Rao K.S - Introduction to Partial Differential Equations (2011).pdf

Introduction to
PARTIAL DIFFERENTIAL EQUATIONS

THIRD EDITION
K. SANKARA RAO
Formerly Professor
Department of Mathematics
Anna University, Chennai
New Delhi-110001
2011
INTRODUCTION TO PARTIAL DIFFERENTIAL EQUATIONS, Third Edition
K. Sankara Rao
© 2011 by PHI Learning Private Limited, New Delhi. All rights reserved. No part of this book may
be reproduced in any form, by mimeograph or any other means, without permission in writing from
the publisher.
ISBN-978-81-203-4222-4
The export rights of this book are vested solely with the publisher.
Eleventh Printing (Third Edition) … … January, 2011
Published by Asoke K. Ghosh, PHI Learning Private Limited, M-97, Connaught Circus,
New Delhi-110001 and Printed by Syndicate Binders, A-20, Hosiery Complex, Noida, Phase-II
Extension, Noida-201305 (N.C.R. Delhi).
This book is dedicated with affection and gratitude to
the memory of my respected Father
(Late) KOMMURI VENKATESWARLU
and
to my respected Mother
SHRIMATI VENKATARATNAMMA
Contents
Preface ix
Preface to the First and Second Edition xi
0. Partial Differential Equations of First Order 1–51

0.1 Introduction 1
0.2 Surfaces and Normals 2
0.3 Curves and Their Tangents 4
0.4 Formation of Partial Differential Equation 7
0.5 Solution of Partial Differential Equations of First Order 11
0.6 Integral Surfaces Passing Through a Given Curve 18
0.7 The Cauchy Problem for First Order Equations 21
0.8 Surfaces Orthogonal to a Given System of Surfaces 22
0.9 First Order Non-linear Equations 23
0.9.1 Cauchy Method of Characteristics 25
0.10 Compatible Systems of First Order Equations 33
0.11 Charpit’s Method 37
0.11.1 Special Types of First Order Equations 42
Exercises 49
1. Fundamental Concepts 52–105

1.1 Introduction 52
1.2 Classification of Second Order PDE 53
1.3 Canonical Forms 53
1.3.1 Canonical Form for Hyperbolic Equation 55
1.3.2 Canonical Form for Parabolic Equation 57
1.3.3 Canonical Form for Elliptic Equation 59
v
vi CONTENTS
1.4 Adjoint Operators 69

1.5 Riemann’s Method 71
1.6 Linear Partial Differential Equations with Constant Coefficiants 84
1.6.1 General Method for Finding CF of Reducible Non-homogeneous
Linear PDE 86
1.6.2 General Method to Find CF of Irreducible Non-homogeneous Linear PDE 89
1.6.3 Methods for Finding the Particular Integral (PI) 90
1.7 Homogeneous Linear PDE with Constant Coefficients 97
1.7.1 Finding the Complementary Function 98
1.7.2 Methods for Finding the PI 99
Exercises 102
2. Elliptic Differential Equations 106–181

2.1 Occurrence of the Laplace and Poisson Equations 106
2.1.1 Derivation of Laplace Equation 106
2.1.2 Derivation of Poisson Equation 108
2.2 Boundary Value Problems (BVPs) 109
2.3 Some Important Mathematical Tools 110
2.4 Properties of Harmonic Functions 111
2.4.1 The Spherical Mean 113
2.4.2 Mean Value Theorem for Harmonic Functions 114
2.4.3 Maximum-Minimum Principle and Consequences 115
2.5 Separation of Variables 122
2.6 Dirichlet Problem for a Rectangle 124
2.7 The Neumann Problem for a Rectangle 126
2.8 Interior Dirichlet Problem for a Circle 128
2.9 Exterior Dirichlet Problem for a Circle 132
2.10 Interior Neumann Problem for a Circle 136
2.11 Solution of Laplace Equation in Cylindrical Coordinates 138
2.12 Solution of Laplace Equation in Spherical Coordinates 146
2.13 Miscellaneous Examples 154
Exercises 178
3. Parabolic Differential Equations 182–231

3.1 Occurrence of the Diffusion Equation 182
3.2 Boundary Conditions 184
3.3 Elementary Solutions of the Diffusion Equation 185
3.4 Dirac Delta Function 189
3.5 Separation of Variables Method 195
3.6 Solution of Diffusion Equation in Cylindrical Coordinates 208
3.7 Solution of Diffusion Equation in Spherical Coordinates 211
3.8 Maximum-Minimum Principle and Consequences 215
CONTENTS vii
3.9 Non-linear Equations (Models) 217
3.9.1 Semilinear Equations 217
3.9.2 Quasi-linear Equations 217
3.9.3 Burger’s Equation 218
3.9.4 Initial Value Problem for Burger’s Equation 219
Exercises 229
4. Hyperbolic Differential Equations 232–281

4.1 Occurrence of the Wave Equation 232
4.2 Derivation of One-dimensional Wave Equation 233
4.3 Solution of One-dimensional Wave Equation by Canonical Reduction 236
4.4 The Initial Value Problem; D’Alembert’s Solution 240
4.5 Vibrating String—Variables Separable Solution 245
4.6 Forced Vibrations—Solution of Non-homogeneous Equation 254
4.7 Boundary and Initial Value Problem for Two-dimensional Wave Equations—
Method of Eigenfunction 257
4.8 Periodic Solution of One-dimensional Wave Equation in Cylindrical Coordinates 260
4.9 Periodic Solution of One-dimensional Wave Equation in Spherical Polar
Coordinates 262
4.10 Vibration of a Circular Membrane 264
4.11 Uniqueness of the Solution for the Wave Equation 266
4.12 Duhamel’s Principle 268
Exercises 279
5. Green’s Function 282–315

5.1 Introduction 282
5.2 Green’s Function for Laplace Equation 289
5.3 The Methods of Images 295
5.4 The Eigenfunction Method 302
5.5 Green’s Function for the Wave Equation—Helmholtz Theorem 305
5.6 Green’s Function for the Diffusion Equation 310
Exercises 314
6. Laplace Transform Methods 316–387

6.2 Transform of Some Elementary Functions 319
6.3 Properties of Laplace Transform 321
6.4 Transform of a Periodic Function 329
6.5 Transform of Error Function 332
6.6 Transform of Bessel’s Function 335
6.7 Transform of Dirac Delta Function 337
viii CONTENTS
6.8 Inverse Transform 337

6.9 Convolution Theorem (Faltung Theorem) 344
6.10 Transform of Unit Step Function 349
6.11 Complex Inversion Formula (Mellin-Fourier Integral) 352
6.12 Solution of Ordinary Differential Equations 356
6.13 Solution of Partial Differential Equations 360
6.13.1 Solution of Diffusion Equation 362
6.13.2 Solution of Wave Equation 367
Exercises 383
7. Fourier Transform Methods 388–446

7.2 Fourier Integral Representations 388
7.2.1 Fourier Integral Theorem 390
7.2.2 Sine and Cosine Integral Representations 394
7.3 Fourier Transform Pairs 395
7.4 Transform of Elementary Functions 396
7.5 Properties of Fourier Trasnform 401
7.6 Convolution Theorem (Faltung Theorem) 412
7.7 Parseval’s Relation 414
7.8 Transform of Dirac Delta Function 416
7.9 Multiple Fourier Transforms 416
7.10 Finite Fourier Transforms 417
7.10.1 Finite Sine Transform 418
7.10.2 Finite Cosine Transform 419
7.11 Solution of Diffusion Equation 421
7.12 Solution of Wave Equation 425
7.13 Solution of Laplace Equation 428
Exercises 443
Bibliography 447–448
Answers and Keys to Exercises 449–484
Index 485–488
Preface
The objective of this third edition is the same as in previous two editions: to provide a broad
coverage of various mathematical techniques that are widely used for solving and to get analytical
solutions to Partial Differential Equations of first and second order, which occur in science and
engineering. In fact, while writing this book, I have been guided by a simple teaching philosophy:
An ideal textbook should teach the students to solve problems. This book contains hundreds of
carefully chosen worked-out examples, which introduce and clarify every new concept. The core
material presented in the second edition remains unchanged.
I have updated the previous edition by adding new material as suggested by my active
colleagues, friends and students.
Chapter 1 has been updated by adding new sections on both homogeneous and non-
homogeneous linear PDEs, with constant coefficients, while Chapter 2 has been repeated as such
with the only addition that a solution to Helmholtz equation using variables separable method is
discussed in detail.
In Chapter 3, few models of non-linear PDEs have been introduced. In particular, the exact
solution of the IVP for non-linear Burger’s equation is obtained using Cole–Hopf function.
Chapter 4 has been updated with additional comments and explanations, for better
understanding of normal modes of vibrations of a stretched string.
Chapters 5–7 remain unchanged.
I wish to express my gratitude to various authors, whose works are referred to while writing
this book, as listed in the Bibliography. Finally, I would like to thank all my old colleagues, friends
and students, whose feedback has helped me to improve over previous two editions.
It is also a pleasure to thank the publisher, PHI Learning, for their careful processing of the
manuscript both at the editorial and production stages.
Any suggestions, remarks and constructive comments for the improvement of text are always
welcome.
K. SANKARA RAO
ix
Preface to the
First and Second Edition
With the remarkable advances made in various branches of science, engineering and technology,
today, more than ever before, the study of partial differential equations has become essential. For,
to have an in-depth understanding of subjects like fluid dynamics and heat transfer, aerodynamics,
elasticity, waves, and electromagnetics, the knowledge of finding solutions to partial differential
equations is absolutely necessary.
This book on Partial Differential Equations is the outcome of a series of lectures delivered by
me, over several years, to the postgraduate students of Applied Mathematics at Anna University,
Chennai. It is written mainly to acquaint the reader with various well-known mathematical
techniques, namely, the variables separable method, integral transform techniques, and Green’s
function approach, so as to solve various boundary value problems involving parabolic, elliptic and
hyperbolic partial differential equations, which arise in many physical situations. In fact, the
Laplace equation, the heat conduction equation and the wave equation have been derived by taking
into account certain physical problems.
The book has been organized in a logical order and the topics are discussed in a systematic
manner. In Chapter 0, partial differential equations of first order are dealt with. In Chapter 1, the
classification of second order partial differential equations, and their canonical forms are given. The
concept of adjoint operators is introduced and illustrated through examples, and Riemann’s method
of solving Cauchy’s problem described. Chapter 2 deals with elliptic differential equations. Also,
basic mathematical tools as well as various properties of harmonic functions are discussed. Further,
the Dirichlet and Neumann boundary value problems are solved using variables separable method
in cartesian, cylindrical and spherical coordinate systems. Chapter 3 is devoted to a discussion on
the solution of boundary value problems describing the parabolic or diffusion equation in various
coordinate systems using the variables separable method. Elementary solutions are also given.
Besides, the maximum-minimum principle is discussed, and the concept of Dirac delta function is
introduced along with a few properties. Chapter 4 provides a detailed study of the wave equation
representing the hyperbolic partial differential equation, and gives D’Alembert’s solution.
In addition, the chapter presents problems like vibrating string, vibration of a circular
membrane, and periodic solutions of wave equation, shows the uniqueness of the solutions, and
illustrates Duhamel’s principle. Chapter 5 introduces the basic concepts in the construction of
xi
xii PREFACE TO THE FIRST AND SECOND EDITION
Green’s function for various boundary value problems using the eigenfunction method and the
method of images. Chapter 6 on Laplace transform method is self-contained since the subject
matter has been developed from the basic definition. Various properties of the transform and
inverse transform are described and detailed proofs are given, besides presenting the convolution
theorem and complex inversion formula. Further, the Laplace transform methods are applied to
solve several initial value, boundary value and initial boundary value problems. Finally in
Chapter 7, the theory of Fourier transform is discussed in detail. Finite Fourier transforms are also
introduced, and their applications to diffusion, wave and Laplace equations have been analyzed.
The text is interspersed with solved examples; also, miscellaneous examples are given in
most of the chapters. Exercises along with hints are provided at the end of each chapter so as to
drill the student in problem-solving. The preprequisites for the book include a knowledge of
advanced calculus, Fourier series, and some understanding about ordinary differential equations
and special functions.
The book is designed as a textbook for a first course on partial differential equations for the
senior undergraduate engineering students and postgraduate students of applied mathematics,
physics and engineering. The various topics covered in the book can be taught either in one
semester or in two semesters depending on the syllabi. The book would also be of interest to
scientists and engineers engaged in research.
In the second edition, I have added a new chapter (Partial Differential Equations of First
Order). Also, some additional examples are included, which are taken from question papers for
GATE in the last 10 years. This, I believe, would surely benefit students intending to appear for the
GATE examination.
I am indebted to many of my colleagues in the Department of Mathematics, particularly to
Prof. N. Muthiyalu, Prof. Prabhamani, R. Patil, Dr. J. Pandurangan, Prof. K. Manivachakan,
for their many useful comments and suggestions. I am also grateful to the authorities of
Anna University, for the encouragement and inspiration provided by them.
I wish to thank Mr. M.M. Thomas for the excellent typing of the manuscript. Besides, my
gratitude and appreciation are due to the Publishers, PHI Learning, for the very careful and
meticulous processing of the manuscript, both during the editorial and production stages.
Finally, I sincerely thank my wife, Leela, daughter Aruna and son-in-law R. Parthasarathi, for
their patience and encouragement while writing this book. I also appreciate the understanding
shown by my granddaughter Sangeetha who had to forego my attention and care during the course
of my book writing.
Any constructive comments for improving the contents of this volume will be warmly
appreciated.
K. SANKARA RAO
CHAPTER 0
Partial Differential Equations

of First Order
0.1 INTRODUCTION
Partial differential equations of first order occur in many practical situations such
as Brownian motion, the theory of stochastic processes, radioactive disintegration, noise in
communication systems, population growth and in many problems dealing with telephone
traffic, traffic flow along a highway and gas dynamics and so on. In fact, their study is
essential to understand the nature of solutions and forms a guide to find the solutions of
higher order partial differential equations.
A first order partial differential equation (usually denoted by PDE) in two independent
variables x, y and one unknown z, also called dependent variable, is an equation of the form
§ wz wz·
F ¨ x, y , z , , 0.
w x w y ¸¹
(0.1)
©
Introducing the notation
wz wz
p , q (0.2)
wx wy
Equation (0.1) can be written in symbolic form as
F ( x, y , z , p , q ) 0. (0.3)
A solution of Eq. (0.1) in some domain Ω of IR 2 is a function z f ( x, y ) defined and is

of C c in Ω should satisfy the following two conditions:
(i) For every ( x, y ) Ω, the point ( x, y, z , p, q ) is in the domain of the function F.
(ii) When z f ( x, y ) is substituted into Eq. (0.1), it should reduce to an identity in x,
y for all ( x, y ) Ω.
1
2 INTRODUCTION TO PARTIAL DIFFERENTIAL EQUATIONS
We classify the PDE of first order depending upon the form of the function F. An
equation of the form
wz wz
P ( x, y , z ) Q ( x, y , z ) R ( x, y , z ) (0.4)
wx wy
is a quasi-linear PDE of first order, if the derivatives w z /w x and w z /w y that appear in the
function F are linear, while the coefficients P, Q and R depend on the independent variables
x, y and also on the dependent variable z. Similarly, an equation of the form
wz wz
P ( x, y ) Q ( x, y ) R ( x, y , z ) (0.5)
wx wy
is called almost linear PDE of first order, if the coefficients P and Q are functions of the inde-
pendent variables only. An equation of the form
wz wz
a ( x, y ) b ( x, y ) c ( x, y ) z d ( x, y ) (0.6)
wx wy
is called a linear PDE of first order, if the function F is linear in w z /w x, w z / w y and z, while
the coefficients a, b, c and d depend only on the independent variables x and y. An equation
which does not fit into any of the above categories is called non-linear. For example,
wz wz
(i) x y nz
wx wy
is a linear PDE of first order.
wz wz
(ii) x y z2
wx wy
is an almost linear PDE of first order.
wz wz
(iii) P( z ) 0
wx wy
is a quasi-linear PDE of first order.
2 2
wz §w z ·
(iv) §¨ ·¸ ¨ ¸ 1
©w x ¹ ©w y ¹
is a non-linear PDE of first order.
Before discussing various methods for finding the solutions of the first order PDEs, we
shall review some of the basic definitions and concepts needed from calculus.
0.2 SURFACES AND NORMALS
Let Ω be a domain in three-dimensional space IR 3 and suppose F ( x, y, z ) is a function in

the class C c (Ω), then the vector valued function grad F can be written as
PARTIAL DIFFERENTIAL EQUATIONS OF FIRST ORDER 3
⎛∂ F ∂ F ∂ F ⎞
grad F = ⎜ , , (0.7)
⎝ ∂ x ∂ y ∂ z ⎟⎠
If we assume that the partial derivatives of F do not vanish simultaneously at any point then
the set of points (x, y, z) in Ω, satisfying the equation
F ( x, y , z ) = C (0.8)
is a surface in Ω for some constant C. This surface denoted by SC is called a level surface
of F. If (x0, y0, z0) is a given point in Ω, then by taking F ( x0 , y0 , z0 ) = C , we get an equation
of the form
F ( x, y , z ) = F ( x0 , y0 , z0 ), (0.9)
which represents a surface in W, passing through the point ( x0 , y0 , z0 ). Here, Eq. (0.8) represents
a one-parameter family of surface in W. The value of grad F is a vector, normal to the level
surface. Now, one may ask, if it is possible to solve Eq. (0.8) for z in terms of x and y. To
answer this question, let us consider a set of relations of the form
x = f1 (u , v ), y = f 2 (u , v ), z = f3 (u , v) (0.10)
Here for every pair of values of u and v, we will have three numbers x, y and z, which
represents a point in space. However, it may be noted that, every point in space need not
correspond to a pair u and v. But, if the Jacobian
∂ ( f1 , f 2 )
≠0 (0.11)
∂ (u , v)
then, the first two equations of (0.10) can be solved and u and v, can be expressed as functions
of x and y like
u = λ ( x, y ), v = μ ( x, y ).
Thus, u and v are obtained once x and y are known, and the third relation of Eq. (0.10)
gives the value of z in the form
z = f3 [λ ( x, y ), μ ( x, y )] (0.12)
This is, of course, a functional relation between the coordinates x, y and z as in Eq. (0.8).
Hence, any point (x, y, z) obtained from Eq. (0.10) always lie on a fixed surface. Equations
(0.10) are also called parametric equations of a surface. It may be noted that the parametric
equation of a surface need not be unique, which can be seen in the following example:
The parametric equations
x = r sin θ cos φ , y = r sin θ sin φ , z = r cos θ
and
(1 − φ 2 ) (1 − φ 2 ) 2rφ
x=r cos θ , y=r sin θ , z=
(1 + φ )
2
(1 + φ )2 1+φ2
both represent the same surface x 2 + y 2 + z 2 = r 2 which is a sphere, where r is a constant.
If the equation of the surface is of the form

z f ( x, y ) (0.13)
Then
F f ( x, y ) z 0. (0.14)
Differentiating partially with respect to x and y, we obtain
wF wF wz wF wF wz
0, 0
wx wz wx wy wz wy
from which we get
wz w F /w x wF
(using 0.14)
wx w F /w z wx
wF
or p.
wx
Similarly, we obtain
wF wF
q and 1.
wy wz
Hence, the direction cosines of the normal to the surface at a point (x, y, z) are given as
§ p q 1 ·
¨ , , ¸ (0.15)
¨© p 2 q 2 1 p2 q2 1 p 2 q 2 1 ¸¹
Now, returning to the level surface given by Eq. (0.8), it is easy to write the equation of the
tangent plane to the surface Sc at a point (x0, y0, z0) as
ªw F º ªw F º ªw F º
( x x0 ) « ( x0 , y0 , z0 )» ( y y0 ) « ( x0 , y0 , z0 ) » ( z z0 ) « ( x0 , y0 , z0 )» 0. (0.16)
¬wx ¼ ¬w y ¼ ¬wz ¼
0.3 CURVES AND THEIR TANGENTS

A curve in three-dimensional space IR 3 can be described in terms of parametric equations.
&
Suppose r denotes the position vector of a point on a curve C, then the vector equation of
C may be written as
& &
r F (t ) for t I , (0.17)
where I is some interval on the real axis. In component form, Eq. (0.17) can be written as
x f1 (t ), y f 2 (t ), z f3 (t ) (0.18)
& &
where r ( x, y, z ) and F (t ) [ f1 (t ), f 2 (t ), f3 (t )] and the functions f1 , f 2 and f3 belongs to
C c ( I ).
Further, we assume that

§ df1 (t ) df 2 (t ) df3 (t ) ·
¨© , , ¸ z (0, 0, 0) (0.19)
dt dt dt ¹
This non-vanishing vector is tangent to the curve C at the point (x, y, z) or at [ f1 (t ), f 2 (t ), f3 (t )]

of the curve C.
Another way of describing a curve in three-dimensional space IR 3 is by using the fact
that the intersection of two surfaces gives rise to a curve.
Let
F1 ( x, y , z ) C1 ½°
and ¾ (0.20)
F2 ( x, y , z ) C2 °¿
are two surfaces. Their intersection, if not empty, is always a curve, provided grad F1 and
grad F2 are not collinear at any point of Ω in IR 3. In other words, the intersection of surfaces
given by Eq. (0.20) is a curve if
grad F1 ( x, y , z ) u grad F2 ( x, y , z ) z (0, 0, 0) (0.21)
for every ( x, y, z ) Ω . For various values of C1 and C2, Eq. (0.20) describes different curves.
The totality of these curves is called a two parameter family of curves. Here, C1 and C2
are referred as parameters of this family. Thus, if we have two surfaces denoted by S1 and S2
whose equations are in the form
F ( x, y , z ) 0½°
and ¾ (0.22)
G ( x, y , z ) 0°¿
Then, the equation of the tangent plane to S1 at a point P ( x0 , y0 , z0 ) is

wF wF wF
( x x0 ) ( y y0 ) ( z z0 ) 0 (0.23)
wx wy wz
Similarly, the equation of the tangent plane to S2 at the point P ( x0 , y0 , z0 ) is

wG wG wG
( x x0 ) ( y y0 ) ( z z0 ) 0. (0.24)
wx wy wz
Here, the partial derivatives w F/w x, w G /w x, etc. are evaluated at P ( x0 , y0 , z0 ). The intersection
of these two tangent planes is the tangent line L at P to the curve C, which is the intersection
of the surfaces S1 and S2. The equation of the tangent line L to the curve C at ( x0 , y0 , z0 ) is
obtained from Eqs. (0.23) and (0.24) as
( x x0 ) ( y y0 ) ( z z0 ) (0.25)
w F wG w F wG w F wG w F wG w F wG w F wG

wy wz wz wy wz wx wx wz wx wy wy wx
or
( x x0 ) ( y y0 ) ( z z0 ) (0.26)
w (F , G) w (F , G) w (F , G)
w ( y, z ) w ( z , x) w ( x, y )
Therefore, the direction cosines of L are proportional to
ªw (F , G) w (F , G) w (F , G) º
« , , ». (0.27)
¬ w ( y , z ) w ( z , x ) w ( x, y ) ¼
For illustration, let us consider the following examples:
EXAMPLE 0.1 Find the tangent vector at (0, 1, π /2) to the helix described by the equation
x cos t , y sin t , z t, t I in IR c .
Solution The tangent vector to the helix at (x, y, z) is

§ dx dy dz ·
¨© , , ¸ ( sin t , cos t , 1).
dt dt dt ¹
We observe that the point (0, 1, π /2) corresponds to t π /2. At this point (0, 1, π /2), the tangent
vector to the given helix is (1, 0, 1).
EXAMPLE 0.2 Find the equation of the tangent line to the space circle
x2 y 2 z 2 1, x yz 0
at the point (1/ 14, 2/ 14, 3/ 14).
Solution The space circle is described as

F ( x, y , z ) x2 y 2 z 2 1 0
G ( x, y , z ) x yz 0
Recalling Eq. (0.25), the equation of the tangent plane at (1/ 14, 2/ 14, 3/ 14) can be
written as
x 1/ 14 y 2/ 14
2 § 3 · § 3 · § 1 ·
2u 2¨ ¸¹ 2 ©¨
© 14 ¹¸
2¨ ¸
14 © 14 14 ¹
z 3/ 14
§ 1 · § 2 ·
2¨ ¸ 2¨
© 14 ¹ © 14 ¹¸
or
x 1/ 14 y 2/ 14 z 3/ 14
.
10/ 14 8/ 14 2/ 14
0.4 FORMATION OF PARTIAL DIFFERENTIAL EQUATION

Suppose u and v are any two given functions of x, y and z. Let F be an arbitrary function of
u and v of the form
F (u , v) 0 (0.28)
We can form a differential equation by eliminating the arbitrary function F. For, we differentiate
Eq. (0.28) partially with respect to x and y to get
w F ªw u w u º w F ªw v w v º
p» p» 0
w u «¬ w x w z ¼ w v «¬ w x w z
(0.29)
¼
and
w F ªw u w u º w F ª w v w v º
q q 0
w u «¬ w y w z »¼ w v «¬ w y w z »¼ (0.30)
Now, eliminatin w F /w u and w F /w v from Eqs. (0.29) and (0.30), we obtain

wu wu wv wv
p p
wx wz wx wz
0
wu wu wv wv
q q
wy wz wy wz
which simplifies to
w (u , v) w (u , v) w (u , v)
p q . (0.31)
w ( y, z ) w ( z , x) w ( x, y )
This is a linear PDE of the type
Pp Qq R, (0.32)
where
w (u , v) w (u , v) w (u , v)
P , Q , R . (0.33)
w ( y, z ) w ( z , x) w ( x, y )
Equation (0.32) is called Lagrange’s PDE of first order. The following examples illustrate the
idea of formation of PDE.
EXAMPLE 0.3 Form the PDE by eliminating the arbitrary function from
(i) z f ( x it ) g ( x it ), where i 1
2 2 2
(ii) f ( x y z, x y z ) 0.
Solution
(i) Given z f ( x it ) g ( x it ) (1)
Differentiating Eq. (1) twice partially with respect to x and t, we get

wz
f c ( x it ) g c ( x it )
wx
w 2z
f cc ( x it ) g cc ( x it ). (2)
w x2
Here, f c indicates derivative of f with respect to ( x it ) and g c indicates derivative of g
with respect to ( x it ). Also, we have
wz
if c ( x it ) ig c( x it )
wt
w 2z
f cc ( x it ) g cc ( x it ). (3)
w t2
From Eqs. (2) and (3), we at once, find that
w 2z w 2z
0, (4)
w x2 w t2
which is the required PDE.
(ii) The given relation is of the form
G (u , v) 0,
where u x y z , v x 2 y 2 z 2
Hence, the required PDE is of the form
Pp Qq R, (Lagrange equation) (1)
where
wu wv
w (u , v) wy wy 1 2y
P 2 ( z y)
w ( y, z ) wu wv 1 2z
wz wz
wu wv
w (u , v) wz wz 1 2z
Q 2 ( x z)
w ( z, x) wu wv 1 2x
wx wx
and
wu wv
w (u , v) wx wx 1 2x
R 2 ( y x)
w ( x, y ) wu wv 1 2y
wy wy
Hence, the required PDE is

2 ( z y) p 2 ( x z) q 2 ( y x)
or
( z y ) p ( x z )q y x.
EXAMPLE 0.4 Eliminate the arbitrary function from the following and hence, obtain the
corresponding partial differential equation:
(i) z xy f ( x 2 y 2 )
(ii) z f ( xy/z ).
Solution
(i) Given z xy f ( x 2 y 2 ) (1)
Differentiating Eq. (1) partially with respect to x and y, we obtain
wz
y 2 xf c( x 2 y 2 ) p (2)
wx
wz
x 2 yf c ( x 2 y 2 ) q (3)
wy
Eliminating f c from Eqs. (2) and (3) we get
yp xq y 2 x2 , (4)
(ii) Given z f ( xy /z ) (1)
Differentiating partially Eq. (1) with respect to x and y, we get
wz y
f c( xy /z ) p (2)
wx z
wz x (3)
f c ( xy/z ) q
wy z
Eliminating f c from Eqs. (2) and (3), we find

xp yq 0
or
px qy (4)

EXAMPLE 0.5 Form the partial differential equation by eliminating the constants from
z ax by ab.
Solution Given z ax by ab (1)

Differentiating Eq. (1) partially with respect to x and y we obtain
wz
a p (2)
wx
wz
b q (3)
wy
Substituting p and q for a and b in Eq. (1), we get the required PDE as
z px qy pq
EXAMPLE 0.6 Find the partial differential equation of the family of planes, the sum of
whose x, y, z intercepts is equal to unity.
x y z
Solution Let 1 be the equation of the plane in intercept form, so
a b c
that a b c 1. Thus, we have
x y z
1 (1)
a b 1 a b
Differentiating Eq. (1) with respect to x and y, we have
1 p p 1
0 or (2)
a 1 a b 1 a b a
and
1 q q 1
0 or (3)
b 1 a b 1 a b b
From Eqs. (2) and (3), we get
p b
(4)
q a
Also, from Eqs. (2) and (4), we get
p
pa a b 1 a a 1
q
or
§ p ·
a ¨1 p ¸ 1.
© q ¹
Therefore,
q
a= (5)
( p + q − pq )
Similarly, from Eqs. (3) and (4), we find
p
b= (6)
( p + q - pq )
Substituting the values of a and b from Eqs. (5) and (6) respectively to Eq. (1), we have
p + q − pq p + q − pq p + q − pq
x+ y+ z =1
q p − pq
x y z 1
or + − = .
q p pq p + q − pq
That is,
pq
px + qy − z = , (7)
p + q − pq
0.5 SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS OF FIRST ORDER

In Section 0.4, we have observed that relations of the form
F ( x, y , z , a , b ) = 0 (0.34)
give rise to PDE of first order of the form
f ( x, y , z , p , q ) = 0 (0.35)
Thus, any relation of the form (0.34) containing two arbitrary constants a and b is a solution
of the PDE of the form (0.35) and is called a complete solution or complete integral.
Consider a first order PDE of the form
∂z ∂z
P ( x, y , z ) + Q ( x, y , z ) = R ( x, y , z ) (0.36)
∂x ∂y
or simply
Pp + Qq = R (0.37)
where x and y are independent variables. The solution of Eq. (0.37) is a surface S lying in
the ( x, y, z ) -space, called an integral surface. If we are given that z = f ( x, y ) is an integral
surface of the PDE (0.37). Then, the normal to this surface will have direction cosines
proportional to (w z /w x, w z /w y , 1) or ( p, q, 1). Therefore, the direction of the normal is

& &
given by n { p, q, 1}. From the PDE (0.37), we observe that the normal n is perpendicular
&
to the direction defined by the vector t {P, Q, R} (see Fig. 0.1).
(x, y, z)
n nt
n
x
Fig. 0.1 Integral surface z f ( x , y ).
Therefore, any integral surface must be tangential to a vector with components {P, Q, R}, and
hence, we will never leave the integral surface or solutions surface. Also, the total differential
dz is given by
wz wz
dz dx dy (0.38)
wx wy
From Eqs. (0.37) and (0.38), we find
{P, Q, R} {dx, dy, dz} (0.39)
Now, the solution to Eq. (0.37) can be obtained using the following theorem:
Theorem 0.1 The general solution of the linear PDE

Pp Qq R
can be written in the form F (u , v) 0, where F is an arbitrary function, and u ( x, y , z ) C1 and

v ( x, y , z ) C2 form a solution of the equation
dx dy dz (0.40)
P ( x, y , z ) Q ( x, y , z ) R ( x, y , z )
Proof We observe that Eq. (0.40) consists of a set of two independent ordinary differential
equations, that is, a two parameter family of curves in space, one such set can be written as
dy Q ( x, y , z ) (0.41)
dx P ( x, y , z )
which is referred to as “characteristic curve”. In quasi-linear case, Eq. (0.41) cannot be
evaluated until z ( x, y ) is known. Recalling Eqs. (0.37) and (0.38), we may recast them using
matrix notation as
ªP Q º § w z /w x · §R·
« dx (0.42)
¬ dy »¼ ¨©w z /w y ¸¹ ¨© dz ¸¹
Both the equations must hold on the integral surface. For the existence of finite solutions of
Eq. (0.42), we must have
P Q P R R Q
0 (0.43)
dx dy dx dz dz dy
on expanding the determinants, we have

dx dy dz (0.44)
P ( x, y , z ) Q ( x, y , z ) R ( x, y , z )
which are called auxiliary equations for a given PDE.
In order to complete the proof of the theorem, we have yet to show that any surface
generated by the integral curves of Eq. (0.44) has an equation of the form F (u, v) 0.
Let
u ( x, y , z ) C1 and v ( x, y , z ) C2 (0.45)
be two independent integrals of the ordinary differential equations (0.44). If Eqs. (0.45)
satisfy Eq. (0.44), then, we have
wu wu wu
dx dy dz du 0
wx wy wz
and
wv wv wv
dx dy dz dv 0.
wx wy wz
Solving these equations, we find
dx dy
wu wv wu wv wu wv wu wv

wy wz wz wy wz wx wx wz
dz
wu wv wu wv

wx wy wy wx
which can be rewritten as

dx dy dz
(0.46)
w (u , v) w (u , v) w (u , v)
w ( y, z ) w ( z , x) w ( x, y )
Now, we may recall from Section 0.4 that the relation F (u , υ ) 0, where F is an arbitrary
function, leads to the partial differential equation
w (u , v) w (u , v) w (u , v)
p q (0.47)
w ( y, z ) w ( z , x) w ( x, y )
By virtue of Eqs. (0.37) and (0.47), Eq. (0.46) can be written as
dx dy dz
P Q R
The solution of these equations are known to be u ( x, y , z ) C1 and v ( x, y , z ) C2 . Hence,

F (u , v ) 0 is the required solution of Eq. (0.37), if u and v are given by Eq. (0.45),
We shall illustrate this method through following examples:
EXAMPLE 0.7 Find the general integral of the following linear partial differential equations:
(i) y 2 p xy q x ( z 2 y)
(ii) ( y zx) p ( x yz ) q x2 y 2 .
Solution
(i) The integral surface of the given PDE is generated by the integral curves of the
auxiliary equation
dx dy dz (1)
y 2 xy x ( z 2 y)
The first two members of the above equation give us
dx dy
or x dx y dy ,
y x
which on integration results in
x2 y2
C or x2 y 2 C1 (2)
2 2
The last two members of Eq. (1) give
dy dz
or z dy 2 y dy y dz
y z 2y
That is,
2 y dy y dz z dy ,
which on integration yields
y2 yz C2 or y 2 yz C2 (3)
Hence, the curves given by Eqs. (2) and (3) generate the required integral surface as
F ( x 2 y 2 , y 2 yz ) 0.
(ii) The integral surface of the given PDE is generated by the integral curves of the
auxiliary equation
dx dy dz (1)
y zx ( x yz ) x y2
2
To get the first integral curve, let us consider the first combination as
x dx y dy dz
2 2
xy zx xy y z x y2
2
or
x dx y dy dz .
2 2
z (x y ) x y2
2
That is,
x dx y dy z dz.
On integration, we get
x2 y 2 z 2
C or x2 y 2 z 2 C1 (2)
2 2 2
Similarly, for getting the second integral curve, let us consider the combination such as
y dx x dy dz
2 2
y xyz x xyz x y2
2
or
y dx x dy dz 0,
which on integration results in
xy z C2 (3)
Thus, the curves given by Eqs. (2) and (3) generate the required integral surface as
F ( x 2 y 2 z 2 , xy z ) 0.
EXAMPLE 0.8 Use Lagrange’s method to solve the equation
x y z
α β γ
0
wz wz
1
wx wy
where z z ( x, y ).
Solution The given PDE can be written as

ª wzº ª w zº ª wz w zº
x « β γ y « α γ z «α β 0
¬ w y »¼ ¬ »
w x¼ ¬ wy w x »¼
or
wz wz
(γ y β z ) (α z γ x) β x α y (1)
wx wy
The corresponding auxiliary equations are
dx dy dz (2)
(γ y β z ) (α z γ x) ( β x α y)
Using multipliers x, y, and z we find that each fraction is
x dx y dy z dz
.
0
Therefore,
x dx y dy z dz 0,
x2 y 2 z 2 C1 (3)
Similarly, using multipliers α , β , and γ , we find from Eq. (2) that each fraction is equal to
α dx β dy γ dz 0,
which on integration gives
αx β y γ z C2 (4)
Thus, the general solution of the given equation is found to be
F ( x2 y 2 z 2 , α x β y γ z) 0
EXAMPLE 0.9 Find the general integrals of the following linear PDEs:
(i) pz qz z 2 ( x y )2
(ii) ( x 2 yz ) p ( y 2 zx) q z 2 xy.
Solution
(i) The integral surface of the given PDE is generated by the integral curves of the
auxiliary equation
dx dy dz
(1)
z z z ( x y )2
2
The first two members of Eq. (1) give

dx dy 0,
x y C1 (2)
Now, considering Eq. (2) and the first and last members of Eq. (1), we obtain
z dz
dx
z C 12
2
2 z dz
or 2 dx,
z C 12
2

ln ( z 2 C 12 ) 2 x C2
or
ln [ z 2 ( x y )2 ] 2 x C2 (3)
Thus, the curves given by Eqs. (2) and (3) generates the integral surface for the given PDE
as
F ( x y, log {x 2 y 2 z 2 2 xy} 2 x) 0
(ii) The integral surface of the given PDE is given by the integral curves of the auxiliary
equation
dx dy dz
(1)
2 2 2
x yz y zx z xy
Equation (1) can be rewritten as
dx dy dy dz dz dx (2)
( x y) ( x y z) ( y z) ( x y z) ( z x) ( x y z )
Considering the first two terms of Eq. (2) and integrating, we get
ln ( x y ) ln ( y z ) ln C1
x y
or C1 (3)
yz
Similarly, considering the last two terms of Eq. (2) and integrating, we obtain
yz
C2 (4)
zx
Thus, the integral curves given by Eqs. (3) and (4) generate the integral surface
§x y y z·
F¨ , 0.
© y z z x ¸¹
0.6 INTEGRAL SURFACES PASSING THROUGH A GIVEN CURVE

In the previous section, we have seen how a general solution for a given linear PDE can be
obtained. Now, we shall make use of this general solution to find an integral surface containing
a given curve as explained below:
Suppose, we have obtained two integral curves described by
u ( x, y , z ) C1 ½°
¾ (0.48)
v ( x, y , z ) C2 °¿
from the auxiliary equations of a given PDE. Then, the solution of the given PDE can be
written in the form
F (u , v) 0 (0.49)
Suppose, we wish to determine an integral surface, containing a given curve C described
by the parametric equations of the form
x x(t ), y y (t ), z z (t ), (0.50)
where t is a parameter. Then, the particular solution (0.48) must be like
u {x(t ), y (t ), z (t )} C1 »¯
¼ (0.51)
v {x(t ), y (t ), z (t )} C2 ¯½
Thus, we have two relations, from which we can eliminate the parameter t to obtain a relation
of the type
F (C1 , C2 ) 0, (0.52)
which leads to the solution given by Eq. (0.49). For illustration, let us consider the following
couple of examples.
EXAMPLE 0.10 Find the integral surface of the linear PDE

x ( y 2 z) p y ( x2 z) q ( x2 y2 ) z
containing the straight line x y 0, z 1.
Solution The auxiliary equations for the given PDE are

dx dy dz
(1)
2 2
x ( y z) y ( x z) (x y2 )z
2
Using the multiplier xyz, we have

yz dx zx dy xy dz 0.
xyz C1 (2)
Suppose, we use the multipliers x, y and z. Then find that each fraction in Eq. (1) is equal
to
x dx y dy z dz 0,
x2 y 2 z 2 C2 (3)
For the curve in question, we have the equations in parametric form as
x t, y t , z 1
Substituting these values in Eqs. (2) and (3), we obtain
t 2C1 ½°
¾ (4)
2t 2 1 C2 °¿
Eliminating the parameter t, we find

1 2C1 C2
or
2C1 C2 1 0
Hence, the required integral surface is
x 2 y 2 z 2 2 xyz 1 0.
EXAMPLE 0.11 Find the integral surface of the linear PDE

xp yq z
which contains the circle defined by
x2 y 2 z 2 4, x yz 2.
Solution The integral surface of the given PDE is generated by the integral curves of
the auxiliary equation
dx dy dz (1)
x y z
Integration of the first two members of Eq. (1) gives
ln x ln y ln C
or
x (2)
C1
y
Similarly, integration of the last two members of Eq. (1) yields
y
C2 (3)
z
Hence, the integral surface of the given PDE is
§ x y· (4)
F¨ , ¸ 0
©y z¹
If this integral surface also contains the given circle, then we have to find a relation between
x/y and y/z.
The equation of the circle is
x2 y2 z 2 4 (5)
x yz 2 (6)
From Eqs. (2) and (3), we have
y x /C1 , z y /C2 x /C1C2
Substituting these values of y and z in Eqs. (5) and (6), we find
x2 x2 § 1 1 ·
x2 4, or x 2 ¨1 2 2 2 ¸ 4 (7)
C 12 C 12C 22 ¨© C 1 C 1 C 2 ¸¹
and
x x § 1 1 ·
x 2, or x ¨1 2 (8)
© ¸
C1 C1C2 C1 C1C2 ¹
From Eqs. (7) and (8) we observe

2
1 1 § 1 1 ·
1 ¨©1 C C C ¸¹ ,
C 12 C 12C 22 1 1 2
which on simplification gives us

2 2 2
2 0
C1 C1C2 C 1 C2
That is,
C1C2 C1 1 0. (9)
Now, replacing C1 by x/y and C2 by y/z, we get the required integral surface as
x y x
1 0,
y z y
x x
or 1 0,
z y
or xy xz yz 0.
0.7 THE CAUCHY PROBLEM FOR FIRST ORDER EQUATIONS
Consider an interval I on the real line. If x0 ( s ), y0 ( s ) and z0 ( s ) are three arbitrary functions
of a single variable s I such that they are continuous in the interval I with their first
derivatives. Then, the Cauchy problem for a first order PDE of the form
F ( x, y , z , p, q ) 0 (0.53)
is to find a region IR in ( x, y ), i.e. the space containing ( x0 ( s ), y0 ( s )) for all s I , and a

solution z φ ( x, y ) of the PDE (0.53) such that
Z [ x0 ( s ), y0 ( s )] Z0 (s)
and φ ( x, y ) together with its partial derivatives with respect to x and y are continuous functions
of x and y in the region IR .
Geometrically, there exists a surface z φ ( x, y ) which passes through the curve Γ, called
datum curve, whose parametric equations are
x x0 ( s ), y y0 ( s ), z z0 ( s )
and at every point of which the direction ( p, q, 1) of the normal is such that
F ( x, y , z , p, q ) 0
This is only one form of the problem of Cauchy.
In order to prove the existence of a solution of Eq. (0.53) containing the curve (, we
have to make further assumptions about the form of the function F and the nature of Γ. Based
on these assumptions, we have a whole class of existence theorems which is beyond the scope
of this book. However, we shall quote one form of the existence theorem without proof,
which is due to Kowalewski (see Senddon, 1986).
Theorem 0.2 If
(i) g(y) and all of its derivatives are continuous for | y y0 | δ ,
(ii) x0 is a given number and z0 g ( y0 ), q0 g c ( y0 ) and f ( x, y, z, q) and all of its partial
derivatives are continuous in a region S defined by
| x x0 | δ , | y y0 | δ , | q q0 | δ ,
then, there exists a unique function φ ( x, y ) such that

(a) φ ( x, y ) and all of its partial derivatives are continuous in a region IR defined by
| x x0 | δ1 , | y y0 | δ 2 ,
(b) For all ( x, y ) in IR, z φ ( x, y ) is a solution of the equation
wz § wz·
f ¨ x, y , z , and
wx © w y ¸¹
(c) For all values of y in the interval | y y0 | δ1 , φ ( x0 , y ) g ( y ).
0.8 SURFACES ORTHOGONAL TO A GIVEN SYSTEM OF SURFACES

One of the useful applications of the theory of linear first order PDE is to find the system
of surfaces orthogonal to a given system of surfaces. Let a one-parameter family of surfaces
is described by the equation
F ( x, y , z ) C (0.54)
Then, the task is to determine the system of surfaces which cut each of the given surfaces
orthogonally. Let ( x, y, z ) be a point on the surface given by Eq. (0.54), where the normal
to the surface will have direction ratios (w F /w x, w F /w y, w F /w z ) which may be denoted by P,
Q, R.
Let
Z φ ( x, y ) (0.55)
be the surface which cuts each of the given system orthogonally (see Fig. 0.2).
Z = φ(x, y)
F = C1
F = C2
F = C3
Fig. 0.2 Orthogonal surface to a given system of surfaces.
Then, its normal at the point ( x, y, z ) will have direction ratios (w z /w x, w z /w y , 1) which, of
course, will be perpendicular to the normal to the surfaces characterized by Eq. (0.54). As a
consequence we have a relation
wz wz
P Q R 0 (0.56)
wx wy
or
Pp Qq R (0.57)
which is a linear PDE of Lagranges type, and can be recast into
wF wz wF wz wF (0.58)

wx wx wy wy wz
Thus, any solution of the linear first order PDE of the type given by either Eq. (0.57) or (0.58)
is orthogonal to every surface of the system described by Eq. (0.54). In other words, the
surfaces orthogonal to the system (0.54) are the surfaces generated by the integral curves of
the auxiliary equations
dx dy dz (0.59)
w F /w x w F /w y w F /w z
0.9 FIRST ORDER NON-LINEAR EQUATIONS

In this section, we will discuss the problem of finding the solution of first order non-linear
partial differential equations (PDEs) in three variables of the form
F ( x, y , z , p , q ) 0, (0.60)
where
wz wz
p , q .
wx wy
We also assume that the function possesses continuous second order derivatives with respect
to its arguments over a domain Ω of ( x, y, z , p, q)-space, and either Fp or Fq is not zero at
every point such that
F p2 F q2 z 0.
The PDE (0.60) establishes the fact that at every point ( x, y, z ) of the region, there exists
a relation between the numbers p and q such that φ ( p, q ) 0, which defines the direction of
&
the normal n { p, q, 1} to the desired integral surface z z ( x, y ) of Eq. (0.60). Thus, the
direction of the normal to the desired integral surface at certain point (x, y, z) is not defined
uniquely. However, a certain cone of admissable directions of the normals exist satisfying the
relation φ ( p, q ) 0 (see Fig. 0.3).
O y
x
Fig. 0.3 Cone of normals to the integral surface.
Therefore, the problem of finding the solution of Eq. (0.60) reduces to finding an integral
surface z z ( x, y ), the normals at every point of which are directed along one of the permissible
directions of the cone of normals at that point.
Thus, the integral or the solution of Eq. (0.60) essentially depends on two arbitrary
constants in the form
f ( x, y , z , a , b ) 0, (0.61)
which is called a complete integral. Hence, we get a two-parameter family of integral surfaces
through the same point.
0.9.1 Cauchy’s Method of Characteristics

The integral surface z = z ( x, y ) of Eq. (0.60) that passes through a given curve x0 = x0(s),
y0 = y0(s), z0 = z0(s) may be visualized as consisting of points lying on a certain one-parameter
family of curves x = x (t , s), y = y (t , s ), z = z (t , s ), where s is a parameter of the family called
characteristics.
Here, we shall discuss the Cauchy’s method for solving Eq. (0.60), which is based
on geometrical considerations. Let z = z ( x, y ) represents an integral surface S of Eq. (0.60)
in (x, y, z)-space. Then, { p, q, − 1} are the direction ratios of the normal to S. Now, the
differential equation (0.60) states that at a given point P ( x0 , y0 , z0 ) on S, the relationship
between p0 and q0, that is F ( x0 , y0 , z0 , p0 , q0 ), need not be necessarily linear. Hence, all the
tangent planes to possible integral surfaces through P form a family of planes enveloping a
conical surface called Monge Cone with P as its vertex. In other words, the problem of
solving the PDE (0.60) is to find surfaces which touch the Monge cone at each point along
a generator. For example, let us consider the non-linear PDE
p 2 − q 2 = 1. (0.62)
At every point of the xyz-space, the relation (0.62) can be expressed parametrically as
p = cosh μ , q = sinh μ , −∞ < μ < ∞ (0.63)
Then, the equation of the tangent planes at ( x0 , y0 , z0 ) can be written as
( x − x0 ) cosh μ + ( y − y0 ) sinh μ − ( z − z0 ) = 0 (0.64)
Let P ( x0 , y0 , z0 ) be the vertex and Q( x, y , z ) be any point on the generator. Then, the direction
ratios of the generator are ( x − x0 ), ( y − y0 ), ( z − z0 ). Now, the direction ratios of the axis of
the cone which is parallel to x-axis are (1, 0, 0) (see Fig. 0.4). Let the semi-verticle angle of
the cone be π /4. Then,
π ( x − x0 )1 + ( y − y0 ) 0 + ( z − z0 ) 0 1
cos = =
4 2 2
( x − x0 ) + ( y − y0 ) + ( z − z0 ) 2 2
or
( x − x0 )2 + ( y − y0 )2 + ( z − z0 )2 = 2( x − x0 )2
or
( x − x0 )2 − ( y − y0 )2 − ( z − z0 )2 = 0 (0.65)
Thus, we see that the Monge cone of the PDE (0.62) is given by Eq. (0.65). This is a right
circular cone with semi-vertical angel π /4 whose axis is the straight line passing through
( x0 , y0 , z0 ) and parallel to z-axis.
z P (x0, y0, z0)
/4
Q (x, y, z)
O y
x
Fig. 0.4 Monge cone.
Since an integral surface is touched by a Monge cone along its generator, we must have a
method to determine the generator of the Monge cone of the PDE (0.60) which is explained
below:
It may be noted that the equation of the tangent plane to the integral surface z z ( x, y ) at
the point ( x0 , y0 , z0 ) is given by
p ( x x0 ) q ( y y0 ) ( z z0 ). (0.66)
Now, the given non-linear PDE (0.60) can be recasted into an equivalent form as
q q ( x0 , y0 , z0 , p ) (0.67)
indicating that p and q are not independent at ( x0 , y0 , z0 ). At each point of the surface S,
there exists a Monge cone which touches the surface along the generator of the cone. The
lines of contact between the tangent planes of the integral surface and the corresponding
cones, that is the generators along which the surface is touched, define a direction field on
the surface S. These directions are called the characteristic directions, also called Monge
directions on S and lie along the generators of the Monge cone. The integral curves of this
field of directions on the integral surface S define a family of curves called characteristic
curves as shown in Fig. 0.5. The Monge cone can be obtained by eliminating p from the
following equations:
Fig. 0.5 Characteristic directions on an integral surface.

p ( x x0 ) q ( x0 , y0 , z0 , p ) ( y y0 ) ( z z0 ) (0.68)
and
dq
( x x0 ) ( y y0 ) 0. (0.69)
dp
observing that q is a function of p and differentiating Eq. (0.60) with respect to p, we get
dF w F w F dq
0. (0.70)
dp w p w q dp
Now, eliminating (dq /dp ) from Eqs. (0.69) and (0.70), we obtain
w F w F ( x x0 )
0
w p w q ( y y0 )
or
x x0 y y0 (0.71)
Fp Fq
Therefore, the equations describing the Monge cone are given by
q q ( x0 , y0 , z0 , p),½°
( x x0 ) p ( y y0 ) q ( z z0 ) °
°°
¾ (0.72)
and °
x x0 y y0 °
. °
Fp Fq °¿
The second and third of Eqs. (0.72) define the generator of the Monge cone. Solving them
for ( x x0 ), ( y y0 ) and ( z z0 ), we get
x x0 y y0 z z0 (0.73)
Fp Fq pFp qFq
Finally, replacing ( x x0 ), ( y y0 ) and ( z z0 ) by dx, dy and dz respectively, which corresponds

to infinitesimal movement from ( x0 , y0 , z0 ) along the generator, Eq. (0.73) becomes
dx dy dz (0.74)
.
Fp Fq pFp qFq
Denoting the ratios in Eq. (0.74) by dt, we observe that the characteristic curves on S can be
obtained by solving the ordinary differential equations
dx
Fp {x, y , z ( x, y ), p ( x, y ), q ( x, y )} (0.75)
dt
and
dy
Fq {x, y, z ( x, y ), p ( x, y ), q ( x, y )}. (0.76)
dt
Also, we note that
dz w z dx w z dy dx dy
p q
dt w x dt w y dt dt dt
Therefore,
dz
pFp qFq (0.77)
dt
Along the characteristic curve, p is a function of t, so that
dp w p dx w p dy

dt w x dt w y dt
Now, using Eqs. (0.75) and (0.76), the above equation becomes
dp w p wF w p wF
.
dt wx w p wy wq
Since z xy z yx or p y qx , we have
dp w p wF wq wF
(0.78)
dt wx w p wx wq
Also, differentiating Eq. (0.60) with respect to x, we find
wF wF wF w p w F wq
p 0 (0.79)
wx wz w p wx wq wx
Using Eq. (0.79), Eq. (0.78) becomes
dp
( Fx pFz ) (0.80)
dt
Similarly, we can show that
dq
( Fy qFz ) (0.81)
dt
Thus, given an integral surface, we have shown that there exists a family of characteristic
curves along which x, y, z, p and q vary according to Eqs. (0.75), (0.76), (0.77), (0.80) and
(0.81). Collecting these results together, we may write
dx ½
Fp , °
dt
°
dy °
Fq , °
dt °
dz °°
pFp qFq , ¾ (0.82)
dt °
dp °
( Fx pFz ) and °
dt °
°
dq °
( Fy qFz ).
dt °¿
These equations are known as characteristic equations of the given PDE (0.60). The last three
equations of (0.82) are also called compatibility conditions. Without knowing the solution
z z ( x, y ) of the PDE (0.60), it is possible to find the functions x (t ), y (t ), z (t ), p(t ), q(t ) from
Eqs. (0.82). That is, we can find the curves x x(t ), y y (t ), z z (t ) called characteristics
and at each point of a characteristic, we can find the numbers p p (t ) and q q(t ) that
determine the direction of the plane
p ( X x) + q(Y y ) = ( Z z ). (0.83)
The characteristics, together with the plane (0.83) referred to each of its points is called a
characteristic strip. The solution x x (t ), y y (t ), z z (t ), p p (t ), q q (t ) of the
characteristic equations (0.82) satisfy the strip condition
dz dx dy
p (t ) q (t ) (0.84)
dt dt dt
It may be noted that not every set of five functions can be interpreted as a strip. A strip should
satisfy that the planes with normals ( p, q, 1) be tangential to the characteristic curve. That is, they
must satisfy the strip condition (0.84) and the normals should vary continuously along the
curve.
An important consequence of the Cauchy’s method of characteristic is stated in the
following theorem.
Theorem 0.3 Along every strip (characteristic strip) of the PDE: F ( x, y, z , p, q ) 0, the
function F ( x, y, z , p, q) is constant.
Proof Along the characteristic strip, we have

d w F dx w F dy w F dz w F dp w F dq
F{ x (t ), y (t ), z (t ), p(t ), q (t )}
dt w x dt w y dt w z dt w p dt w q dt
Now, using the results listed in Eq. (0.82), the right-hand side of the above equation becomes
Fx Fp Fy Fq Fz ( pFp qFq ) Fp ( Fx pFz ) Fq ( Fy qFz ) 0.
Hence, the function F ( x, y, z , p, q) is constant along the strip of the characteristic equations
of the PDE defined by Eq. (0.60).
For illustration, we consider the following examples:
EXAMPLE 0.12 Find the characteristics of the equation pq z and determine the integral
surface which passes through the straight line x 1, z y.
Solution If the initial data curve is given in parametric form as

x0 ( s ) 1, y0 ( s ) s, z0 ( s ) s,
then ordinarily the solution is sought in parametric form as
x x (t , s ), y y (t , s ), z z (t , s ).
Thus, using the given data, the differential equation becomes
p0 ( s )q0 ( s ) s 0 F (1)
and the strip condition gives
1 p0 (0) q0 (1) or q0 1. (2)
Therefore,
q0 1, p0 s (unique initial strip) (3)
Now, the characteristic equations for the given PDE are
dx dy dz dp dq
q, p, 2 pq, p, q (4)
dt dt dt dt dt
p c1 exp (t ), q c2 exp (t ), x c2 exp (t ) c3 ½°

¾ (5)
y c1 exp (t ) c4 , z 2c1c2 exp (2t ) c5 °¿
Now, taking into account the initial conditions
x0 1, y0 s , z0 s, p0 s, q0 1 (6)
we can determine the constants of integration and obtain (since c2 1, c3 0)
p s exp (t ), q exp (t ), x exp (t ) ½°

¾ (7)
y s exp (t ), z s exp (2t ) °¿
Consequently, the required integral surface is obtained from Eq. (7) as

z xy.
EXAMPLE 0.13 Find the characteristics of the equation pq z and hence, determine the
integral surface which passes through the parabola x 0, y 2 z.
Solution The initial data curve is
x0 ( s ) 0, y0 ( s ) s , z0 ( s ) s2.
Using this data, the given PDE becomes
p0 ( s ) q0 ( s) s 2 0 F (1)
The strip condition gives
2s p0 (0) q0 (1) or q0 2 s 0 (2)
Therefore,
s
q0 2s and p0 z0 /q0 s 2/2 s (3)
2
Now, the characteristic equations of the given PDE are given by
dx dy dz dp dq
q, p, 2 pq, p, q (4)
dt dt dt dt dt
On integration, we obtain
p c1 exp (t ), q c2 exp (t ), x c2 exp (t ) c3 ½°

¾ (5)
y c1 exp (t ) c4 , z c1c2 exp (2t ) c5 °¿
Taking into account the initial conditions
x0 0, y0 s, z0 s 2 , p0 s /2, q0 2 s,
we find
c1 s /2, c2 2 s, c3 2 s, c4 s /2, c5 0
Therefore, we have
s ½
p exp (t ), q 2 s exp (t ), °
2
°
s ° (6)
x 2 s [exp (t ) 1], y [exp (t ) 1] ¾
2 °
°
z s 2 exp (2t ) °
°¿
Eliminating s and t from the last three equations of (6), we get
16 z (4 y x)2 .
This is the required integral surface.
EXAMPLE 0.14 Find the characteristics of the PDE
p2 q2 2
and determine the integral surface which passes through x 0, z y.
Solution The initial data curve is

x0 ( s ) 0, y0 ( s ) s , z0 ( s ) s.
Using this data, the given PDE becomes
p 02 q 02 2 0 F (1)
and the strip condition gives

1 p0 (0) q0 (1) or q0 1 0 (2)
Hence,
q0 1, p0 r1 (3)
Now, the characteristic equations for the given PDE are given by
dx dy dz ½
2 p, 2q, 2 p 2 2q 2 4°
dt dt dt ° (4)
¾
dp dq °
0, 0 °¿
dt dt
p c1 , q c2 , x 2c1t c3 ½°
¾ (5)
y 2c2 t c4 , z 4t c5 °¿
Taking into account the initial conditions

x0 0, y0 s , z0 s, p0 r1, q0 1,
we find
p r1, q 1, x r2t ,½°

¾ (6)
y 2t s, z 4t s. °¿
The last-three equations of (6) are parametric equations of the desired integral surface. Eliminating
the parameters s and t, we get
z y r x.
0.10 COMPATIBLE SYSTEMS OF FIRST ORDER EQUATIONS

Two first order PDEs are said to be compatible, if they have a common solution. We shall
now derive the necessary and sufficient conditions for the two partial differential equations
f ( x, y , z , p , q ) 0 (0.85)
and
g ( x, y , z , p , q ) 0 (0.86)
to be compatible.
w ( f , g)
Let J z0 (0.87)
w ( p, q )
Since Eqs. (0.85) and (0.86) have common solution, we can solve them and obtain
explicit expressions for p and q in the form
p φ ( x, y , z ), q ψ ( x, y, z ) (0.88)
and then, the differential relation
p dx q dy dz
or
φ ( x, y, z ) dx ψ ( x, y, z ) dy dz (0.89)
should be integrable, for which the necessary condition is
& &
X curl X 0
where X {φ , ψ , 1}. That is,
iˆ ˆj kˆ
ˆ ˆ ˆ
(φ i ψ j k ) w /w x w /w y w /w z 0
φ ψ 1
or
φ (ψ z ) ψ (φ z ) ψ x φ y
ψ x φψ z φ y ψφ z (0.90)
Now, differentiating Eq. (0.85) with respect to x and z, we get
wp wq
fx f p fq 0
wx wx
and
wp wq
fz f p fq 0
wz wz
But, from Eq. (0.89), we have
w p wφ w q wψ
, and so on.
wx wx wx wx
Using these results, the above equations can be recast into
f x f pφ x f qψ x 0
and
f z f pφ z f qψ z 0.
Multiplying the second one of the above pair by φ and adding to the first one, we readily
obtain
( f x φ f z ) f p (φ x φφ z ) f q (ψ x φψ z ) 0
Similarly, from Eq. (0.86) we can deduce that

( g x φ g z ) g p (φ x φφ z ) g q (ψ x φψ z ) 0
Solving the above pair of equations for (ψ x φψ z ), we have
(ψ x φψ z ) 1 1
f p (gx φ gz ) g p ( fx φ fz ) f q g p gq f p J
or
1
ψ x φψ z [( f p g x g p f x ) φ ( f p g z g p f z )]
J
1 ªw ( f , g ) w ( f , g)º
« φ (0.91)
J ¬ w ( x, p ) w ( z , p ) »¼
where J is defined in Eq. (0.87). Similarly, differentiating Eq. (0.85) with respect to y and z
and using Eq. (0.88), we can show that
1 ªw ( f , g ) w ( f , g) º
φ y ψφ z « ψ (0.92)
J ¬ w ( y, q) w ( z , q) »¼
Finally, substituting the values of ψ x φψ z and φ y ψφ z from Eqs. (0.91) and (0.92) into
Eq. (0.90), we obtain
w ( f , g) w ( f , g) ªw ( f , g ) w ( f , g) º
φ « ψ
w ( x, p ) w ( z, p) ¬ w ( y, q) w ( z , q) »¼
In view of Eqs. (0.88), we can replace φ and ψ by p and q, respectively to get
w ( f , g) w ( f , g) w ( f , g) w ( f , g)
p q 0 (0.93)
w ( x, p ) w ( z , p) w ( y, q) w ( z, q)
This is the desired compatibility condition. For illustration, let us consider the following
examples:
EXAMPLE 0.15 Show that the following PDEs

xp yq x and x2p q xz
are compatible and hence, find their solution.
Solution Suppose, we have
f xp yq x 0, (1)
and
g x 2 p q xz 0. (2)
Then,
w ( f , g) ( p 1) x
px 2 x 2 2 x 2 p xz xz x 2 p x 2 ,
w ( x, p ) (2 xp z ) x2
w ( f , g) 0 x
x2 ,
w ( z, p) x x2
w ( f , g) q y
q,
w ( y, q) 0 1
w ( f , g) 0 y
xy.
w ( z, q) x 1
and we find
w ( f , g) w ( f , g) w ( f , g) w ( f , g)
p q xz x 2 p x 2 px 2 q qxy
w ( x, p ) w ( z, p) w ( y, q) w ( z , q)
xz q qxy x 2
xz q x (qy x)
xz q x 2 p
0
Hence, the given PDEs are compatible.
Now, solving Eqs. (1) and (2) for p and q, we obtain
p q 1
xyz x 3
x x z 2
x x2 y
from which we get
x (1 yz ) 1 yz
p
x (1 xy ) 1 xy
and
x 2 ( z x) x ( z x)
q
x (1 xy ) 1 xy
In order to get the solution of the given system, we have to integrate Eq. (0.89), that is
(1 yz ) x ( z x)
dz dx dy (3)
1 xy 1 xy
or
y ( z x) x ( z x)
dz dx dx dy
1 xy 1 xy
or
dz dx y dx x dy
zx 1 xy
ln ( z x) ln (1 xy ) ln c.
That is,
z x c (1 xy )
Hence, the solution of the given system is found to be
z x c (1 xy ), (4)
which is of one-parameter family.
0.11 CHARPITS METHOD

In this section, we will discuss a general method for finding the complete integral or complete
solution of a nonlinear PDE of first order of the form
f ( x, y , z , p , q ) 0. (0.94)
This method is known as Charpit’s method. The basic idea in Charpit’s method is the introduction
of another PDE of first order of the form
g ( x, y , z , p, q ) 0 (0.95)
and then, solve Eqs. (0.94) and (0.95) for p and q and substitute in
dz p ( x, y , z , a ) dx q ( x, y , z , a ) dy. (0.96)
Now, the solution of Eq. (0.96) if it exists is the complete integral of Eq. (0.94).
The main task is the determination of the second equation (0.95) which is already discussed
in the previous section. Now, what is required, is to seek an equation of the form
g ( x, y , z , p, q ) 0
compatible with the given equation
f ( x, y , z , p , q ) 0
for which the necessary and sufficient condition is
w ( f , g) w ( f , g) w ( f , g) w ( f , g)
p q 0. (0.97)
w ( x, p ) w ( z , p) w ( y, q) w ( z, q)
On expansion, we have
§w f w g w f w g · §w f w g w f w g ·
¨© w x w p w p w x ¸¹ p ¨© w z w p w p w z ¸¹
§w f w g w f w g · §w f w g w f w g ·
¨ ¸ q¨ 0
©w y wq wq w y ¹ © w z w q w q w z ¸¹
which can be recast into
wg wg wg wg wg
fp fq ( pf p qf q ) ( f x pf z ) ( f y qf z ) 0. (0.98)
wx wy wz wp wq
This is a linear PDE, from which we can determine g. The auxiliary equations of (0.98) are
dx dy dz dp
fp fq pf p qf q ( f x pf z )
dq
(0.99)
( f y qf z )
These equations are called Charpit’s equations. Any integral of Eq. (0.99) involving p or q
or both can be taken as the second relation (0.95). Then, the integration of Eq. (0.96) gives
the complete integral as desired. It may be noted that all charpits equations need not be used,
but it is enough to choose the simplest of them. This method is illustrated through the
following examples:
EXAMPLE 0.16 Find the complete integral of

( p2 q2 ) y qz
Solution Suppose
f ( p 2 q 2 ) y qz 0 (1)
then, we have
fx 0, fy p2 q2 , fz q
fp 2 py, fq 2qy z ,
Now, the charpits auxiliary equations are given by

dx dy dz dp dq
fp fq pf p qf q ( f x pf z ) ( f y qf z )
That is,
dx dy dz
2 py 2qy z 2 p y 2q 2 y qz
2
dp dq
(2)
pq [( p q 2 ) q 2 ]
2
From the last two members of Eq. (2), we have

dp dq
pq p2
or
p dp q dq 0
p2 q2 a (constant) (3)
From Eqs. (1) and (3), we obtain
ay qz 0 or q ay/ z ½
and °
° (4)
2 ¾
§ ay · 2 2 2 2 °
p a¨ ¸ (az a y )/z °
© z ¹ ¿
Substituting these values of p and q in

dz p dx q dy,
we get
az 2 a 2 y 2 ay
dz dx dy
z z
or
z dz ay dy az 2 a 2 y 2 dx
d (az 2 a 2 y 2 )1/2
dx
a
On integration, we find
az 2 a 2 y 2
xb
a
or
( x b) 2 ( z 2/ a) y 2
Hence, the complete integral is
( x b) 2 y 2 z 2/ a.
EXAMPLE 0.17 Find the complete integral of the PDE:

z2 pq xy.
Solution In this example, given
f z 2 pq xy. (1)
Then, we have
fx pqy, fy pqx, fz 2z
fp qxy, fq pxy.
Now, the Charpit’s auxiliary equations are given by

dx dy dz dp dq
.
fp fq pf p qf q ( f x pf z ) ( f y qf z )
That is,
dx dy dz dp dq (2)
qxy pxy 2 pq xy pqy 2 pz pqx 2qz
From Eq. (2), it follows that
dp / p dq /q dx /x dy / y
qy 2 z px 2 z qy px
dp / p dq /q dx/x dy / y
qy px qy px
or
dp dq dy dx

p q y x
p x
c (constant)
q y
or
p cqy /x
From the given PDE, we have
z2 pq xy cq 2 y 2
which gives
q2 z 2/ cy 2 or q z/ c y az /y,
where a 1/ c .
Hence,
p z /ax.
dz p dx q dy,
we get
z az
dz dx dy
ax y
or
dz 1 dx dy
a
z a x y
1
ln z ln x a ln y ln b
a
or
z bx1/a y a
which is the complete integral of the given PDE.

x2 p2 y 2 q2 4 0
using Charpit’s method.
Solution The Charpit’s equations for the given PDE can be written as
dx dy dz dp
2 2 2 2 2 2
2x p 2y q 2 (x p y q ) 2 xp 2
dq (1)
2
2 yq
Considering the first and last but one of Eq. (1), we have
dx dp dx dp
2 2
or 0
2x p 2 xp x p
ln ( xp ) ln a or xp a (2)
From the given PDE and using the result (2), we get
y2q2 4 a2 (3)
Substituting one set of p and q values from Eqs. (2) and (3) in
dz p dx q dy,
we find that
dx dy
dz a 4 a2 .
x y
On integration, the complete integral of the given PDE is found to be
z a ln x 4 a 2 ln y b.
0.11.1 Special Types of First Order Equations

Type I Equations Involving p and q only.
That is, equations of the type
f ( p, q ) 0. (0.100)
Let z ax by c 0 is a solution of the given PDE, described by f ( p, q ) 0, then
wz wz
p a, q b.
wx wy
Substituting these values of p and q in the given PDE, we get
f ( a, b ) 0 (0.101)
Solving for b, we get, b φ (a ), say. Then,

z ax φ (a ) y c (0.102)
is the complete integral of the given PDE.
EXAMPLE 0.19 Find a complete integral of the equation

p q 1.
Solution The given PDE is of the form f ( p, q ) 0. Therefore, let us assume the solution
in the form
z ax by c
where
a b 1 or b (1 a )2
Hence, the complete integral is found to be
z ax (1 a )2 y c.
EXAMPLE 0.20 Find the complete integral of the PDE

pq 1.
Solution Since the given PDE is of the form f ( p, q ) 0, we assume the solution in
the form z ax by c, where ab 1 or b 1/ a. Hence, the complete integral is
1
y c.
z ax
a
Type II Equations Not Involving the Independent Variables.
f ( z , p, q ) 0 (0.103)
As a trial solution, let us assume that z is a function of u x ay , where a is an arbitrary

constant. Then,
z f (u ) f ( x ay ) (0.104)
wz dz w u dz
p ¹
wx du w x du
wz dz w u dz
q ¹ a
wy du w y du
Substituting these values of p and q in the given PDE, we get
§ dz dz ·
f ¨ z, , a ¸ 0 (0.105)
© du du ¹
which is an ordinary differential equation of first order.
Solving Eq. (0.105) for dz /du , we obtain
dz
φ ( z , a ) (say)
du
or
dz
du.
φ ( z, a)
dz
³ φ ( z, a) uc
That is,
F ( z , a) uc x ay c (0.106)

p (1 q ) qz
Solution Let us assume the solution in the form

z f (u ) x ay
Then,
dz dz
p , q a
du du
Substituting these values in the given PDE, we get
dz § dz · dz
¨©1 a ¸¹ az
du du du.
That is,
dz dz
a az 1 or a du
du az 1
ln (az 1) u c x ay c
which is the required complete integral.

p2 z 2 q2 1.
Solution Let us assume that z f (u ) x ay is a solution of the given PDE. Then,

dz dz
, q a
p
du du
Substituting these values of p and q in the given PDE, we obtain
2 2
§ dz · 2 2 § dz ·
¨© ¸¹ z a ¨© ¸¹ 1
du du
That is,
2
§ dz · 2 2 dz 1
¨© ¸¹ ( z a ) 1 or
du du z a2
2
or
z 2 a 2 dz du
z z 2 a2 a2 ª z z 2 a2 º
ln « » x ay b
2 2 «¬ a »¼
which is the required complete integral of the given PDE.
Type III Separable Equations.

An equation in which z is absent and the terms containing x and p can be separated from those
containing y and q is called a separable equation.
f ( x, p ) F ( y, q) (0.107)
As a trial solution, let us assume that
f ( x, p ) F ( y, q) a (say) (0.108)
Now, solving them for p and q, we obtain

p φ ( x), q ψ ( y)
Since
wz wz
dz dx dy p dx q dy
wx wy
or
dz φ ( x) dx ψ ( y ) dy
On integration, we get the complete integral in the form
z ³ φ ( x) dx ³ ψ ( y) dy b (0.109)

p 2 y (1 x 2 ) qx 2
Solution The given PDE is of separable type and can be rewritten as

p 2 (1 x 2 ) q
2
a (say), an arbitrary constant.
x y
Then,
ax
p , q ay
1 x2
dz p dx q dy,
we get
ax
dz dx ay dy
1 x2
a 2
z a 1 x2 y b
2

p2 q2 x y
Solution The given PDE is of separable type and can be rewritten as

p2 x y q2 a (say)
Then,
p x a, q y a.
Now, substituting these values of p and q in
dz p dx q dy
we find
dz x a dx y a dy
On integration, the complete integral is found to be
2 2
z ( x a )3/2 ( y a )3/2 b.
3 3
Type IV Clairaut’s Form

A first order PDE is said to be of Clairaut’s form if it can be written as
z px qy f ( p, q ) (0.110)
The corresponding Charpit’s equations are
dx dy dz
x fp y fq px qy pf p qf q
dp dq
(0.111)
p p qq
The integration of the last two equations of (0.111) gives us

p a, q b
Substituting these values of p and q in the given PDE, we get the required complete integral
in the form
z ax by f (a, b) (0.112)
EXAMPLE 0.25 Find the complete integral of the equation
z px qy 1 p 2 q 2
Solution The given PDE is in the Clairaut’s form. Hence, its complete integral is
z ax by 1 a 2 b 2 .

( p q ) ( z xp yq ) 1
Solution The given PDE can be rewritten as

1
z xp yq
pq
which is in the Clairaut’s form,
z px qy f ( p, q )
Hence, the complete integral of the given PDE is
1
z ax by .
ab
EXAMPLE 0.27 Solve the following Cauchy IVP:
PDE: ut + cux = 0, x Î R, t > 0
IC: u(x, 0) = f(x), x Î R.
dt dx du
Solution The characteristic equations for the given problem is given by .
1 c 0
On integration, we get: u = constant and x – ct = x (constant). This linear problem
has a unique solution, given by u(x, t) = f(x – ct).
This is a right travelling wave with speed c, of course with no change in shape. The
characteristic line x = x + ct, gives rise to a system of parallel, straight lines in the (x, t)-plane
as shown in Fig. 0.6.
Fig. 0.6 The family of characteristic lines: x – ct = x.
We observe that one of these lines that passes through the point (x, t) intersects the
x-axis at (x, 0) moving with speed c, having a slope –1/c.
EXAMPLE 0.28 In classical mechanics, the Hamilton–Jacobi equation for the problem of
one-dimensional, Harmonic oscillator is given by the differential equation as (see–Sankara
Rao, 2005).
2
1 È S Ø 1 S
É Ù Kq 2 0,
2 m Ê q Ú 2 t
S
where S = S(p, q, t), p = and K is a constant. Using Charpits method, find S.
q
Solution Following the notation of Eq. (0.94) we rewrite
2
1 È S Ø 1 S
f (t , q, S, St , Sq ) É Ù Kq 2 (1)
Ê
2 m q Ú 2 t
which gives us
Sq
ft 0, fq Kq, f S .
0, f St 1, f Sq
m
Then, the Charpits auxiliary equations (0.99) assumes the following form:
dt dq dS dSt dSq
(2)
1 Sq /m St Sq2 /m 0 Kq
dq dSq
Considering the second and last members, we have .
Sq /m Kq
Sq2 1 2
Kq a (constant of integration).
2m 2
Equation (1) then becomes
St = –a,
È 2a Ø
and Sq2 Km É q2 Ù .
ÊK Ú
Substituting St and Sq into
dS = St dt + Sq dq
and integrating, we arrive at
1/2
È 2a Ø
S at Km É
ÊK Ô
q2 Ù
Ú
dq C
or S Ô
at Km (D 2 q 2 )1/2 dq C
2a
where a2 = and C is another constant of integration.
K
EXERCISES
1. Eliminate the arbitrary function in the following and hence obtain the corresponding
PDE
z x y f ( xy ).
2. Form the PDE from the following by eliminating the constants
z ( x 2 a ) ( y 2 b).
3. Find the integral surface (general solution) of the differential equation
wz wz
x2 y2 ( x y ) z.
wx wy
4. Find the general integrals of the following linear PDEs:
y2 z
(i) p xzq y2
x
(ii) ( y 1) p ( x 1) q z.
5. Find the integral surface of the linear PDE
xp yq z
which contains the circle x 2 y 2 1, z 1.

6. Find the equation of the integral surface of the PDE
2 y ( z 3) p (2 x z ) q y (2 x 3)
which contains the circle x 2 y 2 2 x, z 0.

7. Find the general integral of the PDE
( x y) p ( y x z) q z
which contains the circle x 2 y 2 1, z 1.

8. Find the solution of the equation
1 2
z ( p q 2 ) ( p x) (q y )
2
which passes through the x-axis.
9. Find the characteristics of the equation
pq xy
and determine the integral surface which passes through the curve z x, y 0.
10. Determine the characteristics of the equation
z p2 q2
and find the integral surface which passes through the parabola 4 z x 2 0, y 0.
11. Show that the PDEs

xp yq and z ( xp yq ) 2 xy
are compatible and hence find its solution.
12. Show that the equations
p2 q2 1 and ( p2 q2 ) x pz
are compatible and hence find its solution.
13. Find the complete integral of the equation
( p2 q2 ) x pz
where p w z /w x, q w z /w y.
14. Find the complete integrals of the equations
(i) px5 4q3 x 2 6 x 2 z 2 0
2
(ii) 2 ( z xp yq ) yp .
p q pq.
16. Find the complete integrals of the following equations:
(i) z pq p q
(ii) p 2 q 2 x 2 y 2 x 2 q 2 ( x 2 y 2 ).
17. Find the complete integral of the PDE
z px qy sin ( pq )
18. Find the complete integrals of the following PDEs:
(i) xp3 q 2 yp 2 q3 ( p3 q3 ) zp 2 q 2 0
(ii) p qz p 2 ( xq p 2 ) q 2 ( yp q 2 ).
19. Find the surface which intersects the surfaces of the system
z ( x y) c (3 z 1)
orthogonally and passes through the circle
x2 y2 1, z 1.
( p2 q2 ) x pz
wz wz
where p , q . (GATE-Maths, 1996)
wx wy
21. Find the integral surface of the linear PDE

wz wz
( x y) ( x y z) z
wx wy
which contains z 1 and x 2 y 2 1. (GATE-Maths, 1999)
Choose the correct answer in the following questions:

22. Using the transformation u W / y in the PDE xu x u yu y , the transformed equation
has a solution of the form W =
(A) f ( x /y ) (B) f ( x y)
(C) f ( x y ) (D) f ( xy ). (GATE-Maths, 1997)
23. The complete integral of the partial differential equation
xp3 q 2 yp 2 q3 ( p3 q3 ) zp 2 q 2 0
is z =
2 2
(A) ax by (ab 2 ba 2 ) (B) ax by (ab ba )
(C) ax by (ba 2 ab 2 ) (D) ax by (ab 2 ba 2 ).

(GATE-Maths, 1997)
24. The partial differential equation of the family of surfaces z ( x y ) A ( xy ) is
(A) xp yq 0 (B) xp yq x y
(C) xp yq x y (D) xp yq 0.
(GATE-Maths, 1998)
25. The complete integral of the PDE z px qy sin ( pq ) is
(A) z ax by sin (ab) (B) z ax by sin (ab)
(C) z ax y sin (b) (D) z x by sin (a ).
(GATE-Maths, 2003)
CHAPTER 1
Fundamental Concepts
1.1 INTRODUCTION
Many practical problems in science and engineering, when formulated mathematically, give
rise to partial differential equations (often referred to as PDE). In order to understand the
physical behaviour of the mathematical model, it is necessary to have some knowledge about
the mathematical character, properties, and the solution of the governing PDE. An equation
which involves several independent variables (usually denoted by x, y, z, t, …), a dependent
function u of these variables, and the partial derivatives of the dependent function u with
respect to the independent variables such as
F ( x, y, z , t , } , u x , u y , u z , ut , } , u xx , u yy , } , u xy ,}) 0 (1.1)
is called a partial differential equation. A few well-known examples are:
(i) ut k (u xx u yy u zz ) [linear three-dimensional heat equation]
(ii) u xx u yy u zz 0 [Laplace equation in three dimensions]
(iii) utt c 2 (u xx u yy u zz ) [linear three-dimensional wave equation]
(iv) ut uu x μ u xx [nonlinear one-dimensional Burger equation].

In all these examples, u is the dependent function and the subscripts denote partial differentiation
with respect to these variables.
Definition 1.1 The order of the partial differential equation is the order of the highest derivative
occurring in the equation. Thus the above examples are partial differential equations of
second order, whereas
ut uu xxx sin x
is an example for third order partial differential equation.
52
FUNDAMENTAL CONCEPTS 53
1.2 CLASSIFICATION OF SECOND ORDER PDE

The most general linear second order PDE, with one dependent function u on a domain Ω of
points X ( x1 , x2 , } , xn ), n ! 1, is
n n
¦ aij u xi x j ¦ biuxi F (u) G (1.2)
i, j 1 i 1
The classification of a PDE depends only on the highest order derivatives present.
The classification of PDE is motivated by the classification of the quadratic equation of
the form
Ax 2 Bxy Cy 2 Dx Ey F 0 (1.3)
which is elliptic, parabolic, or hyperbolic according as the discriminant B 2 4 AC is negative,

zero or positive. Thus, we have the following second order linear PDE in two variables
x and y:
Au xx Bu xy Cu yy Du x Eu y Fu G (1.4)
where the coefficients A, B, C, … may be functions of x and y, however, for the sake of
simplicity we assume them to be constants. Equation (1.4) is elliptic, parabolic or hyperbolic
at a point (x0, y0) according as the discriminant
B 2 ( x0 , y0 ) 4 A ( x0 , y0 ) C ( x0 , y0 )
is negative, zero or positive. If this is true at all points in a domain Ω, then Eq. (1.4) is said
to be elliptic, parabolic or hyperbolic in that domain. If the number of independent variables
is two or three, a transformation can always be found to reduce the given PDE to a canonical
form (also called normal form). In general, when the number of independent variables is
greater than 3, it is not always possible to find such a transformation except in certain special
cases. The idea of reducing the given PDE to a canonical form is that the transformed
equation assumes a simple form so that the subsequent analysis of solving the equation is
made easy.
1.3 CANONICAL FORMS

Consider the most general transformation of the independent variables x and y of Eq. (1.4)
to new variables ξ , η , where
ξ ξ ( x, y ), η η ( x, y ) (1.5)
such that the functions ξ and η are continuously differentiable and the Jacobian
w (ξ , η ) ξx ξy
J (ξ xη y ξ yη x ) z 0 (1.6)
w ( x, y ) ηx ηy
in the domain Ω where Eq. (1.4) holds. Using the chain rule of partial differentiation, the
partial derivatives become
ux uξ ξ x uηη x
uy uξ ξ y uηη y
u xx uξξ ξ x2 2uξηξ xη x uηηη x2 uξ ξ xx uηη xx
u xy uξξ ξ xξ y uξη (ξ xη y ξ yη x ) uηηη xη y uξ ξ xy uηηxy
u yy uξξ ξ y2 2uξηξ yη y uηηη 2y uξ ξ yy uηη yy (1.7)
Substituting these expressions into the original differential equation (1.4), we get
Auξξ Buξη Cuηη Duξ Euη Fu G (1.8)
where
A Aξ x2 Bξ xξ y Cξ y2
B 2 Aξ xη x B (ξ xη y ξ yη x ) 2Cξ yη y
C Aη x2 Bη xη y Cη y2
D Aξ xx Bξ xy Cξ yy Dξ x Eξ y
E Aη xx Bη xy Cη yy Dη x Eη y
F F, G G (1.9)
It may be noted that the transformed equation (1.8) has the same form as that of the original
equation (1.4) under the general transformation (1.5).
Since the classification of Eq. (1.4) depends on the coefficients A, B and C, we can also
rewrite the equation in the form
Au xx Bu xy Cu yy H ( x, y , u , u x , u y ) (1.10)
It can be shown easily that under the transformation (1.5), Eq. (1.10) takes one of the
following three canonical forms:
(i) uξξ uηη φ (ξ , η , u, uξ , uη ) (1.11a)
or
uξη φ , (ξ , η , u , uξ , uη ) in the hyperbolic case
(ii) uξξ uηη φ (ξ , η , u, uξ , uη ) in the elliptic case (1.11b)
(iii) uξξ φ (ξ , η , u , uξ , uη ) (1.11c)

or
uηη φ (ξ , η , u, uξ , uη ) in the parabolic case
We shall discuss in detail each of these cases separately.
Using Eq. (1.9) it can also be verified that
B 2 4 AC (ξ xη y ξ yηx )2 ( B 2 4 AC )
and therefore we conclude that the transformation of the independent variables does not
modify the type of PDE.
1.3.1 Canonical Form for Hyperbolic Equation
Since the discriminant B 2 4 AC ! 0 for hyperbolic case, we set A 0 and C 0 in Eq. (1.9),
which will give us the coordinates ξ and η that reduce the given PDE to a canonical form
in which the coefficients of uξξ , uηη are zero. Thus we have
A Aξ x2 Bξ xξ y Cξ y2 0
C Aη x2 Bη xη y Cη 2y 0
which, on rewriting, become
2
§ξ · §ξ ·
A¨ x ¸ B ¨ x ¸ C 0
©ξ y ¹ ©ξ y ¹
2
§η · §η ·
A¨ x ¸ B ¨ x ¸ C 0
©η y ¹ ©ηy ¹
Solving these equations for (ξ x /ξ y ) and (η x /η y ), we get
ξx B B 2 4 AC
ξy 2A
ηx B B 2 4 AC (1.12)
ηy 2A
The condition B 2 ! 4 AC implies that the slopes of the curves ξ ( x, y ) C1 , η ( x, y ) C2 are

real. Thus, if B 2 ! 4 AC , then at any point (x, y), there exists two real directions given by the
two roots (1.12) along which the PDE (1.4) reduces to the canonical form. These are called
characteristic equations. Though there are two solutions for each quadratic, we have considered
only one solution for each. Otherwise we will end up with the same two coordinates.
Along the curve ξ ( x, y ) c1 , we have

dξ ξ x dx ξ y dy 0
Hence,
dy §ξ ·
¨ x ¸ (1.13)
dx ©ξy ¹
Similarly, along the curve η ( x, y ) c2 , we have
dy §η ·
¨ x ¸ (1.14)
dx ©η y ¹
Integrating Eqs. (1.13) and (1.14), we obtain the equations of family of characteristics ξ ( x, y ) c1
and η ( x, y ) c2 , which are called the characteristics of the PDE (1.4). Now to obtain the
canonical form for the given PDE, we substitute the expressions of ξ and η into Eq. (1.8)
which reduces to Eq. (1.11a).
To make the ideas clearer, let us consider the following example:
3u xx 10u xy 3u yy 0
Comparing with the standard PDE (1.4), we have A 3, B 10, C 3, B 2 4 AC 64 ! 0. Hence

the given equation is a hyperbolic PDE. The corresponding characteristics are:
dy §ξ · § B B 2 4 AC · 1
¨ x ¸ ¨ ¸
dx ©ξ y ¹ ¨© 2A ¸¹ 3
dy §η · § B B 2 4 AC ·
¨ x ¸ ¨ ¸ 3
dx ©ηy ¹ ¨© 2A ¸¹
To find ξ and η , we first solve for y by integrating the above equations. Thus, we get
1
y 3 x c1 , y x c2
3
which give the constants as
c1 y 3 x, c2 y x /3
Therefore,
1
ξ y 3x c1 , η y x c2
3
These are the characteristic lines for the given hyperbolic equation. In this example, the
characteristics are found to be straight lines in the (x, y)-plane along which the initial data,
impulses will propagate.
To find the canonical equation, we substitute the expressions for ξ and η into Eq. (1.9) to get
A AY x2 BY xY y CY 2y 3( 3)2 10( 3) (1) 3 0
B 2 AY xIx B (Y xI y Y y Ix ) 2CY y I y
¦ 1µ © ¦ 1µ ¸
2(3) ( 3) § ¶ 10 ª( 3)(1) 1§ ¶ ¹ 2(3) (1) (1)
¨ 3· « ¨ 3· º
¦ 10 µ 100 64
6 10 § ¶ 6 12
¨ 3· 3 3
C 0, D 0, E 0, F 0
Hence, the required canonical form is
64
uξη 0 or uξη 0
3
u (ξ , η ) f (ξ ) g (η )
where f and g are arbitrary. Going back to the original variables, the general solution is
u ( x, y ) f ( y 3x) g ( y x /3)
1.3.2 Canonical Form for Parabolic Equation
For the parabolic equation, the discriminant B 2 4 AC 0, which can be true

if B 0 and A or C is equal to zero. Suppose we set first A 0 in Eq. (1.9). Then we
obtain
or
2
§ξ · §ξ ·
A¨ x ¸ B ¨ x ¸ C 0
©ξy ¹ ©ξ y ¹
which gives
ξx B r B 2 4 AC
ξy 2A
Using the condition for parabolic case, we get

ξx B
=− (1.15)
ξy 2A
Hence, to find the function ξ = ξ ( x, y ) which satisfies Eq. (1.15), we set
dy ξ B
=− x =
dx ξ y 2A
and get the implicit solution
ξ ( x, y ) = C1
In fact, one can verify that A = 0 implies B = 0 as follows:
B = 2 Aξ x η x + B (ξ x η y + ξ y ηx ) + 2Cξ y η y
Since B 2 − 4 AC = 0, the above relation reduces to
B = 2 Aξxηx + 2 AC (ξ xη y + ξ y ηx ) + 2Cξ y η y
= 2 ( Aξ x + C ξ y ) ( Aηx + C η y )
However,
ξx B 2 AC C
=− =− =−
ξy 2A 2A A
Hence,
B = 2 ( Aξ x − Aξ x ) ( Aη x + Cη y ) = 0
We therefore choose ξ in such a way that both A and B are zero. Then η can be chosen in
any way we like as long as it is not parallel to the ξ - coordinate. In other words, we choose η such
that the Jacobian of the transformation is not zero. Thus we can write the canonical equation
for parabolic case by simply substituting ξ and η into Eq. (1.8) which reduces to either of
the forms (1.11c).
To illustrate the procedure, we consider the following example:
x 2u xx − 2 xyu xy + y 2u yy = e x
The discriminant B 2 − 4 AC = 4 x 2 y 2 − 4 x 2 y 2 = 0, and hence the given PDE is parabolic

everywhere. The characteristic equation is
dy ξ B 2 xy y
=− x = =− 2 =−
dx ξy 2A 2x x
On integration, we have
xy c
and hence ξ xy will satisfy the characteristic equation and we can choose η y. To find
the canonical equation, we substitute the expressions for ξ and η into Eq. (1.9) to get
A Ay 2 Bxy cx 2 x2 y 2 2 x2 y 2 y 2 x2 0
2
B 0, C y , D 2 xy
E 0, F 0, G ex
Hence, the transformed equation is
y 2uηη 2 xyuξ ex
or
η 2uηη 2ξ uξ eξ /η
The canonical form is, therefore,
2ξ 1
uηη u
2 ξ
eξ /η
η η2
1.3.3 Canonical Form for Elliptic Equation
Since the discriminant B 2 4 AC 0, for elliptic case, the characteristic equations
dy B B 2 4 AC
dx 2A
dy B B 2 4 AC
dx 2A
give us complex conjugate coordinates, say ξ and η. Now, we make another transformation
from (ξ , η ) to (α , β ) so that
ξ η ξ η
α , β
2 2i
which give us the required canonical equation in the form (1.11b).
To illustrate the procedure, we consider the following example:
u xx x 2u yy 0
The discriminant B 2 4 AC 4 x 2 0. Hence, the given PDE is elliptic. The characteristic

equations are
dy B B 2 4 AC 4 x 2
ix
dx 2A 2
dy B B 2 4 AC
ix
dx 2A
Integration of these equations yields
x2 x2
iy c1 , iy c2
2 2
Hence, we may assume that
1 2 1 2
ξx iy , η x iy
2 2
Now, introducing the second transformation
ξ η ξ η
α , β
2 2i
we obtain
x2
α
, β y
2
The canonical form can now be obtained by computing
A Aα x2 βα xα y cα 2y x2
B 2 Aα x β x B (α x β y α y β x ) 2c (α y β y ) 0
C Aβ x2 B β x β y c β y2 x2
D Aα xx Bα xy cα yy Dα x Eα y 1
E Aβ xx B β xy c β yy D β x E β y 0
F 0, G 0
Thus the required canonical equation is

x 2uαα x 2uββ uα 0
or
uα
uαα uββ
2α
EXAMPLE 1.1 Classify and reduce the relation

y2 x2
y 2u xx − 2 xyu xy + x 2 u yy = ux + u y
x y
to a canonical form and solve it.
Solution The discriminant of the given PDE is
B 2 − 4 AC = 4 x 2 y 2 − 4 x 2 y 2 = 0
Hence the given equation is of a parabolic type. The characteristic equation is
dy ξ B −2 xy x
=− x = = =−
dx ξ y 2 A 2 y2 y
Integration gives x 2 + y 2 = c1. Therefore, ξ = x 2 + y 2 satisfies the characteristic equation. The

η - coordinate can be chosen arbitrarily so that it is not parallel to ξ , i.e. the Jacobian of the
transformation is not zero. Thus we choose
ξ = x2 + y 2 , η = y2
To find the canonical equation, we compute
A = Aξ x2 + Bξxξ y + Cξ 2y = 4 x 2 y 2 − 8 x 2 y 2 + 4 x 2 y 2 = 0
B = 0, C = 4 x2 y 2 , D=E =F =G=0
Hence, the required canonical equation is
4 x 2 y 2uηη = 0 or uηη = 0
To solve this equation, we integrate it twice with respect to η to get
uη = f (ξ ), u = f (ξ )η + g (ξ )
where f (ξ ) and g (ξ ) are arbitrary functions of ξ . Now, going back to the original independent
variables, the required solution is
u = y 2 f ( x2 + y 2 ) + g ( x2 + y 2 )
EXAMPLE 1.2 Reduce the following equation to a canonical form:

(1 + x 2 ) u xx + (1 + y 2 ) u yy + xu x + yu y = 0
B 2 − 4 AC = − 4 (1 + x 2 ) (1 + y 2 ) < 0
Hence the given PDE is an elliptic type. The characteristic equations are
dy B − B 2 − 4 AC − 4(1 + x 2 ) (1 + y 2 ) 1 + y2
= =− = −i
dx 2A 2(1 + x 2 ) 1 + x2
dy B + B 2 − 4 AC 1 + y2
= =i
dx 2A 1 + x2
ξ = ln ( x + x 2 + 1) − i ln ( y + y 2 + 1) = c1
η = ln ( x + x 2 + 1) + i ln ( y + y 2 + 1) = c2
Introducing the second transformation
ξ +η η −ξ
α= , β=
2 2i
we obtain
α = ln ( x + x 2 + 1)
β = ln ( y + y 2 + 1)
Then the canonical form can be obtained by computing
A = Aα x2 + Bα xα y + Cα 2y = 1, B = 0, C = 1, D=E =F =G=0
Thus the canonical equation for the given PDE is
uαα + u ββ = 0
EXAMPLE 1.3 Reduce the following equation to a canonical form and hence solve it:
u xx − 2 sin xu xy − cos2 xu yy − cos xu y = 0
Solution Comparing with the general second order PDE (1.4), we have
A = 1, B = −2 sin x, C = − cos 2 x,
D = 0, E = − cos x, F = 0, G=0
The discriminate B 2 − 4 AC = 4 (sin 2 x + cos 2 x) = 4 > 0. Hence the given PDE is hyperbolic.
The relevant characteristic equations are
dy B − B 2 − 4 AC
= = − sin x − 1
dx 2A
dy B + B 2 − 4 AC
= = 1 − sin x
dx 2A
y cos x x c1 , y cos x x c2
Thus, we choose the characteristic lines as
ξ x y cos x c1 , η x y cos x c2
In order to find the canonical equation, we compute
B 2 Aξ xη x B (ξ xη y ξ yη x ) 2Cξ yη y
2 (sin x 1) (sin x 1) 4 sin 2 x 2 cos 2 x 4
C 0, D 0, E 0, F 0, G 0
Thus, the required canonical equation is
uξη 0
Integrating with respect to ξ , we obtain
uη f (η )
where f is arbitrary. Integrating once again with respect to η , we have
u ³ f (η ) dη g (ξ )
or
u ψ (η ) g (ξ )
where g (ξ ) is another arbitrary function. Returning to the old variables x, y, the solution of
the given PDE is
u ( x, y ) ψ ( y x cos x) g ( y x cos x)
EXAMPLE 1.4 Reduce the Tricomi equation

u xx xu yy 0, xz0
for all x, y to canonical form.
Solution The discriminant B 2 4 AC 4 x. Hence the given PDE is of mixed type:
hyperbolic for x 0 and elliptic for x ! 0.
Case I In the half-plane x 0, the characteristic equations are
dy ξx B B 2 4 AC 2 x
x
dx ξy 2A 2
dy ηx B B 2 4 AC
x
dx ηy 2A
Integration yields
2
y ( x)3/2 c1
3
2
y ( x)3/2 c2
3
Therefore, the new coordinates are
3
ξ ( x, y ) y ( x )3 c1
2
3
η ( x, y ) y ( x )3 c2
2
which are cubic parabolas.
In order to find the canonical equation, we compute
9 9
A Aξ x2 Bξ xξ y Cξ y2 x0 x 0
4 4
3
B 9 x, C 0, D ( x)1/2 E, F G 0
4
Thus, the required canonical equation is
3 3
9 xuξη ( x ) 1/2 uξ ( x )1/2 uη 0
4 4
or
1
uξη (uξ uη )
6 (ξ η )
Case II In the half-plane x > 0, the characteristic equations are given by

dy dy
i x, i x
dx dx
On integration, we have
3 3
ξ ( x, y )
y i ( x )3 , η ( x, y ) y i ( x )3
2 2
Introducing the second transformation
ξ η ξ η
α , β
2 2i
we obtain
3
α y, β ( x )3
2
The corresponding normal or canonical form is

1
uαα u ββ uβ 0
3β
EXAMPLE 1.5 Find the characteristics of the equation

u xx 2u xy sin 2 ( x) u yy u y 0
when it is of hyperbolic type.
Solution The discriminant B2 – 4AC = 4 – 4sin2x = 4cos2x. Hence for all x ¹ (2n – 1)p/2,
the given PDE is of hyperbolic type. The characteristic equations are
dy B # B 2 4 AC
1 # cos x
dx 2A
y x sin x c1 , y x sin x c2
Thus, the characteristic equations are
ξ y x sin x, η y x sin x
EXAMPLE 1.6 Reduce the following equation to a canonical form and hence solve it:
yu xx ( x y ) u xy xu yy 0
Solution The discriminant

B 2 4 AC ( x y )2 4 xy ( x y )2 ! 0
Hence the given PDE is hyperbolic everywhere except along the line y = x; whereas on the
line y = x, it is parabolic. When y z x, the characteristic equations are
dy B # B 2 4 AC ( x y) # ( x y)
dx 2A 2y
Therefore,
dy dy x
1,
dx dx y
y x c1 , y2 x 2 c2
Hence, the characteristic equations are
ξ y x, η y 2 x2
These are straight lines and rectangular hyperbolas. The canonical form can be obtained by
computing
A Aξ x2 Bξ xξ y Cξ y2 yx yx 0, B 2 ( x y )2 ,
C 0, D 0, E 2( x y ), F G 0
Thus, the canonical equation for the given PDE is
2( x y )2 uξη 2( x y ) uη 0
or
2ξ 2uξη 2(ξ ) uη 0
or
w § wu ·
ξ uξη uη ξ 0
w ξ ¨© w η ¸¹
Integration yields
wu
ξ f (η )
wη
Again integrating with respect to η , we obtain
1
u
ξ ³ f (η ) dη g (ξ )
Hence,
1
u
yx ³ f ( y 2 x 2 ) d ( y 2 x 2 ) g ( y x)
is the general solution.

EXAMPLE 1.7 Classify and transform the following equation to a canonical form:
sin 2 ( x) u xx sin (2 x) u xy cos 2 ( x) u yy x
B 2 4 AC sin 2 2 x 4 sin 2 x cos 2 x 0

Hence, the given equation is of parabolic type. The characteristic equation is
dy B
cot x
dx 2A
Integration gives
y ln sin x c1
Hence, the characteristic equations are:

ξ y ln sin x, η y
η is chosen in such a way that the Jacobian of the transformation is nonzero. Now the
canonical form can be obtained by computing
A 0, B 0, C cos 2 x, D 1,
E 0, F 0, G x
Hence, the canonical equation is
cos2 ( x) uηη uξ x
or
[1 e2 (η ξ ) ] uηη sin 1 (eη ξ ) uξ
EXAMPLE 1.8 Show that the equation

2N 1
u xx utt ux
a2 x
where N and a are constants, is hyperbolic and obtain its canonical form.
Solution Comparing with the general PDE (1.4) and replacing y by t, we have A 1, B 0,
C 1/a 2, D 2 N / x, and E F G 0. The discriminant B 2 4 AC 4/ a 2 ! 0. Hence, the

given PDE is hyperbolic. The characteristic equations are
dt B # B 2 4 AC 4/ a 2 1
# #
dx 2A 2 a
Therefore,
dt 1 dt 1
,
dx a dx a
x x
t c1 , t c2
a a
Hence, the characteristic equations are
ξ x at , η x at
The canonical form can be obtained by computing
A Aξ x2 Bξ xξt Cξ t2 0,
B 2 Aξ xη x B (ξ xηt ξtη x ) 2Cξtηt 4,
2N 2N
C 0, D Dξ x Eξt , E Dη x Eηt
x x
Thus, the canonical equation for the given PDE is

2N
4uξη (uξ uη ) 0
x
Expressing x in terms of ξ and η , the required canonical equation is
N
uξη (u u ) 0
ξ η ξ η
EXAMPLE 1.9 Transform the following differential equation to a canonical form:

u xx 2u xy 4u yy 2u x 3u y 0
Solution The discriminant B 2 4 AC 12 0. Hence, the given PDE is elliptic. The
characteristic equations are
dy B B 2 4 AC
1 i 3
dx 2A
dy B B 2 4 AC
1 i 3
dx 2A
Integration of these equations yields
y (1 i 3) x c1 , y (1 i 3) x c2
Hence, we may take the characteristic equations in the form
ξ y (1 i 3) x, η y (1 i 3) x
In order to avoid calculations with complex variables, we introduce the second transformation
ξ η ξ η
α , β
2 2i
Therefore,
α y x, β 3x
The canonical form can now be obtained by computing
A Aα x2 Bα xα y Cα 2y 3
B 2 Aα x β x B (α x β y α y β x ) 2Cα y β y 0
C Aβ x2 B β x β y C β y2 3
D Aα xx Bα xy Cα yy Dα x Eα y 1
E Aβ xx B β xy C β yy D β x E β y 2 3
F 0, G 0
Thus the required canonical form is

3uαα 3u ββ uα 2 3u β 0
or
1
uαα u ββ (uα 2 3u β )
3
1.4 ADJOINT OPERATORS

Let
Lu φ (1.16)
where L is a differential operator given by
dn d n 1
L a0 ( x) a1 ( x)
an ( x)
dx n dx n 1
One way of introducing the adjoint differential operator L* associated with L is to form the
product vLu and integrate it over the interval of interest. Let
B B
³A vLu dx [ ] BA ³A uL * v dx (1.17)
which is obtained after repeated integration by parts. Here, L* is the operator adjoint to L,
where the functions u and v are completely arbitrary except that Lu and L*v should exist.
EXAMPLE 1.10 Let Lu a ( x) (d 2u /dx 2 ) b( x) (du /dx) c( x)u; construct its adjoint L*.
Solution Consider the equation
B B ª d 2u du º
³A vLu dx ³A v « a ( x) 2 b( x)
¬« dx dx
c( x)u » dx
¼»
B d 2u B du B
³A (av)
dx 2
dx ³A (bv)
dx
dx ³A (cv) u dx
However,
B
d 2u B
d
Ô A
(av)
dx 2
dx Ô A
(av)
dx
(u )dx
B
[u va ] BA Ô A
(av) u dx
B
[u av] BA [u (av) ] BA Ô A
u (av) dx
B B
du
Ô A
(bv)
dx
dx [u (bv)] BA Ô A
u (bv) dx
B B
Ô A
(cv) u dx Ô A
u (cv) dx
Therefore,
B B
³A vLu dx [u c (av) u (av) c u (bv)] BA ³A u [(av)cc (bv)c (cv)] dx
Comparing this equation with Eq. (1.17), we get
L*v (av) cc (bv) c (cv) av cc (2a c b) v c (a cc b c c) v

Therefore,
d2 d
L* a 2
(2a c b) (a cc b c c)
dx dx
Consider the partial differential equation
L(u ) Au xx 2 Bu xy Cu yy Du x Eu y Fu φ (1.18)
which is valid in a region S of the xy-plane, where A, B, C, …, φ are functions of x and y.

In addition, linear boundary conditions of the general form
α u β ux f
are prescribed over the boundary curve wS of the region S. Let
³³ vLu dσ [ ] ³³ uL * v dσ
S S
where the integrated part [ ] is a line integral evaluated over w S , the boundary of S, then L*
is called the adjoint operator. In general, a second order linear partial differential operator L
is denoted by
n n
w 2u wu
L(u ) ¦ Aij
w xiw x j i ¦ Bi w xi Cu (1.19)
i, j 1 1
Its adjoint operator is defined by

n n
w2 w
L * (v ) ¦ w xiw x j
( Aij v) ¦ w xi (Bi v) Cv (1.20)
i, j 1 i 1
Here it is assumed that Aij C (2) and Bi C (1). For any pair of functions u , v C (2) , it can
be shown that
w Ë È wu wv Ø È w Aij Ø Û
n n n
vL(u ) uL * (v) Ç Ì
w xi Ì Ç Aij É v
Ê w xj
u Ù
w xj Ú
uv É Bi
Ê
Ç wx ÙÜ
j ÚÜ
(1.21)
i 1 Í j 1 j 1 Ý
This is known as Lagrange’s identity.
EXAMPLE 1.11 Construct an adjoint to the Laplace operator given by

L(u ) u xx u yy (1.22)
Solution Comparing Eq. (1.22) with the general linear PDE (1.19), we have
A11 1, A22 1. From Eq. (1.20), the adjoint of (1.22) is given by
w2 w2
L * (v ) (v ) (v ) vxx v yy
w x2 w y2
Therefore,
L * (u ) u xx u yy
Hence, the Laplace operator is a self-adjoint operator.
EXAMPLE 1.12 Find the adjoint of the differential operator

L (u ) u xx ut (1.23)
Solution Comparing Eq. (1.23) with the general second order PDE (1.19), we have
A11 1, B1 1. From Eq. (1.20), the adjoint of (1.23) is given by
w2 w
L * (v ) (v ) ( v ) vxx vt
wx 2 wt
Therefore,
L * (u ) u xx ut
It may be noted that the diffusion operator is not a self-adjoint operator.
1.5 RIEMANNS METHOD

In Section 1.2, we have noted with interest that a linear second order PDE
L(u ) G ( x, y )
is classified as hyperbolic if B 2 ! 4 AC , and it has two families of real characteristic curves

in the xy-plane whose equations are
ξ f1 ( x, y ) c1 , η f 2 ( x, y ) c2
Here, (ξ , η ) are the natural coordinates for the hyperbolic system. In the xy-plane, the curves
ξ ( x, y ) c1 and η ( x, y ) c2 are the characteristics of the given PDE as shown in Fig. 1.1(a),
while in the ξη -plane, the curves ξ c1 and η c2 are families of straight lines parallel to
the axes as shown in Fig. 1.1(b).
y
I
Characteristics Characteristics
Y = const.
x
O
I = const. Y
O
(a) (b)
Fig. 1.1 Families of characteristic lines.
A linear second order partial differential equation in two variables, once classified as a
hyperbolic equation, can always be reduced to the canonical form
w 2u
F ( x, y , u , u x , u y )
wxw y
In particular, consider an equation which is already reduced to its canonical form in the
variables x, y:
w 2u wu wu
L(u ) a b cu F ( x, y ) (1.24)
wxw y wx wy
where L is a linear differential operator and a, b, c, F are functions of x and y only and are
differentiable in some domain IR .
Let v ( x, y ) be an arbitrary function having continuous second order partial derivatives.
Let us consider the adjoint operator L* of L defined by
w 2v w w
L * (v ) (av) (bv) cv (1.25)
wxw y wx wy
Now we introduce
wv wu
M auv u , N buv v (1.26)
wy wx
then
Mx Ny u x (av) u (av) x u x v y uvxy u y (bv) u (bv) y v y u x vu xy
Adding and subtracting cuv, we get

ª w 2v w w º ª w 2u wu wu º
Mx Ny u « (av) (bv) cv » v « a b cu »
«¬ w x w y w x wy »¼ «¬ w x w y wx wy »¼
i.e.
vLu uL*v Mx Ny (1.27)
This is known as Lagrange identity which will be used in the subsequent discussion. The
operator L is a self-adjoint if and only if L = L*. Now we shall attempt to solve Cauchy’s
problem which is described as follows: Let
L(u ) F ( x, y ) (1.28)
with the condition (Cauchy data)
(i) u f ( x) on Γ, a curve in the xy-plane;
wu
(ii) g ( x) on (.
wn
That is u, and its normal derivatives are prescribed on a curve * which is not a characteristic line.
Let * be a smooth initial curve which is also continuous as shown in Fig. 1.2. Since
Eq. (1.24) is in canonical form, x and y are the characteristic coordinates. We also assume that
the tangent to * is nowhere parallel to the coordinate axes.
Q P(Y, I)
IR
Data curve R
x
O
Fig. 1.2 Cauchy data.
Let P(ξ , η ) be a point at which the solution to the Cauchy problem is sought. Let us draw
the characteristics PQ and PR through P to meet the curve Γ at Q and R. We assume that
u, ux, uy are prescribed along Γ. Let w IR be a closed contour PQRP bounding IR . Since
Eq. (1.28) is already in canonical form, the characteristics are lines parallel to x and y axes.
Using Green’s theorem, we have
ÔÔ ( M x N y )dxdy vÔ IR ( M dy N dx ) (1.29)
R
where ∂ IR is the boundary of IR . Applying this theorem to the surface integral of Eq. (1.27),
we obtain
∫ ∂ IR (M dy − N dx) = ∫∫ [vL(u) − uL * (v)] dx dy (1.30)

IR
In other words,
Ú Γ
( M dy - N dx) + Ú RP
( M dy - N dx) + Ú PQ
( M dy - N dx) = ÚÚ [vL(u) - uL * (v)] dx dy
R
Now using the fact that dy = 0 on PQ and dx = 0 on PR, we have
∫ Γ (M dy − N dx) + ∫ RP M dy − ∫ PQ N dx = ∫∫ [vL(u) − uL * (v)] dx dy (1.31)

IR
From Eq. (1.26), we find that

Q Q
∫ PQ N dx = ∫P buv dx + ∫P vux dx
Integrating by parts the second term on the right-hand side and grouping, the above equation
becomes
Q
ÚPQ
N dx = [uv] Qp + ÚP
u (bv - vx ) dx
Substituting this result into Eq. (1.31), we obtain
[uv] p = [uv]Q + ∫PQ u (bv − vx ) dx − ∫ RP u(av − v y ) dy

− ∫ Γ (M dy − N dx) + ∫∫ [vL(u) − uL * (v)] dx dy (1.32)
IR
Let us choose v ( x, y; ξ , η ) to be a solution of the adjoint equation

L * (v ) = 0 (1.33)
and at the same time satisfy the following conditions:
(i) vx = bv when y = η , i.e., on PQ (1.34a)
(ii) v y = av when x = ξ , i.e., on PR (1.34b)
(iii) v =1 when x = ξ , y =η (1.34c)
We call this function v ( x, y; ξ , η ) as the Riemann function or the Riemann-Green function.

Since L(u ) = F , Eq. (1.32) reduces to
[ u ] P [ uv ]Q ³* [u ( av v y ) dy v (bu u x ) dx ] ³³ (vF ) dx dy (1.35)

IR
This is called the Riemann-Green solution for the Cauchy problem described by Eq. (1.28)
when u and ux are prescribed on *. Equation (1.35) can also be written as
[u ] P [uv]Q Ô *
uv (a dy b dx ) Ô *
(uv y dy vu x dx) ÔÔIR (vF ) dx dy (1.36)
This relation gives us the value of u at a point P when u and ux are prescribed on Γ. But when
u and uy are prescribed on Γ, we obtain
[u ]P [uv]R ³ Γ uv (a dy b dx) ³ Γ (uvx dx vu y dy) ³³ (vF ) dx dy (1.37)

IR
By adding Eqs. (1.36) and (1.37), the value of u at P is given by
1 1
[u ]P
2
{[uv]Q [uv]R }
( ³
uv (a dy b dx)
2 ( ³
u (vx dx v y dy )
1
³
2 (
v (u x dx u y dy ) ³³ (vF ) dx dy (1.38)
IR
Thus, we can see that the solution to the Cauchy problem at a point (ξ , η ) depends only
on the Cauchy data on Γ. The knowledge of the Riemann-Green function therefore enables
us to solve Eq. (1.28) with the Cauchy data prescribed on a noncharacteristic curve.
EXAMPLE 1.13 Obtain the Riemann solution for the equation

w 2u
F ( x, y )
wxw y
given
(i) u f ( x) on Γ
wu
(ii) g ( x) on Γ
wn
where Γ is the curve y x.
Solution Here, the given PDE is
w 2u
L(u ) F ( x, y ) (1.39)
wxw y
We construct the adjoint L* of L as follows: setting

M auv uv y , N buv vu x
and comparing the given equation (1.39) with the standard canonical form of hyperbolic
equation (1.24), we have
a b c 0
Therefore,
M uv y , N vu x (1.40)
and
Mx Ny vu xy uvxy vL(u ) uL * (v)
Thus,
w 2v (1.41)
L * (v )
wxw y
Here, L = L* and is a self-adjoint operator. Using Green’s theorem
³³ (M x N y ) dx dy ³w IR (M dy N dx)
IR
we have
³³ [vL(u) uL * (v)] dx dy ³w IR (M dy N dx)

IR
or
³³ [vF uL * (v)] dx dy ³w IR (M dy N dx) (1.42)

IR
But
³w IR (M dy N dx) ³* (M dy N dx) ³QP (M dy N dx) ³PR (M dy N dx) (1.43)
where
§ wv wu ·
³ Γ (M dy N dx) ³ Γ ¨© u w y dy v w x dx ¸¹
From Fig. 1.3, we have on Γ, x y. Therefore, dx dy. Hence
§ wv wu·
³ Γ (M dy N dx) ³ Γ ¨© u w y v w x ¸¹ dx (1.44)
y
y =η
P(ξ, η) Q
x=ξ x
y=
R
x
O
Fig. 1.3 An illustration of Example 1.13.
since on QP, y = constant. Therefore, dy = 0. Thus,
∂u
∫QP (M dy − N dx) = ∫QP − N dx = ∫QP − v ∂x dx (1.45)
Similarly, on PR, x = constant. Hence, dx = 0. Thus,

∂v
∫PR (M dy − N dx) = ∫PR M dy = ∫PR −u ∂ y dy (1.46)
Substituting Eqs. (1.44)–(1.46) into Eq. (1.43), we obtain from Eq. (1.42), the relation
⎛ ∂v ∂u ⎞ ∂u ∂v
∫∫ [vF − uL * (v)] dx dy = ∫Γ ⎜⎝ −u ∂ y dx − v ∂ x dx ⎟⎠ + ∫QP −v ∂ x dx + ∫PR −u ∂ y dy
IR
But
∂u ∂v
∫QP −v ∂ x dx = [−vu] Q + ∫QP u ∂ x dx
P
Therefore,
⎛ ∂v ∂u ⎞
∫∫ [vF − uL * (v)] dx dy = ∫Γ ⎜⎝ −u ∂ y dx − v ∂ x dx ⎟⎠
IR
∂v ∂v
∫QP u ∂ x dx + ∫PR − u ∂ y dy
P
+ [−vu ] Q + (1.47)
Now choosing v ( x, y; ξ , η ) as the solution of the adjoint equation such that
(i) L * v = 0 throughout the xy-plane

∂v
(ii) =0 when y = η , i.e., on QP
∂x
wv
(iii) 0 when x ξ , i.e., on PR
wy
(iv) v 1 at P (ξ , η ).
Equation (1.47) becomes
§ wv wu ·
³³ (vF ) dx dy ³ ( ¨© u w y dx v w x dx ¸¹ (uv)Q (u)P
IR
or
§ wv wu ·
(u ) P (uv)Q ³ ( ¨© u w y dx v w x dx ¸¹ ³³ (v F ) dx dy (1.48)
IR
However,
© w w ¸
(uv) R
(uv)Q ± ( d (uv) ±( ª w x (uv) dx w y (uv) dy ¹
« º

±( (u x v dx uvx dx u y v dy uv y dy )
Now Eq. (1.48) can be rewritten as
(u ) P (uv) R ³ ( (u y v dy uv y dy) ³³ (vF ) dx dy (1.49)

IR
Finally, adding Eqs. (1.48) and (1.49), we get
1 1
(u ) P
2
[(uv)Q (uv) R ]
2 ( ³
(uvx dx vu x dx)
1
³
2 (
(u y v dy uv y dy ) ³³ (vF ) dx dy
IR
EXAMPLE 1.14 Verify that the Green function for the equation
w 2u 2 §w u w u ·
0
w x w y x y ©¨ w x w y ¹¸
subject to u 0, w u /w x 3x 2 on y x, is given by
( x y ) {2 xy (ξ η ) ( x y ) 2ξη}
v ( x, y; ξ , η )
(ξ η )3
and obtain the solution of the equation in the form
u ( x y ) (2 x 2 xy 2 y 2 )
Solution In the given problem,
w 2u 2 wu 2 wu
L(u ) 0 (1.50)
wxw y x y wx x y w y
Comparing this equation with the standard canonical hyperbolic equation (1.24), we have
2
a b , C 0, F 0
x y
Its adjoint equation is L * (v) 0, where
w 2v w § 2v · w § 2 v ·
L * (v ) . (1.51)
w x w y w x ©¨ x y ¸¹ w y ¨© x y ¸¹
such that
(i) L * v 0 throughout the xy-plane
wv 2
(ii) v on PQ, i.e., on y η
wx x y
wv 2
(iii) v on PR, i.e., on x ξ
wy x y
(1.52)
(iv) v 1 at P(ξ , η ).
If v is defined by
( x y)
v ( x, y; ξ , η ) [2 xy (ξ η ) ( x y ) 2ξη ] (1.53)
(ξ η )3
Then
wv x y ª 2 xy (Y I ) ( x y ) 2YI º
[2 y (Y I )] « »
wx (Y I )3 ¬ (Y I )3 ¼
or
wv 1
[4 xy 2 y 2 2 x (Y I ) 2YI ] (1.54)
wx (Y I ) 3
and
w 2v 4( x y )
(1.55)
wxw y (ξ η )3
wv 1
[4 xy 2 x 2 2 y (ξ η ) 2ξη ] (1.56)
wy (ξ η )3
Using the results described by Eqs. (1.53)–(1.56), Eq. (1.51) becomes
w 2v 2 §wv wv · 4v
L * (v )
w x w y x y ©¨ w x w y ¹¸ ( x y )2
4 ( x y) 2
[4 xy 2 ( x 2 y 2 )]
(ξ η ) 3
( x y ) (ξ η ) 3
or
4 ( x y) 4 ( x y)
L * (v ) 3
0
(Y I ) (Y I )3
Hence condition (i) of Eq. (1.52) is satisfied. Also, on y η.
wv 1
[4 xI 2I 2 2 x (Y I ) 2YI ]
wx (Y I ) 3
or
wv 1
[2η 2 2 x (ξ η ) 2ξη ] (1.57)
wx y η (ξ η ) 3
Also, 2v/( x y ) at y η is given by
2v 2 x η
[2 xη (ξ η ) ( x η ) 2ξη ]
x y x η (ξ η )3
1
[2η 2 2 x (ξ η ) 2ξη ] (1.58)
(ξ η ) 3
From Eqs. (1.57) and (1.58), we get

wv 2
v at y η
wx x y
Thus, property (ii) in Eq. (1.52) has been verified. Similarly, property (iii) can also be
verified. Also, at x ξ , y η ,
ξ η (ξ η ) (ξ η ) 2
v [2ξη (ξ η )2 2ξη ] 1
(ξ η ) 3
(ξ η )3
Thus property (iv) in Eq. (1.52) has also been verified.
From Eqs. (1.50) and (1.51), we have

w 2u w 2v w § 2vu · w § 2vu ·
vL(u ) uL * (v) v u
w x wy w x w y w x ¨© x y ¸¹ w y ¨© x y ¸¹
w § w u · w § w v · w § 2vu · w § 2vu ·
¨v ¸ u
w y © w x ¹ w x ¨© w y ¸¹ w x ¨© x y ¸¹ w y ¨© x y ¸¹
w § 2vu w v · w § 2vu wu ·
¨ u ¸ ¨ v ¸
wx©x y w y¹ w y ©x y wx¹
wM wN

wx wy
where
2uv wv 2vu wu
u M , N v
x y wy x y wx
Now using Green’s theorem, we have
Q
³³ [vL(u) uL * (v)] dx dy ³w IR (M dy N dx) ³R (M dy N dx)
IR
P R
³Q (M dy N dx) ³P (M dy N dx) (1.59)
(see Fig. 1.3) on QP, y C. Hence, dy 0; on PR, x C . Therefore, dx 0

Q ª 2uv wv½ 2uv wu½ º
³R «®
¬¯ x y
u ¾ dy ®
w y¿ ¯x y
v ¾ dx »
w x¿ ¼
P ª 2uv w u ½º R ª 2uv w v ½º
³Q «®
¬¯ x y
v ¾» dx
w x ¿¼ ³P «®
¬¯ x y
u ¾» dy
w y ¿¼
However,
P § 2uv wu · P 2uv P wv
³Q ¨© x y v w x ¸¹ dx ³Q ³Q u w x dx
P
dx (uv) Q
x y
Now, using the condition u 0 on y x, Eq. (1.59) becomes
Q § 2uv wv · Q § 2uv wu ·
³³ [vL(u) uL * (v)] dx dy ³R ¨© x y u w y ¹¸ dy ³R ¨© x y v w x ¹¸ dx
IR
P 2uv P wv
³Q x y
dx (uv) P (uv)Q ³Q u w x dx
P 2uv R § wv ·
³Q x y
dy ³P u
©¨ w y ¹¸
dy
Also, using conditions (ii)–(iv) of Eq. (1.52), the above equation simplifies to
Q wu
(u ) P (uv)Q ³R v
wx
dx
Now using the given condition, viz.

wu
3x 2 on RQ
wx
we obtain
Q ª 2 x [2 x 2 2YI ] º
(u ) P (uv)Q 3 ³R x2 «
«¬ (Y I )
3 » dx
»¼
12 I

(Y I ) 3 ³Y ( x5 x3YI ) dx
12 ª1 6 6 1 4 4 º
«¬ 6 (I Y ) 4 YI (I Y ) »¼
3
(Y I )
Y 2 I2
[2 (Y 4 Y 2I 2 I 4 ) 3YI (Y 2 I 2 )]
(Y I )3
(Y I ) (2Y 2 YI 2I 2 )
Therefore,
u ( x, y ) ( x y ) (2 x 2 xy 2 y 2 )
Hence the result.
EXAMPLE 1.15 Show that the Green’s function for the equation
w 2u
u 0
wxw y
is
v ( x, y; Y , I ) J0 2 ( x Y ) ( y I)
where J0 denotes Bessel’s function of the first kind of order zero.
Solution Comparing with the standard canonical hyperbolic equation (1.24), we have
a b 0, c 1
It is a self-adjoint equation and, therefore, the Green’s function v can be obtained from
w 2v
v 0
wxw y
subject to
wv
0 on y η
wx
wv
0 on x ξ
wy
v 1 at x ξ , y η
Let
φk a ( x ξ ) ( y η)
wv w v wφ
wx wφ w x
But
wφ
kφ k 1 a ( y η)
wx
Therefore,
wφ a 1 k
φ ( y η)
wx k
Thus,
wv w v a 1 k
φ ( y η)
wx wφ k
w 2v w ª w v a 1 k º
« φ ( y η )»
wxw y w y ¬w φ k ¼
a ª 1 k w v w v wφ w 2v w φ º
«φ (1 k ) φ k ( y η ) φ 1 k ( y η ) 2 »
k ¬« wφ wφ w y w φ w y »¼
However,
wφ a 1 k
φ (x ξ )
wy k
Therefore,
w 2v a ª 1 k w v k a 1 k w v 1 k w 2 v a 1 k º
« φ (1 k ) φ ( x ξ ) ( y η ) φ φ ( y η ) φ ( x ξ )»
wxw y k ¬« wφ k wφ wφ k2
¼»
Hence,
∂ 2v
+v=0
∂x∂ y
gives
a ⎡ φ k 2 (1− k ) d 2 v φ 1− k dv dv ⎤
⎢ φ + (1 − k ) + φ 1− k ⎥+v=0
k ⎢⎣ k dφ 2 k dφ dφ ⎥⎦
or
a ⎛ 2− k d 2 v 1− k dv
⎞
⎜ φ + φ +v=0
k2 ⎝ dφ 2 dφ ⎟⎠
or
k2 k
φ 2 v ′′ + φ v ′ + φ v=0
a
Let k = 2, a = 4. Then the above equation reduces to
1
φ 2 v ′′ + φ v ′ + φ 2 v = 0 = v ′′ + v ′ + v (Bessel’s equation)
φ
Its solution is known to be of the form
v = J 0 (φ ) = J 0 2( x − ξ )( y − η)
which is the desired Green’s function.
1.6 LINEAR PARTIAL DIFFERENTIAL EQUATIONS WITH CONSTANT

COEFFICIENTS
An nth order linear PDE with constant coefficients can be written in the form
∂nu ∂nu ∂nu ∂nu

a0 + a1 + a2 + L + an
∂x n ∂x n -1∂y ∂x n - 2 ∂ 2 y ∂y n
= f(x, y) (1.60)
where a0, a1, …, an are constants; u is the dependent variable; x and y are independent
∂ ∂
variables. Introducing the standard differential operator notation, such as D = , D¢ = ,
∂x ∂y
the above equation can be rewritten as
F(D, D¢)u = (a0Dn + a1Dn–1D¢ + a2Dn–2D¢2 + … + anD¢n)u = f(x, y) (1.61)
It can also be written in more compact form as
F ( D, D ¢)u = Â ÂC D D¢
i j
ij
i j
(1.62)
∂ ∂ ∂n ∂n
where Cij are constants, D = , D ¢ = , D n = n , D ¢ n = n , etc.
∂x ∂y ∂x ∂y
As in the case of linear ODE with constant coefficients, the complete solution of Eq. (1.61)
consists of two parts:
(i) the complementary function (CF), which is the most general solution of the
equation F(D, D¢)u = 0, the one containing, n arbitrary functions, where n is the
order of the DE.
(ii) the particular integral (PI), is a particular solution, which is free from arbitrary
constants or functions of the equation F(D, D¢)u = f(x, y).
The complete solution of Eq. (1.61) is then
u = CF + PI (1.63)
It may be noted that, if all the terms on the left-hand side of Eq. (1.61) are of the same
order, it is said to be a homogeneous equation otherwise, it is a non-homogeneous equation.
Now, we shall study few basic theorems as is the case in ODEs.
Theorem 1.1 If uCF and uPI are respectively the complementary function and particular
integral of a linear PDE, then their sum (uCF + uPI) is a general solution of the given PDE.
Proof Since F(D, D¢)uCF = 0,
and F(D, D¢)uPI = f(x, y),
we arrive at
F(D, D¢)uCF + F(D, D¢)uPI = f(x, y).
showing that (uCF + uPI) is in fact a general solution of Eq. (1.61). Hence proved.
Theorem 1.2 If u1, u2, …, un are the solutions of the homogeneous PDE: F(D, D¢)u = 0,
n
then ÂC u , where C
i =1
i i i are arbitrary constants, is also a solution.
Proof Since we observe that

F(D, D¢)(Ciui) = CiF(D, D¢)ui
n n
and F ( D, D ¢) Âi =1
vi = Â F (D, D ¢)v
i =1
i
For any set of functions vi, we find at once,

n n
F ( D, D ) Ç i 1
Ci ui Ç F (D, D )(C u )
i 1
i i
Ç C F (D, D )u
i 1
i i 0
Hence proved.
We shall now classify linear differential operator F(D, D¢) into reducible and irreducible
types, in the sense that F(D, D¢) is reducible if it can be expressed as the product of linear
factors of the form (aD + bD¢ + c), where a, b and c are constants, otherwise F(D, D¢) is
irreducible. For example, the operator
F(D, D¢)u = (D2 – D¢2 + 3D + 2D¢ + 2)u
= (D + D¢ + 1)(D – D¢ + 2)u
is reducible. While the operator F(D, D¢)u = (D2 – D¢), is irreducible, due to the fact that it
cannot be factored into linear factors.
1.6.1 General Method for Finding CF of Reducible Non-homogeneous

Linear PDE
The general strategy adopted for finding the CF of reducible equations is stated in the
following theorems:
Theorem 1.3 If the operator F(D, D¢) is reducible that is, if (aiD + biD¢ + ci) is a factor of
F(D, D¢) and fi(x) is an arbitrary function of a single variable x, then, if ai ¹ 0,
È c Ø
ui = exp É i x Ù fi(bix – aiy) (1.64)
Ê ai Ú
is a solution of the equation F(D, D¢)u = 0 (Sneddon, 1986).
Proof Using product rule of differentiation, Eq. (1.64) gives
È ci Ø È ci Ø È c Ø
Dui ÉÊ a ÙÚ exp ÉÊ a x Ù Ii (bi x ai y) bi exp É i x Ù Ii(bi x ai y)
i i Ú Ê ai Ú
ci È c Ø
ui bi exp É i x Ù I (bi x ai y).
ai Ê ai Ú
Similarly, we get
È c Ø
D ui ai exp É i x Ù I (bi x ai y).
Ê ai Ú
Thus, we observe that

(aiD + biD¢ + ci)ui = 0 (1.65)
That is, if the operator F(D, D¢) is reducible, the order in which the linear factors appear
is immaterial. Thus, if
ÏÔ n ¸Ô
F ( D, D ¢)ui = Ì
ÔÓ
’ ¢(a D + b D ¢ + c )˝Ô (a D + b D ¢ + c )u
j j j i i i i (1.66)
j =1 ˛
where, the prime on the product indicates that the factor corresponding to i = j is omitted.
Combining Eqs. (1.65) and (1.66), we arrive at the result F(D, D¢)ui = 0. Hence proved.
It may be noted that if no two factors of Eq. (1.65) are linearly independent, then the
general solution of Eq. (1.66) is the sum of the general solutions of the equations of the form
(1.65). For illustration, we consider the following examples:
EXAMPLE 1.16 Solve the following equation (D2 + 2DD¢ + D¢2 – 2D – 2D¢)u = 0.
Solution Observe that the given PDE is non-homogeneous and can be factored as
(D2 + 2DD¢ + D¢2 – 2D – 2D¢)u = (D + D¢)(D + D¢ – 2)u.
Using the result of Theorem 1.3, we get the general solution or the CF as
uCF = f1(x – y) + e2xf2(x – y).
On similar lines, we can also establish the following result:
Theorem 1.4 (Sneddon, 1986) Let (bi D¢ + ci) is a factor of F(D, D¢)u, and fi(x) is an
arbitrary function of a single variable x, then, if bi π 0, we have
Ê c ˆ
ui = exp - i
ÁË b y ˜ fi(bix) (1.67)
i ¯
as a solution of the equation F(D, D¢)u = 0.

Proof Suppose, the factorisation of F(D, D¢) = 0 gives rise to a multiple factors of the
form (aiD + biD¢ + ci)n, the solution of F(D, D¢)u = 0 can be obtained by the application of
Theorems 1.3 and 1.4. For example, let us find the solution of
(aiD + biD¢ + ci)2u = 0 (1.68)
We set,
(aiD + biD¢ + ci)u = U,
then, Eq. (1.68) becomes
(aiD + biD¢ + ci)U = 0.
Using Theorem 1.3, its solution is found to be
Ê c ˆ
U = exp Á - i x ˜ fi(bix – aiy).
Ë ai ¯
Further, assume that ai π 0, now, in order to find u, we have to solve
Ê c ˆ
(aiD + biD¢ + ci)u = exp Á - i x ˜ fi(bix – aiy).
Ë ai ¯
This is a first order PDE. Using the Lagrange’s method (Section 0.8), its auxiliary equations
are
dx dy du
= = (1.69)
ai bi Ê ci ˆ
-ci u + exp Á - x ˜ fi (bi x - ai y)
Ë ai ¯
One solution of which is given by
bix – aiy = l (a constant) (1.70)
Substituting this solution into the first and third of the auxiliary equations, we obtain
dx du
=
ai Ê c ˆ
-ci u + exp Á - i x ˜ fi (l )
Ë ai ¯
du ci 1 Ê c ˆ
or + u = exp Á - i x ˜ f i ( l ).
dx ai ai Ë ai ¯
This being an ODE, its solution can be readily written as
Êc ˆ 1
u exp Á i x ˜ = xfi (l ) + m (constant)
Ë ai ¯ ai
1 Ê c ˆ
or u= [ xfi (l ) + m ] exp Á - i x˜ (1.71)
ai Ë ai ¯
Thus, the solution of Eq. (1.68) is given by
u = [xfi(bix – aiy) + yi(bix – aiy]e–ci/aix (1.72)
where fi and yi are arbitrary functions.
In general, if there are n, multiple factors of the form (aiD + biD¢ + ci), then the solution
of (aiD + biD¢ + ci)n u = 0 can be written as
Ê c ˆÈ n ˘
u = exp Á - i
Ë ai
Â
x ˜ Í x j -1fij (bi x - ai y) ˙
¯ ÍÎ j =1 ˙˚
(1.73)
Here follows an example for illustration.

EXAMPLE 1.17 Find the solution of the equation (2D – D¢ + 4)(D + 2D¢ + 1)2u = 0.
Solution The complementary function (CF) corresponding to the factor (2D – D¢ + 4)
is e–2x f(–x – 2y). Similarly, CF corresponding to
(D + 2D¢ + 1)2 is e–x[y1(2x – y) + xy2(2x – y)].
Thus, the CF for the given PDE is given by
u = e–2xf(x + 2y) + e–x[y1(2x – y) + xy2(2x – y)],
where, f, y1, y2 are arbitrary functions.
1.6.2 General Method to Find CF of Irreducible Non-homogeneous Linear PDE

If the operator F(D, D¢) is irreducible, we can find the complementary function, containing
as many arbitrary functions as we wish by a method which is stated in the following theorem:
Theorem 1.5 The solution of irreducible PDE F(D, D¢)u = 0 is

u Ç ci exp(ai x bi y) (1.74)
i 1
Proof Let us assume the solution of F(D, D¢)u = 0 in the form, u = ceax+by, where a, b and
c are constants to be determined. Then, we have
Diu = caieax+by, D¢ju = cb j eax+by,
DiD¢ju = caib jeax+by
Thus, F(D, D¢)u = 0 yields
c[F(a, b)]eax+by = 0
where c is an arbitrary constant, not zero, holds true iff
F(a, b) = 0, (1.75)
indicating that there exists infinite pair of values (ai, bi) satisfying Eq. (1.75). Hence,

u Ç ci ea x b y
i i
(1.76)
i 1
is a solution of irreducible equation

F(D, D¢)u = 0, (1.77)
provided F(ai, bi) = 0 (1.78)
It may be noted that this method is applicable even for reducible equations. Here follows
an example for illustration:
EXAMPLE 1.18 Solve the following equation (2D2 – D¢2 + D)u = 0.
Solution The given equation is an irreducible non-homogeneous PDE. Using the result
of Theorem 1.5, it follows immediately that
•
u = uCF = Â ci ea x +b y
i i
i =1
where ai, bi are related by F(ai, bi) = 0.
That is,
2ai2 – bi2 + ai = 0
which gives bi2 = 2ai2 + ai.
1.6.3 Methods for Finding the Particular Integral (PI)

To find the PI of Eq. (1.61), we rewrite the same in the form
1
u= f ( x, y) (1.79)
F ( D, D ¢)
Very often, the operator F–1(D, D¢) can be expanded, using binomial theorem and interpret
the operators D–1, (D¢)–1 as integrations. That is,
1
D -1 f ( x, y) =
D
f ( x, y) = Ú f ( x, y)dx ,
y constant
1
and
D¢
f ( x , y) = Ú f ( x, y)dy.
x constant
We present below different cases for finding the PI, depending on the nature of f(x, y).
Case I Let f(x, y) = exp(ax + by), then
1 1
eax + by = e ax + by (1.80)
F ( D, D ¢) F (a, b)
By direct differentiation, we find DiD¢jeax+by = aib j eax+by.
In other words,
F(D, D¢)eax+by = F(a, b)eax+by,
that is,
1
e ax + by = F (a, b) e ax + by .
F ( D, D ¢)
Dividing both sides by F(a, b), we get
1 1
eax + by = e ax + by ,
F ( D, D ¢) F (a, b)
provided F(a, b) π 0.
Case II Let f(x, y) = sin(ax + by) or cos(ax + by), where a and b are constants, then, since
D2 sin(ax + by) = –a2 sin(ax + by)
DD¢ sin(ax + by) = –ab sin(ax + by)
D¢2 sin(ax + by) = –b2 sin(ax + by)

We notice that
1
sin(ax + by)
F ( D, D¢)
is obtained by setting, D2 = –a2, DD¢ = –ab, D¢2 = –b2 provided F(D, D¢) π 0. Thus,
F(D2, DD¢, D¢2) sin(ax + by) = F(–a2, –ab, –b2) sin(ax + by)
1 1
or sin( ax + by) = sin( ax + by) (1.81)
F ( D , DD ¢, D ¢ )
2 2
F ( - a , - ab, - b 2 )
2
Similarly,
1 1
cos(ax + by) = cos(ax + by) (1.82)
F ( D , DD ¢, D ¢ )
2 2
F ( - a , - ab, - b2 )
2
Case III Let f(x, y) is of the form xpyq, where p and q are positive integers. Then, the PI
can be obtained by expanding F(D, D¢) in ascending powers of D or D¢.
Case IV Let f(x, y) is of the form eax+by f(x, y).
Then,
F(D, D¢)[eax+by f(x, y)] = eax+by F(D + a, D¢ + b) f(x, y).
Let us recall Leibnitz’s theorem for the nth derivative of a product of functions; thus, we have
n
D n [eax f ] = Â ncr (Dr eax )(Dn-rf )
r =0
Ê n ˆ
Â
= e ax Á ncr ar D n -r f ˜
Ë r =0 ¯
= eax(D + a)nf
Applying this result, we arrive at
F(D, D¢)[eax+byf(x, y)] = eax+by F(D + a, D¢ + b) f(x, y) (1.83)
Hence, it follows that
1 1
[e ax + byf ( x, y)] = e ax + by f ( x, y)
F ( D, D ¢) F ( D + a, D ¢ + b)
1
= e ax ◊ ebyf ( x, y)
F ( D + a, D ¢)
1
= eby ◊ e ax f ( x, y) (1.84)
F ( D, D ¢ + b)
For illustration of various cases to find PI, here follows several examples:
EXAMPLE 1.19 Solve the equation (D2 + 3DD¢ + 2D¢2)u = x + y.
Solution The given PDE is reducible, since it can be factored as
(D + D¢)(D + 2D¢)u = x + y (1)
Therefore,
CF = f1(x – y) + f2(2x – y) (2)
where f1 and f2 are arbitrary functions.
The PI of the given PDE is obtained as follows:
1
PI ( x y)
( D 3DD 2 D 2 )
2
1
( x y)
È D D 2 Ø
D É1 3
2
2 2 Ù
Ê D D Ú
1
1 Ë È D D 2 Ø Û
Ì1 É 3 2 Ù Ü ( x y)
D 2 ÌÍ Ê D D 2 Ú ÜÝ
1 Ë D Û
Ì1 3 D Ü ( x y)
D2 Í Ý
1 Ë 3 Û
2 Ì
x y (1) Ü
D Í D Ý
1 x2 x3
2
[ y 2x] y (3)
D 2 3
Adding Eqs. (2) and (3), we have the complete solution of the given PDE as
x2 x3 .
u I1 ( y x ) I2 ( y 2 x ) y
2 3
EXAMPLE 1.20 Solve the following equation (D – D¢ – 1)(D – D¢ – 2)u = e2x–y + x.
Solution The CF of the given PDE is
CF = exf1(x + y) + e2xf2(x + y) (1)
2x–y
The PI corresponding to the term e is
1 1 2xy
e2 x y e (2)
(2 1 1)(2 1 2) 2
while the PI corresponding to the term x is

1
1 È D D Ø
(1 D D )1 É1 x
2 Ê 2 2 ÙÚ
1Ë È D D ØÛ
Ì (1 D D ) É1 Ù Ü x
2Í Ê 2 2 ÚÝ
1 È 1Ø 1È 3Ø
(1 D D ) É x Ù Éx Ù (3)
2 Ê 2Ú 2Ê 2Ú
Combining Eqs. (1), (2) and (3), the complete solution of the given PDE is found to be
1 2xy x 3
u e x I1 ( x y) e2 x I2 ( x y) e .
2 2 4
EXAMPLE 1.21 Solve the following equation
(D2 + 2DD¢ + D¢2 – 2D – 2D¢)u = sin(x + 2y).
Solution The given PDE can be factored and rewritten as
(D + D¢)(D + D¢ – 2)u = sin(x + 2y) (1)
for which the CF is given by
CF = f1(x – y) + e2xf2(x – y) (2)
while
1
PI sin( x 2 y) .
( D 2 DD D 2 2 D 2 D )
2
Setting D2 = –1, DD¢ = –2, D¢2 = –4, we get

1
PI sin( x 2 y)
(2 D 2 D 9)
[2( D D ) 9]
sin( x 2 y )
[4( D 2 DD D 2 ) 81]
2
2( D D ) 9
sin( x 2 y)
117
1
[2 cos(x + 2y) + 4 cos(x + 2y) – 9 sin(x + 2y)]
117
1
[2 cos(x + 2y) – 3 sin(x + 2y)] (3)
39
Adding Eqs. (1), (2) and (3), we find that the complete solution of the given PDE as
1
u = f1(x – y) + e2xf2(x – y) + [2 cos(x + 2y) – 3 sin(x + 2y)].
39
EXAMPLE 1.22 Solve the following equation
(D2 – DD¢)u = cos x cos 2y.
D(D – D¢)u = cos x cos 2y (1)
Its CF is given by
CF = f1(y) + f2(y + x), (2)
while its PI is given by
1 1
PI ¹ [cos(x + 2y) + cos(x – 2y)]
( D DD )
2 2
1Ë 1 1 Û
Ì cos( x 2 y) cos( x 2 y)Ü
2 Í ( 1 2) ( 1 2) Ý
1 1
cos( x 2 y) cos( x 2 y) (3)
2 6
Hence, the complete solution of the given PDE is given by
1 1
u = f1(y) + f2(y + x) + cos(x + 2y) – cos(x – 2y).
2 6
EXAMPLE 1.23 Find the solution of
(D2 + DD¢ – 6D¢2)u = y cos x.
(D + 3D¢)(D – 2D¢)u = y cos x (1)
Its CF is given by
CF = f1(3x – y) + f2(2x + y) (2)
The PI of the given PDE is
1 1
PI ¹ y cos x (3)
( D 3D ) ( D 2 D )
1
Applying the operator first on y cos x
( D 2 D )
(D – 2D¢)u = y cos x
Its auxiliary equations are

dx dy du .
1 2 y cos x
The first two members give
y + 2x = l (constant).
From the first and the third members, we have
du = y cos x dx = (l – 2x) cos x dx,
on integration, we get
u Ô (O 2 x ) cos x dx
where l is to be replaced by (y + 2x) after integration. Now, Eq. (3) gives
1 Ë(O 2 x ) sin x sin x ( 2)dx Û
PI
( D 3D ) ÍÌ Ô ÝÜ
1
[(O 2 x ) sin x 2 cos x ]
( D 3D )
1
[ y sin x 2 cos x ]
( D 3D )
Ô [(O 3x ) sin x 2 cos x ]dx

Ô
(O 3 x )( cos x ) 3 cos x dx 2 sin x .
Now, replacing l by (y – 3x), we get

PI = –y cos x + sin x (4)
Hence, the solution of the given PDE is found to be
u = CI + PI
= f1(3x – y) + f2(2x + y) – y cos x + sin x.
EXAMPLE 1.24 Show that a linear PDE of the type
i j u
ÇÇ aij x i y j x i y j f ( x, y)
i j
can be reduced to a one with constant coefficients by the substitution

x = log x, h = log y.
Solution
∂u ∂u ∂ x 1 ∂ u
= =
∂x ∂x ∂ x x ∂ x
∂u ∂u
or x =
∂x ∂x
That is,
∂ ∂
x = = D (say) (1)
∂x ∂ x
Therefore,
∂ Ê n -1 ∂ n -1u ˆ n ∂ u
n
n -1 ∂
n -1
u
x Á x - ˜ = x + ( n - 1) x -
∂x Ë ∂x ¯n 1
∂x n
∂x 1n
n -1
∂nu Ê ∂ ˆ n -1 ∂ u
or xn = Á x - n + 1˜ x (2)
∂x n Ë ∂ x ¯ ∂x n -1
By setting n = 2, 3, 4, … in Eq. (2), we obtain
∂2 u ∂u
x2 = ( D - 1) x = D( D - 1)u ,
∂x 2 ∂x
∂3 u
= D( D - 1)( D - 2)u ,
x3
∂x 3
and so on. Similarly, we can show that
∂u ∂u
y = = D ¢u,
∂y ∂h
∂2u
y2 = D ¢( D ¢ - 1)u ,
∂y 2
∂2u
and xy = DD ¢u
∂ x ∂y
and so on. Substituting these results into the given PDE, it becomes
F(D, D¢)u = f(ex, eh) = f(x, h) (3)
where,
∂ ∂
D= , D¢ =
∂x ∂h
Thus, Eq. (3) can be seen as a PDE with constant coefficients.
For illustration, here follows an example.

EXAMPLE 1.25 Solve the following PDE
(x2D2 + 2xy DD¢ + y2D¢2)u = x2y2 (1)
Solution Using the substitution
x = log x, h = log y and using the notation
∂ ∂
= D, = D¢
∂x ∂h
respectively, the given PDE reduces to a PDE with constant coefficients, in the form
[D(D – 1) + 2DD¢ + D¢(D¢ – 1)]u = e2x+2h.
On rewriting, we have
(D + D¢)(D + D¢ – 1)u = e2x+2h (2)
The CF of Eq. (2) is given by
CF = f1(x – h) + ex f2 (x – h), (3)
while, its PI is obtained as
1
PI = e 2 x + 2h
(2 + 2)(2 + 2 - 1)
1 2 x + 2h
= e . (4)
12
Transforming back from (x, h) to (x, y), we find the complete solution of the given PDE as
x 2 y2
u = f1(log x – log y) + x f2(log x – log y) +
12
Ê xˆ Ê xˆ 1
or u = y 1 Á ˜ + x y 2 Á ˜ + x 2 y2.
Ë y¯ Ë y ¯ 12
1.7 HOMOGENEOUS LINEAR PDE WITH CONSTANT COEFFICIENTS

Equation (1.61) is said to be linear PDE of nth order with constant coefficients. It is also
called homogeneous, because all the terms containing derivatives are of the same order. Now,
Eq. (1.61) can be rewritten in operator notation as
[a0Dn + a1Dn–1 D¢ + a2Dn–2D¢2 + … + anD¢n]u = F(D, D¢)u = f(x, y) (1.85)
As in the case of ODE, the complete solution of Eq. (1.85) consists of the sum of CF
and PI.
1.7.1 Finding the Complementary Function

Let us assume that the solution of the equation F(D, D¢)u = 0 in the form
u = f(y + mx) (1.86)
Then,
j
Diu = mifi(y + mx), D¢u = f j(y + mx),
and DiD¢ju = mifi+j (y + mx).
Substituting these results into F(D, D¢)u = 0, we obtain
(a0mn + a1mn–1 + a2mn–2 + … + an)fn(y + mx) = 0,
which will be true, iff
a0mn + a1mn–1 + a2mn–2 + … + an = 0 (1.87)
This equation is called an auxiliary equation for F(D, D¢)u = 0.
Let m1, m2, …, mn are the roots of Eq. (1.87). Depending on the nature of these roots,
several cases arise:
Case I When the roots m1, m2, …, mn are distinct. Corresponding to m = m1, the CF is
u = f1(y + m1x). Similarly, u = f2(y + m2x), u = f3(y + m3x), etc. are all complementary
functions. Since, the given PDE is linear, using the principle of superposition, the CF of
Eq. (1.85) can be written as
CF = f1(y + m1x) + f2(y + m2x) + … + fn(y + mnx)
where, f1, f2, …, fn are arbitrary.
Case II When some of the roots are repeated. Let two roots say m1 and m2 are repeated,
and each equal to m. Consider the equation
(D – mD¢)(D – mD¢)u = 0.
Setting (D – mD¢)u = z, the above equation becomes
(D – mD¢)z = 0,
∂z ∂z
or -m =0
∂x ∂y
which is of Lagrange’s form. Writing down its auxiliary equations, we have
dx dy dz .
= =
1 -m 0
The first two members gives y + mx = constant = a(say).
The third member yields z = constant.
Therefore, z = f(y + mx) is a solution.
Substituting for z, we get
(D – mD¢)u = f(y + mx)
which is again in the Lagranges form, whose auxiliary equations are

dx dy du
1 m I ( y mx )
which gives,
y + mx = constant
and u = xf(y + mx) + constant
Hence, the CF is
u = xf(y + mx) + y(y + mx).
In general, if the root m is repeated r times, then the CF is given by
u = xr–1f1(y + mx) + xr–2f2(y + mx) + … + fr(y + mx)
where, f1, f2, …, fr are arbitrary.
For illustration, we consider the following couple of examples.
EXAMPLE 1.26 Solve the following PDE (D3 – 3D2D¢ + 4D¢3)u = 0.
Solution Observe that the given PDE is a linear homogeneous PDE. Dividing throughout
by D¢ and denoting (D/D¢) by m, its auxiliary equation can be written as
m3 – 3m2 + 4 = (m + 1)(m – 2)2 = 0.
Therefore, the roots of the auxiliary equation are –1, 2, 2. Thus,
CF = f1(y – x) + f2(y + 2x) + x f3(y + 2x).
(D3 + DD¢2 – 10D¢3)u = 0.
Solution Observe that the given equation is a linear homogeneous PDE. Denoting (D/
D¢) by m. The auxiliary equation for the given PDE is given by
m3 + m – 10 = (m – 2)(m2 + 2m + 5) = 0.
Its roots are: 2, (–1 + 2i), (–1 – 2i).
Hence, the required CF is found to be
u = f1(y + 2x) + f2(y – x + 2ix) + f3(y – x – 2ix).
1.7.2 Methods for Finding the PI

Methods for finding the PI, in the case of linear homogeneous PDE’s are, similar to the one’s
developed in the Section 1.6 for linear non-homogeneous PDEs. That is, the PI for the
equation
F(D, D¢)u = f(x, y)
is obtained from
1
uPI = PI = f(x, y).
F ( D, D)
However, when the above stated methods fail we have a general method, which is applicable
whatever may be the form of f(x, y), and is presented below:
We have already assumed that F(D, D¢) can be factorised, in general, say into n linear
factors. Thus,
1
PI = f ( x, y)
F ( D, D ¢)
1
PI = f ( x, y)
( D - m1D ¢)( D - m2 D ¢) K ( D - mn D ¢)
1 1 1
= ◊ L f ( x, y ) .
( D - m1D ¢) ( D - m2 D ¢) ( D - mn D ¢)
In general, to evaluate
1
f ( x, y),
( D - mD ¢)
we consider the equation
(D – mD¢)u = f(x, y)
or p – mq = f(x, y) (Lagrange’s form)
for which the auxiliary equations are
dx dy du
= = .
1 - m f ( x, y)
Its first two members, yield
y + mx = c (constant)
The first and last members gives us
du = f(x, y)dx = f(x, c – mx).
u= Ú f ( x, c - mx )dx
1
or
( D - mD ¢)
f ( x, y) = Ú f ( x, c - mx )dx
After integration, we shall immediately replace c by (y + mx). Applying this procedure
repeatedly, we can find the PI for the given PDE. For illustration, we consider the following
examples:
(D2 – 4DD¢ + 4D¢2)u = e2x+y.
Solution The given equation is a linear homogeneous PDE. Its auxiliary equation can
be written as
m2 – 4m + 4 = (m – 2)2 = 0.
In this example, the roots are repeated and they are 2, 2. The complementary function and
particular integral are obtained as
CF = f1(y + 2x) + xf2(y + 2x) (1)
1
and PI = e2 x + y
( D - 2 D ¢) 2
If we set D = 2, D¢ = 1, we observe that F(D, D¢) = 0, which is a failure case. Therefore,

we shall adopt the general method to find PI. Now, noting that
1
( D - mD ¢)
f ( x, y) = Ú f ( x, c - mx )dx .
1 1
PI = ◊ e2 x + y
( D - 2 D ¢) ( D - 2 D ¢ )
1
=
( D - 2 D ¢) Ú
e(2 x + c - 2 x ) dx
1 1
= xec = xe y + 2 x
( D - 2 D ¢) ( D - 2 D ¢)
Ú
= xe( c - 2 x + 2 x ) dx = ec xdx Ú
x 2 x 2 y +2 x
= ec = e (2)
2 2
From Eqs. (1) and (2), the complete solution of the given PDE is found to be
x 2 y+2x
u = f1(y + 2x) + xf2(y + 2x) + e .
2
EXAMPLE 1.29 Find a real function u(x, y), which reduces to zero when y = 0 and satisfy
the PDE
∂2 u ∂2u
+ = - p ( x 2 + y 2 ).
∂x 2
∂y 2
Solution In symbolic form, the given PDE can be written as

(D2 + D¢2)u = –p(x2 + y2)
Its auxiliary equation is given by
(m2 + 1) = 0, which gives m = ±i.
Hence,
CF = f1(y + ix) + f2(y – ix) (1)
-1 p ( x 2 + y2 )
and PI = p ( x 2 + y2 ) = -
( D 2 + D ¢2 ) D ¢ 2 (1 + D 2 /D ¢ 2 )
-1
p Ê D2 ˆ
=- Á 1 + ˜ ( x 2 + y2 )
D ¢2 Ë D ¢2 ¯
p È D2 ˘ 2
=- Í1 - + L˙ (x + y )
2
D ¢ ÎÍ
2
D¢ 2
˚˙
p 2p
=- ( x 2 + y2 ) +
D¢ 2
D ¢4
p Ê y3 ˆ 2p
=- 2
+ +
D ¢ ÁË 3 ˜¯ D ¢3
x y y
Ê x 2 y 2 y 4 ˆ 2p y2
or PI = - p Á + ˜+ 2
Ë 2 12 ¯ D ¢ 2
Ê x 2 y2 y 4 ˆ y4
= -p Á + ˜ + 2p
Ë 2 12 ¯ 24
p
=- x 2 y2 (2)
2
Hence, the complete solution of the given PDE is found to be
p
u = f1(y + ix) + f2(y – ix) – x2y 2 (3)
2
Finally, the real function satisfying the given PDE is given by
p
u=- x 2 y2 (4)
2
which of course Æ 0 as y Æ 0.
EXERCISES
1. Find the region in the xy-plane in which the following equation is hyperbolic:
[( x − y )2 − 1] u xx + 2u xy + [( x − y )2 − 1] u yy = 0
2. Find the families of characteristics of the PDE

(1 x 2 ) u xx u yy 0
in the elliptic and hyperbolic cases.
3. Reduce the following PDE to a canonical form
u xx xyu yy 0
4. Classify and reduce the following equations to a canonical form:
(a) y 2u xx x 2u yy 0, x ! 0, y ! 0.
(b) u xx 2u xy u yy 0.
(c) e x u xx e y u yy u.
(d) x2u xx 2 xyu xy y 2u yy 0.

(e) 4u xx 5u xy u yy u x u y 2.
5. Reduce the following equation to a canonical form and hence solve it:
3u xx 10u xy 3u yy 0
6. If L(u ) c 2u xx utt , then show that its adjoint operator is given by
L* c 2 vxx vtt
7. Determine the adjoint operator L* corresponding to
L(u ) Au xx Bu xy Cu yy Du x Eu y Fu
where A, B, C, D, E and F are functions of x and y only.

8. Find the solution of the following Cauchy problem
u xy F ( x, y )
given
wu
u f ( x), g ( x) on the line y x
wn
using Riemann’s method which is of the form
1 1 y0
u ( x0 , y0 )
2
[ f ( x0 ) f ( y0 )]
2 ³x
0
g ( x) dx ³³ F ( x, y) dx dy
IR
where IR is the triangular region in the xy-plane bounded by the line y x and the
lines x x0 , y y0 through ( x0 , y0 ).
9. The characteristics of the partial differential equation

w 2z w 2z w 2z wz wz
2 cos 2 x 2 2 3 0
w x2 wxw y wy wx wy
when it is of hyperbolic type are… and… (GATE-Maths, 1997)
10. Using η x y as one of the transformation variable, obtain the canonical form of
w 2u w 2u w 2u
2 0.
w x2 w x w y w y2
(GATE-Maths, 1998)
Choose the correct answer in the following questions (11–14):
11. The PDE
y 3u xx ( x 2 1) u yy 0 is
(A) parabolic in {( x, y ) : x 0}
(B) hyperbolic in {( x, y ) : y ! 0}
(C) elliptic in IR 2
(D) parabolic in {( x, y ) : x ! 0}. (GATE-Maths, 1998)
12. The equation
x 2 ( y 1) z xx x ( y 2 1) z xy y ( y 2 1) z yy z x 0
is hyperbolic in the entire xy-plane except along
(A) x-axis (B) y-axis
(C) A line parallel to y-axis (D) A line parallel to x-axis.
(GATE-Maths, 2000)
13. The characteristic curves of the equation
x 2u xx y 2u yy x 2 y 2 x, x ! 0 are
(A) rectangular hyperbola (B) parabola
(C) circle (D) straight line.
(GATE-Maths, 2000)
14. Pick the region in which the following PDE is hyperbolic:
yu xx 2 xyu xy xu yy ux u y
(A) xy z 1 (B) xy z 0
(C) xy ! 1 (D) xy ! 0.
(GATE-Maths, 2003)
15. Solve the following PDEs:

(i) (D – D¢ – 1)(D – D¢ – 2)u = 0
(ii) (D + D¢ – 1)(D + 2D¢ – 3)u = 0
16. Solve the following PDE:
(D2 + DD¢ + D + D¢ + 1)u = 0
17. Solve the following PDEs:
(i) (D2 – DD¢ + D¢ – 1)u = cos(x + 2y)
(ii) D(D – 2D¢ – 3)u = ex+2y
(iii) (2D + D¢ – 1)2(D – 2D¢ + 2)3 = 0
18. Find the complete solution of the following PDEs:
(i) (x2D2 – 2xy DD¢ + y2D¢2 – xD + 3yD¢)u = 8(y/x)
(ii) D(D – 2D¢ – 3)u = ex+2y
19. Find the complete solution of the following PDEs:
(i) (D2 + 3DD¢ + 2D¢2)u = x + y
(ii) (D2 + D¢2)u = cos px cos qy
(iii) (D2– DD¢ – 2D¢2)u = (y – 1)ex
(iv) (4D2 – 4DD¢ + D¢2)u = 16 log (x + 2y)
(v) (D2 – 3DD¢ + 2D¢2)u = e2x+3y + sin (x – 2y)
CHAPTER 2
Elliptic Differential Equations
2.1 OCCURRENCE OF THE LAPLACE AND POISSON EQUATIONS

In Chapter 1, we have seen the classification of second order partial differential equation into
elliptic, parabolic and hyperbolic types. In this chapter we shall consider various properties
and techniques for solving Laplace and Poisson equations which are elliptic in nature.
Various physical phenomena are governed by the well known Laplace and Poisson equations.
A few of them, frequently encountered in applications are: steady heat conduction, seepage
through porous media, irrotational flow of an ideal fluid, distribution of electrical and magnetic
potential, torsion of prismatic shaft, bending of prismatic beams, distribution of gravitational
potential, etc. In the following two sub-sections, we shall give the derivation of Laplace and
Poisson equations in relation to the most frequently occurring physical situation, namely, the
gravitational potential.
2.1.1 Derivation of Laplace Equation

Consider two particles of masses m and m1 situated at Q and P separated by a distance r as
shown in Fig. 2.1. According to Newton’s universal law of gravitation, the magnitude of the
force, proportional to the product of their masses and inversely proportional to the square of
the distance, between them is given by
mm1
F G (2.1)
r2
)&
where G is the gravitational constant. It r represents the vector PQ, assuming unit mass at
Q and G 1, the force at Q due to the mass at P is given by
m1r §m ·
F ¨ 1¸ (2.2)
r 3 © r ¹
106
ELLIPTIC DIFFERENTIAL EQUATIONS 107
z
Q(m)
r
U
P
m1
S
O y
x
Fig. 2.1 Illustration of Newton’s universal law of gravitation.
which is called the intensity of the gravitational force. Suppose a particle of unit mass moves
under the attraction of a particle of mass m1 at P from infinity up to Q; then the work done
by the force F is
r r §m · m1
³ f F dr ³ f ¨© r1 ¸¹ dr r
(2.3)
This is defined as the potential V at Q due to a particle at P and is denoted by

m1 (2.4)
V
r
From Eq. (2.2), the intensity of the force at P is
F V (2.5)
Now, if we consider a system of particles of masses m1 , m2 , } , mn which are at distances
r1 , r2 , } , rn respectively, then the force of attraction per unit mass at Q due to the system is
n n
mi mi
F ¦
ri
¦ ri
(2.6)
i 1 i 1
The work done by the force acting on the particle is

n
r mi
³f F dr ¦ ri
V (2.7)
i 1
Therefore,
n n
mi m
2V 2 ¦ ri ¦ 2 rii 0, ri z 0 (2.8)
i 1 i 1
where
w2 w2 w2
2 div
w x2 w y2 w z2
is called the Laplace operator.
In the case of continuous distribution of matter of density ρ in a volume τ , we have
ρ (ξ , η , ζ )
V ( x, y , z ) ³³³ r
dτ (2.9)
τ
where r {( x ξ )2 ( y η )2 ( z ζ )2 }1/2 and Q is outside the body. It can be verified that
2V 0 (2.10)
which is called the Laplace equation.
2.1.2 Derivation of Poisson Equation
Consider a closed surface S consisting of particles of masses m1 , m2 , } , mn . Let Q be any

n
point on S. Let Σ mi M be the total mass inside S, and let g1 , g 2 , } , g n be the gravity
i 1 n
field at Q due to the presence of m1 , m2 , } , mn respectively within S. Also, let Σ gi g , the
i 1
entire gravity field at Q. Then, according to Gauss law, we have
³³ g dS 4π GM (2.11)
S
where M ³³³ ρ dτ , ρ is the mass density function and τ is the volume in which the masses
τ
are distributed throughout. Since the gravity field is conservative, we have
g V (2.12)
where V is a scalar potential. But the Gauss divergence theorem states that
³³ g dS ³³³ g dτ (2.13)
S τ
Also, Eq. (2.11) gives
³³ g dS 4π G ³³³ ρ dτ (2.14)

S τ
Combining Eqs. (2.13) and (2.14), we have
∫∫∫ (∇ ⋅ g + 4π G ρ ) dτ = 0
τ
implying
∇ ⋅ g = −4π G ρ = ∇ ⋅ ∇V
Therefore,
∇ 2V = −4π G ρ (2.15)
This equation is known as Poisson’s equation.
2.2 BOUNDARY VALUE PROBLEMS (BVPs)

The function V, whose analytical form we seek for the problems stated in Section 2.1, in
addition to satisfying the Laplace and Poisson equations in a bounded region IR in R3, should
also satisfy certain boundary conditions on the boundary ∂ IR of this region. Such problems
are referred to as boundary value problems (BVPs) for Laplace and Poisson equations. We
shall denote the set of all boundary points of IR by ∂ IR . By the closure of IR, we mean the
set of all interior points of IR together with its boundary points and is denoted
by IR. Symbolically, IR. = IR U ∂ IR .
If a function f ∈ c ( n ) ( f “belongs to” c(n)), then all its derivatives of order n are continuous.
If it belongs to c(0), then we mean f is continuous.
There are mainly three types of boundary value problems for Laplace equation. If f ∈ c (0) and
is specified on the boundary ∂ IR of some finite region IR, the problem of determining a
function ψ ( x, y, z ) such that ∇ 2ψ = 0 within IR and satisfying ψ = f on ∂ IR is called the
boundary value problem of first kind, or the Dirichlet problem. For example, finding the
steady state temperature within the region IR when no heat sources or sinks are present and
when the temperature is prescribed on the boundary ∂ IR, is a Dirichlet problem. Another
example would be to find the potential inside the region IR when the potential is specified
on the boundary ∂ IR . These two examples correspond to the interior Dirichlet problem.
Similarly, if f ∈ c (0) and is prescribed on the boundary ∂ IR of a finite simply
connected region IR, determining a function ψ ( x, y, z ) which satisfies ∇ 2ψ = 0 outside IR and

is such that ψ = f on ∂ IR, is called an exterior Dirichlet problem. For example, determination
of the distribution of the potential outside a body whose surface potential is prescribed, is an
exterior Dirichlet problem. The second type of BVP is associated with von Neumann. The
problem is to determine the function ψ ( x, y, z ) so that 2ψ 0 within IR while w ψ /w n is
specified at every point of w IR, where w Z /w n denotes the normal derivative of the field
variable ψ . This problem is called the Neumann problem. If ψ is the temperature, w ψ /w n is
the heat flux representing the amount of heat crossing per unit volume per unit time along
the normal direction, which is zero when insulated. The third type of BVP is concerned with
the determination of the function ψ ( x, y, z ) such that 2ψ 0 within IR, while a boundary
condition of the form wψ /w n hψ f , where h t 0 is specified at every point of w IR . This
is called a mixed BVP or Churchill’s problem. If we assume Newton’s law of cooling, the
heat lost is hψ , where ψ is the temperature difference from the surrounding medium and h ! 0 is
a constant depending on the medium. The heat f supplied at a point of the boundary is partly
conducted into the medium and partly lost by radiation to the surroundings. Equating these
amounts, we get the third boundary condition.
2.3 SOME IMPORTANT MATHEMATICAL TOOLS

Among the mathematical tools we employ in deriving many important results, the Gauss
divergence theorem plays a vital role, which can be stated as follows: Let w IR be a closed
surface in the xyz-space and IR denote the bounded region enclosed by w IR in which F is
a vector belonging to c(1) in IR and continuous on IR. Then
³³ F n̂ dS ³³³ F dV (2.16)
w IR IR
where dV is an element of volume, dS is an element of surface area, and n̂ the outward drawn
normal.
Green’s identities which follow from divergence theorem are also useful and they can be
derived as follows: Let F fφ , where f is a vector function of position and φ is a scalar
function of position. Then,
³³³ (fφ ) dV ³³ nˆ fφ dS
IR w IR
Using the vector identity

(fφ ) f φ φ f
we have
³³³ f φ dV ³³ nˆ fφ dS ³³³ φ f dV
IR w IR IR
If we choose f ψ , the above equation yields
³³ φ ψ dV ³³ φ nˆ ψ dS ³³³ φ ψ dV (2.17)
2
IR w IR IR
Noting that nˆ ψ is the derivative of ψ in the direction of nˆ, we introduce the notation
nˆ ψ wψ /w n
into Eq. (2.17) to get
wψ
³³³ φ ψ dV ³³ φ w n dS ³³³ φ ψ dV (2.18a)
2
IR w IR IR
This equation is known as Green’s first identity. Of course, it is assumed that

both φ and ψ possess continuous second derivatives.
Interchanging the role of φ and ψ , we obtain from relation (2.18a) the equation
wφ
³³³ ψ φ dV ³³ ψ w n dS ³³³ ψ φ dV
2
(2.18b)
IR w IR IR
Now, subtracting Eq. (2.18b) from Eq. (2.18a), we get

§ wψ wφ ·
³³³ (φ ψ ψ φ ) dV ³³ ©¨φ w n ψ w n ¹¸ dS (2.19)
2 2
IR w IR
This is known as Green’s second identity. If we set φ ψ in Eq. (2.18a) we get

wφ
³³³ (φ ) ³³ φ w n dS ³³³ φ φ dV (2.20)
2 2
dV
IR w IR IR
which is a special case of Green’s first identity.
2.4 PROPERTIES OF HARMONIC FUNCTIONS

Solutions of Laplace equation are called harmonic functions which possess a number of
interesting properties, and they are presented in the following theorems.
Theorem 2.1 If a harmonic function vanishes everywhere on the boundary, then it is identically
zero everywhere.
Proof If φ is a harmonic function, then 2φ 0 in IR . Also, if φ 0 on w IR, we
shall show that φ 0 in IR IR U w IR . Recalling Green’s first identity, i.e., Eq. (2.20), we get
wφ
³³³ (φ ) ³³ φ w n dS ³³³ φ φ dV
2 2
dV
IR w IR IR
and using the above facts we have, at once, the relation
³³³ (φ )
2
dV 0
IR
Since (φ ) 2 is positive, it follows that the integral will be satisfied only if φ 0. This
implies that φ is a constant in IR. Since φ is continuous in IR and φ is zero on w IR, it

follows that φ 0 in IR .
Theorem 2.2 If φ is a harmonic function in IR and w φ /w n 0 on w IR, then φ is a constant

in IR.
Proof Using Green’s first identity and the data of the theorem, we arrive at
³³³ (φ )
2
dV 0
IR
implying φ 0, i.e., φ is a constant in IR . Since the value of φ is not known on the boundary
w IR while w φ /w n 0, it is implied that φ is a constant on w IR and hence on IR.
Theorem 2.3 If the Dirichlet problem for a bounded region has a solution, then it is unique.
Proof If φ1 and φ2 are two solutions of the interior Dirichlet problem, then
2φ1 0 in IR; φ1 f on w IR
2φ2 0 in IR; φ2 f on w IR
Let ψ φ1 φ2 . Then
2ψ 2φ1 2φ2 0 in IR;
ψ φ1 φ2 f f 0 on w IR
Therefore,
2ψ 0 in IR, ψ 0 on w IR
Now using Theorem 2.1, we obtain ψ 0 on IR, which implies that φ1 φ2 . Hence, the solution
of the Dirichlet problem is unique.
Theorem 2.4 If the Neumann problem for a bounded region has a solution, then it is either
unique or it differs from one another by a constant only.
Proof Let φ1 and φ2 be two distinct solutions of the Neumann problem. Then we have
∂ φ1
∇ 2φ1 = 0 in IR; = f on ∂ IR,
∂n
∂ φ2
∇ 2φ2 = 0 in IR; = f on ∂ IR
∂n
Let ψ = φ1 − φ2 . Then
∇ 2ψ = ∇ 2φ1 − ∇ 2φ2 = 0 in IR
∂ ψ ∂ φ1 ∂ φ2
= − =0 on ∂ IR
∂n ∂n ∂n
Hence from Theorem 2.2, ψ is a constant on IR, i.e., φ1 − φ2 = constant. Therefore, the solution
of the Neumann problem is not unique. Thus, the solutions of a certain Neumann problem
can differ from one another by a constant only.
2.4.1 The Spherical Mean

Let IR be a region bounded by ∂ IR and let P ( x, y , z ) be any point in IR . Also,
let S ( P, r ) represent a sphere with centre at P and radius r such that it lies entirely within
the domain IR as depicted in Fig. 2.2. Let u be a continuous function in IR . Then the
spherical mean of u denoted by u is defined as
1
u (r ) =
4π r 2 ∫∫ u (Q) dS (2.21)
S ( P,r )
∂IR
Q (ξ, η, ζ)
r
P
IR
S (P, r)
O y
z
Fig. 2.2 Spherical mean.
where Q(ξ , η , ζ ) is any variable point on the surface of the sphere S ( P, r ) and dS is the
surface element of integration. For a fixed radius r, the value u (r ) is the average of the values
of u taken over the sphere S ( P, r ), and hence it is called the spherical mean. Taking the
origin at P, in terms of spherical polar coordinates, we have
ξ x r sin θ cos φ
η y r sin θ sin φ
ζ z r cos θ
Then, the spherical mean can be written as

1 2π π
u (r )
4π r 2 ³φ 0 ³θ 0
u ( x r sin θ cos φ , y r sin θ sin φ , z r cos θ ) r 2 sin θ dθ dφ
Also, since u is continuous on S ( P, r ), u too is a continuous function of r on some interval

0 r d R, which can be verified as follows:
1 u (Q ) 2π π
u (r )
4π ³³ u (Q ) sin θ dθ d φ
4π ³0 ³0 sin θ dθ d φ u (Q )
Now, taking the limit as r o 0, Q o P, we have
Lt u (r ) u ( P) (2.22)
r o0
Hence, u is continuous in 0 d r d R.
2.4.2 Mean Value Theorem for Harmonic Functions
Theorem 2.5 Let u be harmonic in a region IR . Also, let P(x, y, z) be a given point in IR
and S(P, r) be a sphere with centre at P such that S(P, r) is completely contained in the domain
of harmonicity of u. Then
1
u ( P) u (r )
4π r 2 ³³ u (Q) dS
S ( P ,r )
Proof Since u is harmonic in IR, its spherical mean u (r ) is continuous in IR and is

given by
1 1 2π π
u (r )
4π r 2 ³³ u (Q) dS
4π r 2 ³0 ³0 u (ξ , η , ζ ) r 2 sin θ dθ dφ
S ( P,r )
Therefore,
du (r ) 1 2π π
dr 4π ³0 ³0 (uξ ξ r uηηr uζ ζ r ) sin θ dθ dφ
1 2π π
4π ³0 ³0 (uξ sin θ cos φ uη sin θ sin φ uζ cos θ ) sin θ dθ dφ (2.23)
Noting that sin θ cos φ , sin θ sin φ and cos θ are the direction cosines of the normal n̂ on
S(P, r),
u iuξ juη kuζ , nˆ (in1 , jn2 , kn3 ),
the expression within the parentheses of the integrand of Eq. (2.23) can be written as u nˆ. Thus
du (r ) 1
dr 4π r 2 ³³ ˆ 2 sin θ dθ dφ
u nr
S ( P ,r )
1
4π r 2 ³³ u nˆ dS
S ( P ,r )
1
4π r 2 ³³³ u dV (by divergence theorem)
V ( P,r )
1
³³³ u dV
2
0 (since u is harmonic)
4π r 2 V ( P,r )
du
Therefore, 0, implying u is constant.
dr
Now the continuity of u at r 0 gives, from Eq.(2.22), the relation

1
u (r ) u ( P)
4π r 2 ³³ u (Q) dS (2.24)
S ( P,r )
2.4.3 Maximum-Minimum Principle and Consequences

Theorem 2.6 Let IR be a region bounded by w IR . Also, let u be a function which is
continuous in a closed region IR and satisfies the Laplace equation 2 u 0 in the interior
of IR . Further, if u is not constant everywhere on IR , then the maximum and minimum values
of u must occur only on the boundary w IR.
Proof Suppose u is a harmonic function but not constant everywhere on IR. If possible,
let u attain its maximum value M at some interior point P in IR . Since M is the maximum
of u which is not a constant, there should exist a sphere S ( P, r ) about P such that some of
the values of u on S ( P, r ) must be less than M. But by the mean value property, the value
of u at P is the average of the values of u on S ( P, r ), and hence it is less than M. This
contradicts the assumption that u M at P. Thus u must be constant over the entire
sphere S ( P, r ).
Let Q be any other point inside IR which can be connected to P by an arc lying entirely
within the domain IR . By covering this arc with spheres and using the Heine-Borel theorem
to choose a finite number of covering spheres and repeating the argument given above, we
can arrive at the conclusion that u will have the same constant value at Q as at P. Thus u
cannot attain a maximum value at any point inside the region IR . Therefore, u can attain its
maximum value only on the boundary w IR . A similar argument will lead to the conclusion
that u can attain its minimum value only on the boundary w IR .
Some important consequences of the maximum-minimum principle are given in the following
theorems.
Theorem 2.7 (Stability theorem). The solutions of the Dirichlet problem depend continuously
on the boundary data.
Proof Let u1 and u2 be two solutions of the Dirichlet problem and let f1 and f2 be given
continuous functions on the boundary w IR such that
2u1 0 in IR; u1 f1 on w IR,
2 u2 0 in IR; u2 f2 on w IR
Let u u1 u2 . Then,
2u 2u1 2 u2 0 in IR; u f1 f 2 on w IR
Hence, u is a solution of the Dirichlet problem with boundary condition u f1 f 2 on w IR . By

the maximum-minimum principle, u attains the maximum and minimum values on w IR . Thus
at any interior point in IR, we shall have, for a given ε ! 0,
ε umin d u d umax ε
Therefore,
|u | ε in IR, implying | u1 u2 | ε
Hence, if
| f1 − f 2 | < ε on ∂ IR, then | u1 − u2 | < ε on IR
Thus, small changes in the initial data bring about an arbitrarily small change in the
solution. This completes the proof of the theorem.
Theorem 2.8 Let {fn} be a sequence of functions, each of which is continuous on IR and
harmonic on IR . If the sequence {fn} converges uniformly on ∂ IR, then it converges uniformly
on IR.
Proof Since the sequence {fn} converges uniformly on ∂ IR, for a given ε > 0, we
can always find an integer N such that
| fn − fm | < ε for n, m > N
Hence, from stability theorem, for all n, m > N , it follows immediately that
| fn − fm | < ε in IR
Therefore, {fn} converges uniformly on IR.
EXAMPLE 2.1 Show that if the two-dimensional Laplace equation ∇ 2 u = 0 is transformed

by introducing plane polar coordinates r , θ defined by the relations x = r cos θ , y = r sin θ , it
takes the form
∂ 2u 1 ∂ u 1 ∂ 2u
+ + =0
∂ r 2 r ∂ r r 2 ∂θ 2
Solution In many practical problems, it is necessary to write the Laplace equation
in various coordinate systems. For instance, if the boundary of the region ∂ IR is a circle,
then it is natural to use polar coordinates defined by x = r cos θ , y = r sin θ . Therefore,
r 2 = x2 + y 2 , θ = tan −1 ( y/x)
sin θ cos θ
rx = cos θ , ry = sin θ , θx = − , θy =
r r
since
⎛ sin θ ⎞
u = u (r , θ ) u x = ur rx + uθ θ x = ⎜ ur cos θ − uθ ⎟
⎝ r ⎠
Similarly,
⎛ cos θ ⎞
u y = ur ry + uθ θ y = ⎜ ur sin θ + uθ ⎟
⎝ r ⎠
Now for the second order derivatives,
§ sin θ · § sin θ · § sin θ ·

u xx (u x ) x (u x )r rx (u x )θ θ x ¨© ur cos θ uθ ¸¹ cos θ ¨© ur cos θ uθ ¸ ¨ ¸
r r r ¹θ © r ¹
Therefore,
§ sin θ sin θ ·
u xx ¨© urr cos θ uθ r uθ 2 ¸ cos θ
r r ¹
§ sin θ cos θ · § sin θ ·

¨ urθ cos θ ur sin θ uθθ uθ ¸ ¨ ¸ (2.25)
© r r ¹© r ¹
Similarly, we can show that
§ cos θ cos θ ·
u yy ¨© urr sin θ urθ uθ 2 ¸ sin θ
r r ¹
§ cos θ sin θ · § cos θ ·

¨ urθ sin θ ur cos θ uθθ uθ ¸¨ ¸ (2.26)
© r r ¹© r ¹
By adding Eqs. (2.25) and (2.26) and equating to zero, we get

1 1
u xx u yy urr ur 2 uθθ 0 (2.27)
r r
which is the Laplace equation in polar coordinates. One can observe that the Laplace equation
in Cartesian coordinates has constant coefficients only, whereas in polar coordinates, it has
variable coefficients.
EXAMPLE 2.2 Show that in cylindrical coordinates r , θ , z defined by the relations x r cos θ ,
y r sin θ , z z , the Laplace equation 2 u 0 takes the form
w 2u 1 w u 1 w 2u w 2u
0
w r2 r w r r 2 wθ 2 w z2
Solution The Laplace equation in Cartesian coordinates is
w 2u w 2u w 2u
2u 0
w x2 w y2 w z2
The relations between Cartesian and cylindrical coordinates give
r2 x2 y 2 , θ tan 1 ( y /x), z z
Since
u u (r , T , z )
È sin T Ø
ur rx uT T x u z z x ur cos T uT É
Ê r ÙÚ
ux
È cos T Ø
ur ry uT T y u z z y ur sin T uT É
Ê r ÙÚ
uy
uz ur rz uT T z u z uz
for the second order derivatives, we find
u xx (u x ) x (u x )r rx (u x )θ θ x (u x ) z z x
ª § sin θ · º ª § sin θ · º § sin θ ·

«ur cos θ uθ ©¨ r ¹¸ » cos θ «ur cos θ uθ ¨© r ¸¹ » ¨©
r ¹
¸
¬ ¼r ¬ ¼θ
§ sin θ sin θ ·
¨© urr cos θ urθ uθ 2 ¸ cos θ
r r ¹
§ sin θ cos θ · § sin θ ·

¨ urθ cos θ ur sin θ uθθ uθ ¸ ¨ ¸ (2.28)
© r r ¹© r ¹
Similarly
u yy (u y ) y (u y ) r ry (u y )θ θ y (u y ) z z y
ª cos θ º ª cos θ º § cos θ ·

«¬ur sin θ uθ r »¼ sin θ «¬ur sin θ uθ r »¼ ¨© r ¸¹
r θ
§ cos θ cos θ ·
¨© urr sin θ uθ r uθ 2 ¸ sin θ
r r ¹
§ cos θ sin θ · § cos θ ·

¨ urθ sin θ ur cos θ uθθ uθ ¸¨ ¸ (2.29)
© r r ¹© r ¹
(2.30)
u zz u zz
Adding Eqs. (2.28)–(2.30), we obtain

1 1
2u urr ur 2 uθθ u zz (2.31)
r r
EXAMPLE 2.3 Show that in spherical polar coordinates r , θ , φ defined by the relations
x r sin θ cos φ , y r sin θ sin φ , z r cos θ , the Laplace equations 2 u 0 takes the form
w § 2 wu · 1 w § wu · 1 w 2u
¨© r ¸¹ ¨© sin θ ¸¹ 0
wr wr sin θ w θ wθ sin 2 θ w φ 2
Solution In Cartesian coordinates, the Laplace equation is
2u u xx u yy u zz 0
In spherical coordinates, u u (r , θ , φ ), r 2 x 2 y 2 z 2 , cos θ z/r , tan φ y /x.

It can be easily verified that
cos θ cos φ cos θ sin φ sin θ
θx , θy , θz
r r r
sin φ cos φ
φx , φy , φz 0
r sin θ r sin θ
Now,
cos θ cos φ sin φ

ux ur rx uθ θ x uφ φ x ur sin θ cos φ uθ uφ
r r sin θ
cos θ sin φ uφ cos φ

uy ur ry uθ θ y uφ φ y ur sin θ sin φ uθ
r r sin θ
§ sin θ ·
uz ur rz uθ θ z uφ φ z ur cos θ uθ ¨ ¸
© r ¹
For the second order derivatives,
u xx (u x )r rx (u x )θ θ x (u x )φ φ x
§ cos θ cos φ sin φ ·

¨© ur sin θ cos φ uθ r
uφ
r sin θ ¸¹ r
(sin θ cos φ )
§ cos θ cos φ sin φ · § cos θ cos φ ·

¨ ur sin θ cos φ uθ uφ ¨ ¸¹
© r r sin θ ¸¹θ © r
§ cos θ cos φ sin φ · § sin φ ·

¨ ur sin θ cos φ uθ uφ ¨© r sin θ ¸¹
© r r sin θ ¸¹φ
cos 2 θ cos 2 φ sin 2 φ

(sin 2 θ cos 2 φ ) urr uθθ uφφ
r2 r 2 sin 2 θ
§ 2 sin θ cos θ cos 2 φ · § 2 sin φ cos φ ·

urθ ¨ ¸ urφ ¨© ¸¹
© r ¹ r
§ 2 cos θ cos φ sin φ · § cos 2 θ cos 2 φ sin 2 φ ·

uθφ ¨
¸ r¨u
© r 2 sin θ ¹ © r r ¹¸
§ sin φ cos φ cos 2 θ cos φ sin φ sin φ cos φ ·

uφ ¨ 2 2 ¸
© r2 r 2 sin 2 θ r sin θ ¹
§ cos θ sin 2 φ 2 cos θ sin θ cos 2 φ · (2.32)

uθ ¨ 2 ¸
© r sin θ r2 ¹
§ cos θ sin φ cos φ ·

u yy (u y )r ry (u y )θ θ y (u y )φ φ y ¨© ur sin θ sin φ uθ uφ (sin θ sin φ )
r r sin θ ¸¹ r
§ cos θ sin φ cos φ · cos θ sin φ

¨ ur sin θ sin φ uθ uφ
© r r sin θ ¸¹θ r
§ cos θ sin φ cos φ · cos φ

¨ ur sin θ sin φ uθ uφ
© r r sin θ ¸¹φ r sin θ
cos 2 θ sin 2 φ cos 2 φ

(sin 2 θ sin 2 φ ) urr uθθ uφφ
r2 r 2 sin 2 θ
§ 2 sin θ cos θ sin 2 φ · § 2 cos φ sin φ ·
urθ ¨ ¸ urφ ¨© ¸¹
© r ¹ r
§ 2 cos θ cos φ sin φ · § cos 2 θ sin 2 φ cos 2 φ ·

uθφ ¨ u ¨
¸ r ¸¹
r
© r 2 sin θ ¹ © r
§ 2 sin θ cos θ sin 2 φ cos θ cos 2 φ ·

uθ ¨ ¸
© r2 r 2 sin θ ¹
§ sin φ cos φ sin φ cos φ cos 2 θ sin φ cos φ ·

uφ ¨ 2 2 ¸
© r2 r sin θ r 2 sin 2 θ ¹
(2.33)
Similarly,
u zz (u z )r rz (u z )θ θ z (u z )φ φ z
§ sin θ · § sin θ · § sin θ ·

¨© ur cos θ uθ ¸¹ (cos θ ) ¨© ur cos θ uθ ¸¹ ¨© ¸
r r r θ r ¹
2 sin θ cos θ sin 2 θ

urr cos 2 θ urθ uθθ
r r2
sin 2 θ cos θ sin θ

ur uθ (2.34)
r r2
Adding Eqs. (2.32)–(2.34), we obtain

1 1 2 cos θ
2u urr u
2 θθ
uφφ ur 2 uθ 0
r r sin θ
2 2 r r sin θ
w § 2 wu · 1 w § wu · 1 w 2u
2u ¨© r ¸ ¨ sin θ ¸ 0 (2.35)
wr w r ¹ sin θ w θ © w θ ¹ sin 2 θ w φ 2
2.5 SEPARATION OF VARIABLES

The method of separation of variables is applicable to a large number of classical linear
homogeneous equations. The choice of the coordinate system in general depends on the shape
of the boundary. For example, consider a two-dimensional Laplace equation in Cartesian
coordinates.
2u u xx u yy 0 (2.36)
We assume the solution in the form
u ( x, y ) X ( x) Y ( y ) (2.37)
Substituting in Eq. (2.36), we get
X ccY Y ccX 0
i.e.
X cc Y cc
k
X Y
where k is a separation constant. Three cases arise.
Case I Let k p 2 , p is real. Then
d2X d 2Y
p2 X 0 and p 2Y 0
dx 2 dy 2
whose solution is given by
X c1e px c2 e px
and
Y c3 cos py c4 sin py
Thus, the solution is
u ( x, y ) (c1e px c2 e px ) (c3 cos py c4 sin py ) (2.38)
Case II Let k 0. Then

d2X d 2Y
0 and 0
dx 2 dy 2
Integrating twice, we get
X c5 x c6
and
Y c7 y c8
The solution is therefore,
u ( x, y ) (c5 x c6 ) (c7 y c8 ) (2.39)
Case III Let k p 2 . Proceeding as in Case I, we obtain
X c9 cos px c10 sin px
Y c11e py c12 e py
Hence, the solution in this case is
u ( x, y ) (c9 cos px c10 sin px) (c11e py c12 e py ) (2.40)
In all these cases, ci (i 1, 2, } , 12) refer to integration constants, which are calculated by
using the boundary conditions.
2.6 DIRICHLET PROBLEM FOR A RECTANGLE

The Dirichlet problem for a rectangle is defined as follows:
PDE: 2u 0, 0 d x d a, 0 d y d b
BCs: u ( x, b) u (a, y ) 0, u (0, y ) 0, u ( x, 0) f ( x) (2.41)
This is an interior Dirichlet problem. The general solution of the governing PDE, using the
method of variables separable, is discussed in Section 2.5. The various possible solutions of
the Laplace equation are given by Eqs. (2.38–2.40). Of these three solutions, we have to
choose that solution which is consistent with the physical nature of the problem and the given
boundary conditions as depicted in Fig. 2.3.
y
y=b
u=0
x=0 x=a
u=0 u=0
u = f (x)
x
O y=0
Fig. 2.3 Dirichlet boundary conditions.
Consider the solution given by Eq. (2.38):
u ( x, y ) (c1e px c2 e px ) (c3 cos py c4 sin py )
Using the boundary condition: u (0, y ) 0, we get
(c1 c2 ) (c3 cos py c4 sin py ) 0
which means that either c1 c2 0 or c3 cos py c4 sin py 0. But c3 cos py c4 sin py z 0;

therefore,
c1 c2 0 (2.42)
Again, using the BC; u (a, y ) 0, Eq. (2.38) gives
(c1eap c2 e ap ) (c3 cos py c4 sin py ) 0

implying thereby
c1eap c2 e ap 0 (2.43)
To determine the constants c1, c2, we have to solve Eqs. (2.42) and (2.43); being homogeneous,
the determinant
1 1
0
eap e ap
for the existence of non-trivial solution, which is not the case. Hence, only the trivial solution
u ( x, y ) 0 is possible.
If we consider the solution given by Eq. (2.39) u ( x, y ) (c5 x c6 ) (c7 y c8 ), the boundary
conditions: u (0, y ) u (a, y ) 0 again yield a trivial solution. Hence, the possible solutions
given by Eqs. (2.38) and (2.39) are ruled out. Therefore, the only possible solution obtained
from Eq. (2.40) is
u ( x, y ) (c9 cos px c10 sin px) (c11e py c12 e py )
Using the BC: u (0, y ) 0, we get c9 0. Also, the other BC: u (a, y ) 0 yields
c10 sin pa (c11e py c12 e py ) 0
For non-trivial solution, c10 cannot be zero, implying sin pa 0, which is possible if pa nπ or
p nπ /a, n 1, 2, 3, } Therefore, the possible non-trivial solution after using the superposition
principle is
f
nπ x
u ( x, y ) ¦ sin a
[an exp (nπ y/a) bn exp ( nπ y/a)] (2.44)
n 1
Now, using the BC: u ( x, b) 0, we get
nπ x
sin [ an exp (nπ b/a ) bn exp ( nπ b /a )] 0
a
implying thereby
an exp (nπ b /a ) bn exp ( nπ b/a ) 0
which gives
exp (nπ b /a)
bn an , n 1, 2, } , f
exp ( nπ b/a)
The solution (2.44) now becomes
f
2an sin (nπ x/a ) ª exp {nπ ( y b)/a} exp { nπ ( y b)/a} º
u ( x, y ) ¦ exp ( nπ b /a) «¬ 2 »¼
n 1
f
2a
¦ exp (nnπ b/a) sin (nπ x/a)sin h {nπ ( y b)/a}
n 1
Let 2an /[exp ( nπ b /a )] An . Then the solution can be written in the form
f
u ( x, y ) ¦ An sin (nπ x/a) sinh {nπ ( y b)/a} (2.45)
n 1
Finally, using the non-homogeneous boundary condition: u ( x, 0) f ( x), we get

f
¦ An sin (nπ x/a) sinh (nπ b/a) f ( x)
n 1
which is a half-range Fourier series. Therefore,

2 a
An sinh ( nπ b/a )
a ³0 f ( x) sin (nπ x /a ) dx (2.46)
Thus, the required solution for the given Dirichlet problem is

f
u ( x, y ) ¦ An sin (nπ x/a) sinh {nπ ( y b)/a} (2.47)
n 1
where
2 1 a
An
a sinh ( nπ b /a) ³0 f ( x) sin (nπ x /a ) dx
2.7 THE NEUMANN PROBLEM FOR A RECTANGLE

The Neumann problem for a rectangle is defined as follows:
PDE: 2 u 0, 0 d x d a, 0 d y d b
BCs: u x (0, y ) u x ( a, y ) 0, u y ( x, 0) 0, u y ( x, b ) f ( x) (2.48)
The general solution of the Laplace equation using the method of variables separable is
given in Section 2.5, and is found to be
u ( x, y ) (c1 cos px c2 sin px) (c3e py c4 e py )
The BC: u x (0, y ) 0 gives
0 c2 p (c3e py c4 e py )
implying c2 0. Therefore,
u ( x, y ) c1 cos px (c3e py c4 e py ) (2.49)

The BC: u x ( a, y ) 0 gives
0 c1 p sin pa (c3e py c4 e py )
For non-trivial solution, c1 z 0, implying

nπ
sin pa 0, pa nπ , p (n 0, 1, 2, })
a
Thus the possible solution is
nπ x
u ( x, y ) cos ( Aenπ y /a Be nπ y /a ) (2.50)
a
Now, using the BC: u y ( x, 0) 0, we get
nπ x § nπ nπ ·
0 cos ¨© A B ¸
a a a ¹
implying B A. Thus, the solution is
nπ x
u ( x, y ) A cos [exp (nπ y/a) exp ( nπ y /a)]
a
nπ x nπ y
2 A cos cosh
a a
Using the superposition principle and defining 2 A An , we get
f
nπ x nπ y
u ¦ An cos a
cosh
a
(2.51)
n 0
Finally, using the BC: u y ( x, b) f ( x), we get

f
nπ x nπ nπ b
f ( x) ¦ An cos a a
sinh
a
n 1
which is the half-range Fourier cosine series. Therefore,

nπ nπ b 2 a nπ x
a
sinh
An
a a ³0 f ( x) cos
a
dx
Hence, the required solution is

f
nπ x nπ y
u A0 ¦ An cos a
cosh
a
(2.52)
n 1
where
2 1 a nπ x
An
nπ sinh (nπ b /a ) ³0 f ( x) cos
a
dx
2.8 INTERIOR DIRICHLET PROBLEM FOR A CIRCLE

The Dirichlet problem for the circle is defined as follows:
PDE: 2u 0, 0 d r d a, 0 d θ d 2π
BC: u (a, θ ) f (θ ), 0 d θ d 2π (2.53)
where f (θ ) is a continuous function on w IR . The task is to find the value of u at any point
in the interior of the circle IR in terms of its values on w IR such that u is single valued and
continuous on IR.
In view of circular geometry, it is natural to choose polar coordinates to solve this
problem and then use the variables separable method. The requirement of single-valuedness
of u in IR implies the periodicity condition, i.e.,
u (r , θ 2π ) u (r , θ ), 0 d r d a, (2.54)
From Eq. (2.27), 2 u 0 which in polar coordinates can be written as

1 1
urr ur 2 uθθ 0
r r
If u (r , θ ) R(r ) H (θ ), the above equation reduces to
1 1
R ccH R cH 2 RH cc 0
r r
This equation can be rewritten as
r 2 R cc rR c H cc
k (2.55)
R H
which means that a function of r is equal to a function of θ and, therefore, each must be
equal to a constant k (a separation constant).
Case I Let k λ 2 . Then

r 2 R cc rR c λ 2 R 0 (2.56)
which is a Euler type of equation and can be solved by setting r e z . Its solution is
R c1eλ z c2 e λ z c1r λ c2 r λ
Also,
H cc λ 2 H 0
whose solution is
H c3 cos λθ c4 sin λθ
Therefore,
u (r , θ ) (c1r λ c2 r λ ) (c3 cos λθ c4 sin λθ ) (2.57)
Case II Let k λ 2 . Then

r 2 R cc rR c λ 2 R 0, H cc λ 2 H 0
Their respective solutions are
R c1 cos (λ ln r ) c2 sin (λ ln r )
H c3eλθ c4 e λθ
Thus
u (r , θ ) [c1 cos (λ ln r ) c2 sin (λ ln r )] (c3eλθ c4 e λθ ) (2.58)
Case III Let k 0. Then we have

rR cc R c 0
Setting R c (r ) V (r ), we obtain
dV dV dr
r V 0, i.e., 0
dr V r
Integrating, we get ln Vr ln c1. Therefore,

c1 dR
V
r dr
On integration,
R c1 ln r c2
Also,
H cc 0
After integrating twice, we get
H c3θ c4
Thus,
u (r , θ ) (c1 ln r c2 ) (c3θ c4 ) (2.59)
Now, for the interior problem, r 0 is a point in the domain IR and since ln r is not defined
at r 0, the solutions (2.58) and (2.59) are not acceptable. Thus the required solution is
obtained from Eq. (2.57). The periodicity condition in θ implies
c3 cos λθ c4 sin λθ c3 cos (λ (θ 2π )) c4 sin (λ (θ 2π ))
i.e.
c3 [cos λθ cos (λθ 2λπ )] c4 [sin λθ sin (λθ 2λπ )] 0
or
2 sin λπ [c3 sin (λθ λπ ) c4 cos (λθ λπ )] 0
implying sin λπ 0, λπ nπ , λ n (n 0, 1, 2, }). Using the principle of superposition and

renaming the constants, the acceptable general solution can be written as
f
u (r , θ ) ¦ (cn r n dn r n ) (an cos nθ bn sin nθ ) (2.60)
n 0
At r = 0, the solution should be finite, which requires d n 0. Thus the appropriate solution
assumes the form
f
u (r , θ ) ¦ r n ( An cos nθ Bn sin nθ )
n 0
For n 0, let the constant A0 be A0/2. Then the solution is
A0 f n
u (r , θ )
2 n 1 ¦
r ( An cos nθ Bn sin nθ ) (2.61)
which is a full-range Fourier series. Now we have to determine An and Bn so that the BC:
u (a, θ ) f (θ ) is satisfied, i.e.,
f
f (θ ) ¦ an ( An cos nθ Bn sin nθ )
n 0
Hence,
1 2π
A0
π ³0 f (θ ) dθ
1 2π
a n An
π ³0 f (θ ) cos nθ dθ (2.62)
1 2π
a n Bn
π ³0 f (θ ) sin nθ dθ , n 1, 2, 3, }
In Eqs. (2.62) we replace the dummy variable θ by φ to distinguish this variable from the
current variable θ in Eq. (2.61). Substituting Eq. (2.62) into Eq. (2.61), we obtain the relation
2π f ª r n cos nθ 2π
1
u (r , θ )
2π ³0 f (φ ) dφ ¦ « n
π ³0 cos (nφ ) f (φ ) dφ
n 1«
¬a
r n sin nθ 2π º

an π ³0 sin (nφ ) f (φ ) dφ »
»¼
Interchanging the order of summation and integration, we get
2π 2π f n
1 1 §r·
u (r , θ )
2π ³0 f (φ ) dφ
π ³0 f (φ ) ¦ ¨© ¸¹ {cos nφ cos nθ sin nφ sin nθ } dφ
a
n 1
1 2π ª1 f
§r·
n º
π ³0 f (φ ) « ¦¨ ¸
«¬ 2 n 1 © a ¹
cos n (φ θ ) » dφ
»¼
(2.63)
To obtain an alternative expression for u (r , θ ) in closed integral form, we can proceed as

follows:
Let
f n
§r·
c ¦ ¨© a ¸¹ cos n (φ θ )
n 1
f n
§r·
s ¦ ¨© ¸¹ sin n (φ θ )
a
n 1
so that
f n
ª r i (φ θ ) º
c is ¦ «a e »¼
n 1¬
Since r a, (r/a ) 1 and | ei (φ θ ) | d 1,
f
(r/a ) ei (φ θ )
n
ª § r · i (φ θ ) º
c is ¦ «¨ ¸ e
¬© a ¹
»
¼ [1 (r /a) ei (φ θ ) ]
n 1
(r/a ){ei (φ θ ) (r/a )}

[1 (r/a )ei (φ θ ) ][1 (r/a ) ei (φ θ ) ]
Equating the real part on both sides, we get
[r/a ) cos (φ θ ) (r 2 /a 2 )]
c
[1 (2r/a ) cos (φ θ ) (r 2 /a 2 )]
Thus, the expression in the square brackets of Eq. (2.63) becomes
1 [(r/a ) cos (φ θ ) (r 2 /a 2 )] a2 r 2

2 [1 (2r/a ) cos (φ θ ) ( r 2 /a 2 )] 2[a 2 2ar cos (φ θ ) r 2 ]
Thus, the required solution takes the form
1 2π (a 2 r 2 ) f (φ )
u (r , θ )
2π ³0 [a 2 2ar cos (φ θ ) r 2 ]
dφ (2.64)
This is known as Poisson’s integral formula for a circle, which gives a unique solution for
the Dirichlet problem. The solution (2.64) can be interpreted physically in many ways: It can
be thought of as finding the potential u (r , θ ) as a weighted average of the boundary
potentials f (φ ) weighted by the Poisson kernel P, given by
a2 r 2
P
[a 2 2ar cos (φ θ ) r 2 ]
It can also be thought of as a steady temperature distribution u (r , θ ) in a circular disc, when
the temperature u on its boundary w IR is given by u f (φ ) which is independent of time.
2.9 EXTERIOR DIRICHLET PROBLEM FOR A CIRCLE

The exterior Dirichlet problem is described by
PDE: 2u 0
BC: u (a, θ ) f (θ ) (2.65)
u must be bounded as r o f.
By the method of separation of variables, the general solution (2.60) of 2 u 0 in polar
coordinates can be written as
f
u (r , θ ) ¦ (cn r n dn r n ) (an cos nθ bn sin nθ )
n 0
Now as, r o f, we require u to be bounded, and, therefore, cn 0.

After adjusting the constants, the general solution now reads

∞
u (r , θ ) = ∑ r −n ( An cos nθ + Bn sin nθ )
n=0
With no loss of generality, it can also be written as
A0 ∞ − n
u (r , θ ) = +
2 n =1 ∑
r ( An cos nθ + Bn sin nθ ) (2.66)
Using the BC: u (a, θ ) = f (θ ), we obtain
A0 ∞ − n
f (θ ) = +
2 n =1 ∑
a ( An cos nθ + Bn sin nθ )
This is a full-range Fourier series in f (θ ), where
1 2π
π ∫0
A0 = f (θ ) dθ
1 2π
π ∫0
a − n An = f (θ ) cos nθ dθ (2.67)
1 2π
π ∫0
a − n Bn = f (θ ) sin nθ dθ
In Eq. (2.67) we replace the dummy variable θ by φ so as to distinguish it from the current
variable θ . We then introduce the changed variable into solution (2.66) which becomes
2π ∞ ⎡ r −n an 2π
1
u (r , θ ) =
2π ∫0 f (φ ) dφ + ∑ ⎢⎢⎣ π
cos nθ ∫0 cos (nφ ) f (φ ) dφ
n =1
r −n an 2π ⎤
+
π
sin nθ ∫0 sin (nφ ) f (φ ) dφ ⎥
⎥⎦
or
1 2π ⎡ 1 ∞ ⎛ a ⎞n ⎤
u (r , θ ) =
π ∫0 f (φ ) ⎢ +
⎢⎣ 2 n = 1 r
∑
⎜⎝ ⎟⎠ cos n (φ − θ )⎥ dφ
⎥⎦
(2.68)
Let
• n
Ê aˆ
C= Â ÁË r ˜¯ cos n (f - q )
n =1
n
È aØ
S Ç ÉÊ ÙÚ sin n (I T )
r
n 1
Then,
f n
ª § a · i (φ θ ) º
C iS ¦ « ¨© r ¸¹ e »
n 1¬ ¼
a
Since 1, | ei (φ θ ) | d 1. We have
r
a ei (φ θ ) ( a/r ) [ei φ θ ) ( a/r )]

C iS
r [1 (a/r ) ei (φ θ ) ] [1 ( a/r ) ei (φ θ ) ] [1 (a/r ) e i (φ θ ) ]
Hence,
[( a / r ) cos (φ θ ) (a 2 /r 2 )]
C
[1 (2a / r ) cos (φ θ ) (a 2/ r 2 )]
Thus the quantity in the square brackets on the right-hand side of Eq. (2.68) becomes
1 [(a / r ) cos (φ θ ) (a 2/ r 2 )] r 2 a2

2 [1 (2a/r ) cos (φ θ ) ( a 2/ r 2 )] 2[r 2 2ar cos (φ θ ) a 2 ]
Therefore, the solution of the exterior Dirichlet problem reduces to that of an integral equation
of the form
1 2π (r 2 a 2 ) f (φ )
u (r , θ )
2π ³0 [r 2 2ar cos (φ θ ) a 2 ]
dφ (2.69)
EXAMPLE 2.4 Find the steady state temperature distribution in a semi-circular plate of
radius a, insulated on both the faces with its curved boundary kept at a constant temperature
U0 and its bounding diameter kept at zero temperature as described in Fig. 2.4.
R = Q/2
u = u0
r=a
r
R
R=Q u=0 R=0
Fig. 2.4 Semi-circular plate.
Solution The governing heat flow equation is
ut = ∇ 2 u
In the steady state, the temperature is independent of time; hence ut = 0, and the temperature
satisfies the Laplace equation. The problem can now be stated as follows: To solve
1 1
PDE: ∇ 2 u (r , θ ) = urr + ur + 2 uθθ = 0
r r
BCs: u (a, θ ) = U 0 , u (r , 0) = 0, u (r , π ) = 0
the acceptable general solution is
u (r , θ ) = (cr λ + Dr − λ ) ( A cos λθ + B sin λθ ) (2.70)
From the BC: u (r , 0) = 0, we get A = 0; however, the BC: u (r , π ) = 0 also gives
B sin λπ (cr λ + Dr − λ ) = 0
implying either B = 0 or sin λπ = 0. B = 0 gives a trivial solution. For a non-trivial solution,

we must have sin λπ = 0, implying
λπ = nπ , n = 1, 2, …
meaning thereby λ = n. Hence, the possible solution is
u (r , θ ) = B sin nθ (Cr λ + Dr − λ ) (2.71)
In Eq. (2.71), we observe that as r → 0, the term r − λ → ∞. But the solution should be finite
at r = 0, and so D = 0. Then after adjusting the constants, it follows from the superposition
principle that,
∞
u (r , θ ) = ∑ Bn r n sin nθ
n =1
Finally, using the first BC: u ( a, θ ) = U 0 , we get
∞
u ( a, θ ) = U 0 = ∑ Bn an sin nθ
n =1
which is a half-range Fourier sine series. Therefore,
4U 0
2 π ° nπ , for n 1, 3, }
³0 U 0 sin nθ dθ
n
Bn a ®
π ° 0,
¯ for n 2, 4, }
Hence,
4U 0
Bn , n 1, 3, }
nπ a n
With these values of Bn, the required solution is
f n
4U 0 1 §r·
u (r , θ )
π ¦ n
¨© ¸¹ sin nθ
a
n odd
2.10 INTERIOR NEUMANN PROBLEM FOR A CIRCLE

The interior Neumann problem for a circle is described by
PDE: 2u 0, 0 d r a; 0 d θ d 2π (2.72)
wu w u ( a, θ )
BC: g (θ ), r a
wn wr
Following the method of separation of variables, the general solution (2.60) of equation
2u 0 in polar coordinates is given by
f
u (r , θ ) ¦ (cn r n dn r n ) (an cos nθ bn sin nθ )
n 0
At r = 0, the solution should be finite and, therefore, d n 0. Hence, after adjusting the constants,
the general solution becomes
f
u (r , θ ) ¦ r n ( An cos nθ Bn sin nθ )
n 0
With no loss of generality, this equation can be written as

f
A0
u (r , θ )
2 n
¦ r n ( An cos nθ Bn sin nθ ) (2.73)
1
f
wu
wr ¦ nr n1 ( An cos nθ Bn sin nθ )
n 1
Using the BC:

wu
( a, θ ) g (θ )
wr
we get
f
g (θ ) ¦ nan1 ( An cos nθ Bn sin nθ ) (2.74)
n 1
which is a full-range Fourier series in g (θ ), where
1 2π
na n 1 An
π ³0 g (θ ) cos nθ dθ
1 π
na n 1Bn
π ³0 g (θ ) sin nθ dθ (2.75)
Here, we replace the dummy variable θ by φ to distinguish from the current variable θ in
Eq. (2.74). Now introducing Eq. (2.75) into Eq. (2.73), we obtain
A0 f rn 2π
u (r , θ ) ¦
2 n 1 nπ a n 1 ³0 g (φ ) (cos nφ cos nθ sin nφ sin nθ ) dφ
or
2π f n
A0 §r· a
u (r , θ )
2
³0 g (φ ) ¦ ¨© ¸¹
a nπ
cos n (φ θ ) dφ (2.76)
n 1
This solution can also be expressed in an alternative integral form as follows: Let
n
§r· a
C ¦ ¨© a ¸¹ nπ
cos n (φ θ )
n
§r· a
S ¦ ¨© ¸¹
a nπ
sin n (φ θ )
Therefore,
n f n
§r· a in (φ θ ) a ªr º 1
C iS ¦ ¨© a ¸¹ nπ
e
π ¦ «¬ a ei (φ θ ) »¼n
n 1
ª r i (φ θ ) ½ r i (φ θ ) ½2 r i (φ θ ) ½3 º
«® e ¾ ® e ¾ ® e ¾ »
a «¯a ¿ ¯a ¿ ¯a ¿
»
π « 1 2 3 »
« »
¬« ¼»
or
a ⎡ r ⎤ a ⎡ r r ⎤
C + iS = − ln ⎢1 − ei (φ −θ ) ⎥ = − ln ⎢1 − cos (φ − θ ) − i sin (φ − θ ) ⎥ (2.77)
π ⎣ a ⎦ π ⎣ a a ⎦
To get the real part of ln z, we may note that
W = ln z or z = eW
i.e., x + iy = eu +iv = eu cos v + ieu sin v. Therefore,
x = eu cos v, y = eu sin v
e2u = x 2 + y 2 = | z |2
i.e., u = ln | z | . Therefore,
2 2
⎛ r
a ⎞ ⎛r ⎞
C = − ln ⎜1 − cos (φ − θ ) ⎟ + ⎜ sin (φ − θ ) ⎟
π ⎝ a ⎠ ⎝ a ⎠
a a 2 − 2ar cos (φ − θ ) + r 2
=− ln
π a2
Thus the required solution is
A0 a 2π a 2 − 2ar cos (φ − θ ) + r 2
u (r , θ ) = −
2 π ∫0 ln
a2
g (φ ) d φ (2.78)
which is again an integral equation.
2.11 SOLUTION OF LAPLACE EQUATION IN CYLINDRICAL COORDINATES

The Laplace equation in cylindrical coordinates assumes the following form:
1 1
∇ 2u = urr + ur + 2 uθθ + u zz = 0 (2.79)
r r
We now seek a separable solution of the form
u (r , θ , z ) = F (r , θ ) Z ( z ) (2.80)
Substituting Eq. (2.80) into Eq. (2.79), we get
∂2F 1 ∂F 1 ∂2F d 2Z
Z + Z + Z + F =0
∂ r2 r ∂r r2 ∂ θ2 dz 2
or
⎛ ∂ 2F 1 ∂ F 1 ∂ 2F ⎞ 1 d 2Z 1
⎜ 2 + r ∂r + 2 ⎟ = − = k (say)
⎝ ∂r r ∂θ 2 ⎠ F dz 2 Z
where k is a separation constant. Therefore, either
d 2Z
kZ 0 (2.81)
dz 2
or
w 2F 1 w F 1 w 2F
KF 0 (2.82)
w r2 r w r r 2 wθ 2
If k is real and positive, the solution of Eq. (2.81) is
Z c1 cos k z c2 sin kz
If k is negative, the solution of Eq. (2.81) is
Z c1e kz
c2 e kz
If k is equal to zero, the solution of Eq. (2.81) is

Z c1 z c2
From physical considerations, one would expect a solution which decays with increasing z
and, therefore, the solution corresponding to negative k is acceptable. Let k λ 2 . Then
Z c1eλ z c2 e λ z (2.83)
Equation (2.82) now becomes
w 2F 1 w F 1 w 2F
λ2F 0
Let F (r , θ ) f (r ) H (θ ). Substituting into the above equations, we get
1 1
f ccH f cH 2 f H cc λ 2 f H 0
r r
or
1 H cc
(r 2 f cc r f c λ 2 r 2 f ) k c (say)
f H
From physical consideration, we expect the solution to be periodic in θ , which can be obtained
when k c is positive and k c n 2 . Therefore, the acceptable solution will be
H c3 cos nθ c4 sin nθ (2.84)
When k c n2 , we will also have
d2 f df
r2 2
r (λ 2 r 2 n 2 ) f 0 (2.85)
dr dr
which is a Bessel’s equation whose general solution is given by

f AJ n (λ r ) BYn (λ r ) (2.86)
Here, J n (λ r ) and Yn (λ r ) are the nth order Bessel functions of first and second kind, respectively.
Since Yn (λ r ) o f as r o 0, Yn (λ r ) becomes unbounded at r 0. Continuity of the solution
demands B 0. Hence the most general and acceptable solution of 2 u 0 is
u (r , T , z ) J n (O r ) (c1eO z c2 e O z ) (c3 cos nT c4 sin nT ) (2.87)
EXAMPLE 2.5 A homogeneous thermally conducting cylinder occupies the region 0 d r d a,

0 d θ d 2π , 0 d z d h, where r , θ , z are cylindrical coordinates. The top z h and the lateral
surface r a are held at 0°, while the base z 0 is held at 100°. Assuming that there are no
sources of heat generation within the cylinder, find the steady-temperature distribution within
the cylinder.
Solution The temperature u must be a single valued continuous function. The steady
state temperature satisfies the Laplace equation inside the cylinder. To compute the temperature
distribution inside the cylinder, we have to solve the following BVP:
PDE: 2u 0
BCs: u 0q on z h,
u 0q on r a,
u 100q on z 0
The general solution of the Laplace equation in cylindrical coordinates as given in Section 2.11
is
r (r , θ , z ) J n (λ r ) (c1 cos nθ c2 sin nθ ) (c3eλ z c4 e λ z )
Since the face z 0 is maintained at 100° and since the other face and lateral surface of the
cylinder are maintained at 0°, the temperature at any point inside the cylinder is obviously
independent of θ . This is possible only when n 0 in the general solution. Thus,
u (r , z ) J 0 (λ r ) ( Aeλ z Be λ z )
Using the BC: u 0 on z h, we get
0 J 0 (λ r ) ( Aeλ h Be λ h )
implying thereby Aeλ h Be λ h 0, from which
Aeλ h
B
e λ h
Therefore, the solution is

J 0 (λ r ) A
u (r , z ) λh
[e λ ( z h ) e λ ( z h ) ]
e
or
u (r , z ) J 0 (λ r ) A1 sinh λ ( z h)
where A1 2 A / e λ h . Now using the BC: u 0 on r a, we have
0 A1 J 0 (λ a ) sinh λ ( z h)
implying J 0 (λ a ) 0, which has infinitely many positive roots. Denoting them by ξ n , we

have ξ n λ a, and therefore,
ξn
λ
a
Thus the solution is
§ξ r · ªξ º
u (r , z ) A1 J 0 ¨ n ¸ sinh « n ( z h) » , n 1, 2, }
© a ¹ ¬ a ¼
Using the principle of superposition, we have
f
§ ξn r · ªξ º
u (r , z ) ¦ An J0 ¨© a ¹ ¸ sinh « n ( z h)»
¬a ¼
n 1
The BC: u = 100° on z = 0 gives

f
§ ξn h · § ξn r ·
100 ¦ An sinh ¨© a ¸¹ ¨© a ¸¹
J0
n 1
which is a Fourier-Bessel series. Multiplying both sides with rJ 0 (ξ m r/a ) and integrating, we
get
f
a §ξ r · § ξn h · a § ξn r · § ξm r ·
100 ³0 rJ 0 ¨ m ¸ dr
© a ¹ ¦ An sinh ¨© a ¸¹ ³0 rJ0 ¨© J0
a ¸¹ ¨© a ¸¹
dr
n 1
Using the orthogonality property of Bessel’s function, namely,
0, if i z j
a °
³0 xJ n (αi x) J n (α j x) dx ® a 2
° J 2 (αi ), if i j
¯ 2 n 1
where α i , α j are the zeros at J n ( x ) 0, we have
f
a §ξ r · § ξn h · a2 2
100 ³0 rJ 0 ¨ n ¸ dr
© a ¹ ¦ An sinh ¨© a ¸¹ 2 1
J (ξ n )
n 1
Therefore,
200 a § ξnr ·
An
§ ξ h· ³0 rJ 0 ©¨ a ¹
¸ dr
a 2 sinh ¨ n ¸ J 12 (ξ n )
© a ¹
Setting
ξn r a
x, dr dx
a ξn
the relation for An can also be written as
200 ξn
An
§ ξ h· ³0 xJ 0 ( x) dx
ξ n2 sinh ¨ n ¸ J 12 (ξ n )
© a ¹
Using the recurrence relation
d n
x n J n 1 ( x) [ x J n ( x)],
dx
For n 1, we get
³ xJ 0 ( x) dx xJ1 ( x)
Now, An can be written as

ξ
ª 200 x J1( x) º n 200
An « »
« ξ n2 sinh (ξ n h/a) J 12 (ξ n ) » ξ n sinh (ξ n h /a ) J1 (ξ n )
¬ ¼0
Hence, the required temperature distribution inside the cylinder is
f
J 0 (ξ n r/a) sinh [(ξ n /a) ( z h)]
u (r , z ) 200 ¦ ξ n sinh (ξ n h /a) J1 (ξ n )
n 1
where ξ n are the positive zeros of J 0 (ξ ).

EXAMPLE 2.6 Find the potential u inside the cylinder 0 d r d a, 0 d θ d 2π , 0 d z d h, if

the potential on the top z h, and on the lateral surface r = a is held at zero, while on the
base z = 0, the potential is given by u (r , θ , 0) V0 (1 r 2/ a 2 ), where V0 is a constant; r , θ , z are
cylindrical polar coordinates.
Solution The potential u must be a single-valued continuous function and satisfy the
Laplace equation inside the cylinder. To compute the potential inside the cylinder, we have
to solve the following BVP:
PDE: 2u 0
BCs: u 0 on z h,
u 0 on r a,
§ r2 ·
u V0 ¨1 2 ¸ on z 0
© a ¹
In cylindrical coordinates, the general solution of the Laplace equation as given in Section 2.11
is
u (r , θ , z ) J n (λ r ) (c1 cos nθ c2 sin nθ ) (c3eλ z c4 e λ z )
Since the face z = 0 has potential V0 (1 r 2/ a 2 ), which is purely a function of r and is independent
of θ and since the other faces of the cylinder are at zero potential, the potential at any point
inside the cylinder will obviously be independent of θ . This is possible only when n = 0 in
the general solution. Thus,
u (r , z ) J 0 (λ r ) ( Aeλ z Be λ z )
Using the BC: u = 0 on z = h, we obtain
0 J 0 (λ r ) ( Aeλ h Be λ h )
implying Aeλ h Be λ h 0, which yields
Aeλ h
B
e λ h
Hence, the solution is
A
u (r , z ) λh
J 0 (λ r )[eλ ( z h) e λ ( z h) ]
e
or
u (r , z ) A1 J 0 (λ r ) sinh λ ( z h)
where A1 A / e λ h . Now, using the BC: u = 0 on the lateral surface, i.e., on r = a, we get
0 A1 J 0 (λ a ) sinh λ ( z h)
implying J 0 (λ a ) 0. This has infinitely many positive roots; denoting them by ξ n we shall
have
ξn λa or λ ξ n /a
The solution now takes the form
§ξ r · ªξ º
u (r , z ) A1 J 0 ¨ n ¸ sinh « n ( z h) » , n 1, 2, }
© a ¹ ¬a ¼
The principle of superposition gives
f
§ ξn r · ªξ º
u (r , z ) ¦ An J 0 ¨© a ¹ ¸ sinh « n ( z h) »
¬a ¼
n 1
§ r2 ·
The last BC: u V0 ¨1 2 ¸ on z 0 yields
© a ¹
§ r2 · f
§ ξn h · § ξn r ·
V0 ¨1 2 ¸
© a ¹
¦ An sinh ¨© J0
a ¸¹ ¨© a ¸¹
n 1
This is a Fourier-Bessel’s series. Multiplying both sides by rJ 0 (ξ m r/a ) and integrating, we get
§ r2 · f
a § ξm r · § ξn h · a § ξn r · § ξm r ·
V0 ³0 ¨1 2 ¸ rJ 0 ¨© a ¸¹ dr
© a ¹
¦ An sinh ¨© a ¸¹ ³0 rJ0 ¨© a ¸¹ ¨© a ¸¹
J0 dr
n 1
Using the orthogonality property of the Bessel functions
Î 0, if i j
a
Ñ 2
Ô 0
x J n (D i x) J n (D j x) dx Ïa 2
Ñ J n 1 (D i ), if i j
Ð 2
where α i , α j are the zeros of J n ( x) 0, we get
§ r2 · f
a § ξn r · § ξn h · a2 2
V0 ³0 ¨1 2 ¸ rJ 0 ¨© a ¸¹ dr
© a ¹
¦ An sinh ¨© ¸
a ¹ 2 1
J (ξ n )
n 1
which gives
2V0 a § r2 · § ξnr ·
An
§ ξ h· ³0 ¨1 2 ¸ rJ 0 ¨© a ¸¹ dr
© a ¹
a 2 sinh ¨ n ¸ J 12 (ξ n )
© a ¹
By letting ξ n r /a x, this equation can be modified to
2V0 ξn
An
§ ξ h· ³0 (ξ n2 x 2 ) x J 0 ( x) dx
ξ n4 sinh ¨ n ¸ J 12 (ξ n )
© a ¹
Using the well-known recurrence relation
d a
xα Jα 1 ( x) [ x Jα ( x)] for B 1, 2, }
dx
we get
³ xJ0 ( x) ³x
2
xJ1 ( x), J1 ( x) x 2 J 2 ( x)
From these relations, we obtain

2V0 ξn
An
§ ξ h· ³0 (ξ n2 x 2 ) d [ xJ1 ( x)]
ξ n4 sinh ¨ n ¸ J 12 (ξ n )
© a ¹
Integrating by parts, we get
4V0 ξn
An
ξ n4 sinh (ξ n h /a ) J 12 (ξ n ) 0
³ x 2 J1 ( x) dx
4V0 ξn
ξ n4 sinh (ξ n h /a ) J 12 (ξ n ) ³0
d [ x 2 J 2 ( x)]
4V0 ξ
[ x 2 J 2 ( x)] 0n
ξ n4 sinh (ξ n h /a ) J 12 (ξ n )
Thus,
4V0 J 2 (ξ n )
An
ξ n2 sinh (ξ n h /a ) J 12 (ξ n )
The recurrence relation

2n
J n 1 ( x) J n 1 ( x ) J n ( x)
x
for n 1 gives
2
J 0 (ξ n ) J 2 (ξ n ) J (ξ )
ξn 1 n
Hence,
2
J 2 (ξ n ) J (ξ )
ξn 1 n
since J 0 (ξ n ) 0. Therefore,
8V0 J1 (ξ n )
An
ξ n3 sinh (ξ n h /a ) J 12 (ξ n )
Thus, the required potential inside the cylinder is
§ξ r · ªξ º
f 8V0 J 0 ¨ n ¸ sinh « n ( z h)»
© a ¹ ¬a ¼
u (r , z ) ¦ § ξ h ·
n 1 ξ n3 J1 (ξ n ) sinh ¨ n ¸
© a ¹
2.12 SOLUTION OF LAPLACE EQUATION IN SPHERICAL COORDINATES

In Example 2.3, the Laplace equation is expressed in spherical coordinates and has the
following form:
w § 2 wu · 1 w § wu · 1 w 2u
2u ¨© r ¸¹ ¨© sin θ ¸¹ 0 (2.88)
wr wr sin θ w θ wθ sin 2 θ w φ 2
Let us assume the separable solution in the form
u (r , θ , φ ) R (r ) F (θ , φ ) (2.89)
w È 2 w RØ R w È wFØ R w2F
F É r Ù É sin T Ù 0
w r Ê w r Ú sin T w T Ê w T Ú sin 2 T w I 2
Separation of variables gives
d § 2 dR · 1 ° w § wF· 1 w 2 F ½°
¨r ¸ ® ¨© sin θ ¸¹ ¾
dr © dr ¹ sin θ ¯° w θ wθ sin θ w φ 2 ¿°
μ
R F
where μ is a separation constant. Therefore,
1 d § 2 dR ·
¨r ¸ μ (2.90)
R dr © dr ¹
1 ªw § wF· 1 w 2F º
« ¨© sin θ ¸¹ » μ (2.91)
F sin θ «¬ w θ wθ sin θ w φ 2 »¼
Equation (2.90) gives
d 2R dR
r2 2
2r μR 0
dr dr
which is a Euler’s equation. Hence, using the transformation r e z , the auxiliary equation
can be written as
D ( D 1) 2 D μ D2 D μ 0
where D d /dz. Its roots are given by

1 r 1 4 μ
D
2
Let μ α (α 1); then we get
° § 1 · ½°
2
®1 r 2 ¨©α ¸¹ ¾
°¯ 2 ° 1 § 1·
¿ r ¨α ¸
D
2 2 © 2¹
Hence, D α and (α 1). Therefore, the solution of Euler’s equation is
R c1rα c2 r (α 1) (2.92)

Taking μ α (α 1), Eq. (2.91) becomes
w § wF· 1 w 2F
¨© sin θ ¸ α (α 1) F sin θ 0
wθ w θ ¹ sin θ w φ 2
Inserting
F H (θ ) Φ (φ )
into the above equation and separating the variables, we obtain
sin θ ª d § dH · º 1 d 2Φ
¨© sin θ ¸¹ α (α 1) sin θ H » ν2
H «¬ dθ dθ ¼ Φ dφ 2
where v2 is another separation constant. Then
d 2Φ
ν 2Φ 0 (2.93)
dφ 2
sin θ ª d § dH · º
¨© sin θ ¸¹ α (α 1) sin (θ ) H » ν
2
«
H ¬ dθ dθ
(2.94)
¼
The general solution of Eq. (2.93) is
Φ c3 cos νφ c4 sin νφ (2.95)
provided ν z 0. If ν 0, the solution is independent of φ which corresponds to the axisymmetric

case. Equation (2.94) becomes, for the axisymmetric, case,
d § dH ·
¨© sin θ ¸ α (α 1) sin (θ ) H 0
dθ dθ ¹
Transforming the independent variable θ to x and by letting x cos θ , the abvoe equation
becomes
d § dH dx · dx
¨© sin θ ¸ α (α 1) sin (θ ) H 0
dx dx dθ ¹ dθ
i.e.,
d ª dH º
«¬(1 cos θ ) dx »¼ α (α 1) H
2
0
dx
or
d ª 2 dH º
dx «¬(1 x ) dx »¼ α (α 1) H 0 (2.96)
This is the well-known Legendre equation. Its general solution is given by

H c5 Pα ( x ) c6 Qα ( x ), 1 d x d 1 (2.97)
where Pα , Qα are Legendre functions of the first and second kind respectively. For convenience
let α be a positive integer, say α n. Then
H c5 Pn (cos θ ) c6Qn (cos θ ) (2.98)
Continuity of H (θ ) at θ 0, π implies the continuity of H ( x) at x r1. Since Qn(x) has a

singularity at x = 1, we choose c6 0. Therefore, in axisymmetric case the solution of Laplace
equation in spherical coordinates is given by
u (r , θ , φ ) {c1rα c2 r (α 1) } (c3 ) [c5 Pn (cos θ )]

After renaming the constants and using the principle of superposition, we find the solution
to be
f
u (r , θ ) ¦ [ An r n Bn r (n1) ] Pn (cos θ ) (2.99)
n 0
EXAMPLE 2.7 In a solid sphere of radius ‘a’, the surface is maintained at the temperature
given by
° k cos θ , 0 d θ π /2
f (θ ) ®
°̄ 0, π /2 θ π
Prove that the steady state temperature within the solid is
ª1 1§r· 5 §r·
2
3 §r·
4 º
u (r , θ ) k « P0 (cos θ ) ¨ ¸ P1 (cos θ ) ¨ ¸ P2 (cos θ ) ¨ ¸ P4 (cos θ ) »
¬« 4 2©a¹ 16 © a ¹ 32 © a ¹ ¼»
Solution It is known that the steady state temperature distribution is governed by

the Laplace equation. In spherical polar coordinates, the axisymmetric solution of the Laplace
equation in general with the assumption that the temperature should be finite at the origin is
given by Eq. (2.99) in the form
f
u (r , θ ) ¦ An r n Pn (cos θ ) (2.100)
n 0
Using the given BC: u (a, θ ) f (θ ), we have

f f
u ( a, θ ) f (θ ) ¦ An a n Pn (cos θ ) ¦ bn Pn (cos θ )
n 0 n 0
where An a n bn . This is a Fourier-Legendre series, where
2n 1 1
bn
2 ³ 1 f (θ ) Pn (cos θ ) dθ
In the present problem,
2n 1 1
bn
2 ³0 f (θ ) Pn (cos θ ) dθ
Let cos θ x,
1 1 1 1 k
b0
2 ³0 kx P0 ( x) dx
2 ³0 kx 1 dx 4
Hence, we get
k
A0 =
4
Also,
3 1 k
b1 =
2 ∫0 kx ⋅ x ⋅ dx = 2 = A1a
Therefore,
k ⎛1⎞
A1 = ⎜ ⎟
2 ⎝a⎠
5 1 5 1 3x 2 − 1 5
b2 =
2 ∫0 kxP2 ( x) dx =
2 ∫0 kx
2
dx = k
16
Thus,
5 1
A2 = k⋅ 2
16 a
Further,
7 1 7 1 5 x3 − 3 x
b3 =
2 ∫0 kxP3 ( x) dx =
2 ∫0 kx
2
dx = 0
1
Similarly, noting that P4 ( x) = (35 x 4 − 30 x 2 + 3), we get
8
3
b4 = − k = A4 a 4
32
Hence,
3 1
A4 = − k ⋅ 4 ,L
32 a
Substituting these values of A0, A1, A2,… into Eq. (2.100), we obtain, finally, the required
temperature as
⎡1 1⎛r⎞
u (r , θ ) = k ⎢ P0 (cos θ ) + ⎜ ⎟ P1 (cos θ )
⎣ 4 2⎝a⎠
5 ⎛r⎞
2
⎛ 3 ⎞⎛r ⎞
4 ⎤
+ ⎜ ⎟ 2 P (cos θ ) + ⎜⎝ − ⎟⎜ ⎟ 4 P (cos θ ) + L⎥ .
16 ⎝ a ⎠ 32 ⎠ ⎝ a ⎠ ⎥⎦
EXAMPLE 2.8 Find the potential at all points of space inside and outside of a sphere of
radius R = 1 which is maintained at a constant distribution of electric potential
u ( R, θ ) f (θ ) cos 2θ .
Solution It is known that the potential on the surface of a sphere is governed by the
Laplace equation. The Laplace equation in spherical polar coordinates is
w 2u 2 w u 1 w 2 u cot θ w u 1 w 2u
2 0
w r2 r w r r wθ 2 r 2 w θ r 2 sin 2 θ w φ 2
The possible general solution by variables separable method, after using superposition principle,
is given by Eq. (2.99). Thus we have two possible solutions:
f
u1 (r , θ ) ¦ An r n Pn (cos θ ) (2.101)
n 0
f
B
u2 ( r , θ ) ¦ r nn1 Pn (cos θ ) (2.102)
n 0
For points inside the sphere, we take the series (2.101). Why is this so? Applying the BC: u ( R, θ )
f (θ ) cos 2θ , we obtain
f
f (θ ) ¦ An Rn Pn (cos θ )
n 0
which is a generalized Fourier series of f (θ ) in terms of the Legendre polynomials. Using

the orthogonality property, we get
2n 1 1
An R n
2 ³1 f (θ ) Pn ( x) dx
Let x cos θ . Then we have
2n 1 π
2R
An n ³0
f (θ ) Pn (cos θ ) sin θ dθ
For points outside the sphere, we take the series (2.102). Why is this so? Using the BC:
u ( R, θ ) f (θ ), we get
f
B
f (θ ) ¦ Rnn1 Pn (cos θ )
n 0
Again, using the orthogonality property of Legendre polynomials, we have

2n 1 n 1 π
Bn
2
R ³0 f (θ ) Pn (cos θ ) sin θ dθ
In the present problem, it is assumed that at R 1, f (θ ) cos 2θ 2 cos 2 θ 1 2 x 2 1. Hence,
2n 1 1
³1 (2 x
2
An 1) Pn ( x) dx
2
However,
1
P2 ( x) (3x 2 1)
2
Therefore,
4 1
2x2 1 P2 ( x)
3 3
Thus,
2n 1 ª 4 1 1 1 º
An
2 «¬ 3 ³1 P2 ( x) Pn ( x) dx 3 ³1 P0 ( x) Pn ( x) dx»¼
Using the orthogonality property of Legendre polynomials, all integrals vanish except those
corresponding to n = 0 and n = 2. We obtain, therefore,
1 1 1 1
³1 P0 ( x) dx
2
A0
2 3 3
5 4 1 4
³1 P2 ( x) dx
2
A2
2 3 3
Also,
2n 1 1
³1 (2 x
2
Bn 1) Pn ( x) dx
2
2n 1 ª 4 1 1 1 º
2 «¬ 3 ³1 P2 ( x) Pn ( x) dx 3 ³1 P0 ( x) Pn ( x) dx»¼
which, on using the orthogonality property, gives the non-vanishing coefficients as
1 4
B0 , B2
3 3
Substituting these values of A0 and A2 into Eq. (2.101), we obtain
1 4
u1 (r , θ ) r 2 P2 (cosθ )
3 3
which gives the potential everywhere inside the sphere. Similarly, substituting the values of
B0 and B2 into Eq. (2.102), we get
1 4
u2 (r , θ ) = −
+ P2 (cos θ )
3r 3r 3
which gives the potential outside the sphere.
EXAMPLE 2.9 Find a general spherically symmetric solution of the following Helmholtz
equation:
(∇ 2 − k 2 ) u = 0
Solution In spherical polar coordinates, the Helmholtz equation can be written as
∂ 2 u 2 ∂ u 1 ∂ 2u cot θ ∂ u 1 ∂ 2u
+ + + + − k 2u = 0 (2.103)
∂ r2 r ∂ r r 2 ∂θ 2 r 2 ∂ θ r 2 sin 2 θ ∂ φ 2
In view of spherical symmetry, we look for u to be a function of r alone. Hence, Eq. (2.103)
becomes
∂ 2u 2 ∂ u
+ − k 2u = 0
∂ r2 r ∂ r
Therefore, we have to solve
∂ 2u ∂u
r2 + 2r − k 2 r 2u = 0 (2.104)
∂r 2 ∂r
Let
1
u= F (r )
r
Differentiating twice with respect to r and rearranging, we obtain
∂u F (r )
2r =− + 2 r F ′ (r )
∂r r
∂ 2u 3
r2 2
= − r −1/2 F (r ) − r1/2 F ′ (r ) + r 3/2 F ′′ (r )
∂r 4
Substituting the above relations, Eq. (2.104) becomes
⎛ 1⎞
r 2 F ′′ (r ) + rF ′ (r ) − ⎜ k 2 r 2 + ⎟ F (r ) = 0
⎝ 4⎠
or
⎡ ⎛1⎞ ⎤
2
r 2 F ′′(r ) + rF ′ (r ) + ⎢(ik )2 r 2 − ⎜ ⎟ ⎥ F (r ) = 0
⎢⎣ ⎝2⎠ ⎥
⎦
This is the Bessel equation whose solution is

F (r ) AJ1/2 (ikr ) BY1/2 (ikr )
where J1/2 , Y1/2 are Bessel functions with imaginary arguments, and is rewritten as
F (r ) AI1/2 (kr ) BK1/2 (kr )
Therefore,
u (r ) r 1/2 [ AI1/2 (kr ) BK1/2 (kr )]
But as r o f, the solution should be finite, which is possible only if A = 0. It is also known
that for large z,
π z
K1/2 ( z ) e
2z
Thus the acceptable spherically symmetric solution of the Helmholtz equation is given by
π kr c kr
u (r ) Br 1/2 e e
2kr r
where
π
c B
2k
2.13 MISCELLANEOUS EXAMPLES
EXAMPLE 2.10 Show that the velocity potential for an irrotational flow of an incompressible
fluid satisfies the Laplace solution.
Solution Let us consider a closed surface S enclosing a fixed volume V in the region
occupied by a moving fluid as shown in Fig. 2.5.
∧
n
dS
V
Fig. 2.5 Conservation of mass.

Let ρ be the density of the fluid. If n̂ is a unit vector in the direction of the normal to the
surface element dS and q the velocity of the fluid at that point, then the inward normal velocity
is (q nˆ ). Hence the mass of the fluid entering per unit time through the element dS is (q nˆ ) dS . It
follows therefore that the mass of the fluid entering the surface S in unit time is
³³ ρ (q nˆ) dS
S
Also, the mass of the fluid within S is
³³³ ρ dV
V
So the rate at which the mass goes on increasing is given by
w wS
wt ³³³ S dV ³³³ wt
dV
V V
By conservation of mass, the rate of generation of mass within a given volume under the
assumption that no internal sources are present is equal to the net inflow of mass through the
surface enclosing the given volume. Thus,
wρ
³³³ wt
dV ³³ ρ (q nˆ ) dS
V S
³³³ div ( ρ q) dV [using the divergence theorem]

V
Therefore,
ªw ρ º
³³³ «¬ w t div ( ρq)»¼ dV 0
V
Since the integrand is a continuous function and since this result is true for any arbitrary
volume element dV, it follows that the integrand is zero. Therefore,
wρ
ρq 0
wt
which is called the equation of continuity. For an incompressible fluid, ρ constant and,
therefore,
q 0
Further, if the flow is irrotational, i.e., there exists a velocity potential φ such that
q φ
Hence,
q φ 2φ 0
Thus, an incompressible irrotational fluid satisfies the Laplace equation.
EXAMPLE 2.11 A thin rectangular homogeneous thermally conducting plate lies in the xy-
plane defined by 0 d x d a, 0 d y d b. The edge y 0 is held at the temperature Tx ( x a),
where T is a constant, while the remaining edges are held at 0°. The other faces are insulated
and no internal sources and sinks are present. Find the steady state temperature inside the
plate.
Solution Since no heat sources and sinks are present in the plate, the steady state
temperature u must satisfy 2 u 0. Hence the problem is to solve
PDE: 2u 0
BCs: u (0, y ) 0, u ( a, y ) 0, u ( x, b) 0, u ( x, 0) Tx ( x a )
This is a typical Dirichlet’s problem. The general solution satisfying the first three BCs is
given by Eq. (2.47). Therefore,
f
§ nπ · ª nπ º
u ( x, y ) ¦ An sin ¨© a x ¸ sinh « ( y b)»
¹ ¬ a ¼
n 1
where
nπ b 2 a § nπ ·
An sinh
a a ³0 f ( x) sin ¨
© a
x ¸ dx
¹
Using the last BC: u ( x, 0) Tx ( x a ) f ( x), we get
nπ b 2 a § nπ ·
An sinh
a a ³0 Tx ( x a) sin ¨© a x ¸ dx
¹
2T a § nπ ·
a ³0 x ( x a ) sin ¨
© a
x ¸ dx
¹
a 2T ª a § nπ · ½º

nπ a ¬
« ³0 x ( x a ) d ®cos ¨
¯ ©
x ¸ ¾»
a ¹ ¿¼
a
2T ª § nπ · º a a ª § nπ · º
«
nπ ¬
( x a ) cos ¨
© a
x ¸»
¹ ¼ 0 nπ ³0 (2 x a) d «¬sin ¨© a x ¸»
¹¼
a
2aT ª § nπ ·º a § nπ ·
2 2 «
nπ ¬
(2 x a ) sin ¨
© a
x ¸»
¹¼ 0 ³0 2 sin ¨© a x ¸ dx
¹
2aT ° § nπ · º ½°
a
2a ª
2 2 ®
a sin nπ «cos ¨© a x ¸¹ » ¾
n π ¯° nπ ¬ ¼ 0 °¿
2aT 2a 4a 2T
(cos nπ 1) [(1)n 1]
n 2π 2 nπ nπ3 3
Thus the required temperature distribution is given by

f
§ nπ · 4Ta § nπ · ª nπ
2
º
u ( x, y ) ¦ cosech ¨© a b ¸¹ n3π 3 [(1)n 1] sin ¨© a x ¸ sinh « ( y b)»
¹ ¬ a ¼
n 1
EXAMPLE 2.12 Solve

2u 0, 0 d x d a, 0d ydb
satisfying the BCs:
u (0, y ) 0, u ( x, 0) 0, u ( x, b ) 0
wu πy
( a, y ) T sin 3
wx a
Solution Using the variables separable method, one of the acceptable general solutions
is given by Eq. (2.38). Hence
u ( x, y ) (c1e px c2 e px ) (c3 cos py c4 sin py )
Using the BC: u ( x, 0) 0, we get
0 c3 (c1e px c2 e px )
implying c3 = 0. Therefore,
u ( x, y ) c4 sin py (c1e px c2 e px )
Now, using the BC: u ( x, b) 0, we obtain
0 c4 sin pb (c1e px c2 e px )
c4 z 0 (why?) implying sin pb 0 which gives

nπ
pb nπ or p , n 1, 2, 3, }
b
Thus,
§ nπ ·
c4 sin ¨ y (c e px c2 e px )
© b ¸¹ 1
u ( x, y )
Renaming the constants, we have
§ nπ · ª § nπ · § nπ · º
u ( x, y ) sin ¨ x B exp ¨ n 1, 2,}
© b ¸¹ «¬
y A exp ¨
© b ¸¹ © b ¸¹ »¼
x ,
If we use the BC: u (0, y ) 0, we get
§ nπ ·
sin ¨ y ( A B)
© b ¸¹
0
giving A + B = 0; therefore, A = –B. Thus,
§ nπ ·ª § nπ · § nπ ·º
u ( x, y ) A sin ¨ y ¸ «exp ¨ x ¸ exp ¨ x¸»
© b ¹¬ © b ¹ © b ¹¼
§ nπ · § nπ ·
2 A sin ¨ y ¸ sinh ¨ n 1, 2, }
© b ¸¹
x ,
© b ¹
Differentiating with respect to x, we obtain

wu nπ § nπ · § nπ ·
sin ¨ y ¸ cosh ¨
© b ¸¹
2A x
wx b © b ¹
The last BC yields
πy nπ § nπ · § nπ ·
T sin 3 sin ¨
© b ¸¹
2A y cosh ¨ a¸
a b © b ¹
from which we can determine 2A. Hence, the required solution is
bT πy nπ § nπ ·
u ( x, y ) sin 3 sech a sinh ¨ x¸
nπ a b © b ¹
The principle of superposition gives the required solution as

f
πy § nπ · § nπ ·
¦ nπ sin3
bT
sech ¨ a ¸ sinh ¨
© b ¸¹
u ( x, y ) x
a © b ¹
n 1
EXAMPLE 2.13 Find the potential function u ( x, y , z ) in a rectangular box defined

by 0 d x d a, 0 d y d b, 0 d z d c (see Fig. 2.6), if the potential is zero on all sides and the
bottom, while u f ( x, y ) on the top of the box.
z
c
b y
a O
x
Fig. 2.6 Rectangular box.
Solution The potential distribution in the rectangular box satisfies the Laplace equation.
Thus the problem is to solve
2u u xx u yy u zz 0
subject to the BCs:
u (0, y, z ) u ( a, y , z ) 0
u ( x, 0, z ) u ( x, b, z ) 0
u ( x, y, 0) 0
u ( x, y , c ) f ( x, y )
Following the variables separable method, let us assume the solution in the form
u ( x, y , z ) X ( x) Y ( y ) Z ( z )
Substituting into the Laplace equation, we get
X cc ( x) Y ( y ) Z ( z ) X ( x) Y cc ( y ) Z ( z ) X ( x) Y ( y ) Z cc( z ) 0
which can also be written as
Y cc ( y ) Z cc ( z ) X cc ( x)
λ 12
Y ( y) Z (z) X ( x)
where λ1 is a separation constant. Thus we have
X cc ( x) λ 12 X ( x ) 0 (2.105)
After the second separation, we also have

Z cc ( z ) Y cc ( y )
λ 12 λ 22
Z ( z) Y ( y)
Y cc ( y ) λ 22Y ( y ) 0 (2.106)
Z cc ( z ) λ 32 Z ( z ) 0 (2.107)
where λ 23 λ 12 λ 22 . The general solutions of Eqs. (2.105)–(2.107) are
X ( x) c1 cos λ1 x c2 sin λ1 x
Y ( y) c3 cos λ2 y c4 sin λ2 y
Z ( z) c5 cosh λ3 z c6 sinh λ3 z
From the BCs,

X (0) X (a ) 0
Y (0) Y (b) 0
Z (0) 0
X (0) 0 gives c1 0
X (a) 0 gives λ1a mπ .
Therefore,
mπ ,
λ1 m 1, 2, }
a
Similarly,
Y (0) 0 gives c3 0
Y (b) 0 gives λ2 b nπ
Therefore,
nπ ,
λ2 n 1, 2, }
b
Also, Z (0) 0 gives c5 0. Further, we note that
§ m2 n2 ·
λ 23 λ 12 λ 22 π2 ¨ 2 2 ¸ λ mn
2
(say)
©a b ¹
Then
m2 n2
λ3 π λmn
a2 b2
The solutions now take the form
mπ x ,
X ( x) c2 m sin m 1, 2, }
a
nπ y ,
Y ( y) c4 n sin n 1, 2, }
b
Z ( z) c6 mn sinh λmn z
Let cmn c2 m c4 n c6 mn ; then, after using the principle of superposition, the required solution is
f f
mπ x nπ y
u ( x, y , z ) X ( x) Y ( y ) Z ( z ) ¦ ¦ cmn sin a
sin
b
sinh λmn z (2.108)
m 1 n 1
Using the final BC: f ( x, y ) = u ( x, y , c), we get
mπ x nπ y
f ( x, y ) = ∑ ∑ cmn sinh λmn c sin a
sin
b
which is a double Fourier sine series. Thus, we have
4 a b mπ x nπ y
cmn sinh λmn c =
ab ∫0 ∫0 f ( x, y ) sin
a
sin
b
dx dy (2.109)
Therefore, Eqs. (2.108) and (2.109) constitute the required potential.
EXAMPLE 2.14 Find the electrostatic potential u in the annular region bounded by the
concentric spheres r = a, r = b, 0 < a < b (see Fig. 2.7), if the inner and outer surfaces are kept
at constant potentials u1 and u2, u1 ≠ u2 .
Solution The electrostatic potential satisfies the Laplace equation
∇ 2u = 0
It is natural that we choose spherical polar coordinates. From the problem, it is evident that
we are looking for a solution with spherical symmetry which is independent of θ and φ .
Hence, u = u ( r ).
u = u1
a u = u2
Fig. 2.7 Annular region.
Thus, we have to solve

∂ ⎛ 2 ∂u⎞
PDE: ⎜⎝ r ⎟=0 (2.110)
∂r ∂r ⎠
subject to
BCs: u = u1 at r = a
u = u2 at r = b
Integrating Eq. (2.110) with respect to r, we obtain

∂u
r2 =A (a constant of integration)
∂r
Again, integrating, we get

A
u B
r
Now, using the BCs, we have
A A
u1 B, u2 B
a b
Solving these equations, we get
u2 u1 (u1/b) (u2 /a )
A , B
(1/a ) (1/b) (1/b) (1/ a )
Hence, using these values, the required potential is
1 ª §1 1· § 1 1 ·º
u
(1/ a) (1/b) «u1 ¨© r b ¸¹ u2 ¨© a r ¸¹ »
¬ ¼
EXAMPLE 2.15 A thermally conducting solid bounded by two concentric spheres of radii
a and b as shown in Fig. 2.8, a < b, is such that the internal boundary is kept at f1 (θ ) and
the outer boundary at f 2 (θ ). Find the steady state temperature in the solid.
Solution It is known that the steady temperature T satisfies the Laplace equation. In
the present problem,
T T (r , θ )
a r P
O x
b
y
Fig. 2.8 Region bounded by two concentric spheres.
Thus, we have to solve
PDE: 2T 0
subject to the boundary conditions
T f1 (θ ) at r a
T f 2 (θ ) at r b
In spherical polar coordinates, for axially symmetric case, the solution of the Laplace equation
is given by Eq. (2.99) as follows:
f
§ B ·
T (r , θ ) ¦ ¨© An r n r nn1 ¸¹ Pn (cos θ )
n 0
Using the BCs:

f
§ B ·
f1 (θ ) ¦ ¨© An an ann1 ¸¹ Pn (cos θ ) (2.111)
n 0
f
§ B ·
f 2 (θ ) ¦ ¨© Anbn bnn1 ¸¹ Pn (cos θ ) (2.112)
n 0
In order to find the coefficients An and Bn, we have to express f1 (θ ) and f 2 (θ ) in terms
of Legendre polynomials and compare the coefficients. In this process, the following orthogonality
relation is useful:
0, if m z n
π °
³0 Pm (cos θ ) Pn (cos θ ) sin θ dθ ® 2
° , if m n
¯ 2n 1
Thus, multiplying both sides of Eq. (2.111) by Pm (cos θ ) sin θ and integrating, we obtain
π f π
§ B ·
³0 f1 (θ ) Pm (cos θ )sin θ dθ ¦ ¨© An an ann1 ¸¹ ³0 Pn (cos θ ) Pm (cos θ )sin θ dθ
n 0
§ m Bm · 2 (2.113)
¨© Am a m 1 ¸¹ 2m 1
a
Similarly, Eq. (2.112) gives
π f π
§ B ·
³0 f 2 (θ ) Pm (cos θ )sin θ dθ ¦ ¨© Anbn bnn1 ¸¹ ³0 Pn (cos θ ) Pm (cos θ )sin θ dθ
n 0
§ m Bm · 2 (2.114)
¨© Am b m1 ¸¹ 2m 1
b
Let
2m + 1 π
2 0 ∫
f1 (θ ) Pm (cos θ ) sin θ dθ = Cm
2m + 1 π
2 0∫f 2 (θ ) Pm (cos θ ) sin θ dθ = Dm
Then Eqs. (2.113) and (2.114) reduce to
Bm
Am a m + = Cm
a m +1
Bm
Am bn + = Dm
b m +1
Solving this pair of equations, we obtain
Cm a m+1 − Dm bm+1 (2.115)

Am =
a 2m+1 − b2m +1
a m+1bm+1 (Cm bm − Dm a m ) (2.116)

Bm =
b2m+1 − a 2m+1
Hence, the required steady temperature is
∞
⎛ B ⎞
T (r , θ ) = ∑ ⎜⎝ Am r m + r mm+1 ⎟⎠ Pm (cos θ )
m=0
where Am and Bm are given by Eqs. (2.115) and (2.116).
EXAMPLE 2.16 A thin annulus occupies the region 0 < a ≤ r ≤ b, 0 ≤ θ ≤ 2π . The faces are
insulated. Along the inner edge the temperature is maintained at 0°, while along the outer
edge the temperature is held at T = K cos (θ /2), where K is a constant. Determine the temperature
distribution in the annulus.
Solution Mathematically, the problem is to solve
PDE: ∇ 2T = 0, a ≤ r ≤ b, 0 ≤ θ ≤ 2π
BCs: T (a, θ ) = 0
T (b, θ ) = k cos θ /2
The required general solution is given by Eq. (2.57) in the form
T (r , θ ) = (c1r n + c2 r − n ) (c3 cos nθ + c4 sin nθ )

Using the first BC, we get

0 (c1a n c2 a n ) (c3 cos nθ c4 sin nθ )
implying thereby c1a n c2 a n 0, or c2 c1a 2n . After adjusting the constants suitably, we have
§ n a 2n ·
T (r , θ ) ¨ r n ¸ ( A cos nθ B sin nθ )
© r ¹
The principle of superposition gives
f § n a 2n ·
T (r , θ ) ¦ ¨ r n ¸ ( An cos nθ Bn sin nθ )
n 1© r ¹
Now, using the second boundary condition, we obtain
f
θ
T (b, θ ) K cos
2 ¦ (bn bn a2n ) ( An cos nθ Bn sin nθ )
n 1
which is a full-range Fourier series. Hence,
1 2π θ
An (b n b n a 2 n )
π 0 ³
K cos cos nθ dθ
2
k 2π ª § 1· § 1· º
2π ³0 «cos ¨© n 2 ¸¹ θ cos ¨© n 2 ¸¹ θ » dθ
¬ ¼
2π
ª § 1· § 1· º
sin n ¸ θ sin ¨ n ¸ θ »
k « ¨© 2¹ © 2¹
« »
2π « 1 1 »
n n
«¬ 2 2 »¼
0
0
implying An 0. Also,
k 2π θ
Bn (b n b n a 2 n )
π ³0 cos
2
sin nθ dθ
k 2π ª § 1· § 1· º
2π ³0 «sin ¨© n 2 ¸¹ θ sin ¨© n 2 ¸¹ θ » dθ
¬ ¼
2π
ª § 1· § 1· º
cos ¨ n ¸ θ cos ¨ n ¸ θ »
k « © 2¹ © 2¹
« »
2π « 1 1 »
n n
«¬ 2 2 »¼
0
§ ·
k ¨ 1 1 1 1 ¸

2π ¨ 1 1 1 1¸
¨© n n n n ¸
2 2 2 2¹
§ ·
k ¨ 1 1 ¸ k 2n

π ¨n 1 n 1 ¸ π n2 1
¨© ¸
2 2¹ 4
or
8kn
Bn (b n b n a 2 n )
π (4n 2 1)
Thus the temperature distribution in the annulus is given by
f ª (r / a ) n (a / r ) n º
8k n
T (r , θ )
π ¦ «
4n2 1 ¬« (b / a)n (a /b) n ¼»
» sin nθ
n 1
EXAMPLE 2.17 V is a function of r and θ satisfying the equation
w 2V 1 w V 1 w 2V
0
within the region of the plane bounded by r a, r b, θ 0, θ π /2. Its value along the
boundary r a is θ (π /2 θ ), along the other boundaries is zero. Prove that
f
2 (r / b)4n 2 (b / r )4n 2 ª sin (4n 2)θ º
V
π ¦ (a /b)4n2 (b/ a)4n2 «¬ (2n 1)3 ¼
»
n 1
Solution The task is to solve the PDE
w 2V 1 w V 1 w 2V
0
subject to the following boundary conditions:
(i) V (b, θ ) 0, 0 θ π/2
(ii) V (r , π / 2) 0, ardb
(iii) V (r , 0) 0, ardb
(iv) V (a, θ ) θ (π / 2 θ ), 0 θ π /2.
The three possible solutions (see Section 2.8) are given as follows:
V (c1r p c2 r p ) (c3 cos pθ c4 sin pθ )
V [c1 cos ( p ln r ) c2 sin ( p ln r )] (c3e pθ c4 e pθ )
V (c1 ln r c2 ) (c3θ c4 )
Since the problem is not defined for r = 0, f, the second and third solutions are not acceptable.
Hence, the generally acceptable solution is the first one. The boundary condition (iii) gives
0 c3 (c1r p c2 r p )
implying c3 0. The boundary condition (ii) implies
π
0 c4 sin p (c1r p c2 r p )
2
Therefore,
π n 1, 2, }
sin p 0 or p 2 n,
2
Thus, the possible solution of the given equation has the form
V (r , θ ) c4 sin (2nθ ) (c1r 2 n c2 r 2n )

Now, applying the boundary condition (i), we get
0 c4 sin (2nθ ) (c1b 2 n c2 b 2 n )
which gives c2 c1b 4 n . Therefore,
V (r , θ ) c1c4 sin (2nθ )[r 2n r 2 n b 4 n ]

Superposing all the solutions, we obtain
f
V (r , θ ) ¦ cn sin (2nθ ) (r 2n r 2nb4n )
n 1
Satisfying boundary conditions (iv), we get
§π · § a 4n b 4n ·
θ ¨ θ ¸
©2 ¹ ¦ cn sin (2nθ ) ¨
© a 2n ¹
¸
which is a Fourier sine series. Thus, we have
2 π /2 §π · § a 4n b4n ·
π /2 ³0 θ ¨ θ ¸ sin (2nθ ) cn ¨
©2 ¹ © a 2n ¹
¸
Integrating by parts, we obtain

π/2
ª§ π 2 · cos (2nθ ) ½ § π · sin (2nθ ) ½ cos (2nθ ) ½º π § a 4n b 4n ·
« ©¨ 2 θ θ ¹¸ ® ¾ ¨ 2θ ¸ ®
© ¹
¾ (2) ® ¾» cn ¨ ¸
¬ ¯ 2n ¿ 2 ¯ 4n 2
¿ ¯ 8n 3 ¿ ¼ 0 4 © a 2n ¹
On simplification, we get
1 π § a 4n b4n ·
{(1)n 1} cn
4n 3 4 ¨© a 2 n ¸¹
Thus,
1 ,
π § a 4 n b 4 n · ° 2 n3 for n odd
cn ®
4 ¨© a 2 n ¸¹ °
¯ 0, for n even
f 4n 2 § r 8n 4 b8n 4 ·
2 1 §a·
V (r , θ ) ¦ π (2n 1)3
¨© ¸¹
r
sin (4n 2)θ ¨ 8n 4
©a
¸
b8n 4 ¹
1
which can be recast in the form given in the problem.
EXAMPLE 2.18 Determine the potential of a grounded conducting sphere in a uniform

field defined by
PDE: 2u 0, 0 d r a, 0 θ π , 0 d φ 2π
BCs: (i) u ( a, θ ) 0.
(ii) u o E0 r cos θ as r o f.
Solution In spherical polar coordinates, with axial symmetry, the solution of the Laplace
equation is given by Eq. (2.99) in the form
f
§ B ·
u (r , θ ) ¦ ¨© An r n r nn1 ¸¹ Pn (cos θ )
n 0
Using the boundary condition (ii), we have

f
u (r , θ ) ¦ An r n Pn (cos θ ) E0 r cos θ
n 0
which is true only for n = 1, when P1 (cos θ ) cos θ . Also, An 0 for n t 2. Therefore,
u (r , θ ) A1r cos θ E0 r cos θ

implying A1 = − E0 . Hence,
∞
B
u (r , θ ) = − E0 r cos θ + ∑ r n+n1 Pn (cos θ )
n =1
Now, applying the boundary condition (i), we get

∞
B
0 = − E0 a cos θ + ∑ ann+1 Pn (cos θ )
n =1
Multiplying both sides by Pm (cos θ ) sin θ and integrating between the limits 0 to π , we have
π ∞ π
B
E0 a ∫0 cos (θ ) Pm (cos θ )sin θ dθ = ∑ ann+1 ∫0 Pn (cos θ ) Pm (cos θ ) sin θ dθ (2.117)
n =1
Using the orthogonality property
⎧ 0, for m ≠ n
π ⎪
∫0 Pn (cos θ ) Pm (cos θ )sin θ dθ = ⎨ 2
⎪ , for m = n
⎩ 2m + 1
we obtain
Bm 2 π
a m +1 2m + 1
= E0 a
0∫cos (θ ) Pm (cos θ ) sin θ dθ
or
2m + 1 π
2
Bm =E0 a m + 2
0 ∫
cos (θ ) Pm (cos θ ) sin θ dθ
It can be verified that the integral on the right-hand side of the Eq. (2.117) vanishes for
all m except when m = 1, in which case
B1 = E0 a3
Therefore, the required potential is given by
E0 a3
u (r , θ ) = − E0 r cos θ + cos θ
r2
EXAMPLE 2.19 The steady, two-dimensional, incompressible viscous fluid flow past a circular
cylinder, when the inertial terms are neglected (Stokes flow), is governed by the biharmonic
PDE: ∇ 4ψ = 0, where ψ is the stream function. Find its solution subject to the BCs:
(i) ψ (r , θ ) = ∂ ψ /∂ r = 0 on r = 1
(ii) ψ (r , θ ) → r sin θ as r → ∞.
Solution In view of the cylindrical geometry, we can write
4ψ 2 ( 2ψ ) 0
where
§ w2 1 w 1 w2 ·
2ψ
¨ 2 r wr ¸ψ
©w r r 2 wθ 2 ¹
Using the variables separable method, let us look for a solution of the form
ψ (r , θ ) f (r ) sin θ
Therefore,
wψ wψ
f c (r ) sin θ , f (r ) cos θ
wr wθ
w 2ψ w 2ψ
f cc (r ) sin θ , f (r ) sin θ
w r2 wθ 2
Hence,
ª 1 1 º
2ψ « f cc (r ) r f c (r ) 2 f (r ) » sin θ
¬ r ¼
which can also be written in the form
2ψ F (r ) sin θ
where
1 1
F (r ) f cc(r ) f c (r ) 2 f ( r )
r r
Therefore,
4ψ 2 ( 2ψ ) 2 [ F (r ) sin θ ] 0
i.e.,
ª 1 1 º
« F cc (r ) r F c ( r ) 2 F ( r ) » sin θ 0
¬ r ¼
implying
1 1
F cc(r ) F c(r ) 2 F (r ) 0
r r
Introducing the transformation r ez , D d /dz , the above equation becomes
[ D ( D 1) D 1] F (r ) 0
or
( D 2 1) F (r ) 0
Its complementary function is
B
F (r ) Ae z Be z Ar
r
or
1 1 B
f cc(r ) f c (r ) 2 f (r ) Ar
r r r
i.e.
r 2 f cc (r ) rf c (r ) f (r ) Ar 3 Br
which is a homogeneous ordinary differential equation. Again using the transformation r ez,
D d /dz , we get
[ D ( D 1) D 1] f Ae3 z Be z
or
( D 2 1) f Ae3 z Be z
Its complementary function is
f (r ) Ce z De z
while its particular integral is
1 Ae3 z Bze z
( Ae3 z Be z )
D2 1 8 2
Therefore,
D A 3 B
f (r ) Cr r r ln r
r 8 2
Thus, we have
§A 3 B D·
ψ ¨© r r ln r Cr ¸¹ sin θ
8 2 r
Now to satisfy the BC: ψ o r sin θ , as r o f and from physical considerations, we choose
A = 0, Therefore,
§B D·
ψ ¨© r ln r Cr ¸¹ sin θ
2 r
The boundary condition ψ 0 on r 1 gives (C D) sin θ 0, implying C D. Also, the

boundary condition w ψ /w r 0 on r = 1 gives
B B
C D 0 2D
2 2
implying B = 4D. Hence, the general solution is
§ r 1 ·
ψ 2 D ¨ r ln r ¸ sin θ
© 2 2r ¹
EXAMPLE 2.20 The problem of axisymmetric fluid flow in a semi-infinite or in a finite

circular pipe of radius a is described as follows in cylindrical coordinates:
PDE: 2u 0, 0ra
wu
BCs: (i) 0 at r 0
wr
wu wu
(ii) 0, V (z) at r a
wz wr
Show that the speed of suction is given by
f
V ( z) ¦ αn ( An cosh αn z Bn sinh αn z) J1 (αn a)
n 1
Solution In cylindrical coordinates (r , θ , z ),
w u
1 w u 1 w 2u w 2u
2u
0
w r 2 r w r r 2 wθ 2 w z2
In axisymmetric case, the above equation becomes
w 2u 1 w u w 2u
0 (2.118)
w r2 r w r w z2
Let u (r , z ) f (r ) φ ( z ) which, when substituted into Eq. (2.118), gives
f cc (1/ r ) f c φ cc
α 2 (say)
f φ
Then
φ cc α 2φ 0 (2.119)
1
f cc f c α2 f 0 (2.120)
r
The solution of Eq. (2.119) is

φ A cosh α z B sinh α z
Equation (2.120) can be rewritten as
r 2 f cc rf c α 2 r 2 f 0
which is a Bessel’s equation of zeroth order whose general solution is
f J 0 (α r ) DY0 (α r )
Here, J 0 (α r ) and Y0 (α r ) are zeroth order Bessel functions of first and second kind respectively.
Therefore, the typical solution is
u ( A cosh α z B sinh α z ) [ J 0 (α r ) DY0 (α r )]
Now, Y0 (α r ) is infinite at r = 0, and hence D = 0. Therefore, the possible solution is
u ( A cosh α z B sinh α z ) J 0 (α r )
The condition w u /w r 0 at r = 0 is automatically satisfied, since J 0c (α r ) 0 at r = 0. Now the

boundary condition w u /w z 0 at r = a gives J 0 (α a ) 0, implying that α a are the zeros of the
Bessel function J0. Let these zeros be α n a ( n 0, 1, 2, }). Thus the appropriate solution is
f
u (r , z ) ¦ αn ( An cosh αn z Bn sinh αn z) J0 (αn r )
n 1
Using the fact that, J 0c J1 , the speed of suction is given by

f
§w u ·
V ( z) ¨© ¸¹
wr r a
¦ αn ( An cosh αn z Bn sinh αn z) J1 (αn a)
n 1
EXAMPLE 2.21 Solve the following Poisson equation:

w 2u w 2u
2
w x2 w y2
subject to the boundary conditions
u (0, y ) u (5, y ) u ( x, 0) u ( x, 4) 0
Solution We assume the solution of the form

u v X (2.121)
where v is a particular solution of the Poisson equation and ω is the solution of the corresponding
homogeneous Laplace equation. That is,
∇2v = 2 (2.122)
∇ 2ω = 0 (2.123)
It is customary to assume that v has the form
v ( x, y ) = a + bx + cy + dx 2 + exy + fy 2
Substituting this into Eq. (2.122), we get
2d + 2 f = 2
Let f = 0. Then d = 1. The remaining coefficients can be chosen arbitrarily. Thus we take
v ( x , y ) = −5 x + x 2 (2.124)
so that v reduces to zero (satisfies the boundary conditions) on the sides x = 0 and x = 5.
Now, we shall find ω from
∇ 2ω = 0, 0 < x < 5, 0< y<4 (2.125)

satisfying
ω (0, y ) = −v (0, y) = 0
ω (5, y ) = −v (5, y ) = 0
ω ( x, 0) = −v ( x, 0) = −(−5 x + x 2 )
ω ( x, 4) = −v ( x, 4) = −(−5 x + x 2 )
The above conditions are obtained by using Eqs. (2.121), (2.124) and the given boundary
conditions. By using the superposition principle (see Section 2.5), the general solution of
Eq. (2.125) is found to be
∞
ω ( x, y ) = ∑ sin (nπ x /5) [an exp (nπ y /5) + bn exp (−nπ y/5)] (2.126)
n =1
Now, applying the non-homogeneous BC: ω ( x, 0) = − (−5 x + x 2 ), we get, after renaming the
constants, the equation
ω ( x, 0) = −(−5 x + x 2 ) = ∑ An sin (nπ x /5)
Also, applying the BC: ω ( x, 4) = −(−5 x + x 2 ), Eq. (2.126) can be rewritten in the form
∞
⎛ 4nπ 4nπ ⎞ nπ x
−(−5 x + x 2 ) = ∑ ⎜⎝ an cosh 5
+ bn sinh
5 ⎠
⎟ sin
5
(2.127)
n =1
which gives
2 5 nπ x
an
5 0 ³
(5 x x 2 ) sin
5
dx
Now, integrating by parts, the right-hand side yields
5
2 ª 2 § 5 nπ · § 52 nπ x · § 53 nπ · º
an «(5 x x ) ¨ cos x ¸ (5 2 x) ¨ 2 2 sin ¸ (2) ¨ 3 3 cos x »
5 «¬ © nπ 5 ¹ © nπ 5 ¹ ©n π 5 ¸¹ »¼
0
2 ª 2(53 ) 2 § 53 · º
« 3 3 cos nπ 3 ¨ 3 ¸»
5 «¬ n π n © π ¹ »¼
4 (52 ) ª 1 cos nπ º 4 (5)2 ª 1 (1)n º

« » « 3 »
π 3 ¬ n3 n3 ¼ π 3 «¬ n3 n »¼
Hence,
8(52 )
° 3 3, when n is odd
an ®π n (2.128)
°
¯ 0, when n is even
Also, from Eq. (2.127), we have

4nπ 4nπ 2 5 § nπ ·
³0 (5x x
2
bn sinh )sin ¨
© 5 ¸¹
an cosh x dx an
5 5 5
Therefore,
ª § 4 ·º
an «1 cosh ¨ nπ ¸ »
¬ © 5 ¹¼
bn (2.129)
§ 4 ·
sinh ¨ nπ ¸
©5 ¹
Substituting an, bn from Eqs. (2.128) and (2.129) into Eq. (2.126), we get
f
ª § nQ · §4 · §4 · § nQ ·
X ( x, y ) ¦ an ¬«cosh ¨© 5 y ¸ sinh ¨ nQ ¸ cosh ¨ nQ ¸ sinh ¨
¹ ©5 ¹ ©5 ¹ © 5
y¸
¹
n 1
§ nQ · § nQ · § 4 ·º
sinh ¨ x sinh ¨ nQ ¸ »
© 5 ¸¹ ¨© 5 ¸¹
y sin
© 5 ¹¼
or
∞
⎡ ⎧ nπ ⎫ ⎛ nπ ⎞ ⎤ ⎛ nπ ⎞ ⎛4 ⎞
ω ( x, y ) = ∑ ⎣⎢sinh ⎩⎨ 5 (4 − y)⎭⎬ + sinh ⎜⎝ 5 y ⎟ ⎥ sin ⎜
⎠⎦ ⎝ 5
x ⎟ sinh ⎜ nπ ⎟
⎠ ⎝5 ⎠
(2.130)
n =1
Combining Eqs. (2.124), (2.128) and (2.130), the solution of the given Poisson equation is
u(x, y) =
⎡ ⎤
8×5 ⎢
2
sinh (2n − 1) π (4 − y )/5 + sinh [(2n − 1)π y /5] ⎥ ⎧⎪ sin [(2n − 1) π x /5] ⎫⎪
∞
x ( x − 5) + 3
⎢
π n =1 ⎢
∑ ⎡ 4 ⎤
⎥×⎨
⎥ ⎩⎪ (2n − 1)3
⎬
⎭⎪
sinh ⎢(2n − 1) π ⎥
⎢⎣ ⎣ 5 ⎦ ⎥⎦
EXAMPLE 2.22 Let IR be a region bounded by ∂ IR. Let P( x, y, z ) be any point in the
interior of IR, as shown in Fig. 2.9. Let φ be a harmonic function in IR; also, let ψ = 1/ r , where
r is the distance from P. Applying Green’s second identity, show that
1 ⎡1 ∂φ ∂ ⎛ 1 ⎞⎤
φ ( P) =
4π ∫∫ ⎢⎣ r ∂ n − φ ∂ n ⎜⎝ r ⎟⎠ ⎥⎦ ds
∂ IR
∂IR
Q IR
ε ∂Σε
P
Σε
O x
z
Fig. 2.9 An illustration of Example 2.22.
Solution Since ψ possesses a point of discontinuity in IR at P ( x, y , z ), Green’s second

identity cannot be directly applied to φ and ψ . However, ψ = 1/r is bounded in IR − Σε with
the boundary ∂ IR ∪ ∂ Σε , where Σε is a sphere of radius ε with centre at P. Now applying
Green’s second identity (2.19) to functions φ and ψ in IR– Σε , we get
⎡ 2 1 2 ⎤ ⎡ ∂ ∂φ ⎤
∫∫∫ ⎢⎣φ ∇ (1/ r ) − r ∇ φ ⎥⎦ dV = ∫∫ ⎢⎣φ ∂ n (1/ r ) − (1/ r ) ∂ n ⎥⎦ dS
IR − Σε ∂ IR
∂ ⎛∂φ ⎞
+ ∫∫ φ ∂ n (1/ r )dS − ∫∫ (1/ r ) ⎜⎝ ∂ n ⎟⎠ dS (2.131)
∂ Σε ∂ Σε
From the right-hand side of Eq. (2.131), we observe that the last two integrals depend only
on ε . But in the direction of the exterior normal to ∂ Σε , we find that
∂ ∂ 1
(1/ r ) = − (1/ r ) = 2
∂n ∂ Σε ∂ r r =ε ε
Therefore,
∂ 1 4πε 2
∫∫ φ
∂n
(1/ r ) dS = 2
ε ∫∫ φ dS =
ε2
φ (Q) = 4πφ * (Q)
∂ Σε ∂ Σε
where φ * (Q ) is the average value of φ (Q ) on ∂ Σε . Further, the third integral
∂φ 1 ⎛∂φ ⎞ ⎛∂φ *⎞
− ∫∫ (1/ r )
∂n
dS = −
ε ∫∫ ⎜⎝ ⎠⎟ dS = −4πε ⎝⎜
∂n
⎟
∂n ⎠
∂ Σε ∂ Σε
where (∂ φ /∂ n) * is an average value of the normal derivative on ∂ Σε . Substituting these results
and using the fact that ∇ 2 (1/ r ) = 0 in IR − Σε , we obtain
⎡ ∂ ∂φ ⎤ ⎛∂φ ⎞
∫∫∫ (−1/ r ) ∇ 2φ dV = ∫∫ ⎣⎢φ ∂ n (1/ r ) − (1/ r ) ∂ n ⎦⎥ dS + 4πφ * ( P) + 4πε ⎝⎜ ∂ n ⎠⎟ * (2.132)
IR − Σε ∂ IR
Now, taking the limit as ε → 0, and using the fact that φ is harmonic in IR − Σε , we arrive
at the fundamental result
1 ⎡ ∂φ ∂ ⎤
φ ( P) =
4π ∫∫ ⎢⎣(1/ r ) ∂ n − φ ∂ n (1/ r )⎥⎦ dS (2.133)
∂ IR
Thus, the value of a harmonic function at any point of IR can be obtained in terms of the
values of φ and ∂ φ /∂ n on the boundary ∂ IR of the region IR .
EXAMPLE 2.23 Find the solution of the following Helmholtz equation, using separation of
variables method:
—2u + K2u = uxx + uyy + uzz + K2u = 0 (2.134)
Solution It may be noted that the Laplacian in cartesian coordinates is a PDE with constant
coefficients, while in cylindrical or spherical coordinates, it is a PDE with variable coefficients.
Thus, let us assume the solution of the given Helmholtz equation in the form
u(x, y, z) = X(x) Y(y) Z(z)
where X(x) is a function of x alone, Y(y) is a function of y alone, Z(z) is a function of z only.
Substituting into the given Helmholtz equation, we get
X≤(x) Y(y) Z(z) + X(x) Y≤(y) Z(z) + X(x) Y(y) Z≤(z) + K2X(x) Y(y) Z(z) = 0,

X ( x ) Y ( y) Z ( z )
K2 0.
X(x) Y ( y) Z (z)
This equation is satisfied iff
X ( x ) Y ( y) Z ( z )
K12 , K 22 , K 32 (2.135)
X(x) Y ( y) Z (z)
and K2 K12 K 22 K32

The sign of these separation constants K1, K2 and K3 need not be same, of course depends
on the physical considerations. The solution of the three ODEs in Eq. (2.135) can be written
in the form
X ( x ) C1eiK1x C2 e iK1x Þ
Ñ
Y ( y) C3eiK2 y C4 e iK2 y ß (2.136)
Z ( z ) C5 eiK3 z C6 e iK3 z Ñà
Hence, in general, the solution of Eq. (2.134) can be written as

u(x, y, z) = AeiKr + Be–iKr.
However, if K2 is positive, the solution is of the form
u(x, y, z) = A cos(K1x + K2y + K3z) + B sin(K1x + K2y + K3z),
while, if K2 is negative, the solution is found to be
u(x, y, z) = A cosh(K1x + K2y + K3z) + B sinh(K1x + K2y + K3z).
EXERCISES
1. Solve the following boundary value problem:
PDE: 2u 0, 0 d r d 10, 0 d θ d π
400
BCs: u (10, θ ) (πθ θ 2 )
π
u (r , 0) 0 u (r , π )
u (0, θ ) is finite
2. A homogeneous thermally conducting solid is bounded by the concentric spheres
r = a, r = b, 0 < a < b. There are no heat sources within the solid. The inner surface
r = a is held at constant temperature T1, and at the outer surface there is radiation
into the medium r > b which is at a constant temperature T2. Find the steady temperature
T in the solid.
3. A thermally conducting solid bounded by two concentric spheres of radii a and b,

a < b, is such that the internal boundary is kept at T1 and the outer boundary
at T2 (1 cos θ ). Find the steady state temperature in the solid.
4. A thin annulus occupies the region 0 a d r d b, 0 d θ d 2π , where b > a. The faces
are insulated, and along the inner edge, the temperature is maintained at 0°, while
along the outer edge, the temperature is held at 100°. Find the temperature distribution
in the annulus.
5. A thermally conducting homogeneous disc with insulated faces occupies the
region 0 d r d a in the xy-plane. The temperature u on the rim r a, is
° C , 0 θ α
u ®
°̄ 0, α θ 2π
where α is a given angle 0 α 2π . Find the series expression for temperature at

interior points of the disc. In particular, consider the case when C 100, α π / 2.
6. If ψ is a harmonic function which is zero on the cone θ α and takes the value Σα n r n
on the cone θ β , show that, when α θ β ,
f
Q (cos α ) P (cos θ ) P (cos α ) Q (cos θ ) ½
ψ ¦ αn ¯® Qnn (cos α ) Pnn(cos β ) Pnn(cos α ) Qnn(cos β ) ¿¾ r n
n 0
7. Show that
q ,
ψ (q is constant)
| r rc |
is a solution of the Laplace equation.
8. Solve the following
w 2G 1 w G 1 w 2G
PDE: 0
w r2 r w r r 2 wR 2
wG
BCs: vr 0 at r a
wr
vr U f cos R at r f
1 wG
vR U f sin R
r wR
9. In the theory of elasticity, the stress function ψ , in the problem of torsion of a beam
satisfies the Poisson equation
∂ 2ψ ∂ 2ψ
+ = −2, 0 ≤ x ≤ 1, 0 ≤ y ≤ 1
∂ x2 ∂ y 2
with the boundary conditions ψ = 0 on sides x = 0, x = 1, y = 0 and on y = 1. Find
the stress function ψ .
10. For an infinitely long conducting cylinder of radius a, with its axis coincident with
its z-axis, the voltage u (r , θ ) obeys the Laplace equation
∇2u = 0, 0 ≤ r ≤ ∞, 0 ≤ θ ≤ 2π
Find the voltage u (r , θ ) for r ≥ a if Lt u (r , θ ) = 0, subject to the condition

r→ ∞
∂u u0
= sin 3θ
∂r r =a a
11. Hadamard’s example:

(a) Consider the Cauchy problem for the Laplace equation
u xx + u yy = 0 (E11.1)
1
subject to u ( x, 0) = 0, u y ( x, 0) = sin nx, where n is a positive integer. Show that
n
its solution is
1
un ( x , y ) sinh ny sin nx (E11.2)
n2
(b) Show that for large n, the absolute value of the initial data in (a) can be made
arbitrarily small, while the solution (E11.2) takes arbitrarily large values even at the
points (x, y) with | y | as small as we want.
(c) Let f and g be analytic, and let u1 be the solution to the Cauchy problem described
by
u xx + u yy = 0
subject to
u ( x, 0) = f ( x), u y ( x, 0) = g ( x) (E11.3)
and let u 2 be the solution of the Laplace equation (E11.1) subject to
u ( x, 0) = f ( x), u y ( x, 0) = g ( x) + (1/ n) sin nx. Show that
1
u2 ( x, y ) − u1 ( x, y ) = sinh ny sin nx (E11.4)
n2
(d) Conclude that the solution to the Cauchy problem for Laplace equation does not
depend continuously on the initial data. In other words, the initial value problem
(Cauchy problem) for the Laplace equation is not well-posed. It may be noted that
a problem involving a PDE is well-posed if the following three properties are satisfied:
(i) The solution to the problem exists.
(ii) The solution is unique.
(iii) The solution depends continuously on the data of the problem.
Fortunately, many a physical phenomena give rise to initial or boundary or IBVPs
which are well-posed.
12. Find the solution of the following PDE using separation of variables method
uxx – uy + u = 0.
CHAPTER 3
Parabolic Differential Equations
3.1 OCCURRENCE OF THE DIFFUSION EQUATION

The diffusion phenomena such as conduction of heat in solids and diffusion of vorticity in
the case of viscous fluid flow past a body are governed by a partial differential equation of
parabolic type. For example, the flow of heat in a conducting medium is governed by the
parabolic equation
wT
ρC div ( K T ) H (r , T , t ) (3.1)
wt
where ρ is the density, C is the specific heat of the solid, T is the temperature at a point with
position vector r, K is the thermal conductivity, t is the time, and H (r, T , t ) is the amount of
heat generated per unit time in the element dV situated at a point (x, y, z) whose position
vector is r. This equation is known as diffusion equation or heat equation. We shall now
derive the heat equation from the basic concepts.
Let V be an arbitrary domain bounded by a closed surface S and let V V S . Let T ( x, y,
z , t ) be the temperature at a point (x, y, z) at time t. If the temperature is not constant, heat
flows from a region of high temperature to a region of low temperature and follows the
Fourier law which states that heat flux q (r, t) across the surface element dS with normal n̂ is
proportional to the gradient of the temperature. Therefore,
q (r , t ) K T (r , t ) (3.2)
where K is the thermal conductivity of the body. The negative sign indicates that the heat flux
vector points in the direction of decreasing temperature. Let n̂ be the outward unit normal
vector and q be the heat flux at the surface element dS. Then the rate of heat flowing out
through the elemental surface dS in unit time as shown in Fig. 3.1 is
dQ (q nˆ ) dS (3.3)
182
PARABOLIC DIFFERENTIAL EQUATIONS 183
∧
n
q
dS
Fig. 3.1 The heat flow across a surface.
Heat can be generated due to nuclear reactions or movement of mechanical parts as in inertial
measurement unit (IMU), or due to chemical sources which may be a function of position,
temperature and time and may be denoted by H (r , T , t ). We also define the specific heat of
a substance as the amount of heat needed to raise the temperature of a unit mass by a unit
temperature. Then the amount of heat dQ needed to raise the temperature of the elemental
mass dm = ρ dV to the value T is given by dQ = C ρT dV . Therefore,
Q= ∫∫∫ C ρT dV
V
dQ ∂T
dt
= ∫∫∫ C ρ ∂ t dV
V
The energy balance equation for a small control volume V is: The rate of energy storage in
V is equal to the sum of rate of heat entering V through its bounding surfaces and the rate
of heat generation in V. Thus,
∂ T (r, t )
∫∫∫ Cρ ∂t
dV = − ∫∫ q ⋅ nˆ dS + ∫∫∫ H (r, T , t ) dV (3.4)
V S V
Using the divergence theorem, we get
⎡ ∂T ⎤
∫∫∫ ⎢⎣ Cρ ∂ t (r, t ) + div q (r, t ) − H (r, T , t )⎥⎦ dV = 0 (3.5)
V
Since the volume is arbitrary, we have

∂ T (r , t )
ρC = − div q (r, t ) + H (r, T , t ) (3.6)
∂t
Substituting Eq. (3.2) into Eq. (3.6), we obtain
∂ T (r , t )
ρC = ∇ ⋅ [ K ∇T (r, t )] + H (r, T , t ) (3.7)
∂t
If we define thermal diffusivity of the medium as

K
α
ρC
then the differential equation of heat conduction with heat source is
1 w T (r, t ) H (r, T , t ) (3.8)
2T (r , t )
α wt K
In the absence of heat sources, Eq. (3.8) reduces to
w T (r, t )
α 2T (r, t ) (3.9)
wt
This is called Fourier heat conduction equation or diffusion equation. The fundamental problem
of heat conduction is to obtain the solution of Eq. (3.8) subject to the initial and boundary
conditions which are called initial boundary value problems, hereafter referred to as IBVPs.
3.2 BOUNDARY CONDITIONS

The heat conduction equation may have numerous solutions unless a set of initial and boundary
conditions are specified. The boundary conditions are mainly of three types, which we now
briefly explain.
Boundary Condition I: The temperature is prescribed all over the boundary surface. That
is, the temperature T(r, t) is a function of both position and time. In other words, T G (r, t ) which
is some prescribed function on the boundary. This type of boundary condition is called the
Dirichlet condition. Specification of boundary conditions depends on the problem under
investigation. Sometimes the temperature on the boundary surface is a function of position
only or is a function of time only or a constant. A special case includes T(r, t) = 0 on the
surface of the boundary, which is called a homogeneous boundary condition.
Boundary Condition II: The flux of heat, i.e., the normal derivative of the temperature w T /w n,
is prescribed on the surface of the boundary. It may be a function of both position and time,
i.e.,
wT
f (r , t )
wn
This is called the Neumann condition. Sometimes, the normal derivatives of temperature may
be a function of position only or a function of time only. A special case includes
wT
0 on the boundary
wn
This homogeneous boundary condition is also called insulated boundary condition which
states that the heat flow is zero.
Boundary Condition III: A linear combination of the temperature and its normal derivative
is prescribed on the boundary, i.e.,
∂T
K + hT = G (r, t )
∂n
where K and h are constants. This type of boundary condition is called Robin’s condition. It
means that the boundary surface dissipates heat by convection. Following Newton’s law of
cooling, which states that the rate at which heat is transferred from the body to the surroundings
is proportional to the difference in temperature between the body and the surroundings, we
have
∂T
−K = h (T − Ta )
∂n
As a special case, we may also have
∂T
K + hT = 0
∂n
which is a homogeneous boundary condition. This means that heat is convected by
dissipation from the boundary surface into a surrounding maintained at zero temperature.
The other boundary conditions such as the heat transfer due to radiation obeying the
fourth power temperature law and those associated with change of phase, like melting, ablation,
etc. give rise to non-linear boundary conditions.
3.3 ELEMENTARY SOLUTIONS OF THE DIFFUSION EQUATION

Consider the one-dimensional diffusion equation
∂ 2T 1 ∂ T , − ∞ < x < ∞, t > 0

= (3.10)
∂ x2 α ∂ t
The function
1
T ( x, t ) = exp [ -( x - ξ )2/(4α t )] (3.11)
4πα t
where ξ is an arbitrary real constant, is a solution of Eq. (3.10). It can be verified easily as
follows:
∂T 1 ( x − ξ )2 1
= − exp [ − ( x − ξ )2/(4α t )]
∂t 4πα t 4α t 2 2t
∂T 1 −2 ( x − ξ)
= exp [ − ( x − ξ )2/(4α t )]
∂x 4πα t 4α t
Therefore,
w 2T 1 ª 1 ( x ξ )2 º 1 wT
» exp [( x ξ ) /(4α t )]
2
«
w x2 4πα t «¬ 2α t 4α t »¼
2 2 α wt
which shows that the function (3.11) is a solution of Eq. (3.10). The function (3.11), known
as Kernel, is the elementary solution or the fundamental solution of the heat equation for the
infinite interval. For t > 0, the Kernel T(x, t) is an analytic function of x and t and it can also
be noted that T(x, t) is positive for every x. Therefore, the region of influence for the diffusion
equation includes the entire x-axis. It can be observed that as | x | o f, the amount of heat
transported decreases exponentially.
In order to have an idea about the nature of the solution to the heat equation, consider
a one-dimensional infinite region which is initially at temperature f (x). Thus the problem is
described by
wT w 2T
PDE: α , f x f, t ! 0 (3.12)
wt w x2
IC: T ( x, 0) f ( x), f x f, t 0 (3.13)
Following the method of variables separable, we write

T ( x, t ) X ( x) β (t ) (3.14)
Substituting into Eq. (3.12), we arrive at
X cc 1 βc
λ (3.15)
X α β
where λ is a separation constant. The separated solution for β gives
β Ceαλt (3.16)
If λ ! 0, we have β and, therefore, T growing exponentially with time. From realistic physical
considerations, it is reasonable to assume that f ( x) o 0 as | x | o f, while | T ( x, t )| M as
| x | o f. But, for T(x, t) to remain bounded, λ should be negative and thus we
take λ μ 2 . Now from Eq. (3.15) we have
X cc μ 2 X 0
Its solution is found to be
X c1 cos μ x c2 sin μ x
Hence
2
T ( x, t , μ ) ( A cos μ x B sin μ x) eαμ t (3.17)
is a solution of Eq. (3.12), where A and B are arbitrary constants. Since f (x) is in general not
periodic, it is natural to use Fourier integral instead of Fourier series in the present case. Also,
since A and B are arbitrary, we may consider them as functions of μ and take
A A ( μ ), B B ( μ ). In this particular problem, since we do not have any boundary conditions
which limit our choice of μ , we should consider all possible values. From the principles of
superposition, this summation of all the product solutions will give us the relation
f f 2
T ( x, t ) ³0 T ( x, t , μ ) dμ ³0 [ A ( μ ) cos μ x B ( μ )sin μ x] eαμ t dμ (3.18)
which is the solution of Eq. (3.12). From the initial condition (3.13), we have
f
T ( x, 0) f ( x) ³0 [ A ( μ ) cos μ x B ( μ ) sin μ x] dμ (3.19)
In addition, if we recall the Fourier integral theorem, we have

1 f ª f º
f (t )
π ³0 «¬ ³f f ( x) cos ω (t x) dx » dω
¼
(3.20)
Thus, we may write

1 f ª f º
f ( x)
π ³0 «¬ ³f f ( y) cos μ ( x y) dy »¼ dμ
1 f ª f º
π ³0 ³f f ( y) (cos μ x cos μ y sin μ x sin μ y) dy »¼ dμ
«¬
1 f ª f f º
π ³0 «¬cos μ x ³f f ( y) cos μ y dy sin μ x ³f f ( y) sin μ y dy »¼ dμ (3.21)
Let
1 f
A (μ )
π ³f f ( y) cos μ y dy
1 f
B (μ )
π ³f f ( y) sin μ y dy
Then Eq. (3.21) can be written in the form
f
f ( x) ³0 [ A ( μ ) cos μ x B ( μ )sin μ x] dμ (3.22)
Comparing Eqs. (3.19) and (3.22), we shall write relation (3.19) as

1 f ª f º
T ( x, 0) f ( x)
π ³0 ³f «¬ f ( y ) cos μ ( x y ) dy » dμ
¼
(3.23)
Thus, from Eq. (3.18), we obtain

1 f ª f º
T ( x, t )
π ³0 «¬ ³f f ( y ) cos μ ( x y ) exp ( αμ 2t ) dy » dμ
¼
(3.24)
Assuming that the conditions for the formal interchange of orders of integration are satisfied,
we get
1 f ª f º
T ( x, t )
π ³f f ( y) «
¬ ³0 exp (αμ 2 t ) cos μ ( x y ) d μ » dy
¼
(3.25)
Using the standard known integral

f π
³0 exp ( s 2 ) cos (2bs ) ds
2
exp (b 2 ) (3.26)
Setting s μ α t , and choosing

x y
b
2 αt
Equation (3.26) becomes
f 2 π
³0 eαμ t cos μ ( x y ) dμ exp [ ( x y ) 2/(4α t )] (3.27)
4α t
Substituting Eq. (3.27) into Eq. (3.25), we obtain
1 f
³f f ( y) exp [( x y) /(4α t )] dy
2
T ( x, t ) (3.28)
4απ t
Hence, if f (y) is bounded for all real values of y, Eq. (3.28) is the solution of the problem
described by Eqs. (3.12) and (3.13).
EXAMPLE 3.1 In a one-dimensional infinite solid, f x f, the surface a x b is

initially maintained at temperature T0 and at zero temperature everywhere outside the surface.
Show that
T0 ª § bx · § a x ·º
T ( x, t ) «erf ¨© 4α t ¸¹ erf ¨© 4α t ¸¹ »
2 ¬ ¼
where erf is an error function.
Solution The problem is described as follows:
PDE: Tt αTxx , f xf
IC: T T0 , a xb
0 outside the above region

The general solution of PDE is found to be

1 f
³f f (ξ ) exp [( x ξ ) /(4α t )] dξ
2
T ( x, t )
4πα t
Substituting the IC, we obtain
T0 b
³a exp [( x ξ ) /(4α t )] dξ
2
T ( x, t )
4πα t
Introducing the new independent variable η defined by
x ξ
η
4α t
and hence
dξ 4α t dη
the above equation becomes
T0 (b x ) /√ (4α t )
η 2 T0 ª 2 (b x ) /√ (4α t ) 2 2 ( a x ) /√ (4α t ) 2 º
T ( x, t ) ³(a x) /√(4αt ) e dη « ³0 eη dη ³0 eη dη »
π 2 ¬ π π ¼
Now we introduce the error function defined by
2 z
³0 exp (η ) dη
2
erf ( z )
π
Therefore, the required solution is
T0 ª §bx · § a x ·º
T ( x, t ) «erf ¨© 4α t ¸¹ erf ¨© 4α t ¸¹ »
2 ¬ ¼
3.4 DIRAC DELTA FUNCTION

According to the notion in mechanics, we come across a very large force (ideally infinite)
acting for a short duration (ideally zero time) known as impulsive force. Thus we have a
function which is non-zero in a very short interval. The Dirac delta function may be thought
of as a generalization of this concept. This Dirac delta function and its derivative play a useful
role in the solution of initial boundary value problem (IBVP).
Consider the function having the following property:
° 1/2ε , |t |ε
δ ε (t ) ® (3.29)
°̄ 0, |t |!ε
Thus,
f ε 1
³f δ ε (t ) dt ³ε 2ε dt 1 (3.30)
Let f (t) be any function which is integrable in the interval (ε , ε ). Then using the Mean-
value theorem of integral calculus, we have
f 1 ε
³f f (t ) δ ε (t ) dt 2ε ³ε f (t ) dt f (ξ ), ε ξ ε (3.31)
Thus, we may regard δ (t ) as a limiting function approached by δ ε (t ) as ε o 0, i.e.
δ (t ) Lt δ ε (t ) (3.32)
εo0
As ε o 0, we have, from Eqs. (3.29) and (3.30), the relations
(in the sense of being very large)

° f,
δ (t ) Lt δ ε (t ) ® if t 0 (3.33)
εo0
°
¯ 0, if t z 0
f
³f δ (t ) dt 1 (3.34)
This limiting function δ (t ) defined by Eqs. (3.33) and (3.34) is known as Dirac delta function
or the unit impulse function. Its profile is depicted in Fig. 3.2. Dirac originally called it an
improper function as there is no proper function with these properties. In fact, we can observe
that
f
1 ³f δ (t ) dt Lt ³
ε o 0 |t | ! ε
δ ε (t ) dt
εo0
Lt 0 0
δε
1/2ε
–ε 0 ε t
Fig. 3.2 Profile of Dirac delta function.
Obviously, this contradiction implies that δ (t ) cannot be a function in the ordinary sense.
Some important properties of Dirac delta function are presented now:
f
PROPERTY I: ³f δ (t ) dt 1
PROPERTY II: For any continuous function f (t),

f
³f f (t ) δ (t ) dt f (0)
Proof Consider the equation

f
ε o 0 ³f
Lt f (t ) δ ε (t ) dt Lt f (ξ ), ε ξ ε
ξo0
As ε o 0, we have ξ o 0. Therefore,
f
³f f (t ) δ (t ) dt f (0)
PROPERTY III: Let f (t) be any continuous function. Then

∞
∫ −∞
δ (t − a ) f (t ) dt = f (a )
Proof Consider the function
⎧⎪ 1/ε , a < t < a +ε

δ e (t − a) = ⎨
⎪⎩ 0, elsewhere
Using the mean-value theorem of integral calculus, we have

f 1 a ε
³f δ ε (t a ) f (t ) dt
ε ³a f (t ) dt f (a θε ), 0 θ 1
Now, taking the limit as ε o 0, we obtain

f
³f δ (t a) f (t ) dt f (a )
Thus, the operation of multiplying f (t) by δ (t a) and integrating over all t is equivalent to
substituting a for t in the original function.
PROPERTY IV: δ (t ) δ (t )
1
PROPERTY V: δ (at ) δ (t ), a!0
a
PROPERTY VI: If δ (t ) is a continuously differentiable. Dirac delta function vanishing for large
t, then
∞
∫− ∞ f (t ) δ ′(t ) dt = − f ′(0)
Proof Using the rule of integration by parts, we get
∞ ∞
∫−∞ f (t ) δ ′ (t ) dt = [ f (t ) δ (t )] ∞
−∞ − ∫−∞ f ′(t ) δ (t ) dt
Using Eq. (3.33) and property (III), the above equations becomes
∞
∫− ∞ f (t ) δ ′(t ) dt = − f ′(0)
∞
PROPERTY VII: ∫−∞ δ ′(t − a) f (t ) dt = − f ′(a)
Having discussed the one-dimensional Dirac delta function, we can extend the definition to
two dimensions. Thus, for every f which is continuous over the region S containing the
point (ξ , η ), we define δ ( x − ξ , y − η ) in such a way that
∫∫ δ ( x − ξ , y − η ) f ( x, y) dσ = f (ξ , η ) (3.35)
S
Note that δ ( x − ξ , y − η ) is a formal limit of a sequence of ordinary functions, i.e.,
δ ( x − ξ , y − η ) = Lt δ ε (r ) (3.36)
ε →0
where r 2 = ( x − ξ )2 + ( y − η )2 . Also observe that
∫∫ δ ( x − ξ ) δ ( y − η) f ( x, y) dx dy = f (ξ , η) (3.37)
Now, comparing Eqs. (3.35) and (3.37), we see that

δ ( x − ξ , y − η) = δ (x − ξ )δ ( y − η) (3.38)
Thus, a two-dimensional Dirac delta function can be expressed as the product of two one-
dimensional delta functions. Similarly, the definition can be extended to higher dimensions.
EXAMPLE 3.2 A one-dimensional infinite region − ∞ < x < ∞ is initially kept at zero
temperature. A heat source of strength gs units, situated at x = ξ releases its heat instantaneously
at time t = τ . Determine the temperature in the region for t > τ .
Solution Initially, the region f x f is at zero temperature. Since the heat source
is situated at x ξ and releases heat instantaneously at t τ , the released temperature
at x ξ and t τ is a δ - function type. Thus, the given problem is a boundary value problem
described by
w 2T g ( x, t ) 1 wT
PDE: , f x f, t ! 0
wx 2 k α wt
IC: T ( x, t ) F ( x) 0, f x f, t 0
g ( x, t ) g sδ ( x ξ ) δ (t τ )
The general solution to this problem as given in Example 7.25, after using the initial condition
F ( x) 0, is
α t dt c f
T ( x, t )
k ³t c 0 4πα (t t c ) ³x c f
g ( x c, t c ) exp [ ( x x c)2 /{4α (t t c)}] dx c (3.39)
Since the heat source term is of the Dirac delta function type, substituting
g ( x, t ) g sδ ( x ξ ) δ (t τ )
into Eq. (3.39), and integrating we get, with the help of properties of delta function, the
relation
α gs t exp [( x ξ )2 /{4α (t t c)}]

T ( x, t )
k 4πα ³0 t tc
δ (t τ ) dt c
Therefore, the required temperature is
α g s exp [ ( x ξ )2 /{4α (t τ )}] for t ! τ

T ( x, t )
k 4πα (t τ )
EXAMPLE 3.3 An infinite one-dimensional solid defined by f x f is maintained at

zero temperature initially. There is a heat source of strength g s (t ) units, situated at x ξ , which
releases constant heat continuously for t > 0. Find an expression for the temperature distribution
in the solid for t > 0.
Solution This problem is similar to Example 3.2, except that g ( x, t ) g s (t ) δ ( x ξ ) is

a Dirac delta function type. The solution to this IBVP is
α t g s (t c )
T ( x, t )
K ³t c 0 4πα (t t c )
exp [ ( x ξ )2 /{4α (t t c )}] dt c (3.40)
It is given as g s (t ) constant g s (say). Let us introduce a new variable η defined by
x ξ 1( x ξ ) 2
η or t tc
4α (t t c ) η 2 4α
Therefore,
1 ( x ξ )2
dt c dη
η 3 2α
Thus, Eq. (3.40) becomes
x ξ f exp (η 2 )
T ( x, t ) gs
2K π ³( xξ )/ 4α t η2
dη
However,
d § eη 2 · eη
2
2
¨ ¸ 2eη
dη ¨© η ¸¹ η 2
Hence,
ª § η 2 ·
f º
x ξ « e f
η 2 »
T ( x, t ) gs
2K π « ¨¨ η ¸
¸¹
2 ³( xξ ) / (4αt ) e dη »
«¬ © ( x ξ ) / 4α t »¼
Recalling the definitions of error function and its complement

2 x
η 2
erf ( x)
π ³0 e dη , erf (f) 1
2 § f x ·
³0 exp (η 2 ) dη ³0 exp (η ) dη ¸
2
erfc ( x) 1 erf ( x) ¨
π © ¹
2 f
π ³x exp (η 2 ) dη
the temperature distribution can be expressed as
α gs ª t | x ξ | § x ξ ·º
T ( x, t ) « exp [( x ξ )2 /(4α t )] ¨©1 erf ¸»
K ¬ 2π 2α 4α t ¹ ¼
Alternatively, the required temperature is
α gs ª t | x ξ | x ξ º
T ( x, t ) « exp [ ( x ξ )2 /(4α t )] erfc »
K ¬ 2π 2α 4α t ¼
3.5 SEPARATION OF VARIABLES METHOD

Consider the equation
wT w 2T
α (3.41)
wt w x2
Among the many methods that are available for the solution of the above parabolic partial
differential equation, the method of separation of variables is very effective and straightforward.
We separate the space and time variables of T(x, t) as follows: Let
T ( x, t ) X ( x) β (t ) (3.42)
be a solution of the differential Eq. (3.41). Substituting Eq. (3.42) into (3.41), we obtain
X cc 1 βc
K , a separation constant
X α β
Then we have
d2X
KX 0 (3.43)
dx 2
dβ
αKβ 0 (3.44)
dt
In solving Eqs. (3.43) and (3.44), three distinct cases arise:
Case I When K is positive, say λ 2 , the solution of Eqs. (3.43) and (3.44) will have the form
c1eλ x c2 e λ x ,
2
X β c3eαλ t (3.45)
Case II When K is negative, say λ 2 , then the solution of Eqs. (3.43) and (3.44) will have
the form
2
X c1 cos λ x c2 sin λ x, β c3e αλ t (3.46)
Case III When K is zero, the solution of Eqs. (3.43) and (3.44) can have the form
X c1 x c2 , β c3 (3.47)
Thus, various possible solutions of the heat conduction equation (3.41) could be the following:
2
T ( x, t ) (c1ceλ x c2c e λ x ) eαλ t
2
T ( x, t ) (c1c cos λ x c2c sin λ x) eαλ t
(3.48)
T ( x, t ) c1cx c2c
where
c1c c1c3 , c2c c2 c3
EXAMPLE 3.4 Solve the one-dimensional diffusion equation in the region 0 d x d π , t t 0,
subject to the conditions
(i) T remains finite as t o f
(ii) T 0, if x 0 and π for all t
x, 0 d x d π /2
°
(iii) At t 0, T ® π
°̄ π x, d x d π.
2
Solution Since T should satisfy the diffusion equation, the three possible solutions are:
2
T ( x, t ) (c1eλ x c2 e λ x ) eαλ t
2
T ( x, t ) (c1 cos λ x c2 sin λ x) eαλ t
T ( x, t ) (c1 x c2 )
The first condition demands that T should remain finite as t o f. We therefore reject the first
solution. In view of BC (ii), the third solution gives
0 c1 0 c2 , 0 c1 π c2
implying thereby that both c1 and c2 are zero and hence T = 0 for all t. This is a trivial
solution. Since we are looking for a non-trivial solution, we reject the third solution also.
Thus, the only possible solution satisfying the first condition is
2
T ( x, t ) (c1 cos λ x c2 sin λ x) eαλ t
Using the BC (ii), we have

0 (c1 cos λ x c2 sin λ x ) x 0
implying c1 = 0. Therefore, the possible solution is
2
T ( x, t ) c2 eαλ t sin λ x
Applying the BC: T 0 when x π , we get
sin λπ 0 λπ nπ
where n is an integer. Therefore,
λ n
Hence the solution is found to be of the form
2
T ( x, t ) ceα n t sin nx
Noting that the heat conduction equation is linear, its most general solution is obtained by
applying the principle of superposition. Thus,
f
¦ cneα n t sin nx
2
T ( x, t )
n 1
Using the third condition, we get

f
T ( x, 0) ¦ cn sin nx
n 1
which is a half-range Fourier-sine series and, therefore,
2 π 2ª π /2 π º
cn
π ³0 T ( x, 0) sin nx dx
π «¬ ³0 x sin nx dx ³π /2 (π x)sin nx dx»¼
Integrating by parts, we obtain
2 ª § cos nx sin nx · π /2 π
cos nx sin nx ½ º
cn «¨ x ¸ ® (π x ) ¾ »
π «¬ © n n2 ¹ 0 ¯ n n 2 ¿ π /2 »¼
or
4 sin (nπ /2)
cn
n2π
Thus, the required solution is
f 2
4 eα n t sin (nπ /2)
T ( x, t )
π ¦ n2
sin nx
n 1
EXAMPLE 3.5 A uniform rod of length L whose surface is thermally insulated is initially
at temperature q = q0. At time t = 0, one end is suddenly cooled to q = 0 and subsequently
maintained at this temperature; the other end remains thermally insulated. Find the temperature
distribution q(x, t).
Solution The initial boundary value problem IBVP of heat conduction is given by
wθ w 2θ
PDE: α , 0 d x d L, t ! 0
wt w x2
BCs: θ (0, t ) 0, tt0
wθ
( L, t ) 0, t!0
wx
IC: θ ( x, 0) θ 0 , 0d xdL
From Section 3.5, it can be noted that the physically meaningful and non-trivial solution is
2
θ ( x, t ) eαλ t ( A cos λ x B sin λ x)
Using the first boundary condition, we obtain A 0. Thus the acceptable solution is
2
θ Beαλ t sin λ x
wθ 2
λ Beαλ t cos λ x
wx
Using the second boundary condition, we have
2
0 λ Beαλ t cos λ L
implying cos λ L 0. Therefore,
The eigenvalues and the corresponding eigenfunctions are
(2n 1) π
λn , n 0, 1, 2, }
2L
Thus, the acceptable solution is of the form
§ 2n 1 ·
θ B exp [α {(2n 1) / 2 L}2 π 2 t ] sin ¨ πx
© 2 L ¸¹
Using the principle of superposition, we obtain
f
§ 2n 1 ·
θ ( x, t ) ¦ Bn exp [α {(2n 1)/2L}2π 2t ] sin ¨© 2L
π x¸
¹
n 0
Finally, using the initial condition, we have

f
§ 2n 1 ·
θ0 ¦ Bn sin ¨© 2L
π x¸
¹
n 0
which is a half-range Fourier-sine series and, thus,

2 L § 2n 1 ·
Bn
L ³0 θ 0 sin ¨© 2L
π x ¸ dx
¹
2ª § 2n 1 · ½ º
L
2L
« θ 0 ® ¨
cos π x ¸¹ ¾ »
L«
¬
(2n 1) π ¯ © 2L ¿ 0 »¼
4θ 0 4θ 0
[cos {(2n 1) π /2} cos 0]
(2n 1) π (2n 1) π
Thus, the required temperature distribution is

f
4θ § 2n 1 ·
θ ( x, t ) ¦ (2n 01)π exp [α {(2n 1)/2L}2π 2t ] sin ¨© 2L
π x¸
¹
n 0
EXAMPLE 3.6 A conducting bar of uniform cross-section lies along the x-axis with ends
at x 0 and x L. It is kept initially at temperature 0° and its lateral surface is insulated. There
are no heat sources in the bar. The end x = 0 is kept at 0°, and heat is suddenly applied at
the end x = L, so that there is a constant flux q0 at x = L. Find the temperature distribution
in the bar for t > 0.
Solution The given initial boundary value problem can be described as follows:
wT w 2T
PDE: α
wt w x2
BCs: T (0, t ) 0, t!0
wT
( L, t ) q0 , t!0
wx
IC: T ( x, 0) 0, 0d xdL
Prior to applying heat suddenly to the end x = L, when t = 0, the heat flow in the bar is
independent of time (steady state condition). Let
T ( x, t ) T( s ) ( x) T1 ( x, t )
where T(s) is a steady part and T1 is the transient part of the solution. Therefore,
w 2T( s )
0
w x2
whose general solution is
T( s ) Ax B
when x 0, T( s ) 0, implying B 0. Therefore,
T( s ) Ax
w T( s )
Using the other BC: q0 , we get A q0 . Hence, the steady state solution is
wx
T( s ) q0 x
For the transient part, the BCs and IC are redefined as

(i) T1 (0, t ) T (0, t ) T( s ) (0) 00 0
(ii) w T1 ( L, t )/w x w T ( L, t )/w x w Ts ( L, t )/w x q0 q0 0
(iii) T1 ( x, 0) T ( x, 0) T( s ) ( x) q0 x, 0 x L.
Thus, for the transient part, we have to solve the given PDE subject to these conditions. The
acceptable solution is given by Eq. (3.48), i.e.
2
T1 ( x, t ) eαλ t ( A cos λ x B sin λ x)
Applying the BC (i), we get A 0. Therefore,

2
T1 ( x, t ) Beαλ t sin λ x
and using the BC (ii), we obtain
w T1 2
Bλ eαλ t cos λ L 0
wx x L
π
implying λ L (2n 1) , n 1, 2,} Using the superposition principle, we have
2
f
§ 2n 1 ·
T1 ( x, t ) ¦ Bn exp [α {(2n 1)/2L}2π 2t ]sin ¨© 2L
π x¸
¹
n 1
Now, applying the IC (iii), we obtain

f
§ 2n 1 ·
T1 ( x, 0) q0 x ¦ Bn sin ¨© 2L
π x¸
¹
n 1
§ 2m 1 ·
Multiplying both sides by sin ¨ π x ¸ and integrating between 0 to L and noting that
© 2L ¹
0, nzm
L § 2n 1 · § 2m 1 · °
³0 Bn sin ¨
© 2 L ¸¹
π x sin ¨
© 2L
π x ¸ dx
¹
® Bm L
° , n m
¯ 2
we get at once, after integrating by parts, the equation
4 L2ª § 2m 1 · º L
q0 «sin ¨ π ¸» Bm
(2m 1) π ¬ © 2
2 2 ¹¼ 2
or
4 L2 L
q0 (1) m 1 Bm
(2m 1) π 2 2 2
which gives
(1)m 8 Lq0
Bm
(2m 1) 2 π 2
Hence, the required temperature distribution is
f ª (1)m
8Lq0 § 2m 1 · º
T ( x, t ) q0 x
π2
¦ ¬«« (2m 1)2 exp [α {(2m 1)/ L}2π 2t ] sin ¨© 2L
π x ¸»
¹ ¼»
m 1
EXAMPLE 3.7 The ends A and B of a rod, 10 cm in length, are kept at temperatures 0°C
and 100°C until the steady state condition prevails. Suddenly the temperature at the end A is
increased to 20°C, and the end B is decreased to 60°C. Find the temperature distribution in
the rod at time t.
Solution The problem is described by
wT w 2T
PDE: α , 0 x 10
wt w x2
BCs: T (0, t ) 0, T (10, t ) 100
Prior to change in temperature at the ends of the rod, the heat flow in the rod is independent
of time as steady state condition prevails. For steady state,
d 2T
0
dx 2
whose solution is
T( s ) Ax B
When x 0, T 0, implying B 0. Therefore,
T( s ) Ax
When x 10, T 100, implying A 10. Thus, the initial steady temperature distribution in
the rod is
T( s ) ( x) 10 x
Similarly, when the temperature at the ends A and B are changed to 20 and 60, the final steady
temperature in the rod is
T( s ) ( x) 4 x 20
which will be attained after a long time. To get the temperature distribution T ( x, t ) in the
intermediate period, counting time from the moment the end temperatures were changed, we
assume that
T ( x, t ) = T1 ( x, t ) + T( s ) ( x)
where T1 ( x, t ) is the transient temperature distribution which tends to zero as t → ∞. Now,

T1 ( x, t ) satisfies the given PDE. Hence, its general solution is of the form
2
T ( x, t ) = (4 x + 20) + e−αλ t ( B cos λ x + c sin λ x)
Using the BC: T = 20 when x = 0, we obtain

2
20 = 20 + Be −αλ t
implying B = 0. Using the BC: T = 60 when x = 10, we get

nπ
sin 10λ = 0, implying λ = , n = 1, 2,…
10
The principle of superposition yields
∞
⎛ nπ ⎞
T ( x, t ) = (4 x + 20) + ∑ cn exp [−α (nπ /10)2 t ]sin ⎜⎝ 10 ⎟⎠ x
n =1
Now using the IC: T = 10x, when t = 0, we obtain
⎛ nπ ⎞
10 x = 4 x + 20 + ∑ cn sin ⎜⎝ 10 x ⎟⎠
or
⎛ nπ ⎞
6 x − 20 = ∑ cn sin ⎝⎜ 10 x ⎠⎟
where
2 10 ⎛ nπ ⎞ 1 ⎡ n 800 200 ⎤
cn =
10 ∫0 (6 x − 20) sin ⎜
⎝ 10
x ⎟ dx = −
⎠ 5 ⎢⎣ (−1) nπ − nπ ⎥⎦
Thus, the required solution is

∞ ⎡ ⎛ nπ ⎞ 2 ⎤
1 ⎡ n 800 200 ⎤ ⎛ nπ ⎞
T ( x, t ) = 4 x + 20 − ∑
5 n = 1 ⎢⎣
( −1)
nπ
−
nπ ⎥⎦
exp ⎢ −α ⎜ ⎟ t ⎥ sin ⎜
⎢⎣ ⎝ 10 ⎠ ⎥⎦ ⎝ 10
x⎟ .
⎠
EXAMPLE 3.8 Assuming the surface of the earth to be flat, which is initially at zero
temperature and for times t > 0, the boundary surface is being subjected to a periodic heat
flux g0 cos ωt. Investigate the penetration of these temperature variations into the earth’s
surface and show that at a depth x, the temperature fluctuates and the amplitude of the steady
temperature is given by
g0 2α
exp [ (ω /2α ) x ]
2 ω
Solution The given IBVP is described by
wT w 2T
PDE: α (3.49)
wt w x2
wT
BC: g 0 cos ω t at x 0, t ! 0 (3.50)
wx
IC: T ( x, 0) 0 (3.51)
We shall introduce an auxiliary function T satisfying Eqs. (3.49)–(3.51) and then define the
complex function Z such that
Z T iT
We can easily verify that Z satisfies
wZ w 2Z
PDE: α (3.52)
wt w x2
wZ
BC: g0 eiω t at x 0, t ! 0
wx
IC: Z 0 in the region, t 0
Let us assume the solution of Eq. (3.52) in the form
Z f ( x) eiω t
where f (x) satisfies
d 2 f ( x) ω
i f ( x) 0 (3.53)
dx 2 α
df ( x)
g0 at x 0 (3.54)
dx
Also,
f ( x) is finite for large x. (3.55)
The solution of Eq. (3.53), satisfying the BC (3.55), is
f ( x) A exp [ (iω /α ) x ]
The constant A can be determined by using the BC (3.54). Therefore,
1 α
f ( x) g0 exp [ (iω / α ) x]
i ω
Thus,
α 1
Z g0 exp [iω t (iω / α ) x] (3.56)
ω i
It can be shown for convenience that
1 i 1 1 i
i ,
2 i 2
Thus, Eq. (3.56) can be written in the form
g0 2α § ω · ª § ω ·º
Z exp ¨ x ¸ (1 i ) exp «i ¨ω t x ¸»
2 ω © 2α ¹ ¬ © 2α ¹¼
g0 2α § ω · ª § ω · § ω ·º
exp ¨ x ¸ (1 i) «cos ¨ω t x ¸ i sin ¨ω t x »
2 ω © 2α ¹ ¬ © 2α ¹ © 2α ¸¹ ¼
Its real part gives the fluctuation in temperature is
g0 2α § ω ·ª § ω · § ω ·º
T ( x, t ) exp ¨ x ¸ «cos ¨ω t x ¸ sin ¨ω t x ¸»
2 ω © 2α ¹ ¬ © 2α ¹ © 2α ¹¼
g0 2α § ω · § ω π·
exp ¨ x ¸ cos ¨ω t x ¸
2 ω © 2α ¹ © 2α 4¹
Hence, the amplitude of the steady temperature is given by the factor

g0 2α § ω ·
exp ¨ x
2 ω © 2α ¹¸
EXAMPLE 3.9 Find the solution of the one-dimensional diffusion equation satisfying the
following BCs:
(i) T is bounded as t o f
wT
(ii) 0, for all t
wx x 0
wT
(iii) 0, for all t
wx x a
(iv) T ( x, 0) x (a x), 0 x a.
Solution This is an example with insulated boundary conditions. From Section 3.5, it
can be seen that a physically acceptable general solution of the diffusion equation is
T ( x, t ) exp (αλ 2t ) ( A cos λ x B sin λ x)

Thus,
wT
exp (αλ 2 t ) ( Aλ sin λ x Bλ cos λ x) (3.57)
wx
Using BC (ii), Eq. (3.57), gives B = 0. Since we are looking for a non-trivial solution, the use
of BC (iii) into Eq. (3.57) at once gives
sin λ a 0 implying λ a nπ , n 0, 1, 2,}
Using the principle of superposition, we get
f ª § nπ · 2 º § nπ ·
T ( x, t ) ¦ An exp (αλ 2t ) cos λ x ¦ An exp « α ¨ ¸ t » cos ¨ ¸ x.
«¬ © a ¹ »¼ © a ¹
n 0
The boundary condition (iv) gives

f ª § nπ · 2 º § nπ ·
T ( x, 0) x (a x) A0 ¦ An exp « α ¨ ¸ t » cos ¨ ¸ x
«¬ © a ¹ »¼ © a ¹
n 1
where
2 a a2
A0
a ³0 (ax x 2 ) dx
6
2 a § nπ ·
³0 (ax x
2
An ) cos ¨ x ¸ dx
a © a ¹
2a 2 2a 2
(1 cos nπ ) [1 (1)n ]
n 2π 2 n 2π 2
Therefore,
4a 2
° 2 2 , for n even
An ® nπ
°
¯ 0, for n odd

f ª § nπ · 2 º
a 2 4a 2 1 § nπ ·
T ( x, t )
6
2
π
¦ n2
cos
©¨ a ¹¸
x exp « α ¨ ¸ t »
«¬ © a ¹ »¼
n 2, 4,} even
EXAMPLE 3.10 The boundaries of the rectangle 0 d x d a, 0 d y d b are maintained at zero

temperature. If at t = 0 the temperature T has the prescribed value f (x, y), show that for
t > 0, the temperature at a point within the rectangle is given by
f f
4 mπ x nπ y
T ( x, y , t )
ab ¦ ¦ f (m, n) exp (αλ mn2 t ) sin a
sin
b
m 1 n 1
where
a b mπ x nπ y
f ( m, n ) ³0 ³0 f ( x, y ) sin
a
sin
b
dx dy
and
§ m2 n2 ·
λ mn
2
π2 ¨ 2 2 ¸
©a b ¹
Solution The problem is to solve the diffusion equation described by
wT § w 2T w 2T ·
PDE: α ¨ 2 2 ¸, 0 x a, 0 y b, t ! 0
wt ©w x wy ¹
BCs: T (0, y , t ) T (a, y, t ) 0, 0 y b, t ! 0
T ( x, 0, t ) T ( x, b, t ) 0, 0 x a, t ! 0
IC: T ( x, y, 0) f ( x, y ), 0 x a, 0 y b
Let the separable solution be

T X ( x) Y ( y ) β (t )
Substituting into PDE, we get
X cc Y cc 1 βc
λ 2
X Y α β
Then β c αλ 2 β 0
X cc § Y cc ·
¨λ 2 ¸ p 2 (say)
X © Y ¹
Hence,
X cc p 2 X 0
Y cc
λ 2 p 2 q 2 (say)
Y
Therefore,
Y cc q 2Y 0
Thus, the general solution of the given PDE is
2
T ( x, y , t ) ( A cos px B sin px) (c cos qy D sin qy ) eαλ t
where
λ2 p2 q2
Using the BC: T (0, y, t ) 0, we get A = 0. Then, the solution is of the form
2
T ( x, y , t ) B sin px (c cos qy D sin qy ) eαλ t
Applying the BC: T ( x, 0, t ) 0, we get c = 0. Thus, the solution is given by

2
T ( x, y , t ) BD sin px sin qyeαλ t
Application of the BC: T (a, y, t ) 0 gives
sin pa 0, implying pa nπ
or
nπ
, n 1, 2,}p
a
Using the principle of superposition, the solution can be written in the form
f
§ nπ ·
¦ An sin ¨© a
2
T ( x, y , t ) x ¸ sin qyeαλ t
¹
n 1
Using the last BC: T ( x, b, t ) 0, we obtain

mπ
q , m 1, 2,}
b
Thus, the solution is found to be
f f
§ nπ · § mπ · αλ 2t
T ( x, y , t ) ¦ ¦ Amn sin ¨© a x ¸ sin ¨
¹ © b ¸¹
y e
m 1 n 1
where
§ m2 n2 ·
λ2 p2 q2 π2 ¨ 2 2 ¸
©b a ¹
Finally, using the IC, we get
§ nπ · § mπ ·
T ( x, y, 0) f ( x, y ) ¦ Amn sin ¨© a x ¸ sin ¨
¹ © b ¸¹
y
which is a double Fourier series, where
2 2 a b § mπ · § nπ ·
Amn
a b ³0 ³0 f ( x, y ) sin ¨
© a
x ¸ sin ¨
¹ © b
y ¸ dx dy
¹
Hence, the required general solution is

f f
§ nπ · § mπ ·
¦¦
2
f (m, n) eαλ t sin ¨
© a ¸¹
x sin ¨
© b ¸¹
T ( x, y , t ) y
m 1 n 1
where
4 a b ⎛ mπ ⎞ ⎛ nπ ⎞
f ( m, n ) =
ab ∫0 ∫0 f ( x, y ) sin ⎜
⎝ a
x ⎟ sin ⎜
⎠
y ⎟ dx dy
⎝ b ⎠
and
§ m2 n2 ·
λ2 π2 ¨ 2 2 ¸
©b a ¹
3.6 SOLUTION OF DIFFUSION EQUATION IN CYLINDRICAL

COORDINATES
Consider a three-dimensional diffusion equation
wT
α 2T
wt
In cylindrical coordinates (r , θ , z ), it becomes
1 wT w 2T 1 w T 1 w 2T w 2T (3.58)

α wt w r2 r w r r 2 wθ 2 w z2
where T T (r , θ , z, t ).
Let us assume separation of variables in the form
T (r , θ , z , t ) R(r ) H (θ ) Z ( z ) β (t )
Substituting into Eq. (3.58), it becomes

1 1 βc
R ccHZ β R cHZ β 2 H ccRZ β Z ccRH β RHZ
r r α
or
R cc 1 R c 1 H cc Z cc 1 βc
λ 2
R r R r2 H Z α β
where λ 2 is a separation constant. Then
β c αλ 2 β 0 (3.59)
R cc 1 R c 1 H cc Z cc
λ2 μ 2 (say)
R r R r2 H Z
Thus, the equations determining Z, R and H become
Z cc μ 2 Z 0 (3.60)
R cc 1 R c 1 H cc
λ2 μ2 0
R r R r2 H
or
R cc Rc H cc
r2 r (λ 2 μ 2 ) r 2 v 2 (say)
R R H
Therefore,
H cc v 2 H 0 (3.61)
1 ª v2 º
R cc R c «(λ 2 μ 2 ) 2 » R 0 (3.62)
r ¬« r ¼»
Equations (3.59)–(3.61) have particular solutions of the form
2
β eαλ t
H c cos vθ D sin vθ
Z Ae μ z Be μ z
The differential equation (3.62) is called Bessel’s equation of order v and its general solution
is known as
R (r ) c1 J v ( λ 2 μ 2 r ) c2Yv ( λ 2 μ 2 r )
where Jν (r ) and Yν ( r ) are Bessel functions of order ν of the first and second kind, respectively.
Of course, Eq. (3.62) is singular when r = 0. The physically meaningful solutions must be twice
continuously differentiable in 0 d r d a. Hence, Eq. (3.62) has only one bounded solution, i.e.
R (r ) Jν ( λ 2 μ 2 r )
Finally, the general solution of Eq. (3.58) is given by
2
T (r , θ , z , t ) e αλ t [ Ae μ z Be μ z ] [C cos νθ D sin νθ ] Jν ( λ 2 μ 2 r )
EXAMPLE 3.11 Determine the temperature T (r , t ) in the infinite cylinder 0 d r d a when

the initial temperature is T (r , 0) f (r ), and the surface r = a is maintained at 0° temperature.
Solution The governing PDE from the data of the problem is

wT
α 2T
wt
where T is a function of r and t only. Therefore,
w 2T 1 wT 1 wT (3.63)

wr 2 r wr α wt
The corresponding boundary and initial conditions are given by
BC: T (a, t ) 0 (3.64)
IC: T (r , 0) f (r )
The general solution of Eq. (3.63) is
T (r , t ) A exp (αλ 2 t ) J 0 (λ r )
Using the BC (3.64), we obtain
J 0 (λ a ) 0
which has an infinite number of roots, ξ n a (n 1, 2, } , f). Thus, we get from the superposition
principle the equation
f
T (r , t ) ¦ An exp (αξ n2t ) J0 (ξn r )
n 1
Now using the IC: T (r , 0) f (r ), we get

f
f (r ) ¦ An J 0 (ξn r )
n 1
To compute An, we multiply both sides of the above equation by rJ 0 (ξ m r ) and integrate with
respect to r to get
e
± 0
a
rf (r ) J 0 (Ym r ) dr ¥ A ±
n 1
n
a
0
rJ 0 (Ym r ) J 0 (Yn r ) dr
¬0 for n y m
¯¯
¦ a2 µ
¯ Am § ¶ J 12 (Ym a) for n m
¯® ¨ 2 ·
which gives
2 a
Am
a 2 J 12 (ξ m a ) ³0 uf (u ) J 0 (ξ mu ) du
Hence, the final solution of the problem is given by
f
2 J 0 (ξ m r ) ª a º
T (r , t )
a 2 ¦ J 1 (ξ m a)
2
exp (αξ m2 t ) «
¬ ³0 uf (u) J0 (ξmu) du »¼
m 1
3.7 SOLUTION OF DIFFUSION EQUATION IN SPHERICAL

COORDINATES
In this section, we shall examine the solution of diffusion or heat conduction equation in the
spherical coordinate system. Let us consider the three-dimensional diffusion Eq. (3.9), and let
T T (r , θ , φ , t ). In the spherical coordinate system, Eq. (3.9) can be written as
w 2T 2 wT 1 w § wT · 1 w 2T 1 wT
2 ¨© sin θ ¸¹ 2 2 (3.65)
wr 2 r w r r sin θ w θ wθ r sin θ w φ 2 α wt
This equation is separated by assuming the temperature function T in the form
T R(r ) H (θ ) Φ (φ ) β (t ) (3.66)
R cc 2 R c 1 1 d § dH · 1 d 2Φ 1 βc
¨© sin θ ¸¹ λ 2 (say)
R r R r 2 sin θ H dθ dθ Φr 2 sin 2 θ d φ 2 α β
where λ 2 is a separation constant. Thus,

dβ
λ 2αβ 0
dt
whose solution is
2
β c1eαλ t
(3.67)
Also,
ª 1 § d 2 R 2 dR · 1 d § dH · º 1 d 2Φ
r 2 sin 2 θ « ¨ 2 ¨ sin θ ¸ λ 2
» m2 (say)
¬« R © dr r dr ¸¹ Hr 2 sin θ dθ © dθ ¹ ¼» Φ dφ 2
which gives
d 2Φ
m2Φ 0
dφ 2
whose solution is
Φ (φ ) c1eimφ c2 eimφ (3.68)

Now, the other separated equation is
1 § d 2 R 2 dR · 1 d § dH · m2
¨© sin θ ¸λ
2
¨ ¸
R © dr 2 r dr ¹ Hr sin θ dθ
2 dθ ¹ r 2 sin 2 θ
or
r2 § 2 · m2 1 d § dH ·
¨© R cc R c ¸¹ λ r ¨© sin θ
2 2
¸
R r sin θ
2 H sin θ dθ dθ ¹
n (n 1) (say)
On re-arrangement, this equation can be written as
2 n (n 1) ½
R cc R c ®λ 2 ¾R 0 (3.69)
r ¯ r2 ¿
and
1 § d 2H dH · m2
¨ sin θ cos θ ¸ n (n 1)
H sin θ © dθ 2 dθ ¹ sin 2 θ
or
d 2H dH ° m 2 ½°
cot θ ®n (n 1) 2 ¾ H 0 (3.70)
dθ 2 dθ ¯° sin θ ¿°
Let R (λ r )1/2ψ (r ); then Eq. (3.69) becomes
ª 1 ° (n 1/2)2 °½ º
λ r )1/2 «ψ cc (r ) ψ c (r ) ®λ 2 ¾ψ » 0
«¬ r °¯ r2 °¿ »¼
Since (λ r ) z 0, we have
1 ° (n 1/2)2 °½
ψ cc (r ) ψ c (r ) ®λ 2 ¾ ψ (r ) 0
r °¯ r2 °¿
which is Bessel’s differential equation of order (n 1/2), whose solution is
ψ (r ) AJ n 1/2 (λ r ) BYn 1/2 (λ r )
Therefore,
R(r ) (λ r )1/2 [ AJ n 1/2 (λ r ) BYn 1/2 (λ r )] (3.71)

where Jn and Yn are Bessel functions of first and second kind, respectively. Now, Eq. (3.70)
can be put in a more convenient form by introducing a new independent variable
μ cos θ
so that
cot θ μ/ 1 μ2
dH dH
1 μ2
dθ dμ
d 2H d 2H dH
(1 μ 2 ) μ
dθ 2
dμ 2 dμ
Thus, Eq. (3.70) becomes
d 2H dH ª m2 º
(1 μ 2 ) 2μ « n (n 1) »H 0 (3.72)
dμ2 d μ «¬ 1 μ 2 »¼
which is an associated Legendre differential equation whose solution is
H (θ ) Ac Pnm ( μ ) B cQ mn ( μ ) (3.73)
where Pnm ( μ ) and Q mn ( μ ) are associated Legendre functions of degree n and of order m, of
first and second kind, respectively. Hence the physically meaningful general solution of the
diffusion equation in spherical geometry is of the form
¦
2
T (r , θ , φ , t ) Aλ mn (λ r )1/2 J n 1/2 (λ r ) Pnm (cos θ ) erimφ αλ t
(3.74)
λ , m, n
In this general solution, the functions Q mn ( μ ) and (λ r )1/2 Yn1/2 (λ r ) are excluded because
these functions have poles at μ r1 and r = 0 respectively.
EXAMPLE 3.12 Find the temperature in a sphere of radius a, when its surface is kept at
zero temperature and its initial temperature is f (r , θ ).
Solution Here, the temperature is governed by the three-dimensional heat equation in

spherical polar coordinates independent of φ . Therefore, the task is to find the solution of
PDE
1 wT w 2T 2 wT 1 w § wT ·
2 ¨© sin θ ¸ (3.75)
α wt wr 2 r w r r sin θ w θ wθ ¹
subject to
BC: T (a, θ , t ) 0 (3.76)

IC: T (r , θ , 0) f (r , θ ) (3.77)
The general solution of Eq. (3.75), with the help of Eq. (3.74), can be written as
¦ Aλn (λ r )1/2 J n1/2 (λ r ) Pn (cos θ )eαλ t

2
T (r , θ , t ) (3.78)
λ, n
Applying the BC (3.76), we get

J n 1/2 (λ a ) 0
This equation has infinitely many positive roots. Denoting them by ξi , we have
f f
T (r , θ , t ) ¦ ¦ Ani (ξi r )1/2 J n1/2 (ξi r ) Pn (cos θ ) exp (αξ 2i t ) (3.79)
n 0 i 1
Now, applying the IC and denoting cos θ by μ , we get

f f
f (r , cos 1 ( μ )) ¦ ¦ Ani (ξi r )1/2 J n1/2 (ξi r ) Pn (μ ).
n 0 i 1
Multiplying both sides by Pm ( μ ) d μ and integrating between the limits, –1 to 1, we obtain
1 f f 1
³1 f (r , cos 1 ( μ )) Pm ( μ ) d μ ¦ ¦ Ani (ξi r )1/2 J n1/2 (ξi r ) ³1 Pm (μ ) Pn (μ ) d μ
n 0 i 1
f f
§ 2 ·
¦ ¦ Ani (ξi r )1/2 J n1/2 (ξi r ) ¨© 2n 1 ¸¹
n 0 i 1
or
f
§ 2n 1 · 1
¨©
2 ¹
¸ ³1 Pn ( μ ) f (r , cos 1 ( μ )) d μ ¦ Ani (ξi r )1/2 J n1/2 (ξi r ) for n 0, 1, 2, 3,}
i 1
Now, to evaluate the constants A ni , we multiply both sides of the above equation
by r 3/2 J n 1/2 (ξ j r ) and integrate with respect to r between the limits 0 to a and use the
orthogonality property of Bessel functions to get
§ 2n 1 · a ª 1 º
³0 r ³1 Pn (μ ) f (r, cos
1
ξ 1/2
i ¨
3/2
J n 1/2 (ξ j r ) dr « ( μ )) d μ »
© 2 ¸¹ ¬ ¼
f a
¦ Ani ³0 rJ n1/2 (ξi r ) J n1/2 (ξ j r ) dr
i 1
f
1
2i¦ Ani [ J cn1/2 (ξi r )]2 , n 0, 1, 2, 3,} (3.80)
1
Thus, Eqs. (3.79) and (3.80) together constitutes the solution for the given problem.
3.8 MAXIMUM-MINIMUM PRINCIPLE AND CONSEQUENCES
Theorem 3.1 (Maximum-minimum principle).

Let u ( x, t ) be a continuous function and a solution of
ut α u xx (3.81)
for 0 d x d l , 0 d t d T , where T > 0 is a fixed time. Then the maximum and minimum values
of u are attained either at time t = 0 or at the end points x = 0 and x = l at some time in the
interval 0 d t d T .
Proof To start with, let us assume that the assertion is false. Let the maximum value
of u ( x, t ) for t 0 (0 d x d l ) or for x = 0 or x = l (0 d t d T ) be denoted by M. We shall assume
that the function u ( x, t ) attains its maximum at some interior point (x0, t0), in the rectangle
defined by 0 d x0 d l , 0 d t0 d T , and then arrive at a contradiction. This means that
u ( x0 , t0 ) M ε (3.82)
Now, we shall compare the signs in Eq. (3.81) at the point (x0, t0). It is well known from
calculus that the necessary condition for the function u(x, t) to possess maximum at (x0, t0) is
wu w 2u
( x0 , t0 ) 0, ( x0 , t0 ) d 0 (3.83)
wx w x2
In addition, u ( x0 , t0 ) attains maximum for t = t0, implying
wu
( x0 , t0 ) t 0 (3.84)
wt
Thus, with the help of Eqs. (3.83) and (3.84) we observe that the signs on the left- and right-
hand sides of Eq. (3.81) are different. However, we cannot claim that we have reached a
contradiction, since the left- and right-hand sides can simultaneously be zero.
To complete the proof, let us consider another point (x1, t1) at which w 2 u /w x 2 d 0 and
w u /w t ! 0.
Now, we construct an auxiliary function
v ( x, t ) u ( x, t ) λ (t0 t ) (3.85)
where λ is a constant. Obviously, v ( x0 , t0 ) u ( x0 , t0 ) M ε and λ (t0 t ) d λT . Suppose

we choose λ ! 0, such that λ ε /2T ; then the maximum of v ( x, t ) for t 0 or
for x 0, x l cannot exceed the value M ε /2. But v ( x, t ) is a continuous function and,
therefore, a point ( x1 , t1 ) exists at which it assumes its maximum. It implies
M ε /2 d v ( x1 , t1 ) t v ( x0 , t0 ) M ε
This pair of inequalities is inconsistent and therefore contradicts the assumption that v
takes on its maximum at ( x0 , t0 ). Therefore, the assertion that u attains its maximum either
at t 0 or at the end points is true.
We can establish a similar result for minimum values of u(x, t). If u satisfies Eq. (3.81),
–u also satisfies Eq. (3.81). Hence, both maximum and minimum values are attained either
initially or at the end points. Thus the proof is complete. We shall give some of the consequences
of the maximum-minimum principle in the following theorems.
Theorem 3.2 (Uniqueness theorem). Given a rectangular region defined by 0 d x d l , 0 d t d T ,

and a continuous function u(x, t) defined on the boundary of the rectangle satisfying the heat
equation
ut α u xx
This equation possesses one and only one solution satisfying the initial and boundary conditions
u ( x, 0) f ( x)
u (0, t ) g1 (t ), u (l , t ) g 2 (t )
where f (x), g1(t), g2(t) are continuous on their domains of definition.
Proof Suppose there are two solutions u1 ( x, t ), u2 ( x, t ) satisfying the heat equation as
well as the same initial and boundary conditions. Now let us consider the difference
v ( x, t ) = u2 ( x, t ) − u1 ( x, t )
It is also a solution of the heat conduction equation for 0 ≤ x ≤ l , 0 ≤ t ≤ T and is continuous

in x and t. Also, v ( x, t ) = 0, 0 ≤ x ≤ l and v (0, t ) = v (l , t ) = 0, 0 ≤ t ≤ T . Hence, v ( x, t ) satisfies
the conditions required for the application of maximum-minimum principle. Thus, v ( x, t ) = 0 in
the rectangular region defined by 0 ≤ x ≤ l , 0 ≤ t ≤ T . It follows therefore that u1 ( x, t ) = u2 ( x, t ).
Another important consequence of the maximum-minimum principle is the stability property
which is stated in the following theorem without proof.
Theorem 3.3 (Stability theorem). The solution u ( x, t ) of the Dirichlet problem
ut = α u xx , 0 ≤ x ≤ l, 0 ≤ t ≤ T
u ( x, 0) = f ( x), 0≤ x≤l
u (0, t ) = g (t ), u (l , t ) = h(t ), 0 ≤ t ≤ T
depends continuously on the initial and boundary conditions.
3.9 NON-LINEAR EQUATIONS (MODELS)

Today various studies of fluid behaviour are available which encompass virtually any type of
phenomena of practical importance. However, there are many unresolved important problems
in fluid dynamics due to the non-linear nature of the governing PDEs and due to difficulties
encountered in many of the conventional, analytical and numerical techniques in solving
them. In the following, we shall present few non-linear model equations to have a feel for this
vast field of study.
3.9.1 Semilinear Equations

Reaction–diffusion equations that appear in the literature are frequently semilinear and are of
the form ut = —2u + f(u, x, t).
Typically, they appear as models in population dynamics, with inhomogeneous term
depending on the density of local population. In chemical engineering, f varies with temperature
and/or chemical concentration in a reaction like f(u) = luN.
3.9.2 Quasi-linear Equations

Many problems in fluid mechanics, when formulated mathematically, give rise to quasi-linear
parabolic PDEs. A simple example concerns the flow of compressible fluid through a porous
medium. Let r denotes fluid density. Following Darcy’s law, which relates the velocity V to
the pressure p as
ÊKˆ
V = - Á ˜ —p .
Ëm¯
Then, the equation of conservation of mass is given by

∂r
+ — ◊ ( rV ) = 0
∂t
and the equation of state p = p(r) can be combined to get
∂r
+ — ◊ [ K ( r )— r ]
∂t
dp
where K(r) is proportional to r . The model can then be written as the porous medium
dr
equation in the form
∂r
+ — ◊ ( r n —r )
∂t
where n is a positive constant.
3.9.3 Burger’s Equation

The well-known Burger’s equation is non-linear and finding its solution has been the subject
of active research for many years. For simplicity, let us consider one-dimensional Burger’s
equation in the form
ut + uux = vuxx
Ê 1 ˆ
or ut - Á vux - u2 ˜ = 0 (3.85a)
Ë 2 ¯x
which is actually the non-linear momentum equation in fluid mechanics without the pressure
term. v is the physical viscosity. Here vuxx measures dissipative term and uux measures
convective term, while ut is the unsteady term. Hopf (1950) and Cole (1951) gave independently
the analytical solution for a model problem using a two-step Hopf–Cole transformation described
by
u(x, t) = yx
y = –2v log f(x, t)
That is,
fx
u = - 2v (3.85b)
f
Thus,
Êf ˆ ff
ut = - 2 v Á xt ˜ + 2 v x 2 t ,
Ë f ¯ f
ÈI Ø È I2 Ø
ux 2 v É xx Ù 2 v É x2 Ù
Ê I Ú ÊI Ú
ÈI Ø ÈI I Ø È I3 Ø
uxx 2 v É xxx Ù 6v É x 2xx Ù 4 v É x3 Ù .
Ê I Ú Ê I Ú ÊI Ú
Inserting, these derivative expressions into Eq. (3.85a) and on simplification, we arrive at
Ix
(vfxx – ft) – (vfxx – ft)x = 0. (3.85c)
I
Therefore, we have to solve Eq. (3.85c) to find f(x, t), and using this result in Eq. (3.85b),
we obtain an expression for u(x, t) which of course satisfies Eq. (3.85a). Thus, if f(x, t)
satisfies heat conduction equation
ft = vfxx (3.85d)
which also means solving trivially Eq. (3.85c). This is also called linearised Burger’s equation.
Equivalently, we may introduce the transformation:
\x u, Þ
Ñ
u2 ß (3.85e)
\t vux Ñ
2à
in such a way, satisfying that yxt = ytx. Then, the above transformation can be rewritten as
\ x2
yt = vyxx – (3.85f)
2
Also, Eq. (3.85b) can be recast in the form
f(x, t) = e[–y(x, t)/2v] (3.85g)
Thus, knowing f(x, t), we can find u(x, t) from Eq. (3.85b). It may also be observed that
Eqs. (3.85d) and (3.85f) are equivalent.
Hence, the transformation of non-linear Burger’s equation into heat conduction equation,
made life easy to get analytical solution to the Burger’s equation.
3.9.4 Initial Value Problem for Burgers Equation

The IVP for Burger’s equation can be stated as follows. Solve
PDE: ut + uux = vuxx, – ¥ < x < ¥, t > 0
IC: u(x, 0) = f(x) (3.85h)
Under the transformation defined by Eq. (3.85b) and using (3.85g), the given IVP can be
restated as a Cauchy problem, described by
PDE: ft = vfxx,
Ë 1 x Û
Ì 2v
Í Ô0 f (K ) dK Ü
Ý
IC: f(x, 0) = I ( x ) e (3.85i)
Using, the standard, separation of variables, method of solution, as given in Eq. (3.28), the
solution to Eq. (3.85i) is found to be
1 Ë ( x [ )2 Û
I ( x, t )
4S vt Ô
I ( x )exp Ì
ÍÌ 4 vt ÝÜ
Ü d[ (3.85j)
Substituting the expression for I ( x ) , the above equation can be rewritten as
1 Ë f ([ , x, t ) Û
4S vt Ô
I ( x, t ) exp Ì d[
Í 2 v ÜÝ
where,
( x [ )2
[
f ([ , x , t )
0 Ô 2t
f (D )dD
.
Finally, using Eq. (3.85b), the exact solution of the IVP for Burger’s equation as stated
in Eq. (3.85h) is found to be
Ë (x [) Î f ([ , x, t ) Þ Û
Ì
Í
Ô t
exp Ï
Ð
ß d[
2 v à ÜÝ
u( x, t ) (3.85k)
Ë f ([ , x, t ) Û
exp Ì d[
Í 2 v ÜÝ
Here, the function f(x, x, t) is known as Hopf–Cole function.
3.10 MISCELLANEOUS EXAMPLES

EXAMPLE 3.13 A homogeneous solid sphere of radius R has the initial temperature distribution
f (r ), 0 d r d R, where r is the distance measured from the centre. The surface temperature
is maintained at 0°. Show that the temperature T (r , t ) in the sphere is the solution of
§ 2 ·
Tt c 2 ¨Trr Tr ¸
© r ¹
where c2 is a constant. Show also that the temperature in the sphere for t > 0 is given by
f
1 § nπ · cnπ
T (r , t )
rn ¦ Bn sin ¨© R r ¸¹ exp (λ 2nt ), λn
R
1
Solution The temperature distribution in a solid sphere is governed by the parabolic

heat equation
Tt c 2 2T
From the data given, T is a function of r and t alone. In view of the symmetry of the sphere,
the above equation with the help of Eq. (3.65) reduces to
§ 2 ·
Tt c 2 ¨Trr Tr ¸ (3.86)
© r ¹
Setting v rT , the given BC gives

v ( R, t ) rT ( R, t ) 0
while the IC gives
v (r , 0) rT (r , 0) rf (r )
Since T must be bounded at r 0, we require
v (0, t ) 0
Now,
§v· vr r v
vt rTt , Tr ¨© ¸¹
r r r2
Similarly, finding Trr and substituting into Eq. (3.86), we obtain
vt c 2 vrr
Using the variables separable method, we may write v (r , t ) R(r ) τ (t ) and get
R (r ) A cos kr B sin kr
τ (t ) exp ( c 2 k 2t )
Thus, using the principle of superposition, we get
f
v (r , t ) ¦ ( An cos kr Bn sin kr ) exp (c2 k 2t )
n 1
Also, using v (0, t ) 0, we have
( An cos kr Bn sin kr ) | r 0 0
implying An 0. Also, v ( R, t ) 0 gives Bn sin kR = 0, implying sin kR = 0, as Bn z 0. Therefore,

nπ
kR nπ , k , n 1, 2, }
R
Thus, the possible solution is

f § c 2 n2π 2t ·
§ nπ ·
v (r , t ) ¦ Bn sin ¨© R r ¸¹ exp ¨© R2 ¹
¸
n 1
Finally, applying the IC: v (r , 0) rf (r ), we get
f
§ nπ ·
rf (r ) ¦ Bn sin ¨© R r ¸¹
n 1
which is a half-range Fourier series. Therefore,
2 R § nπ ·
Bn
R ³0 rf (r )sin ¨
© R ¸¹
r dr
But v (r , t ) rT (r , t ). Hence, the temperature in the sphere is given by

f § c 2 n2π 2t ·
1 § nπ ·
T (r , t )
rn ¦ Bn sin ¨© R r ¸¹ exp ¨© R2 ¹
¸
1
EXAMPLE 3.14 A circular cylinder of radius a has its surface kept at a constant temperature
T0. If the initial temperature is zero throughout the cylinder, prove that for t > 0.
° 2
f
J 0 (ξ n a) ½°
T (r , t ) T0 ® 1
°¯ a ¦ ξ J (ξ a)
exp (ξ n2 kt )¾
°¿
n 1 n 1 n
where rξ1 , r ξ 2 ,} , r ξ n are the roots of J 0 (ξ a ) 0, and k is the thermal conductivity which
is a constant.
Solution It is evident that T is a function of r and t alone and, therefore, the PDE to
be solved is
w 2T 1 wT 1 wT (3.87)

wr 2 r wr k wt
subject to
IC : T (r , 0) 0, 0dra
BC : T (a, t ) T0 , tt0
Let
T ( r , t ) T0 T1 ( r , t )
so that
T1 (r , 0) T0 (3.88)
T1 (a, t ) 0 (3.89)
where T1 is the solution of Eq. (3.87). By the variables separable method we have (see
Example 3.11),
T1 (r , t ) AJ 0 (λ r ) exp (λ 2 kt )
Using the BC: T1 ( a, t ) 0, we get
AJ 0 (λ a ) exp (λ 2 kt ) 0
which gives J 0 (λ a ) 0 as A z 0. Let ξ1 , ξ 2 ,} , ξ n , be the roots of J 0 (λ a ) 0. Then the

possible solution using the superposition principle is
f
T1 (r , t ) ¦ An J0 (ξn r ) exp (ξ 2n kt ) (3.90)
n 1
Using the IC: T1 ( r , 0) T0 into Eq. (3.90), we obtain

f
¦ An J 0 (ξn r ) T0
n 1
Multiplying both sides by rJ 0 (ξ m r ) and integrating, we get
a f a
T0 ³0 rJ 0 (ξ m r ) dr ¦ An ³0 rJ 0 (ξm r ) J 0 (ξn r ) dr
n 1
a
³0 rJ 0 (ξm r )
2
Am if m n; otherwise 0
a2 2
Am J 1 (ξ m a)
2
But,
a ξma x dx
T0 ³0 rJ 0 (ξ m r ) dr T0 ³0 ξm
J 0 ( x)
ξm
( x ξm r )
T0 ξm a d

ξ 2m ³0 dx
[ xJ1 ( x)] dx
T0 ξ a aT0
[ xJ1 ( x)] 0m J (ξ a )
ξ 2m ξm 1 m
Therefore,
a2 2 aT0
Am J 1 (ξ m a) J (ξ a )
2 ξm 1 m
or
2T0 1
An
a[ n J1 ([ n a )
Hence, Eq. (3.90) becomes
f
J 0 (ξ n r ) exp (ξ n kt )
2
2
T1 (r , t ) T0
a ¦ J1 (ξ n a) ξn
n 1
Finally, the complete solution is found to be

ª 2 f
J 0 (ξ n r ) exp (ξ n kt ) º
2
T (r , t ) T0 «1
«¬ a
¦ J1 (ξ n a) ξn
»
»¼
n 1
EXAMPLE 3.15 Determine the temperature in a sphere of radius a, when its surface is
maintained at zero temperature while its initial temperature is f (r , θ ).
Solution Here the temperature is governed by the three-dimensional heat equation in

polar coordinates independent of φ , which is given by
1 wu w 2u 2 wu 1 w § wu ·
¨ sin θ ¸ (3.91)
k wt wr 2 r w r r 2 sin θ w θ © wθ ¹
Let
u (r , θ , t ) R(r ) H (θ ) T (t )
By the variables separable method (see Section 3.7), the general solution of Eq. (3.91) is
found to be
u (r , θ , t ) ¦ ¦ Aλn (λ r )1/2 J n1/2 (λ r ) Pn (cos θ ) exp (k λ 2t ) (3.92)

λ n
In the present problem, the boundary and initial conditions are
BC: u (a, θ , t ) 0 (3.93)

IC: u (r , θ , 0) f (r , θ ) (3.94)
Substituting the BC (3.93) into Eq. (3.92), we get
J n 1/2 (λ a ) 0 (3.95)
Let ξ1a, ξ 2 a,… , ξ i a,… be the roots of Eq. (3.95). Then the general solution can be put in the form
∞ ∞
u (r , θ , t ) = ∑ ∑ Ani (ξi r )−1/2 J n+1/2 (ξi r ) Pn (cos θ ) exp (−kξ 2i t ) (3.96)
n =1 i =1
Now using the IC, we obtain

∞ ∞
f (r , θ ) = ∑ ∑ Ani (ξi r )−1/2 J n+1/2 (ξi r ) Pn (cos θ )
n =1 i =1
Multiplying both sides by Pn (cos θ ) d (cos θ ) and integrating, we have

1 ∞ ∞ 1
∫−1 Pn (cos θ ) f (r , θ ) d (cos θ ) = ∑ ∑ Ani (ξi r )−1/2 J n+1/2 (ξi r ) ∫−1 Pn2 (cos θ ) d (cos θ )
n =1 i =1
Using the orthogonality property of Legendre polynomials, we get
∞ ∞
1 ⎛ 2 ⎞
∫−1 Pn (cosθ ) f (r , θ ) d (cos θ ) = ∑ ∑ Ani (ξi r )−1/2 J n+1/2 (ξi r ) ⎜⎝ 2n + 1 ⎟⎠
n =1 i =1
Rearranging and multiplying both sides of the above equation by r 3/2 J n+1/2 (ξi r ) and integrating
between the limits 0 to a with respect to r, we get
2n + 1 a 1 a
∫0 ∫−1 Pn (cos θ ) f (r , θ ) d (cos θ ) = Ani ∫0 rJ n+1/2 (ξi r ) ξ i
2 −1/2
r 3/2 J n +1/2 (ξi r ) dr dr
2
a2
= Ani {J n′ +1/2 (ξi a )}2
2
Therefore,
(2n + 1) ξ 1/2 a 1
∫0 ∫−1 Pn (cos θ ) f (r , θ ) d (cos θ )
i
Ani = 2 2
r 3/2 J n +1/2 (ξi r ) dr (3.97)
a {J n′ +1/2 (ξi a )}
Hence, we obtain the solution to the given problem from Eq. (3.96), where Ani is given by
Eq. (3.97).
EXAMPLE 3.16 The heat conduction in a thin round insulated rod with heat sources present
is described by the PDE
ut − α u xx = F ( x, t )/ρc, 0 < x < l, t > 0 (3.98)
subject to
BCs: u (0, t ) = u (l , t ) = 0
IC: u ( x, 0) = f ( x), 0≤ x≤l (3.99)
where ρ and c are constants and F is a continuous function of x and t. Find u ( x, t ).
Solution It can be noted that the boundary conditions are of homogeneous type. Let
us consider the homogeneous equation
ut α u xx 0 (3.100)
Setting u ( x, t ) X ( x) T (t ), we get
Tc X cc
λ 2 (say) (3.101)
αT X
which gives X cc λ 2 X 0. The corresponding BCs are

X (0) X (l ) 0
The solution of Eq. (3.101) gives the desired eigenfunctions and eigenvalues, which are
2
§ nπ ·
sin λn x, λ 2n n t1 (3.102)
©¨ l ¹¸
X n ( x) ,
For the non-homogeneous problem (3.98), let us propose a solution of the form
f
u ( x, t ) ¦ Tn (t ) X n ( x) (3.103)
n 1
It is clear that Eq. (3.103) satisfies the BCs (3.99). From the orthogonality of eigenfunctions,
it follows that
2 l
Tm (t )
l ³0 u ( x, t ) X m ( x) dx
However,
2 1 § mπ ·
Tm (0)
l ³0 f ( x) sin ¨
© l ¸¹
x dx (3.104)
which is an IC for Tm(t). Introducing Eq. (3.103) into the governing equation (3.98), we get
f f
F ( x, t )
¦ Tnc X n α ¦ Tn X ncc ρc
(3.105)
n 1 n 1
Now, we shall expand F ( x, t )/ρ c, so that it is represented by a convergent series

on 0 x l , t ! 0 in the form
f
¦ qn (t ) X n ( x)
F
(3.106)
ρc n 1
where
2 l F ( x, t ) § nπ ·
qn (t )
l ³0 ρc
sin ¨
© l
x ¸ dx
¹
(3.107)
Thus, qn(t) is known. Now, Eq. (3.105), with the help of Eq. (3.101), becomes
f
¦ X n (Tnc λ 2nαTn qn ) 0
n 1
Therefore, it follows that
Tnc (t ) λ 2nαTn (t ) qn (t ) (3.108)

Its solution with the help of IC (3.104) is
t
Tn (t ) Tn (0) exp (λ 2nα t ) ³0 exp [λ nα (τ t )] qn (τ ) dτ
2
(3.109)
From Eqs. (3.103) and (3.109), the complete solution is found to be

f
ª 1 º
u ( x, t ) ¦ «¬Tn (0) exp (λ 2nα t ) ³0 exp [λ 2nα (τ t ) qn (τ ) dτ »¼ X n ( x)
n 1
In the expanded form, it becomes

f
ª 2 l ½
u ( x, t ) ¦ «¬®¯ l ³0 f (ξ ) X n (ξ ) dξ ¾ exp (λ 2nα t )
¿
n 1
^ `³ F (ξ , τ ) º
2 l l

l ³0 exp λ 2nα (τ t )
0 ρc
X n (ξ ) dξ dτ » X n ( x)
¼
(3.110)
It can be verified that the series in Eq. (3.110) converges uniformly for t > 0. By changing
the order of integration and summation in Eq. (3.110), we get
l ª f exp ( λ 2nα t ) X n ( x) X n (ξ ) º
u ( x, t ) ³0 ¦ «
«¬ n 1 1/2
» f (ξ ) dξ
»¼
l l ª f exp {λ 2nα (t τ )} X n ( x) X n (ξ ) º F (ξ , t )

³0 ³0 ¦ «
«¬ n 1 l /2
»
»¼ ρ c
d ξ dτ
which can also be written in the form

l l l F (ξ , τ )
u ( x, t ) ³0 G ( x, ξ ; t ) f (ξ ) dξ ³0 ³0 G( x, ξ ; t τ ) ρc
d ξ dτ (3.111)
where
f exp (λ 2nα t ) X n ( x) X n (ξ )
G ( x, ξ ; t ) ¦ l /2
n 1
is called Green’s function. More details on Green’s function are given in Chapter 5.
EXAMPLE 3.17 The temperature distribution of a homogeneous thin rod, whose surface is
insulated is described by the following IBVP:
PDE: vt – vxx = 0, 0 < x < L, 0 < t < ¥ (3.112)
BCS: v(0, t) = v(L, t) = 0 (3.113)
IC: v(x, 0) = f(x), 0 £ x £ L (3.114)
Find its formal solution.
Solution Let us assume the solution in the form
v(x, t) = X(x)T(t)
Eq. (3.112) gives
XT¢ = X²T
X T
or D 2 (say)
X T
where a is a positive constant. Then, we have
X² + a2X = 0
and T¢ + a2T = 0
From the BCS
v(0, t) = X(0)T(t) = 0,
and v(L, t) = X(L)T(t) = 0,
we obtain, X(0) = X(L) = 0 for arbitrary t. Thus, we have to solve the eigenvalue problem
X² + a2X = 0
subject to X(0) = X(L) = 0.
The solution of the differential equation is
X(x) = A cos ax + B sin ax.
Since X(0) = 0, A = 0. The second condition yields
X(L) = B sin aL = 0
For non-trivial solution, B ¹ 0 and therefore we have
sin aL = 0, implying a = np/L, for n = 1, 2, 3, …
Thus, the solution is obtained as

np x
Xn(x) = Bn sin .
L
Next, we consider the equation
T¢ + a2T = 0
whose solution can be written as
T (t ) = Ce -a
2
t
Tn (t ) = Cn e-( np /L ) t .
2
or
Hence, the non-trivial solution of the given heat equation satisfying both the boundary
conditions is found to be
Ê np x ˆ
vn ( x, t ) = an e - ( np /L ) t sin Á
2
(3.115)
Ë L ˜¯
where an = bncn (arbitrary constant).
To satisfy the IC, we should have
•
Ê np x ˆ
v( x, 0) = f ( x ) = Â an sin ÁË L ˜¯
n =1
which holds good, if f(x) is representable as Fourier Sine series with Fourier coefficients
2 L Ê np x ˆ
an =
L Ú0 f ( x )sin Á
Ë L ˜¯
dx .
Hence, the required formal solution is

•
È2 npt ˘ - ( np /L )2 t Ê np x ˆ
Â Ú0
L
v( x , t ) = f (t )sin dt ˙ e sin Á
ÍL Ë L ˜¯
.
n =1 Î ˚
L
EXERCISES
1. A conducting bar of uniform cross-section lies along the x-axis, with its ends at x = 0
and x = l. The lateral surface is insulated. There are no heat sources within the body.
The ends are also insulated. The initial temperature is lx − x 2 , 0 ≤ x ≤ l. Find the
temperature distribution in the bar for t > 0.
2. The faces x = 0, x = a of a finite slab are maintained at zero temperature. The initial
distribution of temperature in the slab is given by T ( x, 0) = f ( x), 0 ≤ x ≤ a. Determine
the temperature at subsequent times.
3. Show that the solution of the equation

wT w 2T
wt w x2
satisfying the conditions:
(i) T o 0 as t o f
(ii) T = 0 for x = 0 and x = a for all t > 0
(iii) T = x when t = 0 and 0 < x < a
is
f
2a (1)n 1 § nπ ·
T ( x, t )
π ¦ n
sin ¨ ¸ x exp [(nπ /a)2 t ]
© a ¹
n 1
2
4. Solve the equation w T w T satisfying the conditions:
w t w x2
(i) T = 0 when x = 0 and 1 for all t
2 x, 0 d x d 1/2
°
(ii) T ® 1
° 2 (1 x), d x d 1 when t 0.
¯ 2
5. Solve the diffusion equation
wθ § w 2θ 1 w θ ·
v¨ 2
wt ©w r r w r ¹¸
subject to
r 0, θ is finite, t!0
r a, θ 0, t!0
P 2
θ (a r 2 ), t 0
4μ
Here, P, μ and ν are constants.

6. A homogeneous solid sphere of radius R has the initial temperature
distribution f (r ), 0 d r d R, where r is the distance measured from the centre. The
surface temperature is maintained at 0°. Show that the temperature T(r, t) in the
§ 2 ·
sphere is the solution of Tt c 2 ¨Trr Tr ¸ . Show that the temperature in the sphere
© r ¹
for t > 0 is given by
f
1 § nπ ·
T (r , t )
rn ¦ Bn sin ¨© R r ¸¹ exp (λ 2nt )
1
where λn cnπ /R and c2 is a constant.

7. If φ ( x) is bounded for all real values of x, show that
1 f
³f φ (ξ ) exp [( x ξ ) /(4kt )] dξ
2
T ( x, t )
4π kt
is a solution of Tt kTxx such that T ( x, 0) φ ( x).

8. An infinite homogeneous solid circular cylinder of radius a is thermally insulated to
prevent heat escape. At any time t, the temperature T(r, t) at a distance r from the
axis of symmetry is given by the heat conduction equation with axial symmetry. At
time t = 0, the initial temperature distribution at a distance r from the axis is known
to be a function of r. Find the temperature distribution at any subsequent time.
9. Let r ( x, y , z ) represent a point in three-dimensional Euclidean space R3. Find a
formal solution u (r , t ) which satisfies the diffusion equation
ut α 2u, t!0
and the BC: u (r , 0) f (r ), where r R3 .
wθ w 2θ
10. Solve , 0 d x d a, t ! 0 subject to the conditions
wt w x2
θ (0, t ) θ (a, t ) 0 and θ ( x, 0) θ 0 (constant).
(GATE-Maths, 1996)
CHAPTER 4
Hyperbolic Differential Equations
4.1 OCCURRENCE OF THE WAVE EQUATION

One of the most important and typical homogeneous hyperbolic differential equations is the
wave equation. It is of the form
w 2u
c 2 2u (4.1)
w t2
where c is the wave speed. This differential equation is used in many branches of Physics and
Engineering and is seen in many situations such as transverse vibrations of a string or
membrane, longitudinal vibrations in a bar, propagation of sound waves, electromagnetic
waves, sea waves, elastic waves in solids, and surface waves as in earthquakes. The solution
of a wave equation is called a wave function.
An example for inhomogeneous wave equation is
w 2u
c 2 2u F (4.2)
w t2
where F is a given function of spatial variables and time. In physical problems F represents
an external driving force such as gravity force. Another related equation is
w 2u wu
2γ c 2 2u F (4.3)
wt 2 wt
where γ is a real positive constant. This equation is called a wave equation with damping
term, the amplitude of which decreases exponentially as t increases. In Section 4.2, we shall
derive the partial differential equation describing the transverse vibration of a string.
232
HYPERBOLIC DIFFERENTIAL EQUATIONS 233
4.2 DERIVATION OF ONE-DIMENSIONAL WAVE EQUATION

Suppose a flexible string is stretched under tension τ between two points at a distance L
apart as shown in Fig. 4.1. We assume the following:
1. The motion takes place in one plane only and in this plane each particle moves in
a direction perpendicular to the equilibrium position of the string.
2. The tension τ in the string is constant.
y
Q τ
P ds y = y (x, t)
τ
x
O dx A
Fig. 4.1 Flexible string.
3. The gravitational force is neglected as compared with tension τ of the string.

4. The slope of the deflection curve is small.
Let the two fixed ends of the string be at the origin O and A(L, 0) which lies along the
x-axis in its equilibrium position. Consider an infinitesimal segment PQ of the string. Let ρ be
the mass per unit length of the string. If the string is set vibrating in the xy-plane, the
subsequent displacement, y from the equilibrium position of a point P of the string will be
a function of x and time t, while an element of length dx is stretched into an element of length
ds given by
§w y ·
2 ª 1 § w y ·2 º
ds 1 ¨ ¸ dx | «1 ¨ ¸ » dx
©w x ¹ «¬ 2 © w x ¹ »¼
The elementary elongation is given by

2
1 §w y ·
dL ds dx ¨© ¸ dx
2 wx¹
while the work done by this element against the tension τ is

2
1 §w y ·
U¨ ¸ dx
2 ©w x ¹
Therefore, the total work done, W, for the whole string is
2
1 L §w y ·
W
2 ³0 U¨
© wx¹
¸ dx
If U is the potential energy of the string, then

2
1 L ⎛∂ y ⎞
U =W =
2 ∫0 τ⎜
⎝ ∂ x ⎟⎠
dx
Also, the total kinetic energy T of the string is given by

2
1 L ⎛∂ y ⎞
T=
2 ∫0 ρ⎜
⎝ ∂ t ⎟⎠
dx
Using Hamilton’s principle (See—Sankara Rao, 2005), we have

t1
δ ∫t
0
(T − U ) dt = 0
i.e.
t1
∫t 0
(T − U ) dt
is stationary. In other words,
1 t1 L ⎡ ⎛ ∂ y ⎞2 ⎛∂ y ⎞ ⎤
2
2 ∫t ∫0
0
⎢ ρ ⎜ ⎟ − τ ⎜ ⎟ ⎥ dx dt
⎢⎣ ⎝ ∂ t ⎠ ⎝∂ x ⎠ ⎥
⎦
is stationary, and is of the form
∫∫ F ( x, t, y, yx , yt ) dx dt
Noting that x and t are independent variables, from the Euler-Ostrogradsky equation, we have
∂F ∂ ⎛∂F ⎞ ∂ ⎛ ∂F ⎞
− − =0
∂ y ∂ t ⎜⎝ ∂ yt ⎟⎠ ∂ x ⎜⎝ ∂ y x ⎟⎠
which gives
∂ ⎛ ∂ y⎞ ∂ ⎛ ∂ y⎞
ρ − τ =0
∂ t ⎜⎝ ∂ t ⎟⎠ ∂ x ⎜⎝ ∂ x ⎟⎠
If the string is homogeneous, then ρ and τ are constants, in which case the governing equation
representing the transverse vibration of a string is given by
∂ 2y 2
2∂ y
= c (4.4)
∂ t2 ∂ x2
where
c 2 = τ /ρ (4.5)
EXAMPLE 4.1 Consider Maxwell’s equations of electromagnetic theory given by
E 4πρ
H 0
1 wH
uE
c wt
4π i 1 w E
uH
c c wt
where E is an electric field, ρ is electric charge density, H is the magnetic field, i is the
current density, and c is the velocity of light. Show that in the absence of charges, i.e.,
when ρ i 0, E and H satisfy the wave equations.
Solution Given
1 wH
curl E uE
c wt
Taking its curl again, we get
§ 1 wH· w §1 ·
u ( u E) u ¨ ¨© u H ¸¹
© c w t ¸¹ wt c
1 w §1 wE· 1 w 2E
¨ ¸
c wt ©c wt ¹ c2 w t 2
Moreover, using the identity
u ( u E) ( E) 2 E 2 E
it follows directly that
1 w 2E
2 E
c2 w t 2
Similarly, we can observe that the magnetic field H also satisfies
1 w 2H
2 H
c2 w t 2
which is also a wave equation.
4.3 SOLUTION OF ONE-DIMENSIONAL WAVE EQUATION BY

CANONICAL REDUCTION
The one-dimensional wave equation is
utt c 2u xx 0 (4.6)
Choosing the characteristic lines
Y x ct , I x ct (4.7)
the chain rule of partial differentiation gives
ux uξ ξ x uηη x uξ uη
ut uξ ξt uηηt c(uη uξ )
In the operator notation we have
w w w w §w w ·
, c¨
wx w[ wK wt © wK w[ ¹¸
Thus, we get
2
w 2u §w w ·
¨© w[ wK ¸¹ u uYY 2uYI uII (4.8)
w x2
w 2u
2
c 2 (uYY 2uYI uII ) (4.9)
wt
Substituting Eqs. (4.8) and (4.9) into Eq. (4.6), we obtain
4uYI 0 (4.10)
Integrating, we get
u (Y , I ) G (Y ) Z (I ),
where G and Z are arbitrary functions. Replacing Y and I as defined in Eq. (4.7), we have
the general solution of the wave equation (4.6) in the form
u ( x, t ) G ( x ct ) Z ( x ct ) (4.11)
The two terms in Eq. (4.11) can be interpreted as waves travelling to the right and left,
respectively. Consider
u1 ( x, t ) G ( x ct )
This represents a wave travelling to the right with speed c whose shape does not change as
it travels, the initial shape being given by a known function φ ( x). In fact, by setting t = 0
in the argument of φ , it can be observed that the initial wave profile is given by
u1 ( x, 0) = φ ( x )
At time t = 1/ c,
u1 ( x, 1/c ) = φ ( x − 1)
Let x ′ = x − 1. Then φ ( x − 1) = φ ( x ′ ). That is, the same shape is retained even if the origin
is shifted by one unit along the x-axis. In other words, the graph of u1 ( x, 1/ c ) is the same as
the graph of the original wave profile translated one unit to the right. At t = 2/ c, the graph
of u1 ( x, 2 /c ) is the graph of the wave profile translated two units to the right. Thus, in
particular, at t = 1, we have u1 ( x, 1) = φ ( x − c ). Hence in one unit of time, the profile has
moved c units to the right. Therefore, c is the wave speed or speed of propagation. Using
similar argument, we can conclude that the equation u2 ( x, t ) = ψ ( x + ct ) is also a wave profile
travelling to the left with speed c along the x-axis. Hence the general solution (4.11) of one-
dimensional wave equation represents the superposition of two arbitrary wave profiles, both
of which are travelling with a common speed but in the opposite directions along the x-axis,
while their forms remain unaltered as they travel. This situation is described in Fig. 4.2.
u1(x, 0) = φ (x)
u1(x, 1/c) u1(x, 1)
x
–1 0 (1, 1/c) (c, 1)
Fig. 4.2 Travelling wave profile.
Let k be an arbitrary real parameter. Consider then
u ( x, t ) = φ [k ( x − ct )] + ψ [k ( x + ct )] (4.12)
This is also a solution of the one-dimensional wave equation. Further, let ω = kc. Then
u ( x, t ) = φ [kx − ω t ] + ψ [kx + ω t ] (4.13)

A function of the type given in Eq. (4.13) is a solution of one-dimensional wave equation iff
ω = kc. Therefore, waves travelling with speeds which are not the same as c cannot be
described by the solution of the wave equation (4.6). Here, (kx + ω t ) is called the phase for
the left travelling wave. We have already noted that x ± ct are the characteristics of the one-
dimensional wave equation.
EXAMPLE 4.2 Obtain the periodic solution of the wave equation in the form
u ( x, t ) = Aei ( kx ±ω t )
where i = −1, k = ± ω /c, A is constant; and hence define various terms involved in wave
propagation.
Solution Let u ( x, t ) = f ( x)e±iω t be a solution of the wave equation
utt = c 2u xx
Then
uxx = f ¢¢( x)e± iw t , utt = - f ( x)w 2 e± iw t
Substituting into the wave equation, we get
ω2
f ′′ ( x) + f ( x) = 0
c2
Its general solution is found to be
f ( x ) = c1 exp [i (ω /c ) x ] + c2 exp [−i (ω / c ) x]
Therefore, the required solution of the wave equation is
u ( x, t ) = [c1 exp {i (ω /c) x} + c2 exp {−i (ω /c) x}] e± iω t
Since k = ± ω /c, the time-dependent wave functions are of the form
u ( x, t ) = Aei ( kx ±ω t )
Hence, u ( x, t ) = Aei ( kx ±ω t ) is a solution of the wave equation, and is called a wave function.
It is also called a plane harmonic wave or monochromatic wave. Here, A is called the
amplitude, ω the angular or circular frequency, and k is the wave number, defined as the
number of waves per unit distance. By taking the real and imaginary parts of the solution,
we find the linear combination of terms of the form
A cos (kx ± ω t ), A sin (kx ± ω t )
representing periodic plane waves. For instance, consider the function u ( x, t ) = A sin (kx − ω t ).
This is a sinusoidal wave profile moving towards the right along the x-axis with speed c.
Defining the wave length λ as the length over which one full cycle is completed, we
have λ = 2π /k , thereby implying that k = 2π /λ.
Suppose an observer is stationed at a fixed point x0 ; then,
⎛ λ⎞ ⎛ λ⎞
u ⎜ x0 , t + ⎟ = A sin ⎜ kx0 − ω t − ω ⎟
⎝ c⎠ ⎝ c⎠
= A sin (kx0 − ω t − 2π ) = A sin (kx0 − ω t )
Thus, we have
u ( x0 , t + λ /c ) = u ( x0 , t )
Hence, exactly one complete wave passes the observer in time T = λ /c, which is called the
period of the wave. The reciprocal of the period is called frequency and is denoted by
f = 1/ T
The function, u = A cos (kx − ω t ) = A sin (π /2 + kx − ω t ), also represents a wave train except
that it differs in phase by π /2 from the sinusoidal wave. Now consider the superposition of
the sinusoidal waves having the same amplitude, speed, frequency, but moving in opposite
directions. Thus, we have
u ( x, t ) = A sin [k ( x − ct )] + A sin [k ( x + ct )]
= 2 A sin kx cos (kct ) = 2 A cos (kct ) sin kx
Its amplitude factor [2 A cos (kct )] varies sinusoidally with frequency ω . This situation is
described as a standing wave. The points xn = nπ /k , n = 0, ± 1, ± 2, … are called nodes. No
displacement takes place at a node. Therefore,
u ( xn , t ) = 0 for all t
The nth standing wave profile will have (n − 1) equally spaced nodes in a given interval
as shown in Fig. 4.3.
n=5
n=4
n=3
n=2
n=1
Fig. 4.3 Standing wave profiles.
4.4 THE INITIAL VALUE PROBLEM; D’ALEMBERT’S SOLUTION

Consider the initial value problem of Cauchy type described as
PDE: utt − c 2 u xx = 0, − ∞ < x < ∞, t ≥ 0 (4.14)
ICs: u ( x, 0) = η ( x ), ut ( x, 0) = v ( x ) (4.15)
where the curve on which the initial data η ( x) and v( x) are prescribed is the x-axis. η ( x) and v( x)
are assumed to be twice continuously differentiable. Here, the strin

Sprache

Willkommen bei Scribd!

Rao K.S - Introduction to Partial Differential Equations (2011).pdf

Hochgeladen von

Dokumentinformationen

Datum des Uploads

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Beschreibung:

Copyright:

Verfügbare Formate

Rao K.S - Introduction to Partial Differential Equations (2011).pdf

Hochgeladen von

Beschreibung:

Copyright:

Verfügbare Formate

Verwandte Titel

Introduction to

PARTIAL DIFFERENTIAL EQUATIONS

Eleventh Printing (Third Edition) … … January, 2011

0. Partial Differential Equations of First Order 1–51

1. Fundamental Concepts 52–105

1.4 Adjoint Operators 69

2. Elliptic Differential Equations 106–181

3. Parabolic Differential Equations 182–231

4. Hyperbolic Differential Equations 232–281

5. Green’s Function 282–315

6. Laplace Transform Methods 316–387

6.8 Inverse Transform 337

7. Fourier Transform Methods 388–446

Partial Differential Equations

A solution of Eq. (0.1) in some domain Ω of IR 2 is a function z f ( x, y ) defined and is

0.2 SURFACES AND NORMALS

Let Ω be a domain in three-dimensional space IR 3 and suppose F ( x, y, z ) is a function in

If the equation of the surface is of the form

0.3 CURVES AND THEIR TANGENTS

Further, we assume that

This non-vanishing vector is tangent to the curve C at the point (x, y, z) or at [ f1 (t ), f 2 (t ), f3 (t )]

Then, the equation of the tangent plane to S1 at a point P ( x0 , y0 , z0 ) is

Similarly, the equation of the tangent plane to S2 at the point P ( x0 , y0 , z0 ) is

Solution The tangent vector to the helix at (x, y, z) is

Solution The space circle is described as

0.4 FORMATION OF PARTIAL DIFFERENTIAL EQUATION

Now, eliminatin w F /w u and w F /w v from Eqs. (0.29) and (0.30), we obtain

Differentiating Eq. (1) twice partially with respect to x and t, we get

Hence, the required PDE is

Eliminating f c from Eqs. (2) and (3) we get

Eliminating f c from Eqs. (2) and (3), we find

which is the required PDE.

Solution Given z ax  by  ab (1)

0.5 SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS OF FIRST ORDER

proportional to (w z /w x, w z /w y , 1) or ( p, q, 1). Therefore, the direction of the normal is

Theorem 0.1 The general solution of the linear PDE

can be written in the form F (u , v) 0, where F is an arbitrary function, and u ( x, y , z ) C1 and

on expanding the determinants, we have

which can be rewritten as

The solution of these equations are known to be u ( x, y , z ) C1 and v ( x, y , z ) C2 . Hence,

EXAMPLE 0.8 Use Lagrange’s method to solve the equation

Solution The given PDE can be written as

The first two members of Eq. (1) give

which on integration yields

0.6 INTEGRAL SURFACES PASSING THROUGH A GIVEN CURVE

EXAMPLE 0.10 Find the integral surface of the linear PDE

Solution The auxiliary equations for the given PDE are

Using the multiplier xyz, we have

Eliminating the parameter t, we find

EXAMPLE 0.11 Find the integral surface of the linear PDE

From Eqs. (7) and (8) we observe

which on simplification gives us

0.7 THE CAUCHY PROBLEM FOR FIRST ORDER EQUATIONS

is to find a region IR in ( x, y ), i.e. the space containing ( x0 ( s ), y0 ( s )) for all s  I , and a

then, there exists a unique function φ ( x, y ) such that

(c) For all values of y in the interval | y  y0 |  δ1 , φ ( x0 , y ) g ( y ).

0.8 SURFACES ORTHOGONAL TO A GIVEN SYSTEM OF SURFACES

Fig. 0.2 Orthogonal surface to a given system of surfaces.

Solution Given z ax by ab (1)

proportional to (w z /w x, w z /w y , 1) or ( p, q, 1). Therefore, the direction of the normal is

is to find a region IR in ( x, y ), i.e. the space containing ( x0 ( s ), y0 ( s )) for all s I , and a

(c) For all values of y in the interval | y y0 | δ1 , φ ( x0 , y ) g ( y ).

Finally, replacing ( x x0 ), ( y y0 ) and ( z z0 ) by dx, dy and dz respectively, which corresponds

p c1 exp (t ), q c2 exp (t ), x c2 exp (t ) c3 ½°

p c1 exp (t ), q c2 exp (t ), x c2 exp (t ) c3 ½°

where X {φ , ψ , 1}. That is,

Solving the above pair of equations for (ψ x φψ z ), we have

0.11 CHARPITS METHOD

Let z ax by c 0 is a solution of the given PDE, described by f ( p, q ) 0, then

As a trial solution, let us assume that z is a function of u x ay , where a is an arbitrary

Solution Let us assume that z f (u ) x ay is a solution of the given PDE. Then,

which contains the circle x 2 y 2 1, z 1.

which contains the circle x 2 y 2 2 x, z 0.

which contains the circle x 2 y 2 1, z 1.

which contains z 1 and x 2 y 2 1. (GATE-Maths, 1999)

(C) ax by (ba 2 ab 2 ) (D) ax by (ab 2 ba 2 ).

(ii) u xx u yy u zz 0 [Laplace equation in three dimensions]