V. Ya.

Arsenin

Basic Equations
and
Special Functions
of
Mathematical Physics
Jl
 Basic Equations and Special
Functions of Mathematical Physics
T
 BASIC EQUATIONS AND SPECIAL
FUNCTIONS OF MATHEMATICAL PHYSICS

V. Ya. Arsenin

Translated by
S. Chomet, King’s College, London

English translation edited by

S. Doniach, Imperial College o f
Science and Technology, London

LONDON ILIFFE BOOKS LTD

NEW YORK AMERICAN ELSEVIER PUBLISHING
COMPANY INC.
 First published by Nauka, Moscow, under the title Matematicheskaya
Fizika — Osnovnye Uravneniya i Spctsial’nye Funktsii

English Edition first published in 1968 by Jliffc Books Ltd,

42 Russel] Square, London, W.C.I.

American edition published by American Elsevier Publishing Company Inc.

52 Vanderbilt Avenue, New York, New York 10017

Library of Congress Catalog Card Number 68-57390

Translated and prepared for press by Scripta Technica Ltd,

139a New Bond Street, London, W.l.

(C) Iliffe Books Ltd 1968

Printed in Poland
 Contents

Preface 7

PART 1

Chapter 1 Linear Equations with Two Independent Variables 3

Chapter 2 Partial Differential Equations in Physics.
Boundary-value Problems 13
Chapter 3 The Method of Characteristics 32
Chapter 4 Separation of Variables (Fourier Method) 59
Chapter 5 The Method of Green’s Functions for Parabolic
Equations 107
Chapter 6 The Method of Green’s Functions for Elliptical
Equations 124
Chapter 7 Potentials 146
Chapter 8 Integral Equations 168
Chapter 9 Integral Equations with Symmetric Kernels 188

PART 2
Chapter 10 Gamma Function 215
Chapter 11 Cylindrical Functions 223
Chapter 12 Spherical Harmonics 268
Chapter 13 Chebyshev-Hermite and Chebyshev-Laguerre
Polynomials 291
Appendix Definition of Generalised Functions. The S
Function 303
Answers and Solutions 321
Bibliography 355
Index 357

5 A
 Preface

This book consists of two parts. Part 1 gives an account of the

methods for solving typical problems in mathematical physics and
an introduction to integral equations. Part 2 deals with the appli­
cations of these methods to problems requiring the use of special
functions.
Extensive use is made in this book of the Dirac (5-function.
Generalised functions are introduced and their applications are
described. Each chapter concludes with a number of problems
illustrating the main text (altogether 150 problems with answers
are given).
This text was designed for students of physics and engineering and
was based on a course given at the Department of Theoretical and
Experimental Physics of the Moscow Engineering-Physics Institute.
Special thanks are due to A. N. Tikhonov, who initiated this
course, and to A. A. Samarskii for many discussions and valuable
advice. V. S. Vladimirov and T. F. Volkov have read the manuscript
and put forward a number of important suggestions which have
been incorporated in the text. I am also indebted to the publisher’s
reader, S. A. Shirokova, for many useful suggestions which have
led to an improvement in the presentation of the material.

V. Ya. Arsenin

7
PART ONE

1
 1

Linear Equations with Two

Independent Variables

Many physical problems lead to second-order partial differential

equations. These equations can be written in the general form of
a functional relationship between the independent variables
x l5 x„, the unknown function u and its first- and second-order
partial derivatives uXl, uXlX2, •••, uXiXj, ..., uXnXn:
0 O i , X 2 , • • ■, x n , U, Ux j , u X 2 , , UX n , UXIXI , . . . , UX(Xj , . . . , u XnXn) = 0

Very frequently these equations are linear in the second-order

derivatives, i.e. they are of the form
n n

' ^ ' ^ j aij uXiXJJr F (x,, x„, u, uXi, ...,uXn) = Q

*’=1 >=1
where the coefficients of the highest-order derivatives, atj, are
functions of only the independent variables Xj, x2, ..., xn.
When the function F(xx, ..., x„, u, uXl, ..., uXn) is linear in the
arguments u, uXl, ..., uXn, the differential equation is termed ‘linear’.
Linear equations are of the form
n
S^ S^ \ i aiJuXtXj+ S^ b iuxl+cu = / ( x ls ...,x„) (1)
i= 1j = l

where the coefficients a c are functions only of the independent

variables x l9 ..., x„.
When / ( x , , ...,x„) = 0, Equation (1) is called a homogeneous
3
 Mathematical Physics

linear equation,' and when this identity is not satisfied, it is called

an inhomogeneous equation.
When the coefficients au , 6;, c are constants, Equation (1) is
called a linear equation with constant coefficients.
Equations which are linear in the second-order derivatives or
in both second- and first-order derivatives can be divided into
three classes (types). Each class includes an equation with a particu­
larly simple form which is known as the canonical form. Equations
belonging to a given class have many common properties. To
investigate these properties it is sufficient to consider the canonical
equations. In the next few chapters we shall be concerned with
the properties of the solutions of canonical equations and with
methods for obtaining these solutions.
Whether or not a particular equation belongs to a given class
is determined by the coefficients of the highest-order derivatives.
We shall consider this classification first for equations in which
the unknown function u depends on only two variables, i.e.
u = u(x, y). In this case, equations which are linear in the highest-
order derivatives may be written in the form
anUxx+ 2 al2uxy+ a 22Uyy+F(x, y, u, ux, uy) = 0 (la)
Similarly, linear equations may be written in the form
an u.xX+2au uxy —a22uyy+ b l ux+ b 2uyy-cu = f { x , y ) (2)
where a,-;, bif c are functions of only the independent variables x, y.
Any equations of the form (la) or (2) can be reduced to a more
simple form, the canonical form, by a suitable transformation of
the independent variables. In studying partial differential equations
with two independent variables, we shall therefore be able to confine
our attention to the canonical equations.
Let us transform the independent variables in Equation (la)
in accordance with the formulae

f y), v = v>(x> y) (3)

which define a one-to-one mapping between the points (£, ?;) and
(x, _y). We shall require that the functions <p(x, v) and y(.v, j ) and
their first- and second-order derivatives be continuous. We then have
Ux <fx U'i lpx , lly Typy

uxx = rf l % I-2(px yx Uin + ip2x U,in7 q-xx lti —ipxx Ur,

Uy y = ' l l « { 4 H- 2 ( p y V y u i n + lp ] 11nn + T y y 11i + W y y 11n

Uxy = <Px <?y W« + (<Px Wy + 9 y V x ) 11i n + W x Wy 11nn+ ' f x y H i + W xy Hn

4
 Linear Equations
Substituting these derivatives into (la) and collecting up terms
involving the same derivatives, we obtain the transformed equation
aiiM« + 2 a 12 i/^ + a22M„+-F1(M^, un, u, £, rj) = 0 (4)
where
an = an <p2x-\-2al2<px<py+ci22<p2y

«i2 = an <pxy x+ a l2(<PxV,-\-<PyV>x)+a22(Py1/>y (5)

«22 = an y%+2a12y>xy>,+a32y$

Using (5) we can readily verify that

£(<?; vOl 2
*12 alia22 (^12 all 022) ( 6)

We can now adopt the following classification of equations

of the form (la). If in a given domain D the discriminant
V2 = a\2—an a22 is positive (V 2 > 0 ) then, Equation (la) is hyperbolic
in D. If, on the other hand, V2 < 0 in £>, (la) is elliptic in D.
Finally, when V2 = 0 at all points of D, (la) is parabolic in D.
It follows from the identity given by (6) that the type of an equa­
tion of the form (la) is unaffected by a transformation of the inde­
pendent variables of the form (3). We shall use this procedure of
transformation of the independent variables to simplify (la), i.e.
to reduce it to its canonical form. For each type of equation there
is a particular canonical form.
1. If (la) is hyperbolic in a domain D, there exist in D functions
<p{x, y ) and ip(x, y) such that the transformation given by (3) reduces
(la) to the simpler form

un, u, f, V) = 0 (7)
This is the canonical form.
We shall now summarise the procedure for finding y{x, ^) and
yt(x, y) without discussing the conditions for their existence:
a. If an — a22 — 0 in D, then an ^ 0. Dividing both sides of
(1) by 2ax2 we obtain the canonical form (7).
b. Suppose that *11+ 022 ^ 0 in D. We shall confine our attention
to the case when at least one of the coefficients au , a22 is not iden­
tically zero in a part Dx of D. Suppose an has this property.
We shall take <p(x, y) and ip(x, y) in (3) to be functions which
ensure that the coefficients a n and a22 in the transformed equation (4)
5
 Mathematical Physics
will vanish, i.e. we shall suppose that they are the solutions of the
following equations:
a\\(P2xJr'^a\2(Px(PyJr a22(pl = 0
auW l+^uVxW y+^wl = 0
These yield

<Px _ _ — a i2 ± V V 2 Wx _ — fli2=fcr'V2
(fy Ify Q.\ J
Consequently, each of the equations in (8) can be split into the
following:
<Px+*i(x, y)<py= 0, Wx+h(x, y)v>y = 0 (9)
where

a \2 - l /] V 2 an~\~X V2
h ( x , y) - ?i(x, y) = ( 10)
mi mi
The two equations in (9) are equivalent to
dy dy
S i = i , ( x ’ y)• no

This procedure yields functions <p(x, y) and y>(x, y) which will

ensure that the coefficients a n and a22 will vanish in the case under
consideration. At the same time, a I2 A 0, which follows immediately
from the identity given by (6). Dividing the transformed equation
by 2«12 we obtain the required canonical form.
On integrating (11) we obtain
<p(x, y) = Ci and y>(x, y) = c2
which form two families of curves called the characteristic curves,
or simply characteristics, of (la). We note that two characteristics
belonging to different families will never coincide since ^ 22.
It follows that the above families of characteristic curves form
curvilinear coordinate nets.
2. If (la) is elliptic in D, there exist in D functions cp(x, y) and
(p(x, y) such that (la) can be reduced to the canonical form
u^-ytt^-yFx{it^, Uq, u, £, ij) = 0 (12)
by means of the transformation (3). We shall again confine our
attention simply to a description of the procedure for finding
<p(x, y) and y>(x, y).
6
 Linear Equations

As before, we first reduce the equation formally to the form

+ F y(its, u,, ,/ / ,£ , »7) = 0 (13)
The new variables £ and will now be complex conjugates
f j ) + » > (* , >'), »7 ^ <p{x, y ) — iip(x, y)
since the differential equations for the characteristic curves are
now of the form
dy _ fli; . } / —V2 dy = gi2 | ^-l7 —V2
d.v fln an ’ dx an an
Consequently, an elliptic equation has only imaginary character­
istic curves.
Finally we can transform the independent variables again in
accordance with the formulae

P= y), <* = ^ 2 J~ = v ( x , t )

so that (13) and, consequently, (la) will assume the required canon­
ical form (in the new variables)
upp+iiaa+ F 2(iip, ua, u, p, a) = 0
3. If (la) is parabolic in D, there exist in D functions cp(x, y)
and y)(x, y) such that the transformation given by (3) reduces
Equation (la) to the canonical form
u ^ + F^Ui, un, u, f , >]) = 0 (14)
The procedure for finding the functions 9?(.v, _y) and ip(x, y)
can be summarised as follows. To begin with, we must find the
function <p(x, y) which will ensure that the coefficient a n in the
transformed equation will vanish, i.e. it will be a solution of the
equation
an cpl+2a{2cpxcpy+ a 22(p2y = 0 (15)
As in the case of hyperbolic equations we are assuming that an
is not identically zero in any part of D. We then solve (17) for
fxhfy, but in contrast to the hyperbolic case (see (9)), we obtain
only one equation, namely,
<Px+Kx,y)rPy = 0 (16)
where
 Mathematical Physics
Any solution of (16) which is not identically equal to a constant
can be taken for <y(x, y). The coefficient a 12 in the transformed
equation will then also vanish, which follows from the condition
that (la) is a parabolic equation, and from the identity given by
(6). For y>(x, y) we can take any twice continuously differentiable
function which does not lead to a22 being zero. Dividing the equation
transformed in this way by a22, we obtain the required canonical
form. An equation of the parabolic type has only the single family
of characteristic curves

If the initial equation (la) is linear, the transformed equation will

clearly also be linear.
Thus, the canonical forms of linear equations are
»in+P\ut+^2un+yu = /(£ > V) (hyperbolic)
Ut'+Uw+PiUz+PiUn+yit = / ( £ , v) (elliptic) (17)
Um+PiUc+PiUv+yt* = / ( £ , V) (parabolic)

4. If the differential equation is a linear equation with constant

coefficients, then the coefficients P\,P2,V in the corresponding
canonical equations will also be constants. (The characteristic
curves of a hyperbolic equation will then be straight lines.) The
equations given by (17) can then be simplified still further by
substituting
u — v e ^ +vn (18)
where /<, v are to be determined.
Evaluating the derivatives of the function u and substituting
the result into, say, the first equation in (17), we obtain
vtn-\-(v+Pi)vr \-(ji+PJvn+(j*v+p.p1+vp2-\-y)v = / ( £ , ??)e
If we substitute
M = —Pi, v = —P\
the transformed equation assumes the form
vtn+yiv = / .( £ , v) (19)
where

8
 Linear Equations
Similarly, an elliptic equation will reduce to the form

v iZJr 'v ’m + y i v ~ V) (20)

where

Yi = r - j W + l > l ) , r= -jf> i, / , = / e - ' !- '{

In a parabolic equation, the coefficients of v* and v n cannot

be made to vanish by a suitable choice of y and v since the trans­
formed equation is of the form
z'nn~rfti'v$Jr(2v-rfi2)vnJr(v2jrVfi2JrfJ‘P\Jr'y)v v)

Substituting v = —~ /J 2, ju = —y ) we obtain

( 21 )

As a result of the simplifications of (la) described above, we will

confine ourselves to consideration of methods for solving problems
which can be formulated in the form of canonical equations. In the
case of linear equations with constant coefficients, we may generally
confine our attention to equations of the form given by (19),
(20) and (21).
Let us now consider a number of examples.
Example 1 uxx—yuyy — 0. In this case, an = 1, an = 0,
a22= —y, V2 = a\2—a na22 = y . Consequently, for y > 0 the equation
is of the hyperbolic type, whereas for y < 0 it is of the elliptic type.
a. Consider, to begin with, the region in which the equation
is hyperbolic. The differential equations for the characteristic curves
are of the form

and their general integrals are x ~ 2 ] / y = cl9 x-\-2]/y = c2. Substi­

tuting
i = x —2\/y, v = x + 2 ]/y
we obtain the canonical form of the transformed equation

, 1 1 / ,
“«i + T TS L _•(-(“< - u-i) = 0

9
 Mathematical Physics

The characteristics are the right- and left-hand branches of the

family of parabolae ( x —c)2 = 4y (solid and broken curves in
Fig. 1.1). The minimum points of the parabolae, which lie on the x
axis, do not belong to the characteristic curves since, at these points,
the equation is not of the hyperbolic type (V2 = 0).

Fig. 1.1

b. In the region in which the equation is elliptic < 0), we

shall substitute
£+»? v —£ t /—
p = — = x> — 2r = 2 V -y

The canonical form of the equation is

Mpp^rHao 0
a

Example 2 xu xx—2 )/xyu XyJryUyy+\uy = 0.

In this case, an = x, a12 = —\/xy, a 21 = y, V2 = a\2—ana22 = 0.
Consequently, this equation is hyperbolic at all points. It-has the
single family of characteristic curves given by the differential
equation

dx
i/Z
|/ x
(or —
\
L= —
\ /y
7=^)
| x /
The general integral of this equation is

/ * + Vy = c
We shall therefore substitute
£ = 1Sx + \/y
and >i can be set equal to any function <p(x, y) which does not lead
to «22 being zero in the transformed equation. Let

'/ = I -v
10
 Linear Equations
The canonical form of the equation is then

h ,,— - = 0
7]

5. Whether or not a linear differential equation which does not

contain a mixed derivative of the unknown function, i.e. an equation
of the form
an uxx+ a 22uyy+ b 1ux+ b 2uy+ c u = /(x , j ) (22)
belongs to a particular class, is determined by the signs of an and a22.
More precisely, if «n(*> y) and a22(x, y) have different signs through­
out D (and do not vanish in D), (22) is hyperbolic in D. If, on
the other hand, an (x, j ) and a22(x, y ) have the same signs throughout
D (and do not vanish in D), Equation (22) is elliptic in D. Finally,
if one of the coefficients an , a22 is zero throughout D, (22) is para­
bolic in D.
Similar criteria can be used to classify linear equations of the
form
n n

y aituXiXi+ ^ bkuXk+ c u = /(.Xi, x 2, ■■■, x„) (23)

i=i k =1
with a large number of independent variables (xl5 x 2, ..., x„),
where a,y, bk, c are functions of the variables (xl5 x 2, ..., x n).
Equation (23) can be classified as follows. It is:
elliptic at the point (x?, x°, ..., x°) if all the coefficients au(x?l,
x2, ..., xg) at this point are not zero and have the same sign;
hyperbolic at the point (x?, x2, x°„) if the coefficients ait(x
x 2, ...,x2) at this point are not all zero and all of them, except
one (for example a,-0;0), have the same sign;
hyperbolic at the point (x?, x 2, ..., x“) if the coefficients
a n ( x i , x ...,x2) at this point are all, except one (for example,
a;0,0), different from zero and all have the same sign
ai0i0(x°i> x 2 , ■■■>A ) = 0 and biQ(x\, x°2, ..., xj) ^ 0

PROBLEMS
1. Reduce the following equations to canonical form:
(a) = 0
x z u x x — y 2Uyy

(b) y 2uxx+ x 2Uyy = 0

(c) x 2uxx+ l x y u Xy + y 2Uyy = 0
(d) Uxx+yuyy+ 0.5 Mj, = 0
11
 Mathematical Physics
2. Reduce the following equations to canonical form:
(a) uxx-\-2uxy+ Uyy-\-3ux — 5Ity ^ 4ll = 0
(b) UXX-{-4UXy-{-3Uyy-{- 5ltx-\- Ily + A u = 0
(c) 2uxxJr2uxyJr UyyJr 4ux-j-4Uy-{- u = 0
 2
Partial Differential Equations
in Physics
Boundary-value Problems

2.1 SMALL TRANSVERSE VIBRATIONS OF A STRING

We shall assume that the string is elastic and offers no resistance

to any change of form other than a change in length. This is
expressed mathematically by saying that the tension in the string
is always parallel to the tangent to the instantaneous profile of the

string (Fig. 2.1). We shall also assume that the vibrations are small,
i.e. the square of the displacement can be neglected, and that they
take place in the (x, u) plane.
The assumption that the vibrations are small ensures that the
tension T in the string is independent of time t. To show this,
consider a segment (xl3x2) °f the undisturbed string. Its initial
length is x 2—x 1, whereas the length at time t is
Xl *2
^ ]/1 T ul dx « jj 1 •dx = x2—Xi
X\ x t

It follows that, to within terms of the second order in ux, the length
13
 Mathematical Physics
of a fixed segment of the string is independent of time, i.e. the
segment is not extended. Hence, in view of Hooke’s law, it follows
that the tension T is also independent of time (to within terms
of the second order in ux). Consequently, T can be a function
of only x :
T = T(x)
Since we are considering the transverse vibrations of the string,
we shall only be interested in the component of the tension along
the it axis, which will be denoted by Tu. It is clear that

Tu - Tsina — T tanacosa = T — Ux ■« Tux

l/l+ w *
where a is the angle between the tangent to the curve u — u(x, t)
and the x axis for given t (Fig. 2.1). The momentum of a segment
(x1? x2) of the string at time t is equal to
*2
J w,(£, t ) p (f)d f

where p is the linear density of the string.

According to Newton’s second law, the change in the momen­
tum of the segment (xl5 x2) in a time V2t = t2—ti is equal to the
time integral of the applied forces which, in the present case, consist
of the tension force Tux at the ends of the segment and the external
XI

force J /(£, /)d£, expressed in terms of a force density /(x , t):

X\

Xi

S t#<(f, h )]p (f)d f

Xi

12 >Z*2
= J [7’(x2)wv(x2, t ) — 7’(x1)w,(xi , t)] d r + jj $ / ( I , t) d f dr (1)
h <i xi

This is the integral form of the equation for small transverse vibra­
tions of a segment of a string.
If w(x, t ) has continuous second-order derivatives and T(x)
has a continuous first-order derivative, then using the mean value
theorem for the integrals in (1) we obtain

t/„(fi, rOp(fi) V2t V2x = ~ [ r ( x ) « J , _ {2V2t V2x + /( f 3, r 3)V2t V2x (2)

OX t =Z2

14
 Partial Differential Equations in Physics

where £l512>£3 e [*u -v2], Ti, rz> ri e [*u C]- Dividing both sides of
(2) by V2r V2.x and proceeding to the limit as V2r -> 0 and V2x' -> 0,
we obtain the following differential equation for the small transverse
vibrations of a string

■^7 [Tux]+f(x, t) = p(x)utt (3)

When T = const and p = const, this equation is usually written

in the form
a2uxx+F (x, t) = utt (4)
where a2 = T/p, F(x, t) = (l/p )f(x, t). Equation (4) is called the
one-dimensional wave equation.

2.2 SMALL LONGITUDINAL VIBRATIONS

OF AN ELASTIC ROD

Consider a rod parallel to the x: axis. We shall use the following

notation: 5(.v) is the cross-sectional area of the rod in the plane
perpendicular to the x axis at the point x, k(x) and p(x) is the
Young’s modulus and density of the rod in a plane having the
abscissa x, and u(x, t) is the displacement (in the direction of the
axis of the rod) of the section having the abscissa x at time t. We
shall assume that the displacement of all points on the plane with
given x: is the same. It is clear that the longitudinal vibrations are
completely described by the function u(x, t). Small longitudinal
vibrations will be defined as those for which the stresses produced
in the rod during the vibration always obey Hooke’s law. Consider
the relative extension of a segment (x, x + V 2x) at time t. The coordi­
nates of the end of this segment are
x f f u { x , t), .v+V2x:+m(x:+V2x:, t)
Consequently, the relative change in the length of the segment is
{[x+V 2x:+m(x:+V2x, t)]—[x+ i/(x, t)]}—V2x
ux(x-\- 0 V2x, 1 )
' Vhc
(0 < 0 < 1)

It follows that the relative extension at the point x: at time l is ux(x, t)

and the tension T is, in view of Hooke’s law, given by
T = k(x )S(x)ux(x, t)
15
 Mathematical Physics

If we now apply Newton’s second law to the segment (a,, x2) of the
rod in a time interval V1t = (2- ( i ,w e obtain
*2

*1
n 12x2
= $ {^(.v2)A:(.X2) Ux(x2, t ) —S (* i)£ (* i) ux(x ,, r)} d r + J $/(£, r ) d |d r
n n*i
where /(x , /) is the density of the external force acting on a cross-
section at a distance x from the origin. This is the integral form of the
equation for the small longitudinal vibrations of the rod. Assuming
that the function u(x, t) has continuous second-order derivatives and
that k(x) and iS(x) have continuous first-order derivatives, we can
readily show that the differential equation for the small longitudinal
vibrations of a rod is

— [iSXaO&CxKO, /)]+ /(* > 0 = P (x)S(x)u,«(.v, r) (5)

When >S(x), k(x) and p(x) are constants. Equation (5) assumes the
form
a2i'xx+F {x, t) = u„
where

«2 = f f o = - /( .v . o

Equations (3) and (5) are identical in form and differ only in the
notation (Sk instead of T, and pS instead of p). They are hyperbolic
at all points since, by definition, T(x), S(x) and k(x) are positive.

2.3 SMALL TRANSVERSE VIBRATIONS OF A MEMBRANE

A membrane is defined as a stretched plane sheet which is perfectly

flexible and can be displaced, but offers resistance to stretching.
For example, a plate whose thickness is small in comparison with
its two other linear dimensions can in certain cases be regarded as
a membrane. We shall consider small transverse vibrations of
a membrane at right-angles to its own plane (.v, v) and will assume
that the squares of the displacements ux and uy can be neglected,
where u = u(x, y, t ) is the displacement of the point (x, y) at time t.
16
 Partial Differential Equations in Physics
Let ds be an element of arc of a contour on the surface of the
membrane and let N be a point on this element. The tension force
acting on this element is T ds. The fact that the membrane is per­
fectly flexible is expressed mathematically by saying that the vector
T lies in the plane which is tangential to the surface of the membrane
at the point M and perpendicular to the element ds, and that the
magnitude of T at this point is independent of the direction of ds.
Since the vibrations are small, it follows that the component Tc of
the tension vector T on the (x, y) plane is equal to T. In fact,
Tc — T cos a, where a is the angle between T and the (x, y ) plane.
However, a is not greater than the angle y between the tangent plane
to the surface of the membrane, which contains the vector T, and
the (x, y ) plane (i.e. a < y). It follows that

cos a .^ cosy = 1
- 1
V1 +
Consequently, cos a « 1 and hence Tc « T.
The second consequence of the fact that the vibrations are small
is that the tension T is independent of the time t. To show this,
consider a part S of the undisturbed membrane. This area is equal
to dx d^. At time t this area is equal to
s

It follows that the area of a fixed part of the membrane is independent

of time, and in view of Hooke’s law, T is independent of time. Since
T is perpendicular to the element of arc ds, it follows that T is also

B, B2

yt I
I
Fig. 2.2 1
x, X

independent of x and y. In fact, consider a part A lBlB2A2 of the

undisturbed membrane bounded by lines parallel to the coordinate
axes (Fig. 2.2):
2 17
 Mathematical Physics
the tension force acting on the segment of the surface is equal to

J rd y + \ Tds+ \ Tds+ $ Tds

A.\j4.2 A 2B 2 B 2B 1 Bf A\

Since points of the membrane are not displaced in the direction of

the x and y axes, we have
yz yz
5 T d^+ 5 T ds = 5T (x'2>y) d>’~ ST(x i ’y) d>'
A 2B 2 B tA i _K, J>1

y2
= $ [T{X2 , v)- T{x!, .)■)] d r = 0 (6)
y\

Xi

S T d s + \ T d s = \ l [T(X, y 1) - n x , y 2)]dx = 0 (7)

A\Ai B2Bj . vj

Since the intervals (x,, x 2) and (j^, y 2) are arbitrary, it follows from
(6) and (7) that 7’(x1, j ’) = T(x2, y) and 7’(.v,>’1) = T(x, y2), and
so on.
Let S be a segment of the membrane at time t, bounded by the
contour C (Fig. 2.3). Let S x and C1 be the projections of S and C
on to the (x, >0 plane.18

18
 Partial Differential Equations in Physics

To determine the vertical component Pu of the tension force

acting on C, consider an element d/ of C and a point M on it. Let
T m be the tension vector at the point M at right-angles to d/. The
plane drawn through T m which is perpendicular to the (x, y) plane

will cut the (x, y) plane along the normal n to C1 at the point M x
(Fig. 2.3). Fig. 2.4 shows the curve L along which this plane cuts
the surface S. It is clear that

tana
Tu — T sin a T
|/ l + t a n 2a

Consequently,
r^dU ,, \ x du d /l
Pu $r„d/ = T-^— dl
on £ dn cos/?
where /? is the angle between the elements dl and d/t .
Since /? < y (as in the condition a < y discussed above)
cos /? ^ cos y = 1/ j/ l +m2+iPy w 1. Therefore,

P “ = T lC1 T n i l '

Application of Gauss’s theorem to this integral yields

Pu = $ 5 (uxx+ u yy) dx dy = T jj J V2u dx d>’

s, s,
The equation for the small transverse vibrations of a membrane
can now readily be obtained. Let f( x , y, t) be the density of the
resultant external force acting on the membrane at the point M(x, _y)
at time I along the u axis, and let p(x, y) be the surface density of
the membrane. Applying Newton’s second law to a part S x of the
2* 19
 Mathematical Physics

membrane in a time interval V2t = we obtain the required

integral equation

J J [u,(x,y, h ) —u,(x, y, t})]p(x, dy

s',
h <2
= jj ^ Jj T V2u dx dy dr-}- jj\ f ( x >y> T)dx dy dr
,i Sj I) S,

Assuming the existence and continuity of the corresponding deriva­

tives, we readily find the following differential equation for the small
transverse vibration of a membrane
T V 2u + f ( x , y, t) = p u tt
This is a hyperbolic equation. When p = const, it can be rewritten
in the form
a2V2u + F (x , y, t) = u„ (8)
where a2 = Tip, F{x,y, t ) = (1 /p) f ( x , y , t ) . Equation (8) is called
the two-dimensional wave equation.

2.4 THE EQUATIONS OF HYDRODYNAMICS

AND ACOUSTICS

The motion of a continuous medium can be characterised by a veloc­

ity vector v(x, y, z, t), a pressure p{x, y, z, t) and a density p(x, y,
z, t). We will take the medium to be an ideal liquid (gas). Consider
a volume D of the liquid bounded by the surface S. The pressure
acting on this volume is equal to

^ jj pn ds
s’
where n is a unit vector in the direction of the inward normal to S.
Gauss’s theorem yields

^ pn ds = —jj jj Vp dr
s ‘ b'
where Vp is the pressure gradient. In the absence of external forces,
the equation of motion for the volume D can be written in the form

20
 Partial Differential Equations in Physics

and, since D is arbitrary, this yields the following equation of motion

in Euler’s form

p~+Vp = 0 (9)

where dvjdt is the acceleration of a particle and is equal to

dv dv dv dv
~di+V' t e +Vz~di+V^ z
If there are no sources or sinks within D, the time rate of change of
the amount of liquid in D is equal to the flux of the liquid through
S, i.e.
d_
dt

Using Gauss’s theorem on the right-hand side of this equation, we

obtain

sss — +div(pz;) dr = 0

which yields the continuity equation for the medium

-^y-+div(pa) = 0 CIO)

Consider the adiabatic motion of a gas for which

where y = cp/cv, cp and cv are the specific heats at constant pressure

and constant volume, respectively, and p 0, p 0 are the initial values
of the pressure and density, respectively. The non-linear equations
given by (9 )-(ll) form a complete system of equations describing
the adiabatic motion of an ideal gas. If we substitute

o= p = p 0(l + a) (12)
Po
and confine our attention to small vibrations for which we can
neglect second order terms involving the ratio a, the velocity, and
the gradients of velocity and pressure, then equations (9) and (11)
21
 Mathematical Physics
can be considerably simplified (this is referred to as linearisation).
In fact, subject to these assumptions, we have
1
~~ j 7 ~~ ~~~ 0 O~\~0Z ...) "(1 o')
P Po 1-1-0' Po Po
Po(l+o)y ss Po(l+yo) (13)

±VP x^-V o (Po = const)

Po Po
div(pz>) « po div[(l+a)z>] a; p0 div(z>) (p0 = const)
Neglecting the higher-order terms in (9) and (10) we obtain

z>t+ a 2Vo = 0 a ypo

Po (14)
cr,+div(z;) = 0
Let us apply the divergence operator to the first equation in (14)
and the operator d/dt to the second. Subtracting one from the other,
we obtain
a2V2o = o. (15)
From (12), (13) and (14) we obtain the analogous equations for
p and p
2 V2p = P„
'2V2p = p t
Equations (15) and (16) are the equations of acoustics. They are
of the hyperbolic type. Such equations are also referred to as the
three-dimensional wave equations.
From the first equation in (14) we find that
t

v (x , y, z, t) = v(x, y, z, 0) —a2 Vo dr
6

= ®(-V, j ’, z, 0)—V (jj azo dr)

6
If we assume that at the initial time (t = 0) the velocity field has
a potential /(.v, y, z), i.e.
®!r=o = -V /(.v , y, z)
t
then »(.y, v, z, t) = —V |/(.v, y, z)-\-a2 jj a dr} = —Vu
o
22
 Partial Differential Equations in Physics
Consequently, the velocity field has a potential u when t > 0 and
this is given by

u = . f ( x , y , z)H n2 \<7dr
o

Differentiating this with respect to time, we find that

u, = a2a

a» = a2at

Substituting for crf and v in terms of a into the second equation in

(14), we obtain
a2V2u = utt (17)

It follows that the potential of the velocity field is also a solution

of the wave equation.

2.5 ELECTRIC FIELD IN A VACUUM

Maxwell’s equations in a vacuum may be written in the form

1 dH
curl E =
c dt
div E = 0
(18)
div H = 0
1 dE
curl H =
c dt

where H is the magnetic field and E is the electric field. Application

of the curl operator to the first equation yields

curl curl E = - - - - curl H (19)

c dt
Moreover,
curl curl E = V(div E ) - V 2E

23
 Mathematical Physics

In our case, curl curl E = —V2E, since div E = 0. Substituting this

into (19) and using the last equation in (18), we obtain the wave
equation for E:
c2 V2E = E„ (20)

2.6 THE EQUATIONS OF HEAT TRANSFER

AND DIFFUSION

We shall now derive the equation describing the distribution of

temperature in a body. Let u(M, I) be the temperature of the body
at the point M at time t. In deriving the required equation we shall
use Fourier’s law for the heat flux density zo in the direction n per
unit time

dn
where k is the thermal conductivity. The thermal conductivity may
be a function of temperature, position and time
k = k(u, M, t)
Consider a part D of the body bounded by the surface S. Let f { M , t)
be the density of heat sources and let us evaluate the heat balance
for D in a small time interval V2U

2 , = 5 J J / ( M , r)d rV 2/
D

s
where Qy represents the heat gain from the sources and Qz the heat
lost from D. The derivative dujdn is evaluated in the direction of the
outward normal to the surface S. Finally,

Qz = {jjjjjcpu, dxV2t
D

is the heat lost due to change in temperature, where c is the specific

heat and p the density of the material. Since energy must be con­
served, we have
Q i r 62+63
or

$$^ d0+$ $ \ f ( M ' t ) 4 r = $\ \ ci" h dT

S D D 24

24
 Partial Differential Equations in Physics
Application of Green’s theorem to the first integral yields

jj $ jj [div (k Viv)+/(A/, /)] dr = jj jj jj cpu, dr

*D D

and since D is arbitrary, we obtain the required heat transfer

equation
div(kVu)-\-f(M, t) = cpu, (21)
The diffusion equation can be established in a similar way.
According to Nernst’s law for the flux of matter w in the direction n

where u — u ( M , t ) is the concentration of the diffusing material

(gas or liquid) and D is the diffusion coefficient. In the equation
for we must now replace cp by the porosity c of the medium in
which the diffusion takes place. The diffusion equation is found to
be of the form
div (D Vm) + /( M , /) = cu (22)
Physical considerations indicate that the coefficients k and D
must be positive, and, therefore, Equations (21) and (22) are of the
parabolic type.
Problems involving the determination of the steady-state tem­
perature or concentration lead to the elliptical equation
di v ( k V u ) = —f ( M ) (23)
if k, c, p and / (and, correspondingly, D and c) are independent of
time t.

2.7 FORMULATION OF BOUNDARY-VALUE PROBLEMS

The solution of many problems in physics and other branches of

science by mathematical methods requires a preliminary mathemat­
ical formulation of these problems. This means that one must first
write down the equation or system of equations which the required
function or system of functions describing the phenomenon under
investigation must satisfy. The next step is to write down the addi­
tional conditions which the unknown function must fulfil on the
boundaries of the region within which it is to be determined. The
25
 Mathematical Physics
nature of these additional conditions can be illustrated by consider­
ing the problems considered in the last sections. For example, in
the case of the vibrations of a string or rod (Equations (3) and (5)),
one must prescribe the initial profile
u(x, 0) = <p(x)
and the initial velocity
ut(x, 0) = y>(x)
of all points on the string (rod). These are the initial conditions.
There are analogous conditions for any other wave equation. More­
over, we must write down the conditions at the ends of the string
(rod). Thus, if the law of motion of the ends (x = 0 and x = /) is
specified to be
«(0, 0 = /“i(0> u(l, t) =
these additional conditions involving the value of the variable
(displacement) on the boundary will be called boundary conditions
of type I. If we specify the variation of the force applied to the end
of a string (rod) and acting in the direction of the vibrations, the
conditions at the ends can be written in the form
Eu.x|x=o —_/i(o> Eux\x==i y^(0
or
ux(0, t) = ^ ( 0 , ux{l, t) = v2(t)
These involve the derivative of the variable on the boundary and
are called boundary conditions of type II.
Suppose that a spring is attached to the end of the rod at x = /
and acts in the direction of the x axis. The tension force Eux at the
end will then be balanced by the force a u due to the spring. The
boundary conditions at the end can then be written in the form
Eux(l, t) = —out (I, t)
where a is the spring constant, or
ux(l, t)+hu(l, 0 = 0
If the spring, in its turn, moves in accordance with the law
x= the boundary conditions can be represented by
ux(l, t)+h[u(l, 0 - £ ( 0 ] = 0
This is called a boundary condition of type III. At the left-hand end
(x = 0) it can be written in the form
ux(0, t)—h[u(0, t)—P(t)] = 0
26
 Partial Differential Equations in Physics

For the two- and three-dimensional cases the above three types of
boundary condition are of the form
u 's = ft (M , t) (type I) (24)

y(M, t) (type 11) (25)

huj — P(M, t) (type III) (26)

where dujdn is the derivative along the outward normal to the

surface S.
Similar boundary conditions are encountered in problems leading
to parabolic equations. Thus, if the temperature on the surface of
a body is specified, we have a boundary condition of type I. If we
specify the heat flux k dujdn through the surface S of the body we
have a boundary condition of type II. If, finally, the heat transfer
between the surface of the body and the surrounding medium at
temperature t) is in accordance with Newton’s law

-k^~ =
dn s
we have the boundary condition of type III.
Other types of boundary condition will be discussed later. The
boundary conditions considered above are linear because the
unknown function or its derivatives enter into them in linear forms.
They are called homogeneous if the right-hand sides {pi, v, fi) are
identically zero, and inhomogeneous in all other cases. It is evident
that similar boundary conditions are encountered in problems
leading to elliptic equations. Physical interpretation of each presents
no special difficulties.
We shall now formulate the corresponding three types of boun­
dary condition for equations of the form
div(fcVw) —qu+ f{M , 0 = puu (27)
and
div(fcVu) —qu-rf(M, t) = pu, (28)
where k, q and p are functions of the coordinates of M. All the
equations considered above belong to this type with q = 0.

First boundary-value problem Find the function u(M, t) satisfying

Equation (27) (or (28)) in the domain (M e D, t > 0) and the
additional conditions.27
27
 Mathematical Physics

1. The initial conditions

u(M, 0) = ut(M, 0) = y(M), M eD
(or u(M, 0) = cp(M)).
2. The boundary conditions
u(M, r)|s = n{M, t), t > 0

Second and third boundary-value problems These can be formulated

in a similar way by replacing the boundary condition of type 1 by
a boundary condition of type II or III.

Note All the above types of boundary condition can be represented

by the single relation

When y x — 0 we obtain a boundary condition of type I, y2 = 0

yields a boundary condition of type II and yx # 0, y2 ¥= 0 yield
a boundary condition of type III.
It is easy to imagine problems in which one is interested in the
unknown function u(M, t) at points M which are so far from the
boundary S that the effect of the boundary conditions on these
points can be neglected. This justifies the following formulation.

Cauchy's problem Find the function u(M, t) which satisfies Equa­

tion (27) (or (28)) for t > 0 at any point M of space, and the initial
conditions
u ( M , 0) = ut(M, 0) = ip(M)
(or u(M, 0) =
If we are interested in the values of the function u(M, t) for
large t, the effect of the initial conditions can be neglected. This
leads to the following problem.

Determination o f steady-state conditions It is required to find the

function u(M, t) which satisfies Equation (27) (or (28)) in D and
the boundary condition

(or no initial conditions).

28
 Partial Differential Equations in Physics

For an elliptic equation, the boundary-value problems can be

formulated as follows. It is required to find a function u(M) which
satisfies in the domain D the equation
div[k(M)Vu]~q(M)u = —f ( M )
and the boundary condition

| HM)

on the boundary S. y1 = 0, y2 — 0 and y x A 0, y2 # 0 correspond

to the first, second and third boundary-value problems, respectively.

PROBLEMS

1. The upper end of an elastic, uniform, vertically suspended,

heavy rod of length / is rigidly attached to the ceiling of a freely
falling lift which, having reached a velocity v0, is instantaneously
brought to rest. Formulate the boundary-value problem for the
longitudinal vibrations of the rod.
2. Formulate the boundary-value problem for the small trans­
verse vibrations of a string in a medium with a resistance propor­
tional to velocity, assuming that the ends of the string are rigidly
fixed.
3. An elastic rod (0 < x < /) of variable cross-section S(x) has
its ends fixed elastically, using springs. Formulate the boundary-value
problem for the longitudinal vibrations of the rod, neglecting the
deformation of the transverse cross-sections.
4. Formulate the boundary-value problem for the transverse
vibrations of a heavy string relative to the vertical equilibrium
position, if its upper end (x = 0) is rigidly fixed and the lower end
is free.
5. Consider the previous problem on the assumption that the
string rotates with an angular velocity co — const about the vertical
equilibrium position.
6. A light string rotating around the vertical axis with a constant
angular velocity co is in the horizontal plane and one of its ends
(x = 0) is attached to a point on the axis, whilst the other end is free.
At the initial time t = 0 all the points on the string are given small
deflections and velocities at right-angles to this plane. Formulate
the boundary-value problem for the deflections of points on the
string from the plane of equilibrium motion.
29
 Mathematical Physics
7. An elastic cylinder is displaced from its equilibrium state at
time t = 0 in such a way that its transverse cross-sections are rotated
through small angles 0 within their own planes about the axis of
the cylinder. Formulate the boundary-value problem for the resulting
small torsional vibrations of the cylinder if its ends are rigidly fixed
(or are free).
8. A current l(t) is passed at time t = 0 through a string
(0 < .\ ^ /), whose ends are rigidly fixed and whose resistanceis
negligible. The string is in a magnetic field H. Formulate the boun­
dary-value problem for the transverse vibrations of the string under
the action of the ponderomotive forces.
9. Two semi-infinite, elastic rods with identical transverse cross-
sections are joined end to end and form a single infinite rod. If the
densities and elastic moduli of the two rods are p x, E x and p2, E2,
formulate the boundary problem for the small longitudinal vibra­
tions of the rod under the action of an initial perturbation.
10. Formulate the boundary-value problem for the transverse
vibrations of a string with fixed ends, carrying a localised mass m
at the point x0.
11. Consider the transverse vibrations of an infinite string under
the action of a force F(t) applied at time t — 0 at the point a = 0
and moving along the string with velocity v0.
12. Find the equations for the a.c. voltage and current (i.e. the
equations of telegraphy) in a thin conductor with resistance R,
capacitance C, self-inductance L and leakage loss G per unit length.
Hint: use Ohm’s law and the conservation of charge.
13. Formulate the boundary-value problem for the electrical
oscillations in a conductor with negligible resistance and loss (but
finite inductance and capacitance per unit length) if the ends of the
conductor are earthed, one through a lumped resistance R0 and the
other through a lumped capacitance C0.
14. Consider the previous problem on the assumption that one
of the ends of the conductor (.v = 0) is earthed through a lumped
self-inductance L and an e.m.f. E{t) is applied to the other end
through a lumped self-inductance L£2).
15. Formulate the problem of the electrical oscillations in a loss­
less infinite conductor consisting of two semi-infinite conductors
connected through a lumped capacitance C0.
16. The lateral surface of a thin rod surrounded by a medium
at a temperature u,„ = <p(t) loses heat in accordance with Newton’s
law of cooling. Formulate the boundary-value problem for the
30
 Partial Differential Equations in Physics
temperature distribution in the rod if one end of it is maintained
at the temperature /j(f) and a heat flux q(t) is applied to the other.
17. Formulate the boundary-value problem for the temperature
distribution in a rod carrying a constant electric current / when
the surface of the rod loses heat to the surrounding medium main­
tained at zero temperature according to Newton’s law, and the ends
of the rod are held in large clamps of given thermal capacity and
high thermal conductivity.
18. Derive the diffusion equation for a medium moving with
velocity w(x) in the direction of the x axis if the surfaces of equal
concentration at each instant of time are planes perpendicular to
the x axis.
19. Derive the diffusion equation for a stationary medium whose
particles (a) decay (for example, an unstable gas) at a rate propor­
tional to the concentration, and (b) multiply (for example, neutrons)
at a rate proportional to their concentration.
20. Formulate the problem for the determination of the tem­
perature distribution in an infinite rod produced by joining two semi­
infinite rods of different materials if the two rods are joined (a)
directly and (b) through a massive clamp of thermal capacity C0
and very high thermal conductivity.
21. Formulate the boundary-value problem for a semi-infinite
rod, one end of which burns in such a way that the combustion
front propagates at velocity v and has a temperature <p(t).
22. Formulate the problem of the determination of the heating
of an infinitely thin rod exposed to a point source of heat of strength
Q moving along it with a velocity v0. Assume that all the heat is
communicated to the rod and that the thermal capacity of the source
is negligible.
23. Formulate the boundary-value problem for the cooling of
a thin circular ring, assuming Newton’s law of cooling and that the
temperature of the surrounding medium is u0.
24. Derive the equation for the propagation of a plane electro­
magnetic field in a conducting medium (i.e. in a medium in which
the displacement currents can be neglected in comparison with the
conduction currents).
25. Use Maxwell’s equations to derive the equation for the
electrostatic potential produced by a charged conductor, and for
the potential in a current-carrying conductor.
 3
The Method of Characteristics

In this chapter we shall be mainly concerned with the simplest wave

equation, i.e.
a2uxx+ f { x , t) = u„ (1)
The method of characteristics can be used to obtain solutions for
a number of problems which can be formulated in terms of this
equation. The principle of this method is best understood in terms
of an example involving the solution of Cauchy’s problem (Section
2.7), for the homogeneous wave equation.

3.1 VIBRATIONS OF AN INFINITE STRING

D’ALEMBERT’S FORMULA

3.1.1 It is required to find the function u(x, t) which is continuous

in the closed region (— oo < x < oo, i ^ 0) and satisfies the
equation
a2uxx = u„ ( — oo < .x < o o , t > 0 ) (2 )

and the initial conditions

u(x, 0) = <f(x), u,(x, 0) = ip(x) ( —oo < x < + o o ) (3)

To solve this problem we shall reduce Equation (2) in terms of

its characteristic curves. In the present case
an = a2, ai2 = 0, a22 — 1, V2 = a2
32
 The Method o f Characteristics
Consequently, the differential equations for the characteristic curves
are
dt _ _1 dt_ ___1_
d.v a’ d.v a
and .Y—cit — Cj, x + a t = c2 are their general integrals. Let us
substitute
I = x —at, i] = xf-at
The transformed equation is now of the form
= 0 (4)
If we assume that the required solution exists, then by substituting
it into (2) we obtain an identity. Consequently, the transformed
Equation (4) will also be an identity. Integrating this identity with
respect to rj we obtain
«{ = *i(£) (5)
where 0 1(s) is an arbitrary function. Integrating the identity given
by (5) with respect to £, we obtain
u = $#i(f)d £+F(i]) = ®(£)+F(ri)
or
u (x, t) = 0 (,v—at) -f F (y+ at) (6)
where <?(|) and F(i]) are arbitrary functions.
Thus, by assuming the existence of the solution of Cauchy’s
problem, we have come to the conclusion that it should be of the
form given by (6). To ensure that the functions u(x, t) given by (6)
are, in fact, the solutions of (2), it is necessary that the functions
0{z) and F(z) have first- and second-order derivatives. Subject to
these conditions, direct verification will show that each of the
functions 0 { x —at) and F(x-\-at) is a solution of (2).
Solutions of the form given by (6) include those which satisfy
the prescribed initial conditions (3):
u(x, 0) = rp(x) = 0(x)+F(x)
ut(x, 0) = y) (x) = —a0'(x)Jr aF'(x)
Integrating the last identity we obtain two equations for 0{z) and
F(z)
0(y)+.F(y) = rp(y)
y
— 0(y)+F(y) = — [ y>(z)dz+C (7)
a .)
*03
3 33
 Mathematical Physics

from which we find that

y

*0
Substituting these functions into (6), we obtain d’Alembert’s formula

^ <r(x-aiyv<p(x-r at) , 1 f , ^
u(A, t ) = ------------ 2 J ? '0 ) dz (8)
x — at

Thus, by assuming the existence of the solution of Cauchy’s problem,

we have come to the conclusion that it should be of the form given
by (8). Consequently, it is the only solution. If the function y(x)
possesses first- and second-order derivatives whilst <p(x) possesses
first-order derivatives, then (8) yields the required solution of
Cauchy’s problem, (2)-(3). This can be verified by direct substitution
of the right-hand side of (8) into (2) and (3). By constructing an
explicit solution of Cauchy’s problem we have demonstrated its
existence.
3.1.2 Let us consider now the physical interpretation of the solu­
tion it — <P(x—at). The function u(x, /) will be called the displace­
ment at the point x: at time t. Consider a point a*o and imagine

that an observer begins to move with a velocity a in the positive .v

direction at time t = 0. At time tx it will be at the point x x = x0-\-atx.
The displacement which the observer will see at the point xq at time
L will be it = 0(.x,—at,) = #(.y0). It follows that at any given time34
34
 The Method o f Characteristics
the observer will see a constant displacement &(xo) at the point at
which he is located. It follows that the initial profile u(x, 0) = &(x)
will move with a velocity a in the positive .v direction as if it were
a rigid system which does not undergo a change of form (Fig. 3.1).
In view of these properties, the solution u = &(x—at) is called
the forward travelling wave solution. A similar interpretation can
be given to the solution w = F(x--at). This solution is called the
reverse travelling wave solution. In this case, the profile moves as
if it were a rigid system moving in the negative x direction with
velocity a. It follows that any solution of (2) can be written as
a superposition of forward and reverse travelling waves. The above
method of obtaining the solutions of Cauchy’s problem is called
the method of characteristics or the method of travelling waves.

3.2 CONTINUOUS DEPENDENCE OF THE SOLUTION OF

CAUCHY'S PROBLEM ON THE INITIAL CONDITIONS.
GENERALISED SOLUTION

3.2.1 D ’Alembert’s formula (8) gives the solution of Cauchy’s

problem (2)-(3) on the assumption that the initial functions <p(x)
and yj(x) possess bounded derivatives <p'(x), <p"(x), y>'(x). However,
it is not difficult to find problems for which the initial functions
<p(x) and ip(x) do not have these properties. It is sufficient, for
example, to specify the initial deflection of a string in the form
of the broken line shown in Fig. 3.2.

To show how the solution of Cauchy’s problem can be obtained

in such cases, we shall prove the following theorem.

Theorem Let ux(x,t) and u2(x,t ) be the solutions of Cauchy’s

problem (2)-(3) subject to the initial conditions
Ui(x, 0) = 7 t(x ), uu (x , 0) = (*)
and
w2(x, 0) = rPl( x ) , u2t(x, 0) = yj2(x)
For any e > 0 and (t > 0 we can then find a quantity a > 0,
a function of e and t lt which is such that the inequalities
\<Pi(x)—(pz(x) | < d, Iv’iW —V2(*) I < 8, — oo < x < co

3* 35
 Mathematical Physics

lead to
|Ui(-Y, t ) - U z(x, t) | < £, -CO < .V < CO, t < /,

Proof Using d’Alembert’s formula for u^x, t) and u2(x, t) we

obtain

ux(x, t)—u2(x, t) = ^-[<Pi{x—at)—<p2(x—at)]

+ - y [(pi(x+at)-<p2(x+at)]
x+ a t

+ 2~ W i ( z ) - y 2(z)] dz
x — at

and, consequently,

M * , t)—u2(x, t) | < ^\<px{x—at)—tp2( x —at)\

+ -y l <Pi{x+at)-cp2{x+at)\

x+ at

+ \
x — at
Idz
x+ at

1 1 I f
< T 4 + T 4 + -£r 1
x — at
idz
= <5+<5t < 5 (1 + 0
If we let 5 = e/(\ + b), the inequality
| itX(x, t)—u2(x, t) | < e
will be satisfied for all oo < .v < oo, t ^ /,, which was to be proved.

The above theorem can readily be expressed in words: small

changes in the initial values of Cauchy’s problem lead to small
changes in the solution.
In practice, the initial conditions are deduced from measure­
ments and are therefore subject to uncertainties. The above theorem
shows that small errors in the determination of the initial conditions
36
 The Method o f Characteristics
will lead to small changes in the solution of Cauchy’s problem.
The theorem also shows one of the possible ways of obtaining
solutions of Cauchy’s problem when the initial functions <p(x)
and y(x) do not have the required derivative (Fig. 3.2).

3.2.2 Let us return now to Cauchy’s problem (2)-(3). We shall

assume that the initial functions <p(x) and xp{x) are not identically
zero on finite segments, are continuous everywhere, and that the
function y>(.\') possesses a first-order derivative. These functions
can be uniformly approximated to by the differentiable functions
<p,,(x) and y„(x) so that

<Pn(x) = : <p(x) (n —>00), y n(x) zz: rp{x) (n —>00)

where <pn(x) possesses a first and second derivative and tp„(x)

possesses a first derivative.
If we take <p„(x) and ip„(x) as the initial functions for the Cauchy
problem, they will define a unique solution for the problem, un(x, t).
Consider the difference un+k(x, t)—u„(x, t). Since the sequences
{^(x)} and {y>„(x)} are uniformly convergent for any e > 0 tA > 0,
it is possible to find N such that for any n > N and any positive
integral values of k we have

B 6
I <Pn(x)—<pn+k(x) | < - and | y>„+k(x)—ipn(x) \ <

for all —00 < jc < 00. In view of the theorem proved above, we
have the following inequalities for all t < tx and —00 < x <00

I un+k(x, t )—u„(x, t) | < e

These hold for any n > N and any positive integral k. However,
this means that the sequence of solutions {u„(x, 0} converges
uniformly in the above range of xc, t to some function u(x, t). This
function is called the generalised solution of Cauchy’s problem
(2)-(3). It is given by

u(x, t) = lim un(x, t) = -^-lim [<pn( x —at)+<pn(x-\-at)]37

x + al

x — at

37
 Mathematical Physics

or

This function and its derivative ut(x, t) will assume the prescribed
values cp(x) and y(.v). It follows that in the above case, the d’Alem­
bert formula will also yield the (generalised) solution of Cauchy’s
problem. The problem can also be solved in another way by using
the generalised functions and their convolution (see Appendix,
Section A.l).
Definition The fundamental solution G(x, t) of the wave equation
a2uxx = u,t is defined as the solution of the Cauchy problem
a2Gxx = G,t , G(x, 0) = 0, Gt(x, 0) = <5(x)
It is readily verified by direct substitution that

G(x, t) = — [y]{x+at)-r]{x—.at)\

function. In point of
fact (see Appendix, Section A.l),

Gx(x, t) = — [b{x+at)—b{x~at)\

Gxx(x, t) = — [b'(x+at)—b'{x—at))

Gt(x, t) = — [b {x+at)+b (x—at)}

G„(x, t) = —-[^('V- \-at)— b'(x—>at))

Consequently,
a2Gxx(x, t) = G tt(x, t)

G( x , 0 ) = 2a- h ( x ) - V(x)] = 0

Gt(x, 0) - y [(S(A-)-M(A')] = b (x)

38
 The Method o f Characteristics
The solution of Cauchy’s problem
a2vxx = vtt, v(x, 0) = 0, v,(x, 0) = y>(x)
will be written in the form of the convolution
v(x, t) = G (x, t ) * y(x) (9)
In fact, by evaluating the derivatives of the convolution (see
Appendix, Section A.l), we obtain
vxx = Gxx * V, v,t = G„ * rp
and, consequently,
a2vxx—v„ = (a2Gxx- G tt) * tp = Q*ip = 0
v (x, 0) — G(x, 0) * ip (.x) = 0 * ip (.x) = 0
v t(x, 0) = Gt(x, 0) *y>(x) = d(x) * y ( x ) = y(x)
The convolution G * ip can also be written in the form
oo Qt

v{x,t)= jj G($, t)y>(x—f )d f = ~ jj y>(x—$)d$

— oo — at

Substituting x —$ = z in the last integral, we obtain

x+ at

v ( x , t ) = ~2- J v(z)dz (10)

x —at

We therefore note that in Equation (9) and, consequently, in

Equation (10) the function yj(x) can be any integrable (and even
any generalised) function.
If R (.x, t) is a solution of Cauchy’s problem,
a2Rxx = R,t, R (*, 0) = 0, R , ( x , 0) = cp(.x)

the function w(x, t ) = —R (.x, t) is a solution of the Cauchy problem

a2wxx = ™(x, 0) = cp(x) , w, ( x , 0) = 0

In fact, differentiating the identity

° 2R XX = R ,t

with respect to t, we obtain a2{R,)xx = (Rt)u , i.e.

 Mathematical Physics

Next,
w (.v, 0) = R,(x, 0) = cp(x)
wt(x, 0) = Rtt(x, 0) = G„(x, 0) * <p(x) = 0 * y(x) = 0
If the function cp(z) is continuous, then w(x, t) can be written
in the form
x + at

v ,{ x,t ) = R,(X, t ) = + l ^ \
I X -Q t ]

<p(x—at)+<f (x + a t)
= 2
i.e.
, . cp(x— a t)+ < p { x + a t)
w ( x , t) = --------------2 -------------

The solution of the arbitrary Cauchy problem

a2uxx = u,,, u ( x , 0) = cp(x), u, (x ,0) = y>(x)
where cp(x) is a continuous function and y>(x) is integrable (this
includes piecewise-continuous functions) will be the sum
u — v + w — G ( x , t ) * y ( x ) + G t ( x , t) * cp(x)

or
x+at
/ .x tyix—at)+cp(x+at) 1 f / , ,
« < * , / ) = -------------2------------ + 2a ) w(z)dZ
x — at

It follows that, in this case, the solution can again be deduced

from d’Alembert’s formula. Its derivatives are then treated as
the derivatives of generalised functions and coincide with the
ordinary derivatives when the latter exist.
We note that the expression given by (8) therefore yields the
solution of the Cauchy problem even for arbitrary generalised
initial functions tp(x) and y>(x).

3.3 VIBRATIONS OF A LOADED INFINITE STRING

3.3.1 Now that we know how to obtain the solution of Cauchy’s

problem for the homogeneous wave equation (2), we can readily
establish the solution of this problem for the inhomogeneous wave40
40
 The Method o f Characteristics
equation (1). The method is the same for all linear hyperbolic
equations and we shall therefore confine our attention to the general
equation
div (k V u ) —quf-f(M, t) = putt (11)
where k, q and p are known functions of the coordinates of the
point M.
Thus, suppose that it is required to solve Cauchy’s problem
for (11) subject to the initial conditions
u(M, 0) = ut(M, 0) = y>(M) (12)
We shall split this problem into two parts:
1. Cauchy’s problem for the homogeneous equation
di v (k V v ) —qv = pv„ (13)
subject to the initial conditions
v(M , 0) = <p(M), vt(M, 0) = y>(M) (14)
and
2. Cauchy’s problem for the original equation
div(fcVw)—qw+f{M, t) = pwtt (IT)
subject to the initial conditions
w{M, 0) = 0, wt(M, 0) = 0 (15)
It is evident that it =
Let us suppose that we know how to solve Cauchy’s problem
(13)-(14). The solution of Cauchy’s problem (ll')-(15) can then
be determined as follows. Consider the function W(M, t, r) which
satisfies the homogeneous Equation (13) and the initial conditions
f ( M , r)
Wt_x = 0, Wt\u (16)
p(M)
By hypothesis we know how to solve this problem. The required
solution of Cauchy’s problem (2) will then be of the form

zv(M,t) = \ w ( M , t , T ) d r (17)
o
In fact,

wt(M, t) = W\t=x+ \ w t(M, t, r)d r

o
41
 Mathematical Physics

and using the first of the conditions in (16) we obtain

t
wt{ M ,t) = \ w t( M , t , r ) d r (18)
o
From (17) and (18) it follows at once that w(M ,t) satisfies the
initial conditions. Differentiating (18) with respect to t once again
and using the second of the conditions in (16), we obtain
t t
zvtt = Wt\z=t+ ^ W tt{M, t, r) dr = \ T) dr

Consequently,

pwt t = f { M , t ) + \ p W n dr (19)
o
Let us evaluate the quantity div (kVw)—qw. The operation
div (A:V) can, of course, be carried out under the integral sign.
We have

div ( k V w ) - q w = J {div(& V W ) - q W } dr (20)

o
Since the function W(M, t, r) is a solution of (13), it follows from
(19) and (20) that the function w(M, t) defined by (17) is a solution
of (IT ) and, consequently, a solution of Cauchy’s problem (2).

3.3.2 Let us now use the above method to solve Equation (1).
It is required to solve the Cauchy problem

a ^ x x + f i x , t) = utt
u(x, 0) = <p(x), ut(x, 0) = y>(x)

We shall split this problem into two parts.

1. Cauchy’s problem for the homogeneous equation with given
initial conditions

a2vxx =
fl(-v, o) = <7(a), V,(x, 0) = y>(x)
42
 The Method o f Characteristics

2. Cauchy’s problem for the given equation with zero initial

values
a2wxx+ f ( x , t) = zvlt,
zo{x, 0) = 0, zot(.v, 0) — 0

We then have u = v-\-w.

The function v(x, t) can be obtained from the d’Alembert
formula
x-j-at
<p(x—at)-r<p(x-\~at) 1 f , XA
v ( x , 0 = - -------- - x f1------------ r-2a ) y ( z ) d z
x —at

In accordance with the foregoing discussion

zv(x, t) = ^ W(x, t, t) dr
o
where W(x, t, r) is a solution of Cauchy’s problem
a2Wxx = Wtt,
w \ t=x= 0 , w t\ =z= f ( x , T)
and, consequently, can be written in accordance with the d’Alembert
formula
x X a { t — x)

W(x, f, t) = - ^ 5 / 0 »T) dz
x — a t t — x)

Therefore,
t x - \ - a ( t — t)

0 = 2^ 5 S f{z,x)dzdx
0 x -a (t-x )

3.4 SOLUTION OF BOUNDARY-VALUE PROBLEMS

ON A SEMI-INFINITE STRAIGHT LINE

3.4.1 We shall now consider boundary-value problems on a semi­

infinite line. To begin with, we shall prove the following two lemmas.

Lemma 1 If in Cauchy’s problem (2)-(3) the initial functions

cp(x) and ip(x) are odd with respect to x = 0, the solution of this
problem at x = 0 is
w(0, t) — 043
43
 Mathematical Physics

Proof Substituting x = 0 in d’Alembert’s formula which gives

a solution of Cauchy’s problem (2)-(3), we obtain
at

Since <p(x) is an odd function, it follows that cp(—at) = —<p{at).

Therefore, cp{—at)-\-rp{at) = 0. Since y(x) is odd, the integral
at

— at

is also zero. Consequently, i/(0, t) = 0.

Lemma 2 If in Cauchy’s problem (2)-(3) the initial functions

9>(y) and rp(x) are even with respect to x = 0, the derivative ux(x, ()
of the solution of this problem at v = 0
ux(0, t) = 0
The proof of this lemma is similar to that given above. It is
based on the fact that the derivative <p '( y ) is odd.

3.4.2 Consider the homogeneous boundary-value problem

a2uxx = u„
u(x, 0) = cp(y ), u,(x, 0) = tp(x), 0 < y < oo (21)
i/(0, /) = 0
We shall assume that 95(0) = y>(0) = 0. To solve this problem we
cannot use d ’Alembert’s formula directly because the difference
x —at in this expression may be negative, and the initial functions
cp(x) and ^(y) are not defined for negative values of the argument,
in accordance with (21).
We shall proceed as follows. We shall continue the functions
9?(y) and y(x) along the negative part of the y axis by defining new
functions

The function
x+at

44
 The Method o f Characteristics
will then be a solution of the boundary-value problem. In fact, it
satisfies the homogeneous wave equation because it is a super­
position of forward and reverse waves. It satisfies the boundary
condition in view of Lemma 1. Let us verify the initial conditions:

„(Xj0) = ^ w + ^ j Vi(z)d2

= 9fi(*) = <KlL x>0

= viW = v W . -v > o
It follows that the initial conditions are also satisfied.
The boundary-value problem
Q Ux x ■ Ut t ,

u(x, 0) = <p(x), u,(x, 0) = y>(x), 0 < x < co (22)

ux( 0 , t ) = 0
can be solved in a similar way except that the functions <p(.x) and
yj(x) are now continued as even functions along the negative part
of the straight line.

3.5 REFLECTION OF WAVES AT FIXED

AND FREE ENDS

The solution of the boundary-value problems (21) and (22) can

be written in the form (6):
u(x, t) = &(x—at)-\-F(xf-at)
We shall interpret u as a displacement. The displacement due to
the forward wave is constant along the characteristic x —at = cx,
i.e. &(x—at) = 0(Cj). Along the characteristic x f- a t = c2 the
displacement due to the reverse wave is also constant, i.e. F(xJ\-at)
= T(c2). It follows that the displacements propagate along the
characteristic curves.
Let us draw the two characteristics x —at — x —ato and x-\-at
= x0+ a t0 through the point (x0, t0) in the (x, t) plane. They will
intersect the x axis at points —x x and x 2, respectively (Fig. 3.3).45
45
 Mathematical Physics

The displacement u(x0, l0) at point x0 and time l0 consists of

the displacement due to the reverse wave which arrives from the
point a'2 and the displacement due to the forward wave which
arrives from the point —X[. However, at the point —X! there can
be no displacement at t = 0 because the initial conditions in the

Fig. 3.3

problems defined by (21) and (22) are prescribed only on the semi­
infinite straight line for which x > 0. However, from the boundary
condition for the problem defined by (21) it follows that
0 ( _ z) = - F(Z)
Consequently, 0 ( —xy) = —i 7(x1). It follows that instead of the
forward wave arriving from the point —Xj we can consider the
reverse wave which has left the symmetric point x: at time t = 0.
This reverse wave will reach the point x = 0 in a time t1. From
t = t x onwards, it must be replaced by the forward wave which
has left the point x = 0 at time t = t x and has propagated with
displacement —0 ( —xi). It follows that when the boundary con­
dition »(0, /) = 0 is satisfied at x = 0 we have the phenomenon
of reflection in which the magnitude of the displacement is conserved
but its sign changes.
It can be shown in a similar way that when the boundary con­
dition ux(Q, t) — 0 is satisfied at x = 0, we have the phenomenon
of reflection with the conservation of the magnitude and of the
sign of the displacement.
The method of characteristics can also be used to construct
solutions of homogeneous boundary-value problems on a finite
segment subject to boundary conditions of types I and II. To be
specific, consider the first boundary value problem
a2uxx = utt
u(x, 0) = <p(x), wf(x, 0) = y>(x), 0 < x < / (23)
u(0, t) = 0> r/(/, t) = 0, t> 0

46
 The Method o f Characteristics
To obtain the solution let us continue the initial functions <p(x)
and y(x) along the entire line and assume that they are odd with
respect to x = 0 and x = /. Let 9?2(x) and V>z(.x) represent the
functions continued in this way. The function
x+ at

it(.\, t) = ------------------------- Jv2(z)dz

X — Qt

will then be a solution of the boundary-value problem. This function

satisfies the boundary conditions in view of Lemma 1. The initial
conditions can be verified directly as in the case of the semi­
infinite line.
Note that if the function tp(x) is odd (even) with respect to
two points, say, x = 0 and x = /, then it is periodic with period 21.
In fact, since <p(x) is odd with respect to x = /, we have the identity
<p(I—z) = —9?(/+z). Substituting z = x + /, we obtain <p(—x)
= —9:(x+2/). Since <p(—x) = —tp(x), it follows that (p(x-f2l)
= 9?(x). The initial functions ^(x) and y(x) can therefore be con­
tinued to the segment (—/, 0) as odd functions and then periodically
with a period 21 along the entire straight line.

3.6 PROPAGATION OF THE BOUNDARY VALUE ALONG

A SEMI-INFINITE STRAIGHT LINE
Consider the inhomogeneous boundary-value problem on. a semi­
infinite straight line
a2uxx = u,„
u(x, 0) = 9?(x), w((x, 0) = ip(x), x > 0
i/(0, t) = u(t), t > 0
It can be split into two problems, namely:
1. the homogeneous boundary-value problem
a2vxx = vn ,
v(x, 0) = ff (x), v t(x, 0) = ip (x), x > 0
£>(0, t) = 0
2. the problem involving propagation of the boundary value
a2wxx = a>„,
w{x, 0) = 0, w,(x, 0) = 0
zv(0, t) = u(t), t > 0
We then have u = v-fw.

47
 Mathematical Physics
We already know how to solve the problem for v(x, t). The
propagation of the boundary value will now be considered. Since
the only reason for the appearance of a perturbation is the situation
at the boundaries, we shall seek the solution in the form of the
forward wave
zv(x, t) = 0>{x—at)
From the initial conditions we find that
w(x, 0) - 0(x) = 0, x > 0
It is evident that wt(x, 0) = —a0'(x) = 0 will also be satisfied
for x > 0. From the boundary condition we find
0 ( —at) = I> 0
and therefore
0 z > 0

z < 0

or 0( z) = ??(—zja)/x{—z/a), where ?j(£) is the unit step function

(equal to unity for f > 0 and zero for £ < 0). Consequently,

The propagation of boundary conditions of type II can be

discussed in a similar way: ux(0, t) = v(t).
Using the phenomenon of reflection considered above, we can
readily solve the problem of the propagation of the boundary
conditions of type I or type II along a finite segment.
The method of characteristics can also be used to solve a number
of other problems, which can be formulated in terms of one-dimen­
sional inhomogeneous wave equation. However, the problems
which we have already discussed illustrate all the main features of
the method and we shall therefore confine our attention to them.

3.7 VIBRATIONS OF AN INFINITE MEDIUM.

POISSON’S FORMULA

Many problems leading to the two- or three-dimensional wave

equations can be reduced to problems of the form already discussed.
Let us consider some of them.
48
 The Method o f Characteristics
3.7.1 Cauchy’s problem for the homogeneous wave equation in
three-dimensional space:
a2 V2u = utt (24)
u(M, 0) = <p(M), ut(M, 0) = y>(M) (25)
To solve this problem consider the auxiliary function u(r, t) which
is the average of the required solution over a sphere centred
on the point M and having a radius r:

«(G 0 = 4 ^ 2 - (26)
SM
where P denotes the variable integration point.
Let da) be an element of solid angle subtended by the area
do- at M so that dcr = r2 dco. It follows that we can also write

“(r’° =^SSM(p,° dco (27)

SM
Using the mean value theorem in Equation (26) and letting r tend
to zero, we obtain
77(0, 0 = u(M, t) (28)
It follows that to find the function u(M, t), it is sufficient to find
the function u(r, t). To do this we shall require the following
lemma.

Lemma
V2m =V2(m)
where on the left-hand side the Laplacian V2« is evaluated with
respect to the coordinates of M, and the right-hand side is evaluated
with respect to the coordinates of point r. Henceforth, we shall
omit the subscript r.

Proof Let DT M be a region bounded by the spherical surface S rM.

Using Gauss’s theorem we have

4 49
 Mathematical Physics

Applying (27) to the last integral, we obtain

d 'm

On the other hand,

r r

jj jj jj V2u dr = jj \ ^ 2u dc) dp = $ (4jip2 V2u) dp

° rM ° *Sf °
and consequently,
r

jjp2 W2u dp = r2^

o»t

Differentiating this with respect to r we obtain

V2u = \ ~ ( r 2ur) = V2(u)

r or
This completes the proof of the lemma.

Let us now suppose that the solution of the problem (24)-(25)

exists. Averaging the identity
a2 V2u - utt
over the sphere and using the above lemma, we obtain
a2V2u = un
or
1 d , _
a r i d ? (r Ur) =
or
a2{rurtJr l u r) = rTi„
Substituting v = ru, the last result can be rewritten in the form
a \ \r = v„
It follows that the function v(r, t ) satisfies the one-dimensional
wave equation.50
50
 The Method o f Characteristics

If we take the average of (25), we obtain

»(/•, 0) = <p(r) = jj jj <p(P) da

(29)
ut(r, 0) = y(r) = ^ ^ 2 J J V(p ) da

Let

fM 0 = r<Kr) and V\(?) = ry(r)

It is evident that
®0", 0) = <Pi(r), v t(r, 0) = y>i(r), v(Q, t) = 0

Therefore, for v(r, t) we have the following problem on a semi­

infinite straight line

a-v„ = v„
v(r, 0) = <px(r), vt(r, 0) = y,(r)
t;(0, t) = 0

To solve this problem, the initial functions cpx(r) and y>i(r)should,

in accordance with Section 3.4, be continued as odd functions
along the semi-infinite line (—co, 0), and the d’Alembert formula
should be used for the continued functions tp2(r) and y>2(r). The
continued functions y(r) and y>(r) will be odd [we shall retain the
previous notation for the continued function, namely, qi{r) and ^(r)].
The solution of the problem for v(r, t) will be of the form
r-\-at

^ 0 - » ( f + a ) + * (f- a 0 + T J V2(z)dz
r— at

Consequently,
r+ ai
_ <Pi{r-\-at)+<p2{r—at) , 1
u(r, 0 = ~ ^ "v + J V2(z)l
r
r — at

If we substitute r — 0 in this formula, we obtain w(0, t ) = 0/0.

4* 51
 Mathematical Physics

The function u(0, t) can be evaluated with the aid of 1’Hopital’s

rule (bearing in mind the definition of <p2 and y>2) :

w(0, 0 = ^ { y { a t ) + y ( —a t ) + a i y ' ( a t ) - a t y ( —at)}

+ {aiy(at)+aty)(—at)}

Since the functions y{z) and Tp{z) are even, whereas y'(z) is odd,
it follows that

u(0,t) = y(at)-\-aiy'(at)+ty(at) = -^-{fp(«0} + *v(a 0 (30)

Using (28), (29) and (30), we obtain Poisson’s formula for the
required solution of Cauchy’s problem (24)-(25):
1 d f f y(P)d<r i f f v C f ) da
u(M, t) = (31)
An a dt J J at Ana J J at
cat
«i/
Thus, if we assume the existence of the solution of Cauchy’s
problem for the three-dimensional space, we find that it must be
given by (31). Consequently, this is the only solution.
3.7.2 We can now solve Cauchy’s problem in three-dimensional
space for the inhomogeneous wave equation
a2V2u + f { M , t) = utt
u(M, 0) = <p{M), ut(M, 0) - rp(M)
We can split it into two subsidiary problems:
1. a2V2v — vtt
v (M, 0) = <p(M) , v,(M, 0) — ip (A/)
The solution of this problem is given by Poisson’s formula.
2. a2V2w-\-f(M, t) = wtt
w ( M , 0) = 0, wt( M , 0) = 0
It is evident that u = v+ w .
Problem 2 can be solved by the method described in Section 3.3.
In particular, we must first solve the auxiliary problem
a2V2W = Wtt
w \ , ^ = 0, n a = t= / ( M , t)
52
 The Method o f Characteristics
and then from Poisson’s formula we have

As was shown above,

= J W (M ,t, r)d r
o
or

Substituting a(t—r) = r in the outer integral, we obtain

at

r
do- dr
0 cr

where r is the distance between the integration point P and the point
M, i.e. r = rMP. This integral can, clearly, be written as an integral
over the region D% bounded by the sphere S$:

(32)

If the external driving force f ( M , t) is different from zero only

at the single point M 0, where it is equal to f(t), the wave equation
can be written in the form
az V2u + f(t)d (M , M 0) = utt
where <5(M, M 0) is the <5 function with a singularity at M 0 (see
Appendix).
The solution of this equation which has zero initial values
[and is therefore due only to the point force f(t)\ can be written
in accordance with Equation (32) in the form

/ l/--)(5 (P ,A /o )
u - dv
r

53
 Mathematical Physics
Using the fundamental property of the 8 function, and remembering
that for any continuous function <p(M)
0 if M0 $ D
S $ 5 ? ( p ) W M 0)dV= |
D (p(Mo) if Mo g D
we obtain
0, where at < rMMo
u(M, l) 1_______ 1 /./ r MM (33)
- , where at > rMMo
.4 Jta2 rMMoJ \ a

3.7.3 The solution of Cauchy’s problem for the homogeneous

wave equation in two-dimensional space
a2 V2u = u„
u(x, y, 0) = tp(x, y ) , u,(x, y, 0) = y>(x, y) (34)

can also be obtained from Poisson’s formula. In point of fact,

if the functions <p(P) and ip(P) in Equation (31) are independent
of z, then the integrals over the surface S%} of the sphere can be
reduced to integrals over the major circle of this sphere in the
(x, y) plane (Fig. 3.4). The integral over the upper half of 5"/ is
equal to

<
p(P\) d a
uppers^ a at cosy

54
 The Method of Characteristics

where y is the angle between the normals to the (x, y ) plane and
to the sphere 5 $ at the point P. It is evident that

\P P u IPtMl2
cosy
\MP\ at

\ / ( a t f —( x - £ ) 2- ( y - r ] ) 2
at
where £, rj are the coordinates of the point P r which is the projection
of P or, to the (x, y) plane, and (x, y) are the coordinates of the
point of observation M. Therefore,

cc <P(P) dcr = f f <p(£> V) d^ dr)

JJ at V(etQ2—(x—£)2—(y—v)2
upper

and, similarly,

cc <P(P) dr = f f 9>(f, rj) df dr]

JJ at " j I/a 2t 2- ( x - i f - ( y - r j ) 2
lower

Using a similar transformation for the second integral in Poisson’s

formula, we obtain a solution of Cauchy’s problem (34) in the
form
, —J _ A [[ y(£, v) d£ drj_____
^ 2na dt j 1a2t2—(x —£)2—(y —rj)2

+ f f y(£. V ) & d r )
(35)
Ina j J |/ a2t2— {x — £)2—(y — rj)2
SM
We are now in a position to solve Cauchy’s problem in two-
dimensional space and for the inhomogeneous equation. It can be
reduced to the problem just considered and to Cauchy’s problem
for the inhomogeneous equation with zero initial values. The latter
problem can be solved by the method described in Section 3.3.
We shall not repeat all the steps leading to this solution and shall
merely quote the final result

w(x,y, 0 = _ L f / [[ T)d £ dr (36)

2 n a o L Jt-T) Va2( t - r ) 2- ( x ~ i ) 2- ( y ~ T j ) 2

55
 Mathematical Physics

We leave it as an exercise for the reader to obtain the solution

due to a point force f(t).

3.8 PHYSICAL INTERPRETATION OF POISSON’S FORMULA

Let us return now to Poisson’s formula. Suppose the initial functions

tp(M) and y>(M) are non-zero only within a domain D bounded
by the surface S (Fig. 3.5). Let us observe the state of the medium

at a fixed point M. For sufficiently small t the surface Sgj of the

sphere centred on M will not cut the domain D. The integrals
on the right-hand side of Poisson’s formula will therefore be zero.
Consequently, for these values of t we have u(M, t) = 0 (the distur­
bance has not reached the point M). Let dx be the distance of the
nearest point on the surface S from M, and let d2 be the distance
of the most distant point on S from M. For t e = dxja,
t2 = d2ja) the surface S% of the sphere will cut the domain D.
Therefore, the integrals in Poisson’s formula will not be zero and,
therefore, for these values of t we have u(M, t) # 0. For t > ti
the sphere Sgf will not cut the domain D and, consequently, u(M, t)
will again be zero.
Let us suppose now that from each point in the domain D
disturbances are emitted and propagate with velocity a in all direc­
tions (i.e. the Huygens principle). The above changes in the function
u(M, t) with time can then be physically interpreted as follows.
For t < ti the disturbances have not reached the point M. At time
t = tx the leading edge of the wave reaches the point M and the
wave passes M for t1 < t ^ t2. At time t = t2 the trailing edge
of the wave passes M and from this moment onwards the point
M remains at rest.
Suppose that the initial functions <p(x, >>) and ip(x, y) for the
two-dimensional case are non-zero only in the domain D bounded 56
56
 The Method o f Characteristics

by the curve S (Fig. 3.6). For t < t, the circle Y m does not contain
points belonging to D, and therefore the integrals in (35) are zero so
that u(x, y, t) = 0. For any t > tx the circle Y m contains the domain
D or part of it and, therefore, u(x, y, t) ^ 0 for these values of t.

The times t x and tz are defined as before. Thus, in the two-dimen­

sional case, the wave has a leading edge (it reaches the point M at
time t = tf) but no trailing edge. Huygens’ principle is then invalid.
This is readily understood if we recall that the two-dimensional
case which we have been considering is, in fact, a three-dimensional
problem in which the region of non-zero values of the initial func­
tions <p(M) and \p{M) is an infinite cylinder, whose generators are
parallel to the z axis. It is evident that the spherical surface S%
will cut the cylinder for any t > 11, and therefore the integrals in
Poisson’s formula will be non-zero for all t > tj.

PROBLEM S

1. An infinite string is excited by a triangular initial deflection

in the range (c, 2c) with corners at the points c, 3c/2 and 2c. Draw
the profile of the string for times tk = ck/2a, where k = 1, 2, 3.
2. Solve Problem 1 if the initial deflection is in the form of
a triangle lying in the ranges (—2c, —c) and (c, 2c) with corners
at the points —2c, —1.5c, —c, c, 1.5c, 2c.
3. An infinite string is given an initial transverse velocity v0
= const in the range —c < x < c. Determine the vibrations of
the string. Draw the profile of the string for times tk — ck/2a
(k = 1, 2, 3).
4. A semi-infinite string with one end rigidly fixed is excited
by an initial deflection which is no-zero only on the segment (c, 3c),
where it takes the form of a broken line with comers at the point
c, 2c, 3c. Draw the profile of the string for times tk = ckj2a
( k= 2, 4, 6).57
57
 Mathematical Physics

5. At the initial time t — 0 a semi-infinite string with one end

rigidly fixed is given a transverse impulse at the point x = x0.
Determine the vibrations of the string.
6. An infinite elastic rod consists of two semi-infinite homo­
geneous rods, joined at the point x = 0. The densities and elastic
constant of the two materials are p n E x and p2, Ez, respectively.
Suppose that the wave w,(x, t) = / ( t —x/a,) travels along the rod
from the region x < 0. Find the reflected and refracted waves.
When will they exist? Investigate the solution for E2 -» 0 and
E2 —> oo.
7. A constant e.m.f. E0 is applied for a sufficiently long interval
of time to the end x = 0 of a semi-infinite non-distorting trans­
mission line (GL — CR), so that a stationary distribution of voltage
and current is set up in the conductor. At time t = 0 the end of
the conductor is earthed through a lumped resistance R0. Find the
voltage and current in the conductor for t > 0.
8. The ends x = 0, x — I of a string are rigidly fixed. The
initial deflection is w(x, 0) = A sin {njl)x (0 < x < /) and the
initial velocity is zero. Draw the profile of the string for times
tk = I k f l a (k = 1, 2, 4).
9. Determine the vibrations of an infinite string under the
action of a localised transverse force F{t) (/ > 0) if the point of
application of the force moves along the string at a constant velocity
v0 from the position x = 0 and v0 < a.
10. An e.m.f. E = f ( t) is applied at time / = 0 to the end x = 0
of a semi-infinite conductor with negligible resistance and leakage.
Find the potential distribution w(x, t) in the conductor.
11. A capacitance C0 charged to a potential difference V is
discharged at time t = 0 into an infinite conductor having an
inductance L and a capacitance C. Determine the current in the
conductor.
12. A compression S0 = (p—p0)/p0 is produced at time t = 0
in a gas at rest. The disturbance is localised within a volume bounded
by a given surface a. Find the compression S(M, t ) as function of
the area a, of the part of the surface of the sphere S %} lying inside a.
13. Which linear equations with constant coefficients of
the form an uxxJr2an uxt+ a 22ut(-t-bl ux-\-b1utJrC u = 0 have solu­
tions in the form of arbitrary travelling waves f ( x —at), where
a = const? (There is no dispersion.)
14. Which equations of the form given in Problem 13 have
solutions in the form of arbitrary damped travelling waves of the
form e~ '"/(x —a/)?
 4
Separation of Variables
(Fourier Method)

Typical problems which can be solved by the method of separation

of variables are boundary-value problems for hyperbolic and
parabolic equations in bounded regions. The principle of the
method is best illustrated by the simplest case, i.e. the case of
homogeneous boundary-value problems. We shall consider in
parallel boundary-value problems for hyperbolic and parabolic
equations.

4.1 PRINCIPLE OF THE METHOD. EIGENFUNCTIONS

AND EIGENVALUES

4.1.1 Suppose that it is required to find a function u(M, t) satis­

fying the equation
IP "a
div(& Vu) —qu = < ( 1)
Ip u t
for t > 0 in the domain D bounded by a closed piecewise-smoojF
surface S which is continuous in the closed domain B = {M e D ;
t ^ 0}, where D = D-{-S, subject to the additional conditions

( 2)

and
u{M, 0) = u, (M, 0) = <p,(A0
[respectively u(M , 0) = (3)
59
 Mathematical Physics

Substituting L[u] = diwik'Viij—qu, Equation (1) can be rewrit­

ten in the form
jpun
L[u] = O')
\ pu,
This equation and the boundary conditions given by (2) are linear
and homogeneous. Consequently, if u1 and u2 are the solutions
of (1), which satisfy the condition given by (2), then the functions
u — cl ul-r c2u2
where c: and c2 are constants will also be solutions of (1) subject
to the conditions (2).
We shall try to satisfy the initial conditions (3) by superimposing
all the linearly independent initial solutions of this type (i.e. solu­
tions satisfying the boundary conditions (2)). We shall thus seek
non-trivial solutions of Equation (1), i.e. those which are not
identically zero, which satisfy the boundary conditions (2) among
functions of the form 0 (M ) W{t), where 0 (M ) is continuous in D
and 1¥{t) is continuous for 0 ^ t < oo. Substituting the function
0{M) 1P{t) into (1) and dividing both sides of the equation by
p(M) 0{M) Wit), we obtain
L [0] W" I . , W \
= y esPective|y -gr]
In order that this equation should be an identity, i.e. the function
0 ( M ) W(t) should satisfy (1), it is necessary and sufficient that
both fractions L[0]/p0 and 1¥ " I'F should be equal to the same
constant, i.e.
L[0] _ , W"

The following identities must therefore be satisfied

'ST+l'P = 0 = 0) and L[0]+?.p0 - 0
Consequently, we can take the functions Wit) and 0 (M ) to be
the non-trivial solutions of the equations
W ' + X Y = 0 (respectively W'+XW = 0) (4)
L[0]+Xp0 = 0 (5)
where 0{M) must satisfy the boundary condition
d0 ,
y20 = 0 ( 6)
ri^ r s

60
 Separation o f Variables

The problem defined by (5)-(6) is called the Sturm-Liouville

problem. It does not possess non-trivial solutions for all values of A.

Definition Those values of A for which the problem (5)-(6) has

non-trivial solutions are called the eigenvalues of the boundary-
value problem (5)-(6), and the corresponding non-trivial solutions
0 (M ) of (5) are called the eigenfunctions of the boundary-value
problem (5)-(6).
We shall assume henceforth that k(M), q(M) and p(M) are
continuous in D, k(M) > 0, p (M ) ^ 0 (but p(M) ^ 0), q(l\f) > 0
in D\ yi(Af) and y2(M) > 0 o n S and y\-\-y\ =£ 0.
Subject to these conditions we have the following theorem.

Theorem 1 There exists an infinite set of eigenvalues {A„}

(n = 1 ,2 ,...) and corresponding eigenfunctions {&n{M)} of the
boundary-value problem (5)-(6).
We shall not reproduce the proof of this theorem here.

We shall say that the function f ( M ) has piecewise-continuous

first- and second-order partial derivatives in D if there exists a finite
number of domains Dx, D2, ..., D„ which do not intersect in pairs
and are such that (I) D — D1JrD 2Jr ■■■ + A , and (2) f{ M ) has
continuous and bounded first- and second-order partial derivatives
in each of the domains Dly Dz, ..., D„.
Let A denote the class of functions which (1) are continuous
in D, (2) have piecewise-continuous first- and second-order partial
derivatives in D and (3) satisfy the boundary conditions (6).
The eigenfunctions of the boundary-value problem (5)-(6) will
clearly belong to class A but they do not exhaust all the possible
members of this class of functions.

4.1.2 The eigenvalues and eigenfunctions of the boundary-value

problem (5)-(6) have a number of properties and we shall now
summarise some of them.

Expansion theorem Any function f ( M ) belonging to class A can

be expanded into a Fourier series in terms of the eigenfunctions
of the boundary-value problem (5)-(6), and the expansion will
converge absolutely and uniformly in the domain D.

We shall not reproduce the proof of this theorem here (the one­
dimensional case will be discussed in Chapter 9).61
61
 Mathematical Physics
The Fourier series for the function f ( M ) in terms of the functions
00
{0n(M)} is defined as the series ^ Cn0 n(M), in which the coefficients
n= 1
C„ are given by
C„ = jj/(M )p(M ) 0„(M) dr
3 P ^ndr J
D

Having solved the Sturm-Liouville problem, let us return to

Equation (4). For each eigenvalue 2n we have the general solution

lF„(l) = C„ cos ]/?.nt^-Dn sin )/l„t

(where ^ n(t) — Cn e^;‘^,)
Therefore, partial solutions of (1) which satisfy only the boundary
conditions (2) are functions of the form

un(M, t) = (C„ cos ]?.nl JrD n sin

(where un( M , t) = C„ e~*nt&n(M))
These functions can be written in the form

u„ = Bnsm(],0.nt+ dn) 0 n(M)

where Bn = ]/(% + Dl, 0„ = tan~ \ C nlDn).

Functions of this kind are said to describe natural vibrations
and standing waves; is the fundamental and m2(M, l),
u3( M , l) are the harmonics. The numbers \ 0 . z are the natural
frequencies of the vibrations. These frequencies do not depend on the
initial conditions. This means that the frequencies of natural vibra­
tions are independent of the method used to excite them. They
characterise the properties of the vibrating system itself and are
determined by the material constants of the system (for example,
the velocity of sound in the medium), geometrical factors (shape
and dimensions) and the conditions on the boundary.
The eigenfunction Bn0„{M) specifies the profile of the standing
wave.
If we now take the sum of these special solutions over all the
eigenfunctions

u(M, t) = ^ (C„cos| ?.nt+D„s\n\/Tnt)0„(M) (7)

n=162

62
 Separation o f Variables

we encounter the following problem: is it possible to choose the

coefficients C„ and Dn so that the sum will be the solution of the
problem (l)-(3)? The answer to this problem is given by the fol­
lowing theorem.

Theorem 2 The solution of (l)-(3) in a closed domain B = {M e D ;

t^ 0} can be represented by the series (7) with

Cn ~ \ p ( P ) < r ( P ) i ’,(P) i z r

Dn
II «!

M 2 = \ p ( P ) 0 2n(P) drP
D

The number |]0n|| is called the norm of the function

Consider the functional
R [w, 0] = — wL [0] dtp
D

The following lemma applies to this expression.

Lemma The functional R[w, 0] is symmetric for the class A, i.e.

R[w, 0] = R[0, w], for any functions w and 0 belonging to class A.
Proof Using the well-known result
p div(£) = di\{pE)—E V p , p = w, E — k V$>
we have (see definition of L in Section 4.1.1).

R\w, 0] = jj k(Vw-V0) d r + ^ q w 0 d r— jj div(A>w - V0)dr

D

( d0
In view of Green’s theorem, the last integral is equal to j /cw — dr,
and therefore 5
R[w, 0] = ^ k(7w- V 0 ) 6 t + jj qw0> d r — ^ k z v ^ —da ( 8)
d d s
If we are dealing with the first or second boundary-value problem,
f d0 ld0
we have \ k w ——d r = 0, so that w = 0 I —— = 01. In all
i dn s \ d% 63

63
 Mathematical Physics

these cases R[w, 0] must be symmetric. In the case of the third

boundary-value problem, we have from the boundary conditions
d& _ Z T 0 , and therefore R[w, 0] can be written in the
dn y i Is
form

R[w, 0] = ^ k (Vzu•V0 ) dr-f- ^ q w 0 dr-f- ^-^-&w)0dcr

D D S

This expression is again symmetric.

Remark For functions belonging to class A we must have

R[0, 0] > 0.

Proof o f Theorem 2 Suppose that u(M, t ) is the required solution.

Since this solution is continuous in D and satisfies the boundary
conditions (2) for any t > 0, it must belong to class A. Consequently,
in view of the expansion theorem it can be represented by the Fourier
series
00

W
(M,O= rt=
X^"(')0n
-1 (M) (9)
where

y n(t) = j f ^ \ p ( P ) u ( p , t ) # , ( p ) A T F (io)

Using Equation (5) for 0„(M) we can transform the last formula
to the form
i?[w,0„]
^ (0 = U® nf

and, using the above lemma,

R\&n,u]
0„(P)L[u)dT
h ¥ > n\?

Using (1) we obtain

P U,t d r

64
 Separation o f Variables
and hence, if we compare this result with ( 10) we have

W„(t) = - p - , i.e. K H ' V z e 0

'•n
Therefore the function xP„{t) is a solution of the equation X
P"
+X„W = 0 and, consequently, can be written in the form
¥n(t) = C„ cos \ X nt +T>„sin \/1„ t
where
C„=X
P n{0), Dn I X = 'I'n(0)
Using (10), we find that
C„ = Wn{Q) = \ p (P)u OP, 0)0„(P) dr

11 n11 D

a, = = 1 [p<,P)r i (P)<fn(.P) dr
l /, I 4 ll«'„lr ^
which was to be established.
By assuming the existence of the solution of (l)-(3), we have
come to the conclusion that it can be represented by the series (7)
and, therefore, it is a unique solution.
4.1.3 Let us now consider the following properties of eigenfunc­
tions and eigenvalues.
Property 1 If 0 is an eigenfunction corresponding to an eigenvalue
A, then C 0 (where C is a constant) is also an eigenfunction corre­
sponding to the same eigenvalue.
Property 2 If and <f>2 are eigenfunctions corresponding to the
eigenvalue A, any linear combination such as C1$ 1+ C2$ 2 is also
an eigenfunction corresponding to the same eigenvalue A. The
validity of these statements is obvious.
Property 3 Eigenfunctions (P, and <f>2 corresponding to different
eigenvalues A, and A2 (Ax ^ A2) are orthogonal in the domain D
with a weight p(M), i.e.

] p ( P ) 0 , ( P ) 0 2(P) dr = 0
D 65

5 65
 Mathematical Physics

Proof By definition of eigenfunctions and eigenvalues we have

Multiplying the first of these by 0 2 and the second by 0 y , subtracting

one from the other and integrating the resulting identity over the
domain D, we obtain

- R [ ® 2, , 0 2] = (X2- X y ) \ p 0 i0 2dx
D

Since the functional R is symmetric (the eigenfunction belongs

to class A), the left-hand side of this equation is zero and, consequent­
ly, ])’p 0 10 2dT = 0, since Xy ^ X2, which was to be established.
D
If to a given eigenvalue X there correspond r linearly independent
eigenfunctions 0 y , 0 2, . . . , 0 r , these functions will not necessarily
be orthogonal. It is possible, however, to replace them by other
eigenfunctions 0y, 0 2, ..., 0 r, which are linear combinations of
the original eigenfunctions and these can be made to be orthogonal.
In fact, let us suppose that 0y — 0 X. If §p 0 2dr = 0, then 0 2 = 0 2
D
and A \ p 0 l 0 2dx # 0, then 0 2 =■ 0y-\-B]0 2. The constant By can
D
be found from the condition \ p 0 y 0 2dx = 0, i.e. from the equation
D

jj p 0 \ dX-\-By ^p0y02 dr = 0
D D

If 0 3 are orthogonal to 0y and 0 2, then we can set 0 3 = 0 3 and, if

they are not, we can set 0 3 = 0y + 5 320 2+ B330 3. The constants B32
and B33 can be found from the conditions Jp 0 y 0 3 dr = 0 and
p - * o
) p 0 20 3dx — 0. i.e. from the equations
u

jjp01 dr + £32 \lp 0 y 0 2 dT + 2?33 \i p 0 y 0 3dx — 0

D D D

\ p 0 y 0 2 d x f B 32 p 0 l d T + B 33 ^ p 0 20 3dx = 0
D D D

and so on.
66
 Separation o f Variables
Continuing this orthogonalisation process, we can consider r
eigenfunctions @l} 0 2, ■ d>r corresponding to the given eigenvalue
A and these will now be orthogonal in pairs. Assuming further that
this orthogonalisation process has been performed, we conclude
that any two linearly independent eigenfunctions of the bound­
ary value problem (5)-(6) are orthogonal in the domain D with
a weight p.

Property 4 All the eigenvalues of the problem (5)-(6) are real.

Suppose that A = a+//S (jS # 0) is an eigenvalue and
0 = 0 J+ /0 2 is the corresponding eigenfunction. We must then have
L[<
PlJr i 0 2\ Jr{<xJi-iP)pi0lJriO^) = 0
and, consequently,
L [ 0 1]+a.p01—Pp02 = 0
i{L[&2]Jr<xp@2Jrpp@l } = 0
Subtracting the second identity from the first, we obtain

L [ 0 1—i 0 2]+(ot—iP)p(0l—i 0 2) = 0
Therefore, A = a —ifi and 0 = 0 x—i 0 2 are the eigenvalue and
eigenfunction of the problem, respectively. Using Property 3 we
have

5P O^i+ i —i®2) dr = 0 , where ^ p (0\-\- 0j) dr = 0

D D

which is impossible.

Property 5 All the eigenvalues of the problem (5)-(6) are non­

negative.
To prove this, we multiply the identity L[0 n]-\-?.np 0 n = 0 by 0 n
and integrate the result over the domain D. The result is

\ 0 nL[*n] d r + A ^ p ^ d r ^ O
D D

and hence

ll^ ll2
Since R[0„, 0 „] ^ 0, we have A„ ^ 0.
5* 67
 Mathematical Physics
Remark For the first and third boundary-value problems all the
eigenvalues are positive. For the second eigenvalue problem with
q(M) = 0, we have A = 0 as the eigenvalue and 0 = 1 as the corre­
sponding eigenfunction.

Example 1 Suppose that it is required to solve the problem

a2uxx = utt
u(x, 0) = <p(x), ut(x, 0) = 9?!(*), w(0, t) = u(l, t) = 0 (11)
«p(0) = <p(t) = 0
We recall that we solved this problem in Chapter 3 by the method
of characteristics. We then continued the initial values cp(x) and
(p\{x) to the segment (—1, 0), as odd functions and then periodically
over the whole straight line. We leave it to the reader to show directly
that the solution obtained by the method of characteristics is the
same as that obtained by the method of separation of variables.

Solution Among the functions of the form 0{x)0{t) there are

solutions of (11) which satisfy only the boundary conditions of the
problem. Substituting 0{x )0 (t) into the equation, we obtain
0" yR"
~0 s a20 =
Consequently, x¥ " Jr a2X 0 = 0 and
<p"+X0 = 0 , 0 (0) = 0(l) = 0 (12)
This is the first boundary-value problem. All its eigenvalues are
positive. The general solution of (12) can therefore be written in
the form
0(x) = A cos }/A a:-\-B sin ] A.v
From the boundary condition on the left we find that /t = 0.
Consequently, 0{x) = 5sin y' Xx and B ^ 0. From the boundary
conditions on the right we have A?sin]/A/ = 0. Consequently,
sin ] A/ = 0 and hence y Xl = nn and

J. = — ( » = 1 ,2 ,3 ,...)

These are the eigenvalues. The corresponding eigenfunctions are

0 n(x) = sin (nn/l)x. 68
68
 Separation o f Variaotes
For each 2„ we have

■>f * v ann , ^ . ann

i n{t) = c„ cos — A, Sin —j - t

According to Theorem 2 (Section 4.1.2), the required solution of

the problem is the function
CD

, . \ '/ _ ann _ . ann \ . nn

u(x, 1)= 2__i I Cncos—— /+x?nsin —-—/ Isin — x
n= 1
where
i i
C„ = 4 ^ ( ^ ) sin- r - ^ d t , Dn = — ^ 951(^)sin —J- k d£
/ J / ott/i J /
o o
i
II0„ ||2 = J sin2~ f df = —
o

Example 2 It is required to solve the problem

a2uxx = ut , u(x, 0) = <p(x), ux(0, t) = ux(l, t) = 0
As in the preceding example, we find that

<P"+/0 = 0, 0'(O) = <Z>'(0 = 0, lA + o 2^ = 0 (13)

We thus have the second boundary-value problem q = 0. Conse­
quently, 2 = 0 is an eigenvalue and &(x) = I is the corresponding
eigenfunction.
The remaining eigenvalues and eigenfunctions can be found
as in Example 1:

0(x) = /lc o s ^ /2 x + 5 s in )/2 x

From the condition 0'(O) — 0 we find that B = 0. Consequently,

A ^ 0 and 0(x) — A cos j/2x. From the condition = 0 we
find that sin \/?.l = 0 and consequently j/2 /= nn and 2(1 = n2n2ll2
(n = 1, 2, 3, ...). It follows that

„ n2 4.x2 n2n269

69
 Mathematical Physics

are the eigenvalues and

71 271 Tltl
1 , COS—X, COS-j-X, •■•, COS—-—X, ...

are the eigenfunctions. For each we can find the following eigen­
functions
Vn{t)= C„e-“2^ {n = 0, 1, 2, ...)
The required solution of the problem will, in view of Theorem 2
(Section 4.1.2) be the function
CO

u(x, t) = Cn e - ^ ^ c o s — x
n—0
where
/ t
C0 = I ^ , C„ = - J ^ ) c o S y f d f ( n = 1 ,2 ,3 ...)
6 o
i
ll^oli2 = /, I M 2 = jjcos2— f df = y (« = 1, 2, ...)
0

Example 3 It is required to determine the temperature distribution

along a homogeneous rod of length / whose lateral surface is insula­
ted and is subject to convective heat transfer (Newton’s law) at the
ends. The surrounding media are at constant temperatures and
u2, respectively. The initial temperature is arbitrary.
The mathematical formulation of the problem is as follows:
a2uxx — ut (14)
ux(0, t)—/Z][u(0, t) —ni\= 0 (15)
ux{l, t)+ h1[u{li t)—u2]= 0 (16)
u(x, 0) = <r(x) (17)
We shall seek the solution in the form u(x, t ) = v(x)-\-zv(x, t),
where ^(x) is the solution of (14) satisfying the boundary conditions
(15) and (16), i.e.
v" = 0 (140
u'(0)—/»i[^(0)—zq] = 0 (150
v'(l) 1 lh[v(l)~u2\ = 0 (160
70
 Separation o f Variables
For the functions w(x, t) the problem may be formulated as
follows:
a2wxx = wt (142)
w*(0, O-^i^CO, 0 = 0 (152)
wx(l, t)+h2w(l, t) = 0 (162)
zv(x, 0) = 99j(x) = <p(x)—v(x) (172)
The function v(x) describes the stationary state and w(x, t) a depar­
ture from the state.
Let us solve the problem for v(x) first. The general solution of
(14,) is of the form
?^(x) = C ,x+ C 2
The constants C, and C2 are given by (15,) and (16,):
Ci—hi(,C2—u1) = 0
CiJrh2(C1lJrC 2—u2) = 0
and hence
_ /?,(m2- » i ) r _ , C,
1 hx+ h 2+ h ,h 2V C2
We have thus found the steady-state solution. The problem for
w(x, t) will now be solved by the method of separation of variables.
Among the functions of the form $(x)!P(t) we shall require the
solutions of (142) which satisfy only the boundary conditions (152)
and (162). Substituting the function &(x)lF(t) into (142) and into
the boundary conditions (152) and (162), we obtain
<t>"+X& = 0 (18)
<Z>'(0)-/z,<Z>(0) = 0 (19)
0'(f)+h20(I) = 0 .(20)
W'+a2XW = 0 (21)
In view of Property 5, the problem defined by (18)-(20) will have
only positive eigenvalues. Therefore, the general solution of (18)
can be written in the form
&(x) — Acos\fXx-\-Bsm\/ Xx
From the boundary condition (19) we find that B )/ X = h{A and
consequently
_ _ _
*(x) = T r ( c o s ] / X x+/z,sin ]/X x)71 ( 22)

71
 Mathematical Physics
The factor B {\///,) will be included in yIJ(t). Substituting (22) into
(20), we obtain the equation for the eigenvalues:

cos /i
1 w 2\
7(/?i+/i2) t* I
where ju = ]/Xl. ptx, jx2, ... are the positive roots of this
equation. The eigenvalues will then be the numbers

Xn Pi

The eigenfunctions will be of the form

0„(x) = y c o s y ^ + ^ s i n - J-x

They are orthogonal within the range [0, /] with a weight p = 1.

Let us return now to Equation (21). Its general solution for
X = X„ is of the form
y n{t) = c nz - ° ^
We then have
00

W{X, 0 = 2 C nZ -^'^nix)
n= 1
The coefficients C„ can be found from the initial condition using the
orthogonality of the eigenfunctions 0 „(x):
/

Remark The method discussed in this example can be used for

a broad class of boundary value problems with steady-state, i.e.
time-independent, inhomogeneous parts, either in the equation
itself or in the boundary conditions (or both).

Example 4 Determine the transverse vibrations of a string, one

end of which is rigidly fixed and the other free, if a localised mass
m0 is attached to the free end and the initial excitation is arbitrary.
72
 Separation o f Variables
The mathematical formulation of the problem is as follows:

a2uxx= u„, 2~_ PoT

a

u{\, 0
)=cp(x), ut(x, 0
) - <Pl (x)
w(0,r) = 0, Tux(l,t) -- m0utt(l, t)
Among the class of functions 0(x)W(t) we shall seek solutions which
satisfy only the boundary conditions. Separating the variables we
have
W "+ a22V = 0 (23)
&"+?.& = 0 (24)
0(0) = 0 (25)
The boundary condition on the right can be written in the form
T 0 ,(t)'P(t)-mo0(I)'P"(i) = 0
Replacing W"(t) in (23) by 1F(t) and dividing both sides of the
equation by ^ ( 0 , we obtain
0'(t)+hX0(I) = 0 (26)
where h = o2m0IT.
The solutions of (24) which satisfy (25) are of the form 0(x)
= sin \ flx . From (26) we find the equation for the eigenvalues
K > 0:
-1
tan fi = (A — } / 2 I
IjT ’
The corresponding eigenfunctions are

= siny *

It is readily verified directly that these functions are not orthogonal

with the weight p(x) — 1. This does not conflict with the general
theorem on the orthogonality of eigenfunctions since the boundary
condition (3) is not the usual boundary condition of type III because
it contains explicitly (rather than through the eigenfunction) the
eigenvalue To elucidate this point we note that the equation for
u(x, t) can be written in the form
Tuxx = [p0+ m 0d(x—l)\u„
73
 Mathematical Physics
Consequently, the equation for the eigenfunctions can be written
in the form
T0"+/.p(x)& = 0
where
p(x) = p0+ m 0d(x—l)
The eigenfunctions 0 n(x) will therefore be orthogonal with the
weight p{x). This is readily verified by direct calculation.
The next step is the same as before. First, we find xP n{t) so that
o°
u(x, t) = (c„cos——t+T>„sin— sin—-.x
T1=1
From the initial conditions we can determine the coefficients Cn
and Dn using the orthogonality of the eigenfunctions in terms of
the weight p = p0-{-m0d(x—1):
i
Cn = -p - ||2 jjj Po^(s) &„(£) d |+ m 0cp (/) <Pn (/)J

i
Dn = ---- \L~ii2 \\P0<P1 (0 0 n(f) d^+ Wo-PlCO (O)
aPnWn\\ L> >

l |0 J 2 = P o $ ^ (£ ) d l+ m 0^ 2(/)
o
Let us now return to the properties of the eigenvalues and eigen­
functions. We note, to begin with, that since for the functions 0
belonging to class A we have R[0,<P] ^ 0,
R[0 , 0]
inf = /i ^ o
<PeA im i2

Property 6 (extremal property) This is expressed by the following

theorem.

Theorem 3 If u = inf — — is valid for some function 0

0eA ¥>\\2
belonging to class A, then 0 is an eigenfunction and ,u is the corre­
sponding eigenvalue for the problem (5)-(6). At the same time,
p, will, of course, be the smallest eigenvalue.
74
 Separation o f Variables
i?[0, 0]
Proof For any function 0 belonging to class A we have
~ m T
R[0 0 1
-H > 0. In particular, ?.„= — > p. Consequently, for
II®,,
0 e A, we have
1F[0] = R[0,0]-(X ||0||2 ^ 0
whereas
W[0] = R[0, 0]~ju \\0\\2 = 0
Therefore, the functional 0[!F] reaches a minimum on 0. This is
equivalent to <p(oi) = !F [0 + a/], where f e A reaches a minimum
for a = 0. However, we then have <p'(0) = 0. Let us evaluate this
derivative:

9>'(0) = [® + a/, 0 + a /] - ^ ||0 + a /||2 } a=o

= {&+«./) L [ 0 + a f ] dz-ju jjp(0 + a/ ) 2dr}a=o

n d

= ~ \ { f L [ 0 ] + 0 L [ f ] } d r - 2 p \ P 0 f dr
D D

= - 2 \ f { L [ 0 ] + p p 0 ) dr
D

where we have used the symmetry property of the functional R[f, 0]

for functions belonging to class A.
Therefore, for an arbitrary function / belonging to A we have

\ f { L [ 0 ] + p p 0 } dr = 0
D

Hence, it follows (from the fundamental lemma of the calculus of

variations) that at points at which the function L\0\ is continuous
we must have
L[0]-\-pp0 = 0
which was to be established.

If the minimum of the functional is sought in a class of functions

A k belonging to A and orthogonal with respect to the weight p in the
domain D to the eigenfunctions 0 U 0 2, 0 k- lt this minimum
75
 Mathematical Physics

will be the A-th eigenvalue 7k and the function on which it is reached

will be the corresponding eigenfunction 0 k. The proof of this is
similar to that given above.

Property 7 The eigenvalues do not increase as k(M)[q(M)]

increases. More precisely, if k {(M) > k2(M ) in D, then xj^ ^ x<2).
We shall give the proof for Ax.
For any function 0 e A we have
^ [ 0 , 0 ] ^ R2[0,0]
||0||2 ^ ||0||2
where R { and R 2 are the functionals R corresponding to the functions
ki(M) and k 2(M). Consequently,

x ^ = inf R d * , <P] inf = ;(2)

0 eA || 0 ||2 0 EA || 0||2

For the case qdM) > qi(M) the proof is practically the same as
before.

Property 8 The eigenvalues do not increase with increasing p{M).

More precisely, if P \ ( M ) > p 2 ( M ) in D, then < x^2).
We shall give the proof for x,. For any function 0 belonging
to class A we have
J?[0, 0] ^ R[0 , 0]
\\®\\2Pl ^ m i2
where J|0 ||Pl and ||0||P2 are the norms of the function 0 with weights
p t and p2. We then have
R[0, 0] R [0 , 0]
;.f‘> = inf < inf = x<2>
0 EA 0C-A i\ m 2
which was to be proved.
From Properties 7 and 8 it follows that in the one-dimensional
case the eigenvalues x„ increase as n2 with increasing n. In point
of fact, in addition to the equation

— [k (-v) 0 ' (-v)] - q (x) 0 (,v)+ xp (.v) 0 (,v) = 0 (27)

dx:
let us consider the equations
k20"-V{APx~ q 2)0 = 0 (28)
and
M y '+ (x p 2-<7i)0 = 0 (29)
76
 Separation o f Variables
where k z, q2, p 2 are the maximum values of the functions k(x), q(x),
p(x) in the range [0,/]; k x,q^ and p x are their minimum values
(either sup or inf).
To be specific, consider the first boundary-value problem, i.e.
let us determine the solutions of equations (27), (28) and (29)
subject to the boundary conditions
<P(0) = <P(/) = 0 (30)
Since (28) and (29) have constant coefficients, the eigenvalues of
the problems (28)-(30) and (29)-(30) can readily be found. They
are given by
7i2n2 n2n2
2nn k -i A y
72 ^2
0
pil Pi p2l2 ki Pi
From Properties 7 and 8, the eigenvalues 2„ for the problem (27)-(30)
lie between 2' and 2'', i.e.
'•n <'*n '• ' n
This confirms the validity of the above statement. The fact that
there are an infinite number of eigenvalues and eigenfunctions
in the one-dimensional case (Theorem 1, Section 4.1.1) is also
a consequence of these inequalities. The reader is recommended
to give the proof for the second and third boundary-value problems.

Property 9 The eigenvalues do not decrease with decreasing

domain D, i.e. if D' cz D", then 2' ^ 2„'.
We shall give the proof of this property only for 2j for the
first boundary-value problem. To each domain D' and D" there
correspond their own classes of functions A', A".
Suppose that some function 0 ' belongs to class A'. It is zero
(in view of the boundary conditions on the boundary of D') on
that part L' of the boundary of D' which is contained in D" (Fig. 4.1).

The function 0 " which is equal to 0 ' in D' and zero in D " —D'
(shaded region) will clearly belong to class A " . If we perform this
operation for each of the functions in A', we obtain the new class
77
 Mathematical Physics

of functions A' which is a part of A". For any function 0 £ A'

we have
R"[0, 0] = - $ 0L[0] dr = - 5 0L[0] d r = R'[0, 0]
D" D'

and
^ p 0 2dr = ^ p 0 2 dr
D" D'

since this function is identically equal to zero in D " —D'. Therefore,

R'[0, 0] R"[0, 0] R " [ 0 , 0] -/
X[ = inf =inf >- inf '-1
0 cA' \m \ 0eA' wni \\n i~ 0 eA'
where H^U, and \\0\\2 are the norms of 0 in D' and D", respectively.

Definition An eigenvalue X will be called r-fold degenerate if there

exist r linearly independent eigenfunctions with this eigenvalue.

Definition An eigenvalue X will be called a simple (or non-deg­

enerate) eigenvalue if any two eigenfunctions with this particular /.
are linearly dependent.

Property 10 All eigenvalues of the one-dimensional boundary-

value problem (5)-(6) are simple eigenvalues.

Proof Suppose that 0 f x ) and 0 2{x) are eigenfunctions corre­

sponding to the same real eigenvalue X. Both these functions are
then the solutions of the same equation

— [k0']~ q 0 + X p 0 = 0

and satisfy the same boundary conditions on the left-hand side

72^1 (0 ) = 0

y 10'z (0) y20 2(0) = o

These can be regarded as a system of linear equations for y l and y2.
Since y?+y? # 0, the determinant of the system is zero. However,
this is the Wronskian W(x) for the solutions 0 f x ) and 0 2(x) at
x = 0. It is well known that the Wronskian consisting of the solu­
tions of a given linear homogeneous equation is either identically
zero or does not vanish anywhere. Since, in our case, W(0) = 0,
it follows that W(x) = 0. This means that 0i(x) and 0 2{x) must
78
 Separation o f Variables
be linearly dependent. We note that in multi-dimensional boundary-
value problems this is no longer true.

Example 5 Consider the vibrations of a square membrane with

fixed edges under the action of an initial excitation. The edges of
the square lie along the coordinate axes.
Mathematically this problem may be formulated as follows:
V2h = ttff, u — u{x, y, t) (31)
u(0, y, t) = u(l,y, t ) = 0 (32)

o
u(x, 0, 0 = 3 (33)
>1

II
= <p(x, y). ut( x , y , 0) = tpi{x,y) (34)
Among functions of the form <P(x, y ) IF(r) we shall seek solu­
tions of (31) which satisfy only the boundary conditions (32), (33).
Substituting this function into (31), (32) and (33) and separating
the variables, we obtain the following Sturm-Liouville problem
V2<£+A0 = 0 (35)
0 (O ,y)= 0 (l,y) = O (36)
0 (x,O ) = 0 ( x , l ) = O (37)
This problem can also be solved by the method of separation of
variables. We shall seek solutions of the form &(x, y ) = A(x) B(y).
Substituting this function into (35) and separating the variables
we obtain

A= 0

In order that this should be an identity it is necessary that

A " / A -- —p, and B " /2?+A = p, i.e.

A " + p A = 0 (38)
B"+ (X—p)B = 0 where B"+xB = 0 (39)
From (36) and (37) we find that
A (0) = A (I) — 0 (40)
B ( 0) = B (I) = 0 (41)
79
 Mathematical Physics

Therefore, we have the boundary-value problems (38)-(40) and

(39)-(41). The eigenvalues n and X—/i should be positive (Pro­
perty 5). As in Example 1, we find that
2 2
71 11-
H„ = —p { = 11, 2,
(n T ...)\

A„(x ) = sin -y -x

and
n2k 2
ak p ( * = 1 ,2 ,...)

_ . . . nk
Bk(y) = sin — y

However, ak = A—/x„ and, consequently,

n2
K m = a*-hMn or K.k = -^-(n2+/c2)
where k and n assume the values 1,2, ... independently of each
other. We have thus found the eigenvalues of the problem (35)-(37).
The corresponding eigenfunctions are
, . .n n . nk
®n,k(x,y) = sin— x sin— y

The eigenvalues K m and K,n will clearly be equal and the cor­
responding eigenfunctions
. .n n nk , , . nk . nn
<P„fk = sin — x sin—y y and n ~ sin —-j—x sin —-—

are linearly independent. For example, A1>2 = A2>1 = 5(n2jl2)

@12 = sin-r-x sin—p- v and &2 i = sin—j- x sin — y

/ I J 2,1 / I*
Therefore, the eigenvalues of this particular problem are not simple
eigenvalues.
The solution of (31)-(34) can be represented by the series
00 00

u ( x , y , l) = ^ ^ (Ctli k cos a }/Ani k t

n=lk=l
^ . /-— . . nn . nk
-f D„ik sina |/Anik/)s in —- x s i n — y80

80
 Separation o f Variables

in which the coefficients are given by

i i

oo
i i

4.2 SOME PROPERTIES OF A SET OF EIGENFUNCTIONS

We shall now consider some of the properties of a set of eigen­

functions {0„}-

Definition A set of functions {<?„} which are orthogonal in pairs

in a given domain (with a weight p) is called complete in D if for
any function f(M), which is square integrable in D,
00

(42)
I
where Ck are the Fourier coefficients of the expansion of f{M )
in terms of the functions {0fc}.
A sufficient test for the completeness of the system {#„} If a function
F(M) is continuous in D, and for any e > 0 there exists a linear
combination Sn = ...+a„tf>n for which $ p(F—S„)2 dr < e,
then the set {&„} is complete. D
Consider a fixed e > 0. For any function f ( M ) which is square
integrable in D, we can find a function cp(M) which is continuous
in D and is such that \p(J—?>)2dr < e/4. This requires proof but
D
we shall not give it here. For the function <p(M) and for the
chosen e, it is possible by hypothesis to find a linear combination
S n = ot.l0 1Jr ...Jr oi.n<hn, for which

$ P (<P-Sn)2 d t <
D ^

Consider the integral

n

\p (f- ^ c k®k)2 dr
D

6 81
 Mathematical Physics

in which Ck0 k is the partial sum of the Fourier series for /(A/).
k~l
It is clear that
n 2 n n
$ „ (/-£ a * , d t = \ p f 2i r - 2 Y J Ck J p / ^ d r H - ^ Q I I ^ I i 2
D X k=\ ' D k=\ D k —l

where we use the orthogonality property of the function <Pk. Since

\ p f 0 k dr = Ck\\0kj|2, it follows that
D

n 2 n

\p lf~ Y j dT = $ p / 2d r - ^ a | i 0 t ||2
D ' k^l ' i) k =\

It is well known that the quantity <32 = \ p ( f —S n)2 dr is a minimum

n D
if S n is taken to be ^ Ck0 k. Therefore
k= 1

n n

0 < ^ p f z d r - Y Ck\\0k\\Z = S p (/_ X j C,t0,t) dT

D k=I D k= l'

< j p ( / - 5 „ ) 2d r < 5 p{f— <p+ (p-S ,)2dr

D D

< 2 J p(f-<p)2dr+2 jj p(<p—s„)z dr < 2 * + 2 - = £

D I)

In this expression we have used the well-known inequality

(A + B)2 < 2A2-\-2Bz
It follows that

O < $ p / 2d T - ^ C * 2||0 J 2 < £

D k=l

and hence we have the completeness condition

oo

2 dr = ^ a | | 0 t ||2
D k=I

Expansions of the Fourier type in terms of the complete systems

of functions {0,,} have the following remarkable property.82
82
 Separation o f Variables
Theorem If a set of functions is complete in a domain D,
then the series of the Fourier type for any function which is square
integrable in D can be integrated term by term independently of
whether it is convergent or not, i.e. for any domain D' c D
CO

\f{ M ) dr = ] T c n 5 $„(M) dr
D' n=\ D'

n
Proof Consider the difference d„ = jj / d r — ^ Ck jj 0 k dr
D' k^l D’
n

^ = !!!/-E c^ )dT
D' k= 1

n n

< b'Sf - E
k^i
dr < [ \ f - Y JCk0 k I dr
D k =\

To estimate the last integral we shall use the Cauchy inequality

dr
\ f ~ ^ C k0 k dr Ck&k
> 1 VP
D k= 1

<

The last integral is bounded and the difference

n

$p/2d r - ^ Cf\d>k\\2
D k= I

tends to zero as n -+ oo in view of the completeness condition.

Consequently, dn -*■ 0 as n -> co, which was to be proved.

Theorem The set of eigenfunctions of the boundary value problem

(5)-(6) is complete.83
6* 83
 Mathematical Physics

Proof Consider the function f ( M ) which is arbitrary and con­

tinuous in D. For any number e > 0, the functions belonging
to class A (Section 4.1) will include a function g(M ) such that

j j p ( / - s ) 2dr < — (43)

D

The series of the Fourier typeg(M ) = Ck&k converges uniformly

_ *=i
in D. Consequently, for any £j > 0 we can find N(e2) such that
n

g— ^ c k0 k \ < for n > Nfa) (44)

1
n
We shall show \ p [ f — X! Ck0 kf dr < £. By the above test for
D /c= 1
completeness {$„} must be complete. It is evident that
n n

S p (/- dr = \ p [ f s + S — ^ C fc0 k)2dT

D k= 1 D k=l
n

< 2 \ p { f - g f d r + 2 \ p [ g - ^ C fc0fc)2dr
D D k= 1

Using (43) and (44) we obtain

^ p ( / —^ C k^ k) dr < dr < £
D k= 1 D

where £, ^ ] / e / 2 } / S and B = jjpdr.

D
The last two theorems enable us to integrate the series expansion
in terms of the eigenfunctions of the boundary-value problem
(5)-(6) term by term for any function which is square integrable
without bothering either about the uniform convergence of these
series, or even whether they converge at all.

4.3 SOLUTION OF INHOMOGENEOUS BOUNDARY-VALUE

PROBLEMS BY THE FOURIER METHOD

Knowledge of the set of eigenfunctions {&„} and the corresponding

eigenvalues {2,,} enables us to solve inhomogeneous boundary-
value problems. Let us consider some of these.84
84
 Separation o f Variables
4.3.1 It is required to find the solution of the problem
L[u]+f(M, t) = pu„ (respectively ut) (45)
u(M, 0) = 0, w,(M, 0) = 0 (46)
du
r ^ + f2 ui =o (47)

which is continuous in the closed domain B = {M e D, t > 0}.

The required solution u(M, t) belongs to class A and, therefore,
in accordance with the expansion theorem it can be represented by
series of the Fourier type in terms of the eigenfunctions {0,,} cor­
responding to the inhomogeneous problem (5)-(6):
00

= (48)
n= l

where

V,(t) = ■— l)®n{P)AT (49)

Using Equation (5) for p@n under the integral sign in (49), we
obtain

*P > „,«] l f
L[ m] dr

Substituting for L[u] from (45), we obtain

^ » (0 = - ; I.L 1,2 \ p u a 0 ndT+ 1 Tp (50)

The first term on the right-hand side of (50) is equal to —1

The second term is a known function and will be represented by
/„(/)!/;.„. Therefore,
'K , fn
^ (0
K ^ K
Consequently, Wn(t) is the solution of the equation
'Fn + K 7yn = fn(t) (respectively Wn+ h xF n = /„)
85
 Mathematical Physics

with the additional conditions

V'„(0) = \pu(P, O)0„(P)dr 0

W ) = ~ - | |2- \put{P, 0 ) ^ n( / ,)d r = 0

The solution of this problem is of the form

1
^ (0 = ^ sin 10:n ( t - 0 ) f n(0) d0
o

(respectively ^ ( r ) = ^ e 0)/ n(^) dfl)

o
Substituting the functions ^ ( t ) obtained above into (48), we
obtain the required solution in the form of the series expansion
in the eigenfunctions. If, in particular, /(M , /) = f ( t ) d(M, M 0),
then

/,(<) = W , F $/ ( 0 6 (P' Mo) 0 "(p) dr = m

and

v.it) = - - - - ( sin |/;.;(r —tom m

\ / K W *. If J
t

/respectively lF n(t) = jj e 0)f ( fj)

Consider the problem

t) = p u tt (pit,)
u (M, 0) = <p(M), ut(M, 0) = (fi(M)

r ' ‘In + H = °
We shall seek the solution in the form of the sum

u(M, t) = v (M , t ) Jr w { M , t)

86
 Separation o f Variables

where v ( \ l , t) and w(M, t) are the solutions of the following

problems:
v: L [v\= pv,, (pvt)

v{M, 0) = v,(A/) = ^>l -d— Jr y2v ^ = 0

zv\ L[w ]+ f(M ,t)= pw n (pw,)

I dw \
w(M, 0) = zv,(M, 0) = 0 , h '1 - + y 2w\ = 0

We already know how to solve these problems.

4.3.2 Suppose it is required to find the solution of the problem

L [ u ]+ f( M ,t) = pu„ (pu,) (51)

(y, d£ - + y 2i^s = H M , t ) (52)

u(M, 0) = <p(M), u,(M, 0) = tp^M) (53)

which is continuous in the closed domain B = {M e D; t > 0}.
We shall consider the following method of solving this problem.
Among the functions v(M, t ) which are continuous in B and have
in this domain piecewise continuous first- and second-order partial
derivatives, we shall take the function v x(M, t) which satisfies the
prescribed boundary conditions given by (52). We shall seek the
solution u(M, t) in the form of the sum u — v y(M, t),
where w{M, t) is continuous in B. The problem can then be formu­
lated as follows:
£ M + /i( M , t) = pwtt (pzvt)
w ( M , 0) = =

(^ H r °
where
f i ( M , t) = f ( M , 0+£[*>i]-/Wim
<P(M) = <p(M)—v1(M, 0), , 0)
We have already considered this problem in Section 4.3.1. The
form of the function v x(M, t) is either guessed or it is found by
Duhamel’s method (see Section 4.3.3).
87
 Mathematical Physics
Example 6 It is required to solve the problem
a2uxx = utt
u(x, 0) = rp^x) , ut(x, 0) = cp2(x)
w(0, 0 = /«i(0, i<(/, 0 = M O
We shall assume that the functions p.x(t) and ju2(t) are twice dif­
ferentiable and will take

V {( X , t) = — y —- j« i ( 0 + ' y / u 2 ( 0

The solution will be sought in the form of the sum

u(x, t) = ■y1(x, t)Jrw{x, t)
The function w{x, t) will clearly be the solution of the following
problem:

a2wxx+ ^ j ^ - f i i ( t ) - j / 2 (t) = w,t

w(x, 0) = ( 0 ) - y ,u 2(0) = rPl (x)

wt(x, 0) = (p2(x) + — —f i [ ( 0 ) - ~ p , 2(0) = y 2(x)

zv(0, t ) = w(l, t) = 0
The function w(x, t) will also be sought in the form of the sum
w = R(x, t)+Q(x, t), where R(x, t) is the solution of the homo­
geneous boundary-value problem
a2Rxx = Ra
R ( x y 0) = cpi (.v), = q>2{x), R ( 0, 0 = K(0 0 = 0
and is given by (see Example 1)
co

R(x, 0 = „ cos a }/?.„ t + Dnsina \f h t ) sin —j - x

n= 1
I
n2n2 2 „ /u, . nn .
K = —p r > c » = J ] <Pi(0 s,n — I d£
o
i
Dn = [ y 2(g) sin — § d |
ann .) I
o
88
 Separation o f Variables
and Q(x, t) is the solution of the following problem

<?Qxx+f(X, t) = Qu
Q(x, 0) = <2((*, 0) = 0
Q(P,t) = Q ( l , t ) =0

w h e re/O , 0 = - y ^ z 'O ) -
According to Section 4.3.1,
00

2 0 . 0 = ^ ¥„(/) sin— x
n=i
The functions are given by
r

*„( 0 sin]/A„(t-d)f,(6)d8

where

/,(<>) = t o d f = — K - o v r o o - ^ 'w ]
o

Remark 1 If the boundary conditions are of the form

ctiUx (Q, t) — ^ u ( 0 , t) = ^ i(0 . ct2ux (l, t)-\-P2u(l, t) = fi2{t)

the function v x(x, t), which satisfies these boundary conditions, can
be taken to be
v(pc, 0 = D x?h2(t) C(x /)2^ i (t)
where

2 ^ 1 + P,l2 ’ ~ 2a2/+/32/2

Remark 2 It is sometimes easy to find the function ^iO , 0 which

satisfies not only the prescribed inhomogeneous boundary condi­
tions but also a given equation.89
89
 Mathematical Physics
Example 7 It is required to solve the problem
&Uxx — ttn (54)
«(x, 0) = <Pi(x), «t(x, 0) = (p2 (x) (55)
t \
cT

o' «(/, t) = A sin cot (56)

a

II
A

Among functions of the form F(x) sin cat it is quite easy to

find the solution v x(x, t) of (54), which satisfies the boundary con­
ditions (56). In fact, substituting this function into (54) and dividing
both sides of the equation by sin ait, we obtain the equation for F{x) :
a2F"+a>2F = 0 (57)
From the boundary conditions (56) we find that
F( 0) = 0, F(l) = A (58)
The solution of the problem defined by (57)-(58) will be of the form
. ai
sin— x
a
F(x) = A
■ M /i
sm—
a
and, consequently,
. 01
sm— x
v 2(x, t) = A ------------sin ait
a
The solution of the problem (54)-(56) will be sought in the form
u(x, t) = ^i(x, t)+z«(x, t)
where w(x, t) is the solution of the following problem:
a2wxx = wtt
« ;(0 ,/) = 0, w {l,t) = 0
. ai
sin— x
w(x, 0) = p,(x), w,(x, 0) -- <p2(x)—Aai ------------ - <p2{x)
sin— I
a90

90
 Separation o f Variables

This inhomogeneous boundary-value problem can be solved by

the method of separation of variables.

4.3.3 Duhamel's method This can be used for boundary-value

problems with inhomogeneous boundary conditions which can
be reduced to the form

L [ u ] = p ( M ) \ U,t (59)
\u,

u ( M , 0) = u,(M, 0) = 0 [respectively u(M, 0) = 0] (60)

(a4 nJ r ^ U) s = ^ (6l)
This involves the following steps.
1. First we solve the problem defined by (59)-(61) with a sta­
tionary inhomogeneity on the boundary condition, i.e. with a
boundary condition of the form

| a— , t)

[the stationary inhomogeneity /u(M, r) on the boundary is switched

on at time t = 0], where t is a fixed number. Suppose that w(M, t, r)
is the solution of this problem, which is continuous together with
the first-order derivatives and the derivative wtt in the domain
( M e D , t ^ 0 ) . The solution of this problem must be treated
as a generalised function because rj(t)/n(M,r) is a generalised
function.
The solution of the problem (59)-(61), subject to the boundary
and the initial conditions of the form

«-— +/?«) = n(t— t), a't =z = ut', =T= 0

on /s

[the stationary inhomogeneity n(M, r) is switched on at t = t]

will be the function w(M, t —r, r) rj(t—r). We note that at internal
points M of D we have

w(M, 0, t) = wt( M , 0 , t ) = 0 (62)

91
 Mathematical Physics

2. The solution of the problem defined by (59)-(61), subject

to the boundary and initial conditions

w!t=t = ut\t=z= o

(a — + f i u = fi(M, x)[rj(t-x) — r](t—x-dx)]

[stationary inhomogeneity n(M, x) on the boundary exists only

for the time interval between l — x and t = r+ d r] will be the
function

w(M, t —x , x)r](t—x)—w(M, t —x, x)r](t—x—dx)

- t —x, x)i](t—x)\ dr
ot

3. In the original boundary-value problem (59)-(61) the inhomo­

geneity on the boundary acts only for the time interval between
0 and t. One would therefore expect that the solution of (59)-(61)
will be
t

u(M, t) = ^ ~ [w(M, t—x, x)rj(t—x)\ dx (62a)

o

Direct verification will show that this is so. In point of fact,

this function can also be written in the form
t t

u(M, t ) = \ r j ( t - x ) w t(M, t—x, r ) d r + \ w ( M , t - x , r)<3(/-r) dr

o o
since (d/dt)r](t—r) = 6(t—r). Since r](t—x) — 1 for all r between
0 and t, we have, using the properties of the (5-function,
t

u(M, 0 = 5 wt( M , t —x, x) dxJr w(M, 0, t) (63)

o

From this expression and from the formulae for the derivative
t

ut(M, t ) = \ w tt(M, t—x, x) d r+w,(M, 0, t) (64)

o92
92
 Separation of Variables
it follows directly that the initial conditions (60) are satisfied
[(w(M, 0, t) = wt(M, 0, t) = 0 for internal points of D]. The
boundary condition (61) is also satisfied because

= ^— r)v2(t—r)} dr
0
t

- ^ — {p(M, r)r](t—T)} dr
0
t

= ^ (M , t) — 7](t-r)dr
0
t

= ^ p (M , r)d(t—r) dr = p ( M , f)
o
Let us substitute (62a) into (59) and use Equations (63) and (64).
For internal points in D we have, in view of (62),

u(M, t) = ^ wt( M , t —t , r)d r

o
*
ut( M , 0 = 5 t —T, r) dr
o
Consequently,
r

utt = 5 t —T, t)dr+TO„(M, 0, r)

o
From the identity
L [ to( A / , r — t , t )] = p(M )wtt( M , t — r , t)

and using the continuity of wrr in the region (M e Z), t > 0), we
find that (for t —r -> 0)
L[w(M , 0, t)] = p(M)wtt{M , 0 ,t) = 0
since w(M, 0, t) = 0 for internal points of D and, consequently,
L[zo(M, 0, 0] = 0.93
93
 Mathematical Physics

Therefore,

un(M, t) = \ t t , t) dr
6
so that
t
L[u) -putt = jj d -{L[w\~pwt,} dr = 0
o
since L[w\ = pwtt. This means that (62a) is the solution of (59)-(61).
Consider the special case /u(M, t) = Q(t). Proceeding by analogy
with the above discussion, the solution can be obtained as follows.
1. We first solve Equation (59) subject to the boundary condition

i.e. for Q(t) = 1.

Let R(M, t) be the solution of this problem. The function
Q{t) R(M, t) will then be the solution of the problem with the
boundary condition of the form

(“In + ^ U) s =
where t is a fixed number.
2. The function Q(x) R(M, t —r) >?(/ —t) will be the solution
of Equation (59) subject to the boundary and initial conditions of
the form

(« +P“j s = 2 ( t )
H't = x = Ui\, = x = 0

We note that in view of the initial conditions, for all internal points
of the domain D we have
R(M, 0) ee Rt(M, 0) = 0

3. The function
Q(t)[R(M, t - T ) r / ( t - z ) - R ( M , t - T - d : ) j j ( l - r - d r ) ]

= < 2 (0 ^ [&(M, f —t)->?(/—t)] dr

94
 Separation o f Variables

will be the solution of (59) subject to the boundary and initial

conditions of the form

»r =T= Wf!t =t = o
4. The solution of the original boundary-value problem will
be the function

U(M, 0 = ^ Q ( r ) ~ R ( M , t - t) dr
o
This can be verified by direct substitution as in the previous case.
It is thus sufficient, in this case, to find the solution R(M, t) for
the problem with the very simple (stationary) inhomogeneity on
the boundary condition Q(t) = 1.

Example 8 Find the solution of the problem

a2uxx = l<t, u(x, 0) = 0, u{0, t) = 0, u(l, t) = Q{t)
To begin with, we find the solution R(x, t) of the problem for
Q(t) = 1. The function R(x, t) will be sought in the form of the
sum R = v(x) : P(x, t), in which ?;(x) describes the steady-state
conditions and P(x, t) the departure from the steady state. For
the function v(x) the problem can be formulated as follows:
v ” = 0, ^(0) = 0, v(J) = 1
The solution will be the function xll. For P(x, t) the problem can
be formulated as follows:
a2Pxx = Pt , 0) = - P{0, t) = P(l, 0 = 0

Solving this problem by the method of the separation of variables

(see Example 1), we find that

The coefficients C„ can be determined from the initial condition:

GO

95
 Mathematical Physics

and are given by 2 (-1 )'

Therefore, CO

Consequently, the solution of the original problem will be

U(x, /) = J 2 ( t ) ^ [/?(*, t t) rj(t r)] dr

or

u(x, /) = t\ q (t) - ^ R ( x ’ / —T)dT+ Jq (t)/? (x , t —r)d(t—r) dr

o 0

§ Q (r)-^ R (x , t - r ) d r + Q ( t ) R ( x , 0 )
o
The function R(x, 0) is zero at all internal points within the range
[0, /] and for x = 0, while for x = / we have R(x, 0) = 1.
If it is required to solve the problem
a2uxx = ut , u(x, 0) = 0, u(0, /) = £?i0), u(l,t) = Qz{t)
the solution can be sought in the form of the sum u = v-\-w, where
for?; and w we have the following two problems:
v: a2vxx = vt, v(x, 0) = 0, v(0, t) = v(l, 0 = 0
zu: a2wxx = w,, w (x, 0) = 0, w(0, t) = 0, zo(l, t) — Q2(0
Each of these problems can be solved by Duhamel’s method,
as shown in the above example. This method can also be used
for the solution of boundary-value problems on a semi-infinite
straight line.

4.4 UNIQUENESS OF THE SOLUTIONS OF

BOUNDARY-VALUE PROBLEMS

4.4.1 The uniqueness problem can be resolved as follows. If there

are two continuous solutions of the problem (l)-(3) in a closed
domain B = [M e D, t ^ 0}, their difference
v(M, t) = t)—u2(M, t )
96
 Separation o f Variables
is the solution of the homogeneous boundary-value problem
L [ v ] = p v tt (or pvt)
v(M, 0) = v t(M, 0) - 0
dv
= 0
r ' J n + r *V S
In the class of functions A this problem has the unique solution
v = 0. Consequently, ux = u2. The conclusion that the solution
of the homogeneous boundary-value problem for a function belong­
ing to A was based on the validity of the expansion theorem.
However, the requirement that the expansion theorem is valid is
not necessary for the uniqueness of the solution of boundary-
value problems.
In the examples given below we shall prove the uniqueness of
the solutions of boundary-value problems without referring to
the expansion theorem.

Theorem The solution of the boundary-value problem

t)=pu„

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

«i«*(0, t)—P M 0 , t) = r f t ) , a2ux(l, t)+p2u(l, t) = r2(t)
where a,-, /?; > 0 (/ = 1, 2), k{x) > 0, P(x) > 0, which is continuous
in a closed domain {0 < .v < /; t > 0}, is unique.

Proof Let u f x , t) and u2(x, t) be two solutions of this problem.

The function v = ux—u2 is the solution of the homogeneous
problem

j x (kvx) = p v tt (65)

v(x, 0) = 0, vt(x, 0) = 0, 0< x f I (66)

« i^ (0 , 0 —0i»(O, 0 = 0, ct2vx(l, O + 0 2®(/, 0 = 0 (67)
and is continuous in the domain {0 < x < /, t > 0}. We shall
show that?;(A', t) = 0. Consider the auxiliary function
i
E(t) = y jj [k(x)v2x(x, t)+p(x)vf(x, 0] d.V ( 68)
0
7 97
 Mathematical Physics
It can readily be shown that E(t) is equal to the energy of vibra­
tions described by the equation (dfdx)(kvx) = pvtt for points
within [0, /] and is therefore called the energy integral. We shall
show that E(t) — 0. Consider the integral
i
E'(t) = ^ (kvxvxt+ p v tvtt) dx
6
Integrating the first term by parts, we have
(
|t
E'{t) = kvxvt dx
|o ■1
o
Since the function v(x, t) satisfies Equation (65), the integrand
is zero and, therefore,

E'(t) = k v xvt = k(l)vx(l, t)v,(l, t)—k(0)vx(0, t)v,(0, t) (69)

If we are dealing with the first boundary-value problem, then

vt(0, t ) = v t(l, t) — 0 and, consequently, E'(t) = 0. If, on the other
hand, we are dealing with the second boundary-value problem,
then vx(l, t) = ^*(0, t) = 0 and, consequently, E'(t) — 0. In such
cases, E(t) = const = E(0).
Using Equation (68) and the initial conditions (66), we find
that E(0) = 0. Therefore, E(t) = 0 for both the first and the second
boundary-value problems.
In the case of the third boundary-value problem we must pro­
ceed as follows. From the boundary condition (67) we find that

vx(0, t) = — v ( 0 , t ) , vx{l, t) = — — *>(/, 0

and if we substitute these values into (69), we obtain

E'(t) = - ^ k(t)v(I, t)vt{l, t ) - l ' - k ( 0 ) v ( 0 , t)vt(0, t)

a2 al
It is evident that this expression can also be written in the form

P: k { l ) v \ l , t ) + JPi
± . k ( 0)^2(0, 0
2a 2aj
98
 Separation o f Variables
Integrating this relationship between 0 and t, we find that

E (f)-m = o -*?(/, o)]

o)]

Since E(0) = 0, v(l, 0) = z>(0, 0) = 0, it follows that

£(/)= - - ^ k ( r ) v ‘( l , 0 - - E k ( 0 ) v > ( 0 , r ) <0

However, it follows directly from the definition of E(t) that E{t) > 0.
Consequently, E{t) = 0. Therefore, E{t) = 0 for any solution
of the problem defined by (65)-(67), i.e.
i
^ [kvl-rpv]]dx = 0

Hence, it follows that

k{x)v2x {x, t)+p (x)v2(x, t) = 0
[k{x) > 0 and p(x) > 0], vx(x, t) = 0 and vt(x, t) = 0
Consequently, v{x, t) = const. Since v{x, t) is continuous in a closed
domain, we have
v(x, t) = v(x, 0) = 0
which was to be proved.

Remark The requirement that the solution should be continuous

in a closed domain is important since otherwise the solution would
not be unique. In point of fact, if we add a function u(x, t) to the
solution of, say, the first boundary-value problem, which is equal
to a constant C inside the domain {0 < x < /, / > 0} and is zero
on its boundary, we obtain a solution of this boundary-value
problem for any value of C.

4 . 4.2 Consider now the parabolic equation

— [kux] = p u t (70)

The proof of the uniqueness of the solutions of boundary-value

problems for parabolic equations is based on quite different ideas.9
i* 99
 Mathematical Physics
as we shall see below. To begin with, let us establish the fol­
lowing theorem.

Theorem Any solution u(x, I) of Equation (70) which is continu­

ous in the closed domain D = (0 < a < /, 0 < t < T) assumes
its maximum and minimum values either on the lower boundary
of the domain D (for I = 0) or on end boundaries (a = 0, x = /)•

Proof If u(x, t) = const, the validity of the theorem is obvious.

Therefore, let us suppose that u(x, t) ^ const. To be specific, we
shall prove the theorem for the maximum value. The proof for
the minimum value follows the same course, except that u(x, t)
is replaced by —u(x, t).
Let M r be the maximum value of u{x, t) on the boundary
t = 0, x = 0, x = I, and M D be the maximum value u(x, t) in D.
It is required to show that M r — M D. Let us suppose that M D > Mr .
Consider the auxiliary function
v{x, t) = n(x, t)-\-a(T—t)

where a > 0 and a < (MD—Mr )/2T. The function ^(.v, t) is con­
tinuous in D and, consequently, it reaches its maximum value in
D at some point (x1; tj). It is evident that v(x 1, t 1) ' ^ M D since
zj(a, /) > u(x, t) in D.
The point (xq, /,) cannot lie on any of the three boundaries
/ = 0, x = 0, x = /. In fact,

\v(x, 0)| < \u(x, 0 ) |+ a r < — M r) < M d

|«(0, 01 < |u (0 ,O I+ a (T -O < M r + - ( M D- M r ) < M D

|v(l, 01 < |u(l, 0l + a (r -0 < Mr + j ( M D- M r) < MD

whereas v(x i , t]) > M D.

Therefore, the point (aI9 tt) belongs to the domain D =
(0 < a < /, 0 < t < T) and therefore the function w(a , t) should
satisfy Equation (70) in this domain. However,

H*(.Vi , t{) = ^(A j , fj) = 0, Uxx(.Vj , tt) = vxx(x1, tt) < 0

i/, ( a , , fj) = v f X i , tj)+ a > 0, (or ^,(x, , L) > 0)
100
 Separation o f Variables

and, consequently, u(x, t) does not satisfy Equation (70) at the

point (xl9 q). This contradiction means that the original hypothesis
must be incorrect and, therefore, M D = M r .

This theorem is an expression of the obvious fact that heat

(or diffusing matter) is transmitted only from points at the higher
temperature (concentration) to points at lower temperature. Once
the initial temperature (concentration) is specified on the boundary
t = 0 of D x, this process begins from t — 0 onwards. It is evident
that the temperature at internal points cannot become higher
than the temperature at / = 0. The same applies when the tem­
perature is prescribed at x = 0 or x = /.
Corollary 1 (Uniqueness of the solution of the first boundary-
value problem). The solution of the first boundary-value problem

— (kux) + f ( x , t) = pu, (71)

w(0, t) = u^t), u(l, t) = p 2(t), u(x,0) = <p(x)

which is continuous in the domain D0, D0 = {0 < x < /, / > 0}
is unique.
Suppose that u: and u2 are two solutions of this problem. The
difference u — ul —u2 is the solution of Equation (70)' satisfying
the following boundary and initial conditions
u{0, t) = «(/, t) = 0, u(x, 0) = 0
This solution is continuous in any domain D = {0 < x ^ /,
0 < t < T}, and therefore assumes maximum and minimum
values on the boundaries of the domain (t = 0, x = 0, x = /).
It is evident that these values are equal to zero. Consequently,
u(x, t) = 0 in D, which was to be proved.
Corollary 2 If the solutions iq(x, t) and u2(x, t) of Equation (71)
are continuous in the domain D, and on the boundaries of the
domain (t = 0, x = 0, x = /) satisfy the inequality w^x, t) < uz{x, t),
then the following inequality is satisfied throughout D:
ut(x, t) < u2(x, t)
In fact, the function u(x, t) = u2—ui is a solution of (70); it is
continuous in D and is positive on the boundaries t = 0, x = 0,
x = /. Consequently, the maximum and minimum values of u(x, t)
are positive. It follows that «(x, t) = > 0 throughout D,
which was to be proved.
101
 Mathematical Physics

Corollary 3 (Continuity of the dependence of the solution of

the first boundary-value problem on the boundary and initial
values.) If in the boundary-value problems

j ^ ( k u x) + f ( x , t) =pu,

u(0, t) = u(l, t) = /i 2(t), u(x, 0) = <p{x)

and

j ^ ( k u x) + f ( x , t) = pu,

u(0, = u(l, = u(.x, 0) = v(x)

the functions /q , p2>Fi> <P>9 satisfy the inequalities
IA*i(0—A*i(0 I < e> IPziO—PziO I < £
for all values t ^ 0, and
| <p(x)—<p(x) | < e
at all points within the range 0 < x < /, then for solutions u{x, t)
and u(x, t) which are continuous in D, we have the following
inequality everywhere in D
I u(x, t ) - u ( x , t) I < £
This follows directly from Corollary 2.

PROBLEMS

1. Determine the vibrations of a string 0 ^ x ^ / with rigidly

fixed ends if, up to time t = 0, the string was in equilibrium under
the action of a transverse force F0 = const, applied at x = x0
at right angles to the undisturbed position of the string, and at
t = 0 the force F0 is instantaneously removed.
2. Determine the vibrations of a string with rigidly fixed ends
under the action of an impulse P, applied to the string at time
t = 0 at the point x = x0.
3. A rod with one end (x = 0) rigidly fixed is in equilibrium
under the action of a longitudinal force F0 = const applied at
the end x = /. At time t = 0 the force F0 is instantaneously removed.
Determine the subsequent vibrations of the rod.
102
 Separation o f Variables

4. One end of a rod (x = /) is fixed elastically and the other

(.v = 0) is given a longitudinal impulse P at the initial time t = 0.
Determine the subsequent vibrations of the rod.
5. Determine the temperature of a sphere or radius R whose
surface loses heat in accordance with Newton’s law by convective
heat transfer to the surrounding medium maintained at zero tem­
perature. The initial temperature of the sphere is /(r).
6. Investigate the cooling of a spherical shell R : < r < R 2
which loses heat from its two surfaces in accordance with Newton’s
law to an ambient medium at zero temperature. The initial tem­
perature is u(r, 0) = /(r), < r < R 2.
7. A closed spherical container of 0 < r < R contains dif­
fusing material whose particles multiply at a rate proportional
to the concentration. Determine the dimensions of the container
(critical size) for which the process will diverge if (a) zero concen­
tration is maintained at the surface of the container, (b) the wall
of the container is impenetrable and (c) the wall is semipermeable.
8. Find the eigenvalues and eigenfunctions of a rectangular
membrane with boundary conditions of types I, II and III. Show
that in the case of a square there are two eigenfunctions to each
eigenvalue.
9. Determine the eigenvalues and eigenfunctions of a rectangular
parallelipiped for boundary conditions of types I, II and III.
10. Determine the natural frequencies of acoustic resonators
in the form of (a) a rectangular parallelipiped and (b) a sphere.
11. Solve Problem 1 of Chapter 2.
12. Determine the longitudinal vibrations of a rod 0 < x ^ /,
one end of which is rigidly fixed and a force To = const is applied
to the other at time t = 0.
13. Determine the temperature distribution in a rod 0 ^ x ^
whose ends are maintained at a constant temperature (iq and u2)
and whose surface loses heat, in accordance with Newton’s law, to
the ambient medium maintained at a temperature u3 = const. The
initial temperature can be assumed to be arbitrary.
14. Determine the temperature distribution in a rod 0 < x < /
whose ends and surface lose heat in accordance with Newton’s
law to ambient media at constant temperature. The initial tem­
perature is arbitrary.
15. The pressure and temperature of air in a cylinder 0 < x < /
are kept at standard values. One end of the cylinder is open at
time / = 0 and the other is kept closed. The concentration of
a particular gas in the atmosphere is u0 = const. At time t = 0
103
 Mathematical Physics
the gas diffuses into the cylinder through the open end. Find the
amount of gas Q{t) which diffuses into the cylinder if the initial
concentration in the cylinder is zero.
16. Solve Problem 15 by assuming that the diffusing gas propa­
gates at a velocity proportional to its concentration.
17. Determine the electric field in a conductor 0 < < / with
negligible leakage and self-inductance, one end of which is insulated
and a constant e.m.f. E0 is applied to the other. The initial potential
is v0 = const and the initial current is zero.
18. Determine the electric field in a conductor with negligible
leakage and self-inductance if one end of it, x = /, is earthed and
an e.m.f. E0 is applied to the other through a lumped resistance R0.
Assume that the initial current and potential are both zero.
19. A conducting layer 0 < x < / is initially free of electro­
magnetic fields. At time t = 0 a magnetic field H0, which is parallel
to the layer, is applied to it. Determine the magnetic field in the
layer for t > 0.
20. Determine the temperature of a rod 0 < .x < I with ther­
mally insulated surface if its initial temperature is zero, one end
is maintained at zero temperature and the other is thermally insu­
lated, given that a source of constant strength Q is applied at time
t = 0 at the point x0, where 0 < Xq < I.
21. Determine the temperature of a uniform plate of zero
initial temperature if a heat flux of constant density q is applied
to it through the plate x — 0 beginning with t = 0, and the face
x — I is maintained at a temperature uo = const.
22. A constant current producing heat of density Q = const
is passed through a conductor in the form of a flat plate of thickness
I beginning with time t — 0. Determine the temperature of the plate
for t > 0 if the faces of the plate lose heat in accordance with New­
ton’s law to the surrounding medium. The temperature of the
medium is it0 = const and the initial temperature of the plate is zero.
23. The initial temperature of a sphere 0 < r < R is u0 = const
and its surface loses heat in accordance with Newton’s law to the
surrounding medium maintained at a temperature ux = const.
Determine the temperature of the sphere for t > 0.
24. A heat flux Q (per unit time) enters a beam of semicircular
cross-section through its flat surface and leaves through the remain­
der of the surface. Find the steady-state distribution of temperature
over the cross-section of the beam, assuming that the heat fluxes
entering and leaving are distributed with constant densities.
104
 Separation o f Variables
25. The pointer of an instrument is attached to the end of a rod
of length /, fixed at x = 0. Determine the torsional vibrations of the
rod if at the initial time t = 0 the pointer was deflected through an
angle a and then released with zero initial velocity. The moment of
inertia of the pointer with respect to the axis of rotation is 70.
26. A string, 0 < x < /, with rigidly fixed ends, is subjected at
time r = 0 to a continuously distributed force of linear density (a)
0 = d>0sin<x>r, (b) 0 = 0 O cos cot, where 0 = const. Determine
the vibrations of the string in each case.
27. Determine the vibrations of a string 0 < x < / with rigidly
fixed ends subjected to a force F — F0sincot (and F0cos tot) applied
at the point .y = 0 at time / = 0 in the absence of resonance.
28. Determine the temperature of a rod 0 < x < / with its
surface thermally insulated if at time t = 0 heat sources of density
0(t)sin(7tll)x appear in the rod. The initial temperature is zero and
the ends are maintained at zero temperature.
29. A heat source moves with velocity v0 = const along a rod
0 < x ^ / whose surface loses heat to the surrounding medium
in accordance with Newton’s law. The temperature of this medium
is zero and the amount of heat received by the rod from the source
per unit time is q = Ae~ht, where h is the heat-transfer coefficient
in the heat-transfer equation for the rod ut = a2uxx—hu. Determine
the temperature of the rod if its initial temperature is zero and the
ends are maintained at zero temperature.
30. Determine the longitudinal vibrations of a rod 0 < x < /
if the end x = 0 is rigidly fixed and a force F = A sinojf (and A
cos cot) is applied to the other end (x = /) at time t — 0, where
A = const.
31. Determine the temperature distribution in a sphere 0 < r < R
if its initial temperature is u0 = const and, beginning with t — 0,
a constant heat flux of density q — const enters the sphere through
the surface.
32. A rod 0 < x ^ / with thermally insulated lateral surface
and constant cross-section consists of two homogeneous rods
0 ^ x < x 0, x 0 ^ x ^ / with different physical properties. Deter­
mine the temperature distribution in the rod if the initial temperature
is/(x) and the ends are maintained at zero temperature.
33. Determine the temperature distribution in a homogeneous
rod with its surface thermally insulated containing a thermal capacity
C0 at the point x0. The initial temperature is arbitrary and the ends
are maintained at zero temperature.105
105
 Mathematical Physics
34. Determine the potential distribution in a conductor with
negligible self-inductancc and leakage if one end of it (x = /) is
earthed through a lumped capacitance C0 and a constant e.m.f. E0
is applied to the other (x = 0). The initial potential and current
are zero.
35. Find the solution of the first internal boundary-value prob­
lem in a circle of radius R for the Laplace equation subject to the
boundary conditions:
(a) u(R, <p) = Acostp, (b) u(R, <p) = A-j-Bsimp, (c) ujr=R = A x y ,
(d) u(R, <p) = A s m 2tp-\- B cos2<p.
36. Solve the second internal boundary-value problem in a circle
of radius R for the Laplace equation with boundary conditions:
dUl = A,
(a) dn (b) = A x, (c) — i = A(x2- y 2),
dn ic
du
(d) = A sin <p-\-Bsin3<p. Which of these problems is correctly
dn
formulated?
37. Solve the first internal boundary-value problem in a ring
* i < r < R 2 for the Laplace equation subject to boundary condi­
tions u\r=Rl = Ui, u\r=R2 = uz . Use the solution of the problem to
determine the capacitance of a cylindrical capacitor per unit length.
38. Determine the capacitance of a spherical capacitor filled
with a dielectric of permittivity
e = £j for a < r < c and e = e2 for c < r < b.
39. Determine the electrostatic potential in a sphere of radius
R, which is first charged to a potential u0 and then placed in an
infinite medium of permittivity e = e, for R < r < c and e — e2
for r > c. Consider the special cases c = oo, e2 = oo and £j = e2 —e-
40. Find the solution of the internal boundary-value problems
in a ring R x < r < R2 for the equation V2« = A subject to the boun­
dary conditions (a) u\r=R = ux, »|r=R2 = u2 > (b) u\r=Rl = ux,
du I
dn jr =k2 = “2-
41. Find the solution of the boundary-value problem V2n = 1,
w11-= rj = 0, u|r=R2 = 0 in a spherical layer R x < ;• < R2.
42. Determine the potential distribution u(x, y) inside a box of
rectangular cross-section —a < x < o, —b < y < b if two opposite
faces of the box (x = ±ar) are at a potential V0 and the others are
earthed.
 5
The Method of Green’s Functions
for Parabolic Equations

In Chapter 4 we considered solutions of boundary-value problems

for parabolic equations. In this chapter we shall discuss Cauchy’s
problem for the simplest parabolic equations for which the method
of characteristics will no longer be suitable.

5.1 UNIQUENESS OF SOLUTIONS FOR PROPAGATION OF

HEAT ALONG AN INFINITE STRAIGHT LINE

Uniqueness theorem The solution of Cauchy’s problem

a*Uxx+f(x,t) = ut ( 1)
u(x, 0) = fp(x) (2)

which is continuous and bounded in the closed domain D { = {—oo

< x+oo, t ^ 0} is unique.

Proof Let u f x , t) and u2{x, t) be two solutions of the problem.

By hypothesis, there exists a number M such that |w,| < M and
|w2| < M throughout D Consider the function v(x, t)
= u2(x, t)—Ui(x, t ) which is the solution of the problem

a2vxx = vt (3)
v(x, 0) = 0 (4)
and is continuous in Dlt where \v(x, t ) \ ^ 2 M throughout D{.
107
 Mathematical Physics

Consider the domain Db = {|x| < b, t ^ 0} and the auxiliary

function
A M lx2 ,\
w(x, t) = - f r r \ ~ 2 ^ a t ]

It is evident that zo(x, t) is a solution of (3) which is, continuous in

the domain Db. Moreover, on the boundaries of Db we have
|w(x, t)l ^ w(x, t). In fact,
2M
\v(x, 0)| = 0 < w ( x , 0) = - ^ - x 2

AM l b 2 \ AM
\ v ( ± b , t ) \ < 2M < ® ( ± O /) -

Therefore, Corollary 2 (Section 4.3) applies to the functions

a(.v, 0 and w(x, l) in the domain Db. Accordingly, \v(x, 01 ^ w(x, 0
throughout Db. Consider now an arbitrary point (xy, q) of Zfi. For
any sufficiently large b this point will belong to Db. Consequently,
, , ,, AM / x? , ,
K *i» b)! < + a ti

If we take an arbitrary e > 0 and sufficiently large b , we shall have

. , .. AM I x\ 2 \
K m , b)! < - p - y Y + a h I < e
Consequently, ^(x,, t,) = 0. Since the point (x ,, ti) is arbitrary, we
have v(x, t) = 0, i.e. u2 = i/, throughout Dx.

5.2 THE FUNDAMENTAL SOLUTION (GREEN’S FUNCTION)

ON A STRAIGHT LINE

5.2.1 Definition The fundamental solution G(x—x0, t) of the

simplest equation of heat transfer along an infinite straight line is
defined as that solution of Cauchy’s problem
a2uxx = u, (5)
it (x, 0) = d (.x .v0) (6)
which is continuous throughout Dx except for the point (xn, 0).
The fundamental solution is frequently also referred to as Green's
function.108

108
 Green’s Functions for Parabolic Equations
Let us derive this function. Consider, to begin with, the following
special Cauchy problem
a2uxx = u, (7)
(0, .v < 0
u (x, 0) = tp (.y) = v (x) = x>0 (8)

We shall seek a solution of this problem among functions of the

form f ( x / t 3). Substituting the function u — f ( x / t 3) into (7), we
obtain

■ £ /" (* )= " " / ' t o . for ^= —

In order that this should be an identity in z, it is necessary that

a — 1/2. The equation for f ( z ) is of the form

f"(z)+ ~ f(z) = 0 (9)

From the initial conditions (8) for u(x, t) we have

/ ( —°o) = 0, /(+ c o ) = 1 (10)
This formulates the problem for/(z).
Integrating Equation (9) we obtain

\nf'(z)= for f '( z ) = C e “ ^

and hence
zjla

f(z) = C e f2/4°2 d£ — 2aC e ■J,2di

This function satisfies the first condition in (10). From the second
condition we obtain a relationship for C:
GO

1 = 2aC ^ e~y2dy = 2aC |/n

—CO

from which C = 1/2a\/ ti. Therefore, the solution of the problem

(7)-(9) is of the form

|/4aM
u(x, t ) = f z~yl dy
(t t H ; S

109
 Mathematical Physics

It can also be written in the form

0 j /4 a lt

w (x,/) = ~ { [ dy
\/n J \n J

or

u(x, 0 = y 1+ 0
\ l/4a2t

where 0(z) = ( e- ^2dy is the error function.

v 71 0
If the condition given by (8) is written in the form
u(x, 0) = u0rj(x—x 0)
the solution of (7)-(8) will be the function
l W0 i x —Xo_Y\
u(x, 0 = y 1 + 0

If, on the other hand, u(x, 0) = it0bl(.x—x r) —r](x—x2)],

u0
u(x, t) ( 11)
4a2t 4a t
Suppose now that an amount of heat Q is liberated uniformly
in the segment [xt , x2] at the initial time t — 0. This is equivalent
to specifying an initial temperature

Q
u (.v, 0) = iv ( x - x l)~ r]( x - x 2)}
CP ( x 2 -Yj)

In view of (11), the corresponding solution of Cauchy’s problem is

, 1 A — .Vi \ ^ / A*— A 2
0 ___l —0 f
) 4a2t T 4a2/
u(x, t) — Q_ ( 12)
2cp Xi A"1
If we now allow the segment [xt , x 2\ to contract to a point x0,
keeping Q constant, the function (12) will tend to a limit given by
(-*—*o)2
Q_± 0 x —z Q 1 4 a 2t
cp dz ‘ \ |/4 a 2/ z = Xo CP \ / 4n a 2t

110
 Green’s Functions for Parabolic Equations

Direct verification will show that the function

i (*-*o)2
G ( x - x o, t) = — = e *»2'(12a)
y4 n a 2t

satisfies Equation (5). Moreover, G(x—x0, 0) = <5(x—x0), since

by allowing t to tend to zero in the sequence {?„}, where tn = 1/4a2n,
we obtain a sequence of values of the function G(x—x0, 0
= y/n/7ie~n(-x ~x°) which defines the (5function <5(x—x0) (see Appendix).
The continuity of the function G(x—x0, t) throughout Dlf except
for the point (x0, 0), is self-evident. Therefore, G(x—x0, t ) is a solu­
tion of the problem (5)-(6) and, consequently, is the required fun­
damental solution. The function G(x—£, t) gives the temperature
of an infinite straight line (for example an infinite thin rod) for
t > 0, in which an amount of heat Q — cp is instantaneously released
at t = 0 at the point x = £. When Q ^ cp, the temperature is
Qlcp\G(x-£, /)].

Remark If amounts of heat Q{ and Q2 are released instantaneously

at and £2 at the initial time, the temperature of the infinite straight
line due to these sources will be

—
cp
G { x S u 0+—
cp
G (x -h , 0

5.2.2 Definition The fundamental solution of the simplest

equation of heat transfer (Green’s function) on a semi-infinite
straight line subject to the boundary condition u(0, t) = 0 [or ux(0,t)
— 0] is defined as that solution of the problem

^ Uxx = Mt
u(x, 0) = d(x—x 0) —6 (x + x 0)

[or, correspondingly, of the problem a2uxx = ut, u(x, 0) = <5(;t—x0)

+ (5(x+x0)] which is continuous throughout the closed domain Dx
except at the points (—x 0, 0) and (jc0, 0).
Using the above remark, we find the solution to be

G*(x, x0, t) = G ( x —x 0, t)—G ( x + x 0, t)

( * —■*<])2 _ (* + * o )2'
1 4 a2/ p 4 a 2/
 Mathematical Physics

or, correspondingly,
G**(x, x0, t) = G (x —x0, 0 + G ( a '+ a 0, t)
( * - * o ) 2~
1 4a2/
e 4a2‘ + e
\/ 4na2t

5.3 PROPAGATION OF HEAT ALONG AN INFINITE

STRAIGHT LINE

5.3.1 We are now in a position to formulate the solution of

Cauchy’s problem. Consider, to begin with, the homogeneous
equation
a2uxx = ut (13)
u(x, 0) = <p(x) (14)
We shall use the temperature interpretation of the problem.
Consider an element of length d£ containing the point a = £
on the line t — 0. The amount of heat liberated at time t = 0 in this
element is cptp(£)d£. This amount of heat can be referred to the
point £. We thus have a point source which liberates the amount of
heat dQ = cpq>{£)d£ at time t = 0 at the point x = £. The tempera­
ture distribution along the infinite straight line for t > 0, which
results from this source, is

d^ G ( x - £ , t) = <p{£)G{x-£, 1) d£
cp
This also applies to any other element of length d | along the line
t = 0. In view of the remark in Section 5.2, it is natural to suppose
that the temperature due to the action of all such elements, i.e. the
temperature due to the initial temperature distribution u(x, 0) = y(x)
is given by
oo

U ( x , t ) = $ <r(£)G(x-S,t)d£ (15)
— CO

If this is correct, the function given by (15) will, in fact, be the solu­
tion of Cauchy’s problem (13)~(14). To verify this statement, it is
sufficient to show that this function satisfies Equation (13) for all a
in the range — oo < x < oo and t > 0 and the initial condition
given by (14).12

112
 Green’s Functions for Parabolic Equations

To begin with, let us verify the condition given by (14). According

to (15) and in view of the fact that G(x—£, 0) = 6(x—£), we have
00 00
u (.v, 0) = J <f (£) G (x—£, 0) dir = $ ?(£)d(.x-£)d£ = p(x)
— 00 — 00

The last integral is equal to <p(x) from the fundamental property

of the ^-function. It follows that the function (15) does, in fact,
satisfy the condition given by (14).
To establish that the function (15) is a solution of (13), it is
sufficient to show that this function can be differentiated with res­
pect to .y (twice), and with respect to t, under the integration sign.
In fact, if
00 00

uxx = J <p(£)Gxx(x—£, t)d£ and ut = |j <p(£)Gt(x—I , /) d£

—00 —00
then
00
a2uxx—u, = J <p(£){a2Gxx—G,} d£ = 0
— 00

since the function G(x—£, t) is a solution of (13).

It is clearly sufficient to show that the integral (15) converges,
whereas the integrals
oo oo co

J <p(£) Gt d | , J ?(l)(?xd£ and J y ( |) G « d f (15a)

—00 —CO —00
converge uniformly in De = {— co < .y < oo, t ^ e} for arbitrary
e > 0.
We shall suppose that 9?(x) is bounded, i.e. |<p(x)| ^ Af.
Let us substitute a = ( | —x ) //4 a 2tin the integral (15). We then
have (see Equation (12a))
00
u(x, t ) = — ^ <f(x-\-2aa \/1 ) e~®2da
y 7i —co
CO

Iu(x, 01 < - - ^ 1 dct

]/71 —oo
ou
M f
—]=■ \ e a da = M (16)
\7l '>13

8 113
 Mathematical Physics

Therefore, the integral given by (15) does, in fact, converge (more­

over, it converges uniformly) in the domain Dx and |«| < M. If we
assume, in addition, that <p(x) is continuous throughout, it will
follow that the function (15) is continuous in the closed domain Dx.
(This involves the additional assumption that the function q(x)
is bounded. If cp{x) is piecewise-continuous, the function (15) is
continuous throughout D x except for the points on / = 0 at which
<p'(x) is discontinuous.)

Remark It follows from (16) that if the initial values <fi(x) and
<p2(x) differ by less than e, i.e. \<P\(x)—f 2(x)\ < E f°r ah then the
corresponding solutions of Cauchy’s problem, ux(x, l ) and u2(x, t),
will also differ from each other by less than e, i.e.

|«i(*, t)—u2(x, t)j < £

It follows that the solution of Cauchy’s problem depends continu­
ously on the initial values.
Consider now the first of the integrals in (15a)
ou uu

\ 9>(£)G,d£ = - J — G M ,/) d |

9 ( * —£)2
— 00

The substitution a = (£—x)/]/4 a2t will reduce the first integral to

\ — — 9?(.v+2act]/1 ) e a*da
^ 2 I/ n t
00 *

This integral converges uniformly in the domain Dt for arbitrary

£ > 0, since the function ( M / 2 ] / n e ) e ~ al majorises the integrand and
the integral of the former converges. The second integral in (15a)
°o 2
can be reduced to I —-=-^(A '+2na|//t)e~0t2da by the same substitu-
—Joo \7l.t
*
tion. This integral converges uniformly in the domain Dt for arbi­14
trary £ > 0 since, in this domain, the function (M/e\/ji)<x2e- a2
majorises the integrand and the integral of the former converges.
The third integral in (15a) can be treated in a similar way. We have
114
 Green’s Fund ions for Parabolic Equations
thus shown that (15) does, in fact, give the solution of Cauchy’s
problem (13)-(14).
The solutions of the following problems can be obtained in
a similar way:
1. a2uxx ut , u(x, 0) = (pipe) (0<X<OO), n(0, t) = 0
00
u(x, t) = J ?(g)G*(x, £ ; / ) d £ (17)
o
2. a2uxx ut , u(x, 0) = 0 (0 < x < x), ux(o, 0 = o
00

«(.v,r) = S G * * (* ,£ ;/M f)d £ (18)

o
The proof that the functions given by (17) and (18) are solutions
of Problems 1 and 2 is quite similar to the proof given above.

Remark 1 It follows from (15) that the heat propagates along the
rod instantaneously. In fact, suppose that the initial temperature
<p{x) is positive over a finite segment (xl5 x 2) of an infinite rod and
is zero outside the segment. The temperature at an arbitrary point x
is then given by
•*2
u ( x , t ) = 5 <p(£)G(x—£ , t ) d £

It is evident that this function is positive for any x and any small
t > 0. We have reached this conclusion because the physical assump­
tions which we have used in the formulation of Cauchy’s problem
were not strictly accurate [for example in writing Equation (13)].

Remark 2 Formula (15) can be regarded as the convolution (see

Appendix) of the fundamental solution

1 -J iL
G ( x , 0 = - / = , - e 4a2'
X Ana11
with the initial function y(x), i.e.
u(x, t ) = G(x, t) * <p(x)
If the initial function rp(x) in this formula is taken to be an arbitrary
finite generalised function, then u(x, t) will also be a solution of
Cauchy’s problem.
8* 115
 Mathematical Physics

5.3.2 Consider a number of examples.

Example 1 Solve the Cauchy problem

lui, x < 0
a2uxx = u„ u(x, 0) = <p(x) = x ^ o

In accordance with (15)

00

u(x, 0 = 5 <p(£)G(x —£ >0 ^ 1

Ml ^ G(x—£, 0 d£+w2 J G(x I, t)d£

Substituting a = (x~£)/j/4azt we obtain

y 4a2t

u{xt t ) = - —
\n
[ e“®2d a-----i-
J 1/tt
f
t
e“®2 da

y 4a2t

0 ]/ 4a2, 0 —oc

= 5 ) - )/?t
* U +$
p4a2r
M|+ M2 , »2- i l i f / -V
| 74a2r

Example 2 Solve the Cauchy problem

m2mx* = , !/(*, 0) = A e- *2

From (15) we have

00

u ( x , t ) — A jj e~i2C(x—£, t) d£

116
 Green’s Functions for Parabolic Equations

Substituting a = (£—x)j)/4a2t, we obtain

u ( x , t ) = - ^ jj Q - i x + l a a ^ t )2e - “2 d a
\ ' 71
f J00
— — 00

00

Ae~*2 — ^ g - t a a ^ l - ( 4 a 2f + l ) a 2 ^
]//7l

4 x 2a 7t oo I W T /-—r-r-
, _,2 ------- r- * - I — ' + 1/ 1 + 4 a 2< g
/I e l + 4 a 2< f \ j / i + 4 0 2r

Y 71 \ da

If we now substitute 2ax]/ t j \ /1 + 4 a 2t + ] /1 + 4 a 2t a = /? in the

last integral, we obtain
-*2
yl 1 + 4 sir 1(* ^4 l+4<J2f
*i(.v,r) = - —,e - 7——^ - \ e-^'d/3 = e
1 V l+ 4 a 2f J ]/1 +4rf2t

5.3.3 Consider now the inhomogeneous equation. The solution

of Cauchy’s problem
a2uXx+f(x, t) = ut (19)
u(x, 0) = <p(x) (20)
will be sought in the form of the sum
u(x, t) — v{x, t)-\-w(x, t)
where the functions v and uo are the solutions of the following
problems:
v: azvxx = vt , v(x, 0) = q>(x)
w: a2wxx+ f ( x , t) = w, (19)
w{x, 0) = 0 (21)
The function v(x, t) is given by (15).
To obtain the solution of the problem defined by (19), (21), we
shall use again the temperature interpretation of Equation (19).
In this equation cpf(x, t) is the density of heat sources per unit
time. Consequently, the element of length d£ containing the point £
will liberate the amount of heat dQ = c p f (£, r)d£ dr in the time
interval (r, r+ d r). If this amount of heat is considered to be released17
117
 Mathematical Physics

at the point I instantaneously at time r, the temperature due to

this source is

^ G ( x - £ , t - t) = / ( f , i ) G ( x - f , / - r ) d |d r
cp
In view of the remark in Section 5.2, it is natural to suppose that
the temperature distribution due to all such sources distributed
along the line in the time interval between 0 and t will be given by
t OO
w{x, t) = 5 5 /(£> r)G(x—£, t —t) d£ dr (22)
0 —oo
This formula is, in fact, the solution of the problem (19), (21). The
condition given by (21) will, of course, be satisfied. To verify that
the function given by (22) does, in fact, satisfy Equation (19) can
be confirmed by analogy with the analysis given in connection with
Equation (15).
If the heat source acts only at the point £o, but is a function of
time, the function f(x, t) in (19) is of the form

f ( x , t) = f ( t ) d ( x —£0)
The solution of the problem (19), (21) is then of the form
t 00
w(x, t) = $ d(£—£o)f(j)G(x—£, t —T) dc dr
0 —00
/
= S./It)<j(.y —fo, t ~ r) dT
0
where we have used the properties of the (5-function.

Remark 3 The solution of Cauchy’s problem for the inhomoge­

neous equation which is initially zero,

a2uxx+ f { x , t) = u,, u(x, 0) = 0

can also be written as the convolution (two variables!) of G(x, t)

with f(x, t ) :
t oo
u(x, t) = G ( x , t ) * f ( x , t ) = \ J G(.v—f, / —r)/(£ , r ) d f dr
0 — oo 18

118
 Green's Functions for Parabolic Equations

Remark 4 The solution of the problem (19), (21) on the semi-infi­

nite line with the boundary condition zu(0, /) = 0 [or ^*(0, t) = 0]
can be obtained in a similar way and is given by
f oo

«-’(*, 0 = $ $ /(£> r)G*(x, £, t—r) d i dr

00
/ oo
(or zv = j SfG**d£ dr),
oo

5.4 PROPAGATION OF HEAT IN THREE-DIMENSIONAL

(TWO-DIMENSIONAL) SPACE

Consider now the Cauchy problem for the equation of heat conduc­
tion in two- and three-dimensional space. To begin with, consider
the homogeneous equation
a2 V2u = u, (23)

Definition The fundamental solution G(M, M 0, t) of (23) is defined

to be the solution which
1. satisfies the initial condition
u(M, 0) = d(M, M 0) (24)
and
2. is continuous throughout the closed domain D3 = {—oo
< x, y, z < oo, t > 0}, except for the point (x0, y0, z0, 0). x, y , z
and x0,y 0, z0 are the coordinates of M and M 0, respectively, and
b{M, M0) is the ^-function with the singularity at the point M 0. To
determine G{M, M 0, t) let us establish the following lemma.

Lemma If in the Cauchy problem

a2 V2« = ut , u(M, 0) = <p(M)
the initial function <p(M) is
<P(M) = <P\(x)(p2(y)vfiz)
the solution of the problem will be the function
u(M, t) = U\(x, t)u2(y, t)u3(z, t)
where W[(x, t), u2(y, t), u3(z, t) are the solutions of the corresponding
one-dimensional problems:
a2ulxx = ult, iq(x, 0) = (px(x)
and so on.19
119
 Mathematical Physics
Proof By hypothesis
a2V2(uxu2it3) = u2itici2ulxxJr u lu3a2U2yyJru 1u2a2u3zz
= u2u3uuJt-ul u3u2tJru1u2u3t = (uiii2u3)t
and therefore u = uxu2u3 satisfies the given equation and, clearly,
the initial condition also.

We note that 6(M, M 0) = S(x—x0)d(y—y0)d(z—z0). Applying

the above lemma to the problem (23)-(24), we obtain the required
fundamental solution
G(M, M0, /) = G(x—A'0, t)G(y—y0, t )G(z—z0, t)
Using the formulae for the one-dimensional fundamental solutions
G(x—x0, t), etc., we obtain
3 _ ( x - x o ) 2+ ( y - y 0) 2+ ( z - z 0y
ip 4a2‘

for the three-dimensional space, and

2 ( x - x 0y - + ( y ~ y 0y
e 4a^t

for the two-dimensional space.

If an amount of heat equal to Q is liberated at the point M n at
the initial time t = 0, the temperature at an arbitrary point M due
to this source for t > 0 will be Qlcp[G(M, M 0, f)]. Using this fact,
it is easy to construct the solution of the Cauchy problem for the
homogeneous equation
azV2 it = it,, u ( x , y , z , 0) = ^(.y, y , z )
The solution will be the function
00 00 00

— 00 — 00 — 00

The solution of the problem

a2y 2u-\- f ( x , y , z , t ) = ut , u(x, y, z, 0) = 0120
will be the function
t CO 00 00

0 - 00—00—00
(26)

120
 Green’s Functions for Parabolic Equations

The solution of the Cauchy problem for the inhomogeneous

equation
a2V2 u + f ( M , t) = Uf, u ( M , 0) = <p(M)
is equal to the sum of the functions (25) and (26).
The proof of these statements is almost the same as for the one­
dimensional case and will not be reproduced here.

Remark 5 The solution of the Cauchy problem

a2V2 it = ut , u(M , 0) = <p(x, y, z)
can also be written in the form of the convolution (in three variables!)
of the fundamental solution
3 x 2+ y 2+ z 2
G ( x , y , z, t) = e 4 a2t
\/Ana21
and the initial function <p(x, y, z):
u ( x , y , z, I) = G(x, y, z, t)*<p(x,y, z)

= $ $ $ G(x—£ , y —t], z —£, 0<p(i, v, t)d£ d/?dC

—oo

We have illustrated the application of the method of source

functions to the solution of problems in infinite space and half-space.
It is also possible to use this method to solve boundary-value prob­
lems in bounded (in the space variables) regions. For example,
if we define the fundamental solution of the first boundary-value
problem as the solution of the problem
a2uxx = u,
u(0 , t) — 0 = u(l, I), u(x, 0) = <5(x—x0)
which is continuous throughout the domain Dt — (0 < x < /, t ^0}
except for the point (xo, 0) [we shall denote this solution by G(x, x0,
t)], the solution of the problem
a2uxx = ut
u(0 , t) = u(l, l) = 0 , u(x, 0) = (p{x)
can be written in the form
i
u(x, I) = 59?(f)G (x, t) d£
o12
121
 Mathematical Physics

We note that if we solve this problem by the method of separation

of variables, we find that
00

— . 2 . n n . Tin . Tiant
G(x, x0, t) = -j- y sin ~j~x sin— x0sin ——
n= 1

The situation is quite similar for the second and third boundary-
value problems. However, this method is not ideal for bounded
regions and is not usually employed.

PROBLEMS

1. The initial current and voltage in a semi-infinite uniform con­

ductor 0 < -x < oo are both zero. The self-inductance per unit
length of the conductor is negligible. A constant e.m.f. E0 is applied
to one end of the conductor at time / = 0. Determine the voltage
in the conductor for t > 0.
2. Determine the temperature distribution in infinite space
given that Q uniformly distributed units of heat were released instan­
taneously at time t — 0 on a spherical surface of radius r0.
3. Find the concentration of diffusing matter in infinite space
if the initial concentration of the material is w|t=0 = u0 = const
for 0 < r < R and u\t=0 = 0 for r > R.
4. Solve Problem 3 for the half-space z > 0 assuming that
z0 < R where (0, 0, z0) are the coordinates of the centre of the
sphere in which the initial concentration is u0. Consider the case
where (a) the plane z — 0 is impermeable to the diffusing material
and (b) zero concentration is maintained on the plane z = 0.
5. Determine the temperature distribution in a semi-infinite rod
0 < x < oo with thermally insulated ends and lateral surface, due
to heat sources of density Q{t) on the segment (a, b) (0 < a < b)
beginning with time t — t0.
6. Use the source function found in Problem 2 to solve the
boundary-value problem

a2|wrr- f — ur j + / ( r , t) = ut, 0 < r, t < co

u(r, 0) = 7>(/'), M < oo , r2 = .v2+ y 2+ z 2

7. Find the temperature distribution in infinite space using the
fact that Q uniformly distributed units of heat are instantaneously12
122
 Green’s Functions for Parabolic Equations
released at t = 0 on each unit of length of an infinite cylindrical
surface of radius r0.
8. Use the source function found in Problem 7 to solve the
following boundary-value problem:

a2 — «rj + / ( / - , 0 = ut, 0 < r , t < co

u(r, 0) = <p(r), \u\ < co, r2 = x2+ y 2

9. Determine Green’s function for a point source on an infinite
line for the equation
 6
The Method of Green’s Functions
for Elliptical Equations

In this and in the following chapter we shall be concerned with

the principal methods for the solution of boundary-value problems
for elliptical equations of the form V2u = f{M), and with the unique­
ness of these solutions.

6.1 G R E E N ’S F O R M U L A . SIM PLEST PRO PERTIES

O F H A R M O N IC F U N C T IO N S

All the results which are derived in this chapter can be deduced
from a small number of formulae and relationships. These will
first be derived.
6.1.1 Suppose that the functions u(M) and v{M) have the following
properties:
1. These functions and their first-order derivatives are continuous
throughout a closed domain D, bounded by the surface S, except,
possibly, for a finite number of points.
2. The functions and their first-order derivatives are integrable
in D.
3. The functions have second-order partial derivatives which are
integrable in D.
Subject to these conditions, we have
R [ u , v ] = — J vL[u] dr
D

D D s
124
 Green's Functions for Elliptical Equations

where L[n] = div(k Vif)—qu. Subtracting ^[u,^] from R[v,u], we

obtain Green’s formula
du dv \ ,
^ {vL[ii\ —uL[v\) dr =
= \ k l v ~ ----- u z— I da ( 2)
dn dn
d s
For the one-dimensional case this assumes the form

{vL[u]—uL[vty d.v = k (?;- (20

dx

Corollary Suppose that L[w] = div(AVu). If we set v = 1 for

this operator in (2) and take u{M) to be the solution of the equation
L[u] = f ( M ) which, together with the first-order partial derivatives,
is continuous in the domain D = Df-S, then

(3)
5 ^ d,i=W )dr
For a doubly-connected region D', bounded by two concentric
spheres SR and S Ri(Rx < R), centred on the point M0, Equation (2)
can be written in the form

\^{vL[u\ — uL[v]} dr -- ^ d—
-U—v 1\ dcr
,
dr
d' sR \
du ,
k \ v ----- u ——1dcr (4)
dr dr

The minus sign in front of the integral over S Rl has appeared

because on this surface d/dn = —d/dr.
Suppose that L[u] = V2u. The continuous solutions of the
Laplace equation V2u — 0 are called harmonic functions. Direct
verification will show that in three-dimensional space the function
\/r is harmonic everywhere except for the point r = 0. In the two-
dimensional case, the function In (1 /r) is harmonic everywhere
except for the point r = 0.
Using (3) for functions which are harmonic in a domain D
and are continuous together with their first-order partial derivatives
in D + S , we obtain
du
do = 0 (5)
dn

125
 Mathematical Physics
6.1.2 Mean value theorem If the function u(M) is harmonic in
a spherical domain DR and is continuous, together with its first-
order partial derivatives, in DR = DR-\-SR, then its value at the
centre M 0 of DR is equal to the arithmetic mean of its values on
the sphere S R, i.e.

u (M °) = $ u^ d<7
Sr

Proof Let us substitute L[u] = V2w in (4), so that

v 1 0'm„m = V ( x - x 0)2+ ( y —y0)2- (z—z0)2)

' M„M
and take u(M) to be a function which is harmonic in DR, bounded
by S R and continuous together with ux, uy, uz in DR-\-SR. Subject
to these conditions, the integral over D'(D' <= DR) is equal to
zero. The integrals over S R and S Rl of the product v du/dr are also
zero in view of (5) (the function v is equal to l/R and 1//?, on
these surfaces, respectively, and k — 1). We therefore have

Evaluating the derivatives and applying to the last integral the

mean value theorem for integrals, we obtain

^2 \ u (M ) da = 4nR]u(M% M* e SRl
Sr 1
If we now go to the limit i!, -+ 0 we obtain Equation (6).
We have considered harmonic functions in three-dimensional
space. In two-dimensional space (plane) the mean value theorem
is given by

U(M°) = 2^R S u ( W ds (J)

Cr

where CR is a circle centred on M 0. To derive this formula we

must take v = ln(l /rMoM) in the relationship analogous to (4).

6.2 UNIQUENESS OF SOLUTIONS OF BOUNDARY-VALUE

PROBLEMS
In this section we shall consider the uniqueness of the solutions
of the first and second boundary-value problems. It will be nec­
essary to distinguish between internal and external boundary-value126
126
 Green’s Functions for Elliptical Equations
problems. The formulation of the former problems is given in
Chapter 2. The first external boundary-value problem involves
the determination of the function which satisfies the equation
L[tt] = f ( M ) at points lying outside the closed surface S, and
assumes given values on S, i.e. //|s = <p(M). The second boundary-
value problem is formulated in a similar way.

6.2.1 Consider to begin with the uniqueness of the solutions of

internal boundary-value problems.

Theorem (maximum and minimum values) A function u(M)

which is harmonic in a finite domain D bounded by the closed
surface S and is continuous in D = D + S reaches its maximum
and minimum values on the boundary S.

Proof Let Hs represent the maximum value of u on 5 and HD

the maximum value of u in D. It is required to show that HD = Hs.
Let us suppose that this is not so. We then have HD > Hs and
u(M0) = H d at some point M 0(M0 e D).
Consider the auxiliary function

v (M ) = u(M) + ^— — - [ ( x - x 0)2- f ( y - y 0)2+ ( z ~ z 0)2]

where d is the diameter of D, i.e. the upper limit of the distance

between points in D; (x0, ,y0, z0) and (x, y, z) are the coordinates
of the points M 0 and M, respectively. It is evident that for all
points M e D
( x - x 0)2+ ( y —y Q)2+ ( z - z 0)2 < d 2

and v(M0) = u(M0) — HD. On the other hand, at a point M on

S we have

v (M ) < H S+ H^ Hs

Consequently, the function v(M), which is continuous in D, should

reach its maximum value at some internal point M x of D. At this
point we should have V2v < 0 since, at the point where the maxi­
mum is reached, none of the derivatives vxx, vyy, v Z2 can be positive.
On the other hand,

V2v = V2« + 3 — = 3 ——T — > 0

dL d127
127
 Mathematical Physics
This contradiction shows that the original hypothesis, namely,
H d > Hs, was incorrect. Consequently, HD = Hs. By applying
this result to the function —u, we obtain the proof of the theorem
for the minimum value.

The uniqueness of the solution of the first internal boundary-

value problem for the equation V2w = f { M ) follows directly from
this theorem.

Uniqueness theorem The solution of the first internal boundary-

value problem
V2u = f ( M ) , u\s = cp{M)
which is continuous in the closed domain D — D + S , is unique.

Proof Suppose that the two functions w, and u2 are the solutions
of this problem. Their difference u = ul—u2 is a function which
is harmonic in D, continuous in D and zero on S. In view of the
above theorem, the maximum and minimum values of u are zero
and, consequently, u = ux—u2 = 0 throughout D.

It is readily shown that the solutions of the first internal boun­

dary-value problem depend continuously on the boundary values
for the equation V2w = f(M).

Theorem Let u f M ) and u2{M) be the solutions of the first internal

boundary-value problem for the equation V2u = f(M ), which
are continuous in D and assume the values <pfM) and <p2(M) on
the boundary S of the domain D. If then the inequality |<p:
is satisfied everywhere on S, the inequality
\ul(M)—u2(M)\ < e
is satisfied throughout D.

Proof The function u — ul —u2 is harmonic in D, continuous

in D and u\s = 9?\ —<p2. Since —e < cpi—(p2 < e, it follows from
the previous theorem that the maximum and minimum values
of u(M) lie between —e and e . Consequently |»| < e, i.e.
\itfM ) —u2{M)\ < e. The following theorem is also valid.

Theorem If a sequence of functions u1, u2, ..., u„, ..., which are
continuous in a closed and bounded domain D and are harmonic
in D, converges uniformly on the boundary of the domain, it will
128
 Green’s Functions for Elliptical Equations
also converge uniformly in D. The reader is recommended to
prove this theorem using the Cauchy criterion for the convergence
of the sequence.
6.2.2 Theorem All the solutions of the second internal boun­
dary-value problem

L[u\ = div (kVii)—qu = f ( M ) , = <P(M ) (k > 0, q > 0)

which are continuous together with their first-order partial deriva­

tives in a closed domain D = D-\-S, can differ from each other
only by a constant, i.e. for any two solutions ux and u2, we have
the identity ux—u2 = const. For different pairs of solutions these
constants will, in general, be different.
Proof Let iq and u2 be two solutions of the problem. The function
to = ux— u2 is a solution of the problem

L[w\ = 0 ,~ | = 0
on s
Using (1) for u = w and v — w we obtain

dr = 0

In view of the boundary conditions, the integral over S is zero

and, therefore,
{k(yw)2+ q w 2} dr = 0
D

whence k(Vzv)2-\-qw2 = 0. Consequently, k(Vw)2 = 0 and qw2 = 0.

If q ^ 0, then w = ul—u2 = 0. If <7 = 0, then from the identity
k(Vzo)2 — 0 it follows that Vzv = 0 since k > 0 and, hence,
to = ux—u2 = const. This proves the theorem.
Remark For arbitrary functions q>{M) and f ( M ) (even if they
are continuous), the second boundary-value problem cannot have
a solution. In fact, let L[u] = div (kVu). The relationship given
by (3) must then be satisfied for the solution u(M) of the second
boundary-value problem, which is continuous in D = D-\-S,
du
together with its first-order partial derivatives. Since
dn s
we have
dcr

9
129
 Mathematical Physics

Therefore, the functions f (M ) and 9o{M) must be related by

5 /(-^O d r = jj k H M ) dcr (8)
D S

In particular, if f (M) — 0, the function y ( M ) must satisfy the

condition
jj k(p dcr = 0 (9)
s
The physical significance of (8) and (9) is readily understood
if u{M) is interpreted as the steady-state temperature distribution
and k(M) as the thermal conductivity. The relationship given by (8)
then expresses the following self-evident fact: in order to obtain
a steady-state solution it is necessary that the amount of heat
released by internal sources in the domain D during time At be
equal to the total heat flux which has passed through the boundary
S of the domain during the same interval of time.
The uniqueness of the solution of the third internal boundary-
value problem which is continuous in 15, together with its first-order
derivatives can be proved in a similar way. The reader is recom­
mended to establish the proof for himself.
6.2.3 To ensure the uniqueness of solutions of external boundary-
value problems, these solutions must satisfy additional conditions
with regard to their behaviour at infinity. In fact, if we seek the
solution of external boundary-value problems, these solutions
must satisfy additional conditions with regard to their behaviour
at infinity. For instance, if we seek the solution of the first external
boundary-value problem for r > R, subject to the boundary con­
dition u\r=R = C, where C is a constant, the solutions will be
ux — C,u2 — CrjR and it — A ulJr Bu2, where A and B are arbitrary
constants such that A-\-B — 1.
Let Dx denote the domain of points lying outside the closed
surface S. In three-dimensional space we then have the following
theorem.
Theorem The solution u(M) of the first external boundary-value
problem for the equation V2z/ —/(M ), which is continuous in
the closed domain Dx = DX-\~S and tends uniformly to zero
as M tends to infinity, is unique.
Proof Let iq and u2 be two solutions of the problem. The function
u — Wj—w2 is harmonic in D ,, continuous in Dx and u\s = 0.
Let us construct a sphere S R centred on some given point,
M 0, of D, which is bounded by the surface S and is large enough130
130
 Green’s Functions for Elliptical Equations
for S to lie entirely in D{ (Fig. 6.1). The function it is harmonic
in the domain Dz, bounded by the surfaces S R and S, and is con­
tinuous in Dz. Consequently, in accordance with the theorem on
the maximum and minimum values, the function u assumes maxi­
mum and minimum values at points on the surfaces S R and S.

Fig. 6.1

Let us choose an arbitrary e > 0 and take R to be so large that

S R has \u\ < e . This is possible because u(M) tends uniformly to
zero as M tends to infinity. Since w|s = 0, the theorem just mentioned
will ensure that \u\ < e everywhere in Dz. Since e is arbitrary, this
means that u = ux—uz = 0. This proves the theorem.
In the two-dimensional case, instead of the condition that
u(M) must tend uniformly to zero as M tends to infinity, it is neces­
sary that the solution be bounded in Dx. The proof of the uniqueness
theorem for this case will not be reproduced here.
In the case of the second external boundary-value problem
in three-dimensional space, the solution must both tend uniformly
to zero at infinity and its second-order partial derivatives must
follow a certain prescribed behaviour at infinity. Thus, if the required
solution tends uniformly to zero at infinity, whilst the partial
derivatives ux, uy, uz tend to zero (as M tends to infinity) as A/r2,
the solution of the second external boundary-value problem
is unique.

6.3 THE METHOD OF GREEN’S FUNCTIONS

We shall now consider methods for solving boundary-value problems
for equations of the elliptic type:
L[u} ~ f ( M ) ( 10)

(ID
131
 Mathematical Physics
where
oq = ot^M) , <x2 = ol2(M) , a\ , a2 > 0 and ocj-f-a2 # 0
One of these methods is the method of separation of variables.
We shall illustrate it by the following two examples.

Example 1 Find the function u{r, cp) which is harmonic within

the circle D R of radius R, continuous in the closed domain DR
and assumes on the boundary of this region (r = R) the prescribed
values f((p), i.e.
V2w = 0 (12)
u{R,tp)=f{<p) ' (13)
Since the required solution u(r, <p) is unique, it must be periodic
in <p with a period of 2n, i.e.
u(r, tp^-ln) = u(r, cp) (14)
Since the solution is continuous in the closed domain DR, it follows
that it must be bounded in DR.
We shall seek the solutions of Equation (12) in the form (P(r)
W{cp) which are bounded in DR and are periodic in cp (with a period
of 27i). Let us write down the Laplacian in polar coordinates

-rTor- O ' ^ + r^ ' w = 0 ( 12a)

and separate the variables so that

r i - (?&’)-).& = 0 (15)
dr
W"+XW=0 (16)
Using (14) we find that
V(<P+7ji ) = 'I'{cp) (17)
For X < 0, Equation (16) does not have solutions satisfying (17).
Consequently, X ^ 0. For X > 0 we find that
Wipp) — ylsin \/Xcp-\-Bcos ]/Xtp
From (17) we find that X 2n = Inn and hence Xn = n2, where n is
an arbitrary non-negative integer. Therefore, the eigenvalues of
the problem (16)-(17) are
X„ = n2 (n = 0, 1 ,2 , ...)
132
 Green’s Functions for Elliptical Equations
and the corresponding eigenfunctions are
1, sin<p, cosq>, ..., sinncp, cosn<p, ...
For X = 0 the general solution of (16) is
W{(p) = A ay + B 0
It is only for A 0 = 0 that this will satisfy the condition given by
(17). Therefore, X = 0 corresponds to the eigenfunction W0(<p) = 1.
Consider now Equation (15). For X = n2 we have
r2& " + r & ' - n 2® = 0
The general solution of this equation is of the form

0 n( r ) = CBr " + - ^ (n > 0) (18)

<Z>o(0 = C o + D o ln - ( « = 0) (19)

Since the required solution is bounded, we must set Dn — 0

(n = 0, 1, ...) in Equations (18) and (19).
Therefore, the bounded solutions of (12i) which are of the
form 0(r) ¥{<$) and satisfy the condition given by (14) will be the
functions
nn = r"{A„cosnq)+Bnsmmp)
The solution of the problem (12)-(14) can now be written in the
form of the series
00
rf ) — '^_j r n(Ancosnq)Jr B nsmn<p) ( 20)
n=0

The coefficients A n and Bn can be found from (13) using the ortho­
gonality of the eigenfunctions within the range [0, 2n] with the
weight p — 1:
CO

fifp) = (A„cos,rupJr B nsmn(p)Rn (21)

71=0
2n In

Ao= — J m m , = m cos « f d^
o 6
2n

Bn = — ( / ( I ) sin/7^d| (n = 1 ,2 ,...) (22)

71 J
013
133
 Mathematical Physics
Remark 1 The series (20) with the coefficients given by (22) can
readily be summed. However, we shall not do this here since the
problem (12)-(14) will be solved by another method in Section 5.4.
This method yields the result in a closed form.

Remark 2 The solution of the boundary-value problem (12)-(14)

outside the circle can be represented by the series
00

u(r, <p) = ^ y n (Ancosti(p + Bnsinii<p)

n=0

whose coefficients can be determined from the condition given

by (13). For the annular region between two concentric circles R :
and R2, the solution can be represented by the series
00
u= ic„rn+ y - j (A„cosn<p-j-Bnsinncp) + A0-~B0Inr

whose coefficients (A0, B0, C„A„, C„B„, D„An and D„Bn) can be
determined from the boundary conditions

u (R i , <p) = M y ) , u(R2 , <p) = f 2(<p)

The reader is recommended to write out the corresponding for­
mulae for these coefficients.

Example 2 Solve the boundary-value problem

V2w = 0 (23)
w(0, >0 = «(/,>;) = 0 (24)
u(x, 0) = A(X), u(x, b)=f2(x)(25)

in the rectangular domain (0 < x < /, 0 < y < b}.

We shall seek the solutions of Equation (23) in the form &(x)
F(y) which satisfy only the homogeneous boundary conditions
(24). Substituting this form into (23) and separating the variables,
we obtain
0" 0"
T + Y = 0

This will be an identity if &"/ 0 = —2, F " j xF — /. = const.

134
 Green's Functions for Elliptical Equations

We thus obtain the following equations for the two functions:

0" - \ = 0 (26)
= 0 (27)
From (24) we find that
0(0) = 0(1) = 0 (28)
The problem defined by (26), (28) has only positive eigenvalues:

p (n 1 ,2 ,...)

The corresponding eigenfunctions are 0„(x) = sin(nn/l) x.

Consider now Equation (27). For X = /„ it has the general
solution
¥ n(y) = Cn cosh \X„y-\-Dn sinh ) /X„y
Consequently, the solutions of (23) which satisfy the boundary
condition given by (24) are of the form
un( x , y ) = ^n(xyFn(y)
The solution of the problem defined by (23)-(25) can be represented
by the series
CO

u(x, y) = X j (c «cosh 1 X ^+ ^n S in h }/Xny) sin — x

>i=i 1
The coefficients of this series can be determined from the boundary
conditions (25):

. nn
/ , m = 2 j C„ sin——x
n= 1
and
eo
. nn
fi(x) = y , ( c ncosh ~ b f - D n sinh — 6 sin - f x
n= 1
Flence,
 Mathematical Physics
and
i
„ , jt/2 , „ . , n n , 2 f „ .^ t
C„ cosh — 6 + Z)nsinh — b = - j ] /2(f) sin di
0
In Sections 11.1 and 12.4 we shall quote other examples of the
application of the method of separation of variables for equations
of the elliptic type, which will require the use of special functions.
Another method of solving the boundary-value problems for
elliptic equations is the method of Green’s functions. It consists
of the following. To begin with, one finds the solution of the prob­
lem (10)-(11) for special values of the functions f ( M ) and y{M).
In particular, one solves the problem
L(G) = — d ( M , P) (29)

( - c + “ £ L - 0 (3o)
This solution is called the Green’s function for the problem (10)-(11).
We shall require that the required function G(M, P) be con­
tinuous (together with its first-order partial derivatives if a2 ^ 0)
everywhere in a closed domain D except, perhaps, for the point
P at which G may have a singularity.
If Green’s function has been found, it can be used to find the
solution of the original problem (10)-(11). This can be done with
the aid of Green’s formula as follows

jj {GL[u]~uL[G]} dr = jj k l G^ ~ u j P \ dcr (31)

D S ' *

Here and in the ensuing analysis the derivative d/dn is evaluated

in the direction of the outward normal to S. Since in the domain
D, L [ u ] = f ( M ) and L[G] = —d(M, P), the last expression can
be written in the form
dit
jjR M ) G { M , P) drM+ jju(A/)<5(M, P) dzM = Jfc - u * ° \ da M
dn dn
D D S

The second integral on the left-hand side is equal to u(P) in view of

the properties of the <5 function. The last relationship can therefore
be written in the form

u(P) = jj k l c ~ - u — \ d<7„- jj 0 ( M , d r„ (32)

S ' I D

136
 Green’s Functions for Elliptical Equations

where the integration is performed over the coordinates of the

point M.
For the first boundary-value problem (a! — 1, a2 = 0)
G\s = 0 , U s = <P
and from Equation (32) we obtain the solution of the problem
d O )-(ll):

u ( P ) = - \ k < p { M ) ~ d o M- ^ G { M , P ) f { M ) d z M (33)
S D

For the second boundary problem (oq = 0, a2 = 1)

dG „ du
= 0, <p(M)
dn s dn
and from (32) we obtain the solution of the problem (10)-(11):

u{P) = $ ktp(M)G(M, P) doM- $ G(M, P )/(M ) dzM (34)

S D

For the third boundary-value problem (oq ^ 0, a2 0) we have

dG du , fP(M )
- a iG ----- “ % 14 1
dn S %2 s’ dn S a2 S a 2

and, in this case, (32) yields

u{P) = J G(M, P) dcrM J G(M, P ) f( M ) dzM (35)

Therefore, the original boundary-value problem (10)-(11) has been

reduced to the determination of Green’s function. Methods of
finding these functions will be discussed later.
We shall now consider some of the properties of Green’s func­
tions.
Green’s functions are symmetric, i.e.
G(M, P) = G(P, M)
To prove this let us use Green’s formula for G, = G(M, P,)
and G2 = G(M, P2) where Pj and P2 are arbitrary fixed points in the
domain D. We have

\ {GjL[G2]—G2L[G,]} d z M = \ k |g , - C 2S daM
D S ' n 1137
137
 Mathematical Physics
The left-hand side is equal to

- $ {G(M, P i) d (M , P2)—G ( M , P2) <5( M , P,)} drA,

D

= G(P,, P2) —G(P2, P.)

and, consequently,

do-A/
6' '

The integral on the right-hand side is zero; in fact, if vve are

dealing with the first (or second) boundary-value problem, this
follows from the boundary conditions for Gx and G2 (G\S = 0 or
dGjdn\s — 0). If we are dealing with the third boundary-value
problem, then by expressing dGl/dn and dG2/dn in terms of Gx
and G2, and substituting these values into the integrand, we obtain

r dG2 dGx
- ^ - G xG2- { - ^ G 2Gx = 0
dn a2 a2

Therefore, G(PX, P2) = G(P2, Px).

We must now investigate the properties of Green’s function
at the point P. We shall confine our attention to the case when
L[u] = V2. In this case, Green’s function has a singularity of
the form \ j4nrMP in three-dimensional space or (l/2^)[ln(l lrMP)\ on
a plane.
In view of the structure of the equation V2G = —d(M, P), which
is satisfied by Green’s function, one would expect that this function
may be written in the form

G(M, P) = P)

where v is harmonic in D (as a function of M), and tp(rMP) has

a singularity at the point P, i.e. at rMP = 0, and should satisfy the
equation V2<p = —8(M, P).
To be specific, consider the three-dimensional case. Let DP
represent the spherical region bounded by the surface S P and centred
on the point P. Let the radius of the region be R. If we integrate
the identity
V2V = - 8 { M , P)
138
 Green’s Functions for Elliptical Equations
over the domain Dp (Dp c D), we obtain

\ V2WdrM = - 1

By Green’s formula the integral on the left-hand side is equal to

dip
do-
dr M

and, therefore,

= -1

On the sphere S* the function dip/dr has a constant value and,

therefore,
dip
dr $
r = R rR
d° = - 1 for 4 * tf-d^di? = - 1
p
Hence,

ip(R) =
4nR
It follows that Green’s function G(M, P) is given by

* G(M , P) - ~ ~ + v ( M , P) (36)
mp

and, consequently, it has a singularity of the form 1/4nrMP at the

point P.
On the plane, the function G(M, P) is of the form

G(M, P) = — In + v ( M , P) (360
In ’Mr
The function v(M, P) is defined as the solution of the problem

i dv\ «i 1
V2w = 0, «2
_ r MP dn S 471

It is unique for the first (and the third) boundary-value problem

and is determined to within an additive constant for the second
boundary-value problem.
139
 Mathematical Physics
Green’s function is determined in a similar way for external
boundary-value problems. It is again symmetric and has the same
singularities for L[u\ = W2u-\-qu.
Using Equation (36) one can readily provide a physical interpre­
tation for Green’s function in the case of the operator V2u. We shall
do this for the first boundary-value problem.
Suppose that the surface S bounding the domain D is in the form
of an earthed conductor. Let us place an electric charge of magnitude
1/4it at the point P inside D. This charge will induce a certain charge
distribution on S and the potential in D will be equal to the sum
of the potential \jA7irMP due to the point charge and the potential
v(M, P) due to the induced charge. This sum is, in fact, equal to
G(M, P).
It follows that G(M, P) can be interpreted as the potential due
to a point charge placed inside an earthed closed conducting surface.
In this interpretation the symmetry of Green’s function is an expres­
sion of the reciprocity theorem for the point at which the charge
s located and the point of observation.

Remark 3 Green’s function determined in this way does not

always exist. For example, Green’s function for the second internal
boundary-value problem for the Laplace operator L[u] = V2u does
not exist since one cannot find a function v(M) (G = l/Anr+v)
which is harmonic in D, continuous in D together with its first-
order partial derivatives and satisfies the condition
dv 1
dn I!s 47i dn \ rMP
This is because the necessary condition ) cp&o = 0 is not satisfied.
s
In this case, Green’s function can be determined as the solution
of the boundary-value problem
1
V2G = - p , P),
S~o
where S0 is the area of the surface S. This function exists and is
defined to within an additive constant. Using (32), we can find the
solution u(P) of the second boundary-value problem (10)-(11)

t ( P ) = ^ k ( M ) G ( M , P)tp(M) d<rM
s

- ^ G(M, P)f(M) drM J — dcj

140
 Green’s Functions for Elliptical Equations
or

«(P) = J k(M)G(M, P) <p(M) daM- ^ G(M, P ) f ( M ) drM+ C

5 D

where

C = const da M

6.4 DERIVATION OF GREEN’S FUNCTIONS.

POISSON’S INTEGRAL

One of the methods of constructing Green’s functions is the method

of images. We shall illustrate it by the following examples.

Example 1 Derive Green’s function for the first boundary-value

problem for a half-space bounded by a plane Q (we can assume
without loss of generality that this plane lies in the z = 0 plane).
Let P be the singular point of Green’s function. Since

G{M, P) = — ^— + v
47trMp

the problem reduces to the determination of a function v which

is harmonic in the half-space under consideration (for example,
z > 0) and is equal to — lj4jtrMP on its boundary. It is clear that
this function is v = —1/47irMPl, where P1 and P are symmetric
with respect to the plane Q. In fact, the function —\l4nrMPl is
harmonic in the half-space z > 0 and is equal to \l4nrMP at the
points M e Q since for such points rMP = rMPl. Therefore, the
required Green’s function is

G(M, P) = --------------------
4 7irMP 47irMPl

This method of deriving Green’s function for a half-space

bounded by a plane is suggested by the physical interpretation of
Green’s function given in Section 6.3. In point of fact, if we place
point charges 1j4n and —1/4tt at the symmetric points P and Pj ,
the potential produced by these will then be a function which is
harmonic everywhere except for the points P and P l5 and is zero
on the plane Q.14
141
 Mathematical Physics

Similarly, for a half-plane bounded by a straight line /, Green’s

function is of the form
1 1 1
G (M , P) = In In
r MP 2tz rMPi
where the points Px and P are symmetric with respect to the straight
line I.

Example 2 Derive Green’s functions for the first boundary-value

problem in the case of two straight rays l{ and l2 forming a right-
angle.

Fig. 6.2

Let P be a singular point of Green’s function. The images

of P in /, and l2 will be the points P1 and P2 (Fig. 6.2). Let P3 be the
image of Pj and P2 with respect to the continuations of the two lines
l{ and l2. Green’s function will then be given by

G(M ’ p) = Tr~ln
1.71 T~~
t Mp _ i1.71 l n 7‘A/^P! “ i l n 7'MP2
^ + ^2 l l n rr_\tP3
"
In fact, in this case, the functions is equal to

1 iIn ----------
— -— 1 1— In
i -------
1 b -=—
,1 m
i ------
1
2^ i'mp1 2.71 rMp2 27i rM
It is harmonic within the right-angle D (as a function of the point
M) and is equal to —(l/27r)[In(l/rWP)] on its sides. This follows from
the fact that if M e lx then rMP = rMPl, rMPi = rMPi, if M e l2,
then rMp — rMPl, rMPl = >'mp3-
Example 3 Determine Green’s function for the first (internal)
boundary-value problem for the circle

G (A /,P ) = ^ - l n — + ®
2ai rAfp
142
 Green’s Functions for Elliptical Equations
The problem reduces to the determination of the function v, which
is harmonic within the circle and is equal to —(\/27i)[\n(\jrMP)] on
its boundary.
Let P be the singular point of Green’s function, and let P x be
the image of P with respect to the boundary of the region (the circle
C). Pl is the image of P with respect to the circle C if both these
points lie on the same ray drawn through the centre of the circle

and the product of their distances, p x and p, is equal to the square

of the radius, i .e. ppl = R2. If the point M lies on C, then it is evident
from Fig. 6.3 that
R
(37)

since the triangles OMP\ and OMP are similar. Therefore, the
function

2n PfMPy
is the required function. Consequently, Green’s function for
the first (internal) boundary-value problem for a circle is of the
form

G(M, P) —1—Ii n-----

1 - —1 Ii n -------
R (38)
2ji rmp 2n prMpl

Example 4 Solve the first internal boundary-value problem for

Laplace’s equation V2w = 0 in a circle.
The required solution is given by

u(P) = - 5 < p ( S ) ~ d s (39)

c
143
 Mathematical Physics
which is obtained from (33) with f ( M ) = 0. In the case under con­
sideration, Green’s function G is given by (38). Let us evaluate
dG/dn:
dG dG 1 1 1 1
cos (n, r) = cos (n, rMP) ■ cos ( u , rMPl)
d/i dr K ’ ' 2n rMP v 2n rMP^
From the triangles OMP and OMPx (Fig. 6.3), we find that
R2+ r2MPt ~ p \
cos (n, rMP) = R2+ r 2MP- p 2 cos (n, rMPl) =
2Rr MP 2 Rr MP,
and, therefore,

^T>2TI '..2MPi~~P\
dG 1 1 [ R 2+rl1P- p - „2
dn !IC 2n 2 R r MP
2 2Rr2MPl
Substituting for rMPi from (37) and for p { from the formula
p, = R2/p, we obtain
dG 1 R 2- p 2
dn 2siR rlMP
From the triangle OPM we find that rjfP = R2jr p 2—2Rpcos(0—yi)
and therefore
dG 1 i?2—p2
dn 2tiR R2+ p 2- 2 R p c o s ( 0 - y )
Substituting this value into (39) we obtain the Poisson integral
2rt
(R2- p 2M 6 ) d6
(40)
- s- S R2jrp 2—2Rpcos(Q—ip)

where (p, yf) are the polar coordinates of P, and (R , 0) are the polar
coordinates of the point M on C.

PROBLEMS

1. Derive Green’s function for the first external boundary-value

problem: (a) for a circle and (b) for a sphere (L[»] = V2//).
2. Derive Green’s function for the first internal boundary-value
problem for a circular sector 0 ^ ^
tp n/n (LM =
V2u).
3. Derive Green’s function for the first internal boundary-value
problem for (a) a spherical layer R { < r < R2 and (b) for a ring
i?! < r < R2.

144
 Green’s Functions for Elliptical Equations
4. Derive Green’s function for the first internal boundary-value
problem for a plane layer 0 < z < /* (L[//] = V2//).
5. Using the principle of the maximum and minimum of harmonic
functions, show that Green’s function for the first boundary-value
problem for the domain D is positive in D {L[ii\ = V2w).
6. Using the Poisson integral prove the following theorems.
(a) Any harmonic function which is positive on a plane is equal
to a constant.
(b) Harnak’s theorem. Let {iti(x, >>)}, i = 1,2, ..., be harmonic
functions in a finite domain D bounded by the contour r , which
00

are continuous in D. If the series Uj(x, y) is uniformly convergent

/=i
on F, it will converge uniformly in D and its sum will be a harmonic
function in D.
 7

Potentials

The method of separation of variables and the method of Green’s

functions was discussed in Chapter 6 in connection with elliptic
equations. The third method for solving boundary-value problems
is the method of integral equations, in which the solution is sought
in the form of certain special integrals (potentials) with unknown
charge (mass) density distributions. In the present chapter, we shall
consider the simplest properties of potentials and their application
to the solution of boundary-value problems.

7.1 VOLUME POTENTIAL

7.1.1 The electrostatic potential (in space) due to a charge e at

a point P is given by (at an arbitrary point M)

fMP
where rMP is the distance between M and P.
If there are charges elf e2, ..., e„ at the points P lt P2, Pn,
the electrostatic potential due to these charges is given by
e\ <?2
u(M) = + + 0)
f‘M P l r MP2 r M Pn

Suppose that charges are distributed with a density p(P) in a

domain D. A small volume element drp containing the point P will

146
 Potentials
contain the amount of charge p(P) drP. The potential due to this
charge is approximately equal to
p{P) dr,
I'm p

The potential due to all the charges in D is given by

u ( M) = [ ^ l drp (2)
g r MP

This integral is called the volume potential. In two-dimensional

space (plane) the volume potential assumes the form

u (M )= $p(P)ln(— W (3)

7.1.2 The volume potential is thus an improper integral. Consider

the more general improper integral

u ( M ) = \ f ( M , P ) drP (4)
D

where f ( M , P) is a continuous function of the two points M and

P, M A P, which becomes infinite as M -» P. We shall call the
integral (4) uniformly convergent in the neighbourhood of the point
M qif for any e > 0 there exists a number 5 such that:
1. for any domain DdM<) containing M Q and having a diameter
d less than d, i.e. d(T>m0) < e, and
2. for all points M separated from M 0 by a distance less than
8(MM0 < d).
5 f ( M , P ) drP < e
D b
M0
This idea lies at the basis of proofs of a number of properties of
potentials. The main property of a uniformly convergent improper
integral is expressed by the following theorem.

Theorem An improper integral which is uniformly convergent

in the neighbourhood of a point M 0 is continuous at that point.

Proof Consider the difference

u ( M ) —u(M 0) = u1( M ) - u l(M0)-]-{u2 (M )—U2 (M0)}
10* 147
 Mathematical Physics

where

Mi(M) = J /( M , P) drP, uz(M) = S f ( M , P) drp

n r»^

Since the integral (4) converges uniformly in the neighbourhood

of M0, it follows that for arbitrary e > 0 it is possible to find a num­
ber <5 such that for the domain D6Mo with d (DsMf) < d and for all
points M, separated from M 0 by a distance less than (5, we have
the following inequalities

|wi(M)| < — , Im^M q); < y (5)

Since M 0 £ D — £>m0, the integral uz{M) is not an improper integral

and the function uz(M) is continuous at M 0. Consequently, for the
same e we can find a number <5t such that for all the points M,
separated from M 0 by distances less than <51} we have

\uz( M ) - u z(M0)\ < y ( 6)

Let <5, = min {5, <$,}. For all points M, such that M M 0 < d2, we
then have inequalities (5) and (6) and, consequently, the inequality

u ( M ) —u(M 0) 1 < e
This proves the theorem.

We note that the uniform convergence of an improper integral

ensures that it is convergent at the point M 0.

7.1.3 Let us now consider the simplest properties of the volume

potential with density p(P), | p(P)\ < A.

Property 1 The volume potential exists and is continuous every­

where.
If the point M 0 does not belong to D, the integral u(M0) is not
an improper integral. Since the integrand, regarded as a function
of M, is continuous at M0, the integral u{M) must be continuous
at this point. If M 0 e D, then according to the theorem of Section 7.2
and the remark at the end of Section 7.2 it is sufficient to know that 148
148
 Potentials
the integral converges uniformly in the neighbourhood of M 0.
Consider the integral
P(P) QTp
,
r MP
DM o

It is evident that
! ij p(p) d rP ^ e \p \ drP < A \ dtp < A drP
fMP ^ r MP f MP
r2S ^mp

to
JMo WM0
°Mo "Mo 'M

O
where T$ is a spherical region centred on M and of radius 2d
(.D 5Mo c= T lf) . Transforming to polar coordinates, we have
2d 7t 271
f _^Tp_ _ ^ C C f /-sin 6 dr dddtp = 8And1 (r = rMp)
J
t 2<5
rMp J J
0 0 0
J
1M

Therefore P(P) drP! < 8 A n d 2. To ensure that this integral

r MP
o
is less than a given number e, it is sufficient to take d < j/e/8nA.

Property 2 The first-order partial derivatives of the volume

potential with respect to the coordinates of the point M are every­
where continuous.
If M 0 D, the integral u(M0) is not an improper integral. Since
the integrand, regarded as a function of the point M, has continuous
first-order partial derivatives with respect to the coordinates of M
at the point M 0, the integral u(M) will also have this property and
the derivatives can be evaluated by differentiating under the integral
sign:
du du 0i — y ) p (P) drP
dx \
D
(£ -* )
r MP
p(P) drP,
dy \ rMP

du
dz \ (L r MP
l z2 P (P) drP (7)

where (£, p, Q are the coordinates of P.

If M 0 e D, then it is sufficient to show that the integrals of the
derivatives on the right-hand sides of (7) converge uniformly in the
neighbourhood of M 0. One can then differentiate under the integral
149
 Mathematical Physics

sign and the formulae given by (7) are valid for du/dx, du/dy and
dujdz. To be specific, consider the integral

(i-x)p(P)
d r.
r MP

It is evident that

i C ( f - * ) P (P) drP < A |£—x\ d rP drp

r3 < A
' MP rMP rMP r2Mp
JMa Mq 'Mo
since \£—x/rMP| = |cos(r, n)\ < 1 . Next
2(5 n 2n

A ^\ ^ <P A T ^ ~ 2 —= A ^ ^ sin 0 dr d0 <\<p = SrtAd

6 rMP i s rMP £ o£
'Mo M

To satisfy the inequality

(t-x)p(P)
r3 drp < e
r MP
'M0
it is sufficient to take 6 < e/S tiA .

Property 3 The volume potential is a harmonic function outside

the domain D in which the charges (masses) are located.
This property follows from the fact that for points M £ D the
integral (2) is not an improper integral and, therefore, the Laplace
operator can be taken out from under the integral sign:

since for points M D we have V2(l jrMP) = 0.

Property 4 At points in the domain D the volume potential sat­

isfies the equation

V2it = - 4 n p ( M ) (8)
We shall suppose that p(P) has bounded and integrable first-
order partial derivatives dp/dH, dp/dr], dp/dt,.
150
 Potentials
Proof The second-order derivatives d2u/dx2, d2u/dy2, d2u/dz2 can­
not be obtained by differentiating the right-hand side of (7) under
the integral sign since this yields divergent integrals, for example:
3 ( Z - x ) 2p(P)
—j — di> + d rP
I' m p r MP

To establish the existence of the second-order derivatives we shall

proceed as follows. Suppose that M 0 e D; T dMo is a spherical region
of radius 6 centred on the point M 0 and bounded by the spherical
surface SlIo, where c D. For points M and T 6Mo we can write
u(M) = ux(M)-{-u2(M)
where

ufM ) P_iP) drP, u2{M)

p(£)
dtp
r MP r MP
D

The integral u2(M) is not an improper integral.

In view of Property 3, it is harmonic at the point M 0, i.e.
V2»2Im=m0 = 0. Consequently, V2i/|M=Mo = «iV2|m=m# and there­
fore it is sufficient to consider the function u f M ) .
The derivative
dux j —x
dr i ~ 1 —— 1dr
dx - $
7"i<5
rMP 5 dx \ rMP
1M 0 Mo
can also be written in the form
dp

1l x7
\
J
TM 0
p(P)f t ± - W
d£ \ r\fp I
=
1 Mo
1 - * dT,-
rMP
ji(
1 Mo
P -W
rMP I

Applying Green’s formula to the second integral, we obtain

dp

dux \ P (P)
« d r , - cosa dfTr (9)
dx rMP i r MP
'M0 ^ Af0

where a is the angle between the direction of the outward normal

to S dMo and the x axis.
The first integral on the right of (9) is a volume potential with
charge (mass) density px(P) = dp/d£. Therefore, in accordance
151
 Mathematical Physics

with Property 2 it has a continuous first-order derivative with

respect to x. The second integral is not an improper integral and
therefore possesses a continuous first-order derivative with respect
to x for any internal point M of T ^ 0. Consequently, d u j d x has
a continuous second-order derivative with respect to * in 7 ^ 0.
At the same time,

dp_
( £ — x 0)
d2ul d£ POP) ( I —x0) cos a da>
dri
dx2 M=Mq rpi 5 rM0P - ! rM0P
1 Mo 3M0
However, (£—x0)/rM P = cos a, and therefore
dp_
d2Ui (£ —*o)
dr p P{P) cos2ada> (10)
'dx2 - s \ r2
M = Mo -,5 I' m q P PMqP
1M q *M

Similarly, we find that

d2Ui
dy2 m =m 0
S
[A/n
(.v-yo) O
tP-
Yj
M qP
d tp -
A
\
3Mo
f M0 P
cos2/?da> (11)

dp
d2u i ( C - z 0)
dC
d T p - ( 4 — cosVddp (12)
dz2 M = M q *5 'M qP A rMoP
‘ M q 3M q

where /? and y are the angles between the normal to S\/o and the y
and z axes, respectively.
Combining (10), (11) and (12), we have

dp_ dp_
d£ dij
V2w|m =m0 — V2iti M = Mq J v2 co sad rP + \ —— cos/? drP
T„
<5 n t 0p
- -3 '" o p
J M 0 1 M q

(* ^P_
, \ d£ f P do-p
+ —2 cosy dtp — \ —2 (13)
j i Pm q P •> r M 0p
' M q 'M q

15l
 Potentials

If we repeat the analysis leading to Property 2, we find that each

of the integrals over the domain T sMo in (13) will not exceed 4jzBd,
where B is the upper bound of the functions \dpjd$\, \dp/dr]\,
\dp/dC\, i.e.
J | < 4tcB8 (14)
.5
Mo
Using the mean value theorem in (13) we obtain

—y — da> = 4ttp (P * ) (15)

r M 0P

where P* e S sMo. Proceeding to the limit in (13) as 8 0, and

bearing in mind (14) and (15), we obtain

V2w| m = m 0 — 4jcp(M o)

In the two-dimensional case the analogue of (8) is

V2» = —2tcP(M0) (16)

Properly 5 As the point of observation tends to infinity, the

volume potential tends to zero (in the three-dimensional case, D
is a bounded domain).
To prove this property let us apply the mean value theorem to
(2). We obtain

it ( M ) = ------- \ p drP = ------

rMP* J i~m p *

where P * e D, m = \ p d r P is the total charge. This proves Property 5.

Example Let us find the volume potential due to a uniformly

charged sphere D of radius R. It is evident that the required
potential is a function of the distance R from the centre of the
sphere to the point of observation:
u ( M ) = u{r)
Outside the sphere D we have V2» = 0 and, consequently,
u = Cl/rJr C 2. In view of Property 5 we have u(r) -> 0(r -> co).
Consequently, C2 = 0. Inside the sphere D we have V2u = —4np
o rd (r2w')/dA- = —4;rpA-2. Consequently, u(r)— (—2j3)xr2p-{-(A/r)~\-B
for r < R. Since the volume integral is bounded everywhere, it153
153
 Mathematical Physics

follows that A = 0. From the condition that the potential and its
first-order derivatives must be continuous, we find that
2 , C, 4 C,
— ~ n R 2p + B = -jj- and — —n R p =
R2
and hence Cx — (4/3) ti R 3p , B = 2 n R 2p. Therefore,

-y * (3 R2- r 2)p r< R

u{r) =
4 R3p
— n ----- r >7?
3 r
Remark The volume potential can be written in the form of the
convolution (in the variables x, y, z) of the fundamental solution
\j47tr = (l/47r)(x2+>^-(-z2)“1/2 of the Laplace equation V2m = 0
[V2(1/4jt/-) = —3(x, y, z)] and the function 4n.p(x, y, z):
p ( £ , V, Q d^ dr/dC
\ / ( x — f )2+ ( V■
— v )2+ (z — t ) 2

P (B )
I' m p

All the above properties of the volume potential can be deduced

immediately from this expression. For example,

j j
it = V21— *p = V21— * p = —47tS (M) * p(M) — —4np(M)

since V2(l/r) = —4ti<5(A/) and d(M)*p(M) = p(M).

7.2 POTENTIAL DUE TO A SIMPLE LAYER

Suppose that charges (masses) are distributed on a surface S with

density p(P). The potential due to these charges is given by

v( M ) = [ ^ - d o P (17)
s rMP
This integral is called the potential due to a simple layer. We shall
suppose henceforth that the function p(P) is bounded, \p(P)\ ^ H,
and the surface S' is a Lyapunov surface. The surface S is defined
as a Lyapunov surface if it has the following properties.154
154
 Potentials

1. The surface S possesses a tangential plane at each point on it.

2. For each point P on S there exists a neighbourhood S P which
is such that any straight line parallel to the normal at P intersects
S P not more than twice.
3. The angle y(P, Pi) = (n P, n Pl) between the normals n P and
n Pi at the points P and Pi satisfies the following condition:
y(P, Pi) < ArPPl
where A and d are constants and 0 < (5 < 1.
Let us consider some of the properties of the potential due to
a simple layer.

Property 1 The potential of a simple layer is defined everywhere.

For points M not belonging to the surface S this is self-evident.
If M e S the integral given by (17) is an improper integral in'the
two-dimensional region S. It is known that the improper two-dimen­
sional integral

[ d(TP
' r MP

converges absolutely if a < 2. In our case, a = 1 and, consequently,

(17) converges.

Property 2 The potential due to a simple layer is continuous

everywhere.
If M $ S, the integral given by (17) is not an improper integral
and its continuity follows directly from the continuity of the inte­
grand, 1/rMP.
If M 0 e S, it is sufficient to show the uniform convergence of (17)
in the neighbourhood of M 0. Consider the integral
p (P) drtP
®i
rM P

taken over the surface S m0 (Sm 0 c S ) containing the point M 0

and having a diameter less than 6, dOS^0) < <5. We shall use a set
of coordinates with the origin at M0 and the z axis parallel to the
normal to S' at this point. Let M(x, y, z) be an arbitrary point sepa­
rated from M 0 by a distance less than d(MM0 < (5). Let repre­
sent the projection of the surface S ^ 0 on to the (jc, y) plane and let
Q2m xbe the circle on the (x, .y) plane, centred on the point Mj(x, y, 0)
and of radius 2d. It is evident that c= Qj$r The projection of15
155
 Mathematical Physics

a surface element da on to the (x, _y) plane is ds = da cos y, where y

is the angle between the normal to S and the z axis. It is clear that
daP
IVy{M)\ < H
t)2- h ( y - v ) 2+ ( z = W
o

da p
< H
1/ ( x —£)2- r ( y - rj)2
'Mo
ds
co sy ]/(x —l)2+ (y —v)2
-M0
From the third property of Lyapunov surfaces, <5 can be taken to
be so small that for points P e S 5Mo we have cos y > 1/2. We there­
fore have
ds
V ( x —£)2- r ( y —v)2
JM0

< 2H
s ds
}r( . x - t y + < y - v ?

If we now introduce a polar set of coordinates with the origin at

Mi, we can readily evaluate the last integral. It is equal to
2d 2n
2H ^ = 2 H ^ drd<y = 8 r t//(5
■2m
Q as, o o

In order that the integral |u1(A/)| be less than the given number e,
it is sufficient to take 5 < 1/8 tiH.

Property 3 The potential due to a simple layer is a harmonic

function everywhere except for points on the surface S.
This property follows immediately since for points M S the
integral (17) is not an improper integral and therefore

Property 4 If the surface S is bounded, the potential of the simple

layer tends to zero as M tends to infinity.
156
 Potentials
To prove this, let us apply to (17) the mean value theorem:

c(M) = — (p(P)d(TP = — ■ 08)

I mp* J i'mp*

where P * e S, m — \pdct is the total charge,

s
Property 4 follows directly from (18).

Property 5 The normal derivatives of the potential due to a simple

layer have a discontinuity of the first kind at points on the surface
S , and the magnitude of the discontinuity is 4np{M).
We shall not reproduce the proof of this property here.
For the two-dimensional case (plane) the potential of a simple
layer is of the form
v (M) = \ p (P) In l - ^ - A dsP
O \ 1MP I

Properties 1-3 are valid for this function. As M tends to infinity,

v(M) tends to infinity as In rMP. The discontinuous change in the
normal derivatives at points on the curve C is 2np(M). The proofs
of all these properties are similar to those given for the three-dimen­
sional case and will not be reproduced here.

7.3 POTENTIAL DUE TO A DOUBLE LAYER

7.3.1 Consider charges —e and e at points Pi and P2, respectively

(Fig. 7.1). The electrostatic potential due to this dipole is given by

w(M) — e
rMP2 rMPi
or
zv(M) = <?/?— (—^
on \ rMP p=p*

157
 Mathematical Physics

where P* is a point on the segment P iP2 and the derivative is evalu­

ated in the direction n of the segment drawn from Px to P2 (axis of
the dipole). The quantity h is the distance between P1 and P2. The
dipole moment is defined by eh = v. If we allow the points Px and
Pz to approach each other at constant v (by increasing the magnitude
of the charges e), then in the limit as h -> 0 we obtain a point dipole
located at P. The potential due to this dipole is given by

where the derivative is evaluated with respect to the coordinates of

P and in the direction of the dipole axis.
Let S be a two-sided surface with continuously varying tan­
gential plane. This means that if at some point P on the surface we
choose the positive direction of the normal n P to the surface and
allow P to move along a closed curve (lying on S), then if the direc­
tion of the normal varies continuously, as we return to the initial
point the direction of the normal will become the same as the original
direction. One of the directions of the normal to the surface can be
taken as positive since the unit vector in this direction will be con­
tinuous on the surface. We shall assume that this positive direction
has been chosen.

7.3.2 If on a two-sided surface S there is a distribution of dipoles

with dipole moment density v(P) such that the axes of the dipoles
at each point coincide with the positive direction of the normal,
then the potential due to these dipoles is given by

(19)

Let S be a two-sided surface with chosen positive direction of

the normal. Let us imagine that we have drawn lines of length h

Fig. 7.2

158
 Potentials
along the positive direction of the normal at each point. The geo­
metric locus of the end-points of these lines is a surface S l which
is separated from S by the distance h. Suppose that on 5 there are.
negative charges distributed with density v(P)/h and on S x there are
positive charges with the same density (Fig. 7.2).
We thus have a double layer of charges of opposite signs, which
can be regarded as a set of dipoles distributed over S and S { with
the density v(P)/h. The potential due to the dipole on an element
da of S and S x is v(P)(d/dn)(\/rMP)da. The potential due to all the
dipoles is therefore given by

If we let h tend to zero, we obtain a double layer on S, whose

potential is given by (19).
Since
d / 1 \ cos cp
dn \ ?mp I rltp
where y is the angle between the positive direction of the normal
to S at P and the line P M , the potential due to a double layer can
also be written in the form

iv(M) = ^ v(P)— -daP (20)

g r MP

Let dooMP be the solid angle which d<rP subtends at M so that

f mp doj^p ;= cos cp daP
This formula follows directly from the fact that, by definition, do>MP
is equal to the area of an element on a unit sphere centred on M
which is cut by the cone with apex at M and base daP (Fig. 7.3).
dcoMP is positive if the angle tp is acute, and negative if it is obtuse.

F ig . 7 .3
 Mathematical Physics
The double-layer potential can, therefore, also be written in the
form
w(M) = ^ i'(-P) d(oMP (21)
s
7.3.3 It follows from (21) that the potential due to a double layer
is defined also at the points M on the surface S. We therefore have
the following two properties.
Property 1 The potential due to a double layer is defined every­
where.

Property 2 At points M which do not lie on S, the potential due

to the double layer is a harmonic function.
To show this, we can use Equation (20). If M $ S, the integral
given by (20) is not an improper integral and, therefore,

Property 3 As the point of observation M tends to infinity, the

double-layer potential tends to zero. We are assuming that the
surface S has a finite area and lies in a finite region.
To prove this, let us use Equation (20). Applying the mean value
theorem to this integral, we obtain

w( M) = v ( P ) ^ - \ \ da,,
rMP \P=P* g

where P * e S. The validity of the property follows at once.

We shall assume henceforth that the surface S is closed. The
positive direction of the normal will be taken to be the direction
of the inward normal to S.
Consider the special case of the potential due to a double layer
when the dipole moment density is a constant, )'o. For this potential
we have the following results:
'47U'0, M inside S
zv(M) — 2crr0, M on S
0, M outside S
160
 Potentials
To prove this, let us use (21). Suppose that M lies inside S. Let
us suppose to begin with that a ray drawn from M cuts the surface
S at one point only. The integral ) dcoMP is then equal to the total
s
solid angle subtended by S at an internal point, i.e. 4 n. Consequently,
in this case, w(M) = 4nv0.
If some or all rays drawn from the point M intersect the surface
S i n a finite number ( < k) of points, the solid angles da>MP subtended
by the surface element da> defined by rays drawn from inside S
(Fig. 7.4) will be positive, whereas the solid angles dcoMP subtended

St

by the surface element do> defined by rays drawn from outside S

will be negative. Hence, the angle ep between the inward normal and
the direction of P M will be obtuse and, consequently, cos <p will
be negative. It is, therefore, evident that

( dcojvfp+ ^ dcoMP — 0
Si s2
The algebraic sum of all solid angles dcoMP will therefore also be
equal to 4n.
It follows that, in this case again, w(M) = 4nv0. If the point M
lies outside S, the solid angle dcoMP corresponding to elements d(TP
on .S', (Fig. 7.5) will be negative, whereas the solid angles da)MP
corresponding to elements do> on S2 will be positive. Therefore,

^ d(oMP — J d<x>MP + J dwMP = 0

S Si s2

It follows that if the point M lies outside S, then w{M) = 0. Simi­

larly, it can be shown that w(M) = 2nv0 if M e S.16
li 161
 Mathematical Physics
We can now understand the properties of the potential due to
a double layer in the neighbourhood of the point M lying on the
surface on which the double layer is situated.

Property 4 If the dipole moment density v{P) is continuous on S,

the potential of the double layer, w(M), will have a discontinuity
of the first kind at points on S and the discontinuity is
(M0) - w int(M0) = 4 jiv(M0) , M 0 e S
where wcxl(M0) is the limit of w(M) at Mo, when M tends to M 0
from outside the surface and wiai (M0) is the limit of w(M) at M0
when M tends to M 0 from inside.
We shall suppose, for simplicity, that each ray drawn from the
point M intersects S not more than k times (although this statement
is not valid for an arbitrary Lyapunov surface). Let M0 be a fixed
point on S and consider the auxiliary function

w(M) = ^ {i'(P)—i'(M0)} dcoMP - w ( M ) —w(M) (22)

s
Lemma The function w{M) is continuous at M 0.

Proof Let S' represent the part of S which lies in a ^-neighbour­

hood D s M
o of M 0, and let S ” represent the remaining part of S. We
may then write
w(M) =
where
w f M ) = J (i'(P)—v(M0)} dwMp
S'

w 2{ M ) = ^ {r(P )—v (M q)} dioMP

S "1
2
6

162
 Potentials
The function w2(M) is continuous at M0. Therefore, for arbitrary
e > 0 the quantity \w2{ M )~ w 2(M0)\ will be less than e/3 if M M 0
is sufficiently small. Next,

K (M )| = |5 {v(P)—v(M0)} dtoMP < 5 |v(P)—v(M0)| IdcOAfpl

S' S'

Since v(P) is continuous at M0, the quantity \v(P)—v(M^)\ is less

than e/ 12 k n if d (i.e. the radius of the neighbourhood DsMo of M 0)
is sufficiently small.
Next,
5 |deojtfp| < Ank
S’

and, consequently,

< J \ v ( P ) - v ( M 0)\ \da>MF\ < y

S'

and

|«7,(Af0)l < y
Therefore,
\ w ( M ) - w { M 0)\ < \ w l{M)\ + \wl{MQ)\Ar \w2{M)—w2{MQ)\ < e

if the point M is sufficiently close to M0. This proves the lemma.

Proof o f Property 4 Let us pass to the limit in Equation (22) by

allowing M to approach M0 from outside and inside S. This yields

^ext(Afo) = wext(M0)-A7iv(M 0) = w(M 0)

= w ( M Q) - 2 n v ( M Q) = wlat(M0) = wint (M0)—0

where w(M0) and w(M0) are the values of w{M) and w(M) at M0
on S. From these equations we have

« > e x t(M o ) = W {Mq)+2jiv (M o) (23)

W i n t ( M 0) = w(M 0) —2 n v (Mq) (24)

Wext(M0) - w iDt( M 0) = A n v { M 0) (25)

163
 Mathematical Physics

For the two-dimensional case it can be shown, in a similar way,

that at points M 0 on the carrier curve C we have
w„t(M 0) = w (M0)+ nv (M0) (26)
Wi„t(M0) = a; (M 0) - jiv (M0) (27)
wn t (M0) —wlnt(M0) = 2 n v(M 0) (28)

7.4 APPLICATION OF POTENTIALS TO THE SOLUTION

OF BOUNDARY-VALUE PROBLEMS

The above properties of potentials can be used for the solution of

boundary-value problems. We shall illustrate this by considering
the first internal boundary-value problem
V2w = f ( M ) at D (29)
u\s = <p(M) (30)
where u(M) is continuous in D = D+ S .
A special solution of (29) is the volume potential (in view of
Property 4 in Section 7.1)
1 C /(F )
Wl( M ) = - drP
4?r J rmp

It is therefore natural to seek the solution of (29)-(30) in the form

of the sum
u(M) — ul{M)JrU1{M)
where the boundary-value problem for the function ui{M) will be
formulated as follows:

V2»2 = 0 (31)
u2\s = <p(M)-u1(M)\s = F(M) (32)
We shall seek the solution of this problem in the form of a double-
layer potential

w{M )= U p f - ^ d o p
i rMp
with suitably chosen v(P). For any v(P) this potential is harmonic
in D (in view of Property 2, Section 7.3). To satisfy the boundary
condition (32), it is necessary that the relationship weKt(M) = F(M) 164
164
 Potentials

should be satisfied at points M e S. Using Equation (23) of Section

7.3, this condition can be written in the form
w(M)+2nv(M) - F(M)
or

\ K P ) 4 ~ (— I daP+2nv(Af) = F(M) (33)

e on yrMpj

The solution of the boundary-value problem (31)-(32) will be the

double-layer potential with dipole moment density v(P) satisfying
the condition given by (33).
It follows that the boundary-value problem under consideration
reduces to the solution of the integral equation (33) for v(P).

Example Let us solve the first boundary-value problem for a cir­

cular region of radius R bounded by the circle C:
V2« = 0, w|c = F(.s)
where 5 is the length of arc on the circle.
At points M on the circle (Fig. 21)
cos tp 1
(34)
pm p 2R
We shall seek the solution in the form of the double-layer potential

u d£
c
Using (34), the boundary condition leads to an integral equation
for v(s)\
\ ^ v ( S ) 6 ^ n v { s ) = F(s) (35)
c
We shall seek the solution of this equation in the form

v(s) = — F ( s) ± A
71

where A is an unknown constant. Substituting this function into

(35) we obtain165

A d£+F(s)+JiA = F(s)

165
 Mathematical Physics

and hence

JL $ f« )d f+ 2 ^ = 0 and
c c
It follows that

Consequently, the solution of the boundary-value problem is of

the form

u(M) =

c c c

F($) df
71

or
1 f 2rRcos(p—r:
F(i) di
“ = L71 .)\2 Rr2

166
 Potentials

From the triangle OPM (Fig. 7.7) we find {OM = p):

2 Rrcoscp—r2 R2—p2
2R? = 2R[R2+ p 2- 2 R p c o s ( 0 - i p )]
and therefore
2- (/?2- p 2)F(/?fl)dfl__
u(M) = — \ R2^rp2—2 Rp cos (0—ip)
L 71 J
0

We have thus obtained Poisson’s integral.

The solution of the second boundary-value problem for the
Laplace equation can be sought in the form of a potential due to
a simple layer with unknown density p(P).

PROBLEMS

1. Find the volume potential due to the mass distribution of

density p(r) in the spherical layer Rx < r < R2. Consider also the
case p(r) = p0 = const.
2. Find the potential due to a spherical simple layer of constant
surface charge density po.
3. Find the electrostatic field due to charges distributed uniformly
inside a sphere placed above a perfectly conducting plane z = 0.
4. Find the potential due to a circle of constant line density
of charge.
5. Find the potential due to a simple layer in the form of a seg­
ment of constant charge density.
6. Find the potential due to a double layer in the form of a seg­
ment of constant dipole-moment density.
7. Use the double-layer potential to solve the first boundary-value
problem for the Laplace equation (a) outside the circle and (b) on
a half-plane.
 8

Integral Equations

We have seen (Section 7.4) that the first boundary-value problem

for elliptic equations can be reduced to a linear integral equation.
In this chapter we shall discuss some of the basic properties of linear
integral equations of the second kind. For the sake of simpli­
city, we shall consider only the one-dimensional case (except for
Section 8.1). All the results will also be valid for the multi­
dimensional case.

8.1 CLASSIFICATION OF LINEAR INTEGRAL EQUATIONS

Equations of the form

<p(M)-X \ K ( M , P)<p(P) drP = f ( M )

D
where <p(P)is the required function,/(M ) and K(M, P) are unknown
functions, D is a fixed domain and X is a numerical parameter, are
called Fredholm’s integral equations of the second kind. When
f ( M ) = 0, the equation is called homogeneous; in all other cases
it is referred to as inhomogeneous.
Equations of the form

<p(M)—X J K ( M yP M P ) d r P = f ( M )
D (M )

where D(M) is a variable domain which depends on the point M,

168
 I n te g r a l E q u a tio n s

are called Volterra integral equations of the second kind. For

example, in the one-dimensional case

<p(x)-). J K(x, s)<p(s) ds = f ( x )

When f ( M ) = 0, the equation is called homogeneous; otherwise

it is referred to as inhomogeneous. Equations of the form
]K(M,P)<p(P)dzP = f ( M )

where D is a fixed domain, are called Fredholm’s integral equations

of the first kind.
Equations of the form
$ K (M ,P M P )drP = f ( M )
D(M)
are called Volterra integral equations of the first kind. The function
K(M, P) is called the kernel of the integral equation.

Remark Volterra equations are special cases of Fredholm’s

equations. Thus, if in the one-dimensional case we let
JO, x* < s < b
Ki(x,s)
\K(x,s), a< S < X
the Volterra equation .
X

<p(x)—/. J K(x, s)tp(s) d^ = f( x )

a

can be rewritten as the Fredholm equation with a kernel K^(x, s):

b

<p(x)—?. \ Ki(x, s)<p(s) d^ = f ( x )

a
The kernels K\(x, s) are called Volterra kernels.

8.2 PROBLEMS LEADING TO INTEGRAL EQUATIONS

8.2.1 Cauchy’s problem for a linear ordinary differential equation

of order n can be reduced to a Volterra integral equation of the
second kind. To be specific, consider the second-order equation
y ”+a(x)y' + b{x)y = f ( x ) ( 1)
f(0) - To, / ( 0 ) = To (2)
169
 Mathematical Physics

Let
/ '( * ) = <P(x) (3)
We then have
X

y'(x) = jo + S <p(s) ds (4)

0
x

^ W = J’o + S / © df (5)
o
Substituting (4) into (5) we obtain
X |

>’(*) = Jo^-^J,o+ $ $ rp(s) df

00

If we rearrange the order of integration in this expression we obtain

X

y ( X) = + J ( X - S ) e p ( s ) dS (6)
6
We have thus expressed the function y(x) and its derivatives
y'(x) and y"(X) in terms of the function tp(x) defined by (3), (4) and
(6). Substituting these values into (1) and taking the functions a(x)
and b(x) under the integral sign, we obtain the following Volterra
integral equation of the second kind:

<p(x)+ \ {«W + 0 —5)60)} <p(s) dJ = /lO ) 0)

6
with the kernel K(x, s) = a(x) -lr (x—s)b(.\), where
/iO ) = f ( x ) —yoa(x)—y0b(x)—xyob(x)
The procedure for reducing the Cauchy problem to an integral
equation for the equations of the n-th order is quite similar.

8.2.2 We shall now show that the Sturm-Liouville problem on a

finite segment can be reduced to a Fredholm integral equation of
the second kind. To do this, we shall introduce Green’s function for
the boundary-value problem and investigate its simplest properties.
Definition Green’s function G(x, s) for the boundary-value pro­
blem

£M = = -/(.v )

a i/(0 )-/3 ij(0 ) = 0, a2/(Q+02^(O = 0

170
 Integral Equations
is the solution of the boundary-value problem
L \ y \ = —d(x—s)
aiy (0) ^1j ( 0 ) = o, a2y (0~bPzyiO = o
which is continuous in the segment [0, /]. We shall now summarise
some of the properties of Green’s function G(x, s).
1. Green’s function is symmetric, i.e.
G(x, 5) — G(s, x)
Proof Let us apply Green’s formula to the one-dimensional
case (Section 6.1) to the functions v = Gi = G(x, sx) and u = G2
= G(x, s2). We have
/
I

0 0
( 8)
Using the properties of the <5 function, the integral on the left-
hand side of Equation (8) is equal to G(s1} s2)—G(s2, ^i), whereas
the right-hand side is zero. For the first and second boundary-value
problems this follows directly from the fact that the functions
Gi and Gz, or the derivatives dGJdx and dGJdx, vanish at the
ends of the segment (x = 0 and x = /). For the third boundary-
value problem we can express the derivatives dGi/dx and dG2/dx
at the end of the segment in terms of Gl and G2:
dGi F1n(r. „ dG2
= h— C (0 , j ,). — —■(7(0, s2)
dx x= 0 x=0 0=1

dGi dGi 2
= -~Pl
^ G(l, s^ , G(l, s2)
dx x=I &-2 dx x= l 0=2

and substitute these values into the right-hand side of (8). We

obtain

- ^ G(l, Si)G(l, s2)+ ^ G(l, s2)G(l, s{)

a2 ct2

+ ^ G(0, ^ (7 (0 , s2) - ^ G(0, s 2) G ( 0, Sl) = 0

oq «i
It follows that, in fact,
G(s2,Si) = G(st , s 2)
171
 Mathematical Physics

2. The partial derivative of Green’s function Gx(x, s) has

a discontinuity of the first kind at x = s and the magnitude of
this discontinuity is

<7*0 + 0, S ) - G X( S - 0, S) = — -yjrr (9)

k(s)
To prove this, let us integrate the identity
L[G\ = — d(x—s)
with respect to x between s —e and s+ e, where e > 0. We obtain
i+e
5 L...........
[ C ] dx = ) Gx(x ,s)]
tjj | | d[ A '( a -............. — q (x) G(x,
........................... - s) j dx1 ,

or
j-f e
k(x) Gx(x, s) ^ q(x) G(x, s) dx = —1

On passing to the limit in this equation as e -» 0, we obtain

<7*0+0, s)—Gx(s—0, s)
k(s)

since lim J q (x) G (x, s) dx = 0.

£->0 S— 8

8.2.3 Theorem The Green’s function is unique.

To begin with, let us establish the following two lemmas.

Lemma 1 There exists a solution jyO') of the equation L[y] = 0,

which satisfies the boundary condition aiT'CO)—/3,^(0) = 0.

Proof It is known that Cauchy’s problem for the equation

L\y] = 0 with any initial values ^(0) = y0, / ( 0 ) = y'0 has a solu­
tion. It follows, in particular, that there exists a solution with
initial values _y(0) and j/(0) which are related by ai.y'(0)—/?i v(0) = 0.
This proves the lemma.

Lemma 2 Any two solutions ^ (x ) and J’1(x) of the equation

L\y\ = 0 which satisfy the same boundary condition differ from
each other only by a constant factor, i.e. j>t(x) = C jj^x).
172
 Integral Equations
Proof The functions j',(x) and ji(x ) are solutions of the second'
order linear equation L[y] = 0 and satisfy the conditions
“ i T i' ( O ) — A ttC O ) = 0

w M -P o m = 0 ( }
These relationships can be regarded as a system of equations
for a, and Since at least one of the numbers and /?, is not
zero, the determinant of the system (10) is zero:
Vi'(0) y i (0)
= 0
H(0) J i (0)
This determinant is the value of the Wronskian at x = 0 for the
solutions j,(x ) and >q(x). It is known that the Wronskian made
up of the solutions of a given linear homogeneous equation either
is identically zero, or is everywhere non-zero. Since in our case
w(0) = 0, the Wronskian for _y,(x) and y,(x) is identically zero.
Hence, it follows that jj(x) = Cy^x).

We shall now prove the above theorem. We shall suppose

that /. = 0 is not an eigenvalue of the boundary-value problem
L\y\+?.y = 0
« ,/( 0 ) - A X 0 ) = 0, a2/(/)+ /S 2X 0 = 0 ( a)
Let j,(x ) be the solution of the equation L[y] = 0 which satisfies
the boundary condition a^^O) —fjy(Q) = 0. This solution exists
in view of Lemma 1. Any other solution satisfying the same
boundary condition will be of the form Q j^ x ), in accordance
with Lemma 2. All this shows that there exists the solution y 2(x)
of the equation L[y] = 0 which satisfies the boundary condition
a2y(0+/®2.y(0 = 0. Any solution of L\y] = 0 which satisfies
the same boundary condition must be of the form C^Cx).
The functions y r(x) and y 2(x) are linearly independent. If this
were not so, we would have _y2(x) = C y y{x). However, the function
y2(x) would then be a solution of L[y] = 0, satisfying both boundary
conditions. Consequently, / = 0 would be an eigenvalue of the
boundary-value problem (10a), which contradicts the original
assumption.
By choosing Cx and C2 in a suitable way, we can construct the
Green’s function from the functions C jj^x ) and C2y 2{x). Let

j C iyi(x), x s
G(x, s) on
{C2y 2(x), x > s
173
 Mathematical Physics

Since Green’s function is continuous at x = s, we have

CiJhO) = C2y2(s)
Hence,
Ci C2 ^
M *) ^ i (j )
Consequently, = Cy2(s), C2 = Cy^s). The coefficient C can
be determined from (9) which must be satisfied by Green’s function:

C lhW hW -hW j'i'C i)] = - (12)

The expression in the square brackets is the Wronskian of the

solutions Ji(x) and y 2(x) and is equal to D/k(s) (D = const). Since
the functions Ji(x) and y 2(x) are defined to within constant factors,
they can be chosen so that the Wronskian of the solutions >’1(5)
and j^C?) is equal to —1jk(s), i.e. we can suppose that D — — 1.
The relationship given by (12) then assumes the form

C 1
k{s) k{s)
and hence C = 1. Green’s function is therefore of the form
* < s
(13)
{X’ S ) ~ \ y i(s)y2(x), x> s
From (9) and (13) it follows directly that

Gx(x, x —0 ) ~ G x(x, x + 0 ) = — (14)

k(x)

8.2.4 We shall now prove two theorems due to Hilbert.

Hilbert's first theorem For any integrable function fix'), the

solution y(.v) of the boundary-value problem
Tb] = -/(* ) (15)
a i / ( 0 ) - / 5 i ^ ( 0 ) = 0, a 2/(/)+ & > v (/) = 0 (16)
can be represented by the formula
1

y{x) = \ G { x , t i ) m < V i (17)

0174
174
 Integral Equations
Proof Let us apply Green’s formula for the one-dimensional
case (Section 6.1) to the functions u = y(x) and v = G(x, s).
We obtain
t
J {G(x, s)L[y]—y(x)L[G]} dx = k(x)[G(x, s)y'(x)
o

or
( <
— J G(x, s)f(x) d x + ^ ( x ) d ( x —s) dx = &(x)[(7(x, £)./(■*)
o o
—y(x)Gx(x, s)]£
From the boundary conditions (16) for y (x) and G(x, s) it follows
that the left-hand side of this equation is zero. Consequently,
i
J G(x, s) f ( x ) dx = y(s)
o
Changing the symbol used for the integration variable and using
the symmetry of Green’s function, we obtain the formula stated
in the theorem.

Hilbert's second theorem For any integrable function /(x ), the

function
/
;K*) = $G(x, £)/(£) df (18)
0
is a solution of the boundary-value problem (15)-(16).

Proof It is clear that the function y(x) is continuous in the range

[0, /] and
1
y'(x) = l G x( x , m t ) df (19)
0
It follows that

« i^ '(0 )-^ y (0 ) = J {a,Gx(0, |) - f r G ( 0 , !)}/(£) d£ = 0

o
175
 Mathematical Physics

since, by definition of Green's function, the integrand is identically

zero. Similarly,

w 'iO + P iyV ) = o (20)

It follows that the function .y(x) satisfies the boundary con­
ditions (16).
Let us evaluate L[y\. We have
i i
L[y) = \ L [ G ] m df = - $ <5(* -£ )/(£ ) df = - f i x )
6 o

It follows that _y(x) satisfies (15) and this proves the theorem.

8.2.5 Let us apply Hilbert’s theorem to the Sturm-Liouville

problem
L[y]+?.p(x)y = 0 (21)

= 0, L2/ ( l ) + p 2y(l) = 0
0 (22)

From Hilbert’s first theorem the solution of this problem is given by

i
y(x) = Ajj G(x, s)p(s)y(s) ds (23)
o

i.e. the required solution is a Fredholm integral equation. Con­

versely, by Hilbert’s second theorem, (23) is a solution of the
Sturm-Liouville problem (21)-(22). It follows that the Sturm-
Liouville problem is equivalent to the integral equation (23).
We have considered the case when 2 = 0 was not an eigenvalue
of the boundary-value problem (10a). When 2 = 0 is an eigenvalue
of (10a), Green’s function F(x, 5) for the boundary-value problem

L[y] = —f i x )

« i/( 0 ) —0i.y(O) = °> w 'i O + P i y i i ) = 0

will be defined as the solution of the boundary-value problem

L[y] = —?>{x—s)+<P0(x)<P0(s)
« ./(0 )-/? ,j(0 ) = 0, «2/ ( / ) ] - p 2y ( l ) = 0
176
 Integral Equations
which is continuous in [0, /] and is orthogonal to the eigenfunction
<P0(x) corresponding to the eigenvalue I = 0, i.e. such that
i
^ <P0(x)G(x, s) dx = 0
o
The corresponding properties of Green’s function, in this case,
can be established in a similar way.

8.3 INTEGRAL EQUATIONS WITH DEGENERATE KERNELS

A kernel K(x, s) is called degenerate if it is of the form

K{x, s) = ^ a,(x)bi(s) (24)

i=i
where a,(x) are linearly independent functions.
The solution of Fredholm’s equation of the second kind with
a degenerate kernel can be reduced to the solution of a system of
linear algebraic equations.
In point of fact, substituting (24) into
b
(p(x) = /. J K(x, s)rp(s) ds+ f(x) (25)

we obtain

rP(x ) = z ^ C,ai(x)+f(x) (26)

(=i
b
where C; = $6i(j)9?(j)d.s' are unknown numbers.
a
It follows that the solution of (25) with a degenerate kernel
must be sought in the form given by (26). Substituting this function
into (25) and comparing the coefficients of the same functions
we obtain the following system of linear algebraic equations
n
C, = 7. ^ C jU ij+ p i (/ = 1,2, ...,«) 17 (27)

12 177
 Mathematical Physics

where
b b
ctij = J ai(s)bj(s) dj, pi = dj
a a

By solving this system we can find Cj and, consequently, the

solution of (25).

8.4 EXISTENCE OF SOLUTIONS

8.4.1 If the kernel is degenerate, then the existence of the solution

of the Fredholm integral equation reduces to the question of the
existence of the solution of the corresponding system of algebraic
equations (27). We shall prove the existence of the solution of
(25) (for sufficiently small values of |2|) by the method of successive
approximations in a more general case.
We shall assume, for simplicity, that ' ( l ) the kernel K(x, s)
is continuous for a < x, s < b, in which case it is bounded by
a certain constant A, |AT| < A, and (2) the function f(x) is continuous
in the range [a, b] so that it is bounded in this range by a certain
constant B, \f\ < B. Consider the sequence of functions
9h00> —,<Pn(x), ...
such that
b
<Fi(x) = /(* )+ '• SK(x, s) <p0(s) ds (28)
a

where ^0(^) is an arbitrary fixed continuous function

b
<Piix) = m + X 5 K(x, s)<Pi(s) ds (29)
a

b
<Pn(x) = SK(X, j)9Jn_i(j) ds (30)
a

Theorem The sequence of functions given by (28)-(30) converges

uniformly in the range [a, b] to the function <p(x) which is a solution
of (25).

Proof Substituting y f x ) into the formula for tp2(x) we obtain

b b b
<p2(x) = f ( x ) + X jj K(x, s ) f( s) ds+X2 5 K ( x , s) J K(s, t)(p0(l) dt ds
a a a 871

178
 Integral Equations
If we change the order of integration in the last integral, we obtain
b b
<Pi{x) = / ( * ) + '- $ Ki(x, s ) f ( s ) dS + / } 5 K2(x , t)<p0(t) d t

where

K i( x , s) = K(.x, s), K 2(x , t ) = J Ky(x, s)Kl(s, t ) d s

a

Similarly, we find that

b b
<Pn{x) = f ( x ) + / . J Ky(x, s )f (s ) d^+z.2 J K 2(x, s)f(s) di-+ ...
a a

b b
S s ) f ( s ) d s + A n J K n(x, 0 ^ , ( 0 d t

where K„(x, t) = \ Ky(x, s)Kn_y(s, t)ds. The limit of the function

b
(p„{x), if it exists, is equal to the sum of the series
b b
V(x) = f ( x ) + /. J Ky(x, s)f(s) d.v+ J Kn(x, s)f(s) d^+ ... (31)
a a

We shall now prove the uniform convergence of this series. Consider

the integral

\ k „(x , s ) f ( s ) d^

It is evident that

\K2(x , s ) | < J \Kx(x, s)Ky(s, t)l d^ < A \ b ~ a )

a

b
[A"3(jc, j)| < jj s )K2(s , f)| d j < A 3(b—a)2
a

b
\Kn(x, j)| < S \Kx{x, s)K„_y(s, 0! ds < A \ b - a y - 179

12* 179
 Mathematical Physics

and, therefore,
b b
I $ Kn(x, s)f(s) ds | < An(b—d)n~1$ 1/0)1 ds < AnB { b - a ) n
a a

Consequently, the numerical series

00

^ BAn\A\n(b—a)n (32)
n= 1

is the majorant series for (31). If |A| < 1/A(b—a), then the series
(32) will converge. Consequently, for such A the series (31), and
together with it the sequence of functions y„(x), will converge
uniformly to the function <jp(x). This function is a solution of (25).
In the limit, as n -> oo in (30), we obtain
b
q>(x) = X^Kix, s)(p(s) ds —/ (x)

The transition to the limit under the integral sign is legitimate

since the sequence converges uniformly. We note that the limit
lim <pn(x) = <p(x) is independent of the choice of <p0(s) (zero-order
oo

approximation). The uniqueness of the solution of (25) readily

follows from this. In point of fact, if there exists a further solution
y>(x) of (25), then, assuming <p0(x) = y>(x) in the procedure for
constructing the functions (28)-(30), we obtain
9?,(a') = tp(x), <p2 {x) = ip{x), ..., cpn(x) = y(.v), ...
This sequence has the function ^(.x) as the limit. However, it is
also evident that

lim <p„{x) = ip{x)

n —> oo

and, therefore,

V(x) = V(x)

8.4.2 Since the series given by (31) converges for |A| < \/A(b—a),
00
it follows that for these values of A, the series ^ An\X\n~x{b — a)n~x
n= 1
will also converge. However, this series is the majorant series for180
180
 Integral Equations
00 00
2 A"-1ATn(x, j) so that ]>] Xn~lK„{x, s) converges uniformly. The
n = -1 n= l
series given by (31) can therefore be written in the form
b oo
v(x) = / 0 ) + ;- ‘y)}/(‘y) dy
a n= 1

V(x) = f{x)+X ^ R(x, s, ;.)/(s) d j (33)

00
where the function R(x, s, X) = /,-n_1 Kn(x, s) is called the resol-
71 = 1

vent of (25).
It follows that if we know the resolvent of (25), we can use (33)
to obtain its solution. We have defined the resolvent only for small
values of \X\. However, it can also be defined for any finite domain
in the complex plane of the variable X by analytical continuation
(except, perhaps, for a finite number of singular points in this
domain). If this has been done, then (33) yields the solution of
(25) for any X except for these singular points. We shall not pause
to consider this in any detail.

8.4.3 If we apply the above procedure to the Volterra equation

X

<p(x) = /. 5 K(x, s)rp(s) d s + f i x ) (a < x < b) (34)

we obtain the following sequence of functions

X

<P\(x) = / ( * ) + ' ’■$ K(x, s)<p0(s) dj

a
x

<Pi(x) f i x )+ , s) 'pi(s) d.y

<Pn(x) = /(* ) + * SK { X , S ) tp n^ i s ) dj

181
 Mathematical Physics
The sequence converges uniformly in the range [a, b] for any
values of In point of fact, it is clear that

M-v)l < |/ W | + |A |5 |^ ,5 )||< p 0(5)|d5

a
< B + ^ A B v i x —a), where |<jp0(s)[ < B0
X

l?2(X)| < 1/0)1+ m 5 \K(x, /)!/,(/)! dv

X

< B+\).\A 5 { B + \ M B 0(s-a )} dj

a

In general,

\<Pn(x)\ < 5 + |/ . M 5 ( . x - fl) + . . . + r V ' ,- 15 (- ^ -

0 —1)!
O -a y
+ 1V ^0 n\
00

Since the series B lA\"An(x — a)n/nl converges uniformly in [a, b],

n=1
and its partial sums are the majorants for the functions <p„0)> it
follows that the sequence {<p„0)} will also converge uniformly;
<jp(x) = lim <pn{x) is, clearly, the solution of (34) and, in fact, the
00

unique solution.

8.5 APPROXIMATE METHODS OF SOLUTION

OF FREDHOLM’S INTEGRAL EQUATIONS OF THE SECOND
KIND

The method of successive approximations described in Section 8.4

can serve as an approximate method for the solution of integral
equations. As the approximate solution we must take the functions
<pn{x) given by (28)-(30).
A second method for solving integral equations involves the
approximation of the kernel K(x, .?) by the degenerate kernel
n

K ( X , S) = ^ tf ;( .v ) 6 ;( /)
f= l
182
 I n te g r a l E q u a tio n s

The solution of the equation with the kernel K{x, s) will be the
approximate solution of the original equation.
A third method consists of the following. The ranges [a, b\
of the variables x and s are divided into a grid of n identical parts
b

by points x i}Sj, and the integral ) AT(x, T)<p(.s)ds in the integral

a
equation is replaced by an integral sum. As a result,
n

<p(x) ^
K (x >sj)Vi ^ S j + f i x )
j'=i
Assuming in this expression that x is equal to x {(i = 1, 2, ..., n),
let us consider the set of equations
n

<Pi = Ku<Pj V2sj+ fi ( j = 1, 2, ...,«) (35)

; =i
where
<Pi = < p (xd , K ;j = K ( X i, S j) , f i = /(* ;)> V 2Jy = Sj + 1 — S j

Solving this system for gj,-, we obtain the values of the approxi­
mate solution at the grid points. We shall not pause to consider
these approximate methods in any detail and refer the reader to
specialist literature.

8 .6 F R E D H O L M ’S THEOREM

In this section and in the next chapter we shall confine our attention
to Fredholm’s integral equations of the second kind
b

<p(x)—X J K{x, s)fp(s) ds = f ( x ) (36)

a

8.6.1 The homogeneous equation

b

<p(x) = 2 J K(x, s)<p(s) ds (37)

a

has the trivial solution g?(x) = 0 for any value of the parameter
However, for some values of 2, it may also have non-trivial solutions.

Definition The values of 2, for which Equation (37) has non-trivial

solutions (i.e. solutions which are not identically zero), are called183
183
 Mathematical Physics
the eigenvalues of (37) [the kernel K(x, .s)] and the corresponding
solutions <p(;c) are called the eigenfunctions of the equation (kernel).
We now have the following theorem.

Theorem 1 If in Equation (36) the parameter X is not equal to

the eigenvalue of the corresponding homogeneous Equation (37),
then (36) can have only the unique solution. The existence of the
solution will be established later.

Proof Let y f x ) and cp2(x) be two solutions of (36). We then

have
b
<Px{x)—X 5 K(x, s)<pfs) ds = f ( x )
a
b
tp2(x)—X 5 K(x, s)tp2(s) ds = f { x )

and, hence,
b
(<Pi—<P2) —?-<
\ K(x,s)((pl—<p2)ds = 0
a

Consequently, the difference tp(x) = <pi(x)—<p2(x) is a solution of

the homogeneous equation. Since X is not an eigenvalue, it follows
that <p(x) = <Pi(x)—<Pi(x) = 0- This proves the theorem.

8.6.2 It will now be convenient to recall some theorems on

systems of linear algebraic equations.

Theorem A A necessary and sufficient condition for the homo­

geneous system of equations
n
= 0 (/ = 1, 2, ..., n) (38)
>= 1
to have only a trivial solution, i.e. a solution consisting only of
zeros, is that the determinant of the system should be different
from zero.

Theorem B If the determinant of the homogeneous system (38)

is equal to zero, the system has p = n —r linearly independent
solutions where r is the rank of the matrix of the system.
184
 Integral Equations
Theorem C If the homogeneous system of equations (38) has
only a trivial solution, then the corresponding non-homogeneous
system
X au xJ = bi O' = 1, 2, ..., n) (39)
j-1
has aunique solution for any values of the right-hand side, b r

8.6.3 It was shown in Section 8.5 that the approximate solution

of (36) can be obtained by replacing this equation with the cor­
responding system of linear algebraic equations
n
<Fm—/• X KJm(PJ =/■• (m = 1, 2, ..., n) (40)
; =i
and then solving this system.
Similarly, certain theorems for systems of linear algebraic
equations can be extended to Fredholm’s integral equations of
the second kind. For integral equations these theorems are called
Fredholm’s theorems. We shall indicate below one of the methods
of obtaining Fredholm’s theorems without giving detailed proof.
The function <p„(x) which is equal to the solution of the system
(40) at the corresponding grid points and is linear between them
will be called a polygonal function corresponding to the solution
of the system (40). We then have the following theorem.
Theorem 2 A polygonal function <pn(x) corresponding to the
solution of (40) tends uniformly to the solution of the integral
equation (36) as n -> oo.
We shall not give a proof of this theorem and will proceed to
the method of deducing Fredholm’s theorems.
Suppose that X is not an eigenvalue of the kernel K(x, s). The
homogeneous Equation (37) will then have only the trivial solution.
Therefore, in view of Theorems A and 2, we conclude that the
corresponding system of algebraic equations which replaces the
integral Equation (37), i.e. the system
n
Vm—X ^ K Jmq>jV2Sj = 0 (m = 1, 2, ...,«)
j =i
has a non-zero determinant. Consequently, the system of equations
a

rpm X K j m(fj V Sj f m (m = 1, 2, ...,«)

j =11
85
185
 Mathematical Physics
which replaces the inhomogeneous integral Equation (36) has
a unique solution. The polygonal function <pn(x) corresponding to
this solution for n -* oo will, in accordance with Theorem 2, tend
uniformly to the solution of (36). This means that the following
theorem is valid.

Fredholm's first theorem For any X which is not an eigenvalue,

Equation (36) has a unique solution.

Remark Since the determinants of the system (40) and of the

transposed system
n

( m = 1, 2, ..., n)

coincide, it follows that for any X which is not an eigenvalue of the

kernel K(x, s), the adjoint integral equation
b
yi{x)—X J K(s, x)ip(s) ds = f( x )
a

will also have a unique solution.

Let us consider now the case where X is equal to one of the
eigenvalues. We then have the following theorem.

Fredholm's second theorem If 2 is an eigenvalue of the kernel

K(x, s) then both the homogeneous integral Equation (37) and its
adjoint equation have a finite number of linearly independent
solutions.
This theorem follows from the fact that the homogeneous
system of algebraic equations corresponding to equation (37)
has a finite number of linearly independent solutions, in accordance
with Theorem B.

Fredholm's third theorem Let I be an eigenvalue of the kernel

K(x, s). In order that (36) have a solution, it is necessary and
sufficient that the function f(x) on the right-hand side of (36) be
orthogonal to all the eigenfunctions of the adjoint homogeneous
equation corresponding to this eigenvalue. The necessity of this
condition is readily proved. In fact, if <p(x) is a solution of (36),
then
b
cp(x)—x 5 K(x, s)(p(s) ds = f ( x )
a186
186
 I n te g r a l E q u a tio n s

Let us multiply this identity by the eigenfunction -ip(x) of the adjoint

equation and integrate the result (with respect to x) within the
range [a, b\. We obtain
b b b b

5 f ( x )w(x) dx = jj <p(x)y(x) dx—7. jj \p(x) jj K(x, s)<p(s) ds dx

a a a a
Since
b b b b

a jj i/i(.y) \ K(x, s)tp(s) ds dx = ^ <p(s)X ^ K{x, s)ip(x) dx dj

a a a a
and
b

7, K(x, s)ip(x) dx = ip(s)

a
it follows that
b b b

5 /(x)y>(x) dx = jj <p ( x ) v , ( x ) dx— jj (p(s) ii>(s) ds = 0

a a a

which was to be proved.

The sufficient condition is more difficult to establish. The proof
can be given, for example, first for the corresponding system of
algebraic equations; then, by going to the limit in the polygonal
functions, one can extend this result to integral equations also.
We shall not give this proof here.
Suppose that there are r linearly independent eigenfunctions
corresponding to a given eigenvalue We then have the fol­
lowing theorem.
Theorem If in (36) 7. is one of the eigenvalues and the solution
of (36) exists, i.e. /(x ) is orthogonal to the corresponding eigen­
functions of the adjoint equation, the solution of (36) will also
be any function
r

= Po(*)+ 2 C‘I(M X)
<7=1
where <p0(x) is a solution of (36), <pq(x) are eigenfunctions of the
kernel K(x, j) corresponding to the eigenvalue 7. and Cq are arbitrary
constants.
Remark In Section 8.4 it was shown that the inhomogeneous
Volterra equation has a unique solution for any values of the
parameter /. Consequently, in accordance with Fredholm’s theorems,
the Volterra equation has no eigenvalues.
 9

Integral Equations with Symmetric

Kernels

In this chapter we shall consider Fredholm’s equations with sym­

metric kernels. The kernel K(x, s) is called symmetric if, for all
x and 5, the following identity is satisfied in the square a < .y,
s < b:
K(x, s) — K(s, x)

If the kernel K{x, s) is symmetric, it is evident that all the iterated

kernels K„(x, s) are also symmetric. We recall that, for the sake of
simplicity, we are confining our attention only to kernels continuous
for a < .v, s < b.
Equations with symmetric kernels are frequently encountered
in mathematical physics. They have a number of specific properties,
the chief of which is given by the following theorem.
Theorem 1 Any continuous symmetric kernel, which is not
identically equal to zero, has at least one eigenvalue.
We shall not prove the theorem here and will merely note that
it is possible to find asymmetric kernels which do not possess
eigenvalues. For example, the kernel

K(x, s) = sin.v coss, 0 ^ A', s < 2 tc

and all Volterra kernels are examples of this.
The set of all eigenvalues of a kernel (equation) is called the
eigenvalue spectrum of the kernel (equation) or, simply, the spectrum
of the kernel (equation).18
188
 Integral Equations with Symmetric Kernels
9.1 SIMPLEST PROPERTIES OF EIGENFUNCTIONS
AND EIGENVALUES
It is evident that the following properties must hold.
Property 1 If <f(x) is an eigenfunction corresponding to an eigen­
value A, then C<p(x), where C is an arbitrary constant, is also an
eigenfunction corresponding to the same eigenvalue A.

The constant factor C can be chosen so that the norm of the

/b
eigenfunction C<p(x), i.e. ||C<p||="l/ $ C2<p2(x)dx, is equal to unity,
V a
so that ||CVI| = 1. We shall assume henceforth that all eigen­
functions are normalised to unity in this way.

Property 2 If two eigenfunctions cpi{x) and <pi{x) correspond to

the same eigenvalue A, then for all constants Q and C2 the functions
C\tp\{x)+Czcp2{x) will also be functions corresponding to the same
eigenvalue A.
Let us now establish the following property.
Property 3 The eigenfunctions <fi(x) and <p200 corresponding
to different eigenvalues At and A2 are orthogonal in the range
[a, b], i.e.
6

S<pi(x)vz(x) dx = 0
a

Proof By hypothesis we have

1 b J b
y<Pi 00 = jj K{x, s)<pi(s) ds, 7 ^ 200 = J K (x ’ s) n ( s ) ds

Let us multiply the first of these by <pi{x) and the second by <pi00
and subtract one from the other. If we now integrate the resulting
identity with respect to x within the range [a, b\, we obtain
1 j \b b b
K ------y H Pi 0 0 ^ 2 0 0 dx = jj jj K(x >s)<pi(s)<p2(x) d j d.v
' 1 Z' a a a

bb
' ~ \ \ K (x »J) P2(*) P i00 d s d.v
aa

189
 M a th e m a tic a l P h y s ic s

Changing the order of integration in the second term on the right,

and bearing in mind the fact that the kernel is symmetric, we have
b b bb

K(x, s)cp2(s)cpl(x) ds dx = 5 5 AT(x, s)rp{(x) <p2(s) dx ds

a a aa

b b

= ^ K(x, s)<p\(s)cp1{x) ds dx
a a

Consequently,
b

| ^ dx = 0

The orthogonality property follows at once.

If the eigenfunctions corresponding to a given eigenvalue /. are

made orthogonal, any two linearly independent eigenfunctions
^ (x ) and (p2(x) will be orthogonal.
We shall assume henceforth that the orthogonalisation procedure
has been performed whenever necessary. Consequently, a set of
eigenfunctions can be considered as orthonormal.

Property 4 All eigenvalues of integral equations with symmetric

kernels are real.

Proof Let us suppose that I — a+//?, /? 0 is a complex eigen­

value, and cp{x) = y>i(x)-\-i'tp2(x) is the corresponding eigenfunction.
We then have
b

Wi (x)+z>2(x) = (a+//J) aJ K(x, s)[ipfs) -\-iy>2(s)] ds

and hence
b b

a jj K(x, s)y>i(s) ds—fi jj K(x, s)ip2(s) ds

b b

a jj K(x, s)tp2(s) ds+ p $ K(x, . f y f s ) ds

a a 190

190
 Integral Equations with Symmetric Kernels
Let us multiply the second of these identities by i and subtract
the results from the first identity. We have
b
V>i(.y)-/> 2(.y) = (a— j)[vi(j)—/>2(j)]dJ
a
Therefore
/. = a —//? and <p(x) = (x)—iy2(x)
are, respectively, the eigenvalue and the corresponding eigenfunction.
Since / / /. (/? # 0), it follows from Property 3 that the functions
<p(x) and <p(.v) are orthogonal, i.e.
b b
5 < P ( x ) l p ( x ) dx = jj {y>?(x) Vwl{x)} = 0
a a

Hence, since the functions y f x ) and ip2(x) are continuous, it follows

that Wiix) = y 2(x) = 0- Hence rp(x) = 0, which is impossible.
This proves the property.
Property 5 In each finite range [A, B] there is only a finite number
(including zero) of eigenvalues.
Proof Let us suppose that in a given range [A0, i?0] there is an
infinite set of eigenvalues. Let us select out of this set a certain
infinite sequence of eigenvalues {!„}. Suppose that is the
sequence of corresponding eigenfunctions. Since the family of
functions {<p„(x)} are orthonormal, the Fourier coefficients of the
kernel K(x, s) for this family are (1 /l„)<pn(x). It follows that the
Bessel inequality
oo b

must hold. Consequently, for any integer p > 0

P b

Integrating this inequality with respect to x within the range [a, b\,
we obtain

1 " n a
191
 Mathematical Physics
Since all the Xn lie in the finite range [A0, B0], it follows that all the
numbers X\ are less than B2, X2 < B2, where B2 = max {A%, BQ.
p j
If we replace all the I 2 in the sum ^^'yY by the larger number B2,
n=l *n
we obtain for any integer p
p b b
^ ^ K 2(x, j) d j dx
n=1 a a

which is impossible since the series ^ is divergent and con-

n= 1
1
sequently, for sufficiently large p, the sum 2_, ~S2 wi^ '-)e greater
n = \
bb
than $ jj K 2(x, s)d.y dx.
a a

From Property 5 it follows directly that (1) all the eigenvalues

can be numbered in increasing order of their absolute magni­
tudes, i.e.
I^iI < 1^1 < - < \ K \ < -
and (2) if the eigenvalue spectrum is infinite, then \Xn\ -> oo as
n -» oo.

Property 6 To each eigenvalue X there corresponds a finite number

q of eigenfunctions ^ (x ), y2(x), ..., <pq(x).

Proof Let us suppose that to a given eigenvalue X there corre­

sponds an infinite sequence of eigenfunctions^1(x),^2(x ),... , 9?„(x),...
It follows from the Bessel inequality that for any integer p > 0
we have

Integrating this inequality with respect to x within the range [a, b],
and bearing in mind that the eigenfunctions are normalised, we
have for any integer p > 0
b b b l>
Jf jj jj ^ 2(x, s) d.v dx for p < X2 jj ^ K 2(x, s) ds dx
n = 1 a a a a

192
 Integral Equations with Symmetric Kernels
which is impossible. It follows that to each eigenvalue X there
corresponds only a finite number of eigenfunctions.
It follows from Section 8.3 that a degenerate symmetric kernel
can only have a finite spectrum. In fact, in order that a homogeneous
system of linear equations ((27) for /?, = (); Chapter 8) for the
coefficients Cf should have a non-zero solution, it is necessary
that the determinant of this system D(X) be zero. The eigenvalues
can be found from this equation. It is evident that it has only
a finite number of roots. The converse is also true; if a symmetric
kernel K(x, .?) has a finite spectrum, it is degenerate. In fact, suppose
that /j, X2, ..., is the spectrum of the kernel and <Mx), <p2(x),
... ,<pn(x) is the set of all the corresponding eigenfunctions of the
kernel (complete system). Consider the symmetric continuous
function
n

If it is not identically equal to zero, then by Theorem 1 it has at

least one eigenvalue ,« and the corresponding function is
b
y(x ) = ft K {n\ x , s)ip(s)ds
a

The function y>(x) is orthogonal to all the eigenfunctions (pq(x)

of the kernel K(x, s) since
b b b
$ V (*) fPn(-x) dx = (I 5 <pq (x) 5 K (n)(x, s)y>(s) ds dx
a a a

bb
= ,u $ $ K (n)(x, s)<pq(x)y(s)ds dx
a a

On changing the order of integration we obtain

U -- — v.

/I

13
193
 Mathematical Physics
Since the functions <pp(x) are orthonormal, the last integral
is equal to

when n and cp{x) are the eigenvalue and eigenfunction of the kernel
K(x, s), since
b b n

b
= /i J K M (x, = ip(x)
a

In deriving this expression we have used the orthogonality of

y(x) and <pp(x). Since y>(x) is the eigenfunction of the kernel K(x, s)
and the functions q>x( x \ <p2(x), ..., cpn{x) form a complete system of
eigenfunctions of the kernel K{x, s), it follows that y(x) should be
a linear combination of the functions cp\(x), <Pz{x), ..., <pn(x). This,
however, is impossible since ip(x) is orthogonal to all these functions.
We cannot, therefore, assume that K (n)( x , s ) ^ 0. Consequently,
A^(n)(x, s) = 0 or
n

i.e. the kernel K(x, s) is degenerate. This means that the following
theorem must be valid.

Theorem 2 In order that the spectrum of a symmetric kernel

be finite, it is necessary and sufficient that the kernel be degenerate.

9.2 THE SPECTRUM OF ITERATED KERNELS

Let us substitute
b
Acp = \ K (x, s)q>(s)ds
 Integral Equations with Symmetric Kernels
It follows from the definition of iterated kernels that
b
A (A<p) = A 2<p = ^ K2(x, s)cp (^)dj
a

and, in general,
b
A n<p = A(An~l<p) = \k„ (x, s)<p(s)ds
a

For eigenfunctions tpp(x) and eigenvalues Xp we have

<PP(x) XpA<pp = XPA (Xp A(fp)
b
= X2p A 2<pp = ... = Xnp A ncpp = Xnp 5 K„(x, s)tpp(s)ds
a

from which it follows that the following theorem must be valid.

Theorem 3 If cpp{x) and Xp are the eigenfunctions and eigenvalues

of the kernel K(x, s), respectively, then <pp(x) and Xnp will be the
eigenfunction and eigenvalue of the kernel K„(x, s).
The following theorem is also valid.

Theorem 4 If is an eigenvalue of the kernel K„(x, s), then at

least one of the (real) roots of the n-th power of ^ will be an eigen­
value of the kernel K(x, s).
To prove this result, it will be convenient to introduce the
following lemma.

Lemma If hu h2, ..., hn are the roots of the equation hn — /u,

then
^1+ ^ 2+ ■■•Jrhsn = 0
for s = 1,2, ..., n — 1.

Proof It is well known that hm = ] / fi£m where Y m is a root

of the equation hn = /u, £ = el2*/n. We then have

h \ + h \+ - f h l = V 7 ( \ Jr t s+ e s+ ... + r (n- 1>)^

£sn- l
rjV = o
T =T
since £sn = 1.
13* 195
 Mathematical Physics

Proof o f the theorem Let y(x) be an eigenfunction of the kernel

Kn(x, 5) corresponding to the eigenvalue /i. We shall determine the
functions 99,,(v) from the formula

<PP(x) = -^-(y-\-hpAy>+h2pA2ip f ...4 hn~'An~ \ ) (1)

Summing these equations over p between p = 1 and p — n, and

bearing in mind the above lemma, we obtain
n

v ( x ) = ^ ( p p( x )
P=1

It follows from this identity that among functions <pp(x) there is at

least one which is not identically equal to zero. It is readily seen
that <pp(x) = hpA(pp. In point of fact, if we apply the operator A
to the identity given by (1) and multiply the result by hp, we obtain

hpA(pp - hpAy>+h2p A2xp+ ...Jrhp~lAn-'y’) Jr — hpA ny>

or

hpA<pp <Pp(x ) - ~ - y ( x ) + ^ h npA ny>= <pp{x)

since hp — ,u and p,Any> = y>.

It follows that functions 9tp(x) which are not identically equal
to zero are the eigenfunctions of the kernel K(x, s) and hp are the
corresponding eigenvalues. From Property 4, the kernel K(x, s)
has only real eigenvalues. Consequently, the functions <pp(x) cor­
responding to complex roots hp are identically zero. When n is odd,
there is only one real root (V/i = hp which should, in fact, be the
eigenvalue of the kernel K(x, s) and <pp(x) = y>(x) is its eigenfunction.
If n is odd, there are two real roots. Suppose that ^ (x ) and <p2(x)
are the corresponding eigenfunctions. We then have

v(x) = rPi(x)+y2(x) (2)

It follows that when n is odd, each of the eigenfunctions of the

kernel K(x, s) will also be an eigenfunction of the kernel K(x, s). For
even n each eigenfunction of K(x, s) will either coincide with the
eigenfunction of the kernel K(x, s) (one of the functions ^(.v) or
(p2(x) in (2) may be identically zero) or will be a combination of
the eigenfunctions of K(x, s). This means that if {2p} and196
196
 Integral Equations with Symmetric Kernels
represent the sets of all the eigenvalues and all the eigenfunctions
of the kernel K(x, s), then {A"} and {9?p(;t)} are the sets of all the
eigenvalues and eigenfunctions of K„(x, s).

9.3 EXPANSION OF ITERATED KERNELS

In this section we shall show that for any n ^ 3, the following

expansion is valid

« .* * ) = o)
7=t p
in which the series converges absolutely and uniformly within the
range a < x, s < b.
We shall show, to begin with, that the series on the right of (3)
converges absolutely and uniformly for a < x, s < b. For this let
us estimate the finite series
m+q m+q

y , j j w W P{x)(pP{s)\ ri(s) 1 (4)

% J
We have used the inequality
\ A - B \ < ~ ( A 2+ B 2)

and the fact that \ap\ tends monotonically to zero as p -> oo. From
Bessel’s inequality
oo b

where D = const and D > 0. Therefore, for q > 0

z
m+q
Pj>(*) Pj.fr). D
(5)
p=m
K~2
Since |/ m| oo as m -> oo, it follows from (5) in accordance with
Cauchy’s criterion that the series (3) converge absolutely and uni­
formly.
Let
OO
= J] i PpfrOPpfr)
P- 1 'v

197
 Mathematical Physics
This function is continuous within the range a < s, s < b. We must
show that Kn(x, s) = <P(x, s). Let us suppose that this is not so.
The symmetric function
Q(x, s) = Kn(x, s)~&(x, s)
will by Theorem 1 have an eigenvalue fi and an eigenfunction
ip(x), i.e.
b
y>(x) = n \ Q { x , s)y{s)ds
a

The function y>(x) is orthogonal to all the eigenfunctions <pr(x)

of the kernel K(x, s) since
b bb
$ y>(x)<Pr(x) dx = P jj jj Q(x, s)y>(s)(pr(x) ds dx
a aa

and because <pr(s) — A" $ K„(x, s)cpr(x)dx. The function y>(x) is an

a
eigenfunction of the kernel K„{x, s) since
b
xP{x) = fi]Q(x,s)y>(s)ds
a

b
= [i
a 1 p= 1 P )

= 11\ k „(x,s)y>(s)ds

198
 Integral Equations with Symmetric Kernels
We have used the orthogonality of the function ip{x) to all the eigen­
functions (pp{x). Consequently, as was shown in Section 9.2, ip(x)
should be a linear combination of the functions cpp{x). However,
this is impossible since y>(x) is orthogonal to all the functions <pp{x).
Therefore, it cannot be assumed that Q(x, 5) 0 0.

Remark The expansion given by (3) is valid also for Kz(x, j)

(n — 2) and, under certain additional conditions, for K{x, s). We
shall not prove this here.

9.4 HILBERT-SCHMIDT THEOREM

We shall now prove one of the fundamental theorems of the theory

of linear integral equations which has many applications. This is
the expansion theorem.

Hilbert-Schmidt theorem If the function rp(x) can be represented

by
b
f ( x ) = 5a'(jc,j)/i(j) ds ( 6)
a

where h(s) is piecewise-continuous in [a, b], it can be represented

by an expansion in terms of the eigenfunctions of the kernel K(x, j),
i.e.

/(* ) = X fpVpW
P= 1
(7)

where
b
fp =
a
This series converges absolutely and uniformly in the range [a, b\.
To prove this theorem the following lemma will be convenient.

Lemma In order that a continuous function Q(x) be orthogonal

to the kernel K(x, s), i.e.
b
$A:(x, j ) 0 ( j ) d j = 0 (8)
a91

199
 Mathematical Physics
it is necessary and sufficient that it should be orthogonal to each
eigenfunction of the kernel, i.e.
b
S0(*)Pp(.^d* = O ( p = 1, 2, . . . ) (9)
a

Proof Sufficiency:
b bb
SQ(x)<pp(x) dx = /.p 5 5 K(x, s)<rP(s)Q(x) d.<r d.v
a aa
b b
= Ip 5 <pp(s) { 5 K(x, s)Q(x) dA-} ds = 0
a a

since the inner integral is zero.

Necessity : consider the auxiliary integral
bb
J\ = J \ K 4( x , s ) Q ( x ) Q ( s ) ds dx
a a

It is zero since, using Equation (3) for n = 4 and Equation (9),

we obtain

a = J $ X! ^ d*
a a p =1 p

oo b b

= J 4 J <Pp(x)Q(x) dA J <pp(s)Q(s ) ds = 0
p= 1 ^ a a

Since K4(x, s) = \ K2(x, t)K2{t, s)dt, it follows that

bb b
0 = Ji = 55 IS K 2( x , t)K2(t, s) d/} Q(x)Q(s) ds
aa a
b b b
= 5{[5 02(-v)d.v] [ j Kz(l, s)Q(s) di]} i t
a a a
b b
= 5{S*2(.Y,oe(*)dA )2dt
a a02

200
 Integral Equations with Symmetric Kernels
Consequently,
b
J K2(x , t)Q(x) dx = 0 (10)
a

where we have used the symmetry of the kernel K2{x, 5 ). Multiplying

(10) by <9(0 and integrating the result with respect to t within the
range [a, b\, we obtain
bb
\ \ K z(x,t)Q{x)Q(t) dxdf = 0
aa
b
If we replace Kz(x, t) in this equation by the integral )K(x, £)K(i;,t)dz
a
and perform transformations similar to those carried out above,
we obtain
b
5 K(x, £)Q(x) dx = 0
a

This proves the lemma.

Remark The Fourier coefficients f p of the function /(x) are equal

to hpfXp, where hp are the Fourier coefficients of h(s). In fact,
b b b
f p = \f(x)(pp{x) dx = J (pp(x) \ k (x , s)h(s) ds dx
a a a
b b b
= jj h(s) J K ( x t s)<pp(x) dx = jj h(s) — — ds = —
a a a P P

and, therefore, instead of the series given by (7) we may consider

the series
00

o ')
p

Proof of the theorem We shall first show that (11) converges abso­
lutely and uniformly. From the Cauchy inequality we have

n +q

< ( 12)

201
 Mathematical Physics

From Bessel’s inequality we have

oo 0 oo o

A* < jj/j2(.y) ds and ^ K 2(x, 5) ds < D

p —1 a p —1 p
CO

Consequently, the series ^ A2 converges and therefore the sum

= p i
n-\q
A2 can be made less than e/D, where e is an arbitrary number,
p-n
if n is taken to be sufficiently large. Hence, for large enough n,
n+q
V—1 I L

X ' -a^. O ) < e when x e [a , b]

p=n

which means that (11) converges absolutely and uniformly. Let

00

20) = y > ,p P ) - /P )

The function <2(x) is continuous in [a, A]. It is orthogonal to all the

functions q>pp ). In fact,
b co
5 f(x)<Pr(x) dx = n y - ^ p P ) - / P ) | <Pr(x) dx
a o' - p= 1 P >

= j ; - f' = 0
Consequently, in accordance with the above lemma, it is orthogonal
to the kernel K(x, s), i .e.
b
$tf(x, j)g (x )d x = 0 (13)
a
Next, since the functions Q(x) and <pp(x) are orthogonal,
b b 00

^ 2 2p ) dx = jj fi(x){ ^ — ? > p p )-/(x )| dx

a a t p= i p I

b
= - Se w /w <1*
202
 Integral Equations with Symmetric Kernels

Replacing f(x) in accordance with (6), and using (13), we obtain

b b b
^ Q2(x) dx = — ^ Q(x) ^ K(x, s)h(s) ds dx
a a a
b b
= — ^ h(s) ^ K(x, s)Q(x) d.x ds = 0
a a
Consequently,
00

Q (*) = X X Vpix)—f { x ) = 0
7=i Ap
and this proves the theorem.

9.5 EXPANSION OF THE SOLUTION OF

THE INHOMOGENEOUS EQUATION

Suppose that in the equation

b
<p(x) = I J K(x, s)tp(s) ds + f ( x ) (14)
a

A is not equal to any of the eigenvalues. Then by Fredholm’s first

theorem this equation has the unique solution
?(*) =/(*)+*£(*) 05)
where
b
g(x) = J K(x, s)cp(s) dj
a

By the Hilbert-Schmidt theorem the function g(x) can be repre­

sented by an expansion in terms of the eigenfunctions of the kernel
K(x, s) :
00

£(*) = pX!
—i
o6)
Substituting (15) into (14), we obtain
oo b oo

/( * ) + * X = f (x)+ A 5 K(x, j){ /(j)+ A X

Cp<Pp(s)\ dj
JJ= 1 a p=1203
203
 Mathematical Physics

or
oo b go b
2 c P<pP(x) = s K(x’ s ) f ^ ds+ ?- 2 c p \ K(x, s)<pp(s) ds
p —1 a p=1 a

We now apply the Hilbert-Schmidt theorem to the function

b
J K(x, s ) f( s) ds

and replace )K(x, s)(pp(s) d5 by (pp{x)jXp. We obtain

a

X! < > p M s £ c„ ^ l
P = 1 P= 1

and hence

r
W = —
3 +I — C
1 '-'P for
1U1 C
'“'P = ;—77—
;
'■p Ap '-p '•

It follows that the required solution of (14) can be represented by

the absolutely and uniformly convergent series
CO

?>(*)=/(-'■)+; V - r f r ^ * ) (l7>
r^

If X is equal to an eigenvalue X„ which corresponds to eigenfunctions

9V(*)> ?>,+i(4 <pr+q{x), then Xr = Xr+l = ... = Xr+q.
In this case, it is evident from the formulae for the coefficients
Cp that
f r = f r +l = - = / r +„ = 0

or
b
$ f(* )< P r +t(x) dx = 0 (t = 0, 1, 2, 9)

i.e. the function/(v) should be orthogonal to all the eigenfunctions

of the kernel corresponding to the eigenvalue Xr. The coefficients
204
 Integral Equations with Symmetric Kernels
c „ Cr+l> Cr+q are then undefined (they remain arbitrary) and
the solution of (14) can be written in the form
cp (x) = C r g?r (.Y ) + C r + 1 ? r + i ( . T ) + . . . + C r + q (p r+ q ( .x )

where 'L' represents summation with respect to all values of p except

fo rp = r, r + 1, ...,r+ q .

Remark An equation with an antisymmetric kernel of the form

b
tp(x) = /. jj K(x, s)p(s)cp(s) ds
a

where p{s) is an unknown function, p(s) ^ 0 in [a, b\, and K(x, s)

is a symmetric function, can be reduced to an equation with a sym­
metric kernel with respect to the function y(x) = <p(x) ) / p(x) :
b
yj(x) = /. jj K(x, s) )/p(x)p(s) ip(s) ds

9 .6 E X P A N S IO N THEOREM

It was shown in Section 8.2 that the boundary-value problem

L[&]+kp0 = — [k&']-q0+?.p& = 0 (19)

dx
ai0'(O)-^0(O) = 0, y.20 '(l)+^(O = o (20)
is equivalent to the integral equation
b
0(x) — /. ^ G ( x , s)p(s)0(s)ds (21)
a
or
b
0(x) = 5 Kx( x , s) 0(s) ds (22)
a

where K ^ x , s) = G(x, s)\/p(x)p(s), 0{x) = 0(x)\/p{x) and G(x, s)

is Green’s function for the boundary-value problem (19)-(20).
205
 Mathematical Physics
Consequently, the eigenvalue and eigenfunction of the boundary-
value problem (l9)-(20) coincide with the eigenvalue and eigenfunc­
tion of the kernel /f,(x, 5). This fact enables us to derive the expan­
sion theorem from the Hilbert-Schmidt theorem. In fact, let /(x )
be a function belonging to class A (see Section 4.1), so that

i [ / J = - ^ [ * / '] - ? / = ~F(x)

will be a piecewise-continuous function, and by Hilbert’s first theo­

rem (see Section 8.2)

/(x ) = G(x, 5) F(s) ds

Consequently, by the Hilbert-Schmidt theorem /(x) can be repre­

sented by an absolutely and uniformly convergent series in terms
of the eigenfunctions (<£p(x)} of the boundary-value problem
(19)—(20)
00
f(x) = 2 C„0r(x)
p=l

This proves the expansion theorem for the one-dimensional case

(Section 4.1).

9.7 CLASSIFICATION OF KERNELS

Let us now consider one further application of the Hilbert-Schmidt

theorem. Positive definite (negative definite) kernels are a particu­
larly interesting class of symmetric kernels. The kernel K(x, s) is
called positive definite (negative definite) if for any piecewise-con­
tinuous function h(x,) the integral form
bb
J = jj jj K(x, s)h(x)h(s) ds dx (23)

is positive (negative).
It is readily shown that the necessary and sufficient condition
for the kernel K(x, 5) to be positive (negative) definite is that all its
eigenvalues Xp be positive (negative).
206
 Integral Equations with Symmetric Kernels
In fact, the function
b
f ( x ) = ^ K(x, s)h(s) d?
a

can by the Hilbert-Schmidt theorem be represented by the uniformly

convergent (in the range \a, 6]) series
u uu

f ( x ) = jj ( x , s) h (5 ) ds = ^ — -<PP(x) (24)
a p= 1 P

Multiplying both sides of this equation by h(x) and integrating with

respect to x in the range [a, b], we obtain
bb
J = ^ K ( x , s)h(s)h(x)ds dx = (25)
p= 1

Consequently, if all the eigenvalues Xp are positive (negative) the

form given by (23) is positive (negative). If the form (23) is positive
for any piecewise-continuous function h(x), then for h(x) = <p„(x)
the formula given by (25) yields
bb
K(x, s)<f>„(s)<pn(x) ds dx =
w An

Consequently, Xn > 0 and, similarly, for the negative form. For

positive definite (negative definite) kernels we have the following
theorem.
Theorem If the kernel K{x, s) is positive (negative) definite and
continuous with respect to the variables x, s in the range a < x,
s < b, then it can be represented by the uniformly convergent
series

K(x,s) = Y f M l M
U ^

where epp and are the eigenfunctions and eigenvalues of the kernel.
We shall not prove this theorem here.
It is important to note that all the theorems and facts referring
to the Fredholm equations and described in this chapter and in
207
 Mathematical Physics

Section 8.5 are valid also for the multidimensional case including
kernels of the form
H{P, Q) d
K(P, Q)
\PQ\a ’
where H{P, Q) is a continuous kernel, \PQ\ is the distance between
the points P and Q and d is the dimensionality of the space.

9.8 THE SPECTRUM OF SYMMETRIC KERNELS SPECIFIED

IN AN INFINITE DOMAIN

We have considered integral equations with an infinite range of

integration [a, b\. For integral equations with an infinite range
of integration, the above results will not, in general, be valid. Thus,
for symmetric kernels specified in a finite domain we can establish
the following facts.
1. The spectrum is discrete.
2. The spectrum of a non-degenerate kernel is infinite.
3. To each eigenvalue there corresponds a finite number of
linearly independent eigenfunctions.
For symmetric kernels specified in an infinite domain these
results will not, in general, be valid. This is indicated by the following
example.

Example 1

cp(x) = 2 J si n (ocw) 99(.v) d.v (26)

o
has only two eigenvalues, namely 2, = ^ 2 jci, 2, = — | 2jci, and to
each of them there corresponds an infinite set of linearly independent
eigenfunctions.
To prove this we shall use the following well-known results:
Cl
2
sin(.\\9)e“<7id.v
a2T ,v2

.ysin ( a w ) Cl a.x
d.9 e
a2j\-s2 2

208
 Integral Equations with Symmetric Kernels
where x > 0 and a > 0. Adding and subtracting these equations,
we obtain
CO
71
$ sin (vs) P i / ,_p—as ds = 1 / —
A L' 2 a2+ s 2 l / f [>

Ism(AS)W t e~"~?T?]ds= / f W re~“~

Therefore, 7.a = j^2/m and A2 = — ]/2/ ti are the eigenvalues of
Equation (26) and

<PiM= ] / f e - “ + ~ 5 and >,(v)=j/f e - - ^

9

are the corresponding eigenfunctions for any a. The functions 9?i(x)

[and <p2{x)\ corresponding to different values of the parameter a are
linearly independent. Consequently, to each of the eigenvalues , X2
there corresponds an infinite set of linearly independent eigenfunc­
tions. We shall now show that the equation given by (26) has no
other eigenvalues. To establish this let us substitute <p(s) defined by
this equation into the right-hand side of (26). We obtain
CO 00

<p(x) = 7.2 ^ sin(xs)$ sin(st)<p(t)dtds

o o

Comparing this with the Fourier integral

CO CO

rp{x) = sin(.xs)^ sin (st)<p(/)d/ds

o o
we find that /} = 2 Jn.

Example 2 Consider equations of the form

CO

<p{x) — I Ij H ( \ x —s|)(p(s)d.s (27)

— CO

14 209
 Mathematical Physics

where H(z) has the following properties.

1. It is continuous and positive for all z > 0.
2. There exists a positive number ,4 (^4 < oo) such that the integral
00
jj H(z) cosh a z dz converges for all positive A < a and diverges for
6
a — A.
All the eigenvalues and all the eigenfunctions of this equation can
be found. We shall seek the solution in the form <p(x) = e**. Substi­
tuting this function into (27) we obtain
X 00

eax = A ^ H ( x —s)easds+/. ^ Jf( s—x)e0lsds

— oo X

or, by changing the integration variable (x—s = z in the first integral

and s —x — z in the second), we have
00 00
e = 7. ^ //(z )e a(JC-z)dz+7. ^ 77(z)ea(jt+z)dz
0 o
and hence
00
1 = 2/1 ^ //(z)co sh azd z
o
Therefore, for

= *(“) = ^ -------- ------------- (28)

2 \ //(z)coshazdz
6
the function <p(x) = e00', where a < A, will be a solution of Equation
(27) , i.e. its eigenfunction corresponds to the eigenvalue 7. =7. (a).
Similarly, we find that for a < A the function e-a* will also be
an eigenfunction of Equation (27) corresponding to the same eigen­
value A = 7(a) = /.(—a).
Since coshaz increases monotonically with a, it follows that for
00

2(a) defined by (28), the values of 7.(0) = 1/(2 jj H(z)dz) will increase
6
monotonically and continuously in the range 0 < a < A up to the
value /.(A) = 0. Therefore, to each value 7. e [0, 7.(0)] there corre­
sponds a definite value a (a > 0), which can be determined from
(28) and, consequently, the solutions of (27) are
Q eu + C 2e-“
210
 Integral Equations with Symmetric Kernels
where Cl and C2 are arbitrary constants. The solution corresponding
to 2(0) is
eoor_e-a*
<p(x) = lim = X
oc-»0 2a
which is readily verified by direct substitution.
00

When 2 > 11(2 $ H(z) dz), the solution must be sought in the
6
form e±'Px. Substitution of these functions into (27) then shows
that the values 2 = 2(//?) = 2(—//?), where

> m = K - m = — --------- > -5 -2 —

2 J //(z )co saz d z 2$ H(z)dz

correspond to the real solutions cos fix and sin fix of Equation (27).
Therefore, the spectrum of Equation (27) is continuous; all the
non-negative X are its eigenvalues. Thus, if we take the equation
00

cp(x) = ?. ^ e^lJC_fl9j(j)d j (29)

— CO

then H(z) — t~z, /.(a) = (1—a2)/2, 2(0) = 1/2 and A = 1.

 P A R T TW O

The methods described in Part 1 (for exam­

ple, the method of separation of variables
in cylindrical and spherical polar coordi­
nates), lead to the so-called special func­
tions, i.e. cylindrical, spherical, and other
functions. These functions are solutions of
equations with singular points of the form
d d>'
k(x) q{x)y = 0
dx dx
where k ( x ) vanishes at one or more points
in the range of the variable x. The proper­
ties and applications of cylindrical and sphe­
rical functions as well as certain special
polynomials will be discussed in Part 2.
 10

Gamma Function

The gamma function (or the Euler integral of the second kind) is
defined by
00
r ( z ) = $ e -'f'-M f (1)
6
This function has the following properties.

Property 1 r{z) is bounded and continuous for Re z > 0.

Proof Consider a closed domain D = {0 < 8 < Re z < N } ,

where 5 and N are arbitrary fixed numbers. For all z e D, we have
I/*-1e -,l < / a_1 0 <f<l
|tz-1e“(| < e- 'rw-1 1 < t < oo
Consequently, the function
t s- < / < 1
m = t N~ < t < co
is the majorant for |C~1e~'| in the range 0 < t < co for all z e D.
CO 1 co

Since the integral = jjta-1d r+ $tN-1e~'dt converges, it follows

o o i
00
that the integral j t z~x e~( dt converges uniformly for z e D. It follows
o _
that .T(z) is bounded and continuous in the domain D and, hence,
in 0 < Re z < co, since 6 and N are arbitrary.
215
 Mathematical Physics

Property 2 P{z) is analytic for Re z > 0. To prove this property,

it will be sufficient to show that the integral $ P{z.) dz evaluated over
c
an arbitrary piecewise-smooth closed contour C lying in the domain
D is zero. By M orera’s theorem, r ( z ) will then be analytic in the
domain D and, consequently, for Re z > 0,
00 00

J l'(z)dz = | jj tz“1e“'di‘) dz = jj e~‘ ^ t z~1dzj dt = 0

c co o c
since by Cauchy’s integral theorem

Ut z-1dz = 0
c

The change in the order of integration is justified since the integral

00
$/z-1e~'dt converges uniformly for z e D.
o
Property 3 The identity
A z + i) = z m (2)
is satisfied for Re z > 0. The validity of this property can be estab­
lished directly by integrating by parts:
00 00
T ( z + 1) = jj t ze~’dt = —e~7z|” + z J / z-1e“'d t
o o
00

= z $ / z- le - 'd t = : f ( z )
o
Successive application of (2) yields

_______ r { z + n + \ ) _______
(3 )
z (z + 1)... (z + n —1) (z+«)

Property 4 The function T(z) can be continued analytically to

the entire z plane with the aid of (3) except for the points z = 0,
—1, —2, ..., —n, ..., at which T(z) has first-order poles with resi­
dues given by

(-1 )"
Re f ( —«)
n\
216
 Gamma Function

Property 5 /'(/z + l) = n\
Direct evaluation yields r ( 1) = 1. Using (3) we have

r { n + l ) = n\ (2.)
Therefore, I \ z ) can be regarded as a generalisation of the factorial
function to arbitrary complex numbers.

Property 6 The following relationship is valid:

n
r(z )T (l-z ) (4)
sin n z

Remark It is sufficient to establish the validity of Property 6 for

z = .v, where 0 < x < 1. In view of the uniqueness of analytic
functions this property will then be valid for all z.

Proof

r(x) = jj tx 1e fdt = 2jj u2x 'e “2du (t = u2)

o o
00 00

r ( l —x) = t~xe~{dt = 2 jj v~2x+1e~v2dv (t = v2)

o o
and, consequently,
00 00

r ( x ) r ( l — x) = 4 ^ u2x~Ie~lj2du jj v~2x+le~v2dv

ou
, 2x-l
u
e iu2+v2) I — | du dv
- S0i0
Substituting u = r cos 93 and v = r sin<p, we obtain
03 j i /2

r ( x ) r ( l - x ) = 4 5 \ e~r2r(cot <p)2x~xdr dy
0 0
00 n /2 n /2

= 4 jj e " rV d r jj (cot?)2*-1^ = 2 jj (cot<p)2*-i 3

9

111
 M a th e m a tic a l P h ysics

If in the last integral we substitute cot2<p = we obtain

p ex/?d/3 ji
r ( x)r ( m = ^ T tV = i n ^
— 00

which was to be proved.

Property 7 The function /\z ) has no zeros. Suppose that z0 is

a zero of the gamma function. It is clear that z0 will not be equal
either to a negative integer or to zero. From (4) we have

lim r ( 1—z) = lim - - - ■ = oo

z->Zo z->zo I ( z ) SinTTZ

Therefore, z0 is a singularity of -T(l—z). However, by Property 4

the singular points of the gamma function are only positive integers.
Consequently, 1—z0 = —n, where n is an integer and n ^ 0, while
z0 = 1+n. It follows that

A zo) = -^(rt+l) = n\ ^ 0
Hence, the supposed existence of the zero z0 leads to a contradiction.

Property 8

r (z ) = ei J z _ l \tz *dt (5)

Y
where y is the contour shown in Fig. 10.1.

To prove the validity of (5), let us establish the lemma

^tz~ , e“' d t = ^ t z~lQ~, dt

V1 *2
where y, and y2 are the contours shown in Fig. 10.2.
Consider the integral evaluated over the contour C shown in
Fig. 10.3. This contour encloses a singly-connected domain in which
218
 Gamma Function
e 9* 1 is an analytic function. Consequently, by Cauchy’s integral
theorem

5 e~'tz“]d/ = 0
c
Moreover,

0 = J e“ 'tz_I dt
c

= 5e“'f I - I d r + 5 e 'ft z-1d t + e ' ' / z‘ ‘d ( + 5 e“^ z-1dt

y'2 y[ 'i»2 t3t4

F ig . 1 0 .3

where y\, y'z are shown in Fig. 10.3, and t k = ak-\-ifik (k — 1, 2, 3, 4).
Let us proceed to the limit in this equation as a -> oo, keeping j3k
constant. The integrals jj and $ will tend to $ and —jj , respectively.
K y[ r2 J'i
If we can show that the integrals jj and jj > evaluated over the
> l l2 tjt*
ranges t { t 2 and t3t4, will then tend to zero, this will prove the lemma.
Consider jj , bearing in mind that z = x-{-iy\
tit 2

jj e -ftz- Id t |< J |e“,fz-1||df| = e~a J |/z“ 1li*|

flf 2 <l t2 >l t2

= e-“ J \t\x~le~y!lTg'dt
fl<2

Since |f| < 2a and arg t < 27i, we have

e-a J |f|* -, e- J’ar' , |dr| < e - a(2a)x- 1e23t|j,||r2- ? i

»1»2
219
 Mathematical Physics
For any fixed z the last product tends to zero as a -> co. There­
fore, \ e~tt z ~xdt 0 as a co. The fact that the integral over t3t4
tih
tends to zero can be established in a similar way. This proves the
lemma.
Using the above lemma, we can take the contour y in

F(z) = J e - 'U - ’dt

v
as consisting of a circle y0 of radius r centred on the point t = 0
and the upper and lower edges of the cut along the real axis between
r and oo:
00 oo

F { z)— jj e“Uz_Id t— jj e- t / z-1dt~r § e- r t z-1dt

y r r
(upper edge) (lower edge)

The integral F0(z) = $ e~‘t z~l dt is an analytic function of z every-

yo
where. In fact, F0(z)is continuous in the entire plane and the integral
5 F0(z)d z= \ or1 \ tz~xdzdt evaluated over any closed piecewise
c y0 c
smooth contour C is zero. Therefore, by Morera’s theorem F0(z) is
analytic everywhere.
The integral

Fi(z) = § e {t z ~l dt
r
(upper edge)

can be written in the form of the sum of two integrals:

1 oo
^iO ) = $ + 5
r I

1
The integral jj e~‘tz~xdt is not an improper integral but is a continuous
r
00
function of z. The integral jj e~'tz~xdt converges uniformly in any
1
band —N < Re z < IV, since for all t > 1 we have |e~Uz-1| < e~T‘v_1,
00
and the integral $ c~‘tN~xdt converges. Consequently, the integral
i
00
jj t~‘tz~xdt is also a continuous function of z in any strip for which

220
 Gamma Function
— Re z < N, and hence in the entire plane of the variable z.
Hence, it follows that the function F1(z)is continuous in the entire
plane of the variable z.
Moreover, using Morera’s theorem, we can establish the analy-
city of the function Ffz). The integral \ Fj(z)dz, evaluated over an
c
arbitrary piecewise-smooth closed contour C is zero since tz~ l is an
analytic function of z and

Sf.(‘)d- = CS{i+S!d
C r 1
z
1 oo
= S (S z~ttz~l dz} d f+ § (J e- f / z-1dz) dt — 0
r c 1C
The change in the order of integration is justified because the integral
CO

\ e~'tz~1dt converges uniformly in any strip —N < Re z < N.

i
Finally,
CO CO

F2( z ) = J e ~ '/z_1dt — e2?r'z J e "1*2- 1dt = e2BizF1(z)

r r
(lower edge) (upper edge)

Therefore, F(z) — Fo(z)+(e2riI’z—\)F\{z) is an analytic function of

z everywhere. Therefore, to prove Equation (5) it will be sufficient
to establish this expression for z = x > 0 (see the remark in con­
nection with Property 6).

Proof We have

Fix) = To(xH (e2^ - l ) F , ( x ) (6)

Let y0 contract to a point. Fj(x) will then tend to F(x) and F0(x)
will tend to zero since
271
|T0(x)| < J |e-,/* -1||d/| < \ e - rrxd<p<rxer27i -+ 0
Vo
(t = re i,p on y0)
Consequently, by passing to the limit in (6) as r -> 0, we obtain
Equation (5).

221
 Mathematical Physics

Property 9 From Properties 6 and 8 we have

1
- ‘f - ' - ’dt
r(z + l) 2ni r$e~ '
In fact
1sin (n z + n ) ^ ^ = ~ sin7z:z r ( —z)
T (z+ 1 ) jr 71

'_e‘ r
t~z ldt =
27ii (e - 2jiiz- f ) S e 2ni Y 1

The gamma function is plotted in Fig. 10.4.

F ig . 1 0 .4
 11

Cylindrical Functions

Many problems require the solution of differential equations of the

form given by (1). This equation has to be solved in the case of
problems considered in Chapter 1 by the method of separation of
variables if the problems are expressed in terms of cylindrical (or
polar) coordinates (vibrations of a circular membrane, cooling of
a circular cylinder, and so on). The solutions of this equation will
be investigated in this chapter.

11.1 BESSEL FUNCTIONS

11.1.1 Consider the equation

L[z6\ — z2w" -\-zw' + {z2—v2)w — 0 (1)
Solutions of this equation which are not identically zero are called
cylindrical functions. We shall construct one class of cylindrical
functions as follows. We shall seek the solution of (1) in the form
of the series
w = za{a0+ a l z + a 2z2-\-...) (2)
where a0 ^ 0. We then have
zzo' = z 17[UqO c i i z - \ - c i 2(o-\-2)z2-\- ...]
z2w" = z<I[G0cr((T—l) + <71(cr+l)crz-|-(22(o'+2)(cr+l)z2+ ...]

Substituting these expressions into (1) and collecting terms with

the same powers of z, we obtain
223
 Mathematical Physics

za\c/0a2—a0v2\-i-z17+1[a! (a + I)2—atv2\-i-za+2[a2(a+2)2
—a2v2-\-a0]+ ...J\-za+n[an(o+n)2—anv2-j-an_2]-h ... = 0
If (2) is to be a solution of (1), we must have
a0(a2—v2) = 0
tfitOH-l)2—v2] = 0
az [(cr'b2)2— = 0
fl„[(u+n2) - ^ ] - f a „ . 2 = 0,...
From the first equation we find that
a=
since a0 # 0. Let us take a = v. From the second equation, we then
have ai = 0. Moreover,
a n~Z
an = (io-\-ri)2—v2 ’ n = 2, 3,

Since a = v, it follows that

— a n-2
an =
(2 v-\-n)n
and it is clear that
a2k+l = 0
for all non-negative integers k and
—a2k-2 ( - D^c
a 2k
2\v+ k)k 22> + /c ) ( v - \ - k - 1)... (v+ l)fc!
Substituting a0 = l/2vr(y-\-l) and using Equations (2) and (2{)
of the preceding chapter, we obtain

02k —
2lk+vr ( k + v + \ ) r ( k + \ )
We therefore obtain the formal solution of Equation (1) in the form
of the generalised power series

, M_ V (~it w k*’ m
^ L ixk+ v+ i)t\k+ i)

This series is, in fact, the solution of (1) in the region in which it
converges. It is readily verified using, for example, the d’Alembert
test, that the series converges everywhere except possibly for z — 0.
224
 Cylindrical Functions

Consequently, the function Jv(z) is a solution of (1) everywhere

except, possibly, for z = 0. This function is called Bessel’s function
of order v (sometimes Bessel’s function of the first kind). Since (1)
is unaffected when v is replaced by — v, the function J_ r(z) is also
a solution of (1). When v is not an integer, the functions J,(z) and
J_r(z) are linearly independent because one of them behaves as z v
in the neighbourhood of z = 0 and the other behaves as z~v. On the
other hand, if v is an integer (say, v = n ), then
J_„(z) = (—l)V„(z)
Let us prove this. We have
2k —n 2k-n
00
(-DM- (-l)fchr
J-n(z) =
Z-> r ( k - n + i ) r ( k + 1)
k= 0
z^
k=n
- n+ i ) r ( k + 1)
r ( kk—n

since T{k —n + 1) = oo for all k = 0,1, 2, ..., n— 1.

Substitute k — j + n in the last sum. This yields
. 2s+n

(-l)V „ (z )
r(^ + i)r(^ + » + i)
j= 0

11. 1.2 The behaviour of Bessel’s functions J,(z) and J_v{z) for
non-integral v in the neighbourhood of z = 0 is a manifestation of
a property exhibited by linearly independent solutions of the fol­
lowing class of equations (with singular points)

- t [ k(x)t ] - ^ )y= 0 (4)

in which k(x) = (x—a)(p(x), cp{a) 0 and (p{x) is differentiable
at x — a and in the neighbourhood of this point. This property
will be frequently used below. Let us formulate it more precisely.

Theorem If the solution ^i(x) of Equation (4) is bounded in the

neighbourhood of x = a, then any other solution y 2(x) of (4),
which is linearly independent of ^ (x ), is not bounded in the neigh­
bourhood of x = a. Moreover, i f j ^ x ) ^ (x = a)mw,(x) and w, (0)^0,
then yz(x) has a singularity of the form (x—a)~m at the point x — a,
and if y\{a) ^ 0, then y 2{x) has a logarithmic singularity at the
point x = a.
15 225
 Mathematical Physics
Proof From the properties of the Wronskian, we have

>’2P)M *)->M *)>fiP') = C&0

and hence
d \fi ___ C __
dx y ! k(x)y](x)

Integrating this identity within the range [x, x j, a < x < .vi, we
obtain
*1
yz(xi) _ r Cdt
yi(x) = y f x )
> ip i) J k(t)y\(t)
X

where x x is a fixed number such that the functions <p(x) and n fx )

do not vanish in the range [a, x j.
Consider the first case: ^ (x ) = (x—a)mul(x). We then have

yz(xi) _ r [ _______ d*_______

y 2(x) = y f x )
^i(xO j ( / —fl)2m+V ( 0 « i ( 0

Applying the mean value theorem to this integral, we obtain

yz(x 0 /x Cyi(x) C ____d/

yz(x) =
y,(x,) •lW f i t , ) J (<-«)
2m +1

or
*i
.y2p'i) 1; / x_i_ Cthjx) ( x - a )m (/-rz )-2
>’20')
y f x i ) y i ^' ‘ 2/w?i(f1)i/i(fi)
x < < X], ^ = £ ip)
and, therefore,
j ’2(x) = ^ i O ) > 'i O ) # i P ) 0 —a)
where
>2Oi) |___________£ ______ _ _
A fx)
j ’i (X]) 2m(x\ —a)1"'v(i\)u] 0 )
-c » iO )
B fx)
2ni(p{£i)u\{$\)
226
 Cylindrical Functions
Since ^i(x) and B^x) have no singularities at x — a, this proves
the theorem for the first case. In the second case, similar calculations
lead to the following result
v2(.v) = A 2(.v) y \ (.y)+ B2(x ) \n{x—a)
where
C ln (.Y i— a ) = C
A 2{x )
yi(xi) <p(t2)u2i(h) ’ 2' <p(t2)u\{£2)
x < f2 < * i, h = h(x)
Since the functions A2(x) and Bz(x) have no singularities at x = a
this proves the theorem for the second case.
The above theorem is of considerable importance in formulating
boundary-value problems for Equation (4) within the range [a, b]
where either one or both end-points of this range are singularities
of the equation. If it is required to find the solution ^i(x) which
is bounded in [a, b], then by writing the general solution in the
form
y = Qj'iCjO+C^Ok)
we can find one of the arbitrary constants from the condition
that the solution must be bounded (C2 = 0). Therefore, the con­
dition that the solution must be bounded plays the role of a boun­
dary condition, so that it must be formulated as one of the boundary
conditions. .

11. 1.3 The Bessel functions Jv(Xx) are orthogonal with the
weight p(x) — x. More precisely, for any v > — 1

^ x Jv x j Jv xj dx = 0 when o i^ j3 (5)
0
where both a and /? are the roots of one of the three equations
j v(y) = 0, J'v(y) = 0, yJ'v(y)-\-hJv(y) = 0

Remark The equation

x ly " + x y ' Jr (A2* 2—v2) y = 0 (6)

can be reduced to (1) by substituting ?.x = z. Consequently, the

Bessel function Jv(?.x) is a solution of (6).
15*
227
 Mathematical Physics
Proof of the orthogonality property Consider the two identities

JP(px)- Jv(p x)+ l,u2x — — \Jv{px) = 0

dx2
Let us multiply the first of these by Jv(px) and the second by
Jv(?.x) and then integrate with respect to x in the range [0, /] term
by term. We have
f
^ J dz dz
Jp(?.x)—Jv(?.x) -^ 2 Jv(liX)^ dx

+ jj Jv{Xx)—Jp{Xx) - ^ y , ( JMx)|dx
0
= (jf2—)}) xJp(?.x)JP(px) dx

or

jj "ST {X J f X x ) —Jp{).x) — /„O x)J Jdx

6 /
= ip2—).2) ^ xJp(?.x)J„(px)dx

Evaluation of the integrals yields

( J,(px) — - f ( ? . x ) - J p(?.x) — Jp(px)\ J

= ip2—?2) ^ xJ„(?.x)Jp(px) dx (7)

We shall show that for v > —1 and ). = a/l, p = fi/l, the left-hand
side of Equation (7) will vanish. We note that, using Equation (3),
we can write

J P(?x) = - +XV+2P^ (x) ( 8)

P (v+ \)
I Xx
V \ 2
-xv+iQx(x)
d x JA lx)~ x I \ v + 1)
228
 Cylindrical Functions
where Px{x) and Qx(x) are power series. Using these formulae
we find that

x J v(jj. x ) Jv(Xx)—xJv(Xx) - — J r(jux)

QA* QAT

= ^ w ] (" - 2 ^ + T T + v’+2 Q>(x)

- ['Fr^+iT+x' MPi w ] [” 2 -Av+lf+v't2°«w]

= x 2v+2R 1(x) + x 2v+4R2(x )

where Ri(x) and R2(x) are power series. When v > —1 the last
expression will vanish when * = 0. Substituting X — ajl, / 1 = /?//
in Equation (7), we obtain

jj xJ, l - j - x j J*( ~ T x ) dx = G8) - ^ «) %

0 '
(9)
The orthogonality property follows directly from (9) for the above
values of a and /?. In the third case, the products a d /v(a)/da and
/Sd/„(/?)/d/9 of the right-hand side of (9) must be replaced by —h J v(a)
and —hJf(J3).
Consider now Equation (9) and let us pass to the limit as
/? -> a:

= lira~Er—2 [a/v(PV'v(a)—/? /,(<*)Jp(P)\

P —a P —®

By l’Hopital’s rule we have

/2
= 2 ^ [aJv (a) J'v (a)—Jv (a) J'v (a)—aJv (a) J " (a)] (10)

Next, using the identity

z2J : ( z) + z J ’v(z) + ( z 2- v2) U z) = 0

we obtain

-■*"(«) = - - / ; ( « ) + (

229
 Mathematical Physics
Substituting this expression for J"(v) into (10), we obtain

Jv ( 11)

We note that the square of the norm in the range [a, 6] of any
function T„(2.\;) satisfying Equation (6) can be calculated from
the formula
to.
z d Y v(z)
xY„(?.x)dx Yl(z)
~2?3 dz

where 2 = ?.x. To prove this we use the obvious relationship

b ,
^ x Y v(?.x) Yv (jux) d.v = Y,ljxx)— Y,0.x)

- Y , ( x x ) — Y,(j,x) 1*
L
and then pass to the limit as /u -> In evaluating the limit of the
right-hand side, we must use l’Hopital’s rule in Equation (6) just

as in the calculation of

11. 1.4 We saw in Section 11.1.3 that the orthogonality of Bessel

functions is connected with the zeros of the Bessel functions and
their derivatives. The zeros of the Bessel functions have the fol­
lowing properties.
1. All the zeros are simple, except perhaps for 2 = 0.
2. All the zeros of J„(z) with real subscript v > —1 are real.
3. Any Bessel function has an infinite number of zeros.
The validity of these statements follows from the following
theorems.

Theorem 1 The zeros of any cylindrical function are simple,

except, possibly for z = 0.

Theorem 2 All the zeros of the Bessel functions J,,(z) with real
v > —1 are real.

Theorem 3 Any real cylindrical function has an infinite number

of zeros.

230
 Cylindrical Functions

Proof o f Theorem 1 Suppose that z0 ^ 0 is a zero of order n ^ 2

of the solution y v(z) of Equation (1), which is not identically zero.
We then have y„(z0) = y',(z0) = 0. In view of the uniqueness of
the solution of Cauchy’s problem for Equation (1), y v(z) = 0.
This is in conflict with our hypothesis and, therefore, the zero
must be simple.

Corollary All the zeros of cylindrical functions are isolated zeros.

Proof The fact that z0 = 0 is an isolated zero follows from the

definition of Bessel function J f z ) and from the theorem in Section
11.1.2. In fact, any cylindrical function y v(z) can be written in
the form
y v(z) = CxJ v(z )- t C2N f z )
where Nv(z) is a solution of (1), which is linearly independent of
Jfz ). If y v(z) = CxJv(z), and using Equation (3), we can write this
function in the form y f z ) = Cxz"cp(z), where <p(0) =£ 0. Hence, it
follows that z = 0 is an isolated zero of y v(z). If y v(z) = CxJv{z)
— C 2N v( z ), where C2 # 0, then z = 0 cannot be a zero of this
function since N f z ) becomes infinite at z = 0, in accordance with
the theorem in Section 2.
Let z„ # 0 be a limit point of a sequence of zeros {z„} -> z0
of the cylindrical function y v(z) and y f zo) — 0- It is clear that

y 'v ( z o) = I'm y v(zn)—y f z 0)

We therefore have y v(z0) = 0 and y'(2o) — 0- This means that
z0 is a zero of multiplicity two, i.e. we have arrived at a contra­
diction, by Theorem 1.
It is evident that the above statement is equivalent to the fol­
lowing; in any bounded domain of the variable z, any cylindrical
function y„(z) has a finite number of zeros.
To prove the second theorem, we must first establish the fol­
lowing lemma.
Lemma If a = rel,p is a root of the equation J f y ) = 0, then
the conjugate number a = re -17’ is a root of the same equation.

Proof The Bessel function can be written in the following form

co
where bk = (-r i y

231
 Mathematical Physics

We then have
00

J f r e i9) = rveiv9 ^ bkr2kei2k9

CO CO

( E bkr2kcos2kcp+ i ^ bkr2ksin 2k<p j

or
Jv(rei9) = r veiwp [Afr, <p)-iD,{r, <P)] ]
( 12)
Jv(re~i9) = r ve~lvp[Av(r, <p) + iDv(r, <p)] I
where
00 00

A„(r, <p) = h r 2kcos Ikcp, D fr, 9?) = ] T bkr2ksin 2k<f

The lemma follows immediately from (12). In fact, let a = r e 19

be a root of the equation Jv(y) = 0. Hence,
Jv{a) = Jv(r e'9) = r veiv9 [Av(r, <p)+iDv{r, 99)] = 0
Consequently, A v(r, 90) — Dv{r, <p) = 0. It also follows that 7„(a) = 0.
Proof of Theorem 2 Let a = re'9 be a root of the Bessel function
Jv(z). In accordance with the above lemma, a = re~,<p is also a zero
of this function. In view of the orthogonality property, we then
have

or

(13)

In deriving this expression, we have used Equation (12). How­

ever, the integrand in the last integral is continuous and is not
identically zero. Consequently, the integral itself cannot be equal
232
 Cylindrical Functions
to zero. It follows that the assumed existence of complex roots
of the Bessel function leads to a contradiction.
Theorem 3 will be proved in Section 11.3.9.

Remark It follows from (3) that the roots of the equation Jv{z) = 0
are arranged symmetrically with respect to z = 0 on the z plane.
Using the identities
d 7_r(z) — iQ) \z*Jv(z)\ = zvJv_ f z ) (14)
d: zv
which can be verified directly, the reader can readily prove the
following theorem.

Theorem 4 The functions Jv(z) and / v+1(z) do not have common

roots (except possibly for z = 0).
If we carry out the differentiation in (14) and eliminate /'(z),
we obtain the recurrence relation

Jv+i{z) = Jv(z) (15)

z
If v is equal to an integer n, then we can use the recurrence relation
to express all the functions Jn(z), n > 2 in terms of J0(z) and J fz ) .
Using the above theorems, the positive roots of the equation
Jv(y) = 0, where v is a real number, can be arranged in an increasing
sequence
7l < 7 l < ••• < 7m < 7m +1 < •••

It is evident that these roots are functions of the subscript v, i.e.

7 m = 7 m (v )

Theorem 5 7m(v) are increasing functions of the variable v. if

v > 0.

Proof For any fixed m we have

J v[7 m(v)] = 0 (16)
Differentiating this identity with respect to v and omitting, for
simplicity, the subscript m, we obtain

Jv(Z)\z = Y -h Jv{z)\z=y = 0 (17)

233
 Mathematical Physics
The prime will represent the derivative with respect to 2. Next,
from the first formula in (14) we have
z /'(z )-l'/,(z ) = —zJv+i(z)
and hence, using (16), we obtain
J'v(z)\z =y — Jv+\(y) (18)
Returning now to (17), we have
dy 1
- r - M z)lz-y (19)
dv Jv+i(y) ^
Differentiating the identity

- - [zJ'v(z)i+ I z - — I Jv(z) = 0 (20)

with respect to v, we obtain

_d v2 \ d
- - Jv (2) + 2 Jv(z) = 0 ( 21 )
dz z I dv
If we now multiply (20) and (21) by (djdv)Jv(z) and Jv(z), respectively,
and subtract one from the other, we have, after integrating with
respect to 2 between 0 and y,

i ; ( z ) |i ,( z ) - j ,( 2 ) } / ; ( 2 ) dz dz
ov or
or

\-J'v (z) ~ J v (z )—zJv (z) - - J'v (2)J = —2v (j — J; (2) dz

0
Since
Jl (2) = b2z''-l + z ' ^ P 2(z), J,, (2) = b {z''-:r z r '2P , (2)

Jv(z) = b2z" In 2+C3Z" -!-zv+2 In 2 P3( z ) + z ' ^ 2P4(z)

ov

~ J ' ( z ) = b4z* 1 In 2 C ,:” l + z"+1\n z P5(z) ,-z"-1Pb(z)

ov
where Pk(z) are power series, k = 1, 2, 3, 4, 5, 6, the substitution of
the lower limit of integration 2 = 0 yields zero. Therefore, using
(16) and (18) we obtain
y
2 -J
I JHz)dz = yJv+, ( y ) ~ J y ( z ) 1=y
0
234
 Cylindrical Functions

Using this formula and Equation (19), we have

dym = 2v ?
— Jl{z) dz > 0
dl' ymJv+l(.Yni) 3 z
This proves the theorem.

It is readily verified that •

J \ r t (z) = |/ T sin z and J _ 1/2 (z ) = j / ~ cos z

11. 1.5 Example Consider the cooling of a uniform infinite rod

of circular cross-section (radius R), whose surface is maintained
at zero temperature. The initial temperature at internal points
in the rod is given to be <p(r).
The mathematical formulation of the problem is as follows:
it is required to find the solution u(r, t) of the equation V2« = (1 ja2)ut
for t > 0 and 0 < r < R which satisfies the following initial and
boundary conditions:
it(r, 0) = 9?(r) , u ( R ,t) = 0, |//(0, 01 < 00

Solution Separating the variables so that u{r, t) = &(r) W(t),

we find that

= .&(R) = o, [<z>(0)j/<oo
r
R(t) = C t ~ 'alt for / > 0
The general solution of the equation for <Z>(r) 7can be written
in the form
<P{v)= AJ,(]r?.r)-rBN»(\r).r)
where N0 \ X r is a solution of the equation for 0 (r) which is linearly
independent of J0(]/2r).
By the theorem in Section 11.1.2 r) is unbounded in the
neighbourhood of r = 0. Since the required solution must be
bounded, we have B = 0.
Consequently, &{r) = A J qQ/X r). It is clear that we can set
A = 1. From the boundary condition for r = R, we find the equa­
tion for the eigenvalues

Join) = o, n = \'X R

235
 Mathematical Physics
By Theorems 1-3 (Section 11.1.4) this equation has an infinite
number of simple real roots
Mi < M t< ••• < Mn < •••

These can be used to define the eigenvalues

and the eigenfunctions for the problem JQ rj .

We shall suppose that this set of eigenfunctions is complete and

that the functions cp(r) can be expanded into a series in terms of the
eigenfunctions J0 rj.
The solution of the original problem will be sought in the form
of the series

u(r, t) = ^ C„ e R2 J0 rj

The coefficients C„ will be found using the initial conditions and

the orthogonality of Bessel functions
°°
u(r, o) = q>(r) = ^ c "J° n r r )
n=1

Let us multiply this identity by and integrate the

result with respect to r within the range [0, R]. Using the ortho­
gonality of the Bessel functions and the formulae for the square
of their norm, we obtain
R
. \ r <p(r) r 1dr = C« R -
0 ' '
and, consequently,

2 Mk
Ck dr
R

236
 Cylindrical Functions

Remark For an approximate solution of the problem, it is sufficient

to retain only a few of the first terms in the series, for example

7 -i- i 2 \
u(r, t) ^ Ci e + C2e ^ •/<>(— ->*

Fi and fi2 can be found from tabulated values of J0(x). They are
//, = 2.4048, fxz = 5.5201

11.2 HANKEL FUNCTIONS

11.2.1 Cylindrical functions of the second class are constructed

as follows. We seek the solution of Equation (1) in the form of the
contour integral
w(z) = \ K(z, I M I ) df (22)
c
where K{z, s) is a given function and v(£) is an unknown function.
Substituting for w(z) into the left-hand side of (1), we obtain

L[zo] = J { z 2K zz+ z K z + z 2K - v 2K } v (£ ) d£ (23)

c
We suppose that the contour C and the function K(z, £) are chosen
so that all the above operations can be carried out.
The function K(z, I) is chosen to be the solution of the equation
z *K2Z+ zK z+ z2K + K x = 0 (24)
so that L{zo) can be written in the form

L[w] = - J (v2K + K (i) v ( i ) d f

c
= - \ k {v " + v 2v } d ^ + {K v '+ K . v } V
a

c
This formula is obtained by double integration by parts of the second
term; A and B are the end-points of the contours of integration.
Let us now take the function K{z, I) to be (l/jr)e“ ,ZBlnf and let us
takea(£) as a solution of the equation
v " Jrv2v = 0
for example e”L The contour C is chosen so that all the above
operations can be performed and so that the expression Kv'—Kfi
237
 Mathematical Physics

is equal to zero at the ends of the contour C, i.e. at the points A

and B. We then have

w(z) = — Se -'z sln f+i”1 df (25)

•"* c

11.2.2 If Ci and C2 are the end-points of the contour C (Fig. 11.1),

we obtain the following two cylindrical functions

Ci
(26)
H[2)(z) = — [ e- ‘” ln{+w df
c2
which are called Hankel functions.S
o

So far the Hankel functions have been introduced in a purely formal

manner. It is therefore necessary to show that the functions H (vl)(z)
and Hl2\z), defined by (26), are, in fact, the solutions of (1), i.e.
they have first- and second-order derivatives and that when they are
substituted into Equation (1) the differentiation (both first- and
second-order) can be carried out under the integral sign. It is also
necessary to show that for the above choice of C, and C2, the expres­
sion K v '—Kzv vanishes at the ends of these contours.
We shall now establish some of the properties of Hankel func­
tions.
1. Hankel functions are defined and continuous in the domain
Re z > 0.
To show this, it is sufficient to establish the uniform convergence
of the integrals which define the Hankel functions for Re z <5 > 0,
where b is any positive number. To be specific, consider the function
H[}}(z). On the upper part of the contour C,
£ = —.'7-:-//?(/? > 0), sin£ = —/sinh/S
238
 Cylindrical Functions
On the lower part of Cx
i = //3(/S < 0), sin I = i sinh /3
Consequently, on these parts of Cx the functions
and e5stnhP' ^ will be the majorants of the modulus of the integrand
(v = sJr ‘q), respectively. At the same time, the integrals of these
ao 0

functions jj e~SsirLhfi~sP+riqdP and j e^^^^'^d/S will converge. Hence

0 — oo

the original integral over Cx will converge uniformly for Re z > (5 > 0.
The uniform convergence of the integrals

$ XzVdS, J Kzzv d i , J Ke(v d | (p = 1, 2)

Cp Cp Cp

may be established in a similar way.

2. Hankel functions are analytic for Re z > 0. To establish this,
we note that \ / / („1)(z)dz, evaluated over any piecewise-smooth con-
L
tour L in the region Re z ^ d > 0, is zero, since

$ //<X)(z)dz = J J e - i2sinf+,'^dzd£ = 0
L Ci L

The change of the order of integration is justified because the integral

over Cj converges uniformly and the function e~'2slnf is analytic
everywhere. By Morera’s theorem it follows that H ^ \ z ) is an ana­
lytic function. The fact that H^2)(z) is analytic can be shown in a simi­
lar way.
Since the integrals $ Kzv d£ and J Kzzvd£ converge uniformly
Cp Cp
for Re z > d ((5 > 0), it follows that when one evaluates the deriva­
tives of Hankel functions one can differentiate under the inte­
gral sign.
3. The following limits are valid:
lim [K{z,Z)v'{!Z)-Ks{z,£)v{g)} = 0
| = —n-cip -* — too
Rez > 0 (27)
lim {K (z,£)v'(£) -Ke(z,£)v(£)} = 0
i - ifl — i co

Let us prove the first of these. On the upper part of Cj

|AT(z, S)v'(i)I = |e- i2Slnf+i,’/v| = |v|c-xs^p-sp+xq 0
M oo
\K^(z, f)u(f)| = | - / z c o s |e - i2Siuf- ,',,’ |
= |z| cosh o, f oo
This proves the first relationship.
239
 Mathematical Physics
The second relationship can be established in a similar way. It
follows that the Hankel functions are the solutions of Equation (1)
and are analytic in the half-plane Re z > 0.
4. The Hankel functions are linearly independent. To prove this,
it is sufficient to show the existence of a sequence {z„} in the range
in which the Hankel functions are defined, in which one of the
functions tends to infinity and the other is bounded. Consider
Equations (26). On the upper and lower parts of Q we have £
= —Ti+i/3 and, correspondingly, £ = —ip (ft > 0). The integrals
over these parts of Cx can therefore be written in the form
00 oo

J + iC e - z s i n h 0 -v/re - i V i d 0 and e- 2 sinh/»+Wl d 0

J7T ?! J
0 0
Consequently, the Hankel function H^l){z) can be written in the
form
CO oo

H i ly (z) = — e -* v [ e- zsin^ - vM/9---- - [

n J 7i J
o o
o
+ — [ e-izsinS+htd^r

or
00 00
= - ^ - e ~ ivn[ d/5 _ J _ C Q—zsinhfl+i'fl d ^j
• JZ J 7Z J
0 0
71

+ — ( eizsln{- ,xdf (26i)

7Z J
0
Similarly,
00 00
i/<2>(z) = — t inv f e-zslnll^- ^d/5+ — f e-zsinh/?+>/?d/5
n J n J
0 0
71

+ — ( e - izslnf+i,,|d£ (262)
n J
o
As z tends to infinity, over the sequence z„ = x0+/y„, where x 0 is
fixed, x0 > 0, y n > 0 and y n -> oo, the first two integrals in (26j)
and (262) remain bounded since
00 00
(j e-znsinh^=FV/)d^
0 0
240
 Cylindrical Functions

The integral
71

^ g —^nSinfg/Arosinf-iVI d £

0 0
will also be bounded as y„ -*■ oo. Consequently, remains
bounded for z„ -+ go. The integral

= j e>,nsin1e ~ sin1+‘‘,f d £
o o
in (262) tends to infinity as y„ -> + oo, since sin £ > 0 within the
range of integration. Consequently, i/<2)(z„) -»■ oo as z„ -> co. This
proves the linear independence of Hankel functions.

11.2.3 Direct evaluation will show that the following recurrence

relations are valid:

a w + t f & w = — H<tKz) (* = 1, 2) (28)

Z

= - 2 — H ?\z) (k = 1,2) (29)

In fact,

= — ( e - izsln|(ei(v+1)l+ e i(v- 1)f)d-

7Z J

71 J
Ck

Integrating by parts, we obtain

+ — — - [ e ~ izsm + M f l g = A * J_ [ g - i z sin = — / f ( l )(z)

Z 71 J Z 71 C z
C, C,

16 241
 Mathematical Physics
since the substitution of the limits into the integrand yields zero.
The procedure for //*2)(z) is the same. Next,
]_
H % (z)-H f\(z) ^ e-i*slnf4W2/ Sin fd f
71
Ck
On the other hand,

2 — H i k\ z ) = — [ e“ 'z sln? +/l,’(—/sin^) d£

QZ 7Z J
Ck

Consequently, Equation (29) is also valid.

For Re z > 0 we have
H ^\z)+ H ^\z)
(30)

Since the functions Jv(z) and H (k\ z ) (k — 1,2) are analytic for
Re z > 0, it is sufficient to show that (30) is valid for z = x > 0.

Proof Let j v(z) represent the right-hand side of (30). We then have

jv(x) = —— \ e~ix sin s+,vSd i (31)

2 71 J
Co

where Co is the contour shown in Fig. 11.2.

Fig. 11.2

Let us substitute

into this integral. The half-line (—n + / oo, —tx) of C0 transforms

into the half-line ( + oo, x/2) on the real axis, and the half-line
(ji, 7i-\-i oo) into the same half-line (x/2, + oo) on the real axis, but
oriented in the opposite direction; the segment [—ti, ti\ transforms
into the circle \a\ = .v/2.
242
 Cylindrical Functions

Under this transformation, therefore, C0 will be transformed

into the contour y (see Chapter 10), which is traversed in the opposite
direction. Consequently,

Expanding ex2,ia in a Laurent expansion in powers of a and

integrating term by term, we obtain

«-£(f)'i( k= 0 y
-
i2 k +v

(-i)i-
—- = J v{x)
k=0
(&+V-M) r ( k + \ )
where we have used Equation (7) of Chapter 10.
Therefore,

^ W = y [ ^ V ) + t f ?>(*)]

We have obtained the integral representation of the Bessel

functions
Jv{x) = — [ t ~ iz sin "+ivS di (32)
2 71 J
Co

If we split this into three integrals, we obtain

•j j 'j j

J _ f g/z sin (i0)+nir-vP ____ _l f gi z sin (ifl) —i n v —vpdfi

Jv(z) =
2* ) p 2 7i }

\— L ( e-b ^ '+ '-d oi (f = a+//3)

2 71 J

or
1 e - i z slna + i r a d a
Jv(z) = (e'™—e“iwt) ^ e -2B,nh/'-
2ti
o
16* 243
 Mathematical Physics

or
00 71

Jv(z) = ( e -z sinh d £+ — - [ e~h sln a+,Va da (33)

71 J 2 71 J
0 —71
In particular, for v = n (an integer) we have
71

; „ ( z ) = — ( e- i2Si"“+inIda (34)
2 71 J
— 71

It follows immediately that

\Jn(z)\ < cosh y (z = x + i y )
and
|/„(A')I < 1

11.2.4 The functions

H ? \z)-H ? \z)
N v(z) = (35)
2i
are called Neumann’s functions. They are analytic for Re z > 0
since the functions H ^ { z ) and / / („2)(z) are analytic in this region.
It is clear that Nv(z) is a solution of (1).
It is evident from Equations (30) and (35) that

Hl"(z) = Jv(z)+ iNv(z)

H ? \ z ) = J v( z ) - i N v(z) °

It is readily shown that the functions Jr(z) and N v{z) are linearly
independent. If this were not so, there would exist a constant C
such that for Re z > 0 we would have Nv(z) = CJv{z). However,
from (35j) we would then have

H\}\z) = (1 + i C ) J v(z) and H (,P(z) = ( 1 - / C ) / V(z)

and hence H ^ \ z ) — DH[l\z), where D = (1 —t'C)/(l + /C), which

contradicts the linear independence of the Hankel functions proved
above.
Therefore, the functions T,,(z) and Nv(z) regarded as functions
of and form a fundamental system of solutions of Equation
244
 C y lin d r ic a l F u n c tio n s

(1). From (35J it follows that the recurrence relations (28) and (29)
of this chapter are valid also for J„(z) and iV„(z):

Jv+i(z)+Jv-i (z) = — Jv(z) , Jv+1(z) - Jv_ 1(z) = —2J ’„(z)

z

iW + i^ + A ^ z ) = vV,(z), ^ +1(z)-iV ,.i(z) - - 2 N'v{z)

11.3 ASYMPTOTIC REPRESENTATIONS

OF CYLINDRICAL FUNCTIONS

11.3.1 In many problems encountered in physics it is necessary

to evaluate the steady-state solutions. Mathematically, this leads
to the consideration of functions for large values of the arguments,
i.e. to the asymptotic behaviour of these functions. When studying
the asymptotic behaviour of functions in a region £, they are re­
placed by simpler functions with similar properties. Most frequently,
the simpler functions are taken in the form

l c \ , c2 , , cn
co H - - - - - - - - 1— 2~ + • • •
Z Zz zf

Definition The series Cq+Cj/zH- ... + c„/z"+ ... is called the

asymptotic expansion of a function f(z) in a region £ containing
sequences converging at infinity if

Ck
lim zn = 0 where n = 0, 1, 2, ...
z —>CO
ze£ k=D
The following notation is used:

Cn
/(* )' z
+
^
— +
ZH ^

or
Cl
/(z ) = c0+

It is readily shown that if the asymptotic expansion exists, then it

is unique.

245
 M a th e m a tic a l P h y s ic s

In point of fact, it follows from the definition that when n — 0,

lim {/(z)—c0} = 0, and hence c0 = lim /(z). For n = 1 we have
r —*00 z>
- oo

limz j/(z ) —c0— j = 0

and hence
ci = limz {/(z)—c0}
Z —►00

n—1
\ ^ Ck
c„ = lim z"
Z-*00 ^ > - k2= 0, 7

However, different functions may have the same asymptotic expan­

sions. In fact, if

/ ( * ) ~ c 0+ - ^ + . . • + — +
then also

/(x)-i-e * ~ c0+ —- + + ••• (for x > 0)

If

V (z ) c,
/(* ) = 7n + \
? ( z) Z

then
C|
y>(z) - <p(z) C0+ - i- - + —
z z

will be called the asymptotic representation of y(z).

11.3.2 We shall now find the asymptotic representation of the

error integral
.V

246
 Cylindrical Functions

for large a > 0. It is clear that

0 (A) =1 - ~ \ e (2dt
- J
y it J
V 71

It is therefore sufficient to find the asymptotic representation of

00
the functions / ( a ) = $ e~'2dt.
X '

We have

Integrating by parts, we obtain

1 . 1- 3
e*2/(-v) = ^ - - - T O +
2a 22a 3 ^ 23a 5 -

-(- i r 1 1 ’ 3 2n5^ n-2r ~ 3) + *»(*)

For the remainder
f e* “
2” ) ~ T n~
x

we obtain
1•3 • 5 . . . ( 2« - l )
!*»(*)! < 2" +!x 2n+l ° { ^ Y)
Consequently,

/(*) C~*2[ ^ 22x3 + 23x5

( - i r ' ( 2« - 3) ! !
+ n v2n—1
2" A
and
3
23a5

' ( - l ) ’- '( 2n - 3) ! ! ^ + ( ) |_ _ l_ jj
n . J in — 1

247
 Mathematical Physics
11.3.3 One of the most widely used methods for obtaining asymp­
totic representations is the method of steepest descents. This is
based on the fact that for large values of the variable x the magni­
tude of the integral
/(x ) = $y(£)ex,,(f)d£
c
is determined mainly by the part C„ of the contour of integration
C, on which |e;"p(f)| = ex Rcfp^l is large in comparison with the
values of this modulus on the remainder of C. The integral over Cu
is estimated more readily the shorter this region and the steeper
the fall of the quantity x Re 9?(£). In the method of steepest des­
cents, one tries to deform the contour of integration C into the
most convenient form C in the above sense. By Cauchy’s theorem
this deformation does not affect the value of the integral provided
we remain in the region in which <p(£) and y>(£) are analytic. Since
the function <p(£) = u(a, /?)+iv(a, /?), £ — a+ij8 is analytic, the
direction of the most rapid variation of u(a, /?) is the same as the
direction of the line v{<x, /S) = const. The deformed contour Cn
must contain the point £0 = a0+i'/?0, at which u(ct, /S) reaches its
maximum value (among the values of this function on C).
It is readily shown that ^'(£0) = 0. In fact, the derivative of
u(a, ft) along the line C, taken at the point £0, is equal to zero since,
at this point, the function u(a, /?) reaches its maximum value
(along C).
dv
Next, = 0, since in the neighbourhood of the point
ds f=fo
£ = £0, we have v = const along C. Therefore,
da dv
P '( f o ) +i = 0
ds !f=fo ds Is=£0
The point £0 for the surface u = i/(a, /S) is, of course, a saddle point.
Therefore, in applying the method of steepest descents to the
asymptotic form of the integral \y»(£)ew(f)d£, the path of integration
c
C must be deformed into C which passes through £0 where
9> ' ( £ o ) = 0, and in the neighbourhood of this point coincides
with the line v(a, j3) = const = v ( a 0, fi0).
11.3.4 We shall use the method to deduce the asymptotic behav­
iour of the Hankel functions 7/{,J)(x) and H (v2)(x) for large values
of x (x > 0).
Let us establish the following theorem.
248
 Cylindrical Functions

Theorem Any real solution of the equation

■ i ^ + i ^ + [l- 7 ) y = 0 (lmv= 0)
has an asymptotic representation of the form

, ’(A) = ^ s in (x + '5o)[ 1 + 0 ( ^ ) ]
where A0 and S0 are constants.

Proof Consider the function ji(x ) as defined by

y i(x)
y(x) =
}/ x
The differential equation for ^i(x) is

/ -----:
"ill (36)
y i + \ i■ hi = o

For large x this equation is not very different from

d2zc
-zo = 0
dx2
whose general solution is
w = ylsin(x+<5) (37)
where A and S are constants.
Therefore, for large values of x we shall seek the solution of (36)
in the form
y f x ) = A (x) sin [x+<5(x)]
where A(x) and d(x) are the required functions.
Since there are two required functions, and they are related by
only one condition, the requirement that T(x)sin[x+($(x)] must
satisfy (36), we can subject them to one further condition. We shall
choose this in such a way that the derivative of ^ (x ) is evaluated
as if A(x) and <5(x) were constants. Since

y[ = T cos(x+ 6)+ y45'cos(x+ < 5)+ ^l, sin(x+ 5)

249
 Mathematical Physics

we shall suppose that

^<5'cos(jc+<5)+^'sin(x+d) = 0 (38)
We then have
y[ = y4cos(x+(5) (39)
Evaluating the derivative y[' and substituting this into (36), we
obtain

.4'cos ( a:+ (5) —A I + sin(A-r<S) = 0 (40)

where

Eliminating A and A' from (38) and (40), we obtain

S' (a) = sin2(A+<5) (41)

and hence

sin2[f+«5(f)]d! (42)

For a fixed a and b -> oo, the right-hand side of (42) tends to a limit.
Consequently, the left-hand side will also have a limit:
lim 8{b) = (50
b —►oo

We therefore have
00
s (a) = <30+ ^ ~ 2-s in 2[£+<K£)]d£

However,
CO 00

y 2-sin2(£+<5)d£
x x
X X

and therefore

S (a) — <50+ O
x

250
 Cylindrical Functions
From (38) and (41) we have

(In A)' = — = sin2(.v+ 6)

and consequently

V ^(“ sin2(£+<5) Jf.

In A(b) — \nA{x)-sr —- *-2

Repeating the argument given for <5(.v) and d(b), we arrive at the
conclusion that the limit
lim In A ( b ) = In A 0
b —* oo
exists and

ln,4(x) = \ n A Q+ 0 ^ —~ j
Consequently,

^ ) - ^ [ l + o (l)]
Therefore,

Vi(-v) = A 0 j^l + o | —j jsin j^x+60+ O j

= ^o^in(x+<50) | l + o |- ^

and

y(x) = —^sin(.x+<50) 1+ 0
Vx d )

11.3.5 Let us return now to the consideration of Hankel functions.

To be specific, consider

H[l\ x ) = — [ e~ix sln e+,vSdS = -1 f e c- lx*lnSd$

71 ,) 71
Ci Ci

where <p{£) = — z sin £. The saddle points S0 can be found from the
equation

<p'(£o) = —i cos So = 0, So = y (2k +1)

251
 Mathematical Physics
where k is an integer. Since the contour Cx lies in the strip
—7t < Re £ < 0, we shall be interested only in £„ = —ti/2:
<p(|) = —i sin | = —i sin(a + i/5) = —i sinacosh /? + cos a sinh/?
Therefore, v(a,/?) = —sin a cosh/?. In the neighbourhood of the
point £0 the deformed contour C, has the equation
—sin a cosh/? = —sin a0 cosh/?0 = 1
or
sin a cosh /? = —1
The direction of this curve at the point £0 = —n/2 is determined
by the angular coefficient

_d£_| = _ J _ = _!
da WI = ~ Y" sin a a„_^3L
~ 4

Therefore the contour Ci can be taken as the broken line

( + co, £u £2, £ 3, f 4, — oo), shown in Fig. 11.3.

Fig. 11.3

In accordance with the above discussion we can split this contour

into three parts: C ^ + o o , £1; £2)> Cn (£z, £3) and CIU(£3, £4, — co).
We shall assume that

f2 = « 2 + /ft = — +A'-°e‘ 4"

s£3 - a 3 + / / ? 3 = - - ^ + A ' - ffe _ ‘ 4

where a is a fixed number 1/4 < cr < 1/2.

252
 Cylindrical Functions
11.3.6 Let us now estimate the integrals over C1 and CIH. It is
clear that

-ix sin c+ ^ j ^ g —ix (sin 3 cosh /f+i cos a sinh P) +iv(a+ifl)

er
^ J g* cos 3 sinh 11
CI
_ gX cos 32 sinh 02 ^ gX (cos 3 sinh 0—cos 32 sinh 02)—*P j
^1
_ gX cos 3 2 sinh 021 ^ gX (cos 3 sinh 0—cos 32 sinh 02)—^
(+00, fi)
j ^ gX (cos 3 sinh 0-cos s2sinh 02)-v0 j(Jt| |
(Si. £2)
Each of these two integrals is bounded since cos a sinh /3—cos a2
sinh /?2< 0 for £ ^ £2 and decreases with increasing x. Consequently,

_J_ 1-23

jj = o (e* cos “2sinh = C>(e 2 * )

since

a2 —
— 7^1 1 T^j o2
@
2 X ° \/ 2 x a\/2
and
-1 v —
lx-° .x~ia
cos a2 sinh /?2 = = -+ ...
x a\/2 2 1/2 -3! 1/2 ‘ 3! 2 (/ 2

~ s \ . f\ a I ***
2x?° 6! x6<T
Similarly,
f g — ix sin I + ir | ^ gX cos 3 3 sinh 03
1J I
cm
x (cos 3 sinh 0—cos 33sinh 03) VP ]d^T'i
x 1 1 e
ii3. I.)

_J_ J gX (cos 3 s in h 0—cos 33 sinh 03) VP d/?| — O (c )

( L ,- o c )

253
 Mathematical Physics
11.3.7 Let us estimate the integral over Cu . We have £ = —rt/2
-\-rei3n/4 in the range (£2j—7t/2), 0 < r < x~a, and £ = —7t/2 + re~i:z/4
in the range (—jt/2, ?3), 0 < r < x- ?. These two formulae can be
combined into one:
t 7l

where — ^ r ^ x~~a. Therefore,

r-O ,n . 71 7T
—I—
—w « £_i_:„e i c. f i jc c o s f r e 4 ) -~/4r v ffV re 4 —‘“T ,
Se-x l3i n
C„
^ =
-A :-a
v e 2 e dr

./ n n \ x a —i — —i—
iIac—v ——— J r* f.x[cos(re 4 ) —l]+ /vre 4
= e \ e

Expanding cos(rQ~,!t/4) into a series in powers of r, we obtain

X
— na -n
— I—
,n
— I ---
r ix [cos (r e 4 ) —1] + i»r e 4
it — j e dr
-x~ a
x~° xr2
— ^ e 2 emi(JC’r)et",2(JC’r)d/'

where
vr r6 r l0 p 14
mv(x, r) = —r=- +.v
1/2 6 7 “ 10[+ 144
.2+4 n
,»+i +
(2+ 4 n)\
,,4n
vr r4 r8
ntz(x, r) = —x 2------ — + ... + ( _ l ) n+i--------
|/2 4! 8! ^ ; (4n) !
or

R — ^ e ^ e”11 (cos//i2+ /sinw 2)dr

-A.-*7
Applying the mean value theorem to each of these two integrals,
we obtain
R = {e'"l(x’0,-v“<7)cos/7/2(x, (+x~CT)
X-< xri
~2
+ /e"i1(-v' e2':_<7) sin/7J2 (x, 02x“°) ^ dr

254
 Cylindrical Functions
As .V -* oo, the functions mk(x, OjX a) ( k , j = 1 ,2 ) tend to zero
as x~°. Consequently,
X~a _ jcr_2
R= + S e 2 dr
-x~a
where p^.v) and p2(.v) are bounded as x -*■ oo. Substituting r\Zx/2
= y, we obtain

R — { 1 ( p i + 2‘p2)x a

Using the asymptotic representation of the error integral (see

Section 11.3.2), we obtain
/T _ - l xl ~2a
R = y ~ {\+ p {x)x-°\\/n {\-0 (e 2 )}

— {\+Pi{x)x~oO(c-0-5xX 2<!)}
where
P(x) = pi(x)-rip2(x),
Pi{x) = p ( i ) + / 0 ( e J '5sl 2°)
and the modulus of p3(x) is bounded for large x > 0. Therefore,
for large x > O w e have

Hi%x) = l / — e‘ 4 ^ (1 -Tft(x)x- *} (43)

f 71 X

Similarly, it can be shown that

H {2\ x ) = l / ~ e~,(^ T“ T) {1 + p ( x ) r » } (44)

where p (*) is bounded a s x - » c o .

11.3.8 From (43), (44), (30) and (35) we obtain

IT 71 71
cos I x v —------ { l+ a ,(x )x .a 1 (45)
= V -:

W,(*) = 1 / sinyc—v —----- ^-j{ 1-\-az{x)x a } (46)

V -*
= toM - p (x) ^ fl2(jc) = {pi(x)_ p (x))
2

255
 Mathematical Physics
Using the theorem on the asymptotic representation of real
cylindrical functions (see Section 11.3.4) and the uniqueness of the
asymptotic representation, we conclude that and a2(x)x~a
are small quantities of the order of 1jx. Consequently, p(x)x~a and
p(x)x~a are small quantities of the same order.
Therefore, we obtain the following improved versions of
(43M46):

These are, in fact, the required asymptotic representations.

We note that these formulae are valid for any z: |z| > 1, | argz
< 7i—S, where <5 is an arbitrary small number.
11.3.9 The validity of Theorem 3 of Section 11.1.4 follows from
Equations (49) and (50) and from the linear independence of the
functions J v{z) and N v(z). In fact, because of the linear independence
of these functions, an arbitrary real cylindrical function y,.(z) can
be obtained from the formula
>’,.(z) = Dl J v(z)+D1N v(z)
Consequently, it will have the following asymptotic representation
2
yr(z) = A c o s l z - n~- V - —
1 71 Z

+ Z)2sin - y 1+0

Substituting A/£>2 = tana;, we obtain

71 71
} ’v ( z ) 2+ D \ sinjz —J+- —— + a ; ) { l + 0
= V x z i / D l + D l s 'm {

Theorem 3 of Section 11.1.4 follows directly from this asymptotic

representation, together with the result that the distance between
256
 <* 5 6 7 8 9 10 JJ 12

F i g . 1 1 .5

17
 Mathematical Physics
two neighbouring zeros of the Bessel function Jv(z) (and of the
Neumann function) tends to n as the absolute magnitudes of the
zeros increase without bound. Plots of Bessel and Neumann func­
tions are shown in Figs. 11.4 and 11.5. that

11.3.10 Using the asymptotic behaviour of the Jv(z), J - V(z), Nv(z),

Hi^iz) and 77<,2)(z), one can readily show that
Jv(z) COS V 71 — J_„(z)
N v{z) =
sin v n
and

H ? \Z) = —A [t~inVJv{z) —J l v{z)]

snivel

H?\Z) = —r-— WnlJr{z) —J-v{z)]

sinvjr
If v is an integer, the right-hand sides of these formulae are meaning­
less since they are of the form 0/0. The right-hand sides must then
be regarded as the limits as v -*■ n. These formulae can be taken
as the definitions of Neumann and Hankel functions.

11.4 THE FUNCTIONS Iv(z), Kv(z), etc.

11.4.1 The third class of cylindrical functions can be defined

as follows:
7„(z) = i~vJv{iz) (51)
71

Kv(z) = - H y \ i z ) c 2' (V+i) (52)

Since Jv(z) and H ^ i z ) are linearly independent, it follows that

7„(z) and Kv{z) are also linearly independent. The solutions of the
equation

— zjw = 0 (53)

which is obtained from Equation (1) of this chapter by substituting

z = /£. It is clear that

K (54)
k=a r ( k + v + \ ) r ( k + \ )
258
 Fig. 11.7

17*
 Mathematical Physics
Using Equations (47) and (49), it is readily shown that for large
\z\ and | arg z | < n j l —d,

(55)

(56)

From the properties of the zeros of Bessel functions it follows that

/„(z) has an infinite number of simple, purely imaginary zeros. At
the point z = 0 the function Kv{z) has a singularity of the form
z~v, v 0, and at v = 0 it has a logarithmic singularity. There are
recurrence relations for these functions which correspond to the
formulae for /„(z) and H^l)(z). Plots of Kv(x) and f„(x) are given
in Figs. 11.6 and 11.7.

11.4.2 The following cylindrical functions

ber„(z), bei„(z), kern(z), kei„(z)

are also encountered in applications. For real values of the argu­

ment a-, they are defined as follows:

bern(.Y) = R e[/„(x|/—/)], bei„(.Y) = Im [/n(.Yl/'—/)] (57)

kern(x) = Re[Ar„(x]/—/)], kein(.v) = Im[A'„(.v| - / ) ] (58)

and then analytically continued to the entire plane of the variable z.

These definitions can be used to derive properties analogous to
the corresponding properties of the functions /„(z) and Kv(z). The
derivation of these properties is left to the reader.
It can be shown that

(59)

(60)
=o
260
 Cylindrical Functions

1
beri(z) (61)
I 2 Z nl(n-\-1)!

1
beijCzJ (62)
w n=Q
where E\y\ is the integral part of the number y.

11.4.3 Example 1 Determine the steady-state temperature dis­

tribution in a uniform circular cylindrical rod of length h and radius
R, whose ends are maintained at zero temperature and whose
lateral surface is kept at a temperature f(z).
The mathematical formulation of the problem is as follows.
It is required to find the solutions u(r, z) of the equation Ai/ — 0
for 0 ^ r < R, 0 < z < h satisfying the boundary conditions
u(r, 0) = u(r, h) = 0, |k(0,z)\ < oo, u(R, z) = f{ z)
We shall seek solutions satisfying the first three boundary condi­
tions in the form u = 0(r)lIJ(z). For the functions 0(r) and F(z)
we shall have the following problems:
= 0, W(0) = F(h) = 0 (consequently I > 0)
1
0' 0 ' —A0 = 0, 10(0)! < oo (63)

It is clear that
2 2
net 7in
W„(z) sin ^ z, /„

The general solution of (63) (for / = /.„) can be written in the

form

<Pn(r) = C J o ^ r j + D ^ o i ^ r j

nn \

set D„ = 0. Therefore,
(—j^-r I becomes infinite at r = 0, we musi

nn
0 n(r) = C J 0

261
 Mathematical Physics

The solution of the original problem will be sought in the form

. sin
sin— z
h

where Cn is determined by the last boundary condition (assuming

that the function/(z) can be expanded into a Fourier series in terms
of the eigenfunctions sin (Tin/h)z in the range (0, /?)):
h

Example 2 Determine the steady-state temperature distribution

in an infinite uniform flat plate of thickness h having a circular
aperture of radius R if the faces of the plate are maintained at zero
temperature and the walls of the aperture are kept at a tempera­
ture /(z).
The mathematical formulation of the problem is: it is required
to find the solution u(r, z) of the equation Aw = 0 in the region
7 ? < r < o o , 0 < z < / i satisfying the boundary conditions
u (/•, 0) — u (r, h) = 0, |w(oo, z)l < oo, u (R ,z)= f(z)
We shall seek solutions satisfying the first three boundary conditions
in the form For the functions &(r) and ^(z) we
have the following problems:
t/ y"+ A li / = 0, !/y(0) — ¥(K) = 0 (consequently A > 0)

&" + — 0 '- A 0 = O, |0 ( o o ) |< o o (64)

/•
It is clear that

The general solution of (64) (for A = A„) can be written in the form

262
 Cylindrical Functions

Since the required solution must be bounded in the region

R < r < oo and the function I0 \^j~ r j *s not bounded in this range of
/•[since for large values of r it behaves as | / /;/2^r2« r e(Jin/,,)r(l + 0(l//-)],
we must set Dn = 0. Hence

<£„(/•) = C„K0
? ')
The solution of the original problem will therefore be sought in the
form

u(r, z) = V C„JU — r) sin— z

n=1
where
, , s . nn
c n = y(z) si n -^ -z d z
hK0\ — R\

11.5 AIRY FUNCTIONS

Many problems in physics, for example the problem involving the

motion of a charged particle in a uniform electric field, lead to the
equation
y " —x y = 0 (65)
Substituting
]/xz(x) for x > 0
y(x) =
] / —x z(x) for x < 0
we obtain the following differential equation:

Z" + T z-w _ ( J ? + .lc) z = o ( 66)

To derive the general solution of this equation, let us make the

further substitution

--X '.3/2 for x > 0

t
< -V) 3/2 x |3/2 for x < 0
3
263
 Mathematical Physics

Equation (66) will then become

d 2z 1 dz
-U - + 1 z = 0 for x > 0
dr2 t dt
d2z 1 dz
1 2 z = 0 for x < 0
d t2 t dt
These are the equations for cylindrical functions. Their general
solutions can be written in the following form:
z(t) = C:I_v3(t)+CzI ui(t) for x > 0
z(0 = DiJ_il3(t)+DzJ v3(t) for .V < 0
Consequently, the general solution of (65) can be written in the
form

for x > 0
y& =
for x < 0

If we set

C\ — —C2 = D x = D z =

and

Ci — C2 — D\ — —Do — —

we obtain the Airy functions Ai(x) and Bi(x):

1'
'-1 /3 -A * 4-V» for x ^ 0
3
Ai(x')
l7W
J-m yl.vlw +A» f |.v H for x < 0

for a' ^ 0

for .v < 0

264
 Cylindrical Functions

From the power-series definitions of Iv(t) and Jv{t) it follows that

A i ( x ) - + - i = --------- , Bi(x) tt= ---------

*V ^18 r (2/3) *-o j / l 8 F1(2/3)
Using the same procedure as in Section 11.3.5, we can readily
show that
Ai(x) = — a-‘ 1/4e"2/33[3/2[1 -f 0 (x " 3/2)] for x -> + oo
2\ /n

Ai(x) = - ~ x ~ 1/4 s i n |y |x|3/2+ y j [l + 0 (|x |-7/4)]

for x -*■ —oo

/ /y

Bi{x) = - _ -x “ 1/4 e2/!,3,2[l fO(-'f‘ w)J for x -> +co

y 7i

W(x) = - L i * - 1'4 s in ( - ||x |3" + j J [ l+ 0 ( |x |- W ) l

for x -*■ —oo

The Airy functions have been tabulated (see bibliography).

PROBLEMS

1. Find the temperature of an infinite circular cylinder whose

initial temperature is u(r, 0) = A ( l —r2/R2), and whose surface
is maintained at zero temperature.
2. A homogeneous cylindrical conductor of radius R is heated
for a considerable time by a constant current R. Investigate the
cooling of the conductor after the current has been switched off
if the surface of the conductor loses heat in accordance with
Newton’s law and the surrounding medium is kept at zero tempe­
rature.
3. A constant magnetic field H0 is switched on outside an infinite
circular conducting cylinder 0 < r < R, of conductivity o, at time
t = 0. The magnetic field is parallel to the axis of the cylinder.
Find the magnetic field inside the cylinder if the current is initially
zero. Determine the flux of magnetic induction through the cross-
section of the cylinder.
4. Find the temperature distribution in a cylindrical tube
R x ^ r ^ R2 & heat flux of density q is introduced at time t = 0
265
 Mathematical Physics
through the outer surface of the tube whilst the internal surface
is kept at zero temperature. The initial temperature is zero.
5. Determine the vibrations of a circular membrane with fixed
edges under the action of a uniformly distributed load Q — const,
applied on one side, beginning with t = 0.
6. Solve Problem 5 for (a) Q = A sin cot, (b) Q = A cos cut
and (c) the load Q distributed over the ring < r < Rz (consider
also the case R { = Rz).
7. Determine the vibrations of a circular membrane 0 -<^ r < R,
whose edge moves for t > 0 in accordance with the laws:
(a) u(R, t) = A sin cot, (b) u(R, t) = A cos cot. There is no initial
excitation.
8. Determine the vibrations of a circular membrane 0 ^ r ^ R
with fixed edges under the action of a point impulse P, applied to
the membrane at the point (r0, tp0) at time t = 0.
9. Find the electrostatic potential inside a hollow cylinder of
radius R and height h whose lower base (z = 0) and lateral surface
are kept at the potential V0 whilst the upper base is kept at the
potential Vu
10. A constant current / enters through one end of a cylindrical
conductor made of material having a conductivity a and leaves
at the opposite end. Determine the potential distribution inside
the conductor, assuming that the end contacts are discs of radius
R\ < R, where R is the radius of the cylinder, and that the current
is distributed in the conductor with constant density.
11. A thin wire, heated by constant current and liberating O
units of heat per unit length is introduced into a cylindrical specimen
of radius R and height h. Determine the temperature distribution
in the specimen, assuming that the lateral surface of the cylinder
is kept at zero temperature and its ends lose heat in accordance
with Newton’s law to the surrounding medium at zero temperature.
12. Find the temperature distribution in an infinite circular
cylinder 0 < r < R if its initial temperature is ito — const and
a constant heat flux of density q is applied to its surface beginning
with time t = 0.
13. Determine the natural oscillations (i.e. find the eigenvalues
and eigenfunctions) of a circular cylinder of length /; subject to
the boundary conditions of types I, 11, and III, respectively.
14. Determine the natural vibrations of a membrane in the
form of a circular sector ( / '< / ? , 0 < cp < a) for types 1, II and III
boundary conditions.
266
 Cylindrical Functions
15. Find the temperature distribution in an infinite cylindrical
sector 0 < r < R, 0 < <p < a, whose surface is kept at zero
temperature; the initial temperature is arbitrary.
16. Find the temperature distribution in a finite circular cylinder
whose surface is kept at zero temperature and where the initial
temperature is arbitrary.
17. A circular membrane of radius R is loaded with a point
mass m at its centre. Find the eigenvalues of this membrane.
Compare these with the eigenvalues for m = 0. Consider also
the two cases m small and in large.
18. Find the coefficients of the expansion of the function
e*’2(r- i/r) jn integrai powers of t.
19. Evaluate 7±1/2(x), K±m{x), H % 2(x), N ±m(x).
20. Find the steady-state distribution of the concentration of
an unstable gas inside an infinite circular cylinder if the concen­
tration on its surface is maintained at the constant value u0.
21. Solve Problem 20 for the region outside the cylinder.
22. Find the electrostatic field inside the cylinder 0 < r ^ R,
0 < z < h, whose ends and lateral surface are maintained at
potentials ul} u2 and w0, respectively.
23. Find the steady-state temperature distribution in a circular
cylinder of height h, whose lower face is thermally insulated and
whose upper face loses heat in accordance with Newton’s law to
the medium at zero temperature. The lateral surface is maintained
at temperature u\r=R = f(z).
24. The wall of a cylindrical channel drilled through an infinite
flat plate of thickness h is maintained at the temperature u0 = const.
Find the steady-state temperature distribution in the plate if its
faces are maintained at zero temperature.
25. Find the temperature distribution in a cylinder (0 < r < R,
0 ^ z < h) if its initial temperature is zero, and if from time t — 0
onwards the base of the cylinder z = h is maintained at the tem­
perature uQ= const, while the remaining part of the surface is
kept at zero temperature.
 12

Spherical Harmonics

The simplest class of spherical harmonics is that of the Legendre

polynomials P„(cos0). In this chapter we shall be concerned with
their properties.

12.1 LEGENDRE POLYNOMIALS

12.1.1 The simplest and most rapid derivation of these poly­

nomials is to obtain them by means of a generating function.
Let us expand the function lP (x, t) = (1—2 x t + t 2)~1/2 into
a power series in terms of t. We have
'/'(*, t) = P0(x) + Pl(x)t+ ...+ P n(.v)P+ ... (1)
It will be shown below that the coefficients of this expansion,
P„(x), are the Legendre polynomials.
The function lP(x, t ) is called the generating function for the
Legendre polynomials. If we set x = 1 in the expansion (1), we
obtain

'//(1, 0 = 1 ^ 7 = i + f + •
Consequently, P„(l) = 1.
Substituting .v = —1 into (1), we obtain

n -i,n =

268
 Spherical Harmonics

Consequently, P„(— 1) = (—1)". Tt is clear that

1 dnW
Pn(x) = ( 2)
n\ dtn t=o

On the other hand, the /7-th order derivative with respect to t of

the function W for t = 0 is given by
dnT n! c n v -5 ) ..
( 2i )
dtn t=o

where C is a closed contour surrounding the point £ = 0. In the

integral given by (2i), let us substitute
j/1 - 2 a £ + £ 2= l - £ z
We obtain
1
2nn\ (z2- ! ) "
Pn(x) = dz (3)
( z - . y) " +1

where Cx is a closed contour surrounding the point z = x.

Using the formula for the /7-th derivative of the Cauchy integral,
we obtain

(4)

Therefore, P„(x) is, in fact, the Legendre polynomial of order n.

It follows from Equation (4) that P2k(z) and P2k+\(z) a r e e v e n
and odd functions, respectively. It is clear that

Po(x) = 1, Pi(x) = x, P2(x) = --.v2— —

12.1.2 To obtain the differential equations for P„(x) consider

the function w = (x2— 1)". It is clear that
2nxw
w' - 2nx(x2— \)n 1
I
or
(a'2—\ ) w ' — 2nxzv = 0
Differentiating the above identity n + \ times, we obtain
(a2- 1) [w(n^}" + 2x[w(n)}'-n{n V l)w(n> =
269
 Mathematical Physics

Therefore, the function w{n){x) and, consequently, Pn(x) [since

Pn(x) = (\/2nn\)win)(x)] satisfies the equation
( l - x 2) y " - 2 x y ' + / . y = 0 [A = « (n+ l)] (5)
This is the Legendre differential equation. It can also be written
in the form

d T [( l“ *V l + ; ° ' = 0 W
The second solution of (5), which is linearly independent of
P„(x)> has a logarithmic singularity at the point x = ± 1 , in accord­
ance with the theorem in Section 11.1.2.

Remark The Legendre polynomials can also be generated in

another way. One can seek the solution of Equation (5) which is
bounded in the range [—1,1] in the form of the power series
00

X ck*k- For / = «(«-!-1) this series cuts off at the term of power n,
k=0
i.e. for / = n(n-\-1) the solution will be a polynomial of order n,
i.e. P„(x). It differs from the Legendre polynomial of order n by
a constant factor. This factor can be chosen so that P„( 1) = 1.

12.1.3 The Legendre polynomials are orthogonal in the range

[—1, 1] with weight p(x) = 1, i.e.
1
J P„(x) Pk(x) d.v = 0, if n ^ k
-1
In fact, consider the two identities

- - [ ( l - . Y 2)P ;] + «(// + l)P n(.v) EE 0

- [ ( l - . v 2) P i ] + I ( H l ) P t (.v) = 0

Let us multiply the first of these by Pk(x) and the second by Pn(x).
Let us then subtract one from the other and integrate the result
with respect to x within the range [—1, 1]. The result is

-i v
1
= [k(k+ !)—«(« + !)] S p n(x)Pk(x) dx

270
 Spherical Harmonics

or
i l
\ - - { ( 1 -X^iP'nPk-PnPL)} d-Y= (k ~ n ) (A'+ /J+ 1) [ Pn(x) P(x)dx
J d.v ,
-i ~L
Consequently,

j, = (A;-,)(l+B+l) W-. =0
when n ^ k.

12.1.4 We shall now establish the validity of the following two

recurrence relations:

(w+l)Pn+i(*)—(2 « + l)x P n(x)+H Pn_i(x) = 0 (6)

p .(x) = r d — [/>;+1(x )-P :.,(.v )] (7)

Zn-\-1
Differentiating the expansion given by (1) with respect to t and x
we obtain
dP ( x - t)F
Pl+2P2t + ...
dt l-lxt-p t2
dP tP
P'Q+ p [ t + . . . + P'nt n+ . . .
dx 1—2 xt-\-t2
or

(x —r)(P0- f -Pi t + ... + Pnt n+ •••) = (1—2 x r + r 2)(PH -2P2t + •••

+ n P nt n~1+ ...)

?(P0+ Pl^+ ••• + Pn^"+ •••) = (1 —2xt+ ^ 2)(Po + P l/ + •••

+ P nt"+ -)

Comparison of the coefficients of equal powers of t in these identities

yields
(/i+ l)P n+1(x)—(2rc+ l)xP n(x)+«P „-i(x) = 0 (6)
and
Pn(x) = P'n+i ( x ) - 2 x P 'n( x ) P P ; ^ ( x ) (8)
271
 Mathematical Physics

D iffe r e n tia tin g (6), w e o b ta in

(n+ l ) P U l( x ) - ( 2 n + l ) P „ ( x ) - ( 2 n + \ ) x P ' ( x ) { - n P ^ ( x ) = 0
If we now eliminate x /*„(+) from this result and from (8), we obtain
the identity given by (7).
Remark 1 Using (6) and the formulae
P0( * ) = l , Pl(*) = *
we can determine all the Legendre polynoriiials.
Remark 2 The relationship given by (7) enables us to express
the integral of the Legendre polynomial $ P„(x)dx in terms of the
polynomials P„+i(x) and Pn-i(x)-
12.1.5 Let us now evaluate the square of the norm ||Pn||2
i
= $ Pl(x)dx. If we express one of the factors in the integrand in
terms of P n_j and Pn- 2, in accordance with (6) and replace n by
n — 1, we obtain

IIP-1 Sp - p ’ 6 x =
-i
\ p^ r L
-l
x K - ^ L pA ix

2n-l
jj xP„Pn^ d x
n
where we have used the orthogonality of the polynomials P„ and
Pn_2. In the last integral the product xP n can be expressed in terms
of P„+1 and P„-1 in accordance with (6). This yields

2 2/z—l 1* jrc+ 1 n
l,p”L U " - 'i 2 ^ + T n + , + 2„ + i

In- 1 f
P L i d.v
2/2+1 J
or
2 / 2-1
LPJI2 = n — 1 li (9)
2 / 2+1
Here again, we have used the orthogonality of the polynomials
P--i and Pn+t.
272
 Spherical Harmonics
From (9) we have

3-iiPii;: 2
\\P k\\2 ( 10)
2k+\ 2k+l
and hence ||Pil|2 = 2/3.

12.1.6 Theorem All the zeros of the Legendre polynomials

are simple and lie in the range (—1, 1).
We shall now prove a more general theorem. We shall say that
the polynomials {q„(x)} form a normal system if they include the
polynomials of all non-negative orders.

Theorem If the polynomials {<?„(x)} [qo(x) = 1] are orthogonal

in the range (a, b) with weight p{x) > 0 and form a normal system,
all the zeros of the polynomial qn(x) are simple and lie in the
interval (a, b).

Proof Since the polynomials q„(x) are orthogonal, we have

(for n > 0)
b
$ 1 • qn(x)p(x) dx = 0
a

Consequently, q„(x) changes sign k times in the interval {a, b)

(k 5* 1).
Suppose that this occurs at the point x t , x 2, ..., x k. We then
have qn(x) = (x—x1)(x—x2)...(x —xk)<p„(x), where <p„(x) does not
change sign in (a, b). It is clear that to establish the theorem it is
sufficient to show that k = n.
Let us suppose that k < n. We then have the expansion
R k(x) = (x—x 1) { x —x 2).--(x—x k) = r a kqk(x)
b
in which ak # 0. It is clear that ]j qn{x)Rk{x)p{x) dx = 0 since
a
qn and qr are orthogonal (r = 0, 1, 2, ..., k). On the other hand,
b b
0 = J qn(x)Rk(x)p(x) dx = $ Rl(x)tpn(x)p(x) dx > 0
a a

We have thus reached a contradiction. Consequently, k = n. This

theorem leads to the theorem on the zeros of the Legendre poly­
nomial. Moreover, the zeros of these polynomials are arranged
symmetrically with respect to x = 0.
18 273
 Mathematical Physics

12.1.7 The Legendre polynomials can be represented by

1
Pn(x) = 2“ j [*+* V 1~ *2sin^]" dcp (11)
-i
To obtain this integral representation, let us take C, in Equation
(3) to be a circle of radius R, R = 1—x2 (|a| < 1) centred on

Fig. 12.1

the point z = x and then let us substitute z = .v+ | T —.v2e,(p so

that dz = / ]/1 —x2t iq>dcp. We have
z2—l = a-2—1-r (1 —A'2)e2,?,+2.v p 1—.v2e'<p
= ]/1 —a-2 e'?’[2a*H- l / l - a - 2 (e '^ -e -^ )]
= 2 )^ 1—A'2 e,,p[a+J ]/1 —A'2 sin 9?]
Substituting for z—a%z2~ 1 and dz in (3), we obtain
2.1

Pn(x) = — \ [a:+ / \ / \ —A-2 sin9?]" d<?

2n6
It follows from this that
|P„(a-)| < 1 where .v 6 ( - 1 , 1 ) (12)
Fig. 12.1 shows graphs of the first few Legendre polynomials.
274
 Spherical Harmonics
12.1.8 Example 1 Determine the potential inside a hollow
sphere of radius R which consists of two hemispheres insulated
from each other and charged to potentials v, and vz.
The mathematical formulation of the problem is as follows.
It is required to find the solution u(r, 0) of the equation V2u = 0
in the region 0 < r < R which satisfies the boundary conditions
71
®I , 0 < 0 < —
w(0, 0)| < o o , u(R, 0) =
71
V2 , -y < 0 < 71

Solution Let us find, to begin with, bounded solutions of V2u = 0

of the form u = f(r) y'(0). Separating the variables, we obtain
_d_ 1
— O ' sinO)
dr 0-2f ) sin 0
= X
f V
and, consequently.
_ l ___ d_
dr
(ry ' ) - / / = o , sin 0 dO
(y/ sin 0)-\-Xip = 0

In the last equation let us substitute £ = cos 0, so that

d2ip
(i-a \-Xy> = 0
d£2
which for / = n ( n + 1) has a bounded solution in [—1, 1] in the
form of a Legendre polynomial P„(£). For such values of X the
equation for /( r) has a bounded solution of the form f(r) = r".
The solution of the original problem will now be sought in the form

w0 \ °) = ^T l cnrnPn(cos 0) (13)
nZ0
The coefficients c„ can be determined from the second boundary
condition using the orthogonality of Legendre polynomials:

cn ^ u(R, 0)Pn(cos0) sinG dO

o
o i
2nf
^ B(f)df+*>i d£ )
2

18* 275
 Mathematical Physics

The last two integrals can be evaluated using Equations (7) and (6)
of this chapter. The final result is

2/7+1
cII Pn+m
2 n
In this problem we have used the following theorem on the
expansion of the function 97(f) into a Fourier series in terms of
the Legendre polynomials:
If a function 97(f) is piecewise-continuous together with its
first-order derivative then at each point at which 97(f) is
continuous, its Fourier series in terms of the Legendre polynomials
converges to this function.
We shall not prove this theorem here.

Example 2 Expand the plane wave v = eiAz into a series in terms

of the Legendre polynomials and Bessel functions.

Solution The function v = e'Az = e,Arcos0 is a solution of

Av-\-?.2v — 0
Using spherical polar coordinates, this may be written in the form
1 d_ 1 d
(r2v,)+ (Vo sin 0)+22t/ = 0
dr r2 sinO dO
We shall seek the solution of this equation in the form
v = f(r)v(0)- Separating the variables we obtain

1
^ di

1
(y/ sinO)+/<y/ = 0
sinO dd
Bounded solutions of the equation for ip are obtained when
H = //(/7+1) in the form of the Legendre polynomials
y){0) = P„(cos 0)
Substituting
9 + -)
/(r) =
\ r

276
 Spherical Harmonics
we obtain
>2 n

t/ \ t ■ <p = 0
<J> ^-------- 9" -t-

The bounded solution of this equation is the function Jn+yityr).

Therefore, the equation which the travelling wave under consid­
eration must satisfy has a family of solutions of the form
(l/|'V )/n+1/2(/.r)Pn(cos d). It is thus natural to set

gip cos 0
-% Jn +l/2(p)Pn (COS6) (14)
Z
»= 0 VP

Using the orthogonality of the Legendre polynomials we find

that

l» ± y M = .+ ± 1 $ ew/».{{)df
IP 2
Integrating n times by parts on the right, we obtain

2 c » ^n + l/2 (P ) _ 1 cIpjp ^
[e ^ (W )]L ,
2/1+1 | p iP oipf

This result is valid for any p. For large p we can replace the function
yn+1/2(p) by its asymptotic representation. The result is

2c„
]/ 2 j 1 \ 71
cos p - /! +
2« + l PI 71 I

2 |2 c „ ^
j/ir(2 /i+ l) P

eip*Pn(!d
IP
277
 Mathematical Physics
This shows that
71

2 l/2 c"sin\ P - " 2 1 .

' — !-----------
\/n 2/?+ l i
or

2 l/2 c»sin[ p - " 2 ,

1I n 2/7+1

I *
— 2/"sinlp—n —

Hence, we find that

+ = | / y i ,,(2n + 1) = l /2^/n( « - - ) (15)

Example 3 Determine the perturbation of a plane acoustic wave

ito(M, t) due to the presence of a sphere of radius R with perfectly
rigid walls, i.e. consider the scattering of sound by a sphere.
We shall suppose that the centre of the sphere lies at the origin
of the coordinates. Motion outside the sphere will be described
by the function u(M, t ) = w0(M, t)+v(M, t), where v(M, t) is the
required perturbation. Since the functions u(M, t) and u0(M, t)
are solutions of the equation

a2Au — utt
it follows that v(M, t ) will also be a solution of this equation.
The functions a, u0 and v will be interpreted as velocity potentials.
On the surface of the sphere we must then have = 0.
Therefore, the problem fori/ can be formulated as follows:

a2Av = vtt where r > R

dv ditp
V < CO
dn r —R dn r= R

278
 S p h e r ic a l H a r m o n ic s

It is clear that one can choose a Cartesian set of coordinates so that

the plane wave u0 can be written in the form
«b = Qikze~ikat
We shall seek v(M, l) in the form v = <P(M)e~ikat. The problem
for 0 ( M ) can then be formulated as follows:
A<P-{-k2& = 0 where r >R
d&
|<Z>| < oo
dn r=R dn \r=R ’
where <p0 — e‘fc\
In view of the expansion given by (14), it is natural to seek
&(M) in the form of the series
00
0 =
m=0
where 0 m(M) is the solution of problem
A0 m-\-k20 m — 0 where r> R (15i)
d& \ d
= - - . - { A m(r)Pm(cos6)}r=R, \<?>m\ < o o (152)
(y/i p — OF
where A m{r) Pm{cos 0) is the term of number m in the expansion
(14). If we assume that 0 m = Bm(r) ^,,,(0), we see that, in view of
the boundary condition (152), the function should be equal
to Pm(cos 0).
Substituting 0 m = Bm(r) Pm(cosO) into (150, we obtain

1)1
Bm Bm = 0
r2 \
If we now substitute Bm = Dm/[/r, we obtain

D ••
mT
l k2 (15a)

This is the differential equation for cylindrical functions with

index v — m-\-1/2. A physical interpretation of the problem suggests
that the function v(M, t) should be a superposition of spherical
diverging waves of the form (1 /r)c,k{r~at\ Since for large r the
279
 Mathematical Physics
asymptotic behaviour of this kind is exhibited only by the Hankel
function
71
ik
e

the general solution of (153) must be written in the form

Dm(r) = < ^H £ lvz(kr)+pmH g l ll2(kr)
and only the first term need be retained (it represents a converging
wave). Therefore, we have

Bm(r) = amJim(Icr) where hv(kr) = —^ = H i l m (kr)

ykr

The coefficient ocm can be found from the boundary condition (152)
which yields
BUR) = - a 'U R )
Hence, we find that

-A 'U R ) j'm(kR) nrukR )

GCm— — Cn = - ] /2 nr m
h'UkR) k h'UkR) 2 j h'UkR)
wherej m(p) = (l/|/p ) Jm+i/iip)- Consequently,
00

j'mikR) h j k r )
&(M) = ^ ®UM)
w=0 m —0 ' '
h'UkR)

The functions Pn(cos 0) are a sub-class of a much broader class

of spherical functions which can be defined with the aid of the
associated Legendre functions.

12.2 THE ASSOCIATED LEGENDRE FUNCTIONS

12.2.1 In Section 1 of this chapter we showed that the Legendre

polynomials were solutions of Equation (5) for / = « (n + l). In this
section we shall consider the more general equation

( \ - x 2) y " - 2 x y ' + = 0 ( 16)

where k is a non-negative integer.

280
 Spherical Harmonics

Solutions of (16) which are bounded in [—1, 1] are called the

associated Legendre functions. They can be found by substituting
v = ( l - x !)*'2z(.v) (17)
so that

( 1 - . y2) z" —2 x (A:+1)z' t [ ^ - ^ ( ^ t 1)]z = 0 (18)

or

~ [ ( l - x ^ z ' ] ^ [ X - k ( k + \ ) ] ( \ - x 2)kz = 0 (180

The same expression can be obtained from Equation (5) by

differentiating it k times. The solution of (18) and (18i), which
is bounded in [—1,1] for )<.= n(n-r1), is the function
dfc
z(x) = j - r Pn{x). We thus have the following identity:
d.v

d_ dfc+1
( l - ^ ) * +1 Pn(x)
dx dxft+I

= ~ [n(n+l)-k(k-'rl)] ( l - x 2)k- ~ P n(x)

= ( n - k ) ( n + k + l ) ( l ~ x 2)k- ^ Pn(x) (19)

Consequently, the solution of (16), which is bounded in [—1, 1]

for I = «(«+1), i.e. the associated Legendre function is given by
d*
P k(x) = (1 _ x 2)ft/2— ^ Pn(x), 0 < k < n(20)

It is clear that P%(x) = P„(x).

12.2.2 Henceforth, we shall require only one of the properties

of these functions, namely, their orthogonality.
The associated Legendre functions are orthogonal in the range
[—1, 1] with weight p(x) = 1:
i
§ P%(x)Pk(x)dx = 0 where n/ s ( 21)
-i
281
 Mathematical Physics

Proof Substituting

<s = S

and using Equation (20), we obtain

A k„ , s = J ( l - ^ ) k^ P n( x ) - ^ P s ( x ) d x

dfc ifc-i
= C1- ^ ) dxh

r dk~l , d
d.x
(1~ X ) d ^ P"(x)

Substitution of the limits of integration in the first term yields

zero and, therefore.

f d*~! d
Ai s = - ) -j z t t W n(-V)]<
d.x d.x
-1
The second factor in the integrand can be transformed with the
aid of (19) (replacing k by k — 1). The final result is
l
< • = ( n - k + l)(n+fr) jj
2i '

or
A k„,s = (/i - k + \) { n + k ) A kn:? (22)
Using this last expression for A kf , Ak~ f and so on, we obtain

4k _ (»-r/c)! a0
Ak
CM n s ----- -
— (23)
(n-k)\ n's
Since

A°„,s = $ P„(x)Ps(x)dx = 0 if / U s (24)

- i

282
 Spherical Harmonics

and

Al,„ J Plix)dx 2/;+1

it follows that
1
A k„,s = J P*(x)Pks(x)dx = 0 for n^s
-i
and
(«+£)! 2
= ILP„fcli2 = ( n - k ) \ 2n+l
Remark It follows from (21) and (22) that the /c-th order deriva­
tives of the Legendre functions are orthogonal in the range
[—1, 1] with the weight p(x) = (1—x 2)k.

12.3 SURFACE HARMONICS

12.3.1 We shall seek the solutions of the Laplace equation
V2w = 0, expressed in spherical polar coordinates r, 0, cp, in the
283
 Mathematical Physics
form F(r), 7(0, <p) where the differential equations for F(r), 7(0, <p)
are

(r2F')—XF = 0 (25)
dr

1 d I . . d Y \. 1 d2Y
sin0 + -XY = 0 (26)
sin0 dO d6 sin20 dtp2

Definition The solutions of (26) which are bounded for 0 < 6 < zi,
0 < <p ^ 271 and are such that Y(0, tp-\-2n) = 7(0, <p) are called
surface harmonics (or generalised spherical harmonics).
If the bounded solutions of (26) are sought in the form 0(0) 0 (99),
<P{(p-\-2n) = 0(tp), the differential equations for 0 (0) and 0 (99) are
1 d f*
(0 ' sin0) + X 0= 0 (27)
sin0 d0 sin20
0"-Yn& = 0 (28)
Since the function 0 (9?) is periodic, we find that ju — k z, where k is
an integer and, therefore,
0 (9?) = A cos^9?+ S s\nk<p
If we substitute cos 6 = £ into (27), we obtain
d 2W AW k2
0= 0 (29)
r+ 1- f 2
which is the same as (16). When X — n(n-\-1), this equation has
the solution 0 = P£(|), which is bounded in [—1, 1]. Consequently,
surface harmonics of the form 0 ( 0) 0 (99), where 0 (99+ 2:1) = 0(<p)
are given by
Y k„(0, 99) = P kn(cos0) sink <p
and
f„"l ( 0 , 9?) = P£(cos0) coskcp (30)
Y°„(0, <p) = P°(cos0) = P„(cos0)
These functions are called the fundamental surface harmonics
of order n. It is clear that the functions

yn(0,9>)= 2
k=>—n
c *y "(0>9O (31)

284
 Spherical Harmonics
are also surface harmonics. They are called surface harmonics of
order n. When 2 = n(« + 1), Equation (25) has solutions of the
form
1
F\(r) — /•" and F2(r)

and, consequently,
M r , 0,<p) = rnYn(0, q>)

M r , 0,(p)= Y„(0, <p)

are harmonic functions. They are called spherical functions of

order n.
It follows that surface harmonics of order n, Y„(0, <p), are the
values of spherical functions of order n on a unit sphere.

12.3.2 Surface harmonics are orthogonal on a unit sphere:

\ \ Y„(0, <p)Ys(0, cp) da = 0 if s

Si n
or
2n n

J J Y„(0, <p) Ys(0, <p) sin0 dO d<p = 0 (33)

oo
To prove this we note that the fundamental surface harmonics
are orthogonal:
2 71 7t
<
\j ] Y kn(0,<p)Y§(0,<p)sm0d0d<p = O for (n ,k)^(s,p ) (34)
oo
since
2n n 2rz 1

J \ Y k(0, <p)Y£(0, <p) sinOdOdy = jj cosky cospy d<p jj

oo o - i
= 0
when (n, k) ^ (s, p) (to be specific, we have assumed that k > 0,
p > 0). If k # p, the first integral on the right-hand side is equal
to zero. If, on the other hand, k = p but n # s, the second integral
will be equal to zero.
The orthogonal property given by (33) follows from the ortho­
gonality of fundamental surface harmonics and from Equation (31).

285
 Mathematical Physics
The square of the norm is given by
271 71 2.1 1

^nll2 = S 5 [ W , 9 O ] 2sin 0 d0 d p= $cos2^ d ? $ [P„k(f)]2<^

00 0 -1
and, consequently,
fc 2rr{ n - k ) \ 1, k * 0
(34')
J "" 2 n + l (n-k)\ ’ 2, k = 0

12.3.3 Theorem The spherical functions rnY„(6, (p) are homo­

geneous harmonic polynomials of order n in the variables x ,y, -■

Proof Since
n

V«(0,<p)= X C “ Y kn ( 8,p )
k = —n

it is sufficient to establish this theorem for the functions r„Ykn(0, <p).

To be specific, we shall assume that k > 0. We then have

Ykn(0, tp) = Pk($) cos ktp = ( I - ^ I f ^ ^ c o s ^

dfc \ ^
= (1—£2)fc/2^ r / | aq£"~2q cos ktp
R—0

— (1 - + )fc/2 ^ hqj n~2q~k cos kcp

q= 0

where £ = cos 0. It is clear that it is sufficient to establish the

theorem for functions of the form r" sink 0(cos 0)n~2q~k cos J<rp.
For these functions we have
rns'mk0(cos0)n~2q~k cos k(p = rk s\nk0 Re(e,k,p)r2qrn~2q~k(cos0)n~-r‘~k
= Re(.v--i}f(.v2+ / + : 2)2,z " '2,_t
This is clearly a homogeneous polynomial of order n.

Example Determine the temperature at the internal points of

a sphere of radius R, whose surface is maintained at zero tem­
perature; the initial temperature was f(r, 0, cp).
The mathematical formulation of the problem is as follows.
It is required to find the solution of V2u = (1 /a2)ut in the range
286
 Spherical Harmonics

0 < r < R, 0 < 8 < 7i, 0 < (p < 2n, t > 0, which satisfies the
boundary conditions
u(R, 8, <p, t ) = 0, \u(0, 8, (p, t)| < oo
u(r, 8, cp-'r 2n, t) = u(r, 8, <p, t)
and the initial conditions u(r, <P, 0) = fir, 8, cp).

Solution We shall find solutions of the equation V2u = (1 /a2)wt

in the form A(r) Y(8, cp) B(t) which satisfy the boundary conditions
uniquely. Separating the variables we find that
B ' ~ a 2a B = 0 (35)
_ j _ ± 1 . . m j ___ i _ * r IY = 0
sinft d8 | Sm d8 I 1 sin20 dcp2 (36)
\Y(8,cp)\ < o o , Y(8,<p~2n) = Y(8, cp)

— (r2v4') —(a r2—?)A — 0, M(0)| < CO, A(B) = 0 (37)

The solutions of (36) for A = n (« + 1) are the surface harmonics

Y„(8, <p). If in the equation for A(r) we substitute A(r) = z(r)/\/r,
we obtain

z " + - z '+ = 0
r
The general solution of this can be written in the form
z(r) = M /„+i/2(]/a r)-\-N7_„_1/2(j/a r)
and, consequently,

A(r) = -)- + ^ y- ' g /(f ° ‘-'')-

l> l r
Since ^4(0) is bounded, we find that N = 0; M can be set equal
to unity.
Therefore,
A(r) = — Ja+ift([/cir)
Vr
prom the condition A(R) — 0 we obtain the equation for a:

Jn+1/2O) = 0, n = /a R
287
 Mathematical Physics
Let
MI,n >P'2,ii! •••■>Ms, n5 •••
be the positive roots of this equation. We then have aSi„ = /j 2,sJ R 2
Solutions of (37) are of the form
J
d n , s (r ) —
I r
Returning now to Equation (35) we find that
a^l,n
in '
Bs. n — C s, n 6

Consequently, the required partial solutions of the original problem

which satisfy the boundary condition uniquely are the functions
n
-in '
C s. n £ ■W v)
]>
The solution of the original problem can be written in the form
oo oo n

> ^ J „ + i ^ \ — r\[Cs.n.kYi(d,<p)
n=A _ iI /c=0
0 s— I_A *

+ Ds.n,k Y ; k(0,<p))e R2
The coefficients Cs.„.k and D3 n k can be determined from the
initial conditions using the orthogonality of the functions Yk(0, <p)
and the Bessel functions. The final result is
( « — /c)!(2/? + l )
ne(n-\-k) ! [J!hU2(/<s, n)]2R2
R 2n n
X J jj jj/‘3/2/(/-, O,9p)y„+i/2/ — !-/-Jy5(0>93)sin0drdOd95
66o ' >
n = ___ (/i—A') !(2/»~r 1)
Us" " k neR\nYk)\[J^iz(M s,n)}2
R 2.-t n
X jj J jj r3/2/■(/-, 0, y)Jn+ii?Y,s~Li \ Y„ k(0, (p) s\nO dr do dtp
00 0
288
 Spherical Harmonics

PROBLEMS

1. Calculate P„(0).
2. Determine whether the k-th order derivative of the Legendre
polynomials P2n( x \ where k is fixed, are orthogonal in the range
[0, 1]. If they are, determine the weight.
3. Solve Problem 6 of Chapter 2 for arbitrary initial data,
placing the origin of the coordinates at the fixed end of the string.
4. Determine the electrostatic field inside and outside a hollow
sphere whose upper half is charged to the potential Vx, and lower
half to the potential Vz.
5. Find the expansion in terms of surface harmonics for the
spherical charges induced in a perfectly conducting earthed sphere
by point charge e placed (a) inside the sphere and (b) outside
the sphere.
6. Consider the polarisation of a dielectric sphere of radius R
placed in the field of a point charge if the permittivity is e = ex
for r < R and e = e2 for r > R.
7. Find the potential of a simple layer in the form of a
circular disc.
8. Calculate the potential at all points of a conducting sphere
of conductivity a when a current / enters the sphere at one of its
poles (0 = 0) and leaves at the pole Q = n.
9. A sphere loses heat through its surface to the surrounding
medium which is maintained at zero temperature. A point heat
source of strength Q is placed at an internal point of the sphere.
Find the steady-state distribution of temperature inside the sphere.
10. Determine the potential due to a point charge placed
between two conducting earthed concentric spheres r = Ri and
r = R2, also determine the density of surface charges.
11. Find the steady-state distribution of temperature in a sphere
of radius R, a part of whose surface ^ (0 < a) is at a temperature
u0 = const, while the remainder of the surface S z is maintained
at zero temperature.
12. A sphere of radius R is heated by a plane-parallel beam of
heat of density q incident on its surface. The sphere loses heat
to the surrounding medium maintained at zero temperature in
accordance with Newton’s law of cooling. Find the steady-state
distribution of temperature.
13. Consider the oscillations of a gas in a spherical container
due to small oscillations of the wall, assuming that the oscillations
19
289
 Mathematical Physics

began at / = 0. The displacement of the wall is radial and given

by P„(cos 0)f(t), where /(0 ) = / '( 0 ) = 0.
14. Find the natural oscillations of a sphere subject to types I,
II and III boundary conditions, respectively.
15. Consider the cooling of a sphere of radius R, whose surface
is maintained at zero temperature. The initial temperature was
fir, <
p)•
 13

Chebyshev-Hermite and
Chebyshev- Laguerre
Polynomials

13.1 CHEBYSHEV-HERMITE POLYNOMIALS

We shall now determine two new classes of orthogonal polynomials

which have many applications. There are a number of ways in
which they can be defined; we shall derive them by the method
of generating functions, which provides the quickest route to the
main properties of these polynomials.

13.1.1 As the generating function we shall take H(x, t) = e2xt~t2

and expand it into a power series in terms of t\

H(x, t) — Hn(x) ^ (1)

n= 0

It will be shown below that the expansion coefficients H„(x) are

the Chebyshev-Hermite polynomials. It is clear that

Ot (=0
On the other hand, the n-th order derivative dnHjdtn for t = 0
is given by
dnH 2x 1 - 0
n\ f e2x
-d t
~d?~ t=o 2ni tn+ i

19* 291
 Mathematical Physics

where the closed contour C surrounds the point t = 0, or

, —( x —t)2
n\ e
H„(x) = e*2
2 ? i/J t n+1
n+
c
If we substitute x —t = £ in the last integral, we obtain

Cl

where the contour Cx surrounds the point i — x. Using the formula

for the /7-th derivative of the Cauchy integral, we obtain

Hn( x ) = ( - l ) " e j2d^ ( e ' 12) (2)

It follows from this expression that H n{x) is a polynomial of order

n. H2k(x) and H2k+i(x) are even and odd functions, respectively.
It is clear that H0(x) = 1, H {(x) — 2x, H2(x) = 4x2—2, and so on.

13. 1.2 We shall show that Hn(x) is a solution of the equation

y " —2 x y ,J\-).y — 0 where 2 = 2/7 (3)
In fact, differentiating the function w — e- *2 once, we find that
w,Jc2xw = 0. Differentiating this identity n-rl times, we obtain
[zi>(n)] " + 2 a'[w(,,)]'+ 2 n w(,,) = 0 (4)
Substituting (2) into this identity, we have
w(n) = ( - 1 )nHn(x) e - x2
/ / '' (x) - 2 x H'n(.v) J-2 n I i n(x) = 0
and so on. Equation (3) can be written in the form

~ ( e - * y ) + A e - * 2j ’ = 0 (5)
dx

13. 1.3 Theorem The Chebyshev-Hermite polynomials are ortho­

gonal in the range (—co, co) with weight p(x) = e~x , i.e.

$ H„(x)Hp(x)e-**dx = 0 if n * p ( 6)
— OD

292
 Chebyshev Polynomials
Proof Consider the two identities

- f e~**H„(x) = 0

- f [ e - * ,H;(x)]+2pe-*‘Hr(x) = 0

Multiply the first by Hp(x) and the second by H n(x), subtract one
from the other and integrate the difference with respect to x between
—oo and + o o . The result is
00

— 00
00
= 2( p - n ) 5 Hn(x)Hp( x ) e - x2dx
— 00

The left-hand side of this equation can clearly be written in the

form

jj - ^ { ( H pH'n- H nH'p)z-*2}dx
— CO

and, consequently,
CO
00
2(p—n) jj HnHpt 2dx = (HpH'n- H nH'p) z - x2
—00

Since p + n, this leads directly to Equation (6).

Let us now find the norm ||//„||. To begin with, let us establish
the following recurrence relationships:
Hn+f x ) ~ 2 x H n{x) + 2n H n_ f x ) - 0 (7)
H ’n(x) = 2 n N n_i(x) (8)
To do this we must find the relationship between the generating
function H(x, t) and its partial derivatives dH/dt and dll/dx.
Direct calculation shows that
dH dH
2{x-t)H and 2tH
dt dx
293
 Mathematical Physics

Substituting (1) into these identities, we obtain

(9)
n=l n=0

CO 00

( 10)
n= 0 n=0
Equating the coefficients of equal powers of t in (9) and (10), we
obtain the recurrence relations (7) and (8), respectively.
The identity given by (8) can be used to evaluate the integral

2 (« + l)
The identity (7) can be used to evaluate the square of the norm
OO
ll^Ji2 = s e-xlH 2n dx:
—oo
OO oo

||/7„l!2 = J e~xlH 2a(x)dx = $ H„(x)H„(x)e~x2dx

— 00 — CO

Let us express one of the factors Hn(x) in the integrand by means

of (7) in terms of H„_! and //„_2 by replacing n by n — 1. The
result is
00
|i//„||2 = 5 q~x2H„(x) {2.y//„_!(x)—2(n —1)77„_2(,v)}dx
—oo
CO
=
—00
S
e~x2H„_i(x)2xH„(x)dx

In this expression we have used the orthogonality of the polyno­

mials H„_2 and H n. If we now express 2 x H n(x) in terms of Hn_x{.x)
and Hn+i(x) through Equation (7), we obtain
CO

II/-/JI2 = \ e - ^ /7 1_1(.v){T/,1+1(.v)+2;7//n_I(.v)}d.v
—00
OO

= 2n J Q~x2H„_i(x)dx

or
|7/„||2 = 2n\\H„_i ||2 (11)

294
 Chebyshev Polynomials
In this we have used the orthogonality of the polynomials H„.^
and H n+1. From (11) it follows that
00

[|i/„||2 = 2n-1/z!H#!|]2 = 2nn\ ^ 2 x2e~x2dx = 2"n! \/n (lli)

— 00

Therefore,
\\Hn\\2 = 2"n\\/n (12)
13.1.4 The Chebyshev-Hermite polynomials form a normal
system. Consequently the theorem of Section 12.6 applies to the
polynomials. All the zeros of the polynomials Hn(x) are simple
and real.
13.1.5 The Chebyshev-Hermite functions
Hn(x) - * 2'2
rp„(x) (13)
lltf.ll C
are frequently used. They vanish at infinity. These functions form
an orthogonal system with the weight p(x) = 1:
00

5 Wn(x)xpp(x)dx = 0 if n+ p (14)
— CO

IIVn II = 1 (15)
From Equation (5) for the polynomials H n(x) one can readily
obtain the differential equation for the functions y„(x)\
y>"+(?.~x2)y> = 0 (A = 2 /i+ l) (16)
Example Determine the values of E for which the Schrodinger
equation for the linear harmonic oscillator
j 2mE m2ojl \
V + | - ^ 2 ---------------------- = 0 (1?)

has a solution which is bounded in the range —oo < x < oo.
In this equation in, oj0 , E are the mass, the natural frequency and
the total energy of the oscillator, respectively, and h is Planck’s
constant divided by 2 n . _____
If we substitute z = \/oj0m/hx into (17) this equation reduces
to the form given by (16), in which I = 2E/o)0h:
d2y> 2E
z2\xp — 0 (18)
dz2 o>0fi

295
 Mathematical Physics

When 2E/(joQf\ = 2 n + 1, where n is an integer, Equation (17) has

the solution y>„(z) — W,M' 1^ x\ which is bounded in the range
— oo < z < co. The Chebyshev-Hermite functions provide solu­
tions of the Laplace equation V2u = 0 on separation of variables
in terms of parabolic coordinates. In fact, if we introduce the
parabolic coordinates a, (3, z, which are related to the Cartesian
coordinates x, y, z by

* = y ( a 2~/32), y = cctp, z = z (19)

(where c is a dimensional factor, — oo < a < oo, 0 < /? < oo,

— oo < z < oo), then V2u = 0 takes the form

1 f d 2u d2u d2u \
V2w -c2(a2+ /?2) 0 (20)
c2(a2+ p 2) + ~df? d?\
We shall seek the solution of this equation in the form
it = A(ot) B(fi) D(z). Separating the variables we obtain the fol­
lowing differential equations:
A"-sr(]x~?.2c2ix2)A = 0 (21)
B " - { jjl+X2c2P2)B = 0 (22)
D " + / 2D = 0
where A2 and /a, are unknown parameters. Substituting £ = j/A ca
and 7] = / ] /Ac/9 into (21) and (22), we obtain
d2^
i z \A = 0 (23)
d£2 Ac
and

which is the same as Equation (17).

13.2 CHEBYSHEV-LAGUERRE POLYNOMIALS

13.2.1 As indicated in Section 12.1, we shall derive the Chebyshev-

Laguerre polynomials with the aid of a generating function. We
can take this to be
V
L a(x, t) = e i-t a > —1
( i - t y +l
296
 Chebyshev Polynomials
and expand it into a power series in terms of t :
00
La( x , t ) = Y JLl(x)in (24)
m=o

It will be shown below that the coefficients of the expansion are

the Chebyshev-Laguerre polynomials. These are occasionally
called the generalised Chebyshev-Laguerre polynomials, whilst
the polynomials L°n(x) = L n(x) are called the Chebyshev-Laguerre
polynomials. It is clear that
1 dnLa(x, t) 1 { L \ x , t)
L*n{x) =
n\ dt" 2 7ii ln+l
1=0 c
where C is a closed contour surrounding the point t = 0. Let us
substitute t = \ —xjz in this integral. We obtain
1 f zn+xe~x 1 d"
n+a e“x)
L' (X) = X" eX2 ^ i \ TdZ = ~ ^ X~‘ dxn

The contour Cx surrounds the point z = x. We can then use the

formula for the derivative of the Cauchy integral. Therefore,

K(x) (xn+a e~x) (25)

It follows from this formula that Lx(x) is, in fact, a polynomial

of order n. It is clear that L%(x) — 1, Lxn(x) = 1 + a —x.

13.2.2 We shall show that the polynomial L x(x) is a solution of

the equation
* / ' + ( “ + 1 —x ) y ' + l y = 0 (26)
or

- ^ - ( x a+1e Xj')+Ax®e xy = 0 for X= n (26j)

In fact, differentiating the function to = x n+ae~x once, we obtain

w' = (nJrOL)xn+a~l e“x—.\n+<Ie“x
and this yields the identity
xw'—(n+tx—x)w — v
297
 Mathematical Physics
Differentiating this /z+1 times, we obtain
A-[ze(n)],,+ ( x + l - a ) [ w (n)],+ ( n + l ) ^ n) = 0
Using (25) we have
w(n) = xa e.~xLn(x)n\
and substituting this for w(n) we obtain the identity
x(L“) " + ( a + l - .x ) ( L “)'+ « L “ = 0
We shall now consider some of the properties of the polynomials
K{x).

13.2.3 Theorem The Chebyshev-Laguerre polynomials are

orthogonal within the range (0, oo) with weight p(x) = xa e~x:
CO

J L„(x) Lap(x)xct e“xdx = 0 if n p and a > —1 (27)

o

Proof Consider the two identities

_d_ xa+1 e - x ^ W | + n x a e--L^(x) = 0
dx dx |
_d_
dx
xa+1e~x dLd^ j + p x a e~xLp(x) = 0

Let us multiply the first of these by La(x) and the second by La(x),
subtract one from the other and then integrate with respect to x
within the range (0, oo). The result is

A Ta
x“+1e -x n
dx

= ( p - n ) $ La(x)La(x)xa e -xdx
o
The left-hand side of this equation can clearly be written in the
form

\ - ^ { x a+ , e - x[La(Lay - ( L ayLa]} dx
o
298
 Chebyshev Polynomials

Consequently,

] U W ^ W - ^ e - * dx = — {xx+1c - xm ) ' L l - w m o" = 0

o P n
When x = 0 the integrated part vanishes because of the presence
of x a+l(a > —1), and when x = oo, because of the presence of e~x.
Since p + n, Equation (27) follows at once.

13.2.4 We shall now determine the norm ||Z.“||. First we must

establish the following two recurrence relations:
(rt-j- l)F„+1(x)—(2/7+1 + a —x)L“(x)-)-(«+a) L5|_i(x) = 0 (28)

4 Lan(x) = - L * t \ ( x ) (29)
dx
To do this, we must find the connection between the derivative of
the function and its partial derivatives, dLa/dx and dLa/dt. Direct
calculation shows that

(1 —2 = [ a + 1—x —(a+ l)t]Z .“(x, /)

and

dx
Substituting (24) into these identities, we obtain
00 __
( 1 - 2 t + t z) Y , " L n ( x ) t n- 1 = [ a + l - x - ( a + l ) f ] ^ £ ; ( x ) / B (30)
n= l n= 0

and
CO

- / tnL*+t(x) (31)
«=0 n= 0

Equating the coefficients of equal pqwers of t in (30) and (31), we

obtain Equations (28) and (29), respectively.
From (29) it follows that

5 L a„(x) dx = - L “+l(x) (32)

299
 Mathematical Physics

We shall use (28) to evaluate ||L“||2:

00 00

\\L%||2 J ^ c - x[L%{x)fdx = $ x * c xL*(x)L*(x)d X

0 o'
Using (28) and replacing n by n —1, we obtain
00

]|Z,®|i2 = jj ,vJ e"T„(.x) {(2a- 1+ a - . v ) £ ; . 1(x)

o

—(n- 1 + a) L“_2(x)}— dx
oo
— [ Xa c~xL*_x(x) [—x£“(x)]dx
n J
We have used the orthogonality of the polynomials and £„_2
and the orthogonality of L an and L“_,. If we use (28) again, we obtain
00
|]L“ ||2 = — \ .va e - JCL “_ 1( x ) { ( n + l ) L n
% I(x )
n oJ
— (2 /7 + 1 + o t)Z ,5 S (A ')+ (n + a )L n -i(:v:)} d x
00
= 5 -v * e -'[i;-iW ]2d.v = IHS-.lr

or
l i r a ||2 _ n + a
l+nll — -------- l+n-1 II
lira |,2 (33)

In these expressions we have used the orthogonality of the poly­

nomials
/-n—i and +n+i » L„_i and L n
From Equation (33) it follows that

||L*||2 = (» + «) (» + « - Q ••• (a ^ 2) ||£«|l2

n\
00
r(/i+ « + i)
jj ( 1 + a —x)2xae Xdx
/7!r(a+ 2)

r
7 > + « + l) , = r ( n + a + 1)
n\r(a+2) 1 ( + } n\ "
300
 Chebyshev Polynomials
Therefore,
Ta |,2 T(>Z+ a -H )
(34)
LA ~ n\

It is clear that the Chebyshev polynomials form a normal set. Con­

sequently, the theorem of Section 12.1.6 is valid for them. All the
zeros of the polynomials L„(x) are therefore simple and real and
lie in the interval (0, oo).

13.2.5 The functions

# K * )= m « * v w (35)
L-'n

are frequently used in applications. They vanish at infinity

(x — +oo). These functions have the following property:

jj &*(x)0p(x)dx = 0 if n ^ p and a > —1 (36)

o
From Equation (26i) it follows at once that the functions 0>l(x)
are the solutions of the equation

(37)

where

The Chebyshev-Laguerre polynomials are encountered in the

solution of problems on the propagation of electromagnetic waves
in long lines and in the analysis of the motion of electrons in the
Coulomb field, as well as in certain other problems.

Example Expand the function f( x ) — e-x into a Fourier series

in terms of the Chebyshev-Laguerre polynomials.

Solution In the required expansion

OO

e - * = 2 c nL°(x)
/i = 0

301
 Mathematical Physics

the coefficients cn are given by

co co

c“= i k f ) = izjjiF$
Integrating n times by parts, we obtain
n\ \ _ d”-1
c’ = ? > + « + i ) | e" d y ^ (v e~'>
d n-2 'co
+ e - T:^ r y(x',+ae -JC)| + . . . +
dxr‘ )|
|0
JA-n+ae -2xdA-j
Substitution of the limits yields zero.
Since
00

S ' " ’ e -” d . v = y ^ r / ' ( » + - + I )

0
it follows that
n!
2°+a+1
 Appendix

Definition of Generalised Functions.

The 6 Function

A .l Generalised functions will be introduced by a method similar

to that used when real numbers are introduced with the aid of
sequences of rational numbers. Real numbers are introduced so
that operations such as the extraction of the root or the taking of
logarithms can be carried out. The introduction of generalised
functions ensures that the operation of differentiation can always
be performed.
The sequence of rational numbers {a„} is called fundamental
if for any rational e > 0 there exists n0, such that for all n and
m > m0
l^n @m\ <'- £
The fundamental sequences {a„} and {bn} are called equivalent if
lim |a„~bn\ - 0
n—*■co
Equivalent sequences define a real number.
A.2 Consider continuous functions defined in the interval
(A, B), — oo ^ A < B ^ co. The sequence of functions { /„ (x )}
which are continuous in (A, B) is called fundamental in (A, B) if
there exists an integer k ^ 0 and .another sequence of functions
{F„(x)} which are continuous in (A B) which are such that the
following two properties are satisfied:
1. FW ( a-) = /„ (* );
2. the sequence (F„(x)} converges uniformly in any segment
[a, c (A, B) (Fn(x) :).
The following theorem follows from the definition of a funda­
mental sequence.
303
 Mathematical Physics

Theorem 1 If a sequence of functions {/„(X)} which are continuous

in (A, B) converges uniformly in any segment [a, /?] c= (A, B), then
it is a fundamental sequence. In point of fact, we have Fn(x) = f„(x)
and k — 0.

Theorem 2 If {/„(x)} is a fundamental sequence of functions

having continuous derivatives of order m, f i m)(x), the sequence
{/nm)(-v)} is also fundamental.

Proof For there exists a number k > 0, and a sequence

{/^(x)} which has Properties 1 and 2. For {ffim)(x)} we can take
k + m instead of k and the same sequence {F„(x)}. These will clearly
satisfy Properties 1 and 2. Consequently, {/„(m)(x)} is a fundamental
sequence.
A sequence of functions {/„(x)} is said to be uniformly bounded
in (A, B) if there exists a number M such that \f„(x)\ < M for all
n and for all x e (A, B).

Theorem 3 If a sequence of functions which are continuous in

(A, B) is uniformly bounded in (A, B), and converges uniformly
in any segment [a, /?] c= (A, x0) and in any segment [a, /?] <= (x0, B),
then it is fundamental in (A , B) where x0 e (A, B).

Proof Let us take i 7nW = S/n(0dt and k = 1. Property 1 will

Xq
then be satisfied. It remains to show that Property 2 is satisfied. Let us
take e > 0 and [a, /S] <=(A, B). If [a, /?] c= (A, xo) or [a, ] <=(.Vp, B),
then for these segments Property 2 will be satisfied by the conditions
of the theorem. Suppose that [a,/3] <=(A,B) and A < a < A'0
< /? < B. By hypothesis \fn{t)\ > M. We must show the uniform
convergence of the sequence {F„(.v)} in the segment [a, /?]. Conse­
quently, we must consider x e [a, f\. For such x we have
X

\Fn{ x ) - F m{x)\ = ! 5 { U t ) - f n ( t ) } dt\

Xo

A' /?
< | S|/n (0 -/» .(0 1*1 < S\ M t ) - M 0 \ dr
A-o a

*0- 6:11 JCo+6AJ

S \fn -fm \ dr+ S ) f , - f » \ d t + 5 *
Aq- aq-I- 6.\t
6M
304
 Appendix
Since the sequence {/„(/)} converges uniformly in the segments
[a, .Y0—e/6M] and [.\0t«/6jW , /S], we can find a positive integer n0
such that the following inequalities will be satisfied in the two seg­
ments :

l/n(0-/m(0l < for all m, n ^ n0

3 (/?—ot)
For n, m > n0 we then have

F„(x)—Fm(x)\ <
3 0 - a ) *o — 6M

+ j « * -)< ■
Consequently, the sequence {Fn(x)} converges uniformly'in any
segment [a,/?] B). Property 2 is therefore satisfied and this
proves the theorem.

Remark If the sequence {/„(.v)} is fundamental, the sequence

X

(S/n(0df) is also fundamental.

*0
Example 1 Consider the sequence of functions {gn(.x)}

gn(x) =
1+e"
and the interval (—oo, co) (Fig. A.l).

This sequence is uniformly bounded by the number 1. In any

segment [a, /?] cz(—co, 0) it converges uniformly to zero. In any
segment [a, /?] <=(0, co) it converges uniformly to unity. Therefore,
Theorem 3 applies. Consequently, the sequence (g„(x)} is a funda­
mental sequence.
20 305
 Mathematical Physics

Example 2 Consider the sequence of functions {/„(x)}

and the interval (—co, co) (Fig. A.2). In any segment [a, ft] c=
(— oo,0) or [a,/3] c= (0, oo), the sequence converges uniformly to
zero. However, it is not uniformly bounded. The sequence of func-
tions {F„(x)}, where F„(x) = \ /„(/)d/ will also converge uniformly
— QO

in any segment [a, /?] within the intervals (—co, 0) or (0, oo) and
is uniformly bounded (by the number 1) in the interval (—oo, oo).
Consequently, by Theorem 3 it is a fundamental sequence. If this
is so, then by Theorem 2 the sequence {/„(/)} is also fundamental.
Example 3 Consider the sequence of functions {^n(x)}, where
<p„(x) is a piecewise-linear continuous function equal to zero outside
the interval (—1/«, 1/«) (Fig. A.3). In any segment [a, /?] within
the intervals (—oo, 0) and (0, oo) it converges uniformly to zero.
However, it is not uniformly bounded. The sequence of functions

*
{$„(x)}> 0 n(x) — jj <pn(t)dt, will also converge uniformly in any
— CO

segment [a, /3] in the intervals (—oo, 0) and (0, oo) and is uniformly
bounded (by the number 1) in the interval (—oo, oo). Consequently,
by Theorem 3 it is a fundamental sequence. If this is so, it follows
from Theorem 2 that the sequence {<?„(*)} is also a fundamental
sequence.
Example 4 Consider the sequence of functions {<f „ ( a')} where
71

306
 Appendix

and the interval (—00, 00) (Fig. A.4). Since the sequence of functions
{ ^ (x )} , where !/y„(x) = \ y>„(t)dt, satisfies the conditions of Theo-
—00
rem 3, this sequence is also fundamental. Consequently, by Theorem
2 the sequence (yn(x)} is a fundamental sequence.

A.3 Two fundamental sequences {/„(x)} and {,gn(x)} are called

equivalent, {/n(x)} {g„(x)}, if there exists an integer k ^ 0 and
two other sequences {.Fn(x)}, {(7(x)} such that
1. F p = /„(x), G(nk)(x) = gn(x);
2. in any segment [a, /?] <=.(A, B) the sequence {.Fn(x)—(7n(x)}
converges uniformly to zero:

Thus, the sequences in Examples 2, 3 and 4 are equivalent to one

another. They are also equivalent to the sequence {^(x)} (Exam­
ple 1).

Definition Each class of equivalent fundamental sequences defines

a generalised function /(x ) which can be represented by any other
sequence of this class. We shall also say that a fundamental sequence
{/„(x)} defines a generalised function /(x ) and write f( x ) = {/„(*)}•
Thus, the sequence (g„(x)} (Example 1) defines the unit step
10, x < 0,
function ??(x) = Y > 0 Consequently, the unit step function
rj(x) is a generalised function.
In view of the lemma in Section A.4, and Theorem 1, any function
which is continuous in (A, B) is a generalised function.
The sequences {^(x)}, {/„(*)}, {?„(*)}> W * )} in Examples 1^4
define a generalised function d(x) (with a singularity at x = 0).
This is called the Dirac d function. It is clear that the d function <5(x)
is an even function, i.e. <3(—x) = <5(x).

A.4 A linear combination a/+ /3 ^(a and fi are constants)

of two generalised functions /(x ) and y(x), specified by the
fundamental sequences {/„(x)} and {^(x)}, is defined as the gen­
eralised function F(x) corresponding to the fundamental sequence
{af n(x)-\-P<pn(x)}. In particular, the sum and difference of two
generalised functions /(x ) and <p(x) is also a generalised function.
The product of a generalised function ?y(x—xo), i.e. the unit
function, which is continuous in the segment [a, b], and the function
9?(x), which will be written rj(x—x0)(p(x), will be defined as the
20*
307
 Mathematical Physics
generalised function defined by the fundamental sequence {g„0v—.v0)
^(x)}, where g„(x) is the function in Example 1, and

9'(b), x > b
V(x) = (p{x), a ^ x ^ b
tpifl), x ^ a
It is clear that
x > x0
i ] ( x - X 0)(f.(x)
“ lo, X < x0

A function cp(x) which is piecewise-continuous in [a, b] and has

singularities at x {, xz, ..., x k(a < xi < x 2 < ... < xk < b) can be
written in the form
<F(X) = .v ,) ] f o W + ...
+ V i(x)- ...
+ bi (x - xk) - >](a-- b)] (pk(a-)
where
(p(xi+i), X A ';+1
O' = 0, 1, 2, ...,* )
<Pi(x) (p(x), X{ < -V < .Yi + i
A'0 — a, A't+1 = b
(F{Xi), A < A(
<Pi( a ) are continuous in (— oo, oo). Their products by unit step
functions i](x—x c) are generalised functions. Consequently, an
arbitrary piecewise-continuous function is a linear combination
of generalised functions and is, therefore, also a generalised function.

A.5 Let {e„} be a sequence of positive numbers tending to zero.

Definition A sequence of continuous functions {<5„(x)} is called

a sequence if these functions have the following properties:
1. d„(x) > 0 in (—en, e„) and zero elsewhere;
2. dn(x) has derivatives of all orders everywhere;
00
3. S d„(t)dt = 1.
—00
Consider the function
 A p p e n d ix

It has derivatives of all orders. The function

/S„(-v) = a(.v | En)rj.(Fn—x )
is positive in (—e„, e„) and zero elsewhere and possesses derivatives
of all orders. Let
CO

|'/ ? J = S Pn(t)dt
— 00

We then have
00

—00

Consequently, the sequence ( 1 is a 6 sequence.

{ \\Pn\\ J

Theorem 4 Any d sequence is a fundamental sequence.

Proof The sequence of functions

X

yn(-v) = J d„(t) dt
—co
is uniformly bounded, since 0 ^ y„( x ) ^ 1, and in any segment
[a, /3] which does not contain x — 0 converges uniformly to zero
if [a, p\ c ( —oo,0), and to unity if [a, {$] <=(0, co). In point of
fact, in the former case {[oc, /5] cz(—oo, 0)} we can find a number
no such that for all n > n0 we have en < \(3\. Consequently, the
segment [a, fi] will lie to the left of all intervals ( —e„, en). Therefore,
all functions d„(x) with n > n0 will be zero in [a, /?]. Hence, it follows
that y„(x) — 0 in [a, /?] for n > m . In the second case {[a, /J] <=(0, oo)}
we can find a number n0 such that for n > n0 we have e„ < a. Con­
sequently, the segment [a, /3] will lie to the right of all intervals
(—e„, £»), n > n0. Therefore, for x e [a, /S] and n > n0,
x e„

7n(x) = J dn( t ) d t = J < S „(0 d /= l

— CO — en

Hence Theorem 3 applies and the sequence {y„(.v)} is a fundamental

sequence. If this is so, it follows that by Theorem 2 the sequence
{Ai(X)} is also a fundamental sequence, and this proves the theorem.
Remark It is clear that for all constants C the sequence {Cr)„(x)}
is a fundamental sequence. It is readily shown that all S sequences
are equivalent to each other and to the sequences in Examples 2-4.

309
 Mathematical Physics
The reader is recommended to prove this. Consequently, any 8
sequence {<5„(x)} defines the <5 function <5(x). The <5 sequence
{(5n(x—x0)} defines the <5 function <5(x—x0).
Multi-dimensional 8 sequences are defined in a similar way.
For example, in three-dimensional space the d sequence is defined
as being of the form
{<5,,(x)<5n(y)(5n(z)}
i.e. as consisting of the products Sn(x) <5„(y) <5n(z), whereas the 8
function in three-dimensional space is defined by 6 sequences of
this form. Therefore, by definition, 8(M) = 6(x) 8>(y)d(z) and
8 ( M , M 0) = {<5„(x—x0)8u(y—y0)8u(z—z0)}
= <5(x—x0) <5(y—y 0) 8(z—z„)
where x, y, z are the coordinates of the point M and xo, Jo, zo are
the coordinates of the point M 0.
We have shown above that the sequences {A 8„(x—x0)} are
fundamental sequences. It is possible to prove a more general
statement, namely that for all functions y(x) which are continuous
in the neighbourhood O(x0) of the point x = x 0, the sequence
{ f ( x ) ^ ( .r - x 0)} is fundamental in <9(x0) and is equivalent to the
sequence {9?(x0)^„(x—x0)}.
Proof Consider the functions
X X

Fix) = jj <f(t)6n(t—x 0)dt and 0„(x) = \ (f(xQ)8n( t ~ x 0)dt

— oc — co

It is clear that F'n(x) = (f(x)8„(x—x 0), <£'(x) = ?(xo)<5n(x —x0) for

all points of O(x0). For any e > 0 we can find a number n0(e) such
that
|<p(x)—p(x0)| < e for x e (,y0—e„, x0-fen) and n > n0
Then for any x e O(x0) and n > n0
X *

|F„(x)—0 „(x)| = | \ {<p(t)-(f(xo)}Sn(t—x0)dt\

— 00

X
^ $ 1^ (0 —y(^o)l^»(^—x0) dt
— OC

^ 5 \<p(t)—9K-vo)l 8n(t—x 0) dt
*0+en
< £ jj (5„(/—x0) d t = e
Xo—*n
310
 A p p e n d ix

i.e.
|Fn( x ) - 0 B(x)| < £ (A.l)
Since the sequence {$n(x)} converges uniformly [(0„(x)
= <p(xo)y„(x—x0); see proof of Theorem 4], it follows that the
sequence {.Fn(x)} converges uniformly in any segment [a, /?] c= 0 (xo).
From this and from the inequality given by (A.l) it follows that
the sequence {^(x) <5„(x—xo)} is fundamental in 0(xo) and is equiva­
lent to the sequence {<p(xo) Sn(x2—Xo)}.
The generalised function q>(x) S(x—x0) specified by the funda­
mental sequence {^(x) (5„(x—xo)} is defined as the product of the
function <p{x) which is continuous in the neighbourhood of the
point x = xo and the d function <5(x—xo).
The above statement means that
(f (x) h(x x0) = cp(xQ) <5(x —x 0)

A.6 Theorem 5 Equivalent fundamental sequences defining the

generalised function /(x) include fundamental sequences of differ­
entiable functions (polynomials).
Let us first establish the following lemma.

Lemma For any function F{x) which is continuous in (A, B)

there exists a sequence of polynomials (P n(x)} which converges
uniformly to F(x) in any segment [a, /S] c- (A, B).

Proof Let {An} be a decreasing sequence of numbers converging

to A, and let {Bn} be an increasing number of zeros converging
to B. By the Weierstrass theorem there exists a polynomial Pn(x)
such that |F(x)—F n(x)| < 1jn for all points x in [An, B„], The
sequence of such polynomials has the property indicated in the
lemma. In fact, for any positive number e and any segment
[a, /5] cz (A, B) one can find a positive integer n0 such that for all
n > n0

— < e and [a, p] <= [An, B„]

n
It then follows that for all n > no and all x 6 |a, f)\,

\Pn(x)-F(x)\ < - < s

n
which indicates a uniform convergence of the sequence (P n(x)} to
the function F(x) in the segment [a, /?]. This proves the lemma.
311
 Mathematical Physics
Proof o f the theorem Let {/„(*)} be a fundamental sequence
defining the generalised function fix). By definition of a funda­
mental sequence there exists an integer k > 0 and a sequence of con­
tinuous functions (F„(x)} which converges in (A, B) to a
function F(x), which are such that
F(nkKv) = / ,( * )
The function F(x) is continuous in (A, B) since the sequence {F„(.v)}
converges uniformly to F(x) in any segment [a,/?] <=(A, B). In
accordance with the above lemma, there exists a sequence of poly­
nomials { P „ ( a ) } which converges uniformly to F(x) in any segment
[a, /9] c ( A , B). Therefore, by definition of equivalent fundamental
sequences, the sequence of polynomials
0 „ 0 )} where p n(x) = P(nk)(x)
is a fundamental sequence which is equivalent to {/„O0}- This
proves the theorem.
Therefore, it can always be considered that the generalised
function / ( a ) is defined by the fundamental sequence of differentiable
functions { / „ ( a ) } .
Definition The w-th order derivative of a generalised function
f i x ) = {f„{x)} is defined as the generalised function
f i ' f i x ) = { f i ”%x)}
which is defined by the fundamental sequence of /n-th order deriva­
tives {/„m)0)}- Thus, in particular, the d function (5(.y) possesses
derivatives of all orders. For example, 6'( v) is defined by the fun­
damental sequence (<5'0)}- The derivative of the unit step function
7](x) is the generalised function
VO) = <50)
since the sequence {g'j of Example 1 is equivalent to {/„O0} °f
Example 2, which defines the d function. Consequently, we may
write
X

fix)= j m dt
-- oc
A.7 The integral of the product of fi(.v—.v()) and of an arbitrary
continuous function y{.v), i.e.
b
\ ? (a-) (5(.v-.v„) d.v

312
 A p p e n d ix

is defined as the limit

lim $r(-'')<5nCv—x0)dx

where {(5„(x—xo)} is any 6 sequence defining the d function d(x -xo).

We shall show that this limit exists and that the following formula
is valid
b
<f (x 0), if x0 e ( a , b )
\cf {x)b(x -x 0)dx
0 , if x0£ [ a , 6]
It may be assumed that the functions S„(x—x0) which form the 6
sequence are even with respect to x — xo. For xo = a or xo — b,
we then have
b

$ rf (-v) <5(x—x0) dx = — (p(x0) (Prove this!)

a

Suppose that x0 $ [a, b]. We can then find a number n0 such

that for all n > n0 we have e „ < min {|x0—a\, |x0—b\}. Conse­
quently. for n > n0 we have (5„(x—x0) = 0 in [a, b],
b
Therefore j[ y(x)(3n(x—x0)dx = 0. Consequently
a
b

5 (p(x)d(x—x0) dx = 0

Let x0 e (a, b). We can then find a number n0 such that for n > n0
the intervals (x0—e„, x0+e„) will lie entirely in the segment [a, b\.
Consequently, for such n,
b x0-]-En
5<p(x) S„(x —x0) dx = 5 (fix) 6„(x—x0) dx
a xo £n
Let us apply the mean value theorem to the last integral. We obtain
Wo-Ff/i -TO+f/i
\ (f (x)bn(x — x 0) d x — </■(£„) jj A„(x—x 0) d x = (p(£„)
Xq £n Xo
where i n e [x0—e„,x0+e„]. A sn oo, we have £„ -+ ,v0. Therefore,
since the function<p(x) continues in [a, b\, weobtain
b
lim J <p(x) d„(x—x 0) dx = lim $>(£„) = y(x0)

313
 Mathematical Physics

This proves the above statement. In particular, for 9?(x) = 1,

we have
b
{ 1, *o e [a, b]
^ <5(a —x0) d.Y =
| o , x„ $ [a, b\
The formula

\<p{M)b{M,M*)dxM = \ ,p{-M*) ’ lf M° e D ’ (A.2)

d [0, if M 0 $ D
can be proved in a similar way for any function (p{M) which is
continuous in D.

Remark Each of the sequences {/„}, {99,,}, {y„} of Examples 2^4,

which define the 6 function <5(a), converges to zero at any point
x ^ 0 and to infinity at a = 0. In view of this, we may write

I °« x =£ 0
«K-v) = \ co, x = 0

The function <5(a') becomes infinite at x = 0 so that ^ <5 (.v)dA‘ = 1

' —a
for any a > 0. The expression (A.2) is often used to define the
<5 function as a functional.

A.8 A function m ( a ' ) will be defined as localised if it is identically

equal to zero outside a given interval (a, b). A localised function
will be called smooth if it is continuous and possesses continuous
derivatives of all orders everywhere.
Let f { x ) be a continuous or locally integrable, i.e. integrable
in any finite interval, function. The convolution of f ( x ) and 99(A)
is defined by
O0
/(.v) * ? (.'-)= 5 /(A -/M 0 d '
— 00
(A.3)
It is clear that
00

f i x ) * 99(A) = 99(A) * f(x) = J fit)<pix— t) d t (A. 4)

— 00

Since 99(A) is a localised function, the convolution can also be

written in the form
b
f *m = \ f i x —t)<pit) dt (A.5)
a

314
 A p p e n d ix

within the interval outside which <p(x) = 0. The following are the
simplest properties of the convolution.

Property 1 If the function <p(x) possesses everywhere continuous

derivatives up to A-th order, then the convolution f * <p has deriva­
tives everywhere up to A-th order and

-^ r [/(■*) * ? « ] = U * v i r) = f ( x ) * ^ ' \ x )

( p = 1 ,2 , ..., A)
This follows directly from (A.4) and from the fact that the function
<p(x) is localised. If the function f(x) has continuous derivatives
up to order A and <p(x) is integrable and finite, then
( P = 1 ,2 , ..., A) (A.7)

Property 2 If the sequence of continuous functions {/„(x)} con­

verges uniformly in the range a0—b < x < A0—a to the function
/(x), then for any continuous localised function rp(x) which is
identically zero outside the range {a, b), the sequence {/„(x) * <p(x)}
converges uniformly to the function /(x) * <p(x) in the range [a0, 60].
The validity of this follows directly from (A.5).
Let (A ', B ’) represent the interval consisting of points x' such
that [x'—b, x ' —a\ c= (A, B).

Property 3 If the sequence {/„(x)} is fundamental in (A, B) and

<p{x) is localised and continuous [<p(x) = 0 outside (a, b)\, then
the sequence {/„(x) * <p(x)} is fundamental in (A', B').

Proof Let [a, /?] cz (A' B'). The segment [a—b, /?—a] then belongs
to (A, B). Since {/„(x)} is a fundamental sequence, there exists
an integer A > 0 and another sequence {/^(x)} such that
Fn(k) = f„(x) and {Fn(x)} converges uniformly in [a—A, /S—a].
We have
/„(x) * (p{x) = F {nk\ x ) * <p(x)
and, using Equation (A.7), we obtain

/»(*) * rp(x) = [F„(x) * 9? ( x ) ] (A;) (A.8 )

In view of Property 2, the sequence {/^(x) * 9?(x)} converges uni­
formly in [a, /?]. Hence, and from Equation (A.8), it follows that
the sequence {/„(x) * <p(x)} is a fundamental sequence.
315
 M a th e m a tic a l P h y s ic s

Remark If q>{x) is a smooth function, then Equation (A.8) can

be written in the form
/„(.v) * <p(x) = F„(x) * tpW(x) (A.9)
Since the sequence {/^.(x)} converges uniformly in any segment
[a, /?] <= (A, B), then by Property 2, the sequence {Fn(x) * ^ (fc)(x)}
converges uniformly in any segment [a + 6, px-a] belonging to
(A', B'). Therefore, we have the following property.

Property 4 If the sequence {/„(x)} is fundamental in (A, B),

and <p(x) is a smooth finite function [<p(x) = 0 outside (a, b)\, then
the sequence {/„(x) * ^(x)} converges uniformly in any segment
[ * ' , n <=
This leads naturally to the following.

Definition The convolution of an arbitrary generalised function

/(x), defined by the fundamental sequence {/„(x)}, and a local­
ised continuous function <p(x), is defined as the generalised function
/(x ) *tp(x) defined by the fundamental sequence {/„(x) * ^(x)}. It
is clear that
/(x ) * tp{x) = <p{x) *f(x)

The convolution of the generalised function /(x ) and an arbitrary

localised integrable function 90(x) is defined in a similar way. More­
over, ( f * (pYp) — f (p) * <p, where / (p) is the p-th order derivative
of the generalised function. In particular, if tp(x) is an arbitrary
localised and everywhere continuous function,
<)(x) * f f (x) = <p(x) * ci(x)

Property 5 For a localised and everywhere continuous func­

tion <p(x).
cf(x) * 9(.y) = <f{x)

Lemma Let {<\,(x)} be a 6 sequence and <^(x) be a function which

is continuous in (A, B). The sequence {^(x) * <5„(x)} will then
converge to cp(x) uniformly in any segment [a, /?] c= (A, B).

Proof Let [a, /3] c: {A, B). For any e > 0 one can then find
n0(s) such that for n > n0 and all x in [a, fi] and all t in (—e„, e„),

i^(x—0 —r(Y)i < e

316
 Appendix

Consider the difference y(x) * <5n(.*) —<p(.r):

00
!9'(.y)*(3,1(.v)-V-(-v)I = S (f (y -t)Sn( t ) d t - <f{x)
- oo
CO CO

= | J <p(x~t)d„(t)&t— $ <p(x)dn(t)dt
— 00 — CO

CO

^ jj |9?(.Y—r)—9?(.Y)|<5n(r)dt
— CO

= J i9?(*—o —
—£n
For n > nQ(e) and any x e \<x, /S|, the last integral is less than
en

e J d„(t) dt = e
— en

Therefore, for n > n0(e) and x e [a, /?]

<5„(x) (p{_x)\ < £
This proves the lemma.

Proof of Property 5 Since the functions <p(x) * dn(x) are con­

tinuous in (A, B), and in accordance with the above lemma, the
sequence {y(.v) * d„(x)}, converges uniformly to the function y(x)
in any segment [a, f] c: (A, B), then by Theorem 1 it is fundamental
and defines the generalised function 95(x). On the other hand, by
definition, the fundamental sequence (95(x) * <S„(x)} defines the
convolution ^(x) * <5(x). Therefore,
y(x) * d(x) = y(x)
i
Moreover,
CO

y(x) = lim [^(x)* <5„(x)] = lim jj

n~*cc n —+co _ cC
VD CO
= jj (p{x— t)d{t) dt = ^ y{t)S(x —t)dt
—00 —00
where we have used the definition of an integral of the product
of a continuous function and a 6 function.
317
 Mathematical Physics
Therefore, the convolution of a <5 function and an arbitrary
continuous function <p (a) can be written in the form
00 CO

(p{x) * <5(x) = jj (p{x—/)<5(r)dt= jj (p(t)d(x—/) d/

—oo —CO
Property 6 The convolution of an arbitrary generalised function
f(x) and a smooth function <p(x) has continuous derivatives of
all orders.
Proof Let the generalised function f(x) be defined by the funda­
mental sequence {/„(-v)}. By Property 4 the sequence {/„( a') * <p(x)}
converges uniformly to the continuous function in any segment
[a', /?'] <= (A', B’). From Equation (A.9) and Property 4 it follows
that the sequence {/, * of the /-th-order derivatives (/ = 1,2,...)
will also converge uniformly to continuous functions. Using the
theorem on the term-by-term differentiation of sequences, we
obtain Property 6.
It is possible to define the convolution of an arbitrary generalised
function f(x) and a 5 function as the generalised function f(x) * 8(x),
defined by the fundamental sequence
{/(*) *
where {<5n(x)} is an arbitrary (5 sequence. We then have
f ( x ) * d(x) = f ( x )
The following is a list of the most useful formulae and rela­
tionships involving the <5 function. The reader is recommended
either to prove these results for himself or to refer to specialist
literature.
1. 8 ( - x ) = 6(x)

2. 8(otx) = —<3(a-)
a

- if has only simple zeros x n.

n \<P (-T,)l
4. aA(.y) = 0, <r(x)8{x) = <p(0)d(x), (f(a±x)d(x) = <p(a)d(x)
CC

5. J q ( x ) 8 ( x - x ()) dx = <p(A'o)
— CO

OC
6. jj d(x—t)d(s—t)dt = 8(x—s)
— oc

318
 Appendix

7. 6 \x) = — d(x)
00

8. J <p(x)d'(x—x0) d.v = —<p'(X(i),

—CO
if <p(x) is continuous at.v = x0.
00

9. jj d'(x—t)d(s—t)dt = d'(x—s)

dn n\
10. ^ (5(.v) = ( - l ) n— 6(x)
d.v'1 .v

1L $ <p(x) - ^ d(x ~ s) dx =
—CO
if 9?(n)(x) is continuous at x = s.
CO
12. \ $ $ <p(M)d(M, M 0) dvM = <p(M0)
—CO
13. The Fourier transform of the 6 function 6(,y—.v0) is given by
00

« ({ )= — t <5(.v-.v„)e«'d* = —
j/2.
V 27i J }/ In

and, consequently,

-i$(x-xo)
di
r — 00 -00

For the three-dimensional case

CO CO CO
d(M, Mo) = I—I jj jj J e-iLftx-xoJ+Kj-^+Wr-^df d»/dC
— CO — CO — CO

= — C e-«f(*-*o) d£— [ e - '^ - ^ d r / — f e- i?(z“Zo)dC

J J 2tt J
— GO — 00 — °0

= &(x—x 0)&(y—yo)&(z—z0).
Therefore
d(M, M 0) = d( x —-X0) (5(j^ y0) d(z Zq)

319
 Mathematical Physics

in Cartesian coordinates.

14-. d(M, M0) — ' h(/■—/'o) d(<jp—(fo) in polar coordinates on

a plane.

15. d(M, M0) = |2 — r0) (5((9—(90)<5(^p—t?0) in spherical

coordinates.

16. 6(M, M0) = in arbitrary ortho.

'h 'h >h
gonal curvilinear coordinates (qu q2, <73), where h2, and h3 and
the Lame coefficients.
The generalised functions 6+(x) and <5-(.y) are also encountered
and are defined formally by
OO 0 00

0 — oo 0

It is clear that <5+(x)-h6_(x) = (5(x) and <5+(—x) = <5_(x).

 Answers and Solutions

C H A PT E R 1

1. (a) WM“ -0.5(l/f)w{ = 0 V = t / a >;

>
II

(b) Aw—-W{—w, = 0 (I = t 2> V == x 2y,

(c) in ~= 0
11> (f = t / a, v = y) ;
0.5
(d) (W{ w,) = 0 for _y < 0 (f =
V = x —2 ) /—y), Aw = 0 for t > 0 (£ = x, r) = 2]/y).
2. (a) vnn—2v£ = 0,u = v e"l+"", p = 1.1875, v = 0.25
(f = y —x> n = y+x) ;
(b) = w = ^ eM’+’"', /A = —0.25,- v = - 7 /8
(£ = T—3 a-, V = y —x);
(c) v(SJr vnn— l.5v = 0, u = v c~‘~r> (£ = 2y—x, y = x).

C H A PT E R 2

1. a2uxx+ g = un , u (x, 0) = 0, u ,( x ,0 ) = v o, u(0, () = ux(l,t) = 0.

2. a2uxx—Pu, = utt, u (x , 0) = <p(x), ut(x, 0) = fpi(x),
i/(0, t) = it (I, t) = 0.
3. d/d.r (E S uxx) = p S u „ , u(x, 0) = ^(.r), u,(x, 0) = <pi(.v),
ai«x(0, t)—fixu(0, t) = 0, a2w.v(/, 0+/?2«(^ 0 = °-
21 321
 Mathematical Physics

4. d/dx [(/—x)ux] = (l / g) ua , u( x , 0 ) = <p(x), u,(x, 0) = <pt(x),

u(0,t) = 0, |m(/, t)| < Af.
5. gd/dx [(/—x)wx]+co2w = utf, additional conditions of Problem
4.
6. co2d/dx [(/2—x2)wx] = pu„, additional conditions of Problem 4.
7. a20xx = 0„, 0(x, 0) = <p (x), 0,(x, 0) = <Pi(x),

0(0, t) = 0(7,/) = 0.
8. a2uxx+(H/cp)I(t) = u„, u(x, 0) = ut(x, 0) = 0,
w(0, t) = w(/,/) = 0, c is the velocity of light.
/), x<0
„,(* ,/), *>0 ( '= 1 . 2 ) .
w(x, 0) = <P(x), u,(x, 0) = p,(x), a2i = Et/Pi (i = 1,2),
Mi(0, t) = m2(0, 0 , 0 = E2u1x(0, /).
10. Tuxx = [p+m 0<5(x—x0)] m„, u (x, 0) = <p(x), wf(x, 0) = 9>,(x),
u(0, t) = u(l, t) = 0.
Or:
( W ^X, /), X < X0
u (x,t) = \ a 2 ( U i ) x x = (M,)rr O' = 1,2),
I U2(x, t), X > x 0
m ( , 0) = 9)( x ) ,
x wf(x, 0) = ( P i ( x ) , i/,(0, 0 = 0, u2(l, t) = 0,
0 = w2(^o, t), T [w2x(x0, t) — u l x ( x 0 , /)] = mQutt(xQ, t).
11- Tuxx+ F ( t ) d ( x —v0t) = pu„, u(x, 0) — t/f(x, 0) = 0.
12. vx-\-Li,-\-Ri = 0, ix-\-Cvt+Gv = 0, v (x, 0) = cp (x),
/ (x, 0) = cp!(x).
13. vx+ Li, = 0, /X+ C a, = 0, w(x, 0) = cp (x), / (x, 0) = <p, (x),
—v(0, t) = R0i (0, t), c0v t(l, t) = i (I, t).
14. vx-\-Lit = 0, ix+Cv, = 0, v(x, 0) = cp (x), i(x, 0) = <p,(x),
- v ( 0 , t) = Li»i,(0, t), v(l, / ) - £ ( / ) = L ^ i t(l, t).
. . \ v i (x,t), x < 0 |/i(x , 0 . *<0
15
l®2(*> i), x >0 *(*» 0 — | ^ ^ x > o
(vk)x+ Lk(ik),+Rkik = 0, (ik)x+ Ck(vx), = 0, v ( x , 0 ) = <p(x),
i (x, 0) = 9>!(x), i,(0, t) = i2(0, t),
®2«(0, 0 - ® i « ( 0 , 0 = (l/c o )/,(0 , 0 (k = 1, 2 ) .

322
 Answers and Solutions
For the current i (x, t) :
(ik)xx — CkL k(ik)u + Ck Rkik, (0, t) = z2(0,t).

16. kuxx—h[u—<p(t)] = cput, u(x, 0) = f ( x ) , w(0, r ) = /i(0 >

k u x(l, t) = q(t).
17. k u xx—h u + Q I 2= cput, u(x, 0) = /(x ), kux(0, t) = t),
k u x(l, t) = C2ut(l, t), where Cl} C2 are the thermal capacities
of the holders.
18. djdx (Dux) —d/dx (vu) = cut .
19. (a) djdx (Dux) —flu = cut ;
(b) djdx {Dux)-\-fiu = cut.

u (x, 0) = <p(x), 0 = u2(0, t) ;

(a) kiU^iP, t)—kxulx(0, t) = 0;
(b) k 2u2x(0, t ) —k 1Uix(0, t) = C0ut(0, t ) .
21. d/dx (kux) = cput, u(x, 0) = 0, u(vt, t) = <p(t).
22. d/dx (kux)-\-Qd(x—v0t) = cput, u(x, 0) = <p(x).
23. a2uge—h(u—u0) = ut, u(d, 0) = <p(d), u(0-\-2n, t) = u(6, t),

6 is the polar angle, a2 = kjcpR2, R is the radius of the ring.

dc2 d2E dH _ c2 d2H
dt Ana p d£2 ’ dt Anap d£2 ’
where c is the velocity of light, a is the conductivity of the
medium, p is the magnetic permeability, £ is the distance
measured from a fixed plane, E = £"(£, t), H = //(I, t).
25. (a) A u = —4np;
(b) Au = 0.
21 * 323
 Mathematical Physics
CHAPTER 3

1. See Fig. B.l. Hint : use D’Alembert’s ‘formula.

A=/ k=2
2c 0 c 2c 3c 0 c 2c 3c
Fig. B .l

2. See Fig. B.2.

4=/ k=2
/ N / \ . /X /X / \ / \ / \ / \
~2c ~c 0 c 2c ~3c ~2c -c 0 c 2c 3c

k--3
. y x .
-4c ~3c -2c - c 0 c 2c 3c 4c
Fig. B.2

3. zz(.\\ t ) = \/2 a { F (x+ a t) —F ( x ~ a t ) } ,

0, z < —c
where F(z) = v0(z+c), —c < z < c
2v0c, z > c
4. See Fig. B.3.

A =4

0 c 2c 3c 4c 0 '^ g,'2 c 3c 4c Jc 3c

Fig. B.3

5. n ( x , t ) = \/2a{F(x-\-at) —F ( x —at)},
where F(z) = p/p{r]{z-xQ) - p { z - \ - x 0)}.
Hint: solve the problem: a2uxx = utt, m(0, t) = 0,
u(x, 0) = 0, tt,(x, 0) = (p/p)d(x—x 0) , 0 < a- < go .

6. For —oo'< .x < 0 we have

i/ i (.y, 0 = / ( r - .v /« ,) + j J l |£ |~ f P i £j - / ( ( + .v/a|).
V P\ E\ + | IhFi
324
 Answers and Solutions
^ 2
Refracted wave: u2{x,t) = — ()l 1----- f ( t —xla2).
V p i E i +X/'Pi E,

Reflected wave: urefl = ^ Z-1^ 1— *f 2^ 2 f { t + x j u :).

}/p1£ 1+ ^ p 2£ 2 -
»refl absent for pjZq = p2E2.
7. v (.v, 0) = £ 0 e~ l'G* * i (x, 0) = E0)/ GjZ e - >/G* *
u ( a , 0 = E0 e~^GRxj'j (x—a t), 0 < a, t < oo,
i(x, t) = E ^ / G j L t ~ ^ CRxr](x—at)-

8. u (x, tx) = 0, u(x, t2) = — u(x, 0), u (x, t4) - u(x, 0).
x+af
a + v o

a2- v l —at < x < v0t,

2aT0 jj f(f)d { ,
9. Mj Cv, 0 =
0, x > at,

2 2 $ ^ )d l, v0t < x < a t ,

a —Vp
w2(a', t) =
2aTn 0, x > at,

T0 is the initial tension in the string.

10. u(x, t ) = p (t—x / a ) f ( t —xja), a = l/|/L C .

11. / (x, t) = p ( /- a /« ) ^ ]/C /L

where a — \l\/LC , oc = (1/C0) j/C /T .

12 .

13. Only equations of the hyperbolic type of the form

a \ \ u x x Jr ' 2 - a \ 2 u x t Jr a 2 2 u tt = 0.
The velocity a is determined from the equation
a22a2—l a l2a-\-an = 0.
14. Only equations of the hyperbolic type with coefficients satisfying
the following relationships:
325
 Mathematical Physics
a22 a2~ 2 al2a Jr a u = 0, bx—b2a ~ l n ( a {2—a22a) — 0,
p b 2-\-c == 0.

CHAPTER 4

i ~ 2hl2 V 1 •nn . nna

1. UyX, t ) 2 (1 2 ^ ® / x0sin —j-xcos
0(l—x o)\ Z—
/
7tzx J AT /
n= 1

where A = y ^ F 0x0(/—x0) .

„ , , 2P V 1 . Jin . Tin nan

2. u (x,t) = ----- > — sin—r--x0 sin—p x sin—— t .
7iap iL-l /J /
n= I

Hi nt : u (x, 0)= 0, u,(x, 0)=—r P

6 (x—x0) .
CO

, , . 8 IF0 V ( - 1 ) " . 2/1+1 2/1+1

3- u ^ = ^ L i p ^ i r m ^ r ’, x c o s ^ r a x ‘ -
n = 0

//wit: u (x, 0) = —|rX, u*(x >°) = 0 •

CO cos-^j-x s i n a t
ip v*
4. “ ( + ' ) = — 2_,
n= 1 1+
Pn pl

where ftn are the positive roots of the equation ft tan ft hi.
00

5. u{r, 0 = £ Cne ' a2A" '- s i n — r, where //„ = ,

n= 1
i\
- __ 2 & i t + ( R h - \ ? P s in -A d r,
C” ~ R R*ri,+ ( R h - \ ) R h P A K

H„ are the positive roots of the equation tan p —

u = vjr and a2vrr = v tt.

326
 Answers and Solutions
00

6. u(r,t) = 2 _ i Cne °JV y 0 B(r),

n= 1
where <Z>„(r) = (1 —/?! R{) sin ]/?.n r+ ]/2„ cos]/An r ,
i?2 ^2
C„ = 5 r/(r)4>„(r)dr, J 4>5(r)dr,
i?i Ri
2„ are the positive roots of the equation
, -rr.~ W -h M -Q -h R A
ta n \ 2 (R2 R,) '

7. (a) i?cr = 7i a I \/ 15 ;
(b) i?cr = 0;
(C) i?cr is equal to the smallest positive root of the equation

( 1 - h R ) t a n ^ - R = $ - R.
a a
8. For boundary conditions of type I:
nl n2 p2\ , . 7in .np
= + 0 n.P ( x , y ) = sin— x sin — >>
(n,p = 1, 2, ...)•
In the case of { l — k), ?.np = ji2/l2{n2-\-p2),
K i = ;-2,i = 5 n2/l2.
, , , - . 2 71 . 71 .7 1 .271
but &it2(x, y) = sin— xsm-^-.y ^ ®2 ,i = sin— x sin-y-y.
Therefore, to a single eigenvalue X = 5ti2//2 there correspond
two linearly independent eigenfunctions.
For boundary conditions of type II:

i
K ,p = n
i l n2 , P2 \ r
0 n,p(x,y)
\= nn nP
COS-j-XCOS ~ y y

(n,p = 0, 1, 2, ...);
for boundary conditions of type III:
^ 'n , p P ‘n '~ \~ tX p >

0 n,p(x, y) = ( i V cos | / pn x-\-hi sin ///„ x)

X (y'cCp cos \/a py + h } sin ]/apy),
pn and ap are positive roots of the equations

327
 Mathematical Physics

tan ]/// / tan ]/a k =

hihi—fA /z3/z4—a
9. For boundary conditions of type I :
zz2 ,1 Jpz
^n,p,q — /2 zt2
nn Tip nq
0 n,p,q(x, y, z) = sin— A- sin — y sin z ( n , p , q = 1,2, ...);
/ /C YYl
for boundary conditions of type II:

/•n, P, <7 - ~712' — + ~ + - ‘74 i ,

\ /2 k 2 m2 1
, . Tip nq
0 (x, y, z) = c o s - r- x c o s - r y c o s — z (n,p,q = 0,1,2,...);
/ K m
for boundary conditions of type III:
^'n,p,q kn
0 n,P,q(x,y, z) = (|/^ nco s|//inx + /j1sin]//iB.x)
x ( v / ^ cosj/apy+Zza sin ]/ap j )
X ()/0, cos ]/j8,z+A5sin j//^ z),
lln,&p,Pq are the positive roots of the following equations:

tan I t a n ]/ a £ = ~
/Zj /?2— /z3/z4—a

tan ]//? nz =
hsh6- p

10. (a) <0n,p,q a V^n,P,q 0:1 "J/ J2 k2 HI2

(n,p, q = 1, 2, ...);
(b) c'Vp = tf/<(l>p 1/^, where /^n p is the root of number p of
the equation y '+1/2(/0— ^ - y n+i/200 — 0, R is the radius of
Z/<
the sphere n = 0, 1, 2, ..., p = 1, 2, ..., where Jk{z) are Bessel
functions of order k (see Chapter 11)

2 lg 2/z+J_
n . » (* ,o = ^ { 7rfl(2/7-(-1)3 C° S 21 ° (2/z+l)2
n=0 V

328
 Answers and Solutions

2«+l } . 2/Z+l
rrx— - / -v .
~2I a 2

8F0/ V i : 1)" • 2 « + i
12.
(2/z+ l)2 Sln 2 / ^
/7—0
2 /7 + 1
X cos - 2 r . ^ t.

Hint for Problems 12-23: the solution should be sought in

the form of the sum u = v(x)-}-w(x, t) where v(x) satisfies the
equation and the boundary conditions of the inhomogeneous
boundary-value problem under consideration, whilst w is the
solution of the corresponding homogeneous boundary-value
problem; v(x) describes the steady state, w the departure
from this state.

13. u(x, t) = t/3+ ,c ;(x)+ 'y '<C„ e (a2/ln+/')<sin -^ -x , where

n=i
l [ 17/ ]/1i
v(x) = ----------- — (W|—w3)s in h — (/—x ) + (w2—z/3)smh— x
s in h

n2n2
= c «= y \ sin+ r - d-
12
6

14. u (x , /) = uz-'rv(x)+ c n e (a2Xn+hl)t

n= I
X (/z, sin ) / ?.n x -f- ] / c o s \U n x ) ,

( //l 3 V>>)
w h ere z;(x) = / l i e a + /l2 e a • T h e co efficien ts A\ and
A2 are d e ter m in ed fro m th e fo llo w in g e q u a tio n s:

/7,(«3—Mi)
A\ A2
\/ hi —a h l

}/lh a = / / 2W2,
h2-V e Jr A 21h2
a

329
 Mathematical Physics

c n= jj {<p ( £ ) - u 3 - * > ( £ ) } ® « ( £ ) d^>

II*,
= ht sin \/T nx + \ / Xncos j / x „ x .
_ fa _|_/^ _
A„ are the positive roots of the equation tan \/XI = 1 * |/A.
/ —Al1/^2
I
15. Q(f) = S 'J i/f o /)&*, where S is the area of the transverse
o
cross-section of a cylinder,
00

/ \ \ ' 4W0 2n+l

-a^J sin —
u(x,,)= u „ - 2 , ^ + 1 ) e 5 — nx.
21
n =0

X„ = — ( 2 « + l) 2 or Q (0 = - i j j «,(0, r )d r.
/
16. 2 ( 0 = w(x, O dx ((Puxx—fiu = ut),
0
oo
where u(x, t) = v(x)-\- C„e~a2A,,,sin 71XC -P*
21
/7= 0
aun
v(x) = — cosh -V?
— (/—x ) ,
/- i/fl a
|//9 cosh----- /

2«+1
sin- -71 £ d|.
= - A f) TT

n= 0

where Xn = T ( 2 n + l f / 4 l 2, a2 = l/i?C .

18. v(x, 1) = + 2 EeR> £ e—

sinj/An (l—x)
X —=
A [/?(/?0+ / ? / ) + ^ o ^ c o s A / ’
330
 Answers and Solutions
where a2 = 1/RC, X„ are the positive roots of the equation
R tan],/X„ I = —Rq\/ X .
w
4H, - a 2Xn> • 2 /1 + 1
19. H(x, l) = H0- , .. | i ^ sin six,
7i yZ_J
z—i —2/z4
2/z+ l 21
n=0
where a2 = c2/4 tio/li, Xn — Jiz(2 n+l)z/4l2.
00

20. u(x, o = ^ y {-<P„ (*„) (1 (x),

n«=1

where = — (2/z+l)2, 0„(x) = c o s - ^ ~ —n x .

k u xx + Qd(x— x 0) = cp ut ,
k is the thermal conductivity.

21. „(*, 0 = “o+ t (i - t ) - S 1?

nFI=
- 0ft

Xe —
°2V" —1;—t^tCOS———
2/z+l n x .
(2 /i+ 1)2 21

where Xn = - ^ — ^ —, k is the thermal conductivity.

00

22. a(x, 0 = » W + ^ C ,e " “y» '0 ,(x ),

n= 1

where v(x) = -— x + C 0(*+ /z), C0 = - - | l + ^ j l 1++/z/+/z2'

Fl
I

c"= wVS®(f)0”(f)df’
+/zsinj/2„ x,
2h ] / r
positive roots of the equation tan yX I = ^ 2- .
2

23. '2 's in ^ - r ,

n=>1

where C = ^ — , p„ are the positive roots of the

n pn(p2+/z2i?2-/zi?) ’ F
331
 Mathematical Physics
equation tan ,u — —p,/(Rh — 1), h is the heat transfer coefficient
in the boundary condition ur (R, t)-\-h[u(R, t) —wj = 0.

Qr
24. &u = 0, ur(R,(p) Q
uv (r, 0) = (r, 7i) = -- -Q- r
nkR ’ 2k R 2k R'

a(r, <p) = 0r • Q V / ' - \ 2n+I cos(2n+ l)<p

2kR Smrp r 2 nk Z-J \ R ) n{n+\)

Hint: first find the solution of the Laplace equation of the

form rv(cp) satisfying only the conditions: 0) = Qr/2kR,
uv (v, t i ) = —Qr/2kR, and the deflection zv(r, cp) from it. Then,
u = rv(tp)+w(r, (p).
25. a20XX = 6 tt, 0 (.v, 0) = axjl, 0,(x, 0) = 0,

0(0, 0 = 0, '1,(1, l j = - k a2 = p G j p .

(h—Io/4) sin— x
Ofln
»(.Y,0 = 4 a ] T
Iokn(2pn sin2ia„) I
n=i
where I is the polar moment of inertia of the transverse cross
section of the rod, G is the shear modulus, 7) is the moment
of inertia of the rod, p is the linear density of the rod and it,,
are the positive roots of the equation tan u = I J I 0p-
, 2n-\-1
4a& l2\ ^ 0)1 ---- 1— a^it — (2/z+ \)na sin cot
26. (a) U(x,t ) = —^ r ^ (2« +1 )2[to212—(2 « +1 )27i2a2\
n=1
. 2n~\~ 1
X sin----— x;

(b) Replace sin cot by cos cot in the last formula (co ^ ktxjl,
k = 1 ,2 ,3 ,...).
, . Tin
00
col sin — a t —n a n sin cot
2F0a l S p . nn . T i n
27. u( x,t) sin — t sin — .V
71T Z -J (uo2l2—n2a2n2)n

(oj ^ Tinjl, n — 1 ,2 ,...) and similarly for F0 cos cot.

332
 Answers and Solutions

7T2a2
28 ■ i/(x, t) = —^ s i n — e ~—1,r2 (" '- "+ 0 (t) dr

X---<* —?T2/?2tf2
+ ^C „e /2 'sin — .v,
n=l
2 (* 7Zft
where A' is the thermal conductivity, C„ = ~ \/( + ) s in — £df
, I 0 I

29. „(*, 0 = M r . V L n i ^ , _ V cos. ^ , + V . \

Cp Z— I \ / 71/7 / 77/7 /
n=l
. 77/7
sin—- x
X
l2vl~\-n2n2a2 '
i///7t: the equation for u(x, t) is of the form
/
a2uxx- h u - \ ------ 6 (x—v0t) = ut, 0 < t <
cp vn
30. (a) For co + ^—na (n = 0, 1, 2, ...)
uu
. 2«+l . 2/7 + 1
i/(x, t) — v (x) sin &>?+ ^ | C„ sin— —— nat sin—2 /— A"’
n= 0
. a)
. sin— x
Aa a —4 co f . 2/7+1
« O) C„ =
ES co ' ^ ( 2 F + T ) f (’ )Sm“ 2 / ^ f d i
cos — /
a
/■u\ f 2 /7 0 + l
(b) for co — — — — 710
u (x , t) — z/](x) sin c o t J r V 2 { x ) t cos o > t

2/7+ 1 . 2/7+ 1
4~ 'y i Cn sin —+ Ttis/sin——— n x .
n= 0 ~2.1 21
n^no
2/7+1
where C„
na(2n-\-1)
J[w^i+)+£+<+] sin-
TT“
77^ ds,
 Mathematical Physics

. . 2aA , . o)
V*{X) = 1 E S (~ l) SmV X
and similarly for F = A cos cot.
Hint: the solution should be sought for (a) in the form u = v(x)
sin ait+zo(x, /); in case (b) in the form u = ^ (x ) sin co t
+ v 2(x)t cos cot+w{x, t), where a(x) sin cot (correspondingly,
zq(x) sin cot+v2(x)t cos cot) satisfies the equation and the boun­
dary conditions of the problem.
00

2 qR2 1 &
31. u(r, t) = Fi(r)+tF2(r) e r.
k xr x:
n= 1
pi cos p n

p n are the positive roots of the equation tan p = p,

qR 3R2—5r2 3qa2
Fx( r ) = u 0- F2(r) =
kx 10R2 ~k^R'
Hint: determine the solution of the problem a2v„ = v , ,
v(r, 0) = u0r, k x[Rvr(R, t)—v(R, t)] = q(u = v/r) in the form
v — f x(r)-\-tf2(r)-\-w(r, t), where (fi-\-tf2)jr is the steady state
satisfying the equation and the boundary conditions and
wjr is the deviation from it; w is the solution of the homo­
geneous boundary-value problem.
32. u(x, t ) is the solution of the problem
d/dx [fc(x)i/J = p{x)ut, u(0, t ) = u(l, t) = 0, u(x, 0) = /(x),
0 < x < x0, 0 < x < xQ,
where k(x) = p(x) =
k 2, XQ < X < I, X0 < X < I,

ux(x, t), 0 < x < x0,

or u(x, t) =
u2{x, t). x0 < x < /,

a\(ui)xx = («i)(, a22(u2)xx = (u2)t , a2 = — (/ = 1, 2),

Pi
ui(°> t) = 0, u2(l, t ) = 0, Ui(x0, t) = u2(x0, t),
ki , t ) = k 2u2x(xQ, t ) , u (x, 0) = /( x ) ,
00

u (x, 0 = ^ C , , e ' ,‘»$„(x))

71 = 1

334
 Answers and Solutions

• Pn
sin— x lt 0 ^ x < x0,
pn
•
sin— x0
a,
where # n(x) =
1 Pn
s in a- 2- V—*o)» x0 < x < /,
sin-^—( / - x 0)
a2

c n = p ^ j j / ( * ) p ( * ) 0 n(X)d*,

P __
*1 cot— k2
un are the positive roots of the equation — x0 =
Q-i O-i Q2
cot — (x0—/)• The eigenfunctions &n(x) are orthogonal with
a2
the weight p(x) in the range [0, /].
33. u(x, t) is the solution of the problem
a2uxx = [1 + (C0/C)(5(x—x0)]wt,
u(0, t) = uil, t) = 0, u(x, 0) = /(x ),
( u f x , t), 0 < x < x0,
or u(x, t) = \ . . ,
1 u2(x, t), x0 < x < I,
= (Ui)„ (i = 1, 2). u f 0, 0 = 0 = u2(l, 0,
Ui (x0, t) = u2(x0, t ) , u(x, 0) = f { x ) ,
ku 2xixo, t) k u lx(x0, t) Cqut(x0, t),
00

u(x, t) = Y , C n e - ° 2< l 0 n(x),

n= 1
/
where Cn = — \ p (x)f(x)&n(x) dx,

sin p„x V

0 < x < x0,

sin p nXo
= sin p„(l—x)
X0 < X < /,
sin p n( l - x0) ’
//„ are the positive roots of the equation cot/zx0—cot//(/—x0)
C0
p-
|C P
335
 Mathematical Physics
Eigenfunctions </>„(a ) are orthogonal in the range [0, /] with
the weight p(x) = 1+ — <5(a — a 0) .

00
34. u(x, l) = E0Jr 2 l E uC 'y e 12RC

^-'oM'n sin fxn11 jj C /cos//n|l ^|

x {lCC0+ l 2C2+Clfil)fin sin (A„
where p n are the positive roots of the equation p tan p = Cl/C0.
Hint: the eigenfunctions of the problem are orthogonal in the
Q
range [0, /] with the weight p(x) = 1 + ^ - 6 ( x —l).

A A
35. (a) u(r, <p) = ~R rcos<p = —a;

(b) u = A + - ^ y ;

(c) u = Axy;
AEB B—A 2 2
(d) it = — + - 2 r 2 -~- ( a 2 - v 2) .

Hint for Problems 35-41. See Example 1 of Section 6.3.

36. Problem (a) is incorrectly formulated since the necessary

condition ^ ~ l ds = 0 is not satisfied;

c on
(b) u = A R x \~D ;

(c) „ = ;

(<0 « = + 4 B) y ~ -12^ [ 3 (- v 2+ r ) i - 4 .- )] + c ,
where D is an arbitrary constant.

37. u = u(r) = — The capacitance per unit

In k 2/ R ]
length of the cylindrical capacitor is C = l/(lni?2—In ^i)-
336
 Answers and Solutions

Hint: the capacitance C of the conductor bounded by the sur-

_| j* \
fs.cc S in three dimensions is C — ----- j ----- do*, in two
4 7i uQs dn

dimensions C = - § ——ds, where L is the contour,

I t i Uq l on ’
u0 the potential of the conductor, du/dn = E„ is the normal
component of the electric field stress tensor.

38. C =
1
u0 __ L
i i - i L l + i
La c e2 \ b c,
Hint: solve the problem A«j = 0 for a < r < c, Ah2 = 0
for c < r < b, u,{d) = u0, u2(b) = 0, Wl(c) = u2(c).
du
- £ , ------ , «0 is the potential difference between the
E> r=c dn r —c
plates of the capacitor.
1 I £i - £ 2 1
39. u - u0 for R < r < c.
l , £l - £2 1
/? 1" £2 C
£l 1
e2 '
It = Uq for r > c.
1 , «i — 1
R 1 £■> c
Ui—u2+0.25A(Rl—Rf) R^ .
40. (a) u = lh+ ^ . ( r 1- R ^ + ~‘- "V "" ln ^ ;
\nR7—Ini?, r

(b) u = w, + -^-(r2- ^ ) + /?2|w2- y i ? 2| l n —

1
41- u = I ^ - R ^ - R . R ^ R . + r J — - -

.2/1+1 2n+l
CO
cosh —--— 7ix cos ——— 7iy
2b 2b
42. =
in ,2/j+l
n= 0 (2 n + 1 ) cosh - —— 7ia
2b
Hint : see Example 2 of Section 6.3.
22
337
 Mathematical Physics
CHAPTER 5

RC G1
1. u(x, 0 = ^ — 0| x- | /
41

1 r (r~ ro)2 __(r+_roTn

2. G(r,r0;t) = ---------= e 4‘2' - e ~ 4fl2' .
8n rr0y n t [ J
Hint: reduce the problem to the one-dimensional problem by
r d(r—r0)
substituting u = v jr: a2vrr = vt, v(r, 0) =
4 71
R
• .........> - ? [ • ( r S M iS r

4. (a) u(x, y, z, t ) = « (]/x 2+ y 2+ ( z —z0)2, t)

+ u ( ] / x 2+ T 2+ (z + z 0) 2 , /),

(b) u(x, y, z, t) = u(]/x2-i-y2~h(z—z0)2 , t)

— « ( ] / x 2+ t 2+ ( z + zo)2 >0»
where u{r, /) is the solution of the last problem.

2 j\ \ } /4 a 2(t—T) J \ ] / 4 a 2(t—r)
10

7 f e ) ) l e(r)dr
P-D2 ('•+D2i
6. w(r, /) = ~ 4a2, _ e 4a2/ ^
1s ^ 5 fW)
|/ 4 a27it

1 »» t2 (r-t)1 C+f)2
g 4a2(/ —i ) __g 4a2(/ —t) d | dr.
1/471 a2 ,) ,1 ] /1—t
00 }
338
 Answers and Solutions
00

7. G(r, r0; t) J _ C e~a2X2tJ0 (?.r)J0(Ar0) AdA

2 7Z *J
0
r*+r20
l 4a2/ T rr0 \
y0 2 a2t
4n<J2t

Hint: solve the problem a2\urr-\-— urj = ut,

u(r, 0) = - ^ - — d(r—r0), \u\ < co.

I n r0
The solution must be sought in the form
00 00

m(/-5 = J J U{p, t ) J 0{/.p)J0(Ar)A dAdp (see Chapter 11).

o o

8. u(r,
0 ' '
/ 00
r2+i2 rt
4a2{t-z)Jo d£ dr.
t —r 2 a2(t—t)
00
e -/ir (x-xp)2
9. G(x, x 0; t) = — -e 4a2' .
| / 47ia2r

CHAPTER 6

1. (a) G(M ,P) = — (ln— ----- In—

2n \ r MP t'orM P1 J

where Pj is the point symmetric to P relative to the circle,

fo is the distance of P from the centre of the circle;

(b) G(M, P) — —-----------— I.

4n \ r MP f QrMpl J
n—1
2. G(M,P) = J ] [° Cr{M, P,„)~Gcr(M, P J ] ,
m=0

22 *
339
 Mathematical Physics

where Gcr(M, P) is Green’s function for the interior of the

circle, Pm and P,„ are points with the polar coordinates

and (r°’ ~ ^ 7Z~ <Po) respectively; (r0, y0) are

the coordinates of P.
00

1
3. (a) G (M, P) =
Z_J Arc \ fMP„ ntp' I ’
n= 0

where P„ are the points with coordinates (p„, 0o,<po), P'n are
the points with coordinates (<pn, 0o, tpQ) and (p'0, 0o,<po) are
the coordinates of P,
A ) for n = 2k,
*2,
e„ =
for n = 2k-\-1,
i(f)
A for n = 2k,
(£ ) Po

A A for n = 2&+1,
Ri Po
i 2k
A Po for n = 2k,
Rz
Pn [2k+2
A p0 for n = 2 k-\-1,
A
i 2k
A A for n = 2k,
Ri Po
Pn :
R 2 \ 2k R 2
for ii = 2 k + 1.
Ri Po

(b )C (« ,f) = - V I. l’"------- In

where Pn are points with coordinates ( p n,( p 0), P'n are points
with coordinates (p'n ,<Po), ( P o , T o ) are the coordinates of P;
P n > P n , e n and <?' are determined by the same formulae as in
case (a).

340
 A n s w e r s a n d S o lu tio n s

1
4. G(M, P) -
r MP'n

where Pn are points with coordinates z0+2nh), P'n are

points with coordinates [x0, y0, z0—(2«+l)A], (.x0, y 0, z 0)
are the coordinates of P .

CHAPTER 7

B, rC ,
r

1. u(r) — - 4 ^ S (l/f-l/0 fV (f)d f+ C , R, ^ r < R2,

Ri
Djr, R2 < r.
Rz
where D — £2p(£)d£,

Ri
B = C = D/R2+ 4 n \ ( 1 / |- 1 / R 2) | 2p(f)d£.

_ , ^ I 47zRPo> r < R’
2. wtr) = t ,
\A n R 2p0jr, r > R.
| 27ip0(R2- r 2/ 3 ) - M f r 1, r ^ R,
“( r ) ~ i M O / r - l /r O , r
where A/ = 4/3 7iR*p0, r — ]/x 24-y2-\-(z—h)2,
fi = l /^ + Z + ^ + Z z ) 2,
p0 is the charge density, (0, 0, h) are the coordinates of the
centre of the sphere of radius R.
Hint: calculate the effect of a perfectly conducting plane
z = 0. Reflect the original sphere in the plane z = 0. The
solution in this case then takes the form
M
r < R,
u(r) =
ir - 4rx / r > R,

341
 M ath em atical Physics

C is determined from the condition that the solution must

join at r = R.

m (4 - —In/?—
\ 2 2 R2
4. u(r) =
M In— ,
r
where M = 7iR2p0, p0 is the charge density, R is the radius
of the circle.
5. The potential of the simple layer 0 x < 1 of density p0 is
/
1_____
v(x, y) = p0 jj In d f.
o |/ ( f —x)2jr y 2

«. w( M) = v0(*] —
cos <p
djP=v0>f ,/(f d£ •
2ti
- , * , X J C (p2- ^ 2) /( 6 )d 0
7. (a) ro(p,<p) 2 jiJ / ? 2 _1_ p 2 _ 2 ^ p c o s ( o _ (p) ’

CO

1 f y /(g ) d£
(b)
* —
3OO ( i s - x f + y 2 •

CHAPTER 11

a2<In / \
1. «(/', o = 8/4 \ —3~ - — -e
Z_j a ^ iC ^ n )
/2= 1
where a„ are the positive roots of the equation J0(a) = 0.

2. « (M ) = ^ C „ e r),

where a„ are the positive roots

of the equation ccJ'0(a)-\-hRjQ{oL) = 0,

342
 Answers and Solutions

1
C„ = /H iJ i-tU .

f( r ) is the bounded solution of the problem = o,

/C

/ ’( * ) + ¥ ( * ) = 0, f ( r ) = % d f - r > ) + e .* 2 j „ j q
2h tii —
k
00
a-a21
n
3. //(r, ?) - H0- 2 H 0 V _ i _ R2
1 )
' / _ J

where a„ are the positive roots of the equation J0(<x) = 0. The

flux of the magnetic induction through the cross-section of the
R 27i
cylinder is CP = § jj pH(r, t)rdrd<p, where,« is the permeability.
o o
<JU

4. u(r, t) = /(/■ )+ Cne - fl2^ '0 n(/-),

1
where 0„(r) = ^ ( /^ i ? i ) A r0()/ V ) - A fo ( |/ ^ i ) ^ o ( |/ ^ 0 5
*2

c- = w - o f 5 d r- m = ^ lni - -
/„ are the positive roots of the equation $ '(i? 2) — 0.
5. a2&.u+Q/p = utt)u(R,t) = 0, u(r, 0) = u,(r, 0) = 0, |w |< o o ,

2R2Q S T 1 , /«» \ av-n

r / COS
n=l
where a„ are the positive roots of the equation J0(a) = 0,
R2—r2
m = -Q is the stationary state.
4a2p
i I I • ®/i
2 A „ ,R ^ M s ? '" ! "
6. (a) u(r, t) = f(r ) sin rat-
dp Z - j <xn2((»2R2- a 2ci2
n)Jl(an)

(co is not equal to any of the eigenvalues aan/R),

343
 Mathematical Physics

“ d
\ a I -1
/ to \
70l — R)
a /
T I-=rr
“n \ cos——
anfl t.
70
R R
(b) u(r, t ) = f ( r ) cos to —
r (to2R 2— a2a l ) /] (a„) ’
n^= ,1 an
00

(c) it (r, t ) = ^ ^ J 0 rJ Wn(t),

n=1
t
where \F„(t) = B n ^ sin-^— (/—t) sin tor dr,

2A
, for = R2,
Bn aanJ 2(an)

B.
2A ^ol—' > )
Raa„ Jf(an)
CO

. a„r . a a„
7. (a) w(r, 0 = /( r ) sin a>f— ^ Cn
/J=l

2,4 a>
where C„ = ~ H dr,
a)
aanRJ0\ — RU^a,,) 0

m = a

'•Hr*
a„ arc the positive roots of the equation J0(a) = 0;
00
aotn
(b) u(r, l) = f ( r ) cos to t— ^ DnJ0 -r cos

aRa„
where Dn = C„
to

344
 Answers and Solutions

8. u(r, <p, t) is the solution of the problem a2Au = utt>

u(R,<p,l) = 0, u(r, <p, 0) = 0, iit(r,<p,0) = P^-d(r--ri)d(<p — <pi),

, 2P V ^ ^ aal„,
ru ^ L Z -. 3 .4 " > K « > ) r 7”( J? T m /! '■
[1, n 7^ 0,
where e„ = t a£n) are the positive roots
^ 2, n 0,
of the equations J„(a) = 0.

“ s in h 4 z /J -r'
R
9. u(r, z) = j/0- 2 ( F 0- K 1) y ------- --
7 . _• i &n
n=i sinh-—L/jan,/1(an)
where a„ are the positive roots of the equation / 0(a) = 0-

10. u(r, z) = —~ — ,
v } 7iR2a tiR
- /
2 7xo Z_i s
n= 1
sinh^ zy°(x*‘Wir'') + const,
cosh-—hot2J I M

where a„ are the positive roots of the equation -A(a) = 0.

R h1cosh— z
R
11. u(r, z) = --y ^ ^
27ikZ—i <x2nJf(an)
n=1 a .s in h - ^ h + J tfa cosh /i

X70( » ,
where a„ are the roots of the equation J0(oc) = 0.
Hint: the solution u(r, z) should be sought in the form tt = v(r)
-I~w{r, z).
The formulation of the problems for v(r) and w(r, z) is:

v: Av -j- = 0, \v\ < c o , v(R) = 0,

2n r
w . Aw = 0, \w\ < oo, m(R, z) = 0,
wz (r, —h/2) —h1w(r, —hi 2) = hiv(r),
wz (r, h/2)Jrh lw(r, hj2) = —hYv{r).
345
 Mathematical Physics

The origin of the coordinates should be taken in the centre

of the cylinder.

, , , , V 2 qR J° [ r r) - - - »
n= 1
qR
where F0(r) = u0■
4k

Fy(r) = 2qa2— , «n are the positive roots of the equation

kR
71(a) = 0.
13. (a) = (nk/h)2-h<x^lR)2, are the positive roots of the
equations J„ (a) — 0,
. 7ik la ™ Wsinwy,
\n —r-zJn —- n
&n,m.k(r, <p, z ) = sin
h \ R f (cos ncp.
(b) A„,m,k = (n k/h f+ ia W /R )2, a,(„n) are the positive roots of the
equations J'„{ol) — 0,
nk I «Ln) \ jco sn ?,
0" . «"•2>= (sin
cosT z J ’ \~ R ~ ‘ I n r,
(c) = vl+ Q z^/R )2, «
m)are the positive roots of the
equation aJfa)+RhJ„(ai) = 0, vk are the positive roots of
(hxFh^v
the equation ta n i '/7 = 2 — j —j —.
v —u\ h')
y («)
Am cos ncp,
&„,m.k(r> <p’ Z) = ll\ ( Z) J n R r sin tup.
lF k(z) = vkcosvkz+hi sini'feZ.

/ r (n) \ Tin I r,(») \ 2

14. (a) <P„,m(r,(T) = J„nial — rlsin^-99, ?.n>m = ,

y<>) are the positive roots of the equations Jn^ta.{y) = 0;

/ y(n) \ Tin
(b) (p) = ^ „/al - — r Icos ^ <p, 2„

yb1) are the positive roots of the equations Jh^aiy) — 0;

346
 Answers and Solutions

,(n ) ' 2

n,m = IIL__
ft ,f ’
are th e p o sitiv e r o o ts o f th e e q u a tio n s
yJ'v'n(y)+RhJVn(y) = 0,
v„ are th e p o sitiv e r o o ts o f th e e q u a tio n
(Jh Jrh2)
t a n v a = —5— r - - v .
v —ni h2
00 00

15. w(r, <p, t ) = 2 _ , 2 j Cn,fce

n= 1 /c= 1

C n>* — - d 2„ r r ' ■jj ^ /0 , <P)rJn*ia (~ -r j sin -~ (p d rd < p ,

nj nJ ' '
yin) are th e p o sitiv e r o o ts o f th e e q u a tio n s Jn^aiy) = 0 ,

;(«)
Ak —
— (y P
1R
CO oc 00

16 u(r, <P, z, t) = V (c„jm[t c o s n <P

n= 0fc= 1m= 1
y [n) \ . 71 m ,,
(
w h ere
R h2n
yln) V . Tim
-r r Ism—r—z
\S
000
5r/(r’*z)y-nR / h

X c o s «<p dr d z dtp,
R h 2n ,.
4 (*f* f* / yln) \ . Tim
z>- ‘= j j j r f( r - 9’ , V - M V
X sin n<p dr d z d<p,
y [n) an d 7nm>t are d eterm in ed as in P ro b lem 13.

19. / , /2(x) = l / - —sinhx, /_ i/2W = l / — coshx,

71
K ± m {x) = y JV,/2 ( x ) = - / . 1;2W - |l/ / 71X
- COS X ,
-V -
347
 Mathematical Physics

N - m (x) = J„z(x) = l / - sinx, H $(x) = - > • ] /

tf iW M = y 71X
e ,JC, H V A x) = - • ] / | e -

/ 2 p -\X
7lX

2°. U( r ) ~ ^ R) h ( M , V2U—fi2U - 0, w|r=i? =

w0
21. ?;(r) = K0(fir).
K a(PR)

u2~ u l . 71n
22. u (r, z) = sin—^ z ,
h
n = l '

where C„ = -{(tf0—w1) [ ( - l ) n- l ] + ( - l ) n (u2—« i)} - 1

ij — R
°\ h
u(r, z) is the potential of the field E, i.e. E = —V2u.
Hint: the solution should be sought in the form
n = A(z)+B(r, z), V2A = 0, >4(0) = ux, A(h) = n2.
00

23. u(r,z) = y ^ C ,,/0(— HCOS-— z,

n= J

where C„ = ^ /(z ) cos z dz.

cos-j-zll o
M t * n
\7i{2n — 1)
4 h.
* ”0\ ^ - ^/;---- I . 7i{2n— 1)
24. « ( r , z ) = ^ - \ i - T — ----- — sin ----- ■------ *■
7i L— i r(2 « —
i7ii2n —1)
»=• —' r

-a2 / 2 n2k2\
25. u(r, z, f) = v(r, z)+ ^ ^ C n>*e l a*+- * r y
n= 1 fc=l

/ a n \ . 7ik
y J A ~ R rr a i r z ’
348
 Answers and Solutions
Rh

c »-‘ = \ c (r' z)rJ° ') sin zdzdr’

where a„ are the positive roots of the equation J0(cc) = 0;

00

u0 2Mo V 1 (—1)" r l nn \ o- nn
v(r, z ) = X ^ + — 2_, - 7 — -T M ~ r T nT 2-

A
Hint', the solutions should be sought in the form
u(r, z, t) = v(r, z)+ w (r, z, /)>
where t>(r, z) is the solution of the problem (see Problem 22)-
V zv = 0, v (R ,z ) = 0, v(r, 0) = 0, v(r,h) = u0, \v\ < oo.

CHAPTER 12

* m v
1. ^ ( 0 ) = (- 1 )' 22k(k I)2 ’ P2f t + l ( 0) — 0.

2. They are orthogonal with the weight (1—x2)k. This follows

from the orthogonality of the associated Legendre functions.
uu

3. u{x, t) = £ (C„ cos a ]/2n{2n— 1) t

n=1
+ D„ sin a \/2n(2n—1)t} P2„-\(xl0>

Cn = ( 4 n - l ) ^ ( | ) P 2n - i( y ) df,
o

J3> = ____ y )d { .
a]/2n(2n—\) j]
4. E — -V h , where u(r, 0) is the potential of the field
00
' Z-j
> P2n+1(C0sfl),
2
n=0
u(r, 0) = Oj
V 1- V 2
/ d \ 2,1+2
V, 2 ^ C nU P2n+1(cosO), r p P ’

n=0 ' '

349
 Mathematical Physics

4 « + 3 (-1)"(2«)!
Cn =
2n+ 2 22n( n \f '
00

5. (a) u(r, 0) - - e ^ —— P n(cosQ);

n= 0

\ ' R2n+
(b) u(r, 0) = —e / I yn+l *•«+* COS0).
* * t c\ r
n=0

///«/: the resultant potential V(r, 0), due to the point charge
of the induced charges, should be sought in the form of the sum
V(r, 0) = e j r ^ u i r , 0),
00

r < R,
XjCnnr)
n__n
=0' ’ p^cos^’
where u(r, 0) =
R
v Dn i P„(cosO), r > R,
n= 0

where rx is the distance between the point (r, 0) and the point
(r0, 0) at which the charge is located. Use the expansion

(cos0), r < r0,

1
00 *■
*
Pn(cosO), r > rn.
IS (-
n= 0 '
The coefficients C„ and Dn are found from the condition V{R, 0)
= 0:
er'a eRn
Cn = pn +l » A, =
6. (a) If the charge lies outside the sphere at the point (r0, 0)>
r0 > R , the electrostatic potential is
\ u x(r,0), r^R,
u(r, 0) — j r
where
2«+l
ih{r, 0) = e \ ^ ~ e, + («+l)e2 r£+1 Pn(cosO),
n=0

350
 A nsw ers an d Solutions
CO

, .. e e —e
u2(r, 8) = -------\-e —
£->ri
■y __ «
Z—i n e1-|- («
(n + l)e2 /•on+1'-^ r P '‘(COSO)'
n= 0
Hint', the expansion coefficients are found from the conditions
diii ! du?
u{{R, 8) = u2(R, 8), e2‘
1 dr [r=R z dr j,=* ’
where r, is the distance from the point (r, 8) to the point (r0, 0),
where the charge is located;
(b) If the charge lies outside the sphere (r0 < R), then

e
u‘(r’ e) = ^£i r i - +n = 0 + > + i ) eT ^ p "(cos8)'
CJJ
2/1 + 1
u2(r,, 8) = e ---- ^ P n ic O s O ) .
Z—: «e, + (« + l)£ 2 f
n=0

7.
00

¥ 2 h r ) [£ .(0 )+ i> .-!(0)Ji>,(cos,)) - ^ P 1(cose),

n=0 ^ ^ „
r ^R ,
©( ^ ) = ,

^
| ^ Z_i
2 ( 7) + [£ .(O )+ -P .-2 (O )]P .(co s0 ), r > R.
( n=0

Hint: solve the problem by the method of separation of vari­

ables. Then

Cnr nP„(cos 0), r <R,

„ 1 _ .

The coefficients C„ and are found by com paring these

form ulae with the expansion in powers of z for the potential
at a point on the z axis (which is perpendicular to the disc
and passes through its centre) which is calculated directly:

l/ f e 0) = —

351
 Mathematical Physics

/ y 4 //-f3 / r \ 2"+1
8. v(r, 0) = P 2n+1 (cos 0).
2noR 2—J 2 n + l U ;
/?=0
In view of the symmetry of the problem v(r, ti/2) — 0. There-
00

fore, in the expansion v(r, 0) = CkrkPk(cosO) the coeffi-

n = 0

cients C2n with even indices should vanish. The remaining

coefficients are determined from —ctvJR, 0) = /•

271RA sin 0

, Q Q n + l - R h rn
0rn n , ..
9. u(r, ° ) - ^ + 4- 7 2 j ' n + R h / p +f p »(cos °) >
where is the distance between the point (r, 0) and the source
(ro,0 ).
Hint: the solutions should be sought in the form of the sum
Q -v(r, 0),
u(r, 0) =
4jrA:r]
where a(r, 0) is the solution of the problem
1 Jl \
A , _ 0. * . , ( * 0)+l,v(R, 0) =
Use the expansion of l//'i into a series in terms of the
Legendre polynomials.
p V i 1 i-2,,+1_ /?2”+1
10. u(r,0) = - ----A ^
2 - i [ R 2n+l_ R f71+ I n+1

~L R f ,+l- r 02u+1 Rln+i

RV'+l- m n+ i PAc° 50)-
The density of the induced charges is

, -<? V (2 //+ l)(^l)(tf2,,+1- - - 2n+1

2 ,l+1—,'o"+I) ^"■1p„(cOS 0),
1 = <’ u '" = - 4 r L RV'+X- * ;n+l ,.«+1
H=0
OO
e . ( 2 » + D ( / - r ‘- . R ? +‘)
n= 0

where r ( is the distance between the point (r, 0) and the charge
placed at (ro, 0).
352
 Answers and Solutions

Hint: the solution should be sought in the form

u = elrt + v(r,6), O'= 1,2).

00

11 . «0, °) = - ~ | l - c o s a - ^ [P„+1(cosa)
n= 1
ln
—Pn-!(cos a)] I—I Pn(cos 0) |
■)
„ / rn_ ^ L 1 i r cos0
12‘ ^ 2k \2 h R "h P l + P/z
uu
(4 n + l)P 2n(O)r2nP2n(cos0)
- s (2n+/zP) (2n - 1) (2n+2) i?2n
/J=1
i/w f: the boundary condition for the problem is
f qjk cos 0, 0 ^ 0 < n/2
ur(R, 6)+hu(R, 6) = | 0; sr/2 < 0 < re .

13. w(r, 0, t) = -^^j-ZCO-PnOos 0)

+ Pn(C0S 0) ^ % ( 0 - |^ ^ n +l/2 ,

where afc are the positive roots of the equation

^•Ai+1/200 2'•^n+l/2(a) = ^ ’
t

n ( 0 = — i/"(T)sin-fl“«-((-T )d i,
0Ct. Cl t) iv
* 0

_2 j r M* V . +10( ^ » ) d i-
Ah —
nR n+1 n ( n + 1)
Jn+ 1/2 (a Jt) 1
a*
where w(r, 0, t) is the velocity potential a2Vzu = u„,
u(r, 0,0) = ut(r, 0,0) = 0, ur(R, 0, t) = Pn(cos 0)f(t), |w |<co.
14. ).n m k = l/R 2(a<
i"))2+ k 2 are the eigenvalues, and
23 353
 Mathematical Physics

= ^ y"+‘'J( ^ r)F;(cos9){sintr
are the eigenfunctions, where o4n) are the positive roots of
the following equations:
(a) ^ ,+i/2(a) = 0,
(b) Jn+1/2(k) 2aL^n+ll2(a) =
(c) 2ctJ'a+1p(a)—( l —2Rh)JB+lf2(ci) = 0,
h is the constant in the condition vr{R, 6, q})-{-hv(R, 6, <p)= 0.
co co n
15. w(r, <?, <?) = X X X {Cn,m,kc° s k cPJrD n rn'ks\nk(p}
m—1 n = 0 /c = 0

a2[>}]2
1 / a(n) \ R*
X —^ / n+1/2| — rjP i(co s 0)e

where are the positive roots of the equation 7„+1/2(a) — 0,

R 7i 2n

C„,m,k = 4n,m,k jj \ \ f ( r>6>V ^ J n +m ( ( COs G) COS k<p

00 0
X sin 0 dcp d 6 dr,
R 71 2 71
a (") \
Dn>m,k = A.m.fc JJ^/O , 0, <P)rV2Jn+ll2 —— r |P*(cos 0) sin kq)
0 00
X sin 0 dcp dd dr,
(2 n + l)(n —&)! _ 2, fc = 0,
= x R 2{n+k)\EkJ 2 n +m{a<,n)) ’ £fc \l,
 Bibliography

This book list is included for the benefit of Western readers who
do not have access to the Russian references quoted in the original
text - Editor.
General
morse, P. m ., and feshbach , h ., Methods of theoretical physics
(2 volumes), McGraw-Hill, New York and London (1953).
courant, r ., Methods o f mathematical physics, Volume 2, Inter-
science, New York and London (1962).
Chapters 1-7
epstein , B., Partial differential equations, McGraw-Hill, New York
and London (1962).
Sommerfeld, A., Partial differential equations in physics, Academic
Press, New York and London (1949).
Chapters 8 and 9
mikhlin , s. G., Integral equations, Pergamon Press, Oxford (1957).

Chapters 10-13
Kantorovich , l . v ., and krylov , v. i., Approximate methods of
higher analysis, P. Noordhoff, Groningen (1958).
erdelyi, A., Higher transcendental functions. Volume 3, McGraw-
Hill, New York and London (1955).
Whittaker , E. T., and w atson , G. n ., Modern analysis, Cambridge
University Press, London (Fourth edition, 1927).
Appendix
lighthill , M. J., Introduction to Fourier analysis and generalised
functions, Cambridge University Press, London (1958).
355
 Index

Acoustics 20 Boundary-value problem-continued

equations of 22 parabolic equation 59
Adiabatic motion of ideal gas 21 semi-infinite straight line 43, 47
Airy functions 263-65 solution 45, 121, 131, 134, 146,
Associated Legendre functions 171
280-83 uniqueness of solutions of 96-
graphs of 283 102, 126-31
orthogonality 281
Asymptotic expansion 245
Asymptotic representation 246, Canonical form 4-11
277 Cauchy criterion 129, 197
of cylindrical functions 256 Cauchy inequality 201
of error integral 246, 255 Cauchy integral 269, 292, 297
Cauchy integral theorem 216,219,
248
Bessel functions 223-37, 288 Cauchy problem 28, 32-44, 107-
example 276 21, 169, 170, 172
integral representation 243 Characteristic curves 6-10, 32
orthogonality 227-30, 232, 236 definition 6
zeros of 230, 258, 260 differential equations 7, 9, 33
Bessel inequality 191,192,197,202 method of 32-58, 68
Boundary conditions 46, 47, 48, parabolic equation 8
60, 64, 68, 73, 77, 78, 135, 138, Chebyshev-Hermite functions 295-
261, 262, 287, 288 96
Boundary value, propagation of 47 Chebyshev-Hermite polynomials
Boundary-value problem 25-29, 291-96
68, 77, 80, 122, 137-40, 206, 227 Chebyshev-Laguerre polynomials
application of potentials to 164-67 296-302
eigenvalue and eigenfunction of generalised 297
83-84, 173, 206 Continuous derivatives 315, 318
Green’s function for 170, 176, Continuous functions 303, 304,
205 308, 312, 314-18
homogeneous 88, 97 Convolution 38, 39, 115, 314-18
hyperbolic equation 59 Coulomb field, electron motion in
inhomogeneous 84 301

357
 Index
Cylindrical functions 223-67 Elliptic equations 5, 6, 7, 9, 25
asymptotic representations 245, 29, 131
256 Green’s functions for 124-45
differential equation 279 Energy integral 98
examples 261-63 Equivalent sequence 303,307,311,
312
Error function 110
d’Alembert formula 34, 35, 36, Error integral, asymptotic representa­
38, 40, 43, 44, 51 tion of 246, 255
d’Alembert test 224 Euler integral 215
Degenerate kernels 177, 178, 182, Expansion theorem 61, 97, 199-
194 203, 205-6
Degenerate symmetric kernel 193
(5-function 113, 118, 119, 307,
310-14, 317-19 Force density 14
definition 307 Fourier coefficients 201
<5-sequence 308-10, 313, 316, 318 Fourier integral 209
multi-dimensional 310 Fourier law 24
Differential equation, for cylindrical Fourier method 59-106
functions 279 Fourier series 62, 64
Legendre 270 Fourier transform 319
linear 3 Fredholm integral equations 170,
Differentiation of sequences, term-by- 176, 177, 178, 182-83, 185, 1S8,
term 318 207
Diffusion coefficient 25 definition 168, 169
Diffusion equation 25 Fredholm theorems 183-87, 203
Dipoles, distribution of 158 Fundamental sequence 303-12,
potential due to 158-59 315-18
Dirac (5-function. See (5-function Fundamental solution 108, 111,
Divergence operator 22 119-21
Duhamel’s method 87, 91-96 convolution of 115
Fundamental surface harmonics
284
Eigenfunctions 59-81,83-84,133, orthogonality 285
135, 177, 184, 186, 187, 195-97,
206, 236, 262
definition 61 Gamma functions 215-222
orthonormal 190 Gauss’s theorem 19, 20, 21, 49
properties 65, 74 Generalised Chebyshev-Laguerre
properties of set of 81-84 polynomials 297
simplest properties of 188-94 Generalised functions 38, 303-20
Eigenvalues 59-81, 84, 132, 135, Generalised spherical harmonics 284
173, 184, 186, 187, 195-97, 206, Green’s formula 124-26, 151, 175
235, 236 Green’s function 108-12, 174-77
definition 61 for boundary-value problem 170,
properties 65, 74 176, 205
r-fold degenerate 78 for elliptical equations 124-45
simple (or non-degenerate) 78 for parabolic equations 107-23
simplest properties of 188-94 method of 131-34
Electric field in vacuum 23 partial derivative 172
Electromagnetic waves, propagation Green’s theorem 25, 63
of 301
Electron motion in Coulomb field
301 Hankel functions 237-45, 251,
Electrostatic potential 146 280
358
 Index
Hankel functions-cwzr/wwecf Kernel s , iterated-continued
asymptotic behaviour 248 spectrum of 194-97
definition 258 negative definite 206
linear independence of 241 positive definite 206
Harmonic functions 124—26, 128, symmetric 188-211, 206
130-32, 150, 160, 285 degenerate 193
Harmonics 62 specified in finite domain 208
Heat balance 24 specified in infinite domain
Heat propagation, along infinite 208-11
straight line 107, 112 Volterra 169, 188
in two- and three-dimensional space
119-22
Heat transfer 24 Lame coefficients 320
on semi-infinite straight line 111 Laplace equation 154, 167, 283
Heat transfer equation 25 homogeneous 3
Hilbert theorems 174, 175, 176, inhomogeneous 3
206 with constant coefficients 4
Hilbert-Schmidt theorem 199- with two independent variables
204, 206, 207 3-12
Homogeneous equations 3, 41, Laurent expansion 243
42, 119, 120, 168, 169, 183-86 Legendre differential equation 270
Homogeneous polynomial of order Legendre functions, associated
n 286 280-83
Homogeneous wave equation 32, graphs of 283
49, 54 orthogonality of 281
Hooke’s law 14, 15, 17 Legendre polynomials, examples
Huygens’ principle 57 275-80
Hydrodynamics 20 generating function 268
Hyperbolic equations 5, 7, 8, 10, integral of 272
20, 41, 59 orthogonality 270, 272, 273,
275, 277
properties of 268-80
Improper integral 147-51, 160 zeros of 273
Independent variables, transforma­ l’Hopital’s rule 52, 229, 230
tion of 5, 7 , Linear differential equations 3
Infinite medium, vibrations of 48 Linear harmonic oscillator 295
Inhomogeneous equations 3,117, Linear integral equations 168-69
121, 168, 169 Linearisation 22
expansion of solution 203-5 Localised function 314-16
Inhomogeneous wave equation 40, Lyapunov surface 154, 156, 162
48, 52, 55
Integral equations 146, 168-87
Fredholm’s. See Fredholm Maxwell’s equations 23
linear, classification of 168-69 Mean value theorem 126, 313
problems leading to 169-77 Membrane, transverse vibrations of
Volterra 169, 170, 181, 187
with degenerate kernels 177-78 16-20
with symmetric kernels 188-211 Method of characteristics 32-58,
Iterated kernels. See Kernels 68
Method of separation of variables
59-106, 122, 132
Kernels 182, 183, 184 principle of 59-81
classification 206-8 Method of source functions 121
degenerate 177, 178, 182, 194 Method of steepest descents 248
iterated, expansion of 197-99 Method of travelling waves 35

359
 Index
Morera’s theorem 216, 220, 221 Reflection of waves at fixed and free
ends 45
Resolvent 181
Natural frequencies of vibration 62 Rod, longitudinal vibrations 15-
Negative definite kernels 206 16
Nernst’s law 25 vibrations of 26
Neumann function 244, 258
definition 258
Newton’s law 27
Newton’s second law 14, 16, 19 Schrodinger equation 295
Non-homogeneous equations 185 Semi-infinite straight line 46, 51
Norm 63 boundary-value problems 43, 47
Separation of variables, method of.
See Method of separation of
One-dimensional wave equation 15 variables
Orthogonality 190 Set of functions, definition 81
associated Legendre functions 281 test for completeness of 81
Bessel functions 227-30, 232, Smooth functions 213, 314, 316,
236 318
eigenfunctions 190 Spherical functions 286
fundamental surface harmonics of order n 285
285 Spherical harmonics 268-90
Legendre polynomials 270,272, generalised 284
273, 275, 277 Steady-state conditions, determina­
surface harmonics 285 tion of 25, 28
Steady-state solution 71
Steady-state temperature distribution
Parabolic equations 5, 8, 9, 99 130
boundary-value problems 59 String, vibration of 13-15, 26, 32,
Green’s functions for 107-23 40
Partial differential equations, in Sturm-Liouville problem 61, 79,
physics 13-31 170, 176
second-order 3 Surface harmonics 283-88
Planck’s constant 295 fundamental 284
Poisson formula 52-55 orthogonality 285
physical interpretation of 56 of order n 285
Poisson integral 144, 167 orthogonality 285
Positive definite kernels 206 Symmetric kernels 188-211, 206
Potentials 146-67 degenerate 193
application to boundary-value specified in finitedomain 208
problem 164-67 specified in infinite domain 208-
due to dipoles 158-59 11
due to double layer 157-64
due to simple layer 154-57, 167
volume 146-54
Propagation of electromagnetic waves Temperature distribution 24, 112
301 along homogeneous rod 70
Propagation of heat, along infinite steady-state 130
straight line 107, 112 Term-by-term differentiation of se­
in two- and three-dimensional space quences 318
119-22 Thermal conductivity 24, 130
Three-dimensional space 49, 52,
310
Rational numbers 303 Two-dimensional space 54, 55
Real numbers 303 Two-dimensional wave equation 20
360
 Index
Transformation of independent Volterra integral equations 169,
variables 5, 7 170, 181, 187
Volterra kernels 169, 188
Uniqueness of solutions, for propa­ Volume potential 146-54
gation of heat along infinite properties of 148-54
straight line 107
of boundary-value problems 96-
102, 126-31 Wave equations 23, 26, 32, 38
Uniqueness problem 96 homogeneous 32, 49, 54
Uniqueness theorem 107, 128 inhomogeneous 40, 48, 52, 55
Unit step function 307, 312 one-dimensional 50
two- or three-dimensional 48
Waves, reflection at fixed and free
Vacuum, electric field in 23 ends 45
Vibration, natural frequencies of 62 Weierstrass theorem 311
of elastic rod 15-16 Wronskian 78, 173, 174, 226
of infinite medium 48
of membrane 16-20
of rod 26
of string 13-15, 26, 32, 40, 72 Young’s modulus 15

361