You are on page 1of 455

<

G-^
AN INTRODUCTION TO THE

THEORY OF FUNCTIONS
OF A

COMPLEX VARIABLE

AN
INTRODUCTION TO THE

THEORY OF FUNCTIONS
OF A

COMPLEX VARIABLE
BY

COPSON

E. T.

M.A. (OXON.), D.Sc. (EDIN.)


Professor of Mathematics at University
College, Dundee, in the University of
St.

Andrews

OXFORD
AT THE CLARENDON PRESS
1935

OXFORD
UNIVERSITY PRESS
AMEN HOUSE, E.C. 4
London Edinburgh Glasgow
New York Toronto Melbourne
Capetown Bombay Calcutta
Madras Shanghai

HUMPHREY MIIPORD
PUBLISHER TO THE
UNIVERSITY

FEINTED IN GREAT BEITAIN

PREFACE
This book

is

based on courses of lectures given to undergraduates

in the Universities of

Edinburgh and St. Andrews, and is intended to provide an easy introduction to the methods of the
theory of functions of a complex variable. The reader is assumed
to have a knowledge of the elements of the theory of functions
of a real variable, such as is contained, for example, in Hardy's
Course of Pure Mathematics an acquaintance with the easier
parts of Bromwich's Infinite Series would prove advantageous,
but is not essential.
;

The first six chapters contain an exposition, based on Cauchy's


Theorem, of the properties of one-valued differentiable functions
of a complex variable. In the rest of the book the problem of
conformal representation, the elements of the theory of integral
functions and the behaviour of some of the special functions of

by the methods developed in the earlier


The book concludes with the classical proof of Picard's

analysis are discussed


part.

Theorem.
No attempt has been made to give the book an encyclopaedic character. My object has been to interest the reader
and to encourage him to study further some of the more
advanced parts of the subject suggestions for further reading
have been made at the end of each chapter.
I am especially indebted to Mr. W. L. Ferrar, who read the
manuscript of the whole book in its original and revised forms,
and suggested many improvements. My grateful thanks are
also due to Professor E. T. Whittaker, F.R.S., for his kindly
;

criticism during the early stages of the preparation of this


and for his constant encouragement.

work

have to thank Dr. H. S. Ruse and Professor J. M.


Whittaker for their careful reading of the proof sheets and many
Finally, I

valuable suggestions.

E. T. C.

GREENWICH,
July 1935

CONTENTS
I.

COMPLEX NUMBERS

....

III.

THE CONVERGENCE OP INFINITE SERIES


FUNCTIONS OF A COMPLEX VARIABLE

IV.

CAUCHY'S THEOREM

II.

V.

VI.
VII.
VIII.

IX.

X.
XI.
XII.
XIII.

13

32

....
....
....
....
....

52

UNIFORM CONVERGENCE
THE CALCULUS OF RESIDUES

117

INTEGRAL FUNCTIONS

158

CONFORMAL REPRESENTATION
THE GAMMA FUNCTION

92

180

205

THE HYPERGEOMETRIC FUNCTIONS


LEGENDRE FUNCTIONS

233

BESSEL FUNCTIONS

313

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

XIV. JACOBI'S ELLIPTIC FUNCTIONS

......
......

MODULAR
THEOREM
INDEX

XV. ELLIPTIC

FUNCTIONS

AND

272

345
380

PICARD'S
421

444

CHAPTER

COMPLEX NUMBERS
The introduction

of complex numbers into algebra


In arithmetic, we understand by a real number a magnitude
which can be expressed as a decimal fraction. If the decimal
1.1.

terminates or recurs, the real number is said to be rational, since


then the ratio of two whole numbers. But if the decimal
does not terminate or recur, the number is not the ratio of two

it is

whole numbers and is said to be irrational. We shall assume


that the reader is acquainted with Dedekind's method of founding the theory of rational and irrational numbers on a sound
logical basis.f

Elementary algebra is concerned with the application of the


operations of arithmetic to symbols representing real numbers.
The result of any sequence of such operations is always a real
number.

difficulty,

tions.

If a,

ax2 +2bx-{-c
roots

which

when

=
b2

however, soon arises in the theory of equac are real numbers, the quadratic equation
has two distinct roots if b 2
ca and two equal

and

6,

= ca.

>

But

if b 2

< ca,

there

is

no

real

number x

the equation, since the square of every real


number is positive. It is customary to introduce a new symbol
V( 1), whose square is defined to be 1, and then to show
that the equation is formally satisfied by taking
satisfies

ax

when

b2

< ca.

-bJ(ac-b2 )*J(-l),

When this new symbol has been added to the algebra of real
numbers, every quadratic equation is formally satisfied by two
expressions of the form a+^( 1), where a and are real
numbers. Such expressions are called complex numbers.
If we suppose that the symbol
<J( 1) obeys all the laws of
is 1, we can develop a consistent
algebra of complex numbers which includes the algebra of real
numbers as a particular case and which also possesses a character of completeness which is lacking in the simpler theory. This

algebra, save that its square

t See, for example, Chapter I of Hardy's Pure Mathematics (1921).


4111

_.

COMPLEX NUMBERS

well illustrated by the theorem which states


completeness
of degree n has precisely n roots a result
equation
that every
of complex numbers but is false in
algebra
which is true in the
is

the algebra of real numbers.


The present book is concerned essentially with the application
of the methods of the differential and integral calculus to complex numbers.

An algebra of ordered

1.2.

Before

we proceed

pairs of real

numbers

to elaborate the algebra of complex

num-

very desirable that we should provide a definition of

bers, it is

such numbers which depends only on real numbers, instead


of the formal introduction of the symbol A/( 1) of elementary

Such a

algebra.

definition is suggested

by the

following geo-

metrical considerations.
In the analytical geometry of the Euclidean plane, the equaay define a transformation of any conax, y'
tionsf x'

figuration of points into a

new

configuration which

is

obtained

by a very simple geometrical construction. When a is positive,


the transformation magnifies all distances from the origin in the
whilst when a is negative, there is a magnification in
ratio a
:

1,

the ratio

a:

together with a rotation about the origin

through two right angles.


If

we

(x',y')

write the equations of transformation in the form


(x,y)a, we see that a real number is the symbol of

a certain geometrical transformation.


These transformations are, however, merely particular cases
bx+ay,
ax by, y'
of the more general transformation x'
transformation
This
a
and
6.
parameters
real
depending on two
2
magnifies all distances from the origin in the ratio <](a +b?) 1
1
and rotates all rays from the origin through an angle tan- (&/)
The two equations defining the transformation may be com-

bined by writingj

(*,*)
t
J

(*,v)(_l

*);

The axes of coordinates are supposed to be at right angles.


The notation is that of the algebra of matrices. The reader

against confusing the matrix

("'

matrix has no numerical value, but

is

*) with the determinant

is

|_'

merely an array of numbers.

warned

J.

COMPLEX NUMBERS

here the multipliers for x' occur in the first column of the
tworowed symbol, those for y' in the second. We shall now construct an algebra of these two-rowed symbols and
shall show
that this provides the desired definition of a complex
number.
In the first place, we say that two symbols are equal if the
corresponding transformations are the same; thus

b\/

a,

\b,

a)

a',

b'\

\b',

aj

holds if and only if a


a' and b
b'.
Secondly,
the sum of the two symbols by the equation
a,

From

b\

a',

aj^\-b',

\-b,

b'\(
aj

the definition of addition


a',

\-b',

we

accordingly

\-b,

b-b'\ __

a',

\-b',

a,

a-a'j~[-b,

b\

a)'

by

define subtraction

aj

a+a'f

follows that

it

a-a',

define

b+b'\

\-b~b',

a')^\~b+b',

b\l

a,

b'\

a+a',

we

b'\(
aj

a-a',

\-b+b',

b-b'\

a-aj'

The definition of the product of two symbols is not quite so


obvious, but it is suggested by analogy with the simpler case
in which the transformation (x', y')
(x, y) ah is the result of

applying successively the two transformations (x', y')


(x",y")b
and (x", y")
(x, y) a. Accordingly, we define the product
of the

Symb0lS
I

a,

\-b,

b\

a',

b'\

\-b',

aj

a)'

to be the symbol of the transformation obtained


successively the transformations

w.m = (_"'; ^:).

<*-,,-)

by applying

<*,-,)(_

).

'

It follows thatf
/

a,

\-b,

b\l

a',

a]\-b',

b'\l
aj

aa'bb',

\-ab'-a'b,

ab'+a'b\

aa'-bbj'

t It should be observed that the rule for multiplication is the same as that
for multiplying two determinants ; elements of a row of the first
multiplied

by the corresponding elements

of

a column

symbol are
in the second symbol.

COMPLEX NUMBERS
L'

The symbol

Qj

in this
called the zero symbol, for it plays the part of zero
product
algebra, since the difference of two equal symbols or the

is

of

any symbol and the zero symbol

Similarly

we

call

/j

q\

ij'

is

always the zero symbol.

which corresponds to the identical transformation


{x',y')

(x,y),

multiplicathe unit symbol, since every symbol is unaltered by


symbol.
tion by the unit
Finally, given any non-zero symbol

we can deduce from


namely

3-

the law of multiplication a unique symbol,

a/ (a2 +ft 2 );

\6/(a

+6

2
),

_&/( a2 +6 2)\
a/(a2 +& 2 )/'

whose product with the given symbol is the unit symbol. The
second symbol is said to be the reciprocal of the first. We now
by
define division by a non-zero symbol to be multiplication
its reciprocal.

Having defined the operations of addition and multiplication,


together with the inverse operations of subtraction and division,
we must next consider whether these operations obey the commutative, associative, and distributive laws of algebra.f The
reader will easily verify that this is the case by showing that,
ifA,B,C denote any three of these two-rowed symbols, then

A+B = B+A,
/
AB = BA,
=
A+(B+C),
(A+B)+C
(AB)C = A(BC),
A(B+C) = AB+AC.
Moreover,

if

^^s.
**

AB is identical with the zero symbol,


Algebra, 1 (1910), 2-24.
f See G. Chrystal,

so also

is

at

COMPLEX NUMBERS
least

one of

and B.

this algebra of

It follows, therefore, that operations in

two-rowed symbols are carried out in precisely

the same manner as in ordinary algebra.

In particular, the algebra of symbols of the form

'

is

\0, a)

numbers a. Even though the


two-rowed symbols of this type are logically distinct from the
real numbers, no useful purpose is served by keeping a special
notation for them.| With this convention, we may write
identical with that of the real

a,

b\

\~b,

a)

a-\-ib,

where

We

easily see,

by means of the

rule for multiplication, that

-j-ty

o,

jr.*

has all the properties formally assigned to


*J(1) in elementary algebra.
We now define a complex number to be a two-rowed symbol
in other words,

of the form

or a-{-ib, which obeys the laws of combina-

'

tion prescribed above.

It follows that

we can perform

all the

operations of algebra with complex numbers in exactly the same


way as with real numbers, provided that we treat the symbol i as

a number and replace


1.3.

its

square by

The modulus and argument

whenever

of a

it

occurs.

complex number

The geometrical transformation


(x',y')

associated with the complex

(x

4-1:

number

tances from the origin in the ratio


all

)
a-\-ib

magnifies

-\-*J{a?-\-b

%
)

rays from the origin through a certain angle

1,

a.

all dis-

and rotates
The magni-

f It will be recalled that a similar point arises in connexion with Dedekind's


theory of the real numbers, -where the real rational number x (defined by a
Dedekind section of the rational numbers) is logically distinct from the corresponding rational number. See Hardy, Pure Mathematics (1921), 14.

COMPLEX NUMBERS

fication factor +*J(a

+b 2

is

called the

modulus of the complex

number a-\-ib and is denoted by \a-\-ib\.


The number a which is determined by the two equations

a=

\a-\-ib\

b=

cos a,

|a+i6|sina,

called the argument^ of a+ib, and is written arg(a+i&). If


the modulus of a complex number a-\-ib is not zero, its argument has an infinite number of values; for if a is a value of

is

a+2w7r, n being any integer. The principal


value of arg(a+#>) is defined to be that which satisfies the

arg(a+i&), so also

is

_w <

inequality

aig{a+ib)

ff .

If, however, |a+i6| is zero, then a and b are both zero, and
the equations to determine a are satisfied identically; the argument of the complex number is thus indeterminate.
Let us now suppose that the complex number A-\-iB is the
product of a -\- ib and a'-j-ib'. Since the geometrical transforma-

tion corresponding to

A+iB

is

the result of applying succes-

sively the transformations corresponding to a-\-ib


it

and

a'-\-ib',

follows at once that

\A+iB\
and

\a+ib\.\a'+ib'\,

also that a value of &Tg(A-\-iB)

is

values of the arguments of a+ ib and

the

sum

of the principal

a'-\-ib'.

and imaginary parts of a complex number


It is convenient to denote a complex number x-\- iy by a single
letter, z say. If z = x-j-iy, where x and y are real, we call x and
1 .31

The

real

y the real and imaginary parts of z, and frequently write


x
It

is

= Imz.

easily seen that

From
1.2, it

if

= Rlz,

\z\

< Rlz <

\z\,

\z\

< Imz <

\z\.

the definition of the equality of two of the symbols of


follows that

and only

if

for addition,

Rl(z+z')

Rlz

we

z and z' are equal


Again, from the rule

two complex numbers

= Rlz', Imz = Imz'.

see that

= Rlz + Rlz',

Im(z+z')

= Imz + Imz'.

f Alternatively, the amplitude of a-{-ib.

COMPLEX NUMBERS
Conjugate complex numbers

1.32.

= x-\-iy, where x and y are real, the number x iy is said

If z

to be conjugate to z

conjugate to z
z1

+z 2 and

and

is

is z itself.

z x z 2 are

denoted by z. Obviously, the number


Moreover, the numbers conjugate to

evidently z x -\-z 2 and z x z 2 respectively.

The proofs of many theorems regarding complex numbers can


often be greatly simplified by the use of conjugate complex
numbers, if we remember the easily-proved formulae
|z|

= zz,

First of all,

= z+z,

2ilmz

= z

z.

if z x and z 2 are any two complex numbers, we have

Pl Z 2l

and

2Rlz

=Z

so

= ^l^l Z
=

Z 2 Z1 Z2
\zi z v\

|z x |. |z 2

Z 2 == Pi

z 2\

>

|,

the positive square root being taken since the modulus of a complex number is never negative. Therefore the modulus of the
product of two complex numbers (and hence, by induction, of any
number of complex numbers) is equal to the product of their
moduli, f

Again,

we have

+ 2 =
=
=
<
2

|Zl

=
and sot
Hence

z
l

the

modulus of

by

induction, of any

the

sum

On

(Zl+Z 2 )(Zi

+Z

2)

z 1 z 1 -)-z 1 Z2 _r z i^2 _r22 z 2


|z i

P
2

l2 1

+ 2Rl(z
+21z 1 z 2

(P1I+P2I)

z 2 )+|z 2 |2

|z

2
,

< Kl+tali+
sum of two complex numbers
z 2l

the

(and

so,

number of complex numbers) cannot exceed

of their moduli.

we

the other hand,

Pi-z 2

2
l

=
>
=

also
|z 1
| 1

have
2

-2Rl(z 1 z 2 )+|z 2 |2

|-2|z 1 g a |+|

(pll-p 2

2
l)

f An alternative proof was given in 1.3.


j It should be observed that the sign of equality occurs only when z-^%^ is
real and positive; this is the case if and only if z x and z 2 have the same

argument.

COMPLEX NUMBEKS

so that

\z

2 >
2

Kl^xl

|z 2 |)|,

an important inequality of which frequent use will be made.


Example. Show that argz -f- argz = Imr, where n is an integer

or

zero.

1 .4.

is

The geometrical representation of complex numbers

The usual method of representing real numbers geometrically


by means of points on a line, the real number x corresponding

to the point of abscissa x, according to some given scale, referred


to a fixed origin on the line. In a similar way,

we

shall represent

a complex number 2 by a point in a plane whose rectangular


Cartesian coordinates are (Rlz, Imz); the complex number z is
then called the affix of the point which represents it. The
straight line whose equation is Im z
is called the real axis,
since the affix of each point on it is a real number; similarly we
call the line Rlz
the imaginary axis.
This geometrical picture, which is often called the Argand
diagram, enables us to describe properties of complex numbers
by means of geometrical language, as all the operations of the
algebra of complex numbers have simple geometrical interpretations in this scheme.

Example
show that

If P, Q,
are the points of affix z,z',z+z' respectively,
is a parallelogram.

OPRQ

Example 2. Prove that the polar coordinates of the point of affix z


are (|z|,argz), referred to the origin as pole and the real axis as initial
line. Deduce a geometrical construction for the point of affix zz'.
Example

3. Prove that

\a b\ 2 + \a+b\ 2

2|a| 2 +2|6| 2 .

Interpret

this result geometrically.

1.41.

The point

at infinity

In the Euclidean geometry of the plane, 'points at infin ity'


do not occur; two straight lines intersect in a point except in

when the

the case

lines are parallel.

It

however, usual to

is,

postulate, at a later stage, that there exists

an

infinite

number

of points at infinity, each being defined as the point of intersection of a pencil of parallel lines.

Now,
c in the

if

we take a

Argand

pencil of curves through the point of affix

plane, the transformation

z'

1/z turns

into a pencil of curves through the point of affix

that c

is

them

1/c, provided
not zero. To avoid the difficulty of this exceptional

COMPLEX NUMBEBS
case,

we now

postulate that there

is

single point at infinity in

the Argand plane; this point at infinity is defined to be the point


corresponding to the origin in the transformation 2'
1/2.

The nature of the Argand plane at the point at infinity is


made much clearer by the use of Riemann's spherical representation of complex numbers, which depends on stereographic
projection.

In order to map a spherical surface stereographically on a


we take a point on the sphere as vertex of projection and

plane,

equatorial plane as plane of projection. Then to any point


of the sphere save the vertex of projection, there corresponds

its

a unique point of the plane; conversely, to each point of the


plane, there corresponds a unique point of the sphere.

we

P+^

2
project the points on the sphere
1 stereo+C 2
graphically on the plane
0, taking the point (0, 0, 1) as
vertex of projection, we find that the point (x, y, 0) of the plane

If

corresponds to

on the sphere

(, tj, )

= f/(l-),

if

= VU-a

or, writing z for x-\-iy,

Conversely,
affix z in

(,

t\,

the plane
j.

tf+i?)/(i-a

on the sphere corresponds

to the point of

if

2z

22+1

221
22-j-l

These equations provide a continuous one-to-one correspondence between the complex numbers and points on a spherical
surface.

Now

if (,

7],

corresponds to

corresponds to the point of affix

z',

where
.

since
\z\

\z'\

2
2
+r)

1,

i+iv

W_

z, (,

77,

-^Q

But obviously argz = argz' and


and 2' are inverse points with respect to
with centre at the origin. Thus points of the

1 2

so that z

the unit circle


Argand plane which are inverse with respect to the circle \z
1
correspond to points of the Riemann sphere which are sym\

metrical with respect to the plane

0.

COMPLEX NUMBERS
In particular, when z = 0, z' is, by definition, the point at
infinity, and so the point (0, 0, 1) of the Riemann sphere corre10

sponds to the point at infinity of the Argand plane. We have


thus set up a one-to-one correspondence between the points of
the Argand plane, completed by the addition of the point at
infinity,

and the points of the whole Riemann

sphere.

Any geo-

metrical property involving the point at infinity in the

Argand

plane can thus be easily visualized by means of this spherical


representation of complex numbers.

REFERENCESf
G. H. Habdy, Pure Mathematics^ (Cambridge, 1921), Chap. III.
B. A. W. Russell, Introduction to Mathematical Philosophy (London,
1919), Chap. VII.

MISCELLANEOUS EXAMPLES
,

1.

Show

that the equation of a straight line in the Argand plane


bz

where

and

c are constants, c

+ bz

being

is

c,

Deduce the condition

real.

for the

collinearity of three points.

Prove
'--

2.

also that z' is the reflection of z in this line if bz

+ bz' =

c.

Find the area of the triangle whose vertices are the points of affix

z \> z 2> 2 3*
3.

Show

are similar

4.

that the triangles whose vertices are z 1( z a z 3 and


,

z2

z%

23

Prove that the equation of a

J-

z'z

0.

circle in

azz-\-bz-\-bz-\-c

where a (^

Show

z[, z'

if

the Argand plane

is

0,

and c are constants, a and c being real.


and z' are inverse points with respect to

0), 6

that z

az'z-\-bz-\-bz' -\-c

this circle if

0.

f At the end of each chapter, references will be given to suggest further


reading.
% It is important to notice that, although Hardy's definition of a complex
number is quite different from that adopted here, both depend on constructing
an algebra with certain symbols involving two real numbers arranged in a
certain manner. As the laws of combination are the same for both sets of
symbols, the two constructions lead to the same algebra of complex numbers.
with respect to the degenerate circle
This implies that inversion
Ozz 62+ bz-\- c
is reflection in the line bz-\- bz-\- c
0.

COMPLEX NUMBERS
5.

Show

11

that the equations


arg-

A,

where A and a are variable parameters, represent two orthogonal families


of coaxal
6.

circles.

Prove that the homographic transformation


yz + B'

where

a, , y, are

circles in

complex constants, turns

circles in the z-plane into


the .Z-plane, straight lines being regarded as degenerate circles.

7. Prove that a homographic transformation is a one-to-one transformation of the complete z-plane into the complete Z-plane, which
leaves the angle between any two intersecting curves unaltered. Show
also that the cross -ratio of the affixes of four points
(2 1
(z 1
is

-g 2 )(z 3 -z4
-z3 )(z a 4

)
)

invariant under a homographic transformation.

8. Show that every homographic transformation may be generated


by an even number of inversions with respect to circles or straight lines.
9. Show that there are two points which are invariant with respect
to a homographic transformation. Hence show that, if these invariant
points are distinct, the transformation is expressible in the form

Z ~ Zl

Z z
where k

is

a constant, and that,

the transformation

if

10.

6 is

,
1

a constant.

Determine the regions of the z-plane specified by

1az
where a
11.

~ Zl

zz 2

the invariant points are coincident,

is

Zzx
where

is

<

=1,

1,

a constant of modulus

less

or

>

1,

than unity.

Find the regions of the z-plane for which


z

z+a <
where the

real part of

is

1,

or

1,

>

1,

positive.

12.

Show

13.

Two points, whose affixes are z x and z 2 move independently round


with a focus at the origin. Show that the region in which the

that the equation {az+azf


2(6z+5z) + c, where
represents the most general parabola in the complex plane.

an

ellipse

c is real,

COMPLEX NUMBERS

12

point of affix zjz 2

lies is

bounded by a curve whose polar equation

(l+r )(l-e
2

2
)

2r(l-e

cos0)

is

0,

the eccentricity of the ellipse. Sketch this curve and indicate


the region in which zjz 2 must lie. What is the corresponding result for
[Oxford: Junior Scholarship, 1929.]
a parabola?

where

e is

Prove that, in stereographic projection, straight lines in the plane


correspond to small circles on the sphere which pass through the vertex
14.

of projection.
Show also that the great circles through the ends of a fixed diameter
of a sphere project into a family of coaxal circles with real common
points, and that the small circles whose planes are perpendicular to this
diameter project into the orthogonal system of coaxal circles.
15.

If da be the element of length of a curve on the sphere


i an(i ds the element of length of its stereographic pro-

|2_|_,j2_^2

jection

on the plane

0,

prove that

da

Hence show that stereographic projection maps a sphere conformallyf


on a plane.
16. Show that the points z x z 2 are the stereographic projections of
1.
the ends of a diameter of the Biemann sphere if z x z 2
about this
If the Riemann sphere be rotated through an angle
diameter, prove that the corresponding transformation of the Argand
plane is given by
,

Z
Z

^=
1

Z 2

(cos6+iamd)

Z=^.
Z

Zz

sufficient condition that a transto a rotation of the Riemann


correspond
plane
Argand
the
formation of
sphere about a diameter is that it be of the form
17.

Prove that the necessary and

cz+a
where the constants

a, c are

connected by the relation aa+cc

1.

rotation of the Riemann sphere about a diameter through the


18.
point of spherical polar coordinates (l,6 ,(j> ) moves the point (\,6,<f>)
to (\,0',<f>'). Prove thatt
cot O cosec a cot \Q e^-^"' i

cotp

*.

_
-

cot

cosec a+i cot i0 6*(*-*

+)

Determine the relation between the angle of rotation and the parameter a.

method of mapping is said to be conformed if it preserves the angle of


f
intersection of every pair of intersecting curves. See Chapter VI.
See 3.51.
% e** is here used as a convenient abbreviation for cos <ji-\- i sin <ji.

CHAPTER

II

THE CONVERGENCE OF INFINITE SERIES


Argand plane
chapter, we can describe

2.1. Sets of points in the

As we saw

in the last

properties of

complex numbers in geometrical language by using the Argand


diagram or the Riemann sphere. In the present chapter, we
consider from this geometrical standpoint some of the properties
of sets of complex numbers.
By a neighbourhood of a point z in the Argand plane we mean

<

e; we call e the radius


the set of all points z such that \z z
1/z, the
of this neighbourhood. Under the transformation z'
neighbourhood of the origin of radius e becomes the set of all
\

points z for which \z\ >


bourhood of the point at

we define a neighbe the part of the z-plane


outside a circle )z = B. It should be observed that the portion
of the Riemann sphere which corresponds to this neighbourhood
Accordingly,

1/e.

infinity to

of the point at infinity

by the plane

A point z

is

is

the interior of the small circle cut off


i3.ii.s-qh

{B1)/(R*~+1).

said to be a limiting point of a set of points

the Argand plane

S in

every neighbourhood of z contains a point


For example, the points l+i are
of 8 distinct from z
limiting points of the set of points of affix
if

(- 1 )" + ^XT
whilst the point at infinity

of -affix

n2 +2in

is

(=

1.2.3,...),

a limiting point of the set of points


(n

1, 2, 3,...).

This definition implies that every neighbourhood of a limiting


point z of a set of points S contains an infinite number of points
e contains a point z x of
of S. For the neighbourhood [z
|

distinct

from

contains a point z 2 of

2.11. Closed

The
the

<

next, the neighbourhood \z z

distinct

and open

from

and

so

on

<

\z x

indefinitely.

sets of points

limiting points of a set are not necessarily points of

set.

For example, the limiting points

l+i

of the set

THE CONVERGENCE OF INFINITE SERIES

14

of points of affix

(- 1 )" +

^1

= !> 2,

3,...)

members of the set. If, however, every limiting


set, we say that the set is closed.
Limiting points are divided into two classes, interior points
and boundary points. A limiting point z of a set S is said to be
an interior point if there exists a neighbourhood of z which
are certainly not

point of the set belongs to the

limiting point which is not


consists entirely of points of $.f
an interior point is called a boundary point. Thus, if S consists

of all points for which \z\ < 1, points on the circle \z\ = 1 are
boundary points, whilst each point of the set is an interior point.
A set which consists entirely of interior points is said to be
open. It should be observed that sets exist which are neither
open nor closed; a simple set of this type consists of the point
z = 1 and all points for which \z\ < 1.

Example. Show that, if a set <S" consists of all the limiting points
of a given set S, then a limiting point of S' necessarily belongs to S'.
2.12.

Jordan curves

The equation
x(t) and

where

variable

t,

z
y(t)

x(t)-\-iy(t),

are real continuous functions of the real

denned in the range t ^ t ^ T, determines a set of


Argand plane which we call a continuous arc.

points in the

A point Zj is said to be a multiple point of the arc if the equation

= x(t)-\-iy(t)

is satisfied by more than one value of t in the


given range.
A continuous arc without multiple point is called a Jordan
arc. A simple example of a Jordan arc is the polygonal arc
which consists of a finite chain of straight segments.
A continuous arc which has but one multiple point, a double
point corresponding to the terminal values tQ T of t, is called
a simple closed Jordan curve. For example, the arc

zx

where

^ ^
t

point J being z

cost+isint,

a simple closed Jordan curve, the double


corresponding to t
and t
2tt.

2-n, is

1,

f This, of course, implies that z belongs to S.


J It should, however, be observed that it is not
understood in the theory of higher plane curves.

a double point in the sense

2.13.

THE CONVERGENCE OF INFINITE SERIES


Bounded sets

15

A set of points is said to be bounded if there exists a positive


number

with the property that the inequality

\z\

^K

by the affix z of each point of the set. If there


no such number K, the set is said to be unbounded.
satisfied

2.14.

The

definition of a

set of points in the

every pair of

its

is

exists

domain

Argand plane

is

points can be joined

said to be connex if

by a polygonal

arc

which consists only of points of the set. An open connex set


of points is called an open domain. If we add to an open
domain its boundary points, the resulting set is called a closed
domain.

Two open domains which have no

point in

common

are said

to be separated. Obviously every polygonal arc which joins


a point of an open domain to a point of another open domain
separated from the first must contain boundary points of the
sets.

2.15.

The Jordan curve theorem

shown that the circle \z\


1 divides the Argand
plane into two separated open domains, namely the sets defined
1 and \z\
by the inequalities \z\
1, which have the circle
as common boundary. This result is a particular case of the
Jordan curve theorem, which states that a simple closed Jordan
curve divides the plane into two open domains which have the curve
as common boundary. One of these domains is bounded and is
called the interior domain; the other is unbounded and is called
It is easily

<

>

)"

the exterior domain.

The two domains

into which the plane

closed Jordan curve

If a

property.
returns to

scribes

are distinguished

is divided by a simple
by a simple analytical

a point of the exterior domain, arg(z a)


value when z goes once round C. On the

its original

other hand,
increases

is

by

C in

if

b is a point of the interior domain,

277.

If the increase

is -\-2tt,

we say

arg(z b)
that z de-

the positive or counter-clockwise sense.

Although the

results

we have

just stated

seem quite obvious,

t Jordan's original proof will be found in his Cours d" Analyse,


1887), 587. He assumed the truth of the theorem for a polygon.

(Paris,

THE CONVERGENCE OF INFINITE SERIES

16

their proof is extremely complicated

and

far too difficult to give

here.f For the most part, we shall only use simple closed Jordan
curves composed of straight lines and circular arcs, and we shall
rely upon geometrical intuition, which does not, in fact, lead us
astray. Actually the proofs of the theorem in these elementary
cases are generally quite straightforward.

Nests of intervals and rectangles


The set of all real numbers x such that a < x < b is called
an interval. If we represent the real numbers by points on a

2.2.

straight line in the usual manner, this interval

is

represented

by a segment of the line of length ba, the end-points being


included in the segment. It is convenient to use the same word,
interval, for such a set of real numbers and for the corresponding
segment on a

line.

sequence^ of intervals 71; J2 ,..., /,... is said to form a nes1%


if each interval In+1 consists only of points of In and if the
length of In tends to zero as n -> oo. We shall now show that
there is one and only one point which belongs to all the intervals
of a nest.

For suppose that the interval In consists of all points x


Then, by the definition of a nest,
bn
x
such that an
6 X for all values of n, and so the numbers a n form
an+1
an
a non-decreasing bounded sequence. Hence, as n -> oo, an tends
an+p for every positive
to a finite limit . Moreover, since a n
->
oo,
that
an
making
. Similarly we
we
see,
by
integer p,
p
$', for every
',
and
that
b
limit
to
a
can show that bn tends
n

< <

<

<

<

<

>

finite

value of n.

an tends to zero by hypothesis, an and b n must


tend to the same limit |. We have thus shown that there exists
But

since b n

a number such that the inequality an

< < b n holds for every

The simplest proof which has so far been published seems to be that of
f
Winternitz, Math. Zeitschrift, 1 (1918), 329-40, who gives full references to
previously published work. See also P. Dienes, The Taylor Series (Oxford,
1931), 177-97; G. N. Watson, Complex Integration and Cauchy's Theorem
(Cambridge, 1914), 3-16.
throughout this book, a set of infinitely many
J By a sequence we mean,
objects which can be put into one-to-one correspondence with the set of
positive integers.
This term is suggested in R. C. Young's translation
and Application of Infinite Series (London, 1928).

of

Knopp, Theory

THE CONVERGENCE OF INFINITE SERIES

17

value of n. In other words, the point of coordinate


in every interval of the nest.f
finite

A similar theorem
a geometrical form.

lies

true of complex numbers;

we state it in
of closed rectangles Sv i? ,...,
2
n ,..., whose sides are parallel to the real and imaginary axes,
is called a nest if each rectangle
n+1 consists only of points of
n and if also the lengths of the sides of
n tend to zero as
n -> oo. Then, as we shall now show, there is one and only one
point which lies in all the rectangles of the nest.
Let us suppose that the rectangle
n consists of all points of
affix z such that
is

A sequence

an

< Rlz < b n

a'n

< Imz < b'n

Then, by definition, we have

< +1 < K+x < K>


< < a'n+1 < &;+1 < b'n
a'
where b n n and b'n n both tend to zero as n -+ oo.
an

These conditions, however, imply that the intervals

an

form a

nest.

< x < bn

Hence there
is

1, 2, 3,...)

exactly one point $ on the real axis


b n holds for all values of n.
f
precisely one point i-q on the imaginary axis
is

such that the inequality an


Similarly there

(n

< <

< <

such that a'n


b'
-q
n for all positive integers n. It follows
that the point of affix i+i-q is the one and only point which
lies in all the rectangles of the nest.

Example. A x Bt C is a triangle, which is divided into four congruent


by joining the mid-points of its sides. A 2 B 3 C is one of these
2

triangles

smaller triangles, chosen according to some definite rule. A B C


3
3
3 is
obtained from A 2 B 2 G3 in the same manner, and so on indefinitely.
Prove that there is one and only one point which lies within or on all
the triangles of this sequence.

The Bolzano -Weierstrass theorem


The fundamental property of bounded sets of points is contained in the theorem of Bolzano and Weierstrass, which runs
2.21.

t It is necessary for the truth of the theorem that the end-points of an


interval should belong to the interval. For example, if J denotes the
set of
n
points such that
< x < ljn, there is no point which belongs to all the sets
Jl>

J>

4111

Jfif

18

as

THE CONVERGENCE OF INFINITE SERIES


follows. // a set S is bounded and contains infinitely many

points, then

it

possesses at least one limiting pointy

hypothesis, there exists a closed square


KiK, to which every point of the set

By

R
S

with vertices
belongs.

The

imaginary axis divides this square into two


One of these, regarded as a closed set, must contain an infinite
number of points of S; if both do, we take the right-hand one.
equal rectangles.

Again, the real axis divides the selected rectangle into two congruent squares. One of these squares, regarded as a closed set,
contains an infinite number of points of 8; if both do, we take
the upper square.
obtain a closed square R v of side K, which
an infinite number of
is contained in B and which contains
way. We divide B x
same
in
the
B
points of 8. We now treat
x
parallel
to the imaginary
line
a
by
into two congruent rectangles
set, contains
closed
a
as
regarded
axis. One of these rectangles,
the righttake
we
both
do,
if
an infinite number of points of 8;
equal
two
into
divided
then
hand one. The selected rectangle is

In

this

squares

way we

line parallel to the real axis.

by a

Again, one of these

squares, regarded as a closed set, contains an infinite


of points of *S; if both do, we take the upper one.

number

We

thus

in
obtain a closed square B 2 of side K/2, which is contained
B x and which contains an infinite number of points of S. The
insquare B 2 is then treated in the same manner, and so on
reach
never
since
we
terminates
never
process
The
definitely.
square which contains only a finite number of points of S.
,

the squares B B^..., Bn ,... form a nest, since Bn+1


of each side of
consists only of points of B n and the length
there
Accordingly
->
n
oo.
n
as
zero
to
Bn+1 is Kj2 which tends
of
squares
the
all
in
lies
which
point

exists one and only one


For
if
of
point
8.
limiting
a
is
the nest; we now show that
integer
an
choose
can
we
e is any assigned positive number,
then the
such that the diagonal of the square BN is less than e;
hence
and
square
B
the
contains
N
neighbourhood of of radius e
the
proves
This
of
8.
points
of
contains an infinite number

Now

theorem.
t

Two proofs of this theorem for the case when all the points lie on a straight
given

The proof
be found in Hardy, Pure Mathematics (1921), 30, 134.
second proof.
here in the two-dimensional case is similar to Hardy's
line will

2.3.

THE CONVERGENCE OP INFINITE SERIES


The convergence of complex sequences

19

A sequence of complex numbers z 1 z 2 ,..., zn ,... is said to be


convergent if the corresponding set of points in the Argand
plane has one and only one limiting point, whose affix z is finite;
,

when

number z is called the limit of the


denote this symbolically by writing 'z -> z as
n

this is the case, the

sequence.

n -> oo',

We

or lim z n

z.

The definition implies that z is the limit of the sequence if


and only if every neighbourhood of the point of affix z contains
all but a finite number of points of affix z
n In other words,
.

z is the limit of the sequence z z ,..., z ,...


if, given any positive
v 2
n
number e, we can find an integer
such that the inequality

n~A < e holds for all integers n~^N.

It follows that a convergent sequence is necessarily bounded.


sequence which is not convergent is said to be divergent.
divergent sequence either has more, than one limiting point,
\

in which case it is said to oscillate, or else has a single limiting


point at infinity.
If the points of the sequence z z ,..., z ,..., having a single
v 2
n
limiting point at infinity, lie on a branch of a curve which
has

the line argz


<x as asymptote, it is sometimes
convenient to
denote this fact by writingf z -> oo e ai Thus n+i/n -+
n
+00,
and 1 +ni+ 1/n -> 1 +00 i. This convention does not violate the
postulate that there is but one point at infinity in the Argand
plane; it is merely a simple way of specifying the manner in
which the limiting point at infinity is approached.
.

2.31. Cauchy's principle of convergence


It can be easily shown that a sequence of complex numbers
z v z 2 ,..., z n ,... is convergent if and only if the two sequences
of real numbers xv x 2 ,..., x ,..., and
,...,
where
,...,

xn

= Rlz,

yn

= lmzn

yx

are convergent,

y2

and

yn
so theorems con-

cerning the convergence of complex sequences can be deduced


from the corresponding theorems for real sequences. It is, how-

much more

convenient to

make

principle of convergence,

which runs as

follows.

ever, usually

use of Cauchy's

and
zx ,

The necessary

sufficient condition for the convergence of a complex sequence


z s ,..., z n ,... is that, given any positive number e, there should

M
t e

is

used here to denote cos

a+ i sin a.

THE CONVERGENCE OF INFINITE SERIES

20

an

exist
\z

N, depending on

integer

N+p zN

e,

such that the inequality

< e holds for every positive integer p.

The condition is necessary for if zn - z, there exists an integer


inequality
N, depending on the given number e, such that the
positive
any
N. Hence, if p is
\z n -z\ < ie holds when n >
;

integer,

we have

<

\Zn+p~Zn\

The condition is

|ZiV+;p-z|+l%- Z

also sufficient.

For

<

"

if it is satisfied,

the given

number of its points


hence, by the Bolzano-

sequence is bounded, since all save a


e;
lie in the neighbourhood oiz N of radius
limiting
Weierstrass theorem, the sequence possesses at least one
possess
must
it
point. Now if the sequence is not convergent,
finite

Accordingly,
at least two finite limiting points, z' and z", say.
number <=,
there exist integers q and r, depending on the given

such that

z '-z N+a

With

these values of N,

=
<

z '-z'\
\

<

\z"-zN+r

e,

q, r,

<

e.

we have

\{z'-zN+q )+(zN+a -zN )-{zN+r -zN )-(z"-zN+r )\

\z'-zN+q

N+a -zN

\z

N ^-zN

\z

+ \z"-zN \

<4e.
however, impossible, since z' and z" were supposed to
is quite
be distinct limiting points of the sequence, whilst e
than one
arbitrary. Hence the sequence does not possess more
sufficiency of the
the
proof
of
the
completes
this
point;
limiting

This

is,

condition.

Example

Determine the limiting points of the sequence of com-

plex numbers

where a

lx

n7T

and rational,
and irrational.
If zn ->z,z'n -> z', prove that

is (i)

real

(ii)

real

Example
and

n7T\

*=( 1+ n)( 0OS ^ + * Sm ^i'

2.

also that, if

z' is

not zero, zjz'n ->

zn +z'n

-> z+z',

zn z'n

-> zz',

zjz'.

The maximum and minimum limits of a sequence


of real numbers
We shall now show that a sequence of real numbers x v x 2 ,...,

2.32.

cases
xn ,... possesses a greatest and a least limiting point. In the
one
but
has
sequence
->
the
-oo,
->
x
+oo, n
when xn ->x, xn

THE CONVERGENCE OF INFINITE SERIES


limiting point

and there

is

21

nothing to be proved. Accordingly,

we may

restrict our attention to oscillating sequences.


Let us suppose, in the first place, that the sequence is bounded

above, that

number K such that the


< K holds for all values of n. We now divide the

is,

inequality xn

there exists a real

numbers into two classes L and R in the following manner.


The number x belongs to R if there are only a finite number of
members of the sequence greater than x, and belongs to L in
real

the contrary case.

number

Now
a

finite

x,
xn ->

Evidently the class

greater than

K belongs to

exists, since

every

it.

if every real number belonged to the R class, all save


number of members of the sequence would be less than

no matter how large x was; but this would imply that


co which we are supposing is not the case. Hence the
>

L class also

exists.

Moreover, this classification of the real numbers


conditions for a Dedekind section, namely,
(i) every number belongs to L or R;
(ii) each class exists;

satisfies

the

(iii) any member of L is less than any member of


R.
Hence, by Dedekind's theorem,! there exists a real number

such that all numbers less than it belong to L and all numbers
greater than it to R.

Now

if e is

are, therefore,

any

positive

only a

finite

number,

+^ e

belongs to R; there

number of members of the sequence

and so g+e is certainly not a limiting point.


Hence the sequence does not possess a limiting point greater
greater than |+|e,

than

On

the other hand,

f e

belongs to L; there are, therefore,

an infinite number of members of the sequence between e


and +e, and so is a limiting point of the sequence.
We have thus shown that a sequence of real numbers,
bounded above, possesses a greatest limiting point . We call
g the maximum limit of the sequence, and write
|
If,

however, the sequence

= lima;

ra

unbounded above,

imix = +00.

write

is

r-

t See

Hardy, Pure Mathematics

(1921), 29.

it is

usual to

22

THE CONVERGENCE OF INFINITE SERIES


The minimum limit of the sequence, which is denoted by

lim xn

is

defined

by the equation
lim: n

= lim( xn

\m\xn has a finite value f


the~sequence; but if \rmx n =
If

is

oo,

).

the least limiting point of


the sequence

is

unbounded

below.
Finally,

it is

evident that the equation

= lim:r

]iva.x n

holds only in the cases

TC

when xn -> x, xn -> +co

or x n ->

oo.

2.4. Infinite series

The symbol

+a 1 +a

+...+aK +...,

which involves the addition of an infinite number of complex


numbers, has, in itself, no meaning. In order to assign a meaning to the sum of such an infinite series, we consider the associated sequence of partial sums s s 1} s 2 ,..., s n ,..., where
,

sn

= a + a + a + -+ anl

If this sequence tends to a finite limit


series is

convergent and that

its

sum

an

=s

s,

we say

that the infinite


we write

is s; in this case,

00

the sequence of partial sums does not tend to a finite


limit, we say that the infinite series is divergent.
The necessary and sufficient condition for the convergence of
an infinite series is provided by Cauchy's principle of con-

But

if

There are, however, numerous sufficient conditions


which may be deduced from this general principle.
An alternative procedure is to make the theory of the convergence of complex series depend on the corresponding theory
for real series, with which we shall assume the reader is acquainted.! For the sequence of partial sums s s v s 2 ,..., ,...

vergence.

Mathematics (1921), chapters 4, 8; Bromt See, for example, Hardy, Pure


wich, Theory of Infinite Series (1926), chapters 1-4 ; Knopp, Theory and Application of Infinite Series (1928), chapters 3, 4, 9, 10.

THE CONVERGENCE OF INFINITE SERIES


of a complex series

tend to

finite

23

convergent if and only if Blsn and Ims


limits, that is, if and only if the two real series
is

CO

00

JRIoB>

Xlma

are convergent.

2.41. Absolutely convergent series


00

An

2 an

infinite series

f complex terms

said to be abso-

is

o
00

lutely convergent if the series


] lJ

convergent.

is

We now

show that

absolute convergence implies convergence, but not con-

versely.
00

Let us suppose that the

and
sn

let

series

2o an

*s

absolutely convergent,

us write

= a + ai +a

+...+an

an

|ol+lil+lH--+KI-

Then, given any positive number e, we can find an integer N,


depending on e, such that the inequality
\

aN+p~ ivl

<e

Now
= aN+l-h aN+2 JT

holds for every positive integer p.


s
\

N+p sn\

<
But, since oN+p crN

is

la^+1

-i-

aN+p\

+ lJV+2 + +
1

liV-hjp

positive, this gives

<

aN+p "jvl
e
Hence, by Cauchy's principle of convergence, the absolutely
\

SN+p sn\

GO

convergent series ]T an

*s

convergent,

To complete the proof of the theorem, we need only observe


that the series

.,

.,

1^2 T 3 T ^ n
is

convergent, but not absolutely convergent.

The most important property of an absolutely convergent


sum is not altered by changing the order of its
terms in any manner, a property which is not possessed by nonseries is that its

a f

THE CONVERGENCE OF INFINITE SERIES

24

absolutely convergent series. This follows from the corresponding well-known theorem concerning real series.

2.42. Sufficient conditions for absolute convergence

We

now prove

shall

convergent

absolutely
\a \^ n

>

that the series

lim |aj 1/w

if

<

2 .

f complex terms is
but is divergent if

1,

This test, which is due to Cauchy,


importance in the theory of power series.

fim

For

lim |aj 1/m

if

equality

1.

values of n.
\an

<

= 1 2c,

1 c

There

(1 c) n when

where

< c < |,

exists, therefore,

an

integer

of great

then the

true for all save a finite

is

in-

number of

M such that

n^M.

now, an denotes the

If,

we

<

\a n lln

is

nth. partial

sum

of the series

an \>

^ M,
K +*-tfJ < (l-c)*+n +(l-c)-*+,..+(l-c)
(1 c) n+1
(lc) + (lc) n +P+
^
c
c

have,

when n

Since this last expression tends to zero as

an arbitrary

positive

(> M) such

number

vergence, the series

the other hand,

inequality |aj

n -> go, we may assign

and then choose an integer

1 're rjs

<

e is true for
\aN+p
N
Hence, by Cauchy's principle of con-

that the inequality

all positive integers p.

On

an\

if

1+d

is

convergent.

Hm \an \^ n =
is

l+2d, where d

true for an infinite

> 0,

the

number J of

(l+d) n for an
values of n. This, however, implies that \a n
infinite number of values of n, and so a n does not tend to zero
\

>

series 2 an * s divergent.
which is somewhat easier to apply, is d'Alembert's ratio-test, which states that the series 2 a n f complex

as

->- co.

Another

terms
if

lim

Hence the

test,

is absolutely
\an+1 /a n

>

convergent if ]im
1.

The proof of

<

1, but is divergent
Jan
this is left to the reader. It

\a n+

should, however, be observed that d'Alembert's ratio-test


definitely less powerful

than Cauchy's

nth.

root test.

Sm \an+1 /an > hm" |on |" > lim |an |" > hm \an+1/an
\

is

3?or, since
\,

t See Bromwieh, loc. cit. 71, or Knopp, loe. oit. 138-9.


{ But not necessarily for all sufficiently large values of n.
For a proof of this inequality, see Bromwieh, loc. cit. 421-2; Knopp,
cit. 277-8.

loc.

THE CONVERGENCE OF INFINITE SERIES


25
a series ] an may be absolutely convergent in virtue of the fact
that Mm |aw |V
1, even though fim \an+1 /a
1; the series
n

<

>

l+a+b*+a3 +b i +...,

<

where

<

<

\a\
\b\
1, is a case in point.
applies to the ratio-test for divergence.

If,

however,

fails.

similar

if

remark

tends to unity, d'Alembert's ratio-test

We may then have recourse to Raabe's test,

that the series

For

\a n+1 /an

which states

2 an is absolutely convergent if

the value of this

maximum

limit is

1 2c, where c is

positive, the inequality

n[*-i)<-i.
holds for

all

tainly true

save a

when

finite

n^N,

number of values of n, and


say.

If

we

so is cerwrite this inequality in

the form
c| m

<

and add, we

(m-l)\am \-m\am+1
find that,

when n

4KI + K+il+-+KI} <


But

(m

= N,N+l,...,n)

> N,

(N-l)\aN \-n\an+1

< {N-l)\aN

\.

(JV l)|a^| does not depend on n, the series


2 KJ
of positive terms has bounded partial sums and so is convergent.
Hence 2 an *s absolutely convergent.
since

The three

tests for absolute convergence given in this section


be found quite adequate to deal with the series used in this
book. For a discussion of more refined tests, we refer the reader
to the works of Bromwich and Knopp already cited.
will

2.43.

A test for non -absolute convergence

Of the various

sufficient conditions for the convergence of


a series of complex terms which does not converge absolutely,
the following test, due to Dedekind, is one of the most important. The series
n vn is convergent if

^a

(i)

the series

(ii)

the series

(iii)

vn ->

as

2 an h8 bounded partial sums,


2 (n vn+1 is absolutely convergent,
)

n -*- oo.

and

THE CONVERGENCE OF INFINITE SERIES

26

This test

closely related to the tests of Abel

is

and its

real series, f

proof, like theirs, depends

of partial summation.
Let us write

=a

sn
Then
\s n

and Dirichlet for


on Abel's method

+a1 v

+...+an vn

by hypothesis, a constant

there exists,

such that

< K, for all values of n.

Now

if

is

any

positive integer,

we obtain by

partial

sum-

mation
"n+p

"n

= an+l vn+l JT an+Z vn+2 JT JT an+p vn+p


= S n+l Sn) vn+l~Jr(sn+2 S
m+2+"- +
ji+l) 1,

s n+p-l) vn+p

i" \ s n+p

n+pl

= *+l+ X Sr(VrVr+1 )+Sn+^Vn+p


r=n+l
and so

l^-m-^nl

<K

n+pl

K+ll+

I*!- r+ll

< K lK+i\+K+p\+
But
is

any given

if e is

convergent,

we can

that

r=n+l

is

I, ,+il).

\v,vr+1

Moreover, since vn tends to zero,


%e/K when n
2 such that \v n
the greater of x and 2 we have

convergence, the series

and

2 an vn

that, if n

>

*s

we can also

>N

2.

r=m+l

so,

by Cauchy's

principle of

(r~ sm)(vr

^+1) + (*+ Sm )v+p

n+1 ) is absolutely convergent.

loc. cit.

if

convergent.

in the notation of the previous section.


Deduce that the series
an vn is convergent if
(i) the series
converges, and

t Bromwich,

find

Hence,

m,

Sn+p Sn = (* *m)^n+i+ 2

(ii)

r +\\

such

<e

2
the series 2 (vn

e,

sn+p8n\

for every positive integer p,

Example. Show

<

\vr

e
< ^=
oA.

depending on

lt

>N

an integer

an integer

2,

when n

f.

+ Knl

positive number, then, since


find

58-60 Knopp,
;

loc. eit. 314r-15.

THE CONVEKGENCE OF INFINITE SERIES

27

Double sequences

2.5.

A set of complex numbers smn where m and n take all positive


integral values

is

to the value s

if,

called a double sequence. It

is said to converge
given any positive number e, we can find an
integer N, depending on e, such that the inequality \s m>n s\
e
holds whenever
and n exceed N. The finite number s is then

<

and is denoted by
double sequence which does not converge is said

called the limit of the double sequence

lim sm>n

m,n>go

to be divergent.

The fundamental theorem


lows.

in the theory of double sequences

principle of convergence,

is Stolz's

which

be stated as

fol-

sufficient condition for the convergence of


is that, given any positive number e, there

the double sequence smn


should exist an integer

M depending on

may

The necessary and

sx>,o~ sm,n\

<

holds whenever

The condition

is

e,

such that the inequality

> m > M and q> n > M.


To show that

obviously necessary.

sufficient,

we

*i,i> "*2,2>-->

sn,w- converges to a finite limit

observe that

it

also

it is

implies that the simple sequence


s,

say. Accordingly,

an integer
|s snn < \e when
x such that
n > Nv But, by hypothesis, there also exists an integer N2
such that \sn>n
< |e when p > n > 2 q > n > 2V2
Hence, if
is the greater of
<e
2 we have \s pq
x and
when p > N, q > N, and so sp converges to the limit s.
It follows immediately from the definition that if s mn be a
there

exists

convergent double sequence for which both the repeated limits

lim (lim s m

lim ( lim smn

each repeated limit is equal to the limit of the double sequence.


For if s is the limit of the double sequence sm , the inequality

exist,

sm,n~ s

But

< e holds when m and n exceed a

since

Mm sm>n exists,
\

when

m > M.

certain integer

M.

this gives

]imsm,n s \<

This, however, implies that

lim/limsm J =3.

similar

equal to

argument shows that the other repeated

s.

limit, is

also

THE CONVERGENCE OF INFINITE SERIES

28

The

and equality of the repeated limits of a double


sequence do not imply that the sequence is convergent. For
existence

instance, if

mn

~ (m+nf

Sm n
'

both repeated limits are equal; but since


1

n,n

the double sequence

is

"v.n .
2n,n

4'

not convergent. Again, the example

(- l m W-+-)

v =

shows that a double sequence

\m

may

n)

be convergent even though

neither repeated limit exists.

2.5 1.

By

Double series
a double

series,

we understand an

bers of the form

_
_
a o,o ~r u,i ~r w o,2 ~r
,

+ Ol,0 +

O l,l

array of complex

num-

a l,2+-

+ ,o+a i+ a 2,2+2

2(

there being an infinite

number of rows, each containing an


number of terms. Associated with such an array is the
double sequence of partial sums

infinite

Sp,q

Z.

2,

m=

7i

am,m

where p and q are positive integers. The double series is said


to be convergent if the double sequence
converges to a finite
limit s. We then call s the sum of the series and write
00

= 2 am,nm,n=0

A double series which is not convergent is said to be

divergent.

If the repeated limit

M)= f

Km(lim
exists, its value is called

Similarly, if

lim

(f

am

sum by rows

the

of the double

,,,,
limO= nJ= m=0
2 .)

q->co ^p><x>

'

series.

THE CONVERGENCE OF INFINITE SERIES


exists, it is called the sum by columns. It follows from 2.5
a double

if

columns,

series is convergent

all three

sums are

Example. Prove

and

that the

sum

summable by rows and by

(Pringsheim's theorem.)

equal.

positive terms can be obtained

is

29

that

of a convergent double series of


or by columns.

by summing by rows

2.52. Absolutely convergent double series

2 am f complex terms is said to be abso2 lm,J s convergent. An absolutely con-

double series

,n

lutely convergent if

vergent double series is convergent.

For if we denote by apa the sum of the moduli of the terms


which occur in spq then, by Stolz's principle of convergence,
if we are given a positive number e, we can find an integer
such that the inequality
< a - amn < e holds when
,

p>m>N, q>n>N.
this gives

\s

hence the double

<

But since \s -sm>n <


when p > m > N and q

m>n
series converges.
\

We
be

shall now show that an absolutely convergent


summed by rows or by columns. For since

a^a^,

> n > N;
series

may

oo

2 KJ < r,s=0
2 K
s=0

s \,
00

no matter how large n

may

lutely; let us denote its

be, the series

sum by

r.

2a

rs

converges abso-

Then

oo

go

oo

2 \A\ < r=0


2 K,l,
2 s=0
2 K.I < r=0
2 s=0
r=o
no matter how large
finite, since

m may

But

be.

this last expression is

a convergent double series of positive terms can be

summed by rows. Hence 2 -A r i s absolutely convergent, and so


the sum by rows of^ars exists; similarly the sum by columns
also exists. The required result now follows by Pringsheim's
theorem

2.51).

Example 1 The double


sum S. Show that, if
.

series

^a

^n 2
then Sn > S

as

n ->

oo.

r.s i s

ar.s>

absolutely convergent with

THE CONVERGENCE OF INFINITE SERIES

30

CO

00

Example

Show

2.

that, if

2K

and

B respectively,

vergent series with sums A,

are

two absolutely con-

then

00

2,

(aiK + a 2 bn-l + + an-l b 2 + an b l)

is

and has sum AB. (Cattchy.)

absolutely convergent

REFERENCES
Theory of sets of points:
P. Dienes, The Taylor Series (Oxford, 1931), Chap. VI.
E. W. Hobsost, Functions of a Real Variable, 1 (Cambridge, 1921),
Chap. II.
Convergence of infinite series of complex terms:
T. J. I'aBbomwich, Infinite Series (London, 1926), Chap. X.
E. C. Fbancis and J. E. Littlewood, Examples in Infinite Series,
with Solutions (Cambridge, 1928).
Theory and Application of Infinite Series (Glasgow, 1928),

K.

Kistopp,

Chap. XII.

Double Series:
T.'J. I'a

Beomwioh,

Chap. V.

loc. cit.,

MISCELLANEOUS EXAMPLES
1.

also

Show that the series 2 ^/n converges absolutely when \z\ < 1 and
when \z\ = 1 and Rl/x. > 1, but that it diverges when \z\ > 1.
11

Prove

also

absolutely,

Finally

when
2.

it is

that

show

if

<

when

that,

\z\

Rl;/,

<

when

1,

the series converges, but not


it diverges when Rl/x < 0.
the series converges only if Blfj. > 1,

1,

absolutely convergent.

Prove that,

if

an

is

never zero and

an
where w n is bounded, p
(=7^

0) as

1,

and that

n->

>

1,

nP'

and p is a complex constant, then a% n-v -*- 1

oo.

Hence show that the series 2 a converges


and that it is then absolutely convergent.
3.

Prove that the binomial

if

and only

convergent

if

1,

...

when \z[ < 1 and diverges when


> 1. Show
1, the series is absolutely convergent if Rlr > 0,
1 < Rlv < and z ^ 1, and divergent if Rlv < 1.

converges absolutely

on

>

series

l+vz+v(v-l)J + v(v-l)(v-2)^

also that,

if Rl/ti

\z\

\z\

THE CONVERGENCE OF INFINITE SERIES


Show

4.

31

that the hypergeometric series

/
ft

(a) n = a(a+l)(a+2)...(a+n 1), converges absolutely when


and also when \z\ = 1 and Rl(y a /?) > 0, but that it diverges
when \z\ > 1. Discuss also the convergence of the series on \z\ = 1
when Rl(y a. /?) < 0.
5. Prove by using Dedekind's test that, if
an nr* is convergent and
Biz > Rl^, then X an n~" 's also convergent. Show by an example
that the latter series is not necessarily convergent when Rlz = Rift.

where

<

\z\

Show

6.

that, if

f(z,n)

= z(z-l)(z-2)-(s-n+l)

and z is not zero or a positive integer, the series ^.an f(z,n) and
2 an( l)"/nl+* are either both convergent or both divergent.
Prove

7.

that, if
'

and

z(z+l)(z+2)...(z+n)

'

not zero or a negative integer, the series


an n ~z are either both convergent or both divergent.
z is

Show

8.

that, if

A n > A

as

w ->

oo,

J an F(z, n)

and

then

(4 1 +4,+...+4)/n->4.

More

A n -> A, Bn -> B, then


Bn +A 2 Bn_ 1 +...+A n B )/n^AB.

generally, prove that, if

(A 1

00

The

9.

series

00

2 m> 2 &
l

spectively.

Show

sums

and

2? re-

that, if

cn

and

are convergent with

a 1 bn +a i bn _ 1 + ... + an_ 1 b i +an b 1

= Cl +c

+...+cn

lim(C1 +02 +... + C)/n

then

= ^B.

n->oo
00

Deduce

that, if

2 c is convergent,

its

sum

is

AB.

(Abel's theorem.)

Show

that it is sufficient for the convergence of the series


cn f
that one of the series
a
&' De absolutely convergent.!
(Merten's theorem.)
10.

Ex.

9,

Compare the

result of 2.52,

Ex.

2.

CHAPTER

III

FUNCTIONS OF A COMPLEX VARIABLE


The

3.1.

definition of a function

When we

say that tv is a function of the complex variable z,


denned in a domain D of the Argand plane, we mean that we
can calculate the value of w at each point z of D by a given
rule or set of rules. For example, the greatest integer less than
\z\ is a function of the complex variable z in this general sense.
This definition

however, far too wide for our present purx-\-iy, then w is of the form
u(x, y)-\- iv(x,y), where u and v are real functions of the real
variables x and y of the most general possible type. In the
present chapter we consider how this definition may be modified
so that the methods of the differential calculus of functions of
a single real variable may also be applicable, as far as possible,
poses.

For

is,

implies that, if z

it

to functions of a complex variable.

Continuous functions

3.2.

A function f(z),
Argand

plane,

is

defined in a

bounded closed domain

said to tend to a limit

D of the

as z tends to a point

along any path in D, if, given any positive number e, no


matter how small, we can find a number 8, depending on e and
z such that the inequality
z of

\M-l\
holds for

points z of D, other than z

all

region \z

<

<

When

this is the case,

lim/(z)

which belong to the

we

write

I.

This definition says nothing whatever about the value of f(z)


and, in general, I need not be equal to/(z ). But if/(z)
h
we say that the function is continuous at z In other words,
at z

bounded closed domain D, is said


continuous at a point z of D, if, given any positive number
we can find a number 8, depending on e and z such that the

the function f(z), defined in the


to be
e,

inequality

\f(z)-f(z
holds at all points z of

)\

<e

D in the neighbourhood

\z

< 8 of z

FUNCTIONS OF A COMPLEX VARIABLE


If a function

33

continuous at every point of a bounded closed


domain D, it is said to be continuous in D.
ft should be observed that a function
of the complex variable
z
x+iy, continuous in a domain D, is necessarily
a continuous
function of a; when y is constant and also
a continuous function
of y when x is constant. But continuity
with respect to x and
y
separately does not imply continuity
with respect to z. An
instance of this is provided by the
function
is

f(z)

which

= xy/(x*+y*)
=

(Z

^
= 0),
),

a continuous function of a; and separately.


y
But since
2
) on the line y
mx, f(z) is certainly not a continuous function of z at the origin.

f(z)

is

= m/(l+m

Example

1.

Show
/(z)

that the function denned

= jq^-

is

^0),

/(0)

discontinuous at the origin.

Example

2.

Show
f(z)

is

by the equations

that the function denned


x*j(x*+y*)i
(

^ 0),

by the equations

/(0)

continuous at the origin.

E
iPle 3 " Thefunctions /(2)and^) are
continuous at z
Show
,
that/( 2 )+gr( 2 ) and /(z) gr( 2 ) are also continuous
there. Prove also that
J(z)l9(.z) is continuous at z if g(z
) i s not zero.

wT

3.21.

Uniform continuity

We have seen that, if f(z) is continuous in the bounded closed


domain D, then, given any positive number e and
any point z
of D, we can find a positive number S( e
,z), depending on e and
z,

such that the inequality

!/(*')-/(*)

holds whenever

<*
D

a point of
in the neighbourhood
\z'-z\
8( e ,z) of the point z. In point of fact,
there are an
infinite number of such numbers
S(e, z); for if the property holds
with S(e, z)
8', say, it also holds when
8'. Let us now
8(e, z)
denote by A(e, z) the largestf of the numbers 8(e,
z); it is a positive
function of the complex variable z defined at
each point of D.
z'

is

<

<

the

4111

of the numbers S( e z)
of the real numbers.

lal es *

rnl^
f section
means
of ^rf^T
a Dedekind

is

readily proved

by

FUNCTIONS OF A COMPLEX VAKIABLE

34

of A(e,z)
There are two possibilities regarding the behaviour
positive number ij(e),
in the domain D. Either there exists a
A(e,z) 5? 17(e)
depending only on e, such that the inequality

at
points of
holds at every point of D, or else there exist
please.
as
we
small
as
values
which A(e, z) takes
find a sequence of
If the second alternative is true, we can
is
which
A(e,zJ
(i)._ Since
points z v *,,..., z n! ... of D, for
point
limiting
one
,
least
at
bounded, this sequence possesses
is
since
moreover,
the Bolzano-Weierstrass theorem;

<

by

closed, ^ is

a point of D.

But since/(z) is continuous at ?, there exists a positive

number

8 such that the inequality

\M-ffl\

<h

28. Now
in the region \z-t,\
holds whenever z is a point of
points
all
for
8. Then,
for which |z-
let z be any point of

<

z'

of

D such that

|z'-z|

W)-f(z)\

From

<

<

< 8, we have
!/(*')-/()

this it follows that A(*,z)

>

|z'-|

<

28,

+ 1/(0-/(2)1 <
8,

whenever z

is

and

so

a point of

<

8 of the limiting point . But


in the neighbourhood |z-|
of the sequence z v z 2 ,...
points
are
this is impossible, since there
alternative is thus
second
The
in every neighbourhood of .

untenable.

We have thus shown that,

//(z)

continuous in the bounded

domain D, then given any positive number e, we can find


inequality
a positive number r,(e), depending only on e, such that the

closed

\f(z')f(z)

<

holds for every pair of points

z, z'

of

D for which

other words, we can choose the number 8(c, z)


z
z
,j( e ). In
of D. This property
so that it has the same value at each point
form: a function which is conis usually stated in the following
continuous
tinuous in a bounded closed domain is uniformly

<

'

there.f

3.22.

Bounded functions

A function f(z), defined in a domain D, is said to be bounded


that the inequality
in D if there exists a positive constant K such
|/(z)|

< K holds at each point z of D.

of a real variable, see


+ For the equivalent theorem concerning functions

Hardy, Pure Mathematics (1921), 189-90 (Theorem

II).

/(
.

i8

FUNCTIONS OF A COMPLEX VARIABLE


35
a uotta a bounded closed domain
D,
it is bounded
.T r

2K and this
/("JI
se <* uence Woul possess^
-^
'
*
at
.
lift one limiting
r
least
poznt { belonging to D. This
is, however

ssr*
r?/p
f* to athat
"* }} ^ould
be observed

finite iimit as

true if the

domain
tmuous when \z\ <

* ^
con

1,

The symbols

3.23.

<
*
this result is not necessarily
is not closed; for
example, l/(l- 2 is
.
but

is

and

not bounded there.


o

The notation </(*)= 0(1) as 2 -> a


means that /(z) is bounded
a neighb
h d of the po nt o more predse
ther
positive numbers A and
8, independent of z, such that
|/(z) < A
when z - a < 8 Similar </(z) = O(l)
as 2->oo' means ttal
f(z) is bounded m a certain
neighbourhood of
>

^ 7^

the point at infinity

fj= nii%T
0(1^(2)1).

(1)

}{Z)

2_>a

When, however, /(z)/^(z ) tends


= o[\
Thus z* =

write

m)

as 2-^oo.

We

'

Sha11

retimes

to zero as z -> a it
(, z)) as z
, but
,

is
S

write

usual to

lp)
w
'

'

3.3. Differentiability

We shall now consider whether the definition of the

derivative

a function of a single real variable


is applicable in the theory
of functions of a complex
variable. If/( 2 i s a
ot

f(z) is differentiate at

a point z of

one-valued func-

D if the increment-ratio

z
tends to a finite limit as 2 tends
to z in any manner, provided
tnat 2 always remains a point
of D.
7: We require that there should e
a number
7 with the
I
following property: given any
positive number e no
matter how small, there must exist a
positive

^f?

pending on

^t

and possibly on

number

z-Zo

holds whenever z is a point of

8 ' de-

zQ such that the inequality


,

D in the neighbourhood |z-2 < S


1

FUNCTIONS OF A COMPLEX VARIABLE

36

When this is the case, we call I the derivative


of the point z
/'(z ).
of f(z) at z and denote it by
be differentiable at a certain
should
In order that a function
there; for otherwise the increment
point, it must be continuous
to a finite limit On the other
ratio would certainly not tend
simple
chfferentiabmty.
hand, continuity does not imply
\z\ ,
function
continuous
the
instance of this is provided by
when
For
else.
nowhere
but
which is differentiable at the origin,
.

z^z

and

we have

= >

Z |2_ Zo |2
|

= zz-z^ = g + g
=

!=!o
o

z+z (cos2a isin2a),

not tend to a
axg(z-z ), and this expression does
where a
when z is zero, the
Z , in any manner; but
unique limit
to zero with z.
increment ratio is z, which tends
not differentiable anywhere.
Example. Show that \z\ and argz are

as^

3.31.

The

definition of

an analytic function

point
one-valued and differentiable at every
finite
a
for
possibly
of the Argand plane, save
of a domain
the
in
analytic
that it is
number of exceptional points, we say
points
singular
exceptional points are called the
If a function

is

domain D. The

or singularities of the function.

analytic
is a singularity of the
however, no point of
regular in D.
function, we then say that it is
which are analytic or
functions
When we are dealing with

If,

regular in a domain,

we

shall often find it

use the ordinary notation

more convement to

dm
dz

instead of f'{z).
for the derivative of f(z),

and g(z) are analytic functions, regular


and
/(*)*) ? also
a domain D, then /(*) + <?(*)
Z>
the ordmary rules of the
derivatives may be calculated by

Example
in

1.

Show that,

if/()

^^^

'

that their

"So"' also to*


vanish there.

mm

**"

*D

>

provided that

9{z)

does not

Example

FUNCTIONS OP A COMPLEX VARIABLE


2. Prove that, if wis a regular function of

a regular function of

itself

then

z,

is

dw

dw

dz

d dz'

37
,

which

a regular function of

is

and

dt,

Polynomials and rational functions


One of the simplest analytic functions is zn where n

3.3i2.

positive integer.

This function

domain, and has derivative nz"- 1

2
as z - z

= z-

+2-%+...+zz-2-f 2-i

-> na-i

2,

Similarly,

we can show that

integer, is analytic in every

2-", where n is a positive


bounded domain and has but one

singularity, the origin; its derivative is

is

regular in every bounded


for when z
z

is

nz - " -1

By using the result of 3.31, Ex. 1, we find that the function


-j-a 1 z+a z 2 + ...+a z n where n is a positive integer and the
z
n
,

ar are constants, is also an analytic function, regular


in every bounded domain. Such a function is called a polynomial in 2 of degree n. It will be shown in 6-21 that this
polynomial is expressible uniquely in the formf
coefficients

<Uz-Zi)(z-z 2 )-(z-zJ,
where the numbers zv z2 ,..., zn are certain constants, depending
only on the coefficients ar Since the polynomial vanishes when
.

=z

r,

we

call z 1 , z 2 ,..., z n the zeros

of the polynomial.

The quotient of two polynomials is called a rational function.


Such a function is also an analytic function, its only singularities
being the zeros of the denominator.
3.33.

Power

An infinite
z,

series
series,

proceeding in ascending integral powers of

of the form

+a1 z+a

z2

+ ...+an z n +...,

where the coefficients a a v a 2 ,... are all constant, is called


a power series.% By Cauchy's nth root test ( 2.42), this series
,

t See also Hardy, Pure Mathematics (1921), 83, 433-7.


t The conclusions of this section will also be true of
CO

form

2 an z ~
(

<*)">

tne obvious changes being made.

power

series of the

FUNCTIONS OF A COMPLEX VARIABLE

38

converges absolutely
lim

\a n z n \^ n

>

when lim

Hence,

1.

<

\a n z n \^ n

and diverges when

if

Km |J-iM =

S>

when
< B and divergent
R is called the radius of convergence
of the power series; the circle
= R is its circle of convergence.
There are three cases to be considered, viz.
R = 0,
R
finite,
R infinite. The first case is quite trivial, since the

the series

when

is

absolutely convergent

> R.

\z\

\z\

The number

\z\

(i)

(ii)

(iii)

then convergent only when z


0. In the third case
the series converges for every finite value of z.
series is

We shall now show that, if a power series has a non-zero radius


of convergence,
circle

sum

its

is

an

analytic function regular within

its

of convergence.
CO

Let

f(z)

be the sum of the power

series

^an zn

which has

a non-zero radius of convergence R. Obviously /(z) is one-valued


\z\
R; we have to show that it is continuous and differentiable at every point of the domain \z\
Rx where Rx is

<

when

<

any

number definitely less than R.


Now choose a number R 2 such thatf < R
If |z|< -#1 and \M < S R
2
v we have [z+h\ <
finite positive

f(z+h)-f(z)

<R <
2

R2

R.

and so

= 2an (z+h)--fan z=^

an{(z+h)^+(z+h)^z+...+

+ {z+h)zn - +zn ~
2

From

this, it follows

1
}.

that

\f(z+h)-f(z)\

<

\h\2n\aJB? i
i

But

this series of positive

terms

is

]i^{n\an \R$-iym

Hence
where

\f {z +h)-f{z)\

tinuous

is

independent of z and

when

\z\

<R

convergent, since J

= ^<1.
Ii

< A\h\,
h,

and therefore

f(z) is

con-

t If iS is infinite, it is conventional to interpret the inequality as meaning


that i? 2 is finite and greater than iJ t
% We use here the well-known result that lim n 1 /" =1.
.

FUNCTIONS OF A COMPLEX VARIABLE

39

Now

the increment ratio {f(z+h)f(z)}/h tends formally to


n
nan z -x as h -> 0. Accordingly, we consider the expression

zfflwhere

and A

,^.

same conditions as

satisfy the

use of the binomial theorem,


(z-i~Ji\ n

and

^(feW

we

before.

By

the

see that

zn

so

n z n
{z+h)
v
'

,W-1
nz"

<\M^ nG

r \z\

n -r

\hn

n 2

\MZnCs + 2\z\ n

-S -2

\h\

n-2

<**(- 1)|*| I

n - 2Cs \z\"--*\h\

= Wfl-l)\h\{\z\ + \h\}-*
< ra(rc 1)|A|^< i|A| |(n-l)|aJ2$- = |A|,
2

Hence

|Z|

where

jB is finite and independent of z and A, since the


~2,n(nl)\an \R%- 2 is convergent. But this implies that
/ tends to zero with h. We have thus shown that

say,

series

/'(z)

= |>

<R

z-i,
re

provided that \z\


lt where
x is any number less than the
radius of convergence of the given power series. This completes
the proof of the theorem.

Now,

Urn \nan \-^

n^n z _1 nas the

since

the series
the series

2
J "

lim |aj~

1/m
,

same radius of convergence

whose sum

is f(z).

Hence,

theorem to the derived series, we see that/'(z)


function, regular when \z\
n(n l)an z n ~ 2 and so on.
order p is regular when \z\

<

R, and that

if

is itself

its

we apply

as
the

an analytic

derivative

is

Thus the derivative of/(z) of any


< R, and is given by the formula

d"

FUNCTIONS OF A COMPLEX VARIABLE

40

In other words, a power series can be differentiated term by term


as often as we please at any point within its circle of convergence.

Example. Show that, if \ajan+1 -> R as n -s. oo, then R is the radius
of convergence of the power series
an zn (This result often provides
a simple method of calculating the radius of convergence of a power
1

series.)

The Cauchy-Riemann equations


We have just seen that the sum of a power series

3.4.

zero radius of convergence

with a non-

an analytic function, regular within


its circle of convergence. The converse theorem, that an analytic
is

function, regular in a neighbourhood of a point z


pressed as a power series of the form

can be ex-

f (-o) n
o

with non-zero radius of convergence, is also true, and will be


proved in 4.5. Thus the whole of the theory of analytic functions can be made to depend on power series.
For some purposes it is, however, more convenient to make
use of an alternative method of denning analytic functions
which depends on the theory of continuous functions of two
real variables. We shall suppose that the reader is acquainted
with the elements of this theory .f
Let us suppose that, when z
x+iy, the function f(z) is
expressed in the form u(x, y)-\-iv{x, y), where u and v are real
functions of the two real variables x and y. Now if f(z) is differentiable at a given points, the increment ratio {f(z+h)f(zj}/h
tends to the limit f'(z) as h ->- 0. If we take h to be real, this
implies that the expression

u(x+h ,y)u(x,y)
_
tends to f'(z) as A

must

-*- 0.

.v(x+h,y)v(x,y)
_

+t
,

Hence the two partial derivatives ux vx


and the derivative is then given by
,

exist at the point (x, y),


/'(z)

= ux (x,y)+ivx (x,y).

t See, for example, Dockeray, Elementary Treatise on Pure Mathematics


(London, 1934), Chap. VII; Gibson, Advanced Calculus (London, 1931),
Chap. IV; Hardy, Pure Mathematics (Cambridge, 1921), 274-81; or Phillips,
Course of Analysis (Cambridge, 1930), Chap. IX.

FUNCTIONS OF A COMPLEX VARIABLE


41
find, by taking h to be purely imaginary, that
vy must exist at (x, y) and that

we

Similarly

uy

/'(z)

The two expressions

= vy (x,y)-iuy (x,y).
for f'(z) so obtained, must, however,

be

Equating real and imaginary parts, we find that

identical.

UX

= Vy,

These two relations are called the Cauchy-Biemann differential


We have thus shown that for the function f(z) u+iv
to be differentiable atz
x+iy, it is necessary that the four partial

equations.

ux , vx uy , vy should exist
differential equations.

derivatives

Biemann

and

That the conditions of this theorem are not

by

considering the functionf defined

This function

Cauchy-

satisfy the

sufficient is

shown

by the equations

continuous at the origin, and the four partial


and have valuesj

is

derivatives exist there

ux

satisfying the

uy =-~l,

l,

vx

=l,

Cauchy-Riemann equations;

will easily verify, the derivative /'(0) does

yet, as the reader

not

exist.

3.41. Sufficient conditions for a function to

We
f{z)

now show

shall

= u+iv

an

is

domain D,

if the

continuous,

and

=l,

vy

be regular

that the continuous one-valued function

analytic function of z

= x+iy,

four partial derivatives ux vx


,

satisfy the

uy

regular in a
,

vv

exist,

Cauchy-Biemann equations

are

at each

point of D.

Let
since

z = x+iy and z' = x'+iy' be two


ux uy exist and are continuous, we
,

points of D.

have,

Then,

by the mean-

value theorem for functions of two variables,


ii(x',y')~u{x,y)
t This example
159-60.
% It

is

= {ux (x,y)+ e}(x'-x)+{uy (x,y)+ v }(y'-y),


due to

Pollard, Proc.

should be remembered that ux (0, 0)


lim{u(a;,0)-(0,0)}/a;
-*0

Similarly for the other derivatives.

London Math.
is,

by

28

definition, equal to

= Km (x/x) =
aj-0

Soc. (2)

1.

(1928),
'

A COMPLEX VARIABLE

FUNCTIONS OF

42

where

and

77

tend to zero as

(s',y')-(*,y)

and

z'

->

Similarly,

z.

= {vx (x,y)+ e '}(x'-x) + {vv (x,y)+7)'}(y'-y),


Hence, by the Cauchy-

tend to zero.
Riemann equations, we obtain

where

e'

1/ also

/(')-/()

where

= K(*,y)+*vx {x,y)}&-z)+to,

c +e')(a;' x)+(i7+*V)(y' V)>

m-m _

and so

z'-z

U<x+ ivx-

Now
\y'-y\

\x'x\
z

z <{\A+W\)

(fol+IVI)

< H + le'l+M + IV
Hence
z

as z ->

z.

Thus

each point of D; this

/(z) is differentiable at

completes the proof of the theorem.

Example. The
derivative

is

function /(z)

identically zero.

is

Show

regular in a

that/(s)

The exponential function


We now introduce the exponential,

is

domain D, where

its

constant in D.

3.5.

and trigonometric functions of a complex variable by means of power series.


We shall assume that the reader is well acquainted with the
properties of the corresponding functions of a real variable, and
shall show how these properties can be extended into the complex domain.
The exponential function expz

expz

is

logarithmic,

defined

by the power

series

^zn

=1+2,^-

using the ratio-test ( 3.33, Ex.), we find that the radius of


convergence of this power series is infinite. Hence expz is an
analytic function which has no singularities in any bounded
domain in the z-plane.
When a; is a real number, exp x is identical with the function
ex of elementary algebra. We shall often find it convenient to

By

FUNCTIONS OF A COMPLEX VARIABLE


write

e?

when

for expz,

43

complex, since expz obeys the

z is

multiplication law

= exp(z+z'),

expz expz'
which

is

of the same form as the law of indices in


algebra.

To prove this, we observe that, if we differentiate


term by
term, we find that the derivative of expz
is expz. Hence, if
a be any

finite constant,

/(z)
is

identically zero,

is

found by putting

or, writing z' for

and

the derivative of the function

= expz exp(a z)

so/(z)

a constant whose value, expa,

is

= 0. We have thus shown that


expzexp(a z) = expa,
z

az, that
expz expz'

exp(z+z').
This result is usually called the addition theorem
of the exponential
function, f
An important consequence of the addition theorem is that
expz never vanishes. For if expz vanished when z
z v the
equation

expz 1 exp(

would give an
3.51.

infinite

value for

The trigonometrical

x)

exp(- Zl ), which

impossible.

is

functions

It follows

from the geometrical definition of the trigonometrical functions of an angle of circular measure
x thatj

sin^=

V( 1)"

x2U+1

for all values of the real variable x.

catT

"

'

,w

*n
n x

We now define

metrical functions of a complex variable z


sinz

the trigono-

by the equations

z 2n+l

Since the radius of convergence of each of these power series


is
sinz and cosz are analytic functions of z, regular in

infinite,

every bounded domain of the Argand plane. Moreover,

we

see

t The addition theorem can also be proved by using Cauehy's theorem on


the product of two absolutely convergent series
( 2.52, Ex. 2).
t See, for example,

Hobson, Plane Trigonometry (Cambridge,

1911), 131-4.

FUNCTIONS OF A COMPLEX VARIABLE

44

at once

of sin z

by term-by-term differentiation that the


and cosz are cosz and sinz.

derivatives

by the

are then defined

The other trigonometrical functions


equations

tanz

sinz

cotz

= cosz

secz

cosecz

cosz

sinz

cosz

smz

Obviously sinz, cosecz, tanz and cotz are odd functions of z,


cosz and secz even functions.
If we use the result of 3.31, Ex. 2, we find that tanz and
secz are analytic functions, regular in any domain in which cosz
never vanishes, their derivatives being sec 2 z and secz tanz respectively. Similarly cotz

and

cosec z are regular in

any domain

cosec 2 z and
which sinz never vanishes and have derivatives
coseczcotz respectively
If we denote exp(iz) by eiz in accordance with the convention
of 3.5, the equations defining the sine and cosine become

in

sinz

From

fe

(e

e~

iz

cosz

)/2i,

iz

(e

+e- iz )/2.

and the addition theorem

these formulae

easily follows that the

for expz, it

fundamental identity
sin2 z+cos 2 z

and the addition theorems

= sinzcosz' + coszsinz',
cos(z+z') = cosz cosz' sinz sinz',
sin(z-)-z')

hold when z is complex. As all the elementary identities of


trigonometry are algebraical deductions from the fundamental
identity and the addition theorems, these identities still hold
for the trigonometrical functions of a complex variable.
still

Example. Prove

that, if z
e"

= x-\-iy, where x and y are real,


ex (cos

y+i sin y).

The hyperbolic functions


When z is real, the hyperbolic functions

3.52.

are defined

by the

equations

e~
cosech z = 1/sinh

sinhz

We now

i(e

),

coshz
z,

sech z

l(e

= sinhz/coshz,
coth z = 1/tanh

+e~z ), tanhz

1/cosh z,

define the hyperbolic functions for all real or

z.

complex

FUNCTIONS OF A COMPLEX VARIABLE


by these equations, the symbol e* being now
meaning the function expz.

values of z

preted as

45
inter-

The hyperbolic functions of the complex variable z are all


and their derivatives have the same form as

analytic functions,

in the theory of the hyperbolic functions of a real variable. It

should be observed that sinhz and coshz are regular in every


bounded domain. On the other hand, tanhz and sechz have
singularities at the points where coshz vanishes, and cothz and
cosechz at the points where sinhz vanishes.
The equations

= sinhz,
sinh iz = sinz,
siniz

cos iz

= coshz,

coshiz=cosz,

which the reader will easily prove, are of great importance as


they enable us to deduce the properties of the hyperbolic functions from the corresponding properties of the trigonometrical
functions.

3.53.
If z

The zeros

of sinz

and cosz

= x-\-iy, where x and y are real, we have


sin z = sin x cosh y + cos x sinh y,
i

and

so sinz vanishes if

and only

sin x cosh y

if

cos x sinh y

0,

0.

1 when y is real, the first equation implies


But since coshy
mr, where n is an integer or zero.
that sin x is zero, and so x
0, and this has but
The second equation then becomes sinh 3/
one root, y
0, since sinh y increases steadily with y. We have
tin, where
thus proved that sinz vanishes if and only if z

a positive, zero, or negative integer. Moreover, by using


the addition theorem for sinz, we see that

is

sinz

z~=^-^

iN
V) re

sin(z-m7r)

z=^

which tends to ( l) n as

1V
'

,^,_ iy(?i:H^l

Z}

(2r+l)l

z -> mr.

and cosecz are


The behaviour of these functions near
exhibited by the equations

It follows that the only singularities of cotz

at the points

z = mr.

a singularity

is

(znn) cosecz -+ (1)",

(z nir)cotz ->

as z -> nit; the proof of this is left to the reader.

FUNCTIONS OF A COMPLEX VARIABLE


we can show that cosz vanishes if and only if
z = (n+ 1)tt, where n is an integer or zero. It follows
that the
only singularities of tan z and sec z are at the points z =
(ra-j-f )tt,
and that
46

Similarly

{z (m+J)7r}tanz
as z ->

1,

->

{z (w+i)77}secz

( l)"+i

-^

(n+ 1)tt.

Example

1.

Show

are at the points z


and that, as z > mri,

that the only singularities of cosechz and cothz


where n is a positive, zero or negative integer,

niri,

(zmri)coaechz->( l)\

(z- mri) cothz->

1.

Example

2. Prove that the only singularities of sechz and tanhz


are at the points z
(n+i)m, where n is a positive, zero or negative
integer, and that, as z -s- (n+$)7n,

{z-(n + J) 7n}sechz->i(-l)'+i )

The

3.54.

{z-(n + ),} tanhz ->

1.

periodicity of expz

The

real or complex number


y is said to be a period of the
function /(z) if the equation f(z+y) =/(z) holds
for all values
of z. We shall now show that expz is a

periodic function, its


periods being 2niri, where n is an integer.
For, if y
a+ifi be a period of expz, we have, by the
addition theorem,

expz
so that

expy

e a (cos^+isinj8)

= exp(z+y) = expzexpy,

1.

But

1,

this equation,

when

e<sin 8

we square and add, we find that e2 =


number is thus given by the equations

1,

ecos=l,

If

<*

cos

Hence
3.55.

written in the form

implies that

2mr, where

1,

sin

0.

and

so a

0.

The

0.

n is an integer. This proves the theorem.

The logarithmic function

When x is real and positive,

the equation e = x has one real


which is called the logarithm of x and is denoted
by
log*. When, however, z is complex, the
corresponding equation
expw
z has an infinite number of solutions,
as We shall now
show; each solution of this equation is called
a logarithm
solution,

of z.

FUNCTIONS OF A COMPLEX VARIABLE

To

find the solutions of this equation

where u and v are

this it follows that v is

and so u
thus of the form
e"

\z\

= log

write

47

= u-\-iv,

that

real, so

e u (cosv-\-ism.v)

From

we

|z[.

w=

z.

one of the values of argz, whilst

Every solution of the equation


,

is

log|z]+iargz.

Since argz has an infinite number of values, there are an infinite


number of logarithms of the complex number z, each pair differing by a multiple of 2ni.

We shall write
Logz

= log |z|+i argz,

Logz is a function with an infinite number of values


corresponding to each value of z. Each determination of Logz,
obtained by making a special choice of the many-valued funcso that

tion argz, is called a branch of the logarithm. The most important branch is the principal value of the logarithm of z, which
is obtained by giving to argz its principal value.
We shall
denote this principal value by logz, since it is identical with the
ordinary logarithm when z is real and positive.

Now

the function log|z|

but as z ->

->

is

continuous except at the origin;

oo.

Again the principal value of argz


is continuous except at points on the negative half of the real
axis; for, if x < 0, y > 0, we have
0,

log

|z|

hmarg(x+i?/)

tti,

limarg(a; iy)

ni.

if z
and z' -*-z along any path which does not cross
the negative half of the real axis, then log z' ->- log z. But if the

Hence,

path from

z'

to z crosses the negative half of the real axis once,


where the sign is
or
according as the

logz' -> log z2m,

path crosses from above to below or from below to above. Thus


we can pass from the principal value to any other branch of
Log z by making a sufficient number of circuits about the origin.
For this reason, the origin is called the branch point of Log z.
To avoid this difficulty, we now make a cut along the real
axis from
oo to 0, across which it is impossible to pass. Then,
if D is any bounded domain in this cut plane which does not
possess the origin as boundary point, log z is one-valued and

FUNCTIONS OF A COMPLEX VARIABLE

48

continuous in D. If z and z' are any two points of


and if
w and w' denote log z and log z' respectively, then w' ->w as
z' -* z along any path in D. Hence

w w
z z
as

z'

->

w w

Thus log z

z.

has the function

is

/exp w

expw;
-> l/expw> = 1/z
expw>

exp w

w w

an analytic function, regular

Example 1 Prove that

a value of the logarithm of zz'


not necessarily the principal value.

but that this

is

in

D, which

1/z for derivative.


is

log z

+ log z',

Example 2. Show that log(l z 2 ) is a regular analytic function provided that the z-plane is cut along the real axis from oo to 1 and
from +1 to +oo. Discuss how the function behaves when circuits are
made about the points 1 and 1.
3.56.

The power

series for log(l+z)

We have just seen that log(l-fz) is an analytic function which


regular in the z-plane, supposed cut along the real axis from
to 1, and which has derivative l/(l+z). In the region
\z\
1 this derivative can be expanded as a convergent power

is

oo

<

series

1 z-fz2
-,

But, by

3.33, the

sum

+...

of this power series

is

the derivative of

the function
/(z)

= z-% + *.. + (-!)<


2

'

'

2+l

n+1

'"

whenever the

Now

latter series converges.


the radius of convergence of the power series denning

/(z) is unity.

Hence
F(z)=f(z)-log(l+z)

an analytic function which

and has zero derivative


whose value
is found to be zero by putting z = 0.
We have thus shown
that log(l-j-2) can be represented by a power series
is

when

\z\

<

1.

is

regular

It follows that F(z) is a constant,

Z
,. Z n+1
= is-_z + _...
+ (_i):
S

log(l+z)

convergent when

\z\

<

1.

'

'

n+1

FUNCTIONS OF A COMPLEX VARIABLE

The function

3.57.

49

za

from the definition of Logs that, when


p is any
exp(pLogz) is a one-valued function of z
being equal to zP, even though Logz is
many-valued. If is
It follows

integer, the function

any other

integer, the function

w=

exp/^Logz]

therefore, a solution of the equation


might denote any branch of it by g?K
is,

vfl

= zP,

We

reserve this notation for the principal value


of w,

a*

The other

so that

shall,

we

however,

namely

= exp/l'logz).

solutions of the equation are then of the form

w=

zPlge ik n ilQ
'

"

where k = 1, 2, 3,..., q- 1. Obviously each of these solutions


can
be obtained from gPh by making circuits about the
branch-point
of log z.

We now define z

any

for

equation
this definition the

zzP

complex value of a, by the

= exp(alogz).
,

z<*

With

real or
,

law of indices

still

holds; for

= exp( a log z)exp(,8 logz) = exp{(a+/3)logz} = z<*+0.

We

saw in 3.55 that logz is discontinuous on the negative


half of the real axis, and that it suddenly decreases
by 2ni as

we

cross this fine

> 0,

from above to below.

then

Hence,

if

<

and

(x iy)<*
as

y->0; thus

z<*

is

when
when a is an integer.

also discontinuous

negative, except in the case

z is real

and

however,
one-valued and continuous in every bounded domain
of the
z-plane, supposed cut along the real axis from 00 to
0. Moreover, if we write the equation defining z<* in the form z<*
e<*
It

is,

where = logz, we see, by 3.31, Ex.


D, its derivative being given by

4111

dz a

dei dt

az

a dz
_,

2,

that

z<* is

regular in

FUNCTIONS OF A COMPLEX VARIABLE


Show that it is not necessarily the case that

50

Example

1.

Example

2. Determine the branches

Z?2

tion

(1-z2

1 /2
)

(!)"

and branch -points of the func-

REFERENCES
Power

series:

T. J. I'a Bbomwich, Theory of Infinite Series (London, 1926), Chap. X.


P. Dienes, The Taylor Series (Oxford, 1931), Chap. V.

The

exponential, logarithmic

and

trigonometrical functions:

G. H. Hardy, Pure Mathematics (Cambridge, 1921), Chap. X.


E. W. Hobschst, Plane Trigonometry (Cambridge, 1911), Chap. XV.
E. T. Whittakeb and G. N. Watson, Modern Analysis (Cambridge,
1920),

Appendix.

MISCELLANEOUS EXAMPLES
i.

m=

if

afly(y

ix)

7Z*

<**>.

x a +2/ 2

^)

>

prove that the increment ratio {/(z) /(0)}/z tends to zero as


any radius vector, but not as z -> in any manner.
2.

If z

along

= u+iv, where r, 9, u, v are real and/(z) is a regular


show that the Cauchy-Riemann differential equa-

re9*, f(z)

analytic function,
tions are

Su_\Sv

8v

8r~r86'

8r

~~

8u

r 8$'

3. Verify that the real and imaginary parts of log z satisfy the CauchyRiemann equations when z is not zero.
4.

Prove that the function

/(z^l+J^-Hp^zregular when |z| < 1 and that its derivative is a/(z)/(l+z). Hence
show that the derivative of (l+z)~/(z) is zero, and deduce that
is

/(z)
5.

If

is

any

(l+z).

positive or negative integer or zero,

and if

<

<

\tt,

tends to infinity as z tends to infinity in the angle


z
n tends to zero as z tends to
\rt-\-e. < argz < far e, but that \e /z
infinity in the angle ^77+e < argz < fn- e.

prove that

|e*/z"|

6.

Prove that,

if

is

a positive integer and


nlog(l +

where

\z\

3 = z+/(z)

\fn (z)\

<

i|-\

<

\n,

FUNCTIONS OF A COMPLEX VARIABLE

51

Deduce that
as

n -> co, when z lies in any bounded domain of the z-plane.


Show that the general solution of the equation z = tanio

7.

is

1+&

%z

Prove that each branch of this many-valued function


is an analytic
function, regular in the z-plane, supposed cut
along the imaginary axis
from -co* to i and from i to cot, and that the
derivative of each
such branch is 1/(1+ z 2 ).
Show also that, when \z\ < 1, the principal branch is
z3

z5

Prove that the conjugates of the complex numbers sinz


and cosz

8.

are sinz

and cosz

Isinzp
9.

respectively.

Hence show that

(cosh22/-cos2a;),

|cosz| 8

(cosh22/+cos2a;).

Prove that
sinh|Imz|
'

'

sin

<| cos z <


^

cosh(Imz).

Deduce that

|sinz|

and

either of the angles e

<

<

tend to infinity as z tends to infinity in


argz < - e> +e < aTgz < 2n - e where

|cosz|

<

7T.

10. If f(z)

= u+iv

is

an

analytic function, regular in a

where f'(z) does not vanish, prove that the curves


u
v = constant form two orthogonal families.
Verify that this

is

the case

when

(i)/(z)

z \ (ii)/(z)

domain

constant,

sinz.

CHAPTER IV

CAUCHY'S THEOREM
4.1. Rectifiable arcs
discussing Cauchy's theorem on the integration of
of a complex variable, it is desirable to confunctions
analytic
the
length of a Jordan arc, whose equation is
how
sider briefly

Befobb

= x(t)+iy{t),

< <

t
T, may be defined. Let z z 1;
t
which correspond to the values
of
this
arc
z 2 ,...,z n be the points
t
t
v t 2 ,...,tn of the parameter t, where

where

The polygonal
from

to z 1;

<t1 <t <


2

arc which

from

is

z x to z 2 ,

<

...

tn _ r

<tn =T.

obtained by drawing straight lines


so on, is of length

and

2=2

Iv- V-ll-

2 tends to a unique limit I as n - oo and the greatest of the


numbers tr tr _ x tends to zero, we say that the arc is rectifiable
and that its length is I.
It can be shown that the necessary and sufficient condition
for a Jordan arc to be rectifiable is that the sums 2 should be
bounded for all possible modes of subdivision of the range of
values of the parameter. t In the present book, we shall, in
general, be concerned only with rectifiable arcs of a more special
type, namely Jordan arcs with continuously turning tangent.
Such curves we shall call regular arcs. A regular arc is characterized by the fact that the derivatives x (t) and y(t) exist and
are continuous over the whole range of values of t.
We shall now prove that a regular arc is rectifiable by showing
If

that the sum

2 does, in fact, tend to the limit


T
\

{x 2

+y

2 112

dt.

The

first

step in the proof

is

to

show that

can be made to

t Proofs of this are given, for example, by P. Dienes, The Taylor Series
(Oxford, 1931), 199-201; E. W. Hobson, Functions of a Real Variable (Cambridge, 1921), 1, 318-20; E. G. Phillips, Course of Analysis (Cambridge, 1930),
208-10.

CAUCHY'S THEOREM
differ

by

as

little

as

we

please from the


n

53

sum

1^2Wr)+iy{tr)\(t-tr-x),
r=l

by making the greatest of the numbers tr

We

sufficiently small.

r_x

start with the inequality

|{|Z|-|Z'|}|< \Z-Z'\
of

which obviously implies that

1.32,

<

\{\Z\-\Z'\}\

\m(Z-Z')\ + \Im(Z-Z')\.

Hence we have

Ir=\{

<

\z-Zr-l\

\*(tr )

+ iy(tr)\(t -tr-l)
r

\(Xr-%r-l)-x(tr)(tr-tr-l)\+\(yr-yr-l)-y(Wr-tr-l)l

where xr denotes x(tr ), and so on.


Now, by the mean- value theorem of the
xr xr _ x

where rT and

Ir

= x{r

r )(tr

<

t^j),

he between

r'r

tr _ x

differential calculus,

yr yr -i = y{r'r){tT -tr - x ),
and tr and so
,

(t-tr-i){\A{rr )-x{tr )\

+ \y{r' )-y{t

r )\}.

But, by hypothesis, the functions x(t) and y(t) are continuous


and hence also uniformly continuous. We can therefore assign
arbitrarily a positive number e, as small as we please, and then
choose another positive number , depending only on e, such
that
\x(t)-x(t')\

<

\m-y{t')\<e

e,

whenever \t 1'\ < 8. If the greatest of the numbers


less than S, we find that

Ir

<

tT _ x

is

Htr-tr-l),

and hence that


n

1 ~ 2x < 1 ^ <

2e

2 {t-t^) =

2e(T-t

).

by the definition of the integral of a continuous funcsum ~ tends to the limit

Finally,
tion, the

T
j
as

n tends

\x(t)+iy(t)\ dt

to infinity

JV + ^i,2

dfy

and the greatest of the numbers

tends to zero. Since, however,


2i can
as we please by making 8 sufficiently small,
|

*>e

tr

tr _ 1

ma de as small

2 must also tend

CAUCHY'S THEOREM
We have thus shown that a regular arc

54

to the same limit.

and that

rectifiable

its

length

is

is

T
2
2 112
j {x +y }

dt.

to

4.11. Contours
If

AB

and

respectively,

BG

are

two

rectifiable arcs of lengths

which have only the point

in

and V

common, the

arc

evidently also rectifiable, its length being l+l'. From


Jordan arc which consists of a finite number
of regular arcs is rectifiable, its length being the sum of the
lengths of the regular arcs forming it. Such an arc we call

AG

is

this it follows that a

a contour.
By a closed contour we mean a simple closed Jordan curve
which consists of a finite number of regular arcs. Obviously
a closed contour is rectifiable.

Riemann's definition of integration


The fundamental operation of the calculus known as integration arises in two distinct ways. The determination of the
4.12.

indefinite integral

may

be regarded as the operation inverse to

in the applications of the calculus


to geometry or physics, it is the definite integral, defined as the
limit of a sum, which is of importance. In the theory of func-

that of differentiation.

tions of a

But

complex variable we

start

with the definition of an

integral as the limit of a sum, and, later on, deduce the connexion between the operations of differentiation and integration.

Although this definition of the definite integral had its origin


work of the Greek mathematicians, it only attained a

in the

precise arithmetical form, satisfying

modern standards of rigour,

We

shall
in the nineteenth century at the hands of Biemann.f
now explain Riemann's definition of an integral, not, indeed, in

form, but in one more suited for our present purposes.


a function f(z) of the complex variable z,
consider
Let us
analytic but which has a definite finite
necessarily
which is not
rectifiable arc L. Let the equation of
a
of
point
each
value at
subdivide this
T.
t
where t
x(t)+iy(t),
this arc be z

its original

^ <

We

trigonometric series. This


t In his inaugural address of 1854 on
in his collected works (German edition (1876), 213-51).

is

reprinted

CAUCHY'S THEOREM
arc into

smaller arcs

by the points

55

z v z 2 ,..., zn _1 , z n

(=

Z),

which correspond to the values


*0

of the parameter

"^
t,

*1 *~-

*n-l "^ *n

and then form the sum


n

2 /(W(V-ar-i).
2 = r=l
where

r is

a point of

between

and

zr _ x

zr

tends to a unique limit J as to tends to infinity


and the greatest of the numbers tr tr _ r tends to zero, we say
that/(z) is integrable\ from z to Z along the arc L, and we write
If this

sum

= jf(z) dz.
L

The

direction of integration

affix x(t)+iy(t) describes

is

from

the arc

to Z, since the point of

L in this sense when

increases.

Thus the numbers Z, z play in this theory much the same parts
as the upper and lower limits in the definite integral of a function of a real variable. Nevertheless, we do not write
z

= jf(z)dz;

we shall soon see, the value of J depends, in general, not


only on the initial and final points of the arc L but also on its
actual form.
The complex integral of a function /(z) along a rectifiable arc

for, as

L, defined in this way, exists under quite general conditions,


a sufficient, but not necessary, condition being that f(z) should
be continuous on L. There is no need whatever to assume that
the derivatives x(t) and y(t) exist. We shall not, however, prove
this general result, J but shall only consider in detail the case
when L is a contour.

Example

1.

If

j"

dz

is

any

rectifiable arc joining

the points

and Z,

prove that

= Z-z

jzdz =

%{Z*-zl).

f In particular, if L is a segment of the real axis, this definition reduces to


the ordinary definition of the integral of a bounded function of a real variable.
% For a proof of this, see, for example, Watson, Complex Integration and
Cauchy'a Theorem (Cambridge, 1914), 17-25.

CAUCHY'S THEOREM

56

Both
on L.

integrals exist since the integrand

In the

first case,

in each case, continuous

is,

we have

2= /(r)(Zr-Zr-l) = 2= ( r_i) = Z-Z


r

from which the

a,

result stated follows at once.

In the second, we have to find the limit of


n
2, ir\ zr

This

is

Zi

and

so

zr-l)-

r=l
necessarily the same as the limits of

is

also the

2. zA zr

r=i
same

*(2x

zr-l)>

Zr

~ 2 zr-l\ zr zrl)>
r=l

+2

as the limit of ( 2,

2.)

= ii
(4-4-i) =
=
r

Hence

izdz

)*

But

2
i(^ -^).

"

i(Z 2 zl).

L
be observed that in both these examples the value of the
independent of the path from z to Z.

It should
integral

is

Example
and

BC

2. If f(z)

is

integrable along the

which have only the point

integrable along the arc

in

jf(z)dz= Jf(z)dz
AB

AB

f(z) is

jf(z)dz.
BC

Example 3. If f(z) is integrable along L,


described in the opposite sense, show that

Jf(z)dz=L'

rectifiable arcs

AC and that

AC

Example 4. If /(z)
constants, prove that

two

common, prove that

and

W( z + bg( z
)

jf(z)
L

and

)}

dz

4.13. Integration along

a J7(z) dz
L

L'

is

the same arc

dz.

g(z) are integrable

if

along L, and a and b are

+b j g(z) dz.
L

a regular arc

Let us suppose that f(z) is continuous on the regular arc L


whose equation is z = x(t)-\-iy{t), where t ^.t^.T. We prove
that f(z)

is

integrable along

j f(z)

dz

L and that

T
j F(t){x(t)+iy(t)} dt,

CAUCHY'S THEOREM

57

where F(t) denotes the value of/(z) at the point of L of parameter t.


In the notation of 4.12 we have to consider the sum

2/0^-^-1),
2 = r=X
where r is a point of the arc between zr _ x and zr If rr is the
parameter of ,., tt obviously lies between tr _ x and tr Writing
are real, we find that
F(t) = <f>(t)-\-itp(t), where j> and
.

tfi

2 = r=l
2 ^(T )(av av_i)+r=l
2 0(T )(av av_i) +
r

+% 2 4>(rr )(yr -yr -i)r=l

We

consider these four

By

separately.

the mean- value theorem of the differential calculus the

term

first

sums

2 ^(T )(y -y _i).

is

= 2 ^(T,)(aV a*-l) = 2 ^M^(T

2i

r)

r=l

ft- <*-l).

where r'r lies between r _ x and tr The first step is to show that
2 X can be made to differ by as little as we please from
.

j,=i^)i((H-i)
r=l
by making the greatest of the numbers tr tr _x sufficiently small.
Now, by hypothesis, the functions (f>{t) and x(t) are continuous.
As continuous functions are necessarily bounded, there exists
a positive number K such that the inequalities \<f>{t)\ < K,
\x(t) < K, hold when t < t < T. Moreover, the functions are
|

also uniformly continuous;

a positive number
positive

number

e,

8,

\tt'\

is less

than

<
8,

can, therefore, assign arbitrarily

we

Hence,
we have
8.

<
if

and then choose a


such that

please,

depending only on

\<f>(t)-<}>(t')\

when

we

as small as

e,

\x{t)-X{t')\

6,

<

the greatest of the numbers

t
T

tT _ x

\<j>(Tr )x(r'
r )-<f,(tr )x(tr )\

=
<

\</>(rr

){x(^)-x(tr )}+x(tr ){</>(rr )-4(tr )}\

\<f,(rr )

\x(r'r )-x(tr )

+ |x(

r)

|#Tr )-#g

< 2#e

CAUCHY'S THEOREM

58

and

-2

therefore

\I1

By

\<2Ke(T-t

).

the definition of the integral of a continuous function of


tends to the limit
2

a real variable,

T
j.

<f>{t)x{t)

dt

as n tends to infinity and the greatest of the numbers t


r
r _x
tends to zero. Since, however,
~"
can
De
ma
de
as
small
2i
2
as we please by taking 8 small enough,
must also tend to
the same limit. Similarly the other terms of
tend- to

limits.

Combining these

we

results

that

find

tends to the

limit

T
j

T
(<f>-<f>y)

dt

+i j

to

and

T
(tpx+<f,y) dt

F(t){x(t)+iy{t)}

dt,

J*

t%

to

so f(z) is integrable along the regular arc L.

This result

is

not merely of theoretical importance as an

existence theorem. It

is also of practical use in that it reduces


the problem of evaluating a complex integral to the integration
of two real continuous functions of a real variable.

More

generally, it can be shown without difficulty that, if


continuous on a contour C, it is integrable along C, the
value of its integral being the sum of the integrals of f(z) along
the regular arcs of which C is composed.

f(z) is

Example

1. Find the value of the integral of l/( a) round the


whose equation is \z a\ = JR.
The parametric equation of G is z = a+ Boost + iBsiat, where t
varies from
to 2n as z describes C once in the positive sense. Hence
circle O,

27T

Rd cos + iRD sm
.

( Msint + iBcost)dt

'

o
.TT

dt

2m.

Example
from

2.

1 to +1

Show that the integral of 1/z along a semicircular arc


has the value
or
according as the arc lies above

or below the real axis.

CAUCHY'S THEOREM

The absolute value

4.14.

of a

59

complex integral

C of length

Iff(z) is continuous on a contour


the inequality |/(z)|
M, then

I,

where

it

satisfies

<

jf(z)dz sCMl.

In proving

this

when C

case

Now

if

variable

t,

is

<f>(t)

theorem we evidently only need to consider the


a regular arc.

is

any complex continuous function of the

tmitr-tr-lti
and

so,

real

we have

on proceeding to the
T

<I

mtr)\(t-tr-l),

limit,

<

j<j>(t)dt
to

f \<f>(t)\dt.
to

In the notation of 4.13,/(z)

F(t)

on G, and so

\F(t)
\

<M

Hence
JF(t)(+iy)dt

jf(z)dz

<

<M
4.2.

T
f (x2

2 1 '2
2/

dt

il^?.

Cauchy's theorem

Let /(z) be an analytic function, regular in a domain D. Let


z and z x be two points of D, joined by a rectifiable arc L, every
point of which belongs to D. Then the integral of /(z) along L
certainly exists since f(z) is continuous on L. The fundamental
property on which the theory of analytic functions depends is
that the value of this integral is a function of z and z x alone
and is quite independent of the particular arc L which joins the

two given

An

points.

equivalent form of this result

is

Cauchy's theorem, which

G is a simple closed rectifiable arc lying in D, then


the integral of f(z) round G is zero. For any two points of G
divide it into two rectifiable arcs L and L v say. If L 2 denotes
states that, if

CAUCHY'S THEOREM

60

the arc

along

Lr

described in the opposite sense, the integrals of /(z)


2 are equal. Hence we have, by 4.12, Exx. 2, 3,

L and L

jf(z) dz

jf(z) dz

]*/(*)

dz

("/()

dz

(f(z) dz

= 0.

In the sequel we

shall always suppose that a simple closed


described in the positive or counter-clockwise
sense, unless the contrary is explicitly stated.

rectifiable

It

is

curve

very

is

prove Cauchy's theorem in its most


In the next section we make the additional

difficult to

general form.

assumption that the derivative f{z) is continuous within and


on C and follow the lines of Cauchy's second proof.f We then
show that this additional assumption is not necessary when the
contour is a polygon, and, finally, we indicate briefly how the
extension to the general case

4.21.

may

The elementary proof

be carried out.

of Cauchy's

theorem

We now

prove the simplest and original form of Cauchy's


theorem, that if f{z) is an analytic function whose derivative
f'(z)
exists and is continuous at each point within and on the closed
contour C, then

jf(z) dz

0.

Let D be the closed domain which consists of all points within


C. If we write z = x+iy, f(z)
u+iv, where x, y, u, v

and on

are real,

we

have,

j f(z) dz
a

4.13,

= j (u dx v dy)+i j (v dx +u dy),
'

where, by J
the form

by

(Pdx+Qdy), we mean

c
the

sum

of the integrals of

(Px+Qy)

dt over all the regular arcs composing C.


transform each of these curvilinear integrals by
means of Green's theorem, J which states that if P(x,
Q(x,
f

We now

y),

y),

t Comptes Rendus, 23 (1846), 251-5. This proof is often called Riemann's


proof; he gave it in his inaugural dissertation at Gottingen in 1851 (Ges. Werke
(1876), 3-46).

Cauchy's first proof depended on the calculus of variations. It occurs in his


'Memoire sur les integrates definies, prises entre limites imaginaires' (Paris,
1825), which has been reprinted in Bulletin des sci. math. 7 (1874), 265-304,
8 (1875), 43-55, 148-59.
} See Gibson, Advanced Calculus (London, 1931), 335-6, or Phillips, Course
of Analysis (Cambridge, 1930), 290-1.

THEOREM

CATTCHY'S

61

8Q/8x, dPjdy are continuous functions of both variables


in D,

then

By

hypothesis

f'(z) exists

ever,

/ (z)

we

and

is

continuous in D. Since, how-

= ux +tvx = vy iuv
,

assuming that u and v and their partial derivatives ux vx u v are continuous functions of both variables
y
x
y
and y in D. The conditions of Green's theorem are thus
are, in fact,
,

satisfied.

Hence we

see that

o,

by the Cauchy-Riemann

differential equations. This completes

the proof of Cauchy's theorem.


4.22.

The general form

of Cauchy's

theorem

Although the simple form of Cauchy's theorem which we have


just proved suffices for many of the applications we have in
view,

it is desirable to consider whether the assumptions


we
have made are necessary for the truth of the theorem. The first
step in this direction was taken by Goursat,f who showed that
it is unnecessary to assume the continuity of
f'(z), and that
Cauchy's theorem is true if it is only assumed that f(z) exists
at each point within or on the simple closed rectifiable curve G.

Actually the continuity of the derivative /'(z) and, indeed,


differentiability are consequences of Cauchy's theorem.

More generally

still, it

can be shown that

iff(z) is

its

an analyW}^l ^0

function, continuous within and on the, simple, closed rectifiable


curve C, and iff'(z) exists at each point within C, then

ff(z)dz
a

= 0.

We

shall now indicate briefly the lines of the proof of this


general form of Cauchy's theorem.
t Trans. American Math. Soc. 1 (1900), 14-16. Goursat's proof
found in his Cours d' Analyse, 2 (1918), 74r-8.

will also

be

CAUCHY'S THEOREM

62

The

first

step

is

to

show

that, if f(z)

is

an analytic function

whose derivative f'(z) exists at each point within and on a triangular contour C, the integral of f(z) round C vanishes. Let
us write

f(z) dz

= h,

o
so that h

> 0.

The proof

consists in showing that

we

are led

>

0.
to a contradiction if we assume that h
0. If we join the middle points
Let us suppose, then, that h

>

of the sides of

C by straight lines, the domain within C is divided

into four congruent triangles


say.

Then

whose boundaries are yv y2 y3 y4


,

jf(z)dz=2
since,

on the right-hand

side,

we

j f(z)dz,
are integrating twice in oppo-

site directions over each side of a triangle yr which is not part


of a side of C, and the corresponding integrals cancel. Hence

we have

A<2
r=l

J>)

dz

y.

so that the inequality

jf(z)dz

>lh

V,

one value of r; if it holds for more than


the least. In this way we obtain a triangular contour, Cv say, of half the linear dimensions of C,
with the property that

must hold

for at least

one value of

r,

we take

jf(z)dz

>&

We now treat Cx in

the same manner, and so on indefinitely.


Proceeding thus, we obtain a sequence of triangles C, Cv C2 ,...,
C ,..., each of which is contained in and has half the linear
n

dimensions of its predecessor, with the property that

jf(z)dz >A/4

By

the example of 2.2 there is precisely one point, a say,


which lies within or on every triangle of this sequence. More-

CAUCHY'S THEOKEM

63

by hypothesis, f(z) is regular at a. Hence, given any


positive number e, we can find a neighbourhood of
a, whose
over,

radius depends on

e,

in which the inequality

<

!/(*)/() ( Xr()
e\z-a\
holds. This neighbourhood of a contains all the triangles
of the
sequence for which n
N, where
is an integer depending
J

>

It follows that, if l
n
inequality
e.

/(z)
|

by the

is satisfied

is

We know,

(2

affix z of

it

jf(z) dz

0,

zdz

provided that

= 0,

= j {/()-/()-(*-).f ()} dz.

{{M-f(oc)-(z-)f'(oc)}dz

< ell = J /**


2

1
,

the perimeter of C.
have thus shown that, when
is

< h < 4
This

> JV, we have, by 4.14,

jf(z)dz

We

Gn

every point on

Hence, when n

the

follows that

where

however, thatf

j dz
from which

_/(a) _ _ a)f (a) < ^

> N.

on

the perimeter of the triangle

> N,

dz

<el2

however, impossible, since e is arbitrary; the assumption


positive is thus untenable. But since h
0, we must,
therefore, have h
0. This completes the proof} of Cauchy's
theorem for a triangle within and on which the integrand is
is,

that h

>

is

regular.

The next step is


contour.

any

We

to extend this result to the case of a polygonal

shall

make

use of the fact that the interior of

closed polygon can be divided

See 4.12, Ex.

The proof

499-506.

is

up

into a finite

number of

1.

due to E. H. Moore, Trans. American Math. Soc.

X (1900).

CATJCHY'S

64

triangles.

For,

if

the polygon

THEOREM

is

not convex,

it

can be divided

up into convex polygons by producing the sides sufficiently, and


any convex polygon can be divided into triangles by joining any
interior point to the vertices.

Let us suppose that/(z) is an analytic function regular within


and on a closed polygon C. If C is now divided up into n triangles whose boundaries are Gv C2 ,..., Cn we obtain
,

jf(z)dz

=f

jf(z)dz,

side, we are integrating twice in oppoover each side of a triangle Cr which is not also
part of a side of C, and the corresponding integrals cancel. But
we have shown that the integral of f(z) round each triangular
contour Cr vanishes, and so

since,

on the right-hand

site directions

jf(z)dz=0.
o
We have now reached the really difficult stage in the proof
of Cauchy's theorem in the general form enunciated at the
beginning of this section. We are given that f(z) is regular
within the simple closed rectifiable curve C and continuous
within and on C. These conditions imply that J/(z) dz exists;
c

we have

to

show that

it is zero.

can be proved that, when any positive number e is assigned,


a closed polygon P can be constructed, each point of which lies
within C and at a distance less than e from C. The derivative
f'(z) exists at every point within or on P, so that the integral
of/(z) round P vanishes. The general form of Cauchy's theorem
would, then, be completely proved if we could show that
It

lim

The

/()&=

\f{z)dz.

validity of this passage to the limit

immediately sugit is one of the


most difficult things to prove in the whole of the theory of
functions of a complex variable. A proof has been given by
Pollard which depends on 'some rather delicate theorems of
de la Vallee Poussin, the development of which requires great
care and unusual nicety of thought'. We shall, therefore, con-

gested

by geometrical

intuition.

is

Nevertheless,

CATJCHY'S
tent^ourselves

THEOREM

by relying on geometrical intuition

65

for the validity

of this passage to the limit,


and refer the reader to the o^gmai
original
memoirst for the proof.
4.3.

The deformation

7"

of contours

^g *

n fthe
1
an analytic function round
a !5f
a
closed contour, within which
the function is not necessarily
Can
n be -^PH&d by noticing that

the value of the


contour megral of an analytic
function is unaltered by deformation
of the contour provided that the
contour crosses no singularity
of
J
the integrand during
deformation.

Fig. 1

For simplicity we

shall only consider

d f Prf iS

T^
C are

Tow Vw"-^
show
that jf Ox and
pletely inside C then
v

ff(z)dz

a particular case of this

quite

""all
8eneral
two closed contours, C lying
com .
2

ff(z)dz,

provided thatf(z) is continuous in 'the


closed annulus and regular
in the open annulus bounded by C
and
C2
t
Let us take two points of affixes a
and 6 on Cx and two points
of affixes c and d on C
If we join a to c and b to
2
d by two
potygonal arcs which have no point in
common and which do
not cross
or C2 we form two closed
contours L and L , as
(J figure.
2
shown in the
Now/(z ) is continuous in the closed domain
and regular
the open domain bounded by
Lx and so its
.

f Pollard, Proc. London Math. Soc. (2)21(1923) 456-82- 2 noosi iak * n

^ Watson, Co^,^^^^

4111

CAUCHY'S THEOREM

66

integral

round

L x is

zero; similarly for

J7(z)

Hence

& + J7(s) & =

0.

and L 2 each contain the polygonal arc ac described


senses, so that the two integrals along the arc
opposite
twice in
for the polygonal arc bd. Hence this last
similarly
ac cancel;

But

Lx

equation takes the form

jf(z)dz- jf(z)dz
when we take

0,

into consideration the conventional sense of

description of the contour

C2

4.31. Cauchy's integral

formula

We now show that i//(z) is an analytic function, regular within


a closed contour C and, continuous within and on C, and
any point within C, then
JV

'

2iri

if

is

za

c
Since f(z)

is

regular at a,
f{)

we have

=/(o)+(-o)f(o)+(2-o)l.

where rj is a function of z and a which tends to zero as z -> a.


Hence, given any positive number e, we can find a neighbour8 in which the inequality |ij| < e holds.
hood \za\
Now draw a circle y with centre a and radius r, where r is
less than 8 and is also so small that y lies entirely within G.
Then, since f(z)/(za) is regular in the annulus bounded by y
and C, we have, by the theorem of 4.3,

<

za

J
7

za

27rt/(a)+

r]

dz.

CATJCHY'S

From

THEOKEM

fi7

we deduce that

this

since
this

<

on y. The expression* the


right-hand side of
mequahty tends to zero with r.
The expression on the left
|,|

- '* - -tte

s^vir^'

quite independ

This completes the proof


of Cauchy

2tuJ

'

s integral

formula

which expresses the value of


an analytic function at a
point
closed contour in terms
of its values on the contoC
Example. Let Cx and C be two
closed contour O I ,
2
Plete, y w lthin Cl and let a
be a point between

withm a

C^t sC ZT

srrr st~j t -4.32

The

If it were

tMS frmUla iS

JI7 ^
to^iTsH^f

** - the

derivatives of an analytic
function
known that the process of differentiating

Sh W

lies

Under the c -^ons of

6 8hOTteSt diBtenoe fr

Cauchv's

m to

^ m-

< S the Pint +* also


<\ ^
cZ^/:^
at a distance not
than S from G

wfthin O
withm

'*'

'

less

-By Cauchy's integral


formula,

-^pM = {

we have

2J

f(z)

)*

(- a -A)(g_) az

'

CAUCHY'S THEOBEM

68

and so
l_

/<+*)-/()
h

f.A'Lfc

2id) {z-af

tort

Hence

_ um /(+*)-/<"> = J_

f(a)

dz.

(zah){za)*

J>

a
/()

provided that

J
is

is

dz

-h)(z-af

(a

bounded as fe tends to zero.


Now/(z) is continuous on C, so that an inequahty
satisfied there. Hence

when

z is

J
<c
Z

is

48 3

on C, and so

Ml

/(g)

where

<M

M_

<

(a a h)(z a) 2

\f(z)\

(2 a A)(z af

dz

48 3

the length of C. This completes the proof of Cauchy's

integral formula for /'(a).

The

result

proved has the very remarkable conitself regular within C. To prove this, we

we have

just

sequence that f'(z) is


have to show that the derivative /"(z) of the function /'(z) exists
at each point a within C.
Formal differentiation under the sign of integration gives
J

Accordingly

we

'

tti

(zaf

consider

f(a+h)-f(a) .1

f
TrtJ

/<*)

fJ

tori J

dz

(z a)
1

\{z-a-hf

{z-af

2iri

JK) (z-a-KY(z-af

\dz
(z-arj h
2h

But

it is easily

69
CAUCHY'S THEOREM
by using the same type of argument as
expression tends to zero with h. Hence

seen

before that this last

f(a)

= hm-

exists

and

is

hr->o

given by

2ni] {zaf

c
In a similar manner we can show that

by

O, its derivative f'"(z) being given

r{a)

*L (

f"(z) is regular

within

the formula

JVL*,

any point within C. And so on indefinitely. Hence


if f(z) is an analytic function regular within a closed contour C
and continuous within and on C, it possesses derivatives of all

when a

is

orders which are regular within C, the nth derivative being given by
/<*)()

f{z

-^L

dz.

Example. The function /() is regular within a closed contour C.


Show that, if s = x+iy, the functions log |/(z)|, arg/(2), Rl/(z), Im/(2)
all satisfy

Laplace's equation

dW SW _
+ 8y
dx*
a

(g-2 + ^)

Prove also that

|/|

4|/?.

4.33. Cauchy's inequalities


Let f(z) be

an

analytic function regular within

a and radius B. Then


everywhere on C,
centre

We have proved that


/W
(q) =
J
v

-m

circle

<M

of

holds

Hi. f

2m

if the inequality \f(z)\

{za) n+lx

dz.

< MjBn +

and the length of C


by 4.14.

On

C, \f(z)l(za) n ^\

The

result stated follows immediately

Example. Prove
have a true

1
,

is

2nR.

that the modulus of an analytic function cannot

maximum at a point a if it is regular in a neighbourhood of a.

CAUCHY'S THEOREM

70

4.34. Liouville's

An

theorem on integral functions

analytic function which is regular in every finite region


is called an integral function.

of the z-plane

!f f(z)
l/()l

an integral function which satisfies the inequality


for all values of z,
being a constant, then f(z) is

is

^M

constant.

This theoremf follows at once from Cauchy's inequality concerning the derivative of an analytic function. For, if
a is
any point of the z-plane, f{z) is regular when \za\
R, no
matter how large
may be, and satisfies there the inequality

<

\f(z)\

< M.

Hence we have

L/l<.
R

Making

however, a

tend to
is

we find that f'(a) is zero. Since,


we have thus shown that the derivative

infinity,

arbitrary,

of/(z) vanishes everywhere.

The converse

4.4.

Hence /(z)

of Cauchy's

is

a constant.

theorem

Let f(z) be a one-valued function, continuous


within a closed
contour C. In order that the integral
of f(z) along any contour
within C may depend only on the affixes
of the end-points
that
of

contour,

it is

both necessary

and

sufficient that f(z) be

an

analytic

function, regular within G.

The condition is evidently sufficient. For the difference between the integrals of f(z) along two different contours,
which
lie within C and have the same end-points,
is equal to the
integral of/(z) round a closed contour within C, and
so vanishes.
To show that the condition is also necessary, we consider
the

integral of f(z) along a path within C from a


fixed point a to
a variable point z. Since, by hypothesis, the value
of this
integral is independent of the path, it is a one-valued

function

of z;

let

us denote

it

by
F{z)

= jf{z)dz.
a

We

shall

show that F{z)

is

an analytic function, regular with-

in C.
t x * was given
to Cauehy.

by

Liouville in lectures in 1847, but seems to be really


*

due

CAUCHY'S THEOREM

71

Let b be any point within G. If the shortest distance of b


from C is 28, every point z for which \zb\
8 certainly lies
within C. Hence, if |A|
8,

<

<

b+h

F(b+h)-F(b)

f(z)dz,

where the path of integration

From

is

now taken

to be a straight line.

this it follows that

b+h

F(b+h)-F(b)
-f(b)

= lj{m-f(b)}dz.
6

We

Now, by

hypothesis, f(z) is continuous at 6.


can, therefore, assign arbitrarily a positive number e, and then choose
a positive number ij, depending on e, such that
1/00-/(6)1

when

<

\zb\

-q.

<

Hence we have, by

4.14,

F(b+h)-F{b)
-f(b)

<e,

<

provided that \h\


77, and so F'(b) exists and is equal to/(fe).
Since, however, b was any point within G, we have thus

shown that

F(z)

is

regular inside

f(z). This, however, implies,

within C.

by

and that

its

derivative

is

4.32, that/(z) is itself regular

This completes the proof of the necessity of the

condition.

The theorem we have


really a converse of

just proved is due to Morera.f It is


Cauchy's theorem, and is more usually

stated in the following form.

If f{z)

and

is

continuous and one-valued within a closed contour

if

jf(z)dz
r
for every closed contour
tion, regular

within C, then f(z)

is

an

analytic func-

within G.

we have proved incidentally that


within a closed contour G, the integral

It should be observed that


if f{z) is regular

\f(z)dz,
a

f Rendiconti del

It. 1st.

Lombardo, 19 (1886), 304-7.

72

CAUCHY'S THEOREM

taken along any contour within C, is an analytic


function, regular
within C, and has derivative f(z). This is the
fundamental

theorem of the calculus of analytic functions; it asserts that


the
operations of integration and differentiation are inverse
operations.

Example. Let/(z), g(z) be


Show that

analytic functions, regular within a closed

contour C.

ff(z)9'(z) dz

= f(z)g(z)-f(a)g(a)-

\ f(z)g(z) dz,

when

integration

4.5. Taylor's

is

along any contour within C.

theorem

We saw in 3.33 that the sum of a power series with nonzero radius of convergence is an analytic function,
regular within
the circle of convergence.
now prove the converse theorem,
that iff(z) is an analytic function regular in a
neighbourhood of

We

a,

it is

expansible as a power series of the form

whose radius of convergence

By

hypothesis, there exists a positive


is regular when \z-a\
positive number less than R, and let
x

on the

number

with the
x be any

< R. Let R
R = HR+RJ,
2

ft.

Then

is not zero.

property that/(z)

<R <R <

Y an (za) n

f(z) is certainly regular

C whose equation is \za\ = R2


a+h be any point of the region \za\

so that,

within

and

circle

as

Now let
a+h is within the circle

formula

C,

<R

x.

Then,

we find, by using Cauchy's integral

4.31), that

f(a+h)

2m Jf-^-dz
zah

-a/Hi
hn

A2

+
(z-af {z-af + '" +
A+i

+ z - a )n+i-r (z_o)+i(z_o-A)/
7
(

If

we now

use Cauchy's formula for the derivative of


we obtain

analytic function,

f(a+h) =/()+

2/w ()^+^>

an

CAUCHY'S THEOREM

An

where

=,

*^

73

/(*)

dz
f
2*1 J (z-o)+i(z-a-A)

'

<7

But
there.
|/(2)|

|/(2)| is

and

so

^M

there exists a positive number


on C. Moreover, when \za\

=R

Using the result of

Since, however,

An

is

bounded

such that

2,

\z-a-h\

that

continuous on the circle

Hence

\h\

|{|-o|-|A|}|

4.14,

<

we

^<

tends to zero as

> J?,-^.

find that

i? a , it follows

n tends

from

to infinity

this

and

inequahty

therefore that

f(a+h) =/()+ JT /f">(o)^.


It is, however, possible to prove
rather more than the mere
convergence of this power series. Since |A|
we

<R

x,

\*n\<

R^-R^rJ

have

'

the expression on the right-hand side


of the inequahty being
independent of A. Hence, given any positive
number e, we can
choose an integer N, depending on e but
quite independent of
h, such that \A
e when n
N. We express this property
n
by saymg that the power series converges
uniformlyt with
\

respect to

h when

<

\h\

>

^ Rv

We have now proved that iffa) is an analytic function, regular

<

the neighbourhood \z-a\


It of the point z
a, it can be
expressed tn that neighbourhood as a
convergent power series of

the form

f(z)=f(a)+ffnKa) t=^.
*i

n=l
This expansion

n\

<R

uniformly convergent when \za\


x proR. This result is known as Taylor's theorem
concerning analytic functions of a complex variable.
vided that

R1 <

is

t For an account of the theory of uniform convergence, see Chapter V.

CAUCHY'S THEOREM

74

Example
is

Prove that, when

1.

equal to

\z[

<

1,

the principal value of (1+z)"

Example

Show

2.

V "(g-lMcx-w+l) ^

that,

when

|z|

<

1,

z2

lOg(l+Z)

= Z--+ zg -....

4.51. Zeros

an analytic function which vanishes when z = a and


that a is
is regular in a neighbourhood \za\ < B of a, we say
a zero of f{z). By Taylor's theorem we can expand f(z) as a
power series of the form
If /(z) is

M=l
neighbourhood of a and has no
given
which converges in the
non-zero coefficient in this
first
constant term. If am is the
expansion, we say that a is a zero of order m.
Let us suppose, then, that f{z) has a zero of order m at a.

We can therefore write

f(z)

where

is

<j>(z)

{z-ar

regular

am+n {z-ay

when \za\

<B

(s-a)W),

and does not vanish

= a. We now show that there exists

a neighbourhood
of the point a which contains no other zero of f(z).
that
2c, it follows from the continuity of <}>{z)
For if <f>(a)

when

< 8 in which
|fla)-fla)| <

there exists a region \za\

\c\.

This implies that


|#3)|

>

{ |^(0)

|^(*) ^()

>M

<

there.
8, and so <f>(z) certainly does not vanish
(za) m(f>(z), we have thus shown that the only
Since f(z)
8 is the given
point at which f(z) vanishes in the region \z a\

when \za\

<

zero z = a.

From

this it follows that iff(z) is

having

an

analytic function regular

if z v z 2 ,..., z n> ... is a sequence of zeros off(z)


point
an interior point a of D, thenf(z) vanishes
limiting
as

in a domain D, and
identically in

D.

CAUCHY'S THEOREM

For

75

a continuous function having zeros zn as near


a as we please, f(a) must be zero. Moreover, as f(z) is regular
in the domain D, of which a is an interior point, we can expand
/(z) as a power series
since /(z)

is

/(*)

=i

(*-)*,

converging in a certain neighbourhood of a. Either f(z) is


is a first coefficient, am say, in this
power series which is not zero. But if the latter is the case, we
have just seen that there is a neighbourhood of a which contains
no zero other than a, and this is contrary to the hypothesis that
a is a limiting point of the sequence of zeros z z ,..., z ,...

identically zero, or else there

Hence f(z)

is

identically zero.

4.52. Laurent's

theorem

Let us consider an analytic function f(z) which is regular in


the annulus
\za\
but not regular everywhere in
\za\
R'. Although such a function cannot be expanded as
a power series in z a, it can be expressed, as we now show,
as the sum of two series of the form

R<

<

<

/(*)

= 2 (-)"+ S b n (z-a)-,
o

each series being convergent in the annulus.


Let R x R be any two positive numbers

R<R

<R' <R'

and

let i? 2

\(R+RJ, R'2

such

that

= i(iJ'+JK{).

Then/(z) is certainly regular in the closed annulus bounded by


the circles C,
whose equations are \za\
R2 \za\ R'2

respectively.

Now let a+h be any point of the annulus Rx


\za\
R'^
Then, as a+h lies between the circles C and C", we have, by
the example of 4.31,

<

2rn J

zah

2tti

f
J

zah

Just as in the proof of Taylor's theorem,

2tti
ni

<

f( z ^

zah

dz

we

= fan hn
o

easily

show that


CAUCHY'S THEOBEM

76

where

= -

an

2iri

(za)^1

dz.

<

This series converges uniformly when \h\


R x We cannot,
however, write / (m) (a)/! for an since f(z) is not regular everywhere within
,

Similarly

we have
ah

J z

277^

a
"'

Zhr
where

br

-x

f(z)(za) r

2m

dz

27riJ h n (zah)

But

Jf denotes the greatest value, necessarily


on C, we have
if

MR

Thus n tends to
respect to h when

MR

>
(?*Y <
^
\h\-R^
Aim)
zero as n tends to

\BJ<

\h\

2iti

r
J

/(*)

zah

<

Changing the notation

R'x

infinity,

,-y hnn

uniformly with

'

Z-i

n=1

c-

Rx ^

|/(z)|

and so

1}

the series being uniformly convergent


thus proved that

provided that

of

b.-rM*Y

^R

_j_

finite,

when

\h\

^ R v We have

slightly, this result,

which

is

known

as Laurent's theorem,^ can be expressed in the following form.

If f(z)

is

an

analytic function, regular in the open annulus

f It was published by Laurent in Comptes Rendus, 17 (1843), 348-9.

THEOREM

CATJCHY'S

R<

<

\za\

77

can be expressed there as a convergent

JR', it

of the form

series

CO

This expansion is uniformly convergent in the closed annulus


z a
R 1 R[ R'. The coBi
^i> provided that R
efficients a n are now given by the single formula

<

<

<

2iti
2,-TTl

<

dz,

n+ 1
} {za)
J

where

Y denotes C when n

<

we may take T

to be

any

^ 0.

and C" when n

ever, the integrand is regular in the annulus


circle

\za\

r,

R<

where

Since,

how-

<

R',

R<r<

i?',

\za\

no matter what value n has.


The real importance of Laurent's theorem rests in the fact
that it is an existence theorem. It shows that an analytic function can be expanded, under certain circumstances, as a series
of a given type, but it does not necessarily provide the simplest
method of calculating the coefficients.
Finally, it should be observed that Laurent's theorem will
not provide an expansion of the logarithm of z as a series of
positive and negative powers of z. For Log z is a many- valued
function, whose principal value, logz, is discontinuous along the
negative half of the real axis and so is not regular in any annulus
with centre at the origin.
Example 1 Show that
.

cosh(z

where

an

+ -) =

+ 2 (z" + ^s)>

cos ra#cosh(2 cos 0)d6,

the series being uniformly convergent in any closed annulus with centre
at the origin.
We have to expand the function coshw where w s-f-z-1 Now
eoshw is an integral function of w, whereas w is an analytic function
of z whose only singular point is the origin. Hence cosh(z+z~ 1 ) is an
analytic function of z regular in the annulus R < |z| < R', no matter
how small the positive number may be or how large R' may be.
We can, therefore, apply Laurent's theorem to obtain

oo

cosh(z+z- 1 }

= 2zn

78

THEOBEM

CATJCHY'S

the contour T being any circle \z\


r. If we take r
on r, where 6 varies from to 2ar. Hence we have

an

1,

then

cosh(e ei +e- Bi )e- nei d0

Y
o
2ir

|-

2ir

cosh(2eos0)cosn0d$

+~

cosh(2cos0)sinn#d0.

If

we put 6

2ir <,

we

find that the last integral vanishes,

and so

S7J

cos n# cosh(2 cos 9) dd.

But evidently an

a_ M and therefore
,

coshfc+z- 1 )

00

+ 2 an (zn +z~ n

).

The uniform convergence of this series in every annulus with centre


the origin is an immediate consequence of Laurent's theorem.

Example
function
(iii)

when

2.

Find the Taylor or Laurent

l/{(z a +l)(z+2)},
\z\

>

(i)

when

<

\z\

series
1,

at

which represent the

when

(ii)

<

\z\

<

2,

2.

4.53. Isolated singularities of

an analytic function

Let/(z) be an analytic function with a singular point at z


a.
If there exists a neighbourhood of the point a which contains

no other singularity of

/(z),

the point a

is

an

isolated

\za\

< R in

called

singularity of the function.

If this

which

is

<

the case, there exists an annulus r


and can be represented

f(z) is regular

by the Laurent

series
o

Since, however,

we can make

r as small as

Laurent expansion actually holds when


infinite series

<

we

please, this

\za\

< B.

The

is

= a.

called the principal part of /(z) at the singular point z


There are three cases to be considered. First of all, it

may

CAUCHY'S THEOREM

happen that

all

79

the coefficients b n are zero.

We

then

call z =

a removable singularity of f(z), since we can make


B by suitably denning its value at a. For
when \za\
< \za\ B and F(a) = a
write F(z) = /(z) when

a.

f(z) regular

<

<

function F(z) has the Taylor expansion

2 aAz a Y

if

we
the

an <l so

is

regular

when \za\

<

B. Singularities of this type are of little

importance.
Secondly, the principal part of /(z) at the isolated singularity
may be a terminating series of powers of l/(z a). The singularity is then called a pole. If b m is the last non-zero coefficient
in the principal part, the pole is said to be of order m. Poles
of orders
poles.

1, 2, 3,...

The

are usually called simple, double,

triple,...

coefficient b x is called the residue of f(z) at the pole a.

If f(z) has a pole of order

m at z =

a,

the Laurent series takes

the form
f(z)

{z-a)-{b m +bm _ 1 (z-a)+b m _ z (z-af

+ ...+

+b 1 (z-a) m -i+ | on (a-o)+}

=
where
bm

is

\za\

(z-a)-m^(z),

<

B, and <f>{a), being equal to


is regular when \za\
not zero. We can therefore findf a neighbourhood
8 of the pole in which

(f>{z)

<

1/(2)1

> \\bj.\z-a\-.

Hence, if/(z) has a pole at a, |/(z)| tends to infinity as z tends


to a in any manner.
Moreover, if f(z) has a pole of order m at a, l//(z) is regular
and has a zero of order m there. For

TO = {z-ar/ftz),
where

</>(z)

Similarly

is

regular

and does not vanish when \za\

we can show

F(z) has a zero of order

that the converse

is

<

8.

also true, that if

m at a, ljF(z) has a pole of order m there.

Finally, if the principal part of f(z) at the isolated singularity

is

not a terminating series but has an

called

an

infinite

is
zero coefficients,
this case, a is evidently also a singularity of

t See

number of non-

isolated essential singularity.

4.51.

l//(z).

In

80

Example

CATJCHY'S THEOKEM
The function/(z) has a simple pole at

the residue at this pole


Since a

a simple

is

is

pole,

Example

2.

if/(z)

(z

- a)f(z) =

is

bounded near

Show

= a.

Show that

a and

b^ is

the residue

we have

where i/,(z) is regular in a neighbourhood of


to be determined. Hence
as z -* a, since

lim{(z a)f(z)}.

b x +{z- a)$(z)

-> b t

a.

that the only singularities of cot77z/(z-a) 2 are

Find the residues of the function at these poles, distinguishing


between the cases when a is or is not an integer or zero.
poles.

If

we

write

_ COS7TZ
~ (zaf&mTTz'

COt TTZ

(z-) a

we

see that the function is the quotient of two integral


functions, and
so its only singularities are at the zeros of the denominator.
Thus the
function is regular save for poles at z
a and at z
0, 1, 2,...
If a is not an integer or zero, z
a is a double pole and the rest all
simple poles; but if a is an integer or zero, z
a is a triple

pole.

Let us consider first the pole z = w, when n is an integer or zero,


and
let us suppose that n
a. The residue at this pole is, by Ex.
1,

lim

(z-n)coWz
^ cos

= Km

= ]im gcot77(g+n)

C-*o(+n a)*smnC

ir(n~ a) 2

'

When we

determine the residue at z = a, we have to consider


separately the cases when a is or is not an integer or zero.
If a is an
integer or zero, we obtain by writing za =

cot ttz

and so the residue


cot ttz

(za) i

_
~

cot ttL

is

Jtt. But when a is not an integer, we have


cot Tra
cosecVa
_
~~
|-" cosec

cot7r(a+)

3.

tt

7*

so that the residue at

Example

is

now

Determine the

77-

,2

7racot77a+...,

cosecVa.

singularities of zcosecz,

and

find the

residues at its poles.

Example
Example

5.

Find the residues of z i /(c 2 +z 2 )* at its poles.


Show that e1 /* has an isolated essential singularity at

6.

Show

4.

the origin.

Example

at the origin, which

that sin(l/z) has an isolated essential singularity


the limiting point of the zeros of the function.

is also

CAUCHY'S THEOREM

Example
at z

a.

7. f( z ) is

Show

that,

81

an analytic function with an isolated singularity


if /() = 0(|z-o|-) as z -> a, the
singularity is

a pole of order not exceeding

n.

4.54. Limiting points of zeros or poles

We have already seen that if a is a limiting point of zeros of


an analytic function f(z), regular when
< \z-a\ < R, then
f(z) either

vanishes identically or else has a singularity at


a. In
the latter case the singularity is isolated, but
it is not a pole
since \f(z)\ does not tend to infinity as z
tends to a in any
manner. Thus, apart from the trivial case when/(z) is
zero save
at a, the function has an isolated essential
singularity at the
point a.

however, f(z) is an analytic function whose only


singuthe region
\z~a\
R are poles, infinite in number, having the point a as limiting point, a
is a singularity; for
f(z) is unbounded in every neighbourhood of a. A
singularity
of this type is not a pole, since it is not an
isolated singular
If,

<

larities in

point.

We

<

an essential singularity.
Example. Show that sec(l/z) has simple poles
_ l/{(n+)7r}, where n is an integer or zero, and an
call it

larity at the origin.

Prove also that the residue at l/{(+lWl

The behaviour of a function near an

4.55.
tial

singularity

We

have seen

at the points
essential singu-

that, if

is

isolated essen-

a pole of the function f(z), then


any manner. On
the other hand, the behaviour of a function
near an isolated
essential singularity is of a far more complicated
character; in
fact, in every neighbourhood
of an isolated essential singularity,
there exists a poinff at which the function
differs by as little as
we please from any previously assigned number. This result
is due
|/(z)|

is

increases indefinitely as z tends to a in

to Weierstrass.J

Let us consider, then, an analytic function /(z)


isolated essential singularity at z
a. Let c be

which has an
any number,

t Actually there are an infinite number of such points. For if z, be


such
a point in the neighbourhood \z-a\ < r of the essential
singularity a, there
m
m |0_ l < tt zi~ a aad so on indefinitely.

if/
t Abh. der Preusa. Akod. Wiss. zu Berlin (Math. Klasse) 1876, 11
reprinted in Weierstrasa's TFer&e, 2, 77.
l>

This

is

CAUCHY'S THEOREM

82

real or complex;

numbers

and

r,

we have

show

to

no matter how
< r at which

in the region \za\

that, given

two

positive

small, we can find a point

<

\f(z)c\

z1

e.

If a is a limiting point of zeros of the function f(z)c, the


theorem is obviously true, since we have only to take zx to be
any zero in the given neighbourhood of a. But if a is not a
limiting point of zeros, the function f(z)c has no zeros in the
given neighbourhood, provided that r is sufficiently small, and
hence the function
j

is

<

<

\z a\
r. We have to show that there
when
a point z x in this region at which \g(z) > 1/e.
Let us suppose that, on the contrary, |gr(z)| < 1/e there.
regular

exists

00

Then,

if

the principal part of g(z) at a

K = hi

is

2 K( z ~ a )~ n

fl'C'K*-")""

>

we have

'^

H=P

<p<

where

r,

and hence

Since, however, b n is independent of p, we find, by making p tend


to zero, that b n is zero for all values of n and hence that g(z)

This

impossible; for

is

regular at a.

is

regular at a or else has a pole there.

\g(z)\

<

1/e

when

is

<

\za\

<r

is,

it

would imply that/(z)


The assumption that

therefore, untenable.

We have thus shown that the inequality \g(z) > 1/e must be
satisfied at one point at least in the given neighbourhood of a;
\

this completes the proof of Weierstrass's theorem.

A more striking result still is Picard'sf theorem, which states


that, in every neighbourhood of
there exists

a point at which

an

isolated essential singularity,

any given
1/z and ellz

the function actually attains

For example, sin


have isolated essential singularities at the origin; sin 1/z actually
\z\ < r, no matter how small r may
attains every value in

value with at most one exception.

<

be,

whereas

e llz attains

t Comptes Rendus,

there every value except zero.


88

(1879), 1024-7;

89

(1879), 745-7,

The

CAUCHY'S THEOREM
proof of Picard's theorem
present stage.
4.56.

The point

is

83

too difficult to be given at the

at infinity

The equation

z'
l/z sets up a continuous one-to-one correspondence between the points of the complete z-piane and
the
points of the complete z'-plane. If z is at the point at infinity,
z' is at the origin. Accordingly we shall say
that a function /(z)
has a zero, a pole, or an essential singularity at infinity
if the
function /(z- 1 ) has a zero, a pole, or an essential singularity
at
the origin. For example, the functions sin l/z, (z+1) 2
and ez
have respectively a simple zero, a double pole, and an essential
,

singularity at infinity.
It is important to notice that
if the only singularities of an
analytic function, including possibly the point
at infinity, are
poles, the function is

a rational function.
F(z) be such a function. It can have only a
finite
number of poles since a limiting point of poles is an essential
singularity; let its poles be at a ,,..., a oo,
of orders n n ,...,
x
k

For

let

nk

is

m respectively.
G{z) =

regular in every

tion.

We

x,

It follows that

(z-a 1 )^(z-a 2 ) n*...(z-ak ) n*F(z)

bounded domain, and so is an integral funcexpand G(z) as a Taylor series

can, therefore,

G(z)

= fb n z",
o

which converges for every

But

value of z.

finite

{z-a^iz-a^.^z-a^ has

a pole of order

N = n +n +...+nk
1

at infinity.

N+m.

Hence G(z) also has a pole at infinity, its order being


This, however, implies that the Taylor
series for G(z)

terminates and

is

of the form

N+m
This gives

F(z)

= *|\ }/{ (z~a )-j,


XJ
(

and

so F(z)

is

a rational function.

CAUCHY'S THEOREM

84

4.6. Analytical continuation

Let fx [z) be a regular analytic function denned in a domain

Dv We shall now show that it is sometimes possible to continue


this function analytically.

By this we mean that it is sometimes

an analytic function F(z) which is equal to ft (z)


at each point of D t but is regular in a more extensive domain.
Let D2 be another domain which has in common with D x a set
of points forming a domain A. If there exists a function f2 (z),
regular in D 2 which is equal to f^z) at a set of points having

possible to find

a limiting point belonging to A, the analytical continuation


f fi( z ) is possible.
In the first place, the functions /^z) and f2 (z) are equal at
every point of A. For fx{z) 2 (z) is regular in A and has a set
of zeros with as limiting point; since is a point of A, this
implies that/^z)f2 (z) vanishes everywhere in A.

If we now write F(z) = fx (z) when z is in D x and F(z) = f2 (z)


when z is in D 2 the function F{z) so defined is evidently an
,

analytic function, regular in the

domainf

the function

x -{-D 2 ;

Moref^z) has, therefore, been continued analytically into


2
over this analytical continuation is unique. For if g(z) is

another function, regular in 2 which is equal to fx {z) at a set


of points having a point of A as limiting point, g(z) is equal to
ft (z) at every point of A and hence is also equal to f2 (z) there;
a repetition of the previous argument then shows that f2 (z) and
g(z) are equal everywhere in Z> 2
When f2 (z) has been found in this way, it may be possible to
continue f2 (z) analytically into a domain 3 which overlaps 2
If this is so, there is a function f3 (z), regular in
3 which is
equal to f2 (z) in the common part of 2 and D 3 If 3 overlaps
Z> 1; we should expect that the analytical continuation of /3 (z)
into 1 would be the original function fx {z). If lt 2 and 3 are
circles having a domain A' in common, this conjecture is, in fact,
,

D
D

D D

For in A' we have /x (z) = f2 (z) = /3 (z), and this implies


= f3 (z) in the common part of D3 and D. But if
D x D 2 and D3 have no domain in common, it is not necessarily
the case that the analytical continuation of/3 (z) into D1 is/x (z).
For example, let us denote the domains \z 1| < p, \z o>| < p,

true.

that fx (z)
,

which belong to 1 or to
f By
x +2) 2 we mean the set of points
is also a domain.
both. Since
l and
a are domains, -Dj+Da

or to

CAUCHY'S THEOEEM

< p,

= e '

85

< p < 1, by Dly D2


any two of these domains have in common an
area bounded by two circular arcs, but there is no point common
to all three. The function z* is regular in D if we continue
it
x
analyticaUy into Z>2 from D2 into D and, finally, from D into
3
Dv the function obtained in this way is not 2* but 2*. 3
\z a> 2

where

and $V3

respectively;

4.61.

The general

definition of an analytic function


be
an
analytic function defined only in a certain
f(z)
domain where it is regular. Let us suppose that it is possible

Let

to continue this function analytically outside the given domain.


If we form all the continuations of the function, then all
the
continuations of these continuations, and so on in every possible
way, the complete analytic function /(z) is defined as consisting
of the original function and all the continuations so obtained.

The complete analytic function defined in this way is, of course,


not necessarily a one-valued function.
If/(z) is not an integral function, there will be certain exceptional points which do not lie in any of the domains into
which
the function has been continued. These exceptional points are
called the singularities of the complete analytic function.
It is
evident that the singular points of a one-valued analytic
func-

tion are also singularities in this wider sense.


It may happen that, in this process of continuation,

we

ulti-

mately reach a closed curve across which it is impossible to


continue the function. Such a closed curve is called a natural
boundary of the complete analytic function. An example of a
function with a natural boundary will be found in 4.62, Ex.
2.

4.62. Analytical continuation

by power

series

We shall now consider very briefly the problem of continuing


analytically a function/(z) defined initially as the
series

an (z

n whose circle
of convergence
)

sum of a power

has a

finite

non-zero radius.

The first thing to observe is that, when the continuation has


been carried out, there must be at least one singularity of the
complete analytic function on the circle of convergence C For
if there were not, we could construct, by analytical
continuation, an analytic function which is equal to
f(z) within C but
.

CAUCHY'S THEOBEM

86

regular in a larger concentric circle


function as a Taylor series in powers of

is

The expansion of this


z-z would then con-

this is, however, impossible since


verge everywhere within
the original series, whose circle
be
the series would necessarily
of convergence is C
take any nxed
In order to carry out the continuation, we
and its suc/(a)
of
values
the
point Zl within CQ and calculate
series
power
given
the
from
cessive derivatives at that point
the
form
then
We
by repeated term-by-term differentiation.
;

Taylor

series

>

,,

,i,

whose
Let

circle of

convergence

is

Gx

say.

with centre z x which touches C


new power series is cerinternally. By Taylor's theorem, this
/(as) there. The radius
tainly convergent within Yx and has sum
V There are now
of G cannot, therefore, be less than that of x

rx

denote the

circle

three possibilities:

Cx may have a larger radius than C In this case Cx lies


provides an anapartly outside G and the new power series
.

(i)

We can then take a point z 2 withm

lytical continuation of /(as).

Gx and outside C and repeat the process.


boundary of /(). In this case we
(ii) C may be a natural
C and the circle Cx touches C
outside
cannot continue f(z)
,

CQ was chosen.
when CQ is not a natural
(hi) Cx may touch C internally even
is a singuboundary of /(). The point of contact of C and Gx

internally,

point z x within

no matter what

obtained by the analytilarity of the complete analytic function


For there is
series.
cal continuation of the original power

on

necessarily one singularity

Cx and

this

cannot be withm

not a natural boundary of the


power
(z-Zo) this Process of forming new
function /(z)
anafunction
the
series provides a simple means of continuing
all
deduce
to
It is, therefore, theoretically possible

We

see, then,

that

=2

if

is

TC

>

lytically.!

from the prothe properties of the complete analytic function


series which defines
perties of the coefficients a n of the Taylor
on the theory of analytical continuation by power
f For further information
Oours d' Analyse, 2 (1918), 235-62.
Goursat,
aeries, see, for example,

CAUCHY'S THEOREM
the function

problem

87

This interesting but extremely difficult


however, beyond the scope of the present book.f

is,

Example

initially.

Show

that the function

f(z)= l+z+z* + ...+z+...


can be continued analytically outside its circle of convergence.
The circle of convergence O of this power series has the equation
|z| = 1. Within C the sum of the series is (1 z)~ l
But this function
is an analytic function, regular in any domain which does not contain
.

the point z

I,

and so provides the required

analytical continuation

of/(z).

however, instructive to carry out the continuation by means of


a is any point within C it is easily seen that the power
series expressing /(z) in powers of z a is
It

is,

power

series. If

(za) n

(I o)+i"

\za\
|1 o|.
at z
1, which is, therefore, a
singularity of the complete analytic function denned initially by the
given power series. If, however, a is not real and positive, we have

The

of convergence

circle

Now

<

if

<

1,

Ox of this new power series is

Cy touches

<7

jl a\ > 1 |o|, so that Cx crosses C ; in this case the


provides an analytical continuation of /(z) outside C

new power

series

Example

Show

2.

that the circle of convergence of the power series

/(Z)

= l ++2 +2 +2 +
2

...

a natural boundary.
The circle of convergence C of the power series is \z\ = 1. If the point
e" 4 on C is not a singularity of /(z), then /(re" 4 ) must tend to a finite

is

number

limit as the real

r increases

and tends

to unity. J

Let us consider, then, the behaviour of /(z) as z moves up to C along


the radius through the point of affix e 2"!*", where p and q are integers.
Now we can express /(z) in the form
OD

f{z)

say.

Since

a unique

/1 (re

limit

23wi / a *)

asr->

y2

= I++z + ...+a+ 2 z "


m=a+l
= A() +/(*),
a

is
1.

re 2D/29)

a polynomial in r of degree

2, it

tends to

But

= 2

=8+l

r ae 2i+-j)irf

_ y

r a f

=a+l

An

interesting account of recent researches on this problem is given by


Mandelbrojt, La Serie de Taylor (Collection 'Seientia'; 1926).
See also Dienes, The Taylor Series (Oxford, 1931), Chapter X.
t It is not, however, the case that, if /(re01*) tends to a finite limit, then
e M is not a singularity olf(z). For example, the binomial expansion of (lz)t

Hadamard and

<

converges when |z|


1, and (1 r)i tends to zero as *-*
a singularity (a branch-point) of (1 z)i.

1.

Yet

is

CAUCHY'S THEOREM

88

which tends to infinity as r -> 1. Accordingly the point of affix e 2 ""*/ 24


on Ca is a singularity of /(z).
But any arc of C no matter how small its length, contains a point
whose affix is of the form e2""*/ 2*, where p and q are integers. There
are, therefore, points whose affix is of this form within every circle which
,

crosses

Ca

its circle

, so that it is impossible to
continue f(z) analytically outside
of convergence.

REFERENCES
Complex

and Cauchy's theorem :


d' Analyse, 2 (Paris, 1918), Chap. XIV.
Pollard, Proc. London Math. Soc. (2), 21 (1923), 456-82; 28
integration

E. Goubsat, Cours
S.

(1928),

145-60.

G. N. Watson, Complex Integration and Cauchy's Theorem (Cambridge

Math. Tract, 1914).


Analytical continuation:

E. Goubsat,

Chap. XVT.

loc. cit.,

Hadamabd and

J.

S.

Mandelbbojt, La

Sirie de Taylor (Collection

'Scientia'; Paris, 1926).

MISCELLANEOUS EXAMPLES
1. The function /(z) is regular when
\z a\ < R. Show
< r < R, then

that,

if

2ff

where F(6) denotes the

2.

The function /(z)

/'()

real part

is

regular
|a|

then

fla)
JK
'

2ni

F(6)e-"de,

of/fa+re8*).

when

< R<

\z\

<

R'.

Prove that,

if

R',

B2 ~ aa
f{) d
J(
>
(z-a)(R*-za)
J
f

'

where

is

the circle

\z\

0<r<R,

Deduce Poisson's formula,

R.

that, if

o
3.

|/(z)|
\z\

<

When

<

R, the function/(z) is regular and satisfies the inequality


> 1. Show, by applying Poisson's formula to log/(z), that, if
kR where k < 1, then
|z|

l/WI

<

I/WI^*.

4.

The function

CAUCHY'S THEOREM
when \z\ < E and has

89

/(z) is regular

the Taylor

QO

expansion

o an zn

Show

that, if r

<

R,

= JK|V.

|/(w*)|cW

2^J
o

Hence prove

that, if |/(z)|

M when

<

fo
5.

Deduce Cauchy's

notation of

|a|

iJ 2

|/{M)(a))

<

<

R,

ilf 2

from Ex.

inequalities

4.33,

\z\

4,

and show

that, in the

n MjRn
,

and only if f(z) = Me ai(z a)n/Rn where <* is a real constant.


6. The integral function f(z) satisfies everywhere the inequality
k where A and k are positive
constants. Prove that/(z) is
]f( z )\ < A\z\
a polynomial of degree not exceeding k.
if

that the branch of the function (1 2/^z+z )+ 1 when z = 0, is regular when is less than the
|^\W~ 1)| anc so can be represented by a Taylor series

07. Show

equal to

1 /2

\z\

which

is

smaller of

i+f -P.0*)*".
Prove that the

coefficients in this

expansion are given by

*- J^ -2/4Z+Z

2 )- 1 / 2

efe

where

a closed contour surrounding the origin but not enclosing


either of the points /tVO u,2 ~ ^)8. Prove that the function exp{w(z -1 )} is regular save at the origin
and can be expanded as a Laurent series
is

00
2jt

where

JJu)

( 1)V_()

= J-

cos(0-wsin0)

d#.

The function /(z) is regular when \z\ > R and


-> oo. Prove that f(z) can be expanded as a

9.
|z|

|/(z)| is

bounded as
the form

series of

CO

23~". convergent when


10.
is

\z\

>

R.

The function /(z) is regular in the strip a < Imz < <x, where a
Prove by using Laurent's theorem that, if f(z) is periodic,

positive.

of period

2tt, it

can be expanded in the form

CAUCHY'S THEOREM

90

where

C)}

/(z)er* dz,
o

the series being uniformly convergent in the strip


for every positive value of S ( < a).

a+S < Imz < a

Deduce thatf
00

/(z)

+ 2 (aoosnz + 6sinwz),

2v

where

aB

27T

6 = -

/(z)coswz dz,
o

11.

Each of the

12.

functions

Show

z),

cosec 2 zlog(l

(ii)

has a pole at the

origin.

Find

that the circle

its

\z\

Show

that the

>?* M

sum

acz

T^ +
a when
14.

is

z/(sinz tanz)

1 is

a natural boundary of each of the

^
<i

is

(iii)

order and residue in each case.

functions

13.

/(z)sinraz dz.

cotz,

(i)

\z\

<

1,

but

(a

- c)

is

\z\

<

1,

z2

z*

(z^ri+z^i + z^i + -)

when

Prove that the sum of the

z/(l z 2 ) when

2 ;=?**

<*>

of the series

|z|

>

1.

Why is this?

series

(1+Z)(1+Z M + 1 )

but

is l/(z 2

1) when

\z\

>

1.

The function

f(z) is regular within and on the circle C whose


R, save for simple poles b x 6 2
b n within C. Moreover, /(z) does not vanish on G, but has simple zeros a a
am within
x
%
C. Prove that the function
15.

equation

is

\z\

*>-"> filial ft
8=1

is

regular

(!e9

6=1

and non-zero within and on O, and

also that \F(z)\

|/(z)j

on O.
t This result is Fourier's theorem for analytic functions of a complex
variable.
detailed account of the conditions under which Fourier's theorem
for functions of a real variable is valid will be found in Hobson's Functions
of
a Real Variable, 2 (1926).

CAUCHY'S THEOREM

By applying Poisson's formula to


r

<

log F(z),

show

91
that, if z

= rew where

R, then

iogi/<*)i

log

8=1

2 log li^=^l +
l^=^r S=l
2ir

+ 2^

JJ2

lo

j.2

gl^^>l .ffl-2JfrcoB(g-^)+^

Deduce that
2ff

log

|/(^)| d*

= ftr^

J5^gp

-flog

|/(0,|.

not a zero or pole of f(z).


be modified if f(z) possessed multiple poles
or multiple zeros within C ? (Jensen, j")

provided that the origin

How

would these

t J. L.

is

results

W.

V. Jensen, Acta Math. 22 (1899), 359-64.

CHAPTER V

UNIFORM CONVERGENCE
5.1.

The limiting function

Let

of a sequence of functions
be a sequence of one-valued functions,
each defined in a bounded closed domain D.
At present we do
not assume that these functions are differentiable
or even continuous in D; they are merely functions of the
complex variable
z in the most general sense.
(z),

s x {z), s 2 (z),...

It may happen that, when


is a point of D, the sequence of
complex numbers s (Q, Sl (), S2 (),. tends to a
definite finite
limit.

We

then say that the sequence of functions is


convergent
. If the sequence converges at each point of D,
it is said to
be convergent in D, and the limiting function
s{z) of the sequence
is defined at each point of
by the equation
at

D
s(z) = lim sjz).

This means that, given any point


of D and any positive
number e, no matter how small, we can find an integer
that \s(Z)-sn(Q\

made

definite

< e Wh en

N such

> N.

The

N, which

integer

is

by being taken

as small as possible, will depend,


in general, not only on e but also on the
particular point under
consideration; we denote this fact by writing
N(e,

In

this

N=

way we have

).

associated with the convergent sequence


a function N{e, z) which is defined at each
point of
and which
only takes positive integral values. In
general, this function

will not be bounded in D. But when


N( e ,z) is bounded, we say
that the sequence is uniformly convergent in
the domain D.
The uniform convergence of the sequence implies,
therefore
that there exists an integer
M( e depending

M=

on

),

e alone,'

<

such that N{e,z)


M( e ) at each point z of D. In other words,
the sequence of functions s (z),
Sl (z),... converges uniformly in
the domain
to the limiting function s{z) if, given a
positive
number e, no matter how small, we can find an
integer M,
depending on e alone, such that the inequality
s (z)-s n {z)\
e
holds at each point z of
provided only that
The idea of uniform convergence is of great
importance in

n^M.

<

analysis; for, as

UNIFORM CONVERGENCE
we shall see, it often enables

93

us to deduce properties of the limiting function of a sequence from the common


properties of the members of the sequence.
5.11.

The

principle of uniform convergence

The definition which we have just given of the uniform convergence of a sequence presupposes that the limiting function
of the sequence is known. Before we proceed to discuss the
properties of uniformly convergent sequences, it is desirable to
express the condition for uniform convergence in a form which
does not involve the actual determination of the limiting function. This is provided by the following principle of uniform
convergence, analogous to the principle of convergence for
sequences of numbers. The necessary and sufficient condition for
the

uniform convergence of the sequence of functions s (z), s (z),


x
bounded closed domain
is that, corresponding to

s 2 (z)... in the

any positive number e, there should exist an


on e alone, such that the inequality
\

<

P ^)sJz)\

s m+

integer m(e), depending

holds at each point z of


for every positive integer p.
In the first place, the condition is necessary.

sequence converges uniformly to s(z),


of 8 5.1,
...
...

we

For if the
have, in the notation

D provided that n > M(%e).


m = M(%e) and p is any positive integer,
= \{s(z)-s m (z)}-{s(z)-sm+p
m+p (z)-sm
< Hz)-sm + \s(z)-sm+p
<e
at each point z of

\s

(z)\

(z)\

Hence,

if

(z)}\

(z)\

at each point z of D.

The condition is also sufficient. For if it is satisfied, the


sequence converges, in virtue of the principle of convergence, to
a limiting function s(z). If we make p tend to infinity in the
inequality

we

sm+p\ z )

find that

But

since

\s

sm\ z )\

\s(z)s m (z)\
n (z)

m (z)\

<e

<

<

e>

e.

at each point of

provided that

UNIFORM CONVERGENCE

94

n> m, this inequality gives


\s(z)-s n (z)\

=
<

\{s(z)-s m (z)}+{s m (z)-s n (z)}\

\s(z)sm (z)\ + \s m (z)~s n {z)\

<2e,
when n >,m. Hence the sequence converges uniformly
5.12.

to s(z).

Uniformly convergent sequences of continuous

functions

At first sight we should expect that the limiting function of


a convergent sequence of continuous functions would itself be
continuous. This is, however, not the case. For example, the
sequence of functions 1, z2 z4 z 6 ,... converges at each point
within and on the ellipse x 2 +2y 2
1, yet the limiting function
of the sequence has the value 1 at the points z
1, but is
zero at every other point within or on the ellipse. It can, however, be proved that ifs (z), s^z), s (z),... is a convergent sequence
2
,

of functions, each continuous in a bounded closed domain D, a


sufficient condition for the continuity in

of the limiting function


of the sequence is that the convergence be uniform.
For if the sequence is uniformly convergent, then, given any

s(z)

positive

number

e,

we can

find

an integer M, depending on

alone, such that the inequality

holds at each point z of D, provided only that


and z 2 are any two points of D,

n^z M. Hence,

if z x

\s{z x )-s{z

2)

=
<

lisiz^-s^z^+is^z^-s^z^+ls^z^-siz^}
l*(i) *jf(i)|+

\s(z 2

)-sM (z z )\ + \sM ( Zl )-sM (z 2 )\

<2e+\sM (z 1 )sM {z2 )\.


But

a continuous function, and so is also uniformly


We can therefore find a positive number 8, depending only on e, such that the inequality
s

(z) is

continuous.

<e
of D for

M)w(z')l
holds for each pair of points

Hence,

if

\z 1

z 2

<

8,

z, z'

we have

Kzi) s(z 2 )|
this proves the theorem.

<

3e;

which \zz'\

< 8.

UNIFORM CONVERGENCE

From

we

95

deduce that if s (z), s^z), s 2 (z),... is a


sequence of continuous functions which converges uniformly to s(z)
in a bounded closed domain D, and if L is a contour lying in
D, then
r
r
this

shall

lim

n>co

s n (z) dz

The function

s(z) dz.

5:

Li

Jj

has been shown to be continuous in


and
Now, given any positive number e, we can, by hypothesis, find an integer M, depending on e
alone, such that, when n
M, the inequality \s(z)sn (z)\
e
holds everywhere in
and, in particular, on L. Hence, if I be
the length of L, we have
s(z)

so is certainly integrable along L.

<

j"

s(z)

dz

s n (z) dz

| {s(z)-sn {z)} dz

<d.

Since, however, e

can be as small as we please, this shows that


lim
J

s n (z) dz

Lt

5.13.

f s(z) dz.
L*

Uniformly convergent sequences of analytic func-

tions
Let us suppose that the sequence of functions s

(z),

s^z), s 2 (z),...

domain D within a closed


contour C, and that each member of the sequence is an analytic
function regular within C. Then s(z) is also regular within G and
the sequence s'
n {z) converges uniformly to s\z) in D.
converges uniformly to s(z) in every closed

We know that the limiting function s(z) is continuous within


C; we shall prove that it is also regular there by means of
Morera's theorem ( 4.4).
Let L be any contour, not necessarily closed, lying entirely
within C. Since each member of the sequence is a regular
analytic function, J s n (z) dz depends only on the affixes of the
L

end-points of L.

But since L lies within C, it lies in a closed domain in which


the sequence of functions converges uniformly, and so
lim

The value of

f s(z)

s n {z) dz

s(z) dz.

dz depends, therefore, only on the affixes of

UNIFORM CONVERGENCE

96

the end-points of L. Hence,

by Morera's theorem,

s{z) is

regular

within C.

show that s^(z) converges uniformly to s'(z) in


domain D within C. We construct a closed contour
r which lies within G and yet is definitely outside D; let 8 be
the shortest distance between T and D. Now let a be any point
of -D; then, by Cauchy's integral for the derivative of an analytic
It remains to

any

closed

function,

27rt

()

\z~af

J
r

But, by hypothesis, s n (z) converges uniformly to 5(2) within


and on T. Hence, when we are given any positive number e,
we can find an integer M, depending on c alone, such that
\s(z)s n (z)\ < e when z is any point within or on V, provided
that n >
By the result of 4.14, it follows that, when

tc

> M,
|'(a)-;(o)|

where

is

the length of V.

el1

<

2ttS 2

'

The expression on the right-hand


and therefore s^(z)

side of this inequality is independent of a,

converges uniformly in

D to the limiting function s'(z).

A repetition of the same argument shows that s^(z) converges


uniformly in every closed domain within

to

s"(z),

and so on

indefinitely.

5.2.

Uniformly convergent series

Let each term of the

infinite series

^ un z
(

De a one-valued

We

function of z, defined in a bounded closed domain D.


associate
series the sequence of partial sums s (z), s^z), s (z),...
2

with this

6re
()

If this sequence
s{z),

we say

is s(z).

If,

= o(*)+i()+a() + -+().
D

convergent in
and has the limiting function
that the series converges in
and that its sum
is

further, s n (z) tends to its limiting function uniformly

in D, the infinite series

is

said to be uniformly convergent in

the bounded closed domain D.


If each term of the infinite series

bounded closed domain D, and

2 un {z)

is

continuous in a

if the series converges .uniformly

UNIFORM CONVERGENCE
in D, then the sum of the series s(z)
Moreover, if L is any contour in D,
\ s{z)

dz

= f

is also

un (z)

97

continuous in

dz.

This important result is obtained immediately


by applying the
theorems of 5.12 to the sequence of partial sums
of the infinite

series.

Similarly,
series

we deduce from

5.13 that if each term of the infinite

ujz) is an analytic function, regular within a closed conand if the

tour G,

infinite series converges uniformly in every closed


within C, then the sum of the series s{z) is
also an
analytic function, regular within G. Moreover,

domain

*'(*)

= 2<(z),

the latter series being uniformly convergent


in

D. This result is
sometimes called Weierstrass'sf double-series theorem,
since his
proof of it depended on expressing each term as
a power series

and rearranging the double


It should be

theorem to the

series

so

on

formed.

we can apply

Weierstrass's

u^(z), to obtain
*"(*)

and

series so

observed that

= !<(*),

In other words, a uniformly convergent


can be differentiated term

indefinitely.

series of regular analytic functions

by term

we

as often as

5.21. Weierstrass's

please.

M -test

One of the

simplest sufficient conditions for the uniform convergence of a series is Weierstrass's ilf-test, which
runs as
follows.

The

lutely in

a bounded

an

J ujz) converges uniformly and absodomain D if each term satisfies there


n where
n is independent of z and

infinite series

closed

inequality \un {z)\

2M

<M

n is convergent.

t Monatsberiehte der Preuss. Akad. Wiss. (1880), 719-43. This paper is


Weie 'strass 8 Werke, 2 > 201-30. See also Bromwich,
Infinite Series
, orKnopp,
(lZb), lbb-7,
Theory and Application of Infinite Series (1928); 430-3.
'

no^oif

UNIFORM CONVERGENCE

98

The
that

convergent in D. To prove
uniformly convergent in D, we observe that, if
the nth. partial sum of the series, then

series is evidently absolutely

it is also

s n (z) is

m+p
m+p
\s

m+%
m+p

i
|

2 un(z < m+1


2 K( z
m+1

m+P (z)sm {z)\

)l

m+p

<*>

2 Mn
< m+1
2 Mn < m+1

Now 2 Jfn is

for every positive integral value of p.

hence, given any positive

number

we can choose

e,

2Mn<e

convergent;

m so that

m+1

With

this value of

m, which depends only on

K+*(2 )- s(z )l
where

denotes any point of

The required

result

now

e,

we have

<e

and # is any positive integer.


by the principle of uniform

follows

convergence.
00

Example

The radius of convergence of the power

1.

series

2o an zn

B. Show that the series converges uniformly and absolutely when


< R', provided that R' < R. Deduce that the sum of the series is
an analytic function regular within its circle of convergence.
is

\z\

00

Example

Show

2.

that the series

2 n~

converges absolutely and

uniformly in any bounded closed domain in which Biz

5.22. Further tests for the

>

1.

uniform convergence of an

infinite seriesf

The
closed
(i)

series ]T

domain

uniformly convergent in a bounded

sums of the series

2 (z) o,re uniformly bounded

D,

z
** uniformly and absolutely convergent
{v n( z )~ vn+i( )}

in D,
(iii)

D if

the partial

in
(ii)

an(z ) vn(z )

and

v n (z) tends

to zero

uniformly in D.

The tests of this section are similar to those of Abel and Dirichlet for
f
the ordinary convergence of real series. They were first published by Hardy,
Proo. London Math. Soc. (2), 4 (1907), 247-65.

UNIFORM CONVERGENCE

99

Let us write

= ao(z)+a (z)+...+an
$.() = ao(z)v (z)+a (z)v (z)+...+a (z)v
n
n
*()

(z),

(z).

The first condition implies that there exists a


positive constant
K, such that the inequality \s {z)\
< K holds when z is any
n
point of D and n is any positive integer.
As in 2.43, we obtain by partial summation

8n+p (z)-8n (z)

=
and

n+pl

Z+i Sr{z){v {z)-Vr+1 {z)}+Sn+p {z)vn ^{z),

-*.(*K + l(*)+

therefore

1^(2!)-^)!

< ^{K +1

(z)l

n+

<^{K +

1
\

+l

Vr (z)- Vr+l(z)

+lVn+pm

+ i<Wz + f (a)-wr+1 (a)|}


when z denotes any point of D and p is any positive
integer.
The second condition states that the series
(z)v

converges uniformly in
i(z)l

|r

)l

\v

(z)\

D. Hence, given any positive numlber


we can find an integer lt depending only on e, such
that

n+l

\vr (z)-v

when

r+1 (z)\

>

< e/K

z is any point of
and n
iV3
Moreover, since n () tends to zero uniformly in
Z>, we can
find another integer
depending only on e such that
2
\e\K when n
K(z)\
2 and z is any point of D.
.

N
^N
,

<

bining these results,

and

N,
2

we

Com-

find that, if

l^.-f(a)-^(2)|
,

<

is

the greater of N.

e,

for every positive integral value of


p.

Hence, by the principle


of uniform convergence, the sequence of functions
8 (z), S^z),
S2 (z),... converges uniformly in D. This completes the proof.
In a similar manner it can be provedf that
the series
an(z) vn (z) is uniformly convergent in a bounded closed
domain

Dif

(i)

the series

an {z) converges uniformly in D,

t See 2.43, Ex. Compare the corresponding discussion of the


uniform
convergence of integrals in 5.52.

UNIFORM CONVERGENCE

100
(ii)

the series

which
(iii)

is

K(z) v+i(z )l

*s

and has a sum

convergent

bounded in D, and

the function v (z) is

bounded in D.
00

Example

series ~2,an z n ,

The power

1.

whose radius of convergence

1. Show that the series converges


is unity, converges at the point z
< 8 < \n.
uniformly in the domain |1 z\ < cosS, |arg(l z)\ < 8, if
oo
Deduce that
m

as z -> 1 along any path within the circle of convergence which does
not touch that circle.
We apply the second test of 5.22, with an (z) = an vn (z) = zn Since
a n is convergent and an does not depend on z, the first condition is
.

satisfied.

When

The
\z\

third condition

<

1,

is

also satisfied since v

{z)

1.

we have

2 k(z)-^+1 2 =
(

)|

ri-*l

s
l

l"

=iE$y

Nowthe domain defined by the inequalities |1 z\ <

cos 8, |arg(l z)\

<

S,

a sector of the circle with centre at z = 1 and radius cos 8. If


< 8 < \n, each point of the sector, save z = 1, lies within. the circle
of convergence of the given power series. In this sector we have
1 z = pe, where < p < cosS and 8 < < 8. Hence

is

<j>

|z|

and

1-2/jcos^+p 2

<

l-2pcosS+/3 2

< 1 2pcos8+pcos8 < 1 pcosS + Jp


\z\ < 1 |pcos8.

so

cos 2 8,

It follows, therefore, that at each point of the sector, save z

1,

2M*)-W*>l<i^=2Bec8.
=

also satisfied at z
1, since each term of the series
vanishes there. The second condition of the test is
thus also satisfied, and the proof of the first part of the problem is
completed.
By 5.12, the sum of the series ^,an zn is continuous in the sector

This inequality

Ki( z )

^th-i^ 2

is

)!

under consideration. Hence


as z > 1 along any path in the sector. This conclusion also holds when
1 which does not touch the circle of
z > 1 along any path within |z|
convergence; for, by choosing 8 sufficiently near to \ir, we can ensure
that such a path lies in the given sector.
More generally, if the radius of convergence of the power series 2) an zn
is finite and non-zero, and if the series converges at the point z on the

UNIFORM CONVERGENCE
circle

101

of convergence, then

as z - z along any path which does not touch the circle


of convergence.
For the series an z^zn has radius of convergence unity and

at the point z

power

converges

This result

1.

AbeVs theorem on

is

the continuity of

series.

Example

Give an example to show that

2.

if

the power series

2 an zn has a finite non-zero radius of convergence, and if the sum of the


series tends to a finite limit as z ->
a path which does not touch that

that

2 a z

on the

of convergence along
not necessarily the case

circle

circle, it is

converges.f

Example

3.

Riemann's Zeta-function

denned by the equation

is

CO

(z)

= 2n

~">

wh en Rlz >

1.

this function into the region

that the analytical continuation of

where Rlz

(l-2i-*)(z)

Hence show that the only

Show

>

is

given

by

l-*_2-*+3-*-4-*+....

singularity of (z) in the right-hand half of

a simple pole of residue 1 at the point z = 1.


We have seen ( 5.21, Ex. 2) that the series denning (z) converges
uniformly and absolutely in any bounded closed domain to the right of
the z-plane

is

the line Rlz

Hence, by

1.

when Rlz > 1.


Now, when Rlz
(

>

5.13, (z) is

an analytic function, regular

1,

- 2i~*)(z) = 2 n-% 1 - 2*-*) = w-*- 2 V (2n)-'


l
i
= 1-* 2-*+3-
i

...,

the reordering of the terms of the series being valid by absolute convergence.
now show that the latter series converges uniformly in
any bounded closed domain
in which Rlz > S, provided that S is
positive. We use the first test of 5.22, with

We

a(z)

The
(i) is

partial

satisfied.

sums of

(-l) n

2 an

Condition

vn (z)

(n+l)-.

are alternately 1

(iii) is

and

0,

so that condition

also satisfied, for

K(z)| = (n+l)- E,s < (n+l)-,


and so vn (z) certainly tends to zero uniformly in D.
To show that condition (ii) also holds, we use the formula
m+2
1
()- Vfi() = {n+l)-"-(n+2)-" = z
j r*- dt.
TC+l

f For an account of the conditions under which the converse of Abel's


theorem is true, see, for example, Landau, Darstellung und BegriXndung einiger
neuerer Ergebnisse der Funktionentheorie (Berlin, 1929), 52-67.

UNIFORM CONVERGENCE

102

This gives

+2

<

to.(*)-vn(*)|

+2

/ l

1*1

M*<

m+l

<

2
l

t-*- ldt

+l

lal-fn+l)-

8- 1
.

Since z is bounded in ), it follows by Weierstrass's M-test that


{vn( z )~ vn+i( z )} converges uniformly and absolutely there.
We have thus shown that the series

rj(z)

1-* 2-*+ 3-*- 4-*+...

converges uniformly in any bounded closed domain in which Rlz

Hence
Biz >

an analytic function, regular when Rlz


we have
1 2
= ,, j,,.,,,

-q(z) is
1,

0.

>

0.

But when

(1-2

r)(z)

>

)(z).

Accordingly this equation provides the analytical continuation of (z)


into the region
< Rlz < 1.
Now the function 1 2 1-2 has simple zeros at the points given by

(1 z)Iog2 = 2pm,
where p is any integer or zero. The equation (z) = q(z)j( 1 2 1 - 2 ) shows
that a point of this set is a simple pole of (z), provided that tj(z) does
not vanish there, and also shows that (z) has no other singularities in
the right-hand half of the z-plane.
The point z
1 is a simple pole of (z), since
residue there is
.,.
,,

But no other

zero of

equation

where

^(z)

which the reader

1_z is

l- z +2~

sl

-q(l)

log 2.

The

log2

a pole of

(z).

= i_ 8 i-
(

To show

this,

we use the

)(lg)f

2. 3-* +4-' +5-* 2.6-"+...,

The function ij x (z) is


when Rl z > 0. Hence the poles

will easily prove.

analytic function, regular

1-2 1-*

-i
1

evidently an
of (z) in the

right-hand half of the z-plane are the points of the set

(1 z)log3
where q

2qm,

any

integer or zero, at which ^(z) does not vanish.


If (z) possessed a pole other than z
1 in the right-hand half of the
z-plane, this would imply the existence of integers p and q such that
is

log 3

log2

which

is

impossible.

5.3. Infinite

Hence

(z)

= pq'

has only one pole in Rlz

>

0.

products

The symbol

(l+a1 )(l+aj(l+aa )...(l+an )...,


which involves the multiplication of an infinite number of complex numbers, has, in itself, no meaning. In order to assign

UNIFORM CONVERGENCE
a meaning to the value of such an

sequence of partial products

103

infinite product,

we form the

p v p 2 p 3 ,..., where
,

n
1 + ar)JP=IT(
r=l

If

pn

tends to a

we say

finite

non-zero limit

that the infinite product

is

as

n tends

convergent, and

to infinity,

we

write

p=f[(l+a

r ).

however, p n tends to zero or does not tend to any finite


we say that the infinite product is divergent.
In order that the infinite product may converge it is necessary that no factor should vanish; for if l+m
0, then p n
when n
m. We shall suppose this condition is always satisfied. It is also necessaryf that an ->
as n -> oo, since

If,

limit,

We

shall

Pn = Pn-l+ anPn-V
now show that the necessary and

sufficient condition

for the convergence of the infinite product JJ (l-\-an ) is the convergence of the series 21og(l~t~am ), where each logarithm has its

principal value.
n

Let us write

sn

We then have pn =

= 2 log(l+o

r ).

exp(s m ).

But

since the exponential function

is continuous, s n -> s implies that

pn -> es

This proves the

sufficiency of the condition.

Now

sn

= logp n +2qn

Tri,

where qn is an integer. Since the principal value of the logarithm


of a product is not necessarily the sum of the principal values
of the logarithms of

its factors,

qn

is

not necessarily zero.

We

show that qn is, however, constant for all sufficiently large values
of n; from this the necessity of the given condition will follow
immediately.
Let us write <xn and /3n for the principal values of the arguments of l+an and p n respectively. If the infinite product is
convergent,
is

<x

and j3m

n -*

->

/?,

say, as

n -> oo. The

integer qn

then given by
al

+ ++
<*2

<*n

= Pn + tyn"-

f The example of the infinite product JJ (1+ l/w)> *or which pn


shows that this condition is not sufficient.

(n+1),

UNIFORM CONVERGENCE

104

Hence we have
as n -> co. But since q is an integer, this implies that
n
qn
for all sufficiently large values of n.

Therefore, if p n tends to the finite non-zero limit


it

follows that

and

sn

so the condition

is

as

=q

- co,

-^logp+2qm,
,

also necessary.

5.31. Absolutely convergent infinite products

The infinite product JJ (1+aJ is said to be absolutely convergent if the series


log(l +aj is absolutely convergent.
Evidently an absolutely convergent infinite product is convergent and its value is not altered when its factors are deranged. The necessary and sufficient condition for the absolute

convergence of the infinite product


JJ (l+an ) is the absolute convergence of the series
<V
For since an -> as n *-> co, we can find an integer
such

that

\an

< i when n^N.

Hence we have, when

N
n^ N,

_log(l+am )
a

and so

3^4

J|o B

<

|log(l+J|

'"

< ||aB

|.

This shows that the series 2,log(l+a ) converges absolutely


if
n
and only if the series
Kl is convergent, and the required

result follows at once.

Uniformly convergent infinite products


Let u^z), u 2 (z), u3 (z),... be a sequence of one-valued functions,
defined in a bounded closed domain D, such that the infinite
5.32.

product IIO+mJz)} converges at each point of D.


sequence of partial products

If the

/Js)=TT{i+^(s)}
converges uniformly in D,
verges uniformly in D.

The simplest

we say

test for the

that the infinite product con-

uniform convergence of an

infinite

UNIFORM CONVERGENCE
product

105

the Jf-test, which states that the infinite


product
1
+m(2)}
converges uniformly and absolutely in the bounded
IT{
closed domain
if each function un (z) satisfies there an inequality
\u n (z)\
n where
n is independent of z and
is conis

<M

^Mn

vergent.

The absolute convergence of this product


of 5.31.
Let us write

Pn = JJ (1+M

Then since

r ).

a consequence
*

is

convergent,

r is

Pn tends to a finite limit as n -+ 00.


Now when n >m,we have
!/(*)-/(s)l

But

if

we multiply out

l/m (*)|.

{1

+u

fl{l+ur (z)}-l\.

m+1

(z)}- 1,

we obtain an expression

of the form

I u (z)+ I u {z)u {z)+ 2 u {z)u (z)u {z)+...+


+Um +l(z)Um+2(z )-Un
r

(z),

whose absolute value does not exceed

ZK+ZM M +ZM M Ml+ ...+Mm+1 Mm+2 ...M


r

= fl(l+M )-l.
r

m+l
Hence we have

\L(z)-fm (z)\

< r=l
ft (i+mm n (l+jr,) 1} = pn ~pm
m+1
K

But

since

positive

when

n>m.

when n
e,

Pn

number

tends to a hmit,
e,

we can

and then choose

assign arbitrarily a

m so that < Pn Pm <

This gives

l/(z)-/m(z)l <e
> m and z is any point of D. Since m depends only on

the sequence of functions


fx (z), /,(),
The test is thus established.

/,(),...

Finally, it follows immediately

5.13 that if the infinite

converges uni-

formly.

product

Jl{l+un (z)}

from

converges uniformly to f(z) in every closed


domain within a closed contour C, and if each factor of the product
is an analytic function, regular within G,
thenf(z) is also regular
within C.

UNIFORM CONVERGENCE

106

Discuss the convergence of the infinite products

Example.

(-)('-*)('-<">

('-9('+i)('-l)('+l)--

{(-M('+iH{(-iM((+iHD

in any bounded closed domain

1.

which contains none of the points

\z
< R when z lies in D. Since
convergent, the product (i) converges
uniformly and absolutely in D, by the M-test. If F(z) is the value of
the product, it is an analytic function, t regular in D.
On the other hand, the product (ii) does not converge absolutely in
D, since the series B(l l + i+i+i+i + . ) is divergent. Let us write

There exists a constant

\z

/n 2

< R 2jn s

and

R such that

R /n
2

2 is

+
Fn (z),fn (z) for the nth partial products of

(i)

we have

and

(ii)

respectively.

Then

and so the sequences f^z), f3 (z), /6 (z),... and f^z), /4 (z), /6 (z), both
converge uniformly to F(z). Hence the infinite product (ii) converges
uniformly in
to F(z).
To discuss the third product we write

(l_)e/n=

n/

un (z)

where

l-un (z),

)'

=2

when n > R.

Similarly

we have

(i+)r4=

<

when

where

|tf(z)|

ilf-test

shows that the product

e(R/n) 2

is

(iii)

-(),

any point of 2? and n > R. The


converges uniformly and absolutely

ini).
Finally, since the partial product of (iii) of order 2n
the third product also converges to F(z).

5.4.

is

equal to

Fn(z),

Functions depending on a parameter

Let f{z,a) be a one-valued function of the two complex


variables z

and

f It is well

a,

defined

known

when

that F(z)

z lies in a

sin nz/{nz).

See

bounded closed
6.83,

Ex.

UNIFORM CONVERGENCE
domain

and a

107

<

in the circle !<*-<*


p. If /(z, a ) tends
to a finite limit as a -> a when z is
any point of D, the
limiting function F{z) has a definite finite
value at each point
lies

Let us suppose that this is the case. Then, given any


positive
c we can find a positive number 8{e, z),
depending on
and on z, such that \f(z,<x)-F(z)\
6 when |a-a
8.

number
e

<

We choose S as large as possible.f

Then

<

a one-valued

8(e, z) is

positive function of z defined everywhere in D. If there


exists
a positive number A(e), independent of z, such that S(e, z)

> A(<=)
D to

at each point z of D, we say that/(z, a) tends uniformly in


the limiting function F(z) as a -> <x
.

By

the appropriate modification of the analysis of 5.12, we

can show that iff(z, a) is a continuous function ofz in the bounded


closed

domain

<

for each value of a in |a


p, and if
F(z) uniformly as a -> a then F{z) is continuous in D.
Moreover, if L is any contour lying in D,
|

f(z, a) ->

lim

if,

f(z,*)dz=

F(z)dz.

Again, by an argument similar to that of 5.13, we find that


for every value of a in |a <*
p, the function f(z,ot) is an
|

<

analytic function of z, regular within a closed contour C, and


if
f(z, a) -> F(z) uniformly in every closed domain
within C, then

F(z)

is

regular within

C and
dz

uniformly in D.
5.5. Analytic functions

denned by

definite integrals
Let F(z, t) be a one-valued function of the complex variable

= x+iy

and the

a closed contour

real variable

C and a
F(z,t)

where

<f>

and

tfi

^ <
t

b.

defined

when

z lies within

Let us write

= <f>(x,y,t)+ii/,(x,y,t),

are real functions of the three real variables

t The existence of a largest 8


section of the real numbers.

is

readily proved

by means

of Dedekind's

UNIFORM CONVERGENCE

108
x, y,

If

t.

and

<f>

are continuous functions of position in the

ift

corresponding region of the three-dimensional space in which


(x, y, t) are rectangular cartesian coordinates, we say that F(z, t)
is a continuous function of both variables z and t.

We

shall

now show

both variables

when z

that

lies

if F(z,

t)

is

a continuous function of
C and a ^t ^ b,

within a closed contour

and

if, for each such value of


regular within C, the function

t,

f(z)

F(z,

t)

is

an

analytic function,

F(z,t) dt
J*

is also regular

within

G and

its

derivatives of all orders

may

be

found by differentiating under the sign of integration.


To prove this, we divide the range of integration into n equal
parts

by

points

<t1 <t2 <...<

tn _ x

<tn =

and consider the behaviour of

fn(z)

= ZF(z,Wr-t -l)
r

r=l

as

-> oo. Since F(z,

t)

is

a continuous function of

t,

fn (z)^JF(z,t)dt=f(z)
a

as

-> oo, for each value of z under consideration.

We now show that fn (z) tends to its limiting function f(z)


uniformly in every closed domain
within C. Since F(z, t) is
a continuous function of both variables when z lies in
and

=5* t

^ b, it is uniformly continuous there.

positive

on

e,

number

e,

Hence, given any


we can find an integer to, depending only

such that the inequality


\F(z,

holds

t)-F{z, t')\< e

when z is any point of D provided that

But, since

\t't\

/.(*)-/()

=2

'- 1

tL

{F(z,tr )-F(z,t)}dt,

<

(b~a)/m.

UNIFORM CONVERGENCE

109

we have
l/(*0-/()l<i

> m,

provided that n

\F(z,tr )-F(z,t)\dt<(b~a) e

and so the sequence

uniformly con-

is

vergent.

By

the theorem of

in D.

But

follows that f(z) is

5.13, it

and

function, regular within C,


since

an analytic

also that/4(z) ->f(z) uniformly

AM-T^fc-t-J.
r=l
b

this implies that

/' (2)

8F
f
J

Z
'

dt.

8z

The derivatives of higher orders may be discussed in the same


manner. This completes the proof of the theorem.
Example. Show that the equation
2w

2v

dt

J- 1+zsini

V(i z ")

holds everywhere in the z -plane, supposed cut along the real axis from
-co to 1 and from +1 to +00, provided that the branch of >J(l z 2 )
which reduces to
1 at the origin is taken.
The integrand is a continuous function of both variables, save when

= cosec*. Now as

real axis

from 00 to

varies from

1 and

to

moves along the


increases from 77
and then back to

this point

77,

then back to

00;

as

moves along the real axis from + 00 to +1


Thus the integrand is a continuous function of both variables
when t is real and z lies in the cut z-plane; moreover, for each such value
of t, it is an analytic function whose only singularity is a pole lying on
to

277, it

+ co.

one or other of the

cuts.

Hence, by

5.5,

the value of the integral

an analytic function, regular in the cut plane.


We can, however, show by elementary methods that, when

1 < z <

is

1,

the value of the integral is 2tt/J(1z*), the positive square root being
taken. But the branch of 27r/V(lz 2 ) which is positive when
1 < z < 1
is an analytic function, regular in the cut plane.
We have thus shown that the expressions on each side of the equation

2tt

dt

/1+zsini

277

^/(l

2
)

are regular in the same cut plane and are equal when
1 < z < 1. By
analytical continuation, this equation holds everywhere in the cut plane.

UNIFORM CONVERGENCE
by infinite integralsf

110

5.51. Functions defined


Let the function F(z,
(i)

t)

satisfy the following conditions:

a continuous function of both variables when z lies


within the closed contour C and a
t
T, for every
finite value of T;

it is

< <

(ii)

(iii)

for each such value of


regular within C;
the integral

an

it is

t,

analytic function of

z,

= JF(z,t)dt

f(z)

z lies within C and uniformly conwhen z lies in any closed domain D within C.
Thenf(z) is an analytic function of z, regular within C, whose
derivatives of all orders may be found by differentiating under the
is

convergent

when

vergent

sign of integration.

The

third condition

means

T tends to infinity,

that, as the real positive

number

T
j F(z,t)dt
a

tends to f(z) when z is any point within


vergence is uniform when z lies in D.

C and

that the con-

Let us consider, then, the behaviour of


n

n (z)

= j F(z,t) dt
a

as the integer

n tends to infinity. By

5.5,

fjz)

is

regular within

C and satisfies all the conditions of the theorem of


f\z) is regular

in/).

5.13.

Hence

within C, and f(z) converges uniformly to

f'(z)

But

a
00

and so

/'(*)= [
J

8F{z

'

t]

dt.

8z

This completes the proof of the theorem.


t It is assumed that the reader is acquainted with the theory of the convergence of infinite integrals, as given, for example, by Hardy, Pure Mathematics (1921), Chap. VIII.

UNIFORM CONVERGENCE
uniform convergence

5.52. Tests for the

111

of infinite in-

tegrals

The simplest

test for the

uniform convergence of an

integral is the analogue of Weierstrass's

runs as follows.

-test

infinite

5.21),

and

be a continuous function of t when


z lies in a bounded closed domain
and t^a, satisfying at each
point of
the inequality

Let F(z,

t)

< M(t),

\F(z,t)\

where M(t)

is

a positive function, independent of

CO
J"

Then,

z.

if

oo

M(t) dt converges, the integral J F(z, t) dt

is

uniformly and abso-

lutely convergent in

D.

T
Since F(z,

t) is

a continuous function of

t,
J"

F(z,

t)

dt exists

for every value of


CO

T (> a)

and

for each point z in

D.

By

hypothesis, J
(t) dt converges; we can, therefore, assign arbitrarily a positive number e, and then choose T, independently

of

so that

z,

j M(t)
T

But

if

T'

<

dt

e.

> T, we have
2"

j
T

2"

F{z, t)dt

<j

co

\F{z,

< j M(t) dt < e

<tt

CO

Hence
lies

J F{z,
a

t)

dt converges absolutely

and uniformly when

in D.

To

discuss the uniform convergence of integrals which do not


converge absolutely, the following tests, similar to those of 5.22
for series, are frequently of service. Let u(z, t) and 8v(z, t)/8t be

continuous functions of

when

z lies in

a bounded

closed

domain

CO

D and f^ a. Then the integral J u(z, v(z, dt is uniformly cona


vergent in D if either of the following sets of conditions is satisfied:
t)

I.

(i)

u(z,t) dt

^ K, where K is

t)

independent of z and T,

TINIFOKM CONVEKGENCE

112
QO

~-^ dt

(ii)

uniformly and absolutely convergent in D,

is

St

and
(iii)

v(z,

->

t)

as

-> oo uniformly with respect to

z.

CO

II.

u(z,t) dt is

(i)

uniformly convergent in D,

CO

\8v(z,t)

dt converges

and

is

dt

a bounded function of

z,

and
(iii)

For

v(z,

a) is

a bounded function of z in D.

we shall only prove this theorem under conThe proof under conditions I follows very similar

brevity,

ditions II.
lines.

The operation of partial summation on which the proof of


the corresponding tests for series depended is replaced here by
by

integration

parts.

Let us write
t

V{z,

Then,

if

>

T'

u(z,

t)

T,

v(z,

t)

t)

f u(z,

t) dt.

we have
dt

= {U(z, T')-U(z, T)}v{z, T')~

T
T'

j{U(z,t)-U(z,T)}

^dt.

Now, by condition II (i), we can assign arbitrarily a positive


number e and then choose T, independently of z, such that the
inequality U{z, t)U(z, T) < e holds when t > T and z is any
point of D. Hence
|

3"

u(z,

2"

t) v(z,

t)dt

<

J \v(z, T')

\dv(z,t)\

dt

dt
2"

6v(z,

v(z,a)+

t)

dt
j

8v(z,t)

dt
dt

UNIFORM CONVERGENCE

113

2"

<.{|tfc,)|

+ 2j|*j^

dt\

where, by II

(ii)

independent of

and

Il(iii), the finite positive


number
T, T'. This, however, shows that

z,

u(z, t)v{z,

t)

is

dt

converges uniformly in D.

Example

1.

Show

that

J-*/
when Rlz >

0,

provided that the branch of Vz which

is

positive

on the

real axis is taken.

Let

D be any bounded closed domain which contains part of the real

and

axis

lies entirely

inequality

Rlz

>

8,

to the right of the imaginary axis, so that


the
> 0, is satisfied at each point z of
Now

where 8

00

and Se-M'dt

is

convergent.

Hence, by the

Jtf-test,

converges uniformly and absolutely in D, and


function of z, regular in D.

When z is real
substitution zt 2

its

the given integral

value

and positive, we can evaluate the


w 2 which gives

is

an analytic

integral

by the

/^*=ij^*=iy(i).
the square root being positive. Hence the expressions on each side
of
the equation

/-*~i/(i)
are analytic functions of z, regular in D, which are equal on the
positive
part of the real axis, provided that we take that branch of the square
root which is positive when z > 0. It follows, by the principle
of
analytical continuation, that the equation still subsists
everywhere
ini).
4111

UNIFORM CONVERGENCE

114

Example

2.

that the result of Ex.

Show

holds

when Rlz

>

0,

GO

provided that z

Deduce the value of

0.

J"

cos<

dt.

o
00

We

prove,

of

first

that, if

all,

>

-2" dt converges uniformly


J e

0,

when r
number
|e-*|

<
r

Now

when Rlz

if

[argz[ < w, no matter how small the positive


\z\ < R,
may be. The ilf-test is obviously inapplicable here, since

is

any

0.

positive

number less than r, we have seen that the


when Rlz > 8, \z\ < R. Accordingly it

integral converges uniformly

demonstrate its uniform convergence in the two rectangles


denned by
< Rlz < S, p < |Imz| < R, where p = V^-S 2 ) > 0.
This we do by writing the integral in the form
suffices to

e-xfe-W at,

where z

x+iy, and then applying the second


u(z,t)

Now, when

<

<

Ur

<

S,

iytf

v{z,t)

<

\y\

test of 5.52,

with

e-^Jt.

we have

R,

= ^(e-^'-e-^^) < 1,
12*2/

f u(z,t) dt
J

'I

so that condition I(i)


constant sign and
{

Again

is satisfied.

Bv(z,t)

j _e-*T"

e-

I(ii) also holds; for 8v/dt is

of

er^

as

T -> oo,

uniformly with respect to

Finally, since

z.

\v(z, t)
|

<

1/t,

00

condition I

satisfied,

(iii) is

and

so

e~*

dt

converges uniformly.

a
00

We

have thus shown that

J"

e_J*' dt

converges uniformly^

when

<

< R and

<

and so represents a regular analytic


domain for which Rl z > 8, the value
of the integral is ^(tt/z). Hence, by analytical continuation, this is the
value of the integral when Rl z > 0, provided that z =#= 0.
Putting z = eei we have

\z\

function.

But

|argz|

\n,

in the part of this

oo

exp( e^a )

dt

i-Vjre-* 84 .

t Actually we proved this result with a as lower limit. This, obviously,


does not affect the result but considerably simplifies its proof. It should be
and also when RI z < 0.
observed that the integral diverges when z

UNIFORM CONVERGENCE

115

CO

In particular,

dt

e-**'

JV^e"-"*/ 4,

o
00

and so

CO

cost* dt

jsin a

d<=

/^.

REFERENCE
Bbomwich,
Appendix III.

T. J. I'a

Infinite Series

(London, 1926), Chaps. VII,

X and

MISCELLANEOUS EXAMPLES
CO

Show

1.

that the series 'ze~ nz converges absolutely but not unio

formly in the sector \z\ < B, |argz| < 8, where


<8
convergence uniform when r < \z\ < B, |argz| < 8?

Prove that the sum of the

2.

z
is

<

J7r.

Is the

seriesf

\zn

jL-i

an analytic function whose only


= 0, 1, 2,....

n!

singularities are simple poles at the

points

oo

Show

3.

^an n~ converges
but is not necessarily convergent when Rl z = 0. Prove

that, if ~2,an

convergent, the series

is

when Rl z

>

0,

also that the convergence


for

which Rlz

4.

The

>

series

where 8

2 am (z
a

a non-integral value of
bounded closed domain.

is

uniform in any bounded closed domain

>

0.

)(z 2

Show

z.

)...(z

that

it

2
) is known to converge for
converges uniformly in any

Discuss the convergence of the following infinite products:

5.

wftfiSH

<i>f[(*3.
1

Show

also that,

when

\z\

<

1,

the value of the product

If J

6.

<j>(t)

dt converges,

show that J e~a^>(t)

gent
t

(iv) is

1/(1 z).

CO

CO

when

dt is

uniformly conver-

o
\z\

The accent

<

B, |argz|

<

8, if

indicates that the

<

<

\n.

term corresponding to n

is

omitted.

UNIFORM CONVERGENCE

H6
00

7.

-' 8 + 2rf

Show that

dt is

j e

uniformly and absolutely convergent in

00

any bounded closed domain and that

its

value

is Vire*

Deduce that

00

f e~p

cos 2tydt

Vwe-"'.

00
8.

The function

not tend to a

F(z,

t)

the conditions of 5.5 save that

satisfies

finite limit as

->

it

does

but

a,

6
f F(z, t) dt
a

lim

.F(z,

*)

(S

>

0)

a+o

exists. Show that if this limit is approached uniformly when


any bounded closed domain within C, the function

z lies in

/()

= j F(z,t)dt
a

may be

analytic function, regular within C, whose derivatives


calculated by differentiation under the sign of integration.
is

an

00

OO

it - 1 cost
1

Showthatt

9.

\tz- 1 sintdt

dt,

represent analytic functions of z which are regular

and

1 <

10.

Biz

<

when

<

Rl z

<

respectively.

Prove that the equation


T(z) =; f e-H z 1 dt

an analytic function of z, regular


zT(z).
the difference equation T(z+1)
defines

when Rl >

oo

t Consider

and

separately.

0,

which

satisfies

CHAPTER VI

THE CALCULUS OF RESIDUES


6.1.

Cauchy's theorem of residues

Let f(z) be continuous within and on a closed contour


regular, save for a finite number of poles, within C.
Then
J

f(z)

dz

2irix{sum of residues of f(z) at

and

poles within G).

its

Fio. 2

If we denote

by zx z 2 ,..., zn the poles of/(z) within C, we can


draw a set' of circles Cr of radius e and centre z
r
which do not intersect and all lie within G, provided that e is
sufficiently small. Then f(z) is regular in the domain bounded
externally by C and internally by the circles G We can, therer
,

evidently

fore,

deform

f(z) until it

continuously without crossing a singularity of


consists of the circles Cr joined together by a poly-

gon P, as shown in the

figure.

ff(z)dz= \f{z)dz

We then have
(f(z)dz

=f

ff(z)dz,

the integral round the polygon


vanishing since /(z)
within and on P.
Let zr be a pole of order m, say, so that

f(z)

= <f,(z)+y 7^-

is

regular

THE CALCULUS OF RESIDUES

118

where

<f>(z)

regular within

is

1X
S~

nT
C

Now

on

Cr

= z +ee
we

jf{z) dz

Then

(i.
Cr

where 6 varies from

to

2tt.

Making

find that

= 2=

1_s
s

Cr

=
From

6i

this substitution

and on Cr

2iri

eP--#Hd6
J

X (residue

2ma x

of /(z) at zr ).

this it follows that

(f(z)de

= i
=

f/(z )efe

2 (residue of/(z) at z

2vix

r ).

This completes the proof of Cauchy's theorem of residues.


6.2.

The number

of zeros of

an analytic function

An

important deduction from Cauchy's theorem of residues


a formula for the number of zeros of an analytic function
within a given closed contour. Let f(z) be regular within and
on a closed contour C, save for poles 6 1; b 2 ,..., b n none of which
lie on C. Moreover, let f(z) not vanish on C but have zeros
av a2 ,..., am within C. Then, by the theorem of residues,
is

_L [11
M
M'^dz
J

equal to the sum of the residues of/' (z)//(z) at its singularities


within 0.
Now the only possible singularities of this function are the
poles and zeros of f(z). If a is a zero of order r, we have
is

f(z)

where

<f>(z)

is

regular

of a. This gives

since

(j>'(z)/</>(z)

{z-aYftz)

and non-zero in a
(z)
f(z)

But

is

certain neighbourhood

^
z-a+W
(z)

regular in the neighbourhood of a, the

function f'{z)jf(z) has a simple pole of residue r at z

= a.

Simi-

THE CALCULUS OF RESIDUES

=6

119

a pole of order s of the function /(z), then


z
has
a simple pole of residue
s at z
f'( )/f(z)
b.
If, therefore, the order of the zero at z
ap is rp and the
order of the pole at z
bp is s
p we have shown that
larly, if z

is

=
=

This result takes a particularly simple form if we agree to regard


each zero of order r as equivalent to r simple zeros and each
pole of order s as equivalent to simple poles; for, with this

convention,

_L
2tt

O
,,.

*<*>

dz

J f(z)

c
equal to the excess of the number of zeros of /(z) within
over the number of poles there.

is

If

we

excess

is

actually carry out the integration,

we

find that this

equal to

^Pg/(z)]c
since log /(z)
|

= ^[arg/(z)]c

2^[logl/(Z )l+iarg/(z)] c

returns to its original value

round C. Hence

the excess of the

when we go once

number of zeros over

the

number

of poles of f(z) within C is (l/2n) times the increase in arg/(z) as


z goes once round C. This result is sometimes called the principle
of the argument.

Example 1 . If g(z) is regular within


conditions of 6.2, show that
C

J
c

Example

f'(z)

2. If f(z)

and on G and f(z)

g{z)dz=

is

9(a" }

-^

the

8" 9{b

and on C and does not vanish


of the zeros of /(a) within C is

regular within

on C, show that the sum of the

affixes

_1 [ ZI(

iidj
2TriJ
o

dZ '

f(z
f(z)'

the affix of a multiple zero being repeated according to

6.21. RouchS's

satisfies

its order.

theorem

Rouchef has proved


regular within and on a

that, if /(z) and g(z) are two functions


closed contour C, on which f(z) does not

f Journal de Vtcole Pol.

39

(1862), 217.

THE CALCULUS OF RESIDUES

120

\g(z)\ < \f(z)\, then f{z) and f{z)+g{z) have the


same number of zeros within C.
For let f(z) and f(z)+g(z) have respectively m and n zeros

vanish and also

within C, multiple zeros being counted according to their order.

Then the function

has n zeros and

,.

poles within C, and is regular and never


the principle of the argument, 2n(n m) is
equal to the increase in the argument of F(z) as z goes round G.
But, when z is on C, we have

By

zero on C.

m{F(z)}

^ l

m >0.

Hence

as z goes round C, F(z) describes a closed path entirely


to the right of the imaginary axis, so that its argument returns
to its original value; and this proves Rouche's theorem.

In particular,
f(z)

we take
m
a z
g(z)
if

g()

we have

=a
,

zm

-1

+a 2 z-*+...+am

*m 1

\a z\

which can be made as small as we please by taking \z\ sufficiently large. Hence there exists a circle \z\ E on which /(z)
does not vanish and also \g(z)\ < \f(z)\. But f(z) has one zero of

order
within this circle; therefore, by Rouche's theorem, the
polynomial a zm +a1 zm -1 +...+a has precisely
zeros within
m
the circle \z\
R for all sufficiently large values of R. If these

zeros are z v z 2 ,..., z m multiple zeros being repeated according


to their order, the function
,

a zm +a lZm -i+...+am

(z-z x )(z-z a )...(z-zm )


is

an

which tends to a as \z\ -+ oo, and so has


values of z, by Liouville's theorem. Hence

integral function

the value a for

all

aQ z'+a 1 z^+...+am

= a (z- Zl )(z-z )...(z~zm


2

).

We have thus proved the fundamental theorem of the algebra

THE CALCULUS OF RESIDUES


of complex numbers, that a polynomial
of degree
and can be expressed as a product of
linear

Example

121

m has m zeros

factors.

Prove that a rational function takes any given


value

1.

p tunes where p is the number of its poles, including the point at infinity,
a pole of order

being counted s times.

Example

2. Each function of the sequence


/ (z), /^z), /a (z),... is
regular in the closed domain
bounded by a closed contour C. The
sequence converges uniformly to/(z) in D. Show that,
if/(z) does not
vanish on C, then f(z) and the functions /(z), for all
sufficiently large
values of n, all have the same number of zeros within G.
Prove also that a zero of/(z) is either a zero of /(z) for all sufficiently
large values of n or else is a limiting point of the set
of zeros of the
functions of the sequence. (HuBWiTZ.f)

6.22. Inverse functions

Let f(z) be an analytic function, regular in a neighbourhood


of
at which it takes the value w
The necessary and
sufficient condition that the equation f(z)
w should have a unique
solution z
F(w), regular in a neighbourhood of w is thatf'(z
)
should not vanish.
the point z

The condition is obviously necessary. For if F(w) were regular


F'(w ) would be finite; since, however

in a neighbourhood of w

*"K)

l/f(*o),f{zo) cannot be zero.

The proof of the

sufficiency of the condition is

the one given below

much more

due to Landau.J We shall suppose that z and m> are both zero; for, if they were not, we could
make the transformation z'
zz w'
ww By hypothesis, f(z) is regular when \z\
R, and so can be expressed
there as a convergent Taylor series
difficult;

is

<

= a z+a z +...,
= a > 0.
|/'(0)| = \a
3

f(z)

where

Now

if z x

where A

<

and

1,

z 2 are

any two points of the region

\z\

we have

< XE

/W-/W = L + | an{zr + zn%+ ...+zri


i

t Math. Annalen, 33 (1889), 246-66.


t Berlin SUzungsberichte (1904), 1118-33; (1926), 467-74. Math.

30

(1929), 616-17.

Zeitschriff.
J

'

THE CALCULUS OF RESIDUES

122

Hence

small that

if A is so

f |aJAi-R"- < a,
1

the equation

Moreover,

if

\f'(z)\

=/(z) can have at most one root in \z\ ^ XR.


such a root exists, it is not a multiple root; for

k+ fnan z-A > a- fnlaJA^B"- >


1

But, by hypothesis,

f(z) is regular in

^ M,

and

|/(z)|

< M/R,

\an

<

i?

is

\z\

where
Hence, by Cauchy's inequalities, we have
there an inequality

0.

< M/Rn

and

so satisfies

finite constant.

therefore

- ^ _ A)2 < jj _ A)8


(1
(1
If we now take A = \Ra\M, so that A ^ \, we easily find that
\n\an \Xn - Rn ~ < 8o/9.
Hence the equation m> = /(z) has at most one simple root in the
region
< \R?a\M.
2,K|A

^ 2, ^

if

\z\

This, of course, implies that f(z) has but one zero


\z\

<

\R2 ajM, namely the known

when

simple zero at the origin.

We

deduce from this, by the aid of Rouche's theorem, that/(z) w


has precisely one simple zero in the same region, provided that
w is not too large.
Now when \z\ %R2a/M, we have

|/( 2 )|

>\a1 z\-f\an *\
2
oo

^a\z\-M 2\ Z/R\ n
2

4M

4M

Z,
2

by making use of the

UM

'

4JH-Raj
fact that

Rouche's theorem, that/(s)

Ra

^ 4M

*'

6M

^ M We now see, by using


.

has just as

many

zeros as/(z)

ww

THE CALCULUS OF RESIDUES


in the region

other words,

< iR

\z\

a/M, provided that

\w\

123

< R a?/M.
2

has precisely one simple zero there;


this zero by F(w).
Finally,
for F(w),

it

by

6.2,

Ex.

2,

In

us denote

explicit

formula

namely

= 2 f g- dz,
J f(z)w
c
C denotes the circle
= \R alM. On the contour
z = \Waeu\M,
*<)

where

we can obtain an

let

\z\

C,

where t varies from


function of

to 2v, and so the integrand is a continuous


and w. Hence, by 5.5, F(w) is an analytic func-

M<

tion of w, regular when


\R2a2 \M. This completes the
proof of the sufficiency of the condition.
It should be observed that we have not proved that J.R 2 a2 /Jf
is the radius of convergence of the Taylor series for F(w) in
powers of w. All we know is that the radius of convergence is
not less than ^RW/M.

Example. The function f(z) is regular in a neighbourhood of the


point z at which it takes the value w
Show that, if the first
1
derivatives of/(z) vanish at z the equation /(z)
w has a solution
,

z-z

= 1 bn(w-w
l

)lr,

where the series of powers of (w a ) x lv converges in a neighbourhood


of wa Deduce that z is a jp-valued function of w, having a branch -point
.

at

6.23. Lagrange's

formula for the reversion of series

We have just seen that, if


+
f(z) = w +a 1 (zz )+a 2 {zz
2

is

regular near z

there

is

(%

...

# 0)

a unique function

= zo+bjiw )+b (w +...,


regular near w such that z = F(w) is the solution of the equaF(w)

tion

== f(z).

It is possible to obtain the coefficients b n

ww 2 an z z n *ne power
equating coefficients. A more elegant

substituting in the equation


series for

method

ww
we

o)

and

can, however, be obtained

theorem, as

by

shall

now

show.

by means of Cauchy's

THE CALCULUS OF RESIDUES


6.22, we have

124

In the notation of

where

by

the circle \z

is

integration

by

^ J f(z)w
c
= lB?a/M.

From

this it follows,

parts, that

dz
27rtj f(z)-w'
,

and so

We have,

however, shown that

\m-w > fBW/^


\

when
where

z lies

on the contour G. Hence

if

< <

A
1, the infinite series under the sign of integration in the formula for F'(w) converges uniformly with respect
to z and so can be integrated term by term. This gives
F'(w)

where

nb

= f n6 n (w-w )-i,
n=l
1

dz

2w
2?rt J {/(*)->}*'

Since f(z)w has a simple zero at z and vanishes nowhere


on C, the coefficient nbn is evidently the residue
n a* the point z
of
z ) w
simple method of

else within or
{f(

is

o}~

to proceed as follows. If

<f>(z)

is

regular within

we

write

and on C, and therefore

finding b n

THE CALCULUS OF EESIDUES

125

If we substitute this value of b n in the series for F'(w),


find that iff{z) is regular in a neighbourhood of z and iff{z )

we

f'(z

^ 0, then the equation


f(z)

=w

has a unique solution, regular in a neighbourhood ofw

where

ftz)w

This

is

(zz

of the form

)/<f>(z).

Lagrange's formulaf for the reversion of a power

more general

It is a particular case of the following

series.

expansion

which

is

also

due to Lagrange. The proof of

this is left io the

reader.

6.3.

The evaluation

The

of definite integrals

devoted to one of the first applications which Cauchy made of his residue theorem the evaluation of definite integrals. The method to be adopted in any
particular case should be clear after a consideration of the
typical examples discussed below.
It should, however, be observed that a definite integral which
can be evaluated by Cauchy's method of residues can always
be evaluated by other means, though generally not so simply.
On the other hand, quite simple definite integrals exist which
rest of this chapter is

00

cannot be evaluated by Cauchy's method,

x*
J e~

dx being a case

in point.

We discuss

here three

2ff

/(cos 6, sin 9) d0,

main types of

definite integral,

namely

oo

oo

f(x) dx,

a;"-

/^)

dx.

-co

We also show how the values of certain integrals, usually determined by a complex change of variable, can be
means of Cauchy's theorem.
t Mimoirea de VAcod. Roy. des. Sci. (Berlin)

24

easily

found by

(1768), 251.

THE CALCULUS OF RESIDUES

126

The evaluation

6.4.

of

/(cos0, sin0) dd

f
o

If /(cos 0, sin 0)

and

sin 6

which

is

a rational function of the two variables cos


on the range of integration, we make

is finite

the transformation z

=e

9i

The

becomes

integral

g(z) dz,

where g(z) is a rational function of z, finite on the circle C


whose equation is \z\ 1. The labour of evaluating the residues
of g(z) may often be considerably lightened by preliminary
manipulation of the integral, before introducing the complex
variable

z.

Example

1.

Evaluate

J^ 2aJL
<'
z 22J_ sin

where a

is positive.

If

we denote the

by

required value of the integral

I,

we know

that

saw

_1
~

acUf)

I"

2 J o a +sir
-sin 2 ^'
o

since the integrand

is

periodic, of period

-n.

We may now

evaluate the

by the method of contour integration by making the substitution z = e**. This, however, leads to an integrand containing a polynomial of degree 4 in the denominator, so that the work would be rather
integral

laborious.

A simpler method is to write


TT

_
=

27T

Tt

a ^4>

2
2
J a + sin <

_
~

2ad^>

+ 2a cos2^
2

add

+ 2ra cos0'
2

and then

to put z

em .

In this way we find that


2ai

22

22(l + 2a

rdz,

)+l

where

denotes the circle |z|


I.
The integrand has simple poles at the points l+2a2 2oA/(l + o2 ) and
(l
2a 2 ) 2aA/(l a 2 ). If we take the positive square root, the former
point lies outside C, the latter within C; moreover, the residue at the
pole within O is
,

From

this

it

follows,

2V(l+2 )'
by Cauchy's theorem

= 7r/V(l+a

2
).

of residues, that

THE CALCULUS OF KESIDUES


Example

2.

Evaluate J

e coa9 cos(nB sin0) dd,

127

where n

is

a positive

integer.

Consider

2ir

e ooee {eos(n6 sin0) isin(n9 sm6)}dd

o
2jr

-.

e cos8+islnfl-ei tfQ

e^-"- 1

dz.

The

origin

there

is

-1
z
1 of the function e z~"
, and the residue
277/n!.
Since the integrand has no other singularity, J

a pole of order

is 1/ra!.

n+

Equating real and imaginary parts, we have


27r

f e cos0 C os(n0 sin0)

d0

= 2~,

til
2w

I
6.5.

e>80sin(0 sin0) ^0

The evaluation

of
f

f(x)

0.

dx

save
If the function /(z) is regular in the half-plane Imz
possibly for certain poles which do not he on the real axis, we

can evaluate

00

by considering the

integral of f(z)

round a closed contour, con-

R to R and

a semicircle in the
from
upper half of the z-plane on this segment as diameter, provided
that the integral round the semicircle tends to a limit as R -> co.
sisting of the real axis

The simplest case occurs when the integrand

is

0(\z\~ k )

Since the length of the semicircular


1_fc
it is 0(i?
) when
- 00, prois large, and this will certainly tend to zero as
vided that k
1. When, however, the integrand is not of the
order of \z\~ k , where k
1, a more delicate type of argument
is needed to determine the limit of the integral round the semi-

for large values of

part of the contour

\z\.

is

nR, the integral along

>

>

circle.

THE CALCULUS OF RESIDUES

128

method

It will be observed that this

gives

R
lim

f(x) dx.

00

If

Jti

j f(x)dx
f

exists in the ordinary sense.f it

is

necessarily equal to this limit.

may, however, happen that the

It

even though the


the limit the Cauchy

limit exists,

In this case we

integral does not.

call

00

and write

principal value of J f(x) dx,

lim

dx

f f(x)

=P

~R

f f(x) dx.

oo

Finally, it should be noticed that there is no special merit in


a semicircle. All we need is a curve joining the points iJ
which tends to the point at infinity as R -> oo. For some purposes, the rectangle with vertices i?, R-\-iR is more con-

venient.

Example

1.

Evaluate

x+2

/
We

consider

a;

+10a; 2 + 9

dx.

-J^g*-*,,

/=

r
the contour consisting of the real axis from
to
and
the semicircle in the upper half-plane on this segment as diameter. The
integrand has simple poles at the points *> 3t; when R > 3, the poles
i and Zi, which have residues
(l+i)/16 and (3 7t)/48 respectively,
lie within T. Hence, by Cauchy 's theorem of residues,

where

is

I
t

The

5tt/12.

integral exists in the ordinary sense

when

oo

J f(x) dx

f(x) dx,

oo
converge separately

that

is,

when
s
f

f[x)

dx

B
tends to a limit as It and

jS

tend to infinity independently

THE CALCULUS OF RESIDUES


Now, when

129

the integrand is 0(|z|- 2 ); we should, therefore,


expect that the integral round the semicircle would be
0(1/ R) when
is large, and so would tend to zero as
R -> oo. To put this argument in
a rigorous form, we write z
Rem when z lies on the semicircle, to obtain
\z\ is

large,

x*-x+2

J*4 +
-R

BV"-BV"+2B&

J=

where

(i? 2 e 29*+l)(.R 2 e 29*+9)


o

Remembering the important inequality^


1

ah,^
we

find that the

\a\-\b\-

modulus of the integrand does not exceed

R3 +R* + 2R
(i? 2 -l)(i? 2

when

R>

3.

1
'

so that

We have

=>-

-9)

Hence

as 12 -

(i? 2 -l)(J? 2 -9)'

oo.

thus proved that

-*+2

+ 10a; + 9 ax
2

5v
12

~R

But, since the integrand behaves like

1/a;

for large values of x,

x 2 x+2
zdx
i
i
x
I- +X0x +9
>

exists in the ordinary sense

and

has, therefore, the value &ir/12.

00

efo,

where a

>

0.

16

00

is

z 2 -\-a%

dz

taken round the contour of Ex. 1. When R > a, there is only one pole
of the integrand within T, namely a simple pole at ai of residue
e-/(2ai); hence
2 =

^^

t See
4111

1.23.

*t-8

THE CALCULUS OF RESIDUES


x+iy and y > 0, we have
Now, when z
to-| _ 6 -v
< 1,
e&| = e
We
half-plane.
upper
so that e is bounded in the

130

fa

should, therefore,

Y would be 0(ljR),
R -> oo. To prove this,

expect that the integral round the curved part of

when R is
we write z

and so would tend


Be94 when z lies on the

large,

to zero as

semicircle to obtain

R
~~

J=

where

,dx+J,

+a2

a;

'

ti5cos8 Usinfl

^Be8<

e
- <#
p2 aW +a*
,
JB*e
2fli _i

From

this it follows that


7T

7|

-^

e -SaiaBM

<

so that

as

R -> oo.

We have thus proved that

the integral
0C

x z +a'

exists in the ordinary sense, has the value


and imaginary parts, we find that

which obviously

we now
+,

Ja.

If

>

C cosx j

ire~

Example
r

siax j

-co

_oo

>

-rre

equate real
as

dx,

3. Evaluate

iK sin. oc

-=-;

ir^

s
a'*

dx,

when a

is positive.

We shall find the value of this definite integral by considering the


*
contour integral
fe
T

Here T denotes a closed contour consisting of the segment of the real


axis from R to R and a curve a in the upper half-plane which joins
the ends of this segment and tends to the point at infinity as R -> oo.
Evidently, for sufficiently large values of R, there is no singularity of
the integrand on T and one singularity, namely the simple pole ai,

-iU,

within r.

By

THE CALCULUS OF RESIDUES


Cauchy's theorem of residues,

follows that

it

= me~a

for all sufficiently large


values of R.
can also express / in the

We

form
R
dx +J,

where

zei*z

z*+a~2'

deIicate
of argumentTsIeeded
^
oTllttto^
*
S *hlS
* keeP the semicircular contour a and
l Jordan's
T1 inequality; this explained in
apply

mOTe

1S

fl

is

trtzsT

use a rectanguiar
CaSS
C nSiStS f thr6e
*

th

ifwtfaiyU^

A
w ^**.
tx
^
6.52,

s to

below

more

f this

-ie
is
*

R+iy)e- x-v
{

"(-iJ+M/T+^r*^
say.

We then have
R
2

2
fV(-B +2/ )e-",

U\<

.RV2

f
J

R*J2
so that

^ -^

as i? -> oo; similarly

J3

|Jil<
-is

Adding th6Se
^ * "*
tTt J t
R is increasedfindefinitely.
to zero as

results >

we *"* t^t J

tends

THE CALCULUS OF RESIDUES

13 2

Another simple method of obtaining the limit of J is to keep the


semicircular contour or and integrate by parts, f This gives
f
J

ze

use<M*

f(a

*+&** = " L*M^]-s +


,

We

-s 2 )^

"

and the two terms on the right-hand


to zero as

1^+oY

side of this equation clearly

tend

R -> co.

have thus shown that

B
lim

This, however, implies that

-j-

J x +a
2

= me

da;

the existence of this infinite integral in the ordinary sense being an easy
consequence of Dirichlet's test. J Equating real and imaginary parts,
we obtain
K
OT
f a;cosa;
,

2
2
J x +a
I

,
dx

fxsinx

..

-j-^x dx

0,

2
2
J K +a

CO

From

= ire

a.

00

this it follows that


a;

I
when a

is

sin a;

+ a?

x*

da;

positive.

Example

4.

Prove that
a;

J
provided that Bla
HI a is negative ?

Example

5.

is

Jx

Show

da;

+a?f

What

positive.

=
is

it

8a"3

'

the value of this integral

when

that

dx

(a;

when the

\trer a ,

real parts of

+ aa

2
)

(x 2

+6

a and b are

2
)

TT(2a+b)
2a 3 6(a+6) 2

positive.

I owe this remark to Mr. W. L. Ferrar.


j See Bromwich, Infinite Series (1926), 477. This test is essentially the
second test of 5.52, above, under conditions I with the references to uniformity
omitted.
f

THE CALCULUS OF RESIDUES

133

Example
the upper

6. f(z) is an analytic function whose only singularities in


half-plane are poles, finite in number, and which has no

on the real axis. Show that,


uniformly with respect to 6 when

singularity

R -> oo,

if

Rf(Reei ) tends

<

<

-n,

to zero as

then the principal

CO

value of J f(x) dx
at

its

is

equal to

2iri

times the

sum

of the residues of f(z)

poles in the upper half-plane.

6.51. Cauchy's definition of

an improper integral

The ordinary definition of the integral of a function f(x) of


the real variable x over a finite interval a
x
b presupposes
that f(x) has a definite finite value at each point of the interval.
We shall now explain how Cauchy extended this definition to
cover cases when f(x) is infinite at a finite number of points of
the interval.

< <

It suffices to consider the case when there is only one point c


at which f(x) becomes infinite. If c is not an end-point of the
interval,

we take two

small positive numbers

and

and con-

-q

sider the expression

ce

f(x)

dx

J"

f(x) dx.

j
c+t]

and tends to a unique limit as e and rj


tend to zero independently, we say that the improper integral
of f(x) over the interval exists, its value being defined by
If this expression exists

f(x) dx

lim

e
f

f(x)

dx

lim

-j-

'

f(x) dx.

7)

however, the expression does not tend to a limit as


tend to zero independently, it may still happen that
If,

r)

!c-e
j

f(x)

dx

When

and

>

j f(x) dx\

-j-

c+e

a
exists.

this is the case,

we

Cauchy
by

call this limit the

principal value of the improper integral and denote

it

P j f(x) dx.
a

Finally, if f(x)

becomes

infinite at

an end-point, a

say, of

134

THE CALCULUS OF RESIDUES


we say that f(x) is

the range of integration,

< x < b if

integrable over

lim

e-^ +

f(x)

dx

exists.

When we attempt to determine the value or principal value


of an improper integral by means of Cauchy's theorem of residues, we have the difficulty that the integrand has a singularity
c on the contour of integration. We avoid this
difficulty by
modifying the contour in the following way: we delete from the
area within the contour the portion which also lies within
a
small circle \z c\
e and then integrate round the boundary
of the remaining region. This process is called indenting
the

contour.

The integral round the indented contour is calculated by the


theorem of residues and then the radius of each indentation is
made to tend to zero. This process gives the Cauchy principal
value of the improper integral, and the question of the
existence
of the improper integral in the ordinary sense requires
further
investigation.

The' details of this method will become obvious


from a consideration of the following examples.
Example 1 Prove that, if a > 0,
.

Trsma

We consider
r
where the closed contour T consists of the real axis from
-R to
a semicircle in the upper half of the z-plane on this segment

R and

as diameter.
Smce the integrand has poles at z
a, which lie on this contour, we
modify T by making an indentation of radius e at a and
another of
radius r) at -a. The integrand is now regular within
and on T, and so

is

zero.

Evaluating the various parts of the integral

/"

in9
ecosfl-B
B
-r,

iM*de+j1+ jt+
a .- geW

J,

-O-Tj
/

+
J

we

obtain

o-e

B
/

+
J
a+n

_JL_2(fe)

a+e

where Jt and Js denote the integrals round the indentations


at a and
a respectively.

THE CALCULUS OF RESIDUES


The modulus of the

135

term on the right-hand side of this equation


is less than ttR/(R 2 a 2 ) and so this term tends to zero as R -> oo. To
evaluate Jlt we observe that, on the indentation at a, z = a + ee m where
decreases from it to 0. Hence
r

first

,-,

exp(ta+iee w

eiePdO

2a+eeei

o
Trie*"

2a
as e tends to zero.f

R - oo, -> 0,

Making

J2 tends

Similarly,
ri

-- 0,

we

to

fynitr^ja as

t]

tends to zero.

find that

TO

J
where, so
limits

far,

and

the integral

2a

is

a principal value with respect to the

also with respect to the singularities at

a. But

infinite

since the

integrand behaves like 1/a; 2 for large real values of x, the integral is an
ordinary infinite integral as regards the limits. If, however, we write
x
a+, we find that the integrand behaves like a constant multiple
of l/ when is small. Thus the integral exists only as a Cauchy principal
value with respect to the pole a; a similar remark applies to
a.
Finally, equating real and imaginary parts, we obtain

QO

cosa?
f -g
(fe
2

J a?x
00

00

it

= -sma,
a
.

f -jsina: -dx=0.
2
a; 2
J a

00

Example 2. The function /(z) has a simple pole of residue r at a point


on a simple closed contour T. If T be indented at c, show that the
integral of f(z) round the indentation tends to rod as the radius of
the indentation tends to zero, a being the internal angle between the
two parts of T meeting at c.
c

at

Example 3. By integrating eiz/z round the rectangle


R, R + Ri indented at the origin, prove that
sin a;

/
Example

4.

ax

with vertices

jfit.
,

Obtain the result of Ex. 3 by using an indented semi-

circular contour.

f Evidently J1 has, as limit,


2, below.

See Ex.

ni times the

residue of the integrand at a.

THE CALCULUS OF RESIDUES

136

6.52. Jordan's inequality


Since cos 6 decreases steadily as 6 increases from
to \n,
the mean ordinate of the graph of y
cos a; over the range
x
d also decreases steadily. But this mean ordinate is

< <

cosxdx

e
CJ

sine
-

We

< 6 < \w,

have therefore proved that, when

no

that

is,

s^ sin# <

that

0.

tr

is known as Jordan's inequality.^


The importance of Jordan's inequality

This

lies in the fact that


enables us to evaluate integrals of the type exemplified
by
6.5, Ex. 3, without having to use a rectangular contour or
integration by parts.
it

00

Example

1. Evaluate

<

f *~^4x,

where a

>

0.

~~\~Ctr

00

We have

already proved that

7T

-R

Now the absolute value of the second integral on the left-hand side does

not exceed

>P2

/*

'--

<i^

e-2R6/v

d#

by Jordan's

inequality,

t Comj-8 a? Analyse, 2 (1894), 286.

THE CALCULUS OF RESIDUES


which tends to zero as

R ->

137

We have thus proved that

oo.

lim

-y-

from which the value of the given

dx

= w-",

integral follows at once.

CO

Example

2. Evaluate

f sin a;
I

ax,
dx,

by the

aid of Jordan's inequality.

CO

Example
\z

= B in

Prove Jordan's lemma, that, if a denotes the semicircle


the upper half-plane and if m > 0, then
3.

lim f
J

emizf(z)

dz

0,

B--co

provided that, as

<

<

B -> oo, f(Re

Bi
)

->

uniformly with respect to 6

when

ir.

00

Evaluation of integrals of the form


where a is a real constant

6.6.

xp-^x)

dx,

Let f(x) be a rational function which has either no poles or


else

only simple poles on the positive part of the real axis. Then
00

if

xaf(x) ->

as x

and

->-

x ->

also as

oo, J

a;

a_1/(a;)

dx converges

at the upper and lower limits of integration and possesses a


Cauchy principal value with respect to each singularity on the

range of integration.

There are three main methods of evaluating such integrals.


First method. If we

make the substitution t

CO

ex ,

we find that

QO

tP-yit) dt

= j eaxf(ex

dx.

-co

The

latter integral

may

be evaluated by considering
f e

az

f(e

z
)

dz,

V consists of the segment of the real axisf from


to
R, and the semicircle in the upper half-plane on this segment as
diameter.
But if a is a pole of f(z), the function /(e2) has a pole at each
where

f Indented

if

necessary.

THE CALCULUS OF RESIDUES


loga+2w7rt, where n is any integer. The contour
thus expressed as a sum of an infinite series of resi-

138

point z

integral

is

dues, which often diverges.

This difficulty is overcome by observing that, since f(ez )


a rational function of ez

is

the values of the integrand on the real axis differ from


2w only in the factor e2c"".
those on the line Im z
that

is,

We

consider, therefore,

| **/(*)

dz,

where F is the rectangle with vertices R, R-\-2vi, S-\-2iri, S,


and then make the positive numbers R, S tend to infinity.
Second method. The function za_1/( z) has, in general, a
~
branch-point at the origin. If za x is taken to be real and
positive on the positive part of the real axis, the branch of
2 u_1/( z) so defined is regular, save for poles, in the whole
z-plane cut along the negative part of the real axis.
R to R and the
If, then, r consists of the real axis from
semicircle in the upper half-plane on this segment as diameter,
indented at the origin (and elsewhere, if necessary), we can

evaluate

j zP-yiz) dz
r

by Cauchy's theorem of

residues.

But the contribution

of the

real axis to this contour integral is

j x^fix) dx

+ j {re

a -1

7,i

f(re" i )&' i dr

xa ~ 1f(x) dx

since argz has increased to

tt

ea

r"- 1/^) dr

when z reaches the negative part


The value of the given integral

of the real axis in going round I\


is

then obtained by equating imaginary parts.

f It may, however, be summable by one


summation of divergent series.

of the conventional

methods

of

THE CALCULUS OF &ESIDUES

139

Third method. If V denotes the contourf formed by the circles


\z\
R, \z\
e joined together along the upper and
lower

sides of the cut along the negative part of the real


axis,

we may

evaluate
J

2 a -V(-z) dz,

where z"- 1 has

its principal value, by Cauchy's theorem of


residues, since the integrand is regular, save for
poles, within

and on T. The contributions of the two

e 'r-i/(r) dr
J

since arg z

-^r a -^(r) dr =

J*

sides of the cut are

-2isina7r j ra ~ lf(r)

dr,

equal to iri on the cut. The value of the required


follows at once. It should be observed that the
contour in the third method may be derived from that of the
integral

first

is

now

by the

ez

substitution

f.

These three methods are illustrated in the following examples.


CO

Example

1.

Evaluate

where

jj~dt,

<

<

1.

First method. If we

put

ex,

we have

CO

CO

J e*+l

J 1+*

co

We

consider therefore

<*"

'

..

dz,

r
where
where

the rectangle with vertices at R, R+2m, S+Zrn,


S
R and 8 are positive. The integrand has but one singularity
within T, a simple pole at z
m; the residue is
is

lixn

^=pl =

and so I = 2irie'ri
But we may write

eai

lhn ^2li

= _**

-B

2ot

2jt

so
t Cf. Fig. 4

on

p. 141.

THE CALCULUS OF BESIDUES

140

q<iR

Now

+ iy

2ire aR

< e*-l'

idy

-=

/
o

R > oo,

which tends to zero as

<

since a

2ff

moreover

e-oS +

27re_aS

ridy\<
/ rw, + r"'pil?j'
o

S >

which tends to zero as

oo,

>

since a

Hence we have

0.
CO

2Trieam =(l e 2a"

r
)

00

J e^+l
CO

'

so that

ax

dec,

00

at

J 1+*

dx

J *"+!

sin ffl7T

be observed that, by the principle of analytical continuawhen < Rl a < 1, since the expressions on
each side of the equation are analytic functions of a, regular in this strip.
It should

tion, this equation also holds

Second method.

We

consider the value of the contour integral

IS-

where

2 a_1

has

its

and T is the contour of Fig. 3, the


and 1 being e and r\ respectively. Since
within and on T, I is zero.

principal value

radii of the indentations at

the integrand is regular


This contour integral can be written in the form

~ x dx +J+

~ x dx +

ij

-^ do

Ref*

J
where

l+r

^JR

dr

+
T
Jl-ee
I

denotes the integral round the indentation at z

iRa eaBi

l-Rem
1

which tends to zero as

if?

;p

do,

1.

Now

Ra
<
Hi?-1'

do\

oo,

since a

<

1.

Again, since a

>

0,

we have

THE CALCULUS OF RESIDUES


as e -*

Also,

0.

limit of

Hence,

as
if

tj

by applying the
->

result of 6.51,

Ex.

141
2,

we

find that the

is ni.

we make

i?

s-

oo, e

0,

1-*

77

>

J 1+a:

we

0,

dx

obtain

-\-iri

0,

Fig. 4

the

first integral

singularity

we

1.

being a Cauchy principal value with respect to the


Equating real and imaginary parts in this formula,

find that

sinCMT

IE

0-1

!+*

dx

7T

as before, and also that


00

CO

!-*

dX

C0SO7T

1+*

dx

7TCOta77.

THE CALCULUS OF RESIDUES

142

method

It should be observed that the second


two definite integrals at once.

Third method. If

is

the contour of Fig.

z0-1

when

has

its

we have

2iri

1 dz =

4,

gives us the values of

principal value. Evaluating this contour integral,

we

obtain

Be

1+r

Bi

8*

1+r

and fourth terms arise from integrating along the


upper and lower sides of the cut. If we now make B -> go and e -> 0,

in which the second

we

evidently arrive at the same result as before.

Complex transformations

6.7.

When a given

of definite integrals

definite integral has

been evaluated,

it is

often

possible to deduce formally the values of certain related integrals

by means of a complex change of variable. Cauchy's theorem


of residues provides a simple means of proving the validity of
this formal process. We shall illustrate this by considering two
transformations of the well-known integral

!
In the

first place, let

e~ x' dx

Vtt.

us consider

= Je-*'dz,

being the rectangle whose vertices are R, .R+i&, where


Since the integrand is regular within and on T, / is

b is real.

Hence we have

zero.

e~x* dx

e-^+^'i dy

e-^+w dx

e^-^'i dy.

-B

But

since
b

f e -<RiV?i

dy <-

161

J
o

-R

'+y'dy<

|&|e-*

+*'

THE CALCULUS OF RESIDUES

which tends to zero as

->-

oo, this

143

equation gives

00
2
e -x +b'( cos

2bx isia.2bx) dx

f e-*'

dx

Vtt.

00

we

Thus, equating real and imaginary parts,

obtain

00
("

dx

Jn e

xS
e- sin 26a; dx

0.

e _a;, cos 2bx

-6 ',

oo
oo
I"

OO
Secondly, let us take the same integrand in the case when
R, bounded by radii argz
T is the sector of the circle \z\
As before, / vanishes,
a
Jtt.
a, where
and argz
Jw

and

=
< <

so

B
f

e~ x ' dx

e-^00326 -^'^ 20 iRe0i dd

B
g ! cos2a ir ! sin2agoa ^y_

("

Now
f

e -ficos20-iiJ

sin2e^e9t

rf#

-l?coB2B
.R f e

gQ

<
Hence, making

by Jordan's

R -> oo, we

and

deduce that

oo

inequality.

!
!
sln2a
c -r oos2a-ir

^r

_e

-cci

e -x'

gx

_ e _ai

so
e

I"

-'',oos2o:

cos(r2 sin2a) dr

jVn-cosa,

fVirsina,

00

provided that

e-r

'

oos2a! sin(r 2 sin

Jir < a <

Jw.

2a)

rfr

Vn-,

THE CALCULUS OF KESIDUES

144

Cauchy's expansion of a function as a series of

6.8.

rational functions

One of Cauchy's most important

applications of his theorem


concerned with the expansion, under suitable conditions, of an analytic function as a series of partial fractions.
Let us suppose that

of residues

(i)

is

the function f(z) is regular, save for poles, in

any

finite

region of the z-plane;


(ii)

there exists

an

increasing sequence of positive numbers


n -+ oo, and such that the circle

such that Rn -> co as


whose equation is \z\
for
(iii)

the

passes through no pole of f(z),

any value of n;

upper bound of

\f(z)

on

Cn is itself bounded asn^oo;

->0 as n->co, uniformly with respect to 6 in


< < 2tt, or, more generally, in every portion of this
interval which does not include one of a finite number
of

iv )

9i
!/(-# e )

Then,

if is not

= Rn

Rn
Cn

exceptional values of

6.

a pole of f(z), we have


/()

Sn () is the
of f(z) within Gn
where

sum

= lim#J),

of the residues of /(z)/( z) at the poles

That this theorem does provide the required representation


of /() in partial fractions is easily seen. Tor if a is a pole of

with principal part

f(z)

the

sum 8n (0

is

Ji b (za.)r

r
,

the contribution of

a.

to

2 b (C-cc)-':
r

Let us now suppose that is any point of a bounded closed


domain D which lies in || < R and contains no poles of f(z).
Then since Rn ^co with n, we can choose an integer
such

RN >

R; the point lies, therefore, within all the circles


Cn for which
N. Since the singularities of /(z)/( z) are
and the poles of f(z), we have, by the theorem of residues,
that

n^

55 J

iS* = *>-*

t Exercices de Math. (Paris, 1827)


7, 324-44.

reprinted in (Euvres de Cauchy (He serie)


'

THE CALCULUS OF RESIDUES

145

>

provided that n
N. It remains to show that the integral on
the left-hand side of this equation tends
to zero as n -> oo.
For simplicity, we shall suppose that there is only
one exceptional value of 6, a say; the method
of proof is, however,
applicable in the general case. By hypothesis
(iii), there exists
a positive number M, independent of n, such
that \f(z)\
when z lies on Cn moreover, by (iv), if 8 is any positive
number,
the upper bound e of \f{R e9i
n
)\ when a+S
n
2,7+a-S
tends to zero as n -> oo. It follows that,
when

<M

< <
n^N,

dz

<

AgJ*|/(.Be*)|d0
i

< nfj

+S

/*" +
a

<
and

27r+a-8

/ ,.*
/v_lS

fi

B,

^-S-(2Jf8+27wJ,
R.-.R

HmlfMZ dz < 2MS.

so

k-ko J

But

as 8

is

quite arbitrary, this implies that

lim

/ i-z
It should be observed that

dz

0.

we have shown

incidentally that

Sn (Q converges uniformly to /(C) when lies in the domain D.


By this theorem the function /() is expressed as a series of
the form

00

si(0+

2 {sn+1 (0-sn (0},


n=l

which converges uniformly in any bounded closed domain


which
contains no poles of f(z). But S
n+1 tt)~S (Q is the sum of
"

the
n
residues of /()/( z ) at the poles of
f(z) between Cn and Cn+1
have, therefore, expressed /() as a series of partial
fractions
which converges uniformly, provided that the terms are suitably
bracketed.

We

4111

THE CALCULUS OF RESIDUES

146

6.81.

The expansion of

cosecz as a series of partial frac-

tions
application of the theorem of the previous section, we
consider the expansion of cosecz as a series of partial fractions.
This function has simple poles at the points 0, ir, 2tt,,.., so

As an

whose equation is \z\ (n+^)n, does not


pass through a pole of cosec z for any value of the integer n.
In order to apply the theorem, we must consider the behaviour
of cosec z on this circle.
Let us draw, with each pole as centre, a circle of radius e,
where e is less than \-n. We shall show that cosec z is boundedf

that the circle

in the region

In the

Cn

T exterior to

Icoseczl

all

these circles.

= x-\-iy, we have

first place, if z

2
:

<

so that the inequality [cosecz]

2
.

t-t

= cosech|2/|

< cosecha holds in the part of

<

T outside the strip |Imz| a. On the other hand, |cosecz| is


obviously bounded in the portion of T within the rectangle with
vertices zt%nai, and therefore, by periodicity, is bounded in
which |Imz| < a. Combining these two
holds everyan inequality |cosec z <
being obviously dependent on e.
where in T, the constant
As the points on the circle Cn are all at a distance not less
than \tt e from the circles defining T, this inequality also
holds everywhere on Cn independently of the value of n.
Again, when z = (n-\-%)Tre ei we have
the part of
results,

we

for

find that

< cosech Kra+iVsin^l < cosech{(w+J)7rsinS},


provided that 8 < 6 < ttS or 77+8 < 6 < 2tt 8, where 8
|cosecz|

any small

positive

number.

It follows that, as

n ->

00,

is

cosec z|

tends to zero uniformly with respect to 8 in the two given


angles.

All the conditions of the theorem of 6.8 are thus satisfied.


t The method of proof is taken from the footnote on p. 32 of Lindelof's
Borel tract, Le Calcul des Residus (Paris, 1905).
A similar result can be proved in the same manner for each of the functions
sec z, tan z, cot z, e^'jifi1
a
1)
(0
1 ), in the regions obtained by excluding
their poles by small circles.

< <

THE CALCULUS OF RESIDUES


Hence,

if is

147

not a pole of cosec z,

= lim j (residue of cosec /() at z = mn)

cosec

~n

By Weierstrass's if -test, this series converges uniformly and


absolutely when lies in any bounded part of T,
and so may
be integrated term by term. This gives

= log^+logi+ f (_ 1) m log /
m=l

logtanK

tan

K = MC f[ I-

Since tan A/(A) ->

Hence,

unity.

if

as ->

we

write

fM
m

WV(W )_)

7r

0,

the constant of integration

2z,

is

we have

tan z
1

Hi mvj/ni

(2^1)^)'

which represents tan 2 as a quotient of two


each of which is an integral function.f
6.82.

An

Cauchy

infinite products,

extension of the theorem of 6.8


also showed how the theorem of 6.8

when

the function f(z) satisfies conditions (i),


not condition (iv). In this case the function
F{z)

may

(ii),

and

be used
(iii) but

= fQ
z

is

on

considered;

Cn

it

Bn e

hence, as
to 9 in

obviously

ei

-> ao,

<6<

satisfies

conditions

(i)

and

(ii).

But

and so

F(Rn e ei )
2tt.

tends to zero uniformly with respect

The function F(z)

t Cf. the results of Exx. 33, 36 at the

satisfies, therefore, all

end of

this chapter.

THE CALCULUS OF RESIDUES


6.8, and we conclude that

148

the conditions of

= lim (sum

m-oo

of residues of
t{Z

at poles of^-^ within

t)

particularly important case of this arises

and has only simple

regular at the origin

poles.

Cn

\.
)

when f(z) is
we suppose

If

that the poles a v a 2 a 3 ,..., of residues 6 X b 2 b 3 ,... respectively,


are arranged in order of increasing distance from the origin, so
,

<

that

|z

<

\a 2

at ar oif(t)/(ztP)

<

\a 3

<

...,

= bJ

whilst the residue at the origin

1\

\z ar

aj'

The

is /(0)/z.

partial fraction

now becomes
/(s)

= /(Q)

K
^ a (za

Z-,

and

easily find that the residue

is

ar (zar )

formula for f(z)/z

we

)'

so, finally,!

/w-zro+^kL.+i).
More generally, if the condition (iii) of 6.8 is replaced by the
boundedness, as n -> oo, of the upper bound on Gn of \z~pf(z)\,
where p is a positive integer, the function f(z)/zp+1 satisfies all
the conditions of 6.8, and the representation of f(z) as an
infinite series of partial fractions follows immediately.
6.83.

The representation

infinite

of

an integral function as an

product

Let F(z) be an integral function which does not vanish at


the origin but has simple zeros z lt z 2 z3 ,..., arranged in order of
,

As these
oo as n ->

increasing modulus.
of finite affix,
If

we

\z n

->
\

write F(z)

zeros can have

(z zr )0(z), then

zero in a certain neighbourhood of zr

J\z7

no limiting point

oo.

regular and nonHence we have

<J>(z) is

~ z-z + (zj'
r

t This series converges uniformly in any bounded closed domain which contains none of the poles of /(z), provided that the terms are arranged in groups
corresponding to poles of equal modulus.

THE CALCULUS OP RESIDUES

149

and

so the only singularities of F'(z)JF(z) are simple poles of


residue 1 at the points zr
.

Let us now suppose that F'(z)/F(z)


6.8.

satisfies

the conditions of

It follows that

= 2

CO

-*"()

F(z)

^-i z

n=l
the series being uniformly convergent in any bounded closed
domain which contains none of the zeros of F(z), provided that
the terms are suitably bracketed. If we now integrate term by
term, we find that
log.F( Z )

Hence

= log *-(<>)+
=

F(z)

F(0)

J~J

^V

log^l

(1-1

the infinite product being uniformly convergent in any bounded


closed domain which contains none of the zeros of F(z).
If, however, F'(z)/F(z) satisfies only conditions (i), (ii), (iii) of
6.8, it

follows

by

6.82,

that

r(z)_F'(0)
F(z)

and

F(0)

^/
+ Z\z-z
+ zJ'
n
1\

so

where A is a constant of integration, whose value is easily seen


to be logi^(O). Finally, taking the exponentials of the expressions on each side of this equation, we deduce that
F(z)

F(0)e**"l F

FT

[(l

-]e*4.
nl

n = 1 \\
CO

Example. Prove

that sin7rz

T (l

ttz 1

^J.

Let us consider the integral function


CO

V
'

.Z/

'

(2w+l)!

THE CALCULUS OF RESIDUES

150

which

is equal to
simple zeros at z

(s,imrz)j(TTz)

when

and

to

Moreover,

its

z =

= 1, 2, 3
F'(z)

the conditions of

when

0.

It

has

F(Z)
satisfies all

logarithmic derivative

Hence

6.8.

F'(z)

F(z)

BM-oo

-^

U n

z+nj

Integrating,

we

Z-, z 2 n*
4.

obtain immediately

w-n(.-5>.
i

We cannot,

however, writef

=n'

('-.-)

CO

since this infinite product diverges.

Again, by

6.82, "we

havej

00

00

V/
+ iWV(J
\zn
n)

Zii

IV
n)'

/_, \z-\-n
l

each of these series being uniformly and absolutely convergent in any

bounded closed domain which contains none of the

zeros of F(z).

From

this it follows at once that

i
00

and

sm ttz

so

= ttzTT' N1 -WM,
00

the latter product being absolutely convergent.

REFERENCES
E. Lindelof, Le Calcul des Risidus (Borel tract Paris, 1905).
G. N. Watson, Complex Integration and Cauchy's Theorem (Cambridge,
;

1914).

The accent

% F'(0)/F(0)

indicates that the

lim jwcotwz

term corresponding to n
1

0.

is

omitted.

THE CALCULUS OF EESIDUES

101

MISCELLANEOUS EXAMPLES
If the function /(z)

1.

|/(z)|
it is

= M,

show that

regular within and on the level curve


a double point of this curve if and only if

is
is

a zero of/'(z).

which is
2. If C is a simple closed level curve of the function /(z)
regular within and on C, show that/(z) has at least one zero within C.
1. (Maczeros within C, f'(z) has
Prove also that, if /(a) has

DONAlD.t)
3.

Show

that, if k

>

1,

the equation

2 n 6fc-

has n roots in
4.

\z\

<

The function /(z)

whose modulus

1.

..

is

regular in

than

is less

1 az

I",

<

|z|

Show

1.

where

that, if

,,

has only one zero


the function

has exactly one zero in \z\ < 1, provided that the constant be
ciently small. (Cambridge, 1932.)
5.

Express as a

2 ( g -/x >

Z2
y.

when w

D,
-PW

6.

Show

Hence prove that

0.

(1-2/xw+w2 )-1

where

suffi-

powers of w the solution of

series of

which reduces to

a,

1,

,.,}

it

<

|6|

'2

= ^7

= 2wMP(ja),

dy-1)"
^S

that the solution of the equation


z

which reduces to a when

w=

= a+we

is
2

'w;

the expansion being valid

when
\w\

f Proc. London
(1916), 227-42.

Mi
= a+weP+^2e+...+ ^ n^ea +...,

Math. Soc.

(1),

<

29

e- 1

- 1110

(1898), 576. See also

Watson,

ibid. (2),

15

THE CALCULUS OF RESIDUES

152

Show

7.

that, if

\a\

<

1,

dO

2tt

2
J l+z 2acos0

'

and deduce the value of the

Show

8.

>

when Rl a

that,

integral

when

|o|

m.

>

1.

0,

taxi(d+ai) d0

Show

9.

that, if

m be real and <


1

<

1,

2w

J^m
+

cob 9

2o sin

cos(m sin #) sin(m sin

+ 0)} d6 =

2ttcosio,

o
2jt

+ a - 2a sin
2

^W

sm g

g cos ( m sin # + ^)>

dfl

27rsinma.

(Edinburgh, 1931.)
10.

By

using Cauchy's formula for the derivative of sin a or other-

wise, prove that


2jt

{cos

sin(a

+ cos 0)cosh(sin 0) + sin

cos(<x

+ cos 0)sinh(sin 0)} d0

=
11.

277 cos a.

(Edinburgh, 1930.)

Prove that
CO

/cos x dx
(x 2

+a

)(

+6

-6

e~ a \

77
/e
= ^Zfei\"6

^1

CO

when the

real parts of

a and 6 are positive and a

also that

f cosjcefo

( 2

CO

when the
12.

is

not equal to

b.

Show

OT

real part of

Show

is

+a2

= 7r(l+o)
a

2a 3 e

positive.

that the Cauchy principal value of

dx

00
where

f,

v and
,

are real,

is

Xm^i-e^Hiq),

the value of A being

THE CALCULUS OF RESIDUES


1,

or-1

0,

according as

is

153
greater than, equal to, or less than

unity.

Deduce that
OC

sin m(x-)

smn(x-r))

(x-i)

(x-r,)

/
when n >

m>

ax

_
-T-sinm^-i;)
{

_ v)

'

0.
CO

13.

Prove that

14.

Showthat

J^L dx =

cos^-cos 9.

(Math. Trip., 1919.)

n.

= 7T(p _ g)

<fe

00

when p and

q are positive.

Prove that the residue of the function e^/fa* 2zcosa+l) 2


,
< a. < w ), at the pole which lies in the upper half-plane is
Me"os, where
e- s in(nsina+l)
15.

(n.

>

0,

4sin 3 a

Hence show that


a

x(x*

/
d
16.

+ l)sinnxdz

__ 7rAsin(n cos <x)

(x* -2x* cos 2a -f-l)

4 cos a

(Edinburgh, 1930.)

Jdx

Prove that

"

00

evrinlmnx)

= frr(e-

(Catchy.)

1).

17.

where

By integrating round
c,

R, and

<S

the rectangle with vertices ciB,


are positive, prove that, when a > 0,
c

SiR,

+ coi
f

dz

c coi
the path of integration being a straight line parallel
to the imaginary

Show
18.

also that the integral

Show

that, if

Bla

>

0,

is

zero

\1ma\

when a <

<

0.

n,

c+ooi

*"-&=

SU177Z

2*
l

e-<"

c ooi
the path of integration being the straight line Rlz

c,

where

<

<

I.

THE CALCULUS OF RESIDUES


By
integrating
19.
e'^'cosechirz round the rectangle
R%i, show that, if < a < it,
154

cos(aa: a )

dx

& cos iac,

a#

ismja.

with vertices

00

coshoo:

sinlaa^)

COSh7TX

(Math. Trip., 1932.)


20.

Show

1 <

that, if

<

<

and

x"dx
f-l
I

21.

Prove that,

if

{
J

<

3, b

>

xa- l ain(inabx)

f
J

it,

7rsinoa

+ 2a; cos oc+x"

<

<

xP- 1 sin(iTTabx)-i
C

sina7rsina;

>

0, c

0,

dx
~^-^ =

in<p-*e-<>,

-\-c

- =
Xx=

n-c6c).
oc).
fncr-*2 cos(|iTa
cos(-fcna

(Cattchy.)
22.

Prove that,

<

if

<

u a/

j8

<

and

<

sino7r|

2,

la/

ex

23.

By making

the transformation

eM ,

show that

24. /(z) is a rational function with no poles on the real axis and is
such that zf(z) tends to zero as z > 00 and also as z >- 0. By integrating
logz/( z) round an appropriate contour, prove that
00

QO

flog x {f{x) +f(x)}dx +iri] f(x)dx


is

equal to

2m times the sum of the


How

residues of log z/(

z) at

is this result to be modified


in the upper half-plane.f
are simple poles on the real axis?

t This result is of particular

value when/(z)

is

j"/(a;)

dx, whereas the

method of

poles

an odd function, as it enables

00

us to evaluate

its

when there

8.5 fails.

THE CALCULUS OF RESIDUES


Show

25.

155

that

dx
J

(1

+*)*"

2 2
J (1+* )

*"'

Find the theorem corresponding to that of Ex.


z) round the same contour.
Hence evaluate the integral of Ex. 23.
26.

24, obtained

by

integrating (logz) 2/(

27.

7r,

By integrating z/(l ae_fa round


7T+m, prove that, if a > 1,
)

the rectangle with vertices at

IT

aXSmX

2acosa:+a ,dx
2

= TTlog(l+a-

).

What

is

the value of this integral

<

when

<

(Catjchy.)

1 ?

28. Integrate logsinz round the rectangle with vertices at


and it, and so prove that
iR, indented at

0,

it,

tt+iR,

if

logsinaicfa;

29.

By

integrating eoz /(e-2te

H:

30.

Prove that,

if

n <

= 7rlog2.

1) round a suitable contour, prove that

dy

<

v,

j77Coth Jira Jo.

sinhax

/!sinh7W
00
31.

The

Gamma function is

da;

tan \

denned by
00

T(a)

when a
and

>

0.

Integrate z0-

so prove that, if

^-

>

("

afl- 1 er

round a

and

\tt <

dx

sector, indented at the origin,


ex

<

\tt,

00

xar 1 e- xmBa

(xsina) dx

f
J

Show

that this formula

still

holds

T(a)

(aa).

sin

sin

when

a=47rif0<a<l.

THE CALCULUS OF KESIDUES

156

32. Integrate the principal value of za

and

<

so prove that, if

<

and

>

~V round the contour of Fig. 5,


0,

CO

f eiv (c

+ iyf-

dy

2e~ c sin <nrT(a).

00
-RtiR

ic+FR

c-iR

-R-iR.
Fig. 5

Deduce that
iff
("

co8{te,ne-(l-a)6}seo 1 +a $de

e^sincwrlXa).
(Catjchy.)

33.

<

By

integrating the branch of z~ v (c+az)- q which is real when


0, a
contour of Fig. 5, prove that, if c

>

< c+a, round the


and p < 1 < p+q, then
s

00

00

dy
(c+iyY(a-iy)

34.

Prove that secz

4tt-2

(-l) n (2w+l)/{(2n+l)V 2 -42 2}.

35.

Show

that, if

is

not an integer or zero,


00

sin7r( Z +)

= sin^ Yl {{ 1 + ^h) e~*Hn+h) }-

>

THE CALCULUS OF RESIDUES


36.

Prove that tanz

2o l/{(2n+l)

8z

7r

157

-4z 2 }.

00

Deduce that

cosz

(1

...

37.

Show

<

that, if

e"

l1
f-

~~

ez
38.

<

1,

00

-O 2zoos2mra 4w7r sin 2mra

+ Z,
2*

+ 4nV

aa

Prove that
00

cosh z cosz
39.

Prove that,

if

x"*
^
+ OTZ 2'

l) nncosechn7T
(W7r) + |z
4

^_<

n=l

<

.
2

<

tt

tt,

CO

sin az

sin77Z

-sr*

^_/

it

sin Ma

rs=l
00

cosaz_
sin7rz

40.

Prove

~ ttz

that, if b is not

an

2j

v"

cosna

it

Z-i

zz

'

integer,
CO

sin7r6sin7r(z 6)

suittz

=
41.

lzbn)(b+n)'
oo

A function /(z) is regular for all finite values of z and satisfies the
= n + , where n is an
M

inequality |/(z)cosec7rz| <


on the circles \z\
integer and
is independent of n. Show that

m _ -s [m+

ffi>

-,-(/

.,(i + i) +/<->( -f-i))]s

i^(z) is

an

series of the

integral function such that {F(z)} k can

above form

that, if \F(n)\

then F(z)

is

<

be expanded in a

for every positive integral value of k.

K when n = 0,

1,

2,...,

K being

a constant.

Show

independent of n,
(Cambridge, 1933.)

a periodic analytic function of period tt whose only singu< Rlz < tt are simple poles %, a 2 ,..., an of residue
c 1; c 2 ,..., c n respectively. Show that, if f(x+iy) > as 2/ -> +oo and > i'
as y -> oo, uniformly with respect to x, then
42. /(z) is

larities in

the strip

l'l= 2i%cr

By applying this result to the function /(z)cot(z z), where


a pole of /(z), prove that
/(z)

h(l

+ l')+ 2cr cot(z-or


1

).

is

not

CHAPTER

VII

INTEGRAL FUNCTIONS
The factorization of integral functions
The most important property of a polynomial is
7.1.

that it can be
expressed uniquely as a product of linear factors of the form

-HJHJ-K)

"where

A is a constant, p a positive integer or zero,

z n the points, other

than the

origin, at

and z v z 2 ,...,
which the polynomial

vanishes, multiple zeros being repeated in the set according to


their order.

Conversely, if the zeros are given, the polynomial


determined apart from an arbitrary constant multiplier.
Now a polynomial is an integral function of a very simple
type, its singularity at infinity being a pole. We naturally ask
whether it is possible to exhibit in a similar manner the way in
which any integral function depends on its zeros. It has been
shown in 6.83 that such a factorization is possible under
is

certain circumstances.

There
sidered.

are,

however, two rather serious

The

first is

number of zeros

that

if

difficulties to

be con-

an integral function has an

infinite

z 1 z 2 ,..., z n ,..., say, it is


,

that the infinite product IJ

not necessarily the case

(1 z/zm converges, much


)

less

that

the value of the product


factors.

2,

independent of the order of its


(sin ttz)j(ttz) has zeros 1,
product!

is

For example, the function

3,...;

but the

infinite

GO
defined as the limit of

-M
-M

'

'

and JV tend to infinity independently, diverges. This product can, however, be made to converge by grouping its factors
as

t Here and in the sequel, the accent indicates that the infinite term, given
0, is to be omitted.

by n


INTEGRAL FUNCTIONS
suitably, but,

even

so,

159

the value of the product then depends

on the mode of grouping.!


It was shown by Weierstrass that the appropriate method of
expressing (sin ttz)/(ttz) as a convergent infinite product whose
value does not depend on the order of its factors is
sinn-z
TTZ

frK'-SH-

Evidently it will not suffice, in the solution of our problem, to


consider only simple factors of the form (1 z/z n ).
A second difficulty is due to the fact that there exist integral
functions which never vanish, ez being a simple instance of this.
It follows that a knowledge of the zeros of an integral function
cannot determine the function save for an arbitrary constant
multiplier; the multiplier is now an integral function with no
zeros.

The most general

integral function with

no zeros

is

of the

where g{z) is itself an integral function. For if


integral function which never vanishes, the function
eff(2) ,

form
an

f(z) is

m=miM

is

an integral function. Integrating the expressions on each


of this equation along any path from z to z, we find that

also

side

log f(z)

= log/(z

F(z) dz.

)+
J*
*.

But
is

as the expression on the right-hand side of this equation


an integral function, the result stated is now established.
Example 1 Prove that
.

" *

-('+9(>+JK'-9(>+i)('+3(>-5)
We have

K'+3("+3('-i)-v+En)('+s)M
2n

i
-

See the examples at the end of this section.

INTEGRAL FUNCTIONS

160

n
oo

-co

as

n >

oo.

The required

now

result

follows

by

Weierstrass's formula

for sin 772.

Example

2. Prove that

-('+f)('+l)('+l)('-f)('+I)('+I)('+l)('-D
7.2.

The construction

of

an integral function with given

zeros
If f(z)

is

an integral function with only a

finite

number of

zeros, z x , z 2 ,..., z n , say, the function

f{z)l{zz 1 ){z-z i )...{zz n )


is

an integral function with no


f(z)

where

^(z) is

an

zeros; hence

= (z-z )(- )...(z-B )eP


1

integral function.

however, an integral function has an infinite number of


cannot have a limiting point in any finite
region of the plane, since such a limiting point would be a
singularity of the function. The only limiting point of the set
is, therefore, the point at infinity.
We now prove Weierstrass's theorem, that if z, z 2 ,..., zn ,...
be any sequence of numbers whose only limiting point is the point
at infinity, it is possible to construct an integral function which
vanishes at each of the points z n and nowhere else. The construction involves the use of Weierstrass's primary factors
E(z,0)
(1 z )ez +zt 2 +-+zPIP (p>0).
lz; E(z,p)
If,

zeros, the set of zeros

Each primary

factor

is

zero, a simple zero at z

an integral function which has but one

1.

< we have
E{z,p) = exp{log(l- Z )+2 + +
|

Now when

\z\

1,

...

+ Jj

f Abh. der Preuss. Akad. Wiss. zu Berlin (Math. Klasse) (1876), 11-60;
reprinted in Weierstrass's Werke, 2, 77124.

INTEGBAL FUNCTIONS
and so

log E{z,p)

_^H_^1 _
p+2

p+1

From

this it follows that,


[log E(z,p)\

161

when

\z\

"'

<i

< ^(1 + + 1+...) ^


1

2]e]p+K

We may suppose that the origin is not a zero of the integral


function to be constructed; for if we
require a function which
vanishes at the origin, we need only multiply
the function 0(z)
determined below by an appropriate power
of z. Let the given
zeros z 1 z 2 ,... be arranged in order of
non-decreasing modulus,
multiple zeros being supposed to be repeated
in the set according
to their order. Then, since r
\zn increases indefinitely with
n
n, we can always find a sequence
of positive integers
pv p 2 ,...,
pn ,... such that the series
,

w~
converges for

pn

= n,

all positive

since for

holds for

values of r. In fact, it suffices to take


any given value of r the inequality r/r
n < \

all sufficiently large values


of n.
next assign arbitrarily a positive number
R and then
choose the integer
such that rN
2R
rN+1 Hence, when
n
and \z\
R, we have

We

>N

and so
It follows,

<

<

logWl,^_ij
by

2/^".

Weierstrass's if -test, that the series

os

ii-

converges absolutely and uniformly


that the infinite product

when

Mi-"-ttU

<

lu-

ll

\z\

<

R. This implies

INTEGRAL FUNCTIONS
B, no
converges uniformly and absolutely in the circle \z\ <
integral
an
represents
so
and
matter how large B may be,
162

function G{z).
With the same value of B,

< B < rM+1

that rM

Then

we choose another
all

y\ E (-, Vn -\\
L\

integer

M such

the functions of the sequence

(m

= M+l,M+2,...),

\*

<

B.
\z\
vanish at the points z 1; z 2 ,..., zM and nowhere else in
zeros of
Hence, by Hurwitz's theorem ( 6.21, Ex. 2), the only
0(z) in

means

But since .ft is arbitrary, this


a,,..., z
l5
are the points of the sequence
of
G(z)
zeros
that the only
\z\

< B are z

This completes the proof of the theorem.


function G{z)
Since there are many possible sequences p n the
has
sequence
a
such
when
But
not uniquely determined.
given
the
with
function
integral
the most general

z v z a> ...

is

been

zeros
7.3.

fixed,
is

G(z) e H(z\
-

The

where

(z) is

principle of the

an

integral function.

maximum modulus

and on a
Let f(z) be an analytic function, continuous within
bound
upper
the
be
Let
G.
within
regular
and
closed contour
everywhere
holds
|/(z)|
inequality
of |/(z)| on C. Then the
at a point within C if and only
within C. Moreover, |/(z)|
proof of this result, which
following
iff{z) is a constant. The
modulus, is due to
maximum
is known as the principle of the

<M

=M

Landau.
If n

is

any

positive integer

and a any point within G, we have

by

Cauchy's integral formula. If a

this equation gives

where

is

is

at a distance 8 from C,

it/

the length of C. Hence


/

The expression on the left-hand

Vln
side of this inequality is

independent of

INTEGRAL FUNCTIONS
making tend

163

so that,

n,

to infinity,
J

obtain

the

first

result of the theorem.

Again

=~f

nf'(a){f(a)}n-i

dz
{/(*)}*

and

so

Now

we

n \f( a )

if there exists

\f(a) |*-i

<

a point a within

{z-af

2jtS 2

such that

\f(a)\ ==

this inequality gives

making ->cowe obtain/'(a)

'

0. Similarly we can show that


if
the derivatives of f(z) vanish at z
a By
Taylor's theorem it follows that/(z) is
constant in a neighbourhood of a and hence, by analytical continuation,
is constant
everywhere within and on C. This completes
the proof of the
theorem.

|/(o)|

= M,

all

It should be observed that the principle


of the

maximum

modulus holds also when/(z) is continuous in the


closed annulus
and regular in the open annulus bounded by
two non-intersecting closed contours C and C
For if a is any point of the
x
2
open annulus, we have
.

and the proof follows the same lines as before,


the upper bound of \f(z)\ on G and C
1

7.31.

The maximum modulus

If f(z)

is

the circle

on that
For,

an

circle.

(r) is

<r

an integral function
modulus

steadily increasing

principle of the

is

continuous on

unbounded function.
modulus, we have

maximum

Ifir^Ji

where rx

of

integral function, its

= r and so actually attains its upper bound M(r)

\z\

by the

M being now

2.

< M(r

2)

and hence

M(rx )

< M(r

2 ),

INTEGRAL FUNCTIONS

164

save in the trivial case when/(z) is a constant. Moreover, M{r)


cannot be bounded; for if it were, /(z) would be a constant, by

Thus M(r)

Liouville's theorem.

increases indefinitely with

r.

We can, however, prove very much more than this; in fact,


r)
log if (e^) is a continuous function of, whose graph is convex
a
downwards. To prove this, we consider the behaviour of z f(z)
constant
to
be
r 2 where a is a real
|a|
in the annulus r x

<

<

fixed later.

The function

z af(z) is not, in general,

one-valued.

But

if

we

cut the annulus along the negative part of the real axis, we
obtain a domain in which the principal branch of this function
is regular. The maximum modulus of this branch in the cut

annulus

Now

is

attained on the boundary of the domain.


a is real, all the branches of z af(z) have the same
If we consider a branch of this function which is

since

modulus.

<

<

argz
\tt,
regular in the part of the annulus for which \it
attain
its
maxivalue
cannot
principal
that
the
we see at once
modulus on the cut, and so must attain it on one of the

mum

boundary

when

rx

<

the annulus.

circles of
\z\

<r

We

have thus shown that,

2,

a
|z /(z)|

< maxfrJJfta), f?Jf(r

8 )}.

Using an obvious notation, we deduce at once that


r a M{r)

when

< r <C r

< m.ax{r%Mv rM

2.

We now choose

a.

so that

r\Mx

= r%M

2;

2}

thus

logjMJMJ

We

then have

or, finally,

The

r a M(r)

< r M

{M (x)Y*rtoi

x,

and so

< JfJwhWjiftogCrW.

result contained in this inequaUty is

known

as Hadamard's

three-circles theorem.^
t Bulletin de la Soc. math, de France, 24 (1896), 186. The result obviously
holds not only for integral functions, but also for any function -which is regular
in an annulus.

Since

<M

from which

and

INTEGRAL FUNCTIONS
we may write this inequality

follows that M(r)

it

we may show
rx

(r),

->M1

that Jf(r)->Jf2 as

as r ->

^+0.

r->r 2 0.

r 2 are arbitrary, this implies that

function of

165

in the

form

Similarly

Since, however,

M(r)

is

a continuous

r.

Finally, if

we

write

(&-ih(0

>?()

= log M(e%),

we

when

find that,

< tf.-l)l(fi) + (f-fi)^tfa),


= log M(e^) is convex downwards. We

so that the graph of

17

express this property

by saying that

log M(r)

is

a convex func-

tionf of log r.

Example.
on

\z\

(r) is

and A

r,

is

the

maximum modulus of the integral function/^)


and

a constant between

decreases steadily as r increases,

1.

Show

that M(\r)jM(r)

and that

limM(A-)/Jlf(r)

0,

r><x>

save

when /(z)

this case

is

a polynomial.

What

is

the limit of the quotient in

The order of an integral function


An integral function is said to be of finite order if there exists
real number k, independent of r, such that its maximum

7.4.

modulus M(r) on the

= r satisfies the inequality


logilf(r) < r k

circle

\z\

for all sufficiently large values of

number
If f(z)

k,

the function

is

inequality

of

finite

is

r.

If there exists no such

said to be of infinite order.

order, the constant k occurring in the

log Jf(r)< r*

must be positive. For if it were zero or negative,


be bounded and so f(z) would be a constant, by

(r)

would

Liouville's

theorem. Moreover, if the inequality holds for one value of


evidently holds for all greater values of k.

k,

it

t For an account of the properties of convex functions, see J. L. W. V.


Jensen, Acta Math. 30 (1906), 175, or Hardy, Littlewood, and Polya, Inequalities (Cambridge, 1934), 70-8, 91-6.

INTEGRAL FUNCTIONS

166

By means of a Dedekind section of the real numbers, we


can determine a number p with the property that, for a given
integral function f(z) of finite order, the inequality

log

holds for

when k

(r)

rk

large values of r

all sufficiently

< p.

M <

This number p

is

>p

when k

holds for a sequence of values of r which increase


In other words,

is

loglogM{r)

r->-oo

logr

of infinite order,

indefinitely.

we have

logr

^-*<c

Many

of the elementary functions of analysis are integral

functions of finite order. For example, a polynomial


zero; ez sinz,
,

and cosz are of order

the other hand, ef

Example,

/(z) is

maximum modulus

is

an

that/'(z)

is

cos Vz

1;

is

integral function of finite order p.


on \z\
r. Prove that, if jR

of/'(z)

< mr) <

shall

of order
.

On

>

M^r)

is

the

r,

also of order p.

7.41. Integral functions of finite order with

We

is

of order

of infinite order.

^W- l/l
Deduce

The

called the order of f(z).

< p, the inequality


log M(r) >r k

when

definition implies that,

When /(z)

but not

now show

order with no zeros if

already seen that

that

e H(0> is

an

no zeros

integral function of finite

and only if H(z) is a 'polynomial. We have


is an integral function with no zeros if
an integral function; moreover, if H(z) is-

e ff^>

and only

if H(z) is
a polynomial of degree k, e H& is obviously of finite order k.
To complete the proof of the theorem, it only remains to show
that if the real part of an integral function H(z) satisfies the

inequality

mH{z)

< rP+*

for every positive value of e and all sufficiently large values of r,


H(z) is a polynomial of degree not exceeding p. This analogue of
Liouville's

theorem

is

due to Hadamard.f

t Journal de

Math.

(4)

(1893), 186-7.

INTEGRAL FUNCTIONS

By

Taylor's theorem,

167

we have

H{z) = a +a z+a z +...


r
dz
1
H{z)~,
an =
i

where

being the

circle

\z\

r.

Now, when n

> 0,

*W >-/<-
2w

co

V
=

am rm n e-(m+n)l)ii d6

0,

the term-by-term integration being valid since the series


am z m converges uniformly. By addition, it follows that

Hence we have

K|r"<i

\mH(re6i )\d6.

f
o

On

the other hand


2ir

O
2ff

and

Rl

so

i-

Rl #(re<)

d0.

J
o

Hence we

see that
2rr

2Rla

+KK <

{|Ri#l+Rl#}d0.

But the integrand


or

<

0.

Since

is

equal to 2R1J? or according as


rP + we have, therefore,

Rl# <

B1H >

2Rla +K|r

<

4r>>+

for every positive value of e and all sufficiently large values of r.


If we now write this inequality in the form
4rP+ e - n
r- n

Kl <

2Rla

168

and then make


H(z)

INTEGRAL FUNCTIONS
we see that an = when n

r -> oo,

a polynomial of degree not exceeding


the proof of the theorem.
is

p.

> p,

and

so

This completes

Example. Prove

that, if /(z) i s an integral function of finite


order,
attains any assigned value, with at most
one exception, infinitely
often. Show also that, if the order is a fraction,
there is no exceptional
value.f
it

7.42. Jensen's inequality

Let f(z) be an integral function which does


not vanish at the
Let its zeros, arranged in order of increasing
modulus, be
2 i> z 2 s 3> multiple zeros being repeated.
Then if
origin.
,

>

|%|

RN

we have

\f(0)\

< R < N+1


< M(R)\ Zl z
\z

\,

...

\.

For the function

R*-zz
= f(z)f\ =^_

F{Z)

also an integral function, and \F(z)\ =


]/(z)| when \z\ = R.
Hence when |z|< R, we have, by the principle of the maximum

is

modulus,

< M(R)

\F(z)\

where M(R)
Putting z ==

is
0,

the

we

maximum modulus

of f(z) on

\z\

J?.

find that

as stated above. This resultj

is

known

as Jensen's inequality.

Let us now denote by n{r) the number of zeros of


f(z) in
\z\
r; evidently n(r) is a non-decreasing
function of r which
is constant in any interval which does
not contain the modulus
of a zero of /(z). Then

<

log-^L_

f f^_ f^dx
W_1

The

result contained in this

theorem, to which reference was


is

example

made

in

is

a particular case of Pieard's

4.55.

t The inequality evidently also holds if f(z) is not an integral function, but
regular when |z| < B. See also p. 90, Ex. 15, for a
more general result
Acta Math. 22 (1899), 359-64.

INTEGRAL FUNCTIONS

169

Accordingly Jensen's inequality can be written in the form

^cte< log Jlf(JJ)-log |/(0)|.

From this we deduce that if f(z) is an integral function of


finite order p, then n(r)
0(rP+ e ) for every positive value of e and
all sufficiently large values
r.

For

if

we put

B=

of

2r,

we have

2r

~Y dx < log Jf(2r)-log |/(0)| < ArP+*


o

for every positive value of e

and

all sufficiently large

being a finite constant independent of


a non-decreasing function, we have
2r

n(r)\og2

But

values of r,

since n(x) is

2r

<f

r.

dx^j^ldx< Ar^,

and

so n(r)

0(rP+*).

The exponent of convergence of the zeros of f(z)


Let/(z) be an integral function with zeros z z ,... arranged
x
2
in order of increasing modulus. We associate with this sequence
of zeros a number p x defined by the equation
7.43.

loan

^.

-*<*,

log r

where rn
\z n \.
This number p x is called the exponent of convergence of the zeros of f(z), since it has the following important
property.

and

// Pl

diverges

is finite, the series

when t

< Pl

r~ T converges when t > p

for every real value ofr.

For

if Pl is finite

and t

> Rl

log n

holds for

but if Pl is infinite, the series diverges

the inequality
,

all sufficiently large

rl

values of n, and so

> n1+P,

170

where p

On

INTEGRAL FUNCTIONS
T
(t- Pi )/(t+ Pi ) > 0. Hence 2 * converges when

the other hand,

and t has any

< n.

which r

if Pl is finite

and t

<p

1;

or if Pl

is infinite

real value, there exists a sequence of integers for

Let iV be such a value of n and let m be the


than %N. Then since rn increases with n,

least integer greater

we have

N
A^m

rN

JS

But

as there are values of JV as large as

that

2 fn

we

please, this implies

diverges.

If Pl is finite, the series ] ?n


divergent. For example, if r n

pl

may

be either convergent or

we have p x =
nQogrif, we again have

= n,

diverges; but

if r n

and

Pl =

2 TO_1
but

1,

in this case the series converges.


Finally, if f(z) is an integral function of finite order P
finite and does not exceed P For, by 7.42,

Pl is

p,

tt loen

lim=

J
n*ooAogr
n

^loa(ArP+
w < hms^
= P +e,
= hm logrc(r)
logr
logr
t-+
,

)
-'

j-^oo

for every positive value of

Example. Show

,-t

e,

and

that the series

so Pl

^ p.
00

2V

and tne

integral J w(a;)a;- 1

-T

da;

converge or diverge together.

7.5.

Canonical products

If f(z)

is

an integral function of finite order P with an

infinite

p such that
1 ~ p is convergent. If
integer,
an
is
not
p is the
series
the
2 *"n
px
either
be
may
an
integer,
is
than
if
less
greatest integer
p
Pl
Pl
number of zeros

z lt z 2 ,...,

there exists a least integer

Pl or

By

Pi 1. But,
7.2,

in

any

the infinite

case, Pl

1 <

< Pi < p-

i?

product
CO

0(z)

converges uniformly and absolutely in any finite region of the


plane, and represents an integral function which vanishes if and
only if z is a zero of f{z). We call it the canonical product formed

with the zeros of/(z); the integer p

is

called its genus.

INTEGRAL FUNCTIONS
7.51. Borel's theorems on canonical products

171

Borelf has proved two fundamental theorems on canonical


products which we shall deduce from the following lemma. $
If 0(z) is a canonical product of genus p with zeros zv z 2 ,...,

and

if

N is a positive integer such that


logfj 1-

where I

is

\z

<

2|z|

|% +1

|,

then

-7<log|(?(z)|< J

equal to

zn

(P

(P

> 0),

0),

2ii

+t

being independent of

N+l

ftl

z.

The canonical product

is

We

say.

In

shall denote

n 2 we
,

|io g

have un

|n 2 iK

|io g

\z\,

<
n2

\zn

\ and

<

\,

\z/z n

by

r,

rn

un

respectively.

so

iog^,^) <2 2<+!,


N+l

by the inequality of 7.2.


To obtain inequalities satisfied by U 1 we have to
and p = 0. Now when p
separately the cases p >
\z\ ~^\, we have
\E{z,p)\

and

<

(l

+ |2|)exp(| Z |+i| 2 |3+...+ i| Z

consider

>

and

|*),

so

log \E(z,p)\

<log(l+\z\)+\z\ + i\z\*+...+-\z\*>^A\z\P

t Acta Math. 20 (1897), 357-96.


j The proof of this lemma is a modification of that of Valiron, Integral
Functions (Toulouse, 1923), 53-8.


INTEGRAL FUNCTIONS

172

where

is

independent of z; similarly

\og\E{z,p)\

>log|l- z |-| 2 |_i|z|*.

\z\p

p
^log\l~z\-A\z\p.
Since

un

> \ in

IIj,

we deduce from

these inequalities that

N
z

2 ^g
> 0.

when #

If

-Af << log in; <^fg,


i

we combine

-2

2k

this

with the other inequality

< iog|n < 2JV+l


| g+i,
the result stated in the lemma for p >
follows at once.
When p = 0, we have
+1

JV+1

log

|G(()|

= log n
|

na

[-|-iog
i

N
C

log

+ 2 5>
+

iV

and

< 2^g(l+un )+2^_u


n =
iV+l

I,

also

log\G(z)\>"^log

-2

iV + l

iV

= 2 log

+ Zlog(l+un )-I

> 2 logl -

-/.

This completes the proof of the lemma.


We shall now prove the first of Borel's theorems, that the
order of a canonical product is equal to the exponent of convergence
of

its zeros.

Let jubea number such that p


r~ /* is convergent. Then, when

< < p+ and also such that


p > 0, we have

oo

= A2 -^+2 2
+
2V

<

2^-M

fi

<+2^
1

|+

2V

+1 ""

<,
1

since un

INTEGRAL FUNCTIONS
\ when n ^ N, and un < f when n

(iV

00

iV

173

> N.

This gives

+l

'

A similar proof can be given when p =

and leads to the same

result.

Let the exponent of convergence of the zeros of G(z) be p v


from the lemma that, when p. > pv the inequality

It follows

<ef

10(3)1

holds for all sufficiently large values of r, and so the order p of G(z)

cannot exceed p. Since p

is

any number greater than p v

this

We

have, however, already proved that p x


plies that p
pv
for any integral function. Hence, for a canonical product, p 1

im-

^p
= p.

From this theorem, in conjunction with the lemma, we deduce


and an arbitrary
number of circles of

that, given a canonical product G(z) of order p


positive

number

e,

there exists

arbitrarily large radius

an

infinite

on each of which holds


\G(z)\

>e-l*i p+e

This theorem sets a lower bound to the


say, of \G(z)\

on

\z\

we cannot expect
To overcome
M(r).

minimum

value, m(r)

r.

Since m(r) vanishes whenever r


G(z),

the inequality

is

the modulus of a zero of

that m(r) will behave quite as simply as


this

difficulty,

we

describe a

circle

\z n = r~ h about each zero z n for which rn = \z n


1,
h being any real number greater than p. Since ~^r~ h is convergent, these circles do not cover the whole plane, and so there
exists an infinite number of circles \z\ = r of arbitrarily large
radius which do not intersect any of the small circles having
\

>

zeros of G(z) as centre. It will, therefore, suffice to prove that

the inequality in question holds for all sufficiently large values


\z\, provided that z lies outside all these small circles.
The proof of the theorem depends on the inequality

of

log \G(z)\^ log

fl

1-- -/

established in the lemma.

As we have shown,

for all sufficiently large values of

r.

INTEGRAL FUNCTIONS

174

It will be recalled that the integer

inequality
z/z n

|1

rN

>

<

2r

< rN+1

provided that r
iog

Now when

> 2,

n 1-1

was defined by the


r n < 1, we have

and so

>0.

*n s;i

But when
we have

< rn ^

and

2r

z lies outside all the small circles,

i-l

Hence

log

x
Tr i+n

(2r) x + h

> iV(l+A)log2r.

2J

Krn <2r

We have, however, shown in 7.42 that, for all sufficiently large


values of

N < ArP+**.

r,

Combining these

results,

we

N
logfj
It follows that,
r

find that

>
when

A(l+h)log(2r)rP+* e

z lies outside all the small circles

and

sufficiently large,

is

log \0{z)\

This result
7.6.

is

> rp+i*{l + (A+h)log(2r)} >

-rP+*.

due to Borel.

Hadamard's

factorization

theorem

is an integral function of finite order p which has zeros


and does not vanish at the origin, it can be factorized in

If f(z)
Zy, z 2 ,...

the

form
f(z)

G(z)e*n,

where 6(z) is the canonical product formed with the zeros of f(z)
and H(z) is a polynomial of degree not exceeding p.
It has already been proved that such a factorization is possible, H(z) being an integral function. It remains to show that
H{z) is a polynomial.

Let

/>!

be the exponent of convergence of the zeros of f(z).


0(z) is of order p v and p x does not
lie denotes an arbitrary positive number, there exists,

Then the canonical product


exceed p.

as

we have

INTEGRAL FUNCTIONS
seen, an infinite number of

just

arbitrarily large radius

But

circles

\z\

=r

of

on which the inequality


\G(z)\

is satisfied.

175

>e~rH+e

since f(z) is of order p, the inequality

\m\<erP+e
holds for all sufficiently large values of r.
From these two inequalities it follows that

<e

|e
l

\G(z)\

for a certain set of arbitrarily large values of

principle of

maximum

r.

Hence, by the

modulus,

<

e H(*)|
|

e2r>>

+<

and so e B& is an integral


function which does not vanish and whose order does not exceed
By 7.41, H(z) is a polynomial whose degree does not
p.
exceed p. This completes the proof of Hadamard's factorization
for all sufficiently large values of

r,

theorem.
H{z) of degree q, the greater of the
of f(z). Since^
genus
p,
p and q
integers # and q is called the
It
order.
its
exceed
not
does
the genus of an integral function
can be shown that, when p is not an integer, the genus is the
integer less than p. But when p is an integer, the genus
If 0{z)

is

p and

of genus

<

<

greatest
is

either p or

p 1,

the actual determination in any particular

case being sometimes difficult.

Example. Prove
z |-i-j>log \G(z)\

->

that, if 0(z) is

as

\z\

->

oo.

a canonical product of genus p,

(Poinoabei.)

7.7.

The Taylor

finite

coefficients of

order

The necessary and

to be

an

an integral function

sufficient condition for

integral function of finite order is that

lim
m-co

log(i/Kl)

nlogn

>0j

of

INTEGRAL FUNCTIONS
The condition is necessary. For suppose
order p. Then, when 1c > p, the inequality
!76

M(r)

that/(z)

of finite

is

< e^

by the maximum modulus of f(z) for all sufficiently


values of r. Hence we have, by Cauchy's inequality,

is satisfied

large

The expression on the right-hand


a

maximum

side of this inequality has

value (ek/n) n lk attained


,

when

KK(-j
From

= njh, and so

rk

this it follows that

i^iog(i/Kl)
nlogn
so that the condition
It remains to

show

is

>0

necessary.

that, if

n log n
is positive, f(z) is

an

sider separately the

We

integral function of finite order.

two

cases,

(i)

finite,

(ii)

con-

infinite.

Let us suppose that p. is finite. The definition of p. implies


is an arbitrary positive number less than
p., there exists
an integer iV such that the inequality
that, if e

lo g(VKI)

holds whenever

> N.

>

(p-e)nlogn

This inequality
\an

<

may

be written

%-(/*-*).

But since pe is positive, this implies that |aj 1/n tends


n -> co and hence that f(z) is an integral function.
Now, when r > 1, we have

to zero

as

l/(*)l

< I KK+ f

|an |r

< Ar+

f (JLY

N+X
where

is

a constant. Let us choose an integer

Mv- e
If r

2V+1

is

< 2r <

sufficiently large,

< < if
to

'

(Jf+l)F-e.

will

be greater than N.

we have
r

M such that

< rM < exp{(2r) /(M-)logr}


1

When

INTEGRAL FUNCTIONS

177

and so

^ exp{(2r)>JIogr}j|

2, \^i)

-<^)

< exp{(2r) ^- 'logr} 2 n-^-^


= B exp{( 2r J^-^log r}

where

B is independent
Z,
M+l \iJFv
x

""^

'

of

r.

Also

"^Zl^
M+

2, Iu(jif+i )/-)'
+1

ikf

'

'

1,

Combining these inequahties, we find that

< ArN +Bex V {{2r)Wi>-*\ogr}+l,

|/(2)|

from which

it

follows that

M(r)

< exp{2(2r) /^- >logr}


1

for all sufficiently large values of r. Let us

of f(z) by
p

lo E lo 8

lim

r)

so

Again,

an

order

M ^ j^ log 2 + log(2r)iV-'>+loglog
^
logr

logr

r-*oo

and

now denote the

Then we have

p.

r
'

p <^ l/(p e).


it

infinite

also follows

from the definition of p. that there

number of positive
log(l/KI)

or

\an

integers

<

exists

n such that

(p,+e)nlogn,

>K,-(f+e)m_
\

Using Cauchy's inequality we find that, for each such value of n,

M(r)

we take

If

2nJl + e ,

M(r)

>

lojf

>

{^f.

we now have

>2 n = exp{(ir) '^+ 'log 2}


1

for certain arbitrarily large values of

r. This, however, implies


that p 2s l/(p,+e).
have thus shown that the order p of f(z) satisfies the
inequality

We

__ <p <
p+e
pe
T

for all positive values of e less than

we

find that f(z)

4111

is

p..

of finite order 1/p.

Making

tend to zero,

INTEGKAL FUNCTIONS

178

To

deal with the case

inequality

when

log( !/ 0n
|

/x is

)
|

infinite,

we observe

that the

> Xn log

and all sufficiently large


holds for any given positive value of
part
of the previous argurepetition of the first
values of n.
ment shows that /(a) is an integral function of order not exceedcan be as large as we please, the order of f(z)
ing IjK. Since

This completes the proof of the theorem.


It should be observed that we have proved incidentally that

must be

zero.

y.

nlogn
log(l/KD

is

the order of the integral function /(z).

REFERENCES
E.

R.
E.
G.

Bobel, Lecons sur les fonctions entieres (Paris, 1900).


NEVANLimsrA, Le Theoreme de Picard-Borel (Paris, 1929).
C. Titchmaesh, The Theory of Functions (Oxford, 1932), Chap. VIII.
Valibon, Lectures on the General Theory of Integral Functions (Toulouse, 1923).

See also G. Polya and G. Szego, Aufgaben und Lehrsatze aus der
Analysis (Berlin, 1925), 2, IV. Abschn., Kap. 1.

1.

Prove that,

MISCELLANEOUS EXAMPLES
> 1, the integral function

if ||

CO

is

of order zero.
2.

Show

that, if a

>

1,

the integral function


00

is

of order 1/a.
3.

Prove that.f

KK

if

io g

Iim

+1 j

logi

= K>0)

oo

the function 2) an 2

ig

an

integral function of order not exceeding

l//c.

Show

also that the order is 1/k if

Iim

io g

KK

+1 j

logn

f Use Stirling's approximation to the

and

positive,

See

8.5.

T{\+x)

= K>0

Gamma function,

e-*x*<J(2TTx){l+o(l)}.

that

when x is large

INTEGRAL FUNCTIONS
4.

179

Determine the orders of the following integral functions:


(i)

cos Vz;
00

7rrr

(iv)

2-*J(Vz)

>

0);

l2 "+2n!r(v+n+l)
o

5. /(z) is

an

integral function of finite order p.

Show

that, corre-

sponding to any positive number e, there exists an infinite number of


circles \z\ = r of arbitrarily large radius on which the lower bound m(r)
of

\f(z)\ satisfies

the inequality
m(r)

>

e-'"^.

the order of an integral function is not an integer


or zero, it is equal to the exponent of convergence of its zeros. Deduce
that such a function has an infinite number of zeros.
6.

Prove that,

7.

By means of Hadamard's factorization theorem

if

prove that,

not an integer or zero,


00

Bmir(z+h)

sin7r/iTT
CO

{(

+ ^T^) 6^ /<n+W

00

and

also that

simrz

ttz

j|l

+ -)e~* n
;

j.

)'

if

is

CHAPTER

VIII

CONFORMAL REPRESENTATION
8.1.

Isogonal mapping

that the functions u(x,y) and v(x,y) are continuous and possess continuous partial derivatives of the first
order at each point of a domain 8 in a plane in which x and y are

Let us suppose

u(x, y),
rectangular Cartesian coordinates. The equations u
v(x, y) set up a correspondence between the points of 8 and
v
the points of a set 2 in the (u, v)-plane. The set 2 is evidently

a domain and is called a map of 8. Moreover, since the partial


derivatives of the first order of u and v are continuous, a curve
in 8 which has a continuously turning tangent is mapped on
a curve with the same property in 2. The correspondence
between the two domains is not, however, necessarily a one-to-

one correspondence.

2
v = y2
is given by taking u = x
mapped on the triangle bounded by

simple example of this

The domain x
u = 0, v = 0,

+y <

1 is

u-\-v

1,

but there are four points of the

circle

corresponding to each point of the triangle.

A method of mapping 8 on 2 is said to be isogonal if it


a one-to-one transformation which turns any two intersecting
curves of 8 into two curves of 2 which cut at the same angle. In
order that the correspondence may conserve angle in the sense
just defined, it is both necessary and sufficient that the involution
of orthogonal directions at each point of S be transformed into
the involution of orthogonal directions at the corresponding
point of 2. This is equivalent to saying that the isotropic
directions at each point of 8 transform into the isotropic directions at the corresponding point of 2, and conversely.
Now the differential equation of the isotropic directions in

is

the

(x, 2/)-plane is

{dxf+{dyf

{duf +{dvf

Hence

0.

= h {{dxf + {dyf}
2

where h depends only on x and y and

is

not zero.

To determine the properties of the functions u and v for which


this equation holds, we make the substitution
du

= ux dx +uy dy,

dv

= vx dx +vy dy,

CONFORMAL REPRESENTATION
where

181

denote partial differentiation, and then equate


coefficients. It follows that the functions u and v satisfy the
partial differential equations
suffixes

we satisfy the first two equations by putting


ux = hcosoc,
vx = Asina,
uy = hcosfi,
v y = AsinjS,
we find that the third equation is also satisfied if a /? = ^7r.
Hence, if the correspondence u = u(x,y), v = v(x,y) is one-

If

to-one, it is also isogonal if and only if, at each point of 8, the


four first-order partial derivatives satisfy one of the following
sets of equations:
(a)

ux

= vy

uy

= vx

(b)

ux

= vy

uy

= vx

and do not vanish simultaneously


The equations (a) are, however, the well-known CauchyRiemann differential equations. In virtue of the initial restrictions on u and v, the conditions (a) and (b) are, therefore,
equivalent to
(a')

u+iv

where f{x-\-iy)
since

is

\f(x+iy)\ 2

= f(x+iy),

(V)

uiv=f(x+iy),

an analytic function, regular in 8. Moreover,


= ux -\-vx the derivative f(x-\-iy) vanishes
,

nowhere in S. In other words, the only isogonal transformations


of a domain S of the z-plane into a domain 2 of the w-plane
are of the form w = /(z) orw= f(z), where f(z) is an analytic
function whose derivative is finite and non-zero at each point
of S. Actually

it suffices

to consider only isogonal transforma-

tions of the former type, since the transformation

equivalent to

= f(z)

followed

by a

= f(z)

is

reflection in the real axis

of the w-plane.

Now if f(z) is an analytic function, regular in a neighbourhood


of the point z at which /'(z) does not vanish, we know, by the
inverse-function theorem ( 6.22), that the equation w
f(z)

up a one-to-one correspondence between a certain neighbourhood of z in which /'(z) does not vanish and the region within
a certain closed contour within which the point w = f(z ) lies.
Thus the neighbourhood of z is mapped isogonally on the region
sets

within the closed contour.


It must not, however, be supposed that,

if f'(z) is finite

and

CONFORMAL REPRESENTATION

182

/(z) necessarily
non-zero at each point of a domain S, then w
maps S isogonally on a domain of the w-plane. For, although
a neighbourhood of each point of S is mapped isogonally, it is
not necessarily the case that the mapping of the whole of S
2
is a one-to-one correspondence. For example, the function z has

<

<

\z\
2,
a derivative 2z which is finite and non-zero when 1
2
on
an
overdomain
this
maps
z
argz
fw; yet w
fw
\-n, and so the
argu>
|tt
\w\
4,
lapping region 1
transformation is not one-to-one.

<

<

Example. Show

that

<

<

w=

z/(l

<

z)

8.11.

\z\

real axis

<

from

isogonally on the

oo to J.

Conformal mapping

Let us suppose that


tion,

maps

whole w-plane, supposed cut along the

<

= f(z), where f(z) is an analytic func-

maps a domain S of

the z-plane isogonally on a certain


x(t)+iy(t)
is a regular arc z

of the w-plane. If C
in 8, the equation of its map is

domain

= f{x(t)+iy(t)},

G is mapped on a regular arc T, say.


and respectively the angles which
Let us now denote by
and Y at the points of the same parameter t
the tangents to
make with the real axis. Then we have
so that

ifi

<f>

= tan-

^
x

= &Tg{x(t)+iy(t)}

and
arg

f{x(t)+iy(t)}\

= arg[/'(z){^)+^(<)}]

= <Harg/'(z).
i/j
is equal to arg/'(z) and so depends only on the affix
point
the
z and not on the particular curve through that
of
point. This, however, implies that the transformation conserves
not only the magnitude of the angle of intersection of two curves

Thus

c/>

but also the sense of the angle.

since the transformation w = /(z) is equivalent to


=
w /(z) followed by a reflection in the real axis of the w-plane,

But

this transformation conserves the

intersection of

two

magnitude of the angle of

curves but reverses its sense.

conformal transformation

is

defined to be an isogonal trans-

CONFORMAL REPRESENTATION

183

formation which conserves the sense of an angle as well as its


magnitude. It follows that the only conformal transformations
of a domain of the 2-plane into a domain of the w-plane are of

w f(z), where f(z) is an analytic function.


Returning now to the regular arc C and its map T, let us
denote by s and a the lengths of C and V respectively up to
the point of parameter t. Then we have
the form

~
= V(* +2/ = \m+im\
dt
2

=
dt
da

j
and

jMt)+iy(t)} = \m{m+m}\;
hence we see that
= z n **.
t

dt

Jf'(

dt

We

have thus shown that the conformal transformation


w = /(z) maps a small neighbourhood of a point z on a neighbourhood of the corresponding point w which, correct to the
,

obtained by a magnification in the ratio


and a rotation through the angle arg/'(z ).
first

order,

is

|/'(z

)l

Example 1 . The domain S of the z-plane is mapped conformally on


the domain 2 of the w-plane by to
f(z). Show that the curves in 2
corresponding to x
constant and y
constant form two orthogonal
families. Verify that this is the case when (i) w = coshz, (ii) w
z2 .
Explain the apparent contradiction in (ii) when x
and y
0.

Example

2. Prove that, in the notation of Ex.

11

1,

the area of

is

\f(z)\*dxdy.

Example 3. The function w =/(z) maps ]z\


domain of area A. Prove that A > ir|/'(0)| 2 i? a

<R

conformally on a

Example

4.

Prove that the quadrant

\z\

<

1,

<

argz

mapped conformally on a domain S in the w-plane byui=


Find S and determine the length of its boundary.

<

4/(z

Jw

is

+ l)

The derivative of the function 4/(z+ 1) 2 is finite and non-zero at each


point of the quadrant. Accordingly w
4/(z+l) a maps the quadrant
conformally on a domain S provided that w does not take any value
twice in the quadrant. Now if

4/(*+l)

then either z

Zj

= z

z 1 or z
2 certainly is not.

formally on a domain

4/(zx

+ l)

2
,

2. But if zt is a point of the quadrant,


Hence w = 4/(z+l) 2 maps the quadrant con1

in the w-plane.

CONFORMAL REPRESENTATION

184

The boundary of 2 is evidently a closed contour T. But since w is


infinite only when z = 1, the quadrant is mapped on the domain
within r.

When

z moves along the real axis from


to \, w moves along the
from 4 to 1, and so the length of this part of V is 3.
When z moves from 1 to i along an arc of \z\ = 1, we have z = e 2U
where t increases from to 77. Then w moves from 1 to 2i along the
arc whose parametric equation is w = secHe~ 2ii From this we deduce
that \w\ = 2 Rlw>, so that the path is an arc of the parabola with
focus at the origin and directrix Rlw = 2. The length of this part

real axis

of

is

lt

/4

JI3-J 2aeoHdt = V2+log(l+V2).


Finally

when

2i

to 4.

moves along the imaginary

axis from i to 0, w goes


write z
itexit, where t increases from 77
to 0, we find that the parametric equation of this portion of V is
w 4cos 2ie 2i*. This curve is, however, the inverse of the parabola
w sec 2<e- 2tt with respect to the circle \w\ 2, and so is a cardioid
with cusp at the origin. The length of this portion of V is

from

If

we

=
=

JM*=J

Combining these

8 cos t dt

4V2.

w/4

-it/*

results,

we

see that the length of

is

+ 5V2+log(l + V2).

Simple functions
liw = f(z ) maps a domain S conformally on a certain domain
2 of the w-plane, we say that/(z) is simple^ in 8. The charac-

8.12.

teristic

property of such a function

more than once

in 8.

is that it takes no value


The general theory of simple functions

too difficult to be included in the present book.J It is, howshow that if f(z) is regular within and on
a closed contour C and takes no value more than once on C, then
is

ever, quite easy to

f(z) is

The

simple within and on G.


relation

= f(z) evidently sets up a continuous one-to-

one correspondence between the points of


t

The French and German words

For an account

C and the points of a

are univalente and schlicht respectively.


of the very interesting recent work on this subject, see,

for example, L. Bieberbach, Lehrbuch der Funktionentheorie, 2


(1927), 82-94;
P. Dienes, The Taylor Series (1931), Chap. VIII; E. Landau, Darstellung und
Begriindung einiger neuerer Ergebnisse der Funktionentheorie
(1929), 107-14.

CONFORMAL, REPRESENTATION
185
w is any point of the w-plane which does
on T, the number of zeros of f(z)w within C is

closed contour I\ If

not

lie

f(*)

^_

dw

>

(7

V being that which corresponds to the positive sense of description of 0.


Ifw lies outside T, then
0, and so no point within C is
mapped on a point outside I\ But when w lies within C,
we have
cannot be negative,f the equa1. Since
tion f(z)
w has precisely one root within G when w lies
within r.
To complete the proof of the theorem it only remains to show
that f(z) cannot take a value w x on T at any point z within C.
If this were the case, a neighbourhood of z x would be mapped
conformally on a neighbourhood of w^; this is impossible since
the neighbourhood of wx contains points outside T.
the direction of integration round

m=

m=
=

Riemann's theorem on conformal mapping


The fundamental problem in the theory of conformal mapping

8.2.

concerned with the possibility of transforming conformally


a given domain S of the z-plane into any given domain of
the w-plane. Actually it suffices to consider whether it is possible to map conformally any given domain on the interior of
a circle. For if
1 and w
f(z) maps S on ||
g{Q maps
2 on ||
1, then the equation w
g{f(z)} provides a conformal transformation of S into 5).
It is not, however, possible to map a completely arbitrary
domain on the interior of a circle; there must be some restriction
on the nature of the boundary. For example, we cannot map
a domain whose boundary consists of a single point conformally
on ||
1; for if the domain had a single boundary point,
which we may take to be the point at infinity, | the mapping
function
f(z) would be an integral function which satisfied
1 and so, by Liouville's
everywhere the inequality \f(z)\
theorem, would be a constant. Thus, in order that a domain
is

<

<

<

<

f This implies that the sense of description o'f T is the positive one.
a is the boundary point, it can be transformed into the point at
J If z
infinity by z x
l/{za).

CONFORMAL REPRESENTATION

186

may

be mapped conformally on a
than one boundary point.

must possess more

circle, it

The simplest case occurs when the boundary is a simple closed


Jordan curve. It can be shown that if S is the open domain
bounded by a simple closed Jordan curve G, there exists a unique
analytic function f(z), regular in S, such that w

= f(z) maps 8 con-

and also transforms a point z = a within G


into the origin and a given direction at z = a into the positive
direction of the real axis. This theorem was first stated by
Riemann in his inaugural dissertation^ at Gottingen in 1851, but
his proof, which depended on the Calculus of Variations, was
shown by Weierstrass to be incomplete.

formally on

<

\w\

Whilst

it is difficult to give a rigorous proof of this theorem,


truth will become obvious to the reader who is content to
rely on physical intuition. Let us suppose that such a function

its

does exist and consider what properties it must have.


In the first place, f(z)/(za) is regular and non-zero in S and
so is of the form e^ where <f>(z) is regular within C. Thus
(2,)

f{z)

Moreover, since

\f{z)\

the condition

Now

on

C, the function ${z) satisfies there

B1 ,,
log|z a\-\-tS\<p(z)
,

>

0.

the real part of an analytic function satisfies Laplace's

equation

d2y
dx*

where

{z-a)e^z\

= x-\-iy.

If,

e2

+ ~8yJ=0

'

therefore, it could be

shown that there

exists a unique real solution of Laplace's equation,

vanishes on

and

is finite

within

G save at the point

whose neighbourhood it behaves like log


<f>(z) would be given by the equation

V say, which

js

= a, in

a|, the real part of

= log \z-a\+m<f>(z).

The Cauchy-Riemann differential equations would then enable


us to find the imaginary part of <f>(z), save for an additive
constant.

physical argument renders intuitive the existence of such


t See Riemann's Ges. Werke (1876), 3-43.

CONFORMAL REPRESENTATION

187

a solution of Laplace's equation. For consider an earthed


cylindrical conductor of infinite length whose cross-section is
the curve C. Then V is uniquely determined as being the
electrostatic potential within this cylinder due to a line-charge
axis of the cylinder and
of density
\, which is parallel to the

passes through the point of affix a. Unfortunately, however, it


existence and
is just as difficult to give a rigorous proof of the
uniqueness of the electrostatic potential! as it is to prove Rie-

mann's theorem on conformal representation.


If the reader is prepared to accept this argument from physical intuition, it is easy to complete the 'proof of Riemann's
theorem. It only remains to choose the arbitrary additive constant occurring in the imaginary part of (f>(z) so as to make the
o correspond to the real axis in the
given direction at z
w-plane. We shall not attempt to give here an adequate proof
of Riemann's theorem, which can be found in various easily

accessible works. J

Homographic transformations
We shall now consider whether it is

8.3.

possible to

formally the complete z-plane, apart from a finite


exceptional points, on the complete w-plane. If it
it will be effected

by a relation w

= f(z), where /(z)

is

map

con-

number of
is

possible,

an analytic

function which has only a finite number of singularities.


The function f(z) cannot possess an isolated essential singularity; for, by Weierstrass's theorem ( 4.55), a function ap-

proaches as near as we please to any assigned value in every


neighbourhood of an essential singularity, and this is clearly
inadmissible here. The only singularities of f(z) are, therefore,
poles, of which the point at infinity may be one, and so f(z) is

a rational function ( 4.56).


But, by 6.21, Ex. 1, a rational function takes any assigned
value p times, where p is the number of its poles. Hence the
function /(z) has but one singularity, a simple pole, since a conformal transformation is one-to-one. If the singular point is at
Potential Theory (Berlin, 1929).
f See O. D. Kellogg, Foundations of
Caratheodory, Conformal Representation (Cambridge,
J See, for example, C.
1932), Chap. V; L. R. Ford, Automorphic Functions (New York, 1929),
Chap. VIII; G. Julia, Lecons sur la representation conforme (Paris, 1931),

Chap. III.

CONFORMAL REPRESENTATION

188

finite distance,

the transformation

w
where

and d

a, b, c,

the singularity

if

A (^

formation
8.31.

is

are constants such that

ad be

= 0;

but

at infinity, then

is

and

0)

of the form

(az-\-b)J(cz+d),

w
where

is

= Az+B,

are constants.

In either case the trans-

homographic.

Homographic transformations which leave the

unit

circle invariant

Homographic transformations which map the circle |a| = 1


on \w\
1 are of particular importance in the theory of func-

tions.

type
Jr

The transformation

if

w=

(az+b)/(cz+d) will be of this

az+b
cz+d
,

whenever

\z\

using conjugate complex numbers,

1;

(az+b)(dz+b)

The constants a,

b, c,

(cz+d)(cz+d).

and d must,

therefore, satisfy the equations

= cc+dd,
as well as the inequality ad ^ be.
aa+bb

Now,

if

we

cd~,

not zero, the second equation gives b


cd/a and
Substituting these values in the first equation,

is

find that

and so

ab

= cd/a.

hence b

\a\

When

...

7T
\aacc)(aadd)
,

=
=

\c\

or

\a\

|c|,

0,

\d\.

= a& where y is real. Sub= cd, we deduce that d = bev

we can

write c
stituting this in the equation ab
\a\

we require

Hence, when \a\


\c\, the transformation degenerates into
w e-v*, which is not a homographic transformation.
To deal with the case \a\
\d\, we write d
ae~ Si , where

is real.

We easily find that c =

be~ Si so that the transforma,

For an account of the elementary properties of homographic transformaG. H. Hardy, Pure Mathematics (Cambridge, 1921), 94-6. Some of
these properties are given in examples 6, 7, 8, 9 at the end of Chapter I of the
t

tions, see

present book.

CONFORMAL REPRESENTATION
tion

189

of the form

is

.az-\-b

As

a-\-bz

where

\a\

|6|.

A similar investigation shows that this formula

when a

also holds

0,

particularly simple form

the transformation then taking the

w=

eai /z,

where a

is real.

When w and z are connected by the homographic transformation

can be shown without

(^l), it

(1 ww)\a+bz\*
Hence when

>

\a\

(^1)

In

<
<

\z\

1,

we have

difficulty that

(1 zz){aa 66).

\w\

<

or \w\

>

according as

Thus the homographic transformation


maps \z\ < 1 conformally on \w\ < 1 only if \a\ > |6|.
8.33 we show that no other conformal transformation has
|6|

or

\a\

|6|.

this property.

8.32. Schwarz's

lemma

The proof of the result stated at the end of the previous


on the following lemma due to Schwarz.f //

section depends

regular in
satisfies the inequality
< R, where
< M, and iff(0) = 0, then the inequality \f(z)\ < M\z\/R
holds whenever
< B. Moreover, equality can occur only when
is a real constant.
f(z) = Mze /E, where

f(z)

is

it

\z\

|/(z)|

\z\

ai

a.

from the data of the lemma that f(z) is expansible


as a power series a 1 z+a 2 z2 +..., whose radius of convergence is
not less than R. The function <j>(z)
f(z)/z is, therefore, regular
It follows

when

\z\

<

R.

Let a be any number whose modulus is less than R. If we


choose r so that \a\
r
R, then, by the principle of the
maximum modulus, the function <f>(z) attains its maximum
r. Hence
modulus in \z\
r at some point on the circle \z\

< <

we have
\<t>{a)\

and so

< max

M = -max

|^(a)|

|/(2)|

< M/r.

This last inequality holds no matter how near r is to R, and


Making r -> R, we deduce that
<f>(a) is independent of r.
M/R, the first result of the lemma.
\<f>(a)
|

<
t

H. A. Schwarz

(1869), Ges.

Math. Abhandlungen,

2, 109-10.

CONFORMAL REPRESENTATION

190

number a of modulus less than B, such that


we know that <j>(z) is a constant of modulus
M/B. Hence we have f(z) = Mze ai/B, where a is a real conIf there exists a

= M/B,

\<j>(a)\

This completes the proof of the lemma.

stant.

Example. Prove that, if /(z) satisfies the conditions of Schwarz's


Show also that equality occurs only when
|/'(0)| < M/B.
ai
f(z) = Mze /B, where a is a real constant.
lemma,

Riemann's theorem for a

8.33.

circle

Let us suppose that w


f(z) maps \z\ < 1 conformally on
\w\ < 1 and turns the interior point z
c into the origin. We
shall now show that such a transformation is necessarily homo-

graphic.

We know
on
z,

\w\

we obtain a transformation w

\w\
all

=sC

and conserves the

= <^() which maps

origin.

This function

||

on

<() satisfies

the conditions of Schwarz's lemma, and so

when || < 1.
But we can

that
(z c)/(l cz) maps \z\
1 conformally
and transforms z
c into the origin. Eliminating

also write

this

transformation in the form

= O(w), where O(w) again satisfies the conditions of Schwarz's

lemma. Hence we have


ia

when

\w\

<

when || <
lemma now
the

1.

1,

and

so \w\

||.

If,

inequalities,

we

find that

The second part of Schwarz's


is a real constant. Thus

where a

gives

w
is

Combining these two

w = e a%
transformation w = f(z)

and so

<

is

of the form

1 cz

homographic.

in addition,

we

require that the direction arg(s

c) =

/?

at

be transformed into the positive direction of the real axis,


we must take a = jS.
We can now show that there is a unique function w = f(z)
which maps \za\
B conformally on \w\ 1 and also transc is to

<


CONFORMAL REPRESENTATION
forms an interior point

191

= c into the origin and a given direc-

For the
1,
(za)/B maps \za\ < B on \z'\
c into an interior point z' d and leaves the
transforms z
prescribed direction unaltered. We have only to apply a
homographic transformation to get the result stated. This completes the proof of Biemann's theorem for a circle.
Example. Show that the region |z a| < jR is mapped conformally
on |w| < 1 by
Rlz-c)

tion at c into the positive direction of the real axis.

transformation

z'

<

JR 2 -(z-a)(5-a)

where a is real and z

c is the point

'

which is transformed into the origin.

The conformal representation

8.34.

of a half-plane

on

a circle

We shall now consider the problem of mapping the half-plane


Imz

<

>

1 and simultaneously transconformally on \w\


0.
forming a given point c of that half-plane into the origin w
the
By choosing R sufficiently large, we can make c he within
circle \ziB\
B. As B tends to infinity, the interior of this

circle increases until it

Now the

up the whole

fills

half-plane.

transformation
in

Be ai (z c)
-(z-iB){c+iB)
1

maps \ziB\ < B conformally on \w\ < 1 and turns z


w = 0. If we write this transformation in the form
ie ai (z
c)
v

(1 icjB)z c

and then make B->co, we

find that

w=
where
\w

<

|A|

1,

maps the

and turns

zc

half-plane

= c into w =

It should be observed that,

make any

= c into

by

Imz

>

conformally on

0.

suitably choosing

prescribed direction at z

positive direction of the real axis at

priate changes in the argument of

A,

we can

= c correspond to the
= 0. Making the appro-

8.33,

we

easily

show that

COKFOBMAL REPRESENTATION

192

Riemann's theorem on conformal representation

is

also true for

a half-plane.
Finally, by making a preliminary translation and rotation,
any half-plane can be mapped conformally on the interior of

circle.

Example

1. Show that Imz >


is mapped conformally on Imio >
w = (az + b)/(cz + d), provided that a, b, c, and d are real and
ad be > 0. What happens if ad be < 0?
Example 2. Show that w = (i g )/(l + z) maps
< 1 conformally
on Rlw > 0. Determine the curves in the to-plane which correspond to

by

\z\

|z|

and to argz

a.

Schwarz's principle of symmetry

8.4.

Let

AB

be a closed contour consisting of a segment


of the
and a curve in the upper half-plane joining the ends
of that segment. Let f(z) be an analytic function which is
regular within T and continuous within and on T, and which
real axis

also takes real values

continue

on AB.

We

f(z) analytically across

/( z )=/(z)-

This result

known

is

shall

now show

AB by means

that

we can

of the equation

as Schwarz's principle of

symmetry.f
If

Tt

we

reflect

in the real axis,

consisting of the segment

We

plane.

when
on

within
within

on T,

By hypothesis,
Ii; for

r and

when

<f>(z)

is

regular within T. It

and z+h

lie

within

rx

we denote by C

is

also regular

and z+R he

=m-

the closed contour formed by the curved


<f>(z) is evidently continuous within and on G.
the
5.5,
function

r and L),

Hence, by

so

If

lower half-

<f>(z)

= lim M+R)-m
parts of

closed contour

by the equation <f>(z) = f(z)


and by (f>(z) = f(z) when z is within or

define the function

z is within or

T\.

we obtain a

AB and a curve in the

-f tz

2tti

a
t Journal fur

Math, 70 (1869), 106-7.

CONFORMAL REPRESENTATION
when z lies within F, we have

193

regular within G. But,

is

2m

2m

r
the value of the

first

Hence we

vanishes.

and, similarly, also

J
ii

integral

is <f>(z)

see that F(z)

and the second

<j)(z)

when z lies within 1^.

when

integral

z lies within T,

Finally,

by continuity,

when z is an interior
point of the segment AB, but not necessarily when z is an endfollows that this equation holds also

it

point of that segment.

We have
C.

Since

thus shown that the function <f>(z) is regular within


equal to /(z) within T, it provides the required

it is

analytical continuation of f(z) across


proof of the principle of symmetry.

Example

AB.

This completes the

r is a closed contour consisting of an arc AB of the


and a curve within that circle joining A to B. The function /(z) is regular within T and continuous within and on I\ By
mapping the arc AB on a segment of the real axis, prove that, if f(z) is
real on AB, it can be continued analytically across AB by means of the
circle

\z\

relation /(z)

1.

= /(l/z).

Example 2. The function /(z) satisfies the conditions of Ex. 1, save


that it now takes values of modulus 1 on the arc AB. Show that the
analytical continuation of/(z) across

Example

3.

Show

<

that, if

air <
is

<

argz

mapped conformally on the upper

8.5.

AB is given by/(z) =
1,

<

l//(l/z).

the angular region


era-

half-plane

by

w=

z 1 /! 2 "'.

The conformal representation of a polygon on a half-

plane

Let us suppose that the relation w


f(z) enables us to map
the domain within a closed polygon T in the z-plane conformally
on the region lmw>0. We propose to investigate the properties of this function f(z).

Actually it turns out to be more


convenient to consider, not /(z), but the inverse function
z
F(w).
Now the function F(w) is regular and F'(w) does not vanish

when

Imw >

0.

From
d

this it follows that


,

_,,.

F"(w)

CONFORMAL REPRESENTATION

194
is

regular in the upper half-plane.

also true at

any point on the

line

We next
Imw =

show that

this is

which does not

correspond to a vertex of V.
Let L be a side of Y making an angle 8 with the real axis.
Then if b is any point of L which is not a vertex of Y, (zb)e~ 0i
b,
is real on L. Hence if w
/} is the point corresponding to z
6i
segment
the function {F(w)b}e~ is real and continuous on the

A of the real axis corresponding to L, and is regular when


Imw > 0. By Schwarz's principle of symmetry, this function
can be continued analytically across A it is, therefore, regular
and can be expressed as a Taylor series
of the form
a>
{F(w)-b}e~ ei
2 ar {w-P) r
;

in a neighbourhood of j8

r=l

where the

ar are real. Now if the first non-zero


expansion were ap this would imply that the

coefficients

coefficient in this

two segments into which L is divided by the point 6 intersect


at an angle v/p. Hence a x is not zero, and so dlogF'(w)jdw is
regular in a neighbourhood of p and is real when w is real.

When
finity,t

where the

coefficients cr are real

in this case

dlogF'(w)/dw

and

is

L
z'

w~r

cx

the point at in-

is

^ 0.

tt

real

us suppose that

let

and

F'(w)

intersecting in the vertex

Next,

= fc

F"{w)

of the point at infinity

is

similarly

{F(w)b}e- 6i

Hence

=b

the point corresponding to z

we have

This gives

vf

regular in a neighbourhood

when

w is real.

and
are consecutive sides of
where Y has an angle cm. The

function F(w) cannot be regular in a neighbourhood of the point


z', since an angle am at z' is mapped
w' corresponding to z
on an angle it at w' But since arg{(z' z)e~ ei} is equal to zero

on L and L' respectively, the function [{z' F(w)}e~ ei ] l!a


is real and continuous on the segment of the real axis corresponding to the consecutive sides L and L'. Moreover, this

and

cm-

f In this case
length.

consists of

two segments

of the real axis, each of infinite

function

CONFORMAL REPRESENTATION
regular when lmw>0, since F(w)z'

is

and does not vanish


before,

we

there.

195
is

regular

Applying the same argument as

see that

[{z'-F(w)}e- 8iYI
regular in a neighbourhood of w' and can be expressed as
a Taylor series of the form

is

[{z'-F(w)}e- Bi]U<

where the

= f b (w-w'Y,
r=l

coefficients b r are real

gives

= z'+e

F{w)

ei

and

(ww')<*

2 b' (w-w')

r=0

where

b' is

r
,

not zero. Hence we have


F'(w)

where O(w)

not zero. But this

b x is

is

regular

(W M?') a -1 <I>(w;),

and does not vanish

in a

neighbourhood

of w', and so

t logi^(w)
&
v

dw

=
5_^.
w w '^<b{w)

Hence

if w' corresponds to a point at which V has an angle car,


the function dlogF'(w)/dw has there a simple pole of residue

a 1.
In a similar manner the reader will readily show that, if the
point at infinity corresponds to a vertex of T of angle oar, then

iogF'(u,)

aw

= -?+y*L.
w

'

/L,vf
2

Now suppose

that the points a, b, c,..., I (all of finite affix) lie


on the real axis of the w-plane and correspond to the vertices
of r, and let
.
om,

prr, yrr,..., Arr

be the corresponding angles of V. Then the function

dlogF'(w)/dw

must
(i)

satisfy the following conditions:

It is regular

when Imw^O save


01, y 1,..., A 1

residue oc1,
c,..., I
(ii)

respectively.

It is real

when w

is real.

for simple poles of

at the points a,

b,

CONFORMAL REPRESENTATION

196

It possesses

(iii)

an expansion of the form


&

dw

w^ Z*vf
=

valid in a neighbourhood of the point at infinity.

>

and real when


Since the function is regular when Imw
0, we can continue it analytically across the real axis
by Schwarz's principle of symmetry, and we find that the only

Imj

singularities of the function in the

complete w-plane are the

prescribed poles on the real axis. Hence, by


is a rational function and so is of the form

Uo

F'{w)
dw g

where

= ^1+^i

wo

+...+

4.56,

wl

the function

+(w)

<&(w) is a polynomial.

Now when

\w\ is large,

we have

d\ogF'(w)=<!>(w)+y^r

dw

-<

a -l)+(jS-l)

where

c(

But

cwr, jSw, y-rr,...,

since

polygon,

vf

+ (y-l)+... + (A-l).

Xv are the internal angles of a closed

(i_ a ) 7r+ (i_ 8) 7r+ ... + (l-A) 7r=

2tt

and so c[ = 2. Hence in order to satisfy condition (iii), <E>(w>)


must be identically zero.
We have thus shown that the mapping function F(w) satisfies
the differential equation

dw

r,,,

log F' {w)

Integrating this equation,

F(w)

A
B 1
a h~
w bi + H w Ir

1
= a

we deduce

= A\ (wa) a -

that

(w-b)P- 1 ...(wl) x - 1 dw

+B,

A and B are constants of integration. The integrand is


not one-valued in the complete w-plane, but has branch-points
a, b, c,..., I on the real axis. But if we take a definite branch of
the integrand and suppose that the path of integration never
passes below a branch-point, F(w) is regular when Imw
0,
save at the points corresponding to the vertices of F.
It remains to show that the constants A, B, a, &,..., I can be

where

CONFORMAL REPRESENTATION
197
determined. Now, with any set of real numbers a, b,
the correct order, the relation z = F(w) maps the line
on a closed polygon V, not necessarily simple, whose

suitably
in

c,..., I

Im w

same as those of I\ But, in order to construct


a polygon similar to a given polygon of n sides, it does not
suffice to make corresponding angles equal; there are n 3 other
conditions to be satisfied. We can, therefore, choose three of
the numbers a, b, c,..., I arbitrarily and then choose the rest to
angles are the

make

T' similar to I\ This done, the scale, orientation, and


I"" are determined by \A\,axgA, and
respec-

position of
tively.

We

can therefore choose

and

to

make

V coincide

with r.
Let us denote by C the closed contour in the w-plane which
consists of the segment of the real axis from
R to R, indented
at the points a, b, c,..., I, and the semicircle in the upper halfplane on this segment as diameter. The function

= i| {wa) a -

(wb)P- 1 ...(wl) x - 1 dw

+B

and on C. Moreover, when w goes round C,


polygon T, indented at the vertices and also at

regular within

is

z describes the

the point corresponding to w


oo. It follows, by 8.12, that
the region within C is mapped conformally on the region
within r. If we now make R tend to infinity and the radius
of each indentation of C tend to zero, we find that the region

Imic

>

polygon

is

F
z

mapped conformally on

of angles

cut, /fo,..., A7r,

= A j (wa)<*-

the interior of the closed

by the relation

(wb)P- 1 ...(wl) x - 1 dw

+B,

provided that the constants are suitably determined. This result


was discovered independently by Schwarzf and Christoffel.J

a is the point at infinity.


has to be modified; we now require

It is often convenient to suppose that

In

this case, condition

(iii)

that the expansion


a

+V^
^ Z, vf

#-logJ'=Aw &
w
v

'

t Journal Jiir Math. 70 (1869), 105-20.


% Annali di Math. (2) 1 (1867), 95-103; 4 (1871), 1-9.

CONFORMAL REPRESENTATION

198

should hold near the point at infinity. Making this change, the
reader will readily prove that

F(w)

=A

Examples

8.51.

(wb)P- 1 (wc)v- 1 ...(wl) x - 1 dw

+ B.

of the use of the Schwarz-Christoffel

formula
The region within a

a+fi+y

1, is

by taking

ABC,

triangle

mapped conformally on

=D

of angles oltt, /3tt, yrr where


the upper half of the w-plane

(wa^Hwb)*- 1 ^ c)t-i dw + E.

The constants a, b, c can be chosen arbitrarily, and D and E are then


fixed by the scale, orientation, and position of the triangle. In particular,
if we take 6 at infinity, the relation connecting w and z becomes
z

= D f (w ay-^w c)''-1 dw + E.

The only non-degenerate triangles for which this integral can be


simply evaluatedf are the equilateral triangle, the right-angled isosceles
triangle, and the right-angled triangle with one angle \tt. In each of
these cases w is an elliptic function of z.
There are, however, certain interesting degenerate triangles for which
the integral can be easily evaluated. These are discussed below.

AB

fixed and make y


(i) Let us keep A and C and the direction of
tend to 1. The triangle becomes the region between two straight lines
intersecting at A at an angle oltt. Taking a = 0, the relation connecting
w and z becomes

=
z

The reader

will' easily

Fw +E&

verify that this relation enables us to

map

the

an conformally on Iiaw > 0.


(ii) Let us keep A and C fixed and make jS tend to zero. The triangle
becomes the region between two parallel lines on one side of a transversal the corresponding relation between w and z is

<

angle

arg{(z E)/F}

<

nj,(wa)'- (wcy- 1 dw
1

where oL+y

+18,

1.

The simplest case occurs when the transversal cuts the parallel lines
l,we then have
at right angles, so that a
1, a
y
J. Taking

c=

dw
+E
a
-i)

-/:V(^
To

w=

verify that this


coshz. Writing z

is

the case,

x+iy,

w=

coshacosj/,

Dcozhri-w+E.

we take D =
u+iv, we find
v

1,

E=

0,

that

sinhxsiny.

f See Love, American Journal of Math. 11 (1889), 158-71.

so that

CONFORMAL REPRESENTATION
Keeping the positive number x fixed, let y vary from
evidently moves round the upper half of the ellipse

+ sinh
;

cosh 2a;

199
to

it.

Then

a;

to oo, this arc of the ellipse sweeps out


Moreover, as x increases from
the upper half of the w-plane. Thus w = coshs maps the half-strip
Rlz > 0, < Imz < ?r conformally on Imw > 0. The discussion of the
correspondence of the boundaries of the two regions is simple and is left
to the reader.

AB

fixed, and then make a and /8


(iii) Let us keep O and the direction
tend to zero and y tend to 1. In this way the triangle degenerates into
the strip between two parallel lines, which is mapped conformally on the
upper half of the w-plane by taking

(wa)

dw +E.

= 1,22 = 0, a = 0,
us consider the case when
to tt,
is kept fixed and Imz increases from
w moves round the upper half of the circle \w\ = e^1 ". Moreover, as
Rlz increases from oo to oo, this circular arc sweeps out the upper
< Imz < w conhalf of the w-plane. Thus w = ez maps the strip
formally on Imw > 0.
As a

verification, let

so that

w=

8.52.

The conformal representation

ez .

Rlz

If

of a rectangle

on a

half-plane
Christoffel shows that it is possible to
the interior of a rectangle conformally on the upper half of the
w-plane by a transformation of the form.

The theorem of Schwarz and

map

=D

[(w-a){w b^w-c^w-d)]-1 2 dw +E.


!

We shall now consider the particular transformation of this type


10

z=

[(l-w2 )(l kHv2 )]- 1 2 dw,


!

< k < 1.
This integral is an elliptic integral and its value cannot be expressed
in terms of the elementary functions of analysis. At the present stage
we cannot discuss the transformation by expressing w explicitly in terms
of z, as this involves a knowledge of the theory of elliptic functions.
Let us suppose that the integrand is positive on the real axis when
1 < w < 1, and that it is defined elsewhere on the real axis by continuation along a path in the upper half-plane. It is easily seen that the
where

2
1 2
integrand is equal to i[(w 2 1)(1 khv )']- ! when 1 < w < ljk, and is
-1 / 2 when w
1/k. We now consider how z
>
l)(A 2w 2
[(w 2
equal to
l)]

CONFORMAL REPRESENTATION

200

behaves when w describes the complete boundary of the positive quadrant


of \w\ < R, where R > 1/fc.
As w moves along the real axis from to 1, 2 moves along the real
axis from
to K, where
1

K=

[(l-w 2 )(l-& 2w> 2 )]-i/ 2

<Z

This integral converges, and so


is finite and positive.
When 1 < w < 1/k, the corresponding path of z is given by

[(l-^jfl-W)]- 1 /! dw +i

f [(M> 2 -l)(l-ifc 2 tt>2)]-l/2

dw.

The

first integral is

from

to

K'

as

K,

equal to

increases

whilst the second


1 to 1/k, where

is

real

and

increases

from

Ilk

K'

f [(w> 2

-l)(l-JW)]-1 /2

The number K'

is

from

moves along the

to l/k, z

When w >

evidently

finite.

the path of z

1/k,

dw.

Thus as w moves along the real


Rlz = K from K to K+iK'.
given by

axis

line

is

w
s

= K+iK'-

[(^-lXiV-I)]-'/' dw.
J"
Ilk

But

if

we

write

kw

f [(w 2 -l)(^ 2

\/t

in this integral,

w2 -l)]- 1 2 dw =

1/*

find that

[(l-^Xl-W)]-1 2 (ft

we

l/ftw

= K-

2
2 )]~i/ 2
j [(l- )(l-tt

<ft,

llkw

and so
Hence as
line

Imz

w moves

#'+

[(l-PXl-kH*)]- 1!*

dt.

(A)

along the real axis from 1/k to R, z moves along the


to the point of affix

K' from K+iK'

llkB

iK'+

[(l-* 2 )(l- 2 )]- 1 / 2 ^.

By

analytical continuation, the formula (A) holds everywhere in


1/k. Hence z is regular there and can be represented by a convergent series of inverse powers of w; in fact, if we expand the integrand
\w\

and

>

integrate term

by term, we obtain

(1+P)

kw

6k aw 3

CONFORMAL, REPRESENTATION
ziK'

This shows that

201

= ^- (l + ofr^))

when \w\ is large, and so the path of z, as w moves from R to iR, lies
within a circle of centre iK' whose radius tends to zero as R-> oo.
Moreover, the argument oiziK' decreases by %w in moving along this
path.
Finally

when w

iv

where v

is

we have

positive,

i f

[(l+v 2 )(l+fcV)]- 1 / 2

dv.

Since this integral is real and positive and decreases as v decreases,


moves along the imaginary axis from the point near iK' to the origin
as w moves along the imaginary axis from iR to 0.
have thus

shown

We
w moves round the complete boundary of the positive
\w\ < R, z moves round the rectangle with vertices 0, K,

that, as

quadrant of

K+iK',

iK', indented at iK', and that the radius of the indentation


-*- oo. From this it follows, by 8.12, that the region
tends to zero as
< argw < \n is mapped conformally on the interior of this rectangle.
Now z is purely imaginary when Rlw 0. By using Schwarz's principle of symmetry, we find that the region %n < argw < -n is mapped
conformally on the interior of the rectangle with vertices 0, iK',
K+iK', K. Combining these two results, we see that the relation

w
z

[(

-w

2
)(

- kho

)]- 1 ! 2

dw

enables us to

K+iK',

map the interior of the rectangle with vertices K, K+iK',

K conformally on the upper half of the w-plane.


REFERENCES

The
C.

theory of conformaf. representation:

Cab.athodoby, Conformal Representation (Cambridge Math. Tract,


1932).

A. R. Fobsyth, Theory of Functions (Cambridge, 1918), Chaps.

XIX,

XX.
G. Julia, Lecons sur la representation conforme (Paris, 1931).
G. Julia, Principes giometriques d 'analyse, 1 (Paris, 1930), 2 (Paris,
1932).

The application of conformal representation to physical problems:


H. Batbman, Partial Differential Equations of Mathematical Physics
(Cambridge, 1932), Chap. IV.
A. R. Fobsyth, loc. cit., 639-52.
H. Lamb, Hydrodynamics (Cambridge, 1916), Chap. IV.
M. Walker, Conjugate Functions for Engineers (Oxford. 1933).

CONFORMAL REPRESENTATION

202

MISCELLANEOUS EXAMPLES
1. Show that stereographic projection
conformally on a plane.

maps the

surface of a sphere

2. Prove that it is possible to map a surface of revolution conformally


on a plane in such a manner that the meridians and parallels on the

surface correspond to lines parallel to the axes of coordinates.


Show that, in the case of the sphere

X=
where

<

provided by
3.

Y=

i?sin#cos<,

<

<^<

Prove that, when

parallel to the real

w=

RcosO,

mapping of

2n, a conformal
Mercator's projection x
y
<f>,
-n-,

Z=

Rsm<?sin<,

this

type

is

logtan J0.

tanhz, the curves corresponding to lines


in the z-plane form two families

and imaginary axes

of coaxal circles.
Show that this relation maps the strip
the upper half of the to-plane.

< Im z <

\tt

conformally on

4. Prove that w
cosz maps the strip
< Rlz < ir conformally on
the whole w-plane, supposed cut along the real axis from
oo to 1
and from
1 to
oo.

5. The circles |z 1| = V2, |z+l|


V2 divide the z-plane into four
domains, of which one contains the origin; show that this domain is
mapped conformally on \w\ < 1 by w = 2z/(l 2 ).
(Math. Tripos, 1932.)
6.

Find a function

w = f(z)

which maps

|z

<

conformally on the

w=

exterior of the parabola v 2


correu-\-iv, with z
4m, where
sponding to w
3 and the positive directions of the real axes at these

points corresponding to one another.


7.

Prove that,

if

>

>

0,

(Math. Tripos, 1932.)

the equation

(a+b)z*2wz+(ab)

provides a conformal map of the annulus *J{(a b)/(a + b)} < |z| < 1
on the interior of the ellipse x2/a2 +2/ 2 /6 2
1, cut along the real axis

between

its foci.

Discuss the representation in the w-plane of the curves \z\ = r,


a, and, in particular, the behaviour of the transformation at
points on the boundary of the annulus.

argz

8.

Prove that

w=

(1+z 3 2 j(l z3
(l+z 3 ) 2 +i(l-z 3
)

2
)

2
)

the region \z\ < 1,


< argz < J77 conformally on \w\ < 1. Discuss the correspondence between the boundaries of the two regions.

maps

(Edinburgh, 1932.)

ww

CONFORMAL REPRESENTATION

203

domain S containing the circle \z\ = 1 is mapped conformally


on a region by w = f(z), the circle being transformed into a curve T.
9.

Show

that the curvature of

is

[i+bi{zT( 2 )//'(z)};k|/'(z)|.
w
10.

Show

that

2 )]" 1 / 2

[w(l

dw

maps the upper

half of the w-plane conformally

on the

interior of

in

square of side J ^/(cosec 6) dd.


TV

11.

Prove that

=
f

maps

<

\w\

(1

dw

4 )- 1 / 2

conformally on the interior of a square of diagonal

2 $ (lw*)- 1 !* dw.
o

12. Show that the region outside a square


mapped conformally on \w\ < 1 by

13.

The function

inequality

/(z) is regular in

< M; moreover,/(0) =

|/(z)j

to the function

prove that

14.

\z\

<

Show

plane Rlz

<

B.

,jy

15.

|z|

<

"

>

ft > 0, the relation


conformally on the strip

satisfies

the

';

',
,

conformally on

7rat> = 2ft log z


h < Rlw <

maps the
ft.

half-

Deduce that

1+z
01.1log
= 2ft
__

h < Elw <

h.

The function /(z) is regular in \z\ < B, where its real part
an inequality |Rl/(z)| < ft; moreover, /(0) = 0. Show that the

satisfies

inequality

^
llm/WK-log^
2ft

|z

holds

it

(CabatheodOey.)

TTIW

maps

B, where

By applying Schwarz's lemma

M*-cf(z)'

that, if

>

<

|z|

c.

jtf{/(z)-c}
(Z)

when

dw.

in the z-plane can be

when

\z\

<

B. (Cabathodoby.)

CONFORMAL REPRESENTATION

204
16.

The function /(z)

is

negative; moreover, /(0)

\z\ < B, where its real part


considering the function

regular in

1.

By

1-/(8)

show that the

inequalities

B+\ z \< VW\ < R-\ z


IT

hold when

\z\

<

tl

R. (Lindelof.)

2R

Z
\

is

never

CHAPTER IX

THE GAMMA FUNCTION


The definition of T(z)
The problem of finding a function

9.1.

when x

continuous

was

a positive integer,

showed that the

x n)
>

first

of a real variable x which


to x\ when x

and reduces
solved by Eulerf

positive

is

in 1729,

limiting function^ U(x) of the sequence

H(x,n)

is

when he

lZ

(W

/ Ti w
i
-l.
x
(*+l)(a;+2)...(a;+TO)

*'

2 *>"*
'

has the desired property. To prove this we observe that,


x,
a; is a positive integer and n
tt,

is

n\x\

>
~n =

when

x\nx

(+l)(n+2)...(n+a;)

(+*)!

-/IHFS-H)
as

-> oo.

Euler also gave a definite integral formula for this function


II (a), which is most simply obtained by considering the expression

(lr) n - 1 rc

dr,

which is now called an Eulerian integral of the first kind. When


n is a positive integer and x > 1, it is easily shown, by in-

by parts, that the value of this integral is


Writing t = nr, we deduce that
tegration

II {x,

n)/nx+i

U{x,ii)= f

(l

-J""

1
t

x dt,

from which

it

follows

(cf.

9.21) that
OO

U{x)=

e-H x

dt.

He announced his discoveries in two letters addressed to Goldbach. These


are printed in Correspondance matMmatique et physique de quelques ceT&bres
geometry du XVIIIi sttcle (Saint-Petersbourg, 1843), 1, 1-18.
% This notation is due to Gauss.
(

THE GAMMA FUNCTION

206

This infinite integral is called the Eulerian integral of the


second kind.
In the present chapter we consider the properties of the
Gamma functionf of a complex variable, defined by the
Eulerian integral of the second kind
00

= J e~U*-i dt

r(z)

whenever
has

its

this integral converges

principal value),

and

(it

defined

being understood that V~x

by

analytical continuation

elsewhere.

Example. Show

The

9.11.

To

that T(n)

(w 1)! when n

is

a positive integer.

analytical character of V{z)

discuss the convergence of the integral defining T(z), it is

most convenient to write


V{z)

= <D(z)+T(z),

where

<S>{z)

oo

= f c^-

dt,

T(z)

e-~ x

dt.

We

consider the function Y(z) first.


Let us suppose that z lies in a quite arbitrary bounded closed
domain. Then there exists a constant A such that Biz ^ A
when z lies in this domain, and so

when

1.

But

since e

^A_1 tends to

zero as tf-> +oo, there

depending on A, such that A_1


Hence we have

exists a constant C,
t

1.

when

e -^-i|

<

Ce when

<je -u

given domain, and so, by the


-test ( 5.52),
is uniformly and absolutely convergent. It follows, by 5.51, that T(z) is an analytic function,
z lies in the

the integral defining T(z)


regular in every

bounded closed domain, and

so

is

an integral

function.
f The notation T(z) is due to Legendre. For some purposes it is still found
more convenient to use the older notation of Gauss and write 11(2 1) for the

function denoted here

by

T(z),

THE GAMMA FUNCTION


To -deal with
transformation

the integral for


t

\ju,

*(*)

207

simplest to

<J>(z), it is

make

the

which gives

= j e-

u-z ~x du.

xlu

<

0. It is,
This integral obviously does not converge when Rl z
in
any
bounded
convergent
however, uniformly and absolutely
closed domain which lies definitely to the right of the imaginary

axis.

For in such a domain an inequality Biz


and hence

8,

where S

>

0,

is satisfied,

e -lluu -S-l

<^
I

M -S-l

> 1. The result stated follows immediately by the


and implies that <5(z) is an analytic function, regular
when Biz > 0.
We have thus shown that T(z) is the sum of an integral function Y(z) and an analytic function <I>(z) which is regular to the
right of the imaginary axis. We next show how to continue
when u
Jf-test,

$(z) analytically across the imaginary axis.

If in the formula

<D(z)

e-'F- 1 dt
,

j
o

Taylor series and then integrate term by


term, as we obviously may when Rlz
0, we find that

we

replace er*

by

its

>

o>( Z )

But

(-1)"
n\ (z-\-n)

and absolutely convergent


domain which contains none of the points 0, 1,

this series is uniformly

closed

in

any

2,...,

and so provides the analytical continuation of $(z).


The Gamma function is therefore of the form

where Y(z)

is

an

integral function.

'

'

The only singularities of

T(z) are thus simple poles at the points 0,


at z
n being ( l) n/n\

1,

2,..., the residue

9.2.

Tannery's theorem

In order to prove the identity of the Gamma function and


Euler's limit ( 9.1), we need the following lemma, which is the

THE GAMMA FUNCTION

208

analogue of a well-known theorem concerning


Tannery.f

Km /(*,) =

Jf

Km

then

n><x>

provided that

f(t,

f
J

f(t,n)dt=

n) tends to

its

such that

\f(t,n)\

due to

HmAw =+oo,

9(t),

and provided also that

interval,

series,

f g(t) dt,
J

any fixed

limit g(t) uniformly in

a positive function M(t)


values of n and t, and also such

there exists

< M(t) for all

00

that J M(t) dt is convergent.


a

Let r be any number greater than


enough to make A > t, we have

Then

a.

if

is

chosen

large

An

GO

jf(t,n)dt-jg(t)dt

+ ffdt-

j(f-9)dt

jgdt
A,

<
But

j U(t,n)-g(t)} dt

since \f(t,n)\

< M(t).

\g(t)\

< M{t)

+j

\f(t,n)\ dt

+j

for all values of n,

\g(t)\ dt.

we

also

have

This gives

00

jf(t,n)dt-jg(t)dt
T

<

00

{/(*,

n)-g(t)}

dt\

J"

Now f(t, n)
a

< <
t

t,

2J

M(t)

dt.

converges uniformly to g(t) in the fixed interval


so the first term on the right-hand side of this

and

inequality tends to zero.


A

lim
nj-oo
t See

Hence we have
oo

f(t,

n)dt

J*

Bromwich,

g(t) dt

<

j M(t)>dt.

Infinite Series (2nd edition), 49, 172.

THE GAMMA FUNCTION

209

00

we can make

Since the integral J M{t) dt converges,

the expres-

on the right-hand side of this inequality as small as we


by making t sufficiently large. But as the expression on

sion

please

the left-hand side of the inequality does not involve

t,

this

implies that
00

\l

lim
m>oo

and

so the

lemma

f(t,n) dt

- j g(t) dt

is

proved.

Example. Let F(n)

= 2 v (n),

where

r=0

with

n.

Then if v,(n) ->

lim.F(n)

provided that ]vr (n)\ <


converges. (Tannery.)

M,.,

where

= fw
r=0

r,

r is

formula for
now show that, if

9.21. Euler's limit


shall

tends steadily to infinity

wr as n -> oo for each fixed value of r, prove that


->oo

We

0,

independent of n and

~Mr

V{z)

then F(z, n) tends to P(z) as n -> oo, the convergence being uniform in any bounded closed domain D which contains none of
the singularities of Y(z). The proof falls into two parts.
In the first place, we have

But

since

i+

K)'('+rwhen

r is large, the infinite

,+o

product

converges uniformly and absolutely in


to an analytic function
F(z), say. This, however, implies that Y{z, n) tends uniformly
to F(z) in D. It only remains to show that F(z) is identical with
4111

T>

V
THE GAMMA FUNCTION

210

by the theory of analytical continuation, it


prove this whenf Biz
1.
if n is a positive integer and Rlz
1, it is easily shown

r(z); moreover,

>

suffices to

Now
by

>

by parts that

integration

V{z,n)

=n

(1r)"^-1 dr

(l--Yt^ 1 dt.
f
o

It will follow that


00

T(z, n) -

eHf- 1

dt

T(z),

provided that we can show that the conditions of Tannery's

theorem are

From

satisfied.

the equation

ww-r**follows that,

it

^ ^ n,

when

2n
o

Hence,

if

x denotes the

real part of z,
ix+l

and so
as

(l

- -]

-+

n -> oo, uniformly with respect to

first

e-*-1
t

condition of Tannery's theorem

Again,

we

also

in
is

any fixed
thus

interval.

The

satisfied.

have

00

But

since

>

1, J e

-^* -1

dt converges

and the second condition

of Tannery's theorem is satisfied. This completes the proof of


Euler's limit formula for T(z).
f

We

take Rl z

>

limit of integration.

1 instead of

Rl z

>

to avoid difficulties at the lower

THE GAMMA FUNCTION


we have shown

211

It should be observed that

incidentally that

This formula

is also due to Euler.


Example. Show, by vising Hurwitz's theorem

6.21,

Ex,

2),

that

T(z) never vanishes.

9.22.

Two

We

shall

important identities

now deduce from Euler's limit formula the two


important identities

= z r(z)
= Trcosec

r{z+i)

and

r(z)r(l-z)

To prove

the

T(z+1)

ttz.

we observe that

identity,

first

^^

= lim

+ 2)...(z++l)

-,(z+l)(z

n\ns

= zlim

nr-MoZ(3+l)(z+2)...(z

+ ) z+n+1

= zT(z).
From

this it follows

integer,

Again,

p(n)

we

also

immediately that, when n

_ 1}| p(l) =

"lUL

->ooZ(z+l)...(z

TC

^_^

mlml ~z

lim
72,)

m^oo(l z)(2 z)...(m+l z)

nlnln

= lim

-*=oZ(l 2

-Z2 )(2 2 -Z2 )...(2 -Z

Hence, by the example of

1.

Prove that T(i)

2.

Show

= TrcosecTTZ.
Vtt.

that
(2n)!

6.83,

T(z)r(l z)

Example
Example

a positive

have

= hm -

r(z)r(l-z)

(n

is

2 inrHT(n

+ i)/^TT.

)(rc+l_ Z )

THE GAMMA FUNCTION

212

Example

3.

Show

that, if

a; is

r ^i

real,

JLm^Y

Example 4. Prove that T(z,n) = nT(n+l)T(z)IT(n+z+l).


duce that n z T(n)IT(n+z) -* 1 as n -> co.

C Example

5. If

De-

denotes the derivative of log T(z), show that

<Jj{z)

tjj(lz)

l]j(z)

9.23. Legendre's duplication

7T

COt 773.

formula

was shown by Legendxe that T(2z) can be simply expressed


in terms of T(z) and T(z+|). To prove this result, we observe
that, by 9.22, Ex. 2,
It

r(2z, 2)

=
If

2z(2z+ i)(2z+ 2)...(22+2)


2 2z

~1

n2zn\ T(n+%)

\/nz{z+l)(z+2)...{z+n)(z+^)(z+l)...(z+n-^)
22Z_1

nz n)T(z+

we now make n -> oo and

find that

^V^z)

n)

r n + X^ z +i+ n
(

use the result of


2

2 *- i

9.22,

Ex.

4,

we

r(z)r(z+i).

Legendre's duplication formula.

This

is

9.3.

The Eulerian

integral of the first kind

The Beta function B(p,q) is a function of two complex


variables p and q, defined by the Eulerian integral of the first
kind

B(p,q)

1
j tP-^l-t)^-

dt

whenever

this integral converges, it being


p,-i

_e

(i_)g-i

(p-i)\ogi
}

understood that

=e

(a-i)iog(i-0

where the logarithms have their principal values. This equation


defines B(^, q) when the real parts of p and q are positive. For
other values of the variables, the Beta function is defined by
analytical continuation.
As we have just seen, there

is

a close connexion between the

THE GAMMA FUNCTION


shall

now

213

kind and the Gamma function. We


exhibit this connexion more fully by showing that

Eulerian integral of the

first

q'

F(p+q)

Tor simplicity of proof, we assume temporarily that the real


parts of p and q exceed unity.
Let us denote by SR the square bounded by the lines x = 0,
x B,y = Q,y=R. We then have
CO

T{p)T(q)

j"

00

e- xaP~ l dx

e-^- 1 dy

= hm

R^J

= lim

dx

dy e-x -yxp -1y^- 1

e-x-vxP-hfl- 1 dxdy,

f f

R-hx>JJ

Sb

the double integral being equal to the repeated integral since


the integrand is a continuous function of both variables. We
next show that

T(p)T(q)

lim

f f

erx-vxv-ryi-\ dxdy,

Tr

TB denotes the triangle bounded by the axes and the line


x+y = J?.
where

For,

if

we

call the

integrand f(x, y),

/(*, V)

dxdy

J" J"

we have

jj f{x, y) dxdy
tr

Sb

jj fix,y)dxdy
1x-Ts

<
<

jj \f{x,y)\dxdy
jj
Sb

But

dxdy- jj

\f\

\f

dxdy

= jj
sK

dxdy

SrI%

this last expression tends to zero as

jj
Sji

\f\

e --va>Rij>-iyHi-i

B -* oo,

since

dxdy -> r(Rl^)r(Rlg).

THE GAMMA FUNCTION

214

Hence

T(p)r(q)

lim

f f e -*--J'a;-

-q,

R
T(p)Y{q)

= Km

**%
the substitution

^-i(l q)-i

<Zq

e-^+a-1 d X

^- x (l <) a_1 <&

e-ff-w- 1 df

CO

V"

**-T,
equation we now make
we obtain

If in this last

x+y

V(p+q)B(p,q).

when

the real parts of p and q exceed


whenever the integral
however,
unity. The result still holds,
the
real parts of p and
defining B(p,q) converges, that is, when
q are positive. For the expressions on each side of the equation
are regular analytic functions of each variable.
We have thus proved that

This proves the theorem

w*
Bip

'

\
q)

- TMEM
~ W+rt

when the real parts of p and q are positive. For


of p and q, this equation is to be regarded as the

other values
definition of

the Beta function.


An important generalization of the Eulerian integral of the
first

kind

is

due to Pochhammer,f who showed that, with

this

general definition of the Beta function,

zP-^l zY~x dz

4:e^+^i smTrpsimTqB(p,q),

denotes a rather complicated contour encircling each


and 1 twice in opposite directions. For the
of the points
details of this we refer the reader to Pochhammer's original

where

memoir.
9.4. Euler's constant

We now show that the numbers


= 1 +l + l + -+l~ 1S n (n=

1,2,3,...)

f Math. Annalen, 35 (1890), 495. See also Whittaker and Watson, Modern
Analysis (1920), 256-7.

THE GAMMA FUNCTION

215

form a convergent sequence whose limit y lies between and 1.


This constant y, which is approximately equal to 0-5772, is of
great importance in the theory of the Gamma function and is

known as Euler's constant.


To prove this, we observe, in the first place, that the sequence

usually

is

a steadily decreasing one. For

*+i-

u.

On

= r-j-log(n+l)+log% = _L+Iog/l
n+1

n-f-i

the other hand, since

1/t

decreases as

+J->

1+1 + 1+

dt

>

4.

L\ < 0.

n+1)

increases,
l

+1

1
.

>

from which

it

follows that
1

< un <

1.

The sequence is, therefore, bounded and steadily


and so tends to a limit y. Moreover, by the last
y must lie between and 1.
9.41.

The canonical product

decreasing,
inequality,

for 1/T(z)

We have already seen that T(z) is an analytic function which


has no zeros and has simple poles at the points 0, 1, 2,...
From this it follows that l/r(z) is an integral function with
simple zeros at the points 0,-1,-2,.... We now exhibit the
property more clearly by proving that

TO-^n(HM'
is an integral function of order
due to Schlomilchf and F. W. Newman. J
To prove this, we start with the equation

so that l/r(z)
is

This formula

_ z(z+l)(z+2)...(z+n)

r(z,ra)
t Archiv der Math,
t

1.

n\nz

und Phys. 4 (1844), 171.


Cambridge and Dublin Math. Journal, 3 (1848), 57-60.

216

in the notation of

r(^j =

2 1
(

THE GAMMA FUNCTION


9.21. Hence we have
1

+l)( +l)-(

+$ e

-2log "

= 2 exp(2 (l + I + ^ + + I_log W J} {(l + |-^


))
...

as n -> oo, by the result of


the required result.
9.5.

Asymptotic expansions

A series

+^+^ + ...+^ +

which

may

zn

z2

all

Since T(z, n) -> V(z), this gives

9.3.

...,

either converge for large values of

values of z,

is

F(z), valid in a given range of values of arg z,


value of n, the expression

z4F{ Z )-A

When

\z

or diverge for

if,

for every fixed

-^l-^-...-M
n
z2

tends to zero as

|z|

an asymptotic expansion of the function

called

-> oo, whilst arg z remains in the given range.

this is the case,

function and the series

we denote the
by writing

relationship

between the

F(z)~A +^+^+....
This definition, which

is due to Poincare,f implies that the


between ^(2) and the sum of n terms of its asymptotic
expansion is of the same order as the (+l)th term when \z\ is
large, a fact which often renders an asymptotic expansion more
suited for numerical computation than a convergent series.
If a function F(z) possesses an asymptotic expansion

difference

F(z)~A +^l+^+...,
2
z

t Acta

Math. 8

(1886), 295-344.

THE GAMMA FUNCTION


the coefficients are determined successively

lim F(z)

=A

217

by the equations

]imz{F(z)-A

=A

1,

\z\>-QO

lim

#{f(z)-Ai-M = A 2

isl-

and so on. On the other hand, a knowledge of the asymptotic


expansion does not determine the function, since two functions
may possess the same asymptotic expansion. For example, ellz
and ellz -\-e~z have the same asymptotic expansion
z 2 .2!

z.l!

valid in the angle |argz|

9.51.

<

\it.

A more general definition of an asymptotic expan-

sion
It

may happen

possess

that, even though the function F(z) does not


an asymptotic expansion, there exists a function G(z)

such that

VM

G(z)

z2

in a certain range of values of arg z. In this case,

The term A

G(z) is called the

we

shall write

dominant term of this asymptotic

representation of F(z).

Example

1.

Show

that, if /(z)

~ 2 Am z~m and

9( z )

~ 2 Bm z~m

***

the same range of values of argz, then

Cm = A Bm +A x B^+Az Bm_^+...+A m B

where

Example

2.

Prove that,

if

f{x)

<~2-^m a;
a

positive,

then

Z* mxm
i

~OT

when x

is

large

and

THE GAMMA FUNCTION

218

9.52. Watson's

lemma

The following lemma, due to G. N. Watson,f enables us to


determine very simply the asymptotic expansion of a function
defined by a definite integral. Although the type of integral
considered is apparently very special, the lemma does, in fact,
cover many of the cases which occur in analysis.
Let us suppose that f(t) is an analytic function, regular, save
possibly for a branch-point at the origin,
a and 8 are positive, and let

when

< a+8,

\t\

= 2 am ^)-x

/(*)

when \t\^.a,r being positive. Let us suppose, further,


t is positive and t^ a,
1/(01

where

that,

when

< K4*

K and b are positive numbers independent of


'

where

t.

Then

GO

F(z)

f e-*f(t)
o

am V{mlr)z-m
dt~f
m=1

'r

and |argz| < %tt A, where A is an arbitrary


positive number. The lemma states that the asymptotic expansion is obtained by substituting in the integral the series for/(<)
and then integrating formally term by term.
To prove this lemma, we observe that, having fixed a positive

when

\z\

integer

is large

M, we can

find a constant

M-l

f(t)
is

true

such that the inequality

am #mM-A

<

CHfmsi-ijt

when t^0,no matter whether

M-l "

F(z)

=2

< a or not.

er*am &&-i dt

+BM

Hence,

if

M-l

= 1 am T(mlr)z-l*+RM
m=l
we have

to

show that

M BM is bounded as

t Proc. London Math. Soc.


Functions (1922), 236.

>r

(2),

17 (1918), 133.

\z\

-> oo.

See also Watson, Bessel

Writing z

THE GAMMA FUNCTION


x-\-iy, we find that

\Rm\

=
<

219

j e-*[f(t)-2a*fi*-*}

dt

e-xi ClfM^- 1 e bt dt

_
~

CT(M/r)
{xb) M r

'

xb is positive. But from the condition


\n
A,
we have x ^ Z |sinA, so that x 6 is positive
<
> tcosecA. Hence, when |argz| < \tt A < Jw and

provided that
|argz|

if

Z
|

\z\

> fecosecA, we have


Mlr

|Z

**'

<

CT{MJr)\z\^
(| Z

|sinA-6)^

this completes the proof of the

9.53.

We

now

(1)

'

lemma.

The asymptotic expansion


shall

_
~

of T(z)

discuss the behaviour of T(z) for large values

by finding the leading terms in its asymptotic expansion,


valid when [argz| < it. One way of doing this involves proving
of

\z\

first Stirling's!

logr(z)

In

asymptotic series

Z _i)l 0gz

_ z+i

0g (2.)+

this formula, the coefficients

defined

|^-Jtg^-.

are Bernoulli's

numbers

by the expansion
izcothiz

= 1+2 (- 1 r_ljB'-(^)!)

T S= X

The required asymptotic expansion of T(z) can be deduced from


its logarithm, though it is difficult to give any simple

that of

formula for the coefficient of the general term.


As this methodj involves a number of difficult preliminary
transformations, we obtain here the asymptotic expansion of
First published in his Method/us differentialis (London, 1730), 135.
j See Whittaker and Watson, Modern. Analysis (1920), 246-53.
f

THE GAMMA FUNCTION

220

T(z) directly from the definition as an integral, by a method


based on Watson'sf discussion of a closely related problem of

Ramanujan.

When

and

z is real

positive, the transformation

T(z)

and

= - r(l+z) = -

Although we have proved


real

f e-'t* dt

z~zez V(z)

so

and

= z er
z

f (ne 1 -")*

du

(we 1 -") 2 du.

by supposing that z is
by the theory of analytical

this formula

positive, it will also hold,

any closed domain

continuation, in

zu gives

00

00

in

which z~zez Y(z)

is

regular

and the integral is uniformly convergent. The formula is, therefore, true when Rlz > 0.
Let us suppose, then, that Rlz > and that \z\ is large. We
from
from

to
1

to

1,

u increases steadily

from
and then decreases steadily from

observe that

ue>-~

as

to

u increases
u increases

We now write

oo.

oo

-T()

(we 1 -")*

du

+ j (Ue - V y dU,
x

it

to 1 as
1

being convenient to use different symbols for the variables

in the different parts of the range of integration.


we make the
If, in the second integral,
e~'

from

Ue1 - 17
1

to oo

increases steadily

on the other hand,

from
if

to oo as

we put

er

first integral, t

decreases steadily from oo to

from

this

to

1.

In

way we

substitution

increases

= ue ~ u in
x

as

the

increases

obtain the equation


00

to which

we may apply Watson's lemma.

Now U and u

are the
t

t Proe. LoncUm Math. Soc.


end of the present chapter.

two

real solutions of the equation

= u 1 logtt.
(2),

29

(1929), 293-308.

See also Ex. 18 at the

THE GAMMA FUNCTION

221

In order to determine the nature of these solutions we consider

& = w-logQ+w),

the equation

denning

as a function of the complex variable .


regarded
as a function of w, is two-valued in the
,
neighbourhood of the origin, its two branches being

Now

Since each branch

<

= w(l |w-f-|w;
2

an analytic function of w, regular when

is

with a simple zero at


that the equation
\w\

1,

.)i/a.

w=

follows,

0, it

= w{\ %w+lw
2

...)

by

6.22, 6.23,

1/2

possesses a unique solution

regular in a neighbourhood

||

<p

of the origin.

As we saw

in 6.23, the coefficients a n are given by the fact that


residue of t,~ n at
0; in particular,

a2

3>

ai

86'

270'

We shall denote this solution by w-^jC).

is

the

4320-

Similarly, the solution of

= w (l fw+1102
where w
< p, is w = w

...)l/2,

= w i( )
||
2 ()>
2 ()
have thus shown that the function w{t) defined by the

regular in

We

ah

na n

equation

= w log(l+w)
has two branches w = w (0> w = w
each regular in a neigh,

.,.
2

2 (),

bourhood of the
larities

origin.

The only branch-points or other singu-

of m>() are the points at which dwjdt,

Since

dw

1.

zero or infinite.

_ Ul+w)
w

d
these are

is

and the points corresponding to

Of these,

is

w=
w (),

not a branch-point of

and
and

the value of corresponding to w


1 is infinite; thus the
only singularities of w^jX) of finite affix are given by 2
imri,

where n

is

a positive or negative integer. Similarly for

w> 2 ().

THE GAMMA FUNCTION


we now put 2 = 2t, we see that the functions U, u are
regular when \Lmt\ < 2n, save for a branch-point at the origin,
and that, when \t\ < 2tt,
222

If

U=

l+(2tyl*+a 2 (2t)+a3 (2t) s l*+ai (2ty+...,

l-(2t)^+a (2t)-a3 (2tfl 2 +ai (2t^-...,

;i

the square roots being positive

when

is

real

and

positive.

Lastly, since

dt

d(Uu)/dt is bounded when

lemma

son's

> e > 0.

The conditions of Watsatisfied, and the asymptotic

are, therefore, all

expansion of

is

Ul + l-u'

dt~U-l~i~l-u

obtained by substituting for

and u

their expansions in

powers of ^j{2t) and integrating formally term by term. From


this it follows that, when |argz| < \n A < \n and \z\ is large,
z-e?r(z)

~ V( 2 r (4)+ 3r (f)3(2/2)+5r(l)a
~ V(2W2)[l+3 /z+lW3 +-]-

6 (2/

/2)[

Z )2

...],

Hence the required asymptotic expansion


1

The range

9.54.

of

*>~-r/(7)[

is

+i5+se +

--]-

of validity of the asymptotic expansion

We next prove that this asymptotic

expansion of T(z)

is

valid

\z\, except when z lies in an arbitrarily


small angle enclosing the negative part of the real axis.
followed from
The validity of the expansion when El z

for all large values of

>

applying Watson's

lemma

to the formula
CO

z~zeV{z)

e~d F{t) dt,

where

F(t) = d(Uu)/dt. Now by applying Cauchy's theorem

to the integral of e-^Fty) round the complete boundary of the

THE GAMMA FUNCTION

< R bounded by axgt =


show that, if \n < a < \n, then
sector of

e-<*F(t) dt

= j exp(zte

223

and argt

|*|

ai

<x,

we

easily

)F{te ai )e ai dt

when

the real parts of z and ze ai are positive. But the latter


integral converges uniformly in any closed domain for which
Rl(ze ai )
and represents an analytic function. The expres-

>

on the left-hand side of this equation is equal to z~seT(z)


when Biz > 0. Since the half-planes Rlz > and Bl(ze orf ) >
have an area in common, it follows, by analytical continuation,
sion

that

oo

z-se*r(z)

exp( zteai)F(te ai )e ai dt

when m(ze ai )

> 0, provided that fn- < a <

Writing temporarily

ai
for ze ,

far.

we obtain, by Watson's lemma,


00

a<
i
af
j exp(-z*e )-f(te )e
o

dt

e -^F(te ai )e ai dt
J*

\z/

From

12

i"288^"7

T 12z T 288zaT

this it follows that

V\/\
when

|arg(ze a *)|

12z^288z2^-

<

Jtt A, where A is an arbitrary positive


number. But since a is any number between frr, this implies
that the asymptotic expansion is valid under the single restriction |argz|
77 8, where 8 is any positive number.

<

9.55.

The asymptotic behaviour

of \Y(x+iy)\

when y

is

large
Since T(z) takes conjugate complex values at points which
are symmetrical with respect to the real axis,f we have

|iW =
t This
it is

real

r(z)r(z-),

obvious from the canonical product for l/r(). Alternatively


a consequence of Schwarz's principle of symmetry ( 8.4), since T(z) is

when

is

z is real.

THE GAMMA FUNCTION

224

and

when

so,

<

|argz|

it,

|r(z)l

Writing

= x+iy,

\T{x+iy)\

~27re-*-V25/|z|

this gives

2TTe- 2x (x+iy) x+iv (xiy)

= 2ire-2x (x +y
2

Then

~
?/

is

large

l\z\

2ire-2

"Oyix-^-ny+ix

r(x+iy)

results,

\T(x+iy)\
is finite

e-^+2 arctan

27rt/2*-i e -7'.

~ 2tt( yf*-H

\T(x+iy)\ =

Combining these two

when x

2
|

we remember

follows immediately if

+tl

and negative, the corresponding asymptotic

formula

9.56.

e- 2vaT8z

(~2\a;-l/2
1

When

x - iv

suppose that y increases indefinitely, whilst x remains

Now
finite.

x - ll2

and

\y\

Another proof

that

\T(x-iy)\.

we

~ V(2^)

have, finally,
\y\*-V*er*yU*

large.

of Legendre's duplication

formula

We now

give an alternative proof of Legendre's duplication


formula, depending on the use of Liouville's theorem.
Let us consider the function

= 2 *r(z)r(z+i)/r(22
2

<f>(z)

).

Its only possible singularities are the poles of the

numerator and

the zeros of the denominator. Now the numerator has simple


0, 1, 2, 3,...; but since these
poles at the points 2z
points are also simple poles of the denominator, they are not
singularities of <(z). The denominator T(2z) never vanishes.

Hence

<(z)

has no singularity of finite

affix

and

so

is

an

integral

function.

Again

<j>(z)

is

a periodic function of period

flz+1)

fa)

2 s(s+i)
(2z+l)2z

1,

since

by the recurrence formula for the Gamma function. Hence, if


we can show that fa) is bounded when Rlz > 1, say, it will

THE GAMMA FUNCTION


by periodicity, that it is bounded all over the
But, by the asymptotic expansion of r(z), we have
follow,

<f>(z)

1/2

(l

2(j)

+ ^( +
1

y=

225

z-plane.

2^1 + 0(1/ W)}

when \z\ is large and |argz| < w Hence <j>{z) is certainly


bounded when Rlz > 1.
We have thus shown that </>(z) is an integral function which
remains bounded as |z|^oo in any manner. By
Liouville's
.

theorem, this implies that </>(z) is a constant, its value


2Vtt being
found by making |z|->oo. This completes the
proof of Legendre's dupHcation formula
77*r(2z)

Example. Prove Gauss's


tive integer,

r( Z )r( z

2^-ir(z)r(Z +|).

multiplication theorem that, if

+I)r(z+ ^)...r(z + ^-1 =


)

n is a posiv

(2w )*<"~*-r(n*).

Hankel's contour integral for l/r(z)


The Eulerian integral of the second kind denned
T(z) only
when the real part of z was positive. We now
introduce a
9.6.

contour integral, due to Hankel.f which represents


the
function under

much

Gamma

less restrictive conditions.

Let us consider the contour integral

dt-z dt.

The integrand has, in general, a branch-point at the


origin, but
each branch is a one-valued function of t, regular
in any domain
of the complex plane supposed cut along the

-co

to

0.

real axis

We choose the branch for which


(ft-z

= e'-s

from

10 !?',

where log* has its principal value.


Let D be a contour in the cut plane which starts at
p on
the lower edge of the cut, goes round the origin
in the positive
sense, and returns to
p on the upper edge of the cut. This
contour can be continuously deformed without crossing
the cut,

t Zeitschriftfur Math. u. Phys. 9 (1864).


4111

7.

THE GAMMA FUNCTION

226

until

from
p to 8,
edge of
upper
and then the

consists of the lower edge of the cut

it

< 8 < p, the circle

where

the cut from

to

\t\

8,

p. This deformation does not

alter the

value of the integral.


Now on the upper edge of the cut, we have t uevi where
u is real and positive, and so the integrand takes there the value

g uzlogu ZTTl == g ^z qS7.

Similarly

grand

is

on the lower edge of the cut

now

e~ w u~ z eZ7ri

= ue-, and the inte-

Hence we have
p
f e't~z

dt

(e* e-zni )

e- uu~ z

du +1,

I=j exp(8tP)&-&-*H d0.

where

77

But

if z

= x-\-iy,
IT

\I\

<

&--xe* oose+y6 dd

<

2TT^-- x e

h +^y\

7T

and so / tends to zero with


thus shown that

8,

provided that x

<

1.

We

have

p
z

f eft~

dt

2ismrrz

e~ u u~ z dz,

>

0.
provided that Rl(l-z)
obtained from
contour
the
denote
Now let C

p tend

to infinity.

We

by making

then have
00

[4t-z

dt

= 2isin7TZ

&
Since r(z)r(l z)

e~ uu-z

du

2&'sin7rzr(l z).

= TrcosecTrz, it follows that

^J_ = JL
1

^(Z)

f <}t- z dt.

2771 J
-o

>

0,
Although we have only proved this formula when Rl(l z)
analytical
theory
of
the
of
z,
by
values
aU
for
it does, in fact, hold
continuation, since the expressions on each side of the equation

are integral functions.

THE GAMMA FUNCTION

227

This completes the proof of HankePs formula, that

o
for all values of

where the contour

z,

starts at the point

on the real axis, encircles the origin once in the


and returns to its starting-point.t
It should be observed that
(2 )

save

0,

we have shown

o^

2^ sin 772 J\

J*-1

oo

positive sense,

incidentally that

dt

for all values of

z,

1,

2,...

REFERENCES
N. Nielsen, Handbuch der Theorie der Gamma-funktion
(Leinzis
8

1906).

'

E. LnsTDBLOF, Le calcul des residus (Paris, 1905), Chap.


IV.

MISCELLANEOUS EXAMPLES
By

integrating p-ier* round the complete boundary of a


quadrant
of a circle, indented at the origin, prove that
1.

OO

^e-* dt =

e-"zi / 2 r(z)

provided that

<

<

Rlz

Deduce the values of

1.

00

00

tf^cos t

f t'-tsin t dt.

dt,

o
2.

By

integrating

~H~ round a
l

suitable closed contour,

show that

CO

f e -stt*-i dt

s~T(z)

provided that the real parts of s and z are positive. Deduce the
values of
00

j*

^ie-coscos (A<silla)

df

when A > 0, x > 0, and Jtt < a <


3. Show that, when x is positive,
which occurs between
t

and

%tt.

T(x) has a single

2.

We shall frequently write

(0

+
f

oo

for

minimum

value

THE GAMMA FUNCTION

228

CO

Show

4.

B(p,q)

that

provided that the real parts of


T(Z)T{1Z) = TTOOSeOTTZ.

Prove that

5.

jf^^+i'

p and

Deduce that

q are positive.

, ,.

/ 2

provided

that,

Show

6.

the real parts of p and g are positive.

that,

when the

real parts of p

and q are

positive,

(o+6)W

(a+&*)"+

J
o

provided that a/6

by

Prove,

7.

uvw

z,

is

not a real number between

and

1.

y + z,
using the transformation u = x+y+z, uv
if /(*) is continuous and a, /3, y are positive,

or otherwise, that,

where the triple integral is over the volume bounded by the coordinate
planes and the plane x+y + z = 1. (Dibicbxet.)

t,y = t+u, or otherwise,


Prove, by using the transformation xy
are
positive,
and
and
a.
continuous
j8
is
that, if /(*)
8.

f dx f

9.

ww-xrvii-^

Show

that Euler's constant y

is

ff+ffl

J/w(

-*)* fP ~ 1 *

given by

*--s[/{-(-^?-J(-3"?
L

10. If

tji(z)

denotes the derivative of logics), prove that


oo

*>=-y-jJ+2-sii)-

THE GAMMA FUNCTION

229

Deduce that

0(1)

= -y+ ^f-

0(n+l)

-y,

7-=l
00

Show

also that

il>'(z)

rs
A* (n+zf
>

-;

n=

11.

Show

that
0(4)

12.

-y-21og2,

Prove that, when Rlz

*'>

>

fcr.

0,

= J(t-iS*)*-

Deduce that

oo

0(z+l)
13.

0'(|)

Prove that,

if

= +k>gz
Rlz

>

(jcoth^ -je^cft.

0,

221

i+'
J

d<

Deduce that

- dt

(l

+ |)-A0(z).

i0(z

+ )log

T(x+i)T(y)

log
&

r(y+4)r(a;)

provided that the real parts of x and


14.

Show

are positive.

2/

and

that, if the real parts of a

6 are positive,

1-a;
o

Deduce that
l

(a^-a^Xl-as")

_,

(l-x)loga;

T(a + c)r(6)
B r(6 c)r(a)'
+

provided that the real parts of


15.

c+o,

a, b,

The logarithmic -integral function

li(z)

is

+c

loi

OC

Prove that

li(e_z )

all positive.

denned by

dt

are

du,

THE GAMMA FUNCTION


when |argz| < ^7r A < \tt,

230

and deduce

that,

(1

16.

Show

when

that,

\tt A

<

|arg2|

2<

1!

<

3'

\tt,

CO

f _.

.,

J\

a-1

(a-l)(a-2)

(Leqendbb.)

17.

The

error function

defined

is

by
00

Erfcz

Prove that, when

|argz|

<

<

77 A

f ert'dt.

tt

/i_j_+i^_i^_5 +...).
41-J-+
2 z
18.

Show

2V

3 6

2 2z 3

l2s

that, if
1

+ V. + 2l + - + (n-l)l + n\n

*6

'

co

01 (we1-")" cftt '

Deduce

that, as

n ->

(Ue l~ u ) n dV\.
'

'

00,

f'~

1,_*
-r

135n

_ "*"""'

2835w 2

(Ramanttjan, Watson.)
19.

Prove that,

if

1- + -... + <-<>--,

<*=!_

then

Hence show that, as n


1 to , and that

e -

1_-1

dw,

H'-3'-

increases

from

to

00,

<f>

n decreases steadily

from

^~2 + 8^ + 32^ + -20.

Deduce from Ex.

12 that,

if \z\ is

large

and

|argz|

<

\tt,

then

Hence show that|


logr(z+l)

~(z + i)logz-z + ilog27r+

J^T^i-

t Use 9.51, Ex. 2. The constant \ log 2tt is determined by using the
dominant term of the asymptotic expansion of F(2+ 1).

THE GAMMA FUNCTION


Find an asymptotic expansion, valid when

21.

Rl z

231

>

0,

is

\z\

and

large

of the function

V(i+*)'

Extend the range

of validity of this expansion

= 4- (^ cos z - Y sin z),

2L* dt

Vz

V*

and deduce that

Z
00

= -V(Xsinz+Fcosz),
^dt
Vz
Vi
where,

when

\z

is

>

and Rl z

large

0,

1.3.5.7.9

1.3.5

X ~2z~lW +
v
X
22.

Show

that,

when

13

l-3-g-7

2_t"
i

(2z)
is

\z\

large

cos(t-z)

(2z) 5

(2z) 4

and

,,

|argz|

<

Jot A

<

Jot,

5!

3!

*2

z4

z6

2!

4!

z3

z5

Z
OC

sin(i-z)

,,

23.

By

integrating

ttz-*

coaeCTTSJT(\s), where z

the rectangle with vertices


that

2m
Prove that
S

>

this

is positive,

aiB, RiR, where a

is

positive,

round

show

(7+ooi

z~ s T(s) ds

e- z

J
a ooi

formula also holds when |argz|

<

J77 8, where

0.

24.

Deduce from Hankel's contour

= 2m
_L

T(z)

where a

>

0,

25. /() is

where c

is

provided that Rlz

>

integral that

a+coi

Jtr*dt,

J
0.

(Laplace.)

an analytic function of z, regular in the half-plane Rlz < c,


ar denotes the value of the rth derivative of /(z) at

positive;

THE GAMMA FUKCTION

232

Show

the origin.

that the equation

(0+)

\mzf-

a- 1

dz

2isina w

f>- 1

[f(-t)-a

+a t1

^+

...

+ (-l)*+^t*}dt

o
is

provided that the integer k

true,

lies

between Rl( a) and Rl( a 1 ).


(Cattchy.)

26.

Prove that, when Rl z

<

0,

00

where k

is

the integer which

between Rl( z) and Rl( z I).

lies

(Saalschutz.)
27.

Show

- =

that the equation

l(z

infinite sets of roots,

(logB)

as

n ->

oo,

28.

where

which behave asymptotically


-rc{l

integer n.

+ l)

c,

^ 0,

has three

like

+ o(l)},

,i'T^i,;Tj^;

one root of each set being associated with each large positive
(Habdy.)

Show

that the inverse factorial series


00

z(3+l).. .(z+n)

converges whenever J,an n~ z is convergent, provided that z is not a


negative integer or zero; and conversely.
Prove also that the convergence of either series is uniform in a bounded
closed domain D, if it is uniform for the other, provided that
contains
none of the points 0, 1, 2,...
(Landatj.)

CHAPTER

THE HYPERGEOMETRIC FUNCTIONS


10.1.

Homogeneous

Before we

linear differential equations

discuss in detail the properties of several im-

portant analytic functions denned by differential equations, it


is desirable to consider whether there exists an analytic function
w(z) which satisfies the homogeneous linear differential equation

dnw

.& n

-x

.dw

where the coefficients p-^z), p 2 (z),..., p n (z) are analytic functions


whose only singularities of finite affix are poles, and, further,
if such a solution does exist, to ask what effect the singularities
of the coefficients have on the nature of the solution.
The simplest equation of this type
d

+p(z)w

presents
solution

On

little difficulty.
is

Its variables are separable,

= exp
i

and the

p(z) dz\.

the other hand, the equation of order two

^+p(Z

Tz

+q(z)w=0

has no such simple solution. We restrict our attention to this


equation for two reasons; firstly, the analysis in this case is
easily extended to the more general case, and, secondly, the
particular functions with which we deal in the sequel do, in
fact, satisfy

point z

second-order equations.
is

said to be

an ordinary point of

this differential

the functions p(z) and q(z) are regular in a neighbourhood of z all other points are called singular points of the

equation

if

differential equation.

10.11.

The

solution near an ordinary point

We now show that if z G is an ordinary point of the equation


d

dii)

in)

+piz)

Tz

+q{z)w==0

THE HYPERGEOMETRIC FUNCTIONS

234

if a Q and a x are two arbitrary constants, there exists a unique


function w(z) which is regular and satisfies the differential equation
in a certain neighbourhood of z and which also satisfies the initial
conditions w(z ) = a w'(z ) = a v This theorem, which is due
to Fuchs,f shows that the only possible singularities of the

and

function defined

by the

the coefficients p{z) and

differential equation are the poles of


q(z).

For simplicity we suppose that z is zero.| Then since p(z)


and q(z) are regular in a neighbourhood \z\ < R of the origin,
they are expansible as Taylor

series of the

form

CO

00

the radius of convergence of each series being not less than R.


find a formal solution by substituting

We now try to

=a

r a 1 z-\-a 2 z

-J

-\-...

in the equation

d2 w

\p

dw

-st

and equating coefficients. This gives

2a = a p +a q
2.3a = 2a # +a +a
x

1 ;p 1

1 g'

+a

g1

and, generally,

(nl)nan

(nl)an _ 1 p +(n2)an _ 2 p 1 +...+a 1 p n _ 2 +

+ an-2Qo + an-3.Ql+- + a

a n-3+ a On-2-

These equations determine the coefficients an successively as


linear combinations of a and a x
We next show that this power series which satisfies the differential equation formally and also satisfies the given initial
conditions, has a radius of convergence which is not less than R.
The proof of this is rather difficult, since we know very little
about the coefficients pr and qr
and
the maximum values, necessarily
Let us denote by
.

f Journal fiir Math. 66 (1866), 121. See also Forsyth, Theory of Differential
Equations, 4 (1902), Chap. I, and Bromwich, Infinite Series (1926), 162.
} This involves no loss of generality. For, if z a is not zero, we make the
transformation z'
zzn

THE HYFERGEOMETRIC FUNCTIONS


finite,

of

\p(z)\

and

on the

\q(z)\

circle

\z\

235

where

r,

<

R.

Then, by Cauchy's inequality, we have


\p n

and so

\p n

< Mir",

\qn

< K/r",

\qn

< N/r\
< K\r^\

M and Nr.

where K is the greater of


Writing b and 6 X for |a

2K|
and

so

and
x

we have

respectively,

< ^M+^oltfol < b K+b


\a
<b
26 = 2b K+b
2

where

\a x

K\r

<

K+b

2b 1

KJr,

K/r.

Similarly,

< 2|o
|+6
H-6 1 |+6
< 2b K+2b Kr- +b Kr~
< 3b K+2b Kr- +b Kr\a
and so
^6
2.36 = 36 Z+26 ^r- +6 ^r"
where
Proceeding in this way we find that \an < b n where
(n-l)nb n = nb n _ K+(n-l)b n _ Kr-i+...+b Kr- n +K
2.3|o,|

2 ||i>

>

|j 1

1 |g

From

this equation

we

see that the coefficients b n are con-

nected by the recurrence formula

{n-l)nb n -{n-2){n-\)b n _ 1 r^

But
as

n-2

K-x

nr

this gives

->-

2 b n zn

oo,
is

= wh n _

K.

n-

so that the radius of convergence of the power series


6 m it follows by the comr. Since, however, \a n
\

<

parison test that the radius of convergence of

2 an zn cannot be

than r; moreover, as r was any number


shows that 2 an z converges when \z < R.

less

less

than R, this

The function

w(z)

=^?an z n
o

is,

\z\ < R and satisfies the prescribed


The formal processes of term-by -term
multiplication, and rearrangement of power

therefore, regular

when

conditions at the origin.


differentiation,

THE HYPERGEOMETRIC FUNCTIONS

236

by which this function was made

to satisfy the differential


equation are now seen to be completely justified, since all the
series involved converge uniformly and absolutely in every
closed domain within \z\
R. This completes the proof of the
theorem.
Since a n is a linear function of a and a x we can express the
series

solution

we have just found in the form w(z)

The function

(z)

=a w

(z)-\-a 1

w1 (z).

a solution of the differential equation

is

which satisfies the initial conditions w (0) = 1, Wq(0) = 0,


whereas w-^z) is a solution satisfying the conditions w^O)
0,
Wi(0) = 1. Every solution of the differential equation regular
in the neighbourhood of the origin is, therefore, a linear combination of the solutions w (z) and w x (z), which we call a fundamental pair of solutions. Obviously w (z) and wx (z) are linearly
independent; by this we mean that there is no linear combination of them which is identically zero.
So far, the functions w (z) and w^z) are defined only in a
neighbourhood of the origin. When we continue these functions
analytically, they remain linearly independent solutions of the
differential equation. The continuation can be carried out along
any path which does not pass through a singular point of the
differential equation, and so the solution will ultimately be
defined all over the z-plane. In the next section we show that

is not one-valued, by proving that a singularity


of a differential equation is, in general, a branch-point of its

the continuation

complete solution.

Example 1 Prove that the necessary and sufficient condition that


the two functions /(z) and g(z) should be linearly independent is that the
determinant
.

A(/,|/)

should not vanish.

two

g(z)\

/(*)

(This determinant

is

called the

Wronskian of the

functions.)

Example

2.

Show

w"-\-p(z)w'+q(z)w

0,

that, if

^A(w
so that

Deduce

&.(w
that, if

(z)

(z)

and w^z) are two

solutions of

then

,w 1 )+p(z)Mw ,w 1 )

,w x

and

= CexpJ

w>i(z)

become

0,

p(z) dz

Q (z)

and

x (z)

after analytical

THE HYPERGEOMETRIC FUNCTIONS

237

continuation round a closed curve T,

A(

where

10.12.

is

the

sum

W W

t)

A(w wjer***,
,

of the residues of p{z) at

The nature

its

poles within T.

of the solution near a regular singu-

larity

a singularity of the differential equation


if it is a pole of one or both of the
w" +p(z)w' +q(z)w
functions p(z) and q(z). We call it a regular singularity if it
is not a singularity of either of the functions (z z )p(z) and

The point

is

(zz fq(z); otherwise it is called an irregular singularity.


The reason for this distinction is simply explained. In a
neighbourhood of a regular singularity, the differential equation
possesses, as we shall shortly prove, two linearly independent
solutions which are regular save possibly for a branch-point at
the singularity. But near an irregular singularity, the method of
solution by series breaks down and the singularity of the complete
solution is of a much more complicated character. A general
investigation of the behaviour of the complete solution near an
irregular singularity is beyond the scope of the present book.f
If the origin

is

a regular singularity of the differential equa-

tion under consideration, the functions zp(z) and z\{z) are


of the origin and so possess
regular in a neighbourhood \z
|

<R

convergent Taylor expansions of the form


CO

00

WW = J,Pr

ZT

Z2 2(Z)

>

= 2 ?r *>

where the coefficients p q and q x are not all zero. We now


show that, in general, the equation possesses two linearly independent solutions of the form
,

CO

w(z)

= z a '2,a

zr ,

a root of a certain quadratic equation. When we


power series in the differential equation and
equate coefficients, we find that this expression is a formal
solution of the equation if a and the coefficients ar satisfy the

where a

is

substitute these

conditions

ev a
a
o* ( )
>

n
>

t See Forsyth, Theory of Differential Equations,


Differential Equations (1927), 417-37.

(1902), or Ince, Ordinary

THE HYPERGEOMETRIC FUNCTIONS

238

re-i

and

an F(ac+n)

= - a {(<x+s)Pn _ +qn _
s=o
s

s}

>

1),

where

jp'(a) denotes the quadratic a(a lj+jjoa+g,,.


The first equation is satisfied by choosing a arbitrarily and
making a a root of the quadratic equation F(a) = 0. This equation is called the indicial equation and its roots the exponents
of the regular singularity under consideration. The remaining

equations determine successively the coefficients as as constant


provided that F{a+n) does not vanish for any

multiples of a

positive integral value of n. Hence, if the indicial equation has


distinct roots which do not differ by an integer, this process

two formal

gives

solutions,

one corresponding to each root of

the indicial equation.

however, the roots of the indicial equation are equal or


by an integer, we may obtain only one formal solution. We
leave this case for the moment as it presents certain difficulties
which do not occur in the general case.
If,

differ

10.13.

The convergence

of the series solution near a

regular singularity

We

have just seen that the

differential equation

w"-{-p(z)w'+q(z)w

=
00

always has one formal solution

= z a 2

u valid near the

regular singularity at the origin,

when the
To prove

and that

it

has two such solu-

an integer or
that this formal series does represent a solution
of the equation, we have to show either that the series }T an z n
terminates or else that it has a non-zero radius of convergence.
Let us suppose that the series does not terminate. We show
that if zp(z) and z2 q(z) are regular when \z\
B, the radius of
tions
zero.

difference of the exponents is not

<

convergence of ^,an zn is not less than B. The proof is very


similar to that of 10.11.
If a is the other root of the indicial equation, the coefficients
an are given by
-i
n(n+oc-oc')an
as{(oc+s}pn _s +qn _s}.

Let us write bn

\an

=J
s=o
when < n < 8, where S =

|a a' and
J

THE HYPERGEOMETMC FUNCTIONS


T

239

m is the least integer greater than 8, we have


m(m 8)|a,| ^ |m(m+a m

Then

|a|.

if

ot.')a

ImX
2 {(a+s)# m -s +gw
s

'

<2
But
of
r

M and N denote the maximum values, necessarily

if

\zp(z)\

<

and

\z

q(z)\

on the

respectively

circle

\z\

r,

finite,

where

R, Cauchy's inequality gives


\p n

and

b s{(s+r)\p m -s \+\q m -s\}-

so

where

[p n

< Jf/i*
< Kir*,

\q n
\q n

< N/r*
< tf/r,

K is the greater of M and JV. Substituting these bounds


and n we find that \am < b m where

for \p n

|g

|,

m(w-S)6m

(s+r+l)bs/rm -*.

=^ 2
=
s

Similarly

we can show that


n(n~S)6 n

From

this equation,

\a n

-K

< 6m when n^m,

f (s+r+l)b

8=0

we

/r

where

n ~\

see that the coefficients b n satisfy the

recurrence formula

n(n-8)b n -(n-l)(n-l-8)b n ^/r

But

= K(n+r)b n _

x lr.

this gives

bn

The radius of convergence of the


Since, however,

n{n8)r

Mm (6 n /6 m _1 =

and so

\a n

K(n+r)
n{n 8)r'

(n-l)(-l-8)

b n -i

<

series

6 m it follows
,

1/r.

2 b n zn

is,

therefore,

by the comparison

r.

test

that the radius of convergence of }T an z n cannot be less than r;


moreover, as r was any number less than R, this implies that
this series is convergent

when

\z\

<. B.

^an zn

za
The formal processes which made w
justified
since
completely
are now seen to be

all

& solution

the series in-

THE HYPERGEOMETRIC FUNCTIONS

240

volved converge uniformly and absolutely in every closed


domain within \z\
B.

When the exponent-difference is not an integer or zero, we


derive in a similar manner a second independent solution
w z a ']T a'n z n corresponding to the other root of the indicial
equation. In this case at least one of the solutions has a branchpoint at the origin.

10.14. Solutions valid for large values of

To

discuss the nature of the solution in the neighbourhood

we make

of the point at infinity,

The

\z\

the transformation z

Ijt.

differential equation

d2w

.dw

M +p{z)

dz-

+ q{z)w

then becomes

d*w

(2

/1\]

dw

/1\

[j-kijjYi+T^
The behaviour of the solution for large values of \z\ is now
determined by solving the transformed equation in the neighbourhood of the

origin.

Accordingly we say that the point at infinity


point if
2

/1\

i- p [l)'
are regular at the origin, that

an ordinary

q \t

is, if

2zz2p(z),
are regular at infinity.

is

z 4 q(z)

The complete

solution of the equation,

valid in a neighbourhood of the point at infinity,

is

in this case

of the form

where a

a are arbitrary constants.


Again, the point t
is a regular singularity of the trans,

formed equation

if

-tWt)'

-^1\~f\

are regular there;

we

say,

therefore, that the point at infinity is a regular singularity if


zp(z)

and

z 2 q(z) are regular there.

In

this case, p(z)

and

q(z)

are

THE HYPERGEOMETBIC FUNCTIONS


expansible,

241

Laurent's theorem, in series of the form

by

(Jg )

= 2 + a + ft+...,
23

a2

M = +++
>

i? of the point at infinity.


convergent in a neighbourhood |z|
It may now be shown, just as in 10.13, that there exist in
this neighbourhood two linearly independent solutions

= z- a ^an z~ n

w
where a and

a!

= z~ a 2 K z~ n
'

are the roots of the indicial equation

provided that these roots do not differ by an integer or zero.

The

solution
integer or zero

10.15.

When a a'

when

the exponent -difference

is

an

where s is a positive integer or zero, the


For if s = 0, the two solutions become
whilst
s
identical,
if
> 0, all the coefficients in one of the solutions from some point onwards are either infinite or indeterminate. It is, however, well knownf that a knowledge of one
solution of a linear differential equation of order n enables us
to depress the order to n 1. In our case, we obtain in this
way a linear equation of the first order which can be integrated
solution of 10.13

8,

fails.

immediately.

To

effect this depression of order,

we make,

according to the

usual rule, the change of independent variable

=w

(z)v,

where w (z) is the known solution of exponent


v is found to satisfy the equation
dz 2

whose solution
v{z)

\w

a.

The function

dz

is

= A+B

t See, for example, Forsyth, Treatise on Differential Equations (1914), 130-3.


4111

THE HYPEKGEOMETBIC FUNCTIONS

242

where

and

are arbitrary constants.

second solution, valid near the origin,


"

iv(z)

Now

=w

Hence the required

is
z

a and cxs are the roots of the indicial equation

so that

w^

exv

l-}-s

2a.

Hence we have

{~i p{z)dz)
+a lZ +...)2

z*(a

exPi

"
J

-Px-V^--\

dz\

= K+C+...)^ exp {- J <*!+**+-) *}


= z-t-ogiz),
where ^(0) = 1/ag. Since a ^ 0, the function (ao-fa^H-...)is

regular in a neighbourhood of the origin.

regular there
g(z)

Hence

and can be expanded as a convergent Taylor

= 2 gn zn

Substituting this series for

second solution

w=w

we

g(z),

g(z) is also

series

find that the

is

(z)

\
J

z 1 -*

y gn zn dz
o

= w (z)[y^^ + g
\/i ns
g

y ^l
ns

logz+

Z-i

m=0

n.=8+l

In particular, when the exponent-difference

).
j

s is zero, this

solution can be written in the form

=gw

w
As g

CO

{z)\ogz+zi + 1

2 b n zn

not zero, this solution possesses a logarithmic branchWhen the exponent-difference s is a positive
integer, the second solution takes the form
is

point at the origin.

w
If

it

happens, as

=gw

may

{z)\ogz+z a

'

2 cn z n

be the case, that gs

is

solution does not involve a logarithmic term.

zero, the

second

THE HYPERGEOMETRIC FUNCTIONS

An

method of determining the second solution


is an integer or zero is due to

alternative

when the

243

exponent-difference

Frobenius.f

Example

1.

Determine two linearly independent solutions of the

equation
valid

(i)

z*(z+l)w"-z*w' + (3z+ \)w

near z

(ii)

0,

near

1.

(Fobsyth.)

Example 2. Determine two linearly independent solutions of Bessel's


e<luation

zw" + w' + zw

valid near the origin.

Example

3.

Prove that the equation


z<*w"

+ (z+l)w'+w =

has an irregular singularity at the origin, and show that the method of
solution by series gives only one integral of the equation valid near
the origin.

Example

4.

Show

that the equation


z s w"

+ z w'-\-w =
2

has an irregular singularity at the origin, and that


find a series solution valid near the origin.

10.2.

The second -order

it is

impossible to

differential equation with three

regular singularities
If the only singularities of the differential equation

w" J[-p(z)w' -\-q(z)w


are regular singularities at
P(z)
Q(Z)

=
=

f,

tj,

and

, the functions

-)(zn )(z-t,)<p{z),

(z-^z-rjfiz-lMz)

are integral functions. Moreover, since the point at infinity is


not a singularity of the differential equation, the functions

2zz2p(z) and z 4q(z) are regular at infinity. This is the case if


and only if P(z) and Q(z) are quadratics in z and the coefficient
of z 2 in P(z)

is 2.

Accordingly we can write


p(z)

= A

z-C zn

C
z-tJ

t Journal fur Math. 76 (1873), 214-24. See also Forsyth, Treatise on


Differential Equations (1914), 24358; Ince, Ordinary Differential Equations
(1927), 396-403.

THE HYPERGEOMETRIC FUNCTIONS

244

where

A+B+C =

2,

and

= D

(z-i)(z- v )(z-Og(z)

F
zV

z7]

here the capital letters denote constants depending on the


exponents of the singularities.

The

indicial equation relating to the singularity


f

the form

If the exponents at are- a and

A=
y

17

and
Moreover,

0.

this gives

(_,)(_)'.

are

ft,

and those at

ft'

are

E = {-n-Oiv-iW,

i-ft-ft',

C=l-y-/,
since .4 + jB+C ==

trary, but are connected

J>
2,

by the

(J-^-^Jyy'.

the six exponents are not arbirelation

+ a'+ft+ft'+y+y' =

1.

differential equation is therefore of the

d2 w
dz*

takes

we have

B=

The

<x' ,

D=

1-a-oc',

Similarly if the exponents at


y,

]c(k-l)+Ak+D/{(Z- v )(Z-Q}

now

a<xoc'
[

z-i

1 ft

ft'

z~^T^

'

form

lyy'\dw
"i=jfe"~
yy

(z-$){-q-t.)

'

x
*

(z-QGn)
(t-y)(v-ZM-J) w

{z-v)(-)

(z-i)(z-rj)( Z

-0

In the notation introduced by Riemann, we express the fact


wisa solution of this differential equation by writing

that

Pla.

W
Example

ft

z\.

ft'

linear differential equation of the second order

but one singularity, a regular singularity at the


equation is zw"-\-2w' = 0.

Example

2.

origin.

Show

has

that the

linear differential equation of the second order has

These are a regular singularity of exponents tx and


the origin and a regular singularity of exponents jS and ft' at infinity.
Find the equation and show that a+a'+ft+ft'
0, oca.'
ftft'.

but two

singularities.

a' at

THE HYPEKGEOMETRIC FUNCTIONS


Example

3.

Show

245

that the differential equation satisfied

function

by the

w=

co

Plcc

W p

y'

is

dtw
(

1 g '

y y'\ dw

+1

Example

4.

Prove,

by transforming the

oo

z*(l-z)ep a

)3

0'

y'

=p

differential equation, that


co

a+p

j3-p- 9
la'+p P'p q

-I

y+<?

y'+q

also that
oo

'

[0

Plcc

U'

Example

(-co

=P

U'

,8

JS'

y'

1/(1-2).
J

Prove that

5.

l^JU^KK
U
1

and

+^+-i=r")(i=ij(i=) =

_^

(0

U'

and

z\

y'

= P\a+p

p- p - q

y+q

[a'+p

P'~p-q

y'

+q

also that
Plot.

P'

y'

=P

z
J

where z lt &, Vl , & are derived from


homographic transformation.

z,

p
p>

yy

v respectively by the same


,

The hyper geometric equation


we make the homographic transformation

10.21.
If

and use the

results of

Exx.

5 of the preceding section,

4,

we

find that
(

P\<*

y'

(0

[=P
J

[a'

oo

t)

y>

= t%l~t)yp\

[tx'

oo

ct+ji+y

A.

ot+fi'+y

y'y

"J

THE HYPERGEOMETRIC FUNCTIONS

246

in order to determine the properties of the solution of

Hence,

the second-order equation with three regular singularities,


suffices to discuss the behaviour of the function
oo

w=

Pl

[lc
which

satisfies

a
b

t\,

cab

the equation

t(l-t)w"+{c-{a+b+l)t}w'-abw
This

is

it

0.

called the differential equation of the hypergeometric

the hypergeometric equation.


the possible occurrence of
regarding
To avoid difficulties
of this equation, we
solution
complete
logarithmic terms in the
of the exponentnone
that
shall suppose in the present chapter

more

function, or,

differences

briefly,

c 1, ab, a+bc

is

an integer or zero.f

The generalized hypergeometric equation


we introduce the operator & = z djdz, we find that

10.22.
If

the

hypergeometric equation
z{

l- z) *E+{c-(a+b+l)z}^-ab,

takes the simple form

= z(&+a)(&+b)w,

&(&+cl)w
a fact which
equation by

is

of importance in the formal solution of the

series.

be of generalized hypercan be written in the form

differential equation is said to

geometric type if

it

= z{&+oc
the order of this equation
10.3.

To

is

){&+oc 2 )...(&+ocp )w;

equal to the greater of p and q+1.

The hypergeometric function


discuss the nature of the solution of the hypergeometric

equation

&{&+c-l)w

= z(&+a){&+b)w

The case of zero or integer exponent-differences has been discussed by


-f
Proc.
Iandelof, Acta Soc. Scient. Fennicae, 19 (1893), No. 1, and W. L. Ferrar,
Edinburgh Math. Soc. (1), 43 (1925), 39-47.

THE HYPERGEOMETRIC FUNCTIONS


247
near the regular singularity at the origin, we substitute formally
the series

= z 2 a 2
o

and equate

We

coefficients.

find that the exponent

oc

satisfies

the indicial equation

a(a+C 1)
and that the

an

coefficients

(<x+n){cx+cl+n)an

From

where

satisfy the recurrence

formula

= a + a +nl)(ot+b+nl)an _ v
(

the recurrence formula,

an

we deduce

that

r(oc+a+n)r(oc+b+n)
r(a +l-\-n)r(a+c+n)

'

an arbitrary constant, and so we obtain two formal

is

solutions

y r(a+n)T(b+n)
r(c+n)n\
z

Z-,

gl _ c
'

^ r(a-c+l+)r(6-c+l-fn)
Z
r(2-c+n)n\

Zn

'

But

since the two power series converge absolutely and uniformly with respect to z in every closed domain within the unit
circle,

the validity of these solutions

when

\z\

<

follows im-

mediately.

We now

define the hypergeometric function F(a,b;c;z) by the

equation

T(a)V(b)

r{a+n)V{b + n)
V(c+n)
71=0

F{a,b;c;z) = y
FW~^^'
^'~Z
V

when

<

'

l
z

n\

and by analytical continuation when \z\ > 1; it is an


z, which is certainly regularf when \z\ < 1.
In terms of this new function, the two solutions of the hyper\z\

1,

analytic function of

geometric equation valid near the regular singularity at the


origin are

F{a,

Example

zi-F{l+a-c, l+b-c; 2-c; z).

b; c; z),

1.

Show

that

(l-z)-
\o

-~ =
gT
6

=
=

.F( a ,j8;jS;z),

zF(l,l;2;z),

limWa,fi;S;-).

We shall see later that it has branch-points at

and

oo.

THE HYPERGEOMETRIC FUNCTIONS

248

Example

2. Prove that

^-F(a,b;c;z)=
dz

Example

3.

Show

that,

c F(a+l,b + l;c+l;z).

when

the hypergeometric equation, valid

two independent

1,

when

\z\

<

1,

solutions of

are F(a,b; l;z) and

00

T(a)T(b)F(a,b;Uz)logz+

*'

^T

J^'r
l

where

sT

^ (j+Z=I + b+^1 ~ n}
1

10.31.

The

An

integral representation of F(a,b;c;z)

series

the hypergeometric function can be

defining

written in the form

Effl

;e;z)

r^=

5j^r(a+JB(6+,c-6)5.

Using the Eulerian integral of the

when Rlc

first

kind,

we deduce

that,

> R16 > 0,

r(c)
i

it

being understood that the

sign of integration are

made

arg

many -valued
definite

= arg(l =
t)

functions under the

by taking
0.

Y{a+n)zntnjn\ converges unithe series


formly with respect to t over the whole range of integration,
and so we can invert the order of integration and summation

Now when

\z\

<

1,

to obtain

!W)j(a,M;,) =
V

c)

Jf

*-i(l-r

<& *
f ^%
T(c

J<

6)

n\

T(a)
<ft,

r(c-6)

249

THE HYPERGEOMETRIC FUNCTIONS


where (l-zt)- a has its principal value.
that, when \z\ < 1 and Rlc > R1& > 0,

T^E^^l a,b;c;z)
F{
r (c)

We

have thus shown

= h- H l-tr-^(l-zt)-dt.
o

on the right-hand side of this equation is, howdomain of the z-plane,


ever, uniformly convergent in any closed
and so represupposed cut along the real axis from 1 to +oo,
plane. Hence this
sents an analytic function, regular in the cut

The

integral

the analytical
integral representation of F(a,b;c;z) provides
case when
the
in
function
hypergeometric
continuation of the

Rlc

> R16 >

If

0.

double-circuit integral for the

we had used Pochhammer's

we
Beta function (see 9.3) instead of the Eulerian integral,
F(a,b;c;z),
for
integral
double-circuit
should have obtained a
on the paravalid in the cut plane, without any restrictions
reader to
the
refer
we
this,
meters b and c. For the details of
Pochhammer's memoir already
10.32.

The value

of F(a,b;c;

Since T(a+n)ir(n)

nave

na

cited.
1)

when Rl(c-a-6)

when n

r(a+n)V(b+n)
r(c+) n\

is

large

> 0.

and positive^ we

1__
~ nc-o-b+r

Hence the hypergeometric series F(a,b;c;l) converges if and


only if Rl(c a b) > 0. When this condition is satisfied, it
follows from Abel's theorem on the continuity of power series $
that

F{a,b;c;l)

lim F{a,b;c;x).
s-*l-0

If,

in addition, the condition

deduce from

Rlc

> R16 >

is satisfied,

the integral formula of 10.31 that

r(c)
K

r(b)r(c--b)J
6

~
t See

r(c)r(c-o-6)
T{ca)r(c-b)'

9.22,

Ex.

4.

t 5- 22 >

Ex

we

THE HYPERGEOMETRIC FUNCTIONS

250

The

additional restrictions on the real


parts of b
however, quite unnecessary for the truth of
this

and c are
and only

result,

arise on account of the particular


method of proof adopted.
Although the restrictions can be removed by an

principle of analytical continuation,


proof, as follows.

If

we

it is

appeal to the
simpler to give a direct

write

F(a,b;c;z)

= A n z",

F(a,b;c+l;z)

= f Bn z n

it is

easily verified that

c{c~a-b)A n
from which

it

e(e-a-b)

But

since

(c~a)(c-b)Bn +cnA n ~c(n+l)A n+1

follows that

I^ n =

(c-a)(c-b)

(N+1)A N+1 ~ l/Nc - a -,

| Bn -c(N+l)A N+1

this equation gives

c(cab)
provided that Rl(c o 6)

Y{c~a)Y{c~b)

_,

>

'

From this we easily deduce that

0.

rW^=T) F(a b;c;1)


'

V(c+ma)V(c+mb)
L

for every positive integral value of

r(c)rc-- F)
Let us now write

F{a b;C
'

'

F(fl,b;c+m;z)
Then, since
integer

Cn

1}

<

'

=l+fcn zn

= Rl(c a b).

1).

larg6j

we can

when n

> N,

is

that,

<- _JL _J_


nK+m+l ^ jym nK+l'
Hence we have

Ttf+1

'

m, and hence

~ l/e-a-++i when n
1

where k

= ]F(a,b;c+ m

N, independent of m, such
\Cn

_,

(c-\-m)r(c+m a b)

find

an

251

THE HYPERGEOMETRIC FUNCTIONS


where

Cn

is

of n,
independent of m. Since, for every fixed value

asm-co,

tends to zero

also tends to zero

this inequality implies that ]

Cn

and hence that


\imF(a,b;c+m;l)

1.

m> ro
This completes the proof that the equation

F ^ b

,,

is

true

when

'

c'

1)=

the real part of

r(c)r(c o 6)

r(c-a)rJ^b)'

cab

is

positive.

analytical continuation of F(a,b;c;z)


1 * in the hypergeometric equation satisfied
If we put z
by F{a, b;c;z), we find that

The

10.33.

t(l-t)~+{(l+a+b-c)-(a+b+l)t}^-abw
which

also of hypergeometric type.

is

From

0,

this it follows that

the hypergeometric equation has two linearly independent solutions

F(a,b;l+a+bc;lz),
(lz)

c -a - b

F(ca,cb; 1+cab; 1z)

when |1 z\ < 1. We make the second solution definite


~ by giving to (1 z)c a h its principal value.
The region in which these two solutions are defined has an
valid

area in

was

common

with the region

defined. In this

common area,

\z\

<

1 in

which F{a,b;c;z)

the three solutions must be

connected by a linear relation


F{a,b;c;z)

= AF{a,b;l+a+bc;lz)+

To determine
Rl(a+6)

+B(lz)c - a - b F(ca,c~b;l+cab;lz).
the constants A and B, we suppose that

< Rlc <

1,

so that the three series

F(a,b;l+a+b-c;l),

F(a,b;c;l),

F(ca,cb; 1+cab;
are convergent.
If we make z tend to

1)

along the real axis,

F(a,b;c;l)

= A.

we

find that

252

Similarly,
1

By

THE HYPERGEOMETRIC FUNCTIONS


making z tend to zero, we have

= AF(a,b; l+a+b-c; l) + BF(c-a,c~b; 1+c-a-b;

1).

10.32, the first equation gives

r(c)r(c-o-6)

T(c a)T{c by

The second equation now becomes

r(c)r(i-c)r(c-a-6)r(i+o+6- C
r(c-o)r(i+o-a)Tv=6jF(i+6-c) +
)

+5 r(i- C )r(i+c-o-6)
F(I-6)r(i-o"j

from which

it is

not

difficult to

deducef that

r( C )r(a+6- C )

ir(6)

We have thus proved that the relation

+
,

V(c)T{a+b-c)
'

r(o)r(6)

when

holds

|z|

<

1,

^-^-"-"^-^-M+c-a-M-z)

|l-z|

<

El(o+6)

1,

provided that

<

Rlc

<

1.

This restriction on the parameters a, b, c is not necessary for


the truth of this result, and arises as a result of the method of
proof adopted. An alternative proof, valid for all values of the
parameters, will be found in Barnes's memoir cited below.
The importance of the result lies in the fact that it provides
the analytical continuation of F(a,b;c;z) into the region
1
2
1 and shows that this function has a branch -point at
I

<
=
1

1.

10.4. Barnes's contour integral for F(a,b;c;z)

Barnes,J following Pincherle and Mellin,|| has represented


the hypergeometric function by means of a contour integral
t

By

||

using the formula T(z)F(lz)


v eosec CT z.
London Math. Soc. (2), 6 (1908), 141-77;
Atti d. E. Accademia dei Lincei, Rendiconti (4), 4
(1888), 694-700, 792-9.
Acta Soc. Scient. Fennicae, 20 (1895), No. 7.

% Proc.

THE HYPERGEOMETKIC FUNCTIONS

Gamma

whose integrand involves


of Barnes's formula

253

A particular case

functions,

is

CT+001

ztti

sin tts

ct cot

<

<

when \z\
|arg( z)|
ir, the path of integration
1,
being the straight line Bis
a, where
1
a
0; this has
already been given as an exercise for the reader.f
Now if & denotes the operator z d/dz, we have formally

valid

JWff)
r(c)

^a

'' c ' z)

< <

r(+*)r(6+0)

-r^+#)nT+^)

+ + +

''

-)

j_ r>+#)r(6+fl)
r(c+0)r(l+p)

2w

--

conns

a cot

a+xi

r(a+ s )r(b+s)
r(c+ 5 )r(i+s)

sin?

( z) s ds

(7 OOl

2?r*

r(c+)

'

This

is Barnes's contour integral for F(a,b;c;z).


Let us consider, then, the integral
i

277*

in the case

r( a+S )r(6+,)r(-,)
r(c-f-s)

<

< tt e, where e is an
The path of integration is the
modified, if necessary, by loops to make the
lie to the right of the path and those of

when

\z\

1,

|arg( z)|

arbitrary positive number.

imaginary
poles of

axis,

r( s)

r(a+s)r(b+s)

to its left; this

neither a nor 6

is

is always possible, provided that


a negative integer.} We suppose, as before,

that none of the exponent-differences c 1, a b, a+bc is an


integer or zero.
This integral may be evaluated by Cauchy's theory of resi-

x
z, z by
s, in Ex. 18 on page 153.
t Replace e~ by
t The case when a or 6 is a negative integer is unimportant, since F(a, b;c;z)

then a polynomial.
Barnes's formula does, in fact, provide a solution of the hypergeometric
equation when this condition is not satisfied.

is

THE HYPERGEOMETKIC FUNCTIONS

254

dues. Let C denote the closed contour formed by three sides


of the rectangle whose vertices are iN, iV+i*iV, together
iN.
with the portion of the path of integration from iN to

We

take
to be an integer, greater than |Ima| and |Im6|, so
that no poles of the integrand lie on C. We then have

j_ rr(o+)r(H*)r(-)'-(z)*ds
(

2tt

r(c+s)

(z)
= sum of residues of r(a+s)T{b+s)T{-s)
r{c+s)

at the

poles within

N
2, r(c+n)r(l+n)*

'

and so

iN

r(a+s)r(b+s)r(-s)
T(c+s)

i_

2ni J

-iN

N r(a+n)T(b+n)

zn

T{c+n)

n=

:J,
2,-Tri

J denotes the integral along the remainder of C. We


now show that J tends to zero as N -> oo.
where

We find it

convenient to write
-iN + N + i N + h + iN

iN + N + i

+ N + i-iN
S

-iN

= J -\-J2J3,
1

iN

say,

and to consider Jv J2 J3 separately.


Now, when |s| is large and |args|
,

r(a+s)r(b+s)

<

77,

+tf[l+o(l)],

r(c+ S )r(i+s)
where

= a+bcl,

and

r(o+a)r(6+)
r(c+s)r(i+s)

<x-\-if3

Hence, in

Jx

</2 ,

so
|s|

a e-^ args [l+o(l)].

J3 we have
,

T(a+s)r(b+ S )
r(c+s)r(i+s)
where

is

a constant, independent of N.

shall

But

Jlt

in

THE HYPERGEOMETRIC FUNCTIONS


= oiN, so that, when JV is large,

255

ir( *)r(l+)|

Z) =
s

|(

<

z \ e Nem-z)

<

4^e- nN

Iz^ir-ON^

These inequalities give immediately

<

IJJ

iirAN ae- eN j \z\do

<

2{2N+l)nAN<*e-*N

since

In

and

<

|z|

infinity.

Ja

1; this

Jx

implies that

tends to zero as JV tends to

In a similar manner we prove that J tends to zero.


3
however, we have s = N+^+it, where
JV
t
JV,

< <

so

ir(-)r(i+)i
"

lain

and

|(_ z ) s

ff (JV

+*.+*)
\z\

i
|co S h7r|

<

2tte-^
n&

'

N+ie-tax&.-$.

It follows that
\J2

<

2 7r^JV a |z| iv'+*

e -nlU-ta.Tg(-z) dt

-JV

< 2TrAN a

\z\

N+i

N
j e-^dt

-JV

<
But

since

|z|

<

1,

4fnAN a \z\ N+*lc

J2

this implies that

tends to zero as JV tends

to infinity.

We have

thus shown that

JVi

n+)r(6+)r(-)

JL

2ni

r(c+s)

-Ni

z)

as

JV

Z
as JV->

r(c)

oo,

so that,f

when

\z\

<

27ri

J
ooi

r(c+n)

and |arg( z)|

r(c+)

n!

<

tt,

z)

t That the integral on the right-hand side actually converges and


a Cauchy principal value, will be easily proved by the reader.

as

is

not

THE HYEERGEOMETBIC FUNCTIONS

256

Finally,

we show

that the integral

ooi

r(a+s)V(b+s)r(-s)
r(c+ 5 )

ds

analytic function, regular in the z-plane supposed


to 00, and so provides the analytical
cut along the real axis from

represents

an

continuation of

all over this cut

r(a)T(b)

plane.

For in any closed domain of the out plane, there holds an


inequality |arg( z)\ < -n e, where e is a positive constant.
it where t is real and \t\ large, then
It easily follows that, if s

r(a+s)r(b+s)V(-s)

K is independent of z and

where

t.

Km-m

This implies that Barnes's

repreintegral converges uniformly in the closed domain and so


integral
the
of
value
the
As
function.
analytic
sents a regular
1,
has been shown to be r(a)T(b)F(a,b;c;z)/r(c) when \z\
immediately.
the result stated follows

<

The behaviour

10.41.

of F{a,b;c;z) near the point at

infinity
similar to that of 10.4, it can be
is a positive integer,
difficulty that, if

By an argument
without

much

shown

ooi

2iri

T(a+s)T{b+s\r{-: s)

J_

T(c+s)

_ z)8ds

coi
m+ ooi

j_
2ni

c
J

r(a+8)r(b+a)r(-s)
T(c+s)

yt , a

m oi

+ y (residue at s = an)+ J, (residue at s =

bn),

to
where p and q are integers, not exceeding m, which tend
integral
second
the
infinity with m. The path of integration in

obtained by translating that of the first integral a distance


new contour
to the left. If a or 6 is a positive integer, the

is

is

THE HYPERGEOMETRIC FUNCTIONS

257

indented so that the pole, which would otherwise


to its

lie

on

Now
v

= a n is

the residue at s

r{a+n)V{b-a-n)
r{l+n)r(c-a-n)

'

'

/_,)- I>+)r(l+ c+n) simr(a+nc) _ n


Z

r(l+n)r(l+a-6+) simr(a+n-b)
sin7r(a-c) _
r(o+)r(l+a c+n) _ n
Z
sin7r(a by
r(l+n)r(l+ab+n)
'

|arg( z)\

where, as usual,
s

it, lies

left.

= b n.

'

<

77.

similar result holds at

Hence we have

ir(&)

-F(a,b;c;z)I

Y(c)

sinTr(q-c)

y a \p

r(a+n)r(l+ac+n) z^

^ r(6+n)r(l+6-c+^) z-

sin7r(6-c)

r sin77(6-a)^

Z^

r(l+b a+n)

'

n\

where
-m+ct>i

J=

na+sWib+sWi-s)

2t

T(c+5)

2 OOZ

001

= -J- 2 2ni

r(a-m+ a )r(6-m+ S
f

ooi

z~ m J,

r(c-m+s)r(l-m+s)

simrs

'

say.

But when |arg( z)\ < tt e where e > 0, J is a bounded function of m and z; hence J ->
as m -> 00 provided that |z| > 1.
We have thus shown that, when z lies in the part of the cut
plane for which

\z\

>

1,

r(a)T(b)
F(a,b;c;z)
r(c)

r(a)F(b-a)
(-z)~ aF(a,
r(ca)
.

'

4111

l+o c; l+o 6;sg-1 )+

r(6)r(o-6)
(-z)-"F{b,
r(c-6)

1+6 c; 1+b-aiz-1

).

THE HYPERGEOMETRIC FUNCTIONS

258

This equation provides the analytical continuation of F(a, b; c; z)


outside its circle of convergence and shows that the function
so defined possesses a branch-point at infinity.
We have now proved that the function F(a, b;c;z) defined

by the power

series

T(a+n)T{b+n)
T(c+n)
V{a)V{b) 4,
n=0
<*

V(c)

zn

n\

<

1, and defined by analytical continuation when


a one-valued analytic function, regular in the whole
plane supposed cut along the real axis from 1 to +co.

when
\z\

\z\

1, is

The

10.5.

relations

between contiguous hypergeometric

functions

Let us denote by P(z) a solution of a second-order linear


with three regular singularities. If A and
of this differential equation, the funcexponents
any
two
are
fi
in P(z) by A+l
tion Pa+i, -i( z )> obtained by replacing A and
m
and i respectively, is said to be contiguous to P(z). Since
P(z) has two exponents at each of its singularities, there are
thirty functions contiguous to P(z). Biemannf proved that the
differential equation

jjl

ju.

function P{z) and any two functions contiguous to it are connected by a linear relation whose coefficients are polynomials
in 2. The recurrence formulae satisfied by the Legendre functions are particular cases of this general theorem.

Now if we transform

P(z) and

its

contiguous functions to the

hypergeometric form
1

00

1 c

cab

as in 10.21, we find that Riemann's theorem is a consequence


of an earlier one due to Gauss. According to Gauss, the function
F(a',b';c';z) is contiguous to F{a,b;c;z) if it is obtained by
increasing or decreasing one and only one of the parameters
a, b, c by unity. There are, then, six hypergeometric functions

denoted by

and these may be conveniently


Fb _, Fc+ Fc _ in an obvious notation.

F{a,b;c;z),

contiguous to

Fa+ Fa _, Fb+
,

(Math. Klasse),
t Abh. d. Ges. d. Wiss. zu Gottmgen

(1857), 1-24.

THE HYPERGEOMETRIC FUNCTIONS

259

Gaussf showed that, between F(a,b;c;z) and any two hypergeometric functions contiguous to it, there exists a linear relation
with polynomial coefficients. There will then be fifteen such linear
relations.

The simplest method of proof consists in determining the


required linear relation when \z\
1 from the series definition
of the hypergeometric function, and deducing the result in

general by analytical continuation.

<

For example,

,,rr(&)_.
~T7c)

r(c+n)

4*

y
Z

r(c+n)
{ aa

r(c)

Thus the equation


H
holds

when

10.6.

The

If

n\

r a + n + 1 W(b+n) z_

\z\

<

1,

n\

y r(a+n)r(b+n+l)
Z
r(c+n)

zn

n\

+~ bJ>b+},

XT
(a-b)F
,

,
= aFa+ -bF
b+

and hence generally.

generalized hypergeometric function

we attempt

to solve the generalized hypergeometric equa-

tion

&(&+Pi~ l)(&+p 2 - 1)...(&+ Ps - l)w

= Z(d+<x )($+a )...(&+*p )w


by a series of the form z? cn zn we find that
1

00

w=

yV

r (Y+<Xi+n)r(y+<x 2 +n)...r(y+(Xp +n)


F(Y+Pi+n)nr+P2+n)...r(y+pa +nj

Z
n\

provides a formal solution

when y is a root of the equation


y(y+ Pl -l)(y+p 2 -l)...(y+pa -l) = 0.

Apart from the case when the

series terminates and so represents a polynomial multiplied by zv, this formal series solution
will be valid only within its circle of convergence. Now the
t Ges. Werke, 3, 130.

THE HYPERGEOMETRIC FUNCTIONS

260

radius of convergence of the series

is 0,

1,

or oo according as

< q. p 1 > q, the formal solution breaks


down completely. When pl < q, we obtain q+1 solutions,

p1

>, =,

If

or

provided that none of the coefficients becomes infinite or indeterfunction,


minate. If pl
q, each solution is an integral
valid
all over
is
and
so
of
z,
power
a
multiplied, possibly, by

<

q, the series solutions are valid


the z-plane. But when pl
must be defined outside this
1
and
only within the circle \z\
circle by analytical continuation.
The solution of exponent zero at the origin is usually denoted

by pFq (<x 1 ,a2 ,...,<xp

Pl ,p 2 ,... >Pa ;z),

where

r(pi)r(p,)...r(p)

r( 1 +)r( +n)...r( p +n) *


r( Pl +n)r(pz+n)...r( Pq +n) n!"
ls

2=

In

the ordinary hypergeometric function

notation,

this

is

2 Ji(a,6;c;z).

In dealing with asymptotic expansions we shall find it conq, even though the
venient to use this notation when p 1

>

series diverges.

10.61.

The

function ^(ajpjz)

The function

F {ol;p\z)

satisfies

&[&+piyw
If

we

= {&+a)w.

write this in the form


z

we

the differential equation

see that it has

- z)
w
Tz-
l* +{p
an

>

irregular singularity at infinity

and 1p

regular singularity of exponents

at the origin.

and a

When

solutions of this
p is not an integer, two linearly independent
equation are ^(a-.pjz) and z^^F^a p+l;2 p\z)\ these are

valid

over the z-plane, since


easily shown that

all

It is

!*!(; p;z)

FX is

an integral function.

= VmF{*,P;p;zlP),
J3-J-CO

so that the irregular singularity at infinity arises from the 'confluence' of the singularities at j3 and infinity of the equation

satisfied

THE HYPERGEOMETRIC FUNCTIONS


Many of the properties

by

F(cc,l3;p;z/fS).

261

of ^(a;^;^),

which is usually

called a confluent hypergeometric function, can


be obtained by this limiting process, though a rigorous proof is
often difficult. But since a direct proof by the methods of this
chapter is generally quite straightforward, the most important
formulae and certain deductions from them are given as exercises for the reader.

Example

Prove that, when Rla

1.

>

and Bi(p-a)

>

0,

TMTip-ot^F^ocp-.z)

T(p) (

e^-^l-t)"-"- 1

dt,

where

*" _1

and (1 ty-"- 1 have

Example 2. Prove,
or otherwise, that

by a transformation

x Ji(a;p;2)

provided that p

their principal values.

of the differential equation

e^-F^p a;p; z),

not a negative integer or zero.f (Rummer's}

is

first

transformation.

Example

Show that

3.

the function

dw
W
dz-*+Pdz-

d?w
Z

w=

^(piz)

satisfies

the equa-

Hence show that


tF^at;

provided that

2ot is

2<x;

2z)

e\Fx (<x+h hz\

not a negative integer or zero. (Kummer's second

transformation.)

Example 4. Prove that,


integer or zero,

when

|arg( z)\

<

\n and a. is not a negative

wi

ooi
where the path of integration
sary,

by loops

ones to

to

make

its right.

Example

5.

the imaginary axis, modified, if necesthe negative poles lie to its left and the positive

(Baknes.)

Show

that,

^
Deduce Kummer's

is

first

when

- Si

|argz|

In proving

T(n s)z s

transformation.

9.55.

4.42.

this result, it is

77,

ds.

f For a discussion of the case when

Bessel Functions (1922),

<

(Babnes.)

a negative integer, see Watson,


J Journal fur Math. 15 (1836), 138-41.
necessary to use the asymptotic formula of
p is

THE HYPERGEOMETRIC FUNCTIONS

262

Example

Prove that, when

6.

<

|args|

fn-

and p

is

not an integer,

Wfr

T(- S )T(l-p-s)T{*+s)z ds
J

T(l-p)T(ot) 1 F1 (ot; P ;z) + T{p-l)T(l+ot-p)z^-i' l F1 (l + a-p;2-p;z)

its right

and

exists. The contour is the imaginary


loops to make the positive poles lie to
the negative poles to its left. (Barnes.)

10.62.

An

asymptotic expansion

provided that the right-hand side


axis, modified, if necessary,

It is possible to obtain

when

\z\

is

large

and Rlz

by

an asymptotic expansion of -^(c^pjz)


by applying the method of 10.41

<

to the integral

r() rw ...,,_
;2
>
r(p)

J^
f

r(+).

r (-^- z

and then deducing the corresponding expansion when Rlz >


by the aid of Kummer's first transformation. If, however, we
use the result of Ex. 6 of the previous section, namely that,

when

|argz|

77 e

<

fw,

cot

2ttI
TTl

r(-s)r(i- P -s)r(oc+s)zs ds

coi

r(l-p)r(oc) 1F1 (oc; P ;z)+

+r(p-l)r(l+<x- P y-P1F1 (l + c-p;2-p;z),


we can determine the required asymptotic expansion over a
much wider range of values of arg z.
Let us consider,

)"

in

then, the integral

r (-s)r(i-p-)r( a +s)z da,

C is the rectangle with vertices at iM, K-{-iM,


KiN, iN, modified in the following manner. In the first

where

place, the right-hand side of the contour


if necessary, so
its right

is

provided with loops,

that the positive poles of the integrand

and the negative ones to

its left.

Secondly,

lie

to

we take

| The analysis which follows is a slightly modified form of the work of


Barnes, Cambridge Phil. Trans. 20 (1904-8), 259-61.

THE HYPERGEOMETRIC FUNCTIONS

K Rla

263

= m,

a positive integer, and indent the left-hand side


the
pole
a lies within the contour.
so that
By Cauchy's theory of residues, we have

sum

^ r(+)r(i +-/>+) /_

= z_ a
If

a n within C

of residues of integrand at poles

we can show

i\"

that the integrals along the sides parallel to the


tend to infinity, it will
and

real axis tend to zero as

follow that

.
2m

T(-s)r{lps)r(<x+s)zs ds

001

= z_ x

y r(+)r(i+-/+) /_i\
+

-K+cci

r(-a)r{l-p-a)r(+a)* da.

K<ni

Now on the upper side of the rectangle G, we have s =

where
^ a ^ 0.
formula of 9.55 that,
then

K
if

being kept fixed,

is

and

large

it

s lies

follows

on

o-\-iM

from the

this side of C,

\r(- S)r(i-p-s)r(+8)z\

(2 7r )

<

(277)3^2-71- Im(p +a )/2 e -eMJfRl(a-p-l/2) Z


/M\,

since ]argz|

3/2

e- ,rIm(/ + a " 2 c- M(arg;S + 3 'r

JfE1(a -P- 1

' 2)

'2

'|z/if I"

< fw e <

fir.

Hence the absolute value of the

integral along the upper side of

does not exceed


o

AM m(cc-p-ime -m

f Z/M\d<r,
\

-K
where

M
M -=

independent of

and

As

this expression
does the integral
along the upper side of C. In a similar way it can be shown
that the integral along the lower side of C tends to zero as
2V->oo.
is

evidently tends to zero as

z.

oo, so also

THE HYPERGEOMETRIC FUNCTIONS

264

We have
i,

thus proved that, when |argz|

<

\tt,

r(-8)r(i-p-8)r(*+a)* ds

2m J
-coi

a -p+n)
V n+n)r(l+
+
Z
nl(z)
n\(-z)
n=0

where

'

'
'

a m+coii

J=*2m

r(-s)r(i- P -s)r(<x+s)z*ds

ocmcci

r(oc+ms)r(l+ocp+ms)r(sm)za ds

ooi

when

0(1)

\z\

is large.

Hence we

see that

T(cc)r(l-p) 1F1 (a; P ;z)+

+ r(l+*-p)r(p-l)zi-P F (l+ a -p;2- P ;z)


1

~ r(a)r(l+-p)a- F L + a-p; -IJ,


l

the series on the right-hand side being an asymptotic expan-

when \z is large and |argz| < \tt; for the


by terminating the series at the mth term is
sion valid

^ m*+m]r(\+c
(

L- p+m)
m
m!
l)

which

is

10.63.
If

we

of the order of the

error caused

''

term omitted.

first

The asymptotic expansion

of jF^aipiz)

write

= P,
r(l +*-p)T(p- 1)^(1 + a -p; 2-p; z) =
r( a )r(l+a- P ^
1+oc-p; -Z' =
r(<x)r(l- P ) 1F1 (oc;p;z)

)2

(a,

the result of the previous section

P+z>-pQ
for large values of

\z\.

is

when

|argz|

this holds

when

that,

~ z-i?

In particular,
J7T

Q,

JR,

< argz ^

77.

<

tt,

THE HYPERGEOMETRIC FUNCTIONS


265
But when \tt < argz < ir, we have fir < SkTg(ze~ 2ni < n,
and so this asymptotic formula holds when z is replaced by
)

ze-z. This gives


P-f- z i-p e -2(i-/>)Q

since P, Q,

B are unaltered.
Psin7rp

In a similar manner

ia

R,

Ehminating Q, we find that

e7r<xiz-<x Rs,m.TT{p a).

can be shown that, when

it

77

< argz <

\it,

~ e-ntiz-tEsmirip a)

Psimrp
for large values of

~ z- e^

These two results can be combined into

\z\.

the single formula

Psinvp
valid whenRlz

<

0,

~ {z)- aB sin tt(pol)

provided that(z)- a hasits principal value.

We have thus proved that, when Rlz <


1

F1 (oc;p;z) ~ *M_(_ z )-^( a


I

(p a)

0,

1+oc-p; -z-i).

From this we deduce, by Kummer's first transformation


2), that when Rlz > 0,

10.61,

Ex.

xJi(;

provided that

P ;s)

zfi-*

~ J^-^ip-oc, l-;-i),

has

its

principal value.

REFERENCES
Linear differential equations:
A. R. Forsyth, Theory of Differential Equations, 4 (Cambridge, 1902).
E. L. Ince, Ordinary Differential Equations (London, 1927).
E. T. Whittakeh and G. N. Watson, Modern Analysis (Cambridge,
1920), Chap. X.

The hypergeometric function:


F. Klein, Vorlesungen uber

die hypergeometrische

Funktion (Berlin,

1933).

E. T.

Whittaker and G. N. Watson,

loc. cit.

Chap. XIV.

The generalized hypergeometric functions


H. Bateman, Partial Differential Equations of Mathematical Physics
(Cambridge, 1932), Chap. IX.
E. T. Whittakeb and G. N. Watson, loc. cit. Chap. XVI.

THE HYPERGEOMETRIC FUNCTIONS


MISCELLANEOUS EXAMPLES

266

1. Show that the hypergeometric


whenRl(l + c a 6) > 0.
Prove that, when R16 < 1,

2.

Prove that, when

z is

Denoting by

>

Rl(z+A+1)

Clt C8

4.

convergent

(2n)!

=o

T(l+z)jT(l+z a.), show

that,

if

0,

[z+hY
where C
powers of

is

Z*2*nn,\n\z+n

the ratio

[]

b;c; 1)

not a negative integer or zero,

nz)T(j)
T(z+i)
3.

series F(a,

,...

C[z?+C1 [z]-W + tf2 l>]"--W + ...,

are the coefficients in the expansion of (1+A) a in

h.

Show

>

when Rl z

that,

0,

r(z)Vz

T(z+i)
*'
=
~ V/fi + 4(z+l) + 4
I

5.

If the two

12,32
.

(z+ l)(z + 2)

where

infinity

(z+ l)(z+2)(z+3)

+ "T
\

way

that

< x|i+!+-+,|,

independent of n, prove that


lim

6i

+ 6 + -+6 =
2

provided that the right-hand limit

Prove that an = T(a+n)/n\


when Rl a. > or when a. = 0.
6.

Deduce

8 12

|ai|+K| + -+K|

|,

with n in such a

KI+KI+-+KI
is

1.3.5
.

sums

K+o,+...+on
both tend to

+4

that, if sn denotes the

exists.

satisfies

sum

lin,

(Jensen.)
the conditions of Ex. 5 only

of the

first

n terms of the

series

F(a,b;c;l), then
r(e)n"+

*n

when Rl(c a b) <

0,

ft

-'!

~ T(a)T(b)(a+b-c)

and

T(a+b)

~WjW) losn

when

a+b.

(M. J. M. Hin.f)

t Proc. London Math. Soc. (2), 5 (1907), 335; 6 (1908), 339. The proof
indicated in this exercise is a modification of one due to Bromwich, ibid.

(1909), 101.

7.

THE HYPERGEOMETRIC FUNCTIONS


converge when \z\ < 1, and
2' 2 K

If the series ^l

z<i

over, the coefficients an satisfy the conditions that


(")

2 W^ <

2 Om

-K
I

<

when

31 "!

<

(i)
|%l
the constant

1,

is

267
if,

more-

divergent,

X being in-

dependent of x, prove that

km ~
<--i o

z<

"

a a;w

= hm
*

CT

provided fhat the right-hand limit exists.f


8.

Show

Rl(c a 6)

that, if

<

0,

F(a,b;c;x)

T(c)T(a+b-c)

Jt (l-xf-"-<>.
Prove also .,
that
t.

9.

Show

lim

T(a)T(b)

F(a,b;a+b;x)
,hm / rl /M V,
a^-i-o lg[l/(l x )\

when Rl(c a b)

that,

<

T(a+b)

r
,\rih\
T(a)r(6)

Gauss.)

0,

\F(a, b;c;x)

V,
Z/

T(a+b-c-n)T(o-a+n)F(c-b+n)T(c)
'

___

K+c_ a

_J

n<ric-a)r(c-b)T(a)T(b)

'

T(c-a-b)T(c)
T(c-a)T(c-b)'

k being the integer such that k

< Rl(a+6 c) <

ifc+1.

(Hardy.)

10. Determine the fifteen relations with polynomial coefficients which


connect F(a, b;c;z) with a pair of contiguous hypergeometric functions.
(Gauss.)
11.

Show

that the complete solution, valid

equation

,.

z(l_z)_ + i( a+J8+l)(l-2Z

when \z |

<

i,

of the

-0^ =

is

w = ^F{i a ,ij3;i;(l-2^} + B(l-2 2 )J{i(a+l),i( 8+l);|;(l-2Z


where A and B are arbitrary constants.
/

12.

Show

that, if

Rl(o b)

r (e +A)
when

|A| is

large

and

00

JT

8=0

fc

Ts + 6

>

0,

6' c

+A

|argA|

<

-1

e- T

(1

2
) },

)-ZTp^

the coefficients ks being defined by

77,

'- 1

e- T

-w

(l z+ze-7 )-".

t For an account of this and other generalizations of


of Cesaro, see Pringsheim, Acta Math. 28 (1904), 1.

a well-known theorem

THE HYPERGEOMETRIC FUNCTIONS

268

Show

also that the asymptotic expansion is valid

where 8

>

0,

provided that |1 zr l

Deduce from Ex.

13.

12,

<

when

<

|argA|

it S,

(Watson.-)-)

1.

the asymptotic expansion of

F(a+\,b+\;c+\;z).
14.{

Show

when

that,

<

\z\

1,
00 00

T(a)T(b)F(a,b;i;z)

|"

I"

e""-" cosh 2<J(uvz) u"- 1 ^' 1 dudv,

provided that the real parts of a and b are positive. (Whittakeb.)

Prove that, when |arg^|

15.

T{b)

r(c)

Show

-n 8, where 8

>

0,

F a,b;c;z- h)
{

T(-s)T(a+s)T(b+s)
"'W-IIV'"
F(a+s,b+s;c+s;z)h
~
T(c+s)
"'

that,

'

2m

provided that |arg(l z)|


16.

<

when

<

8.

ds,

(Whittakeb.)

< Rls < Rla <

R16,

T(s)T(a-s)T(b-s)
F(a s, b s;c s;z)
T(c-s)
T(a)T(b)

F(a,b;

T{c)

Show

17.

c;izt)?- 1 dt.

that, if

(l-z) a + b - c F(2a,2b;2c;z)
then

= f>n ,
o

=I^2
00

I6 ;c

+ i; ^W c-a,c-6;c + i; ^)

18.

of v

is

(Whittakeb.)

The Incomplete Gamma Function


positive, by the equation
y(v,z)

denned,

is

I|TST)

when the

S,,

real part

f e-ty-i-dt.
o

Show

that

V y{y, z)

= F (v; v + 1 z).
1

t Cambridge Phil. Trans. 22 (1912-23), 277-308. In this paper, Watson


determines the asymptotic expansion of F(a+e 1 X, & + e 2 A; c+e A;z), where
3
*i, 2> *3 take the values 0, 1, or 1 for large values of |A|.
t The results contained in Exx. 14, 15, 16 were obtained by E. T. Whittaker
by applying contact transformations to the solution of differential equations
by definite integrals: Proc. Edinburgh Math. Soc. (2), 2 (1931), 189-204.

THE HYPERGEOMETEJC FUNCTIONS


19.

269

Prove that the Error Function


z

Erfz

fexp( t*)dt
o

is

equal to z 1 .F'1 (J;f ;


20.

a 2

).

The generalized Laguerre polynomial!

n being a

i'(z)

is

defined

by the

positive integer or zero. Prove that

__
z

= e z~

!>(*)

(i)

rJ n

(er^f),
(0+)

"

(n)

(2)=

mahri

-*)'

00
21.

Show

that,

when

];|

<

1,

00

22.

Prove that the integral


00

converges when Rla > 1 and that it has the value r(ct+n+l)/n! or
and n are equal or unequal.
zero according as the integers

00

Deduce that, if/(s) can be expanded as a series of the formj

and term-by -term integration

is

valid,

2o cn ?'(*)

then

00

23.

Show

that, if m-\-n is a positive integer or zero,


1 \m p e rfm+n
/

Deduce that

= > zO^s)

(Dbbuyts.)

r=
The polynomial discussed by Laguerre (Bull. Soc. math, de France, 7
by putting a = 0. The generalized Laguerre polynomial was introduced by Sonine, Math. Annalen, 16 (1880), 1-80 (41-2).
X The question of the convergence of such a series may often be settled by
f

(1879), 72-81) is obtained

the use of Fejer's asymptotic formula (Comptes Rendus, 147 (1908), 1040),

L^\z)

y(^^')coB{2V()-(2+lW4}+0(n/->/),
/

valid

when n

is

large

and

z finite.

THE HYPERGEOMETRIC FUNCTIONS

270
24.

Prove that

IV+n+lXs-ir
dt =
le-^mt)
"
nlz"+ n+1

J
o

provided that the real parts of z and


25.

a+ 1

are positive. (Soninb.)

Prove that

provided that the real parts of p and cx+1 are positive. (Koshliakov.)
26.

Show

L^(2z)

that

r(+l+)|i^^j^).

M. Wnarar.)

(B.

r=0
27.

Show

that

*--()

=^

e-H+lJa {2j(zt)}dt,
J

provided that the real part of n + a+1


the Bessel function

is

positive.

Here Ja (u) denotes

{-wr
**>"2*i\T(ix+n+iy
o

28.

Prove that
(i)

(ii)

(iii)

Deduce

LW-IgLjz) = L^\z),
^S?

>

(*)= -i&tj'fc),

nL\z)

that, if

is

(2n+ a -l-z)<f2 1 ( 2 )-(n+ Q; -l)<f2 2 ( z ).

not a negative integer,

iw^W) mx)my)
(n+1)!

T(n+oc+l)(x-y)
29.

Show

that

f
30.

tL^(z)/T(a+n+l)

4 (zt)-iJ{2<J(zt)}.

Prove thatf

m=

where

{L%\x)L%Uy )-L%Ux)L%\y)}.

/.()

t See Hardy, Journal


(1933), 189.

= (M *

J ^rfSfl)

London Maths Soc. 7

(1932), 138;

Watson,

ibid.

THE HYPERGEOMETRIC FUNCTIONS


31.

The Hermite polynomials

are defined

exp(2te-* 2 )

= 1 Hn (z)tn/nl

Prove that

,
(i)

(ii)

(Hi)

Show

32.

Hn = (-l)exp ^exp(- Z
HM = (-ir2n'.L<-*\z
Hin+1 = (-l)"2WnUiLj*>(*)2

(z)

(ii)

),

is

2 n n!
34.

Vw

),

(z)

w=

^-Hn(z) = 2nHn_ 1 {z),

fl^z)- 2zHn (z) + 2nHn_1 (z) =

ffn (z)

is

0.

a solution of

dz a

Show

(z)

d*>
-^-

33.

that
(i)

Deduce that

271

by the expansion

_ dw
2z-z+ 2nw =
dz

0.

that the value of the integral

or zero

j e-^Hm {x)Hn (x)dx


00
according as the integers m and n are equal or unequal.

Prove that,t when

||

<

1,

+ See Watson, loc.

eit.

CHAPTER XI

LEGENDRE FUNCTIONS
11.1. Legendre's differential equation
last chapter we discussed the solution of the hypergeometric equation under the assumption that none of the
exponent-differences was an integer or zero. We shall now consider in detail an equation of hypergeometric type in which this
condition is not satisfied, namely Legendre's differential equation

In the

=
(l-z*)^-2z^+n(n+l)w
1

0,

dz

dz

where n

is an integer or zero.
This equation, which is of importance in mathematical
physics, arises when one tries to find a solution of Laplace's

equation

o 2 t/

8W

2217

dX 2 ^8Y2 ^dZ 2
a polynomial of degree n in X, Y, and Z. Such a soluharmonic of degree n. If we transform this
partial differential equation to spherical polar coordinates
(r, d, (f>) defined by the equations

which

is

tion

called a solid

is

X = rsin0cos<,
it

Y=

Z=

rsin0sin<,

rcosfl,

becomes
8r 2

+r

8r

+r

d 02

r2

+ r gm
2

^~

It follows that every solid harmonic of degree n is of the form


r n S n (9,<j>), where S n (d,<f>) is a polynomial in sin#, cos#, sin0,
cos <f> and satisfies the partial differential equation

cosec0^sm0^)+cosec 20^+rc(w+l)#n
Sn (8,

<f>)

is

0.

usually called a spherical harmonic of degree n.

In particular,

if

Sn (e,<f>)=f(0)coBm^+ e ),
where

m and e are constants, we find

tion of

that

= f(d)

is

a solu-

LEGENDKE FUNCTIONS
cosecd^(sm9-j?\+{n(n+l)
or,

putting z

273

cosec 2 d}w

0,

= cos 8,

2\d
n
{1 - Z)
2

d^-

o
2z

dw

+
di

n{n+1)

-l^] W =
\

This differential equation, which reduces to Legendre's equation


0, is called the associated Legendre equation. Its solution is seen, without much difficulty, to be of the form

when m

w= P
Riemann's notation ( 10.2).
In the problem of determining solid harmonics of degree n,
the parameter m can take only the values 0, 1, 2,..., n, and this
in

is

the case of greatest interest since the exponent-differences are

then all integers. We shall show that the associated Legendre


equation has, in this case, only one solution which is a polynomial in z and ^/(l z2 ); it is denoted by P%(z) when to is not
zero and by PJz) when to is zero. It follows from this that
there are only 2n+ 1 linearly independent spherical harmonics
of degree n, namely
PJcosfl),

P^(cose)cosw4,

P^(cos0)sinm<,

Any other spherical harmonic of degree n is


of harmonics of this set.
In the present chapter,

we

(m

1,2,.. .,n).

a linear combination

consider the form of the complete


and of the associated equation
n are positive integers or zero and

solutions of Legendre's equation

only in the case when

m and

m does not exceed n.

The Legendre polynomials


When n is a positive integer or zero,

11.11.

the point at infinity is


a regular singularity of Legendre's differential equation, the
exponents there being n+1 and n. If we attempt to satisfy
the equation by a series of the form

w=
41U

z-c{a

+a1 z-1 +a2 zrj

+...},

LEGENDRE FUNCTIONS

274

we

n is

find that the solution of exponent

where

say,

^4

and

JB

(n+l)(+2)(+3)(n+4)
2.4(2n+3)(2w+5)

"

two arbitrary

are

constants.

"

r "7|

The second

from the fact that the equation which should


constant B
out to be an identity. The solution of
turns
determine a2n+1
exponent n-\- 1 is found to be
arises

w = Cw
C is an arbitrary constant. We have thus found formally
2,

where

two

linearly independent solutions of Legendre's equation, viz.

w = Awv

Aw x

Cw2

a polynomial in z of degree n and


so certainly satisfies Legendre's equation for all values of z. The
second solution is a non-terminating series of descending powers
1; the formal processes which
of z which converges when \z\
made it a solution are thus completely justified. A discussion
of the analytical continuation of the second solution within the

The

first

solution

is

>

unit circle

is

deferred for the

moment.

(2n)\/{2 n (n\) 2 }, we can write the polynomial soluTaking A


tion in the form w = Pn (z), where

(-l)r(2n-2r)l
#.
Z, 2"r!(n r)!(n 2r)l

z
'

2r

is \n or \{n 1) according as n is even or


Legendre's polynomial of degree n. In parti-

where the integer p


odd.

We call Pn (z)

cular,

P {z)=l, P1 (z) =
From

z,

this definition

p f\
*n\*>

P(*0

2
i(3 -l),

we deduce
>T

x^

P,()

that

r_

Z,2"r!(n-r)I<fe

(
v

z2n-2r)'

i(&?-3).

LEGENDRE FUNCTIONS
dn

l) n\

cfe^
r!(n r)\
=

"%!

275

2 n n\ dz n
1

Pn (z) =

and so

-=-

dn
-f-

2*r\(nr)\

(z 2 -l).

This

is Rodrigues's formula for Legendre's polynomial.


Using Cauchy 's formula for the nth derivative of an analytic
function, we obtain from Rodrigues's formula the following contour integral formula, due to Schlafli:

p(z)-

- ^
1

dt

where

C is

Example

a closed contour surrounding the point

Pn (z) are all real


Example

= ^Sf^-^'i-i^i-^"

Show

3.

Pn
Example

4.

according as n

Example

5.

is

__^__ ^ _i n+i(in+1
s

f;z2)

even or odd.

Show, by using Rodrigues's formula and integrating by

j z*Pn (z) dz

0, 1, 2,...,

n 1.

Deduce that
l

Pm{*)Pn(z)

dz=0

-l

when

)-

Pn (z) is equal to

-i

F(n+\,~n;l;l(l-z)).

parts, that

when k

z.

that
(.z)

Prove that

(_l)(-i)/ 2

or

2. Prove that
Pn{z)

Example

Deduce from Rodrigues's formula that the n zeros of


and lie between ; 1.

1.

the integers

m and n are unequal.

LEGENDRE FUNCTIONS

276

Example

6.

Prove that
1

fz

2^!)*
Pn('

(2w+l)

-i
c

Deduce that

{-P(z)} 2

dz

2n + l

Example 7. Show that a polynomial/(z)


in the

form

of degree

n can be expressed

f(z)

= 2a P (z),
T

r=0
1

where

2r4- 1
-
=

or

f(z)PT (z)

dz.

-l

More

generally, prove that if /(z)

expanded as a

an analytic function which can be

is

series

r=0

which converges uniformly 'when


given by the above formula.

1 <

11.12. Laplace's integral for the

<

1,

the coefficients ar are

Legendre polynomials

We shall now deduce from Schlafli's integral a definite -integral


Pn (z), due to Laplace. Let us take as the contour

formula for

the circle \t z\

= <J\z 1|.
2

Then on C we have

= z+(z 1)W,
2

tt to tt. Evidently it is immaterial which


where varies from
branch of *J(z2 1) is taken.
Making this change of variable, we find that
<f>

=
1

2(z 2

l)h^{z+(z l)*cos<} =
2

and hence that


'

J ( z )

2{tz){z+(flfco&4>}

^J{

z +( z2

-1

icos

^M ^-

IT

Since the integrand

is

an even function of

<f>,

it

follows that

TT

pn {z) = -I

f {z+(z 2

- l)*cos# <ty.

This formula is known as Laplace 's first integral for the Legendre
polynomial Pn {z).

LEGENDRE FUNCTIONS
Example. By making the transformation
{z + (z 2 -l) 1 2 COS^}{z-(2 2 lJ^cOSl/.}
when z is real and greater than unity, show that
'

pn (z) =when

>

1.

By

{2

+ (2

-l) 1/2 cosoj}-- 1

277

daj

appealing to the principle of analytical continuation,

prove Laplace's second integral formula, that


w

4(z)

M {z+(z*-l)

1 li

cosw}- n - 1 dw,

where the upper or lower sign

Show

positive or negative.

when

z is

is taken according as the real part of z is


also that Laplace's second integral diverges

purely imaginary.

A generating function for the Legendre polynomials


The Legendre polynomials can also be denned as the coefficients in the expansion of (1 2hz+h 2 )~ 1 2 as a series of ascending
1 1 .2.

powers of h, a result which enables us to determine very simply


many of the properties of Pn (z). To prove this, we sum the
series 2 h nPn ( z )Using Laplace's first integral, we find that

|>

nP(z)

-^
= ~2
i

^{z+(z2 -l) } cos<}^

yhn{z+{z ~l)icoa<t>}n
2

d<f>,

provided that we can justify this inversion of the order of


integration and summation.
Let us suppose that

|&|<(l-e)/(| 2 |+V|z 2 -l|),

where

<e<

1.

We

then have

\h{z+(z*-l)icos<f>}\

<

\h\{\z\

00

and

so the series

2 hn

{z-\- (z 2

+ -l j*}<
|z

l-,

1 )*cos </>}"

converges absolutely and uniformly with respect to the real

LEGENDRE FUNCTIONS
The inversion of order is thus

278

variable

<j>.

valid.

It follows

that
00

2 hnPn {z) =
n=

TT

{l-hz-hizz-l^coscf,}-1

d<j>

by the fact that


when h = 0.
Now (1 2Az+A2 ) -1 2 regarded as a function of h, has but two

the branch of this function being determined

equal to

it is

'

and these are branch-points

at h
z(z2 1)*.
Hence, by Taylor's theorem, this function is expansible as a
series of powers of h whose radius of convergence is the smaller
f z i(2 2 !)*[ But we have seen that, for sufficiently small
values of h, the function is expansible in the form
hnPn ( z )As the Taylor expansion of an analytic function is unique, we
have thus shown that
singularities,

00

(l-2hz+h 2 )-^

<

provided that \h\


convergence of this

Example
P(l)

2.

to prove that

series is

{*)

In particular, the radius of

1)*|.

1 ^ z ^

unity when

1.

Prove that

P.(-l)

l,

Example

\z(z 2

= 2 hnPn

P2W (0) =

(-l),

Use the identity

P (cos 6)

is

TC

(1

"^y

2hcos6+h =
2

*Vi(0)

1.3(2n)(2n-2)

...

+ 2.4(2n-l)(2n-3) 2c S( "- 4)e+ -|Deduce that

Example

|P(cos0)|

3.

Show

<

when

that, if

(1+ft

2
)

(l-2hz+h*)

Deduce that
2"(w!) a

is real.

h+ 1/h =

2/k,

then

(1-zk)'

a" is equal to
t

(2n+I)!(

,,,
ln
, - ,.2n+l
(2n+1)P
3) (z)+(2n
2

-'

0.

(1 heK ){\ he-61 )

equal to

LEGENDRE FUNCTIONS
The recurrence formulae
We shall now find the recurrence formulae

279

11.21.

connecting Legendre polynomials of different orders by using their generating

y^ h) =

function

This generating function


differential equation

i_2hz+ hi)-il>.
easily seen to satisfy the partial

is

(l-2hz+h?) I-

Hence we have

(z-h)V.

(1-Zhz+V)

| nh^PJz) =

(z-h)

where the

<

\(z a -l) i \.

hn ~ x we find that

coefficients of

nPn (z)-(2n-l)zPn ^(z)+{n-l)Pn _ 2 (z)


which

is

the

first

of the recurrence formulae.

we now

^)_^M = ^
,

that
Z

If

differentiate

(dPn (z)

dPB _ x (z)l

dz

dz

JTt

(i)

(Z ).

with respect to

z,

dPn _Jz)

dPn ^(z)\

dz

dz

it

follows

by formula

dPJz)

From

dPn

these three formulae,

Jz)

it is

g Pn+i(g) _ggt=i(g)

(Z

(ii)

(with

for n), that

,...,

easy to deduce that

(2n+l)Pn (g),

IM = nzP {z)-nP _M,

2_ 1} d

(2n-l)Pn _ 1 (),

n 1
.

ii)

we obtain

=
from which

(i)

0,

from the relation

Similarly,

we deduce

(z),

converge absolutely provided that

infinite series

1*1

Equating

f h"Pn
o

(iv)

(v)

LEGENDRE FUNCTIONS

280

2_i)^) =

-(n+l)zPn (z)+(n+l)Pn+1 (z).

(3

(vi)

It should be observed that the six recurrence formulae of this


paragraph are really particular cases of the formulae of Gauss

connecting contiguous hypergeometric functions


1 1 .22.

10.5).

The integral of a product of Legendre polynomials

The polynomials PJz),

Pn {z) satisfy the differential equations

^{(l-* 2 )P;j+m(m+l)Pm

0,

{(l-z*)P'n} +n {n+l)Pn

0,

where dashes denote differentiation with respect to z. If we


multiply these equations by Pn and P respectively and subm

we obtain

tract,

(m-n)(m+n+l)Pm Pn

= Pm ^{(l-*)P;}-Pn ^{(l-*)P;J

Hence, by integration, we have

(m-n)(m+n+l) j Pm Pn dz

= [(l-2 )(Pm P;-P^PJ]^.

But, by the recurrence formula

(l-^)(Pm P;_p;pj
so that,

when

(v),

= nPm Pn _ -mPm _ Pn ^{m-n)zPm Pn


x

m ^ n,

fpm Pn dz =
I

K
L

f-i-<-i Pn + (m-n)zPm Pn
(m )(m+n+l)

-\

J Si

a formula which enables us to integrate the product of two


different Legendre polynomials between any limits.
To determine the corresponding formula for the integral of
the square of a Legendre polynomial,! we observe that, by the
t See Hargreaves, Proc. London Math. Soc.

(1),

29

(1898), 115-23.

LEGENDRE FUNCTIONS
recurrence formulae

and

(ii)

(iii),

2Pn (zPn Pm _i)

2Pn -l\"n

= -jZlz Pn~\~ z "n-l


and

z * n-l) "T Pn~r*n-1

*"n-l"n\>

so

(2+l) \ P\ dz -(2-l) P*_ x


J

we now put n

1, 2,...,

= [zP|+zP|- -2PK _
1

1 Pj;.

m and add, we obtain

fpi dz - fpi dz =

(2m+l)

<fe

a,

a,

If

281

z p2 +2z (Pf+P|+...+p2 l _ 1 )+

a,

ai

+ z p^-2(p p +p p +...+pm _
P =
1

or, since

P (z) =

1,

x (z)

(2m+l) jPldz

pj]*;,

z,

[2 Z (P|+Pl+...+PJl _ 1 )+

ai

+zP* -2(P1 P2 +P2 P3 +...+Pm _ 1 Pm ).


l

In particular, since

PK (1) =

1,

JPM z )PB z
(

fe

Pn ( 1) =

= 1 _2j:i

( 1), we

see that

orO

m and n are equal or unequal. This


can also be proved by using Rodrigues's formula and
integrating by parts, as suggested in 11.11, Exx. 7 and 8.
according as the integers

result

Example

Show

that, if

;>

n and

m n is even,

j Pm (z)Pn (z)
is

equal to

or l/(2n+l) according as

Example

2. Prove that

dz

m>

or

n.

j Pir (z)P3s+ i(z) dz

(-l)r++i(2r)!(2a+l)!

4f+(2r 2s l){2r+2s+2)(r!)*(s\)''

(Rayleigh.)

LEGENDRE FUNCTIONS

282

Example
J

3.

Show

(\-z*)P'm (z)P'n (z) dz

that

[(l-z*)Pm (z)P^z)fz l+n(n+l)

j Pm(z)Pn (z) dz.

Deduce the value of


l

[ (1

-z 2 )P^(z)P;(z) dz.

(Habgbeaves.)

-l

Example

4.

P'n (z)

Show

that

(2ra-l)PM_ 1 +(2n-5)P_ 3 +(2n-9)PM _ 5 +...,

the last term being 3PX or


according as n
otherwise show that, if
> n,

is

even or odd. Hence or

j P'm (z)P'n {z) dz

n(n+ 1)

or

-l
according as

11.3.

m n is even or odd.

Legendre polynomials of large degree

In many of the applications of Legendre polynomials it is


necessary to discuss the convergence of series of the form

y,an Pn (z), and therefore to know how Pn (z) behaves when its
degree is large. We shall now give an account of Darboux'sf
method of deducing the asymptotic expansion of the Legendre
polynomials of large degree from their generating function.
Let us consider, in the first place, the case when the point z
does not lie on the real axis between 1, so that z
cosh,
where Rl
0. We then have

>

_
provided that \h\
|e f|. The generating function of the polynomials PM (cosh) is, therefore, of the form (e~ A)-*F(fe),
where F(h) is regular when \h\
|e$|.
We can, however, write

<

<

F(h)

is

= (2sinhS+e-?-A)= f (-i,r)(e-f-A)'/(2sinh0'+,
1 '2

f Journal de Math. (3), 4 (1878), 5 and 377. It will be seen that the method
applicable to many of the functions defined by a generating function.

LEGENDKE FUNCTIONS
where

(a, r)

283

denotes the binomial coefficient

r(a+l)/{'!r(+l-r)}.
This power series converges
this it follows that, for

any

when

|e

h\

<

2|sinh|.

From

positive integral value of p,

where G(h) is regular when |A|


We have thus shown that

<

|e^[.

(e-~h)v+iG(h)

= 2 APm (coshO- f (-ir)(e--ftrV(2sinh) r+i


=0
r=0
= 2 ^(oo H h - (-t,r)(r-l,
r?o

=
say.

By

yA_}

!.**
Taylor's theorem, this expansion
\h\

The

coefficients

an

<

is

when

valid

|e-|.

am are given by

= J_.

e -C_fc)i>+*Q!(A)-**

c
denotes a circle concentric with and interior to the
circle of convergence. We can, however, deform the contour,

where

without altering the value of the integral, until it consists of


_
the circle of convergence \h\
|e S| indented at the singularity
e ^. Moreover, the integral round the indentation tends to zero
with the radius of the indentation.
Accordingly, if we write h
e~^ +6i on the modified contour,
where 6 varies from to 2it, we obtain

an

= e n-p-M
(

ATT

f
J

The

first

0i )*>+lG{e-+8i

dd

2w

eJn--C
= _L
Z7T
J

say.

6-^(1
e- n8if(6)

derivatives of f(9) are continuous, whilst its

derivative of order

p+1,

being of the form

(1 e6i )~*g(8), where

LEGENDRE FUNCTIONS

284
g{6) is

continuous,

fore integrate

p+1

and

discontinuous at

is

by

times

e(n-p-i)Q

2w

{ni)P+ 1

We can

2tt.

there-

parts to obtain
r e - ei (l eei )-ig(6) d6.
o

But

this integral

remains

n -> oo, and

finite as

so

np+1 e~ n ^an

is

uniformly boundedf as n -> oo, provided that lies in a closed


domain where the inequality Rl
is satisfied.

We

>
> 0,

when Rl

therefore see that,

PJcosh)

2 ( hr)(r
_

},)-

f r(j+f)r(i+r)

9X

since the

is large,

this equation

(-i)V*+x

!V(27rsinh)45

/ee\
,

(2 sinh ) r+ *

'

term corresponding to r = p is

PB (0OBh)

'

e<+^>e

V7rZ.TC!r!r(|-TO+r) k

But

(2 sinh )-+*

0(e n ^/n p+i

\^+V'
) when

can be written

r(i+r)r(j+r) /
r! T(^ n+r) \

e-e

y+

2sinh/

^\
-

^t+*/

In other words,

^ CS

4;

~ ! V(277sinh

2,

rl

r(J-n-fr)

2 sinh J

'

the series on the right-hand side being an asymptotic expansion


in Poincare's sense when Rl
0, since the error caused by
terminating the series at any term has the same order of magni-

>

tude as the first term omitted. It should be observed that this


asymptotic expansion may be written in the form

'<- t

0~^^'(=-^*'-SEEW

and, in particular, that

p (c h

-;pSW){ 1+0

This analysis has to be modified slightly


t It can, in fact, be shown that n^+'e
O in the next equation can be replaced
portance here.

when

z is

a real

^ tends to zero uniformly, so that


by

o.

This

is,

however, of no im-

LEGENDRE FUNCTIONS
number

lying between

285

since the branch-points of the

1,

generating function then both he on the circle

< < n,
case we write z = cos0, where

\h\

==

1.

In

this

and make use of the

fact that

Z/

'

'

(2sin0) r +*

;
'

(-2isin^r+i

regular save at the branch-points e 6i moreover, at e ei this


function behaves like (e ei h) p+i G(h), where 0(h) is regular
is

when

\e

Bi

h\ <

in powers of h
find that, if e

2sin0,

and

similarly at e~ 8i

If we

and apply the same argument

now expand
we

as before,

^ 6 ^ it e where e > 0, then

PM (cos0)

\7rsm0J

/2

(-1)"

{r(r+|)y

cos{(tt-r+i)fl+i(2r--l) ff}

2vr!r(r+J-)

%!

^sin-fl

and, in particular, that

^
This approximation

any

is

oos{{n+\)e-in}+0{n-W).

valid only

when

^ 6 < 7r

positive number. It can, however, be

'

p (cos&)

e,

where

e is

shown that

l<V(^)

under the less restrictive condition < 8 < -n. A proof of this
will be found in Hobson's book, Spherical and Ellipsoidal
Harmonics.
11.4.

is

The complete solution of Legendre's equation when


an integer

of Legendre's differential
equation at the points
1 are both zero, the complete solution
has a logarithmic singularity at these points. To discuss the
nature of the complete solution near these singular points we
apply the usual rule for solving a linear differential equation
when one solution is known, that is, we make the substitution

Since the exponent-differences

w=

PJz)u.

LEGENDRE FUNCTIONS

286

We

find that

satisfies

the equation

(l-z*)Pn (z)^ + 2{(l-z*)P^(z)-zPn (z)}^

from which

it

0,

follows that
z

where

and

= A + Bj.

dz

(z*-l){Pn {z)Y'

Hence the complete

are arbitrary constants.

solution of Legendre's differential equation

Now Pn (z)

is

is

a polynomial of degree n, with n distinct zeros


no zero being equal to 1. Accordingly we

z v z 2 ,..., z n , say,

have,

by the theory of partial

(z 2 -l){P(z)} 2

Here ar

Pn (z) =

is

2(z-l)

the residue of

fractions,

2( Z +1)

+ Z\z-z + {z-z
f\r

l/[(z 2 l){Pn (z)} 2 ]

=z

at z

r.

If we write

(zzr )F(z), we easily find that


a

2{zr

F(zr )-(l-z*)F'(zr )}
{l-zl){F{zr )f

But

if

we

substitute (zzr )F(z) for

ferential equation,

we

Pn (z)

in Legendre's dif-

obtain

h(z~zr ){(l~zi )F"{z)-2zF'(z)+n(n+l)F(z)}+

+{(l-z*)F'(z)-zF(z)}

= z we find that a = 0. Substituting the value of


l){Pm (z)} so obtained, we have as the complete solution

putting z
l/[(z 2

0;

r,

of Legendre's differential equation

= APn (z)-BUpn (z)!og?+Pn

(z)

J zz,

r=l

'

= AP^-B^P^log^-W^z)},
where A and B are arbitrary constants and W^^z)
nomial of degree n 1.

is

a poly-

LEGENDRE FUNCTIONS
The second

287

solution of Legendre's differential equation

which we have just determined, has logarithmic branch-points


1, but any branch of the function is one-valued and
1
regular in the z-plane supposed cut along the real axis from
at z

to

1.

We now

define

n (z)

in this cut plane

by

assigning to

the logarithms their principal values; we call this solution the


Legendre function of the second kind. This is equivalent to taking
2

where the path of integration does not cross the cut.


The cut is a line of discontinuity of Qn {z). For, if 1< n
we have
lim

When n

= iP.^W^Tiril-WUG*)

Qn ^ei)

Qn (^+0i) Qn {fi Oi)

and so
is

< 1,

rriPJfi,).

a point on the cut, it is convenient to define Qn (ix)


mean of Qn ([j.0i). With this convention

as the arithmetic

we have
when

1 < <
/x

1.

The behaviour of Qn (z) at infinity


When \z\ > 1, we can expand the function

1 1 .41

Qn)

\Pn {z)\0g

-^-W
n _^z)

Z
z

as a series of the form


Qniz)

= PJ 2 )(i + ^ + i + -}-^-^)
= z"-i{a +a z- +a z-*+...}
3

ar being constants. If am is the first nonvanishing coefficient, this solution has the exponent
n-\-\
say, the coefficients

at infinity.

We saw, however, in

11.11 that the exponents of Legendre's

LEGENDRE FUNCTIONS

288

and

solution of exponent

n.

equation at the regular singular point at infinity are

w+1. Evidently Qn (z) cannot be the


Accordingly, we identify Q n (z) with
+l, viz.

the solution of exponent

(n+l)(n+2)(n+3)(n+4)
"^

where

(7 is

4"

2.4(2ra+3)(2w+5)

+ ""'/'

a constant.

a2n is the first non-zero coefficient,


a fact which enables us to determine the polynomial Wn^z).
To determine C, we equate coefficients of z~ n - 1 in the two
expansions. This gives
It should be observed that

= a 2n
(2n)\
_
~~

2(w!)2(2w+l

n{n 1)

n(rc l)(n 2)(w 3)

2(2w 1) 2+ 2.4(2w l)(2n-3) 2

"J

We

have thus proved that, when

*W -

2+ir( W +|)

\z\

1+

Z
\

>

1,

2(2^+3)

2.4(2n+3)(2+5)

a result which can be expressed most simply in the form

Example

Show

that

.(*)

= ilog^y,
Z I

Q,(*)

= *P (ss)log?i-f*,

.(*)

= i-PaWloggy-^+f

-1,
Q t (z) = izlogp^
Z 1

LEGENDRE FUNCTIONS
Example

2.

Show

^=
)

289

that

(Siy.

(z

- 1)

li?,

n + 1 n + 1;2w + 2:
'

the latter formula being valid when \z


sponding formula, valid when |z+ 1 > 2 ?

1|

>

2.

ii)'

What

is

the corre-

Example 3. Prove

that, if we replace the factorials in the functions


functions, the functions so obtained are linearly
independent solutions of Legendre's differential equation for general
values of the parameter n. [The functions n (z) and Q n (z) so defined are

by

of Ex. 2

Gamma

called the Legendre functions of the first

11.42.

An

integral formula for

and second kinds

respectively.]

Qn (z)

Using the duplication formula for the Gamma function, we


find that we can write the hypergeometric series for Q {z)
n
( 11.41) in the form
fc)
QnW
*

when \z\
we have

>

1.

2z )
(

we

4j

2s )

But, by the Eulerian integral of the

an(i so

If

V{S+i)
V {n+2S)l r(w+s+f)
s-

= _L^
M+1

oo

insert vanishing terms of the

first

kind,

form

(n+2s+l)\
n\ (2s+

$!*-ir*
-l

in this infinite series,

we

find that

We can invert the order of integration and summation in this


last equation; for, since

\z\

I
4111

>

1,

the series

(n+r)\(t\ r
n\r\

\z

LEGENDRE FUNCTIONS
when 1 ^ t ^ 1. It

290

converges uniformly

Q n (z)

or, finally,

?>
_^+i
^3^

-1
This formula, which

follows that

the analogue of Schlafli's integral for

is

Pn (z), has been proved under the assumption that

\z\

>

1.

The

however, uniformly convergent when 2 lies in any


closed domain of the z-plane supposed cut along the real axis from
1 to +1.
Hence, by the principle of analytical continuation,
this formula for Q n (z) holds everywhere in the cut plane.
integral

is,

Example. Prove
type as those

that

satisfied

by

Qn (z)

satisfies

11.43. Heine's integrals for

We

shall

recurrence formulae of the same

Pn{z).

now deduce from

Q n (z)

the formula
1

two

^- P ^
2^1 J (zt)^1
-1

Qui*)

integral representations of

Laplace's integrals for

Qn (z) which

a real one. If

as

varies from

= z l)h
(z

we take the

to

1,

positive square root,

6 varies from a to

iUog
2
& 2

direct substitution that,

QJ?)

z is real

and greater

we

find that,

a, where

= arccothz,

the real value of the logarithm being taken.

by

are analogous to

Pn (z).

Let us suppose, in the first place, that


than unity, so that the transformation

is

dt

when

>

From this it follows

1,

{-(- l)*oosh0}

de,

LEGENDRE FUNCTIONS
and

so, since

the integrand

291

an even function of 0,

is

QnW =

2
/ {z-(z -l)*cosh0}

dd.

This formula is important since it enables us to calculate simply


the functions Qn {z). For example,

(z)

=*=

ilog^+J,
z

Qt (z)

= za-(z

-l)hmhac

frlogt+1-l,
z

and

so on.

For the more

when 2 is complex, we
If we now make the
{z+(z

we

difficult discussion

of this formula

refer the reader elsewhere.

substitution

l)*cosh}{z-( z2 l)icosh0} =

find that, as 9 varies

and hence,

from

to a,

<f>

varies

1,

from

after a little easy manipulation, that,

when

to
z

-f-oo,

>

1,

00

QnW =

{2

+ (z

-l)*COSh <}--!

This formula, which is due to Heine, $


place's second integral for PJz).

d<f>.

the analogue of La-

is

we wish to extend this formula to the case when z is comwe must first ensure that the integral is one-valued and
convergent. This is the case if we take the branch of (z 2 1)*
which is positive when z > 1, and make a cut in the z-plane
If

plex,

along the real axis from oo to 1. If z lies


domain within the cut plane, the integral

in a

bounded closed

00

(z-f- ( 2

2_ l)icosh J--l

converges uniformly with respect to z, and so represents a reguwhich is equal to Qn (z) when z
1. From

lar analytic function

>

t See, for example, Hobson's memoir published in Phil. Trans. (A) 187
(1896), 443-531, or his Spherical and Ellipsoidal Harmonics (Cambridge,
1931),

Chap. V.
t Journal fur Math. 42 (1851), 73-5.

The cut from - 1 to 1 is required to make (z 2 - 1)* one-valued, whilst that


from -oo to -1 ensures that z+(z 2 -l)*co8h ^ never vanishes when
> 0.
<j,

LEGBNDRE FUNCTIONS

292

this it follows,

by

analytical continuation, that the formula


CO

holds everywhere in the z-plane, supposed cut along the real axis
2
1)* has its principal
1, it being understood that (z
rom oo to

value. The value of QJz) on the cut is determined by the


equation
(-ijn+igjz)

QJ _ z) =

when z
11.5.
F.

<

1,

and by the convention of

when

1.4

Neumann's integral for QJz)


Neumannf showed that if z is any

<z<
1

1.

point of the z-plane,

supposed cut along the real axis from 1 to +1, QJz) can be
expressed in terms of Legendre's polynomial by the relation
i

QJZ) =

We

du

J1

PnW zu

deduce this result from Cauchy's integral formula

4.31,

Ex.).

Let z be any fixed point of the w-plane which does not lie
on the real axis between 1 and +1. We choose a closed
contour Cx which surrounds the points 1 but not the point z,
and which does not cross the cut; we also take the circle Ca
whose equation is \w\ = R, where R > \z\. Since QJw) is
regular in the annulus bounded by Cx and C2 Cauchy's formula
,

gives
1

dio

c,

r.

<

dw

modify the expression on the right-hand side of this


equation by making R tend to infinity.
It follows from 11.41 that, for all sufficiently large values

We now
of

H'

\QJw)\^A/\w\"+\

A=
.

where

n\

Vw

2 r(n+f)

t Journal JUr Math. 37 (1848), 24.

LEGENDRE FUNCTIONS

293

Hence we have
2ttA

Qn( w );
J

and

R n (B-\z\Y

wz

this implies that the integral

round

C2

tends to zero as

R tends to infinity.
We have thus shown that
QnW =

dw

^J

.()

is regular in the cut plane, we can deform the


without altering the value of the integral, until it
rj joined
consists of the two sides of the cut from
1-f-e to 1
ij.
Moreover, since
e, |io
by the circles |w+l|
1|

But

Qn (w)

since

contour

1;

W
the integrals round these small circles tend to zero as

and

ij

tend to zero. It follows that


i

QnW =

du
{Qn (u-0i)-Qn (u+0i)}^u

Finally, since

Qn (u-0i)-Q n (u+0i)
when 1

<u<

1,

this

= *iPn (u)

formula can be written in the form

-1

which

is

Neumann's

result.

11.51. Heine's expansion of (zu)- 1 as a series of Le-

gendre polynomials
Let us suppose that it is possible to expand
of Legendre polynomials of the form
1

m=0

am Pm {u).

l/(z

u) as a series

LEGENDRE FUNCTIONS

294

Then, by

11.11,

Ex.

the coefficients am are given

7,

by

dw

du
D/
j Pn (u)^- = (2m+l)Qm (z).

2m-\-\

I"

-i

This suggests that the formula


1

U = m2=o (2m+l)Qm (z)Pn (n)

probably true, provided that certain restrictions are made


z and u. This result, which is due to Heine, f is of
great importance; for on it depends the theory of the expansion
of a class of analytic functions as series of Legendre polynomials.
The proof given below is due to Christoffel. J
When
1, it follows from the recurrence formulae

is

concerning

m^

=
(2m+l)uPm (u) =
(2m+l)zQJz)

(m+l)Qm+1 (z)+mQm ^(z),


{m+l)Pm+1 (u)+mPm _ 1 (u),

that

(2m+l)(z-u)Q m (z)Pm (u)


= (m+l){Qm+1 (z)Pm (u)-Qm (z)Pm+1 (u)}-

m{Qm (z)Pm _ 1 (u)-Qm _ 1 (z)Pm (u)}.


The corresponding formula when
(z-u)Q

From

(z)P (u)

m=

is

= {dCOPoM-GoOOAMRi.

these two formulae

we

have,

by

addition,

m=0

(2m+l)Qm (z)Pm (u)

= %Z^
+ l^)Pn (u)-Q n (z)Pn+1 (u)}.
u +P^{Qn
%%
l

To prove Heine's formula, we have

therefore to

show that

(n+l){Qn+1 (z)Pn (u)-Q n (z)Pn+1 (u)}


tends to zero as n tends to

infinity.

Now, if z = cosh(a+ijS) where a > 0, and if ^ is real, we have


[z+(z l^cosh^l = !cosh(a-H/3)+sinh(a+^3)cosh<|
= ^{^(cosh 2a+cos 2j8)+sinh 2acosh^+
2

-f-

J(cosh

2a cos 2/J)cosh2 <}

^ ^/{cosh2a+sinh2acosh^} ^ ea
t

Journal filr Math. 42 (1851), 72.

% Ibid -

55

(1858), 61-82.

LEGENDRE FUNCTIONS
from Heine's integral

It follows

!(*)

{z+(z2

295

11.43) that

l^cosh^}-"-

&j>

J"

{cosh 2a+sinh2a cosh ^}-<"+ 1

>'

2
d<f>

J*

o
00

e -(-i)a f { C osh

2a+sinh 2a cosh ^} _1

d<

= e-(m - Q (cosh 2).


if u = cosh(y+iS) where
1)a

Similarly,

> 0,

and

if

is real,

we

have

+ (M -l)icOS0|
= |cosh(y+tS)+sinh(y+iS)cos0|
= V{K cosn 2y+cos 28)+sinh 2y cos0+ J(cosh 2y cos 2S)cos 0}
< ^{cosl^+sinh^} = ev,
2

|*t

and

by Laplace's

so,

1^)1

=-

Now suppose that


number

(+ 1)

less

<
=
<

first

than

a.

integral

11.12),

{u+(u2 I)*cos0}d0

< en

v.

and that e is an arbitrary positive


Then, when
< 7 < a e, we find that

a.

is

fixed

Qn+lWnM- Qn(z)Pn + l{u)

{n+ l){e-"Q (cosh 2a)emy+e-<w - 1 a Q


(n+ 1 ) #o(cosh 2a)e"<y-a'{l ea +r}
(w+l)Q (cosh2a)e- Ke {l+e2a!}.
>

(cosh 2a)e<'l + 1>r}

This last expression does not depend on j3, 7, or 8, and tends to


zero as n tends to infinity. It follows that the series

f(2n+l)Q n (z)Pn (u)


converges uniformly with respect to j8, 7, and 8 and has sum
l/(z-).
The simplest way of stating this result is to use geometrical language. If we keep a. fixed and vary /?, the point
cosh(a+ij8) traces out an ellipse with foci at the points of
z

affix

+1

and major

axis of length 2 cosh a.

The condition

LEGENDRE FUNCTIONS
u = cosh(y+i8) lies

296

^y<

a. e means that

within or on the
smaller confocal ellipse of major axis 2cosh(a e). The result

we have

just proved can therefore be stated in the following

form. The series

Z(2n+l)Qn (z)Pn (u)


o

and u when z lies on a fixed


^ 1 and u lies in any closed

converges uniformly with respect to z


ellipse

C with foci at the points of affix

domain

definitely within C.

11.52.

Neumann's expansion theorem

If f( z )

is

an

The sum of

the series is l/(zu).

and on an ellipse
can be expanded as a series

analytic function, regular within

C with foci at the points of affix


of Legendre polynomials

1,

it

M = fa P

n (z)

which converges uniformly when z


ellipse

For

Cv
if

lies

within or on a smaller

confocal with G.
is

any point within or on

Cv we

have

-u

=
=

2^- JV(s)
o
2~-

| (2n+l)Qn (z)Pn (u) du

2 cj m(2n+l)Q

n (z)Pn (u) du,

the inversion of the order of integration and summation being


valid since the infinite series under the sign of integration is
uniformly convergent with respect to z and u when z lies on C.

From

this it follows that f(u)

convergent series

can be expressed as a uniformly


,

f(u)

= J,an Pn {u),
o

where

=
2

'
JJMQ
c

n (z)dz.

This completes the proof of K. Neumann's expansion theorem.


The actual determination of the constants an is best carried
out by deforming the contour G, as in 11.5, until it consists

of the two sides

by the

LEGENDRE FUNCTIONS
of the cut from 1+e to 1 ij,

|z+l|

circles

e,

value of the integral for a n

\zl\

rj.

297

joined together

This does not alter the

It easily follows that

= ^^

f f(x){Q(x-Oi)-Q(x+Oi)} dx
-l

and hence that


a

2ra+l
f(x)Pn (x) dx.

we can

Alternatively

J*

term by term in the equation

integrate

f(x)Pn (x) dx

(*

_i r=0

_i

ar Pr (x)Pn (x)dx

which leads to the same result; this process is valid on account


of the uniform convergence of Neumann's series.
For theorems of a similar nature relating to functions of a
real variable, we refer the reader to Chapter VII of Hobson's
Spherical and Ellipsoidal Harmonics.

Example. The

function f(z)

an analytic function regular in an

is

annulus bounded externally by an ellipse C 2 with foci at the points of


affix J: 1, and internally by a smaller confocal ellipse. Show that /(z)
can be expanded as a series of the form
/(z)

where

= f an Pn (z)+f bn Qn (z),
2n+l
2~=

f(z)Q(z) dz,

J
c,

2n+l
K= 2
Prove also that the

series

jf(z)Pn (z)dz.

converges uniformly in any closed domain


(K. Neumann.)

definitely within the annulus.

11.6.

The

associated Legendre functions

We saw in

11.1 that the problem of finding all the spherical


harmonics of degree n depends on solving the associated Legendre differential equation

w
dw
- z).d^2z
2

(1

^+r

" +1)

2
)

"i=^r =0.

LEGENDRE FUNCTIONS

298

where n and

(^

n) are positive integers or zero.

This

dif-

ferential equation has three regular singularities, its complete

solution being of the form


oo

(1
If

we make

fyn

\m

n-\-\

the substitution

are changed to

mn and m+-f

1.

n
w

\m
\m

(z2

z\.
)

l) imu,

the exponents at

and m, and the exponents at infinity to


But this function u satisfies the differential

equation
(in

(1 z 2 ) 2(m+l)z
az-

and

this,

by

-f-(M m)(n-\-m-\-l)u =
dz
fitj

0,

Leibniz's theorem, can be written in the form

dm

,d v
dv
(,,
^-)-.|+-(-+i).}-.
a

dl?

= dmvjdzm

where u

Hence,

if

is

any solution of Legendre's

differential equation, the function

w=
satisfies

2 -l)*
(z
v

'

^
dz m

the associated Legendre equation.

Accordingly

we take as the two linearly independent solutions

of this equation the functions

P(z)

= Z *_l)**g^,
(

QZ(z)

being supposed that the z-plane


from
1 to 1 and that (z 2
1)* has

it

is

its

2_i)**^l),

cut along the real axis


principal value.

These

functions are called the associated Legendre functions.

When m

We

is odd, P%{z) is undefined on the cut from


1 to
complete the definition when
1
x
1 by writing

1.

< <

where the positive value of the square root is taken; it must,


however, be remembered that, although P(z) is now defined
everywhere,

P(x)

it is

discontinuous across the cut, since

lim
+

c--

el m7ri P%(x+ie)

= Km er^
e-*+0

mr,i

F^{x ie).

LEGENDRE FUNCTIONS

When

is

even, P%(z)

a polynomial of degree n, and so

is

regular everywhere; but as


in definition according as

when

1 < x <

1,

all

Similarly

even or odd, we suppose that,

is

(-l)(l-a*)*?Lg^

values of ra.f
convenient to define <2(x),

it is

by the equation
( l) m Q%(%)

is

undesirable to have differences

it is

the equation

PZ(x)
holds for

299

= i hm {e-i

m7ri

when

<x<
1

1,

Q%(x+ei)+e* m7H QZ(x ei)},

e->-+0

Q n (x) when

which agrees with the definition of

m=

The

0.

functions P%{x), Q(x) so defined evidently satisfy the differential equation for real values of x.

Example

Prove that P()

is

even or odd according as n

is

even

or odd.

Example

Prove that

2.

J \TO/2

z2

(pl+m

Deduce that

^" (8,-

2.n!(+)!

2.(2n-l)

'

(n-wt)(w-m-l)(n-w-2)(n-OT-3)

_4

"*"

2.4.(2n 1)(2 3)

Example

Show

3.

Show

4.

that

"

-l
according as

Example

n
5.

is

or

is

^-^^
2n+l

not equal to

orO

(ra m)!

r.

Prove that

\^
-l
J

according as

/z4-l\/ 2 dn

fi-( Z )IT( 2 ) dz
J

_ '"/'

that

Example

n x)Pl(x)
*

J^
= L^^l
m(nm)\

a;

or0

m is or is not equal to h.

Hobson's definition. The factor ( 1) is sometimes omitted; to


distinguish the two cases, it is usual to write T%(x) = (lx 1 ) m 2 dmPn (x)jdxm
a notation due to Ferrers (Spherical Harmonics (1877), 76).
t This

is

300

Example

Show

6.

LEGENDRE FUNCTIONS
< n,

that, if r

xP%(x)P?(x) dx

-l

when

vanishes, save

= n 1,

when

value

its

4w 2 1 (n m

Example

Show

7.

that, if

(-ir!?(*0i)
11.61.

is

(n+m)l

2tt

1)!'

1 < x < 1,
= e m" {(x)Ti7nf?(a;)}.
2

A lemma of Jacobi

Before

we

discuss the representation of P(z)

integrals, it is convenient to

prove that,

r = (_

Jm-lgfapm-lJ,

rffi-

if

fj.

l)m-l (2m)!

l.

2 m m!

to

an interesting result due to Jacobi. f


The result is true when to = 2; for, in
shown that

by

definite

cos <j>,

sin ra<A,
r

this case, it

is

easily

d8in3<f>

We

shall

show that the lemma

is

true generally

by the method

of induction.

lemma

Let us suppose that the


value of to. Then, writing
Jr

true for

some

particular

(-l)-i(2m)l
2 mTO!

TO

we deduce

is

>
'

that

dmsin2m_1 <

d sin m<f>

dp

^tocosto^

__

sin <f>

dfj,

and
dm+1 sin2m -i<
d/i"+

~ m sinTOcsin<-|-- WiCOSTO, cos


2

<
'

sin3 <

Hence, by Leibniz's theorem, we have

dm+1 sm2m +1
dfxm + 1

m+
2m _
d {(l j?)sin
m+
1

(f>

dfj,

1
<f>}

= K{ sinm<f> TOcosTO^cot^+
+ 2to(to+ l)cos m<f> cot 0~to(to+ l)sinTO^}
2

t Journal filr Math. 15 (1836), 3-4. Tho proof given here


Proc. London Math. Soc. (2), 23 (1925), (Records) xxx.

is

that of Ferrar,

LEGENDBE FUNCTIONS

301

= Km(2m+l)cos(m+l)(f>lsin<f>
m+1

d(j.{

If

we now

integrate with respect to

dm

2m+1

fi,

we

find that

J2m+2)!
= (=1
m+1 2+ (m+l)!

<f>

d/*

the constant of integration vanishing since the expression on


each side of the equation is an odd function of <f>.
We have thus shown that, if the lemma is true for any particular value of m,

true for to =

true for the next greater.

it is

But

since it is

true generally.

2, it is

11.62. Integral representations of P%(z)

The simplest

integral formula for P%(z)

Schlafli's integral for the

Pn {z >

is

the analogue of

Legendre polynomial, namely

- ^2^r (z

1}

zri J (*-)-h-+i

'

C is a simple closed contour within which the point z


This follows immediately from the generalized Rodrigues's
formula
lUm Jn+m
2
where
lies.

pm(~\

by using Cauchy's

L>

Vs

integral for the

2_ \\n

(n+m)th

derivative.

When z does not he on the real axis between 1, we


as C the circle \t~z\ = Vl z2 ~ *! Then on C we have
t

where
It

(z

now

can take

= z+(z l)*e<^,
2

1)* has its principal value and

<f>

varies from

follows, exactly as in 11.12, that

P( Z )

(W+

)!

n\

I
it

n
2
f {z+(z -lfcOScf>} COSmcf,
J
o

Now by

the

lemma

cos mi
r

we have
rf^sin 2 " 1 y
^ da
^
,w 2 mm!
l)"-i(

-,
v
m
d^
rf/x
(2m)!

of 11.61

d*f>.

to 2n.

LEGENDRE FUNCTIONS

302

where

denotes

fx

We

coscf>.

(2m)\nl

by parts

Integrating

Finally, putting

fx.

rr

we deduce

that

we

cos (/>,

dp

m times,

(n m)\

{2m)\

can, therefore, write this definite

form

integral for P%(z) in the

obtain a second definite-integral

formula for P%(z), namely

(2m)l(nm)\

it

J
o

vahd when the z-plane is cut along the real axis from 1 to 1
and (z2 1)* has its principal value.
It should be observed that the two integrals of this paragraph
reduce to Laplace's

Example
(z 2

integral for the Legendre polynomial

first

m is zero.

when

l)

1 /2

has

Show

that,

when

the z-plane

is

cut from

to

and

principal value,

its

{z+(z 2 -l)i/ 2 cos^r

P.(a)

+2

y -^
+ r

.PS{z)oosmd,.
T

Z-, 7(n

m)l

m=l

Example

By

2.

{z

when

>

applying the transformation

+ (z

-l) 1 2 cos^}{z-(z 2 -l)i/ 2 C osi/r}


/

the second definite integral of

1 to

11.62,

show that

IT

PS(Z)

2 m m' (n-\-m)*

(2^!

(r^\

(zi

- l)m

'\)

{*+(**-

ll2 osco}---i

S ma>da>.

Deduce

Z)

that,

when the

^|~^(

z-plane

22

-1

where the upper or lower sign

m
)

is

also that the integral diverges

is

cut from

'2

ij{

to

1,

-1

Z +( Z2 -l) 1/2 ^-}-''-M

taken according as Rl z

when Rl z

0.

>

or

^ ^^,

<

0.

Show

LEGENDRE FUNCTIONS
Example

By

3.

303

using Jacobi's lemma, show that


IT

-FT(z)

1)*"
,

n
n!

(nm)\

77

'~

f
{z

+ (z l)
2

1/2

- 1 cosmtodto,

ooso)}- n

J
o

where the upper or lower sign

Example

4.

Show

is

taken according as Rl z

<

<

that, if

>

or

<

0.

n,

(coadisin9coB<l>) n

-Pn (cos0)

+ 2 2 (l)me~ml
m=i

-^PJ^cosflJcosm^.

2-

(n-\-m)\

11.63. Integral representations of <2(z)

We have

seen in

axis between

11.42 that,

and

when

does not

lie

on the

real

1,
i

-i

From

by

this it easily follows

differentiation

under the sign of

integration that the equation

Oft)
mV

Jl=gT

(-l)m(+")'
(g ._l)i.
v

f
n+m+1
J (zt)

2 n+1 n\

'

'

i
holds everywhere in the cut z-plane. This
definite-integral

representation

of Q%(z),

generalizations of Heine's integrals

As the analysis runs on exactly


and 11.62, we state the results
Example
Q%(z)

where a

Prove that, when

(-irfo+m)!

it

the

11.43) can be deduced.

similar lines to that of 11.43

as exercises for the reader.

z is real

and greater than unity,

^_ (z2 _ 1)

i/2

cosh ^ coshm0d ^

2.

Show

that, if
.

coshmfl

Hence prove

that,

when

W)!W
= ~!r
?.t
(2m)! (n my.
(

the fundamental

and from

arccothz.

Example

Q%(z)

is

ft

!l

/u

coth0,

= r-sinh0.d
2""m!

(2m)!

>

msinhsm- 1
-j-
m
dfx,

1,

2m
(z

-l) m;2

{z-iz^-l^coshe^-^smh^ede.

LEGENDRE FUNCTIONS

304

Example
_

Deduce from Ex. 3 that the equations


l l) m (n4-m)lm<2 m
2
2
',
(z - ir' x
y \.j
(2m)!
(n m)\

3.
<

CO

f {z

+ (2 - 1
s

l
)

/*cosh <}--- 1 sinh 2m<

d<f>,

CO

Q"(z)

t\\mn
*

,,"

{z

+ (

-l) 1 2 cosh^}--icoshm^d^
/

hold everywhere in the cut z-plane.

The addition -theorem for the Legendre polynomials

1 1 .7.

OA and the radius

denotes the angle between a fixed line

If

vector to the point of spherical polar coordinates

(r,d,<f>), the
a spherical harmonic of degree n. Accordingly we can express this harmonic as a series of the form

function.

Pm (cos0)

Pm (cos0) =
where the

is

a o PK (cos0)-f

2 {am C0Bm(
m=l

f>

Jr

m sinm(f>)P%-(cos9),

am and b m are independent of 6 and


In
has polar coordinates (r v 6 X 0), this expansion

coefficients

particular, if

(f>.

becomes

Pm (cos 6 cos ^-f sin# sinflj cos

=a

<f>)

n
o

P(cos0)-f-

a result which

m=l

(a m cosm^-f & m sinm0)P(cos0),

called the addition-theorem for the

is

Legendre

polynomials.
If

we

and

write z

z x for

cos# and cos^ respectively, this

addition-theorem takes the form

Pn (0 =
where

Pn {z)+ J

= zz

(a m cos m</>+b m sin m<j>)P^(z),

(zi l) i (z\~l)icos<j>.

In the present section we propose to determine the coefficients


am and b m for general complex values of z and zx
Let us suppose, in the first instance, that z and z x are real
and greater than unity and that ^ ^ ir, so that is certainly positive. The expansion
.

<f>

CO

2 ^pm (0 =
m=0

(1-2^+^-1,2

LEGENDRE FUNCTIONS

305

then holds for all sufficiently small values of h. We now express


the function on the right-hand side of this equation as a definite
integral and deduce the addition-theorem by equating coefficients of powers of h.
When h is real, it is easily shown that

(l-2h+h2 )-W

= [(z-hz

1)

-{(z2 -l)-2h(z2 -l)i(zl-l)icos<f>+h*(zl-l)}]-l

IT

[( z

-^i)+{( 22 - 1 )- 2^22 - l

i
)

!- l

icos<

A+

+h2 (zl-l)}icos(w-oc)]for every real value of a. In particular, if

tana

da>

we take

ft(s|-l)W

=
(z

we

-l)*-A(zf-l)*cos<'

obtain

(1 2AC+&

=^

)- 1 ' 2

[{z-hz 1 )

+ {(z

-l)i~h(zl~l)icOS<j>}cOSOJ

IT
-\-h(z\ l)*sin0 sin co]- 1 dco
tt

==

kn\

z +( z2

1 ) icosw

- h z i+( zi{

1 ) icos

("-i>)}]~ ld "-

IT

This formula also holds, by the theory of analytical continuation, for all sufficiently small complex values of h.
For any fixed value of h which satisfies the inequality

we can expand the integrand


series in

<
77

a)

of the last integral as a powerh which converges uniformly with respect to a> when
Accordingly we can integrate term by term to
it.

obtain

iAPB ()=(l-2MH-tf)-i'

=o

V,I f {z +(zl-l)i C S -<f>)}n


= Z/
2tt
{z+(z -l)*cosco} +i
(qJ

4111

re

LEGENDRE FUNCTIONS
Equating coefficients, we deduce that
306

J_ f

n ^'

27T

{ Zl

ir
If

we

+ (zl-l)icO S (<Q-4>)y j^

{+(*- 1)*COSO>}+1

substitute in this equation the value of


{z 1 +(zf-l)*cos(o> 4>)}

given in

PJ z i)

11.62,

Ex.

we

Pn (t) is equal to

find that

*"

f
2
J {z+(z

2t7

1,

l) cosw}"+
i

'

7T

TT

^(ra+m)!
(rc+m)!
^l

77

{z+(z 2

- l^costo}^

But
cosm(ft)-^)dw

77

{(!

cos mo>

(;* l^COSft)}"* 1

~ COSm ^

c?a>

"

{z+^-lJicog^+l"

7T
7T

+ Smm ^
,

TT
== 2Tr(-l) m

by

11.62,

Ex.

3,

sin mco dio

f
{2

+ (22-l)*cosa,r +

^^P(z)cosm^,

the integral involving sinmaj vanishing since

integrand is an odd function of


We have thus shown that, when

its

co.

and

z x are real

and greater

than unity,
P{ii-(s?- !)*(!- 1)*oob^}

= Pn (z)Pn Zl )+2
(

J ^^(-l)
m=l
i

m P^(z)P^(^)cosm<l>.

'"

the proof of the addition-theorem, we have only


to extend this result to general complex values of z, z 1 and <j>.
Let us suppose that the Argand plane is cut along the real

To complete

and that (z2 1)* and (zf 1)* have their


principal values. When z is any point of this cut plane, z 1 (> 1)
and being real, the expressions on each side of the last equation are regular analytic functions of z which are equal when
axis from

<j>

to

LEGENDRE FUNCTIONS
z is real

and greater than

By

unity.

307

the principle of analytical

still holds for such complex values


and < remaining real. The restriction on zx is next
removed in the same way. Finally it is not necessary to suppose

continuation, the equation

of

zx

z,

that

<f>

is real,

since the addition-theorem expresses the equality

of two polynomials in e^ 1 and e~^ 1

We have thus shown that if z and

any two points of the


axis from 1 to 1, and

z x are

Argand plane supposed, cut along the real


is any complex number, the equation
if
<f>

Pn (0 = Pn (z)Pn

( Zl

)+2

g=^(_i r p-( z )p

(zi )cosm0,

=i

where

and

= zz l)l{z\ l^cos^, holds provided


1)* have their principal values.

(z

that (z 2

1)*

(z\

Example. Prove
P(cos 6 cos d x

+ sin

that,

when

6X

6,

and

<j>

are real,

sin 6 t cos <j>)

= Pw (cose)P(cos6' + 2
1)

S p^^ PS(
l

CoS

e)PS(ooad 1 )oo S m4.

n=l

Legendre functions of non -integral order


In the present chapter we have only discussed the

11.8.

solution

of the associated Legendre equation

- z)o\d w 2z dw n{n+i) m
d*- Tz+{
-i=*r=
case when m ^ n and m and n are positive integers
,,

(1

in the

or

which is of most frequent occurrence in physical


problems. For a discussion of the solution in the general case
when m and n have any real or complex values, we refer the
reader to the original memoirs, in particular those of Barnesf
and HobsonJ.
zero, the case

REFERENCES
H. Bateman, Partial

Differential Equations of Mathematical Physics


(Cambridge, 1932), Chap. VI.
E. W. Hobson, The Theory of Spherical and Ellipsoidal Harmonics
(Cambridge, 1931).
E. T. "Whittakeb and G. N. Watson, Modern Analysis (Cambridge,
1920), Chap. XV.

t Quarterly Journal of Math. 39 (1908), 97-204.


% Phil. Trans. (A), 187 (1896), 443-531. See also Hobson's book cited above.

LEGENDRE FUNCTIONS
MISCELLANEOUS EXAMPLES

308

1.

Show that the value

of the integral

j z(l-z*)P'm (z)P'n (z)

dz

-l
is

zero unless

m = n^

Find the value of the

1.

integral in these excep-

tional cases.
2.

Prove that,

if

m>

n,

= i^^l_{3m(m+l)-n(n+l) + 6}

j P'^z)P'n (z)dz

or

-l
according as
3.

m n

even or odd.

is

Prove that

Pl
nKZ)

dh

~~2-rd) hn + 1 (l-2hz+h 2 ) 1

'

where

is

a simple closed contour enclosing the origin but neither of


2 ; the branch of (1 2hz+h i ) 1 2 is that which

the points z(z 2 l) 1


reduces to I when h

4.

Deduce from Ex.

'

0.

3 that

hndh

rt^-A.
nK
~ 2mi

f
(l-2hz+h2 ) 1 2
J
r
where Y is a simple closed contour enclosing both the points
the branch of (1 2hz + h i fl i is specified by the condition
'

'

z (z 2 l)1

'

2
;

(l-2hz+h*) 1 2 /h->l
l

> oo.
Hence prove

as h

<

when

that,

v.,
o\
R.(COS0)
BV

'

<

V2 f
77

it,

+ 37775 "4>,
,,
r
COS0)

ooa(n
;

(COS0j

j)<f>

1'2

the positive square root being taken, f


5.

Show

where R
and Z.

is

that

if

R = X + Y* + Z
2

2
,

the numbers involved being real,

regarded as a function of the independent variables X,

To
t This is the Diriohlet-Mehler integral for the Legendre polynomials. 9
prove it, deform, the contour Y until it consists of two small circles about e ',
joined by an arc of \h\ = 1, covered twice.

LEGENDBE FUNCTIONS
Prove that the polynomial

6.

W_
n

l (z)

309

the differential equation

satisfies

(l-z*)^-2z^ + n(n+l)w =

2P'n (z).

Deduce that

2n 1

2w

2ra 5

(Chbistoffel.)

Show

7.

that

Wn_^(z) = ^i' (*>P_ (8)+


1

P (*_
1

(8)+...+P _ 1 (>P ().


1

Prove the following generalization of Rodrigues's formula:

8.

= - 1 r +1 2"n

Qn(z)

where

Z>

_1

D-^-^z 2 - 1 J-"" 1

denotes integration from the point at infinity to

z.

Show that

9.

P{*)Qr^MPn-i(*)Qn(z) =
Deduce that

.w =

tp.(. )

iog!

-pllW 2

r^km-

Prove that

10.

00

00

(2n+l){Qn (z)Y dz

(8-lKfl-i(s)} ,

fe

= -^.

(2n

+ l){Qn Z )}Mz (

J"

(2n-l){(2_ 1 ( Z )} 2 dz

=1.

Hence show that


CO

CO

j (2n+l){Qn (z)}*dz=^?,
+i

i
1

CO

(2n+ l){Qn (z)Y

fe

= _

J I.
(Habgbeaves.)

Prove that

11.

p.(*)q;(8)-p;(*)q,()

Show

12.

i/(i-z 2 )-

that

(l-^MPia-P^Q^.!) =
Deduce

that,

when

j
i

(2n

+ 3)Pn+1

does not

lie

(w+l) 2 (P+i<2+1 -P<2Jbetween 1,

- j (2n+ 1)P Q dz
= z ("+l Vjh-1+P 0)~(P Cn+i+Pn+

dz

$)

LEGENDRE FUNCTIONS

310

Show

also that
i

(2n+l) j

Pn (x)Qn (x) dx -(2n + 3)

Pm+1 M<3 M+1 (z) =

o
1

Hence evaluate

PK (a:)C n

(Habgbeaves.)

d*.

(a;)

13.

Prove that
(2t&4- \)z

= ^piy-

pw(*)Q-i(*)--P-i(*)0h()
14.

Show

that

(l+x)m +nPm (x)Pn(x) dx

{mlni} 2 {2m + 2n+lj\'

-l
15.

Show

(TlTCHMABSH.)
that
1

(m-n)(m+n+l) j Pm (x)Q n {x)

dx

l_(-l)+.

-l

Deduce

that, if m-\-n

odd,

is

{m-n)(m+n+ 1)

Pm (x)Q n (x) dx =

1.

(Nicholson.)

16.

Prove that
l

(2m-2n)(2m+2n+l)

P2m {x)Q m (x) dx

(2n)! [2 mm\

l_(_l\m+^
,

17.

d_

(2m)!l2n!

(Nicholson.)
v

Show, by the aid of the addition theorem or otherwise, that


l

{x)

dx

Vd--a;r
2

^{Pn (0)}K

-l

Hence show that

Deduce Catalan's expansion


sin-^

br

'

""O
18.

Prove that,

if

2 .4.

! r )

,{P

(a!)

~ P'*- l(a!)}

(Nicholson.)

n > m,

P(cos0)cosm0<H?

=7,

-^
,

,--,,

n-r

or

LEGENDRE FUNCTIONS
according as

n-m is

311

even or odd. Deduce that


oo

n=m
and
sin{(2m+ l)cos -1 z}

=
19.

Show

,.,V
(r,
{P
(m+i)
r(n+m+2)r(n-m+l)
K_m
T(n+m+i)T(n-m+l)

(z)

- Pa" (Z)}

(Nicholson.)

that

r(|m+in+i)r(lm-n)

P(cos0)si
sinm0 dd
o

when

m>

n and

m n is odd, and that the value of the integral is zero


(Heine.)

in all other cases.


20.

The equation of a nearly

a{l

spherical surface of revolution

+ eP(cos#)}, where e is small. Show that,

if e 3

be neglected,

is

its

volume and surface are respectively

21.

Show

that

if

m>

n and

>

f.()P,(*)V*(*)<fc
J

m\A m =

where

0,

= (2^+2^-25+1)^^/

.3 .5...

(2m 1).

00

22.

where

| h-Pn (z)Pn {zh) = ^ (1 _^

Prove that

(Febbebs.)

27T

2)1/2

X denotes the complete elliptic integralf of modulus


(Gebonimus.)

J(lE^}-

23.

Show

that, if

^f

sufficiently small,

\h\ is

hn *w

Q n {z)

=
-

(1
v-

--!-
2fts+A

24.
{z

! )- 1 / a
i

/l

~S

arccosh
" 7(^-l)i/2-

Prove that

+ (z

-l) 1 /2}--i
00

= -(+D 2,
f

See

The complete
14.42.

elliptic

2 7r.m!(m++l)!

integral of

modulus k

1
is f <ft/\/{(l

^^^
2

)(l W)}.

LEGENDRE FUNCTIONS

312
25. If

>

and y

*J(x

1), show that

j Pn (l+2y*am.

!i

d<j>
<f>)

(Nicholson.)

TT{Pn (x)}".

26.

Prove that,

if

1 <

xt

<

and

dj,

j dx p.&jp.tf.)
-1

where

^=
&=

1 <

2ff

aa . i

+ (1 _ a

S) i/2(

x2

<

1,

~- p
x

n d),

_ aa)i/2 0OS (^_^

i)>

aa;,+(l-) "(l-a|) 1 /Scos(^-^),


1

CHAPTER XII

BESSEL FUNCTIONS
12.1. Bessel's differential

equation

In the present chapter we determine the complete solution of


Bessel's differential equation

d2w
dz

dw

v2 \

^zdz^\

z2 )

where v is a constant. It is an equation of the confluent hypergeometric type considered in 10.61, and has a regular singularity of exponents y at the origin and an irregular singularity
at infinity.

This differential equation is of frequent occurrence in physical


problems. Eor example, it arises in the determination of the
solutions of Laplace's equation associated with the circular

and so its solutions are sometimes called cylinder funcFor suppose we transform Laplace's equation

cylinder,
tions.

8W

82 V

+ + _
8X* 8Y* 8Z*

82

to cylindrical coordinates

(p,

<f>,

Z) defined by the equations

= psin<f>;
1 8W
8W n
8W -1 _
+ 8V + - + W2 = 0.
X = pcos<l>,

41

this gives

It follows that the expression

V
where

Jc,

v,

and

= ekZw{p)coa(v(j>-\-e),

are constants,

is

a solution of Laplace's equa-

w satisfies Bessel's
1 dw
dw

tion provided that

equation

;+r+(*-)=o
+ pdp~

lf?

with independent variable hp.


In most physical problems V must be a one-valued function
of position in space, and so v must be an integer. In this case
the exponent-difference at the origin is an integer and it turns
out that it is impossible to express the complete solution in
powers of z. It will, therefore, be necessary to distinguish

'

314
BESSEL FUNCTIONS
between formulae which hold for general values of v and those
which are valid only when v is an integer. We do this by-

adopting the convention, usual in this subject, that the parameter occurring in BesseVs equation will be denoted by a Roman
letter instead of a Greek one in any formula which holds only
for
integral values of that parameter.

12.11.
If

The Bessel functions

we

write z 2

Bessel's equation

4x and denote by # the operator xdjdx,

becomes
(&

This equation

\v

)w-{-xw

satisfied formally

is

0.

by the

series

CO

X<x

2 cr&>

r=0

provided that a

is a root of the indicial equation a2


\v2
the coefficients cr being connected by the recurrence formula

{(a+ff-i^K+Cr-i

Now

if

we take

\v,

is satisfied

is

0,

M-n

r!r>+H-l)'

independent of

Hence

r.

w
is

+cr _ 1

by taking
cr

where

0.

the recurrence formula becomes

(v+r)7-cr

which

0,

Z,r!r(v+r+l)
=6

equation in question. But since this


converges uniformly and absolutely in every circle
R, the formal processes by which this solution was ob-

series solution of the

infinite series
\x\

<

tained are valid

all

over the x-plane.

^i

and, similarly,

where

z2

1xY
r\ T{v+r+l)

w
Ax, are

Bx-^

two

We have thus shown that

y
Z^rWi-v+r+l)'
(

x)r

solutions, not necessarily independent,

of Bessel's differential equation of order

v.

BESSEL FUNCTIONS

We now
v to

315

define the Bessel function of the

be the function

first

kind of order

Jv (z)-( 2 z)

Zr\r(v+r+l)'

r=

where
value.

{\z) v

exp(vlog^z),

The function

the logarithm having

(\z)~ v Jv (z) is

its

principal

thus an integral function.

In this notation, the expression

w
is

= AJ (z)+BJ_
v

v (z)

a solution of Bessel' s differential equation depending on two


it remains to determine whether it is the

arbitrary constants;

complete solution or not. In the first place, it is obviously not


the complete solution when v
0, since it then reduces to a
constant multiple of J (z). Again, if n is a positive integer,

we have

,.,,.
{\)
^
2< rlFl n+r+1)
,

J-n( z )

(fc)~

{\zf

r=0

= (W-2h r(-+r+l)

when t is a negative integer or zero. The


we have just obtained is therefore not the complete
solution when v is an integer, since it then reduces to a constant
multiple of Jv {z). The nature of the complete solution in this
since 1/T(i) vanishes

solution

case accordingly requires further investigation.


When v is not an integer each coefficient in the expansions

of Jv (z) and J- V (z)

J- V (z)

is finite

and non-zero. The functions Jv (z) and

are, in this case, evidently linearly

independent solutions

of Bessel's differential equation.

Example. Prove

that the Wronskian of


A(J,J_)

J {z) and
v

J_(z) is

ttz

By 10.11,
Now when

Ex.
\z\ is

2,

we have

small,

A(^,,

J_ v )

C/z,

where

C is

a constant.

BESSEL FUNCTIONS

316

Hence
A(j j_>

j,( 2 )j:,,( Z )-j;(2)jl,( Z )

r(i-v)r(v)JU

[r(i+v)r(-i.)

Comparing the two expressions

for the Wronskian,

A,J_ r =

we

+ 0(z) J

find that

2sinv7T

This result shows that


not an integer or zero.

12.12.

Jv (z) and

77Z

<7_,,(z)

The recurrence formulae

As Jv (z)

is

are linearly independent

for

if v is

Jv (z)

the limiting form of a certain hypergeometric func-

we should naturally expect that Gauss's formulae connecting contiguous hypergeometric functions would provide, in
tion,

the limit, recurrence formulae connecting Bessel functions. This


is indeed the case, but it is much simpler to prove these recurrence formulae directly from the definition of

The

first

Jv (z).

recurrence formula states that

Jv -1 (z)+Jv+1 (z)

= ~JM-

For

(i)

^-i(2)+^+i(2)

(wiV
(2Z)
r!r(v+r) +
Z
(~^

2 r
)

(-i*

2 r
)

L^^r^+r)

{iz)

lW)
r_

+2

Zr!i>4-r4-2)J

(Hf>+^)-(r-l)!r(v4-r+l)|

(_ ^2)
J

.'ir(v+r+l)

the reordering of the terms of the two infinite series being


on account of their absolute convergence.

justifiable

BESSEL FUNCTIONS

we

Similarly

~
= ^ Y [W) +
1

and

* }

317

see that

\r\

T(v+r)

2jrir(v+r+l)l

+ (r-l)! r>+r+l)J(~l)

4J.'

^-i( 2 ) Jv+i( z )

so

2JV
'

Z ),

where the accent denotes differentiation with respect to

From

(i)

and

u)
z.

the remaining recurrence formulae

(ii),

-Jv {z)+J'v (z)

Jv _ x {z),

(iii)

Jv+1 (z),

(iv)

-Jv {z)J'v {z)

and

may

be easily obtained.

Example

Prove by induction that, when

(T =
Example

2.

m is an integer,

(m+2n)(m+n 1)!
.(*)

m is any positive integer,


~J {z>J (z)} = zJ _m (z),
{z JAz)) = (-^""^-Wz)-

Show

(d

that, if
\

~''

izizY

Example 3. Prove that, if v is real, there lies precisely one zero of


z~"Jv+1 (z) between any two consecutive real zeros of z~ v Jv (z).
Example

4.

Show

Example

5.

Prove that

J* (Z)

that

CI)

ShlZ

J- i(Z) =

'

(S

C aZ

Deduce the values of J %(z).

Example

6.

<-j3)

Show

that, if a

f3

and

> 1,

ftJAotVm dt = iBJ^oa)*^-^^)*^),
a:

2a 2

t{J(ocb)Y dt

2
(

a -^){J,((tf)}H{^)

(LOMMEL.)

BESSEL FUNCTIONS

318

Example 7. Prove
of Jy (x), then

> 1

that, if v

and a and

f3

are distinct zeros

tJv (ctt)Jv (f3t)

dt

0,

o
l

^ t{JMW dt

l{Jv+1 (a)}K

Hence show that, when v > 1, the zeros of Jv (x), apart from x
all real and distinct. (Lommei.)
12.13. Schlafli's contour integral for
If

we

0,

are

Jv (z)

use Hankel' s formula


(0+)

00

where

|arg<|

77

on the path of integration, we obtain

'V

We

consider, therefore, the function

00

obtained by inverting the order of integration and summation


in this last expression.

By

5.51, we find that F(z) is an


whose derivatives of all orders can be obtained
by differentiating under the sign of integration. It follows that

applying the theorem of

integral function

the Taylor expansion of F(z)


f(z)

is

2 <-*r-^

***

(\z)- v Jv (z) are integral functions with the same


Taylor expansion and so are identical. Hence we have

Thus F(z) and

BESSEL FUNCTIONS
where

^ v on the path of integration.

|arg|

319

This result

is

due

to Sehlafli.t

The Bessel functions of integral order


When n is a positive or negative integer or zero,

12.2.

integral

Schlafli's

(0+)

exp/i
Ju*>-<*fpH3*-*

has as integrand a one-valued analytic function of t whose only


singularity of finite affix is an essential singularity at the origin.
Hence we can deform the path of integration, without altering
the value of the integral, until it consists of the circle \t\
provided that z is not zero. Making the substitution

we

= \\z\R,
\zu,
t

find that this gives

Jn( z

exp{z( tt-i)}--i du,

2^1}
o

where

denotes the circle

holds also

when C

is

\u\

This formula evidently

_R.

any simple closed contour

encircling the

origin.

In particular,
Bi
where

if

we take

<
Jn

=e

<

jt

77,

for

we

the

circle \u\

1,

on which

obtain the formula

IT

(Z )

2-rr

-in0+izBine JQ
f e
J

IT
TT

and hence
This

is

Jn (z)

/g inB+izsind igind izsm$\ JO

Bessel's| integral for

Example

Show
Jn( z )

t Ann. di Mat.

(2),

C os(nd zsind) d6.

Jn (z).

that

= 77~

cos(zcosd%mr)cosn0d9.
J

5 (1873), 204.

$ Berliner Abh. (1824), 1-24.

BESSEL FUNCTIONS

320

Deduce, by the use of Jacobi's lemma, Poisson's integral formula

where n

>

0.

Example

Show from

2.

the series definition that


n

T^wm j -(--^-^

provided that Rl(v+)

Example

3.

>

0.

Prove that,

if z

and

x-\-iy

than \,

if

is

real

and not

less

lizl-e"

Example

4.

Show

that

cos(v# zsin#) dd
satisfies Bessel's

Example

5.

equation of order v only

Show

L\

where the path of integration

12.21.

We

when

v is

an

integer.

that

is

e zsinh(-n( rf^

any curve joining the points

The generating function

-^-ni.

Jn {z)

for

have just proved that

Jv

= iif e*

du
U n +V

-!/)_

C denotes any circle with centre at the origin. Since the


only singularity of the integrand is at the origin, it follows
immediately from Laurent's theorem ( 4.52) that
where

oo

e hz(u-llu)

= 2 UnJn

(z),

GO

the series converging uniformly with respect to


it

lies in

any annulus

<R ^
1

\u\

bounded closed domain.


The functions Jn (z) are therefore the

^ R2

and

u and

when

z lies in

coefficients in the

any

expan-

BESSEL FUNCTIONS

321

sion of the generating function e^"- 1/") as a Laurent series


in u,
and, for this reason, are sometimes called the Bessel
coefficients.

This expansion was, in fact, Schlomilch'sf definition of the


Bessel coefficients, and from it he derived many of their principal properties.

Example 1. Show by means of Schlomilch's generating function


that the Bessel coefficients satisfy the recurrence formulae of
12.12,
and hence that they are solutions of Bessel's differential equation.
Show

also that

Example
efesine

J_ n (z)

= Jn - z) = - 1
(

nJ

n (z).

2. Prove that

rn 4QAG

= Jo{z) + 2 | JM (z)cos2n6+2i f Jsn+1 (z)sm(2n +


1

and deduce the Fourier

Example

3.

series for cos(zsin0)

and

sin(zsin0).

m
(Jacobi.)

Prove that

= 2| (2n+ 1)^(2),
izcosz = J (z)-9J3 (z) + 25J(z)-...,
izsiaz = 4J (z)-16J4 (z) + 36J,(z)-....
z

(Lommel.)

Example

4.

Show

that
00

Jn (z + z')

= 2 Jm z )Jn-m(.z')
(

(K.

NEUMANN.)

00

12.3.

The

solution of Bessel's equation

by complex

in-

tegrals

We have already seen


equation

12.2,

Ex.

5) that Bessel's differential

,,

rri

is satisfied

by

ezslnht

- vt

dt

iri

when

v is

an

integer.

We now

propose to determine whether

there exist solutions of the form


6

sainht - vt

f e

dt

for general values of

v.

t Zeitsdirift fur Math,


4111

und Phys. 2

(1857), 137-65.

BESSEL FUNCTIONS

322

integral funcIt should be observed that the integrand is an


particular
the
of
independent
is
integral
tion of t, so that the
It
is, moreand
6.
a
points
the
path of integration which joins

over, useless to consider integrals of this type

round closed

contours since they all vanish identically.


When the limits a and 6 are finite, we can differentiate
under the sign of integration to obtain

d2 w

dw

>.

dz

dz*

*stah<-W( z2 cos h2-|e

ZSmh V2 )dt

This result also holds when one or both of the limits are infinite
provided that the integral for w converges uniformly. If we can
choose a and b so that this last expression vanishes, w is a solution of Bessel's equation.

When the

wi, we have

limits are

2Sinh '- '<(zcosh+v)]"

[e

7ri

2i{zv)sinvTT.

ni

w=

Hence

e**^-" dt
J"
iri

is

a solution of Bessel's equation

if

and only

if v is

an integer

or zero.

not an integer or zero, the appropriate values of


a and b are given by the considerations which follow.
Let us consider now a path of integration consisting of the
<x, the imaginary axis
negative part of the straight line Imt

When

v is

Im =

+oo+
pi in the
The limits of integration are then ao+oci and
usual notation. We suppose, moreover, that z = re lies in a
bounded closed domain D for which ^7r+a+S < 8 < \tt-\-ol S,
from

oci

to

/3t,

and the

positive part of the straight line

ei

where

8 is

When R

an arbitrary positive number.


is positive, we have

Rl{zsinh( R+ai)}

^re- Rcos(d+a.)

lre Rcos(eoc) <

\rer R

Jre^sinS.

fi.

BESSEL FUNCTIONS
If v

[gs Blah

<-W(2 cogh

<

<+ v )

[e*

R -> co,

when

<= _ B+af

(rcoshi?+ |v|)exp(|re- Jre^sinS+A^+jua).

Since by hypothesis r

as

323

= X+i/j,, it follows that

bounded

is

8inlu - w

in

D, we deduce that

(zcosbJ+v)]<= _

R+a

-*

the convergence being uniform with respect to


z
D. In a similar manner we can show that

2 lies in

[e* slnh

'- w

(zcosh*+v)]

<=B+iJ

as .B->oo, provided that p =


a the convergence being
uniform with respect to z when z lies in D.
Lastly, it can be shown without much difficulty that,
under

these conditions, the integral

oo+ai

converges uniformly with respect to

z when z lies in D. Hence


an analytic function, regular in D, whose derivatives can
be found by differentiation under the sign of integration, and

is

w satisfies all the conditions necessary to


of Bessel's equation of order v.
We have thus shown that the functions

so

+ (w- a )t

fy

-=o + ai

a solution

ee S inhl-vl^ t

-oo + ai

are solutions of Bessel's equation of order


in the angle

it

oo-(7r+a)i

gzainh l-vl

make

^w+a+S < argz <

arbitrary positive number.


paths of integration can be

It

v,

valid

when

^4- a _s, where

z lies
is

an

evident, moreover, that the


any contours which are asymptotic
to the particular contours defined above.

12.31.

is

The Hankel functions

In particular, we

find,

by taking a

Trt

0,

that

BESSEL FUNCTIONS

32 4

iri

oo

-X

Hf\z)=--

e*^-"

dt,

valid when
are solutions of Bessel's equation of order v,
functions of
Bessel
\-n. These functions are called the
|argz|
of order v.
the third hind, or, more briefly, the Hankel functions^
ranges
other
for
The analytical continuation of these functions

<

of values of arg z

is

provided by the formulae


oo

+ (7r a)i
J

co + ai
00 (7T + 0j)t

TTl

ao + ai

where -|7r+a+S <

<

argz

proved

ment by which this is


the case when a. = \tt.
H^\z) is defined when

177+a-S and

may be

|argz|
co

> 0.

\tt,

The arguby considering

whilst

+ iiri
f

TTl

<

illustrated

e*

einht - vl dt

0O + J7U

<

<

ir. We have
argz
a solution of Bessel's equation when
in the first
identical
are
functions
two
these
that
show
to
definition.
of
region
common
their
quadrant,
\rt 8. If we apply
argz
8
Let us suppose that

is

0< <

<

Cauchy's theorem to the rectangle with vertices \ni, B-\-\ni,


M+Tri, tti, we find that
R+Ti
R+im
R+m
zsinhl - vl dt
f
zsmht-vtdt=
ezsinbt vt dt.
e
e
R+ini
im
lm

But, on account of our restrictions on argz, the second integral


on the right-hand side tends to zero as B - oo, and so
f
irri

zetnh.t-vt

fa

ez sinh l-vt fa.

iwi

Hankel who first realized their importance. The


t So called because it was
definition adopted here is not the usual one, but is essentially equivalent to
1-16.
that of Hopf and Sommerfeld, Archiv der Math, und Phys. (3), 18 (1911),

BESSEL FUNCTIONS
Similarly

we can show
I"

ez sinh '-"' dt

00

Hence, when

the

first

+ Tri
(*

ez sinh

'~

CO + $TTi

z lies in
co

325

that

ezsinhl

- vl

dt

co

quadrant,
co + $ttI

e s:s,nh

'- w

df,

co \iri

which is the required result.


In this way the Hankel functions are continued analytically
all over the z-plane. It will, however, be observed that the
functions so defined are not one-valued, but possess a branchpoint at the origin, even in the case when v is an integer.

Example 1. Show that H^l(z), fll2 >(z) are also solutions of Bessel's
equation of order v by proving that
H%(z)

eH{z),

H1\(z)

erHf\z).

Example 2. Prove that the Hankel functions H(z), Hf{z) satisfy


the same recurrence formulae as Jv (z).
Example

3.

Show

that $

H(ze?*)

Hf^ze)

=
=

-e-miH?>(z),
2cosvrrH^(z) + e' 7'iH^(z).

Hence prove that


fl(ze2 )

Hf\ze*)

=
=

-i?< 1, (z)-2cosv7re-'"rffff(z),
(4cos 2 v7r-l)H(z)

+ 2cosnre""' Hi
i

1)

(z).

(Wbbeb; Gbaj\)
12.32.

The connexion between

the Bessel and Hankel

functions

We

shall

now show

that the Hankel and Bessel functions are

connected by the relations

= H?>(z)+H?\z),
= e^ HW(z)+e-^ Hi \z),
<

2Jv (z)

2J_ v (z)
provided that |argz|

it.

t See Ex. 3.
Jw
j It is simplest to prove these formulae in the first instance when arg z
arg(ze"*)
in this case \n
fir. The formulae for general values of z follow
by analytical continuation.
This restriction is needed since the Bessel functions are not defined outside
this range of values of arg z. The formulae we are about to prove may be used
|

<

<

to extend their range of definition.

<

BESSEL FUNCTIONS

326

If z

re

8i
,

where

w < <

section give

w, the

formulae of the previous

.+<*-*

H<?\z)

dt,

J
-co

#<,2) (z)

exp(re 04 sinh-v<)

771

+ Oi

TTl

exT>(re ei smhtvt) dt.

-x + Oi
^-h+ih

Hence we have

\{H\z)+H<?\z)}

exV (reeisinht-vt)

[
co

dt,

di ni

the path of integration being the positive part of the line


Imi
tt8, the imaginary axis from
(n+d)i to (TT6)i,
and the positive part of the line Im t
tt6. If we make, in

turn, the

two substitutions
00

$[H( Z )+H(z)}

= -L

= u0i,

ew

we obtain

2v/r,

+ TTl

e^{reei sinh(u-ei)-vu+vei} du

2tti
00

7TI
oo + jrt

If ^

=
2-rn

29i
exp{Are lre
e- u }
2

_
_

7TI

(0+ >
C

2ni

(0

But

Jv (z)

From

this

J_(a)

is

r v e v6i

e vu

dv

2 v v v ~v

4i;

exp{t>

vBi
ev6i

-rr(izr

this last expression

fore

_r e

2 20i

Pj
00

'

CO

du

Schlafli's integral for

-.

7(z),

and there-

{J?<( z )+#t>(*)}.

we deduce that
4{H i>F (z)+fi<!),()}
(

a result which reduces to J_ n (z)

i{e-^)( 2 )+e *#<,2 >(z)},

( l)V(z) when n

is

an

integer.

12.33.

The complete

solution of Bessel's equation


Although the Hankel functions are merely linear combinations of the functions J(z) and J_(z) when v is not an integer,

BESSEL FUNCTIONS

327

remarkable property of
they possess, as we
being linearly independent solutions of Bessel's equation for all
values of the parameter v, so that the complete solution is
shall prove, the

= AH<?\z)+BH<?\z),

where A and B are arbitrary constants.


For when v is not an integer, the Wronskianf of Jv (z) and
J-V (z) is given by
A(J, J_)

= J (z)JL (z)-J'v(z)J-M
i4 {H^z)+H^(z)}^{e^HW(z)+e-^H^(z)}
v

-{e^m^\z)+e-^m^\z)}^H^\z)+H^\z)}\^

and

by the example of

so,

12.11,

But, for any fixed value of z, the Hankel functions are integral
functions of v, and so, by continuity, the Wronskian of H(z)
and v2) (z) is 4/(7nz) for all values of v. Since the Wronskian
never vanishes, H'^ {z) and H^{z) are linearly independent solu-

tions of Bessel's equation.

12.4.

is

The Bessel

function of the second kind

In view of the fact that the behaviour of Jv {z) near the origin
much simpler than that of the Hankel functions, it is very

desirable to retain it as one of the standard solutions of Bessel's

equation.
defined

As a second

solution

by the equation

Yy (z)

= {H (z)-H (z)},
v

for all values of the parameter

Jv (z)
we

we take Weber's J function Yv (z),

v.

Remembering that

h{Hl(z)+Hi(z)},

see that the linear independence of

t See

10.11,

Exx. 1, 2.
Math. 76

Jv (z) and Yv (z)

follows

(1873), 9; Math. Ann. 6 (1873), 148. There are


several other definitions of a second solution of Bessel's equation, the present
one being the most satisfactory. See Watson, Bessel Functions, 3.5 et seq.
X

Journal

fiir

BESSEL FUNCTIONS
from that of the Hankel functions. Yv (z)
328

is usually called the


Bessel function of the second kind. Evidently
v (z) satisfies the
same recurrence formulae as Jv {z).

When

v is

not an integer,

12.32 that

follows from the formulae of

it

coaJTlz

Y (~)

v(

T
)J-v(z)

sinvrr

When

v is an integer, this formula fails because the numerator


and denominator are then both zero. But since the Hankel
functions are continuous functions of v, so also is Y (z). Hence
v
the value of Weber's function of integral order is given by the

equation

_
T
= ]im COSV7rJ - J^ zl

If

we

The

Sin VTT

v-vn

12.41.

" {z)

Yn (z)

Yn (z)

series for

substitute in the formula

Y(z)
*

= COSV7tJ

v( z

) j-v(z)

sin vtt

the power series for the Bessel functions,

we obtain the expan-

sion

V -r }7f>
TM = cot (W Z^r\T(v+r+l)
vtt

r=0

cosec

vtt

(^-"2,r \r(-v+r+l)
r

vahd when

_,

not an integer. This formula

v is

fails

when

v is

an

integer.

Now by

the ordinary rule for finding the limit of a quotient,

we have

v
Yn (z)
,

J_(z)
= hm cos vttJJz)
ti-i
?P-,.

SU1V7T

v-*n

= hm

cos vtt

v
-

= Ilim (^_
TT

v-n

dv

when n

is

._
(

sin vtt Jv

-=?

dv )l

{tt

cos vtt}'

)^
dv

l) n

(
n (z),
positive or zero.

Moreover, since Y_ n (z)


the case

tt

ov

v-*n\

it suffices

to consider only

BESSEL FUNCTIONS
Denoting the derivative of log
{iz)V

dv

T(t)

by

329

<p(t),

we

find that

2 r!r(v;r+l) ^ ^-^+ r+1)


* -^ +r+1)}
2 r47^Sy!
r!(+r)!'
(

>

r=0

* Z)W

{l0g{ 2)

r=0

as v -> n.

When n is positive, we have to consider separately the first


n terms of the series for 8J_J8v since T( v+r+1) and
ifj( v+r+1) have poles at v = n when r takes the values
0, 1, 2,..., Ti1.
This difficulty does not arise when n is zero.
When r = 0,1, 2,..., n 1, we have
8_

( l) r (Jz)-"+2
(%z)- v+2r

'--

r!T( v+r+l)J
p( v

_^

8v

r v - r
(

Ttr\

[sin V7T log \z-\-tt cos V7r+sin vct i/r(v r)]

( l)(i-z)-"+

\l z) - V+ 2r sin

( r 1)!

2,-5

r!

as v ->

7i.
Treating the terms for which r
forward way, we see that

_i).^^"ffcr=l)l( W
ov

^-^

rl

2=

- 1)s

ti-

in the straight-

^+
iily!^ +1)

- log( *s)}

as v -> n. Hence,

when n

can be expanded as a

Yn

Z)

is

a positive integer, Weber's function

series of the

^ 2 ^feS^

2io

form

g^)-^+ )-^+ r +
i

m-1

1 )}-

(nriy.

r=0

This formula also holds when n


0, the terms in the second
line being omitted. It is sometimes useful to modify this formula

by

writing

#1)

=-y,

#+l)=l + i-|-i + + I-y,

where y denotes Euler's constant.

...

BESSEL FUNCTIONS

330

The asymptotic expansions of the Bessel functions


Although we have now represented the complete solution of

12.5.

Bessel's equation

some

in

cases

by

the convergence

by

is,

ascending powers of z (multiplied


which are convergent for all values of z,

series of

logz),

unfortunately, so slow

when

\z\

is

large that

an examination of the initial terms of such a series gives


practically no information regarding the value of its sum. To
overcome this difficulty, we now propose to determine, by means
of the method of steepest descents,! expansions of the Bessel
functions in inverse powers of z which are asymptotic in Poincare's sense, even though they are divergent for all values of z.
The method of steepest descents, J which was devised by
Debye to obtain asymptotic expansions of Bessel functions of
large order, consists essentially in representing the function

under consideration by an integral along a path with a particular


geometrical property.

Let us suppose that we wish to find the asymptotic behaviour,


\z\ is large, of a function defined by a contour integral of

when

the form

e?fWF{w)dw

which converges when Rlz is positive, the functions f(w) and


F(w) being analytic functions of w, regular in a region of the
w-plane which contains the path of integration. The path of
integration

is

to be deformed, if possible, until the following

conditions are satisfied:


(i)

(h)

To

the path of integration goes through a zero


the imaginary part of f(w)

is

obtain a geometrical picture of this,

face defined

off'(w);

we

consider the sur-

by the equations

w=
in a space in

constant on the path.

which

(u, v,

u-\-iv,
t)

R1/(m>)

are rectangular Cartesian coordinates.

For an account of other methods of determining these asymptotic expanWatson, Bessel Functions, Chap. VII. The only application of the
method of steepest descents to the present problem seems to be that of Meijer
(Proc. Kon. Akad. Wet. Amsterdam, 35 (1932), 657-67, 852-66), although
Sommerfeld and Hopf (loc. cit., p. 324) used it to find the dominant term.
} Sometimes called the saddle-point method.
Math. Ann. 67 (1909), 535-58; Miinchener Sitzungsberichte (5), 40 (1910).
|

sions, see

BESSEL FUNCTIONS
331
The height of a point on this surface above the plane t = can
never be a true maximum or minimum, since t satisfies the
partial differential equation

dH
8u2

The

dH
+ 8v* _
~

'

may, however, possess points at which the tangent


t = 0, and these are necessarily saddle
points. At such a saddle-point, dt/du and dtjdv both vanish, and
hence, by the Cauchy-Riemann conditions, f'{w) also vanishes
surface

plane

is

parallel to

there.

If

we

construct a

means of contour

map

lines,

of this surface on the

the curves of this

(u,

v)-plane

map on which

by
the

imaginary part of f(w) is constant are everywhere orthogonal


to the contour lines and so are the projections of the paths of
steepest ascent or descent

on the

surface.

Let us suppose that it is possible to modify the path of


integration in this way without altering the value of the integral
and so to express the function as a sum of integrals along paths
of steepest descent starting at a saddle -point w
Integrals along
paths of steepest ascent cannot occur as they evidently diverge.
On a path of steepest descent, f(w) f(w )r, where r is real
and positive and increases indefinitely as we move away from
the saddle-point. We have,therefore, to consider the behaviour
of a sum of integrals of the form
.

00

when

\z\

is large.

To each of these we can, in general, apply


lemma ( 9.52), and so deduce the re-

the result of Watson's

quired asymptotic expansion.

be

This brief explanation of the method of steepest descents will


made clearer by a consideration of the following section.

12.51. The asymptotic expansions of the


tions

Hankel func-

In the present section we find an asymptotic expansion of


the Hankel function Hf^z) by applying the method of steepest

BESSEL FUNCTIONS

332

descents to the integral

jjW{z)
v

e zalnht

iri

~ vt

dt

co+ai

where

|arg(ze _ai

<

)!

\n.

From this

the asymptotic formulae for

the other solutions of Bessel's equations are easily deduced.


and
The saddle-points are given by the equation coshtf

so are

where n

(n-{-\)Tri,

on the path of integration

lies

nie ivni Hf\z)

the substitution

an

is

integer.

Of

these,

If

for all values of a.

\-rri

we make

\m-\-w, we find that


0O-t-i(7T oc)i
exi>{ze i7Ti cosh.wvw}dw

|(ir cc)i
co + J(w a)i

co

exp{ze i,Ti cosh w} cosh vw dw.

2
J

To

bring this formula into agreement with the theory of the


e ai so that |arg|
\tt. This
we write z

previous section,
gives

<

co+( W -/3)i
v ni
'

Trie^

H^
l

xi

{le'

exp^e^'coshw^coshi/wdw;,

ol-\-\tt.
where /?
We have now to see whether there exists
a path of steepest descent which starts at the origin and goes
to infinity in the required direction.
2tt. The steepest paths through
Let us suppose that
/3
the origin are given by the equation

< <

e^coshw
where r

is real.

If

we

e w __

efi

t,

we find that
i_ Te -#_( T2 e -2#_2Te-#)i/2,
solve for w,

so that argw>- |-(7r ,8) as r->+0 and argw->^j8 as


t ->
0. There are therefore four steepest paths through w
0,

the paths of steepest descent being given


as t -* +oo,

eM

by

^ 0.

Moreover,

^ + a^^^i
j

so w -> oo+(7r fi)i or oo-(77 fi)i. Hence there is a


path of steepest descent which leaves the origin in the direc-

and

tion

&rgw

2(77 /J)

and tends

to infinity in the direction


333
BESSEL FUNCTIONS
oo-f-(TT /3)i; we can take this as path of integration without
altering the value of the integral. Hence we have

irie*

H y{te

vvi

ai
)

dw

e-^coshvw -^

2exp(e^)

"T
d-

o
00

= -exp(eP

e~^ T

J
)

In order to apply Watson's lemma to

show that

first

-=-

this integral,

dsmhvw

regarded as a function of the complex variable t,


conditions of the lemma; these are as follows:

(ii)

satisfies all

the

regular, save possibly for a branch-point at the


a+8, where a and 8 are positive;
origin, when |t|

F(t)

is

\F(t)
I

<
< KebT holds when t > a, K and b being positive

numbers independent of
(hi)

we must

= -ST-'

E{r)

(i)

dr

.F(t)

can be expanded as a
F(r)

valid

when

|t|

r;

series of the

form

= f am T<-*
m=l

< a, r being positive.

Now, from the equation


e^coshw

we

eP l

= ew -^V(|-r).
positive on the path
as t ^ +0, V(+t)

sinh \w

find that

Since arg >,->- \{tt j8)

is

of integration. Moreover, sinh^w is an analytic function of r,


regular save for the branch-point at the origin.
On the other hand, sinh vw is an analytic function of sinh \w,
regular

when

From

|sinh;|

<

1.

Its Taylor expansion is

=!
r=l

this it follows that

sinhvw

is

an analytic function of

t,

BESSEL FUNCTIONS

334

regular, apart

from a branch-point at

0,

when

|t|

<

2.

Also,

for such values of r the expansion

y (4v2-rw-

X
is

-I

This implies that, apart from a branch-point at the


when |t|
2 and can then be represented

valid.

<

origin, F(r) is regular

bv the

3 2),.,(4,*-(2,-i)2)

series

F(r)

V(2t)

Hence conditions

when r

Finally,

2-(2r)!

(i)

is

and

of Watson's

(iii)

and

large

^ l-\-2rd

ew

lemma

are satisfied.

positive,
7r

-P> i

*?~.

and

dr

It follows that there exists a positive

number

t x with the pro-

perty that
\e

when

>t

<

3t,

But

x.

<

|e-|

1/r,

|dto/cTr|

<

2/r

since

= vcoshvw

F(t)

ar
this implies that there exists
\F(r)\

when t
dition

^ Tl

(ii)

lemma

is

constant

such that

< At"- < Ae


1

Since \F(t)\

of the

a-

is

bounded when

<r<r

x,

con-

also satisfied.

By the formal term-by- term integration whose validity is a


consequence of Watson's lemma, we now deduce that, when
||

is

large

and
7ri&

|arg|

v"i

<

\tt,

H<?-\t,e ai )

2
--=

-exp(e#)

e-^F(r) dr

BESSEL FUNCTIONS

335

~ /?V' exp{e^+^-m[r() +
(4^-P)(4^-3^)...(4^-(2r--l)^) /e<^)Y
^tv.,
(+i)
1 ,

{2rJ\

r=l

V
JU
x
Z,
"^

(4v 2

,
*

-l 2 )(4v2 -3

)...(4v 2

2r

r! (2e

r=l

-(2r-l) 2 )

ai r
)

In obtaining the expression in the last line we have used the


duplication formula for the Gamma function.
If

we

restore the original variable

(O

we

find that

\l/2

\
X

is

z,

exip{i(z-ivTT-iTr)}x

2,

2^r\V2izf

the asymptotic expansion of H^\z), valid

and

|arg(ze _orf )|

<

when

\z\

is

large

\n.

<

ai
\tt,
In proving this formula we assumed that !arg(ze- )|
a+%n must he
where the only restriction upon a. is that (3
and 2n. If we allow /? to vary over this range, we
between
asymptotic expansion is valid when \z\ is large,
the
find that
2tt.
argz
that
tt
only
provided
argz
2tt and \z\ is
that,
when
w
shown
We have now
asymptotically
by
represented
is
H^\z)
large, the Hankel function

<

<

<

the series

H<(z)

where

~ (A)

M=

<

^
^-1')^-^^-^-!^

1/2

expz-W-^)}[l+

,(V,r)

(2iz) r

This result, for general complex values of v and z, was first


obtained by Hankel.f
Since the Hankel functions are connected by the relation
#<2>( Z )
it

= e^H^ze^),

follows immediately from the asymptotic expansion of H<P(z)


t Math.

Ann.

1 (1869), 491-5.

336

that,

BESSEL FUNCTIONS
when 2tt< argz < it and \z\ is

large,

H (z)
(

has the

asymptotic expansion

The asymptotic expansions of the Hankel functions for other


ranges of values of argz may be deduced from the formulae of
Ex. 3.
Although these asymptotic expansions are in the form of
divergent series, they may be used for computing the value of
the Hankel functions, since the error incurred by terminating

12.31,

the series at any point

is of the same order of magnitude as the


term omitted. More precise information on the magnitude
of the error will be found in 7.3 of Watson's treatise and also

first

in the series of papers

Example

by Meijer already

cited.

Prove that the formulae

r=0

hold for large values of

Example

\z\

provided that |argz|

<

-n.

Show that the asymptotic expansions for the Hankel


functions terminate when v is half of an odd integer. Deduce that, if
v = n-\- where n is a positive integer or zero,

JM =
"v

'

2.

(iy
\ttz)

/2

[ sm(z
v

_ 2imr)

'

(-P'("+2r)i

jLt (2r)i(n-2r)l(2z) 2r

(n-l)/2

+ cos( Z -in,T

(-iy(n+ar+l)l

(2r+l !(ro-2r-l)!(2;
2r l)!(22) 2r+ 1
Z, (2r+l)!(n

<m/2

\m!

^*

v
I

'

i
-sMz+inrr)
,

Z,

V
^

(2r)!(n-2r)!(2z)

-l'n+2r+l)!
(2r+l)l(n-2r-l)!(2,)n

J-

BESSEL FUNCTIONS
12.6.

337

The Neumann polynomials

We now propose to investigate the properties of certain polynomials associated with the Bessel coefficients which are of
importance in the theory of the expansion of an arbitrary
analytic function /(z) as a series of the form

= 2 aA(z

/(z)

These are the


expansion

Neumannf polynomials On (t), denned by

the

J (z)O

2 Jn {z)On

(t)+2

We

)-

{t).

71=1

from the formula

start

00

(t z)-1

f e-<l

- sinhw
coshw
!l

'>

dw

which

valid

is

when

~R\(t z)

> 0,

and use Schlomilch's gene-

rating function in the form

= jo(z)+ ^
n=l

e*sinh,

{ e'+(-l)e-'}Jre (z).

This gives
00

(tz)- 1

f e~' Bi* hw

2 {enw +( l) ne~ nw}Jn {z)\ dw

coshw\j {z)+

n=1

00

=J

Q {z)

e- tBtahwcosh

w dw

+ 2 Jni z
provided that it
and summation.

is

?
)

e~ teblbwco8h.w{e nw +(l) n e- nw} dw,

permissible to invert the order of integration

sufficient condition for the validity of this process is the

convergence of the

series

\e- tBinhw coshw{e nw

+{l) ne-nw}JJz)\dw.

t K. Neumann, Journal fur Math. 67 (1867), 310. The Neumann polynomial


was formerly called the Bessel function of the second kind, by analogy with
the corresponding formula in the theory of the Legendre polynomials. But the

terminology
4111

is

misleading as

On (t)

is

not a solution of Bessel's equation.


Z

BESSEL FUNCTIONS
and Im z = y, we have

338

But

if

Rl t

=t>

00

f \e~lsi]lhw co8hw{e nw -\-(l) n e-

nw}J

dw

n (z)\

o
00

o
00

<

4|J (z)|e* T

c-i TCV't + 1 'w

dw

00

= 4|J(z)|eiT (2/r)' +%!


< 8\z\ n elT +v/rn+1
l

by

12.2,

\Z\

< T.

Ex.

The interchange

3.

We have thus shown that,


{t- z )-i

is

thus certainly valid

when Hit

= J (z)O

>

when

\z\,

(t)+22 U*)on {t)


=i

where

OJt)
n

i*

e-' 8lnhwcosh;

J
o

cosil
cosh
.

smh

nw dw,

the upper or lower function being taken according as

even

is

or odd.

The functions
Rl

> 0. We

n (t) have, so far, been denned only


their region of definition and

now extend

that the expansion of (t z) _1


condition

<

\z\

Now, when n
cos

^nw

sinh

is

when
show

valid under the less restrictive

\t\..

> 0, we know thatf


2 n -4smhvw
\

+
+ 2(2n
T
ff*~ *\ sinh"- w ^
2)
a

n(n-l)(n-2)(n-3)
2.4(2 2)(2 4)

^J
J

the upper or lower function being taken according as n is even


sinhw, we find that,
or odd. If we make the substitution u

f See Hobson, Plane Trigonometry (1911), 105.

BESSEL FUNCTIONS

>

when n

339

> 0,

and Bit
00

n(n-l)(n-2)(-3)
and

'

so
/0

2 "- lw!

m-

*
f!

*+*

2
*

~'~2(2n-2)

+ 2.4.(2n-2)(2n-4J~ K

the series terminating just before a zero or negative


term would
occur in the denominator. On the other hand,
00

(t)

8lnhw cosh
f c-'

w dw

-.
'

We now

mean

define OJt) to

these polynomials in

l/t,

for all

values of arg t.

From

we

this definition

\n(t)\

so that,

by

12.2,

Ex.

see that,

when n

>

0,

< ~W|*|2),

3,
Z n

< -^^e^dW+Imz).
I

\Jn {*)O n {t)\

Hence, when |z| < j? and t lies in a bounded closed domain for
which |*| > R+e, where e is positive, the series
4>(*)0 o (<)+2

| 4(z)O(0

converges absolutely and uniformly with respect to z and t.


its sum is (t-z)- 1 when Rl*
|z|, it follows by analytical
continuation that this will still be the sum when \t\
\z\.

>

Since

We have

>

thus shown that


(-)-!

provided only that

Example

1.

= J (z)O (t)+2fjn (z)On


<

\z\

(t),

\t\.

Prove that the Neumann polynomials are connected

by the recurrence formulae


(-l)0+1W+(+l)O^(f)-?(^-l)OW

0^(0-0,^(0 =
-OiW =

20'n (t)

= |n8inlfMr
(

>

1),

OJ(0.

where accents denote differentiation with respect to

t.

>

1),

BESSEL FUNCTIONS
O n (t) satisfies the differential
3dw /, na -l\

340

Example

2.

Show
d?w

where g n (t) denotes

equation

that

\jt

...

or n/fi according as

even or odd.

is

Example 3. Show that, if C denotes any simple closed contour about


the origin,

Om (z)O n (z)

dz

Jm(z)O n(z)

dz

0,

J*

(m

^ n),

(m

>

c
j"

Jm (z)Om{z) dz =

tri

1).

Neumann's expansion theorem


K. Neumann showed that iff(z) is an analytic function, regular

12.61.

when

\z\

<

B,

it

can be expanded as a
f(z)

=a

series of the

form

J (z)+2fan Jn (z),
i

which

The

converges uniformly in every closed

coefficients

an are given by

an==

domain within

\z\

B.

\z\

B,

the relation

iL /

n{t)

dL

\t\~R

For
we have

if z lies

in a closed

domain within the

circle

K1-.R

= 2^
\t\

J
=R

f^[Jo( Z

o(

t)

+ 2 I Jn(^n(t)}dt.

Since the series occurring under the sign of integration converges uniformly with respect to z and t, we can integrate it

term by term to obtain Neumann's

result.

REFERENCES
H. Bateman,

Partial Differential Equations of Mathematical Physics

(Cambridge, 1932), Chap. VII.


A. Gkay, G. B. Mathews, and T. M. MaoRobeet,
Bessel Functions (London, 1922).

Treatise

on

BESSEL FUNCTIONS
G. N. Watson,

on

Treatise

the

341

Theory of Bessel Functions (Cam-

bridge, 1922).

E. T.

Whittakeb and G. N. Watson, Modern Analysis (Cambridge,

1920), Chap.

XVII.

MISCELLANEOUS EXAMPLES
hm (_

Prove that

1.

n>oo ^

provided that x

Show

2.

is

is

an

integer,

= J(?EyJn+i{z)

'ePn (t)dt

and deduce that

Prove that

j
CO

equal to 2/(2n+l) or
or unequal.
is

Show

4.

that,

00

Vfi)
3.

positive.

when n

that,

I)"W CO s-) = Jm x
nf

ni

when

2 {2n+

1)iMJ
(z)PW+i

(Baubb.)

J^^J^x) dx
X

according as the integers

and n are equal

cosfl is positive,
oo

= AL~
I

P(cos0)

\m

e-zvxej^xsinejx" dx.

(Hobson.)

Prove that the product J^zjJ^z)

5.

W** y
n-0

Show

6.

is

equal to

r^+ v +2n+l)(-zy4r
rtja+v+n+ljrdu+n+ljr^+n + l)'
(ScHxJunx)

that
00

7T

Jv {z)

=-

cos(z sin 0-v0) d0

- ^^

=-

d<,

oo

jr

^(z)

f e-*^*-"*

sin(zsin0-v0)d0--

provided that the real part of z

is

e-^^VHe-^cosra-) d#

positive.

(Schlafli.)

Prove, by term-by -term integration, that the equation

7.

e
J

4<*)f-

d*

2v

^r( v +l)^l~2~'

2' v+1

'

-*)

BESSEL FUNCTIONS
> 0, Rla > and |6| < \a\. Apply

342

holds when Rl(/*-f y)


of analytical continuation to show that the equation

>

less restrictive conditions ~R\(fx-{-v)

Show

8.

the principle
true under the
and 111(0^*6) > 0. (Hankei,.)
is

that

provided that |argp|

<

\tt

and

Prove that, when Rlv

9.

>

Rl(/i+i-)

0.

(Hankel.)

> 1,
c

+ ooi

ooi

the path of integration being the straight line RI(


10.

R1j>

= c>

By using the formula for Jv+1 (tu) given in Ex.

> 1, u >

0, c

>

9,

0.

(Sontne.)

prove that, when

0,

e+ooi

oo

c ooi
e

Deduce that

+1 (m) t&

J,,(f )^,

where

e is

equal to

0, $,

or

according as

is less

than, equal to, or

greater than unity.


11.

Prove that the relafton


00

F(x)

+1

=
j

(jJ

Jv+1 (xtW(t) dt

oo

implies that

0(t)

(?X

+1

Jv+1 {tu)F'(u) du

provided that
integration

we can

differentiate the first integral

and invert the order of a

under the sign of

certain repeated integral.

[This is

the Hankel Transform Theorem. Sufficient conditions for its validity are
given by Burkill, Proc. London Math. Soc. (2), 25 (1926), 513-24. See
also Titchmarsh, Proc. Camb. Phil. Soc. 21 (1922-3), 463-73.
The more usual form of the theorem is obtained by writing
F'(x)

stf+yix),

*'(*)

00

it

then states that

f(x) ==

t</>(t)Jv (xt)

dt

t"+^(t);

BESSEL FUNCTIONS

343

00

implies that

</>(t)

uf(u)Jv (tu)

du."\

12.

Prove that, when the

/a+ 1 and v are positive,

real parts of
2

= ^-

z i+r)/2JAi+|,(2Vz)
((

Hence show

that,

when n

is

an

"/ 2 J(2V'u)(zu)"- 1 dv.

integer,
z

z z

z (M+n)/V

M+ (2Vz)

integration being performed


13.

Show

z W^(2Vs) dz,
Jj"...j*

times. (Sonine.)

that

(z+h)~"H{2^z+h)}

^ ^r^m=

<,'

+m)/2

<W2Vz),

00

*<-/ V,,(2V*),

m=0

when

the latter expansion being valid only


14.

Show

that,

when Rl(v+J) >

\h\

<

\z\.

> 1,

RI/a

00

= Jtf+g J^_

(t+z)-"l*Jv{2j(t+z)}t dt

(2^ 5 ).

(Sonine.)

15.

Show

that,

when Rl(v+ J) >

0,
l

[ eia lfi

\y- x l 2

dt

licosirrr f

where the contour C, on which |arg(/ 2 1)


goes round t = 1 once positively and t =

Deduce that

Jv (z)

r(

g)

)(

|T^

provided only that

<

77, is

a figure of eight which

once negatively.

"

<(*
j

16.

coszm(1 u^f- 1 ! 2 du,

- l)-

of Ex.

5,

dt,

(Hankel.)

v+J is not a positive integer.

By using the result

1'2

'

prove that

IT

J (2zsmijt)cos2niJ> difi

and

also that

J2n (2zsini/t)

dip

7r{J(z)} 2 ,

tt{JJz)}

2
.

BESSEL FUNCTIONS

344
1 7.

Show

that, if Z =

z2

+ z\ - %zz

J (Z) =
18.

By

that, if

cos B,

| J (z)Jm (z% )eM.


00 m

integrating round a semicircle in the upper half-plane, show-

>

a,

a,
"

_
~ m.a?
H
(au)li
x*'
/
HPjxu) du
(

this being

a Cauchy principal value.


CO

C JJau)YJxu) JJxu)YJau) du

-r* j
mi
D6dUCe
that
j.

J
19.

Show

Jv {au)Y + {iA au)Y

irav

" 5?

that, for all values of the parameter

v,

,,

(T-OHMABSH.)

the complete primi-

tive of the differential equation

>ldw
dw
+

d 2w
dz*

v2 \

'

zHz-\ l + zj) W =

w=--AUz)+BKv(z),

is

where

'

m=

KM =

m!\T(v+m+\)'
.

Jirie'"

20.

Prove that, when

|argz|

21.

Show

|argz|

<

rf

a
i2F>(e"*/ z).

2
/

\n and

\z\

is large,

V(2t7Z)'

that,

when

<

jt,

oo

4( z )

00

K(z)

+ wi

==
~*
2w
J

e^coshic-raidw,

7li

e- zcosllw coshvw dw,

the path of integration in the former integral being along straight lines
from oo 7ri to
iri, from
-ni to iri, and from
to oo+wi.

CHAPTER

XIII

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS


13.1. Periodic functions

function

/(z) is said to

be periodic

if

there exists a non-

zero constant 2a such that the equation

f(z+2a>)
holds for

all

values of

z.

= f(z)

The number 2w

is

called a period of

f(z); evidently, if n is a positive or negative integer, 2wco is also


call 2a> a fundamental period if no submultiple
a period.

We

of

a period.

it is

periodic function which has only one fundamental period

For example, ez is simply-periodic,


being
2ni. A function which possesses
fundamental
period
its
more than one fundamental period is said to be multiplyperiodic. In the present chapter we show that multiply-periodic

is

said to be simply-periodic.

functions exist

by constructing Weierstrass's doubly-periodic

function p(z), which has two fundamental periods whose ratio


is not real; every other period is a sum of multiples of these

two fundamental periods.


The existence of such a function naturally suggests the

fol-

lowing questions:
(i)

Does there exist an analytic function, regular save for


poles, which has two fundamental periods whose ratio
is

real?

Does there exist an analytic function, regular save for


poles, which possesses more than two independent
fundamental periods?

(ii)

These questions were first asked by Jacobi,f who showed in


each case that such a function is necessarily constant.

Example. The
that, if

function /(z) has two periods

2ma) 1 + 2nco 2

is also

and

2co 2 .

Show

a period.

f See the earlier part of Jacobi's paper, Journal fur


is reprinted in his Qes. Werke, 2 (1882), 25-50.

which

2coi

m and n denote positive or negative integers or zero, the number


Math. 13

(1835), 55-78,

346

13.11.

THE ELLIPTIC FTOJCTIONS OF WEIERSTRASS


The lower bound of the periods of an analytic

function

As a preliminary step
we show that a

Jacobi,

in provingf the results established

by

periodic analytic function which is not

a constant cannot possess arbitrarily small periods.


Let/(z) be an analytic function which has a set of periods 2a>,
where the lower bound of \co\ is zero. If z is a point at which
f(z) is regular, the function f(z)f(z

points z +2j.

But

has a zero at each of the

bound of

since the lower

exists a point of the set z -f-2a> in every

This

is

impossible J unless f(z)f(z

is

|a>|

is zero,

there

neighbourhood of

identically zero.

Hence

the result.

13.12. Jacobi 's first question

We now answer Jacobi 's first


analytic function f(z), other than

question

by showing that

constant, has

if

an

of periods
a multiple of

set

where A is real, each of these periods is


fundamental period.
If the only values which A takes are positive or negative
integers, there is nothing to be proved; 2a> is then the fundamental period.
2co,

2Aco,

single

however, there are periods 2Aw with non-integral values


each such period is expressible in the form 2mco+2ixco,
where
is a positive or negative integer or zero and
a
1.
Evidently 2aa> is itself a period, since it is the difference between
the periods 2Ad> and 2m<a. Thus corresponding to every period
2Aco which is not a multiple of 2a>, there exists a period 2<x<o,
where
a
1.
There can, however, be only a finite number of periods of the
form 2<xoj. For if there were an infinite number, they would
possess a limiting-point in virtue of the Bolzano-Weierstrass
theorem; as the difference of any two of the periods 2aco is also
a period, this would imply that there exist periods of arbitrarily
small modulus, which is impossible by 13.11.
Let us write 2<x> x
2<x x w, where a x is the least of the numbers
If,

of

A,

< <

< <

f In framing the general argument of 13.H-13.13 I am materially indebted to Mr. W. L. Ferrar and Mr. J. Hodgkinson, and I wish to take this
opportunity of thanking them.
% See 4.51.

THE ELLIPTIC FUNCTIONS OF WEIERSTBASS


Then

347

and each of the periods 2Aw are multiples of 20^.


For if not, by a repetition of the previous argument we could
find a period 2a 2 co x where
< a 2 < 1, and this is impossible
by the definition of 2a> v This completes the proof of the
a.

2cu

theorem.
It should be observed that we have shown incidentally that
an
analytic function, other than a constant, has distinct periods
if

and 2Aco, where A is real, then A is necessarily rational. For 2oj


and 2Xco are multiples of a fundamental period 2w x
Finally, it is evident that if all the periods of an analytic
function are of the form 2Aw, where A is real, the function is
2co

simply-periodic.

13.13. Jacobi's second question


Let/(z) be an analytic function, other than a constant, which

two fundamental periods 2w x and 2a> 2 whose ratio is


not real. Such a function cannot be simply -periodic. We now
answer Jacobi's second question by showing that it must be
doubly-periodic, in the sense that every period is a sum of
multiples of a certain pair of fundamental periods, not necespossesses

sarily 2co 1

and 2w 2

Since ct^/a^

is

not

real,

we can express any period


and

in the form 2Aco 1 +2/o 2 where A


,

fx

are real,

2a>

by

uniquely

solving the

simultaneous equations!

If

Rico

=ARla) 1 -(-jU.Rlaj 2

Ima>

= Almwx+juJmajg.

turns out that the only values of A and

it

which occur are


nothing more to
of multiples of 2to 1

positive or negative integers or zero, there

be proved, since every period

and

2co 2

If,

2Zco 1

is

a sum

is

fj.

is

which are not of the form

however-, there are periods

+2mco 2 where I and m are integers or zero, each such period


,

expressible in the form


2la) 1 -\-2ma) z -\-2oto) 1 -\-2fSa) 2 ;

t These equations do determine A and

and

this,

by

hypothesis,

is

not zero.

/t,

since their determinant is

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

348

here

a and

and

m denote integers or zero, and


0< a< 1,
<<

1,

not being simultaneously zero. Evidently the number

/?

2ao> 1 +2co 2
is itself

a period.

We now see

that neither

<x

nor

f$

can be zero;

for if /? were zero, a

would not be zero, by hypothesis, and f(z)


would have periods 2aj t and 2aa> 1 This is impossible, by 13.12,
since 2tu x is given to be a fundamental period.
We have thus proved that, corresponding to each period 2w
which is not of the form 2Zoj 1 +2maj 2 where I and m are integers,
there exists a point 2aco 1 +2^o) 2 which represents a period and
lies within the parallelogram with vertices 0, 2cu
1; 2o> 1 +2oj 2 2cu 2
There can be only a finite number of such points; for, if there
were an infinite number of them, it is easily seenf that f(z)
would have periods of arbitrarily small modulus, which is impossible. Moreover, no two periods of the set have the same
value of /?; for if 2ao) 1 +2to 2 and 2a'co 1 +2co 2 were two such
periods, 2\ix~a'\w 1 would also be a period, and this is im.

possible.

We

that there exists a unique point of the set


which the value of fl is least. Let us call it 2&> 2
Every period of f(z) can be expressed as a sum of multiples of
2u> x and 2w 2
For if not, a repetition of the preceding argument
shows that we can find a period of the form 2a'a> 1 +2/}'co 2 where
see, then,

2a<o 1 -f 2j8w 2 for

< a' <

1,

< $' <

1.

2 a 'a> 1 +2 8'a4
i

But

since

2(<x'+a 8')a> 1 +2#S'a> 2 ,


i

We have thus proved


that if f(z) is an analytic function, other than a constant, which
possesses periods whose ratio is not real, it is necessarily doublythis contradicts the definition of 2co 2

periodic.

If 2co and 2a/ are two periods of a doubly-periodic function


with the property that every other period is a sum of multiples
of 2a> and 2o/, we say that 2o> and 2a>' form a pair of primitive
periods.

The periods

2<o 1

and

2o> 2

evidently form such a pair.

When a pair of primitive periods is known, an unlimited number


t Cf.

13.12.

THE ELLIPTIC FUNCTIONS OF WEIERSTKASS


if we write

349

of other pairs can be found. For

O =
x

where

Coj x -\-dco 2 ,

and d are integers connected by the relationf


then 2Q X and 2Q 2 are also periods. Moreover

a, b, c,

ad be =

1,

ui^

= dQ-i bQ

= cQ

w2

2>

As any period is expressible


it is

Q =

fltoj-t-feajg,

as a

x -|-aQ 2 .

sum of multiples of 2od and 2w 2


sum of multiples of 2Q X and
>

therefore also expressible as a

2Q 2 thus 2Q X and 2Q 2 form a


;

Example. Show

pair of primitive periods.

that any three periods 2to 1( 2co 2

periodic analytic function are connected

by a

2la) 1 +2mco i +2na) 3

where
13.2.

I,

m, and n are

The

2w 3

of a multiply-

relation
0,

integers.

an

definition of

An elliptic function is

elliptic function

defined to be a doubly-periodic analytic

function whose only possible singular points in the finite part


of the plane are poles.
elliptic

functions,

it

is

Before we actually construct special


convenient to consider some of the

properties of elliptic functions in general.

Let/(z) be an elliptic function with a pair of primitive periods


2a) x

and

pose, as

With

2a> 2

The imaginary part of

we may without

co 2 /to 1 is

not zero; we sup-

loss of generality, that it is positive, f

this convention, the points 0, 2oj v 2cu 1 +2co 2 , 2co 2 ,

taken

in order, are the vertices of a parallelogram described in the


positive sense.
/(z);

We

call it

a primitive period-parallelogram of

number of
we have proved,

there are evidently an unlimited

period-parallelograms.

Since, as

primitive

the only

and n are
periods of/(z) are of the form 2mcj 1 -\-2nco i where
integers, the vertices are the only points within or on a primitive
,

period-parallelogram whose affixes are periods.

Now mark

in the

Argand plane the points of

^ m,n =

2ma> 1 +2na) 2

affix

t Evidently the integers a and 6 must be prime to each other. When a and
and d can be found by the process of repeated division used in
determining the G.C.M. of two numbers. See Chrystal, Algebra, 1 (1910), 45;
2 (1919), 436.
primitive
J For if lTa(uiJai 1 ) is negative, we consider instead the pair of
periods 1w x and
2tu 2
b are fixed, c

THE ELLIPTIC FUNCTIONS OF WEIERSTBASS

350

where

points

Qpq

and n take the values


,

0.p+lq , Oj, +lg+1 ,

1, 2,... Then the four


are the vertices of a parallelo-

0,

>8+1

gram, which is obtained from the primitive period-parallelogram


by a translation without rotation; it is called a periodparallelogram, or, more briefly, a mesh. The Argand plane is
completely covered by this system of non-overlapping meshes.
The points z-\-Qpq and z+Q.rs obviously lie in different
meshes; if we translate one mesh until it coincides with the
second, these two points become coincident. Accordingly we say
that the point z+Clpq is congruent to the point z+Q,
Then
rs
since Q r>s QpQ is a period of/(z), it follows that/(z) takes the
same value at every one of a set of congruent points. The
behaviour of an elliptic function is therefore completely determined by a knowledge of its values in a primitive period.

parallelogram.

From this it follows that an elliptic function must possess poles.


For

an

if f(z) is

elliptic

period-parallelogram,

where

a finite constant.

is

mesh the values


/(z) is

function which

it satisfies

it

But

\f(z)\

< K;

regular in a primitive
\f(z)\

< K,

since f(z) repeats in every

takes in a primitive period-parallelogram,

an integral function which

equality

is

there an inequality

hence,

by

satisfies

everywhere the ina con-

Liouville's theorem, it is

stant.

An

elliptic

mesh. For

function has only a finite number of poles in any


had an infinite number, the set of poles would

if it

possess at least one limiting -point,

a limiting-point of poles

we

see that

an

elliptic

an

and

this is impossible since

essential singularity.

Similarly
function has only a finite number of zeros
is

in any mesh.

When we

wish to calculate the number of poles (or zeros) of


it is inconvenient to have
poles (or zeros) on the boundary of the mesh. But as there are
only a finite number of poles and zeros in each mesh, we can
always translate the mesh without rotation until no pole or zero
lies on its boundary. The parallelogram obtained in this way is
called a cell. The set of poles (or zeros) in a given cell is called

an

elliptic

function in a given mesh,

an irreducible set.
Example 1. Show
f(z).

that, if f(z)

is

an

elliptic function, so also is

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS


Example

2.

with a given

elliptic

The

13.21.

351

Show that all primitive period-parallelograms associated


function are of the same area.

and zeros

irreducible poles

of

an

elliptic

function

We prove, first of all, that the sum of the residues of an elliptic


function atjts poles in any cell is zero. For if C is a cell, the sum
of the residues of the elliptic function /(z) at its poles within G is
c

and

this is zero, since the integrals along opposite sides of

cancel on account of the periodicity of f(z).

The number of poles of an elliptic function in any cell, each


pole being counted according to its multiplicity, is called the
order-f

The order of an

of the function.

least two, since

an

elliptic function is at
function of order one would have one

elliptic

irreducible pole of residue zero,

We

next show that an

in each

cell,

multiplicity.

zeros in a

which

is

impossible.

has
function of order
zeros
multiple zeros being counted according to their

For

if

the

nm

cell,

elliptic

elliptic

is

function f(z) of order


has n
sum of the residues of

equal to the

a* its poles in the celLJ But/'(z) is obviously an elliptic


function with the same periods as f(z), and therefore so also is

f'( z )/f( z )

f'( z )/f( z )',

nm =

hence
0, which proves the theorem.
we prove that the sum of the affixes of the zeros of an

Finally

cell\\ exceeds the sum of the affixes of its


by a period. For if f(z) is an elliptic function
with primitive periods 2cj x and 2a> 2 the sum of the affixes of
its zeros in the cell C exceeds the sum of the affixes of its poles

elliptic

function in any

poles in that

cell

there bytt

tl
Cz

,
1

2m J
2?nJ

z)

dz.

f(z)

should be observed that the word order has different meanings in the
theory of elliptic functions and the theory of integral functions.
t See 6.2.
More generally, a being any constant, the elliptic function /(z) of order
takes the value a
times in each cell.
Multiple zeros (or poles) are repeated in these sums according to their
t It

||

multiplicity.
-ft

See

6.2,

Ex.

1.

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

352

If the vertices of
difference

L
2i

are

t-\-2ui 1 ,

t,

t-\-2w 1 -\-2ui 2 , t-\-2<o 2

this

is

f >/'(*)

(z+2a> 2 )f (z+2co 2 ) \

\M

ds

f(z+2co 2 )

(z+2co 1 )f'(z+2co 1 ) }

fs/'fr)

to*

\M

/(+2i)

2^1

=
since f(z)

f
J

^z-2
M

2^{2co 1 [log/( Z )|

+2

f&
/(*)

--2 W2 [log/( Z )| +2 ""}

and /'(z) have the periods 2&> x and 2w 2


takes the same value at each vertex of
.

Now f(z)
the
the

sum
sum

of the affixes of the zeros of f(z) in the


of the affixes of its poles there by
.{2m7rio} 1 2mricD 2 }

2mo> 1

2na>

cell

C.

Hence

exceeds

2,

Itl

where

and n

required result

are integers.

is

now

13.3. Weierstrass's

As 2mco 1 2na> 2

is

a period, the

established.

Sigma

function

Let f(z) be a simply-periodic analytic function of period it,


which has simple zeros z 1; z 2 ,..., zm and simple poles p v p 2 ,,p n
in the strip
Rlz
It is easily seen that
it.

<

m
II sin(z z,)

/(z)

= g{z)e**> -I

IJ sin(z

-,

r)

or 1
where g(z) is an integral function of period it and k
according as
n is even or odd.
If we wish to exhibit in a similar manner the way in which
an elliptic function with primitive periods 2u> 1 and 2a> 2 depends
on its zeros and poles, we must first construct an integral function with simple zeros at the points l mn which is to play a
part similar to that of sinz in the theory of simply -periodic

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS


functions.

353

The simplest function with the required property is

Weierstrass's canonical product.


In order to construct this canonical product, we
mine the exponent of convergence of its zeros. If 9

must

deter-

= arg(o> /w
2

1 ),

we have
Imwj+wwsjl 2

=m

\cj 1

+n

\o)

+2mncoae\co 1 (u 2

\.

< 6 < cos0 = fi, where < < Hence


Imwj+w^l = TO |w +, |w = 2mTC
= (l-jL6)(m K| +i U2 )+ t(TO]a; |wh
> (l- M )(m Wl +rc W2
> (1 (m +n

But

since

77,

/j,

ii|a) 1 co
j
2

From

|)

),

the smaller of

loijl

iTO^+wwal
where 6

fj,)a

is

|c

where a

1.

<

and

|co 2

|.

we

Similarly

see that

(l+ /i )6 2 (m 2 +r2,2 ),

the greater of la^l and |cu 2 |.


these two inequahties it follows that the double series

is

2*

where summation

\^m,n\~

'

extended over all positive and negative


integral and zero values of m and n, save m
n
0, converges
or diverges with the series
is

= =

2' (m2 +n?)-<l2

But it

is

shown by the

easily

integral testf that the latter series

>

<

convergent when a
2 and divergent when a
the exponent of convergence of the zeros Q m m is 2.

is

2.

Hence

It now follows from the general theory of canonical products


J
that the doubly-infinite product

where multiplication

is extended over all positive and negative


and zero values of m and n, save m = n = 0, converges
uniformly and absolutely in any bounded closed domain of the
z-plane which contains none of the points l m n and represents
an integral function a(z\a> 1 w 2 ), of order 2, with simple zeros at
the points Q. m>n This function is Weierstrass's Sigma function.

integral

t See, for example,

Bromwich,

t 7.2.
4111

Aa

Infinite Series (1926), 86.

THE ELLIPTIC FUNCTIONS OF WEIERSTKASS


When there is no need to emphasize the parameters w x and w 2
on which the Sigma function depends, we shall denote it more
354

by

briefly

a{z).

be observed that a{z) is not an elliptic function; for


would be identically zero in virtue of the theorem
of 13.2, and this is certainly not the case.
In the canonical product for a(z) we can arrange the factors
in pairs, such as
It should

if it

were,

it

~er-)

Mcr~ + 25r)'

1+ i~) exp _
if~ + 2i5"4
o
(

the second factor being derived from the first by replacing


and n. But since these two factors interchange
and n by
when we replace 2 by z, it follows that o{z) is an odd func-

tion of

z.

13.31. Weierstrass 's elliptic function p(z)

Before

we show how

elliptic functions

can be constructed as

quotients of products of Sigma functions,


discuss the properties of Weierstrass 's

which

is

it is

elliptic

convenient to
function @{z),

very closely connected with a{z).


log a(z) can be written as a double

The function

series

which converges absolutely and uniformly in any bounded


closed domain D which contains none of the points D mj .
Weierstrass' s Z eta function^ is denned by the equation
C{z) =^

-^oga(z);

term-by -term differentiation gives at once

i+2'{di-+a

L +ff-

the double series being uniformly and absolutely convergent in


the domain D, since
t This function

is

its

general term

is

0(|Q mn

-3
).
|

not to be confused with Riemann's Zeta function


Us)

which

is

= f n-,
1

of importance in the analytic theory of numbers.

(z) is

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS


therefore

each of the points


If

we

355

an analytic function with a simple pole of residue

replace

becomes

Q m .

and n by

and n, the

at

series for (z)

l,_z_\

,_

J[

Hence
C{

z)

so that (z)

is

+ Z \~z+Q
am<n o*J
m:n

an odd function.

(z) is

since the residue at each of its poles is

residues of

an

elliptic

-m,

not an elliptic function,


whereas the sum of the

1,

function at a set of irreducible poles

is

zero.
Weierstrass's'f elliptic function p(z) is

equation
*<*>

now denned by

the

^w-

It follows that p{z) is represented

by a double

series

"" = 3+ 2' {l^y-ik-}


which converges uniformly and absolutely in every bounded
closed domain containing none of the points 0. m>n Hence p(z)
is an even analytic function whose only singularities are double
.

poles of residue zero at each of the points

To show that

p(z)

is

an

mn
we consider the
increased by 2a> x or 2a> 2
0.

elliptic function,

behaviour of its derivative when z is


This derivative p'(z) is given by the equation

where summation is now extended over all positive and negative


and zero values of m and n without exception. From

integral
this,

we

see that
,

ey

2
but since the set of points
this equation becomes

0, m _ ln is

p'(z+2w 1 )

t Weierstrass, Ges. Werke,

the same as the set

$>'(*).

(1895), 245-55.

l m>n ,

THE ELLIPTIC FUNCTIONS OF WEIEKSTEASS


has the period 2a) x similarly we can show that it has

356

Thus

p'(z)

the period 2oj 2


function, for it

The function

is

doubly-periodic with a triple pole of residue

zero at each of the points t2 mw

2^

and

p'(z)

is,

an

therefore,

elliptic

form a pair of primitive periods of


p'(z).
For if not, there would exist a network of periodparallelograms, each having a smaller area than the parallelogram
with vertices 0, 2a> 1; 20^+ 2a> 2 2a> 2 some of these would evidently contain no singularity of p'(z), and this is impossible.
Moreover,

2o> 2

By

integrating the equation

p'(z+2a> 1 )

we obtain
The value

of the constant

Since p(z)

is

p(z-\-2u> x )

= p'(z),
= p(z)+C.

C can be found by putting z = a^.


an even function and p(<^i) is finite, this gives
G = p{u>J-p{-a> x ) = 0,

a period. Similarly 2co 2 is also a period. Hence


p(z) is a doubly -periodic function whose only singularities are
poles, and so is an elliptic function. Evidently 2cd x and 2a> 2
form a pair of primitive periods of p{z).
When it is desirable to put in evidence the primitive periods
2(n x and 2a> 2 with which p(z) is constructed, we denote it by

so that

2w 1

p(z\o) v

w 2 ).

13.32.

The pseudo -periodicity

If

we

is

(z)

integrate the equation p(z-{-2oj 1 )

^+2^) =

relation

where
have

of

2r) x is

t{z)

p(z),

we

+ 2^

a constant of integration; putting

obtain the

a> lt

we

= ( Wl )-(-c^) = 2K),
Hence 1 = (a>
Similarly
Uz+2w = (a) + 2^

is

odd.

x ).

7j

2)

then,

a{z)

2 Vx

since (z)

where

and

ij

= ^(to

2)-

Neither

by symmetry,

since (z)

is

not an

t/

rj

would

2,

nor ^ 2 i s zero; for if


also be zero, which

elliptic function.

The

rj

were zero,

impossible
function (z) has,
is

therefore, a pseudo-periodicity, in that the function is repro-

duced, apart from an additive constant,

a period of

p(z).

when

z is

increased

by

THE ELLIPTIC FUNCTIONS OF WEIEESTKASS


The constants

^ and

t]

7]i (v

by the

are connected

357

relation

= i -

2 rl2 0} i

For the only singularity of (z) within or on the cell C of


w 1 ui 2 co 1 +to 2 to 1 +cu 2 oj 1 co 2 is a simple pole at

vertices

the origin of residue


2iri

(z)

hence

dz

J"

=
=

j {Uz)-t(z-2u> 1 )}dz+
2t]x
toi

dz

2t} 2

a>2

{t(z)-t(z+2co 2 )} dz

cfe

a>i a>2

4^6024^2 0!,

which gives the required

result.

It is frequently convenient to

make

use of the period

where eoj+tug+wg = 0. The pseudo-periodicity of


respect to 2a> 3 is then expressed by the equation

C{z+2w a
where 7^+772+773

0.

the equation (z+2co x )


tegration
that
/
o
^
s

= (z) +

= ^4e
Putting z =
,

(7(3+20)!)

is

a constant.

Similarly

a(z+2cu 1 )

we can show

>,>

2l

,
e

2t7 1;

we deduce by

a(z),

co lt we have

fa^)

that

ct(z+2co 2 )
(j(z+2co 3 )

= e >^+^a{z),
= e ^(s+<"
2r

3 )cT(z).

These three equations exhibit the pseudo-periodicity of


z is increased by a period of p(z).

when

Example. Show
772603

that

77360,5

in-

= e^fc+Mrfc).

Ct(

so that

2tu 3 ,

with

= i(z) + 27,

From

where

(z)

77360!

Tj 1

OJ 3

77 1

60 2

772 00! =

%ni.

a(z)

THE ELLIPTIC FUNCTIONS OF WEIERSTKASS


The algebraic relation connecting two elliptic func-

358

13.4.

tions

We shall now prove the important theorem that if two elliptic


functions have a pair of common periods whose ratio
they are connected by an algebraic relation.

Let f(z) and

be two

g(z)

elliptic

is

not real,

functions having in

common

not real. From


a
primitive pair
as
in
13.13,
these periods we can construct,

property
that
every
common
of periods 2co 1 and 2oj 2 with the
period of f(z) and g(z) is of the form 2ma> 1 2ncu 2 where m and
n are integers. We are not necessarily supposing that 2a 1 and
2a> 2 form a pair of primitive periods for both functions; the
theorem will still be true, for example, if <a x and 2oj 2 are a
primitive pair of periods of /(z), and 2to x and w 2 are a primitive

2Q 2 whose

a pair of periods 2t2 1 and

ratio

is

pair of periods of

g(z).

Denote by a v a 2 ,..., am the points in the parallelogram with


vertices 0, 2oj 1; 2o> 1 +2oj 2 2cu 2 which are poles either of f(z)
or of g(z) or of both functions. Let fi r be the order of the pole
at aT if ar is a pole of both functions, let /xr be the greater
,

order.

We now consider a polynomial F(,

ij),

of degree

in

and

ij,

which has no constant term; such a polynomial involves


^n{n-\-Z) arbitrary constants.

<b(z)

therefore,

is,

2oj 2 ,

an

elliptic

The function
F{f{z),g(z)}

function of primitive periods

having poles at some

or all of the points ar

2oj x

and

Since the order of the pole ar cannot exceed n/u.r we can make
the principal part of O(z) at each pole identically zero by
,

choosing the \n(n-\-Z) coefficients to satisfy a certain set of


w(/x 1 +ju, 2 +...+

um )

homogeneous

n+3 >

linear equations.

If

2( /x 1 +/x 2 +...+jO>

there will be fewer equations than coefficients to be determined


so a suitable set of coefficients can always be found.l But
the coefficients are chosen in this manner, O(z) is an elliptic

and
if

f In

some

cases

it is

possible to find suitable ooeificients

n+3 =
An

example

of this is provided

2(ii 1 +ii t

by the

+ ...+pm

differential

by taking

).

equation of

13.41.

THE ELLIPTIC FUNCTIONS OP WEIERSTRASS


function without singularities

and

is

359

therefore a constant.

other words, there exists a polynomial F(,

In

such that

rj)

F{f(*),g(*)}
is

a constant, which

13.41.

The

is

the required result.

by

differential equation satisfied

p(z)

from the general theorem of 13.4 that there exists

It follows

relation connecting the two elliptic functions p(z),


they have the same primitive periods 2coj and 2w 2
This relation, which we shall now determine, is a differential
equation satisfied by the function p(z).
The function p(z) z~ 2 is regular in a neighbourhood of the
origin and so can be represented there by a power series

an algebraic

p'(z), since

CO

>(z)-Z-2

= 2X Z"

from Taylor's theorem that


a 2n+1 vanish and that
It follows

and

so on.

0,

fl

= 3r^,

hood of the

coefficients

z"+0(z),

regular in a neighbour-

is

and has a zero of order n at the

origin

origin.

this equation it follows that


p'{z)

We now
is

= 2z-

+2a 2 z+4a4 z3 +0(z 5

).

construct a polynomial in p(z) and

regular at the origin.

(z)

p' 2 (z)

To do

this,

p' 2 (z)-4p 3 (z)

),

= -20a
= 20a

its

2
),

z- 2 -28a 4 +O(z 2 )

p(z) 28a4 +0(z2 ).

have thus proved that the function


<D(z)

= p'2(z)-4p*(z)+20a

derivative

we observe

= z-+3a z- +3a +0(z


= 4z- -8a z- 16a +C(z

whence

We

odd

4=5I'^

= 2- +a 2Z +a

where 0(zn ) denotes a function which

which

the

Thus
p(z)

From

all

p(z)+28ai

that

THE ELLIPTIC FUNCTIONS OP WEIERSTRASS

360

regular in a neighbourhood of the origin


zero at the origin.
is

Now
Hence

mn

0.

an elliptic function with periods 2a> x and 2a> 2


regular in a neighbourhood of each of the points
as the only possible singularities of <3>(z) are the
0(2) is an elliptic function with no singularities,

<D(z) is
it is

But

Qffln

points

and so

and has a double


.

a constant. The value of this constant is zero, since


<I>(z) has a double zero at the origin. Hence p(z) is a solution of
the differential equation
is

@'\z)

where

a2

' Q^*

For many purposes


g2

20a 2

4p(2)-20a a p(z)-28o4

T O^V

more convenient

it is

60 ' &%,

93

to write

28a4

140 1' Cl'%,

the constants g 2 and g z being called the invariants-j- of p(z). We


have thus proved that w = p(z) satisfies the differential equation
4vJs

Example
d 2 w/dz*

6tt>

Show

that

w=

g 2 w-

p(z) satisfies the differential equation

-Jft.
The function p(z)z~ % possesses a Taylor expansion of the form

p(z)-z-*

Show

valid near the origin.

{n-2){2n + 3)c n
for

whose
a

3,4,5,....

coefficients

Example

2.

c1 z 2

+c 2 zH.-+cM s 2

',

+...

that the coefficients are connected by the

+c

3(c 1 cm _ 2

Hence prove that

c n_ 3

+ + cM_
...

c1 )

a polynomial in ft and g s
are positive rational numbers.
cn is

Prove that
<t(z)

z+b 1 z< +b 2 z '+...+bn z+3+...,


1

where the coefficients b n are polynomials in ft and g3 whose coefficients


are rational numbers. In particular, show that

W=
Example

3.

Show

62

-ft/240,

-ft/840.

that

(T(A2|AaJ 1 ,Ac0 2 )

(Az|Aco 1 ,Aaj a )

^(Az|Acu 1 ,Aco 2 )

= A(7(z|o) ,CU
= A-^zlco^eo,),
= A- p(z|cu a)
1

2 ),

1(

2 ).

Prove also that the invariants of p(Az|Aa) 1 ,Ao) 2 ) are A~ 4ft, A_6 ft.
f

The reason

for the

name

will

be evident after reading

13.7.

THE ELLIPTIC FUNCTIONS OP WEIEBSTRASS


13.42. The constants e 1; e 2 and e 3
We shall now show that, if g2 and g 3 are the invariants

361

ciated with
4,tv

$){z),

the three roots e 1;

g 2 wg 3

To prove

and

an

since p'(z)

an odd

is

each of the points

wv

<xi

we have

elliptic function,

vanishes at

it

oj 2 .

But

function of order 3 with a triple pole at

elliptic

TO

sum

consequently the

its irreducible zeros is

points

asso-

of the equation

so p'{z) vanishes at o^; similarly

p'(z) is

e3

of the fact that e v e 2 and e 3 are


the points where its derivative

this

Now

vanishes.

and

are all distinct.

we make use
taken by p(z) at

the values

e2

a period.

and w 3 form a

Since

of the affixes of

+to 2 + ttl 3

aj 1

0,

the

set of irreducible zeros of p'{z).

It follows that
ei

e2

@if\>,

= p(a>

= p(w

e3

2 ),

3)

are the three roots of the cubic equation in question.

Now

p(z)e 1 being an
,

double zero at z
primitive
ex

^e

cu 1;

elliptic function of order 2 with a


cannot vanish at any other point in the

period-parallelogram;

Similarly

constants

13.43.

ev e2

The

in

we can show that

and

particular,
e2

=fi

e3

e1

^e

and

so that the three

e 3 are all distinct.

solution of a differential equation

Let us consider the problem of finding the function


by the differential equation

_=

(4:w 3

g 2 wgs

defined

1l2
)

where g 2 and g3 are given constants. This problem can be presented in a somewhat different form;
of w by the equation
z-\-a.

where a

is

f (4w>

g wg
2

z is

given as a function

)~ 112

dw,

a constant of integration. If we could carry out the


and solve the resulting equation, w would be deter-

integration

mined.
If the discriminant g\Z7g\ of the cubic

4uP

w g3

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

362

wQ say, and the integral

vanishes, the cubic has a multiple zero,

becomes

dw

which can be evaluated by the elementary methods of the

in-

tegral calculus. It follows that, in this case, z can be expressed


in terms of

w by means

of the elementary functions of analysis.

If the discriminant of the cubic


distinct linear factors.

numbers

io x

We

and w 2 whose
,

g2

shall

ratio

is

is

60 2' O^t,

not zero, the cubic possesses


that, if we can find two

show

g3

not

real,

such thatf

140 2' K&,

the integral can be evaluated^ by means of the transformation

Assuming, then, that a^ and

z+

and

so

= j {^(Q-g

p{\u) lt

oj 2 )

w 2 have been
2

found,

we have

p(0-9s}- ll2 P'a) dC

(p{z-\-a.\<a x ,

w2

).

00

In particular,

(4ws g 2 wg3 )

~ 1/2

dw

w
equal to z+Q mn where

z is the point at which p takes the


value w. The period Q mn depends on the manner in which
the path of integration loops round the branch-points e x e 2 and

is

es

of the integrand.

13.5.

The

The addition-theorem for


elliptic

p{z)

functions p(u) and p(u+v), regarded as functions

of the complex variable u, have the same pair of primitive


periods and so, by 13.4, are connected by an algebraic relation.

To determine

this relation
f{z) =

where
order

A
3,

we

consider the function

p'{z)+Ap{z)+B,

This is an elliptic function of


with a triple pole at each of the points Q mn it has,

and

are constants.

27g\ is not zero, is proved


t The existence of such numbers w lt <o 2 , when g\
in 15.31.
% It cannot, however, be evaluated in terms of the elementary functions
of analysis. See, for example, Hardy, Integration of Functions of a Single
Variable (Cambridge tract, 1905).

363
THE ELLIPTIC FUNCTIONS OF WE1ERSTRASS
is
affixes
whose
therefore, three irreducible zeros, the sum of

a period.

Now
is

if

u and

v are such that

a period of p(z),

none of the numbers u,v,uv


the constants A and B so

we can choose

=
p'(v)+Ap(v) + B =

that

p'(u)+Ap(u)+B

0,
0.

When this is done, the function f(z) has simple zeros at the
points congruent to u and v; the third irreducible zero is therefore congruent to

uv. Hence we have

p'{ u v)+Ap( u v) + B = 0.
Eliminating A and B from these equations, we obtain
p{u)

p'(u)

p{v)

P'(v)

p{u+v)

p'(u+v)

=0.

Now the derivatives occurring in this equation can be expressed


p(u+v) by means of the
differential equation for p(z), and so we have really expressed
p(u+v) algebraically in terms of p(u) and p(v).

algebraically in terms of p(u), p(v),

An

analytic function F(z)

theorem

if

said to possess an addition-

there exists a formula which expresses

braically in terms of F(u)

to express

is

p(u+v)

and

F(v).

F{u+v)

Although we only

algebraically in terms of p(u),

alge-

set

out

we have

actually proved the important result that p(z) possesses an


addition-theorem. It should be noticed that Weierstrassf

proved in his lectures that a function F(z) which possesses an


addition-theorem is either an algebraic function of z, or an algebraic function of exp(7rzi/co), where o> is a suitably chosen constant, or

an algebraic function of

)(z\cj v co 2 ), oj 1

and

cu 2

being

suitably chosen.

An

alternative form of the addition-theorem for p(z) can be

obtained by considering the


F(z)

elliptic

function

= p'\z)-{Ap{z)+Bf
= p3(z)-A*p*(z)-(2AB+g

2 )p(z)-(g 3

+B*),

account of Weierstrass's lectures is given by Schwarz, Formeln und


t
Lehrsatze zum Gebrauehe der elliptischen Funktionen (Berlin, 1893). The
theorem to which we have just referred ia stated there without proof. Proofs
have been published by Phragmen, Acta math. 7 (1885), 33-42, Koebe, Berlin
Dissertation (1905), and Falk, Nova Acta Soc. Upsal. (4), 1 (1907).

An

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

364

where the constants A and B have the same values as before.


This is an elliptic function of order 6, with six irreducible zeros
at the points u, v, (u+v). Since p(z) is an even function,
it

follows that the cubic equation

4p3-AZpZ-(2AB+g 2 )p-(g3 +B*)

has the three roots p(u), p(v), and p(u+v).


deduce that
p(u) + p(v)+p(U +v)
\A*.

From

this

we

But

if

we

solve the equations defining

and B, we obtain

P'M-P'W

A=

'

p(u)-p{v)

Hence we have

a relation which expresses

and

p(u+v)

explicitly in terms of p(u)

p(v).

Example

Prove that the functions

/()=

(i)

P(z)

p'(z)

p(a)
p(z + a)

p'(a)

II
1

-p'(z + <x)

l|.

are elliptic functions with no singularities. Deduce the two forms of the
addition-theorem.

Example

(This result

Example

2.

is

Show

called the duplication formula for p(z).)

3.

Prove
p(u)

where

that

u+v+w =

that, if

=p

0,

p(v)

lt

The formula

By the second form


***+.>

2,

p(w)

=p

3,

then

(Pi+P2+P 3 )(PiP 2 p 3 -g z
13.51.

=p

(PiPa+PiPs+PaPi + ig^.

for piz+cvj in

terms

of p(z)

of the addition-theorem for p(z),

\m^\-^-^
5

'

[p(z)- ei}*

-p(z)-e 1

we have

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

365

{p(z)-e2}{p(z)-e3}

= ei+e e g)(z)(e
3

+e3

p(z)-e 1

But since

e 1 -\-e 2 -{-e 3

piz+otj)
Similarly

relation can be written in the

0, this

we can show

e1 -

p{z+oj 3 )

e3 -

of

an

The expression
Sigma functions

e2 )(e 1 e3

p(z)-e 1

that

p(z+a> 2 )

13.6.

(e 1

form

e2 -

(e 2

1 )(e 2

P(z)-e 2
(e 3

3)

'

e^fa

2)

elliptic function in

terms

of

In the next few pages we shall consider the problem of


expressing a general elliptic function in terms of the periodic or

pseudo -periodic functions of Weierstrass. The simplest method


to express such a function as a quotient of products of Sigma
functions, the resulting expression being analogous to the
formula
n
m
g{z)e ik * XI sin(z-zr ) f[ sin(z-p r )
is

r=l

'

r=l

for a simply-periodic function with assigned zeros

Let us consider then an

elliptic

2w 1; 2w 2> of which

and

poles.

function of order

n and

a set of irreducible zeros, a multiple zero being repeated in the set according
to its order. If p x Pi,---, p n -\, p'n i s a se ^ f irreducible poles,
a multiple pole being repeated according to its order, we know
primitive periods

z x ,z 2 ,...,z n is

Zl

+Z2 +...+Zn

= p +P +-+Pn-l+Pk+^>
1

where Q

is a period. If we replace p'


n -\-Q by p n we obtain a set
of irreducible poles p ls p 2 ,..., p n the sum of whose affixes is equal
to the sum of the affixes of the given set of zeros.
,

We now

construct the function

1_1 o(zpr )

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

366

which has the same poles and zeros as f(z). Hence f(z)jF(z) is
an integral function. But, by the pseudo-periodicity of a(z),
we have
1_1 exp{2^ 1 (z-^ r

F(z)

+ Wl

)}

is of period 2w x
similarly 2oj 2 is also a period.
Hence .F(z) is an elliptic function. This implies that f(z)/F(z)
is an elliptic function with no singularities and so is a constant,
A say. We have thus proved that

so that F(z)

a(z-pr )

1 1

a formula which exhibits the manner in which the


tion /(z) depends on its zeros and poles.

Example,

/(z) is

zeros z v z 2 ,..., z n ,

an

and

z i+ z 2+

zu

p lt p 2 ,..., p n Show
.

= Pi+Pt + +p

re

then

where

f(z)

= AeWviWJ* TT

that, if
,

o(z-z T )

a constant.

p(z) p(a)

As an example on the theorem of


elliptic

+2ico 1 + 2ma) 2

'=1
is

A formula for

13.61.

func-

of order n, with irreducible

elliptic function,

irreducible poles

elliptic

F (z) =

function

where a

is

13.6,

we

shall express the

p{z)-p{a),

not a period, as a quotient of products of Sigma

functions.

We
Then

suppose, in the

first

instance, that 2a

is

not a period.

of order 2, and has a pair of irreducible simple


zeros a and a. Since F(z) has a double pole at the origin and
all

F(z)

is

congruent points,

it

follows at once that

){z)-p{<x)

=A

J\^

'

where A is a constant.
To determine the value of A, we consider how the expressions
on each side of this equation behave near the origin. For
sufficiently small values of

P{z)-p{a)

\z\,

= Z~* -$)(*) + 0(z*).

THE ELLIPTIC FUNCTIONS OF WEIERSTBASS

367

Moreover, by Taylor's theorem,

= <t(oc)+Zo'(oc) + 0{z*)
= o-(a)+z<r'(a)+0(z

a(z+ot)

and

a(z oc)

Using the result of

Ex.

13.41,

cr(zoc)a(z+a.)
2

o-

Equating

coefficients of

A = l/o-

find that

Z2

in the equation

-z~ %

= A-

(a).

g (g

is

'

We have

&(z)-&(ol)
P(Z)
P(a)_

provided that 2a

follows that

^Wiqh)

(z)

p(z)-p(a)

we

2, it

mn

thus shown that

- a g s +)
)

-'

o(s)o*(eO

side of this equation are,

ever, analytic functions of a, regular save


Cl

>

z)

not a period.

The expressions on each


of the set

).

when a

is

how-

a point

Hence, by the principle of analytical con<x is not a period

tinuation, the formula holds provided that

of p(z).

Example
(!)

Prove that

= -JTIT'

(2)

(")

z)

37~T~?

ww

Deduce that
<r(2z)

Example

2{a{z)a(zo} 1 )a(zoD i )o(zoi t )}l{a(u} 1 )<j{m i )a((a i )}.

2.

Show

(r(2z)

that
'

2 CT 3 (z)-3CT(z)CT'(z)a"(z)+CT 2 (z)cr"'(2).

*(*)

13.61 1

The functions

We now

define

{p{z)er}^

{p(z) ej1

'

as

which has a simple pole of residue


principal part of p'(z)

is

2/z

meaning that square root

+1

at the origin. Since the


near the origin, this definition

implies that
p'(z)

-2{p(z)- ei}^{p(z)-e 2yi%p(z)-e3yiK

In order to express {p(z)eryi2 explicitly in terms of Sigma


we write a = a>r in the formula of the last section

functions,

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

368

and obtain
a"(z)a-2 (a) r )

by the pseudo-periodicity of
{fP(z)-er yi*

Hence we have

cr(z).

0*^)0* (co r )

e-Vr'Z&+pl
o(z)o(a> r )

A consideration of the behaviour of the function near the origin


shows that the upper sign must be taken, and so
{p(z)-er}W

= e-*^^

(r

1, 2, 3).

a(z)(j(<v r )

We

see

from

the points

this

formula that

{p{z)e^2 has

and simple zeros at the points

l m>n

simple poles at

oi r -\-Q.

In particular, we shall always understand by


value taken by {p(z) cj 1 2 at the point co,.; thus

(er

mn

s)

^ the

'

(er

Example

-es

112
)

= e-v<r

"

Prove that {p(z) ej 1 2 is an elliptic function of periods


2a) x and 4a> 2 and of order two, which has 0, 2co 2 as a set of irreducible
poles and a> lt <o 1 + 2o) 2 as a set of irreducible zeros. Determine also the
corresponding results for {p(z) ej 1 2 and {p(z) e^lK
1.

'

'

Example 2. Show that


(ei-e^ 2 = i{,e % -ex yl\
(es-ex)

Example

3. Prove,

pilai-L+aiz)

in terms of

4.

5.

=
=

e1
e1

= _

The functions

1 /2

-63 )}i/ 2
1 2
2 )(e 1 -e 3 )}
2 )(e 1

13.51

by

expressing

^(i<ui)-g>("'i) |'
-

g3(z)-p(aJl )

ar (z)

the introduction of the functions

ar (z)

functions, that

+ {(6 -e
-{(e

that

p'{z)

By

-e 2

functions.

Show

*a'(z+">i)

13.612.

t(e 3

Prove the formula of

Sigma

Example

-es )V* =
= t(e1 -e3 )V 2
(e 2

by using Sigma

P(hi)

Example

1'2

= e-^^+f^

(r-

1,2,8),

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

369

the formulae of the preceding section take the simple forms

= <$,

ft*,)-*)*
(er

-e

P = a-p,
a{w
r)

which

prove to be of great importance in the theory of

will

Jacobi's elliptic functions.

The function ar (z)

an

integral function with simple zeros


Moreover,
it possesses pseudo-periodic
mn
properties analogous to those of a(z); these are expressed by
the two equations

at the points

is

co r -\-l

ar {z+2os )

where r

2l

Prove that oT (z)

is left

to the reader.

an even function and that

is

its

Taylor

is

aT (z)

Example

2.

(i)
(ii)

Show

l_ie,a+Js (

fl 11

_6e?)2

..

v.

that

o*00-o =
(e a -e,)of( Z )

(es

-er )<j%z),

+ (e -e
s

1 )oi(z)

The expression of an
Zeta functions

13.62.

It

z+ai

The proof of these formulae

s.

Example
expansion

= e ^ '\{z),
= e ^+<'>ov(z),
2

oy(z+2v)

+ (e1 -e

8 )of(z)

0.

elliptic function in

terms

of

can be shown that, if f(z) is a simply -periodic function of


it whose only singularities in the strip
Rlz < w are

<

period

simple poles p v p 2 ,..., p n of residue c 1; c 2 ,..., c n respectively, thenf


n

f(z)

= g(z)+ J c
r=l

cot(z~pr ),

where g(z) is a simply-periodic integral function. We now propose to determine the analogous formula in the theory of elliptic
functions, where Weierstrass's pseudo-periodic function (z)
plays a part similar to that of cotz in the simpler theory J.
Now although (z) is not an elliptic function, we can easily
Ex. 42 on p. 157.
J (z) is the logarithmic derivative of o(z), just as eotz is the logarithmic
derivative of sin z.
t See

4111

bb

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

370

choose the constants ar so that


n

r=l
is

doubly-periodic. For
n

= ^a {i{z+2co -p )-^z-p

<f>(z+2a> 1 )-<l>(z)

r=l

2l?i

2a =
r

r )}

0,

n
if

2 r

zero; similarly

is

<t,(z+2a> 2 )-<f>(z)

Thus

<(z) is

of the set

pr

an

elliptic

+Q min

0.

function with simple poles at the points

Using this result, we can express any elliptic function in


terms of Zeta functions and their derivatives if we are given
a set of irreducible poles and the principal part of the function
near each pole.
To prove this, consider an elliptic function f(z) of primitive
periods 2w x and 2o> 2 having a set of irreducible poles p x p 2 ,--,
p n If the principal part of f(z) in the neighbourhood of p k is
,

**

1a

s=l

ktS

{z-p k Ys

T ak 1 =

we must have

sum

since the

0,

of the residues of f(z) at a set of irreducible poles


we have just seen, that

It follows, as

is zero.

2 ak,i(z-Pk)
k=l
is

an

elliptic function.

We now construct the function


n

where
that

<s)

(z)

(1) (z)

m*

denotes the sth derivative of {(2;). If

= p{z), we see at once that F(z)

is

we remember

an

elliptic

func-

moreover, an integral function, since the principal


part of F(z) at each irreducible pole has been made identically
tion.

It

is,

'

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

371

Hence, by Liouville's theorem, F(z) is a constant,


Thus the function /(z) can be expanded in the form

zero.

= i+jy

Az)

fc=l

This result
integrate

an

f /(z)

is

=fa

dz

fc(1

5 is the
1

when we wish

i )s i

= l S=2

7~fii^'"

2)

^-^)+^^+^

constant of integration.

P(z)-M*)
if

u+v+w =

= ^+)-2C()-2).
0,

() +(;()+(;()}+{'()+{'() +'()

Example

to

immediately

Prove that

Deduce thatf,

fc ).

loga(z- pr )+

22 (fc

Example

of particular importance
It gives

say.

'*

elliptic function.

+
where

(-i)*-i^i^-i>(3 -,p

s=l

0.

2. Prove that

g>(z-ac)p(z-P)

p(a)p(iS)

+ p(Q;- 8){^(z-a)+p(z-P)-p(a)-p(^)} +
J

+#'( ^KC( )(* >3)+C()C(/8.

is

Example

3.

an

function of periods 2ui x and 2o> 2

elliptic

Show

that

C(*-)-J(-j8)-{(-)8)+{(2-^8)
Prove that
.

it is

equal to

g(g-2a+j8)g(z--2j8+oi)
ct(

13.63.

2 - 2a)o-(z - a)a(z -ft)

The expression

of

an

elliptic function in

terms of

Let us consider, in the first instance, an even elliptic function


of primitive periods 2co x and 2o> 2 which is regular and non-zero
at each point of the set Q mn The order of such a function is
,

necessarily

an even

integer, 2k say.
a zero of /(z) in a certain cell, the point in the
cell congruent to
zr is also a zero of the same order as zr We
can, therefore, choose k zeros z 1; z 2 ,..., z in this cell, each
k

Now

if z r is

f This result is a quasi-addition theorem for (z). It


theorem, since t,'(z) is not an algebraic function of {(2).

is

not a true addition

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

372

multiple zero being repeated according to its order, in such a


way that they, together with the points in the cell congruent
to

z z
1}

2 ,...,

z k form an
,

irreducible set.

Similarly

we can

choose k poles p 1} p 2 ,..., Pk> so that they, together with the points
of the cell congruent to
p v p 2 ,..., p k form an irreducible

set of poles.

When

the zeros and poles have been chosen in this manner,

the function

{)

l\ W-PiPrY

where p(z) has primitive periods 2co x and 2w 2 is an elliptic


function having the same poles and zeros as/(z). Hence f(z)jF(z)
is an elliptic function with no singularities and so is a constant
A. Thus
k
P< g >-frfo>.
,

f(z)

-All

We

next remove the restriction that f(z) is regular and nonQ m>n If f(z) has a pole (or zero) at
such a pole (or zero) must be
points,
congruent
the origin and
positive
or negative integer s be
if
the
of even order. Hence,
zero at each of the points

suitably chosen, f{z){p (z)} s is an even elliptic function which is


regular and non-zero at each of the points O m>m and so is expres,

elliptic

We

have thus proved that an even


function of periods 2cn x and 2w 2 can be expressed as a

sible in

the above form.

rational function of p(z\co 1 ,co 2 ).

From

this it

is

easy to deduce that an

elliptic

function of

primitive periods 2a> x and 2co 2 can be expressed in the form


F{p(z)}+$)'(z)G{$>(z)},
2<u x and 2<u 2 and F(X)
For any elliptic function

where p{z) has the same primitive periods

and

G(X) denote rational functions ofX.

f(z)

can be written in the form

M=

l{/(^)+/(-2)}+i{/(2)-/(-2)}.

term on the right-hand side of this equation is an even


and so is a rational function F{p(z)} of p(z).
The second term is, however, odd; but since p'{z) is also odd,
is, therefore, a rational func\{f( z )f( z )}l&'i z ) i s even, and

The

first

elliptic function,

tion G{p{z)} of @(z).

The

result stated

now follows immediately.

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

373

An important consequence of this result is that every elliptic


function possesses an addition theorem. For an elliptic function
f(z)

is,

as

we have

seen, expressible in the


f(z)

form

F{p{z),g>'(z)},

F denotes a rational function. Hence,


p(v) = p 2
P() = Pi,
P'M = pi

where

writing

we have

=
/() =
/(+) =
f(u)

= p'

p'(v)

2,

F(Pl ,p{),

(i)

F{p 2 ,p'2 ),

(ii)

F{p(u+v),p'(u+v)}.

But by the addition theorem for p{z), we can express p(u+v)


and p'(u-\-v) as rational functions of p x p 2 p'x and p'2 and so
,

f(u+v)
where

C?

Pi

GipvPvpipz),

denotes a rational function.

Pv> Pi' Pi' an<i Pz fr


aid of the identities

*^e equations

4pt9aPi9

P'i

If

(i),

(hi)

we now

(ii),

and

eliminate

(hi),

by the

4pl9iPag

we

obtain an algebraic relation connecting f(u-\-v), f(u), and


f(v); this proves the theorem.

Example. Show that


p(z-o)-p(z + x)

>'(z)/(a){p(2)-p(a)}- 2 .

The evaluation of elliptic integrals


An integral of the form F(t, u) dt, where F denotes a rational

13.7.

J"

functionf of t and u, and where

u2
is

=a

tl

+4a1 P+6a2 P+4aa t+ai

a quartic of cubic function of t without repeated factor, %


an elliptic integral. For example, the integral

is

called

a
which

is

{\-hH2 )dt

2
J V{(i- )(i-^

)}'

equal to the length of arc of an ellipse of eccentricity

It must be genuinely a rational function of t and u, and not a rational


f
function of t and u 1 In the latter case the integral can be expressed in terms
of the elementary functions of analysis.
2
J If u has repeated factor, we can evaluate the integral by means of ele.

mentary functions. See Hardy's tract

cited

on

p. 362.

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

374

it is from this fact that the


name.
class of elliptic integrals derives its
We shall now prove the important theorem that any elliptic
integral can be evaluated in terms of Weierstrass's periodic and
pseudo-periodic functions, combined with the elementary func-

k and major axis 2a,

The

tions of analysis.

of this type;

is

first

integral into a canonical

step in the proof

to transform the

is

form
G(w, v) dw,

where

denotes a rational function of

v2

w and

4w3 g 2 wg3

v,

and

where g\21g\ is not zero.


Let us consider how the expression

=a

xi +4:a l x3y+6a 2 x2y2 +4:a3 xy3 +ai yi

= l'X-\-m'Y,

lX+mY,y
behaves under the transformation x
where A
lm' I'm is not zero. The quantities

g2

=a

ai 4 a 1 a3 +Bal,

ax

2
a3

a2
are

a2
a3
a4

known to be invariantsf with respect to this transformation.

This means that,

/
and

if

=A X

if

+4:A 1 Xs Y+6A 2 X 2 Yz +4:A 3 XYs +A i

G 2 and G3

are the

A s as g 2 and g3 are in a

same expressions

Y*,

in the coefficients

then G 2
A*g 2 and G3
A 6 <73 Mores
over, since / has no repeated factors, gl27gl is not zero. We
now show that the transformation can be chosen so that A

and
It

A 2 vanish
is

easily seen that

vanishes

if

l't

where

is

a root

of the equation
4>{t)

When
of

=a

+4:a 1 t3 +6a 2 t2 +4a3 t+ai

has been chosen in this way,

A2

0.

vanishes

if

the ratio

m to m' is given by
^m^+(6^-2t

<f,;)mm'+(tUo-Qt

KW

0,

t See, for example, Elliott, Algebra of Qvantics (Oxford, 1913), 8, 21.

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS


where a denotes the value of d(f>/dt when t = t and
<j>'

for

<j)g.

m't

value of

first

which makes
second value

<f>Q

and gives the transformation

non-trivial

= (X+XY)- Y,

The corresponding value of/ is

Since

write x

y^(t)

where

Finally, if

values of

we

= X+XY,

\<j>'

we now

trivial transformation
not zero by hypothesis, the

is

is

gives the other coefficients

If

m'{t a <j>l~ &<j>' ).

m' provides a

Since

4i 1 I3 7+4i 3

Ax

m$|

or

zero.

= $5/0o-

where 6A

where

ty,

=/ee

IP+4

-^4

Qz\A\.

X = wY/A v we find that


7*{4ufi-g2 v>-g t )IAl
t

"

substitute for y in terms

where

Y*,

the invariance of g 2 and g3

1,

A y and A, we

A 3 = \g 2\A x

Y
= A^
-(w-^XAj),
"

w+XA 1
of Y and

replace the

obtain

w Mo)

= *o+-^-

With this change of variable, the general


brought to the canonical form
J

where

similarly

Hence

m=
The

375

elliptic integral is

G(w, v) dw,

w and v, and
4^g^wg^

denotes a rational function of


v

where g\%1g\ is not zero. Assuming that we can determine


numbers co x and co 2 whose ratio is not real, such thatj"
g,

60 2'

Q -*,

g3

we make the transformation w


j G(w, v)dw
t

A proof

140 '

Q -,

p{z\w v co 2 ), which gives

= J G{p(z), p'(z)}p'(z) dz.

of the validity of this

assumption

is

given in Chapter

XV.

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS


integrand is now an elliptic function of periods

376

As the

2^

and

can be evaluated in terms of Weierstrass's periodic and


pseudo-periodic functions by the method of 13.62. This proves
the theorem.
2cu 2 , it

Example. Prove

that, in the

above notation, the equation

i
implies that

= + ^'(t
ta

){p(l, gi ,g3 )^f(t )}-\

REFERENCES
H. A. Schwabz, Formeln und Lehrsatze zum Oebrauche

der elliptisehen

Funktionen (Berlin, 1893).


J.

Tannery and

1.

Prove,

J.

Molk, Fonctions

elliptiques (Paris, 1893-1902).

MISCELLANEOUS EXAMPLES
p(z)

by the use of Liouville's theorem

or otherwise, that

+ {p(2z)-e }^{p{2z)-e3}^+
+ {p(2z) - e }W{p(2z) - ejVH {p(2z) - e }^{p{2z) - e
p(2z)

yl*.

Deduce that
p'dcoj
2.

Show

-2(e1 -ea

that, if 3a be

1/ 2

(ei-e 3 ) 1

a period of

{(e 1

-e 2

1/ 2

+(61 -e3

Prove that.f
(-)

if z 1

+z 2 -f83 +4

A {^(Zr)-e

1}

1' 2

}.

p{z),

{P(z)-p(oc)}Mz+a)-p(oc)}{p(z + 2oi)-p(oc)}
3.

1/2
)

= -p'2

(oc).

0,

+(e3 -e1

n {pM-*J^+
+ (ei-e )n{P(Zr)-e
r
4

3}

1' 2

= K )(e 3 )(e e^.


4. Prove that, if oc+fi+y = 0,
P(P)P'(y)-$>'W)P(y) = P(y)P'(oc)-p'(y)p(oi) = P(*)P'(p)-p'(a)p(p)
2

&(P)-&(y)

P(y)~P(oc)

p(a)-p(P)

= P'() +
P'iy - ${P(ot) +P(P)+ My)}
= {ip(cc)p(p)p(y)-g ylK
a'(i8)

)}

3' 2

5.

Show

that
&'{e)&'(fi+aii)&'(z+a t )p'(z+a> t )

where g 2 and g3 are the invariants of

g\-21g%,

p(z).

The simplest method is to write z x = z, z 2 = a, z 3 = 0, z 4 = yz, where


= 0, and then to regard the expression on the left-hand side as an
elliptic function of z, whilst a, jS, and y are constant parameters.
t

a +/?+7

THE
6.

ELLIPTIC FUNCTIONS OF WEIERSTRASS


if n is an integer, p{nu) can be expressed as a

377

Prove that,

rational

function of p(u). Show, in particular, that

where

tji

7.

If

.F(z)

(i)

p(2u)

(ii)

p(3u)

p' (u)[p(u)
2

4p 3 -ff2p-g a

p(u)

p(2u)].

a(z+a)a(z a)a(b + c)a(b c)+a(z+b)a(z b)a(c+a)a(c a) +

+ a{z + c)cr(z
by

prove,

c)cr(a+ b)a(ab),

considering the function F{z)/a\z) or otherwise, that F(z)

is

identically zero.
8.

Prove that
1

P(z)

P'(z)

p(a)
p(b)

p'(a)

9.

Show

~2 o(z+a + b)<j(za)o(zb)o(ab)
3
3
a (z)a s (a)o

p'(b)

'

(b)

that
p'{u

P(u

^'"-"(Wo)

p(Mi)

p'iuj

pln-Vfa)

P(^i)

#>'()

P{un )

p'{un )

j\re(-l)/2

P ^(un
{

1!2!.. .n\o(u
ct"+ 1 (m

+u

)ct

+...+un II g(^A~ mm )
n+1 (u
(mi) -. a
n
)

m+1

where the product is extended over all pairs of integers A and


to n, with the restriction that A < /*.

/j.

from

Deduce that
p(u)-p(v)
p\u)

p'(u)-p'{v)
p"(u)

p"(u)-p"(v)

p'(v)

p"(v)

p'"(v)

p"'(u)

A (u

- - 12 o(2u+2v)o

v)

(Oxford, 1925.)
10.

Prove that
P'{u)

P"W

P'>)

p'"(u)

p(n-l)( M

pW(u)

p<*\u)

= 1)- {1!2!... (n1

a2
iy.y

_cr(nu)
{<j{u)}'-

(KXETEBX.

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS

378
11.

Prove that

Oi(2w)

CTj(2u)

o- 2 (2m)

<t 2 (2v)

ct 3 (2m)

<j 3 (2v)

=
12.

otX2w)
(2w)
a 3 (2w)

ct 2

i 1y.1 -,4-, :

Prove that,
(i)
(ii)

(iii)

^
if

n ,,i 'r(

M -''Wi+' W''+'W-)<r(K+MM'-)
)

the numbers

y denote

<x, jS,

1, 2,

3 in

some

order,

= a (u)crl(a) oKu)a\a),
aj,u+a)aj.u a) = CT|(w)<7i(a) (e e^)(e y (M)t7 (a),
oa (u+a)oa (u a) = CTj(w)<7|(a) (e a
(a)<T (M).
o(u+a)o(u a)

Deduce the

)CT

ef,)t7

result of

Ex.

from equation

(iii).

(Tannery and Molk.)


13.

Prove that a(z+a)/a(a)

is

equal to

m.n

where multiplication

is

extended over

and zero values of m and n without


Deduce that
or (z)

= exp(-Kz TT ((l-o- 5

14.

is

an

that, if q

e"i<u K so

and negative

integral

that

|g|

p(o

<

1,

+\

"'

)}

the function

same zeros as a(z).


an elliptic function without

integral function with the

By

showing that
deduce that

S(s)/cr(z) is

- *-*<W)
15.

) ex

Show

positive

all

exception.

n-

29

;i-^-

singularities,

+ |

Prove that
/'?i a2 \

/1T3

* = Mfe) cos fe)


ff2(2)

6XP

teJ
2

,()

/*?iZ \
= expg-)

1 1

"TT

(l

[ }

T^T" fl

1 1

+ 2g2B coS7Tz/co +24B

1:
(

+g

<l-g 1)

+ 2qin-

cosTTz/w l +q i

(1+g2 -V

)'

i
\

THE ELLIPTIC FUNCTIONS OF WEIERSTRASS


Show that, if \q\ < 1 and

16.

= n(l+2 ").
2

2i

).

= n(i-9 2n-

?3

then 2ig 2 ?3
Prove that,
1

-e,)V

= ri(i+28n=
n

1
).

e"W"\

(es-^'^-^V'W.

= ^-g8flS.

16ggf

Prove that,

are real, and e

>

if

>

'*-^^-

q%q\-

= iX, where A is real and positive, then e

wjm^

e3

(ei-*) 1

Sigma

Prove that

19.
.

2coj

/i7iz \

TT

ttz

71

( 2 )

p(s)

= JI_ +
w2

20.

=1

ot-+3 >
2^ 2o)! Zw
CO

= s?r 3+

Prove that

V
2

cosec

o^zljtu^aii)

Deduce that
21.

where

p(z|ico 1 ,co 2 )

Prove that

G2 =
Show

w=

60e?-4gr 2,

that

if

CO

2a) x

coi

sin 2 ( 7rz/2a. 1 )

COt

2<u!

,7r(z 2no> 2 )

2^

=
=

+ COt

CO

!73

e2 , e3

i\,

>!

V' cosec wco

2l

^r

CO

exp(^ 1 s 2 )cr(s)<T1 (z).

p(z)+^(z+co!)

x.

p(z\^o) 1 ,oj 2 ) satisfies the differential equation

<?3

4w

z+b = p 2 (w), where


=0.

then

x,

e2.

18. Determine the limiting forms, as A-> +oo, of the four


nXi.
functions with parameters cu 1
it, cd 2

22.

).

if

Deduce that
17.

s-2

r=l

m=l

(fl

CO

CO

CO

?o=Il(l-22n

379

146x^+22^3. (Oxford, 1922.)


dz
J (z+

the invariants of p(w) are

gr

4(6 a) and

CHAPTER XIV
JACOBI'S ELLIPTIC FUNCTIONS

The construction of elliptic functions with two


simple poles in each cell
In the previous chapter we saw that the order of an elliptic
function cannot be less than two and we considered in detail
Weierstrass's function ){z), which is of order two and has one
14.1.

We

double pole in each cell.


now introduce three other elliptic
functions of order two, which differ from p(z) in that each of
them has two simple poles in every cell; these are essentially
the

elliptic

functions of Jacobi. Whilst Jacobi's functions are

more complicated than

p(z), they possess, as we shall


advantages when we wish to obtain numerical
results in problems involving elliptic functions.
Let us consider the functions

slightly
see,

distinct

S(z)

(et-ejM

= *@,

C{z)
<7

2 (Z

)'

in the notation of the previous chapter. If

we

pseudo-periodic properties of a(z)


creased by 2cor

S(z), C{z),

D(z)

= SM,

0-2(2)

(*r

we make use of the


when z is in-

find that,

and D(z) are reproduced, save

for

a multiplier 1, as indicated below.


2 Wl

2w 2

2m 3

-1

C(z)

-1
-1

D(z)

S(z)

-1
-1

+1
-1

It follows that each of the functions is doubly-periodic,

though
each has a different primitive period-parallelogram.
The only singularities of 8{z) are the points at which a 2 (z)
vanishes,
co 2

+Qmm

and these are simple poles at the points of the set


Thus S(z), and similarly C(z) and D{z), are elliptic

functions.

The three functions are not, however, independent. For since


and 4a> 2 are periods of each of them, it follows by the
theorem of 13.4 that C{z) and D(z) can be expressed as alge4a)!

JACOBI'S ELLIPTIC FUNCTIONS

381

The explicit formulae are easily obtained

braic functions of S(z).

when we observe that


S{z)

-[m=j

C(z)

'

We find,

D{z)

-(pm=^I

2)

'

~ W)-ej

in fact, that

= {l-S^z)}

C(z)

1 '2

D()

1 '2

fl_^Z3

fif8(

z ))

the square roots being chosen so that C(z) and D(z) have the

value

+1

The

at the origin, where 8(z) vanishes.

latter relation is usually written in the

D(z)

where

&

= {l-&

S2 (z)} 1/2

= L "
K-e

form

1/2

2)

&

called the

is

modulus of the functions

Since e x e 2 and e 3 are distinct,


,

values

or

Similarly,

^1.
it must

is finite

1c

#(z), G(z), and -D(z).


and cannot have the

also be possible to express the derivative

of S(z) algebraically in terms of the three functions.

and

so,

by

From

this

2{p(z)-e 2fl*

>

'

13.611,

*>M
which gives

Now

(e 1

-e^/ 2 {g)(z)-e 1}i' 2 {g>( 2 )-e3}i' 2

'()

we deduce

(e 1

-e 2

1 /2
)

C(z)Z>( Z ).

that

=
D'(z) =
C'(z)

-(e 1 -e 2 )^D(z)S(z),
-(e1 -e 2 ) 1 2 A;2 5(2)C(z),
/

by using the formulae expressing

C(z)

and D(z)

in terms of

S(z).

14.11. General description of the functions 8(z), C(z),


D{z)

and

The function $(z) is an elliptic function with periods 4a> x and


2w 2 the points 0, 4a> 1; 4co 1 +2oj 2 2co 2 taken in order, being the
,

JACOBI'S ELLIPTIC FUNCTIONS

382

vertices of a primitive period-parallelogram. It

odd function;

K-e^^L^
=

8{-z)

CT

since a(z)

is,

moreover, an

for
(ei

Z)

2(

-e 2 )i/^
CT

-8(z),

(Z)

odd and a2 (z) even.

is

when <r(z)
0; thus the only zeros of S(z)
are simple ones at the points of the set Q. m>n As two of these
zeros occur in every cell, 8{z) is an elliptic function of order
two. Similarly the only singularities of 8(z) are the points at
8(z) vanishes only

which

o-

2 (z)

gruent to

vanishes,

a> 2

and

so are simple poles at the points con-

or t Sa^-f w 2

The residue of

8(z) at

lim Ue 1 -e i )^o{z)
z-+uj,

But
and
If

a2 {z)
so

o'z (o) 2 )

we remember

(mod 4^, 2w 2 ).

co 2 is

equal to

Z
^l)
= (e.-e^^A

a2 \Z)

= e^
=

(co 2 )

z
(j(a) 2 z)/a(co ),
2
r? 2a,2

/ (T ( a>2)-

that

(ei

_e 2 )i/2

= _ e -^ K)
ff(co )a(cu )
1
2

we

see that the required residue

is

Moreover, since the sum of the residues of an elliptic function at


poles in any cell is zero, the residue at 2co 1 +oj 2 is (e 3 eg)- 1 2
The function C(z), however, is obviously an even elliptic function with periods 4a> 2 and 2a> 3 The points 0, 4w 2 4co 2 +2o> 3 2a> 3
taken in order, are the vertices of a primitive period-parallelogram for C(z); the latticef of period-parallelograms associated
with C(z) is thus different from that of S(z). The only zeros
of C(z) are found to be simple ones at the points of the set

its

'

f The reader is advised to draw a diagram of the lattice of period-parallelograms associated with each of the functions S{z), C(z), and D(z). He will see

that the three lattices considered here are not the only possible ones for
example, we might have taken the points 0, iw^ iui^ 2a> 3 ,
2 co 3 as the vertices
of a primitive period-parallelogram of C{z).
The three lattices of period-parallelograms considered here are probably the
simplest, since each is derived from the preceding one by a cyclic permutation
;

of

to lt

w z and
,

to R ,

JACOBI'S ELLIPTIC FUNCTIONS


aj 1

+Omn

two of these in each

as there are

383
C(z) is of

cell,

order two.

The

two

singularities of C(z) fall into

poles at points congruent to


residue at

cu 2

<r

(z)
2

^KW^3.)

)2

a(o) 2 +cu 3 )

a^(co 2 )

From

this it follows that the residue at

set

i(e3 e 2 )~112

is

The

is

^ g iK) = _ e^o

they are simple


,

and congruent points

Urn (g->i(g) ]

*~<4

sets;

or 3co 2 (mod4a> 2 2w 3 ).

a> 2

i(e 3

_e y

1/2

each pole of the second

Finally, as the reader will easily show, D(z) is an even elliptic


function with periods 4co 3 and 2w 1; the points 0, 4oj 3 40)3+20^,
,

taken in order, being the vertices of a primitive periodparallelogram. It has a simple zero at each point of the set
a>3+Q mn as two zeros he in every cell, D(z) is of order two.
Moreover, its only singularities are simple poles at points congruent to co 2 or to o> 2 +2w 3 (mod4oj 3 20^), the residue at each
-1 2
pole of the first set being i(e 1
and at each pole of the
2)
second set i(e x e 2 )- 1 2
2co 1;

'

14.12.

'

The complementary modulus

The complementary modulus k' associated with


and D(z) is denned by the equation

_
and so

is

8(z), C(z),

fa-e,)*/*

connected with the modulus k by the relation


k2 +k' 2

Moreover,

k' is finite

Example. Prove

and

is

1.

not equal to

or

1.

that

K) = 1,
S(co 3 = 1/k,

C(coj)

C(o) 3 )

=
=

0,

Z(coj)

ik'/k,

D(o 3

=
=

k',

0.

14.2. Jacobi's elliptic functions

The occurrence of the

factor

derivatives of S(z), C(z),

{e x

112

2)

in the formulae for the

and D{z) suggests that

it

would be

advantageous to change the independent variable from


where u
(e 1 e 2 ) ll2 z.

z to u,

JACOBI'S ELLIPTIC FUNCTIONS

384

we

Accordingly,

define Jacobi's elliptic functionsf

by the

equations

sn

dnu
and the

d
-=-sn
du

(1 an2 **) 1

we add 4w 1 (e1

numbers

dntt
1/2
2)

(1 ^sn3 ^) 1 2

d
cnw
du

= dnMsnK,

dnu

or 4a> 2 (e 1

1/2

to u, the three func-

2)

and dntt are unaltered. Accordingly we

^^^

R=

14.1

-k 2 sia.ucnu.

iK

the quarter-periods:]: of Jacobi's

from

C{(e l e 2 )- 1 l 2 u},

Dlfae^-Wu},

A
cnuanu,

du
tions snu, cn,

off 14.1 now take the simple forms

identities

cna=

If

cna

<S{(ei e^-Wu},

when u

that,

is

>

the

= vfa-ejM

elliptic functions.

increased

call

by a

It follows

half-period 2K, 2iK',

2K Jr 2iK', snw, cnw, and dnu are reproduced, save for a


multiplier 1, as shown below.

or

SI1M

CUM
dntt

2K
-1

2iK'

2K+2W

+1

-1

1
+1

-1

+1

-1

-1

Finally, the equations denning Jacobi's


take the form

sniKz/ajj)

0>!

Example

snO
cnO
dnO

a{z)
2 (Z)'

Show

=
=
=

cn(Kz/coj)
(7

elliptic

functions

d^Kz/u^
<j

2 (z)'

now
,(*)
.(*)

that

1,

snK =
cnK =

0,

1,

dnK =

k',

0,

sn(K+iK')
ca(K+iK')
dn(K+iK')

1,

=
=
=

1/k,

-ik'/k,
0.

t This notation is due to Gudermann (1838). Jacobi wrote sin am u,


cos am u, A am u for sn u, en u, dn u, when he published his discoveries in
his Fundamenta Nova Theoriae Functionum Ellipticarum in 1829 ; he regarded
sn u, en u, and dn u as the sine, cosine, and derivative of a function am u
defined

amw
by the equation J (1 ifc'sin'fl)- 1 2 dB
/

% It should

be observed that ~Rl{K'IK)

>

u.

Cf. 14.41.

since Txaiwjwj)

>

0.

JACOBI'S ELLIPTIC FUNCTIONS

Example

Prove that Jacobi's

2.

385

functions have the Taylor

elliptic

expansions

cnw

l_i- M2+l(i +4fc 2 )M*-.,

valid near the origin.

Example
k

Example

3.

Show

that

e^-iaKo^coJ/o-Vj,),

4.

Show
^

that.f if t

4a

k'

cua/a^

e<"t-fl.K t r(a)/ff(<D

and g

2B )/(l+3 are- 1

2B 1

frrrf[

{(l

4
)}

)}

frr

iK'

e,

n {(l+2

n {(i-2 "- )/(i+g K = n {(l-g^m+g "*'

1 ),

)*},

g^jsji+gM-i)^

m=l

Deduce that, when t is purely imaginary, the four numbers &, ,


are real and positive and that k, k' are less than unity.

Example

5.

k->0,
14.21.

K, K'

Prove that.t as Imr-s- +oo,

k'-^l,

K->%rr,

The general properties

K'-log(i/k) ->

0.

of the functions snu, cnu,

and dau
It is convenient at this stage to translate into the new notation the properties of S(z), C(z), and D{z) discussed in 14.11.

The only

real difficulty is the determination of the residues.

We have

already seen that


S(z)

Use the

We

i-<p

result of Ex. 16 on p. 379.


shall see in 14.4 that, when k is given, the principal value of
t is
determined, and that, when h -* 0, the imaginary part of this principal value

tends to infinity. Thus k


4111

--

implies that

CC

K -* Jw and K' Iog(4/fc) ^> 0.

JACOBI'S ELLIPTIC FUNCTIONS

386

where

<D

denotes a function which

is

regular at 2

o> 2 .

It

follows that
-

where

<1>

(ei-e 2) lf2

+<]>,
+<d
*(-*')
(c-e,) ^-^')^
Thus snw. has a simple pole
is regular atw =

ni
sntt
,

1'2

i-K"'.

of

residue 1/fc at every point congruent to iK' (mod ^K,2iK');


1/k at every
similarly, it also has a simple pole of residue
and dnw are
of
cnu
residues
point congruent to 2K+iK'. The

obtained in the same way.


We see, therefore, that snu is an odd elliptic function of order
two with primitive periods 4K and 2iK'. Its only singularities
are at points congruent to iK' or to 2K+iK' {moc\K,2iK');

1/k respecthese are simple poles with residues 1/k and


tively. Its only zeros are simple ones at points congruent to

or 2K.

A slight

difficulty occurs

when we come

to translate into the

new notation the properties of en u and dn u. It will be remembered that the lattices of period-parallelograms of C(z) and D(z)
were obtained from that of S(z) by cyclic permutations of the
parameters
there

is

<o l5 co 2 ,

no special

and co 3 In the theory of Jacobi's functions


symbol for co^ e 2 ) 1/2 and a certain lack of
.

We

we wished, take the parallelovertices 0, UK', 2K+2iK', 2K2iK' as fundamental for cnu. But since every period of cnu is of the form
4mK+2n(K+iK'), it is usually found more convenient to take
the points 0, 4K, 4K+2{K+iK'), 2K+2iK' as the vertices of

symmetry
gram with

results.

could, if

the primitive period-parallelogram of cnu.


The function cn is, then, an even elliptic function of order
two with primitive periods 4iJT and 2K+2iK'. Its only singularities are at points congruent to iK' or to 2K+iK' (mo&lK,
2K+2iK'); these are simple poles, with residues i/k, i/k
respectively. Its only zeros are simple ones at points congruent

to

K or 3K

elliptic function of order two with


Its only singular points are
and
UK'.
primitive periods 2K
2if, 4i.fi:'); these are simple
(mod
UK'
congruent to iK' or to
Its only zeros are
respectively.
poles of residue i and +i

Finally,

dnu

is

an even

simple ones at points congruent to

K+iK'

or

K+BiK'.

JACOBI'S ELLIPTIC FUNCTIONS

387

14.22. Glaisher's notation for quotients of Jacobi's functions

convenient to have a notation for the reciprocals and


quotients of Jacobi's elliptic functions. We shall write
It

is

sn/cn
cnu/snu

1/cnu

= sew,
= csu,

This notation

= ndtt,
cnit/dnw = cdw,
dnu/cnu = dcu.

= new,
sntt/dnw = sd,
dnu/snu = dsu,

1/snu ='nsu,

l/dn

due to Glaisher.f

is

The addition theorem


Since snu, enw, and dnu are

14.3.

elhptic functions, each of these


functions possesses an addition theorem whose form we shall

now

determine.:]:

Let us consider the function


F(u)

where

= en ucn(u + A snusn(u

and a are

ex)

constants.

It is

2K and

2iK'.

primitive periods are

an

elliptic

at),

function whose

not congruent to
K, iK' or K+iK' (mod2K,2iK'), each term in F(u) has two
simple poles in each cell at points congruent to iK' and u+iK'.
We can, however, choose the constant A so that the principal
part of F(u) at iK' and congruent points is identically zero;
thus F(u) is an elliptic function with only one pole in each cell
and so is a constant, by Liouville's theorem. Putting u
0, we
see that F{u)
en a.
The appropriate value of the constant A is determined by
observing that, when \u\ is small,
If

a.

is

F(u)

Hence

A=

cna+ttdnasna uA

dn a and

so.

<x+0(u2 ).

so

cnucn(u a)+musxx(u a)dna

= cnac.

(i)

This formula is true provided that a is not congruent to iK',


the restrictions that a. is not congruent to
or K-\-iK' being
relaxed by an appeal to the principle of analytical continuation.

t Messenger of Math. 11 (1882), 86.


t The method of proof given here is a slight modification of that of Hurwitz
and Courant, Funktionentheorie (3rd edition) (Berlin, 1929).

&&

JACOBI'S ELLIPTIC FUNCTIONS

388

In a similar manner we can show that, when a


to iK',

(ii)

= u+v in equations and


we obtain
c c s s dn(u+v) = cn(u-\-v),
d d lc s s cn(u+v) = dn(it+^))
brevity, we have written s = sna, d = dnv, and

we now put

(i)

a,

and

not congruent

= dna.

daudn(u a)+k2 snusa(u a)cna


If

is

(ii),

where, for

so on. These equations give at once

{u+v) =

Finally, writing a
s1

from which

= v in equation
d sa(u-\-v) = c c
2

this last equation

(i),

we

'

i-vbk
obtain

cn(u-\-v),

follows that

it

sniu+v)

Now

{u+v)

l-wr'

sn(u+v)=

8l (l-a

i-VAi

can be written in the form

)^(l-^l)V+* a (l-a;P(l-^) 1/a

j-^

-,

on using the identities connecting the functions sau, can, and


dnw. This equation expresses sn(u+v) as an algebraic function
of sn u and sn v, and so is the addition theorem for sn u. Similar
remarks apply to the formulae for ca(u-\-v) and dn(u-\-v).
We have thus shown that the Jacobian elliptic functions
possess the addition theorems

snttcnwdn^+sni'ciiadnM
sa(u-\-v)

sn2 ttsn2 v

ca{u-\-v)

cnucnv dnwdnflsnwsnw
1 2 sn2 sn2

dn(w+v)

Aa.uAa.v- Wsnusn.vcin.ucnv
1

Example

ea(u-\-K)

fc

sn2 sn2

Prove that
cdw,

sn(u+K+iK')

on(u-\-K)

k'&du,

dn(u-\-K)

k-idcu,
cn(u+K+iK')
dn(u+K+iK') = ik'scu.

k'ndu,

ih'k^ncu,

JACOBI'S ELLIPTIC FUNCTIONS

Example

2.

Deduce from Ex.

sn(u+iK')

Hence show

that,

when
sn(

_1 nsM,

cn(u+iK')

dn(u-{-iK')

= icsu.

fc

\u\ is

,.

dsu,

1.2

= --t< + os *+

3. Prove that

cn(M + a)cn(w a)

=
=

dn(u+<x)dn(u a)

sn(w + a)sn(M a)

Example

2F-1.

dn 2m

i^+...,

dn(+iK')

Example

= ik~

small,

MW = ^ +

389

that

4. Writing

(sn 2

(cn

sn a)/(l &
2

a dn asn
2

sn 2 Msn 2 a),

M)/(l

&

sn 2 wsn 2 a:),

(dn <xk cn <xsn u)/(lk 2 sn i usn 2a).

s, c, d,

D for snw, cnw, dn, sn2w, en 2m,

S, C,

respectively, prove that

_ 1 2s + k s
2

_2scd_

"l-W

_l 2k s + k s
1 AV

l-fc 2s*

2 2

'

i
"

Deduce that

Example
sn iK

Example

Show

5.

1/V( 1

6.

* = ?? *>

" = ?

'"ItS'
thatf

en pC

fc'),

VA'/V( 1

+ &'),

dn Jit

V/b'.

Prove that
sn(w+K:)

u + en u dn w
cn 2 M + fc'sn 2 w

fc'sn

(l+&')~ 1/22 ~

14.4. Jacobi's elliptic functions of given

modulus

Weierstrass's elliptic function

obviously a homogeneous function of the three variables z, wj,


ai 2 of degree
2; in particular, the constants e x e 2 and e 3

is

and

homogeneous functions of
Hence the function
are

aj x

and

8{z)

co 2
1/2

= -^-^

f In this result, the branches of the square roots of k' and


requiring them to be positive when
k
1.

by

<

of the same degree.

<

1+k' are defined

JACOBI'S ELLIPTIC FUNCTIONS


homogeneous of degree zero in z, a^, and <o 2 and so
form F(z/co v t), where t = coa/o^. But this gives
390

is

is

snu

^(ej-ea)-1 2}
/

of the

Flt^e^-^/a^T}.

1 2
a) 1 (e 1 eg)
is a function of t alone, it follows that snu depends only on the independent variable u and the parameter t;
similarly for cnu and dn.
This means that, when t is assigned, we obtain the same
Jacobian elliptic functions, no matter what pair of parameters

As

and w 2 we employ in

cu 1

their construction, so long as the ratio

equal to r.
Similarly the quantities

u) 2 /coi is

(e3

-e 2 )W

K = co^-ea)

1' 2
,

iK'

(%-e,)*"

= tofa-ejW

are functions of t alone; moreover, the explicit formulae, found


in

14.2,

Ex.

4,

Conversely,
1,

show that they are analytic

functions, regular

> 0.

when Imr

when k2 has a given value

c different

from

or

the Jacobian elliptic functions are uniquely determined. (This

fact

is

often emphasized

so on.f)
(e 3

To prove

2 )/(e 1

is positive,

2)

and

this,

by denoting them by
we need to show that

= c always has a root

sn(w, k),

and

the equation

t whose imaginary part

also that every root of this equation leads to

the same set of Jacobian elliptic functions.


This problem is completely solved in Chapter

XV.

We

show

there that the equation always has a root r lying in a certain

'fundamental region' of the upper half of the r-plane, and that


any other solution is of the form (ar +2&)/(2cT +^), where a, b,
c, and d are integers such that ad4bc = 1. From this it is
deduced that the functions sn(u, k), cn(u,k), and dn(u,k) are
uniquely determined when k2 is given.
The equation (e3 e 2 )/(e 1
c defines t as a many-valued
2)
function of c, with branch-points at c
0, 1, oo. Each branch
is regular in the c-plane supposed cut along the real axis from
and from 1 to +oo. As c -> the principal branch t
oo to
tends to infinity along a line parallel to the imaginary axis.

The notation

is

rather unfortunate, as k 3

is

given, not k.

JACOBI'S ELLIPTIC FUNCTIONS

391

These branch-points of r, regarded as a function of c, are such


co, and is
1 and c
that eviT has branch-points only at c
axis from
the
real
regular in the c-plane supposed cut only along

to +00.
Finally, sincef

*K' = tK,
K = \* ft {(l-g (l+2 where q = e" these results imply that K is an analytic function of
regular in the cut c-plane, and that K is also regular
2rl

27l 2

1 4
)

},

iT

c,

'

provided that an additional cut

from oo to
14.41.

When

Jc

is

an

of

not equal to
f

is

madej along the

real axis

0.

The evaluation
2

is

an
or

elliptic integral
1,

the expression

(l_ Z2)-l/2(l_fc2 Z2)-l/2

elliptic integral.

We now

show

d,Z

that,

when

the path of

integration does not pass through a branch-point of the integrand, we can evaluate this integral by means of the substitution
z

= sn(w,

Jc).

= m(w, k) defines w as a many-valued funcFor if w f{z) is any solution of this equation, then
w = f{z)+4mK+2niK'
w = 2Kf(z)+4mK+2niK'

The equation
tion of

z.

and

are also solutions for all positive

values of

m and n.

and negative

= (l- 2)-i/2(i_F2
dz
z

the only singularities of


Z =

and zero

integral

Since

2 )- 1 / 2

are branch-points at z

and

l/lc.

There

is,

however, a unique solution of the equation which

regular in a neighbourhood of the origin and vanishes at the


F(z), say, is a conorigin. The existence of this solution w

is

sequence of the inverse-function theorem of 6.22, since sn(w>, k)


has a simple zero at the origin and is regular when \w is small.
\

t See 14.2, Ex. 4.


% Actually the cut from
See 14.44.

c= ltoc=

+oois unnecessary

in the case of K'.

392

JACOBI'S ELLIPTIC FUNCTIONS


As the path of integration does not pass through a branchpoint of the many-valued function w{z), we can continue F{z)

analytically along

it. Accordingly the transformation w


F(z)
determines a definite path of integration in the w-plane, which
starts at the origin and ends at a point v, where v is a certain
solution of the equation sn(v, k)
s.
The particular solution
v which has to be taken evidently depends on the manner in
which the path of integration in the z-plane passes between the
branch-points 1 and l//fc.

If we now make the substitution z


m(w, k) and suppose
that the integrand has the value
1 at the origin, we obtain

S
j*

Example

1)

(l-S)-l/(l-jy2*)-V dz

= j dtU = V.

Prove that
i

(i)

(l-t*)r 1 l*(k' i +kH*)- 1 l*

dt,

cnu
1
(ii)

(iii)

= j (**+ l)-i/ty+Jfe')-i/ dt,

(l* 8 )-1/*^ &')-!/(&,

00

CSM

sdw
(iv)

J"

- &'*)-!/*( 1 +

2 2 )- 1 / 2

A; *

dt.

Example

2.

Show

that, if

k2

is

not equal to

or

1,

(l

WsmHyWdt =

v,

where v is a solution of the equation sn(v, k) = sin 0. (This equation


was used by Jacobi as the definition of his function 6 = amo.)

The expression

14.42.

We

saw

in 14.4 that

of

K in terms of k

K is an analytic function of c

(= k2 ),

regular in the c-plane supposed cut along the real axis from
shall now show that the equation

to +00-

We

j (1
o

^^(l ta)-y* dx = K,

JACOBI'S ELLIPTIC FUNCTIONS

where the path of integration


where in this cut c-plane.

By

14.41,

where
1

393

is a straight line, holds every-

the value of this integral

is u-\-4mK-\-2niK'
a zero of 1 snu. But as
function of order two with a double zero

m and n are integers and u is

sn is an elliptic
u = K, we have

at

(1 a*)-V*(l ce)-i/ dx

(4m+l)K+2niK',

J"

where m and n are integers to be determined.


Let us consider first the case when
<c

<

value of the integral

In

principal value of t

K' are

is

and

also real

real

is

and

positive.

this case the

purely imaginary, and therefore


This implies that

0.

2)-i/2(l- Ca; 2 )- 1 2 dx
j (l_ x
/

K and

m^

positive.

But if m were positive, there would


between and 1, such that

so that the

1,

and

a number x

exist

= K;

we should then have sniC

m=

=x <

1,

which is impossible. Hence

0.

We have

thus shown that


i
2 )-i/2(i_
2)-i/2
ca;
dz
j (l-a;

=K

< c < 1. The truth of this formula when c lies anywhere in the cut plane immediately follows by the principle of

when

analytical continuation.

K expressed as a hyper geometric function


If we write x sin 5 in the integral-formula for K,

14.43.

that

we

find

j,

K = j (l-csinW)-

1'2

dd,

<

when c lies in the cut plane. When |c|


1, we can expand the
integrand as a series of ascending powers of c which converges
uniformlyf with respect to 6 and so can be integrated term by
f

By Weierstrass's

ilf-test,

since

<

sin2

<

1.

JACOBI'S ELLIPTIC FUNCTIONS

394

term. This gives immediately

where

k2

By the principle of analytical continuation, this formula holds


everywhere in the c-plane, supposed cut along the real axis from
1 to +o, since the expressions on each side of the equation
are both regular all over the cut plane.

The expression of K' in terms of k'


We now prove that K' is the same function

14.44.

of

k'

as

K is of

k, by considering the effect of renaming the parameters a^ and


<a 2 which occur in the definition of Jacobi's elliptic functions.

we

= o^,

toi+coj+cog
^> * ne se^ ^
points 2m'a>[-\-2n'<o'2 where m' and n' take all positive and
negative integral and zero values, is identical with the set of
periods 2mco 1 -\-2nu> 2 of Weierstrass's function p(z|to 1 ,a) 2 ). It

If

write

co^

o> 2 ,

o> 2

follows that
p{z\u[,a> 2 )

= p(z\w v

and

a(z\co[, oj'2 )

l(z\w[,oi'2 )

a) 2 ),

a(z\w 1

(z|aj 1; a> 2 ),

cu 2 ).

On the other hand, the constants which occur in the equations


w 2 and

expressing the pseudo-periodicity of (z|wi,


are no longer tj 1 and ij 2 For, if

a(z\io[,o)2)

= 2r][+Z(z\co[,a>
= (
w =
% = +7j

Z,{z+2a>[\u}'1 ,<jj'2 )

then

and

7]'

(coi|co{,to
v

2)

similarly

r}'

Hence we obtain a
if

we

r]

2 ),

co 2 |aj 15

2rj 2

2)

rj 2 ,

different set of Jacobian elliptic functions

use the parameters co'x and co 2 instead of <x} 1 and co 2


find the modulus and quarter-periods of this new set
.

Next we
of

elliptic functions.

Using an obvious notation, we have

k(a>[, co 2 )

3-ea ) 1/2

(ei-e^l*

{pM-e&l* =

{pCoD-W

{pfa-W!)-^!*
{P(-^2)-%}

by the definition of (er ea )^ 2 Now {p{z) ej 1 2


function with periods 2ta x and 4w 2 hence
/

is

1/a

'

an odd elliptic

since

V
(e 3

2)

- ei

)i' 2

{PW-JW ~ "(e^) " K-e,)


= ifo-e,) ", (e,-^) ^ = -i{e -e fl\

1'2

1'2

'

It

JACOBI'S ELLIPTIC FUNCTIONS


follows, therefore, that k{w[, w'2 )

k'ico^

395

w 2 ), and

similarly that

= k(u) v co 2 ). Thus the modulus and complementary


modulus are interchanged by the interchange of parameters.
Again, we have
k'(w[, co 2 )

K(a>[,a>'2 )

= uttei-etfl* = -oApi-uJ-eJ* =

vfa-ejV*.

which gives at once

= K(a>i,

K' (co[, ca'2 )

similarly

Hence, when k
K'(a> r

o) 2 )

not real and greater than unity,

(coi,a> 2 ) is

= K(w{,

co 2 ).

ojg)

= f (l-a*)-*l{l-&(a>i,to3a?}-i

l*

dx

o
i

= {l-x^-^l-k'^co^^x }2

1'2

dx,

K'

that

is,

save

when

k' 2

= j (l-x )- /2(i_fc'2
2

1.

a;

2)-i/2

dX}

This proves the required result.


see that K'

by using the result of 14.43, we


function of c (= k 2 ), regular in the

Further,
analytic

is

an

c-plane supposed

cut along the real axis from


oo to 0; the additional cut from
1 to +oo, imposed in 14.4, is unnecessary. The explicit

formula for K' as a function of

14.45. Jacobi's

c is

imaginary transformation

We

have just seen that the primitive periods of sn(, k')


are 4K and 2iK, and hence those of sn(m, k') are 2K and AiK'
Therefore the functions sn(w, k) and sn(w, k') have a common
pair of periods 4K and UK', and this implies that they are connected by an algebraic relation which we shall now determine.
By definition, we have
'

7(z\co[,w'2 )
'

K[

_ K' o(z\w[,

z (z\coi,co 2 )

co{

cr(z)aK)

_ Z'

(i)'

)a(a)2 \u>[, u> 2 )

e-^V(z+U2|a)i,co2)
g(g)

JACOBI'S ELLIPTIC FUNCTIONS

396

But iK'JK

tojw^ and so K'/to^

K'/w 2

= iK/w^

Hence

snl
\ 0>!

Putting ifz

CO!

oi 1 u,

or

2 (z)/

is

\o>!

2 (z)

isc(, &),

From

the required algebraic relation.

= nc(u, h),

cn(iu, k')

we obtain
sn(ra, k')

which

ff

da{iu,k')

this it follows

dc(u,k).

These three equations constitute Jacobi's imaginary transformationf of his elliptic functions.

Example
equation

1.

Deduce Jaoobi's imaginary transformation from the


8n(.Jfc'>

(l-z a )-*l*(l-k'*zi )- 1 2 dz

iu,

by the

substitution z

Example

i/(l

2 1/ 2
)

2. Prove

by the aid of Jacobi's imaginary transformation

*/V*.

en \iK'

that

sn %iK'

the square roots being positive

Example

V( 1

when

dn fyK'

k)Wk,

<

<

V( 1

+ *),

1.

3. Prove that

14.46. Landen's transformation

The theory of the transformations of elliptic functions is concerned with the expression of the Jacobian functions of parameter t' in terms of those of parameter r, where
T

'

(aT+b)/(CT+d),

and d being integers such that ad be is positive. A very


simple instance^ of this is provided by Jacobi's imaginary
transformation, in which r'
1/t.
a, b, c,

The general theory of these transformations is beyond the


scope of the present book; we shall consider only Landen's
t Fundamenta Nova, 34-5.
} For others, see Ex. 18 on p. 416.
See,

195-232.

however, Tannery and Molk, Fonctions

elliptiques,

(Paris, 1896),

JACOBI'S ELLIPTIC FUNCTIONS

= 2t.

transformation, defined by r

This

397

of historical interest,

is

being the first case ever discussed, t


In order to determine Landen's transformation, J we consider
first the connexion between Weierstrass's Sigma functions

and

ff(z|a> 1 ,a) 2 )

o-(z|coi,a> 2 ),

where

these give

fl
It

= ^w

and

<o 2

= a>

2,

since

= 2^? = 2t.

shown that

easily

is

co{

being understood that the parameters of a(z) are w 1 and a> 2


unless the contrary is explicitly indicated. From this it follows
that
it

a|coi,oi)

e li

8-, 1 +()+S(z+co 1

)>

and hence that

and

= eiCOg

172

i7 1

+S(w 2 )+^(a)1 +w2 )

= Cj 0)2+2^.

Let us now suppose that the Jacobian elliptic functions con(j(z\a)[, w 2 ) have quarter-periods L, iL' and modu-

structed from
lus

I.

Then

m-m

sn

(2zL

_ _L

a{z \o}[,

wi

= 2_L

q)a)g(tt)a |a>i, ta 2 )

e-^^crCz+coaloji.wa)

y+

= 2iy e

2r7, 8

&>!

>(*,>(, +,

'h z+ie
Z+
Z

exp{(2

^ +eia>a)2+Kz2 _

"I h ^2 ici(2+a) 2 )

+ ^(z-f w

2 )}

<T
a> l) Cr a> 2) g
l+ a, 2)
a^z+^aiz+^+ai^ia)])

(7 (

PM.

tt)

Trans. 65 (1775), 285. Landeil, however, deals Only with


the idea of 'inverting' an elliptic integral to obtain an elliptic
function (due to Abel and Jacobi) not having been introduced until 1826.
t A simpler but less obvious method of finding Landen's transformation is
suggested in Ex. 2 at the end of this section.
See Ex. 20 on p. 379.
t Laaden,

elliptic integrals,

398

JACOBI'S ELLIPTIC FUNCTIONS


_^

2L

a{z)a1 (z)

ai t

a 2 {z)a 3 {z)

= K?4 d (?4
s

Kz

Writing

cojtt,

we have

Ksn(2Lu/K, I)

2Lsn(u, k)cd(u, k).

(i)

It remains to determine the relations connecting the quarter-

periods and moduli of the

two

sets

of Jacobian functions.

The

relations connecting the quarter-periods are


putting u
in this equation. This gives

= \K

Ksn(L,l)

and hence, by

14.3,

Ex.

found by

2sn(pr,&)cd(pr,fc),

5,

L=

l{l+k')K.

Using this value of L/K, equation

sn{(l+>,Z}

(i)

becomes

(l+ft')sn(,*)od(,ife).

Moreover, since

2L/L'

we

also

have

Again,

if

L'

we

2ico[/a>^

write equation

v-\-iK'

we

L'dc(v,k)

kHn(v,k)

= K/K',

(l+k')K'.
(i)

L' m(, k)cd{u, k)

and then put u

ia>Jco 2

in the

form

= K' sn(L'u/K',

I)

obtain

K'
_
~ lBa.(L'v/K',iy

Multiplying the expressions on each side of this equation by v


and then making v tend to zero, we have

^l
K
B~

L'V

and so

=
i' 2

{l+k'f

1+*'

We have thus proved that


&n{{\+k')u,l)

where

(l+k')sn(u,k)cd(u,k)

(lk')J(l+k'). This
transformation.
I

is

the result

known as Landen's

JACOBI'S ELLIPTIC FUNCTIONS

Example

cn{(l+k')u,l}

dn{(l+fc>.0

Example
f

2.

399

Prove that, in the notation of the previous section,

Show

=
=

{l-(l+k')sa%u,k)}nd(u,k),
{l-(l-k')sD?(u,k)}nd(u,k).

that Landen's transformation

(l-PsinV)-1 8
'

^=

is

equivalent to

(l-^sin^)"1 2 ^,

(1+fc')

'

J
o

sin&

where

(14-fc )sinfl1

oo8^(l-* , sinei )-1

'i>.

Obtain this result also by a direct transformation of the

14.5.

Legendre's three kinds of

integrals.

elliptic integral

We shall now show that the problem of evaluating the general


elliptic integral

J F{t, u) dt,

and v?

where

F(t, u) denotes a rational

a quartic or cubic polynomial in


t without repeated factors, can be reduced to the evaluation of
three very simple types of elliptic integral.
Since F(t, u) is a rational function of t and u, we can write

function of

and

u,

F(t,u)

where iV and

is

= N(t,u)/D(t,u),

denote polynomials in

and

u.

Now

u) is a polynomial in t and u2 since it does not


alter when we change the sign of u, and so becomes a polynomial
in t alone when we substitute for v? in terms of t. Thus
D(t, u) D(t,

uF{t,u) = uN(t,u)D(t, u)jG(t),


denotes a polynomial in

where

we multiply out uN(t,u)D(t,


terms of t, we obtain ultimately
If

uN(t,u)D(t,

where

<I>

and

t.

u)

'

for

u%

in

) = {t)+uY(t),

\F denote polynomials in

F(t u)

and substitute

t.

Hence

^uW) + WY

As the second term in this expression for F(t,u) can be integrated by means of elementary functions, the problem of
evaluating the general elliptic integral has been reduced to the
discussion of the simpler integral
T $() dt

J G(t)u

400

JACOBI'S ELLIPTIC FUNCTIONS

Let us suppose, in the


v

where

<x,

/?,

and

y,

first instance,

that

(t-ac)(t-l3)(t-y)(t-Z),

8 are,

by

Making the

hypothesis, distinct.

transformation!
z2

we

= {08-8)(<-)}/{(a-8)(<-)8)},

easily find that

V{(a-y)(/2-S)}^

where

A;

cubic in

say

,,.

,.

the appropriate transformation


at once
,,

v( a

where k2

^(lrj ff_^

{(a 8)(j8 y)}/{( y)(j8 8)}.

==

7) dt

is z 2

If,

however, 2

where

<J{(l-z 2 )(l-k 2 z2 )}'

H f = / ^Ha-^a-*

.R(z 2 ) is

*)}- 1 *

or

2 be a

&>

is

finite

con-

1.

The next step is to express R(z 2 )


The general elliptic integral

in z2

a rational function of z2 and k2

stant, not equal to

(t<x)/(tfi); this gives

(/3 y)/(cc y). Hence, no matter whether


quartic or cubic, we have now shown that

is

2dz

,,,

sum of partial fractions


thus reduced to a sum of

as a
is

integrals of the following types,


2 2 '-{(l-3 2 )(l-/fc 2 2 2 )}-i/2 dz,
J"

(l+w2 )-8{(l-z2 )(l-A; 2z2 )}-1

/2

dz,

J*

where r and

s are integers, s being positive,

and v

is

a con-

stant.

by using well-known methods of the integral calculus,


easy to obtain reduction formulae by means of which each
of the above integrals can be expressed in terms of one of the
Finally,

it is

t For an alternative method which only involves the use of a homographic


transformation, see Ex. 29 at the end of the present chapter.

&

JACOBI'S ELLIPTIC FUNCTIONS

401

three standard forms of Legendre:


2
2
1
j {(l-z )(l-Fz )}-

/2

dz,

z 2 {(l-z2 )(l-&V)}- 1 ' 2 dz,

2 )- 1
{(l-z2 )(l-Pz2 )}-1
j (l+vz

These are called the


first

The

elliptic integral of

we make

dz.

first,

The evaluation of the

kind has already been discussed in

14.51.
If

of the

elliptic integrals

third kinds respectively.

'

and

14.41.

the second kind

the substitution z

-1
2
2
2 2
z )} ' 2 dz,
J z {(l )(l
2
J sn (, k) du. But since

= u

sn(w, k) in Legendre's in-

we

tegral

dn2 (u, k) du

second,

integral of the

2
(

find

that

it

becomes

sn?(u, k) du,

the latter integral could equally well be taken as the standard


elliptic integral of the second kind, and this proves to be more
convenient. It

is

usual to write
u

E(u, k)

dn2 (, k)

du,

being supposed that the path of integration does not pass


through a pole of the integrand. When it is unnecessary to
emphasize the modulus k, we shall denote the function more
it

briefly

We

by

E(u).

shall

now show

that E(u)

is

an odd analytic function of

u, regular save for simple poles of residue

2mK+(2n+l)iK'.
To prove this, we observe that dn2 u

is

+1

an even

at the points
elliptic

tion of periods 2K, 2iK' whose only singularity in

any

funccell is

a double pole of residue zero at the point congruent to iK'. It


round any closed contour which
does not pass through a pole is zero, and hence that E(u) is
a one-valued function. Moreover, since the derivative of E(u)
is dn 2 M, E(u) is an odd function whose only possible singularities
are the poles of dn 2 u. We know, however, that, if a is a pole
follows that the integral of dn2 u

ofdn2 w,
4111

6xx 2

u=

-(u-a)-2 +<f>(u),
Dd.

JACOBI'S ELLIPTIC FUNCTIONS

402

where

<f>(u) is

where O(m)
residue

regular at

is

hence

a.;

E(u)

(ucx^+Qiu),

regular at

a.

Thus a

is

a simple pole of E(u), of

+1.

is not an elliptic function,


any
set of its poles can never
since the sum
when u is increased by
that,
be zero. We can, however, show
2K or 2iK', E{u) is reproduced, save for an additive constant.
For we have
2K
U+2K
E{u+2K)E(u) = f dn2 w du = j dn2 w du,

It should be noticed that E(u)

of its residues at

u
since

a period of the integrand, and hence

2K is

E{u+2K)-E(u)
Putting u

= -K, we find that

Similarly

E(2K)

we can show that


E(u+2iK')-E(u)

E(2K).

2E{K).

E(2iK'),

but no simpler form of the constant can be obtained at the


present stage, since iK' is a pole of E(u).
14.511.

The complete

elliptic integral of the

second kind

The constant E(K) is called the complete elliptic integral of


the second kind, and is usually denoted by E. If we write
$nu = sin# in the equation

K
f

dn2

du,

o
ot/2

we

E=

find that

1 2
j (l-csin^)
'

dd,

is an analytic
k2 From this it can be shownf that
c
function of c, regular in the c-plane supposed cut along the real
1 to +oo, and that
axis from

where

We
k'

as

Since E' is the same function of


regular in the c-plane supposed cut along

write E' for E(K',k').

is

of

k, it is

t Cf.

14.43.

JACOBI'S ELLIPTIC FUNCTIONS

the real axis from

Example

oo to 0,

Show

Example

->

that, as h

dc

ks

Example

(K-E)/k3 -> 77.

9
= E ~ Kc
dc~

c'

= k'

c'

'

cc'

'

2 = K'~ W
c

form

2. Prove that

~
2~
=E K
dc

where

in the

0,

E'->1,

E-^i-rr,

403

and can be written

'

dK _ K'c-E'
'

<>

dc

'

cc'

'

2.

3. Show,

by using the

results of Ex. 2, that

EK'+E'K KK'
is

independent of

c,

and

find its value

by making

-v

0.

14.512. Jacobi's imaginary transformation of E(u)


To find the effect of Jacobi's imaginary transformation on
the function E(u, k), we write
in

E(iu, k)

= J dn

u
2
(,

k)dt

and so

ij dn 2 (iu, k) du

2
j dc (, k')

du,

E{iu, k)+iE(u,

k')

[dc 2 (,

k')+dn^u, k')] du

-ij\l+^{sn(u,k')do(u,k')}

From

du.

this it follows that

= iu+isn(u,k')dc(u,k')iE(u,k').
particular, by putting u = 2^', we obtain the formula
E(2iK',k) = 2iK'-iE(2K',k') = 2i(K'-E').
E(iu,k)

In

Hence, by

14.51,

E(u+2iK',k)

E(u,k)+2i(K'E').

14.513. Legendre's relation


Since the only singularities of E(u) are simple poles of residue
at the points 2mK+{2n+l)iK',

+1

E(u) du

2m,

JACOBI'S ELLIPTIC FUNCTIONS

404

when C

is

with vertices K,

the parallelogram

K+2iK',

K+2iK', and K. Hence


2Tri=

{E(u)E(u+2iK')} du

{E(u+K)E(uK)}du.

-K

When we make
this

use of the pseudo-periodic properties of E(u),

becomes
2wt

= -UK{K'-E')+UK'E,
EK'+E'K-KK' =
\tt.

or

This relation, which


the formula

i\

is

r^wx
x w2

due to Legendre, is the analogue of


\n which occurs in the theory of the

integral (z) of Weierstrass's function p{z).

The quasi -addition theorem

14.514.

for E(u)

Let us consider the expression


F(u)

E(u)+E(v)-E(u+v),

regarded as a function of u. By the pseudo-periodicity of E(u),


it is an elliptic function with periods 2K and 2iK'. It has a pair
of irreducible simple poles iK', iK'v, and two irreducible
simple zeros 0, v. But snusn(u+v) is an elliptic function of

same irreducible poles and zeros


as F(u); thus snusn(u+v)/F(u) is an elliptic function without
singularity and so is a constant.
When i*| is small, we have

periods

2K and

2iK', with the

snusn(u+v)

u{l-S'()}+0(w2 )

f(S)

and

usnv+Q(u2

snt;

1-oW ^ Q
f

.,

"'

therefore

snusn{u-\-v)

l'(it)

sni;

1 dn2

t>

k2 snv'

for all values of u.

We

have thus shown that

E{u)+E(v) E(u+v)

k 2 anu$nvsn(u+v),

the quasi-addition theorem for E(u). There can be no


true addition theorem, as E(u) is only pseudo-periodic.

which

is

JACOBI'S ELLIPTIC FUNCTIONS

405

14.515. Jacobi's Zeta function Z(w)

For some purposes it is desirable to have a function which is


of the same general character as E(u), but is simply-periodic.
Such a function

is

Jacobi's function Z(u), denned

tlon

Z()

by the equa-

E(u)-uEIK.

It is easily verified that

Z(u+2K)

Z(ic+2iK')

Z(),

on taking into account Legendre's

Z(u)iri/K,

relation.

Thus Z(u)

is

simply-periodic with period 2K.

Example. Show

that

Z(u)+Z(v) Z(u+v)

14.52. Jacobi's

k 2 Bnuanvsn(u-{- v).

Theta function 0()

We

now introduce Jacobi's function Q(u), which plays a part


in the present theory similar to that of o{z) in the work of
Weierstrass.f This function

is

defined

by the equation

S{u)

= 0(0) exp | Z(tt) du,


o

a constant which we shall choose later. We show


that (u) is an even integral function of period 2K, whose only
zeros are simple ones at the points 2mK-\-{2n-\-\)iK'
For if aisapoleof Z(u) we know that Z(u)
(u<x)- 1 +^(u),

where 0(0)

is

where

is

<f>(u)

regular at

<x,

and

so

two determinations of

u
/

Z(u)du, along different paths of integration,

multiple of 2m. Hence log0(w)

&(u)
(u)

itself is
is

a many- valued function, but

one-valued. Moreover, as Z(w)

is

an odd function,

possible singularities of Q(u) are the poles of Z(u).

log(M)

= log(w a)+

0>{u) is regular at a,

()
t In fact,
factor.

by a

even.

The only

where

is

differ

we shall show that

f <j>(u)

du

= \og{u a)+$,

and so

( )e>.

(Kzl<o x ) and a 2 (z)

differ

only in an exponential

JACOBI'S ELLIPTIC FUNCTIONS


Hence {u) has a simple zero at each pole of Z{u). Since, however, any pole or zero of Q(u) is a polef of Z(u), it follows that
() is an integral function whose only zeros are simple ones
406

at the poles of Z(u).

Z(u+2K)

Finally, since

= Z(u), we have

(u+2K) = A&{u),
where A is a constant. If we put u = K and remember that
@(u) is an even function, we find, that A = 1; thus {u) is
simply-periodic J with period 2K. This completes the proof of

the fundamental properties of @(u).

14.521.

The Fourier

series for Q(u)

is an integral function of period 2K,


by a Fourier series

Since 0(w)

represented

Q(u)

= |oe'"

ri l

it

can be

z
,

00

which converges uniformly and absolutely in any bounded


closed domain of the ii-plane. We shall deduce the values of
the coefficients an from the important identity

@{u+2iK')
where q

er7,K^

To prove

= g-ie-'^*^),

this identity

we

integrate the equation

= Z(u)7ri/K.
Q(u+2iK') = Ae-iu K
Z(u+2iK')

This gives

where

is

<s>{u),

a constant of integration.

To

find

A, we put

= v %K'\ we then have


-iK = (iK'+v) _ Q(iK'+v)
Aoe
*
(-iK'+v) ~ Q(iK'-v)'
But, when
is smaU, @{iK'+v) = '(iK')v+0(v*),

\v\

and

&(iK') does not vanish, since Q(u) has only a simple zero at
iK'. Hence, when v -> 0, we obtain Aq
1, and so

&(u+2iK')
t Since Z(m)

'()/().
% If it were doubly -periodic, it
the case.

See Ex. 10

= q-H-vlK {u).

on

p. 89.

would be a constant, which

is

certainly not

JACOBI'S ELLIPTIC FUNCTIONS

407

Substituting in this identity the two series

@(u)

= f an en u K

Q(u+2iK')

CO

From

CO

coefficients of e{n

and equating

= 2 an q2ne nviu K

~ 1)7ri utK
,

an

this it follows that

we

'

find that

rft.n-1

(l) na qn\

for all positive

and

negative values of n. Hence

= a 2 (l) nqn

Q(u)

*e

n" iu K
'

CO

The constant a

at our disposal, since, in the definition of

is

Jacobi's Theta function, 0(0) was left undetermined.


fix 0(0) by taking a
1. This gives

We now

CO

CO

= 2 ( i)grV

Q(u)

Mri '*

=1+22 ( W'cosimru/K).
1

CO

This series provides a valuable method of computing 0(tt);


for since 'Rl(K'/K) > 0, we have \q\ < 1, so that the series converges rapidly.

The

14.53.

elliptic integral of

Making the substitution


integral of the third kind,

(1+ks

2 )- 1

2
2 2
1
J {(l-2; )(l-)k 2 )}

If v

= 1

of the
of

v,

Z
'

z =

the third kind


sn(u, k) in Legendre's elliptic

we obtain

_ C
~J

du

l+rsn2

_M _

sn%
U
2
l+vsn

"J
r

'

the evaluation can be completed in terms


of the second kind. For other values
k 2 sn2 a, and we have to evaluate
choose a so that v
or

elliptic integral

we

+&2 sn2 a

C
J

ST1 It

We shall take as the fundamental


kind Jacobi's function
u
,

of the third

elliptic integral

snacnadnasn
lk sn asn u

f &2

,
II(,o)=

du.

k l snto sn%

du,

and

this

can be expressed in terms of Jacobi's Theta function.

JACOBI'S ELLIPTIC FUNCTIONS

408

For, using the addition theorems for

snu and

Z(u),

we have

U(u,a)

k 2 snusna{sn(u-\-a) Jr sn(u a)}du

ij {Z{u-a)-Z{u+a)+2Z(a)} du.
o

Hence

II(,a)

= Jlog^^-J + Z(a).

The expressions for E(u) and Z() in terms of


Weierstrass 's Zeta function

14.6.

The function &n?u

is

an

elliptic

IK and

function of periods

2iK', with one double pole per cell at the point congruent to
iK', the principal part at iK' being

and suppose that o^/^i = i^-'IK, we find that


(K/co^ 2 Aa?(Kzlu)y) is an elliptic function of periods 2co v 2a> 2 with
one double pole ]per cell at the point congruent to co 2 As the
principal part of this function at to 2 is (a co 2 ) -2 it follows, by

= Kz\u>

{uiK')~ % If we substitute

x,

13.62,

where

that

a constant.
we integrate the expressions on each side of this equa-

J. is

When
tion,

K2

^dn = C'(a+,)+4,

we

find thai;

El^)=Uz+w )+Az-r,
a

a,

the constant of integration being determined by the fact that


0. To find the constant A, we put z
a^. This gives
12(0)

EK/wx

t,{ai

1 -\-oi 2 )

J Au)
r
1

and so

We

= Au>
EK
rj

x -ri 2 ri z

= A(a

+7j v

rj.

have thus obtained the formula


.ST

w IKz\

EK

Vl

wl

wl

w l Wl/
which expresses the

elliptic integral

of Weierstrass's Zeta function.

of the second kind in terms

JACOBI'S ELLIPTIC FUNCTIONS

From

this equation,

w1
14.61.

o> x

\oi x }

The connexion between (u) and

a(z)

the connexion between {u) and


last equation of 14.6 in the form

To show

log
dz

and

\cjoJ

j-logc
= dz

a ()

and

integrate between the limits

we now use the result

we

a{z),

write the

z.

^-,
wi

This gives

= 0(O)exp( foxsV^iW 2

(Kz/toj)
If

409

follows at once that

it

of Ex. 15 on p. 378,

)-

we deduce from

this equation that

()

= G ft {l-292m - cos(7m/X)+g4m 1

2
},

= e~vK IK and O

depends only on q. As this formula


is of great importance, we give in the next section a proof of
it which is independent of the theory of Sigma functions and
which gives incidentally a very simple expression for 0.
'

where q

14.62.
If

\q\

The

<

1,

infinite

product formula for @(u)

the infinite product


F(t)

n
1
n
fl{(l-q* -H)(l-q* -H- )}
l

converges uniformly and absolutely in any bounded closed


domain not containing the origin. Hence F(t) is regular in every
\t\
a
annulus
j8 and can be represented there by a uni-

< <

formly and absolutely convergent Laurent series


F(t)

= 2"Jn

F{t), we can show, as in


Moreover, since qtF(q2 t)
n
n
to determine a
remains
It
( \) q \.
that an
Now the function
,
(i-?2 "-1

/(*)

= IT

*)

14.521,

JACOBI'S ELLIPTIC FUNCTIONS


an integral function, expansible as a Taylor series

410
is

f(t)

= fb n

t-,

1
where b
1.
Since F(t)
/(fifty- ), we can express a n in
terms of the coefficients br by multiplying together the series
for/(<) and/( _1 ) and then equating powers of t n this gives
;

an

= K + hK+l + b

2 b n+2+-

But if we substitute the Taylor series in the identity


(1 qt)f(qH) = fty) and equate coefficients, we find that the
coefficients b n are connected by the recurrence formula

k=
from which

it

-&n-i<r-7(i-<72w ),

follows that

*.= (-WA>,
where

cn

Hence

(-l)"ga

and so

(1 q2 )(l

= =
o

4
tf

)...(l q 2n ).

(-1)"

f-

cn.

~{

+ V ^

3
cr c n+r

M,

As o is independent of n, it is necessarily equal to the limit as


n tends to infinity of the expression on the right-hand side of
this equation.

Since

and so

\q\

<

l/c TC is

1, c n tends to a finite non-zero limit as %->


a bounded function of n. Hence we have

oo,

q2(nr+r')

cr cn+r

where

is

<^i<?i

independent of n. Since

2n

2i<?r>

2r
l<zl

is

convergent, this

inequality shows that

Km
n-*

and so

y1

^i

= 0j

Cr C n+r

= lim l/cn =1/11 (lq2n


'

m--oo

We have thus

shown that

fl{{l-q2n )(l g 2 *- 1 *)^


1

).

"- 1 * -1

)}

= 00
2 ( lJV"*".

JACOBI'S ELLIPTIC FUNCTIONS

411

= exp{TriujK), we obtain
2n - co&('7Tu/K)+q in 2n
(*) = f[{{l-q )(l~2q

Putting

2
)},

the required formula for @(u).

14.63. Jacobi's Eta function H{u)


Closely associated with Jacobi's function @(u)

his

Eta

denned by

function,

H()

The

is

relation

-^WexptJwwt/Z) (+')

between (u) and H(it)


0()

is

a reciprocal one; for

iqW&cpQriuWKlu+iK'),

as the reader will easily verify

by using the formula

for

(u+2iK').
H() is an odd integral function which possesses simple zeros
at the points of the set 2mK-\-2niK' The effect of increasing
.

u by 2K or 2iK' is given by the equations


B.(u+2iK') = q~ 1 errt "l K H.{u).
B.(u+2K) = H(),
Thus H()

is

a simply-periodic function of period

Example. Show

that

(i)

H()

= 22 iy*q'*a+1)
n=0

(ii)

H(w)

14.64.

'l

am{(n + i)iru/K},

2q1 i sin(iTrulK) fl {{lq* n )(l-2q* n cos(TTu/K) + q ln )}.


l

The expression

terms

4K

of 0(it)

of Jacobi's elliptic functions in

and H(w)

It follows at once

from the formulae of 14.521, 14.63 that

1 or -fthe function H()/0(w) is reproduced with multipliers


when u is increased by 2K or 2iK' respectively, and so is an

elliptic

function of periods

K and 2iK'.

In any

cell it

has two

simple zeros (the zeros of H()) at points congruent to

and

2K, and two simple poles (the zeros of ()) at points congruent
to iK' and 2K-\-iK'. Hence snuQ(u)j'H.(u) is an elliptic function with no singularities, and so is a constant whose value
0(^)/H() is found by putting u
K. We have thus shown

that

8WU

(K)H.lu)

'.

B.{K)(u)

JACOBI'S ELLIPTIC FUNCTIONS

412

Similarly

it

CILU

can be proved that

H(Z)0

Example. Prove
v&

dnU

'

n(K)/@(K),

dnw

14.7.

(K)@(u)

that

Vfc'

and deduce that


snu = V(l/*)H()/0(tt),

Show

also that

cn

(o)/0(js:),

V(*7*)H(+JC)/0(t),

VA'0(m+U:)/0(m).

*J(2K/tt)

(K).

The numerical evaluation

of elliptic functions

The simplest method of computing the numerical values of


Jacobi's elliptic functions, when q is given, is provided by the
formulae of 14.64, which express them as quotients of
and

H functions.

For, in the formulae


QO

Z-,

Vfcsnw

= -2=2
l

+2

2K

J Woos^
(~

<

we have \q\
1 so that the initial terms of these rapidly
convergent series give very accurate approximations to the
values of snu, cnu, and dnw.
Usually, however, it is the modulus k, and not q, which is
given. When this is the case we have to determine the values
of
and q before we can calculate the values of the elliptic
etc.,

functions.

Now

possible to calculate

it is

K and K'

in terms

of k by using the hypergeometric series

when

\k 2 and |1 k 2 are less than unity, and thence to find


from
the formula q = e-" r' But as k2 and 1 k2 are never
q
small simultaneously, at least one of these series converges far
too slowly to be of much use.
An alternative procedure, due to Weierstrass, gives q immediately in terms of k. We know that
\

V*'

= - 2g+2g-2g+...
M
l+2q+2q*+2q
Q{K)
1

+...

JACOBI'S ELLIPTIC FUNCTIONS


l

andso

we

If

write

methods

2g+2g'+y+...
4
l
+2g16 +...

= jfr

+
+ 2g
2e = (1 VF)/(1 +
1

Vfc'

Vfc')

for the reversal of series,

413

= e+2e

+15e

we

and apply the ordinary


find thatf

-f-150e

13
-f-...,

term omitted being of the order of e17 This series is


convergent when [e| < \, a condition which is certainly satisfied
in the case of greatest practical importance, namely that in
which 1c lies between and 1.
The power of this method for calculating q is seen from the
fact that, when
< k2 < \, e < 1/23. The second term in the
series for q is thus less than 10- 6 and can, in general, be omitted.
J
When q has been found, K is then calculated from the formula
J(2K/ir) = (K) = l + 2q+2qi+2q+...,
the

first

and

finally

K' from the equation

nK'

= Klogq-

1
.

Having found q, K, and K', the computation of Jacobi's elliptic


functions can be readily carried out.
14.8.

The four Theta

functions

In his later work, Jacobi denoted the functions H(w), K(u+K),

&(u+K), and @(u) of the Fundamenta Nova by &x(z,q), &2 (z,q),


where 2Kz = ttu. For our
#3(3,(7), and &4{z,q) respectively,
||

purposes Jacobi's older notation was preferable.


Another method of introducing the elliptic functions of Jacobi
is to determine, from first principles, the properties of the four

Theta functions defined by means of infinite series, and thence


to deduce the whole theory of Jacobi's elliptic functions, defined
as quotients of Theta functions. For an account of this very
t See Weierstrass, Werke, 2 (1895), 276, or Schwarz's account of WeierFormeln und Lehrsatze zum Gebrauche der ettiptischen Funk-

strass's lectures,

tionen (Berlin, 1893), 54.


k2
1, it is probably best to interchange k and k' and determine
J If i
e~"KIK', and thence find q.
For an example of this, see Whittaker's discussion of a problem in rigid
dynamics, soluble in terms of elliptic functions, in his Analytical Dynamics
(Cambridge, 1917), 144-9.
This notation, which is a slight modification of Jacobi's, is the one now
in general use. Jacobi himself wrote &(z, q) for & t (z, q).

<

||

<

JACOBI'S ELLIPTIC FUNCTIONS

414

beautiful theory,

we must, with

regret, refer the reader else-

where.!

REFERENCES
C. G. J. Jacobi,

Fundamenta Nova Theoriae Functionum Ellipticarum

(Konigsberg, 1829); reprinted in his Ges. Math. Werke, 1 (1881),


49-239.
A. Caylby, Elliptic Functions (London, 1895).
J. Tannery and J. Mole, Fonctions elliptiques (Paris, 1893-1902).
E. T. Whittakbb and G. N. Watson, Modern Analysis (Cambridge,
1920), Chaps.

XXI-XXII.

MISCELLANEOUS EXAMPLES
1

Show that sn w, en , and dn u tend to sin , cos u, and

as k ->
2.

0,

and to tanhu, sechw, and

respectively

sechw. respectively as k

->

1.

Prove that
/

&' 2

cn 3 w

sn "3u
M\ 2

cnwdni
elm/

\snMdn

are elliptic functions of periods

K,

'

3
da3u
dn
M

2
sn 3 w\ 2
k2k
k
k' 22BD3u

-4.

enwdnu/

IsnucnM

'

iK'.

z =

< k < 1, the curves in the


sna (w, fc), where
3. Show that, if
z-plane on which the real and imaginary parts of w = u+iv are constant
form two orthogonal families of Cartesian ovals, whose equations are
|z 1| |z|dn(2w,&)
\z-l\cn(2v,k')+\z\dn(2v,k')

=
=

cn(2M,&),
1.

< k < 1, the transformation z = ^Jk saw maps


4. Prove that, if
conformally the interior of the rectangle with vertices w = ^K+^iK'
on the interior of the circle \z\ = 1, provided with cuts along the real
axis from

V&

to

and from

VA; to

1.

Show that the part of the w -plane within the ellipse ui lai +vijbi = 1,
where w = u+iv, is mapped conformally on the interior of the circle
|z| = 1 by the relation
5.

^V^sn^arcsin-^-^],
where Jacobi's parameter q is given by q = (a 6) /(a+6)
2

6.

2
.

(Schwabz.})

Prove that

{l + dn(w+j)}{l + dn(w v)} =


where D = 1 sn wsn .
{l+sn(w+u)}{l+sn(w v)}
{l+cn(w+u)}{l+cn(w v)}
2

(cnv+snudnv) 2/D,
(cnM+cnv) 2/>,
(dnu+dnv) 3 /D,

f See, for example, Whittaker and Watson, Modern Analysis (Cambridge),


Chapter XXI, or Tannery and Molk, Fonctions elliptiques, 2 (Paris, 1896), The
notation in the latter book is slightly different from that given above.
% See Forsyth, Theory of Functions (1918), 614-15.

JACOBI'S ELLIPTIC FUNCTIONS

Show

7.

415

that

sn(u+v)cn( uv)+sn(uv)on(u+v)
L

2snenwdnt;/(l

2
jfc

sn 2 M srA>),

en(u+v)cxi(uv)+sa(u+v)3n(uv)

= (cn t! sn*udn*M)/(I A^sn^wsnfy).


if tanh U = fcsn , tanhF = fcsn y, then
tanh(U F) = fcsn(M+)sn(M v).
2

Show

8.

that,

Deduce that

sn{v+w)m(vw)+sn(w+u)sn(wu)+sn(u+v)sn(uv) +

+k

sn(v+w)sn(tv+u)sn(u+v)sa(vw)sii(wu)sn(u~v)

0.

(Glaisher.)
9.

circle

AB is a fixed diameter of a circle. P, Q are variable points on the


such that Z-PAB = am(u+a), LQAB = am(-a), where a is a

constant. Prove that the chord

PQ

envelopes a

circle.

(jACOBi.f)

10. P, Q are points on the ellipse a;


+2/ /& = 1 of eccentric angle
am(<ia). Show that, if a is constant and u variable, the locus of the
pole of PQ is the ellipse
a

11.

Prove

that

snpC+cnp:=
is

12.

Show

an

elliptic

sum

/a 2

^aa'a+^m^K-a) =

1.

snfX

K = k' sd $K.

cdJX, en

Deduce that

1.

snM+acnM+/?dnw-fy, where a, /S, y are constants,


function of periods 4K, UK' and of order 4 and that the
of the affixes of its zeros in any cell is congruent to (mod4K, UK').

Deduce

that

that, if

m1 +m 2 +m3 +m4

0,

*!

0,

dt

= snur cr = cnur dT = daw,.


Hence or otherwise show that

where sT

c t <22

dr

c 3 rf4

c t d3

(Cayley.)

0.

"*4
13.

Determine the periods and order of the

elliptic

function

snusn(u+K)+cxsn(u+K)+l3sn.u+y,
t This result is of importance in Jaeobi's discussion of Poncelet's poristio
polygons {Journal filr Math. 3 (1828), 376). See also Forsyth, Messenger of
Math. 12 (1882), 100.

c d

416

where

a,

/J,

y are

JACOBI'S ELLIPTIC FUNCTIONS


Hence show that, if m1 +m 2 +ms +m4

constants.

*i^i
2 1*2

14.

Ml

Prove,

2K,

<*i

^a

**2 "'a

$3 C 3

^3^*3

C3

^3

S4 C4

4 rf4

C4

d4

by using

0.

(Oxford, 1923.)

theorem or otherwise, that,

Liouville's

+ Ma + M + M4 = 0,
k' kV^s^s^ Si+h ^ c^c^^d^ d

if

d4

0.

(GUDEBMANN.)
15.

Deduce from Ex. 14

k^s^CgC^

(i)
(ii)

eZ1

(iii)

kc c s s
= c^^d^k^s^s^,
= d d 3 s4 c4
2

d2

ci

d 2 c.,c 4 4-&' 2s 1 a 2
s1 s 2 d3 di

= 0,
d d

m1 +m2 +m 3 +m4

that, if

i,

(H. J.
16.

Show

that, if snu, enu,

and dnM are denoted by

S.

s,

Smith.)

c,

and d

respectively,

_ 1+s n2s+2k a 3 -k2s i \ 2


l+sn3~ l-sll + 2s-2fc2 s3 fcW
l-sn3w

l-cn3w

_ 1-

I k'

'

+ 2A:' c + 2fc c + k c*
2

+k c
l-dn3u _ l-d / /2 + 2A;' 2rf-2d3 -dn 2
~ \ + d\k' -2k'H+2d 3 -d l
l + dn3w
l+cn3w~

l+c\k' -2k' e-2k


2

i)

2
'

fc

'

17. If the nine roots of en 3w


are c x , c 2 ,...,c 9 , prove that

3k* fl

a,

<V+ifc'

18.

T'

By

considering

= 1/(1+t),

(iii)

the
t'

2 <V

four

transformations

= (1+t)/t,

snl&w, tJ

snii&'w,

0.

(ii)

regarded as an equation in en u,

=
I

(iv) t'

&sn(,

(i)

t'

fc),

t&'sc(M,&),

e,miku,-j--\

= ifcsd(M,

snl&'w.-rr;)

t/(1+t),

1+t, prove that

fc),

k'sd(u,k).

Deduce the corresponding transformations of enw and dnw.

JACOBI'S ELLIPTIC FUNCTIONS


417
Prove the results of Ex. 18 by direct transformation
of the integral
formula
mM)
19.

= j

(l_ 22)-l/2( 1 _ &2z2 )-i/2

dz.

Prove Gauss's transformation,

viz. that if A = 2Vfc/(l+fc),


= (l+k)sn(u,k)/{l+ksn*(u,k)},
cn{(l+A) M) A} = cn(u,k)dn(u,k)/{l+kan^u,k)},
dn{(l + k)u,X} = {1 - k sn (u, k)}/{l + k sn M
k)}.

20.

sn{(l+&) M ,A}

Show also

that, if 4A, 2iA' are the periods of sn(w,A),

A=

A' = i(l+k)K'.
o and 6 are two real positive numbers. Two
sequences of
numbers, a 1( a 2 a a ,... and b lt b b
are defined by the equations
2
3

(l+k)K,

21.

= i(<*+t>),
= i(a +b

ar+i

Show
say.
22.

that, as

real

n ->

(This limit

bt
6r+1

r ),

= J(ab),
= J(a b
r

r ).

o and bn tend to the same finite limit,


M(a,b)
called the arithmetico -geometric mean of a
and b.)

is

00,

Prove that, in the notation of Ex. 21,


de
f
J (a*cosW+b*sw.'0)W

Deduce

that, if

<

<

f
a
J (a?cos

df

+ 6fsinV)

1 / 2-

1,

K = frr/M(l,k'),

K' = iTrJM{l,k).
Show that the arc s of the lemniscate r 2 = a2 cos 20, measured
from
the point of polar coordinates (a, 0), is given by r =
a cn(W2/a, 1/V2).
23.

Prove that, for the

elliptic functions of modulus 1/V2,


0-04321
q
1-854 > correct to four significant figures.
Deduce that the
flength of a complete loop of the lemniscate is 2-62a,
approximately.

=K =
'

A rigid

body is performing finite oscillations about a fixed horizontal axis, under the action of gravity. Show that, if
its angle of swing
is 2a, its period of oscillation is 4Kj(l/g),
where k
sin<x and I is the
length of the equivalent simple pendulum.
If gravity is suddenly reversed when the pendulum is
at the end of
a swing, show that the new period is 4JC'*](ljg).
24.

25.

Show

that

2k

^dau+kcnu'

cnudu

arctan(&sdw),

dn u du

= am w.
Ee

JACOBI'S ELLIPTIC FUNCTIONS


Deduce the integrals of the other elliptic functions of Jacobi by
creasing u in turn by K, iK', K-\-iK'.
418

26.

cd

w,

in-

2
2
2
2
2
Prove that the integrals of ns 2 w, nc 2 w, nd w, sc w, cs w, sd w, ds w,
func2
elliptic
Jacobi's
E{u),
terms
of
expressed
in
dc M can all be

and

tions,

u.

Show

27.

-r- (A;

du

that

sn m en u dn"-1 ^)

(n+ l)dn+2M-(2n- l)(l + fc' 2 )dn2tt+(n- l)dn- 2M.

Hence obtain reduction formulae


j

dnmit

for

dnmM du.

du,
J

Find

n
also similar formulae for sn u,

cnnw.

Prove that

28.

2X

2K

und?u du

= ?^*L,

| und'u du

J*

= ^p(l +fc'

)X,

2K

j undhidu

|^{2(l + fc' 2 }.E+fc'a.K}.

that any quartic or cubic polynomial in x without repeated


factor can be expressed in the form
29.

Show

w =
2

{A(x-<xy + B(x-pY}{C(x-af+D(x-fiY},

where A, B, C, D,

a,

and

are real

/2

if

the coefficients of the quartic

are real.

Hence prove, by using the nomographic transformation


t

(as-a)/(-j8),

evaluated in terms of
J B{x, w) dx can be
Legendre's three kinds of elliptic integral, together with the elementary
functions of analysis.^

that an

30.

elliptic integral

Express x in terms of u when


X

2 -2*)- 1/ 2 dt.
[ (1-f t

(Math. Trip., 1914.)

31.

Prove that
U(u,v) THv,u)

32.

Show

,_,.

H(u,a) + TI(v,a) U.lu+v,a)


f

uZ(v)vZ(u).

that

The advantage

modulus when the

of this

method

coefficients in

w1

k 2 snasnusnvsii(u-\-v a)

ilog

,_

r.

of reduction is that it always gives a real


are real, unlike the reduction of 14.5.

JACOBI'S ELLIPTIC FUNCTIONS


33.

Prove that

34.

Show

II(w,ia+K,k)

419

Xl(u,a+K',k').

that

Il(u,a)+Il(u,b)~U(u,a+b)

_
~
35.

,.

* lo S

By

k saasa.bsnu8n{a+b u)
r-JT,r+t
M _, _
i+rsnasnosnMsn(a-f-A-fM)
2

1
i

snasn6sn(a+6).
l

'

'

integrating round a cell of snu, prove that

[ snu eiqp(imriu/K) du = JT^* {1 - ( - 1 )"}.


l
* '
-2K
Deduce that snu is expansible as a Fourier series

2n

g"+i/2

\-
o

<

valid in the strip \Ixn(u/K)\


36.

TZl(K'/K).

Prove that the expansions

cnu

2n

a* 1 / 2

"ST*

Kk 2, T+^+i cos M +*W^,


<

hold in the strip

|Im(/iir)|

37.

Deduce from Exx.

38.

Show

<

m(K'/K).

36 the Fourier series for cd u, sd u, and nd w.

35,

that

ns M

cosec

+ - ^j^^^n+^nu/K,
o

the Fourier series on the right-hand side being convergent in the strip
\Ixa(u/K)\
39.

< 2m{K'/K).

Prove that the expansion


,

611

= E + 2tt
K K?

holds in the strip \Im(u/K)\


Deduce that the formula

<

wo"

"V
r^iS
Z

C08W /#

B,1(K'/K).

^
00

Z(u)

2t7
=
j>
l

is

valid in the

same

strip.

n
-

sin totw/JC

JACOBI'S ELLIPTIC FUNCTIONS

420

= 0,
"O c. d
, 8, So s,
zk)+zk)+zk)+zk) = - 1+
Z-7T
d d d s d 4 V*r c
E^CgC,^ ^d s ^
~ Wc^CzCi k' 2 o
V +d d dl di jL, d
Prove that,

40.

if

t^+Mj+Ma+M*

fc

^Wi
fc

r
'

II H(w
K) + r=l
r=l
transformation 2u' = u^+u^u^u^ 2u

II

invariant under the

is

Prove that

41.

r)

r.

Deduce that

K)(m 2 )H(m 3 )H(m4 )+H(m1 )H(m 2 )0(m3)0K)


and
(u 1 +K)(u i +K)(u a )(u i )-TI(u 1 +K)TI(u 2 +K)H-(u 3 )H.(u i )
are also invariant.

Show that, in the notation of Ex. 41,


(u1 +K)(u 2 +K)(u3 )(ui )+TI{u 1 +K)H.(u 2 +K)Tl{ua )E.(ui )
42.

= 0K+jr)0K+x)0W)0W)+HW+x)HW+x)H(J)H(i).
(1 fcsnM snM )(l fcsnM snM

43.
is

ksn(t x)en(y-z)
+ ksn(t-x)sn{y-z)

equal to unity.

4)

ksnUiSnU2){ i +ksnU3SIlUi )

invariant under the transformation of Ex. 41.


1

is

Prove that

Deduce that

l-fcsn(f 2/)sn(z x) lksn(tz)sn(xy)


l+&sn(*-2/)sn(z-a>) l+ksn(tz)sn(xy)

CHAPTER XV

THE ELLIPTIC MODULAR FUNCTIONS AND


PICARD'S THEOREM
The construction
modulus

15.1.

of elliptic functions

given

of

It was seen in the previous chapter that the problem of evaluating an elliptic integral is solved if we can construct Jacobian
elliptic functions when c, the square of the modulus, is given.
By the first formula of Ex. 4 on p. 385, we have to show
that, when c has any given value different from
or 1, we can
find a number t whose imaginary part is positive, which satisfies
<
the equation
/ ,
, x R
,

where q

e, and, moreover, that each solution of this equa-

same set of Jacobian elliptic functions.


equivalent form of this problem is the determination of
two parameters <o t and u> 2 , such that Im{a> 2/cD 1 ) is positive,
which satisfy the equation
tion leads to the

An

p(w t + 0)2^,0)^ p(u) 2 \a) v W 2 )


P(">iK,o> 2 ) p(a> 2 |o. 1 ,w a )
Actually a precise value for

wx

^
~

cannot be determined from this

equation, since the expression on the left-hand side is a function


of t
co 2 ju> x alone; but this is immaterial, since a) disappears

when we

construct the Jacobian elliptic functions


Weierstrass's Sigma functions.

by means of

however, we reduce the elliptic integral to Weierstrass's


13.7, we are given three constants e 1; e 2 and e3
whose sum is zero, and have to determine Wj and w 2 such that
If,

form, as in

= P(i),
and Im(a)Ju} > 0.
ei

1)

e 2 = p(w 2 ),
But when we take

e3

geneity of p(z), we find that this problem


as the one we have just stated.
15.2.

The primitive periods

of

an

= pi^+ws)

into account the


is

homosame

precisely the

elliptic

function

the periods 2^ and 2w 2 of an elliptic function were


said to be a pair of primitive periods if Im(co /cu ) is positive and
2
1

In

13.2,

THEOREM
if every other period of the function is of the form 2mw 1 +2ww 2
where m and n are positive or negative integers or zero. When
MODULAR FUNCTIONS AND

ELLIPTIC

422

PICARD'S

this is the case, the parallelogram with vertices 0,

2w 1; 2w 1 +2w 2

2w 2 is a primitive period-parallelogram; if the affix of a point


within or on this parallelogram is a period, the point is necesone of the vertices.
There are obviously an infinite number of pairs of primitive
periods. If 2u> x and 2o> 2 form one pair, the numbers
sarily

2o>i

form another

and only

if
<a 1

if

= 2bw +2aw
> and
= rco^+sui^,

2w 2

2doj 1 +2cco 2 ,

Im(w 2 /oui)

= pco[+qco

a> 2

2,

where p, q, r, and s are integers. It is easily seen that Im(w 2 /wi)


has the same sign as ad be, and so ad be > 0. If we solve
the first set of equations for co 1 and w 2 we obtain
,

(ad bc)a) 1

and

= aaj{cco'
a
-

so

(adbc)o) 2

2,

=
=
q
r
b

d
-

ba>{+dw 2

= (ad be),
.

(ad bc)(ps qr) = 1 and hence that


ad be = 1. We have thus shown that if 2w x and 2w 2 form a
pair of primitive periods of an elliptic function, so also do

From

this it follows that

2coi

if

2dco 1 +2cco 2 ,

2a> 2

26a> 1 +2ao) 2

and only if a, b, c, and d are integers such that ad be 1.


Out of the infinite number of pairs of primitive periods of the

given elliptic function, we now choose a particular pair of special


importance, in the following manner.
In the first place, there exists a unique period, 2w say, which
is such that the inequalities
|2w|

<

hold whenever 2Q

|2Q|,

< arg 2co < arg 2Q

any other period of the function. Further,


number of periods 2io' such that 2w and
2a>' form a pair of primitive periods, and these all lie on a
straight line parallel to the line through the points cu and -co.
Of these, at most two are nearer to the origin than the rest. If
there is only one, call it 2cot; if two, let 2cot be the one which
is

there are an infinite

MODULAR FUNCTIONS AND

ELLIPTIC

THEOREM

423

The unique

pair

PICARD'S

gives the smaller (positive) value to arg(a//w).

of primitive periods chosen in this

way is

called the fundamental

pair of primitive periods of the given elliptic function, and the


corresponding period-parallelogram the fundamental primitive
period-parallelogram.

>

<

<

Obviously |t|
argr
1, and, if \r\
1, then
%tt.
Moreover, as 2a> and 2co+2cot also form a pair of primitive
periods of the function, we have jl+r|
|r|;
but since
arg(l+r)
argT when Intr
0, the sign of equality is not

<

|t 1|

|t|,

>

we can show

that

where, however, the sign of equality can

now

and so |1+t|

permissible,

>

>

<

Similarly

|t|.

occur.

Therefore if 2co and 2cot form the fundamental pair of primitive periods of an elliptic function, t lies in the region defined

by the

inequalities

|t|

>

1,

>

|t+1|

|t|,

|t 1|

>

|t|.

This

bounded by an arc of the circle T = 1 and the


straight lines RIt = i; the only boundary points which
belong to the region are those for which the real part of t
region

is

is

not negative.

15.21.

The modular group

of transformations

If 2oo, 2ojt and 2a/, 2a>V are two pairs of primitive periods
of an elliptic function, there exist, as we have just seen, integers
a, b, c,

and

d,

connected by the relation


co't'

When,

therefore,

= aoyr-\-bw,

cm'

ad be

1,

such that

= Carr-\-dco.

we change from one

pair of primitive periods

to another, the quotient r undergoes the homographic trans-

formation

ar+b
CT+d
,

A transformation of this type,


connected by the relation

where

ad be

a, b, c,

1, is

and d

are integers

called a modular trans-

formation.

The
it

(i)

modular transformations forms a group, since


two characteristic properties of a group, namely:

set of all

possesses the
If t'
is

= (a,T-\-b)l(cT-\-d)

is

a modular transformation, so also

the inverse transformation t

= (dr' b)/(cr' a).

ELLIPTIC

424

(ii)

If t x

MODULAR FUNCTIONS AND PICARD'S THEOREM


(a 1 T+b 1 )/(c 1 r+d 1 t 2 = (a T +6 )/(c r +rf
are
2 1
2
2 1
2
),

two modular transformations, so also is the transforma2

(^iC 2 +c 1 d2 )T+(b 1 c 2

+d1 d2 )'

which

is obtained by the successive application of the


given transformations.
The group of modular transformations is called the modular

group.

Two points in the T-plane are said to be congruent with


respect to the modular group whenever their affixes are connected by a modular transformation. If Imr
0, there is
evidently no point congruent to t in the lower half-plane;

>

moreover,

if

a point

is

congruent to a point on the real axis,

We

either the point at infinity or else lies

on the real axis.


shall only consider the effect of modular transformations on sets
it is

of points in the upper half-plane, as the corresponding results


for the lower half-plane can be obtained by a reflection in the
real axis.
It follows immediately

from the work of the preceding section


which is congruent
to t' with respect to the modular group and which lies in the
region defined by (i) |t| > I and | < Rlr <
or (ii) T > 1
and < R1t < |. This region is called the fundamental region
of the modular group.
We have not, however, shown that r is the only point of the
fundamental region congruent to t'. Now if tx were another
point of the fundamental region congruent to r', r would be
x
that

if

Imr'

is

positive, there exists a point t

congruent to t; this is impossible, as we shall now show.


Let us suppose, then, that tx and t are connected by a modular
transformation rx
(aT +b)/(cT+d) and that r lies in the funda2
mental region; we consider separately the cases c
0, c
1,

>
If c = 0,

and

c2

1.

the transformation becomes r t = t+w, where n is


an integer, and so tx does not lie in the fundamental region.f
The point at infinity is invariant under this transformation.
Ifc2 = ^thetransformationisoftheformTi
l/(rn),

m=

t It must be remembered here that boundary points on the


do not belong to the fundamental region.

line

Rl

= J

ELLIPTIC

MODULAR FUNCTIONS AND

PICARD'S

THEOREM

425

m and n

where

are positive or negative integers or zero. Now


the inequality \rn\
1 holds for all points of the fundamental
region if n
or 1, and for all save certain boundary points if

>

<

In these cases |tj m|


1, so that t is certainly
outside the fundamental region. When n
0, we have to consider the boundary points t
e $i where %rr
\n, and so
Tx
d"-^. It follows that r belongs to the fundamental
or

1.

< <

m =

when ra =
and 6 = \n, or when m = 1 and
= i^; in both cases, t is identical with t. Again, when =
k > 1 save when t = e ' this gives r m = e ^
so that t lies in the fundamental region only when m = 0, in
region only

1,

1|

77

which case tx and t are


Finally,

when

c2

>

identical.

1,

the form

the transformation can be written in

But

so that

hence tx

We

Ti

is

d
+ >

V3c2

<*W

V3

2V3

'

certainly not in the fundamental region.


that, if Imr' is positive, there exists a

have thus shown

unique point t which lies in the fundamental region of the modular


group and is congruent to r' with respect to that group.

modular transformation

t'
Z7(t), being homographic,
region conformally on a region of the
upper half-plane bounded by three circular arcs (or straight
lines) intersecting at angles \tt, Jtt, 0. It is convenient to denote
this region by U, which is also the symbol of the transformation

maps the fundamental

producing it; we denote the fundamental region by /.


Every point of the upper half-plane lies in at least one of the
regions congruent to /, so that they completely fill the upper half-

we shall show that they do so without any overlapping.


For suppose that r'x is a point common to the regions U and
V, produced from / by the modular transformations t' = U(r),
t' = V(r); let t x be the point of / congruent to t^ Then, if
t = ?7 _1 (t') is the transformation inverse to r' = U(r), we have
plane;

Tl

U-\r'x )

U-WirJ,

426

ELLIPTIC

so that t x

is

MODULAR FUNCTIONS AND

a point of I which

transformation

t'

?7

_1

is

PICARD'S

THEOREM

invariant under the modular

F(t).

We

have, however, just seen that at most one point of I is


invariant under a modular transformation and that such a point

has

Hence the regions U and V can have


common, and so do not overlap.

affix e 7' 3 , e wi/2 , or co.

at most one point in

/+

\+
\
~+~-\\

/+

I//''""

In the above figure are shown the fundamental region and a


of the regions congruent to it. In each case the dotted
curve is congruent to the imaginary axis; it divides each region
into two parts and corresponding parts are marked with crosses.
At first sight, it appears that two adjacent regions would have
a whole side in common, contrary to what we have just proved.
This is not the case, since part of the boundary of each region
does not belong to it.

number

15.22.

The

elliptic

modular functions

An analytic function /(t), regular save for poles when


Im t > 0, is called an elliptic modular function^ if there exists
an algebraic relation between the functions /(r) and/(r') whenever t and t' are connected by a transformation of the modular
f The standard work on elliptic modular functions
Theorie der ettiptischen Modulfunktionen (Leipzig, 1890).

is

Klein and Fricke,

ELLIPTIC

MODULAR FUNCTIONS AND

= /(t')

PICARD'S

THEOREM

427

modular transformations, /(t) is


said to be an automorphic function^ of the modular group.
We can easily construct an elliptic modular function out of
the invariants g s g3 of Weierstrass's function p(z\co, cor), which
are unaltered when we replace 2co and 2o>t by any other pair
of primitive periods 2a/ and 2o>V. But since

group.

If /(t)

for all

g z (w,

cor)

w- 4gr 2 (l, t),

g3 (co, cor)

= co~ g
6

3 (l, r),

follows gl/gl is a function of t which is invariant when r undergoes a modular transformation. As g\]g\ can be shown to be
regular in the fundamental region, save for one simple pole, it
it

is

an automorphic function of the modular group.

On

the other hand, the function

^=^t\m^
=

where q
eniT is an elliptic modular function, but not an automorphic function of the modular group. To prove this, we
,

observe that, as

we saw

in

15.1,

A(t)

(e 3

2 )/(e 1

2 ),

and e 3 are the values taken by @{z\to, cot) when


and <o+o>t. Now when we change to a new pair of
primitive periods, 2a>' and 2a>V say, the numbers e{, e 2 e'z are
where

e v e2 ,

co, tor,

a permutation of e 1; e 2 e3 Hence, when r is connected with


t by a modular transformation, A(t') is equal to one of the six
.

functions

Mr),

l-A(r),

W)

^-^ w^-

i>

l~

wy

obtained by permuting e v e 2 and e3 Since A(t) is regular when


1, that is, when Imr
0, we have thus shown that A(t)
\q\
,

<

is

an

>

elliptic

15.23.

modular function.

TheA-group

group of transformations A is said to be a sub-group of


another group B, if every transformation of A belongs to B
although there are transformations of B which do not belong
f For the theory of automorphic functions in general, see, for example,
L. R. Ford, Automorphic Functions (New York, 1929), or Forsyth, Theory of
Functions (Cambridge, 1918).

428

ELLIPTIC

MODULAR FUNCTIONS AND

We now

PICARD'S

THEOREM

modular function A(t) is


an automorphic function of a certain sub-group of the modular
to A.

show that the

elliptic

group.

We have seen that A(co 2 /to 1 ) A(a> 2 /a>i) if 2co[, 2a> 2 form a pair
of primitive periods of p(z \a) v a) 2 ), such that p{<*>[)
e lt
This
case
e
and
p(a>'
e
is
the
if
P(oj 2 )
2
s)
3

co[

= dco

1 -\-2cu>2,

u>2

26co 1 -f-aa> 2 >

where a, b, c, and d are integers such that adibc = 1, a and


d being odd.f Writing a> 2
u> x t, eo 2 = oj[t'
we find that
A(t')
A(t) when t'
(ar+2&)/(2cT-f-rf), where a, b, c, and d are

integers such that

We

',

shall call a

ad4bc

1.

modular transformation of

this special type


a A-transformation. The set of all A-transformations evidently
constitutes a group, which is a sub-group of the modular group;
we call it the A-group. The elliptic modular function A(t) is,
therefore, an automorphic function of the A-group.

15.24.

Two

The fundamental region

of the A-group

points in the r-plane are said to be congruent with

respect to the A-group if their affixes are connected

transformation.

on the

As two congruent

real axis or

by a

A-

points are necessarily both

both on the same side of that

axis, it suffices

to consider only the effect of A-transformations on sets of points


in the region Im t
0.

We show first of all that the A-group possesses a fundamental


region

by

2Q 2 and 2Q2 of

considering the periods

such that ^(Qy)

=e

p(z\oj x ,

cu 2 )

the ratio co^co^ being an


2
assigned number t' whose imaginary part is positive.
The fundamental primitive period-parallelogram of p(z) contains either one or

we

call it

oj.

ev

>(:Q 2 )

two points of the

set Cl v If there is only one,

If there are two, they are the midpoints of opposite

we take as <o the one nearer the


uniquely defined and satisfies the con-

sides of the parallelogram;


origin.

dition

We

In
|o>|

way, co
|w+2mre

this

is
|,

where

0.

mn

is

any period of

p{z).

next consider the points co' of the set Q 2 such that 2co
and 2a/ form a pair of primitive periods of p(z). These points
u>' he on a straight line parallel to the line through a>, and at
f It will be seen that the condition
are odd.

ad 46c =

1 itself implies

that o and d

ELLIPTIC MODULAR FUNCTIONS AND PICARD'S THEOREM 429


most two of them are nearer to the origin than the rest. If there
is

only one,

we

call it cot.

If there are two,

we take

as cot the

one which gives the smaller (positive) value of arg(a//w). Thus


cot is uniquely denned and satisfies the condition
|wt|

<

|WT+Qm>n

|,

where Q. m>n is any period of p(z).


In this way we have found a number t which is congruent
to the given number t with respect to the A-group, and which
satisfies

the inequalities
1

<

|2m+l+2*r|,

|r|

<

|2p+(2g+l)r|,

for all positive or negative integer or zero values of m, n, p,

Each of these

q.

and

inequalities is satisfied if
1

<

|2t+1|,

1<|2t-11,

|t|

<

|t+2|,

|t|<|t-2|.

The point t hes, therefore, in a region bounded by the straight


RIt = 1 and the semicircles |2t1| = 1; moreover, it
is easily verifiedf that the only boundary points which belong
to the region are those for which RIt > 0. We call the region
lines

denned in this way the fundamental region of the A-group.


We have thus shown that, when Inn-' > 0, there exists a
point in this fundamental region which is congruent to t with
is not, however, obvious that t is the
only point of the fundamental region congruent to t'. If there
were another such point t1; t and tx would be two points in the
fundamental region congruent with respect to the A-group, and
0, there
this can be shown to be impossible. J Hence, iflmr'
exists a unique point t in the fundamental region of the X-group

respect to the A-group. It

>

which

is

congruent to t with respect to that group.

Each A-transformation maps the fundamental region conformally on a region in the upper half-plane, bounded by four
semicircles (or straight lines) all orthogonal to the real axis. It

can be shown, just as in the case of the modular group, that


these 'quadrilaterals' congruent to the fundamental region fill
the upper half-plane completely without any overlapping. Two
t

By

finding t

when

t'

= l+iy and when

2t'

= l+eM

omitted, since it is a simple modification of that given in


15.21 in connexion with the modular group.
J

The proof

is

+
l'+.

-t

to

!+s

-<

--<--

ELLIPTIC

MODULAR FUNCTIONS AND

PICARD'S

THEOREM

431

quadrilaterals can

have at most one point in common, and such


a common point is congruent to one of the points 0, 1, or oo of
the fundamental region.f
The figure given opposite shows the fundamental region and
a number of the regions congruent to

it

with respect to the

A-group.

The behaviour of A(t) in the fundamental region


The function A(t) is regular in any bounded closed domain in
the upper half of the r-plane and never takes there the values
and 1. We now show that these exceptional values 0, 1, and oo
are the limits to which A(t) tends as r moves in the fundamental
region of the A-group up to one of the corners 0, 1, or oo.
15.3.

Now

X(t)

where q

=&

rtT

Hence,

16q

if

= u-\-iv,

t
co

where

|A(r)
|

is

where v

Now as
-*-

independent of u and

we have

Therefore \{u-\-iv)

v.

In particular,

->0as

A(t) ->

t moves to the origin in the fundamental region, the

1/t tends to infinity in the

Jacobi's imaginary transformation,


A(t)

and so

1,

-\lp-2nir \8

v -> oo, uniformly with respect to u.


as t -> oo in the fundamental region.

point

< Mr- fj (^JW) < Ar ">


,

so

'

= e- < e-w

\q\

and

[l+q^-ij

A(t) ->'l as t ->

same

region.

But, by

we have

l-A(-l/r),

(i)

in the fundamental region.

Again, sincef
A(T)

A(r+l)-l

= A(r-l)-l

the function A(t) tends to infinity as t ->

'

in the

(U)

fundamental

region.

When t lies on the imaginary axis,


also real.

But X(iv)

t The proofs
by the reader.
{ See

is

q is real and hence A(t) is


a continuous function of the real variable

of these statements are omitted, as they can be easily supplied

Ex. 18 on p. 416.

ELLIPTIC

432

MODULAR FUNCTIONS AND

THEOREM

PICARD'S

and 1 on the
values between
v; therefore A(t) assumes
that
deduce
A(t) is also
imaginary axis. Using formulae (ii), we
on each of the lines
real and takes all values between
oo and
all real

RIt

by formula (i), we see that A(t)


values between 1 and + on each of the

1.

takes all
|2rl|

Finally,

real

and

l.

The zerosf

15.31.

is

semicircles

of A(t) c

a closed contour in the upper half of the T-plane on


which A(t) never takes an assigned value c, different from or 1,
If

is

gVf'

the function A(t)- -c has

N zeros within T, where


J A(t)-c

We suppose, in the first place, that c is not real, so that A(t) c


has no zeros on the boundary of the fundamental region, and
we take V to be the boundary of that part of the fundamental
region for which

Imr
where
")

The

<

> 0.

l/2e,

|t ie\

This contour

analysis of this section

is

>

\r\ ie\

e,

shown

>

e,

in the figure above. Since

is essentially

the same as that given

Whittaker and Watson, Modern Analysis (Cambridge, 1920), 481-3.

by

ELLIPTIC

MODULAR FUNCTIONS AND

A(t) -> 0, 1, oo as t -> oo, 0,

PICARD'S

THEOREM

433

respectively in the fundamental

we can make e so small that no zero of the function


occurs in the part of the fundamental region outside T. The
value of the contour integral then remains constant as e deregion,

creases;

we

find its value

by

considering the limit as

->

0.

Now when t moves along the arc F'E', its affix is \-\-\e n -a)i
(~'

The congruent point of the fundamental


region has affix l+i&ei and so describes the arc FE. It follows
that the integrals along F'E' and EF cancel. Similarly the sum
of the integrals along OH and H'G' is zero.
The value of the contour integral is thus equal to the sum of
the integrals along HH', G'F', E'E, and FG. We consider these
where 6

increases.

parts separately.

Now when Imr


therefore express

it

as a

A(t)

regular since

1, A(t) is

power

\q\

^e

_,r
.

can

16q(l+a1 q+a a ^+...)

which converges uniformly with respect to t when


Hence we have
A'(r)

We

series

Imr

>

1.

167rig(l+2o 1 gr+3o 1 gi+...),

and so the function


A'(t)

167Mg(l+2ffl 1 g+3a 2 g 2 +...)

__

A(t)-c ~

-c+16q(l+aiq +a 2 q2+...)

tends to zero as Idit-> oo, uniformly with respect to Rlr, provided that c is not zero. From this it follows that the integral
along HH' tends to zero with e, since Imr
l/(2e) on HH'.

When

t moves along the arc E'E, the point Tl


scribes the straight line HH', and hence
f
J

^W ar
J -

\{T)-c~a^

E'E

<SA(t)

aw^^T
BB'

so

J
E'E

(t)

AW=-C

= 1 A(t

1 ),

^J

A(Tl)

_ (1 _ c) *-i-

BB'

By the same argument as before,


1 c is, by hypothesis, not zero.
41U

lm

But, by Jacobi's imaginary transformation, A(t)

and

= 1/t de-

yf

this tends to zero with

e,

since

434

ELLIPTIC

Again,

MODULAR FUNCTIONS AND

when t describes the arc FG,

moves along the right-hand half of the


f

dX{r)

THEOREM

PICABD'S

= 1/(t 1)

the point r x
line

HH

c*A(t)

so that

',

But, by Ex. 18 of p. 416,

A(t)

AK)'

which gives
A'(t)

f
J

1
A'fa)
f
)-l
(l-c)A(r
A( Tl )
J
1

A(t)-c

by

HK

FG
Similarly,

using the transformation r x

= 1/(t+1),

we

can show that


A'(r)

AW=c

Combining these
that
9^-iAT
27ri^

dT

results,

lim
= lim

__J__A'(t

__

{l-c)\{rx )-l

we

find that, so far,

1)

A( Tl )

^
,

we have shown

jL^-^^th

(l-c)A(r)-l

A(t)

HH'

The integrand

can, however, be written in the


77

and

form

(l+2a lg +3a 2 ga +...)


(l+a iq +a^+...){l~(l-c)X(T)y

this tends to

iri

as

Imr-^ +oo,

uniformly with respect

1.
2m, and so
have thus proved that, when c is not real, the function
A(t) c has one zero in the fundamental region of the A-group.
It remains to consider the case when c is a real number not
equal to or 1. In this case, A(t) c will have a finitef number
of zeros on the sides EF, E'F', GH, and G'H' of Y, and none
in the part of the fundamental region outside T, provided that
e be sufficiently small. Now F'E' does not belong to the fundamental region, but is congruent to FE with respect to the
A-group; a similar remark applies to the sides G'H' and GH.
to R1t.

From this

it

follows that 2-niN

We

f It cannot have an infinite number, since A(t) c

onI\

is

regular within

and

ELLIPTIC MODULAR FUNCTIONS AND PICARD'S THEOREM 435


Accordingly these zeros occur in pairs which are symmetrically
placed with respect to the imaginary axis.

We now

cut out each zero on E'F' and O'H' by an indentawhose radius is so small that it cuts out no other zerof of
X(t)c. The zeros on EF and OH are then brought within V
tion

by

circular arcs congruent to these indentations, as

the figure.

We

region, this

means that the function has

shown in

then apply the same argument to the modified


contour, observing that the integrals along the indentations and
the congruent circular arcs cancel, and conclude that A(t) c
has one zero within the modified contour. Since, however,
points on E'F' or O'H' do not belong to the fundamental
precisely one zero in

that region.

Hence

it

follows that, if c

is

any number not equal

or

to

1,

modular function A(t) takes the value c once in the


fundamental region of the X-group.
In particular, we see that A(t) increases steadily from
oo to
as t moves along the straight line from 1 to 1+ooi, then
from to 1 as t describes the imaginary axis from cci to 0, and
finally from 1 to +oo as t moves along the semicircle joining
the elliptic

the points of affix

and

1.

The solution of the problem of 15.1


We now show that the Jacobian elliptic functions are uniquely

15.32.

determined when the square of the modulus is assigned.


For we know that the equation A(t)
k 2 has an infinite number of roots; each root t is connected with the root t in the
fundamental region of the A-group by an equation

where

(ar+2b)/(2cr+d),

and d are integers such that adibc


1.
construct two sets of Jacobian elliptic functions from
the Weierstrassian functions p(z\2cT+d,ar+2b) and p(z\l,r),
the square of the modulus k has the same value for each set.
If

But
two

a, b, c,

we

since the
sets of

when k2

two Weierstrassian functions are

identical, the
functions are also identical. Thus,
given, sn(u,k), cn(u,k), and dn(w,&) are uniquely

Jacobian

is

elliptic

determined.
t This

is possible,

since the zeros of

a regular analytic function

are isolated.

MODULAR FUNCTIONS AND


Conformal mapping by A(r)

ELLIPTIC

436

15.4.

PICARD'S

THEOREM

In the fundamental region of the A-group, the function A(t)


OA C. Hence
is real only on the sides of the curvilinear trianglef
the imaginary part of A(t) is always of the same sign within this
'triangle'. To determine this sign, we observe that

A(r)

made

f(q) can be
sufficiently large, and so

where
if t

lftrfl +/(?)},

as small as we please by taking Imr


Im A(t) has the same sign as Im q. But

= u-\-iv,
Imq

= Im(e

7riu -'"')

OA C.

and this is positive within

sin ttu e~

Thus

the

m
,

imaginary part of A(t)

vanishes on
is positive within the curvilinear 'triangle' OAG, and
of A(t)
part
imaginary
that
the
we
see
its boundary. Similarly
vanishes
and
OA'G
'triangle'
curvilinear
is negative within the

on

its

boundary; in

fact,

by Schwarz's

principle of

symmetry, J

A(t) takes conjugate complex values at points which are symmetrical with respect to the imaginary axis.
Since A(t) is regular and simple within the fundamental region

and takes there every value save real values between oo and
and between 1 and -foo, the transformation z = A(t) maps
the interior of this region conformally on the whole z-plane,
and from 1 to
supposed cut along the real axis from -co to
+oo. The upper sides of the two cuts correspond to A C and OA
respectively, the lower sides to A'G and OA'. There exists,
therefore, a unique inverse function t

the cut z-plane, and maps


region of the A-group.
15.41.

The

it

= v(z) which is regular in

conformally on the fundamental

inverse function

v{z)

A(t) defines t as a many- valued function of


z
the function v(z), which we have just defined, the
fundamental branch of t; any branch of t is connected with
v{z) by a A-transformation. We now consider how these other
branches of r can be obtained by continuing analytically the

The equation

z.

We

call

function

v{z)

across the cuts in the z-plane.

certain results in

Chapter

t See the figure of

We do this by using

XIV which depended on showing that

15.24.

% See 8.4.

ELLIPTIC

MODULAR FUNCTIONS AND

PICARD'S

THEOREM

437

the Jacobian elliptic functions are uniquely determined when


the square of the modulus is given; the work of the present

chapter has completely justified this assumption.


Now v(z) iK'/K, where and iK' are the quarter-periods
of the Jacobian elliptic functions of modulus z1 2 But
has

'

branch-points at z

and

= oo, and K' at z =

and z

K
= oo;

from their expressions as hypergeometric functions.


has branch-points at 0, 1, oo.
To determine the nature of these branch-points, we use the
equation
,^
this follows

Hence

v(z)

V{Z)

dz\K]

4iK*z(l-z)'

which follows immediately from the result of 14.511, Ex.


|z| < 1, this equation can be written in the formf

2.

When

v'(z)

-j-a

TtVZ

the infinite series having


v{z)

\z\

+a 1 z+a 2 z
1

+...,

as circle of convergence.

Hence

= log z+$(z),
TTl

where

<j>(z)

is

regular

The function

when

\z\

<

1.

a logarithmic branch-point
once round
z =
in the positive sense, we obtain a new branch t = v(z)+2
of the inverse function J; this branch maps the cut z-plane conformally on the interior of the 'quadrilateral' CA0 1 A x of the
at the origin.

v(z) has, therefore,

When we

continue

v(z) analytically

figure of 15.24.

On

the other hand,

d(

we have
v'{z)

deduce from

this, in

_L
V(Z)

where

i/i(z)

is

regular

t Since 1/{(1 z)K'}

is

AiK'H{\-z)

v (z)

dz\v{z)J

We

TT

a similar manner, that

= !.iog(i-z)+^z),
TTl

<

when |1 zj
regular

when

\z\

<

1.
1

and has the value 4/w 2 at the

origin.

should be observed that this implies that e'r8T , regarded as a function


has no branch-point at the origin, but has a simple zero there.

t It

of

z,

MODULAR FUNCTIONS AND

ELLIPTIC

438

The

at z

singularity of v(z)

PICABD'S

1 is therefore

THEOREM

most simply

explained by saying that l/v(z) has a logarithmic branch-point


1 once
1. When we continue v(z) analytically round z
at z
inverse
r
of
the
in the positive sense, we obtain a new branch

function, given

by the equation

*t

so that

+2

2i>(z)+l

This branch maps the cut plane conformally on the interior of


the 'quadrilateral' OA C2 A 2 of the figure, which is obtained from
t/(2t+1).
the fundamental region by the A-transformation r'
The point at infinity is the only other branch-point of v(z).
But as a circuit about the point at infinity is the same as a

circuit

any

about

and

1,

there

is

no need to consider

this

further.

by the equation
by repeated circuits about
appropriate orders. Each such branch maps

All the branches of the function t defined


A(r)

=z

can therefore be obtained

and z

in

the cut z-plane conformally on the interior of a 'quadrilateral'


congruent to the fundamental region of the A-group. It follows

that every transformation of the A-group can be generated by


t+2 and
repeated applications of the two transformations t'
t/(2t+ 1). For this reason, these transformations are called
t

the generating transformations of the A-group.

theorem
It is well known that a polynomial
value n times, where n is the degree
15.5. Picard's

takes any assigned finite

As a
polynomial is an integral function with a pole at infinity, we
naturally ask whether every integral function actually attains
any given value.
The theorem of Weierstrass, proved in 4.55, does not really
answer this question. It merely shows that we can find a
sequence of points at which the integral function approaches as
near as we please to the given value.
The answer to this question was first given by Pieard,f who
of the polynomial.

t Oomptes rendue, 88 (1879), 1024-7.

MODULAB FUNCTIONS AND PICARD'S THEOEEM

ELLIPTIC

proved that an

439

integral function actually attains every finite value,

This result

save one e