G Stephenson - P M Radmore - Advanced Mathematical Methods For Engineering and Science Students-Cambridge University Press (1990) PDF

Advanced mathematical methods for
engineering and science students

Advanced mathematical
methods for
engineering and science
students
G. STEPHENSON
Emeritus Reader in Mathematics,
Imperial College, London
and
P. M. RADMORE
Lecturer, Department of Electronic and
Electrical Engineering,
University College London
CAMBRIDGE
UNIVERSITY PRESS
Published by the Press Syndicate of the University of Cambridge
The Pitt Building, Trumpington Street, Cambridge CB2 1RP
40 West 20th Street, New York, NY 10011^1211, USA
10 Stamford Road, Oakleigh, Melbourne 3166, Australia
© Cambridge University Press 1990
First published 1990

Reprinted 1993
British Library Cataloguing in Publication Data

Stephenson, G. (Geoffrey), 1927-
Advanced mathematical methods for engineering and
science students.
1. Mathematics
I. Title II. Radmore, P. M.
510
Library of Congress Cataloguing in Publication Data

Stephenson, G. (Geoffrey), 1927-
Advanced mathematical methods for engineering and science
students / G. Stephenson and P. M. Radmore
p. cm.
ISBN 0-521-36312-8. — ISBN 0-521-36860-X (pbk)
1. Mathematics. I. Radmore, P. M. II. Title
QA39.2.S745 1990
51O-Kic2O 89-36870 CIP
ISBN 0 521 36312 8 hardback

ISBN 0 521 36860 X paperback
Transferred to digital printing 2002

Contents
Preface ix
Notes to the reader xi
1 Suffix notation and tensor algebra 1
1.1 Summation convention 1
1.2 Free and dummy indices 3
1.3 Special symbols 4
1.4 Vector identities 6
1.5 Vector operators 8
1.6 Orthogonal coordinate systems 9
1.7 General coordinate transformations 13
1.8 Contra variant vectors 15
1.9 Kronecker delta symbol 15
1.10 Scalars and covariant vectors 16
1.11 Tensors of higher rank 18
Problems 1 19
2 Special functions 22
2.1 Origins 22
2.2 The gamma-function 23
2.3 The exponential integral and related functions 28
2.4 The error function 28
2.5 The Bessel function 31
2.6 The Bessel function Yv 36
2.7 Generating function 39
2.8 Modified Bessel functions 42
2.9 Summary of the main properties of special functions 44
Problems 2 50
3 Non-linear ordinary differential equations 53
3.1 Introduction 53
3.2 Equations with separable variables 54
vi Contents
3.3 Equations reducible to linear form 57

3.4 Bernoulli's equation 59
3.5 Riccati's equation 61
3.6 Special forms of Riccati's equation 63
3.7 The Lane-Emden equation 65
3.8 The non-linear pendulum 66
3.9 Duffing's equation 68
Problems 3 73
4 Approximate solutions of ordinary differential equations 75
4.1 Power series 75
4.2 Frobenius series 78
4.3 Picard iterative method 79
4.4 Perturbation series 82
4.5 Normal form 83
4.6 The W.K.B. (Wentzel-Kramers-Brillouin) approximation 84
4.7 Eigenvalue problems 88
4.8 The Liouville-Green technique 90
Problems 4 96
5 Contour integration 98
5.1 Functions of a complex variable 98
5.2 Exponential and logarithmic functions 101
5.3 The derivative of a complex function 103
5.4 The Cauchy-Riemann equations 105
5.5 Derivatives of multi-valued functions 108
5.6 Complex integration 109
5.7 Cauchy's Theorem (First Integral Theorem) 112
5.8 Cauchy Integral Formula (Second Integral Theorem) 116
5.9 Derivatives of an analytic function 118
5.10 Taylor and Laurent series 120
5.11 Singularities and residues 125
5.12 Cauchy Residue Theorem 131
Problems 5 134
6 Applications of contour integration 137

6.1 Introduction 137
6.2 Calculation of real integrals 137
6.3 An alternative contour 144
6.4 Poles on the real axis: the principal value of integrals 146
6.5 Branch points and integrals of many-valued functions 150
6.6 Summation of series 154
Problems 6 158
7 Laplace and Fourier transforms 161

Contents vii
7.2 The Laplace transform 162

7.3 Three basic theorems 163
7.4 The calculation of an integral 166
7.5 Laplace transform of an error function 167
7.6 Transforms of the Heaviside step function and the Dirac
delta-function 169
7.7 Transforms of derivatives 171
7.8 Inversion 173
7.9 Inversions of functions with branch points 179
7.10 Solution of ordinary differential equations 181
7.11 Solution of a Volterra integral equation 182
7.12 The Fourier transform 183
Problems 7 188
Partial differential equations 190

8.2 Principle of Superposition 190
8.3 Some important equations 191
8.4 Linear second-order equations in two independent variables 193
8.5 Boundary conditions 194
8.6 Method of characteristics 195
8.7 Separation of variables 205
8.8 Integral transform techniques 214
Problems 8 218
Calculus of variations 224

9.2 Euler's equation 225
9.3 Alternative forms of Euler's equation 229
9.4 Extremal curves with discontinuous derivatives 233
9.5 Several dependent variables 233
9.6 Lagrange's equations of dynamics 234
9.7 Integrals involving derivatives higher than the first 237
9.8 Problems with constraints 238
9.9 Direct methods and eigenvalue problems 240
Problems 9 246
Answers to problems 248

Index 253
Preface
This book contains a selection of advanced topics suitable for final

year undergraduates in science and engineering, and is based on
courses of lectures given by one of us (G. S.) to various groups of third
year engineering and science students at Imperial College over the
past 15 years. It is assumed that the student has a good understanding
of basic ancillary mathematics. The emphasis in the text is principally
on the analytical understanding of the topics which is a vital
prerequisite to any subsequent numerical and computational work. In
no sense does the book pretend to be a comprehensive or highly
rigorous account, but rather attempts to provide an accessible working
knowledge of some of the current important analytical tools required
in modern physics and engineering. The text may also provide a useful
revision and reference guide for postgraduates.
Each chapter concludes with a selection of problems to be worked,
some of which have been taken from Imperial College examination
papers over the last ten years. Answers are given at the end of the
book.
We wish to thank Dr Tony Dowson and Dr Noel Baker for reading
the manuscript and making a number of helpful suggestions.
G.S.
Imperial College, London
P.M.R.
University College London
IX
Notes to the reader
1. The symbol In denotes logarithm to base e.

2. The end of a worked example is denoted by
XI
1
Suffix notation and tensor
algebra
1.1 Summation convention

We consider a rectangular cartesian coordinate system with unit
vectors i, j and k along the three coordinate axes x, y and z
respectively. For convenience, we relabel these unit vectors el5 e2 and
e3 and denote the coordinate axes by xlf x2 and x3 (Figure 1.1). A
typical vector a with components aly a2 and a3 in cartesian
coordinates can then be written as
Instead of writing the summation sign in (1.1) every time we have an

expression of this kind, we can adopt the summation convention:
whenever an index occurs precisely twice in a term, it is understood
that the index is to be summed over its full range of possible values
without the need for explicitly writing the summation sign £. Hence
(1.1), with this convention, is
a= fl/Cf, (1.2)
where summation over i is implied (i = 1, 2, 3). Since the components

at of a are given by the dot-product of a with each of the unit vectors
e,, then at•, = a . e, (i = 1, 2, 3) and (1.2) can be written
a = (a.e,)e l , (1.3)
again adopting the summation convention.
2 Suffix notation and tensor algebra
Example 1 If vectors c and d have components c, and dt (i = 1, 2, 3)

respectively, then
3
c . d = 2 ctdif = C/d/. ^ (1.4)
Example 2
asxs = alx1 + a2x2 = a^. (1.5)
Example 3 Consider the term a^bibj in which / and j both occur twice.
The summation convention implies summation over / and ;
independently. Hence, if / and j run from 1 to 2,
ttijbibj = axxbxbx + a12b1b2 + a2lb2bx + a22b2b2 (1.6)
? | a 21 ). ^ (1.7)
The usual rules apply when the summation convention is being used,
that is,
(1.8)
and
We see in (1.9) that although i occurs three times in the left-hand side
it only occurs twice in each term and therefore summation over i is
implied. Also
(atCjXbidj), (1.10)
Figure 1.1
1.2 Free and dummy indices 3
that is, the order of the factors is unimportant and summation is

implied over both i and j . It would be wrong, however, to write (1.10)
as ciibiCidi since the index occurs four times and therefore no
summation over / would be implied.
Example 4
/ 2 x 2
2 <*sxs = (a1xl + a2x2)2 an)
/
\s = l
= a\x\ #2*2- (1.12)
Using the summation convention we can write this as
( t «/*.-)(2 "ft) = Wjxj, (1-13)

where both / and j appear twice and consequently are both summed
over the values 1 and 2. ^d
1.2 Free and dummy indices

Consider the following set of n linear equations for the quantities
Xi,x2> . • . , xn with (constant) coefficients ally aX2y . . . , alny
a
&2n> • • • y nn-
021*1 + 022*2 + . . . + a2nXn = C2y
where cly c2y . . . , cn are given constants. This set of equations can be
written as
02;*; = C2y
(1.15)
7= 1
By introducing the index i, (1.15) can be expressed in the more

compact form
dtjXj = ciy (/ = 1, 2, . . . , n). (1.16)

Using the summation convention, we can write (1.16) as

a^Xj^Ci. (1.17)
Indices which occur twice, so that summation over them is implied (y
in (1.17)), are called dummy indices, while indices which can have any
value (/ in (1.17)) are called free indices. We note that (1.17) could
equally have been written as apqxq = cpy where summation over the
dummy index q is implied and p is free.
Example 5 Suppose we are given constants atj (i, j = 1, 2, 3) and the

function (p = aijxixJ (where summation over both i and y is implied).
We wish to calculate the quantities d(p/dxsy where s is a free index
(equal to 1, 2 or 3). Then by the chain rule for differentiation
-fl = aij-£LXj + aijXi-^1. (1.18)

dxs dxs dxs
Since the xt are independent variables, dXj/dxs is 1 if i=s and zero
otherwise. Hence in the first term the /-summation has only one
non-zero term (when i = s). Similarly the /-summation in the second
term has only one non-zero term (when j = s). Thus
d(pldxs = asjXj + aisxiy (1.19)
where j is summed in the first term and / in the second. These dummy
indices can be given any letter we choose so that, replacing the dummy
index y by / in the first term of (1.19),
— = asiXi + disXi = (asi + a^xt, (1.20)
where / is now the dummy index and s is the free index. If ais is
symmetric so that a^ = asi then
d(j>ldxs = 2aisxh (1.21)
whereas if a^ is skew- (or anti-) symmetric so that a^ = —asi then
0 = 0 and d(f)/dxs = 0. ^ (1.22)
1.3 Special symbols
1. Kronecker delta
The Kronecker delta symbol <5,y is defined by
1.3 Special symbols 5
We note that <5/yey- = dnei + <5/2e2 + <5/3e3 and that only one of these
three terms is non-zero depending on /; for example <5n = 1,
<5i2 - ^13 = 0. Hence <5/y-ey = e,-. Also, as in (1.3),
. (1.24)
Consider the quantity <5iy<5yA:. Then
dijdjk = (ei.ej)(ej.ek) (1.25)
= (ef-. eOfo. ek) + (e,. e2)(e2 . ek)
+ (e / .e 3 )(e 3 .e,). (1.26)
Now if i i= ky then at most one of the two brackets in each term can be
non-zero and hence each term is zero. If i = k then one term has both
brackets non-zero and equal to 1. Hence 6ij6Jk is zero if / =£ k and is 1
if / = k. This is just the definition of 6ik (see (1.23)) and so
Further consider the expression SrsApqs, where r, /? and g are free

indices. Then
SrsApq5 = SrlApgl + < 5 r 2 ^ 2 + Sr3Apq3. (1.28)

Only one of these terms is non-zero depending on the value of r. Hence
6rsApqs=Apqr. (1.29)
2. TTre alternating symbol

The alternating symbol eiik is defined as
6 ^ = e,-. (ey X efc). (1-30)

Hence if /, j and k are all different, then ey X efc = ±e,-, the plus sign
being taken if /, /, k form a cyclic permutation o f l , 2 , 3 ( l , 2 , 3 o r 3 ,
1, 2 or 2, 3, 1) and the minus sign if they form an anticyclic
permutation (3, 2, 1 or 2, 1, 3 or 1, 3, 2). Hence, from (1.30),
eijk = +1 if /, y, k are all different and cyclic, and eijk = — 1 if i, j , k are
all different and anticyclic. If j = k} then the cross-product is zero in
(1.30) and consequently so is eijk. If either j or k equals i then ey X e^ is
at right-angles to e, and eijk will again be zero. We have finally
+ 1 if /, y, k are a cyclic permutation of 1, 2, 3,

eijk = ^ — 1 if i, y, A: are an anticyclic permutation of 1,2, 3, (1.31)
0 if any two (or all) indices are equal.
Hence cyclically permuting the indices i, j , k leaves eijk unaffected,

whereas interchanging any two indices changes its sign:
€
ijk ~ ekij = €
jki> (1.32)
Cjik=-eijk- (1-33)
1.4 Vector identities

Consider the vector product c = a X b . Then from (1.2)
c = (afr) X (bjej) = afifa X ey). (1.34)
Now, in general, c = (c.e*)efc using (1.3). Hence substituting for c
from (1.34) gives
c = a X b = ( ^ ( e , X C y ) . ek)ek (1.35)
= aibj€kiJek, (1.36)
using (1.30). In (1.36), summation over the indices i, j and k is implied
so that
a X b = aibjCujei + aibJ€2ije2 + aibje3ije3 (1.37)
and the rth component of a X b is therefore
(a X b) r = eâfij, (r = 1, 2, 3). (1.38)
For the scalar triple product
(a X b ) . c = (afijet X e 7 ). ckek (1.39)
= aibJckek.(eiXeJ) (1.40)
= ekijaibjCk. (1.41)
Using (1.32), we have finally
(aXb).c = ^ Q - (1-42)
The scalar triple product does not depend on the order of the dot
and cross operations since a . (b X c) = (b X c ) . a = €pqrbpcqar =
erpqarbpcq = (a X b ) . c.
We can use a result from vector algebra to derive an important
identity involving the Kronecker delta and the alternating symbol. We
have the standard result
(a X b ) . (c X d) = (a . c)(b . d) - (a . d)(b . c). (1.43)
Putting a = e,-, b = ey, c = ek and d = eh then
(e, X e y ). (e* X ez) = (ef-. e*)(ey • e7) - (ef-. e/)(ey. ek). (1.44)
1.4 Vector identities 7
Now since for any vector A, A = (A . e m )e m (as in (1.3)),

(e, X e,) = [(e, X e , ) . e m ]e m = emiJem = eiJmem. (1.45)
Similarly
(ekXel) = eklpep. (1.46)
Hence the left-hand side of (1.44) becomes
(e, X e y ). (e* X e7) = €iJmeklpem • ep (1.47)
= Cijm€kiPdmP (1-48)
= eiimeklm, (1.49)
using (1.23) and (1.29). Substituting this into (1.44) and expressing all
the dot-products on the right-hand side using (1.23), we have
îjm^kim = dikfyi - 8a&jk- (1.50)
Example 6 A matrix 0 has elements

0/* = rtifl* + euknh (1-51)
where «, are the components of a unit vector. Show that the elements
of the matrix <p>2 are given by
(0 2 ) iy = 2if f ./i y -a o , (1.52)
Now
(02)*y = 0/*0*y = («/«* + £ukni)(nknj + €kmjnm) (1.53)
(1.54)
Since e ^ is an antisymmetric symbol under interchange of any two
indices, quantities such as eilknink are zero because for any pair of
values / and A:, two terms (with opposite signs) result from the
summations over / and k (for example, €il2nln2 cancels with
£/2i#2rti = —^12^1^2)- Hence, since nknk = 1,
( 0 % = /1,-ziy 4- eilk€km/filnm (1.55)
= /1,-rty + eilkemjkriinm. (1.56)
Using (1.50), (1.56) becomes

( 0 % = ntnj + (<5/m(5/y - S^Jn^ (1.57)
= /1,-rty + rijTii - bijtiî = irtiYij - dijy (1.58)
as required. ^
Suffix notation and tensor algebra
1.5 Vector operators

Defining the operator V, by
d
vf- = dXi (1.59)
and the vector operator V by

(1.60)
the gradient of a scalar function (f> = <j)(xly x2, x3) is
3d)
grad d) = V<b = e.V^ = ex h e2 1- e 3 — . (1-61)
dxx dx2 dx3
We note that the quantities dxjdxs in Example 5 above can be
written
axI./3xJ = VJ*l- = S£l. (1.62)
The divergence of a vector function 2i(xXy x2, x3) is
•, ~ 5«! da 5^3
div a = V . a = V,a, = — + — 2 4- — - . (1.63)
a* 9JC 5JC
The curl of a vector function a(xlt x2, x3) can be expressed, using
(1.38), as
da^
(V X a),- = eijkVjak = eijk — , (1.64)
giving
curl a = V X a = e ^ ^ V y ^ . (1.65)
The curl of a vector (in cartesian coordinates) can easily be written
down in terms of determinants as follows: if in (1.41) a . ( b X c ) is
written out in full using the definition (1.31) of eijk, we find
a.(bXc) = e ^ , V * ( L66 )
= a1b2c3 — alb3c2+ a2b3cl — a2bxc3 + a3blc2— a3b2clf (1.67)
which can be written as the determinant
ax a2 a3
a . (b X c) = b\ b2 b3 (1.68)
cx c2 c3
Hence
VXa= eiîVjUk (1.69)
i e2 e3
x V2 V3 (1.70;
ax a2 a3
Various identities involving the vector operator can be derived using

the above results (and the summation convention). We illustrate this
with three examples.
Example 7
(1.71)
(1.72)
^ (1.73)
Example 8
(1.74)
(1.75)
= 0 c u r l b + e / ^(V</>) 7 ^ (1.76)
where in the last term we have used (1.38) with a = V0. ^
Example 9
curl curl A = e^V/curl A)* (1.78)
= ei<EijkVj(eklmVlAm) (1.79)
= ei€ijk€klmVlVfAm (1.80)
(1.81)
(1.82)
= eiVi(VmAm)-VjVJeiAi (1.83)
2
= grad(div A) - V A, (1.84)
where
v =v Vy =
'" a ^ + aî + acl' ^
1.6 Orthogonal coordinate systems

So far we have considered only cartesian coordinates xr, x2 and x3. We
will require, in later chapters, expressions for the operator divgrad
(=V2) in coordinate systems based on cylindrical and spherical polar
coordinates.
Consider two points with cartesian coordinates (xlf x2, x3) and
(*! 4- dxly x2 4- dx2, x3 4- dx3), where dxt are small displacements. The
infinitesimal distance ds between these two points is given by

ds2 = dx\ + dx\ + dx\ = dxt dxt (1.85)
(using the summation convention). We now transform to a new
coordinate system, say qlt q2 and q3, the xt being functions of the qt.
If ds2 can be written in the form
ds2 = h\ dq2 + hi dqi + h2 dq\ = h2 dq2f (1.86)

then the new coordinates form an orthogonal coordinate system. For
cartesian coordinates hl — h2 = h3 = l and qt = xt (i = 1, 2, 3). We now
give, without proof, expressions for the gradient, divergence and curl
in the new coordinate system in terms of the quantities hh
If <£ is a scalar and A = e ^ , is a vector then
e
2
(1.87)
h3 dq3
div A
= TVT \T~ 4dq 2
n1n2n3ld
(1.88)
hxex h2e2
1 d d d
curl A =
hxh2h3 dqx dq2 dq3
h3A3
It is important to realise that in the above expressions the vectors e,
are unit vectors which are directed along the three new coordinate
axes and point in the direction of increasing coordinate values. Further
V2O^ i I" a (h2h3d<t>\ | d /hxh3d<s>\ ( d /M 2 a<s>\-|

hxh2h3 Idqi \ hi dqj dq2\ h2 dq2) dq3\ h3 dq3)Y
(1.90)
We now specialise these results to two particular coordinate systems
which are of importance in later chapters.
1. Cylindrical polar coordinates

We specify z, the distance of the point from the xlf x2 plane, and the
polar coordinates p and 0 of the projection of the point in this plane.
Thus qx — p, q2 — $, q3 = z (see Figure 1.2). The vectors e, point in the
directions of increasing p, 0 and z. The relationships between
cartesian and cylindrical polar coordinates are

Xi = p cos (j>, x2 = p sin (f>, x3 = z. (1.91)
Hence
ax\ ax\ ax\ (1.92)
dp dip az
— cos ip dp — p sin ip dip. (1.93)
Similarly
dx2 = sin ip dp + p cos 0 dip, (1.94)
dx3 = dz. (1.95)
Hence
<£s2 = dxt dxt = (cos ip dp — p sin ip dip)2
+ (sin ipdp -\- p cos 0 d0) 2 + dz2 (1.96)
(1.97)
This is therefore an orthogonal coordinate system with
h1 = l, h2 = p, h3 = l. (1.98)
Hence, from (1.87)-(1.90),
(1.99)
dp + p dip7
3
(1.100)
ei pe 2 e3
a d d
curl A = — (1.101)
P dp dip dz
Al pA2 A3
Figure 1.2
e,, = e,, e^ = e2, e, =

and
(1.102)
p I dp \ dp) p dd>2
2. Spherical polar coordinates

The coordinates specify the distance r of the point from the origin, and
the 'latitude' 6 and longitude 0, so that qi = r, q2=0y q3 = (t> (see
Figure 1.3). The relationships between cartesian and spherical polar
coordinates are
= r sin 6 cos (f), x2 = r sin 6 sin (p, x3 = r cos 6. (1.103)
Hence
(1.104)
= sin 6 cos (f)dr + r cos 6 cos 0 dO — r sin 6 sin (p a (1.105)

Similarly
dx2 = sin 6 sin (pdr + r cos 6 sin $ dd + r sin 6 cos 0 < , (1.106)
dx3 = cos 6 dr — r sin 6 dd.
(1.107)
From (1.105)-(1.107)
<fc2 = dx, dxt = dr2 + r 2 dd2 + r 2 sin2 0 d(p2.
(1.108)
This is therefore an orthogonal coordinate system with
h1 = l, h2 = r, h3 = r sin 6.
(1.109)
Figure 1.3
1.7 General coordinate transformations 13
Hence, from (1.87)-(1.90),
(1.110)
r 96 r sin 6 9(f)
l) + r—(smeA2) + , (LIU)
r sin
5i re 2 r sin 6 e 3
1
curl A = (1.112)
r2 sin 6
Ax rA2
and
V2<$> =2 -r-—[sintf — (r2 — ) + — (sin0 — ) + ^ — 2^ 1 V (1.113)

}
r sineL 3r\ 9r I 98\ 96/ sin 6 96 \
1.7 General coordinate transformations

The tensor notation and the summation convention developed and
used in the preceding sections is useful especially in the general theory
of tensors. This is a large topic and here we only give a brief
introduction.
Tensor theory is concerned with the way particular mathematical
quantities transform under general coordinate transformations.
Suppose we have an n-dimensional coordinate system (the jc-system),
a point in this space having coordinates xXy x2, . . . , xn. Now suppose
that we carry out a transformation of these coordinates to a new
coordinate system (the i-system) where xx is related to the x-
coordinates by
i, X2, - • • , Xn). (1.114)
Similarly,
x2=f2(xlyx2,
(1.115)
where f\,f2, . . . ,fn are functions specifying how the coordinates xt are
related to the coordinates xt. Using a free index /, (1.114) and (1.115)
may be written
Xi =fi(x\> x2, . . . , xn)y (i = 1, 2, . . . , n)y (1.116)
or more compactly,
*i=fi(xs), (*,i = l , 2 , . . . , / i ) . (1.117)
For example, the transformations (1.91) from cylindrical polar coordi-

nates (p, 0, z) to cartesian coordinates {xly x2, x3) are of this form, as
are those in (1.103) from spherical polar coordinates (r, 6, 0) to
cartesian coordinates {xx, x2, x3).
Now assuming that the functions /• are differentiable (up to whatever
order we require), we can differentiate (1.117) to obtain
dxi = -^dxl + -^dx2 + ...+-^dxn (1.118)

ox i ox2 oxn
= fk> (1.119)
or
dx (1.120)
where, using the summation convention, summation is over the index;

(from 1 to n). Using (1.120), we have n equations for the dxt in terms
of the dxt. In order that the dxt may be uniquely expressed in terms of
the dxif we must require that the Jacobian determinant / (with
element dXi/dxj in the ith row and/th column) satisfies
J = \dXi/dxj\±0. (1.121)
When this condition is satisfied, the coordinate transformation (1.117)

is called an allowable transformation. It then follows that we can solve
(1.120) to give
dx^^dxj, (I = 1 , 2 , . . . , / I ) . (1.122)
Geometrically, we can easily see the significance of these transfor-

mations. Suppose in the jc-coordinate system P is a point with
coordinates xt and Q is a neighbouring point with coordinates xt + dxt.
Then the quantities dxt are the components of an infinitesimal vector
joining P to Q. When the coordinate system is transformed to the
x-coordinates, P has coordinates xt and Q has coordinates xt + dxt.
Then (1.120) represents how the vector dxt from P to Q in the
x-system is expressed in terms of the vector dxt in the x-system.
1.9 Kronecker delta symbol 15
1.8 Contravariant vectors

We now use the basic form of the transformation law (1.120) to define
a contravariant vector. It is convention that the indices on the xt
and xt in (1.120) are raised into superscript position so that we write
(1.120) as
Definition A set of functions A1 of the coordinates is said to form a

contravariant vector if under coordinate transformation from xl to x\
A1 transforms to A1 where
With this definition we see that the dxl of (1.123) form a contravariant
vector.
Quantities with indices in the superscript position are contravariant
objects and a quantity with one free index (A1) is called a tensor of
rank one, the rank being equal to the number of free indices.
Accordingly, if (1.124) holds, A1 is a contravariant tensor of rank one,
a rank one object of this type being a vector.
1.9 Kronecker delta symbol

We now define the Kronecker delta symbol (see (1.23)) by putting one
index into superscript position so that
Then, using the summation convention,

b)A=A\ (1.126)
<5)(5£=<5L (1.127)
b\ = n (the dimension of the space). (1.128)
Using the Kronecker delta, we can now invert the transformation
(1.124) for a contravariant vector as follows: multiplying (1.124) on
both sides by dxk/dxl (and summing over i), we have
**-££* (1.I2W
dx dx dxJ
dxk
= -—AJ = 6fAJ = Ak. (1.130)
Hence
v-g* am,
is the inverse transformation of (1.124) (provided that (1.121) holds).
1.10 Scalars and co variant vectors

Any quantity (p(xl) (a function of the coordinates xl) which has the
same value under the transformation of coordinates from xl to xl, in
the sense that
4>(xi) = <j>(xi), (1.132)
is called a scalar quantity. We say that a scalar is invariant under
coordinate transformations.
Now consider
Ai = d<f>ldxi (1.133)
and
Ai = d<j>ldxi. (1.134)
Then by the chain rule of differentiation,
aââF'aâf7' (L135)
using (1.132) (summation over / is implied). Hence the objects At

defined by (1.133) transform as
- dxJ
A
< = Âi> (* = 1,2,. . . , * ) . (1.136)
Such an object is called a co variant vector. We note the difference

between (1.124) and (1.136). However, if we transform from one
rectangular cartesian coordinate system to another, linearly related to
the first, it can be shown that there is no distinction between covariant
and contravariant tensors. Such tensors are called cartesian tensors.
The inverse of (1.136) may be easily obtained, since
dx1 - dx1
(1.137)
dxk ' ~ dxk dx
= d{Aj = Ak, (1.138)
whence
Ak=^-kAi. (1.139)
1.10 Scalars and covariant vectors 17
Suppose A1 is a contravariant vector (or contravariant tensor of rank

one) and Bt is a covariant vector (or covariant tensor of rank one).
Then, using (1.124) and (1.136),
where summation over both s and p is implied. Hence

ABi = SPASBP = ApBp = AlBh (1.141)
since p is a dummy index and may be written as /. The quantity AlBt is
therefore equal to AlBt under coordinate transformation and hence is a
scalar called the scalar product of the vectors A1 and Bt. Sometimes
this scalar is called the inner product of A1 and Bh whereas the object
AlBj is not a scalar but a tensor of rank two and is called the outer
product of A1 and Bj. In elementary discussions, a vector is defined as
an object which has both magnitude and direction. Strictly speaking, a
vector (or tensor) is defined in terms of its transformation properties
as in (1.124) or (1.136).
Example 10 Show that the vector A1 with components

Al=x+y, A2 = y-xy A3 = 0 (1-142)
has the same form under rotation about the z-axis.
Under rotation about the z-axis, the coordinates x, y and z
transform to
x = x cos 6 4- y sin 6, (1.143)
y = -x sin 6 4- y cos 0, (1.144)
z=z. (1.145)
With
X1 = Xy x2 = yy x3 = z (1.146)
and
x1 = x, x yf x J z, (1.147)
transformation law (1.124) to obtain
= dXAl + 3XA2
+ A3 (1.148)
dx dy ¥Z
= (x+y)cos6 + (y —x) sin 6 (1.149)
= (x cos 6 + y sin 6) 4- (— x sin6 + y cos 6) (1.150)
= x +y. (1.151)
Similarly
(1.152)
Equations (1.151) and (1.152) are of the same form as (1.142), as
required. ^
1.11 Tensors of higher rank

Tensors of higher rank (meaning more than one free index) may be
defined by a simple extension of the transformations (1.124) and
(1.136), so that every superscript index transforms as (1.124) and every
subscript as (1.136). For example, the objects with the following
transformation laws are all tensors of rank two, but of different types:
Tij = —-k —j Tki, (contravariant, rank 2) (1.153)

dx ox
T) = - ^ - ^ Tkh (mixed, rank 2) (1.154)
T^ = —— —— Tkh (covariant, rank 2) (1.155)
where the mixed tensor has one contravariant and one covariant
index. Tensors of higher rank may be defined in an obvious way.
The algebra of tensors is straightforward since we may readily define
the sum and difference of two tensors using the basic transformation
laws. For example,
A1 ± Bl = —. (AJ ± Bj)y (1.156)
so that if A1 and Bl are contravariant vectors their sum and difference

are also contravariant vectors. Similarly, it can be seen from the
appropriate transformation law that the sum and difference of two
tensors of the same type are tensors of that type.
An important property of all tensor equations is that if the
components of a tensor are zero in one coordinate system, then they
remain zero in any other coordinate system (as can be seen from the
transformation laws).
The calculus of tensors, which we omit here, involves another
quantity called the affine connection which does not possess tensorial
properties: it may have zero components in one coordinate system but
non-zero components in another.
Problems 1 19
Finally, we consider the infinitesimal distance ds between two

neighbouring points. In cartesian coordinates, ds2 is given by (1.85). In
a general coordinate system, we may write
ds2 = gijdxidxi, (1.157)
where gtj is a covariant rank two tensor. Expression (1.157) is known
as the metric and the elements gtj (i, j = 1, 2, . . . , n) are the metric
coefficients. Consider a coordinate transformation from xl to xl. Then
using (1.123) and (1.155),
ds2 = & dxldxJ (1.158)
= bkpblqgkldxpdx« (1.161)
= gkldxkdxl = ds2. (1.162)
Hence, the quantity ds is a scalar since it is invariant under a general
coordinate transformation.
Tensor algebra and its associated calculus are important tools in the
study of continuum mechanics and in the general theory of relativity.
Problems 1
1. Write out aikXiXk in expanded form, assuming aik — akiy and
i, k = 1, 2, 3.
2. Over which indices (if any) in the following expressions is
summation implied?
(i) dijbj, (ii) dijbjj, (iii) atibny (iv) au = bu.
3. Find the values of 6iy6iy, <5,y(5yVAm<5/m, ejklAkAh and dikeikmy all
indices ranging from 1 to 3.
4. Evaluate €iklejki and €ijkeijky all indices ranging from 1 to 3.
5. The object 7^; is related to 5iy by the relation
7;,. = (adijdu + PdxdjdSu,
where a and /? are constants. Find 5^ in terms of Tijy all indices
ranging from 1 to 3. (Hint: first evaluate Tu.)
6. The square matrices A and B have elements aik and bik
respectively. Use the suffix notation to show that (AB)T = BTAT,
where T denotes the transpose operation. If A is a 3 x 3 matrix

show that eijkaipajqakr = epqr \A\> where \A\ denotes the deter-
minant of A.
7. Show that the ith component of the vector a X (b X a) may be
written as c^bj, where bj are the components of b. Determine the
form of ctj in terms of the components of a and show that it is a
symmetric object.
8. Use the suffix notation to show that
(i) div(A X B) = B . curl A - A . curl B,
(ii) curl(A X B) = A div B - B div A + (B . V)A - (A . V)B,
where A and B are vectors.
9. If fk = xixjeijk + XiXiXk> show that
dfk/dxs = 2xsxk + XiXtdte.
Find also d2fk/dxr dxs, and deduce that d2fkldx2s = 2xk + 4xs8ksy no
summation over s being implied. Verify the last result directly for
the cases k = 1, s = 1, and k = 1, s = 2.
10. If Ak =xnxnxmxmxk, show that
dAk/dxp = ocxpxkxmxm + xnxnxmxmdkp,
and determine the constant a. Show also that
= fixnxnxmxm
and determine /3 for the cases (i) when all indices range from 1 to
3, and (ii) when all indices range from 1 to 7.
11. Using the form
d ( d(b
div(/grad0) = — [f-f
aX \ o
and assuming that both / and 0 are functions of u =xnxn only,

show that
12. Verify that, if 4>{xt) satisfies the equation V20 + K2<j) = 0, where K
is a constant, then the second-order tensor
Tik = (VfV* - SikV2)<t>,
where V, = d/dxif is a solution of

eiJkeamVjVmTkp-K2Tlp = 0.
Problems 1 21
13. The symmetric tensor gik is related to the three-index object Yijk
which is symmetric in its last two indices (that is, Tijk = TikJ) by the
relation
_ a T — Q V = 0
> ft/5x skm Bsk1- sim y}
'
Show that
= l/dgim dgkm dgik\
2 \ dxk dxt dxj
2
Special functions
2.1 Origins
This chapter is principally concerned with some of the special
functions which occur in the solution of differential equations both
ordinary and partial. These functions are more complicated than those
met in elementary mathematical methods (for example, the sine,
cosine, exponential and logarithmic functions) and often arise in the
solutions of linear second-order differential equations of the form
d2y dy
For example, the equation

9 d2y dy , 9 ~ ,
x —- + x — + K(x— v )y^v = 0 K(2 2^
dx2 dx ' ]
where v is a constant, is called Bessel's equation, the solutions of

which are the Bessel functions. We discuss the solution of this
equation in detail in Section 2.5 and derive various properties of the
Bessel functions in order to illustrate the general methods of solution.
These methods may be employed to derive the properties of other
special functions and, rather than repeating the analysis, we will
simply list some of their important properties for reference purposes
and for use in later chapters. For an extensive account of special
functions and their tabulated values, the reader should consult a
standard reference workf. By building up an extensive list of such
+
M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions
(Dover, New York, 1964).
22
special functions and their properties, we may regard a problem or

equation as being solved if its solution can be expressed in terms of
these functions.
Besides arising as solutions of differential equations, other special
functions are defined in terms of integrals. These occur in some of the
analysis which follows in later chapters. The next three sections deal
briefly with these functions.
2.2 The gamma-function

We define the gamma-function T(x) by
=fJo
(2.3)
where, in order for the integral to converge, x>0. Now consider

T(x + 1). Integrating by parts, we have
= f txe-tdt = [-txe-']Z + x \ 1
dt. (2.4)
Jo Jo
In (2.4), the integrated term is zero because at the upper limit the
exponential dominates the tx term, while at the lower limit tx is zero
since x > 0 . Hence, using (2.3),
r(x + l ) = * f tx-lQ-tdt = xT(x). (2.5)

Jo
Equation (2.5) is a recurrence relation which connects T(x 4-1) with
T(x). Now suppose that x = n> where n ^ 1 is an integer. Then
T(n + 1) = nT(n) = n(n - l)T(n - 1) = . . . = w! T(l). (2.6)
But
r(l)= f e-'* = l, (2.7)

Jo
so that
r(/i + l) = /i! (2.8)
Sometimes T(x 4-1) is denoted by x\ and referred to as the factorial
function. Hence we may define 0! = F(l) = 1. Another useful value to
calculate is
(2.9)
Jo Jo
24 Special functions
Here we have transformed the variable of integration to u, where

t = u2, and used the standard result
(2.10)
Hence, from equations (2.5) and (2.9),

(2.11)
(2.12)
and so on for other half-integer arguments.
The recurrence relation (2.5) may be used to define the gamma-
function for negative values of its argument. Taking
(2.13)
we have
(2.14)
(2.15)
Figure 2.1
T(x)
4-
-l
-4
From equation (2.13), we see that T(x) becomes infinite at x = 0 and

hence also at x = — 1, - 2 , —3, . . . as shown in Figure 2.1. Using the
gamma-function, it is possible to evaluate various integrals, as shown
by the following two examples.
Example 1 Consider
/,= f [\n(l/x)]Ux. (2.16)
Putting JC = e~', we find
h= I f*e-'<& = r(i) = 3Vtf, (2.17)

h
using (2.12). ^ i
Example 2 Consider
I2= f jcie-*3<k. (2.18)

Jo
Putting JC3 = y, we find
h = \ \ y-^-ydy = 13ra) = Wjz, (2.19)
using (2.9). ^
Other properties of the gamma-function are also of interest.
Differentiating (2.3) with respect to x we have
Now
Hence, from (2.20),
T'{x)= \ tx-\\nt)Q-1 dt. (2.22)

Jo
Evaluating this at JC = 1 gives
r(l)= f e-'\ntdt=-Y, (2.23)

Jo
where y is called Euler's constant and has the value (to four decimal
places) 0-5772. This value can be obtained either by numerical
integration in (2.22) or from the alternative definition
(2.24)
A property of the gamma-function, which we give without proof, is

T(jc)r(l -x) = Jt/sin(7Tx) (2.25)
which is known as the reflection formula.
Related to the gamma-function is the integral
f1
B(m,/i)= xm-\l-x)n-ldx, (2.26)
Jo
with m > 0, n > 0. This is called the beta-function. It is straightforward
to show (see reference on page 22) that
An alternative form of B(ra, n) can be found from equation (2.26) by

letting x = sin2 0. Then, since dx = 2 sin 0 cos 0 dQ, we have
rJt/2
J 1
0
sin2"1"1 0 cos2""1 0^0. (2.28)
Example 3 Consider
f-Jt/2
Then
J o
V(tan 6) dd. (2.29)
rJt/2
/ = f sin^ 0 cos'i 0 dd = iB(|, i). (2.30)
Jo
Using (2.27), and T(l) = 1, this can be written
where we have used (2.25) with x = \. ^
The incomplete gamma-function T(x, a) is defined by
T(x,a)= \ tx-lz-<dty (2.32)

where a ^ 0, x > 0. This can be written as
T(x,a) = f tx-lz-ldt- \ f-^-'dt (2.33)
= T(jc)- f tx-lQ-ldt. (2.34)

Jo
Finally we obtain an approximation to T(x + 1) for large x known as
Stirling's formula. Consider, from (2.3),
= f feT'dt. (2.35)
Jo
We change the variable of integration from t to x, where t = x 4- rV*-
Then
Y{x 4-1) = \ (x 4- x^x)xc-ix+ry/xWx dx. (2.36)

J—V*
This can be written
ix (2.37)
\ p [ ( ^ ) (2.38)
—\/x L \ vx/ J
Writing (2.38) as the sum of two integrals and replacing the logarithm
term by its MacLaurin expansion in — V* <r< V*, we have
+ f exp[-Wjc+Jclnf 1+77-)] dr. (2.39)
For large x, the second integral has a vanishingly small range of

integration, whilst the first integral gives
(2.40)
where we have again used the standard result (2.10). When x — n, a

positive integer, T(n + 1) = n\ so that, from (2.40), an approximation
to n! for large n is
)e- n rt"^. (2.41)
This result is surprisingly accurate even for modest values of n. For

example, if n = 3, n\ — 6 compared with an approximate value of 5-836
from (2.41). Similarly, when n = 10, n\ = 3 628 800 compared with the
approximate value from (2.41) of 3 598 695-618, which is within 1% of
the exact value.
2.3 The exponential integral and related functions

The three integrals Ex(x)y Ci(x), and Si(x) are defined for x > 0 by
E1(x)=\ ^-dt, (2.42)

jx t
Ci(x) = - \ ^ d t , (2.43)
Si(x)=f — dt. (2.44)

Jo t
The graphs of these functions are shown in Figures 2.2(a) and 2.2(b).
These integrals arise in the Laplace inversion of logarithmic functions,
in particular
( L ^) (2.45)
(2.46)
where ££ denotes the Laplace transform (see Chapter 7). It is

important to realise that the definitions (2.42)-(2.44) vary slightly
between texts. Ci(x) is sometimes defined without the minus sign.
There also exists a function Ei(x) which must not be confused with
Ex(x), and the function si(x) defined by
fS^ (2.47)
2.4 The error function

The error function, denoted by erf x, is denned by the integral
2
2 r
=^- e"M du. (2.48)
\JT Jo
2.4 The error function 29
Again using the standard result (2.10), we see that erf jt-» 1 as x-^^.
The graph of the error function is shown in Figure 2.3.
The complement of the error function, denoted erfex, is defined by
erfc x = 1 — erf x = —7— d« (2.49)
(see also Figure 2.3). Both of these functions arise from the solution of
partial differential equations and are connected with the Laplace
transforms of particular functions (see Chapters 7 and 8).
It is relatively easy for these functions to generate what is called an
Figure 2.2
+-X
(a)
asymptotic series. We first give a definition of such a series. The series
where a0, ax, a2, . . . are constants, is said to be an asymptotic series of

the function/(x), written /(x) ~ ^(x), provided that, for any ny the
error involved in truncating the series at the term ajxn tends to zero
faster than ajxn as x—>°°. In other words we require
limx"[/(x)-S(x)] = 0. (2.51)
Asymptotic series may be added, multiplied and integrated to obtain

asymptotic series for the sum, product and integral of the correspond-
ing functions. It is instructive to point out the difference between a
convergent series and an asymptotic series: a convergent series tends
to the corresponding function/(x) as nôo for a given x, whereas an
asymptotic series tends to f(x) as JC->» for a given n (that is,
truncating the series at the term an/xn).
Consider now equation (2.49) for the complement of the error
function and rewrite the integral
•2)du. (2.52)
Integrating the right-hand side of (2.52) by parts we have
(2.53)
—re " du. (2.54)

2x L 2u2
Figure 2.3
-+-X
The last integral may be integrated by parts as follows:
\du (2.55)
-^-"2du (2.56)
=
l?" | /3 e "" 2 r f M " (2 57)
'
Continuing in this way, we eventually find
1.3 1.3.5
(2JC2)2 (2X2)3 ' ' '
+ ( - 1 ) " ——:—' " ''} H

- + Rn+i , (2.58)
where
2
[wrû2du- (2 59)
-
Equation (2.58) is the asymptotic series for ^Vêrfcx since the ratio
of consecutive terms is (2n — 1)/2JC2 which tends to zero as x^> °° for a
given n. However, for a given x, the series diverges as /i—>o°. By
choosing a suitable n, the resulting finite series is a good approxima-
tion to the integral in (2.58). Accordingly, using (2.49) and (2.58), it is
seen that the behaviour of erfcx for large x and to first order in 1/x is
(2.60)
2.5 The Bessel function

The Bessel equation of order v (where v ^ O is a constant) is (see
(2.2))
* 2 ^ + * £ + (*2-v2)y = 0. (2.61)
We attempt a series solution of this equation by using the Frobenius

series
oc
y =xm(a0 + a1x + a2x2 + . . . ) = 2 arxm+r, (2.62)
where, without loss of generality, we shall assume a o ^ 0 . (If aQ were

zero, we could redefine m to preserve the form of the series (2.62).)
The conditions under which a series solution of this kind is valid are
discussed in Chapter 4. In (2.62), m and the ar are constants which
must be found so that (2.62) satisfies (2.61). By this method, we will
find two possible values of m, giving two solutions of (2.61). The
solutions may not be independent if the values of m are identical or
differ by an integer (see Chapter 4) but the method will always
generate one solution. The method for finding a second solution in this
case will be discussed later. The Frobenius series is more general than
a Taylor or MacLaurin series since m may be found to have
non-integer values and the resulting Frobenius series will then contain
non-integer powers of x. This contrasts with the MacLaurin series
which only contains integer powers of x.
Differentiating (2.62) we obtain
£ = 2 *(,* + * - • ' - (2.63)

dx r=0
and
7 ^ = 2 ar{m + r)(m + r - l)xm+r~2. (2.64)
ax r=0
Inserting these into (2.61) we have
x2 2 ar(m + r)(m + r - l)xm+r~2 + x 2 ar(m + r ) x m + r ' 1

r=0 r=0
oo
+ (x2 - v2) 2 arxm+r = 0. (2.65)
Collecting together like powers of x gives

oo
2 [ar(m + r)(m + r - 1) + ar(m + r) - v2ar]xm+r

oo
+ 2 «r* m + r + 2 = 0. (2.66)
This simplifies to
00 OO
2 ar[(m + rf - v2]xm+r + X arxm+r+2 = 0. (2.67)
Expanding the first summation by writing out the first two terms
explicitly we have
ao(m2 - v2)xm + ai[(m + I) 2 - v2]xm+1
oo oo
2 m+r
+ 2 ar[(m + rf - v ]x + 2 arxm+r+2 = 0. (2.68)
r=2 r=0
The first summation contains powers of x from x2+m upwards, so that

putting r — s + 2 in this sum gives
2 ar[{m + r) 2 - v2}xm+r = £ a5+2[(m + j + 2)2 - v

2
]xm+^2 (2.69)
= 2 <*r+2[(m + r + 2)2 - v2]jcm+r+2, (2.70)

r=0
where the dummy index has been changed from s to r. Hence (2.68)
can be written with the two summations combined as
aQ(m2 - v2)xm + ax[{m + I) 2 - v2]xm+l
00
+ 2 {flr+2[(m + r + 2)2 - v2] + a r }x m+ '' +2 = 0. (2.71)
Now, since flo^0^ w e must have m2 — v2 = 0 in order that the sole

term in xm vanishes. Hence
m = ±v. (2.72)
Likewise, the second term is the only one in xm+1 so that
a\(m + I) 2 - v2] = fll(l ± 2v) = 0. (2.73)
Hence ax = 0 unless v — \y but as mentioned earlier, this case will have
values of m which differ by an integer and so may need to be
considered separately. Further, from (2.71), the coefficient of the
general term in jc m+r+2 must vanish giving
ar+2 = - % 22 5 2
2) -v
for r = 0, 1, 2,
We now consider the two possible values of the constant m
separately.
Case m = v In this case (2.74) becomes
(v + r + 2)2 - v2 (r + 2)(2v + r + 2)'

so that
and so on for all values of r. The final result for y(x) is, from (2.62),
~2 _4
•2) 2 . 4(2v + 2)(2v + 4)

x6
•+ ...]. (2.77)
2 . 4 . 6(2v + 2)(2v + 4)(2v + 6)
To define the Bessel function of order v uniquely it is conventional
to choose a specific value of a0. By taking
ao = l / 2 T ( v + l) (2.78)
and using the recurrence relation (2.5) for the gamma-function,
r(v + 1) = vF(v), we may write the series (2.77) as
where Jv(x) is the Bessel function of order v and 0! is unity by (2.8).

When v = n, where n is a positive integer or zero, then T(n + r + 1) =
(rc 4- r)\ from (2.8), and (2.79) becomes
In particular
x4 x6
x x3 x5 x1
from which we see that

dfo(x)/dx = -Jl(x). (2.83)
From the series (2.80) it is easy to see that all the Jn(x) satisfy Jn(0) = 0
for n a positive integer, whereas if n = 0, /0(0) = 1. The graphs of the
first two, J0(x) and /I(JC), are shown in Figure 2.4.
We now have to consider the second possible value of m obtained
in (2.72).
Case m = —v In this case we obtain the solution by simply changing

the sign of v in (2.77) and (2.79). If v = 0, 1, 2, . . . then changing the
sign of v will result in diverging coefficients in (2.77). For example,
(2.77) with v replaced by — v contains coefficients l/(—2v + 4)
which diverge if v = 2. Further, the definition of a0 in (2.78)

would be 2V/F(—v +1) which is zero if v is an integer. Hence
the cases v = 0, 1, 2, . . must be treated separately. If then v
0, 1, 2, . . . we have
- v + 2r
(-1)'
(2.84)
The two solutions (2.79) and (2.84) are independent since (2.79)
contains only positive powers of x whereas (2.84) contains both
positive and negative powers of x. The first is zero at x = 0 (except for
the case v = 0) whilst the second is infinite at x = 0.
If v = n, an integer, then (2.73) still holds and al = 0. From (2.71),
the recurrence relation is
- ( r + 2)] = <ir. (2.85)
Writing out these in full gives
a2.2(2n-2) =
a4.4(2n-4) = (2.86)
a6. 6(2n-6) =
and similar equations for #8, a10, . . . . Only the even values of r need
to be considered since (2.85) with ax = 0 implies that all ar with r odd
are zero. From (2.86) we see that if, say, n = 3 then a4 = 0 which,
working up the list, implies a2 and hence a0 are zero. In general then
all the ar preceding that for r = 2n are zero. We then have from (2.85)
(2.87)
2(2n + 2)'
a a
2n+2 _ 2n
(2.88)
' 4{2n + 4) " 2 . 4{2n 4- 2){2n + 4)'
Figure 2.4
• • *
and so on. The series for y{x) is
and hence
From (2.80) we see that Jn{x) and J-n(x) are linearly dependent. We
normally choose a2n so that
J-n(x) = (-l)nJn(x). (2.92)
The general solution of the Bessel equation may be written as
y(x)=AJy(x) + BJ-v(x), (2.93)
where A and B are constants, only when v is not an integer. When
v = n w e need to obtain a second solution (Jn(x) being one solution)
using the method in the next section.
2.6 The Bessel function Fv

To find the second solution of the Bessel equation when v = n, we
write
y(x) = u{x)Jn{x) (2.94)
and substitute this into the Bessel equation (2.61). This gives
u{x2K + xJ'n + (x2 - v2)Jn) + 2x2u'J'n + x2u"Jn + xu'Jn = 0, (2.95)
where a prime denotes differentiation with respect to x. Since /„
satisfies Bessel's equation (2.61) the first bracket in (2.95) is zero.
Consequently
u" V 1
^ + 2 ^ + - = 0. (2.96)
u Jn x
Integrating with respect to x gives
In u' + In/;; + Inx = In Ay (2.97)
where for convenience we have chosen the arbitrary constant to be
In A Hence
u'=AlxJ2n (2.98)
2.6 The Bessel function Yv 37
and therefore the second solution, often denoted by Yn(x), is

dx
Yn(x) = u(x)Jn(x)=AJn(x)j (2.99)
xJl{x)
The general solution of the Bessel equation is then
y{x) = AJn{x) J —p^, + BJn{x), (2.100)
where A and B are arbitrary constants.

When n = 0, for example, we have
(2.101)
an
Inserting the series (2.81) for /o0O d taking only the first two terms,
we find for small x
Y0(x)~AJ0(x)l±(l-Jj) dx. (2.102)

Performing the integral gives
Y0(x)~AJ0(x)[lnx + P(x)L (2.103)
where P{x) is a polynomial in x. Hence the second solution is
logarithmically divergent at x = 0. In fact, all the Yn(x) are singular at
x = 0. The graphs of the first two are shown in Figure 2.5.
Alternative forms exist for the function Yn(x). Consider the
Neumann or Weber function defined by
COS(VJT)/ V (X)-/_ V (X)
Yv(x)= (2.104)
0.4-
-•*
When v=£n, this is just a linear combination of Jv(x) and J-V(x) and
so is a solution of Bessel's equation (independent of Jv(x)). When
v = n> the numerator and denominator are both zero, so the limit as
v->n must be taken using L'Hopital's rule. Hence
— n sin(v;r)/v(jc) + COS(VJT) — —
Yn{x) = lim I — — I (2.105)
v^n(^ JT COS(VJT) J
= - lim \ ^ - stc(vjv) ^ 1 (2.106)
[ ( r 4 (2107)
To see why this is a solution of Bessel's equation consider the solutions
Jv and /_ v . These satisfy
x2 ^ 4- x ^ + (x2 - v 2 )/_ v = 0. (2.109)
We differentiate each of these equations with respect to v. Then

multiplying the second by (—l)n+1, where n is an integer, and adding it
to the first, we obtain
-(-l)"/_v], (2.110)
ux ux
where
We now let v-*n. Then from (2.92), the right-hand side of (2.110)
tends to zero and we find that
wn = limwv (2.112)
v—*n
also satisfies Bessel's equation and is therefore the required second

solution. Now from (2.107)
Yn=^"n (2.H3)
so that Yn(x) defined by (2.104) is also a solution of Bessel's equation.

2.7 Generating function

It is possible to generate Bessel functions of integral order as the
coefficients of particular series. Consider the expansion of the func-
tions Qxt/2 and Q~xl2t in powers of t:
Taking the product of the two series gives
r =o 5=0
We now examine the coefficients of various powers of t. The

coefficient of t° is obtained from (2.116) by putting r = s and evaluating
the sum of all such terms. The result is
using (2.81). Likewise the coefficient of tn is Jn(x) and that of t~n is

J-n{x). Hence we may write
oo
e i*<»-i/0 = ^ >"/„(*). (2.118)

« = — 00
The function on the left-hand side of (2.118) is called the generating

function for the Bessel functions and the series on the right is called a
Laurent series (see Chapter 5) rather than a MacLaurin series, since
the summation includes positive and negative powers of t. Various
properties of the Bessel functions can be obtained from the generating
function as follows.
1. Recurrence relations
(a) Differentiating (2.118) with respect to x gives
M
'- 1 / 0 = 2 *"/;(*) (2.119)
rt = — oo
or, substituting for the generating function,
hit-lit) i tnJn(x)= i t"J'n(x). (2.120)

Hence
\ i t"+%(x)~i
n — — °°
£/-i/n(*>= i tnK(x). (2.121)
Comparing like powers of t on each side of (2.121) gives
j[Jn-l(x) - / n + 1 ( * ) ] =/;,(*), (2.122)

whence
2J'n(x) = / n -i(^)-/ w +i(ac). (2.123)
From this relation with n = 0, and using (2.92), we have

2/^t)=/_1(JC) — Ji(x) = —2/a(x) (2.124)
so that
(2.125)
as in (2.83).
(b) Differentiating (2.118) with respect to t gives
^ ( l + l/f 2 )e^ ( '~ 1/0 = £ ntn-%(x) (2.126)

/t ^ — ° °
and hence
\x 2 i"/n(^) + ^ i r- 2 / n (^)= i «r-Vn(^). (2.127)

n = — °° n = —<x> n = — oo
Comparing like powers of f on both sides gives
1(x) = nJn(x), (2.128)

whence
In
— Jn{x) = /n_i(jc) + Jn+1(x). (2.129)
(c) We have
— \xnJ (x)] =xnJ'(x) 4- nxn~lJ (x) (2 130^
using (2.123). Substituting for Jn+1(x) from (2.129) gives
[A- •'ny-^'/J — 2-* •'n — l v * v 2-^ «'n\-^/ —•'n — l v * /
+ /UC"-7 B (JC). (2.132)

Hence
£[xVn(x)]=xnJn_l(x). (2.133)
Example 4
I = [ x2Jx(x) dx = x2J2(x) + C, (2.134)
where C is a constant, using (2.133). ^Jk
2. Integral formula for Bessel functions

In equation (2.118) we put t = tld to simplify the following analysis.
Since t - 1/t = 2i sin 0, we have from this equation
eixsin6= 2 tnJn{x)= 2 eiAT0/n(jc). (2.135)
Expanding the final summation and using the relation (2.92) we find
gix sin 0 _ j / j . \ _^_ 2ry /^\ cos(20) + / (x) cos(40) + 1
+ 2\[Jl(x) sin 0 +/ 3 (x) sin(30) + . . . ] . (2.136)
Taking the real and imaginary parts of (2.136) we have
COS(JC sin 0) = J0(x) + 2 X hn(x) cos(2«0), (2.137)
sin(;c sin 0) = 2 ]£ /zn+iW sin[(2n + 1)0]. (2.138)

n=0
Multiplying (2.137) by cos(m0) and (2.138) by sin(m0), where m is an

integer, integrating with respect to 0 from 0 to n and using the
standard integrals
rn rJt
cos(m0) cos(«0) dO = s i n ( m 0 ) sin(n0) dO = \nbmny (2.139)

Jo Jo
(where bmn is t h e Kronecker delta symbol defined in (1.23)), we have
m X W e n m 1S e v e n
- f COS(JC sin 0) c o s ( m 0 ) d9 = ( ^ \ (2.140)
7i jQ 1 0 when m is odd,
W e n m iS
i fsin(x sinfl)sin(,n0) d6 = f ^ ° ' ^ (2.141)
;r Jo ^m(^) when m is odd.
Hence, for all n, we have
Jn(x) = — [COS(JC sin 6) cos(n6)

JT Jo
+ sin(jc sin 6) sin(n6)] dd (2.142)
= - I cos(nO - x sin 6) dd. (2.143)

JT Jo
This integral form is useful in evaluating Jn(x) numerically and can

also be used to derive other relations. For example, we can obtain
(2.92) from (2.143) by writing
j_n(x) = - I cos(-nd - x sin 6) dd (2.144)

JT Jo
= - f cos(nd + x sin 6) dd. (2.145)

JT Jo
Now put 6 = JT - (f> in (2.145). Then

1 f°
/_„(*)=- cos[n(jr-(f))-\-x sin cj)](-d(t>) (2.146)
JT Jn
=- COS(JC sin (p - n<t> + njr) d<j). (2.147)

JT Jo
Now cos(A + nn) = (—l)n cos A, so that
J_n(x) = -(-!)" [ cosing - x sin </)) d(f> = (-l)nJn(x). (2.148)

JF ô
2.8 Modified Bessel functions

The general solution of the equation
x2 —^ + x j - 4- {k2x2 - v2)y = 0 (2.149)
can be found from the standard form (2.61) of the Bessel equation by
writing t = kx. Then (2.149) becomes
t2 - | + t-j- + (t2 - v2)y = 0 (2.150)

2 v )y v }
dt dt
which has the general solution
y=AJv(t) + !?Yv(0 = AJv(fct) + BYv(kx). (2.151)
2.8 Modified Bessel functions 43
An important case arises in particular physical problems when

k2= - 1 . We then have the modified Bessel equation
x2 —— 4- x — — (x2 4- v2}v = 0 (2 152^
with the general solution (since k — ±i)

y = A/V(i;c) + £ Yv(i;t). (2.153)
We now define a new function
/v(jt) = r v / v (ijc) (2.154)
and use the series (2.79) for Jv. Then
< 215S >
<2156>
which is a real function of x. Similar considerations apply to Kv{x)
which is the second solution of the modified Bessel equation (2.152).
Iv(x) and Kv(x) are called modified Bessel functions and their
properties can be obtained in a similar way to those of Jv(x) and Yv(x).
The main properties of these functions are given in the next section.
We note finally that the differential equation
d2yldx2 = (1 - l/4x2)y (2.157)
can be solved in terms of the modified Bessel functions. Putting
y =xû, we obtain the equation
x2—^ + x-^-x2u = 0 (2.158)

dx dx
which, on comparison with (2.152), has solutions /0(JC) and K0(x).
Hence the general solution of (2.157) is
y(x) = Axko(x) + Bx±K0(x). (2.159)
Another equation which we shall meet in later chapters is
(2160)
(V>
Again putting y = xû, we find
2
x ^ + x^-(l4x2+v2)u = 0, (2.161)
which the substitution t = x/2 converts into the modified Bessel

equation in t:
'2^+'f-('2+v>=0- (2162)
The solutions, from (2.152), are therefore Iv(t) and Ky(t) so that the
solution of (2.160) is
y(x) =Ax%(x/2) + Bx*Kv(x/2). (2.163)
2.9 Summary of the main properties of special functions
1. Bessel functions
Bessel's equation of order v is
x2 —^ + x j - + (x2 - v2)y = 0. (2.164)
The independent solutions when v is not zero or an integer n are Jv(x)

and /_v(jc), where
When v = n,
J-n(x) = (-l)nJn(x), (2.166)
(2.167)
A second solution Yn(x) exists, in particular (see (2.103))

v-2
x4 , „ x6
2 2v (2.16R)
2. 4 " 2 . 4 2 . 62 v
2 z
where y is Euler's constant (see (2.23) and (2.24)).

The generating function is
e4x(/-i/«)= 2 *"/„(*). (2.169)

2.9 Summary 45
Important recurrence relations are

2J'n(x)=Jn-l(x)-Jn + l(x), (2.170)
jJn(x)=Jn^(x)+Jn+l(x), (2.171)
(note dJ0(x)/dx = -J{(x)).

The integral form is
Jn(x) = - f cos(n0 - x sin 0) d a (2.172)

71 Jo
The following are known as the Lommel integrals:
j'xJn(px)Jn(qx) dx = ^ - p [pJn(ql)J'n(pl) - qJn(pl)J'n(ql)] (2.173)
for p ^q, and

l
^ [2 ( ^1)2J (2.174)
2. Modified Bessel functions

The solutions of the equation
(2.175)
are the modified Bessel functions Iv(x) and Kv{x), where
Iv(x) = rvJv(ix), Ky(x) = - iv+1[iYv(ix) + Jv(ix)]. (2.176)
In particular
x2 x4 x6
- —+ . . . . (2.178)
The graphs of I{)(x) and K0(x) are given in Figure 2.6.
3. Legendre polynomials
Legendre's equation is
2
o d y dy
(l-x2)—^-2x-f- + 1(1 + l)y = 0. (2.179)
If / is not an integer, both solutions diverge at x = ±l. If l = n =

0, 1, 2, . . . the solutions are polynomials Pn(x) called the Legendre
polynomials, where
P (x) = (x2- l)n (2.180)
In particular
Po(x) = l, (2.181)
(2.182)
Recurrence relations are
Pn+l(x) = (2.183)
(2.184)
(l-x2)P'n(x) = -nxPn(x) + nPn.x{x). (2.185)
Figure 2.6
2.9 Summary 47
The integral form is

1 fn
Pn(x) = - [x + V(*2 -f 1) cos 6]n dO. (2.186)
n Jo
We also have the following integral:
f1 2
Pn(x)Pm(x)dx=——dnm. (2.187)
j-i z/i + i
Graphs of the first three Legendre polynomials are shown in

Figure 2.7.
Further, the associated Legendre functions are defined by
dm+n
(2.188)
- n v-/ 2 n n! dxm+n y
where i AZ. In Chapter 8 we shall meet the equation
The solutions are bounded only ifA = - « ( « 4 - l ) and have the form
y = PjT'Ccos 6). (2.190)
Figure 2.7
4. Laguerre polynomials
The Laguerre equation
x2 —^ + (1 - x) - / + ny = 0, (2.191)
dx dx
where n = 0, 1, 2, . . . , has polynomial solutions Ln{x), called the
Laguerre polynomials, given by
n! dx
In particular
2
L0(JC) = 1, L^*) = 1 - x9 L2(x) = \(x - 4x + 2). (2.193)

xt/(l-t)
-T-— =J.tnLn{x). (2.194)

l —t n = 0
A useful recurrence relation is

xL'n(x) = nLn(x) - nLîx). (2.195)
We have the following integral:
f t-xLm(x)Ln(x) dx = dmn. (2.196)

Jo
The associated Laguerre polynomials are defined by
L™(x)=—-Ln(x) (2.197)
for n ^ m and satisfy

d2y dy
dx2 dx
5. Hermite polynomials
The Hermite equation
has polynomial solutions Hn{x)> called the Hermite polynomials, when

n = 0, 1, 2, . . . , given by
//„(*) = (-l)"e* 2 -— (e"*2). (2.200)

2.9 Summary 49
In particular
H0(x) = 1, Hx(x) = 2x, H2(x) = 4x2 - 2. (2.201)
(2.202)
Important recurrence relations are

Hn+l(x) - 2xHn(x) + 2nHn_1(x) = 0, (2.203)
(2.204)
We also have the following integral:
(\-x2H(x)H (x)a (2.205)
The Weber-Hermite function

(2.206)
satisfies the differential equation
(2.207)
dx2
where
A = In + 1. (2.208)
If A # 2« + 1 in (2.207), then y is not finite as x -> ±00.
Figure 2.8
Bi(x)
1—
!fsrf\rf\ff \Ai(x)
-jW\iJ\xjf"
6. Airy functions
The Airy equation
d2y
(2209)
has solutions called the Airy functions, Ai(x) and Bi(x), where
Ai(x) = \/(xl3)[I-&c*) - mhl)l (2.210)
Bi(x) = V(x/3)[/_i(|je^ + /j(M)]. (2.211)
The integral form for Ai(x) is
A*(jt) = - f cos(^ 3 + xO A. (2.212)

JT Jo
The graphs of Ai(x) and 5i(jt) are given in Figure 2.8.
Problems 2
1. Use the gamma-function to evaluate
(i) f x*e-xdx, (ii) f ^ A , (a

Jo JO yt
2. Show that
3. Show that
» m>
4. Evaluate J Q ^ J C / ( 1 — JC 3 )^ using t h e beta-function.

5. Given
show by integration by parts that
6. Use the beta-function to evaluate
f
(i) f V(sin0)d0,
Jo
f
(ii) f
Jo
V(cot(9)d0.
Problems 2 51
7. Show using equation (2.79) that
COSJC
and that
8. Show, using (2.80) with n =0, that, for or >0,

1
Jo
Hence show that, in the limit a—»0,
f /„(*) dr = 1.
Jo
9. Transform the equation
where v is a constant, by writing z = e*. Hence obtain the general

solution of this equation.
10. Show that Bessel's equation
dx2 *dx*^X 'y ~

can be transformed into
d2u
where y—x û. Obtain the general solution when v = ±\. Use
this result to show that for x2 » v2 — \
Jv(x) — i - (A sin x + B cos x).
11. Show that the general solution of
dx2
is
y = Vx[AJ1(2\/x) + BYx{2\/x)l
12. Evaluate J J\{x^) dx, using Example 4.
13. Evaluate the following integrals using (2.180) and integration by
parts:
(i) { xP3(x)dx, (ii) f P2(x)\n(l-x)dx.
14. Show, using the standard integrals

COS U JT
d
sin u ji
d 0 < p < 2
'
and (2.212), that
3ir(§).
3
Non-linear ordinary differential
equations
3.1 Introduction
The differential equations met in Chapter 2 were linear in the sense
that they were special cases of the general nth order linear equation
where fly f2, . . . , fn, g are given functions of x. For such equations the
Principle of Superposition applies: an arbitrary linear combination of
individual solutions is also a solution.
Equations which cannot be written in the form of (3.1) are called
non-linear and, for such equations, the Principle of Superposition does
not apply. A typical first-order non-linear equation is
whereas
d2y , dy
is a second-order non-linear equation.

There is no general method of solving non-linear equations analyti-
cally and numerical procedures are frequently the only techniques
available. However, some limited types can be solved analytically by
special methods. These are discussed in subsequent sections.
53
54 Non-linear ordinary differential equations
3.2 Equations with separable variables

In general, the first-order equation
dy/dx=f(x,y)/g(x,y), (3.4)
where f(x, y) and g(x, y) are given functions, will be non-linear.
However, if f(x, y) and g(x, y) are separable so that f(x, y) =
X(x)Y(y) and g(x, y) = U(x)V(y), then
dy/dx = F(x)/G(y), (3.5)
where F(x) = X(x)/U(x) and G(y) = V(y)/Y(y). If we multiply (3.5)
by G(y) and integrate both sides with respect to x, we obtain
\
G(y) dy
"I dx + C (3.6)
where C is a constant.
Example 1
dy CO
dx
(3.7)
X
Separating the variables gives
I sec2 ydy=i — + C, (3.8)
where C is a constant. Carrying out the integrations and writing

C = \nA, where A is another constant, we obtain
tany =\nx + \nA (3.9)
so that
y = tan-1[ln(^jc)]. ^ (3.10)
A case of a first-order non-linear equation which can be reduced to

separable form is
dy/dx = K(y/x), (3.11)
where K is a given function of y/x only. Writing y = xu(x) gives
x^ + u = K(u), (3.12)
ax
or
^ =^ ^ . (3.13)
dx x
This equation is now of separable form.
3.2 Equations with separable variables 55
Example 2
d JL
i
dx x—y'
(3-14)
Writing (3.14) in the form of (3.11), we have
dy _ y/x
(3.15)
dx 1 —y/x
Putting y = xu gives
du u
(3.16)
~ dx 1-u'
or
du u2
(3.17)
dx x{\ — u)
Separating the variables leads to
f — ^ d u = I — + ln,4, (3.18)
J u J x
where A is a constant. Integrating both sides, we have
- - - I n n = ln(i4*), (3.19)
so that the relationship between x and y is, on substituting for u=y/x,
(y) ( ) (y) O. ^ (3.20)
We note that in Example 2 we cannot determine y explicitly in terms

of x. Although the separation of variables and integration can always,
in principle, be carried out for equations of the type (3.5), there is no
guarantee that y may be found explicitly in terms of x.
It may be possible, after a first integration, to reduce a second-order
non-linear equation to a first-order separable type. We now give an
example to illustrate this.
Example 3
xg-(,-0* (3.21,
given that y = 1 and dy/dx = 1 at x = 0.
Since
'(&.& +* (3.22)
dx \ dxl dx dx
then (3.21) can be written
x w
dx\ dx) dx " 'dx' "~'
or, simplifying,
d I dy\ dy d , ~
dx \ dx) dx dx
Integrating with respect to x gives
2
X^=$y + A, (3.25)
where A is a constant. Applying the boundary conditions, we find that

A = — \y so that (3.25) becomes
x-^ = $y2-$. (3.26)

dx
Separating the variables gives
^ = j j + \nB, (3.27)
where B is a constant. The left-hand side can be integrated by

expressing (y2 — I)" 1 in partial fractions. We have
(3 28)
iiîJTi) ' -
or
Hence
y - l = Bx(y + l), (3.30)
or
3.3 Equations reducible to linear form 57
To find B, we again apply the boundary conditions. The condition

y = 1 at x = 0 is satisfied by (3.31), and to satisfy the condition
dy/dx = 1 at x = 0 we need to find dy/dx. From (3.31),
2B
£ (3 32)
dx (\-Bxf
At x = 0, 1 = 2B, giving B = 3. The complete solution, from (3.31), is
therefore
3.3 Equations reducible to linear form

In some cases a non-linear equation may be reduced to linear form
and hence solved by standard techniques. The substitution p = dy/dx
may help to carry out this reduction.
Example 4 Consider
Differentiating with respect to x, we have

dy d2y\ I dy\ dy d2y dy
dx / \ dx! dx dx dx
Eliminating the logarithm between (3.34) and (3.35) gives, on multi-
plying through by y + dy Idx,
dyd2y (dy\2 dy
-T^ri + A^) +y-r = ®- ( 3 - 36 )
2
dxdx \dx) dx
Hence either dy/dx = 0 or d2y/dx2 + 2 dy/dx + y = 0, both of which
are linear equations. The possible solutions are therefore
y=C (3.37)
and
y = e~x(Ax + B), (3.38)
where A, B and C are constants.
We now determine the values of these constants so that (3.37) and
(3.38) satisfy (3.34). Inserting y = C into (3.34) gives
ClnC + C = 0, (3.39)
so that C = 0 or C = e~1. In the case C = 0 the expression C l n C is

undefined. Hence the solution is C = e" 1 giving
y = e" 1 . (3.40)
For y = e~x(Ax + B), y + dy/dx = Ae~x. Inserting these into (3.34),
we have
*) + e~x(Ax + B) = 0. (3.41)
Cancelling e~* and using \n(Ac~x) = In A -x, we find
A(\n A - x) + Ax + B = 0, (3.42)
so that
,41n,4 + £ = 0. (3.43)
Substituting back for B> we have a solution
y = e~x(Ax - A In A), (3.44)

where A is an arbitrary constant which can be found by imposing a
boundary condition on y. ^4
Example 5 Consider
where y = \ when x = 0 and dy/dx = \ when y — 0. Putting
p= dy/dx, (3.46)
we have
d2y/dx2=pdp/dy. (3.47)
Now (3.45) becomes
pdp/dy + 2p2 = y2. (3.48)
This can be written
^ ( P 2 ) + 2p2 = / > (3.49)
which is a linear first-order equation in p2. Putting z=p2 in (3.49),

we find
4z = 2y2, (3.50)
which may be solved by multiplying by the integrating factor e4y.
3.4 Bernoulli's equation 59
Accordingly,
— (ze4y) = 2y2e4y, (3.51)

dy
and hence
ze4y = 2\ y2Q4y dy + C, (3.52)
where C is a constant. Performing the integration and dividing by

e4y gives
(3.53)
From the boundary conditions, z = (dy/dx)2 = je when y = 0. Hence

C = 0 and
P 2 = b> 2 -b> + * . (3.54)
Taking the square root,
P = f = ^V[(r-l)2 + ^ (3.55)
the positive sign being chosen so that p = \ when y = 0. The variables
may now be separated giving
(3.56)
where A is a constant. Hence
V2 sinh'M y-T1) =x + A (3.57)
Now y = i at x = 0, so A = 0. The solution for y is therefore
y = \[l + sinm>/V2)]. ^ (3.58)
3.4 Bernoulli's equation

A linear equation of the type
(3.59)
may be solved, as in Example 5 above, by multiplying by the
integrating factor exp(J P(x) dx). The non-linear equation
dy/dx+ P(x)y=y"Q(x), (3.60)

where ni=0, 1, is often called Bernoulli's equation. Using the

substitution
Hyn~l = z, (3.61)
then
and (3.60) becomes
~ + P(x)z = Q(x). (3.63)

1 — ndx
This is now of the linear form (3.59) and hence may be solved by the
integrating factor method.
Example 6 Consider
dy/dx + 2y=xy3. (3.64)
Hence
1 ^ + 1 = *, (3.65)
y dx y
and therefore
1d/1 \ 1.
2 (3.66)
2dx\y ) y2
Writing z = 1/y2, we have
dz/dx-4z = -2x. (3.67)
Multiplying by the integrating factor e~4x, we find
e~4xz = -2 [xe~Axdx + Cy (3.68)
where C is a constant. Performing the integration and multiplying by

e4* gives
z = \ = & + i + Ce4x. ^ (3.69)
Sometimes it is not obvious how an equation can be converted into

Bernoulli form. It may be helpful to invert the equation, that is, to
regard it as an equation for x in terms of y (where y is the independent
variable and x is the dependent variable). To illustrate this, we give
the following example.
3.5 Riccati's equation 61
Example 7 Consider
x(l - xy2) — + 2y = 0, (3.70)
whence
dx x{\-xy2Y {iJl)
This would seem to be difficult to solve but, by inverting, we obtain
dx _ x(l— xy2) _ x x2y

dy 2y 2y 2
and so
=y (3 73)
j+Y ~T- *
We see that (3.73) is an equation of Bernoulli type when viewed as an
equation for x in terms of y. Letting z = 1/x, then
= (3 74)
T~T ~o> '
dy 2y 2
for which the integrating factor is y~K Hence
zy~± = -[ -y-t dy + C, (3.75)
giving
(3.76)
The relationship between x and y is therefore
-=-&2 + Cyi. ^ (3.77)

x
3.5 Riccati's equation

Riccati's equation is
dy/dx =p(x)y2 + q(x)y + r(x), (3.78)

where /?, ^ and r are given functions of x. Suppose one solution of
(3.78) is known, say y = S(x). We then put
y = S(x) + l/u(x). (3.79)

From (3.79) and the differential equation (3.78), we obtain
*- + ^ . (3.80)
dx u dx
But, since S is a solution of (3.78),
dS/dx =p(x)S2 + q(x)S + r(x), (3.81)
and (3.80) becomes
du/dx + [2p(x)S + q(x)]u + p(x) = 0. (3.82)
This equation is of standard first-order linear form which may be
solved by the integrating factor method. Here the integrating factor is
v = exp{ J [2p(x)S(x) + q(x)] dx}, (3.83)

and consequently u(x) may be obtained.
Example 8 Consider
Here y = 1 = S(x) is a solution. Hence putting

y = 1 + 1/M, (3.85)
as in (3.79), we find
du/dx + (2 + l/x)u = - 1 . (3.86)
For this equation the integrating factor is
exp[[ (2 + IIx) dx] = xe2*, (3.87)

whence
= j-x^dx +Q (3.88)
where C is a constant. Integrating and dividing by xe 2
(3.89)
From (3.85) finally

1
(3.90)
3.6 Special forms of RiccatVs equation 63
3.6 Special forms of Riccati's equation
1.
In (3.78) we consider first the case of p(x) = — 1 so that
dyldx +y2 = q(x)y + r(x). (3.91)
Writing
so that
T- = - T 4 - A ( ? ) > (3-93)
dx zdx z \dxl
(3.91) becomes 2
dz dz
—2q{x)
-q(x)--r(x)z=0. (3.94)
d?- d-x
This is a linear second-order equation for z and appropriate methods
may now be applied (for example, the Frobenius series method used in
Chapter 2 if q and r are simple polynomials). We note that if
p(x) = 4-1 in (3.78), then the substitution
will give the equation

d2z / dz ,x
—2 - q{x) — 4- r{x)z = 0. (3.96)
This is also a linear equation.
2.
Secondly we consider the case of q(x) = 0 in (3.78) so that
dyldx =p(x)y2 4- r(x). (3.97)
We now make a change of the independent variable x to x', where xr
is defined by
dxf/dx=p(x)y (3.98)
or
x'= \p(x)dx. (3.99)
Then
dy dy dx' dy
dx dx' dx dx'
Hence from (3.97) we have
p{x) — = p{x)y2 + r(x). (3.101)
Equation (3.101) may be written as

dy/dxr=y2+f(x'), (3.102)
where
f{x') = r[x{x')]lp[x{x')} (3.103)
is found using the relationship (3.99) between x and x' (this may be
difficult to obtain in many cases since x' is explicitly given in terms of
x, but not vice-versa). Equation (3.102) may be transformed to linear
form since it is a Riccati equation with p(x) = +l as discussed in
1 above.
Example 9 Consider
dy/dx = 2xy2 - 2x3. (3.104)
This is a case of (3.78) with q{x) = 0. We therefore change the variable
from x to
x' = J2xdx=x2, (3.105)

following (3.99). Then
% = %dfx=2x% (3 106)
-
and substituting this into (3.104) and cancelling 2x gives
dy/dx' = y2 - x2 = y2 - x'. (3.107)
f
This is now a Riccati equation for y in terms of x' with p(x ) in (3.78)
equal to +1. Following (3.95), we put
1 dz
y—- (3.108)
and obtain from (3.107)
—4 = *'*- (3.109)
This is Airy's equation discussed in Section 2.9 and hence a solution
for z in terms of x' may be found in terms of Airy functions. The
solution for y is then obtained from (3.108) and finally substituting x2
for*' from (3.105). ^
3.7 The Lane-Emden equation 65
3.7 The Lane-Emden equation

The form of this equation is
d2y 2 dy
dx xdx
where a is a constant. Solutions which satisfy the boundary conditions
y = l, dy/dx = 0 at x = 0, (3.111)
are called Lane-Emden functions. Exact solutions are known when

a = 0, 1 and 5. In the cases a = 0 and a = 1, the equation is linear and
the solutions are easily found. When a = 5, let
Then after some algebra we find
4d2u/dt2-u + u5 = 0. (3.113)
Now writing v = du/dt, so that d2u/dt2 = v dv/du, (3.113) becomes
4vdv/du - M + M5 = 0. (3.114)
This equation may be integrated with respect to u to give
2v2-\u2 + Ub=C, (3.115)
where C is a constant. From the boundary conditions (3.111) and the

transformations (3.112), we see that x = 0 corresponds to f—»oo and
hence u and du/dt must tend to zero as t—»o°. Hence u = v=0 as
/—>oo and so, from (3.115), C = 0 giving
U2 = iw2_x^ (3.116)
or
v = du/dt = £MV(l - 5"4)- (3.117)
Separating the variables and integrating, we have
where A is a constant. To evaluate the integral on the left-hand side,

we put u4 = 3 cos2 6 so that
t + A = - J sec 6 d& = -ln(sec 6 + tan 6). (3.119)

Taking the exponential of both sides yields, with B = e~A,

Be1 = sec 0 + tan 0. (3.120)
Using cos 0 = w2/V3 and x = e~', we find
(3 m)
$V(^) -
the solution of which is
From (3.112), the solution for y is

V$B
l
Applying the boundary condition (3.111) that y = 1 when x = 0 gives
5 = 1/V3 and, finally,
y = 1/V(1 + k 2 )- (3.124)
3.8 The non-linear pendulum

The non-linear equation
d2d/dt2 + o)2 sin 0 = 0, (3.125)
where a> is a constant, is called the non-linear pendulum equation and
may be solved in terms of a class of integrals known as elliptic
integrals. Multiplying (3.125) by 2dd/dt and integrating, we have
{dOldtf = 2co2 cos 0 + C, (3.126)
where C is a constant of integration. Assume, for example, that
dd/dt = 0 when 0 = a. Then C = —2(o2 cos a and
(dd/dtf = 2co2(cos 0 - cos a). (3.127)
Taking the square root, separating the variables and integrating, we
have
d6
w\l2 Jo V(cos 0 - cos a)'
assuming 0 = 0 when t = 0. If t = T at 0 = 6lt then
a>V2J<> V(cos 0 - cos a)'

3.8 The non-linear pendulum 67
Using
cos 0 = 1 - 2 sin2 (0/2), cos a = 1 - 2 sin2 (or/2), (3.130)
(3.129) becomes
2a> Jo V[sin 2 (ar/2) - sin

sir2(<9/2)]' v
Now writing
sin(0/2) = £ s i n 0 , (3.132)
where k = sin(ar/2), then
in which k sin 0X = sin(0!/2). The integral in (3.133) is referred to as

an elliptic integral of the first kind and is usually denoted by
(3 134)
'w4>)' "
with OÂ;^1. Similarly, the elliptic integral of the second kind is
defined by
E(k, 0 0 = f ' V(l - k2 sin2 0) d(j>y (3.135)
where, as before, 0 ^ k ^ 1. In both cases the elliptic integrals are said

to be complete if 0 t = JT/2 and are then denoted by F(k) and E(k)>
respectively. Graphs of F{k, 0 0 and E(k, 0 0 are shown in Figures
3.1(a) and 3.1(b). We see that (3.133) gives T explicitly in terms of 0X
but not vice-versa.
We mention here that other integrals may be expressed in terms of
the above elliptic integrals and hence evaluated by consulting tables of
values (see reference on page 22).
Example 10 Evaluate
d<t>
1
Jo V(l - 4 sin
Putting 4 sin2 0 = sin2 6, we obtain
rjz/2
Example 11 Evaluate
dx
(3.138)
Putting x = sin 0, we obtain

rjr/6
h J rn
0
cos
(3.139)
However, 3 — 4 sin 2 <p + sin 4 <p = (3 — sin 2 (p){\ — sin 2 <p) = (3 — sin 2
cos 2 <p. H e n c e
_d$ = _L-/ 1 (3.140)

h = -rrz
- 3 sin 0) v3
3.9 Buffing's equation

In its simplest form Duffing's equation is
d2y/dx2 + ay + by3 = 0, (3.141)
where a and b are constants. We proceed, as in the case of the
Figure 3.1
k=0
(a)
3.9 Buffing's equation 69
non-linear pendulum equation, by multiplying (3.141) by 2dy/dx.

Then
This equation may be integrated directly to give
=C-ay2-lby\ (3.143)
where C is a constant of integration. Taking the square root of (3.143)

and separating the variables, we find
wrx+B- <3144)
where B is another constant. For particular values of a, b and C, the
left-hand side can be expressed in terms of an elliptic integral.
Example 12 If
f l = - l y + y \ (3.145)
with boundary conditions y = 0 and dy/dx = l when x = 0, find the
x-value for which y = 1.
Proceeding as above, we have
= 2(—\y +y ) — , (3.146)
which on integration gives

= C-ly2 + hy\ (3.147)
where C is a constant. Since y = 0 when dy/dx = 1, we find C = 1 and
ÂkWa-iZ + l/). (3.148)
Separating the variables and using y = 0 when JC = 0,
dy
f (3.149)
The value of x, JC0 say, for which y = 1 is
3?2. i..4,- (3-150)
Now putting y = sin <py we have dy = cos 0 d0 and

1 " b2 + 2 / = 2 ( 2 - 3 sin2 0 + sin4 0) = (1 - \ sin2 (p) cos2 0. (3.151)
Hence
— J sin <p)
using the definition of the complete elliptic integral of the first
kind. ^
In general, it is extremely difficult to obtain analytic (closed form)

solutions to Duffing's equation. However, a technique known as the
phase-plane method is useful in finding the nature of the solutions of
(3.141) for any a and b values and given initial conditions. We define
dy/dx = w, (3.153)
so that (3.141) becomes
dw/dx = -ay-by3. (3.154)
The equations (3.153) and (3.154) are now a particular case of the
general system
dy/dx = P(y,w),}
(3.155)
where, for Duffing's equation,

P(y, w) = w, Q(y, w) = -ay - by3. (3.156)
It is for a system of the form (3.155) that the phase-plane method is
appropriate. By eliminating x from (3.155), we have
P(y, w)
(3.157)
dw Q{y,w)'
In particular, for the Duffing equation,
dy w
(3.158)
dw ay + by3'
Separating the variables and integrating, we find the solution of
(3.158) to be
\ay2 + \ C, (3.159)
where C is a constant determined by the initial conditions. The
(y, w) = (yy dyldx) plane is called the phase-plane and the nature of
the solution (3.159) of (3.158), as represented by the curves of y and w
in the phase-plane for different values of C, yields some information
about the solution of Duffing's equation, but does not give the solution
3.9 Buffing's equation 71
itself. For example, it can be shown that the existence of simple

(non-intersecting) closed curves in the phase-plane implies the exist-
ence of periodic solutions of the original equation. Specifically,
Duffing's equation will have periodic solutions if there exist simple
closed (y, w) curves described by (3.159). To illustrate this, we
consider the following values of a and b:
1. a = l, b=2
In this case (3.159) becomes
/ + y2 + w2 = 2C. (3.160)
The graph of this family of curves is shown in Figure 3.2. From
(3.160), we see that it is not possible to have C < 0 for any initial
conditions. For any value of C ^ 0, the curve is closed, which indicates
that only periodic solutions of the equation exist for these values of a
and b.
2. a = 1, b = - 2
In this case (3.159) becomes
y4-y2-w2=-2C. (3.161)
Figure 3.2
C=2
The graph of this family of curves is shown in Figure 3.3. IfO

both open and closed curves exist in the phase-plane, whereas if C < 0
or C ^ J only open curves exist. Hence, periodic solutions of the
equation exist only if the initial conditions are such that 0 ^ C < J, and
at t = 0 the point lies within the shaded region of Figure 3.3. An
example of this is shown for C = ^ . The phase-plane method is also of
importance in problems of stability of solutions of non-linear systems
of the form (3.155). The technique requires knowledge of the fixed or
critical points (y0, w0) of (3.155) defined by
Wo) = 0. (3.162)
The stability problem is not central to our discussion of obtaining
solutions of non-linear equations and, accordingly, we shall not pursue
it here.
For the Duffing equation, the differential relation (3.158) defining
the curves in the phase-plane is simple to integrate since the variables
may be separated. This may not be the case for other non-linear
Figure 3.3
c—i-
Problems 3 73
equations. For example, the van der Pol equation
- l ) ^ + >>=0, (3.163)
where fi is a constant, has curves in the phase-plane defined by

dy w
This cannot be integrated by separation of variables but solution

curves can be sketched using, for example, the method of isoclines1.
For a more detailed account of analytic techniques for studying
non-linear differential equations, including stability problems and
chaotic behaviour, the reader is referred to a standard text*. In
general, however, non-linear equations require numerical methods for
their solution.
Problems 3
1. Obtain a solution of the equation
given y(0) = 0 and / ( 0 ) = \ll.

2. By writing dy/dx = p, solve
given y = dy/dx = 1 when x = 1.

3. By writing xy' — y = v(x), solve
x3y"={xy' -yf.
y
4. By writing t = z, solve
In x x In x
5. Show that the equation
y" = xy + 2y3 + \
f
E. Kreysig, Advanced Engineering Mathematics (Wiley, New York, 1988)
Section 1.10.
* D. W. Jordan and P. Smith, Nonlinear Ordinary Differential Equations,
(O.U.P, Oxford, 1987)
has a first integral y' = \x + y2. By finding the general solution of

this Riccati equation in terms of Airy functions, obtain a solution of
the original equation containing one arbitrary constant.
6. Verify that
+
' W2
where C is a constant, satisfies Pinney's equation
y"' + p(x)y = C/y3,
where u and v are independent solutions of
z" + p(x)z = Q
and W = uv' — vu'. (For equations with no first derivative term
dz/dx, W, the Wronskian, is a constant). Hence show that
is a solution of
y" + y = 2/y3.
7. Show that
(l)
J2/V3 V(2M 4 - 3M2 + 1) " V2 \ V2 ' 3 / '
where F and £ are elliptic integrals of the first and second kinds,
respectively.
8. Show that the inhomogeneous Duffing equation
y" + ay 4- by3 = A cos(3cot)
has an exact solution
y = \-g-j cos(o>o,
provided
4
Approximate solutions of
ordinary differential equations
4.1 Power series

We begin by recalling briefly some elementary ideas. Linear equations
of the form
dy
+f(x)y =Q (4-i)
or
—2+p( x ) \~(l(x)y = Q (4.2)

dx dx
can be solved in many cases by substituting a Taylor series expansion
about some fixed point x0 of the form
(4.3)
n=0
Provided f(x), p(x) and q(x) are also expressible as Taylor series
about x0, then the coefficients of like powers of x — x0 may be equated
in the differential equation and, by solving for the coefficients an, the
solutions of the equation may be obtained. We shall mostly be
concerned here with the case x0 = 0 for which the series (4.3) becomes
a MacLaurin series.
Example 1 Consider
dy/dx = 2e-x-y, (4.4)
given that>>(0) = 0.
75
76 Approximate solutions of differential equations
Writing y = Y^ânxny substituting this into (4.4), and expanding

e~* as a power series, gives
(4.5)
n=0
The lowest power of x on each side is x°, and equating coefficients of

powers of JC°, xl, X2, X3, . . . gives, respectively,
= 2-a0, (4.6)
2a2 = -2-au (4.7)
3«3 = (4.8)
2!"^
2
4« 4 (4.9)
~~~J\~a3'
and so on. Hence, since y(0) = 0, ao = 0 and therefore
ax = 2y a2 — (4.10)
giving
y(x) = 2x-~ Z X "i ^ 2>X T~ . . . . (4.11)
The coefficient of xn for n ^ 1 is easily seen to be
Hence, applying the ratio test for convergence, we find

\an+jan\ = \ln. (4.13)
Consequently the series (4.11) converges for all values of x since the
ratio of the (n + l)th term to the nth is x/n, which tends to zero for
any finite x as n^>o°. The exact solution of (4.4) is easily found by the
integrating factor method to be y{x) = 2xz~x, of which (4.11) is the
MacLaurin expansion.
This method is equivalent to assuming a MacLaurin series
x2
y(x)=y(0)+xy'(0) + -y"(0) + ... (4.14)
and obtaining the derivatives of y by successive differentiation of the

equation. From (4.4),
d2y/dx2 = -2e~x - dy/dx (4.15)
4.1 Power series 77
and, in general,
Hence, at x = 0,
y(n)(0) = 2(-1)"" 1 - / " - " ( 0 ) . (4.17)
Since y(0) = 0, we find from (4.17) that / ( 0 ) = 2, y"(0) = - 4 and so

on. On substituting these into (4.12) we again obtain the solution
(4.11). ^
Although the power series method is readily justified in the case of

linear equations, it is more difficult to justify its use for non-linear
equations of the type dy/dx =f(x, y). Proofs of convergence and of
the existence of solutions are complicated in the case of non-linear
equations, and power series methods should be used with caution. We
illustrate this with an example.
Example 2 Consider
dy/dx =x-y2y (4.18)
given that>>(0) = 0.
Writing, as before, y = E^=o0«*n and substituting in (4.18) gives
ax-\-2a2x-\-3a3x2-\-. . . =x - (ao + axx 4-. . .) 2 . (4.19)
Comparing coefficients of like powers of x gives
ax = ~al (4.20)
(4.21)
\, (4.22)
4a 4 = — 2axa2 — 2aoa3, (4.23)
5a5 = -2ala3 — a\ — 2aoa4, (4.24)
and so on. Since y(0) = 0, we have from the original series ao = 0.
Hence, from (4.20)-(4.24),
0i = O, fl2 = i «3 = 0, 04 = O, 05 = - ^ , (4.25)
and so on. We have finally
y(x) = 12x2-T0x5 +.... (4.26)

Although a series solution can be generated in this way, it is difficult to

find an expression for the general term and hence to prove
convergence. ^
4.2 Frobenius series

In the case of the second-order equation
d2y dy
—2^-p(x) \- q(x)y = 0, (4-27)
where p(x) and q(x) are given functions, a power series solution does
not necessarily exist. However, provided p(x) and q(x) are
differentiate and single-valued at a point x0 (that is, they are regular
at x0), then x0 is called an ordinary point of the equation and a Taylor
series centred at x0 will provide the solutions. If, however, x0 is not an
ordinary point but nevertheless
lim (x-xo)p(x) (4.28)
and
lim (x-xo)2q(x) (4.29)
are finite at x0, then x0 is called a regular singular point of the

equation and near this point it is possible to find at least one solution
of the form
y(x) = E <*n(x - *o) m+ ", (4.30)
where m is some number. If (4.28) and (4.29) are not finite, then x0 is
an irregular singular point. The series (4.30) is known as a Frobenius
series and has been used in Section 2.5 to develop the solution of the
Bessel equation. In applying this method, certain special cases arise
depending on the two values mx and m2 of m. If these values are
identical or differ by an integer, then only one solution can be found
by this method. This was the case with the Bessel equation of integer
order n for which mx — m2 is an integer 2n and (see (2.92))
We now give a few examples of equations, taking x0 = 0 as the point

of expansion.
For this equation x = 0 is an ordinary point and power series

solutions exist.
d2y 3 dy 1
(ii) —i + - — + - 2 . y = 0 - (4-32)
Here * = 0 is a regular singular point since \imx_+ox(3/x) = 3 and

\imxôx2(l/x2) = 1 are finite. At least one Frobenius series
solution exists.
d2y 1 dy
(iii) ~T~2 + ~2lT + xy = 0 - (4.33)
ax x ax
In this case x = 0 is an irregular singular point since
\imxôx(l/x2) does not exist.
We note that x = 0 is a regular singular point of the Bessel equation
(2.61) since \imxôxp(x) = \imxôx(l/x) = l and l i m ^ o * 2 ^ * ) =
\imx_>0(x2 — v2) =—v2 are finite. The application of the Frobenius
series method follows that of Section 2.5 and we shall not illustrate the
method further.
4.3 Picard iterative method

Consider the standard first-order equation
dy/dx=f(x,y), (4.34)
with y(x0) =y0 given. Equation (4.34) has the formal solution
y(x)=A+ff(x,y)dx, (4.35)
x
0
where, since y(x0) = y0, A = y0.

A sequence of functions ylf y2, . . . , yn is generated as follows: first
we insert y0 into the right-hand side of (4.34) to get
dy1/dx=f(x,y0). (4.36)
Solving for yx and inserting this into (4.34) gives
dy2/dx=f(x,yi), (4.37)
and so on. Hence we form the sequence by iterating the relation
f(x,yn)dxy (4.38)
i
where n = 0, 1, 2, . . . , from (4.35). It can be proved that this

sequence tends to a solution of (4.34) on some interval of x, except in
particular circumstances (see Example 6).
Example 3 Consider
f'*-^, (4.39)
dx x
given y(0) = 0. (This is a Riccati equation - see Section 3.5.) Then
(4.40)
J0 A
Hence
yi(x)= I X-^-dx=X-y (4.41)

Jo x z
rxxX2 ++( X(x^2/o\2
2
/ /2)
Z J
2 V2
x 2 V44
x
y2(x )= dx=X- + X
-y (4.42)
Jo X Z 10
x2 x4 x6 x8
(4 43)
- -
Each of these approximations is a finite series. This sequence
approaches the exact solution as n —• °°. ^
Example 4 Consider
dy/dx = JC + siny, (4.44)
with >>(0) = 7i 12. Then
yrt+1(x) = —h (JC 4- siny n ) dx. (4.45)

Z JQ
Hence
rx 2
7t l 71 X
However,
*-2 \ i
(4.47)
2
cannot be integrated analytically, and so the next function in the
sequence cannot be found except, perhaps, for sufficiently small values
of x for which the sine term in (4.47) may be expanded. ^
The Picard method of successive approximation may be used on a

system of equations. Suppose we want to solve for y(x) and z(x),
where
dy/dx=f(x9y,z), dz/dx = g(x, y, z), (4.48)
given, say, _y(0) = yQ and z(0) = z0. Then the iterated sequence is given
by
yn + i=yo+\ f{x,yn,zn)dx, zn+l = z0+\ g(x,yn,zn)dx.

A) A)
(4.49)
Example 5 Consider
dy/dx=x+z, dz/dx=x-y2, (4.50)
where y(0) = 2, z(0) = 1. Then
r 1 9
yj(jc) = 2 + (JC + 1) djc = 2 + JC + 2* , (4.51)
Jo
ZI (JC) = 1 + [ (JC - 22) dx = 1 - 4JC + \x2. (4.52)
Jo
Similarly
y2(x) = 2 + (JC + 1 - 4 J C + ^JC ) dx = 2 + JC — |JC + ^JC , (4.53)

Jo
Z2 ( x ) = 1+ f ( J C - [ 2 + JC + \x2}2) dx = l-4x-3ix2-x3- \xA - ^x5,

Jo
(4.54)
and so on. ^
The Picard method does not always provide a non-trivial solution as

we demonstrate in the next example.
Example 6 Consider
dy/dx = \?y, (4.55)
given y(0) = 0. Exact solutions of this equation are y = 0 and y = \x2.
Applying the Picard method, we find
yn+l(x) = y(0) + fy/[yn(x)] dx. (4.56)

Jo
Hence
yi(x) = 0+ I 0dx = 0, (4.57)

Jo
y2(JC) = 0 + I 0dx = 0, (4.58)
Jo
and so on. For this equation, the Picard method only generates the
trivial solution y = 0. ^d
The failure in Example 6 to produce the non-trivial solution is

related to the Lipschitz condition which states that for the general
equation (4.34), \df/dy\ must be finite in the range of integration for a
non-trivial solution to be generated. In Example 6, f(x, y) = Vy and
hence \df/dy\ = 1/2Vy which tends to infinity as y—»0.
4.4 Perturbation series

An important technique for solving non-linear ordinary differential
equations is the perturbation method. This method requires that the
equation is non-linear in virtue of a small parameter, say e. By
expanding the dependent variable y(t), say, as a power series in £ we
generate a solution to any desired order in e. Consider the following
typical example.
Example 7 Duffing's equation (see Section 3.9) has the form

d2y/dt2 + ay + by3 = 0. (4.59)
Suppose we take the case where a = 1 and b = e, where e is a small
parameter. Suppose also that y(0) = 1 and y '(0) = 0. We expand y as a
series in e, as follows:
y(t) = yo(t) + eyx(t) + e2y2{t) + . . . . (4.60)
Substituting this series into (4.59) gives
dt2 dt2 dt1

+ (ya + eyi + e2y2 + . . . ) + ^ o + ey, + . . . ) 3 = 0. (4.61)
Equating to zero the coefficients of successive powers of e gives
^ ^ (4.62)
, (4.63)
0, (4.64)
and so on. Now, given y(0) = 1 and y'(Q) = 0, we have from (4.60) on
equating powers of e, yo(0) = 1, yo(0) = 0,
*(0)=y2(0) = . . . = 0 (4.65)
and
y[(0)=ym = ---=0. (4.66)
Hence (4.62) gives
(4.67)
4.5 Normal form 83
whilst (4.63) gives

d2yjdt2 + y1 + cos31 = 0. (4.68)
This is a linear equation with constant coefficients which can be solved
by elementary methods. Writing cos31 = ^(cos 3t + 3 cos t), we find
using (4.65) and (4.66) that
yx(t) = - ^(cos t - cos 3t) - 11 sin t. (4.69)
This may be inserted into (4.64), together with (4.67), to give y2(t) and
so on, giving a solution of the form (4.60) to whatever order in e is
required. ^
The previous example is only meant to illustrate the general

approach. The method may also be applied to the van der Pol
equation (see (3.163)) where ji is a small parameter. To proceed
further would take us too far into technical details (see reference * on
page 73).
4.5 Normal form

In the following sections we shall be dealing with approximation
methods for solving second-order linear differential equations in
normal form for which there is no first derivative term. This is no real
restriction since any second-order linear differential equation can be
cast into normal form as we now show.
Consider the equation
Now let y = u(x)v(x). Then

u"v + 2w V 4- uv"+p(uv' + vu') + quv = 0. (4.71)
We now choose u so that the first derivative term in v' vanishes by
putting
2u'+pu = 0. (4.72)
Hence
u'lu = -\py (4.73)
giving
- ^ | p(x)dx}. (4.74)
Here A is an arbitrary constant of integration which may be taken

equal to unity without loss of generality since u occurs in every term of
(4.71). Inserting u' and u", obtained from (4.74), into (4.71), we find
dx2 V 2dx *A
Hence the solution of (4.70) may be obtained from (4.74) and the
solution of (4.75), which is an equation in normal form.
Example 8 Reduce
— ! + x — + \x2y = 0 (4.76)
dx dx
to normal form and hence obtain its general solution.
Proceeding as above with p(x)=x and q(x) = \x2, we find from
(4.74)
u — exp"j 2 | % &x [ — ^ y \^''')

I J J
and from (4.75),
d2v/dx2 + {\x2 - \ - \x2)v = 0 (4.78)
or
d2v/dx2-^v=0. (4.79)
Hence
v = AzxN2 + Bt~xN2 (4.80)
and finally therefore
y = uv = e-ix\Aex/y/2 + Be~x/V2), (4.81)
where A and B are arbitrary constants, ^d
4.6 The W.K.B. (Wentzel-Kramers-Brillouin) approximation

This is a method of obtaining an approximate solution to any linear
second-order differential equation in normal form when the second-
order derivative term is multiplied by a small parameter. The form we
shall consider is
e2d2y/dx2=f(x)yy (4.82)
where 6 is a small parameter, f(x) is a given function, and y is given
at, say, two particular x-values in some range. An important occur-
rence of (4.82) is the Schrodinger equation in quantum mechanics. We
note that, as e—>0, the equation is singular in the sense that the order
decreases from second to zero order when e is set equal to zero. We
must therefore expect the solution to be singular as e—»0. For
example, consider
e2d2y/dx2 + y=0, (4.83)
subject to y(0) = 0, y(l) = 1. This equation has the exact solution
sin(x/6)
y (4 84)
^û7y -
Both sin(jt/£) and sin(l/e) rapidly oscillate as 6-^0 for any given x in
the range and (4.84) becomes undefined at e = 0.
For equations of the type (4.82), the W.K.B. method gives an
approximation which is often close to the exact solution over much of
the range. In attempting to find an expression for y(x) as a power
series in e, we must have a series which is singular as e —» 0. We adopt
the trial expression for y(x) in the form
y(x) = exp{^ | [S0(t) + 65,(0 + €2S2(t) + . . . ] * } , (4.85)
which is singular as e—»0. Here SQ(t), S{(t), . . . are functions which

will be determined by the method. We could have a multiplying
constant in front of the exponential in (4.85) if we wished but, instead,
by leaving the integral as an indefinite one, we can always fix the lower
limit by knowing some initial or boundary conditions on y(x). From
(4.85),
dy/dx=- [So(x)+ eStix) + e %(x) + ]y(x) (4.86)
and
d2y 2
[±-2[S0(x) + el S^x) + 62S2(X) + ..
dx2 ]
1
+ ) + eS[(x) + €2S2(X) x). (4.87)
Hence, inserting (4.87) into (4.82) and cancelling y(x) on both sides,
[S()(x) + eSl(x) + e2S2(x) + ...]2
+ e[So(x) + eS[(x) + e2S'2{x) + . . .} =f(x). (4.88)
Equating like powers of e on each side of (4.88) gives an infinite set of

equations from which So, Su S2, . . . can be determined. The first two
of these are
S2Q(x)=f(x), (4.89)
25 (x)S (x) + S'(x) = 0 (4 90)
Hence, from (4.89),
S0(x) = ±[f(x)]K (4.91)
while from (4.90)
i. (4.92)
In the expression for y(x) in (4.85), we require a term
expl^S^t) di\. The integral of Sx from (4.92) is simply -\ ln[S0(x)] so
that
exp[|' 5,(0 dt\ = [$,(*)]"* = [f(x)]-\ (4.93)
using (4.91) (apart from constants which again can be combined into
the specification of the lower limit of integration).
The W.K.B. approximation consists of neglecting all terms in (4.85)
of order e and higher, so that only S0(x) and S^x) are required.
Hence, using (4.91) and (4.93), we have the two W.K.B. solutions
y(x) = [/(*)]"' exp{± i f [/(<)]* dt\ (4.94)
The general solution of (4.82) within the W.K.B. approximation is

therefore a linear combination of the two solutions:
y(x)
- iTSjH; /*[/w|i •") + i7SF xp K f i«'»i! 4
(4.95)
where A and B are constants.
The solution (4.95) will differ from the exact solution to (4.82) by
terms of order e whenever f(x)¥z0. However, if points exist where
f(x) = 0 (called the turning points of the equation), (4.95) will diverge
at these points whereas a numerical integration of (4.82) will give a
finite solution there. The W.K.B. method, although a good ap-
proximation over much of the range, therefore exhibits the wrong
behaviour close to the turning points. We discuss this is more detail in
Section 4.8. Solutions which are more accurate away from turning
points may be found by including the higher order terms S2(x), S3(x)
and so on. This will not concern us here.
Example 9 Consider
e2 d2y/dx2 = (1 + x2fyy (4.96)
subject to y(0) = 0 and y(1) = 1. Then f(x) = (1 4- x 2 ) 2 and (4.95) gives
(4 97)
' -
where we have chosen the lower limits of integration to be zero.
Hence imposing y(0) = 0 gives
0 = A exp| - . ol + B exp I"- - . ol (4.98)

or
A + B=0. (4.99)
Substituting B = —A into (4.97) and performing the integration, we

find
Applying the boundary condition y(l) = 0 gives
l = y/2A sinh(4/36). (4.101)

Hence
^ = l/V2sinh(4/36) (4.102)
and from (4.100)
Example 10 Consider the Airy-like equation
e2d2y/dx2 + xy = 0 (4.104)
(compare (2.209)).
Then/(jc) = -x and (4.95) gives
(4.105)
whence
'3e i -2Lv 3 / 2 /3
(4- 106)
X*
— x$ + d\ (4.107)
where C and 6 are arbitrary constants. ^
4.7 Eigenvalue problems

An important problem in physics and engineering is that of finding the
eigenvalues of a differential equation. Suppose we take
d2y/dx2 + Xy=0, (4.108)
where y(0) = 0, y(jz) = 0, and A is a constant. Then clearly y - 0 is a
trivial solution. For particular values of A, non-zero solutions will
exist. These values of A are called the eigenvalues and the correspond-
ing solutions are the eigenfunctions. Now the solution of (4.108) is
y(x) = A cos(xVA) + B sin(jt VA), (4.109)
so for y(0) = 0 we have A = 0, whilst y{n) = 0 gives
B sin(WA) = 0. (4.110)
Now, if we choose B = 0, (4.110) is satisfied but, since ^4=0, the
solution for y is the trivial one. Accordingly, we require
sim>VA) = 0 (4.111)
for a non-trivial solution. Hence VA = n, with w = 1, 2, 3, . . . and the
eigenvalues are therefore
X = n2, (4.112)
where n — 1, 2, 3, . . . . The corresponding eigenfunctions are
sinjc, sin(2x), sin(3x), . . . .
The properties of various special functions discussed and listed in
Chapter 2 are often useful in finding the eigenvalues of some types of
linear second-order differential equations. For example
d2y/dx2 + (A - x2)y = 0, (4.113)
where -OO<JC<OO, has non-zero solutions which tend to zero as
x^> ±oo only when A = 2n + 1, n = 0, 1, 2, . . . . These solutions are the
Weber-Hermite functions given by yn(x) = e~x2/2Hn(x), where Hn(x)
are the Hermite polynomials (see (2.200)). There is an infinite number
4.7 Eigenvalue problems 89
of eigenvalues. Likewise the equation

d2yldx2 + (A - x)y = 0, (4.114)
where 0 ^ x < o°, has solutions which are the Airy functions
y(x) = Ai(x - A), Bi(x - A) (4.115)
(see (2.209)-(2.212)). As x^>™, only the Ai solution tends to zero
(the Bi solution tends to infinity as *—»oo). Hence, if we wish to solve
(4.114) subject to the boundary conditions
0, y(oo) = 0, (4.116)
we see that
Ai(-X) = 0. (4.117)
This determines the set of A values which are the eigenvalues of the
problem. From tables (see reference on page 22), the A values are
approximately 2.34, 4.09, 5.52, . . . . There is an infinite number of
values of A for which (4.117) is true (see, for example, the graph of
Ai{x) in Figure 2.8).
In general, although we will not prove it here, any differential
equation of the form
Jx [P{X) £
with y(a) -y{b) = 0, where a and b are constants (or ±°°), has an
infinite number of real positive eigenvalues provided p(x) > 0 , q(x) ^
0 and r(x) > 0. Such a differential equation (with these boundary
conditions) is called a Sturm-Liouville system. The calculation of
exact eigenvalues is difficult, if not impossible, and approximation
techniques are valuable. The following example shows the use of the
W.K.B. method in this connection.
Example 11 Consider
d2y/dx2 + Xx2y = 0, (4.119)
given y(0) = 0, y(jz) = 0. Writing this as
\d2y
id—xy (4 120)
-
and comparing with the basic form (4.82), we see that f(x) = —x2 and
e2 = I/A. Since e is assumed to be small, we expect the method to give
the large eigenvalues A with good accuracy. The general solution of
(4.119) is, from (4.95),
>>(*) =-y-expjVxJ f*)+-êxp(-iVAj tdty (4.121)

or equivalently,
y(x) = -^sin(VA* 2 /2) + -^-cos(VA* 2 /2), (4.122)
where A> B, C and D are arbitrary constants. To satisfy the boundary

condition y(0) = 0, we must have D=0 in (4.122). The remaining
boundary condition y(jt) = 0 then gives
S- sin(VA JT 2 /2) = 0, (4.123)
or
VkJT2/2 = njt, (4.124)
where ft is an integer (which must be large in order that A itself is
large). Hence approximate eigenvalues are given by
)2, (4.125)
for large n. ^
It should also be mentioned that Sturm-Liouville theory for (4.118)

shows that the eigenfunctions yn can be made, by suitable multiplying
constants, to satisfy the relations
I ym(x)yn(x)r(x)dx = dmn, (4.126)
where 6mn is the Kronecker delta (see (1.23)). Such eigenfunctions are
said to form an orthonormal set. Since the W.K.B. method of
Example 11 gives the eigenfunctions ((4.122) with D = 0 and A given
by (4.125)) up to an arbitrary multiplicative constant, this constant
may be fixed by (4.126). Hence the eigenfunctions and eigenvalues for
large n are uniquely determined.
4.8 The Liouville-Green technique

This technique is an attempt to transform the normal form equation
d2y/dx2=f(x)y (4.127)
into another, also in normal form with the same number of turning
points, which is soluble exactly or approximately in terms of known
functions.
In (4.127), we make a transformation from x to a new variable £ by
writing x=x(%)> and from y{x) to a new dependent variable G(§),
where
y(x)= (4-i28)
W?'
with f-' = d^/dx. The inclusion of the factor (§')~* ensures that the
equation for G(£) is also in normal form, as we now show. From
(4.128),
Further,
the terms in dG/dt; cancelling. Inserting (4.131) into (4.127) and using
(4.128), we find
(f(x) , A i^ (4 132)
,2 ^ , 2
where
(4-133)
is commonly called the Schwarzian derivative. We still have to specify

the relationship between x and § from which, in principle, the terms in
brackets in (4.132) can be expressed in terms of § alone. In particular
cases this term can be retained in its entirety and (4.132) solved
exactly, giving an exact solution to (4.127). In others, the function A
can be shown to be small compared to the term f{x)%'~2 and may
be neglected, allowing an approximate solution of (4.132), and of
(4.127), to be found. We note here that if A is neglected and £' 2 is
chosen to be f(x) in (4.132) then the approximate solutions so
generated are precisely those of the W.K.B. approximation. In this
case (4.132) becomes
2 2
=G (4.134)
and hence
Bz~\ (4.135)
Since %'2=f(x), we have
]* dt. (4.136)
Combining (4.135) and (4.136) with (4.128) gives the W.K.B. result
(4.95) (with e formally equal to 1). However, if the equation (4.127)
has turning points, that is, if f(x) has zeros in the range of interest,
then we clearly should not choose %'2=f(x) since £' 2 is always
positive whereas/(JC) changes sign. This is the origin of the divergence
in the W.K.B. solutions at the turning points of the equation. The
correct way to proceed is to choose some function h(%) having the
same number of zeros as/(x) in the given jc-range, and then let
h(%)%f2=f(x). (4.137)
The choice of h(%) should be such that A can be neglected in (4.132)
and such that the resulting approximate equation for G(§),
d2G/d^2 = h(%)G, (4.138)
is soluble in terms of known functions. We now illustrate various
features of this technique.
Example 12 Consider
d2y/dx2 = exy, (4.139)
where -oo<jt<o°. Then performing the Liouville-Green transforma-
tion (4.128), we find
d2G I e* \
~d]=2 = \]=;2*A)G' (4.140)
We now choose
£' 2 = e*, (4.141)
which gives on integration
§ = 2e*/2. (4.142)
Hence, for -oo<x<oo, w e have 0 < £ < o o . From (4.142) the deriva-
tives up to £'" can be calculated and substituted into the expression
(4.133) for A. We find
Hence (4.140) becomes
(4.144)
This equation was discussed in Chapter 2 (see (2.157) and (2.159))

where it was shown that the general solution is
where A and B are arbitrary constants and / 0 and Ko are the modified
Bessel functions. Now, from (4.141) and (4.142), (§')"* = e~*/4 =
(2/£)i From the inverse transformation (4.128) relating G(§) to y(x),
we have finally
y(x) = AI0(2ex/2) + M 0 (2e x / 2 ), (4.146)
where A and B are arbitrary constants. ^
Example 13 Consider
d2y/dx2 = x4y (4.147)
for x > 0. Transforming as before, we find
Now choosing
%'2 = x\ (4.149)
we have
? = 3*3 (4.150)
and, from (4.133),
A = -2/9§ 2 . (4.151)
f-('-£)°-
Writing %=d/2, (4.152) becomes
d2G /. 2 \
^^ 2 \ 9^ 2 / v }
Comparing (4.153) with (2.160) and its solution (2.163), we see that
v2 - \ = —\ giving v = \, and the solution of (4.153) is therefore
(4.154)
(4.155)
Using the transformation (4.128) to find y(x), we have

y(x) = C\Txh(W) + D\Tx K^x3). (4.156)
We note here that the general form
d2y/dx2 = xny (4.157)
for x > 0 , n =£2, may be solved by the method of this example. Two
simple cases arise:
Case 1 n = 0,
y = A sinh x + B cosh JC. (4.158)
Case 2 n = —4,
( ) ( Y ^ (4.159)
Example 14 As we showed in Chapter 3, the Riccati equation

dy/dx=y2-ex (4.160)
can be transformed to linear form by writing
y—\%. (4.161)
After some algebra, (4.160) becomes

d2z/dx2 = exz, (4.162)
which can then be solved as in Example 12, and the result substituted
into (4.161) to find y. This illustrates the use of the Liouville-Green
technique in solving a non-linear equation. ^
The above examples demonstrated the use of the Liouville-Green

technique in solving equations exactly. In many cases this is not
possible and we now demonstrate its use in solving equations
approximately. We will not discuss equations with no turning points,
since, as shown above, the solutions obtained are the same as those of
the W.K.B. method. We therefore apply the technique to equations
with one or two turning points.
Example 15 Consider
d2y/dx2 = N2x(l + x2fyy (4.163)
with -oo <x < oo, and TV large. This equation has a single turning point
(since the function x(l + x2)2 changes sign only at x = 0). Carrying out
the Liouville-Green transformation, we find
We require (4.164) to have one turning point (if possible) and to be

analytically soluble for G if A is small enough to be neglected.
Choosing
%r2£ = N2x{l+x2)2y (4.165)
we obtain the Airy equation
d2G/d%2=%G, (4.166)
provided A can be neglected. From (4.165), we have
%'\^ = N\x\Hl+x2) (4.167)
so that on integration,
§ | ^ = N(§|*|Ml^). (4.168)
Hence
§ = Aflc(l + Wf. (4.169)
The solution of (4.166) is
+ D &"(§), (4.170)
and hence the approximate solution for y{x) is found from y(x) =
(£')~^G(§) since (£')"* can be evaluated in terms of x, and § can be
substituted using (4.169). Further, from (4.169), A can be calculated.
It is found that A is bounded (that is, finite for all x) and that it has an
overall factor N~% which is a small quantity for large N. It is therefore
small over the whole range of x> and may be neglected in (4.164). ^
For problems in which the equation has two turning points, we

consider an important class of eigenvalues problems for which the
equation takes the form
d2yldx2 = [-E + V(x)]y, (4.171)

where — OO<JC<OO, V{x) is finite for finite x> and y—»0 as x—» ±o°.
Here E is the eigenvalue. Transforming equation (4.171), we obtain
d2Gldlf=\ - : , ; v ~ ' + AG. (4.172)

By assumption, the function —E + V(x) changes sign twice in the

range. The simplest equation with two turning points which is soluble
in terms of known functions is that satisfied by the Weber-Hermite
functions (see (2.206) and (2.207)),
d2G/d%2 = (-A + §2)G, (4.173)
where A = 2/? + l (n an integer) for finite G as §—»±<». Hence we

choose
§' 2 (-A + §2) = -E + V(*) (4.174)
in (4.172). Neglecting A, (4.172) becomes (4.173) which can be solved

in terms of the Weber-Hermite functions. We note, however, that on
taking the square root and integrating (4.174), the relationship
between £ and x can only be found implicitly (that is, we cannot write
£ explicitly as a function of x). Nevertheless, an approximate formula
giving the eigenvalues En can be obtained from (4.174). The turning
points, say x1 and x2, correspond to £ = ±VA. Hence integrating
(4.174) between the turning points after taking the square root gives
- + VA rX2
J V(§ 2 - A) d% = y/[E - V(x)] dx. (4.175)

-VA JXl
Performing the integration on the left-hand side and substituting
A = In + 1, we find
f2 y/[En - V(x)] dx = n(n + \\ (4.176)
This is known as the Bohr-Sommerfeld formula. It can also be derived

within the W.K.B. approximation using so-called connection formulae
which are not discussed here. For each value of n, a value of En can be
found (if necessary, numerically) and this value is an approximation to
the corresponding exact eigenvalue of the equation. In fact, for a
particular class of potentials V(x), it can be shown that the formula
provides the exact eigenvalues of the equation.
Problems 4
1. Find the first three non-zero terms in the series solution of the Airy
equation
d2y/dx2 = xy,
given y = 0 and dy/dx = 1 at x = 0.

Problems 4 97
2. Find the Frobenius series solutions of the differential equation
dx2 dx
in the form
Show that, if y(0) = 0, then

x x2
is the solution (A arbitrary).

3. Obtain by Picard's method the first three terms of the series
solution of the equation
dy/dx = 1 + 2xy + y2,
given y(0) = 0.
4. Obtain by Picard's method the first three approximations to the
solution of the simultaneous equations
dy/dx = 2x + z,
dz/dx = 3xy -\-x2z,
given>>(0) = 2, z(0) = 0.
5. Use the perturbation series method of Example 7 to obtain a
solution, to first order in the small parameter \iy of the van der Pol
equation
taking yo(t) = cos t.

6. Reduce the equation
dx2 dx
to normal form, and hence find its general solution.
7. Use the W.K.B. method to find approximate solutions of the
equation
d2y/dx2 = N2x4y.
8. Use the Liouville-Green transformation to obtain solutions of
d2y/dx2 = xny
for x > 0 , where n is a positive constant, in terms of the modified
Bessel functions. (See Example 13.)
5
Contour integration
5.1 Functions of a complex variable

It will be necessary in this chapter to obtain a number of important
results, the applications of which will be discussed in Chapter 6. We
begin with some elementary ideas and lead up to the main topic of
complex integration in Section 5.6.
A complex number has the form
z=x + \yy (5.1)
where x and y are real and i 2 =— 1, and may be represented
geometrically in the complex (Argand) plane as shown in Figure 5.1.
We shall subsequently consider functions of the complex number z.
One of the many reasons for wanting to do this may be illustrated by
the following example.
Consider the function f(x) = (1 + x 2 ) ~ \ where x is real. The power
series expansion of this function in powers of x only converges for
|x| < 1. This may seem strange since/(x) is finite and well-behaved for
all values of real x. However, in the complex plane, / ( z ) = (1 + z 2 )" 1
diverges as z approaches the points +i and — i which are at unit
distance from the origin. For all z inside the circle \z\ = l , / ( z ) is well
behaved. The radius of convergence of the power series expansion in z
o f / ( z ) is therefore unity. Hence, when z=x (real), the radius of
convergence is also unity. The behaviour of f(x) is more clearly
understood by examining that of / ( z ) in the complex plane.
From Figure 5.1, an alternative form for z is obtained by writing
x = rcos6, y = rsin6, (5.2)
98
5.1 Functions of a complex variable 99
where 0 ^ 6 < 2n or any other interval of width 2JT which includes

6 = 0, for example —JT<6^JZ. Using Euler's formula, we have
z = r(cos 6 + i sin 6) = re i(0+2n7r) , (5.3)
where n = 0, ±1, ±2, . . . , since each time we change 6 by 2JT we
return to the original point. In (5.1), x and y are called the real and
imaginary parts of z and are denoted Rez and Imz, respectively;
r = \z\ = V(^2 + y2) is the modulus (or magnitude) of z and 6 is the
phase or argument of z, denoted by arg z. We note here that the set of
points x, y satisfying, say, | z | = 4 form a circle, centre the origin,
radius 4, whilst those points x, y satisfying, say, \z — 2| = 1 form a
circle, centre (2,0), radius 1.
The complex conjugate of z is denoted by z* (or sometimes by z)
and is defined by changing the sign of i in (5.1) to give
z* =x — \y. (5.4)
Consequently, using (5.1) and (5.4),
1
x = $(z + z*), y=-(z-z*). (5.5)
We define a function of z,f{z)> to have the form

(o = f(z) = u(x, y) + iv(x, y), (5.6)
where u{x> y), v(x, y) are real functions. In general we have the
modulus of (o given by
M = |/(z)| = V(a2 + i>2). (5.7)
Furthermore, a> is single-valued if to each value of z (in some region
of the complex plane) there is precisely one value of w. Consider the
following standard examples:
co = z2 = (x + iy)2 = (x2 - y2) 4- \{2xy) (5.8)
Figure 5.1
100 Contour integration
or equivalently
(5.9)
is a single-valued function of z. On the other hand, co = y/z (z ^ 0 ) is

not a single-valued function since, writing z = r e l ( 0 + 2 "" r ) , we have
co = yz = (5.10)
The values n = 0 and 1 give two distinct solutions
= Vr e i0/2 = cos(0/2) + i sin(0/2)] (5.11)

and
(o2 = V7
i(0/2+jr)
7e = cos(0/2 + JT) + i sin(0/2 + (5.12)
These two functions are called the two branches of co = y/z, and
a) = y/z is termed a multi-valued function. If we give r and 0 a
succession of different values then o)x traces out some curve in the
complex plane; a)2 is obtained by rotating the curve through n about
the origin according to (5.11) and (5.12) (see Figure 5.2). Two values
of \lz exist for all z except z = 0, which is called the branch point. It is
important to realize that the two curves in Figure 5.2 are not the
graphs of co = \jz but only the paths along which z moves as we
change r and 6 (or equivalently x and y). By analogy with the above,
co = z^ will have three branches with a branch point at z = 0, while
(o = y/[(z — a){z — b)] will have two branches with two branch points
at z = a and z = b. We shall expand the discussion of branch points
and multi-valued functions in Chapter 6.
Figure 5.2
- • • * •
5.2 Exponential and logarithmic functions 101
5.2 Exponential and logarithmic functions

We define
using Euler's formula. This definition satisfies the obvious requirement

that ez should become e* when z is real. Further
ez = l + z + ^ + ^ + . . . . (5.14)
From (5.13) when x = 0,

eiy = cos >> + i sin y (5.15)
and consequently
e~iy = cos y — i sin y. (5.16)
It follows from (5.15) and (5.16) that
cosy = {(zxy + e-xy)y (5.17)
siny = i ( e * - e - * ) , (5.18)
where y is a real variable. We may now extend these formulae to

define the trigonometric functions of a complex variable z by
cosz = Keiz + e- iz ), (5.19)
sinz = ^ ( e i z - e - i z ) . (5.20)
From (5.19) and (5.20), we can immediately define tanz, secz,

cosecz, cotz as in the case of real variable theory. Similarly, the
hyperbolic functions cosh z and sinh z can be defined as
cosh z = i(e 2 + e" z ), (5.21)
sinhz = ^ ( e z - e ~ z ) , (5.22)
which reduce to the usual definitions when z is real. The other
hyperbolic functions are defined in an obvious way. It is readily
seen that
sin2z + cos2z = l, (5.23)
cosh2 z - sinh2 z = l, (5.24)
with the other identities of real variable theory remaining true when
the real variable x is replaced by the complex variable z.
Closely related to the exponential function e2 is the logarithmic
function In z. If we write
z = eM (5.25)
then
w = \nz = \n(rê+2njt)), (5.26)
from (5.3). Hence
a) = In r + i(0 + Inn), (5.27)
where n = 0, ±1, ±2, . . . , and we see that co = In z is a multivalued
function. Writing co = u + iv, we have
u = \nr, v = d + 2nji (5.28)
for the real and imaginary parts of co. When n = 0 and —Jt<6^jz, co
takes its principal value and co = In z is then a single-valued function.
Since r = \z\ and 6 = arg z, we can write
a> = In \z\ + i(arg z + 2AZJT). (5.29)
Sometimes, when the principal value of co is taken, we write w = Lnz,
the capital letter L meaning that n has been put equal to zero in
(5.29). We now give examples and include some elementary results.
Example 1
(a) ln(-2) = In | - 2 | + i(arg(-2) + Inn). (5.30)
Hence, since —2 = 2e™, |—2| = 2 and arg(—2) = ny and therefore
ln(-2) = In 2 + i(jr 4- 2AZJT). (5.31)
(b) Ln(-i) = In | - i | + i arg(-i). (5.32)
72
Hence, since —i = le"™ , |—i| = 1 and arg(—i) = — jr/2, and therefore
Ln(-i) = In 1 + i(-jr/2) = -jri/2. (5.33)
(Similarly Ln(i) = jri/2).
(c) Complex powers a of z can be established using the relation
za = ealnz, (5.34)
where In z is defined by (5.27). For example the principal value of i1 is
given by
ii = eiLni = ei(.i/2) = e - ^ (535)
which (surprisingly perhaps) is real. ^

5.3 The derivative of a complex function 103
Example 2 To determine all possible values of

a) = sin~lz (5.36)
when z = 2. (We note that for real variables sin"1 2 makes no sense
since the sine function is bounded by ±1.) Now
2 s iin f t , ^ ^ (5.37)
Hence
e 2 i w -2ie i f t > z-l = 0. (5.38)
lft
This is a quadratic equation for the quantity e \ so that
eia> = ^[2iz ± V(~4z 2 + 4)] = iz± V(l - z 2 ). (5.39)
Taking the logarithm of (5.39), we find
\o) = ln[iz ± V(l - z 2 )]. (5.40)
Now, putting z = 2,
ia) = ln[(2±V3)i]. (5.41)
Hence
co = - Tln(2 ± V3) + i ( | + 2nn\\ (5.42)
since the points (2 ± V3)i have arguments JT/2. Finally then
w = | + 2rtjr - i ln(2 ± V3). (5.43)
Since(2-V3)(2
ln(2 - V3) + ln(2 + V3) = 0 (5.44)
and therefore (5.43) may be written
^ î (5.45)
5.3 The derivative of a complex function

We say that a function / ( z ) is analytic at a point z0 if it has a unique
derivative at that point. By comparison with the real variable
definition of a derivative, we now write
(5.46)
oz
Equivalently, we have
(5.47)
Z — Zo
If in addition/(z 0 ) has a single finite value, and/'(z 0 ) is unique, then

the function is said to be regular at z = z0. In (5.46) and (5.47), the
limit must be independent of the path along which <5z—>0 (that is,
z-*z0). Consider Figure 5.3 and assume / ( z ) is defined in some
(shaded) region R. Then if z = z0 + 6z is some point in this region it
can be made to approach z0 along any one of the infinite number of
paths joining z to z0 (Cly C2, and C3 are three such paths). If f'(z0) is
to have a unique value, then (5.46) and (5.47) must be independent of
the way z approaches z0 (or dz—>0). We note that in real variable
theory we do not have the freedom that complex variable theory
produces for, if x0 is some point on the jc-axis, a neighbouring point x
may approach x0 in only two possible ways - from the left and from
the right. The requirement that the derivative be independent of the
path in the Argand plane restricts the type of functions which possess a
derivative. On the assumption that/(z) does have a unique derivative
at z0, it follows from (5.47) that it must be continuous at z0 in the
sense that
/(zo+(5z)->/(zo) as (5.48)
along any path. Functions which do not satisfy (5.48) are discon-
tinuous at z0; z0 is then said to be a singularity of the function. We
shall discuss this concept in more detail later.
Figure 5.3
5.4 The Cauchy-Riemann equations

We now obtain the conditions under which a function
w = f(z) = u(x> y) + ">(*, y) (5.49)
has a unique derivative as given by (5.46) or (5.47). Writing
8z = dx + idy and z0 = x0 + ry0 then
f(z0 + <5z) - ~ u(x0, y0)]
6y) - v(xQ, y0)] (5.50)
(5.51)
Hence
/ (z0) = hm = hm
hm (5.52)
do oz
dx-+o \ ox + \oy
If the limit is to be path-independent (as required), we can take the
limit in two particular ways: (i) allowing 8x—>0 and then Sy-^0 (path
C1 in Figure 5.4), and (ii) allowing 8y —»0 followed by <5JC-^0 (path C2
in Figure 5.4). We require that the two results should be identical. As
dy^>0, we have from (5.52)
dv
f'(zo)= lim (5.53)
60 8x
whereas when 8x -^ 0 we have
idv dv . du
(5.54)
idy dy dy
For (5.53) and (5.54) to be identical, we require
du .dv dv . du
(5.55)
dx dx dy dy '
Figure 5.4
it
2, + dz
c,
c, c2
2,, c2
o
or equating real and imaginary parts (remembering that u and v are

real functions), we find
3u/dx = dv/dy, du/dy = -dv/3x. (5.56)
The equations (5.56) are called the Cauchy-Riemann equations and
are the necessary conditions for the limit expression for the derivative
to have a unique value. However, these conditions are not sufficient
since we have considered only two possible paths out of the infinity of
possible paths joining z0 + <5z to z0. It can be proved (although we
shall not do this) that provided the first-order partial derivatives
du/dx, dv/dx, du/dy, dv/dy are continuous at z0 then (5.56) are
necessary and sufficient for the limit expression to be path-
independent and for the derivative to be unique. We now give some
examples of the use of the Cauchy-Riemann equations.
Example 3 The function

CO = Z 2 = ( x + [yy = ^ 2 _ y2) + 2bcy (5 57)
is analytic since u = x2 — y2, v = 2xy, and

du/dx = 2x = dv/dy, (5.58)
du/dy = -2y = -dv/dx. (5.59)
The Cauchy-Riemann equations are satisfied and the first derivatives
are continuous. Hence
da) du dv
— = — + i — = 2x + i2y=2z.
y v(5.60)
dz dx dx '
This is precisely the result which would have been obtained by
differentiating co = z2 assuming co and z were real. In fact, the usual
rules of differentiation are still valid when we are dealing with complex
analytic functions. For example, if G = F(a>) and w = / ( z ) then
f-ff. ^ (5.61)
dz dco dz
Example 4 Consider
(0 — e2 = ex+iy = e*(cos y + i sin y). (5.62)
Then
u=ex cos y, v = e* sin y (5.63)
and
du/dx = e* cosy = dv/dy, (5.64)
x
du/dy = —Q siny = —dv/dx. (5.65)
Again the Cauchy-Riemann equations are satisfied and the derivatives

in (5.64) and (5.65) are continuous. Hence co = ez is an analytic
function and
dco du . dv ,
—- = — + I — = e cos y + le* sin y (5.66)
dz dx dx
+
= e* * = e*. (5.67)
Consequently, as expected,
£(e') = e2. ^ (5.68)
Example 5 Consider
o) = l/z. (5.69)
Then
= == =
0) ; ^ ~ ^ ^— 1 ^ r. (5./U)
2 2
x + iy x +y x2 + y2 x2-\-y2 v
Hence
and
du j?->? dv
2 2 2 J
dx {x + y ) dy' "
P j
3y (x + / ) 5JC'
Therefore, the Cauchy-Riemann equations are satisfied except when

x=y = 0 (since the derivatives in (5.72) and (5.73) are not defined
there). Consequently, co = 1/z is an analytic function everywhere
except at z = 0, this point being termed a singularity of the
function. ^
Example 6 The last example illustrated a function which is analytic

everywhere except at one point. Some functions are not analytic
anywhere. Consider
(o = z* =x -iy. (5.74)
Then u—x and v = — y, so
du/dx = l, dvldy = -ly du/dy=0, dv/dx = 0. (5.75)
Consequently, du/dx^ dv/dy and the function is not analytic for any
x and y.
Similarly, consider
co = \z\2 = zz* = (x + iy)(x — iy) =x2 + y2. (5.76)
Hence u =x2 -\-y2 and v = 0 and the Cauchy-Riemann equations are

therefore not satisfied except at x = y = 0. These results clearly imply
that the limit in the expression for the derivative depends on the path
along which z-*z0 (or <5z—»0). To find the derivative of co = \z\2 (see
(5.47)) we need the limit of
|z| 2 -lzol 2 _zz*-z o zo*
z -z0 z - z0
= z , + z 0 z*-z 0 z 0 * (5?8)
z — z0
*: 7-*\
(5 79)
v i^tr -
H e n c e , if z — z 0 = re 1 0 ,
Z ZQ Z ZQ A C
and
\z\2~\z,\2
= z* + z0[cos(2^) - i sin(20)]. (5.81)
z-z0
As z^z0, the limit depends on 6, the angle at which z approaches
z0. ^
5.5 Derivatives of multi-valued functions

When a function is not single-valued (for example \lz and In z) then
the derivative may be obtained only on a particular branch (see
Section 5.2). If, for example, co = lnz = lnr + i(0 + 2njr), then by
taking a particular value of n (say n = 0) we have a single-valued
function z = ew. For this branch
= e(° (5.82)
which implies
da)/dz = e-M = l/z, (5.83)
except at z = 0 (In z is not analytic at z = 0).
5.6 Complex integration

We first indicate the meaning to be attached to J c / ( z ) dz, where C is
some path in the complex plane with end points A(zx) and B(z2) (see
Figure 5.5). Since f(z) may be written as /(z) = u(x, y) + iv(x, y) and
dz = dx + i dyy then f(z) dz = (u + iu)(dx + i dy) and
f /(z) dz = \ (u dx - v dy) + i [ (v dx + u dy). (5.84)

Jc Jc Jc
Each integral on the right of (5.84) is a real line integral and may be
evaluated by integrating from A to B along the specified path C (the
equation of which will, in general, be given as y = y(x), say). Since the
complex integral can be put in the form (5.84), the usual rules of
integration which apply to real integrals must also apply to complex
ones. Accordingly, we have
(i) f [/(z) + g(z)] dz = f f(z) dz + \ g(z) dz, (5.85)

Jc Jc Jc
where / and g are any two integrable functions;
(ii) f Kf(z)dz = K[ f(z)dz, (5.86)

Jc Jc
where K is an arbitrary complex constant;
("0 f f{z)dz=\ f{z)dz+\ f(z)dz, (5.87)

JAEB JAE JEB
where E is some point on the curve C lying between A and B

(see Figure 5.5).
An important result follows if / ( z ) is analytic in a region R
Figure 5.5
B(z2)
- • *
O
containing the points A(zx) and B(z2). In this case we can show that
/'(z) dz = [/(z)E =/(z2) -f(Zl), (5.88)
which implies that the integral is independent of the path C since the
result depends only on the values of / ( z ) at the two end points.
Consider
(5 89)
-
f
using the form for f (z) given in (5.53). Then
Now, since f(z) is assumed to be analytic, the Cauchy-Riemann

equations
du/dx = dv/dy, du/dy = -dv/dx (5.91)
hold. Consequently the right-hand side of (5.90) becomes
= j 2 d u + i[2dv, (5.93)
since the integrands in (5.92) are the total derivatives of u and v.

Finally, (5.93) integrates to give
Jcf'(z)dz = [u]% + i[v]%=[u + iv]Z
= [/(z)E=/fe)-/(*i). (5-94)
This proves the result that the integral is independent of the path. The
following examples illustrate these ideas.
Example 7 Evaluate
[ z2dz, (5.95)
Jc
where
(i) C = Cx is the straight line y = x from A(0, 0) to B(l, 1),
(ii) C = C2 is the x-axis from (0,0) to A(l, 0) and the straight line
from A(l, 0) to B(l, 1),
(see Figure 5.6).
I z2dz= (x2 -y2 + 2ixy)(dx + i

JC\ JC\
=J [(x2-y2)dx-2xydy]
cx
+ i I [2xy dx + (x2 - y2) dy]. (5.96)
Now on Cly y = x. Inserting this into (5.96) we find (using dy = dx)
I z2dz= I [(x2-x2)dx-2x2dx]
Jcx Jo
+i f [2JC2rfjc-h (x2 - x2) dx] (5.97)
= (i - 1) I 2x2dx = i ( - l + i). (5.98)

Jo
(ii) f z2dz=\ [(x2-y2)dx-2xydy]

Jc2 Jc2
-h i f [2xydx + (x2 - y2) dy], (5.99)

Jc2
as in (5.96). Splitting this into two parts, one for each section of the
path C2, we have
f z2dz=i \{x2-y2)dx-2xydy]
Jc2 J(o,o)
[2xy dx + (JC2 - y2) dy]

(0,0)
+r i) [(^->' 2 )^-^^]
J(l,0)
r(l,l)
J (1,0)
[2xy dx + (x2 - y2) dy], (5.100)
Figure 5.6
jt
C2
0(0,0) C2
where the integrals are along the straight lines joining the end points.
Along OA, y = 0 and hence dy =0. The first two integrals in (5.100)
therefore give
f xx22dx
dx ++ i\i I 0dx = l (5.101)
Jo Jo
Along AB, x = 1 and dx = 0 and the second pair of integrals in (5.100)
gives
f -2ydy + i f (1 -y2) dy = - 1 4- fi. (5.102)

Jo Jo
Hence
z2dz = i - 1 -h |i = §(-1 -h i). (5.103)
The results (5.98) and (5.103) are the same. This is to be expected
since z 3 /3 (the function which differentiates to give z2) is an
analytic function and hence the value of the integral is \{\ 4- i)3 =
5.7 Cauchy's Theorem (First Integral Theorem)

This may be stated as follows: If /(z) is single-valued and analytic
throughout a simply connected region R (that is, a region without
holes) then, if C is any closed path within R,
if(z)dz = 0, (5.104)
where the symbol ^ c denotes integration around a simple (non-

intersecting) closed curve C (see Figure 5.7). To prove this we refer to
Figure 5.7
-•*
5.7 Cauchy's theorem {first integral theorem) 113
(5.84) which states that
4 f(z) dz = j) (udx-vdy) + i<b (vdx + u dy). (5.105)
Now if, in general, P(x, y) and Q(x, y) are two continuous functions
with continuous first partial derivatives (at least), then
j^Pdx + Qdy)= -ff (f^-f^) dxdy, (5.106)

where D is the region enclosed by the curve C. This result (called
Green's Theorem) may be used to transform the right-hand side of
(5.105). Accordingly, we find
f ((fdv 9u\ ff/du dv\
jf(z) dz = - (- + -) * *
J
+ i\ \ - - - J
dx dy. (5.107)
D D
Now, since f(z) is analytic, the Cauchy-Riemann equations are

satisfied and each integrand on the right-hand side of (5.107) is zero.
Hence Cauchy's Theorem (5.104) holds as stated.
Example 8 Evaluate
I=lzzzdz, (5.108)
where C is a unit circle, centred at the origin.

Since ze 2 is an analytic function (as may be checked by verifying
that it satisfies the Cauchy-Riemann equations), 7 = 0 by Cauchy's
Theorem. Â
Example 9 Consider
(5.109)
— Z
where Cx is the circle \z\ = 1 (see Figure 5.8).

Since l/(z — 2) is analytic everywhere except at z = 2 (which is
outside the circle \z\ = 1), Il = 0by Cauchy's Theorem.
Now consider
h=i -^r, (5.110)

Jc2 z — z
where C2 is a circle, radius p, centred at z = 2 (see Figure 5.9).
Writing
(5.111)
where 0 ^ 6 ^ 2JT, we find using dz = piew d6y

p\eedG
] (5.112)
c2 pe°
The non-zero result arises because l/(z—2) is not analytic at z=2
and this point lies inside the circle C 2 , thus violating the conditions of
Cauchy's Theorem. ^
Example 10 The integral
I=4>zndz, (5.113)
where n is a positive integer, is zero by Cauchy's Theorem since zn is

analytic in the whole finite plane. Consider the circle, radius a, with
equation |z|=a. Then / can be evaluated directly using polar
Figure 5.9
5.7 Cauchy's theorem (first integral theorem) 115
coordinates (as in Example 9). Putting z = aeld, we have

n ind
1=1 a e aeiei dO = an+1i f V» + 1 >* dO
f2jz
= an+1i [cos(n + 1)0+ isin(n +1)6] dO = 0, (5.114)

Jo
since both the cosine and sine terms integrate to zero, ^d
We now discuss an important result which follows from Cauchy's

Theorem. Consider two closed curves Cx and C2 in the complex plane
and a function f(z) which is analytic on the curves and in the region
between them. We form a single closed curve by breaking Cx and C2
and joining them by two straight lines AB and EF (see Figure 5.10).
Then, by Cauchy's Theorem,
f (5.115)
since f(z) is analytic on and inside the closed curve where Cx is

described anticlockwise and C2 clockwise. Now the two straight lines
can be made as close as we please and are integrated along in opposite
directions. In the limit where these lines coincide their contributions to
(5.115) cancel and we have
f(z) dz+ i f(z) dz = 0. (5.116)
Hence
f(z) dz = -{ f(z) dz = l f(z) dz. (5.117)
Figure 5.10
-•*
o
The integral of / ( z ) around any closed path between Cx and C2 is

therefore independent of the path.
5.8 Cauchy Integral Formula (Second Integral Theorem)

Suppose we now have a function of the form
g(z)=f(z)/(z-zo)y (5.118)
where/(z) is assumed to be analytic within a region R and z0 lies in R.
Assuming /(z 0 ) =£ 0, then g(z) is not analytic at z = z0. An example of
this type of function was considered in Example 9 where we integrated
the function l/(z - 2) (which is of the form (5.118) with / ( z ) = 1 and
z0 = 2). The Second Cauchy Integral Theorem states that
£^-dz, (5.119)
Jc z - z0
where C is any closed curve which lies in R and includes the point z0.
To prove this, we surround the point z0 by a small circle Cx, radius a.
By the argument at the end of Section 5.7, we have
f f(z) f f(z)
4> jL±-L dz = 4) - ^ ^ - dz, (5.120)
z
Jc — z0 Jcx z — z 0
since the whole function f(z)/(z — z0) is analytic between C and Cx.
Now let z = z0 4- aeld describe the circle Cx so that on the right-hand
side of (5.120), dz = iaewd6 and z-zo = aeie. Then
IS L
4 - dz=\ f{zo + aee)—td8, (5.121)
J Cl z - z0 Jo ae
which, as the small circle Cx shrinks to zero, becomes
/(z 0 ) i dd = 2jri/(z0). (5.122)
We have used the fact that, since / ( z ) is a continuous function,

f{z)-*f{zQ) as z->z0. Hence, using (5.120) and (5.122), we have
proved the result (5.119). The remarkable property of analytic
functions which follows from this result is that the value of / ( z ) at any
point inside a closed contour is completely determined by its values on
the contour.
Example 11 Evaluate
I=f ~^dz, (5.123)

Jc z — 1
where C is the circle \z — 2\ = 2.
5.8 Cauchy integral formula (second integral theorem) 117
Clearly the singular point of the integrand is at z = 1 which lies

within C (see Figure 5.11). Furthermore /(z) = ez is analytic for any
finite point z in the plane. Accordingly, by the Cauchy Integral
Formula,
1= dz = 2mf(z{) = 1) = 2jrie. (5.124)
Jc z — 1
Example 12 Evaluate
2z-l
/= -dz9
where (i) C = Cx is the circle \z\ = 4 and (ii) C = C 2 is the circle \z\ = \
(see Figure 5.12).
Figure 5.11
We first use partial fractions to write the integrand as
2z - 1 _ 1 1
+ (5.126,
(z + l ) ( z - 2 ) ~ z + l z-2'
Then
Z
I _2\dz==j) ~ ^ 7 + ^ ~Zy' (5.127)
(i) For Clf both singular points z = —1 and z = 2 lie within Cx.
Hence, by the Cauchy Integral Formula,
r 2jri.l. (5.128)
v 7
l 2 z-2
Hence
/ = 2jri(l + l) = 4jri. (5.129)
(ii) For C 2 , only the singular point z = — 1 lies within C 2 . Hence the
first integral is
^ ^ 1. (5.130)
Jc2 z + 1
Since l/(z — 2) is analytic within C 2 , then by Cauchy's Theorem
• = 0. (5.131)
c2 z ~ z
Hence
^ (5.132)
5.9 Derivatives of an analytic function

We have from (5.46) with 8z = h
f'(z0) = lim — —. (5.133)
Now, by (5.119),
1 r f(7\
(5.134)
Jcz-zo
and
<5 135)
-
5.9 Derivatives of an analytic function 119
Substituting (5.134) and (5.135) into (5.133), we have
dz (5.136)
z -zo-h z - zj h
= lim - ^ i ~ ^ 7 7 dz- ( 5 - 137 )

hôzjti Jc (z — z0 — h)(z — ZQ)
Reversing the order of the limit and the integration, we have
<5 138)
^?*- -
In a similar way, we can obtain the general result
where n is a positive integer and/ ( n ) denotes the nth derivative. Since

the analytic function f(z) possesses derivatives f'(zo),f"(zo), .. . ,
these derivatives must also be analytic functions. Further, if f(z) =
u+iv then the existence of/" implies that / ' is differentiate and
hence that du/dx and dv/dy may be differentiated. From the
Cauchy-Riemann equations (5.56) we then have
d2u/dx2 = d2v/dx dy = -32uldy2 (5.140)
so that
V2w=!^ + ^ = 0. (5.141)
Similarly
d2v 32v
V 2 u=—o2 + ^ = 0. (5.142)
dx dy2 v
'
Example 13 Evaluate
Q
I=j> _U2dz> (5.143)
where C is the circle \z\ — 2.

Here the singular point of the integrand is at z = 1, and f(z) = e2z.
Comparing with (5.138), we see that
C
<f> 2 dz = 2mf'(z0 = 1) = 2;ri . 2e2 = 4;rie2. ^ (5.144)
Jc {z — 1)
5*10 Taylor and Laurent series

We now state, without proof, two important theorems.
1. Taylor's Theorem
If /(z) is analytic within and on a circle C of radius R and centre
z0, then
Kz ~ zzo)
o) • (5.145)
n=0
This series converges within the circle, centre z0, of radius R equal to
the distance from z0 to the nearest singularity of /(z). In general,
/ (n) (z 0 ) can be found from (5.139).
Example 14 To find the Taylor series for /(z) — sin z about z0 = 0, we

first evaluate the derivatives
/'(z) = cosz, /'( (5.146)
/"(z) = -sinz, / (5.147)
f'"(z) = -cos z, (5.148)
and so on. Now from (5.145)-(5.148),
sin z = - n\ 3! 5! (2n +1)!

0
(5.149)
which is the familiar series for the sine function. ^
Example 15 The function 1/z2 is not analytic in a region containing

the origin z = 0 . However, inside the circle \z — 1| = 1 (see Figure
5.13) the function is analytic and may therefore be expanded as a
Figure 5.13
>• x
Taylor series about z = 1. By differentiation,
/<">(*) = ( - 1 ) " ^ ^ (5.150)

and hence
/(">(l) = ( - i y ( « + l)! (5.151)
Accordingly the Taylor series (5.145) gives
\= 2 (-l)"^-^(z - 1)" = i (-l)"(n + l)(z - 1)". (5.152)

Z n=0 «! n =o
For convergence, | z - l | < l since the distance from the centre of

expansion (z() = 1) to the nearest singularity of 1/z2 (at z = 0) is
unity. ^
Before stating the second theorem, we first explain what is meant by

an isolated singularity. Consider a single-valued function /(z) with a
singularity at z = z0. We surround the point z = z0 by a circle centred
on z0. If the function is analytic everywhere within some sufficiently
small circle, except at the single point z = z0, then z0 is termed an
isolated singularity. For example, /(z) = 1/z2 has an isolated sin-
gularity at z = 0 since within an arbitrarily small circle enclosing this
point the function is analytic except at z = 0. However, consider the
function /(z) = cosec(l/z) = l/sin(l/z). This function has singularities
at z = 0 and at z = linn, n = ±1, ±2, . . . . If we move along the jc-axis
towards the origin, we encounter an ever-increasing density of
singularities, that is, the singularities accumulate onto the point z =0.
Each of the singularities at l/nn is isolated since an infinitesimally
small circle can be placed around each one so that the circle contains
only one singular point. However, the singularity at z = 0 is not
isolated since any circle, however small, centred on z = 0 will contain
an infinite number of singular points. For multi-valued functions such
as \lz and In z, the concept of an isolated singularity is not applicable.
2. Laurent's Theorem
This is an extension of the preceding theorem and states the following:
If/(z) is analytic inside and on the boundaries of the circular annulus
defined by \z — zo\=Rr and \z — zo\ = R2 with R2<R\ (see Figure
5.14), then/(z) can be represented by the Laurent series
f(z) = i an{z - zoy + S -^f- , (5.153)

n=0 n = \ \Z — ZQ)
where
dz (5.154)
and
(5.155)
In the case when z0 is an isolated singularity of/(z), the series (5.153)

consists of a Taylor series (the an series, convergent for \z — zo\ <Ri),
and the a_n series which contains terms which are singular at z0. This
latter series, called the principal part of the Laurent series and
denoted by F(z), converges for all \z - zo| > 0 (that is, R2 may be
taken to be zero).
Example 16 Consider
(5.156)
which has an isolated singularity at z = 0. Then since
£; zn
e - + z + —+ . . . + — + . . . , (5.157)
we have, for all z ¥= 0,

, . . 1 1 1 1 1 z
(5.158)
This is the Laurent series expansion (5.153) about z = 0. The principal

part of the series is
„. . 1 1 1 1
(5.159)
which is a finite series.
Figure 5.14
Example 17 Not all principal parts of Laurent series are finite series.
Consider, for example,
( ^ ^ ^ ) (5.160)
for all z =£ 0. This function has
P(z) = — + - - ^ + . . . + f w_1 + ..., (5.162)
which is an infinite series. ^
Example 18 Both the previous examples had singularities at z = 0.

Suppose now
/ ( z ) = e 2 /(z - I) 2 . (5.163)
Then / ( z ) is singular at z = 1. Hence let z — 1 = u> say, so that
e2 e1+w e ' -2 -3
(5.164)
giving
which is the Laurent series expansion of (5.163) about z = 1. The

principal part is
(5 166)
[(^(îj] ^ -
Example 19 Consider
'W=;<IVFV? < 5167 >

The singularities of / ( z ) are at z = 0 and z = 1. Hence there are two
regions in which the function is analytic and for which we can write a
Laurent series in powers of z: (i) 0 < | z | < l and (ii) K | z | . We
consider these regions separately,
(i) F o r O < | z | < l ,
/(z) = - ^ - ( l - z ) - \ (5.168)
Now, since 0 < | z | < l , the second term can be expanded by the
Binomial Theorem to give
- - ( l + z + z2 + z3 + . . . ) • (5.169)
(ii) For | z | > l ,
Again, since \z\ > 1, the second term can be expanded by the Binomial
Theorem giving
=
? + ? + 7 + -'-- (5 172)
-
In the above we have chosen z0 = 0 as the centre of expansion. If we
choose instead zo=l, then we obtain two series in powers of z — 1, one
for 0 < \z — 1| < 1 and one for \z — 1| > 1. The first of these is obtained
by putting z — 1 = u so that
X
^ {l-u + u2-...)y (5.173)
z(z - 1 ) u 1+w u
since \u\ < 1, giving
-174)
Similarly, we may obtain the series for \u\ > 1. ^
Example 20 Consider
(^) (5.175)
for |z| > 1 (recalling that the capital letter indicates the principal value
of the logarithm). Then
Hence
l/z 2 ). (5.177)
The principal value of Ln(—1) is \n while the second term can be

expanded using the series for Ln(l - 1/z2) since |l/z 2 | < 1. We find
z2 2z 4 3z e
1 1 1
(5-178)
5.11 Singularities and residues

We have seen in Section 5.10 that if / ( z ) has an isolated singularity at
z = z0 then in the neighbourhood of this point / ( z ) can be represented
by the Laurent series (5.153). The principal part of the Laurent series
(the terms with negative powers of z - z()) may contain either a finite
number of terms (as in Example 16 above) or an infinite number of
terms (as in Example 17 above). We have the following definitions in
each case:
Finite number of terms. In this case the singularity at z() is called a

pole of order m, where m is the highest negative power of z — z0.
Hence, in Example 16, / ( z ) has a pole of order 4 at z = 0 whilst in
Example 18, / ( z ) has a pole of order 2 at z = 1. Poles of order 1 are
called simple poles. Hence, in Example 19, / ( z ) has a simple pole at
z = 0 (see the Laurent series (5.169) about z = 0 for 0 < \z\ < 1) and a
simple pole at z = 1 (see the Laurent series (5.174) about z = 1 for
Infinite number of terms. When the Laurent series in the neighbour-

hood of a singular point contains infinitely many terms with negative
powers of z — z0, the point z = z0 is termed an essential singularity.
For example
/ ( z ) = e' /2 = l + ^ + ^ + ... (5.179)
has an infinity of terms in negative powers of z. Hence z = 0 is an

essential singularity of e1/z. Similarly
r3i^+5i^-" (5 180)
-
has an essential singularity at z = 0.
An important number, as we shall see in Section 5.12, is the

coefficient a_x in the Laurent series. This coefficient is called the
residue of the function / ( z ) at z = z0 and we now show how it may be
calculated in different cases.
1. Simple poles
Clearly if we have a simple pole at z = z0, then the Laurent series
(5.153) will have the form
f(z) = - ^ - +ian(z- z0)". (5.181)

Z — ZQ n=Q
By multiplying through by z — z0, we have

(z - zo)f(z) = a_x + ao(z - z0) + ax(z - z0)2 + . . . . (5.182)
Hence taking the limit as z-^>z0, we find
\im[(z-zo)f(z)] = a_u (5.183)
which enables the residue to be calculated.
Example 21 To find the residues at z = 0 and z = 1 of

/(z) = l / z ( z - l ) . (5.184)
Method 1 In Example 19 above we showed that the Laurent series of

f(z) in the neighbourhood of z = 0 is (see (5.169))
f(z) = 1-z-z2- (5.185)
The coefficient of 1/z is by definition the residue at z = 0 and is

therefore — 1. Further we showed (see (5.174)) that the Laurent series
o f / ( z ) in the neighbourhood of z = 1 is
/ ( z ) = ^ - l +(z-l)-(z-l)2 +.... (5.186)
Hence the residue at z = 1 (the coefficient of l/(z - 1)) is 1.
Method 2 We now use the formula (5.183). For z = 0,
\ \ -l, (5.187)
z-*oL z(z — 1)J
as before.
For z = 1,
a
-^diz-l)^hj]=l' (5 188)
-
as above. ^
Example 22 To show that z = 3JT/2 is a simple pole of

2z
/ ( z ) = e tanz, (5.189)
and to determine the residue there.
Let z - 3JT/2 = u. Then expressing (5.189) in terms of u we have
f(u) = e{u+37tl2) tan(i* + 3JT/2)
cos(w + 3JT/2)
sin
We now expand the functions cosw and sinw as power series in u
so that
2! 4!
We may further expand e2" as a power series and use the Binomial
Theorem to expand the last factor. Then
x ^ l + - - . . . j. (5.192)
Since all the expansions involve only positive (or zero) powers of u>
the only term with a negative power of u is —e3jr/w. Hence the
principal part of the Laurent series of / ( z ) in the neighbourhood of
z = 3JT/2 is
(5.193)
The pole is therefore a simple one. The residue is the coefficient of

l/(z - 3JT/2), which is -e 3 j r .
Alternatively, the residue could be calculated using (5.183). We
obtain
fl_x= lim [(z - 3;r/2)e2z tan z]
Z-*3JT/2
= lim [ ( z - 3 j r / 2 ) e 2 z ~ l . (5.194)
Z^>3JT/2L COSZJ
Now both z — 3JT/2 and cos z tend to zero as z—> 3;r/2. Hence
tf _i = lim ( )( lim e 2z )( lim sinz)

Z->3JI/2\ COSZ J\Z^3JZI2 / \Z->3JT/2 /
= lim (—!— W*)(-l), (5.195)

z->3n/2\— Sin Z/
where we have used L'HopitaPs rule in the first term. We find

therefore
-e3jT
e 3 ^, (5.196)
-sm(3jr/2)
as above. ^
Besides the two methods described above, the residue at a simple

pole may be calculated by another method. Suppose
m (5 197)
'W) -
and that z = z0 is a simple zero of g(z) (that is, g(z) behaves like z — z0
as Z ^ ' Z Q ) . Then z = z0 is a simple pole of / ( z ) provided p(z 0 ) =£0 and
that p(z) is not singular at z = z0. Using (5.183) we can now write
, (5.198)
since ^(z0) = 0 by definition. Hence
a_x = lim i
z^z 0
z -z0 J
using the limit definition of the derivative of q(z) at z = z0. For a

simple pole, therefore,
a_l=p(z0)/qf(z0). (5.200)
Example 23 Consider
/(z) = jrp- (5-201)
The pole at z = 1 is simple (since z = 1 is a simple zero of
1 - z 3 = (z - 1 ) ( - 1 - z - z 2 ) ) . Hence, since p(z) = l and q(z) =
1-z 3 ,
a-i = , A =-i ^ (5.202)
2. Po/es 0 / order m
The Laurent series (5.153) for a function f(z) with a pole of order m
at z - z0 has the form
Now multiplying (5.203) by (z - zo)m we have
(z - zb)m/(z) = i aB(z - zb)m+" + a_,(z - zb)1""1
+ a_2(z-zor"2 + ...+a_m. (5.204)

To find fl_j, we differentiate (5.204) m — \ times to obtain
^ Z T [(z ~ zb)m/(z)] = (m - 1)! a . ! + m! ao(z - z0) + . . . . (5.205)

Hence
a_! = lim — m_J(z - z o ) m /(z)]. (5.206)
When m = 1 we obtain the result (5.183) for a simple pole.
Example 24 Consider
f(^) = {jZri) • (5.207)

Then z = 1 is a double pole (pole of order 2). We can obtain the
residue at this point by finding the Laurent series, or by using (5.206).
Method 1 Let z - 1 = u. Then

2
/z V O + l) z l + 2w + wz 1 2
T =- ^ - = 5 =- +-+1. (5.208
\z-l/ u u u u
Hence the Laurent series is
1 2
(5.209)
The coefficient of l/(z — 1) is 2 which is therefore the residue at z = 1.
Method 2 Using (5.206) with m = 2,
(5.210)
= lim-f (z2) = 2, (5.211)
as in method 1. ^k
Example 25 Consider
Then / ( z ) has a pole of order 2 at z = 0. Using the expansion of e2, we

find
(5 213)
-?-)+-- -
From the Laurent series (5.213), we see that the residue of / ( z ) at
z = 0 is —1. Alternatively, using (5.206), we have
<5214)
Both the numerator and the denominator of this expression tend to

zero as z —»0. The value of the limit may be obtained by expanding tz
5.12 Cauchy residue theorem 131
as follows:
a_1 = lim (5.216)
-z3-|z4
lim (5.217)
z—•()
= lim =-1. (5.218)
Hence the residue of /(z) at the double pole z = 0 is —1, as

before. ^
5.12 Cauchy Residue Theorem

We have seen that, in the neighbourhood of a singularity z0, f(z) may
be represented by the Laurent series
m
n °°
/(*) = 2 , ~\n +2<*n(z- Z0)", (5.219)
where the singularity is termed a pole of order m for any finite m, and
an essential singularity for m infinite. Suppose we require
=1 f{z)dz, (5.220)
where C is some closed curve surrounding z0. We showed in Section

5.7 that the integral around C is equal to the integral around a circle y
of radius p, centre z0. This was done by cutting C and inserting two
parallel straight lines of equal length joining C to y, the contributions
from these lines (the cut) cancelling as the lines are brought into
coincidence. Hence
f(z) dz = i f(z) dz. (5.221)

J
Using (5.221) and (5.219) we have
f(z) dz = i [ £ T^\n + S an(z - zo)» 1 dz. (5.222)

The integral of each term in the second summation is zero by Cauchy's

Theorem since each term is an analytic function. Now consider a
typical term in the first summation, say,
-dz.
Since y is a circle of radius p (say), then on y, z = zo + pe10, and

(5.223) can be evaluated directly:
If n ^ 1, this integral is simply
-ÊT {cos[(n - 1)0] - i sin[(/i - 1)6]} dd = 0. (5.225)

P Jo
If n = 1, then
1 (5.226)
T y (z - z 0 ) Jo
Accordingly,
i f(z)dz = 2 (5.227)
Jc
From Section 5.11, a_x is the residue of f(z) at the pole of order m at
z = z0. Since we know how to calculate a_x for a given function/(z), it
follows that we can now calculate $cf(z) dz around any closed
contour C surrounding a singularity at z0. The result (5.227), known as
the Cauchy Residue Theorem, may be readily extended to the case
where /(z) has a number of poles z0, zx, z2, . . . , zk within some closed
contour C (the case of three singularities is shown in Figure 5.15).
Figure 5.15
-•*
5.12 Cauchy residue theorem 133
Forming the cuts in the usual way, we see that the integral around C is
just the sum of the integrals around circles centred on each singularity.
Hence
f(z) dz = 2jri(a_ _x 4 - . . . ) , (5.228)
where a_ly b_u c_i, . . . are the residues at z0, zx, z2, . . . . We there-
fore have
n> f(z) dz = 2m x (sum of the residues at (5.229)

c
the poles within C).
Example 26 Consider the function

(5.230)
Then f(z) has two simple poles, one at z =0 and one at z = 1. At
z = 0, using (5.183), the residue is
•--Sl'STl)]- 1 - (5.231)
whilst at z = 1, the residue is
(5.232)
We now evaluate the integral of ez/z(z — 1) around three different

closed curves C,, C2, and C3 (see Figure 5.16). Consider first
-dz, (5.233)
Jc, z(z - 1)
where Cx is some closed curve not containing a pole. Then the
Figure 5.16
function is analytic inside and on Cx and hence by Cauchy's Theorem

the integral is zero. On the other hand, by the Cauchy Residue
Theorem,
I e2
4) — -dz = 2 ; r i ( - l ) , (5.234)
where C2 encloses the singular point z = 0 (at which the residue is - 1 )

but not the singular point z = 1. Finally,
I e2
4> — — dz=2;ri(e-l), (5.235)
Jc,z(z-1)
where C3 encloses both poles at z = 0 and z = 1 at which the residues

are respectively —1 and e. ^
The Cauchy Residue Theorem enables many real definite integrals

to be evaluated without performing any integration. This and other
applications are dealt with in the next chapter.
Problems 5
1. If z=x + \y> show that, if a>0, \az\=ax, where the principal

value of az is taken.
2. Determine all possible values of
(i) l n ( l - i ) , (ii) tan" 1 (2i).
3. Show that the function e2 has no zeros, and locate the zeros of
cosh z.
4. Given that
is the real part of an analytic function / ( z ) , determine the

corresponding imaginary part v and the form of / ( z ) .
5. If/(z) is an analytic function, show that
Use this result to find the most general forms of g(z) and h{z)
given that g(z) and h(z) are analytic for all z, and that
(p = |g(z)| 2 + \h(z)\2 is harmonic (that is, it satisfies Laplace's
equation V20 = 0).
Problems 5 135
6. Evaluate J (z 2 + 1) dz
(i) along the straight line joining the origin, z = 0, to z = 1 + i,
(ii) along the parabola x = t> y = t2 where 0 =^ t ^ 1,
(iii) along the straight line from the origin, z = 0, to z = 1, and
then along the straight line from z = l to z = 1 + i.
Why are the results the same?
7. (i) Evaluate
dz
_ z-2'
where C is (a) the circle \z\ — 1; (b) the circle \z + i| = 4.
(ii) By writing z = e10, evaluate jcLnzdz and J c z 3 L n z d z ,
where C is the curve defined by —JZ<Q<JZ.
8. Evaluate the following integrals using the Cauchy Integral For-
mula, where C is the circle \z\ = 2:
r sinz
(o
9. Evaluate the following integrals using the Cauchy Integral For-
mula, where C is the boundary of a square the sides of which are
defined by x = ±2, y = ±2:
tan(z/2)
• (z - J T / 2 ) 2 Z
'
10. Find the Taylor series about z = 0 and its radius of convergence
for each of the following functions:
(i) e z s i n z , (ii) s i n f - ^ - ) , (iii)
11. Locate the singularities of the following functions and determine

their nature and the appropriate Laurent series about these
points:
cos 2z
r\ • /i/ \
(i) z sin(l/z), r-\
(n) ( ) , /(in)
-x e z s i n z—z.
z—K \z — nil)
12. Find the Laurent series of l/z(z — I) 2 about
(i) z = 0 , 0 < | z | < l ,
(ii) 2 = 0 , |Z|>1,
(iii) 2 = 1, 0 < | z - l | < l ,
(iv) 2 = 1 , | 2 - 1 | > 1 .
13. Find the residues of each function at the specified poles:
(i) at all the poles,
(ii) cot z at z = 3JT,
(Ui)
z\z - l) 2 (z - 2) a t a U t h C P lle S
° *
lz
14. Locate the poles of z/(l — ae~ ), where a > 1 is a real constant,
and hence deduce that
where C is any closed contour in the upper half-plane.

15. Show that
where C is a contour lying in the upper half-plane consisting of the

semicircle of radius R>1, centred at the origin, and the portion
of the jc-axis between —R and + R.
6
Applications of contour
integration
6.1 Introduction
Using the basic theorems of the last chapter, in particular the Cauchy
Residue Theorem (5.229), we may now apply contour integration to
two specific problems: (i) the calculation of real integrals, and (ii) the
summation of infinite series.
6.2 Calculation of real integrals
1. Integrals of the type fl"f(cos 0, sin 6) dd

The integrand /(cos 6, sin 0) is first transformed into a function of
z = e10 by writing
e ie + e- i e z + l/z
cos0 = = —-—, (6.1)
and dz = ieied0 = izdd. The integral from 0 = 0 to 6 = 2n is then

equivalent to integrating around the unit circle C, defined by z = el6>
or \z\ = 1, and may be evaluated using the Cauchy Residue Theorem.
Example 1 Evaluate
dd
137
138 Applications of contour integration
Choosing the unit circle C (as above) for which z = e10, we have
dO = dz/iz. (6.4)
Using (6.1) and (6.4), (6.3) becomes
dz/iz dz
/ = 2 (6.5)
2-\(z + Hz) iJcz -4z
Now the poles of the integrand occur where z2 — 4z + 1 = 0, that is, at
z = 2 ± V3. Both are simple poles since z2 — 4z + 1 = [z — (2 —
V3)][z - (2 + V3)], but only the pole at z = 2 - V3 is within C (see
Figure 6.1). Hence, to evaluate the integral by the Cauchy Residue
Theorem, we require only the residue at z = 2 - V3 which is (by
(5.183))
Jin,^ {[z - (2 - V3)] [ z _ ( 2 _ V 3 ) ] 1 [ z -
. (6.6)
(2-V3)-(2 2V3
Hence
dz
(6.7)
by the Cauchy Residue Theorem (5.229). Finally

. ( _2JT
(6.8)
2 - cos 6 ~ ~ i \ 2V:
Example 2 Evaluate
** cos(2^)
(6.9)
l0 5 - 4 c o s ^ '
Using the fact that
cos(20) = *6 = Z2+ 1/Z2

2 , (6.10)
together with (6.1) and (6.4), / becomes
z2+l/z2 dz If z4+l
rfz, (6.11)
iz 4i Jc z\z - i)(z - 2)
where the contour C is the circle \z\ = 1. The poles of the integrand
are as follows:
(i) a pole of order 2 at z = 0;
(ii) a simple pole at z = \\
(iii) a simple pole at z = 2.
Of these, only the first two lie within C (see Figure 6.2) and hence we
need the residues at these two poles only.
Residue at z =0 Expanding the integrand as a Laurent series about

z = 0, we have
-1
z/2 + z2/4 (6.12)
The coefficient of 1/z is therefore 2 + ^ = §, which is the residue at

z = 0. Alternatively, since z = 0 is a pole of order 2, we could have
used (5.206) with m = 2 to obtain the same result.
Figure 6.2
Residue at z =\ Here the pole is simple and the residue is (by (5.183))
(6.13)
Finally then, summing the two residues within C,

z4+l
dz = (6.14)
Jcz\z-\)(z-2)
and consequently, by (6.11),
[
cos(20) 7T
(6.15)
5 - 4 cos 6 4i 6*
2. Integrals of the type J_oo/(x) dx

We replace x by z and consider the closed contour C in the upper
half-plane consisting of the semicircle C" of radius R, for which
z = Reie, OÔ^JZ, and the part of the x-axis from -R to +R (see
Figure 6.3). We further assume that the single-valued function /(z)
satisfies two conditions:
(i) /(z) is analytic in the upper half-plane except for a finite number
of poles, none of which lie on the real axis (the case of poles on
the real axis is dealt with in Section 6.4);
(ii) |/(z)| ^ MIR2, where M > 0 is some constant, as \z\ = R tends to
infinity on C" (the curved part of C).
Now, by the Cauchy Residue Theorem,
lf{z)dz=\R f(x)dx+f f(z)dz

JC J-R JC
= 2n\ x (sum of residues at the poles of /(z) inside C).

(6.16)
Figure 6.3
Further
f f(x)dx=\imi f(x)dx. (6.17)
Hence, provided
lim [ f(z)dz=0, (6.18)
as we expand the contour by letting /?—»oo, (6.16) gives (since, as

R^>™, the interior of C becomes the upper half-plane)
f(x) dx = 2m
c
x (sum of residues at the poles of / ( z ) in the upper half-plane).

(6.19)
Condition (ii) above is sufficient to ensure that (6.18) is satisfied, for
lim If f(z)dz =lim

R
\ f(Reid)[Reiddd
fi^°° I J o -*°° Jo
^ lim [ \f(Rci6)\Rdd
fn At
^ lim -1Rd6 = 0, (6.20)
R-+°c JQ R
where we have used the fact that the modulus of the integral of a
function is less than or equal to the integral of the modulus of that
function. The vanishing of the modulus of the integral o n C ' a s / ? ^ o o
implies that the integral itself is zero.
Example 3 To evaluate
r
-rrn> (6-21)
where a is a real positive constant.

Consider
dz
(6.22)
where C is the closed contour of Figure 6.3. The conditions (i) and (ii)
above are satisfied since f(z) = (z4 + a4)~l has simple poles where
z 4 = -a\ and |/(z)| behaves like R~4 for large \z\=R. The four poles
are at
zk = ae{jt+2kn)l\ k = 0, 1, 2, 3, (6.23)
and only two of these lie in the upper half-plane:
z 0 = aelJT/4, zx = ae3jtl/4. (6.24)
To find the residues at these poles, we use (5.200) with p(z) = 1 and
q(z) = z 4 + a4. Hence the residue at z0 = azxnl4 is
_J_ 1
3 (6.25)
~4z
whilst that at zx = ae3jTl/4 is
Piz) J_ (6.26)
:
4z 3 4a3
Finally, using (6.19), we find
f
i)1
4a L 3
V2 V2 J
n
(6.27)
J. Integrals of the type / " . / ( ^ e ^ d r (

These integrals occur in the theory of Fourier transforms (see Chapter
7), and can be evaluated using the same contour C as in Figure 6.3.
We assume that
(i) f(z) is analytic in the upper half-plane except for a finite number
of poles, none of which lie on the real axis (as before);
(ii) \f(z)\ ^ M/R, where M > 0 is some constant, as \z\ = R tends to
infinity on C .
Condition (ii) ensures that the integral of f(z)elkz on C tends to zero
as R -> oo, for
at
lim
R
f(z)elkz dz
I \f(Reie)\ \eikReie\ R dd
•nil
— Q ~kRsind Rdd = lim —kR sin
(6.28)
Since s i n 0 ^ 2 0 / ; r for 0 ^ d^n/2, the last integral in (6.28) is less

than or equal to
2M lim f z-2kR6l« dO = lim ^ (1 - e"**), (6.29)
which tends to zero as /?—><». The vanishing of the modulus of the

integral on C" as /?—»o° implies that the integral itself is zero. Using
the Cauchy Residue Theorem for the contour C, as in 2 of this section,
we have finally
f f(x)elkx dx = 2jti x (sum of residues of f(z)elkz at the poles of f(z)
in the upper half-plane). (6.30)
Example 4 Evaluate
where k > 0, a > 0.

-r cos(kx)
(6.31)
Here
7 = jRe -2dxy (6.32)
where Re stands for the real part of the integral. The integral in (6.32)
now has the form of the left-hand side of (6.30). Consequently we
consider
j> -^—-2dz, (6.33)
where C is the contour of Figure 6.3, as before. The conditions (i) and
(ii) are satisfied since \f(z)\ = |l/(z 2 + a2)\ behaves like R~2 on \z\ = R
as /?—> ». The poles of / ( z ) are at z = ±\a of which only z — \a lies in
the upper half-plane. The residue at this pole is (using (5.183))
T eikz I e~ka
lim v(z - in) — — =—-. v(6.34)
î«L (z-ifl)(z + ifl)J 2\a '
Hence
r zxkx (z~ka\ n v
"1 5 ^ = 2;ri —— = - e , (6.35)
Loo x2 + a2 \2\a) a v }
and, since this is real, (6.32) gives
l = ^~ka. (6.36)
We note here that if k < 0 then the integral of (6.33) becomes

unbounded on the curved part C since
.**e«| =e -A
(6.37)
which, for k < 0 and 0 < 6 < n> diverges as R —> <». In this case we use
the contour in the lower half-plane (see Figure 6.4), the integral on the
semicircular part tending to zero as R—>™ by an analysis similar to
that in (6.28) and (6.29). Calculating the residue at z = — \a> we find
K
= — eka, k<0. (6.38)
2a
Hence, for all k,
[C^dX=Yai (6.39)
6.3 An alternative contour

The contours of Sections 6.1 and 6.2 have been either circles or
semicircles but other contours may be more suitable for a given
integral. The following example shows how a rectangular contour may
be used to obtain the value of an integral from a known result, without
expanding the contour to infinity in all directions.
Example 5 Suppose we want to evaluate
/= e~*2 cos(2ax) dx, (6.40)
where a > 0 is a real constant. Attempting the method of the last

section and replacing x by z, we find that on a semicircle z =Rel°,
Figure 6.4
• • x
6.3 An alternative contour 145
|e~22cos(2az)|^<*> as \z\ = R^>°° when jt/4<6<3jt/4 (since |e"z2| =

e-Ricos(2e) a n d c o s ( 2 ^) < o for JT/4 < 6 < 3JT/4). Hence the integral on
this semicircle cannot tend to zero. Suppose instead we choose a
rectangular contour C as shown in Figure 6.5, where the various
straight portions are denoted by Clf C2, C3 and C4. Now consider the
integral
dz. (6.41)
The function e z2 has no poles and it follows therefore from Cauchy's

Theorem that this integral is zero. Hence
+ ^2 dx +i e-(-"1+ij;)2 i dy = 0, (6.42)
•'Vi
since on C 2 , e" z2 = e"(jfl+i^)2; on C 3 , e " 2 ' = e"(x+i>;i)2 and on C 4 ,

e~z2 = e~(~Ari+1>;)2. We find, after some algebra, that
P e""2 dx - 2e~x2 f ' y) dy - e -C+2U W ) dx = 0
(6.43)
By putting yl = a and letting JCI—> o°, we have
: = e*2 e~*2[cos(2#jt) — i sin(2a;c)] dx

J — oo
= efl2 f e" x2 cos(2ax) dx, (6.44)
Figure 6.5
4-iy,
Cj •f 1
since sin(2«;c) is an odd function. But, from (2.10), we deduce
f e~x2dx = ^jz (6.45)
and hence
/ = [ e~*2 cos(2ox) dx = V^r e~a\ (6.46)

J — 00
By differentiating with respect to a> we note that
\ ^ (6.47)
J — oo
6.4 Poles on the real axis: the principal value of integrals

We have assumed in the previous sections that none of the poles of the
integrand lie on the real axis. When a pole does lie on the real axis,
we can still obtain a value for the integral J~oc/(jt) dx. However, since
we cannot integrate through a singularity (f(x) becomes infinite at a
pole), we must assign a so-called principal value to the integral.
Suppose a pole exists at x = x0 (a < x0 < b). Then the Cauchy principal
value of $af(x) dx (denoted by P) is defined by
P [ f(x) dx = lim [ [ ° /(*) dx + [ f(x) dx\ (6.48)

Ja 6—»0 LJa Jxo+d -I
Example 6 The integrand of

dx
is singular within the range of integration at x = 2. The principal value

is therefore
(6.53)
We now suppose that / ^ ^ / ( J C ) dx is to be evaluated using the

techniques in 2 of Section 6.2. The function is assumed to have the
same properties as stated in (i) and (ii) of that section, with the
exception that a simple pole is now allowed at x = x0 on the real axis.
We indent the contour C by constructing a small semicircle y of radius 6
centred at x0 (see Figure 6.6). Now (with C" being the curved part of C)
I f(z) dz = f° "fix) dx+fJ f(x) dx + \ f(z) dz + I f(z) dz

JC J-R x0+6 Jc Jy
(6.54)
= 2n\ x (sum of residues at the poles of f(z) inside C). (6.55)
To evaluate the last integral in (6.54), we use the Laurent expansion
of f(z) about z =x0 in the form
/(z) = — + a0 + ax(z - x0) + (6.56)

z — x0
Writing z — x0 = de'e, we find
\f{z)dz= \fêdieiedd + a0 f diedd

Jy jn oe Jn
+ terms involving higher powers of 6. (6.57)

Consequently,
lim f /(z) dz = -ri\a_x = -2m(^), (6.58)

6-^0 J v \ 2 /
where a_x is the residue at z =x0. The basic results (6.54) and (6.55)
become, as /?—»o° and <5—»0, using (6.58),
r rxo-6 rR -i
lim /(JC) dx + I /(JC) rfx = 2jri x (sum of the residues at the

/?-*oo U_R JXo+d J
-R
(a-\\
poles of/(z) in the upper half-plane) + 2m\— I, (6.59)
since J c /(z)dz—»0 as /?—»o° (as before). Hence, using (6.48), the

principal value is given by
P I f(x) dx = 2m x (sum of the residues at the poles of /(z)

J — oc
(a-\\
in the upper half-plane) + 2jri( ). (6.60)
If more than one pole exists on the real axis, (6.60) generalises to
pf f(x)dx = 2m
x [(sum of the residues at the poles of /(z) in the upper half-plane)
+ ^(sum of the residues at the poles of /(z) on the real axis)]. (6.61)
Example 7 A standard example is to evaluate
— dx, (6.62)
where m > 0 is a real constant.

Writing the integral in complex form we have
P
dz, (6.63)
where C is the indented closed contour shown in Figure 6.7. The

integrand has a pole at z = 0 and this is outside C because of the
indentation (a small semicircle, radius 6, centred on z = 0). The
function eimz/z has no poles within C and the residue at z = 0 is
(6.64)
z—0 \
Hence, using (6.61), we have, on letting /?-*<» and d—>0,

f00 QimX
P dx = 2jti(0 + \ . 1) = jri. (6.65)
J — CC X
Taking the real and imaginary parts of (6.65) we find

cos(rax)
and
•£ -dx = 0 (6.66)
sin(mx)
£ dx = n. (6.67)
We note that we have omitted the principal value symbol in (6.67)

since sin(m;t)/jc does not diverge as |x|—»0 and therefore does not
have a pole on the real axis. If m < 0, the contribution on the large
semicircle does not tend to zero in the upper half-plane but, as before,
we may use the contour in the lower half-plane shown in Figure 6.8.
We find in the same way, by letting R—»o° and <5—>0, that
sin(rax)
£ dx = —JZ, m <0. (6.68)
This integral will occur again in Chapter 7 and may be written as an

integral from 0 to <»,
3 JT/2, m>0
sin(rajt)
dx = i 0, m = 0, (6.69)
-nil, m<0
since the integrand in (6.68) is an even function of x.
Example 8 By integrating e imz /z(z 2 + a2) around a semicircular

contour in the upper half-plane, indented at the origin (see Figure
6.7), show using (6.61) that
where a and m are positive constants.

Since the integrand tends to zero sufficiently fast as R —> o°,
——; ^- djt = 2jri x (sum of the residues of the integrand
2 2
-xx(x +a ) . . . - u 1* 1 \
v 7
at its poles in the upper half-plane)
+ jri x (residue of the integrand at z = 0). (6.71)
In the upper half-plane there is a simple pole at z = \a with residue
T eimz 1 1
lim (z - in) — — — = - T-22 e"mfl. v(6.72)
}
z-iaL z(z + ia)(z-ifl)J 2a
For the simple pole at z = 0, the residue is
f eimz 1 1
lim z——22 =- = -=. (6.73)7
z-oL z(z + a 2 )J 2
a v
We have from (6.71) therefore
( ) la1
= P(l-e-"»). (6.74)
Taking the imaginary part of each side and noticing that the integrand
is an even function, we have
By differentiating with respect to the parameter m we find
I ^° 2 2 dx=y e~mfl, (6.76)
as in (6.36). ^
6.5 Branch points and integrals of many-valued functions

Consider now, for example,
f xj(x)dx, (6.77)
where or is a constant but not an integer. Writing
izj(z)dz, (6.78)
Jc
we notice that z a is a multi-valued function and that z = 0 is a branch
point. In Section 5.1, we discussed the functions z^ and z^ and found
that if the argument of z is changed by 2JZ (in other words, if a branch
point is encircled) then we obtain another branch of the function and a
different value from the starting value. In using the contour integration
technique, therefore, the contour must be chosen to exclude the
branch point so avoiding multiple values. Some integrals involving
multi-valued functions can be evaluated using a suitably indented
semicircle, and we illustrate this with two examples.
Example 9 Consider
;dx, (6.79)
1+jt 2
where 0 < a < 1.

Accordingly, we evaluate
(6.80)
where C is the indented closed contour shown in Figure 6.7. The small
circle, radius d, centred on z = 0 ensures that the branch point z = 0 is
outside the contour. As /?-><» on the large semicircle z = Rew,
f \[za/(l + z2)]dz\ behaves like Ra~l which tends to zero. On the
small semicircle z = deld
which tends to zero as <5 —> 0. Hence

r xa c° xa
2 dx + i 2 dx = 2jri x
[residue of za/(l + z 2 ) at z = i].
Jo 1 H~ x J_oc 1 + x
(6.82)
m
By changing x to — x = xe in the second integral and evaluating the
residue of za/(l 4- z 2 ) at the simple pole z = i, we find
r Xa
1+jt 2
~ dx +
f™ (xe*1)"
Jo 1+jt 2
T- dx = 2K'\
/e^ri/2\a
2i
:— . (6.83)
Finally then
(1 + e*8") I -^—2 dx = JZZ"M2, (6.84)

JQ -L ~r X
giving
a
C x J nzn'xocl2 n
5 dx = —- = —
Jo 1 + x 1 4- e 2 COS(JI
Example 10 Consider
where a > 0.
f x +a
(6.86)
Using the same indented semicircular contour (see Figure 6.7) to

avoid the branch point at z = 0, we now examine
j> zln_Ja2dz- ( 6 - 87 )
On the large semicircle, J |[(ln z)/(z 2 + a2)] dz\ behaves like R~1\nR
which tends to zero as /?—»oo. On the small semicircle z = dew,
lnz
which tends to zero as 6 —> 0. Hence

f00 lnx f° lnjc
2 2 = ZJTI
Jo JC + a J_oo
x [residue of In z/(z2 + a2) at z = ia], (6.89)
so that changing x to — x =xem in the second integral, we obtain
2 * ^ . (,90,
Jo Jt2 + fl2 Jo JC2 + « 2
Expanding the logarithms into real and imaginary parts, and taking the
principal value of ln(ia) since all arguments within C lie between 0 and
jt, we have
lnfl+i
2 -2 -2dx + n\\ — 2=- T I (6.91)
Jo JC2 + a 2 J o Jt2 + a 2 aV 2/
The real part of (6.91) gives
^lna. ^ (6.92)
v
2a '
It should be noted in the above two examples that the integral from
-oo to 0 included, or was a multiple of, the required integral from 0 to
oo. This arose because the single-valued parts of the integrands
(1/(1 + x2) and Il(x2 + a2)y respectively) were even functions of x. If
this is not the case then we require an alternative contour which also
excludes the branch point. A suitable contour is shown in Figure 6.9.
The contour C is made up of the large broken circle (radius R), the
two straight lines along the x-axis separated by an infinitesimal
distance (the 'branch cut'), and the small broken circle (radius 5)
surrounding the branch point z = 0. We shall see that, due to the
multi-valued nature of the integrand, the contributions above and
below the branch cut do not cancel. The following example illustrates
the use of this contour.
Example 11 Evaluate
/ = :dx. (6.93)
We therefore evaluate
V2
(6.94)
where C is the contour of Figure 6.9. Both contributions from the

large circle of radius R and the small circle of radius S tend to zero (as
R—»oo and S—>0, respectively). The contributions from above and
below the cut are, respectively, the required integral / in (6.93) and
(putting z — XQ2JIX on the line below the cut)
:<Z2"'dx = -t -tdx = (6.95)

Hence
r o dz = 21 = 2jri
r
c 1 + z3
x [sum of the residues of Vz/(1 + z3) at all its poles]. (6.96)
The poles occur at z 3 = - 1 = e wi+2 ^ wi , giving
zx = ^ ' \
z2 = e ^ = - l , (6.97)
z3 — e
Since these are simple poles, the residues are, by (5.200),

(^z/3z2)z=Zk = \zf\ for k = 1, 2, 3. From (6.96) we have therefore
/ = ™ (e-»i/2 + e -3.i/2 + e -5.i/ 2 ) (6 93)
= ^ ( - i + i-i) = | . ^ (6.99)
We note here that the integral in Example 9 can also be evaluated

using the cut circular contour of Example 11. However, the logarith-
mic integral in Example 10 cannot be evaluated in this way. We leave
the reader to attempt this and to verify that, although the contribu-
tions from above and below the branch cut do not entirely cancel,
those terms containing the required integral do cancel.
6.6 Summation of series

Contour integration can be used to sum particular infinite series.
Suppose / ( z ) is a function which is analytic at the integers z =
0, ±1, ±2, . . . and tends to zero at least as fast as |z|~ 2 as |z|—»<».
Now consider the function F{z) = JZ cot(jrz)f(z). This function F(z)
has simple poles at z=n (n=0, ±1, ±2, . . .) with residues (using
(5.200) with p(z) = JT COS(JTZ)/(Z) and q(z) = sin(jrz)) given by
jt COS(JTZ)/(Z)
= /(/!). (6.100)
(d/dz) sin(jrz)
We now integrate n cot(jzz)f(z) around a square S with corners at the
points z = (N + 2)(±1 ± i)> where N is a positive integer (see Figure
6.10). Using the Cauchy Residue Theorem,
4> n cot(jtz)f(z) dz = 2ri\ x [sum of the residues at the poles of

s
K cot(jrz)/(z) inside the square]
f v^ ,, x [sum of the residues of JT cot(JTZ)/(Z)1 ]

= 2m
L?? / ( B ) +
Lat the po.es of /<z) inside the square]!'
square]!' (6101)
where the first term is the sum of the residues of n cot(jrz)/(z) at the
poles of cot(Jtz). On the horizontal sides of S> z=x + iy where
y = AT + i > i. Then
\JIZ
|C0t(7Tz)| = e~H
— e"ijrz
= coth(jr |y I) < coth(;r/2), (6.102)

e "'-<
since y > \. Further, on the vertical sides of 5, z = ±(N + \) + \y so
that
= \cot{jz[±{N
= |tan(jrry)| = | nh(^y)| < coth(jr/2). (6.103)
Hence, from (6.102) and (6.103), cot(jrz) is bounded on S as N-*oc
and consequently J \jzcot(jzz)f(z)dz\ tends to zero on S as TV—»oo
because of the behaviour o f / ( z ) . Letting N->c° in (6.101), we see
that the integral on the left-hand side tends to zero giving
+ 00
2 fin)
n = —°°
= -[sum of the residues of n cot(jrz)/(z) at the poles of/(z)],
(6.104)
Figure 6.10
i
\ Q
which enables the series on the left-hand side of (6.104) to be

summed. Similarly, if we consider the function n cosec(jrz)/(z), we
find that the residues at the integers are (—l)nf(n). The corresponding
result to (6.104) is
= —[sum of the residues of n cosec(jrz)/(z) at the poles of/(z)],

(6.105)
which enables series with alternating signs to be summed.
Example 12 Consider
i-2XT=1 +i+i+
* + ---- <6-106)
n=on +1
To express this in the form of a sum from — oo to +o°, we write

(6107)
Now, using (6.104), we have

+ 00 -I
2 7 = ~[sum of residues of n cot(jrz)/(z 2 + 1) at z = ±i].

2
n = — o +1
(6.108)
Both poles at z = ±i are simple. The residue at z = +i is therefore
n
=-- coth n, (6.109)
whilst that at z = - i is
n
= - - coth n. (6.110)
Hence, from (6.107) and (6.108),
v-^ 1 + 71
2 (6.111)
n%n +l " 2
Similarly for the series
(6.112)
we write
(6113)
i^^=i+^ ? Sri-
Using (6.104) we have
2 —2 = —[sum of the residues of jt cosec(jtz)/(z2 + 1) at z = ±i]

(6.114)
^jrcosec(jri) Jt
2i -2i / sinh,
Hence, from (6.113),
(6.116)
Often the series we wish to sum begins with the n = 1 term, and the
function / ( z ) has a pole at z = 0, corresponding to the n = 0 term. In
this case, since the n = 0 term diverges in the sums on the left-hand
sides of (6.104) and (6.105), we omit the term n = 0 and the residue at
z = 0 is included on the right-hand sides. We illustrate this with an
example.
Example 13 Consider
i ^ = l +H U f 6 + .... (6.117)
n=\ n
Then / ( z ) = 1/z2 has a pole at z = 0. Hence
2 - 2 = - r e s i d u e of ^—-atz=0 , (6.118)
n=-^n L z J
since z = 0 is the only pole of/(z). To find the residue here, we expand
n cot(jrz)/z2 as a Laurent series about z = 0:
n cot(Jtz) n COS(JTZ)
z2 z 2 sin(jrz) z2/ JTV
(
The residue (the coefficient of 1/z) at this pole (of order 3) is therefore
- J T 2 / 3 . From (6.118) we have
2 l/n2 = n2/3, (6.120)
so that
oc
X l / « 2 = i 2 /6. (6.121)
Similarly for the series

oc
2 (-1)"-7/I2 = 1 - *+* - £ +..., (6.122)
we have
oc
2 {-l)nln2 = -[residue of n cosec(jrz)/z2 at z = 0]. (6.123)
Expanding about z = 0 to find the residue, we have - 1

jt cosec( JZZ) n
z2
(6.124)
/
Hence the residue (the coefficient of 1/z) is JT2/6 and consequently
S ^ - = " 2 2 ^ = ^ - -^ (6-125)
n =i n rt=_oc n 12
Problems 6
1. Evaluate by contour integration around the unit circle \z\ = 1:
Jo 5 - 4 c o s 0 Jo 2 - c o s 0
2. Show by contour integration that
2
f2"" dd In
Jo l -
Problems 6 159
Hence show that

r2jt
dO 2K
f 2 2
Jo (l + a ) - 4 a c o s 02 2
3. Evaluate, using a semicircular contour in the upper half-plane

1-a4' ' '
centred at the origin,

dx
2
_x + 2x
2
x dx ,. x r x2dx
4. By integrating
where a > b > 0, around a semicircle in the upper half-plane

centred at the origin, show that
r cos* _ it /e~b e~a\
L (x2 + a2)(x2 + b2) " {a2 - b2) \V~~a~r
5. By integrating the function
e 3iz
(z + 4)2 2
around a semicircular contour in the upper half-plane centred at

the origin, show that
I {x2 + AfdX 32 e~6'

6. Find the principal value of the integral
dx
7. Show that, if a =£fc,the function
has a simple pole on the real axis. By integrating this function

around a suitably indented semicircular contour, show that
f00 cos(fcx) - cos(q*)
I 2 dx = Jt{a-b)y
where a and b are positive constants.

8. Show that the zeros of 1 — 2e~ lz lie in the lower half-plane

(Im z < 0). Evaluate
f zdz
/=
where C is the rectangular contour with corners at the points

(-JT, 0), Or, 0), O , iK) and ( - J T , i/T). Show, by letting Kôo,
that
rx^nx_dx = jr
Jo 5 - 4 cos x 2
9. Evaluate
where 0 < a < 1, where C is the contour in Figure 6.9 consisting of

concentric circles of radii R > 1 and <5 < 1 with centres at the
origin joined by a cut along the positive real axis. Deduce, that, as
R —> oo and <5 —> 0,
'/"' , n
•dx =•
Jo 1 + x sin(jrar)
10. Show, by using the same contour as in Problem 9, that
(i)
I (l+x2)2(l+x)dX =
2\/2\l~^2)'
(ii) f (1+x)2 = it.
11. Evaluate
a) i i/(«2+i)2, (n) i(-ir/« 4 .

n=0 n=l
7
Laplace and Fourier transforms
7.1 Introduction
In the mathematical analysis of many linear problems it is often useful
to define an integral transform of a function f(x). The general form of
such a transform is given by
f(s) = j f(x)K(s,x)dx, (7.1)
where/(5) is the integral transform of/(jc) with respect to the kernel

K(s, x) (a given function of two variables x and s)y a and b being real
constants. There are a number of important transforms (for example,
Laplace, Fourier, Hankel and Mellin) obtained by choosing different
forms for K(s, x) and different values for a and b. The operation of
taking the integral transform, as in (7.1), exhibits a linearity property.
Suppose we let /{ } denote the operation of taking the transform of
whatever function occurs inside the curly brackets. Then
/{/(*)} =/(*) = ff(x)K(s, x) dx (7.2)

and clearly
I{af(x)} = al{f(x)}, (7.3)
whilst
I{af(x) + pg(x)} = j f W t o + Pg(x)]K(s, x) dx (7.4)
I}, (7.5)
161
162 Laplace and Fourier transforms
where a and j3 are arbitrary constants. Equations (7.3) and (7.5) show
that /{ } is a linear operator. In order that f(x) may be obtained if
f(s) is given, we now introduce the inverse operator I~1{ } which is
such that if
•} = /(*) (7.6)
then
f(x) = rl{f(s)}. (7.7)
Accordingly 7/~1 = /~ 1 / = l (the unit operator). It can be shown that

7"1 is also a linear operator. In this chapter, we concentrate on two of
the most commonly used transforms - the Laplace transform and the
Fourier transform.
7.2 The Laplace transform

We suppose that f(x) is defined for x ^ 0. The Laplace transform of
f(x) is then
= /(*)=[ f(x)e~sxdx. (7.8)
In general, the variable s may be complex. We shall assume, however,

that s is real until 3 of Section 7.8. It can be proved that provided
\f(x)e~ax\ ^M as x^>™, where M and a are suitable constants, then
(7.8) will exist. Functions satisfying this condition are said to be of
exponential order. We note that the function e*2 is not of exponential
order since ex e~ax increases without limit as x—»°° for all a. It is not
possible, therefore, to define the Laplace transform of this function.
The Laplace transforms of some elementary functions are given in
Table 7.1. The conditions on s given in Table 7.1 ensure that the
integral in (7.8) exists.
The proofs of these results all follow by integration from the basic
definition (7.8), and from use of the linearity properties (7.3)-(7.5).
For example
ax _ e~ax>>
{ ]=
__,, i i \
\s — a s + a/ s —a
using the standard transform of eax (see Table 7.1).

Table 7.1
2{f(x)} = /(*)
1 1/5 (5>0)
x n ( / i = O , 1,2, . . . )
1
5 —a
5 / ^ A\
COS(OJC)
sinh(ajc)
5 2 -a 2 (
^|J|)
s
cosh(fljc)
s2-a2 (j
'|a|)
7.3 Three basic theorems
1. The Shift Theorem

This states that if
W(x)}=/W (7.10)
then
(7.11)
Proof Using (7.8)
(7.12)
Jo
= f e-(s+a)xf(x)dx=f(s + a), (7.13)

Jo
since this integral is again (7.8) with s replaced by s + a.
2.
If
2{f(x)}=f(s) (7.14)
then
2{xJ(x)} = (-!)"£; f{s), (7.15)
where n = 1, 2, 3 . . . .
Proof Using (7.8),

f{s)=\ e-sxf(x)dx (7.16)
*/o
so that
= f (-l)Ve-"/(*) dx = (-\)n£{xnf(x)}. (7.18)

Jo
Hence (7.15) follows.
3. The Convolution Theorem

This states that if
2{f(x)}=f(s) (7.19)
and
<Plg(x\\ = e(s} (1 2(V>
then
i f f f /(JC - «)g(«) </M| = / ( * ) | ( S ) . (7.21)
(The integral JS/(x — u)g(u) du is usually called the convolution

of / and g and is denoted by /*g. Clearly, from (7.21), £{f*g} =
Proof Using (7.8), (7.21) takes the form
where the region of integration in the (w, x) plane is shown in Figure

7.1. The integration in (7.22) is first performed with respect to u from
u = 0 to u = x up the vertical strip and then from x = 0 to o° by moving
the vertical strip from x = 0 outwards to cover the whole (infinite)
shaded region.
We now change the order of integration so that we integrate first
along the horizontal strip from x = u to o° and then from u = 0 to o° by
moving the horizontal strip vertically from u = 0 upwards. Then (7.22)
becomes
g(u)du I e-xsf(x-u)dx. (7.23)

In the integral farthest to the right in (7.23), u may be regarded as a

constant. Hence putting x - u = t, so that dx = dt, (7.23) becomes
f g(u)du \ c-û+t)f(t)dt. (7.24)

Jo Jt=o
Equation (7.24) can be rewritten as
f g(u)e-sudu f f{t)t~stdt (7.25)
simply by collecting the terms involving u together. Each integral in

(7.25) is recognizable as the integral defining the Laplace transform.
Hence
•=/(*)«(*). (7.26)
We now give three examples of the use of these theorems.
Example 1 Since
2
^{sinClx)} = -0 , (7.27)
S! ~f~ 4
using the standard result of Table 7.1, then the Shift Theorem states
that
(7.28)
(s + a)2 + 4 '
Example 2 Since
1
(7.29)
' 5+2
Figure 7.1
•+•*
from Table 7.1, then the second theorem above states that
^{xe-2*} = ~ (_L) = _ L _ (7.30)

1 J
ds\s + 2/ (s + 2)2 v
'
and
Example 3 Given
/ = i eau cos[b(x - u)] du, (7.32)

Jo
we write f(x -u) = cos[b(x - u)] and g(u) = eau. Then f(x) = cos(bx)
and g(x) = Qax so that
f(s) = -^-T2> ^ ) = 7 i - - (7-33)
Hence by the convolution theorem
%\\\aucos[b(x-u)]du\=f{s)g(s) = - S
^ (7.34)
KJQ J \S — a)\s -TU)
7.4 The calculation of an integral

A slightly more sophisticated problem rests on the result (which we
assume without proof)
!t\^ f(x, u) du} = | #{/(*, u)} du, (7.35)
where J? is taken with respect to the parameter x {x > 0). Consider
Now
(7 38)
-
7.5 Laplace transform of an error function 167
Hence
Now since (using Table 7.1) J£{JZ/2} = it 12sy we have, for JC > 0,
Clearly if x = 0 then
r *M*u)du = 0 (742)
Jo u
and if x < 0 , from (7.41),
it
du = - - . (7.43)
A convenient way to represent the results (7.41)-(7.43) is by defining

the sign function, sgnjc, as
1, x > 0
0, x=0 (7.44)
L-l, x<0.
Then
r°°sim>w) it
du= — sgn x. (7.45)
JQ U 2
The Laplace transform approach to integration is a useful technique
and may be applied to more complicated integrals.
7.5 Laplace transform of an error function

In Chapter 8 we shall require the Laplace transform of a complemen-
tary error function defined in (2.49) by
= - ^ - f e~u2du. (7.46)
yji Jy
The particular transform needed is
V}, (7.47)
where a > 0 is a real constant. Now, from (7.46) with y = a/isjx,
The region of integration in the (w, x) plane is shown in Figure 7.2.

Instead of integrating with respect to u first (along the vertical strip),
we change the order of integration and first integrate with respect to x
(along the horizontal strip). Then (7.48) becomes
<t\erfc(-A-)} = -?- f e""2( f e"« <&) rfw (7.49)
= - ? - f e- w2 e- fl25/4 " 2 (7.50)

5 V ^ Jo
(7.51)
Now writing
u2 + ^ 42 = O - aVs/2u)2 + a> r7 ^
4M
and making the substitution

v = u- aVsllu, (7.53)
we find
u = \[v + V(^ 2 + 2a V (7.54)
and
(7.55)
Figure 7.2
i a
\
7.6 Transforms of Heaviside and Dirac functions 169
The range 0û<°° becomes — oo<u<oo (from (7.53)). Hence the

integral in (7.51) transforms to
dv . (7.56)
The first integral has the value V'n (see (6.45)),
J
and the second
integral is zero since the integrand is an odd function of v. Accordingly
from (7.56)
a \ ] e"
-SP erfc - (7.57)
7.6 Transforms of the Heaviside step function and the

Dirac delta-function
The Heaviside unit step function situated at x = a is defined by
x>a
H(x-a) = (7.58)
10, x<a.
This is a suitable (discontinuous) function for describing an 'off-on'
process in physical modelling and is shown in Figure 7.3. If x is a time
variable then prior to x =a the function is zero, whereas after x = a
the function has a non-zero (unit) effect. Due to its occurrence in this
way, we require its Laplace transform (see Example 10 of this
chapter). Consider
-a)}=\ H(x - a)e~sx dx (7.59)
= \ H(x - a)z~sx dx + [ H(x - a)e~sx dx (7.60)

Jo
e-sxdx = e-sa/s, (7.61)
using the definition (7.58).
Figure 7.3
H(x-a)
i
1 i
o a
Similarly, if y is some arbitrary (Laplace transformable) function

then
2{H(x - a)y(x - a)} = [ e~sxH(x - a)y(x - a) dx (7.62)
= f e-sxH(x-a)y(x-a)dx
Jo
+ [ e~sxH(x - a)y(x - a) dx (7.63)

•'a
= 0 + [ Q~sxy(x - a) dx, (7.64)
again using (7.58). Now writing x — a = v, the integral (7.64) becomes
+
f Q-s(v °)y(v) dv = e~say(s), (7.65)
Jo
where y(s) is the Laplace transform of y{x). Hence

2{H(x - a)y(x - a)} = e~say(s). (7.66)
The Dirac delta-function is defined by 6(x — a) = 0, x ¥= a, and
[ d(x-a)dx=\9 (7.67)
^ — 00
I 6(x-a)f(x)dx=f(a). (7.68)
The delta-function may be considered as the limit of a sequence of

ordinary functions. For example, defining
dn(x) = \/(n/jt)e-nx2 (n = 1,2, 3, . . .), (7.69)
then
f \/(n/jt)e-nx2dx = l, (7.70)
J — 00
using the standard integral /"«, e~nx2 dx = yj(jzlri). Hence
lim \ 6n(x)dx=i d(x)dx = l, (7.71)

n—»» J_oo J—oo
and the delta-function is seen to be the limit of a sequence of Gaussian

functions with decreasing widths.
If we consider the integral of S(x — a) over the interval — oo to xf,
7.7 Transforms of derivatives 111
then
= H(x'-a), (7.73)
using (7.58). Hence by formally differentiating each side we obtain
S(x-a)=-^-H(x-a). (7.74)
This shows that the delta-function is not a proper function in the usual
sense since the right-hand side of (7.74) is the 'differential' of a
discontinuous function.
The Laplace transform of the delta-function is easily obtained as
follows:
2{6(x -a)}=\ e~sxd(x -a)dx = e"Jfl, a>0, (7.75)
using the property (7.68).
7.7 Transforms of derivatives

One of the main applications of the Laplace transform is to the
solution of differential equations, both ordinary and partial, and we
therefore require the transforms of the derivatives of the dependent
variable. For the purposes of this discussion, we do not go beyond the
second differential coefficient.
1. Ordinary derivatives
Suppose y =y{x) is a Laplace transformable function. Then
•2>{^} = [ e - ^ d j c = Ue-**l +s f ye~sxdx (7.76)

idx) JQ dx L Jo Jo
= -y(0)+sy(s), (7.77)
where y(s) is the Laplace transform of y(x) and y(0) is the value of y
at x = 0. We have assumed here that ye~sx-> 0 as x —> <». Similarly
(7.79)
(7-80)
Table 7.2
2{f(*)} =/(*)
dy
sy(s)-y(O)
dx
s2y(s)-sy(O)-y'(O)
dx~2
dy(s)
xX dy S
dx ds }(S)
d2y{s) dy{s)
xX 2 d y
dx ds2 ds
X
dx2
2 ^W dy(s)+y.
X
dx2 ds2 ds ' ^yv>)
using (7.77), where we have assumed e sxy'(x)->0 as JC—>oo. Hence
y'{0), (7.81)
where y'(0) is the value of dy/dx at JC = 0.

The transforms of terms of the form xmdny/dxn may be obtained
using (7.15). For example
(7-83)
Some of these results are collected together in Table 7.2.
2. Partial derivatives
We now suppose that W(JC, t) is an arbitrary function of x and t> where
a ^ x ^ b and t ^ 0 , a and b being constants. Since f ranges from 0 to
oo, we may transform with respect to this variable as follows:
= [u(x, t)e~st]o + s i u{xy t)c~stdt (7.85)

Jo
= su(x,s)-u(x,0), (7.86)
7.8 Inversion 173
where u(x, s) is the Laplace transform of u(x, t) with respect to t.

Further
du(xf t)
s u{x, s) — su(x, 0) — (7.87)
dt t=0
Again we have assumed that u and its first derivative (with respect to
t) are of exponential order and are such that at infinity the integrated
terms are zero.
Besides the derivatives du/dt and d2u/dt2, we have the other
derivatives du/dx, d2u/dx2, and d2u/dx dt. For the first of these, we
have
= — u(x,s), (7.89)
where s is treated as a parameter. Similarly

u s
<e\—"} = ^' (790)
and
7.8 Inversion
Before applying the Laplace transform to the solution of differential
equations, it is necessary to discuss in detail how to find/(jc) from the
transform/(s). Such a process was indicated in (7.7) but we now need
specific techniques. One of the simplest and most obvious is to read
the inversion from a list of transforms so that from Table 7.1, for
example,
fe}cos(a*)- (7 92)
-
Other methods use partial fractions, the Convolution Theorem, and
the general inversion formula based on contour integration.
1. Partial fraction method

When the function/(s) has the form
f(s) = P(s)/Q(s), (7.93)
where P(s) and Q(s) are polynomials in s, the degree of P being less
than that of Q, the method of partial fractions may be used to express
f(s) as the sum of terms the inversions of which are readily
found.
Example 4 Consider
2s2 + 3s-4
Jy (7-94)
' (s- 2)(s2 + 2s + 2)
Then by partial fractions we may write this as
1 s+3
(7 95)
-
Hence
= e^ + e"^ cos x 4- 2e"* sin JC, (7.98)

using the standard results of Table 7.1 and the Shift Theorem. ^
2. Inversion using the Convolution Theorem

From (7.26) we may write
2-l{f(s)g(s)} = l'f(x ~ u)g(u) du. (7.99)
Hence if some function, say F(s), is given which can be written as the
product of two functions f(s) and g(s), the inversions of which are
known, we may then use (7.99) to obtain the inversion of F(s).
Example 5 To evaluate
i (7 ioo)
W\ -
7.8 Inversion 175
we write
/ ( ) ) (
^
Then from the standard results and the Shift Theorem we find
f(x)=<e-l{f(s)} = 4x, (7.102)
1 2
g(x) = %- {g(s)}=xe- *. (7.103)
2u
Hence from (7.99), since f(x — u) = 4(x — u) and g(u) = ue~ , we
have
î7(hf}=4[(x-u)ue~2udu' (7 io4)
-
which, by integration by parts, gives
vhfH^"2^-1- ^ (7 105)
-
Example 6 Given that
f(x) = 5£-l{f{s)}, (7.106)
we can show that
{ s } j t . (7.107)
Writing
g(s) = l/s, (7.108)
we have that g(x) = 1. Hence, using the Convolution Theorem,
W - / ( * ) j = ("fix - u)g(u) du (7.109)
= f f(x-u)du, (7.110)
Jo
since g(w) = 1. Writing x — u = ty we have finally
3. Inversion by contour integration: the general inversion

formula
It may be shown in general that
=-^r esj(s)ds, (7.112)

ln\ Jr_ioo
where s is now a complex variable. The path of integration is the

straight line L in the complex s-plane with equation s = y + iv,
-oo < u < oo, as shown in Figure 7.4, y being chosen so that all the
singularities of the integrand lie to the left of this line. This line is
known as the Bromwich contour. In practice, we evaluate the integral
in (7.112) by taking a finite straight line, closing the contour in a
suitable way and then letting the closed contour expand to infinity,
while using the Cauchy Residue Theorem. There are a number of
ways of closing the contour. Here we concentrate on the method
shown in Figure 7.5 in which the finite straight line L' is closed by
means of the part of the circle C, centre the origin. Then, provided
that |/(5)| < aR~k on s = Reie, where a and k are positive constants,
the integral of esxf(s) on C tends to zero as /?->», as we now show.
Consider first the part of C which is a semicircle S to the left of the
Figure 7.4
-•Res
Singularity •
of/(z)
Res
7.8 Inversion 111
y-axis. On this part s = Reld, where Jt/2 < 0 < 3JT/2. Hence on S
\esxf(s) ds\ = \eRx(cosd+isine)f(Reie)Reidid6\
= eRx cos e \f(Reie)\ R dd. (7.113)
By assumption / behaves as a power of R on this semicircle and
cos0<O. Hence as R-+™, (7.113) tends to zero since it includes a
decreasing exponential (with R) multiplied by a power of R.
Now consider the upper arc of the circle Ax in the first quadrant.
The contribution from this arc to the total integral is bounded as
follows:
\ tsxf{s)ds <\ \esxf(s)ds\<[ eRxcosd—kRd6, (7.114)

JAX JAX Jd0 R
where cos 60= y/R (see Figure 7.5). Now since 60^ d^Jt/2, it
follows that cos 6 ^ cos 60 = y/R. Hence (7.114) is less than
a f71'2 aeyx
Rx /R
- ^ e v d0 = -^(jt/2- e0). (7.115)
K JeQ R
In (7.115), the quantity Jt/2- 00 = Jt/2-cos~\y/R). For small t, the
function y = cos"11 has the expansion
y =y(0) + ^y'(0) + . . . = | - r + . . . . (7.H6)
Hence for large R, t = y/R is small and (7.116) gives
Finally then, for large R, (7.115) behaves as

are7* y
which tends to zero as R —• oo since k > 0.

Similarly the integral on the lower arc A2 in the fourth quadrant
tends to zero as R —» o°. By the Cauchy Residue Theorem we therefore
have
f e 5 7(s) ds + f e"/(j)rf5+ f e 5 y(5) <fc + f e5A/(5) ds

h JAI JA2 JU
sx
= 2it\ x [sum of the residues of e f(s) inside the contour]. (7.119)
Letting /?—><», the first three integrals in (7.119) tend to zero and the
finite line U becomes the Bromwich contour L. Further L is chosen to
lie to the right of all the singularities of f(s). Hence
= [sum of the residues of esxf(s) at the poles of f(s)], (7.120)

sx
since e has no poles.
Example 7 Consider the function
which has simple poles at s = ±2i. Using the formula for the residue at
a simple pole (5.183), we have
(residue at s = 2i) = lim Us - 2i) *f° 1 = ^~ (7.122)

5^2iL (s — 2\)(s 4- 2i) J 2
and
(residue at s = -2i) = lim (s 4- 2i) ^ = . (7.123)
Hence
f(x) = (sum of residues) = \(z2bc 4- e"2Lr) = cos(2x), (7.124)
which is the inversion of (7.121) (as can be seen from Table 7.1). ^4
Example 8 Consider the function in Example 5

4
which has double poles at s = 0 and s = — 2. Using the residue formula

for double poles ((5.206) with m = 2) gives
d f 4ciSX 1
2
(residue at 5 = 0) = lim — s -^- -r = JC - 1 (7.126)
v
' , - o d i L 52(5 + 2)2J v
'
and
d \ 2 4esx 1
(v r e s id
i d u e att s = - 22)
)7 = llii m — \(s
U 4-+ 22)) 2 -2r - -r = (A
(JC ^
5—2rfs L ' s (s + 2)2} v
(7.127)
Hence
/(JC) = (sum of residues) = (x + l)^2* + x - 1, (7.128)
as in (7.105). ^
7.9 Inversions of functions with branch points 179
7.9 Inversions of functions with branch points

When the function to be inverted possesses a branch point, we must
proceed as in Section 6.5 by making a cut in the s-plane so as to
exclude this point from the contour. The method of closing the
Bromwich contour therefore requires modification. Consider the
following example:
Example 9 Suppose we require

(7.129)
We first anticipate the result by evaluating the Laplace transform of
1/V* by elementary methods.
(7.130)
= f -e~su22udu = 2 f e~su2du= J(-). (7.131)

Jo u Jo y\s/
Hence
2-l{\Ns} = \N(nx). (7.132)
Now f(s) = 1/Vs has a branch point at s = 0. We can allow for this by
cutting the circular closure of the Bromwich contour as shown in
Figure 7.6 and surrounding the branch point at s = 0 by a small circle
C (radius p). Denoting the large curved part of radius R by C and the
two straight parts defining the cut by L1 and L2, then
since 1/Vs1 has no singularities inside the total contour. We showed in

the last section that the integral on C tends to zero as i? —> o°, whilst
that on L' tends to the required Bromwich integral. We now examine
the remaining three integrals in (7.133).
On C , we have s = pew and
^xp(cos 0 + isin 0)
-piewd0
Vpei e a
(7.134)
•f
which tends to zero as p —»0.
On Llf s = re31 = -r so that y/s = Vrein/2 = Wr. Hence
-rds=\ -rds=\ -1-{-dr) = -'\\ —j-dr. (7.135)
On L 2 , s = re" ijr = -r so that Vs = Vre~iJt/2 = -Y\/r. Hence
-r&= -rds=\ r(-dr) =-i\ —j-dr. (7.136)

V V
JL2^S JO V^ JO - i V r ' Jo V r ^
The cut constrains the argument of s to lie between — n and jr giving
an argument of n above the cut and — n below it. Consequently using
(7.133)-(7.136) we have, as /?-*«> and p ^ O ,
1 ry+i™ esx 1 f°° e~rx
/(*)=—T - r * = -r-^2(-i) ^-dr (7.137)
7 v 7 v 7
2JTI Jy-ioc V^ 2JTI JO Vr
1 r°°e~rjr
=- —rdr. (7.138)
V 7
jrJ 0 Vr
2
Putting rx = w , say, (7.138) becomes
/(*)
y w =- f - ^ r - — du=-^r f e - |<2 dM=-7^—, (7.139)
JTJ 0 (W/VJC)X WJCJO VW)
as found in (7.132). ^ f
The following two results have occurred previously and may be

obtained by the contour integration method:
(7 140)
-
(see (7.57)).
(2) , a>0, (7.141)
(see (2.45)).
7.10 Solution of ordinary differential equations 181
7.10 Solution of ordinary differential equations

It is often convenient to use Laplace transforms to solve ordinary
differential equations using the results of Table 7.2.
Example 10 Consider
d2y
(7.142)
given that y(0) = 0 and y'(0) = 2.

Taking the Laplace transform of the equation term by term and
using the results (7.61), (7.77), and (7.80), we obtain
2
(7.143)
s s
Hence
(s2 + 3s + 2)y (s) - 2 = - - — (7.144)
s s
so that
1 r2(* + l) 2e"H
(s2 + 35 + 2) L 5
2 2e" J
s(s+2) s(s + \){s + 2)'
Writing this expression in partial fractions, we find
(7147)
Inverting (7.147) using the Shift Theorem (7.11) and result (7.66), we
have
y(x) = 1 - e"2* + H(x - 1)[-1 + 2e" ( *- 1) - e " ^ - ^ ] . ^ (7.148)
The attraction of this method is that it enables us to solve the
differential equation by solving only an algebraic equation (7.144) and
iising known results for Laplace inversions. It is sometimes convenient
|p use the differential equation defining a special function to evaluate
its Laplace transform. Consider the following example:
Example 11 Suppose we require the Laplace transform of the Bessel

function of order zero, /0(*)- The differential equation satisfied by
y =J0(x) is (see (2.61) with v = 0)
d2v dy
Taking the Laplace transform of this equation, using Table 7.2, we

obtain
- f = 0, (7.150)
which simplifies to
(7.151)
as
Separating the variables and integrating gives
(7 152)
)' -
where A is an arbitrary constant which must be evaluated. Hence the
Laplace transform of / 0 (*) is
Putting 5 = 0 gives
=\\ J0(x)dx = l, (7.154)

o
using the result of Problem 8, Chapter 2. Hence the Laplace transform

of/0(JC) is 1/V(1 +s 2 ). We note that this result could have been found
by substituting the series for J0(x) into the Laplace integral and
integrating term by term. Summing the resulting series gives the
expansion of 1/V(1 + s2). However, using this method, we can prove
convergence only for |s| > 1. ^
7.11 Solution of a Volterra integral equation

The equation
y(x) =/(*) + fy(u)g(x - u) du (7.155)
is a type of Volterra integral equation for y{x)> where f(x) and g(x)
are given functions. The Laplace transform approach is particularly
suited to this form of integral equation since the integral is a
convolution of y and g. Hence transforming and using the Convolution
Theorem (7.21), we have
y(s)=f(s)+y(s)g(s), (7.156)
whence
Inverting (7.157) gives y(x).
Example 12 Consider the integral equation
y(x) = sin x + | sin[2(x - u)]y(u) du. (7.158)
Taking the Laplace transform gives
m=
TT?+*hm (7 159)
-
or
Inverting gives
>;(jc) = 3sinx-V2sin(V2jc). ^ (7.161)
7.12 The Fourier transform

In this section, we give the corresponding results to the Laplace
transform for the Fourier transform defined by
f(s)=\ f(x)e-™dx, (7.162)

J — 00
where the notation &{f(x)} =f(s) denotes the Fourier transform of

f(x). It can be shown that the inverse transform ZP~l{f(s)} is given by
{s)^ds. (7.163)
We note that in some books the definitions differ from those above by
having different multiplicative factors in front of the integrals. Any
two such factors the product of which is 1/2JT can be used.
1. Even functions
If f(x) is an even function (f(—x)=f(x) in the range —oo<x<oo)
then we may define the Fourier cosine transform ^c{f(x)} as
&c{f(x)} = / C ( J ) = f fix) cos(sx) dx (7.164)

with the inversion
l
{Us)} =/(*) = \ [fc(s) cos(sx) ds. (7.165)
2. Odd functions
If f(x) is an odd function (/(—*)= —/(*) in the range -oo<x<oo)
then we may define the Fourier sine transform ^ s {/(*)} a s
f sin(s;c) <fc (7.166)

Jo
with the inversion
=/(*) = - f Us) sin(sx) ds. (7.167)

Jt Jo
These three transforms are linear in the sense of (7.3) and (7.5).
Functions defined for x ^ 0 which are neither even nor odd can be
extended over the whole range by defining them to be either even or
odd, as in the case of Fourier series.
Example 13 Consider the Fourier sine transform of e~* for x > 0 (so
that the function is — e* for x < 0, thus making it odd). Then
= fJo dx = -^—-. (7.168)

s +1
Using the inversion result (7.167) for the Fourier sine transform, we
have for x > 0
e"* = - f -^— sin(sjt) ds. (7.169)

JT Jo S ~ h l
Choosing some constant value of JC, say x = k (and calling the dummy s
of integration JC) we have for k > 0
0
x sin(kx)
r°° JC sin(fct JT .
2v -dx=-e~k. (7.170)
Jo * + l
For k < 0 we have
r rtan (kr\ x = _ x k ^ ^ ( 7 _m)

3. Convolution Theorem
The corresponding result to (7.21) is
(* - K)g(u) dw} =f(s)g(s), (7.172)
where /(s) and g(s) are the Fourier transforms of f(x) and g(x) as in
(7.162). As with the Laplace transform we denote the convolution
integral by
| f(x-u)g(u)du=f*g (7.173)
with the property

/}. (7.174)
4. Transforms of derivatives
(a) Fourier transform If y =y(x), we may obtain the Fourier trans-

forms of the derivatives by simple integration by parts:
= [y(x)e-isx]Z^ + is [ yix^-^dx (7.176)

= isy(s), (7-177)
provided y(x)—*0 as JC—» ±°°.
Similarly
^ { ^ } = (w)2y(*) (7-178)
and, in general, provided y (n ~ 1) (jc)^»O as *—> ±00,
(w)"y(*)- (7-179)
(b) Fourier cosine transform Using the definition (7.164) and integrat-
ing by parts, we have
= [y(x)cos(sx)]% + s I y(x)sin(sx)dx (7.181)
+ sys(s), (7.182)
again assuming that y—»0 as JC—>o° and using (7.166). Further
| ^ } = ~y'(0) - s2yc(s)> (7.183)
with >>(*) andyf(x)->0 as JC-*O°.
(c) Fourier sine transform Here
*«->* < 7 - 184 >
) sin(^jc)]o - 5 f y (*) cos(sx) dx (7.185)
and
&s{^} = sy'(0)-s2Ms), (7.187)

provided y(jc) and yf(x)^>0 as JC—>oo.
All the above transformations of derivatives may be extended to
partial derivatives in the same way as in 2 of Section 7.7.
5 . Differential equations
The application of Fourier transforms to the solution of partial
differential equations is dealt with in the next chapter.
6. Integral equations
The integral equation
y(x)=f{x) + j y(u)g(x-u)du, (7.188)
w h e r e / a n d g are given, is a type of Fredholm equation (the important

point is that the limits are fixed whereas the Volterra equation of
Section 7.11 has a variable upper limit of integration). Fourier
transforming (7.188) and using the Convolution Theorem (7.172) gives
g(s)> (7-189)
whence
When f{s) and g(s) are known, y(s) may be inverted to give y(x).
Example 14 Consider the integral equation
(x-u)f(u)du=^-^. (7.191)
Taking the Fourier transform, we find
&{j J(x - u)f(u) du] =/(*)/(*) = ^ ( ^ ; ) - (7-192)

Now
(7.WM)
The integral in (7.170) and (7.171) is the derivative of that in (7.194)

with k replacing s. Hence
(?TiHfe~w' (7 195)
-
and from (7.192)
-]s]'2. (7.196)
Inverting (7.196) we have, using (7.163),
r f V » e e f d s (7.197)
V I \ e~s/2+>sxds+f es/2+is*ds) (7.198)
j(U+i)
r- ( f <?<**-*> ds+l e- ds) (7.199)
y/ji \J0 j0 I
^ ) (7- 200)
«-
Problems 7
1. Evaluate (i) ${cosh{ax)cos{ax)}, (ii) #{**}.
2. Use the Shift Theorem to obtain Z£{ex sin 2 x}.
3. Find the Laplace inversions of the following functions:
s x
n D
4. Use the Laplace transform to evaluate
(i) I e~x sinjc dx, (ii) I e~xx sinxdx,

Jo Jo
(iii) f t-*J0(x)dx.
Jo
5. Show that
where a> — 1 is a real constant.

6. Use the Convolution Theorem to find the function/(JC) such that
fiU)
f -«) .*,.!.
ioix>0.
7. Use the Convolution Theorem to evaluate $lJ0(u)J0(x — u) du.
8. Show that
for a > 0.
9. Solve by the Laplace transform method, where D = d/dx:
(i) (D2 + 3D + 2)y = 4e~3* with y(0) = 1, / ( 0 ) = 2,
(ii) (D 2 + % = i/(x - 2) with y(0) = 1, y'(0) = 2,
(iii) (D + l)y = 6(x - a) with y(0) = l,a>0.
10. Use the Laplace transform to solve the equations
(D - 2)x + Dy = 2y
(2D - l)x - {AD - 3)y = 7
for x(t) and y(t)y where D = d/A, with JC(O) = 0, y(0) = 1.
11. Express the solution of
Problems 7 189
where A is a real constant, in the form of an inverse Laplace

transform, and hence find the solution when A = 2.
12. Use the Bromwich contour method to find the inverse Laplace
transforms of
1 s
(0 ((ii i )
(s + l)(s-2f ( , + l) (5-l)2-
3
13. Find, using the Convolution Theorem, the solution of the integral
equation
y(x) = sin(3;c) + | sin(jc — u)y(u) du.
14. Using the result
COS(SJC) dx — y/(ft/
Jo
find the Fourier transform of e~^ . Hence derive the Fourier

transform of the delta-function d(x) by a limiting process using
the definition
15. Show that the Fourier transform of e~ aW (a > 0 ) is 2a/(s2 + a2)

and hence derive the transform of x2e~a[xK
16. Find the Fourier sine and cosine transforms of
(i) e ^ , (ii) 1/JC.
17. Using the first result of Problem 15 with a = l, show from the
Inversion Theorem that
COSJC , n
2
dx= —.
_OC1+JC e
18. Using the Fourier Convolution Theorem and the first result of
Problem 15, solve the integral equation
£d! • * . -
8
Partial differential equations
8.1 Introduction
Partial differential equations occur frequently in the formulation of
basic laws of nature and in the mathematical study of a wide variety of
problems. In general, since we live in a universe of three space
dimensions and one time dimension, the dependent variable, u, say, is
a function of at most four independent variables. Relative to a
cartesian coordinate system, therefore, we have
u = u{x, y, z, t). (8.1)
The relationship between u and its partial derivatives ux = du/dx,
uxy = d2u/dx dy, . . . ,
f(x, y , z, t\ u, uxy uy, uxx) uxy,...) = 0, (8.2)
where / i s some function, is called a partial differential equation. If/is
linear in each of the variables u, ux, uy, uxx, . . . and if the coefficients
of each of these variables are functions only of the independent
variables x, y, z, t, then the equation (8.2) is said to be linear.
Equations which are not of this type are said to be non-linear. In
general, the order of the partial differential equation is defined by the
order of the highest-order partial derivative in the equation.
8.2 Principle of Superposition

As with linear ordinary differential equations (see Section 3.1), the
Principle of Superposition applies also to linear partial differential
equations. As an illustration, consider the one-dimensional wave
8.3 Some important equations 191
equation
d2u l d2u
- 2,,2, (8.3)
dx 2
where c is a constant. It is easily verified that the two solutions of (8.3)
are
ux=f{x + ct), u2 = g(x-ct), (8.4)
where / and g are arbitrary functions, at least twice differentiate, oi
the arguments (x + ct) and (x — ct), respectively. Accordingly, by the
Principle of Superposition,
u=f(x + ct) + g(x - ct) (8.5)
is also a solution of (8.3). We shall show in Section 8.6 that (8.5) is, in
fact, the general solution of (8.3).
The Principle of Superposition, however, does not apply to non-
linear equations. There is no general method of obtaining analytical
solutions of non-linear partial differential equations and numerical
procedures are usually required for their solution. Because of the
difficulty in solving such equations, approximations are sometimes
made which linearise the equations. Provided the physical implications
of such approximations are understood, the analytical solutions of the
linearised equation (which are much easier to obtain) often give a
valuable insight into the solution of the original problem.
8.3 Some important equations

Many physical processes are described to some degree of accuracy by
linear second-order partial differential equations and it is principally
the analytical solution of this type of equation with which we shall be
concerned in this chapter. Some of these equations are listed here,
making use of the conventional notation that V2 represents the
Laplacian operator which in three-dimensional cartesian coordinates
takes the form
The form of V2 in other coordinate systems has been discussed in

Chapter 1.
(a) The wave equation
where c is a constant.
192 Partial differential equations
(b) The heat conduction, or diffusion, equation
where A' is a constant.

(c) Laplace's equation
V2w = 0. (8.9)
(d) Helmholtz's equation
V 2 w+Aw=0, (8.10)
where A is a constant.
(e) Poisson's equation
V2u=f(xyyyz). (8.11)
(f) Schrodinger's equation
V2u + a[E - V(xy yy z)]u = 0, (8.12)
where a and E are constants and V is the potential function.
(g) The Klein-Gordon equation
Dw + Aw = 0, (8.13)
where the operator • , the d'Alembertian, is defined by
and A is a constant.
Higher-order equations also arise. For example, the biharmonic wave
equation
V 4 « = V 2 (V 2 M) = - 1 | ^ , (8.15)
where K is a constant, describes the transverse oscillations of a thin

plate.
First-order equations of the general form
du du
P{xyy) — +Q(xyy) — = R(xyy)un (8.16)
also occur in a variety of physical problems (for example, in fluid flow,

kinetic theory of gases, transport processes). We note that (8.16) is
linear in u if n = 0 or 1, but non-linear otherwise.
8.4 Linear second-order equations 193
8.4 Linear second-order equations in two

independent variables
In the previous section, a number of physically important equations
were listed. Many of these equations, when only two independent
variables x and y are present, are special cases of the general linear
second-order equation
d2u d2u d2u du du
a—^ + 2h + b —-} + 2 / — + 2g — + cu = F(x, y). (8.17)
dx dx dy dy dx dy
The coefficients of u and its derivatives, namely a, h> b, f, g and c, are
constants or given functions of x and y, and F(x, y) is a given function.
The physical interpretation of the independent variables depends on
the particular situation. For example, Laplace's equation (8.9) in two
space dimensions, x and y, is a special case of (8.17) obtained by
putting 0 = 1, A = 0, 6 = 1, f = g = c = 0, F(x, y) = 0. On the other
hand, the heat conduction equation (8.8) in one space dimension, xy is
obtained from (8.17) by associating the variable y in (8.17) with time t
and then taking a = 1, h = 0, b =f = c = 0, g = -1/2K, F(x, t) = 0.
We note that the homogeneous form of (8.17) (obtained by putting
F(x, y) = 0) resembles the equation of a general conic, that is,
ax2 + 2hxy + by2 + 2fx + 2gy + c = 0. (8.18)
Equation (8.18) represents an ellipse (the circle being a special case),
parabola or hyperbola if ab —h2 is positive, zero or negative,
respectively. Hence we term (8.17) as being of
elliptic type if ab-h2>0, (8.19)
parabolic type if ab-h2 = 0, (8.20)
or
hyperbolic type if ab-h2<0. (8.21)
Clearly Laplace's equation (8.9) for which 0 = 1, h = 0, and b = 1 (as
above) has ab - h2 = 1 > 0 and is therefore of elliptic type. Similarly,
the heat conduction equation (8.8) in x and t for which (as we have
seen) 0 = 1, h = 0, and b = 0 has ab — h2 = 0 and is therefore of
parabolic type. In a similar way, it is found that the wave equation
(8.7) in x and t is of hyperbolic type.
When a, b and h are functions of x and y, the nature of the equation
may change from one region of the (JC, y) plane to another. For
example, the equation y d2u/dx2 + 2x d2u/dx dy + x d2u/dy2 = 0 for
which ab — h2 = xy — x2 = x(y — x), is elliptic if xy — x2 > 0 (that is, if
x > 0 and y — x > 0, or if x < 0 and y — x < 0), and is hyperbolic if

jcy - x2 < 0 (that is, if x > 0 and y - * < 0, or if JC < 0 and y - x > 0).
8.5 Boundary conditions

The general solutions of partial differential equations involve arbitrary
functions (see (8.5)), whereas the general solutions of ordinary
differential equations involve arbitrary constants. This shows that
there is an infinity of possible solutions and only by specifying the
boundary conditions can we obtain a specific solution. For this to
represent a physical situation, we require the problem as defined by
the partial differential equation and its boundary conditions to be well
posed. This criterion is satisfied if (a) the solution is unique, and (b)
small changes in the boundary conditions and the various coefficients
in the equation give rise to only small changes in the solution. When
(b) holds the solution is said to be stable.
Much work has been carried out to determine the various types of
boundary conditions which, when imposed on a partial differential
equation, lead to a unique stable solution. In general, there are three
main types of boundary conditions which arise:
(^4) Dirichlet conditions These specify the dependent variable u at
each point of the boundary of the region within which a solution
is required. For example, in two dimensions, u is specified at
every point on a boundary curve C of a plane region R.
The boundary value problem requiring the solution of V2u = 0
(in the appropriate number of dimensions) within a region R,
subject to u being given on the boundary of R, is called the
Dirichlet problem.
(B) Neumann conditions In this case, the values of the normal
derivative du/dn on the boundary of R are specified.
(C) Cauchy conditions These arise mainly in time-developing situa-
tions where u and du/dt are specified along the line t = 0. Such
conditions, therefore, are called initial conditions.
Cauchy conditions are generally met in conjunction with hyperbolic
equations, whereas Dirichlet and Neumann conditions relate specifi-
cally to elliptic and parabolic equations. Applying the wrong type of
boundary conditions to a particular equation can lead to an over-
determined system or an under-determined system. We shall not
attempt to discuss these problems here since we are more concerned
with the techniques of solution assuming that the given boundary
conditions are of the appropriate type for a well-posed problem.
8.6 Method of characteristics
1. First-order equations
The method of characteristics can be applied usefully to both linear
and non-linear equations. To illustrate the technique, we take the
simplest form of (8.16) by letting n = 0, so that
du du
P(x,y) — + Q(x,y) — = R(x,y). (8.22)
ox oy
Consider now the family of curves defined by the equation
^ ^ (8.23)
where we write P = P(x, y) and Q = Q(x, y). Then considering the

infinitesimal triangle (Figure 8.1), where ds is the infinitesimal element
of arc length of a typical curve C of the family given by
ds2 = dx2 + dy2, (8.24)
we have, using (8.23),
(8.25)
Q2
(8.26)
Hence
dx _dy ds
=
= dt. (8.27)
~P ~Q'
Now from (8.23), the curves are defined by
dy _Q
dx~~P' (8.28)
Figure 8.1
Multiplying (8.22) by ds and using (8.27), we have
Pds — +Qds — = Rds (8.29)

dx dy
and hence
f Q2) dx^ + V(P

v
2
+Q2)dy^ = R ds. (8.30)
dx ~ ' J dy
The total differential of u{x, y) is
du du
du= — dx + — dy (8.31)
dx dy
and so (8.30) takes the form
=
^ V(P 2 ^-6 2 )^' (8 32)
*
Using (8.27) this becomes
du=Rdt. (8.33)
Finally then (8.23) may be written as
*-!-!<-*>•
from which w can be found by integration. This integration is clearly
along any curve C defined by (8.23). These curves form a special
family relating to the given differential equation and are called the
characteristic curves (or, simply, the characteristics). The method of
characteristics requires that the region over which (8.22) is to be
solved should be filled with characteristic curves. The following
examples illustrate the method in some simple cases.
Example 1 To solve
given u = 2 on y = x2.
The basic equation (8.34) gives (since P — xy Q =y, R =2ry)
- =^ =^ (8.36)
x y 2xy
and the characteristics are defined by
dy/dx=y/x (8.37)
(see (8.28)). The solution of (8.37) is the family of curves

y/x = K, (8.38)
where A' is a parameter. Hence the characteristics are straight lines
through the origin with arbitrary gradients. From (8.36) we also have
on any characteristic curve
dullxy = dyly (8.39)
whence, cancelling y and using (8.38) to obtain dy = Kdx, we have
du=2xKdx. (8.40)
Integrating we find
u = Kx2 + A, (8.41)
where A is an arbitrary constant of integration. This constant is, in
general, different on each characteristic curve. Further, each charac-
teristic curve has a different value of K. Hence, as K varies, A varies
and we may write A =f(K), where / is an arbitrary function to be
determined. Writing A=f(K) in (8.41) and eliminating K using
(8.38), we see that the general solution of (8.35) is
u=xy+f(y/x). (8.42)
Since on the curve y =x2, u = 2 (given), we have
2 = *3+/(*). (8.43)
Hence
f(x) = 2-x3 (8.44)
giving
f(y/x) = 2-(y/x)3. (8.45)
Substituting (8.45) into (8.42), we see that the required solution of
(8.35) is
u=xy + 2-{ylxf. ^ (8.46)
Example 2 Consider now the homogeneous equation
v 2
^x ' 2' By
^ (8.47)
Equation (8.34) becomes (since R = 0)
dx _ dy du
2 2 (8.48)
2xy (jc + y ) 0
Hence the characteristics are defined by

dy(x2 + y2)
(8.49)
dx 2xy '
which on putting y — vx gives
dv (1 - v2)
(850)
Separating the variables and integrating, we find

l-v2 = K/x, (8.51)
where A' is a parameter, and hence
x2-y2
- = K. (8.52)
Also (8.48) requires, for finiteness, that
du=0 (8.53)
or
u = constant =f(K), (8.54)
as explained in Example 1. Hence, using (8.52), we find that the
general solution is
(8.55)
Suppose now that some boundary condition is given, say
Then
(8.57)
X /
Letting (2x — l)/x = w, say, we have

f(w) = ey", (8.58)
which defines the function /. Consequently (8.55) subject to (8.56)
becomes, using the form of/found in (8.58),
u = ^x2-y\ ^ (8.59)
At the beginning of this section, we considered the simple form of

(8.16) with n = 0. Equations of the general type (8.16) can easily be
reduced to the form (8.22) by letting ul~n = v, for n =£ 1, so that

(8.16) becomes
dv dv
P-—+Q — =(l- n)R. (8.60)
This equation can be solved by the method described in this

subsection. If n = 1, then v = \nu will reduce equation (8.16) to the
required form of (8.22). More generally, the method is applicable to
equations where R is a function of x, y and u. We illustrate this by an
example.
Example 3 To solve
du du
with u = 0 on y = x2.
Following (8.34) with R =xe~u, we have
- =^ ~ - (8-62)
x y xe
As in Example 1, the characteristics are
y/x = K. (8.63)
We also have from (8.62)
dx = eudu, (8.64)
which on integration gives
eu=x+f(K). (8.65)
Hence using (8.63),
eu=x+f(y/x). (8.66)
Now u = 0 when y = x2, giving
f(x) = l-x. (8.67)
From (8.66), therefore, we have
eu=x + l-y/x, (8.68)

whence
u = ln(x + l-y/x). ^ (8.69)
2. Second-order equations
One of the simplest applications of characteristics to second-order
equations occurs when the equation may be factorised into two
first-order equations. We illustrate this by the following example.
Example 4 To solve
s?u d2u 2d
2
u du_
dx2 dx dy dy2 dy
given that du/dx = y2 on x = 0 and u = ey on x = 0.
Consider the operator
L = ^- + x^-. (8.71)
dx dy
Then
3\ 32u d2u 2d
2
u du
(d+ x
d
) ( d , .,. U (8.72)
\dx dy/\dx dy) 3x2 + dx dy X
dy2' dy'
and hence (8.70) may be written as
L2u = 0 (8.73)
Now writing
Lu = w (8.74)
(8.73) becomes
L 2u = L(Lu) = Lw = 0, (8.75)
or, using (8.71),
dw dw
= 0. (8.76)
dx dy
The characteristics of (8.76) are found from
whence
y = \x2 + K. (8.78)
Also dw = 0 from (8.77), which implies
*=f(K), (8-79)
where / is an arbitrary function. The general solution of (8.76) is
therefore
w=f(y-hx2). (8.80)
Since du/dx =y2 on x = 0, we have, from (8.74),

du du du 9
w = — + x — = — = y2 on x = 0. (8.81)
dx dy dx
Using this in (8.80) gives
y2=f(y) (8.82)
and hence, again from (8.80), we find
w = {y-\x2)2. (8.83)
Having found w, we can now find u from (8.74) using the method of
characteristics. Writing out (8.74), using (8.71) and (8.83), we have
— + x — = (y - \x2fy (8.84)
whence
^ = __^£_. (g.85)
The characteristics are therefore

y = \x2 + Ky (8.86)
where K is a new constant of integration. Also
du = {y-\x2fdx (8.87)
2 2 2 2
= (I* + K - \x ) dx = K dxy (8.88)
whence
u = K2x+g(K)y (8.89)
where g is an arbitrary function. Substituting for K using (8.86) we
have the general solution
u=x(y- \x2)2 + g(y - \x2). (8.90)
Now u = ey on x = 0 (given) and hence from (8.90)
ey=g(y). (8.91)
Substituting for g(y) in (8.90), we have finally
u=x(y-\x2)2 + z{y-±x2\ ^ (8.92)
We now discuss the method of characteristics as applied to second-

order constant coefficient equations which are not factorisable into two
first-order equations as in Example 4. For this purpose we return to
the general linear form (8.17) and specialise t o / = g = c = 0 and F = 0

to obtain
(8.93)
dx dx dy dy
This equation, known as Euler's equation, is of one of the types
described by (8.19)-(8.21). Here we take ay h and b to be constants.
Defining new independent variables
£ = px + qyy rj = rx + syy (8.94)

where py qy r and s are constants, we have
du du dE du drj du du
L
— = —r — + = p—z. + r—, (8.95)
du du dt; du drj du du
Hence
d2u d (du\ i d d\i du „.
d2u +2pr
^ stu+r 2d
2
u (8, 98)x
w wr, ^> -
d2u d (du\
—? = — I — I = I q hs—
dy dy \dy) \ dg dr
,32u
(8.100)
<jg e g urj oi\
and
d2u d (du\ ( d d\( du c
dx dy
+rs
d2u 32u (8102)
^, w
Using (8.98), (8.100) and (8.102), (8.93) becomes
(ap2 + 2hpq + bq2) -^ + 2[apr + bsq + h{rq + sp)]
^ = 0. (8.103)
dr)
We may now choose the constants p> q> r and s to simplify the
equation. Let p = r = 1 and choose q and s to be the two roots kx and
A2, say, of the quadratic equation
6A2 = 0. (8.104)
Then (8.103) becomes
2[a +/i(A1 + A2) +feAaA2]- | ^ - = 0. (8.105)

From (8.104),
Aj + A2= -2h/b, (8.106)
k1X2 = a/b. (8.107)
Hence (8.105) finally has the form
^-*2)a0Ta (8 1O8)
-
Provided now b =£ 0 and the equation is not parabolic in nature
(ab-h2*0), (8.108) becomes
3r/=0, (8.109)
where, making use of our choices of/?, qy r and s,
% = x + X1y, r/=x + A2>^. (8.110)
Now (8.109) may be integrated immediately, first with respect to, say,
§ giving du/drj =m(rj), where m(rj) is an arbitrary function, and then
with respect to rj giving
where / and g are arbitrary functions. Using (8.110), we have the

general solution
u =f(x + Xxy) + g(x + A2y). (8.112)
The curves of constant § and constant r] are called the characteristic
curves. Hence
x + Xxy = constant, x + X2y = constant (8.113)
are the characteristics of the Euler equation (8.93) with a> b and h as
constants, and the form (8.109) is said to be the canonic form of the
original equation. We see that by choosing particular curves (the
characteristics defined by (8.113)) the equation can be brought into a
simpler form which may be integrated more easily.
The same type of analysis can be applied when the equation is

parabolic (ab — h2 = 0), but a different choice of p, q, r and s is
needed to simplify the equation. We shall not discuss this here.
Another feature illustrated by this analysis is that the basic equation
(8.104) defining A clearly has real roots if ab—h2<0 and complex
roots if ab — h2 > 0. Hence we see that hyperbolic equations have real
characteristics, whereas elliptic equations have no real characteristic
curves.
We conclude this section with two examples.
Example 5 The wave equation
Â-\^ =0 (8.114)
ox c at
is a special case of (8.17) with 0 = 1, h = 0 and b = -l/c2. Hence
ab — h2= — l / c 2 < 0 and the equation is therefore hyperbolic (as
mentioned in Section 8.4). Accordingly (8.104) becomes
l - A 2 / c 2 = 0, (8.115)
giving
A1 = c , A2=-c. (8.116)
The characteristics are therefore
§ = x + ct = constant, (8.117)
rj=x — ct = constant, (8.118)
and the canonic form is
Using (8.112) we have the general solution

u =f(x + ct) + g(x - ct), (8.120)
where / and g are arbitrary functions, as in (8.5). ^
Example 6 Consider the equation

_ d2u _ d2u d2u
— = y. (8.121)
C7A CA ay dy
Then ab -h2 = 2- ( - | ) 2 = - J < 0 and so (8.121) is hyperbolic. The

equation giving the values of A, (8.104), is
2 - 3 A + A2 = 0, (8.122)
whence
A = l or A = 2. (8.123)
The characteristic curves are
§ = x + 2y = constant, (8.124)
x) = x + y = constant. (8.125)
The equation reduces, by (8.108), to the canonic form
\{~i)wk=y=^~n' (8126)
or
d2u/dt;dr) = ri-t=. (8.127)
This equation may be readily integrated, first with respect to 77 giving
3u/d% = \Y]2 - %rj +/(§), (8.128)
where/is an arbitrary function, and secondly with respect to f giving
(8.129)
where F and G are arbitrary functions. Expressed in terms of x and y;
using (8.124) and (8.125), we have
(8.130)
8.7 Separation of variables

This method applies only to linear equations and makes use of the
Principle of Superposition (see Section 8.2) in building up a linear
combination of individual solutions to form a solution satisfying the
boundary conditions. The basic approach in attempting to solve
equations (in, say, two independent variables x and y) in this way is to
write the dependent variable u(x, y) in the separable form
u{x,y) = X{x)Y(y), (8.131)
where X and Y are functions only of x and of y respectively. In many
cases the partial differential equation reduces to two ordinary
differential equations for X and Y. Similar considerations apply to
equations in three or more independent variables. The following
examples illustrate the method both for first-order and second-order
equations.
Example 7 To solve
given u(x, 0) = cosh x3.

Now writing
u(x,y) = X(x)Y(y), (8.133)
equation (8.132) becomes
X'Y-x2XY' = 2x2yXY, (8.134)
where the primes denote ordinary differentiation with respect to the
appropriate variable (Xf = dX/dxy Yr = dY/dy). Equation (8.134)
may be written as
1 Yf V1
— — = 2y+ — y (8.135)
from which we see that the left-hand side is a function of x only, whilst
the right-hand side is a function of y only. Hence the equation has
been separated with terms dependent only on x on one side and
only on y on the other. For (8.135) to be satisfied, each side must be
equal to a constant (or, say) so that
i yt y
-A= > T+2y = a-

a
( 8 - 136 )
This is a pair of ordinary differential equations for the dependent
variables X(x) and Y(y). Integrating the first of the equations in
(8.136), we find
X(x)=Aeax3/3, (8.137)
while integrating the second, we have
Y(y) = Beay-y2, (8.138)
where A and B are arbitrary constants. Hence, from (8.133),
u(x, y) = X(x)Y(y) = ABe~y2ea(x3/3+y) (8.139)
y2 a(x3/3 y
= Ce- e +\ (8.140)
where C is an arbitrary constant. We can have an infinity of such
solutions with different choices of C and a. By the Principle of
Superposition, the linear combination
u-e 2J c e
(8.141;
all C
all a
is the general solution of (8.132). The values of C and a in (8.141) are

arbitrary at this stage and must be found by applying the boundary
condition. It is more convenient, however, to write (8.141) as
u = e-y2F(x3/3+y), (8.142)
3
since the sum is some function F of x /3 + y.
Now u(x, 0) = coshx3 so that, from (8.142),
(8.143)
Letting x3/3 = 0, then
F(0) = cosh(30). (8.144)
Hence
F(JC 3 /3 + y) = cosh(jc3 4- 3y) (8.145)
and the required solution is, by (8.142),
u = e~y2 cosh(x3 + 3y). ^d (8.146)
The separation of a partial differential equation into ordinary

differential equations, as in (8.136), is not always possible, as the
following example shows.
Example 8 To show that the equation
| ^ + 0 + ( * 2 + /) 2 M = O (8.147)
is not separable in x and y but is separable in plane polar coordinates
(r, 0).
Writing u=X(x)Y(y), (8.147) becomes
X" Y"
— + — + x4 + 2x2y2 + / = 0 (8.148)
JC Y
or
4
( ) V ( / ) 0 . (8.149)
The first bracket is a function of x only and the second bracket a

function of y only, but the product 2x2y2 contains both x and y. Hence
the equation is not separable.
From (1.102), the form of V2w in plane polar coordinates (r, 6) is
2 1 3

l d2u 4
Writing u = R(r)@(0) we have
The terms in square brackets are functions of r only and the remaining
term is a function of 6 only. Hence (8.152) is separable into two
ordinary differential equations for R(r) and O(0),
r d ( dR\ 6 1 d 2e
where a is called the separation constant. Â
Example 9 The temperature distribution T(r, t) in a homogeneous

sphere of radius a satisfies the equation
V2r = - ^ y , (8.154)
for 0 ^ r ^ a, t>0, where A' is a positive constant and V2 is in spherical

polar coordinates (r, 0, 0). Show that the substitution u(r, t) = rT(r, t)
transforms this equation into
32u 1 du
— = -—, (8.155)
and hence that, if T is finite (bounded) at the centre of the sphere and
ST/dr = 0 on the surface r = a for t > 0, the solution is of the form
T{r, t) = A + - 2) BnfTKioh sin(wnr), (8.156)

7" n = \
where A, Bn are constants, and o)n (n = 1, 2, 3, . . . ) are determined

by the positive roots of
tan((ona) = wna. (8.157)
To solve this problem, we use the form of V2 in spherical polar
coordinates (r, 6, 0) given in (1.84). We have, for T = T{ry t),
\ ^ . (8.158)
Hence
dr2 r dr K dt
Now putting u = rT, we have
3T \du 1U
2 (8.160)
dr r dr r2 y
'
and
d T 11
(8.161)
dr2 r dr2 r2 dr r
Substituting into (8.159), we find, as required by (8.155),
d2u l du
(8 162)
-&-K*' -
and writing u(r, t) in separable form as
u(r,t) = R(r)S(t), (8.163)
we have
R" 1 5'
where a is the separation constant. For convenience, we write

a = ±o)2 to indicate that the separation constant may be of either sign.
The solution of (8.164) for 5 with w ^ 0 is
S=Ae±Ka)2t. (8.165)
In order that the temperature falls off exponentially with time rather
than increasing without limit (a non-physical solution), we must exclude
the solution with the positive sign in (8.165). Hence the solutions of
(8.164) for w ^ O a r e
R = B cos(o>r) + C sin(cor), S = Ae~K(a2t, (8.166)
where A, B and C are constants. Hence, for co ^ 0 ,

u = [B cos(cor) + C sm(cor)]e-K(°2t, (8.167)
where the constant A has been absorbed into B and C. For co = 0,
S = £>, (8.168)
R = Er + F, (8.169)
where Dy E and F are constants, giving

(8.170)
where D has been absorbed into E and F.
We may now write the general solution as a superposition of the
infinity of possible solutions, one for each value of co (co = 0, and
(o = coi, co = co2, • • • which are yet to be found). Hence
u(r, t) = Er + F+j?J[Bn cos(conr) + Cn sm(corr)]c-K^\ (8.171)

n= \
Accordingly, since u = rT, we have
T(r, t) = E + - + - S [Bn cos(conr) + Cn sin(a> n r)]e-^ 2 '. (8.172)
Now for T to be finite at the centre of the sphere, we must remove any
term behaving as 1/r as r—»0 since this will diverge. Hence F = 0, and
Bn = 0 (for all n) since cos(ft>nr)/r—•<» as r—>0. Consequently
T(ry t) = E + -j^Cn sin(conr)c-K^f (8.173)
and for dT/dr = 0 on r = a> we have
^ ) ] =0, (8.174)
or
[rcon cos(conr) - sin(conr)]r=a = 0, (8.175)
giving
tan(cona) = cona, (8.176)
as required. An infinity of con values satisfy this transcendental
equation and these values must be used in (8.173) to form the solution
T(r, t).
By imposing some initial condition in the form T(r, 0 ) = / ( r ) , we
may determine the constants E and Cn as follows. From (8.173),
inserting t = 0, we have
T{r, 0) =f(r) = E + - 2 Cn sin(conr). (8.177)
Hence
rf(r) = Er+2,Cn sin(<onr). (8.178)

n= \
Multiplying by sin(a> m r) and integrating from r = 0 to r = a gives
rf(r) sin(a)mr) dr = E r sin(a)mr) dr

h Jo
+ 2 Cnsm((onr)sin(wmr)dr. (8.179)
n=l h
Now integrating by parts and using (8.176), we find
r
r sm(comr) dr = 0. (8.180)
Jo
Further
m
sin((Dnr) sin(a)mr) dr = - ( 2 2)dmn, (8.181)
Jo l \ l + o) m a /
for m, « = 1, 2, 3, . . . , using sin(cona) = cona/^(l 4- ft>^«2) and
cos(aâ) = 1/V(1 + G ^ 0 2 ) from (8.176). Hence from (8.179) we find
rf(r) sin(co m r) dr = -[ 2 2 )Cm, (8.182)

Jo 2 VI 4- (x)znal)
which determines Cm in terms of an integral.
Similarly, multiplying (8.178) by r a n d integrating from r = 0 to
r = «, we have
I r2f(r)dr = E { r2 dr + j?, Cn [ r sin((onr) dr (8.183)

Jo JO n=\ Jo
= ?y, (8.184)
using (8.180). Hence
£ = 4 f r2f(r)dr. (8.185)
fl Jo
Equations (8.182) and (8.185) determine the constants in the solution

(8.173) in terms of the given function/(r). ^
In many problems, the solution of a partial differential equation

obtained using the separation of variables technique involves some of
the special functions discussed in Chapter 2. The concluding example
of this section is of this type, and concerns the Schrodinger equation of
quantum mechanics (see (8.12)).
Example 10 To solve
0, (8.186)
where or is a constant and E is the energy parameter, subject to the
conditions that u is a single-valued function and is bounded (finite)
everywhere within a sphere of radius a, given u = 0 on r = a.
Expressing V2 in spherical polar coordinates (r, 0, (p), using (1.84),
(8.186) becomes
1 9 / ~ du\ 1 9 / du\ 1 d2u
-~ — \r— +-5 s i n 0 — + -; ^ ~
r 2 d r \ <9r/ r 2 sin0<90\ 9 0 / r2 sin2 d d(f>2
(8.187)
Writing
u(r, 0, <p) = / ? ( r ) B ( 0 ) * ( 0 ) , (8.188)
we have
1 d / 2dR\ 1 I d / d0\ 1 1
e
^ 7 V ) )
(8.189)
Letting
— — 2 = -m2> (8.190)
where m is a constant, we have

4> = Ae im *, (8.191)
which, provided m = 0 , ± 1 , ±2, . . . is such that for every rotation of
2JV we return to the same value of O. This is required by the condition
of single-valuedness of the solution for u. Using (8.190) in (8.189), we
find
>dRX • "" [ l d
' ' — (8.192)
The terms in r and 0 have now been separated so we may write
= A (8193)
- ^ '
where A is the separation constant. This equation is discussed in
Chapter 2 (see (2.189)) where it is noted that bounded solutions exist
only if
A = - / ( / + !), / = 0,1,2,..., (8.194)
and that these solutions are given in terms of the associated Legendre
functions
0 = £P}m|(cos 0), (8.195)
where |ra|=^/, and B is an arbitrary constant. Finally we solve the
equation for R(r) which, from (8.192) and (8.193), is
This equation may be put into a recognisable form by the substitutions
V(ar£)r=s, R(r) = -rP(s), (8.197)

\s
giving
s2 -£ + s - / + [s2 - (/ + ^)2]P = 0. (8.198)
ds ds
This is Bessel's equation of order v = / 4- \ (see (2.164)) which has the
bounded solution
P{s)=Jl+h(s)y (8.199)
the second solution Yi+^(s) being unbounded as s—>0. Hence
(8.200)
Combining the results (8.191), (8.195) and (8.200), we finally have
u{ry 0, <p) = Ktim<t>P\m\cos 0 W / / + i ( V ( a r £ ) r ) , (8.201)
where K is an arbitrary constant which can be determined by imposing

a further condition on u. In quantum mechanics this requirement is
that
[[[ u*ur2sin 0drd8d(f> = ly (8.202)

all space
thus normalising u*u = \u\2 to unity when integrated over all space.
Now if u = 0 on r = ay we have
JlH(\/(aE)a) = 0. (8.203)
For each given / value, there will be an infinity of roots rly r2y r3y . . .
such that // + i(r / ) = 0 (the values of the r, being the zeros of the Bessel
function). Hence
= rlr\yrl...y (8.204)
and the energy E therefore takes on an infinite set of discrete values
(this property being known as the quantisation of energy). ^
We conclude this section by remarking that the examples given here

are only representative of a whole variety of problems which can be
solved in this way. For reasons of space, we do not discuss non-
homogeneous equations or non-homogeneous boundary conditions
(but see Problem 9 of this chapter).
8.8 Integral transform techniques

The integral transforms discussed in Chapter 7 are useful in solving a
variety of partial differential equations, the choice of the most
appropriate transform depending on the type of boundary or initial
conditions to be imposed on the equation. The following examples
illustrate the use of the Laplace and Fourier transforms.
1. Laplace transforms
Example 11 To solve the diffusion equation
dx2 K dt'
for x > 0, t > 0 given that u = u0 (a constant) on x = 0 for f > 0, and
u = 0 for x > 0, f = 0.
Taking the Laplace transform with respect to t gives
U
\ 2'S =-psu(x,s), (8.206)
dx /C
using (7.86) with u(x, 0) = 0. Also
M(0, S) = f u(0, 0 e " " df = MQ/5. (8.207)

Jo
Hence solving (8.206) gives

W(JC, s) = A(s)e~xy/(s/K) + £(s)e W ( 5 / / °. (8.208)
To exclude solutions becoming infinite as JC—»o°, we take B(s) = 0.
Then using (8.207) we find
(8.209)
and hence
a(x,s) = ê-xy/<s/K\ (8.210)
The inversion of this function with respect to Ms a standard result

which we obtained in Section 7.5. From (7.57) with a=x/^K we
therefore have
(8 2n)
iW^] -
where erfc is defined in (2.49). ^k
Example 12 To solve the equation

d2u d 2u
(8.212)
dt2 dx2 dx2 dt2
for x > 0, t > 0 (K being a real constant) given that
u = du/dt = 0 for t = 0, x>0, (8.213)
M =l for x = 0, f > 0 , (8.214)
wÔ as x-+«>, t>0, (8.215)
and to show that
(8.216)
We recall that the Laplace transform with respect to t of time-

derivatives of u involve precisely the values of u and du/dt at t = 0
which are given in (8.213). Hence taking the Laplace transform with
respect to t is an appropriate method of solving the problem.
Using the results of Chapter 7 (see (7.87) and (7.90)), (8.212)
becomes
s2u(x, s) — su{x, 0) - (—)

\ dt/(=o
(8.217)
From (8.213), u(x, 0) = (du/dt),=o = 0. Hence (8.217) becomes

d 2u s2
where u = u(x, s). Solving this equation gives

U(JC, s) = A(s)e-sxl^l+Kls2) + B(s)esx/^l+K2s2\ (8.219)
Since u(xf f)—»0 as JC->O°, it follows that W(JC, s)—»0 as JC—»O° and we
must therefore have B(s) = 0. Accordingly
M(JC, 5) = A(5)e- 5 x / V ( 1 + ^ 2 ) . (8.220)
Now w(0, 0 = 1 by (8.214), so that
M(0, 5) = f 1. e~ 5 '^ = 1/5. (8.221)
Therefore in (8.220), A(s) = l/s and

i
ilfv c^ — ~—sx/V(l + K2s2) / o >yy-)\
U\Xy o) — e . yo.LLL)
S
Using the Bromwich contour (see Section 7.8) to evaluate this

integral, we have the solution in the form
i /•y + ioo i
u(x, t) = —\ - e -«^ + r t V(fe. (8.223)

2m Jy _ioo s
From (8.223)
<8224)
(8 225)
Uv)^ -
(8 226)
-
where s' = J^5. Since the Laplace inversion of 1/V(1 + s2) is / 0 (^) (see
Chapter 7, Example 11), (8.226) gives
S h = 4 y ° W K ) - -* <8227>
-
2. Fourier transforms
Finally we illustrate the use of the Fourier transform in solving
particular types of boundary value problems. In Chapter 7, it was
found that the Fourier sine transform of a second derivative term gave
the result
, —2 sin(sjt) dx = -s2us(s, t) + sw(0, r), (8.228)
Jo ox
whereas the cosine transform gave

f00 d2u / du\
J o
2 C O S ^ A J UX — —S UQ{S, I) — 1 — I , yp.AZry)
o a* \dx/x=o
ws and wc being respectively the Fourier sine and cosine transforms of
u{x> t) with respect to x. The choice of a suitable transform is
therefore dependent on the nature of the given boundary conditions.
If u(0y t) is specified, but not (du/dx)x=0, then the sine transform is
appropriate, whereas if (du/dx)x=0 is specified but not u(0, t) then the
cosine transform is appropriate. We now give an example.
Example 13 To show, using the basic result
w
e""2 cos(Aw) du = — e"A2/4, (8.230)
that the Fourier sine transform of erfc(fljt) with respect to x is
(1 - e~s2/4a2)/s, and hence to solve
d2u 1 du
for x >0, t>0 given that u = u0 (a constant) on x = 0 for t >0, and

u = 0 for x > 0, t = 0. (We note that this is the same problem as in
Example 11 which we now solve by taking the Fourier transform with
respect to x instead of taking the Laplace transform with respect
to t.)
Using the definition (2.49),
2 f°°
)=-M e~u2du. (8.232)
VK Jax
Hence its Fourier sine transform is
^s{erfc(fljc)}=-^- f sin(sx) f e~u2dudx. (8.233)

\Jt Jo Jax
Inverting the order of integration, we have
= ^ Q-u2du\ sm(sx)dx (8.234)

\JT Jo Jo
f-^ye-*2'4*2) (8-236)
= - (1 - e- j2/4a2 ), (8.237)
s
using (8.230).
Since w(0, t) = u0 is specified, the Fourier sine transform is

appropriate in this case. From (8.231), using (8.228), we therefore
have
-s2us(s, t) + su(0, 0 = 4 ^ ^ ' <8-238)

K at
or
^ + Ks2us = Ksu0. (8.239)
at
The solution of this equation is easily found to be
us(s, t) = A(s)e~Ks2t + ujs. (8.240)
Since u(xy 0) = 0, we have Q{sy 0) = 0. From (8.240) it follows that
0 = A(s) + uo/s, (8.241)

which determines A(s). Hence
Qs(s, t) = (uo/s)(l - e~Ks2t). (8.242)
Using the result (8.237), we obtain
u(x, t) = u0 erfc[x/2^(Kt)], (8.243)
which was the solution found in Example 11. ^4
Problems 8
1. Use characteristics to solve
du du
given u — cos x on x1 + y2 — 1,
given u —> e^ as x —> <»,
(in) x — -y — = u2
ax ay
given u = 1/lnjc on JC2^ = 1,
given w(0, >^) = sinh y/y.

Problems 8 219
J
dx2 XJ
' dx dy J
dy2 dy J
dx
can be written as L1L2(p = 0, where

d d d d
L1=x-—-^y — f L2 = y — -x — .
dx dy dx dy
By writing xp = L2<j>, obtain by the method of characteristics the
solution of the original equation subject to the conditions
(t> = ly2 on x = 1, d(j)/dx = \ on x = 1.
d2u du d 2u
dx2 dx dy2
can be transformed into
d 2v d2v
by putting u = e~û. Hence obtain the general solution of the

equation for u.
4. Solve
du du o
x — -2 — = 2x2 + 2-4y
dx dy
subject to the condition M(1, y) = 1+ y2 4- ey.
5. By writing 0 = \p + ^(JC2 + y2), solve the equation
dd> d<b 0
subject to the boundary condition

cf) = Q-y2 cos y + \{l + y2) on JC = 1.
6. Show that U(JC, f) =/(f)e*2g(/) satisfies the diffusion equation
if/and g are solutions of the equations

Hence show that if u(0, i) = 1/V(1 +1), then

2
v(x,t)= }
7. Prove that if AT is a positive constant, any solution u = u(r, t) of
the equation
s2u idu \ du
which is of the separable form u(r, t) = R(r)T(t) and which tends

to zero as t—> <» is of the form
where A, B and A are constants, and /0(*) and Y0(x) are the
zero-order Bessel functions satisfying
xy" + y' +xy = 0.
Given u(a, t) = 0 and u(r,0) = Jo(a>r/a), where a> is defined by
/0(cw) = 0, find the value of t such that
u(o, o = Mo, o).
8. An infinite cylinder with circular cross-section has radius a = 2
units. The temperature distribution within the cylinder is T(r, t).
The surface of the cylinder is kept at zero temperature for all
time, and at / = 0,
r(r,0)=/ 0 (1.2r),
where Jo is the Bessel function of zero order. Writing the heat
conduction equation
K dt
in cylindrical coordinates, show that the temperature distribution
is approximately
9. If the diffusion equation

d2u _ l du
3x2 ~ K dt
is to be solved for u(x, t) subject to the inhomogeneous boundary
Problems 8 221
conditions
u(0,t) = Uo, t>0y
u(l,t) = Uu t>0,
where Uo, Ux and / are given constants, and
U(JC,0) =/(*), O^JC^/,
where f(x) is a given function, show that the substitution
u(x, t) = v{x) + w(x, t)

can lead to a homogeneous boundary value problem for w(xf t) by
suitably choosing v(x).
Show that the solution of the diffusion equation when Uo = 0,
I/1 = l, 1 = 1 and/(;c) = 0 i s
u(x, t) = x + - f ) ^Z^- s i 2 2
^
10. The function u(r, t) satisfies the wave equation
c2 dt2 ~ dr2* r dr
within the sphere 0 =^ r < a where c is a constant and t > 0. Given
w=0 and du/dt = l at f = 0;

u = 0 on r = a;
u is finite as r->0,
show, by taking the Laplace transform with respect to ty that
Using the result
A1 sinh(sjt)] xt 2a ^ (—l)n (nnx\ (nnt\

Sin Sin
% j~2 • u / J = - + l ZJ ( ( }>
Is sinh(5fl)J a n n = 1 /i \ a ) \ a/
obtain a series expansion for u(r, t).
11. Show that, if u(x, t) satisfies the differential equation
32u _ du
a—5 + b e cx — — , (JC 52 0, fÔ),
5JC 5^
where ay b and c are constants, and the boundary conditions are

u=0 at t = 0 for all x\
w—>0 as JC—»o° for all f;
du/dx = 0 at x = 0 for all t,
then
{w(x, 0} = u(x, s) = 2 Irrrr^

as(cz-s/a) ly(s/a)
Using the result
find W(JC, 0-
12. Show that the solution of the heat conduction equation
d2u l du ,
subject to the conditions

du
— = 0 for x = 0, all t\
dx
u{l, t) = u0 for all t,
has the Laplace transform
u0
"(X' S)
ss cosh[y/(s/K)l] '
By finding the residues of u(x, s)est at the poles 5 = 0 and
s = -(2n- 1)2(JZ2K/412), where n = 1, 2, 3, . . . , show that
13. Solve, using the Fourier transform, Helmholtz's equation

V2w = u,
where — oo < * < oo, O ^ y ^ l , given
(du/3y)y=0 = 0 for all x;

0 2
u(x, 1) = e" ^ (a positive).
Problems 8 223
14. Show that the solutions to Burger's equation

du du d2u
V u — = v—r v (v = constant)
dt dx dx2 ;
may be obtained from solutions of dd/dt = v d26/dx2 by making u

proportional to (d/dx)(\n 6). Derive this constant of pro-
portionality.
Use this transformation, followed by separation of variables for
the linear ^-equation, to solve Burger's equation for
t > 0 with the initial condition
where u() is a constant, and the boundary conditions

9
Calculus of variations
9.1 Introduction
One of the elementary uses of differential calculus is in finding the
stationary values of functions of one or more variables. The calculus of
variations considers the more complicated problem of finding the
stationary values of integrals. A simple physical example which
illustrates the general type of mathematical problem involved is as
follows: suppose P(xly y{) and Q(x2, y2) are two given fixed points in a
cartesian coordinate system (see Figure 9.1). We wish to determine
the equation of the curve joining these two points such that when the
curve is rotated through 2n about the jc-axis to form a surface, the
surface area so generated has a minimum value. Suppose v =y(x) is
some curve joining P and Q. Then, if ds is the element of arc length,
the surface area S so formed is given by
\ y(x)ds = 2jt f2 y^[l + (dy/dx)2]dx. (9.1)
S is not a function in the usual sense since it depends on the form of

the curve y=y(x) in the whole range xl^x^x2. It is commonly
called a functional and is written as £[)>(*)], the square brackets
signifying that if the form of y(x) is known then the value of S is
known.
It is relatively straightforward to find the conditions under which
(9.1) takes on a stationary value, but not usually easy to determine
whether this is a maximum or a minimum, or some other type of
stationary value. In many problems, however, the solution may be
seen by physical arguments to be either a maximum or a minimum, and
224
accordingly we shall not attempt here to give the detailed mathemati-

cal analysis determining the nature of the stationary values.
9.2 Euler's equation

The simplest problem is to find a function y(x) such that the integral
I[y{x)]=Pf(x,y,y')dx (9.2)
is stationary with respect to small changes in y(x). Here y' denotes

dy/dx, and JCX and x2 are given fixed limits. We see that, in the case of
(9.1), the function/(JC, y, y') has the form
f(x,y,y') = 2jry\/(l+y'2). (9.3)
Suppose y=y(x) is some curve joining the points P{xlyyx) and
Q(x2, yi) as in Figure 9.1. Consider a small variation in y(x) so that
the new curve has the form
y{xy e)=y(x) + €ri(x), (9.4)
where 6 is a small parameter and rj(x) is an arbitrary function chosen
so that
Q. (9.5)
In this way the new curve (9.4) also passes through the end points P
and Q. The term er){x) is denoted by dy and is called the variation of
y. This variation may be one of two types - weak or strong - in the
following sense. Suppose that the curve Co represents y =y(x) (Figure
9.2). The neighbouring curve Cx corresponds to a weak variation since
Figure 9.1
-•*
226 Calculus of variations
the difference between the gradients of Co and Cx, at any value of x in

the range xx ^ x ^ x2y is a small quantity. In other words
d
6y' = — (6y) (9.6)
is small. However, the curve C2 (continuous but with a discontinuous

derivative) corresponds to a strong variation since, although 6y is itself
small, the derivative 6y' is no longer small due to the abrupt changes
in the slope of the curve. In most cases we deal only with weak
variations, and assume that the curve y(x) + 6y is similar to Cx and is
differentiate to all desired orders.
Inserting (9.4) into (9.2) we can write
= Ff{x,y{x)- er,'(x))dx, (9.7)
whence
61(6) =1(6)-I
(9.8)
Assuming now that the first term of the integrand may be expanded in
powers of 6, we have (using Taylor's expansion of a function of two
variables)
6
61(6) = 6ll+ — I2 + terms in 63 and higher orders, (9.9)
where
(9.10)
Figure 9.2
and
To the first order in e, the variation dl(e) is given by

6I(e) = €ll9 (9.12)
and the sign of dl(e) therefore depends on the sign of e. The value of
^[.yC*)] m (9.2) therefore may increase for some variations and
decrease for others. However, for the integral to have a maximum or
minimum, all variations with respect to 6 should respectively decrease
or increase its value. This can be achieved by requiring in (9.12) that
lx = 0, that is
Integrating the second term in (9.13) by parts we have
However r/(xj) = r/(jt2) = 0 (see (9.5)) and the first term on the right of
(9.14) is therefore zero. Inserting (9.14) into (9.13) gives
- *
Now rj(x) is an arbitrary function, apart from the conditions (9.5).
Hence we may make use of a fundamental result which states that if
rj(x) is continuous and has continuous derivatives up to at least second
order in the range xx ^x ^x2 with r){xx) = rj(x2) = 0 and rj(x) =£0 for
xx<x<x2y and if g(x) is itself a continuous function and
X2
fl(x)g(x)dx=0 (9.16)
for arbitrary (every) rj(x) satisfying the conditions above, then

g(x) = 0. Since (9.15) holds for arbitrary rj(x), it follows from the
above fundamental result that a necessary condition for the integral to
be stationary is
gfimO. (9.17)
This equation is known as Euler's equation and its solution leads to
the form of y(x) which produces the stationary value of /. The curves
y =y(x) so obtained are called extremals of (9.2).
Example 1 To find the stationary value of
I[y(x)\ = \°p [ ( £ ) ' - 2yx + y2] dx, (9.18)

where P = (0, 0) and Q = (1, 1).
From (9.18)
f(x,y,y') = y'2-2yx+y2 (9.19)
and hence
The Euler equation (9.17) is therefore
-2jt + 2 v - £ ( 2 y ' ) = 0 (9.21)
or
d2y
(9.22)
The solution of this equation is

y = A coshx + B sinhx + x. (9.23)
By requiring this curve to pass through P(0, 0), we have
,4 = 0, (9.24)
while for it to pass through Q(l, 1) we must have
B = 0. (9.25)
Hence
y=x (9.26)
is the extremal curve.
To evaluate the stationary value of (9.18), we insert (9.26) into the
integrand of (9.18) and integrate. Then
/ = [ (l-2x2 + x2)dx (9.27)
= (l-x2)dx = l ^ (9.28)
9.3 Alternative forms of Euler's equation

Euler's equation (9.17) assumes simpler forms when one or more of
the variables x, y or y' are absent from the function /.
Case 1 If y is explicitly absent from / then

f(x,y,y')=f(x,y') (9.29)
and hence df/dy = 0. The Euler equation (9.17) now becomes
dx\dy /
giving
df/dyf = constant. (9.31)
Example 2 Suppose we want to find the equation of the curve of

shortest length which joins two points P and Q in the plane. If
y =y(x) i s some curve joining P to Q, then the element of arc length
along the curve is given by
ds = yj(\+y'2)dx. (9.32)
Hence the total length of the curve from P to Q is
rQ
S=\ \/(l+y'2)dx. (9.33)
Jp
We can now apply the Euler equation to this integral and, since the
integrand is explicitly independent of y, we can use the form (9.31).
The Euler equation therefore has the form
j - 9 [V(l +y' 2 )] = constant (9.34)
or
y'
vYi—^ = c ( 9 * 35 )
Hence
y'=Ay (9.36)
where A is a constant, and the equation of the curve of shortest length
joining P and Q is therefore
y=Ax + By (9.37)
the constants A and B chosen so that (9.37) passes through the end
points P and Q. ^k
Case 2 If y' is explicitly absent from /, then /(JC, y, y') =/(*, y).
Therefore df/dy' = 0 and the Euler equation (9.17) becomes
df/dy=Q. (9.38)
Example 3 To find the extremal curve of the integral
1=1 (2y4-xy)dx, (9.39)
we use (9.38) to obtain

df/dy = Sy3-x = 0 (9.40)
or
y = ±xK (9.41)
The stationary value of /, found by inserting (9.41) into (9.39) and
integrating between the given limits, is / = — j ^ . ^
Case 3 If JC is explicitly absent from / then /(JC, y, y') =f(y, y') and
hence df/dx = 0. To see the effect this has on the Euler equation
(9.17), we note that the total differential of /(JC, y, y') with respect to JC
is given by
^31 3ldy K ±
dx dx dydx dy'dxKy} K }
using df/dx =0.

Now multiplying Euler's equation (9.17) by y', we have
Eliminating y' df/dy between (9.43) and (9.44) gives
y y 0 (9 45)
dx dy' dxKdy'
or
£</->•§)-*
Integrating (9.46) we find
f~y'§ = C, (9.47)
where C is a constant.
The following two examples illustrate the use of this form of the
Euler equation.
Example 4 An ancient problem is to find the curve joining two given

points in a vertical plane for which the time of descent of a particle
down the curve (under a constant gravitational field) is a minimum.
Such a curve is called the 'curve of quickest descent' or the
brachistochrone (from two Greek words - brachistos meaning 'shor-
test' and chronos meaning 'time'). Suppose a particle starts from rest
at a point O (see Figure 9.3) and slides down the curve OA, frictional
forces being neglected. Then at a typical point P(x, y)y the velocity of
the particle is
^ (9.48)
where g is the acceleration due to gravity. The time 8t taken to travel

from P to a neighbouring point P' is therefore
8t=8sN(2gy), (9.49)
where 8s is the length P to P'. Allowing P' to tend to P, we have
ds V(l + v' 2 K
<ijc, (9.50)
and therefore the total time T from O(0, 0) to A(xlf yr) is
(9.51)
V(2g)
We now wish to find y(x) such that T is a minimum. The integrand of
(9.51) is explicitly independent of x and hence we may use the form
Figure 9.3
o
ds
(9.47) of the Euler equation. Accordingly, we have
or
/ / l + v' 2 \ y' 2
V --r~77 2T = C - (9-53)
V\ y ) vyy(l+y )
This simplifies to
y ( l + y ' 2 ) = D, (9.54)
where D is a constant.
Making the substitution y' = cot Q> we have
and
^ = -,^z = D tan 0 sin(20) = D[l -cos(20)], (9.56)

at7 y do
whence
(9.57)
Equations (9.55) and (9.57) give the parametric form of the equation
of the curve, and define a cycloid. ^
Example 5 We now return to the problem discussed in Section 9.1 of

the generation of a surface of minimum area. The integral to be
minimised is given in (9.1) and has an integrand which is explicitly
independent of x. Hence, using the form (9.47) of Euler's equation,
we obtain
w^)=a (9 58)
-
After simplification, this becomes
y2 = C 2 ( l + / 2 ) , (9.59)
which integrates to give
(^^) (9.60)
9.5 Several dependent variables 233
where the constants A and C must be chosen so that the curve passes
through the end points P and Q. This curve is known as a catenary,
and the surface of minimum area generated by this curve is known as a
catenoid. ^
9.4 Extremal curves with discontinuous derivatives

When the extremal is found to consist of two or more branches, we
must check whether it is possible for the given end points to lie on one
and the same part of the curve. If this is not possible then the curve is
not the required extremal since the end points are not joined by a
continuous curve. We can illustrate this by the following example.
Example 6 Suppose we require the curve y =y(x) which minimises

cQ
1= x2yf2dx, (9.61)
Jp
where P = (1, 2) and Q = (-2, - 1 ) .
The Euler equation, from (9.31), is
2x2y' = Cy (9.62)
from which, by integration, we have
y=Alx + B. (9.63)
The curve on which P and Q lie is therefore y = 2/x, but P is one
branch of this curve and Q is on the other branch (see Figure 9.4).
Hence this is not the required extremal (in fact, evaluating / with this
curve gives infinity), and no extremal curve of the form (9.63) exists.
However, provided extremal curves with discontinuous derivatives are
allowed, we may obtain an extremal curve joining P to Q by noting
that the integrand in (9.61) is a perfect square and hence its minimum
value is zero. This is achieved by taking y'=0, or y— constant. A
possible extremal is shown in Figure 9.4, for which y' = 0 except at
9.5 Several dependent variables

The integral (9.2) and its associated Euler equation (9.17) may be
generalised to n dependent variables y^x), y2(x), . . . , yn(x) by writing
= Pf[x;yi(x), y2(x), . . . , yn(x);y[(x)9 y2(x), . . . , y'n(x)] dx. (9.64)

Defining (as an extension of (9.4))

(9.65)
where i = 1, 2, . . . , n, and where the rj^x) are n independent ar-
bitrary functions satisfying
tfc(*i) = *fc(*2) = 0, (9.66)
we obtain, in a similar way to the proof given in Section 9.2, the set of
n Euler equations
<9-67>
All of these equations must be satisfied simultaneously for (9.64) to
have a stationary value.
9.6 Lagrange's equations of dynamics

Consider a dynamical system made up of a given number of particles.
Let #i, #2> • • • > <Zn be fl independent quantities (functions of time t) in
terms of which it is possible to specify uniquely the position of each
particle of the system. Such quantities are called the generalised
coordinates of the system and may be, for example, distances or
angles, and n is called the number of degrees of freedom of the
system. For a single particle in three-dimensional space we need only
Figure 9.4
-•*
O
9.6 Lagrange's equations of dynamics 235
three coordinates to specify uniquely its position at time t. Clearly we

could choose qx=xy q2 = y> q3~z or, in terms of spherical polar
coordinates (see (1.84)), q\ = r, q2= 6, q3 = (p. Similarly, for two
particles we require six coordinates. In the case of a rigid body, say a
solid ellipsoid, six generalised coordinates are required: three to
specify a point in the body, two to specify an axis through the point
and a further one to represent the rotation of the body about this axis.
It can be shown that the Newtonian equations of motion for a
system with kinetic energy T and potential energy V can be derived by
requiring the quantity Jjj (T — V) dt to be stationary, where T and V
are, in general, functions of the qly q2y . . . , qn and their time
derivatives. It is common to call T — V = L the Lagrangian of the
system. The condition for
f\r-V)dt= I' Ldt (9.68)
to be stationary is (by Section 9.5) that L satisfies the set of n Euler

equations
ftKft)* (9.69)
where L = L{t\ qlf q2, . . . , qn\q\, qi, - - , qn) and qt = dqjdt.
Equations (9.69) are known as Lagrange's equations and provide a
powerful technique for deriving the equations of motion of a dynami-
cal system. This is illustrated by the following two examples.
Example 7 We consider the equation of motion of a simple pendulum

which was discussed in Chapter 3. A particle of mass m is suspended
from a point O by a weightless inextensible string of length / (see
Figure 9.5) and allowed to oscillate in the vertical plane. The position
of the particle at time t is uniquely specified by the angle 6 made by
the string with the vertical and hence the system has one degree of
freedom. We therefore choose 6 as the generalised coordinate to
describe the motion.
The kinetic energy T of the system is
T = \mv2 = \m(l6)2y (9.70)
where v = 16 = Id6/dt is the velocity of the particle. Likewise, the

potential energy measured from the horizontal through O is
V = -mgl cos 0. (9.71)

Hence
L=T-V = + mgl cos 0. (9.72)
Since there is only one generalised coordinate, the single Lagrange
equation (from (9.69)) is
9L dL(dL\_
(9.73)
~dO~~dt\d8)~ '
which, using (9.72), gives
-mgl sin 6 (m/20) = 0 (9.74)

at
or
d 2 e * .
(9.75)
-d7 + lsn
This is the equation of motion discussed in Section 3.9 and has elliptic
integral solutions. ^
Example 8 Three masses mly m2 and m3 lie on a straight line and are
connected by two springs of stiffness K for which the potential energy
is (by Hooke's Law) V = \K (extension)2. No other forces act. If xx, x2
and x3 are the displacements when the system is disturbed from its
equilibrium position then
L = T - V = \mxx\ + \m2x\ + hm3x\ - \K{x2 - xxf - \K(x3 - x2f.
(9.76)
The Lagrange equations (9.69) give (with qt = xt)
dL d_(dL\_
(9.77)
~dxi~Jt\dlJ~
Figure 9.5
10
9.7 Integrals involving derivatives higher than the first 237
for i = 1, 2, or 3, whence
m1x1 - K(x2 - JCO = 0, (9.78)
m2x2 + K(x2 - JCO - K(jt3 - *2) = 0, (9.79)
m3x3 + AT(JC3 - JC2) = 0. (9.80)
This coupled set of three second-order differential equations defines
the motion of the system, and elementary methods may be used to
obtain xlf x2 and x3 as functions of time. ^
9.7 Integrals involving derivatives higher than the first

Suppose we require that
\X2f{x\y, y\y\ ...,y ( n ) )dx, (9.81)
where y^n) = dnyIdxn> should be stationary subject to the end

conditions
y(Xl) = au y\xx) = a2,..., y^KxJ = an, (9.82)
{n l
y(x2) = bu y'(x2) = b2,..., y ~ \x2) = bn. (9.83)
Then by an analysis similar to that of Section 9.3 we find the
generalised Euler equation (given here without proof)
( ) + ( ) + ( i r ( ) 0 (984)
Example 9 Consider the integral
I[y{x)] = _[* (y"2 ~3y'2- 4y2) dx, (9.85)

given
y(0) = 0, y'(0) = 0, (9.86)
y(ji) = 0, y'{n) = \. (9.87)
The Euler equation (9.84) now gives
~8y~ic(~6y')+is{2y")=0> (9 88)
-
or
d\ d2y
J + iJ-4y.O. (9.89)
The solution of this equation is

y = Aex + Be~x + C cos(2x) + D sin(2x), (9.90)
where A, B, C and D are to be found by imposing the given conditions
(9.86) and (9.87). Hence
l l s i n h j r
A = B = c = D = i
4
2(1-e*)' 2(1 -e-*)' 2 ( 1 - cosh * ) ' *
(9.91)
The stationary value of / is obtained by inserting (9.90) and (9.91) into
(9.85). ^
9.8 Problems with constraints

Many physical problems (especially within the area of control theory)
require that an integral be maximised or minimised subject to
constraints. Such problems are often called isoperimetric, the name
given to the problem of finding the closed curve with a given fixed
perimeter which bounds a maximum area. The solution of this
problem is known to be a circle, a result which may be proved using
the following analysis.
We require that
I[y{x)]=\X2f(x,y,y')dx (9.92)
(see (9.2)) be stationary subject to the integral constraint
J[y(x)]=Pg(x,y,y')dx = C, (9.93)
where g(x, y> y') is a given function and C is a given constant. As with
the conventional Lagrange multiplier approach for finding the station-
ary points of a function subject to a constraint, we now form the linear
combination
/ + Ag, (9.94)
where A is a constant (Lagrange) multiplier, and determine the
extremals of the integral
\X2 x. (9.95)
9.8 Problems with constraints 239
The Euler equation appropriate to (9.95) is
which when solved leads to extremal curves of the general form

y=y{x,X,CuC2), (9.97)
where Cx and C2 are constants. The three constants A, C\ and C2 are
uniquely determined by the three requirements that (9.97) passes
through the end points xx and x2, and that / has the prescribed value C
as given by (9.93).
Example 10 To find the extremal curve y =y(x) of the integral
I[y(x)]=\\y'2 + 2yy')dx, (9.98)
with y(0) = 0 and y(l) = 0, and where y(x) is subject to the constraint
f y(x)dx = h (9.99)
and to determine the stationary value of

As in (9.95) we now consider the integral
(y'2 ' + ky)dx. (9.100)

Jo
Since the integrand of (9.100) is explicitly independent of x, we find,
using (9.47), that the Euler equation is
y'2 + 2yy' + Ay - y' — (y'2 + 2yy' + ky) = constant. (9.101)

dy
Then
y'2-ky = Cy (9.102)
where C is a constant, so that
y' = \/(C + ky). (9.103)
Separating the variables and integrating, we have
^ j , (9.104)
where A is a constant.
Now _y = 0 when x = 0. Hence

2
kA2 _Cr
4 ~ A' (9.105)
Also y = 0 when x = 1. Hence

C
4) 2 = (9.106)
Solving (9. 105) anc1(9. ,106) gives

2
A = -\, C = A 716, (9.107)
and consequently
) = 4(^-2) = ). (9.108)
~16 4^
Now by (9..99) we require
f y{x) dx = - \ (x2 - x) dr = i (9.109)

Jo ^ Jo
from which we find A = — 4. The solution for the extremal curve is

therefore
y=x-x2. (9.110)
Inserting (9.110) into (9.98) we obtain the extremal value / = \. ^
9.9 Direct methods and eigenvalue problems

The Euler equations obtained by requiring integrals to be stationary
are, in general, non-linear ordinary differential equations. These
frequently require numerical solution. An alternative and more direct
approach to finding the minimum value of an integral was first
developed by Rayleigh and Ritz. Suppose we again consider for
simplicity the integral (9.2)
rf{x,y,y')dx, (9.111)
where y{xx) = a and y(x2) = /?, oc and /? being fixed. Now consider a
sequence of functions
yi(x),y2{x)y...yyn{x),..., (9.112)
each satisfying the end conditions yn{xx) = or, yn{x2) = /?, for all n, such
that
\im I[yn(x)] = I0 (9.113)
«—•<»
and
\imyn(x)=y0(x), (9.114)
where 70 is the minimum value of / and yo(x) is the solution of the
Euler equation producing the value 70. The sequence (9.112) is called a
minimum sequence. The simplest case is if a = )3 = 0, and then the end
conditions can be satisfied if each function yn is written as
yn(x) = Cx4>i{x) + C 2 0 2 (x) + . . . + Cn<t>n(x), (9.115)
where 0I(JC), . . . , (f>n{x) are n independent functions each satisfying
0/(î) = 0/(*2) = 0, and Clf . . . , Cn are arbitrary constants. Inserting
the nth member of the sequence (9.115) into (9.111) we find
I[yn{x)]=\Xlf(x,yn,y'n)dx (9.116)
= F(CU C2, . . . , Cn). (9.117)

By the usual methods of determining the maxima and minima, we now
find the stationary value of /[}>„(*)] by equating the n partial
derivatives of F with respect to Clf . . . , Cn to zero:
dF/dC1 = dF/3C2 = . . . = dF/dCn = 0. (9.118)
We now have n equations for the n values Cly . . . , Cn, and the form of
the extremal curve y =yn(x) can then be found from (9.115). Provided
the sequence converges to the correct extremal curve y=yo(x) as
n->oof we can obtain an increasingly accurate approximation to the
true stationary value 70.
Example 11 Consider
I[y(x)]=\\y'2 + xy)dx, (9.119)
withy(0) = 0,
Suppose we choose (p1(x)=x(l — x) since this satisfies the end
conditions. Then the first member of a minimum sequence is
y1(x) = C1x(l-x). (9.120)
Hence
(9.121)
and
yi(x)] = f [c\(\ - 2x)2 + d x ^ l - x)] dx (9.122)
= IC21 + &C1 = F(C1). (9.123)

Hence
(9.124)
gives
(9.125)
and with this value, (9.123) gives
I[yi(x)] = " ife. (9.126)

To find the next member of the sequence y2(x)> we now choose
02(*) = A:2(1 — X) which again satisfies the end conditions. Then
y2{x) = Cxx(\ -x) + C2x\\ - x) (9.127)
and
yi(x) = Cx(l - 2x) + C2(2JC - 3JC2). (9.128)
w e
Calculating /[)>2(*)L find
I[yi(x)] = ^C2 + iC22 + J d C 2 + ^ d + ^ C 2 . (9.129)
Hence
dF
^ - = §^ + ^2 + ^ = 0 (9.130)
and
^^ =^ + ^ 2 + ^ = 0. (9.131)
Solving these two equations for Cx and C2, we find

d = C2=-A, (9.132)
giving
y2(x) = -Ml-x)-T2X2(l-x) = Ux3-x). (9.133)

From (9.129) and (9.132), the extremal value of / is
/b 2 (*)] = -lfe. (9.134)

We see that the Euler equation for / in (9.119) is
*-£(2/) = 0 (9.135)
or
y" = \x. (9.136)
Integrating we have
y' = \x2 + A (9.137)
and
y = ±x3+Ax + B. (9.138)
Applying the end conditions gives B = 0, A = — -^ and hence we find
y = Ux3~x). (9.139)
This is precisely the function found in (9.133), and hence the value of /
in (9.134) obtained by the Rayleigh-Ritz method is exact in this case.
We emphasise that this depended on our choice of (p1 and 0 2 - F ° r this
simple example, the Euler equation gives the extremal curve very
easily, but nevertheless this example illustrates the use of the
Rayleigh-Ritz method when the value of the extremal curve is zero at
both ends. ^
If the end values of the curve are not both zero, then a linear
combination of the form (9.115) can still be written but we must
ensure that each yn(x) still satisfies the end conditions. This usually
results in relationships between the arbitrary constants, thereby
reducing the number of independent constants to less than n. The
stationary value of I[yn] is then found by minimising with respect to
the independent constants remaining. We give an example to illustrate
this.
Example 12 Consider
I[y(x)]= f {y'2 + xy2)dx, (9.140)

Jo
with>>(0) = 0,
Suppose we choose (/>i(x) = x and $ 2 (*) = *2- K we first put
yi(x) = Ci0!(x) so that )>i(0) = 0 then we must have CX = 1 in order
that yl(l) = l. This choice leaves no free constants. Next if we put
yi(x) = Cxx + C2x2, (9.141)
which immediately satisfies y2(0) = 0, then at x = 1, we have
y2(l) = l = C1 + C2, (9.142)
so that
C2=l-C1. (9.143)
Hence
y2(x) = C,x + (1 - Cx)x2, (9.144)
which satisfies the end conditions for all Cx. Using (9.144), we find
I[y2(x)] = [ { [ d + 2(1 - Cx)x]2 + x[Cxx + (1 - CX)JC2]2} dx, (9.145)

h
and integrating gives
(9.146)
Hence
dF/dC1 = jfcd - | = 0, (9.147)
giving
d =f (9.148)
and the extremal value
I[y2(x)] = §g. (9.149)
We have therefore generated an approximation to the extremal value
and an approximation to the extremal curve, which from (9.144) and
(9.148) is
yi(x) = 7* + 7*2. (9.150)
We note that the Euler equation corresponding to (9.140) is simply
y" = xy, which has a solution in terms of Airy functions (see Section
2.9). However, although the extremal curve can be found by applying
the end conditions, the evaluation of (9.140) is difficult. ^
We conclude this section with an example to illustrate the use of the

method in evaluating eigenvalues.
Example 13 Consider the simple equation

d2y/dx2 + Ay = 0, (9.151)
subject to the boundary conditions y(0) = 0, y(l) = 0.
The exact solution of (9.151) is
y = A cos(VA X) + B sin(VX X). (9.152)
Imposing the boundary conditions, we find A = 0 and
X = n2jz2, (9.153)
where n = 1, 2, 3, . . . . These values of A (an infinite discrete set) are
called the eigenvalues of the equation. Now consider
I[y(x)]=l\y'2-ky2)dx, (9.154)
which has (9.151) as its Euler equation. We choose some trial function
passing through the end points, say
x). (9.155)
Then
IbiM] = C2S - Â) = F ( d ) . (9.156)
Hence
dF/dd = 0 = 2 d G " Â), (9.157)
which gives A = 10 as an estimate for the lowest eigenvalue. This is to
be compared with the exact value A = ;r 2 ~9.87 obtained by putting
n = \ in (9.153).
In order to improve the approximation, we choose, say
y2(x) = d*(l - x) + C2x\\ - x) (9.158)
which also passes through the end points. We now find
Kyiix)] = \c\ + \cxc2 + ic22 - A[ic? + i d d + ifecl] = F(d, d).
(9.159)
Hence
5 F / 3 d = (f " * A ) d + G ~ *A)C2 = 0 (9.160)
and
9F/dC2 = G - i A ) d + ( B - ifeA)C2 = 0. (9.161)
Equations (9.160) and (9.161) are a pair of homogeneous linear
equations for Cx and C2. For a non-trivial solution we require that the
determinant of the coefficients is zero. This gives
A2 - 52A + 420 = (A - 42)(A - 10) = 0 (9.162)
and hence
A = 10, A = 42. (9.163)
The first of these is the value we found above corresponding to n = 1
(for which the exact value is n2^ 9.87), while the second corresponds
to n = 2 (and an exact value of A = 4JT2 ~ 39.48). ^
Problems 9
1. Find the extremal curves and the stationary values of the following
integrals:
(i)
where P = (0, 0) and Q = (JZ/2, JT/2),
rQ
(ii) [2ysmx + (dyldx)2]dx,
JP
where P = (0, JT) and Q = (JT, 0).

2. Find the extremal curve of the integral
where P = (0, 0) and Q = (1, 2).

3. Find the extremal curves of the integral
Q
y'2
IT?*>
where P = (0, 0) and Q = (1, 2).
4. Find the function y(x) which makes the following integral
stationary:
Jo
where y(0) = 0, y(jt/2) = 1, and y(x) is subject to the constraint

rJt/2
J 1
o
y dx = JT/2 - 1.
5. Minimise the integral
where the function jc(r) is subject to the constraint
V[l + {dxldtf] dt = constant.
Show that x(t)=A + Bcosh[(t-C)/B], where A, B and C are

constants.
Problems 9 247
6. A curve y =y(x) meets the x-axis at x = ±ay and has length na

between these points. Show that the curve which encloses maxi-
mum area between itself and the x-axis is the semicircle x2 4- y2 = a2
7. Theflight-pathof an aircraft lies in the (x, y) plane, and the aircraft

moves with a speed v(y) (that is, a function of y only). If its path
(assumed to be smooth) between two fixed points is such as to
minimise the flight-time, show that the equation of the path is given
by
v(y)dy , „
-I
where A and B are constants.
8. A particle moving in the (JC, y) plane has a speed u(y) depending
only on its distance from the jc-axis; its direction of motion makes
an angle 6 with the jc-axis which can be controlled to give the
minimum time of transit between two points. If u(y) = uoe~y/h,
where u0 and h are constants, and if the particle, starting at x = 0,
y = 0, is required to reach {\jzh, h) in least time, prove that its
initial and final directions must be 60 = tan - 1 (l — V2/e) and
^1 = tan" 1 (eV2-l).
9. Find a functional which has
y" + y+x = 0
as its Euler equation. Hence obtain an approximate solution of the
equation given y(0) =y(l) = 0 by using the trial functions
(i) y = Cx(l-x),
(ii)y = Cxx{l-x) + C2x2(\-x)y
and finding the values of C, Cx and C2.
ANSWERS TO PROBLEMS
Chapter 1
1. auxi + a22xl + a33xj + 2{aX3xxx3 + ai2x1x2 + a23x2x3).
2. (i) j , (ii) none, (iii) i and /', (iv) i on each side.
3. 3, 3, all three components zero, all three components zero.
4. 26,1, 6.
5. S,, = -
7. c,y = 5,yapap - a^j.

9. 32fk/3xr dxs = 2drsxk
10. or = 4,(i) j8 = 7, (ii) 0 = 11.
Chapter 2
1. (i) T(5) = 24, (ii) V(w/a).
4. 2w/3V3, using (2.25) and (2.27).
6. (i) |B(|, | ) , (ii) w/V2, using (2.25) and (2.27).
9. AJv(ex) + BYv(ex).
10. —r- (A cos x + B sin JC).
12.
13. (i) 0, (ii) -I
Chapter 3
2. 4/(x-3)2.
3. _y = x In Ax.
Cx + 1
In A:
4. y = I n
x+C
where A is an arbitrary constant.

248
Chapter 4
1.
3. Jt+x 3 + | r \
4. yx = 2 + x2y zx = 3x2,
y2 = 2 + x2 + x3, z2 = 3x2 4- \x
>>3 = 2 + jt 2 4 - x 3 4-|>jt 5 4 - ^ j t 6 ,
z3 = 3JC2 4- |JC4 4- %x5 4- 4 * 7 + i
5. cos t 4- ju[|r cos t — ^ sin t — ^
6. e"*3/6[,4 COS(2JC) + 5 sin(2x)].
7. i
8-
Chapter 5
2. (i) \ In 2 4- i(-JT/4 + 2A;JT), where A: = 0, ±1, ±2, . . . ,
(ii) nil + kjt + (i/2) In 3, where k = 0, ±1, ±2,
3. i(jr/2 + A:JT), where k = 0, ±1, ±2,
4. u = 2jry + Ay - 2x 4- C, /(z) = z2 + (4 - 2i)z + C.
5. g and /i constant.
6. All parts 5 4- f i; z 2 4-1 is analytic for all z.
7. (i) (a) 0, (b) 2m,
(ii) -2jri, jri/2.
8. (i) 2;risinhl, (ii) 4^:ie/3.
9. (i) In, (ii) 2m/9, (iii) 2jri.
10. (i) l + z 2 + ^z 4 + . . . ( / ? = oo),
(ii) sin 1 4- (cos l)z + (2 cos 1 - sin l)z 2 /2 4- . . . ( / ? = 1),
(iii) In 2 4- ^z + \z2 + . . . (R = JT).
11. (i) 1 — ——2 + 7^—4 "" • • • (essential singularity at z = 0),
(ii) l/(z -n)- 2(z - n) + \(z - rif - . . . (simple pole at z = JT),
• • •] (double pole
at z = n/2).
250 Answers to problems
12. (i) -
1 2 3 4
13. (i) - 1 ,
00 1,
(iii) simple pole at z = 2, residue | ,
double pole at z = 1, residue 2,
pole of order 3 at z = 0, residue — ^ .
14. 2kjt — Una, where k = 0, ± 1 , ±2, . . . .
Chapter 6
1. (i) r, (ii) (4-2V3
3. 0) . (») JT, (iii) iTT, (iv) \n.
6.
U
JV2>
8. J = 0.
9. e™(
11. (i) i i cosech2 n + iJT COth JT,
(ii) TIC
Chapter 7
3
s
1. (i) (ii)
+ a2][(s-\-af + a2]'
i
2
3. (i) i[cos*-cos(2*)], (ii) - l + e*-*e* + i

4. (i) \, (ii) i (iii) 1/V2.
6. /(*) = * +1/W*.
7. sinjc.
9. (i) 6e^ - 7c-21 + 2e~3x,
(ii) l//(x - 2){1 - cos[2(* - 2)]} + COS(2JC)
(iii) e-^1 + e"H(x - a)].
10. x(t) = -l + te' + e',y(t) = 2 + te' - e'.
11. y(x) = C^~l{(l-s)x/s1+k}> where C is a constant; when A = 2,

2
y = C(l - 2x + i* ).
12. (i) |e~' + ê 2 ' - £e2', (ii) ê"'(l - It2) + êf(2r - 1).
13. i* + § sin(3*).
14. ^/(jt/X)e~s2/4;
is. *?y-y>.
16. (i)
(ii) (^/2) sgn5, the cosine transform does not exist.

1
18.
2JT(1+JC 2 )'
Chapter 8
(iii) w = - l / l n ^ V ) , (iv) w = - sinh(jc2 + y).
2. 0 = ^( ; y )
3. M = e " ^ / ^ + i>^) 4- g{x — iy)], f, g arbitrary functions.
2 2
4. w = j t V
5 . 0 = i(
7. r = fl2
in t , 2a2 " ( - l ) n . /n^r\ . /njrcA
10. w(r, 0 = - - 2 ~ Z 1 - ^ L s i n sin .
nrcr n=i n \ a / \ a )
13 M(
- ^» = 2V(b)f,coshva^^ 2 ) c o s h [ V ( 1 + s 2 ) y ] e i
14. u(x,t) = - ^
where
a )
252 Answers to problems
Chapter 9
1. (i) y=x,/ = k ( l + i^ 2 ),
(ii) y = n — x — sin xy I = §JT.
2. y = 2x.
3. y = sinh(jcsinh~12).
4. y = 1 — COSJC.
9. (i) C = 4 ( i i ) ^ = ^ ^ 2 = * .
Index
Airy equation 50, 64, 87 constraints 238-40

Airy functions 50, 89, 244 contour integration 98-136
alternating symbol 5-6 contravariant vectors 15-18
analytic functions 103 convolution theorem 164-6, 185
associated Legendre functions 47, 213 coordinate systems 9-13
asymptotic series 30 coordinate transformations 13-14
covariant vectors 16-18
curl of a vector 8-11, 13
Bernoulli's equation 59-61 cycloid 232
Bessel equation 22, 31, 44, 213 cylindrical polar coordinates 10-12
Bessel functions 22, 31-45, 181-2, 213,
215-16
Bessel functions (modified) 42-5 differential equations
beta function 26 non-linear 53-73
biharmonic equation 192 approximate solution of 75-96
Bohr-Sommerfeld formula 96 partial 190-218
boundary conditions 194 diffusion equation 192, 208-11, 214-18
brachistochrone 231 Dirac delta functions 170-1
branch points 100, 150-4, 179-80 direct methods (in the calculus of
branches of a complex function 100, 108 variations) 240-5
Bromwich contour 175-80, 216 Dirichlet conditions 194
divergence of a vector 8-13
Duffing's equation 68-73
calculus of variations 224-45 dummy indices 3-4
canonic form 203
catenary 233
catenoid 233 eigenvalues 88-90, 244-5
Cauchy conditions 194 elliptic equations 193-4
Cauchy Integral Formula 116 elliptic integrals 67-70
Cauchy Residue Theorem 131-4 error function 28-31, 167-9
Cauchy-Riemann equations 105 Euler's constant 26, 44
Cauchy's Theorem 112-16 Euler's equation 202-5
characteristics 195-205 in the calculus of variations 225-35
C/function 28 Euler's formula 99, 101
complementary error function 29-31, exponential integral function 28
167-9,215, 217-18 extremal curve 227, 233
complex derivatives 103-8
complex functions 98-136
complex integration 109-12 factorial function 23
complex numbers 98-9 Fourier cosine transform 183-6
253
254 Index
Fourier sine transform 184-6, 217 Lagrange's equations 234-7

Fourier transform 161,183-7 Lagrangian 235
Fredholm equation 186 Laguerre polynomials 48
free indices 3-4 Lane-Emden equation 65-6
Frobenius series 31-6, 78-9 Laplace inversion 173-80
functional 224 Laplace transform 161-83
functions Laplace's equation 192
Airy 50, 89, 244 Laurent series 120-5
associated Legendre 47, 213 Laurent's Theorem 121-5
Bessel 22, 31-45, 181-2, 213, 215-16 Legendre polynomials 45-7
beta 26 Liouville-Green technique 90-6
a 28 Lipschitz condition 82
complementary error function 29-31, logarithmic (complex) function 102
167-9, 215, 217-18 Lommel integrals 45
Dirac delta 170-1
error 28-31,167-9,216-18
exponential (complex) 101-2 metric 19
exponential integral {Ex) 28 minimum sequence 241
factorial 23 modified Bessel functions 42-5
gamma 23-8 multi-valued functions 100, 108
Heaviside step 169-71
incomplete gamma 26-7
Legendre (associated) 47, 213 Neumann conditions 194
logarithmic (complex) 102 Neumann function 37
Neumann 37 non-linear differential equations 53-73
Si 28 non-linear pendulum 66-8, 235-6
sign 167 normal form 83-4, 91
Weber 37
Weber-Hermite 49, 88
ordinary differential equations
special functions 22-50
non-linear 53-73
gamma function 23-8 approximate solution of 75-96
incomplete 26-7 orthogonal coordinate systems 9-13
generalised coordinates 234-6
generating function 39, 44, 46, 48-9
gradient of a scalar 8-13 parabolic equations 193-4, 203
Green's Theorem 113 partial differential equations 190-218
perturbation series 82-3
phase-plane 70-3
heat conduction equation 192, 208-11, Picard iterative method 79-82
214-18 Poisson equation 192
Heaviside step function 169-71 poles 125-34
Helmholtz equation 192 principal part (of a Laurent series)
Hermite polynomials 48-9, 88 122-5
hyperbolic equations 193-4, 204 principal value
of a complex function 102
of an integral 146-50
incomplete gamma function 26-7 Principle of Superposition 53, 205
integral equations 182-3, 186-7
integral transforms 161,214-18
integration (contour) 98-134, 137-58 Rayleigh-Ritz method 240-3
inverse transforms 162, 173-80, 183-4 recurrence relations 23-4, 39-41, 45-6,
isolated singularity 121 48-9
reflection formula 26
regular functions 104
kernel 161 residue theorem (Cauchy) 131-4
Klein-Gordon equation 192 residues (calculation of) 125-31
Kronecker delta 4-5, 15-16 Ricatti equation 61-4, 80
Index 255
scalar 16-17, 19 Taylor series 120-5

scalar (dot) product of vectors 1 Taylor's Theorem 120-1
Schrodinger equation 192, 211-14 tensor algebra 1-19
Schwarzian derivative 91
separation of variables 205-14
shift theorem 163-4
Si function 28 van der Pol equation 73
sign function 167 vector operators 8-9
simple poles 126-9 vector product 6
singularities 104, 107, 120-34 Volterra equation 182-3
isolated 121
special functions 22-50
spherical polar coordinates 12-13
Stirling's formula 27-8 wave equation 191-2, 204
Sturm-Liouville form 89-90 Weber function 37
suffix notation 1-19 Weber-Hermite function 49, 88
summation convention 1-3 Wentzel-Kramers-Brillouin (WKB)
summation of series 154-8 method 84-8

G Stephenson - P M Radmore - Advanced Mathematical Methods For Engineering and Science Students-Cambridge University Press (1990) PDF

Uploaded by

Copyright:

Available Formats

G Stephenson - P M Radmore - Advanced Mathematical Methods For Engineering and Science Students-Cambridge University Press (1990) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

G Stephenson - P M Radmore - Advanced Mathematical Methods For Engineering and Science Students-Cambridge University Press (1990) PDF

Uploaded by

Copyright:

Available Formats

Advanced mathematical methods for

engineering and science students

© Cambridge University Press 1990

First published 1990

British Library Cataloguing in Publication Data

Library of Congress Cataloguing in Publication Data

ISBN 0 521 36312 8 hardback

Transferred to digital printing 2002

3.3 Equations reducible to linear form 57

6 Applications of contour integration 137

7 Laplace and Fourier transforms 161

7.2 The Laplace transform 162

Partial differential equations 190

Calculus of variations 224

Answers to problems 248

This book contains a selection of advanced topics suitable for final

1. The symbol In denotes logarithm to base e.

1.1 Summation convention

Instead of writing the summation sign in (1.1) every time we have an

where summation over i is implied (i = 1, 2, 3). Since the components

Example 1 If vectors c and d have components c, and dt (i = 1, 2, 3)

asxs = alx1 + a2x2 = a^. (1.5)

that is, the order of the factors is unimportant and summation is

( t «/*.-)(2 "ft) = Wjxj, (1-13)

1.2 Free and dummy indices

021*1 + 022*2 + . . . + a2nXn = C2y

By introducing the index i, (1.15) can be expressed in the more

dtjXj = ciy (/ = 1, 2, . . . , n). (1.16)

Using the summation convention, we can write (1.16) as

Example 5 Suppose we are given constants atj (i, j = 1, 2, 3) and the

-fl = aij-£LXj + aijXi-^1. (1.18)

— = asiXi + disXi = (asi + a^xt, (1.20)

1.3 Special symbols

Further consider the expression SrsApqs, where r, /? and g are free

SrsApq5 = SrlApgl + < 5 r 2 ^ 2 + Sr3Apq3. (1.28)

2. TTre alternating symbol

6 ^ = e,-. (ey X efc). (1-30)

+ 1 if /, y, k are a cyclic permutation of 1, 2, 3,

Hence cyclically permuting the indices i, j , k leaves eijk unaffected,

1.4 Vector identities

Now since for any vector A, A = (A . e m )e m (as in (1.3)),

Example 6 A matrix 0 has elements

(02)*y = 0/*0*y = («/«* + £ukni)(nknj + €kmjnm) (1.53)

Using (1.50), (1.56) becomes

1.5 Vector operators

and the vector operator V by

Various identities involving the vector operator can be derived using

where in the last term we have used (1.38) with a = V0. ^

1.6 Orthogonal coordinate systems

infinitesimal distance ds between these two points is given by

ds2 = h\ dq2 + hi dqi + h2 dq\ = h2 dq2f (1.86)

V2O^ i I" a (h2h3d<t>\ | d /hxh3d<s>\ ( d /M 2 a<s>\-|

1. Cylindrical polar coordinates

cartesian and cylindrical polar coordinates are

e,, = e,, e^ = e2, e, =

2. Spherical polar coordinates

= sin 6 cos (f)dr + r cos 6 cos 0 dO — r sin 6 sin (p a (1.105)

Hence, from (1.87)-(1.90),

V2<$> =2 -r-—[sintf — (r2 — ) + — (sin0 — ) + ^ — 2^ 1 V (1.113)

0211 + 0222 + . . . + a2nXn = C2y

(02)y = 0/0y = («/« + £ukni)(nknj + €kmjnm) (1.53)

i=fi(xs), (,i = l , 2 , . . . , / i ) . (1.117)