Sunteți pe pagina 1din 177

.

Complex Variable Methods


1n

Science and Technology

JOHN CUNNINGHAM
M.A., M.SC., PH.D.

Lecturer in Applied Mathematics


University College of North Wales

D. VAN NOSTRAND COMPANY LTD


LONDON
NEW YORK

TORONTO
PRINCETON, NEW JERSEY

D. VAN NOSTRAND COMPANY LTD.


358 Kensington High Street, London, W.14
D. VAN NOSTRAND COMPANY INC.
120 Alexander Street, Princeton, New Jersey
24 West 40th Street, New York 18
D. VAN NOSTRAND COMPANY (CANADA) LTD.
25 Hollinger Road, Toronto 16

Copyright 1965
JOHN CUNNINGHAM

Library of Congress Catalog Card No. 65-20159

PRINTED IN GREAT BRITAIN BY


J. W. ARROWSMITH LTD., BRISTOL

Preface
This book is based largely on lectures delivered by the author at the
University College of North Wales to honours undergraduates in the
departments

of

Applied

Mathematics

and

Electronic

Engineering

during their second and third years.


The purpose of the text is to provide students who are not primarily
pure mathematicians with the basic tools of complex analysis for use
in the theoretical study of physical

problems.

Throughout the work

undue emphasis on mathematical rigour has been avoided: the theory


is built up step by step and the main ideas are backed up by proofs
and plausibility arguments which are illustrated by numerous diagrams.
The author believes that the surest method of acquiring mathematical
skills is to study examples and then to try to apply . the methods exem
plified to fresh problems. The structure of the book is based on this
plan: the text itself deals in depth with essentials only: the examples
then demonstrate the theory in action: finally, the problems, for which
hints and answers are provided, give the reader an opportunity to
discover for himself many interesting results of complex variable theory
and become expert in the application of this powerful mathematical tool.
Thus the book offers in a concise form a complete course in the
theory and practice of an important and exciting mathematical method.
U.C.N.W.

J.C.

1964

Acknowledgements
The author wishes to thank his wife for her assistance in the preparation
of the original typescript from the author's lecture notes.
The author is grateful to the examining bodies of the undernoted
Universities for their permission to reproduce examination questions
set by them:
The University of Durham
The University of Edinburgh
The University of Hull
The University of Sheffield
The University of Wales (U.C.N.W., Bangor).
When known, the source of each question has been appended to it
together with the date of the examination. The responsibility for any
errors in the answers or solutions provided rests with the author.

vi

Contents

Preface
CHAPTER

CHAPTER 2

CHAPTER 3

CHAPTER 4

CHAPTER 5

REAL VARIABLE THEORY


1.1

Functions of One Variable

l.2

Functions of Two Variables

1.3

Partial Differentiation

l.4

Jacobians

11

l.5

Multiple Integrals

12

l.6

Repeated Integrals

15

l.7

Change of Variables

19

l.8

Green's Theorem

21

COMPLEX NUMBERS
2.1

The Square Root of Minus One

2.2

The Argand Diagram

28

2.3

Complt::x Conjugation

30

27

2.4

De Moivre's Theorem

32

2.5

Formal Considerations

37

ANALYTICITY
3.1

General Principle of Convergence

40

3.2

Continuity

43

3.3

Differentiation

45

3.4

The Exponential Function

51

3.5

Potential Problems

54

3.6

The Logarithmic Function

55

CONFORMAL TRANSFORMATIONS
4.1

Mapping

60

4.2

Joukowski's Transformation

63

4.3

The Schwarz-Christoffel Theorem

67

INTEGRATION
74

5.1

The Limit of a Sum

5.2

Cauchy's Theorem

76

5.3

Cauchy's Integral Formulae

79

5.4

Connection with Differentiation

87

5.5

The Theorem of Residues

88

5.6

Trigonometric Integrals

93

Vil

viii
CHAPTER 6

CHAPTER 7

CHAPTER 8

CHAPTER 9

CONTENTS
IMPROPER INTEGRALS
6.1

Infinite Integrals

6.2

Infinite Integrands

101

6.3

The Evaluation of Infinite Integrals

102

6.4

Jordan's Inequality

106

6.5

Expansion in Series

110

99

MANY-VALUED FUNCTIONS
7.1

Logarithmic Branch Points

114

7.2

The Principle of the Argument

117

7.3

Rouche's Theorem

121

7.4

Powers of

123

7.5

Integrals Involving Logarithms

127

BETA, GAMMA AND DELTA FUNCTIONS


8.1

Convergence

8.2

The Factorial Function

137

8.3

Analytic Continuation

139
141

135

8.4

The Continuation of the Factorial Function

8.5

The Gamma Function

142

8.6

The Hankel Function

143

8.7

The Beta Function

146

8.8

The Dirac Delta Function

148

DIFFERENTIAL EQUATIONS
9.1

Physical Considerations

9.2

Solution in Series

160

9.3

The Formulae of Rodrigue and Schlafii

164

9.4

Contour Integral Solutions

169

156

References

176

Index

177

CHAPTER 1

Real Variable Theory


It is difficult to embark on any systematic treatment of complex variable
theory without first studying the properties of functions of real variables.
Thus, in this preliminary chapter, we give an elementary discussion of
some of the aspects of real variable theory. Our main object is to discuss
the concept of continuity and to establish certain results from the
theories of partial differentiation and multiple integration.

1.1

Functions of One Variable

The reader is no doubt familiar with the notion that continuity has
something to do with 'smoothness'. If we draw the graph of a plane
curve y = f(x) and assert thatf(x) is a continuous function of x between
the points x =a and x

==

b, then we mean that y =f(x) 'goes smoothly'

from x =a to x = b. The curve will have some such shape as that


drawn in Fig. 1.1, and certainly not the sort of shape shown in Fig. 1.2.

FIG. I.I

Fm. 1.2

Let us make this idea a little more precise. We say thatf(x) is continuous
at x = x0 if f(x) is uniquely defined at x = x0, and as x tends towards
x0 in any manner, f(x) tends towards the value f(x0). More concisely

[Ch. 1

REAL VARIABLE THEORY

we may write this definition in the following manner:

Limf(x)=f(xo)

(1.1.1)

For the function drawn in Fig. 1.2 the above limit is not uniquely
defined and so f(x) is clearly not a continuous function of

at

x= x0.

Let us now introduce the sophistication of right-handed and left


handed limits. This is to say, let us define limits in which

approaches

x0 only from the right and only from the left respectively. We shall

x -+ x0 + and x -+ x0 - and, assuming that the


appropriate limits actually exist, we write:

denote such limits as

Lim

f(x)= Lim f(x0 +e )=f(xo +)

(1.1.2)

Lim

f(x)= Lim f(x0 - e)=f(x0 - )

(1.1.3)

x-+xo+

x-+xo-

where

-+0

e-+0

is a positive quantity.

If f(x) is actually continuous at

x= x0,

as is the case in the example

of Fig. 1.1, then it is always true that f(x0 +)=f(x0 - ).

Let us consider a specific example. A function which commonly

occurs in mathematical physics is the so called 'step function' which is


often defined as follows :

e(x)= +1,

for

x>O

e(x)= 0,

for

x=O

e(x)= -1,

for

<

0.

The shape of this function is shown in Fig. 1.3.

----y
---1 -1

Fm. 1.3

Clearly the step function, although uniquely defined at


continuous at

x= 0.

For, if we let

x= 0,

is not

tend towards zero from the right

Sec. 1.1)

FUNCTIONS OF ONE VARIABLE

and from the left respectively, we obtain


Lim e{x) =e{O+) =+1
X-+0+

Lim e{x) =e(O-) = -1.


xo-

These two values are unequal and consequently the function is


discontinuous. Incidentally, neither value is equal to that which we have
chosen as the value e{O).
It is now a fairly easy step to extend our ideas of continuity at a point
to continuity in an interval. By saying that x lies in an interval {a,b) we
mean that x may assume any value between x =a and x =b. We may
or may not include the end points of the interval, i.e. the points x =a

and x =b. Let us call an interval a< x<b which explicitly excludes
the end points x =a and x =b, the open interval (a,b): if we do include
the end points we call a x b the closed interval (a,b). The closed
interval {a,b) consists of the open interval {a,b) together with both end
points x =a and x =b. Now, by saying that f{x) is continuous in an
interval {a,b) we are implying that f{x) is continuous at each point of
(a,b). If the interval is open we have no difficulty in making this definition
precise. For a closed interval, however, the end points are rather different
from the other points of the interval: we require that, at x =a, f(x) be
continuous only on the right, and at x =b, only on the left. That is to
say: if f(x) is continuous at each point x0 such that a<x0<b, and if
the limits f(a+) and f(b-) both exist and are unique, then {provided
we take these unique limiting values to be the definition of the function
f(x) at the end points of {a,b))f{x) is said to be continuous in the closed
interval {a,b).
This point may be clarified by studying the example of the step
function e{x) which is depicted in Fig. 1.3. The reader will readily convince
himself that e{x) is not continuous in any of the intervals -1 x 0,
-1 x + 1, 0 x +1 because at x =0 neither the left-handed
nor the right-handed limit {though they both exist) is equal to the value
of the function at x =0.
If, however, we define a new step function e0>(x) as follows:
e(ll(x) =+1,

for

x;;:,: 0

e<1>(x) = -1,

for

x<O

then, although not continuous in the intervals -1 x 0, -1 x +1,


it is continuous in 0 x + 1 because the right-handed limit does
exist, is unique, and is equal to the value of the function at x =0.

[Ch.

REAL VARIABLE T'ffEOR Y

Similarly a step function

1P>(x)

would be continuous in

>
1P(x)
defined as:
=

+l,

-1x0

for

for

xO

> 0

but not in

-1 x + 1

nor in

Ox +l.
We conclude this section by discussing two important theorems
concerning continuous functions.
THEOREM:

If f(x) is a continuous function of x in the closed interval (a,b),


then a point x0, a x0 b, can be found such that the derivative f'(x0),
if it exists, is given by
'
f (Xo )

f(b)-f(a)
b-a

This theorem is usually known as the

(1.1.4)

mean-value theorem and plays a

central role in the differential calculus. Its proof is beyond the scope of
our present brief discussion but its content is rendered abundantly clear
from graphical considerations (see Fig.

1.4).

110

FIG. l.4

The expression (
while

f(b)-f(a))/(b-a) is just the gradient of the chord AB


f'(x0) is the gradient of the tangent at P. Thus our intuitive notions

of continuity imply that at least one point P exists between A and B on


the arc such that the tangent at P is parallel to AB.
Before proceeding to the enunciation of our second theorem, which is
a trifle more subtle, the author would like to point out certain features
of the behaviour of continuous functions in specific examples.
Consider the trigonometric sine function y
in the closed interval 0x n (see Fig.
function in the interval is

1.5).

sin

x which is continuous

The greatest value of the

1, occurring at x
n/2 and the least
x
0 and at x
n. We say that, in the
closed interval (O,n), sin xis bounded above by 1 and bounded below by 0.
If we had considered the same function y
sin x in the open interval
value is

0, occurring at

Sec. 1.1]

FUNCTIONS OF ONE VARIABLE

x;r/2

x=T

FIG. 1.5

(O,n) it is still true that sin x is bounded above by 1 and bounded below
by 0. There is, however, this difference: in the first case the function
actually takes on the values 0 and 1 while in the second case, since the
points x = 0 and x = n are explicitly excluded, the lower bound 0 is
never actually reached in the interval.
Let us take another example of a trigonometric function-y
tan x
which is graphed in Fig. 1.6. Now tan n/2 = ro so that the limit
=

xT/2
FIG. 1.6

Lim tan x does not exist. So although tan x is continuous in every

x-+(n/2)

closed interval 0 x a where a < n/2 it is not continuous in the closed


interval 0 x n/2. Furthermore it is bounded below in the closed
/
interval (O,n/2) but above it is unbounded.
The following theorem on the bounds of continuous functions can be
proved.
If a function f(x) is continuous in a closed interval (a,b) then it
is bounded both above and below and the bounds are values actually taken
on by f (x) somewhere in a x b.
THEOREM:

[Ch. 1

REAL VARIABLE THEORY

In this statement we have assumed implicitly that if a function is


bounded above then there exists a least bound which we call the least
upper bound or more simply just the upper bound. Similarly we have

assumed that a function which is bounded below has a greatest lower


bound which we call the lower bound. This assumption is by no means

obvious but for our purposes a fuller discussion is not necessary.


The reader is urged to return to the examples y
and y

sin x in 0 x n

tan x in 0 x n/ 2 and to rediscuss these functions in the

light of the above theorem.


1.2

Functions of Two Variables

For a function of two real variables we say that f(x,y) is continuous


at (x0,y0) if/(x0,y0) is uniquely defined and/(x,y) tends towards/(x0,y0)
as (x,y) tends towards (x0,y0) in any manner. In concise mathematical
language we write
Lim

(x,y)-+(xo,yo)

The functionf(x,y)

/(x,y)

f(x0,Yo).

(1.2.1)

(x2-y2)/(x2 + y2) provides us with an instructive

example of a function of two variables. Let us examine the limit off(x,y)


as x and y both tend towards zero. If we let y tend towards zero and
then x we obtain

x2-y2
Lim --y-+0 X2 + y2
x-+0

x2
Lim2
x-+O X

1.

(1.2.2)

Reversing the order of the limiting process we find that the result is
quite different
x2-y2
Lim --0 X 2 + Y2
X-+0
y-+

-y2
Lim -y-+0 Y2

-1.

(1.2.3)

Thus the order of taking limits is critically important and the function
is not continuous in the sense of (1.2.1).
The author wishes to emphasize tnat the definition (1.2.1) of continuity
in two real variables is very much stronger than continuity in each
variable separately.

The reader should satisfy

himself that, while

(x2 - y2)/(x2 + y2) is not continuous in the two variables x and y at the
origin, it is continuous in x and y separately when they are zero. That
is to say that (x2 -a2)/(x2 + a2) and (b2 - y2)/(b2 + y2), where a and b are
constants, are continuous functions of x and y respectively.
By analogy with single variable theory we can now introduce the
concepts of continuity in open and closed regions.

Sec. 1.3]

PARTIAL DIFFER ENTIATIO N

For example the open region x2 + y2

< 1 comprises all the interior


1, and the closed region x2 + y2 1
consists of the open region x2 + y2 < 1 together with its boundary points,
namely the points of the circle x2 + y2
l.
A function f (x,y) is said to be continuous in an open region if it is

points of the circle x2 + y2

continuous at each point of the region. When we talk of continuity in a


closed region we mean that f(x,y) is continuous at each point of the
corresponding open region and that, on the boundary, the definition

(l.2.1) is satisfied for all limits in which (x,y) --. (x0,y0} from interior
points of the region.
There is only one instance when we shall require to distinguish
between open and closed regions in our present discussions. It is in
connection with the two dimensional analogue of the theorem on bounds
stated in Section 1.1 which we now enunciate without proof.
THEOREM: If f(x,y) is a continuous function of two real variables x and y
in a closed region A then the function is bounded both above and below
and f(x,y) actually assumes the values of the bounds somewhere in A.
1.3

Partial Differentiation

Suppose that we have a functionf(x,y,z, ... ) of several real variables


and that we wish to differentiate with respect to one of them treating
the other variables as if they were constant. This is the operation which
we usually call partial differentiation.
The partial derivative of f(x,y,z, . . .) with respect to x is defined, at
the point (x0,y0,z0,

), as
fixo,Yo,zo,

x-+xo

=Lim

f(X,Yo,Zo,

.) =

[0f]x=xo
y=yo
z=zo
X

.)-f(Xo,Yo,zo,
X-X0

.)

(1.3.l)

In partial differentiation care must be exercised o avoid the confusion


which can arise when there is more than one set of independent variables.
This point is illustrated by means of an example.
We shall discuss z = x2 -y2 where z is a function of the independent
variables x and y. We call z the dependent variable, since if we fix the
values of x and y arbitrarily then z is fixed through the functional
dependence of z upon x and y. Suppose, now, that u

x+ y and v

x-y.

It is then possible to write own a functional dependence of z upon the


independent variables u and v, namely z = uv. Indeed it is a simple

REAL VARIABLE THEORY

[Ch. 1

matter to express z in a variety of ways: two of these are:


z =x2-y2
z

2xu-u2.

(1.3.2)

Now calculating oz/ox from these two expressions we obtain respec


tively
oz
-=2x
ox
oz

(1.3.3)

=2u =2(x+ y).


ox
It is evident then, that in general we do not obtain the same result
when different variables are held constant, so that we should write

(::)Y
(!:)"

=2x
(1.3.4)
=2(x+ y)

the suffix denoting that variable which has been held constant. In the
first case x and y have been treated as the independent variables and
in the second x and u.
In most examples it will be clear from the context what the derivatives
mean and often we will dispense with suffixes: but, at all costs, confusion
must be avoided and the reader is advised to use suffixes until he has
become experienced in the manipulation of partial derivatives.
Even in the theory of functions of a single real variable partial deriv
atives play a significant part. For example we might be faced with a
function f(x,y), ostensibly a function of two variables, but actually a
function of.one variable u because of functional dependences of x and y
upon u. Under the assumption that f(x,y) is continuous in each variable
separately and that x and y are continuous functions of u, x = x(u),
y =y(u), we can prove that
df
of dx of dy
-=- -+- . du
ox' du oy du
providing that the relevant derivatives exist and that fx and
continuous functions.

(1.3.5)

fy

are

Sec. 1.3]

PARTIAL DIFFER ENTIATIO N

total derivative df/du


u = u0 which corresponds to the point (x0,y0)

We see this by considering the definition of the


at the point

df
du

=Lim
u-+uo

=Lim
U-+UQ

Since

f (x,y)-f (x0,Yo)
U-Uo
f (x,y)-f (xo,y)+f (xo,y)-f (x0,y0)
_
U-Uo

f (x,y) is assumed to be continuous in each variable separately,


(x0,y0), we may use the mean value theorem (see

at and near the point


Section 1.1) to write

df
f x(Xi.Y) (x-xo)+J;,(Xo,Yi). (y-yo)
=Lim
du
u-+uo
u-u0

where

Xi lies between x0 and x, and Yi between Yo and y.

Thus

df

Now if

=L.Im
u-+uo

-+

fxXi,Y
(
)

y(u)- y(u0)
x(u)-x(u0) f,
)
+ y(Xo.Y1
U-Uo
U-Uo

u0 then by continuity x

-+

x0, y

so that we obtain the result (1.3.5), provided that


functions in the neighbourhood of

y0, Xi -+ x0, Yi -+ Yo
f x and fyare continuous

-+

(x0,y0).

The natural generalization of this result to an arbitrary number of


variables is simply the following.

If we have a function f (xi,X2, ...,x") such that each X; is a


continuous function of u, X; = x; (u) for i = 1, 2,.. ., n, then under suitable
assumptions regarding continuity we have

THEOREM:

df
du

i= 1

of dx;
ox; du

(1.3.6)

The proof of this theorem proceeds in exactly the same manner as we


have already shown in our two-variable example.
We can further extend the result of equation (1.3.6) by replacing total
derivatives by partial ones.

X; = X;( v1,v2, . .. vm) for i = 1,2, ..., n, and let vi


vi( u) for
m.
If we think of our function f as a function the variables vi each one of
which has a functional dependence upon u we can use the above theorem
Let

j = 1,2, . .,
.

to write:
(1.3.7)
2

10

[Ch.

REAL VARIABLE THEORY

Furthermore we may apply the theorem to each of the functions

u through vj:
dx; - ox; dvi
du jI avj du .

1
X;

which are also functions of

Now we can employ equation

(1.3.8)

(1.3.8)

to rewrite equation

This result, on comparison with equation

(1.3.7),

(1.3.6)

thus:

gives us

(1.3.9)
assuming that the ordering of the summations may be interchanged
which will always be a valid operation for finite sums.
Finally we remark that higher partial derivatives may be defined in
the natural manner. To exemplify this we consider very briefly second
derivatives of a function

f(x,y) of two real variables x and y. There are

four possible such derivatives:

: :x (ix)

t,

a2f a (af
oyox - oy \ax
a2f = !_(of
oxoy OX \ay
!..._ af
= a2f
ay2 ay oy

xx

xy

f,
yx

f,
yy

()

)
)

(1.3.10)

In many applications, however, it turns out that xy


x Investigating
this matter from first principles we see that these derivatives are certainly
=

equal if
Lim
x-+xo
y-+yo

F(x,y) =Lim F(x,y)


y-+yo
x-+xo

where

(
F x,y)

f(x,y)-f(xo,y)-f(x,Yo)+f(xo,Yo) .
(x-x0)(y-y0)

(l.3.11)

Sec. 1.4)

II

JACOBIANS

Equality (1.3.11) will certainly hold good if


function of the two variables

F(x,y ) is a continuous
x and y at the point (x0,y0 ).

Although this condition is a sufficient one it is not necessary and the


result can be derived on the basis of somewhat less restrictive continuity
assumptions.
1.4

Jacobians

In many of the problems of physics the mathematics of co-ordinate


transformations plays an essential role. In this context we shall be
interested in a determinant called a

Jacobian.

If we have two equivalent sets of independent variables


and

u1, u2,

o(X1,Xz, ...,Xn)
o(u1,Uz, ...,Un)

OX1

OX1

ou1

ou2

ox2

x1, x2 , ..., Xn

, U n then the determinant

OX1
OUn

(1.4.1)

OU1

::; :::. . . . :::1


.

whose (i,j)th element is

oxjoui, is called the Jacobian of the transforma


u to the x. The Jacobian of the reciprocal trans
formation, that from the variables x to the u , has (i,j)th element 0u;/oxi,

tion from the variables


and is written as :

OU1
ox1
o(U1,Uz, ...,Un)
O(X1,Xz, ...,Xn)

OU1
ox2

OU1
OXn

ou2

ax;:
OUn
ox1

oun
ox2

(l.4.2)

OUn
OXn

A very important property of Jacobians is given in the following


theorem.

If we have two equivalent sets of independent !'ariables


x1, x2, ... Xn, and u1, u2, ... Un, then

THEOREM:

o(U1,Uz, ...,Un) o(X1 ,Xz, ...,Xn)


.
O(X1,Xz, ...,Xn) o(U1,U2, ...,Un)

1.

(1.4.3)

12

[Ch. 1

REAL V ARlABLE THEORY

The proof is a trivial consequence of the 'row into column' multiplica

tion theorem for determinants. The


minant is

L ou;/oxk. oxJouj
k

unity when
1.5

(i,j)th

element of the product deter

which, using the results of Section 1.3, is

and zero otherwise.

Multiple Integrals

In defining a definite integral as the limit of a sum we are readily


led to an interpretation of the limit in terms of 'the area under a curve'.
We might refer to an expression such as

J! /(x) dx

as a single integral,

the epithet 'single' referring to the fact that only one variable, namely
is involved.

x,

The 'limit of a sum' type of definition is readily generalized to give


natural definitions of

double and triple integrals-indeed multiple integrals

of any order whatsoever. The geometric interpretations, on the other

hand, very rapidly assume dimensions too large to be contained in


ordinary three dimensional space: this, however, does not worry the
geometer who is quite prepared to carry over into higher dimensions
the language of three dimensional geometry.
We shall be largely concerned with double integrals which we will

relate to 'the volume under a surface' in three dimensions-a concept


which should not tax our imagination too greatly.
Suppose that we have some functional relationship z =

f(x,y).

The

geometer calls this the equation of a surface in three dimensional space.


Let us measure the direction of the z axis vertically upwards. Coni;ider

a region of the surface z =

f (x,y) {shaded in

Fig. 1.7) and its projection A

FIG. 1.7 '

upon the horizontal plane z = 0. What we have constructed in Fig. 1. 7


is a right cylinder on A as base, and the volume enclosed by this cylinder
between z = 0 and z =

f (x,y)

is what we shall call the volume under the

prescribed region of the surface.

Sec. 1.5]

MULTIPLE INTEGRALS

13

Let us divide A into closed subregions A; of area LiA; and let (x;,y;)
be any point of A;. Then we define the double integral off(x,y) over A as

f1

f(x,y) dxdy = Lim

;ti f(x;,Y;)LiA;.

(1.5.1 )

The limit in (1.5.1 ) is taken as n becomes infinite and the diameter of


the subregion shrinks to zero (the diameter of a region is the greatest
possible distance between two points of the region). There is, of course,
no necessity that this limit should exist at all. It certainly does exist if

f(x,y) is a continuous function of the two variables x and y in the closed


region A. For then it must be continuous in each closed subregion,
and so, by the theorem of Section 1.2, it is bounded in each subregion
by values actually taken on by the function. If we call the upper and
lower bounds of the function in the ith subregion M; and m; respectively,
it is evident that
n

m;M;

i=l

i=l

i=l

I f(X;,Y;)LiA; I M;LiA;.

(1.5.2 )

The upper and lower sums are readily interpreted as volumes (see
Fig. 1.7). As the diameter of A; shrinks to zero the bounds M; and m;
tend towards equality and we can interpret J SA f(x,y) dx dy as the volume
under the surface z = f(x,y).
It is important to emphasize two points-firstly that in the definition

(1.5.1 ) (x;,Y;) is any point of the subregion A;, and secondly that the
. division of A into subregions A; is perfectly arbitrary.
Finally there is one slight complication which may be a pitfall for the
unwary. It concerns the geometrical interpretation of the double integral
in the case when f(x,y) has values in A which are both positive and
negative. When f(x,y) is negative it is clear that the volume which lies
below the plane z = 0 gives a negative contribution. Thus, if we are
using double integrals to calculate volumes, we should consider separately
the regions of positive and negative
Example 1.

z.

Give a geometric interpretation of J L dx dy.

In this example

we are dealing

J SAf(x,y)dxdy where f(x,y)

with

an integral of the type

1. Thus we can interpret the integral as


the volume under that portion of the plane z = 1 whose projection on
z = 0 is the region A. Clearly this volume is just 1 . LiA = LiA. So we
conclude that J L dx dy over any region A simply gives the area LiA of
that region.

14

[Ch. 1

REAL VARI ABLE THEORY

Find an expression in terms of multiple integrals for the


height above the plane z = 0 of the mass centre of a uniform solid obtained
by projecting a portion of the surface z = f(x,y) onto the plane z = 0.

Example 2.

Let us call the projection of the given portion of the surface


onto

Ai.

z=0 A

(see Fig.

1.7)

and as before, divide

below by values of the function in each


the volume on

A;,

z; be

f(x,y)

f(x,y)

into subregions

thatf(x,y);;:::: 0 in A.

We shall assume for the present

continuity assumptions we can infer that


M; and mi. Then, if

Under suitable

is bounded above and

A;: Jet us denote these bounds by


z 0 of the mass centre of

the height above

it is clear that

(1)
Now

z,

the height of the mass centre of the whole solid above the

z=0

plane, is given by
n

I pV;z;

Z=

i= 1
n

iI
=1

(2)

pV;

where v; is the volume of the ith subvolume and

the density of the

solid. But we have already noted that:

(3)
Thus combining inequalities

(1) and (3)

we see that
(4)

Now, since we are assuming that


and

in the closed region

f(x,y)

the bounds

is a continuous function of

m;

and M; of

f(x,y)

in each

A;

are actual values of the function. This being so, when we pass to the
limit as

becomes infinite and the diameters of the subregions shrink

to zero, we see from (4) that

ff

{f(x,y)}2 dx dy

V;zi t ff {f(x,y)}2 dx dy

Lim
.1=1

which implies that


Lim

;ti V;zi

Jt

Finally, proceeding to the limit in

z=

(2),

(5)

{f(x,y)}2 dx dy.
and cancelling the

t SL {f(x,y)}2 dx dy
SL f(x,y) dx dy

p,

we obtain

(6)

Sec. 1.6]

15

REPEATED INTEGRALS

If, however,f(x,y) is not positive throughout


the regions of positive and negative

A 1 and A
2

A let us consider separately

which we will call respectively

We have

t J L, {J(x,y)}2 dx dy
Zi =
J L, f(x,y) dx dy

(7)

t J L2 {J(x,y)}2 dx dy
J SAJ(x,y) dx dy

(8)

and

Zz =
Notice that

f(x,y) 0 in A If we call the


2
z = 0 V1 and V respectively then
2
centre z given as

because

above and below the plane


the height of the mass

volumes
we have

1 + V2Z2
z = V Z1
V1 + Vz
where V1
equation

(9)

= J SAJ(x,y) dx dy and V = - J LJ(x,y) dx dy. Writing out


2
(9) in full we have
J J {J(x,y)}2 dx dy- J L2 {j(x,y)}2 dx dy
(lO)
z = A,
2
J LJ(x,y) dx dy- J JAJ(x,y) dx dy

The denominator in equation

(10)

is positive while the numerator can

be either positive or negative.

Write down an expression for the volume of a sphere of


radius a as a definite integral.
The equation of a sphere radius a and centre the origin is

Example 3.

x2+y2+z2 = a2

(I)

which may be rewritten as

.J(a2-x2-y2).

This presents us with a new problem:

(2)

z is two-valued, one value positive,

one negative.

z=0
2 J L .j(a2 - x2-y2) dx dy

It is, however, clear that there is as much of the volume above


as there is below it. Thus the expression V
where

is the closed region

x2+ y2

a2

correctly gives the value for

the volume of a sphere.

1.6

Repeated Integrals

It is all very well to be able to write down expressions for volumes,


positions of mass centres, and so on, but it is of little value to us in

16

[Ch.

REAL VARIABLE THEORY

practice until we learn the mechanics of evaluating double integrals.


We shall discover below that the methods of ordinary single integration
can be employed repeatedly to evaluate multiple integrals. So \Ye now
introduce the concept of

repeated integration.

Let us start with a very simple type of double integral-namely one


in which the region A is a rectangle. The division of A into subregions
is at our choice so Jet us elect to deal with rectangular subregions (see
Fig.

1.8). In this way it will be possible to express the area of each sub
L\xi . L\yi. Each subregion is now labelled by two
suffixes i and j but it is clear that we could invent a single suffix notation
if we wished. If i takes the values 1, 2, ... , I and j the values 1, 2, ..., m
then altogether there are n rectangles where n
I . m.
region in the form

y =,B-------

y=aL---+--+---- -
x=b
x=o

l:;.x,
FIG. 1.8
Our definition of a double integral now-takes the form:

fI

where

f(x,y) dx dy

Lim

ii=I
m

i f(x;,31)Lix; . L\yi

1-+oo i=l
m-+oo j= 1

'
(1.6.1)

..

(x;,yj) is any point in the shaded rectangle in Fig. 1.8.

In general ambiguity exists in the interpretation of the right-hand


side of equation
m .._. oo and then

(1.6.1). For example, we might first perform the limit


I

.._. oo :

1!1 J1 { itl f(xi,y)Liyi} Lix; 1!1 JJJ: f (x;,Y) dy} Lix;


=

The final form of equation

J: {J:

dy f(x,y).

(1.6.2)

(1.6.2) is a bit clumsy and we prefer to


y integration is performed first

write the repeated integral in which the


as s: dx s

f(x,y) dy dx.

Sec.

1.6)

REP EA TED INTEGRALS

17

Now we could invert the order of the limiting operation to obtain


the repeated.integral

J dy J dxj (x,y).

If we study specific examples we discover that the value of a repeated


integral may or may not depend on the ordering of the integrations.
However, if the corresponding double integral exists-the continuity of

f(x,y)

is a sufficient condition for this-then both orders of performing

the successive integrations lead to the same result, namely the value of
the double integral.

Evaluate in two different ways J L x2(x2-y2) dx d y where


is the square 0 x 1, 0 y l.

Example 4.
A

(1)

fL

x2(x2 - y 2) dx dy

(2)

fI

x2(x2 -y2) dx dy

It will be noticed that

t t
t
dx

dy(x4 -x2y2)

(x4-ix2) dx

-ts

t I
I
dy

dx(x4-x2y2)

(t-fr2)d y

-ts

x2 (x2-y2) is a continuous function of x and y

over the square and consequently the double integral exists.

Unfortunately matters are not always as simple as the rectangular


case!
In general we must be prepared to deal with regions the shapes of
which are perfectly arbitrary. Let us take as typical the region A de
picted in Fig. 1.9 and let us enclose it in the smallest possible rectangular
R. Now define the discontinuous function

F(x,y)
F(x,y)

F(x,y) as follows

f (x,y)

for

(x,y) in

for

(x,y) elsewhere in R.

Then, as we might expect intuitively, it can be shown that

fJR

F(x,y) dx dy

fL f

(x,y) dx dy.

18

REAL VARIABLE THEORY

[Ch. 1

We use this result to evaluate a double integral as follows:

fL

f(x,y) dx dy

fL
f f:
J: f:

F(x,y) dx dy

dy

dy

dx F(x,y)

(1.6.3)

dxf(x,y).

where x1 and x2 are functions of y. Let us clarify the point by means


of an example.

._
__::::a.,-=.____,u
y=a L-....__
x=u
FIG. 1.9

Example 5.

0, y

Evaluate
0, and x+ y

SSA xy dx dy
=

over the triangle bounded by the lines

1 (see Fig.

1.10).

x =O
FIG. 1.10

Suppose we perform the x integration first as in equations (1.6.3). As y


varies between 0 and 1, x varies from 0 up to its value on the line x+ y

1.

Sec.

1.7]
x1

Thus

CHANGE OF VARIABLES
=

0 and

x2

1-y

f1

xydxdy

I: I:-y
f

19

dx xy

dy

(1-y)2y

dy

l4

Clearly we could have reversed the order of the integrations and obtained

fL
1.7

xydxdy

I: I:-x
f
dx

dy xy

(1-x)2x

dx

Change of Variables

In multiple integration, just as in the evaluation of single integrals,


an intelligent choice of integration variables is an important factor in
determining the ease with which we can work out the value of the integral.
It is obvious, then, that we must learn how to effect a change of variables
in a multiple integral.
As can be seen from our definition of a double integral in equation

(1.5.1)

the problem of change of variables is basically a question of how

elements of area transform under variable changes.


Y (x,y+6y)

X(x+6x,y)

O(x,y)
FIG. I.I I

(x,y) plane drawn in Fig. 1.11,


x = x(u,v), y = y(u,v). If dx
and dy are very small the new element of area in the (u,v) plane will be
approximately a parallelogram (see Fig. 1.1). Now all the points on OX
correspond to the same (constant) value of y so that the point X' in the
Consider the rectangular region of the

and let us make a co-ordinate transformation

20

[Ch. 1

REAL VARIABLE THEORY

u
Fm. 1.12

(u,v) plane corresponding to

X in the

(x,y) plane is completely specified

relative to O' by

(1.7.1)

Similarly the point Y' is specified relative to O' by

Auy
Avy

OU
=

oy

. Ay
(l.7.2)

av
. Ay.
oy

Now by considering the areal equality O'X'Y'

O'UY' + WX'Y' +

UVX'W-O'VX', and noting that the area of the parallelogram is twice

the area of triangle O'X'Y', we can write down the following expression
for the area of the transformed region :

the sign being chosen to give a positive value.


Thus, in terms of Ax and Ay, the area of the parallelogram is given by

'

'

au av au av
.
Ax . Ay
ox . oy - oy ox .

Then, by definition, we obtain the result

fLr

JT(u,v) du dv

f1

o(u,v)
Ax . Ay.
o(x,y) .

f (x,y)

I!: I

dx dy

. (1.7.3)

(l.7.4)

Sec.

GREEN'S THEOREM

1.8]

where

fT

21

denotes the transformed function and

AT

the new region of

integration. Recalling that

o(u,v) o(x,y)
=
.
o(x,y) o(u,v)

we can also conclude that

f1

f(x,y)dxdy =

fLT

fT(u,v).

, :::;I

dudv.

(l.7.5)

Find the volume of a sphere of radius a.

Example 6.

From Example 3 we have that the required volume is


V=

fl

.J(a2-x2-y2)dxdy

(1)

over the region


Put

Then

A: x2 + y2 a2.
r cos () and y r sin 0.
Jo(x,y)/o(r,O)I r and the transformed
=

V=
where 0

a and

Now to evaluate

JJ ,,j(a2-r2)rdrd()

(2)

2tr.

0 ..:;;; ()

(2)

we express Vas a repeated integral

V=

1.8

integral becomes

{" f:
d()

dr.J(a2-r2)r

]a/()

f,_,,, [.,
:

-:i(a2-r2)3/2

4na3

-3

Green's Theorem

Suppose that we have a plane curve

y = g(x) which we shall call C;


lin e integral with respect to x along C from the point
of the function f(x,y) as

then we define the

A to

the point B

where

x= a

and

x=

f(x.y)dx

J:

f(x,g(x))dx

b correspond to the points

(1.8.1)

A and

B respectively

of the curve C. This definition is only meaningful if the functional

relationship

g(x) is

one-to-one.

22

[Ch. 1

REAL VARIABLE THEORY

In circumstances when the relationship is not one-to-one we must

split the curve

C1, Cz, ...


1.13). Then

into portions

in each (see for example Fig.

such that

g(x)

is one-valued

the line integral along

is

y
8

FIG. 1.13

interpreted as

f(x,y)dx

f(x,y)dx+

f(x,y)dx

(1.8.2)

C2

C1

Example 7.
Calculate the line integ ral of y with respect to x round the
circle x2 +y2 a2 from (0, - a) to (0, +a) in both directions.
(1) Firstly let usintegrate round C in an anticlockwise sense.
=

ydx

We set

a cos()

I:
r

J(a2-x2)dx+

J(a2-x2)dx

J(a2-x2)dx

to obtain the final result

ydx

1t
- 2
.

(2) In exactly similar fashion we can integrate round the circle from
(0,

-a) to

(0,

+a)

in the clockwise sense to obtain

ydx

na2
+2.

Sec.

1.8]

23

GREEN'S THEO RE M

The above example suggests that it may often be useful to use a

C. Suppose we have x
x(u),
y(u) as the parametric equations of C; then we can rewrite the
definition (1.8.1) as

parametric representation for a curve

f(x,y) dx

where

u0 and u
C.

fub

(1.8.3)

f(x(u),y(u))x'(u) du

Ua

ub correspond to the points A and

B respectively

of the curve

x
a cos() and y
x2 + y2
a2

In Example 7 we could conveniently use


as the parametric equations of the circle
Should a curve

a sin()

C be closed we split the curve into sections as before

.and take as convention that the positive sense of describing the curve
is anticlockwise. When the curve is closed we usually modify the sign
of integration by superposing a small circle thus:

f.

We are now in a psition to establish a connection between double


integrals and line integrals. This important result, known as Green's
theorem, will be seen in later chapters to have special relevance in the
discussion of the integral properties of functions of a complex variable.

Suppose that we have a closed region A lying in a plane and


bounded by a closed curve C which has no double points; if we have two
continuous functions of x and y, F(x,y) and G(x,y), whose partial derivatives,
oF/oy and oG/ox, are also continuous in A, then
THEOREM:

(1.8.4)

Let us enclose the region A in the smallest possible rectangle as shown


in Fig.

1.14. Now consider the first term on the right-hand side of


R

FIG. 1.14

24

[Ch. 1

REAL VARIABLE THEORY

equation

(1.8.4):

fl dxdy=fdy Idx
=I: [c(x,y ::dy
=f {c(x2,y)-G(xi,y)}dy
=I G(x,y)dy- f G(x,y)dy
=J G(x,y)dy + f G(x,y)dy
= G(x,y)dy.
PQR

PSR

PQR

RSP

In exactly similar fashion we find that

ft;dxdy = - F(x,y)dx.
The assumptions of continuity which have been made guarantee that
,all the relevant integrals exist.
. .

Ci C2A=Ai +A2

Suppose that we consider a composite region

bounding curve

as shown in Fig.

1.15. If

and

with

are the bounding

FIG. 1.15

A2, CiAi C2A2

curves of
and

and

and

respectively, then along the boundary between

Ai

have a common portion but run in opposite senses.

Ex. I]

25

REAL VARIABLE THEORY

It is clear that the following symbolic relationships will be valid:

(1.8.5)

EXERCISE 1

y)
n x
y
(A.x,Ax"y)f(x,
= A."f(x,y)
0)
f(x,
y
)
FEuler's
(y/x) theorem:
x 0. x(of/ox)+y(of/oy)= nf.
2.
(x-y)/(x+y)3
l dx l dy- x-y
l dx- x-y .
.
dy
f 3 J (x+y)3
J J (x+y)3
SSA(x-y)/(x+y) dxdy
0 x
0 y
(Answer: t, -!)
y2= a1,ux=, y2x2/y a2x,
x2v ==y2b1y,
x2
= b y (a
b
b1
0)
2 2
2
/x
SL dxdy
(Answer: (a -a1)(b -b 1)/3)
2 2
x2+y2)n dxdy
(
ff a2 b2
n 0
x2/a2+y2/b2 z = 0.
(Answer: nab/(n+
' f:sine
( 2 d(} dr J(l:-r2f
(Answer: 2(nz= 0
z = x+y
x2
+
y2
=
(Answer:
1. If

is a homogeneous function of degree

for A.

form

for

show that

>

in

and

(that is

can be expressed in the

>

Prove

at the origin.

Discuss the behaviour of the function

Evaluate the repeated integrals

Does

the

1,

double

and

integral

over

the

square

1 exist?

3. Find the area enclosed by the four parabolae


> a1 >

O;

in the double integral

>

>

by setting

and

taken over the required area.


(U.C.N.W. 1963)

4. Evaluate

for

>

over the region

1,

1))

(U.C.N.W. 1963)

5. Evaluate the repeated integral

2))

6. Find the volume in the first octant between the planes


which is inside the cylinder

and

16.

128/3)

7. Use Green's theorem to show that the area

of a plane region

bounded by a curve C is given by

A=
3

fc xdy= -ydx = t(xdy-ydx).

REAL VARIABLE THEORY

26

8. Evaluate the integral

f x dy where C is the perimeter of the triangle


c

formed by the points (0,0), (0,a), (a,b); hence verify that the area of the
triangle is given by this integral.
9. If x
r sin (}. cos (}, y = r sin (}. sin </J, z r cos (} (spherical polar
co-ordinates) show that
=

I o(r,(},</J) I
o(x y,z)
,

I r2 sm e1.

10. Evaluate the triple integral SJ Jv .j(a2 -x2 - y2 - z2) dx dy dz over


the spherical region x2 +y2 +z2 a2 by transforming to spherical polar
co-ordinates.
(Answer: n2a4/4)
Note: A triple integral of a function f (x,y,z) of three variables over a region V of
three dimensional space may be defined by analogy with equation

(1.5.1)

as the

limit of sum :

fff f(x,y,z)dxdydz

=Lim

Such

an

.I f(x;,y;,z;).M;.

1=1

integral may be evaluated by repeated integration, and the natural gener

alizations of the formulae for change of variables given in Section

1.7

are valid.

11. By using the substitution y = tx, or otherwise, find the area


enclosed by the arc of the curve
X=

at

--

l+t3'

at2
1 +t3

from the point O(t = 0) to the point A(t


T), together with the straight
line AO. By letting T-+ oo prove that the area of the loop is a2/6.
=

(HULL 1960)
12. Evaluate

ff

dx dy
(x2 +y +3a2)2

taken over the interior of the parabola y2 = 4ax.


(Hint: Change the variables to polar co-ordinates)
(EDINBURGH 1959)

CHAPTER 2

Complex Numbers
The concept of an 'imaginary number' such as the square root of minus
one was first introduced, with great scepticism, late in the sixteenth
century. Mathematicians and physicists soon discovered that the device
led to many simplifications in difficult problems, and having proved its
worth in practical situations the imaginary number became a reputable
and powerful mathematical tool. Today the study of imaginary numbers
has become one of the most fascinating of all branches of mathematics
and has applications ranging from simple algebraic problems to the
complicated problems of modem classical and quantum physics. In
this chapter we shall embark on our study of this important topic.

2.1

The Square

Root of Minus

One

The art of mathematics is largely concerned with symmetries and


patterns. Let us pick up any school textbook on the solution of quadratic
equations. We are likely to find the following type of statement 'the
equation x2 - 9

0 has two roots x

3, but the equation x2 + 9

has no roots'. The mathematician finds this sort of asymmetry rather


unpalatable; he would prefer an algebra in which all quadratic equations
had the same number of roots. This can in fact be achieved by adding
to the real number system the imaginary number which is the square
root of minus one. If we denote such a quantity by i (in some textbooks,
especially engineering ones, the symbol j is employed) then we have

i2

-1 or i

j(-1). Having taken this very radical step we can

rewrite the equation x2 + 9

0 as x2 - 9i2

0 which has solutions

3i. This formal manoeuvre certainly puts the equations x2 - 9

and x2 + 9

0 on an equivalent footing and it is a straightforward

matter to show that all quadratic equations have two roots provided
the entity i is added to our number system. The general quadratic
equation is

ax2+bx+c
27

(2.1.1)

[Ch. 2

COMPLEX NUMBERS

28

and has formal solutions


- b..j (b2 - 4ac)
x = 2a
If b2 ;;:::: 4ac theri there are two real roots

(2.1.2)

(in the case b2 = 4ac these

two roots coincide). On the other hand if b2 < 4ac the quantity i can be
introduced to write the solution as
x =

-bi ..j(4ac-b2)

2a

(2.1.3)

In this way the original notion of an imaginary number gives way to


the concept of a complex number. We define a complex number z in
terms of two real numbers x and y as
z = x+iy.

(2.1.4)

It is evident that a real number is a special type of complex number


with y = 0 and an imaginary number one with x = 0. It is customary
to call x and y the real and imaginary parts of the complex number z.
We write:
x = .!Jlz,

y = fz.

(2.1.5)

There are many methods by which the concept of a complex number


could be introduced but from a practical standpoint there is really no
logical justification for such a step except that, having taken it, one
discovers new and more effective methods of coping with practical
problems. Mathematically, however, we are at liberty to build whatso
ever theories we please provided that they can be derived consistently
from hypotheses which are, of course, at our choice. We shall return to
this point in Section 2.5.
2.2

The Argand Diagram

An instructive method of coming to terms with the notion of complex


numbers is to regard the introduction of i as a mathematical device
which enables us to handle simultaneously two real quantities. The
algebra relating to a complex number

automatically describes both x

and y at the same time. Thus the algebra which we shall develop is
really an algebra of real number pairs.
Often a real number pair (x,y) is represented by a point in a plane:
the plane of points representing complex numbers is sometimes called
the Argand diagram, but we shall usually refer to it as the complex
z-plane. The point P in Fig.- 2.1-whose coordinates are (x,y) represents

Sec. 2.2]
the complex number
ordinates

29

THE ARGAND DIAGRAM

z = x+iy; P can also be


z = r(cos (}+i sin fJ).

specified by polar co

r and (} so that

P(x,y)

x
FIG. 2.1

The distance

OP is called the modulus of the


OP m akes with the x-axis is

the angle (} which

complex number
called the

and

argument (or

amplitude) of z. We shall adopt the following notation:

= lzl = +.j(x2+y2)
8
arg z = tan-1y/x

(2.2.1)
;

(2.2.2)

where it is understood that tan


angle e, namely

y/x

denotes the

principal
value of the
.

'It < (} 'It.

Now let us make the assumption that the entity


defined by the equation

i which we have

i2+1 = 0 is subject to the same rules of algebra

as are the real numbers.


y

FIG. 2.2

If we add together the complex numbers


obvious notation

z1

and

z
2

we have in an

(2.2.3)
As we can see from Fig. 2.2 the figure

OP1PP

is a parallelogram.

Thus the rule (2.2.3) for adding complex numbers corresponds, in the

30

COMPLEX NUMBERS

[Ch. 2

complex z-plane, to a parallelogram law for addition. This is the vector


law of addition which is frequently met with in physics as the combin
ation law of forces, velocities, etc.
If, now, we multiply together z1 and z2 the result is rather more
complicated :
Z1Z2 = (X1X2 -Y1Yi)+i(X1Y2+X2Y1)
=

r 1 r2{ cos(O1+82)+i sin(01+02)}

(2.2.4)

where X;
r; cos O;, Y;
r; sin O;, for i = 1, 2.
Thus the modulus of a product is equal to the product of the moduli
of the factors. If
n < 01 + 82 n then the argument of the product is
just the sum of the arguments of the factors. However, 01+02 may lie
outside this range and in this case the argument of the product will be
given by one of the two expressions 81+82 2n. We shall dwell further
on this matter in Chapter 7.
=

2.3

Complex Conjugation

In general a functionf(z) of a complex variable z can be expressed, in


terms of two real functions of two real variables x and y, as
f(z)

u(x,y) + iv(x,y)

(2.3.1)

where we call u and v the real and imaginary parts respectively of f(z),
and we employ the same notation as in equation (2.1.5), namely
u

v =.ff

fJ1l f,

(2.3.2)

We also extend our previous definitions of modulus and argument to


include functions of z, thus
(2.3.3)

arg f

tan

1
-

v/u.

(2.3.4)

Let us now introduce a special notation z* to denote the quantity


x - iy which we call the complex conjugate of z (in some textbooks the
notation z is used). Complex conjugation involves replacement of i by

-i.

Sec. 2.3]

31

COMPLEX CONJUGATION

Example 1.

Show tha t bo th zz* and z+z* are real qu an tities.

(1)

zz*

(2)

z+z*

(x+iy) (x -iy)

x2+y2

(x+iy)+(x-iy)

2 x.

We shall call f*(z) the complex conjugate function of f(z) and define
it as

f*(z)

(2.3. 5 )

u (x,y)-iv(x, y).

In other words we replace each i which occurs (either implicitly or


explicitly) in f(z) by -i. In general f*(z) will be a different function
from f(z*). For example if f(z)

(1+i) (l+z), then f*(z)


(1-i)(l+z*)
(1+i)(l+z*).
f(z*) then f(z) is called a real function of the complex

whereas f(z*)
If f*(z)

variable z. To gain some insight into the meaning of 'real' in this defini
tion it is instructive to consider the example of an algebraic function :
in this case the condition f*(z)

f(z*) implies that all the coefficients

in the expression for f(z) are real numbers. Clearly f (z) = (1+i)(l+z) is
not a real function.
Example 2.

functionf(z)
(1)

Find the real p art, the mod ul u s, and the amplit ude of the
=

(l+z)/(1-z). Is f(z) a real function?


f(z)

1+x+ y
=

1-x-zy
l+x+iy 1-x+iy
1-x-iy 1-x+iy
1-x2 -y2+2iy

= -----

(l-x)2+y2

Hence

(2)

l-x2 -y2

.OAf(z)

l f l2

(l-x)2+y2
f*f
l+x+iy l+x-iy
1-x-iy 1-x -iy

(1+x)2+y2
(l-x)2+y2

COMPLEX NUMBERS

32

[Ch.

Thus

(3)

arg f

(1+x?+y2
(1-x)2+y2

j
-

'!I = +

tan

=tan -

f
.o/tf
5

2y
1 -x2 -y2

using

(1)

1+z *
f*(z) = {
1- z

(4)

l+z*
1-z*
= f(z*)
Hence

f (z)

is a real function of the complex variable

coefficients in the expression for


2.4

f(z)

z. Notice

that the

are real.

De Moivre's Theorem

We have asserted in the preceding chapters that, having conceived


the notion

of introducing

complex

numbers,

mathematicians

and

physicists persevered with the idea because it proved so fruitful in


practice. While many of the major achievements of complex variable
theory must necessarily be deferred to later chapters we can quite
convincingly demonstrate the power of the method in one or two
simple problems.
It is well known that the functions
as infinite series in

ex,

cos

x, and sin x

can be expanded

which converge for all real values of

x.

These

series are

x2
ex= l+x+-+ ...
2!
x"
= I n=o n!
oo

cosx =

x2 x4
1--+-- ...
2! 4!
oo

nO

x2"(-1)"
(2n)!

(2.4.1)

(2.4.2)

Sec. 2.4]

33

DE MOIVRE'S TH E O RE M

x3 xs
sin x =x--+-3!
5!
Cl()

= n=O

x2

.
(2.4.3)

n+ 1 -1)"
(

(2n+ 1)'

Very early on in the history of complex variable theory a startling


formal connection between these three functions was discovered. It was

noticed that the substitution of i(J for x in equation (2.4.1) led to the

following result :

ei8

2
1 -(J + (J4 - ..
2!
4!

=cos

+ /e-(}3 + (J5 3! 5!

(}+ i sin (}.

.. )

(2.4.4)

This formula actually appeared to have some mystical significance!


In the history of mathematics there had been great initial opposition to
the introduction of negative numbers and of
such as n and

transcendental numbers

(a transcendental number is one which cannot be ob

tained as the solution of an algebraic equation with rational coefficients).


Now here we have

i,

the imaginary number, which had received most

bitter opposition, effecting a bridge be.tween all three unwanted numbers!


For if we put

(}

=n in equation (2.4.4) we obtain the result

ein

= -1!

Now we shall have a great deal to say about the exponential function
with imaginary argument in later chapters and we shall not give any
detailed account of the matter at this stage.
Let us proceed, however, as did the mathematicians of former days,
and study equation (2.4.4).
Now if (e;e)" =eine (the usual law of indices for real exponents) we
discover yet another remarkable result :

n(J + i sin n(J.


This result is known as De Moivre's theorem. We
proof of this theorem (for rational values of n) which
THEOREM:

(cos

(} + i sin (})"

=cos

shall now give a


is independent of

the argument which we have used above to derive the result: the method
of proof which we shall employ is called the method of

induction (for an alternative method

mathematical

of proof see Exercise 2, Question 2).

Let us suppose that the result which we are trying to prove holds for
some positive integer

= k. Then we may write

(cos e+ i sin

=cos k + i sin k .
(J
(J

34

[Ch. 2

COMPLEX NUMBERS

Now multiplying both sides of this equation by (cos()+i sin()) we


obtain
(cose+i sin (J)k+ 1

(cos kO+i sin kO) (cosO+i sin 0)

(cos k() . cos 0- sin k() . sin 0)


+i(sin k() . cos()+cos k() . sin 9)

cos(k + 1)()+i sin(k + 1)().

Thus we have proved that if the theorem is true for


true for n
k + 1.
But the theorem is obviously true for n
1 since

k it is also

(cosO+i sin 0)1

cos(l. O)+i sin(l. 0).

Hence we can conclude that the theorem is true for n


2; but if
the theorem is true for n = 2 it is also true for n
3; the argument may
be repeated indefinitely and we see that the theorem must be true for
all values of n.
It is a simple matter to extend this result to embrace negative integers.
Suppose that p is an integer such that p > 0 and we wish to evaluate
(COS{)+i sin{))- P,
=

(cos()+i sine)- p

.
(cos()+1. sm())P

1
= ------

cos p()+i sin p()


cos pO - i sin p()
cos2p()+ sin2p(J

cos( - p()) +i sin( - pO).

Finally we can extend the theorem to include any rational index


p/q where p and q are integers, q i= 0. We have:

(cos</>+i sin<f>)Pfq

(cos p<f>+i sin p<f>)1fq.

Then raising both sides to the power q we obtain


(cos q<f>+i sin q</> )Pfq

Thus writing ()

cos p<f>+i sin p<f>.

q<f> we get the required result

(cos()+i sineytq

cos !?.e+i sin !?.e.


q
q

DE MOIVRE'S THEOREM

Sec. 2.4]
Example 3.

35

Find cos 38 and sin 30 as functions of sin 8 and cos 8.

cos 38+ i sin 30 = (cos 0+isin 0)3


=cos30+3icos28. sin 0-3 cos 0. sin20-isin30.
Hence
cos 30 = cos30-3 cos 0. sin20
=4 cos38-3 cos 8
and
sin 30 =3 cos20. sin 0-sin20
=3 sin 0-4 sin30.
Example 4.

Factorize the expression zs - l.

We shall solve this problem by finding five linear factors for zs -1.
Now it is easily. verified from equation (2.4.4 ) that e2"Ni = 1 for all
integral values of N. Then we have:
zs =cos 2nN+ i sin 2nN
z =(cos 2nN+i sin 2nN )1fs

=cos(2nN/5)+i sin(2nN/5)
where N may take on any value. This does not mean however that there
is an unlimited number of values of z. Because of the periodicity of the
trigonometric functions, we obtain exactly the same result if we set
N =k+5 as we would have obtained by setting N =k.
cos

2nk
2n(k+5) . . 2n(k+5)
=cos --+ 2
+z sm
5
5
5

.. (

2nk
+1 sm --+2
5

= cos(2nk/5)+ i sin(2nk/5).
Let us therefore consider only N =0, 1, 2, 3, 4.
When N = 0 we obtain

= cos O+ i sin 0 = l.

When N = 1 we obtain z = cos(2n/5)+i sin(2n/5).


When N = 2 we obtain

= cos(4n/5)+isin(4n/5).

When N =3 we obtain

= cos(6n/5) +isin(6n/5).

When N =4 we obtain z =cos(8n/5)+ i sin(8n/5).


Now cos(2n/5) =cos(8n/5) and sin(2n/5) = -sin(8n/5) so that the
solutions corresponding to N = 1 and N = 4 are complex conjugates.
Similarly the roots corresponding to N =2 and N =3 are also complex

36

[Ch. 2

COMPLEX NUMBERS

conjugates. Thus we may write the factors of zs-1 as:


(z -1){z - cos(2n/5)-i sin(2n/5)}{z-cos(2n/5) +i sin(2n/5)}
{z-cos(4n/5)-i sin(4n/5)}{z-cos(4n/5)+i sin(4n/5)}
= (z-l)[{z-cos(2n/5)}2 + sin2(2n/5)][{z-cos(4n/5)}2+sin2(4n/5)]
= (z- l){z2-2 cos(2n/5). z+ 1 }{z2 -2 cos(4n/5) .z+1 }.

It is easily proved that the roots of an algebraic equation with real


coefficients always occur in complex conjugate pairs. For suppose we
have some polynominal of degree n in z which we shall call f(z) such
that f *(z)= f (z*): that is to say the coefficients in the equation f (z)= 0
are real. Now the equation f (z) = 0 is equivalent to two real equations
u= 0 and v=0 so that f (z) = 0 implies f *(z)= 0. Thus since f(z) is a
real function of z we have that f (z*) = 0. Hence if z= z0 satisfies
f (z) = 0 so also does z
z.
=

r=n

Example 5.

For any algebraic equation

a.z= 0, where each a. is

r=O

real, show that the coefficient of z"-1 is equal to minus the sum of the
roots: because a" # 0 we have taken an= 1. Deduce that
1 +cos(2n/5)+cos(4n/5)+cos(6n/5)+cos(8n/5) = 0.

If the roots of the equation are z1, z2, , zn, then we may write the
equation in the form (z-zi)(z-z )
(z-z )
0. It is evident, then, by
"
2
multiplying out this expression and comparing it with the original form
of the equation, that the coefficient of z"-1 gives minus the sum of the
roots. In Example 4 we considered the equation zs =1. The coefficient
of z4 is zero, so that the sum of its roots must be zero. Hence we obtain
the required result.
.

Example 6.

Find the sum of the series

cos () cos 2() cos 3()


--+
3+
+
4'
4

. .

Let
cos () cos 2()
c=--+--+
42
4
and let
s=

sine
--

sin 2()

--

42

+ ...

Sec.

2.5]

FORMAL CONSIDERATIONS

37

Then
c+is =

1 9 1 2 9
4e + 42 e ' +

If now we assume that the usual formula for the sum to infinity of a
geometric series is valid we obtain
c+is =

eiB

4-e'.8

--

(4-ei8)(4-e-i8)
ei8(4- e-iB)
11-8 cos e

Now taking the real part only of both sides we obtain the final result
c=

2.5

4cos

0-1

17-8 cos(}

Formal Considerations

In the foregoing sections we have been discussing the idea of complex


numbers in a somewhat empirical manner. Having defined the symbol
i as

,j(-1) we have developed an algebra of complex numbers on the

assumption that i obeys all the usual rules of the algebra of real numbers.
This is indubitably the manner in which complex numbers were first
introduced-we have followed this type of argument thus far in order
to understand the beginnings of the subject. Now, however, let us start
afresh.
In many textbooks on pure mathematics complex numbers are intro
duced in abstract fashion in the context of an algebra of number pairs.
The algebra of complex numbers is defined in terms of ordered real

(x,y) which obey the following rules of combination:


1. (x1,y1)= (x2,y2) if, and only if, x1 = x2 and y1= y2
2. (x1,yi)+(x2,Y2) = (x1+x2,Y1+Y2)
3. A.(x,y) = (A.x,A.y) for any real A.
4
. (x1,Y1HX2,Y2) = (x1X2-Y1Y2,X1Y2+X2Y1).
Rule 1 defines what we mean by e quality, rule 2 what we mean by
4 what we mean by multiplication.
addition, rules 3 and
number pairs

It is readily verifiable that this algebra is commutative: in other words


it does not matter in which order the additions and multiplications are
performed.

[Ch.

COMPLEX NUMBERS

38

The zero element in such an algebra is (0,0) because it is the only

element satisfying (x,y)+(O,O)

(x,y).

The unit element turns out to be

(l,Q) because no other element has

(x,y).
the property (1,0)(x,y)
What then, we may well ask, is the role of (0,1) in this algebra?
=

We can use the given rules to establish that (O,l)(x,y)

that (0,1)(-y,x)

( - y,x) and

(-x,-y). Hence, if we denote the unit element (1,0)

by 1 and the element (0,1) by i we have the relationship


i2

(2.5.1)

-1.

Also, any complex number (x,y) may be written in the form


(x,y)

x(l,O)+ y(O,l)

(2.5.2)

xl+ yi.

Thus all additions and multiplications involve only the rules of


manipulation of real numbers plus the rules 1 to 4 for the elements

1 and i. Clearly all the results of this algebra could be obtained by

writing the complex number (x,y) as x+iy and treating i as an ordinary


number in all respects except that i2

-1.

Evidently there is a case to be made against the terminology


'imaginary' and 'number' when one approaches the subject of complex
numbers in this manner. The matter is not one of vital importance,
however, and we shall, in this book, adhere to the historical nomencla
ture. In particular we will refer to x and y as the real and imaginary
parts of z

x + iy and we shall call y

0 the real axis and x

0 the

imaginary axis.

EXERCISE 2
1. Find all the roots of the equation z3

1. If w is one of the complex

cube roots of unity show that the other is w2 Verify that 1+w+w2

2. Suppose that we have a set of n complex numbers zi


+ i sin 8) for j
1, 2, ..., n. By considering the product z1z
=

De Moivre's theorem for positive integral index.

. .

0.

ri<cos (Ji
Zn prove

3. Find the sum to n terms of the series


1 +cos x+ cos

2x +

(Answer: sin tnx . cos t{n-l)x.cosec tx)

4. Use the method of mathematical induction to prove the result

of Question 3.

5. Use De Moivre's theorem to show that


cos 5()

16 cos58- 20 cos3()+ 5 cos()

Ex. 2]

39

COMPLEX NUMBERS

2
6. Show that an expression of the form z " -1 can always be written
as a product of n real quadratic factors.
n-1
2
2
(Answer: (z -1) n { z -2z. cos(rn/n)+1})
1
r=
7. By representing the complex numbers z1 and z2 as points of the
complex z-plane show that lz1 + z2I lz1 l+lz2I
8. By applying the parallelogram rule to z1 +( -z2 ) show on a diagram
that arg(z1 - z2) is an angle between P 1 P 2 and the real axis, where
P 1 and P 2 are the points of the complex plane which represent z1 and z2
respectively.
9. Show that the locus of z defined by the equation lz-al
lz-bl is
the perpendicular bisector of the line AB, where A and B are the points
in the z-plane which represent the complex numbers a and b respectively.
Deduce that
=

arg

2z-a-b
a-b

10. If z1, z2, z3, and z4 are four distinct points of the complex z-plane
show that they lie on a circle or a straight line if and only if
(z3-z1)(z4 -z2)/(z3-z2)(z4 -z1) is real..
11. If x0 + iy0 is one complex root of the quintic equation z5 1
=

1.
prove that 4xoMi -x6)
Show further than the other roots are x0 -iy0 and those of the cubic
0.
z3+2x0z2 + (3x y)z+4x0(x -y )
=

(U.C.N.W. 1964)

CHAPTER 3

Analyticity
We are now in a position to formulate precise definitions of continuity
and differentiability for functions of a complex variable. This we shall
do by direct analogy with real variable theory. As we have already seen

assertions concerning a function f (z) of a complex variable


imply certain properties of the functions
respectively the real and imaginary parts

z x + iy
u(x,y) and v(x,y) which are
of f(z). The implications of
=

continuity and differentiability place severe restrictions on the class of


functions u and

v,

and this circumstanc will be shown to have consider

able importance in two dimensional potential problems.


3.1

General Principle of Convergence

If we have a sequence of complex numbers


said to converge to the limit

z 1, z , ..., zn then Zn is
zn2 2 tends towards

if the modulus of

zero as n tends towards infinity. This definition implies that the real and
imaginary parts of
conversely if

xn

and

zn converge to the limits 9l2


Yn converge to limits then so

J2 respectively;
zn. We can infer

and
does

the truth of these assertions by inspection of the equality

lzn-21

l(xn-.2)+i(yn-J2)1
+J{(xn-.2)2+(yn-J2)2}.

(3.1.1)

Very often it is inconvenient to have to deal explicitly with the limit

because we may know of its existence without having a knowledge of


its value. We may well wish to test a given sequence of complex numbers
for convergence, and a test which involves A. explicitly presupposes a
knowledge of the answer to our quest.
The statement

'lzn-21

-+

0 as n

large the distance of the point

-+

oo' means that as n becomes very

zn from the point

.A. in the complex plane

becomes very small. Thus if we choose some distance


wish, we can cause the separation of
40

zn

e,

as small as we

and .A. to be less than

simply

GENERAL PR INCIPLE OF CO NVERGENCE

41

by choosing n large enough. This is, in point of fact, the basis of the
formal definition of convergence.
The problem which we have set ourselves is to reformulate this
definition to eliminate any direct reference to A..
Now provided A. does actually exist the two points zm and zn can be
made arbitrarily close simply by choosing m and n to be large enough:
this must be so because both zm and zn can be made arbitrarily near to A.
by expedient of choosing m and n sufficiently large. Conversely, if by
choosing m and n sufficiently large the distance between zm and zn can
be made as small as we please then a limit A. does exist. This argument
can be made mathematically rigorous and leads to a result known as
the General Principle of Convergence (G.P.C.).
THEOREM: A sequence z. of complex numbers converges if and only if,

given any positive number e, a number N can be found such that l zm - zn l < e
whenever both m and n exceed N.

.
Let us now turn our attention to the series sn

t, whose terms t,

r=O

are complex numbers. A series with an infinite number of terms is said


to converge provided that the sequence of partial sums sn converges to
some limit s as n tends towards infinity. We call s the sum to irifinity of
00

the series and write s

t,.

r=O

00

The series

= L

00

t, is said to be absolutely convergent if

L l t,I

is con-

r=O

r=O

vergent. If we denote by u n the sum to n + 1 terms of this series of moduli


then we have, form> n, that

l sm-snl = l tn+l +tn+2+

l tn+ ii+ l tn+ 21 +


=

+tm l

+ ltm l

lum-un l

where we have used the result of Exercise 2, Question 7, which is often


referred to as the triangle inequality.
00

Now if

l um - un i

L l t,I

is convergent then it follows from the G.P.C. that

r=O

can be made arbitrarily small by choosing m and n sufficiently

large. Clearly I sm- snl can be made as small as we wish by choosing


m

and n large enough. Thus, by the G.P.C., sn is convergent to some

limit s. In other words we have proved that an absolutely convergent


series s convergent.
4

42
But

[Ch. 3

ANALYTICITY

CX)

L jt,j

r=o

is a series of real terms so that we can use the usual con-

vergence tests for series of real positive terms to test for the absolute
convergence of series of complex terms.

power series.

An important example is that of


function

f(z) is

Suppose that the

defined as the sum to infinity of a series of powers of

f(z)= a0+a1z+a2z2+
The series for

f(z)

CX)

= L a,z'

z:

(3.1.2)

r=O

will be absolutely convergent provided the series

of real terms

is convergent. We can use the usual ratio test to conclude that

f(z)

is

absolutely convergent if
Lim
r-+oo

a, z 1
+i
=A.<
a,z'

(3.1.3)

l.

The condition (3.1.3) for absolute convergence may be restated as


follows:

lzl <
Thus we see that
hence convergent) for

1 a,+ 1 1 =

Lim

r-+oo

(3.1.4)

p.

f(z) will certainly be absolutely convergent (and


lzl < p where p is called the radius of convergence

of the power series.


It can be further shown that for

lzl >

p the series for

f(z)

is not con

vergent. No rule can be given regarding the convergence of the power


series when
Example 1.

lzl =

p.

Find the radius of convergence of the power series


CX)

l+z+z2/2!+
Now we have

a,= 1/r!

so that p

= L z'/r!.

r=O

Lim(r+l)!/r! which is infinite.


-+oo

r
Hence the given series is absolutely convergent
(and so convergent) for

all values of

z.

It will be noticed that for

z real this series reduces

to the

usual exponential function of elementary algebra We shall denote the


CX)

series

L z'/r! by exp z.

r=O

Sec. 3.2)
Example 2.

If

a,z' is a convergent series, show that

r=O

Lim
n-+0

Let sn

43

CONTINUITY
00

L
r=

'

I
oo

a,z' and let s

a;z'

r=n + 1

'

a,z'. Then, since the given series con-

r=O

verges, lsn-sl--.. 0 as n--..

oo.

0.

But lsn-sl

/ /
=

a,z' which proves the

required result.

3.2 Continuity
The function f(z) is said to be continuous at z
defined for z

z0 iff(z) is uniquely

z0 and, as z tends towards z0, in any manner,f(z) tends

towards the value f(z0). We may write the content of this definition in
concise form as :
Limf(z)

(3.2.1)

f(zo)

In other words lf(z)-f(z0)I may be made arbitrarily small by choosing


the point z sufficiently close to the point z0.
For example let us consider the continuity of/(z)

z*/z.

Writingf(z) in the form u(x,y)+iv(x,y) we obtain in the usual manner

u(x,y)
v(x,y}

x2-y2
-2--2
x +y

(3.2.2)

-2xy
x- 2-+y2

From equations (3.2.2) it is fairly easy to show that, provided z i= 0,

equation (3.2.1) is satisfied: at any rate let us concentrate our attention


on the point z

0.

Let us proceed to examine the limits as z--.. 0 in various ways. We


might, for example, approach the origin in the complex plane along
straight lines of arbitrary gradient. If we put y

mx in equations (3.2.2)

we readily obtain

u(x,y)
v(x,y)

1 -m2
-1+m-2
-2m
.
-1+m
--2

(3.2.3)

44

[Ch. 3

ANALYTICITY

Thus as z-+ 0 along the line y = mx we see that the limit (3.2.1)
becomes

1-m2

z*

Lim-=
-2i--.
z-+0 z
l+m2
l+m2

(3.2.4)

--

Clearly, then, by choosing two different straight lines y


m1x and
m x we can obtain two different values for this limit. Therefore,
2
since the limit is not unique we conclude that the function z*/z is not
continuous at z
0. We could have anticipated this result somewhat
because we have already seen in Section 1.2 that u(x,y) is not a continuous
function of (x,y) at (0,0).
e now ask ourselves the general questions: what does the continuity
ofj(z) imply about u(x,y) and v(x,y): conversely what does the continuity
of u(x,y) and v(x,y) imply aboutf(z)? The answers to these two questions
are summed up in the following theorem:
=

"f

THEOREM: Afunction f(z) is continuous at z = z0 if and only if its real and

imaginary parts are continuous functions of x and y at the point (x,y)

(Xo,Yo).

First we shall suppose f(z) to be continuous at z = z0. In this case


u(x0,y0) and v(x0,y0) are both uniquely specified, and the inequality

0 iu(x,y)-u(x0,y0)I IJ(z)-f(z0)I

(3.2.5)

is satisfied because

lf(z)-f(z0)I = +J[{u(x,y)-u(x0,y0)}2 +{v(x,y)-v(x0,y0)}2].


Now, if we allow z -+ z0 in (3.2.5) it is evident that u(x,y) -+ u(x0,y0).
Thus the double limit

u(x0,y0) =

Lim
(x,y)-+(xo,yo)

u(x,y)

(3.2.6)

exists independently of how we let x-x0 and y - y0 tend towards zero.


Equation (3.2.6) is just a concise statement of the fact that u(x,y) is a
continuous function of x and y at the point (x0,y0).
In exactly similar fashion v(x,y) can be shown to be continuous.
Thus we have shown that the continuity of u and v is a necessary
condition forf(z) to be continuous.
We now suppose that u(x,y) and v(x,y) are continuous and we wish to
show that this is a sufficient condition thatf(z) be continuous at z = z0.
The truth of this is self-evident, for if u(x,y)-+ u(x0,y0) and v(x,y) -+
v(x0,y0), then, clearly,f(z) = u(x,y)+ iv(x,y) -+ u(x0,y0)+iv(x0,y0) = f(z0).

Sec. 3.3]

It is readily demonstrated that if


=

z0

45

DIFFERENTIATION

f+g, Jg,

then so are

f(z)

and

g (z)

are continuous at

and, except perhaps when

g(z0)

0, 1/g.

We shall not prove these results.

Example 3.

r=OI a,z' for lzl

Show that the function f(z) which is defined by the series

00

is continuous.

< p

Using the triangle inequality, we have

l , to a,(z'-)+r=+i a,(z'-z0) '


l,t a,(z'-) 1+ 1 r= +1 a,z'l+l r=+1 a,zol

IJ(z)-f(z0)I =

Now since we are assuming that both

lr=+i a,.z' I l,+i a,.z0 '


n
I ,t a,(z'-z0) 1
and

sums

choosing

lzl

and

lzol

are less than p the

can be made arbitrarily small by

sufficiently large (see Example

2). Furthermore the expression

can be made arbitrarily small by choosing z close enough

to

z0

since

(z'-z0)

is a continuous function of z. Thus

be made as small as we wish by fixing


taking

f(z)

-+

z sufficiently
f(zo).

Example 4.

close to

z0

n at a suitably

IJ(z)-f(z0)I

can

large value and

Hence we may infer that, as

z-+ z0,

Show that exp z is a continuous function of z.

By Example 1 the radius of convergence of the power series defining

exp z is infinite. Hence by Example 3 exp z is continuous for all finite

values of lzl.

3.3 Differentiation
Let us now proceed to an exact definition of differentiation.
If the limit

A.= Lim
z-+zo

exists independently, of how


at z =

z0

and .A. is called the

f(z)-f(zo)
Z-Zo

(3.3.1)

z-+ z0 then f(z) is said to


derivative ofj(z) at z z0.
df(z)
= f'(z) = A.
=

where .A. is a unique, finite, complex number.

be

differentiable

We write
(33

46

[Ch. 3

ANALYTICITY

A one-valued function which is differentiable in the above sense is


called a

regular or analytic function. Some authors make a distinction in

the meanings of these two words using the terminology 'analytic at a


point' and 'regular in a region': we shall not make this distinction and
shall regard the two words as synonymous.
An important consequence of equation (3.3.1) is that any regular
function must also be continuous in the sense of equation (3.2.1). This
is obvious by a method of proof which has been used in mathematics
since its very earliest days-reduction to absurdity. For if we assume
that

f(z) is a regular function but not continuous, equation (3.3.1) gives

us an absurdity: it is inconceivable that the quotient on the left-hand


side of (3.3.1) could possess a limit (finite) as z-z0 -+ 0 without also

f(z)- f(z0)-+ 0. The converse statement is just not true-it is quite

possible for a continuous function to be non-analytic.


In the theory of functions of a real variable there is a very celebrated
example of a curve which is everywhere continuous but nowhere differ
entiable! The reader, if interested, is referred to textbooks on pure
mathematics such as

Theory of Functions of a Real Variable by E. W.

Hobson.

f (z)
g(z) are analytic at z
z0, then so are f+g,fg, and, except perhaps
when g (z0)
0, l/g, with derivatives f' +g', f'g +Jg;, and -g'/g2 respec
It is readily demonstrated, just as in real variable theory, that if

and

tively. We shall not give proofs of these results.


Let us now inquire what the analyticity off(z) implies for the functions
A.
=
z0 a unique, finite value of

u(x,y) and v(x,y). Sincef(z) is regular at z

will be obtained from the limit (3.3.1) regardless of the path chosen in
the z-plane for approaching z0 We are quite at liberty to take a path
=
such that z-z0 is real as z -+ z0, i.e. take z0
x0 + iy0, and z
x +iy0
=

Then we obtain:

u(x,y 0) +iv(x, y0)-u(x0,y0)-iv(x0,y0)


X-Xo

f (z)-f (z0)
Z-Z o
=

u(x,Yo)-u(x0,y0) . v(x,y0)-v(Xo,Yo)
+z
.
X-X
. o
X-X0

If now we let z -+ z0 we find that the value of A. is

avl
+i
.
ax a;J z=zo

A.= au

(3.3.3)

Equally 'well we might have chosen z-z0 to be purely imaginary:


=
=
x0 +iy and, proceeding
x0 +iy0 and z

let us now do this. Take z0

Sec.

3.3]

47

DIFFERENTIATION

as above, we have
. u(x0,y)-u(x0,y0) v(x0,y)-v(x0,y0)
+

f(z )-f (zo)

-l

Z-Zo
and letting z

-+

------

Y-Yo

Y-Yo

z0 we obtain, for the value of A., the result

(3.3.4)
Comparing equations

(3.3.3)

and

(3.3.4)

and equating real and imagi

nary parts (since A. is unique) we obtain the Cauchy-Riemann relatii>ns


au
av
-=-,
ax
oy

(3.3.5)

Thus we have established a fundamental theorem of complex analysis:


THEOREM: If f(z) is analytic at z = z0 then u(x,y) and v( x,y) satisfy the
Cauchy-Riemann relations.
The converse of this theorem is false. The fact that two functions u
and v satisfy the Cauchy-Riemann equations does not necessarily imply
that an analytic function f = u+iv exists. For example we consider the
analyticity at z = 0 of the function
. f(z) =

x3(1 +i)-y3(1-i)
,
2
2
x +y

f(O) = 0.
The reader may readily convince himself that the partial derivatives
all exist and satisfy the Cauchy-Riemann conditions. None the less a
unique va_lue A. of the limit
f(z )-f (O)

Let us take the limit z


-+

0 getting

does not exist.

x4-x3y+y3x+ y4
x4+x3y-y3x+y4
+i
2
2 2
2
2 2
+
y
x
(
)
(x + y )

z-0

then y

(3.3.1)

-+

0 in various ways. First of all we let x


.

Y_
.
f(z)-f(O)
.
- L1m (l+z.)_ 2
L 1m
2
y o
z -0
o
(y )
y-+0
_

...

...

=l+i.

-+

48

x y, then y 0 giving

[Ch. 3

ANALYTICITY

. f(z)-f(O)
.
. 2y4
Lim
=Lim (1+1)
X-+y
Z
y-+0
0
(2Y2)2

Secondly let us take the limit

y-+0

z =0.

=!(1 +i).

Clearly then A. does depend on the way in which


not differentiable at

z0

and so j(z) is

Although the converse of the above theorem is false the following

If two functions u(x,y) and v(x,y) satisfy the Cauchy


Riemann conditions at some point then, provided ou/ox, ou/oy, ov/ox, ov/oy
all exist and are continuous there,f(z) = u(x,y)+iv(x,y) is analytic at that
point.
result can be proved.

PARTIAL CONVERSE:

Show that lzl2 is not analytic except at z=0.


Letf (z)=lzl2=x2+y2.
Then we have u(x,y) =x2+y2 and v(x,y) =0. Hence the
partial derivatives of u and v are

Example 5.

-=2x
ox

first order

-=2y
oy

OU

OU

av
=0
ox

Clearly the Cauchy-Riemann equations are not satisfied except at z =0.

Thus by the above theoremf(z) is not analytic except perhaps at

and are continuous at z=0 the partial converse implies that

z=0.
jzj2

Since the partial derivatives do exist, satisfy Cauchy-Riemann conditions,

shall endeavour to establish analyticity at z=0 by another method.

is

analytic at the origin. However, we did not prove this result so that we
Let us calculate A. :

lf(z)-/(0)
z-0 1

llzl2- 021
z-0

=
1:e:61

1;61
=r,
where z = x+iy =re;6
Now as z 0 in any manner, r 0 so that A. exists and is equal to zero.

Sec. 3.3]

49

DIFFERENTIATION

Show that the real and imaginary parts of an analytic


function satisfy the Laplace equation in two dimensions, i.e. V2u
V2v
0,

Example 6.

where

a2
=

ox2

Hence show that it is possible for u

a2
oy2

x3 +y3 -3x2y-3xy2 +2xy to be a

real part of an analytic function, and calculate the corresponding imaginary


part v.

From equations (3.3.5) we have


o2v

o2u
ox2

o2v

ox oy

oyox

o2 u
oy2

so that
o2u
ax2

o2u

oy 2

0.

In similar fashion we can obtain


o2v
ax2
Now if u

o2v
oy2

0.

x3 +y3 -3x2y-3xy2+2xy we find that


o2u
ox2

6x-6y

and
o2 u
oy2

6y-6x

so that Laplace's equation is satisfied by u and so it is possible that u


is the real part of an analytic function. We next employ the Cauchy
Riemann relations to try to find v.
av
-

oy
ov

ox

OU
=

ax

3x2-6xy-3y2+2y

OU
oy

-3y2 +3x2 +6xy-2x.

Integrating the first of these two equations we obtain

50

[Ch.

ANALYTICITY

where

A(x) is an arbitrary function of x. But, by differentiating this

result, we obtain

ov

ox=

6xy-3y2+A' (x).
ov/ox we see that
x3 -x2 apart from an arbitrary constant.

Comparing this with our previous expression for

A'(x)

3x2-2x. Thus A(x)

Finally we obtain the required result namely

v= 3x2y-3xy2-y3+y2+x3-x2
We conclude this section by proving a theorem on the term by term
differentiation of a power series.
THEOREM:
00

The derivative f'(z) of a functi on f(z) defined by a power series

a,z' with radi us of c onvergence

is given by

00

L
r=O

r=O

ra,z'-1 for lzl

00

Using the ratio test it is evident that the series

r=O

of convergence

p.

Now, proceeding to the calculation of

< p.

ra,z'-1 has radius

f'(z) from first principles, we

have

where

sn

f(z)-f(z0)
Z-Zo

r=O

sn(z) -sn(z0) +
=

L
r=O

r=n+l

a,

z'-z(i
Z-Zo

a,z'.

Let us denote the sums to


00

Z-Zo

n+ 1 terms and to infinity of the series

ra,z'-1 by an and a respectively.

Then we have

f(z)-f(zo)
Z-Zo

a(zo)

Now for

l
l

sn(z)-sn(zo)
z'-z0
+an(zo)-an<zo)- a(zo)
+ I a,
Z- Z o
Z-Zo
r=n+l
00

--

sn(z)-sn(z0)
Z-Zo

aiz0) + Ian(z0) -a(z0) I+

z'-zo
a,--.
r=n+l Z-Zo

L,,

z sufficiently close to z0 the first term on the right-hand side

of the above inequality can be made arbitrarily small. Furthermore,

Sec.

3.4]

51

THE EXPONENTIAL FUNCTION

the second term can be made as small as we please simply by choosing


n large enough.

The behaviour of the third term is a trifle more complicated. However,


we may write:

00

r=n+

L ja,j(jzj'-1+lzl'-2lzol+ ... +lzol'-1).

z0

Now near to the point

we must satisfy two inequalities which

ensure that both the series under discussion are absolutely convergent:

lzl

< p and

that

lzl

jz0j

< p. Thus it will be possible to select a number

< p and

lzol R

I r=n+l
00

L a,

00

Since

such

< p. Then we have

z'-z'0
Z-Z o

--

I r=n+l
00

L rja,jR'-1.

lzl
r=OI ra,z'-1
r=n+L l rja,jR'-1
l,=+l a,(z'-To)/(z-z0)1
is absolutely convergent for

< p it follows from

"-'

tends towards zero as n tends towards

Example 2 that

infinity. Thus we can make

arbitrarily small

by choosing n sufficiently large.

Then, for n sufficiently large, and

sufficiently close to

make the quantity

f(z)-f(zo)
u( z)
Z-Zo

z0,

we can

as small as we wish. Hence we can infer that

u(z0)

f'(z0).

3.4 The Exponential Function


The function exp
00

power series

r=OL z'/r!
z

which reduces, for

which was defined in Example 1 by the infinite


is a continuous function of

(see Example

4)

real, to the usual exponential function of elementary

algebra. We shall now demonstrate that exp z has many properties in


common with its real variable counterpart.

52

[Ch. 3

ANALYTICITY

Using the results of Section 33


. it is clear by term by term differentia00

tion of the power series

L z'/r !

that

r=O

d
dz

( 3 .4 1 )

expz = expz.

We can use this result to establish an important property of expz:


THEOREM: expz1 . expz2 = exp(z1 +z2)for all z1 and z2
Consider the function
j(z) = expz . exp(z0-z)
where z0 is a constant complex number.
Then we have
j'(z) = expz . exp(z0 -z)-exp z. exp(z0 -z) = 0.
Hence we can deduce thatj(z) is a constant. Now we can see from the
definition off (z) that
f (0) = exp0 . expz0
= expz0
since exp0 = 1.
Thus expz . exp(z0 -z) = expz0 for all z.
If finally we write z = z1 and z0-z = z2 we obtain the required
result.
Now, not only does expz reduce for i real to the usual exponential
function of elementary algebra but it obeys the same law of multiplica
tion. It is, therefore, convenient to write expz =

ez

If now we define the functions cosz and sinz by the relationships

cosz =

sinz =

eiz +e-iz

2
eiz -e-iz

2i

(3.4.2)

(3.4.3)

we can readily verify, using the series (2.4.2) and (2.4.3), that these
functions reduce for z real to the usual trigonometric functions of
elementary algebra. It is a simple matter to prove from these definitions
and the properties of
( see Exercise 3).

ez

that the usual trigonometric identities are valid

Sec. 3.4]

53

THE EXPONENTIAL FUNCTION

We can also define hyperbolic functions coshz and sinhz by the


equations
(3.4.4)
ez-e-z
.
smhz=
.
2

Example 7.

(3.4.5)

i
Show that e z= cosz + i sinz and deduce that ez= ex(cosy

+ i siny).

cosz+ i sinz =

ei z

- iz

+i

i - -i
ez e z
---

2i

Hence
= ex(cosy+ i siny)

Example 8.

Show that the equation eZ = ez2 implies

Z1 -Z2=

2Nn

where N is an integer.
If ez= 1 then, by Example 5, we have
ex(cosy+ i siny) = l.

(1)

Therefore we obtain by equating the real and imaginary parts of (1)


ex cosy= 1

(2)

ex siny= 0.

(3)

Squaring (2) and (3) and adding we obtain


e2x

l.

(4)

Since x is real, equation ( 4) implies that x = 0. Thus, from (2) and (3),
we see that y must be an integer multiple of 2n.

Example 9.

If we define tanz as sinz/cosz and tanhz as sinh z/coshz

establish a connection between tanz and tanh iz.

tanz =

=
=

sinz

;-i/

--

cosz

e iz

ejz+ e

2
.sinh iz

1 ---

cosh iz

= -i tanh iz.

jz

54

[Ch. 3

ANALYTICITY

3.S Potential Problems


A fundamental equation in electrostatics is Poisson's equation
(3.5.1)
where

v2 =

a2

a2

ax2

+o

y2

for a two dimensional problem, <P is the electrostatic potential which


gives rise to an electric field with components

(Ex,Ey)

(- -}
,

p is the density of the electric charge distribution, and

In the case of free space p

is a constant.

= 0 and we obtain Laplace's equation V2</J

0,

the solutions of which are called harmonic functions.


THEOREM:

The real and imaginary parts of an analytic function are


harmonic functions (for a proof of this theorem see Example 6 of Section

3.4).
In electrostatics we define a complex potential
w

= </>+it/I = f (z)

where f (z) is analytic. Then as a direct consequence of the above theorem


(3.5.2)
If we now decide to take <P as our electrostatic potential, the curves

</J(x,y) = constant will be equipotentials.


The

t/J(x,y)

function
=

tjJ

is

called

the

charge

function

and

the

lines

constant are clearly the lines of force of the field. This is so

because, using equations (3.3.5) we deduce trivially that


(3.5.3)
and the curves <P

= constant and tjJ = constant are evidently orthogonal,

their gradients being

_a<1> a<1>
ax ay
respectively.

and

_at/I a"'
ax o y

Sec. 3.6]

55

THE LOGARITHMIC FUNCTION

Now ifOJ= <P+ it/I we may write

dOJ o</J .o</J


-=--ldz
OX
oy

ot/J
oy

.or/I
ox

=-+i

dOJ *

( z
d

= Ex+iE1.

(3.5.4)

Equation (3.5.4) gives both the magnitude and direction of the field.
Now the theory of conformal mapping (see Chapter 4) enables us to
set up a correspondence between two different electrostatic problems
using complex potentials, and in this way several complicated problems
are easily solved.
Laplace's equation also plays a leading role in two dimensional
steady current flow problems in electricity and non-viscous, irrotational
flow problems in hydrodynamics. The above techniques are used in
these spheres in much the same manner as described in this section.

Example 10. Find a complex potential to d escribe a uniform electrostatic


field E at an angle ri to the x-axis
The field components may be written down at once

Ex= E cos ri,

E1 = E sin ri.

Then by using equation (3.5.4) we have

-(;)

* = Ee;

dOJ
dz
apart from a constant.
HenceOJ= -Ee-iz

3.6

The Logarithmic Function

Many two dimensional potential problems involve the exponential


function and the closely related logarithmic function. If expOJ =z and
z # 0 we define Log z to be the quantityOJ:

OJ= Logz.

tJ6 I l

56

[Ch.

ANALYTICITY

Let

= u.f-iv and

= rei6 Then we have, in an elementary fashion,

(3.6.2)
Using the result of Example 8 we obtain the relationships e" = r and

v =

0+2Nn

where

is any integer. Hence we may write


u = loger = logejzj

(3.6.3)

v = 0+2Nn = arg

z+2Nn

(3.6.4)

and

Log

= lo&lzl+ i arg z+2Nni

(3.6.5)

where 'log' denotes the usual logarithm of elementary algebra. Log

is thus a many-valued function. Let us define a single-valued logarithm

of a complex number as follows:

log
or

Log

where we recall that arg

Iogelzl

logelzl + i arg z

(3.6.6)

=log z+2Nni

(3.6.7)

defines an angle between

and

+n

and

is simply the logarithm of a real positive number. Clearly log

is discontinuous for
Example 11.

fJ.z

:::;; 0, z = 0.

If f(w) is analytic at w = Wo and f'(w0) =I= 0 show that


w(z) which is supposed to be uniquely defined by
derivative 1/f'(w0) at the point z = z0 where z0
f(w0).

the inverse function

= f(w) has

We denote the inverse function by


dw
dz

=Li

=Lim
= Lim

f-1(z0).

Now

o_
w__w _
f(w)-f(wo)

__

ro-+roo

Z-Zo

z-+zo

Example 12.

w0

m f-1(z)-f-1(zo)

z-+zo

= f-1(z),

f(w)-f(wo)
W-Wo

1/f'(w0).

Find the derivative of log

for

<

arg

<

n.

Now we know that the function f(w) = e"' has derivative e"'. Thus if
= log

log

(one-valued logarithm) then

will be

1/e"'

= 1/z.

e"' and the derivative of

Ex.

J]

57

ANALYTIC ITY

Example 13.

Show that the complex potential which describes a line

charge of unit strength per unit length perpendicular to the (x,y) plane is
w

(X

--log(z-z0)
2
n

where z0 is a constant.
The components of the field are
Ex
where z

(X

- cos (),
2
nr

(X

()
-sm
2
nr

re;o.

Thus

Therefore:
dw

ix

dz

n
2 z

So:
w

(X

n
2

log z +constant

(X

-- log(z -z0)
2
n
where we have chosen w

= oo

at z

z0.

EXERCISE

1. By considering the power series for cos z and sin z show that both
are continuous functions of z for all finite- z. Show further that
d
dz

cos z

.
-sm z,

d
dz

Slll Z = COS Z

2
. Use the definitions (3.4.2) and (3.4.3) of cos z and sin z to show that
2
2
(i) cos z +sin z
1.
=

(ii) cos(z1 +z2)


(iii) sin(z1 +z2)
S'-

=
=

cos z1 . cos z2 - sin z1 . sin z2.


sin z1. cos z2 +cos z1. sin z2

58

[Ch. 3

ANALYTICITY

3. Show that cos z = cosh iz and cosh z =cos iz. Write down the
corresponding relationship between sin iz and sinh z.

(Answer: sin iz =i sinh z)


4. Find the real and imaginary parts of cos z.
(Answer: qt cos z =cos x . cosh y, J cos z = - sin x . sinh y)
5. Show that cos z = 0 if and only if z = (N+!)n where N is an

integer. When is sin z =O?

(Answer:

z = Nn with Nan integer)

6. Show that
jcos z j 2 =cos z . cos z* =!(cosh 2y+cos 2x)
7. Verify that the Cauchy-Riemann equations are satisfied at every
point of the z-plane for the functions z3, ez, cos z.
8. Show that
z-plane.

lzl,

Jz, and arg z are not regular at any point of the

9. If f(z) is a regular function such that


OU

(u2+v2)ox

is constant show that

= (u2+v2)- =0.
oy

Deduce that f(z)is everywhere constant.

10. If z = rei0, and f(z) = u +iv, where

f(z)is

If(z)j

OU

r, (), u

and

are all real and

regular, show that

. au

av

1 av

ar

-;: ae'

u = r2 cos 2(); find the


(Answer: r2 sin W)
1 1 . If f(z) is a regular function show that

A regular function has

(' .

'

ax 2

12. Show that the function

u+iv as a function of
(Answer: u =!{x 2-y2), u +iv

and write

v.

=xy satisfies Laplace's equation and


having

=xy. Find

=!z2)

rO (:)

(cx-r-1)/r!

has radius of convergence unity.

u +iv

z only.

13. Show that the binomial series


= cx(cx-1) ...

corresponding function

\ IJl2 =4lf'l2
ay2)

deduce that there exists a regular function

(:)

au

z' where

and

()

=1

u,

Ex.

3]

59

ANALYTICITY

If f(z) is the sum of the binomial series for lzl < 1


(d/dz)[f(z)/(1 + z)"] 0 and deduce that f(z) (1 + z)" for lzl
=

14.

n-1

Show that

L z'

(l -z")/(1-z) and deduce that, for

lzl

<

1,

1/(1-z).

r=O

00

Verify that the geometric series

15.

show that
< 1.

r=O

00

L z'

Prove that, for

lzl

<

1,.

L z' has radius of convergence unity.


r=O

00

2: lzln cos ne -

n=O

I z I " sm ne

, n =O

1-lzl cos e
1-21Z I COS e + IZ12

lzl sin e
.
1-21 Z I COS e + IZ 12
00

(Hint: Take the real and imaginary parts of

L z")
n=O

CHAPTER 4

Con/ormal Transformations
The type of transformation which is known by the name 'conformal'
plays an important role in many two-dimensional problems in classical
electrostatics, electric current flow, and non-viscous hydrodynamics.
While we do not intend to give a detailed account of any of these appli
cations in this chapter, we shall provide the necessary mathematical
tools for tackling such problems and shall indicate just how one puts
these tools into use.

4.1

Mapping

Consider the equation z= f(O where z and ( are complex numbers.


Such an equation may be regarded as a transformation of points of the
complex (-plane into points of the complex z-plane. That is, given a
point (in the (-plane a point (or set of points) of the zp
- lane is specified.
We say that a region of one plane is mapped by the other. For example,
we say that the equation z= (2 maps the upper half (-plane onto the
whole zp
- lane. We see this by taking real and imaginary parts of the
equation as follows:

x = e2-ri2
Y

= 2e,,

(4.1.1)

where z= x+iy and'= e+iri.


The upper half (-plane is defined by the inequalities

o::::;,,::::; +oo
-oo::::; e::::; +oo.
Thus the equations

(4.1 2)
.

(4.1.1) define quantities x and y which satisfy


-oo::::; x::::; +oo
- CX) ::::; y

::::; + CX).

60

(4.1.3)

61

MAPPING

Consider two neighbouring points of the complex (-plane: ( and (+A(.


Let the corresponding points of the z-plane be z and z+Az (we are
supposing that to each ( there corresponds but one z). If now f(O is a
regular function of ' the derivative rm exists and we may write '
(4.1.4)
the approximation being valid provided that Liz and AC are very small
and r ( O =I= 0. We may take the modulus and argument of equation
(4.1.4) to obtain the results:

IAzl
argAz

lrmllACI

(4.l.5)

argr(C)+ argA(

(4.1.6)

Equation (4.1.5) is a statement of the fact that an element of length is


magnified by a scale factor M
lrml which is called the modulus of
the transformation. Also equation (4.1.6) tells us that the angles which
corresponding elements of length make with their respective real axes
differ by an amount argrm radians. The name conformal transformation
is applied to such a mapping since elements of area are simply scaled by
a factor M2 and rotated bodily through an angle arg rm.
When rm = 0 the transformation is said to be singular and the
conformal property breaks down. In such a case, at the corresponding
point of the inverse transformation (
f 1(z) the derivative d(/dz does
not exist and we are dealing with non-analytic behaviour off- 1(z).
In Section 3.5 we dealt with the idea of a complex potential in electro
statics. Let such a potential function be w = </>+ii/I: we know that,
when w is a regular function of z, V2<f>
V2l/J = 0 so that </> (or l/J) may
be used as the electrostatic potential at a point where there is no charge:
in the presence of charge, however, the electrostatic potential satisfies
Poisson's equation (3.5.1) and consequently cannot be represented by
the real or the imaginary part of an analytic function. Thus the points
at which a potential function w(z) is not regular can be interpreted as
points where there exists some non-zero distribution of electric charge.
If we apply a conformal transformation z = f(0 to the complex
potential w(z) we obtain
=

w(z)

( f ( O)

= w

(4.1.7)

and w(f(()) is a complex potential describing some electrical problem


in the (-plane.
It can be shown (see Exercise 4, Question 1) that
(4.1.8)

62

[Ch. 4

CONFORMAL TRANSFORMA TIONS

Then, provided z

f(O is a regular function of ( with

lf'(()i

=I- 0, it

can be deduced that areas of the z-plane which correspond to charge


free regions of the (-plane are also free of charge. In this way theoretical
physicists have been able to set up relationships between complicated
electrical problems on the one hand and simple well-understood ones
on the other by means of suitable transformations.
In actual practice singular transformations are widely used although
care must be exercised in dealing with the spurious charge distributions
which can arise. This is to say that the use of a singular transformation
may give us charges in one plane which have no counterpart in the other.
For example, suppose that there is a non-zero charge distribution at
z

z0 in the z-plane: if at the corresponding point (0 of the (-plane

/'((0)

0, it is possible that the charge distribution at (

(0 vanishes

(see equation (4.1.8)).


The same techniques are available for dealing with two -dimensional
current flow problems in electricity and hydrodynamics-indeed all
problems based on Laplace's equation are amenable to the methods
described above.

Example 1. Show that the conformal transformation z (", n =I- 1 can


be used to map the sector 0 r oo, 0 (:} mi:, n < 2, onto the upper
half (-plane, where z rei8.
=

dz

d(

nrn-1
.,,

Thus

and
arg Az

arg n(" - 1 +arg A(

arg ("-1 +arg A(

(n- l)arg (+arg A(.

Now it is evident that on the line PA( 0, J(

0, both arg (and arg A(


r, 0 r oo.
0, both arg (and arg A(

are zero so that this line is the map of the line z


It can be seen that on the line PA( 0, J(

are equal to

n: thus at every point of this line the corresponding point


nn: we deduce that the negative real (-axis
is the map of the line z
rei"", 0 r oo.

in the z-plane has arg Az

In giving this solution of the question we have followed the argument


detailed in the above section: the result is more readily obtained by

Sec.

4.2]

JOUKOWSKI'S TRANSFORMATION

63

simply considering z= 'n in undifferentiated form:

argz

narg,.

From these equations it is at once evident that the effect of the trans
formation is to rotate radii through the origin by the prescribed amount.

4.2

Joukowski's Transformation

One of the most famous problems which has been treated by means
of conformal transformations is the problem offluidflow past a Joukowski

aerofoil. Such an aerofoil section is obtained from a circle by means of a


certain conformal mapping which we shall study below.
This is a two-dimensional problem, and so as a model of an aircraft
wing it is unrealistic in that the model wing is of uniform cross-section
and infinite in extent. This defect in the model is not too serious if we
stipulate that any conclusions reached refer only to parts of the aerofoil
which are distant from the wing tips: in this way the end effects will be

of minimal importance.

Let us study the transformation


z
If we write z= x + i
parts of equation

'+

a2

(4.2.1)

4(

and ' = Rei9, and then equate real and imaginary

(4.2.1)
y

we find that

;;)
(
:)

x= R+

= Ry

cos0

(4.2.2)
sin e.

Now the circles R= constant in the '-plane correpond in the z-plane


to a family of confocal ellipses:

( x'r ( r
R+4R

1.

(4.2.3)

y
R-R
4'.,

As is well known, the ellipse may be written in standard form as

x2 Y2
+ =l
A1 B2
where the foci are the points

( .J(A 2-B2),0).

(4.2.4)

64

CONFORMAL TRANSFORMA TIONS

[Ch. 4

Comparing equations (4.2.3) and (4.2.4) we see that the foci of the
family of ellipses under consideration are ( a,0).
Now the mapping of the z-plane onto the (-plane defined by (4.2.1) is
not one-to-one. While each point of the (-plane gives rise to a unique
point in the z-plane, it is not true that each point of the z-plane corre
sponds to a unique point in the (-plane. We can see this very simply by
noticing that the points (R,0) and (a2/4R,-0) of the (-plane both give
rise to the same point (x,y) in the z-plane (see equations (4.2.2)). Thus
the region outside the critical circle R

a/2 is mapped onto the whole

z-plane. So also the region inside the critical circle corresponds to the
whole z-plane.
What happens is this: as the radius R of circles in the (-plane increases
from zero the confocal ellipses in the z-plane become smaller and smaller
in area until, when R

a/2, the system of ellipses collapses into the

straight line joining the foci ( a,O): then as R increases further, the
ellipses increase in area until ultimately they again occupy the whole
z-plane (see Fig. 4.1).

2o
FIG. 4.1

To proceed further with our studies it is useful to rearrange the


transformation (4.2.1) into the form:

z+a
z

' +a/2 2
C-a/2

(4.2.5)

Sec. 4.2]

JOUKOWSKI'S TRANSFORMATION

65

Similarly let us set ( +a/2 = R1e;e, and (-a/2 = R2e;8 2 (see Fig. 4.3).
If we use this notation the formula (4.2.5) becomes
r1 ei<B1-Bi>

()
R1

Ri

2
e i(e, -e2>.
.

(4.2.6)

The loci 61 -62 = cf>= constant and 01 - 02 =<I>= constant are


circles in the z and (-planes respectively. Let us consider a circle in the
p

A
-a

FIG. 4.2

(-plane which passes through the critical points R1 = 0 and R2 = 0 as


shown in Fig. 4.4. If we trace out this circle in an anticlockwise sense
starting at B' we notice that, at the point A', the angle 01 changes
P'

FIG. 4.3

discontinuously by an amount n. At the point A, in the z-plane, however,


an inspection of equation (4.2.6) tells us that the angle 61 also changes
discontinuously but by an amount 2n. Thus as we trace out the lower
arc A'B' in the (-plane, the corresponding point in the z-plane merely
retraces the arc AB. In particular if the circle in the (-plane has centre
the origin it will be the critical circle and the arc AB becomes a straight
line (see dotted loci in Fig. 4.4).
If now we draw a second circle in the (-plane to touch the original
circle at the point A' and completely surrounding it, the map of this
circle onto the z-plane will be some curve which entirely surrounds the
arc AB (see Fig. 4.5). If the original circle is the criical circle then its
map onto the z-plane will give the symmetrical picture drawn in Fig. 4.6.

66

CONFORMAL TRANSFORMATIONS

[Ch. 4

FIG. 4.4

The curves obtained in this manner are called Joukowski aerofoil


sections.
The sharp edge at point A which corresponds to the critical point A'
can be eliminated by choosing a circle which passes near to A' in the

FIG. 4.5

(-plane without actually passing through it. An infinite variety of


similar shapes can be obtained by varying the size and orientation of
our circles in the (-plane.

FIG. 4.6

This transformation allows the problem of flow past a Joukowski


aerofoil to be related to flow past a circular cylinder-this latter problem
being very well understood in hydrodynamics.

Sec. 4.3]
4.3

67

THE SCHWARZ-CHRISTOFFEL THEOREM

The Schwarz-Christoffel Theorem

In Example 1 we observed that the transformation z


the sector 0

(}

mt

(" mapped

of the z-plane onto the upper half (-plane. This is,

in fact, a special case of a very powerful theorem which shows how any
polygonal arc may be mapped into a straight line.
The transformation which is attributed to Schwarz and

Christoffel

has differential form

dz
-

d(

CT ((-e,)r
r=

(4.3.1)

where A is a complex constant,

a, = ((}, - n)/n, each e, is real, and each


(}, 2n. It will be shown that this trans
, (Jn
formation maps a simple closed polygon with interior angles (}1, (}2,

(},

is an angle satisfying 0

in the z-plane int the real axis of the (-plane, the points '

e, cor

responding to vertices of the polygon.


Now it is at once apparent that, at (

(4.3.1) is either zero

e,,

the right-hand side of

or infinity so that at these points the transformation

is not conformal. Because of this circumstance we will consider not the


real axis of the (-plane but a curve r which coincides with the real axis

at all of its points except those in the vicinity of the points (


the points '

..

e,.

Near

r differs from the real axis in that it has small semi

circular indentations which by-pass the points (


clearly illustrated in Fig.

4.7

e,.

This matter is

where, for convenience, we have assumed

FIG. 4.7

that

ei <e2 < <en

We shall, of course, be interested in the limit in

which r tends to coincidence with the real (-axis. Now we shall show
that r corresponds in the z-plane to a polygon-like figure (see Fig.

4.8)

which has small indentations, in the shape of arcs of a circle, instead of

vertices. In the limit as r tends into coincidence with the real (-axis this
polygon-like figure will tend to become the polygon

P 1P2 ... Pn.

Further

more we shall demonstrate that the deviations which r makes into the

68

[Ch. 4

CONFORMAL TRANSFORMATIONS

upper half (-plane correspond in the z-plane to excursions into the


interior of the polygon; we shall conclude, therefore, that interior points
of the polygon are mapped by points of the upper half (-plane.
y

Jt

FIG. 4.8

Let us examine in some detail the neighbourhood of a typical point

, en which lie on the real (-axis. On the semi


( = e. of the set e1, e2,
circular indentations we write ( - e. = Rei8 where 0 e n : this
clearly does give us a circular arc since (e- e.)2+172 = R 2 Now since R
.

is small we may treat (( - e,) as a constant when ( = e. + Rei8 for all


except
r = s

r = s:

in other words we set (( -e,) = (e. -e,) for all

except

when we write (( -e,) = Rei8. In this way it will be possible to

rewrite equation (4.3.1) as follows:

dz
=
d(

BRa exp(ieix.)

where Bis some new complex constant and

dz
d(

ix.=

(4.3.2)
(O.-n)/n. Now

dz de
dEJdf

so that equation (4.3.2) may be modified to become


dz

de

iBR61" exp(ieOJn).

(4.3.3)

Now it is evident that equation (4.3.3) could be obtained by differentia


tion from
z = z.+

n
.
BR6'" exp(zeO.Jn)
0
s

where z. is a constant.

(4.3.4)

Sec. 4.3]
As R

69

THE SCHWARZ-CHRISTOFFEL THEOREM


-+

0, z

-+

z, and we shall identify the point z = z. with the

vertex P, of the polygon in the z-plane.


Taking the argument of equation (4.3.4) we can derive in a straight
forward manner the result

ee

(4.3.5)

arg(z-z,) = argB+--
7t

Now on the semicircular arc of r which is near (

e. in the (-plane

the angle e decreases from e = 7t to e = 0 as we move around '

e.

from left to right (see Fig. 4.7). Thus in the z-plane we see from equation

(4.3.5) that arg(z-z,) decreases by an amount e. (see Fig. 4.9 and


Exercise 2, Question 8).

',
'P,$

,.,.
,.,.

'
'

/arg 8

'

',

arg B +15

FIG. 4.9

On each section e.- i < ( < e. of the real (-axis we can see that
arg z is a constant because

argz = argA+

(4.3.6)

(0,-n).

r=s

Equation (4.3.6) can be deduced from equation (4.3.1) by writing


argz-arg(

argA+

n (} -n

r= 1

1t

and noting that, for e.- 1 < ' < e., arg '
r =

1, 2, .. ., s- 1, and arg(( -e,)

What

have

we

= 7t

(4.3.7)

'- arg((-e,)
=

0, arg((-e,)

0 for

for r = s, s + 1, ..., n.

learned? We can now

assert

that

the

sections

e.-1 < ' < e. and e. < ' < e.+ 1 of the real (-axis correspond to the
straight lines P._ 1 P. and P.P.+ 1 in the z-plane which meet at an angle e.
Before we can say that at the polygon in z-plane does in fact map
into the real (-axis under the Schwarz-Christoffel transformation we

70

CONFORMAL TRANSFORMATIONS

[Ch.

must explain how the portions ( < el and ( > en of the real (-axis
correspond to the side PnP1 of the polygon in the z-plane. Let us draw
a large semicircle of radius R as shown in Fig.

4.10; on this semicircle

(
1'1<;. 4.10

we may write (

E>

Re;9, 0

n.

may make the approximation (( - ,)


can be rewritten as

z
d(

Now because R is very large we


=

Re;e for all

Equation ( 4.3.1)

r.

AR -2 exp(-2iE>)

where we have use9 the fact that the angles 91, 92,
give

(4.3.8)

, (Jn add up to

(n-2)n radians. Proceeding in much the same manner as above


we can modify equation (4.3.8) so that it reads
d

dE>

iA -;a
e

(4.3.9)

which can be obtained by differentiation from

(4.3.10)
where we shall interpret the constant z0 as some point P0 on Pnp1

(consider the limit R-+

as indicated by Fig.

co). Now as ( goes round the large semicircle,


4.10, E> increases by amount n so that arg(z -z0)

decreases by amount n because


arg(z-z0) = argA-E>.

(4.3.11)

This corresponds to the point z on the side PnP 1 of the polygon going

round the point P0 (see Fig.

4.11).

Sec. 4.3)

THE SCHWARZ-CHRISTOFFEL THEOREM

71

FIG. 4.11

Show that by suitable choice of A we may adapt the Schwarz


Christojfel formula to allow en = oo.

Example 2.

A = A'( -en)-a" in equation (4.3.1) we obtain:


"flt
dz
' ((-en\ a"
=A
(( - e,r.
d(
\-en!
we let en
00, { ((-en)/( -en)}
1 and we have

Setting

r=l

If now

-+

-+

the required

result:

dz
d(
If

en

-+

= A'

"f{ (( - e,r.
r=l

00 we can simply forget about it!

Use the Schwarz-Christoffel theorem to transform the shaded


region in Fig. 4.12 into the real axis by treating it as a simple closed
polygon whose vertices P1 and P3 are at infinity.

Example 3.

FIG. 4.12

Using the formula (4.3.1) we obtain

dz
d(

where we have chosen

A((-e2>-1'2((-e3)-1((-e4)112
e1

oo. and used the result of Example 2. Let us

CONFORMAL TRANSFORMATIONS

72

. [Ch. 4

set e = -1, e3 = 0, and e4 = 1. Then we have


2

dz
= A c1
d(
.

(-1
'+ 1

which can be integrated by elementary methods.


The physical significance of this particular example involves the
interpretation of Fig. 4.12 as the cross-section of the bend in a Leyden
jar. A full account of this problem can be found, for example, in Electricity
and Magnetism by Sir James Jeans.

EXERCISE

1. By using the identity


o
oe

ox

oy

oe ox+ oe oy

show that

02
oe2

(ox

\)e

()

02
02x o 02y o
oy 2 02
ox oy 02
. OX2 + 2 oe. oe. oxoy+ oe . oy 2 + oe2 OX+ oe2 oy.

If z = x+ iy is a regular function of ( = e i11 use the Cauchy-Riemann


+
relations to show that

(:;2 ::i) {() () } (::2 :;2).


+

2. Show that the transformation


z=

(+1

-(-1

maps circles in the (-plane into circles in the z-plane. What exceptions
are there?

(Answer: The ratio of xercise 2, Question 10, is unaltered by the


transformation, so that

circle in the (-plane becomes a circle or a

straight line in the z-plane. Circles which pass through ( = 1 are mapped
into straight lines because their images are unbounded)

3. Show that the transformation z = (2 transforms the right half-plane


fJt( 0 onto the whole plane, and third quadrant fJt( 0, J( 0 onto
the upper half-plane Jz 0.
4. Show that the transformation
z=
maps the half-disc J( 0,

1(1 1

( )

( +l 2
(-1

onto the upper half-plane Jz 0

Ex. 4]

CONFORMAL TRANSFORMATIONS

73

5. Show that the strip - oo x oo, - nc < y nc, is mapped by


the transformation z = c log(, where c. is real, onto the whole (-plane
while the strip
oo x oo, 0 y nc, is mapped onto the upper
half-plane J( 0.
6. Show that the polygonal arc r which comprises the straight lines
e= tn, 0 ,, 00 and -tn e tn. ,,= 0 is mapped by the
transformation z1= sin ( into the real axis of the complex z1-plane.
Show also that the transformation z2= cot tC+( maps r into that
portion of the real axis in the z2-plane which satisfies z
l 2I !n+1
together with the curves x2= Hnsech 1/). Yi= ,,+tanh ,,, with
'I 0. Sketch the complex zi-plane.

Note: In hydrodynamics this pair of transfonnations has been used to relate the
problem offluid issuing from a slot of width n + 2 in the z2-plane to a flow problem
in the half-plane Jz1 0.

7. Find a transformation which maps the strip 0 y 1, x 0 into


the real (-axis.
1
(Answer: dz/d(= A((2-1)- '2 where Ji is a constant and the points
(0,0) and (0,1) in the z-plane correspnd to the points (-1,0) and (l,Q)
in the (-plane)
8. If D denotes the set of points which lie inside the circle l z -1I = ,J2
and outside the circle lz+1 I = ,J2 find the image of D under the trans
formation
'=

{;:}

Also find the region of thez-plane whose image under this transformation
is the circular disc ICI < 2. Evaluate

fl

dx dy.
(1+ +y2)2

, (Answer: e 0, - e 1/ e; l3z +Sil

> 4; n/4)
(DURHAM 1958)
9. If m= u+iv, z = x+y
i show that under the transformation m2= z
the region y2 > 4(1-x) of the z-plane corresponds to the regions u > 1
and u < -1 of the m-plane.
If

'= --1
<.O

determine what regions of the (-plane correspond to the region of the


z-plane given by y 2 > 4(1-x).
(Answer: ICI < 1, IC+ 21 < 1)
(HULL 1962)
Note: Further examples of conformal transformations
Dictionary of Conformal Representations by H. Kober.
6

can

be found in the

CHAPTER 5

Integration
The idea of integration as the inverse process to differentiation has
already been smuggled into this book in Chapters 3 and 4, but the

point in our study of complex variable theory has now been reached
when we must discuss in general terms what we mean by the integral
of a function of a complex variable.
5.1

The Limit of a Sum

In Section

1.5

we undertook to generalize the usual 'limit of a sum'

type of definition of an ordinary single integral to embrace several

real variables. We shall now use our experience in this matter to set up

a natural definition of the integral Jc f(z) dz of a complex function f(z)

with respect to z along some path C in the complex z-plane. Let us

suppose that the curve C is a simple curve such athat drawn in Fig.
Zn

5.1

FIG. 5.1

and let us divide it into

f,

portions by points z0, z 1,

. , z"

as shown. If

now we choose any point zt'' of the curve which lies between z,_1 and
z, we may write

m I /(z<'l)(z,-z,_1)
f(z)dz =Li
n-+cor=l
74

(5.1.1)

75

THE LIMIT OF A SUM

where the limit is taken as (z,-z,_ 1)

-+

0 for all

r. There is, of course,

no need for such a limit to exist: when it does,f(z) is said to be integrable,


and the expression (5.1.1) defines a

contour integral.

Now by trivial manipulation we can rewrite the sum on the right-hand


side of equation (5.1.1) as
n

L {u(x<r>,y<'>)(x,-X,_ 1)-v(x<r>,y<'>)(y, - Y -1)}


r
r= 1
n

+i L {v(x<'>,y<'>)(x,- x,_1)+u(x<'>,y<'>)(y,-y,_ 1)}


r=l
which leads us to conclude that

ic
wheref(z)

f(z)dz

fc

(udx-vdy)+i

lc

(vdx+vdy)

(5.1.2)

u(x,y)+iv(x,y) and the line integrals of(5.1.2) are interpreted

in the sense of Section 1.8.


We shall normally use the equation (5.1.2) as our working definition
of the integral

Scf(z)dz.

Evaluate Scdz where


a and z = b.
//

Example 1.

dz= Lim

1
r=

1 . z,-z,_1)

b-a.

Evaluate Sc zdz where


a and z = b.

Example 2.

points z
If I
n

Sc z dz

is a curve running between the points

is a curve running between the

then I does not depend on which point z<r> we choose in

L z<'>(z,-z,_ 1) to define the integral. Taking z<r>


1

r=

z, we obtain

Lim

L z,(z,-z,_ 1 )
r= 1

(1)

whereas if z<r> = z,_ 1 we have


n

I = Lim

L z,_1(z,-z,_1).
r=l

(2)

[Ch. 5

INTEGRATION

76

By adding (1) and (2) we finc;l that


2I =Lim

(z;-z;_1)

r= 1

= z;-z

S.2

Cauchy's Theorem

In Examples 1 and 2 we notice that if C is a closed curve (a


both

b) then

fc dz and fc z dz vanish identically. These results are special cases

of Cauchy's theorem which we shall now discuss.

THEOREM: Iff(z) is regular, and f'(z) continuous , both inside and on a


simple closed curve C then

f(z) dz

(5.2.1)

0.

We have seen that the regularity of f(z) implies its continuity (see
Section 3.3) and so by the theorem of Section 3.2 we can conclude that
u(x,y) and v(x,y) are continuous functions of x and y. Also we have been
given thatf'(z) is continuous which impliet ou/ox, ou/oy , ov/ox, and
ov/oy are continuous. Thus Green's theorem (Section 1.8) asserts that

(u dx-v dy)

(v dx+u dy)

Jl (!: :)

dxdy

Ji(;:;)

dxdy

and

where A is the. region of the z-plane Whose boundary is the curve C.


But the Cauchy-Riemann equations (3.3.5) tell us that both the integrals
in the above double integrals vanish because f(z) is regular.
Hence we have provd the result stated in equation (5.2.1).
An important corollary of Cauchy's theorem is the following:
GoROLLARY: Iff(z) is regular in an annular region between two closed
curves C 1 and C 2 then

J. f(z) dz J. f(z) dz.


:rc2
:rc. =.

(5.2.2)

Sec. 5.2]

CAUCHY'S THEOREM

77

The closed contour C drawn in Fig. 5.2 coincides, apart from a narrow
neck P1P2, with curves C1 and C2 shown in Fig. 5.3. Now since /(z)
is regular in the annular region (shaded in Fig. 5.3) it must be regular

FIG. 5.3

FIG. 5.2

inside the closed contour C (shaded in Fig. 5.2). Thus by Cauchy's


theorem we must have

f(z)dz= 0.

Now, using our knowledge of the properties of line integrals, we


can split the integral !OUnd C into parts : then, letting the two portions
of the neck P 1P 2 shrlnk into coincidence, we obtain the required result
as follc:Jws :

(l

_J. + f + f
J. f(z)dz=
Jc
Jc, Jc, JP,P2 JP2P1
=

(, -f )
,

f(z)dz

f (z)dz

= 0.

A point wher.e f(z) is non-analytic is generally called a singular point


or simply a singularity off(z).
Example 3. Show that, Jc.f(z)dz= Sc,!(z)dz where C1 and C2 are
any two curves. which join the points a arid b of the complex plane and do
not enclose between them any singularity of the integrand f(z).
'

The result follows immediately from Cauchy's theorem by treating


the two curves as one closed curve. If the curves C1 and C2 intersect
one another we simply apply Cauchy's theorem repeatedly to the several
t
closed regions (see Fig. 5.4).

78

[Ch. 5

I_NTEG RATION

It is clear that when we are dealing with the integral of a function of a


complex variable we may deform our contour of integration in any
manner we please without altering the value of the integral provided

FIG. 5.4

that we do not interfere with the end points of the contour and provided
that we do not cross a singularity of the integrand.
Example 4.

Show that

contour which surrounds z

fc dz/zn+
=

0 for n

>

0 and C any closed

0.

Cauchy's theorem cannot be invoked here because, as the reader may


verify, 1/zn+ 1 is not regular at z

0. There are, however, no other

singularities of 1/zn+ 1 so that C can be any cuyve which surrounds

z
z

==

0. Let us choose a circle whose radius is r: in other words let us set

rei11:

where we have used the fact that an integral is the limit of a sum to
infer, from the triangle inequality, that the modulu of an integral is
less than the integral of the modulus of the integrand. Thus we have the
result

which we cannot recolici1e with the fact that the left-hand side is independent of r unless

f,c 4z/zn+

0 because as r becomes very large the;

right-hand expression tends towards :zero.

Sec.

5.3]

79

'
CAUCHY S INTEGRAL FORMULAE

Example 5.

Evaluate l dz/z round any closed contour


:t'c

C.

There are two cases to consider since the contour may or may not
enclose the singular point z= 0. If C does not enclose t=0 the matter is
'trivial and
a

=0

f dz/z=0 by Cauchy's theorem. If, however, C does enclose


c

we use the method of Example 4:

"

f -dz=i J0 d8
=
..
1

Example 6.

2ni

If f(i) is a (continuous) function of a complex variable z


such that lf(z)I is bounded above by M on C show that IScf(z)dzl Ms
where s is the length of the curve C.

dy

FIG. 5.5

From Fig.

5.5

we s that (dx/dt)2 +(dy/dt)2 =(ds/dt)2 so that, if we

int_roduce a real parameter

11

t,

we may write

/ I L f(z)( + i ) dt l
{ 1f(z)1J{() 2 + (r} dt
L M ds

/(z)dz =

=Ms.
5.3

Cauchy's Integral Formulae

THEOREM: If f(z) is regular inside and on a simple closed contour


and if z0 is any point within C, then
f(z0)= 1
2m

f(z)

Jc z-z0 dz.

C,

(5.3.1)

80

[Ch.

INTEGRATION

First of all we remark that the integrand on the right-hand side of

equation (5.3.1) is regular at all points inside C except perhaps at

Thus we can replace the curve C by a small circle of radius

z= z0
r which

z= z0 Also because of the regularity of f(z) we may infer


z z0 Hence for z on the circle of radius r we can make
lf(z)-/(z0)I as small as we wish by choosing r sufficiently small. In

surrounds

its continuity at
other

words,

l/(z)- /(z0)I

<

given
e

any

small

by choosing

number

e,

we

r0

can

make

5, we have
/(z0) + /(z)-f(z0)\ d
J.
z
Jiz- z o l=r z-zo Z-Zo )

But, using the result of Example

J. /(z) dz=
jcz-zo

positive

small enough, say

Hence, by Example 6, we may deduce that

f(z)-f(zo) d
zl
I
_fl:Ldz-f(zo) I 1J.
J.
2m fi1z-zo -r ,.z-z0
2m J; z-z0
=

1_

- .-. 2nr
2n r

e.

The right-hand side of inequality

for

r0
(5.3.2)

(5.3.2)

is completely arbitrary and

can be made as small as we please. The left-hand side, however, does


not depend on

and so our only consistent conclusion is that this

expression vanishe.

It may be shown further that the derivatives of a function f(z0) which

(5.3.1) all exist and are given


f(z) d
z.
f(n)(zo)= 2-.
,
1tl
(
Z-Zo)n+l
C

can be expressed in the form

n! f

by

(5.3.3)

Sec.

CAUCHY'S INTEGRAL FORMULAE

5.3]

If now

z1

-+

z0

81

we obtain an expression for j<k+,1l(z0):

j<k+ll(zo) = ,(f

211:i'j

(z)(k+1)(z -z0)k
dz
(z - zo)2k+ 2

(k+ l)!,( /(z)


=
2ni J (z -z0)k+ 2 dz.
The reader may very easily check the truth of formula

k =1

and deduce the truth of the formula for all

(5.3.3)

for

by the method of

mathematical induction.
The formulae

(5.3.1) and (5.,33) are

Cauchy's integral formulae.

called

Let us consider the region A of the z-plane which is bounded by the


two concentric circles

C1

and

C2

drawn in Fig.

5.6

and let us suppose

FIG. 5.6

that A is embedded in a region where f(z) is a regular function. Then


if

z0 be

the common centre of these. two circles and

z1

be any point of A,

we shall prove that

f(z1) =

n=

+oo

L ah1 -zo'f

n=

(5.3.4)

co

with

an
where

-2
7tl

f(z)
+ 1 dz
(
z
c
-zor

is any circle concentric with

C1

and

(5.3.5)
C2

'
and tying between

them.
Now suppose that we draw a circle

C3

z1 lying wholly within


5.7, 5.8 and 5.9 it is evident

centre

the region A. From a consideration of Figs.

82

[Ch. 5

INTEGRATION

that the curve in Fig. 5. 7 can be deformed without crossing a singularity


of

/(z)/(z-z0) into the circle C3 depicted in Fig.

FIG. 5.7

5.9. Hence we may use

FIG. 5.8

FIG. 5.9

5.2 to deduce that


,( f(z) dz= ,( f(z) dz - ,( f(z) dz.
Jc3 Z-Z1
Yc , Z-Z1
!, z-z1

the arguments of Section

But, by Cauchy's integral formula (5.3.1), we have

f -f(z) dz= 2nif(z1).


z-z1
c,

Hence we may write

,( /(z) dz- J. f(z) dz ,


f(z1.)=
2m !, Z-Z1
2m !. Z-Z1
1
= -m
2 .U1 +I2).

(5.3.6)
say.

CAUCHY'S INTEGRAL FORMULAE

Sec. 5.3]

83

Now, using the formula for the sum of geometric series, we may
rewrite 1 1 and 12 as
11

12

,( f(z) i+(z1-zo)+ ... +(z1-zo>" az+R1


(z-z0)"
fc, (z-z0) (z-z0)
,( f(z) + (z-zo) + ... + (z-zo)" dz-R2
(z1-zo)"
1, (z1 -zo) (z1-zo)

}
}

{i

where the 'remainder' terms

R1
R1

R1 and R2 are given by the expressions


1
/(z)(z1-zor+ d
z
1
(Z-Zo)"+ (z-z1 )
1
/(z)(z-z0)"+ d .
z
(z1 -z0) (z-z1)

f,
f

c,

n+l

c,

The radius of the circle C 1 clearly exceeds the distance of z1 from

lz1-z0I < lz-zol Thus, as n -+


0 and we conclude that R1
0.

so that on the circle C1 we may write

l(z1-z0)/(z-z0)I"+1

-+

z0
oo,

-+

Also the radius of the circle C2 is always smaller than the distance

of

z1 from z0 so

that, on the circle C2, we may write

i(z-z0)/(z1-z0)j"+1 ...:+ 0

Then, as n -+ oo,

R2-+ 0.

lz1-z0I > lz-zol

and we may argue that

Thus we expect that/(zi) may be written in the form

2nif(z1)

oo

n=O

f f(z). . (z-z

)"

dz+ L
(Z Zo) +

c,

oo

n=l

f f(z)(z-z_0)"- 1 d..z
c,

(Z1 Zo)"

The above argument / n in fact be made the basis of a mathematically


rigorous derivation of equatiort (5.3.7).
Now since there is no singularity of the integrand in equation (5.3.7)

between C1 and C2 we may deform both these contours (see Section (5.2))

into a single contour C which lies somewhere between them to obtain


.

2nif(z1)

(z1-z )"
,(1 /(z)(z '.... zo)"o+l

n=+
<XJ
n

<X)

(5.3.8)

which is just the result stated in equations (5.3.4) and (5.3.5).


It is quite evident that the curve C need not be a circle and could

be any curve whatsoever which lies wholly inside A and encircles the
point

z0.

INTEGRATION
[Ch. 5
n= + oo
L an(z-z0r is called the Laurent expansion of the
The series
n= - co
function f (z) about the point z0.
Let us now suppose that f(z) is actually analytic at each point inside
C1 (see Fig. 5.6) except z
z0: in such a case the radius of C2 may be
made arbitrarily small: f (z) possesses a Laurent expansion about the
point z = z0 for every z in the neighbourhood of this point We say that
f (z) has an isolated singularity at z = z0.
If f(z) is regular at every point inside C1 (see Fig. 5.6) including
z = z0 then the expression (5.3.5) for an vanishes for n - 1 because
the product f(z). (z-z0)-n-l of two regular functions has zero integral

84

(by Cauchy's theorem) round any closed contour in A. Furthermore we


can rewrite the exJ"<:SSion
formulae

(5.3.5) for
(5.3.1) and (5.3.3), to obtain
f (z)

wherej<n>(z0) denotes

"'

I pn>(zo)

n=O

the nth derivative

n 0, using Cauchy's integral


/

(z-lzor

(5.3.9)

n!

ofj(z) at z

of a regular function is called the Taylor series for f

z0 This expansion
(z) about z z0.
=

At this juncture it is convenient to introduce some special terminology.

n= -1
L an(z-zor in a Laurent expansion about an isolated
n= - oo
singularity at z
z0 is known as the principal part of f(z) and the
coefficient a_ 1 is called the residue of f (z) at z
z0
The series

If the principal part terminates (let us say that an

then we refer to the isolated singularity at

z0

0 for all n < - p)

as a pole (of order p).

We talk of simple, double, and triple poles when p =


spectively.

1, 2, and 3 re

If the principal part is non-terminating then we speak of an isolated

essential singularity ofj(z) at z

z0.

Example 7. Show that if a function f (z) can be expanded in a series


n= +oo
L biz-z0)" then bn =an where On is defined by equation (5.3.5).
n= - oo
We have from (5.3.5) that
f(z)
2man = .
+ 1 dz
C ( Z-Zo r

. f
:::
=

b,(z-z0y-n-l dz.

Now if it is permissible to interchange the order of performing the


summation and integration operations we obtain the required result

Sec. 5.3]

'
CAUCHY S INTEGRAL FORMULAE

85

by using Examples 4 and 5 (or Exercise 5; Question 3):


2nian

i:.

r= oo

r= oo
-

b,

J. (z-z0)r-n-t dz
Ye

= 2nibn.
To endow this 'proof' of the uniqueness of a Laurent expansion with
mathematical rigour would involve us in a short discussion of uniformity
of convergence. We shall not proceed further in this matter, but, if the
reader is interested, he may refer, for example, to Theory of Functions
of a Complex Variable by E. T. Copson.
Example 8.

Use the uniqueness of the Laurent series about z = 0 to

calculate the principal part and residue there for co sec z and sec z.

1
cosec z = -sin z
1
z-z3/3!+

1
z(l-z2/3!+
=

(i
+

+higher powers of

This last step has been made using the binomial theorem which we know
to be valid for lzl

<

1 (see Exercise 3, Question 13). Since the Laurent

expansion is unique it must be that so found. The principal part is thus


1/z and we conclude that the residue of cosec z at z = 0 is + 1.
Similarly we find that
z2
sec z = 1+ ! +higher powers of z
2
so that sec z has no principal part and hence its residue at z = 0 is zero.
Notice that ec z is regular at and near z

0.

Show that e1J{ has an isolated essential singularity at z


and calculate the residue f zn- l e1/z at z = 0.
Example 9.

e l fz

1
1
1
1 +-+--+ ...+--+ ..
l!z 2!z2
n!z"

The principal part is non-terminating so that the isolated singularity


of e11z at z = 0 is essential.

86

[Ch. 5

INTEGRATION

Now z-1 e11' has Laurent expansion

z-2 z-3
1 1
z-1 el/z =z-1 +--+--+ ...+- -+ ...
1!
2!
n! z
so that the residue at z=0 is 1/n!.
Example 10. Expand 1/(1-z2) as a Laurent series about the points
z-Oandz=l.

Using the results about geometric series obtained in Exercise 3,


Question 14 we may write
1

1-z2

=l +z2+z4+

--

where lzl < 1.


The expansion h. no principal part and is valid at all points inside
the circle lzl =1 : this corresponds to a Taylor expansion inside the
circle lzl = 1 (see the shaded region of Fig. 5.10). The function 1/(1-z2)

_,

+I

FIG. 5.10

which we have expanded is a regular function in this region-indeed its


only singularities are on the circle lzl =1 at z= 1. We cannot really
be surprised, then, that the singularities at z = 1 place a restriction on
our expansion.
(2) Proceeding in much the same way as in (it we obtain

1
1 1
1 1
+
=
1-z2 2 1-z 2 l +z

where lz-11

<

2.

(z-1) (z-1)2
_! _l_ !
+
+ l2 z-1 4
2
22

Sec. 5.4]

CONNECTION WITH DIFFERENTIATION

87

Our principal part is -!/(z-1) which corresponds to a simple pole of


residue -! at z
1. The remainder of the Laurent series corresponds to
the Taylor expansion of !(/ 1+ z) about z
1 at and near which it is
regular: it is in fact valid for lz - 1I < 2. This relates to the fact that
1/(l-z2) is regular inside the circle lz-11
2 except for the simple pole
at z
1: the validity of the expansion is restricted by the presence on
the circle lz-11 2 of the singularity at z
-1(see Fig. 5.10.)
=

Example ll. Show that the residue at z


isolated singularity there is given by

z0 of a functionf(z) having an

1 1f(z)dz

2ni

_ c

where C is any simple closed curve which surrounds z

z0.

The Laurent expansion for such a function is


f(z)

n= +oo
=

n= -oo

an(z - z0)"

where
an
If n

S.4

f(z)
1
_ 1
dz.
2ni (z-zo)"+l
c

-1we obtain the required result.

Connection with Differentiation

Suppose that F(z) Sc f(Od( where C is any simple arc which connects
the points (
z0 and ( z and lies wholly within a region A of analyti
city off((.) Let us try to find the derivative of F (z) at z
z1:
=

s. J(() d(

(5.4.1)

Now we know that/(() is continuous in A because it is regular there:


this implies that, for any given positive e, lf(0-f(z1)1 < e whenever (is
chosen close enough to z1 Thus, provided l(-z11 is sufficiently small, we

88

[Ch.

INTEGRATION

may write

f(z i)

l I
=

J;, f(()d' f(z


Z-Z1

i>/

J;, {/(0-f(z1)} d'


Z-Z1

where we have used the results of Examples

and _6. We conclude,

therefore, that

d
F(z)
dz
for every point

(5.4.2)

f(z)

lying in the region A.

This result can be reformulated as follows:

(5.4.3)
where

5.5

F'(O

f(,).

The Theorem of Residues

Example

11

is a particular instance of the

theorem of residues

which

we shall now enunciate and prove.


THEOREM: Iff(z) is regular inside and on a simple closed contour C except
for a finite number ofisolated singularities at the points z,, r
1, 2, . . . , n,
then
=

f(z)dz

2ni

t R,

, i

(5.5.1)

where R, is the residue off(z) at the point z,.


The proof consists in showing the equivalence of the contour C to the
set of contours C, as depicted in Fig.
obvious from the construction of Fig.
the contour of Fig.
have

5.12

5.11. This equivalence is fairly


5.12. The integral of f(z) round

is clearly zero by Cauchy's theorem and so we

'(kJ--' r=lk
f i)

f(z)dz

0.

Sec.

5.5)

89

THE THEOREM OF RESIDUES

8 82

Cz

Ci

"'

()
FIG. 5.11

Example 12.

If at z

FIG. 5.12

z0 f(z) has a simple pole of residue R show that

R =Lim (z-z0)f(z).
z-+zo

Now since f(z) has a simple pole at z

z0 it has Laurent expansion

of the form
R

f(z) = -+a0 +a1(z-z0)+

Z-Zo

Thus we may write

(z-z0)f(z)

R +a0(z-z0)+a1(z-z0)2 +

= R + (z-z0)F(z)
where F(z) is regular at z

z0 since it has a Taylor expansion about this

point.
Hence

Lim(z-z0)f(z)

R.

z-zo

Example 13.

f(z0) =g(z0)

Show

that

Limf(z)/g(z)

Limf'(z)/g'(z)

z-+zo

z-+zo

when

0 provided this limit actually exists.

The implication is that f(z) and g(z) are regular functions of z at the
point z

z0. Hence both possess Taylor expansions in the vicinity of

z = z0. Let us write


f(z)

f(z0)+f'(z0)(z-z0)+

and

g(z) =g(z0)+g'(z0)(z- z0)+


7

90

[Ch. 5

INTEGRATION

But bothf(z0) and g(z0) are zero so that


f(z)
g(z)

f'(z0)+(z-z0)F(z)
g'(z0)+(z-z0)G(z)

where both F(z) and G(z) are regular at z= z0. Letting z

-+

z0 we obtain

the required result which is usually called L 'Hopital's rule.


Example 14. Use the method of Example 12 to find the residue of
1/(l-z2 ) at z= 1.

From Example 11 we know that the pole at z = 1 is simple and has


residue -!. Let us, however, tackle the prblem afresh from first prin
ciples. We can write our function in the form
1

1
1
=
.-2
l-z
1-z l+z

--

--

1
z-1

F(z)

where F(z) is regular at and near z= 1 and so can be expanded in a


Taylor series

Hence, near z= 1, 1/(1-z2) behaves like a0/(z-1) and so possesses a


simple pole at this point.
The residue may thus be found using the formula of Example 12:
R= Lim
z-1

z-1
2

--

1-z

-1
=Lim-
z-1 1 +z

= -t.
Notice that we could have evaluated the above limit using L 'Hopital's
rule:
R=Lim
z-1

z-1

1 -z2

1
= L.im-
z-1 -2z

Sec. 5.5]

91

THE THEOREM OF RESIDUES

If f(z) is regular at the point z


z0 show that the residues
of f(z)/(z - z0) and f(z)/(z-z0)2 are respectively f(z0) and f'(z0).

Example 15.

Now because

where

an

f(z) is regular it has a Taylor expansion

f(z) = ao+a1(z-z0)+
n
(l/ n !)f< l(z0).

+an(z-z0t+

Thus we obtain immediately that

f(z)

ao
= --+a1 +
Z-Zo
Z-Zo

so that the residue of

powers of (z-z0)

f(z)/(z-z0) at z
z0 is a1
z

residue off(z)/(z-z0)2 at

z0 is a0
f' (z0).

f(z0). Similarly the

Show that if a r.egular function f(z) has a zero of order n at


z0 then the function {J(z)}-1 has a pole of order n at z
z0

Example 16.

When we say that f(z) has a zero of order n at z


f(z) can be expressed in the form

f(z)

z0 we mean that

(z- z0tF(z)

where

F(z) is regular and non-zero at and near z

where

G(z) is regular at and near z = z0. The required result is now


G(z) has a Taylor expansion about z
z0.

z0. Thus we find that

self evident since

Show that sin z has a simple zero at the point z


Nn
where N is an integer and calculate the residue of cosec z at that point.

Example 17.

Let us write z- Nn
the point

w: in other words

manoeuvre.
sin

w + Nn)
(-lt sin w
sin(

w
(-lt(w\ 3!3 +
( - 1t w

(i

Hence, by the definition of a zero, sin


i.e. at

let us change the origin to

Nn under consideration; this is often a worthwhile

- w2
3!

)
)

z has a simple zero at

Nn. Thus the residue of cosec z at z

0,

Nn is given by the

92

[Ch . 5

INTEGRATION

formula of Example

12
R

= Lim(z-Nn)cosecz
z-+N1t
=Lim

z-Nn

z-+N"

--
Sin Z

1
-
z-+N" COS Z

.
Lim

( -l f.

Alternatively we could calculate the residue directly from the Laurent


expansion of cosec z about the point z = Nn:
cosecz =

1
-
.

smz

=(-l)NW-l 1=

Example 18.

- l)N ( W2
1+ + ..
\ 3!
(l)

Find the residue

!+

of cosecz/z at z

-1

0.

Now 1z
/ has a simple pole at z = 0. since z has a simple zero there:
also cosec z has a simple ,pole at this point by Example
cosec zz
/ has a double pole at the origin.

Thus

cosecz
z

17.

(1-+ )
3!
1
)
=--' 1+-+ ...
zsinz

z-2

-1

z2

3!

z2

1 1
2+- + powers
3,.

ofz.

Hence we conclude that the required residue is zero because there is no


term in 1z
/ .

Notice that cosec z/z will have simple poles at z = Nn for all non-zero

integral values of N ( see Example

17)

and themethod of Example

could be used to evaluate the corresponding residues.

12

Sec. 5.6]
S.6

TRIGONOMETRIC INTEGRALS

93

Trigonometric Integrals

Suppose that we have to work out some trigonometric integral such


as Jf(cos 0, sin 0) dO or Jf(cos20, sin20) dO: quite often these integrals can
be evaluated by means of the substitutions t = tan to or t = tan 0 but
sometimes this evaluation may prove difficult or even impossible with
the methods at our disposal. The theorem of residues provides us with
a simple but powerful method of dealing with such problems when the
limits of integration are prescribed. In theoretical physics the limits of
integration are usually prescribed by the physical boundary conditions
of the problem; the method, then, Ilas many practical applications.
What we do is to make the formal substitution z = ei6 which has the
effect 9f replacing cos 0 by t(z+1/z), sin 0 by -!i(z-1/z), and dO by
dz/iz. If the angle 0 varies over a range of 2n then the point z in the
complex z-plane describes a circle of unit rads. Then we have

f"+2"
11

where

C is

/(cos 0, sin O)dO =

the circle lzl

Example 19.

Ic 1f{1 {
+) , - 12i f.-1)} dz
2
J

l.

Evaluate by contour integration

We may, write

f6" (5 + 2 cos 0)- 1 dO.

where z = e;6 and C is the circle lzl = l.


Now the integrand in (1) has two simple poles at the points
z = -!(- 5 ..}21). Only one of these poles actually lies inside the contour
C: z = -!(- 5+..}21). Now the residue at this point is easily found by
using the method of Examples 12 and 13.
R

=Lim

{z+t(5-..}21)}
z2+5z+1

.
= L1m

1
2z+5

94

[Ch.

INTEGRATION

Thus by the theorem of residues

2n
1
(5+2cose)-1de= -i.2ni.
,J21

2n
,J21

(2)

Now it is quite easy to adapt the method of Example

19

to obtain

the values of trigonometric integrals whose range of integration is not

2n.

In this context the following result is often very useful.

THEOREM:

Iff(e)=f(a.-e) then J0f(e)de = H2f(e)de.

We can readily c nvince ourselves of the truth of this theorem by


interpreting the integrals as areas under a curve and appealing to

FIG. 5.13

Fig.

5.13.

Evidently the curve is symmetric about the line

e= !a.

and so

the two shaded regions are equal in area.


Example 20.

Evaluate

So (5+2co:> e)-1de.

Using the above theorem and the result


immediately that

Example 21.

f:

Evaluate

(5+2 cos e)-1de=

rnn (5+2cos2e)-1de.

f:

Example

19

it follows

,J:.

J012 (5+2cos2e)-1de.

If we proceed as in Examples
of

(2) of

19 and 20 we are led to the consideration


z = e;9 and, as above, we obtain:

Let us put

n
(5+2cos2e)-1de= -2i

z(z4+12z 2+1)-1dz .

(1)

Sec. 5.6]

TRIGONOMETRIC INTEGRALS

95

The integrand in (1) has four simple poles at the points z


i.j(6.J35 )
of which only i.J(6-.J35 ) actually lie inside the unit circle C. The
required residues are obtained by the method of Example 12: at both
points we obtain l/4.J35. Thus our final result is
=

J:

'2

J:"

(5+2cos2e)-1de=!

(5+2cos2e)-1 de

=!. (-2i). 2ni.

2
4.J35
(2)

Let us pause, however, to compare equation (1) in this example with


equation (1) in Example 19. Our observation is that it is rather more
complicated in the present instance in that we have had to deal with a
quartic denominator rather than a quadratic one.
A simple change of variable could have avoided this added difficulty
as we shall now demonstrate.

n/2

(5+2cos2e)-1de=!

where</>

f"
21t}
f

(5+(1+cosW))-1de
0

6+cos </J)-1d<f>

(3)

W. If now we set z = ei<f> we obtain

f
0

"f

(5+2cos2e)-1de= _!_
2

(z2+12z+1)-1dz.

(4)

The integrand in (4) has poles at z


-6.J35 of which only that at
z = -6+.J35 actually lies inside the unit circle C. The appropriate
residue is 1/2.J35 so that our final result is
=

n/2

. 0

-i
1
(5+2cos2e)-1de= 2. 2ni.
2.J35
(5)

EXERCISE

[Ch.

INTEGRATION

96

l. Find the poles and corresponding residues of the functions (i) tan

z/z3.
(Answers:

(ii) cot

(i) Simple poles of r:esidue -1 at

an integer; (ii) Simple poles of residue

1/N3n3

at

(N +!)n where N is
Nn where N is a
=

z = 0)
{z2(z -1)(2z-1)}-1

non-zero integer : also a quadruple pole with zero residue at

2.

Use the results of Example

15

to express

in

partial fractions.

(Answer: z-2 + 3z-1 +(z-1)-1-8(2z-1)-1)


3. If n is an integer and C the circle lz-z0I

wise, show that

R, described anticlock-

fc (z-z0)"-1 dz vanishes for all non-zero values of n.


n

What is the value of the integral when

(Answer: 2ni)

O?

z and if C is any closed curve


z2 show that
(z -2U:
f(z)
f(z2)+ 12m Jc (z-z )(z-z ) dz.
2
1

4. Iff(z) is regular for all finite values of


enclosing both the points

f(z1)

z1

and

Note: In mathematical physics we sometimes obtain results in the form


f(z1)

,( f(z) dz
2m !z-z1

and wish to let parts of the curve C to recede to infinity. In such circumstances
it is often necessary to improve the convergence of the integral by introducing a
so-called subtraction termf(z2) as in Question 4. When the functionf(z) diverges
as lz l -+ oo we can always construct a convergent integral by introducing a finite
number of subtraction terms provided the asymptotic behaviour off(z) is no worse
than a polynomial of finite degree in z; the subtraction terms may represent
parameters in a physical theory which require to be determined experimentally;
examples of this technique are to be found in the quantum mechanical theory of
dispersion relations which relates to the scattering of elementary particles.

5. Liouville's theorem

states that a function wnich is regular and

bounded for all finite values of

z is a constant.

Prove this result by taking C in Question 4 to be the circle


letting R -+ oo.
6. By expanding

exp{!z0(z-z-1)}

as a Laurent series about

show that

exp{!z0(z-z-1)}

lzl

+.<X>
=

L Jn(z0)z"

-oo

R nd

Ex. 5]

97

INTEGRATION

where

Jh0)

2n

2"

cos(ne-z0sin e)de.

7. By considering J zi-1 e11z dz round lzl=1 show that, for n a


positive integer,

{"

e008cos(ne-sin e) de

(Hint: See Example 9)


8. By means of the substitution t

2n/n !

'

tan }e show that

n
de=--,, 1-2a sin e+ a2
1-a2'
,2
-----

la l

<

1.

9. Obtain the result of Question 8 by the method of contour integra


tion.
10. By means of the substitution t=tan e show that

11. Show that

,,12
1+2sin2e

1+2cos2e

"12 1+2sin2e

1+2cos2e

de

de=

n(4,J3- 3)/6.

2"2-cos</J

2+cos</J

d</J

and hence verify the result of Question 10 by the method of contour


integration.
12. Use the theorem of residues to show that
lt/2

and deduce that

a2+cos2e

---

de=

2a ,j(a2+1)

(You may assume that it is permissible to differentiate with respect to


a under the sign of integration.)

98

INTEGRATION

13: If F(z) is a regular function of z inside and on tlfe circle lzl


Use Cauchy's theorem to show that
F(Zo)

1
- 2ni

lzl

R.

(R2 -z0z)F(z)
dZ
*
(R2 -zz0)(z-z0)
=R

where lzol < R.


Deduce that
F(rei6) =

2"

(R2 - r2)F(Re;.;)
d</J
2n 0 R2 -2Rr cos({}-</J) + r2
1

where r < R. This result is known as Poisson's formula.


14. By expanding 1/(1 +z2) as a Laurent series about z = !i obtain
an expression of 1/(1 +z2) which i:; valid in the annulus!< lz-!ij < l
(Answer: an = (2irf3n+ 1, n :;;:: 0; an (2i)"( -1)"+ 1, n < 0).
=

CHAPTER 6

Improper Integrals
The calculus of residues described in Chapter 5 is a powerful tool in
the evaluation of many of the integrals which occur in real variable
theory. The author feels that this is one of the most impressive and
important aspects of the practical pplication of complex variable
theory in applied mathematics.
6.1

Infinite Integrals

An integral is called

improper

if its range of integration is infinite

or if the integrand is infinite at a point of the integration range. For the


present we shall assume that we are dealing with the ordinary integral

of a real function f (x) of a real variable


of the closed interval R

which is finite at each point

S. Let us write

I(R,S)

J:

f(x) dx.

(6.1.1)

Clearly we may cause the range of integration to become infinite by


letting R or S tend towards infinity. If the limits
Lim I(R,S)

R-+-oo

exist then we say that the

and

Lim/(R,S)
soo

(6.1.2)

irifinite integrals
(6.1.3)

and

exist and are equal respectively to the limiting values.


It is also evident that we obtain an infinite range of integration by
letting both R and S become infinite. If the limit
Lim

(R.S)(-00,00)
99

I(R,S)

(6.1.4)

100

[Ch.

IMPROPER INTEGRALS

exists then we say that the infinite integral

(6.1.5)
exists and is equal to the value of the limit
Such a definition of

J 00f(x)dx

(6.1.4).

is equivalent to insisting that both

and

f(x)dx

exist and that

the choice of a being immaterial,


However, in applying contour integration to calculate values of such
integrals in later sections of this current chapter, we shall find time and
time again that it is most convenient to take the limit (R,S) -+
in a very special fashion: namely R-+ -S followed by S-+

( - oo,oo)

oo.

Now it is quite conceivable that the limit

(6.1.6)

LimJ(-S,S)
S-+oo

may exist while the limit

(6.1.4)

does exist then so does the limit


The value of the limit

(6.1.6),

does not; of course, if the limit

(6.1.6)

(6.1.4)

and the two are equal.

if it exists, is called the Cauchy principal

value of the integral and denoted by

(6.1.7)
Thus we conclude that, if the integral

(6.1.5)

exists then so does the

Cauchy principal value of the integral and the two are equal:

However, the integral


does not.

(6.1.7)

may exist even when the integral

(6.1.5)

Sec.

6.2]

Example 1.

Discuss the improper integral Sf x""dx.

Consider the integral

n x""dx.

x""dx =

sa+l_l

oc+ 1

oc#-1

=loges,

OC=-1.

Now Lim(S""+1-1)/(oc + 1) exists only for

S-+oo
-(oc+ 1)-1:

Jf x"" dx

fo1

INFINITE INTEGRANDS

oc + 1

0 when its value is

<

also Lim logeS does not exist. Thus we conclude that

S-+oo

exists only for

oc < -1.

Example 2. Discuss the existence ofthe improper integrals J00 sinh x dx


and p s00 sinh x dx.

. J:

sinh x dx = cosh S - cosh R.

It is quite evident that the double limit

does not exist so that the integral

Lim

(R,S)-+(-00,00)

J00 sinh x dx

{ cosh S -cosh R}

does not exist. However,

since cosh( - S) = cosh S, Lith { cosh S - cosh R}

R-+-S
S-+oo

does exist and has

value zero. Thus the principal value integral exists while the integral

itself does not.

6.2

Infinite Integrands

As remarked in

(6.1) the infinite integral is not the only kind of integral

which is classified as improper. If the integrand is infinite at some point

(or points) of the range of integration then the integral is also termed

improper.

Without any loss of generality we may consider the integral of a real

function f(x) of a real variable x over the finite range a ::>.;; x ::>.;; c such

that f(x) is finite at every point of the interval (a,c) except the point
x =b, a < b < c. Then the integral J f(x) dx is said to exist provided
the limits

b-1

Lim

1 ...... 0

f(x) dx

and

Lim c
f(x) dx
b+t2

i-+O

(6.2.1)

both exist and the value of the improper integral is defined by the sum
of these two limits.

102

[Ch. 6

IMPROPER INTEGRALS

Now it may happen that the improper integral does not exist but the
limit

(6.2.2)
does. The value of this latter limit is called the
the integral and is denoted by

Cauchy principal value of

f(x)dx.

Clearly if the integral Sf(x)dx exists then the principal value exists
and the two are equal.
If the point b actually coincides with an end point of the region of
integration, say b
c, then the integral Jf(x)dx is said to exist if
Lim J:- 'f(x)dx exists.
=

-o

Discuss the existence of the improper integral Si, xdx.


The integrand is infinite at x
0 when ex < 0 and so the lower end

Example. 3.

of the region of integration must be carefully examined.

xdx=

l-ea+1

ex + 1

logee,

ex=!= -1
ex

-1.

Now Lim(l-e+1)/(ex + l) exists only for ex + l >


t-+O

0 when its value is

(ex+ 1 ) - 1 : also Lim logee does not exist. We conclude, therefore, that
.

E-0

n xdx exists only for ex


Example 4.

>

- 1.

Discuss the existence of the improper integral J: x - 1 dx.


1
J: x-1 dx does not exist.

It is clear from Example 3 that the integral


However the principal value integral does.

Pf

-1

x-1 dx

= f f - ' x-'l dx+ x-1 dx


1J _ l
= Lim { loge j -e j-logej e j}
1

-+0

6.3

0.

The Ev ua on of Infinite Integrals

Suppose that f(z) is a function of a complex variable z such that f(z)


is regular in th,e half-plane fz 0 apart from a finite number of isolated

Sec. 6.3)

103

THE EVALUATION OF INFINITE INTEGRALS

singularities. Let us take a contour

C comprising a portion

-R

of the real z-axis together with a semicircle of radius R as shown in

f,cf(z) dz and letting R

Fig. 6.1. Then, by considering

'.,)

possible to deduce the value of the integral

-:-+OO

it is often

J':: 00 f(x) dx.

FIG, 6.1

Example 5.

Evaluate

J':: (1+x2r1 dx.

00

Now, by elementary methods, we see that the value of this integral is

f00

(l+x2)-1 dx

Lim

(R,S)-+(-co,co)

_00
=

{tan-1S-tan-1R }

n.

However, for purposes of illustratin, let us proceed in the fashion


outlined above, and consider
Fig. 6.1. Now
pole at

(1+z2)-1

dz

where

is regular inside and on

z
(1+z2)-1 at

(the pole at

The residue of

fc (1+z2)

- lies outside

is the contour of

except for a simple

C).

is given in the usual fashion as

(z-i
Lim
z-+i 1+z

.\

2i

Thus by the theorem of residues we obtain

(1+z2)-1

dz

2ni. i
n.

However, we may split our contour integral up into two parts as follows:

[Ch. 6

IMPROPER INTEGRALS

104

where

is the integral of

(1+ z2)-1

I II

Clearly, then, if

R-+

oo,

f"

around the semicircular part of C.

Rieie
d()
i
R e2ie+ 1

"

R
d()
R2-1
O
rr:R
.
R2-1

III -+ 0 and we obtain the result

foo (l+x2)-1 dx

n:.

Provided that the given integral is known to exist, the Cauchy principal
value integral evaluated above gives us the required result. As we shall
see in the followin examples it is a simple matter to exploit the symmetry
of the integrand to show that

J 00(1+x2)-1 dx and JQ'(l+x2)- 1 dx both

have value rr:/2. Then it is evident that the given integral does exist and
has value rr: (see Section 6. 1).

An important feature of the method of contour integration as ex

emplified above is the vanishing of the integral around the infinite


semicircle in the upper half-plane. It is clearly a trivial matter to assert
this property if the modulus of the integrand behaves like
n >

for

lzl

lzj-n

where

large. The method can, however, still be used to evaluate

integrals whose integrands do not possess this property.

Evaluate SO' cos x/(1+x2)3 dx.


Now, since leiI
e'fY, it is evident that, although leiI 1 for y 0,
le-iI i_s unbounded in the upper half-plane. We chose therefore to treat
f/iz/(1+z2)3 dz rather than f cos z/(1+z2)3 dz.

Example 6.

The integrand

ei'/(1+z2)3 is
z

except for a triple pole at

regular inside and on the contour C


i (the pole at z

i lies outside the

contour). Because the pole is not simple we must employ an expansion

Sec.

6.3]

THE EVALUATION OF INFINITE INTEGRALS

method of determining the residue. Let us set

e iz

8;"'(1+w/2W 3
------(1+z2)3
e(2iw)3
=

... \

-=-! Ji + iw ---- w2 ..
8iew3

2!

105

z ---- i:

) \J

(1 ---- 3w ---- 3w2..


2i
2!

1
5
7
---w-3 ---w -2+-w -1 ...
8ie
16ie
16e

Thus, by the theorem of residues, we have

,f eiz
Jc (z2+1)3dz

7
2ni.
16ie
7n
8e

Proceeding as in Example 5 we obtain

. f 2
(z

z
l)3d

I: 9:2:iln
c

x .
dx+I

(1)

where

----

nR

(R2 --1)3.

noticing that cos x/(l+x2)3 is an even function and that


x/(1+x2)3 is an odd function we may rewrite the equation (1) as

Now,
sin

follows

,f

ei

z
Jc(z2+1)3d

cos x
2 o
x+J.
(x2+1)3d

If now we proceed to the limit as


required result:

oo

R-+

COS X

oo,

I-+ 0, and we obtain the

x
'!!!_
(x2+1)3d ---- 16e
__

IMPROPER INTEGRALS

106

Notice that if we take the limit R


pressed in the form (1) we obtain
p

fro

--+ oo

cos x
dx+iP
2+l)3
-ro(x

[Ch. 6

with our contour integral ex

ro

sin x
-oo(x2+1)3dx

7n
8e"

Taking real

and imaginary parts of this equation we verify that


Pfrosin x/(x2+1)3 dx
0 and find that Pf ro cos x/(x2+1)3
7n/8e.
If further we know that the given integral actually exists we can deduce
=

its value from the latter result.


6.4

Jordan's Inequality

In Example 6 we avoided difficulties arising from the behaviour of

icos zl

on the infinite semicircle by considering a modified integrand

which vanished like lzl-", n > 1, as lzl--+

oo. There are problems, however,

which can be treated by the method of the last section even in the case
of an integrand which vanishes like

lzl-1

as

lzl --+ oo:

this is done either

by choosing a different integration contour or by means of Jordan's


inequality. This latter result is simply
.
2()
() (),
-sm

(6.4.1)

7t

We shall not offer a formal proof of this result but merely remark
that it is obvious from geometrical considerations. In Fig. 6.2 it is clear

0
FIG. 6.2

FIG. 6.3

from the convexity of the graph that y


y

sin () always lies above the line

W/n provided 0 () tn. In Fig. 6.3 AB is an arc of a circle centre

0. Since AB ;;:,:: AC it is evident that ()

AB/r always exceeds sin()

AC/r.

'
JORDAN S INEQUALITY

Sec. 6.4]
Example 7.

Evaluate

107

f':' 00 x sin x/(x2+1) dx.

Consider the integral

fc zeiz(z2+1)-1 dz

around the contour C of

Fig. 6.1. The integrand is regular inside and on C except for a simple
pole at z

i: the residue at tl].is pole is given as follows:


. (z-i)zeiz
L1m
z2+1
z;

"
z-i
L1m
e z z2+1
i
i

--

In the usual manner we obtain

where

x(cos x+i sin x)


X2 + 1

-R

I"

dx+l

ni
=

(1)

Rei9 . exp(iRei9) . iRei9


d(J.
R2 2111+1

Now, proceeding as in Examples 5 and 6, we find that


R2n

Ill R2-1
which inequality is insufficient for us to conclude that

Ill -+ 0 as R -+

We require to make a more refined argument as follows:

111

I"

R2
R2 -1

l exp{ iR(cos (}+i sin e)}I d(J

Now, by appealing to Jordan's inequality, we see that

Ill

R [
R -1

nR
=

/2

R2-1

e-2R9/" d(J

(1-e R)

-+ 0 as R

-+ oo.

oo.

108

Now,

sin

[Ch. 6

IMPROPER INTEGRALS

noticing

x/(x2+1)

follows:

that

cos

x/(x2 + 1)

is

an odd function and

is an even function, we may rewrite equation

JR
-RX

x sin x
dx
2
1
+

_
-

JR

2z

ni
=--

that

(1)

as

sin x
dx
2+ 1
(2)

I.

Finally we proceed to the limit R -+ oo in equation (2) and obtain the

f00

required result:

xsinx
dx
x2+1
-oo

Thus .far our examples have been of integrands with singularities


occurring only at complex points. If a pole does Ii<; on the real axis then
a different contour of integration must be used.
Example

8.

Evaluate f0 x -1 sin x dx.

Just as in Example 6 we ar led to the consideration of the integral

fc z-1eiz dz

rather than

fc z-1 sin z dz

where C is a suitable closed

contour. The curve C must differ from that drawn in Fig,


cannot allow C to pass through the point

6.1

in that we

0. We indent our contour

to avoid this point as shown in Fig. 6.4. The indentation takes the form

FIG. 6.4

of a semicircular arc of radius

r.

Now no poles of the integrand lie

inside our contour of integration: thus since

eiz/z

is regular both inside

and on our contour C we may conclude that

J.. eiz
Jc z dz

(f-r JR)
R
+

.
eix
dx +I+J
x
(1)

Sec.

109

'
JORDAN S INEQUALITY

6.4]

where I and

are respectively the integrals of

semicircles of radii

R and r in

eiz/z

taken round the

the senses indicated in Fig. 6.4.

As in Example 7 we can use Jordan's inequality to show that I


vanishes in the limit

R-+

oo :
7t

IJI R ( l -

..

e -R)

(2)

Now let us concentrate on the evaluation of J for this is essentially


what the problem has been reduced to. We have

J=

eiz

-dz
z
C'

reiB, 0 () n.
i -dz+
1
i eiz_ dz
J=

where C' is the semi-circle


parts as follows:

Let us split

into two

=
.

--

1
.
8ire'8d()+J',
re'

say

"

= -ni+J'.

(eiz -1)/z-+ 0 lzl -+ 0


l(eiz - )/ l M
M

Now

as

where

Chapter 5 we have

Clearly then
inequality as

so thl\t, provided

IJ'I

is small enough,

nrM

(4)

is the only conclusion consistent with the above

Let us now proceed to the limit as


Using the results

is a positive constant. Thus by Example 6 of

J' -+ 0
r-+ 0.

(3)

r-+ 0

and

R-+

eix dx+ R eix dx:\


Rx I x }
(-J ei dx+ f :x d x)
eixe-ix dx
=
)
m(f
00
xd
.
=I
2
x
J

(2), (3),

and

(4)

we obtain

( r- r
ni =Lim \J
_

'

=Lim

Li

sin

--

oo in equation

(1).

110

Finally we conclude that

sin x

6.5

[Ch. 6

IMPROPER INTEGRALS

dx

Expansion in Series

An extremely important application of the theorem of residues


concerns the expansion in series of partial fractions of analytic functions.
Let us consider an example of this application.

Show that the function l cot zl is bounded, for N an arbi


(N+!)tr. By
trarily large integer, on the lines 9lz
(N+!)tr, ..h
considering the integral
cot z/z(z-z0) dz, z0 =F Ntr, where C is the
c
boundary of the square defined by the above straifht lines, show that
Example 9.

On the line

Blz

00

cot z

-+2z

L 2 2 2,
n=lZ -n tr

=F

ntr.

(N+!)tr we have

i cot zl

i(eiz+e-iz)
(eiz_e-iz)

:::;;; 1.

On the line .h

(N+!)tr we obtain

--+

eixe - (N +t)x+e-ixe(N+t)x
eixe-(N +t)x_e-ixe(N +t)x

as

--+ oo.

Thus, given M any fixed number greater than unity, J cot zl :::;;; M provided
N is chosen sufficiently large. Similar arguments lead to identical
conclusions for the .lines

9lz

-(N+!)tr and

Jz

-(N+!)tr re

spectively and we conclude that, on C, provided N is a large enough


integer, J cot z l :::;;; M.

Ex. 6]

111

IMPROPER INTEGRALS

Now let us consider the integrand cot z/z(z -z0) which is regular

z = 0, a simple pole at
z= mr, n 1, 2, .... The
-1/za, cot z0/z0, and 1/mr(nn-z0),

inside and on C except for a double pole at

z= z0,

and simple poles at the points

usual methods yield the residues,

1, 2, ... respectively.

Thus, by the theorem of residues, we obtain

_1_ J. cot z d
z=
2ni k z(z - z0)

_.!_ +cot z0+


z

z0

I
- nn(nn-z0)

N
(n,00)

But, on C, we know from the above argument that jcot zl

I,z(z:-z0) I
cotz

M
(N+!)n{(N+!)n-jz01}

so that

provided N is sufficie qtlY, large and we have used the fact that both
j{(N+!)2n2+x2} and' j{(N+!}2n2 +y2} exceed (N+!)n. Clearly then

by Example 6 of Chapter 5

,(

cot z

dz

I fcz(z-z0) I

M4(2N+ l)n
(N+!)n{(N+!)n-jz01}

-+ 0

as

N-+

oo.

Hence we obtain our final result


cotz0

1
=z0

oo

Zo
L
= nn(nn - z0)
0
00

(n,0 )
_

Zo
f
Zo
= z0+ =l nn(z0-n) nn(z0+nn)
n
00
1
2z
=-+I
2 27t2
Zo n=l Zo-n

Note: Arguments involving uniformity of convergence actually permit the above

result to be integrated term by term to yield ail infinite product for sin z/z. Such a
device is frequently used in both pure and applied mathematics.

EXERCISE 6
1. Show that, for 0

(!)3'-ltx-1
holds provided

<

1.

1,

!, the inequality

(l-t)'-ltx-1 tx-1

and write down the corresponding result for

112

IMPROPER INTEGRALS

[Ch. 6

Deduce that the improper integral

I:

tx-1 (1-t)Y-1 dt

can exist only for x > 0 and y > 0.


(Hint: Split the range of integration
u

Ji+Ji and use the variable change

1-t in the

second integral)
2. Use the theorem of residues to show that

Joo
o

X2
dx
x4+1

Jczeiz(z2+z+l)-1 dz

3. Show that

7t
2..; 2

vanishes when C is the infinite

semicircle in the upper half-plane. Hence deduce that

. [-oo x2x+x+x l dx
cos

(1)
(ii)

[oo x::i::1 dx

( )
../33/2 ( D
2n
2n 1
cos 3-2
3
,;312
e
J
2n

sin

(You may assume that both the given integrals exist.)

4. Show that

jeizl

is bounded in the upper half-plane whereas

oo

is not. Show by contour integration, that

COS X
- 2-- 2 dx
x +a

7t
2a

e-a,

je-izl

a>O

and, by differentiating this result with respect to a, or otherwise, show


that

oo

COS X
dx
(x2+1)2

Joo (x2+1)X dx.


COS

(U.C.N.W. 196 3)

5. Use the method of Example 9 to show that

(1).

cosec2 z

(ii) tan

'-'

_00

2z

1
'
(z-nn)2

I (n+1/n2 -z2,
o

6. By integrating cot

z/z2 round
1
n2
00

=I= nn,

,z (n+!)n.
the contour of Example 9 show that
=

7t2
6.

Ex. 6]

IMPROPER INTEGRALS

113

7. By integrating
(z2 +a1)(e2"z - 1 )

round a rectangle, show that, when a i s not a n integer,

n cos a(t-n)
1
2a cos nt
.
- -+L.
2
2
sm an
a
) a -n
a
_

Deduce that

n cot an

n cosec an
.

By considering the case a

cos nt
L... --2n
i

1
00
2a
-+ 2--2;
a
1 a -n

1
=

(0 t 2n).

00 ( -1)"2a
2 .
2
a -n

L1

0, prove that

1 2
1 2
6n -tn +4t

(0 t 2n).
(SHEFFIELD 1952)

8. By integrating the function

sinh z
around the rectangle with vertices
necessary, prove that

foo .
0

sm x d
x

-:-smh x

R,

R +in,

indented where

n
n
- tanh - .
2
2
(HULL 1958)

9. By integrating the function

l+x"

around the perimeter of the sector of the circle z


2
r and z
re "i/n show that

lines z

rei6 bounded by the

Joo
0

1
_
x_m_-_ d
x
1 +x"

n
__

_
_
=

. nm
nsm
n

where both m and n are positive integers.

m < n

CHAPTER 7

Many- Valued Functions


The theory of many-valued functions has no good analogue in real
variable theory: it is a distinctive feature of complex analysis. Even
although its concepts lie outside the span of ordinary real analysis (and
for this reason is sometimes regarded as a 'difficult' topic for engineers
and scientists) the theory of many-valued functions has consequences of
great interest to practical men. In this chapter the basic ideas of this
branch of complex variable theory will be introduced 'via a detailed
study of the many-valued logarithmic function.
7.1

Logarithmic Branch Points

In our efforts to establish contact, in this mathematical treatise, with


the practical problems of electricity (see Sections 3.5 and 3.6) we intro
duced the logarithmic function Logz as the inverse function of exp w :
further we had noticed that Logz was not one-valued but many-valued,
and, indeed, we had defined a single-valued logarithm logz by means of
the equations (3.6.6) and (3.6.7) which we reproduce below
logz

Logz

log z
l l + i argz

(3.6.6)

logz+2nNi.

(3.6.7)

Now, because argz has been defined as the principal value of tan
it lies in the range - n

<

y/x

argz n, so that the function logz is not

continuous at the points of the negative real axis. This is clearly the case
because
Lim

zxoie

where

argz

n,

Xo <

is some positive quantity which we assume vanishes as the limit

is taken. Thus logz cannot be analytic on !Hz 0, Jz

0 because, if it

were, it would also be continuous there. We rewrite the result of Example


114

LOGARITHMIC BRANCH POINTS .

115

12 of Chapter 3 as
d
1
-logz =dz
z

(7.1.1)

provided z is not real and negative.


Let us consider the evaluation of the definite integral fi c- 1 d( which
we know must exist provided the path C from 1 to z in the (-plane does
not pass through the origin. Now, since d/dz logz does not exist for
argz =n, we cannot, in general, evoke the result of Section 5.4 to assert
that

fz 'f1

d( =logz.

(7. 1.2)

This result is valid only if C does not cross the negative real axis. Suppose,
then, that C does cross arg( =n once as shown in Fig. 7.1. By using

FIG. 7.1

Cauchy's theorem we can deform C continuously into C' without altering


the value of the integral (see Fig. 7.1): C' comprises that portion of the
real ( axis which connects ( =1 to ( =lzl together with the appropriate
ar\: of the circle !Cl =lzl.
The integral may thus be split into two parts as follows:

r !d, = r !d,
Jc (
JC' (

flzl ( fz (
lzl
l

-d(+

-d(

1.

say.

(7.1.3)

116

[Ch.

MANY-VALUED FUNCTIONS

I 1 can be evaluated immediately by using formula


I1

(7.1.2) as

log z
l l.

(7.1.4)

The evaluation of I2 is also quite straightforward: let us substitute


C

Rei9 and write z

z
l l e;11 :

12

Now () > n (see Fig.


argz

f:

dE>

i().

(7.1.5)

7.1) so that () =I= arg z: we have, in fact,

{)-2n.

Finally we obtain the value of the integral:

{f

d(

If our path from C

log z
l l+i argz+2ni
logz+2ni.

1 to C

(7.1.6)

z had crossed the negative real axis n

times in a clockwise sense and m times in an anticlockwise one, we


should have obtained the result

{f

d(

logz+2ni(m-n).

(7.1.7)

It is a simple matter to extend this result to any definite integral of

Zz

z-1:

z1

dz

logz2-logz1+2ni(m-n).

(7.1.8)

The many-valued function


Logz

logz+2nNi

is said to have an infinite number of branches corresponding to N

0,

1, 2, .. .. We have singled out for special mention the particular


branch which corresponds to N

0, namely logz. This branch, the

principal branch, of Logz has been seen to be regular in the z-plane


apart from argz

n. Similarly each of the other branches is also a

regular function of z for all z which are not real and negative. We call
the line PAz 0, fz

0, a branch cut; it is so named because the line,

as it were, cuts the plane in such a ,way as to divide our function Logz
into one-valued branches.

Sec.

7.2]

117

THE PRINCIPLE OF THE ARGUMENT

We say that each branch of Log

z is regular in the cut plane.

Ostensibly the cut appears to play a very significant role but, in an


important sense, this is not really so. The cut is really fairly arbitrary:
indeed,

any

line joining the origin to infinity would suffice to split Log z

up into one-valued branches.


How then have we arrived so unambiguously at the cut arg

n?

Well, it stems simply from our insistance that we choose our one
valued argument to be the principal value of tan - i

y/z

i.e.

-n

< (}

n.

If we had decided to define our single-valued argument in, say, the


range 0 < (}
real axis.

2n we would have arrived at a plane cut along the positive

At the very outset when we wrote

rei8

many-valued argument Arg z of which arg

we could have defined a


was but the principal

branch:
Arg z
where

arg

z + 2Nn

N is any integer.

The fact that we chose the value of arg

(7.1.9)

to lie in the range

is quite arbitrary. We could have effected the division of Arg


one-valued branches by selecting any range of

2n

principal value.
In this new notation we can rewrite equation
Log z

log

and calling this the

(3.3.7)

in the form

lzl + i Arg z

which has precisely the same form as equation

( - n,n)
z into

(7.1.10)

(3.6.6).

It is now fairly evident that the points which a branch cut joins rather
than the cut itself are the fundamental entities: we call such points

branch points.

In the present discussion the origin and infinity are the

relevant branch points. Any cut drawn from the origin to infinity divides
Arg z up into single-valued branches.

7.2

The Principle of the Argument

Equation

(7.1.8) is

very often rewritten in terms of Log

z2

Zt

1
-; dz =

[Log z]c

where [Log zJc denotes the change in Log


In particular if

z1

z2

as

as follows:

(7.2.1)
z

describes the curve C.

(i.e. the curve C begins and ends at the same

118

[Ch . 7

MANY-VALUED FUNCTIONS

point) then we write


(7.2.2)

where [Argz]c= 21t(m-n), the change in Argz as C goes round the


origin

(m-n) times. When we say that C goes round the origin (m-n)
m anticlockwise and n clockwise circuits have been

times we mean that


made.

We obtain a generalization of equation (7.2.1) by setting w

= f(z)

as

follows:

fz2
Zt

f'(z)
--

f(z)

dz=

1
-dw

J(C) (J)

=[Logw]J<q
= [Logf(z)]c

(7.2.3)

where f(C) is our integration contour in the ro-plane. In particular if


z1 =z2 we obtain the equation corresponding to equation (7.2.2):

J. f'(z) dz= [i Argf(z)]c.


Jc f(z)

(7.2.4)

A function which has, in a certain region, a finite number of zeros and


is regular apart from a finite number of poles is called

meromorphic.

If we have such a function in a region A whose boundary is the closed


curve C then, if

p be the sum of the orders of zeros inside A and

q the

sum of the orders of poles, we have


1

p-q=-

2ni

f.

f'(z)
--dz
c f(z)

1
[Argf(z)]c
21t

(7.2.5)

provided no pole nor zero actually lies on C.


Suppose that the zeros and poles off(z) occur at the points z=zi,

j= 1, 2, ... ; then, at and near z= zi we may writef(z) in the form


(7.2.6)
where g(z) is regular and non-zero at z= . z i. If Pi > 0 we have a zero
of order Pi and if Pi < 0 a pole of order - Pi We can calculate the

Sec.

7.2]

residue of

119

THE PRINCIPLE OF THE ARGUMENT

f'(z)/ f(z)

at

z=zj

pj:

by inspection: it is simply

f'(z) _EL__ g'(z)


=
+
.
f(z)
z-zj g(z)

(7.2.7)

Thus, by applying the theorem of residues we obtain

.
J. f'(z)
dz=2
f(z)
m

Jc

=2ni

t pj

I + 4:. \pj
p;>O
p;<o)
.
=2ni(p - q).
This result is usually known as the

principle of the argument.

Example 1. If f(z)=z5+A.z2+1, where A. ;;::i: 0, show that f(z) has one


zero or no zeros in the octant 0 arg z in according as A. < 2 - 115 or
A. > r 115 If A.=r 115, find two zeros of f(z).
(DURHAM 1959)
This problem is typical of

an

important class of problems which

yield solutions very readily on application of the principle of the argu


ment.
As a first step it is best to ascertain whether or not there are zeros

off(z) on the boundary of the region in which we are interested. Certainly


there are no real positive zeros because f(z) ;;::i: 1 when z is real and
positive (A. ;;::i: 0): furthermore when z=rei"/4 the equation f(z)=0
becomes

-r5(2)-1'2(1+i)+A.ir2+1=0

which, on taking real and imaginary parts, cannot be satisfied for real
unless

A.=2-115.

FIG. 7.2

If then we exclude the case

A.=2-115 and
7.2 (R--. oo),

where C is the curve drawn in Fig.

tion will give us the sum of the orders of zeros of

f(z)

clearly has no poles.

[Argf(z)]c/2n,

calculate

the result of our calcula

f(z)

inside C because

120

[Ch. 7

MANY-VALUED FUNCTIONS

As we vary z along OA [Arg f(z)]


point on OA. On the arc AB we have:
[Argf(z)]

Argzs

0 since arg f(z)

0 at every

(i ;; ]
[ ( )]
+

[Argzs ]+ Arg l+

1
+ s
2

where we have used the result of Exercise 7, Question 1. Thus, setting


z = Rei6, we obtain
1
[Argf(z)] =[ArgRse5i9]+ Arg t+ + 5
2
2
=

[ ( )]
[ :)l

5n
+ Arg
4

As R oo we see that [Argf(z)] =5n/.


4 We could have obtained the
same result as follows:
Rs sin 50 + A.R2 sin 20
Argf(z)
tan-l
Rs cos 50 + A.R2 cos 20+1
=

tan-1 tan 50
Now on the line BO we have z
write
Arg f(z) =tan

tan- 1

as

R oo.

rei"14 where 0 r R so that we may

rs sin 5n/4 + A.r2 sin n/2


rs cos 5n/4

+ A.r2

cos n/2 +1

r2(A.- 2-112r3)
1

--

tan-1 g(r),

2 112rs
say.

We notice that the numerator of g(r) vanishes at r


O and at r
21'6A.1'3
0
111
(
r say).
( =r1 say) while the denominator vanishes once at r =2
2
Now r1 r according as A. 2-1/s so we must distinguish the two
2
cases. Let us tabulate values of g(r) and [Argf(z)] as follows:
(1) A.> r11s
=

00

oo> r > r1

rl

g(r)

[ArgJJ

-n
4

r1 > r > r
2

-n
2

r1

r > r> 0
2

00

-n
2

Sec. 7.3]

121

ROUCHE'S TH E O R E M

Thus as r varies from


1/5
(2) }. < 2 -

00

g(r)

oo >

R( oo)
r

>

r 2 r2 r2

[Arg f]

to 0 along B O [Arg

>

> r1

r1

00

TC

TC

f(z)]
r1

>

2TC

[Arg

if

if

> 0

0
0

f(z )]

r
+

Thus as r decreases from R( oo) to 0 along BO [Argf (z)]


Hence we obtain the final result:

- Sn/4.

3n/4.

The second part of this problem concerns the limiting case A.


2-115
in which a zero of ](z) actually lies on the line arg z
TC/4. What is
actually happening is that a zero off(z) crosses arg z
n/4 as the value
of the parameter A. is allowed to vary through r l/5.
We simply set z
rei"/4 in the equation f(z) 0 and take real and
imaginary parts to obtain the solution z
r 215(1 +i) (we have in fact
already done this above). Finally since f(z) is a real function the roots
of f(z)
0 occur in complex conjugate pairs: thus z
r 215(1- i) is
also a solution.
=

7 .3

Rouche's Theorem

If f(z) and g(z) are regular within and on a simple closed


contour C and lg(z)J < lf(z)I on C then f(z) and f(z) + g(z) have the same
number of zeros within C.
This result, usually called Rouche's theorem; is a simple' corollary of

THEOREM:

the principle of the argument.


Since lg/ fl < 1 on C it follows that the function 1 +g/ f assumes
values only inside the disc Jz- l J < 1 while z varies on C. Thus
0. Rouche's theorem then follows directly:
[Arg(l + g/ f)] c
=

[Arg (f +g)]c

[Argf(l +glf)]c

:= [Arg f 1c + [Arg(l + glf)]c


[Argf1c
where we have used the result of Exercise 7, Question
=

1.

122

[Ch. 7

MANY-VALUED FUNCTIONS

Prove the fundamental theorem of algebra that a poly


nominal of degree n 1 has exactly n zeros.
n
Let fn(z)
L a.z where an =I- 0.

Example 2.

r=O

Now it is obvious that


.
LIm

lzl->ao

'

fn-1(z)
n
anz

,- _
_

'
LIm

lzl->ao

IanIIzI

=0.

an-1 +

an-2
z

. . . 1
+

n
z -1

Thus if we choose f (z)


anz" and g(z) =f,, 1(z) in Rouche's theorem
and take the curve C to be jzj
R (R _... oo) we have that jg/fj < 1 for
R large enough and the fundamental theorem of algebra follows from the
observation thatf(z) vanishes n times at z
0.
=

If C

is real and greater than 2, show that the equation


C z4 =sin z
has one zero root, one positive root, and two complex roots within the
circle lzl
1.
(SHEFFIELD 1951)

Example 3.

Clearly z
0 satisfies the given equation and furthermore it is evident
from Fig. 7.3 that there also exists a positive root less than unity. There
=

Fm. 7.3

is no other real root. If we can show that j Cz4 j > j sin zj on jzj
1 then
it will follow from Rouche's theorem that Cz4 - sin z has four zeros
=

Sec.

.
POWERS OF

7.4]

123

inside lzl 1 since Cz4 has. In other words we wish to show that
!sin zl 2 on lzl 1 because 1Cz41 C > 2 on this circle.
eiz -iz
!sin zl
=

I ; I

-!<leizl + le - izl)

coshy.
Now as y increases (or decreases) from zero, coshy increases from
unity so that its maximum value (subject to lzl 1) occurs at y 1
(or -1). Thus we have
=

!sin zl

+).

The value of e clearly exceeds unity and, because n !


we may write
e

<

1
1
1
1 +-+-+-+
1 ! 2! 2!

+-+

n!

> 2"-1

for n 3,

(> 1)

1 1
1
1+1+-+-+ ... +--+
...
n
2 -l
2 22

1
=1+1-t
=

3.

Hence we conclude that


!sin zl

<

t(3+1)

=2

which is what we require.


Actually e can be limited a great deal more precisely than above
(e.g. to 5 decimal places e 271828) so that the value C 2 is not the
'best' value of C for which there exist four roots of Cz4 sin z inside
the unit circle.
=

Powers of z
If z is real and positive, and p and q are integers then zPlq is the unique
real positive number whose qth power is zP. The following relationship
holds
7 .4

zPlq

exp{(p/q) log z}.

(7.4.1)

124

[Ch . 7

MANY-VALUED FUNCTIONS

Let us use this expression (7.4.1) to define


complex numbers:
a
z

t'"

when z and rx. are general

exp(rx.logz),

0,

z = 0,

(X # 0

1,

z = 0,

(X = 0.

(7.4.2)

The quantity za so defined is clearly one-valued because log z is one


valued, but because log z is discontinuous on arg z
n it follows that
a
z is also discontinuous there.
The usual index law holds:
=

(7.4.3)
This is clearly so because by (7.4.2) we have
a' a2
z z

However, it is

not

exp(rx.1 logz)exp(rx.2logz)

exp{(rx1 +rx2)log z}

necessarily true that

because
z = exp{rx.(log z 1 +logz2) }

= exp{ rx(logz1z2:+ 2Nni)},


N7ti

= (z1z2 rea

(7.4.4)

where N has one of the values -1, 0, 1 (see Exercise 7, Question 1).
Now suppose that we wish to solve the equation w za for z. One
1
solution is clearly w fa but there may be others because
=

exp(logw) = exp(rxlog z)
implies that
logw+2nNi = rxlogz
where N is any integer.
Therefore we may write
2nNi
1
-logw+-(X
(X

logz

Sec. 7.4]
or

125

POWERS OF Z

1
- Logw

log z.

IX

Taking exponentials we have


exp{(1/ix)logw + 2nNi/ix}

exp(log z)

w1faexp(2nNi/ix)

exp{(l/ix)Logw}

z.

or

We shall denote by w the many-valued function exp{(1/ix)Loiw}


and refer to the one-valued function w1ta = exp{(l/ix)logw} as the
principal value of w corresponding as it does to the principal value
of Logw.
Example 4.

Discuss the function on z.


z

exp(! Log z)

exp(! log z) . exp(nNi)

z1/2 . e"Ni.

The function z112 is one-valued in the z-plane cut from z = 0 along the
negative real axis: e"Ni has values 1 corresponding to even and odd
values of N: thusz can be split into two one-valued branches z112
both regular in the plane cut along arg z = n. The derivatives may be
calculated as in Exercise 7, Question 3.
w
If we had set z
re we would have found
=

exp(! log r + !ie)


rlf2ei8/2

Thus if (} increases by 2n we move from one branch of z to the other.


However, as (} increases by 4n z returns to its original value. In this way
we see that an odd number of circuits of the origin corresponds to a
change of branch while an even number of circuits simply reproduces
the original value. Clearly, then, any line through the origin which runs
to infinity will divide the plane in such a way that z is one-valued
(see Fig. 7.4).

126

[Ch. 7

MANY-VALUED FUNCTIONS

FIG. 7.4

Example 5.

Discuss the function ;j(z2- l).


;j(z2 -1)

exp{! Log(z2 -1)}.


exp{! log(z2 -l)}exp(n:Ni)
12
(z2 -1) 1 e"Ni.

As in Example 4, ;j(z2 -1) is seen to be a two-valued function. Cuts


along those lines on which z2 -1 is real and negative will divide the
12
plane into regions where the branches (z2 -1 ) 1 are regular. However,
these cuts are rather more complicated than in Example 4; z2 -1 is
real and negative for z
0, -1 fYtz 1 and for z
0 (see Fig.
7.5).
=

-I

FIG. 7.5

The cut along the imaginary axis is not in fact necessary. Let us put
z+l
r 1e;81 and z-1
r2ei82; then we may write:
=

Log(z2-l)

Log(z -l) (z+l)


log r1r2 + i Arg(z2 -1)
log r1r + i(0 1 + 0 )
2
2

Sec. 7.5]

127

INTEGRALS INVOLVING LOGARITHMS

Thus we have

( z2- l)

0 1 + 02

Clearly if

+ )}
2). exp {ti(01 02
(r1r2)112 exp {ti(0 +Oz)}.

=exp(!logr1r

(z2-l)

increases by the amount 4n.then

returns to

its original value while an increase of amount 2n corresponds to a change


of branch. Thus if the number of circuits of the branch points

is even, we end up on the same branch as that on which we started.


If the number is odd we change our branch. It is evident, then, that we
need only divide our plane in such a way as to prohibit an odd number
of

circuits

-1 &tz

to

split

1,

( z2-1)

into

one-valued

0 ensures that

both

branches.

points

The

cut

1 must be

encircled together and so the number of circuits will always be even


(alternatively we could cut the plane along the real axis

&tz

1 to prohibit any circuits at all).

Example 6.

Find the function z such that w

=sin

&tz

-1 and

z.

By definition we may write

w
Setting

A.

eiz we readily

---

2i

obtain

(A.-iw)2

1-w2

Thus, using the notation of the above discussion, we obtain

A. =iw+(1-w2)
which, on taking logarithms, gives us our final result

z
This function is usually
of

ro.

= -i

Log{iro+.q1(1-ro2)}
1w and called
denoted by sin
-

the inverse sine

We may write

z =-i[log{iro(l-ro 2)1'2} +2nNi ]


=-ilogi

{ro(ro2-1)1'2} +2nN

where N is any integer.


The branch corresponding to the plus sign and

=0 is called the

principal branch.

7.5

Integrals Involving Logarithms

It will have no doubt occurred to the reader that the techniques of


contour integration developed in Chapters 5 and 6 may require modi
fication if the integrand involves many-valued functions such as
'

,Yz and

128

MANY-VALUED FUNCTIONS

[Ch. 7

Log z. In point of fact a little care and understanding is all that is required
to extend our existing

techniques as

the following examples will


::

demonstrate.
Example 7.

Evaluate

f x0-1(1 +x)-1 dx

where 0 <a< 1.

This is a standard integral which arises in the mathematical theory of


gamma functions. Accounts of its evaluation may be found in most
textbooks on complex variable theory. If we make the substitution

x = e we obtain the result

f fro :
dx =

and in so doing we have, as it were, removed the logarithm. Let us treat

the contour integral

a
e z

z
l+ez d

--

where C is the contour drawn in Fig. 7.6.

-s

R
FIG. 7.6

Our choice of contour has been influenced by the location of the poles

of the integrand at z = (2N+ l) n: i where N is any integer

( = -1 when
x = 0, y = (2N+ l)n:). This choice of contour encloses only the pole at
z = ni where the residue is given in the usual manner as
.

(z -ni)eaz

z-+ni

1 +ez

Lim

= -ea"i.

Let us now apply the theorem of residues :


R

I-s
f-s

l+

dx+

a(x+2ni)
e

1 +e
...x

2"

I0

ea<R+iy)

l++iyd

dx+

e(-S+iy)

1 +e -S+iy
2n

dy - -2nie0";.

sec. 7.5]

INTEGRALS INVOLVING LOGARITHMS

129

Now , clearly, the second and fourth integrals on the left-hand side of the
above equition vanish as R and S tend towards infinity because
.

IJ
and

IJ

2n ea(R+iy)

e"R

d
1+eR+iy y

--+

ea(-S+iy)

2n

1+e

Thus we obtain

oo

d
-S + iy y

e "x
--

1+ex

dx

eR

-1

. 2n

0 as R

--+ oo

sincea < 1,

e-aS

1-e-S. 27t
-

--+

0 as S

--+ oo

since a > 0.

- 2nie" i
1 e2an1
"

- co

2ni
e""i-e-ani
7t

sin na
The solution of this problem , as outlined above, circumvents the problem
of coping with the logarithm rather than dealing with it directly. Let us

+I

FIG. 7.7

then attack the problem afresh without avoiding the issue. Consider the
contour of Fig. 7.7 which avoids the cut which we have associated with
logz. The function z"-1/(1-z)
exp{(a-l)logz}/(1-z) has a simple
=

130

MANY-VALUED

pole of residue

-1

at the point

[Ch. 7

FUNCTIONS

1.

z =

Thus integrating this function

round the contour of Fig. 7.7 we obtain

f- r
f
-r

1-x

-R

-R

lxl +in)}
dx+

exp{(a- l)(log

f-"
f"

exp{(a- l)log e;9}. rie;9

exp{(a- l)(log

1-x

lxl-in)}
dx+
=

d(}

1-re'9

"

exp{(a- l)log Re;9}. Rie;9


1 -R eie

-n

d(}

-2ni.

If now we change our variable in the first and third integrals of the
left-hand side from

x to -x we
ei"(a-1)
.

find that

I
R
I

xa-1
_
_
d
x+I
l+x
r

-e-in(a-1)

where

a-1

d
x+J
l+x

-2ni

2nr0
1-r

--

--+ 0

as

r --+ 0

since a

>

0,

and

IJ I=

If"

exp{(a-l )(log R:i O)}.

1-Re'9

-x

Rie;9
d(}

2nR0
R- 1

-
""

--+ 0
Thus, letting

r--+ 0

as

and

R --+

R --+

oo

oo,

2i sin n(a- 1)

since a

<

1.

we obtain the result:

a-1

dx
l +x

-2ni

Sec. 7.5]

INTEGRALS INVOLVING LOGARITHMS

oo

or

xa-1
1 +x dx

131

7t
=

sin na

The reader may readily verify that the above result can also be obtained
by integrating

za- 1/(1- z) round the contour of Fig. 7.8. Proceeding in

Fm. 7.8

[-

the usual manner one is led to the equation

.
xa-1
1_ x dx-e'''

[ --

xa-1
dx+ni = 0
1+x

from which one may deduce, by equating real and imaginary parts that

xa-1

oo

and

1-x

7t

dx= --

tan na

7t
Xa-1
--dx=-sin na
1 +x

As a means of solving the original problem the first method is evidently


the most straightforward. However, the third method outlined above
does provide the additional result not obtainable by either of the other
methods.
Example 8.

Evaluate

So )x log x(l +x2)- 1 dx.

There is no new principle involved in this example and an outline


solution is provided here without comment. We shall integrate the

132

MANY-VALUED

[Ch. 7

FUNCTIONS

function etoaz log z(l+z2)-1 round the contour of Fig. 7.7.

J:
J:"
J:
r"
g

exp {!<log lxl+in)}. (log lxl+ in)(l +x2 )-1 dx

exp {!(log re;6)}. log re;6 (1 +r2 e2;6)-1riei6 d()

exp {!(log lxl-in)}. (log lxl-in)(l+x2)-1 dx

exp {!(log Re;6)}. log Re;6 (1 + R2e2i6)-1Riei6 d()

2ni

exp(t log i)log i -

i exp(t log - i)log -i}

Proceeding to the limit (r,R)-+ (O,oo) we obtain the result

[
0

.Jx logx(l+x2)-1 dx

n2
2.J2

where we have noted that

2nr3'2{(logr)2+n2}1'2
1-r2

-+ 0 as r-+ 0,
and

I I"

exp {!(log Re;6)}. log Re;6 (1+R2 e2;6 )-1 Riei6 d()

""

2nR3f2 { (log R)2+nz}1/2


---=R2-l

---

-+ 0 as R-+ oo.
(Recall that x" log x, in the limit as x tends to zero or infinity, is dom
inated by x" for all values of n.)

Ex. 7]

MANY-VA LUED FUNCTIONS

13 3

EXERCISE 7
1. Use the definitions of Argz, argz, Logz, and logz to prove that
(i) argz1z
argz1 +argz +2Nni
2
2
(ii) logz1z
logz1 +logz +2Nni
2
2
(iii) Argz1z
Argz1 +Argz
2
2
(iv) Logz1z
Logz1+Logz
2
2
where N has one of the values -1, 0, 1.
2. Prove that every value of n Logw is a value of Logwn. Is the converse true?
(Answer: No. For example Logw2 2 logw+2nNi while 2 Logw
2 logw+4nNi for any integer N)
3 . Find the regions where the functions z" and rr are regular and
calculate their derivatives there.
1
(Answer: (Xz"- , (Xz log (X )
4. Prove that
z4+2z3+3z2+4z+5
0
=

has no real or purely imaginary roots and that it has one complex root
in each quadrant
(EDINBURGH 1954)
5. Prove that, if a is real, kzn - cosh az
0 has n roots inside the
circle lzl
1 if !kl > cosh a .
Deduce that 2z3
coshz has three roots inside the circle lzl
1 and
determine, graphically or otherwise, whether any of the three roots is
(U.C.N.W. 1963 )
real.
6. If the many-valued function z
cos-1w is defined by the equation
w
cosz prove that
cos-1w
-iLog{w+(w2-ljJ
=

7. By considering the function (logz)2/(z2+4), or otherwise, prove


by contour integration that

(x2+4)- 1 log x dx

log 2.
(HULL 1960)

8. By considering the integral of (logz)2/(z2 - 3z+2) round the contour


drawn in Fig. 7.7 show that

oo

oo

log x
dx
x2+3 x+2
(log x)2
dx
x2+3 x+2

1
Wog
2)2

.
=

t log 2{ n2+(log 2)2}.


_

134
9. Prove that

MANY-VALUED FUNCTIONS

{(z-1)/z}

is a two-valued function, each branch of

which is regular in the plane cut along the real axis from 0 to 1.

(z+t)-2{(z-1)/z}112 taken round


prove that, if t is real and positive

By considering the integral of

closed contour in the cut plane,

1 (1-x)112
n
x - ----.=---,...,.=---=- d
___,.
=
0 x111 (x + t) 2
- 2 t 3f2(l + t)112

(EDINBURGH 1956)

CHAPTER 8

Beta, Gamma and Delta


Functions
The gamma function, and the closely related beta function, are of very
common occurrence amongst the improper integrals which arise in the
solution of physical problems. Dirac's delta function is of a very different
nature; indeed, strictly speaking, it is not a function at all. It is a com
parative newcomer to the mathematical scene but is already a well
established tool of the engineer and mathematical physicist; for this
reason it merits inclusion in a book such as this.
8.1

Convergence

In the present chapter we shall deal frequently with functions defined


by improper integrals and it is necessary that we should have some
criterion whereby we can judge, without explicit evaluation, whether or
not a given integral exists. A useful technique involves the generalization
of the G.P.C. introduced for series and sequences in Section 3.1.

f(z) of a complex variable z


F(z,t), where t is real, by the

Let us suppose that we have a function


which is defined in terms of a function
relationship

f (z)

F(z,t) dt.

(8.1.1)

Further let us suppose that the integral is improper in some respect


at

t0, a

t0

b, where

small positive number

e,

or b may be infinite. If now we choose any

and we can derive the inequality

I I:

F(z,t) dt

whenever both R1 arid R2 are sufficiently close to


integral

J F(z,t) dt

(Kl.2)

t0,

exists. We offer no proof of this result.


135

then the

136
Example

BETA, GAMMA AND DELTA FUNCTIONS

1.

Show that f0 e-rtz dt exists only for

z >

[Ch.

-1.

The given integral is improper in two respects. The range is infinite


and the integrand may be infinite at

0. It is evident that

t tends
t/t vanishes. Thus x log t/t will certainly be less than
t provided x is finite and t is chosen sufficiently large.
Because rn e-tt dt vanishes as R1 and R2 tend towards infinity we
Let us consider first the upper limit of the integration range. As

towards infinity log

may assert that, given any small positive number e,

whenever R1 and R2 are chosen sufficiently large. This is the criterion


stated in equation

(8.1.2).

We may therefore conclude that

J;:i e-1tz dt, a

> 0, exists for all finite

values of z.

t
e-1 tends towards unity. Consequently, for small

Let us now consider the lower limit of the integration range. As


tends towards zero,
enough values of

t the following relationships hold

where both R1 and R2 are chosen to be sufficiently small. When x >

-1

we can readily check that

Therefore we can use the G.P.C. to infer the existence of the integral

J'O e-1tz dt, provided z > -1 and a is finite. In the case x -1 the
integral J'O r dt does not exist and the reader will easily convince himself
that this implies the non-existence of the integral J'O e-1tz dt for z -1.

Sec.

8.2]

8.2

The Factorial Function

137

THE FACTORIAL FUNCTION

It is a simple matter to establish, by repeated integration by parts,


that

n!
where

f00

(8.2.1)

e-1t"dt

is a positive integer. This last proviso seems hardly necessary

because we do not normally ascribe a meaning to

n!

unless

n is a positive

integer. Restrictions of any kind irk the pure mathematician and he is

ever at pains to remove them as we saw in Chapter


introduced i

to generalize

when we first

J-1. The formula (8.2.1) provides a golden opportunity

n!

to almost all complex values of

z.

Let us define

z!

by

the formula

z!

e-ttz dt,

The restriction to values of

with

>

-1.

>

-1

arises in connection with

the convergence of the given integral (see Example


We shall show in Section
The function

z!

regular function of

(8.2.2)

for

derivative

(z

of Section 8.1).

>

z -1.
-1 is in fact

how z ! may be defined for

8.4

defined by equation

z with

(8.2.2)

!)

e-1tz log t dt

(8.2.3)

This result, obtainable by differentiation under the integration sign.


can be established rigorously. We shall assume without proof the truth
of this result.

Example 2.

Show that

(-!)!
and hence deduce that ( -t) !
The function ( -!)! occurs

exp(-u2)du

Jn.

in the theory of errors and in many other

problems of a statistical nature.


By definition we have

( -!}!
If, now, we set

u2,

e-1t-112 dt

exp(-u2)du

we obtain
(-!) !

10

21

say.

138

BETA, GAMMA AND DELTA FUNCTIONS

[Ch. 8

But we can write 12 as a double integral as follows:


12

f
ff

du exp( -u2)

oo.

f J exp(

v2)

d8

Let us change our variables to


sin 8, so obtaining the result

r2)r dr d8

: f
J
12

dv exp(

exp{-{u2+v2)} du dv

where A is the region 0 ::::;; u,v ::::;;


polar co-ordinates u
r cos 8, v
12

f,

exp(

r2)r dr

7t

=4.
Hence we obtain our final answer
(

t)!

Example 3.

By setting t

x+
x!

show that

.e

-x

+
x1 t

f"'

-..;

2J()
Jn.

J x in

the formula

x
t e-1 dt

exp{ --rJ x+x log(l + -rx-t)} d-r


x

Deduce Stirling's approximation for n ! :


n larg.e.

Stirling's formula occurs frequently in statistics and in classical


statistical mechanics.
The first part of this example is'simply a matter of change of variable.
To obtain Stirling's formula it is necessary to expand log{l +-rx-t)
in a series as follows

Sec.

8.3]

so that, for

139

ANALYTICAL CONTINUATION
x

large, we have

f"'

i
exp( - -r

/2) d-r

J(2n)

where we have used the result of Example


Hence we obtain

2.

Stirling's approximation as enunciated in the

question.

8.3

Analytical Continuation

Suppose that we have a functionfi(z) which is regular in some region


Ai and a second functionfi(z) regular in another region Ai (see Fig.

8.1).

FIG. 8.1

If, now.fi(z)

fi(z) in the region which is common to both Ai and

Ai we caII fi(z) the analytic continuation of fi(z) into Ai. Equaily we


might refer to fi(z) as the analytic continuation of fi(z) into Ai It is
usual to regard the aggregate of values offi(z) andfi(z) in Ai and Ai as
values of a single functionf(z) analytic in Ai and Ai:
f(z)

fi(z)

for

fi(z)

for

The reader may suspect that the matter discussed above is a trivial
one: the author will endeavour to disabuse him of any such sentiment.
It can be proven that fi(z)

fi(z) throughout a suitable region A in

which both are regular provided there exists some sequence

Zn

of points

140

BETA,

[Ch. 8

GAMMA AND DELTA FUNCTIONS

of A such that (i)fi(zn)

f2(zn) for all n, (ii)

Zn --+

A. as n

--+

oo, where A. is

also a point of A.
Suppose then that we have a function f0(z) defined and regular in A0
and that we continue it analytically into regions Ai and A2 (see Fig. 8.2)
by means of the functionsfi(z) andf2(z) respectively. Let A be the region
common to Ai and A2 If A overlaps A0, as drawn in Fig. 8.2, and the

FIG. 8.2

dotted portion of A possesses the property stated in the previous para


graph, then fi(z)

f2(z) throughout A. Consequently it is immaterial

which function fi(z) or f2(z) which we use to continue f0(z) into A. The
power of the method of analytic continuation lies in this uniqueness property.
In this chapter the technique of analytic continuation will be demon
strated in action. It will be used to extend the definition of functions and
it will be used to generalize the validity of formulae proved for restrictive
values only.
2
By considering the Laurent expansions of 1/(l -z ) about

Example 4.
the points z

1 show how the function


0< 1 1-z l < 2

may be continued into the region 0< I 1 + zl < 2.

Sec.

8.4)

141

CONTINUATION OF THE FACTORIAL FUNCTION

By partial fractions we obtain

1 -z2

=
=
=

1 1

2 1-z

\
1 +z}

(1-zr ----2 n +1-z}


4 n=O

!(f
!(f

0<11-zl < 2

(l+zr \
n +
l+;
2
4 n=O

where the restrictions arise from the requirement that the geometric
series converge. As we can see from Fig. 8.3 the regions 0< I 1- zl < 2

FIG. 8.3

and

0< I 1 +zl < 2

overlap and in this common region the function


f-1(z) =

( +z)"
l n +
\
1+
2
n=O

!(f
4

is equal to the given function /1(z) since both are equal to 1/(1- z2).
Hencef_1(z) can be regarded as the continuation off1(z) into the region
0< ll+zl < 2.
Clearly the function 1/(1-z2) might be used to continue either f1(z)
orf 1(z) into the whole z-plane apart from the points z
1.
_

8.4

The Continuation of the Factorial Function

By integrating equation

(8.2.2) once by parts we obtain the result


z! = z(z-1)!
(8.4.1)

which is the familiar property used to define n! for positive integral

n.

142

BETA, GAMMA AND DELTA FUNCTIONS

The function /1(z)

[Ch.

z! is defined and regular in the region A1 which


91z
-1.
The function /2(z)
(z+1)!/(z+1) is defined and regular (apart from
a pole at z
-1) in the region A2 comprising all points to the right of
the line 91z
- 2.
Furthermore in A1 we have, by (8.4.1), that/1(z)
/2(z).
We may therefore continue /1(z) into the strip -2 < 91z < -1 by
=

comprises all the points to the right of the line

means of the formula

z I.

(z+ l)!

---;+1

The process is illustrated in Fig.

(8.4.2)

8.4.

f1 f2, both functions


defined and regular
=

FIG. 8.4

The process may be repeated indefinitely and

91z

of the complex plane to the left of the line

z! extended to all points


-1 by means of the
=

formula

zl.

(z+n)!
=

-------

(z+n)(z+n-1)

(z+1)

The factorial function so defined is regular for all


poles at the points
8.5

-1,

-2,

(8.4.3)
z apart from simple

-3, . . . .

The Gamma Function

The gamma function is defined as follows :

r(z)

(z-1)!

(8.5.1)

This function is not significantly different from the factorial function.

Sec.

8.6]

THE HANKEL

143

FUNCTION

It is usually introduced by means of the improper integral

r(z)

I:

e-ttz- l dt,

fJlz

>

(8.5.2)

and continued analytically by means of the recurrence formula

r(z)

(8.5.3)

(z-l)r(z-1).

Some writers are conducting a campaign to get rid of the historical


gamma function terminology in favour of the factorial nomenclature.
The author has no strong sentiments in this matter. He has been in
the habit of using the gamma notation but cannot deny the logic under
lying the desire of others to change. The reader can decide for himself.
Both notations will be found in this book.
8.6

The Hankel Function

Let us look briefly at another method which has been employed in

n !. If
(n !)-1 we

an attempt to generalize the notion of

we notice that the residue

of the function

see that

e1t-<n+ 1>

at

_!_
n!

0 is

_ J. ett-<n+lldt
_1
2niJc

where C is a circle enclosing the point


The

Hankel function

is defined as

H(z)

- 1.
2m

where C is the contour drawn in Fig.

(8.6.1)

0 and

e1t-z dt

is a positive integer.

(8.6.2)

8.5. This

definition is valid for all

FIG. 8.5

finite
of

as the contour integral can be shown to exist for every finite value

(see Exercise

8,

Question

1).

If

z is

an integer the integrals along the

144

.BETA, GAMMA AND DELTA FUNCTIONS

[Ch. 8

negative real axis cancel out and the theorem of residues gives us
H(z)= {(z-l )!}-1,
= 0,

z> O
z:::;; 0.

(8.6.3)

The function H(z) defined by equation (8.6.2) for all finite z can be
shown to be a regular function of z. This we shall assume.
A pertinent question is to decide whether or not the Hankel and
gamma functions are reciprocals. The remainder of this section is devoted
to the answering of this question.
Let us split up the contour C of Fig. 8.5 into three portions as follows:
2niH(z)=

f
f,,

exp{t-z(log ltl-in)} d t
exp{rei8-z(log rei8)}ir ei8dO+

= (ei"z-e-"i z)

f,00

exp{t-z(log ltl+in)}dt

e-tt-z dt+l

where

--+ 0 as r--+ 0

if ffiz<1.

Hence we may deduce that


nH(z)= sin nzI'(l -z), ffiz<1.

(8.6.4)

H(z) and sin nz are both regular functions of z for all finite z. I'(l -z)
is regular apart from simple poles at the positive integers. Thus we can
deduce that right-hand side of equation (8.6.4) is regular apart from the
points z= 1, 2, .... It is clear that we may use the method of analytic
continuation to assert the validity of formula (8.6.4) for all finite values
of z; we must, of course, use the values of H(z) at the positive integers to
fill in the missing values on the right-hand side; as it were, the zeros of
sin nz cancel the poles of I'(l -z) at the positive integers. Let us write
nH(z)

sin nzI'(l -z).

(8.6.5)

In the following Section 8.7 on the beta function we shall prove that
I'(x)I'(l -x)=

fro
0

t-x
-dt,
1 +t

O<x<l

(8.6.6)

Sec.

8.6]

145

THE HANKEL FUNCTION

and for the present we shall assume the truth of this result which may
be rewritten using Example 7 of Chapter

r(x)r(l - x)

7t
=

. -

smnx

as follows:

O<x<l.

(8.6.7)

This formula can be extended using the method of analytic continu


ation to complex values of

z;

the result is

r(z)r(l-z)
provided

7t
=

(8.6.8)

. -

smnz

is not an integer. To see this we need only note that the

infinite sequence

x" n/(2n+ 1), n 1, 2, ... , of points of the interval


0<x< 1 has limit !. Thus, by the criterion of Section 8.3, equation
(8.6.7) may be extended to all finite values of z for which both sides of
the equation are defined and regular: this leads to the result (8.6.8).
=

v
x

-I

FIG. 8.6

If now we multiply both sides of equation

nr(z)H(z)

where we have used equation

(8.6.5)

,,,,:,,

sin nzr(z)r(l-z)

sin nz

(8.6.8).

7t
.-

smnz

by

r(z) we

obtain

146

BETA,

[Ch.

GAMMA AND DELTA FUNCTIONS

Therefore, provided

is non integral, we have proved that

r(z)H(z)

1.

(8.6.9)

The matter of filling in the points of integral

is trivial as the right

z. We already know
1
(z-1)! and H(z)
[(z-l)W at the positive integers and
equation (8.6.9) itself may be used to fill in the points z
0, -1,
2, . . .
We know that H(z) vanishes for integral z 0 and that r(z) has simple
poles at these points. Clearly the zeros of H(z) must cancel the poles of
r(z).
Incidentally since H(z) is regular for all finite z, r(z) can have no
hand side always exists regardless of the value of

that

r(z)

-.

finite zero.
A graph of
8. 7

r(z),

for

real, is drawn in Fig.

8.6.

The Beta Function

Almost every schoolboy knows how to evaluate the definite integral

tt/2

B(m,n)
where

and

sin

m-

1e

co s

- 1ede

(8.7.1)

are positive integers or semi-integers, by means of a

recurrence formula obtained by integrating twice by parts. Subsequently


as an engineer or physicist he will encounter such integrals repeatedly
although

and

will assume more general values.

For example, in the classical problem of a portion of uniform width


of a sloping roof being cleared of snow, the time taken for the roof to
be cleared is proportional, under suitable assumptions, to
The

beta function

B(z1,z2)
where both

B(!,-j).

is defined as follows :

2'z 1 and 2'z2

'2

J:

sin2z' - iecos2z2-1ede

(8. 7.2)

must be positive for the existence of the integral.

A more usual form of the beta function is obtained via the transformation

sin2e:

B(z1,zz)

1
tz'- (1-ty>-1dt.

This latter integral was studied, for


Question

z1

and

z2

(8.7.3)
real, in Exercise

6,

1.

We shall now prove, for

z1

and

z2

real and positive, that

(8.7.4)

Sec. 8.7]

147

THE BET A FUNCTION

Let us consider the product

r(z1)r(z2)=

tz1-1e-1d t

uz2-1e-udu

as a double integral over the positive quadrant as follows:


r(z l)r(z2) =

ff tz1 - luz,-1 e-<t+u) dt du.

Let us now change the variables from t and u to


2
2
of the transformation t = OJ cos (), u=OJ sin 0.

(8.7.5)

OJ and () by means

Noticing that the Jacobian of the transformation is given by


.

8(t,u)

20J cos e sm e
a(OJ,O) =
we obtain

II OJZl +z,- le-"' cos2z1 - 1() sin2z2- l(J dOJ d()

r(z )r(z2) = 2
l
=
.

oo

OJZ1 +z,- le-"' dOJ. 2

= r(z1 +z2)B(z1,z2).

/2

2
cos z1 - le sin2z,- l(J d()

Now, because the right-hand side of formula (8.7.4) is defined and


regular provided neither z1 nor z2 is a non-positive integer we may
extend this formula to complex values of z1 and z2 with positive real

parts: furthermore it can be used to continue B (z1,z2) into the left half
z1 and z2-planes.

Example 5.

Show that r(x)r(l -x) = So t-x(l + t)-1 dt, 0 < x < 1.

Using result (8.7.4) we obtain


r(x)r(l-x) = r(l)B(x,1-x)
1'/2

J
[

= 2

2
cos x-1e sin l- 2x()d()

where the use of the integral form of the beta function is justified since
2
both x and 1- x are real and positive. Let us now set t=tan e.
/2
r(x)r(l-x)=2

oo

J
[

= 2

tan1-2x()d()
1
dt
tt-x. 2 lf l
2
t ( +t)

t-x
-dt.
1 +t

148

BETA,

8.8

[Ch.

GAMMA AND DELTA FUNCTIONS

The Dirac Delta Function

Let us consider the step function f(x) which is defined as follows

f(x)
f(x)

1/e,

0,

-e/2

e/2

otherwise.

The area enclosed by such a function and the x-axis will clearly be
unity (see Fig.

8.7).

If now we consider the limit of the function f(x) as


f(x)

-/2

FIG. 8.7

-+

perty
that

we obtain the so-called Dirac delta function which has the pro
<5(x) 0 provided x # 0, <5(0) oo, together with the condition
J <5(x) dx 1 in order that the 'area under the curve' remains
=

unity. Clearly this is not a function in the usual sense of the word but
physical parallels are legion.
Let us suppose that we have a physical system which comprises all
points of a straight line; suppose further that we have no electrical
charge at any point of the line except

whereat is located a unit

charge. Poisson's equation which must be satisfied by the electrical


potential function </> is

82
ox2 </>
where

i s a constant and

all points of the line except


the line is unity so that
that

J p(x) dx
00

ap(x)

(8.8 1)
.

is the line density of charge. But

at

and furthermore the total charge on

p(x) is infinite

at

in such a way as to insist

1. Clearly then our physical situation involves the

Sec. 8.8]

TH E DIRAC D ELTA FUNCTION

149

solution of the equation


fJ2

ox2 </> = ab(x).

(8.8.2)

The restriction of the argument to a one dimensional system is simply


a matter of convenience.

For any suit a bly w ell-behaved functi on f(x) the foll owing
r esults hold:
(i) J!f(x)b(x) dx =f(O) if a 0 b
= 0 otherwise
(ii) S:f(x)b(x-b)dx =f(b) if a b c
= 0 otherwise
(iii) Jf(x)b(x-b) dx = Sf(b)b(x-b) dx
(iv) J:f(x)b(bx)dx = lbl- 1 J: f(x)b(x) dx
(v) Jf(x)b'(x) dx = -f'(O) if a 0 b
0 otherwise.
THEOREM:

Let us consider the assertion (i). If zero is not included in the closed
interval (a,b) then the fact that b(x) = 0 implies (for suitable functions
f(x)) that the integrand and hence the integral vanishes. Let us suppose
that zero does lie in the range of integration. Given any positive number
e which may be chosen to be arbitrarily small, it is clear that

f(x)b(x)dx =

f::,
2

f(x)b(x)dx

. (8.8.3)

because, for lxl ;;::: e/2, b(x) = 0.


If e is chosen small enough, and f(x) is suitably well-behaved, f(x)
will be effectively constant in the interval ( - e/2,e/2). Therefore we may
write:

r2

e/

r2
f

f(x)b(x)dx =f(O)

e/ 2

=f(O)

b(x)dx
b(x)dx

00

=f(O).
Results (ii), (iii) and (iv) are more or less obvious and are left as an
exercise to the reader. Result (v) may be established by integrating once
by parts.

BETA, GAMMA AND DELTA FUNCTIONS

150

[Ch. 8

While it is important to realize that the delta function and its deriva
tives have meaning only under a sign of integration, it would be very
convenient in practice to have an explicit form for the function.
THEOREM:

2m5(x) =

Lim J00 exp(itx -!e2t2)

dt provided we interpret the

e-+O

given expression under an integration sign with respect to x.


Let us examine ,the prorties of the given limit.

f00 exp(itx-!e2t2) dt

= exp( -x2/2e2)

[00

= exp (-x2/2e2)J,

exp{ -!e2(t-ix/e2)2}

dt

say.

In order to evaluate I it is necessary to make the change of variable

u=
We obtain

I=

J2 -:).

J .Je2 e-ui du
c

where C is the contour drawn in Fig. 8.8 in the limit R--+ oo.

-ER/./2

ER/./2

FIG. 8.8

Now, provided the integrals of

e-u2

along the dotted lines in Fig. 8.8

vanish, we may write, in the limit R --+ oo,

I=

2 [
1

00

e-ui du

= -.j(2n)
e .
where we have used the theory of the gamma function (see Example

2).

Sec. 8.8]

THE

DIRAC

151

DELTA FUNCTION

Let us now verify that the appropriate contour integrals do indeed


vanish. For example let us consider the integral of

Blu= eR/J2 from Ju=

0 to

Ju= -x/eJ2.

e-"2 along the line

e the right-hand side


R becomes infinite.

For any fixed positive number


equality tends towards zero as

of the above in

We conclude, therefore, that

[00

exp(itx-!e2t2) dt=

00
J- 00

J(2n)
8

exp (-x2/2e2).

Thus we have immediately that


Lim
e-o

exp(itx-!e2t2) dt=

= 00

if

xO

if

x=

0.

Furthermore, by using the theory of the gamma function, we obtain

f: 00{ J: 00

exp(itx -!e2t2) dt dx=

J n)

= 2n.

00
oo
. J-oo

00

exp( -x2/2e2) dx

'--)

exp(itx-!e2t2) dt

Therefore the 'function' Lim f


e-o

is zero everywhere

except at the origin where it is infinite in such a way that its integral
is

2n.

Clearly then we have the required result

c5(x)= -

27t

Lim
e-o

which we write formally as


1

(x)=
c5
2n

exp(itx-!e2t2) dt

oo
J- oo

eitx dt.

(8'.8.4)

(8.8.5)

Example 6. If the function g(k) is defined implicitly by the formula


(x)=

J( n)

Joo

g(k)eikx dk

152

show that

BETA, GAMMA AND DELTA FUNCTIONS

1
oo
oo f(x)e-ikxdx.
J(2n) J
Fourier's inversion theorem

g(k) =

This is known as

[Ch. 8

and has numerous

applications to physical problems.

By definition of the delta function we may write

g(k) = J00 b(k- k')g(k')dk'

1
=J dk' n[ dx e-i(k-k'Jxg(k').
oo
oo 2

where we have used the result of the above theorem. Assuming that the
implicit limiting processes involved behave suitably we may reverse the
order of the improper integrations to obtain

1.

1
oo
e ikxf(x)dx.
J(2n) J-oo -

EXERCISE 8

Discuss the convergence of the integral

(Answer: The integral


2. Derive the result

H(z) =
2m

where C is the contour depicted in Fig.

By setting

fl/2
0

show that

provided

tx-1( 1 -w-1 dt,

J(n)r(2x) = 22x-1r(x)r(x+!).

Show that the curve

>

r
0.

(8.6.8) for the case x = k.


x2 + y4 = 1 encloses an area

Hence verify the truth of formula

3.

8.5.

exists for every finite

{r(x)}2 = 2r(2x)

t =!-!Ju

e1t-z dt

>

0)

Ex. 8]
4.

BETA, GAMMA AND DELTA FUNCTIONS

153

Show that

.. f"12

(u)

(iii)
(iv)

rr/2

J
f:
0

tan3x+tan5x
dx
exp(tan2x)
y'(tanx) dx

(1-x4)-t dx

l.

!B(i,i)
!y'(n )r{!)/r(!).

By integrating tz- ie-1 round the boundaries of quadrants of a


circle (indented at the origin) whose bounding radii are parts of the
real and imaginary axes show that
5.

(i)
(ii)

I
f

tz-le-it

iz- 1e i1

dt

proed 0 < fllz


Deduce that

and

dt

<

e-irrz/2r(z)

eirrz/2r(z)

1.

J
f

tz- l

cost d t

tz- 1

sint dt

r(z)cos !n z

r(z)sin !nz.

(Hint: Take contours lying (i) in the first quadrant (ii) in the fourth
quadrant)
6. Prove that

lnix)I
where x is real.
11

j( . )
n

x sm

h nx

154

BETA, GAMMA AND DELTA FUNCTIONS

[Ch. 8

(Hint: Consider r(ix) and r(- ix) and use the result that
2
r(ix)r(- x) lr<ix}j )

7. The Fourier transform (F.T.) of the function f(t) is defined by the


relationship

f(t)

y'(n) J

oo

g(k)eikt dk.

Show that the F.T. off(t)eu' is g(k- l).


If
f(t)

-J/2

for

otherwise

show that
g(k)

()

<

<

{J/2

<

sin( J/2)
.

Deduce that the F.T. of F(t) where


F(t)

cos lt

-J/2

for

<

{J/2

otherwise

is given by
t{g(k- l)+ g(k

+ l)}.

(U.C.N.W. 1963)
8. If g 1(k) and g 2(k) are respectively the Fourier transforms of the real

functionsf1(x) andf2(x) show that

00

d J

e-iktg 1(k)g!(k) k

00

f1(x)f2(x

t) dx.

9. If C+ is the contour drawn in Fig. 8.9 and we agree to adopt the


notation

where

.r
Jc+

(z) dz . J_00z-z0+ie
fz-z0
00 (z) dz
f

is a small positive quantity, establish the formal relationship

z-z0+ie ---i
z-z0 ntJ(z-z0)
1

--.

where

stands for Cauchy's principal value.

Ex. 8]

BETA, GAMMA AND DELTA FUNCTIONS

155

10. Use the substitution u = (1 +k)t/(1 +kt) to show that, if k 0 and

m,n

,1

1,

l tm-1(1-t)n-l
1
dt =
. (1+ktr+n+p
(1
+kr+v
0

um-1(l-u)"-1(l +k-ku)"du.

By first transforming to new variables

v and w, given by x+y = v,

FrG. 8.9

f1 :.

x = vw, prove that, if f is a continuous function, the double integral


m

f(x + y) dx dy,

taken over the triangle A whose sides are x = 0, y = 0, x+y = 1, is


equal to

r(m)r(n) l vm+n-1
.f(v)dv.
r(m +n) 0 (1 + vr

ff

Hence show that, if p 1,

xm-lyn-1(1-x-y)p-l
r(m)r(n)r(p)
---,-------,-- dx dy - ----- .
n +p
+x+
+xr+"(l
y)
2m + n r(m +n + p)
(1
A

(EDINBURGH
11. Let

1957)

= JS x-40y2"(1-x2+y2)t-a dx dy taken over the region in


m < 1) and
the hyperbolae x2-y2 = 1, x2-y2 = c2 (0 < c < 1).
(i) Transform the integral by the change of variables x = u sec v,
y =utan v.
(ii) Evaluate the integral when a = 1.
(iii) Using properties of the gamma function, show that, if 0 < a < !,
I

the first quadrant bounded by the lines y = 0, y = mx (0 <

J-+
as

m -+ 1

n(l -2a)cot 2na

------

(1-4a)(3-4a)

and c -+ 0.

(EDINBURGH

12. If the equation f(x) = 0 has

that

b(f(x))

distinct roots x1 , x ,

" b(x-x;)
.
if'(x;)j

J,

1962)

x show
"

CHAPTER 9

Differential Equations
Very often the mathematical model for a physical process is a differential
equation. It is the applied mathematician's job to solve such equations
and the general problem lies quite outside the scope of the present text.
However, in this chapter, an account will be given of certain aspects of
the solution of some of the more important equations of mathematical
physics.
The reader is referred to the text books on differential equations
should he wish to pursue the matter in detail. For example he may
consult

9.1

Ordinary Differential and Difference Equations

by F. Chorlton.

Physical Considerations

Laplace's equation is fundamental in continuum mechanics and can


be written out in terms of spherical polar co-ordinates

V2</J

!!__ (r2 o</J +-2_1_ !!__


r or \ or
r sin () o()

= 0.

fs\in e o()<P_)\+r

(r,fJ,t/I) as

1 z
sin ()

o2
ot/J

follows:

(9.1.1)

FIG. 9.1

A partial differential equation of this type is very often attacked by


the method of

separation of variables.

156

Basically this means that we seek

157

PHYSICAL CONSIDERATIONS

a solution of the form

<P =R(r)E>(e)'l'(l/t):

d2'
d (. . dE>\
1
e }+
n
s
dii
+ e sine de
' i 2e dijt2

r2V2</J r2 d2R 2 dR
+
-<P- = R dr2 -;:- dr

\ID

=0.
The structure of the above equation is such that

(9.1.2)

'-1(d2'/dl/t2)

is

constant ; let us write

-m2
m

where

is a constant. We can see that this must be the case because,

were we to fix

(9.1.2):

(9.1.3)

and

and vary l/t only,

<P

would still have to satisfy

this would be impossible unless equation

The constant

is called a separation

constant.

(9.1.3) held

good.

We now obtain the following simplified differential equation:

r2 d2R
R dr2

2 dR

) ---

- -+--+

r dr

d
0 sine de
1

d
m2
---=0
de
sin2e

-sine-

(9.1.4)

In precisely the same manner as above we may argue that

r2 (d2R 2 dR
=n(n+l)
+
R dr2 -;:-dr

where

n(n+1) is

(9.1.5)

a separation constant.

Thus the original partial differential equation ts replaced by three


uncoupled ordinary differential equations:

r2

d2R
dR
+2r -n(n+l)R =0
dr
dr2

(9.1.6)

m2
d (, . dE>
e
n(n+l)-sin
0=o.
+
)
m
2e
de
sine de
1

In the axially symmetric case (m = 0) the third equation of


be rewritten as

d
dz
where

w =e

and

f.p

-z2)

z =cos e.

dw
+n(n+l)w =0
dz

(9.1.6) may

(9.1.7)

158

[Ch.

DIFFERENTIAL EQUATIONS

(9.1.7) is called Legendre's equation and is of very great importance in


this and other physical problems. In quantum theory, for example, the

separation of Schrodinger's wave equation in spherical polars leads one


to the study of such an equation.
Legendre's equation is generally studied in conjunction with two

similar equations which are associated with the names of Hermite and

Laguerre. These equations are respectively:


d
dz

exp(

- z2)

dw
dz

+2n exp(

- z2)w

(9.1.8)
(9.1.9)

Important instances of the occurrence in physical problems of these


latter two equations are respectively the simple harmonic oscillator
problem in wave mechanics and the solution of the radial part of
Schrodinger's equation for the hydrogen atom.

In the theory of sound and wave disturbances in general the wave

equation

v2d

_.!:.._

o2d

(9.1.10)

c2 ot2

where c has the dimensions of a velocity, is met with regularly. If d


measures the displacement of a point from a plane the wave equation

for vibrations of frequency n may be rewritten as

(9.1.11)
where (r,8) are polar co-ordinates in the plane.
The method of variable separation, as indicated above, can be applied
to the equation

(9.1.11) to yield
d20
d82

+s2

d2R

s2
dR
+!
+ (1 dz2
z dz
z2

where d

R(r)0(8), z

(9.1.12)
R

nr/c, and s2 is a separation constant. The

second equation of the pair

(9.1.12) is called Bessel's equation of order s.

It occurs in many other physical problems notably in the quantum


mechanical theory of potential scattering.

PHYSICAL CONSIDERATIONS

Sec. 9.1]

159)

Example 1. A uniform flexible chain hangs vertically under gravity from


a fixed point. Obtain a differential equation which describes the small
transverse oscillations of the chain.
x

y
FiG. 9.2

Let us measure co-ordinates x and

vertically and horizontally

respectively from the free end of the chain. If we consider the motion of
an element bx of the chain the equation of motion for this element is

b(px)

a1y
at2

b(Tsin t/l)

where Tis the tension in the chain, p its line density, and t/l the angle
which the element tangent to the chain makes with the positive direction
of the x-axis.
But T

pgx where

is the acceleration due to gravity, and because

the oscillations are small


,/,
. ,/,
sm o/ tan
o/

dy
=

dx.

Therefore the required differential equation is

a1y
at2

a
=

If we seek a solution of frequency

d2y

g ax

( )

2nJ(x/g).

ay

ax

we obtain

dy

dz2 + ;- dz +
where

This is just a Bessel equation of order zero.

160
9.2

[Ch.

DIFFERENTIAL EQUATIONS

Solution in Series

We shall restrict our attention to second order linear differential


equations of the form

w" +f (z)w' + g(z)w

(9.2.1)

which are expected, on general grounds, to have two independent


solution functions w1 (z) and w2(z). In physical applications z is usually,
but not always, a real number. The importance of many physical differ
ential equations derives from their solution functions w(z) and we have
seen that our power to handle w(z) mathematically is greatly enhanced
by considering complex values of z. In this present section it matters
little whether the reader thinks of z as real or complex; if he prefers to
take the former viewpoint the technique of analytic continuation will
always be available to extend our results.
A common method of attempting to solve such a differential equation
as

(9.2.1)

is to try to fit an infinite set of numbers p, and a,,

r =

0, 1, 2, ...

such that
00

OJ

"
L..,

arzr+p'

(9.2.2)

r;O

satisfies equation

(9.2.1).

Of course the success or failure of such an

attempt depends critically on the functions f (z) and g(z) in a manner


which can be made precise. In many cases the result of substituting

(9.2.2)

into

(9.2.1)

is of the form

(9.2.3)
where the b, depend on p, a0, a1, a2, .... We now set
b,

0,

and we obtain solutions w(z) for

0, 1, 2, ...

(9.2.1)

(9.2.4)

valid for values of z lying inside

the circle of convergence of the power series


00

a,z'

r;O

(see Section

3.1).

The assumption a0 =/= 0 generally means that the equation b0

depends only on p and is called the indicial equation. For equations of


the type (9.2.1) the indicial equation is a quadratic

(9.2.5)

Sec.

9.2)

161

SOLUTION IN SERIES

The following examples will show that we may or may not be able to
find two independent solution functions by this method.

Example 2. Use the method of series to find two independent solution


functions for the Legendre equation
2
(l -z )w" -2zw'+n(n+ l)w = 0

and show that there exist polynomial solutions if n is a positive integer.


00

Making the substitution w

L a,.z'+P we

r=0

obtain:

00

L a,[(p+r)(p+r - l)zp+r-i - {(p+r)(p +r+ 1)-n(n+ l)}zP+']

r=O

The equations b,

ar+i -

and

a1

(2)

a1p(p+1)

(3)

{(p+r)(p+r+1)-n(n+l)}a,
-'-

--------

(p+r+ 2)(p+r+ 1)

'

(4)

r 0.

(2) has roots p


0 and 1.
0 both equations (2) and (3) will be satisfied with a0
=

arbitrary.

p
1 we must choose a1
0 and so obtain a solution of the form
00

w
where

(1)

a0p(p - 1)

If on the other hand we choose


choose

0.

0 are:

The indicial equation


If we choose

a2,

'\'
L..,

r=O

2
a 2rz '+

is related to the constant

00

'\'
L..,

2
a 2r+ 1 z r+ 1

a0

and

r=O

a2,+1

0. Let us

(5)
to the constant

a1

Since we have now obtained two independent solution functions the

p
1 cannot yield any new independent solution.
n
2N where N is a positive integer the coefficients a2, 0 for
r > N. If n
2N+ 1 where N is a positive integer the coefficients
a2,+ 1
0 for r > N. In either case one or other of the series in (5)

choice
If

terminates and a polynomial solution is found.

Show that the method of Example 2 when applied to the


Laguerre equation
zw"+(l -z)w' +nw
0
Example 3.

does not yield two independent solutions, and discuss the case when n is a
positive integer.

162

DIFFERENTIAL EQUATIONS

[Ch. 9

The equations b, =0 are:

aop2 =0
a,+ 1 =
We are forced to choose

(1)

(p+r-n)a,
, r 0.
( p+r)2

(2)

p =0 and obtain the single solution function

W1 - ao

L..

r=O

(r-n)(r-n-1) .. . z'
(r ')2

(3)

n is a positive integer leads to polynomial solutions.


w2 can actually be obtained by the useful device of
setting w2 = uw1 and solving the appropriate differential equation for u:

Clearly the case when


(A second solution

u" 2w1' 1
-+-+--1 =0.)
u'
W1 Z
Example 4.

Show that the indicial equation for Bessel's equation of

order s

( )

1
s2
w"+-w' 4- 1 --2 w =0
z
z

is (p-s)(p+s) 0. Show further that the solutions of the differential


equation corresponding to p = s are not independent ifs is an integer.
=

The equations b, =0 are

a0(p 2-s2 )=0

(1)

a1{(p+l)2-s2} =0

(2)

a
a, = 2 ,_2 2 , r 2.
s - (p+r)
Let us first choose

(3)

p = s. In this case we obtain


a2,

( - l)'a0
-2--------2 'r!(r+s)(r+s-1) ... (s+1)
a2r+1 =0.

(4)
(5)

a0 = 2-H(s+ 1). With this


a0 this solution is called the Bessel function of the

As a matter of convenience we shall choose


particular choice of
first kind of order

s. It is usual to write
J.(z) =

oo

2:

r=O

( - 1 )'(z/2)2 ' + s
r .'( r+s) .
'

(6)

Sec. 9.2]

163

SOLUTION IN SERIES

If we choose p

s we obtain similarly
J -s(z)

"' (-l)'(z/2)2r-s

r.I( r

r=O

S)I.

(7)

Provided s is not an integer J.(z) and J _5(z) are in fact independent


solution functions of Bessel's equation.
If, however, s is an integer we see that
J _5(z)

(-1) J5(z)

(8)

and consequently J.(z) and J_5(z) are not independent. We can prove
the result

(8)

0,

-1,

=uJ.(z)

and

quite simply by noting that H(z) vanishes for z

-2, ... :
oo

J_.(z)

=I

r=O
00

r=o

(-l )'(z/2)2r-s
r.'( r -s) .'
(-l )'(z/2)2'-H(r-s+l)
r!

-----

5
"' (-l )'(z/2)2 '- H(r-s+ 1)

=I

r=
oo

r=s

r!

-----

(-l)'(z/2)2r-s

'

r.( r-s) .'

oo

(-l)'+(z/2)2r+s

r= 0

( r + s )'.r.'

=I

= ( -1)' J.(z).
(The second solution can always be obtained by setting

solving a differential equation for u:


u"
2 J'
1
----,-+-+u
z
J.

=0.)

In Example 3 we experienced the coincidence of roots of the indicial


equation. A method (due to Frobenius) has been devised for deriving a
second solution from the one first obtained. Let us suppose that the
function w(z,p) is a solution of the differential equation (9.2.1) when

=p1 and p2, the roots of the indicial equation being assumed distinct.
Then we have:

(9.2.7)

164

[Ch. 9

DIFFERENTIAL EQUATIONS

(9 .2.8)

Subtracting these two equations and dividing through by p2-p1 we

obtain

dz
(z) + g
(z)
2+ f
dz
dz

l
J

.!:_

(z, p z)-(z,pi)
w
P2 -Pi

If now we consider the limit as p1

0.

(9.2
.9)

p2 we may conclude that both

-+

and
are solutions of the differential equation.
A similar method applicable to cases when p1 and p2 differ by an
integer will be found discussed in textbooks on differential equations.
9.3

The Formulae of Rodrigue and Schliifti

In Example

2 we saw that one of the independent solutions of

n is a positive integer) is a polynomial of


n containing either odd or even powers of z but not both. This

Legendre's equation (when


degree

terminating series solution can be written in the closed form


(9.3.1)

To show this we set


n

=(z2-l)"

(9.3.2
)

Clearly n satisfies the first order differential equation


dQ
2
(z -1)
dz

2nz Q.

(9.3.3)

If now we differentiate(9.3.3) ( n+ 1) times with respect to z we obtain


n+t
z
" g
d+
d g
d" Q
0
2
(z -l) n+2 +2z
=
n+t -n(n+l) d
dz
dz
z"
2
which is just Legendre's equation for the function d Q/dz .
"

It is usual to define the

Legendre polynomial of degree n as

()
Pn
Z =

1
2" n.I

d" 2
(Z -1)"
n
z

-d

(9.3.4)

where the arbitrary multiplicative constant in the solution function has

Sec. 9.3]

THE FORMULAE OF RODRIGUE AND SCHLAFLI

165

been chosen conveniently. With this choice of constant the coefficient


of the term x" is (2n) !/2"(n!)2 This coefficient is called the leading
coefficient.
Both Hermite's and Laguerre's equations admit polynomial solutions
when n is a positive integer (see Exercise 9, Question 1, and Example 3)
and these too can be simply expressed in closed form.
We define the Hermite polynomial of degree n as
d"

(-l)" exp(z2)d {exp(-z2)}


z"
and the Laguerre polynomial of degree n as
Hn(z)

(9.3.5)

(9.3.6)
The leading coefficients of H (z) and L (z) are thus 2" and ( -1)"
"
"
respectively.
The definitions (9.3.4), (9.3.5) and (9.3.6) are usually known as
Rodrigue's formulae and can be rewritten using Cauchy's integral
formulae (see Section 5.3) in a form due to Schliijli:
Pn(z)
HnZ
( )
L (z)
"

':""

1 J. (,2 -1)"
d(
2n+1 7ti c(,-z)" + 1

(9.3.7)

exp(-(2) d r
(-1)" exp(z2)n!
n+l '>
2n i
c((-z)

(9.3.8)

ez n!J. '"e-
d(
2ni c( (-z)" +I

(9.3.9)

where C is a simple closed curve in the complex '-plane which encloses


the point ' = z.

Example

5.

Show that
ILn(x)I en !(1 +x)"

for all real non-negative values of x.


Let us take C in Schlafli's formula for L (x) to be the unit circle in
"
the (-plane surrounding' x. In other words let us set (
x+e;o.
=

Ln(x)

exn!
("e-
d(
2ni c (,-x)"+ 1

J.
J

exn! " (x+cos O+i sin O)"e-;"9


dO.
.
- n- exp( x+cos (}+i sm 0)
_,,

166

[Ch. 9

DIFFERENTIAL EQUATIONS

Thus taking moduli we find that

"
(x2+2x cos()+1) 12
d()
ILn(x)I 2n -,,
exp(x +cos())
exn !f"

"
exn ! (x2+2x+1) 12

-.

exp(x-1)

2n

.2n

en!(l +xf.

Similar results can be obtained for Legendre and Hermite polynomials


(see Exercise 9, Questions 3 and 4).

6. Show that the Hermite polynomials satisfy the following


orthogonality and normalization conditions in ( - oo,oo):

Example

where

c5mn

0, .
1,

m '"" n

m = n.

The orthogonality condition follows at once from the differential


equations

{exp(-x2)H(x)}'+2n exp(-x2)Hn(x)
{exp(-x2)H;,.(x)}'+2m exp(-x2)Hm(x)
Ifwe multiply equation

(1)

0.

(2 )

(1) by Hm(x) and equation (2) by Hix) subtraction

yields:

(m-n)exp(-x2)Hm(x)H"(x)
=

Hm(x){exp(-x2)H(x)}'-H"(x){exp(-x2)H;,.(x)}'.

Thus, integrating both sides with respect to

(m-n)

J00

we obtain:

exp(-x2)Hm(x)H"(x) dx
=

[(Hm(x)H(x)-H;,.(x)Hn(x))exp(-x2)]

00

where we have integrated once by parts and noted that the unintegrated
term vanishes identically. Hence we may conclude that

J00

exp(-x2)Hm(x)H"(x)dx

0,

m '" n.

Sec.

9.3]

167

THE FORMULAE OF RODRIGUE AND SCHLAFLI

The normalization property can be simply obtained by means of the

oo
J

oo
exp(-x2)H;(x)dx= J
f

Rodrigue formula

oo

(9.3.5):

(- l tHn(x)

-oo

In obtaining formula

(3)

dn {
n

dx

exp ( - x2)} dx

exp(-x2)H">(x) dx.

(3)

times. Now by

00

we hae integrated by parts

making use of the fact that the leading coefficient of


that

H">(x)

2"n!.

Hn(x)

is

2"

we see

Thus we may write down the final result

exp( -x2)H;(x) dx

n
2 n!

exp( -x2)dx
00

(n)2"n!

where we have used the result of Example

of Chapter 8. Similar results

may be obtained for Legendre and Laguerre polynomials (see Exercise


Questions

9,

and 6).

The orthogonal polynomials are exceedingly useful as a basis for the


expansion of various functions. The technique is widespread both in
theoretical and in practical applications.

Example 7. Determine the expansion coefficients for an arbitrary func


tion f (x) expanded as a series of Hermite polynomials.
Let us write

f(x)
where the

a,

00

L a,H,(x)

are the required coefficients.

Now multiplying both sides of equation

ooJ oo

(1)

r= 0

integrating over the range


exp(

(- oo,oo)

x2)Hn(x)f(x)dx

ooJ oo

(1)

by

we obtain

00

00

a,

exp(-x2)H (x)
"

and

exp (-x2)Hn(x)H,(x )dx

L a,(n)2"n!()rn
r=O

where we have assumed that it is permissible for us to integrate the


infinite series term by term.

168

DIFFERENTIAL EQUATIONS

Therefore we conclude that

1
,J(n)2nn !

oo

exp(-x2)Hix)

00

[Ch.

f (x) dx.

Example 8. Show that, for x real with lxl < 1 and ltl < 1, the Legendre
polynomial of degree n is the coefficient of tn in the power series expansion
of (1-2tx+t2)-t.
00

L P,(x)t'

Let us evaluate the sum

r=O

00

00

L P,(x)t'

r=O
Let us set'=

r=O

x+(x2-1)1'2ei8

2r+ l 1tl.

f ((2-1)' d(.

in formula

((-Xy+ l

(1)

(9.3.4):
(1)

to obtain

00

00

r O

t'

using the Schliifli formula

P,(x)t'

f" {x+(x2-1)112 ey de
t'
{x+(x2-l)112cose}'de
f"
1 "
f I;t'{x+(x2-1)1'2cose}'de

t'
rO 2n

cos

-n

00

r o 0
00

1t

0 r=0

(2)

where we have assumed that we can in fact interchange the summation


and integration operations. Now the geometric series converges provided

(3)
We are given that
that

lxl+lx2-11112

ltl

<

and it is clear from graphical considerations

does not exceed unity for

lxl

<

1.

We may now perform both sum and integral in equation


the final result :
00

r-0

P,(x)t'

"

o 1 -tx-t(x2-1)112 cos 8

(2) to obtain

de

(l-2xt+t2)112
Similar results can be obtained for Hermite and Laguerre polynomials
(see Exercise

9, Question 9).

Sec. 9.4]

CONTOUR INTEGRAL SOLUTIONS

169

The function (1-2xt + t2)- t is called a generatingfunction for Legendre


polynomials: the reason for the nomenclature is self-evident (see Example

8). The main importance of generating functions is in the derivation of


recurrence formulae which are useful both in theoretical manipulations
and in problems of a numerical nature. Examples of this application are
given in Exercise 9, Questions 7, 8,

9.4

10, 11, and 12.

Contour Integral Solutions

The Schlli.fli formulae obtained in 9.3 as the polynomial solution


functions of the Legendre, Hermite, and Laguerre equations in the case
when n is a positive integer can be used as a basis for finding both in
dependent solutions of these equations even when n is not an integer.
The matter is perhaps a trifle academic but the reader is invited to
study the following example.

Example 9. Show that both solution functions of Legendre's equation


can be written in the form

m(n,z)

l
J

((2-1)"

2 n +1 7ti c ((-z)"+1

d(

where C is a contour to be specified and n is an arbitrary real number.


Since n is an arbitrary real number we must be careful to specify

((2-l)"/((-z)"+1. Let us regard


((2-l)"/((-z)"+1 as a many-sheetedfunction and let us start at a specific
point of a specific branch and describe the contour C: on closing up
our contour our function ((2-l)"/((-z)"+1 may or may not return to
exactly what we mean by the expression

its original value depending on how we have threaded our way round
the various branch points. With this understanding let us substitute the
given integral into Legendre's equation and try to choose C in order that
it is satisfied. After some elementary manipulation in which we assume
that we may differentiate with respect to
we obtain:

where A.=
12

((2-1)"+1/((-z)"+2.

z under the sign of integration,

170

[Ch. 9

DIFFERENTIAL EQUATIONS

Thus we must choose C in order that


[(n+l)Arg((2-l)]c

[(n+2)Arg((-z)]c+2Nn

(1)

where N is any integer. If we do this A. will return to its original value on


completing a circuit of C.

-I

FIG. 9.3

FIG. 9.4

If we choose the contour C 1 drawn in Fig. 9.3 we note that


[Arg((+l)]c

0,

[Arg((- l)]c

[Arg((-z)]c

21t.

(2)

Again if we choose the contour C2 drawn in Fig. 9.4 we obtain


[Arg((+ l)]c

- [Arg((- l)]c

2n,

[Arg((- z)]c

(3)

In either case equation (1) is satisfied and since C 1 and C2 are not
deformable one into the other, the two solutions must be independent.
No other choice of contour yields any further independent solution. In
the case of n a positive integer, the solution corresponding to C 1 is Pn(z).

S-ar putea să vă placă și