x2 + x - 1 =
n=1
n 2
Figure (6.17) shows the graph of the polynomial 1(x) = x2 + x - 1 on the interval
[0, 2]. Figure (6.18) is the graph of the Fourier sine series representation of the
polynomial 1(x) = x2 + x - 1 on the interval [0,2]. Since 1(0) 1(2) 0, the
Fourier sine series representation of 1(x) = x2 + x - 1 will converge at 1(0) and
1(2) to 2. Therefore, the Fourier sine series representation of the IC 1(x) exhibits
the Gibbs phenomenon at the endpoints. Knowing bn, we can state the specific

x
x
1 1.5

Figure 6.17: The graph of f (x) = Figure 6.18: The Fourier sine series
x2 + x - 1. representation of 1(x) = x2 + x - 1.

solution of the problem. It is

((_)fl
t- 116 3 3 - 1) _ 2(1 +n (-1))1 e- 2 )2sin n2x
n-1 L J

Figure (6.19) shows the graph of u(x, t) for time 0 < t < 0.5. Please note how the

0.4 t

u(x, t)

Figure 6.19: u(x, t) for 0 < t < 0.5.

 220 Chapter 6: The Calculus of Fourier series

boundaries are fixed at 0. Also, the surface of the function has the curve x2 + x - 1
for 0 <x <2 subject to the decreasing exponent. The function 1(x) = x2 + x - 1
is shown as a bold line where it was produced by a Fourier series partial sum of
n = 500. Note how the IC does not quite follow the curve of u(x, t) at either end
at time t = 0.

EXERCISES 6.5

6.5.1. Consider the PDE

au(x,t) = 1.14a2u(x,t)
axe
at .

For each set of BCs and ICs, solve the initial value problem; using your favorite
mathematical software, graph the solution for 0 < t < 2. Clearly indicate the
initial temperature distribution on your graph and determine if the Fourier se-
ries expansion of the initial temperature distribution is accurately represented
on the graph of u (x, t).
u(o, t)= o
(1) Bcs: and IC: u(x, 0) = x-4.
t) = o
u(0, t) = 0
(2) BCs: and IC: u(x, 0) = x3 +2x- 1.
0
8x
8u(0, t) _ 0
8x
(s) Bcs: and IC: u(x, 0) = x2 -1.
/3 \ 0
I u 2,t)=
8u(0, t)
=0
8x
(4) Bcs: and IC: u(x, 0) = x2 + 2x + 4.
8u(ir, t)
8x
-0
6.5.2. Consider the PDE
a2u(x, t) a2u(x, t)
=1 axe
ate
For each set of BCs and ICs, solve the initial value problem. Using your
favorite mathematical software, graph the solution for 0 < t < 1. Clearly
indicate the initial displacement on your graph and determine if the Fourier
series expansion of the initial temperature distribution is accurately repre-
sented on the graph of u(x, t).
 Section 6.5: Fourier series and the Gibbs phenomenon 221

u(0, t) = 0 u(x,0) =x2 -3x+5

(1) Bcs: and ICs:
{ u(ir, t) = 0 8u(x,0) _ o
at
u(0, t) = 0 u(x, 0) = cos x
(a) Bcs: and ICs
Du 2(,t) Du(x,0)
= 0.
ax
=0 l atf
8u (0, t) u(x, 0) = 0
=0
(s) Bcs: ax and ICs:
{ 8u(x, 0)
at
_- 3 sin(5x).
u (3, t) =0

au (o, t)
-x; 0<x<.5
=0 J u(x, o) _
(4) BCs:
8x
and IC: x-1; .5<x<1
8u (1, t)
J dx =0.
at
6.5.3. Let a metallic rod 45 cm long be heated to an initial temperature which
is modeled by (x sin 2x + 3)°C. Suppose at t = 0, the ends of the rod are
plunged into an ice bath of 0°C and thereafter maintain this temperature.
Also, suppose no heat is allowed to escape from the lateral surface of the rod.
Note: You must model and explain your choice of boundary conditions for
this experiment.
(1) Solve the initial value problem if the rod is made of gold. Note: The
thermal diffusivity of these materials may be found in Appendix E.
(2) Using your favorite mathematical software, graph the solution for each
material, then determine the approximate solution for the following three
cases:
(a) Use only the first term in the series for u(x, t) to find the approxi-
mate temperature at x = 15 cm when t = 30 secs.
(b) Use the first three terms in the series for u(x, t) to find the approx-
imate temperature at x = 15 cm when t = 30 secs. Compare this
answer with the previous answer.
(c) Use the first 25 terms in the series for u(x, t) to find the approximate
temperature at x = 15 cm when t = 30 secs. Compare this answer
with the previous two answers.
(d) Explain, in a short essay, the differences in the above three approx-
imations.
(3) Determine the amount of time required for the temperature of the rod
to reach zero degrees at x = 15 cm.
6.5.4. Solve Exercise 6.5.3 if the rod is made of steel.
 222 Chapter 6: The Calculus of Fourier series

6.5.5. Solve Exercise 6.5.3 if the rod is made of tungsten.

6.5.6. Consider the following information:
(a) a perfectly flexible string of length 2ir ft,
(b) tension of 50 lbs/ft,
(c) mass density of 0.02 lbs/ft,
(d) vibrating motion which is entirely vertical,
(e) fixed left end at 0,
(f) free right end,
(g) initial displacement of 2x - 5,
(h) initial velocity of 0, and
(i) no gravity effects.

(1) Set up the mathematical model.

(2) Solve your mathematical model for any time t.
(3) Check that this solution satisfies the equation, the BCs, and the ICs.
(4) Using your favorite mathematical software, graph the solution. Then
write a short paragraph discussing the surface at the following times:
t = 1 sec, t = 5 secs, and t = 10 secs.
Chapter 7
Separation of Variables: The
Nonhomogeneous Problem
7.1 INTRODUCTION
In Chapter 5, we discussed the technique of separation of variables for the linear
homogeneous problem using homogeneous Dirichlet and Neumann boundary con-
ditions. We summarize this solution method with the following:

1. Determine that the PDE and BCs are linear and homogeneous.
2. Assume that the solution is a product of two functions. One function is
possibly of time, while the other function is composed of spatial variables.
3. Substitute this product into the PDE and move all functions of the time
variable to one side, while placing all other components on the opposite side
of the equation.
4. Set both sides equal to the same separation constant, thus breaking the orig-
inal equation into two new equations and separating the variables.
5. Repeat steps (2), (3), and (4) on equations with more than one spatial variable
until you have an ODE for each spatial variable.
6. Substitute the product form of the solution into the BCs to obtain BCs for
the spatial ODEs.
7. Solve the ODE boundary value problem(s) to obtain the eigenvalues and eigen-
functions.
8. Solve the time ODE using the previously found eigenvalues.
9. Use the principle of superposition to get the general solution as an infinite
series.

223
224 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

10. Use the ICs and orthogonality to determine the constants for the specific
solution.

In this chapter, we discuss the separation of variables technique when applied to

various linear nonhomogeneous problems. The nonhomogeneous problem is a very
realistic mathematical model of experimental physical phenomena. For instance,
we can use it to model a forcing function for the wave equation and an internal
heat source for the heat equation. Also, we can model homogeneous heat and wave
equations whose solutions are not always dependent on homogeneous boundary
conditions. These examples only scratch the surface of a host of scenarios that we
could also model. I've selected a few of these, and the following are the ones we will
consider remember all equations are linear:

nonhomogeneous PDEs with homogeneous BCs,

homogeneous PDEs with nonhomogeneous constant BCs,
homogeneous PDEs with nonhomogeneous variable BCs, and
nonhomogeneous PDEs with nonhomogeneous (constant or variable) BCs.

In the next section, we start our discussion of the nonhomogeneous problem by

solving the problem of the nonhomogeneous PDEs with homogeneous BCs.

7.2 NONHOMOGENEOUS PDES

WITH HOMOGENEOUS BCS
A typical problem of this type is solving for the temperature distribution in a
uniform rod of length L, with perfect lateral insulation and some type of constant
or nonconstant internal heat source. The mathematical model is
0126 012 ZG

01t 01x2 + Q(x't).

There are a host of different BCs from which we could choose, but for purposes of
this problem we chose homogeneous Dirichlet BCs modeled as

u(O,t) = 0
J
u(L,t) = 0.
Also, there must be an IC, and we model it as

u(x, 0) = f (x).

The general technique for solving these types of problems is discovered by re-
membering how we solved a linear second-order nonhomogeneous ODE: First, solve
the homogeneous part. Second, solve the particular part. However, solving the
 Section 7.2: Nonhomogeneous PDEs with Homogeneous BCs 225

homogeneous part of a PDE by separation of variables requires the BCs to be ho-

mogeneous. This means we must develop some new and interesting strategies. The
problem that I chose to solve in this section does not demonstrate these strategies,
but it sets the stage necessary to develop them.
Step 1: Solve the related homogeneous problem by changing Equation (7.1) to

0126

at -' 012 ZG

ax2

Remembering our techniques from Chapter 5, we assume the solution u(x, t) is of

the form u(x, t) = G(t)q5(x). Performing the necessary differentiations and replacing
u(x, t) in Equation (7.4), we obtain

G'(t)q5(x) = kG(t)q"(x).

Separating variables and setting the resulting equation equal to a separation con-
stant, we obtain
G'(t) - qY'(x) -
kG(t) fi(x)

This can be written as two separate equations,

G'(t)
kG t
and

qY'(x) -
fi(x)

Now, remembering that the BCs must be separated, we get

5(0) = 0

(L) =0.
We now have two simple ODEs: a spatial ODE with BCs and a time ODE. Up
to now, this is familiar, but here is where we enter new territory. Basically, we
throw away the time ODE and only solve the spatial ODE for its eigenvalues and
eigenfunctions.
The eigenvalues and eigenfunctions that form the solution to the spatial ODE
can be found in Chapter 5. They are
nor 2
eigenvalues: An =
L
n=1,2,3 , .... (7.5)
n7rx
eigenfunctions: q5( x) = cn sin L
226 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

Step 2: Assume that the general solution for Equation (7.1) should be a Fourier
series. In particular, we assume that it will be a Fourier series expanded in the
eigenfunctions, Equation (7.5), multiplied by some function of time, t. Therefore,
assume the general solution for Equation (7.1) is
00
nzx
u(x, t) = cn (t) sin
L

Notice in Equation (7.6) the former constant, cn, is now required to be a function
of time, t.
Step 3: Start working on the particular part by expanding Q(x, t) from Equation
(7.1) in a Fourier series using the eigenvalues and eigenfunctions that were found in
Step 1. This means,
00
n7rx
Q(x, t) _ do sin L
n=

Since we are already using the name cn in Equation (7.6), we selected the constants
do for Equation (7.7).
We must determine the constants do . From Chapter 4, we know

2 L
d72=-
L Q(x, t) sin L dx.
0

From Chapter 5, we know that do must in fact be a function of time, t, and only
constant with respect to x. Thus, if we assume that we can perform the integral
and let an (t) = dn, we can write Equation (7.7) as
00

Q(x, t) _ ant) sin (nLx) .

What we just did in Step 3 is called eigenfunction expansion. This means we

expanded the nonhomogeneous term, Q(x, t), from Equation (7.1) in the eigenfunc-
tions of the related homogeneous problem, Equation (7.4).
ZG
Step 4: Using Equation (7.6), we find and axe .
at
In this step, we must satisfy all the conditions required to do term-by-term
differentiation of both the Fourier sine series and the Fourier cosine series. Since
we assume that time is a continuous function with no sharp bends, we can say
oo 00
a26 a d n7rx
cn (t) sin = - (cn (t)) sin
C7t
n=1
L
n=1
dt L
00
n7rx
n (t) sin
L
 Section 7.2: Nonhomogeneous PDEs with Homogeneous BCs 227

Since the BCs are homogeneous and we assume the rod to be continuous, we
have the Fourier sine series as a continuous function. Thus, if we differentiate the
Fourier sine series with respect to x, we have
00
nLx 11
cn (t) sin (nLx) I = cn (t) (sin C
8x 8x ll

00
nor n7rx
L cn (t) cos L
n=1

Since the BCs are homogeneous and we assume the heat flow to be continuous, we
have the Fourier cosine series as a continuous function. Thus, we can differentiate
with respect to x the Fourier cosine series. This yields
a2u a °°
ax2
=
ax n-1
[1cn(t)cos()]
12?C

L
n?Cx

00

_
n=1
L cn (t) (cos
())
ffl7r
\ L / 2 cn (t) sin (nLx) .

Step 5: Solve Equation (7.1) by replacing the terms of the equation by what we
have determined they equal in Steps 1 through 4. That is,
au _ a2u
k
at ax2 + Q(x,t)
becomes
00 00

Lxl --k \ L /Lcn(t)sin/ nLx)

n=1

00
nix
an (t) sin (-;-).
n=1

We now have the Fourier sine series expansions over the same interval, [0, L], in
all three terms of the equation. This means that the Fourier coefficients must be
equal. Therefore, we only have to solve the ODE
2
c(t) = -k () c(t) + a(t).
228 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

nor 2
Since L = we may write

c( t) _ -kAnCn (t) + an (t)

or

cn(t) + kAncn(t) = an(t). (7.11)

Equation (7.11) is a first-order ODE. To develop the solution for cn (t), we use
the integrating factor
t.

Doing so yields a solution for cn (t):

t
cn(t) = e-pant ekAnran(r) dr + e-kantbn

Note that the integral does not have a lower limit. We assume that when both sides
of the equation are integrated, we will get constants of integration on both sides of
the equation, which we combine and call b.
Step 6: Write the solution to Equation (7.1) and apply the IC, Equation (7.3).
We have
t
u(x, t) = [e_nt ekanTan(r) dr + e-kantbn sin n7rx (7.12)
L
n=1

Applying the IC, we find a general form of u (x, 0) as

00
n7Cx
u(x,O) = f(x) = An (0) + bra sin
n=
(-z)'
where
t
An(0) = ekanTan(T) dT
t=o
This implies
L
bn = -An(0)+ L f0 f (x) sin (nLx) dx. (7.13)

Therefore, the complete solution to Equation (7.1), subject to the BCs, Equa-
tions (7.2), and IC, Equation (7.3), is

u(x, t) _
n= 1
[et t
ekanTan(r) dT +- e-kantbn sin n7rx
L
,

where the constant bra is given by Equation (7.13).

The following example illustrates this method. Also, see the graphical represen-
tation of the solution provided in Figure (7.1).
 Section 7.2: Nonhomogeneous PDEs with Homogeneous BCs 229

EXAMPLE 7.1. Consider

012 u u
012
(7.14)
ate axe + Q(x)'
where

Jx ,

Q(x) _
2r -x r < x2
< ir, ,

subject to
u(o, t) = o
(7.15)
{ u(2ir, t) = 0
and
u(x, 0) = x2 - 27rx

8u(x, 0)
l = 0.
at
Find the time-dependent solution for u(x, t). For this problem, the BCs are linear
and homogeneous. However, the PDE in Equation (7.14) is not homogeneous.
Therefore, we need to follow the method developed in this section.
Step 1: Solve the related homogeneous problem. We basically ignore the term Q(x),
and we derive the eigenvalues and corresponding eigenfunctions for
012 u 012 u
at2 01x2

subject to the BCs, Equation (7.15). Remember, we are only solving for eigenval-
ues and corresponding eigenfunctions of the spatial problem after we separate the
variables. Thus, we must solve
o"(x) = -A(x). (7.17)

Equation (7.17), subject to the BCs indicated in Equation (7.15), has eigenvalues
and corresponding eigenfunctions of

nx
Pn (x) = bn sin
2

Step 2: Assume the general solution for Equation (7.14) has the form
00

u(x, t) _ bn(t) sin (2) . (7.18)

n=
230 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

Step 3: Expand the term Q(x) in a Fourier series using the eigenvalues and corre-
sponding eigenfunctions found in Step 1. That is,
00
nx
Q(x) _ an sin 2) .

n=1

In Equation (7.19), the an term is not a function of time, t. The reason for this is
that the function Q(x) is only a function of x. Solving for an yields
f2
an = J Q(x)sin(2) dx
0

2r

_ x sin (2) dx + J (2ir - x) sin (2 ) dx }

l J

nor
8 sin 2

Therefore,
nit
00 8 sin nx
Q(x) _ -J rn2
2 sin --).
2
(7.20)
n=
a2 u a2 u
Step 4: Using Equation (7.18) find ate and axe . Remember we must be able to
justify term-by-term differentiation. For the derivatives with respect to time, t, we
expect time to be continuous. Thus,
00
a2u nx
ate
n (t)sin().
2
(7.21)
n=

Since the string is continuous and the BCs, Equation (7.15), are homogeneous, we
know that we can perform term-by-term differentiation with respect to the spatial
variable, x. It is
a2 u oo

ax2
= - -n 2 bn (t) sin
2
nx
2
(7.22)
n=1

Step 5: Solve Equation (7.14) by using Equations (7.20, 7.21, and 7.22). We have
00 00
()2

bn(t) sin (-i-)

bn(t) sin (2 I - n=1 -
+ 00 8 sin (2 ) (nx l .
sin
n=1
ant \2l
 Section 7.2: Nonhomogeneous PDEs with Homogeneous BCs 231

Since the Fourier series are all in the same eigenvalues and corresponding eigen-
functions over the same interval, we know the Fourier coefficients must be equal.
Therefore,
nit
fl\2 8 sin ( 2
b(t) = - () b(t) + itfl2
or, written in a more familiar form,
bii t fl \2 8 sin (2 )
() + () b(t) =
This is a second-order linear nonhomogeneous constant coefficient ODE, with a
solution of

b (t) =ate,
cos/

I
nt l
2
/\
)+dnsin)+ I
nt
2
32in
s
n4
7rn4
nit
2 )

Step 6: Write the solution for u(x, t) and apply the ICs (Equation (7.16)). We have

n=1
ate, cos 12 I + do sin (2
nit

)
32sin(_-)1
itn4 2 jsin(). nx+
2

Applying the first IC, u(x, 0) = x2 - 2irx, yields

u(x, 0) = x2 - 2?Cx =
00

n=1
[aj +
flit
32sin()]
2

n
ur4 s).
in(2
Thus,
nit
- -32 sin 2 2ir
nx
an -
in4
+ it-1 (x2_2itx)sin
2
dx
o

-32 sinn2 ((-i)n - 1).

+ 16
in4 n3it
au(x, 0)
App1Ying the second IC = 0, yields
Y
at
00
au(at o) = o = flx\
n=1
[() dam] sin (2
which indicates do = 0 for all n. Therefore, the complete solution is
nor
32 sin nx
u(x, t) _
°O
an cos (2 fit
-- n4T) 1 sin (2 )
n=1
 232 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

where an is given by Equation (7.23).

In Figure (7.1), the solution is shown for 0 < t < 20. Note, the boundaries at x = 0
and x = 2ir remain fixed at 0, the initial displacement is graphically illustrated by a
bold-faced curve, and the repeating curve caused by the ICs and damping function
Q(x)

Figure 7.1: u(x, t) for 0 < t < 20.

The general method and problem we have solved in this section, nonhomoge-
neous PDEs with homogeneous BCs, is the basis for the other nonhomogeneous
problems we will solve later in this chapter. So, it is well worth your time to fully
understand the method. My suggestion is to work through the problem by picking
a value for k, such as 1, and some simple function for Q(x, t), such as xt. After
solving the problem with values for k and Q(x, t), you will notice that we only add
a couple of steps to the solution method of separation of variables, from the chapter
introduction. At the end of this chapter, we summarize the steps to follow when
solving the nonhomogeneous problem by the technique of separation of variables.

EXERCISES 7.2

7.2.1. Consider the PDE

z
tht
+ Q (x' t)'
8t k 8x2
 Section 7.2: Nonhomogeneous PDEs with Homogeneous BCs 233

Find the time-dependent solution for u(x, t) subject to the IC

u(x, 0) = f (x),
and the following sets of BCs:
5u(L't)
(1) au(0't) =0 and =0 .
8x 8x
(2) u(0, t) =0 and 0.

(3) t) = 0 and u(L, t) = 0.

7.2.2. Consider the PDE

au a2u
-t
at ax2 + xe ,

subject to

BCs: and IC:u(x, 0) = x.

au(ir, t) _ 0
ax
Find the time-dependent solution for u(x, t), and using your favorite mathe-
matical software, plot your solution. By inspection, determine if the graph of
your solution is reasonable for the problem.
7.2.3. Consider the PDE
au _ a2 u
at ax 2 -x cos t,
subject to

x, 0<x<ir
BCs: and IC: u(x,0) _
2ir - x it < x < 2ir.

Find the time-dependent solution for u(x, t), and using your favorite mathe-
matical software, plot your solution. By inspection, determine if the graph of
your solution is reasonable for the problem.
7.2.4. Consider the PDE
z 2 z
c + Q(x, t).
ate - ax2
Find the time-dependent solution for u(x, t) subject to the ICs
5u(x' 0)
u(x, 0) = f (x) and = g(x),
and the following sets of BCs:
 234 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

(1) u(0, t) =0 and 0.

5u(0'
(2) t) = 0 and au(L' t) = 0.
8x 8x
(3) a 0' t) = 0 and u(L, t) = 0.
7.2.5. Consider the PDE
92u 52u
ate axe -}- x sin t,

subject to the BCs

as°'t)=0andtht(5x't)

=o
and to the ICs
-x, 0<x<1.5 8u(x, 0)
u(x, 0) _ and at =0.
5x-9 1.5<x<2
Find the time-dependent solution for u(x, t), and using your favorite mathe-
matical software, plot your solution. By inspection, determine if the graph of
your solution is reasonable for the problem.
7.2.6. Consider the PDE
Du D2u
16 -}- y sin t,
at a2
subject to the BCs
8u(0, t) _ 0
ay
and u (2 t) = 0 ,
and IC

u(y, 0) = y
Find the time-dependent solution for u(y, t), and using your favorite mathe-
matical software, plot your solution. By inspection, determine if the graph of
your solution is reasonable for the problem.
7.2.7. Consider the PDE
Du D2u D2u -t
y) e
Dx 2 + p`l y 2 + `x + ,
Dt

subject to the BCs

y, t) _ 0 Du(x, 0, t)
u (0 ' y't) = 0 = 0 and u (x t) =0
Dx ' Dx
 Section 7.2: Nonhomogeneous PDEs with Homogeneous BCs 235

and IC

u(x,y,0) =x+y.
Find the time-dependent solution for u(x, y, t), and using your favorite mathe-
matical software, plot your solution for several different times t. By inspection,
determine if the graph of your solution is reasonable for the problem.
7.2.8. Consider the PDE
a2a a2u a2u
ate axe + a 2 + sin t cos xy,
y
subject to the BCs
au(ir, y, t) = 0
ax '
u ,y,t
2
=0
au(x, 0, t) _ 0
u(x, 2ir, t) = 0
ax '

and the ICs

au(x, y, 0) _ 0
ux 0 = sin x and
Find the time-dependent solution for u(x, y, t), and using your favorite mathe-
matical software, plot your solution for several different times t. By inspection,
determine if the graph of your solution is reasonable for the problem.
7.2.9. Consider the PDE
a2 u a2 u au
4 axe -2 + g, where g is gravity,
ate at
subject to the BCs
u(0, t) = 0 and u(2ir, t) = 0
and the ICs
6x
0<x< 71

' au(x, 0)
u(x,0) = andd =0.
at
2 (2ir x) 71 <x < 27r
71 2 -
Find the time-dependent solution for u (x, t), and using your favorite mathe-
matical software, plot your solution. By inspection, determine if the graph of
your solution is reasonable for the problem. This problem reflects vibrations
which arise from a gravity field in a medium with resistance proportional to
the velocity; the ends of the string are fixed at the same height.
 236 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

7.2.10. Consider the PDE

a2u a2u
ate 9 a-e + g, where g is gravity,

subject to the BCs

au
u 0 t )=0 and t) =0
( ax
and to the ICs
5u(x' 0)
u(x, 0) = 0 and = 100.

Find the time-dependent solution for u(x, t), and using your favorite mathe-
matical software, plot your solution. By inspection, determine if the graph of
your solution is reasonable for the problem. This problem reflects a vertical
flexible rod of length it (with its left end rigidly attached to a freely falling
lift, while its right end is free) that, having attained a velocity of 100 ft/sec,
stops instantaneously.

7.2.11. Given a one-dimensional rod of length it m. A heater moves along the sur-
face of the rod with constant velocity of 1 cm/min. Thus, a convective heat
exchange takes place. The flow of heat from the heater to the rod is 6e-at
watts where $ is the convection coefficient. Suppose the ends of the rod are
held at 0°C, and the initial temperature distribution is 3 sin(5x).
(1) Mathematically model this problem.
(2) Suppose $ _ .12, and solve the time-dependent problem.
(3) Graph the time-dependent problem.
(4) Solve the steady-state problem.

7.3 HOMOGENEOUS PDE WITH

NONHOMOGENEOUS BCS
In this section, we cover two different types of nonhomogeneous BCs. First, we
consider constant, nonzero BCs. Second, we examine BCs that contain a variable.
Although the methods are similar for each, the first is straightforward while the
second requires some intuition.

7.3.1 Homogeneous PDE Nonhomogeneous Constant BCs

Consider a perfect laterally insulated uniform rod of length L, with thermal con-
ductivity k, and some type of initial condition. Also, suppose at x = 0, the end of
the rod is in a bath held at a constant temperature of T1; at the end x = L, the rod
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 237

has some type of insulation, which constricts the flow of heat to a constant value of
A2. The mathematical model for this problem is
z
(7.24)
8t 8x2 >

subject to the BCs

u(0, t) = Tl

8u(L, t) _ A2
8x
with IC
u(x,O) = f(x). (7.26)

In the last section, we said the nonhomogeneous PDE with homogeneous BCs
problem was the basis for all the other nonhomogeneous problems we will solve
in this chapter. However, how is a homogeneous PDE with nonhomogeneous BCs
turned into the basic nonhomogeneous problem? The answer is by creating a new
function. So what does this new function look like? To answer this question, we
must consider all the things that the new function must do.
The new function must have homogeneous BCs. It must replace the function
u(x, t) with no loss of information. Finally, the new function must turn the non-
homogeneous BCs into a source term that makes the PDE nonhomogeneous. The
question that still remains is how? To answer this question, consider what we have
learned so far in the solution techniques for PDEs, particularly PDEs that involve
temperature distribution in a one-dimensional rod. So far, two techniques have
been studied: separation of variables and steady-state. We know that we cannot
use the separation of variables technique yet. So, by default, that leaves us with the
steady-state technique. Let's investigate this method to determine what it gives us.
Basically, in the steady-state problem, we assume that u(x, t) = uSS (x) This
.

a 2uSS = d 2uSS .
assumption implies that cu ss = 0 and The general steady-state
at axe dx2
solution for Equation (7.24) is
u38 (x) = C1 X + C2.
(7.27)

Applying the BCs to Equation (7.27) yields

u88 (x) = A2x + T1. (7.28)

We now want to create a new function, using Equation (7.28), that has homo-
geneous BCs. Suppose we create a function v(x, t) such that
v(x,t) = u(x,t) - uss(x). (7.29)

Then
v(0, t) = u(0, t) - u33(0) = T1 - T1 = 0
238 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

and
av(L,t) _ au(L,t) - &u33(L)
=A2-A2=0.
ax ax ax
Therefore, v(x, t) is a function that has homogeneous BCs. Ideally, we'd like to
replace u(x, t) with v(x, t) in Equation (7.24). To do this, we must determine
au D2 u
u(x, t), , and axe . From Equation (7.29), we have
at
u(x, t) = v(x, t) + u33 (x). (7.30)
Using Equation (7.30) and Equation (7.28), we find that
au av 9u88 (x) _ av a(A2x + T1) _ av
- at (7.31)
at - at + at - at + at
and
a2Z.6 a22J a2uss (x) a2v a2 (A2x +T1) aax2v
. (7.32)
axe + axe
axe axe + axe e

Thus, u(x, t) and v(x, t) are interchangeable in Equation (7.24). However, instead
of ending up with a nonhomogeneous PDE with homogeneous BCs, we have a
homogeneous PDE with homogeneous BCs, as summarized below:
av a2 v
(7.33)
at - ax2 ,
subject to
v(0, t) = o
(7.34)
av(L, t) _ 0
ax
We know from our studies in Chapter 5 that Equation (7.33), subject to Equation
(7.34), has the solution
r (2n- 1) irx 1
sin (7.35)
L 21'
n=1

r ,n=1,2,3,....
n= [(2n_1)]2
2L
However, Equation (7.35) only gives us the solution for v(x, t), not u(x, t). Equation
(7.30) and Equation (7.35) determine the general solution for u(x, t). Thus, the
general solution is
u(x, t) = v(x, t) + u33(x)

_nt (2n - 1) 7rx

bne sin + A2x + T1. (7.36)
n=1
2L
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 239

To avoid any confusion as to the contents within the summation sign, this may be
rewritten as
00
u(x, t) = A2x + Tl - bne-t sin [[ (2n 1
. (7.37)
n= 1 J

Applying the IC
u(x,0) = f(x),

we can determine the specific solution

00
u(x 0) = f (x) = AZx + Tl + bn sin
r(2n-1)x1
n=1 L
ZL

which can be written as

00
r (2n - 1) irx 1
f (x) - AZx - Tl = bn sin I
n-
2L J

Using orthogonality, we find

L
2n 2L) l
b= L
f [f (x) - AZx - Tl] sinLf dx. (7.38)

Thus, the specific solution to the problem in Equation (7.24), subject to the
BCs, Equation (7.25), and the IC, Equation (7.26), is
00
(2n - 1) 7rx
u(x't)=A2x+T1 + >bn e-pant sin 2L
n=1

where bn is given by Equation (7.38).

The following example illustrates this method. Also, there is a graphic repre-
sentation of the solution.
EXAMPLE 7.2. Consider
5u _ 52u
(7.39)
5t 35x2'
subject to
u(o,t) _ -1
(7.40)
u(7r, t) = 1

and

u(x, 0) _ - cos 7x. (7.41)

240 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

Find the time-dependent solution for u (x, t).

Since the PDE is linear and homogeneous, and the BCs are linear but not homo-
geneous, we must determine the steady-state solution, uss (x). The general steady
state solution is
u33 (x) = c1x + C2.

Applying the first boundary condition, u (O) _ -1, yields c2 = -1. Applying the
second boundary condition, u(it) = 1, yields c1 = 2 . Therefore, the specific steady-
state solution is
2
u33 (x) _ -x - 1. (7.42)
71

Letting the function v(x, t) = u(x, t) - u33 (x), we find v(0, t) = 0 and v(ir, t) = 0.
Also,
av au
at -
2v 52u
.
axe axe
Thus, we have linear homogeneous PDE and BCs in terms of v(x, t). We can now
apply the separation of variables technique. We should recognize from Chapter 5
that
av a2 v
3
at = axe
subject to
v(0, t) = o

v(ir, t) = 0,
has, as a general solution,
00
bne-3n2t
v(x, t) _ sin nx.
n=1
Therefore,
00

u(x, t) = -x - 1+ bne-3n2t
sin nx.
71
n=1

Applying the IC, Equation (7.41), yields

00
2
u(x, 0) _ - cos 7x = -x -1 + >bsinnx,
r n=1
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 241

which can be written as

- cos 7x - - x + 1 = bn sin nx.

r n=1

Using orthogonality of the Fourier sine series provides the value for bn, which is

u(x, t)

00

Figure 7.2: u(x,t) for 0 < t < 0.05.

= 2 1 - 2 x - cos 7x) sin nx dx = 98(n -

(-1))
7r Joo 71 n(n2-49)
7Cn (n 49)

Note: When n = 7, bn = 0 by orthogonality. Therefore, the complete solution for

u(x,t) is

49 ( e_3tsinx e-12t sin 2x e-27t sin 3x

u(x,t) =7r2 -x- 1- _
4
+
15
+
5

49 1 e-48t sin 4x e-75t sin 5x 5e-108t sin 6x

?f 22 + 10
+
39
I+

00
98(n- 2-(_e_3n2tsinnx;
7rn (n 49)
n=8
242 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

or it may be written as
6 n
u(x, t) _ x -1 + 98 n 4g) e-32t sinnx

00

+
98(n-(-1)) e -3n2t sinnx.
n=8
urn (n2 - 49)

In Figure (7.2), the solution for u(x, t) is shown for 0 < t < 0.05. Please note the
boundaries are at -1 when x = 0 and 1 when x = ?C. Also, the initial temperature
distribution is indicated by a bold-faced curve.
Now, a question that should come to your mind is: The method works great
for the heat equation, but how does it apply to the wave equation? This is a very
reasonable question considering the wave equation does not have a steady state
solution. But, the wave equation does have an equilibrium, and the equilibrium
state for the wave equation is equivalent to the steady-state solution of the heat
equation. Thus, this technique will work with the wave equation.
Although the homogeneous PDE with constant nonhomogeneous BCs is an in-
teresting problem, the homogeneous PDE with variable nonhomogeneous BCs is far
more interesting. We solve this problem in the next subsection.

7.3.2 Homogeneous PDE Nonhomogeneous Variable BCs

Consider the temperature distribution in a one-dimensional uniform rod with perfect
lateral insulation. Suppose that the boundary, x = 0, is placed in perfect thermal
contact with a medium that changes temperature with time t; at the boundary
x = L, the flow of heat is described by a function that changes with time t. Also,
suppose we know the initial temperature distribution in the rod.
The mathematical model that describes this physical situation is
26
(7.43)
at = ax2 ,
subject to
u(0,t) = a(t)
(7.44)
8u(L, t)
ax
and

u(x,0) = f(x). (7.45)

As in the previous subsection, our goal is to determine a function that has

homogeneous boundary conditions and can model the temperature distribution in
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 243

the rod. The approach we will take is called construction. As the name suggests, we
will construct a function v(x, t) such that v(x, t) = u(x, t) - r(x, t), with v(0, t) = 0
av(L t)
and = 0. This, in turn, requires r(x, t) to have the following properties:
ax
r(0, t) = a(t)
(7.46)
ar(L,t)
8x
-
Using Equation (7.46) as a starting point, we realize that integrating both sides
ar(x t)
of = $(t) with respect to x yields r(x, t) = x$(t) + c(t). (Note: As
ax
far as the variable x is concerned the function c(t) is a constant.) Furthermore,
applying the first BC, r(0, t) = a(t), we find a(t) = c. Thus, it would seem
reasonable to assume that r(x, t) = x$(t) + a(t). This representation of r(x, t)
yields the necessary properties described in Equation (7.46). However, there are an
infinite number of functions r (x, t) that may be generated or picked. For instance,
rx
r (x, t= a(t) x cos L+ cos x- -$(t) sin will also be a continuous function
that satisfies the necessary properties in Equation (7.46). Although r1(x, t) will
work, it is a much more complicated equation than r(x, t) = x$(t) + a(t), which we
will now use. We may now construct v(x, t) and replace u(x, t) in Equations (7.43
and 7.44) to yield

v(x,t) = u(x,t) - r(x,t) = u(x,t) - (x$(t) + a(t)). (7.47)

This implies that

8v - 8u 8r
- 8u (x$'(t) + a'(t)).

Solving for yields

ields
at

at - at + (x$'(t) + a '(t)),
a2 u .
while axe yields

a2v a2u a2r a2u

ax2 ax2 ax2 ax2.
Therefore, replacing u(x, t) in Equation (7.43) with an equivalent equation in v(x, t)
gives
2
+ (x/3'(t) + a'(t)) =
244 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

Thus, we now solve the nonhomogeneous PDE

av a2v
= k 2 - (x$ ,(t) + a '(t)) , (7.48)
at ax
subject to the homogeneous BCs
v(0,t) =0
(7.49)
av(L, t) _ o
ax
To do so, we apply what we learned in section two of this chapter to Equation
(7.48), subject to BCs, Equation (7.49).
Step 1: Solve the related homogeneous problem,
av a2 v
at k axe
for eigenvalues and eigenfunctions. From Chapter 4, we know the eigenvalues and
eigenfunctions are

2L [(2n_1)7r12 n= 1,2,3,...
nlx = Cn, Slri [/Xx]
Step 2: Assume the solution for v(x, t) is a Fourier series in the eigenfunctions
just found. We have

cn(t) sin [/cx], (7.50)

n=1

where An =
[(2n1) 2

n = 1, 2, 3,... Remember cn is a function of time t.

,
2L
Step 3: Expand the term (x$' (t) + a' (t)) in a Fourier series using the eigenvalues
and eigenfunctions found in Step 1. This yields
00

(x$'(t) + a'(t)) = d sin

where
L 1

CZn = L (x$'(t) + a(t)) Slri [x] CL(t).

0

(x$'(t) + a'(t)) a(t) sin [x]. (7.51)

=
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 245

av a2 v
Step 4: Find and axe . Remember, this requires us to satisfy all of the con-
at
ditions necessary to do term-by-term differentiation of a Fourier series. Assuming
all the conditions were met, we obtain
av °O
_ c( t) sin [\/x] (7.52)
n=1

and
00
a2v
axe =-: n=1
AnCn (t)Slll [\/cx] . (7.53)

Step 5: Solve Equation (7.49) by replacing the terms of the equation with what
we determined they equal in Steps 1 through 4. Thus,
av a2 v
_ (x$'(t) + a (t) )
at ax2
becomes
00 00 00

c' (t) Slri [\/cx] = -1C Ac (t) Slri [\/ix] - Cln,(t) Slri [\/cx]
n=1 n=1 n=1
We now have the Fourier series expansions with respect to the same eigenvalues over
the interval [0, L], in all three terms of the equation. This means that the Fourier
coefficients must be equal. Extracting the coefficients yields the ODE
c(t) _ -1CAnCn(t) - an(t)
or

c' (t) + kAncn (t) _ -an (t) . (7.54)

Equation (7.54) is a first-order ODE. We develop the solution by using the

integrating factor

Thus, the formal solution for cn (t) is

t
cn(t) _ -e-leant ekanTan(T) dT +

Note that the integral does not have a lower limit. We assume that when both sides
of the equation are integrated, we get constants of integration on both sides of the
equation, which we combine and call b.
Step 6: Write the solution to Equation (7.49). We have
t
v(x, t) _ -e-leant ekanTan(T) dT + bne-karat

sin nx
n=1
246 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

We cannot apply the initial condition since this is the solution to v(x, t), and we
want the solution to u(x, t). But, from Equation (7.47) we know

u(x,t) = v(x,t) + r(x,t) = v(x,t) + x,6(t) + a(t).

Substituting, we get

u(x,t) = x$(t) +a(t) +

t
_e-leant ekanTan(T) dT + sin [x] (7 .55
n=1

We now apply the IC, Equation (7.45), to Equation (7.55) yielding

u(x, 0) = f (x) = x,Q(0) + a(0) + A(0) + bn sin [x]

n=1
t
where A(0) = ekanTan (T) dr t=o

Rewriting this equation yields

f (x) - (x,3(0) + a(0)) _ An,(0) + bn, sin [x].

n=1

Using the orthogonality of the sine function, we see that

fL
bn = -A(0) + L [1(x) - (x$(0) + a(0))] sin [x] dx. (7.56)

So now the complete solution to Equation (7.43), subject to BCs, Equations

(7.44), and an IC, Equation (7.45), is

u(x,t) =x$(t)+a(t) +

n-1
[_et t
dT + e-kantbn sin [x]
where bn is given by Equation (7.56).
The following example illustrates this method, and it provides a graphic repre-
sentation of the solution.
EXAMPLE 7.3. Find the time-dependent solution for u(x, t) if

8u 82u
(7.57)
8t 8x2'
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 247

subject to
u(0, t) = a(t) = e_t
(7.58)
/71 \
u(,t) =$(t)=t
and
u(x, 0) = cos 3x. (7.59)

The PDE in Equation (7.57) is linear and homogeneous. However, the BCs, Equa-
tion (7.58), are linear but not homogeneous. Thus, to use the separation of variables
technique, we must develop a new PDE in terms of v(x, t), where
v(x, t) = u(x, t) - r(x, t). (7.60)

This means the function r(x, t) must equal e-t when x = 0. It must equal t when
x = . The BCs in Equation (7.58) suggest that r(x, t) = a(t) + x (3(t) - a(t)) =
2 L
e-t + 2x t - e_t . (Note: Although r x t is one of the easier continuous functions
to work with, there are an infinite number which may be found.) Thus, r(0, t) = e-t
andr r2,t =t.
Since v(x, t) = u(x, t) - r(x, t), we have v(0, t) = 0 = v
r2 We must
(,t). now
replace u(x, t) in Equation (7.57) with v(x, t). Using Equation (7.60) yields

u(x, t) = v(x, t) + r(x, t) = v(x, t) + e-t + 2 (t - e-t).

71

Thus,
au
at
=-+-=
av ar av - e -t + 2x 1 +
at at at ?C (
e
_t
)
and

a2u a2v a2r a2v

ax2 ax2 + ax2 ax2
a2r
since = 0.
ax2
We can now rewrite Equation (7.57), subject to the BCs in Equation (7.58), in
terms of v(x, t). We find that
av -e _ t 2x _t )_ax2
a2 v
at + r (1+e

or, written in the usual form

av _ a2v e_ tt 2x e_)
t
(1 + , (7.61)
at ax2 + 71
248 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

subject to

v(O,t) = 0

(7.62)
v ,t2 =0,
2x
where the source term Q(x, t) equals e - t -
r ( 1 + e - t). We now have a linear P DE
with linear and homogeneous BCs.
Step 1: Solve the related homogeneous problem, which is

av a2 v
at ax2
subject to

v(0, t) = o

r =0.
v 2,t
Solving the related homogeneous problem for eigenvalues and eigenfunctions yields

n = 4n2
,n=1,2,3,...
n (x) = sin 2nx J
Step 2: Assume the solution for v (x, t) is a Fourier series in the eigenfunctions
described earlier. We have

v(x, t) _ >bn(t)sin2nx. (7.s3)

n=1

Step 3: Expand Q(x, t) in a Fourier series using the eigenfunctions found in Step 1.
This yields
00
e
-t - 2x (1 + e-)
t
= an(t) sin 2nx.
71
n=1

Remember to use orthogonality when solving for an (t). Thus, we have

an (t) = -
o
2
(e_t - x (1 + e-t) sin 2nx dx

=-(e-t+ (-i)).
2
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 249

Therefore,
o0
e
-t - 2x (1 + e-t_t _ an (t) sin 2nx
?f
n=1

00

_ n_1 n7f (e_t + (-1)n) sin2nx. (7.64)

av a2v
Step 4: Using Equation (7.63), find and axe. Remember, this requires us to
at
satisfy all of the conditions necessary to do term-by-term differentiation of a Fourier
series. Assuming that all the conditions are met, we have
av °O
_ bn (t) sin 2nx (7.65)
at n=1

and
00
a2v
_ -4n2bn(t) sin 2nx. (7.66)
axe n=1

Step 5: Solve Equation (7.61) using Equations (7.64, 7.65, and 7.66). Thus,
av _ (92?)
+e
_t
- 2x (1+e _t )
at axe
becomes
00 00

b' (t) sin 2nx = -4n2bn (t) sin 2nx

n=1 n=1

00

+ (e_t -i- (-1)n) sin2nx. (7.67)

n_1 n7f

In Equation (7.67), all the Fourier series expansions are over the same interval [o,
2
and are expanded in the same eigenfunctions. Since Fourier series representation
of a function is unique, we may solve the ODE formed by the Fourier coefficients,
which is

b' t = -4n2b t+ 2 (e_t +

nor
(-i)).
n
(7.68)

The ODE in Equation (7.68) can be solved by methods reviewed in Appendix C.

The solution is
bn(t) _ [4n2e_t + (4n2 - 1)(-1)+ dne-4n2t1
.
n2(4n2 - 1)71
250 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

Therefore,

v(x, t) =
00 14ne_t + (4n2
n2(4n2 -1)71 - 1)(-1)n
+ dne -4n2t sin 2nx.
n= 1
Thus, the general solution for u(x, t) is
u(x,t) = r(x,t) + v(x,t)

=e_t + 2x
r (t - e
-t )

00 4n2e-t
1)(-1)n
+ + (4n2 - +dne -4n2t sin 2nx. (7.69)
n=1
n (4n 2 -
2

Equation (7.69) indicates the general solution for u(x, t). We now find the specific
solution for u(x, t) by applying the IC (Equation (7.59)), u(x, 0) = cos 3x. Thus,
u(x, 0) = cos 3x
_ 2x °° + (4n2 - 1)(-1)n
4n2e-t
-4n2t sin 2nx,
1- r + n=1 n2(4n2 _ 1)71 +dne
or
00
cos 3x - 1 +
n=1
n2(4n2 - 1)+ dnJ sin 2nx. (7.70)

Using orthogonality, the constants do in Equation (7.70) are

4n2 + (4n2 - 1)(-1)n
d12 n2(4n2 - 1)71

4 2

+ cos 3x - 1 + x sin 2nx dx

o ?f

_ 4n2 + (4n2 - 1) (-1) n 18

(7.71)
n2(4n2 - 1)it + n(4n2 - 9)7r
Therefore, the complete solution for u(x, t) is

u(x, t) = e-t + 2x (t - e -t )
r
00

+ 4n2e-t + (4n2 - 1)(-1)n +dne -4n2t stn 2nx,

2 2- n (4n 1)7r
n=1
where do is given in Equation (7.71).
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 251

Figure 7.3: The graph of the solution for u(x, t) for 0 < t < 2.

A graph of the solution is given in Figure (7.3) for 0 < t < 2, and the initial
temperature distribution is indicated by a bold-faced curve. Note how the BCs
affect the surface as time progresses.
In the previous two sections, we solved the problems of the nonhomogeneous
PDE with homogeneous BCs and the homogeneous PDE with nonhomogeneous
BCs. Both of these problem types are useful in solving physical models. More
importantly, they help us solve the more general problem of a nonhomogeneous PDE
with nonhomogeneous BCs. The nonhomogeneous PDE with nonhomogeneous BCs
applies to many more physical models. We address this general problem in the next
section.

EXERCISES 7.3

7.3.1. Consider the heat equation

au _ 52 u
axe
subject to
u(x,0) = f(x).
Solve for a time-dependent solution given the following nonhomogeneous bound-
ary conditions:
 252 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

8u(0, t) au(L, t)
(1) = a and
8x
=3.
8x
t)
(2) u(0, t) = a and =8.
_
(3)
8u(0, t)
ax - a(t) and 8u(L,
ax
t)
- ,Q(t)

(4) a O' t) = a(t) and u(L, t) _ ,6.

7.3.2. Consider the wave equation
a2u a2u
=c 2 ax2 ,

subject to

u(x, 0) = 1(x) and aunt' = g(x).

Solve for a time-dependent solution given the following nonhomogeneous bound-

ary conditions:
(1) u(0, t) = a and u(L, t) = ,3.
au(0, t)
(2) = a and u(L, t) = /.
ax
(3) u(0, t) = a(t) and u(L, t) = ,3(t).
t) _ /3(t).
(4) u 0 t= a and
au(L,

7.3.3. Liquid is input at time t = 0 by the amount of 3 drops/sec into the right end
of a tube of length 2. The left end of the tube is connected to a large tank
in which liquid pressure remains invariant. Assuming that, until the change
of input at the end x = 2, the pressure and input to the tube are constant,
find the change of input into the tube for t > 0 and the change in pressure in
the section x = 2 for t > 0. The boundary value problem that governs this
system is
ap _ aw
ax at - 2w
0<x<2,0<t,
ap aw
at - ax
subject to
p(O,t) = 0 and w(2, t) = 3 drops/second

and
w(x, 0) = 0

p(x,0) = 0.
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 253

(Hint: Eliminate p(x, t) in the equations previously listed and obtain a second-
order wave equation in w(x, t). If you are unsure how to eliminate p(x, t),
review Chapter 2.)
7.3.4. Consider the mathematical model
a2u a2u
=4
at2 ax2
subject to
u(0, t) = 6 and u(ir, t) = 1
and
au(x, 0)
u x 0 = 6 cos x ? x and =0.
Find the time-dependent solution for u(x, t), and using your favorite mathe-
matical software, plot your solution. By inspection, determine if the graph of
your solution is reasonable for the problem.
7.3.5. Consider the mathematical model
a2u 2 a2u
=C
ate axe
subject to
u(0, t) = coswt and u(L, t) = 0
and

au(x, 0)
u x 0= 0 and =0.
at
Find the time-dependent solution for u(x, t).
7.3.6. Consider the mathematical model
a2 u a2 u
= 16 axe
ate
subject to
u(0, t) = 0 and u(5, t) = sin 3t
and

u(x, 0) = 0 and aunt' =0.

Find the time-dependent solution for u(x, t), and using your favorite mathe-
matical software, plot your solution. By inspection, determine if the graph of
your solution is reasonable for the problem.
 254 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

7.3.7. Consider the heat equation

au a2u
at axe'

subject to
Du(L, t
u 0 t = 0 and = sin wt
(' ) ax
and

u(x, 0) = f (x).
Find the time-dependent solution for u(x, t).
7.3.8. Consider the heat equation
au 1.14a2u

at - axe '

subject to
au(0, t) = au(7, t)
sinwt and =0
ax ax
and

-x, 0<x<2
u(x, 0) = -.5x2 + 4.5x - 9, 2 < x <- 5

1, 5<x<7.
Find the time-dependent solution.
7.3.9. Consider the PDE
av
a2v
axe
-LC a2v
ate - (RC + GL)
at
- GRv = 0,

where v is voltage, L is self-inductance, R is resistance, G is leakage conduc-

tance, and C is capacitance. Find the voltage if
v(0, t) = E (a constant electromagnetic force)

8v(8, t)
ax = 0 (perfect insulation)
and

av (x, 0
(')
v x 0 = 0 and
at
=0.
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 255

7.3.10. Given the PDE

av
a2v
axe
-LC a2v
ate - (RC + GL)
at
- GRv = 0,

where v is voltage, L is self-inductance, R is resistance, G is leakage conduc-

tance, and C is capacitance. Find the voltage if
8v(0, t)
ax = 0 (perfect insulation) and v(ir , t) = 5 sinwot
and
av(x, 0)
v x 0 = 0 and =0.
at
7.3.11. Consider a rectangular plate made of nickel with length 2 yards and width
1.5 yards. Further, suppose the plate has perfect lateral insulation, and the
initial temperature distribution is given by
u(x, y, 0) = sin(xy).
At the boundaries x = 0 and y = 0 the plate is held in a 0°C bath, the
boundary at x = 2 is imperfectly insulated, which is known to vary not only
with time, t, but with position, y, and is estimated to follow the equation
ye-o.2t; and the boundary at y = 1.5 is perfectly insulated. Perform the
following:

(1) Mathematically model this problem. Note: The thermal diffusivity of

nickel may be found in Appendix E.
(2) Find a time-dependent solution of your mathematical model.
(3) Graph the solution of your mathematical model for several fixed times,
t.

7.3.12. Consider vibrations in a perfectly flexible rectangular membrane of length 2

m and width 3 m. Suppose the tension is 15 kgs, the density is 1 kg/m, the
initial position is given by
xy; 0<x<1 0<y<1.5
x(3-y); 0<x<1 1.5<y<3
u(x,y,0) = f(x,y) =
(2-x)y; 1<x<2 0<y<1.5
(2-x)(3-y); 1<x<2 1.5<y<3,
and the initial velocity is zero. Further, suppose that the edge x = 0 is
rigidly fixed at 0.5 m, the edges x = 2 and y = 3 are freely moving, and the
edge y = 0 oscillates to the function 0.5 cost. Determine the time-dependent
solution. Plot your solution for several different times t.
 256 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

7.3.13. Given a perfectly flexible rectangular membrane of length 4 ft and width 2

ft, if the tension is 12.5 lbs, the density is 2.5 lbs/ft (light rubber), the initial
displacement is f (x, y) = 0.1(4x - x2) (2y - y2) feet, the initial velocity is 0,
the edge x = 0 is fixed at 2, the edge x = 4 is fixed at zero, and the edges
y = 0 and y = 2 are free moving:
(1) Find the time-dependent solution.
(2) Using your favorite mathematical software, plot your solution. By in-
spection, determine if the graph of your solution is reasonable for the
problem.
(3) Use only the first term in the series for the vertical vibrations u(x, y, t)
to find the approximate solution at t = 30 secs, x = 1 ft, and y = 0.5 ft.
(4) Use the first two terms in the series for the vertical vibrations u(x, y, t)
to find the approximate solution at t = 30 secs, x = 1 ft, and y = 0.5 ft.
Compare this answer with the previous answer.
7.3.14. Consider
au(x, t) 1sa2a(x,t)

_ + sx

subject to
u(o, c) = o
BC: and IC: u(x, 0) = 9xex-4.
u(4, t) = 36

(1) Give a physical interpretation of this problem.

(2) For very large time find
(a) the solution,
(b) the heat energy generated per unit time inside the entire rod,
(c) the heat energy flowing out of the rod per unit time at each end,
and
(d) state the relationship between parts (b) and (c).
(3) After Part (2) is completed, the heat source is turned off. Perform the
following:
(a) state the equations that describe the mathematical model at this
time and
(b) using the equations that you just stated, solve the heat flow problem
for any time t.
7.3.15. A uniform string with mass density 3 lbs/ft and tension 300 lbs is fixed at the
left end at 0 and driven at the right end by an oscillator with amplitude and
angular frequency of 1. The string has length 200 and is initially at rest with
linear displacement from 0 to 1.
 Section 7.3: Homogeneous PDE with Nonhomogeneous BCs 257

(1) Model this problem mathematically.

(2) Solve the mathematical model.
(3) Write the solution with all known quantities substituted into it.
(4) Check that this solution satisfies the equation, the boundary conditions,
and the initial conditions.
(5) Using your favorite mathematical software, plot your solution. By in-
spection, determine if the graph of your solution is reasonable for the
problem.
7.3.16. Consider the PDE
52z D2z - 2 Dz
Dt2 = 9 Dx2 Dt'
subject to
8z(0, t)
=0 andz(10,t)=5
Dx
and
Du(x, 0)
u x 0 = 0 and =0 .
D

Find the time-dependent solution for z (x, t), and using your favorite mathe-
matical software, plot your solution. By inspection, determine if the graph of
your solution is reasonable for the problem.
This problem reflects the input of liquid into the right end of a tube of length
10 ft dropping at t = 0 by an amount of 5 gals/sec. The left end of the tube
is attached to a large tank in which the liquid pressure remains invariant. If
you assume that until the change of input into the right end, the pressure
and input were constant, we can find the change in input for t > 0 using the
equations listed previously, and the change in pressure in the section x = 10
for t > 0 by the following equation
Dp _ Dz
+ 2z .
Dx Dt

7.3.17. Consider the PDE

Du D2 u
Dt Dx2'
subject to
u(0, t) = 0 and u(5, t) = 2t
and
u(x, 0) = x2 + 2x -4.
Find the time-dependent solution for u(x, t).
258 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

7.4 NONHOMOGENEOUS PDE AND BCs

Suppose we have transverse vibrations in a perfectly flexible uniform one-dimensional
string with a time-dependent forcing function. Also, suppose that the boundary
x = 0 is fixed at some constant height and, at the boundary x = L, a time-dependent
force is being applied. Finally, to complete the description of the physical problem,
there is an initial displacement and velocity.
The mathematical model which describes this problem is

jC ax +
z
252u
Q(x,t)> (7.7z)

subject to BCs
u(0, t) = A
(7.73)
8u(L, t) = B(t)
8x
and ICs
u(x,0) = f(x)

(7.74)
8u(x,0)
at
Equations (7.72-7.74) contain parts of every problem we have discussed in this
chapter to this point. We have a nonhomogeneous PDE, and the nonhomogeneous
BCs contain both a nonzero constant and a function of time t. Therefore, to solve
this problem a solution strategy is mapped out in the following steps:

1. Replace u(x, t) with a function v(x, t) which has homogeneous BCs. This
8u(L , t)
requires constructing a function r(x, t) so that u(0, t) -r(0, t) = 0 =
ax
ar(L, t)
ax
2. Solve the related homogeneous PDE in v(x, t) for eigenvalues and eigenfunc-
tions.
3. Assume a Fourier series solution for v(x, t), where the coefficient in the Fourier
series is a function of time t.
4. Perform an eigenfunction expansion of the time-dependent function, Q(x, t),
using the eigenfunctions found in Step 2.
5. Determine the derivatives of the Fourier series solution for v (x, t).
6. Replace the PDE by its equivalent Fourier series representation and determine
the ODE, which must be solved.
 Section 7.4: Nonhomogeneous PDE and BCs 259

7. Solve the ODE.

8. Replace the coefficient in the Fourier series representation for v(x, t) with the
solution from the ODE.
9. Determine the general solution for u(x, t) from the solution for v(x, t).
10. Determine the constants in the general solution of u(x, t) by using the ICs
and orthogonality.

This is our plan. Now, we will carry it out.

Step 1: Replace u(x, t) with v(x, t) where v(x, t) has homogeneous boundary
conditions. This requires the construction of a function r(x, t) so that u(0, t) -
r(0 , t) = 0 =
au(L, t) - ar(L, t)
ax ax
Dr (L, t)
To construct a function r x t where r(0, t= A and = B(t), we simply
ar(x t)
integrate = B (t) with respect to x. This gives r(x, t) = xB (t) + c. Now,
ax
evaluating r(x, t) = xB(t) + c at x = 0 implies c = A. Therefore, the function we
want is r(x, t) = xB(t) + A. Remember that
v(x, t) = u(x, t) - r(x, t), (7.75)

and replacing r(x, t) with xB(t) + A in Equation (7.75) we generate

v(x, t) = u(x, t) - r(x, t) = u(x, t) - xB(t) - A. (7.76)

This means the BCs

v(0, t) = o
(7.77)
8v(L, t) _ 0
8x
are homogeneous. Rewriting Equation (7.76), we get

u(x, t) = v(x, t) + xB(t) + A. (7.78)

Using Equation (7.78), we can replace u (x, t) in Equation (7.72) with v (x, t). This
a2u a2u
requires the determination of ate and axe , since

a2u a2v
xB (t) (7.7s)
ate =ate +
and
a2u a2v
axe ' (7.80)
Dx2
260 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

2 2 2

ate
=c + Q(x, t)
becomes

a2v xBii`t) a2v

= c2axe
ate + + Q(x, t)
or

a2v a2v
c2 axe +Q(x,t) - xB ii(t), (7.81)
ate =
subject to BCs, Equation (7.77).
Step 2: Solve the related homogeneous problem in terms of v(x, t) for eigenvalues
and eigenfunctions.
The related homogeneous problem is

a2v a2v
5=C 2

axe
subject to BCs, Equation (7.77). After separating variables and applying the BCs
to the spatial problem, we conclude that

r (2n - 1J
)=
n I
ZL
n=1,2,3 ,.... (7.82)
r (2n-1)irxl
cPn,(x)= bn, Sln
2L

Step 3: Assume a Fourier series solution for v(x, t), where the coefficient in the
Fourier series is a function of time t.
We have

[(2n - 1)irxl
v(x, t) _ b(t) sin
n =1
2L

_ bn(t) sin (7.83)

n=1

Step 4: Perform an eigenfunction expansion of the time-dependent function,

Q(x, t), using the eigenfunctions found in Step 2.
We must remember the time-dependent forcing function in Equation (7.72) was
changed to the time-dependent forcing function Q (x, t) - xB" (t) in Equation (7.81).
Therefore, we have

Q(x, t) - xB"(t) = do sin (x),

n=1
 Section 7.4: Nonhomogeneous PDE and BCs 261

where, after integration, the constant do becomes a function of time t. Thus, as-
suming we can perform the required integration, we have

Q(x, t) - xB"(t) _ dn(t) sin (x). (7.84)

n=1

Step 5: Determine the derivatives of the Fourier series solution for v (x, t).
Remember, we must satisfy all the conditions that allow term-by-term differ-
entiation of a Fourier sine series, which is covered in Chapter 5. Assuming these
conditions are satisfied, we have
00
av(x,t)
ax
_ Ld
n=1
nbn(t)COS 1 nx) .

The Fourier cosine series of a continuous function is continuous; thus, term-by-term

differentiation is valid, and
00
82v(x, t)
ax2
_- Anbn(t) sin (\/x). (7.85)
n=1

Since we always expect time to be continuous, and therefore any function of

time to be continuous in both the first and second derivatives, we see that

a2v(x, t) °O

at2
= L1 Ul) Sill V AnJJ.
n=1

Step 6: Replace the PDE by its equivalent Fourier series representation and
determine the ODE that must be solved.
Using Equations (7.84, 7.85, and 7.86), Equation (7.81)

a2v a2v
c2 + Q(x, t) - xBii(t)
at2 = ax2
becomes
00 00

b( t) sin (/x) = -c2 Ab(t) sin (/x)

n=1 n=1
00

+ do (t) sin (x). (7.87)

n=1

Equation (7.87) equates one Fourier series to the sum of two Fourier series
all expanded in the same eigenfunctions over the same interval. This means the
constants must be equal. Extracting coefficients yields

b(t) _ -c2Anbn(t) + dn(t). (7.88)

262 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

This is a second-order nonhomogeneous ODE.

Step 7: Solve the ODE. We solve a second-order nonhomogeneous ODE by solv-
ing the related homogeneous part, solving the particular part, and then summing
the two parts together. In general, unless we know the form of the functions bn (t)
and do (t) in Equation (7.88), we may not be able to determine a solution. However,
if we assume that a function of time, bn (t), is positive for all time t, and we know
c and A are positive, then the solution to the homogeneous part is

bnh (t) = crn cos [c/t] + ,3n sin [c'5t].

We will assume the particular part has a solution and call it bnP (t). Therefore, the
solution to Equation (7.88) is

bn(t) = bnh (t) + bnP (t) = cYn cos [c/t] + ,3n sin [c'5t] + bnP (t). (7.89)

Step 8: Replace the coefficient in the Fourier series representation for v(x, t)
with the solution from the ODE.
From Equation (7.83), we have
00

v(x, t) _ bn(t) sin (/cx) (7.90)

n=1

tl
[an cos [C ant] + ,Qn Sin [ct] + bnp (t)] sin (x)
Step 9: Determine the solution for u(x, t) from the solution for v(x, t).
Equation (7.78), which is

u(x, t) = v(x, t) + xB(t) + A = xB(t) + A + v(x, t),

yields the general form of u(x, t). Replacing v(x, t) in Equation (7.78) then gives
the solution for u(x, t). Thus, we have

u(x,t) = xB(t)+ A (7.91)

+
n=1
Lan cos[ct] + n sin [ct] + bnp (t)J sin (x)
Step 10: Use the ICs and orthogonality to determine the constants in the general
solution of u(x, t).
Equations (7.74) tell us

u(x,0) = f(x)

8u(x,0) _ g(x)
at
 Section 7.4: Nonhomogeneous PDE and BCs 263

Thus, we find that the first IC implies

00

u(x, 0) = 1(x) = xB(0) + A + [an + bnp (0)] sin (x)

n=1
or

f (x) - (xB(O) + A) _ [a + (0)] sin (x).

By using orthogonality, the constant within the summation becomes
2
fL
an +bnp(0) = [ f (x) - (xB(0) + A)] sin ( x) dx,

which means that

L
an = L [1(x) - (xB(0) + A)] sin (x) dx - b(0). (7.92)
0

The second IC implies that

t
8u(x, 0)
= g(x) = xB' (0) +
n=1
°°
[/3n + b'ap (0)] sin (x).
Therefore, in a similar method as described earlier, we have
fL
/3n = L [g( x) - xB'(0)] sin ( x) dx - b(0). (7.93)

We now have the complete solution for u (x, t) from Equation (7.72) subject to
BCs, Equation (7.73), and ICs, Equation (7.74). It is
u(x, t) = xB(t) + A

+
n=1
Lan cos [ct] + /3n sin [ct] + bnp (t)] sin (x),
where an and /3n are given in Equation (7.92) and Equation (7.93), respectively,
and An is given in Equation ( 7.82) .
EXAMPLE 7.4. Consider
au a2u
- xt, (7.94)
at = axe
subject to
au(0, t) _ 0
8x
(7.95)
8u(7r, t)
8x
-t
264 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

and

11x .
u x 0) = sin (7.96)
) 2

Solve for the time-dependent solution using the ten steps developed in this section
(Nonhomogeneous PDE with Nonhomogeneous BCs).
Step 1: Replace u(x, t) with a function v(x, t).
x2t
Let v(x, t) = u(x, t) - r(x, t) where r(x, t) = which implies u(x, t) = v(x, t) +
r(x, t). Therefore,

au av x2
+ (7.s7)
at at 271

and

a2 u a2v t
ax2 ax2 + 71.

Replacing u(x, t) with v(x, t) in Equation (7.94) yields

av x2 a2v t
at + ax2 + - xt
or

- x t--
av a2v 1 x2
(7.99)
at ax2 -
which has the following BCs
av(0, t) _ o
8x
av(71,t)
-0
ax

as corresponding homogeneous BCs. We have completed Step 1, and we can move

on to Step 2.
Step 2: Solve the related homogeneous PDE in v (x, t) for eigenvalues and eigen-
functions only.
The related homogeneous PDE in v(x, t) is

av a2 v
(7.101)
at = ax2
subject to BCs, Equations (7.100). We have solved for the eigenvalues and eigen-
functions several times in the past chapters. The eigenvalues are

n = n2, n = 0, 1, 2, 3, ... (7.102)

 Section 7.4: Nonhomogeneous PDE and BCs 265

The related eigenfunctions are

(7.103)
an
ancoscosnxnx , n> 0.
Step 3: Assume a Fourier series solution for v(x, t) where the Fourier coefficients
are functions of the variable t.
The Fourier series solution is

v(x, t) = ao(t) -{- an(t) cos nx. (7.104)

n=1

Remember, the Fourier coefficient must be a function of time, t. We now move on

to Step 4.
Step 4: Do an eigenfunction expansion of the time-dependent forcing function
Q(x, t). Using the eigenfunctions stated in Equation (7.103), we have

1 x2
C -x) t- 2 = bo(t) + bn(t)cosnx,
n=1

where

o \ /
1 7C 71

and

x2
b(t)=-r2 o
-?r1 - x t - 271
cos nx dx

[(1 - (-i)) t - (-i)].

Thus,

x2 _ 1 7C 71

t 271 7r 2 t- 6
+

00
2
-i [(1 - (-1)") t - (-1)] cosrtx. (7.105)
n=1
266 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

Step 5: Determine the derivatives of the Fourier series solution for v (x, t).
Using Equation (7.104) and assuming the required conditions hold, we have
00
8v(x, t)
at - a o (t) + an (t) cos nx (7.106)
n=1
and
00
a2v(x,t)
= -n a (t) cos nx. (7.107)
8x2 n=1

Step 6: Replacing
av a2v 1
_x t_ x2
at = axe +
with its equivalent Fourier series representation yields

ao (t) + j an (t) cos nx = -n2an (t) cos nx + (7.108)

n=1 n=1
1 r r
°° 2
n2 [(1 - (-1)n) t - (-i)] cos nx.

From Chapter 4 we know that two Fourier series that are equal must have equal
coefficients. Therefore, Equation (7.108) can be reduced to the ODEs

ao(t)=
r ---
2
t--6 1 7r 71
(7.109)

and

a n ( t) _ -n 2 an( t) + rn2 [( 1 - ( -1)n) t - ( -1)n]

or

an t ) + n 2 an( t) n2 [( 1 - (- i)) t - ( -i ) ] . (7 . 110)

Step 7: Solve the ODE.

The ODE in Equation (7.109) is a first-order ODE, and it has the solution
_ 1 71 t2 71t
(7.111)

The ODE in Equation (7.110) is a first-order ODE. We arrive at the solution by

using an integrating factor. Thus, we have
[e_n2tan(t)]/ 2 e-n2t.
= urn 2
[(1 - (-1)n) t - (-1)n] (7.112)
 Section 7.4: Nonhomogeneous PDE and BCs 267

Solving for an (t) yields

an(t)
2
71n4 (1 (-1 )fl) Ct - 2
n
I - (_i)fl] +dne-n"zt (7.113)

Step 8: Replace the coefficients in the Fourier series representation for v (x, t).
Replacing ao (t) and an (t), given in Equations (7.111 and 7.113), in Equation (7.104)
yields
2
v(xt)=(--)---+do+ 7rt
(7.114)

[(1 - (-i)) (t (_1)n + dne -n 2t cos nx.

n-1 - -
Step 9: Determine the general solution for u(x, t).
Remembering u(x, t) = v(x, t) + r(x, t) = r(x, t) + v(x, t) implies
_ x2 t 1 7r t2 art
(7.115)
u(x' t) 2 2

00

{ n 2
4
[(1 - (-1)n) (t n (_1)n + dne2
-n t cos nx.

Step 10: Determine the constants in the general solution of u(x, t) by using the ICs
and orthogonality.
Equation (7.96) indicates the initial temperature distribution in the one-dimensional
rod is u(x, 0) = 0. Therefore,

2 = d0+
u(x,0) =sin 5x

:{; [_(1_(_1y) - (_i)n] + cosnx,

J

which implies

d0=-j
1 sin
11x dx_2
r o 2 1171

and

dn = 2 11 x
cos nx dx +
2 1 (i)) + (_i)fl]
71j0
sin
2 7rn4 n
44+ n4 r
1) )
n 1
+ (-1)nJ . (7.116)
r (4n2-121)
268 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

Figure 7.4: The graph of u(x, t) for 0 < t < 2.

Thus, our complete solution for u(x, t) is

t)_ x2 t 1 7r t2 art 2
u(x, (7.117)
271 + 7r 2 2 6+

2
urn
1- (-i)) n t- n - (_i)fl]+ d e-net
1 cos nx
n=1

where do is given in Equation (7.116). Figure (7.4) is a graph of the solution

presented in Equation (7.117) 0 < t < 2. The bold-faced curve indicates the initial
temperature distribution.
For a final example, we solve a multidimensional problem.
EXAMPLE 7.5. Consider
au(x, y, t) a2u(x, y, t)
+ a2u(x,
a2
y, t) + sin t,
at axe
subject to the BCs
y, t)
u0 t =0and a
= 1 watt (7.119)

au(x' O t)
= 0 and u(x, 27, t) = x°C, (7.120)
ay
 Section 7.4: Nonhomogeneous PDE and BCs 269

and IC

u(x, y, 0) = 0°C (7.121)

A possible physical model is temperature distribution in a perfect laterally insu-

lated rectangle of length 27r and width r. The initial temperature is 0°C. At the
start of the experiment the boundary y = 0 is perfectly insulated, the boundary
x = 7r is imperfectly insulated, the boundary at x = 0 is held in a 0°C bath, and
the boundary at y = 27r has a temperature that varies as x does from 0 to 7r. Also,
there is a source/sink term depending on time, t.

Solution: First, we must construct a function r(x, y, t) so that r is a continuous

function and it has the same BCs as u(x, y, t). One possible function is

r(x,y,t) = x. (7.122)

ar(x, y, t) _ 1 ar(x, y, t)
Since Equation (7.122) yields r (0 ,y,t) = 0, = 0 and
ax ay
r(x, 27r, t) = x. Thus, v(x, y, t) becomes

v(x, y, t) = u(x, y, t) - r(x, y, t)

and

av(x, y, t) au(x, y, t) ar(x, y, t) au(x, y, t)

(7.123)
at = at - at = at

a2v(x, y, t) a2u(x, y, t) a2r(x, y, t) a2u(x, y, t)

(7.124)
ax2 ax2 ax2 ax2
and

a2v(x, y, t) a2u(x, y, t) a2r(x, y, t) a2u(x, y, t

(7.125)
aye aye aye aye

Rewriting Equation (7.118) in terms of Equations (7.123, 7.124, and 7.125) yields

av (x, y, t) 52v (x, y, t) 52v (x, y, t) + sin t.

(7.126)
at = axe + ay2

subject to the BCs

y, t)
v 0
( ' y')t= 0 and ax
=0 (7.127)

and

av(x, 0, t)
= 0 and v(x, 27r, t) = 0. (7.128)
ay
270 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

Now, we solve the homogeneous problem.

av(x, y, t) a2v(x, y, t) a2v(x, y, t)
+ (7 129)
.

at = ax2 aye

subject to the BCs Equations (7.127 and 7.128).

Applying separation of variables yields the spatial equations
X"(x) = -rX(x) (7.130)

and

Y"(y) = -eY(y) (7.131)

where = A - T, which implies A = + T, subject to the BCs
X(0) = 0 and X' (7r) = 0 (7.132)

and

Y'(0) = 0 and Y(27r) = 0, (7.133)

respectively. We know that the solution to Equations (7.130 and 7.131) subject to
Equations (7.132 and 7.133) are
[(2n - 1)x [(2m_ 1)y
X(x) = sin and Y(y) = cos (7.134)

where
r (2n-1
T =
L 2

and

r(2m-1)12
L 2 ,m1, 2,3,.
j
Therefore,
12
2 2 2 (2n-1) 2 (2m-1
n, m = 1, 2,3, .. .
2 + 2

Thus, we write the solution for v(x, y, t), which is

00 00
Lr(2m4 (2n 2 1)x
v(x, y, t) = anm(t)cos 1)yJ sin f (7.135)
n=1 m=1 L

Second, we must expand the new source term of Equation (7.126) in terms of the
eigenvalues and eigenfunctions found for the homogeneous problem. We have
°° °° 1(2m - 1)y [(2n - 1)x
sint = bnm(t) cos sin
4 2
n= 1 m= 1
 Section 7.4: Nonhomogeneous PDE and BCs 271

where
2
2 2n 1)xl (2m
bn-m, (t = 2
si n
Slri t Sl
sin 2 ]cosL 4 1)y]dxdY
n
o n
o

16(_1)m+1 sint
ir2(2m - 1)(2n - 1

= bnm Sin t,

where bnm - r2(2m16(-1)m+1

- 1)(2n - 1)
av(x, y, t) a2v(x, y, t) a2v(x, , t)
Third, we must find axe , and , and they are
at a2
00 00
av(x, y, t) r (2m - 1)yl f (2n - 1)x l
a'n,,t(t) cos sin
at n=1 m=1 L 4 J l 2 J

00 00
a2v(x,y,t) 2n - 1) 1`2 (2m-1)yl r(2n- 1)x1
ax2 L -art(t)
- n=1 m=1i
L 2 J
cos L 4
]sinL 2

and

y
y,t)
z
. 00

I
n=1 m=1
00

I -anm(t) I
1(2m_1)
L J
2 cos I (2m 4 1)yJ sin
L
(2n 2 1)x
L

Next, we put it all the pieces together and solve the resulting ODE. We have
00 00

a(t) cos f (2m - 1)yl sin (2n - 1)x1

n=1 m=1 L
4 J L
2 J

00 00

n=1 m=1
-anen(t)
1(2n - 1)12
L 2 J
cos
r (2m - 1)yl
L 4 J
sin
L2
[(2n - 1)x1

°°
+ :i:°° -Clnm, t r(2m - 1)12 1(2m_l)yl (2n-1)x1
() I 4
j cos 4 J sin 2
n=1 m=1 J

+ bnm Sin t,
which yields the ODE

1)12
anm(t) _ -Cbnm(t) [r(2n2 1)J2 -Clnm(t) (2 + bnm Sin t (7.136)
L J

Equation (7.136) may be written as

(tP2 + bnm Sin t.
anm (t) = -anm
272 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

solving the first-order ODE yields

A2 sin t - cost a2t

anm (t) = bnm A2 1 + Cnme_

Therefore,

(2m [(2n 2 1)x

v(x, y, t) _ anm (t) cos 4 1)y] sin
n=1 m=1 L L

becomes

a2 sin t - cost
v(x,y,t) bnm
n=1 m=1 [( { A2+1

+ Cnme_ \2t
cos f (2m-1)yl s in f (2n-1)x 11
L 4 J L 2 JJ

We now may state the general solution for u(x, y, t), which is

u(x, y, t) = v(x, y, t) + r(x, y, t) = r(x, y, t) + v(x, y, t) = x + v(x, y, t).

Applying the IC u(x, y, 0) = 0, we have

u(x, y, 0) = 0 = x + v(x,y,0)

x+ r({-b+1}
nm
n=1 m=1 L

(2m4
+ Cnm)cos 1)yJ sin L(2n
L 2 1)x1J

or

00 00
(2m4 Lr(2n_
+Cn,m,l cos I 1)yJ sin 2 1)xJ
n=1 m=1 / L

which means

2
f2 f (-x) (2m (2n
Cnm = cos 4 1)yJ sin 2 1)x dxdy + + 1.
J
I

Jo o L L J

Thus,

32(-1)m+1(_1)n bnm
Cnm = (7.137)
(2m-1)(2n-1)22 + A2+1
 Section 7.4: Nonhomogeneous PDE and BCs 273

Therefore, the solution for u(x, y, t) is

00 00
A2 sin t - cost
u(x, y, t) =x+ [bnm{
A2+1
n=1 m=1

CnmC-a2tCOS
(2m4 Lr(2n_
+ I 1)yJ sin
L 2 1)xJJ
where Cnm is given by Equation (7.137).
In the next section, we summarize the solution technique, called the eigenfunc-
tion expansion technique for solving a nonhomogeneous PDE problem by separation
of variables.

EXERCISES 7.4

7.4.1. Consider the heat equation

au a2 u
a
t = k a x2 + Q(x, t),

subject to
u(x,O) = f(x).
Solve for a time-dependent solution given the following nonhomogeneous bound-
ary conditions:
au (0' au (L'
(1) t) = A and t) -B
au (L, t)
(2) u 0 t = A and =B.
ax
(3) u(0, t) = A(t) and u(L, t) = B(t).
au(L, t)
(4) u 0 t= A and = B (t).
a
7.4.2. Consider the wave equation
a221 207221
Q(x, t),
ate - 07x2 +
subject to

u(x, 0) = f (x) and aunt' = g(x).

Solve for a time-dependent solution given the following nonhomogeneous bound-

ary conditions:
 274 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

(1) u(0, t) = A and u(L, t) = B.

(2) a o't) =A and u(L, t) = B.
(3) u(0, t) = A(t) and u(L, t) = B(t).
(4) u(0, t) = A(t) and B(t).

7.4.3. Consider the following information:

(a) heat conduction in a one-dimensional rod of length 2 m

(b) perfect thermal insulation on the lateral sides,

(c) initial condition of x + cos x for 0 < x < and of x - sin x for
4
3ir 3ir
4
<x<
(d) left boundary is time-dependent and is cost,
3ir+2
(e) right end is held at a constant temperature of 2 degrees,
(f) a uniform rod made of pure iron, and
(g) no source.

(1) Set up the mathematical model. Note: The thermal diffusivity of pure
iron may be found in Appendix E.
(2) Give a series solution of your mathematical model.
(3) Using your favorite mathematical software, plot your solution. By in-
spection, determine if the graph of your solution is reasonable for the
problem.
(4) Check that this solution satisfies the equation, the BCs, and the ICs.

7.4.4. Consider the following information:

(a) a perfectly flexible string of length 2ir ft,
(b) tension of 50 lbs,
(c) mass density of 0.02 lbs/ft,
(d) vibrating motion which is entirely vertical,
(e) fixed left end at 1,
(f) fixed right end at 1,
(g) initial displacement of x+1 for 0 <x < it and of 2ir-x+1 for it < x < 27r,
(h) initial velocity of 0, and
(i) with gravitational effects of 32 ft/sec/sec at the start of the experiment
that decreases exponentially as time increases.
 Section 7.4: Nonhomogeneous PDE and BCs 275

(1) Set up the mathematical model.

(2) Solve your mathematical model for any time t.
(3) Using your favorite mathematical software, plot your solution. By in-
spection, determine if the graph of your solution is reasonable for the
problem.
(4) Check that this solution satisfies the equation, the BCs, and ICs.
7.4.5. Consider a uniform one-dimensional rod, without a source, which is not later-
ally insulated. (Heat can flow in and out across the lateral surface.) By exper-
imentation, you have discovered that heat is flowing across the lateral bound-
ary at a rate proportional to the difference between the temperature u(x, t)
and the surrounding medium that is kept at lOx degrees where 0 < x < 2ir ft.
(1) Given that the convection constant of proportionality is .071 and the end
at x = 0 is fixed at one degree, the end x = 2ir is fixed at e2 degrees,
and the rod is composed of zinc, set up the mathematical model. Note:
The thermal diffusivity of zinc may be found in Appendix E.
(2) Suppose the initial temperature of the rod is ex. Give a series solution
of your mathematical model.
(3) Using your favorite mathematical software, plot your solution. By in-
spection, determine if the graph of your solution is reasonable for the
problem.
(4) Check whether this solution satisfies the equation, the BCs, and ICs.
7.4.6. Consider a vibrating string with friction and a time-periodic forcing function
a2 u 2 a2 u au
c 3+ x cos wt,
ate axe - at
subject to
u(0, t) = 0 and u(ir, t) _ r

and

u(x, 0) = x and aunt' =0.

(1) Solve the initial value problem if ,3 is small (0 <3 < 2c).
(2) Suppose that the length of the string was arbitrary, that is, L. What
would the bounds be on 3 in this case?
(3) Suppose 3> 2c. What would happen in this case?
7.4.7. Solve the initial-value problem
au(x, y, t) a2u(x, y, t) a2u(x, y, t)
at - axe + a2
+ t,
 276 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

subject to the BCs

au (ir'
u0 t= and y' t) =0,
and
au(x, 0, t)
= 1 and u(x, 2ir, t) = 0,
ay

and IC

u(x, y, 0) = 0.

7.4.8. Solve the wave equation in a rectangular sheet with Dirichlet BCs and a time-
periodic forcing function, cos t. Assume that the sheet is initially at rest with
displacement

xy; 0<x<1 0<y<1.5

x(3-y); 0<x<1 1.5<y<3
u(x,y,0) = f(x,y) =
(2-x)y; 1<x<2 0<y<1.5
(2-x)(3-y); 1<x<2 1.5<y<3.
7.4.9. Solve the wave equation in a rectangular sheet with Neumann BCs and a
time-periodic forcing function, sin t. Assume that the sheet is initially at rest
with displacement

xy; 0<x<1 0<y<1.5

x(3-y); 0<x<1 1.5<y<3
u(x,y,0) = f(x,y) =
(2-x)y; 1<x<2 0<y<1.5
(2-x)(3-y); 1<x<2 1.5<y<3.
7.4.10. Consider heat conduction on a 3 ft one-dimensional uniform oak rod with no
lateral or boundary insulation, hence the lateral and boundary surfaces are
in direct contact with a time/position-dependent medium modeled as cos(xt),
where x is in inches and t is in seconds. The initial temperature distribution
of the rod is given as 1.
(1) State the mathematical model. Note: The thermal diffusivity of oak may
be found in Appendix E.
(2) Solve the mathematical model.
 Section 7.5: Summary 277

(3) Using your favorite mathematical software, plot your solution. By in-
spection, determine if the graph of your solution is reasonable for the
problem.
(4) Check whether the solution satisfies the equation, the boundary condi-
tions, and the initial conditions.
7.4.11. The space shuttle heat shields are shaped as rectangular boxes with curved
surfaces matching the curvature of the body of the shuttle. The greater ma-
jority of the heat shields are surrounded by other heat shields that are fitted
tightly together. One surface is against the body of the shuttle, while the
opposite side faces space.
The typical heat shield may be modeled adequately by a rectangular box with
no curvature. The model is assumed to have perfect insulation on its four
sides, while the side against the shuttle is assumed to be held at 14°C. The
temperature of the side open to space at time of re-entry is proportional to
the time in seconds, required for re-entry. On re-entry into earth's atmosphere
this function is modeled by

f(t)= h+1 (e()_i).

Here h is calculated using distance in meters from the front of, and from the
center of, the bottom of the shuttle. Note: h decreases to 0 as you approach
the front of the shuttle.
For this problem, we will use the thermal diffusivity of asbestos, which may
be found in Appendix E. The heat shields are sized by height = z = 7.5 cm,
width = y = 15 cm, and length = x = 50 cm. Initial temperature distribution
is
z
u(x, y, z, 0) _ -e(1) + 15.
Note: The side facing space is oriented in the positive z direction.
(1) Construct the mathematical model.
(2) Solve the mathematical model.
(3) Write the solution in a completed form.
(4) Suppose h = .019. Determine the temperature distribution on this heat
shield for a 3.5 min re-entry.

7.5 SUMMARY
In this chapter, the problems solved by the eigenfunction expansion technique were
all linear PDEs with linear BCs. It is important to realize that this method only
works on linear PDEs with linear BCs.
We summarize our eigenfunction expansion technique as follows:
278 Chapter 7: Separation of Variables: The Nonhomogeneous Problem

1. Determine that the PDE and BCs are linear.

2. If BCs are not homogeneous, construct a function that, when replacing the
function in the PDE, will have homogeneous BCs.
3. Determine the related homogeneous PDE and BCs and solve for the eigenval-
ues and eigenfunctions.

(a) Assume the solution is a product of two functions. One function is pos-
sibly of time, while the other function is composed of spatial variables.
(b) Substitute this product into the PDE and move all functions of time to
one side while placing all other components on the opposite side of the
equation.
(c) Set both sides equal to the same separation constant, breaking the orig-
inal equation into two new equations, thus separating the variables.
(d) Repeat steps (a), (b), and (c) on equations with more than one spatial
variable until you have an ODE for each spatial variable.
(e) Substitute the product form of the solution into the BCs to obtain BCs
for the ODEs.
(f) Solve the ODE boundary value problem(s) to obtain the eigenvalues and
eigenfunctions.

4. Assume the solution is a Fourier series expanded in the related homogeneous

eigenfunctions with time-dependent coefficients.
5. Determine a Fourier series expansion in the eigenfunctions found in Step 2 for
any nonhomogeneous terms in the PDE.
6. Differentiate, term-by-term, the Fourier series from Step 4. Remember, all
conditions necessary to do term-by-term differentiating must be satisfied.
7. Replace the PDE with its Fourier series equivalent.
8. Equate the coefficients to obtain and solve an ODE for the constant coeffi-
cients.

9. If a constructed function was used, obtain the solution for the original function
of the original PDE.
10. Use the ICs and orthogonality to determine the constants for the specific
solution.
Chapter 8

The Sturm-Liouville
Eigenvalue Problem
8.1 INTRODUCTION
The Sturm-Liouville eigenvalue problem is a second-order homogeneous ODE with
variable coefficients subject to homogeneous boundary conditions. You should re-
member second-order ODEs produce eigenvalues and eigenfunctions for the solution
space where the ODE is defined. The study of this class of ODEs is important in
itself. However, applying this class of ODEs to PDEs is our primary interest.
Consider the following examples:
EXAMPLE 8.1. Given the heat equation
0721 C72 2l

at ax 2
subject to
u(o, t) = o

u(L, t) = 0
and

u(x, 0) = f (x),
we know the separation of variables technique applies. Letting u(x, t) = p(x)G(t),
taking the appropriate derivatives and separating the PDE yields the time equation
G'(t) = -AkG(t)
and the spatial eigenvalue problem

p"(x) = -Ap(x),

279
280 Chapter 8: The Sturm-Liouville Eigenvalue Problem

subject to
(p(0) = 0

p(L) =0.
EXAMPLE 8.2. Given the wave equation
a2u
=c
29u
ate axe
subject to
8u(0, t) _ 0
8x
8u(L,t) _ o
ax
and

u(x, 0) = f(x)

8u(x, 0)
at = 9(x)
we again apply the separation of variables technique resulting in the time equation

G" (t) _ -Ac2G(t),

and the spatial equation

cp"(x) = -A,o(x),

subject to

We are not concerned with the solution for these examples. What we are con-
cerned with is the spatial problems, which resulted from using the separation of
variables technique. In particular, we notice the spatial equations, Equation (8.1)
and Equation (8.3), are identical. However, the BCs, Equation (8.2) and Equation
(8.4), are different.
Suppose we rewrote the BCs in the following way:

aip(0)+a2p'(O)=O

a3,o(L) + =0.
 Section 8.2: Definition of the Sturm-Liouville Eigenvalue Problem 281

Then, we need a2 and a4 to equal 0, and a1 and a3 to equal 1 for Equation (8.2)
to be true; for Equation (8.4) to be true, we need a1 and a3 to equal 0, and a2 and
a4 to equal 1. Thus, Equation (8.5) is really a generalized expression of the BCs in
Equations (8.2 and 8.4). Therefore, we may restate the spatial eigenvalue problem
given by Equation (8.1), subject to the BCs, Equations (8.2 and 8.4), as the general
spatial problem

subject to
ai(0) + a2(p'(O) = 0

a3co(L) + a4cp'(L) =0.

This chapter develops known facts about the generalized spatial problem. In
the next section, we develop the complete formulation of the generalized spatial
problem, define its characteristics in theorems, and prove several of those theorems.
The section titled, "Rayleigh Quotient," proves another theorem and discusses some
uses of the Rayleigh quotient. Next, we solve a general heat flow/wave motion
problem. We conclude this chapter by considering PDE problems involving Robin's
boundary conditions.

8.2 DEFINITION OF THE STURM-LIOUVILLE

EIGENVALUE PROBLEM
The generalized spatial problem is important for two major reasons. First, we can
study an entire class of ODEs and develop a solution method for that class. Second,
it allows us to go beyond the problems of heat in a one-dimensional uniform rod, or
wave motion in a one-dimensional uniform horizontal string. For instance, consider
the following example.
EXAMPLE 8.3. Suppose we have heat flow in a nonuniform one-dimensional rod
of length L with a source term dependent on the temperature distribution multiplied
by a proportionality spatial function, and subject to homogeneous Robin's boundary
conditions on both ends, and an appropriate initial condition. The mathematical
model for these conditions is

c(x)n(x)
at - ax 1Ko(x)1
ax, + q(x)u(x, t), (8.8)

subject to
Ou(O, t)
-K0(0) = -hiu(0,t)
(8.9)
KO(L)OLt) = -h2u(L,t)
282 Chapter 8: The Sturm-Liouville Eigenvalue Problem

and
u(x, 0) = f (x).
Since the PDE and BCs are homogeneous, the separation of variables technique ap-
plies. Letting u(x, t) =cp(x)G(t), taking the appropriate derivatives, and rewriting
Equation (8.8) yields

c(x)p(x)cp(x)G'(t) = dx [Ko(x)cp'(x)] G(t) + q(x)cp(x)G(t). (8.10)

Equation (8.10) separates into a time equation,

G'(t) = -AG(t),
and a spatial equation,

dx
The spatial equation may be written as

=0. (8.11)
dx
Applying u(x, t) = (x)G(t) to the BCs yields
-Ko(0)p'(O) _

-h2(L)
after separation. The BCs also may be rewritten as
h1 cP(0) - K0(0)'(0) = 0
(8.12)
h2(L) + Ko (L)co'(L) =0.
Thus, we have the spatial equation, Equation (8.11), subject to the BCs, Equation
(8.12).
At this point, we usually solve the spatial problem for eigenvalues and eigen-
functions. However, the spatial problem described by Equations (8.11 and 8.12) is
not our usual spatial problem. Therefore, we must determine if a solution exists,
and if we can find it.
Equations (8.11 and 8.12) may look somewhat messy, but, by making the fol-
lowing substitutions, Equations (8.11 and 8.12) become more compact. First, let
K0(x) = p(x). Second, let c(x)p(x) = a(x). Finally, since h1, K0(0), h2, and
K0 (L) are all constants, we let them equal a1, a2, a3, and a4, respectively. These
substitutions yield the spatial problem

(x)'(x)] + q(x)(x) + Aa(x)(x) = 0, 0 x L, (8.13)

 Section 8.2: Definition of the Sturm-Liouville Eigenvalue Problem 283

subject to
al(0)+a2'(0) =0
(8.14)
a3(L) + a4cp'(L) =0.
If p(x), q(x), and a(x) are real continuous functions on the interval 0 < x < L;
p(x) and a(x) are always greater than 0 on the interval 0 < x < L; and ai, i = 1
to 4 are real numbers, then Equations (8.13 and 8.14) are known as the regular
Sturm-Liouville eigenvalue problem on the interval 0 < x < L. A more
generalized version of the regular Sturm-Liouville eigenvalue problem is stated on
the interval a < x < b. However, for our purposes we generally use the interval
0<x<L.
The fact that Equations (8.13 and 8.14) are regular Sturm-Liouville eigenvalue
problems doesn't mean we have a solution. It means we know the class that ODE
Equations (8.13 and 8.14) belong to, and there are four very important theorems
which apply to any Regular Sturm-Liouville Eigenvalue Problem. These theorems
are as follows :
Theorem 47. There exists an infinite number of eigenvalues, An, n = 1, 2, 3, .. .
Furthermore, the eigenvalues, An, n = 1, 2, 3, ..., have the following properties
All eigenvalues, An, n = 1, 2, 3, ..., are real numbers.
There is a smallest eigenvalue, usually denoted A1.
There is no largest eigenvalue. This means An - oo as n -f 00.
For the eigenvalues An and Am, we have An Am unless n = m. Therefore,
the eigenvalues may be ordered in the following manner:
Ai <A2 <A <... <An <...
Corresponding to each eigenvalue, An, n = 1, 2, 3, ..., there is an unique
eigenfunction, denoted co (x), n = 1, 2, 3, ..., which has exactly n - 1 zeros
on0<x<L.
We will only prove selected parts of Theorem (47). For a complete set of proofs,
I refer the reader to An Introduction to Partial Differential Equations by Michael
Renardy and Robert C. Rogers. We prove that all eigenvalues are real numbers,
and eigenfunctions corresponding to each eigenvalue are unique, after we state and
prove the next theorem.
Theorem 48. The eigenfunctions, (Pn (x), n = 1, 2, 3, ..., corresponding to differ-
ent eigenvalues, An, n = 1, 2, 3, ..., are orthogonal relative to the weight function
a(x). This means
L

f (Pn (x)(Pm (x)U(x) dx = 0, whenever An Am.

It should be noted that eigenfunctions are unique up to an arbitrary constant mul-

tiplier.
284 Chapter 8: The Sturm-Liouville Eigenvalue Problem

Proof. Let cPn (x) and Pm (x), with corresponding eigenvalues An Am, be solutions
to

x (x)'(x)] + q(x)cp(x) + AQ(x)cp(x) =0, 0 < x < L,

subject to

ai'P(0) + a2'P(0) = 0

a3cp(L) + =0.
Thus, we have

q(x)cPn(x) +Ano(x)cpn(x) = 0, 0 < x < L, (8.15)

dx
subject to

ai'Pn(0) + a2pn(0) = 0

a3pn(L) + =0,
and

(x) =0, 0 < x < L, (8.17)

dx
subject to

al 'm (0) + a2 'm (0) = 0

a3'm (L) + a4( L) =0.

Multiplying Equation (8.15) by c0m (x) and Equation (8.17) by 'n (x) yields

x (x)(x)] 'Pm(x) + 9(x)'Pn(2)cPm(x) + =0 (8.19)

and

(x)(x)] 'Pn(x) + q(x)'Pm(x)'Pn(x) + AmU(x)'Pm(x)'Pn(x) = 0, (8.20)

dx
respectively. Subtracting Equation (8.19) from Equation (8.20) yields

0= (x)(x)] n(X) - (x)(x)] m(X)

+ AmQ(x)com(x)con(x) - AnU(x)'Pn(x)'Pm(x). (8.21)

 Section 8.2: Definition of the Sturm-Liouville Eigenvalue Problem 285

Equation (8.21) may be rewritten as

(An Am) (X)n(X)m(X) = n(X)

d {p(x)(x)] com(x).

Integrating both sides of Equation (8.22) with respect to x from 0 to L yields

(An - Am)J L a(x)n(x)m(x) L dx

(x)(x)] <Pn(X) dx

-f L dx dx. (8.23)

We may integrate the right side of Equation (8.23) by using integration-by-parts.

Thus, we have

(An - Am)J lax

0

L
= f 0
dx

L
+J 0
dx+

which becomes

(An - Am)J L dx
0

= (s.24)

Evaluating the right side of Equation (8.24) yields

L
(An - Am)J d2
0

-p(L)Pn(L)Prn(L) +1()On(O)Pm(O),
286 Chapter 8: The Sturm-Liouville Eigenvalue Problem

which may be written as

L
(An - Am)J d2
0

= p(L)

p(O) (8.25)

Using Equations (8.16 and 8.18) in the forms

aicon(O) = -a2(p(O)

(k3cPn(L) _
and

(kl(Prn(O) _ -a2(O)

a3com (L) = -a4(Pm (L),

we evaluate the right side of Equation (8.25). The evaluation yields
L
(fin - Am)f dx = 0.
0

Since An Am, we must have

L

thus completing the proof.

f dx=O,

We can now prove, from Theorem 47, that all eigenvalues are real and that
eigenfunctions corresponding to each eigenvalue are unique.
Proof. Suppose the eigenvalues, An, n = 1, 2, 3, ..., are not real numbers. Hence,
they are complex. Thus, each eigenvalue, An, n = 1, 2, 3, ..., has the form
a + bi. Let A be a complex eigenvalue and cp(x) be the corresponding eigenfunction.
The function, cp(x), must be complex since the differential equation defining the
eigenfunction is complex. Thus, (x), with its corresponding eigenvalue A, is a
solution of

dx (x)'(x)] + o, o < x < L, (8.26)

subject to
ai(O) + a2'(0) = 0
(s.27)
a3(L) + =0.
 Section 8.2: Definition of the Sturm-Liouville Eigenvalue Problem 287

Please note the functions p(x), q(x), and a(x) are real functions. Since A is a com-
plex eigenvalue with corresponding eigenfunction cp x ,then A (complex conjugate)
is an eigenvalue with corresponding eigenfunction (x). This statement is devel-
oped from a similar theorem about complex roots of quadratic equations. Thus,
(x), with its corresponding eigenvalue A, is a solution of Equations (8.26 and 8.27).
Hence, we have

dx [p(x)] + =0,0 < x < L, (8.28)

subject to
al(0)+a2'(0) =0
a3(L) + CY4cP'(L) =0.

By Theorem (48) we know o(x) and (x) must be orthogonal. Therefore,

L
(A - A) I dx = 0, (8.30)
0

which may be written as

(A_)fa(x) (x)2 dx = 0

I2
since (x)co(x) = I cp(x) > 0. Also, a(x) is defined to be greater than 0 for all x on
the interval. Thus,
L

f a(x) (x)2 dx > 0. (8.31)

Equation (8.31) may only equal 0 if cp(x) - 0 (identically equal to 0), which we
know can't be true since we may only have n - 1 zeros for 0 < x < L. Hence, we
must have (A - A) = 0. Therefore, A = A, which implies A is real.
Finally, we prove eigenfunctions corresponding to each eigenvalue are unique up
to an arbitrary constant multiplier.
Proof. Suppose cp1(x) and cp2 (x) are eigenfunctions corresponding to the same eigen-
value, A. Then, we have

dx (x) (x)] + (x) + (x) = o, o < x < r,, (8.32)

subject to
(O) + (0) = 0
(8.33)
(L) + a4( L) _ 0,
288 Chapter 8: The Sturm-Liouville Eigenvalue Problem

and

dx
(x)(x)] + q(x)2(x) + o, o < x < L, (8.34)

subject to
al2(0) + a2(p(0) = 0
(8.35)
(8. 35)
a4(L) = 0.
If we multiply Equation (8.32) by cp2(x), then subtract from it Equation (8.34)
multiplied by cpl (x), we have, after some algebraic manipulation,

(x)(x)] pa (x) - dx (x)(x)] i(X) _ 0,

dx
which may be written as

(x)(x)] Pa(x) - dx (x)(x)] i(x).

dx
We may integrate Equation (8.36) with respect to x over the interval 0 $ x $ L.
We have
fL fL
dx (x)(x)] Pa(x) dx - dx
dx.

Performing the integration by parts yields

L
Pa(x) (x)(x)] Pa(x) dx
0

L
_ Pi (x) (x)] o -f 0
(x) dx+

which becomes

o = v2(x) (x)(x)] to - vi(x)

f 0
L (x)
0
(x) dx,

or just

o = v2 (x) to - i(x) (x)(x)] to (s.s7)

Equation (8.37) may be written as

p(x) 0.
 Section 8.2: Definition of the Sturm-Liouville Eigenvalue Problem 289

By definition, p(x) is greater than 0 for all x on the interval, 0 < x < L. Therefore,
we may divide both sides of Equation (8.38) by p(x). This yields

(2(x)(x) - vi (x)(x)) to =0. (8.39)

Multiplying both sides of Equation (8.40) by 1 2 Yields

(x)]
L
(p2(x)p1(x) - co1(x)p2(x))
= o. (8.40)
(x)12
[Pa 0

Equation (8.40) may now be rewritten as

L
d
=0,
dx cp2(x)
0

which implies

llx = C(p2(x)
Thus, cot (x) is a constant multiplier of co1(x). Therefore, eigenfunctions corre-
sponding to each eigenvalue are unique up to an arbitrary constant multiplier. This
completes the proof.

Theorem 49. Any piecewise smooth function may be represented by a Fourier se-
ries of the eigenfunctions con (x) . This means

f (x) ancPn(x),
n=1

where
L
f(x)p(x)a(x) dx
0
Cln = J fL
cpn(x)o(x) dx
0
J
Furthermore, the infinite series converges to f (x) wherever f (x) is continuous on
f (x f (x ) where f(x) has
- -
0 < x < L, and it convey9 es to ) 2
2
discontinuity.
Sometimes the eigenfunctions, con (x), n = 1, 2, 3, ..., are referred to as forming a
"complete" set.

The proof of Theorem (49) is beyond the scope of this course. A generalized
proof may be found in Introductory Functional Analysis with Applications, Seventh
Edition by Erwin Kreyszig.
 290 Chapter 8: The Sturm-Liouville Eigenvalue Problem

Theorem 50. The expression

+ fL [p(x) ((x))2 - dx
L

f cpn(x)Q(x) dx

called the Rayleigh quotient, relates any eigenvalue, An, to its corresponding
eigenfunction, (pn(x), for n = 1, 2, 3, .. .
Proof of the Rayleigh quotient is provided in the next section.
A more general expression of the above theorems results if we state the theorems
with the bounds a <x <b instead of 0 <x <L.

EXERCISES 8.2

8.2.1. Consider the eigenvalue problem

x2 (p" (x)
+ x(p' (x) + A(o(x) = 0, (8.42)

subject to
(1) = 0 and (b) = 0, 1 <b.

(1) Show that multiplying Equation (8.42) by 1 and performing some al-
gebraic manipulation allows you to put Equation (8.42) in the regular
Sturm-Liouville form.
(2) Show that A > 0.
8.2.2. Consider the non-Sturm-Liouville differential equation

P"(x) + A 3(x)co(x) = 0.

Multiply this equation by H (x) . Then, determine H (x) so that the equation
may be reduced to the standard Sturm-Liouville form:

(x)'(x)] + q(x)(x) + Aa(x)(x) =0.

8.2.3. Consider the following equations:

Equation Name Equation
(1) Bessel's x2 u" + xu' + (x2 - n2) u = 0,

(2) Chebyshev's (1 - x2) u" - xu' + n2u = 0,

(3) Euler's ax2u" + bxu' + cu = 0,

 Section 8.2: Definition of the Sturm-Liouville Eigenvalue Problem 291

Equation Name Equation

(4) Hermite's u" - 2xu' + 2nu = 0,

(5) Legendre's (1 - x2) u" - 2xu' + n (n + 1) u = 0.

Reduce them to the standard Sturm-Liouville form.
8.2.4. Generate a set of orthogonal polynomials from the sequence 1, x, x2, x3,
on the (-1, 1). Are the orthogonal polynomials orthonormal?
8.2.5. Given
co"(x) + A o(x) = 0,
find the eigenvalues and eigenfunctions and graph the first few (more than 3)
of the eigenfunctions for each of the following boundary conditions:
(1) p'(0) = 0 and p(7r) = 0.

(2) cp(O) = 0 and cp'(27r) = 0.

(3) p(0) + p'(0) = 0 and p(l)=O.

(4) cp(O) - p'(0) = 0 and cp'(2) = 0.

(5) cp(O) = 0 and cp(7r) + cp'(7r) = 0.

8.2.6. Consider the following Legendre Polynomials, Pn (x):

Po = 1,

P1 = x

3x2 - 1
P2 =
2

5x3 - 3x
P3 2 ,and

35x4 - 30x2 + 3
P4
8

(1) Show by direct calculation that they are orthogonal to each other on
-1<x<1.
(2) Show by direct calculation that the integral of the square of each of the
Legendre polynomials on -1 <x < 1 is equal to
2
2n + 1'
where n is the subscript of the respective Legendre polynomial.
 292 Chapter 8: The Sturm-Liouville Eigenvalue Problem

8.2.7. Consider the following Laguerre polynomials, Ln (x):

Lo (x) = 1,

L1(x) = 1-x,
L2 (x) = x2-4x+2, and
L3 (x) = -x3 + 9x2 - 18x + 6.
(1) Show by direct calculation that they are orthogonal to each other on
0 <x < oc with weight function e-x.
(2) Show by direct calculation that the integral of the square of each of the
Laguerre polynomials on 0 <x < Do with weight function e-x is equal
to
(n!)2

where n is the subscript of the respective Laguerre polynomial.

8.2.8. Consider the following Hermite polynomials, Hn (x) :

Ho(x) = 1,

Hl (x) = 2x,

H2(x) = 4x2-2, and

H3(x) = 8x3 - 12x.

(1) Show by direct calculation that they are orthogonal to each other on
-oc <x < oc with weight function e-X2.
(2) Show by direct calculation that the integral of the square of each of the
e_x2
Hermite polynomials on -oc <x < oc with weight function is equal
to

2nnk./,
where n is the subscript of the respective Hermite polynomial.
8.2.9. Consider

(x'(x))'+A()(x) 1

subject to

cp(l) = 0 and cp(5) = 0.

 Section 8.3: Rayleigh Quotient 293

The general solution is

o (x) = a cos (/lnx) + b sin (/5lnx).

Find the eigenvalues and eigenfunctions and verify that the eigenfunctions are
orthogonal directly by integration.

8.3 RAYLEIGH QUOTIENT

In the previous section, we stated the Rayleigh quotient without proving it. In
this section, we restate and prove the Rayleigh quotient. Also, we discuss several
properties of the Rayleigh quotient, and some ways the Rayleigh quotient is used.
Given the regular Sturm-Liouville eigenvalue problem

d (x)'(x)] + q(x)(p(x) + o, o < x < L, (8.43)

subject to
al(p(0)+a2(,o'(O) =0
(8.44)
a3(L) + a4(p'(L) _ 0,
we show each eigenvalue, An, n = 1, 2, 3, ..., is related to its eigenfunction, c ' (x),
n = 1, 2, 3..... Thus, we show
L
f [((p ,(x))2 - q(x)cp2(x), dx
L
cp2(x)Q(x) dx
Jo

Proof. Let cpn(x) be a solution of Equation (8.43). Then, Equation (8.43) becomes

(x)(x)] + 9(x)<Pn(x) + Anv(x)cpn,(x) =0, 0 < x < L. (8.45)

Multiplying Equation (8.45) by cp(X) yields

dx
(x)(x)] (x) + 0,

which may be written as

9(x)(pn,(x) + An,Q(x)(pn.(x) _ -dx (x)(p(x)] Pn,(x). (8.46)

Integrating both sides of Equation (8.46) with respect to x from 0 to L gives us

f0
L dx = f L
0
-d (x)(p(x)] Pn(x) dx. (8.47)
294 Chapter 8: The Sturm-Liouville Eigenvalue Problem

Rearranging the left side of Equation (8.47), and noting An is a constant, yields

f L dx + An
0
L (x)(x) dx = f L -dx (x)(x)] Pn(x) dx. (8.48)
0

Using integration by parts on the right side of Equation (8.47), Equation (8.48)
becomes

An IL
J0L 0

L
_ f
0
p(x) ((p (x))2 dx. (8.49)

Performing some algebra manipulations on Equation (8.49) results in

+f L [p(x) ((p(x))2 dx
L

f co(x)o(x) dx

Thus, we have shown the Rayleigh quotient relates each eigenvalue An with its
corresponding eigenfunction (pn (x). D

One use of the Rayleigh quotient is to directly prove that An, n = 1, 2, 3, ... is
greater than or equal to 0 if

1) 0 and

2) q(x) < 0 for 0 < x < L.

A second use of the Rayleigh quotient occurs when the eigenfunctions of the Sturm-
Liouville eigenvalue problem cannot be determined. In this case, we can generally
find a lower bound for the first eigenvalue. Hopefully, it is 0. Also, we can always
find an upper bound for the first eigenvalue. The proof for finding the upper bound
for the first eigenvalue is shown using the theory of an applied branch of mathematics
called calculus of variations. We will not do the proof here. However, we state the
resulting theorem.
Theorem 51. The minimum value of the Rayleigh quotient over all continuous
functions satisfying the boundary conditions, al(p(0) + a2(p'(0) = 0 and a3(p(L) +
a4(p' (L) = 0, is the lowest eigenvalue, A1.

We will demonstrate the theorem using a simple example.

EXAMPLE 8.4. Given
(p'(x) + a(p(x) = o, (8.50)
 Section 8.3: Rayleigh Quotient 295

subject to

(8.51)
o'(1) = 0.

From previous work, we know Al =

2
2.467 and cp1(x) = sin However,
rx
4 ¶.

the point of this example is finding a continuous function satisfying the boundary
conditions, which is not cp1(x). First, we find the lower bound of the eigenvalues.
From Equation (8.50), we immediately determine p(x) = 1, (x) = 1, and q(x) = 0.
This, in turn, means the Rayleigh quotient becomes
1

f ((x)) dx
(8.52)
p(x) dx
L
Using the BCs stated in Equation (8.51), we have cpn (x)cpn (x) Io = 0. Thus, since
(p(x))2 > 0 and SPn (x) 0 for all n, we know Equation (8.52) is greater than
or equal to 0. Therefore, the lower bound for the first eigenvalue is 0. To find an
upper bound, we must find a continuous function that satisfies the BCs, Equation
(8.51). Consider the trial function

ft(x) = x2 - 2x.

We immediately see that ft(O) = 0 and f t (1) = 0. Thus, ft(x) is a continuous

function which satisfies the BCs, Equation (8.51). Next, using the Rayleigh quotient
with ft(x) replacing cp(x), we find an upper bound. Since ft(x)f(x) o = 0, we have
1 1 1

(f(x))dx (2x-2)dx 4x2 - 8x + 4 dx

o = o = o
1 1 1

f 2 (x) dx (x2 - 2x)2 dx x4 - 4x3 + 4x2 dx

0 0 0

or

3 1
43 - 4x2 + 4x
0<A1< - x4 + 4x3 1
0=2.5.
5 3 IO

..Therefore, 0 < Al < 2.5. We note the exact value of Al = ti 2.467 < 2.5.
4

Next, a slightly more complicated example using the more general form of the
Rayleigh quotient is considered.
296 Chapter 8: The Sturm-Liouville Eigenvalue Problem

EXAMPLE 8.5. Given

(x'(x))' + A () (x) =0, 1 x 2, (8.53)

subject to

(8.54)
cp(2) = 0,

find the lower and upper bound for the first eigenvalue.

Solution: First determine p(x) = x, (x) = 1,

x and q(x) = 0, and set up the
Rayleigh quotient, which is
2

f
x

fin 1
x- x<2,
- x('(x))-
>0on1 Pn 2 > 0on1 - x<2foralln
- ,

2
and f(x)
n 1x dx > 0, we must determine ifpn n >- 0. 2
2
-x(x)'(x) I 1

We have -x(pn(x)cpn(x)= (1)(1) by the

second BC. Using the first BC we know p' (l) = cp (1) . This means cpn (1) cp'n (1)
becomes p(l)p(l) = p(1) > 0. Thus, 0 is the lower bound for A1.
Next, we determine a trial function and find the upper bound. Determining a trial
function, while sometimes not easy, is always possible. We let ft(x) = Axe + Bx + C
and apply the BCs to determine A, B, and C. We have from the first BC f t (1) -
ft(i) = 0 or 2A + B - A - B - C = 0, which implies A = C. Thus, by the first BC
we have ft(x) = A(x2 +1) + Bx. The second BC tells us ft(2) = 0 or 5A + 2B = 0,
5
which implies B = A. If we let A = -2 then B = 5' and our trial function
2
becomes ft(x) = -2 (x2 + 1)+5x. Now, we substitute ft(x) for p(x) in the Rayleigh
quotient. We have

0 < Al <
-x (-2(x2 + 1) + 5x) (-4x + 5) + f x (-4x + 5)2 dx ti 17.05.
2
(_2(x2+1)+5x)2 1 dx
1
x

Therefore, 0 < Al < 17.05. The large gap between 0 and 17.05 may mean our trial
function was not very good. If we had chosen as our trial function x3 - 5x + 2,
then our upper bound would become approximately 11.52, which decreases the gap.
 Section 8.3: Rayleigh Quotient 297

That is, 0 < Al < 11.52 < 17.05. A far better approximation occurs with the trial
function

J x, 1<x<1.5
fe(x) _
-3x+6, 1.5<x<2 ,

which, when used in the Rayleigh quotient, yields an upper bound of 0.583734.
Today, one of the major uses of the Rayleigh quotient is a numerical estimate
on the beginning bounds of eigenvalues when solving a variable coefficient PDE
numerically with a computer. We now develop a general solution to a variable
coefficient PDE.

EXERCISES 8.3

8.3.1. Use the Rayleigh quotient to obtain a reasonably accurate upper bound for
the lowest eigenvalue of the following:
(1) cp"(x) - x2cp(x) + Acp(x) = 0 with cp'(0) = 0 and cp(1) = 0.

(2) cp"(x) - xcp(x) + Acp(x) = 0 with p'(0) = 0 and p'(1) + 2p(l) = 0.

(3) cp"(x) + 0 with cp(O) = 0 and p'(1) + p(1) = 0.

8.3.2. Given

(x'(x))'+A()(x)=0, 1 < x < 2,

subject to
p'(1) - cp(l) = 0 and cp(2) = 0,
show that the trial function
x, 1<x<1.5
ft(x) _ J
-3x+6, 1.5<x<2
yields an upper bound on Al of 0.583734.
8.3.3. Consider the eigenvalue problem
cp"(x)
- x2co(x) + Acp(x) = 0,
subject to
p'(0) = 0 and p'(1) = 0.
Show that A> 0.
 298 Chapter 8: The Sturm-Liouville Eigenvalue Problem

8.3.4. For the eigenvalue problem

d4 o _
dx2 - - Ae (p,
subject to the boundary conditions
p(0) = 0, (p(1) = 0,

do(0) _ d2(p(1)
dx - 0, and dx2 = 0.

Show that the eigenvalues are less than or equal to 0. Also, answer the problem
if A = 0?
8.3.5. Determine all negative eigenvalues (if any) for
(p"(x) + 5o(x) + A(p(x) = 0,
subject to
cp(0) = 0 and cp(7r) = 0.

8.4 THE GENERAL PDE EXAMPLE

Consider a one-dimensional rod of length L with perfect lateral insulation. Sup-
pose the rod is composed of several materials that are not uniformly mixed. Thus,
thermal conductivity, mass density, and specific heat are functions of the spatial
variable. Also, suppose there is a sink term, which is proportional to the temper-
ature distribution. Determine the temperature distribution for all time in the rod
if it is known the boundary conditions are homogeneous Dirichlet conditions, and
the initial temperature distribution is f (x) .

The problem just stated has the following equations:

c(x)n(x)
at - ax IKo(x)l
ax] - au(x, t), (8.55)

subject to
u(0, t) = o
(8.56)
u(L, t) = 0
an d
u(x, 0) = f (x). (8.57)

Since Equation (8.55) and BCs, Equation (8.56), are linear and homogeneous,
we know separation of variables is a valid solution method. Thus, letting u(x, t) =
(p(x)G(t) yields the time equation of
G' (t) = -AG(t), (8.58)
 Section 8.4: The General PDE Example 299

which we know has a solution

G(t) = ae-)t (8.59)

and spatial equation

[Ko(x)o'(x)]' - ao(x) + Ac(x)p(x)o(x) =0, (8.60)

with spa ti a l BC s

(p(O) = 0
(8.61)

We know from the description of the problem that Ko (x) and c(x) p(x) are positive
functions for the length of the rod. Therefore, Equations (8.60 and 8.61) are of the
regular Sturm-Liouville type where Ko (x) is p(x), -a is q(x), and c(x) p(x) is o(x).
Thus, we are guaranteed the existence of eigenvalues and eigenfunctions, although
we may not be able to identify them. Setting up the Rayleigh quotient, with the
appropriate functions replaced, results in

[Ko(x) ((x))2 - a(x)] dx

0

f(x)c(x)p(x) dx

Applying the BCs, Equation (8.61), tells us An is greater than 0. Thus, we have
a lower bound on our first eigenvalue. Using the solution to the time equation,
Equation (8.59), and what we know about the solution to the spatial equation,
yields the general solution for u(x, t), which is

u(x, t) = (8.62)
n=1

We find the specific solution by applying the initial condition. Thus, we have

u(2,0) = f (x) -
n=1

or simply

.f(x) _

An by applying orthogonality. This means we multiply both sides of Equa-

tion (8.63) by cpm (x) and integrate over the interval where the rod is defined. Re-
member, we must also multiply by (x) = c(x)p(x), which is the weight function.
 300 Chapter 8: The Sturm-Liouville Eigenvalue Problem

Thus, we have

f
L
f dx = fA()()()()
n=1

_ An L L cpn(x)cp,n,(x)c(x)p(x) dx.
n=1

By orthogonality,

O,n#m
L
Pn(x)m(x)c(x)p(x) dx = fr. (8.64)
L cpn(x)c(x)p(x) dx, n = m.
Jo

Thus, the form of the solution for An is

` 4n -
f f(x)p(x)c(x)p(x) dx
0
L
L (8.65)
(pn(x)C(x)P(x) dx
Jo

Therefore, the solution to Equations (8.55, 8.56, and 8.57) is Equation (8.62) where
An is given by Equation (8.65).

EXERCISES 8.4

8.4.1. Given
a2u ar aul
(x)ax -Bu,
p(x) at2 - ax T e

subject to

a a°' t) = 0 and =0.

and

au(x, 0)
u(x, 0) = f (x) and at = g(x),

determine the general solution to include the Rayleigh quotient for An, the
orthogonality relationships, and the equations for the coefficients of the gen-
eralized Fourier series. You may call the eigenfunctions cp(x).
 Section 8.4: The General PDE Example 301

8.4.2. Consider

p(x) at - a T (x)ax Q(x),

subject to
u(0, t) = Tl and u(L, t) = T2
and

u(x,O) = f(x).
(1) Let v(x) be a solution of the problem
[T(x)v']' - /3(x)v = -Q(x),
subject to
v(0) = Tl and v(L) = T2.
If z(x, t) = u(x, t) - v(x), find the boundary value problem satisfied by
z (x, t).
(2) Consider the same PDE, only now subject to the BCs
au(O,t)- andau(L,t)
au(0' t) =T1 + a 2u(L' t) =T2.
ax 1 ax
Find the general solution.
(3) Using the methods of this problem solve
au a2 u
3,
at =axe -
subject to
u(0, t) = 2 and u(7r, t) _ -5,
and

u(x, 0) = x2 +2x -1.

(4) Using the methods of this problem solve
au a2u
at axe -sin 2x,
subject to
au (7r, t)
u
()
0 t = 0 and
ax
=1
and

u(x,O) = cosx.
302 Chapter 8: The Sturm-Liouville Eigenvalue Problem

8.5 PROBLEMS INVOLVING HOMOGENEOUS BCS OF

THE THIRD KIND
Boundary conditions of the third kind, or Robin's boundary conditions, were last
discussed in Chapter 3. In this section, we thoroughly investigate Robin's boundary
conditions. While we discuss homogeneous Robin's BCs, as shown in Chapters 2 and
3, Robin's BCs may be nonhomogeneous. Some problems in the exercise section of
this chapter have nonhomogeneous Robin's BCs. We start with a general problem.
Consider a thin metallic rod made of a uniform material of length, L, with perfect
lateral insulation and an initial temperature distribution, which is an arbitrary
function of x. Also, suppose that both ends are subject to homogeneous Robin's
Boundary Conditions. Determine the solution of the initial value problem.
The mathematical model for this physical situation is
au(x,t)
at -
ka2uax(x2,t) (.)
866

subject to the IC
u(x,0) = f(x) (8.67)

and BCs
8u(0, t)
+ hou(0, t) = 0

KLt) +hLU(L,t) =0.

In Equation (8.68), the thermal diffusivity of both ends, K0 and KL, and the
coefficient of heat transfer at both ends, h0 and hL, may be the same or different.
However, since the physical description of the problem does not state specifically
that they are the same, we must model them as being different. Normally, K0 would
be modeled as -K0. However, for this theoretical model, we want to look at all
possible eigenvalues. If we modeled K0 as -K0, then we would only get positive
eigenvalues, which are generally the normal case. Also, note that the thermal
diffusivity of the rod may be different from that of the ends. Thus, the thermal
diffusivity of the rod is modeled as k.
The mathematical model is linear and homogeneous for both the heat equation
and the BCs. Therefore, separation of variables may be applied. Letting u(x, t) _
G(t)cp(x), performing the necessary derivatives, and separating the equations yields
the time equation
G'(t) _ -AkG(t) (8.69)

and the spatial equation

= -A(x), (8.70)
 Section 8.5: Problems involving Homogeneous BCs of the Third Kind 303

subject to the BCs

xo'(o) + o

KLco'(L) + hLco(L) =0,

which may be written as
-'(0)
HLco(I) _ -p '(L).
hL
In Equation (8.72), Ho = h° and HL = K
K0 L
From experience, we know that Equation (8.69) has the solution
(t) = Cne-lean t
Gn

where the an are the eigenvalues found when solving the spatial equation.
When solving the spatial equation, Equation (8.70) subject to the BCs, Equation
(8.72), we must remember to consider all three cases for A. That is, A <0, A = 0,
andA>0.
Case 1: A < 0: Let A = -s, where s > 0. Then Equation (8.70) becomes
(8.74)

which has solution

,o(x) = Cl cosh(-x) + C2 sinh(-x). (8.75)

Applying the first BC of Equation (8.72) to Equation (8.75), we find that

Cl =
- C2
H0
Applying the second BC of Equation (8.72) to Equation (8.75), we get the tran-
scendental equation
s-H0
'vJ
tanh(,/TT.l -

Letting z = \/L, we can separate the transcendental function into two functions
f1(z) = tanh z
and

z2 - HoL
f2(z) = z (Ho - HL)
304 Chapter 8: The Sturm-Liouville Eigenvalue Problem

flz)

z2 - HoL
Figure 8.1: Gra ph of .fi(z) = tanh z and fadz ) -
z(Ho - HL)

Remember that s > 0; then z > 0. We graph the two functions and look for inter-
section points. Figure (8.1) shows one possible intersection point. The intersection
point indicates the first eigenvalue. It is
z2
(s.77)

where the corresponding eigenfunction may be given as

pl(x) = dl (_cosh(x) + Ho (8.78)

Case 2: Let A = 0. Then, Equation (8.70) becomes

c o ( x) = 0,
which has solution
(x) = Clx + C2.
Applying the first BC of Equation (8.72) to Equation (8.80), we find that
Cl = - H0 C2 .

Applying the second BC of Equation (8.72) to Equation (8.80), we get

HLC2(1 - HoL) = -H0C2,
which implies C2 = 0, since we must assume that HL, Ho, and L are not identically
0.

Case 3: Let A> 0. Then the solution of Equation (8.70) is

(x) = Cl cos (v5x) + C2 sin (v5x). (8.81)

 Section 8.5: Problems involving Homogeneous BCs of the Third Kind 305

Applying the first BC of Equation (8.72) to Equation (8.81) yields

C2 X
C1 = - H
0

Applying the second BC of Equation (8.72) to Equation (8.81) gives us the tran-
scendental equation

tan(VL) = V'(HL - H0) (s.82)

A+ HoHL

Letting z = we can separate the transcendental function into two functions

fl(z) = tanz
and

Z(HL - H0)
f2(z)=
z2 + HLHoL

We graph the two functions and look for intersection points. Figure (8.2) gives a
C2 A
possible representation of the intersection points. Also, we know that Cl =- H0 .

z(HL - Ho)
Figure 8.2: Graph of fl(z) = tanz and f2(z) =
z2+HLHoL

Thus, we may write the eigenfunction equation as

cpn(x) = do (_cos nx + Ho sin (x)), n = 1,2,3,... (8.83)

We now have the complete solution to the spatial problem. Therefore, the general
306 Chapter 8: The Sturm-Liouville Eigenvalue Problem

solution to Equation (8.66) subject to the BCs, Equation (8.68), may be given by

u(x, t) = Gn (t)in (x) _ (x)

n=1

= Dle-'t (_cosh(x) + Ho
(s.84)

+ Dne-pant (- n cos Ho sin ()),

n=2

z2
where Dn = Cndn, D1 = C1d1, and Al = - L . We may find Dn by using the IC,
Equation (8.67), and the orthogonality of the eigenfunctions, and it is
L
(pn(x) f (x) dx
Dn = J° L (8.85)

where n (x) is given by Equations (8.78 and 8.83).

Using Figure (8.2), we may set up a table of the intersection points of the graphs
(HL
of the two functions f1(z) = tan z and f2() z=z
z2 +HLH0L
H°) . The intersection
points are the actual points where the two functions are equal. However, we are
interested in bounding the intersection points. Bounding the intersection points will
yield upper and lower bounds of the eigenvalues. Thus, the following table contains
bounding values for each zn starting with n = 2:
Lower Bound zn = nL Upper Bound

0 < z2= X2L < -2

7r

3ir
it < z3 = X L < 2

Sir
2ir < z4 = a4L <
2

7ir
3ir < z5= VX5 L < 2

Sir
4ir < z6= XL < 2

(2n-3)ir
(n - 2)ir < zn,XL < 2
 Section 8.5: Problems involving Homogeneous BCs of the Third Kind 307

Using Table (8.86), we may develop upper and lower bounds for each An starting
with n = 2. In general,

C(n - 2)ir12
L )
<< 1 (2n -
2L
(s.87)

(n - 2)irl 2
Again , using Figure (8.2) , it is interesting to note that , as n -- oo , an --
L J
We finish this section by presenting an example.
EXAMPLE 8.6. Consider heat conduction in a one-dimensional rod with perfect
lateral insulation of length 2ir and no internal source. Suppose the boundary at 2ir is
maintained at the constant temperature of 0°C, and the boundary at 0 is subject to
au(0' t)
Newton's law of cooling where it is known to have the relationship u(0, t) _
ax
Also, the initial condition is known to be
x+1 0<x<ir
J \'J
'-'1
-(it + 1)x
+ 2 (ir + 1) 7r < x < 27r.
r
The mathematical model for this problem is
au(x,t) a2u(x,t)

subject to
au(0, t)
u(0, t) =
8x
u(2ir, t) = 0

and the IC, Equation (8.88). The PDE and BCs are linear and homogeneous.
Therefore, we may apply the separation of variables technique. If we let u(x, t) _
G(t)cp(x), and we take the correct partials and substitute into Equation (8.89), we
have

G'(t)cp(x) = G(t)cp"(x). (8.91)

Separating Equation (8.91) and setting the equations equal to an arbitrary constant
yields the time equation

G'(t) _ -AG(t) (8.92)

and the spatial equation

(8 .93)
308 Chapter 8: The Sturm-Liouville Eigenvalue Problem

Applying the same technique to the BCs produces the BCs for the spatial problem,
which are
(O) =
(8.94)
(2ir) = 0.

Equations (8.93 and 8.94) are of the regular Sturm-Liouville type where p(x) = 1 =
o(x), q(x) = 0, c1 = 1 = a3, c2 = -1, and c4 = 0. Thus, we may use the Rayleigh
quotient to find a lower bound for A. The Rayleigh quotient for this problem is
2ir
f0
dx

f[(x)]2dx
We would like A > 0. Since ['(x)]2 > 0 and [(x)]2 > 0, we need only check
-cp(x)cp'(x). Since cp(2ir) = 0, then
2ir
(x) = (0) (0).

Also, we have (0) = cp' (0), which means

=
[()]2 >0.
Thus, A > 0. Therefore, we need only solve the eigenvalue problem for A > 0. We
know that cp"(x) _ -A(x) has the solution
(x) = cl cos V5x + c2 sin fx. (8.95)

The first BC indicates

(0) = c1 = '(0) = C2,
which implies that Equation (8.95) becomes

(x) =C2 (cos'Xx+sin'Xx).

Applying the second BC to (x) yields

(2ir) = 0 = c2 (5cos sJ52ir + sin

or

0 = /5cos V27r + sin sJ52ir. (8.96)

Equation (8.96) may be written as

- V'Xcos V27r = sin /27r,

 Section 8.5: Problems involving Homogeneous BCs of the Third Kind 309

or in the preferred form

sin 27r = tan yA27r.
- vA _ cos V27r
(8.97)

Equation (8.97) is known as a transcendental equation. If we let z = /27r, Equa-

tion (8.97) may be written as
-z
= tanz. (8.98)
27r
Our horizontal axis is the z-axis, and we have two functions, Ii (z) = tan z and
12(z) _ -f. If we graph these two equations, then the intersection points will be
27r
the solution of Equation (8.98), which means we can determine An. Figure (8.3) is
the graph of f1(z) and f 2 (z) .

It 27L 3ir 4ir 57t

ZI Z3

Z2
Z4
Z5

Figure 8.3: Graph of fl (z) = tan z and 12(z) _ -z .

27r

Using Figure (8.3), we may set up a table of the intersection points of the graphs
of the two functions f 1(z) = tan z and f2 (z) = -f--. The intersection points are
27r
the actual points that the two functions are equal. However, we are interested in
bounding the intersection points. Bounding the intersection points will yield upper
and lower bounds of the eigenvalues. Thus, the table of bounding values for each
zn is as follows :
Lower Bound zn = Xn 27r Upper Bound
7r
< zl = X 27r < 7r
2

37r
< z2 = %27r < 27r
2

57r
< z3 = X 27r < 37r
2
310 Chapter 8: The Sturm-Liouville Eigenvalue Problem

Lower Bound zn = n 27r Upper Bound

7ir
< Z4=\/)27r < 47r
2

Sir (8.99)
< z5 = X527r < 5ir
2

(2n- 1)ir
< zn = /Xri2ir < nir
2

Using Table (8.99), we may develop the table for An, which is really what we want.
Therefore, Table (8.100) contains upper and lower bounds for each An:

Lower Bound An Upper Bound

1 < Al < 1

16 4

9
1
16 < A2 <

25 9
< As <
16 4 (8.100)
49
A4 < 4
16 <
81 < < 25
A5
16 4

(2n- 1)2 n2
16
< An < 4

It i s i nteres ti ng t o no t e th a t , as n - 00, -- (2 n - 1)
We have an id ea o f
An
2

16
the bounds on the eigenvalues and of the eigenfunctions, n (X). Also, we know the
solution to the time problem Equation (8.92), which is

G(t) =

Therefore, the general solution for u(x, t) is

u(x, t) = (8.101)
n=1
 Section 8.5: Problems involving Homogeneous BCs of the Third Kind 311

The complete solution is Equation (8.101), with the equation for an, which is
f2n
f (x)cpn(x) dx
an =
J
2 dx

where f (x) is given by Equation (8.88).

EXERCISES 8.5

8.5.1. Consider the eigenvalue problem

cP"(x) + ) p(x) = 0,
subject to
cp'(0) =0 and cp'(1) + 2cp(1) =0.
(1) Prove that A > 0.
(2) Does A = 0?
(3) Determine all eigenvalues graphically. Obtain lower and upper bounds.
Estimate the large eigenvalues.
8.5.2. Consider the eigenvalue problem

cP" (x) + Ap(x) = 0,

subject to
cp'(0) = 0 and cp'(1) - 2cp(1) = 0.
(1) Prove that A> 0.
(2) Determine all eigenvalues graphically. Obtain lower and upper bounds.
Estimate the large eigenvalues.

01u 01226
- 9u,
at = 9ax2

u(0,t) _ au(o, t) = 0 and u(1, t) - t -o

8x 8x

u(x, 0) = f(x).
 312 Chapter 8: The Sturm-Liouville Eigenvalue Problem

(1) Graphically determine the eigenvalues.

(2) Estimate the large eigenvalues.
(3) Determine the solution.
(4) What happens after a long time?

au 0.402 u
= axe
at

au( a0,t)
u(0, t) - = 0 and u(1, t)+ a ax't = o

u(x, 0) = f (x).
Note: You may call the relevant eigenfunctions co (x) and assume that they
are known.

012 u 012 u
01x2
ate

a ao, t)
= u(o, t) and
u( 0, t _ -u(400, t)

au(x, 0) = O.Olx
u x 0 =0 and - 0.000025x2.

(1) Estimate the large eigenvalues.

(2) Estimate A1, and using the first term of the series, describe the solution
physically.
8.5.6. Consider
a2u a2u au
=3
ate axe - 3 at '
the equation of a vibrating string in a medium that resists the motion, subject
to

au(0, t) t)
u(0, t) = and u(7r, t) -
8x 8x
and
au (x, 0)
u x 0 = x2 2x 1 and =2x+2.
 Section 8.5: Problems involving Homogeneous BCs of the Third Kind 313

(1) Graphically determine the eigenvalues.

(2) Estimate the large eigenvalues.
(3) Determine the solution.
(4) What happens after a long time?
8.5.7. Consider
52 u a2 u 5u
ate
_652 - 6 at '
subject to

u 0 t = _ au (0, t) and u 5 t = _ 10 au (5, t)

and
au(x, 0)
u(x, 0) = ex and = sin x.
at
(1) Graphically determine the eigenvalues.
(2) Estimate the large eigenvalues.
(3) Determine the solution.
(4) What happens after a long time?
8.5.8. Consider
a2 u a2 u au
ate ax2 - at - u'
subject to
5uax' t) t
and u(, t) = 3 auax'
u(o, t) _
and
au(x, 0) = 3x2
u x 0 = x3 - 3x2 2x 1 and - 6x 2.

(1) Graphically determine the eigenvalues.

(2) Estimate the large eigenvalues.
(3) Determine the solution.
(4) What happens after a long time?
8.5.9. Consider
a2 u a2 u au au
ate ax2 at ax - u'
 314 Chapter 8: The Sturm-Liouville Eigenvalue Problem

subject to

u (0 , t) =
8u(0, t)
ax an d u (L , t) _ - 8u(L,
ax
t)

and

au(x, 0
u(x, 0) = 1(x) and
at
(1) Graphically determine the eigenvalues.
(2) Estimate the large eigenvalues.
(3) Determine the solution.
(4) What happens after a long time?
8.5.10. Consider the PDE
au(x, t) _ a2u(x, t) au(x, t)
at axe + ax
Solve the initial value problem subject to the BCs
au (7r,
u( 0't)= 0 and
ax
t) =0
and IC

u(x, 0) _ - sinx.
In five sentences or less, give a physical description of this problem.
8.5.11. Consider the PDE
a2u(x,t) __ a2u(x,t) au(x,t)
at2 5x2 + ax
Solve the initial value problem subject to the BCs
a O,t) =0
and
(31\)2
and =x2- .

In five sentences or less, give a physical description of this problem.

8.5.12. Consider the PDE
a2u(x, t) - a2u(x, t) au(x, t) au(x, t)
ate - axe - at + ax
 Section 8.5: Problems involving Homogeneous BCs of the Third Kind 315

Solve the initial value problem subject to the BCs

5u (2ir , t)
u(0, t) = 0 and =0
and IC

u(x, 0) = 1 and
au(x, 0) -x
- 2
.
at
In five sentences or less, give a physical description of this problem.
8.5.13. Consider the PDE
au(x,t) __ a2u(x,t) au(x,t)
9 4
at axe + ax
Solve the initial value problem subject to the BCs
au(0, t) au (2ir, t)
at
= 0 and t)_
ax
=0
and IC

u(x,0) = -cosx.
In five sentences or less, give a physical description of this problem.
Chapter 9

Solution of Linear
Homogeneous
Variable-Coefficient ODE
9.1 INTRODUCTION
In Appendix C, the general second-order linear homogeneous variable-coefficient
ODE is given as follows:
S(x)u"(x) + K(x)u'(x) + H(x)u(x) = 0. (9.1)

In this chapter, we will refer to Equation (9.1) as a general second-order ODE.

Common examples of this type of ODE are
Equation Name Equation

Airy's u" - xu = 0,

Bessel's x2u" + xu' + (x2 - n2) u = 0,

C hebyshev's (1 - x2) u" - xu' + n2 u = 0,

Euler's ax2u" + bxu' + cu = 0,

Hermite's u" - 2xu' + gnu = 0,

Legendre's (1 - x2) u" - 2xu' + n (n + 1) u = 0.

(9.2)

A few examples where these equations arise in problems are astronomy, quan-
tum physics, electromagnetism, propagation of electromagnetic radiation through a

317
318 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

coaxial cable, heat distribution in a circular membrane and a sphere, Schrodinger's

equation for a harmonic oscillator, and vibrations of a circular drum head. All the
Equations (9.2) are known to have series solution except Euler's equation.
This chapter starts with a brief discussion of Equation (9.1). Then, we solve
Euler's equation. (Euler's equation does not require power series methods to de-
velop the solution.) Next, we develop the power series solution method. For many
students taking this course, this material may be unfamiliar. Therefore, the theo-
rems are presented without proof. However, a reference for the complete proof will
be provided for all theorems, most of which may be found in almost any ODE text.
Please remember this is a brief overview and IS not meant as a replacement for an
in-depth study of this material.

9.2 SOME FACTS ABOUT THE GENERAL

SECOND-ORDER ODE
Given the equation
S(x)u"(x) + K(x)u'(x) + H(x)u(x) = 0, (9.3)

we assume that S(x), K(x), and H(x) are polynomials having no common factors.
Then xo is said to be an ordinary point of Equation (9.3) if S(xo) 0. If x0 is
not an ordinary point, then it is a singular point. For instance, every real number,
x0, is an ordinary point of
u"+xu=0.
Whereas, xo = ±1 are singular points of
(1-x2)u" -xu'+n2u= 0.
If x0 is an ordinary point of Equation (9.3), then Equation (9.3) may be rewritten
as

u" (x) + ic(x)u' (x) + (x)u(x) = 0,

H(x)
where ic(x) = K(x) and (x) =
Sx Sx
If, for some x0, S(xo) = 0 in Equation (9.1), then x0 is either a regular singular
point or an irregular singular point. You can determine if x0 is a regular singular
point if the following holds true:

lim(x - x
X-4xo
S is finite

and

lim (x - xo)2
x-xo
5 is finite. (9.7)
 Section 9.2: The General second-order ODE 319

For example, consider Euler's equation

x2 u" + /3xu' + xu = 0.
We have

K(x) = ,ix, and

H(x) = x.
Hence, for x0 = 0, we have S (xo) = 0;
K (x)
lim - x 0) x) = x fix = > a finite constant;
x-xo S(x) x -O O x2

H(x)
lim x - x0) 2 = lim x2 x = x a finite constant.
x-xo ( S(x) x-0 x2
Therefore, Euler's equation is an example of an equation that has a regular singu-
larity at x0 = 0. Legendre's equation has regular singular points at x0 = ±1, which
is left as an exercise.
This text examines those variable coefficient ODEs that have only ordinary
points or regular singular points, and are relevant to PDEs. ODEs that have ir-
regular singular points are extremely difficult to solve, and their study is left to an
advanced course in ODEs.

EXERCISES 9.2

9.2.1. Show that the following equations have ordinary or regular singular points:
(1) u"-xu=0.
(2) x2u" + xu' + (x2 - n2) u = 0.
(3) (1 - x2) u" - xu' + n2u = 0.
(4) u" - 2xu' + gnu = 0.
(5) (1 -x2)u" - 2xu' +n(n+ 1)u = 0.
9.2.2. Determine if the following equations have an ordinary point, a regular singular
point, or an irregular singular point at 0 and 1:
(1) xu"+(1-x)u'+xu=0.
(2) x2 (1 - x2) u" + 5xu' + 8u = 0.
(3) x5(1-x2)u"-3xu'+u=0.
(4) 2x4 (1 - x2) u" - 2 u' + 3x2u = 0.
(5) u" + 1+x u' + 4 (1 + x) u = 0.
(6) (x + 2) u" + 2xu' - (1 - x2) u = 0.
320 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

9.3 EULER'S EQUATION

Euler's equations is any equation of the form
ax2u" + bxu' + cu = 0,
where a, b, and c are constants. Since a, b, and c are constants, the usual form of
Euler's equation is
x2 u" + ,3xu' + xu = 0. (9.s)

You can determine the solutions by power series methods. However, sometimes the
solution may be determined by a proper guess, which is the way most second order
constant-coefficient linear ODEs were originally solved.
For our guess, we'll let u(x) = xr, where x 0, since 0 is a singular point of
Equation (9.8). Then, u'(x) = rxr-1 and u"(x) = r(r - 1)xr-2. Substituting u(x),
u' (x), and u" (x) into Equation (9.8) yields
-
x2r(r 1)xr-2 + flxrxr-1 + xxr =
0,

which, after some algebraic manipulations, becomes

r2+(/3-1)r+x=0.
Equation (9.9) looks exactly like the characteristic equations you found in your
ODE course for second-order constant-coefficient linear ODEs, and may be solved
in the same way. Thus, we have

r= -(-1)±-1)2-4x --ryf8,
2
where

2 and

(a-1)2-4x
2
Therefore, the roots of the characteristic equation may be expressed as

rl = -(-1)+(-1)2-4
2
x =-7+8 (9.10)

and

r2 =
-(-1)--1)2-4 (9.11)
2

We have three cases to investigate:

1 Leonhard Euler (1707-1783) was the one of the most famous mathematicians who came from
Switzerland. He published over 500 books and papers during his lifetime.
 Section 9.3: Euler's equation 321

Case 1: (/3 - 1)2 - 4x > 0, produces two real and unequal roots.
Case 2: (/3 - 1) 2 - 4x = 0, produces real and equal roots.
Case 3: (/3 - 1) 2 - 4x <0, produces complex roots.
The following examples illustrate the type of solution for each case.
EXAMPLE 9.1. Consider
x2"xu'
u - -6u=0.
If we assume a solution u(x) = xr, where x > 0, then the characteristic equation
becomes
r(r - 1) - r - 3 = 0,
which is
r2-2r-3=0.
Thus, we have Case 1: two real and unequal roots, where the roots are -1 and
3. By our assumption, u(x) = xr, we have the two solutions u1(x) = x-14 and
u2 (x) = x3. Therefore, we know that the general solution is
u(x) = C1x-1
+ C2x3. U

Case 2 is somewhat more interesting. Though we don't cover reduction of order

in this text you should be familiar with the process. It is easily found and explained
in most ODE texts.
EXAMPLE 9.2. Consider
x2u" + 5xu' + 4u = 0.
Again, we assume a solution u (x) = xr where x > 0, then the characteristic
,

equation becomes
r(r - 1)+5r+4=r2 +4r+4 = (r+2)2 = 0.
Thus, we have real and equal roots r=-2 with one solution u1 (x) = x-2. The second
solution may be found by using reduction of order, and it is u2 (x) = x-2 ln(x), which
means the general solution is
u(x) =x2 (c1 + c2 ln(x)) .

By far, the most interesting case is Case 3. To arrive at the proper solution for
Case 3, we must remember some basic mathematics. Since you already know that
the solution has complex roots, you probably expect sine and cosine functions to
appear. The question is how? Remember that
xr = er In X
From this simple formula, we arrive at the mathematically cleanest solution. We
now proceed with the example.
322 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

EXAMPLE 9.3. Consider

x2u"+2xu'+u = 0.
As before, if we assume a solution u(x) = xr, where x > 0, then the characteristic
equation becomes

r(r- 1)+2r+1 =r2+r+1 =0.

Using the quadratic formula, we find that

rl =
2

r2 =
-1 2- 2
Thus,

u1(x) = xr1

-1+t
= x 2

-l lnx if lnx
= e 2 e 2

= x 2 e iylnx
-1
2

Similarly,
-1 -2ylnx
u2 (x) = x 2 e 2

Thus,

u(x) = ciu1(x) + c2u2 (x)

-1 2f1nx -1 -iylnx
cl x 2 e 2 +c2 x2e 2

-1 iyln x -iyln x
x2 cl e 2 + c2 e 2

x 21 (dl cos ( 2 In x d2 sin 2 In x ,

which completes the solution for Case 3.

 Section 9.3: Euler's equation 323

Note: In Examples (9.1, 9.2, and 9.3) we assumed x > 0. For the complete
solution given in Proposition (52) for all x 0, we must use absolute values.

Proposition 52. Given Euler's equation

x2u" + /3xu' + xu = 0.

We can solve Equation (9.12) in any interval such that x 0 by substituting u(x) =
xr and its derivatives into Equation (9.12), solving for the roots rl and r2 of the
resulting characteristic equation

r2 + ( - 1)r + X = 0.

If the roots of the characteristic equation are real and unequal, then

u(x) = cl I xl rl + c2 Ixlr2 (9.13)

is the solution to Euler's equation.

If the roots of the characteristic equation are real and equal, then

u(x) _ (ci + c2 1n r, (9.14)

where r = r, r2 is the solution to Euler's equation.

If the roots of the characteristic equation are complex, then

u(x) _ {ci cos (bin c2 sin (bin x)], (9.15)

where r = --y ± ib is the solution to Euler's equation.

EXERCISES 9.3

9.3.1. In each of the following problems determine the general solution.

(1) x2u" + 2xu' - u = 0.

(2) x2u"+xu'-9u=0.
(3) x2u" + xu' + 4u = 0.
(4) x+1)2u" + x+1)u' + 1u=0.
(5) x2u"-xu'+u=0.
(6) x-1 2u"-3x-1 u'+9u=0.
(7) x2u" - xu' + 2u = 0.
 324 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

9.3.2. Consider the interval x > 0. Let x = ey. Knowing that Euler's equation is

x2 u" + /3xu' + xxu = 0,

show that

du 1 du
(9.16)
dx x dy

and

d2 u _ 1 d2 u l du
(9.17)
dx2 x2 dy2 x2 dy

Next, using Equations (9.16 and 9.17), show that Euler's equation becomes

d2 u du
d2+(/3-1)d +u=O. (9.18)
y y

Since Equation (9.18) is now a linear second-order constant-coefficient ODE,

find the characteristic equation and show that the roots generate the identical
cases as Proposition 52.

9.3.3. In Example (9.2) it was said that a second solution to the real and equal roots
case may be found by reduction of order. Given Euler's equation

axe u" + bxu' + cu = 0,

and the fact that rl = r2 are real roots of the characteristic equation

ar(r-1)+br+c=0.

Thus, ul (x) = xnl is one solution of Euler's equation. Use the method of
reduction of order to show that u2 (x) = xnl In x is the second solution.
 Section 9.4: Brief Review of Power Series 325

9.4 BRIEF REVIEW OF POWER SERIES

The power series you are most familiar with are Taylor and Maclaurin series. Some
common Maclaurin series are

ex

00 n2n+1
sin x
(2n+1)!
00
/-l\nx2n
l
cos x
n_O (2n)1

00
1

1-x
-1<x< 1, and
n=0

-1)n-}-lx2n
n
-1<x<1.
n=0

You should have encountered these series in your study of Calculus. However, you
may not have encountered a general definition of a power series. It is as follows:
Definition 53. An infinite series of the form
00

bn(x - xo)n (9.19)

n=0

where xo and b0, b1, ... , bn, ... are constants, is called a power series in x - x0.

Knowing the general definition of a power series only helps us if we know when
and how a power series converges. The power series given in Equation (9.19) is said
to converge if
N
lim bn(x - x0)n
N-oo n=0

exists; otherwise it is said to diverge. If x = x0, the power series must converge.
For values of x x0, we may not know whether the series converges. However, if
the power series converges, the following theorem applies:
Theorem 54. If the power series

>bn(x_xo)
n=0
326 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

converges, then either

(1) the power series converges only at x = xo .

(2) the power series converges absolutely and uniformly for all values of x.

(3) there exists a positive number R (called the radius of convergence) such that
the power series is absolutely convergent for all Ix - xo I < R, and divergent for
x - xo > R.
The proof of this theorem may be found in the text Infinite Series by Earl D.
Rainville. Also, note that in the radius of convergence, the power series is often
written as

f(x) - bn(x - (9.20)

n=0

Since we are developing a power series solution for a general second-order ODE,
it would seem useful to know if a power series may be differentiated. This is given
in the following theorem. The proof may also be found in the text Infinite Series
by Earl D. Rainville.
Theorem 55. If the power series
00

f(x) _ bn(x - xp)n, in Ix - xpl <R, (9.21)

n=o

then
00

fi(x) = nbn(x - xo)n-1, in x - xo) <R. (9.22)

n=1
Note: This theorem indicates that within the radius of convergence, we can take
as many derivatives as we need. Thus, it is possible to have a power series as a
solution to a general second-order ODE. However, there is a difference between the
series' given in Equations (9.21 and 9.22). The sums have a different starting point.
In Equation (9.21), the series starts at n = 0, and in Equation (9.22), the series
starts at n = 1. This is easily solved by shifting the indices, which is shown in most
calculus texts. That is,
Remark 1. For any integer m, the power series
00

bn (x _ xo )n-m
n=no
may be rewritten as
00

bn+m (x - xo) n .
n=no-m
 Section 9.5: The Power Series Solution Method 327

Thus, Equation (9.22) may be rewritten as

00

f '(x) _ (n+ 1) bn+l(x - xp), in Ix - xp <R.

n=0

In the next section, we show how power series may be used to solve Equation
(9.1) at an ordinary point.

EXERCISES 9.4

9.4.1. Find the Maclaurin series expansion of the following functions:

(1) ln(1 + x).
1
(2)

(3)

(9)

(5)

9.4.2. Find the Taylor series expansion of the following functions:

(1) e2x in powers of (x + 1).
(2) cos x in powers of (x - 7r).
(3) sin 5x in powers of x +
2
(4) ln(x + 1) in powers of (x + 7).
(5) 3 in powers of (x+ 1).
2x - 1

9.5 THE POWER SERIES SOLUTION METHOD

The power series solution method is the standard method for solving general second
order ODES. The idea behind solving a variable coefficient second-order ODE at
an ordinary point is to
assume a power series solution with unknown coefficients,
find all necessary derivatives,
replace the derivatives in the original equation,
combine the power series, and
set the coefficients equal to zero and solve.
328 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

As an example, consider the following:

EXAMPLE 9.4. Solve
u'-2u=0
at x0=0.
Solution: Let
u(x) _ >bnX.
n=0
Find u' (x), which is

u' (x) _ >flbnX1.

n=1
This may be rewritten as

u'(x) _ (n + 1)bn+lxn.
n=0
Replace the assumed solution and the derivative in the original equation, and we
have

(n+ 1)bn+lxn - 2 bnxn = 0

or

[(n+1)b+i -2b]x =0.

n=0
Next, set the coefficient equal to 0, and we get
(n+1)b+i -2b n =0,
which implies that

b = 2bn (9.23)

Equation (9.23) is called a recurrence relation. This means that all coefficients may
be expressed in terms of b0. For example,
b1 = 2b0,

b2 = b1=2b0,

b3 = 2 b2 = 2 (2bo ),
3 3

b4 = Zbs = 3Zbo)>
 Section 9.5: The Power Series Solution Method 329

Thus, our solution is

u(x) = b° 1 + 2x + 2x2 + 4x3 + 2x4 + ... = b°e2x

3 3 '

since

C1+2x+2x2+ 3x3+ 3x4+...

is the Maclaurin Series for e2x

Another example is Airy's equation2.

EXAMPLE 9.5. Find a series solution of Airy's equation

u"-xu=0 (9.24)

in powers of x - 2. Since Airy's equation is a second-order ODE, we expect two

linearly independent solutions.

Solution: Assume that

u(x) _ bn(x - 2)n. (9.25)

n=0

Then

u"(x) _ n(n - 1)bn(x - 2)n-2'

n=2

which may be rewritten as

u"(x) _ (n + 2)(n + 2). (9.26)

n=0

Substituting Equations (9.25 and 9.26) for u and u", respectively, in Equation (9.24)
yields

(n + 2)(n + 1)bn+2(x - 2)- x 2)= 0.

n=0 n=0

Here, we cannot simply combine the sums. To express Airy's equation in powers of
x -2, we must express x in front of the second sum in powers of x -2. This is done
20. B. Airy (1801-1892) was the Astronomer Royal of England. He made many contributions
to the study of series and integration.
330 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

by rewriting x as 2 + (x - 2). Thus, we have

(n+ 2)(n+ 1)bn+2(x - 2)n = [2 + (x - 2)]b (x -2)

n
n=0 n=0

00 00

=2 bn(x - 2)n + bn(x - 2)n+1. (9.27)

n=0 n=0

Shifting the index of the last sum in Equation (9.27), we obtain

n=0
(n + 2) (n + 1)bn+2 (x - 2)n = 2
n=0
bn (x 2)n +
n=1
bn-1(x - 2)n.
Equating like powers of x - 2 yields
(n+2)(n+ 1)bn+2 = 2bn+bn-1,
which implies

b2 = bo,

6b3 = 2b1 + bo,

12b4 = 2b2 + b1,

The general recurrence relation for n > 3 is

2bn-2 + bn-3
b =
n(n - 1)

Thus, we have
b2 = bo,

b3
= b1 + bo
3 6'

b2 bl b1
b4 = + = b° +
6 12 6 12'

b3 b2 b1
b5 = + = + b°
10 20 30 15'

Thus,
 Section 9.5: The Power Series Solution Method 331

Figure 9.1: Graph of Airy's function.

/ (x-2) 3 (x-2) 4 (x-2) 5 + ...

u(x) = bo I 1 + (x - 2)2 +
6 + 6 + 15 )

/ (x- 3
(x- 4
(x- 5 l
+ bl I (x - 2) + 302) + ...
3 2) + 122) + J

= boul(x) + biu2(x),
where
=/ (x 3 4 5
ul (x) I 1 + (x - 2)2 +
6 2) + (x 6 2) + (x 152) + ... )
and
(x - 2)3 (x - 2)4 (x - 2)5
u2(x) _ C(x - 2) + + ... f

3 + 12 + 30

and bo and b1 are arbitrary. Figure (9.1) shows the graph of Airy's function from
-10 to 5. Notice the oscillatory nature for x <0, whereas for x > 0 the function is
unbounded.
Example (9.5) demonstrates the following theorem:
Theorem 56. If xo is an ordinary point of
u"(x) + ic(x)u'(x) + e(x)u(x) = 0, (9.28)
K(x) H(x)
then icx
( ) =
Sx and e(x) = Sx have Taylor series expansions at xo> then> the
general solution of Equation (9.28) is given by

u(x) = bn(x - xo) = ciu1(x) + C2u2(x)

n=0
 332 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

where c1 and c2 are arbitrary, and u1(x) and u2 (x) are linearly independent se-
ries solutions, which have Taylor series expansions at x0. Also, the Taylor Series
expansions of u1(x) and u2 (x) have radius of convergence at least as large as the
minimum of the Taylor series expansions of ic(x) and e(x).3
Unlike Airy's equation, which may be expanded in powers of any real number,
Legendre's equation, given by

(1- x2 u" - 2xu' + n (n + 1) u = 0, (9.29)

has singularities at ±1. In the next section, we look at Legendre's Equation and its
solution.

EXERCISES 9.5

9.5.1. Find a solution to the following ODEs using the Power series method at
x0 = 0:
(1) u' = 5u.
(2) u"-2u'-3u=0.
(3) u'-2u=0.
(4) u"+4u'+4u=0.
(5) u' + 6xu = 0.
(6) u"+4u=0.
(7) u'-3u+2=0.
(8) u"+u'+u=0.
9.5.2. Find the power series solution for the following ODEs at the indicated value
of x0. That is, find the series solution in powers of x - x0.
(1) (1 + x2) u" + 3xu' + u = 0, xo = 1.
(2) (1 + x2)u" + 6xu' + 2u = 0, x0 = -1.
(3) (1 + 2x2)u" + 3x2u' - u = 0, x0 = 1.
2

(4) (1 + x2 )u" - 5xu' + 14u = 0, x0 = 2.

9.5.3. Find a series solution for Airy's equation in powers of x - 3.
9.5.4. Find a series solution for Airy's equation in powers of x + 1.
9.5.5. Find a series solution for Airy's equation in powers of x + 0.
3A proof of a more general form of this theorem was completed by Immanuel L. Fuchs (1833-
1902) in 1866.
 Section 9.5: The Power Series Solution Method 333

9.5.6. In Hermite's equation,

u" - 2xu' + nu = 0,
like Airy's equation, in which every real number xo is an ordinary point. Find
a series solution to Hermite's equation. The solutions are called Hermite
polynomials.

9.5.7. Another way of generating Hermite polynomials is by using Rodrigues' for-

mula, which is

2dne_x2
Hn(x) _ (_1)nex
dxn
Generate the first five Hermite polynomials using Rodrigues' formula.

9.5.8. Consider Ho (x) = 1 and Hl (x) = 2x.

(1) Use the recurrence formula Hn+1(x) = 2xHn (x) - 2nHn_ 1(x) to find the
next five Hermite polynomials.
(2) Show that the Hermite polynomials H2 (x) through H6 found in Part (1)
satisfy the recurrence formula Hn (x) = 2nHn _ 1(x) .

9.5.9. This problem shows how a piecewise smooth function may be expanded in a
generalized Fourier series in Hermite polynomials. The weight function for
the Hermite polynomials is e-x2. Also,
00
e-x2Hn (x) dx = 2m!/.

Let

1(x) _ AnHn (x).

n=0

Show that
1 e_x2
An = f (x)Hn (x) dx.
2nn!

9.5.10. Let

Find the generalized Fourier series in Hermite polynomials. Using your fa-
vorite mathematical software graph the function on the interval (-10,10),
and the generalized Fourier series in Hermite polynomials for n = 10, 25, 50.
 334 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

9.5.11. Let

Find the generalized Fourier series in Hermite polynomials. Using your fa-
vorite mathematical software graph the function on the interval (-5, 5), and
the generalized Fourier series in Hermite polynomials for n = 10, 25, 50.
9.5.12. Find a power series solution to Weber's equation

x2
u"+ n+ 21 - 4 u=0,
where n = 0, 1, 2, .. .

9.6 LEGENDER'S EQUATION AND

LEGENDRE POLYNOMIALS
The solutions to Legendre's equation are called Legendre polynomials, and they are
of a class of functions known as special functions. Other special functions arise from
the solutions of Bessel, Hermite, and Laguerre equations. The solutions to Legendre,
Bessel, Hermite, and Laguerre equations form a complete set. Thus, they are used
in generalized Fourier series. For instance, Legendre's equation usually occurs when
solving a PDE which involves spherical geometry, and Bessel's equation occurs when
solving a PDE which involves polar or cylindrical coordinates. Hence, these are
very important equations to solve and understand. They will be used extensively
in Chapter 10.
We now proceed to solve Legendre's equation
(1 - x2)u"-2xu'+n(n+l)u=0. (9.30)

Since S (x) = (1 - x2) is nonzero at 0 we may use the methods we have already
developed in the preceeding sections. Here, the radius of convergence is the open
interval (-1, 1). Thus, we assume that the solution has the form

u(x) = bmxm. (9.31)

m=0
Notice that the index has changed from n to m. We must do this because n already
occurs in Equation (9.30). Therefore, the derivatives of Equation (9.31) are

= (m+1)bm+ixm
m=1 m=0
and
00 00

u"(x) = m(m -
1)bmxm-2

= (in + 2)(m + 1)bm+2xm.

m=2 m=0
 Section 9.6: Legendre's equation and Legendre polynomials 335

Substituting into Equation (9.30), Equation (9.31), and the correct derivatives of
Equation (9.31) yields
00 00

0 = (1 - x2) (in + 2)(m + 1)bm+2x"' - 2x (in + 1)bm+lxm

00

+ n(n + 1) bmxm ,
m=0

which becomes
00 00

0= (in + 2)(m + 1)bm+2xm - (m + 2)(m + 1)bm+2x"'+2

m=0 m=0
00 00
-2 (m + 1)bm+lxm+1 + n(n + 1) bmxm. (9.32)
m=0 m=0

Expanding the first few terms of Equation (9.32) we obtain

0= 2b2 + (3)(2)b3x + (4)(3)b4x2 +

- 2b2x2 + .

(9.33)
- 2blx - (2)(2)b2x2 +

+n(n + 1)bo + n(n + 1)bix + n(n + 1)b2x2 + . .

Since x 0 for all values within (-1, 1), we must have

2b2 + n(n + 1)bo =0,

n(n + 1)
ba = - 2 bo (9.34)

(3)(2)b3+[ri(n+1)-2]bi =0

(n-1)(n+2)
bi (9.35)
b3 (3)(2)

The general recursion formula is

(n+k+1)(n-k)
bk+2 = - bk,fork = 0,1,2,3,... (9.36)
(k+2)(k +1)
336 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

Thus, the coefficients of the infinite series are

(n+1)n
b2 = bo
2 '

b3
(n+2)(n-
(3)(2)

(n+3)(n-2) (n+3)(n+1)n(n-2)
(9 37)
.
b o,
b4 (4)(3) b2 - (4)(3)(2)

(n + 4)(n - (n + 4)(n + 2)(n - 1)(n -

b5 3b 3bbi ,

(5)(4) b3 (5)(4)(3)(2)

Therefore, one solution to Legendre's equation, Equation (9.30), may be written as

u(x)=bo(1_ (n+1)n2 (n+3)(n+1)n(n-2)4±)

(4)(3)(2)

C (n+2)(n-1) 3 (n+4)(n+2)(n-1)(n--3)
+ bl x - f ...
(3)(2) x + (5)(4)(3)(2)

= boul(x) + blu2(x) (9.38)

where

u1(x) = 1- (!2+ (n+3)(n+1)n(n2)4±

and

(n + 2)(n - 1)x3 + (n + 4)(n + 2)(n - 1)(n -3)

U2(X) = x - ±
3! 5!

Generally, we will work with Legendre's equation when n is a nonnegative inte-

ger. Therefore, whenever n = k, either u1 (x) or u2 (x) will be a finite sum while the
other is an infinite sum. For example, suppose n = 2; we see that in u1 (x) every
term after the x2 term has n - 2 = 0 in the product of the numerator. Thus,

u1(x) = 1 - (n+1)n x2 I

2.

and

(n + 2)(n - 1)x3 + (n + 4)(n + 2)(n - 1)(n -3)

u2(x) - x -
3! 5!
 Section 9.6: Legendre's equation and Legendre polynomials 337

So whenever n is an odd integer the sum for u2 (x) is finite; whenever n is an

even integer the sum for u1 (x) is finite. These finite polynomials multiplied by
constants are called the Legendre polynomials. In most cases, we normalize the
Legendre polynomials. That is, for n = 0, we let bo = 1; for n = 1, 2, 3, ... , we let
bn = (2n
2(n')2
n
The normalization allows the polynomials to have a value of 1 when
x = 1. The formula for the Legendre polynomials is

K 2k!((2n-2k)!
P(x) _ (-l)k
- k)!( -
2k)!xn-2k

k=o

n n
where K = or 1, whichever is an integer. Thus, a simplified solution to
2 2
Equation (9.30) is

u(x) = c1Pn(x).

The first five Legendre polynomials are

P0(X)

Figure 9.2: The plot of the first five Legendre polynomials, Pn (x).

Po = 1,

P1 = x,

3x2 -
P2
2

5x3 - 3x
P3 2 ,and

35x4 - 30x2 + 3
P4
8
338 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

which are shown in Figure (9.2).

Another solution to Legendre's equation, Equation (9.30), is Legendre functions
of the second kind, denoted Qn (x). Legendre functions of the second kind are
unbounded at ± 1. The first five Legendre functions of the second kind are

Qo
l In
2
l+x
1-x '

-1+Zln(i+xl

Q1
l

Q2 = - 3x2 + 3x24 - 1 In 11_x

+x

4 - 15x2 5x2 - 3x l+x

and
Q3
6 l 4 1-x
_ 55x - 105x3 (35x4 - 30x2 + 3
In
1+x
Q4
24 + 16 1-x)'
which are shown in Figure (9.3).

Figure 9.3: The plot of the first five Legendre functions, Qn(x).

Thus, the complete solution to Legendre's equation, Equation (9.30), is

u(x) = a1Pn(x) + a2Qn(x).

Two other ways of generating the Legendre polynomials are

1
n
Pn (x) = n (x2 - 1)n, (9.39)
2 n. dx n !
 Section 9.6: Legendre's equation and Legendre polynomials 339

and

=::
1
Pn (x)tn . (9.40)
1 - 2xt + t2 n=0

Equation (9.39) is called the Rodrigues'4 formula. Equation (9.40) is known as a

generating function for Legendre polynomials.
Since Legendre polynomials are used in generalized Fourier series, they must
have the same properties as the trigonometric Fourier series. In particular, Legendre
polynomials form a complete set and are orthogonal on the interval -1 < x < 1,
which means
1

dx =0, if m n
-1
and
1 2
[P(x)]2 dx =
_1 2n± 1
These are left for you to show in the exercises. This short introduction to Legendre
polynomials completes our coverage of this topic. If you are interested in learning
more, I suggest you study the text Special Functions by Earl D. Rainville. In the
next section, we examine Bessel's equation.

EXERCISES 9.6

9.6.1. Generate the first five Legendre polynomials using the formula
n
Pn(x)=
l -p--(
2
x2 x2 - l n.
dxn

9.6.2. Show that

00
1
Pn (x)tn .
1 - 2 xt + t 2
n=0
Hint: Use the binomial theorem on (1 + w )P.
9.6.3. Show that Legendre's equation has regular singular points at ±1.
9.6.4. Prove the recurrence formula
2nd-1
Pn+1(x) = xPn(x) - n 1( )
n+ 1 n+ I n

Hint: Use the generating function and differentiate with respect to t. Then
multiply both sides of the equation by 1 - 2xt + t2.
4Olinde Rodrigues (1794-1851), a bench economist and mathematician.
 340 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

9.6.5. Given P1(x) = x and P2 (x) =

(32-
x2 1),
find P3 (x), P4 (x), and P5 (x) using
the recurrence relation given in the previous problem.
9.6.6. This problem involves a generalized Fourier series in Legendre polynomials.
If

f (x) _ AnPn (x), -1<x<1,

n=0
show that
2n+1 1
An =
nil P(x)f(x) dx.
9.6.7. Let
x+1, -1<x<0
{ 1-x, 0-<x<1.
Expand 1(x) in a generalized Fourier series in Legendre polynomials. Use
your favorite mathematical software to graph the function, 1(x), and the
generalized Fourier series in Legendre polynomials for n = 10, 25, 50, and
100.

9.6.8. Let
-1-x, -1 <x <0
x-1, 0<x<1.
Expand g(x) in a generalized Fourier series in Legendre polynomials. Use
your favorite mathematical software to graph the function, g(x), and the gen-
eralized Fourier series in Legendre polynomials for n = 10, 25, 50, and 100.
9.6.9. Let
h(x) = sin x, - 1 <x < 1.
Expand h(x) in a generalized Fourier series in Legendre polynomials. Use
your favorite mathematical software to graph the function, h(x), and the
generalized Fourier series in Legendre polynomials for n = 10, 25, 50, and
100.

9.6.10. Let
f(x)=cosx, -1<x<1.
Expand 1(x) in a generalized Fourier series in Legendre polynomials. Use
your favorite mathematical software to graph the function, 1(x), and the
generalized Fourier series in Legendre polynomials for n = 10, 25, 50, and
100.
 Section 9.7: Method of Frobenius and Bessel's equation 341

9.6.11. Let

g(x) = tanh x, -1<x<1.

Expand g(x) in a generalized Fourier series in Legendre polynomials. Use
your favorite mathematical software to graph the function, g(x), and the gen-
eralized Fourier series in Legendre polynomials for n = 10, 25, 50, and 100.

9.7 METHOD OF FROBENIUS

AND BESSEL'S EQUATION
The last equation we will examine is Bessel's equation. Bessel's equation requires
a completely new method of solution. That method is known as the method of
Frobenius.5 the method of Frobenius, like the power series method, is another
method for solving variable-coefficient ODEs. This method is briefly discussed in
this section.
The method of Frobenius is an extremely interesting and important method for
an applied mathematician, engineer, and physicist to learn. However, a complete
study of this method would require almost a complete chapter by itself. We have
neither the space nor the time to present such an in-depth method. If you would
like to learn more than just the basics of the method of Frobenius, I would refer
you to either Elementary Differential Equations and Boundary Value Problems by
Boyce and Diprima or Elementary Differential Equations by William Trench. Both
texts cover this topic extensively and present some very interesting problems. We
start our presentation with a short discussion immediately followed by the primary
"lengthy" theorem, which we require for Bessel's equation. Then we work through
Bessel's equation.
In general, the method of Frobenius is studied at regular singular points, which
occur at the origin. In fact, if a regular singular point occurs at a point different
than the origin it is easily translated to the origin for convenience. Also, the method
of Frobenius requires the use of a Frobenius Series
00

IXIr>bflXn, bo 0. (9.41)
n=0

Under certain conditions, which we state later, Equation (9.41) always yields one
solution. The second solution may or may not exist. The problems we study
generally have two solutions. The solutions are either immediately available by
knowing that two roots exist or a second solution is "easily" (Remember: Everything
is relative) constructed. Since it is easier for the presentation, and it does not effect
the method for determining r and all bn, we only consider x > 0. Thus, Equation

5G. Frobenius (1849-1917), a German mathematician who also made important contributions
in group theory.
342 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

(9.41) may be written as

00

bnxn+r, bo 0. (9.42)
n=0

As a way to explain the method of Frobenius, we solve a typical problem.

EXAMPLE 9.6. Consider the ODE
2x2u" (x) + (3x + 2x2 )u' (x) - (1 - x)u (x) = 0. (9.43)

First, we determine if there is a regular singular point at xo = 0. Since S (xo) =

xo = 0 at xo = 0, we know that Equation (9.43) is a possible candidate. Now, we
must determine if Equations (9.6 and 9.7) are finite at xo = 0. Equation (9.6) gives
us

K(x) (3x + 2x2) 3

lim x = lim x = -, which is a finite constant,
x- O S(x) x- O 2x2 2'
and Equation (9.7) tells us that
H(x) -(1 - x) _ _ 1 which is also a finite constant.
lim x2 = lim x2
x- O S(x) x- O 2x2 2'
Thus, xo = 0 is a regular singular point. Another way to check to see if you have a
regular singular point is determining if xK (x) and x2 H (x) have Maclaurin Series.
We know that both xK(x) = x(3x+2x2) and x2H(x) = -(1-x)x2 have Maclaurin
series. Thus, at xo = 0, we have a regular singularity.
Second, assume the solution

u(x) = bnxn+r

(9.44)
n=0

and find the appropriate derivatives of Equation (9.44), then replace them in Equa-
tion (9.43). We have

u'(x) = (n + r)bnxn+r-1

(9.45)
n=0

and
00
1)bnxn+r-2

u" (x) = >(n + r) (n + r - (9.46)

n=0

Note: Equations (9.45 and 9.46) seem to not follow the rule for differentiation of
a power series. However, when n = 0 in Equations (9.44) we have boar and the
derivatives of this first term are rboxr-l and r(r - 1)boxr-2
 Section 9.7: Method of Frobenius and Bessel's equation 343

Replacing in u(x), u' (x), and u" (x) in Equation (9.43) with Equations (9.44, 9.45,
and 9.46) yields

0 = 2x2 (n + r)(n + r - 1)bxn+r-2 + (3x + 2x2 E(n + r)bnxn+r-1

00
-(1 _ x) bnxn+T ,
n=0

which becomes
00 00
3(n r)bnxn+T
0= 2(n + r)(n + r -
1)bnxn+T
+ +
n=0 n=0
00 00 00

+ 2(n + r)bnxn+7 +1 - bnxn+T

+ bnxn+T+1 (9.47)
n=0 n=0 n=0

Canceling out the common x'' term in Equation (9.47) and combining like sums
yields

0 = E[2(n + r)(n + r - 1) + 3(n + r) - 1] bnx

=o
00

+ [2(n + r) + 1] bnx1. (9.48)

=o

Performing some algebra and writing Equation (9.48) differently yields

0= [2r(r - 1) + 3r - 1] bo

+ °° 1(n r 1) (2n + 2r - 1)bnxn (9.49)

+ >1(2n + 2r - 1)bn_1xn.

The bo term of Equation (9.49) yields the equation

[2r(r - 1) + 3r - 1] = 0,

called the indicial equation. It tells us that r = 1 -1. The second and third lines
2'
of Equation (9.49) yield the relationship of the bn's. We have

(n+r+1)(2n+2r-1)bn = -(2n + 2r - 1)bn-1,

which yields

b"` (n+r+l)_1 . (9.50)

344 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

Thus,

ul(x)
n=0

where b 1 2 3bn_1 1 and

n(2) 2n (2)'

U2(X) = x-1 bn(-l)xn>

n=0

where bn (-1) = -1n b_1

n (-1). Also, u1 (x) and u2 (x) form a fundamental set of
solutions. Therefore,

u(x) = Clul(x) + C2u2(x) = C1x2 b ( I x+ C2x-1 bn(-1)x".

n=0 n=0
One way of finding a general solution is to define b0 (r) = 1, then to use the recur-
rence relation, Equation (9.50). This process yields

u(x) = Clx2
f- 2+ 35x2 - 385x3 + .
1 1
+CZx-1 1-x+2x2- 6x3+...

We now state the general theorem for the method of Frobenius.

Theorem 57. Given the ODE
S(x)u"(x) + K(x)u'(x) + H(x)u(x) = 0,
where x = 0 is a regular singular point,
K(x)
lim x is finite,
x --+ 0 8(x)
and
H(x)
lim x2 is finite,
x-+0 8(x)
let r1 and r2, with r1 > r2, be roots of the indicial equation
r(r-1)+kor+ho=0.
Then, we have the following three cases:
(1) If r1 - r2 is not an integer or 0, we have
 Section 9.7: Method of Frobenius and Bessel's equation 345

and
00

u2( x) I 1 /
n=1

(2) If r1 = r2 = r, we have
00

and

u2 (x) = ul (x) In x + xT B (r)x"`.

n=1

(3) If r1 - r2 = N, a positive integer, we have

u1(x) = xri [1+bn(ri)xn

n=1
and

u2(x) = Cu1(2) 1ri x + xT2 1 + Bn(r2)xn

n=1

where c may equal 0.

Proof of this theorem may be found in Appendix 4 of Advanced Engineering
Mathematics, Seventh Edition by Erwin Kreyszig. We may now consider Bessel's
equation.
Bessel's equation arises in the study of heat diffusion in a cylinder or a circular
plate and in vibrations of a circular drum head. Bessel's equation is
x2u"+xu'+ (x2 -n2),u=0 (9.51)
where, for our purposes, n is a nonnegative integer. Also, x = 0 is a regular singular
point. We leave you to show this in an exercise. Therefore, letting

u(x) = bnxn+r
n=0

and substituting u(x) and first and second derivatives of u(x) into Equation (9.51)
yields

(m + r) (m + r -
m=0
00 00

+x (m+r)bmxml + (x2 - n2) bmxm+r ,

m=0 m=0
346 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

which becomes, after some algebraic manipulation and canceling the common x''
term,
00

0=
[(m+r+n)(m+r-n)]bmxm

m=0
+ bmxm+2. (9.52)
m=0

Equation (9.52) may be rewritten as

0= (r+n)(r-n)bo+(l+r+n)(l+r-n)blx
+(2+r+n)(2+r-n)b2x2+
+box2+
Thus, the indicial equation, which comes from the coefficient of the x0 term, is

(r+n)(r-n) =0, (9.53)

which means r1 = n> 0 and r2 = -n. Hence Ti - r2 = 2n > 0. The next equation
is

(1+r+n)(1+r-n)b1 =0,
which must equal 0 for all choices of nonnegative integers n and values of r. There-
fore, b1 must equal 0. The recurrence relation is given by the equation

(2+r+n)(2+r-n)b2+bo =0,
which can be rewritten as

bm
m 1 b m_2 (9.54)
(m+r+n)(m+r-n)
Since b1 = 0 and n is a nonnegative integer, the recurrence relation indicates b3 =
b5 = = 0. Thus, m is always even, so we may replace it with 2m in Equation
(9.54), which becomes

(9.55)
(2m + r + n)-1(2m + r - n) b2m 2
b 2m _

If we make the assumption that r = n, Equation (9.55) becomes

b 2m
2m 1
(2m + 2n)2m 2m 2 b_= 22m(m + n) b_ m=1,2,3,...
2m 2,
1
 Section 9.7: Method of Frobenius and Bessel's equation 347

Thus,
bo
b2
22(1+n)

-b2 bo
b4
22(2)(2 + n) 24(2)(2 + n)(1 + n)'

-b4 -bo
b6
22(3)(3 + n) 26(3)(2)(3 + n)(2 + n)(1 + n)'

The general formula is

(- 1)mb
22mm!(1 + n)(2 + n)(3 + n) (m + n)
Therefore, if we let n = 0, we have
r 00 (_1)mX2ml
ui(x)=bo 1+ 22m(m!)2j'
L m=1

which is the Bessel function of the first kind of order zero, and is denoted as Jo (x) .
Since n = 0, we have case (2) of Theorem 57, a double root. Hence, the second
solution has the form

-0.4

Figure 9.4: Bessel's function of the first kind for n = 0, 1, 2, 3.

Figure (9.4) shows the graph of Bessel's function of the first kind for n = 0, 1, 2, 3.

U2(X) = ul(x) lnx + Bmx'"` = Jp(x) lnx + Bmxm

m=1 m=1
 348 Chapter 9: Solution of Linear Homogeneous Variable-Coefficient ODE

Bm can be determined by taking first and second derivatives of u2 (x) and replacing
u2 (x), u2 (x), and u2 (x) in Equation (9.51). The solution obtained, after some
algebraic manipulation, is

Y0(x) Y1(x)
Y2(x)

Figure 9.5: Bessel's function of the second kind for n = 0, 1, 2, 3.

(ry + In 2) Jo(x) In x +
°O
(1)1G
2(m!)2"''

m=1

where

Gm = l+ 1 +...+l+l
m m-1 2

and 'y = 0.577215664..., which is called Euler's constant. Yo (x) is called Bessel's
function of the second kind of order zero. Therefore, the general solution to Equa-
tion (9.51) when n = 0 is

u(x) = C1Y0(x) + C2J0(x).

Figure (9.5) shows the graph of Bessel's function of the second kind for n =
0,1,2,3.

EXERCISES 9.7

9.7.1. Use the method of Frobenius to find the indicial equation for each of the
following ODES:

(1) xu" - 2xu' + u = 0.

(2) x2u" + xu' = 0.
(3) 3xu" - (x + 2)u' - 2u = 0.
 Section 9.7: Method of Frobenius and Bessel's equation 349

(4) x2u" + xu = 0.
(5) x(u" - 2u - lu) = 0.
(6) x2u" - x2u' + 3xu = 0.
9.7.2. Another set of polynomials is Laguerre's polynomials. They can be found by
using the method of Frobenius on Laguerre's equation

xu" + (1 - x) u' + nu = 0, n= 0,1,2,3,...

(1) Show that x = 0 is a regular singular point of Laguerre's Equation.
(2) Using the method of Frobenius, develop one fundamental solution of
Laguerre's equation.
9.7.3. Use the method of Frobenius to determine one solution to the following
Bessel's equation,

x2u" + xu' + x2 - 14 u = 0.

This is known as the Bessel equation of order one-half.

9.7.4. Show that

f 1 x [Jo(Ax)]2 dx =
J0
t [Jo(t)]2 dt = a {[Jo(A)]2 + [JOF(A)]2}.

Hint: Consider the ODE y"+ t, +y = 0, where y =Jo(t). Multiply the ODE
by 2t2y'.
9.7.5. The Hypergeometric equation, sometimes called Gauss hypergeometric equa-
tion,

x(1-x)u"+[c-(a+b+1)x]u'-abu=0,
is solved by the method of Frobenius.
(1) Show that the hypergeometric equation has a regular singular point at
x=0.
(2) Show that the hypergeometric equation has a regular singular point at

(3) Use the method of Frobenius to solve the hypergeometric equation.

Chapter 10

Classical PDE Problems

10.1 INTRODUCTION
In this chapter, we consider solution techniques for some of the classical problems
of partial differential equations. They include Laplace's equation in cylindrical and
spherical coordinates, the transverse vibrations of a thin beam, heat conduction
in and vibrations of a thin circular membrane, Schrodinger's equation, the Telegra-
pher's equation, and some interesting problems in diffusion. We start with Laplace's
equation.

10.2 LAPLACE'S EQUATION

In Chapter 5, Laplace's equation in a rectangle was discussed. In this chapter,
we expand Laplace's equation to other coordinate systems. Again, Laplace's is an
extremely important equation in mathematical physics, and it naturally arises in
electrostatics, steady-state temperature field, magnetostatics, and potential flow of
an incompressible liquid. Laplace's equation has the following forms in Cartesian
coordinates,
52 u 52
226 =
ax2
- ayu - azu = 0;
2
a2

2
(10.1)

in cylindrical coordinates, Figure (10.1) provides a frame of reference,

2 15(5u)
u - r ar r ar
a2u 1
59-+--=0;
52u
(10.2)
+ r2 2

O2 u=
1 a r2 - 1 t9 / sin 8 au +
1
52 u
r2 ar ar + r2 sin 9 ae ae r2 sin2 8 acp2

351
352 Chapter 10: Classical Problems of PDEs

Figure 10.1: Cylindrical coordinate system.

Figure 10.2: Spherical coordinate system.

10.2.1 Laplace's equation in the Polar Coordinate System

We consider steady-state heat flow in a circular plate. Thus, the use of a suitable
coordinate system is required. We use the polar coordinate system, which is the
two-dimensional version of the cylindrical coordinate system. Therefore, Laplace's
equation becomes
V2u_la rau 1 a2u
O. (10.4)
r ar Dr + r2 a92 =
We assume the boundary of the circular plate is a function of 9. Thus,
u(a, 9) = f(9), (10.5)

where a is the radius of the circular plate. Figure (10.3) shows the cylindrical plate.
It appears we only have one boundary condition. However, Equation (10.4) is
in polar coordinates, which allows us to make some assumptions.
 Section 10.2: Laplace's equation 353

Figure 10.3: Laplace's equation in a circular plate.

When working with a rectangular plate, in Cartesian coordinates, it is natural to

assume that the lower left corner is at the origin. Thus, 0 < x < L and 0 < y < H.
But for a circular plate, in polar coordinates, it is natural to assume the origin is
at the center of the plate, and for any value of r, -it < 8 < it. This implies that
au(r, -ir) _ Du(r, ir)
heat flow at -7C and it must be the same. Therefore, , and for
a8 a8
any value of r, we have -7C and it as the same point on the circular plate. Thus, we
must have u(r, -7C) = u(r, 7C). These boundary conditions for the variable u(r, 8) are
identical to the conditions we encountered for the one-dimensional circular wire, and
they are the natural boundary conditions to assume for the variable 8. Therefore,
we may, for practical purposes, restate the problem given by Equations (10.4 and
10.5) as

2
u=
15(5u)
r
152u
=0,
r ar Sr + r2 582
subject to
u(r, -7C) = u(r, 7C)

au(r, -ir) au(r, ir)

and IC

u(ca, 8) = f(9). (10.8)

Also, we must note that when r = 0, Equation (10.6) would become unbounded.
Therefore, it is usual to assume that as r -* 0, we have u (r, 8) < oc, which means
that Equation (10.6) is bounded as the radius approaches 0.
The single boundary condition, which is nonhomogeneous, immediately tells us
that the "time-like" condition is in terms of r. Therefore, we proceed as before,
354 Chapter 10: Classical Problems of PDEs

using separation of variables on Equation (10.6). Let u(r, 9) = R(r)O(9). Then,

Equation (10.6) becomes

1d dR(r) 1 420(8)
O(e) R(r) = 0, (10.9)
r dr r dr + r2 d92

which may be written as

-r d dR(r) _ 1 d20(9) _
/\' (10.10)
r dr r dr O (9) d92

We now have two separate ODES-one "time-like" equation

r d (rd1T)) = AR (r)'
dr dr
which may be written as

r
2 d2R(r)
dr2
+r dR(r)
dr
-AR(r)=0,
subject to the condition

R(r) <oo asr -* 0, (10.12)

and the homogeneous equation

d2 (9)
_ -SO(B). (10.13)
d92

After separating the BCs, Equation (10.7), the homogeneous equation, Equation
(10.13), is subject to the BCs

(10.14)
= O'(ir).

Equation (10.13), subject to BCs, Equation (10.14), is well known to us and has
the solution
Ao = 0, e0(9) = ao

An = n 2 (10.15)
,n=1,2,3,...
On (8) = an cos n9 + bn sin n9

The "time-like" equation, Equation (10.11), we studied in Chapter 9 Section

9.3. The form we used there is

x2y"+/3xy'+xy=0,
 Section 10.2: Laplace's equation 355

known as the Euler-Cauchy equation.

We have two cases to consider. The first case is ) = n2 for n = 1, 2, 3, .. .
Therefore, Equation (10.11) becomes

r2 R" (r) + rR' (r) - n2 R(r) = 0.

Letting R(r) = r8, and finding the appropriate derivatives yields the characteristic
equation

[s(s_l)+s_n2]r8 =0
or s = ±n. Thus, the solution is
R(r) = clrn + c2r-n, n = 1, 2, 3,... (10.16)

Applying the condition given by Equation (10.12), IR(r) < oo as r -* 0, implies

c2 = 0. Therefore, the solution for Equation (10.16) becomes

R(r) = clrn. (10.17)

The second case is Ao = 0. If we follow the same format as for An = n2, we

would end up with the characteristic equation

[s(s-1)-i-s]r8 =0,
which yields s = 0 or a solution of R(r) = c3. However, we know that a second-order
ODE has two solutions. Let us rewrite Equation (10.11) as

d
rd r dR(r) = AR(r) = 0.
r dr
Thus, we have

d (rd')) 0
dr dr '

which implies that

r dR(r)
= C4, where C4
4 is a constant.
dr
This in turn implies that
dR(r) c4
dr r'
which implies that a second solution is

R(r) = c41n rI.

356 Chapter 10: Classical Problems of PDEs

Thus, our complete solution for Equation (10.11) for the case Ao = 0 is

R(r) = c3 + c4 In r. (10.18)

Applying the condition given by Equation (10.12), R(r) <00 as r -* 0 implies that
c4 = 0. Therefore, the solution for Equation (10.11) for the case Ao = 0 becomes

R(r) = c3. (10.19)

For the general solution to Equation (10.6), subject to the BCs, Equation (10.7),
we have u(r, 8) = O(8)R(r) = aoc3 = Ao given by Equations (10.15 and 10.19) for
A = 0, and u(r, 8) = O(8)R(r) = (aa cos n8 + ba sin n8) (clrn) = Aarn cos n8 +
Barn sin n8 given by Equations (10.15 and 10.17). Summing all of our solutions,
we arrive at

u(r, 8) = Ao + Aarn cos n8 + Barn sin n8. (10.20)

n=1

Applying our one "time-like" condition, Equation (10.8), yields

u(a, 9) = f(9) = Ao + Aaan cos n8 + Baan sin n8.

n=1

Using orthogonality we have

A0 = ff(9) d8 (10.21)

1
Aa ff(9)cosne dB, (10.22)
?ran

and
ir
1
Ba f (B)sin nB dB. (10.23)
7rctia J-
We now solve Laplace's equation in spherical coordinates.

10.2.2 Laplace's equation in the Spherical

Coordinate System
For a sphere, as shown in Figure (10.4), Laplace's equation in spherical coordinates
is given by

2 1a 2 au 1 a au 1 52n
u r2 Sr r ar + 2i959 sin 8
a8
+
r2i29
sin 2
= 0. (10.24)
 Section 10.2: Laplace's equation 357

Figure 10.4: The sphere of radius, r.

However, we will consider the problem where u(r, 8, cp) = u(r, 8). This means we
are considering Laplace's equation in spherical coordinates, which is independent of
cp. This is not an unrealistic assumption, since many problems in electrostatics are
solved in this manner. It is done by setting up the problem in such a way that it is
independent of the variable cp. Therefore, Equation (10.24) becomes
Vu_ 15/25u\
r 1
a
si n 8- = 0. (10.25)
r2 ar ar + r2 sin 8 88 a8
Like the polar coordinate system, as r -f 0, u(r, 8) < oo. Again, this means
that the function u is bounded as r approaches 0. Also, we assume the boundary
of the sphere is now a function of 8. Thus,
u(a, 8) = f(9), (10.26)
where a is the radius of the sphere. This is the "time-like" condition, which will be
used to find the coefficients of the general Fourier series. Since Equation (10.25) is
homogeneous, we let u(r, 8) = R(r)O(e) and proceed with separation of variables.
Thus, Equation (10.25) becomes
R
1r r2 R' ' O + r sin 8
sin 80' ' = 0
which when separated becomes

1 r2R''=- 1
sin 8
Therefore, we have two ODEs,
(r2R)' _ -AR (10.27)

and
(sin 80')' = -a0 sin 8. (10.28)
358 Chapter 10: Classical Problems of PDEs

We will work with Equation (10.27) first. After completing the differentiation,
it becomes

r2 R" + 2rR' + AR = 0. (10.29)

Equation (10.29) we have seen before; it is Euler's equation. (See Chapter 9.) The
usual solution method is to assume the solution has the form R = r'3, and then,
take the appropriate derivatives and determine the characteristic equation. Using
our assumed solution, we have

r2p(p- 1)r'°-2 +2rpr'°-1

+Ar= p(p - 1)r+2pr+Ar= 0.

Dividing by rP, since we assume the radius of the sphere is not 0, we get the
characteristic equation

p2+p+A=0,
which has solutions
-lf 1-4a
p 2

or, written in an easier fashion,

-1 /1
pl- 2 +U4 A (l0.30)

and

A. (10.31)

If we let Equation (10.30) equal n, then Equation (10.32) becomes -n - 1. Also, if

we multiply Equation (10.30) by Equation (10.32) we find that the eigenvalues are
= n(n + 1). (10.32)

Thus, the solution to Equation (10.29) is

C2
=C1rn+ .
=C1rn+C2r-n-1
R(r) rn+ 1 (10.33)

Since u (r, 9) < oc as r -f 0, we must have C2 = 0 in Equation (10.33). Therefore,

we have

Rn (r) = Cnr ' (10.34)

as our final solution for R(r). Next we solve Equation (10.28).

Given

(sin 90' )' = - AO sin 8.

 Section 10.2: Laplace's equation 359

We first replace A by n(n + 1), which gives us

(sin 80' )' + n (n + 1) O sin 9 = 0. (10.35)

Next, we let w = cos 9. Then, using the chain rule, we get

d0 _ d0 dw _ d0
d8 dw d8 - -sin 8 dw .

Therefore,
d0 .
sin 80 = sin 8 = - sine 9
d9 dw '
applying a trigonometric identity followed by substitution yields

- sine ed = cost 8 - 1
d _ w2 - 1 d0

Thus,
d d0 d r d0 dw
(w -1)- =;
2 2
(sinee')' = I (w -1)
dB dw dB
d de
[(w2 - 1) sin B.
dw J

-
Hence, Equation (10.35) becomes

which becomes, after canceling sin 9,

dw (
1-wed
) dw
+n(n+1)0=0. (10.36)

Replacing e with v and w with x in Equation (10.36) yields

[(1 - x2 )v']' + n(n + 1)v = 0

or

(1 - x2)v" - 2xv' + n(n + 1)v = 0. (10.37)

Equation (10.37) is Legendre's equation. (See Chapter 9.) We know it has solutions

v(x) = AnPn (x) + BnQn (x) ,

which may be written as

v(w) = AnPn (w) + BnQn (w) ,

 360 Chapter 10: Classical Problems of PDEs

or, in the preferred form,

O(cos 9) = AnPn(cos 9) + BnQn(cos 9). (10.38)

Since Legendre functions of the second kind are not finite at ± 1, cos 0 = 1, and
cos it = -1, we have Qn (cos 9) undefined at 0 and it. Therefore, we must have
Bn = 0. Thus, Equation (10.38) becomes

O(cos 9) = AnPn(cos 9). (10.39)

Combining Equations (10.34 and 10.39) we have our solution for u(r, 9), which
is

00
u(r, 9) = R(r)O(9) = cnrnPn(cos 9). (10.40)
n=0

Applying the IC, Equation (10.26), we find

u(a, 9) = f(9) = cnanPn(cos 9).

n=0

Thus, the coefficients are found by using orthogonality, and the equation for them
is
ir
cn =
2n+1 f (9)Pn (cos 9) sin 9 d9.
2(x '2 J0
Note:
2
Pn (cos 9) sin 9 d9 =
2n+1

EXERCISES 10.2

10.2.1. Given Laplace's equation in polar coordinates

la
° 2u = r ar r au 152u
0,
ar + r2 ae2 =
find the solution for each of the following BCs:
(1) u(2,9)=5.
(2) u(1, 9) = 3 sin 9.
(3) u(1,9) = 2 cos (29).
(4) u(0.5,9) = 29 - 1.
 Section 10.2: Laplace's equation 361

(5) u(ir, 0) = 2 cos 0 - sin 0.

(6) u(3,0) = 02_ 20+1.
10.2.2. Given Laplace's equation in spherical coordinates, which is independent of the
variable cp,
15/25U
ZU ra (r 8r) + r2 s n 8 a (sin B B ) = 0,
find the solution for each of the following BCs:
(1) u(2, 0) = 0.5.
(2) u(1,0) =sin (3B).
(3) u(1,0) = 2cos(0).
(4) u(0.5,0) = 50+1.
(5) u(ir, B) = 3 sin B - 4 cos B.
(6) u(3,0) = 02_30+2.
10.2.3. Show that the exterior boundary value problem
52u(x, 52u(x,
+ = x in S2,
y

u(x,y) = g(x,y) on BSt,

and

(< a, a a constant in SZ
has at most one solution.
10.2.4. Let 1 be a bounded region. Show that the Neumann problem
au
V2u+ au = f in SZ and = 9on 5SZ
an
has at most one solution if a <0 in SZ.
10.2.5. Let 1 be a bounded region. Show for the Neumann problem
V2
u =fin 1 and au =9on 5SZ .
an
Show that any two solutions differ by a constant.
10.2.6. If we write Laplace's equation in cylindrical coordinates, then
assume that it is axially symmetric (no dependence on 0), we arrive at
a2u(r, z) 1 a2u(r, z)
= 0.
are + r az2
Use separation of variables where u (r, z) = R(r)Z(z) and show that R and Z
satisfy
rR"(r) + R'(r) + ArR(r) = 0 and Z"(z) - AZ(z) = 0.
 362 Chapter 10: Classical Problems of PDEs

(1) Identify both of these ODEs.

(2) Pick reasonable BCs and give a series solution.
10.2.7. Determine the three ODEs obtained by separation of variables for Laplace's
equation in spherical coordinates.
10.2.8. Solve Laplace's equation in a slot (a vertical strip) given by
D2u(x, y) D2u(x, y)
Dx2 + D2
y - °'
subject to
u(O,y) = fi(y), O<y<H;
u(x,O) = 12(X), 0<x;
and

u(x, H) = f3(x), 0<x.

u(x, y) must remain bounded as x - oc. Hint: Let u(x, y) = u1(x, y) +
u2(x, y).
10.2.9. Solve Laplace's equation in a slot given by
D2u(x, y) D2u(x, y)
+ ay2 =0,
axe
subject to
u(0,y) = fi(y), o<Y;
u(L,y) = 12(X), O<Y;
and

u(x, 0) = f3(x), 0 < x < L,

where u(x, y) must remain bounded as y - oo. Hint: Let u(x, y) = ul (x, y) +
u2(x, y)
10.2.10. State the complete Laplace's equation in the first quadrant. Write a short
essay on how to solve this problem.
10.2.11. Poisson's equation,
a2u(x,y) a2u(x,y)
V2u(x,y) -
y
is sometimes easy to solve if f (x, y) is a constant or depends on just one
variable. Solve Poisson's equation if
 Section 10.2: Laplace's equation 363

(1) V2u(x,y) = -3.

(2) V2u(x,y)=x.
(3) V2u(x,y) = y2.
10.2.12. Solve Laplace's equation inside a semicircle of radius 1 where 0 < 0 < 7r,
subject to the following BCs
(1) The diameter is a homogenous Dirichlet condition and u(1, 0) = 02 + 1.
(2) The diameter is perfectly insulated and u(1, 0) = cos 0.
?r
10.2.13. Solve Laplace's equation inside the wedge 0 < 0 < of radius a, subject to
4
the following BCs

(1) u(r,0) = u (r, = 0, u(a,0) = 0.

4

8u (r, )
(2) u ( r, 0) = 0, ae 4 = 0 , u ( a, B) = s i n 8 .

Du (r, 0) au (r, 4
(3) = 0, = 0, u(a, 0) = cos 0.
a0 a0
10.2.14. Suppose u (x, y, z) satisfies Laplace's equation. Show that the value of u (x, y, z)
at any point (x, y, z) is approximately equal to the average of its values at the
six surrounding points (x ±6, y, z), (x, y ±6, z), and (x, y, z ± b). Hint: Calcu-
late the Taylor series expansion of u(x + b, y, z) to the term 63, similarly for
the other five points.
10.2.15. In the text, we solved the first boundary-value problem for Laplace's equation
inside a sphere of radius a, which is independent of cp. Solve the first boundary-
value problem for Laplace's equation outside the sphere. Note: The first
boundary-value problem is a Dirichlet BC.
10.2.16. Complete the following:
(1) Solve Laplace's equation for the general second boundary-value problem
inside a sphere of radius a, which is independent of the variable cp. Note:
The second boundary-value problem is a Neumann BC.
(2) Solve Laplace's equation for the second boundary-value problem inside
a sphere of radius a, which is independent of the variable cp, where the
BC is
au
= a cos 0.
an
 364 Chapter 10: Classical Problems of PDEs

10.2.17. Solve Laplace's equation for the second boundary-value problem outside a
sphere of radius a, which is independent of the variable cp.
10.2.18. Solve Laplace's equation for the general Sturm-Liouville BCs inside a sphere
of radius a, which is independent of the variable cp.
10.2.19. Find the potential in the interior and exterior of a sphere of radius a if the
upper half of the sphere is charged to a potential of qi and the lower half of
the sphere is charged to a potential 2.
10.2.20. Find the potential in the interior and exterior of a sphere of radius 2 when
one-half of the surface of the sphere is charged to potential o = 0 and the
other half of the sphere has potential of 0.

10.3 TRANSVERSE VIBRATIONS OF A THIN BEAM

In this section, we investigate transverse vibrations of a thin beam. We start with
the derivation of the equation. Then we solve the problem of transverse vibrations
of a thin beam, which is simply supported.

10.3.1 Derivation of the Beam Equation

u(x, t)

Figure 10.5: An elastic beam.

We derived the wave equation for a tightly stretched perfectly flexible horizontal
string in Chapter 3. Those equations also allowed us to model longitudinal vibra-
tions of a uniform flexible rod. (See Problem 4.4.9.) However, transverse vibrations
of a thin beam require a new set of equations since, unlike transverse vibrations of a
tightly stretched perfectly flexible horizontal string, the thin beam offers resistance
to bending.
Consider an elastic beam with uniform rectangular cross section, which lies with
the x-axis centered down the beam of length L, as shown in Figure (10.5).
 Section 10.3: Transverse Vibrations of a Thin Beam 365

Given a variable load, w(x, t), is on top of the beam, which produces a small
downward deflection of the beam, we have for any small section (x, x + Ox) of the
beam a bending moment, M (x, t). The bending moment is usually given as
M(x, t) = -E(x)I(x)C, (10.41)

where E(x)I(x) is the flexural rigidity and is composed of the Young's modulus
of elasticity, E(x), which depends on the material and the moment of inertia,
1(x). Also, the curvature, C, of the beam under a variable load w(x, t) per unit
length x can be found from calculus and is given as
D2u(x, t)
C= axe 3 . (10.42)
2 2
1 (Du(xt)
+ Dx

When we assume a small slope in the curvature, which is the usual assumption, we
have
Du(x, t) 0
ti
Dx

Thus, Equation (10.42) is approximated as

_ D2u(x, t)
C Dx2

and Equation (10.41) may be given as

a2u( 2, t
M(x, t) _ -E(x)r(x) (10.43)
8x

Newton's law of motion, applied to a small section of the beam, Ox, is given as
the sum of the forces in the u direction equal to mass of the beam times acceleration
of the beam in the u direction,
2
(x, t)
F =may = ma t2't) where m =
9

and the sum of the moments of bending equal to the moments of inertia times the
angular acceleration,

M(x,t) = 1w.
We assume the angular acceleration, w, to be 0. We make this assumption be-
cause we ignore angular acceleration just as we ignored horizontal vibrations in the
derivation of the wave equation for a vibrating string. (See Chapter 3.) Thus,

M(x,t) = 0.
366 Chapter 10: Classical Problems of PDEs

M MX+&

SX+&

Figure 10.6: A small section (x, x + Ox) of an elastic beam.

We also have shear forces acting on the cross section of the small section of the
beam, Ox, shown in Figure (10.6). We denote shear forces as S(x, t). The sum of
the forces in the u-direction under a continuous load z7(x, t) on the small section of
beam, Ox is
DS(x, t)
F= S x t- (S(x t
Dx
tv x t Ox = 0. (10.44)

Dividing Equation (10.44) by Ox and taking the limit as Ox -+ 0 yields

DS(x, t) D2u(x, t)
x + Z7(x, t) = m 9t2 (10.45)
D

If we take the sum of the moments counterclockwise about the point x, we arrive
at

0= M(x, t) _ -M(x, t) + (M(x, t) + a at) ox)

(s(x t) + as(x, t) ox) ox + tv(x, t)ox °x . (10.46)

Again, dividing by Lx and taking the limit as Ox -* 0 yields

DM(x, t)
=Sxt . (10.47)

Replacing M (x, t) in Equation (10.47) with its equivalent value found in Equation
(10.43) yields

(-E(x)I(x) a2 s(x, t). (10.48)

ax ax2'
 Section 10.3: Transverse Vibrations of a Thin Beam 367

Taking the derivative of Equation (10.48) with respect to x and adding uv (x, t) to
both sides gives us
a2 _ a2u(x, t) DS(x, t) a2u(x, t)
1 x2
E(x)I(x)
1x2
)+zz7(xt)= ax + z7(x, t) = m 92
which becomes
82u(x, t) 82 D2u(x,
CE(x)r(x) -tz7(x't). (10.49)
m(x) 8t2 + 8x2
If the beam is uniform, then E(x) and 1(x) are constant. Also, if the load uv(x, t)
is constant and does not vary with time, t, Equation (10.49) becomes
D2u(x, t) D4u(x, t)
+ c2 ax4 9, (10.50)
ate

where c2 = EI. Since the gravity, 9, is small compared with the internal forces
m
of the beam, it may be neglected in most applications. Thus, Equation (10.50)
becomes
D2u(x, t) D4u(x, t)
+2
which is the homogeneous equation for transverse vibrations in a beam.

10.3.2 Transverse Vibrations of a Simply

Supported Thin Beam
"Simply supported" means that the ends are fixed to support brackets, which allows
the slope of the endpoints to change. Figure (10.7) shows a simply supported beam.
Since riveted connections are elastic, a good example of a simply supported beam is
a girder made either of wood or steel in the construction of a building or a bridge.

Figure 10.7: Simply supported beam.

We know that the mathematical equation for transverse vibrations of a simply

supported thin beam is
a ua 2, t)
+ 2 a4a 4't) = 0. (10.51)
368 Chapter 10: Classical Problems of PDEs

Simply supported means that the endpoints are subject to

u(0, t) = 0, u(L, t) = 0,

82u(0, t) 82u(L, t)
=0, and =0,
8x2 x2
and
0)
u(x, 0) = f (x) and au(t' = g(x). (10.53)

Equations (10.51 and 10.52) are linear and homogeneous. Therefore separation
of variables applies. If we let u(x, t) = p(x)G(t) and replace it in Equation (10.51)
we find that Equation (10.51) becomes
cp(x)G"(t) + c2cpiv (x)G(t) = 0. (10.54)

Separating Equation (10.54) and setting it equal to a separation constant in the

usual manner yields
G"(t) - IV(X)
-
c2G(t) cp(x)

which becomes the time equation

G"(t) = -Ac2G(t) (10.55)

and the spatial equation

iv (x) _
Separating the boundary conditions yields
(0) = 0, cp(L) = 0,
(10.57)
cp" (0) = 0, and "(L) = 0.
Equation (10.56) subject to the BCs given in Equation (10.57) presents a slightly
different problem than the equations we got from separation of variables in Chapter
5. Since Equation (10.56) is a fourth-order ODE, we should expect four linearly
independent solutions. Also, if A is negative the time equation will force the de-
struction of the thin beam because the time equation will produce hyperbolic sine
and cosine functions. If A = 0, then the spatial equation only has a trivial solution.
Thus, if we assume that A> 0 and (x) = eT is a solution of Equation (10.56), we
get the equation
r4e= A& X.
Canceling like terms yields the characteristic equation
r4-A=0,
 Section 10.3: Transverse Vibrations of a Thin Beam 369

which, when factored, becomes

(r (A))(r + (A))(r - i (A) 4) (r + i (A) 4) =0.

Therefore, we have four roots, r = ±1, ±i. From Chapter 5 or Appendix C, you
should recognize that the solution is

x = cie(A) + t + C3ez(A) 4 + c4e-i(A) 4 x

. (10.58)

An easier form of Equation (10.58) is found by using the noncomplex and complex
forms of Euler's equations found in Chapter 5. Thus Equation (10.58) becomes

(x) = C1 cosh(A) 4 x + C2 sinh(A) 4 x + C3 cos(A) 4 x + C4 sin(A) 4 x. (10.59)

Applying the first BC of Equation (10.57), (0) = 0, yields

Cl = -C3.
Applying the third BC of Equation (10.57), cp"(0) = 0, yields

C1 = C3.

Thus, C1 = C3 = 0, and Equation (10.59) becomes

(x) = C2 sinh(A) 4 x + C4 sin(A) 4 x. (10.60)

Applying the second and fourth BCs of Equation (10.57), cp(L) = 0 and cp"(L) = 0,
yields

0 = C2 sinh(A) 4 L + C4 sin(k) 4 L (10.61)

and

0 = C2 sinh(A) 4 L - C4 sin(A) 4 L. (10.62)

Adding Equations (10.61 and 10.62) together yields

0 = 2C2 sinh(A) 4 L,

which implies C2 = 0 because the hyperbolic sine function equals 0 only at x = 0.

Therefore, we are left with

0 = C4 sin(A) 4 L and 0 = -C4 sin(A) 4 L,

which both yield the same eigenvalue

nor 4
n= L
, n = 1,2,3,..., (10.63)
370 Chapter 10: Classical Problems of PDEs

and eigenfunctions
n7rx
cpn (x) =sin L (10.64)

The Time equation, Equation (10.55), becomes

n7r 4
Gn (t) - (i;-) cG(t),
which has the solution

Gn(t) = D1 cosL[() 2 t + D2 sin L[() 2 t (10.65)

The general solution for u(x, t) becomes

u(x, t) = Al cos __ 2
t] + A2 sin
2
'n /nffS\
n=1

Applying the first IC from Equation (10.53) to Equation (10.66) yields

00

u(x, 0) = f (x) _ Al sin (nLx) .

n=1

Using the orthogonality of the sine function, we arrive at

fG
Al = L f (x) sin (nLx) dx.
J0
Applying the second IC from Equation (10.53) to Equation (10.66) yields

8u(x, 0) _ x_ I n 12 AZ sin r n7rx 1 .

8t g() - L J \ L l

Again using the orthogonality of the sine function, we arrive at

G
A2 = 2L 2
g(x)sin (nLx) dx. (10.68)
(/n7r) o

Therefore, the complete answer for transverse vibrations in a simply supported thin
beam is given by Equations (10.66, 10.67, and 10.68).
Other possible boundary conditions are

1. The cantilever beam, Figure (10.8), which has one end rigidly fixed and the
other as a free end. An example of this model is an airplane wing. It has the
 Section 10.3: Transverse Vibrations of a Thin Beam 371

Figure 10.8: The cantilever beam: rigidly fixed on the left end and free on the
right end.

following BCs:

( u(0,t) =0
Rigidly fixed at x = 0 end
8u(0, t)
l 0

0
Free end at x = L end
0.

Figure 10.9: Beam rigidly fixed at both ends.

2. Beam rigidly fixed at both ends, Figure (10.9). A very good example of this
type of beam is the World War II fighter, Lockheed P-38 Lightning. The P-38
was designed as a twin engine fighter where the engines were attached to long
booms, the wings extend from the outside part of the boom, and between the
two booms was the central nacelle, which contained the pilot and armament.
 372 Chapter 10: Classical Problems of PDEs

Such a rigidly fixed beam has the following BCs:

u(0, t) =o
Rigidly fixed at x = 0 end
{ 8u(0, t)
0
Ox

u(0, t) = o
Rigidly fixed at x = L end
Ou(0,t)
0 .

m
x
A 'r

EXERCISES 10.3

10.3.1. Solve the simply supported beam problem of length 27r when u(x, 0) = x(27r -
au(x, 0)
x) and =0.
at
10.3.2. Solve the simply supported beam problem of length 7r when u (x, 0) = x and
au(x, 0) _ 0
at

10.3.3. Solve the simply supported beam problem of length 1 when u(x, 0) = 0 and
au(x, 0)
at
-1-x.
-1-x2
10.3.4. State the BCs for a beam which is simply supported on one end and rigidly
fixed on the other end. Also, try to determine a physical application for these
BCs.

10.3.5. Given a cantilever beam where the right end is free, we know that after sep-
arating variables the spatial equation is

(x) = Cl cosh( ) 4 x + C2 4 x + C3 cos(h) 4 x + C4 sin(k) 4 x. (10.69)

Show that Equation (10.69) satisfies the cantilever beam BCs if (\) 4 L is a
1 1

root of cosh (,\) 4 L cos (,\) 4 L = -1.

10.3.6. Given a beam that is rigidly fixed at both ends, we know that after separating
variables the spatial equation is Equation (10.69). Show that Equation (10.69)
satisfies rigidly fixed beam BCs if (,\) 4 L is a root of cosh 7) 4 L cos 7) 4 L = 1.
 Section 10.4: Heat Conduction in a Circular Plate 373

10.4 HEAT CONDUCTION IN A CIRCULAR PLATE

Heat conduction in a circular plate is really no different from heat conduction in a
rectangular plate, which has the mathematical equation
au(x, y, t)
_ kv2u(x, y, t)
at

_ (O2u(xYt) +
02u(x, y, t)
(10.70)
axe 82
y
subject to the homogeneous Dirichlet BCs
u(0, y, t) = u(L, y, t) = u(x, 0, t) = u(x, H, t) = 0
and
u(x, y, t) = f (x, y).
If you notice, I used the Laplacian (V2), then wrote out the partial derivatives with

Figure 10.10: Polar coordinate system.

respect to x and y. This makes the transition from a rectangular plate to a circular
plate transparent. The only thing that changes is the coordinate system. In a
circular plate, the preferred coordinate system is the cylindrical coordinate system,
which is called polar form in two dimensions, Figure (10.10).Thus, Equation (10.70)
becomes
8u(r, B, t)
_ k02u(r, B, t)
at

k
rl 8 (rO?(1',O,t)1 + 1 82u(r,B,t)1
(10.71)
Lr 8r or J r2 802 '
374 Chapter 10: Classical Problems of PDEs

subject to
u(a, B, t) = 0

and

u(r, 0,0) = g(r, 0). (10.73)

Just as in Section 10.2.1, it appears that we only have one boundary condition.
However, as in Section 10.2.1 the other BCs come from the geometry of the problem,
and they are
u(r, -ir, t) = u(r, ir, t)

and (10.74)

8u(r, -ir, t) 8u(r, ir, t)

00 80
and

li m u(r, B, t) < oo. (10.75)

Equation (10.75) is not so much a BC as it is a natural condition on temperature at

the origin of the plate. If the temperature were allowed to go to infinity the plate
would melt down. An unwanted condition in most experiments.
Equation (10.71) and its corresponding BCs are linear and homogeneous. There-
fore, we may use separation of variables. We let u(r, 0, t) = G(t)Sp(r, 0). After taking
the necessary derivatives and separating Equation (10.71) and the BCs, Equation
(10.72, 10.74, and 10.75), we arrive at the time ODE
G'(t) _ -\kG(t), (10.76)

and the spatial PDE

e) 1 0zae2'
_ e), (10.77)
r 8r )+ 2 e)
subject to
cp(a, 0, t) = 0,

limn-,o Ip(r,9) 00

p(r, -ir) = p(r, ir),

and

8cp(r, -ir) 8cp(r, -ir)

00 00
 Section 10.4: Heat Conduction in a Circular Plate 375

Equation (10.77) is a linear homogeneous PDE with linear homogeneous BCs,

Equation (10.78). Thus, we can use separation of variables. Letting (p(r, e) =
R(r)O(9), taking the necessary derivatives, and separating the equations, we have
o"(e) = -,ro(e), (10.79)

subject to
= e(7r)

and

and after some algebraic manipulation,

r2R"(r) + rR'(r) + r2 - µ)R(r) = 0, (10.81)

subject to
R(a) = 0

and (10.82)

limn->o R(r) I < 0.

We have encountered Equation (10.79) subject to BCs, Equation (10.80), and

know its solution, which is
µo = 0, 40(9) = Ao

(10.83)
m = 1,2,3,...
Om(e) = Am cos me + Bm sin me
We must solve Equation (10.81), restated as
r2R(r) + rR(r) + (\r2 - m2)R,,, (r) = 0, (10.84)

subject to
R,,,, (a) = 0

and (10.85)

I
< 0.

First, we should recognize Equation (10.84) is a form of Bessel's equation, Equation

(9.51),

x2u"+xu'+ (x2 -n2)u=0,

376 Chapter 10: Classical Problems of PDEs

found in Chapter 8, and we know the general solution, which, for Equation (9.51),
is

u(x) = CiY(x) + C2J(x).

Thus, the general solution to Equation (10.84) is

Rm,(r) = C1Y,,,, (/r) +C2J-,,, (Vr) .

Since we know that limo Rm (r) < 0 and as r * 0, Ym (V5r) 00, we must
have C1 = 0. Therefore, we are left with

Rm(r) = C2Jm (Vr). (10.86)

Applying the second BC, Rm (a) = 0 to Equation (10.86) yields

Rm (a) = C2Jm (Va).

For each Bessel function of the first kind of order m, we have an infinite number of
Os. Hence, just like a sine or cosine function, we have an infinite number of possible
solutions. Since each Bessel function of the first kind of each order m is a solution to
a linear homogeneous ODE, we know the sum of all the solutions is also a solution.
Thus, letting ymn represent the Os of Bessel function of the first kind of each order
m, we have

which implies

Amn

Thus, the general solution to Equation (10.86) is

00

Rm(r) = CnJm (/r)

n=1

where
fo R,,,(r)Jm (/Xr) r dr
Cn -
fo Jm 2 r dr

Therefore, the general solution of the PDE given in Equation (10.77) is

00 00

p(r, 0) = Jm (/5r) [amn cos m9 + /3mn sin m9], (10.87)

m=0 n=1
 Section 10.4: Heat Conduction in a Circular Plate 377

where amn = cn Am and /3mn = Cn Bm

We must now solve the time ODE, Equation (10.76),

Gmn (t) - \mrtGmrt (t)

which is

Gmn(t) = drone -Amnkt

We now can give the general solution for u(r, 0, t), which is

u(r, 0, t) _ e-Amnkt Jm (,/r) [Dmn cos m0 + Emn sin m0] (10.89)

m=0 n=1

where Dmn = dmn amn and Emn = dmn l3mn

Applying the IC, Equation (10.73),
u(r, B, 0) = g(r, B)

yields

u(r, 0, 0) = g(r, 0) _ >Jm (/r) [Dmncosm0+Emnsinm0l

m=0 n=1
(10.90)

JO (v'5r) D0n + Jm r Dmn cos m0

n=1 m=1 n=1
(10.91)

+ Jm (/r)mn
Emn sin m0. (10.92)
m=1 n=1

Thus,

Don _ ffo g(r, B)Jo (\/Xr) r dr dB

ff[Jo(r)] rdrdO 2
(10.93)

f " f ao 9(r, B)J.,,,, ('/Xr) cos (mO) r dr dB

Dmn = > 0; (10.94)
f fo a [Jm (/Xmnr) cos (mO)]2 r dr dB

and
f ", f o g(r, ('/Xr) sin (mO) r dr dB
E,,,, _ , m > 0. (10.95)
f fo [J,,,, (\/Ar) sin (mO)]2 r dr dB
 378 Chapter 10: Classical Problems of PDEs

Therefore, the solution to the circular heat problem is Equation (10.89) where
the coefficients are given by Equations (10.93, 10.94, and 10.95).
A similar method is used to solve the wave equation for vibrations of a drumhead.

EXERCISES 10.4

10.4.1. Consider temperature in a circular plate of radius r = a where the initial

temperature of the boundary of the plate is a homogeneous Dirichlet condition.
Given a non0 initial temperature distribution, briefly explain what occurs
when we take u(r, 0, t).
10.4.2. Solve for temperature distribution in a semicircle plate of radius 1 where
0 < 0 < it with initial condition of u(r, 0,0) = g(r, 0) and the following BCs:
(1) u(r, 0, t) = 0, u(r, ir, t) = 0, u(1, 0, t) = 0.

2 au(r, 0, t) _ 0 au(r, 7r, t) _ 0 au(1, 0, t) _ 0

() a0 a0 a0

au(r, 0, t) Ou(r, t)
(3)
ae
-0
- ae
,
- 0 u(1, 0, t) = 0.

For each case briefly explain what occurs when we take u(r, 0, t).

10.4.3. State and solve the general vibrating circular drumhead problem.
10.4.4. Solve the vibrating circular drumhead problem when the radius r = 1, the
boundary is fixed, the initial velocity is 0, and the initial displacement is
f (r) sin 20.

10.4.5. Solve the vibrating circular drumhead problem when the radius r = 7r, the
boundary is free, the initial velocity is 0, and the initial displacement is
f (r) cos 0.

10.4.6. Consider vertical vibrations of a wedge 0 < 0 < with radius 2. Determine
4
the solution if the BCs are as follows:
(1) u(r, 0, t) = 0, u (r, , t = 0, u(2, 0, t) = 0.
4

au(r' au r, 7r ,t= au(2'

(2) 0, t) -0 0 0, t) - 0.
80 ' a0 ' 80
 Section 10.4: Heat Conduction in a Circular Plate 379

(3) 8u(r, 0, t) - 0 8u (r, 4 , t)

(2,0,t) = 0.
ae ae
For all cases, assume the initial velocity is 0 and the initial displacement is a
function of the radius and the angle.
10.4.7. State and solve temperature distribution in a circular plate where there is
a2u
circular symmetry, i.e., = 0.
a02
10.4.8. Consider
au(r,0,z,t
= kV2u(r, 0, z, t),
at
subject to
u(r, B, z, 0) = f (r, B, z)

and the following BCs:

(1) u(r, 0, 0, t) = 0, u(r, 0,1, t) = 0, u(2, 0, z, t) = 0.

(2) u r 0 0 t= 0 au(r,0,2,t) =0 u 0.5 0 z t= 0.

au(r,0,0,t) _ 0 au(r,0,3,t)
()3 ,0,z,t)=0.
az az
10.4.9. Consider
V2u=0
in a cylinder. Find the solution with the following BCs:
(1) u(r, 0, 0) = 0, u(r, 0, 5) = r sin 20, u(1, 0, z) = 0.

8u(r, 0,0) _ 0 8u(r, B, 2)

(2) = 0, u(ir B' z) = 0.
8z 8z

8u( B, 0)
(3) _ 0, u(r, B, r) = r cos B, u(1, B, z) = 0.

=r202au(r,0,2)=0u0.50z=0.
(4) 0,0)
Oz
Does the solution always exist?
 380 Chapter 10: Classical Problems of PDEs

10.4.10. Complete the following:

(1) Solve for temperature distribution inside a uniform sphere of radius a,
which is independent of the variable ( if the initial temperature distrib-
ution inside the sphere is constant throughout the sphere and the BC is
a homogeneous Dirichlet condition.
(2) Solve the problem in Part (1) if the radius a = 5 feet.
(3) Show that as time, t, goes to infinity and that the solution to Part (1)
becomes the solution of Laplace's equation in the sphere of radius a with
homogeneous Dirichlet condition.
10.4.11. A uniform hemisphere of radius 1 m has its convex surface temperature kept
at 100°C while its base is kept at a temperature of 0°C.
(1) Find the temperature distribution inside the hemisphere for all time.
(2) Find the steady-state temperature inside the hemisphere.

10.5 SCHRODINGER'S EQUATION

Schrodinger's equation is very important in the study of quantum mechanics, and
most of you have heard of it. In this section, we develop the time-independent
Schrodinger Equation. Also, in the exercises, several projects are presented that
involve nuclear transport. We start our derivation with the wave equation in three-
dimensional Cartesian space
a2 u
= k2V2u. (10.96)
ate
We are using the constant k2 instead of the constant c2 since we want to reserve
c2 for the speed of light. Helmholtzl formulated solutions to Equation (10.96) by
using separation of variables, where the separation constant is -2. Helmholtz's
solution is in the form
u = Spe-it, (10.97)

where = 27rv and v is the frequency of the radiation. The derivation of = 27rv
is far beyond the scope of this text. However, it is easy to show that G(t) = e-it
is a solution of
G"(t) = -2G(t).
The function ( is the solution of the time-independent equation
02 = -s2 c= . (10.98)

'Herman Helmholtz (1821-1894) started his professional life in physiology, then became inter
ested in mathematical physics. His studies were primarily in acoustics and he published results in
a work titled On the Sensations of Tone.
 Section 10.5: Schrodinger's equation 381

Since we are interested in the time-independent Schrodinger Equation, we want to

focus on Equation (10.98), which can be written as
V2p + S2p =0. (10.99)

Equation (10.99) is the time-independent wave equation, which is sometimes called

the Helmholtz equation. The solutions of Equation (10.99) are called monochro-
matic waves. From Helmholtz's equation, we want to develop the time-independent
Schrodinger Equation. We start by considering Einstein's2 famous equation

E=mc2. (10.100)

Einstein's equation relates energy to mass. From it we see that

E
m= . (10.101)
c2

The photon is a quantity (quantum) of electromagnetic radiation. We know it has

speed and energy where the energy is given by

E = hi', (10.102)

where h is Planck's3 constant and v is the frequency of the radiation. This implies
that
e-it = e-2lrivt =
e

which implies that the eigenvalues depend on the total energy, E. This is shown
later to be exactly the case.
Replacing energy, E, in Equation (10.101) with the formula for energy given in
Equation (10.102), we find that

(10.103)

Since a photon has speed and energy, it has momentum. Momentum for a single
particle of mass (a photon) is given by the equation

p = mv, (10.104)

where v is the velocity (usually expressed as a vector). Traditionally, the velocity

of light is given as c. Thus, the momentum of a photon is

p = Inc. (10.105)

2Albert Einstein (1879-1955), a brilliant physicist, developed the theory of general relativity,
which brought the field of differential geometry back to center stage in the mathematical world.
3 Max Planck (1858-1947) a German physicist who in 1901 took the first steps toward quantum
mechanics.
382 Chapter 10: Classical Problems of PDEs

Using Equation (10.103), we find momentum for a photon is given by

_ by by
p c2 c= . (10.106)
c

In the mid 1920s, de Broglie's hypothesis was presented. It suggested that waves
were associated with material particles, with wavelength, A, and momentum, p,
where
A-h_ h _c (10.107)
p Inc v

Thus,
by h
p (10.108)
c A'

which can be written as

by h h 27r
p (10.109)
c A 2ir A
h
Note: is called the Dirac-h denoted h. However, we continue the derivation
2ir '
h
usin g . Since k is the velocity of the wave, which is the velocity of light and
therefore c, we have
2irv 2ir

2irmc
Using Equations (10.105 and 10.107) we have c = Replacing s in Equation
h
2irmv
(10.99) by yields
h

V2 + (2 c1J 2 _ +
0. (10.110)

After rewriting Equation (10.110) slightly, it becomes

22(P=0. (10.111)

From physics, we know total that the energy, E, is equal to the sum of the
kinetic energy, Ek, and the potential energy, E. Also, we know the kinetic energy
mc2
is equal to 2 . Thus,

mc2
E=Ek+Ep= 2 +Ep,
 Section 10.5: Schrodinger's equation 383

which means
mc2
2 =E-E. (10.112)

Note: In Equation (10.112), the potential energy, is a function of position.

mv2
Using Equation (10.112), we may replace in Equation (10.111) with E - E.
2
Thus, we have
8ir2m
h2
(E-E)p=0, (10.113)

which is the time-independent Schrodinger's equation. In one space dimension,

Equation (10.113) may be written as
2
cp" (x)+ 8 h m (E-E(x))cp(x)=0. (10.114)

Equation (10.114) allows us to employ our knowledge of Sturm-Liouville theory.

We can rewrite Equation (10.114) as

SP"(x) + q(x)So(x) + ASo(x) = 0,

8ir2 m 8ir2 m
where q (x) _ - h2 E( x) and A = h2 E. Thus, the eigenvalues are based on
the total energy. Also, each eigenvalue, En, n = 1,2,3,..., has a corresponding
eigenfunction, Spn, n = 1,2,3,...

EXERCISES 10.5

10.5.1. PROJECT: Apply Schrodinger's equation to a harmonic oscillator-quantum

harmonic oscillator. Consider a physical body of mass in attached to the
bottom of a spring, and the top of the spring is attached to some immovable
object, such as a ceiling. We have no external force or damping. Let u
represent the displacement of the mass, in, from equilibrium or the at-rest
position; then the classical differential equation is given by
mu + kv. = 0, (10.115)

where k is the spring constant and in is the mass. Note: ku represents the
restoring force. Dividing through by in in Equation (10.115) yields
u+wou=0
where wo 2 = -' which implies k = mwo2.
m
(1) Solve the ODE.
 384 Chapter 10: Classical Problems of PDEs

(2) Derive the potential, Ep, by integrating the restoring force, ku. Note:
We must choose the constant of integration so that Ep = 0 when the
spring-mass system is at equilibrium, u = 0.
(3) Replace Ep in Equation (10.114).
(4) Multiply Equation (10.114) by h
2irmcao

(5) Set A _ Oh and obtain the equation

h +ASP- 2irmwo
u2
2irmwo SP
SP=0.

(6) Introduce a new independent variable =

u
2lrmwo
and let cp(u) _
cp(uv). Then find that

+ (A - cp = 0. (10.116)

(7) We now look for solutions to Equation (10.116) on (-oo, oo) by substi-
tuting e 2 2 v(x) into Equation (10.116) and dividing out e 2 2 .
(8) Now let A = 2n + 1 and our equation becomes identical to Hermite's
equation given in Chapter 10.
(9) Solve and graph several Hermite polynomials on the interval (-5, 5).
10.5.2. PROJECT: Consider nuclear transport
(1) The Euler differential equation is one of the few differential equations
with variable coefficients that can be solved using a change of variables.
The change of variables necessary to transform the Euler differential
equation into a linear constant-coefficient differential equation is x =
ez . With this transformation, find the complete solution to the following
equations:
(a) x2 y" - xy' + y = x5.
(b) x2 y" - xy' + 2y = 1 + (lnx)2.
(2) You are working in the joint counter-terrorism unit with other federal
agencies. While searching for the plutonium brick (which was found
using information you supplied about the maximum size of the brick),
airport security forces discovered that the same group of terrorists were
trying to smuggle a 6.0 cm radius solid sphere of Pu-239 into the U.S.
disguised as a child's toy ball. The FBI wants to ship the ball back to a
Department of Energy (DOE) lab for evaluation. Since there is a Coast
Guard cutter docked at a nearby marina, they plan to send the Pu ball
packed in a crate on board ship. An FBI agent goes and gets a big crate
 Section 10.5: Schrodinger's equation 385

full of small foam packing peanuts. (In the center of which he/she plans
to place the Pu sphere for shipping.)
The steady-state neutron diffusion equation with neutron multiplication
is:
v.7f SP
D V2 ( - a Sp + =0,
where

V2 p = 1 (r2 assuming spherical symmetry.

r2 oar oar

The following nuclear data may be useful:

For the Plutonium Ball: D = 1.263 cm, a = 0.0819 cm-1, vcp f = 0.214
cm-1 and density p = 15.4 g/cc.
For the Foam Packing Peanuts: D 2.1 cm, 0a = 10-6 cm-1, vcp f = 0
cm-1, and density p = 1.0 g/cc.
Notes: The steady-state neutron diffusion (Helmholtz) equation repre-
sents a balance of leakage, absorption, and production from fission re-
actions in the sphere using the fundamental mode eigenvalue. When
we apply the "bare sphere" boundary condition that Sp (R + 2D) = 0 at
the radius of the sphere (R) plus the nuclear "extrapolation" dimension
(2D) to make the physics correct, we are essentially assuming that no
neutrons are reflected back into the sphere. In fact, because atmospheric
air reflects back a few neutrons (r-' 0) that escape the sphere surface, this
is a good assumption; therefore, we need not consider solving the diffu-
sion equation in the air immediately surrounding the sphere. However, if
we replace the material surrounding the Pu sphere with a hydrocarbon
(foam packing peanuts), we could get some very nice reflection of neu-
trons back into the sphere, and we will have to account for the diffusion
of neutrons in the region outside of the sphere.
The multiplication factor (k) of the 6.0 cm radius unreflected ("bare")
sphere in open air is k = 0.8446. This means that in this case, the
steady-state loss rate of neutrons by absorption and leakage from the
sphere is greater than the rate neutrons are produced inside the sphere
from fission. If we add to the Pu, or change the geometry to ramp up
the production from fission reactions to make k = 1.0, then a steady
state balance of production from fission and loss due to absorption and
leakage occur. (Notice how the "k" in the steady-state diffusion equa-
tion "adjusts" the production term to satisfy the overall equation. When
k exactly equals 1.0, this is known as "criticality," and the mass caus-
ing this condition is known as "critical mass" not a good thing to be
standing next to, since radiation coming directly out of a critical mass is
lethal.)
(a) What is the critical size (when k = 1.0) of a "bare sphere?"
(b) Determine the critical size (k = 1.0) for a Pu sphere reflected by
an infinite thickness of packing peanuts. (Use (1(R) = Sp2 (R), and
 386 Chapter 10: Classical Problems of PDEs

Jl (R) = J2 (R) (where J = -DV p r) at the interface of the ball

and the packing peanuts.)
(c) Based on your answer, should the FBI agent pack the 6.0 cm radius
sphere in the crate as planned, and if he/she does, is it safe to stand
nearby when he/she packs it? Is using an infinite thickness of packing
peanuts a conservative approach? Why or why not? Explain.
(d) Demonstrate that k = 0.8446 for the 6.0 cm radius "bare sphere."

10.5.3. PROJECT: It is July 1945, and the future is uncertain. You are designing
the Little Boy atomic weapon. Little Boy is a "gun-type" weapon that uses
two identical subcritical cylindrical halves. One cylinder of U-235 remains
stationary at one end of the device, while the other (identical) subcritical
cylinder is fired using a chemical explosive as a projectile down a gun barrel
directly into the stationary cylinder. Provided other components are installed
correctly in the device, a nuclear yield results. Using the time-dependent
neutron diffusion equation with neutron multiplication you find that

1
a Sp Sp =
vo at

where

2 10(0(;o
r8r r8r+8z2
For boundary conditions, the origin can be taken to be the center of the
cylinder. The flux can be assumed to be 0 at the extrapolated boundaries,
and the initial (t = 0) flux profile is given as

Sp (ii, z, t) = 0 (R = R + 2D) cylinder extrapolated radius

p r +a t =0 a
± =+
a
+ 2D cylinder half height
2 ' 2 2

and

Sp(r, z, 0) = coo initial flux distribution.

Note that Sp > 0 everywhere in the cylinder. Also, the following nuclear data
 Section 10.5: Schrodinger's equation 387

for U-235 metal may be useful:

diffusion constant D = 1.3175 cm

mean neutron velocity Vo = 1.4 x 109 cm/s

cm-1
absorption probability 0a = 0.0722

fission neutron production probability vo f = 0,1687 cm-1

fission probability f = 0.0649 cm-1

metal density p = 18.0 g/cc

(1) Solve the time-dependent diffusion equation in cylindrical geometry using

separation of variables. (Hint: To solve this, begin separation of variables
with µ = -A for the time-separation constant).
(2) Identify the largest (just e = 1% below critical) mass of U-235 that
can be safely used for each half cylinder using diffusion theory. Assume
that each half cylinder is a square extrapolated cylinder, where twice the
extrapolated radius is equal to the total extrapolated height (2R = a).
(Recall that criticality is achieved when a stable fundamental mode flux
is obtained, (A1 -+ 0) that is, the fundamental mode flux does not change
with time, and other modes will die out with time).
(3) Assuming coo -p 100 n/cm2 /s, plot, using your favorite mathematical
software, the time dependence of neutron flux in the center of the assem-
bled device at Sp (0,0, t) right after triggering the detonator right when
the two cylindrical halves-each of a size you found in part (b)-meet to-
gether to form a single, supercritical cylinder. Note: Time should be in
the microsecond range.
(4) The heat power released in the supercritical assembly as a function of
time is determined by integrating the volumetric fission rate over the
cylindrical volume:

P(t) in watts = 3.204 x 10-11 (f)(co(r, z, t)) dV.

J
The total energy released, in Joules, is

Er = P(t) dt.

Use your favorite mathematical software to compute the energy released

after 1 µsec and 2.5 µsec using kiloton TNT equivalents, where 1 kiloton
TNT = 4.18 x 1012 Joules.
 388 Chapter 10: Classical Problems of PDEs

Figure 10.11: A fixed point in three-dimensional space

10.5.4. PROJECT: The time-independent Schrodinger's equation

8ir2m
h2
(E-E)p=0
may be developed in spherical coordinates if the potential Ep depends only
on a distance r from some fixed point in space. Given the spherical r, e, and
as shown in Figure (10.11) where

xl = r sin e cos , x2 = r sin a sin , and x3 = r cos e,

show that
1 2

V2
r a (r2 8+ r Sin e ae
(sin0
-) + sin 8 (10.117)

(1) Show that Equation (10.117) may also be written in the form

2 a2 (,O 2( 1 La2 SP
e
a
+ 1 a2 SP

ae2 + cot ae
2
v 'P ar2 + r ar + r2 sine a q52
(2) Using Schrodinger's equation in spherical coordinates, show that sepa-
rating variables results in

R (r) + 2R (r) + (E Ep) R(r)

r
R(r) 0, (10.118)
r h2
where A is the separation constant and

i a eaW(e,
c5)1 + i aw(e, ) + aW(e, q) = 0. (10.119)
Sin eae ` ae J sine B a2
 Section 10.6: The Telegrapher's equation 389

(3) Using Equation (10.119), applying separation of variables results in

T
o"(e)+ got eo'(B)+ (A_ 2 (B) = 0 (10.120)
sin

and

I:(q5) = 0. (10.121)

ti
(4) Show that I(q) = m = ±1, ±2, ±3, ±... is a solution of Equation
2
(10.121). Note: r = m What conclusions can you make about this
solution?
2
(5) Solve Equation (10.120) where r = m Also, set x = cos 0, O(8) _
y(x); only consider the case where m = 0 and A = n(n + 1), n =
0,1,2,3,... Graph the solution for several values of n. (Hint: After
replacing x = cos 0 and O(0) = y(x), your new equation should be
Legendre's equation; see Chapter 9.)

10.6 THE TELEGRAPHER'S EQUATION

The telegrapher's equation has a long history. It was first developed in 1855 by Lord
Kelvin, who studied the Atlantic telegraph cable, which was completed in 1858. It
is used today by modern physicists in the analysis of the time-dependent Boltzmann
equation in the theory of neutron transport. Another use is in mathematical phys-
iology. Here, the telegrapher's equation models electrical tranmission in a neuron.
Once, we develop the standard telegrapher's equation, we will develop the neuron
application.

10.6.1 Development of the Telegrapher's Equation

We develop the telegrapher's equation by considering the flow of electricity in a long
transmission line. A transmission line is imperfectly insulated so that there is both
capacitance and current leakage to the ground. The transmission line runs along
the x-axis, and the line will contain the same amount of resistance, inductance,
and capacitance for any small length, dx. The following is a list of the essential
components, quite similar to the list given in Chapter 3:

dx is the length of line from pl to p2.

v(x, t) and v(x + dx, t) are the potential or voltage at any point of the line.
i (x, t) and i (x + dx, t) are the current along the line.
R is the resistance per unit length.
L is the inductance per unit length.
390 Chapter 10: Classical Problems of PDEs

G is the conductance to ground per unit length of the line.

C is the capacitance to ground per unit length of line.

We will now develop the telegrapher's equation. The potential at p2 is equal to

the potential at pl minus the drop of the potential along the element pl p2 . Thus,
using Kirchoff's first law we have
ai (x, t)
v x - dx t vx t R dx i x t L dx ( 10 122 )
.

which may be written as

v ( x + d x, t) - v ( x, t ) _ -R dx i( x, t ) - L dx a2
(x,
t . ( 10 . 123 )

Dividing by dx and taking the limit as dx -p 0, yields

8v(x t) _ 8i(x t)
-R i(x , t) - L (10 124)
ax' at .

Similarly, the current at p2 is equal to the current at pl minus the current loss
through leakage to the ground and the current loss due to the capacitance. Thus,
according to Kirchoff's second law we have
8v(x, t)
i(x + dx,t) = i(x,t) - G dx v(x,t) - C dx at (10.125)

which may be written as

av(x,t)

i x + dx,t i x,t G dx v x,t C dx (10.126)

Dividing by dx and taking the limit as dx -p 0, yields

aZ(x, t) _ av (x,t)
-G v(x, t) - C (10.127)
8x at
If we differentiate Equation (10.124) with respect to t and multiply it by C, and
differentiate Equation (10.127) with respect to x, we have
a2v (x, t) ix, t) - Lc
C
atax - - Rc
RC
at
a22 (x, t)
ate
(10.128)
a2i(x, t) v(x, t) a2v(x, t)
8x2 - -G ax -C axat
If we subtract the first equation in Equation (10.128) from the second equation in
Equation (10.128) and assume the mixed partials are identically equal, we get

a2z(x,t)
ax2
- Lc a2i(x,t)
at2
+ RC ailx,t>
at
-G ax v(x,t)
(10.129)
 Section 10.6: The Telegrapher's equation 391

v(x, t)
In Equation (10.129) if we substitute Equation 10.124( ) for the resulting
equation is the telegrapher's equation for i (x, t), which is

52i(x, t) 52i(x, t) ai(x, t)

= LC ate + (RC + GL) + RGi(x, t). (10.130)
a x2 at
Similarly, we can find the telegrapher's equation for v(x, t), which is given in Chapter
3.
If we are working with submarine cables, then it is assumed that L = G = 0,
and Equation (10.130) becomes

522(x't) Rca2(x,t)

= (10.131)
8x2 rat

which is known as the telegrapher's equation for submarine cables.

10.6.2 Application of the Telegrapher's Equation

to a Neuron
So that we are all using the same terminology, the following is a short description
of a neuron and its functions. It is directly followed by the development of the
telegrapher's equation for a neuron.4
A neuron is composed of three parts: the cell body, called soma, which contains
the nucleus; the dendrites, which are numerous protoplasmic outgrowths from the
soma; and the axon, a single extension longer than the dendrites, which ends in a
brushlike filament.
The axon is covered with a fat-containing myelin sheath. The myelin sheath
may be looked at as insulation. The axon carries the output of the neuron. The
brush-like filaments at the end are called the synapse. The synapse are specialized
for the transmission of an electrical signal to other neurons. This transmission to
many other neurons is called divergence. The electrical transmission in the axon is
active.
The dendrites are the receivers of the neuron. A single neuron may receive input
from many other neurons, and this action is called convergence. Electrical current
spreading through the dendrites is passive. Thus, the telegrapher's equation may
be used to study the electrical current.
This is somewhat different than the normal telegrapher's equation because for
a neuron we must consider the intracellular space, the extracellular space, and
the cell membrane. In a neuron, the cell is viewed as a membrane surrounding
the cytoplasm. The cytoplasm is the transmission line (denoted line), which is
considered one-dimensional, and for our purposes the line runs along the x-axis.
Wilfrid Rall called this the "core conductance assumption." The line is separated
into small pieces of length dx. Along any length, all currents must balance.
4This subsection is adapted from James Keener and James Sneyd, Mathematical Physiology,
©1998 by Springer-Verlag, New York, pp. 251-256. Reprinted by permission.
392 Chapter 10: Classical Problems of PDEs

The cell membrane acts or takes on the roll of the capacitor, denoted C = q
v
where q is the charge across the capacitor and v is the voltage potential. Since
we have both intracellular and extracellular space, we must consider voltage and
resistance of both. For the intracellular space, the voltage and resistance are denoted
as v2 (x) and v2 (x + dx) and Ri dx. For the extracellular space, they are denoted
as ve (x) and ve (x + dx) and Re dx. Also, there are two currents to consider;
the transmembrane current denoted It dx, which is the current running across
the membrane, and the axial current denoted Ia, which has components in both
the intracellular and extracellular space, denoted Iai and Iae , respectively, and
considered linear functions of the voltage. There also exists an ionic current, denoted
II (Note: When a neutral atom loses or gains one or more electrons, the loss of
electrons results in a positively charged ion and the gain of electrons results in a
negatively charged electron-gains and losses may occur during chemical reactions),
which occurs on the cell membrane. The different functions are as follows:

dx-small length of line

vi (x, t) and v2 (x + dx, t)-intracellular space voltage along the small length dx
Ri-intracellular space resistance along the small length dx
Iai (x, t)-intracellular space axial current
ve (x, t) and ve (x + dx, t)-extracellular space voltage along the small length dx
Re-extracellular space resistance along the small length dx
Iae (x, t)-extracellular space axial current
It-transmembrane current
C-cell membrane, which acts like a capacitor
RCi -resistance of the cytoplasma in the intracellular space
Ai-cross-sectional area of the line in the intracellular space
RCe -resistance of the cytoplasma in the extracellular space
Ae-cross-sectional area of the line in the extracellular space, which may be
very large in comparison with Ai
II-ionic current on the cell membrane

We start with derivation with the axial currents. The equations are similar to
Equation (10.122), and they are

vi (x + dx) = vi (x) - Iai (x, t)Ridx

(10.132)
ve (x +dx) = ve (x)- Iae (x,t)Redx,
 Section 10.6: The Telegrapher's equation 393

which may be written as

vi (x + dx) - vi (x) = -Iai (x, t)Ridx

(10.133)
ve (x + dx) - ve (x) = -Iae (x, t)Redx.

Dividing the equations in Equation (10.133) by dx and taking the limit as dx - 0

yields

8vz(x, t)
-Iai (x, t)Ri
ax
(10.134)
8ve(x, t) -Iae (x, t)Re.
ax

Thus,

1 8vz (x, t)
Iai (x, t) = Ri ax
(10.135)
Z 5ve(x,t)
Iae (x, t)
Re ax

We use Kirchoff's second law to calculate the change in axial current due to trans-
membrane current. We must have

Iai (x, t) - Iai (x + dx, t) = Itdx = Iae (x + dx, t) - Iae (x, t). (10.136)

Dividing Equation (10.136) by dx and taking the limit as dx - 0 yields

- " Iai (x, t) = I = alae (x, t) (10.137)

ax ax

We also have that the total axial current IaT (x, t) = Iai (x, t) + Iae (x, t) if no other
sources of current exist in the line. Therefore, using the expressions for Iae and Iae
given in Equation (10.135), we have

I _
aT
- 1 avi (x, t) _ 1 ave (x, t)
Ri ax Re ax
or

-IaT _ 1 avi (x, t) 1 ave a(x x, t) .

(10.138)
Rz ax + Re
394 Chapter 10: Classical Problems of PDEs

1 avi (x, t) 1 5v2 (x, t)

Adding a factor of 1 = to Equation (10.139) yields
Re ax Re ax
_I _
aT
1 avz (x, t) 1 ave (x, t) 1 avz (x, t) - 1 avz (x, t)
Ri ax + Re ax + Re ax Re ax

_ 1 avz (x, t) 1 avz (x, t) 1 ave (x, t) 1 avz (x, t)

Ri Dx + Re Dx + Re ax Re Dx

1 1 5v(x,t)
(x, t) 1 (5Ve(X,t) 5v(x,t)
(x, t)
Ri + Re 8x + Re 8x 8x

C Rt + Re 1 8v2 (x, t) _ 1 (5v(x,t) _ 8ve (x, t)1

(10.139)
R;,RQ J 8x Re 8x 8x
Also, we know that the total voltage is given by VT (X, t) = v2 (x, t) - ve (x, t). There-
fore, Equation (10.139) may be written as

_IaT _ (Ri+Re'\ 5vi (x, t) _ 1 5VT (x, t)

( 10.140 )
Ri Re 5x Re
Equation (10.141) may be rewritten as
-ReIaT 1 Dvi (x, t) 1 DVT (x, t)
Ri + Re Ri CIx Ri + Re 5x
or

(x, t) _
1 5v(x,t) 1 5VT (x, t) - Re I aT. (10.141)
Ri ax Ri + Re ax Ri + Re
1 5v( x, t)
Since IaZ (x, t) _ - Equation (10.141) becomes
Ri ax
_ai
I ( xt) = 1 (x
A , T ax t _ RP.
RZ + Re
)
Rz + Re
IaT . (10.142)

If we assume that in Equation (10.142) IaT is constant, which is a reasonable as-

sumption, and use Equation (10.137), we arrive at
8vT(x, t)
1
(10.143)
R; + Re 8x
The transmembrane current, It, is equal to the capacitive current and the ionic
(x t)
CavTat,
current where the capacitive current is given by and the ionic current
is given by II. Thus, Equation (10.143) may be written as
avT fix, t> a avT (x, t)
jt - p C ax + I` - ax (_1
R, + Re ax
 Section 10.6: The Telegrapher's equation 395

which becomes
(c') II +
_ a 1

Z + Re
avT (x, t)
(10.144)

Note: Equation (10.144) is very similar to the telegraph equation for submarine
cables, Equation (10.131). Thus, Equation (10.144) is the telegraph or line equation
for a neuron.
If we assume that RZ and Re are constants, then Equation (10.144) becomes
5VT(x, t) _
II_p(RZ+Re)
1 a2VT(x, t)
(10.145)
C at + axe
Equation (10.145) may be normalized by taking into account certain constraints on
the membrane resistivity. Ignoring the extracellular resistance, and then nondimen-
sionalizing the resulting equation. Thus, we get a line equation for a neuron, which
is somewhat easier to work with; it is
C7v
a2 Z1
at - axe +f (v, t), (10.146)

where f (v, t) is a function of voltage and time. (Note: This is in the true tradition
of mathematical modeling. If you try to keep everything you sometimes end up
with an unworkable model, or a model that predicts nothing or the wrong answer
quite accurately. It follows the old military axiom: He who protects everything,
protects nothing at all.)
Many times f (v, t) can be modeled as -v when we are working with passive
activity, such as dendrites. For many other cells, the activity is called passive only
if the membrane potential is sufficiently small. Thus, Equation (10.146) becomes
C7v
a2 Z1
at - ax e - v. (10.147)

This concludes our development of the line equation for a neuron.

EXERCISES 10.6

10.6.1. Consider the telegrapher's equation

a2u(x, t) __ a2u(x, t) au(x, t)
a
C7t2 a x2 - at - 3u(x' t)' (10.148)

where c and 3 are considered nonnegative constants, subject to

u(0, t) = 0 and u(L, t) = 0 (10.149)
or

a 0, t)
=0 and 0 (10.150)
 396 Chapter 10: Classical Problems of PDEs

and

u(x, 0) = 1(x) and g(x). (10.151)

(1) Solve the Telegrapher's equation for both sets of BCs, Equations (10.149
and 10.150).
au(x' t)

22)[()
(2) Multiply Equation (10.148) by 2 and derive the differential iden-
tity

a au au a au 2 au\ 2
+ but - 2a au 2
+( =0.
ax at ax - at I

(3) Prove that if u(x, t) satisfies Equation (10.148) and either BCs, Equation
(10.149), or BCs, Equation (10.150), then

but fL a12 + but dx.

L L (ax) 2 + (at) 2 + t-To
dx <_
L() 2

l t-
(4) State and prove the uniqueness theorem for the initial value problem
given by Equations (10.148, 10.149, and 10.151).
10.6.2. Given the Telegrapher's equation
1 a2u(x,t) a2u(x,t) au(x,t)
C2 at2 = ax2 -a at - 3u (x, t).
au(x, t) au(x, t)
Let u 1(x t) = u(x t), u2(x,t) = ax and u 3(x t) =
an and show
at
that ul (x, t), u2 (x, t), and u3 (x, t) satisfy the following system of three equa-
tions:
aul (x, t)
u3 (x , t) 0
at
au2(x, t) au3(x, t)
0
at at
au (x,t) _2 thL2(x,t)
+au3(xt)+ui(xt)) = 0.
at C ax

10.6.3. Solve the Telegrapher's equation

a2u(x, t) a2u(x, t) au(x, t)
ate = ax2 - at - u (x, t) ,

subject to
u(O,t) = 1 and u(L,t) =cost
 Section 10.6: The Telegrapher's equation 397

and

u(x,0) = x3 - 2x + 1 and
tht' 0) =0.

10.6.4. Solve the Telegrapher's equation

a2u(x, t) - a2u(x, t) a(x, t)
_ u(x, t
ate ax e - at
subject to

u(0, t) = sin t and t = cost

and
au(x' 0)
u(x,0) = x and at =1.

10.6.5. PROJECT:5 Consider the line equation for a neuron

av _ a2 v
axe - v, (10.152)
at
subject to the IC
v(x, 0) = 0. (10.153)
(1) Suppose the left boundary at x = 0 is a voltage-clamp BC. That is,
v(0, t) = vb,
where vb = volts, and at the right boundary, x = L cm, there is a
current injection. Thus, the BC is given as
av(L, t)
ax _ -Riciml(t),
where t is given in microseconds, Ri is the intracellular resistance, 7m is
in cm, and 1(t) is the current injection.
(2) Another set of BCs would be
av(0,t) _ 0
ax '

which means the injected current, 1(t) = 0, and

av(L,t) _ R
v(L, t) - 'Y x s
a
where the soma acts like a resistance. However, the soma membrane
potential is the same at all points. These are two of the three assumptions
of the "Ball lumped-soma model." Note: The BCs for this part of the
problem are simplified.
5Adapted from James Keener and James Sneyd, Mathematical Physiology, ©1998 by Springer-
Verlag, New York, pp. 251-256. Reprinted by permission.
398 Chapter 10: Classical Problems of PDEs

10.7 INTERESTING PROBLEMS IN DIFFUSION

The most famous diffusion example you know is the heat equation, which was
derived in Chapter 2. However, the diffusion equation can be used in many areas
other than just the heat equation. This section is devoted to problems of this type.
Although many of them do involve temperature distribution directly, you really
want to stay focused on the primary structure of these problems, which is diffusion.
We start and end our discussion with a rather simple example.
EXAMPLE 10.1. Suppose a tungsten rod of length 7r ft is perfectly insulated on
its lateral surface and has an initial temperature distribution of

x, 0<x<4;
7r
4,
7r
x<4,
37r

and
37r
7r - x, -j-< x< 7r.
Further suppose that the rod has no insulation on its ends and is plunged into a
bath, which is held at the temperature of 0° F.
(1) Find the temperature distribution of the rod for all time.
(2) Consider the point x = and estimate the error made in replacing the series
2
by its partial sum. Then, determine the time required for which the ratio of the
sum of all terms, starting with the second, to the first term, is less than s > 0.

Solution: Part (1), we mathematically model the physical problem. We have

C72 26
2.39 axe , (10.154)
at
subject to
u(0, t) =o
(10.155)
u(ir,t) =0
and

x, 0<x<4r
ir ir < 3ir
x< (10.156)
4' 4- 4

3ir
it - x, < -x< ir.
4
 Section 10.7: Interesting Problems in Diffusion 399

Equation (10.155) subject to the homogeneous Dirichlet BCs has eigenvalues of

An = n2, n = 1, 2, 3, .... (10.157)
Thus, the general solution for u(x, t) is

u(x, t) = brie-2.39n2t
sinnx. (10.158)
n=1
Using the IC, we solve for a specific solution, which is
00
4 nor nor
uxt= n 7r
(sin
2
cos
4
e-2.39n2t
sinnx. (10.159)
n=1

Part (2), we solve for u , t), which is

2
00

(__-4--)
sin cos
nor
2
nor e-2.39n2t
sin
nor
2
. (10.160)

Equation (10.160) is 0 whenever n is even. Thus, we can rewrite Equation (10.160)

as
00
4(-1)m
u
?C

2 t - m=0 (2m+ 1)2 it sin

2m2 1 it
cos
2m+1)ir e-2.39(2m+1)2t. (10.161)

The series on the right side of Equation (10.161) satisfies the alternating series test.
Therefore, we know that if s is the sum of the series and sn is the nth partial sum,
we have

I s - sn I <- zn+1
where zn+1 is the n + 1 term of the series. Therefore,
r2,t 00
4(_1)m
Sln
2m+1 it
cos
2m+1 it e-2.39(2m+1)2t
(2m+1) it 2 4
m=n+1

(sin 2n+3 2n+3 e-2.39(2n+3)21

< 4
- (2n+3) 2
cos 4

?r
where Is - sn I = Rn 2, t

We can now estimate the ratio of the sum of all the terms of the series starting with
the second term to the first term. This yields
R0 (,t) 2
4 (:) e-2.39(3)21
2

e-2.3st - (2l e-2.39t

I I

1 e-2.39(26)t
<
1 e-2.39(52)t <.
9 -9
 400 Chapter 10: Classical Problems of PDEs

Thus, for t > t = -1 In 9Ewe have the ratio of the sum of all terms starting
with the second term to the first term less than > 0.

EXERCISES 10.7

10.7.1. Given a rod made of nickel with perfect lateral insulation and perfect insula-
tion on the end x = 0. Suppose the other end of the rod, at x = 15 ft, has a
convective heat exchange with a medium whose temperature is 0°F.
(1) Find the temperature distribution in the rod if the initial temperature
distribution is 100°F.
(2) Estimate the error made in replacing the sum of the series, representing
the point x = 15 by its partial sum.
2

Determine the time at which a steady-state will occur at x = 15 to a

(3) 2
degree of accuracy .
10.7.2. Given a rod made of lead with no lateral insulation of length Sir m. Suppose
the rod is held in a medium of 0°C, and the ends of the rod are uninsulated..
(1) Find the temperature distribution of the rod if the initial temperature
of the rod is given as
x, 0 < x < 3ir

f (x) = 37r, 37r < x < 4ir

57r-x, 47r < x < 57r

and the convective constant of proportionality is 0.208.

(2) Given the same rod, medium, and initial temperature distribution, sup-
pose a heater moves with constant velocity of 5 cm/s along the rod. The
flow of heat from the heater to the rod is Q (t) = 350e-0.208t degrees
Celsius. Find the temperature distribution in the rod for all time.
10.7.3. PROJECT: Give the mathematical model for the cooling of a uniformly
heated rod having the shape of a right circular cone of height L and base
radius r. Assume the temperature is constant over the cross sectional area of
the rod. Suppose the ends of the rod are perfectly insulated, and there is a
lateral heat exchange between the surface of the rod and the medium the rod
is in, whose temperature is 0°C.
(1) Suppose the height of the right circular cone is 2ir m and its radius 0.5
m, and it is made of aluminum. Also, suppose there is perfect thermal
 Section 10.7: Interesting Problems in Diffusion 401

insulation on the lateral surface of the rod and that the ends of the rod are
held at 0°C. Find the temperature distribution if the initial temperature
is uniformly 100°C throughout the rod.
(2) Suppose the cone is made by rotating the curve y = 3e-x, 0 < x < 5
about the x-axis. Suppose this cone is made of copper and is uniformly
heated to a temperature of 300°C. Further, suppose the lateral surface
area and the end x = 5 is perfectly insulated and the end x = 0 is held
at 0°C. Find the temperature distribution in the cone for all time.
10.7.4. Given a parallelepiped made of asbestos with 0 < x < it ft, 0 < y < 2ir
ft, and 0 < z < 3ir ft, find the temperature distribution if the sides of the
parallelepiped are maintained at 0°F and the initial temperature distribution
is 100°F. Find the time at which a steady-state will occur at the center of the
parallelepiped with relative accuracy > 0.
10.7.5. Consider a spherical shell made of mild steel with inner radius r1 = it m and
outer radius of r2 = 2 m. Suppose the inner and outer surfaces of the sphere
have a convective heat exchange with a medium whose temperature is 0°C,
and that the initial temperature of the spherical shell is r2 + 1 for r1 <r <r2.
10.7.6. Find the temperature distribution of a rod of length 2ir m, that has perfect
lateral insulation, consisting of two homogeneous materials: the first half of
the rod is made of aluminum, the second half of the rod is made of silver.
Suppose the left end of the rod is held at 0°C and the right end has a heat
flow of sin t watts, where t is time, and the initial temperature of the rod is
sin x, 0 < x < it

6(cosx-1),
10.7.7. A spherical vessel filled with gas moves uniformly for a long time with ve-
locity v0, and then at time t = 0 it is stopped instantaneously and remains
stationary. Find the vibrations of the gas in the vessel.
10.7.8. Find the vibrations of a gas in a spherical vessel of radius r = 3 produced by
small deformations of the wall, beginning at time t = 0, if the velocities of the
wall are radial and equal

6Pn(cos 8) cos(wt).

Note: Pn (cos 8) is a Legendre polynomial.

Chapter 11

Fourier Integrals
and Transform Methods
11.1 INTRODUCTION
In Chapter 3, we discussed d'Alembert's solution to the two-dimensional wave equa-
tion. The particular solution developed was for an infinite string. So far this is the
only method we discussed for infinite boundaries. In the current chapter, we will
develop methods that allow us to solve linear second-order PDEs with semi-infinite
and infinite boundaries. These methods involve transforms.
You were first introduced to transforms in your ordinary differential equation
course. There you solved linear initial-valued ODEs with Laplace transforms. The
method involved transforming an initial-valued ODE into a space where the equa-
tion could be solved algebraically. Once the algebraic solution was found, the inverse
Laplace transform is applied returning you to the space of the original problem. In
ODEs, tables of Laplace transforms were used to simplify the process. Later, in this
chapter, we will again visit Laplace transforms. Here, we use them as another solu-
tion method for initial-valued PDEs with finite and semi-infinite boundaries. Again
tables of Laplace transforms are provided to simplify the process. However, unlike
solving ODEs by Laplace transforms, solving PDEs by Laplace transforms does not
generate an algebraic equation to be solved but an ODE one to be solved. Before
we solve PDEs with Laplace transforms, we will introduce the Fourier integral.
The Fourier integral is a natural extension of Fourier series. By extension, we
mean the representation of a piecewise-smooth function with a Fourier integral
where the domain of the function is semi-infinite or infinite. This is quite different
than Fourier series where the domain is typically [-L, L]. However, as in Fourier
series representation of a function, there are certain restrictions on a function rep-
resented by a Fourier integral. These restrictions are discussed in some detail.
Once the Fourier integral is developed, the Fourier sine and cosine integrals are
discussed. The Fourier sine and cosine integrals are the natural extensions of the
Fourier sine and cosine series, respectively. Next, we introduce transform solution

403
404 Chapter 11: Fourier Integrals and Transform Methods

methods by considering the Laplace transform of PDEs with finite or semi-infinite

boundaries. We then move on to a discussion of the Fourier transform. Finally,
we use the Fourier and the Fourier sine and cosine transforms to solve PDEs with
semi-infinite and infinite boundaries.

11.2 THE FOURIER INTEGRAL

In Chapter 4, we started our discussion of Fourier series. There, we discovered
that the Fourier series representation of a piecewise-smooth function on [-L, L]
converged to the function wherever the function was continuous and to the average
of the left and right values of the function at any jump discontinuity. In Chapter
6, we started with a discussion of Fourier series as the representation of a function
and the Fourier series as a function itself. For instance, consider the function
x3 - 3x2 - 2x + 3 on the interval [-2, 2] shown in Figure (11.1). The Fourier series

Figure 11.1: Graph of x3 - 3x2 - 2x + 3 on [-G, 2].

representation of the function x3 - 3x2 - 2x + 3 is given by

00
8 (- 1 ) n L-6n nlrxll
-1 +
(nir)3
f
cos (---) + (12 - (nir)2) sin t 2
(11.1)
n=1

which is shown in Figure (11.2). Please note that Figure (11.2) exhibits the Gibbs
phenomenon, and that at x = ±2 the Fourier series converges to -9.
The Fourier series representation of x3 - 3x2 - 2x+ 3 as a function is considered is
periodic on [-oo, oo]. Figure (11.3) shows this periodic nature of the Fourier series
function using x3 - 3x2 - 2x + 3 on [-2, 2] as the original function represented.
Thus, we speak of a Fourier series involving "periodic" functions. The question
which arises is; what do we do with a nonperiodic function on [0, oo] or [-oo, oc],
such as

0, -2<xorx>2
f (x) _ (11.2)
x3-3x2-2x+3, -2<x<2.
 Section 11.2: The Fourier Integral 405

Figure 11.2: Graph of Fourier series representation of x3 - 3x2 - 2x +3 on [-2, 2]

where n = 150.

Figure 11.3: Graph of Fourier series function on [-oo, 00] where n = 150.

Since the Function (11.2) is not periodic, it can't be represented by a Fourier

series. However, a solution method is needed, and Fourier integrals were developed
to handle exactly these type of cases.

11.2.1 Development of the Fourier Integral

We develop The Fourier integral with an intuitive approach. However, at the con-
clusion of our approach, we state the Fourier integral theorem. The Fourier integral
theorem states all the sufficient conditions necessary for the Fourier integral repre-
sentation of a function. The proof may be found in Fourier series and Boundary
Value Problems by Ruel Churchill. We now start with our intuitive approach,
Suppose we have an arbitrary piecewise-smooth function 1(x) on the interval
[-L, L]. Then, we may represent 1(x) by the Fourier series

00

f (x) = ao+ an cos (nLx) + bn sin (nLx) , (11.3)

n=1
406 Chapter 11: Fourier Integrals and Transform Methods

where

ao
2L
1 f L
L
1(x) dx,

fL
an L f (x) cos (nLx) dx,
L
and
1 L
bn 1(x) sin (L ) dx.
L J L
nor
If we let 8 = , and replace ao, an, and bn in Equation (11.3) with their integral
L
representations, we have

f (x) dx +[fl(x)
1 L o0 1 L
f (x) =
2L f L
- cos (ex) dx cos (ex)
L
n=1 I'

+
[fL 1(x) sin (Ox) dx sin (Ox)}. (11.4)
L

Now, letting 08 = en+1 - en _ (n+1)ir L- L =Land substituting O8 into

Equation (11.4) yields
00
L
+ °e L f(x) Los (ex) dx Los (ex)
1(x) - 2Lf Lf (x) dx n=1 f L

fL
+ e 1(x) sin (Ox) dx sin (Ox)
J L

1 00 L
1 L
2L, 1(x) dx +-ir IfJ 1(x) cos (Ox) dx OB cos (Ox)
n=1 L

fL
+ 1(x) sin (8x) dx OB sin (Ox)
L
J
[fL
L
1(x) dx + L
1(x) cos (Ox) dx cos (Ox)

+ L 1(x) sin (8x) dx sin (Ox)1 OB. (11.5)

J L J
Taking the limit of Equation (11.5) as L -> oo results in some interesting discoveries.
First, in Equation (11.5),
1
L1(x)dx=o.
 Section 11.2: The Fourier Integral 407

Second,

{ 1(x) cos (8x) dxJ cos (Ox) + 1(x) sin (Ox) dxJ sin (Ox) } D8
ll J
I

LJ J

looks very similar to

00
1
1 F(B)OB
n=1
where

F(B) = 1(x) cos (Ox) dxJ cos (Ox) + I 1(x) sin (8x) dxJ sin (Ox)
LJ 0o L J o0

Since L -* oo implies 08 -* 0, we could look at Equation (11.6) as

00
1
- lim >2F(O)LO. (11.7)
?r
n=1
Equation (11.7) is very suggestive of the definition of the improper integral

1 F(B) dB,
Jo
which becomes

1(x) _ { I f (x) cos (Ox) dxJ cos (Ox)

oJ00 l lJ

+ [f:f(x) sin (Ox) dx sin (Bx) d8. (11.8)

Equation (11.8) is called the Fourier integral.

Remember, this was an intuitive approach and left many details out. These
details are stated in Fourier's integral theorem.
Theorem 58. (Fourier's integral theorem If 1(x) is piecewise-smooth on every
finite interval on the x-axis, and at every point of jump discontinuity the left and
right hand limits exist; furthermore, suppose that 1(x) is absolutely integrable, that
is,

J- f (x) dx
exists. Then, at every point x where the one-sided derivatives of f (x) exist, the
function f (x) may be represented by the Fourier integral

.f(x) _ f00 cos (Of) d] cos (ex)

o { [J-00

+ d sin (Bx) dB. (11.9)

408 Chapter 11: Fourier Integrals and Transform Methods

Please note, in Theorem (58), that the absolutely integrable function, 1(x),
eliminates all periodic functions except 1(x) = 0. Also, a more convienient form of
Equation (11.9) is

f(x) = f {a(e) cos(ex) +B(O) Sin(ex)} de (11.1o)

0

where

A(B) _ (11.11)
f:f()cos(o)d
and

B(B) = 1 ff()sin(O)d (11.12)

We demonstrate Theorem (58) with three examples.

EXAMPLE 11.1. Consider

0, -1 <xorx> 1
f (x) = 1, -1<x<0
2, 0<c<1,
shown in Figure (11.4). The Fourier integral of the function, 1(x), given in Equation

2-
2.5

1.5

0.5
x
-10.5

-1

Figure 11.4: Graph of 1(x).

(11.13) is

1(x) = {A(B) cos(Bx) +B(O) sin(Bx)} dB,

J0
 Section 11.2: The Fourier Integral 409

where

A(B) _ f d
J
(J °i d+ 12 d}
{f° Jo J

3 sin(B)
eir

and

B(e) - f
1/00
f d
(J ° sin(O) dx + sin(e) d }
i
{f° Jo 12 J

1-cos(O)
eir

Therefore,

f (x) = 3 Sew e) cos(Bx) + 1 sin(Ox) dB. (11.14)

0
J
Most Fourier integrals are difficult to integrate. Thus, they are generally left in
their integral form. However, many algebraic software packages can evaluate the
integral numerically for plotting purposes. Figure (11.5) shows the graph of the
Fourier integral representation of 1(x). The vertical bars from 0 to 1 at x = -1,
from 1 to 2 at x = 0, and from 2 to 0 at x = 1 are a result of the mathematical
software being used.

3
2.5

2
1.5

Figure 11.5: Graph of the Fourier integral of 1(x)

410 Chapter 11: Fourier Integrals and Transform Methods

-3

Figure 11.6: Graph of g (x) .

EXAMPLE 11.2. Consider

0, -2<xorx>2
g(x) _
-2, -2 < x < 2,
shown in Figure (11.6). The Fourier integral of the function, g (x), given in Equation
(11.15) is

g(x) = {A(B) cos(Bx) +B(O) sin(Bx)} dB

J0
where

A(B) _
J
d
- 1
p2
- 2 cos(ec) d

4 sin(2B)
Bir

and

B(B) _ fg(e)sin(O)d

l2 - 2
1 J2 d
=0.
Therefore,

g(x) = J
/'°O

0
-e 4 sin(2e)
cos(Bx) d8. (11.16)
 Section 11.2: The Fourier Integral 411

2 2
x

-3

Figure 11.7: Graph of the Fourier integral of g(x).

Figure (11.7) shows the graph of the Fourier integral representation of g (x) . The
vertical bars from -2 to 0 at x = ±2 are a result of the mathematical software being
used.

EXAMPLE 11.3. Consider the function

0, x<-lorx>l
h(x) = -1, -1<x<0
1, 0<x<1.
Figure (11.8) shows the graph of the Equation (11.17). The Fourier integral of the

2
1.5
1

0.5
x
-1 1

-0.5

-1.5
-2

Figure 11.8: Graph of h(x).

function h(x) given in Equation (11.15) is

h(x) = f {a(e) cos(ex) + B(e) Sin(ex)} de,

412 Chapter 11: Fourier Integrals and Transform Methods

where
1 f°O
d

lf - d+f1
o
de}

and

B(B) = 1 d
-00
J °
_{f1

l i
- d+
Jo
1 sin(9)d}
J

2r1- s(8)1
l J

Therefore,
°° 2 1 - cos(B)
h(x) = sin(ex) de. (11.18)
0 a

Figure (11.9) shows the graph of the Fourier integral representation of h(x). The
vertical bars from -1 to 0 at x = -1 and 1 to 0 at x = 1 are a result of the
mathematical software being used.

x
1
0.5

-1.5

-2

Figure 11.9: Graph of The Fourier integral of h(x).

Fourier's integral,

f(x) = f
0
{A(O) cos(ex) + B(e)Sin(ex)} de, (11.19)

where

A(8) = 1
J f()
d (11.20)
 Section 11.2: The Fourier Integral 413

and

B(8) = 1 ff()sin(O)d, (11.21)

has other convenient forms. For instance,

x= 1
is another form of Fourier's integral.
o f f() cos B x- d d8 (11.22

We develop Equation (11.22) from Equations (11.19, 11.20), and 11.21) in the
following manner. First, Equations (11.20 and 11.21) for A(e) and B(e) are

A (O) 1 Il: f() cos B d (11.23)

and

B(B) = 1 f() sin(O) (11.24)

J
Next, we place Equations (11.23 and 11.24), A(e) and B(e), in Equation (11.19).
This yields

.f (x) = 1of { [ff cos (Of) dam] cos (ex)

+ sin (8x)1 } dB. (11.25)

Equation (11.25) may be rewritten as

f(x) _ - ff .f() {cos (ef) Los (ex)

+sin sin (Bx)} (11.26)

Now, we consider part of the integrand of Equation (11.26), which is

cos () cos (8x) + sin (O) sin (Ox). (11.27)

Equation (11.27) reduces to

cos(e - ex) = cos(9( - x)) = cos(B (x - )) (11.28)

by the cosine of the difference of two angles' identity and the fact that the cosine
function is an even function. Finally, we place Equation (11.28) in its proper spot
in Equation (11.26), which yields

f(x) _ (11.29)
ir J J
414 Chapter 11: Fourier Integrals and Transform Methods

Equation (11.29) also has the form

1(x) 1 L f f() cos B x- d d8. (11.30)

Two other useful forms of Fourier's integral are

f()eZ°(x-)
1(x) _ (11.31)

and
oo oo
f(x) - 2ir f eiexde f (e)e-ZB (11.32)

Equations (11.31 and 11.32) are known as the complex form of Fourier's integral,
and their derivation is left as exercises.
Since we have considered a Fourier integral as the limiting case of a Fourier
series, it is natural to assume that the limiting case for the Fourier sine and cosine
series are Fourier sine and cosine integrals. This topic is discussed in the next
subsection.

11.2.2 The Fourier Sine and Cosine Integrals

In the previous subsection, we stated Fourier's integral theorem. Also, we gave the
preferred form of Fourier's integral, which is

f(x) = f
0
{a(e) cos(ex) +B(9) Sin(ex)} de, (11.33)

where

A(B) = 1
f:fcosod (11.34)

and

B(B) = 1 (11.35)

In Example (11.2), the function,

0, -2<xorx>2
g(x) _ (11.36)
-2, -2<x<2
was represented by a Fourier integral. However, since g (x) is an even function, B (O)
in Equation (11.35) equaled 0. Thus, A(e) is an integral of g(x) cos(Bx), which is
 Section 11.2: The Fourier Integral 415

the product of two even functions. Therefore, the integral A(B) could be written as

A(8) _
itJ0
d

o J -z d
2 f2

J- 2 (11.37)

Using Equation (11.37), we rewrite Equation (11.33) as

g(x) = A(9) cos(ex) d8. (11.38)

0

Equations (11.38 and 11.37) are an example of the Fourier cosine integral. For-
mally, the Fourier cosine integral for even functions, 1(x), is given as

1(x) = A(9) cos(9x) dB (11.39)

J0
where

A(8) = 2 ff()cos(ee)d. (11.40)

Close examination of Example (11.3) indicates that h(x) is an odd function.

Thus, A(e) = 0. This means that Equations (11.33 and 11.35) form the Fourier
sine integral. Formally, the Fourier sine integral for odd functions, 1(x), is given as

f (x) = B(8) sin(9x) d8, (11.41)

J0
where

B(8) = 2 ff()sin(O)d. (11.42)

In Example (11.4), we examine the Fourier cosine and sine integrals with the
function 1(x) = 2e-x, where 0 <_ x.

EXAMPLE 11.4. Consider the function

1(x) = 2e-x where 0 <x (11.43)

shown in Figure (11.10). First, we find the Fourier cosine integral of 1(x). This
416 Chapter 11: Fourier Integrals and Transform Methods

Figure 11.10: Graph of 1(x) = 2e-x, where 0 < x.

means, we compute

A(8) _ ff() cos(9e) d

71

b
4
- lim e- cos(Be) d
7r b-9oo U

-6 \
{B sin(b9) - cos(bB)}
7r (1 + 92 + e 1 + 92
41 (11.44)
r 1 + 9V

Thus, the Fourier cosine integral of 1(x) = 2e-x is

Figure 11.11: Graph of The Fourier cosine integral of 1(x) = 2e-x, where 0 < x.
 Section 11.2: The Fourier Integral 417

f (x) = fA(e)cos(ex)dO

Jo 1+92 cos(ex) dO
cos(9x) dB, (11.45)
0 1 + B2
which is shown in Figure (11.11). Now, we find the Fourier sine integral of 1(x).
This means, we compute the following:

B(9) = 2 ff()sin(9)d

o2e- sin(9) d

fe _ sin(g) d
b
4
lim
7r b-9oo 0

4 /1+92
9 -b \
{sin(bO) +9 cos(b9)}
n i. 1 + 92

49 (11.46)
7r l+92 .
Thus, the Fourier sine integral of 1(x) = 2e-x is

Figure 11.12: Graph of The Fourier sine integral of 1(x) = 2e-x, where 0 < x.

1(x) = fB(9) sin(9x) d8

P004 e
sin(9x) d9
J0 ?r 1 + 92

4 °° e
sin(9x) d9, (11.47)
?r o 1 + 92
 418 Chapter 11: Fourier Integrals and Transform Methods

which is shown in Figure (11.12).

Another important form of the Fourier cosine integral is

1(x) _ it f cos(ex) de f .f cos(ee) (11.4s)

0 0

where the variable of integration in the second integral was changed to for conve-
nience. Also, an important form of the Fourier sine integral is

1(x) = 2 sin(Ox) d8 f() sin(O) d (11.49)

0 0

Again, the variable of integration in the second integral was changed for convenience.
Before we proceed with the Fourier transform, our next section discusses Laplace
transforms and how they apply to PDEs.

EXERCISES 11.2

11.2.1. Find the Fourier integral representation of the following functions and graph
the solution using an algebraic software package:
0, x>33
(1)

(3)

(4) f(x)=
 Section 11.2: The Fourier Integral 419

x, 1<x<0
(5) x2, 0<-x<1
0, x>1.1.
sin x, IxI < it
(6) 1(x) {
0, x > 7r.

cos x, x< 7r7r

(7)
{
0, x> 7r.
7r.

(8) f(x)=

xI> 2.

x2-1, -3<-x<0
x-1, 0<-x<l
(9) f(x)=
-x2, 1<x<3
(xI> 3.

0, x<-1
(10) f(x)=1 -x-l, -1<- x<0
_e-x, 0<- x.
11.2.2. Starting with Fourier's integral, Equation (11.9), show that

-
1 °° °°
1(x) = f ()ei9(x-) ddb
is another form of Fourier's integral.
11.2.3. Starting with Fourier's integral, Equation (11.9), show that
1 oo oo

1(x) = 27r _
J
eZex
d8
_00
f ()e-Zed
d
 420 Chapter 11: Fourier Integrals and Transform Methods

is another form of Fourier's integral.

11.2.4. This problem develops the Sine Integral,

z
Si (z) =
15rn9 dB.
0 8

(1) Find the Fourier integral representation of

x2-1, (xI< 1

0, (xI> 1.

(2) Show that the solution to part (1),

2 0<x<1
-
f cos (Ox' sin
e

0,
x=1
1<x.

(3) Evaluate the integral from part (2) at x = 0.

(4) Determine that the integral in part (3) is the limit of the sine integral,
and using an algebraic software package, graph the sine integral.

11.2.5. Find the Fourier sine and cosine integral representation of the following func-
tions and graph the solution using an algebraic software package:

1, 0<x<1
(1)
{ 0, 1<x.

x3, 0<x<1
(2) 1(x) { 0, 1<x.

ex, 0<x<1
(3)
{ 0, 1<x.

25-x2, 0<x<5
{ 0, 5<x.
 Section 11.3: The Laplace Transform and PDEs 421

x, 0<x<2
(s) f(x)= 4-x, 2<x<4
0, 4<x.
1
1<x<2
x
(6) f(x)
={ 0, otherwise.
(7) f(x)=e_x+e_2x, 0<x.

(8) f(x)=e ,0<x.

sin x, 0 < x < it
(s)
0, otherwise.

cos x, 0 < x < it

(10) f(x) =
l 0, otherwise.
11.2.6. Starting with Fourier's cosine integral, Equation (11.39), show that

f (x) = 2 fcos(Ox)dOff(cos(O)d

is another form of Fourier's cosine integral.

11.2.7. Starting with Fourier's sine integral, Equation (11.41), show that

?
J0 f() sin(O)
f (x) _ sin(Ox) dB
Jo
is another form of Fourier's sine integral.

11.3 THE LAPLACE TRANSFORM

When you first encountered Laplace transforms, you learned that they were used for
solving ODEs. There you learned to take the transform of the ODE, which turned
the ODE into an algebraic problem. You solved the algebraic problem. Then, took
the inverse Laplace transform to obtain the answer for the ODE. In this section,
we are going to solve PDEs with both finite and semi-infinite spatial boundaries
with Laplace transforms. The method is very similar to that for ODEs. However,
there are a few minor differences. Instead of getting an algebraic expression after
transforming a PDE, we get an ODE, which we must solved.
422 Chapter 11: Fourier Integrals and Transform Methods

Before we solve a PDE with Laplace transforms, we'll state several important
theorems, list some known properties, and refresh our memories by solving an ODE.
(Note: The proofs for all the properties may be found in any standard ODE text.)
Next, we discuss the error function, which is also known as the probability integral.
Finally, we solve PDEs using the Laplace transform.

11.3.1 Laplace transform Solution Method of ODEs

You should remember that the basis of a Laplace transform is an integral equa-
tion. In fact, the integral is an improper one. This means that the integral over
the unbounded interval is defined as a limit of integrals over finite intervals. For
example,
00

f f (t) dt = lim
6-.00 0J
b f (t) dt,

where b is a positive real number. If the integral exists from 0 to /3 for each /3 < b,
and if the limb+00 exists, then the integral is said to converge to the limiting value.
If any part the previous statement is false, then the integral is said to diverge.
Given this brief description of an improper integral, we state the primary theorem
for Laplace transforms.
Theorem 59. Suppose that f is a piecewise-smooth function on the interval 0 <
t < T for any T E g+ and f (t) < beat when t > M and where a E Ilk and M and
k E Ilk+. Then, the Laplace transform is given by

G {f(t)} = F(s) = e-st f (t) dt, (11.50)

J0
where s>a.
Theorem (59) defines the Laplace transform. However, when we solve initial
value problems, it is useful to have the following theorem:
Theorem 60. Suppose that the functions f, f', f", ..., f fl are continuous and that
f n+1 is piecewise-smooth on the interval 0 < t < T for any T E + . Also, suppose
that there exist real constants k, a, and M such that f (t) I < beat , If' (t) I < treat
... , I fn(t) I < treat when t > M. Then, £ {f+'(t)} exists for s > a and is given
by

£ {f+'(t)} = sn+1 {f(t)} - sn f (O) - ... - s f n-1(0) - f n (0) . (11.51)

The following list contains several more properties of Laplace transforms. You
should be familiar with all of them.
Properties of Laplace transforms

G{af(t)}=aG{f(t)}.
£ {f(t) + g(t)} = £ {f(t)} + £ {g(t)}.
 Section 11.3: The Laplace Transform and PDEs 423

Properties of Laplace transforms (continued)

£ {f'(t)} = sF(s) - f(0).

G { f"(t)} = s2F(s) - s f (0) - f'(0).

£ {ebtf(t)} = F(s - b).

f(t')dt' = F(s).
( o /
£{tf(t)} - _dF(s)

£ {f(t)}
t
= fF(s') ds'.

Theorems (59 and 60) and the other properties of Laplace transforms provide
the necessary tools for solving initial-value ODE problems. Consider the following
example:
EXAMPLE 11.5. Consider
u"(t) + 2u'(t) - 3u(t) = 0, (11.52)
subject to
u(0) = 1 and u'(0) _ -2. (11.53)
The Laplace transform of Equation (11.52) is
G {u"(t)} + 2G {u'(t)} - 3G {u(t)} = 0.
Using Theorem (60) and the initial conditions, Equation (11.53), the Laplace trans-
form of Equation (11.52) becomes
s2G {u(t)} - su(0) - u'(0) + 2 [sL {u(t)} - u(0)] - 3G {u(t)} = 0
or

(s2 + 2s -3) F(s) - (s + 2)u(0) - u'(0).

Solving for F(s) yields

s = s
F(s) = 52+25_3
(s+3)(s-1)
Therefore, taking the inverse Laplace transform we have
s 3e-3t + et
u(t)= -1
l(s+3)(s-1)1 4
where we used the table of Laplace transforms in Appendix F.
424 Chapter 11: Fourier Integrals and Transform Methods

It should be noted that transforms come in pairs. Theorem (59) gives a definition
of the Laplace transform. The inverse Laplace transform requires knowledge of
contour integration. However, the inverse Laplace transform is

f (t) = G-1 {F(s)} F(s)est ds. (11.54)

J
This short review is certainly not meant to be complete, just a refresher. For a
complete review of this material, I suggest that you read your ODE text. We now
move on to an important integral, the error function, which is needed for even an
introduction to Laplace transform methods for PDEs.

11.3.2 The Error Function

The error function plays a very important role in many areas of engineering and
science. The error function is denoted by erf (x), where

2 f°°e_u2du.
erf (x) _ e-"z du = 1 -
J J0
Also, there is a complementary error function, denoted erfc (x), where

erfc (x) _ e_u du = 1 - erf (x).

It should be noted that erf (0) = 0 and erfc (0) = 1. Following is a proof of erfc
(0) = 1, from which we find that erf (0) = 0.

Proof. Let

e-x,
00
2
I= d
2

Jo

and consider I2. Thus,

I2 = 2 I/'O° e-X2 dx e_y2 dy

Jo o

where we changed the variable of integration in the second integral. I2 may also
be written as
4 fOC OC
dx dy, ( 11.55 )
o J
transferring to polar coordinates where r2 = x2 + y2 and 0 = tan- 1 y . Also,
x
changing to polar coordinates requires a change to the limits of integration. Thus,
 Section 11.3: The Laplace Transform and PDEs 425

for r, we have 0 <r < oo, and for 0, we have 0 <0 < Thus, Equation (11.55)
2
becomes
4 2 00
_r,2 4/1:1 1 00

e rdrd8=_
-2e_T2 de
0 J
0 0 0

2
=2 d8
it o

Figure (11.13) shows the graph of erf (x) and erfc (x).

Figure 11.13: The Graph of erf (x) and erfc (x).

In Figure (11.13), please note that as x - Oo, erf (x) - 1 and erfc (x) - 0. Also,
in Appendix F, there are several useful Laplace transforms of the complementary
error function. These transforms will be needed as you solve the problems at the
end of this section and chapter. Next, we look at Laplace transforms and PDEs.

11.3.3 Laplace Transform Solution Method of PDEs

For this section, we work with PDEs that are functions of two variables, x and
t. We immediately note that a Laplace transform is designed to work with one
variable. However, we learned in Chapter 6, The Calculus of Fourier series, that we
could take the derivative of a Fourier series with respect to a parameter t. We will
use that same type of technique here. Only now the parameter is x. Thus, using
Theorem (59), the definition of the Laplace transform of a function of two variables,
u (x, t) becomes

G {u(x, t)} = U(x, s) = e-stu(x, t) dt. (11.56)

J0
426 Chapter 11: Fourier Integrals and Transform Methods

Definition (11.56) is demonstrated by two examples. Remember, that an integral

over an unbounded interval is called an improper integral, and it must be defined
as the limit of integrals over finite intervals.
EXAMPLE 11.6. Find the Laplace transform of u(x, t) = e-3t sin(irx).
By Definition (11.56), we have
0o b
C {u(x, t)} = e-ste-3t sin(irx) dt = sin(irx) lim e-(8+3)t dt
J0 b- +o0 0

b
1
= sin 7rx lim -1 e-(s+3)t
( )b-+oo [s+3 0

- sin(irx) 1l = sin(irx)
lim U
s+3 t-+L J s+3
EXAMPLE 11.7. Find the Laplace transform of cos(x+t). By Definition (11.56),
we have

G {u(x, t)} = e_St cos(x + t) dt

J0

f
b

= lim e-st (cos x cost - sin x sin t) dt

b- +o0

b b
= lim
b-> o0 f e-St cos x cos t dt -
0
e-St sin x sin t dt

b b
= cos x lim a-St cos t dt - sin x lim a-St sin t dt
b- +o0 0 b- +o0 0

e-St b

= cos x lim (-s cost + sin t)

b->oo s2 - 1 0

e-st b

- sin x lim (-s sin t - cost)

b- oo s 2 -I- 1 0

cos x 1e-Sb
lim (-s cos b + sin b) + s]
s2 + 1 boo
 Section 11.3: The Laplace Transform and PDEs 427

sin x
lira [e-Sb (-s sin b - cos b) +1]
s 2 -I- 1 b- +o0

s cos x sin x s cos x- sin x

s2+1 82+1 82+1
Examples (11.6 and 11.7) demonstrate the Laplace transform of a function of
two variables where one variable is a parameter. We now move on to the Laplace
transform solution method of a PDE.
If we assume Theorem (60) holds true for the Laplace transform of a PDE, which
au(x,t) a2u(x,t) a2u(x,t)
it does, then we have for and axe the Laplace transforms;
at , ate ,
au(x,t)
a
=s,C {u(x, t- ux0=sUx s- ux0(11.57)

{ a ua 2' t)
j =Sec {u(x, t)} - Su(x, o) -
auat, °)

0)
= s2U(x, s) - su(x, 0) - au(x,
at
(11.58)

and
,oo
£1
82u(x, t)
e
-st a2u(x,
axe
t)
dt
8x2 0

p00 a2
JO

foo
dx2
[eu(x,t)] dt d=2U d ,s (11.59)
0

respectively.
A close examination of Equations (11.57 and 11.58) indicates that the Laplace
transform is ideally suited for PDEs with initial conditions, in particular, the heat
and wave equations in one spatial dimension where the spatial variable has either
finite or semi-infinite BCs. Also, Equation (11.59) informs us that the Laplace
transform of a linear second-order PDE is a linear second-order ODE. The first two
examples demonstrate this important concept.
EXAMPLE 11.8. Suppose that we have a very thin rod with perfect lateral insu-
lation of length L. Further suppose that the rod has an initial temperature distri-
bution of a sin
x and the ends are held in a 0° bath. Find the time-dependent
L
428 Chapter 11: Fourier Integrals and Transform Methods

solution for u(x, t).

The physical model may be described by the mathematical equations
au(x, t) a2u(x, t)
at - axe
subject to
u(O,t) = 0 and u(L,t) = 0

and
u(x, 0) = a sin L)
7rx
.

First, we find the Laplace transforms of all equations. We have, by Equations (11.57
and 11.59),

£1J t)1 l
= Sc {u(x, t)} - u(x, o) = Su(x, S) - u(x, o)

= sU(x, s) - a sin L
7rx
(11.60)

and

G f 82u(x, t) d2U(x, s)
(11 . 61)
8x2 } dx2

Also, the BCs become

£ {u(0, t)} = U(0, s) = 0 and £{u(L,t)} = U(L, s) = 0. (11.62)

Next, we combine Equations (11.60 and 11.61). This yields a nonhomogeneous

linear second-order ODE, which is
d2U(x, s) 7rx
- sU(x, s) _ -a sin L . (11.63)
dx 2
We solve Equation (11.63) by breaking it into homogeneous and particular parts.
The homogeneous part is
d2Uh(x, s)
- SUh(x, s) = O,
dx2
which has the solution
Uh(x, s) = c1 sinh(vx) + c2 cosh(vx). (11.64)

For the particular part, we use the method of undetermined coefficients. This
method may be found in Appendix C. The solution for the particular part is
aL2 irx
Up(x, s) = sL2
2 sin L (11.65)
 Section 11.3: The Laplace Transform and PDEs 429

Combining Equations (11.64 and 11.65) yields

U(x, s) = Uh (x, s) + Up (x, s)

2
irx
= cl sinh(/x) + c2 cosh(x) + SLa+ ir2
sin (-i-). (11.66)

Applying the BCs, Equation (11.62), implies c1 = c2 = 0. Therefore,

2
(lrXL
sin (11.67)
U(x' S) sLa + 2 1

Finally, we find the inverse Laplace transform of Equation (11.67), which yields the
time-dependent solution u (x, t). Note: A table of inverse Laplace transforms may
be found in Appendix F. We have
2 l
u(x, t) = G-1 {U(x, s)} = G-1 I sin (L)
SLa+ 2 J

2t '7X
= ae-T sin L (11.68)

Equation (11.68) is exactly what we should expect as the answer.

EXAMPLE 11.9. Consider a perfectly flexible string of length L with the ends
attached to frictionless sleeves, which move vertically up and down. Further, sup-
pose the string has no initial displacement. However, the initial velocity is given as
7rx
cos (). Find the displacement u (x, t).
L
The physical model may be described by the mathematical equations

a2u(x, t) a2u(x, t)
ate - axe

subject to

a a(0, t)
= 0 and =0

and

u(x, 0) = 0 and cos (L) .

First, we find the Laplace transforms of all equations. We have, by Equations (11.58
430 Chapter 11: Fourier Integrals and Transform Methods

and 11.59),

J a ua 2, t> l =Sec {u(x, t)} - Su(x, o) - auat,°)

l I

= s2U(x, s) - su(x, 0) - 8u(x, 0)

=s U(x,s)_cos._
x (11.69)

and

c a2u(x, t) d2u(x, s)
(11.70)
{ 8x2 } dx2
Also, the BCs become
8u(0, t)
8x
_ dU(0, s)
dx
= 0 and C
8u(L, t)
ax
Next, we combine Equations (11.69 and 11.70). This yields a nonhomogeneous
_ dU(L, s) _ 0.
dx ()
11.71

linear second-order ODE, which is

d2U(x, s) x
dx2
s2U(x,s) = _cos(-L-) (11.72)

We solve Equation (11.72) by breaking it into homogeneous and particular parts.

The homogeneous part is

d2Uh(x, s) 2
Uh(x, s) = 0,
dx2
which has the solution

Uh(x, s) = cl sinh(sx) + c2 cosh(sx). (11.73)

For the particular part, we use the method of undetermined coefficients. This
method may be found in Appendix C. The solution for the particular part is
L2 irx
Up(x, s) = s2L2 2 cos (11.74)
L
Combining Equations (11.73 and 11.74) yields

U(x, s) = Uh(x, s) + Up(x, s)

L2 (irx)
= cl sinh(sx) + c2 cosh(sx) + s2L2 + 2 cos . (11.75)
 Section 11.3: The Laplace Transform and PDEs 431

Applying the BCs, Equation (11.71), implies cl = c2 = 0. Therefore,

U(x, s) = s2L2 + 2 cos

z
(!). (11.76)

Finally, we find the inverse Laplace transform of Equation (11.76), which yields the
time-dependent solution u(x, t). We have

u(x,t) = £'{U(x,s)} =
L2 xl 1
cos (L / J
'{ L2 + ir2

L
sinrirt
(L cos l L (11.77)

Again, Equation (11.77) is exactly what we should expect as the answer.

In the next example, we solve a linear second-order PDE where the spatial
variable has semi-infinite boundaries. A semi-infinite boundary has one boundary
at x = 0. The other boundary is at infinity.
EXAMPLE 11.10. Consider a semi-infinite solid. The temperature distribution
is determined from the heat equation if the flow is one-dimensional. Suppose that
this is the case, and the initial temperature of the solid is 0°. Further suppose that
the left boundary, x = 0, has a constant flux of heat into the solid of -a. Solve for
the temperature distribution, u(x, t) in the solid.
The physical model may be described by the mathematical equations
au(x, t) _ t
at axe
subject to
5u(0, t)
-a and lim u(x, t) = 0
and
u(x, 0) = 0
First, we find the Laplace transforms of all equations. We have, by Equations (11.57
and 11.59),
auat,
c{ t> } = Sc {u(x, t)} - u(x, o) = Sv(x, s) - u(x, o)

= sU(x, s) (11.78)
and
G 82u(x, t) kd2U(x, s)
(11.79)
{ 8x2 } dx2
432 Chapter 11: Fourier Integrals and Transform Methods

Also, the BCs become

G {8u(0, t)1 _ dU(0, s) _ _ a

li m u(x t)} = lim U(xs) = 0. (11.80)
and G {xoo
ax J dx s '

Next, we combine Equations (11.78 and 11.79). This yields a homogeneous linear
second-order ODE, which is

kdU(x, s) - sU(x, s) =0.

dx2

We usually write the solution of Equation (11.81) in terms of hyperbolic sine and
hyperbolic cosine functions. However, in this case, it is easier to write the solution
in terms of exponential functions. Therefore, we have

U(x, s) = c2e. (11.82)

s) a from Equation (11.80) to Equation (11.82)

Applying the first BC,
yields

svT

Thus, U(x, s) may be written as

U(x, s) =
s + c2 + (11.83)

Applying the second BC, lim0 U(x, s) = 0, from Equation (11.80) to Equation
(11.82) implies c2 = 0. Thus, the solution is

U(x S) = (11.84)
sf
Finally, we find the inverse Laplace transform of Equation (11.84), which yields the
time-dependent solution u (x, t). We have

u(x, t) = £' {U(x, s)} = ' {ae s

_a 2-x()}
LV
2
{erfC (11.85)

where the inverse Laplace transform as found in Appendix F.

 Section 11.3: The Laplace Transform and PDEs 433

The last two sections of this chapter deal with the Fourier transform. In the
next section, we discuss the Fourier transform and the properties of the Fourier
transform, which are similar to those of the Laplace transform. In the final section,
we use Fourier transforms to solve PDEs.

EXERCISES 11.3

11.3.1. Find the Laplace transform of the following ODEs:

(1) u"(x) - u'(x) - 4u(x) = 0; u(0) = 1, u'(0) _ -1.
(2) u"(x) - 3u'(x) + u(x) = 0; u(0) = 1, u'(0) = 0.
(3) u"(x) + 2u'(x) - 5u(x) = 0; u(0) = 3, u'(O) = 2.
(4) u"(x) - 4u'(x) + 4u(x) = 0; u(0) = 1, u'(O) = 1.
(5) u"(x) - 2u'(x) + 4u(x) = u(0) = 0, u'(0) = 2.
11.3.2. Show that
e-a/
S

is the Laplace transform of

erfc (i).
11.3.3. Show that
e-a/

is Laplace transform of

-a{erfc
()}.
11.3.4. Show that
e_'
is the Laplace transform of

{erfc (bj
J
 434 Chapter 11: Fourier Integrals and Transform Methods

11.3.5. Show that

be-a1

is the Laplace transform of

_e°'be62t {erfc (bv' + 2 +erfc

J
11.3.6. Prove the following properties about the error function:
(1) erf (-x) _ -erf (x).
z

(2) d (erf (x))) = 2 .

(3) ferf (x) dx = xerf (x) + + c.

2 (1)flx2fltl
(4) erf (x) _ 7= n_o
n!(2n + 1)
11.3.7. Solve the following boundary value problems

(1)
8u x,
at t 2
4a axe t), x > 0, t > 0,
subject to u(0, t) = 4; lim u(x, t) = 0 and u(x, 0) = 10.

au(t,t)
a2a(2,t)
(2) _ ko , x > o, t > o,
subject to u(0, t) = a; lim u(x, t) _ Q and u(x, 0) = y.

auat (x,t)
azax2,t)
(3) 2 , x >0, t >0,
subject to a t = 0; lim u(x, t) = 0 and u(x, 0) _ -40.
(o,

0716 2 , a2u(x,t)
8u(0, t)
subject to ax - a; lim u(x, t) = ,3 and u(x, 0) = 'Y.

(5)
8u x,
at
t 2
aax2, t)
, x > 0, t > 0,

subject to a 0, t) _ t)
= 0 and u(x, 0) = 25.
 Section 11.3: The Laplace Transform and PDEs 435

au(t,t)
a2ax2,t)
(s> = ko , x > o, t > o,
subject to au(0, t) = a' lim au(x' t)
_ ,6 and u(x 0) = ry.
(7x (7x

au(t,t) azaxe'
(7) _ .22
t> >
x >0, t >0,
subject to u(0, t) = 0; lim a t _ -10 and u(x, 0) = y.
(x,

au(t,t)
a2ax2,t)
(a) _ ko , x > o, t > o,
8u(x, t)
subject to u(0, t) = a; lim ax Q and u(x, 0) = ry.

11.3.8. Solve the following boundary value problems:

a2u(x,t) a2u(x,t)
(1) ate -4 axe , x > o, t > o,
au(x, t)
subject to u 0 t = 0 lim = 0 and u x 0 = 0
Du(x,0)
at
--24

'2' a ua 2' t) _2 t) , x > o, t > o,

8u(x't)
subject to u(0, t) = a; lim ax ,Q and u(x, 0) = 0;
Du(x, 0)

(3)
D2u(x,t) - 9D2u(x,t)
t)
subject to = 0; lim au(x, t) = 0 and u(x, 0) = 0;
au(0,

8x 8x
Du(x,0)
at

a2u(x,t) _ 2a2u(x,t)
(4) -C , x > 0, t > 0,
ate axe
au(0, t) au(x, t) =
subject
J to = a lim
x-oo ax
and u x 0)) = 0
ax C

au(x, 0)

D2u(x, t) D2u(x, t)
() Dt2 Dx2
 436 Chapter 11: Fourier Integrals and Transform Methods

subject to au(0, t) = 0 lim u x t = 0 and u x 0 = 0

ax X -4 00
8u(x, 0)
_ -cos(y).
at

a2u(x, t) 2 a2u(x, t
at2
=c ax2
x>0 t>0
,

au(0, t)
subject
J to =a; lim
x- +o0
u(x,t)= and u(x 0)= 0
ax
8u(x, 0)
8t = -vo4.

a2u(x, t) a2u(x, t)
VL" Vim"
8u(0, t)
subject to ax - 0; lim u(x, t) = 0 and u(x, 0) =
8u(x,0) _ 0
at

,o, a2u(x, t) 2 a2u(x, t)

at2
c ax2, x>O t>O , ,

au(0, t) _a
subject to lim u(x, t) _ ,6 and u(x, 0) = f (x);
ax
8u(x, 0)
=
8t
11.3.9. Suppose you have a uniform tightly stretched string from x = 0 to x = L.
rx
Find the displacement of the string if the initial displacement is A sin L
and the initial velocity is 0.

11.3.10. A uniform bar is clamped at x = 0 and is initially at rest. If a constant force

of 10 lbs is applied to the free end at x = 2ir, the longitudinal displacement
u(x, t) of a cross section of the bar is determined from

a2u(x, t) a2u(x, t)
at2 - ax2
<x<2ir,t>0,
subject to

u(o, t = o; au(ax't) = 10 and u(x, o) = o; = o.

o>

Solve for u(x, t) using Laplace transforms.

11.3.11. An infinite porous slab of width 2 is immersed in a solution of constant con-

centration Co. A dissolved substance in the solution diffuses into the slab at a
constant diffusion rate of D. Determine the concentration C(x, t) in the slab.
 Section 11.3: The Laplace Transform and PDEs 437

11.3.12. A flash burn at x = 0 on an uniform rod of semi-infinte length may be de-

scribed mathematically as
au(o, t)
= -AS(t),
8x
where b(t) is the Dirac delta function. Suppose you have an uniform steel rod
of semi-infinte length, which has perfect lateral insulation, and there is a flash
burn of 1200° C (approximate temperature of burning magnesium mixed with
white gun powder) at the end x = 0. Find the time-dependent solution.
11.3.13. Given a semi-infinite perfectly elastic uniform string where the string is ini-
tially at rest. Determine the displacement, u(x, t), if the end x = 0 has a
boundary condition of

sin(irt), 0 < t < 1

f (t) _
0, t > 1.

11.3.14. Solve the boundary value problem

Du(x, t) D2u(x, t)
Dt Dx2

subject to
8u(0, t)
= 100 - u(0, t); lim u(x, t) = 0 and u(x, 0) = 0.
8x X-* o0

11.3.15. Consider a one-dimensional uniform rod of length L, which has no lateral

insulation in a medium of temperature uo. Determine the temperature dis-
tribution in the rod, u(x, t) if the initial temperature in the rod is a constant
ul.
11.3.16. Consider a very long telephone transmission line initially at a constant poten-
tial u0. If the line is grounded at x = 0 and insulated at the far end, find the
potential u(x, t) for any time t. Hint: See Equation (3.15) where L = 0.
11.3.17. PROJECT: A semi-infinite uniform tightly-stretched string is initially at
rest. Suppose the end at x = 0 is fixed at 0 and a concentrated load of
magnitude Fo moves along the string with a constant velocity v, starting at
the point x =0 at t= 0.
(1) Find the displacement of the string for any time t if the force is acting
positive y-direction where v c. Hint: This is a nonhomogeneous wave
equation where the where the external force is give as
x
Fob t--).
v
 438 Chapter 11: Fourier Integrals and Transform Methods

(2) Find the displacement of the string for any time t of the force is acting
negative y-direction where v c.
(3) Repeat part (1) and (2) where v = c.
(4) Suppose Fo = 1. Use mathematical software of your choice to graph the
solution when v = 1 c when v = c when v = 3 c.
2' 2
11.3.18. PROJECT: Find the steady state temperature distribution in a thin uni-
formly insulated sheet which corresponds with the upper half plane where the
temperature is maintained at

0 IxI>ir
and u(x, y) -* 0 as y -p oo. Hint: Use Laplace's equation. Also, there is not
an infinite set of eigenvalues, instead there is a continuous family of solutions.
11.3.19. PROJECT:
(1) Suppose you have a homogeneous uniform one-dimensional rod with per-
fect lateral insulation and its ends, at x = 0 and x = L, in 0 degree baths.
Further suppose that the initial temperature of the rod is ax, for a a real
constant. Show that the temperature distribution in the rod, u(x, t) may
be expressed both by a Fourier series and by using Laplace transforms.
Hint: The Laplace transform should be the infinite sum of the difference
of two error functions, erf.
(2) In a one page paper, discuss which solution you think will converge more
quickly for a short period of time and for a long period of time.
(3) Suppose your rod is made of cast iron and is 10 cm long, calculate the
temperature of the midpoint of the rod when
(a) a = 50, t = 10 seconds, t = 1 min, t = 10 min, and t = 50 min.
(b) a = -50, t = 10 sec, t = 1 min, t = 10 min, and t = 50 min.
In each case, determine if your suspected convergence in Part (2) is cor-
rect.
(4) Suppose your rod is made of cast iron and is 10 cm long, calculate the
temperature of the midpoint of the rod when the initial condition is given
as

x,0<x<5
u(x,0) _
10-x,5<x<10
 Section 11.4: The Fourier Transform 439

11.3.20. PROJECT:'
(1) Find the solution of the semi-infinite cable equation for a neuron
av a2 v
- v, (11.86)
at - axe
subject to the clamped voltage IC
v(x, 0) = 0

with current injection at x = 0

8v(0, t) _ joH(t)
8x
where t, is given in microseconds, RZ is the intracellular resistance, 7m is
in cm, and I0 H (t) is the current injection where H (t) is the heavysided
function.
Hint: Use the identity
2 1 1

s s+l s+l- s+l s+l+ s+l

(2) Show that ast -+o0
v(x t) - Ri mI0e-

11.4 THE FOURIER TRANSFORM

In the previous section, we mentioned that transforms come in pairs. For instance,
the Laplace transform of a function, 1(x), is

G {f(t)} = F(s) = e-st f (t) dt, (11.88)

J
and the Inverse Laplace transform of F(s) is
1 r«+ioo
f (t) = G-1 {F(s)} F(s)est ds. (11.89)
J
The pair concept is very true for Fourier transforms. However, the Fourier transform
pair is somewhat more obvious than the Laplace transform pair. If we consider the

f (x) =
Fourier integral form in Equation (11.32), which is restated here

1
2ir
f(x)=J00

eZexde J d,
'Adapted from James Keener and James Sneyd, Mathematical Physiology, ©1998 by Springer-
(11.90)

Verlag, New York, pp. 261-262. Reprinted by permission.

440 Chapter 11: Fourier Integrals and Transform Methods

then the integral

f f (S)e-ie£'"S
is a function of 0 and is known as the Fourier transform of 1(x). Thus, the Fourier
transform is stated as in the following definition:

Definition 61.

{.f (x)} = F'(9) = f:fei0d (11.91)

Since Equation (11.90) produces the Fourier integral of f (x), the inverse Fourier
transform is given by Definition (62)

Definition 62.

f(x) _ 9O)' ° F'(e)eiex d8. (11.92)

2ir J
Just as in the Laplace transform pair, finding the Fourier transform of a function,
1(x), is much easier than finding the inverse Fourier transform, which involves
contour integration. Contour integration is a standard topic in a complex variables
course and will not be discussed here. However, we do include several examples
where we find and graph the Fourier transform of a function.

EXAMPLE 11.11. Suppose

f(x)= (11.93)
0, Ix>5.
Then, the Fourier transform is

10
3

x
-S

x
-2
2
-3

Figure 11.14: The graph of 1(x) .

Figure 11.15: The Fourier trans-
form, F(B) of f (x).
 Section 11.4: The Fourier Transform 441

F{f(x)}=F(O) =
5 e_iee s
£ t
-ied
-5S e -2B

ei59 _ e-i59 _ sin 58

(11.94)
i8 2 8

where B # 0. For B = 0, F(9) = 10. Figure (11.14) is the graph of the function
f (x). Figure (11.15) is the graph of the Fourier transform, F(9), of the function.
In general, if

11, IxI<a
f(x) _ (11.95)
0, IxI>a,
then, the Fourier transform is
eia9 _ e-ia9 sin a8
(11.96)

where B # 0. For B = 0, F(9) = 2a.

EXAMPLE 11.12. Consider the function
IxI<a
f(x) _ (11.97)
0, IxI>a.
Then, the Fourier transform of f (x) is

.F{S(am)} = F(B) _ f d

f e-e-ie d
-a
a

f -a
e-(1+ie) d

a
_e-(1+ie)
1+2O -a
e(1+i9)a _ e-(1+i9)a

1 + iB
- e 2 i(e-i)°' _ 2511i(e(e 2)). (11.98)
i
442 Chapter 11: Fourier Integrals and Transform Methods

From the previous example, we see that the Fourier transform of a relatively
common function quickly becomes complicated. For your convenience, there is a
table of Fourier transforms in Appendix F. We now consider Fourier cosine and
sine transform pairs.

11.4.1 Fourier Cosine and Sine Transforms

We start by stating the Fourier cosine and sine transform pairs. Then, we use them
on the same function in an example.
From the Fourier cosine integral,

f(x) =
o
2f cos(ex) de f 0
cos(ee) ae, (11.99)

we see that the integral

f() cos(O) d

is a function of 9. This integral is known as the Fourier cosine transform of 1(x).

Equation (11.100) is the manner in which it is usually stated as

{f(x)} = F(e) = f() cos(B) d. (11.100)

0

Note: In Equation (11.100), the variable is changed from u to x, and we use 1c to

indicate that we are taking the Fourier cosine transform. Since Equation (11.99)
produces the Fourier cosine integral of 1(x), the inverse Fourier cosine transform is
given by

1(x) _ {F(O)} _ F(8) cos(Bx) d8. (11.101)

ir 0J
Similarly, the Fourier sine transform is derived from the Fourier sine integral,

J f() sin(0)
1(x) _ sin(9x) d8 (11.102)
J
Again, the integral

f() sin(Be) d
J0
is a function of 8, and it is known as the Fourier sine transform. Its usual form is

{f(x)} = F(B) = f() sin(9e) (11.103)

J0
The inverse Fourier sine transform is given by

1(x) _ 1 {F(9)} _ f F(8) sin(Ox) dB. (11.104)

 Section 11.4: The Fourier Transform 443

EXAMPLE 11.13. Find the Fourier cosine and sine transform of the function, as
shown in Figure (11.16),

1, 0<x<a
f (x) _ (11.105)
L 0, a < x.

We start with the Fourier cosine transform. We have

2.5
2
1.5

0.5
x
4
-0.5
-1

Figure 11.16: Graph of 1(x).

d
1:1(e)
jcos(Oe)
a
d

a sin(a8)
(11.106)
0
e

For the Fourier sine transform, we have

Figure 11.17: The Fourier cosine Figure 11.18: The Fourier sine
transform, F(B) of 1(x). transform, F(9) of 1(x).
444 Chapter 11: Fourier Integrals and Transform Methods

S { f (x)} = F(8) = ff(e)sin(Oe)de

ja
s in(9e) d

CoS(ee) a 1 - CoS(aB)
(11.107)
B o B

Figure (11.16) is the graph of the function 1(x) for a = 4, Figure (11.17) is the
graph of the Fourier cosine transform of 1(x) for a = 4, and Figure (11.18) is the
graph of the Fourier sine transform of 1(x) for a = 4.
We now develop some necessary theorems about Fourier transforms, which we
need when solving PDEs.

11.4.2 Fourier Transform Theorems

Before we can solve any PDEs with Fourier transforms, we must understand certain
related theorems. For instance, what is the Fourier transform of a first, second, or
even nth derivative? Also, do the same properties hold true for Fourier transforms
that hold true for Laplace transforms? These are important questions, and we start
by stating a theorem.
Theorem 63. Given functions f (x) and g(x) whose Fourier transforms exist and
any constants, Cl and c2 we have
' {cl f (x) + c2g(x)} = f (x)} + c21' {g(x)} ,

which means Fourier transforms are a linear operator.

Proof.

. {cif(x) + c2s(x)} = f (cf() + e-iBX d. (11.108)

Since integration is a linear operator, the right side of Equation (11.88) becomes

cl f (_i0 d + C2
g(e)e-Zed
d,
which equals
elf { f (x)} + c2J:'{9(x)}.
D

Fourier transforms like Laplace transforms are a linear operator. Since Fourier
transforms are going to be used to solve PDEs, it would seem reasonable to hope that
they transform a first, second, and nth derivatives similar to Laplace transforms.
We find this out in the next theorem.
 Section 11.4: The Fourier Transform 445

Theorem 64. Suppose 1(x) is a continuous function on the x-axis and 1(x) - 0
as fxf -p oo. Also, suppose f'(x) is absolutely integrable on the x-axis. Then,

F{ f'(x)} = iBJ' { f (x)} . (11.109)

Proof. Since f'(x) is absolutely integrable, we know that a Fourier transform exists.
Therefore, we have

f .fF(e)e-ZB£d

Integrating by parts yields

[f(ee°°° - (-i8) f(e)e° i£

J
f (x) is continuous and f (x) -> 0 as x -> oo implies that

=0.
Therefore,

{f'(x)} = iB f(e)e° i£ d = {f(x)}.

J
0
For second derivatives, we use Theorem (64) twice. This yields

f"(x)} = i9.F{ f'(x)} _ -82.x' { f (x)} . (11.110)

In general, if 1(x) and its n - 1 derivatives, f k (x), k = 1, 2, 3,... ,n - 1 are

continuous functions on the x-axis, and 1(x) and its n - 1 derivatives go to 0 as
I xf - oo, then, if fn(x) is absolutely integrable on the x-axis, we have

(iO)J'{f(x)}. (11.111)

We next state the convolution theorem for Fourier transforms.

Theorem 65. If 1(x) and g(x) are piecewise-smooth on every finite interval on the
x-axis and absolutely integrable, then

1'{f *g}=1'{f}J'{g}. (11.112)

Proof. For our purposes a convolution f * g of the function f and g is defined by

h(x) _ (.f * 9)(x) = f f(y)9(x - y) dy = f f (x - y)9(y) dy.

446 Chapter 11: Fourier Integrals and Transform Methods

Thus, by the definition of a Fourier transform, we have

{f * 9} -L L - y)e-ZB£ dyd.e. (11.113)

Interchanging the order of integration yields

* 9} = Li-: - y)e-ie dedy. (11.114)

Letting s = - y, we have = s + y and Equation (11.114) becomes

00 00
{f * g} = f (y)g(s)e-Ze(s+y) dsdy,

which can be stated as

{f * g} = f (y)e-iey dy
g(s)e-Zes
ds.
i_:
Thus,

F{f*g} =J:{f}J:{g},
which completes the proof. D

Note: The method of convolution is used extensively when solving the heat
equation in an infinite spatial variable domain. If we let J {f} = 1(0) and J {g} =
g (B), then we have

Taking the inverse transform of both sides yields

j(9
(.f * 9)(x) - 1-00
(0)eiOX dB.

Theorem (64) is similar for Fourier cosine and sine transforms, which we state
here for future reference. For the Fourier cosine transform, we have

{f'(x)} - e.'Fs {f(x)} -f(o) (11.115)

and

1c {f"(x)} = 02c {f(x)} - fF() (11.116)

For the Fourier sine transform, we have

1'S {f'(x)} _ {f(x)} (11.117)

 Section 11.4: The Fourier Transform 447

and

1s {f"(x)} = {f (x)} + Of (O). (11.118)

In the next section, we investigate solving PDEs using Fourier transforms.

EXERCISES 11.4

11.4.1. Compute and graph the Fourier transforms of the following functions:
2, f0,
(1)
{
5<x.
e-2x1,
xf < 1
(2) 1(x) {
0, otherwise.

e-x, x>0
(3)
{
0, otherwise.

ex, x<0
(4) f(=) {
0, otherwise.

xe-2x, x > 0
(s)
{
0, otherwise.

-1, -1<x<0
1, 0<x<1
0, otherwise.

1
!]1 flml -
x2+25
x, 0<x<2
(8) f (x) = 2x-4, 2<x<4
0, otherwise.
 448 Chapter 11: Fourier Integrals and Transform Methods

11.4.2. Compute and graph the Fourier cosine transforms for the following functions:

(1)

(2) f(x) =

(3) f(x) =x-112

(4) f(x) = x-4

3
(s)
x2 +9 .

x, 0<x<1
(6) J(=) { 0, otherwise.

x, 0<x<2
(7) 2x-4, 2<x<4
0, otherwise.

11.4.3. Compute and graph the Fourier sine transforms for the following functions:

(1)

0, otherwise.

(2) 1(x) = ex2

(3) 1(x) = x112.

(4) 1(x) = x.
 Section 11.4: The Fourier Transform 449

x
(s)
x2 +9 .

x, 0<x<1
(6) f(x)= {
0, otherwise.

x, 0<x<2
(7) 2x-4, 2<x<4
0, otherwise.

-1, 0<x<1
(8) f(x)= 1, 1<x<2
0, otherwise.
11.4.4. This exercise develops the Fourier transform of a shift of a function on the
x-axis. Show that if 1(x) has a Fourier transform, then f (x - a) for some
a E 1R, and that

{.f (x - a) } = e-aaeJ {f (x)

11.4.5. This exercise develops a shift on the 0-axis. Show that if f(0) is the Fourier
transform of 1(x), then f(0 - a) is the Fourier transform of eiax f(x)
11.4.6. This exercise develops the basic properties of convolutions. Prove the following
properties of convolutions:
(1) Commutativity of convolutions: f * g = g * f.
(2) Associativity of convolutions: f * (g * h) = (f * g) * h.
(3) Commutation with translation: Given a E R, let fa (x) = f (x - a). Show
that fa * g = f * ga = (f * g)a.

11.4.7. Let 1(x) = e-x2/2 and g(x) = e-I xI

(1) Find the Fourier transform of 1(x) and g (x) .
(2) Find the Fourier transform of the convolution of f and g.
(3) Let a = 2. Find the Fourier transform of fa * g. Hint: Reference the
previous exercise.
11.4.8. Another definition of the Fourier transform is

f(0) _ {f(x)} = (11.119)

2J .f (e)e-ZB de,
 450 Chapter 11: Fourier Integrals aid Transform Methods

and of the inverse Fourier transform is

.f (x) _ -1 {f(e)} = f J(e)eiex dB. (11.120)

27f

(1) Show that the above two definitions are equivalent to Definitions (61 and
62).

(2) Using Definition (11.119) find the Fourier transform of e-ac2 for a> 0.
Compare your answer with that in the table of Fourier transforms in
Appendix F.
(3) Using Definition (11.119) find the Fourier transform of

1, ff(x)

_
0, f

Compare your answer with that in the table of Fourier transforms in

Appendix F. Explain the differences.

11.4.9. Show that the inverse Fourier transform is a linear operator.

11.5 FOURIER TRANSFORM SOLUTION METHOD

OF PDES
This section deals with PDEs, which are functions of two variables, x and t. Again,
we note that a Fourier transform is designed to work with one variable. However,
we learned in Chapter 6, The Calculus of Fourier series, that we could take the
derivative of a Fourier series with respect to a parameter t. We will use that same
type of technique here, and again the parameter is t. Thus, when using Definition
(11.91), the definition of the Fourier transform of a function of two variables, u (x, t)
becomes

{u(x, t)} = u(B, t) _ u(e, t)e-ie£ d.

We demonstrate this new definition with an example.

EXAMPLE 11.14. Suppose

et sin x, IxI<ir
u(x, t) _
0, otherwise.
 Section 11.5: Fourier Transform Solution Method of PDEs 451

Then, the Fourier transform is

u(B, t) _ .F{u(x, t) } _ F{et sin x}
00
sine-Zed
et d
too
sine-Zed
et d

sin
et
J_lr
sine- Ze
d = 2ie t e2 -
1

sin (ir0)
Thus u (9, t ) = 2iet 82-1 .

If we assume Theorem (64) holds true for the Fourier transform of a PDE, which
au(x, t) a2u(x, t) a2u(x, t)
it does, then for , ate , and axe where - oo <x < oo and t > 0
at
we have the Fourier transforms;
f8u(x,t)
at } (11.121)

. f82u(x,t)
ate }
__ d2
dt2
u(8, t), (11.122)

and
a2(, t)
, ' ax2 = (i9)21i(9, t), (11.123)

respectively. Also, we have

, anu(x, t) = (i9)ii(9,t). (11.124)

The method for solving PDEs by Fourier transforms is quite similar to that of
Laplace transforms. We will use the notation 11(0, t) introduced in this section to
simplify the equations. Thus, we have the following steps:
(1) Given a PDE with infinite boundaries and initial conditions, take the Fourier
transform of the infinite boundary value problem while holding t as a para-
meter. You will get an ODE in the variable t, with function 11(0, t). Note:
The ODE only depends on the variable t.
(2) Solve the ODE and find 11(8, t).
(3) Take the inverse Fourier transform of u(B, t) to find u(x, t).
452 Chapter 11: Fourier Integrals and Transform Methods

Example (11.15) illustrates the use of the previous steps.

EXAMPLE 11.15. Given a uniform infinite rod with perfect lateral insulation,
thermal diffusivity of 1, and initial temperature distribution of solve the fol-
lowing problem using Fourier transforms.
First, we must state the mathematical formulation of the problem. We have

au(x, t) _ a2u(x, t)
<x < oo and t > 0, (11.125)
at axe ' - °O

subject to

u(x, 0) = f(x) = (11.126)

Step 1. Take the Fourier transform of the entire problem, which is from Equations
(11.121 and 11.123) and Equation (11.91),

du(9, t) = -9211(9,t),
(11.127)
dt

.F{u(x, o)} = u(9,0)

00
e-2 e-2 e-ie
= = d

= (11.128)
1(9),

where the Fourier transform of the IC was found in Appendix F. Although the
Fourier transform of the IC is 1 e-92'4 the preferred form is f(9). This form
becomes important when we must do the inverse Fourier transform.
Step 2. Equation (11.127) is a first-order ODE in t. Its solution is

i(O, t) = A(9)e_02t. (11.129)

Applying the IC, Equation (11.128), to Equation (11.129) yields

u(B, 0) = (11.130)

Therefore, our solution is

u(8, t) = e-e2/4e-92t
= j(9)e_02t. (11.131)
4r
 Section 11.5: Fourier Transform Solution Method of PDEs 453

Step 3. Find the inverse Fourier transform of Equation (11.131), which is, by
Equation (11.92),

100 u (9, t) ed8

2ir 00

1001 e-e2/4e-e2teiex d8
2ir 1
°°
1
f (e)eJ-e2t eZee d8. (11.132)

If we let g(8) = e_e2t in Equation (11.132), then we have

u(x, t) _ j(O)(O)eiOx de. (11.133)

1-00
Equation (11.133) may be recognizable as a form of Equation (11.119), which is

(f * g)(x) =
f j(O)(O)ei0X dO.

By the definition of convolution,

(.f * 9)(x) = f 00
f(y)9(x - y) dye (11.134)

we must find the inverse Fourier transform of (O). Note that

ff 1 e-02/4a
l I Ta
If we let t = 1
4a'
then a = 1
4t
. Then, we can solve for the inverse Fourier transform
'
of (O). We have

./
{_x2/4t} = t o-92t
= t o

Thus, the inverse Fourier transform of g(8) is

t
Replacing x with x - y in Equation (11.134), we get the final answer of

u(x> t) _ (f * 9)(x) - f .f y) t 2/4t Cry

1too
f (y)e2' dy

1 e-y2e-(x-y)2/4t dy. (11.135)

2 art J_
454 Chapter 11: Fourier Integrals and Transform Methods

In integral equation theory, g(8) = e-e2t, in Example (11.15) is known as a

kernel. In this case, it is typically known as the heat kernel or Gauss's kernel. The
Gaussian is the typical bell-shaped curve well known in statistics and probability.
In general, the solution of the heat equation on an infinite domain, - oo <x < oo,
with an initial temperature distribution, f (x), is the convolution of the heat kernel,
e-e2t
g(O) = with the initial temperature distribution, f (x). Thus, given the general
heat equation
8u(x y) 82u(x t)
' =k0 '
at axe
subject to the initial temperature distribution of

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

our solution is

(y)e--y)2/41cot dy. (11.136)

u(x,t) = 2/] ot J
1 .f

Our next example will make use of Equation (11.136) and the error function,
which was introduced in Section 11.3, Laplace transforms.
EXAMPLE 11.16. Consider a uniform infinite rod with perfect lateral insulation,
thermal diffusivity of iron, and initial temperature distribution of 30°C for x <1
and 0 otherwise. Solve the problem using Fourier transforms and the error function.
First, we must state the mathematical formulation of the problem. We have

au(x, t) _ a2u(x, t)
-. 22 axe - oo < x < oo and t > 0, (11.137)

subject to

(30, x<1 1
u(x,0) = f(x) = (11.138)
0, x>1.1.
Using Equation (11.136),

(,) 2 kot -00

f (y) y,

we know that the solution as an inverse Fourier transform is

[00
1
dy
2 r.22t _00

15 f1
e y (11.139)
22t
r.22t f_i
 Section 11.5: Fourier Transform Solution Method of PDEs 455

Note: The integral is from -1 to 1 since the integral is 0 for x > 1.

If we let z = x y then dz = 1 dy = 1 dy. Also the limits of integration
.88t' .88t 2 .22t
change to x 1 and x + 1 Thus, Equation (11.139) becomes
.
2 .22t 2 .22t

u(x, t)
15 2x-1ti e_z2
-2 .22t dz
r.22t 2 vi x+1

x+1
30 2 vi 2
Ie_Z dz
2 .22t

x+1 x-1
30
o
2 .22t
e-z dz
2
- o
2 .22t
e-z2 dz

15 ( erf
x+1
erf
[x_1l\ (11.140)
L2vi
.22t 2 .22t

Figure (11.19) graphs the temperature distribution, u (x, t) for several values of t. U

t=( 2

r = 1.6

15
t=3.2
\ 10

y 5

Figure 11.19: Temperature distribution, u(x, t), for several values of t.

Our last example for this section is the wave equation.

EXAMPLE 11.17. Consider a tightly stretched uniform one-dimensional string
of infinite length. Suppose the initial displacement is f (x) = 1 1 and the initial
x2
velocity is g(x) = 0. Determine the solution as an inverse Fourier transform.
The mathematical formulation of the problem is
a2u(x, t) a2u(x, t)
at2
=
a x2
, -oo<x<oo,t>0 (11.141)
456 Chapter 11: Fourier Integrals and Transform Methods

subject to
1

' 1+x2
au(x, 0)
at
=0 (11.142)

Step 1. Find the Fourier transform of Equations (11.141 and 11.142). We have
d2u(B,t)
_ -92 (9,t), (11.143)
dt2
subject to
.F{u(x, o)} = u(e, o)

{1X2} -,f 12e -zeE d

1 1

ire0 = J(9). (11.144)

.l
and
J aunt, o 1 _ (9). (11.145)
1
Step 2. Solve the ODE in t subject to the ICs, f(9) and (9). Equation (11.143)
is a second-order ODE that has the solution
u(e, t) = a(e)cos(et) + B(e)sin(9t). (11.146)

Applying the ICs to Equation (11.146), we arrive at A(B) = f(9) and B(B) = 0.
Therefore, Equation (11.146) becomes
u(B, t) = f(9) cos(Bt) _ ire-0 cos(9t). (11.147)

Step 3. Find the inverse Fourier transform. We have

1
u(8, t)eZex d8
2ir
00
1
ire-0 cos(9t)exe d8
2ir f00

= - Looe_0cos(0t00* (11.148)

Equation (11.148) expresses the solution in terms of an inverse Fourier transform.

Many times these expressions can be evaluated. However, the inverse Fourier trans-
form solution of wave equation with infinite boundaries may be manipulated into
d'Alembert's solution. This is left as an exercise.
 Section 11.5: Fourier Transform Solution Method of PDEs 457

EXERCISES 11.5

11.5.1. Solve the following heat problems using Fourier transform method:

(1)
t) _ 82ua( 2, t)
-Do <x < oo and t > 0, subject to

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

(2)
t> _ t)
<x < oo and t > 0, subject to

-25, -1<x<0
u(x,0) = f(x) =
25, 0,x<1
0, 1<x.
au(x, t) _ a2u(x, t)
(3 ax2 ' -oo < x < oo and t > 0, subject to
at
0, x<-1
u(x,0)=f(x)= x, -1<x<1
0, 1<x.
au(x, t) _ 4a2u(x, t)
(9) -Do < x < oo and t> 0, subject to
at 8x2 '

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

11.5.2. Solve the following wave problems using Fourier transforms:

a2u(x,t) - a2u(x,t) ' -00 < x < 00, t> o,

(1)
ate - axe u(x,0) 1
subject to u(x, 0) = 0 and a
at - 1 + x2
 458 Chapter 11: Fourier Integrals and Transform Methods

a2u(x,t) = a2u(x,t)
(2) - < <, >o,
sin(irx) au(x, 0)
subject to u(x, 0) = and =0.

(3) a2u(x,t) - a2u(x,t)

subject to u(x, 0) = 0 and aunt' =

(4) a2u(x, t) - a2u(x, t) -oo <x < oo, t > 0,

8t2 8x2
subject to u(x, 0) =
1 + x2
and auk, 0) -
11.5.3. Consider
a2u(x, t) 2 _2u(x, t)
ate - ax2 '

subject to

u(x, 0) = f (x) and aunt' = g(x).

(1) Solve for u (x, t) using the Fourier transform method.

(2) d'Alembert's solution may be derived from the answer to part (1).

11.5.4. Solve the diffusion equation with convection. That is, solve

au(x, t) _ a2u(x, t) au(x, t)

k° axe +h -00<x<oo,
at 8x
subject to

u(x,0) = f(x).

11.5.5. Solve the linearized Korteweg-deVries equation. That is, solve

au(x, t) a3u(x, t)
at C7x3
' - 00 <x < 00,
subject to

u(x,0) _
0, IxI>2.
 Section 11.5: Fourier Transform Solution Method of PDEs 459

11.5.6. Solve by Fourier transforms and graph the solution of

a2u(x, t) a2u(x, t)
atax = axe - OO< x < oo,
subject to
u(x, 0) = e-ICI.

11.5.7. Consider
a2u(x, t) a4u(x, t
t2 - 00 <X <00,
8x4
subject to
1
u(x, 0) _
1+x2
(1) Find the Fourier transform of the boundary value problem.
(2) Solve the resulting ODE.
(3) Justify a reasonable assumption about the boundedness of the solution
of the ODE.
(4) Apply the IC.
(5) Give u(x, t) in the form of an inverse Fourier transform form.
11.5.8. Solve
82u(x, t) x,t) - oo <X <00
ax4
8t2 -a 8t82x -b
subject to
0)
u(x, 0) = f (x) and aunt' = g(x)

for u(x, t) and leave the answer in inverse Fourier transform form.
11.5.9. Given
au(x, t) _ a2u(x, t)
at
ko x2 - oo < x < oo and t > 0,

subject to
ul/x OJ ` _ fl/xJ ` _ e-amt

show that the solution is

u(x, t) = 1 e-O'x2/9(t) where g(t) = 4akot + 1.

g(t)
 460 Chapter 11: Fourier Integrals and Transform Methods

11.5.10. Solve the following variable-coefficient PDEs using the Fourier transform method:

1au(x, t) - _tau( x, t) subject to u(x, o)

e-x, x > 0
at ax >
{
0, x<0.
au(x, t) _ au(x, t)
(2) , subject to u(x, o) _ cosx 2 .
at ax

8u(x, t) _ t8u(x, t)
(3) , subject to u(x, 0) =sin x2.
at ax

au(t,t) auax,t)
- Sin t , subject to u(x, 0) = sin x2.

au(x, t)
= cost
au(x t
lU/ , subject to u(x , 0) = cos x2 .
at ax

ki
au(x,t) - _tau(x,t)
at axe
subject to u(x,0) = 256.

11.5.11. PROJECT: Solve the Klein-Gordon equation in one dimension. That is,
solve

8 u x, t)
a 2 - 2a
ax2,t)
- m2u(x, t), - Do <x < oo,

subject to
0)
u(x, 0) = 0 and
au(x,

8t
Note: The Klein-Gordon equation is one of a number of equations which
model elementary particles; electrons, mesons, quarks, etc. where m is the
mass of the particle
a 2u(x, t) 2 x,t
(1) Find the Fourier transforms of u( x, t) , and the ICs.
ate '
Note: The Fourier transform of u(x, t) is 11(8, t).
sin tc2B2 + m2
(2) Solve the resulting ODE and apply the ICs. Hint: 11(9, t) _
c2B2 + m2
(3) Setup the inverse Fourier Transform.
(4) From the theory of cylindrical functions it is known that
sin r 1
Jo (r sin sin T) ear cos cos T
sin T dT
r 2 o
 Section 11.5: Fourier Transform Solution Method of PDEs 461

(5) Make the following substitutions:

r2 = t2 (c2e2 + m2), r sin = imt, and r cos = cOt

(6) After working this part of the problem, you should arrive at the following
answer :
2
u(x,t) = 2c Jo m t2 - 2 for x < ct.

11.5.12. PROJECT: Solve the nonhomogeneous Klein-Gordon equation. That is,

solve
82u(x, t) __ 82u(x, t)
ate c2 axe - m 2u(x, t) -}- xt, - oo <x <00,

subject to

u(x, 0) = 0 and aunt' =0.

Hint: Follow the steps of the previous problem.
Appendix A

Summary of the Spatial

Problem
In this appendix, we give a summary of the solutions of the spatial problem for
homogeneous Dirichlet and Neumann boundary conditions. The eigenvalues and
corresponding eigenfunctions are given for each of the four possible sets of boundary
conditions.

463
464 Appendix A: Summary of the Spatial Problem

Given

-Ao(x),
8x2
we have the following results:

Boundary Conditions Eigenvalues Eigenfunctions

(flit z

-lLl pn (x) = sin L

n7rx

n=1, 2, 3,...

CA
(2n- 1)
2L (2n - 1) 7rx
cpn (x) =sin
2L
n = 1,2,3,...

ag(o) = o ((2n_1)ir
2L 2n- 1) 7rx
pn(x) = cos
2L
pSp(L) = 0 n=1,2,3,.

Ao = 0 p0(x) _ Co
( z
=
L)
0p(L)
(L)
Ox
- f0
= n = 1, 2, 3,. p(x) = cos
L

p0(x) _ Co
( p(-L)=cp(L)
(flit 2
cosnitx
L
&p(-L) &p(L)
L)
ax ax n=1,2,3,. pn(x) =sinnix
L
Appendix B

Proofs of Related Theorems

Basically, the flow of this appendix is the statement of the theorem and then the
proof, with only a few comments. Any assumptions about your prior knowledge of
these theorems are given before the statement of the theorem to be proved.
There are five theorems in the text that have proofs in this appendix. Two
are from Chapter 2, two are from Chapter 4, and one is from Chapter 10. The
appendix is sectioned according to chapter and sectioned within chapter according
to theorem.

B.1 THEOREMS FROM CHAPTER 2

B.1.1 Leibniz's Formula
For the proof of Leibniz's formula, I assume that you are familiar with various
theorems from calculus, namely, the fundamental theorem of integral calculus from
Calculus I, and the integral interchange theorem from Calculus III. Also, I assume
that you understand the concept of integration depending on a parameter, which
you may have encountered in Calculus III.
a f (x'
Theorem 66. Suppose f (x, t) and the partial derivative t) are continuous in
at
some region of the xt-plane where a < x < b and c < t < d, then
b b

do
f f (x, t) dx
a = Ifa f att dx.

9(t) = f b of at't dx, c < c < d.

a

Since
a f (x'
at
t) is continuous, we know that g() --
t is continuous for c < t < d. Thus

465
466 Chapter 2: Proofs of Related Theorems

for c < t <d, we have

Idg(t) dt = f d f a6 of ate t) dx dt.

c

Therefore, by the integral interchange theorem, we may switch the order of integra-
tion in Equation (B.1). We have
d d b of (x, t) b cd
of (x, t)
g(t) dt = dx dt = dt dx
I c a at a at
b b b
[f(x,d)-f(x,c)] dx = f (x, d) dx - f (x, c) dx
a a a
= F(d) - F(c),
where F(d) and F(c) are integrations depending on a parameter. If we let d be a
variable t, we have
t
F(t) - F(c) = g(u) du. (B.2)
c

Equation (B.2) can now be differentiated with respect to t. Hence, by the funda-
mental theorem of integral calculus we have

fb af(x,t)
F'(t) = g(t) = dx.
Ia at
Note:

F' (t) = - fbf(x,t)

d
dx.
dt
Thus, Liebnitz's formula is proved.

B.1.2 Maximum-Minimum Theorem

Suppose we are given
au(x, t) _ ka2u(x, t)
(8.3)
at axe
subject to Dirichlet BCs
u(0, t) = g(t)
(8.4)
u(L, t) = h(t)
and IC

u(x,O) = f(x). (B.5)

 Section 2.1: Theorems from Chapter 2 467

Equation (B.1) tells us that for 0 <x <L and 0 <t

8u(x t) 82u (x t) _
8t
- k Dx2 ' 0. - (B.6)

We want to show that the maximum-minimum value occurs on the sides of the
rectangle, R, given by
0<x<Land0<t<T
as shown in Figure B.1. Since k is a positive constant in Equation (B.6), we will

0 L

Figure B.1: u(x, t) in the rectangle 0 < x < L and 0 < t < T.

assume that it has the value of 1. Therefore, Equation (B.6) becomes

8u(x, t) 82u(x, t)
=0 . (8.7)
at axe
The proof of the maximum-minimum theorem is done by contradiction. Also,
the proof of the minimum part is provided. The maximum part is similar.
Theorem 67. (Maximum-Minimum Theorem) Let T be an element of the real
numbers such that T> 0. Suppose the function u(x, t) is continuous in a rectangle,
R, given by
0<x<L and0<t<T,
as shown in Figure (B.1), and satisfies the heat equation given in Equation (B.3) in
the interior of the rectangle. (Note: T is considered in the interior of the rectangle,
since any Tl > T world imply T is in the interior.) Then, u(x, t) attains its
maximum or minimum on the base of the rectangle t = 0 or on the vertical sides of
the rectangle x = 0 or x = L.
Proof Let m be the minimum value of u(x, t) in R and suppose that the minimum
value of u(x, t) on the lower base and vertical sides of R is m + E where f> 0. Let
468 Chapter 2: Proofs of Related Theorems

(XO, to), where 0 < xo <L and 0 < to, be a point where u(x, t) attains minimum in
R. Thus, u(xo, to) = m. Consider the function
(x,t) = u(x, t) - 4L2 (x - xo)2 . (B.8)

Note:
aµ(x, t) 8u(x, t
at - at
(B.9)

and
a2µ(x,t) _ a2u(x,t) _ E
.
axe axe 2L2
On the lower base and vertical sides of R,

x t >m+E- E =m+-4

(B.11)

whereas ,a(xo, to) = m. Therefore, the minimum value of ,u in R is not achieved

on the lower base and vertical sides of R. Let (Xi, ti), where 0 < x 1 < L and
0 <t1 <T, be a point where ,u attains its minimum. At (x1, t1), ,u must satisfy the
necessary conditions for a minimum, that is,
Dµ(x1, t1)
at
-o

82µ(xl,t
8x2
Hence, at (x1, t1),
aµ(xi, ti) a2µ(xi, ti) o.
at 8x2
However,
aµ(xi, ti) a2µ(xi, ti) aµ(x, t)
at - axe - at
au(x, t) a2µ(x, t)

a2u(x, t) E
= axe + 2L2 > 0.

Thus, we have a contradiction. Therefore the minimum value of u(x, t) must be

attained on the lower base or vertical sides of the rectangle R.
 Section 2.2: Theorems from Chapter 4 469

B.2 THEOREMS FROM CHAPTER 4

You may not have previously encountered the next theorems in this form, in a
linear algebra course. In your ODE course, you may have seen something similar.
However, this material, in an ODE course, is concentrated on solving systems of
linear ODES.

B.2.1 Eigenvectors of Distinct Eigenvalues

Are Linearly Independent
The proof of Theorem (68) is done by induction. For those unfamiliar with this
type of proof, I suggest you study a text on logic and proof.
Theorem 68. Let a1, A2, A3,... , an be distinct eigenvalues of an n x n matrix.
xxx
Then the corresponding ei9envectors 1 , 2 , 3 , . xnform a linearly independent
set of vectors. That is, c1 xl + C2 x2 + c3 x3 + ... + cn n = O if and only if c2 = O for
1 <i <n.
Proof. Let A1, A2, A3,... , an be distinct eigenvalues of an n x n matrix. We know
that for each eigenvalue, there exists a corresponding eigenvector, usually expressed
as xi , x2 , x3 , ... n . For n = 1, we have one eigenvalue and one eigenvector and the
result follows from normal algebra on the real numbers R. Now assume that every
set of (n -1) eigenvectors, which correspond to the (n -1) distinct eigenvalues of a
given square matrix, A, are linearly independent. Let xi , x2 , x3 , ... , xni be the
eigenvectors of A corresponding to the distinct eigenvalues A1, A2, A3,... ,
Thus, ai A3 3 for i xx x
j. Suppose that 1 , 2 , 3 ... , xn-1 x aren linearly
dependent. Thus, there exist constants ci for 1 < i < n that are not all equal to 0,
such that
C1xx1 + c2x2 + C3x3 + ... + Cnxn = 0. (B.12)

We assume that c1 0. Now,

0 = (A - 1In) (C1 1 ' C2 x2 ' C3 3 ' ... -}' Cnxn)

= c1(Ax1 -A1i)+c2(A 2 -A1 2)+...+cn(Axn-A1 n)

= C1 (A1 1 - Al 1) + C2 (Ae2 - Al x2) + ... + Cn (Axn - Al ) .

Since

C1 (A 1 - Al 1) = C1 (\1 1 - \1 1)
by Definition (15), we have

Q = C2 (Ax2 - 1 x2) -}- ... -}- Cn (Axn - 1xn) .

470 Chapter 2: Proofs of Related Theorems

But A2 through an are (n -1) distinct eigenvalues, so by the induction hypothesis,

x2 , x3, ... , n are linearly independent. Therefore,

c2(Ai-A2)=...=c(Ai-A2).
But each ai - A3 0. Hence, c2 = c3 = ... = cn = 0. Thus, Equation (B.12)
becomes

cl x1 = 0 ,

which implies xi = 0 since c1 0. This is a contradiction. Therefore, the eigen-

vectors xi , x2 , x3 , ... , n are linearly independent.

B.2.2 Eigenvectors of Distinct Eigenvalues of an n by n

Matrix Form a Basis for Rn
Generally, Theorem (69) is considered a corollary or lemma of Theorem (68). As
such it has a very short proof.
Theorem 69. If an n x n matrix A has n distinct eigenvalues, then the correspond-
ing eigenvectors form a basis for W'.
Proof. Let xi , x2, x3, ... n be the corresponding eigenvectors of the distinct eigen-
values A1, A2, A3,. . . an of the n x n matrix A. By Theorem (68), we know that the

eigenvectors are linearly independent. We must show that they span W'. Let y
be any vector in W'. Then, the set of vectors {T' x2, x3, ... n , y } is linearly
dependent because each vector is a linear combination of any basis vectors of W'.
Thus, the constants c1, c2, c3,... , cn, and c exist where at least one is not 0, such
that
c1 x 1 -}- c2 x2 -}- c3 x3 -}- ... -}- cn n -}- c y = O. (B.13)

Since xi , x2, x3, ... , n are linearly independent, we must have c 0 in Equation
(B.13). Therefore,

c1xxi+c2x2+c3x3+...+cn n=-cy.
-4 -4 -4
Since y was an arbitrary vector of W', x1, x2 x3,... n span Rn . Hence,
2 22
, , , n is a basis of R.

B.3 THEOREM FROM CHAPTER 5

Since there is only one theorem from Chapter 5, we will not separate this section
into subsections.
The maximum principle has two parts-maximum value occuring on the bound-
ary or a minimum value occuring on the boundary. We prove the minimum value
occuring on the boundary. The proof of the maximum value occuring on the bound-
ary is similar.
 Section 2.3: Theorem from Chapter 5 471

Laplace's equation in two dimensions is

a2 u a2 u
axe + a y2 = o.
(B.14)

Theorem 70. (Maximum Principle) Let 1 be a bounded set in l2. Let u(x, y) be a
harmonic function in 1, while u(x, y) must be continuous in the union of 1 and the
boundary of 1, denoted 91k. Then, the maximum and minimum values of u(x, y)
are attained on the aSZ, unless u(x, y) is identically equal to a constant.
Proof. Let E > 0 be given. Let µ(x, y) = u(x, y) - E (x2 + y2). Then,

a2µ(x, y)
ax2 +
a2µ(x, y)
ay2
_
_
a2u(x, y)
ax2 +
a2u(x, y)
aye
- 4E=0-rE<0.

But a2µ(x,y)
axe + a2µ(x'y) > 0 at an interior minimum point, by the second
- deriv_

a y2
ative test from calculus. Thus, µ(x, y) has no interior minimum in f Since µ(x, y)
is a continuous function it must have a minimum in SZ U as Let the minimum of
µ(x, y) be attained at (xO, yo) on %. Then, for all (x, y) in 1 we have
u(x, y) > µ(x, y) > µ(xo, yo) = u(xo, yo) + (x2 + y2) > min u + Er2,
asp

where r is the radius of the circle centered at the origin that contains SZ U of Since
this is true for all f > 0, we have
u (x, y) > min u
asp

for all (x, y) in ft

Appendix C

Basics from Ordinary

Differential Equations
C.1 SOME SOLUTION METHODS FOR FIRST-ORDER
ODES
A first-order ordinary differential equation (ODE) has the form

S(t) d dtt + K(t)u(t) = F(t).

If F(t) = 0, then the equation is homogeneous. For the homogeneous case, u(t) - 0
(where - means identically equal to) is the trivial solution. For a solution to exist,
it is important that S(t), K(t), and F(t) be continuous functions on some interval
a <t <b where the ODE is defined.
If S(t) is not equal to 0 anywhere on the interval a <t <b, then Equation (C.1)
may be written in the more familiar form as Equation (C.2),

u'(t) + k(t)u(t) = f(t). (C.2)

Equation (C.2) is said to be linear since u'(t) and u(t) are linear. Therefore, let
us first consider Equation (C.2) where k(t) is a constant and then complete this
section by discussing Equation (C.2) where k(t) is a more interesting function.

C.1.1 First-Order ODE Where k(t) Is a Constant

The following are two examples of commonly occurring first-order ODEs. These
particular ODEs are very important when solving PDEs where k(t) is a constant.
EXAMPLE C.1. Consider
u '(t) = au(a) (C.3)

473
474 Appendix C: Basics from Ordinary Differential Equations

where A is a constant. Note: In Equation (C.3), the function, 1(t), is the constant
0. Assuming u (t) 0 for any t, we can rewrite Equation (C.3) as

(c.4)

You arrive at the solution by remembering that

ddtt)
u'(t) __
A'
ut ut
which can be written as

u(t) _ Mt.

Assuming that u(t) > 0, we can integrate. The solution is

ln[u(t)] _ At + a.
Now remembering some basic calculus, we arrive at the more compact and useful
solution to Equation (C.4), which is

u(t) =cert.
EXAMPLE C.2. Consider
u'(t) + au(t) = A. (C.5)

Note: In Equation (C.5), the function f (t) is the constant A, and a is a nonzero
multiplicative constant of u(t). We solve Equation (C.5) using the usual technique.
That is, we find an integrating factor. Therefore, we must determine a function g(t)
such that when you multiply Equation (C.5) by g(t), the left side is recognized as a
derivative of a product. Many times you can quickly discover the function g(t) by
remembering your calculus. In this case, if we choose g(t) = eat, we can form the
product eatu(t). When we then take the derivative of this product, we have

dt
[eatu(t)] - eatul(t) + ae°'tu(t) = eat [+ au(t)]. (C.6)

The left side of Equation (C.5) and the contents of the brackets on the right side of
Equation (C.6) are equal. This means that we have the correct integrating factor.
Thus, we can multiply both sides of Equation (C.5) by eat and get

eat [u' (t) + au(t)] _ Aeat.

Making use of the work we did to produce Equation (C.6), we have

eat [uI (t) + au(t)] _ [u(t)eat]I

 Section C.1: Some Solution Methods for First-Order ODES 475

resulting in
[u(t)eat]I = Aeat.

Integrating both sides with respect to t yields

Aeat
u(t)eat = + c.
a

And, finally, solving for u(t), we find that

A
u(t) = ce-at
+ -. U
a

In both examples we have an arbitrary constant c. When c is unknown, we have

a general solution to the ODE and there exists a family of curves as solutions. For
example, if we let A = 1 and a = 12 in Example C.2, then the following graph
(Figure (C.1)) depicts some members of the family of curves that solve Equation
(C.5).

u(t)

Figure C.1: Plot of ce 2t + 2 for c = -2, -1, 1, 2.

In most applications, the solution u(t) is known for some value of t = to where
u (to) = no. This is generally called an initial condition (IC). The IC actually
describes a point (to, u0) through which the solution curve passes, resulting in a
specific solution. For example, if in u(t) = ce-at + - (Example C.2) we had A = 1
a
a = 1 and an IC of n(O) = 1 then our solution would become n(t) = -e 2t + 2.
This is graphed in Figure (C.2).

C.1.2 First-Order ODE Where k(t) Is a Function

In this section, we consider solution methods for the first-order ODE Equation (C.2)
where k(t) is a continuous function on some interval a < t < b where the ODE is
defined.
476 Appendix C: Basics from Ordinary Differential Equations

u(t)
2F

-4 2 4
-2

-4

-6

-8

-10

Figure C.2: Plot of -e 2t + 2.

EXAMPLE C.3. Consider the following ODE subject to the stated IC:
u'(t) - 2tu(t) = t, u(0) = 1. (C.7)

There are three ways of solving this ODE. First, we could use separation of variables
and solve the related problem

u'(t)
1 + 2u(t) t
A second method is to treat the original problem as nonhomogeneous. In this case,
we solve the homogeneous equation uh (t) and then solve for a particular solution
up (t) . Adding uh (t) and up (t) yields the solution for u (t) . Although this method
works, I do not recommend it, because we solve the same problem twice and do
twice the amount of work.
The third method is the integrating factor method. In Equation (C.7), we have
k(t) = -2t, which suggests that an appropriate integrating factor is f (t) = e-t2.
To prove this, consider
d
dt
e-t2u(t) = e-t2u'(t) - 2te-t2u(t) = e-t2 [u'(t) - 2tu(t)].

As in the previous subsection, we see that f (t) = e-t2

is the desired integrating
factor. Now following our process used in Example C.2, we form
e-t2 to-t2 ,
[u' (t) - 2tu(t)] =

which becomes

[e_t2u(t)] = to-t2.

Integrating both sides, we find

u(t)e_t2 l _t2
 Section Cl: .Some Solution Methods for First-Order ODES 477

And then solving for u(t) yields

1 t2
u(t)=-2+ce ,

the general solution. Using the given IC u(0) = 1, we find c = 2. Therefore, the
specific solution of Equation (C.7) is
3 t2 1
u(t) _ 2e -,
which is graphed in Figure (C.3).

u(t)

1.4

1.2

0.8

0.6

0.4

0.2

-
t
0.1 0.2 0.3 0.4 0.5

3
Figure C.3: The graph of u(t) = 2 et2 - 12 for 0 < t < 0.5. --
The ability to solve first-order ODES is an interesting mental exercise, but the
real benefit comes with applying the ability. Applications occur in engineering,
biology, chemistry, environmental studies, and many other disciplines. Later, we
give an example of an application in elementary mechanics.
In elementary mechanics, we assume motion of a rigid body is along a straight
line. Thus, Newton's law of motion applies, giving us "mass times acceleration
equals the sum of the external forces." In mathematical terms, this translates to
F=ma.
Knowing this fact, we proceed to our example.
EXAMPLE C.4. Suppose a body of mass 10 kgs is projected upward with an
initial velocity of 100 m/s. If we assume the gravitational attraction of the earth
is constant, and we neglect all other forces acting on the body, we can find the
maximum height attained by the body, the time at which the maximum height is
reached, and the time it takes the body to return to its starting point.
Solution: We begin by assuming the positive direction is upward and the origin is
the surface of the earth. We have, from Newton's law, F = ma, (force equals the
478 Appendix C: Basics from Ordinary Differential Equations

mass of the object times acceleration due to gravity), (lOkg)(-9.8 m/s2). Force
dv
also equals the mass of the object times acceleration, which is (10 k9) . Thus we
dt
must solve

(10 kg) d _ (10 kg)(-9.8 m/s2),

which can be written as
dv
dt - (-9.8 m/s2).
This means
v(t) _ (-9.8 m/s2)t+c.
Knowing the initial velocity is 100 m/s, we find

v(t) _ (-9.8 m/s2)t + 100 m/s.

Now realizing velocity is the rate of change of position with respect to time, we have
dx
= v(t) _ (-9.8 m/s2)t + 100 m/s.
-t
dtdx
This means the maximum height will occur when = 0. This occurs when
t_ 100
s which answers the second question when does the maximum height
98
occur? Solving for x(t) yields

(-9'8 /S2)t2
x(t) _ + 100t m/s.
2
Note: Since initial distance is at the surface of the earth, implying x(0) = 0, the
constant of integration c equals 0. Applying t = 100 s' Yields
g.8
1002
xmax = 2(9.8) m.

To answer the third question, the time it takes the body to return to its starting
point, we multiply the time it takes to reach maximum height by two, giving us
200 s
2t= 9.8
As the above example shows, using basic laws to acquire an answer to a problem
often results in solving a first-order ODE. This usually means that the problem is de-
scribed by a first-order ODE. For instance, consider problems describing radioactive
decay. Here the ODE is Q'(t) = kQ(t), where k is the constant of proportionality.
Other examples of problems that are described by first-order ODEs are determining
 Section C.2: Some Solution Methods for Second-Order ODES 479

compound interest, studying epidemics or population growth, solving optimization

problems, and modeling electrical circuits.
Not all physical systems are conveniently modeled by first-order ODES. For
instance, electrical and mechanical vibrations, which are also governed by Newton's
law, F = ma, are more easily modeled by a second-order ODE. We take up the task
of reviewing second-order ODEs in the next section.

C.2 SOME SOLUTION METHODS OF

SECOND-ORDER ODES
A second-order ODE has the form
d2u(t) du(t)
S(t) dt2 + K(t) + H(t)u(t) = F(t). (C.8)
dt
If F(t) = 0, then the equation is homogeneous. The homogeneous case, u(t) - 0, is
known as the trivial solution. As in first-order ODES, for the solution to exist it is
important that S(t), K(t), G(t), and F(t) be continuous functions on some interval
a <t </3. Also, if we assume that S(t) does not equal 0 anywhere on the interval
a < t </3, we can convert Equation (C.8) to the familiar form of

u"(t) + k(t)u'(t) + h(t)u(t) = f(t). (C.9)

Solutions for Equation (C.9) depend on the nature of k(t), h(t), and f (t); I will
limit this section to solutions of second-order linear-constant coefficient (k(t) and
h(t) are constants) ODEs,

u'(t) + vu'(t) + cu(t) = f(t). (C.10)

C.2.1 Second-Order Linear Homogeneous ODES

From your previous study in linear homogeneous second-order constant-coefficient
ODEs, two independent solutions, ul (t) and u2 (t), are expected. By the principle
of superposition, any arbitrary linear combination of the two independent solutions
(clue (t) + c2u2 (t)) is also a solution. Given the equation

u" (t) + bu' (t) + cu(t) = 0, (C.11)

we assume that the solution is of the form u (t) = &'t. Finding both the first and
second derivatives of u(t) and plugging them into Equation (C.11) yields

r2 e''t + br& t + cent = 0. (C.12)

Factoring out e''t and multiplying Equation (C.12) by e-''t gives us

r2+br+c=0. (C.13)
480 Appendix C: Basics from Ordinary Differential Equations

Equation (C.13) is known as the characteristic equation of Equation (C.12) .

We solve Equation (C.13) for roots rl and r2 by using the quadratic formula. This
yields

-b+(b2-4(1)(c))2 -b+(b2-4c)2
rl =
2(1) 2

and

-b- (b2 -4(1)(c)) 2 -b- (b2 -4c) 2

r2
2(1) 2

The discriminant (b2 - 4c) determines the solution type. We have three possible
solution types:
real and unequal roots: b2 - 4c > 0;
real and equal roots: b2 - 4c = 0; and
complex roots: b2 - 4c < 0.

Real and Unequal Roots

For real and unequal roots, rl r2. Therefore, the solution of Equation (C.11) is
u(t) = cl&'1 + C2&'2t. (C.14)

However, there are other ways of doing the linear combination. For instance,
consider the form of rl and r2 wherein
rl _ -b + (b2 - 4c) 2 _ _b + (b2_4c) 2
2 2 2

and

-b - (b2 - 4c) 2 -b (b2 - 4c 2

r2= 2 =2- 2

Then, Equation (C.14) could be written as

e_k [di&tt -E- d2 e--yt (b2 - 4c) 2

u(t) = 2 2 where = 2

or

m2e-t (b2 - 4c) 2

u(t)=e -
2t
2 where = 2

If dl = d2 = ml = m2 = 1 in the above two linear combinations, we immediately

notice that we can use forms of Euler's equations, which are
ear + e-aX
cosh ax =
2
 Section C.2: Some Solution Methods for Second-Order ODEs 481

and
ear - e-ate
sinh ax =
2

called the hyperbolic sine and hyperbolic cosine functions. Doing so allows us to
form the new linear combination

t (b2_4c)12
u(t) = e 2 a cosh yt + /3 sinh 7t]; = 2 ,

in which a and /3 are arbitrary constants. We use this form of the solution to
Equation (C.11) with real roots, because it is much more advantageous in our study
of PDEs. The advantage is discussed in both Chapters 4 and 5. Note: If the term
b2 - 4c, is a perfect square, then we will not have the hyperbolic sine and hyperbolic
cosine as part of the solution.

EXAMPLE C.S. Solve

u"(t) + 3u'(t) + u(t) = 0.

Assuming u(t) = ert, we get the characteristic equation

r2+3r+1 =0.
The quadratic formula yields the following roots:

-3+(9-4) 2 -3+v"
rl 2 2

and

-3-(9-4)2 -3-\/g 3+\/g

r2= 2 = 2 =- 2 .

Letting rj = 2 and = -3
2 , r1 = + rj and r2 = - rj. This means that the
solution is

u(t) = cle(E+n)t + c2e(E-n)t =eat (ce?lt + c2e-"lt) (C.15)

or

u(t) = e 23 t o cosh 2 t + Q sinh (2 t )] ,

which is the preferred form.

482 Appendix C: Basics from Ordinary Differential Equations

Real and Equal Roots

We have real and equal roots whenever the discriminant b2 - 4c = 0 applies. Then,
r1 = r2 = r. To have a complete solution, we must develop a second solution.
We do this by using the reduction of order method. For a full discussion on the
reduction of order method, please refer to Elementary Differential Equations, by
Boyce and DiPrima. However, since this is not a course on ODEs, we jump to the
general solution of Equation (C.11), which is

u(t) = aert + /3tert.

Note: This is the form of the solution only for ODEs with constant coefficients.
EXAMPLE C.6. Solve
u" (t) + 4u' (t) + 4u (t) = 0.

Assuming u(t) = ert, we develop the characteristic equation

r2 +4r+4=0,
which factors into

Thus, we have real and equal roots

which yield one solution

So our second solution is

Remember, our second solution depends on using the reduction of order method
and the ODE having constant coefficients. Combining the two solutions yields
u(t) = ae_2t + /3te_2t.

Complex Roots
For complex roots, both r1 and r2 are complex numbers. That is, each root has the
form A + Bi, where A is the real and Bi is the complex part. Note: A and B are
real constants, thus, the roots to the characteristic equation, Equation (C.12) are

= -b -24c 2
Ib2
-b + Ib2 - 4c 2 i
r1 = 2 2+ i
 Section C.2: Some Solution Methods for Second-Order ODES 483

and

_ -b -Ib2 - 4c12i _ -b Ib2_4c12

r2 2.
2 2 2

Thus, we know the two solutions of Equation (C.11) with complex roots are
_7it Ib2
_ 2t 7it _ 2t - 4C) 2
u1(t) = e e and u2 (t) = e e where 'y = 2

Therefore, one possible solution of Equation (C.11) with complex roots is

_ bt Ib2 - 4C) 2
u(t) = cie 2 e7it + c2e - 2t e -7it , where 'y = 2 (C.16)

However, there are other ways of doing the linear combination. For instance, you
could do the linear combination as
bt d1
e7it + d2 e-lit I b2 - 4cl 2
u (t) = e- 2 2 where 'y = 2

or

bt
[mie7it - m2 e-lit I b2 - 4cI 2
u(t) = e- 2 where 'y =
2i 2

If we let d1 = d2 = m1 = m2 = 1 in the above two linear combinations, we

immediately notice that we can use Euler's equations, which are
eiax + e-tax
cos ax =
2

and
eiax - e-iax
sin ax =
2i

Doing so forms the new linear combination

- bt b2 - 4cI 2
u(t) = e 2 [a cos'yt + /3 sin'yt] ; 'y =
2 '

in which a and /3 are arbitrary constants. We will use this form of the solution to
Equation (C.11) with complex roots, because it is much more advantageous in our
study of PDEs. Again, the advantage is described in Chapters 4 and 5.
EXAMPLE C.7. Solve
u"(t) + 6u'(t) + lou(t) = 0.
Assuming u(t) = ert, the characteristic equation is
r2 + 6r + 10 = 0,
484 Appendix C: Basics from Ordinary Differential Equations

which has roots

r1=-3+i
and

r2 = -3-i.
These roots yield the general solution
u(t) = e-3t [a cos (t) + 3 sin (t)]. U

We have covered all three possible general solutions to Equation (C.11). For a
specific solution, we must determine the coefficients. Since the basic idea behind
the solution to a linear homogeneous second-order ODE is two integrations, you
can see that we have two constants of integration to solve for. These constants
of integration have become, through various manipulations, the coefficients of our
general solution (a and /3 in the previous). Thus, a well-defined solution needs two
initial conditions. If the ODE has time, t, as an independent variable, then the
first IC is usually the initial position of the physical body you are modeling. The
second IC is generally the initial velocity of the body and, as such, is the derivative
of the function describing the position of the body. For instance, u (to) = a and
du(to)
= b. Notice that a and b are constants and not functions.
dt
Another way of describing the beginning state of a body with a second-order
ODE without using initial conditions is to describe its boundaries. In this case, the
problem is called a boundary value problem (BVP). Boundary conditions (BCs) are
generally given when the equation is restricted to an interval and the endpoints are
known. Like initial conditions, we need two BCs. Examples of BCs for 0 <t < L
are the following:

u(0) = a, u(L) = b;
u' (0) = a, u' (L) = b;
u(0) = a, u'(L) = b;
or

u'(O) = a, u(L) = b.

As in first-order ODEs, the ability to solve second-order ODEs is an interesting

mental exercise, but the real benefit comes with applying the ability. Two prob-
lems are mechanical and electrical oscillations. Mechanical oscillations can further
be broken into two areas: those with, and those without, external force. Those
that have no external force are called free vibrations. A free vibration equation is
homogeneous. Those that involve an external force are not homogeneous and are
called forced vibrations or forced motion. In the following example, we describe a
free vibration application in mechanics.
 Section C.2: Some Solution Methods for Second-Order ODES 485

EXAMPLE C.8. Consider a weight of 16 lbs attached to a steel spring that has
a natural length of 2 feet. The mass stretches the spring 0.1 ft. Suppose the system
is started in motion by stretching the spring an additional 0.25 ft downward; then
it's released. Determine and solve the resulting equation of motion, neglecting air
resistance.

Solution: Our basic equation is derived from Newton's law of motion, F = ma.
If we let u(t) represent the displacement of the mass from equilibrium, we have
d2u weight 16 lbs 2
ma = m where m = = = 0.5 lbs-s /ft. Though we have no
dt2 ' gravity 32 (ft/s2)
external driving force, we do have a force due to the spring. This force we call FS ,
and it always acts to restore the spring to equilibrium. The force FS is proportional
to the displacement of the spring, u; this is called Hooke's law. FS = ku, where k is
the spring constant. We determine the "spring modulus" k by dividing the weight
16 lbs = 160-j.s
by
y the distance the weight stretches the spring. This gives k =
0.1 ft
Putting it all together results in
d2u
0.5 -160u
dt2

lbs-s2/ft lbs/ft

or

d2u
dt2 + 320u/s2 = 0,
which has the solution

u(t) = cl cos(8Vt) + c2 sin(8Vt) ft.

We know the initial velocity is 0, thus c2 = 0. The initial displacement is 0.25 feet,
which means c1 = 0.25. Therefore,

u(t) = 0.25 cos(8vt) ft

is the final answer.
We complete our review of second-order linear constant-coefficient ODES with
a short discussion of the nonhomogeneous problem.

C.2.2 Second-Order Linear Nonhomogeneous ODES

All second-order linear constant-coefficient nonhomogeneous ODEs are solved using
a basic premise. This premise is that the ODE may be separated into a homogeneous
part, uh (t), and a particular part, up (t) . Consider Equation (C.17),

u'(t) + vu'(t) + cu(t) = f(t). (c.17)

486 Appendix C: Basics from Ordinary Differential Equations

The homogeneous part of Equation (C.17) is

t4( t) + buh (t) + Cuh (t) = 0,

where uh (t) refers to the homogeneous part of the nonhomogeneous ODE. We dis-
covered in the previous section that the solution to the homogeneous equation is

uh(t) = Cluhi (t) + C2uh2 (t)

where uhl (t) and uh2 (t) are two linearly independent solutions. The particular
part of Equation (C.17) really refers to any solution, up (t), of the nonhomogeneous
equation, Equation (C.17), which is different from uhl (t) and uh2 (t). Thus, the
complete solution to Equation (C.17) is

u(t) = up(t) + uh(t) = up(t) + Cluhi (t) + C2uh2 (t)

Hence, we must develop a methodology for obtaining up(t).
There are two major solution methods to consider when solving for up (t). The
first is the method of variation of parameters, which is a general method. It may
be used to solve for up (t) in any second-order linear constant-coefficient nonhomo-
geneous ODE. The second is the method of undetermined coefficients, a specific
method that uses an educated guess of the form of the solution up (t) based on
f (t) in Equation (C.17) . This method will only work for a select few second-order
linear constant-coefficient nonhomogeneous ODEs. In this course, I discuss only the
method of undetermined coefficients. Also, the discussion is example based.
Consider the following examples:
EXAMPLE C.9. Suppose
u"(t) + 3u'(t) + 2u(t) = 3t2 + 2t - 1. (C.18)

We solve Equation (C.18) by first solving the homogeneous part, which is

u( t) + 3u( t) + 2uh (t) = 0, (C.19)

where uh (t) refers to the homogeneous solution. Assuming uh (t) = ert, we form the
characteristic equation

r2 + 3r + 2 = 0,
which factors into

(r+2)(r+1) =0. (C.20)

Equation (C.20) has real and unequal roots of r = -2 and r = -1. Therefore, we
know that uhl (t) = e-2t and uh2 (t) = e-t. Thus,

uh(t) - Cluhi (t) + C2un,2(t) = cle-at + c2e-t. (C.21)

 Section C.2: Some Solution Methods for Second-Order ODES 487

(Note: Equation (C.21) is not in the form of hyperbolic sine and hyperbolic cosine
functions because b2 - 4c is a perfect square).
We must now determine the particular solution, up (t) . First, we notice that 1(t) is
a polynomial in the independent variable t, 1(t) = 3t2 + 2t - 1. This suggests that
a polynomial may be the form of the particular solution, up(t). Thus, we let

up (t) = At2 + Bt + C. (C.22)

We then find the first and second derivatives of Equation (C.22), which are

up (t) = 2At + B (C.23)

and

i4(t) = 2A. (C.24)

Then, we replace u' (t) with up (t) and u" (t) with u , (t) in Equation (C.18), resulting
in

u(t) + 3u(t) + 2up(t) = 3t2 + 2t - 1,

which becomes

2A + 3 (2At + B) + 2 (At2 + Bt + C) = 3t2 + 2t - 1. (C.25)

We solve for the constants A, B, and C in Equation (C.25). By matching the

coefficients of similar terms of the variable t on both sides of the equation, we have

2A + 3B + 2C = -1, (C.26)
6A + 2B = 2 (C.27)

and

2A = 3. (C.28)

3
From Equation 2 Thus, by replacing A by 2 in Equation
(C.28), we know A = 3.

(C.27), we find B = -7.

2 Knowing A and B and using Equation (C.26) we find

C = 13. Therefore
4 '

u(t)=t2_t+
7 13
4 (C.29)

Combining Equation (C.21) and Equation (C.29), we get

u(t) = up(t) + uh(t) = 2t2 - 7t + 13 + cle-ae + c2e-t.

488 Appendix C: Basics from Ordinary Differential Equations

EXAMPLE C.10. Suppose

u" (t) + 3u' (t) + 2u (t) = sin t.

We know from the previous example that

cle-2t + c2e-t.
uh(t) = cluhi (t) + c2uh2 (t) =
Since f (t) = sin t, we assume up (t) = A cost + B sin t. Thus,

up(t) = -A sin t + B cos t

and

u,(t)=-A cost - Bsint. (C.32)

Replacing u" (t) with u , (t), u' (t) with up (t), and u(t) with up (t) in Equation (C.30)
yields

u( t) + 3u(t) + 2uP(t) = sin t,

which becomes

-A cost - B sin t + 3 (-A sin t + B cost) + 2 (A cost + B sin t) = sin t. (C.33)

We solve for the constants A and B in Equation (C.33). By matching the coefficients
of similar terms of the function sin t and cos t on both sides of the equation, we have

-B-3A+2B=B-3A=1 (C.34)

and

-A + 3B + 2A = A + 3B =0. (C.35)
-3
Solving Equations (C.35 and C.34) simultaneously yields A = and B = 1 .
10 10
Therefore, the solution to Equation (C.30) is

u(t) =u(t)
p +uh(t) =-3cost + 1
sint+c1e-2t+c2e-t.
10 10

Note: Using up(t) = A sin t would not have given you the correct answer for u(t).
This completes our review of ODEs. For an extensive review of this topic, I
suggest Elementary Differential Equations, by Boyce and DiPrima.
Appendix D

Mathematical Notation
Throughout this text, we use mathematical notation that may be new to you,
such as IIg to represent the real number line, commonly called the Reals. Also, in
your study of mathematics, you may have thought mathematicians are notation
happy. However, this is not the case. Notation, and knowing the correct definition
of notation, is very important. Another example that shows the importance of
notation is

f
1

x dx.

In English, this notation says integrate the function x on the interval 0 to 1 with
respect to x. I think the notation is much clearer than the English version and
certainly easier to write. I hope you agree. Thus, you easily see it is important
in your mathematical education to learn some basic notation, particularly notation
found in many text books, such as names of spaces and mapping notation.
The real number line is known by the symbol IIg and is called one-space. The
natural numbers, a subset of the real numbers, use the symbol N. The xy-coordinate
plane is commonly known as IIg2 and is called two-space. Three-dimensional space,
the xyz-coordinate system, is known as IIg3 and is called three-space. Both
and IIg3 are easily identified. They are formed by taking copies of the real number
line, then intersecting them at right angles to each other. Since we live in three
dimensional space, it is hard to imagine what 1R5, or R where n E N, meaning
n is any natural number, look like. But mathematically they exist, and they are
quite useful. Also, the spaces that are formed by copies of the real number line,
such as IIg2, IIg3 or R, are known as finite-dimensional spaces. One reason they are
considered finite is the number in the superscript; it indicates a finite number of
copies of the real number line. The reason for the superscript (hence, why they are
finite dimensional) will be addressed in the next section.
The other notation that is important to know is mapping notation. Mapping
notation is very important. It is closely tied to functions, and is easily understood.
For instance, when you add the number 2 to the number 3, 2 + 3, you are actually
dealing with a function. The function is addition. It would be silly to write mapping

489
490 Appendix D: Mathematical Notation

notation for the addition of 2 and 3. However, if you wanted to indicate addition
of all real numbers to each other, then mapping notation is much easier and more
compact. The mapping notation indicating that you can add all real numbers to
each other is

+:IIgxIIg-*II.
It says that you are adding a real number from one copy of IIg to a real number from
another copy of Ilk, and you expect the answer to be a third copy of JIB.
Generally, we do not use mapping notation for addition. We reserve it for other
things. For example, suppose you wanted to discuss, in general terms, all the
functions that are in Mapping notation is a great help. The notation would
simply be

f : IIg -* III.
Notice the mapping has two single copies of III. Adding their superscripts to-
gether indicates the graphs of the functions are living in two-space. What would
the mapping look like for functions in II? One might naturally imagine it is
f: li x li -* IIg.

This indicates functions that take a point from the xy-plane and map it to a
value on the z-axis giving us graphs in the normal IIg3 space. But we could have a
mapping which looks like

f: IIg * IIg x III.

This mapping indicates functions which take a value on the z-axis to a point on
the xy-plane. It is another copy of but with a slightly different twist. The
important thing is knowing where the mapping originates and where it terminates.
Appendix E

Summary of Thermal
Diffusivity of Common
Materials
In this appendix, a summary of thermal diffusivity, k, for common materials is
given. Remember, thermal diffusivity is given by

k = K°
cp

where K° is the thermal conductivity, c is the specific heat, and p is the mass density
of the given material.
Thermal conductivity, K°, is considered temperature independent provided the
temperature variation is not too great. If the temperature variation is great enough,
then the formula for calculating the thermal conductivity is given by
1 Ta
Ki°
=T _T K°dx,
2 1 Tl

where Ko is the new thermal conductivity. However, for most problems in this text,
we will assume that thermal conductivity is temperature independent.

491
492 Appendix E: Summary of Thermal Diffusivity of Common Materials

Thermal Diffusivity of Common Materials

Material k in ft2 /hr k in cm2 /sec

Metals:

Aluminum 3.33 .852

Copper 4.42 1.12
Gold 4.68 1.18
Iron, cast 0.66 0.12
Iron, pure 0.7 0.22
Lead 0.95 0.246
Mercury 0.172 0.044
Nickel 0.6 0.141
Silver 6.6 1.616
Steel, mild 0.48 0.124
Tungsten 2.39 0.725
Zinc 1.6 0.403

Non-metals:

Asbestos 0.01 0.003

Brick, fire clay 0.02 0.005
Cork 0.006 0.002
Glass 0.023 0.006
Granite 0.05 0.014
Ice 0.046 0.011
Oak, across grain 0.0062 0.002
Pine, across grain 0.0059 0.002
Quartz Sand, dry 0.008 0.002
Rubber 0.003 0.001
Water 0.005 0.001
Appendix F

Tables of Fourier and Laplace

Transforms
In this appendix, we provide common Fourier and Laplace transforms. The trans-
forms on the following pages may have a slightly different form than those that you
have seen before. It is an interesting exercise to show that both forms are correct.

493
494 Appendix F: Tables of Fourier and Laplace Transforms

F.1 TABLES OF FOURIER, FOURIER COSINE,

AND FOURIER SINE TRANSFORMS
Table of Fourier Transforms

f (x) J {f (x) } = F(9)

1, x< aa
1 -1 sin a9
(
9
)
0, a < Ix

Ire-a9
1
2 x2 + a2 , a > a

ex, a>O,x>0 1
3. J
0, otherwise a + iO

ex, a>0,x<0 1
4.
0, otherwise a - i9

1
5. a-ax 2 , a > 0 e_02/4a
ra

2a
6. a-aIxI, a> 0
a2 + 02

ire-a0
7. x ,a>0
x2 -}- a2

2a
8.
x2 + a2
, a>0 e -alel
 Section F.1: Table of Fourier Transforms 495

Table of Fourier Transforms (continued)

f(x) J {f(x)} = F(8)

7r, 9<a
sin(ax) 0
9
x
a> ?r 8= a

0, >a

10. cos(ax ), a> 0 cos

Va 4a 4

?r 82 ?r
11. sin(ax2 ), a > 0 cos + iJ
a 4a

Table of Fourier Cosine Transforms

f(x) Fc {f(x)} = F(8)

1
1, 0<x<a sin(a8)
J
' e
lo, a<x

ire-a6
a
2. a2a>
x2 -}- 2a

a
3. a-ax, a > 0
a2 + 82
496 Appendix F: Tables of Fourier and Laplace Transforms

Table of Fourier Cosine Transforms (continued)

f (x) {f(x)} = F(e)

a2 - e2
4. xe-ax, a> 0
(a2
+ e2)2

5. a-axe a> 0 e-62 /4a

4a

xn-le-axe a > 0 F(n) cos (n tan-1(e/a))

6. (a2 + e2)n/2

7. -ax a> 0
cos(ax)e, ae2 + 2a3
4a4 -}.., e4

8. - a>0
sin(ax)eax, 2a3 - a612
4a4 -}.., e4

cos x, 0< x < a 1 sin(a(1 - e)) sin(a(1 + e))

9. 2 1 -e +
0, a<x 1+e

sin ax
ir/2, e<a
10. ,a>0 7r/4, e=a
x 0, e > a

11. sin (ax2 ) , a > 0 cos e2 - sin e2

8a 4a 4a

e2
12. cos(ax2 ), a>0 cos e2 + sin
8a 4a 4a
 Section F.1: Table of Fourier Transforms 497

Table of Fourier Sine Transforms

f(x) {f(x)} = F(e)

1, 0<x<a 1 - cos(ae)
1.
e
0, a < x

x 7I'e-ae
2 x2 a> 2
+a 2,

a-ax , a > 0 e
3. a2 + e2

tae
4. xe-ax, a> 0 (a2
+ e2 )2

xe-ax2 ee_ea/4a
5. a 0
4a3/2
' >

xn-le-axe a > 0 F(n) sin (n tan-1(e/a))

s. (a2 + e2)n/2

e3
7. cos(ax)e-aX, a> 0
4a4 -}.., e4

2a2e
8. sin(ax)e -a x , a > 0
4a4 -}.., e4

sin x, 0<x<a 1 sin(a(1 - e)) sin(a(1 + e))

9.
0, a<x 2 1-e + 1+8
498 Appendix F: Tables of Fourier and Laplace Transforms

Table of Fourier Sine Transforms (continued)

f(x) {f(x)} = F(O)

sin ax
10. a> 0 1 In f B+ a 1
x 2 9-a]
0, e < a
cos(ax)
11. ,a>O ir/4, e=a
x ir/2, B>a

x
12.
ex
,a>0 -
tan1
\/a

F.2 TABLE OF LAPLACE TRANSFORMS

Table of Laplace Transforms

f(t) £ {f(t)} = F(s)

1. 1 -s ,s>0
1

1
2. t 2
s

3. t,nEN n!n,s>0
S

1 1
4.
art
 Section F.2: Table of Laplace Transforms 499

Table of Laplace Transforms (continued)

f(t) £ {f(t)} = F(s)

1
5. a«t
s-a , s > a
a
6. sin(at) ,s>0
s2 a2

F(p+l)
p sp+1

S
8. cos(at) ,s>0
s2 a2

a
9. sinh(at) s2 - a2 , s > a)

s
10. cosh(at) s2 - a2 , s > Ia

11. e«t sin t '

(s - a)2 + '32 s > a

s- a
12. a«t cos(,3t) , s> a
(s-a)2 +132

n.
13. the«t, n E N , s> a
(s - a)n+1

e-«S
14. u(t - a), unit step function , s > Ox
s

15. u(t - a) f (t - a) e-«8F(s)

500 Appendix F: Tables of Fourier and Laplace Transforms

Table of Laplace Transforms (continued)

f (t) £ {f(t)} = F(s)

16. 6(t - a), Dirac delta function e-«s

17. e«t f (t) F(s - a)

t
18. f (t - u)g(u) du, Convolution Integral F(s)G(s)
0

1 -a2 e-a
19. a 4t
art

20.
a
e 4t
2
e-a
2 't3

21. er f c
2 a
, Complementary Error Function
e
s

t -a2 a
22. 2
VT_
a- a er f c

23. eabeb2t {erfc b +

a e
-a

24.
2
-eabeb t {erfc b -}-
a
+ er f c
(a) be-a1

2) J
e-as
25. H (t - a)
S
Bibliography
1. Asmar, Nakhle H. Partial Differential Equations and Boundary Value Prob-
lems. Englewood Cliffs: Prentice-Hall, 2000.
2. Bartle, Robert G. The Elements of Real Analysis, 2nd ed. New York: Wiley,
1976.

3. Boyce, William E. and DiPrima, Richard C. Elementary Differential Equa-

tions, 2nd ed. New York: Wiley, 1969.
4. Boyer, Carl B. A History of Mathematics, 2' ed. New York: Wiley, 1991
5. Byron, Frederick W. Jr. and Fuller, Robert W. Mathematics of Classical and
Quantum Physics. New York: Dover, 1992.
6. Budak, B. M., Samarskii, A. A., and Tikhonov, A. N. a Collection of Problems
of Mathematical Physics, New York: Dover, 1988.
7. Churchill, Ruel V. Fourier Series and Boundary Value Problems, 2nd ed. New
York: McGraw-Hill, 1969.
8. Courant, R. and Hilbert D. Methods of Mathematical Physics. New York:
Interscience, 1953.
9. CRC Standard Mathematical Tables, 18th ed. Cleveland: CRC Press, 1970.
10. Dettman, John W. Mathematical Methods in Physics and Engineering. New
York: Dover, 1988.
11. Duchateau, Paul and Zachmann, David W. Partial Differential Equations.
New York: McGraw-Hill, 1986.
12. Farlow, Stanley J. Partial Differential Equations for Scientists and Engineers.
New York: Dover, 1993.
13. Feynman, Richard P., Leighton, Robert B., and Sands, Matthew the Feynman
Lectures on Physics. Reading, MA: Addison-Wesley, 1966
14. Haberman, Richard Elementary Applied Partial Differential Equation with
Fourier Series and Boundary Value Problems 2nd ed. Englewood Cliffs:
Prentice-Hall, 1987.

501
502 Bibliography

15. Hayt, William H. Jr. Engineering Electromagnetics, 5th ed. New York,
McGraw-Hill, 1989.
16. Keener, James and Sneyd, James Mathematical Physiology. New York: Springer-
Verlag, 1998.

17. Korner, T. W. Fourier Analysis. Cambridge, England: Cambridge University

Press, 1995.
18. Kreyszig, Erwin Advanced Engineering Mathematics, 7th ed. New York: Wi-
ley, 1993.

19. Kreyszig, Erwin Introductory Functional Analysis with Applications. New

York: Wiley, 1978.
20. Lay, Steven R. Analysis with an Introduction to Proof, 2nd. New Jersey:
Prentice Hall, 1990.
21. O'Neil, Peter V. Advanced Engineering Mathematics, 4th ed. Boston: PWS-
Kent, 1995.
22. Ozisik, M. Necati Heat Conduction. New York: Wiley, 1980.
23. Powers, David L. Boundary Value Problems, 2nd ed. New York: Academic
Press, 1979.
24. Rainville, Earl D. Infinite Series. New York: Macmillian, 1967.
25. Rainville, Earl D. Special Functions. New York: Macmillian, 1960.
26. Renardy, Michael and Rogers, Robert C. An Introduction to Partial Differen-
tial Equations. New York: Springer-Verlag, 1993.
27. Sagen, Hans Boundary and Eigenvalue Problems in Mathematical Physics.
New York: Dover, 1989.
28. Spiegel, Murray R. Theory and Problems of Fourier Analysis with Applications
to Boundary Value Problems. New York: McGraw-Hill, 1974.
29. Spiegel, Murray R. and Liu, John Mathematical Handbook of Formulas and
Tables, 2nd ed. New York: McGraw-Hi11,1999.
30. Strauss, Walter A. Partial Differential Equations An Introduction. New York:
Wiley, 1992.

31. Taylor, A. B. Mathematical Models in Applied Mechanics. Oxford, England:

Clarendon Press, 1986.
32. Tolstov, Georgi P. Fourier Series. New York: Dover, 1976.
33. Trench, William F. Elementary Differential Equations. Pacific Grove: Brooks/Cole,
2000.
 Bibliography 503

34. Trim, Donald W. Applied Partial Differential Equations. Boston: PWS-Kent,

1990.

35. Walker, P. L. The Theory of Fourier Series and Integrals. New York: Wiley-
Interscience, 1986.
36. Wolfram, Stephen The Mathematica Book, 3rd ed. Champaign, IL.: Wolfram
Media and Cambridge University Press, 1996.
37. Wylie, Ray C. and Barrett, Louis C. Advanced Engineering Mathematics, 6th
ed. New York: McGraw-Hill, 1995.
38. Zachmanoglou, E. C. and Thoe, Dale W. Introduction to Partial Differential
Equations with Applications. New York: Dover, 1986.
39. Zill, Dennis G. and Cullen, Michael R. Advanced Engineering Mathematics.
Sudbury: Jones and Barlett, 2000.
Index

affine equation, 74 d'Alembert, Jean Le Rond

Airy's equation, 329 d'Alembert's solution, 64
Airy, G. B., 329 Del, 164
Del squared, 164
basis vectors, 85 Dirichlet boundary conditions, 17, 50
Bessel's equation, 341, 345 Dirichlet, P. G. L., 17
Boltzmann's equation, 44, 389 discriminant, 2
boundary conditions, 17
Dirichlet conditions, 17 eigenfunction, 144
Neumann conditions, 18 eigenfunction expansion, 100
Robin's conditions, 19 eigenfunctions, 98
cantilever beam, 370 eigenvalue, 83
Cauchy, Augustin-Louis, 63 eigenvalues, 83, 144
Cauchy-Riemann equations, 39 eigenvector, 84
characteristic, 57 eigenvectors, 83
characteristic equation, 480 Einstein, Albert, 381
characteristic polynomial, 84 elliptic class of PDES
Chebyshev's equation, 317 Laplace's equation, 3
coefficient of convective heat transfer, equilibrium solution
19 steady-state, 5, 31
complementary error function, 424 error function, 424
complex roots, 480, 482 Euler's Constant, 348
conduction Euler's equation, 320
heat transfer, 7 Euler's equations, 480, 483
conservation of thermal energy law, Euler, Leonhard, 320
10, 11 even function, 120
continuity of Fourier cosine series, 127, even or odd extension, 121
194
continuity of Fourier series, 115, 194 Fick's law, 39
continuity of Fourier sine series, 127, finite dimensional, 85
194 first-order ODE, 473
convection flexural rigidity, 365
heat transfer, 6 forced vibrations, 484
convection proportionality function, 9 Fourier coefficients, 100
cross-section, 6 Fourier cosine integral, 415
Fourier cosine transform, 442
d'Alembert's solution, 41, 66 Fourier integral, 405

505
506 Index

Fourier sine integral, 415 wave equation, 3

Fourier sine transform, 442
Fourier transform, 440 initial condition
Fourier transforms initial displacement
convolution theorem, 445 initial velocity, 46
Fourier's Convergence Theorem, 113 initial temperature distribution,
Fourier's integral 15
complex form, 414 initial temperature distribution, 15
Fourier's integral theorem, 407 inner product, 79, 92
Fourier's law of heat conduction, 12, insulation
18 lateral
Fourier, Joseph boundary, 7, 8, 18
Fourier's law of heat conduction, integral equation, 422
7 integrating factor, 474, 476
free vibrations, 484 internal heat source, 8
Frobenius Series, 341 inverse Fourier cosine transform, 442
Frobenius, G., 341 inverse Fourier sine transform, 442
Fuchs, Immanuel L., 332 inverse Fourier transform, 440
function space of piecewise smooth func- irregular singular point
tions, 91 singular point, 318
fundamental solution set, 98
jump discontinuity, 90
Gauss-Jordan reduction method, 84
general second-order linear homoge- Keener, James, 39
neous variable-coefficient ODE, Kirchoff, Gustav, 44
317
Gibbs phenomenon, 126, 193, 213 Laplace transforms, 421
Gibbs, Josiah, 218 Laplace's equation, 181
Gram-Schmidt orthonormalization, 88 Laplace's equation in polar coordinates,
352
harmonic function, 181 Laplace's equation in spherical coor-
heat conduction in a circular plate, dinates, 356
373 Laplacian, 27, 28
heat conduction laws, 7, 11 lateral insulation, 164
heat flow equation, 11 perfect lateral insulation, 164
heat operator, 133 Legendre polynomials, 334, 337
heat sink, 15 Legendre's equation
Helmholtz equation, 381 special functions, 334
Helmholtz, Herman, 380 Leibniz's formula, 13, 198
Hermite's equation, 317 Leibniz, Gottfried, 5
homogeneous, 473, 479 linear combination, 76
homogeneous equation for transverse linear equation, 74, 136
vibrations in a beam, 367 linear homogeneous equation, 137
Hooke's law, 52, 485 linear operator, 132
Hooke, R., 52 linearly dependent, 77
hyperbolic class of PDEs linearly independent, 77
 Index 507

mass density, 8 resistance force, 43

maximum-minimum theorem, 25 restoring force, 42
Method of Frobenius, 341 rigidly fixed beam, 371
moment of inertia, 365 Robin's boundary conditions, 19, 50,
multi-dimensional steady-state equa- 302
tion, 39 Robin, V. G., 19
Rodrigues, Olinde, 339
Neumann boundary conditions, 18, 50
Neumann, Franz, 18 Schrodinger's equation, 380
Newton's law of cooling, 19, 307 separation of variables, 139, 476
Newton's law of motion, 42, 43, 477, simply supported beam, 367
485 singular point, 318
Newton, Isaac, 5 Sneyd, James, 39
nonhomogeneous, 137 specific heat, 8
specified temperatures, 17
odd function, 120 standard basis, 78, 85
ordinary point, 318 function space of piecewise smooth
orthogonal, 78, 79 functions, 95
orthogonal sequence, 98 standard inner product
orthonormal, 86 function space of piecewise smooth
functions, 94
parabolic class of PDEs
steady-state
heat equation, 3
equilibrium solution, 5, 31
perfectly flexible, 42
Sturm-Liouville eigenvalue problem,
periodic extension, 112
279
piecewise smooth function, 91
Planck, Max, 381 Taylor series, 331
Poisson's equation, 181 Telegrapher's equation, 44, 389
Poisson, Simeon-Denis, 181 temperature distribution, 8
power series solution method, 327 tension, 42
Principle of Conservation of Energy, thermal conductivity, 9
54 thermal difusivity, 14
principle of superposition, 137 thin beam, 364
Properties of Laplace transforms, 422 transcendental equation, 309
Trigonometric Fourier series, 100
quadratic equation, 2
trigonometric Fourier series
quadratic formula, 2
Fourier series, 103
radiation trivial solution, 137
heat transfer, 7
radioactive decay, 478 vector space, 75
Rayleigh quotient, 290, 293 real vector space, 75
real and equal roots, 480, 482 Weierstrass M-test, 114
real and unequal roots, 480 Weierstrass, Karl, 114
regular singular point
singular point, 318
regular Sturm-Liouville problem, 283
PEARSON Prentice Hall
Uppper Saddle River, NJ 07458
Education 1 1 1' Fi'i1i